diff options
Diffstat (limited to 'drivers/md')
54 files changed, 28483 insertions, 0 deletions
diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig new file mode 100644 index 00000000000..ac43f98062f --- /dev/null +++ b/drivers/md/Kconfig @@ -0,0 +1,240 @@ +# +# Block device driver configuration +# + +menu "Multi-device support (RAID and LVM)" + +config MD + bool "Multiple devices driver support (RAID and LVM)" + help + Support multiple physical spindles through a single logical device. + Required for RAID and logical volume management. + +config BLK_DEV_MD + tristate "RAID support" + depends on MD + ---help--- + This driver lets you combine several hard disk partitions into one + logical block device. This can be used to simply append one + partition to another one or to combine several redundant hard disks + into a RAID1/4/5 device so as to provide protection against hard + disk failures. This is called "Software RAID" since the combining of + the partitions is done by the kernel. "Hardware RAID" means that the + combining is done by a dedicated controller; if you have such a + controller, you do not need to say Y here. + + More information about Software RAID on Linux is contained in the + Software RAID mini-HOWTO, available from + <http://www.tldp.org/docs.html#howto>. There you will also learn + where to get the supporting user space utilities raidtools. + + If unsure, say N. + +config MD_LINEAR + tristate "Linear (append) mode" + depends on BLK_DEV_MD + ---help--- + If you say Y here, then your multiple devices driver will be able to + use the so-called linear mode, i.e. it will combine the hard disk + partitions by simply appending one to the other. + + To compile this as a module, choose M here: the module + will be called linear. + + If unsure, say Y. + +config MD_RAID0 + tristate "RAID-0 (striping) mode" + depends on BLK_DEV_MD + ---help--- + If you say Y here, then your multiple devices driver will be able to + use the so-called raid0 mode, i.e. it will combine the hard disk + partitions into one logical device in such a fashion as to fill them + up evenly, one chunk here and one chunk there. This will increase + the throughput rate if the partitions reside on distinct disks. + + Information about Software RAID on Linux is contained in the + Software-RAID mini-HOWTO, available from + <http://www.tldp.org/docs.html#howto>. There you will also + learn where to get the supporting user space utilities raidtools. + + To compile this as a module, choose M here: the module + will be called raid0. + + If unsure, say Y. + +config MD_RAID1 + tristate "RAID-1 (mirroring) mode" + depends on BLK_DEV_MD + ---help--- + A RAID-1 set consists of several disk drives which are exact copies + of each other. In the event of a mirror failure, the RAID driver + will continue to use the operational mirrors in the set, providing + an error free MD (multiple device) to the higher levels of the + kernel. In a set with N drives, the available space is the capacity + of a single drive, and the set protects against a failure of (N - 1) + drives. + + Information about Software RAID on Linux is contained in the + Software-RAID mini-HOWTO, available from + <http://www.tldp.org/docs.html#howto>. There you will also + learn where to get the supporting user space utilities raidtools. + + If you want to use such a RAID-1 set, say Y. To compile this code + as a module, choose M here: the module will be called raid1. + + If unsure, say Y. + +config MD_RAID10 + tristate "RAID-10 (mirrored striping) mode (EXPERIMENTAL)" + depends on BLK_DEV_MD && EXPERIMENTAL + ---help--- + RAID-10 provides a combination of striping (RAID-0) and + mirroring (RAID-1) with easier configuration and more flexable + layout. + Unlike RAID-0, but like RAID-1, RAID-10 requires all devices to + be the same size (or at least, only as much as the smallest device + will be used). + RAID-10 provides a variety of layouts that provide different levels + of redundancy and performance. + + RAID-10 requires mdadm-1.7.0 or later, available at: + + ftp://ftp.kernel.org/pub/linux/utils/raid/mdadm/ + + If unsure, say Y. + +config MD_RAID5 + tristate "RAID-4/RAID-5 mode" + depends on BLK_DEV_MD + ---help--- + A RAID-5 set of N drives with a capacity of C MB per drive provides + the capacity of C * (N - 1) MB, and protects against a failure + of a single drive. For a given sector (row) number, (N - 1) drives + contain data sectors, and one drive contains the parity protection. + For a RAID-4 set, the parity blocks are present on a single drive, + while a RAID-5 set distributes the parity across the drives in one + of the available parity distribution methods. + + Information about Software RAID on Linux is contained in the + Software-RAID mini-HOWTO, available from + <http://www.tldp.org/docs.html#howto>. There you will also + learn where to get the supporting user space utilities raidtools. + + If you want to use such a RAID-4/RAID-5 set, say Y. To + compile this code as a module, choose M here: the module + will be called raid5. + + If unsure, say Y. + +config MD_RAID6 + tristate "RAID-6 mode" + depends on BLK_DEV_MD + ---help--- + A RAID-6 set of N drives with a capacity of C MB per drive + provides the capacity of C * (N - 2) MB, and protects + against a failure of any two drives. For a given sector + (row) number, (N - 2) drives contain data sectors, and two + drives contains two independent redundancy syndromes. Like + RAID-5, RAID-6 distributes the syndromes across the drives + in one of the available parity distribution methods. + + RAID-6 requires mdadm-1.5.0 or later, available at: + + ftp://ftp.kernel.org/pub/linux/utils/raid/mdadm/ + + If you want to use such a RAID-6 set, say Y. To compile + this code as a module, choose M here: the module will be + called raid6. + + If unsure, say Y. + +config MD_MULTIPATH + tristate "Multipath I/O support" + depends on BLK_DEV_MD + help + Multipath-IO is the ability of certain devices to address the same + physical disk over multiple 'IO paths'. The code ensures that such + paths can be defined and handled at runtime, and ensures that a + transparent failover to the backup path(s) happens if a IO errors + arrives on the primary path. + + If unsure, say N. + +config MD_FAULTY + tristate "Faulty test module for MD" + depends on BLK_DEV_MD + help + The "faulty" module allows for a block device that occasionally returns + read or write errors. It is useful for testing. + + In unsure, say N. + +config BLK_DEV_DM + tristate "Device mapper support" + depends on MD + ---help--- + Device-mapper is a low level volume manager. It works by allowing + people to specify mappings for ranges of logical sectors. Various + mapping types are available, in addition people may write their own + modules containing custom mappings if they wish. + + Higher level volume managers such as LVM2 use this driver. + + To compile this as a module, choose M here: the module will be + called dm-mod. + + If unsure, say N. + +config DM_CRYPT + tristate "Crypt target support" + depends on BLK_DEV_DM && EXPERIMENTAL + select CRYPTO + ---help--- + This device-mapper target allows you to create a device that + transparently encrypts the data on it. You'll need to activate + the ciphers you're going to use in the cryptoapi configuration. + + Information on how to use dm-crypt can be found on + + <http://www.saout.de/misc/dm-crypt/> + + To compile this code as a module, choose M here: the module will + be called dm-crypt. + + If unsure, say N. + +config DM_SNAPSHOT + tristate "Snapshot target (EXPERIMENTAL)" + depends on BLK_DEV_DM && EXPERIMENTAL + ---help--- + Allow volume managers to take writeable snapshots of a device. + +config DM_MIRROR + tristate "Mirror target (EXPERIMENTAL)" + depends on BLK_DEV_DM && EXPERIMENTAL + ---help--- + Allow volume managers to mirror logical volumes, also + needed for live data migration tools such as 'pvmove'. + +config DM_ZERO + tristate "Zero target (EXPERIMENTAL)" + depends on BLK_DEV_DM && EXPERIMENTAL + ---help--- + A target that discards writes, and returns all zeroes for + reads. Useful in some recovery situations. + +config DM_MULTIPATH + tristate "Multipath target (EXPERIMENTAL)" + depends on BLK_DEV_DM && EXPERIMENTAL + ---help--- + Allow volume managers to support multipath hardware. + +config DM_MULTIPATH_EMC + tristate "EMC CX/AX multipath support (EXPERIMENTAL)" + depends on DM_MULTIPATH && BLK_DEV_DM && EXPERIMENTAL + ---help--- + Multipath support for EMC CX/AX series hardware. + +endmenu + diff --git a/drivers/md/Makefile b/drivers/md/Makefile new file mode 100644 index 00000000000..90de9c146a5 --- /dev/null +++ b/drivers/md/Makefile @@ -0,0 +1,107 @@ +# +# Makefile for the kernel software RAID and LVM drivers. +# + +dm-mod-objs := dm.o dm-table.o dm-target.o dm-linear.o dm-stripe.o \ + dm-ioctl.o dm-io.o kcopyd.o +dm-multipath-objs := dm-hw-handler.o dm-path-selector.o dm-mpath.o +dm-snapshot-objs := dm-snap.o dm-exception-store.o +dm-mirror-objs := dm-log.o dm-raid1.o +raid6-objs := raid6main.o raid6algos.o raid6recov.o raid6tables.o \ + raid6int1.o raid6int2.o raid6int4.o \ + raid6int8.o raid6int16.o raid6int32.o \ + raid6altivec1.o raid6altivec2.o raid6altivec4.o \ + raid6altivec8.o \ + raid6mmx.o raid6sse1.o raid6sse2.o +hostprogs-y := mktables + +# Note: link order is important. All raid personalities +# and xor.o must come before md.o, as they each initialise +# themselves, and md.o may use the personalities when it +# auto-initialised. + +obj-$(CONFIG_MD_LINEAR) += linear.o +obj-$(CONFIG_MD_RAID0) += raid0.o +obj-$(CONFIG_MD_RAID1) += raid1.o +obj-$(CONFIG_MD_RAID10) += raid10.o +obj-$(CONFIG_MD_RAID5) += raid5.o xor.o +obj-$(CONFIG_MD_RAID6) += raid6.o xor.o +obj-$(CONFIG_MD_MULTIPATH) += multipath.o +obj-$(CONFIG_MD_FAULTY) += faulty.o +obj-$(CONFIG_BLK_DEV_MD) += md.o +obj-$(CONFIG_BLK_DEV_DM) += dm-mod.o +obj-$(CONFIG_DM_CRYPT) += dm-crypt.o +obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o +obj-$(CONFIG_DM_MULTIPATH_EMC) += dm-emc.o +obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o +obj-$(CONFIG_DM_MIRROR) += dm-mirror.o +obj-$(CONFIG_DM_ZERO) += dm-zero.o + +quiet_cmd_unroll = UNROLL $@ + cmd_unroll = $(PERL) $(srctree)/$(src)/unroll.pl $(UNROLL) \ + < $< > $@ || ( rm -f $@ && exit 1 ) + +ifeq ($(CONFIG_ALTIVEC),y) +altivec_flags := -maltivec -mabi=altivec +endif + +targets += raid6int1.c +$(obj)/raid6int1.c: UNROLL := 1 +$(obj)/raid6int1.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +targets += raid6int2.c +$(obj)/raid6int2.c: UNROLL := 2 +$(obj)/raid6int2.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +targets += raid6int4.c +$(obj)/raid6int4.c: UNROLL := 4 +$(obj)/raid6int4.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +targets += raid6int8.c +$(obj)/raid6int8.c: UNROLL := 8 +$(obj)/raid6int8.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +targets += raid6int16.c +$(obj)/raid6int16.c: UNROLL := 16 +$(obj)/raid6int16.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +targets += raid6int32.c +$(obj)/raid6int32.c: UNROLL := 32 +$(obj)/raid6int32.c: $(src)/raid6int.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +CFLAGS_raid6altivec1.o += $(altivec_flags) +targets += raid6altivec1.c +$(obj)/raid6altivec1.c: UNROLL := 1 +$(obj)/raid6altivec1.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +CFLAGS_raid6altivec2.o += $(altivec_flags) +targets += raid6altivec2.c +$(obj)/raid6altivec2.c: UNROLL := 2 +$(obj)/raid6altivec2.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +CFLAGS_raid6altivec4.o += $(altivec_flags) +targets += raid6altivec4.c +$(obj)/raid6altivec4.c: UNROLL := 4 +$(obj)/raid6altivec4.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +CFLAGS_raid6altivec8.o += $(altivec_flags) +targets += raid6altivec8.c +$(obj)/raid6altivec8.c: UNROLL := 8 +$(obj)/raid6altivec8.c: $(src)/raid6altivec.uc $(src)/unroll.pl FORCE + $(call if_changed,unroll) + +quiet_cmd_mktable = TABLE $@ + cmd_mktable = $(obj)/mktables > $@ || ( rm -f $@ && exit 1 ) + +targets += raid6tables.c +$(obj)/raid6tables.c: $(obj)/mktables FORCE + $(call if_changed,mktable) diff --git a/drivers/md/dm-bio-list.h b/drivers/md/dm-bio-list.h new file mode 100644 index 00000000000..bc021e1fd4d --- /dev/null +++ b/drivers/md/dm-bio-list.h @@ -0,0 +1,68 @@ +/* + * Copyright (C) 2004 Red Hat UK Ltd. + * + * This file is released under the GPL. + */ + +#ifndef DM_BIO_LIST_H +#define DM_BIO_LIST_H + +#include <linux/bio.h> + +struct bio_list { + struct bio *head; + struct bio *tail; +}; + +static inline void bio_list_init(struct bio_list *bl) +{ + bl->head = bl->tail = NULL; +} + +static inline void bio_list_add(struct bio_list *bl, struct bio *bio) +{ + bio->bi_next = NULL; + + if (bl->tail) + bl->tail->bi_next = bio; + else + bl->head = bio; + + bl->tail = bio; +} + +static inline void bio_list_merge(struct bio_list *bl, struct bio_list *bl2) +{ + if (bl->tail) + bl->tail->bi_next = bl2->head; + else + bl->head = bl2->head; + + bl->tail = bl2->tail; +} + +static inline struct bio *bio_list_pop(struct bio_list *bl) +{ + struct bio *bio = bl->head; + + if (bio) { + bl->head = bl->head->bi_next; + if (!bl->head) + bl->tail = NULL; + + bio->bi_next = NULL; + } + + return bio; +} + +static inline struct bio *bio_list_get(struct bio_list *bl) +{ + struct bio *bio = bl->head; + + bl->head = bl->tail = NULL; + + return bio; +} + +#endif diff --git a/drivers/md/dm-bio-record.h b/drivers/md/dm-bio-record.h new file mode 100644 index 00000000000..d3ec217847d --- /dev/null +++ b/drivers/md/dm-bio-record.h @@ -0,0 +1,45 @@ +/* + * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + */ + +#ifndef DM_BIO_RECORD_H +#define DM_BIO_RECORD_H + +#include <linux/bio.h> + +/* + * There are lots of mutable fields in the bio struct that get + * changed by the lower levels of the block layer. Some targets, + * such as multipath, may wish to resubmit a bio on error. The + * functions in this file help the target record and restore the + * original bio state. + */ +struct dm_bio_details { + sector_t bi_sector; + struct block_device *bi_bdev; + unsigned int bi_size; + unsigned short bi_idx; + unsigned long bi_flags; +}; + +static inline void dm_bio_record(struct dm_bio_details *bd, struct bio *bio) +{ + bd->bi_sector = bio->bi_sector; + bd->bi_bdev = bio->bi_bdev; + bd->bi_size = bio->bi_size; + bd->bi_idx = bio->bi_idx; + bd->bi_flags = bio->bi_flags; +} + +static inline void dm_bio_restore(struct dm_bio_details *bd, struct bio *bio) +{ + bio->bi_sector = bd->bi_sector; + bio->bi_bdev = bd->bi_bdev; + bio->bi_size = bd->bi_size; + bio->bi_idx = bd->bi_idx; + bio->bi_flags = bd->bi_flags; +} + +#endif diff --git a/drivers/md/dm-crypt.c b/drivers/md/dm-crypt.c new file mode 100644 index 00000000000..77619a56e2b --- /dev/null +++ b/drivers/md/dm-crypt.c @@ -0,0 +1,977 @@ +/* + * Copyright (C) 2003 Christophe Saout <christophe@saout.de> + * Copyright (C) 2004 Clemens Fruhwirth <clemens@endorphin.org> + * + * This file is released under the GPL. + */ + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/bio.h> +#include <linux/blkdev.h> +#include <linux/mempool.h> +#include <linux/slab.h> +#include <linux/crypto.h> +#include <linux/workqueue.h> +#include <asm/atomic.h> +#include <asm/scatterlist.h> +#include <asm/page.h> + +#include "dm.h" + +#define PFX "crypt: " + +/* + * per bio private data + */ +struct crypt_io { + struct dm_target *target; + struct bio *bio; + struct bio *first_clone; + struct work_struct work; + atomic_t pending; + int error; +}; + +/* + * context holding the current state of a multi-part conversion + */ +struct convert_context { + struct bio *bio_in; + struct bio *bio_out; + unsigned int offset_in; + unsigned int offset_out; + unsigned int idx_in; + unsigned int idx_out; + sector_t sector; + int write; +}; + +struct crypt_config; + +struct crypt_iv_operations { + int (*ctr)(struct crypt_config *cc, struct dm_target *ti, + const char *opts); + void (*dtr)(struct crypt_config *cc); + const char *(*status)(struct crypt_config *cc); + int (*generator)(struct crypt_config *cc, u8 *iv, sector_t sector); +}; + +/* + * Crypt: maps a linear range of a block device + * and encrypts / decrypts at the same time. + */ +struct crypt_config { + struct dm_dev *dev; + sector_t start; + + /* + * pool for per bio private data and + * for encryption buffer pages + */ + mempool_t *io_pool; + mempool_t *page_pool; + + /* + * crypto related data + */ + struct crypt_iv_operations *iv_gen_ops; + char *iv_mode; + void *iv_gen_private; + sector_t iv_offset; + unsigned int iv_size; + + struct crypto_tfm *tfm; + unsigned int key_size; + u8 key[0]; +}; + +#define MIN_IOS 256 +#define MIN_POOL_PAGES 32 +#define MIN_BIO_PAGES 8 + +static kmem_cache_t *_crypt_io_pool; + +/* + * Mempool alloc and free functions for the page + */ +static void *mempool_alloc_page(unsigned int __nocast gfp_mask, void *data) +{ + return alloc_page(gfp_mask); +} + +static void mempool_free_page(void *page, void *data) +{ + __free_page(page); +} + + +/* + * Different IV generation algorithms: + * + * plain: the initial vector is the 32-bit low-endian version of the sector + * number, padded with zeros if neccessary. + * + * ess_iv: "encrypted sector|salt initial vector", the sector number is + * encrypted with the bulk cipher using a salt as key. The salt + * should be derived from the bulk cipher's key via hashing. + * + * plumb: unimplemented, see: + * http://article.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/454 + */ + +static int crypt_iv_plain_gen(struct crypt_config *cc, u8 *iv, sector_t sector) +{ + memset(iv, 0, cc->iv_size); + *(u32 *)iv = cpu_to_le32(sector & 0xffffffff); + + return 0; +} + +static int crypt_iv_essiv_ctr(struct crypt_config *cc, struct dm_target *ti, + const char *opts) +{ + struct crypto_tfm *essiv_tfm; + struct crypto_tfm *hash_tfm; + struct scatterlist sg; + unsigned int saltsize; + u8 *salt; + + if (opts == NULL) { + ti->error = PFX "Digest algorithm missing for ESSIV mode"; + return -EINVAL; + } + + /* Hash the cipher key with the given hash algorithm */ + hash_tfm = crypto_alloc_tfm(opts, 0); + if (hash_tfm == NULL) { + ti->error = PFX "Error initializing ESSIV hash"; + return -EINVAL; + } + + if (crypto_tfm_alg_type(hash_tfm) != CRYPTO_ALG_TYPE_DIGEST) { + ti->error = PFX "Expected digest algorithm for ESSIV hash"; + crypto_free_tfm(hash_tfm); + return -EINVAL; + } + + saltsize = crypto_tfm_alg_digestsize(hash_tfm); + salt = kmalloc(saltsize, GFP_KERNEL); + if (salt == NULL) { + ti->error = PFX "Error kmallocing salt storage in ESSIV"; + crypto_free_tfm(hash_tfm); + return -ENOMEM; + } + + sg.page = virt_to_page(cc->key); + sg.offset = offset_in_page(cc->key); + sg.length = cc->key_size; + crypto_digest_digest(hash_tfm, &sg, 1, salt); + crypto_free_tfm(hash_tfm); + + /* Setup the essiv_tfm with the given salt */ + essiv_tfm = crypto_alloc_tfm(crypto_tfm_alg_name(cc->tfm), + CRYPTO_TFM_MODE_ECB); + if (essiv_tfm == NULL) { + ti->error = PFX "Error allocating crypto tfm for ESSIV"; + kfree(salt); + return -EINVAL; + } + if (crypto_tfm_alg_blocksize(essiv_tfm) + != crypto_tfm_alg_ivsize(cc->tfm)) { + ti->error = PFX "Block size of ESSIV cipher does " + "not match IV size of block cipher"; + crypto_free_tfm(essiv_tfm); + kfree(salt); + return -EINVAL; + } + if (crypto_cipher_setkey(essiv_tfm, salt, saltsize) < 0) { + ti->error = PFX "Failed to set key for ESSIV cipher"; + crypto_free_tfm(essiv_tfm); + kfree(salt); + return -EINVAL; + } + kfree(salt); + + cc->iv_gen_private = (void *)essiv_tfm; + return 0; +} + +static void crypt_iv_essiv_dtr(struct crypt_config *cc) +{ + crypto_free_tfm((struct crypto_tfm *)cc->iv_gen_private); + cc->iv_gen_private = NULL; +} + +static int crypt_iv_essiv_gen(struct crypt_config *cc, u8 *iv, sector_t sector) +{ + struct scatterlist sg = { NULL, }; + + memset(iv, 0, cc->iv_size); + *(u64 *)iv = cpu_to_le64(sector); + + sg.page = virt_to_page(iv); + sg.offset = offset_in_page(iv); + sg.length = cc->iv_size; + crypto_cipher_encrypt((struct crypto_tfm *)cc->iv_gen_private, + &sg, &sg, cc->iv_size); + + return 0; +} + +static struct crypt_iv_operations crypt_iv_plain_ops = { + .generator = crypt_iv_plain_gen +}; + +static struct crypt_iv_operations crypt_iv_essiv_ops = { + .ctr = crypt_iv_essiv_ctr, + .dtr = crypt_iv_essiv_dtr, + .generator = crypt_iv_essiv_gen +}; + + +static inline int +crypt_convert_scatterlist(struct crypt_config *cc, struct scatterlist *out, + struct scatterlist *in, unsigned int length, + int write, sector_t sector) +{ + u8 iv[cc->iv_size]; + int r; + + if (cc->iv_gen_ops) { + r = cc->iv_gen_ops->generator(cc, iv, sector); + if (r < 0) + return r; + + if (write) + r = crypto_cipher_encrypt_iv(cc->tfm, out, in, length, iv); + else + r = crypto_cipher_decrypt_iv(cc->tfm, out, in, length, iv); + } else { + if (write) + r = crypto_cipher_encrypt(cc->tfm, out, in, length); + else + r = crypto_cipher_decrypt(cc->tfm, out, in, length); + } + + return r; +} + +static void +crypt_convert_init(struct crypt_config *cc, struct convert_context *ctx, + struct bio *bio_out, struct bio *bio_in, + sector_t sector, int write) +{ + ctx->bio_in = bio_in; + ctx->bio_out = bio_out; + ctx->offset_in = 0; + ctx->offset_out = 0; + ctx->idx_in = bio_in ? bio_in->bi_idx : 0; + ctx->idx_out = bio_out ? bio_out->bi_idx : 0; + ctx->sector = sector + cc->iv_offset; + ctx->write = write; +} + +/* + * Encrypt / decrypt data from one bio to another one (can be the same one) + */ +static int crypt_convert(struct crypt_config *cc, + struct convert_context *ctx) +{ + int r = 0; + + while(ctx->idx_in < ctx->bio_in->bi_vcnt && + ctx->idx_out < ctx->bio_out->bi_vcnt) { + struct bio_vec *bv_in = bio_iovec_idx(ctx->bio_in, ctx->idx_in); + struct bio_vec *bv_out = bio_iovec_idx(ctx->bio_out, ctx->idx_out); + struct scatterlist sg_in = { + .page = bv_in->bv_page, + .offset = bv_in->bv_offset + ctx->offset_in, + .length = 1 << SECTOR_SHIFT + }; + struct scatterlist sg_out = { + .page = bv_out->bv_page, + .offset = bv_out->bv_offset + ctx->offset_out, + .length = 1 << SECTOR_SHIFT + }; + + ctx->offset_in += sg_in.length; + if (ctx->offset_in >= bv_in->bv_len) { + ctx->offset_in = 0; + ctx->idx_in++; + } + + ctx->offset_out += sg_out.length; + if (ctx->offset_out >= bv_out->bv_len) { + ctx->offset_out = 0; + ctx->idx_out++; + } + + r = crypt_convert_scatterlist(cc, &sg_out, &sg_in, sg_in.length, + ctx->write, ctx->sector); + if (r < 0) + break; + + ctx->sector++; + } + + return r; +} + +/* + * Generate a new unfragmented bio with the given size + * This should never violate the device limitations + * May return a smaller bio when running out of pages + */ +static struct bio * +crypt_alloc_buffer(struct crypt_config *cc, unsigned int size, + struct bio *base_bio, unsigned int *bio_vec_idx) +{ + struct bio *bio; + unsigned int nr_iovecs = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; + int gfp_mask = GFP_NOIO | __GFP_HIGHMEM; + unsigned long flags = current->flags; + unsigned int i; + + /* + * Tell VM to act less aggressively and fail earlier. + * This is not necessary but increases throughput. + * FIXME: Is this really intelligent? + */ + current->flags &= ~PF_MEMALLOC; + + if (base_bio) + bio = bio_clone(base_bio, GFP_NOIO); + else + bio = bio_alloc(GFP_NOIO, nr_iovecs); + if (!bio) { + if (flags & PF_MEMALLOC) + current->flags |= PF_MEMALLOC; + return NULL; + } + + /* if the last bio was not complete, continue where that one ended */ + bio->bi_idx = *bio_vec_idx; + bio->bi_vcnt = *bio_vec_idx; + bio->bi_size = 0; + bio->bi_flags &= ~(1 << BIO_SEG_VALID); + + /* bio->bi_idx pages have already been allocated */ + size -= bio->bi_idx * PAGE_SIZE; + + for(i = bio->bi_idx; i < nr_iovecs; i++) { + struct bio_vec *bv = bio_iovec_idx(bio, i); + + bv->bv_page = mempool_alloc(cc->page_pool, gfp_mask); + if (!bv->bv_page) + break; + + /* + * if additional pages cannot be allocated without waiting, + * return a partially allocated bio, the caller will then try + * to allocate additional bios while submitting this partial bio + */ + if ((i - bio->bi_idx) == (MIN_BIO_PAGES - 1)) + gfp_mask = (gfp_mask | __GFP_NOWARN) & ~__GFP_WAIT; + + bv->bv_offset = 0; + if (size > PAGE_SIZE) + bv->bv_len = PAGE_SIZE; + else + bv->bv_len = size; + + bio->bi_size += bv->bv_len; + bio->bi_vcnt++; + size -= bv->bv_len; + } + + if (flags & PF_MEMALLOC) + current->flags |= PF_MEMALLOC; + + if (!bio->bi_size) { + bio_put(bio); + return NULL; + } + + /* + * Remember the last bio_vec allocated to be able + * to correctly continue after the splitting. + */ + *bio_vec_idx = bio->bi_vcnt; + + return bio; +} + +static void crypt_free_buffer_pages(struct crypt_config *cc, + struct bio *bio, unsigned int bytes) +{ + unsigned int i, start, end; + struct bio_vec *bv; + + /* + * This is ugly, but Jens Axboe thinks that using bi_idx in the + * endio function is too dangerous at the moment, so I calculate the + * correct position using bi_vcnt and bi_size. + * The bv_offset and bv_len fields might already be modified but we + * know that we always allocated whole pages. + * A fix to the bi_idx issue in the kernel is in the works, so + * we will hopefully be able to revert to the cleaner solution soon. + */ + i = bio->bi_vcnt - 1; + bv = bio_iovec_idx(bio, i); + end = (i << PAGE_SHIFT) + (bv->bv_offset + bv->bv_len) - bio->bi_size; + start = end - bytes; + + start >>= PAGE_SHIFT; + if (!bio->bi_size) + end = bio->bi_vcnt; + else + end >>= PAGE_SHIFT; + + for(i = start; i < end; i++) { + bv = bio_iovec_idx(bio, i); + BUG_ON(!bv->bv_page); + mempool_free(bv->bv_page, cc->page_pool); + bv->bv_page = NULL; + } +} + +/* + * One of the bios was finished. Check for completion of + * the whole request and correctly clean up the buffer. + */ +static void dec_pending(struct crypt_io *io, int error) +{ + struct crypt_config *cc = (struct crypt_config *) io->target->private; + + if (error < 0) + io->error = error; + + if (!atomic_dec_and_test(&io->pending)) + return; + + if (io->first_clone) + bio_put(io->first_clone); + + bio_endio(io->bio, io->bio->bi_size, io->error); + + mempool_free(io, cc->io_pool); +} + +/* + * kcryptd: + * + * Needed because it would be very unwise to do decryption in an + * interrupt context, so bios returning from read requests get + * queued here. + */ +static struct workqueue_struct *_kcryptd_workqueue; + +static void kcryptd_do_work(void *data) +{ + struct crypt_io *io = (struct crypt_io *) data; + struct crypt_config *cc = (struct crypt_config *) io->target->private; + struct convert_context ctx; + int r; + + crypt_convert_init(cc, &ctx, io->bio, io->bio, + io->bio->bi_sector - io->target->begin, 0); + r = crypt_convert(cc, &ctx); + + dec_pending(io, r); +} + +static void kcryptd_queue_io(struct crypt_io *io) +{ + INIT_WORK(&io->work, kcryptd_do_work, io); + queue_work(_kcryptd_workqueue, &io->work); +} + +/* + * Decode key from its hex representation + */ +static int crypt_decode_key(u8 *key, char *hex, unsigned int size) +{ + char buffer[3]; + char *endp; + unsigned int i; + + buffer[2] = '\0'; + + for(i = 0; i < size; i++) { + buffer[0] = *hex++; + buffer[1] = *hex++; + + key[i] = (u8)simple_strtoul(buffer, &endp, 16); + + if (endp != &buffer[2]) + return -EINVAL; + } + + if (*hex != '\0') + return -EINVAL; + + return 0; +} + +/* + * Encode key into its hex representation + */ +static void crypt_encode_key(char *hex, u8 *key, unsigned int size) +{ + unsigned int i; + + for(i = 0; i < size; i++) { + sprintf(hex, "%02x", *key); + hex += 2; + key++; + } +} + +/* + * Construct an encryption mapping: + * <cipher> <key> <iv_offset> <dev_path> <start> + */ +static int crypt_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + struct crypt_config *cc; + struct crypto_tfm *tfm; + char *tmp; + char *cipher; + char *chainmode; + char *ivmode; + char *ivopts; + unsigned int crypto_flags; + unsigned int key_size; + + if (argc != 5) { + ti->error = PFX "Not enough arguments"; + return -EINVAL; + } + + tmp = argv[0]; + cipher = strsep(&tmp, "-"); + chainmode = strsep(&tmp, "-"); + ivopts = strsep(&tmp, "-"); + ivmode = strsep(&ivopts, ":"); + + if (tmp) + DMWARN(PFX "Unexpected additional cipher options"); + + key_size = strlen(argv[1]) >> 1; + + cc = kmalloc(sizeof(*cc) + key_size * sizeof(u8), GFP_KERNEL); + if (cc == NULL) { + ti->error = + PFX "Cannot allocate transparent encryption context"; + return -ENOMEM; + } + + cc->key_size = key_size; + if ((!key_size && strcmp(argv[1], "-") != 0) || + (key_size && crypt_decode_key(cc->key, argv[1], key_size) < 0)) { + ti->error = PFX "Error decoding key"; + goto bad1; + } + + /* Compatiblity mode for old dm-crypt cipher strings */ + if (!chainmode || (strcmp(chainmode, "plain") == 0 && !ivmode)) { + chainmode = "cbc"; + ivmode = "plain"; + } + + /* Choose crypto_flags according to chainmode */ + if (strcmp(chainmode, "cbc") == 0) + crypto_flags = CRYPTO_TFM_MODE_CBC; + else if (strcmp(chainmode, "ecb") == 0) + crypto_flags = CRYPTO_TFM_MODE_ECB; + else { + ti->error = PFX "Unknown chaining mode"; + goto bad1; + } + + if (crypto_flags != CRYPTO_TFM_MODE_ECB && !ivmode) { + ti->error = PFX "This chaining mode requires an IV mechanism"; + goto bad1; + } + + tfm = crypto_alloc_tfm(cipher, crypto_flags); + if (!tfm) { + ti->error = PFX "Error allocating crypto tfm"; + goto bad1; + } + if (crypto_tfm_alg_type(tfm) != CRYPTO_ALG_TYPE_CIPHER) { + ti->error = PFX "Expected cipher algorithm"; + goto bad2; + } + + cc->tfm = tfm; + + /* + * Choose ivmode. Valid modes: "plain", "essiv:<esshash>". + * See comments at iv code + */ + + if (ivmode == NULL) + cc->iv_gen_ops = NULL; + else if (strcmp(ivmode, "plain") == 0) + cc->iv_gen_ops = &crypt_iv_plain_ops; + else if (strcmp(ivmode, "essiv") == 0) + cc->iv_gen_ops = &crypt_iv_essiv_ops; + else { + ti->error = PFX "Invalid IV mode"; + goto bad2; + } + + if (cc->iv_gen_ops && cc->iv_gen_ops->ctr && + cc->iv_gen_ops->ctr(cc, ti, ivopts) < 0) + goto bad2; + + if (tfm->crt_cipher.cit_decrypt_iv && tfm->crt_cipher.cit_encrypt_iv) + /* at least a 64 bit sector number should fit in our buffer */ + cc->iv_size = max(crypto_tfm_alg_ivsize(tfm), + (unsigned int)(sizeof(u64) / sizeof(u8))); + else { + cc->iv_size = 0; + if (cc->iv_gen_ops) { + DMWARN(PFX "Selected cipher does not support IVs"); + if (cc->iv_gen_ops->dtr) + cc->iv_gen_ops->dtr(cc); + cc->iv_gen_ops = NULL; + } + } + + cc->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab, + mempool_free_slab, _crypt_io_pool); + if (!cc->io_pool) { + ti->error = PFX "Cannot allocate crypt io mempool"; + goto bad3; + } + + cc->page_pool = mempool_create(MIN_POOL_PAGES, mempool_alloc_page, + mempool_free_page, NULL); + if (!cc->page_pool) { + ti->error = PFX "Cannot allocate page mempool"; + goto bad4; + } + + if (tfm->crt_cipher.cit_setkey(tfm, cc->key, key_size) < 0) { + ti->error = PFX "Error setting key"; + goto bad5; + } + + if (sscanf(argv[2], SECTOR_FORMAT, &cc->iv_offset) != 1) { + ti->error = PFX "Invalid iv_offset sector"; + goto bad5; + } + + if (sscanf(argv[4], SECTOR_FORMAT, &cc->start) != 1) { + ti->error = PFX "Invalid device sector"; + goto bad5; + } + + if (dm_get_device(ti, argv[3], cc->start, ti->len, + dm_table_get_mode(ti->table), &cc->dev)) { + ti->error = PFX "Device lookup failed"; + goto bad5; + } + + if (ivmode && cc->iv_gen_ops) { + if (ivopts) + *(ivopts - 1) = ':'; + cc->iv_mode = kmalloc(strlen(ivmode) + 1, GFP_KERNEL); + if (!cc->iv_mode) { + ti->error = PFX "Error kmallocing iv_mode string"; + goto bad5; + } + strcpy(cc->iv_mode, ivmode); + } else + cc->iv_mode = NULL; + + ti->private = cc; + return 0; + +bad5: + mempool_destroy(cc->page_pool); +bad4: + mempool_destroy(cc->io_pool); +bad3: + if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) + cc->iv_gen_ops->dtr(cc); +bad2: + crypto_free_tfm(tfm); +bad1: + kfree(cc); + return -EINVAL; +} + +static void crypt_dtr(struct dm_target *ti) +{ + struct crypt_config *cc = (struct crypt_config *) ti->private; + + mempool_destroy(cc->page_pool); + mempool_destroy(cc->io_pool); + + if (cc->iv_mode) + kfree(cc->iv_mode); + if (cc->iv_gen_ops && cc->iv_gen_ops->dtr) + cc->iv_gen_ops->dtr(cc); + crypto_free_tfm(cc->tfm); + dm_put_device(ti, cc->dev); + kfree(cc); +} + +static int crypt_endio(struct bio *bio, unsigned int done, int error) +{ + struct crypt_io *io = (struct crypt_io *) bio->bi_private; + struct crypt_config *cc = (struct crypt_config *) io->target->private; + + if (bio_data_dir(bio) == WRITE) { + /* + * free the processed pages, even if + * it's only a partially completed write + */ + crypt_free_buffer_pages(cc, bio, done); + } + + if (bio->bi_size) + return 1; + + bio_put(bio); + + /* + * successful reads are decrypted by the worker thread + */ + if ((bio_data_dir(bio) == READ) + && bio_flagged(bio, BIO_UPTODATE)) { + kcryptd_queue_io(io); + return 0; + } + + dec_pending(io, error); + return error; +} + +static inline struct bio * +crypt_clone(struct crypt_config *cc, struct crypt_io *io, struct bio *bio, + sector_t sector, unsigned int *bvec_idx, + struct convert_context *ctx) +{ + struct bio *clone; + + if (bio_data_dir(bio) == WRITE) { + clone = crypt_alloc_buffer(cc, bio->bi_size, + io->first_clone, bvec_idx); + if (clone) { + ctx->bio_out = clone; + if (crypt_convert(cc, ctx) < 0) { + crypt_free_buffer_pages(cc, clone, + clone->bi_size); + bio_put(clone); + return NULL; + } + } + } else { + /* + * The block layer might modify the bvec array, so always + * copy the required bvecs because we need the original + * one in order to decrypt the whole bio data *afterwards*. + */ + clone = bio_alloc(GFP_NOIO, bio_segments(bio)); + if (clone) { + clone->bi_idx = 0; + clone->bi_vcnt = bio_segments(bio); + clone->bi_size = bio->bi_size; + memcpy(clone->bi_io_vec, bio_iovec(bio), + sizeof(struct bio_vec) * clone->bi_vcnt); + } + } + + if (!clone) + return NULL; + + clone->bi_private = io; + clone->bi_end_io = crypt_endio; + clone->bi_bdev = cc->dev->bdev; + clone->bi_sector = cc->start + sector; + clone->bi_rw = bio->bi_rw; + + return clone; +} + +static int crypt_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + struct crypt_config *cc = (struct crypt_config *) ti->private; + struct crypt_io *io = mempool_alloc(cc->io_pool, GFP_NOIO); + struct convert_context ctx; + struct bio *clone; + unsigned int remaining = bio->bi_size; + sector_t sector = bio->bi_sector - ti->begin; + unsigned int bvec_idx = 0; + + io->target = ti; + io->bio = bio; + io->first_clone = NULL; + io->error = 0; + atomic_set(&io->pending, 1); /* hold a reference */ + + if (bio_data_dir(bio) == WRITE) + crypt_convert_init(cc, &ctx, NULL, bio, sector, 1); + + /* + * The allocated buffers can be smaller than the whole bio, + * so repeat the whole process until all the data can be handled. + */ + while (remaining) { + clone = crypt_clone(cc, io, bio, sector, &bvec_idx, &ctx); + if (!clone) + goto cleanup; + + if (!io->first_clone) { + /* + * hold a reference to the first clone, because it + * holds the bio_vec array and that can't be freed + * before all other clones are released + */ + bio_get(clone); + io->first_clone = clone; + } + atomic_inc(&io->pending); + + remaining -= clone->bi_size; + sector += bio_sectors(clone); + + generic_make_request(clone); + + /* out of memory -> run queues */ + if (remaining) + blk_congestion_wait(bio_data_dir(clone), HZ/100); + } + + /* drop reference, clones could have returned before we reach this */ + dec_pending(io, 0); + return 0; + +cleanup: + if (io->first_clone) { + dec_pending(io, -ENOMEM); + return 0; + } + + /* if no bio has been dispatched yet, we can directly return the error */ + mempool_free(io, cc->io_pool); + return -ENOMEM; +} + +static int crypt_status(struct dm_target *ti, status_type_t type, + char *result, unsigned int maxlen) +{ + struct crypt_config *cc = (struct crypt_config *) ti->private; + const char *cipher; + const char *chainmode = NULL; + unsigned int sz = 0; + + switch (type) { + case STATUSTYPE_INFO: + result[0] = '\0'; + break; + + case STATUSTYPE_TABLE: + cipher = crypto_tfm_alg_name(cc->tfm); + + switch(cc->tfm->crt_cipher.cit_mode) { + case CRYPTO_TFM_MODE_CBC: + chainmode = "cbc"; + break; + case CRYPTO_TFM_MODE_ECB: + chainmode = "ecb"; + break; + default: + BUG(); + } + + if (cc->iv_mode) + DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode); + else + DMEMIT("%s-%s ", cipher, chainmode); + + if (cc->key_size > 0) { + if ((maxlen - sz) < ((cc->key_size << 1) + 1)) + return -ENOMEM; + + crypt_encode_key(result + sz, cc->key, cc->key_size); + sz += cc->key_size << 1; + } else { + if (sz >= maxlen) + return -ENOMEM; + result[sz++] = '-'; + } + + DMEMIT(" " SECTOR_FORMAT " %s " SECTOR_FORMAT, + cc->iv_offset, cc->dev->name, cc->start); + break; + } + return 0; +} + +static struct target_type crypt_target = { + .name = "crypt", + .version= {1, 1, 0}, + .module = THIS_MODULE, + .ctr = crypt_ctr, + .dtr = crypt_dtr, + .map = crypt_map, + .status = crypt_status, +}; + +static int __init dm_crypt_init(void) +{ + int r; + + _crypt_io_pool = kmem_cache_create("dm-crypt_io", + sizeof(struct crypt_io), + 0, 0, NULL, NULL); + if (!_crypt_io_pool) + return -ENOMEM; + + _kcryptd_workqueue = create_workqueue("kcryptd"); + if (!_kcryptd_workqueue) { + r = -ENOMEM; + DMERR(PFX "couldn't create kcryptd"); + goto bad1; + } + + r = dm_register_target(&crypt_target); + if (r < 0) { + DMERR(PFX "register failed %d", r); + goto bad2; + } + + return 0; + +bad2: + destroy_workqueue(_kcryptd_workqueue); +bad1: + kmem_cache_destroy(_crypt_io_pool); + return r; +} + +static void __exit dm_crypt_exit(void) +{ + int r = dm_unregister_target(&crypt_target); + + if (r < 0) + DMERR(PFX "unregister failed %d", r); + + destroy_workqueue(_kcryptd_workqueue); + kmem_cache_destroy(_crypt_io_pool); +} + +module_init(dm_crypt_init); +module_exit(dm_crypt_exit); + +MODULE_AUTHOR("Christophe Saout <christophe@saout.de>"); +MODULE_DESCRIPTION(DM_NAME " target for transparent encryption / decryption"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-emc.c b/drivers/md/dm-emc.c new file mode 100644 index 00000000000..70065866459 --- /dev/null +++ b/drivers/md/dm-emc.c @@ -0,0 +1,359 @@ +/* + * Copyright (C) 2004 SUSE LINUX Products GmbH. All rights reserved. + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + * + * Multipath support for EMC CLARiiON AX/CX-series hardware. + */ + +#include "dm.h" +#include "dm-hw-handler.h" +#include <scsi/scsi.h> +#include <scsi/scsi_cmnd.h> + +struct emc_handler { + spinlock_t lock; + + /* Whether we should send the short trespass command (FC-series) + * or the long version (default for AX/CX CLARiiON arrays). */ + unsigned short_trespass; + /* Whether or not to honor SCSI reservations when initiating a + * switch-over. Default: Don't. */ + unsigned hr; + + unsigned char sense[SCSI_SENSE_BUFFERSIZE]; +}; + +#define TRESPASS_PAGE 0x22 +#define EMC_FAILOVER_TIMEOUT (60 * HZ) + +/* Code borrowed from dm-lsi-rdac by Mike Christie */ + +static inline void free_bio(struct bio *bio) +{ + __free_page(bio->bi_io_vec[0].bv_page); + bio_put(bio); +} + +static int emc_endio(struct bio *bio, unsigned int bytes_done, int error) +{ + struct path *path = bio->bi_private; + + if (bio->bi_size) + return 1; + + /* We also need to look at the sense keys here whether or not to + * switch to the next PG etc. + * + * For now simple logic: either it works or it doesn't. + */ + if (error) + dm_pg_init_complete(path, MP_FAIL_PATH); + else + dm_pg_init_complete(path, 0); + + /* request is freed in block layer */ + free_bio(bio); + + return 0; +} + +static struct bio *get_failover_bio(struct path *path, unsigned data_size) +{ + struct bio *bio; + struct page *page; + + bio = bio_alloc(GFP_ATOMIC, 1); + if (!bio) { + DMERR("dm-emc: get_failover_bio: bio_alloc() failed."); + return NULL; + } + + bio->bi_rw |= (1 << BIO_RW); + bio->bi_bdev = path->dev->bdev; + bio->bi_sector = 0; + bio->bi_private = path; + bio->bi_end_io = emc_endio; + + page = alloc_page(GFP_ATOMIC); + if (!page) { + DMERR("dm-emc: get_failover_bio: alloc_page() failed."); + bio_put(bio); + return NULL; + } + + if (bio_add_page(bio, page, data_size, 0) != data_size) { + DMERR("dm-emc: get_failover_bio: alloc_page() failed."); + __free_page(page); + bio_put(bio); + return NULL; + } + + return bio; +} + +static struct request *get_failover_req(struct emc_handler *h, + struct bio *bio, struct path *path) +{ + struct request *rq; + struct block_device *bdev = bio->bi_bdev; + struct request_queue *q = bdev_get_queue(bdev); + + /* FIXME: Figure out why it fails with GFP_ATOMIC. */ + rq = blk_get_request(q, WRITE, __GFP_WAIT); + if (!rq) { + DMERR("dm-emc: get_failover_req: blk_get_request failed"); + return NULL; + } + + rq->bio = rq->biotail = bio; + blk_rq_bio_prep(q, rq, bio); + + rq->rq_disk = bdev->bd_contains->bd_disk; + + /* bio backed don't set data */ + rq->buffer = rq->data = NULL; + /* rq data_len used for pc cmd's request_bufflen */ + rq->data_len = bio->bi_size; + + rq->sense = h->sense; + memset(rq->sense, 0, SCSI_SENSE_BUFFERSIZE); + rq->sense_len = 0; + + memset(&rq->cmd, 0, BLK_MAX_CDB); + + rq->timeout = EMC_FAILOVER_TIMEOUT; + rq->flags |= (REQ_BLOCK_PC | REQ_FAILFAST | REQ_NOMERGE); + + return rq; +} + +static struct request *emc_trespass_get(struct emc_handler *h, + struct path *path) +{ + struct bio *bio; + struct request *rq; + unsigned char *page22; + unsigned char long_trespass_pg[] = { + 0, 0, 0, 0, + TRESPASS_PAGE, /* Page code */ + 0x09, /* Page length - 2 */ + h->hr ? 0x01 : 0x81, /* Trespass code + Honor reservation bit */ + 0xff, 0xff, /* Trespass target */ + 0, 0, 0, 0, 0, 0 /* Reserved bytes / unknown */ + }; + unsigned char short_trespass_pg[] = { + 0, 0, 0, 0, + TRESPASS_PAGE, /* Page code */ + 0x02, /* Page length - 2 */ + h->hr ? 0x01 : 0x81, /* Trespass code + Honor reservation bit */ + 0xff, /* Trespass target */ + }; + unsigned data_size = h->short_trespass ? sizeof(short_trespass_pg) : + sizeof(long_trespass_pg); + + /* get bio backing */ + if (data_size > PAGE_SIZE) + /* this should never happen */ + return NULL; + + bio = get_failover_bio(path, data_size); + if (!bio) { + DMERR("dm-emc: emc_trespass_get: no bio"); + return NULL; + } + + page22 = (unsigned char *)bio_data(bio); + memset(page22, 0, data_size); + + memcpy(page22, h->short_trespass ? + short_trespass_pg : long_trespass_pg, data_size); + + /* get request for block layer packet command */ + rq = get_failover_req(h, bio, path); + if (!rq) { + DMERR("dm-emc: emc_trespass_get: no rq"); + free_bio(bio); + return NULL; + } + + /* Prepare the command. */ + rq->cmd[0] = MODE_SELECT; + rq->cmd[1] = 0x10; + rq->cmd[4] = data_size; + rq->cmd_len = COMMAND_SIZE(rq->cmd[0]); + + return rq; +} + +static void emc_pg_init(struct hw_handler *hwh, unsigned bypassed, + struct path *path) +{ + struct request *rq; + struct request_queue *q = bdev_get_queue(path->dev->bdev); + + /* + * We can either blindly init the pg (then look at the sense), + * or we can send some commands to get the state here (then + * possibly send the fo cmnd), or we can also have the + * initial state passed into us and then get an update here. + */ + if (!q) { + DMINFO("dm-emc: emc_pg_init: no queue"); + goto fail_path; + } + + /* FIXME: The request should be pre-allocated. */ + rq = emc_trespass_get(hwh->context, path); + if (!rq) { + DMERR("dm-emc: emc_pg_init: no rq"); + goto fail_path; + } + + DMINFO("dm-emc: emc_pg_init: sending switch-over command"); + elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 1); + return; + +fail_path: + dm_pg_init_complete(path, MP_FAIL_PATH); +} + +static struct emc_handler *alloc_emc_handler(void) +{ + struct emc_handler *h = kmalloc(sizeof(*h), GFP_KERNEL); + + if (h) + spin_lock_init(&h->lock); + + return h; +} + +static int emc_create(struct hw_handler *hwh, unsigned argc, char **argv) +{ + struct emc_handler *h; + unsigned hr, short_trespass; + + if (argc == 0) { + /* No arguments: use defaults */ + hr = 0; + short_trespass = 0; + } else if (argc != 2) { + DMWARN("dm-emc hwhandler: incorrect number of arguments"); + return -EINVAL; + } else { + if ((sscanf(argv[0], "%u", &short_trespass) != 1) + || (short_trespass > 1)) { + DMWARN("dm-emc: invalid trespass mode selected"); + return -EINVAL; + } + + if ((sscanf(argv[1], "%u", &hr) != 1) + || (hr > 1)) { + DMWARN("dm-emc: invalid honor reservation flag selected"); + return -EINVAL; + } + } + + h = alloc_emc_handler(); + if (!h) + return -ENOMEM; + + memset(h, 0, sizeof(*h)); + + hwh->context = h; + + if ((h->short_trespass = short_trespass)) + DMWARN("dm-emc: short trespass command will be send"); + else + DMWARN("dm-emc: long trespass command will be send"); + + if ((h->hr = hr)) + DMWARN("dm-emc: honor reservation bit will be set"); + else + DMWARN("dm-emc: honor reservation bit will not be set (default)"); + + return 0; +} + +static void emc_destroy(struct hw_handler *hwh) +{ + struct emc_handler *h = (struct emc_handler *) hwh->context; + + kfree(h); + hwh->context = NULL; +} + +static unsigned emc_error(struct hw_handler *hwh, struct bio *bio) +{ + /* FIXME: Patch from axboe still missing */ +#if 0 + int sense; + + if (bio->bi_error & BIO_SENSE) { + sense = bio->bi_error & 0xffffff; /* sense key / asc / ascq */ + + if (sense == 0x020403) { + /* LUN Not Ready - Manual Intervention Required + * indicates this is a passive path. + * + * FIXME: However, if this is seen and EVPD C0 + * indicates that this is due to a NDU in + * progress, we should set FAIL_PATH too. + * This indicates we might have to do a SCSI + * inquiry in the end_io path. Ugh. */ + return MP_BYPASS_PG | MP_RETRY_IO; + } else if (sense == 0x052501) { + /* An array based copy is in progress. Do not + * fail the path, do not bypass to another PG, + * do not retry. Fail the IO immediately. + * (Actually this is the same conclusion as in + * the default handler, but lets make sure.) */ + return 0; + } else if (sense == 0x062900) { + /* Unit Attention Code. This is the first IO + * to the new path, so just retry. */ + return MP_RETRY_IO; + } + } +#endif + + /* Try default handler */ + return dm_scsi_err_handler(hwh, bio); +} + +static struct hw_handler_type emc_hwh = { + .name = "emc", + .module = THIS_MODULE, + .create = emc_create, + .destroy = emc_destroy, + .pg_init = emc_pg_init, + .error = emc_error, +}; + +static int __init dm_emc_init(void) +{ + int r = dm_register_hw_handler(&emc_hwh); + + if (r < 0) + DMERR("emc: register failed %d", r); + + DMINFO("dm-emc version 0.0.3 loaded"); + + return r; +} + +static void __exit dm_emc_exit(void) +{ + int r = dm_unregister_hw_handler(&emc_hwh); + + if (r < 0) + DMERR("emc: unregister failed %d", r); +} + +module_init(dm_emc_init); +module_exit(dm_emc_exit); + +MODULE_DESCRIPTION(DM_NAME " EMC CX/AX/FC-family multipath"); +MODULE_AUTHOR("Lars Marowsky-Bree <lmb@suse.de>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-exception-store.c b/drivers/md/dm-exception-store.c new file mode 100644 index 00000000000..17212b4201a --- /dev/null +++ b/drivers/md/dm-exception-store.c @@ -0,0 +1,648 @@ +/* + * dm-snapshot.c + * + * Copyright (C) 2001-2002 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + */ + +#include "dm.h" +#include "dm-snap.h" +#include "dm-io.h" +#include "kcopyd.h" + +#include <linux/mm.h> +#include <linux/pagemap.h> +#include <linux/vmalloc.h> +#include <linux/slab.h> + +/*----------------------------------------------------------------- + * Persistent snapshots, by persistent we mean that the snapshot + * will survive a reboot. + *---------------------------------------------------------------*/ + +/* + * We need to store a record of which parts of the origin have + * been copied to the snapshot device. The snapshot code + * requires that we copy exception chunks to chunk aligned areas + * of the COW store. It makes sense therefore, to store the + * metadata in chunk size blocks. + * + * There is no backward or forward compatibility implemented, + * snapshots with different disk versions than the kernel will + * not be usable. It is expected that "lvcreate" will blank out + * the start of a fresh COW device before calling the snapshot + * constructor. + * + * The first chunk of the COW device just contains the header. + * After this there is a chunk filled with exception metadata, + * followed by as many exception chunks as can fit in the + * metadata areas. + * + * All on disk structures are in little-endian format. The end + * of the exceptions info is indicated by an exception with a + * new_chunk of 0, which is invalid since it would point to the + * header chunk. + */ + +/* + * Magic for persistent snapshots: "SnAp" - Feeble isn't it. + */ +#define SNAP_MAGIC 0x70416e53 + +/* + * The on-disk version of the metadata. + */ +#define SNAPSHOT_DISK_VERSION 1 + +struct disk_header { + uint32_t magic; + + /* + * Is this snapshot valid. There is no way of recovering + * an invalid snapshot. + */ + uint32_t valid; + + /* + * Simple, incrementing version. no backward + * compatibility. + */ + uint32_t version; + + /* In sectors */ + uint32_t chunk_size; +}; + +struct disk_exception { + uint64_t old_chunk; + uint64_t new_chunk; +}; + +struct commit_callback { + void (*callback)(void *, int success); + void *context; +}; + +/* + * The top level structure for a persistent exception store. + */ +struct pstore { + struct dm_snapshot *snap; /* up pointer to my snapshot */ + int version; + int valid; + uint32_t chunk_size; + uint32_t exceptions_per_area; + + /* + * Now that we have an asynchronous kcopyd there is no + * need for large chunk sizes, so it wont hurt to have a + * whole chunks worth of metadata in memory at once. + */ + void *area; + + /* + * Used to keep track of which metadata area the data in + * 'chunk' refers to. + */ + uint32_t current_area; + + /* + * The next free chunk for an exception. + */ + uint32_t next_free; + + /* + * The index of next free exception in the current + * metadata area. + */ + uint32_t current_committed; + + atomic_t pending_count; + uint32_t callback_count; + struct commit_callback *callbacks; +}; + +static inline unsigned int sectors_to_pages(unsigned int sectors) +{ + return sectors / (PAGE_SIZE >> 9); +} + +static int alloc_area(struct pstore *ps) +{ + int r = -ENOMEM; + size_t len; + + len = ps->chunk_size << SECTOR_SHIFT; + + /* + * Allocate the chunk_size block of memory that will hold + * a single metadata area. + */ + ps->area = vmalloc(len); + if (!ps->area) + return r; + + return 0; +} + +static void free_area(struct pstore *ps) +{ + vfree(ps->area); +} + +/* + * Read or write a chunk aligned and sized block of data from a device. + */ +static int chunk_io(struct pstore *ps, uint32_t chunk, int rw) +{ + struct io_region where; + unsigned long bits; + + where.bdev = ps->snap->cow->bdev; + where.sector = ps->chunk_size * chunk; + where.count = ps->chunk_size; + + return dm_io_sync_vm(1, &where, rw, ps->area, &bits); +} + +/* + * Read or write a metadata area. Remembering to skip the first + * chunk which holds the header. + */ +static int area_io(struct pstore *ps, uint32_t area, int rw) +{ + int r; + uint32_t chunk; + + /* convert a metadata area index to a chunk index */ + chunk = 1 + ((ps->exceptions_per_area + 1) * area); + + r = chunk_io(ps, chunk, rw); + if (r) + return r; + + ps->current_area = area; + return 0; +} + +static int zero_area(struct pstore *ps, uint32_t area) +{ + memset(ps->area, 0, ps->chunk_size << SECTOR_SHIFT); + return area_io(ps, area, WRITE); +} + +static int read_header(struct pstore *ps, int *new_snapshot) +{ + int r; + struct disk_header *dh; + + r = chunk_io(ps, 0, READ); + if (r) + return r; + + dh = (struct disk_header *) ps->area; + + if (le32_to_cpu(dh->magic) == 0) { + *new_snapshot = 1; + + } else if (le32_to_cpu(dh->magic) == SNAP_MAGIC) { + *new_snapshot = 0; + ps->valid = le32_to_cpu(dh->valid); + ps->version = le32_to_cpu(dh->version); + ps->chunk_size = le32_to_cpu(dh->chunk_size); + + } else { + DMWARN("Invalid/corrupt snapshot"); + r = -ENXIO; + } + + return r; +} + +static int write_header(struct pstore *ps) +{ + struct disk_header *dh; + + memset(ps->area, 0, ps->chunk_size << SECTOR_SHIFT); + + dh = (struct disk_header *) ps->area; + dh->magic = cpu_to_le32(SNAP_MAGIC); + dh->valid = cpu_to_le32(ps->valid); + dh->version = cpu_to_le32(ps->version); + dh->chunk_size = cpu_to_le32(ps->chunk_size); + + return chunk_io(ps, 0, WRITE); +} + +/* + * Access functions for the disk exceptions, these do the endian conversions. + */ +static struct disk_exception *get_exception(struct pstore *ps, uint32_t index) +{ + if (index >= ps->exceptions_per_area) + return NULL; + + return ((struct disk_exception *) ps->area) + index; +} + +static int read_exception(struct pstore *ps, + uint32_t index, struct disk_exception *result) +{ + struct disk_exception *e; + + e = get_exception(ps, index); + if (!e) + return -EINVAL; + + /* copy it */ + result->old_chunk = le64_to_cpu(e->old_chunk); + result->new_chunk = le64_to_cpu(e->new_chunk); + + return 0; +} + +static int write_exception(struct pstore *ps, + uint32_t index, struct disk_exception *de) +{ + struct disk_exception *e; + + e = get_exception(ps, index); + if (!e) + return -EINVAL; + + /* copy it */ + e->old_chunk = cpu_to_le64(de->old_chunk); + e->new_chunk = cpu_to_le64(de->new_chunk); + + return 0; +} + +/* + * Registers the exceptions that are present in the current area. + * 'full' is filled in to indicate if the area has been + * filled. + */ +static int insert_exceptions(struct pstore *ps, int *full) +{ + int r; + unsigned int i; + struct disk_exception de; + + /* presume the area is full */ + *full = 1; + + for (i = 0; i < ps->exceptions_per_area; i++) { + r = read_exception(ps, i, &de); + + if (r) + return r; + + /* + * If the new_chunk is pointing at the start of + * the COW device, where the first metadata area + * is we know that we've hit the end of the + * exceptions. Therefore the area is not full. + */ + if (de.new_chunk == 0LL) { + ps->current_committed = i; + *full = 0; + break; + } + + /* + * Keep track of the start of the free chunks. + */ + if (ps->next_free <= de.new_chunk) + ps->next_free = de.new_chunk + 1; + + /* + * Otherwise we add the exception to the snapshot. + */ + r = dm_add_exception(ps->snap, de.old_chunk, de.new_chunk); + if (r) + return r; + } + + return 0; +} + +static int read_exceptions(struct pstore *ps) +{ + uint32_t area; + int r, full = 1; + + /* + * Keeping reading chunks and inserting exceptions until + * we find a partially full area. + */ + for (area = 0; full; area++) { + r = area_io(ps, area, READ); + if (r) + return r; + + r = insert_exceptions(ps, &full); + if (r) + return r; + } + + return 0; +} + +static inline struct pstore *get_info(struct exception_store *store) +{ + return (struct pstore *) store->context; +} + +static void persistent_fraction_full(struct exception_store *store, + sector_t *numerator, sector_t *denominator) +{ + *numerator = get_info(store)->next_free * store->snap->chunk_size; + *denominator = get_dev_size(store->snap->cow->bdev); +} + +static void persistent_destroy(struct exception_store *store) +{ + struct pstore *ps = get_info(store); + + dm_io_put(sectors_to_pages(ps->chunk_size)); + vfree(ps->callbacks); + free_area(ps); + kfree(ps); +} + +static int persistent_read_metadata(struct exception_store *store) +{ + int r, new_snapshot; + struct pstore *ps = get_info(store); + + /* + * Read the snapshot header. + */ + r = read_header(ps, &new_snapshot); + if (r) + return r; + + /* + * Do we need to setup a new snapshot ? + */ + if (new_snapshot) { + r = write_header(ps); + if (r) { + DMWARN("write_header failed"); + return r; + } + + r = zero_area(ps, 0); + if (r) { + DMWARN("zero_area(0) failed"); + return r; + } + + } else { + /* + * Sanity checks. + */ + if (!ps->valid) { + DMWARN("snapshot is marked invalid"); + return -EINVAL; + } + + if (ps->version != SNAPSHOT_DISK_VERSION) { + DMWARN("unable to handle snapshot disk version %d", + ps->version); + return -EINVAL; + } + + /* + * Read the metadata. + */ + r = read_exceptions(ps); + if (r) + return r; + } + + return 0; +} + +static int persistent_prepare(struct exception_store *store, + struct exception *e) +{ + struct pstore *ps = get_info(store); + uint32_t stride; + sector_t size = get_dev_size(store->snap->cow->bdev); + + /* Is there enough room ? */ + if (size < ((ps->next_free + 1) * store->snap->chunk_size)) + return -ENOSPC; + + e->new_chunk = ps->next_free; + + /* + * Move onto the next free pending, making sure to take + * into account the location of the metadata chunks. + */ + stride = (ps->exceptions_per_area + 1); + if ((++ps->next_free % stride) == 1) + ps->next_free++; + + atomic_inc(&ps->pending_count); + return 0; +} + +static void persistent_commit(struct exception_store *store, + struct exception *e, + void (*callback) (void *, int success), + void *callback_context) +{ + int r; + unsigned int i; + struct pstore *ps = get_info(store); + struct disk_exception de; + struct commit_callback *cb; + + de.old_chunk = e->old_chunk; + de.new_chunk = e->new_chunk; + write_exception(ps, ps->current_committed++, &de); + + /* + * Add the callback to the back of the array. This code + * is the only place where the callback array is + * manipulated, and we know that it will never be called + * multiple times concurrently. + */ + cb = ps->callbacks + ps->callback_count++; + cb->callback = callback; + cb->context = callback_context; + + /* + * If there are no more exceptions in flight, or we have + * filled this metadata area we commit the exceptions to + * disk. + */ + if (atomic_dec_and_test(&ps->pending_count) || + (ps->current_committed == ps->exceptions_per_area)) { + r = area_io(ps, ps->current_area, WRITE); + if (r) + ps->valid = 0; + + for (i = 0; i < ps->callback_count; i++) { + cb = ps->callbacks + i; + cb->callback(cb->context, r == 0 ? 1 : 0); + } + + ps->callback_count = 0; + } + + /* + * Have we completely filled the current area ? + */ + if (ps->current_committed == ps->exceptions_per_area) { + ps->current_committed = 0; + r = zero_area(ps, ps->current_area + 1); + if (r) + ps->valid = 0; + } +} + +static void persistent_drop(struct exception_store *store) +{ + struct pstore *ps = get_info(store); + + ps->valid = 0; + if (write_header(ps)) + DMWARN("write header failed"); +} + +int dm_create_persistent(struct exception_store *store, uint32_t chunk_size) +{ + int r; + struct pstore *ps; + + r = dm_io_get(sectors_to_pages(chunk_size)); + if (r) + return r; + + /* allocate the pstore */ + ps = kmalloc(sizeof(*ps), GFP_KERNEL); + if (!ps) { + r = -ENOMEM; + goto bad; + } + + ps->snap = store->snap; + ps->valid = 1; + ps->version = SNAPSHOT_DISK_VERSION; + ps->chunk_size = chunk_size; + ps->exceptions_per_area = (chunk_size << SECTOR_SHIFT) / + sizeof(struct disk_exception); + ps->next_free = 2; /* skipping the header and first area */ + ps->current_committed = 0; + + r = alloc_area(ps); + if (r) + goto bad; + + /* + * Allocate space for all the callbacks. + */ + ps->callback_count = 0; + atomic_set(&ps->pending_count, 0); + ps->callbacks = dm_vcalloc(ps->exceptions_per_area, + sizeof(*ps->callbacks)); + + if (!ps->callbacks) { + r = -ENOMEM; + goto bad; + } + + store->destroy = persistent_destroy; + store->read_metadata = persistent_read_metadata; + store->prepare_exception = persistent_prepare; + store->commit_exception = persistent_commit; + store->drop_snapshot = persistent_drop; + store->fraction_full = persistent_fraction_full; + store->context = ps; + + return 0; + + bad: + dm_io_put(sectors_to_pages(chunk_size)); + if (ps) { + if (ps->area) + free_area(ps); + + kfree(ps); + } + return r; +} + +/*----------------------------------------------------------------- + * Implementation of the store for non-persistent snapshots. + *---------------------------------------------------------------*/ +struct transient_c { + sector_t next_free; +}; + +static void transient_destroy(struct exception_store *store) +{ + kfree(store->context); +} + +static int transient_read_metadata(struct exception_store *store) +{ + return 0; +} + +static int transient_prepare(struct exception_store *store, struct exception *e) +{ + struct transient_c *tc = (struct transient_c *) store->context; + sector_t size = get_dev_size(store->snap->cow->bdev); + + if (size < (tc->next_free + store->snap->chunk_size)) + return -1; + + e->new_chunk = sector_to_chunk(store->snap, tc->next_free); + tc->next_free += store->snap->chunk_size; + + return 0; +} + +static void transient_commit(struct exception_store *store, + struct exception *e, + void (*callback) (void *, int success), + void *callback_context) +{ + /* Just succeed */ + callback(callback_context, 1); +} + +static void transient_fraction_full(struct exception_store *store, + sector_t *numerator, sector_t *denominator) +{ + *numerator = ((struct transient_c *) store->context)->next_free; + *denominator = get_dev_size(store->snap->cow->bdev); +} + +int dm_create_transient(struct exception_store *store, + struct dm_snapshot *s, int blocksize) +{ + struct transient_c *tc; + + memset(store, 0, sizeof(*store)); + store->destroy = transient_destroy; + store->read_metadata = transient_read_metadata; + store->prepare_exception = transient_prepare; + store->commit_exception = transient_commit; + store->fraction_full = transient_fraction_full; + store->snap = s; + + tc = kmalloc(sizeof(struct transient_c), GFP_KERNEL); + if (!tc) + return -ENOMEM; + + tc->next_free = 0; + store->context = tc; + + return 0; +} diff --git a/drivers/md/dm-hw-handler.c b/drivers/md/dm-hw-handler.c new file mode 100644 index 00000000000..ae63772e44c --- /dev/null +++ b/drivers/md/dm-hw-handler.c @@ -0,0 +1,216 @@ +/* + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + * + * Multipath hardware handler registration. + */ + +#include "dm.h" +#include "dm-hw-handler.h" + +#include <linux/slab.h> + +struct hwh_internal { + struct hw_handler_type hwht; + + struct list_head list; + long use; +}; + +#define hwht_to_hwhi(__hwht) container_of((__hwht), struct hwh_internal, hwht) + +static LIST_HEAD(_hw_handlers); +static DECLARE_RWSEM(_hwh_lock); + +struct hwh_internal *__find_hw_handler_type(const char *name) +{ + struct hwh_internal *hwhi; + + list_for_each_entry(hwhi, &_hw_handlers, list) { + if (!strcmp(name, hwhi->hwht.name)) + return hwhi; + } + + return NULL; +} + +static struct hwh_internal *get_hw_handler(const char *name) +{ + struct hwh_internal *hwhi; + + down_read(&_hwh_lock); + hwhi = __find_hw_handler_type(name); + if (hwhi) { + if ((hwhi->use == 0) && !try_module_get(hwhi->hwht.module)) + hwhi = NULL; + else + hwhi->use++; + } + up_read(&_hwh_lock); + + return hwhi; +} + +struct hw_handler_type *dm_get_hw_handler(const char *name) +{ + struct hwh_internal *hwhi; + + if (!name) + return NULL; + + hwhi = get_hw_handler(name); + if (!hwhi) { + request_module("dm-%s", name); + hwhi = get_hw_handler(name); + } + + return hwhi ? &hwhi->hwht : NULL; +} + +void dm_put_hw_handler(struct hw_handler_type *hwht) +{ + struct hwh_internal *hwhi; + + if (!hwht) + return; + + down_read(&_hwh_lock); + hwhi = __find_hw_handler_type(hwht->name); + if (!hwhi) + goto out; + + if (--hwhi->use == 0) + module_put(hwhi->hwht.module); + + if (hwhi->use < 0) + BUG(); + + out: + up_read(&_hwh_lock); +} + +static struct hwh_internal *_alloc_hw_handler(struct hw_handler_type *hwht) +{ + struct hwh_internal *hwhi = kmalloc(sizeof(*hwhi), GFP_KERNEL); + + if (hwhi) { + memset(hwhi, 0, sizeof(*hwhi)); + hwhi->hwht = *hwht; + } + + return hwhi; +} + +int dm_register_hw_handler(struct hw_handler_type *hwht) +{ + int r = 0; + struct hwh_internal *hwhi = _alloc_hw_handler(hwht); + + if (!hwhi) + return -ENOMEM; + + down_write(&_hwh_lock); + + if (__find_hw_handler_type(hwht->name)) { + kfree(hwhi); + r = -EEXIST; + } else + list_add(&hwhi->list, &_hw_handlers); + + up_write(&_hwh_lock); + + return r; +} + +int dm_unregister_hw_handler(struct hw_handler_type *hwht) +{ + struct hwh_internal *hwhi; + + down_write(&_hwh_lock); + + hwhi = __find_hw_handler_type(hwht->name); + if (!hwhi) { + up_write(&_hwh_lock); + return -EINVAL; + } + + if (hwhi->use) { + up_write(&_hwh_lock); + return -ETXTBSY; + } + + list_del(&hwhi->list); + + up_write(&_hwh_lock); + + kfree(hwhi); + + return 0; +} + +unsigned dm_scsi_err_handler(struct hw_handler *hwh, struct bio *bio) +{ +#if 0 + int sense_key, asc, ascq; + + if (bio->bi_error & BIO_SENSE) { + /* FIXME: This is just an initial guess. */ + /* key / asc / ascq */ + sense_key = (bio->bi_error >> 16) & 0xff; + asc = (bio->bi_error >> 8) & 0xff; + ascq = bio->bi_error & 0xff; + + switch (sense_key) { + /* This block as a whole comes from the device. + * So no point retrying on another path. */ + case 0x03: /* Medium error */ + case 0x05: /* Illegal request */ + case 0x07: /* Data protect */ + case 0x08: /* Blank check */ + case 0x0a: /* copy aborted */ + case 0x0c: /* obsolete - no clue ;-) */ + case 0x0d: /* volume overflow */ + case 0x0e: /* data miscompare */ + case 0x0f: /* reserved - no idea either. */ + return MP_ERROR_IO; + + /* For these errors it's unclear whether they + * come from the device or the controller. + * So just lets try a different path, and if + * it eventually succeeds, user-space will clear + * the paths again... */ + case 0x02: /* Not ready */ + case 0x04: /* Hardware error */ + case 0x09: /* vendor specific */ + case 0x0b: /* Aborted command */ + return MP_FAIL_PATH; + + case 0x06: /* Unit attention - might want to decode */ + if (asc == 0x04 && ascq == 0x01) + /* "Unit in the process of + * becoming ready" */ + return 0; + return MP_FAIL_PATH; + + /* FIXME: For Unit Not Ready we may want + * to have a generic pg activation + * feature (START_UNIT). */ + + /* Should these two ever end up in the + * error path? I don't think so. */ + case 0x00: /* No sense */ + case 0x01: /* Recovered error */ + return 0; + } + } +#endif + + /* We got no idea how to decode the other kinds of errors -> + * assume generic error condition. */ + return MP_FAIL_PATH; +} + +EXPORT_SYMBOL_GPL(dm_register_hw_handler); +EXPORT_SYMBOL_GPL(dm_unregister_hw_handler); +EXPORT_SYMBOL_GPL(dm_scsi_err_handler); diff --git a/drivers/md/dm-hw-handler.h b/drivers/md/dm-hw-handler.h new file mode 100644 index 00000000000..15f5629e231 --- /dev/null +++ b/drivers/md/dm-hw-handler.h @@ -0,0 +1,61 @@ +/* + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + * + * Multipath hardware handler registration. + */ + +#ifndef DM_HW_HANDLER_H +#define DM_HW_HANDLER_H + +#include <linux/device-mapper.h> + +#include "dm-mpath.h" + +struct hw_handler_type; +struct hw_handler { + struct hw_handler_type *type; + void *context; +}; + +/* + * Constructs a hardware handler object, takes custom arguments + */ +/* Information about a hardware handler type */ +struct hw_handler_type { + char *name; + struct module *module; + + int (*create) (struct hw_handler *handler, unsigned int argc, + char **argv); + void (*destroy) (struct hw_handler *hwh); + + void (*pg_init) (struct hw_handler *hwh, unsigned bypassed, + struct path *path); + unsigned (*error) (struct hw_handler *hwh, struct bio *bio); + int (*status) (struct hw_handler *hwh, status_type_t type, + char *result, unsigned int maxlen); +}; + +/* Register a hardware handler */ +int dm_register_hw_handler(struct hw_handler_type *type); + +/* Unregister a hardware handler */ +int dm_unregister_hw_handler(struct hw_handler_type *type); + +/* Returns a registered hardware handler type */ +struct hw_handler_type *dm_get_hw_handler(const char *name); + +/* Releases a hardware handler */ +void dm_put_hw_handler(struct hw_handler_type *hwht); + +/* Default err function */ +unsigned dm_scsi_err_handler(struct hw_handler *hwh, struct bio *bio); + +/* Error flags for err and dm_pg_init_complete */ +#define MP_FAIL_PATH 1 +#define MP_BYPASS_PG 2 +#define MP_ERROR_IO 4 /* Don't retry this I/O */ + +#endif diff --git a/drivers/md/dm-io.c b/drivers/md/dm-io.c new file mode 100644 index 00000000000..45754bb6a79 --- /dev/null +++ b/drivers/md/dm-io.c @@ -0,0 +1,426 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the GPL. + */ + +#include "dm-io.h" + +#include <linux/bio.h> +#include <linux/mempool.h> +#include <linux/module.h> +#include <linux/sched.h> +#include <linux/slab.h> + +static struct bio_set *_bios; + +/* FIXME: can we shrink this ? */ +struct io { + unsigned long error; + atomic_t count; + struct task_struct *sleeper; + io_notify_fn callback; + void *context; +}; + +/* + * io contexts are only dynamically allocated for asynchronous + * io. Since async io is likely to be the majority of io we'll + * have the same number of io contexts as buffer heads ! (FIXME: + * must reduce this). + */ +static unsigned _num_ios; +static mempool_t *_io_pool; + +static void *alloc_io(unsigned int __nocast gfp_mask, void *pool_data) +{ + return kmalloc(sizeof(struct io), gfp_mask); +} + +static void free_io(void *element, void *pool_data) +{ + kfree(element); +} + +static unsigned int pages_to_ios(unsigned int pages) +{ + return 4 * pages; /* too many ? */ +} + +static int resize_pool(unsigned int new_ios) +{ + int r = 0; + + if (_io_pool) { + if (new_ios == 0) { + /* free off the pool */ + mempool_destroy(_io_pool); + _io_pool = NULL; + bioset_free(_bios); + + } else { + /* resize the pool */ + r = mempool_resize(_io_pool, new_ios, GFP_KERNEL); + } + + } else { + /* create new pool */ + _io_pool = mempool_create(new_ios, alloc_io, free_io, NULL); + if (!_io_pool) + return -ENOMEM; + + _bios = bioset_create(16, 16, 4); + if (!_bios) { + mempool_destroy(_io_pool); + _io_pool = NULL; + return -ENOMEM; + } + } + + if (!r) + _num_ios = new_ios; + + return r; +} + +int dm_io_get(unsigned int num_pages) +{ + return resize_pool(_num_ios + pages_to_ios(num_pages)); +} + +void dm_io_put(unsigned int num_pages) +{ + resize_pool(_num_ios - pages_to_ios(num_pages)); +} + +/*----------------------------------------------------------------- + * We need to keep track of which region a bio is doing io for. + * In order to save a memory allocation we store this the last + * bvec which we know is unused (blech). + * XXX This is ugly and can OOPS with some configs... find another way. + *---------------------------------------------------------------*/ +static inline void bio_set_region(struct bio *bio, unsigned region) +{ + bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len = region; +} + +static inline unsigned bio_get_region(struct bio *bio) +{ + return bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len; +} + +/*----------------------------------------------------------------- + * We need an io object to keep track of the number of bios that + * have been dispatched for a particular io. + *---------------------------------------------------------------*/ +static void dec_count(struct io *io, unsigned int region, int error) +{ + if (error) + set_bit(region, &io->error); + + if (atomic_dec_and_test(&io->count)) { + if (io->sleeper) + wake_up_process(io->sleeper); + + else { + int r = io->error; + io_notify_fn fn = io->callback; + void *context = io->context; + + mempool_free(io, _io_pool); + fn(r, context); + } + } +} + +static int endio(struct bio *bio, unsigned int done, int error) +{ + struct io *io = (struct io *) bio->bi_private; + + /* keep going until we've finished */ + if (bio->bi_size) + return 1; + + if (error && bio_data_dir(bio) == READ) + zero_fill_bio(bio); + + dec_count(io, bio_get_region(bio), error); + bio_put(bio); + + return 0; +} + +/*----------------------------------------------------------------- + * These little objects provide an abstraction for getting a new + * destination page for io. + *---------------------------------------------------------------*/ +struct dpages { + void (*get_page)(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset); + void (*next_page)(struct dpages *dp); + + unsigned context_u; + void *context_ptr; +}; + +/* + * Functions for getting the pages from a list. + */ +static void list_get_page(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset) +{ + unsigned o = dp->context_u; + struct page_list *pl = (struct page_list *) dp->context_ptr; + + *p = pl->page; + *len = PAGE_SIZE - o; + *offset = o; +} + +static void list_next_page(struct dpages *dp) +{ + struct page_list *pl = (struct page_list *) dp->context_ptr; + dp->context_ptr = pl->next; + dp->context_u = 0; +} + +static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset) +{ + dp->get_page = list_get_page; + dp->next_page = list_next_page; + dp->context_u = offset; + dp->context_ptr = pl; +} + +/* + * Functions for getting the pages from a bvec. + */ +static void bvec_get_page(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset) +{ + struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; + *p = bvec->bv_page; + *len = bvec->bv_len; + *offset = bvec->bv_offset; +} + +static void bvec_next_page(struct dpages *dp) +{ + struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; + dp->context_ptr = bvec + 1; +} + +static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec) +{ + dp->get_page = bvec_get_page; + dp->next_page = bvec_next_page; + dp->context_ptr = bvec; +} + +static void vm_get_page(struct dpages *dp, + struct page **p, unsigned long *len, unsigned *offset) +{ + *p = vmalloc_to_page(dp->context_ptr); + *offset = dp->context_u; + *len = PAGE_SIZE - dp->context_u; +} + +static void vm_next_page(struct dpages *dp) +{ + dp->context_ptr += PAGE_SIZE - dp->context_u; + dp->context_u = 0; +} + +static void vm_dp_init(struct dpages *dp, void *data) +{ + dp->get_page = vm_get_page; + dp->next_page = vm_next_page; + dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1); + dp->context_ptr = data; +} + +/*----------------------------------------------------------------- + * IO routines that accept a list of pages. + *---------------------------------------------------------------*/ +static void do_region(int rw, unsigned int region, struct io_region *where, + struct dpages *dp, struct io *io) +{ + struct bio *bio; + struct page *page; + unsigned long len; + unsigned offset; + unsigned num_bvecs; + sector_t remaining = where->count; + + while (remaining) { + /* + * Allocate a suitably sized bio, we add an extra + * bvec for bio_get/set_region(). + */ + num_bvecs = (remaining / (PAGE_SIZE >> 9)) + 2; + bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, _bios); + bio->bi_sector = where->sector + (where->count - remaining); + bio->bi_bdev = where->bdev; + bio->bi_end_io = endio; + bio->bi_private = io; + bio_set_region(bio, region); + + /* + * Try and add as many pages as possible. + */ + while (remaining) { + dp->get_page(dp, &page, &len, &offset); + len = min(len, to_bytes(remaining)); + if (!bio_add_page(bio, page, len, offset)) + break; + + offset = 0; + remaining -= to_sector(len); + dp->next_page(dp); + } + + atomic_inc(&io->count); + submit_bio(rw, bio); + } +} + +static void dispatch_io(int rw, unsigned int num_regions, + struct io_region *where, struct dpages *dp, + struct io *io, int sync) +{ + int i; + struct dpages old_pages = *dp; + + if (sync) + rw |= (1 << BIO_RW_SYNC); + + /* + * For multiple regions we need to be careful to rewind + * the dp object for each call to do_region. + */ + for (i = 0; i < num_regions; i++) { + *dp = old_pages; + if (where[i].count) + do_region(rw, i, where + i, dp, io); + } + + /* + * Drop the extra refence that we were holding to avoid + * the io being completed too early. + */ + dec_count(io, 0, 0); +} + +static int sync_io(unsigned int num_regions, struct io_region *where, + int rw, struct dpages *dp, unsigned long *error_bits) +{ + struct io io; + + if (num_regions > 1 && rw != WRITE) { + WARN_ON(1); + return -EIO; + } + + io.error = 0; + atomic_set(&io.count, 1); /* see dispatch_io() */ + io.sleeper = current; + + dispatch_io(rw, num_regions, where, dp, &io, 1); + + while (1) { + set_current_state(TASK_UNINTERRUPTIBLE); + + if (!atomic_read(&io.count) || signal_pending(current)) + break; + + io_schedule(); + } + set_current_state(TASK_RUNNING); + + if (atomic_read(&io.count)) + return -EINTR; + + *error_bits = io.error; + return io.error ? -EIO : 0; +} + +static int async_io(unsigned int num_regions, struct io_region *where, int rw, + struct dpages *dp, io_notify_fn fn, void *context) +{ + struct io *io; + + if (num_regions > 1 && rw != WRITE) { + WARN_ON(1); + fn(1, context); + return -EIO; + } + + io = mempool_alloc(_io_pool, GFP_NOIO); + io->error = 0; + atomic_set(&io->count, 1); /* see dispatch_io() */ + io->sleeper = NULL; + io->callback = fn; + io->context = context; + + dispatch_io(rw, num_regions, where, dp, io, 0); + return 0; +} + +int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw, + struct page_list *pl, unsigned int offset, + unsigned long *error_bits) +{ + struct dpages dp; + list_dp_init(&dp, pl, offset); + return sync_io(num_regions, where, rw, &dp, error_bits); +} + +int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, unsigned long *error_bits) +{ + struct dpages dp; + bvec_dp_init(&dp, bvec); + return sync_io(num_regions, where, rw, &dp, error_bits); +} + +int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw, + void *data, unsigned long *error_bits) +{ + struct dpages dp; + vm_dp_init(&dp, data); + return sync_io(num_regions, where, rw, &dp, error_bits); +} + +int dm_io_async(unsigned int num_regions, struct io_region *where, int rw, + struct page_list *pl, unsigned int offset, + io_notify_fn fn, void *context) +{ + struct dpages dp; + list_dp_init(&dp, pl, offset); + return async_io(num_regions, where, rw, &dp, fn, context); +} + +int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, io_notify_fn fn, void *context) +{ + struct dpages dp; + bvec_dp_init(&dp, bvec); + return async_io(num_regions, where, rw, &dp, fn, context); +} + +int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw, + void *data, io_notify_fn fn, void *context) +{ + struct dpages dp; + vm_dp_init(&dp, data); + return async_io(num_regions, where, rw, &dp, fn, context); +} + +EXPORT_SYMBOL(dm_io_get); +EXPORT_SYMBOL(dm_io_put); +EXPORT_SYMBOL(dm_io_sync); +EXPORT_SYMBOL(dm_io_async); +EXPORT_SYMBOL(dm_io_sync_bvec); +EXPORT_SYMBOL(dm_io_async_bvec); +EXPORT_SYMBOL(dm_io_sync_vm); +EXPORT_SYMBOL(dm_io_async_vm); diff --git a/drivers/md/dm-io.h b/drivers/md/dm-io.h new file mode 100644 index 00000000000..1a77f326570 --- /dev/null +++ b/drivers/md/dm-io.h @@ -0,0 +1,77 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the GPL. + */ + +#ifndef _DM_IO_H +#define _DM_IO_H + +#include "dm.h" + +/* FIXME make this configurable */ +#define DM_MAX_IO_REGIONS 8 + +struct io_region { + struct block_device *bdev; + sector_t sector; + sector_t count; +}; + +struct page_list { + struct page_list *next; + struct page *page; +}; + + +/* + * 'error' is a bitset, with each bit indicating whether an error + * occurred doing io to the corresponding region. + */ +typedef void (*io_notify_fn)(unsigned long error, void *context); + + +/* + * Before anyone uses the IO interface they should call + * dm_io_get(), specifying roughly how many pages they are + * expecting to perform io on concurrently. + * + * This function may block. + */ +int dm_io_get(unsigned int num_pages); +void dm_io_put(unsigned int num_pages); + +/* + * Synchronous IO. + * + * Please ensure that the rw flag in the next two functions is + * either READ or WRITE, ie. we don't take READA. Any + * regions with a zero count field will be ignored. + */ +int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw, + struct page_list *pl, unsigned int offset, + unsigned long *error_bits); + +int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, unsigned long *error_bits); + +int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw, + void *data, unsigned long *error_bits); + +/* + * Aynchronous IO. + * + * The 'where' array may be safely allocated on the stack since + * the function takes a copy. + */ +int dm_io_async(unsigned int num_regions, struct io_region *where, int rw, + struct page_list *pl, unsigned int offset, + io_notify_fn fn, void *context); + +int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw, + struct bio_vec *bvec, io_notify_fn fn, void *context); + +int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw, + void *data, io_notify_fn fn, void *context); + +#endif diff --git a/drivers/md/dm-ioctl.c b/drivers/md/dm-ioctl.c new file mode 100644 index 00000000000..ee3c869d970 --- /dev/null +++ b/drivers/md/dm-ioctl.c @@ -0,0 +1,1416 @@ +/* + * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. + * Copyright (C) 2004 - 2005 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/vmalloc.h> +#include <linux/miscdevice.h> +#include <linux/init.h> +#include <linux/wait.h> +#include <linux/slab.h> +#include <linux/devfs_fs_kernel.h> +#include <linux/dm-ioctl.h> + +#include <asm/uaccess.h> + +#define DM_DRIVER_EMAIL "dm-devel@redhat.com" + +/*----------------------------------------------------------------- + * The ioctl interface needs to be able to look up devices by + * name or uuid. + *---------------------------------------------------------------*/ +struct hash_cell { + struct list_head name_list; + struct list_head uuid_list; + + char *name; + char *uuid; + struct mapped_device *md; + struct dm_table *new_map; +}; + +struct vers_iter { + size_t param_size; + struct dm_target_versions *vers, *old_vers; + char *end; + uint32_t flags; +}; + + +#define NUM_BUCKETS 64 +#define MASK_BUCKETS (NUM_BUCKETS - 1) +static struct list_head _name_buckets[NUM_BUCKETS]; +static struct list_head _uuid_buckets[NUM_BUCKETS]; + +static void dm_hash_remove_all(void); + +/* + * Guards access to both hash tables. + */ +static DECLARE_RWSEM(_hash_lock); + +static void init_buckets(struct list_head *buckets) +{ + unsigned int i; + + for (i = 0; i < NUM_BUCKETS; i++) + INIT_LIST_HEAD(buckets + i); +} + +static int dm_hash_init(void) +{ + init_buckets(_name_buckets); + init_buckets(_uuid_buckets); + devfs_mk_dir(DM_DIR); + return 0; +} + +static void dm_hash_exit(void) +{ + dm_hash_remove_all(); + devfs_remove(DM_DIR); +} + +/*----------------------------------------------------------------- + * Hash function: + * We're not really concerned with the str hash function being + * fast since it's only used by the ioctl interface. + *---------------------------------------------------------------*/ +static unsigned int hash_str(const char *str) +{ + const unsigned int hash_mult = 2654435387U; + unsigned int h = 0; + + while (*str) + h = (h + (unsigned int) *str++) * hash_mult; + + return h & MASK_BUCKETS; +} + +/*----------------------------------------------------------------- + * Code for looking up a device by name + *---------------------------------------------------------------*/ +static struct hash_cell *__get_name_cell(const char *str) +{ + struct hash_cell *hc; + unsigned int h = hash_str(str); + + list_for_each_entry (hc, _name_buckets + h, name_list) + if (!strcmp(hc->name, str)) + return hc; + + return NULL; +} + +static struct hash_cell *__get_uuid_cell(const char *str) +{ + struct hash_cell *hc; + unsigned int h = hash_str(str); + + list_for_each_entry (hc, _uuid_buckets + h, uuid_list) + if (!strcmp(hc->uuid, str)) + return hc; + + return NULL; +} + +/*----------------------------------------------------------------- + * Inserting, removing and renaming a device. + *---------------------------------------------------------------*/ +static inline char *kstrdup(const char *str) +{ + char *r = kmalloc(strlen(str) + 1, GFP_KERNEL); + if (r) + strcpy(r, str); + return r; +} + +static struct hash_cell *alloc_cell(const char *name, const char *uuid, + struct mapped_device *md) +{ + struct hash_cell *hc; + + hc = kmalloc(sizeof(*hc), GFP_KERNEL); + if (!hc) + return NULL; + + hc->name = kstrdup(name); + if (!hc->name) { + kfree(hc); + return NULL; + } + + if (!uuid) + hc->uuid = NULL; + + else { + hc->uuid = kstrdup(uuid); + if (!hc->uuid) { + kfree(hc->name); + kfree(hc); + return NULL; + } + } + + INIT_LIST_HEAD(&hc->name_list); + INIT_LIST_HEAD(&hc->uuid_list); + hc->md = md; + hc->new_map = NULL; + return hc; +} + +static void free_cell(struct hash_cell *hc) +{ + if (hc) { + kfree(hc->name); + kfree(hc->uuid); + kfree(hc); + } +} + +/* + * devfs stuff. + */ +static int register_with_devfs(struct hash_cell *hc) +{ + struct gendisk *disk = dm_disk(hc->md); + + devfs_mk_bdev(MKDEV(disk->major, disk->first_minor), + S_IFBLK | S_IRUSR | S_IWUSR | S_IRGRP, + DM_DIR "/%s", hc->name); + return 0; +} + +static int unregister_with_devfs(struct hash_cell *hc) +{ + devfs_remove(DM_DIR"/%s", hc->name); + return 0; +} + +/* + * The kdev_t and uuid of a device can never change once it is + * initially inserted. + */ +static int dm_hash_insert(const char *name, const char *uuid, struct mapped_device *md) +{ + struct hash_cell *cell; + + /* + * Allocate the new cells. + */ + cell = alloc_cell(name, uuid, md); + if (!cell) + return -ENOMEM; + + /* + * Insert the cell into both hash tables. + */ + down_write(&_hash_lock); + if (__get_name_cell(name)) + goto bad; + + list_add(&cell->name_list, _name_buckets + hash_str(name)); + + if (uuid) { + if (__get_uuid_cell(uuid)) { + list_del(&cell->name_list); + goto bad; + } + list_add(&cell->uuid_list, _uuid_buckets + hash_str(uuid)); + } + register_with_devfs(cell); + dm_get(md); + dm_set_mdptr(md, cell); + up_write(&_hash_lock); + + return 0; + + bad: + up_write(&_hash_lock); + free_cell(cell); + return -EBUSY; +} + +static void __hash_remove(struct hash_cell *hc) +{ + /* remove from the dev hash */ + list_del(&hc->uuid_list); + list_del(&hc->name_list); + unregister_with_devfs(hc); + dm_set_mdptr(hc->md, NULL); + dm_put(hc->md); + if (hc->new_map) + dm_table_put(hc->new_map); + free_cell(hc); +} + +static void dm_hash_remove_all(void) +{ + int i; + struct hash_cell *hc; + struct list_head *tmp, *n; + + down_write(&_hash_lock); + for (i = 0; i < NUM_BUCKETS; i++) { + list_for_each_safe (tmp, n, _name_buckets + i) { + hc = list_entry(tmp, struct hash_cell, name_list); + __hash_remove(hc); + } + } + up_write(&_hash_lock); +} + +static int dm_hash_rename(const char *old, const char *new) +{ + char *new_name, *old_name; + struct hash_cell *hc; + + /* + * duplicate new. + */ + new_name = kstrdup(new); + if (!new_name) + return -ENOMEM; + + down_write(&_hash_lock); + + /* + * Is new free ? + */ + hc = __get_name_cell(new); + if (hc) { + DMWARN("asked to rename to an already existing name %s -> %s", + old, new); + up_write(&_hash_lock); + kfree(new_name); + return -EBUSY; + } + + /* + * Is there such a device as 'old' ? + */ + hc = __get_name_cell(old); + if (!hc) { + DMWARN("asked to rename a non existent device %s -> %s", + old, new); + up_write(&_hash_lock); + kfree(new_name); + return -ENXIO; + } + + /* + * rename and move the name cell. + */ + unregister_with_devfs(hc); + + list_del(&hc->name_list); + old_name = hc->name; + hc->name = new_name; + list_add(&hc->name_list, _name_buckets + hash_str(new_name)); + + /* rename the device node in devfs */ + register_with_devfs(hc); + + up_write(&_hash_lock); + kfree(old_name); + return 0; +} + +/*----------------------------------------------------------------- + * Implementation of the ioctl commands + *---------------------------------------------------------------*/ +/* + * All the ioctl commands get dispatched to functions with this + * prototype. + */ +typedef int (*ioctl_fn)(struct dm_ioctl *param, size_t param_size); + +static int remove_all(struct dm_ioctl *param, size_t param_size) +{ + dm_hash_remove_all(); + param->data_size = 0; + return 0; +} + +/* + * Round up the ptr to an 8-byte boundary. + */ +#define ALIGN_MASK 7 +static inline void *align_ptr(void *ptr) +{ + return (void *) (((size_t) (ptr + ALIGN_MASK)) & ~ALIGN_MASK); +} + +/* + * Retrieves the data payload buffer from an already allocated + * struct dm_ioctl. + */ +static void *get_result_buffer(struct dm_ioctl *param, size_t param_size, + size_t *len) +{ + param->data_start = align_ptr(param + 1) - (void *) param; + + if (param->data_start < param_size) + *len = param_size - param->data_start; + else + *len = 0; + + return ((void *) param) + param->data_start; +} + +static int list_devices(struct dm_ioctl *param, size_t param_size) +{ + unsigned int i; + struct hash_cell *hc; + size_t len, needed = 0; + struct gendisk *disk; + struct dm_name_list *nl, *old_nl = NULL; + + down_write(&_hash_lock); + + /* + * Loop through all the devices working out how much + * space we need. + */ + for (i = 0; i < NUM_BUCKETS; i++) { + list_for_each_entry (hc, _name_buckets + i, name_list) { + needed += sizeof(struct dm_name_list); + needed += strlen(hc->name) + 1; + needed += ALIGN_MASK; + } + } + + /* + * Grab our output buffer. + */ + nl = get_result_buffer(param, param_size, &len); + if (len < needed) { + param->flags |= DM_BUFFER_FULL_FLAG; + goto out; + } + param->data_size = param->data_start + needed; + + nl->dev = 0; /* Flags no data */ + + /* + * Now loop through filling out the names. + */ + for (i = 0; i < NUM_BUCKETS; i++) { + list_for_each_entry (hc, _name_buckets + i, name_list) { + if (old_nl) + old_nl->next = (uint32_t) ((void *) nl - + (void *) old_nl); + disk = dm_disk(hc->md); + nl->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor)); + nl->next = 0; + strcpy(nl->name, hc->name); + + old_nl = nl; + nl = align_ptr(((void *) ++nl) + strlen(hc->name) + 1); + } + } + + out: + up_write(&_hash_lock); + return 0; +} + +static void list_version_get_needed(struct target_type *tt, void *needed_param) +{ + size_t *needed = needed_param; + + *needed += strlen(tt->name); + *needed += sizeof(tt->version); + *needed += ALIGN_MASK; +} + +static void list_version_get_info(struct target_type *tt, void *param) +{ + struct vers_iter *info = param; + + /* Check space - it might have changed since the first iteration */ + if ((char *)info->vers + sizeof(tt->version) + strlen(tt->name) + 1 > + info->end) { + + info->flags = DM_BUFFER_FULL_FLAG; + return; + } + + if (info->old_vers) + info->old_vers->next = (uint32_t) ((void *)info->vers - + (void *)info->old_vers); + info->vers->version[0] = tt->version[0]; + info->vers->version[1] = tt->version[1]; + info->vers->version[2] = tt->version[2]; + info->vers->next = 0; + strcpy(info->vers->name, tt->name); + + info->old_vers = info->vers; + info->vers = align_ptr(((void *) ++info->vers) + strlen(tt->name) + 1); +} + +static int list_versions(struct dm_ioctl *param, size_t param_size) +{ + size_t len, needed = 0; + struct dm_target_versions *vers; + struct vers_iter iter_info; + + /* + * Loop through all the devices working out how much + * space we need. + */ + dm_target_iterate(list_version_get_needed, &needed); + + /* + * Grab our output buffer. + */ + vers = get_result_buffer(param, param_size, &len); + if (len < needed) { + param->flags |= DM_BUFFER_FULL_FLAG; + goto out; + } + param->data_size = param->data_start + needed; + + iter_info.param_size = param_size; + iter_info.old_vers = NULL; + iter_info.vers = vers; + iter_info.flags = 0; + iter_info.end = (char *)vers+len; + + /* + * Now loop through filling out the names & versions. + */ + dm_target_iterate(list_version_get_info, &iter_info); + param->flags |= iter_info.flags; + + out: + return 0; +} + + + +static int check_name(const char *name) +{ + if (strchr(name, '/')) { + DMWARN("invalid device name"); + return -EINVAL; + } + + return 0; +} + +/* + * Fills in a dm_ioctl structure, ready for sending back to + * userland. + */ +static int __dev_status(struct mapped_device *md, struct dm_ioctl *param) +{ + struct gendisk *disk = dm_disk(md); + struct dm_table *table; + struct block_device *bdev; + + param->flags &= ~(DM_SUSPEND_FLAG | DM_READONLY_FLAG | + DM_ACTIVE_PRESENT_FLAG); + + if (dm_suspended(md)) + param->flags |= DM_SUSPEND_FLAG; + + param->dev = huge_encode_dev(MKDEV(disk->major, disk->first_minor)); + + if (!(param->flags & DM_SKIP_BDGET_FLAG)) { + bdev = bdget_disk(disk, 0); + if (!bdev) + return -ENXIO; + + /* + * Yes, this will be out of date by the time it gets back + * to userland, but it is still very useful for + * debugging. + */ + param->open_count = bdev->bd_openers; + bdput(bdev); + } else + param->open_count = -1; + + if (disk->policy) + param->flags |= DM_READONLY_FLAG; + + param->event_nr = dm_get_event_nr(md); + + table = dm_get_table(md); + if (table) { + param->flags |= DM_ACTIVE_PRESENT_FLAG; + param->target_count = dm_table_get_num_targets(table); + dm_table_put(table); + } else + param->target_count = 0; + + return 0; +} + +static int dev_create(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct mapped_device *md; + + r = check_name(param->name); + if (r) + return r; + + if (param->flags & DM_PERSISTENT_DEV_FLAG) + r = dm_create_with_minor(MINOR(huge_decode_dev(param->dev)), &md); + else + r = dm_create(&md); + + if (r) + return r; + + r = dm_hash_insert(param->name, *param->uuid ? param->uuid : NULL, md); + if (r) { + dm_put(md); + return r; + } + + param->flags &= ~DM_INACTIVE_PRESENT_FLAG; + + r = __dev_status(md, param); + dm_put(md); + + return r; +} + +/* + * Always use UUID for lookups if it's present, otherwise use name or dev. + */ +static inline struct hash_cell *__find_device_hash_cell(struct dm_ioctl *param) +{ + if (*param->uuid) + return __get_uuid_cell(param->uuid); + else if (*param->name) + return __get_name_cell(param->name); + else + return dm_get_mdptr(huge_decode_dev(param->dev)); +} + +static inline struct mapped_device *find_device(struct dm_ioctl *param) +{ + struct hash_cell *hc; + struct mapped_device *md = NULL; + + down_read(&_hash_lock); + hc = __find_device_hash_cell(param); + if (hc) { + md = hc->md; + dm_get(md); + + /* + * Sneakily write in both the name and the uuid + * while we have the cell. + */ + strncpy(param->name, hc->name, sizeof(param->name)); + if (hc->uuid) + strncpy(param->uuid, hc->uuid, sizeof(param->uuid)-1); + else + param->uuid[0] = '\0'; + + if (hc->new_map) + param->flags |= DM_INACTIVE_PRESENT_FLAG; + else + param->flags &= ~DM_INACTIVE_PRESENT_FLAG; + } + up_read(&_hash_lock); + + return md; +} + +static int dev_remove(struct dm_ioctl *param, size_t param_size) +{ + struct hash_cell *hc; + + down_write(&_hash_lock); + hc = __find_device_hash_cell(param); + + if (!hc) { + DMWARN("device doesn't appear to be in the dev hash table."); + up_write(&_hash_lock); + return -ENXIO; + } + + __hash_remove(hc); + up_write(&_hash_lock); + param->data_size = 0; + return 0; +} + +/* + * Check a string doesn't overrun the chunk of + * memory we copied from userland. + */ +static int invalid_str(char *str, void *end) +{ + while ((void *) str < end) + if (!*str++) + return 0; + + return -EINVAL; +} + +static int dev_rename(struct dm_ioctl *param, size_t param_size) +{ + int r; + char *new_name = (char *) param + param->data_start; + + if (new_name < (char *) (param + 1) || + invalid_str(new_name, (void *) param + param_size)) { + DMWARN("Invalid new logical volume name supplied."); + return -EINVAL; + } + + r = check_name(new_name); + if (r) + return r; + + param->data_size = 0; + return dm_hash_rename(param->name, new_name); +} + +static int do_suspend(struct dm_ioctl *param) +{ + int r = 0; + struct mapped_device *md; + + md = find_device(param); + if (!md) + return -ENXIO; + + if (!dm_suspended(md)) + r = dm_suspend(md); + + if (!r) + r = __dev_status(md, param); + + dm_put(md); + return r; +} + +static int do_resume(struct dm_ioctl *param) +{ + int r = 0; + struct hash_cell *hc; + struct mapped_device *md; + struct dm_table *new_map; + + down_write(&_hash_lock); + + hc = __find_device_hash_cell(param); + if (!hc) { + DMWARN("device doesn't appear to be in the dev hash table."); + up_write(&_hash_lock); + return -ENXIO; + } + + md = hc->md; + dm_get(md); + + new_map = hc->new_map; + hc->new_map = NULL; + param->flags &= ~DM_INACTIVE_PRESENT_FLAG; + + up_write(&_hash_lock); + + /* Do we need to load a new map ? */ + if (new_map) { + /* Suspend if it isn't already suspended */ + if (!dm_suspended(md)) + dm_suspend(md); + + r = dm_swap_table(md, new_map); + if (r) { + dm_put(md); + dm_table_put(new_map); + return r; + } + + if (dm_table_get_mode(new_map) & FMODE_WRITE) + set_disk_ro(dm_disk(md), 0); + else + set_disk_ro(dm_disk(md), 1); + + dm_table_put(new_map); + } + + if (dm_suspended(md)) + r = dm_resume(md); + + if (!r) + r = __dev_status(md, param); + + dm_put(md); + return r; +} + +/* + * Set or unset the suspension state of a device. + * If the device already is in the requested state we just return its status. + */ +static int dev_suspend(struct dm_ioctl *param, size_t param_size) +{ + if (param->flags & DM_SUSPEND_FLAG) + return do_suspend(param); + + return do_resume(param); +} + +/* + * Copies device info back to user space, used by + * the create and info ioctls. + */ +static int dev_status(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct mapped_device *md; + + md = find_device(param); + if (!md) + return -ENXIO; + + r = __dev_status(md, param); + dm_put(md); + return r; +} + +/* + * Build up the status struct for each target + */ +static void retrieve_status(struct dm_table *table, + struct dm_ioctl *param, size_t param_size) +{ + unsigned int i, num_targets; + struct dm_target_spec *spec; + char *outbuf, *outptr; + status_type_t type; + size_t remaining, len, used = 0; + + outptr = outbuf = get_result_buffer(param, param_size, &len); + + if (param->flags & DM_STATUS_TABLE_FLAG) + type = STATUSTYPE_TABLE; + else + type = STATUSTYPE_INFO; + + /* Get all the target info */ + num_targets = dm_table_get_num_targets(table); + for (i = 0; i < num_targets; i++) { + struct dm_target *ti = dm_table_get_target(table, i); + + remaining = len - (outptr - outbuf); + if (remaining <= sizeof(struct dm_target_spec)) { + param->flags |= DM_BUFFER_FULL_FLAG; + break; + } + + spec = (struct dm_target_spec *) outptr; + + spec->status = 0; + spec->sector_start = ti->begin; + spec->length = ti->len; + strncpy(spec->target_type, ti->type->name, + sizeof(spec->target_type)); + + outptr += sizeof(struct dm_target_spec); + remaining = len - (outptr - outbuf); + if (remaining <= 0) { + param->flags |= DM_BUFFER_FULL_FLAG; + break; + } + + /* Get the status/table string from the target driver */ + if (ti->type->status) { + if (ti->type->status(ti, type, outptr, remaining)) { + param->flags |= DM_BUFFER_FULL_FLAG; + break; + } + } else + outptr[0] = '\0'; + + outptr += strlen(outptr) + 1; + used = param->data_start + (outptr - outbuf); + + outptr = align_ptr(outptr); + spec->next = outptr - outbuf; + } + + if (used) + param->data_size = used; + + param->target_count = num_targets; +} + +/* + * Wait for a device to report an event + */ +static int dev_wait(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct mapped_device *md; + struct dm_table *table; + + md = find_device(param); + if (!md) + return -ENXIO; + + /* + * Wait for a notification event + */ + if (dm_wait_event(md, param->event_nr)) { + r = -ERESTARTSYS; + goto out; + } + + /* + * The userland program is going to want to know what + * changed to trigger the event, so we may as well tell + * him and save an ioctl. + */ + r = __dev_status(md, param); + if (r) + goto out; + + table = dm_get_table(md); + if (table) { + retrieve_status(table, param, param_size); + dm_table_put(table); + } + + out: + dm_put(md); + return r; +} + +static inline int get_mode(struct dm_ioctl *param) +{ + int mode = FMODE_READ | FMODE_WRITE; + + if (param->flags & DM_READONLY_FLAG) + mode = FMODE_READ; + + return mode; +} + +static int next_target(struct dm_target_spec *last, uint32_t next, void *end, + struct dm_target_spec **spec, char **target_params) +{ + *spec = (struct dm_target_spec *) ((unsigned char *) last + next); + *target_params = (char *) (*spec + 1); + + if (*spec < (last + 1)) + return -EINVAL; + + return invalid_str(*target_params, end); +} + +static int populate_table(struct dm_table *table, + struct dm_ioctl *param, size_t param_size) +{ + int r; + unsigned int i = 0; + struct dm_target_spec *spec = (struct dm_target_spec *) param; + uint32_t next = param->data_start; + void *end = (void *) param + param_size; + char *target_params; + + if (!param->target_count) { + DMWARN("populate_table: no targets specified"); + return -EINVAL; + } + + for (i = 0; i < param->target_count; i++) { + + r = next_target(spec, next, end, &spec, &target_params); + if (r) { + DMWARN("unable to find target"); + return r; + } + + r = dm_table_add_target(table, spec->target_type, + (sector_t) spec->sector_start, + (sector_t) spec->length, + target_params); + if (r) { + DMWARN("error adding target to table"); + return r; + } + + next = spec->next; + } + + return dm_table_complete(table); +} + +static int table_load(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct hash_cell *hc; + struct dm_table *t; + + r = dm_table_create(&t, get_mode(param), param->target_count); + if (r) + return r; + + r = populate_table(t, param, param_size); + if (r) { + dm_table_put(t); + return r; + } + + down_write(&_hash_lock); + hc = __find_device_hash_cell(param); + if (!hc) { + DMWARN("device doesn't appear to be in the dev hash table."); + up_write(&_hash_lock); + return -ENXIO; + } + + if (hc->new_map) + dm_table_put(hc->new_map); + hc->new_map = t; + param->flags |= DM_INACTIVE_PRESENT_FLAG; + + r = __dev_status(hc->md, param); + up_write(&_hash_lock); + return r; +} + +static int table_clear(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct hash_cell *hc; + + down_write(&_hash_lock); + + hc = __find_device_hash_cell(param); + if (!hc) { + DMWARN("device doesn't appear to be in the dev hash table."); + up_write(&_hash_lock); + return -ENXIO; + } + + if (hc->new_map) { + dm_table_put(hc->new_map); + hc->new_map = NULL; + } + + param->flags &= ~DM_INACTIVE_PRESENT_FLAG; + + r = __dev_status(hc->md, param); + up_write(&_hash_lock); + return r; +} + +/* + * Retrieves a list of devices used by a particular dm device. + */ +static void retrieve_deps(struct dm_table *table, + struct dm_ioctl *param, size_t param_size) +{ + unsigned int count = 0; + struct list_head *tmp; + size_t len, needed; + struct dm_dev *dd; + struct dm_target_deps *deps; + + deps = get_result_buffer(param, param_size, &len); + + /* + * Count the devices. + */ + list_for_each (tmp, dm_table_get_devices(table)) + count++; + + /* + * Check we have enough space. + */ + needed = sizeof(*deps) + (sizeof(*deps->dev) * count); + if (len < needed) { + param->flags |= DM_BUFFER_FULL_FLAG; + return; + } + + /* + * Fill in the devices. + */ + deps->count = count; + count = 0; + list_for_each_entry (dd, dm_table_get_devices(table), list) + deps->dev[count++] = huge_encode_dev(dd->bdev->bd_dev); + + param->data_size = param->data_start + needed; +} + +static int table_deps(struct dm_ioctl *param, size_t param_size) +{ + int r = 0; + struct mapped_device *md; + struct dm_table *table; + + md = find_device(param); + if (!md) + return -ENXIO; + + r = __dev_status(md, param); + if (r) + goto out; + + table = dm_get_table(md); + if (table) { + retrieve_deps(table, param, param_size); + dm_table_put(table); + } + + out: + dm_put(md); + return r; +} + +/* + * Return the status of a device as a text string for each + * target. + */ +static int table_status(struct dm_ioctl *param, size_t param_size) +{ + int r; + struct mapped_device *md; + struct dm_table *table; + + md = find_device(param); + if (!md) + return -ENXIO; + + r = __dev_status(md, param); + if (r) + goto out; + + table = dm_get_table(md); + if (table) { + retrieve_status(table, param, param_size); + dm_table_put(table); + } + + out: + dm_put(md); + return r; +} + +/* + * Pass a message to the target that's at the supplied device offset. + */ +static int target_message(struct dm_ioctl *param, size_t param_size) +{ + int r, argc; + char **argv; + struct mapped_device *md; + struct dm_table *table; + struct dm_target *ti; + struct dm_target_msg *tmsg = (void *) param + param->data_start; + + md = find_device(param); + if (!md) + return -ENXIO; + + r = __dev_status(md, param); + if (r) + goto out; + + if (tmsg < (struct dm_target_msg *) (param + 1) || + invalid_str(tmsg->message, (void *) param + param_size)) { + DMWARN("Invalid target message parameters."); + r = -EINVAL; + goto out; + } + + r = dm_split_args(&argc, &argv, tmsg->message); + if (r) { + DMWARN("Failed to split target message parameters"); + goto out; + } + + table = dm_get_table(md); + if (!table) + goto out_argv; + + if (tmsg->sector >= dm_table_get_size(table)) { + DMWARN("Target message sector outside device."); + r = -EINVAL; + goto out_table; + } + + ti = dm_table_find_target(table, tmsg->sector); + if (ti->type->message) + r = ti->type->message(ti, argc, argv); + else { + DMWARN("Target type does not support messages"); + r = -EINVAL; + } + + out_table: + dm_table_put(table); + out_argv: + kfree(argv); + out: + param->data_size = 0; + dm_put(md); + return r; +} + +/*----------------------------------------------------------------- + * Implementation of open/close/ioctl on the special char + * device. + *---------------------------------------------------------------*/ +static ioctl_fn lookup_ioctl(unsigned int cmd) +{ + static struct { + int cmd; + ioctl_fn fn; + } _ioctls[] = { + {DM_VERSION_CMD, NULL}, /* version is dealt with elsewhere */ + {DM_REMOVE_ALL_CMD, remove_all}, + {DM_LIST_DEVICES_CMD, list_devices}, + + {DM_DEV_CREATE_CMD, dev_create}, + {DM_DEV_REMOVE_CMD, dev_remove}, + {DM_DEV_RENAME_CMD, dev_rename}, + {DM_DEV_SUSPEND_CMD, dev_suspend}, + {DM_DEV_STATUS_CMD, dev_status}, + {DM_DEV_WAIT_CMD, dev_wait}, + + {DM_TABLE_LOAD_CMD, table_load}, + {DM_TABLE_CLEAR_CMD, table_clear}, + {DM_TABLE_DEPS_CMD, table_deps}, + {DM_TABLE_STATUS_CMD, table_status}, + + {DM_LIST_VERSIONS_CMD, list_versions}, + + {DM_TARGET_MSG_CMD, target_message} + }; + + return (cmd >= ARRAY_SIZE(_ioctls)) ? NULL : _ioctls[cmd].fn; +} + +/* + * As well as checking the version compatibility this always + * copies the kernel interface version out. + */ +static int check_version(unsigned int cmd, struct dm_ioctl __user *user) +{ + uint32_t version[3]; + int r = 0; + + if (copy_from_user(version, user->version, sizeof(version))) + return -EFAULT; + + if ((DM_VERSION_MAJOR != version[0]) || + (DM_VERSION_MINOR < version[1])) { + DMWARN("ioctl interface mismatch: " + "kernel(%u.%u.%u), user(%u.%u.%u), cmd(%d)", + DM_VERSION_MAJOR, DM_VERSION_MINOR, + DM_VERSION_PATCHLEVEL, + version[0], version[1], version[2], cmd); + r = -EINVAL; + } + + /* + * Fill in the kernel version. + */ + version[0] = DM_VERSION_MAJOR; + version[1] = DM_VERSION_MINOR; + version[2] = DM_VERSION_PATCHLEVEL; + if (copy_to_user(user->version, version, sizeof(version))) + return -EFAULT; + + return r; +} + +static void free_params(struct dm_ioctl *param) +{ + vfree(param); +} + +static int copy_params(struct dm_ioctl __user *user, struct dm_ioctl **param) +{ + struct dm_ioctl tmp, *dmi; + + if (copy_from_user(&tmp, user, sizeof(tmp))) + return -EFAULT; + + if (tmp.data_size < sizeof(tmp)) + return -EINVAL; + + dmi = (struct dm_ioctl *) vmalloc(tmp.data_size); + if (!dmi) + return -ENOMEM; + + if (copy_from_user(dmi, user, tmp.data_size)) { + vfree(dmi); + return -EFAULT; + } + + *param = dmi; + return 0; +} + +static int validate_params(uint cmd, struct dm_ioctl *param) +{ + /* Always clear this flag */ + param->flags &= ~DM_BUFFER_FULL_FLAG; + + /* Ignores parameters */ + if (cmd == DM_REMOVE_ALL_CMD || + cmd == DM_LIST_DEVICES_CMD || + cmd == DM_LIST_VERSIONS_CMD) + return 0; + + if ((cmd == DM_DEV_CREATE_CMD)) { + if (!*param->name) { + DMWARN("name not supplied when creating device"); + return -EINVAL; + } + } else if ((*param->uuid && *param->name)) { + DMWARN("only supply one of name or uuid, cmd(%u)", cmd); + return -EINVAL; + } + + /* Ensure strings are terminated */ + param->name[DM_NAME_LEN - 1] = '\0'; + param->uuid[DM_UUID_LEN - 1] = '\0'; + + return 0; +} + +static int ctl_ioctl(struct inode *inode, struct file *file, + uint command, ulong u) +{ + int r = 0; + unsigned int cmd; + struct dm_ioctl *param; + struct dm_ioctl __user *user = (struct dm_ioctl __user *) u; + ioctl_fn fn = NULL; + size_t param_size; + + /* only root can play with this */ + if (!capable(CAP_SYS_ADMIN)) + return -EACCES; + + if (_IOC_TYPE(command) != DM_IOCTL) + return -ENOTTY; + + cmd = _IOC_NR(command); + + /* + * Check the interface version passed in. This also + * writes out the kernel's interface version. + */ + r = check_version(cmd, user); + if (r) + return r; + + /* + * Nothing more to do for the version command. + */ + if (cmd == DM_VERSION_CMD) + return 0; + + fn = lookup_ioctl(cmd); + if (!fn) { + DMWARN("dm_ctl_ioctl: unknown command 0x%x", command); + return -ENOTTY; + } + + /* + * Trying to avoid low memory issues when a device is + * suspended. + */ + current->flags |= PF_MEMALLOC; + + /* + * Copy the parameters into kernel space. + */ + r = copy_params(user, ¶m); + if (r) { + current->flags &= ~PF_MEMALLOC; + return r; + } + + /* + * FIXME: eventually we will remove the PF_MEMALLOC flag + * here. However the tools still do nasty things like + * 'load' while a device is suspended. + */ + + r = validate_params(cmd, param); + if (r) + goto out; + + param_size = param->data_size; + param->data_size = sizeof(*param); + r = fn(param, param_size); + + /* + * Copy the results back to userland. + */ + if (!r && copy_to_user(user, param, param->data_size)) + r = -EFAULT; + + out: + free_params(param); + current->flags &= ~PF_MEMALLOC; + return r; +} + +static struct file_operations _ctl_fops = { + .ioctl = ctl_ioctl, + .owner = THIS_MODULE, +}; + +static struct miscdevice _dm_misc = { + .minor = MISC_DYNAMIC_MINOR, + .name = DM_NAME, + .devfs_name = "mapper/control", + .fops = &_ctl_fops +}; + +/* + * Create misc character device and link to DM_DIR/control. + */ +int __init dm_interface_init(void) +{ + int r; + + r = dm_hash_init(); + if (r) + return r; + + r = misc_register(&_dm_misc); + if (r) { + DMERR("misc_register failed for control device"); + dm_hash_exit(); + return r; + } + + DMINFO("%d.%d.%d%s initialised: %s", DM_VERSION_MAJOR, + DM_VERSION_MINOR, DM_VERSION_PATCHLEVEL, DM_VERSION_EXTRA, + DM_DRIVER_EMAIL); + return 0; +} + +void dm_interface_exit(void) +{ + if (misc_deregister(&_dm_misc) < 0) + DMERR("misc_deregister failed for control device"); + + dm_hash_exit(); +} diff --git a/drivers/md/dm-linear.c b/drivers/md/dm-linear.c new file mode 100644 index 00000000000..6a2cd5dc8a6 --- /dev/null +++ b/drivers/md/dm-linear.c @@ -0,0 +1,123 @@ +/* + * Copyright (C) 2001-2003 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/blkdev.h> +#include <linux/bio.h> +#include <linux/slab.h> + +/* + * Linear: maps a linear range of a device. + */ +struct linear_c { + struct dm_dev *dev; + sector_t start; +}; + +/* + * Construct a linear mapping: <dev_path> <offset> + */ +static int linear_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + struct linear_c *lc; + + if (argc != 2) { + ti->error = "dm-linear: Invalid argument count"; + return -EINVAL; + } + + lc = kmalloc(sizeof(*lc), GFP_KERNEL); + if (lc == NULL) { + ti->error = "dm-linear: Cannot allocate linear context"; + return -ENOMEM; + } + + if (sscanf(argv[1], SECTOR_FORMAT, &lc->start) != 1) { + ti->error = "dm-linear: Invalid device sector"; + goto bad; + } + + if (dm_get_device(ti, argv[0], lc->start, ti->len, + dm_table_get_mode(ti->table), &lc->dev)) { + ti->error = "dm-linear: Device lookup failed"; + goto bad; + } + + ti->private = lc; + return 0; + + bad: + kfree(lc); + return -EINVAL; +} + +static void linear_dtr(struct dm_target *ti) +{ + struct linear_c *lc = (struct linear_c *) ti->private; + + dm_put_device(ti, lc->dev); + kfree(lc); +} + +static int linear_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + struct linear_c *lc = (struct linear_c *) ti->private; + + bio->bi_bdev = lc->dev->bdev; + bio->bi_sector = lc->start + (bio->bi_sector - ti->begin); + + return 1; +} + +static int linear_status(struct dm_target *ti, status_type_t type, + char *result, unsigned int maxlen) +{ + struct linear_c *lc = (struct linear_c *) ti->private; + + switch (type) { + case STATUSTYPE_INFO: + result[0] = '\0'; + break; + + case STATUSTYPE_TABLE: + snprintf(result, maxlen, "%s " SECTOR_FORMAT, lc->dev->name, + lc->start); + break; + } + return 0; +} + +static struct target_type linear_target = { + .name = "linear", + .version= {1, 0, 1}, + .module = THIS_MODULE, + .ctr = linear_ctr, + .dtr = linear_dtr, + .map = linear_map, + .status = linear_status, +}; + +int __init dm_linear_init(void) +{ + int r = dm_register_target(&linear_target); + + if (r < 0) + DMERR("linear: register failed %d", r); + + return r; +} + +void dm_linear_exit(void) +{ + int r = dm_unregister_target(&linear_target); + + if (r < 0) + DMERR("linear: unregister failed %d", r); +} diff --git a/drivers/md/dm-log.c b/drivers/md/dm-log.c new file mode 100644 index 00000000000..e110655eabd --- /dev/null +++ b/drivers/md/dm-log.c @@ -0,0 +1,711 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the LGPL. + */ + +#include <linux/init.h> +#include <linux/slab.h> +#include <linux/module.h> +#include <linux/vmalloc.h> + +#include "dm-log.h" +#include "dm-io.h" + +static LIST_HEAD(_log_types); +static DEFINE_SPINLOCK(_lock); + +int dm_register_dirty_log_type(struct dirty_log_type *type) +{ + spin_lock(&_lock); + type->use_count = 0; + list_add(&type->list, &_log_types); + spin_unlock(&_lock); + + return 0; +} + +int dm_unregister_dirty_log_type(struct dirty_log_type *type) +{ + spin_lock(&_lock); + + if (type->use_count) + DMWARN("Attempt to unregister a log type that is still in use"); + else + list_del(&type->list); + + spin_unlock(&_lock); + + return 0; +} + +static struct dirty_log_type *get_type(const char *type_name) +{ + struct dirty_log_type *type; + + spin_lock(&_lock); + list_for_each_entry (type, &_log_types, list) + if (!strcmp(type_name, type->name)) { + if (!type->use_count && !try_module_get(type->module)){ + spin_unlock(&_lock); + return NULL; + } + type->use_count++; + spin_unlock(&_lock); + return type; + } + + spin_unlock(&_lock); + return NULL; +} + +static void put_type(struct dirty_log_type *type) +{ + spin_lock(&_lock); + if (!--type->use_count) + module_put(type->module); + spin_unlock(&_lock); +} + +struct dirty_log *dm_create_dirty_log(const char *type_name, struct dm_target *ti, + unsigned int argc, char **argv) +{ + struct dirty_log_type *type; + struct dirty_log *log; + + log = kmalloc(sizeof(*log), GFP_KERNEL); + if (!log) + return NULL; + + type = get_type(type_name); + if (!type) { + kfree(log); + return NULL; + } + + log->type = type; + if (type->ctr(log, ti, argc, argv)) { + kfree(log); + put_type(type); + return NULL; + } + + return log; +} + +void dm_destroy_dirty_log(struct dirty_log *log) +{ + log->type->dtr(log); + put_type(log->type); + kfree(log); +} + +/*----------------------------------------------------------------- + * Persistent and core logs share a lot of their implementation. + * FIXME: need a reload method to be called from a resume + *---------------------------------------------------------------*/ +/* + * Magic for persistent mirrors: "MiRr" + */ +#define MIRROR_MAGIC 0x4D695272 + +/* + * The on-disk version of the metadata. + */ +#define MIRROR_DISK_VERSION 1 +#define LOG_OFFSET 2 + +struct log_header { + uint32_t magic; + + /* + * Simple, incrementing version. no backward + * compatibility. + */ + uint32_t version; + sector_t nr_regions; +}; + +struct log_c { + struct dm_target *ti; + int touched; + uint32_t region_size; + unsigned int region_count; + region_t sync_count; + + unsigned bitset_uint32_count; + uint32_t *clean_bits; + uint32_t *sync_bits; + uint32_t *recovering_bits; /* FIXME: this seems excessive */ + + int sync_search; + + /* Resync flag */ + enum sync { + DEFAULTSYNC, /* Synchronize if necessary */ + NOSYNC, /* Devices known to be already in sync */ + FORCESYNC, /* Force a sync to happen */ + } sync; + + /* + * Disk log fields + */ + struct dm_dev *log_dev; + struct log_header header; + + struct io_region header_location; + struct log_header *disk_header; + + struct io_region bits_location; + uint32_t *disk_bits; +}; + +/* + * The touched member needs to be updated every time we access + * one of the bitsets. + */ +static inline int log_test_bit(uint32_t *bs, unsigned bit) +{ + return test_bit(bit, (unsigned long *) bs) ? 1 : 0; +} + +static inline void log_set_bit(struct log_c *l, + uint32_t *bs, unsigned bit) +{ + set_bit(bit, (unsigned long *) bs); + l->touched = 1; +} + +static inline void log_clear_bit(struct log_c *l, + uint32_t *bs, unsigned bit) +{ + clear_bit(bit, (unsigned long *) bs); + l->touched = 1; +} + +/*---------------------------------------------------------------- + * Header IO + *--------------------------------------------------------------*/ +static void header_to_disk(struct log_header *core, struct log_header *disk) +{ + disk->magic = cpu_to_le32(core->magic); + disk->version = cpu_to_le32(core->version); + disk->nr_regions = cpu_to_le64(core->nr_regions); +} + +static void header_from_disk(struct log_header *core, struct log_header *disk) +{ + core->magic = le32_to_cpu(disk->magic); + core->version = le32_to_cpu(disk->version); + core->nr_regions = le64_to_cpu(disk->nr_regions); +} + +static int read_header(struct log_c *log) +{ + int r; + unsigned long ebits; + + r = dm_io_sync_vm(1, &log->header_location, READ, + log->disk_header, &ebits); + if (r) + return r; + + header_from_disk(&log->header, log->disk_header); + + /* New log required? */ + if (log->sync != DEFAULTSYNC || log->header.magic != MIRROR_MAGIC) { + log->header.magic = MIRROR_MAGIC; + log->header.version = MIRROR_DISK_VERSION; + log->header.nr_regions = 0; + } + + if (log->header.version != MIRROR_DISK_VERSION) { + DMWARN("incompatible disk log version"); + return -EINVAL; + } + + return 0; +} + +static inline int write_header(struct log_c *log) +{ + unsigned long ebits; + + header_to_disk(&log->header, log->disk_header); + return dm_io_sync_vm(1, &log->header_location, WRITE, + log->disk_header, &ebits); +} + +/*---------------------------------------------------------------- + * Bits IO + *--------------------------------------------------------------*/ +static inline void bits_to_core(uint32_t *core, uint32_t *disk, unsigned count) +{ + unsigned i; + + for (i = 0; i < count; i++) + core[i] = le32_to_cpu(disk[i]); +} + +static inline void bits_to_disk(uint32_t *core, uint32_t *disk, unsigned count) +{ + unsigned i; + + /* copy across the clean/dirty bitset */ + for (i = 0; i < count; i++) + disk[i] = cpu_to_le32(core[i]); +} + +static int read_bits(struct log_c *log) +{ + int r; + unsigned long ebits; + + r = dm_io_sync_vm(1, &log->bits_location, READ, + log->disk_bits, &ebits); + if (r) + return r; + + bits_to_core(log->clean_bits, log->disk_bits, + log->bitset_uint32_count); + return 0; +} + +static int write_bits(struct log_c *log) +{ + unsigned long ebits; + bits_to_disk(log->clean_bits, log->disk_bits, + log->bitset_uint32_count); + return dm_io_sync_vm(1, &log->bits_location, WRITE, + log->disk_bits, &ebits); +} + +/*---------------------------------------------------------------- + * core log constructor/destructor + * + * argv contains region_size followed optionally by [no]sync + *--------------------------------------------------------------*/ +#define BYTE_SHIFT 3 +static int core_ctr(struct dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv) +{ + enum sync sync = DEFAULTSYNC; + + struct log_c *lc; + uint32_t region_size; + unsigned int region_count; + size_t bitset_size; + + if (argc < 1 || argc > 2) { + DMWARN("wrong number of arguments to mirror log"); + return -EINVAL; + } + + if (argc > 1) { + if (!strcmp(argv[1], "sync")) + sync = FORCESYNC; + else if (!strcmp(argv[1], "nosync")) + sync = NOSYNC; + else { + DMWARN("unrecognised sync argument to mirror log: %s", + argv[1]); + return -EINVAL; + } + } + + if (sscanf(argv[0], "%u", ®ion_size) != 1) { + DMWARN("invalid region size string"); + return -EINVAL; + } + + region_count = dm_sector_div_up(ti->len, region_size); + + lc = kmalloc(sizeof(*lc), GFP_KERNEL); + if (!lc) { + DMWARN("couldn't allocate core log"); + return -ENOMEM; + } + + lc->ti = ti; + lc->touched = 0; + lc->region_size = region_size; + lc->region_count = region_count; + lc->sync = sync; + + /* + * Work out how many words we need to hold the bitset. + */ + bitset_size = dm_round_up(region_count, + sizeof(*lc->clean_bits) << BYTE_SHIFT); + bitset_size >>= BYTE_SHIFT; + + lc->bitset_uint32_count = bitset_size / 4; + lc->clean_bits = vmalloc(bitset_size); + if (!lc->clean_bits) { + DMWARN("couldn't allocate clean bitset"); + kfree(lc); + return -ENOMEM; + } + memset(lc->clean_bits, -1, bitset_size); + + lc->sync_bits = vmalloc(bitset_size); + if (!lc->sync_bits) { + DMWARN("couldn't allocate sync bitset"); + vfree(lc->clean_bits); + kfree(lc); + return -ENOMEM; + } + memset(lc->sync_bits, (sync == NOSYNC) ? -1 : 0, bitset_size); + lc->sync_count = (sync == NOSYNC) ? region_count : 0; + + lc->recovering_bits = vmalloc(bitset_size); + if (!lc->recovering_bits) { + DMWARN("couldn't allocate sync bitset"); + vfree(lc->sync_bits); + vfree(lc->clean_bits); + kfree(lc); + return -ENOMEM; + } + memset(lc->recovering_bits, 0, bitset_size); + lc->sync_search = 0; + log->context = lc; + return 0; +} + +static void core_dtr(struct dirty_log *log) +{ + struct log_c *lc = (struct log_c *) log->context; + vfree(lc->clean_bits); + vfree(lc->sync_bits); + vfree(lc->recovering_bits); + kfree(lc); +} + +/*---------------------------------------------------------------- + * disk log constructor/destructor + * + * argv contains log_device region_size followed optionally by [no]sync + *--------------------------------------------------------------*/ +static int disk_ctr(struct dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv) +{ + int r; + size_t size; + struct log_c *lc; + struct dm_dev *dev; + + if (argc < 2 || argc > 3) { + DMWARN("wrong number of arguments to disk mirror log"); + return -EINVAL; + } + + r = dm_get_device(ti, argv[0], 0, 0 /* FIXME */, + FMODE_READ | FMODE_WRITE, &dev); + if (r) + return r; + + r = core_ctr(log, ti, argc - 1, argv + 1); + if (r) { + dm_put_device(ti, dev); + return r; + } + + lc = (struct log_c *) log->context; + lc->log_dev = dev; + + /* setup the disk header fields */ + lc->header_location.bdev = lc->log_dev->bdev; + lc->header_location.sector = 0; + lc->header_location.count = 1; + + /* + * We can't read less than this amount, even though we'll + * not be using most of this space. + */ + lc->disk_header = vmalloc(1 << SECTOR_SHIFT); + if (!lc->disk_header) + goto bad; + + /* setup the disk bitset fields */ + lc->bits_location.bdev = lc->log_dev->bdev; + lc->bits_location.sector = LOG_OFFSET; + + size = dm_round_up(lc->bitset_uint32_count * sizeof(uint32_t), + 1 << SECTOR_SHIFT); + lc->bits_location.count = size >> SECTOR_SHIFT; + lc->disk_bits = vmalloc(size); + if (!lc->disk_bits) { + vfree(lc->disk_header); + goto bad; + } + return 0; + + bad: + dm_put_device(ti, lc->log_dev); + core_dtr(log); + return -ENOMEM; +} + +static void disk_dtr(struct dirty_log *log) +{ + struct log_c *lc = (struct log_c *) log->context; + dm_put_device(lc->ti, lc->log_dev); + vfree(lc->disk_header); + vfree(lc->disk_bits); + core_dtr(log); +} + +static int count_bits32(uint32_t *addr, unsigned size) +{ + int count = 0, i; + + for (i = 0; i < size; i++) { + count += hweight32(*(addr+i)); + } + return count; +} + +static int disk_resume(struct dirty_log *log) +{ + int r; + unsigned i; + struct log_c *lc = (struct log_c *) log->context; + size_t size = lc->bitset_uint32_count * sizeof(uint32_t); + + /* read the disk header */ + r = read_header(lc); + if (r) + return r; + + /* read the bits */ + r = read_bits(lc); + if (r) + return r; + + /* set or clear any new bits */ + if (lc->sync == NOSYNC) + for (i = lc->header.nr_regions; i < lc->region_count; i++) + /* FIXME: amazingly inefficient */ + log_set_bit(lc, lc->clean_bits, i); + else + for (i = lc->header.nr_regions; i < lc->region_count; i++) + /* FIXME: amazingly inefficient */ + log_clear_bit(lc, lc->clean_bits, i); + + /* copy clean across to sync */ + memcpy(lc->sync_bits, lc->clean_bits, size); + lc->sync_count = count_bits32(lc->clean_bits, lc->bitset_uint32_count); + + /* write the bits */ + r = write_bits(lc); + if (r) + return r; + + /* set the correct number of regions in the header */ + lc->header.nr_regions = lc->region_count; + + /* write the new header */ + return write_header(lc); +} + +static uint32_t core_get_region_size(struct dirty_log *log) +{ + struct log_c *lc = (struct log_c *) log->context; + return lc->region_size; +} + +static int core_is_clean(struct dirty_log *log, region_t region) +{ + struct log_c *lc = (struct log_c *) log->context; + return log_test_bit(lc->clean_bits, region); +} + +static int core_in_sync(struct dirty_log *log, region_t region, int block) +{ + struct log_c *lc = (struct log_c *) log->context; + return log_test_bit(lc->sync_bits, region); +} + +static int core_flush(struct dirty_log *log) +{ + /* no op */ + return 0; +} + +static int disk_flush(struct dirty_log *log) +{ + int r; + struct log_c *lc = (struct log_c *) log->context; + + /* only write if the log has changed */ + if (!lc->touched) + return 0; + + r = write_bits(lc); + if (!r) + lc->touched = 0; + + return r; +} + +static void core_mark_region(struct dirty_log *log, region_t region) +{ + struct log_c *lc = (struct log_c *) log->context; + log_clear_bit(lc, lc->clean_bits, region); +} + +static void core_clear_region(struct dirty_log *log, region_t region) +{ + struct log_c *lc = (struct log_c *) log->context; + log_set_bit(lc, lc->clean_bits, region); +} + +static int core_get_resync_work(struct dirty_log *log, region_t *region) +{ + struct log_c *lc = (struct log_c *) log->context; + + if (lc->sync_search >= lc->region_count) + return 0; + + do { + *region = find_next_zero_bit((unsigned long *) lc->sync_bits, + lc->region_count, + lc->sync_search); + lc->sync_search = *region + 1; + + if (*region == lc->region_count) + return 0; + + } while (log_test_bit(lc->recovering_bits, *region)); + + log_set_bit(lc, lc->recovering_bits, *region); + return 1; +} + +static void core_complete_resync_work(struct dirty_log *log, region_t region, + int success) +{ + struct log_c *lc = (struct log_c *) log->context; + + log_clear_bit(lc, lc->recovering_bits, region); + if (success) { + log_set_bit(lc, lc->sync_bits, region); + lc->sync_count++; + } +} + +static region_t core_get_sync_count(struct dirty_log *log) +{ + struct log_c *lc = (struct log_c *) log->context; + + return lc->sync_count; +} + +#define DMEMIT_SYNC \ + if (lc->sync != DEFAULTSYNC) \ + DMEMIT("%ssync ", lc->sync == NOSYNC ? "no" : "") + +static int core_status(struct dirty_log *log, status_type_t status, + char *result, unsigned int maxlen) +{ + int sz = 0; + struct log_c *lc = log->context; + + switch(status) { + case STATUSTYPE_INFO: + break; + + case STATUSTYPE_TABLE: + DMEMIT("%s %u %u ", log->type->name, + lc->sync == DEFAULTSYNC ? 1 : 2, lc->region_size); + DMEMIT_SYNC; + } + + return sz; +} + +static int disk_status(struct dirty_log *log, status_type_t status, + char *result, unsigned int maxlen) +{ + int sz = 0; + char buffer[16]; + struct log_c *lc = log->context; + + switch(status) { + case STATUSTYPE_INFO: + break; + + case STATUSTYPE_TABLE: + format_dev_t(buffer, lc->log_dev->bdev->bd_dev); + DMEMIT("%s %u %s %u ", log->type->name, + lc->sync == DEFAULTSYNC ? 2 : 3, buffer, + lc->region_size); + DMEMIT_SYNC; + } + + return sz; +} + +static struct dirty_log_type _core_type = { + .name = "core", + .module = THIS_MODULE, + .ctr = core_ctr, + .dtr = core_dtr, + .get_region_size = core_get_region_size, + .is_clean = core_is_clean, + .in_sync = core_in_sync, + .flush = core_flush, + .mark_region = core_mark_region, + .clear_region = core_clear_region, + .get_resync_work = core_get_resync_work, + .complete_resync_work = core_complete_resync_work, + .get_sync_count = core_get_sync_count, + .status = core_status, +}; + +static struct dirty_log_type _disk_type = { + .name = "disk", + .module = THIS_MODULE, + .ctr = disk_ctr, + .dtr = disk_dtr, + .suspend = disk_flush, + .resume = disk_resume, + .get_region_size = core_get_region_size, + .is_clean = core_is_clean, + .in_sync = core_in_sync, + .flush = disk_flush, + .mark_region = core_mark_region, + .clear_region = core_clear_region, + .get_resync_work = core_get_resync_work, + .complete_resync_work = core_complete_resync_work, + .get_sync_count = core_get_sync_count, + .status = disk_status, +}; + +int __init dm_dirty_log_init(void) +{ + int r; + + r = dm_register_dirty_log_type(&_core_type); + if (r) + DMWARN("couldn't register core log"); + + r = dm_register_dirty_log_type(&_disk_type); + if (r) { + DMWARN("couldn't register disk type"); + dm_unregister_dirty_log_type(&_core_type); + } + + return r; +} + +void dm_dirty_log_exit(void) +{ + dm_unregister_dirty_log_type(&_disk_type); + dm_unregister_dirty_log_type(&_core_type); +} + +EXPORT_SYMBOL(dm_register_dirty_log_type); +EXPORT_SYMBOL(dm_unregister_dirty_log_type); +EXPORT_SYMBOL(dm_create_dirty_log); +EXPORT_SYMBOL(dm_destroy_dirty_log); diff --git a/drivers/md/dm-log.h b/drivers/md/dm-log.h new file mode 100644 index 00000000000..5ae5309ebf2 --- /dev/null +++ b/drivers/md/dm-log.h @@ -0,0 +1,130 @@ +/* + * Copyright (C) 2003 Sistina Software + * + * This file is released under the LGPL. + */ + +#ifndef DM_DIRTY_LOG +#define DM_DIRTY_LOG + +#include "dm.h" + +typedef sector_t region_t; + +struct dirty_log_type; + +struct dirty_log { + struct dirty_log_type *type; + void *context; +}; + +struct dirty_log_type { + struct list_head list; + const char *name; + struct module *module; + unsigned int use_count; + + int (*ctr)(struct dirty_log *log, struct dm_target *ti, + unsigned int argc, char **argv); + void (*dtr)(struct dirty_log *log); + + /* + * There are times when we don't want the log to touch + * the disk. + */ + int (*suspend)(struct dirty_log *log); + int (*resume)(struct dirty_log *log); + + /* + * Retrieves the smallest size of region that the log can + * deal with. + */ + uint32_t (*get_region_size)(struct dirty_log *log); + + /* + * A predicate to say whether a region is clean or not. + * May block. + */ + int (*is_clean)(struct dirty_log *log, region_t region); + + /* + * Returns: 0, 1, -EWOULDBLOCK, < 0 + * + * A predicate function to check the area given by + * [sector, sector + len) is in sync. + * + * If -EWOULDBLOCK is returned the state of the region is + * unknown, typically this will result in a read being + * passed to a daemon to deal with, since a daemon is + * allowed to block. + */ + int (*in_sync)(struct dirty_log *log, region_t region, int can_block); + + /* + * Flush the current log state (eg, to disk). This + * function may block. + */ + int (*flush)(struct dirty_log *log); + + /* + * Mark an area as clean or dirty. These functions may + * block, though for performance reasons blocking should + * be extremely rare (eg, allocating another chunk of + * memory for some reason). + */ + void (*mark_region)(struct dirty_log *log, region_t region); + void (*clear_region)(struct dirty_log *log, region_t region); + + /* + * Returns: <0 (error), 0 (no region), 1 (region) + * + * The mirrord will need perform recovery on regions of + * the mirror that are in the NOSYNC state. This + * function asks the log to tell the caller about the + * next region that this machine should recover. + * + * Do not confuse this function with 'in_sync()', one + * tells you if an area is synchronised, the other + * assigns recovery work. + */ + int (*get_resync_work)(struct dirty_log *log, region_t *region); + + /* + * This notifies the log that the resync of an area has + * been completed. The log should then mark this region + * as CLEAN. + */ + void (*complete_resync_work)(struct dirty_log *log, + region_t region, int success); + + /* + * Returns the number of regions that are in sync. + */ + region_t (*get_sync_count)(struct dirty_log *log); + + /* + * Support function for mirror status requests. + */ + int (*status)(struct dirty_log *log, status_type_t status_type, + char *result, unsigned int maxlen); +}; + +int dm_register_dirty_log_type(struct dirty_log_type *type); +int dm_unregister_dirty_log_type(struct dirty_log_type *type); + + +/* + * Make sure you use these two functions, rather than calling + * type->constructor/destructor() directly. + */ +struct dirty_log *dm_create_dirty_log(const char *type_name, struct dm_target *ti, + unsigned int argc, char **argv); +void dm_destroy_dirty_log(struct dirty_log *log); + +/* + * init/exit functions. + */ +int dm_dirty_log_init(void); +void dm_dirty_log_exit(void); + +#endif diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c new file mode 100644 index 00000000000..43763a0bd09 --- /dev/null +++ b/drivers/md/dm-mpath.c @@ -0,0 +1,1302 @@ +/* + * Copyright (C) 2003 Sistina Software Limited. + * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + */ + +#include "dm.h" +#include "dm-path-selector.h" +#include "dm-hw-handler.h" +#include "dm-bio-list.h" +#include "dm-bio-record.h" + +#include <linux/ctype.h> +#include <linux/init.h> +#include <linux/mempool.h> +#include <linux/module.h> +#include <linux/pagemap.h> +#include <linux/slab.h> +#include <linux/time.h> +#include <linux/workqueue.h> +#include <asm/atomic.h> + +#define MESG_STR(x) x, sizeof(x) + +/* Path properties */ +struct pgpath { + struct list_head list; + + struct priority_group *pg; /* Owning PG */ + unsigned fail_count; /* Cumulative failure count */ + + struct path path; +}; + +#define path_to_pgpath(__pgp) container_of((__pgp), struct pgpath, path) + +/* + * Paths are grouped into Priority Groups and numbered from 1 upwards. + * Each has a path selector which controls which path gets used. + */ +struct priority_group { + struct list_head list; + + struct multipath *m; /* Owning multipath instance */ + struct path_selector ps; + + unsigned pg_num; /* Reference number */ + unsigned bypassed; /* Temporarily bypass this PG? */ + + unsigned nr_pgpaths; /* Number of paths in PG */ + struct list_head pgpaths; +}; + +/* Multipath context */ +struct multipath { + struct list_head list; + struct dm_target *ti; + + spinlock_t lock; + + struct hw_handler hw_handler; + unsigned nr_priority_groups; + struct list_head priority_groups; + unsigned pg_init_required; /* pg_init needs calling? */ + + unsigned nr_valid_paths; /* Total number of usable paths */ + struct pgpath *current_pgpath; + struct priority_group *current_pg; + struct priority_group *next_pg; /* Switch to this PG if set */ + unsigned repeat_count; /* I/Os left before calling PS again */ + + unsigned queue_io; /* Must we queue all I/O? */ + unsigned queue_if_no_path; /* Queue I/O if last path fails? */ + unsigned suspended; /* Has dm core suspended our I/O? */ + + struct work_struct process_queued_ios; + struct bio_list queued_ios; + unsigned queue_size; + + struct work_struct trigger_event; + + /* + * We must use a mempool of mpath_io structs so that we + * can resubmit bios on error. + */ + mempool_t *mpio_pool; +}; + +/* + * Context information attached to each bio we process. + */ +struct mpath_io { + struct pgpath *pgpath; + struct dm_bio_details details; +}; + +typedef int (*action_fn) (struct pgpath *pgpath); + +#define MIN_IOS 256 /* Mempool size */ + +static kmem_cache_t *_mpio_cache; + +static void process_queued_ios(void *data); +static void trigger_event(void *data); + + +/*----------------------------------------------- + * Allocation routines + *-----------------------------------------------*/ + +static struct pgpath *alloc_pgpath(void) +{ + struct pgpath *pgpath = kmalloc(sizeof(*pgpath), GFP_KERNEL); + + if (pgpath) { + memset(pgpath, 0, sizeof(*pgpath)); + pgpath->path.is_active = 1; + } + + return pgpath; +} + +static inline void free_pgpath(struct pgpath *pgpath) +{ + kfree(pgpath); +} + +static struct priority_group *alloc_priority_group(void) +{ + struct priority_group *pg; + + pg = kmalloc(sizeof(*pg), GFP_KERNEL); + if (!pg) + return NULL; + + memset(pg, 0, sizeof(*pg)); + INIT_LIST_HEAD(&pg->pgpaths); + + return pg; +} + +static void free_pgpaths(struct list_head *pgpaths, struct dm_target *ti) +{ + struct pgpath *pgpath, *tmp; + + list_for_each_entry_safe(pgpath, tmp, pgpaths, list) { + list_del(&pgpath->list); + dm_put_device(ti, pgpath->path.dev); + free_pgpath(pgpath); + } +} + +static void free_priority_group(struct priority_group *pg, + struct dm_target *ti) +{ + struct path_selector *ps = &pg->ps; + + if (ps->type) { + ps->type->destroy(ps); + dm_put_path_selector(ps->type); + } + + free_pgpaths(&pg->pgpaths, ti); + kfree(pg); +} + +static struct multipath *alloc_multipath(void) +{ + struct multipath *m; + + m = kmalloc(sizeof(*m), GFP_KERNEL); + if (m) { + memset(m, 0, sizeof(*m)); + INIT_LIST_HEAD(&m->priority_groups); + spin_lock_init(&m->lock); + m->queue_io = 1; + INIT_WORK(&m->process_queued_ios, process_queued_ios, m); + INIT_WORK(&m->trigger_event, trigger_event, m); + m->mpio_pool = mempool_create(MIN_IOS, mempool_alloc_slab, + mempool_free_slab, _mpio_cache); + if (!m->mpio_pool) { + kfree(m); + return NULL; + } + } + + return m; +} + +static void free_multipath(struct multipath *m) +{ + struct priority_group *pg, *tmp; + struct hw_handler *hwh = &m->hw_handler; + + list_for_each_entry_safe(pg, tmp, &m->priority_groups, list) { + list_del(&pg->list); + free_priority_group(pg, m->ti); + } + + if (hwh->type) { + hwh->type->destroy(hwh); + dm_put_hw_handler(hwh->type); + } + + mempool_destroy(m->mpio_pool); + kfree(m); +} + + +/*----------------------------------------------- + * Path selection + *-----------------------------------------------*/ + +static void __switch_pg(struct multipath *m, struct pgpath *pgpath) +{ + struct hw_handler *hwh = &m->hw_handler; + + m->current_pg = pgpath->pg; + + /* Must we initialise the PG first, and queue I/O till it's ready? */ + if (hwh->type && hwh->type->pg_init) { + m->pg_init_required = 1; + m->queue_io = 1; + } else { + m->pg_init_required = 0; + m->queue_io = 0; + } +} + +static int __choose_path_in_pg(struct multipath *m, struct priority_group *pg) +{ + struct path *path; + + path = pg->ps.type->select_path(&pg->ps, &m->repeat_count); + if (!path) + return -ENXIO; + + m->current_pgpath = path_to_pgpath(path); + + if (m->current_pg != pg) + __switch_pg(m, m->current_pgpath); + + return 0; +} + +static void __choose_pgpath(struct multipath *m) +{ + struct priority_group *pg; + unsigned bypassed = 1; + + if (!m->nr_valid_paths) + goto failed; + + /* Were we instructed to switch PG? */ + if (m->next_pg) { + pg = m->next_pg; + m->next_pg = NULL; + if (!__choose_path_in_pg(m, pg)) + return; + } + + /* Don't change PG until it has no remaining paths */ + if (m->current_pg && !__choose_path_in_pg(m, m->current_pg)) + return; + + /* + * Loop through priority groups until we find a valid path. + * First time we skip PGs marked 'bypassed'. + * Second time we only try the ones we skipped. + */ + do { + list_for_each_entry(pg, &m->priority_groups, list) { + if (pg->bypassed == bypassed) + continue; + if (!__choose_path_in_pg(m, pg)) + return; + } + } while (bypassed--); + +failed: + m->current_pgpath = NULL; + m->current_pg = NULL; +} + +static int map_io(struct multipath *m, struct bio *bio, struct mpath_io *mpio, + unsigned was_queued) +{ + int r = 1; + unsigned long flags; + struct pgpath *pgpath; + + spin_lock_irqsave(&m->lock, flags); + + /* Do we need to select a new pgpath? */ + if (!m->current_pgpath || + (!m->queue_io && (m->repeat_count && --m->repeat_count == 0))) + __choose_pgpath(m); + + pgpath = m->current_pgpath; + + if (was_queued) + m->queue_size--; + + if ((pgpath && m->queue_io) || + (!pgpath && m->queue_if_no_path && !m->suspended)) { + /* Queue for the daemon to resubmit */ + bio_list_add(&m->queued_ios, bio); + m->queue_size++; + if (m->pg_init_required || !m->queue_io) + schedule_work(&m->process_queued_ios); + pgpath = NULL; + r = 0; + } else if (!pgpath) + r = -EIO; /* Failed */ + else + bio->bi_bdev = pgpath->path.dev->bdev; + + mpio->pgpath = pgpath; + + spin_unlock_irqrestore(&m->lock, flags); + + return r; +} + +/* + * If we run out of usable paths, should we queue I/O or error it? + */ +static int queue_if_no_path(struct multipath *m, unsigned queue_if_no_path) +{ + unsigned long flags; + + spin_lock_irqsave(&m->lock, flags); + + m->queue_if_no_path = queue_if_no_path; + if (!m->queue_if_no_path) + schedule_work(&m->process_queued_ios); + + spin_unlock_irqrestore(&m->lock, flags); + + return 0; +} + +/*----------------------------------------------------------------- + * The multipath daemon is responsible for resubmitting queued ios. + *---------------------------------------------------------------*/ + +static void dispatch_queued_ios(struct multipath *m) +{ + int r; + unsigned long flags; + struct bio *bio = NULL, *next; + struct mpath_io *mpio; + union map_info *info; + + spin_lock_irqsave(&m->lock, flags); + bio = bio_list_get(&m->queued_ios); + spin_unlock_irqrestore(&m->lock, flags); + + while (bio) { + next = bio->bi_next; + bio->bi_next = NULL; + + info = dm_get_mapinfo(bio); + mpio = info->ptr; + + r = map_io(m, bio, mpio, 1); + if (r < 0) + bio_endio(bio, bio->bi_size, r); + else if (r == 1) + generic_make_request(bio); + + bio = next; + } +} + +static void process_queued_ios(void *data) +{ + struct multipath *m = (struct multipath *) data; + struct hw_handler *hwh = &m->hw_handler; + struct pgpath *pgpath; + unsigned init_required, must_queue = 0; + unsigned long flags; + + spin_lock_irqsave(&m->lock, flags); + + if (!m->current_pgpath) + __choose_pgpath(m); + + pgpath = m->current_pgpath; + + if ((pgpath && m->queue_io) || + (!pgpath && m->queue_if_no_path && !m->suspended)) + must_queue = 1; + + init_required = m->pg_init_required; + if (init_required) + m->pg_init_required = 0; + + spin_unlock_irqrestore(&m->lock, flags); + + if (init_required) + hwh->type->pg_init(hwh, pgpath->pg->bypassed, &pgpath->path); + + if (!must_queue) + dispatch_queued_ios(m); +} + +/* + * An event is triggered whenever a path is taken out of use. + * Includes path failure and PG bypass. + */ +static void trigger_event(void *data) +{ + struct multipath *m = (struct multipath *) data; + + dm_table_event(m->ti->table); +} + +/*----------------------------------------------------------------- + * Constructor/argument parsing: + * <#multipath feature args> [<arg>]* + * <#hw_handler args> [hw_handler [<arg>]*] + * <#priority groups> + * <initial priority group> + * [<selector> <#selector args> [<arg>]* + * <#paths> <#per-path selector args> + * [<path> [<arg>]* ]+ ]+ + *---------------------------------------------------------------*/ +struct param { + unsigned min; + unsigned max; + char *error; +}; + +#define ESTR(s) ("dm-multipath: " s) + +static int read_param(struct param *param, char *str, unsigned *v, char **error) +{ + if (!str || + (sscanf(str, "%u", v) != 1) || + (*v < param->min) || + (*v > param->max)) { + *error = param->error; + return -EINVAL; + } + + return 0; +} + +struct arg_set { + unsigned argc; + char **argv; +}; + +static char *shift(struct arg_set *as) +{ + char *r; + + if (as->argc) { + as->argc--; + r = *as->argv; + as->argv++; + return r; + } + + return NULL; +} + +static void consume(struct arg_set *as, unsigned n) +{ + BUG_ON (as->argc < n); + as->argc -= n; + as->argv += n; +} + +static int parse_path_selector(struct arg_set *as, struct priority_group *pg, + struct dm_target *ti) +{ + int r; + struct path_selector_type *pst; + unsigned ps_argc; + + static struct param _params[] = { + {0, 1024, ESTR("invalid number of path selector args")}, + }; + + pst = dm_get_path_selector(shift(as)); + if (!pst) { + ti->error = ESTR("unknown path selector type"); + return -EINVAL; + } + + r = read_param(_params, shift(as), &ps_argc, &ti->error); + if (r) + return -EINVAL; + + r = pst->create(&pg->ps, ps_argc, as->argv); + if (r) { + dm_put_path_selector(pst); + ti->error = ESTR("path selector constructor failed"); + return r; + } + + pg->ps.type = pst; + consume(as, ps_argc); + + return 0; +} + +static struct pgpath *parse_path(struct arg_set *as, struct path_selector *ps, + struct dm_target *ti) +{ + int r; + struct pgpath *p; + + /* we need at least a path arg */ + if (as->argc < 1) { + ti->error = ESTR("no device given"); + return NULL; + } + + p = alloc_pgpath(); + if (!p) + return NULL; + + r = dm_get_device(ti, shift(as), ti->begin, ti->len, + dm_table_get_mode(ti->table), &p->path.dev); + if (r) { + ti->error = ESTR("error getting device"); + goto bad; + } + + r = ps->type->add_path(ps, &p->path, as->argc, as->argv, &ti->error); + if (r) { + dm_put_device(ti, p->path.dev); + goto bad; + } + + return p; + + bad: + free_pgpath(p); + return NULL; +} + +static struct priority_group *parse_priority_group(struct arg_set *as, + struct multipath *m, + struct dm_target *ti) +{ + static struct param _params[] = { + {1, 1024, ESTR("invalid number of paths")}, + {0, 1024, ESTR("invalid number of selector args")} + }; + + int r; + unsigned i, nr_selector_args, nr_params; + struct priority_group *pg; + + if (as->argc < 2) { + as->argc = 0; + ti->error = ESTR("not enough priority group aruments"); + return NULL; + } + + pg = alloc_priority_group(); + if (!pg) { + ti->error = ESTR("couldn't allocate priority group"); + return NULL; + } + pg->m = m; + + r = parse_path_selector(as, pg, ti); + if (r) + goto bad; + + /* + * read the paths + */ + r = read_param(_params, shift(as), &pg->nr_pgpaths, &ti->error); + if (r) + goto bad; + + r = read_param(_params + 1, shift(as), &nr_selector_args, &ti->error); + if (r) + goto bad; + + nr_params = 1 + nr_selector_args; + for (i = 0; i < pg->nr_pgpaths; i++) { + struct pgpath *pgpath; + struct arg_set path_args; + + if (as->argc < nr_params) + goto bad; + + path_args.argc = nr_params; + path_args.argv = as->argv; + + pgpath = parse_path(&path_args, &pg->ps, ti); + if (!pgpath) + goto bad; + + pgpath->pg = pg; + list_add_tail(&pgpath->list, &pg->pgpaths); + consume(as, nr_params); + } + + return pg; + + bad: + free_priority_group(pg, ti); + return NULL; +} + +static int parse_hw_handler(struct arg_set *as, struct multipath *m, + struct dm_target *ti) +{ + int r; + struct hw_handler_type *hwht; + unsigned hw_argc; + + static struct param _params[] = { + {0, 1024, ESTR("invalid number of hardware handler args")}, + }; + + r = read_param(_params, shift(as), &hw_argc, &ti->error); + if (r) + return -EINVAL; + + if (!hw_argc) + return 0; + + hwht = dm_get_hw_handler(shift(as)); + if (!hwht) { + ti->error = ESTR("unknown hardware handler type"); + return -EINVAL; + } + + r = hwht->create(&m->hw_handler, hw_argc - 1, as->argv); + if (r) { + dm_put_hw_handler(hwht); + ti->error = ESTR("hardware handler constructor failed"); + return r; + } + + m->hw_handler.type = hwht; + consume(as, hw_argc - 1); + + return 0; +} + +static int parse_features(struct arg_set *as, struct multipath *m, + struct dm_target *ti) +{ + int r; + unsigned argc; + + static struct param _params[] = { + {0, 1, ESTR("invalid number of feature args")}, + }; + + r = read_param(_params, shift(as), &argc, &ti->error); + if (r) + return -EINVAL; + + if (!argc) + return 0; + + if (!strnicmp(shift(as), MESG_STR("queue_if_no_path"))) + return queue_if_no_path(m, 1); + else { + ti->error = "Unrecognised multipath feature request"; + return -EINVAL; + } +} + +static int multipath_ctr(struct dm_target *ti, unsigned int argc, + char **argv) +{ + /* target parameters */ + static struct param _params[] = { + {1, 1024, ESTR("invalid number of priority groups")}, + {1, 1024, ESTR("invalid initial priority group number")}, + }; + + int r; + struct multipath *m; + struct arg_set as; + unsigned pg_count = 0; + unsigned next_pg_num; + + as.argc = argc; + as.argv = argv; + + m = alloc_multipath(); + if (!m) { + ti->error = ESTR("can't allocate multipath"); + return -EINVAL; + } + + r = parse_features(&as, m, ti); + if (r) + goto bad; + + r = parse_hw_handler(&as, m, ti); + if (r) + goto bad; + + r = read_param(_params, shift(&as), &m->nr_priority_groups, &ti->error); + if (r) + goto bad; + + r = read_param(_params + 1, shift(&as), &next_pg_num, &ti->error); + if (r) + goto bad; + + /* parse the priority groups */ + while (as.argc) { + struct priority_group *pg; + + pg = parse_priority_group(&as, m, ti); + if (!pg) { + r = -EINVAL; + goto bad; + } + + m->nr_valid_paths += pg->nr_pgpaths; + list_add_tail(&pg->list, &m->priority_groups); + pg_count++; + pg->pg_num = pg_count; + if (!--next_pg_num) + m->next_pg = pg; + } + + if (pg_count != m->nr_priority_groups) { + ti->error = ESTR("priority group count mismatch"); + r = -EINVAL; + goto bad; + } + + ti->private = m; + m->ti = ti; + + return 0; + + bad: + free_multipath(m); + return r; +} + +static void multipath_dtr(struct dm_target *ti) +{ + struct multipath *m = (struct multipath *) ti->private; + free_multipath(m); +} + +/* + * Map bios, recording original fields for later in case we have to resubmit + */ +static int multipath_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + int r; + struct mpath_io *mpio; + struct multipath *m = (struct multipath *) ti->private; + + mpio = mempool_alloc(m->mpio_pool, GFP_NOIO); + dm_bio_record(&mpio->details, bio); + + map_context->ptr = mpio; + bio->bi_rw |= (1 << BIO_RW_FAILFAST); + r = map_io(m, bio, mpio, 0); + if (r < 0) + mempool_free(mpio, m->mpio_pool); + + return r; +} + +/* + * Take a path out of use. + */ +static int fail_path(struct pgpath *pgpath) +{ + unsigned long flags; + struct multipath *m = pgpath->pg->m; + + spin_lock_irqsave(&m->lock, flags); + + if (!pgpath->path.is_active) + goto out; + + DMWARN("dm-multipath: Failing path %s.", pgpath->path.dev->name); + + pgpath->pg->ps.type->fail_path(&pgpath->pg->ps, &pgpath->path); + pgpath->path.is_active = 0; + pgpath->fail_count++; + + m->nr_valid_paths--; + + if (pgpath == m->current_pgpath) + m->current_pgpath = NULL; + + schedule_work(&m->trigger_event); + +out: + spin_unlock_irqrestore(&m->lock, flags); + + return 0; +} + +/* + * Reinstate a previously-failed path + */ +static int reinstate_path(struct pgpath *pgpath) +{ + int r = 0; + unsigned long flags; + struct multipath *m = pgpath->pg->m; + + spin_lock_irqsave(&m->lock, flags); + + if (pgpath->path.is_active) + goto out; + + if (!pgpath->pg->ps.type) { + DMWARN("Reinstate path not supported by path selector %s", + pgpath->pg->ps.type->name); + r = -EINVAL; + goto out; + } + + r = pgpath->pg->ps.type->reinstate_path(&pgpath->pg->ps, &pgpath->path); + if (r) + goto out; + + pgpath->path.is_active = 1; + + m->current_pgpath = NULL; + if (!m->nr_valid_paths++) + schedule_work(&m->process_queued_ios); + + schedule_work(&m->trigger_event); + +out: + spin_unlock_irqrestore(&m->lock, flags); + + return r; +} + +/* + * Fail or reinstate all paths that match the provided struct dm_dev. + */ +static int action_dev(struct multipath *m, struct dm_dev *dev, + action_fn action) +{ + int r = 0; + struct pgpath *pgpath; + struct priority_group *pg; + + list_for_each_entry(pg, &m->priority_groups, list) { + list_for_each_entry(pgpath, &pg->pgpaths, list) { + if (pgpath->path.dev == dev) + r = action(pgpath); + } + } + + return r; +} + +/* + * Temporarily try to avoid having to use the specified PG + */ +static void bypass_pg(struct multipath *m, struct priority_group *pg, + int bypassed) +{ + unsigned long flags; + + spin_lock_irqsave(&m->lock, flags); + + pg->bypassed = bypassed; + m->current_pgpath = NULL; + m->current_pg = NULL; + + spin_unlock_irqrestore(&m->lock, flags); + + schedule_work(&m->trigger_event); +} + +/* + * Switch to using the specified PG from the next I/O that gets mapped + */ +static int switch_pg_num(struct multipath *m, const char *pgstr) +{ + struct priority_group *pg; + unsigned pgnum; + unsigned long flags; + + if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || + (pgnum > m->nr_priority_groups)) { + DMWARN("invalid PG number supplied to switch_pg_num"); + return -EINVAL; + } + + spin_lock_irqsave(&m->lock, flags); + list_for_each_entry(pg, &m->priority_groups, list) { + pg->bypassed = 0; + if (--pgnum) + continue; + + m->current_pgpath = NULL; + m->current_pg = NULL; + m->next_pg = pg; + } + spin_unlock_irqrestore(&m->lock, flags); + + schedule_work(&m->trigger_event); + return 0; +} + +/* + * Set/clear bypassed status of a PG. + * PGs are numbered upwards from 1 in the order they were declared. + */ +static int bypass_pg_num(struct multipath *m, const char *pgstr, int bypassed) +{ + struct priority_group *pg; + unsigned pgnum; + + if (!pgstr || (sscanf(pgstr, "%u", &pgnum) != 1) || !pgnum || + (pgnum > m->nr_priority_groups)) { + DMWARN("invalid PG number supplied to bypass_pg"); + return -EINVAL; + } + + list_for_each_entry(pg, &m->priority_groups, list) { + if (!--pgnum) + break; + } + + bypass_pg(m, pg, bypassed); + return 0; +} + +/* + * pg_init must call this when it has completed its initialisation + */ +void dm_pg_init_complete(struct path *path, unsigned err_flags) +{ + struct pgpath *pgpath = path_to_pgpath(path); + struct priority_group *pg = pgpath->pg; + struct multipath *m = pg->m; + unsigned long flags; + + /* We insist on failing the path if the PG is already bypassed. */ + if (err_flags && pg->bypassed) + err_flags |= MP_FAIL_PATH; + + if (err_flags & MP_FAIL_PATH) + fail_path(pgpath); + + if (err_flags & MP_BYPASS_PG) + bypass_pg(m, pg, 1); + + spin_lock_irqsave(&m->lock, flags); + if (!err_flags) + m->queue_io = 0; + else { + m->current_pgpath = NULL; + m->current_pg = NULL; + } + schedule_work(&m->process_queued_ios); + spin_unlock_irqrestore(&m->lock, flags); +} + +/* + * end_io handling + */ +static int do_end_io(struct multipath *m, struct bio *bio, + int error, struct mpath_io *mpio) +{ + struct hw_handler *hwh = &m->hw_handler; + unsigned err_flags = MP_FAIL_PATH; /* Default behavior */ + + if (!error) + return 0; /* I/O complete */ + + spin_lock(&m->lock); + if (!m->nr_valid_paths) { + if (!m->queue_if_no_path || m->suspended) { + spin_unlock(&m->lock); + return -EIO; + } else { + spin_unlock(&m->lock); + goto requeue; + } + } + spin_unlock(&m->lock); + + if (hwh->type && hwh->type->error) + err_flags = hwh->type->error(hwh, bio); + + if (mpio->pgpath) { + if (err_flags & MP_FAIL_PATH) + fail_path(mpio->pgpath); + + if (err_flags & MP_BYPASS_PG) + bypass_pg(m, mpio->pgpath->pg, 1); + } + + if (err_flags & MP_ERROR_IO) + return -EIO; + + requeue: + dm_bio_restore(&mpio->details, bio); + + /* queue for the daemon to resubmit or fail */ + spin_lock(&m->lock); + bio_list_add(&m->queued_ios, bio); + m->queue_size++; + if (!m->queue_io) + schedule_work(&m->process_queued_ios); + spin_unlock(&m->lock); + + return 1; /* io not complete */ +} + +static int multipath_end_io(struct dm_target *ti, struct bio *bio, + int error, union map_info *map_context) +{ + struct multipath *m = (struct multipath *) ti->private; + struct mpath_io *mpio = (struct mpath_io *) map_context->ptr; + struct pgpath *pgpath = mpio->pgpath; + struct path_selector *ps; + int r; + + r = do_end_io(m, bio, error, mpio); + if (pgpath) { + ps = &pgpath->pg->ps; + if (ps->type->end_io) + ps->type->end_io(ps, &pgpath->path); + } + if (r <= 0) + mempool_free(mpio, m->mpio_pool); + + return r; +} + +/* + * Suspend can't complete until all the I/O is processed so if + * the last path failed we will now error any queued I/O. + */ +static void multipath_presuspend(struct dm_target *ti) +{ + struct multipath *m = (struct multipath *) ti->private; + unsigned long flags; + + spin_lock_irqsave(&m->lock, flags); + m->suspended = 1; + if (m->queue_if_no_path) + schedule_work(&m->process_queued_ios); + spin_unlock_irqrestore(&m->lock, flags); +} + +static void multipath_resume(struct dm_target *ti) +{ + struct multipath *m = (struct multipath *) ti->private; + unsigned long flags; + + spin_lock_irqsave(&m->lock, flags); + m->suspended = 0; + spin_unlock_irqrestore(&m->lock, flags); +} + +/* + * Info output has the following format: + * num_multipath_feature_args [multipath_feature_args]* + * num_handler_status_args [handler_status_args]* + * num_groups init_group_number + * [A|D|E num_ps_status_args [ps_status_args]* + * num_paths num_selector_args + * [path_dev A|F fail_count [selector_args]* ]+ ]+ + * + * Table output has the following format (identical to the constructor string): + * num_feature_args [features_args]* + * num_handler_args hw_handler [hw_handler_args]* + * num_groups init_group_number + * [priority selector-name num_ps_args [ps_args]* + * num_paths num_selector_args [path_dev [selector_args]* ]+ ]+ + */ +static int multipath_status(struct dm_target *ti, status_type_t type, + char *result, unsigned int maxlen) +{ + int sz = 0; + unsigned long flags; + struct multipath *m = (struct multipath *) ti->private; + struct hw_handler *hwh = &m->hw_handler; + struct priority_group *pg; + struct pgpath *p; + unsigned pg_num; + char state; + + spin_lock_irqsave(&m->lock, flags); + + /* Features */ + if (type == STATUSTYPE_INFO) + DMEMIT("1 %u ", m->queue_size); + else if (m->queue_if_no_path) + DMEMIT("1 queue_if_no_path "); + else + DMEMIT("0 "); + + if (hwh->type && hwh->type->status) + sz += hwh->type->status(hwh, type, result + sz, maxlen - sz); + else if (!hwh->type || type == STATUSTYPE_INFO) + DMEMIT("0 "); + else + DMEMIT("1 %s ", hwh->type->name); + + DMEMIT("%u ", m->nr_priority_groups); + + if (m->next_pg) + pg_num = m->next_pg->pg_num; + else if (m->current_pg) + pg_num = m->current_pg->pg_num; + else + pg_num = 1; + + DMEMIT("%u ", pg_num); + + switch (type) { + case STATUSTYPE_INFO: + list_for_each_entry(pg, &m->priority_groups, list) { + if (pg->bypassed) + state = 'D'; /* Disabled */ + else if (pg == m->current_pg) + state = 'A'; /* Currently Active */ + else + state = 'E'; /* Enabled */ + + DMEMIT("%c ", state); + + if (pg->ps.type->status) + sz += pg->ps.type->status(&pg->ps, NULL, type, + result + sz, + maxlen - sz); + else + DMEMIT("0 "); + + DMEMIT("%u %u ", pg->nr_pgpaths, + pg->ps.type->info_args); + + list_for_each_entry(p, &pg->pgpaths, list) { + DMEMIT("%s %s %u ", p->path.dev->name, + p->path.is_active ? "A" : "F", + p->fail_count); + if (pg->ps.type->status) + sz += pg->ps.type->status(&pg->ps, + &p->path, type, result + sz, + maxlen - sz); + } + } + break; + + case STATUSTYPE_TABLE: + list_for_each_entry(pg, &m->priority_groups, list) { + DMEMIT("%s ", pg->ps.type->name); + + if (pg->ps.type->status) + sz += pg->ps.type->status(&pg->ps, NULL, type, + result + sz, + maxlen - sz); + else + DMEMIT("0 "); + + DMEMIT("%u %u ", pg->nr_pgpaths, + pg->ps.type->table_args); + + list_for_each_entry(p, &pg->pgpaths, list) { + DMEMIT("%s ", p->path.dev->name); + if (pg->ps.type->status) + sz += pg->ps.type->status(&pg->ps, + &p->path, type, result + sz, + maxlen - sz); + } + } + break; + } + + spin_unlock_irqrestore(&m->lock, flags); + + return 0; +} + +static int multipath_message(struct dm_target *ti, unsigned argc, char **argv) +{ + int r; + struct dm_dev *dev; + struct multipath *m = (struct multipath *) ti->private; + action_fn action; + + if (argc == 1) { + if (!strnicmp(argv[0], MESG_STR("queue_if_no_path"))) + return queue_if_no_path(m, 1); + else if (!strnicmp(argv[0], MESG_STR("fail_if_no_path"))) + return queue_if_no_path(m, 0); + } + + if (argc != 2) + goto error; + + if (!strnicmp(argv[0], MESG_STR("disable_group"))) + return bypass_pg_num(m, argv[1], 1); + else if (!strnicmp(argv[0], MESG_STR("enable_group"))) + return bypass_pg_num(m, argv[1], 0); + else if (!strnicmp(argv[0], MESG_STR("switch_group"))) + return switch_pg_num(m, argv[1]); + else if (!strnicmp(argv[0], MESG_STR("reinstate_path"))) + action = reinstate_path; + else if (!strnicmp(argv[0], MESG_STR("fail_path"))) + action = fail_path; + else + goto error; + + r = dm_get_device(ti, argv[1], ti->begin, ti->len, + dm_table_get_mode(ti->table), &dev); + if (r) { + DMWARN("dm-multipath message: error getting device %s", + argv[1]); + return -EINVAL; + } + + r = action_dev(m, dev, action); + + dm_put_device(ti, dev); + + return r; + +error: + DMWARN("Unrecognised multipath message received."); + return -EINVAL; +} + +/*----------------------------------------------------------------- + * Module setup + *---------------------------------------------------------------*/ +static struct target_type multipath_target = { + .name = "multipath", + .version = {1, 0, 4}, + .module = THIS_MODULE, + .ctr = multipath_ctr, + .dtr = multipath_dtr, + .map = multipath_map, + .end_io = multipath_end_io, + .presuspend = multipath_presuspend, + .resume = multipath_resume, + .status = multipath_status, + .message = multipath_message, +}; + +static int __init dm_multipath_init(void) +{ + int r; + + /* allocate a slab for the dm_ios */ + _mpio_cache = kmem_cache_create("dm_mpath", sizeof(struct mpath_io), + 0, 0, NULL, NULL); + if (!_mpio_cache) + return -ENOMEM; + + r = dm_register_target(&multipath_target); + if (r < 0) { + DMERR("%s: register failed %d", multipath_target.name, r); + kmem_cache_destroy(_mpio_cache); + return -EINVAL; + } + + DMINFO("dm-multipath version %u.%u.%u loaded", + multipath_target.version[0], multipath_target.version[1], + multipath_target.version[2]); + + return r; +} + +static void __exit dm_multipath_exit(void) +{ + int r; + + r = dm_unregister_target(&multipath_target); + if (r < 0) + DMERR("%s: target unregister failed %d", + multipath_target.name, r); + kmem_cache_destroy(_mpio_cache); +} + +EXPORT_SYMBOL_GPL(dm_pg_init_complete); + +module_init(dm_multipath_init); +module_exit(dm_multipath_exit); + +MODULE_DESCRIPTION(DM_NAME " multipath target"); +MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-mpath.h b/drivers/md/dm-mpath.h new file mode 100644 index 00000000000..8a4bf2b6d52 --- /dev/null +++ b/drivers/md/dm-mpath.h @@ -0,0 +1,25 @@ +/* + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + * + * Multipath. + */ + +#ifndef DM_MPATH_H +#define DM_MPATH_H + +struct dm_dev; + +struct path { + struct dm_dev *dev; /* Read-only */ + unsigned is_active; /* Read-only */ + + void *pscontext; /* For path-selector use */ + void *hwhcontext; /* For hw-handler use */ +}; + +/* Callback for hwh_pg_init_fn to use when complete */ +void dm_pg_init_complete(struct path *path, unsigned err_flags); + +#endif diff --git a/drivers/md/dm-path-selector.c b/drivers/md/dm-path-selector.c new file mode 100644 index 00000000000..ac5c4bbec6c --- /dev/null +++ b/drivers/md/dm-path-selector.c @@ -0,0 +1,156 @@ +/* + * Copyright (C) 2003 Sistina Software. + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * Module Author: Heinz Mauelshagen + * + * This file is released under the GPL. + * + * Path selector registration. + */ + +#include "dm.h" +#include "dm-path-selector.h" + +#include <linux/slab.h> + +struct ps_internal { + struct path_selector_type pst; + + struct list_head list; + long use; +}; + +#define pst_to_psi(__pst) container_of((__pst), struct ps_internal, pst) + +static LIST_HEAD(_path_selectors); +static DECLARE_RWSEM(_ps_lock); + +struct ps_internal *__find_path_selector_type(const char *name) +{ + struct ps_internal *psi; + + list_for_each_entry(psi, &_path_selectors, list) { + if (!strcmp(name, psi->pst.name)) + return psi; + } + + return NULL; +} + +static struct ps_internal *get_path_selector(const char *name) +{ + struct ps_internal *psi; + + down_read(&_ps_lock); + psi = __find_path_selector_type(name); + if (psi) { + if ((psi->use == 0) && !try_module_get(psi->pst.module)) + psi = NULL; + else + psi->use++; + } + up_read(&_ps_lock); + + return psi; +} + +struct path_selector_type *dm_get_path_selector(const char *name) +{ + struct ps_internal *psi; + + if (!name) + return NULL; + + psi = get_path_selector(name); + if (!psi) { + request_module("dm-%s", name); + psi = get_path_selector(name); + } + + return psi ? &psi->pst : NULL; +} + +void dm_put_path_selector(struct path_selector_type *pst) +{ + struct ps_internal *psi; + + if (!pst) + return; + + down_read(&_ps_lock); + psi = __find_path_selector_type(pst->name); + if (!psi) + goto out; + + if (--psi->use == 0) + module_put(psi->pst.module); + + if (psi->use < 0) + BUG(); + +out: + up_read(&_ps_lock); +} + +static struct ps_internal *_alloc_path_selector(struct path_selector_type *pst) +{ + struct ps_internal *psi = kmalloc(sizeof(*psi), GFP_KERNEL); + + if (psi) { + memset(psi, 0, sizeof(*psi)); + psi->pst = *pst; + } + + return psi; +} + +int dm_register_path_selector(struct path_selector_type *pst) +{ + int r = 0; + struct ps_internal *psi = _alloc_path_selector(pst); + + if (!psi) + return -ENOMEM; + + down_write(&_ps_lock); + + if (__find_path_selector_type(pst->name)) { + kfree(psi); + r = -EEXIST; + } else + list_add(&psi->list, &_path_selectors); + + up_write(&_ps_lock); + + return r; +} + +int dm_unregister_path_selector(struct path_selector_type *pst) +{ + struct ps_internal *psi; + + down_write(&_ps_lock); + + psi = __find_path_selector_type(pst->name); + if (!psi) { + up_write(&_ps_lock); + return -EINVAL; + } + + if (psi->use) { + up_write(&_ps_lock); + return -ETXTBSY; + } + + list_del(&psi->list); + + up_write(&_ps_lock); + + kfree(psi); + + return 0; +} + +EXPORT_SYMBOL_GPL(dm_register_path_selector); +EXPORT_SYMBOL_GPL(dm_unregister_path_selector); diff --git a/drivers/md/dm-path-selector.h b/drivers/md/dm-path-selector.h new file mode 100644 index 00000000000..732d06a84f8 --- /dev/null +++ b/drivers/md/dm-path-selector.h @@ -0,0 +1,93 @@ +/* + * Copyright (C) 2003 Sistina Software. + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * Module Author: Heinz Mauelshagen + * + * This file is released under the GPL. + * + * Path-Selector registration. + */ + +#ifndef DM_PATH_SELECTOR_H +#define DM_PATH_SELECTOR_H + +#include <linux/device-mapper.h> + +#include "dm-mpath.h" + +/* + * We provide an abstraction for the code that chooses which path + * to send some io down. + */ +struct path_selector_type; +struct path_selector { + struct path_selector_type *type; + void *context; +}; + +/* Information about a path selector type */ +struct path_selector_type { + char *name; + struct module *module; + + unsigned int table_args; + unsigned int info_args; + + /* + * Constructs a path selector object, takes custom arguments + */ + int (*create) (struct path_selector *ps, unsigned argc, char **argv); + void (*destroy) (struct path_selector *ps); + + /* + * Add an opaque path object, along with some selector specific + * path args (eg, path priority). + */ + int (*add_path) (struct path_selector *ps, struct path *path, + int argc, char **argv, char **error); + + /* + * Chooses a path for this io, if no paths are available then + * NULL will be returned. + * + * repeat_count is the number of times to use the path before + * calling the function again. 0 means don't call it again unless + * the path fails. + */ + struct path *(*select_path) (struct path_selector *ps, + unsigned *repeat_count); + + /* + * Notify the selector that a path has failed. + */ + void (*fail_path) (struct path_selector *ps, struct path *p); + + /* + * Ask selector to reinstate a path. + */ + int (*reinstate_path) (struct path_selector *ps, struct path *p); + + /* + * Table content based on parameters added in ps_add_path_fn + * or path selector status + */ + int (*status) (struct path_selector *ps, struct path *path, + status_type_t type, char *result, unsigned int maxlen); + + int (*end_io) (struct path_selector *ps, struct path *path); +}; + +/* Register a path selector */ +int dm_register_path_selector(struct path_selector_type *type); + +/* Unregister a path selector */ +int dm_unregister_path_selector(struct path_selector_type *type); + +/* Returns a registered path selector type */ +struct path_selector_type *dm_get_path_selector(const char *name); + +/* Releases a path selector */ +void dm_put_path_selector(struct path_selector_type *pst); + +#endif diff --git a/drivers/md/dm-raid1.c b/drivers/md/dm-raid1.c new file mode 100644 index 00000000000..6e3cf7e1345 --- /dev/null +++ b/drivers/md/dm-raid1.c @@ -0,0 +1,1269 @@ +/* + * Copyright (C) 2003 Sistina Software Limited. + * + * This file is released under the GPL. + */ + +#include "dm.h" +#include "dm-bio-list.h" +#include "dm-io.h" +#include "dm-log.h" +#include "kcopyd.h" + +#include <linux/ctype.h> +#include <linux/init.h> +#include <linux/mempool.h> +#include <linux/module.h> +#include <linux/pagemap.h> +#include <linux/slab.h> +#include <linux/time.h> +#include <linux/vmalloc.h> +#include <linux/workqueue.h> + +static struct workqueue_struct *_kmirrord_wq; +static struct work_struct _kmirrord_work; + +static inline void wake(void) +{ + queue_work(_kmirrord_wq, &_kmirrord_work); +} + +/*----------------------------------------------------------------- + * Region hash + * + * The mirror splits itself up into discrete regions. Each + * region can be in one of three states: clean, dirty, + * nosync. There is no need to put clean regions in the hash. + * + * In addition to being present in the hash table a region _may_ + * be present on one of three lists. + * + * clean_regions: Regions on this list have no io pending to + * them, they are in sync, we are no longer interested in them, + * they are dull. rh_update_states() will remove them from the + * hash table. + * + * quiesced_regions: These regions have been spun down, ready + * for recovery. rh_recovery_start() will remove regions from + * this list and hand them to kmirrord, which will schedule the + * recovery io with kcopyd. + * + * recovered_regions: Regions that kcopyd has successfully + * recovered. rh_update_states() will now schedule any delayed + * io, up the recovery_count, and remove the region from the + * hash. + * + * There are 2 locks: + * A rw spin lock 'hash_lock' protects just the hash table, + * this is never held in write mode from interrupt context, + * which I believe means that we only have to disable irqs when + * doing a write lock. + * + * An ordinary spin lock 'region_lock' that protects the three + * lists in the region_hash, with the 'state', 'list' and + * 'bhs_delayed' fields of the regions. This is used from irq + * context, so all other uses will have to suspend local irqs. + *---------------------------------------------------------------*/ +struct mirror_set; +struct region_hash { + struct mirror_set *ms; + uint32_t region_size; + unsigned region_shift; + + /* holds persistent region state */ + struct dirty_log *log; + + /* hash table */ + rwlock_t hash_lock; + mempool_t *region_pool; + unsigned int mask; + unsigned int nr_buckets; + struct list_head *buckets; + + spinlock_t region_lock; + struct semaphore recovery_count; + struct list_head clean_regions; + struct list_head quiesced_regions; + struct list_head recovered_regions; +}; + +enum { + RH_CLEAN, + RH_DIRTY, + RH_NOSYNC, + RH_RECOVERING +}; + +struct region { + struct region_hash *rh; /* FIXME: can we get rid of this ? */ + region_t key; + int state; + + struct list_head hash_list; + struct list_head list; + + atomic_t pending; + struct bio_list delayed_bios; +}; + +/* + * Conversion fns + */ +static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio) +{ + return bio->bi_sector >> rh->region_shift; +} + +static inline sector_t region_to_sector(struct region_hash *rh, region_t region) +{ + return region << rh->region_shift; +} + +/* FIXME move this */ +static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw); + +static void *region_alloc(unsigned int __nocast gfp_mask, void *pool_data) +{ + return kmalloc(sizeof(struct region), gfp_mask); +} + +static void region_free(void *element, void *pool_data) +{ + kfree(element); +} + +#define MIN_REGIONS 64 +#define MAX_RECOVERY 1 +static int rh_init(struct region_hash *rh, struct mirror_set *ms, + struct dirty_log *log, uint32_t region_size, + region_t nr_regions) +{ + unsigned int nr_buckets, max_buckets; + size_t i; + + /* + * Calculate a suitable number of buckets for our hash + * table. + */ + max_buckets = nr_regions >> 6; + for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1) + ; + nr_buckets >>= 1; + + rh->ms = ms; + rh->log = log; + rh->region_size = region_size; + rh->region_shift = ffs(region_size) - 1; + rwlock_init(&rh->hash_lock); + rh->mask = nr_buckets - 1; + rh->nr_buckets = nr_buckets; + + rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets)); + if (!rh->buckets) { + DMERR("unable to allocate region hash memory"); + return -ENOMEM; + } + + for (i = 0; i < nr_buckets; i++) + INIT_LIST_HEAD(rh->buckets + i); + + spin_lock_init(&rh->region_lock); + sema_init(&rh->recovery_count, 0); + INIT_LIST_HEAD(&rh->clean_regions); + INIT_LIST_HEAD(&rh->quiesced_regions); + INIT_LIST_HEAD(&rh->recovered_regions); + + rh->region_pool = mempool_create(MIN_REGIONS, region_alloc, + region_free, NULL); + if (!rh->region_pool) { + vfree(rh->buckets); + rh->buckets = NULL; + return -ENOMEM; + } + + return 0; +} + +static void rh_exit(struct region_hash *rh) +{ + unsigned int h; + struct region *reg, *nreg; + + BUG_ON(!list_empty(&rh->quiesced_regions)); + for (h = 0; h < rh->nr_buckets; h++) { + list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) { + BUG_ON(atomic_read(®->pending)); + mempool_free(reg, rh->region_pool); + } + } + + if (rh->log) + dm_destroy_dirty_log(rh->log); + if (rh->region_pool) + mempool_destroy(rh->region_pool); + vfree(rh->buckets); +} + +#define RH_HASH_MULT 2654435387U + +static inline unsigned int rh_hash(struct region_hash *rh, region_t region) +{ + return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask; +} + +static struct region *__rh_lookup(struct region_hash *rh, region_t region) +{ + struct region *reg; + + list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list) + if (reg->key == region) + return reg; + + return NULL; +} + +static void __rh_insert(struct region_hash *rh, struct region *reg) +{ + unsigned int h = rh_hash(rh, reg->key); + list_add(®->hash_list, rh->buckets + h); +} + +static struct region *__rh_alloc(struct region_hash *rh, region_t region) +{ + struct region *reg, *nreg; + + read_unlock(&rh->hash_lock); + nreg = mempool_alloc(rh->region_pool, GFP_NOIO); + nreg->state = rh->log->type->in_sync(rh->log, region, 1) ? + RH_CLEAN : RH_NOSYNC; + nreg->rh = rh; + nreg->key = region; + + INIT_LIST_HEAD(&nreg->list); + + atomic_set(&nreg->pending, 0); + bio_list_init(&nreg->delayed_bios); + write_lock_irq(&rh->hash_lock); + + reg = __rh_lookup(rh, region); + if (reg) + /* we lost the race */ + mempool_free(nreg, rh->region_pool); + + else { + __rh_insert(rh, nreg); + if (nreg->state == RH_CLEAN) { + spin_lock(&rh->region_lock); + list_add(&nreg->list, &rh->clean_regions); + spin_unlock(&rh->region_lock); + } + reg = nreg; + } + write_unlock_irq(&rh->hash_lock); + read_lock(&rh->hash_lock); + + return reg; +} + +static inline struct region *__rh_find(struct region_hash *rh, region_t region) +{ + struct region *reg; + + reg = __rh_lookup(rh, region); + if (!reg) + reg = __rh_alloc(rh, region); + + return reg; +} + +static int rh_state(struct region_hash *rh, region_t region, int may_block) +{ + int r; + struct region *reg; + + read_lock(&rh->hash_lock); + reg = __rh_lookup(rh, region); + read_unlock(&rh->hash_lock); + + if (reg) + return reg->state; + + /* + * The region wasn't in the hash, so we fall back to the + * dirty log. + */ + r = rh->log->type->in_sync(rh->log, region, may_block); + + /* + * Any error from the dirty log (eg. -EWOULDBLOCK) gets + * taken as a RH_NOSYNC + */ + return r == 1 ? RH_CLEAN : RH_NOSYNC; +} + +static inline int rh_in_sync(struct region_hash *rh, + region_t region, int may_block) +{ + int state = rh_state(rh, region, may_block); + return state == RH_CLEAN || state == RH_DIRTY; +} + +static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list) +{ + struct bio *bio; + + while ((bio = bio_list_pop(bio_list))) { + queue_bio(ms, bio, WRITE); + } +} + +static void rh_update_states(struct region_hash *rh) +{ + struct region *reg, *next; + + LIST_HEAD(clean); + LIST_HEAD(recovered); + + /* + * Quickly grab the lists. + */ + write_lock_irq(&rh->hash_lock); + spin_lock(&rh->region_lock); + if (!list_empty(&rh->clean_regions)) { + list_splice(&rh->clean_regions, &clean); + INIT_LIST_HEAD(&rh->clean_regions); + + list_for_each_entry (reg, &clean, list) { + rh->log->type->clear_region(rh->log, reg->key); + list_del(®->hash_list); + } + } + + if (!list_empty(&rh->recovered_regions)) { + list_splice(&rh->recovered_regions, &recovered); + INIT_LIST_HEAD(&rh->recovered_regions); + + list_for_each_entry (reg, &recovered, list) + list_del(®->hash_list); + } + spin_unlock(&rh->region_lock); + write_unlock_irq(&rh->hash_lock); + + /* + * All the regions on the recovered and clean lists have + * now been pulled out of the system, so no need to do + * any more locking. + */ + list_for_each_entry_safe (reg, next, &recovered, list) { + rh->log->type->clear_region(rh->log, reg->key); + rh->log->type->complete_resync_work(rh->log, reg->key, 1); + dispatch_bios(rh->ms, ®->delayed_bios); + up(&rh->recovery_count); + mempool_free(reg, rh->region_pool); + } + + if (!list_empty(&recovered)) + rh->log->type->flush(rh->log); + + list_for_each_entry_safe (reg, next, &clean, list) + mempool_free(reg, rh->region_pool); +} + +static void rh_inc(struct region_hash *rh, region_t region) +{ + struct region *reg; + + read_lock(&rh->hash_lock); + reg = __rh_find(rh, region); + if (reg->state == RH_CLEAN) { + rh->log->type->mark_region(rh->log, reg->key); + + spin_lock_irq(&rh->region_lock); + reg->state = RH_DIRTY; + list_del_init(®->list); /* take off the clean list */ + spin_unlock_irq(&rh->region_lock); + } + + atomic_inc(®->pending); + read_unlock(&rh->hash_lock); +} + +static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios) +{ + struct bio *bio; + + for (bio = bios->head; bio; bio = bio->bi_next) + rh_inc(rh, bio_to_region(rh, bio)); +} + +static void rh_dec(struct region_hash *rh, region_t region) +{ + unsigned long flags; + struct region *reg; + int should_wake = 0; + + read_lock(&rh->hash_lock); + reg = __rh_lookup(rh, region); + read_unlock(&rh->hash_lock); + + if (atomic_dec_and_test(®->pending)) { + spin_lock_irqsave(&rh->region_lock, flags); + if (reg->state == RH_RECOVERING) { + list_add_tail(®->list, &rh->quiesced_regions); + } else { + reg->state = RH_CLEAN; + list_add(®->list, &rh->clean_regions); + } + spin_unlock_irqrestore(&rh->region_lock, flags); + should_wake = 1; + } + + if (should_wake) + wake(); +} + +/* + * Starts quiescing a region in preparation for recovery. + */ +static int __rh_recovery_prepare(struct region_hash *rh) +{ + int r; + struct region *reg; + region_t region; + + /* + * Ask the dirty log what's next. + */ + r = rh->log->type->get_resync_work(rh->log, ®ion); + if (r <= 0) + return r; + + /* + * Get this region, and start it quiescing by setting the + * recovering flag. + */ + read_lock(&rh->hash_lock); + reg = __rh_find(rh, region); + read_unlock(&rh->hash_lock); + + spin_lock_irq(&rh->region_lock); + reg->state = RH_RECOVERING; + + /* Already quiesced ? */ + if (atomic_read(®->pending)) + list_del_init(®->list); + + else { + list_del_init(®->list); + list_add(®->list, &rh->quiesced_regions); + } + spin_unlock_irq(&rh->region_lock); + + return 1; +} + +static void rh_recovery_prepare(struct region_hash *rh) +{ + while (!down_trylock(&rh->recovery_count)) + if (__rh_recovery_prepare(rh) <= 0) { + up(&rh->recovery_count); + break; + } +} + +/* + * Returns any quiesced regions. + */ +static struct region *rh_recovery_start(struct region_hash *rh) +{ + struct region *reg = NULL; + + spin_lock_irq(&rh->region_lock); + if (!list_empty(&rh->quiesced_regions)) { + reg = list_entry(rh->quiesced_regions.next, + struct region, list); + list_del_init(®->list); /* remove from the quiesced list */ + } + spin_unlock_irq(&rh->region_lock); + + return reg; +} + +/* FIXME: success ignored for now */ +static void rh_recovery_end(struct region *reg, int success) +{ + struct region_hash *rh = reg->rh; + + spin_lock_irq(&rh->region_lock); + list_add(®->list, ®->rh->recovered_regions); + spin_unlock_irq(&rh->region_lock); + + wake(); +} + +static void rh_flush(struct region_hash *rh) +{ + rh->log->type->flush(rh->log); +} + +static void rh_delay(struct region_hash *rh, struct bio *bio) +{ + struct region *reg; + + read_lock(&rh->hash_lock); + reg = __rh_find(rh, bio_to_region(rh, bio)); + bio_list_add(®->delayed_bios, bio); + read_unlock(&rh->hash_lock); +} + +static void rh_stop_recovery(struct region_hash *rh) +{ + int i; + + /* wait for any recovering regions */ + for (i = 0; i < MAX_RECOVERY; i++) + down(&rh->recovery_count); +} + +static void rh_start_recovery(struct region_hash *rh) +{ + int i; + + for (i = 0; i < MAX_RECOVERY; i++) + up(&rh->recovery_count); + + wake(); +} + +/*----------------------------------------------------------------- + * Mirror set structures. + *---------------------------------------------------------------*/ +struct mirror { + atomic_t error_count; + struct dm_dev *dev; + sector_t offset; +}; + +struct mirror_set { + struct dm_target *ti; + struct list_head list; + struct region_hash rh; + struct kcopyd_client *kcopyd_client; + + spinlock_t lock; /* protects the next two lists */ + struct bio_list reads; + struct bio_list writes; + + /* recovery */ + region_t nr_regions; + int in_sync; + + unsigned int nr_mirrors; + struct mirror mirror[0]; +}; + +/* + * Every mirror should look like this one. + */ +#define DEFAULT_MIRROR 0 + +/* + * This is yucky. We squirrel the mirror_set struct away inside + * bi_next for write buffers. This is safe since the bh + * doesn't get submitted to the lower levels of block layer. + */ +static struct mirror_set *bio_get_ms(struct bio *bio) +{ + return (struct mirror_set *) bio->bi_next; +} + +static void bio_set_ms(struct bio *bio, struct mirror_set *ms) +{ + bio->bi_next = (struct bio *) ms; +} + +/*----------------------------------------------------------------- + * Recovery. + * + * When a mirror is first activated we may find that some regions + * are in the no-sync state. We have to recover these by + * recopying from the default mirror to all the others. + *---------------------------------------------------------------*/ +static void recovery_complete(int read_err, unsigned int write_err, + void *context) +{ + struct region *reg = (struct region *) context; + + /* FIXME: better error handling */ + rh_recovery_end(reg, read_err || write_err); +} + +static int recover(struct mirror_set *ms, struct region *reg) +{ + int r; + unsigned int i; + struct io_region from, to[KCOPYD_MAX_REGIONS], *dest; + struct mirror *m; + unsigned long flags = 0; + + /* fill in the source */ + m = ms->mirror + DEFAULT_MIRROR; + from.bdev = m->dev->bdev; + from.sector = m->offset + region_to_sector(reg->rh, reg->key); + if (reg->key == (ms->nr_regions - 1)) { + /* + * The final region may be smaller than + * region_size. + */ + from.count = ms->ti->len & (reg->rh->region_size - 1); + if (!from.count) + from.count = reg->rh->region_size; + } else + from.count = reg->rh->region_size; + + /* fill in the destinations */ + for (i = 0, dest = to; i < ms->nr_mirrors; i++) { + if (i == DEFAULT_MIRROR) + continue; + + m = ms->mirror + i; + dest->bdev = m->dev->bdev; + dest->sector = m->offset + region_to_sector(reg->rh, reg->key); + dest->count = from.count; + dest++; + } + + /* hand to kcopyd */ + set_bit(KCOPYD_IGNORE_ERROR, &flags); + r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags, + recovery_complete, reg); + + return r; +} + +static void do_recovery(struct mirror_set *ms) +{ + int r; + struct region *reg; + struct dirty_log *log = ms->rh.log; + + /* + * Start quiescing some regions. + */ + rh_recovery_prepare(&ms->rh); + + /* + * Copy any already quiesced regions. + */ + while ((reg = rh_recovery_start(&ms->rh))) { + r = recover(ms, reg); + if (r) + rh_recovery_end(reg, 0); + } + + /* + * Update the in sync flag. + */ + if (!ms->in_sync && + (log->type->get_sync_count(log) == ms->nr_regions)) { + /* the sync is complete */ + dm_table_event(ms->ti->table); + ms->in_sync = 1; + } +} + +/*----------------------------------------------------------------- + * Reads + *---------------------------------------------------------------*/ +static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) +{ + /* FIXME: add read balancing */ + return ms->mirror + DEFAULT_MIRROR; +} + +/* + * remap a buffer to a particular mirror. + */ +static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio) +{ + bio->bi_bdev = m->dev->bdev; + bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin); +} + +static void do_reads(struct mirror_set *ms, struct bio_list *reads) +{ + region_t region; + struct bio *bio; + struct mirror *m; + + while ((bio = bio_list_pop(reads))) { + region = bio_to_region(&ms->rh, bio); + + /* + * We can only read balance if the region is in sync. + */ + if (rh_in_sync(&ms->rh, region, 0)) + m = choose_mirror(ms, bio->bi_sector); + else + m = ms->mirror + DEFAULT_MIRROR; + + map_bio(ms, m, bio); + generic_make_request(bio); + } +} + +/*----------------------------------------------------------------- + * Writes. + * + * We do different things with the write io depending on the + * state of the region that it's in: + * + * SYNC: increment pending, use kcopyd to write to *all* mirrors + * RECOVERING: delay the io until recovery completes + * NOSYNC: increment pending, just write to the default mirror + *---------------------------------------------------------------*/ +static void write_callback(unsigned long error, void *context) +{ + unsigned int i; + int uptodate = 1; + struct bio *bio = (struct bio *) context; + struct mirror_set *ms; + + ms = bio_get_ms(bio); + bio_set_ms(bio, NULL); + + /* + * NOTE: We don't decrement the pending count here, + * instead it is done by the targets endio function. + * This way we handle both writes to SYNC and NOSYNC + * regions with the same code. + */ + + if (error) { + /* + * only error the io if all mirrors failed. + * FIXME: bogus + */ + uptodate = 0; + for (i = 0; i < ms->nr_mirrors; i++) + if (!test_bit(i, &error)) { + uptodate = 1; + break; + } + } + bio_endio(bio, bio->bi_size, 0); +} + +static void do_write(struct mirror_set *ms, struct bio *bio) +{ + unsigned int i; + struct io_region io[KCOPYD_MAX_REGIONS+1]; + struct mirror *m; + + for (i = 0; i < ms->nr_mirrors; i++) { + m = ms->mirror + i; + + io[i].bdev = m->dev->bdev; + io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin); + io[i].count = bio->bi_size >> 9; + } + + bio_set_ms(bio, ms); + dm_io_async_bvec(ms->nr_mirrors, io, WRITE, + bio->bi_io_vec + bio->bi_idx, + write_callback, bio); +} + +static void do_writes(struct mirror_set *ms, struct bio_list *writes) +{ + int state; + struct bio *bio; + struct bio_list sync, nosync, recover, *this_list = NULL; + + if (!writes->head) + return; + + /* + * Classify each write. + */ + bio_list_init(&sync); + bio_list_init(&nosync); + bio_list_init(&recover); + + while ((bio = bio_list_pop(writes))) { + state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1); + switch (state) { + case RH_CLEAN: + case RH_DIRTY: + this_list = &sync; + break; + + case RH_NOSYNC: + this_list = &nosync; + break; + + case RH_RECOVERING: + this_list = &recover; + break; + } + + bio_list_add(this_list, bio); + } + + /* + * Increment the pending counts for any regions that will + * be written to (writes to recover regions are going to + * be delayed). + */ + rh_inc_pending(&ms->rh, &sync); + rh_inc_pending(&ms->rh, &nosync); + rh_flush(&ms->rh); + + /* + * Dispatch io. + */ + while ((bio = bio_list_pop(&sync))) + do_write(ms, bio); + + while ((bio = bio_list_pop(&recover))) + rh_delay(&ms->rh, bio); + + while ((bio = bio_list_pop(&nosync))) { + map_bio(ms, ms->mirror + DEFAULT_MIRROR, bio); + generic_make_request(bio); + } +} + +/*----------------------------------------------------------------- + * kmirrord + *---------------------------------------------------------------*/ +static LIST_HEAD(_mirror_sets); +static DECLARE_RWSEM(_mirror_sets_lock); + +static void do_mirror(struct mirror_set *ms) +{ + struct bio_list reads, writes; + + spin_lock(&ms->lock); + reads = ms->reads; + writes = ms->writes; + bio_list_init(&ms->reads); + bio_list_init(&ms->writes); + spin_unlock(&ms->lock); + + rh_update_states(&ms->rh); + do_recovery(ms); + do_reads(ms, &reads); + do_writes(ms, &writes); +} + +static void do_work(void *ignored) +{ + struct mirror_set *ms; + + down_read(&_mirror_sets_lock); + list_for_each_entry (ms, &_mirror_sets, list) + do_mirror(ms); + up_read(&_mirror_sets_lock); +} + +/*----------------------------------------------------------------- + * Target functions + *---------------------------------------------------------------*/ +static struct mirror_set *alloc_context(unsigned int nr_mirrors, + uint32_t region_size, + struct dm_target *ti, + struct dirty_log *dl) +{ + size_t len; + struct mirror_set *ms = NULL; + + if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors)) + return NULL; + + len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); + + ms = kmalloc(len, GFP_KERNEL); + if (!ms) { + ti->error = "dm-mirror: Cannot allocate mirror context"; + return NULL; + } + + memset(ms, 0, len); + spin_lock_init(&ms->lock); + + ms->ti = ti; + ms->nr_mirrors = nr_mirrors; + ms->nr_regions = dm_sector_div_up(ti->len, region_size); + ms->in_sync = 0; + + if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) { + ti->error = "dm-mirror: Error creating dirty region hash"; + kfree(ms); + return NULL; + } + + return ms; +} + +static void free_context(struct mirror_set *ms, struct dm_target *ti, + unsigned int m) +{ + while (m--) + dm_put_device(ti, ms->mirror[m].dev); + + rh_exit(&ms->rh); + kfree(ms); +} + +static inline int _check_region_size(struct dm_target *ti, uint32_t size) +{ + return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) || + size > ti->len); +} + +static int get_mirror(struct mirror_set *ms, struct dm_target *ti, + unsigned int mirror, char **argv) +{ + sector_t offset; + + if (sscanf(argv[1], SECTOR_FORMAT, &offset) != 1) { + ti->error = "dm-mirror: Invalid offset"; + return -EINVAL; + } + + if (dm_get_device(ti, argv[0], offset, ti->len, + dm_table_get_mode(ti->table), + &ms->mirror[mirror].dev)) { + ti->error = "dm-mirror: Device lookup failure"; + return -ENXIO; + } + + ms->mirror[mirror].offset = offset; + + return 0; +} + +static int add_mirror_set(struct mirror_set *ms) +{ + down_write(&_mirror_sets_lock); + list_add_tail(&ms->list, &_mirror_sets); + up_write(&_mirror_sets_lock); + wake(); + + return 0; +} + +static void del_mirror_set(struct mirror_set *ms) +{ + down_write(&_mirror_sets_lock); + list_del(&ms->list); + up_write(&_mirror_sets_lock); +} + +/* + * Create dirty log: log_type #log_params <log_params> + */ +static struct dirty_log *create_dirty_log(struct dm_target *ti, + unsigned int argc, char **argv, + unsigned int *args_used) +{ + unsigned int param_count; + struct dirty_log *dl; + + if (argc < 2) { + ti->error = "dm-mirror: Insufficient mirror log arguments"; + return NULL; + } + + if (sscanf(argv[1], "%u", ¶m_count) != 1) { + ti->error = "dm-mirror: Invalid mirror log argument count"; + return NULL; + } + + *args_used = 2 + param_count; + + if (argc < *args_used) { + ti->error = "dm-mirror: Insufficient mirror log arguments"; + return NULL; + } + + dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2); + if (!dl) { + ti->error = "dm-mirror: Error creating mirror dirty log"; + return NULL; + } + + if (!_check_region_size(ti, dl->type->get_region_size(dl))) { + ti->error = "dm-mirror: Invalid region size"; + dm_destroy_dirty_log(dl); + return NULL; + } + + return dl; +} + +/* + * Construct a mirror mapping: + * + * log_type #log_params <log_params> + * #mirrors [mirror_path offset]{2,} + * + * log_type is "core" or "disk" + * #log_params is between 1 and 3 + */ +#define DM_IO_PAGES 64 +static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + int r; + unsigned int nr_mirrors, m, args_used; + struct mirror_set *ms; + struct dirty_log *dl; + + dl = create_dirty_log(ti, argc, argv, &args_used); + if (!dl) + return -EINVAL; + + argv += args_used; + argc -= args_used; + + if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || + nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) { + ti->error = "dm-mirror: Invalid number of mirrors"; + dm_destroy_dirty_log(dl); + return -EINVAL; + } + + argv++, argc--; + + if (argc != nr_mirrors * 2) { + ti->error = "dm-mirror: Wrong number of mirror arguments"; + dm_destroy_dirty_log(dl); + return -EINVAL; + } + + ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); + if (!ms) { + dm_destroy_dirty_log(dl); + return -ENOMEM; + } + + /* Get the mirror parameter sets */ + for (m = 0; m < nr_mirrors; m++) { + r = get_mirror(ms, ti, m, argv); + if (r) { + free_context(ms, ti, m); + return r; + } + argv += 2; + argc -= 2; + } + + ti->private = ms; + + r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client); + if (r) { + free_context(ms, ti, ms->nr_mirrors); + return r; + } + + add_mirror_set(ms); + return 0; +} + +static void mirror_dtr(struct dm_target *ti) +{ + struct mirror_set *ms = (struct mirror_set *) ti->private; + + del_mirror_set(ms); + kcopyd_client_destroy(ms->kcopyd_client); + free_context(ms, ti, ms->nr_mirrors); +} + +static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) +{ + int should_wake = 0; + struct bio_list *bl; + + bl = (rw == WRITE) ? &ms->writes : &ms->reads; + spin_lock(&ms->lock); + should_wake = !(bl->head); + bio_list_add(bl, bio); + spin_unlock(&ms->lock); + + if (should_wake) + wake(); +} + +/* + * Mirror mapping function + */ +static int mirror_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + int r, rw = bio_rw(bio); + struct mirror *m; + struct mirror_set *ms = ti->private; + + map_context->ll = bio->bi_sector >> ms->rh.region_shift; + + if (rw == WRITE) { + queue_bio(ms, bio, rw); + return 0; + } + + r = ms->rh.log->type->in_sync(ms->rh.log, + bio_to_region(&ms->rh, bio), 0); + if (r < 0 && r != -EWOULDBLOCK) + return r; + + if (r == -EWOULDBLOCK) /* FIXME: ugly */ + r = 0; + + /* + * We don't want to fast track a recovery just for a read + * ahead. So we just let it silently fail. + * FIXME: get rid of this. + */ + if (!r && rw == READA) + return -EIO; + + if (!r) { + /* Pass this io over to the daemon */ + queue_bio(ms, bio, rw); + return 0; + } + + m = choose_mirror(ms, bio->bi_sector); + if (!m) + return -EIO; + + map_bio(ms, m, bio); + return 1; +} + +static int mirror_end_io(struct dm_target *ti, struct bio *bio, + int error, union map_info *map_context) +{ + int rw = bio_rw(bio); + struct mirror_set *ms = (struct mirror_set *) ti->private; + region_t region = map_context->ll; + + /* + * We need to dec pending if this was a write. + */ + if (rw == WRITE) + rh_dec(&ms->rh, region); + + return 0; +} + +static void mirror_postsuspend(struct dm_target *ti) +{ + struct mirror_set *ms = (struct mirror_set *) ti->private; + struct dirty_log *log = ms->rh.log; + + rh_stop_recovery(&ms->rh); + if (log->type->suspend && log->type->suspend(log)) + /* FIXME: need better error handling */ + DMWARN("log suspend failed"); +} + +static void mirror_resume(struct dm_target *ti) +{ + struct mirror_set *ms = (struct mirror_set *) ti->private; + struct dirty_log *log = ms->rh.log; + if (log->type->resume && log->type->resume(log)) + /* FIXME: need better error handling */ + DMWARN("log resume failed"); + rh_start_recovery(&ms->rh); +} + +static int mirror_status(struct dm_target *ti, status_type_t type, + char *result, unsigned int maxlen) +{ + unsigned int m, sz; + struct mirror_set *ms = (struct mirror_set *) ti->private; + + sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen); + + switch (type) { + case STATUSTYPE_INFO: + DMEMIT("%d ", ms->nr_mirrors); + for (m = 0; m < ms->nr_mirrors; m++) + DMEMIT("%s ", ms->mirror[m].dev->name); + + DMEMIT(SECTOR_FORMAT "/" SECTOR_FORMAT, + ms->rh.log->type->get_sync_count(ms->rh.log), + ms->nr_regions); + break; + + case STATUSTYPE_TABLE: + DMEMIT("%d ", ms->nr_mirrors); + for (m = 0; m < ms->nr_mirrors; m++) + DMEMIT("%s " SECTOR_FORMAT " ", + ms->mirror[m].dev->name, ms->mirror[m].offset); + } + + return 0; +} + +static struct target_type mirror_target = { + .name = "mirror", + .version = {1, 0, 1}, + .module = THIS_MODULE, + .ctr = mirror_ctr, + .dtr = mirror_dtr, + .map = mirror_map, + .end_io = mirror_end_io, + .postsuspend = mirror_postsuspend, + .resume = mirror_resume, + .status = mirror_status, +}; + +static int __init dm_mirror_init(void) +{ + int r; + + r = dm_dirty_log_init(); + if (r) + return r; + + _kmirrord_wq = create_workqueue("kmirrord"); + if (!_kmirrord_wq) { + DMERR("couldn't start kmirrord"); + dm_dirty_log_exit(); + return r; + } + INIT_WORK(&_kmirrord_work, do_work, NULL); + + r = dm_register_target(&mirror_target); + if (r < 0) { + DMERR("%s: Failed to register mirror target", + mirror_target.name); + dm_dirty_log_exit(); + destroy_workqueue(_kmirrord_wq); + } + + return r; +} + +static void __exit dm_mirror_exit(void) +{ + int r; + + r = dm_unregister_target(&mirror_target); + if (r < 0) + DMERR("%s: unregister failed %d", mirror_target.name, r); + + destroy_workqueue(_kmirrord_wq); + dm_dirty_log_exit(); +} + +/* Module hooks */ +module_init(dm_mirror_init); +module_exit(dm_mirror_exit); + +MODULE_DESCRIPTION(DM_NAME " mirror target"); +MODULE_AUTHOR("Joe Thornber"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-round-robin.c b/drivers/md/dm-round-robin.c new file mode 100644 index 00000000000..d0024865a78 --- /dev/null +++ b/drivers/md/dm-round-robin.c @@ -0,0 +1,214 @@ +/* + * Copyright (C) 2003 Sistina Software. + * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. + * + * Module Author: Heinz Mauelshagen + * + * This file is released under the GPL. + * + * Round-robin path selector. + */ + +#include "dm.h" +#include "dm-path-selector.h" + +#include <linux/slab.h> + +/*----------------------------------------------------------------- + * Path-handling code, paths are held in lists + *---------------------------------------------------------------*/ +struct path_info { + struct list_head list; + struct path *path; + unsigned repeat_count; +}; + +static void free_paths(struct list_head *paths) +{ + struct path_info *pi, *next; + + list_for_each_entry_safe(pi, next, paths, list) { + list_del(&pi->list); + kfree(pi); + } +} + +/*----------------------------------------------------------------- + * Round-robin selector + *---------------------------------------------------------------*/ + +#define RR_MIN_IO 1000 + +struct selector { + struct list_head valid_paths; + struct list_head invalid_paths; +}; + +static struct selector *alloc_selector(void) +{ + struct selector *s = kmalloc(sizeof(*s), GFP_KERNEL); + + if (s) { + INIT_LIST_HEAD(&s->valid_paths); + INIT_LIST_HEAD(&s->invalid_paths); + } + + return s; +} + +static int rr_create(struct path_selector *ps, unsigned argc, char **argv) +{ + struct selector *s; + + s = alloc_selector(); + if (!s) + return -ENOMEM; + + ps->context = s; + return 0; +} + +static void rr_destroy(struct path_selector *ps) +{ + struct selector *s = (struct selector *) ps->context; + + free_paths(&s->valid_paths); + free_paths(&s->invalid_paths); + kfree(s); + ps->context = NULL; +} + +static int rr_status(struct path_selector *ps, struct path *path, + status_type_t type, char *result, unsigned int maxlen) +{ + struct path_info *pi; + int sz = 0; + + if (!path) + DMEMIT("0 "); + else { + switch(type) { + case STATUSTYPE_INFO: + break; + case STATUSTYPE_TABLE: + pi = path->pscontext; + DMEMIT("%u ", pi->repeat_count); + break; + } + } + + return sz; +} + +/* + * Called during initialisation to register each path with an + * optional repeat_count. + */ +static int rr_add_path(struct path_selector *ps, struct path *path, + int argc, char **argv, char **error) +{ + struct selector *s = (struct selector *) ps->context; + struct path_info *pi; + unsigned repeat_count = RR_MIN_IO; + + if (argc > 1) { + *error = "round-robin ps: incorrect number of arguments"; + return -EINVAL; + } + + /* First path argument is number of I/Os before switching path */ + if ((argc == 1) && (sscanf(argv[0], "%u", &repeat_count) != 1)) { + *error = "round-robin ps: invalid repeat count"; + return -EINVAL; + } + + /* allocate the path */ + pi = kmalloc(sizeof(*pi), GFP_KERNEL); + if (!pi) { + *error = "round-robin ps: Error allocating path context"; + return -ENOMEM; + } + + pi->path = path; + pi->repeat_count = repeat_count; + + path->pscontext = pi; + + list_add(&pi->list, &s->valid_paths); + + return 0; +} + +static void rr_fail_path(struct path_selector *ps, struct path *p) +{ + struct selector *s = (struct selector *) ps->context; + struct path_info *pi = p->pscontext; + + list_move(&pi->list, &s->invalid_paths); +} + +static int rr_reinstate_path(struct path_selector *ps, struct path *p) +{ + struct selector *s = (struct selector *) ps->context; + struct path_info *pi = p->pscontext; + + list_move(&pi->list, &s->valid_paths); + + return 0; +} + +static struct path *rr_select_path(struct path_selector *ps, + unsigned *repeat_count) +{ + struct selector *s = (struct selector *) ps->context; + struct path_info *pi = NULL; + + if (!list_empty(&s->valid_paths)) { + pi = list_entry(s->valid_paths.next, struct path_info, list); + list_move_tail(&pi->list, &s->valid_paths); + *repeat_count = pi->repeat_count; + } + + return pi ? pi->path : NULL; +} + +static struct path_selector_type rr_ps = { + .name = "round-robin", + .module = THIS_MODULE, + .table_args = 1, + .info_args = 0, + .create = rr_create, + .destroy = rr_destroy, + .status = rr_status, + .add_path = rr_add_path, + .fail_path = rr_fail_path, + .reinstate_path = rr_reinstate_path, + .select_path = rr_select_path, +}; + +static int __init dm_rr_init(void) +{ + int r = dm_register_path_selector(&rr_ps); + + if (r < 0) + DMERR("round-robin: register failed %d", r); + + DMINFO("dm-round-robin version 1.0.0 loaded"); + + return r; +} + +static void __exit dm_rr_exit(void) +{ + int r = dm_unregister_path_selector(&rr_ps); + + if (r < 0) + DMERR("round-robin: unregister failed %d", r); +} + +module_init(dm_rr_init); +module_exit(dm_rr_exit); + +MODULE_DESCRIPTION(DM_NAME " round-robin multipath path selector"); +MODULE_AUTHOR("Sistina Software <dm-devel@redhat.com>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-snap.c b/drivers/md/dm-snap.c new file mode 100644 index 00000000000..7e691ab9a74 --- /dev/null +++ b/drivers/md/dm-snap.c @@ -0,0 +1,1208 @@ +/* + * dm-snapshot.c + * + * Copyright (C) 2001-2002 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + */ + +#include <linux/blkdev.h> +#include <linux/config.h> +#include <linux/ctype.h> +#include <linux/device-mapper.h> +#include <linux/fs.h> +#include <linux/init.h> +#include <linux/kdev_t.h> +#include <linux/list.h> +#include <linux/mempool.h> +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/vmalloc.h> + +#include "dm-snap.h" +#include "dm-bio-list.h" +#include "kcopyd.h" + +/* + * The percentage increment we will wake up users at + */ +#define WAKE_UP_PERCENT 5 + +/* + * kcopyd priority of snapshot operations + */ +#define SNAPSHOT_COPY_PRIORITY 2 + +/* + * Each snapshot reserves this many pages for io + */ +#define SNAPSHOT_PAGES 256 + +struct pending_exception { + struct exception e; + + /* + * Origin buffers waiting for this to complete are held + * in a bio list + */ + struct bio_list origin_bios; + struct bio_list snapshot_bios; + + /* + * Other pending_exceptions that are processing this + * chunk. When this list is empty, we know we can + * complete the origins. + */ + struct list_head siblings; + + /* Pointer back to snapshot context */ + struct dm_snapshot *snap; + + /* + * 1 indicates the exception has already been sent to + * kcopyd. + */ + int started; +}; + +/* + * Hash table mapping origin volumes to lists of snapshots and + * a lock to protect it + */ +static kmem_cache_t *exception_cache; +static kmem_cache_t *pending_cache; +static mempool_t *pending_pool; + +/* + * One of these per registered origin, held in the snapshot_origins hash + */ +struct origin { + /* The origin device */ + struct block_device *bdev; + + struct list_head hash_list; + + /* List of snapshots for this origin */ + struct list_head snapshots; +}; + +/* + * Size of the hash table for origin volumes. If we make this + * the size of the minors list then it should be nearly perfect + */ +#define ORIGIN_HASH_SIZE 256 +#define ORIGIN_MASK 0xFF +static struct list_head *_origins; +static struct rw_semaphore _origins_lock; + +static int init_origin_hash(void) +{ + int i; + + _origins = kmalloc(ORIGIN_HASH_SIZE * sizeof(struct list_head), + GFP_KERNEL); + if (!_origins) { + DMERR("Device mapper: Snapshot: unable to allocate memory"); + return -ENOMEM; + } + + for (i = 0; i < ORIGIN_HASH_SIZE; i++) + INIT_LIST_HEAD(_origins + i); + init_rwsem(&_origins_lock); + + return 0; +} + +static void exit_origin_hash(void) +{ + kfree(_origins); +} + +static inline unsigned int origin_hash(struct block_device *bdev) +{ + return bdev->bd_dev & ORIGIN_MASK; +} + +static struct origin *__lookup_origin(struct block_device *origin) +{ + struct list_head *ol; + struct origin *o; + + ol = &_origins[origin_hash(origin)]; + list_for_each_entry (o, ol, hash_list) + if (bdev_equal(o->bdev, origin)) + return o; + + return NULL; +} + +static void __insert_origin(struct origin *o) +{ + struct list_head *sl = &_origins[origin_hash(o->bdev)]; + list_add_tail(&o->hash_list, sl); +} + +/* + * Make a note of the snapshot and its origin so we can look it + * up when the origin has a write on it. + */ +static int register_snapshot(struct dm_snapshot *snap) +{ + struct origin *o; + struct block_device *bdev = snap->origin->bdev; + + down_write(&_origins_lock); + o = __lookup_origin(bdev); + + if (!o) { + /* New origin */ + o = kmalloc(sizeof(*o), GFP_KERNEL); + if (!o) { + up_write(&_origins_lock); + return -ENOMEM; + } + + /* Initialise the struct */ + INIT_LIST_HEAD(&o->snapshots); + o->bdev = bdev; + + __insert_origin(o); + } + + list_add_tail(&snap->list, &o->snapshots); + + up_write(&_origins_lock); + return 0; +} + +static void unregister_snapshot(struct dm_snapshot *s) +{ + struct origin *o; + + down_write(&_origins_lock); + o = __lookup_origin(s->origin->bdev); + + list_del(&s->list); + if (list_empty(&o->snapshots)) { + list_del(&o->hash_list); + kfree(o); + } + + up_write(&_origins_lock); +} + +/* + * Implementation of the exception hash tables. + */ +static int init_exception_table(struct exception_table *et, uint32_t size) +{ + unsigned int i; + + et->hash_mask = size - 1; + et->table = dm_vcalloc(size, sizeof(struct list_head)); + if (!et->table) + return -ENOMEM; + + for (i = 0; i < size; i++) + INIT_LIST_HEAD(et->table + i); + + return 0; +} + +static void exit_exception_table(struct exception_table *et, kmem_cache_t *mem) +{ + struct list_head *slot; + struct exception *ex, *next; + int i, size; + + size = et->hash_mask + 1; + for (i = 0; i < size; i++) { + slot = et->table + i; + + list_for_each_entry_safe (ex, next, slot, hash_list) + kmem_cache_free(mem, ex); + } + + vfree(et->table); +} + +static inline uint32_t exception_hash(struct exception_table *et, chunk_t chunk) +{ + return chunk & et->hash_mask; +} + +static void insert_exception(struct exception_table *eh, struct exception *e) +{ + struct list_head *l = &eh->table[exception_hash(eh, e->old_chunk)]; + list_add(&e->hash_list, l); +} + +static inline void remove_exception(struct exception *e) +{ + list_del(&e->hash_list); +} + +/* + * Return the exception data for a sector, or NULL if not + * remapped. + */ +static struct exception *lookup_exception(struct exception_table *et, + chunk_t chunk) +{ + struct list_head *slot; + struct exception *e; + + slot = &et->table[exception_hash(et, chunk)]; + list_for_each_entry (e, slot, hash_list) + if (e->old_chunk == chunk) + return e; + + return NULL; +} + +static inline struct exception *alloc_exception(void) +{ + struct exception *e; + + e = kmem_cache_alloc(exception_cache, GFP_NOIO); + if (!e) + e = kmem_cache_alloc(exception_cache, GFP_ATOMIC); + + return e; +} + +static inline void free_exception(struct exception *e) +{ + kmem_cache_free(exception_cache, e); +} + +static inline struct pending_exception *alloc_pending_exception(void) +{ + return mempool_alloc(pending_pool, GFP_NOIO); +} + +static inline void free_pending_exception(struct pending_exception *pe) +{ + mempool_free(pe, pending_pool); +} + +int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new) +{ + struct exception *e; + + e = alloc_exception(); + if (!e) + return -ENOMEM; + + e->old_chunk = old; + e->new_chunk = new; + insert_exception(&s->complete, e); + return 0; +} + +/* + * Hard coded magic. + */ +static int calc_max_buckets(void) +{ + /* use a fixed size of 2MB */ + unsigned long mem = 2 * 1024 * 1024; + mem /= sizeof(struct list_head); + + return mem; +} + +/* + * Rounds a number down to a power of 2. + */ +static inline uint32_t round_down(uint32_t n) +{ + while (n & (n - 1)) + n &= (n - 1); + return n; +} + +/* + * Allocate room for a suitable hash table. + */ +static int init_hash_tables(struct dm_snapshot *s) +{ + sector_t hash_size, cow_dev_size, origin_dev_size, max_buckets; + + /* + * Calculate based on the size of the original volume or + * the COW volume... + */ + cow_dev_size = get_dev_size(s->cow->bdev); + origin_dev_size = get_dev_size(s->origin->bdev); + max_buckets = calc_max_buckets(); + + hash_size = min(origin_dev_size, cow_dev_size) >> s->chunk_shift; + hash_size = min(hash_size, max_buckets); + + /* Round it down to a power of 2 */ + hash_size = round_down(hash_size); + if (init_exception_table(&s->complete, hash_size)) + return -ENOMEM; + + /* + * Allocate hash table for in-flight exceptions + * Make this smaller than the real hash table + */ + hash_size >>= 3; + if (hash_size < 64) + hash_size = 64; + + if (init_exception_table(&s->pending, hash_size)) { + exit_exception_table(&s->complete, exception_cache); + return -ENOMEM; + } + + return 0; +} + +/* + * Round a number up to the nearest 'size' boundary. size must + * be a power of 2. + */ +static inline ulong round_up(ulong n, ulong size) +{ + size--; + return (n + size) & ~size; +} + +/* + * Construct a snapshot mapping: <origin_dev> <COW-dev> <p/n> <chunk-size> + */ +static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + struct dm_snapshot *s; + unsigned long chunk_size; + int r = -EINVAL; + char persistent; + char *origin_path; + char *cow_path; + char *value; + int blocksize; + + if (argc < 4) { + ti->error = "dm-snapshot: requires exactly 4 arguments"; + r = -EINVAL; + goto bad1; + } + + origin_path = argv[0]; + cow_path = argv[1]; + persistent = toupper(*argv[2]); + + if (persistent != 'P' && persistent != 'N') { + ti->error = "Persistent flag is not P or N"; + r = -EINVAL; + goto bad1; + } + + chunk_size = simple_strtoul(argv[3], &value, 10); + if (chunk_size == 0 || value == NULL) { + ti->error = "Invalid chunk size"; + r = -EINVAL; + goto bad1; + } + + s = kmalloc(sizeof(*s), GFP_KERNEL); + if (s == NULL) { + ti->error = "Cannot allocate snapshot context private " + "structure"; + r = -ENOMEM; + goto bad1; + } + + r = dm_get_device(ti, origin_path, 0, ti->len, FMODE_READ, &s->origin); + if (r) { + ti->error = "Cannot get origin device"; + goto bad2; + } + + r = dm_get_device(ti, cow_path, 0, 0, + FMODE_READ | FMODE_WRITE, &s->cow); + if (r) { + dm_put_device(ti, s->origin); + ti->error = "Cannot get COW device"; + goto bad2; + } + + /* + * Chunk size must be multiple of page size. Silently + * round up if it's not. + */ + chunk_size = round_up(chunk_size, PAGE_SIZE >> 9); + + /* Validate the chunk size against the device block size */ + blocksize = s->cow->bdev->bd_disk->queue->hardsect_size; + if (chunk_size % (blocksize >> 9)) { + ti->error = "Chunk size is not a multiple of device blocksize"; + r = -EINVAL; + goto bad3; + } + + /* Check chunk_size is a power of 2 */ + if (chunk_size & (chunk_size - 1)) { + ti->error = "Chunk size is not a power of 2"; + r = -EINVAL; + goto bad3; + } + + s->chunk_size = chunk_size; + s->chunk_mask = chunk_size - 1; + s->type = persistent; + s->chunk_shift = ffs(chunk_size) - 1; + + s->valid = 1; + s->have_metadata = 0; + s->last_percent = 0; + init_rwsem(&s->lock); + s->table = ti->table; + + /* Allocate hash table for COW data */ + if (init_hash_tables(s)) { + ti->error = "Unable to allocate hash table space"; + r = -ENOMEM; + goto bad3; + } + + /* + * Check the persistent flag - done here because we need the iobuf + * to check the LV header + */ + s->store.snap = s; + + if (persistent == 'P') + r = dm_create_persistent(&s->store, chunk_size); + else + r = dm_create_transient(&s->store, s, blocksize); + + if (r) { + ti->error = "Couldn't create exception store"; + r = -EINVAL; + goto bad4; + } + + r = kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client); + if (r) { + ti->error = "Could not create kcopyd client"; + goto bad5; + } + + /* Add snapshot to the list of snapshots for this origin */ + if (register_snapshot(s)) { + r = -EINVAL; + ti->error = "Cannot register snapshot origin"; + goto bad6; + } + + ti->private = s; + ti->split_io = chunk_size; + + return 0; + + bad6: + kcopyd_client_destroy(s->kcopyd_client); + + bad5: + s->store.destroy(&s->store); + + bad4: + exit_exception_table(&s->pending, pending_cache); + exit_exception_table(&s->complete, exception_cache); + + bad3: + dm_put_device(ti, s->cow); + dm_put_device(ti, s->origin); + + bad2: + kfree(s); + + bad1: + return r; +} + +static void snapshot_dtr(struct dm_target *ti) +{ + struct dm_snapshot *s = (struct dm_snapshot *) ti->private; + + unregister_snapshot(s); + + exit_exception_table(&s->pending, pending_cache); + exit_exception_table(&s->complete, exception_cache); + + /* Deallocate memory used */ + s->store.destroy(&s->store); + + dm_put_device(ti, s->origin); + dm_put_device(ti, s->cow); + kcopyd_client_destroy(s->kcopyd_client); + kfree(s); +} + +/* + * Flush a list of buffers. + */ +static void flush_bios(struct bio *bio) +{ + struct bio *n; + + while (bio) { + n = bio->bi_next; + bio->bi_next = NULL; + generic_make_request(bio); + bio = n; + } +} + +/* + * Error a list of buffers. + */ +static void error_bios(struct bio *bio) +{ + struct bio *n; + + while (bio) { + n = bio->bi_next; + bio->bi_next = NULL; + bio_io_error(bio, bio->bi_size); + bio = n; + } +} + +static struct bio *__flush_bios(struct pending_exception *pe) +{ + struct pending_exception *sibling; + + if (list_empty(&pe->siblings)) + return bio_list_get(&pe->origin_bios); + + sibling = list_entry(pe->siblings.next, + struct pending_exception, siblings); + + list_del(&pe->siblings); + + /* This is fine as long as kcopyd is single-threaded. If kcopyd + * becomes multi-threaded, we'll need some locking here. + */ + bio_list_merge(&sibling->origin_bios, &pe->origin_bios); + + return NULL; +} + +static void pending_complete(struct pending_exception *pe, int success) +{ + struct exception *e; + struct dm_snapshot *s = pe->snap; + struct bio *flush = NULL; + + if (success) { + e = alloc_exception(); + if (!e) { + DMWARN("Unable to allocate exception."); + down_write(&s->lock); + s->store.drop_snapshot(&s->store); + s->valid = 0; + flush = __flush_bios(pe); + up_write(&s->lock); + + error_bios(bio_list_get(&pe->snapshot_bios)); + goto out; + } + *e = pe->e; + + /* + * Add a proper exception, and remove the + * in-flight exception from the list. + */ + down_write(&s->lock); + insert_exception(&s->complete, e); + remove_exception(&pe->e); + flush = __flush_bios(pe); + + /* Submit any pending write bios */ + up_write(&s->lock); + + flush_bios(bio_list_get(&pe->snapshot_bios)); + } else { + /* Read/write error - snapshot is unusable */ + down_write(&s->lock); + if (s->valid) + DMERR("Error reading/writing snapshot"); + s->store.drop_snapshot(&s->store); + s->valid = 0; + remove_exception(&pe->e); + flush = __flush_bios(pe); + up_write(&s->lock); + + error_bios(bio_list_get(&pe->snapshot_bios)); + + dm_table_event(s->table); + } + + out: + free_pending_exception(pe); + + if (flush) + flush_bios(flush); +} + +static void commit_callback(void *context, int success) +{ + struct pending_exception *pe = (struct pending_exception *) context; + pending_complete(pe, success); +} + +/* + * Called when the copy I/O has finished. kcopyd actually runs + * this code so don't block. + */ +static void copy_callback(int read_err, unsigned int write_err, void *context) +{ + struct pending_exception *pe = (struct pending_exception *) context; + struct dm_snapshot *s = pe->snap; + + if (read_err || write_err) + pending_complete(pe, 0); + + else + /* Update the metadata if we are persistent */ + s->store.commit_exception(&s->store, &pe->e, commit_callback, + pe); +} + +/* + * Dispatches the copy operation to kcopyd. + */ +static inline void start_copy(struct pending_exception *pe) +{ + struct dm_snapshot *s = pe->snap; + struct io_region src, dest; + struct block_device *bdev = s->origin->bdev; + sector_t dev_size; + + dev_size = get_dev_size(bdev); + + src.bdev = bdev; + src.sector = chunk_to_sector(s, pe->e.old_chunk); + src.count = min(s->chunk_size, dev_size - src.sector); + + dest.bdev = s->cow->bdev; + dest.sector = chunk_to_sector(s, pe->e.new_chunk); + dest.count = src.count; + + /* Hand over to kcopyd */ + kcopyd_copy(s->kcopyd_client, + &src, 1, &dest, 0, copy_callback, pe); +} + +/* + * Looks to see if this snapshot already has a pending exception + * for this chunk, otherwise it allocates a new one and inserts + * it into the pending table. + * + * NOTE: a write lock must be held on snap->lock before calling + * this. + */ +static struct pending_exception * +__find_pending_exception(struct dm_snapshot *s, struct bio *bio) +{ + struct exception *e; + struct pending_exception *pe; + chunk_t chunk = sector_to_chunk(s, bio->bi_sector); + + /* + * Is there a pending exception for this already ? + */ + e = lookup_exception(&s->pending, chunk); + if (e) { + /* cast the exception to a pending exception */ + pe = container_of(e, struct pending_exception, e); + + } else { + /* + * Create a new pending exception, we don't want + * to hold the lock while we do this. + */ + up_write(&s->lock); + pe = alloc_pending_exception(); + down_write(&s->lock); + + e = lookup_exception(&s->pending, chunk); + if (e) { + free_pending_exception(pe); + pe = container_of(e, struct pending_exception, e); + } else { + pe->e.old_chunk = chunk; + bio_list_init(&pe->origin_bios); + bio_list_init(&pe->snapshot_bios); + INIT_LIST_HEAD(&pe->siblings); + pe->snap = s; + pe->started = 0; + + if (s->store.prepare_exception(&s->store, &pe->e)) { + free_pending_exception(pe); + s->valid = 0; + return NULL; + } + + insert_exception(&s->pending, &pe->e); + } + } + + return pe; +} + +static inline void remap_exception(struct dm_snapshot *s, struct exception *e, + struct bio *bio) +{ + bio->bi_bdev = s->cow->bdev; + bio->bi_sector = chunk_to_sector(s, e->new_chunk) + + (bio->bi_sector & s->chunk_mask); +} + +static int snapshot_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + struct exception *e; + struct dm_snapshot *s = (struct dm_snapshot *) ti->private; + int r = 1; + chunk_t chunk; + struct pending_exception *pe; + + chunk = sector_to_chunk(s, bio->bi_sector); + + /* Full snapshots are not usable */ + if (!s->valid) + return -1; + + /* + * Write to snapshot - higher level takes care of RW/RO + * flags so we should only get this if we are + * writeable. + */ + if (bio_rw(bio) == WRITE) { + + /* FIXME: should only take write lock if we need + * to copy an exception */ + down_write(&s->lock); + + /* If the block is already remapped - use that, else remap it */ + e = lookup_exception(&s->complete, chunk); + if (e) { + remap_exception(s, e, bio); + up_write(&s->lock); + + } else { + pe = __find_pending_exception(s, bio); + + if (!pe) { + if (s->store.drop_snapshot) + s->store.drop_snapshot(&s->store); + s->valid = 0; + r = -EIO; + up_write(&s->lock); + } else { + remap_exception(s, &pe->e, bio); + bio_list_add(&pe->snapshot_bios, bio); + + if (!pe->started) { + /* this is protected by snap->lock */ + pe->started = 1; + up_write(&s->lock); + start_copy(pe); + } else + up_write(&s->lock); + r = 0; + } + } + + } else { + /* + * FIXME: this read path scares me because we + * always use the origin when we have a pending + * exception. However I can't think of a + * situation where this is wrong - ejt. + */ + + /* Do reads */ + down_read(&s->lock); + + /* See if it it has been remapped */ + e = lookup_exception(&s->complete, chunk); + if (e) + remap_exception(s, e, bio); + else + bio->bi_bdev = s->origin->bdev; + + up_read(&s->lock); + } + + return r; +} + +static void snapshot_resume(struct dm_target *ti) +{ + struct dm_snapshot *s = (struct dm_snapshot *) ti->private; + + if (s->have_metadata) + return; + + if (s->store.read_metadata(&s->store)) { + down_write(&s->lock); + s->valid = 0; + up_write(&s->lock); + } + + s->have_metadata = 1; +} + +static int snapshot_status(struct dm_target *ti, status_type_t type, + char *result, unsigned int maxlen) +{ + struct dm_snapshot *snap = (struct dm_snapshot *) ti->private; + + switch (type) { + case STATUSTYPE_INFO: + if (!snap->valid) + snprintf(result, maxlen, "Invalid"); + else { + if (snap->store.fraction_full) { + sector_t numerator, denominator; + snap->store.fraction_full(&snap->store, + &numerator, + &denominator); + snprintf(result, maxlen, + SECTOR_FORMAT "/" SECTOR_FORMAT, + numerator, denominator); + } + else + snprintf(result, maxlen, "Unknown"); + } + break; + + case STATUSTYPE_TABLE: + /* + * kdevname returns a static pointer so we need + * to make private copies if the output is to + * make sense. + */ + snprintf(result, maxlen, "%s %s %c " SECTOR_FORMAT, + snap->origin->name, snap->cow->name, + snap->type, snap->chunk_size); + break; + } + + return 0; +} + +/*----------------------------------------------------------------- + * Origin methods + *---------------------------------------------------------------*/ +static void list_merge(struct list_head *l1, struct list_head *l2) +{ + struct list_head *l1_n, *l2_p; + + l1_n = l1->next; + l2_p = l2->prev; + + l1->next = l2; + l2->prev = l1; + + l2_p->next = l1_n; + l1_n->prev = l2_p; +} + +static int __origin_write(struct list_head *snapshots, struct bio *bio) +{ + int r = 1, first = 1; + struct dm_snapshot *snap; + struct exception *e; + struct pending_exception *pe, *last = NULL; + chunk_t chunk; + + /* Do all the snapshots on this origin */ + list_for_each_entry (snap, snapshots, list) { + + /* Only deal with valid snapshots */ + if (!snap->valid) + continue; + + down_write(&snap->lock); + + /* + * Remember, different snapshots can have + * different chunk sizes. + */ + chunk = sector_to_chunk(snap, bio->bi_sector); + + /* + * Check exception table to see if block + * is already remapped in this snapshot + * and trigger an exception if not. + */ + e = lookup_exception(&snap->complete, chunk); + if (!e) { + pe = __find_pending_exception(snap, bio); + if (!pe) { + snap->store.drop_snapshot(&snap->store); + snap->valid = 0; + + } else { + if (last) + list_merge(&pe->siblings, + &last->siblings); + + last = pe; + r = 0; + } + } + + up_write(&snap->lock); + } + + /* + * Now that we have a complete pe list we can start the copying. + */ + if (last) { + pe = last; + do { + down_write(&pe->snap->lock); + if (first) + bio_list_add(&pe->origin_bios, bio); + if (!pe->started) { + pe->started = 1; + up_write(&pe->snap->lock); + start_copy(pe); + } else + up_write(&pe->snap->lock); + first = 0; + pe = list_entry(pe->siblings.next, + struct pending_exception, siblings); + + } while (pe != last); + } + + return r; +} + +/* + * Called on a write from the origin driver. + */ +static int do_origin(struct dm_dev *origin, struct bio *bio) +{ + struct origin *o; + int r = 1; + + down_read(&_origins_lock); + o = __lookup_origin(origin->bdev); + if (o) + r = __origin_write(&o->snapshots, bio); + up_read(&_origins_lock); + + return r; +} + +/* + * Origin: maps a linear range of a device, with hooks for snapshotting. + */ + +/* + * Construct an origin mapping: <dev_path> + * The context for an origin is merely a 'struct dm_dev *' + * pointing to the real device. + */ +static int origin_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + int r; + struct dm_dev *dev; + + if (argc != 1) { + ti->error = "dm-origin: incorrect number of arguments"; + return -EINVAL; + } + + r = dm_get_device(ti, argv[0], 0, ti->len, + dm_table_get_mode(ti->table), &dev); + if (r) { + ti->error = "Cannot get target device"; + return r; + } + + ti->private = dev; + return 0; +} + +static void origin_dtr(struct dm_target *ti) +{ + struct dm_dev *dev = (struct dm_dev *) ti->private; + dm_put_device(ti, dev); +} + +static int origin_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + struct dm_dev *dev = (struct dm_dev *) ti->private; + bio->bi_bdev = dev->bdev; + + /* Only tell snapshots if this is a write */ + return (bio_rw(bio) == WRITE) ? do_origin(dev, bio) : 1; +} + +#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) + +/* + * Set the target "split_io" field to the minimum of all the snapshots' + * chunk sizes. + */ +static void origin_resume(struct dm_target *ti) +{ + struct dm_dev *dev = (struct dm_dev *) ti->private; + struct dm_snapshot *snap; + struct origin *o; + chunk_t chunk_size = 0; + + down_read(&_origins_lock); + o = __lookup_origin(dev->bdev); + if (o) + list_for_each_entry (snap, &o->snapshots, list) + chunk_size = min_not_zero(chunk_size, snap->chunk_size); + up_read(&_origins_lock); + + ti->split_io = chunk_size; +} + +static int origin_status(struct dm_target *ti, status_type_t type, char *result, + unsigned int maxlen) +{ + struct dm_dev *dev = (struct dm_dev *) ti->private; + + switch (type) { + case STATUSTYPE_INFO: + result[0] = '\0'; + break; + + case STATUSTYPE_TABLE: + snprintf(result, maxlen, "%s", dev->name); + break; + } + + return 0; +} + +static struct target_type origin_target = { + .name = "snapshot-origin", + .version = {1, 0, 1}, + .module = THIS_MODULE, + .ctr = origin_ctr, + .dtr = origin_dtr, + .map = origin_map, + .resume = origin_resume, + .status = origin_status, +}; + +static struct target_type snapshot_target = { + .name = "snapshot", + .version = {1, 0, 1}, + .module = THIS_MODULE, + .ctr = snapshot_ctr, + .dtr = snapshot_dtr, + .map = snapshot_map, + .resume = snapshot_resume, + .status = snapshot_status, +}; + +static int __init dm_snapshot_init(void) +{ + int r; + + r = dm_register_target(&snapshot_target); + if (r) { + DMERR("snapshot target register failed %d", r); + return r; + } + + r = dm_register_target(&origin_target); + if (r < 0) { + DMERR("Device mapper: Origin: register failed %d\n", r); + goto bad1; + } + + r = init_origin_hash(); + if (r) { + DMERR("init_origin_hash failed."); + goto bad2; + } + + exception_cache = kmem_cache_create("dm-snapshot-ex", + sizeof(struct exception), + __alignof__(struct exception), + 0, NULL, NULL); + if (!exception_cache) { + DMERR("Couldn't create exception cache."); + r = -ENOMEM; + goto bad3; + } + + pending_cache = + kmem_cache_create("dm-snapshot-in", + sizeof(struct pending_exception), + __alignof__(struct pending_exception), + 0, NULL, NULL); + if (!pending_cache) { + DMERR("Couldn't create pending cache."); + r = -ENOMEM; + goto bad4; + } + + pending_pool = mempool_create(128, mempool_alloc_slab, + mempool_free_slab, pending_cache); + if (!pending_pool) { + DMERR("Couldn't create pending pool."); + r = -ENOMEM; + goto bad5; + } + + return 0; + + bad5: + kmem_cache_destroy(pending_cache); + bad4: + kmem_cache_destroy(exception_cache); + bad3: + exit_origin_hash(); + bad2: + dm_unregister_target(&origin_target); + bad1: + dm_unregister_target(&snapshot_target); + return r; +} + +static void __exit dm_snapshot_exit(void) +{ + int r; + + r = dm_unregister_target(&snapshot_target); + if (r) + DMERR("snapshot unregister failed %d", r); + + r = dm_unregister_target(&origin_target); + if (r) + DMERR("origin unregister failed %d", r); + + exit_origin_hash(); + mempool_destroy(pending_pool); + kmem_cache_destroy(pending_cache); + kmem_cache_destroy(exception_cache); +} + +/* Module hooks */ +module_init(dm_snapshot_init); +module_exit(dm_snapshot_exit); + +MODULE_DESCRIPTION(DM_NAME " snapshot target"); +MODULE_AUTHOR("Joe Thornber"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm-snap.h b/drivers/md/dm-snap.h new file mode 100644 index 00000000000..375aa24d4d7 --- /dev/null +++ b/drivers/md/dm-snap.h @@ -0,0 +1,161 @@ +/* + * dm-snapshot.c + * + * Copyright (C) 2001-2002 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + */ + +#ifndef DM_SNAPSHOT_H +#define DM_SNAPSHOT_H + +#include "dm.h" +#include <linux/blkdev.h> + +struct exception_table { + uint32_t hash_mask; + struct list_head *table; +}; + +/* + * The snapshot code deals with largish chunks of the disk at a + * time. Typically 64k - 256k. + */ +/* FIXME: can we get away with limiting these to a uint32_t ? */ +typedef sector_t chunk_t; + +/* + * An exception is used where an old chunk of data has been + * replaced by a new one. + */ +struct exception { + struct list_head hash_list; + + chunk_t old_chunk; + chunk_t new_chunk; +}; + +/* + * Abstraction to handle the meta/layout of exception stores (the + * COW device). + */ +struct exception_store { + + /* + * Destroys this object when you've finished with it. + */ + void (*destroy) (struct exception_store *store); + + /* + * The target shouldn't read the COW device until this is + * called. + */ + int (*read_metadata) (struct exception_store *store); + + /* + * Find somewhere to store the next exception. + */ + int (*prepare_exception) (struct exception_store *store, + struct exception *e); + + /* + * Update the metadata with this exception. + */ + void (*commit_exception) (struct exception_store *store, + struct exception *e, + void (*callback) (void *, int success), + void *callback_context); + + /* + * The snapshot is invalid, note this in the metadata. + */ + void (*drop_snapshot) (struct exception_store *store); + + /* + * Return how full the snapshot is. + */ + void (*fraction_full) (struct exception_store *store, + sector_t *numerator, + sector_t *denominator); + + struct dm_snapshot *snap; + void *context; +}; + +struct dm_snapshot { + struct rw_semaphore lock; + struct dm_table *table; + + struct dm_dev *origin; + struct dm_dev *cow; + + /* List of snapshots per Origin */ + struct list_head list; + + /* Size of data blocks saved - must be a power of 2 */ + chunk_t chunk_size; + chunk_t chunk_mask; + chunk_t chunk_shift; + + /* You can't use a snapshot if this is 0 (e.g. if full) */ + int valid; + int have_metadata; + + /* Used for display of table */ + char type; + + /* The last percentage we notified */ + int last_percent; + + struct exception_table pending; + struct exception_table complete; + + /* The on disk metadata handler */ + struct exception_store store; + + struct kcopyd_client *kcopyd_client; +}; + +/* + * Used by the exception stores to load exceptions hen + * initialising. + */ +int dm_add_exception(struct dm_snapshot *s, chunk_t old, chunk_t new); + +/* + * Constructor and destructor for the default persistent + * store. + */ +int dm_create_persistent(struct exception_store *store, uint32_t chunk_size); + +int dm_create_transient(struct exception_store *store, + struct dm_snapshot *s, int blocksize); + +/* + * Return the number of sectors in the device. + */ +static inline sector_t get_dev_size(struct block_device *bdev) +{ + return bdev->bd_inode->i_size >> SECTOR_SHIFT; +} + +static inline chunk_t sector_to_chunk(struct dm_snapshot *s, sector_t sector) +{ + return (sector & ~s->chunk_mask) >> s->chunk_shift; +} + +static inline sector_t chunk_to_sector(struct dm_snapshot *s, chunk_t chunk) +{ + return chunk << s->chunk_shift; +} + +static inline int bdev_equal(struct block_device *lhs, struct block_device *rhs) +{ + /* + * There is only ever one instance of a particular block + * device so we can compare pointers safely. + */ + return lhs == rhs; +} + +#endif diff --git a/drivers/md/dm-stripe.c b/drivers/md/dm-stripe.c new file mode 100644 index 00000000000..ab89278a56b --- /dev/null +++ b/drivers/md/dm-stripe.c @@ -0,0 +1,234 @@ +/* + * Copyright (C) 2001-2003 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/blkdev.h> +#include <linux/bio.h> +#include <linux/slab.h> + +struct stripe { + struct dm_dev *dev; + sector_t physical_start; +}; + +struct stripe_c { + uint32_t stripes; + + /* The size of this target / num. stripes */ + sector_t stripe_width; + + /* stripe chunk size */ + uint32_t chunk_shift; + sector_t chunk_mask; + + struct stripe stripe[0]; +}; + +static inline struct stripe_c *alloc_context(unsigned int stripes) +{ + size_t len; + + if (array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), + stripes)) + return NULL; + + len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); + + return kmalloc(len, GFP_KERNEL); +} + +/* + * Parse a single <dev> <sector> pair + */ +static int get_stripe(struct dm_target *ti, struct stripe_c *sc, + unsigned int stripe, char **argv) +{ + sector_t start; + + if (sscanf(argv[1], SECTOR_FORMAT, &start) != 1) + return -EINVAL; + + if (dm_get_device(ti, argv[0], start, sc->stripe_width, + dm_table_get_mode(ti->table), + &sc->stripe[stripe].dev)) + return -ENXIO; + + sc->stripe[stripe].physical_start = start; + return 0; +} + +/* + * Construct a striped mapping. + * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+ + */ +static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + struct stripe_c *sc; + sector_t width; + uint32_t stripes; + uint32_t chunk_size; + char *end; + int r; + unsigned int i; + + if (argc < 2) { + ti->error = "dm-stripe: Not enough arguments"; + return -EINVAL; + } + + stripes = simple_strtoul(argv[0], &end, 10); + if (*end) { + ti->error = "dm-stripe: Invalid stripe count"; + return -EINVAL; + } + + chunk_size = simple_strtoul(argv[1], &end, 10); + if (*end) { + ti->error = "dm-stripe: Invalid chunk_size"; + return -EINVAL; + } + + /* + * chunk_size is a power of two + */ + if (!chunk_size || (chunk_size & (chunk_size - 1)) || + (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) { + ti->error = "dm-stripe: Invalid chunk size"; + return -EINVAL; + } + + width = ti->len; + if (sector_div(width, stripes)) { + ti->error = "dm-stripe: Target length not divisable by " + "number of stripes"; + return -EINVAL; + } + + /* + * Do we have enough arguments for that many stripes ? + */ + if (argc != (2 + 2 * stripes)) { + ti->error = "dm-stripe: Not enough destinations " + "specified"; + return -EINVAL; + } + + sc = alloc_context(stripes); + if (!sc) { + ti->error = "dm-stripe: Memory allocation for striped context " + "failed"; + return -ENOMEM; + } + + sc->stripes = stripes; + sc->stripe_width = width; + ti->split_io = chunk_size; + + sc->chunk_mask = ((sector_t) chunk_size) - 1; + for (sc->chunk_shift = 0; chunk_size; sc->chunk_shift++) + chunk_size >>= 1; + sc->chunk_shift--; + + /* + * Get the stripe destinations. + */ + for (i = 0; i < stripes; i++) { + argv += 2; + + r = get_stripe(ti, sc, i, argv); + if (r < 0) { + ti->error = "dm-stripe: Couldn't parse stripe " + "destination"; + while (i--) + dm_put_device(ti, sc->stripe[i].dev); + kfree(sc); + return r; + } + } + + ti->private = sc; + return 0; +} + +static void stripe_dtr(struct dm_target *ti) +{ + unsigned int i; + struct stripe_c *sc = (struct stripe_c *) ti->private; + + for (i = 0; i < sc->stripes; i++) + dm_put_device(ti, sc->stripe[i].dev); + + kfree(sc); +} + +static int stripe_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + struct stripe_c *sc = (struct stripe_c *) ti->private; + + sector_t offset = bio->bi_sector - ti->begin; + sector_t chunk = offset >> sc->chunk_shift; + uint32_t stripe = sector_div(chunk, sc->stripes); + + bio->bi_bdev = sc->stripe[stripe].dev->bdev; + bio->bi_sector = sc->stripe[stripe].physical_start + + (chunk << sc->chunk_shift) + (offset & sc->chunk_mask); + return 1; +} + +static int stripe_status(struct dm_target *ti, + status_type_t type, char *result, unsigned int maxlen) +{ + struct stripe_c *sc = (struct stripe_c *) ti->private; + unsigned int sz = 0; + unsigned int i; + + switch (type) { + case STATUSTYPE_INFO: + result[0] = '\0'; + break; + + case STATUSTYPE_TABLE: + DMEMIT("%d " SECTOR_FORMAT, sc->stripes, sc->chunk_mask + 1); + for (i = 0; i < sc->stripes; i++) + DMEMIT(" %s " SECTOR_FORMAT, sc->stripe[i].dev->name, + sc->stripe[i].physical_start); + break; + } + return 0; +} + +static struct target_type stripe_target = { + .name = "striped", + .version= {1, 0, 2}, + .module = THIS_MODULE, + .ctr = stripe_ctr, + .dtr = stripe_dtr, + .map = stripe_map, + .status = stripe_status, +}; + +int __init dm_stripe_init(void) +{ + int r; + + r = dm_register_target(&stripe_target); + if (r < 0) + DMWARN("striped target registration failed"); + + return r; +} + +void dm_stripe_exit(void) +{ + if (dm_unregister_target(&stripe_target)) + DMWARN("striped target unregistration failed"); + + return; +} diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c new file mode 100644 index 00000000000..ee175d4906c --- /dev/null +++ b/drivers/md/dm-table.c @@ -0,0 +1,950 @@ +/* + * Copyright (C) 2001 Sistina Software (UK) Limited. + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/vmalloc.h> +#include <linux/blkdev.h> +#include <linux/namei.h> +#include <linux/ctype.h> +#include <linux/slab.h> +#include <linux/interrupt.h> +#include <asm/atomic.h> + +#define MAX_DEPTH 16 +#define NODE_SIZE L1_CACHE_BYTES +#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t)) +#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1) + +struct dm_table { + atomic_t holders; + + /* btree table */ + unsigned int depth; + unsigned int counts[MAX_DEPTH]; /* in nodes */ + sector_t *index[MAX_DEPTH]; + + unsigned int num_targets; + unsigned int num_allocated; + sector_t *highs; + struct dm_target *targets; + + /* + * Indicates the rw permissions for the new logical + * device. This should be a combination of FMODE_READ + * and FMODE_WRITE. + */ + int mode; + + /* a list of devices used by this table */ + struct list_head devices; + + /* + * These are optimistic limits taken from all the + * targets, some targets will need smaller limits. + */ + struct io_restrictions limits; + + /* events get handed up using this callback */ + void (*event_fn)(void *); + void *event_context; +}; + +/* + * Similar to ceiling(log_size(n)) + */ +static unsigned int int_log(unsigned int n, unsigned int base) +{ + int result = 0; + + while (n > 1) { + n = dm_div_up(n, base); + result++; + } + + return result; +} + +/* + * Returns the minimum that is _not_ zero, unless both are zero. + */ +#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) + +/* + * Combine two io_restrictions, always taking the lower value. + */ +static void combine_restrictions_low(struct io_restrictions *lhs, + struct io_restrictions *rhs) +{ + lhs->max_sectors = + min_not_zero(lhs->max_sectors, rhs->max_sectors); + + lhs->max_phys_segments = + min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments); + + lhs->max_hw_segments = + min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments); + + lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size); + + lhs->max_segment_size = + min_not_zero(lhs->max_segment_size, rhs->max_segment_size); + + lhs->seg_boundary_mask = + min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask); +} + +/* + * Calculate the index of the child node of the n'th node k'th key. + */ +static inline unsigned int get_child(unsigned int n, unsigned int k) +{ + return (n * CHILDREN_PER_NODE) + k; +} + +/* + * Return the n'th node of level l from table t. + */ +static inline sector_t *get_node(struct dm_table *t, + unsigned int l, unsigned int n) +{ + return t->index[l] + (n * KEYS_PER_NODE); +} + +/* + * Return the highest key that you could lookup from the n'th + * node on level l of the btree. + */ +static sector_t high(struct dm_table *t, unsigned int l, unsigned int n) +{ + for (; l < t->depth - 1; l++) + n = get_child(n, CHILDREN_PER_NODE - 1); + + if (n >= t->counts[l]) + return (sector_t) - 1; + + return get_node(t, l, n)[KEYS_PER_NODE - 1]; +} + +/* + * Fills in a level of the btree based on the highs of the level + * below it. + */ +static int setup_btree_index(unsigned int l, struct dm_table *t) +{ + unsigned int n, k; + sector_t *node; + + for (n = 0U; n < t->counts[l]; n++) { + node = get_node(t, l, n); + + for (k = 0U; k < KEYS_PER_NODE; k++) + node[k] = high(t, l + 1, get_child(n, k)); + } + + return 0; +} + +void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size) +{ + unsigned long size; + void *addr; + + /* + * Check that we're not going to overflow. + */ + if (nmemb > (ULONG_MAX / elem_size)) + return NULL; + + size = nmemb * elem_size; + addr = vmalloc(size); + if (addr) + memset(addr, 0, size); + + return addr; +} + +/* + * highs, and targets are managed as dynamic arrays during a + * table load. + */ +static int alloc_targets(struct dm_table *t, unsigned int num) +{ + sector_t *n_highs; + struct dm_target *n_targets; + int n = t->num_targets; + + /* + * Allocate both the target array and offset array at once. + */ + n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) + + sizeof(sector_t)); + if (!n_highs) + return -ENOMEM; + + n_targets = (struct dm_target *) (n_highs + num); + + if (n) { + memcpy(n_highs, t->highs, sizeof(*n_highs) * n); + memcpy(n_targets, t->targets, sizeof(*n_targets) * n); + } + + memset(n_highs + n, -1, sizeof(*n_highs) * (num - n)); + vfree(t->highs); + + t->num_allocated = num; + t->highs = n_highs; + t->targets = n_targets; + + return 0; +} + +int dm_table_create(struct dm_table **result, int mode, unsigned num_targets) +{ + struct dm_table *t = kmalloc(sizeof(*t), GFP_KERNEL); + + if (!t) + return -ENOMEM; + + memset(t, 0, sizeof(*t)); + INIT_LIST_HEAD(&t->devices); + atomic_set(&t->holders, 1); + + if (!num_targets) + num_targets = KEYS_PER_NODE; + + num_targets = dm_round_up(num_targets, KEYS_PER_NODE); + + if (alloc_targets(t, num_targets)) { + kfree(t); + t = NULL; + return -ENOMEM; + } + + t->mode = mode; + *result = t; + return 0; +} + +static void free_devices(struct list_head *devices) +{ + struct list_head *tmp, *next; + + for (tmp = devices->next; tmp != devices; tmp = next) { + struct dm_dev *dd = list_entry(tmp, struct dm_dev, list); + next = tmp->next; + kfree(dd); + } +} + +void table_destroy(struct dm_table *t) +{ + unsigned int i; + + /* free the indexes (see dm_table_complete) */ + if (t->depth >= 2) + vfree(t->index[t->depth - 2]); + + /* free the targets */ + for (i = 0; i < t->num_targets; i++) { + struct dm_target *tgt = t->targets + i; + + if (tgt->type->dtr) + tgt->type->dtr(tgt); + + dm_put_target_type(tgt->type); + } + + vfree(t->highs); + + /* free the device list */ + if (t->devices.next != &t->devices) { + DMWARN("devices still present during destroy: " + "dm_table_remove_device calls missing"); + + free_devices(&t->devices); + } + + kfree(t); +} + +void dm_table_get(struct dm_table *t) +{ + atomic_inc(&t->holders); +} + +void dm_table_put(struct dm_table *t) +{ + if (!t) + return; + + if (atomic_dec_and_test(&t->holders)) + table_destroy(t); +} + +/* + * Checks to see if we need to extend highs or targets. + */ +static inline int check_space(struct dm_table *t) +{ + if (t->num_targets >= t->num_allocated) + return alloc_targets(t, t->num_allocated * 2); + + return 0; +} + +/* + * Convert a device path to a dev_t. + */ +static int lookup_device(const char *path, dev_t *dev) +{ + int r; + struct nameidata nd; + struct inode *inode; + + if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd))) + return r; + + inode = nd.dentry->d_inode; + if (!inode) { + r = -ENOENT; + goto out; + } + + if (!S_ISBLK(inode->i_mode)) { + r = -ENOTBLK; + goto out; + } + + *dev = inode->i_rdev; + + out: + path_release(&nd); + return r; +} + +/* + * See if we've already got a device in the list. + */ +static struct dm_dev *find_device(struct list_head *l, dev_t dev) +{ + struct dm_dev *dd; + + list_for_each_entry (dd, l, list) + if (dd->bdev->bd_dev == dev) + return dd; + + return NULL; +} + +/* + * Open a device so we can use it as a map destination. + */ +static int open_dev(struct dm_dev *d, dev_t dev) +{ + static char *_claim_ptr = "I belong to device-mapper"; + struct block_device *bdev; + + int r; + + if (d->bdev) + BUG(); + + bdev = open_by_devnum(dev, d->mode); + if (IS_ERR(bdev)) + return PTR_ERR(bdev); + r = bd_claim(bdev, _claim_ptr); + if (r) + blkdev_put(bdev); + else + d->bdev = bdev; + return r; +} + +/* + * Close a device that we've been using. + */ +static void close_dev(struct dm_dev *d) +{ + if (!d->bdev) + return; + + bd_release(d->bdev); + blkdev_put(d->bdev); + d->bdev = NULL; +} + +/* + * If possible (ie. blk_size[major] is set), this checks an area + * of a destination device is valid. + */ +static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len) +{ + sector_t dev_size; + dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT; + return ((start < dev_size) && (len <= (dev_size - start))); +} + +/* + * This upgrades the mode on an already open dm_dev. Being + * careful to leave things as they were if we fail to reopen the + * device. + */ +static int upgrade_mode(struct dm_dev *dd, int new_mode) +{ + int r; + struct dm_dev dd_copy; + dev_t dev = dd->bdev->bd_dev; + + dd_copy = *dd; + + dd->mode |= new_mode; + dd->bdev = NULL; + r = open_dev(dd, dev); + if (!r) + close_dev(&dd_copy); + else + *dd = dd_copy; + + return r; +} + +/* + * Add a device to the list, or just increment the usage count if + * it's already present. + */ +static int __table_get_device(struct dm_table *t, struct dm_target *ti, + const char *path, sector_t start, sector_t len, + int mode, struct dm_dev **result) +{ + int r; + dev_t dev; + struct dm_dev *dd; + unsigned int major, minor; + + if (!t) + BUG(); + + if (sscanf(path, "%u:%u", &major, &minor) == 2) { + /* Extract the major/minor numbers */ + dev = MKDEV(major, minor); + if (MAJOR(dev) != major || MINOR(dev) != minor) + return -EOVERFLOW; + } else { + /* convert the path to a device */ + if ((r = lookup_device(path, &dev))) + return r; + } + + dd = find_device(&t->devices, dev); + if (!dd) { + dd = kmalloc(sizeof(*dd), GFP_KERNEL); + if (!dd) + return -ENOMEM; + + dd->mode = mode; + dd->bdev = NULL; + + if ((r = open_dev(dd, dev))) { + kfree(dd); + return r; + } + + format_dev_t(dd->name, dev); + + atomic_set(&dd->count, 0); + list_add(&dd->list, &t->devices); + + } else if (dd->mode != (mode | dd->mode)) { + r = upgrade_mode(dd, mode); + if (r) + return r; + } + atomic_inc(&dd->count); + + if (!check_device_area(dd, start, len)) { + DMWARN("device %s too small for target", path); + dm_put_device(ti, dd); + return -EINVAL; + } + + *result = dd; + + return 0; +} + + +int dm_get_device(struct dm_target *ti, const char *path, sector_t start, + sector_t len, int mode, struct dm_dev **result) +{ + int r = __table_get_device(ti->table, ti, path, + start, len, mode, result); + if (!r) { + request_queue_t *q = bdev_get_queue((*result)->bdev); + struct io_restrictions *rs = &ti->limits; + + /* + * Combine the device limits low. + * + * FIXME: if we move an io_restriction struct + * into q this would just be a call to + * combine_restrictions_low() + */ + rs->max_sectors = + min_not_zero(rs->max_sectors, q->max_sectors); + + /* FIXME: Device-Mapper on top of RAID-0 breaks because DM + * currently doesn't honor MD's merge_bvec_fn routine. + * In this case, we'll force DM to use PAGE_SIZE or + * smaller I/O, just to be safe. A better fix is in the + * works, but add this for the time being so it will at + * least operate correctly. + */ + if (q->merge_bvec_fn) + rs->max_sectors = + min_not_zero(rs->max_sectors, + (unsigned short)(PAGE_SIZE >> 9)); + + rs->max_phys_segments = + min_not_zero(rs->max_phys_segments, + q->max_phys_segments); + + rs->max_hw_segments = + min_not_zero(rs->max_hw_segments, q->max_hw_segments); + + rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size); + + rs->max_segment_size = + min_not_zero(rs->max_segment_size, q->max_segment_size); + + rs->seg_boundary_mask = + min_not_zero(rs->seg_boundary_mask, + q->seg_boundary_mask); + } + + return r; +} + +/* + * Decrement a devices use count and remove it if necessary. + */ +void dm_put_device(struct dm_target *ti, struct dm_dev *dd) +{ + if (atomic_dec_and_test(&dd->count)) { + close_dev(dd); + list_del(&dd->list); + kfree(dd); + } +} + +/* + * Checks to see if the target joins onto the end of the table. + */ +static int adjoin(struct dm_table *table, struct dm_target *ti) +{ + struct dm_target *prev; + + if (!table->num_targets) + return !ti->begin; + + prev = &table->targets[table->num_targets - 1]; + return (ti->begin == (prev->begin + prev->len)); +} + +/* + * Used to dynamically allocate the arg array. + */ +static char **realloc_argv(unsigned *array_size, char **old_argv) +{ + char **argv; + unsigned new_size; + + new_size = *array_size ? *array_size * 2 : 64; + argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL); + if (argv) { + memcpy(argv, old_argv, *array_size * sizeof(*argv)); + *array_size = new_size; + } + + kfree(old_argv); + return argv; +} + +/* + * Destructively splits up the argument list to pass to ctr. + */ +int dm_split_args(int *argc, char ***argvp, char *input) +{ + char *start, *end = input, *out, **argv = NULL; + unsigned array_size = 0; + + *argc = 0; + argv = realloc_argv(&array_size, argv); + if (!argv) + return -ENOMEM; + + while (1) { + start = end; + + /* Skip whitespace */ + while (*start && isspace(*start)) + start++; + + if (!*start) + break; /* success, we hit the end */ + + /* 'out' is used to remove any back-quotes */ + end = out = start; + while (*end) { + /* Everything apart from '\0' can be quoted */ + if (*end == '\\' && *(end + 1)) { + *out++ = *(end + 1); + end += 2; + continue; + } + + if (isspace(*end)) + break; /* end of token */ + + *out++ = *end++; + } + + /* have we already filled the array ? */ + if ((*argc + 1) > array_size) { + argv = realloc_argv(&array_size, argv); + if (!argv) + return -ENOMEM; + } + + /* we know this is whitespace */ + if (*end) + end++; + + /* terminate the string and put it in the array */ + *out = '\0'; + argv[*argc] = start; + (*argc)++; + } + + *argvp = argv; + return 0; +} + +static void check_for_valid_limits(struct io_restrictions *rs) +{ + if (!rs->max_sectors) + rs->max_sectors = MAX_SECTORS; + if (!rs->max_phys_segments) + rs->max_phys_segments = MAX_PHYS_SEGMENTS; + if (!rs->max_hw_segments) + rs->max_hw_segments = MAX_HW_SEGMENTS; + if (!rs->hardsect_size) + rs->hardsect_size = 1 << SECTOR_SHIFT; + if (!rs->max_segment_size) + rs->max_segment_size = MAX_SEGMENT_SIZE; + if (!rs->seg_boundary_mask) + rs->seg_boundary_mask = -1; +} + +int dm_table_add_target(struct dm_table *t, const char *type, + sector_t start, sector_t len, char *params) +{ + int r = -EINVAL, argc; + char **argv; + struct dm_target *tgt; + + if ((r = check_space(t))) + return r; + + tgt = t->targets + t->num_targets; + memset(tgt, 0, sizeof(*tgt)); + + if (!len) { + tgt->error = "zero-length target"; + DMERR("%s", tgt->error); + return -EINVAL; + } + + tgt->type = dm_get_target_type(type); + if (!tgt->type) { + tgt->error = "unknown target type"; + DMERR("%s", tgt->error); + return -EINVAL; + } + + tgt->table = t; + tgt->begin = start; + tgt->len = len; + tgt->error = "Unknown error"; + + /* + * Does this target adjoin the previous one ? + */ + if (!adjoin(t, tgt)) { + tgt->error = "Gap in table"; + r = -EINVAL; + goto bad; + } + + r = dm_split_args(&argc, &argv, params); + if (r) { + tgt->error = "couldn't split parameters (insufficient memory)"; + goto bad; + } + + r = tgt->type->ctr(tgt, argc, argv); + kfree(argv); + if (r) + goto bad; + + t->highs[t->num_targets++] = tgt->begin + tgt->len - 1; + + /* FIXME: the plan is to combine high here and then have + * the merge fn apply the target level restrictions. */ + combine_restrictions_low(&t->limits, &tgt->limits); + return 0; + + bad: + DMERR("%s", tgt->error); + dm_put_target_type(tgt->type); + return r; +} + +static int setup_indexes(struct dm_table *t) +{ + int i; + unsigned int total = 0; + sector_t *indexes; + + /* allocate the space for *all* the indexes */ + for (i = t->depth - 2; i >= 0; i--) { + t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE); + total += t->counts[i]; + } + + indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE); + if (!indexes) + return -ENOMEM; + + /* set up internal nodes, bottom-up */ + for (i = t->depth - 2, total = 0; i >= 0; i--) { + t->index[i] = indexes; + indexes += (KEYS_PER_NODE * t->counts[i]); + setup_btree_index(i, t); + } + + return 0; +} + +/* + * Builds the btree to index the map. + */ +int dm_table_complete(struct dm_table *t) +{ + int r = 0; + unsigned int leaf_nodes; + + check_for_valid_limits(&t->limits); + + /* how many indexes will the btree have ? */ + leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE); + t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE); + + /* leaf layer has already been set up */ + t->counts[t->depth - 1] = leaf_nodes; + t->index[t->depth - 1] = t->highs; + + if (t->depth >= 2) + r = setup_indexes(t); + + return r; +} + +static DECLARE_MUTEX(_event_lock); +void dm_table_event_callback(struct dm_table *t, + void (*fn)(void *), void *context) +{ + down(&_event_lock); + t->event_fn = fn; + t->event_context = context; + up(&_event_lock); +} + +void dm_table_event(struct dm_table *t) +{ + /* + * You can no longer call dm_table_event() from interrupt + * context, use a bottom half instead. + */ + BUG_ON(in_interrupt()); + + down(&_event_lock); + if (t->event_fn) + t->event_fn(t->event_context); + up(&_event_lock); +} + +sector_t dm_table_get_size(struct dm_table *t) +{ + return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0; +} + +struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index) +{ + if (index > t->num_targets) + return NULL; + + return t->targets + index; +} + +/* + * Search the btree for the correct target. + */ +struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector) +{ + unsigned int l, n = 0, k = 0; + sector_t *node; + + for (l = 0; l < t->depth; l++) { + n = get_child(n, k); + node = get_node(t, l, n); + + for (k = 0; k < KEYS_PER_NODE; k++) + if (node[k] >= sector) + break; + } + + return &t->targets[(KEYS_PER_NODE * n) + k]; +} + +void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q) +{ + /* + * Make sure we obey the optimistic sub devices + * restrictions. + */ + blk_queue_max_sectors(q, t->limits.max_sectors); + q->max_phys_segments = t->limits.max_phys_segments; + q->max_hw_segments = t->limits.max_hw_segments; + q->hardsect_size = t->limits.hardsect_size; + q->max_segment_size = t->limits.max_segment_size; + q->seg_boundary_mask = t->limits.seg_boundary_mask; +} + +unsigned int dm_table_get_num_targets(struct dm_table *t) +{ + return t->num_targets; +} + +struct list_head *dm_table_get_devices(struct dm_table *t) +{ + return &t->devices; +} + +int dm_table_get_mode(struct dm_table *t) +{ + return t->mode; +} + +static void suspend_targets(struct dm_table *t, unsigned postsuspend) +{ + int i = t->num_targets; + struct dm_target *ti = t->targets; + + while (i--) { + if (postsuspend) { + if (ti->type->postsuspend) + ti->type->postsuspend(ti); + } else if (ti->type->presuspend) + ti->type->presuspend(ti); + + ti++; + } +} + +void dm_table_presuspend_targets(struct dm_table *t) +{ + return suspend_targets(t, 0); +} + +void dm_table_postsuspend_targets(struct dm_table *t) +{ + return suspend_targets(t, 1); +} + +void dm_table_resume_targets(struct dm_table *t) +{ + int i; + + for (i = 0; i < t->num_targets; i++) { + struct dm_target *ti = t->targets + i; + + if (ti->type->resume) + ti->type->resume(ti); + } +} + +int dm_table_any_congested(struct dm_table *t, int bdi_bits) +{ + struct list_head *d, *devices; + int r = 0; + + devices = dm_table_get_devices(t); + for (d = devices->next; d != devices; d = d->next) { + struct dm_dev *dd = list_entry(d, struct dm_dev, list); + request_queue_t *q = bdev_get_queue(dd->bdev); + r |= bdi_congested(&q->backing_dev_info, bdi_bits); + } + + return r; +} + +void dm_table_unplug_all(struct dm_table *t) +{ + struct list_head *d, *devices = dm_table_get_devices(t); + + for (d = devices->next; d != devices; d = d->next) { + struct dm_dev *dd = list_entry(d, struct dm_dev, list); + request_queue_t *q = bdev_get_queue(dd->bdev); + + if (q->unplug_fn) + q->unplug_fn(q); + } +} + +int dm_table_flush_all(struct dm_table *t) +{ + struct list_head *d, *devices = dm_table_get_devices(t); + int ret = 0; + + for (d = devices->next; d != devices; d = d->next) { + struct dm_dev *dd = list_entry(d, struct dm_dev, list); + request_queue_t *q = bdev_get_queue(dd->bdev); + int err; + + if (!q->issue_flush_fn) + err = -EOPNOTSUPP; + else + err = q->issue_flush_fn(q, dd->bdev->bd_disk, NULL); + + if (!ret) + ret = err; + } + + return ret; +} + +EXPORT_SYMBOL(dm_vcalloc); +EXPORT_SYMBOL(dm_get_device); +EXPORT_SYMBOL(dm_put_device); +EXPORT_SYMBOL(dm_table_event); +EXPORT_SYMBOL(dm_table_get_mode); +EXPORT_SYMBOL(dm_table_put); +EXPORT_SYMBOL(dm_table_get); +EXPORT_SYMBOL(dm_table_unplug_all); +EXPORT_SYMBOL(dm_table_flush_all); diff --git a/drivers/md/dm-target.c b/drivers/md/dm-target.c new file mode 100644 index 00000000000..aecd9e0c261 --- /dev/null +++ b/drivers/md/dm-target.c @@ -0,0 +1,196 @@ +/* + * Copyright (C) 2001 Sistina Software (UK) Limited + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/kmod.h> +#include <linux/bio.h> +#include <linux/slab.h> + +struct tt_internal { + struct target_type tt; + + struct list_head list; + long use; +}; + +static LIST_HEAD(_targets); +static DECLARE_RWSEM(_lock); + +#define DM_MOD_NAME_SIZE 32 + +static inline struct tt_internal *__find_target_type(const char *name) +{ + struct tt_internal *ti; + + list_for_each_entry (ti, &_targets, list) + if (!strcmp(name, ti->tt.name)) + return ti; + + return NULL; +} + +static struct tt_internal *get_target_type(const char *name) +{ + struct tt_internal *ti; + + down_read(&_lock); + + ti = __find_target_type(name); + if (ti) { + if ((ti->use == 0) && !try_module_get(ti->tt.module)) + ti = NULL; + else + ti->use++; + } + + up_read(&_lock); + return ti; +} + +static void load_module(const char *name) +{ + request_module("dm-%s", name); +} + +struct target_type *dm_get_target_type(const char *name) +{ + struct tt_internal *ti = get_target_type(name); + + if (!ti) { + load_module(name); + ti = get_target_type(name); + } + + return ti ? &ti->tt : NULL; +} + +void dm_put_target_type(struct target_type *t) +{ + struct tt_internal *ti = (struct tt_internal *) t; + + down_read(&_lock); + if (--ti->use == 0) + module_put(ti->tt.module); + + if (ti->use < 0) + BUG(); + up_read(&_lock); + + return; +} + +static struct tt_internal *alloc_target(struct target_type *t) +{ + struct tt_internal *ti = kmalloc(sizeof(*ti), GFP_KERNEL); + + if (ti) { + memset(ti, 0, sizeof(*ti)); + ti->tt = *t; + } + + return ti; +} + + +int dm_target_iterate(void (*iter_func)(struct target_type *tt, + void *param), void *param) +{ + struct tt_internal *ti; + + down_read(&_lock); + list_for_each_entry (ti, &_targets, list) + iter_func(&ti->tt, param); + up_read(&_lock); + + return 0; +} + +int dm_register_target(struct target_type *t) +{ + int rv = 0; + struct tt_internal *ti = alloc_target(t); + + if (!ti) + return -ENOMEM; + + down_write(&_lock); + if (__find_target_type(t->name)) + rv = -EEXIST; + else + list_add(&ti->list, &_targets); + + up_write(&_lock); + if (rv) + kfree(ti); + return rv; +} + +int dm_unregister_target(struct target_type *t) +{ + struct tt_internal *ti; + + down_write(&_lock); + if (!(ti = __find_target_type(t->name))) { + up_write(&_lock); + return -EINVAL; + } + + if (ti->use) { + up_write(&_lock); + return -ETXTBSY; + } + + list_del(&ti->list); + kfree(ti); + + up_write(&_lock); + return 0; +} + +/* + * io-err: always fails an io, useful for bringing + * up LVs that have holes in them. + */ +static int io_err_ctr(struct dm_target *ti, unsigned int argc, char **args) +{ + return 0; +} + +static void io_err_dtr(struct dm_target *ti) +{ + /* empty */ +} + +static int io_err_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + return -EIO; +} + +static struct target_type error_target = { + .name = "error", + .version = {1, 0, 1}, + .ctr = io_err_ctr, + .dtr = io_err_dtr, + .map = io_err_map, +}; + +int __init dm_target_init(void) +{ + return dm_register_target(&error_target); +} + +void dm_target_exit(void) +{ + if (dm_unregister_target(&error_target)) + DMWARN("error target unregistration failed"); +} + +EXPORT_SYMBOL(dm_register_target); +EXPORT_SYMBOL(dm_unregister_target); diff --git a/drivers/md/dm-zero.c b/drivers/md/dm-zero.c new file mode 100644 index 00000000000..7febc2cac73 --- /dev/null +++ b/drivers/md/dm-zero.c @@ -0,0 +1,81 @@ +/* + * Copyright (C) 2003 Christophe Saout <christophe@saout.de> + * + * This file is released under the GPL. + */ + +#include "dm.h" + +#include <linux/module.h> +#include <linux/init.h> +#include <linux/bio.h> + +/* + * Construct a dummy mapping that only returns zeros + */ +static int zero_ctr(struct dm_target *ti, unsigned int argc, char **argv) +{ + if (argc != 0) { + ti->error = "dm-zero: No arguments required"; + return -EINVAL; + } + + return 0; +} + +/* + * Return zeros only on reads + */ +static int zero_map(struct dm_target *ti, struct bio *bio, + union map_info *map_context) +{ + switch(bio_rw(bio)) { + case READ: + zero_fill_bio(bio); + break; + case READA: + /* readahead of null bytes only wastes buffer cache */ + return -EIO; + case WRITE: + /* writes get silently dropped */ + break; + } + + bio_endio(bio, bio->bi_size, 0); + + /* accepted bio, don't make new request */ + return 0; +} + +static struct target_type zero_target = { + .name = "zero", + .version = {1, 0, 0}, + .module = THIS_MODULE, + .ctr = zero_ctr, + .map = zero_map, +}; + +int __init dm_zero_init(void) +{ + int r = dm_register_target(&zero_target); + + if (r < 0) + DMERR("zero: register failed %d", r); + + return r; +} + +void __exit dm_zero_exit(void) +{ + int r = dm_unregister_target(&zero_target); + + if (r < 0) + DMERR("zero: unregister failed %d", r); +} + +module_init(dm_zero_init) +module_exit(dm_zero_exit) + +MODULE_AUTHOR("Christophe Saout <christophe@saout.de>"); +MODULE_DESCRIPTION(DM_NAME " dummy target returning zeros"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm.c b/drivers/md/dm.c new file mode 100644 index 00000000000..243ff6884e8 --- /dev/null +++ b/drivers/md/dm.c @@ -0,0 +1,1194 @@ +/* + * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the GPL. + */ + +#include "dm.h" +#include "dm-bio-list.h" + +#include <linux/init.h> +#include <linux/module.h> +#include <linux/moduleparam.h> +#include <linux/blkpg.h> +#include <linux/bio.h> +#include <linux/buffer_head.h> +#include <linux/mempool.h> +#include <linux/slab.h> +#include <linux/idr.h> + +static const char *_name = DM_NAME; + +static unsigned int major = 0; +static unsigned int _major = 0; + +/* + * One of these is allocated per bio. + */ +struct dm_io { + struct mapped_device *md; + int error; + struct bio *bio; + atomic_t io_count; +}; + +/* + * One of these is allocated per target within a bio. Hopefully + * this will be simplified out one day. + */ +struct target_io { + struct dm_io *io; + struct dm_target *ti; + union map_info info; +}; + +union map_info *dm_get_mapinfo(struct bio *bio) +{ + if (bio && bio->bi_private) + return &((struct target_io *)bio->bi_private)->info; + return NULL; +} + +/* + * Bits for the md->flags field. + */ +#define DMF_BLOCK_IO 0 +#define DMF_SUSPENDED 1 +#define DMF_FS_LOCKED 2 + +struct mapped_device { + struct rw_semaphore lock; + rwlock_t map_lock; + atomic_t holders; + + unsigned long flags; + + request_queue_t *queue; + struct gendisk *disk; + + void *interface_ptr; + + /* + * A list of ios that arrived while we were suspended. + */ + atomic_t pending; + wait_queue_head_t wait; + struct bio_list deferred; + + /* + * The current mapping. + */ + struct dm_table *map; + + /* + * io objects are allocated from here. + */ + mempool_t *io_pool; + mempool_t *tio_pool; + + /* + * Event handling. + */ + atomic_t event_nr; + wait_queue_head_t eventq; + + /* + * freeze/thaw support require holding onto a super block + */ + struct super_block *frozen_sb; +}; + +#define MIN_IOS 256 +static kmem_cache_t *_io_cache; +static kmem_cache_t *_tio_cache; + +static struct bio_set *dm_set; + +static int __init local_init(void) +{ + int r; + + dm_set = bioset_create(16, 16, 4); + if (!dm_set) + return -ENOMEM; + + /* allocate a slab for the dm_ios */ + _io_cache = kmem_cache_create("dm_io", + sizeof(struct dm_io), 0, 0, NULL, NULL); + if (!_io_cache) + return -ENOMEM; + + /* allocate a slab for the target ios */ + _tio_cache = kmem_cache_create("dm_tio", sizeof(struct target_io), + 0, 0, NULL, NULL); + if (!_tio_cache) { + kmem_cache_destroy(_io_cache); + return -ENOMEM; + } + + _major = major; + r = register_blkdev(_major, _name); + if (r < 0) { + kmem_cache_destroy(_tio_cache); + kmem_cache_destroy(_io_cache); + return r; + } + + if (!_major) + _major = r; + + return 0; +} + +static void local_exit(void) +{ + kmem_cache_destroy(_tio_cache); + kmem_cache_destroy(_io_cache); + + bioset_free(dm_set); + + if (unregister_blkdev(_major, _name) < 0) + DMERR("devfs_unregister_blkdev failed"); + + _major = 0; + + DMINFO("cleaned up"); +} + +int (*_inits[])(void) __initdata = { + local_init, + dm_target_init, + dm_linear_init, + dm_stripe_init, + dm_interface_init, +}; + +void (*_exits[])(void) = { + local_exit, + dm_target_exit, + dm_linear_exit, + dm_stripe_exit, + dm_interface_exit, +}; + +static int __init dm_init(void) +{ + const int count = ARRAY_SIZE(_inits); + + int r, i; + + for (i = 0; i < count; i++) { + r = _inits[i](); + if (r) + goto bad; + } + + return 0; + + bad: + while (i--) + _exits[i](); + + return r; +} + +static void __exit dm_exit(void) +{ + int i = ARRAY_SIZE(_exits); + + while (i--) + _exits[i](); +} + +/* + * Block device functions + */ +static int dm_blk_open(struct inode *inode, struct file *file) +{ + struct mapped_device *md; + + md = inode->i_bdev->bd_disk->private_data; + dm_get(md); + return 0; +} + +static int dm_blk_close(struct inode *inode, struct file *file) +{ + struct mapped_device *md; + + md = inode->i_bdev->bd_disk->private_data; + dm_put(md); + return 0; +} + +static inline struct dm_io *alloc_io(struct mapped_device *md) +{ + return mempool_alloc(md->io_pool, GFP_NOIO); +} + +static inline void free_io(struct mapped_device *md, struct dm_io *io) +{ + mempool_free(io, md->io_pool); +} + +static inline struct target_io *alloc_tio(struct mapped_device *md) +{ + return mempool_alloc(md->tio_pool, GFP_NOIO); +} + +static inline void free_tio(struct mapped_device *md, struct target_io *tio) +{ + mempool_free(tio, md->tio_pool); +} + +/* + * Add the bio to the list of deferred io. + */ +static int queue_io(struct mapped_device *md, struct bio *bio) +{ + down_write(&md->lock); + + if (!test_bit(DMF_BLOCK_IO, &md->flags)) { + up_write(&md->lock); + return 1; + } + + bio_list_add(&md->deferred, bio); + + up_write(&md->lock); + return 0; /* deferred successfully */ +} + +/* + * Everyone (including functions in this file), should use this + * function to access the md->map field, and make sure they call + * dm_table_put() when finished. + */ +struct dm_table *dm_get_table(struct mapped_device *md) +{ + struct dm_table *t; + + read_lock(&md->map_lock); + t = md->map; + if (t) + dm_table_get(t); + read_unlock(&md->map_lock); + + return t; +} + +/*----------------------------------------------------------------- + * CRUD START: + * A more elegant soln is in the works that uses the queue + * merge fn, unfortunately there are a couple of changes to + * the block layer that I want to make for this. So in the + * interests of getting something for people to use I give + * you this clearly demarcated crap. + *---------------------------------------------------------------*/ + +/* + * Decrements the number of outstanding ios that a bio has been + * cloned into, completing the original io if necc. + */ +static inline void dec_pending(struct dm_io *io, int error) +{ + if (error) + io->error = error; + + if (atomic_dec_and_test(&io->io_count)) { + if (atomic_dec_and_test(&io->md->pending)) + /* nudge anyone waiting on suspend queue */ + wake_up(&io->md->wait); + + bio_endio(io->bio, io->bio->bi_size, io->error); + free_io(io->md, io); + } +} + +static int clone_endio(struct bio *bio, unsigned int done, int error) +{ + int r = 0; + struct target_io *tio = bio->bi_private; + struct dm_io *io = tio->io; + dm_endio_fn endio = tio->ti->type->end_io; + + if (bio->bi_size) + return 1; + + if (!bio_flagged(bio, BIO_UPTODATE) && !error) + error = -EIO; + + if (endio) { + r = endio(tio->ti, bio, error, &tio->info); + if (r < 0) + error = r; + + else if (r > 0) + /* the target wants another shot at the io */ + return 1; + } + + free_tio(io->md, tio); + dec_pending(io, error); + bio_put(bio); + return r; +} + +static sector_t max_io_len(struct mapped_device *md, + sector_t sector, struct dm_target *ti) +{ + sector_t offset = sector - ti->begin; + sector_t len = ti->len - offset; + + /* + * Does the target need to split even further ? + */ + if (ti->split_io) { + sector_t boundary; + boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) + - offset; + if (len > boundary) + len = boundary; + } + + return len; +} + +static void __map_bio(struct dm_target *ti, struct bio *clone, + struct target_io *tio) +{ + int r; + + /* + * Sanity checks. + */ + BUG_ON(!clone->bi_size); + + clone->bi_end_io = clone_endio; + clone->bi_private = tio; + + /* + * Map the clone. If r == 0 we don't need to do + * anything, the target has assumed ownership of + * this io. + */ + atomic_inc(&tio->io->io_count); + r = ti->type->map(ti, clone, &tio->info); + if (r > 0) + /* the bio has been remapped so dispatch it */ + generic_make_request(clone); + + else if (r < 0) { + /* error the io and bail out */ + struct dm_io *io = tio->io; + free_tio(tio->io->md, tio); + dec_pending(io, -EIO); + bio_put(clone); + } +} + +struct clone_info { + struct mapped_device *md; + struct dm_table *map; + struct bio *bio; + struct dm_io *io; + sector_t sector; + sector_t sector_count; + unsigned short idx; +}; + +/* + * Creates a little bio that is just does part of a bvec. + */ +static struct bio *split_bvec(struct bio *bio, sector_t sector, + unsigned short idx, unsigned int offset, + unsigned int len) +{ + struct bio *clone; + struct bio_vec *bv = bio->bi_io_vec + idx; + + clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set); + *clone->bi_io_vec = *bv; + + clone->bi_sector = sector; + clone->bi_bdev = bio->bi_bdev; + clone->bi_rw = bio->bi_rw; + clone->bi_vcnt = 1; + clone->bi_size = to_bytes(len); + clone->bi_io_vec->bv_offset = offset; + clone->bi_io_vec->bv_len = clone->bi_size; + + return clone; +} + +/* + * Creates a bio that consists of range of complete bvecs. + */ +static struct bio *clone_bio(struct bio *bio, sector_t sector, + unsigned short idx, unsigned short bv_count, + unsigned int len) +{ + struct bio *clone; + + clone = bio_clone(bio, GFP_NOIO); + clone->bi_sector = sector; + clone->bi_idx = idx; + clone->bi_vcnt = idx + bv_count; + clone->bi_size = to_bytes(len); + clone->bi_flags &= ~(1 << BIO_SEG_VALID); + + return clone; +} + +static void __clone_and_map(struct clone_info *ci) +{ + struct bio *clone, *bio = ci->bio; + struct dm_target *ti = dm_table_find_target(ci->map, ci->sector); + sector_t len = 0, max = max_io_len(ci->md, ci->sector, ti); + struct target_io *tio; + + /* + * Allocate a target io object. + */ + tio = alloc_tio(ci->md); + tio->io = ci->io; + tio->ti = ti; + memset(&tio->info, 0, sizeof(tio->info)); + + if (ci->sector_count <= max) { + /* + * Optimise for the simple case where we can do all of + * the remaining io with a single clone. + */ + clone = clone_bio(bio, ci->sector, ci->idx, + bio->bi_vcnt - ci->idx, ci->sector_count); + __map_bio(ti, clone, tio); + ci->sector_count = 0; + + } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { + /* + * There are some bvecs that don't span targets. + * Do as many of these as possible. + */ + int i; + sector_t remaining = max; + sector_t bv_len; + + for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { + bv_len = to_sector(bio->bi_io_vec[i].bv_len); + + if (bv_len > remaining) + break; + + remaining -= bv_len; + len += bv_len; + } + + clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len); + __map_bio(ti, clone, tio); + + ci->sector += len; + ci->sector_count -= len; + ci->idx = i; + + } else { + /* + * Create two copy bios to deal with io that has + * been split across a target. + */ + struct bio_vec *bv = bio->bi_io_vec + ci->idx; + + clone = split_bvec(bio, ci->sector, ci->idx, + bv->bv_offset, max); + __map_bio(ti, clone, tio); + + ci->sector += max; + ci->sector_count -= max; + ti = dm_table_find_target(ci->map, ci->sector); + + len = to_sector(bv->bv_len) - max; + clone = split_bvec(bio, ci->sector, ci->idx, + bv->bv_offset + to_bytes(max), len); + tio = alloc_tio(ci->md); + tio->io = ci->io; + tio->ti = ti; + memset(&tio->info, 0, sizeof(tio->info)); + __map_bio(ti, clone, tio); + + ci->sector += len; + ci->sector_count -= len; + ci->idx++; + } +} + +/* + * Split the bio into several clones. + */ +static void __split_bio(struct mapped_device *md, struct bio *bio) +{ + struct clone_info ci; + + ci.map = dm_get_table(md); + if (!ci.map) { + bio_io_error(bio, bio->bi_size); + return; + } + + ci.md = md; + ci.bio = bio; + ci.io = alloc_io(md); + ci.io->error = 0; + atomic_set(&ci.io->io_count, 1); + ci.io->bio = bio; + ci.io->md = md; + ci.sector = bio->bi_sector; + ci.sector_count = bio_sectors(bio); + ci.idx = bio->bi_idx; + + atomic_inc(&md->pending); + while (ci.sector_count) + __clone_and_map(&ci); + + /* drop the extra reference count */ + dec_pending(ci.io, 0); + dm_table_put(ci.map); +} +/*----------------------------------------------------------------- + * CRUD END + *---------------------------------------------------------------*/ + +/* + * The request function that just remaps the bio built up by + * dm_merge_bvec. + */ +static int dm_request(request_queue_t *q, struct bio *bio) +{ + int r; + struct mapped_device *md = q->queuedata; + + down_read(&md->lock); + + /* + * If we're suspended we have to queue + * this io for later. + */ + while (test_bit(DMF_BLOCK_IO, &md->flags)) { + up_read(&md->lock); + + if (bio_rw(bio) == READA) { + bio_io_error(bio, bio->bi_size); + return 0; + } + + r = queue_io(md, bio); + if (r < 0) { + bio_io_error(bio, bio->bi_size); + return 0; + + } else if (r == 0) + return 0; /* deferred successfully */ + + /* + * We're in a while loop, because someone could suspend + * before we get to the following read lock. + */ + down_read(&md->lock); + } + + __split_bio(md, bio); + up_read(&md->lock); + return 0; +} + +static int dm_flush_all(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + struct mapped_device *md = q->queuedata; + struct dm_table *map = dm_get_table(md); + int ret = -ENXIO; + + if (map) { + ret = dm_table_flush_all(md->map); + dm_table_put(map); + } + + return ret; +} + +static void dm_unplug_all(request_queue_t *q) +{ + struct mapped_device *md = q->queuedata; + struct dm_table *map = dm_get_table(md); + + if (map) { + dm_table_unplug_all(map); + dm_table_put(map); + } +} + +static int dm_any_congested(void *congested_data, int bdi_bits) +{ + int r; + struct mapped_device *md = (struct mapped_device *) congested_data; + struct dm_table *map = dm_get_table(md); + + if (!map || test_bit(DMF_BLOCK_IO, &md->flags)) + r = bdi_bits; + else + r = dm_table_any_congested(map, bdi_bits); + + dm_table_put(map); + return r; +} + +/*----------------------------------------------------------------- + * An IDR is used to keep track of allocated minor numbers. + *---------------------------------------------------------------*/ +static DECLARE_MUTEX(_minor_lock); +static DEFINE_IDR(_minor_idr); + +static void free_minor(unsigned int minor) +{ + down(&_minor_lock); + idr_remove(&_minor_idr, minor); + up(&_minor_lock); +} + +/* + * See if the device with a specific minor # is free. + */ +static int specific_minor(struct mapped_device *md, unsigned int minor) +{ + int r, m; + + if (minor >= (1 << MINORBITS)) + return -EINVAL; + + down(&_minor_lock); + + if (idr_find(&_minor_idr, minor)) { + r = -EBUSY; + goto out; + } + + r = idr_pre_get(&_minor_idr, GFP_KERNEL); + if (!r) { + r = -ENOMEM; + goto out; + } + + r = idr_get_new_above(&_minor_idr, md, minor, &m); + if (r) { + goto out; + } + + if (m != minor) { + idr_remove(&_minor_idr, m); + r = -EBUSY; + goto out; + } + +out: + up(&_minor_lock); + return r; +} + +static int next_free_minor(struct mapped_device *md, unsigned int *minor) +{ + int r; + unsigned int m; + + down(&_minor_lock); + + r = idr_pre_get(&_minor_idr, GFP_KERNEL); + if (!r) { + r = -ENOMEM; + goto out; + } + + r = idr_get_new(&_minor_idr, md, &m); + if (r) { + goto out; + } + + if (m >= (1 << MINORBITS)) { + idr_remove(&_minor_idr, m); + r = -ENOSPC; + goto out; + } + + *minor = m; + +out: + up(&_minor_lock); + return r; +} + +static struct block_device_operations dm_blk_dops; + +/* + * Allocate and initialise a blank device with a given minor. + */ +static struct mapped_device *alloc_dev(unsigned int minor, int persistent) +{ + int r; + struct mapped_device *md = kmalloc(sizeof(*md), GFP_KERNEL); + + if (!md) { + DMWARN("unable to allocate device, out of memory."); + return NULL; + } + + /* get a minor number for the dev */ + r = persistent ? specific_minor(md, minor) : next_free_minor(md, &minor); + if (r < 0) + goto bad1; + + memset(md, 0, sizeof(*md)); + init_rwsem(&md->lock); + rwlock_init(&md->map_lock); + atomic_set(&md->holders, 1); + atomic_set(&md->event_nr, 0); + + md->queue = blk_alloc_queue(GFP_KERNEL); + if (!md->queue) + goto bad1; + + md->queue->queuedata = md; + md->queue->backing_dev_info.congested_fn = dm_any_congested; + md->queue->backing_dev_info.congested_data = md; + blk_queue_make_request(md->queue, dm_request); + md->queue->unplug_fn = dm_unplug_all; + md->queue->issue_flush_fn = dm_flush_all; + + md->io_pool = mempool_create(MIN_IOS, mempool_alloc_slab, + mempool_free_slab, _io_cache); + if (!md->io_pool) + goto bad2; + + md->tio_pool = mempool_create(MIN_IOS, mempool_alloc_slab, + mempool_free_slab, _tio_cache); + if (!md->tio_pool) + goto bad3; + + md->disk = alloc_disk(1); + if (!md->disk) + goto bad4; + + md->disk->major = _major; + md->disk->first_minor = minor; + md->disk->fops = &dm_blk_dops; + md->disk->queue = md->queue; + md->disk->private_data = md; + sprintf(md->disk->disk_name, "dm-%d", minor); + add_disk(md->disk); + + atomic_set(&md->pending, 0); + init_waitqueue_head(&md->wait); + init_waitqueue_head(&md->eventq); + + return md; + + bad4: + mempool_destroy(md->tio_pool); + bad3: + mempool_destroy(md->io_pool); + bad2: + blk_put_queue(md->queue); + free_minor(minor); + bad1: + kfree(md); + return NULL; +} + +static void free_dev(struct mapped_device *md) +{ + free_minor(md->disk->first_minor); + mempool_destroy(md->tio_pool); + mempool_destroy(md->io_pool); + del_gendisk(md->disk); + put_disk(md->disk); + blk_put_queue(md->queue); + kfree(md); +} + +/* + * Bind a table to the device. + */ +static void event_callback(void *context) +{ + struct mapped_device *md = (struct mapped_device *) context; + + atomic_inc(&md->event_nr); + wake_up(&md->eventq); +} + +static void __set_size(struct gendisk *disk, sector_t size) +{ + struct block_device *bdev; + + set_capacity(disk, size); + bdev = bdget_disk(disk, 0); + if (bdev) { + down(&bdev->bd_inode->i_sem); + i_size_write(bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); + up(&bdev->bd_inode->i_sem); + bdput(bdev); + } +} + +static int __bind(struct mapped_device *md, struct dm_table *t) +{ + request_queue_t *q = md->queue; + sector_t size; + + size = dm_table_get_size(t); + __set_size(md->disk, size); + if (size == 0) + return 0; + + write_lock(&md->map_lock); + md->map = t; + write_unlock(&md->map_lock); + + dm_table_get(t); + dm_table_event_callback(md->map, event_callback, md); + dm_table_set_restrictions(t, q); + return 0; +} + +static void __unbind(struct mapped_device *md) +{ + struct dm_table *map = md->map; + + if (!map) + return; + + dm_table_event_callback(map, NULL, NULL); + write_lock(&md->map_lock); + md->map = NULL; + write_unlock(&md->map_lock); + dm_table_put(map); +} + +/* + * Constructor for a new device. + */ +static int create_aux(unsigned int minor, int persistent, + struct mapped_device **result) +{ + struct mapped_device *md; + + md = alloc_dev(minor, persistent); + if (!md) + return -ENXIO; + + *result = md; + return 0; +} + +int dm_create(struct mapped_device **result) +{ + return create_aux(0, 0, result); +} + +int dm_create_with_minor(unsigned int minor, struct mapped_device **result) +{ + return create_aux(minor, 1, result); +} + +void *dm_get_mdptr(dev_t dev) +{ + struct mapped_device *md; + void *mdptr = NULL; + unsigned minor = MINOR(dev); + + if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) + return NULL; + + down(&_minor_lock); + + md = idr_find(&_minor_idr, minor); + + if (md && (dm_disk(md)->first_minor == minor)) + mdptr = md->interface_ptr; + + up(&_minor_lock); + + return mdptr; +} + +void dm_set_mdptr(struct mapped_device *md, void *ptr) +{ + md->interface_ptr = ptr; +} + +void dm_get(struct mapped_device *md) +{ + atomic_inc(&md->holders); +} + +void dm_put(struct mapped_device *md) +{ + struct dm_table *map = dm_get_table(md); + + if (atomic_dec_and_test(&md->holders)) { + if (!test_bit(DMF_SUSPENDED, &md->flags) && map) { + dm_table_presuspend_targets(map); + dm_table_postsuspend_targets(map); + } + __unbind(md); + free_dev(md); + } + + dm_table_put(map); +} + +/* + * Process the deferred bios + */ +static void __flush_deferred_io(struct mapped_device *md, struct bio *c) +{ + struct bio *n; + + while (c) { + n = c->bi_next; + c->bi_next = NULL; + __split_bio(md, c); + c = n; + } +} + +/* + * Swap in a new table (destroying old one). + */ +int dm_swap_table(struct mapped_device *md, struct dm_table *table) +{ + int r; + + down_write(&md->lock); + + /* device must be suspended */ + if (!test_bit(DMF_SUSPENDED, &md->flags)) { + up_write(&md->lock); + return -EPERM; + } + + __unbind(md); + r = __bind(md, table); + if (r) + return r; + + up_write(&md->lock); + return 0; +} + +/* + * Functions to lock and unlock any filesystem running on the + * device. + */ +static int __lock_fs(struct mapped_device *md) +{ + struct block_device *bdev; + + if (test_and_set_bit(DMF_FS_LOCKED, &md->flags)) + return 0; + + bdev = bdget_disk(md->disk, 0); + if (!bdev) { + DMWARN("bdget failed in __lock_fs"); + return -ENOMEM; + } + + WARN_ON(md->frozen_sb); + md->frozen_sb = freeze_bdev(bdev); + /* don't bdput right now, we don't want the bdev + * to go away while it is locked. We'll bdput + * in __unlock_fs + */ + return 0; +} + +static int __unlock_fs(struct mapped_device *md) +{ + struct block_device *bdev; + + if (!test_and_clear_bit(DMF_FS_LOCKED, &md->flags)) + return 0; + + bdev = bdget_disk(md->disk, 0); + if (!bdev) { + DMWARN("bdget failed in __unlock_fs"); + return -ENOMEM; + } + + thaw_bdev(bdev, md->frozen_sb); + md->frozen_sb = NULL; + bdput(bdev); + bdput(bdev); + return 0; +} + +/* + * We need to be able to change a mapping table under a mounted + * filesystem. For example we might want to move some data in + * the background. Before the table can be swapped with + * dm_bind_table, dm_suspend must be called to flush any in + * flight bios and ensure that any further io gets deferred. + */ +int dm_suspend(struct mapped_device *md) +{ + struct dm_table *map; + DECLARE_WAITQUEUE(wait, current); + + /* Flush I/O to the device. */ + down_read(&md->lock); + if (test_bit(DMF_BLOCK_IO, &md->flags)) { + up_read(&md->lock); + return -EINVAL; + } + + map = dm_get_table(md); + if (map) + dm_table_presuspend_targets(map); + __lock_fs(md); + + up_read(&md->lock); + + /* + * First we set the BLOCK_IO flag so no more ios will be + * mapped. + */ + down_write(&md->lock); + if (test_bit(DMF_BLOCK_IO, &md->flags)) { + /* + * If we get here we know another thread is + * trying to suspend as well, so we leave the fs + * locked for this thread. + */ + up_write(&md->lock); + return -EINVAL; + } + + set_bit(DMF_BLOCK_IO, &md->flags); + add_wait_queue(&md->wait, &wait); + up_write(&md->lock); + + /* unplug */ + if (map) { + dm_table_unplug_all(map); + dm_table_put(map); + } + + /* + * Then we wait for the already mapped ios to + * complete. + */ + while (1) { + set_current_state(TASK_INTERRUPTIBLE); + + if (!atomic_read(&md->pending) || signal_pending(current)) + break; + + io_schedule(); + } + set_current_state(TASK_RUNNING); + + down_write(&md->lock); + remove_wait_queue(&md->wait, &wait); + + /* were we interrupted ? */ + if (atomic_read(&md->pending)) { + __unlock_fs(md); + clear_bit(DMF_BLOCK_IO, &md->flags); + up_write(&md->lock); + return -EINTR; + } + + set_bit(DMF_SUSPENDED, &md->flags); + + map = dm_get_table(md); + if (map) + dm_table_postsuspend_targets(map); + dm_table_put(map); + up_write(&md->lock); + + return 0; +} + +int dm_resume(struct mapped_device *md) +{ + struct bio *def; + struct dm_table *map = dm_get_table(md); + + down_write(&md->lock); + if (!map || + !test_bit(DMF_SUSPENDED, &md->flags) || + !dm_table_get_size(map)) { + up_write(&md->lock); + dm_table_put(map); + return -EINVAL; + } + + dm_table_resume_targets(map); + clear_bit(DMF_SUSPENDED, &md->flags); + clear_bit(DMF_BLOCK_IO, &md->flags); + + def = bio_list_get(&md->deferred); + __flush_deferred_io(md, def); + up_write(&md->lock); + __unlock_fs(md); + dm_table_unplug_all(map); + dm_table_put(map); + + return 0; +} + +/*----------------------------------------------------------------- + * Event notification. + *---------------------------------------------------------------*/ +uint32_t dm_get_event_nr(struct mapped_device *md) +{ + return atomic_read(&md->event_nr); +} + +int dm_wait_event(struct mapped_device *md, int event_nr) +{ + return wait_event_interruptible(md->eventq, + (event_nr != atomic_read(&md->event_nr))); +} + +/* + * The gendisk is only valid as long as you have a reference + * count on 'md'. + */ +struct gendisk *dm_disk(struct mapped_device *md) +{ + return md->disk; +} + +int dm_suspended(struct mapped_device *md) +{ + return test_bit(DMF_SUSPENDED, &md->flags); +} + +static struct block_device_operations dm_blk_dops = { + .open = dm_blk_open, + .release = dm_blk_close, + .owner = THIS_MODULE +}; + +EXPORT_SYMBOL(dm_get_mapinfo); + +/* + * module hooks + */ +module_init(dm_init); +module_exit(dm_exit); + +module_param(major, uint, 0); +MODULE_PARM_DESC(major, "The major number of the device mapper"); +MODULE_DESCRIPTION(DM_NAME " driver"); +MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/dm.h b/drivers/md/dm.h new file mode 100644 index 00000000000..e38c3fc1a1d --- /dev/null +++ b/drivers/md/dm.h @@ -0,0 +1,195 @@ +/* + * Internal header file for device mapper + * + * Copyright (C) 2001, 2002 Sistina Software + * Copyright (C) 2004 Red Hat, Inc. All rights reserved. + * + * This file is released under the LGPL. + */ + +#ifndef DM_INTERNAL_H +#define DM_INTERNAL_H + +#include <linux/fs.h> +#include <linux/device-mapper.h> +#include <linux/list.h> +#include <linux/blkdev.h> + +#define DM_NAME "device-mapper" +#define DMWARN(f, x...) printk(KERN_WARNING DM_NAME ": " f "\n" , ## x) +#define DMERR(f, x...) printk(KERN_ERR DM_NAME ": " f "\n" , ## x) +#define DMINFO(f, x...) printk(KERN_INFO DM_NAME ": " f "\n" , ## x) + +#define DMEMIT(x...) sz += ((sz >= maxlen) ? \ + 0 : scnprintf(result + sz, maxlen - sz, x)) + +/* + * FIXME: I think this should be with the definition of sector_t + * in types.h. + */ +#ifdef CONFIG_LBD +#define SECTOR_FORMAT "%Lu" +#else +#define SECTOR_FORMAT "%lu" +#endif + +#define SECTOR_SHIFT 9 + +/* + * List of devices that a metadevice uses and should open/close. + */ +struct dm_dev { + struct list_head list; + + atomic_t count; + int mode; + struct block_device *bdev; + char name[16]; +}; + +struct dm_table; +struct mapped_device; + +/*----------------------------------------------------------------- + * Functions for manipulating a struct mapped_device. + * Drop the reference with dm_put when you finish with the object. + *---------------------------------------------------------------*/ +int dm_create(struct mapped_device **md); +int dm_create_with_minor(unsigned int minor, struct mapped_device **md); +void dm_set_mdptr(struct mapped_device *md, void *ptr); +void *dm_get_mdptr(dev_t dev); + +/* + * Reference counting for md. + */ +void dm_get(struct mapped_device *md); +void dm_put(struct mapped_device *md); + +/* + * A device can still be used while suspended, but I/O is deferred. + */ +int dm_suspend(struct mapped_device *md); +int dm_resume(struct mapped_device *md); + +/* + * The device must be suspended before calling this method. + */ +int dm_swap_table(struct mapped_device *md, struct dm_table *t); + +/* + * Drop a reference on the table when you've finished with the + * result. + */ +struct dm_table *dm_get_table(struct mapped_device *md); + +/* + * Event functions. + */ +uint32_t dm_get_event_nr(struct mapped_device *md); +int dm_wait_event(struct mapped_device *md, int event_nr); + +/* + * Info functions. + */ +struct gendisk *dm_disk(struct mapped_device *md); +int dm_suspended(struct mapped_device *md); + +/*----------------------------------------------------------------- + * Functions for manipulating a table. Tables are also reference + * counted. + *---------------------------------------------------------------*/ +int dm_table_create(struct dm_table **result, int mode, unsigned num_targets); + +void dm_table_get(struct dm_table *t); +void dm_table_put(struct dm_table *t); + +int dm_table_add_target(struct dm_table *t, const char *type, + sector_t start, sector_t len, char *params); +int dm_table_complete(struct dm_table *t); +void dm_table_event_callback(struct dm_table *t, + void (*fn)(void *), void *context); +void dm_table_event(struct dm_table *t); +sector_t dm_table_get_size(struct dm_table *t); +struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index); +struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector); +void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q); +unsigned int dm_table_get_num_targets(struct dm_table *t); +struct list_head *dm_table_get_devices(struct dm_table *t); +int dm_table_get_mode(struct dm_table *t); +void dm_table_presuspend_targets(struct dm_table *t); +void dm_table_postsuspend_targets(struct dm_table *t); +void dm_table_resume_targets(struct dm_table *t); +int dm_table_any_congested(struct dm_table *t, int bdi_bits); +void dm_table_unplug_all(struct dm_table *t); +int dm_table_flush_all(struct dm_table *t); + +/*----------------------------------------------------------------- + * A registry of target types. + *---------------------------------------------------------------*/ +int dm_target_init(void); +void dm_target_exit(void); +struct target_type *dm_get_target_type(const char *name); +void dm_put_target_type(struct target_type *t); +int dm_target_iterate(void (*iter_func)(struct target_type *tt, + void *param), void *param); + + +/*----------------------------------------------------------------- + * Useful inlines. + *---------------------------------------------------------------*/ +static inline int array_too_big(unsigned long fixed, unsigned long obj, + unsigned long num) +{ + return (num > (ULONG_MAX - fixed) / obj); +} + +/* + * Ceiling(n / sz) + */ +#define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz)) + +#define dm_sector_div_up(n, sz) ( \ +{ \ + sector_t _r = ((n) + (sz) - 1); \ + sector_div(_r, (sz)); \ + _r; \ +} \ +) + +/* + * ceiling(n / size) * size + */ +#define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz)) + +static inline sector_t to_sector(unsigned long n) +{ + return (n >> 9); +} + +static inline unsigned long to_bytes(sector_t n) +{ + return (n << 9); +} + +int dm_split_args(int *argc, char ***argvp, char *input); + +/* + * The device-mapper can be driven through one of two interfaces; + * ioctl or filesystem, depending which patch you have applied. + */ +int dm_interface_init(void); +void dm_interface_exit(void); + +/* + * Targets for linear and striped mappings + */ +int dm_linear_init(void); +void dm_linear_exit(void); + +int dm_stripe_init(void); +void dm_stripe_exit(void); + +void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size); +union map_info *dm_get_mapinfo(struct bio *bio); + +#endif diff --git a/drivers/md/faulty.c b/drivers/md/faulty.c new file mode 100644 index 00000000000..0248f8e7eac --- /dev/null +++ b/drivers/md/faulty.c @@ -0,0 +1,343 @@ +/* + * faulty.c : Multiple Devices driver for Linux + * + * Copyright (C) 2004 Neil Brown + * + * fautly-device-simulator personality for md + * + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + + +/* + * The "faulty" personality causes some requests to fail. + * + * Possible failure modes are: + * reads fail "randomly" but succeed on retry + * writes fail "randomly" but succeed on retry + * reads for some address fail and then persist until a write + * reads for some address fail and then persist irrespective of write + * writes for some address fail and persist + * all writes fail + * + * Different modes can be active at a time, but only + * one can be set at array creation. Others can be added later. + * A mode can be one-shot or recurrent with the recurrance being + * once in every N requests. + * The bottom 5 bits of the "layout" indicate the mode. The + * remainder indicate a period, or 0 for one-shot. + * + * There is an implementation limit on the number of concurrently + * persisting-faulty blocks. When a new fault is requested that would + * exceed the limit, it is ignored. + * All current faults can be clear using a layout of "0". + * + * Requests are always sent to the device. If they are to fail, + * we clone the bio and insert a new b_end_io into the chain. + */ + +#define WriteTransient 0 +#define ReadTransient 1 +#define WritePersistent 2 +#define ReadPersistent 3 +#define WriteAll 4 /* doesn't go to device */ +#define ReadFixable 5 +#define Modes 6 + +#define ClearErrors 31 +#define ClearFaults 30 + +#define AllPersist 100 /* internal use only */ +#define NoPersist 101 + +#define ModeMask 0x1f +#define ModeShift 5 + +#define MaxFault 50 +#include <linux/raid/md.h> + + +static int faulty_fail(struct bio *bio, unsigned int bytes_done, int error) +{ + struct bio *b = bio->bi_private; + + b->bi_size = bio->bi_size; + b->bi_sector = bio->bi_sector; + + if (bio->bi_size == 0) + bio_put(bio); + + clear_bit(BIO_UPTODATE, &b->bi_flags); + return (b->bi_end_io)(b, bytes_done, -EIO); +} + +typedef struct faulty_conf { + int period[Modes]; + atomic_t counters[Modes]; + sector_t faults[MaxFault]; + int modes[MaxFault]; + int nfaults; + mdk_rdev_t *rdev; +} conf_t; + +static int check_mode(conf_t *conf, int mode) +{ + if (conf->period[mode] == 0 && + atomic_read(&conf->counters[mode]) <= 0) + return 0; /* no failure, no decrement */ + + + if (atomic_dec_and_test(&conf->counters[mode])) { + if (conf->period[mode]) + atomic_set(&conf->counters[mode], conf->period[mode]); + return 1; + } + return 0; +} + +static int check_sector(conf_t *conf, sector_t start, sector_t end, int dir) +{ + /* If we find a ReadFixable sector, we fix it ... */ + int i; + for (i=0; i<conf->nfaults; i++) + if (conf->faults[i] >= start && + conf->faults[i] < end) { + /* found it ... */ + switch (conf->modes[i] * 2 + dir) { + case WritePersistent*2+WRITE: return 1; + case ReadPersistent*2+READ: return 1; + case ReadFixable*2+READ: return 1; + case ReadFixable*2+WRITE: + conf->modes[i] = NoPersist; + return 0; + case AllPersist*2+READ: + case AllPersist*2+WRITE: return 1; + default: + return 0; + } + } + return 0; +} + +static void add_sector(conf_t *conf, sector_t start, int mode) +{ + int i; + int n = conf->nfaults; + for (i=0; i<conf->nfaults; i++) + if (conf->faults[i] == start) { + switch(mode) { + case NoPersist: conf->modes[i] = mode; return; + case WritePersistent: + if (conf->modes[i] == ReadPersistent || + conf->modes[i] == ReadFixable) + conf->modes[i] = AllPersist; + else + conf->modes[i] = WritePersistent; + return; + case ReadPersistent: + if (conf->modes[i] == WritePersistent) + conf->modes[i] = AllPersist; + else + conf->modes[i] = ReadPersistent; + return; + case ReadFixable: + if (conf->modes[i] == WritePersistent || + conf->modes[i] == ReadPersistent) + conf->modes[i] = AllPersist; + else + conf->modes[i] = ReadFixable; + return; + } + } else if (conf->modes[i] == NoPersist) + n = i; + + if (n >= MaxFault) + return; + conf->faults[n] = start; + conf->modes[n] = mode; + if (conf->nfaults == n) + conf->nfaults = n+1; +} + +static int make_request(request_queue_t *q, struct bio *bio) +{ + mddev_t *mddev = q->queuedata; + conf_t *conf = (conf_t*)mddev->private; + int failit = 0; + + if (bio->bi_rw & 1) { + /* write request */ + if (atomic_read(&conf->counters[WriteAll])) { + /* special case - don't decrement, don't generic_make_request, + * just fail immediately + */ + bio_endio(bio, bio->bi_size, -EIO); + return 0; + } + + if (check_sector(conf, bio->bi_sector, bio->bi_sector+(bio->bi_size>>9), + WRITE)) + failit = 1; + if (check_mode(conf, WritePersistent)) { + add_sector(conf, bio->bi_sector, WritePersistent); + failit = 1; + } + if (check_mode(conf, WriteTransient)) + failit = 1; + } else { + /* read request */ + if (check_sector(conf, bio->bi_sector, bio->bi_sector + (bio->bi_size>>9), + READ)) + failit = 1; + if (check_mode(conf, ReadTransient)) + failit = 1; + if (check_mode(conf, ReadPersistent)) { + add_sector(conf, bio->bi_sector, ReadPersistent); + failit = 1; + } + if (check_mode(conf, ReadFixable)) { + add_sector(conf, bio->bi_sector, ReadFixable); + failit = 1; + } + } + if (failit) { + struct bio *b = bio_clone(bio, GFP_NOIO); + b->bi_bdev = conf->rdev->bdev; + b->bi_private = bio; + b->bi_end_io = faulty_fail; + generic_make_request(b); + return 0; + } else { + bio->bi_bdev = conf->rdev->bdev; + return 1; + } +} + +static void status(struct seq_file *seq, mddev_t *mddev) +{ + conf_t *conf = (conf_t*)mddev->private; + int n; + + if ((n=atomic_read(&conf->counters[WriteTransient])) != 0) + seq_printf(seq, " WriteTransient=%d(%d)", + n, conf->period[WriteTransient]); + + if ((n=atomic_read(&conf->counters[ReadTransient])) != 0) + seq_printf(seq, " ReadTransient=%d(%d)", + n, conf->period[ReadTransient]); + + if ((n=atomic_read(&conf->counters[WritePersistent])) != 0) + seq_printf(seq, " WritePersistent=%d(%d)", + n, conf->period[WritePersistent]); + + if ((n=atomic_read(&conf->counters[ReadPersistent])) != 0) + seq_printf(seq, " ReadPersistent=%d(%d)", + n, conf->period[ReadPersistent]); + + + if ((n=atomic_read(&conf->counters[ReadFixable])) != 0) + seq_printf(seq, " ReadFixable=%d(%d)", + n, conf->period[ReadFixable]); + + if ((n=atomic_read(&conf->counters[WriteAll])) != 0) + seq_printf(seq, " WriteAll"); + + seq_printf(seq, " nfaults=%d", conf->nfaults); +} + + +static int reconfig(mddev_t *mddev, int layout, int chunk_size) +{ + int mode = layout & ModeMask; + int count = layout >> ModeShift; + conf_t *conf = mddev->private; + + if (chunk_size != -1) + return -EINVAL; + + /* new layout */ + if (mode == ClearFaults) + conf->nfaults = 0; + else if (mode == ClearErrors) { + int i; + for (i=0 ; i < Modes ; i++) { + conf->period[i] = 0; + atomic_set(&conf->counters[i], 0); + } + } else if (mode < Modes) { + conf->period[mode] = count; + if (!count) count++; + atomic_set(&conf->counters[mode], count); + } else + return -EINVAL; + mddev->layout = -1; /* makes sure further changes come through */ + return 0; +} + +static int run(mddev_t *mddev) +{ + mdk_rdev_t *rdev; + struct list_head *tmp; + int i; + + conf_t *conf = kmalloc(sizeof(*conf), GFP_KERNEL); + + for (i=0; i<Modes; i++) { + atomic_set(&conf->counters[i], 0); + conf->period[i] = 0; + } + conf->nfaults = 0; + + ITERATE_RDEV(mddev, rdev, tmp) + conf->rdev = rdev; + + mddev->array_size = mddev->size; + mddev->private = conf; + + reconfig(mddev, mddev->layout, -1); + + return 0; +} + +static int stop(mddev_t *mddev) +{ + conf_t *conf = (conf_t *)mddev->private; + + kfree(conf); + mddev->private = NULL; + return 0; +} + +static mdk_personality_t faulty_personality = +{ + .name = "faulty", + .owner = THIS_MODULE, + .make_request = make_request, + .run = run, + .stop = stop, + .status = status, + .reconfig = reconfig, +}; + +static int __init raid_init(void) +{ + return register_md_personality(FAULTY, &faulty_personality); +} + +static void raid_exit(void) +{ + unregister_md_personality(FAULTY); +} + +module_init(raid_init); +module_exit(raid_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-10"); /* faulty */ diff --git a/drivers/md/kcopyd.c b/drivers/md/kcopyd.c new file mode 100644 index 00000000000..eb703648597 --- /dev/null +++ b/drivers/md/kcopyd.c @@ -0,0 +1,687 @@ +/* + * Copyright (C) 2002 Sistina Software (UK) Limited. + * + * This file is released under the GPL. + * + * Kcopyd provides a simple interface for copying an area of one + * block-device to one or more other block-devices, with an asynchronous + * completion notification. + */ + +#include <asm/atomic.h> + +#include <linux/blkdev.h> +#include <linux/config.h> +#include <linux/fs.h> +#include <linux/init.h> +#include <linux/list.h> +#include <linux/mempool.h> +#include <linux/module.h> +#include <linux/pagemap.h> +#include <linux/slab.h> +#include <linux/vmalloc.h> +#include <linux/workqueue.h> + +#include "kcopyd.h" + +static struct workqueue_struct *_kcopyd_wq; +static struct work_struct _kcopyd_work; + +static inline void wake(void) +{ + queue_work(_kcopyd_wq, &_kcopyd_work); +} + +/*----------------------------------------------------------------- + * Each kcopyd client has its own little pool of preallocated + * pages for kcopyd io. + *---------------------------------------------------------------*/ +struct kcopyd_client { + struct list_head list; + + spinlock_t lock; + struct page_list *pages; + unsigned int nr_pages; + unsigned int nr_free_pages; +}; + +static struct page_list *alloc_pl(void) +{ + struct page_list *pl; + + pl = kmalloc(sizeof(*pl), GFP_KERNEL); + if (!pl) + return NULL; + + pl->page = alloc_page(GFP_KERNEL); + if (!pl->page) { + kfree(pl); + return NULL; + } + + return pl; +} + +static void free_pl(struct page_list *pl) +{ + __free_page(pl->page); + kfree(pl); +} + +static int kcopyd_get_pages(struct kcopyd_client *kc, + unsigned int nr, struct page_list **pages) +{ + struct page_list *pl; + + spin_lock(&kc->lock); + if (kc->nr_free_pages < nr) { + spin_unlock(&kc->lock); + return -ENOMEM; + } + + kc->nr_free_pages -= nr; + for (*pages = pl = kc->pages; --nr; pl = pl->next) + ; + + kc->pages = pl->next; + pl->next = NULL; + + spin_unlock(&kc->lock); + + return 0; +} + +static void kcopyd_put_pages(struct kcopyd_client *kc, struct page_list *pl) +{ + struct page_list *cursor; + + spin_lock(&kc->lock); + for (cursor = pl; cursor->next; cursor = cursor->next) + kc->nr_free_pages++; + + kc->nr_free_pages++; + cursor->next = kc->pages; + kc->pages = pl; + spin_unlock(&kc->lock); +} + +/* + * These three functions resize the page pool. + */ +static void drop_pages(struct page_list *pl) +{ + struct page_list *next; + + while (pl) { + next = pl->next; + free_pl(pl); + pl = next; + } +} + +static int client_alloc_pages(struct kcopyd_client *kc, unsigned int nr) +{ + unsigned int i; + struct page_list *pl = NULL, *next; + + for (i = 0; i < nr; i++) { + next = alloc_pl(); + if (!next) { + if (pl) + drop_pages(pl); + return -ENOMEM; + } + next->next = pl; + pl = next; + } + + kcopyd_put_pages(kc, pl); + kc->nr_pages += nr; + return 0; +} + +static void client_free_pages(struct kcopyd_client *kc) +{ + BUG_ON(kc->nr_free_pages != kc->nr_pages); + drop_pages(kc->pages); + kc->pages = NULL; + kc->nr_free_pages = kc->nr_pages = 0; +} + +/*----------------------------------------------------------------- + * kcopyd_jobs need to be allocated by the *clients* of kcopyd, + * for this reason we use a mempool to prevent the client from + * ever having to do io (which could cause a deadlock). + *---------------------------------------------------------------*/ +struct kcopyd_job { + struct kcopyd_client *kc; + struct list_head list; + unsigned long flags; + + /* + * Error state of the job. + */ + int read_err; + unsigned int write_err; + + /* + * Either READ or WRITE + */ + int rw; + struct io_region source; + + /* + * The destinations for the transfer. + */ + unsigned int num_dests; + struct io_region dests[KCOPYD_MAX_REGIONS]; + + sector_t offset; + unsigned int nr_pages; + struct page_list *pages; + + /* + * Set this to ensure you are notified when the job has + * completed. 'context' is for callback to use. + */ + kcopyd_notify_fn fn; + void *context; + + /* + * These fields are only used if the job has been split + * into more manageable parts. + */ + struct semaphore lock; + atomic_t sub_jobs; + sector_t progress; +}; + +/* FIXME: this should scale with the number of pages */ +#define MIN_JOBS 512 + +static kmem_cache_t *_job_cache; +static mempool_t *_job_pool; + +/* + * We maintain three lists of jobs: + * + * i) jobs waiting for pages + * ii) jobs that have pages, and are waiting for the io to be issued. + * iii) jobs that have completed. + * + * All three of these are protected by job_lock. + */ +static DEFINE_SPINLOCK(_job_lock); + +static LIST_HEAD(_complete_jobs); +static LIST_HEAD(_io_jobs); +static LIST_HEAD(_pages_jobs); + +static int jobs_init(void) +{ + _job_cache = kmem_cache_create("kcopyd-jobs", + sizeof(struct kcopyd_job), + __alignof__(struct kcopyd_job), + 0, NULL, NULL); + if (!_job_cache) + return -ENOMEM; + + _job_pool = mempool_create(MIN_JOBS, mempool_alloc_slab, + mempool_free_slab, _job_cache); + if (!_job_pool) { + kmem_cache_destroy(_job_cache); + return -ENOMEM; + } + + return 0; +} + +static void jobs_exit(void) +{ + BUG_ON(!list_empty(&_complete_jobs)); + BUG_ON(!list_empty(&_io_jobs)); + BUG_ON(!list_empty(&_pages_jobs)); + + mempool_destroy(_job_pool); + kmem_cache_destroy(_job_cache); + _job_pool = NULL; + _job_cache = NULL; +} + +/* + * Functions to push and pop a job onto the head of a given job + * list. + */ +static inline struct kcopyd_job *pop(struct list_head *jobs) +{ + struct kcopyd_job *job = NULL; + unsigned long flags; + + spin_lock_irqsave(&_job_lock, flags); + + if (!list_empty(jobs)) { + job = list_entry(jobs->next, struct kcopyd_job, list); + list_del(&job->list); + } + spin_unlock_irqrestore(&_job_lock, flags); + + return job; +} + +static inline void push(struct list_head *jobs, struct kcopyd_job *job) +{ + unsigned long flags; + + spin_lock_irqsave(&_job_lock, flags); + list_add_tail(&job->list, jobs); + spin_unlock_irqrestore(&_job_lock, flags); +} + +/* + * These three functions process 1 item from the corresponding + * job list. + * + * They return: + * < 0: error + * 0: success + * > 0: can't process yet. + */ +static int run_complete_job(struct kcopyd_job *job) +{ + void *context = job->context; + int read_err = job->read_err; + unsigned int write_err = job->write_err; + kcopyd_notify_fn fn = job->fn; + + kcopyd_put_pages(job->kc, job->pages); + mempool_free(job, _job_pool); + fn(read_err, write_err, context); + return 0; +} + +static void complete_io(unsigned long error, void *context) +{ + struct kcopyd_job *job = (struct kcopyd_job *) context; + + if (error) { + if (job->rw == WRITE) + job->write_err &= error; + else + job->read_err = 1; + + if (!test_bit(KCOPYD_IGNORE_ERROR, &job->flags)) { + push(&_complete_jobs, job); + wake(); + return; + } + } + + if (job->rw == WRITE) + push(&_complete_jobs, job); + + else { + job->rw = WRITE; + push(&_io_jobs, job); + } + + wake(); +} + +/* + * Request io on as many buffer heads as we can currently get for + * a particular job. + */ +static int run_io_job(struct kcopyd_job *job) +{ + int r; + + if (job->rw == READ) + r = dm_io_async(1, &job->source, job->rw, + job->pages, + job->offset, complete_io, job); + + else + r = dm_io_async(job->num_dests, job->dests, job->rw, + job->pages, + job->offset, complete_io, job); + + return r; +} + +static int run_pages_job(struct kcopyd_job *job) +{ + int r; + + job->nr_pages = dm_div_up(job->dests[0].count + job->offset, + PAGE_SIZE >> 9); + r = kcopyd_get_pages(job->kc, job->nr_pages, &job->pages); + if (!r) { + /* this job is ready for io */ + push(&_io_jobs, job); + return 0; + } + + if (r == -ENOMEM) + /* can't complete now */ + return 1; + + return r; +} + +/* + * Run through a list for as long as possible. Returns the count + * of successful jobs. + */ +static int process_jobs(struct list_head *jobs, int (*fn) (struct kcopyd_job *)) +{ + struct kcopyd_job *job; + int r, count = 0; + + while ((job = pop(jobs))) { + + r = fn(job); + + if (r < 0) { + /* error this rogue job */ + if (job->rw == WRITE) + job->write_err = (unsigned int) -1; + else + job->read_err = 1; + push(&_complete_jobs, job); + break; + } + + if (r > 0) { + /* + * We couldn't service this job ATM, so + * push this job back onto the list. + */ + push(jobs, job); + break; + } + + count++; + } + + return count; +} + +/* + * kcopyd does this every time it's woken up. + */ +static void do_work(void *ignored) +{ + /* + * The order that these are called is *very* important. + * complete jobs can free some pages for pages jobs. + * Pages jobs when successful will jump onto the io jobs + * list. io jobs call wake when they complete and it all + * starts again. + */ + process_jobs(&_complete_jobs, run_complete_job); + process_jobs(&_pages_jobs, run_pages_job); + process_jobs(&_io_jobs, run_io_job); +} + +/* + * If we are copying a small region we just dispatch a single job + * to do the copy, otherwise the io has to be split up into many + * jobs. + */ +static void dispatch_job(struct kcopyd_job *job) +{ + push(&_pages_jobs, job); + wake(); +} + +#define SUB_JOB_SIZE 128 +static void segment_complete(int read_err, + unsigned int write_err, void *context) +{ + /* FIXME: tidy this function */ + sector_t progress = 0; + sector_t count = 0; + struct kcopyd_job *job = (struct kcopyd_job *) context; + + down(&job->lock); + + /* update the error */ + if (read_err) + job->read_err = 1; + + if (write_err) + job->write_err &= write_err; + + /* + * Only dispatch more work if there hasn't been an error. + */ + if ((!job->read_err && !job->write_err) || + test_bit(KCOPYD_IGNORE_ERROR, &job->flags)) { + /* get the next chunk of work */ + progress = job->progress; + count = job->source.count - progress; + if (count) { + if (count > SUB_JOB_SIZE) + count = SUB_JOB_SIZE; + + job->progress += count; + } + } + up(&job->lock); + + if (count) { + int i; + struct kcopyd_job *sub_job = mempool_alloc(_job_pool, GFP_NOIO); + + *sub_job = *job; + sub_job->source.sector += progress; + sub_job->source.count = count; + + for (i = 0; i < job->num_dests; i++) { + sub_job->dests[i].sector += progress; + sub_job->dests[i].count = count; + } + + sub_job->fn = segment_complete; + sub_job->context = job; + dispatch_job(sub_job); + + } else if (atomic_dec_and_test(&job->sub_jobs)) { + + /* + * To avoid a race we must keep the job around + * until after the notify function has completed. + * Otherwise the client may try and stop the job + * after we've completed. + */ + job->fn(read_err, write_err, job->context); + mempool_free(job, _job_pool); + } +} + +/* + * Create some little jobs that will do the move between + * them. + */ +#define SPLIT_COUNT 8 +static void split_job(struct kcopyd_job *job) +{ + int i; + + atomic_set(&job->sub_jobs, SPLIT_COUNT); + for (i = 0; i < SPLIT_COUNT; i++) + segment_complete(0, 0u, job); +} + +int kcopyd_copy(struct kcopyd_client *kc, struct io_region *from, + unsigned int num_dests, struct io_region *dests, + unsigned int flags, kcopyd_notify_fn fn, void *context) +{ + struct kcopyd_job *job; + + /* + * Allocate a new job. + */ + job = mempool_alloc(_job_pool, GFP_NOIO); + + /* + * set up for the read. + */ + job->kc = kc; + job->flags = flags; + job->read_err = 0; + job->write_err = 0; + job->rw = READ; + + job->source = *from; + + job->num_dests = num_dests; + memcpy(&job->dests, dests, sizeof(*dests) * num_dests); + + job->offset = 0; + job->nr_pages = 0; + job->pages = NULL; + + job->fn = fn; + job->context = context; + + if (job->source.count < SUB_JOB_SIZE) + dispatch_job(job); + + else { + init_MUTEX(&job->lock); + job->progress = 0; + split_job(job); + } + + return 0; +} + +/* + * Cancels a kcopyd job, eg. someone might be deactivating a + * mirror. + */ +int kcopyd_cancel(struct kcopyd_job *job, int block) +{ + /* FIXME: finish */ + return -1; +} + +/*----------------------------------------------------------------- + * Unit setup + *---------------------------------------------------------------*/ +static DECLARE_MUTEX(_client_lock); +static LIST_HEAD(_clients); + +static void client_add(struct kcopyd_client *kc) +{ + down(&_client_lock); + list_add(&kc->list, &_clients); + up(&_client_lock); +} + +static void client_del(struct kcopyd_client *kc) +{ + down(&_client_lock); + list_del(&kc->list); + up(&_client_lock); +} + +static DECLARE_MUTEX(kcopyd_init_lock); +static int kcopyd_clients = 0; + +static int kcopyd_init(void) +{ + int r; + + down(&kcopyd_init_lock); + + if (kcopyd_clients) { + /* Already initialized. */ + kcopyd_clients++; + up(&kcopyd_init_lock); + return 0; + } + + r = jobs_init(); + if (r) { + up(&kcopyd_init_lock); + return r; + } + + _kcopyd_wq = create_singlethread_workqueue("kcopyd"); + if (!_kcopyd_wq) { + jobs_exit(); + up(&kcopyd_init_lock); + return -ENOMEM; + } + + kcopyd_clients++; + INIT_WORK(&_kcopyd_work, do_work, NULL); + up(&kcopyd_init_lock); + return 0; +} + +static void kcopyd_exit(void) +{ + down(&kcopyd_init_lock); + kcopyd_clients--; + if (!kcopyd_clients) { + jobs_exit(); + destroy_workqueue(_kcopyd_wq); + _kcopyd_wq = NULL; + } + up(&kcopyd_init_lock); +} + +int kcopyd_client_create(unsigned int nr_pages, struct kcopyd_client **result) +{ + int r = 0; + struct kcopyd_client *kc; + + r = kcopyd_init(); + if (r) + return r; + + kc = kmalloc(sizeof(*kc), GFP_KERNEL); + if (!kc) { + kcopyd_exit(); + return -ENOMEM; + } + + spin_lock_init(&kc->lock); + kc->pages = NULL; + kc->nr_pages = kc->nr_free_pages = 0; + r = client_alloc_pages(kc, nr_pages); + if (r) { + kfree(kc); + kcopyd_exit(); + return r; + } + + r = dm_io_get(nr_pages); + if (r) { + client_free_pages(kc); + kfree(kc); + kcopyd_exit(); + return r; + } + + client_add(kc); + *result = kc; + return 0; +} + +void kcopyd_client_destroy(struct kcopyd_client *kc) +{ + dm_io_put(kc->nr_pages); + client_free_pages(kc); + client_del(kc); + kfree(kc); + kcopyd_exit(); +} + +EXPORT_SYMBOL(kcopyd_client_create); +EXPORT_SYMBOL(kcopyd_client_destroy); +EXPORT_SYMBOL(kcopyd_copy); +EXPORT_SYMBOL(kcopyd_cancel); diff --git a/drivers/md/kcopyd.h b/drivers/md/kcopyd.h new file mode 100644 index 00000000000..4621ea055c0 --- /dev/null +++ b/drivers/md/kcopyd.h @@ -0,0 +1,42 @@ +/* + * Copyright (C) 2001 Sistina Software + * + * This file is released under the GPL. + * + * Kcopyd provides a simple interface for copying an area of one + * block-device to one or more other block-devices, with an asynchronous + * completion notification. + */ + +#ifndef DM_KCOPYD_H +#define DM_KCOPYD_H + +#include "dm-io.h" + +/* FIXME: make this configurable */ +#define KCOPYD_MAX_REGIONS 8 + +#define KCOPYD_IGNORE_ERROR 1 + +/* + * To use kcopyd you must first create a kcopyd client object. + */ +struct kcopyd_client; +int kcopyd_client_create(unsigned int num_pages, struct kcopyd_client **result); +void kcopyd_client_destroy(struct kcopyd_client *kc); + +/* + * Submit a copy job to kcopyd. This is built on top of the + * previous three fns. + * + * read_err is a boolean, + * write_err is a bitset, with 1 bit for each destination region + */ +typedef void (*kcopyd_notify_fn)(int read_err, + unsigned int write_err, void *context); + +int kcopyd_copy(struct kcopyd_client *kc, struct io_region *from, + unsigned int num_dests, struct io_region *dests, + unsigned int flags, kcopyd_notify_fn fn, void *context); + +#endif diff --git a/drivers/md/linear.c b/drivers/md/linear.c new file mode 100644 index 00000000000..161e9aa8729 --- /dev/null +++ b/drivers/md/linear.c @@ -0,0 +1,343 @@ +/* + linear.c : Multiple Devices driver for Linux + Copyright (C) 1994-96 Marc ZYNGIER + <zyngier@ufr-info-p7.ibp.fr> or + <maz@gloups.fdn.fr> + + Linear mode management functions. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + You should have received a copy of the GNU General Public License + (for example /usr/src/linux/COPYING); if not, write to the Free + Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. +*/ + +#include <linux/module.h> + +#include <linux/raid/md.h> +#include <linux/slab.h> +#include <linux/raid/linear.h> + +#define MAJOR_NR MD_MAJOR +#define MD_DRIVER +#define MD_PERSONALITY + +/* + * find which device holds a particular offset + */ +static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector) +{ + dev_info_t *hash; + linear_conf_t *conf = mddev_to_conf(mddev); + sector_t block = sector >> 1; + + /* + * sector_div(a,b) returns the remainer and sets a to a/b + */ + (void)sector_div(block, conf->smallest->size); + hash = conf->hash_table[block]; + + while ((sector>>1) >= (hash->size + hash->offset)) + hash++; + return hash; +} + +/** + * linear_mergeable_bvec -- tell bio layer if a two requests can be merged + * @q: request queue + * @bio: the buffer head that's been built up so far + * @biovec: the request that could be merged to it. + * + * Return amount of bytes we can take at this offset + */ +static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec) +{ + mddev_t *mddev = q->queuedata; + dev_info_t *dev0; + unsigned long maxsectors, bio_sectors = bio->bi_size >> 9; + sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); + + dev0 = which_dev(mddev, sector); + maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1)); + + if (maxsectors < bio_sectors) + maxsectors = 0; + else + maxsectors -= bio_sectors; + + if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0) + return biovec->bv_len; + /* The bytes available at this offset could be really big, + * so we cap at 2^31 to avoid overflow */ + if (maxsectors > (1 << (31-9))) + return 1<<31; + return maxsectors << 9; +} + +static void linear_unplug(request_queue_t *q) +{ + mddev_t *mddev = q->queuedata; + linear_conf_t *conf = mddev_to_conf(mddev); + int i; + + for (i=0; i < mddev->raid_disks; i++) { + request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev); + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + } +} + +static int linear_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + linear_conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + for (i=0; i < mddev->raid_disks && ret == 0; i++) { + struct block_device *bdev = conf->disks[i].rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector); + } + return ret; +} + +static int linear_run (mddev_t *mddev) +{ + linear_conf_t *conf; + dev_info_t **table; + mdk_rdev_t *rdev; + int i, nb_zone, cnt; + sector_t start; + sector_t curr_offset; + struct list_head *tmp; + + conf = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t), + GFP_KERNEL); + if (!conf) + goto out; + memset(conf, 0, sizeof(*conf) + mddev->raid_disks*sizeof(dev_info_t)); + mddev->private = conf; + + /* + * Find the smallest device. + */ + + conf->smallest = NULL; + cnt = 0; + mddev->array_size = 0; + + ITERATE_RDEV(mddev,rdev,tmp) { + int j = rdev->raid_disk; + dev_info_t *disk = conf->disks + j; + + if (j < 0 || j > mddev->raid_disks || disk->rdev) { + printk("linear: disk numbering problem. Aborting!\n"); + goto out; + } + + disk->rdev = rdev; + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + disk->size = rdev->size; + mddev->array_size += rdev->size; + + if (!conf->smallest || (disk->size < conf->smallest->size)) + conf->smallest = disk; + cnt++; + } + if (cnt != mddev->raid_disks) { + printk("linear: not enough drives present. Aborting!\n"); + goto out; + } + + /* + * This code was restructured to work around a gcc-2.95.3 internal + * compiler error. Alter it with care. + */ + { + sector_t sz; + unsigned round; + unsigned long base; + + sz = mddev->array_size; + base = conf->smallest->size; + round = sector_div(sz, base); + nb_zone = conf->nr_zones = sz + (round ? 1 : 0); + } + + conf->hash_table = kmalloc (sizeof (dev_info_t*) * nb_zone, + GFP_KERNEL); + if (!conf->hash_table) + goto out; + + /* + * Here we generate the linear hash table + */ + table = conf->hash_table; + start = 0; + curr_offset = 0; + for (i = 0; i < cnt; i++) { + dev_info_t *disk = conf->disks + i; + + disk->offset = curr_offset; + curr_offset += disk->size; + + /* 'curr_offset' is the end of this disk + * 'start' is the start of table + */ + while (start < curr_offset) { + *table++ = disk; + start += conf->smallest->size; + } + } + if (table-conf->hash_table != nb_zone) + BUG(); + + blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec); + mddev->queue->unplug_fn = linear_unplug; + mddev->queue->issue_flush_fn = linear_issue_flush; + return 0; + +out: + if (conf) + kfree(conf); + return 1; +} + +static int linear_stop (mddev_t *mddev) +{ + linear_conf_t *conf = mddev_to_conf(mddev); + + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + kfree(conf->hash_table); + kfree(conf); + + return 0; +} + +static int linear_make_request (request_queue_t *q, struct bio *bio) +{ + mddev_t *mddev = q->queuedata; + dev_info_t *tmp_dev; + sector_t block; + + if (bio_data_dir(bio)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); + } + + tmp_dev = which_dev(mddev, bio->bi_sector); + block = bio->bi_sector >> 1; + + if (unlikely(block >= (tmp_dev->size + tmp_dev->offset) + || block < tmp_dev->offset)) { + char b[BDEVNAME_SIZE]; + + printk("linear_make_request: Block %llu out of bounds on " + "dev %s size %llu offset %llu\n", + (unsigned long long)block, + bdevname(tmp_dev->rdev->bdev, b), + (unsigned long long)tmp_dev->size, + (unsigned long long)tmp_dev->offset); + bio_io_error(bio, bio->bi_size); + return 0; + } + if (unlikely(bio->bi_sector + (bio->bi_size >> 9) > + (tmp_dev->offset + tmp_dev->size)<<1)) { + /* This bio crosses a device boundary, so we have to + * split it. + */ + struct bio_pair *bp; + bp = bio_split(bio, bio_split_pool, + (bio->bi_sector + (bio->bi_size >> 9) - + (tmp_dev->offset + tmp_dev->size))<<1); + if (linear_make_request(q, &bp->bio1)) + generic_make_request(&bp->bio1); + if (linear_make_request(q, &bp->bio2)) + generic_make_request(&bp->bio2); + bio_pair_release(bp); + return 0; + } + + bio->bi_bdev = tmp_dev->rdev->bdev; + bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset; + + return 1; +} + +static void linear_status (struct seq_file *seq, mddev_t *mddev) +{ + +#undef MD_DEBUG +#ifdef MD_DEBUG + int j; + linear_conf_t *conf = mddev_to_conf(mddev); + sector_t s = 0; + + seq_printf(seq, " "); + for (j = 0; j < conf->nr_zones; j++) + { + char b[BDEVNAME_SIZE]; + s += conf->smallest_size; + seq_printf(seq, "[%s", + bdevname(conf->hash_table[j][0].rdev->bdev,b)); + + while (s > conf->hash_table[j][0].offset + + conf->hash_table[j][0].size) + seq_printf(seq, "/%s] ", + bdevname(conf->hash_table[j][1].rdev->bdev,b)); + else + seq_printf(seq, "] "); + } + seq_printf(seq, "\n"); +#endif + seq_printf(seq, " %dk rounding", mddev->chunk_size/1024); +} + + +static mdk_personality_t linear_personality= +{ + .name = "linear", + .owner = THIS_MODULE, + .make_request = linear_make_request, + .run = linear_run, + .stop = linear_stop, + .status = linear_status, +}; + +static int __init linear_init (void) +{ + return register_md_personality (LINEAR, &linear_personality); +} + +static void linear_exit (void) +{ + unregister_md_personality (LINEAR); +} + + +module_init(linear_init); +module_exit(linear_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-1"); /* LINEAR */ diff --git a/drivers/md/md.c b/drivers/md/md.c new file mode 100644 index 00000000000..04562add192 --- /dev/null +++ b/drivers/md/md.c @@ -0,0 +1,3766 @@ +/* + md.c : Multiple Devices driver for Linux + Copyright (C) 1998, 1999, 2000 Ingo Molnar + + completely rewritten, based on the MD driver code from Marc Zyngier + + Changes: + + - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar + - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com> + - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net> + - kerneld support by Boris Tobotras <boris@xtalk.msk.su> + - kmod support by: Cyrus Durgin + - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com> + - Devfs support by Richard Gooch <rgooch@atnf.csiro.au> + + - lots of fixes and improvements to the RAID1/RAID5 and generic + RAID code (such as request based resynchronization): + + Neil Brown <neilb@cse.unsw.edu.au>. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + You should have received a copy of the GNU General Public License + (for example /usr/src/linux/COPYING); if not, write to the Free + Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. +*/ + +#include <linux/module.h> +#include <linux/config.h> +#include <linux/linkage.h> +#include <linux/raid/md.h> +#include <linux/sysctl.h> +#include <linux/devfs_fs_kernel.h> +#include <linux/buffer_head.h> /* for invalidate_bdev */ +#include <linux/suspend.h> + +#include <linux/init.h> + +#ifdef CONFIG_KMOD +#include <linux/kmod.h> +#endif + +#include <asm/unaligned.h> + +#define MAJOR_NR MD_MAJOR +#define MD_DRIVER + +/* 63 partitions with the alternate major number (mdp) */ +#define MdpMinorShift 6 + +#define DEBUG 0 +#define dprintk(x...) ((void)(DEBUG && printk(x))) + + +#ifndef MODULE +static void autostart_arrays (int part); +#endif + +static mdk_personality_t *pers[MAX_PERSONALITY]; +static DEFINE_SPINLOCK(pers_lock); + +/* + * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit' + * is 1000 KB/sec, so the extra system load does not show up that much. + * Increase it if you want to have more _guaranteed_ speed. Note that + * the RAID driver will use the maximum available bandwith if the IO + * subsystem is idle. There is also an 'absolute maximum' reconstruction + * speed limit - in case reconstruction slows down your system despite + * idle IO detection. + * + * you can change it via /proc/sys/dev/raid/speed_limit_min and _max. + */ + +static int sysctl_speed_limit_min = 1000; +static int sysctl_speed_limit_max = 200000; + +static struct ctl_table_header *raid_table_header; + +static ctl_table raid_table[] = { + { + .ctl_name = DEV_RAID_SPEED_LIMIT_MIN, + .procname = "speed_limit_min", + .data = &sysctl_speed_limit_min, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, + { + .ctl_name = DEV_RAID_SPEED_LIMIT_MAX, + .procname = "speed_limit_max", + .data = &sysctl_speed_limit_max, + .maxlen = sizeof(int), + .mode = 0644, + .proc_handler = &proc_dointvec, + }, + { .ctl_name = 0 } +}; + +static ctl_table raid_dir_table[] = { + { + .ctl_name = DEV_RAID, + .procname = "raid", + .maxlen = 0, + .mode = 0555, + .child = raid_table, + }, + { .ctl_name = 0 } +}; + +static ctl_table raid_root_table[] = { + { + .ctl_name = CTL_DEV, + .procname = "dev", + .maxlen = 0, + .mode = 0555, + .child = raid_dir_table, + }, + { .ctl_name = 0 } +}; + +static struct block_device_operations md_fops; + +/* + * Enables to iterate over all existing md arrays + * all_mddevs_lock protects this list. + */ +static LIST_HEAD(all_mddevs); +static DEFINE_SPINLOCK(all_mddevs_lock); + + +/* + * iterates through all used mddevs in the system. + * We take care to grab the all_mddevs_lock whenever navigating + * the list, and to always hold a refcount when unlocked. + * Any code which breaks out of this loop while own + * a reference to the current mddev and must mddev_put it. + */ +#define ITERATE_MDDEV(mddev,tmp) \ + \ + for (({ spin_lock(&all_mddevs_lock); \ + tmp = all_mddevs.next; \ + mddev = NULL;}); \ + ({ if (tmp != &all_mddevs) \ + mddev_get(list_entry(tmp, mddev_t, all_mddevs));\ + spin_unlock(&all_mddevs_lock); \ + if (mddev) mddev_put(mddev); \ + mddev = list_entry(tmp, mddev_t, all_mddevs); \ + tmp != &all_mddevs;}); \ + ({ spin_lock(&all_mddevs_lock); \ + tmp = tmp->next;}) \ + ) + + +static int md_fail_request (request_queue_t *q, struct bio *bio) +{ + bio_io_error(bio, bio->bi_size); + return 0; +} + +static inline mddev_t *mddev_get(mddev_t *mddev) +{ + atomic_inc(&mddev->active); + return mddev; +} + +static void mddev_put(mddev_t *mddev) +{ + if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock)) + return; + if (!mddev->raid_disks && list_empty(&mddev->disks)) { + list_del(&mddev->all_mddevs); + blk_put_queue(mddev->queue); + kfree(mddev); + } + spin_unlock(&all_mddevs_lock); +} + +static mddev_t * mddev_find(dev_t unit) +{ + mddev_t *mddev, *new = NULL; + + retry: + spin_lock(&all_mddevs_lock); + list_for_each_entry(mddev, &all_mddevs, all_mddevs) + if (mddev->unit == unit) { + mddev_get(mddev); + spin_unlock(&all_mddevs_lock); + if (new) + kfree(new); + return mddev; + } + + if (new) { + list_add(&new->all_mddevs, &all_mddevs); + spin_unlock(&all_mddevs_lock); + return new; + } + spin_unlock(&all_mddevs_lock); + + new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL); + if (!new) + return NULL; + + memset(new, 0, sizeof(*new)); + + new->unit = unit; + if (MAJOR(unit) == MD_MAJOR) + new->md_minor = MINOR(unit); + else + new->md_minor = MINOR(unit) >> MdpMinorShift; + + init_MUTEX(&new->reconfig_sem); + INIT_LIST_HEAD(&new->disks); + INIT_LIST_HEAD(&new->all_mddevs); + init_timer(&new->safemode_timer); + atomic_set(&new->active, 1); + + new->queue = blk_alloc_queue(GFP_KERNEL); + if (!new->queue) { + kfree(new); + return NULL; + } + + blk_queue_make_request(new->queue, md_fail_request); + + goto retry; +} + +static inline int mddev_lock(mddev_t * mddev) +{ + return down_interruptible(&mddev->reconfig_sem); +} + +static inline void mddev_lock_uninterruptible(mddev_t * mddev) +{ + down(&mddev->reconfig_sem); +} + +static inline int mddev_trylock(mddev_t * mddev) +{ + return down_trylock(&mddev->reconfig_sem); +} + +static inline void mddev_unlock(mddev_t * mddev) +{ + up(&mddev->reconfig_sem); + + if (mddev->thread) + md_wakeup_thread(mddev->thread); +} + +mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr) +{ + mdk_rdev_t * rdev; + struct list_head *tmp; + + ITERATE_RDEV(mddev,rdev,tmp) { + if (rdev->desc_nr == nr) + return rdev; + } + return NULL; +} + +static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev) +{ + struct list_head *tmp; + mdk_rdev_t *rdev; + + ITERATE_RDEV(mddev,rdev,tmp) { + if (rdev->bdev->bd_dev == dev) + return rdev; + } + return NULL; +} + +inline static sector_t calc_dev_sboffset(struct block_device *bdev) +{ + sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; + return MD_NEW_SIZE_BLOCKS(size); +} + +static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size) +{ + sector_t size; + + size = rdev->sb_offset; + + if (chunk_size) + size &= ~((sector_t)chunk_size/1024 - 1); + return size; +} + +static int alloc_disk_sb(mdk_rdev_t * rdev) +{ + if (rdev->sb_page) + MD_BUG(); + + rdev->sb_page = alloc_page(GFP_KERNEL); + if (!rdev->sb_page) { + printk(KERN_ALERT "md: out of memory.\n"); + return -EINVAL; + } + + return 0; +} + +static void free_disk_sb(mdk_rdev_t * rdev) +{ + if (rdev->sb_page) { + page_cache_release(rdev->sb_page); + rdev->sb_loaded = 0; + rdev->sb_page = NULL; + rdev->sb_offset = 0; + rdev->size = 0; + } +} + + +static int bi_complete(struct bio *bio, unsigned int bytes_done, int error) +{ + if (bio->bi_size) + return 1; + + complete((struct completion*)bio->bi_private); + return 0; +} + +static int sync_page_io(struct block_device *bdev, sector_t sector, int size, + struct page *page, int rw) +{ + struct bio *bio = bio_alloc(GFP_KERNEL, 1); + struct completion event; + int ret; + + rw |= (1 << BIO_RW_SYNC); + + bio->bi_bdev = bdev; + bio->bi_sector = sector; + bio_add_page(bio, page, size, 0); + init_completion(&event); + bio->bi_private = &event; + bio->bi_end_io = bi_complete; + submit_bio(rw, bio); + wait_for_completion(&event); + + ret = test_bit(BIO_UPTODATE, &bio->bi_flags); + bio_put(bio); + return ret; +} + +static int read_disk_sb(mdk_rdev_t * rdev) +{ + char b[BDEVNAME_SIZE]; + if (!rdev->sb_page) { + MD_BUG(); + return -EINVAL; + } + if (rdev->sb_loaded) + return 0; + + + if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ)) + goto fail; + rdev->sb_loaded = 1; + return 0; + +fail: + printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n", + bdevname(rdev->bdev,b)); + return -EINVAL; +} + +static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2) +{ + if ( (sb1->set_uuid0 == sb2->set_uuid0) && + (sb1->set_uuid1 == sb2->set_uuid1) && + (sb1->set_uuid2 == sb2->set_uuid2) && + (sb1->set_uuid3 == sb2->set_uuid3)) + + return 1; + + return 0; +} + + +static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2) +{ + int ret; + mdp_super_t *tmp1, *tmp2; + + tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL); + tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL); + + if (!tmp1 || !tmp2) { + ret = 0; + printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n"); + goto abort; + } + + *tmp1 = *sb1; + *tmp2 = *sb2; + + /* + * nr_disks is not constant + */ + tmp1->nr_disks = 0; + tmp2->nr_disks = 0; + + if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4)) + ret = 0; + else + ret = 1; + +abort: + if (tmp1) + kfree(tmp1); + if (tmp2) + kfree(tmp2); + + return ret; +} + +static unsigned int calc_sb_csum(mdp_super_t * sb) +{ + unsigned int disk_csum, csum; + + disk_csum = sb->sb_csum; + sb->sb_csum = 0; + csum = csum_partial((void *)sb, MD_SB_BYTES, 0); + sb->sb_csum = disk_csum; + return csum; +} + + +/* + * Handle superblock details. + * We want to be able to handle multiple superblock formats + * so we have a common interface to them all, and an array of + * different handlers. + * We rely on user-space to write the initial superblock, and support + * reading and updating of superblocks. + * Interface methods are: + * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version) + * loads and validates a superblock on dev. + * if refdev != NULL, compare superblocks on both devices + * Return: + * 0 - dev has a superblock that is compatible with refdev + * 1 - dev has a superblock that is compatible and newer than refdev + * so dev should be used as the refdev in future + * -EINVAL superblock incompatible or invalid + * -othererror e.g. -EIO + * + * int validate_super(mddev_t *mddev, mdk_rdev_t *dev) + * Verify that dev is acceptable into mddev. + * The first time, mddev->raid_disks will be 0, and data from + * dev should be merged in. Subsequent calls check that dev + * is new enough. Return 0 or -EINVAL + * + * void sync_super(mddev_t *mddev, mdk_rdev_t *dev) + * Update the superblock for rdev with data in mddev + * This does not write to disc. + * + */ + +struct super_type { + char *name; + struct module *owner; + int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version); + int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev); + void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev); +}; + +/* + * load_super for 0.90.0 + */ +static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) +{ + char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; + mdp_super_t *sb; + int ret; + sector_t sb_offset; + + /* + * Calculate the position of the superblock, + * it's at the end of the disk. + * + * It also happens to be a multiple of 4Kb. + */ + sb_offset = calc_dev_sboffset(rdev->bdev); + rdev->sb_offset = sb_offset; + + ret = read_disk_sb(rdev); + if (ret) return ret; + + ret = -EINVAL; + + bdevname(rdev->bdev, b); + sb = (mdp_super_t*)page_address(rdev->sb_page); + + if (sb->md_magic != MD_SB_MAGIC) { + printk(KERN_ERR "md: invalid raid superblock magic on %s\n", + b); + goto abort; + } + + if (sb->major_version != 0 || + sb->minor_version != 90) { + printk(KERN_WARNING "Bad version number %d.%d on %s\n", + sb->major_version, sb->minor_version, + b); + goto abort; + } + + if (sb->raid_disks <= 0) + goto abort; + + if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) { + printk(KERN_WARNING "md: invalid superblock checksum on %s\n", + b); + goto abort; + } + + rdev->preferred_minor = sb->md_minor; + rdev->data_offset = 0; + + if (sb->level == LEVEL_MULTIPATH) + rdev->desc_nr = -1; + else + rdev->desc_nr = sb->this_disk.number; + + if (refdev == 0) + ret = 1; + else { + __u64 ev1, ev2; + mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page); + if (!uuid_equal(refsb, sb)) { + printk(KERN_WARNING "md: %s has different UUID to %s\n", + b, bdevname(refdev->bdev,b2)); + goto abort; + } + if (!sb_equal(refsb, sb)) { + printk(KERN_WARNING "md: %s has same UUID" + " but different superblock to %s\n", + b, bdevname(refdev->bdev, b2)); + goto abort; + } + ev1 = md_event(sb); + ev2 = md_event(refsb); + if (ev1 > ev2) + ret = 1; + else + ret = 0; + } + rdev->size = calc_dev_size(rdev, sb->chunk_size); + + abort: + return ret; +} + +/* + * validate_super for 0.90.0 + */ +static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev) +{ + mdp_disk_t *desc; + mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page); + + if (mddev->raid_disks == 0) { + mddev->major_version = 0; + mddev->minor_version = sb->minor_version; + mddev->patch_version = sb->patch_version; + mddev->persistent = ! sb->not_persistent; + mddev->chunk_size = sb->chunk_size; + mddev->ctime = sb->ctime; + mddev->utime = sb->utime; + mddev->level = sb->level; + mddev->layout = sb->layout; + mddev->raid_disks = sb->raid_disks; + mddev->size = sb->size; + mddev->events = md_event(sb); + + if (sb->state & (1<<MD_SB_CLEAN)) + mddev->recovery_cp = MaxSector; + else { + if (sb->events_hi == sb->cp_events_hi && + sb->events_lo == sb->cp_events_lo) { + mddev->recovery_cp = sb->recovery_cp; + } else + mddev->recovery_cp = 0; + } + + memcpy(mddev->uuid+0, &sb->set_uuid0, 4); + memcpy(mddev->uuid+4, &sb->set_uuid1, 4); + memcpy(mddev->uuid+8, &sb->set_uuid2, 4); + memcpy(mddev->uuid+12,&sb->set_uuid3, 4); + + mddev->max_disks = MD_SB_DISKS; + } else { + __u64 ev1; + ev1 = md_event(sb); + ++ev1; + if (ev1 < mddev->events) + return -EINVAL; + } + if (mddev->level != LEVEL_MULTIPATH) { + rdev->raid_disk = -1; + rdev->in_sync = rdev->faulty = 0; + desc = sb->disks + rdev->desc_nr; + + if (desc->state & (1<<MD_DISK_FAULTY)) + rdev->faulty = 1; + else if (desc->state & (1<<MD_DISK_SYNC) && + desc->raid_disk < mddev->raid_disks) { + rdev->in_sync = 1; + rdev->raid_disk = desc->raid_disk; + } + } + return 0; +} + +/* + * sync_super for 0.90.0 + */ +static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev) +{ + mdp_super_t *sb; + struct list_head *tmp; + mdk_rdev_t *rdev2; + int next_spare = mddev->raid_disks; + + /* make rdev->sb match mddev data.. + * + * 1/ zero out disks + * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare); + * 3/ any empty disks < next_spare become removed + * + * disks[0] gets initialised to REMOVED because + * we cannot be sure from other fields if it has + * been initialised or not. + */ + int i; + int active=0, working=0,failed=0,spare=0,nr_disks=0; + + sb = (mdp_super_t*)page_address(rdev->sb_page); + + memset(sb, 0, sizeof(*sb)); + + sb->md_magic = MD_SB_MAGIC; + sb->major_version = mddev->major_version; + sb->minor_version = mddev->minor_version; + sb->patch_version = mddev->patch_version; + sb->gvalid_words = 0; /* ignored */ + memcpy(&sb->set_uuid0, mddev->uuid+0, 4); + memcpy(&sb->set_uuid1, mddev->uuid+4, 4); + memcpy(&sb->set_uuid2, mddev->uuid+8, 4); + memcpy(&sb->set_uuid3, mddev->uuid+12,4); + + sb->ctime = mddev->ctime; + sb->level = mddev->level; + sb->size = mddev->size; + sb->raid_disks = mddev->raid_disks; + sb->md_minor = mddev->md_minor; + sb->not_persistent = !mddev->persistent; + sb->utime = mddev->utime; + sb->state = 0; + sb->events_hi = (mddev->events>>32); + sb->events_lo = (u32)mddev->events; + + if (mddev->in_sync) + { + sb->recovery_cp = mddev->recovery_cp; + sb->cp_events_hi = (mddev->events>>32); + sb->cp_events_lo = (u32)mddev->events; + if (mddev->recovery_cp == MaxSector) + sb->state = (1<< MD_SB_CLEAN); + } else + sb->recovery_cp = 0; + + sb->layout = mddev->layout; + sb->chunk_size = mddev->chunk_size; + + sb->disks[0].state = (1<<MD_DISK_REMOVED); + ITERATE_RDEV(mddev,rdev2,tmp) { + mdp_disk_t *d; + if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty) + rdev2->desc_nr = rdev2->raid_disk; + else + rdev2->desc_nr = next_spare++; + d = &sb->disks[rdev2->desc_nr]; + nr_disks++; + d->number = rdev2->desc_nr; + d->major = MAJOR(rdev2->bdev->bd_dev); + d->minor = MINOR(rdev2->bdev->bd_dev); + if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty) + d->raid_disk = rdev2->raid_disk; + else + d->raid_disk = rdev2->desc_nr; /* compatibility */ + if (rdev2->faulty) { + d->state = (1<<MD_DISK_FAULTY); + failed++; + } else if (rdev2->in_sync) { + d->state = (1<<MD_DISK_ACTIVE); + d->state |= (1<<MD_DISK_SYNC); + active++; + working++; + } else { + d->state = 0; + spare++; + working++; + } + } + + /* now set the "removed" and "faulty" bits on any missing devices */ + for (i=0 ; i < mddev->raid_disks ; i++) { + mdp_disk_t *d = &sb->disks[i]; + if (d->state == 0 && d->number == 0) { + d->number = i; + d->raid_disk = i; + d->state = (1<<MD_DISK_REMOVED); + d->state |= (1<<MD_DISK_FAULTY); + failed++; + } + } + sb->nr_disks = nr_disks; + sb->active_disks = active; + sb->working_disks = working; + sb->failed_disks = failed; + sb->spare_disks = spare; + + sb->this_disk = sb->disks[rdev->desc_nr]; + sb->sb_csum = calc_sb_csum(sb); +} + +/* + * version 1 superblock + */ + +static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb) +{ + unsigned int disk_csum, csum; + unsigned long long newcsum; + int size = 256 + le32_to_cpu(sb->max_dev)*2; + unsigned int *isuper = (unsigned int*)sb; + int i; + + disk_csum = sb->sb_csum; + sb->sb_csum = 0; + newcsum = 0; + for (i=0; size>=4; size -= 4 ) + newcsum += le32_to_cpu(*isuper++); + + if (size == 2) + newcsum += le16_to_cpu(*(unsigned short*) isuper); + + csum = (newcsum & 0xffffffff) + (newcsum >> 32); + sb->sb_csum = disk_csum; + return cpu_to_le32(csum); +} + +static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version) +{ + struct mdp_superblock_1 *sb; + int ret; + sector_t sb_offset; + char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; + + /* + * Calculate the position of the superblock. + * It is always aligned to a 4K boundary and + * depeding on minor_version, it can be: + * 0: At least 8K, but less than 12K, from end of device + * 1: At start of device + * 2: 4K from start of device. + */ + switch(minor_version) { + case 0: + sb_offset = rdev->bdev->bd_inode->i_size >> 9; + sb_offset -= 8*2; + sb_offset &= ~(4*2-1); + /* convert from sectors to K */ + sb_offset /= 2; + break; + case 1: + sb_offset = 0; + break; + case 2: + sb_offset = 4; + break; + default: + return -EINVAL; + } + rdev->sb_offset = sb_offset; + + ret = read_disk_sb(rdev); + if (ret) return ret; + + + sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); + + if (sb->magic != cpu_to_le32(MD_SB_MAGIC) || + sb->major_version != cpu_to_le32(1) || + le32_to_cpu(sb->max_dev) > (4096-256)/2 || + le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) || + sb->feature_map != 0) + return -EINVAL; + + if (calc_sb_1_csum(sb) != sb->sb_csum) { + printk("md: invalid superblock checksum on %s\n", + bdevname(rdev->bdev,b)); + return -EINVAL; + } + if (le64_to_cpu(sb->data_size) < 10) { + printk("md: data_size too small on %s\n", + bdevname(rdev->bdev,b)); + return -EINVAL; + } + rdev->preferred_minor = 0xffff; + rdev->data_offset = le64_to_cpu(sb->data_offset); + + if (refdev == 0) + return 1; + else { + __u64 ev1, ev2; + struct mdp_superblock_1 *refsb = + (struct mdp_superblock_1*)page_address(refdev->sb_page); + + if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 || + sb->level != refsb->level || + sb->layout != refsb->layout || + sb->chunksize != refsb->chunksize) { + printk(KERN_WARNING "md: %s has strangely different" + " superblock to %s\n", + bdevname(rdev->bdev,b), + bdevname(refdev->bdev,b2)); + return -EINVAL; + } + ev1 = le64_to_cpu(sb->events); + ev2 = le64_to_cpu(refsb->events); + + if (ev1 > ev2) + return 1; + } + if (minor_version) + rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2; + else + rdev->size = rdev->sb_offset; + if (rdev->size < le64_to_cpu(sb->data_size)/2) + return -EINVAL; + rdev->size = le64_to_cpu(sb->data_size)/2; + if (le32_to_cpu(sb->chunksize)) + rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1); + return 0; +} + +static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev) +{ + struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); + + if (mddev->raid_disks == 0) { + mddev->major_version = 1; + mddev->patch_version = 0; + mddev->persistent = 1; + mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9; + mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1); + mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1); + mddev->level = le32_to_cpu(sb->level); + mddev->layout = le32_to_cpu(sb->layout); + mddev->raid_disks = le32_to_cpu(sb->raid_disks); + mddev->size = le64_to_cpu(sb->size)/2; + mddev->events = le64_to_cpu(sb->events); + + mddev->recovery_cp = le64_to_cpu(sb->resync_offset); + memcpy(mddev->uuid, sb->set_uuid, 16); + + mddev->max_disks = (4096-256)/2; + } else { + __u64 ev1; + ev1 = le64_to_cpu(sb->events); + ++ev1; + if (ev1 < mddev->events) + return -EINVAL; + } + + if (mddev->level != LEVEL_MULTIPATH) { + int role; + rdev->desc_nr = le32_to_cpu(sb->dev_number); + role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]); + switch(role) { + case 0xffff: /* spare */ + rdev->in_sync = 0; + rdev->faulty = 0; + rdev->raid_disk = -1; + break; + case 0xfffe: /* faulty */ + rdev->in_sync = 0; + rdev->faulty = 1; + rdev->raid_disk = -1; + break; + default: + rdev->in_sync = 1; + rdev->faulty = 0; + rdev->raid_disk = role; + break; + } + } + return 0; +} + +static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev) +{ + struct mdp_superblock_1 *sb; + struct list_head *tmp; + mdk_rdev_t *rdev2; + int max_dev, i; + /* make rdev->sb match mddev and rdev data. */ + + sb = (struct mdp_superblock_1*)page_address(rdev->sb_page); + + sb->feature_map = 0; + sb->pad0 = 0; + memset(sb->pad1, 0, sizeof(sb->pad1)); + memset(sb->pad2, 0, sizeof(sb->pad2)); + memset(sb->pad3, 0, sizeof(sb->pad3)); + + sb->utime = cpu_to_le64((__u64)mddev->utime); + sb->events = cpu_to_le64(mddev->events); + if (mddev->in_sync) + sb->resync_offset = cpu_to_le64(mddev->recovery_cp); + else + sb->resync_offset = cpu_to_le64(0); + + max_dev = 0; + ITERATE_RDEV(mddev,rdev2,tmp) + if (rdev2->desc_nr+1 > max_dev) + max_dev = rdev2->desc_nr+1; + + sb->max_dev = cpu_to_le32(max_dev); + for (i=0; i<max_dev;i++) + sb->dev_roles[i] = cpu_to_le16(0xfffe); + + ITERATE_RDEV(mddev,rdev2,tmp) { + i = rdev2->desc_nr; + if (rdev2->faulty) + sb->dev_roles[i] = cpu_to_le16(0xfffe); + else if (rdev2->in_sync) + sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk); + else + sb->dev_roles[i] = cpu_to_le16(0xffff); + } + + sb->recovery_offset = cpu_to_le64(0); /* not supported yet */ + sb->sb_csum = calc_sb_1_csum(sb); +} + + +struct super_type super_types[] = { + [0] = { + .name = "0.90.0", + .owner = THIS_MODULE, + .load_super = super_90_load, + .validate_super = super_90_validate, + .sync_super = super_90_sync, + }, + [1] = { + .name = "md-1", + .owner = THIS_MODULE, + .load_super = super_1_load, + .validate_super = super_1_validate, + .sync_super = super_1_sync, + }, +}; + +static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev) +{ + struct list_head *tmp; + mdk_rdev_t *rdev; + + ITERATE_RDEV(mddev,rdev,tmp) + if (rdev->bdev->bd_contains == dev->bdev->bd_contains) + return rdev; + + return NULL; +} + +static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2) +{ + struct list_head *tmp; + mdk_rdev_t *rdev; + + ITERATE_RDEV(mddev1,rdev,tmp) + if (match_dev_unit(mddev2, rdev)) + return 1; + + return 0; +} + +static LIST_HEAD(pending_raid_disks); + +static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev) +{ + mdk_rdev_t *same_pdev; + char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; + + if (rdev->mddev) { + MD_BUG(); + return -EINVAL; + } + same_pdev = match_dev_unit(mddev, rdev); + if (same_pdev) + printk(KERN_WARNING + "%s: WARNING: %s appears to be on the same physical" + " disk as %s. True\n protection against single-disk" + " failure might be compromised.\n", + mdname(mddev), bdevname(rdev->bdev,b), + bdevname(same_pdev->bdev,b2)); + + /* Verify rdev->desc_nr is unique. + * If it is -1, assign a free number, else + * check number is not in use + */ + if (rdev->desc_nr < 0) { + int choice = 0; + if (mddev->pers) choice = mddev->raid_disks; + while (find_rdev_nr(mddev, choice)) + choice++; + rdev->desc_nr = choice; + } else { + if (find_rdev_nr(mddev, rdev->desc_nr)) + return -EBUSY; + } + + list_add(&rdev->same_set, &mddev->disks); + rdev->mddev = mddev; + printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b)); + return 0; +} + +static void unbind_rdev_from_array(mdk_rdev_t * rdev) +{ + char b[BDEVNAME_SIZE]; + if (!rdev->mddev) { + MD_BUG(); + return; + } + list_del_init(&rdev->same_set); + printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b)); + rdev->mddev = NULL; +} + +/* + * prevent the device from being mounted, repartitioned or + * otherwise reused by a RAID array (or any other kernel + * subsystem), by bd_claiming the device. + */ +static int lock_rdev(mdk_rdev_t *rdev, dev_t dev) +{ + int err = 0; + struct block_device *bdev; + char b[BDEVNAME_SIZE]; + + bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE); + if (IS_ERR(bdev)) { + printk(KERN_ERR "md: could not open %s.\n", + __bdevname(dev, b)); + return PTR_ERR(bdev); + } + err = bd_claim(bdev, rdev); + if (err) { + printk(KERN_ERR "md: could not bd_claim %s.\n", + bdevname(bdev, b)); + blkdev_put(bdev); + return err; + } + rdev->bdev = bdev; + return err; +} + +static void unlock_rdev(mdk_rdev_t *rdev) +{ + struct block_device *bdev = rdev->bdev; + rdev->bdev = NULL; + if (!bdev) + MD_BUG(); + bd_release(bdev); + blkdev_put(bdev); +} + +void md_autodetect_dev(dev_t dev); + +static void export_rdev(mdk_rdev_t * rdev) +{ + char b[BDEVNAME_SIZE]; + printk(KERN_INFO "md: export_rdev(%s)\n", + bdevname(rdev->bdev,b)); + if (rdev->mddev) + MD_BUG(); + free_disk_sb(rdev); + list_del_init(&rdev->same_set); +#ifndef MODULE + md_autodetect_dev(rdev->bdev->bd_dev); +#endif + unlock_rdev(rdev); + kfree(rdev); +} + +static void kick_rdev_from_array(mdk_rdev_t * rdev) +{ + unbind_rdev_from_array(rdev); + export_rdev(rdev); +} + +static void export_array(mddev_t *mddev) +{ + struct list_head *tmp; + mdk_rdev_t *rdev; + + ITERATE_RDEV(mddev,rdev,tmp) { + if (!rdev->mddev) { + MD_BUG(); + continue; + } + kick_rdev_from_array(rdev); + } + if (!list_empty(&mddev->disks)) + MD_BUG(); + mddev->raid_disks = 0; + mddev->major_version = 0; +} + +static void print_desc(mdp_disk_t *desc) +{ + printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number, + desc->major,desc->minor,desc->raid_disk,desc->state); +} + +static void print_sb(mdp_super_t *sb) +{ + int i; + + printk(KERN_INFO + "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n", + sb->major_version, sb->minor_version, sb->patch_version, + sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3, + sb->ctime); + printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n", + sb->level, sb->size, sb->nr_disks, sb->raid_disks, + sb->md_minor, sb->layout, sb->chunk_size); + printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d" + " FD:%d SD:%d CSUM:%08x E:%08lx\n", + sb->utime, sb->state, sb->active_disks, sb->working_disks, + sb->failed_disks, sb->spare_disks, + sb->sb_csum, (unsigned long)sb->events_lo); + + printk(KERN_INFO); + for (i = 0; i < MD_SB_DISKS; i++) { + mdp_disk_t *desc; + + desc = sb->disks + i; + if (desc->number || desc->major || desc->minor || + desc->raid_disk || (desc->state && (desc->state != 4))) { + printk(" D %2d: ", i); + print_desc(desc); + } + } + printk(KERN_INFO "md: THIS: "); + print_desc(&sb->this_disk); + +} + +static void print_rdev(mdk_rdev_t *rdev) +{ + char b[BDEVNAME_SIZE]; + printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n", + bdevname(rdev->bdev,b), (unsigned long long)rdev->size, + rdev->faulty, rdev->in_sync, rdev->desc_nr); + if (rdev->sb_loaded) { + printk(KERN_INFO "md: rdev superblock:\n"); + print_sb((mdp_super_t*)page_address(rdev->sb_page)); + } else + printk(KERN_INFO "md: no rdev superblock!\n"); +} + +void md_print_devices(void) +{ + struct list_head *tmp, *tmp2; + mdk_rdev_t *rdev; + mddev_t *mddev; + char b[BDEVNAME_SIZE]; + + printk("\n"); + printk("md: **********************************\n"); + printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n"); + printk("md: **********************************\n"); + ITERATE_MDDEV(mddev,tmp) { + printk("%s: ", mdname(mddev)); + + ITERATE_RDEV(mddev,rdev,tmp2) + printk("<%s>", bdevname(rdev->bdev,b)); + printk("\n"); + + ITERATE_RDEV(mddev,rdev,tmp2) + print_rdev(rdev); + } + printk("md: **********************************\n"); + printk("\n"); +} + + +static int write_disk_sb(mdk_rdev_t * rdev) +{ + char b[BDEVNAME_SIZE]; + if (!rdev->sb_loaded) { + MD_BUG(); + return 1; + } + if (rdev->faulty) { + MD_BUG(); + return 1; + } + + dprintk(KERN_INFO "(write) %s's sb offset: %llu\n", + bdevname(rdev->bdev,b), + (unsigned long long)rdev->sb_offset); + + if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE)) + return 0; + + printk("md: write_disk_sb failed for device %s\n", + bdevname(rdev->bdev,b)); + return 1; +} + +static void sync_sbs(mddev_t * mddev) +{ + mdk_rdev_t *rdev; + struct list_head *tmp; + + ITERATE_RDEV(mddev,rdev,tmp) { + super_types[mddev->major_version]. + sync_super(mddev, rdev); + rdev->sb_loaded = 1; + } +} + +static void md_update_sb(mddev_t * mddev) +{ + int err, count = 100; + struct list_head *tmp; + mdk_rdev_t *rdev; + + mddev->sb_dirty = 0; +repeat: + mddev->utime = get_seconds(); + mddev->events ++; + + if (!mddev->events) { + /* + * oops, this 64-bit counter should never wrap. + * Either we are in around ~1 trillion A.C., assuming + * 1 reboot per second, or we have a bug: + */ + MD_BUG(); + mddev->events --; + } + sync_sbs(mddev); + + /* + * do not write anything to disk if using + * nonpersistent superblocks + */ + if (!mddev->persistent) + return; + + dprintk(KERN_INFO + "md: updating %s RAID superblock on device (in sync %d)\n", + mdname(mddev),mddev->in_sync); + + err = 0; + ITERATE_RDEV(mddev,rdev,tmp) { + char b[BDEVNAME_SIZE]; + dprintk(KERN_INFO "md: "); + if (rdev->faulty) + dprintk("(skipping faulty "); + + dprintk("%s ", bdevname(rdev->bdev,b)); + if (!rdev->faulty) { + err += write_disk_sb(rdev); + } else + dprintk(")\n"); + if (!err && mddev->level == LEVEL_MULTIPATH) + /* only need to write one superblock... */ + break; + } + if (err) { + if (--count) { + printk(KERN_ERR "md: errors occurred during superblock" + " update, repeating\n"); + goto repeat; + } + printk(KERN_ERR \ + "md: excessive errors occurred during superblock update, exiting\n"); + } +} + +/* + * Import a device. If 'super_format' >= 0, then sanity check the superblock + * + * mark the device faulty if: + * + * - the device is nonexistent (zero size) + * - the device has no valid superblock + * + * a faulty rdev _never_ has rdev->sb set. + */ +static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor) +{ + char b[BDEVNAME_SIZE]; + int err; + mdk_rdev_t *rdev; + sector_t size; + + rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL); + if (!rdev) { + printk(KERN_ERR "md: could not alloc mem for new device!\n"); + return ERR_PTR(-ENOMEM); + } + memset(rdev, 0, sizeof(*rdev)); + + if ((err = alloc_disk_sb(rdev))) + goto abort_free; + + err = lock_rdev(rdev, newdev); + if (err) + goto abort_free; + + rdev->desc_nr = -1; + rdev->faulty = 0; + rdev->in_sync = 0; + rdev->data_offset = 0; + atomic_set(&rdev->nr_pending, 0); + + size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; + if (!size) { + printk(KERN_WARNING + "md: %s has zero or unknown size, marking faulty!\n", + bdevname(rdev->bdev,b)); + err = -EINVAL; + goto abort_free; + } + + if (super_format >= 0) { + err = super_types[super_format]. + load_super(rdev, NULL, super_minor); + if (err == -EINVAL) { + printk(KERN_WARNING + "md: %s has invalid sb, not importing!\n", + bdevname(rdev->bdev,b)); + goto abort_free; + } + if (err < 0) { + printk(KERN_WARNING + "md: could not read %s's sb, not importing!\n", + bdevname(rdev->bdev,b)); + goto abort_free; + } + } + INIT_LIST_HEAD(&rdev->same_set); + + return rdev; + +abort_free: + if (rdev->sb_page) { + if (rdev->bdev) + unlock_rdev(rdev); + free_disk_sb(rdev); + } + kfree(rdev); + return ERR_PTR(err); +} + +/* + * Check a full RAID array for plausibility + */ + + +static int analyze_sbs(mddev_t * mddev) +{ + int i; + struct list_head *tmp; + mdk_rdev_t *rdev, *freshest; + char b[BDEVNAME_SIZE]; + + freshest = NULL; + ITERATE_RDEV(mddev,rdev,tmp) + switch (super_types[mddev->major_version]. + load_super(rdev, freshest, mddev->minor_version)) { + case 1: + freshest = rdev; + break; + case 0: + break; + default: + printk( KERN_ERR \ + "md: fatal superblock inconsistency in %s" + " -- removing from array\n", + bdevname(rdev->bdev,b)); + kick_rdev_from_array(rdev); + } + + + super_types[mddev->major_version]. + validate_super(mddev, freshest); + + i = 0; + ITERATE_RDEV(mddev,rdev,tmp) { + if (rdev != freshest) + if (super_types[mddev->major_version]. + validate_super(mddev, rdev)) { + printk(KERN_WARNING "md: kicking non-fresh %s" + " from array!\n", + bdevname(rdev->bdev,b)); + kick_rdev_from_array(rdev); + continue; + } + if (mddev->level == LEVEL_MULTIPATH) { + rdev->desc_nr = i++; + rdev->raid_disk = rdev->desc_nr; + rdev->in_sync = 1; + } + } + + + + if (mddev->recovery_cp != MaxSector && + mddev->level >= 1) + printk(KERN_ERR "md: %s: raid array is not clean" + " -- starting background reconstruction\n", + mdname(mddev)); + + return 0; +} + +int mdp_major = 0; + +static struct kobject *md_probe(dev_t dev, int *part, void *data) +{ + static DECLARE_MUTEX(disks_sem); + mddev_t *mddev = mddev_find(dev); + struct gendisk *disk; + int partitioned = (MAJOR(dev) != MD_MAJOR); + int shift = partitioned ? MdpMinorShift : 0; + int unit = MINOR(dev) >> shift; + + if (!mddev) + return NULL; + + down(&disks_sem); + if (mddev->gendisk) { + up(&disks_sem); + mddev_put(mddev); + return NULL; + } + disk = alloc_disk(1 << shift); + if (!disk) { + up(&disks_sem); + mddev_put(mddev); + return NULL; + } + disk->major = MAJOR(dev); + disk->first_minor = unit << shift; + if (partitioned) { + sprintf(disk->disk_name, "md_d%d", unit); + sprintf(disk->devfs_name, "md/d%d", unit); + } else { + sprintf(disk->disk_name, "md%d", unit); + sprintf(disk->devfs_name, "md/%d", unit); + } + disk->fops = &md_fops; + disk->private_data = mddev; + disk->queue = mddev->queue; + add_disk(disk); + mddev->gendisk = disk; + up(&disks_sem); + return NULL; +} + +void md_wakeup_thread(mdk_thread_t *thread); + +static void md_safemode_timeout(unsigned long data) +{ + mddev_t *mddev = (mddev_t *) data; + + mddev->safemode = 1; + md_wakeup_thread(mddev->thread); +} + + +static int do_md_run(mddev_t * mddev) +{ + int pnum, err; + int chunk_size; + struct list_head *tmp; + mdk_rdev_t *rdev; + struct gendisk *disk; + char b[BDEVNAME_SIZE]; + + if (list_empty(&mddev->disks)) { + MD_BUG(); + return -EINVAL; + } + + if (mddev->pers) + return -EBUSY; + + /* + * Analyze all RAID superblock(s) + */ + if (!mddev->raid_disks && analyze_sbs(mddev)) { + MD_BUG(); + return -EINVAL; + } + + chunk_size = mddev->chunk_size; + pnum = level_to_pers(mddev->level); + + if ((pnum != MULTIPATH) && (pnum != RAID1)) { + if (!chunk_size) { + /* + * 'default chunksize' in the old md code used to + * be PAGE_SIZE, baaad. + * we abort here to be on the safe side. We don't + * want to continue the bad practice. + */ + printk(KERN_ERR + "no chunksize specified, see 'man raidtab'\n"); + return -EINVAL; + } + if (chunk_size > MAX_CHUNK_SIZE) { + printk(KERN_ERR "too big chunk_size: %d > %d\n", + chunk_size, MAX_CHUNK_SIZE); + return -EINVAL; + } + /* + * chunk-size has to be a power of 2 and multiples of PAGE_SIZE + */ + if ( (1 << ffz(~chunk_size)) != chunk_size) { + MD_BUG(); + return -EINVAL; + } + if (chunk_size < PAGE_SIZE) { + printk(KERN_ERR "too small chunk_size: %d < %ld\n", + chunk_size, PAGE_SIZE); + return -EINVAL; + } + + /* devices must have minimum size of one chunk */ + ITERATE_RDEV(mddev,rdev,tmp) { + if (rdev->faulty) + continue; + if (rdev->size < chunk_size / 1024) { + printk(KERN_WARNING + "md: Dev %s smaller than chunk_size:" + " %lluk < %dk\n", + bdevname(rdev->bdev,b), + (unsigned long long)rdev->size, + chunk_size / 1024); + return -EINVAL; + } + } + } + + if (pnum >= MAX_PERSONALITY) { + MD_BUG(); + return -EINVAL; + } + +#ifdef CONFIG_KMOD + if (!pers[pnum]) + { + request_module("md-personality-%d", pnum); + } +#endif + + /* + * Drop all container device buffers, from now on + * the only valid external interface is through the md + * device. + * Also find largest hardsector size + */ + ITERATE_RDEV(mddev,rdev,tmp) { + if (rdev->faulty) + continue; + sync_blockdev(rdev->bdev); + invalidate_bdev(rdev->bdev, 0); + } + + md_probe(mddev->unit, NULL, NULL); + disk = mddev->gendisk; + if (!disk) + return -ENOMEM; + + spin_lock(&pers_lock); + if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) { + spin_unlock(&pers_lock); + printk(KERN_WARNING "md: personality %d is not loaded!\n", + pnum); + return -EINVAL; + } + + mddev->pers = pers[pnum]; + spin_unlock(&pers_lock); + + mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */ + + err = mddev->pers->run(mddev); + if (err) { + printk(KERN_ERR "md: pers->run() failed ...\n"); + module_put(mddev->pers->owner); + mddev->pers = NULL; + return -EINVAL; + } + atomic_set(&mddev->writes_pending,0); + mddev->safemode = 0; + mddev->safemode_timer.function = md_safemode_timeout; + mddev->safemode_timer.data = (unsigned long) mddev; + mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */ + mddev->in_sync = 1; + + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + + if (mddev->sb_dirty) + md_update_sb(mddev); + + set_capacity(disk, mddev->array_size<<1); + + /* If we call blk_queue_make_request here, it will + * re-initialise max_sectors etc which may have been + * refined inside -> run. So just set the bits we need to set. + * Most initialisation happended when we called + * blk_queue_make_request(..., md_fail_request) + * earlier. + */ + mddev->queue->queuedata = mddev; + mddev->queue->make_request_fn = mddev->pers->make_request; + + mddev->changed = 1; + return 0; +} + +static int restart_array(mddev_t *mddev) +{ + struct gendisk *disk = mddev->gendisk; + int err; + + /* + * Complain if it has no devices + */ + err = -ENXIO; + if (list_empty(&mddev->disks)) + goto out; + + if (mddev->pers) { + err = -EBUSY; + if (!mddev->ro) + goto out; + + mddev->safemode = 0; + mddev->ro = 0; + set_disk_ro(disk, 0); + + printk(KERN_INFO "md: %s switched to read-write mode.\n", + mdname(mddev)); + /* + * Kick recovery or resync if necessary + */ + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + md_wakeup_thread(mddev->thread); + err = 0; + } else { + printk(KERN_ERR "md: %s has no personality assigned.\n", + mdname(mddev)); + err = -EINVAL; + } + +out: + return err; +} + +static int do_md_stop(mddev_t * mddev, int ro) +{ + int err = 0; + struct gendisk *disk = mddev->gendisk; + + if (mddev->pers) { + if (atomic_read(&mddev->active)>2) { + printk("md: %s still in use.\n",mdname(mddev)); + return -EBUSY; + } + + if (mddev->sync_thread) { + set_bit(MD_RECOVERY_INTR, &mddev->recovery); + md_unregister_thread(mddev->sync_thread); + mddev->sync_thread = NULL; + } + + del_timer_sync(&mddev->safemode_timer); + + invalidate_partition(disk, 0); + + if (ro) { + err = -ENXIO; + if (mddev->ro) + goto out; + mddev->ro = 1; + } else { + if (mddev->ro) + set_disk_ro(disk, 0); + blk_queue_make_request(mddev->queue, md_fail_request); + mddev->pers->stop(mddev); + module_put(mddev->pers->owner); + mddev->pers = NULL; + if (mddev->ro) + mddev->ro = 0; + } + if (!mddev->in_sync) { + /* mark array as shutdown cleanly */ + mddev->in_sync = 1; + md_update_sb(mddev); + } + if (ro) + set_disk_ro(disk, 1); + } + /* + * Free resources if final stop + */ + if (!ro) { + struct gendisk *disk; + printk(KERN_INFO "md: %s stopped.\n", mdname(mddev)); + + export_array(mddev); + + mddev->array_size = 0; + disk = mddev->gendisk; + if (disk) + set_capacity(disk, 0); + mddev->changed = 1; + } else + printk(KERN_INFO "md: %s switched to read-only mode.\n", + mdname(mddev)); + err = 0; +out: + return err; +} + +static void autorun_array(mddev_t *mddev) +{ + mdk_rdev_t *rdev; + struct list_head *tmp; + int err; + + if (list_empty(&mddev->disks)) { + MD_BUG(); + return; + } + + printk(KERN_INFO "md: running: "); + + ITERATE_RDEV(mddev,rdev,tmp) { + char b[BDEVNAME_SIZE]; + printk("<%s>", bdevname(rdev->bdev,b)); + } + printk("\n"); + + err = do_md_run (mddev); + if (err) { + printk(KERN_WARNING "md: do_md_run() returned %d\n", err); + do_md_stop (mddev, 0); + } +} + +/* + * lets try to run arrays based on all disks that have arrived + * until now. (those are in pending_raid_disks) + * + * the method: pick the first pending disk, collect all disks with + * the same UUID, remove all from the pending list and put them into + * the 'same_array' list. Then order this list based on superblock + * update time (freshest comes first), kick out 'old' disks and + * compare superblocks. If everything's fine then run it. + * + * If "unit" is allocated, then bump its reference count + */ +static void autorun_devices(int part) +{ + struct list_head candidates; + struct list_head *tmp; + mdk_rdev_t *rdev0, *rdev; + mddev_t *mddev; + char b[BDEVNAME_SIZE]; + + printk(KERN_INFO "md: autorun ...\n"); + while (!list_empty(&pending_raid_disks)) { + dev_t dev; + rdev0 = list_entry(pending_raid_disks.next, + mdk_rdev_t, same_set); + + printk(KERN_INFO "md: considering %s ...\n", + bdevname(rdev0->bdev,b)); + INIT_LIST_HEAD(&candidates); + ITERATE_RDEV_PENDING(rdev,tmp) + if (super_90_load(rdev, rdev0, 0) >= 0) { + printk(KERN_INFO "md: adding %s ...\n", + bdevname(rdev->bdev,b)); + list_move(&rdev->same_set, &candidates); + } + /* + * now we have a set of devices, with all of them having + * mostly sane superblocks. It's time to allocate the + * mddev. + */ + if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) { + printk(KERN_INFO "md: unit number in %s is bad: %d\n", + bdevname(rdev0->bdev, b), rdev0->preferred_minor); + break; + } + if (part) + dev = MKDEV(mdp_major, + rdev0->preferred_minor << MdpMinorShift); + else + dev = MKDEV(MD_MAJOR, rdev0->preferred_minor); + + md_probe(dev, NULL, NULL); + mddev = mddev_find(dev); + if (!mddev) { + printk(KERN_ERR + "md: cannot allocate memory for md drive.\n"); + break; + } + if (mddev_lock(mddev)) + printk(KERN_WARNING "md: %s locked, cannot run\n", + mdname(mddev)); + else if (mddev->raid_disks || mddev->major_version + || !list_empty(&mddev->disks)) { + printk(KERN_WARNING + "md: %s already running, cannot run %s\n", + mdname(mddev), bdevname(rdev0->bdev,b)); + mddev_unlock(mddev); + } else { + printk(KERN_INFO "md: created %s\n", mdname(mddev)); + ITERATE_RDEV_GENERIC(candidates,rdev,tmp) { + list_del_init(&rdev->same_set); + if (bind_rdev_to_array(rdev, mddev)) + export_rdev(rdev); + } + autorun_array(mddev); + mddev_unlock(mddev); + } + /* on success, candidates will be empty, on error + * it won't... + */ + ITERATE_RDEV_GENERIC(candidates,rdev,tmp) + export_rdev(rdev); + mddev_put(mddev); + } + printk(KERN_INFO "md: ... autorun DONE.\n"); +} + +/* + * import RAID devices based on one partition + * if possible, the array gets run as well. + */ + +static int autostart_array(dev_t startdev) +{ + char b[BDEVNAME_SIZE]; + int err = -EINVAL, i; + mdp_super_t *sb = NULL; + mdk_rdev_t *start_rdev = NULL, *rdev; + + start_rdev = md_import_device(startdev, 0, 0); + if (IS_ERR(start_rdev)) + return err; + + + /* NOTE: this can only work for 0.90.0 superblocks */ + sb = (mdp_super_t*)page_address(start_rdev->sb_page); + if (sb->major_version != 0 || + sb->minor_version != 90 ) { + printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n"); + export_rdev(start_rdev); + return err; + } + + if (start_rdev->faulty) { + printk(KERN_WARNING + "md: can not autostart based on faulty %s!\n", + bdevname(start_rdev->bdev,b)); + export_rdev(start_rdev); + return err; + } + list_add(&start_rdev->same_set, &pending_raid_disks); + + for (i = 0; i < MD_SB_DISKS; i++) { + mdp_disk_t *desc = sb->disks + i; + dev_t dev = MKDEV(desc->major, desc->minor); + + if (!dev) + continue; + if (dev == startdev) + continue; + if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor) + continue; + rdev = md_import_device(dev, 0, 0); + if (IS_ERR(rdev)) + continue; + + list_add(&rdev->same_set, &pending_raid_disks); + } + + /* + * possibly return codes + */ + autorun_devices(0); + return 0; + +} + + +static int get_version(void __user * arg) +{ + mdu_version_t ver; + + ver.major = MD_MAJOR_VERSION; + ver.minor = MD_MINOR_VERSION; + ver.patchlevel = MD_PATCHLEVEL_VERSION; + + if (copy_to_user(arg, &ver, sizeof(ver))) + return -EFAULT; + + return 0; +} + +static int get_array_info(mddev_t * mddev, void __user * arg) +{ + mdu_array_info_t info; + int nr,working,active,failed,spare; + mdk_rdev_t *rdev; + struct list_head *tmp; + + nr=working=active=failed=spare=0; + ITERATE_RDEV(mddev,rdev,tmp) { + nr++; + if (rdev->faulty) + failed++; + else { + working++; + if (rdev->in_sync) + active++; + else + spare++; + } + } + + info.major_version = mddev->major_version; + info.minor_version = mddev->minor_version; + info.patch_version = MD_PATCHLEVEL_VERSION; + info.ctime = mddev->ctime; + info.level = mddev->level; + info.size = mddev->size; + info.nr_disks = nr; + info.raid_disks = mddev->raid_disks; + info.md_minor = mddev->md_minor; + info.not_persistent= !mddev->persistent; + + info.utime = mddev->utime; + info.state = 0; + if (mddev->in_sync) + info.state = (1<<MD_SB_CLEAN); + info.active_disks = active; + info.working_disks = working; + info.failed_disks = failed; + info.spare_disks = spare; + + info.layout = mddev->layout; + info.chunk_size = mddev->chunk_size; + + if (copy_to_user(arg, &info, sizeof(info))) + return -EFAULT; + + return 0; +} + +static int get_disk_info(mddev_t * mddev, void __user * arg) +{ + mdu_disk_info_t info; + unsigned int nr; + mdk_rdev_t *rdev; + + if (copy_from_user(&info, arg, sizeof(info))) + return -EFAULT; + + nr = info.number; + + rdev = find_rdev_nr(mddev, nr); + if (rdev) { + info.major = MAJOR(rdev->bdev->bd_dev); + info.minor = MINOR(rdev->bdev->bd_dev); + info.raid_disk = rdev->raid_disk; + info.state = 0; + if (rdev->faulty) + info.state |= (1<<MD_DISK_FAULTY); + else if (rdev->in_sync) { + info.state |= (1<<MD_DISK_ACTIVE); + info.state |= (1<<MD_DISK_SYNC); + } + } else { + info.major = info.minor = 0; + info.raid_disk = -1; + info.state = (1<<MD_DISK_REMOVED); + } + + if (copy_to_user(arg, &info, sizeof(info))) + return -EFAULT; + + return 0; +} + +static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info) +{ + char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE]; + mdk_rdev_t *rdev; + dev_t dev = MKDEV(info->major,info->minor); + + if (info->major != MAJOR(dev) || info->minor != MINOR(dev)) + return -EOVERFLOW; + + if (!mddev->raid_disks) { + int err; + /* expecting a device which has a superblock */ + rdev = md_import_device(dev, mddev->major_version, mddev->minor_version); + if (IS_ERR(rdev)) { + printk(KERN_WARNING + "md: md_import_device returned %ld\n", + PTR_ERR(rdev)); + return PTR_ERR(rdev); + } + if (!list_empty(&mddev->disks)) { + mdk_rdev_t *rdev0 = list_entry(mddev->disks.next, + mdk_rdev_t, same_set); + int err = super_types[mddev->major_version] + .load_super(rdev, rdev0, mddev->minor_version); + if (err < 0) { + printk(KERN_WARNING + "md: %s has different UUID to %s\n", + bdevname(rdev->bdev,b), + bdevname(rdev0->bdev,b2)); + export_rdev(rdev); + return -EINVAL; + } + } + err = bind_rdev_to_array(rdev, mddev); + if (err) + export_rdev(rdev); + return err; + } + + /* + * add_new_disk can be used once the array is assembled + * to add "hot spares". They must already have a superblock + * written + */ + if (mddev->pers) { + int err; + if (!mddev->pers->hot_add_disk) { + printk(KERN_WARNING + "%s: personality does not support diskops!\n", + mdname(mddev)); + return -EINVAL; + } + rdev = md_import_device(dev, mddev->major_version, + mddev->minor_version); + if (IS_ERR(rdev)) { + printk(KERN_WARNING + "md: md_import_device returned %ld\n", + PTR_ERR(rdev)); + return PTR_ERR(rdev); + } + rdev->in_sync = 0; /* just to be sure */ + rdev->raid_disk = -1; + err = bind_rdev_to_array(rdev, mddev); + if (err) + export_rdev(rdev); + if (mddev->thread) + md_wakeup_thread(mddev->thread); + return err; + } + + /* otherwise, add_new_disk is only allowed + * for major_version==0 superblocks + */ + if (mddev->major_version != 0) { + printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n", + mdname(mddev)); + return -EINVAL; + } + + if (!(info->state & (1<<MD_DISK_FAULTY))) { + int err; + rdev = md_import_device (dev, -1, 0); + if (IS_ERR(rdev)) { + printk(KERN_WARNING + "md: error, md_import_device() returned %ld\n", + PTR_ERR(rdev)); + return PTR_ERR(rdev); + } + rdev->desc_nr = info->number; + if (info->raid_disk < mddev->raid_disks) + rdev->raid_disk = info->raid_disk; + else + rdev->raid_disk = -1; + + rdev->faulty = 0; + if (rdev->raid_disk < mddev->raid_disks) + rdev->in_sync = (info->state & (1<<MD_DISK_SYNC)); + else + rdev->in_sync = 0; + + err = bind_rdev_to_array(rdev, mddev); + if (err) { + export_rdev(rdev); + return err; + } + + if (!mddev->persistent) { + printk(KERN_INFO "md: nonpersistent superblock ...\n"); + rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; + } else + rdev->sb_offset = calc_dev_sboffset(rdev->bdev); + rdev->size = calc_dev_size(rdev, mddev->chunk_size); + + if (!mddev->size || (mddev->size > rdev->size)) + mddev->size = rdev->size; + } + + return 0; +} + +static int hot_remove_disk(mddev_t * mddev, dev_t dev) +{ + char b[BDEVNAME_SIZE]; + mdk_rdev_t *rdev; + + if (!mddev->pers) + return -ENODEV; + + rdev = find_rdev(mddev, dev); + if (!rdev) + return -ENXIO; + + if (rdev->raid_disk >= 0) + goto busy; + + kick_rdev_from_array(rdev); + md_update_sb(mddev); + + return 0; +busy: + printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n", + bdevname(rdev->bdev,b), mdname(mddev)); + return -EBUSY; +} + +static int hot_add_disk(mddev_t * mddev, dev_t dev) +{ + char b[BDEVNAME_SIZE]; + int err; + unsigned int size; + mdk_rdev_t *rdev; + + if (!mddev->pers) + return -ENODEV; + + if (mddev->major_version != 0) { + printk(KERN_WARNING "%s: HOT_ADD may only be used with" + " version-0 superblocks.\n", + mdname(mddev)); + return -EINVAL; + } + if (!mddev->pers->hot_add_disk) { + printk(KERN_WARNING + "%s: personality does not support diskops!\n", + mdname(mddev)); + return -EINVAL; + } + + rdev = md_import_device (dev, -1, 0); + if (IS_ERR(rdev)) { + printk(KERN_WARNING + "md: error, md_import_device() returned %ld\n", + PTR_ERR(rdev)); + return -EINVAL; + } + + if (mddev->persistent) + rdev->sb_offset = calc_dev_sboffset(rdev->bdev); + else + rdev->sb_offset = + rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS; + + size = calc_dev_size(rdev, mddev->chunk_size); + rdev->size = size; + + if (size < mddev->size) { + printk(KERN_WARNING + "%s: disk size %llu blocks < array size %llu\n", + mdname(mddev), (unsigned long long)size, + (unsigned long long)mddev->size); + err = -ENOSPC; + goto abort_export; + } + + if (rdev->faulty) { + printk(KERN_WARNING + "md: can not hot-add faulty %s disk to %s!\n", + bdevname(rdev->bdev,b), mdname(mddev)); + err = -EINVAL; + goto abort_export; + } + rdev->in_sync = 0; + rdev->desc_nr = -1; + bind_rdev_to_array(rdev, mddev); + + /* + * The rest should better be atomic, we can have disk failures + * noticed in interrupt contexts ... + */ + + if (rdev->desc_nr == mddev->max_disks) { + printk(KERN_WARNING "%s: can not hot-add to full array!\n", + mdname(mddev)); + err = -EBUSY; + goto abort_unbind_export; + } + + rdev->raid_disk = -1; + + md_update_sb(mddev); + + /* + * Kick recovery, maybe this spare has to be added to the + * array immediately. + */ + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + md_wakeup_thread(mddev->thread); + + return 0; + +abort_unbind_export: + unbind_rdev_from_array(rdev); + +abort_export: + export_rdev(rdev); + return err; +} + +/* + * set_array_info is used two different ways + * The original usage is when creating a new array. + * In this usage, raid_disks is > 0 and it together with + * level, size, not_persistent,layout,chunksize determine the + * shape of the array. + * This will always create an array with a type-0.90.0 superblock. + * The newer usage is when assembling an array. + * In this case raid_disks will be 0, and the major_version field is + * use to determine which style super-blocks are to be found on the devices. + * The minor and patch _version numbers are also kept incase the + * super_block handler wishes to interpret them. + */ +static int set_array_info(mddev_t * mddev, mdu_array_info_t *info) +{ + + if (info->raid_disks == 0) { + /* just setting version number for superblock loading */ + if (info->major_version < 0 || + info->major_version >= sizeof(super_types)/sizeof(super_types[0]) || + super_types[info->major_version].name == NULL) { + /* maybe try to auto-load a module? */ + printk(KERN_INFO + "md: superblock version %d not known\n", + info->major_version); + return -EINVAL; + } + mddev->major_version = info->major_version; + mddev->minor_version = info->minor_version; + mddev->patch_version = info->patch_version; + return 0; + } + mddev->major_version = MD_MAJOR_VERSION; + mddev->minor_version = MD_MINOR_VERSION; + mddev->patch_version = MD_PATCHLEVEL_VERSION; + mddev->ctime = get_seconds(); + + mddev->level = info->level; + mddev->size = info->size; + mddev->raid_disks = info->raid_disks; + /* don't set md_minor, it is determined by which /dev/md* was + * openned + */ + if (info->state & (1<<MD_SB_CLEAN)) + mddev->recovery_cp = MaxSector; + else + mddev->recovery_cp = 0; + mddev->persistent = ! info->not_persistent; + + mddev->layout = info->layout; + mddev->chunk_size = info->chunk_size; + + mddev->max_disks = MD_SB_DISKS; + + mddev->sb_dirty = 1; + + /* + * Generate a 128 bit UUID + */ + get_random_bytes(mddev->uuid, 16); + + return 0; +} + +/* + * update_array_info is used to change the configuration of an + * on-line array. + * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size + * fields in the info are checked against the array. + * Any differences that cannot be handled will cause an error. + * Normally, only one change can be managed at a time. + */ +static int update_array_info(mddev_t *mddev, mdu_array_info_t *info) +{ + int rv = 0; + int cnt = 0; + + if (mddev->major_version != info->major_version || + mddev->minor_version != info->minor_version || +/* mddev->patch_version != info->patch_version || */ + mddev->ctime != info->ctime || + mddev->level != info->level || +/* mddev->layout != info->layout || */ + !mddev->persistent != info->not_persistent|| + mddev->chunk_size != info->chunk_size ) + return -EINVAL; + /* Check there is only one change */ + if (mddev->size != info->size) cnt++; + if (mddev->raid_disks != info->raid_disks) cnt++; + if (mddev->layout != info->layout) cnt++; + if (cnt == 0) return 0; + if (cnt > 1) return -EINVAL; + + if (mddev->layout != info->layout) { + /* Change layout + * we don't need to do anything at the md level, the + * personality will take care of it all. + */ + if (mddev->pers->reconfig == NULL) + return -EINVAL; + else + return mddev->pers->reconfig(mddev, info->layout, -1); + } + if (mddev->size != info->size) { + mdk_rdev_t * rdev; + struct list_head *tmp; + if (mddev->pers->resize == NULL) + return -EINVAL; + /* The "size" is the amount of each device that is used. + * This can only make sense for arrays with redundancy. + * linear and raid0 always use whatever space is available + * We can only consider changing the size if no resync + * or reconstruction is happening, and if the new size + * is acceptable. It must fit before the sb_offset or, + * if that is <data_offset, it must fit before the + * size of each device. + * If size is zero, we find the largest size that fits. + */ + if (mddev->sync_thread) + return -EBUSY; + ITERATE_RDEV(mddev,rdev,tmp) { + sector_t avail; + int fit = (info->size == 0); + if (rdev->sb_offset > rdev->data_offset) + avail = (rdev->sb_offset*2) - rdev->data_offset; + else + avail = get_capacity(rdev->bdev->bd_disk) + - rdev->data_offset; + if (fit && (info->size == 0 || info->size > avail/2)) + info->size = avail/2; + if (avail < ((sector_t)info->size << 1)) + return -ENOSPC; + } + rv = mddev->pers->resize(mddev, (sector_t)info->size *2); + if (!rv) { + struct block_device *bdev; + + bdev = bdget_disk(mddev->gendisk, 0); + if (bdev) { + down(&bdev->bd_inode->i_sem); + i_size_write(bdev->bd_inode, mddev->array_size << 10); + up(&bdev->bd_inode->i_sem); + bdput(bdev); + } + } + } + if (mddev->raid_disks != info->raid_disks) { + /* change the number of raid disks */ + if (mddev->pers->reshape == NULL) + return -EINVAL; + if (info->raid_disks <= 0 || + info->raid_disks >= mddev->max_disks) + return -EINVAL; + if (mddev->sync_thread) + return -EBUSY; + rv = mddev->pers->reshape(mddev, info->raid_disks); + if (!rv) { + struct block_device *bdev; + + bdev = bdget_disk(mddev->gendisk, 0); + if (bdev) { + down(&bdev->bd_inode->i_sem); + i_size_write(bdev->bd_inode, mddev->array_size << 10); + up(&bdev->bd_inode->i_sem); + bdput(bdev); + } + } + } + md_update_sb(mddev); + return rv; +} + +static int set_disk_faulty(mddev_t *mddev, dev_t dev) +{ + mdk_rdev_t *rdev; + + if (mddev->pers == NULL) + return -ENODEV; + + rdev = find_rdev(mddev, dev); + if (!rdev) + return -ENODEV; + + md_error(mddev, rdev); + return 0; +} + +static int md_ioctl(struct inode *inode, struct file *file, + unsigned int cmd, unsigned long arg) +{ + int err = 0; + void __user *argp = (void __user *)arg; + struct hd_geometry __user *loc = argp; + mddev_t *mddev = NULL; + + if (!capable(CAP_SYS_ADMIN)) + return -EACCES; + + /* + * Commands dealing with the RAID driver but not any + * particular array: + */ + switch (cmd) + { + case RAID_VERSION: + err = get_version(argp); + goto done; + + case PRINT_RAID_DEBUG: + err = 0; + md_print_devices(); + goto done; + +#ifndef MODULE + case RAID_AUTORUN: + err = 0; + autostart_arrays(arg); + goto done; +#endif + default:; + } + + /* + * Commands creating/starting a new array: + */ + + mddev = inode->i_bdev->bd_disk->private_data; + + if (!mddev) { + BUG(); + goto abort; + } + + + if (cmd == START_ARRAY) { + /* START_ARRAY doesn't need to lock the array as autostart_array + * does the locking, and it could even be a different array + */ + static int cnt = 3; + if (cnt > 0 ) { + printk(KERN_WARNING + "md: %s(pid %d) used deprecated START_ARRAY ioctl. " + "This will not be supported beyond 2.6\n", + current->comm, current->pid); + cnt--; + } + err = autostart_array(new_decode_dev(arg)); + if (err) { + printk(KERN_WARNING "md: autostart failed!\n"); + goto abort; + } + goto done; + } + + err = mddev_lock(mddev); + if (err) { + printk(KERN_INFO + "md: ioctl lock interrupted, reason %d, cmd %d\n", + err, cmd); + goto abort; + } + + switch (cmd) + { + case SET_ARRAY_INFO: + { + mdu_array_info_t info; + if (!arg) + memset(&info, 0, sizeof(info)); + else if (copy_from_user(&info, argp, sizeof(info))) { + err = -EFAULT; + goto abort_unlock; + } + if (mddev->pers) { + err = update_array_info(mddev, &info); + if (err) { + printk(KERN_WARNING "md: couldn't update" + " array info. %d\n", err); + goto abort_unlock; + } + goto done_unlock; + } + if (!list_empty(&mddev->disks)) { + printk(KERN_WARNING + "md: array %s already has disks!\n", + mdname(mddev)); + err = -EBUSY; + goto abort_unlock; + } + if (mddev->raid_disks) { + printk(KERN_WARNING + "md: array %s already initialised!\n", + mdname(mddev)); + err = -EBUSY; + goto abort_unlock; + } + err = set_array_info(mddev, &info); + if (err) { + printk(KERN_WARNING "md: couldn't set" + " array info. %d\n", err); + goto abort_unlock; + } + } + goto done_unlock; + + default:; + } + + /* + * Commands querying/configuring an existing array: + */ + /* if we are initialised yet, only ADD_NEW_DISK or STOP_ARRAY is allowed */ + if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY && cmd != RUN_ARRAY) { + err = -ENODEV; + goto abort_unlock; + } + + /* + * Commands even a read-only array can execute: + */ + switch (cmd) + { + case GET_ARRAY_INFO: + err = get_array_info(mddev, argp); + goto done_unlock; + + case GET_DISK_INFO: + err = get_disk_info(mddev, argp); + goto done_unlock; + + case RESTART_ARRAY_RW: + err = restart_array(mddev); + goto done_unlock; + + case STOP_ARRAY: + err = do_md_stop (mddev, 0); + goto done_unlock; + + case STOP_ARRAY_RO: + err = do_md_stop (mddev, 1); + goto done_unlock; + + /* + * We have a problem here : there is no easy way to give a CHS + * virtual geometry. We currently pretend that we have a 2 heads + * 4 sectors (with a BIG number of cylinders...). This drives + * dosfs just mad... ;-) + */ + case HDIO_GETGEO: + if (!loc) { + err = -EINVAL; + goto abort_unlock; + } + err = put_user (2, (char __user *) &loc->heads); + if (err) + goto abort_unlock; + err = put_user (4, (char __user *) &loc->sectors); + if (err) + goto abort_unlock; + err = put_user(get_capacity(mddev->gendisk)/8, + (short __user *) &loc->cylinders); + if (err) + goto abort_unlock; + err = put_user (get_start_sect(inode->i_bdev), + (long __user *) &loc->start); + goto done_unlock; + } + + /* + * The remaining ioctls are changing the state of the + * superblock, so we do not allow read-only arrays + * here: + */ + if (mddev->ro) { + err = -EROFS; + goto abort_unlock; + } + + switch (cmd) + { + case ADD_NEW_DISK: + { + mdu_disk_info_t info; + if (copy_from_user(&info, argp, sizeof(info))) + err = -EFAULT; + else + err = add_new_disk(mddev, &info); + goto done_unlock; + } + + case HOT_REMOVE_DISK: + err = hot_remove_disk(mddev, new_decode_dev(arg)); + goto done_unlock; + + case HOT_ADD_DISK: + err = hot_add_disk(mddev, new_decode_dev(arg)); + goto done_unlock; + + case SET_DISK_FAULTY: + err = set_disk_faulty(mddev, new_decode_dev(arg)); + goto done_unlock; + + case RUN_ARRAY: + err = do_md_run (mddev); + goto done_unlock; + + default: + if (_IOC_TYPE(cmd) == MD_MAJOR) + printk(KERN_WARNING "md: %s(pid %d) used" + " obsolete MD ioctl, upgrade your" + " software to use new ictls.\n", + current->comm, current->pid); + err = -EINVAL; + goto abort_unlock; + } + +done_unlock: +abort_unlock: + mddev_unlock(mddev); + + return err; +done: + if (err) + MD_BUG(); +abort: + return err; +} + +static int md_open(struct inode *inode, struct file *file) +{ + /* + * Succeed if we can lock the mddev, which confirms that + * it isn't being stopped right now. + */ + mddev_t *mddev = inode->i_bdev->bd_disk->private_data; + int err; + + if ((err = mddev_lock(mddev))) + goto out; + + err = 0; + mddev_get(mddev); + mddev_unlock(mddev); + + check_disk_change(inode->i_bdev); + out: + return err; +} + +static int md_release(struct inode *inode, struct file * file) +{ + mddev_t *mddev = inode->i_bdev->bd_disk->private_data; + + if (!mddev) + BUG(); + mddev_put(mddev); + + return 0; +} + +static int md_media_changed(struct gendisk *disk) +{ + mddev_t *mddev = disk->private_data; + + return mddev->changed; +} + +static int md_revalidate(struct gendisk *disk) +{ + mddev_t *mddev = disk->private_data; + + mddev->changed = 0; + return 0; +} +static struct block_device_operations md_fops = +{ + .owner = THIS_MODULE, + .open = md_open, + .release = md_release, + .ioctl = md_ioctl, + .media_changed = md_media_changed, + .revalidate_disk= md_revalidate, +}; + +int md_thread(void * arg) +{ + mdk_thread_t *thread = arg; + + lock_kernel(); + + /* + * Detach thread + */ + + daemonize(thread->name, mdname(thread->mddev)); + + current->exit_signal = SIGCHLD; + allow_signal(SIGKILL); + thread->tsk = current; + + /* + * md_thread is a 'system-thread', it's priority should be very + * high. We avoid resource deadlocks individually in each + * raid personality. (RAID5 does preallocation) We also use RR and + * the very same RT priority as kswapd, thus we will never get + * into a priority inversion deadlock. + * + * we definitely have to have equal or higher priority than + * bdflush, otherwise bdflush will deadlock if there are too + * many dirty RAID5 blocks. + */ + unlock_kernel(); + + complete(thread->event); + while (thread->run) { + void (*run)(mddev_t *); + + wait_event_interruptible(thread->wqueue, + test_bit(THREAD_WAKEUP, &thread->flags)); + if (current->flags & PF_FREEZE) + refrigerator(PF_FREEZE); + + clear_bit(THREAD_WAKEUP, &thread->flags); + + run = thread->run; + if (run) + run(thread->mddev); + + if (signal_pending(current)) + flush_signals(current); + } + complete(thread->event); + return 0; +} + +void md_wakeup_thread(mdk_thread_t *thread) +{ + if (thread) { + dprintk("md: waking up MD thread %s.\n", thread->tsk->comm); + set_bit(THREAD_WAKEUP, &thread->flags); + wake_up(&thread->wqueue); + } +} + +mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev, + const char *name) +{ + mdk_thread_t *thread; + int ret; + struct completion event; + + thread = (mdk_thread_t *) kmalloc + (sizeof(mdk_thread_t), GFP_KERNEL); + if (!thread) + return NULL; + + memset(thread, 0, sizeof(mdk_thread_t)); + init_waitqueue_head(&thread->wqueue); + + init_completion(&event); + thread->event = &event; + thread->run = run; + thread->mddev = mddev; + thread->name = name; + ret = kernel_thread(md_thread, thread, 0); + if (ret < 0) { + kfree(thread); + return NULL; + } + wait_for_completion(&event); + return thread; +} + +static void md_interrupt_thread(mdk_thread_t *thread) +{ + if (!thread->tsk) { + MD_BUG(); + return; + } + dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid); + send_sig(SIGKILL, thread->tsk, 1); +} + +void md_unregister_thread(mdk_thread_t *thread) +{ + struct completion event; + + init_completion(&event); + + thread->event = &event; + thread->run = NULL; + thread->name = NULL; + md_interrupt_thread(thread); + wait_for_completion(&event); + kfree(thread); +} + +void md_error(mddev_t *mddev, mdk_rdev_t *rdev) +{ + if (!mddev) { + MD_BUG(); + return; + } + + if (!rdev || rdev->faulty) + return; + + dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n", + mdname(mddev), + MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev), + __builtin_return_address(0),__builtin_return_address(1), + __builtin_return_address(2),__builtin_return_address(3)); + + if (!mddev->pers->error_handler) + return; + mddev->pers->error_handler(mddev,rdev); + set_bit(MD_RECOVERY_INTR, &mddev->recovery); + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + md_wakeup_thread(mddev->thread); +} + +/* seq_file implementation /proc/mdstat */ + +static void status_unused(struct seq_file *seq) +{ + int i = 0; + mdk_rdev_t *rdev; + struct list_head *tmp; + + seq_printf(seq, "unused devices: "); + + ITERATE_RDEV_PENDING(rdev,tmp) { + char b[BDEVNAME_SIZE]; + i++; + seq_printf(seq, "%s ", + bdevname(rdev->bdev,b)); + } + if (!i) + seq_printf(seq, "<none>"); + + seq_printf(seq, "\n"); +} + + +static void status_resync(struct seq_file *seq, mddev_t * mddev) +{ + unsigned long max_blocks, resync, res, dt, db, rt; + + resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2; + + if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) + max_blocks = mddev->resync_max_sectors >> 1; + else + max_blocks = mddev->size; + + /* + * Should not happen. + */ + if (!max_blocks) { + MD_BUG(); + return; + } + res = (resync/1024)*1000/(max_blocks/1024 + 1); + { + int i, x = res/50, y = 20-x; + seq_printf(seq, "["); + for (i = 0; i < x; i++) + seq_printf(seq, "="); + seq_printf(seq, ">"); + for (i = 0; i < y; i++) + seq_printf(seq, "."); + seq_printf(seq, "] "); + } + seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)", + (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? + "resync" : "recovery"), + res/10, res % 10, resync, max_blocks); + + /* + * We do not want to overflow, so the order of operands and + * the * 100 / 100 trick are important. We do a +1 to be + * safe against division by zero. We only estimate anyway. + * + * dt: time from mark until now + * db: blocks written from mark until now + * rt: remaining time + */ + dt = ((jiffies - mddev->resync_mark) / HZ); + if (!dt) dt++; + db = resync - (mddev->resync_mark_cnt/2); + rt = (dt * ((max_blocks-resync) / (db/100+1)))/100; + + seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6); + + seq_printf(seq, " speed=%ldK/sec", db/dt); +} + +static void *md_seq_start(struct seq_file *seq, loff_t *pos) +{ + struct list_head *tmp; + loff_t l = *pos; + mddev_t *mddev; + + if (l >= 0x10000) + return NULL; + if (!l--) + /* header */ + return (void*)1; + + spin_lock(&all_mddevs_lock); + list_for_each(tmp,&all_mddevs) + if (!l--) { + mddev = list_entry(tmp, mddev_t, all_mddevs); + mddev_get(mddev); + spin_unlock(&all_mddevs_lock); + return mddev; + } + spin_unlock(&all_mddevs_lock); + if (!l--) + return (void*)2;/* tail */ + return NULL; +} + +static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos) +{ + struct list_head *tmp; + mddev_t *next_mddev, *mddev = v; + + ++*pos; + if (v == (void*)2) + return NULL; + + spin_lock(&all_mddevs_lock); + if (v == (void*)1) + tmp = all_mddevs.next; + else + tmp = mddev->all_mddevs.next; + if (tmp != &all_mddevs) + next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs)); + else { + next_mddev = (void*)2; + *pos = 0x10000; + } + spin_unlock(&all_mddevs_lock); + + if (v != (void*)1) + mddev_put(mddev); + return next_mddev; + +} + +static void md_seq_stop(struct seq_file *seq, void *v) +{ + mddev_t *mddev = v; + + if (mddev && v != (void*)1 && v != (void*)2) + mddev_put(mddev); +} + +static int md_seq_show(struct seq_file *seq, void *v) +{ + mddev_t *mddev = v; + sector_t size; + struct list_head *tmp2; + mdk_rdev_t *rdev; + int i; + + if (v == (void*)1) { + seq_printf(seq, "Personalities : "); + spin_lock(&pers_lock); + for (i = 0; i < MAX_PERSONALITY; i++) + if (pers[i]) + seq_printf(seq, "[%s] ", pers[i]->name); + + spin_unlock(&pers_lock); + seq_printf(seq, "\n"); + return 0; + } + if (v == (void*)2) { + status_unused(seq); + return 0; + } + + if (mddev_lock(mddev)!=0) + return -EINTR; + if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) { + seq_printf(seq, "%s : %sactive", mdname(mddev), + mddev->pers ? "" : "in"); + if (mddev->pers) { + if (mddev->ro) + seq_printf(seq, " (read-only)"); + seq_printf(seq, " %s", mddev->pers->name); + } + + size = 0; + ITERATE_RDEV(mddev,rdev,tmp2) { + char b[BDEVNAME_SIZE]; + seq_printf(seq, " %s[%d]", + bdevname(rdev->bdev,b), rdev->desc_nr); + if (rdev->faulty) { + seq_printf(seq, "(F)"); + continue; + } + size += rdev->size; + } + + if (!list_empty(&mddev->disks)) { + if (mddev->pers) + seq_printf(seq, "\n %llu blocks", + (unsigned long long)mddev->array_size); + else + seq_printf(seq, "\n %llu blocks", + (unsigned long long)size); + } + + if (mddev->pers) { + mddev->pers->status (seq, mddev); + seq_printf(seq, "\n "); + if (mddev->curr_resync > 2) + status_resync (seq, mddev); + else if (mddev->curr_resync == 1 || mddev->curr_resync == 2) + seq_printf(seq, " resync=DELAYED"); + } + + seq_printf(seq, "\n"); + } + mddev_unlock(mddev); + + return 0; +} + +static struct seq_operations md_seq_ops = { + .start = md_seq_start, + .next = md_seq_next, + .stop = md_seq_stop, + .show = md_seq_show, +}; + +static int md_seq_open(struct inode *inode, struct file *file) +{ + int error; + + error = seq_open(file, &md_seq_ops); + return error; +} + +static struct file_operations md_seq_fops = { + .open = md_seq_open, + .read = seq_read, + .llseek = seq_lseek, + .release = seq_release, +}; + +int register_md_personality(int pnum, mdk_personality_t *p) +{ + if (pnum >= MAX_PERSONALITY) { + printk(KERN_ERR + "md: tried to install personality %s as nr %d, but max is %lu\n", + p->name, pnum, MAX_PERSONALITY-1); + return -EINVAL; + } + + spin_lock(&pers_lock); + if (pers[pnum]) { + spin_unlock(&pers_lock); + MD_BUG(); + return -EBUSY; + } + + pers[pnum] = p; + printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum); + spin_unlock(&pers_lock); + return 0; +} + +int unregister_md_personality(int pnum) +{ + if (pnum >= MAX_PERSONALITY) { + MD_BUG(); + return -EINVAL; + } + + printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name); + spin_lock(&pers_lock); + pers[pnum] = NULL; + spin_unlock(&pers_lock); + return 0; +} + +static int is_mddev_idle(mddev_t *mddev) +{ + mdk_rdev_t * rdev; + struct list_head *tmp; + int idle; + unsigned long curr_events; + + idle = 1; + ITERATE_RDEV(mddev,rdev,tmp) { + struct gendisk *disk = rdev->bdev->bd_contains->bd_disk; + curr_events = disk_stat_read(disk, read_sectors) + + disk_stat_read(disk, write_sectors) - + atomic_read(&disk->sync_io); + /* Allow some slack between valud of curr_events and last_events, + * as there are some uninteresting races. + * Note: the following is an unsigned comparison. + */ + if ((curr_events - rdev->last_events + 32) > 64) { + rdev->last_events = curr_events; + idle = 0; + } + } + return idle; +} + +void md_done_sync(mddev_t *mddev, int blocks, int ok) +{ + /* another "blocks" (512byte) blocks have been synced */ + atomic_sub(blocks, &mddev->recovery_active); + wake_up(&mddev->recovery_wait); + if (!ok) { + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + md_wakeup_thread(mddev->thread); + // stop recovery, signal do_sync .... + } +} + + +void md_write_start(mddev_t *mddev) +{ + if (!atomic_read(&mddev->writes_pending)) { + mddev_lock_uninterruptible(mddev); + if (mddev->in_sync) { + mddev->in_sync = 0; + del_timer(&mddev->safemode_timer); + md_update_sb(mddev); + } + atomic_inc(&mddev->writes_pending); + mddev_unlock(mddev); + } else + atomic_inc(&mddev->writes_pending); +} + +void md_write_end(mddev_t *mddev) +{ + if (atomic_dec_and_test(&mddev->writes_pending)) { + if (mddev->safemode == 2) + md_wakeup_thread(mddev->thread); + else + mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay); + } +} + +static inline void md_enter_safemode(mddev_t *mddev) +{ + if (!mddev->safemode) return; + if (mddev->safemode == 2 && + (atomic_read(&mddev->writes_pending) || mddev->in_sync || + mddev->recovery_cp != MaxSector)) + return; /* avoid the lock */ + mddev_lock_uninterruptible(mddev); + if (mddev->safemode && !atomic_read(&mddev->writes_pending) && + !mddev->in_sync && mddev->recovery_cp == MaxSector) { + mddev->in_sync = 1; + md_update_sb(mddev); + } + mddev_unlock(mddev); + + if (mddev->safemode == 1) + mddev->safemode = 0; +} + +void md_handle_safemode(mddev_t *mddev) +{ + if (signal_pending(current)) { + printk(KERN_INFO "md: %s in immediate safe mode\n", + mdname(mddev)); + mddev->safemode = 2; + flush_signals(current); + } + md_enter_safemode(mddev); +} + + +DECLARE_WAIT_QUEUE_HEAD(resync_wait); + +#define SYNC_MARKS 10 +#define SYNC_MARK_STEP (3*HZ) +static void md_do_sync(mddev_t *mddev) +{ + mddev_t *mddev2; + unsigned int currspeed = 0, + window; + sector_t max_sectors,j; + unsigned long mark[SYNC_MARKS]; + sector_t mark_cnt[SYNC_MARKS]; + int last_mark,m; + struct list_head *tmp; + sector_t last_check; + + /* just incase thread restarts... */ + if (test_bit(MD_RECOVERY_DONE, &mddev->recovery)) + return; + + /* we overload curr_resync somewhat here. + * 0 == not engaged in resync at all + * 2 == checking that there is no conflict with another sync + * 1 == like 2, but have yielded to allow conflicting resync to + * commense + * other == active in resync - this many blocks + * + * Before starting a resync we must have set curr_resync to + * 2, and then checked that every "conflicting" array has curr_resync + * less than ours. When we find one that is the same or higher + * we wait on resync_wait. To avoid deadlock, we reduce curr_resync + * to 1 if we choose to yield (based arbitrarily on address of mddev structure). + * This will mean we have to start checking from the beginning again. + * + */ + + do { + mddev->curr_resync = 2; + + try_again: + if (signal_pending(current)) { + flush_signals(current); + goto skip; + } + ITERATE_MDDEV(mddev2,tmp) { + printk("."); + if (mddev2 == mddev) + continue; + if (mddev2->curr_resync && + match_mddev_units(mddev,mddev2)) { + DEFINE_WAIT(wq); + if (mddev < mddev2 && mddev->curr_resync == 2) { + /* arbitrarily yield */ + mddev->curr_resync = 1; + wake_up(&resync_wait); + } + if (mddev > mddev2 && mddev->curr_resync == 1) + /* no need to wait here, we can wait the next + * time 'round when curr_resync == 2 + */ + continue; + prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE); + if (!signal_pending(current) + && mddev2->curr_resync >= mddev->curr_resync) { + printk(KERN_INFO "md: delaying resync of %s" + " until %s has finished resync (they" + " share one or more physical units)\n", + mdname(mddev), mdname(mddev2)); + mddev_put(mddev2); + schedule(); + finish_wait(&resync_wait, &wq); + goto try_again; + } + finish_wait(&resync_wait, &wq); + } + } + } while (mddev->curr_resync < 2); + + if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) + /* resync follows the size requested by the personality, + * which default to physical size, but can be virtual size + */ + max_sectors = mddev->resync_max_sectors; + else + /* recovery follows the physical size of devices */ + max_sectors = mddev->size << 1; + + printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev)); + printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:" + " %d KB/sec/disc.\n", sysctl_speed_limit_min); + printk(KERN_INFO "md: using maximum available idle IO bandwith " + "(but not more than %d KB/sec) for reconstruction.\n", + sysctl_speed_limit_max); + + is_mddev_idle(mddev); /* this also initializes IO event counters */ + if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) + j = mddev->recovery_cp; + else + j = 0; + for (m = 0; m < SYNC_MARKS; m++) { + mark[m] = jiffies; + mark_cnt[m] = j; + } + last_mark = 0; + mddev->resync_mark = mark[last_mark]; + mddev->resync_mark_cnt = mark_cnt[last_mark]; + + /* + * Tune reconstruction: + */ + window = 32*(PAGE_SIZE/512); + printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n", + window/2,(unsigned long long) max_sectors/2); + + atomic_set(&mddev->recovery_active, 0); + init_waitqueue_head(&mddev->recovery_wait); + last_check = 0; + + if (j>2) { + printk(KERN_INFO + "md: resuming recovery of %s from checkpoint.\n", + mdname(mddev)); + mddev->curr_resync = j; + } + + while (j < max_sectors) { + int sectors; + + sectors = mddev->pers->sync_request(mddev, j, currspeed < sysctl_speed_limit_min); + if (sectors < 0) { + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + goto out; + } + atomic_add(sectors, &mddev->recovery_active); + j += sectors; + if (j>1) mddev->curr_resync = j; + + if (last_check + window > j || j == max_sectors) + continue; + + last_check = j; + + if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) || + test_bit(MD_RECOVERY_ERR, &mddev->recovery)) + break; + + repeat: + if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) { + /* step marks */ + int next = (last_mark+1) % SYNC_MARKS; + + mddev->resync_mark = mark[next]; + mddev->resync_mark_cnt = mark_cnt[next]; + mark[next] = jiffies; + mark_cnt[next] = j - atomic_read(&mddev->recovery_active); + last_mark = next; + } + + + if (signal_pending(current)) { + /* + * got a signal, exit. + */ + printk(KERN_INFO + "md: md_do_sync() got signal ... exiting\n"); + flush_signals(current); + set_bit(MD_RECOVERY_INTR, &mddev->recovery); + goto out; + } + + /* + * this loop exits only if either when we are slower than + * the 'hard' speed limit, or the system was IO-idle for + * a jiffy. + * the system might be non-idle CPU-wise, but we only care + * about not overloading the IO subsystem. (things like an + * e2fsck being done on the RAID array should execute fast) + */ + mddev->queue->unplug_fn(mddev->queue); + cond_resched(); + + currspeed = ((unsigned long)(j-mddev->resync_mark_cnt))/2/((jiffies-mddev->resync_mark)/HZ +1) +1; + + if (currspeed > sysctl_speed_limit_min) { + if ((currspeed > sysctl_speed_limit_max) || + !is_mddev_idle(mddev)) { + msleep_interruptible(250); + goto repeat; + } + } + } + printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev)); + /* + * this also signals 'finished resyncing' to md_stop + */ + out: + mddev->queue->unplug_fn(mddev->queue); + + wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active)); + + /* tell personality that we are finished */ + mddev->pers->sync_request(mddev, max_sectors, 1); + + if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && + mddev->curr_resync > 2 && + mddev->curr_resync >= mddev->recovery_cp) { + if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { + printk(KERN_INFO + "md: checkpointing recovery of %s.\n", + mdname(mddev)); + mddev->recovery_cp = mddev->curr_resync; + } else + mddev->recovery_cp = MaxSector; + } + + md_enter_safemode(mddev); + skip: + mddev->curr_resync = 0; + wake_up(&resync_wait); + set_bit(MD_RECOVERY_DONE, &mddev->recovery); + md_wakeup_thread(mddev->thread); +} + + +/* + * This routine is regularly called by all per-raid-array threads to + * deal with generic issues like resync and super-block update. + * Raid personalities that don't have a thread (linear/raid0) do not + * need this as they never do any recovery or update the superblock. + * + * It does not do any resync itself, but rather "forks" off other threads + * to do that as needed. + * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in + * "->recovery" and create a thread at ->sync_thread. + * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR) + * and wakeups up this thread which will reap the thread and finish up. + * This thread also removes any faulty devices (with nr_pending == 0). + * + * The overall approach is: + * 1/ if the superblock needs updating, update it. + * 2/ If a recovery thread is running, don't do anything else. + * 3/ If recovery has finished, clean up, possibly marking spares active. + * 4/ If there are any faulty devices, remove them. + * 5/ If array is degraded, try to add spares devices + * 6/ If array has spares or is not in-sync, start a resync thread. + */ +void md_check_recovery(mddev_t *mddev) +{ + mdk_rdev_t *rdev; + struct list_head *rtmp; + + + dprintk(KERN_INFO "md: recovery thread got woken up ...\n"); + + if (mddev->ro) + return; + if ( ! ( + mddev->sb_dirty || + test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) || + test_bit(MD_RECOVERY_DONE, &mddev->recovery) + )) + return; + if (mddev_trylock(mddev)==0) { + int spares =0; + if (mddev->sb_dirty) + md_update_sb(mddev); + if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) && + !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) { + /* resync/recovery still happening */ + clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + goto unlock; + } + if (mddev->sync_thread) { + /* resync has finished, collect result */ + md_unregister_thread(mddev->sync_thread); + mddev->sync_thread = NULL; + if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) && + !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { + /* success...*/ + /* activate any spares */ + mddev->pers->spare_active(mddev); + } + md_update_sb(mddev); + mddev->recovery = 0; + /* flag recovery needed just to double check */ + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + goto unlock; + } + if (mddev->recovery) + /* probably just the RECOVERY_NEEDED flag */ + mddev->recovery = 0; + + /* no recovery is running. + * remove any failed drives, then + * add spares if possible. + * Spare are also removed and re-added, to allow + * the personality to fail the re-add. + */ + ITERATE_RDEV(mddev,rdev,rtmp) + if (rdev->raid_disk >= 0 && + (rdev->faulty || ! rdev->in_sync) && + atomic_read(&rdev->nr_pending)==0) { + if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) + rdev->raid_disk = -1; + } + + if (mddev->degraded) { + ITERATE_RDEV(mddev,rdev,rtmp) + if (rdev->raid_disk < 0 + && !rdev->faulty) { + if (mddev->pers->hot_add_disk(mddev,rdev)) + spares++; + else + break; + } + } + + if (!spares && (mddev->recovery_cp == MaxSector )) { + /* nothing we can do ... */ + goto unlock; + } + if (mddev->pers->sync_request) { + set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); + if (!spares) + set_bit(MD_RECOVERY_SYNC, &mddev->recovery); + mddev->sync_thread = md_register_thread(md_do_sync, + mddev, + "%s_resync"); + if (!mddev->sync_thread) { + printk(KERN_ERR "%s: could not start resync" + " thread...\n", + mdname(mddev)); + /* leave the spares where they are, it shouldn't hurt */ + mddev->recovery = 0; + } else { + md_wakeup_thread(mddev->sync_thread); + } + } + unlock: + mddev_unlock(mddev); + } +} + +int md_notify_reboot(struct notifier_block *this, + unsigned long code, void *x) +{ + struct list_head *tmp; + mddev_t *mddev; + + if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) { + + printk(KERN_INFO "md: stopping all md devices.\n"); + + ITERATE_MDDEV(mddev,tmp) + if (mddev_trylock(mddev)==0) + do_md_stop (mddev, 1); + /* + * certain more exotic SCSI devices are known to be + * volatile wrt too early system reboots. While the + * right place to handle this issue is the given + * driver, we do want to have a safe RAID driver ... + */ + mdelay(1000*1); + } + return NOTIFY_DONE; +} + +struct notifier_block md_notifier = { + .notifier_call = md_notify_reboot, + .next = NULL, + .priority = INT_MAX, /* before any real devices */ +}; + +static void md_geninit(void) +{ + struct proc_dir_entry *p; + + dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t)); + + p = create_proc_entry("mdstat", S_IRUGO, NULL); + if (p) + p->proc_fops = &md_seq_fops; +} + +int __init md_init(void) +{ + int minor; + + printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d," + " MD_SB_DISKS=%d\n", + MD_MAJOR_VERSION, MD_MINOR_VERSION, + MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS); + + if (register_blkdev(MAJOR_NR, "md")) + return -1; + if ((mdp_major=register_blkdev(0, "mdp"))<=0) { + unregister_blkdev(MAJOR_NR, "md"); + return -1; + } + devfs_mk_dir("md"); + blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE, + md_probe, NULL, NULL); + blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE, + md_probe, NULL, NULL); + + for (minor=0; minor < MAX_MD_DEVS; ++minor) + devfs_mk_bdev(MKDEV(MAJOR_NR, minor), + S_IFBLK|S_IRUSR|S_IWUSR, + "md/%d", minor); + + for (minor=0; minor < MAX_MD_DEVS; ++minor) + devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift), + S_IFBLK|S_IRUSR|S_IWUSR, + "md/mdp%d", minor); + + + register_reboot_notifier(&md_notifier); + raid_table_header = register_sysctl_table(raid_root_table, 1); + + md_geninit(); + return (0); +} + + +#ifndef MODULE + +/* + * Searches all registered partitions for autorun RAID arrays + * at boot time. + */ +static dev_t detected_devices[128]; +static int dev_cnt; + +void md_autodetect_dev(dev_t dev) +{ + if (dev_cnt >= 0 && dev_cnt < 127) + detected_devices[dev_cnt++] = dev; +} + + +static void autostart_arrays(int part) +{ + mdk_rdev_t *rdev; + int i; + + printk(KERN_INFO "md: Autodetecting RAID arrays.\n"); + + for (i = 0; i < dev_cnt; i++) { + dev_t dev = detected_devices[i]; + + rdev = md_import_device(dev,0, 0); + if (IS_ERR(rdev)) + continue; + + if (rdev->faulty) { + MD_BUG(); + continue; + } + list_add(&rdev->same_set, &pending_raid_disks); + } + dev_cnt = 0; + + autorun_devices(part); +} + +#endif + +static __exit void md_exit(void) +{ + mddev_t *mddev; + struct list_head *tmp; + int i; + blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS); + blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift); + for (i=0; i < MAX_MD_DEVS; i++) + devfs_remove("md/%d", i); + for (i=0; i < MAX_MD_DEVS; i++) + devfs_remove("md/d%d", i); + + devfs_remove("md"); + + unregister_blkdev(MAJOR_NR,"md"); + unregister_blkdev(mdp_major, "mdp"); + unregister_reboot_notifier(&md_notifier); + unregister_sysctl_table(raid_table_header); + remove_proc_entry("mdstat", NULL); + ITERATE_MDDEV(mddev,tmp) { + struct gendisk *disk = mddev->gendisk; + if (!disk) + continue; + export_array(mddev); + del_gendisk(disk); + put_disk(disk); + mddev->gendisk = NULL; + mddev_put(mddev); + } +} + +module_init(md_init) +module_exit(md_exit) + +EXPORT_SYMBOL(register_md_personality); +EXPORT_SYMBOL(unregister_md_personality); +EXPORT_SYMBOL(md_error); +EXPORT_SYMBOL(md_done_sync); +EXPORT_SYMBOL(md_write_start); +EXPORT_SYMBOL(md_write_end); +EXPORT_SYMBOL(md_handle_safemode); +EXPORT_SYMBOL(md_register_thread); +EXPORT_SYMBOL(md_unregister_thread); +EXPORT_SYMBOL(md_wakeup_thread); +EXPORT_SYMBOL(md_print_devices); +EXPORT_SYMBOL(md_check_recovery); +MODULE_LICENSE("GPL"); diff --git a/drivers/md/mktables.c b/drivers/md/mktables.c new file mode 100644 index 00000000000..adef299908c --- /dev/null +++ b/drivers/md/mktables.c @@ -0,0 +1,125 @@ +#ident "$Id: mktables.c,v 1.2 2002/12/12 22:41:27 hpa Exp $" +/* ----------------------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * mktables.c + * + * Make RAID-6 tables. This is a host user space program to be run at + * compile time. + */ + +#include <stdio.h> +#include <string.h> +#include <inttypes.h> +#include <stdlib.h> +#include <time.h> + +static uint8_t gfmul(uint8_t a, uint8_t b) +{ + uint8_t v = 0; + + while ( b ) { + if ( b & 1 ) v ^= a; + a = (a << 1) ^ (a & 0x80 ? 0x1d : 0); + b >>= 1; + } + return v; +} + +static uint8_t gfpow(uint8_t a, int b) +{ + uint8_t v = 1; + + b %= 255; + if ( b < 0 ) + b += 255; + + while ( b ) { + if ( b & 1 ) v = gfmul(v,a); + a = gfmul(a,a); + b >>= 1; + } + return v; +} + +int main(int argc, char *argv[]) +{ + int i, j, k; + uint8_t v; + uint8_t exptbl[256], invtbl[256]; + + printf("#include \"raid6.h\"\n"); + + /* Compute multiplication table */ + printf("\nconst u8 __attribute__((aligned(256)))\n" + "raid6_gfmul[256][256] =\n" + "{\n"); + for ( i = 0 ; i < 256 ; i++ ) { + printf("\t{\n"); + for ( j = 0 ; j < 256 ; j += 8 ) { + printf("\t\t"); + for ( k = 0 ; k < 8 ; k++ ) { + printf("0x%02x, ", gfmul(i,j+k)); + } + printf("\n"); + } + printf("\t},\n"); + } + printf("};\n"); + + /* Compute power-of-2 table (exponent) */ + v = 1; + printf("\nconst u8 __attribute__((aligned(256)))\n" + "raid6_gfexp[256] =\n" + "{\n"); + for ( i = 0 ; i < 256 ; i += 8 ) { + printf("\t"); + for ( j = 0 ; j < 8 ; j++ ) { + exptbl[i+j] = v; + printf("0x%02x, ", v); + v = gfmul(v,2); + if ( v == 1 ) v = 0; /* For entry 255, not a real entry */ + } + printf("\n"); + } + printf("};\n"); + + /* Compute inverse table x^-1 == x^254 */ + printf("\nconst u8 __attribute__((aligned(256)))\n" + "raid6_gfinv[256] =\n" + "{\n"); + for ( i = 0 ; i < 256 ; i += 8 ) { + printf("\t"); + for ( j = 0 ; j < 8 ; j++ ) { + invtbl[i+j] = v = gfpow(i+j,254); + printf("0x%02x, ", v); + } + printf("\n"); + } + printf("};\n"); + + /* Compute inv(2^x + 1) (exponent-xor-inverse) table */ + printf("\nconst u8 __attribute__((aligned(256)))\n" + "raid6_gfexi[256] =\n" + "{\n"); + for ( i = 0 ; i < 256 ; i += 8 ) { + printf("\t"); + for ( j = 0 ; j < 8 ; j++ ) { + printf("0x%02x, ", invtbl[exptbl[i+j]^1]); + } + printf("\n"); + } + printf("};\n\n"); + + return 0; +} diff --git a/drivers/md/multipath.c b/drivers/md/multipath.c new file mode 100644 index 00000000000..c9b134cd153 --- /dev/null +++ b/drivers/md/multipath.c @@ -0,0 +1,584 @@ +/* + * multipath.c : Multiple Devices driver for Linux + * + * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat + * + * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman + * + * MULTIPATH management functions. + * + * derived from raid1.c. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/spinlock.h> +#include <linux/raid/multipath.h> +#include <linux/buffer_head.h> +#include <asm/atomic.h> + +#define MAJOR_NR MD_MAJOR +#define MD_DRIVER +#define MD_PERSONALITY + +#define MAX_WORK_PER_DISK 128 + +#define NR_RESERVED_BUFS 32 + + +static mdk_personality_t multipath_personality; + + +static void *mp_pool_alloc(unsigned int __nocast gfp_flags, void *data) +{ + struct multipath_bh *mpb; + mpb = kmalloc(sizeof(*mpb), gfp_flags); + if (mpb) + memset(mpb, 0, sizeof(*mpb)); + return mpb; +} + +static void mp_pool_free(void *mpb, void *data) +{ + kfree(mpb); +} + +static int multipath_map (multipath_conf_t *conf) +{ + int i, disks = conf->raid_disks; + + /* + * Later we do read balancing on the read side + * now we use the first available disk. + */ + + rcu_read_lock(); + for (i = 0; i < disks; i++) { + mdk_rdev_t *rdev = conf->multipaths[i].rdev; + if (rdev && rdev->in_sync) { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + return i; + } + } + rcu_read_unlock(); + + printk(KERN_ERR "multipath_map(): no more operational IO paths?\n"); + return (-1); +} + +static void multipath_reschedule_retry (struct multipath_bh *mp_bh) +{ + unsigned long flags; + mddev_t *mddev = mp_bh->mddev; + multipath_conf_t *conf = mddev_to_conf(mddev); + + spin_lock_irqsave(&conf->device_lock, flags); + list_add(&mp_bh->retry_list, &conf->retry_list); + spin_unlock_irqrestore(&conf->device_lock, flags); + md_wakeup_thread(mddev->thread); +} + + +/* + * multipath_end_bh_io() is called when we have finished servicing a multipathed + * operation and are ready to return a success/failure code to the buffer + * cache layer. + */ +static void multipath_end_bh_io (struct multipath_bh *mp_bh, int err) +{ + struct bio *bio = mp_bh->master_bio; + multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev); + + bio_endio(bio, bio->bi_size, err); + mempool_free(mp_bh, conf->pool); +} + +int multipath_end_request(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + struct multipath_bh * mp_bh = (struct multipath_bh *)(bio->bi_private); + multipath_conf_t *conf = mddev_to_conf(mp_bh->mddev); + mdk_rdev_t *rdev = conf->multipaths[mp_bh->path].rdev; + + if (bio->bi_size) + return 1; + + if (uptodate) + multipath_end_bh_io(mp_bh, 0); + else if (!bio_rw_ahead(bio)) { + /* + * oops, IO error: + */ + char b[BDEVNAME_SIZE]; + md_error (mp_bh->mddev, rdev); + printk(KERN_ERR "multipath: %s: rescheduling sector %llu\n", + bdevname(rdev->bdev,b), + (unsigned long long)bio->bi_sector); + multipath_reschedule_retry(mp_bh); + } else + multipath_end_bh_io(mp_bh, error); + rdev_dec_pending(rdev, conf->mddev); + return 0; +} + +static void unplug_slaves(mddev_t *mddev) +{ + multipath_conf_t *conf = mddev_to_conf(mddev); + int i; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks; i++) { + mdk_rdev_t *rdev = conf->multipaths[i].rdev; + if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) { + request_queue_t *r_queue = bdev_get_queue(rdev->bdev); + + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + rcu_read_unlock(); +} + +static void multipath_unplug(request_queue_t *q) +{ + unplug_slaves(q->queuedata); +} + + +static int multipath_make_request (request_queue_t *q, struct bio * bio) +{ + mddev_t *mddev = q->queuedata; + multipath_conf_t *conf = mddev_to_conf(mddev); + struct multipath_bh * mp_bh; + struct multipath_info *multipath; + + mp_bh = mempool_alloc(conf->pool, GFP_NOIO); + + mp_bh->master_bio = bio; + mp_bh->mddev = mddev; + + if (bio_data_dir(bio)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); + } + + mp_bh->path = multipath_map(conf); + if (mp_bh->path < 0) { + bio_endio(bio, bio->bi_size, -EIO); + mempool_free(mp_bh, conf->pool); + return 0; + } + multipath = conf->multipaths + mp_bh->path; + + mp_bh->bio = *bio; + mp_bh->bio.bi_sector += multipath->rdev->data_offset; + mp_bh->bio.bi_bdev = multipath->rdev->bdev; + mp_bh->bio.bi_rw |= (1 << BIO_RW_FAILFAST); + mp_bh->bio.bi_end_io = multipath_end_request; + mp_bh->bio.bi_private = mp_bh; + generic_make_request(&mp_bh->bio); + return 0; +} + +static void multipath_status (struct seq_file *seq, mddev_t *mddev) +{ + multipath_conf_t *conf = mddev_to_conf(mddev); + int i; + + seq_printf (seq, " [%d/%d] [", conf->raid_disks, + conf->working_disks); + for (i = 0; i < conf->raid_disks; i++) + seq_printf (seq, "%s", + conf->multipaths[i].rdev && + conf->multipaths[i].rdev->in_sync ? "U" : "_"); + seq_printf (seq, "]"); +} + +static int multipath_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + multipath_conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + mdk_rdev_t *rdev = conf->multipaths[i].rdev; + if (rdev && !rdev->faulty) { + struct block_device *bdev = rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, + error_sector); + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + } + rcu_read_unlock(); + return ret; +} + +/* + * Careful, this can execute in IRQ contexts as well! + */ +static void multipath_error (mddev_t *mddev, mdk_rdev_t *rdev) +{ + multipath_conf_t *conf = mddev_to_conf(mddev); + + if (conf->working_disks <= 1) { + /* + * Uh oh, we can do nothing if this is our last path, but + * first check if this is a queued request for a device + * which has just failed. + */ + printk(KERN_ALERT + "multipath: only one IO path left and IO error.\n"); + /* leave it active... it's all we have */ + } else { + /* + * Mark disk as unusable + */ + if (!rdev->faulty) { + char b[BDEVNAME_SIZE]; + rdev->in_sync = 0; + rdev->faulty = 1; + mddev->sb_dirty = 1; + conf->working_disks--; + printk(KERN_ALERT "multipath: IO failure on %s," + " disabling IO path. \n Operation continuing" + " on %d IO paths.\n", + bdevname (rdev->bdev,b), + conf->working_disks); + } + } +} + +static void print_multipath_conf (multipath_conf_t *conf) +{ + int i; + struct multipath_info *tmp; + + printk("MULTIPATH conf printout:\n"); + if (!conf) { + printk("(conf==NULL)\n"); + return; + } + printk(" --- wd:%d rd:%d\n", conf->working_disks, + conf->raid_disks); + + for (i = 0; i < conf->raid_disks; i++) { + char b[BDEVNAME_SIZE]; + tmp = conf->multipaths + i; + if (tmp->rdev) + printk(" disk%d, o:%d, dev:%s\n", + i,!tmp->rdev->faulty, + bdevname(tmp->rdev->bdev,b)); + } +} + + +static int multipath_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) +{ + multipath_conf_t *conf = mddev->private; + int found = 0; + int path; + struct multipath_info *p; + + print_multipath_conf(conf); + + for (path=0; path<mddev->raid_disks; path++) + if ((p=conf->multipaths+path)->rdev == NULL) { + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + * (Note: it is very unlikely that a device with + * merge_bvec_fn will be involved in multipath.) + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + conf->working_disks++; + rdev->raid_disk = path; + rdev->in_sync = 1; + p->rdev = rdev; + found = 1; + } + + print_multipath_conf(conf); + return found; +} + +static int multipath_remove_disk(mddev_t *mddev, int number) +{ + multipath_conf_t *conf = mddev->private; + int err = 0; + mdk_rdev_t *rdev; + struct multipath_info *p = conf->multipaths + number; + + print_multipath_conf(conf); + + rdev = p->rdev; + if (rdev) { + if (rdev->in_sync || + atomic_read(&rdev->nr_pending)) { + printk(KERN_ERR "hot-remove-disk, slot %d is identified" " but is still operational!\n", number); + err = -EBUSY; + goto abort; + } + p->rdev = NULL; + synchronize_kernel(); + if (atomic_read(&rdev->nr_pending)) { + /* lost the race, try later */ + err = -EBUSY; + p->rdev = rdev; + } + } +abort: + + print_multipath_conf(conf); + return err; +} + + + +/* + * This is a kernel thread which: + * + * 1. Retries failed read operations on working multipaths. + * 2. Updates the raid superblock when problems encounter. + * 3. Performs writes following reads for array syncronising. + */ + +static void multipathd (mddev_t *mddev) +{ + struct multipath_bh *mp_bh; + struct bio *bio; + unsigned long flags; + multipath_conf_t *conf = mddev_to_conf(mddev); + struct list_head *head = &conf->retry_list; + + md_check_recovery(mddev); + for (;;) { + char b[BDEVNAME_SIZE]; + spin_lock_irqsave(&conf->device_lock, flags); + if (list_empty(head)) + break; + mp_bh = list_entry(head->prev, struct multipath_bh, retry_list); + list_del(head->prev); + spin_unlock_irqrestore(&conf->device_lock, flags); + + bio = &mp_bh->bio; + bio->bi_sector = mp_bh->master_bio->bi_sector; + + if ((mp_bh->path = multipath_map (conf))<0) { + printk(KERN_ALERT "multipath: %s: unrecoverable IO read" + " error for block %llu\n", + bdevname(bio->bi_bdev,b), + (unsigned long long)bio->bi_sector); + multipath_end_bh_io(mp_bh, -EIO); + } else { + printk(KERN_ERR "multipath: %s: redirecting sector %llu" + " to another IO path\n", + bdevname(bio->bi_bdev,b), + (unsigned long long)bio->bi_sector); + *bio = *(mp_bh->master_bio); + bio->bi_sector += conf->multipaths[mp_bh->path].rdev->data_offset; + bio->bi_bdev = conf->multipaths[mp_bh->path].rdev->bdev; + bio->bi_rw |= (1 << BIO_RW_FAILFAST); + bio->bi_end_io = multipath_end_request; + bio->bi_private = mp_bh; + generic_make_request(bio); + } + } + spin_unlock_irqrestore(&conf->device_lock, flags); +} + +static int multipath_run (mddev_t *mddev) +{ + multipath_conf_t *conf; + int disk_idx; + struct multipath_info *disk; + mdk_rdev_t *rdev; + struct list_head *tmp; + + if (mddev->level != LEVEL_MULTIPATH) { + printk("multipath: %s: raid level not set to multipath IO (%d)\n", + mdname(mddev), mddev->level); + goto out; + } + /* + * copy the already verified devices into our private MULTIPATH + * bookkeeping area. [whatever we allocate in multipath_run(), + * should be freed in multipath_stop()] + */ + + conf = kmalloc(sizeof(multipath_conf_t), GFP_KERNEL); + mddev->private = conf; + if (!conf) { + printk(KERN_ERR + "multipath: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out; + } + memset(conf, 0, sizeof(*conf)); + + conf->multipaths = kmalloc(sizeof(struct multipath_info)*mddev->raid_disks, + GFP_KERNEL); + if (!conf->multipaths) { + printk(KERN_ERR + "multipath: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out_free_conf; + } + memset(conf->multipaths, 0, sizeof(struct multipath_info)*mddev->raid_disks); + + mddev->queue->unplug_fn = multipath_unplug; + + mddev->queue->issue_flush_fn = multipath_issue_flush; + + conf->working_disks = 0; + ITERATE_RDEV(mddev,rdev,tmp) { + disk_idx = rdev->raid_disk; + if (disk_idx < 0 || + disk_idx >= mddev->raid_disks) + continue; + + disk = conf->multipaths + disk_idx; + disk->rdev = rdev; + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, not that we ever expect a device with + * a merge_bvec_fn to be involved in multipath */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + if (!rdev->faulty) + conf->working_disks++; + } + + conf->raid_disks = mddev->raid_disks; + mddev->sb_dirty = 1; + conf->mddev = mddev; + spin_lock_init(&conf->device_lock); + INIT_LIST_HEAD(&conf->retry_list); + + if (!conf->working_disks) { + printk(KERN_ERR "multipath: no operational IO paths for %s\n", + mdname(mddev)); + goto out_free_conf; + } + mddev->degraded = conf->raid_disks = conf->working_disks; + + conf->pool = mempool_create(NR_RESERVED_BUFS, + mp_pool_alloc, mp_pool_free, + NULL); + if (conf->pool == NULL) { + printk(KERN_ERR + "multipath: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out_free_conf; + } + + { + mddev->thread = md_register_thread(multipathd, mddev, "%s_multipath"); + if (!mddev->thread) { + printk(KERN_ERR "multipath: couldn't allocate thread" + " for %s\n", mdname(mddev)); + goto out_free_conf; + } + } + + printk(KERN_INFO + "multipath: array %s active with %d out of %d IO paths\n", + mdname(mddev), conf->working_disks, mddev->raid_disks); + /* + * Ok, everything is just fine now + */ + mddev->array_size = mddev->size; + return 0; + +out_free_conf: + if (conf->pool) + mempool_destroy(conf->pool); + if (conf->multipaths) + kfree(conf->multipaths); + kfree(conf); + mddev->private = NULL; +out: + return -EIO; +} + + +static int multipath_stop (mddev_t *mddev) +{ + multipath_conf_t *conf = mddev_to_conf(mddev); + + md_unregister_thread(mddev->thread); + mddev->thread = NULL; + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + mempool_destroy(conf->pool); + kfree(conf->multipaths); + kfree(conf); + mddev->private = NULL; + return 0; +} + +static mdk_personality_t multipath_personality= +{ + .name = "multipath", + .owner = THIS_MODULE, + .make_request = multipath_make_request, + .run = multipath_run, + .stop = multipath_stop, + .status = multipath_status, + .error_handler = multipath_error, + .hot_add_disk = multipath_add_disk, + .hot_remove_disk= multipath_remove_disk, +}; + +static int __init multipath_init (void) +{ + return register_md_personality (MULTIPATH, &multipath_personality); +} + +static void __exit multipath_exit (void) +{ + unregister_md_personality (MULTIPATH); +} + +module_init(multipath_init); +module_exit(multipath_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-7"); /* MULTIPATH */ diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c new file mode 100644 index 00000000000..e7d934eca06 --- /dev/null +++ b/drivers/md/raid0.c @@ -0,0 +1,539 @@ +/* + raid0.c : Multiple Devices driver for Linux + Copyright (C) 1994-96 Marc ZYNGIER + <zyngier@ufr-info-p7.ibp.fr> or + <maz@gloups.fdn.fr> + Copyright (C) 1999, 2000 Ingo Molnar, Red Hat + + + RAID-0 management functions. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + You should have received a copy of the GNU General Public License + (for example /usr/src/linux/COPYING); if not, write to the Free + Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. +*/ + +#include <linux/module.h> +#include <linux/raid/raid0.h> + +#define MAJOR_NR MD_MAJOR +#define MD_DRIVER +#define MD_PERSONALITY + +static void raid0_unplug(request_queue_t *q) +{ + mddev_t *mddev = q->queuedata; + raid0_conf_t *conf = mddev_to_conf(mddev); + mdk_rdev_t **devlist = conf->strip_zone[0].dev; + int i; + + for (i=0; i<mddev->raid_disks; i++) { + request_queue_t *r_queue = bdev_get_queue(devlist[i]->bdev); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + } +} + +static int raid0_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + raid0_conf_t *conf = mddev_to_conf(mddev); + mdk_rdev_t **devlist = conf->strip_zone[0].dev; + int i, ret = 0; + + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + struct block_device *bdev = devlist[i]->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector); + } + return ret; +} + + +static int create_strip_zones (mddev_t *mddev) +{ + int i, c, j; + sector_t current_offset, curr_zone_offset; + sector_t min_spacing; + raid0_conf_t *conf = mddev_to_conf(mddev); + mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev; + struct list_head *tmp1, *tmp2; + struct strip_zone *zone; + int cnt; + char b[BDEVNAME_SIZE]; + + /* + * The number of 'same size groups' + */ + conf->nr_strip_zones = 0; + + ITERATE_RDEV(mddev,rdev1,tmp1) { + printk("raid0: looking at %s\n", + bdevname(rdev1->bdev,b)); + c = 0; + ITERATE_RDEV(mddev,rdev2,tmp2) { + printk("raid0: comparing %s(%llu)", + bdevname(rdev1->bdev,b), + (unsigned long long)rdev1->size); + printk(" with %s(%llu)\n", + bdevname(rdev2->bdev,b), + (unsigned long long)rdev2->size); + if (rdev2 == rdev1) { + printk("raid0: END\n"); + break; + } + if (rdev2->size == rdev1->size) + { + /* + * Not unique, don't count it as a new + * group + */ + printk("raid0: EQUAL\n"); + c = 1; + break; + } + printk("raid0: NOT EQUAL\n"); + } + if (!c) { + printk("raid0: ==> UNIQUE\n"); + conf->nr_strip_zones++; + printk("raid0: %d zones\n", conf->nr_strip_zones); + } + } + printk("raid0: FINAL %d zones\n", conf->nr_strip_zones); + + conf->strip_zone = kmalloc(sizeof(struct strip_zone)* + conf->nr_strip_zones, GFP_KERNEL); + if (!conf->strip_zone) + return 1; + conf->devlist = kmalloc(sizeof(mdk_rdev_t*)* + conf->nr_strip_zones*mddev->raid_disks, + GFP_KERNEL); + if (!conf->devlist) + return 1; + + memset(conf->strip_zone, 0,sizeof(struct strip_zone)* + conf->nr_strip_zones); + memset(conf->devlist, 0, + sizeof(mdk_rdev_t*) * conf->nr_strip_zones * mddev->raid_disks); + + /* The first zone must contain all devices, so here we check that + * there is a proper alignment of slots to devices and find them all + */ + zone = &conf->strip_zone[0]; + cnt = 0; + smallest = NULL; + zone->dev = conf->devlist; + ITERATE_RDEV(mddev, rdev1, tmp1) { + int j = rdev1->raid_disk; + + if (j < 0 || j >= mddev->raid_disks) { + printk("raid0: bad disk number %d - aborting!\n", j); + goto abort; + } + if (zone->dev[j]) { + printk("raid0: multiple devices for %d - aborting!\n", + j); + goto abort; + } + zone->dev[j] = rdev1; + + blk_queue_stack_limits(mddev->queue, + rdev1->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + + if (rdev1->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + if (!smallest || (rdev1->size <smallest->size)) + smallest = rdev1; + cnt++; + } + if (cnt != mddev->raid_disks) { + printk("raid0: too few disks (%d of %d) - aborting!\n", + cnt, mddev->raid_disks); + goto abort; + } + zone->nb_dev = cnt; + zone->size = smallest->size * cnt; + zone->zone_offset = 0; + + current_offset = smallest->size; + curr_zone_offset = zone->size; + + /* now do the other zones */ + for (i = 1; i < conf->nr_strip_zones; i++) + { + zone = conf->strip_zone + i; + zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks; + + printk("raid0: zone %d\n", i); + zone->dev_offset = current_offset; + smallest = NULL; + c = 0; + + for (j=0; j<cnt; j++) { + char b[BDEVNAME_SIZE]; + rdev = conf->strip_zone[0].dev[j]; + printk("raid0: checking %s ...", bdevname(rdev->bdev,b)); + if (rdev->size > current_offset) + { + printk(" contained as device %d\n", c); + zone->dev[c] = rdev; + c++; + if (!smallest || (rdev->size <smallest->size)) { + smallest = rdev; + printk(" (%llu) is smallest!.\n", + (unsigned long long)rdev->size); + } + } else + printk(" nope.\n"); + } + + zone->nb_dev = c; + zone->size = (smallest->size - current_offset) * c; + printk("raid0: zone->nb_dev: %d, size: %llu\n", + zone->nb_dev, (unsigned long long)zone->size); + + zone->zone_offset = curr_zone_offset; + curr_zone_offset += zone->size; + + current_offset = smallest->size; + printk("raid0: current zone offset: %llu\n", + (unsigned long long)current_offset); + } + + /* Now find appropriate hash spacing. + * We want a number which causes most hash entries to cover + * at most two strips, but the hash table must be at most + * 1 PAGE. We choose the smallest strip, or contiguous collection + * of strips, that has big enough size. We never consider the last + * strip though as it's size has no bearing on the efficacy of the hash + * table. + */ + conf->hash_spacing = curr_zone_offset; + min_spacing = curr_zone_offset; + sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*)); + for (i=0; i < conf->nr_strip_zones-1; i++) { + sector_t sz = 0; + for (j=i; j<conf->nr_strip_zones-1 && + sz < min_spacing ; j++) + sz += conf->strip_zone[j].size; + if (sz >= min_spacing && sz < conf->hash_spacing) + conf->hash_spacing = sz; + } + + mddev->queue->unplug_fn = raid0_unplug; + + mddev->queue->issue_flush_fn = raid0_issue_flush; + + printk("raid0: done.\n"); + return 0; + abort: + return 1; +} + +/** + * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged + * @q: request queue + * @bio: the buffer head that's been built up so far + * @biovec: the request that could be merged to it. + * + * Return amount of bytes we can accept at this offset + */ +static int raid0_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec) +{ + mddev_t *mddev = q->queuedata; + sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); + int max; + unsigned int chunk_sectors = mddev->chunk_size >> 9; + unsigned int bio_sectors = bio->bi_size >> 9; + + max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; + if (max < 0) max = 0; /* bio_add cannot handle a negative return */ + if (max <= biovec->bv_len && bio_sectors == 0) + return biovec->bv_len; + else + return max; +} + +static int raid0_run (mddev_t *mddev) +{ + unsigned cur=0, i=0, nb_zone; + s64 size; + raid0_conf_t *conf; + mdk_rdev_t *rdev; + struct list_head *tmp; + + printk("%s: setting max_sectors to %d, segment boundary to %d\n", + mdname(mddev), + mddev->chunk_size >> 9, + (mddev->chunk_size>>1)-1); + blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9); + blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1); + + conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL); + if (!conf) + goto out; + mddev->private = (void *)conf; + + conf->strip_zone = NULL; + conf->devlist = NULL; + if (create_strip_zones (mddev)) + goto out_free_conf; + + /* calculate array device size */ + mddev->array_size = 0; + ITERATE_RDEV(mddev,rdev,tmp) + mddev->array_size += rdev->size; + + printk("raid0 : md_size is %llu blocks.\n", + (unsigned long long)mddev->array_size); + printk("raid0 : conf->hash_spacing is %llu blocks.\n", + (unsigned long long)conf->hash_spacing); + { +#if __GNUC__ < 3 + volatile +#endif + sector_t s = mddev->array_size; + sector_t space = conf->hash_spacing; + int round; + conf->preshift = 0; + if (sizeof(sector_t) > sizeof(unsigned long)) { + /*shift down space and s so that sector_div will work */ + while (space > (sector_t) (~(unsigned long)0)) { + s >>= 1; + space >>= 1; + s += 1; /* force round-up */ + conf->preshift++; + } + } + round = sector_div(s, (unsigned long)space) ? 1 : 0; + nb_zone = s + round; + } + printk("raid0 : nb_zone is %d.\n", nb_zone); + + printk("raid0 : Allocating %Zd bytes for hash.\n", + nb_zone*sizeof(struct strip_zone*)); + conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL); + if (!conf->hash_table) + goto out_free_conf; + size = conf->strip_zone[cur].size; + + for (i=0; i< nb_zone; i++) { + conf->hash_table[i] = conf->strip_zone + cur; + while (size <= conf->hash_spacing) { + cur++; + size += conf->strip_zone[cur].size; + } + size -= conf->hash_spacing; + } + if (conf->preshift) { + conf->hash_spacing >>= conf->preshift; + /* round hash_spacing up so when we divide by it, we + * err on the side of too-low, which is safest + */ + conf->hash_spacing++; + } + + /* calculate the max read-ahead size. + * For read-ahead of large files to be effective, we need to + * readahead at least twice a whole stripe. i.e. number of devices + * multiplied by chunk size times 2. + * If an individual device has an ra_pages greater than the + * chunk size, then we will not drive that device as hard as it + * wants. We consider this a configuration error: a larger + * chunksize should be used in that case. + */ + { + int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_CACHE_SIZE; + if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) + mddev->queue->backing_dev_info.ra_pages = 2* stripe; + } + + + blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec); + return 0; + +out_free_conf: + if (conf->strip_zone) + kfree(conf->strip_zone); + if (conf->devlist) + kfree (conf->devlist); + kfree(conf); + mddev->private = NULL; +out: + return 1; +} + +static int raid0_stop (mddev_t *mddev) +{ + raid0_conf_t *conf = mddev_to_conf(mddev); + + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + kfree (conf->hash_table); + conf->hash_table = NULL; + kfree (conf->strip_zone); + conf->strip_zone = NULL; + kfree (conf); + mddev->private = NULL; + + return 0; +} + +static int raid0_make_request (request_queue_t *q, struct bio *bio) +{ + mddev_t *mddev = q->queuedata; + unsigned int sect_in_chunk, chunksize_bits, chunk_size, chunk_sects; + raid0_conf_t *conf = mddev_to_conf(mddev); + struct strip_zone *zone; + mdk_rdev_t *tmp_dev; + unsigned long chunk; + sector_t block, rsect; + + if (bio_data_dir(bio)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); + } + + chunk_size = mddev->chunk_size >> 10; + chunk_sects = mddev->chunk_size >> 9; + chunksize_bits = ffz(~chunk_size); + block = bio->bi_sector >> 1; + + + if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) { + struct bio_pair *bp; + /* Sanity check -- queue functions should prevent this happening */ + if (bio->bi_vcnt != 1 || + bio->bi_idx != 0) + goto bad_map; + /* This is a one page bio that upper layers + * refuse to split for us, so we need to split it. + */ + bp = bio_split(bio, bio_split_pool, chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); + if (raid0_make_request(q, &bp->bio1)) + generic_make_request(&bp->bio1); + if (raid0_make_request(q, &bp->bio2)) + generic_make_request(&bp->bio2); + + bio_pair_release(bp); + return 0; + } + + + { +#if __GNUC__ < 3 + volatile +#endif + sector_t x = block >> conf->preshift; + sector_div(x, (unsigned long)conf->hash_spacing); + zone = conf->hash_table[x]; + } + + while (block >= (zone->zone_offset + zone->size)) + zone++; + + sect_in_chunk = bio->bi_sector & ((chunk_size<<1) -1); + + + { + sector_t x = (block - zone->zone_offset) >> chunksize_bits; + + sector_div(x, zone->nb_dev); + chunk = x; + BUG_ON(x != (sector_t)chunk); + + x = block >> chunksize_bits; + tmp_dev = zone->dev[sector_div(x, zone->nb_dev)]; + } + rsect = (((chunk << chunksize_bits) + zone->dev_offset)<<1) + + sect_in_chunk; + + bio->bi_bdev = tmp_dev->bdev; + bio->bi_sector = rsect + tmp_dev->data_offset; + + /* + * Let the main block layer submit the IO and resolve recursion: + */ + return 1; + +bad_map: + printk("raid0_make_request bug: can't convert block across chunks" + " or bigger than %dk %llu %d\n", chunk_size, + (unsigned long long)bio->bi_sector, bio->bi_size >> 10); + + bio_io_error(bio, bio->bi_size); + return 0; +} + +static void raid0_status (struct seq_file *seq, mddev_t *mddev) +{ +#undef MD_DEBUG +#ifdef MD_DEBUG + int j, k, h; + char b[BDEVNAME_SIZE]; + raid0_conf_t *conf = mddev_to_conf(mddev); + + h = 0; + for (j = 0; j < conf->nr_strip_zones; j++) { + seq_printf(seq, " z%d", j); + if (conf->hash_table[h] == conf->strip_zone+j) + seq_printf("(h%d)", h++); + seq_printf(seq, "=["); + for (k = 0; k < conf->strip_zone[j].nb_dev; k++) + seq_printf (seq, "%s/", bdevname( + conf->strip_zone[j].dev[k]->bdev,b)); + + seq_printf (seq, "] zo=%d do=%d s=%d\n", + conf->strip_zone[j].zone_offset, + conf->strip_zone[j].dev_offset, + conf->strip_zone[j].size); + } +#endif + seq_printf(seq, " %dk chunks", mddev->chunk_size/1024); + return; +} + +static mdk_personality_t raid0_personality= +{ + .name = "raid0", + .owner = THIS_MODULE, + .make_request = raid0_make_request, + .run = raid0_run, + .stop = raid0_stop, + .status = raid0_status, +}; + +static int __init raid0_init (void) +{ + return register_md_personality (RAID0, &raid0_personality); +} + +static void raid0_exit (void) +{ + unregister_md_personality (RAID0); +} + +module_init(raid0_init); +module_exit(raid0_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-2"); /* RAID0 */ diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c new file mode 100644 index 00000000000..a389394b52f --- /dev/null +++ b/drivers/md/raid1.c @@ -0,0 +1,1449 @@ +/* + * raid1.c : Multiple Devices driver for Linux + * + * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat + * + * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman + * + * RAID-1 management functions. + * + * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000 + * + * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk> + * Various fixes by Neil Brown <neilb@cse.unsw.edu.au> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +#include <linux/raid/raid1.h> + +/* + * Number of guaranteed r1bios in case of extreme VM load: + */ +#define NR_RAID1_BIOS 256 + +static mdk_personality_t raid1_personality; + +static void unplug_slaves(mddev_t *mddev); + + +static void * r1bio_pool_alloc(unsigned int __nocast gfp_flags, void *data) +{ + struct pool_info *pi = data; + r1bio_t *r1_bio; + int size = offsetof(r1bio_t, bios[pi->raid_disks]); + + /* allocate a r1bio with room for raid_disks entries in the bios array */ + r1_bio = kmalloc(size, gfp_flags); + if (r1_bio) + memset(r1_bio, 0, size); + else + unplug_slaves(pi->mddev); + + return r1_bio; +} + +static void r1bio_pool_free(void *r1_bio, void *data) +{ + kfree(r1_bio); +} + +#define RESYNC_BLOCK_SIZE (64*1024) +//#define RESYNC_BLOCK_SIZE PAGE_SIZE +#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) +#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) +#define RESYNC_WINDOW (2048*1024) + +static void * r1buf_pool_alloc(unsigned int __nocast gfp_flags, void *data) +{ + struct pool_info *pi = data; + struct page *page; + r1bio_t *r1_bio; + struct bio *bio; + int i, j; + + r1_bio = r1bio_pool_alloc(gfp_flags, pi); + if (!r1_bio) { + unplug_slaves(pi->mddev); + return NULL; + } + + /* + * Allocate bios : 1 for reading, n-1 for writing + */ + for (j = pi->raid_disks ; j-- ; ) { + bio = bio_alloc(gfp_flags, RESYNC_PAGES); + if (!bio) + goto out_free_bio; + r1_bio->bios[j] = bio; + } + /* + * Allocate RESYNC_PAGES data pages and attach them to + * the first bio; + */ + bio = r1_bio->bios[0]; + for (i = 0; i < RESYNC_PAGES; i++) { + page = alloc_page(gfp_flags); + if (unlikely(!page)) + goto out_free_pages; + + bio->bi_io_vec[i].bv_page = page; + } + + r1_bio->master_bio = NULL; + + return r1_bio; + +out_free_pages: + for ( ; i > 0 ; i--) + __free_page(bio->bi_io_vec[i-1].bv_page); +out_free_bio: + while ( ++j < pi->raid_disks ) + bio_put(r1_bio->bios[j]); + r1bio_pool_free(r1_bio, data); + return NULL; +} + +static void r1buf_pool_free(void *__r1_bio, void *data) +{ + struct pool_info *pi = data; + int i; + r1bio_t *r1bio = __r1_bio; + struct bio *bio = r1bio->bios[0]; + + for (i = 0; i < RESYNC_PAGES; i++) { + __free_page(bio->bi_io_vec[i].bv_page); + bio->bi_io_vec[i].bv_page = NULL; + } + for (i=0 ; i < pi->raid_disks; i++) + bio_put(r1bio->bios[i]); + + r1bio_pool_free(r1bio, data); +} + +static void put_all_bios(conf_t *conf, r1bio_t *r1_bio) +{ + int i; + + for (i = 0; i < conf->raid_disks; i++) { + struct bio **bio = r1_bio->bios + i; + if (*bio) + bio_put(*bio); + *bio = NULL; + } +} + +static inline void free_r1bio(r1bio_t *r1_bio) +{ + unsigned long flags; + + conf_t *conf = mddev_to_conf(r1_bio->mddev); + + /* + * Wake up any possible resync thread that waits for the device + * to go idle. + */ + spin_lock_irqsave(&conf->resync_lock, flags); + if (!--conf->nr_pending) { + wake_up(&conf->wait_idle); + wake_up(&conf->wait_resume); + } + spin_unlock_irqrestore(&conf->resync_lock, flags); + + put_all_bios(conf, r1_bio); + mempool_free(r1_bio, conf->r1bio_pool); +} + +static inline void put_buf(r1bio_t *r1_bio) +{ + conf_t *conf = mddev_to_conf(r1_bio->mddev); + unsigned long flags; + + mempool_free(r1_bio, conf->r1buf_pool); + + spin_lock_irqsave(&conf->resync_lock, flags); + if (!conf->barrier) + BUG(); + --conf->barrier; + wake_up(&conf->wait_resume); + wake_up(&conf->wait_idle); + + if (!--conf->nr_pending) { + wake_up(&conf->wait_idle); + wake_up(&conf->wait_resume); + } + spin_unlock_irqrestore(&conf->resync_lock, flags); +} + +static void reschedule_retry(r1bio_t *r1_bio) +{ + unsigned long flags; + mddev_t *mddev = r1_bio->mddev; + conf_t *conf = mddev_to_conf(mddev); + + spin_lock_irqsave(&conf->device_lock, flags); + list_add(&r1_bio->retry_list, &conf->retry_list); + spin_unlock_irqrestore(&conf->device_lock, flags); + + md_wakeup_thread(mddev->thread); +} + +/* + * raid_end_bio_io() is called when we have finished servicing a mirrored + * operation and are ready to return a success/failure code to the buffer + * cache layer. + */ +static void raid_end_bio_io(r1bio_t *r1_bio) +{ + struct bio *bio = r1_bio->master_bio; + + bio_endio(bio, bio->bi_size, + test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO); + free_r1bio(r1_bio); +} + +/* + * Update disk head position estimator based on IRQ completion info. + */ +static inline void update_head_pos(int disk, r1bio_t *r1_bio) +{ + conf_t *conf = mddev_to_conf(r1_bio->mddev); + + conf->mirrors[disk].head_position = + r1_bio->sector + (r1_bio->sectors); +} + +static int raid1_end_read_request(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); + int mirror; + conf_t *conf = mddev_to_conf(r1_bio->mddev); + + if (bio->bi_size) + return 1; + + mirror = r1_bio->read_disk; + /* + * this branch is our 'one mirror IO has finished' event handler: + */ + if (!uptodate) + md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); + else + /* + * Set R1BIO_Uptodate in our master bio, so that + * we will return a good error code for to the higher + * levels even if IO on some other mirrored buffer fails. + * + * The 'master' represents the composite IO operation to + * user-side. So if something waits for IO, then it will + * wait for the 'master' bio. + */ + set_bit(R1BIO_Uptodate, &r1_bio->state); + + update_head_pos(mirror, r1_bio); + + /* + * we have only one bio on the read side + */ + if (uptodate) + raid_end_bio_io(r1_bio); + else { + /* + * oops, read error: + */ + char b[BDEVNAME_SIZE]; + if (printk_ratelimit()) + printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n", + bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector); + reschedule_retry(r1_bio); + } + + rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); + return 0; +} + +static int raid1_end_write_request(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); + int mirror; + conf_t *conf = mddev_to_conf(r1_bio->mddev); + + if (bio->bi_size) + return 1; + + for (mirror = 0; mirror < conf->raid_disks; mirror++) + if (r1_bio->bios[mirror] == bio) + break; + + /* + * this branch is our 'one mirror IO has finished' event handler: + */ + if (!uptodate) + md_error(r1_bio->mddev, conf->mirrors[mirror].rdev); + else + /* + * Set R1BIO_Uptodate in our master bio, so that + * we will return a good error code for to the higher + * levels even if IO on some other mirrored buffer fails. + * + * The 'master' represents the composite IO operation to + * user-side. So if something waits for IO, then it will + * wait for the 'master' bio. + */ + set_bit(R1BIO_Uptodate, &r1_bio->state); + + update_head_pos(mirror, r1_bio); + + /* + * + * Let's see if all mirrored write operations have finished + * already. + */ + if (atomic_dec_and_test(&r1_bio->remaining)) { + md_write_end(r1_bio->mddev); + raid_end_bio_io(r1_bio); + } + + rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev); + return 0; +} + + +/* + * This routine returns the disk from which the requested read should + * be done. There is a per-array 'next expected sequential IO' sector + * number - if this matches on the next IO then we use the last disk. + * There is also a per-disk 'last know head position' sector that is + * maintained from IRQ contexts, both the normal and the resync IO + * completion handlers update this position correctly. If there is no + * perfect sequential match then we pick the disk whose head is closest. + * + * If there are 2 mirrors in the same 2 devices, performance degrades + * because position is mirror, not device based. + * + * The rdev for the device selected will have nr_pending incremented. + */ +static int read_balance(conf_t *conf, r1bio_t *r1_bio) +{ + const unsigned long this_sector = r1_bio->sector; + int new_disk = conf->last_used, disk = new_disk; + const int sectors = r1_bio->sectors; + sector_t new_distance, current_distance; + mdk_rdev_t *new_rdev, *rdev; + + rcu_read_lock(); + /* + * Check if it if we can balance. We can balance on the whole + * device if no resync is going on, or below the resync window. + * We take the first readable disk when above the resync window. + */ + retry: + if (conf->mddev->recovery_cp < MaxSector && + (this_sector + sectors >= conf->next_resync)) { + /* Choose the first operation device, for consistancy */ + new_disk = 0; + + while ((new_rdev=conf->mirrors[new_disk].rdev) == NULL || + !new_rdev->in_sync) { + new_disk++; + if (new_disk == conf->raid_disks) { + new_disk = -1; + break; + } + } + goto rb_out; + } + + + /* make sure the disk is operational */ + while ((new_rdev=conf->mirrors[new_disk].rdev) == NULL || + !new_rdev->in_sync) { + if (new_disk <= 0) + new_disk = conf->raid_disks; + new_disk--; + if (new_disk == disk) { + new_disk = -1; + goto rb_out; + } + } + disk = new_disk; + /* now disk == new_disk == starting point for search */ + + /* + * Don't change to another disk for sequential reads: + */ + if (conf->next_seq_sect == this_sector) + goto rb_out; + if (this_sector == conf->mirrors[new_disk].head_position) + goto rb_out; + + current_distance = abs(this_sector - conf->mirrors[disk].head_position); + + /* Find the disk whose head is closest */ + + do { + if (disk <= 0) + disk = conf->raid_disks; + disk--; + + if ((rdev=conf->mirrors[disk].rdev) == NULL || + !rdev->in_sync) + continue; + + if (!atomic_read(&rdev->nr_pending)) { + new_disk = disk; + new_rdev = rdev; + break; + } + new_distance = abs(this_sector - conf->mirrors[disk].head_position); + if (new_distance < current_distance) { + current_distance = new_distance; + new_disk = disk; + new_rdev = rdev; + } + } while (disk != conf->last_used); + +rb_out: + + + if (new_disk >= 0) { + conf->next_seq_sect = this_sector + sectors; + conf->last_used = new_disk; + atomic_inc(&new_rdev->nr_pending); + if (!new_rdev->in_sync) { + /* cannot risk returning a device that failed + * before we inc'ed nr_pending + */ + atomic_dec(&new_rdev->nr_pending); + goto retry; + } + } + rcu_read_unlock(); + + return new_disk; +} + +static void unplug_slaves(mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + int i; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks; i++) { + mdk_rdev_t *rdev = conf->mirrors[i].rdev; + if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) { + request_queue_t *r_queue = bdev_get_queue(rdev->bdev); + + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + rcu_read_unlock(); +} + +static void raid1_unplug(request_queue_t *q) +{ + unplug_slaves(q->queuedata); +} + +static int raid1_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + mdk_rdev_t *rdev = conf->mirrors[i].rdev; + if (rdev && !rdev->faulty) { + struct block_device *bdev = rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, + error_sector); + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + } + rcu_read_unlock(); + return ret; +} + +/* + * Throttle resync depth, so that we can both get proper overlapping of + * requests, but are still able to handle normal requests quickly. + */ +#define RESYNC_DEPTH 32 + +static void device_barrier(conf_t *conf, sector_t sect) +{ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_idle, !waitqueue_active(&conf->wait_resume), + conf->resync_lock, unplug_slaves(conf->mddev)); + + if (!conf->barrier++) { + wait_event_lock_irq(conf->wait_idle, !conf->nr_pending, + conf->resync_lock, unplug_slaves(conf->mddev)); + if (conf->nr_pending) + BUG(); + } + wait_event_lock_irq(conf->wait_resume, conf->barrier < RESYNC_DEPTH, + conf->resync_lock, unplug_slaves(conf->mddev)); + conf->next_resync = sect; + spin_unlock_irq(&conf->resync_lock); +} + +static int make_request(request_queue_t *q, struct bio * bio) +{ + mddev_t *mddev = q->queuedata; + conf_t *conf = mddev_to_conf(mddev); + mirror_info_t *mirror; + r1bio_t *r1_bio; + struct bio *read_bio; + int i, disks; + mdk_rdev_t *rdev; + + /* + * Register the new request and wait if the reconstruction + * thread has put up a bar for new requests. + * Continue immediately if no resync is active currently. + */ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_resume, !conf->barrier, conf->resync_lock, ); + conf->nr_pending++; + spin_unlock_irq(&conf->resync_lock); + + if (bio_data_dir(bio)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); + } + + /* + * make_request() can abort the operation when READA is being + * used and no empty request is available. + * + */ + r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO); + + r1_bio->master_bio = bio; + r1_bio->sectors = bio->bi_size >> 9; + + r1_bio->mddev = mddev; + r1_bio->sector = bio->bi_sector; + + r1_bio->state = 0; + + if (bio_data_dir(bio) == READ) { + /* + * read balancing logic: + */ + int rdisk = read_balance(conf, r1_bio); + + if (rdisk < 0) { + /* couldn't find anywhere to read from */ + raid_end_bio_io(r1_bio); + return 0; + } + mirror = conf->mirrors + rdisk; + + r1_bio->read_disk = rdisk; + + read_bio = bio_clone(bio, GFP_NOIO); + + r1_bio->bios[rdisk] = read_bio; + + read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset; + read_bio->bi_bdev = mirror->rdev->bdev; + read_bio->bi_end_io = raid1_end_read_request; + read_bio->bi_rw = READ; + read_bio->bi_private = r1_bio; + + generic_make_request(read_bio); + return 0; + } + + /* + * WRITE: + */ + /* first select target devices under spinlock and + * inc refcount on their rdev. Record them by setting + * bios[x] to bio + */ + disks = conf->raid_disks; + rcu_read_lock(); + for (i = 0; i < disks; i++) { + if ((rdev=conf->mirrors[i].rdev) != NULL && + !rdev->faulty) { + atomic_inc(&rdev->nr_pending); + if (rdev->faulty) { + atomic_dec(&rdev->nr_pending); + r1_bio->bios[i] = NULL; + } else + r1_bio->bios[i] = bio; + } else + r1_bio->bios[i] = NULL; + } + rcu_read_unlock(); + + atomic_set(&r1_bio->remaining, 1); + md_write_start(mddev); + for (i = 0; i < disks; i++) { + struct bio *mbio; + if (!r1_bio->bios[i]) + continue; + + mbio = bio_clone(bio, GFP_NOIO); + r1_bio->bios[i] = mbio; + + mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset; + mbio->bi_bdev = conf->mirrors[i].rdev->bdev; + mbio->bi_end_io = raid1_end_write_request; + mbio->bi_rw = WRITE; + mbio->bi_private = r1_bio; + + atomic_inc(&r1_bio->remaining); + generic_make_request(mbio); + } + + if (atomic_dec_and_test(&r1_bio->remaining)) { + md_write_end(mddev); + raid_end_bio_io(r1_bio); + } + + return 0; +} + +static void status(struct seq_file *seq, mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + int i; + + seq_printf(seq, " [%d/%d] [", conf->raid_disks, + conf->working_disks); + for (i = 0; i < conf->raid_disks; i++) + seq_printf(seq, "%s", + conf->mirrors[i].rdev && + conf->mirrors[i].rdev->in_sync ? "U" : "_"); + seq_printf(seq, "]"); +} + + +static void error(mddev_t *mddev, mdk_rdev_t *rdev) +{ + char b[BDEVNAME_SIZE]; + conf_t *conf = mddev_to_conf(mddev); + + /* + * If it is not operational, then we have already marked it as dead + * else if it is the last working disks, ignore the error, let the + * next level up know. + * else mark the drive as failed + */ + if (rdev->in_sync + && conf->working_disks == 1) + /* + * Don't fail the drive, act as though we were just a + * normal single drive + */ + return; + if (rdev->in_sync) { + mddev->degraded++; + conf->working_disks--; + /* + * if recovery is running, make sure it aborts. + */ + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + } + rdev->in_sync = 0; + rdev->faulty = 1; + mddev->sb_dirty = 1; + printk(KERN_ALERT "raid1: Disk failure on %s, disabling device. \n" + " Operation continuing on %d devices\n", + bdevname(rdev->bdev,b), conf->working_disks); +} + +static void print_conf(conf_t *conf) +{ + int i; + mirror_info_t *tmp; + + printk("RAID1 conf printout:\n"); + if (!conf) { + printk("(!conf)\n"); + return; + } + printk(" --- wd:%d rd:%d\n", conf->working_disks, + conf->raid_disks); + + for (i = 0; i < conf->raid_disks; i++) { + char b[BDEVNAME_SIZE]; + tmp = conf->mirrors + i; + if (tmp->rdev) + printk(" disk %d, wo:%d, o:%d, dev:%s\n", + i, !tmp->rdev->in_sync, !tmp->rdev->faulty, + bdevname(tmp->rdev->bdev,b)); + } +} + +static void close_sync(conf_t *conf) +{ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_resume, !conf->barrier, + conf->resync_lock, unplug_slaves(conf->mddev)); + spin_unlock_irq(&conf->resync_lock); + + if (conf->barrier) BUG(); + if (waitqueue_active(&conf->wait_idle)) BUG(); + + mempool_destroy(conf->r1buf_pool); + conf->r1buf_pool = NULL; +} + +static int raid1_spare_active(mddev_t *mddev) +{ + int i; + conf_t *conf = mddev->private; + mirror_info_t *tmp; + + /* + * Find all failed disks within the RAID1 configuration + * and mark them readable + */ + for (i = 0; i < conf->raid_disks; i++) { + tmp = conf->mirrors + i; + if (tmp->rdev + && !tmp->rdev->faulty + && !tmp->rdev->in_sync) { + conf->working_disks++; + mddev->degraded--; + tmp->rdev->in_sync = 1; + } + } + + print_conf(conf); + return 0; +} + + +static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) +{ + conf_t *conf = mddev->private; + int found = 0; + int mirror; + mirror_info_t *p; + + for (mirror=0; mirror < mddev->raid_disks; mirror++) + if ( !(p=conf->mirrors+mirror)->rdev) { + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + p->head_position = 0; + rdev->raid_disk = mirror; + found = 1; + p->rdev = rdev; + break; + } + + print_conf(conf); + return found; +} + +static int raid1_remove_disk(mddev_t *mddev, int number) +{ + conf_t *conf = mddev->private; + int err = 0; + mdk_rdev_t *rdev; + mirror_info_t *p = conf->mirrors+ number; + + print_conf(conf); + rdev = p->rdev; + if (rdev) { + if (rdev->in_sync || + atomic_read(&rdev->nr_pending)) { + err = -EBUSY; + goto abort; + } + p->rdev = NULL; + synchronize_kernel(); + if (atomic_read(&rdev->nr_pending)) { + /* lost the race, try later */ + err = -EBUSY; + p->rdev = rdev; + } + } +abort: + + print_conf(conf); + return err; +} + + +static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); + conf_t *conf = mddev_to_conf(r1_bio->mddev); + + if (bio->bi_size) + return 1; + + if (r1_bio->bios[r1_bio->read_disk] != bio) + BUG(); + update_head_pos(r1_bio->read_disk, r1_bio); + /* + * we have read a block, now it needs to be re-written, + * or re-read if the read failed. + * We don't do much here, just schedule handling by raid1d + */ + if (!uptodate) + md_error(r1_bio->mddev, + conf->mirrors[r1_bio->read_disk].rdev); + else + set_bit(R1BIO_Uptodate, &r1_bio->state); + rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev); + reschedule_retry(r1_bio); + return 0; +} + +static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private); + mddev_t *mddev = r1_bio->mddev; + conf_t *conf = mddev_to_conf(mddev); + int i; + int mirror=0; + + if (bio->bi_size) + return 1; + + for (i = 0; i < conf->raid_disks; i++) + if (r1_bio->bios[i] == bio) { + mirror = i; + break; + } + if (!uptodate) + md_error(mddev, conf->mirrors[mirror].rdev); + update_head_pos(mirror, r1_bio); + + if (atomic_dec_and_test(&r1_bio->remaining)) { + md_done_sync(mddev, r1_bio->sectors, uptodate); + put_buf(r1_bio); + } + rdev_dec_pending(conf->mirrors[mirror].rdev, mddev); + return 0; +} + +static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio) +{ + conf_t *conf = mddev_to_conf(mddev); + int i; + int disks = conf->raid_disks; + struct bio *bio, *wbio; + + bio = r1_bio->bios[r1_bio->read_disk]; + + /* + * schedule writes + */ + if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { + /* + * There is no point trying a read-for-reconstruct as + * reconstruct is about to be aborted + */ + char b[BDEVNAME_SIZE]; + printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error" + " for block %llu\n", + bdevname(bio->bi_bdev,b), + (unsigned long long)r1_bio->sector); + md_done_sync(mddev, r1_bio->sectors, 0); + put_buf(r1_bio); + return; + } + + atomic_set(&r1_bio->remaining, 1); + for (i = 0; i < disks ; i++) { + wbio = r1_bio->bios[i]; + if (wbio->bi_end_io != end_sync_write) + continue; + + atomic_inc(&conf->mirrors[i].rdev->nr_pending); + atomic_inc(&r1_bio->remaining); + md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9); + generic_make_request(wbio); + } + + if (atomic_dec_and_test(&r1_bio->remaining)) { + md_done_sync(mddev, r1_bio->sectors, 1); + put_buf(r1_bio); + } +} + +/* + * This is a kernel thread which: + * + * 1. Retries failed read operations on working mirrors. + * 2. Updates the raid superblock when problems encounter. + * 3. Performs writes following reads for array syncronising. + */ + +static void raid1d(mddev_t *mddev) +{ + r1bio_t *r1_bio; + struct bio *bio; + unsigned long flags; + conf_t *conf = mddev_to_conf(mddev); + struct list_head *head = &conf->retry_list; + int unplug=0; + mdk_rdev_t *rdev; + + md_check_recovery(mddev); + md_handle_safemode(mddev); + + for (;;) { + char b[BDEVNAME_SIZE]; + spin_lock_irqsave(&conf->device_lock, flags); + if (list_empty(head)) + break; + r1_bio = list_entry(head->prev, r1bio_t, retry_list); + list_del(head->prev); + spin_unlock_irqrestore(&conf->device_lock, flags); + + mddev = r1_bio->mddev; + conf = mddev_to_conf(mddev); + if (test_bit(R1BIO_IsSync, &r1_bio->state)) { + sync_request_write(mddev, r1_bio); + unplug = 1; + } else { + int disk; + bio = r1_bio->bios[r1_bio->read_disk]; + if ((disk=read_balance(conf, r1_bio)) == -1) { + printk(KERN_ALERT "raid1: %s: unrecoverable I/O" + " read error for block %llu\n", + bdevname(bio->bi_bdev,b), + (unsigned long long)r1_bio->sector); + raid_end_bio_io(r1_bio); + } else { + r1_bio->bios[r1_bio->read_disk] = NULL; + r1_bio->read_disk = disk; + bio_put(bio); + bio = bio_clone(r1_bio->master_bio, GFP_NOIO); + r1_bio->bios[r1_bio->read_disk] = bio; + rdev = conf->mirrors[disk].rdev; + if (printk_ratelimit()) + printk(KERN_ERR "raid1: %s: redirecting sector %llu to" + " another mirror\n", + bdevname(rdev->bdev,b), + (unsigned long long)r1_bio->sector); + bio->bi_sector = r1_bio->sector + rdev->data_offset; + bio->bi_bdev = rdev->bdev; + bio->bi_end_io = raid1_end_read_request; + bio->bi_rw = READ; + bio->bi_private = r1_bio; + unplug = 1; + generic_make_request(bio); + } + } + } + spin_unlock_irqrestore(&conf->device_lock, flags); + if (unplug) + unplug_slaves(mddev); +} + + +static int init_resync(conf_t *conf) +{ + int buffs; + + buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; + if (conf->r1buf_pool) + BUG(); + conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free, + conf->poolinfo); + if (!conf->r1buf_pool) + return -ENOMEM; + conf->next_resync = 0; + return 0; +} + +/* + * perform a "sync" on one "block" + * + * We need to make sure that no normal I/O request - particularly write + * requests - conflict with active sync requests. + * + * This is achieved by tracking pending requests and a 'barrier' concept + * that can be installed to exclude normal IO requests. + */ + +static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) +{ + conf_t *conf = mddev_to_conf(mddev); + mirror_info_t *mirror; + r1bio_t *r1_bio; + struct bio *bio; + sector_t max_sector, nr_sectors; + int disk; + int i; + int write_targets = 0; + + if (!conf->r1buf_pool) + if (init_resync(conf)) + return -ENOMEM; + + max_sector = mddev->size << 1; + if (sector_nr >= max_sector) { + close_sync(conf); + return 0; + } + + /* + * If there is non-resync activity waiting for us then + * put in a delay to throttle resync. + */ + if (!go_faster && waitqueue_active(&conf->wait_resume)) + msleep_interruptible(1000); + device_barrier(conf, sector_nr + RESYNC_SECTORS); + + /* + * If reconstructing, and >1 working disc, + * could dedicate one to rebuild and others to + * service read requests .. + */ + disk = conf->last_used; + /* make sure disk is operational */ + + while (conf->mirrors[disk].rdev == NULL || + !conf->mirrors[disk].rdev->in_sync) { + if (disk <= 0) + disk = conf->raid_disks; + disk--; + if (disk == conf->last_used) + break; + } + conf->last_used = disk; + atomic_inc(&conf->mirrors[disk].rdev->nr_pending); + + + mirror = conf->mirrors + disk; + + r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO); + + spin_lock_irq(&conf->resync_lock); + conf->nr_pending++; + spin_unlock_irq(&conf->resync_lock); + + r1_bio->mddev = mddev; + r1_bio->sector = sector_nr; + set_bit(R1BIO_IsSync, &r1_bio->state); + r1_bio->read_disk = disk; + + for (i=0; i < conf->raid_disks; i++) { + bio = r1_bio->bios[i]; + + /* take from bio_init */ + bio->bi_next = NULL; + bio->bi_flags |= 1 << BIO_UPTODATE; + bio->bi_rw = 0; + bio->bi_vcnt = 0; + bio->bi_idx = 0; + bio->bi_phys_segments = 0; + bio->bi_hw_segments = 0; + bio->bi_size = 0; + bio->bi_end_io = NULL; + bio->bi_private = NULL; + + if (i == disk) { + bio->bi_rw = READ; + bio->bi_end_io = end_sync_read; + } else if (conf->mirrors[i].rdev && + !conf->mirrors[i].rdev->faulty && + (!conf->mirrors[i].rdev->in_sync || + sector_nr + RESYNC_SECTORS > mddev->recovery_cp)) { + bio->bi_rw = WRITE; + bio->bi_end_io = end_sync_write; + write_targets ++; + } else + continue; + bio->bi_sector = sector_nr + conf->mirrors[i].rdev->data_offset; + bio->bi_bdev = conf->mirrors[i].rdev->bdev; + bio->bi_private = r1_bio; + } + if (write_targets == 0) { + /* There is nowhere to write, so all non-sync + * drives must be failed - so we are finished + */ + int rv = max_sector - sector_nr; + md_done_sync(mddev, rv, 1); + put_buf(r1_bio); + rdev_dec_pending(conf->mirrors[disk].rdev, mddev); + return rv; + } + + nr_sectors = 0; + do { + struct page *page; + int len = PAGE_SIZE; + if (sector_nr + (len>>9) > max_sector) + len = (max_sector - sector_nr) << 9; + if (len == 0) + break; + for (i=0 ; i < conf->raid_disks; i++) { + bio = r1_bio->bios[i]; + if (bio->bi_end_io) { + page = r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page; + if (bio_add_page(bio, page, len, 0) == 0) { + /* stop here */ + r1_bio->bios[0]->bi_io_vec[bio->bi_vcnt].bv_page = page; + while (i > 0) { + i--; + bio = r1_bio->bios[i]; + if (bio->bi_end_io==NULL) continue; + /* remove last page from this bio */ + bio->bi_vcnt--; + bio->bi_size -= len; + bio->bi_flags &= ~(1<< BIO_SEG_VALID); + } + goto bio_full; + } + } + } + nr_sectors += len>>9; + sector_nr += len>>9; + } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES); + bio_full: + bio = r1_bio->bios[disk]; + r1_bio->sectors = nr_sectors; + + md_sync_acct(mirror->rdev->bdev, nr_sectors); + + generic_make_request(bio); + + return nr_sectors; +} + +static int run(mddev_t *mddev) +{ + conf_t *conf; + int i, j, disk_idx; + mirror_info_t *disk; + mdk_rdev_t *rdev; + struct list_head *tmp; + + if (mddev->level != 1) { + printk("raid1: %s: raid level not set to mirroring (%d)\n", + mdname(mddev), mddev->level); + goto out; + } + /* + * copy the already verified devices into our private RAID1 + * bookkeeping area. [whatever we allocate in run(), + * should be freed in stop()] + */ + conf = kmalloc(sizeof(conf_t), GFP_KERNEL); + mddev->private = conf; + if (!conf) + goto out_no_mem; + + memset(conf, 0, sizeof(*conf)); + conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks, + GFP_KERNEL); + if (!conf->mirrors) + goto out_no_mem; + + memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks); + + conf->poolinfo = kmalloc(sizeof(*conf->poolinfo), GFP_KERNEL); + if (!conf->poolinfo) + goto out_no_mem; + conf->poolinfo->mddev = mddev; + conf->poolinfo->raid_disks = mddev->raid_disks; + conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, + r1bio_pool_free, + conf->poolinfo); + if (!conf->r1bio_pool) + goto out_no_mem; + + mddev->queue->unplug_fn = raid1_unplug; + + mddev->queue->issue_flush_fn = raid1_issue_flush; + + ITERATE_RDEV(mddev, rdev, tmp) { + disk_idx = rdev->raid_disk; + if (disk_idx >= mddev->raid_disks + || disk_idx < 0) + continue; + disk = conf->mirrors + disk_idx; + + disk->rdev = rdev; + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9); + + disk->head_position = 0; + if (!rdev->faulty && rdev->in_sync) + conf->working_disks++; + } + conf->raid_disks = mddev->raid_disks; + conf->mddev = mddev; + spin_lock_init(&conf->device_lock); + INIT_LIST_HEAD(&conf->retry_list); + if (conf->working_disks == 1) + mddev->recovery_cp = MaxSector; + + spin_lock_init(&conf->resync_lock); + init_waitqueue_head(&conf->wait_idle); + init_waitqueue_head(&conf->wait_resume); + + if (!conf->working_disks) { + printk(KERN_ERR "raid1: no operational mirrors for %s\n", + mdname(mddev)); + goto out_free_conf; + } + + mddev->degraded = 0; + for (i = 0; i < conf->raid_disks; i++) { + + disk = conf->mirrors + i; + + if (!disk->rdev) { + disk->head_position = 0; + mddev->degraded++; + } + } + + /* + * find the first working one and use it as a starting point + * to read balancing. + */ + for (j = 0; j < conf->raid_disks && + (!conf->mirrors[j].rdev || + !conf->mirrors[j].rdev->in_sync) ; j++) + /* nothing */; + conf->last_used = j; + + + + { + mddev->thread = md_register_thread(raid1d, mddev, "%s_raid1"); + if (!mddev->thread) { + printk(KERN_ERR + "raid1: couldn't allocate thread for %s\n", + mdname(mddev)); + goto out_free_conf; + } + } + printk(KERN_INFO + "raid1: raid set %s active with %d out of %d mirrors\n", + mdname(mddev), mddev->raid_disks - mddev->degraded, + mddev->raid_disks); + /* + * Ok, everything is just fine now + */ + mddev->array_size = mddev->size; + + return 0; + +out_no_mem: + printk(KERN_ERR "raid1: couldn't allocate memory for %s\n", + mdname(mddev)); + +out_free_conf: + if (conf) { + if (conf->r1bio_pool) + mempool_destroy(conf->r1bio_pool); + if (conf->mirrors) + kfree(conf->mirrors); + if (conf->poolinfo) + kfree(conf->poolinfo); + kfree(conf); + mddev->private = NULL; + } +out: + return -EIO; +} + +static int stop(mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + + md_unregister_thread(mddev->thread); + mddev->thread = NULL; + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + if (conf->r1bio_pool) + mempool_destroy(conf->r1bio_pool); + if (conf->mirrors) + kfree(conf->mirrors); + if (conf->poolinfo) + kfree(conf->poolinfo); + kfree(conf); + mddev->private = NULL; + return 0; +} + +static int raid1_resize(mddev_t *mddev, sector_t sectors) +{ + /* no resync is happening, and there is enough space + * on all devices, so we can resize. + * We need to make sure resync covers any new space. + * If the array is shrinking we should possibly wait until + * any io in the removed space completes, but it hardly seems + * worth it. + */ + mddev->array_size = sectors>>1; + set_capacity(mddev->gendisk, mddev->array_size << 1); + mddev->changed = 1; + if (mddev->array_size > mddev->size && mddev->recovery_cp == MaxSector) { + mddev->recovery_cp = mddev->size << 1; + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + } + mddev->size = mddev->array_size; + return 0; +} + +static int raid1_reshape(mddev_t *mddev, int raid_disks) +{ + /* We need to: + * 1/ resize the r1bio_pool + * 2/ resize conf->mirrors + * + * We allocate a new r1bio_pool if we can. + * Then raise a device barrier and wait until all IO stops. + * Then resize conf->mirrors and swap in the new r1bio pool. + */ + mempool_t *newpool, *oldpool; + struct pool_info *newpoolinfo; + mirror_info_t *newmirrors; + conf_t *conf = mddev_to_conf(mddev); + + int d; + + for (d= raid_disks; d < conf->raid_disks; d++) + if (conf->mirrors[d].rdev) + return -EBUSY; + + newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL); + if (!newpoolinfo) + return -ENOMEM; + newpoolinfo->mddev = mddev; + newpoolinfo->raid_disks = raid_disks; + + newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc, + r1bio_pool_free, newpoolinfo); + if (!newpool) { + kfree(newpoolinfo); + return -ENOMEM; + } + newmirrors = kmalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL); + if (!newmirrors) { + kfree(newpoolinfo); + mempool_destroy(newpool); + return -ENOMEM; + } + memset(newmirrors, 0, sizeof(struct mirror_info)*raid_disks); + + spin_lock_irq(&conf->resync_lock); + conf->barrier++; + wait_event_lock_irq(conf->wait_idle, !conf->nr_pending, + conf->resync_lock, unplug_slaves(mddev)); + spin_unlock_irq(&conf->resync_lock); + + /* ok, everything is stopped */ + oldpool = conf->r1bio_pool; + conf->r1bio_pool = newpool; + for (d=0; d < raid_disks && d < conf->raid_disks; d++) + newmirrors[d] = conf->mirrors[d]; + kfree(conf->mirrors); + conf->mirrors = newmirrors; + kfree(conf->poolinfo); + conf->poolinfo = newpoolinfo; + + mddev->degraded += (raid_disks - conf->raid_disks); + conf->raid_disks = mddev->raid_disks = raid_disks; + + spin_lock_irq(&conf->resync_lock); + conf->barrier--; + spin_unlock_irq(&conf->resync_lock); + wake_up(&conf->wait_resume); + wake_up(&conf->wait_idle); + + + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + md_wakeup_thread(mddev->thread); + + mempool_destroy(oldpool); + return 0; +} + + +static mdk_personality_t raid1_personality = +{ + .name = "raid1", + .owner = THIS_MODULE, + .make_request = make_request, + .run = run, + .stop = stop, + .status = status, + .error_handler = error, + .hot_add_disk = raid1_add_disk, + .hot_remove_disk= raid1_remove_disk, + .spare_active = raid1_spare_active, + .sync_request = sync_request, + .resize = raid1_resize, + .reshape = raid1_reshape, +}; + +static int __init raid_init(void) +{ + return register_md_personality(RAID1, &raid1_personality); +} + +static void raid_exit(void) +{ + unregister_md_personality(RAID1); +} + +module_init(raid_init); +module_exit(raid_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-3"); /* RAID1 */ diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c new file mode 100644 index 00000000000..b100bfe4fdc --- /dev/null +++ b/drivers/md/raid10.c @@ -0,0 +1,1787 @@ +/* + * raid10.c : Multiple Devices driver for Linux + * + * Copyright (C) 2000-2004 Neil Brown + * + * RAID-10 support for md. + * + * Base on code in raid1.c. See raid1.c for futher copyright information. + * + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +#include <linux/raid/raid10.h> + +/* + * RAID10 provides a combination of RAID0 and RAID1 functionality. + * The layout of data is defined by + * chunk_size + * raid_disks + * near_copies (stored in low byte of layout) + * far_copies (stored in second byte of layout) + * + * The data to be stored is divided into chunks using chunksize. + * Each device is divided into far_copies sections. + * In each section, chunks are laid out in a style similar to raid0, but + * near_copies copies of each chunk is stored (each on a different drive). + * The starting device for each section is offset near_copies from the starting + * device of the previous section. + * Thus there are (near_copies*far_copies) of each chunk, and each is on a different + * drive. + * near_copies and far_copies must be at least one, and their product is at most + * raid_disks. + */ + +/* + * Number of guaranteed r10bios in case of extreme VM load: + */ +#define NR_RAID10_BIOS 256 + +static void unplug_slaves(mddev_t *mddev); + +static void * r10bio_pool_alloc(unsigned int __nocast gfp_flags, void *data) +{ + conf_t *conf = data; + r10bio_t *r10_bio; + int size = offsetof(struct r10bio_s, devs[conf->copies]); + + /* allocate a r10bio with room for raid_disks entries in the bios array */ + r10_bio = kmalloc(size, gfp_flags); + if (r10_bio) + memset(r10_bio, 0, size); + else + unplug_slaves(conf->mddev); + + return r10_bio; +} + +static void r10bio_pool_free(void *r10_bio, void *data) +{ + kfree(r10_bio); +} + +#define RESYNC_BLOCK_SIZE (64*1024) +//#define RESYNC_BLOCK_SIZE PAGE_SIZE +#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9) +#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE) +#define RESYNC_WINDOW (2048*1024) + +/* + * When performing a resync, we need to read and compare, so + * we need as many pages are there are copies. + * When performing a recovery, we need 2 bios, one for read, + * one for write (we recover only one drive per r10buf) + * + */ +static void * r10buf_pool_alloc(unsigned int __nocast gfp_flags, void *data) +{ + conf_t *conf = data; + struct page *page; + r10bio_t *r10_bio; + struct bio *bio; + int i, j; + int nalloc; + + r10_bio = r10bio_pool_alloc(gfp_flags, conf); + if (!r10_bio) { + unplug_slaves(conf->mddev); + return NULL; + } + + if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery)) + nalloc = conf->copies; /* resync */ + else + nalloc = 2; /* recovery */ + + /* + * Allocate bios. + */ + for (j = nalloc ; j-- ; ) { + bio = bio_alloc(gfp_flags, RESYNC_PAGES); + if (!bio) + goto out_free_bio; + r10_bio->devs[j].bio = bio; + } + /* + * Allocate RESYNC_PAGES data pages and attach them + * where needed. + */ + for (j = 0 ; j < nalloc; j++) { + bio = r10_bio->devs[j].bio; + for (i = 0; i < RESYNC_PAGES; i++) { + page = alloc_page(gfp_flags); + if (unlikely(!page)) + goto out_free_pages; + + bio->bi_io_vec[i].bv_page = page; + } + } + + return r10_bio; + +out_free_pages: + for ( ; i > 0 ; i--) + __free_page(bio->bi_io_vec[i-1].bv_page); + while (j--) + for (i = 0; i < RESYNC_PAGES ; i++) + __free_page(r10_bio->devs[j].bio->bi_io_vec[i].bv_page); + j = -1; +out_free_bio: + while ( ++j < nalloc ) + bio_put(r10_bio->devs[j].bio); + r10bio_pool_free(r10_bio, conf); + return NULL; +} + +static void r10buf_pool_free(void *__r10_bio, void *data) +{ + int i; + conf_t *conf = data; + r10bio_t *r10bio = __r10_bio; + int j; + + for (j=0; j < conf->copies; j++) { + struct bio *bio = r10bio->devs[j].bio; + if (bio) { + for (i = 0; i < RESYNC_PAGES; i++) { + __free_page(bio->bi_io_vec[i].bv_page); + bio->bi_io_vec[i].bv_page = NULL; + } + bio_put(bio); + } + } + r10bio_pool_free(r10bio, conf); +} + +static void put_all_bios(conf_t *conf, r10bio_t *r10_bio) +{ + int i; + + for (i = 0; i < conf->copies; i++) { + struct bio **bio = & r10_bio->devs[i].bio; + if (*bio) + bio_put(*bio); + *bio = NULL; + } +} + +static inline void free_r10bio(r10bio_t *r10_bio) +{ + unsigned long flags; + + conf_t *conf = mddev_to_conf(r10_bio->mddev); + + /* + * Wake up any possible resync thread that waits for the device + * to go idle. + */ + spin_lock_irqsave(&conf->resync_lock, flags); + if (!--conf->nr_pending) { + wake_up(&conf->wait_idle); + wake_up(&conf->wait_resume); + } + spin_unlock_irqrestore(&conf->resync_lock, flags); + + put_all_bios(conf, r10_bio); + mempool_free(r10_bio, conf->r10bio_pool); +} + +static inline void put_buf(r10bio_t *r10_bio) +{ + conf_t *conf = mddev_to_conf(r10_bio->mddev); + unsigned long flags; + + mempool_free(r10_bio, conf->r10buf_pool); + + spin_lock_irqsave(&conf->resync_lock, flags); + if (!conf->barrier) + BUG(); + --conf->barrier; + wake_up(&conf->wait_resume); + wake_up(&conf->wait_idle); + + if (!--conf->nr_pending) { + wake_up(&conf->wait_idle); + wake_up(&conf->wait_resume); + } + spin_unlock_irqrestore(&conf->resync_lock, flags); +} + +static void reschedule_retry(r10bio_t *r10_bio) +{ + unsigned long flags; + mddev_t *mddev = r10_bio->mddev; + conf_t *conf = mddev_to_conf(mddev); + + spin_lock_irqsave(&conf->device_lock, flags); + list_add(&r10_bio->retry_list, &conf->retry_list); + spin_unlock_irqrestore(&conf->device_lock, flags); + + md_wakeup_thread(mddev->thread); +} + +/* + * raid_end_bio_io() is called when we have finished servicing a mirrored + * operation and are ready to return a success/failure code to the buffer + * cache layer. + */ +static void raid_end_bio_io(r10bio_t *r10_bio) +{ + struct bio *bio = r10_bio->master_bio; + + bio_endio(bio, bio->bi_size, + test_bit(R10BIO_Uptodate, &r10_bio->state) ? 0 : -EIO); + free_r10bio(r10_bio); +} + +/* + * Update disk head position estimator based on IRQ completion info. + */ +static inline void update_head_pos(int slot, r10bio_t *r10_bio) +{ + conf_t *conf = mddev_to_conf(r10_bio->mddev); + + conf->mirrors[r10_bio->devs[slot].devnum].head_position = + r10_bio->devs[slot].addr + (r10_bio->sectors); +} + +static int raid10_end_read_request(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private); + int slot, dev; + conf_t *conf = mddev_to_conf(r10_bio->mddev); + + if (bio->bi_size) + return 1; + + slot = r10_bio->read_slot; + dev = r10_bio->devs[slot].devnum; + /* + * this branch is our 'one mirror IO has finished' event handler: + */ + if (!uptodate) + md_error(r10_bio->mddev, conf->mirrors[dev].rdev); + else + /* + * Set R10BIO_Uptodate in our master bio, so that + * we will return a good error code to the higher + * levels even if IO on some other mirrored buffer fails. + * + * The 'master' represents the composite IO operation to + * user-side. So if something waits for IO, then it will + * wait for the 'master' bio. + */ + set_bit(R10BIO_Uptodate, &r10_bio->state); + + update_head_pos(slot, r10_bio); + + /* + * we have only one bio on the read side + */ + if (uptodate) + raid_end_bio_io(r10_bio); + else { + /* + * oops, read error: + */ + char b[BDEVNAME_SIZE]; + if (printk_ratelimit()) + printk(KERN_ERR "raid10: %s: rescheduling sector %llu\n", + bdevname(conf->mirrors[dev].rdev->bdev,b), (unsigned long long)r10_bio->sector); + reschedule_retry(r10_bio); + } + + rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); + return 0; +} + +static int raid10_end_write_request(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private); + int slot, dev; + conf_t *conf = mddev_to_conf(r10_bio->mddev); + + if (bio->bi_size) + return 1; + + for (slot = 0; slot < conf->copies; slot++) + if (r10_bio->devs[slot].bio == bio) + break; + dev = r10_bio->devs[slot].devnum; + + /* + * this branch is our 'one mirror IO has finished' event handler: + */ + if (!uptodate) + md_error(r10_bio->mddev, conf->mirrors[dev].rdev); + else + /* + * Set R10BIO_Uptodate in our master bio, so that + * we will return a good error code for to the higher + * levels even if IO on some other mirrored buffer fails. + * + * The 'master' represents the composite IO operation to + * user-side. So if something waits for IO, then it will + * wait for the 'master' bio. + */ + set_bit(R10BIO_Uptodate, &r10_bio->state); + + update_head_pos(slot, r10_bio); + + /* + * + * Let's see if all mirrored write operations have finished + * already. + */ + if (atomic_dec_and_test(&r10_bio->remaining)) { + md_write_end(r10_bio->mddev); + raid_end_bio_io(r10_bio); + } + + rdev_dec_pending(conf->mirrors[dev].rdev, conf->mddev); + return 0; +} + + +/* + * RAID10 layout manager + * Aswell as the chunksize and raid_disks count, there are two + * parameters: near_copies and far_copies. + * near_copies * far_copies must be <= raid_disks. + * Normally one of these will be 1. + * If both are 1, we get raid0. + * If near_copies == raid_disks, we get raid1. + * + * Chunks are layed out in raid0 style with near_copies copies of the + * first chunk, followed by near_copies copies of the next chunk and + * so on. + * If far_copies > 1, then after 1/far_copies of the array has been assigned + * as described above, we start again with a device offset of near_copies. + * So we effectively have another copy of the whole array further down all + * the drives, but with blocks on different drives. + * With this layout, and block is never stored twice on the one device. + * + * raid10_find_phys finds the sector offset of a given virtual sector + * on each device that it is on. If a block isn't on a device, + * that entry in the array is set to MaxSector. + * + * raid10_find_virt does the reverse mapping, from a device and a + * sector offset to a virtual address + */ + +static void raid10_find_phys(conf_t *conf, r10bio_t *r10bio) +{ + int n,f; + sector_t sector; + sector_t chunk; + sector_t stripe; + int dev; + + int slot = 0; + + /* now calculate first sector/dev */ + chunk = r10bio->sector >> conf->chunk_shift; + sector = r10bio->sector & conf->chunk_mask; + + chunk *= conf->near_copies; + stripe = chunk; + dev = sector_div(stripe, conf->raid_disks); + + sector += stripe << conf->chunk_shift; + + /* and calculate all the others */ + for (n=0; n < conf->near_copies; n++) { + int d = dev; + sector_t s = sector; + r10bio->devs[slot].addr = sector; + r10bio->devs[slot].devnum = d; + slot++; + + for (f = 1; f < conf->far_copies; f++) { + d += conf->near_copies; + if (d >= conf->raid_disks) + d -= conf->raid_disks; + s += conf->stride; + r10bio->devs[slot].devnum = d; + r10bio->devs[slot].addr = s; + slot++; + } + dev++; + if (dev >= conf->raid_disks) { + dev = 0; + sector += (conf->chunk_mask + 1); + } + } + BUG_ON(slot != conf->copies); +} + +static sector_t raid10_find_virt(conf_t *conf, sector_t sector, int dev) +{ + sector_t offset, chunk, vchunk; + + while (sector > conf->stride) { + sector -= conf->stride; + if (dev < conf->near_copies) + dev += conf->raid_disks - conf->near_copies; + else + dev -= conf->near_copies; + } + + offset = sector & conf->chunk_mask; + chunk = sector >> conf->chunk_shift; + vchunk = chunk * conf->raid_disks + dev; + sector_div(vchunk, conf->near_copies); + return (vchunk << conf->chunk_shift) + offset; +} + +/** + * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged + * @q: request queue + * @bio: the buffer head that's been built up so far + * @biovec: the request that could be merged to it. + * + * Return amount of bytes we can accept at this offset + * If near_copies == raid_disk, there are no striping issues, + * but in that case, the function isn't called at all. + */ +static int raid10_mergeable_bvec(request_queue_t *q, struct bio *bio, + struct bio_vec *bio_vec) +{ + mddev_t *mddev = q->queuedata; + sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev); + int max; + unsigned int chunk_sectors = mddev->chunk_size >> 9; + unsigned int bio_sectors = bio->bi_size >> 9; + + max = (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9; + if (max < 0) max = 0; /* bio_add cannot handle a negative return */ + if (max <= bio_vec->bv_len && bio_sectors == 0) + return bio_vec->bv_len; + else + return max; +} + +/* + * This routine returns the disk from which the requested read should + * be done. There is a per-array 'next expected sequential IO' sector + * number - if this matches on the next IO then we use the last disk. + * There is also a per-disk 'last know head position' sector that is + * maintained from IRQ contexts, both the normal and the resync IO + * completion handlers update this position correctly. If there is no + * perfect sequential match then we pick the disk whose head is closest. + * + * If there are 2 mirrors in the same 2 devices, performance degrades + * because position is mirror, not device based. + * + * The rdev for the device selected will have nr_pending incremented. + */ + +/* + * FIXME: possibly should rethink readbalancing and do it differently + * depending on near_copies / far_copies geometry. + */ +static int read_balance(conf_t *conf, r10bio_t *r10_bio) +{ + const unsigned long this_sector = r10_bio->sector; + int disk, slot, nslot; + const int sectors = r10_bio->sectors; + sector_t new_distance, current_distance; + + raid10_find_phys(conf, r10_bio); + rcu_read_lock(); + /* + * Check if we can balance. We can balance on the whole + * device if no resync is going on, or below the resync window. + * We take the first readable disk when above the resync window. + */ + if (conf->mddev->recovery_cp < MaxSector + && (this_sector + sectors >= conf->next_resync)) { + /* make sure that disk is operational */ + slot = 0; + disk = r10_bio->devs[slot].devnum; + + while (!conf->mirrors[disk].rdev || + !conf->mirrors[disk].rdev->in_sync) { + slot++; + if (slot == conf->copies) { + slot = 0; + disk = -1; + break; + } + disk = r10_bio->devs[slot].devnum; + } + goto rb_out; + } + + + /* make sure the disk is operational */ + slot = 0; + disk = r10_bio->devs[slot].devnum; + while (!conf->mirrors[disk].rdev || + !conf->mirrors[disk].rdev->in_sync) { + slot ++; + if (slot == conf->copies) { + disk = -1; + goto rb_out; + } + disk = r10_bio->devs[slot].devnum; + } + + + current_distance = abs(this_sector - conf->mirrors[disk].head_position); + + /* Find the disk whose head is closest */ + + for (nslot = slot; nslot < conf->copies; nslot++) { + int ndisk = r10_bio->devs[nslot].devnum; + + + if (!conf->mirrors[ndisk].rdev || + !conf->mirrors[ndisk].rdev->in_sync) + continue; + + if (!atomic_read(&conf->mirrors[ndisk].rdev->nr_pending)) { + disk = ndisk; + slot = nslot; + break; + } + new_distance = abs(r10_bio->devs[nslot].addr - + conf->mirrors[ndisk].head_position); + if (new_distance < current_distance) { + current_distance = new_distance; + disk = ndisk; + slot = nslot; + } + } + +rb_out: + r10_bio->read_slot = slot; +/* conf->next_seq_sect = this_sector + sectors;*/ + + if (disk >= 0 && conf->mirrors[disk].rdev) + atomic_inc(&conf->mirrors[disk].rdev->nr_pending); + rcu_read_unlock(); + + return disk; +} + +static void unplug_slaves(mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + int i; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks; i++) { + mdk_rdev_t *rdev = conf->mirrors[i].rdev; + if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) { + request_queue_t *r_queue = bdev_get_queue(rdev->bdev); + + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + rcu_read_unlock(); +} + +static void raid10_unplug(request_queue_t *q) +{ + unplug_slaves(q->queuedata); +} + +static int raid10_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + mdk_rdev_t *rdev = conf->mirrors[i].rdev; + if (rdev && !rdev->faulty) { + struct block_device *bdev = rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, + error_sector); + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + } + rcu_read_unlock(); + return ret; +} + +/* + * Throttle resync depth, so that we can both get proper overlapping of + * requests, but are still able to handle normal requests quickly. + */ +#define RESYNC_DEPTH 32 + +static void device_barrier(conf_t *conf, sector_t sect) +{ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_idle, !waitqueue_active(&conf->wait_resume), + conf->resync_lock, unplug_slaves(conf->mddev)); + + if (!conf->barrier++) { + wait_event_lock_irq(conf->wait_idle, !conf->nr_pending, + conf->resync_lock, unplug_slaves(conf->mddev)); + if (conf->nr_pending) + BUG(); + } + wait_event_lock_irq(conf->wait_resume, conf->barrier < RESYNC_DEPTH, + conf->resync_lock, unplug_slaves(conf->mddev)); + conf->next_resync = sect; + spin_unlock_irq(&conf->resync_lock); +} + +static int make_request(request_queue_t *q, struct bio * bio) +{ + mddev_t *mddev = q->queuedata; + conf_t *conf = mddev_to_conf(mddev); + mirror_info_t *mirror; + r10bio_t *r10_bio; + struct bio *read_bio; + int i; + int chunk_sects = conf->chunk_mask + 1; + + /* If this request crosses a chunk boundary, we need to + * split it. This will only happen for 1 PAGE (or less) requests. + */ + if (unlikely( (bio->bi_sector & conf->chunk_mask) + (bio->bi_size >> 9) + > chunk_sects && + conf->near_copies < conf->raid_disks)) { + struct bio_pair *bp; + /* Sanity check -- queue functions should prevent this happening */ + if (bio->bi_vcnt != 1 || + bio->bi_idx != 0) + goto bad_map; + /* This is a one page bio that upper layers + * refuse to split for us, so we need to split it. + */ + bp = bio_split(bio, bio_split_pool, + chunk_sects - (bio->bi_sector & (chunk_sects - 1)) ); + if (make_request(q, &bp->bio1)) + generic_make_request(&bp->bio1); + if (make_request(q, &bp->bio2)) + generic_make_request(&bp->bio2); + + bio_pair_release(bp); + return 0; + bad_map: + printk("raid10_make_request bug: can't convert block across chunks" + " or bigger than %dk %llu %d\n", chunk_sects/2, + (unsigned long long)bio->bi_sector, bio->bi_size >> 10); + + bio_io_error(bio, bio->bi_size); + return 0; + } + + /* + * Register the new request and wait if the reconstruction + * thread has put up a bar for new requests. + * Continue immediately if no resync is active currently. + */ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_resume, !conf->barrier, conf->resync_lock, ); + conf->nr_pending++; + spin_unlock_irq(&conf->resync_lock); + + if (bio_data_dir(bio)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio)); + } + + r10_bio = mempool_alloc(conf->r10bio_pool, GFP_NOIO); + + r10_bio->master_bio = bio; + r10_bio->sectors = bio->bi_size >> 9; + + r10_bio->mddev = mddev; + r10_bio->sector = bio->bi_sector; + + if (bio_data_dir(bio) == READ) { + /* + * read balancing logic: + */ + int disk = read_balance(conf, r10_bio); + int slot = r10_bio->read_slot; + if (disk < 0) { + raid_end_bio_io(r10_bio); + return 0; + } + mirror = conf->mirrors + disk; + + read_bio = bio_clone(bio, GFP_NOIO); + + r10_bio->devs[slot].bio = read_bio; + + read_bio->bi_sector = r10_bio->devs[slot].addr + + mirror->rdev->data_offset; + read_bio->bi_bdev = mirror->rdev->bdev; + read_bio->bi_end_io = raid10_end_read_request; + read_bio->bi_rw = READ; + read_bio->bi_private = r10_bio; + + generic_make_request(read_bio); + return 0; + } + + /* + * WRITE: + */ + /* first select target devices under spinlock and + * inc refcount on their rdev. Record them by setting + * bios[x] to bio + */ + raid10_find_phys(conf, r10_bio); + rcu_read_lock(); + for (i = 0; i < conf->copies; i++) { + int d = r10_bio->devs[i].devnum; + if (conf->mirrors[d].rdev && + !conf->mirrors[d].rdev->faulty) { + atomic_inc(&conf->mirrors[d].rdev->nr_pending); + r10_bio->devs[i].bio = bio; + } else + r10_bio->devs[i].bio = NULL; + } + rcu_read_unlock(); + + atomic_set(&r10_bio->remaining, 1); + md_write_start(mddev); + for (i = 0; i < conf->copies; i++) { + struct bio *mbio; + int d = r10_bio->devs[i].devnum; + if (!r10_bio->devs[i].bio) + continue; + + mbio = bio_clone(bio, GFP_NOIO); + r10_bio->devs[i].bio = mbio; + + mbio->bi_sector = r10_bio->devs[i].addr+ + conf->mirrors[d].rdev->data_offset; + mbio->bi_bdev = conf->mirrors[d].rdev->bdev; + mbio->bi_end_io = raid10_end_write_request; + mbio->bi_rw = WRITE; + mbio->bi_private = r10_bio; + + atomic_inc(&r10_bio->remaining); + generic_make_request(mbio); + } + + if (atomic_dec_and_test(&r10_bio->remaining)) { + md_write_end(mddev); + raid_end_bio_io(r10_bio); + } + + return 0; +} + +static void status(struct seq_file *seq, mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + int i; + + if (conf->near_copies < conf->raid_disks) + seq_printf(seq, " %dK chunks", mddev->chunk_size/1024); + if (conf->near_copies > 1) + seq_printf(seq, " %d near-copies", conf->near_copies); + if (conf->far_copies > 1) + seq_printf(seq, " %d far-copies", conf->far_copies); + + seq_printf(seq, " [%d/%d] [", conf->raid_disks, + conf->working_disks); + for (i = 0; i < conf->raid_disks; i++) + seq_printf(seq, "%s", + conf->mirrors[i].rdev && + conf->mirrors[i].rdev->in_sync ? "U" : "_"); + seq_printf(seq, "]"); +} + +static void error(mddev_t *mddev, mdk_rdev_t *rdev) +{ + char b[BDEVNAME_SIZE]; + conf_t *conf = mddev_to_conf(mddev); + + /* + * If it is not operational, then we have already marked it as dead + * else if it is the last working disks, ignore the error, let the + * next level up know. + * else mark the drive as failed + */ + if (rdev->in_sync + && conf->working_disks == 1) + /* + * Don't fail the drive, just return an IO error. + * The test should really be more sophisticated than + * "working_disks == 1", but it isn't critical, and + * can wait until we do more sophisticated "is the drive + * really dead" tests... + */ + return; + if (rdev->in_sync) { + mddev->degraded++; + conf->working_disks--; + /* + * if recovery is running, make sure it aborts. + */ + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + } + rdev->in_sync = 0; + rdev->faulty = 1; + mddev->sb_dirty = 1; + printk(KERN_ALERT "raid10: Disk failure on %s, disabling device. \n" + " Operation continuing on %d devices\n", + bdevname(rdev->bdev,b), conf->working_disks); +} + +static void print_conf(conf_t *conf) +{ + int i; + mirror_info_t *tmp; + + printk("RAID10 conf printout:\n"); + if (!conf) { + printk("(!conf)\n"); + return; + } + printk(" --- wd:%d rd:%d\n", conf->working_disks, + conf->raid_disks); + + for (i = 0; i < conf->raid_disks; i++) { + char b[BDEVNAME_SIZE]; + tmp = conf->mirrors + i; + if (tmp->rdev) + printk(" disk %d, wo:%d, o:%d, dev:%s\n", + i, !tmp->rdev->in_sync, !tmp->rdev->faulty, + bdevname(tmp->rdev->bdev,b)); + } +} + +static void close_sync(conf_t *conf) +{ + spin_lock_irq(&conf->resync_lock); + wait_event_lock_irq(conf->wait_resume, !conf->barrier, + conf->resync_lock, unplug_slaves(conf->mddev)); + spin_unlock_irq(&conf->resync_lock); + + if (conf->barrier) BUG(); + if (waitqueue_active(&conf->wait_idle)) BUG(); + + mempool_destroy(conf->r10buf_pool); + conf->r10buf_pool = NULL; +} + +static int raid10_spare_active(mddev_t *mddev) +{ + int i; + conf_t *conf = mddev->private; + mirror_info_t *tmp; + + /* + * Find all non-in_sync disks within the RAID10 configuration + * and mark them in_sync + */ + for (i = 0; i < conf->raid_disks; i++) { + tmp = conf->mirrors + i; + if (tmp->rdev + && !tmp->rdev->faulty + && !tmp->rdev->in_sync) { + conf->working_disks++; + mddev->degraded--; + tmp->rdev->in_sync = 1; + } + } + + print_conf(conf); + return 0; +} + + +static int raid10_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) +{ + conf_t *conf = mddev->private; + int found = 0; + int mirror; + mirror_info_t *p; + + if (mddev->recovery_cp < MaxSector) + /* only hot-add to in-sync arrays, as recovery is + * very different from resync + */ + return 0; + + for (mirror=0; mirror < mddev->raid_disks; mirror++) + if ( !(p=conf->mirrors+mirror)->rdev) { + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + mddev->queue->max_sectors = (PAGE_SIZE>>9); + + p->head_position = 0; + rdev->raid_disk = mirror; + found = 1; + p->rdev = rdev; + break; + } + + print_conf(conf); + return found; +} + +static int raid10_remove_disk(mddev_t *mddev, int number) +{ + conf_t *conf = mddev->private; + int err = 0; + mdk_rdev_t *rdev; + mirror_info_t *p = conf->mirrors+ number; + + print_conf(conf); + rdev = p->rdev; + if (rdev) { + if (rdev->in_sync || + atomic_read(&rdev->nr_pending)) { + err = -EBUSY; + goto abort; + } + p->rdev = NULL; + synchronize_kernel(); + if (atomic_read(&rdev->nr_pending)) { + /* lost the race, try later */ + err = -EBUSY; + p->rdev = rdev; + } + } +abort: + + print_conf(conf); + return err; +} + + +static int end_sync_read(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private); + conf_t *conf = mddev_to_conf(r10_bio->mddev); + int i,d; + + if (bio->bi_size) + return 1; + + for (i=0; i<conf->copies; i++) + if (r10_bio->devs[i].bio == bio) + break; + if (i == conf->copies) + BUG(); + update_head_pos(i, r10_bio); + d = r10_bio->devs[i].devnum; + if (!uptodate) + md_error(r10_bio->mddev, + conf->mirrors[d].rdev); + + /* for reconstruct, we always reschedule after a read. + * for resync, only after all reads + */ + if (test_bit(R10BIO_IsRecover, &r10_bio->state) || + atomic_dec_and_test(&r10_bio->remaining)) { + /* we have read all the blocks, + * do the comparison in process context in raid10d + */ + reschedule_retry(r10_bio); + } + rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); + return 0; +} + +static int end_sync_write(struct bio *bio, unsigned int bytes_done, int error) +{ + int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); + r10bio_t * r10_bio = (r10bio_t *)(bio->bi_private); + mddev_t *mddev = r10_bio->mddev; + conf_t *conf = mddev_to_conf(mddev); + int i,d; + + if (bio->bi_size) + return 1; + + for (i = 0; i < conf->copies; i++) + if (r10_bio->devs[i].bio == bio) + break; + d = r10_bio->devs[i].devnum; + + if (!uptodate) + md_error(mddev, conf->mirrors[d].rdev); + update_head_pos(i, r10_bio); + + while (atomic_dec_and_test(&r10_bio->remaining)) { + if (r10_bio->master_bio == NULL) { + /* the primary of several recovery bios */ + md_done_sync(mddev, r10_bio->sectors, 1); + put_buf(r10_bio); + break; + } else { + r10bio_t *r10_bio2 = (r10bio_t *)r10_bio->master_bio; + put_buf(r10_bio); + r10_bio = r10_bio2; + } + } + rdev_dec_pending(conf->mirrors[d].rdev, mddev); + return 0; +} + +/* + * Note: sync and recover and handled very differently for raid10 + * This code is for resync. + * For resync, we read through virtual addresses and read all blocks. + * If there is any error, we schedule a write. The lowest numbered + * drive is authoritative. + * However requests come for physical address, so we need to map. + * For every physical address there are raid_disks/copies virtual addresses, + * which is always are least one, but is not necessarly an integer. + * This means that a physical address can span multiple chunks, so we may + * have to submit multiple io requests for a single sync request. + */ +/* + * We check if all blocks are in-sync and only write to blocks that + * aren't in sync + */ +static void sync_request_write(mddev_t *mddev, r10bio_t *r10_bio) +{ + conf_t *conf = mddev_to_conf(mddev); + int i, first; + struct bio *tbio, *fbio; + + atomic_set(&r10_bio->remaining, 1); + + /* find the first device with a block */ + for (i=0; i<conf->copies; i++) + if (test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) + break; + + if (i == conf->copies) + goto done; + + first = i; + fbio = r10_bio->devs[i].bio; + + /* now find blocks with errors */ + for (i=first+1 ; i < conf->copies ; i++) { + int vcnt, j, d; + + if (!test_bit(BIO_UPTODATE, &r10_bio->devs[i].bio->bi_flags)) + continue; + /* We know that the bi_io_vec layout is the same for + * both 'first' and 'i', so we just compare them. + * All vec entries are PAGE_SIZE; + */ + tbio = r10_bio->devs[i].bio; + vcnt = r10_bio->sectors >> (PAGE_SHIFT-9); + for (j = 0; j < vcnt; j++) + if (memcmp(page_address(fbio->bi_io_vec[j].bv_page), + page_address(tbio->bi_io_vec[j].bv_page), + PAGE_SIZE)) + break; + if (j == vcnt) + continue; + /* Ok, we need to write this bio + * First we need to fixup bv_offset, bv_len and + * bi_vecs, as the read request might have corrupted these + */ + tbio->bi_vcnt = vcnt; + tbio->bi_size = r10_bio->sectors << 9; + tbio->bi_idx = 0; + tbio->bi_phys_segments = 0; + tbio->bi_hw_segments = 0; + tbio->bi_hw_front_size = 0; + tbio->bi_hw_back_size = 0; + tbio->bi_flags &= ~(BIO_POOL_MASK - 1); + tbio->bi_flags |= 1 << BIO_UPTODATE; + tbio->bi_next = NULL; + tbio->bi_rw = WRITE; + tbio->bi_private = r10_bio; + tbio->bi_sector = r10_bio->devs[i].addr; + + for (j=0; j < vcnt ; j++) { + tbio->bi_io_vec[j].bv_offset = 0; + tbio->bi_io_vec[j].bv_len = PAGE_SIZE; + + memcpy(page_address(tbio->bi_io_vec[j].bv_page), + page_address(fbio->bi_io_vec[j].bv_page), + PAGE_SIZE); + } + tbio->bi_end_io = end_sync_write; + + d = r10_bio->devs[i].devnum; + atomic_inc(&conf->mirrors[d].rdev->nr_pending); + atomic_inc(&r10_bio->remaining); + md_sync_acct(conf->mirrors[d].rdev->bdev, tbio->bi_size >> 9); + + tbio->bi_sector += conf->mirrors[d].rdev->data_offset; + tbio->bi_bdev = conf->mirrors[d].rdev->bdev; + generic_make_request(tbio); + } + +done: + if (atomic_dec_and_test(&r10_bio->remaining)) { + md_done_sync(mddev, r10_bio->sectors, 1); + put_buf(r10_bio); + } +} + +/* + * Now for the recovery code. + * Recovery happens across physical sectors. + * We recover all non-is_sync drives by finding the virtual address of + * each, and then choose a working drive that also has that virt address. + * There is a separate r10_bio for each non-in_sync drive. + * Only the first two slots are in use. The first for reading, + * The second for writing. + * + */ + +static void recovery_request_write(mddev_t *mddev, r10bio_t *r10_bio) +{ + conf_t *conf = mddev_to_conf(mddev); + int i, d; + struct bio *bio, *wbio; + + + /* move the pages across to the second bio + * and submit the write request + */ + bio = r10_bio->devs[0].bio; + wbio = r10_bio->devs[1].bio; + for (i=0; i < wbio->bi_vcnt; i++) { + struct page *p = bio->bi_io_vec[i].bv_page; + bio->bi_io_vec[i].bv_page = wbio->bi_io_vec[i].bv_page; + wbio->bi_io_vec[i].bv_page = p; + } + d = r10_bio->devs[1].devnum; + + atomic_inc(&conf->mirrors[d].rdev->nr_pending); + md_sync_acct(conf->mirrors[d].rdev->bdev, wbio->bi_size >> 9); + generic_make_request(wbio); +} + + +/* + * This is a kernel thread which: + * + * 1. Retries failed read operations on working mirrors. + * 2. Updates the raid superblock when problems encounter. + * 3. Performs writes following reads for array syncronising. + */ + +static void raid10d(mddev_t *mddev) +{ + r10bio_t *r10_bio; + struct bio *bio; + unsigned long flags; + conf_t *conf = mddev_to_conf(mddev); + struct list_head *head = &conf->retry_list; + int unplug=0; + mdk_rdev_t *rdev; + + md_check_recovery(mddev); + md_handle_safemode(mddev); + + for (;;) { + char b[BDEVNAME_SIZE]; + spin_lock_irqsave(&conf->device_lock, flags); + if (list_empty(head)) + break; + r10_bio = list_entry(head->prev, r10bio_t, retry_list); + list_del(head->prev); + spin_unlock_irqrestore(&conf->device_lock, flags); + + mddev = r10_bio->mddev; + conf = mddev_to_conf(mddev); + if (test_bit(R10BIO_IsSync, &r10_bio->state)) { + sync_request_write(mddev, r10_bio); + unplug = 1; + } else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) { + recovery_request_write(mddev, r10_bio); + unplug = 1; + } else { + int mirror; + bio = r10_bio->devs[r10_bio->read_slot].bio; + r10_bio->devs[r10_bio->read_slot].bio = NULL; + bio_put(bio); + mirror = read_balance(conf, r10_bio); + if (mirror == -1) { + printk(KERN_ALERT "raid10: %s: unrecoverable I/O" + " read error for block %llu\n", + bdevname(bio->bi_bdev,b), + (unsigned long long)r10_bio->sector); + raid_end_bio_io(r10_bio); + } else { + rdev = conf->mirrors[mirror].rdev; + if (printk_ratelimit()) + printk(KERN_ERR "raid10: %s: redirecting sector %llu to" + " another mirror\n", + bdevname(rdev->bdev,b), + (unsigned long long)r10_bio->sector); + bio = bio_clone(r10_bio->master_bio, GFP_NOIO); + r10_bio->devs[r10_bio->read_slot].bio = bio; + bio->bi_sector = r10_bio->devs[r10_bio->read_slot].addr + + rdev->data_offset; + bio->bi_bdev = rdev->bdev; + bio->bi_rw = READ; + bio->bi_private = r10_bio; + bio->bi_end_io = raid10_end_read_request; + unplug = 1; + generic_make_request(bio); + } + } + } + spin_unlock_irqrestore(&conf->device_lock, flags); + if (unplug) + unplug_slaves(mddev); +} + + +static int init_resync(conf_t *conf) +{ + int buffs; + + buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE; + if (conf->r10buf_pool) + BUG(); + conf->r10buf_pool = mempool_create(buffs, r10buf_pool_alloc, r10buf_pool_free, conf); + if (!conf->r10buf_pool) + return -ENOMEM; + conf->next_resync = 0; + return 0; +} + +/* + * perform a "sync" on one "block" + * + * We need to make sure that no normal I/O request - particularly write + * requests - conflict with active sync requests. + * + * This is achieved by tracking pending requests and a 'barrier' concept + * that can be installed to exclude normal IO requests. + * + * Resync and recovery are handled very differently. + * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery. + * + * For resync, we iterate over virtual addresses, read all copies, + * and update if there are differences. If only one copy is live, + * skip it. + * For recovery, we iterate over physical addresses, read a good + * value for each non-in_sync drive, and over-write. + * + * So, for recovery we may have several outstanding complex requests for a + * given address, one for each out-of-sync device. We model this by allocating + * a number of r10_bio structures, one for each out-of-sync device. + * As we setup these structures, we collect all bio's together into a list + * which we then process collectively to add pages, and then process again + * to pass to generic_make_request. + * + * The r10_bio structures are linked using a borrowed master_bio pointer. + * This link is counted in ->remaining. When the r10_bio that points to NULL + * has its remaining count decremented to 0, the whole complex operation + * is complete. + * + */ + +static int sync_request(mddev_t *mddev, sector_t sector_nr, int go_faster) +{ + conf_t *conf = mddev_to_conf(mddev); + r10bio_t *r10_bio; + struct bio *biolist = NULL, *bio; + sector_t max_sector, nr_sectors; + int disk; + int i; + + sector_t sectors_skipped = 0; + int chunks_skipped = 0; + + if (!conf->r10buf_pool) + if (init_resync(conf)) + return -ENOMEM; + + skipped: + max_sector = mddev->size << 1; + if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) + max_sector = mddev->resync_max_sectors; + if (sector_nr >= max_sector) { + close_sync(conf); + return sectors_skipped; + } + if (chunks_skipped >= conf->raid_disks) { + /* if there has been nothing to do on any drive, + * then there is nothing to do at all.. + */ + sector_t sec = max_sector - sector_nr; + md_done_sync(mddev, sec, 1); + return sec + sectors_skipped; + } + + /* make sure whole request will fit in a chunk - if chunks + * are meaningful + */ + if (conf->near_copies < conf->raid_disks && + max_sector > (sector_nr | conf->chunk_mask)) + max_sector = (sector_nr | conf->chunk_mask) + 1; + /* + * If there is non-resync activity waiting for us then + * put in a delay to throttle resync. + */ + if (!go_faster && waitqueue_active(&conf->wait_resume)) + msleep_interruptible(1000); + device_barrier(conf, sector_nr + RESYNC_SECTORS); + + /* Again, very different code for resync and recovery. + * Both must result in an r10bio with a list of bios that + * have bi_end_io, bi_sector, bi_bdev set, + * and bi_private set to the r10bio. + * For recovery, we may actually create several r10bios + * with 2 bios in each, that correspond to the bios in the main one. + * In this case, the subordinate r10bios link back through a + * borrowed master_bio pointer, and the counter in the master + * includes a ref from each subordinate. + */ + /* First, we decide what to do and set ->bi_end_io + * To end_sync_read if we want to read, and + * end_sync_write if we will want to write. + */ + + if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { + /* recovery... the complicated one */ + int i, j, k; + r10_bio = NULL; + + for (i=0 ; i<conf->raid_disks; i++) + if (conf->mirrors[i].rdev && + !conf->mirrors[i].rdev->in_sync) { + /* want to reconstruct this device */ + r10bio_t *rb2 = r10_bio; + + r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); + spin_lock_irq(&conf->resync_lock); + conf->nr_pending++; + if (rb2) conf->barrier++; + spin_unlock_irq(&conf->resync_lock); + atomic_set(&r10_bio->remaining, 0); + + r10_bio->master_bio = (struct bio*)rb2; + if (rb2) + atomic_inc(&rb2->remaining); + r10_bio->mddev = mddev; + set_bit(R10BIO_IsRecover, &r10_bio->state); + r10_bio->sector = raid10_find_virt(conf, sector_nr, i); + raid10_find_phys(conf, r10_bio); + for (j=0; j<conf->copies;j++) { + int d = r10_bio->devs[j].devnum; + if (conf->mirrors[d].rdev && + conf->mirrors[d].rdev->in_sync) { + /* This is where we read from */ + bio = r10_bio->devs[0].bio; + bio->bi_next = biolist; + biolist = bio; + bio->bi_private = r10_bio; + bio->bi_end_io = end_sync_read; + bio->bi_rw = 0; + bio->bi_sector = r10_bio->devs[j].addr + + conf->mirrors[d].rdev->data_offset; + bio->bi_bdev = conf->mirrors[d].rdev->bdev; + atomic_inc(&conf->mirrors[d].rdev->nr_pending); + atomic_inc(&r10_bio->remaining); + /* and we write to 'i' */ + + for (k=0; k<conf->copies; k++) + if (r10_bio->devs[k].devnum == i) + break; + bio = r10_bio->devs[1].bio; + bio->bi_next = biolist; + biolist = bio; + bio->bi_private = r10_bio; + bio->bi_end_io = end_sync_write; + bio->bi_rw = 1; + bio->bi_sector = r10_bio->devs[k].addr + + conf->mirrors[i].rdev->data_offset; + bio->bi_bdev = conf->mirrors[i].rdev->bdev; + + r10_bio->devs[0].devnum = d; + r10_bio->devs[1].devnum = i; + + break; + } + } + if (j == conf->copies) { + BUG(); + } + } + if (biolist == NULL) { + while (r10_bio) { + r10bio_t *rb2 = r10_bio; + r10_bio = (r10bio_t*) rb2->master_bio; + rb2->master_bio = NULL; + put_buf(rb2); + } + goto giveup; + } + } else { + /* resync. Schedule a read for every block at this virt offset */ + int count = 0; + r10_bio = mempool_alloc(conf->r10buf_pool, GFP_NOIO); + + spin_lock_irq(&conf->resync_lock); + conf->nr_pending++; + spin_unlock_irq(&conf->resync_lock); + + r10_bio->mddev = mddev; + atomic_set(&r10_bio->remaining, 0); + + r10_bio->master_bio = NULL; + r10_bio->sector = sector_nr; + set_bit(R10BIO_IsSync, &r10_bio->state); + raid10_find_phys(conf, r10_bio); + r10_bio->sectors = (sector_nr | conf->chunk_mask) - sector_nr +1; + + for (i=0; i<conf->copies; i++) { + int d = r10_bio->devs[i].devnum; + bio = r10_bio->devs[i].bio; + bio->bi_end_io = NULL; + if (conf->mirrors[d].rdev == NULL || + conf->mirrors[d].rdev->faulty) + continue; + atomic_inc(&conf->mirrors[d].rdev->nr_pending); + atomic_inc(&r10_bio->remaining); + bio->bi_next = biolist; + biolist = bio; + bio->bi_private = r10_bio; + bio->bi_end_io = end_sync_read; + bio->bi_rw = 0; + bio->bi_sector = r10_bio->devs[i].addr + + conf->mirrors[d].rdev->data_offset; + bio->bi_bdev = conf->mirrors[d].rdev->bdev; + count++; + } + + if (count < 2) { + for (i=0; i<conf->copies; i++) { + int d = r10_bio->devs[i].devnum; + if (r10_bio->devs[i].bio->bi_end_io) + rdev_dec_pending(conf->mirrors[d].rdev, mddev); + } + put_buf(r10_bio); + biolist = NULL; + goto giveup; + } + } + + for (bio = biolist; bio ; bio=bio->bi_next) { + + bio->bi_flags &= ~(BIO_POOL_MASK - 1); + if (bio->bi_end_io) + bio->bi_flags |= 1 << BIO_UPTODATE; + bio->bi_vcnt = 0; + bio->bi_idx = 0; + bio->bi_phys_segments = 0; + bio->bi_hw_segments = 0; + bio->bi_size = 0; + } + + nr_sectors = 0; + do { + struct page *page; + int len = PAGE_SIZE; + disk = 0; + if (sector_nr + (len>>9) > max_sector) + len = (max_sector - sector_nr) << 9; + if (len == 0) + break; + for (bio= biolist ; bio ; bio=bio->bi_next) { + page = bio->bi_io_vec[bio->bi_vcnt].bv_page; + if (bio_add_page(bio, page, len, 0) == 0) { + /* stop here */ + struct bio *bio2; + bio->bi_io_vec[bio->bi_vcnt].bv_page = page; + for (bio2 = biolist; bio2 && bio2 != bio; bio2 = bio2->bi_next) { + /* remove last page from this bio */ + bio2->bi_vcnt--; + bio2->bi_size -= len; + bio2->bi_flags &= ~(1<< BIO_SEG_VALID); + } + goto bio_full; + } + disk = i; + } + nr_sectors += len>>9; + sector_nr += len>>9; + } while (biolist->bi_vcnt < RESYNC_PAGES); + bio_full: + r10_bio->sectors = nr_sectors; + + while (biolist) { + bio = biolist; + biolist = biolist->bi_next; + + bio->bi_next = NULL; + r10_bio = bio->bi_private; + r10_bio->sectors = nr_sectors; + + if (bio->bi_end_io == end_sync_read) { + md_sync_acct(bio->bi_bdev, nr_sectors); + generic_make_request(bio); + } + } + + return sectors_skipped + nr_sectors; + giveup: + /* There is nowhere to write, so all non-sync + * drives must be failed, so try the next chunk... + */ + { + int sec = max_sector - sector_nr; + sectors_skipped += sec; + chunks_skipped ++; + sector_nr = max_sector; + md_done_sync(mddev, sec, 1); + goto skipped; + } +} + +static int run(mddev_t *mddev) +{ + conf_t *conf; + int i, disk_idx; + mirror_info_t *disk; + mdk_rdev_t *rdev; + struct list_head *tmp; + int nc, fc; + sector_t stride, size; + + if (mddev->level != 10) { + printk(KERN_ERR "raid10: %s: raid level not set correctly... (%d)\n", + mdname(mddev), mddev->level); + goto out; + } + nc = mddev->layout & 255; + fc = (mddev->layout >> 8) & 255; + if ((nc*fc) <2 || (nc*fc) > mddev->raid_disks || + (mddev->layout >> 16)) { + printk(KERN_ERR "raid10: %s: unsupported raid10 layout: 0x%8x\n", + mdname(mddev), mddev->layout); + goto out; + } + /* + * copy the already verified devices into our private RAID10 + * bookkeeping area. [whatever we allocate in run(), + * should be freed in stop()] + */ + conf = kmalloc(sizeof(conf_t), GFP_KERNEL); + mddev->private = conf; + if (!conf) { + printk(KERN_ERR "raid10: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out; + } + memset(conf, 0, sizeof(*conf)); + conf->mirrors = kmalloc(sizeof(struct mirror_info)*mddev->raid_disks, + GFP_KERNEL); + if (!conf->mirrors) { + printk(KERN_ERR "raid10: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out_free_conf; + } + memset(conf->mirrors, 0, sizeof(struct mirror_info)*mddev->raid_disks); + + conf->near_copies = nc; + conf->far_copies = fc; + conf->copies = nc*fc; + conf->chunk_mask = (sector_t)(mddev->chunk_size>>9)-1; + conf->chunk_shift = ffz(~mddev->chunk_size) - 9; + stride = mddev->size >> (conf->chunk_shift-1); + sector_div(stride, fc); + conf->stride = stride << conf->chunk_shift; + + conf->r10bio_pool = mempool_create(NR_RAID10_BIOS, r10bio_pool_alloc, + r10bio_pool_free, conf); + if (!conf->r10bio_pool) { + printk(KERN_ERR "raid10: couldn't allocate memory for %s\n", + mdname(mddev)); + goto out_free_conf; + } + mddev->queue->unplug_fn = raid10_unplug; + + mddev->queue->issue_flush_fn = raid10_issue_flush; + + ITERATE_RDEV(mddev, rdev, tmp) { + disk_idx = rdev->raid_disk; + if (disk_idx >= mddev->raid_disks + || disk_idx < 0) + continue; + disk = conf->mirrors + disk_idx; + + disk->rdev = rdev; + + blk_queue_stack_limits(mddev->queue, + rdev->bdev->bd_disk->queue); + /* as we don't honour merge_bvec_fn, we must never risk + * violating it, so limit ->max_sector to one PAGE, as + * a one page request is never in violation. + */ + if (rdev->bdev->bd_disk->queue->merge_bvec_fn && + mddev->queue->max_sectors > (PAGE_SIZE>>9)) + mddev->queue->max_sectors = (PAGE_SIZE>>9); + + disk->head_position = 0; + if (!rdev->faulty && rdev->in_sync) + conf->working_disks++; + } + conf->raid_disks = mddev->raid_disks; + conf->mddev = mddev; + spin_lock_init(&conf->device_lock); + INIT_LIST_HEAD(&conf->retry_list); + + spin_lock_init(&conf->resync_lock); + init_waitqueue_head(&conf->wait_idle); + init_waitqueue_head(&conf->wait_resume); + + if (!conf->working_disks) { + printk(KERN_ERR "raid10: no operational mirrors for %s\n", + mdname(mddev)); + goto out_free_conf; + } + + mddev->degraded = 0; + for (i = 0; i < conf->raid_disks; i++) { + + disk = conf->mirrors + i; + + if (!disk->rdev) { + disk->head_position = 0; + mddev->degraded++; + } + } + + + mddev->thread = md_register_thread(raid10d, mddev, "%s_raid10"); + if (!mddev->thread) { + printk(KERN_ERR + "raid10: couldn't allocate thread for %s\n", + mdname(mddev)); + goto out_free_conf; + } + + printk(KERN_INFO + "raid10: raid set %s active with %d out of %d devices\n", + mdname(mddev), mddev->raid_disks - mddev->degraded, + mddev->raid_disks); + /* + * Ok, everything is just fine now + */ + size = conf->stride * conf->raid_disks; + sector_div(size, conf->near_copies); + mddev->array_size = size/2; + mddev->resync_max_sectors = size; + + /* Calculate max read-ahead size. + * We need to readahead at least twice a whole stripe.... + * maybe... + */ + { + int stripe = conf->raid_disks * mddev->chunk_size / PAGE_CACHE_SIZE; + stripe /= conf->near_copies; + if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) + mddev->queue->backing_dev_info.ra_pages = 2* stripe; + } + + if (conf->near_copies < mddev->raid_disks) + blk_queue_merge_bvec(mddev->queue, raid10_mergeable_bvec); + return 0; + +out_free_conf: + if (conf->r10bio_pool) + mempool_destroy(conf->r10bio_pool); + if (conf->mirrors) + kfree(conf->mirrors); + kfree(conf); + mddev->private = NULL; +out: + return -EIO; +} + +static int stop(mddev_t *mddev) +{ + conf_t *conf = mddev_to_conf(mddev); + + md_unregister_thread(mddev->thread); + mddev->thread = NULL; + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + if (conf->r10bio_pool) + mempool_destroy(conf->r10bio_pool); + if (conf->mirrors) + kfree(conf->mirrors); + kfree(conf); + mddev->private = NULL; + return 0; +} + + +static mdk_personality_t raid10_personality = +{ + .name = "raid10", + .owner = THIS_MODULE, + .make_request = make_request, + .run = run, + .stop = stop, + .status = status, + .error_handler = error, + .hot_add_disk = raid10_add_disk, + .hot_remove_disk= raid10_remove_disk, + .spare_active = raid10_spare_active, + .sync_request = sync_request, +}; + +static int __init raid_init(void) +{ + return register_md_personality(RAID10, &raid10_personality); +} + +static void raid_exit(void) +{ + unregister_md_personality(RAID10); +} + +module_init(raid_init); +module_exit(raid_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-9"); /* RAID10 */ diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c new file mode 100644 index 00000000000..52c3a81c4aa --- /dev/null +++ b/drivers/md/raid5.c @@ -0,0 +1,1965 @@ +/* + * raid5.c : Multiple Devices driver for Linux + * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman + * Copyright (C) 1999, 2000 Ingo Molnar + * + * RAID-5 management functions. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + + +#include <linux/config.h> +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/raid/raid5.h> +#include <linux/highmem.h> +#include <linux/bitops.h> +#include <asm/atomic.h> + +/* + * Stripe cache + */ + +#define NR_STRIPES 256 +#define STRIPE_SIZE PAGE_SIZE +#define STRIPE_SHIFT (PAGE_SHIFT - 9) +#define STRIPE_SECTORS (STRIPE_SIZE>>9) +#define IO_THRESHOLD 1 +#define HASH_PAGES 1 +#define HASH_PAGES_ORDER 0 +#define NR_HASH (HASH_PAGES * PAGE_SIZE / sizeof(struct stripe_head *)) +#define HASH_MASK (NR_HASH - 1) + +#define stripe_hash(conf, sect) ((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]) + +/* bio's attached to a stripe+device for I/O are linked together in bi_sector + * order without overlap. There may be several bio's per stripe+device, and + * a bio could span several devices. + * When walking this list for a particular stripe+device, we must never proceed + * beyond a bio that extends past this device, as the next bio might no longer + * be valid. + * This macro is used to determine the 'next' bio in the list, given the sector + * of the current stripe+device + */ +#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) +/* + * The following can be used to debug the driver + */ +#define RAID5_DEBUG 0 +#define RAID5_PARANOIA 1 +#if RAID5_PARANOIA && defined(CONFIG_SMP) +# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) +#else +# define CHECK_DEVLOCK() +#endif + +#define PRINTK(x...) ((void)(RAID5_DEBUG && printk(x))) +#if RAID5_DEBUG +#define inline +#define __inline__ +#endif + +static void print_raid5_conf (raid5_conf_t *conf); + +static inline void __release_stripe(raid5_conf_t *conf, struct stripe_head *sh) +{ + if (atomic_dec_and_test(&sh->count)) { + if (!list_empty(&sh->lru)) + BUG(); + if (atomic_read(&conf->active_stripes)==0) + BUG(); + if (test_bit(STRIPE_HANDLE, &sh->state)) { + if (test_bit(STRIPE_DELAYED, &sh->state)) + list_add_tail(&sh->lru, &conf->delayed_list); + else + list_add_tail(&sh->lru, &conf->handle_list); + md_wakeup_thread(conf->mddev->thread); + } else { + if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { + atomic_dec(&conf->preread_active_stripes); + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) + md_wakeup_thread(conf->mddev->thread); + } + list_add_tail(&sh->lru, &conf->inactive_list); + atomic_dec(&conf->active_stripes); + if (!conf->inactive_blocked || + atomic_read(&conf->active_stripes) < (NR_STRIPES*3/4)) + wake_up(&conf->wait_for_stripe); + } + } +} +static void release_stripe(struct stripe_head *sh) +{ + raid5_conf_t *conf = sh->raid_conf; + unsigned long flags; + + spin_lock_irqsave(&conf->device_lock, flags); + __release_stripe(conf, sh); + spin_unlock_irqrestore(&conf->device_lock, flags); +} + +static void remove_hash(struct stripe_head *sh) +{ + PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh->sector); + + if (sh->hash_pprev) { + if (sh->hash_next) + sh->hash_next->hash_pprev = sh->hash_pprev; + *sh->hash_pprev = sh->hash_next; + sh->hash_pprev = NULL; + } +} + +static __inline__ void insert_hash(raid5_conf_t *conf, struct stripe_head *sh) +{ + struct stripe_head **shp = &stripe_hash(conf, sh->sector); + + PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh->sector); + + CHECK_DEVLOCK(); + if ((sh->hash_next = *shp) != NULL) + (*shp)->hash_pprev = &sh->hash_next; + *shp = sh; + sh->hash_pprev = shp; +} + + +/* find an idle stripe, make sure it is unhashed, and return it. */ +static struct stripe_head *get_free_stripe(raid5_conf_t *conf) +{ + struct stripe_head *sh = NULL; + struct list_head *first; + + CHECK_DEVLOCK(); + if (list_empty(&conf->inactive_list)) + goto out; + first = conf->inactive_list.next; + sh = list_entry(first, struct stripe_head, lru); + list_del_init(first); + remove_hash(sh); + atomic_inc(&conf->active_stripes); +out: + return sh; +} + +static void shrink_buffers(struct stripe_head *sh, int num) +{ + struct page *p; + int i; + + for (i=0; i<num ; i++) { + p = sh->dev[i].page; + if (!p) + continue; + sh->dev[i].page = NULL; + page_cache_release(p); + } +} + +static int grow_buffers(struct stripe_head *sh, int num) +{ + int i; + + for (i=0; i<num; i++) { + struct page *page; + + if (!(page = alloc_page(GFP_KERNEL))) { + return 1; + } + sh->dev[i].page = page; + } + return 0; +} + +static void raid5_build_block (struct stripe_head *sh, int i); + +static inline void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx) +{ + raid5_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + + if (atomic_read(&sh->count) != 0) + BUG(); + if (test_bit(STRIPE_HANDLE, &sh->state)) + BUG(); + + CHECK_DEVLOCK(); + PRINTK("init_stripe called, stripe %llu\n", + (unsigned long long)sh->sector); + + remove_hash(sh); + + sh->sector = sector; + sh->pd_idx = pd_idx; + sh->state = 0; + + for (i=disks; i--; ) { + struct r5dev *dev = &sh->dev[i]; + + if (dev->toread || dev->towrite || dev->written || + test_bit(R5_LOCKED, &dev->flags)) { + printk("sector=%llx i=%d %p %p %p %d\n", + (unsigned long long)sh->sector, i, dev->toread, + dev->towrite, dev->written, + test_bit(R5_LOCKED, &dev->flags)); + BUG(); + } + dev->flags = 0; + raid5_build_block(sh, i); + } + insert_hash(conf, sh); +} + +static struct stripe_head *__find_stripe(raid5_conf_t *conf, sector_t sector) +{ + struct stripe_head *sh; + + CHECK_DEVLOCK(); + PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector); + for (sh = stripe_hash(conf, sector); sh; sh = sh->hash_next) + if (sh->sector == sector) + return sh; + PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector); + return NULL; +} + +static void unplug_slaves(mddev_t *mddev); +static void raid5_unplug_device(request_queue_t *q); + +static struct stripe_head *get_active_stripe(raid5_conf_t *conf, sector_t sector, + int pd_idx, int noblock) +{ + struct stripe_head *sh; + + PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector); + + spin_lock_irq(&conf->device_lock); + + do { + sh = __find_stripe(conf, sector); + if (!sh) { + if (!conf->inactive_blocked) + sh = get_free_stripe(conf); + if (noblock && sh == NULL) + break; + if (!sh) { + conf->inactive_blocked = 1; + wait_event_lock_irq(conf->wait_for_stripe, + !list_empty(&conf->inactive_list) && + (atomic_read(&conf->active_stripes) < (NR_STRIPES *3/4) + || !conf->inactive_blocked), + conf->device_lock, + unplug_slaves(conf->mddev); + ); + conf->inactive_blocked = 0; + } else + init_stripe(sh, sector, pd_idx); + } else { + if (atomic_read(&sh->count)) { + if (!list_empty(&sh->lru)) + BUG(); + } else { + if (!test_bit(STRIPE_HANDLE, &sh->state)) + atomic_inc(&conf->active_stripes); + if (list_empty(&sh->lru)) + BUG(); + list_del_init(&sh->lru); + } + } + } while (sh == NULL); + + if (sh) + atomic_inc(&sh->count); + + spin_unlock_irq(&conf->device_lock); + return sh; +} + +static int grow_stripes(raid5_conf_t *conf, int num) +{ + struct stripe_head *sh; + kmem_cache_t *sc; + int devs = conf->raid_disks; + + sprintf(conf->cache_name, "raid5/%s", mdname(conf->mddev)); + + sc = kmem_cache_create(conf->cache_name, + sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), + 0, 0, NULL, NULL); + if (!sc) + return 1; + conf->slab_cache = sc; + while (num--) { + sh = kmem_cache_alloc(sc, GFP_KERNEL); + if (!sh) + return 1; + memset(sh, 0, sizeof(*sh) + (devs-1)*sizeof(struct r5dev)); + sh->raid_conf = conf; + spin_lock_init(&sh->lock); + + if (grow_buffers(sh, conf->raid_disks)) { + shrink_buffers(sh, conf->raid_disks); + kmem_cache_free(sc, sh); + return 1; + } + /* we just created an active stripe so... */ + atomic_set(&sh->count, 1); + atomic_inc(&conf->active_stripes); + INIT_LIST_HEAD(&sh->lru); + release_stripe(sh); + } + return 0; +} + +static void shrink_stripes(raid5_conf_t *conf) +{ + struct stripe_head *sh; + + while (1) { + spin_lock_irq(&conf->device_lock); + sh = get_free_stripe(conf); + spin_unlock_irq(&conf->device_lock); + if (!sh) + break; + if (atomic_read(&sh->count)) + BUG(); + shrink_buffers(sh, conf->raid_disks); + kmem_cache_free(conf->slab_cache, sh); + atomic_dec(&conf->active_stripes); + } + kmem_cache_destroy(conf->slab_cache); + conf->slab_cache = NULL; +} + +static int raid5_end_read_request (struct bio * bi, unsigned int bytes_done, + int error) +{ + struct stripe_head *sh = bi->bi_private; + raid5_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); + + if (bi->bi_size) + return 1; + + for (i=0 ; i<disks; i++) + if (bi == &sh->dev[i].req) + break; + + PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n", + (unsigned long long)sh->sector, i, atomic_read(&sh->count), + uptodate); + if (i == disks) { + BUG(); + return 0; + } + + if (uptodate) { +#if 0 + struct bio *bio; + unsigned long flags; + spin_lock_irqsave(&conf->device_lock, flags); + /* we can return a buffer if we bypassed the cache or + * if the top buffer is not in highmem. If there are + * multiple buffers, leave the extra work to + * handle_stripe + */ + buffer = sh->bh_read[i]; + if (buffer && + (!PageHighMem(buffer->b_page) + || buffer->b_page == bh->b_page ) + ) { + sh->bh_read[i] = buffer->b_reqnext; + buffer->b_reqnext = NULL; + } else + buffer = NULL; + spin_unlock_irqrestore(&conf->device_lock, flags); + if (sh->bh_page[i]==bh->b_page) + set_buffer_uptodate(bh); + if (buffer) { + if (buffer->b_page != bh->b_page) + memcpy(buffer->b_data, bh->b_data, bh->b_size); + buffer->b_end_io(buffer, 1); + } +#else + set_bit(R5_UPTODATE, &sh->dev[i].flags); +#endif + } else { + md_error(conf->mddev, conf->disks[i].rdev); + clear_bit(R5_UPTODATE, &sh->dev[i].flags); + } + rdev_dec_pending(conf->disks[i].rdev, conf->mddev); +#if 0 + /* must restore b_page before unlocking buffer... */ + if (sh->bh_page[i] != bh->b_page) { + bh->b_page = sh->bh_page[i]; + bh->b_data = page_address(bh->b_page); + clear_buffer_uptodate(bh); + } +#endif + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + release_stripe(sh); + return 0; +} + +static int raid5_end_write_request (struct bio *bi, unsigned int bytes_done, + int error) +{ + struct stripe_head *sh = bi->bi_private; + raid5_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + unsigned long flags; + int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); + + if (bi->bi_size) + return 1; + + for (i=0 ; i<disks; i++) + if (bi == &sh->dev[i].req) + break; + + PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n", + (unsigned long long)sh->sector, i, atomic_read(&sh->count), + uptodate); + if (i == disks) { + BUG(); + return 0; + } + + spin_lock_irqsave(&conf->device_lock, flags); + if (!uptodate) + md_error(conf->mddev, conf->disks[i].rdev); + + rdev_dec_pending(conf->disks[i].rdev, conf->mddev); + + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + __release_stripe(conf, sh); + spin_unlock_irqrestore(&conf->device_lock, flags); + return 0; +} + + +static sector_t compute_blocknr(struct stripe_head *sh, int i); + +static void raid5_build_block (struct stripe_head *sh, int i) +{ + struct r5dev *dev = &sh->dev[i]; + + bio_init(&dev->req); + dev->req.bi_io_vec = &dev->vec; + dev->req.bi_vcnt++; + dev->req.bi_max_vecs++; + dev->vec.bv_page = dev->page; + dev->vec.bv_len = STRIPE_SIZE; + dev->vec.bv_offset = 0; + + dev->req.bi_sector = sh->sector; + dev->req.bi_private = sh; + + dev->flags = 0; + if (i != sh->pd_idx) + dev->sector = compute_blocknr(sh, i); +} + +static void error(mddev_t *mddev, mdk_rdev_t *rdev) +{ + char b[BDEVNAME_SIZE]; + raid5_conf_t *conf = (raid5_conf_t *) mddev->private; + PRINTK("raid5: error called\n"); + + if (!rdev->faulty) { + mddev->sb_dirty = 1; + if (rdev->in_sync) { + conf->working_disks--; + mddev->degraded++; + conf->failed_disks++; + rdev->in_sync = 0; + /* + * if recovery was running, make sure it aborts. + */ + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + } + rdev->faulty = 1; + printk (KERN_ALERT + "raid5: Disk failure on %s, disabling device." + " Operation continuing on %d devices\n", + bdevname(rdev->bdev,b), conf->working_disks); + } +} + +/* + * Input: a 'big' sector number, + * Output: index of the data and parity disk, and the sector # in them. + */ +static sector_t raid5_compute_sector(sector_t r_sector, unsigned int raid_disks, + unsigned int data_disks, unsigned int * dd_idx, + unsigned int * pd_idx, raid5_conf_t *conf) +{ + long stripe; + unsigned long chunk_number; + unsigned int chunk_offset; + sector_t new_sector; + int sectors_per_chunk = conf->chunk_size >> 9; + + /* First compute the information on this sector */ + + /* + * Compute the chunk number and the sector offset inside the chunk + */ + chunk_offset = sector_div(r_sector, sectors_per_chunk); + chunk_number = r_sector; + BUG_ON(r_sector != chunk_number); + + /* + * Compute the stripe number + */ + stripe = chunk_number / data_disks; + + /* + * Compute the data disk and parity disk indexes inside the stripe + */ + *dd_idx = chunk_number % data_disks; + + /* + * Select the parity disk based on the user selected algorithm. + */ + if (conf->level == 4) + *pd_idx = data_disks; + else switch (conf->algorithm) { + case ALGORITHM_LEFT_ASYMMETRIC: + *pd_idx = data_disks - stripe % raid_disks; + if (*dd_idx >= *pd_idx) + (*dd_idx)++; + break; + case ALGORITHM_RIGHT_ASYMMETRIC: + *pd_idx = stripe % raid_disks; + if (*dd_idx >= *pd_idx) + (*dd_idx)++; + break; + case ALGORITHM_LEFT_SYMMETRIC: + *pd_idx = data_disks - stripe % raid_disks; + *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; + break; + case ALGORITHM_RIGHT_SYMMETRIC: + *pd_idx = stripe % raid_disks; + *dd_idx = (*pd_idx + 1 + *dd_idx) % raid_disks; + break; + default: + printk("raid5: unsupported algorithm %d\n", + conf->algorithm); + } + + /* + * Finally, compute the new sector number + */ + new_sector = (sector_t)stripe * sectors_per_chunk + chunk_offset; + return new_sector; +} + + +static sector_t compute_blocknr(struct stripe_head *sh, int i) +{ + raid5_conf_t *conf = sh->raid_conf; + int raid_disks = conf->raid_disks, data_disks = raid_disks - 1; + sector_t new_sector = sh->sector, check; + int sectors_per_chunk = conf->chunk_size >> 9; + sector_t stripe; + int chunk_offset; + int chunk_number, dummy1, dummy2, dd_idx = i; + sector_t r_sector; + + chunk_offset = sector_div(new_sector, sectors_per_chunk); + stripe = new_sector; + BUG_ON(new_sector != stripe); + + + switch (conf->algorithm) { + case ALGORITHM_LEFT_ASYMMETRIC: + case ALGORITHM_RIGHT_ASYMMETRIC: + if (i > sh->pd_idx) + i--; + break; + case ALGORITHM_LEFT_SYMMETRIC: + case ALGORITHM_RIGHT_SYMMETRIC: + if (i < sh->pd_idx) + i += raid_disks; + i -= (sh->pd_idx + 1); + break; + default: + printk("raid5: unsupported algorithm %d\n", + conf->algorithm); + } + + chunk_number = stripe * data_disks + i; + r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; + + check = raid5_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf); + if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { + printk("compute_blocknr: map not correct\n"); + return 0; + } + return r_sector; +} + + + +/* + * Copy data between a page in the stripe cache, and a bio. + * There are no alignment or size guarantees between the page or the + * bio except that there is some overlap. + * All iovecs in the bio must be considered. + */ +static void copy_data(int frombio, struct bio *bio, + struct page *page, + sector_t sector) +{ + char *pa = page_address(page); + struct bio_vec *bvl; + int i; + int page_offset; + + if (bio->bi_sector >= sector) + page_offset = (signed)(bio->bi_sector - sector) * 512; + else + page_offset = (signed)(sector - bio->bi_sector) * -512; + bio_for_each_segment(bvl, bio, i) { + int len = bio_iovec_idx(bio,i)->bv_len; + int clen; + int b_offset = 0; + + if (page_offset < 0) { + b_offset = -page_offset; + page_offset += b_offset; + len -= b_offset; + } + + if (len > 0 && page_offset + len > STRIPE_SIZE) + clen = STRIPE_SIZE - page_offset; + else clen = len; + + if (clen > 0) { + char *ba = __bio_kmap_atomic(bio, i, KM_USER0); + if (frombio) + memcpy(pa+page_offset, ba+b_offset, clen); + else + memcpy(ba+b_offset, pa+page_offset, clen); + __bio_kunmap_atomic(ba, KM_USER0); + } + if (clen < len) /* hit end of page */ + break; + page_offset += len; + } +} + +#define check_xor() do { \ + if (count == MAX_XOR_BLOCKS) { \ + xor_block(count, STRIPE_SIZE, ptr); \ + count = 1; \ + } \ + } while(0) + + +static void compute_block(struct stripe_head *sh, int dd_idx) +{ + raid5_conf_t *conf = sh->raid_conf; + int i, count, disks = conf->raid_disks; + void *ptr[MAX_XOR_BLOCKS], *p; + + PRINTK("compute_block, stripe %llu, idx %d\n", + (unsigned long long)sh->sector, dd_idx); + + ptr[0] = page_address(sh->dev[dd_idx].page); + memset(ptr[0], 0, STRIPE_SIZE); + count = 1; + for (i = disks ; i--; ) { + if (i == dd_idx) + continue; + p = page_address(sh->dev[i].page); + if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) + ptr[count++] = p; + else + printk("compute_block() %d, stripe %llu, %d" + " not present\n", dd_idx, + (unsigned long long)sh->sector, i); + + check_xor(); + } + if (count != 1) + xor_block(count, STRIPE_SIZE, ptr); + set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); +} + +static void compute_parity(struct stripe_head *sh, int method) +{ + raid5_conf_t *conf = sh->raid_conf; + int i, pd_idx = sh->pd_idx, disks = conf->raid_disks, count; + void *ptr[MAX_XOR_BLOCKS]; + struct bio *chosen; + + PRINTK("compute_parity, stripe %llu, method %d\n", + (unsigned long long)sh->sector, method); + + count = 1; + ptr[0] = page_address(sh->dev[pd_idx].page); + switch(method) { + case READ_MODIFY_WRITE: + if (!test_bit(R5_UPTODATE, &sh->dev[pd_idx].flags)) + BUG(); + for (i=disks ; i-- ;) { + if (i==pd_idx) + continue; + if (sh->dev[i].towrite && + test_bit(R5_UPTODATE, &sh->dev[i].flags)) { + ptr[count++] = page_address(sh->dev[i].page); + chosen = sh->dev[i].towrite; + sh->dev[i].towrite = NULL; + + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + + if (sh->dev[i].written) BUG(); + sh->dev[i].written = chosen; + check_xor(); + } + } + break; + case RECONSTRUCT_WRITE: + memset(ptr[0], 0, STRIPE_SIZE); + for (i= disks; i-- ;) + if (i!=pd_idx && sh->dev[i].towrite) { + chosen = sh->dev[i].towrite; + sh->dev[i].towrite = NULL; + + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + + if (sh->dev[i].written) BUG(); + sh->dev[i].written = chosen; + } + break; + case CHECK_PARITY: + break; + } + if (count>1) { + xor_block(count, STRIPE_SIZE, ptr); + count = 1; + } + + for (i = disks; i--;) + if (sh->dev[i].written) { + sector_t sector = sh->dev[i].sector; + struct bio *wbi = sh->dev[i].written; + while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { + copy_data(1, wbi, sh->dev[i].page, sector); + wbi = r5_next_bio(wbi, sector); + } + + set_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(R5_UPTODATE, &sh->dev[i].flags); + } + + switch(method) { + case RECONSTRUCT_WRITE: + case CHECK_PARITY: + for (i=disks; i--;) + if (i != pd_idx) { + ptr[count++] = page_address(sh->dev[i].page); + check_xor(); + } + break; + case READ_MODIFY_WRITE: + for (i = disks; i--;) + if (sh->dev[i].written) { + ptr[count++] = page_address(sh->dev[i].page); + check_xor(); + } + } + if (count != 1) + xor_block(count, STRIPE_SIZE, ptr); + + if (method != CHECK_PARITY) { + set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); + set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); + } else + clear_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); +} + +/* + * Each stripe/dev can have one or more bion attached. + * toread/towrite point to the first in a chain. + * The bi_next chain must be in order. + */ +static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) +{ + struct bio **bip; + raid5_conf_t *conf = sh->raid_conf; + + PRINTK("adding bh b#%llu to stripe s#%llu\n", + (unsigned long long)bi->bi_sector, + (unsigned long long)sh->sector); + + + spin_lock(&sh->lock); + spin_lock_irq(&conf->device_lock); + if (forwrite) + bip = &sh->dev[dd_idx].towrite; + else + bip = &sh->dev[dd_idx].toread; + while (*bip && (*bip)->bi_sector < bi->bi_sector) { + if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) + goto overlap; + bip = & (*bip)->bi_next; + } + if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) + goto overlap; + + if (*bip && bi->bi_next && (*bip) != bi->bi_next) + BUG(); + if (*bip) + bi->bi_next = *bip; + *bip = bi; + bi->bi_phys_segments ++; + spin_unlock_irq(&conf->device_lock); + spin_unlock(&sh->lock); + + PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n", + (unsigned long long)bi->bi_sector, + (unsigned long long)sh->sector, dd_idx); + + if (forwrite) { + /* check if page is covered */ + sector_t sector = sh->dev[dd_idx].sector; + for (bi=sh->dev[dd_idx].towrite; + sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && + bi && bi->bi_sector <= sector; + bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { + if (bi->bi_sector + (bi->bi_size>>9) >= sector) + sector = bi->bi_sector + (bi->bi_size>>9); + } + if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) + set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); + } + return 1; + + overlap: + set_bit(R5_Overlap, &sh->dev[dd_idx].flags); + spin_unlock_irq(&conf->device_lock); + spin_unlock(&sh->lock); + return 0; +} + + +/* + * handle_stripe - do things to a stripe. + * + * We lock the stripe and then examine the state of various bits + * to see what needs to be done. + * Possible results: + * return some read request which now have data + * return some write requests which are safely on disc + * schedule a read on some buffers + * schedule a write of some buffers + * return confirmation of parity correctness + * + * Parity calculations are done inside the stripe lock + * buffers are taken off read_list or write_list, and bh_cache buffers + * get BH_Lock set before the stripe lock is released. + * + */ + +static void handle_stripe(struct stripe_head *sh) +{ + raid5_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks; + struct bio *return_bi= NULL; + struct bio *bi; + int i; + int syncing; + int locked=0, uptodate=0, to_read=0, to_write=0, failed=0, written=0; + int non_overwrite = 0; + int failed_num=0; + struct r5dev *dev; + + PRINTK("handling stripe %llu, cnt=%d, pd_idx=%d\n", + (unsigned long long)sh->sector, atomic_read(&sh->count), + sh->pd_idx); + + spin_lock(&sh->lock); + clear_bit(STRIPE_HANDLE, &sh->state); + clear_bit(STRIPE_DELAYED, &sh->state); + + syncing = test_bit(STRIPE_SYNCING, &sh->state); + /* Now to look around and see what can be done */ + + for (i=disks; i--; ) { + mdk_rdev_t *rdev; + dev = &sh->dev[i]; + clear_bit(R5_Insync, &dev->flags); + clear_bit(R5_Syncio, &dev->flags); + + PRINTK("check %d: state 0x%lx read %p write %p written %p\n", + i, dev->flags, dev->toread, dev->towrite, dev->written); + /* maybe we can reply to a read */ + if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { + struct bio *rbi, *rbi2; + PRINTK("Return read for disc %d\n", i); + spin_lock_irq(&conf->device_lock); + rbi = dev->toread; + dev->toread = NULL; + if (test_and_clear_bit(R5_Overlap, &dev->flags)) + wake_up(&conf->wait_for_overlap); + spin_unlock_irq(&conf->device_lock); + while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { + copy_data(0, rbi, dev->page, dev->sector); + rbi2 = r5_next_bio(rbi, dev->sector); + spin_lock_irq(&conf->device_lock); + if (--rbi->bi_phys_segments == 0) { + rbi->bi_next = return_bi; + return_bi = rbi; + } + spin_unlock_irq(&conf->device_lock); + rbi = rbi2; + } + } + + /* now count some things */ + if (test_bit(R5_LOCKED, &dev->flags)) locked++; + if (test_bit(R5_UPTODATE, &dev->flags)) uptodate++; + + + if (dev->toread) to_read++; + if (dev->towrite) { + to_write++; + if (!test_bit(R5_OVERWRITE, &dev->flags)) + non_overwrite++; + } + if (dev->written) written++; + rdev = conf->disks[i].rdev; /* FIXME, should I be looking rdev */ + if (!rdev || !rdev->in_sync) { + failed++; + failed_num = i; + } else + set_bit(R5_Insync, &dev->flags); + } + PRINTK("locked=%d uptodate=%d to_read=%d" + " to_write=%d failed=%d failed_num=%d\n", + locked, uptodate, to_read, to_write, failed, failed_num); + /* check if the array has lost two devices and, if so, some requests might + * need to be failed + */ + if (failed > 1 && to_read+to_write+written) { + spin_lock_irq(&conf->device_lock); + for (i=disks; i--; ) { + /* fail all writes first */ + bi = sh->dev[i].towrite; + sh->dev[i].towrite = NULL; + if (bi) to_write--; + + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ + struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + bi->bi_next = return_bi; + return_bi = bi; + } + bi = nextbi; + } + /* and fail all 'written' */ + bi = sh->dev[i].written; + sh->dev[i].written = NULL; + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS) { + struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + bi->bi_next = return_bi; + return_bi = bi; + } + bi = bi2; + } + + /* fail any reads if this device is non-operational */ + if (!test_bit(R5_Insync, &sh->dev[i].flags)) { + bi = sh->dev[i].toread; + sh->dev[i].toread = NULL; + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + if (bi) to_read--; + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ + struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + bi->bi_next = return_bi; + return_bi = bi; + } + bi = nextbi; + } + } + } + spin_unlock_irq(&conf->device_lock); + } + if (failed > 1 && syncing) { + md_done_sync(conf->mddev, STRIPE_SECTORS,0); + clear_bit(STRIPE_SYNCING, &sh->state); + syncing = 0; + } + + /* might be able to return some write requests if the parity block + * is safe, or on a failed drive + */ + dev = &sh->dev[sh->pd_idx]; + if ( written && + ( (test_bit(R5_Insync, &dev->flags) && !test_bit(R5_LOCKED, &dev->flags) && + test_bit(R5_UPTODATE, &dev->flags)) + || (failed == 1 && failed_num == sh->pd_idx)) + ) { + /* any written block on an uptodate or failed drive can be returned. + * Note that if we 'wrote' to a failed drive, it will be UPTODATE, but + * never LOCKED, so we don't need to test 'failed' directly. + */ + for (i=disks; i--; ) + if (sh->dev[i].written) { + dev = &sh->dev[i]; + if (!test_bit(R5_LOCKED, &dev->flags) && + test_bit(R5_UPTODATE, &dev->flags) ) { + /* We can return any write requests */ + struct bio *wbi, *wbi2; + PRINTK("Return write for disc %d\n", i); + spin_lock_irq(&conf->device_lock); + wbi = dev->written; + dev->written = NULL; + while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) { + wbi2 = r5_next_bio(wbi, dev->sector); + if (--wbi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + wbi->bi_next = return_bi; + return_bi = wbi; + } + wbi = wbi2; + } + spin_unlock_irq(&conf->device_lock); + } + } + } + + /* Now we might consider reading some blocks, either to check/generate + * parity, or to satisfy requests + * or to load a block that is being partially written. + */ + if (to_read || non_overwrite || (syncing && (uptodate < disks))) { + for (i=disks; i--;) { + dev = &sh->dev[i]; + if (!test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && + (dev->toread || + (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || + syncing || + (failed && (sh->dev[failed_num].toread || + (sh->dev[failed_num].towrite && !test_bit(R5_OVERWRITE, &sh->dev[failed_num].flags)))) + ) + ) { + /* we would like to get this block, possibly + * by computing it, but we might not be able to + */ + if (uptodate == disks-1) { + PRINTK("Computing block %d\n", i); + compute_block(sh, i); + uptodate++; + } else if (test_bit(R5_Insync, &dev->flags)) { + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantread, &dev->flags); +#if 0 + /* if I am just reading this block and we don't have + a failed drive, or any pending writes then sidestep the cache */ + if (sh->bh_read[i] && !sh->bh_read[i]->b_reqnext && + ! syncing && !failed && !to_write) { + sh->bh_cache[i]->b_page = sh->bh_read[i]->b_page; + sh->bh_cache[i]->b_data = sh->bh_read[i]->b_data; + } +#endif + locked++; + PRINTK("Reading block %d (sync=%d)\n", + i, syncing); + if (syncing) + md_sync_acct(conf->disks[i].rdev->bdev, + STRIPE_SECTORS); + } + } + } + set_bit(STRIPE_HANDLE, &sh->state); + } + + /* now to consider writing and what else, if anything should be read */ + if (to_write) { + int rmw=0, rcw=0; + for (i=disks ; i--;) { + /* would I have to read this buffer for read_modify_write */ + dev = &sh->dev[i]; + if ((dev->towrite || i == sh->pd_idx) && + (!test_bit(R5_LOCKED, &dev->flags) +#if 0 +|| sh->bh_page[i]!=bh->b_page +#endif + ) && + !test_bit(R5_UPTODATE, &dev->flags)) { + if (test_bit(R5_Insync, &dev->flags) +/* && !(!mddev->insync && i == sh->pd_idx) */ + ) + rmw++; + else rmw += 2*disks; /* cannot read it */ + } + /* Would I have to read this buffer for reconstruct_write */ + if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && + (!test_bit(R5_LOCKED, &dev->flags) +#if 0 +|| sh->bh_page[i] != bh->b_page +#endif + ) && + !test_bit(R5_UPTODATE, &dev->flags)) { + if (test_bit(R5_Insync, &dev->flags)) rcw++; + else rcw += 2*disks; + } + } + PRINTK("for sector %llu, rmw=%d rcw=%d\n", + (unsigned long long)sh->sector, rmw, rcw); + set_bit(STRIPE_HANDLE, &sh->state); + if (rmw < rcw && rmw > 0) + /* prefer read-modify-write, but need to get some data */ + for (i=disks; i--;) { + dev = &sh->dev[i]; + if ((dev->towrite || i == sh->pd_idx) && + !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && + test_bit(R5_Insync, &dev->flags)) { + if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + { + PRINTK("Read_old block %d for r-m-w\n", i); + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantread, &dev->flags); + locked++; + } else { + set_bit(STRIPE_DELAYED, &sh->state); + set_bit(STRIPE_HANDLE, &sh->state); + } + } + } + if (rcw <= rmw && rcw > 0) + /* want reconstruct write, but need to get some data */ + for (i=disks; i--;) { + dev = &sh->dev[i]; + if (!test_bit(R5_OVERWRITE, &dev->flags) && i != sh->pd_idx && + !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && + test_bit(R5_Insync, &dev->flags)) { + if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + { + PRINTK("Read_old block %d for Reconstruct\n", i); + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantread, &dev->flags); + locked++; + } else { + set_bit(STRIPE_DELAYED, &sh->state); + set_bit(STRIPE_HANDLE, &sh->state); + } + } + } + /* now if nothing is locked, and if we have enough data, we can start a write request */ + if (locked == 0 && (rcw == 0 ||rmw == 0)) { + PRINTK("Computing parity...\n"); + compute_parity(sh, rcw==0 ? RECONSTRUCT_WRITE : READ_MODIFY_WRITE); + /* now every locked buffer is ready to be written */ + for (i=disks; i--;) + if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { + PRINTK("Writing block %d\n", i); + locked++; + set_bit(R5_Wantwrite, &sh->dev[i].flags); + if (!test_bit(R5_Insync, &sh->dev[i].flags) + || (i==sh->pd_idx && failed == 0)) + set_bit(STRIPE_INSYNC, &sh->state); + } + if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { + atomic_dec(&conf->preread_active_stripes); + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) + md_wakeup_thread(conf->mddev->thread); + } + } + } + + /* maybe we need to check and possibly fix the parity for this stripe + * Any reads will already have been scheduled, so we just see if enough data + * is available + */ + if (syncing && locked == 0 && + !test_bit(STRIPE_INSYNC, &sh->state) && failed <= 1) { + set_bit(STRIPE_HANDLE, &sh->state); + if (failed == 0) { + char *pagea; + if (uptodate != disks) + BUG(); + compute_parity(sh, CHECK_PARITY); + uptodate--; + pagea = page_address(sh->dev[sh->pd_idx].page); + if ((*(u32*)pagea) == 0 && + !memcmp(pagea, pagea+4, STRIPE_SIZE-4)) { + /* parity is correct (on disc, not in buffer any more) */ + set_bit(STRIPE_INSYNC, &sh->state); + } + } + if (!test_bit(STRIPE_INSYNC, &sh->state)) { + if (failed==0) + failed_num = sh->pd_idx; + /* should be able to compute the missing block and write it to spare */ + if (!test_bit(R5_UPTODATE, &sh->dev[failed_num].flags)) { + if (uptodate+1 != disks) + BUG(); + compute_block(sh, failed_num); + uptodate++; + } + if (uptodate != disks) + BUG(); + dev = &sh->dev[failed_num]; + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantwrite, &dev->flags); + locked++; + set_bit(STRIPE_INSYNC, &sh->state); + set_bit(R5_Syncio, &dev->flags); + } + } + if (syncing && locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { + md_done_sync(conf->mddev, STRIPE_SECTORS,1); + clear_bit(STRIPE_SYNCING, &sh->state); + } + + spin_unlock(&sh->lock); + + while ((bi=return_bi)) { + int bytes = bi->bi_size; + + return_bi = bi->bi_next; + bi->bi_next = NULL; + bi->bi_size = 0; + bi->bi_end_io(bi, bytes, 0); + } + for (i=disks; i-- ;) { + int rw; + struct bio *bi; + mdk_rdev_t *rdev; + if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) + rw = 1; + else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) + rw = 0; + else + continue; + + bi = &sh->dev[i].req; + + bi->bi_rw = rw; + if (rw) + bi->bi_end_io = raid5_end_write_request; + else + bi->bi_end_io = raid5_end_read_request; + + rcu_read_lock(); + rdev = conf->disks[i].rdev; + if (rdev && rdev->faulty) + rdev = NULL; + if (rdev) + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (rdev) { + if (test_bit(R5_Syncio, &sh->dev[i].flags)) + md_sync_acct(rdev->bdev, STRIPE_SECTORS); + + bi->bi_bdev = rdev->bdev; + PRINTK("for %llu schedule op %ld on disc %d\n", + (unsigned long long)sh->sector, bi->bi_rw, i); + atomic_inc(&sh->count); + bi->bi_sector = sh->sector + rdev->data_offset; + bi->bi_flags = 1 << BIO_UPTODATE; + bi->bi_vcnt = 1; + bi->bi_max_vecs = 1; + bi->bi_idx = 0; + bi->bi_io_vec = &sh->dev[i].vec; + bi->bi_io_vec[0].bv_len = STRIPE_SIZE; + bi->bi_io_vec[0].bv_offset = 0; + bi->bi_size = STRIPE_SIZE; + bi->bi_next = NULL; + generic_make_request(bi); + } else { + PRINTK("skip op %ld on disc %d for sector %llu\n", + bi->bi_rw, i, (unsigned long long)sh->sector); + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + } + } +} + +static inline void raid5_activate_delayed(raid5_conf_t *conf) +{ + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { + while (!list_empty(&conf->delayed_list)) { + struct list_head *l = conf->delayed_list.next; + struct stripe_head *sh; + sh = list_entry(l, struct stripe_head, lru); + list_del_init(l); + clear_bit(STRIPE_DELAYED, &sh->state); + if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + atomic_inc(&conf->preread_active_stripes); + list_add_tail(&sh->lru, &conf->handle_list); + } + } +} + +static void unplug_slaves(mddev_t *mddev) +{ + raid5_conf_t *conf = mddev_to_conf(mddev); + int i; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks; i++) { + mdk_rdev_t *rdev = conf->disks[i].rdev; + if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) { + request_queue_t *r_queue = bdev_get_queue(rdev->bdev); + + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + rcu_read_unlock(); +} + +static void raid5_unplug_device(request_queue_t *q) +{ + mddev_t *mddev = q->queuedata; + raid5_conf_t *conf = mddev_to_conf(mddev); + unsigned long flags; + + spin_lock_irqsave(&conf->device_lock, flags); + + if (blk_remove_plug(q)) + raid5_activate_delayed(conf); + md_wakeup_thread(mddev->thread); + + spin_unlock_irqrestore(&conf->device_lock, flags); + + unplug_slaves(mddev); +} + +static int raid5_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + raid5_conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + mdk_rdev_t *rdev = conf->disks[i].rdev; + if (rdev && !rdev->faulty) { + struct block_device *bdev = rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, + error_sector); + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + } + rcu_read_unlock(); + return ret; +} + +static inline void raid5_plug_device(raid5_conf_t *conf) +{ + spin_lock_irq(&conf->device_lock); + blk_plug_device(conf->mddev->queue); + spin_unlock_irq(&conf->device_lock); +} + +static int make_request (request_queue_t *q, struct bio * bi) +{ + mddev_t *mddev = q->queuedata; + raid5_conf_t *conf = mddev_to_conf(mddev); + const unsigned int raid_disks = conf->raid_disks; + const unsigned int data_disks = raid_disks - 1; + unsigned int dd_idx, pd_idx; + sector_t new_sector; + sector_t logical_sector, last_sector; + struct stripe_head *sh; + + if (bio_data_dir(bi)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bi)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bi)); + } + + logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); + last_sector = bi->bi_sector + (bi->bi_size>>9); + bi->bi_next = NULL; + bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ + if ( bio_data_dir(bi) == WRITE ) + md_write_start(mddev); + for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { + DEFINE_WAIT(w); + + new_sector = raid5_compute_sector(logical_sector, + raid_disks, data_disks, &dd_idx, &pd_idx, conf); + + PRINTK("raid5: make_request, sector %llu logical %llu\n", + (unsigned long long)new_sector, + (unsigned long long)logical_sector); + + retry: + prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); + sh = get_active_stripe(conf, new_sector, pd_idx, (bi->bi_rw&RWA_MASK)); + if (sh) { + if (!add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { + /* Add failed due to overlap. Flush everything + * and wait a while + */ + raid5_unplug_device(mddev->queue); + release_stripe(sh); + schedule(); + goto retry; + } + finish_wait(&conf->wait_for_overlap, &w); + raid5_plug_device(conf); + handle_stripe(sh); + release_stripe(sh); + + } else { + /* cannot get stripe for read-ahead, just give-up */ + clear_bit(BIO_UPTODATE, &bi->bi_flags); + finish_wait(&conf->wait_for_overlap, &w); + break; + } + + } + spin_lock_irq(&conf->device_lock); + if (--bi->bi_phys_segments == 0) { + int bytes = bi->bi_size; + + if ( bio_data_dir(bi) == WRITE ) + md_write_end(mddev); + bi->bi_size = 0; + bi->bi_end_io(bi, bytes, 0); + } + spin_unlock_irq(&conf->device_lock); + return 0; +} + +/* FIXME go_faster isn't used */ +static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) +{ + raid5_conf_t *conf = (raid5_conf_t *) mddev->private; + struct stripe_head *sh; + int sectors_per_chunk = conf->chunk_size >> 9; + sector_t x; + unsigned long stripe; + int chunk_offset; + int dd_idx, pd_idx; + sector_t first_sector; + int raid_disks = conf->raid_disks; + int data_disks = raid_disks-1; + + if (sector_nr >= mddev->size <<1) { + /* just being told to finish up .. nothing much to do */ + unplug_slaves(mddev); + return 0; + } + /* if there is 1 or more failed drives and we are trying + * to resync, then assert that we are finished, because there is + * nothing we can do. + */ + if (mddev->degraded >= 1 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { + int rv = (mddev->size << 1) - sector_nr; + md_done_sync(mddev, rv, 1); + return rv; + } + + x = sector_nr; + chunk_offset = sector_div(x, sectors_per_chunk); + stripe = x; + BUG_ON(x != stripe); + + first_sector = raid5_compute_sector((sector_t)stripe*data_disks*sectors_per_chunk + + chunk_offset, raid_disks, data_disks, &dd_idx, &pd_idx, conf); + sh = get_active_stripe(conf, sector_nr, pd_idx, 1); + if (sh == NULL) { + sh = get_active_stripe(conf, sector_nr, pd_idx, 0); + /* make sure we don't swamp the stripe cache if someone else + * is trying to get access + */ + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(1); + } + spin_lock(&sh->lock); + set_bit(STRIPE_SYNCING, &sh->state); + clear_bit(STRIPE_INSYNC, &sh->state); + spin_unlock(&sh->lock); + + handle_stripe(sh); + release_stripe(sh); + + return STRIPE_SECTORS; +} + +/* + * This is our raid5 kernel thread. + * + * We scan the hash table for stripes which can be handled now. + * During the scan, completed stripes are saved for us by the interrupt + * handler, so that they will not have to wait for our next wakeup. + */ +static void raid5d (mddev_t *mddev) +{ + struct stripe_head *sh; + raid5_conf_t *conf = mddev_to_conf(mddev); + int handled; + + PRINTK("+++ raid5d active\n"); + + md_check_recovery(mddev); + md_handle_safemode(mddev); + + handled = 0; + spin_lock_irq(&conf->device_lock); + while (1) { + struct list_head *first; + + if (list_empty(&conf->handle_list) && + atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD && + !blk_queue_plugged(mddev->queue) && + !list_empty(&conf->delayed_list)) + raid5_activate_delayed(conf); + + if (list_empty(&conf->handle_list)) + break; + + first = conf->handle_list.next; + sh = list_entry(first, struct stripe_head, lru); + + list_del_init(first); + atomic_inc(&sh->count); + if (atomic_read(&sh->count)!= 1) + BUG(); + spin_unlock_irq(&conf->device_lock); + + handled++; + handle_stripe(sh); + release_stripe(sh); + + spin_lock_irq(&conf->device_lock); + } + PRINTK("%d stripes handled\n", handled); + + spin_unlock_irq(&conf->device_lock); + + unplug_slaves(mddev); + + PRINTK("--- raid5d inactive\n"); +} + +static int run (mddev_t *mddev) +{ + raid5_conf_t *conf; + int raid_disk, memory; + mdk_rdev_t *rdev; + struct disk_info *disk; + struct list_head *tmp; + + if (mddev->level != 5 && mddev->level != 4) { + printk("raid5: %s: raid level not set to 4/5 (%d)\n", mdname(mddev), mddev->level); + return -EIO; + } + + mddev->private = kmalloc (sizeof (raid5_conf_t) + + mddev->raid_disks * sizeof(struct disk_info), + GFP_KERNEL); + if ((conf = mddev->private) == NULL) + goto abort; + memset (conf, 0, sizeof (*conf) + mddev->raid_disks * sizeof(struct disk_info) ); + conf->mddev = mddev; + + if ((conf->stripe_hashtbl = (struct stripe_head **) __get_free_pages(GFP_ATOMIC, HASH_PAGES_ORDER)) == NULL) + goto abort; + memset(conf->stripe_hashtbl, 0, HASH_PAGES * PAGE_SIZE); + + spin_lock_init(&conf->device_lock); + init_waitqueue_head(&conf->wait_for_stripe); + init_waitqueue_head(&conf->wait_for_overlap); + INIT_LIST_HEAD(&conf->handle_list); + INIT_LIST_HEAD(&conf->delayed_list); + INIT_LIST_HEAD(&conf->inactive_list); + atomic_set(&conf->active_stripes, 0); + atomic_set(&conf->preread_active_stripes, 0); + + mddev->queue->unplug_fn = raid5_unplug_device; + mddev->queue->issue_flush_fn = raid5_issue_flush; + + PRINTK("raid5: run(%s) called.\n", mdname(mddev)); + + ITERATE_RDEV(mddev,rdev,tmp) { + raid_disk = rdev->raid_disk; + if (raid_disk >= mddev->raid_disks + || raid_disk < 0) + continue; + disk = conf->disks + raid_disk; + + disk->rdev = rdev; + + if (rdev->in_sync) { + char b[BDEVNAME_SIZE]; + printk(KERN_INFO "raid5: device %s operational as raid" + " disk %d\n", bdevname(rdev->bdev,b), + raid_disk); + conf->working_disks++; + } + } + + conf->raid_disks = mddev->raid_disks; + /* + * 0 for a fully functional array, 1 for a degraded array. + */ + mddev->degraded = conf->failed_disks = conf->raid_disks - conf->working_disks; + conf->mddev = mddev; + conf->chunk_size = mddev->chunk_size; + conf->level = mddev->level; + conf->algorithm = mddev->layout; + conf->max_nr_stripes = NR_STRIPES; + + /* device size must be a multiple of chunk size */ + mddev->size &= ~(mddev->chunk_size/1024 -1); + + if (!conf->chunk_size || conf->chunk_size % 4) { + printk(KERN_ERR "raid5: invalid chunk size %d for %s\n", + conf->chunk_size, mdname(mddev)); + goto abort; + } + if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { + printk(KERN_ERR + "raid5: unsupported parity algorithm %d for %s\n", + conf->algorithm, mdname(mddev)); + goto abort; + } + if (mddev->degraded > 1) { + printk(KERN_ERR "raid5: not enough operational devices for %s" + " (%d/%d failed)\n", + mdname(mddev), conf->failed_disks, conf->raid_disks); + goto abort; + } + + if (mddev->degraded == 1 && + mddev->recovery_cp != MaxSector) { + printk(KERN_ERR + "raid5: cannot start dirty degraded array for %s\n", + mdname(mddev)); + goto abort; + } + + { + mddev->thread = md_register_thread(raid5d, mddev, "%s_raid5"); + if (!mddev->thread) { + printk(KERN_ERR + "raid5: couldn't allocate thread for %s\n", + mdname(mddev)); + goto abort; + } + } +memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + + conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; + if (grow_stripes(conf, conf->max_nr_stripes)) { + printk(KERN_ERR + "raid5: couldn't allocate %dkB for buffers\n", memory); + shrink_stripes(conf); + md_unregister_thread(mddev->thread); + goto abort; + } else + printk(KERN_INFO "raid5: allocated %dkB for %s\n", + memory, mdname(mddev)); + + if (mddev->degraded == 0) + printk("raid5: raid level %d set %s active with %d out of %d" + " devices, algorithm %d\n", conf->level, mdname(mddev), + mddev->raid_disks-mddev->degraded, mddev->raid_disks, + conf->algorithm); + else + printk(KERN_ALERT "raid5: raid level %d set %s active with %d" + " out of %d devices, algorithm %d\n", conf->level, + mdname(mddev), mddev->raid_disks - mddev->degraded, + mddev->raid_disks, conf->algorithm); + + print_raid5_conf(conf); + + /* read-ahead size must cover two whole stripes, which is + * 2 * (n-1) * chunksize where 'n' is the number of raid devices + */ + { + int stripe = (mddev->raid_disks-1) * mddev->chunk_size + / PAGE_CACHE_SIZE; + if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) + mddev->queue->backing_dev_info.ra_pages = 2 * stripe; + } + + /* Ok, everything is just fine now */ + mddev->array_size = mddev->size * (mddev->raid_disks - 1); + return 0; +abort: + if (conf) { + print_raid5_conf(conf); + if (conf->stripe_hashtbl) + free_pages((unsigned long) conf->stripe_hashtbl, + HASH_PAGES_ORDER); + kfree(conf); + } + mddev->private = NULL; + printk(KERN_ALERT "raid5: failed to run raid set %s\n", mdname(mddev)); + return -EIO; +} + + + +static int stop (mddev_t *mddev) +{ + raid5_conf_t *conf = (raid5_conf_t *) mddev->private; + + md_unregister_thread(mddev->thread); + mddev->thread = NULL; + shrink_stripes(conf); + free_pages((unsigned long) conf->stripe_hashtbl, HASH_PAGES_ORDER); + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + kfree(conf); + mddev->private = NULL; + return 0; +} + +#if RAID5_DEBUG +static void print_sh (struct stripe_head *sh) +{ + int i; + + printk("sh %llu, pd_idx %d, state %ld.\n", + (unsigned long long)sh->sector, sh->pd_idx, sh->state); + printk("sh %llu, count %d.\n", + (unsigned long long)sh->sector, atomic_read(&sh->count)); + printk("sh %llu, ", (unsigned long long)sh->sector); + for (i = 0; i < sh->raid_conf->raid_disks; i++) { + printk("(cache%d: %p %ld) ", + i, sh->dev[i].page, sh->dev[i].flags); + } + printk("\n"); +} + +static void printall (raid5_conf_t *conf) +{ + struct stripe_head *sh; + int i; + + spin_lock_irq(&conf->device_lock); + for (i = 0; i < NR_HASH; i++) { + sh = conf->stripe_hashtbl[i]; + for (; sh; sh = sh->hash_next) { + if (sh->raid_conf != conf) + continue; + print_sh(sh); + } + } + spin_unlock_irq(&conf->device_lock); +} +#endif + +static void status (struct seq_file *seq, mddev_t *mddev) +{ + raid5_conf_t *conf = (raid5_conf_t *) mddev->private; + int i; + + seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); + seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->working_disks); + for (i = 0; i < conf->raid_disks; i++) + seq_printf (seq, "%s", + conf->disks[i].rdev && + conf->disks[i].rdev->in_sync ? "U" : "_"); + seq_printf (seq, "]"); +#if RAID5_DEBUG +#define D(x) \ + seq_printf (seq, "<"#x":%d>", atomic_read(&conf->x)) + printall(conf); +#endif +} + +static void print_raid5_conf (raid5_conf_t *conf) +{ + int i; + struct disk_info *tmp; + + printk("RAID5 conf printout:\n"); + if (!conf) { + printk("(conf==NULL)\n"); + return; + } + printk(" --- rd:%d wd:%d fd:%d\n", conf->raid_disks, + conf->working_disks, conf->failed_disks); + + for (i = 0; i < conf->raid_disks; i++) { + char b[BDEVNAME_SIZE]; + tmp = conf->disks + i; + if (tmp->rdev) + printk(" disk %d, o:%d, dev:%s\n", + i, !tmp->rdev->faulty, + bdevname(tmp->rdev->bdev,b)); + } +} + +static int raid5_spare_active(mddev_t *mddev) +{ + int i; + raid5_conf_t *conf = mddev->private; + struct disk_info *tmp; + + for (i = 0; i < conf->raid_disks; i++) { + tmp = conf->disks + i; + if (tmp->rdev + && !tmp->rdev->faulty + && !tmp->rdev->in_sync) { + mddev->degraded--; + conf->failed_disks--; + conf->working_disks++; + tmp->rdev->in_sync = 1; + } + } + print_raid5_conf(conf); + return 0; +} + +static int raid5_remove_disk(mddev_t *mddev, int number) +{ + raid5_conf_t *conf = mddev->private; + int err = 0; + mdk_rdev_t *rdev; + struct disk_info *p = conf->disks + number; + + print_raid5_conf(conf); + rdev = p->rdev; + if (rdev) { + if (rdev->in_sync || + atomic_read(&rdev->nr_pending)) { + err = -EBUSY; + goto abort; + } + p->rdev = NULL; + synchronize_kernel(); + if (atomic_read(&rdev->nr_pending)) { + /* lost the race, try later */ + err = -EBUSY; + p->rdev = rdev; + } + } +abort: + + print_raid5_conf(conf); + return err; +} + +static int raid5_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) +{ + raid5_conf_t *conf = mddev->private; + int found = 0; + int disk; + struct disk_info *p; + + if (mddev->degraded > 1) + /* no point adding a device */ + return 0; + + /* + * find the disk ... + */ + for (disk=0; disk < mddev->raid_disks; disk++) + if ((p=conf->disks + disk)->rdev == NULL) { + rdev->in_sync = 0; + rdev->raid_disk = disk; + found = 1; + p->rdev = rdev; + break; + } + print_raid5_conf(conf); + return found; +} + +static int raid5_resize(mddev_t *mddev, sector_t sectors) +{ + /* no resync is happening, and there is enough space + * on all devices, so we can resize. + * We need to make sure resync covers any new space. + * If the array is shrinking we should possibly wait until + * any io in the removed space completes, but it hardly seems + * worth it. + */ + sectors &= ~((sector_t)mddev->chunk_size/512 - 1); + mddev->array_size = (sectors * (mddev->raid_disks-1))>>1; + set_capacity(mddev->gendisk, mddev->array_size << 1); + mddev->changed = 1; + if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) { + mddev->recovery_cp = mddev->size << 1; + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + } + mddev->size = sectors /2; + return 0; +} + +static mdk_personality_t raid5_personality= +{ + .name = "raid5", + .owner = THIS_MODULE, + .make_request = make_request, + .run = run, + .stop = stop, + .status = status, + .error_handler = error, + .hot_add_disk = raid5_add_disk, + .hot_remove_disk= raid5_remove_disk, + .spare_active = raid5_spare_active, + .sync_request = sync_request, + .resize = raid5_resize, +}; + +static int __init raid5_init (void) +{ + return register_md_personality (RAID5, &raid5_personality); +} + +static void raid5_exit (void) +{ + unregister_md_personality (RAID5); +} + +module_init(raid5_init); +module_exit(raid5_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-4"); /* RAID5 */ diff --git a/drivers/md/raid6.h b/drivers/md/raid6.h new file mode 100644 index 00000000000..f80ee6350ed --- /dev/null +++ b/drivers/md/raid6.h @@ -0,0 +1,135 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2003 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +#ifndef LINUX_RAID_RAID6_H +#define LINUX_RAID_RAID6_H + +#ifdef __KERNEL__ + +/* Set to 1 to use kernel-wide empty_zero_page */ +#define RAID6_USE_EMPTY_ZERO_PAGE 0 + +#include <linux/module.h> +#include <linux/stddef.h> +#include <linux/compiler.h> +#include <linux/types.h> +#include <linux/kernel.h> +#include <linux/errno.h> +#include <linux/mempool.h> +#include <linux/list.h> +#include <linux/vmalloc.h> +#include <linux/raid/md.h> +#include <linux/raid/raid5.h> + +typedef raid5_conf_t raid6_conf_t; /* Same configuration */ + +/* Additional compute_parity mode -- updates the parity w/o LOCKING */ +#define UPDATE_PARITY 4 + +/* We need a pre-zeroed page... if we don't want to use the kernel-provided + one define it here */ +#if RAID6_USE_EMPTY_ZERO_PAGE +# define raid6_empty_zero_page empty_zero_page +#else +extern const char raid6_empty_zero_page[PAGE_SIZE]; +#endif + +#else /* ! __KERNEL__ */ +/* Used for testing in user space */ + +#include <errno.h> +#include <inttypes.h> +#include <limits.h> +#include <stddef.h> +#include <sys/mman.h> +#include <sys/types.h> + +/* Not standard, but glibc defines it */ +#define BITS_PER_LONG __WORDSIZE + +typedef uint8_t u8; +typedef uint16_t u16; +typedef uint32_t u32; +typedef uint64_t u64; + +#ifndef PAGE_SIZE +# define PAGE_SIZE 4096 +#endif +extern const char raid6_empty_zero_page[PAGE_SIZE]; + +#define __init +#define __exit +#define __attribute_const__ __attribute__((const)) + +#define preempt_enable() +#define preempt_disable() + +#endif /* __KERNEL__ */ + +/* Routine choices */ +struct raid6_calls { + void (*gen_syndrome)(int, size_t, void **); + int (*valid)(void); /* Returns 1 if this routine set is usable */ + const char *name; /* Name of this routine set */ + int prefer; /* Has special performance attribute */ +}; + +/* Selected algorithm */ +extern struct raid6_calls raid6_call; + +/* Algorithm list */ +extern const struct raid6_calls * const raid6_algos[]; +int raid6_select_algo(void); + +/* Return values from chk_syndrome */ +#define RAID6_OK 0 +#define RAID6_P_BAD 1 +#define RAID6_Q_BAD 2 +#define RAID6_PQ_BAD 3 + +/* Galois field tables */ +extern const u8 raid6_gfmul[256][256] __attribute__((aligned(256))); +extern const u8 raid6_gfexp[256] __attribute__((aligned(256))); +extern const u8 raid6_gfinv[256] __attribute__((aligned(256))); +extern const u8 raid6_gfexi[256] __attribute__((aligned(256))); + +/* Recovery routines */ +void raid6_2data_recov(int disks, size_t bytes, int faila, int failb, void **ptrs); +void raid6_datap_recov(int disks, size_t bytes, int faila, void **ptrs); +void raid6_dual_recov(int disks, size_t bytes, int faila, int failb, void **ptrs); + +/* Some definitions to allow code to be compiled for testing in userspace */ +#ifndef __KERNEL__ + +# define jiffies raid6_jiffies() +# define printk printf +# define GFP_KERNEL 0 +# define __get_free_pages(x,y) ((unsigned long)mmap(NULL, PAGE_SIZE << (y), PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, 0, 0)) +# define free_pages(x,y) munmap((void *)(x), (y)*PAGE_SIZE) + +static inline void cpu_relax(void) +{ + /* Nothing */ +} + +#undef HZ +#define HZ 1000 +static inline uint32_t raid6_jiffies(void) +{ + struct timeval tv; + gettimeofday(&tv, NULL); + return tv.tv_sec*1000 + tv.tv_usec/1000; +} + +#endif /* ! __KERNEL__ */ + +#endif /* LINUX_RAID_RAID6_H */ diff --git a/drivers/md/raid6algos.c b/drivers/md/raid6algos.c new file mode 100644 index 00000000000..acf386fc4b4 --- /dev/null +++ b/drivers/md/raid6algos.c @@ -0,0 +1,153 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6algos.c + * + * Algorithm list and algorithm selection for RAID-6 + */ + +#include "raid6.h" +#ifndef __KERNEL__ +#include <sys/mman.h> +#endif + +struct raid6_calls raid6_call; + +/* Various routine sets */ +extern const struct raid6_calls raid6_intx1; +extern const struct raid6_calls raid6_intx2; +extern const struct raid6_calls raid6_intx4; +extern const struct raid6_calls raid6_intx8; +extern const struct raid6_calls raid6_intx16; +extern const struct raid6_calls raid6_intx32; +extern const struct raid6_calls raid6_mmxx1; +extern const struct raid6_calls raid6_mmxx2; +extern const struct raid6_calls raid6_sse1x1; +extern const struct raid6_calls raid6_sse1x2; +extern const struct raid6_calls raid6_sse2x1; +extern const struct raid6_calls raid6_sse2x2; +extern const struct raid6_calls raid6_sse2x4; +extern const struct raid6_calls raid6_altivec1; +extern const struct raid6_calls raid6_altivec2; +extern const struct raid6_calls raid6_altivec4; +extern const struct raid6_calls raid6_altivec8; + +const struct raid6_calls * const raid6_algos[] = { + &raid6_intx1, + &raid6_intx2, + &raid6_intx4, + &raid6_intx8, +#if defined(__ia64__) + &raid6_intx16, + &raid6_intx32, +#endif +#if defined(__i386__) + &raid6_mmxx1, + &raid6_mmxx2, + &raid6_sse1x1, + &raid6_sse1x2, + &raid6_sse2x1, + &raid6_sse2x2, +#endif +#if defined(__x86_64__) + &raid6_sse2x1, + &raid6_sse2x2, + &raid6_sse2x4, +#endif +#ifdef CONFIG_ALTIVEC + &raid6_altivec1, + &raid6_altivec2, + &raid6_altivec4, + &raid6_altivec8, +#endif + NULL +}; + +#ifdef __KERNEL__ +#define RAID6_TIME_JIFFIES_LG2 4 +#else +/* Need more time to be stable in userspace */ +#define RAID6_TIME_JIFFIES_LG2 9 +#endif + +/* Try to pick the best algorithm */ +/* This code uses the gfmul table as convenient data set to abuse */ + +int __init raid6_select_algo(void) +{ + const struct raid6_calls * const * algo; + const struct raid6_calls * best; + char *syndromes; + void *dptrs[(65536/PAGE_SIZE)+2]; + int i, disks; + unsigned long perf, bestperf; + int bestprefer; + unsigned long j0, j1; + + disks = (65536/PAGE_SIZE)+2; + for ( i = 0 ; i < disks-2 ; i++ ) { + dptrs[i] = ((char *)raid6_gfmul) + PAGE_SIZE*i; + } + + /* Normal code - use a 2-page allocation to avoid D$ conflict */ + syndromes = (void *) __get_free_pages(GFP_KERNEL, 1); + + if ( !syndromes ) { + printk("raid6: Yikes! No memory available.\n"); + return -ENOMEM; + } + + dptrs[disks-2] = syndromes; + dptrs[disks-1] = syndromes + PAGE_SIZE; + + bestperf = 0; bestprefer = 0; best = NULL; + + for ( algo = raid6_algos ; *algo ; algo++ ) { + if ( !(*algo)->valid || (*algo)->valid() ) { + perf = 0; + + preempt_disable(); + j0 = jiffies; + while ( (j1 = jiffies) == j0 ) + cpu_relax(); + while ( (jiffies-j1) < (1 << RAID6_TIME_JIFFIES_LG2) ) { + (*algo)->gen_syndrome(disks, PAGE_SIZE, dptrs); + perf++; + } + preempt_enable(); + + if ( (*algo)->prefer > bestprefer || + ((*algo)->prefer == bestprefer && + perf > bestperf) ) { + best = *algo; + bestprefer = best->prefer; + bestperf = perf; + } + printk("raid6: %-8s %5ld MB/s\n", (*algo)->name, + (perf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2)); + } + } + + if ( best ) + printk("raid6: using algorithm %s (%ld MB/s)\n", + best->name, + (bestperf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2)); + else + printk("raid6: Yikes! No algorithm found!\n"); + + raid6_call = *best; + + free_pages((unsigned long)syndromes, 1); + + return best ? 0 : -EINVAL; +} diff --git a/drivers/md/raid6altivec.uc b/drivers/md/raid6altivec.uc new file mode 100644 index 00000000000..1de8f030eee --- /dev/null +++ b/drivers/md/raid6altivec.uc @@ -0,0 +1,122 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6altivec$#.c + * + * $#-way unrolled portable integer math RAID-6 instruction set + * + * This file is postprocessed using unroll.pl + * + * <benh> hpa: in process, + * you can just "steal" the vec unit with enable_kernel_altivec() (but + * bracked this with preempt_disable/enable or in a lock) + */ + +#include "raid6.h" + +#ifdef CONFIG_ALTIVEC + +#include <altivec.h> +#include <asm/system.h> +#include <asm/cputable.h> + +/* + * This is the C data type to use + */ + +typedef vector unsigned char unative_t; + +#define NBYTES(x) ((vector unsigned char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x}) +#define NSIZE sizeof(unative_t) + +/* + * The SHLBYTE() operation shifts each byte left by 1, *not* + * rolling over into the next byte + */ +static inline __attribute_const__ unative_t SHLBYTE(unative_t v) +{ + return vec_add(v,v); +} + +/* + * The MASK() operation returns 0xFF in any byte for which the high + * bit is 1, 0x00 for any byte for which the high bit is 0. + */ +static inline __attribute_const__ unative_t MASK(unative_t v) +{ + unative_t zv = NBYTES(0); + + /* vec_cmpgt returns a vector bool char; thus the need for the cast */ + return (unative_t)vec_cmpgt(zv, v); +} + + +/* This is noinline to make damned sure that gcc doesn't move any of the + Altivec code around the enable/disable code */ +static void noinline +raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + + unative_t wd$$, wq$$, wp$$, w1$$, w2$$; + unative_t x1d = NBYTES(0x1d); + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + for ( d = 0 ; d < bytes ; d += NSIZE*$# ) { + wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; + for ( z = z0-1 ; z >= 0 ; z-- ) { + wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; + wp$$ = vec_xor(wp$$, wd$$); + w2$$ = MASK(wq$$); + w1$$ = SHLBYTE(wq$$); + w2$$ = vec_and(w2$$, x1d); + w1$$ = vec_xor(w1$$, w2$$); + wq$$ = vec_xor(w1$$, wd$$); + } + *(unative_t *)&p[d+NSIZE*$$] = wp$$; + *(unative_t *)&q[d+NSIZE*$$] = wq$$; + } +} + +static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + preempt_disable(); + enable_kernel_altivec(); + + raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs); + + preempt_enable(); +} + +int raid6_have_altivec(void); +#if $# == 1 +int raid6_have_altivec(void) +{ + /* This assumes either all CPUs have Altivec or none does */ + return cpu_has_feature(CPU_FTR_ALTIVEC); +} +#endif + +const struct raid6_calls raid6_altivec$# = { + raid6_altivec$#_gen_syndrome, + raid6_have_altivec, + "altivecx$#", + 0 +}; + +#endif /* CONFIG_ALTIVEC */ diff --git a/drivers/md/raid6int.uc b/drivers/md/raid6int.uc new file mode 100644 index 00000000000..ad004cee0e2 --- /dev/null +++ b/drivers/md/raid6int.uc @@ -0,0 +1,117 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6int$#.c + * + * $#-way unrolled portable integer math RAID-6 instruction set + * + * This file is postprocessed using unroll.pl + */ + +#include "raid6.h" + +/* + * This is the C data type to use + */ + +/* Change this from BITS_PER_LONG if there is something better... */ +#if BITS_PER_LONG == 64 +# define NBYTES(x) ((x) * 0x0101010101010101UL) +# define NSIZE 8 +# define NSHIFT 3 +# define NSTRING "64" +typedef u64 unative_t; +#else +# define NBYTES(x) ((x) * 0x01010101U) +# define NSIZE 4 +# define NSHIFT 2 +# define NSTRING "32" +typedef u32 unative_t; +#endif + + + +/* + * IA-64 wants insane amounts of unrolling. On other architectures that + * is just a waste of space. + */ +#if ($# <= 8) || defined(__ia64__) + + +/* + * These sub-operations are separate inlines since they can sometimes be + * specially optimized using architecture-specific hacks. + */ + +/* + * The SHLBYTE() operation shifts each byte left by 1, *not* + * rolling over into the next byte + */ +static inline __attribute_const__ unative_t SHLBYTE(unative_t v) +{ + unative_t vv; + + vv = (v << 1) & NBYTES(0xfe); + return vv; +} + +/* + * The MASK() operation returns 0xFF in any byte for which the high + * bit is 1, 0x00 for any byte for which the high bit is 0. + */ +static inline __attribute_const__ unative_t MASK(unative_t v) +{ + unative_t vv; + + vv = v & NBYTES(0x80); + vv = (vv << 1) - (vv >> 7); /* Overflow on the top bit is OK */ + return vv; +} + + +static void raid6_int$#_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + + unative_t wd$$, wq$$, wp$$, w1$$, w2$$; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + for ( d = 0 ; d < bytes ; d += NSIZE*$# ) { + wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; + for ( z = z0-1 ; z >= 0 ; z-- ) { + wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; + wp$$ ^= wd$$; + w2$$ = MASK(wq$$); + w1$$ = SHLBYTE(wq$$); + w2$$ &= NBYTES(0x1d); + w1$$ ^= w2$$; + wq$$ = w1$$ ^ wd$$; + } + *(unative_t *)&p[d+NSIZE*$$] = wp$$; + *(unative_t *)&q[d+NSIZE*$$] = wq$$; + } +} + +const struct raid6_calls raid6_intx$# = { + raid6_int$#_gen_syndrome, + NULL, /* always valid */ + "int" NSTRING "x$#", + 0 +}; + +#endif diff --git a/drivers/md/raid6main.c b/drivers/md/raid6main.c new file mode 100644 index 00000000000..7e30ab29691 --- /dev/null +++ b/drivers/md/raid6main.c @@ -0,0 +1,2136 @@ +/* + * raid6main.c : Multiple Devices driver for Linux + * Copyright (C) 1996, 1997 Ingo Molnar, Miguel de Icaza, Gadi Oxman + * Copyright (C) 1999, 2000 Ingo Molnar + * Copyright (C) 2002, 2003 H. Peter Anvin + * + * RAID-6 management functions. This code is derived from raid5.c. + * Last merge from raid5.c bkcvs version 1.79 (kernel 2.6.1). + * + * Thanks to Penguin Computing for making the RAID-6 development possible + * by donating a test server! + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + + +#include <linux/config.h> +#include <linux/module.h> +#include <linux/slab.h> +#include <linux/highmem.h> +#include <linux/bitops.h> +#include <asm/atomic.h> +#include "raid6.h" + +/* + * Stripe cache + */ + +#define NR_STRIPES 256 +#define STRIPE_SIZE PAGE_SIZE +#define STRIPE_SHIFT (PAGE_SHIFT - 9) +#define STRIPE_SECTORS (STRIPE_SIZE>>9) +#define IO_THRESHOLD 1 +#define HASH_PAGES 1 +#define HASH_PAGES_ORDER 0 +#define NR_HASH (HASH_PAGES * PAGE_SIZE / sizeof(struct stripe_head *)) +#define HASH_MASK (NR_HASH - 1) + +#define stripe_hash(conf, sect) ((conf)->stripe_hashtbl[((sect) >> STRIPE_SHIFT) & HASH_MASK]) + +/* bio's attached to a stripe+device for I/O are linked together in bi_sector + * order without overlap. There may be several bio's per stripe+device, and + * a bio could span several devices. + * When walking this list for a particular stripe+device, we must never proceed + * beyond a bio that extends past this device, as the next bio might no longer + * be valid. + * This macro is used to determine the 'next' bio in the list, given the sector + * of the current stripe+device + */ +#define r5_next_bio(bio, sect) ( ( (bio)->bi_sector + ((bio)->bi_size>>9) < sect + STRIPE_SECTORS) ? (bio)->bi_next : NULL) +/* + * The following can be used to debug the driver + */ +#define RAID6_DEBUG 0 /* Extremely verbose printk */ +#define RAID6_PARANOIA 1 /* Check spinlocks */ +#define RAID6_DUMPSTATE 0 /* Include stripe cache state in /proc/mdstat */ +#if RAID6_PARANOIA && defined(CONFIG_SMP) +# define CHECK_DEVLOCK() assert_spin_locked(&conf->device_lock) +#else +# define CHECK_DEVLOCK() +#endif + +#define PRINTK(x...) ((void)(RAID6_DEBUG && printk(KERN_DEBUG x))) +#if RAID6_DEBUG +#undef inline +#undef __inline__ +#define inline +#define __inline__ +#endif + +#if !RAID6_USE_EMPTY_ZERO_PAGE +/* In .bss so it's zeroed */ +const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); +#endif + +static inline int raid6_next_disk(int disk, int raid_disks) +{ + disk++; + return (disk < raid_disks) ? disk : 0; +} + +static void print_raid6_conf (raid6_conf_t *conf); + +static inline void __release_stripe(raid6_conf_t *conf, struct stripe_head *sh) +{ + if (atomic_dec_and_test(&sh->count)) { + if (!list_empty(&sh->lru)) + BUG(); + if (atomic_read(&conf->active_stripes)==0) + BUG(); + if (test_bit(STRIPE_HANDLE, &sh->state)) { + if (test_bit(STRIPE_DELAYED, &sh->state)) + list_add_tail(&sh->lru, &conf->delayed_list); + else + list_add_tail(&sh->lru, &conf->handle_list); + md_wakeup_thread(conf->mddev->thread); + } else { + if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { + atomic_dec(&conf->preread_active_stripes); + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) + md_wakeup_thread(conf->mddev->thread); + } + list_add_tail(&sh->lru, &conf->inactive_list); + atomic_dec(&conf->active_stripes); + if (!conf->inactive_blocked || + atomic_read(&conf->active_stripes) < (NR_STRIPES*3/4)) + wake_up(&conf->wait_for_stripe); + } + } +} +static void release_stripe(struct stripe_head *sh) +{ + raid6_conf_t *conf = sh->raid_conf; + unsigned long flags; + + spin_lock_irqsave(&conf->device_lock, flags); + __release_stripe(conf, sh); + spin_unlock_irqrestore(&conf->device_lock, flags); +} + +static void remove_hash(struct stripe_head *sh) +{ + PRINTK("remove_hash(), stripe %llu\n", (unsigned long long)sh->sector); + + if (sh->hash_pprev) { + if (sh->hash_next) + sh->hash_next->hash_pprev = sh->hash_pprev; + *sh->hash_pprev = sh->hash_next; + sh->hash_pprev = NULL; + } +} + +static __inline__ void insert_hash(raid6_conf_t *conf, struct stripe_head *sh) +{ + struct stripe_head **shp = &stripe_hash(conf, sh->sector); + + PRINTK("insert_hash(), stripe %llu\n", (unsigned long long)sh->sector); + + CHECK_DEVLOCK(); + if ((sh->hash_next = *shp) != NULL) + (*shp)->hash_pprev = &sh->hash_next; + *shp = sh; + sh->hash_pprev = shp; +} + + +/* find an idle stripe, make sure it is unhashed, and return it. */ +static struct stripe_head *get_free_stripe(raid6_conf_t *conf) +{ + struct stripe_head *sh = NULL; + struct list_head *first; + + CHECK_DEVLOCK(); + if (list_empty(&conf->inactive_list)) + goto out; + first = conf->inactive_list.next; + sh = list_entry(first, struct stripe_head, lru); + list_del_init(first); + remove_hash(sh); + atomic_inc(&conf->active_stripes); +out: + return sh; +} + +static void shrink_buffers(struct stripe_head *sh, int num) +{ + struct page *p; + int i; + + for (i=0; i<num ; i++) { + p = sh->dev[i].page; + if (!p) + continue; + sh->dev[i].page = NULL; + page_cache_release(p); + } +} + +static int grow_buffers(struct stripe_head *sh, int num) +{ + int i; + + for (i=0; i<num; i++) { + struct page *page; + + if (!(page = alloc_page(GFP_KERNEL))) { + return 1; + } + sh->dev[i].page = page; + } + return 0; +} + +static void raid6_build_block (struct stripe_head *sh, int i); + +static inline void init_stripe(struct stripe_head *sh, sector_t sector, int pd_idx) +{ + raid6_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + + if (atomic_read(&sh->count) != 0) + BUG(); + if (test_bit(STRIPE_HANDLE, &sh->state)) + BUG(); + + CHECK_DEVLOCK(); + PRINTK("init_stripe called, stripe %llu\n", + (unsigned long long)sh->sector); + + remove_hash(sh); + + sh->sector = sector; + sh->pd_idx = pd_idx; + sh->state = 0; + + for (i=disks; i--; ) { + struct r5dev *dev = &sh->dev[i]; + + if (dev->toread || dev->towrite || dev->written || + test_bit(R5_LOCKED, &dev->flags)) { + PRINTK("sector=%llx i=%d %p %p %p %d\n", + (unsigned long long)sh->sector, i, dev->toread, + dev->towrite, dev->written, + test_bit(R5_LOCKED, &dev->flags)); + BUG(); + } + dev->flags = 0; + raid6_build_block(sh, i); + } + insert_hash(conf, sh); +} + +static struct stripe_head *__find_stripe(raid6_conf_t *conf, sector_t sector) +{ + struct stripe_head *sh; + + CHECK_DEVLOCK(); + PRINTK("__find_stripe, sector %llu\n", (unsigned long long)sector); + for (sh = stripe_hash(conf, sector); sh; sh = sh->hash_next) + if (sh->sector == sector) + return sh; + PRINTK("__stripe %llu not in cache\n", (unsigned long long)sector); + return NULL; +} + +static void unplug_slaves(mddev_t *mddev); + +static struct stripe_head *get_active_stripe(raid6_conf_t *conf, sector_t sector, + int pd_idx, int noblock) +{ + struct stripe_head *sh; + + PRINTK("get_stripe, sector %llu\n", (unsigned long long)sector); + + spin_lock_irq(&conf->device_lock); + + do { + sh = __find_stripe(conf, sector); + if (!sh) { + if (!conf->inactive_blocked) + sh = get_free_stripe(conf); + if (noblock && sh == NULL) + break; + if (!sh) { + conf->inactive_blocked = 1; + wait_event_lock_irq(conf->wait_for_stripe, + !list_empty(&conf->inactive_list) && + (atomic_read(&conf->active_stripes) < (NR_STRIPES *3/4) + || !conf->inactive_blocked), + conf->device_lock, + unplug_slaves(conf->mddev); + ); + conf->inactive_blocked = 0; + } else + init_stripe(sh, sector, pd_idx); + } else { + if (atomic_read(&sh->count)) { + if (!list_empty(&sh->lru)) + BUG(); + } else { + if (!test_bit(STRIPE_HANDLE, &sh->state)) + atomic_inc(&conf->active_stripes); + if (list_empty(&sh->lru)) + BUG(); + list_del_init(&sh->lru); + } + } + } while (sh == NULL); + + if (sh) + atomic_inc(&sh->count); + + spin_unlock_irq(&conf->device_lock); + return sh; +} + +static int grow_stripes(raid6_conf_t *conf, int num) +{ + struct stripe_head *sh; + kmem_cache_t *sc; + int devs = conf->raid_disks; + + sprintf(conf->cache_name, "raid6/%s", mdname(conf->mddev)); + + sc = kmem_cache_create(conf->cache_name, + sizeof(struct stripe_head)+(devs-1)*sizeof(struct r5dev), + 0, 0, NULL, NULL); + if (!sc) + return 1; + conf->slab_cache = sc; + while (num--) { + sh = kmem_cache_alloc(sc, GFP_KERNEL); + if (!sh) + return 1; + memset(sh, 0, sizeof(*sh) + (devs-1)*sizeof(struct r5dev)); + sh->raid_conf = conf; + spin_lock_init(&sh->lock); + + if (grow_buffers(sh, conf->raid_disks)) { + shrink_buffers(sh, conf->raid_disks); + kmem_cache_free(sc, sh); + return 1; + } + /* we just created an active stripe so... */ + atomic_set(&sh->count, 1); + atomic_inc(&conf->active_stripes); + INIT_LIST_HEAD(&sh->lru); + release_stripe(sh); + } + return 0; +} + +static void shrink_stripes(raid6_conf_t *conf) +{ + struct stripe_head *sh; + + while (1) { + spin_lock_irq(&conf->device_lock); + sh = get_free_stripe(conf); + spin_unlock_irq(&conf->device_lock); + if (!sh) + break; + if (atomic_read(&sh->count)) + BUG(); + shrink_buffers(sh, conf->raid_disks); + kmem_cache_free(conf->slab_cache, sh); + atomic_dec(&conf->active_stripes); + } + kmem_cache_destroy(conf->slab_cache); + conf->slab_cache = NULL; +} + +static int raid6_end_read_request (struct bio * bi, unsigned int bytes_done, + int error) +{ + struct stripe_head *sh = bi->bi_private; + raid6_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); + + if (bi->bi_size) + return 1; + + for (i=0 ; i<disks; i++) + if (bi == &sh->dev[i].req) + break; + + PRINTK("end_read_request %llu/%d, count: %d, uptodate %d.\n", + (unsigned long long)sh->sector, i, atomic_read(&sh->count), + uptodate); + if (i == disks) { + BUG(); + return 0; + } + + if (uptodate) { +#if 0 + struct bio *bio; + unsigned long flags; + spin_lock_irqsave(&conf->device_lock, flags); + /* we can return a buffer if we bypassed the cache or + * if the top buffer is not in highmem. If there are + * multiple buffers, leave the extra work to + * handle_stripe + */ + buffer = sh->bh_read[i]; + if (buffer && + (!PageHighMem(buffer->b_page) + || buffer->b_page == bh->b_page ) + ) { + sh->bh_read[i] = buffer->b_reqnext; + buffer->b_reqnext = NULL; + } else + buffer = NULL; + spin_unlock_irqrestore(&conf->device_lock, flags); + if (sh->bh_page[i]==bh->b_page) + set_buffer_uptodate(bh); + if (buffer) { + if (buffer->b_page != bh->b_page) + memcpy(buffer->b_data, bh->b_data, bh->b_size); + buffer->b_end_io(buffer, 1); + } +#else + set_bit(R5_UPTODATE, &sh->dev[i].flags); +#endif + } else { + md_error(conf->mddev, conf->disks[i].rdev); + clear_bit(R5_UPTODATE, &sh->dev[i].flags); + } + rdev_dec_pending(conf->disks[i].rdev, conf->mddev); +#if 0 + /* must restore b_page before unlocking buffer... */ + if (sh->bh_page[i] != bh->b_page) { + bh->b_page = sh->bh_page[i]; + bh->b_data = page_address(bh->b_page); + clear_buffer_uptodate(bh); + } +#endif + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + release_stripe(sh); + return 0; +} + +static int raid6_end_write_request (struct bio *bi, unsigned int bytes_done, + int error) +{ + struct stripe_head *sh = bi->bi_private; + raid6_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks, i; + unsigned long flags; + int uptodate = test_bit(BIO_UPTODATE, &bi->bi_flags); + + if (bi->bi_size) + return 1; + + for (i=0 ; i<disks; i++) + if (bi == &sh->dev[i].req) + break; + + PRINTK("end_write_request %llu/%d, count %d, uptodate: %d.\n", + (unsigned long long)sh->sector, i, atomic_read(&sh->count), + uptodate); + if (i == disks) { + BUG(); + return 0; + } + + spin_lock_irqsave(&conf->device_lock, flags); + if (!uptodate) + md_error(conf->mddev, conf->disks[i].rdev); + + rdev_dec_pending(conf->disks[i].rdev, conf->mddev); + + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + __release_stripe(conf, sh); + spin_unlock_irqrestore(&conf->device_lock, flags); + return 0; +} + + +static sector_t compute_blocknr(struct stripe_head *sh, int i); + +static void raid6_build_block (struct stripe_head *sh, int i) +{ + struct r5dev *dev = &sh->dev[i]; + int pd_idx = sh->pd_idx; + int qd_idx = raid6_next_disk(pd_idx, sh->raid_conf->raid_disks); + + bio_init(&dev->req); + dev->req.bi_io_vec = &dev->vec; + dev->req.bi_vcnt++; + dev->req.bi_max_vecs++; + dev->vec.bv_page = dev->page; + dev->vec.bv_len = STRIPE_SIZE; + dev->vec.bv_offset = 0; + + dev->req.bi_sector = sh->sector; + dev->req.bi_private = sh; + + dev->flags = 0; + if (i != pd_idx && i != qd_idx) + dev->sector = compute_blocknr(sh, i); +} + +static void error(mddev_t *mddev, mdk_rdev_t *rdev) +{ + char b[BDEVNAME_SIZE]; + raid6_conf_t *conf = (raid6_conf_t *) mddev->private; + PRINTK("raid6: error called\n"); + + if (!rdev->faulty) { + mddev->sb_dirty = 1; + if (rdev->in_sync) { + conf->working_disks--; + mddev->degraded++; + conf->failed_disks++; + rdev->in_sync = 0; + /* + * if recovery was running, make sure it aborts. + */ + set_bit(MD_RECOVERY_ERR, &mddev->recovery); + } + rdev->faulty = 1; + printk (KERN_ALERT + "raid6: Disk failure on %s, disabling device." + " Operation continuing on %d devices\n", + bdevname(rdev->bdev,b), conf->working_disks); + } +} + +/* + * Input: a 'big' sector number, + * Output: index of the data and parity disk, and the sector # in them. + */ +static sector_t raid6_compute_sector(sector_t r_sector, unsigned int raid_disks, + unsigned int data_disks, unsigned int * dd_idx, + unsigned int * pd_idx, raid6_conf_t *conf) +{ + long stripe; + unsigned long chunk_number; + unsigned int chunk_offset; + sector_t new_sector; + int sectors_per_chunk = conf->chunk_size >> 9; + + /* First compute the information on this sector */ + + /* + * Compute the chunk number and the sector offset inside the chunk + */ + chunk_offset = sector_div(r_sector, sectors_per_chunk); + chunk_number = r_sector; + if ( r_sector != chunk_number ) { + printk(KERN_CRIT "raid6: ERROR: r_sector = %llu, chunk_number = %lu\n", + (unsigned long long)r_sector, (unsigned long)chunk_number); + BUG(); + } + + /* + * Compute the stripe number + */ + stripe = chunk_number / data_disks; + + /* + * Compute the data disk and parity disk indexes inside the stripe + */ + *dd_idx = chunk_number % data_disks; + + /* + * Select the parity disk based on the user selected algorithm. + */ + + /**** FIX THIS ****/ + switch (conf->algorithm) { + case ALGORITHM_LEFT_ASYMMETRIC: + *pd_idx = raid_disks - 1 - (stripe % raid_disks); + if (*pd_idx == raid_disks-1) + (*dd_idx)++; /* Q D D D P */ + else if (*dd_idx >= *pd_idx) + (*dd_idx) += 2; /* D D P Q D */ + break; + case ALGORITHM_RIGHT_ASYMMETRIC: + *pd_idx = stripe % raid_disks; + if (*pd_idx == raid_disks-1) + (*dd_idx)++; /* Q D D D P */ + else if (*dd_idx >= *pd_idx) + (*dd_idx) += 2; /* D D P Q D */ + break; + case ALGORITHM_LEFT_SYMMETRIC: + *pd_idx = raid_disks - 1 - (stripe % raid_disks); + *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; + break; + case ALGORITHM_RIGHT_SYMMETRIC: + *pd_idx = stripe % raid_disks; + *dd_idx = (*pd_idx + 2 + *dd_idx) % raid_disks; + break; + default: + printk (KERN_CRIT "raid6: unsupported algorithm %d\n", + conf->algorithm); + } + + PRINTK("raid6: chunk_number = %lu, pd_idx = %u, dd_idx = %u\n", + chunk_number, *pd_idx, *dd_idx); + + /* + * Finally, compute the new sector number + */ + new_sector = (sector_t) stripe * sectors_per_chunk + chunk_offset; + return new_sector; +} + + +static sector_t compute_blocknr(struct stripe_head *sh, int i) +{ + raid6_conf_t *conf = sh->raid_conf; + int raid_disks = conf->raid_disks, data_disks = raid_disks - 2; + sector_t new_sector = sh->sector, check; + int sectors_per_chunk = conf->chunk_size >> 9; + sector_t stripe; + int chunk_offset; + int chunk_number, dummy1, dummy2, dd_idx = i; + sector_t r_sector; + int i0 = i; + + chunk_offset = sector_div(new_sector, sectors_per_chunk); + stripe = new_sector; + if ( new_sector != stripe ) { + printk(KERN_CRIT "raid6: ERROR: new_sector = %llu, stripe = %lu\n", + (unsigned long long)new_sector, (unsigned long)stripe); + BUG(); + } + + switch (conf->algorithm) { + case ALGORITHM_LEFT_ASYMMETRIC: + case ALGORITHM_RIGHT_ASYMMETRIC: + if (sh->pd_idx == raid_disks-1) + i--; /* Q D D D P */ + else if (i > sh->pd_idx) + i -= 2; /* D D P Q D */ + break; + case ALGORITHM_LEFT_SYMMETRIC: + case ALGORITHM_RIGHT_SYMMETRIC: + if (sh->pd_idx == raid_disks-1) + i--; /* Q D D D P */ + else { + /* D D P Q D */ + if (i < sh->pd_idx) + i += raid_disks; + i -= (sh->pd_idx + 2); + } + break; + default: + printk (KERN_CRIT "raid6: unsupported algorithm %d\n", + conf->algorithm); + } + + PRINTK("raid6: compute_blocknr: pd_idx = %u, i0 = %u, i = %u\n", sh->pd_idx, i0, i); + + chunk_number = stripe * data_disks + i; + r_sector = (sector_t)chunk_number * sectors_per_chunk + chunk_offset; + + check = raid6_compute_sector (r_sector, raid_disks, data_disks, &dummy1, &dummy2, conf); + if (check != sh->sector || dummy1 != dd_idx || dummy2 != sh->pd_idx) { + printk(KERN_CRIT "raid6: compute_blocknr: map not correct\n"); + return 0; + } + return r_sector; +} + + + +/* + * Copy data between a page in the stripe cache, and one or more bion + * The page could align with the middle of the bio, or there could be + * several bion, each with several bio_vecs, which cover part of the page + * Multiple bion are linked together on bi_next. There may be extras + * at the end of this list. We ignore them. + */ +static void copy_data(int frombio, struct bio *bio, + struct page *page, + sector_t sector) +{ + char *pa = page_address(page); + struct bio_vec *bvl; + int i; + int page_offset; + + if (bio->bi_sector >= sector) + page_offset = (signed)(bio->bi_sector - sector) * 512; + else + page_offset = (signed)(sector - bio->bi_sector) * -512; + bio_for_each_segment(bvl, bio, i) { + int len = bio_iovec_idx(bio,i)->bv_len; + int clen; + int b_offset = 0; + + if (page_offset < 0) { + b_offset = -page_offset; + page_offset += b_offset; + len -= b_offset; + } + + if (len > 0 && page_offset + len > STRIPE_SIZE) + clen = STRIPE_SIZE - page_offset; + else clen = len; + + if (clen > 0) { + char *ba = __bio_kmap_atomic(bio, i, KM_USER0); + if (frombio) + memcpy(pa+page_offset, ba+b_offset, clen); + else + memcpy(ba+b_offset, pa+page_offset, clen); + __bio_kunmap_atomic(ba, KM_USER0); + } + if (clen < len) /* hit end of page */ + break; + page_offset += len; + } +} + +#define check_xor() do { \ + if (count == MAX_XOR_BLOCKS) { \ + xor_block(count, STRIPE_SIZE, ptr); \ + count = 1; \ + } \ + } while(0) + +/* Compute P and Q syndromes */ +static void compute_parity(struct stripe_head *sh, int method) +{ + raid6_conf_t *conf = sh->raid_conf; + int i, pd_idx = sh->pd_idx, qd_idx, d0_idx, disks = conf->raid_disks, count; + struct bio *chosen; + /**** FIX THIS: This could be very bad if disks is close to 256 ****/ + void *ptrs[disks]; + + qd_idx = raid6_next_disk(pd_idx, disks); + d0_idx = raid6_next_disk(qd_idx, disks); + + PRINTK("compute_parity, stripe %llu, method %d\n", + (unsigned long long)sh->sector, method); + + switch(method) { + case READ_MODIFY_WRITE: + BUG(); /* READ_MODIFY_WRITE N/A for RAID-6 */ + case RECONSTRUCT_WRITE: + for (i= disks; i-- ;) + if ( i != pd_idx && i != qd_idx && sh->dev[i].towrite ) { + chosen = sh->dev[i].towrite; + sh->dev[i].towrite = NULL; + + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + + if (sh->dev[i].written) BUG(); + sh->dev[i].written = chosen; + } + break; + case CHECK_PARITY: + BUG(); /* Not implemented yet */ + } + + for (i = disks; i--;) + if (sh->dev[i].written) { + sector_t sector = sh->dev[i].sector; + struct bio *wbi = sh->dev[i].written; + while (wbi && wbi->bi_sector < sector + STRIPE_SECTORS) { + copy_data(1, wbi, sh->dev[i].page, sector); + wbi = r5_next_bio(wbi, sector); + } + + set_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(R5_UPTODATE, &sh->dev[i].flags); + } + +// switch(method) { +// case RECONSTRUCT_WRITE: +// case CHECK_PARITY: +// case UPDATE_PARITY: + /* Note that unlike RAID-5, the ordering of the disks matters greatly. */ + /* FIX: Is this ordering of drives even remotely optimal? */ + count = 0; + i = d0_idx; + do { + ptrs[count++] = page_address(sh->dev[i].page); + if (count <= disks-2 && !test_bit(R5_UPTODATE, &sh->dev[i].flags)) + printk("block %d/%d not uptodate on parity calc\n", i,count); + i = raid6_next_disk(i, disks); + } while ( i != d0_idx ); +// break; +// } + + raid6_call.gen_syndrome(disks, STRIPE_SIZE, ptrs); + + switch(method) { + case RECONSTRUCT_WRITE: + set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); + set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); + set_bit(R5_LOCKED, &sh->dev[pd_idx].flags); + set_bit(R5_LOCKED, &sh->dev[qd_idx].flags); + break; + case UPDATE_PARITY: + set_bit(R5_UPTODATE, &sh->dev[pd_idx].flags); + set_bit(R5_UPTODATE, &sh->dev[qd_idx].flags); + break; + } +} + +/* Compute one missing block */ +static void compute_block_1(struct stripe_head *sh, int dd_idx) +{ + raid6_conf_t *conf = sh->raid_conf; + int i, count, disks = conf->raid_disks; + void *ptr[MAX_XOR_BLOCKS], *p; + int pd_idx = sh->pd_idx; + int qd_idx = raid6_next_disk(pd_idx, disks); + + PRINTK("compute_block_1, stripe %llu, idx %d\n", + (unsigned long long)sh->sector, dd_idx); + + if ( dd_idx == qd_idx ) { + /* We're actually computing the Q drive */ + compute_parity(sh, UPDATE_PARITY); + } else { + ptr[0] = page_address(sh->dev[dd_idx].page); + memset(ptr[0], 0, STRIPE_SIZE); + count = 1; + for (i = disks ; i--; ) { + if (i == dd_idx || i == qd_idx) + continue; + p = page_address(sh->dev[i].page); + if (test_bit(R5_UPTODATE, &sh->dev[i].flags)) + ptr[count++] = p; + else + printk("compute_block() %d, stripe %llu, %d" + " not present\n", dd_idx, + (unsigned long long)sh->sector, i); + + check_xor(); + } + if (count != 1) + xor_block(count, STRIPE_SIZE, ptr); + set_bit(R5_UPTODATE, &sh->dev[dd_idx].flags); + } +} + +/* Compute two missing blocks */ +static void compute_block_2(struct stripe_head *sh, int dd_idx1, int dd_idx2) +{ + raid6_conf_t *conf = sh->raid_conf; + int i, count, disks = conf->raid_disks; + int pd_idx = sh->pd_idx; + int qd_idx = raid6_next_disk(pd_idx, disks); + int d0_idx = raid6_next_disk(qd_idx, disks); + int faila, failb; + + /* faila and failb are disk numbers relative to d0_idx */ + /* pd_idx become disks-2 and qd_idx become disks-1 */ + faila = (dd_idx1 < d0_idx) ? dd_idx1+(disks-d0_idx) : dd_idx1-d0_idx; + failb = (dd_idx2 < d0_idx) ? dd_idx2+(disks-d0_idx) : dd_idx2-d0_idx; + + BUG_ON(faila == failb); + if ( failb < faila ) { int tmp = faila; faila = failb; failb = tmp; } + + PRINTK("compute_block_2, stripe %llu, idx %d,%d (%d,%d)\n", + (unsigned long long)sh->sector, dd_idx1, dd_idx2, faila, failb); + + if ( failb == disks-1 ) { + /* Q disk is one of the missing disks */ + if ( faila == disks-2 ) { + /* Missing P+Q, just recompute */ + compute_parity(sh, UPDATE_PARITY); + return; + } else { + /* We're missing D+Q; recompute D from P */ + compute_block_1(sh, (dd_idx1 == qd_idx) ? dd_idx2 : dd_idx1); + compute_parity(sh, UPDATE_PARITY); /* Is this necessary? */ + return; + } + } + + /* We're missing D+P or D+D; build pointer table */ + { + /**** FIX THIS: This could be very bad if disks is close to 256 ****/ + void *ptrs[disks]; + + count = 0; + i = d0_idx; + do { + ptrs[count++] = page_address(sh->dev[i].page); + i = raid6_next_disk(i, disks); + if (i != dd_idx1 && i != dd_idx2 && + !test_bit(R5_UPTODATE, &sh->dev[i].flags)) + printk("compute_2 with missing block %d/%d\n", count, i); + } while ( i != d0_idx ); + + if ( failb == disks-2 ) { + /* We're missing D+P. */ + raid6_datap_recov(disks, STRIPE_SIZE, faila, ptrs); + } else { + /* We're missing D+D. */ + raid6_2data_recov(disks, STRIPE_SIZE, faila, failb, ptrs); + } + + /* Both the above update both missing blocks */ + set_bit(R5_UPTODATE, &sh->dev[dd_idx1].flags); + set_bit(R5_UPTODATE, &sh->dev[dd_idx2].flags); + } +} + + +/* + * Each stripe/dev can have one or more bion attached. + * toread/towrite point to the first in a chain. + * The bi_next chain must be in order. + */ +static int add_stripe_bio(struct stripe_head *sh, struct bio *bi, int dd_idx, int forwrite) +{ + struct bio **bip; + raid6_conf_t *conf = sh->raid_conf; + + PRINTK("adding bh b#%llu to stripe s#%llu\n", + (unsigned long long)bi->bi_sector, + (unsigned long long)sh->sector); + + + spin_lock(&sh->lock); + spin_lock_irq(&conf->device_lock); + if (forwrite) + bip = &sh->dev[dd_idx].towrite; + else + bip = &sh->dev[dd_idx].toread; + while (*bip && (*bip)->bi_sector < bi->bi_sector) { + if ((*bip)->bi_sector + ((*bip)->bi_size >> 9) > bi->bi_sector) + goto overlap; + bip = &(*bip)->bi_next; + } + if (*bip && (*bip)->bi_sector < bi->bi_sector + ((bi->bi_size)>>9)) + goto overlap; + + if (*bip && bi->bi_next && (*bip) != bi->bi_next) + BUG(); + if (*bip) + bi->bi_next = *bip; + *bip = bi; + bi->bi_phys_segments ++; + spin_unlock_irq(&conf->device_lock); + spin_unlock(&sh->lock); + + PRINTK("added bi b#%llu to stripe s#%llu, disk %d.\n", + (unsigned long long)bi->bi_sector, + (unsigned long long)sh->sector, dd_idx); + + if (forwrite) { + /* check if page is covered */ + sector_t sector = sh->dev[dd_idx].sector; + for (bi=sh->dev[dd_idx].towrite; + sector < sh->dev[dd_idx].sector + STRIPE_SECTORS && + bi && bi->bi_sector <= sector; + bi = r5_next_bio(bi, sh->dev[dd_idx].sector)) { + if (bi->bi_sector + (bi->bi_size>>9) >= sector) + sector = bi->bi_sector + (bi->bi_size>>9); + } + if (sector >= sh->dev[dd_idx].sector + STRIPE_SECTORS) + set_bit(R5_OVERWRITE, &sh->dev[dd_idx].flags); + } + return 1; + + overlap: + set_bit(R5_Overlap, &sh->dev[dd_idx].flags); + spin_unlock_irq(&conf->device_lock); + spin_unlock(&sh->lock); + return 0; +} + + +/* + * handle_stripe - do things to a stripe. + * + * We lock the stripe and then examine the state of various bits + * to see what needs to be done. + * Possible results: + * return some read request which now have data + * return some write requests which are safely on disc + * schedule a read on some buffers + * schedule a write of some buffers + * return confirmation of parity correctness + * + * Parity calculations are done inside the stripe lock + * buffers are taken off read_list or write_list, and bh_cache buffers + * get BH_Lock set before the stripe lock is released. + * + */ + +static void handle_stripe(struct stripe_head *sh) +{ + raid6_conf_t *conf = sh->raid_conf; + int disks = conf->raid_disks; + struct bio *return_bi= NULL; + struct bio *bi; + int i; + int syncing; + int locked=0, uptodate=0, to_read=0, to_write=0, failed=0, written=0; + int non_overwrite = 0; + int failed_num[2] = {0, 0}; + struct r5dev *dev, *pdev, *qdev; + int pd_idx = sh->pd_idx; + int qd_idx = raid6_next_disk(pd_idx, disks); + int p_failed, q_failed; + + PRINTK("handling stripe %llu, state=%#lx cnt=%d, pd_idx=%d, qd_idx=%d\n", + (unsigned long long)sh->sector, sh->state, atomic_read(&sh->count), + pd_idx, qd_idx); + + spin_lock(&sh->lock); + clear_bit(STRIPE_HANDLE, &sh->state); + clear_bit(STRIPE_DELAYED, &sh->state); + + syncing = test_bit(STRIPE_SYNCING, &sh->state); + /* Now to look around and see what can be done */ + + for (i=disks; i--; ) { + mdk_rdev_t *rdev; + dev = &sh->dev[i]; + clear_bit(R5_Insync, &dev->flags); + clear_bit(R5_Syncio, &dev->flags); + + PRINTK("check %d: state 0x%lx read %p write %p written %p\n", + i, dev->flags, dev->toread, dev->towrite, dev->written); + /* maybe we can reply to a read */ + if (test_bit(R5_UPTODATE, &dev->flags) && dev->toread) { + struct bio *rbi, *rbi2; + PRINTK("Return read for disc %d\n", i); + spin_lock_irq(&conf->device_lock); + rbi = dev->toread; + dev->toread = NULL; + if (test_and_clear_bit(R5_Overlap, &dev->flags)) + wake_up(&conf->wait_for_overlap); + spin_unlock_irq(&conf->device_lock); + while (rbi && rbi->bi_sector < dev->sector + STRIPE_SECTORS) { + copy_data(0, rbi, dev->page, dev->sector); + rbi2 = r5_next_bio(rbi, dev->sector); + spin_lock_irq(&conf->device_lock); + if (--rbi->bi_phys_segments == 0) { + rbi->bi_next = return_bi; + return_bi = rbi; + } + spin_unlock_irq(&conf->device_lock); + rbi = rbi2; + } + } + + /* now count some things */ + if (test_bit(R5_LOCKED, &dev->flags)) locked++; + if (test_bit(R5_UPTODATE, &dev->flags)) uptodate++; + + + if (dev->toread) to_read++; + if (dev->towrite) { + to_write++; + if (!test_bit(R5_OVERWRITE, &dev->flags)) + non_overwrite++; + } + if (dev->written) written++; + rdev = conf->disks[i].rdev; /* FIXME, should I be looking rdev */ + if (!rdev || !rdev->in_sync) { + if ( failed < 2 ) + failed_num[failed] = i; + failed++; + } else + set_bit(R5_Insync, &dev->flags); + } + PRINTK("locked=%d uptodate=%d to_read=%d" + " to_write=%d failed=%d failed_num=%d,%d\n", + locked, uptodate, to_read, to_write, failed, + failed_num[0], failed_num[1]); + /* check if the array has lost >2 devices and, if so, some requests might + * need to be failed + */ + if (failed > 2 && to_read+to_write+written) { + spin_lock_irq(&conf->device_lock); + for (i=disks; i--; ) { + /* fail all writes first */ + bi = sh->dev[i].towrite; + sh->dev[i].towrite = NULL; + if (bi) to_write--; + + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ + struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + bi->bi_next = return_bi; + return_bi = bi; + } + bi = nextbi; + } + /* and fail all 'written' */ + bi = sh->dev[i].written; + sh->dev[i].written = NULL; + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS) { + struct bio *bi2 = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + bi->bi_next = return_bi; + return_bi = bi; + } + bi = bi2; + } + + /* fail any reads if this device is non-operational */ + if (!test_bit(R5_Insync, &sh->dev[i].flags)) { + bi = sh->dev[i].toread; + sh->dev[i].toread = NULL; + if (test_and_clear_bit(R5_Overlap, &sh->dev[i].flags)) + wake_up(&conf->wait_for_overlap); + if (bi) to_read--; + while (bi && bi->bi_sector < sh->dev[i].sector + STRIPE_SECTORS){ + struct bio *nextbi = r5_next_bio(bi, sh->dev[i].sector); + clear_bit(BIO_UPTODATE, &bi->bi_flags); + if (--bi->bi_phys_segments == 0) { + bi->bi_next = return_bi; + return_bi = bi; + } + bi = nextbi; + } + } + } + spin_unlock_irq(&conf->device_lock); + } + if (failed > 2 && syncing) { + md_done_sync(conf->mddev, STRIPE_SECTORS,0); + clear_bit(STRIPE_SYNCING, &sh->state); + syncing = 0; + } + + /* + * might be able to return some write requests if the parity blocks + * are safe, or on a failed drive + */ + pdev = &sh->dev[pd_idx]; + p_failed = (failed >= 1 && failed_num[0] == pd_idx) + || (failed >= 2 && failed_num[1] == pd_idx); + qdev = &sh->dev[qd_idx]; + q_failed = (failed >= 1 && failed_num[0] == qd_idx) + || (failed >= 2 && failed_num[1] == qd_idx); + + if ( written && + ( p_failed || ((test_bit(R5_Insync, &pdev->flags) + && !test_bit(R5_LOCKED, &pdev->flags) + && test_bit(R5_UPTODATE, &pdev->flags))) ) && + ( q_failed || ((test_bit(R5_Insync, &qdev->flags) + && !test_bit(R5_LOCKED, &qdev->flags) + && test_bit(R5_UPTODATE, &qdev->flags))) ) ) { + /* any written block on an uptodate or failed drive can be + * returned. Note that if we 'wrote' to a failed drive, + * it will be UPTODATE, but never LOCKED, so we don't need + * to test 'failed' directly. + */ + for (i=disks; i--; ) + if (sh->dev[i].written) { + dev = &sh->dev[i]; + if (!test_bit(R5_LOCKED, &dev->flags) && + test_bit(R5_UPTODATE, &dev->flags) ) { + /* We can return any write requests */ + struct bio *wbi, *wbi2; + PRINTK("Return write for stripe %llu disc %d\n", + (unsigned long long)sh->sector, i); + spin_lock_irq(&conf->device_lock); + wbi = dev->written; + dev->written = NULL; + while (wbi && wbi->bi_sector < dev->sector + STRIPE_SECTORS) { + wbi2 = r5_next_bio(wbi, dev->sector); + if (--wbi->bi_phys_segments == 0) { + md_write_end(conf->mddev); + wbi->bi_next = return_bi; + return_bi = wbi; + } + wbi = wbi2; + } + spin_unlock_irq(&conf->device_lock); + } + } + } + + /* Now we might consider reading some blocks, either to check/generate + * parity, or to satisfy requests + * or to load a block that is being partially written. + */ + if (to_read || non_overwrite || (to_write && failed) || (syncing && (uptodate < disks))) { + for (i=disks; i--;) { + dev = &sh->dev[i]; + if (!test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && + (dev->toread || + (dev->towrite && !test_bit(R5_OVERWRITE, &dev->flags)) || + syncing || + (failed >= 1 && (sh->dev[failed_num[0]].toread || to_write)) || + (failed >= 2 && (sh->dev[failed_num[1]].toread || to_write)) + ) + ) { + /* we would like to get this block, possibly + * by computing it, but we might not be able to + */ + if (uptodate == disks-1) { + PRINTK("Computing stripe %llu block %d\n", + (unsigned long long)sh->sector, i); + compute_block_1(sh, i); + uptodate++; + } else if ( uptodate == disks-2 && failed >= 2 ) { + /* Computing 2-failure is *very* expensive; only do it if failed >= 2 */ + int other; + for (other=disks; other--;) { + if ( other == i ) + continue; + if ( !test_bit(R5_UPTODATE, &sh->dev[other].flags) ) + break; + } + BUG_ON(other < 0); + PRINTK("Computing stripe %llu blocks %d,%d\n", + (unsigned long long)sh->sector, i, other); + compute_block_2(sh, i, other); + uptodate += 2; + } else if (test_bit(R5_Insync, &dev->flags)) { + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantread, &dev->flags); +#if 0 + /* if I am just reading this block and we don't have + a failed drive, or any pending writes then sidestep the cache */ + if (sh->bh_read[i] && !sh->bh_read[i]->b_reqnext && + ! syncing && !failed && !to_write) { + sh->bh_cache[i]->b_page = sh->bh_read[i]->b_page; + sh->bh_cache[i]->b_data = sh->bh_read[i]->b_data; + } +#endif + locked++; + PRINTK("Reading block %d (sync=%d)\n", + i, syncing); + if (syncing) + md_sync_acct(conf->disks[i].rdev->bdev, + STRIPE_SECTORS); + } + } + } + set_bit(STRIPE_HANDLE, &sh->state); + } + + /* now to consider writing and what else, if anything should be read */ + if (to_write) { + int rcw=0, must_compute=0; + for (i=disks ; i--;) { + dev = &sh->dev[i]; + /* Would I have to read this buffer for reconstruct_write */ + if (!test_bit(R5_OVERWRITE, &dev->flags) + && i != pd_idx && i != qd_idx + && (!test_bit(R5_LOCKED, &dev->flags) +#if 0 + || sh->bh_page[i] != bh->b_page +#endif + ) && + !test_bit(R5_UPTODATE, &dev->flags)) { + if (test_bit(R5_Insync, &dev->flags)) rcw++; + else { + PRINTK("raid6: must_compute: disk %d flags=%#lx\n", i, dev->flags); + must_compute++; + } + } + } + PRINTK("for sector %llu, rcw=%d, must_compute=%d\n", + (unsigned long long)sh->sector, rcw, must_compute); + set_bit(STRIPE_HANDLE, &sh->state); + + if (rcw > 0) + /* want reconstruct write, but need to get some data */ + for (i=disks; i--;) { + dev = &sh->dev[i]; + if (!test_bit(R5_OVERWRITE, &dev->flags) + && !(failed == 0 && (i == pd_idx || i == qd_idx)) + && !test_bit(R5_LOCKED, &dev->flags) && !test_bit(R5_UPTODATE, &dev->flags) && + test_bit(R5_Insync, &dev->flags)) { + if (test_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + { + PRINTK("Read_old stripe %llu block %d for Reconstruct\n", + (unsigned long long)sh->sector, i); + set_bit(R5_LOCKED, &dev->flags); + set_bit(R5_Wantread, &dev->flags); + locked++; + } else { + PRINTK("Request delayed stripe %llu block %d for Reconstruct\n", + (unsigned long long)sh->sector, i); + set_bit(STRIPE_DELAYED, &sh->state); + set_bit(STRIPE_HANDLE, &sh->state); + } + } + } + /* now if nothing is locked, and if we have enough data, we can start a write request */ + if (locked == 0 && rcw == 0) { + if ( must_compute > 0 ) { + /* We have failed blocks and need to compute them */ + switch ( failed ) { + case 0: BUG(); + case 1: compute_block_1(sh, failed_num[0]); break; + case 2: compute_block_2(sh, failed_num[0], failed_num[1]); break; + default: BUG(); /* This request should have been failed? */ + } + } + + PRINTK("Computing parity for stripe %llu\n", (unsigned long long)sh->sector); + compute_parity(sh, RECONSTRUCT_WRITE); + /* now every locked buffer is ready to be written */ + for (i=disks; i--;) + if (test_bit(R5_LOCKED, &sh->dev[i].flags)) { + PRINTK("Writing stripe %llu block %d\n", + (unsigned long long)sh->sector, i); + locked++; + set_bit(R5_Wantwrite, &sh->dev[i].flags); +#if 0 /**** FIX: I don't understand the logic here... ****/ + if (!test_bit(R5_Insync, &sh->dev[i].flags) + || ((i==pd_idx || i==qd_idx) && failed == 0)) /* FIX? */ + set_bit(STRIPE_INSYNC, &sh->state); +#endif + } + if (test_and_clear_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) { + atomic_dec(&conf->preread_active_stripes); + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) + md_wakeup_thread(conf->mddev->thread); + } + } + } + + /* maybe we need to check and possibly fix the parity for this stripe + * Any reads will already have been scheduled, so we just see if enough data + * is available + */ + if (syncing && locked == 0 && + !test_bit(STRIPE_INSYNC, &sh->state) && failed <= 2) { + set_bit(STRIPE_HANDLE, &sh->state); +#if 0 /* RAID-6: Don't support CHECK PARITY yet */ + if (failed == 0) { + char *pagea; + if (uptodate != disks) + BUG(); + compute_parity(sh, CHECK_PARITY); + uptodate--; + pagea = page_address(sh->dev[pd_idx].page); + if ((*(u32*)pagea) == 0 && + !memcmp(pagea, pagea+4, STRIPE_SIZE-4)) { + /* parity is correct (on disc, not in buffer any more) */ + set_bit(STRIPE_INSYNC, &sh->state); + } + } +#endif + if (!test_bit(STRIPE_INSYNC, &sh->state)) { + int failed_needupdate[2]; + struct r5dev *adev, *bdev; + + if ( failed < 1 ) + failed_num[0] = pd_idx; + if ( failed < 2 ) + failed_num[1] = (failed_num[0] == qd_idx) ? pd_idx : qd_idx; + + failed_needupdate[0] = !test_bit(R5_UPTODATE, &sh->dev[failed_num[0]].flags); + failed_needupdate[1] = !test_bit(R5_UPTODATE, &sh->dev[failed_num[1]].flags); + + PRINTK("sync: failed=%d num=%d,%d fnu=%u%u\n", + failed, failed_num[0], failed_num[1], failed_needupdate[0], failed_needupdate[1]); + +#if 0 /* RAID-6: This code seems to require that CHECK_PARITY destroys the uptodateness of the parity */ + /* should be able to compute the missing block(s) and write to spare */ + if ( failed_needupdate[0] ^ failed_needupdate[1] ) { + if (uptodate+1 != disks) + BUG(); + compute_block_1(sh, failed_needupdate[0] ? failed_num[0] : failed_num[1]); + uptodate++; + } else if ( failed_needupdate[0] & failed_needupdate[1] ) { + if (uptodate+2 != disks) + BUG(); + compute_block_2(sh, failed_num[0], failed_num[1]); + uptodate += 2; + } +#else + compute_block_2(sh, failed_num[0], failed_num[1]); + uptodate += failed_needupdate[0] + failed_needupdate[1]; +#endif + + if (uptodate != disks) + BUG(); + + PRINTK("Marking for sync stripe %llu blocks %d,%d\n", + (unsigned long long)sh->sector, failed_num[0], failed_num[1]); + + /**** FIX: Should we really do both of these unconditionally? ****/ + adev = &sh->dev[failed_num[0]]; + locked += !test_bit(R5_LOCKED, &adev->flags); + set_bit(R5_LOCKED, &adev->flags); + set_bit(R5_Wantwrite, &adev->flags); + bdev = &sh->dev[failed_num[1]]; + locked += !test_bit(R5_LOCKED, &bdev->flags); + set_bit(R5_LOCKED, &bdev->flags); + set_bit(R5_Wantwrite, &bdev->flags); + + set_bit(STRIPE_INSYNC, &sh->state); + set_bit(R5_Syncio, &adev->flags); + set_bit(R5_Syncio, &bdev->flags); + } + } + if (syncing && locked == 0 && test_bit(STRIPE_INSYNC, &sh->state)) { + md_done_sync(conf->mddev, STRIPE_SECTORS,1); + clear_bit(STRIPE_SYNCING, &sh->state); + } + + spin_unlock(&sh->lock); + + while ((bi=return_bi)) { + int bytes = bi->bi_size; + + return_bi = bi->bi_next; + bi->bi_next = NULL; + bi->bi_size = 0; + bi->bi_end_io(bi, bytes, 0); + } + for (i=disks; i-- ;) { + int rw; + struct bio *bi; + mdk_rdev_t *rdev; + if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) + rw = 1; + else if (test_and_clear_bit(R5_Wantread, &sh->dev[i].flags)) + rw = 0; + else + continue; + + bi = &sh->dev[i].req; + + bi->bi_rw = rw; + if (rw) + bi->bi_end_io = raid6_end_write_request; + else + bi->bi_end_io = raid6_end_read_request; + + rcu_read_lock(); + rdev = conf->disks[i].rdev; + if (rdev && rdev->faulty) + rdev = NULL; + if (rdev) + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (rdev) { + if (test_bit(R5_Syncio, &sh->dev[i].flags)) + md_sync_acct(rdev->bdev, STRIPE_SECTORS); + + bi->bi_bdev = rdev->bdev; + PRINTK("for %llu schedule op %ld on disc %d\n", + (unsigned long long)sh->sector, bi->bi_rw, i); + atomic_inc(&sh->count); + bi->bi_sector = sh->sector + rdev->data_offset; + bi->bi_flags = 1 << BIO_UPTODATE; + bi->bi_vcnt = 1; + bi->bi_max_vecs = 1; + bi->bi_idx = 0; + bi->bi_io_vec = &sh->dev[i].vec; + bi->bi_io_vec[0].bv_len = STRIPE_SIZE; + bi->bi_io_vec[0].bv_offset = 0; + bi->bi_size = STRIPE_SIZE; + bi->bi_next = NULL; + generic_make_request(bi); + } else { + PRINTK("skip op %ld on disc %d for sector %llu\n", + bi->bi_rw, i, (unsigned long long)sh->sector); + clear_bit(R5_LOCKED, &sh->dev[i].flags); + set_bit(STRIPE_HANDLE, &sh->state); + } + } +} + +static inline void raid6_activate_delayed(raid6_conf_t *conf) +{ + if (atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD) { + while (!list_empty(&conf->delayed_list)) { + struct list_head *l = conf->delayed_list.next; + struct stripe_head *sh; + sh = list_entry(l, struct stripe_head, lru); + list_del_init(l); + clear_bit(STRIPE_DELAYED, &sh->state); + if (!test_and_set_bit(STRIPE_PREREAD_ACTIVE, &sh->state)) + atomic_inc(&conf->preread_active_stripes); + list_add_tail(&sh->lru, &conf->handle_list); + } + } +} + +static void unplug_slaves(mddev_t *mddev) +{ + raid6_conf_t *conf = mddev_to_conf(mddev); + int i; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks; i++) { + mdk_rdev_t *rdev = conf->disks[i].rdev; + if (rdev && !rdev->faulty && atomic_read(&rdev->nr_pending)) { + request_queue_t *r_queue = bdev_get_queue(rdev->bdev); + + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + + if (r_queue->unplug_fn) + r_queue->unplug_fn(r_queue); + + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + rcu_read_unlock(); +} + +static void raid6_unplug_device(request_queue_t *q) +{ + mddev_t *mddev = q->queuedata; + raid6_conf_t *conf = mddev_to_conf(mddev); + unsigned long flags; + + spin_lock_irqsave(&conf->device_lock, flags); + + if (blk_remove_plug(q)) + raid6_activate_delayed(conf); + md_wakeup_thread(mddev->thread); + + spin_unlock_irqrestore(&conf->device_lock, flags); + + unplug_slaves(mddev); +} + +static int raid6_issue_flush(request_queue_t *q, struct gendisk *disk, + sector_t *error_sector) +{ + mddev_t *mddev = q->queuedata; + raid6_conf_t *conf = mddev_to_conf(mddev); + int i, ret = 0; + + rcu_read_lock(); + for (i=0; i<mddev->raid_disks && ret == 0; i++) { + mdk_rdev_t *rdev = conf->disks[i].rdev; + if (rdev && !rdev->faulty) { + struct block_device *bdev = rdev->bdev; + request_queue_t *r_queue = bdev_get_queue(bdev); + + if (!r_queue->issue_flush_fn) + ret = -EOPNOTSUPP; + else { + atomic_inc(&rdev->nr_pending); + rcu_read_unlock(); + ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, + error_sector); + rdev_dec_pending(rdev, mddev); + rcu_read_lock(); + } + } + } + rcu_read_unlock(); + return ret; +} + +static inline void raid6_plug_device(raid6_conf_t *conf) +{ + spin_lock_irq(&conf->device_lock); + blk_plug_device(conf->mddev->queue); + spin_unlock_irq(&conf->device_lock); +} + +static int make_request (request_queue_t *q, struct bio * bi) +{ + mddev_t *mddev = q->queuedata; + raid6_conf_t *conf = mddev_to_conf(mddev); + const unsigned int raid_disks = conf->raid_disks; + const unsigned int data_disks = raid_disks - 2; + unsigned int dd_idx, pd_idx; + sector_t new_sector; + sector_t logical_sector, last_sector; + struct stripe_head *sh; + + if (bio_data_dir(bi)==WRITE) { + disk_stat_inc(mddev->gendisk, writes); + disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bi)); + } else { + disk_stat_inc(mddev->gendisk, reads); + disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bi)); + } + + logical_sector = bi->bi_sector & ~((sector_t)STRIPE_SECTORS-1); + last_sector = bi->bi_sector + (bi->bi_size>>9); + + bi->bi_next = NULL; + bi->bi_phys_segments = 1; /* over-loaded to count active stripes */ + if ( bio_data_dir(bi) == WRITE ) + md_write_start(mddev); + for (;logical_sector < last_sector; logical_sector += STRIPE_SECTORS) { + DEFINE_WAIT(w); + + new_sector = raid6_compute_sector(logical_sector, + raid_disks, data_disks, &dd_idx, &pd_idx, conf); + + PRINTK("raid6: make_request, sector %llu logical %llu\n", + (unsigned long long)new_sector, + (unsigned long long)logical_sector); + + retry: + prepare_to_wait(&conf->wait_for_overlap, &w, TASK_UNINTERRUPTIBLE); + sh = get_active_stripe(conf, new_sector, pd_idx, (bi->bi_rw&RWA_MASK)); + if (sh) { + if (!add_stripe_bio(sh, bi, dd_idx, (bi->bi_rw&RW_MASK))) { + /* Add failed due to overlap. Flush everything + * and wait a while + */ + raid6_unplug_device(mddev->queue); + release_stripe(sh); + schedule(); + goto retry; + } + finish_wait(&conf->wait_for_overlap, &w); + raid6_plug_device(conf); + handle_stripe(sh); + release_stripe(sh); + } else { + /* cannot get stripe for read-ahead, just give-up */ + clear_bit(BIO_UPTODATE, &bi->bi_flags); + finish_wait(&conf->wait_for_overlap, &w); + break; + } + + } + spin_lock_irq(&conf->device_lock); + if (--bi->bi_phys_segments == 0) { + int bytes = bi->bi_size; + + if ( bio_data_dir(bi) == WRITE ) + md_write_end(mddev); + bi->bi_size = 0; + bi->bi_end_io(bi, bytes, 0); + } + spin_unlock_irq(&conf->device_lock); + return 0; +} + +/* FIXME go_faster isn't used */ +static int sync_request (mddev_t *mddev, sector_t sector_nr, int go_faster) +{ + raid6_conf_t *conf = (raid6_conf_t *) mddev->private; + struct stripe_head *sh; + int sectors_per_chunk = conf->chunk_size >> 9; + sector_t x; + unsigned long stripe; + int chunk_offset; + int dd_idx, pd_idx; + sector_t first_sector; + int raid_disks = conf->raid_disks; + int data_disks = raid_disks - 2; + + if (sector_nr >= mddev->size <<1) { + /* just being told to finish up .. nothing much to do */ + unplug_slaves(mddev); + return 0; + } + /* if there are 2 or more failed drives and we are trying + * to resync, then assert that we are finished, because there is + * nothing we can do. + */ + if (mddev->degraded >= 2 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { + int rv = (mddev->size << 1) - sector_nr; + md_done_sync(mddev, rv, 1); + return rv; + } + + x = sector_nr; + chunk_offset = sector_div(x, sectors_per_chunk); + stripe = x; + BUG_ON(x != stripe); + + first_sector = raid6_compute_sector((sector_t)stripe*data_disks*sectors_per_chunk + + chunk_offset, raid_disks, data_disks, &dd_idx, &pd_idx, conf); + sh = get_active_stripe(conf, sector_nr, pd_idx, 1); + if (sh == NULL) { + sh = get_active_stripe(conf, sector_nr, pd_idx, 0); + /* make sure we don't swamp the stripe cache if someone else + * is trying to get access + */ + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(1); + } + spin_lock(&sh->lock); + set_bit(STRIPE_SYNCING, &sh->state); + clear_bit(STRIPE_INSYNC, &sh->state); + spin_unlock(&sh->lock); + + handle_stripe(sh); + release_stripe(sh); + + return STRIPE_SECTORS; +} + +/* + * This is our raid6 kernel thread. + * + * We scan the hash table for stripes which can be handled now. + * During the scan, completed stripes are saved for us by the interrupt + * handler, so that they will not have to wait for our next wakeup. + */ +static void raid6d (mddev_t *mddev) +{ + struct stripe_head *sh; + raid6_conf_t *conf = mddev_to_conf(mddev); + int handled; + + PRINTK("+++ raid6d active\n"); + + md_check_recovery(mddev); + md_handle_safemode(mddev); + + handled = 0; + spin_lock_irq(&conf->device_lock); + while (1) { + struct list_head *first; + + if (list_empty(&conf->handle_list) && + atomic_read(&conf->preread_active_stripes) < IO_THRESHOLD && + !blk_queue_plugged(mddev->queue) && + !list_empty(&conf->delayed_list)) + raid6_activate_delayed(conf); + + if (list_empty(&conf->handle_list)) + break; + + first = conf->handle_list.next; + sh = list_entry(first, struct stripe_head, lru); + + list_del_init(first); + atomic_inc(&sh->count); + if (atomic_read(&sh->count)!= 1) + BUG(); + spin_unlock_irq(&conf->device_lock); + + handled++; + handle_stripe(sh); + release_stripe(sh); + + spin_lock_irq(&conf->device_lock); + } + PRINTK("%d stripes handled\n", handled); + + spin_unlock_irq(&conf->device_lock); + + unplug_slaves(mddev); + + PRINTK("--- raid6d inactive\n"); +} + +static int run (mddev_t *mddev) +{ + raid6_conf_t *conf; + int raid_disk, memory; + mdk_rdev_t *rdev; + struct disk_info *disk; + struct list_head *tmp; + + if (mddev->level != 6) { + PRINTK("raid6: %s: raid level not set to 6 (%d)\n", mdname(mddev), mddev->level); + return -EIO; + } + + mddev->private = kmalloc (sizeof (raid6_conf_t) + + mddev->raid_disks * sizeof(struct disk_info), + GFP_KERNEL); + if ((conf = mddev->private) == NULL) + goto abort; + memset (conf, 0, sizeof (*conf) + mddev->raid_disks * sizeof(struct disk_info) ); + conf->mddev = mddev; + + if ((conf->stripe_hashtbl = (struct stripe_head **) __get_free_pages(GFP_ATOMIC, HASH_PAGES_ORDER)) == NULL) + goto abort; + memset(conf->stripe_hashtbl, 0, HASH_PAGES * PAGE_SIZE); + + spin_lock_init(&conf->device_lock); + init_waitqueue_head(&conf->wait_for_stripe); + init_waitqueue_head(&conf->wait_for_overlap); + INIT_LIST_HEAD(&conf->handle_list); + INIT_LIST_HEAD(&conf->delayed_list); + INIT_LIST_HEAD(&conf->inactive_list); + atomic_set(&conf->active_stripes, 0); + atomic_set(&conf->preread_active_stripes, 0); + + mddev->queue->unplug_fn = raid6_unplug_device; + mddev->queue->issue_flush_fn = raid6_issue_flush; + + PRINTK("raid6: run(%s) called.\n", mdname(mddev)); + + ITERATE_RDEV(mddev,rdev,tmp) { + raid_disk = rdev->raid_disk; + if (raid_disk >= mddev->raid_disks + || raid_disk < 0) + continue; + disk = conf->disks + raid_disk; + + disk->rdev = rdev; + + if (rdev->in_sync) { + char b[BDEVNAME_SIZE]; + printk(KERN_INFO "raid6: device %s operational as raid" + " disk %d\n", bdevname(rdev->bdev,b), + raid_disk); + conf->working_disks++; + } + } + + conf->raid_disks = mddev->raid_disks; + + /* + * 0 for a fully functional array, 1 or 2 for a degraded array. + */ + mddev->degraded = conf->failed_disks = conf->raid_disks - conf->working_disks; + conf->mddev = mddev; + conf->chunk_size = mddev->chunk_size; + conf->level = mddev->level; + conf->algorithm = mddev->layout; + conf->max_nr_stripes = NR_STRIPES; + + /* device size must be a multiple of chunk size */ + mddev->size &= ~(mddev->chunk_size/1024 -1); + + if (conf->raid_disks < 4) { + printk(KERN_ERR "raid6: not enough configured devices for %s (%d, minimum 4)\n", + mdname(mddev), conf->raid_disks); + goto abort; + } + if (!conf->chunk_size || conf->chunk_size % 4) { + printk(KERN_ERR "raid6: invalid chunk size %d for %s\n", + conf->chunk_size, mdname(mddev)); + goto abort; + } + if (conf->algorithm > ALGORITHM_RIGHT_SYMMETRIC) { + printk(KERN_ERR + "raid6: unsupported parity algorithm %d for %s\n", + conf->algorithm, mdname(mddev)); + goto abort; + } + if (mddev->degraded > 2) { + printk(KERN_ERR "raid6: not enough operational devices for %s" + " (%d/%d failed)\n", + mdname(mddev), conf->failed_disks, conf->raid_disks); + goto abort; + } + +#if 0 /* FIX: For now */ + if (mddev->degraded > 0 && + mddev->recovery_cp != MaxSector) { + printk(KERN_ERR "raid6: cannot start dirty degraded array for %s\n", mdname(mddev)); + goto abort; + } +#endif + + { + mddev->thread = md_register_thread(raid6d, mddev, "%s_raid6"); + if (!mddev->thread) { + printk(KERN_ERR + "raid6: couldn't allocate thread for %s\n", + mdname(mddev)); + goto abort; + } + } + + memory = conf->max_nr_stripes * (sizeof(struct stripe_head) + + conf->raid_disks * ((sizeof(struct bio) + PAGE_SIZE))) / 1024; + if (grow_stripes(conf, conf->max_nr_stripes)) { + printk(KERN_ERR + "raid6: couldn't allocate %dkB for buffers\n", memory); + shrink_stripes(conf); + md_unregister_thread(mddev->thread); + goto abort; + } else + printk(KERN_INFO "raid6: allocated %dkB for %s\n", + memory, mdname(mddev)); + + if (mddev->degraded == 0) + printk(KERN_INFO "raid6: raid level %d set %s active with %d out of %d" + " devices, algorithm %d\n", conf->level, mdname(mddev), + mddev->raid_disks-mddev->degraded, mddev->raid_disks, + conf->algorithm); + else + printk(KERN_ALERT "raid6: raid level %d set %s active with %d" + " out of %d devices, algorithm %d\n", conf->level, + mdname(mddev), mddev->raid_disks - mddev->degraded, + mddev->raid_disks, conf->algorithm); + + print_raid6_conf(conf); + + /* read-ahead size must cover two whole stripes, which is + * 2 * (n-2) * chunksize where 'n' is the number of raid devices + */ + { + int stripe = (mddev->raid_disks-2) * mddev->chunk_size + / PAGE_CACHE_SIZE; + if (mddev->queue->backing_dev_info.ra_pages < 2 * stripe) + mddev->queue->backing_dev_info.ra_pages = 2 * stripe; + } + + /* Ok, everything is just fine now */ + mddev->array_size = mddev->size * (mddev->raid_disks - 2); + return 0; +abort: + if (conf) { + print_raid6_conf(conf); + if (conf->stripe_hashtbl) + free_pages((unsigned long) conf->stripe_hashtbl, + HASH_PAGES_ORDER); + kfree(conf); + } + mddev->private = NULL; + printk(KERN_ALERT "raid6: failed to run raid set %s\n", mdname(mddev)); + return -EIO; +} + + + +static int stop (mddev_t *mddev) +{ + raid6_conf_t *conf = (raid6_conf_t *) mddev->private; + + md_unregister_thread(mddev->thread); + mddev->thread = NULL; + shrink_stripes(conf); + free_pages((unsigned long) conf->stripe_hashtbl, HASH_PAGES_ORDER); + blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/ + kfree(conf); + mddev->private = NULL; + return 0; +} + +#if RAID6_DUMPSTATE +static void print_sh (struct seq_file *seq, struct stripe_head *sh) +{ + int i; + + seq_printf(seq, "sh %llu, pd_idx %d, state %ld.\n", + (unsigned long long)sh->sector, sh->pd_idx, sh->state); + seq_printf(seq, "sh %llu, count %d.\n", + (unsigned long long)sh->sector, atomic_read(&sh->count)); + seq_printf(seq, "sh %llu, ", (unsigned long long)sh->sector); + for (i = 0; i < sh->raid_conf->raid_disks; i++) { + seq_printf(seq, "(cache%d: %p %ld) ", + i, sh->dev[i].page, sh->dev[i].flags); + } + seq_printf(seq, "\n"); +} + +static void printall (struct seq_file *seq, raid6_conf_t *conf) +{ + struct stripe_head *sh; + int i; + + spin_lock_irq(&conf->device_lock); + for (i = 0; i < NR_HASH; i++) { + sh = conf->stripe_hashtbl[i]; + for (; sh; sh = sh->hash_next) { + if (sh->raid_conf != conf) + continue; + print_sh(seq, sh); + } + } + spin_unlock_irq(&conf->device_lock); +} +#endif + +static void status (struct seq_file *seq, mddev_t *mddev) +{ + raid6_conf_t *conf = (raid6_conf_t *) mddev->private; + int i; + + seq_printf (seq, " level %d, %dk chunk, algorithm %d", mddev->level, mddev->chunk_size >> 10, mddev->layout); + seq_printf (seq, " [%d/%d] [", conf->raid_disks, conf->working_disks); + for (i = 0; i < conf->raid_disks; i++) + seq_printf (seq, "%s", + conf->disks[i].rdev && + conf->disks[i].rdev->in_sync ? "U" : "_"); + seq_printf (seq, "]"); +#if RAID6_DUMPSTATE + seq_printf (seq, "\n"); + printall(seq, conf); +#endif +} + +static void print_raid6_conf (raid6_conf_t *conf) +{ + int i; + struct disk_info *tmp; + + printk("RAID6 conf printout:\n"); + if (!conf) { + printk("(conf==NULL)\n"); + return; + } + printk(" --- rd:%d wd:%d fd:%d\n", conf->raid_disks, + conf->working_disks, conf->failed_disks); + + for (i = 0; i < conf->raid_disks; i++) { + char b[BDEVNAME_SIZE]; + tmp = conf->disks + i; + if (tmp->rdev) + printk(" disk %d, o:%d, dev:%s\n", + i, !tmp->rdev->faulty, + bdevname(tmp->rdev->bdev,b)); + } +} + +static int raid6_spare_active(mddev_t *mddev) +{ + int i; + raid6_conf_t *conf = mddev->private; + struct disk_info *tmp; + + for (i = 0; i < conf->raid_disks; i++) { + tmp = conf->disks + i; + if (tmp->rdev + && !tmp->rdev->faulty + && !tmp->rdev->in_sync) { + mddev->degraded--; + conf->failed_disks--; + conf->working_disks++; + tmp->rdev->in_sync = 1; + } + } + print_raid6_conf(conf); + return 0; +} + +static int raid6_remove_disk(mddev_t *mddev, int number) +{ + raid6_conf_t *conf = mddev->private; + int err = 0; + mdk_rdev_t *rdev; + struct disk_info *p = conf->disks + number; + + print_raid6_conf(conf); + rdev = p->rdev; + if (rdev) { + if (rdev->in_sync || + atomic_read(&rdev->nr_pending)) { + err = -EBUSY; + goto abort; + } + p->rdev = NULL; + synchronize_kernel(); + if (atomic_read(&rdev->nr_pending)) { + /* lost the race, try later */ + err = -EBUSY; + p->rdev = rdev; + } + } + +abort: + + print_raid6_conf(conf); + return err; +} + +static int raid6_add_disk(mddev_t *mddev, mdk_rdev_t *rdev) +{ + raid6_conf_t *conf = mddev->private; + int found = 0; + int disk; + struct disk_info *p; + + if (mddev->degraded > 2) + /* no point adding a device */ + return 0; + /* + * find the disk ... + */ + for (disk=0; disk < mddev->raid_disks; disk++) + if ((p=conf->disks + disk)->rdev == NULL) { + rdev->in_sync = 0; + rdev->raid_disk = disk; + found = 1; + p->rdev = rdev; + break; + } + print_raid6_conf(conf); + return found; +} + +static int raid6_resize(mddev_t *mddev, sector_t sectors) +{ + /* no resync is happening, and there is enough space + * on all devices, so we can resize. + * We need to make sure resync covers any new space. + * If the array is shrinking we should possibly wait until + * any io in the removed space completes, but it hardly seems + * worth it. + */ + sectors &= ~((sector_t)mddev->chunk_size/512 - 1); + mddev->array_size = (sectors * (mddev->raid_disks-2))>>1; + set_capacity(mddev->gendisk, mddev->array_size << 1); + mddev->changed = 1; + if (sectors/2 > mddev->size && mddev->recovery_cp == MaxSector) { + mddev->recovery_cp = mddev->size << 1; + set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); + } + mddev->size = sectors /2; + return 0; +} + +static mdk_personality_t raid6_personality= +{ + .name = "raid6", + .owner = THIS_MODULE, + .make_request = make_request, + .run = run, + .stop = stop, + .status = status, + .error_handler = error, + .hot_add_disk = raid6_add_disk, + .hot_remove_disk= raid6_remove_disk, + .spare_active = raid6_spare_active, + .sync_request = sync_request, + .resize = raid6_resize, +}; + +static int __init raid6_init (void) +{ + int e; + + e = raid6_select_algo(); + if ( e ) + return e; + + return register_md_personality (RAID6, &raid6_personality); +} + +static void raid6_exit (void) +{ + unregister_md_personality (RAID6); +} + +module_init(raid6_init); +module_exit(raid6_exit); +MODULE_LICENSE("GPL"); +MODULE_ALIAS("md-personality-8"); /* RAID6 */ diff --git a/drivers/md/raid6mmx.c b/drivers/md/raid6mmx.c new file mode 100644 index 00000000000..359157aaf9e --- /dev/null +++ b/drivers/md/raid6mmx.c @@ -0,0 +1,150 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6mmx.c + * + * MMX implementation of RAID-6 syndrome functions + */ + +#if defined(__i386__) + +#include "raid6.h" +#include "raid6x86.h" + +/* Shared with raid6sse1.c */ +const struct raid6_mmx_constants { + u64 x1d; +} raid6_mmx_constants = { + 0x1d1d1d1d1d1d1d1dULL, +}; + +static int raid6_have_mmx(void) +{ +#ifdef __KERNEL__ + /* Not really "boot_cpu" but "all_cpus" */ + return boot_cpu_has(X86_FEATURE_MMX); +#else + /* User space test code */ + u32 features = cpuid_features(); + return ( (features & (1<<23)) == (1<<23) ); +#endif +} + +/* + * Plain MMX implementation + */ +static void raid6_mmx1_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_mmx_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_mmx(&sa); + + asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d)); + asm volatile("pxor %mm5,%mm5"); /* Zero temp */ + + for ( d = 0 ; d < bytes ; d += 8 ) { + asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("movq %mm2,%mm4"); /* Q[0] */ + for ( z = z0-1 ; z >= 0 ; z-- ) { + asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d])); + asm volatile("pcmpgtb %mm4,%mm5"); + asm volatile("paddb %mm4,%mm4"); + asm volatile("pand %mm0,%mm5"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm5,%mm5"); + asm volatile("pxor %mm6,%mm2"); + asm volatile("pxor %mm6,%mm4"); + } + asm volatile("movq %%mm2,%0" : "=m" (p[d])); + asm volatile("pxor %mm2,%mm2"); + asm volatile("movq %%mm4,%0" : "=m" (q[d])); + asm volatile("pxor %mm4,%mm4"); + } + + raid6_after_mmx(&sa); +} + +const struct raid6_calls raid6_mmxx1 = { + raid6_mmx1_gen_syndrome, + raid6_have_mmx, + "mmxx1", + 0 +}; + +/* + * Unrolled-by-2 MMX implementation + */ +static void raid6_mmx2_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_mmx_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_mmx(&sa); + + asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d)); + asm volatile("pxor %mm5,%mm5"); /* Zero temp */ + asm volatile("pxor %mm7,%mm7"); /* Zero temp */ + + for ( d = 0 ; d < bytes ; d += 16 ) { + asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8])); + asm volatile("movq %mm2,%mm4"); /* Q[0] */ + asm volatile("movq %mm3,%mm6"); /* Q[1] */ + for ( z = z0-1 ; z >= 0 ; z-- ) { + asm volatile("pcmpgtb %mm4,%mm5"); + asm volatile("pcmpgtb %mm6,%mm7"); + asm volatile("paddb %mm4,%mm4"); + asm volatile("paddb %mm6,%mm6"); + asm volatile("pand %mm0,%mm5"); + asm volatile("pand %mm0,%mm7"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm7,%mm6"); + asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d])); + asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8])); + asm volatile("pxor %mm5,%mm2"); + asm volatile("pxor %mm7,%mm3"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm7,%mm6"); + asm volatile("pxor %mm5,%mm5"); + asm volatile("pxor %mm7,%mm7"); + } + asm volatile("movq %%mm2,%0" : "=m" (p[d])); + asm volatile("movq %%mm3,%0" : "=m" (p[d+8])); + asm volatile("movq %%mm4,%0" : "=m" (q[d])); + asm volatile("movq %%mm6,%0" : "=m" (q[d+8])); + } + + raid6_after_mmx(&sa); +} + +const struct raid6_calls raid6_mmxx2 = { + raid6_mmx2_gen_syndrome, + raid6_have_mmx, + "mmxx2", + 0 +}; + +#endif diff --git a/drivers/md/raid6recov.c b/drivers/md/raid6recov.c new file mode 100644 index 00000000000..a8c4d9451bd --- /dev/null +++ b/drivers/md/raid6recov.c @@ -0,0 +1,133 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6recov.c + * + * RAID-6 data recovery in dual failure mode. In single failure mode, + * use the RAID-5 algorithm (or, in the case of Q failure, just reconstruct + * the syndrome.) + */ + +#include "raid6.h" + +/* Recover two failed data blocks. */ +void raid6_2data_recov(int disks, size_t bytes, int faila, int failb, + void **ptrs) +{ + u8 *p, *q, *dp, *dq; + u8 px, qx, db; + const u8 *pbmul; /* P multiplier table for B data */ + const u8 *qmul; /* Q multiplier table (for both) */ + + p = (u8 *)ptrs[disks-2]; + q = (u8 *)ptrs[disks-1]; + + /* Compute syndrome with zero for the missing data pages + Use the dead data pages as temporary storage for + delta p and delta q */ + dp = (u8 *)ptrs[faila]; + ptrs[faila] = (void *)raid6_empty_zero_page; + ptrs[disks-2] = dp; + dq = (u8 *)ptrs[failb]; + ptrs[failb] = (void *)raid6_empty_zero_page; + ptrs[disks-1] = dq; + + raid6_call.gen_syndrome(disks, bytes, ptrs); + + /* Restore pointer table */ + ptrs[faila] = dp; + ptrs[failb] = dq; + ptrs[disks-2] = p; + ptrs[disks-1] = q; + + /* Now, pick the proper data tables */ + pbmul = raid6_gfmul[raid6_gfexi[failb-faila]]; + qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]]]; + + /* Now do it... */ + while ( bytes-- ) { + px = *p ^ *dp; + qx = qmul[*q ^ *dq]; + *dq++ = db = pbmul[px] ^ qx; /* Reconstructed B */ + *dp++ = db ^ px; /* Reconstructed A */ + p++; q++; + } +} + + + + +/* Recover failure of one data block plus the P block */ +void raid6_datap_recov(int disks, size_t bytes, int faila, void **ptrs) +{ + u8 *p, *q, *dq; + const u8 *qmul; /* Q multiplier table */ + + p = (u8 *)ptrs[disks-2]; + q = (u8 *)ptrs[disks-1]; + + /* Compute syndrome with zero for the missing data page + Use the dead data page as temporary storage for delta q */ + dq = (u8 *)ptrs[faila]; + ptrs[faila] = (void *)raid6_empty_zero_page; + ptrs[disks-1] = dq; + + raid6_call.gen_syndrome(disks, bytes, ptrs); + + /* Restore pointer table */ + ptrs[faila] = dq; + ptrs[disks-1] = q; + + /* Now, pick the proper data tables */ + qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]]]; + + /* Now do it... */ + while ( bytes-- ) { + *p++ ^= *dq = qmul[*q ^ *dq]; + q++; dq++; + } +} + + +#ifndef __KERNEL__ /* Testing only */ + +/* Recover two failed blocks. */ +void raid6_dual_recov(int disks, size_t bytes, int faila, int failb, void **ptrs) +{ + if ( faila > failb ) { + int tmp = faila; + faila = failb; + failb = tmp; + } + + if ( failb == disks-1 ) { + if ( faila == disks-2 ) { + /* P+Q failure. Just rebuild the syndrome. */ + raid6_call.gen_syndrome(disks, bytes, ptrs); + } else { + /* data+Q failure. Reconstruct data from P, + then rebuild syndrome. */ + /* NOT IMPLEMENTED - equivalent to RAID-5 */ + } + } else { + if ( failb == disks-2 ) { + /* data+P failure. */ + raid6_datap_recov(disks, bytes, faila, ptrs); + } else { + /* data+data failure. */ + raid6_2data_recov(disks, bytes, faila, failb, ptrs); + } + } +} + +#endif diff --git a/drivers/md/raid6sse1.c b/drivers/md/raid6sse1.c new file mode 100644 index 00000000000..f7e7859f71a --- /dev/null +++ b/drivers/md/raid6sse1.c @@ -0,0 +1,171 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6sse1.c + * + * SSE-1/MMXEXT implementation of RAID-6 syndrome functions + * + * This is really an MMX implementation, but it requires SSE-1 or + * AMD MMXEXT for prefetch support and a few other features. The + * support for nontemporal memory accesses is enough to make this + * worthwhile as a separate implementation. + */ + +#if defined(__i386__) + +#include "raid6.h" +#include "raid6x86.h" + +/* Defined in raid6mmx.c */ +extern const struct raid6_mmx_constants { + u64 x1d; +} raid6_mmx_constants; + +static int raid6_have_sse1_or_mmxext(void) +{ +#ifdef __KERNEL__ + /* Not really boot_cpu but "all_cpus" */ + return boot_cpu_has(X86_FEATURE_MMX) && + (boot_cpu_has(X86_FEATURE_XMM) || + boot_cpu_has(X86_FEATURE_MMXEXT)); +#else + /* User space test code - this incorrectly breaks on some Athlons */ + u32 features = cpuid_features(); + return ( (features & (5<<23)) == (5<<23) ); +#endif +} + +/* + * Plain SSE1 implementation + */ +static void raid6_sse11_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_mmx_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + /* This is really MMX code, not SSE */ + raid6_before_mmx(&sa); + + asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d)); + asm volatile("pxor %mm5,%mm5"); /* Zero temp */ + + for ( d = 0 ; d < bytes ; d += 8 ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z0][d])); + asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d])); + asm volatile("movq %mm2,%mm4"); /* Q[0] */ + asm volatile("movq %0,%%mm6" : : "m" (dptr[z0-1][d])); + for ( z = z0-2 ; z >= 0 ; z-- ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z][d])); + asm volatile("pcmpgtb %mm4,%mm5"); + asm volatile("paddb %mm4,%mm4"); + asm volatile("pand %mm0,%mm5"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm5,%mm5"); + asm volatile("pxor %mm6,%mm2"); + asm volatile("pxor %mm6,%mm4"); + asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d])); + } + asm volatile("pcmpgtb %mm4,%mm5"); + asm volatile("paddb %mm4,%mm4"); + asm volatile("pand %mm0,%mm5"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm5,%mm5"); + asm volatile("pxor %mm6,%mm2"); + asm volatile("pxor %mm6,%mm4"); + + asm volatile("movntq %%mm2,%0" : "=m" (p[d])); + asm volatile("movntq %%mm4,%0" : "=m" (q[d])); + } + + raid6_after_mmx(&sa); + asm volatile("sfence" : : : "memory"); +} + +const struct raid6_calls raid6_sse1x1 = { + raid6_sse11_gen_syndrome, + raid6_have_sse1_or_mmxext, + "sse1x1", + 1 /* Has cache hints */ +}; + +/* + * Unrolled-by-2 SSE1 implementation + */ +static void raid6_sse12_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_mmx_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_mmx(&sa); + + asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d)); + asm volatile("pxor %mm5,%mm5"); /* Zero temp */ + asm volatile("pxor %mm7,%mm7"); /* Zero temp */ + + /* We uniformly assume a single prefetch covers at least 16 bytes */ + for ( d = 0 ; d < bytes ; d += 16 ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z0][d])); + asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8])); /* P[1] */ + asm volatile("movq %mm2,%mm4"); /* Q[0] */ + asm volatile("movq %mm3,%mm6"); /* Q[1] */ + for ( z = z0-1 ; z >= 0 ; z-- ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z][d])); + asm volatile("pcmpgtb %mm4,%mm5"); + asm volatile("pcmpgtb %mm6,%mm7"); + asm volatile("paddb %mm4,%mm4"); + asm volatile("paddb %mm6,%mm6"); + asm volatile("pand %mm0,%mm5"); + asm volatile("pand %mm0,%mm7"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm7,%mm6"); + asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d])); + asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8])); + asm volatile("pxor %mm5,%mm2"); + asm volatile("pxor %mm7,%mm3"); + asm volatile("pxor %mm5,%mm4"); + asm volatile("pxor %mm7,%mm6"); + asm volatile("pxor %mm5,%mm5"); + asm volatile("pxor %mm7,%mm7"); + } + asm volatile("movntq %%mm2,%0" : "=m" (p[d])); + asm volatile("movntq %%mm3,%0" : "=m" (p[d+8])); + asm volatile("movntq %%mm4,%0" : "=m" (q[d])); + asm volatile("movntq %%mm6,%0" : "=m" (q[d+8])); + } + + raid6_after_mmx(&sa); + asm volatile("sfence" : :: "memory"); +} + +const struct raid6_calls raid6_sse1x2 = { + raid6_sse12_gen_syndrome, + raid6_have_sse1_or_mmxext, + "sse1x2", + 1 /* Has cache hints */ +}; + +#endif diff --git a/drivers/md/raid6sse2.c b/drivers/md/raid6sse2.c new file mode 100644 index 00000000000..b3aa7fe0877 --- /dev/null +++ b/drivers/md/raid6sse2.c @@ -0,0 +1,270 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6sse2.c + * + * SSE-2 implementation of RAID-6 syndrome functions + * + */ + +#if defined(__i386__) || defined(__x86_64__) + +#include "raid6.h" +#include "raid6x86.h" + +static const struct raid6_sse_constants { + u64 x1d[2]; +} raid6_sse_constants __attribute__((aligned(16))) = { + { 0x1d1d1d1d1d1d1d1dULL, 0x1d1d1d1d1d1d1d1dULL }, +}; + +static int raid6_have_sse2(void) +{ +#ifdef __KERNEL__ + /* Not really boot_cpu but "all_cpus" */ + return boot_cpu_has(X86_FEATURE_MMX) && + boot_cpu_has(X86_FEATURE_FXSR) && + boot_cpu_has(X86_FEATURE_XMM) && + boot_cpu_has(X86_FEATURE_XMM2); +#else + /* User space test code */ + u32 features = cpuid_features(); + return ( (features & (15<<23)) == (15<<23) ); +#endif +} + +/* + * Plain SSE2 implementation + */ +static void raid6_sse21_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_sse_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_sse2(&sa); + + asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0])); + asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */ + + for ( d = 0 ; d < bytes ; d += 16 ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z0][d])); + asm volatile("movdqa %0,%%xmm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d])); + asm volatile("movdqa %xmm2,%xmm4"); /* Q[0] */ + asm volatile("movdqa %0,%%xmm6" : : "m" (dptr[z0-1][d])); + for ( z = z0-2 ; z >= 0 ; z-- ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z][d])); + asm volatile("pcmpgtb %xmm4,%xmm5"); + asm volatile("paddb %xmm4,%xmm4"); + asm volatile("pand %xmm0,%xmm5"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm5,%xmm5"); + asm volatile("pxor %xmm6,%xmm2"); + asm volatile("pxor %xmm6,%xmm4"); + asm volatile("movdqa %0,%%xmm6" : : "m" (dptr[z][d])); + } + asm volatile("pcmpgtb %xmm4,%xmm5"); + asm volatile("paddb %xmm4,%xmm4"); + asm volatile("pand %xmm0,%xmm5"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm5,%xmm5"); + asm volatile("pxor %xmm6,%xmm2"); + asm volatile("pxor %xmm6,%xmm4"); + + asm volatile("movntdq %%xmm2,%0" : "=m" (p[d])); + asm volatile("pxor %xmm2,%xmm2"); + asm volatile("movntdq %%xmm4,%0" : "=m" (q[d])); + asm volatile("pxor %xmm4,%xmm4"); + } + + raid6_after_sse2(&sa); + asm volatile("sfence" : : : "memory"); +} + +const struct raid6_calls raid6_sse2x1 = { + raid6_sse21_gen_syndrome, + raid6_have_sse2, + "sse2x1", + 1 /* Has cache hints */ +}; + +/* + * Unrolled-by-2 SSE2 implementation + */ +static void raid6_sse22_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_sse_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_sse2(&sa); + + asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0])); + asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */ + asm volatile("pxor %xmm7,%xmm7"); /* Zero temp */ + + /* We uniformly assume a single prefetch covers at least 32 bytes */ + for ( d = 0 ; d < bytes ; d += 32 ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z0][d])); + asm volatile("movdqa %0,%%xmm2" : : "m" (dptr[z0][d])); /* P[0] */ + asm volatile("movdqa %0,%%xmm3" : : "m" (dptr[z0][d+16])); /* P[1] */ + asm volatile("movdqa %xmm2,%xmm4"); /* Q[0] */ + asm volatile("movdqa %xmm3,%xmm6"); /* Q[1] */ + for ( z = z0-1 ; z >= 0 ; z-- ) { + asm volatile("prefetchnta %0" : : "m" (dptr[z][d])); + asm volatile("pcmpgtb %xmm4,%xmm5"); + asm volatile("pcmpgtb %xmm6,%xmm7"); + asm volatile("paddb %xmm4,%xmm4"); + asm volatile("paddb %xmm6,%xmm6"); + asm volatile("pand %xmm0,%xmm5"); + asm volatile("pand %xmm0,%xmm7"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm7,%xmm6"); + asm volatile("movdqa %0,%%xmm5" : : "m" (dptr[z][d])); + asm volatile("movdqa %0,%%xmm7" : : "m" (dptr[z][d+16])); + asm volatile("pxor %xmm5,%xmm2"); + asm volatile("pxor %xmm7,%xmm3"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm7,%xmm6"); + asm volatile("pxor %xmm5,%xmm5"); + asm volatile("pxor %xmm7,%xmm7"); + } + asm volatile("movntdq %%xmm2,%0" : "=m" (p[d])); + asm volatile("movntdq %%xmm3,%0" : "=m" (p[d+16])); + asm volatile("movntdq %%xmm4,%0" : "=m" (q[d])); + asm volatile("movntdq %%xmm6,%0" : "=m" (q[d+16])); + } + + raid6_after_sse2(&sa); + asm volatile("sfence" : : : "memory"); +} + +const struct raid6_calls raid6_sse2x2 = { + raid6_sse22_gen_syndrome, + raid6_have_sse2, + "sse2x2", + 1 /* Has cache hints */ +}; + +#endif + +#ifdef __x86_64__ + +/* + * Unrolled-by-4 SSE2 implementation + */ +static void raid6_sse24_gen_syndrome(int disks, size_t bytes, void **ptrs) +{ + u8 **dptr = (u8 **)ptrs; + u8 *p, *q; + int d, z, z0; + raid6_sse16_save_t sa; + + z0 = disks - 3; /* Highest data disk */ + p = dptr[z0+1]; /* XOR parity */ + q = dptr[z0+2]; /* RS syndrome */ + + raid6_before_sse16(&sa); + + asm volatile("movdqa %0,%%xmm0" :: "m" (raid6_sse_constants.x1d[0])); + asm volatile("pxor %xmm2,%xmm2"); /* P[0] */ + asm volatile("pxor %xmm3,%xmm3"); /* P[1] */ + asm volatile("pxor %xmm4,%xmm4"); /* Q[0] */ + asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */ + asm volatile("pxor %xmm6,%xmm6"); /* Q[1] */ + asm volatile("pxor %xmm7,%xmm7"); /* Zero temp */ + asm volatile("pxor %xmm10,%xmm10"); /* P[2] */ + asm volatile("pxor %xmm11,%xmm11"); /* P[3] */ + asm volatile("pxor %xmm12,%xmm12"); /* Q[2] */ + asm volatile("pxor %xmm13,%xmm13"); /* Zero temp */ + asm volatile("pxor %xmm14,%xmm14"); /* Q[3] */ + asm volatile("pxor %xmm15,%xmm15"); /* Zero temp */ + + for ( d = 0 ; d < bytes ; d += 64 ) { + for ( z = z0 ; z >= 0 ; z-- ) { + /* The second prefetch seems to improve performance... */ + asm volatile("prefetchnta %0" :: "m" (dptr[z][d])); + asm volatile("prefetchnta %0" :: "m" (dptr[z][d+32])); + asm volatile("pcmpgtb %xmm4,%xmm5"); + asm volatile("pcmpgtb %xmm6,%xmm7"); + asm volatile("pcmpgtb %xmm12,%xmm13"); + asm volatile("pcmpgtb %xmm14,%xmm15"); + asm volatile("paddb %xmm4,%xmm4"); + asm volatile("paddb %xmm6,%xmm6"); + asm volatile("paddb %xmm12,%xmm12"); + asm volatile("paddb %xmm14,%xmm14"); + asm volatile("pand %xmm0,%xmm5"); + asm volatile("pand %xmm0,%xmm7"); + asm volatile("pand %xmm0,%xmm13"); + asm volatile("pand %xmm0,%xmm15"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm7,%xmm6"); + asm volatile("pxor %xmm13,%xmm12"); + asm volatile("pxor %xmm15,%xmm14"); + asm volatile("movdqa %0,%%xmm5" :: "m" (dptr[z][d])); + asm volatile("movdqa %0,%%xmm7" :: "m" (dptr[z][d+16])); + asm volatile("movdqa %0,%%xmm13" :: "m" (dptr[z][d+32])); + asm volatile("movdqa %0,%%xmm15" :: "m" (dptr[z][d+48])); + asm volatile("pxor %xmm5,%xmm2"); + asm volatile("pxor %xmm7,%xmm3"); + asm volatile("pxor %xmm13,%xmm10"); + asm volatile("pxor %xmm15,%xmm11"); + asm volatile("pxor %xmm5,%xmm4"); + asm volatile("pxor %xmm7,%xmm6"); + asm volatile("pxor %xmm13,%xmm12"); + asm volatile("pxor %xmm15,%xmm14"); + asm volatile("pxor %xmm5,%xmm5"); + asm volatile("pxor %xmm7,%xmm7"); + asm volatile("pxor %xmm13,%xmm13"); + asm volatile("pxor %xmm15,%xmm15"); + } + asm volatile("movntdq %%xmm2,%0" : "=m" (p[d])); + asm volatile("pxor %xmm2,%xmm2"); + asm volatile("movntdq %%xmm3,%0" : "=m" (p[d+16])); + asm volatile("pxor %xmm3,%xmm3"); + asm volatile("movntdq %%xmm10,%0" : "=m" (p[d+32])); + asm volatile("pxor %xmm10,%xmm10"); + asm volatile("movntdq %%xmm11,%0" : "=m" (p[d+48])); + asm volatile("pxor %xmm11,%xmm11"); + asm volatile("movntdq %%xmm4,%0" : "=m" (q[d])); + asm volatile("pxor %xmm4,%xmm4"); + asm volatile("movntdq %%xmm6,%0" : "=m" (q[d+16])); + asm volatile("pxor %xmm6,%xmm6"); + asm volatile("movntdq %%xmm12,%0" : "=m" (q[d+32])); + asm volatile("pxor %xmm12,%xmm12"); + asm volatile("movntdq %%xmm14,%0" : "=m" (q[d+48])); + asm volatile("pxor %xmm14,%xmm14"); + } + asm volatile("sfence" : : : "memory"); + raid6_after_sse16(&sa); +} + +const struct raid6_calls raid6_sse2x4 = { + raid6_sse24_gen_syndrome, + raid6_have_sse2, + "sse2x4", + 1 /* Has cache hints */ +}; + +#endif diff --git a/drivers/md/raid6test/Makefile b/drivers/md/raid6test/Makefile new file mode 100644 index 00000000000..55780672860 --- /dev/null +++ b/drivers/md/raid6test/Makefile @@ -0,0 +1,58 @@ +# +# This is a simple Makefile to test some of the RAID-6 code +# from userspace. +# + +CC = gcc +OPTFLAGS = -O2 # Adjust as desired +CFLAGS = -I.. -g $(OPTFLAGS) +LD = ld +PERL = perl + +.c.o: + $(CC) $(CFLAGS) -c -o $@ $< + +%.c: ../%.c + cp -f $< $@ + +%.uc: ../%.uc + cp -f $< $@ + +all: raid6.o raid6test + +raid6.o: raid6int1.o raid6int2.o raid6int4.o raid6int8.o raid6int16.o \ + raid6int32.o \ + raid6mmx.o raid6sse1.o raid6sse2.o \ + raid6recov.o raid6algos.o \ + raid6tables.o + $(LD) -r -o $@ $^ + +raid6test: raid6.o test.c + $(CC) $(CFLAGS) -o raid6test $^ + +raid6int1.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 1 < raid6int.uc > $@ + +raid6int2.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 2 < raid6int.uc > $@ + +raid6int4.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 4 < raid6int.uc > $@ + +raid6int8.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 8 < raid6int.uc > $@ + +raid6int16.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 16 < raid6int.uc > $@ + +raid6int32.c: raid6int.uc ../unroll.pl + $(PERL) ../unroll.pl 32 < raid6int.uc > $@ + +raid6tables.c: mktables + ./mktables > raid6tables.c + +clean: + rm -f *.o mktables mktables.c raid6int.uc raid6*.c raid6test + +spotless: clean + rm -f *~ diff --git a/drivers/md/raid6test/test.c b/drivers/md/raid6test/test.c new file mode 100644 index 00000000000..0d5cd57accd --- /dev/null +++ b/drivers/md/raid6test/test.c @@ -0,0 +1,103 @@ +/* -*- linux-c -*- ------------------------------------------------------- * + * + * Copyright 2002 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6test.c + * + * Test RAID-6 recovery with various algorithms + */ + +#include <stdlib.h> +#include <stdio.h> +#include <string.h> +#include "raid6.h" + +#define NDISKS 16 /* Including P and Q */ + +const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256))); +struct raid6_calls raid6_call; + +char *dataptrs[NDISKS]; +char data[NDISKS][PAGE_SIZE]; +char recovi[PAGE_SIZE], recovj[PAGE_SIZE]; + +void makedata(void) +{ + int i, j; + + for ( i = 0 ; i < NDISKS ; i++ ) { + for ( j = 0 ; j < PAGE_SIZE ; j++ ) { + data[i][j] = rand(); + } + dataptrs[i] = data[i]; + } +} + +int main(int argc, char *argv[]) +{ + const struct raid6_calls * const * algo; + int i, j; + int erra, errb; + + makedata(); + + for ( algo = raid6_algos ; *algo ; algo++ ) { + if ( !(*algo)->valid || (*algo)->valid() ) { + raid6_call = **algo; + + /* Nuke syndromes */ + memset(data[NDISKS-2], 0xee, 2*PAGE_SIZE); + + /* Generate assumed good syndrome */ + raid6_call.gen_syndrome(NDISKS, PAGE_SIZE, (void **)&dataptrs); + + for ( i = 0 ; i < NDISKS-1 ; i++ ) { + for ( j = i+1 ; j < NDISKS ; j++ ) { + memset(recovi, 0xf0, PAGE_SIZE); + memset(recovj, 0xba, PAGE_SIZE); + + dataptrs[i] = recovi; + dataptrs[j] = recovj; + + raid6_dual_recov(NDISKS, PAGE_SIZE, i, j, (void **)&dataptrs); + + erra = memcmp(data[i], recovi, PAGE_SIZE); + errb = memcmp(data[j], recovj, PAGE_SIZE); + + if ( i < NDISKS-2 && j == NDISKS-1 ) { + /* We don't implement the DQ failure scenario, since it's + equivalent to a RAID-5 failure (XOR, then recompute Q) */ + } else { + printf("algo=%-8s faila=%3d(%c) failb=%3d(%c) %s\n", + raid6_call.name, + i, (i==NDISKS-2)?'P':'D', + j, (j==NDISKS-1)?'Q':(j==NDISKS-2)?'P':'D', + (!erra && !errb) ? "OK" : + !erra ? "ERRB" : + !errb ? "ERRA" : + "ERRAB"); + } + + dataptrs[i] = data[i]; + dataptrs[j] = data[j]; + } + } + } + printf("\n"); + } + + printf("\n"); + /* Pick the best algorithm test */ + raid6_select_algo(); + + return 0; +} diff --git a/drivers/md/raid6x86.h b/drivers/md/raid6x86.h new file mode 100644 index 00000000000..4cf20534fe4 --- /dev/null +++ b/drivers/md/raid6x86.h @@ -0,0 +1,245 @@ +/* ----------------------------------------------------------------------- * + * + * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, Inc., 53 Temple Place Ste 330, + * Bostom MA 02111-1307, USA; either version 2 of the License, or + * (at your option) any later version; incorporated herein by reference. + * + * ----------------------------------------------------------------------- */ + +/* + * raid6x86.h + * + * Definitions common to x86 and x86-64 RAID-6 code only + */ + +#ifndef LINUX_RAID_RAID6X86_H +#define LINUX_RAID_RAID6X86_H + +#if defined(__i386__) || defined(__x86_64__) + +#ifdef __x86_64__ + +typedef struct { + unsigned int fsave[27]; + unsigned long cr0; +} raid6_mmx_save_t __attribute__((aligned(16))); + +/* N.B.: For SSE we only save %xmm0-%xmm7 even for x86-64, since + the code doesn't know about the additional x86-64 registers */ +typedef struct { + unsigned int sarea[8*4+2]; + unsigned long cr0; +} raid6_sse_save_t __attribute__((aligned(16))); + +/* This is for x86-64-specific code which uses all 16 XMM registers */ +typedef struct { + unsigned int sarea[16*4+2]; + unsigned long cr0; +} raid6_sse16_save_t __attribute__((aligned(16))); + +/* On x86-64 the stack *SHOULD* be 16-byte aligned, but currently this + is buggy in the kernel and it's only 8-byte aligned in places, so + we need to do this anyway. Sigh. */ +#define SAREA(x) ((unsigned int *)((((unsigned long)&(x)->sarea)+15) & ~15)) + +#else /* __i386__ */ + +typedef struct { + unsigned int fsave[27]; + unsigned long cr0; +} raid6_mmx_save_t; + +/* On i386, the stack is only 8-byte aligned, but SSE requires 16-byte + alignment. The +3 is so we have the slack space to manually align + a properly-sized area correctly. */ +typedef struct { + unsigned int sarea[8*4+3]; + unsigned long cr0; +} raid6_sse_save_t; + +/* Find the 16-byte aligned save area */ +#define SAREA(x) ((unsigned int *)((((unsigned long)&(x)->sarea)+15) & ~15)) + +#endif + +#ifdef __KERNEL__ /* Real code */ + +/* Note: %cr0 is 32 bits on i386 and 64 bits on x86-64 */ + +static inline unsigned long raid6_get_fpu(void) +{ + unsigned long cr0; + + preempt_disable(); + asm volatile("mov %%cr0,%0 ; clts" : "=r" (cr0)); + return cr0; +} + +static inline void raid6_put_fpu(unsigned long cr0) +{ + asm volatile("mov %0,%%cr0" : : "r" (cr0)); + preempt_enable(); +} + +#else /* Dummy code for user space testing */ + +static inline unsigned long raid6_get_fpu(void) +{ + return 0xf00ba6; +} + +static inline void raid6_put_fpu(unsigned long cr0) +{ + (void)cr0; +} + +#endif + +static inline void raid6_before_mmx(raid6_mmx_save_t *s) +{ + s->cr0 = raid6_get_fpu(); + asm volatile("fsave %0 ; fwait" : "=m" (s->fsave[0])); +} + +static inline void raid6_after_mmx(raid6_mmx_save_t *s) +{ + asm volatile("frstor %0" : : "m" (s->fsave[0])); + raid6_put_fpu(s->cr0); +} + +static inline void raid6_before_sse(raid6_sse_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + s->cr0 = raid6_get_fpu(); + + asm volatile("movaps %%xmm0,%0" : "=m" (rsa[0])); + asm volatile("movaps %%xmm1,%0" : "=m" (rsa[4])); + asm volatile("movaps %%xmm2,%0" : "=m" (rsa[8])); + asm volatile("movaps %%xmm3,%0" : "=m" (rsa[12])); + asm volatile("movaps %%xmm4,%0" : "=m" (rsa[16])); + asm volatile("movaps %%xmm5,%0" : "=m" (rsa[20])); + asm volatile("movaps %%xmm6,%0" : "=m" (rsa[24])); + asm volatile("movaps %%xmm7,%0" : "=m" (rsa[28])); +} + +static inline void raid6_after_sse(raid6_sse_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + asm volatile("movaps %0,%%xmm0" : : "m" (rsa[0])); + asm volatile("movaps %0,%%xmm1" : : "m" (rsa[4])); + asm volatile("movaps %0,%%xmm2" : : "m" (rsa[8])); + asm volatile("movaps %0,%%xmm3" : : "m" (rsa[12])); + asm volatile("movaps %0,%%xmm4" : : "m" (rsa[16])); + asm volatile("movaps %0,%%xmm5" : : "m" (rsa[20])); + asm volatile("movaps %0,%%xmm6" : : "m" (rsa[24])); + asm volatile("movaps %0,%%xmm7" : : "m" (rsa[28])); + + raid6_put_fpu(s->cr0); +} + +static inline void raid6_before_sse2(raid6_sse_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + s->cr0 = raid6_get_fpu(); + + asm volatile("movdqa %%xmm0,%0" : "=m" (rsa[0])); + asm volatile("movdqa %%xmm1,%0" : "=m" (rsa[4])); + asm volatile("movdqa %%xmm2,%0" : "=m" (rsa[8])); + asm volatile("movdqa %%xmm3,%0" : "=m" (rsa[12])); + asm volatile("movdqa %%xmm4,%0" : "=m" (rsa[16])); + asm volatile("movdqa %%xmm5,%0" : "=m" (rsa[20])); + asm volatile("movdqa %%xmm6,%0" : "=m" (rsa[24])); + asm volatile("movdqa %%xmm7,%0" : "=m" (rsa[28])); +} + +static inline void raid6_after_sse2(raid6_sse_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + asm volatile("movdqa %0,%%xmm0" : : "m" (rsa[0])); + asm volatile("movdqa %0,%%xmm1" : : "m" (rsa[4])); + asm volatile("movdqa %0,%%xmm2" : : "m" (rsa[8])); + asm volatile("movdqa %0,%%xmm3" : : "m" (rsa[12])); + asm volatile("movdqa %0,%%xmm4" : : "m" (rsa[16])); + asm volatile("movdqa %0,%%xmm5" : : "m" (rsa[20])); + asm volatile("movdqa %0,%%xmm6" : : "m" (rsa[24])); + asm volatile("movdqa %0,%%xmm7" : : "m" (rsa[28])); + + raid6_put_fpu(s->cr0); +} + +#ifdef __x86_64__ + +static inline void raid6_before_sse16(raid6_sse16_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + s->cr0 = raid6_get_fpu(); + + asm volatile("movdqa %%xmm0,%0" : "=m" (rsa[0])); + asm volatile("movdqa %%xmm1,%0" : "=m" (rsa[4])); + asm volatile("movdqa %%xmm2,%0" : "=m" (rsa[8])); + asm volatile("movdqa %%xmm3,%0" : "=m" (rsa[12])); + asm volatile("movdqa %%xmm4,%0" : "=m" (rsa[16])); + asm volatile("movdqa %%xmm5,%0" : "=m" (rsa[20])); + asm volatile("movdqa %%xmm6,%0" : "=m" (rsa[24])); + asm volatile("movdqa %%xmm7,%0" : "=m" (rsa[28])); + asm volatile("movdqa %%xmm8,%0" : "=m" (rsa[32])); + asm volatile("movdqa %%xmm9,%0" : "=m" (rsa[36])); + asm volatile("movdqa %%xmm10,%0" : "=m" (rsa[40])); + asm volatile("movdqa %%xmm11,%0" : "=m" (rsa[44])); + asm volatile("movdqa %%xmm12,%0" : "=m" (rsa[48])); + asm volatile("movdqa %%xmm13,%0" : "=m" (rsa[52])); + asm volatile("movdqa %%xmm14,%0" : "=m" (rsa[56])); + asm volatile("movdqa %%xmm15,%0" : "=m" (rsa[60])); +} + +static inline void raid6_after_sse16(raid6_sse16_save_t *s) +{ + unsigned int *rsa = SAREA(s); + + asm volatile("movdqa %0,%%xmm0" : : "m" (rsa[0])); + asm volatile("movdqa %0,%%xmm1" : : "m" (rsa[4])); + asm volatile("movdqa %0,%%xmm2" : : "m" (rsa[8])); + asm volatile("movdqa %0,%%xmm3" : : "m" (rsa[12])); + asm volatile("movdqa %0,%%xmm4" : : "m" (rsa[16])); + asm volatile("movdqa %0,%%xmm5" : : "m" (rsa[20])); + asm volatile("movdqa %0,%%xmm6" : : "m" (rsa[24])); + asm volatile("movdqa %0,%%xmm7" : : "m" (rsa[28])); + asm volatile("movdqa %0,%%xmm8" : : "m" (rsa[32])); + asm volatile("movdqa %0,%%xmm9" : : "m" (rsa[36])); + asm volatile("movdqa %0,%%xmm10" : : "m" (rsa[40])); + asm volatile("movdqa %0,%%xmm11" : : "m" (rsa[44])); + asm volatile("movdqa %0,%%xmm12" : : "m" (rsa[48])); + asm volatile("movdqa %0,%%xmm13" : : "m" (rsa[52])); + asm volatile("movdqa %0,%%xmm14" : : "m" (rsa[56])); + asm volatile("movdqa %0,%%xmm15" : : "m" (rsa[60])); + + raid6_put_fpu(s->cr0); +} + +#endif /* __x86_64__ */ + +/* User space test hack */ +#ifndef __KERNEL__ +static inline int cpuid_features(void) +{ + u32 eax = 1; + u32 ebx, ecx, edx; + + asm volatile("cpuid" : + "+a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)); + + return edx; +} +#endif /* ndef __KERNEL__ */ + +#endif +#endif diff --git a/drivers/md/unroll.pl b/drivers/md/unroll.pl new file mode 100644 index 00000000000..3acc710a20e --- /dev/null +++ b/drivers/md/unroll.pl @@ -0,0 +1,24 @@ +#!/usr/bin/perl +# +# Take a piece of C code and for each line which contains the sequence $$ +# repeat n times with $ replaced by 0...n-1; the sequence $# is replaced +# by the unrolling factor, and $* with a single $ +# + +($n) = @ARGV; +$n += 0; + +while ( defined($line = <STDIN>) ) { + if ( $line =~ /\$\$/ ) { + $rep = $n; + } else { + $rep = 1; + } + for ( $i = 0 ; $i < $rep ; $i++ ) { + $tmp = $line; + $tmp =~ s/\$\$/$i/g; + $tmp =~ s/\$\#/$n/g; + $tmp =~ s/\$\*/\$/g; + print $tmp; + } +} diff --git a/drivers/md/xor.c b/drivers/md/xor.c new file mode 100644 index 00000000000..324897c4be4 --- /dev/null +++ b/drivers/md/xor.c @@ -0,0 +1,154 @@ +/* + * xor.c : Multiple Devices driver for Linux + * + * Copyright (C) 1996, 1997, 1998, 1999, 2000, + * Ingo Molnar, Matti Aarnio, Jakub Jelinek, Richard Henderson. + * + * Dispatch optimized RAID-5 checksumming functions. + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2, or (at your option) + * any later version. + * + * You should have received a copy of the GNU General Public License + * (for example /usr/src/linux/COPYING); if not, write to the Free + * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +#define BH_TRACE 0 +#include <linux/module.h> +#include <linux/raid/md.h> +#include <linux/raid/xor.h> +#include <asm/xor.h> + +/* The xor routines to use. */ +static struct xor_block_template *active_template; + +void +xor_block(unsigned int count, unsigned int bytes, void **ptr) +{ + unsigned long *p0, *p1, *p2, *p3, *p4; + + p0 = (unsigned long *) ptr[0]; + p1 = (unsigned long *) ptr[1]; + if (count == 2) { + active_template->do_2(bytes, p0, p1); + return; + } + + p2 = (unsigned long *) ptr[2]; + if (count == 3) { + active_template->do_3(bytes, p0, p1, p2); + return; + } + + p3 = (unsigned long *) ptr[3]; + if (count == 4) { + active_template->do_4(bytes, p0, p1, p2, p3); + return; + } + + p4 = (unsigned long *) ptr[4]; + active_template->do_5(bytes, p0, p1, p2, p3, p4); +} + +/* Set of all registered templates. */ +static struct xor_block_template *template_list; + +#define BENCH_SIZE (PAGE_SIZE) + +static void +do_xor_speed(struct xor_block_template *tmpl, void *b1, void *b2) +{ + int speed; + unsigned long now; + int i, count, max; + + tmpl->next = template_list; + template_list = tmpl; + + /* + * Count the number of XORs done during a whole jiffy, and use + * this to calculate the speed of checksumming. We use a 2-page + * allocation to have guaranteed color L1-cache layout. + */ + max = 0; + for (i = 0; i < 5; i++) { + now = jiffies; + count = 0; + while (jiffies == now) { + mb(); + tmpl->do_2(BENCH_SIZE, b1, b2); + mb(); + count++; + mb(); + } + if (count > max) + max = count; + } + + speed = max * (HZ * BENCH_SIZE / 1024); + tmpl->speed = speed; + + printk(" %-10s: %5d.%03d MB/sec\n", tmpl->name, + speed / 1000, speed % 1000); +} + +static int +calibrate_xor_block(void) +{ + void *b1, *b2; + struct xor_block_template *f, *fastest; + + b1 = (void *) __get_free_pages(GFP_KERNEL, 2); + if (! b1) { + printk("raid5: Yikes! No memory available.\n"); + return -ENOMEM; + } + b2 = b1 + 2*PAGE_SIZE + BENCH_SIZE; + + /* + * If this arch/cpu has a short-circuited selection, don't loop through all + * the possible functions, just test the best one + */ + + fastest = NULL; + +#ifdef XOR_SELECT_TEMPLATE + fastest = XOR_SELECT_TEMPLATE(fastest); +#endif + +#define xor_speed(templ) do_xor_speed((templ), b1, b2) + + if (fastest) { + printk(KERN_INFO "raid5: automatically using best checksumming function: %s\n", + fastest->name); + xor_speed(fastest); + } else { + printk(KERN_INFO "raid5: measuring checksumming speed\n"); + XOR_TRY_TEMPLATES; + fastest = template_list; + for (f = fastest; f; f = f->next) + if (f->speed > fastest->speed) + fastest = f; + } + + printk("raid5: using function: %s (%d.%03d MB/sec)\n", + fastest->name, fastest->speed / 1000, fastest->speed % 1000); + +#undef xor_speed + + free_pages((unsigned long)b1, 2); + + active_template = fastest; + return 0; +} + +static __exit void xor_exit(void) { } + +EXPORT_SYMBOL(xor_block); +MODULE_LICENSE("GPL"); + +module_init(calibrate_xor_block); +module_exit(xor_exit); |