[PATCH] FUSE - direct I/O

This patch adds support for the "direct_io" mount option of FUSE.

When this mount option is specified, the page cache is bypassed for
read and write operations.  This is useful for example, if the
filesystem doesn't know the size of files before reading them, or when
any kind of caching is harmful.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

3 files changed, 146 insertions, 2 deletions

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 86ffb6db5fe..6bc3fb26de3 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -363,6 +363,118 @@ static int fuse_commit_write(struct file *file, struct page *page,
 	return err;
 }
 
+static void fuse_release_user_pages(struct fuse_req *req, int write)
+{
+	unsigned i;
+
+	for (i = 0; i < req->num_pages; i++) {
+		struct page *page = req->pages[i];
+		if (write)
+			set_page_dirty_lock(page);
+		put_page(page);
+	}
+}
+
+static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
+			       unsigned nbytes, int write)
+{
+	unsigned long user_addr = (unsigned long) buf;
+	unsigned offset = user_addr & ~PAGE_MASK;
+	int npages;
+
+	/* This doesn't work with nfsd */
+	if (!current->mm)
+		return -EPERM;
+
+	nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
+	npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
+	npages = min(npages, FUSE_MAX_PAGES_PER_REQ);
+	down_read(&current->mm->mmap_sem);
+	npages = get_user_pages(current, current->mm, user_addr, npages, write,
+				0, req->pages, NULL);
+	up_read(&current->mm->mmap_sem);
+	if (npages < 0)
+		return npages;
+
+	req->num_pages = npages;
+	req->page_offset = offset;
+	return 0;
+}
+
+static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
+			      size_t count, loff_t *ppos, int write)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	struct fuse_conn *fc = get_fuse_conn(inode);
+	size_t nmax = write ? fc->max_write : fc->max_read;
+	loff_t pos = *ppos;
+	ssize_t res = 0;
+	struct fuse_req *req = fuse_get_request(fc);
+	if (!req)
+		return -ERESTARTSYS;
+
+	while (count) {
+		size_t tmp;
+		size_t nres;
+		size_t nbytes = min(count, nmax);
+		int err = fuse_get_user_pages(req, buf, nbytes, !write);
+		if (err) {
+			res = err;
+			break;
+		}
+		tmp = (req->num_pages << PAGE_SHIFT) - req->page_offset;
+		nbytes = min(nbytes, tmp);
+		if (write)
+			nres = fuse_send_write(req, file, inode, pos, nbytes);
+		else
+			nres = fuse_send_read(req, file, inode, pos, nbytes);
+		fuse_release_user_pages(req, !write);
+		if (req->out.h.error) {
+			if (!res)
+				res = req->out.h.error;
+			break;
+		} else if (nres > nbytes) {
+			res = -EIO;
+			break;
+		}
+		count -= nres;
+		res += nres;
+		pos += nres;
+		buf += nres;
+		if (nres != nbytes)
+			break;
+		if (count)
+			fuse_reset_request(req);
+	}
+	fuse_put_request(fc, req);
+	if (res > 0) {
+		if (write && pos > i_size_read(inode))
+			i_size_write(inode, pos);
+		*ppos = pos;
+	} else if (write && (res == -EINTR || res == -EIO))
+		fuse_invalidate_attr(inode);
+
+	return res;
+}
+
+static ssize_t fuse_direct_read(struct file *file, char __user *buf,
+				     size_t count, loff_t *ppos)
+{
+	return fuse_direct_io(file, buf, count, ppos, 0);
+}
+
+static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	struct inode *inode = file->f_dentry->d_inode;
+	ssize_t res;
+	/* Don't allow parallel writes to the same file */
+	down(&inode->i_sem);
+	res = fuse_direct_io(file, buf, count, ppos, 1);
+	up(&inode->i_sem);
+	return res;
+}
+
 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	if ((vma->vm_flags & VM_SHARED)) {
@@ -393,6 +505,17 @@ static struct file_operations fuse_file_operations = {
 	.sendfile	= generic_file_sendfile,
 };
 
+static struct file_operations fuse_direct_io_file_operations = {
+	.llseek		= generic_file_llseek,
+	.read		= fuse_direct_read,
+	.write		= fuse_direct_write,
+	.open		= fuse_open,
+	.flush		= fuse_flush,
+	.release	= fuse_release,
+	.fsync		= fuse_fsync,
+	/* no mmap and sendfile */
+};
+
 static struct address_space_operations fuse_file_aops  = {
 	.readpage	= fuse_readpage,
 	.prepare_write	= fuse_prepare_write,
@@ -403,6 +526,12 @@ static struct address_space_operations fuse_file_aops  = {
 
 void fuse_init_file_inode(struct inode *inode)
 {
-	inode->i_fop = &fuse_file_operations;
-	inode->i_data.a_ops = &fuse_file_aops;
+	struct fuse_conn *fc = get_fuse_conn(inode);
+
+	if (fc->flags & FUSE_DIRECT_IO)
+		inode->i_fop = &fuse_direct_io_file_operations;
+	else {
+		inode->i_fop = &fuse_file_operations;
+		inode->i_data.a_ops = &fuse_file_aops;
+	}
 }
diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
index 3ec2aff3fdb..0af1ac64692 100644
--- a/fs/fuse/fuse_i.h
+++ b/fs/fuse/fuse_i.h
@@ -34,6 +34,9 @@
     be flushed on open */
 #define FUSE_KERNEL_CACHE        (1 << 2)
 
+/** Bypass the page cache for read and write operations  */
+#define FUSE_DIRECT_IO           (1 << 3)
+
 /** FUSE inode */
 struct fuse_inode {
 	/** Inode data */
@@ -207,6 +210,9 @@ struct fuse_conn {
 	/** Maximum read size */
 	unsigned max_read;
 
+	/** Maximum write size */
+	unsigned max_write;
+
 	/** Readers of the connection are waiting on this */
 	wait_queue_head_t waitq;
 
diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
index 298c1d4c153..652c9d5df97 100644
--- a/fs/fuse/inode.c
+++ b/fs/fuse/inode.c
@@ -258,6 +258,7 @@ enum {
 	OPT_DEFAULT_PERMISSIONS,
 	OPT_ALLOW_OTHER,
 	OPT_KERNEL_CACHE,
+	OPT_DIRECT_IO,
 	OPT_MAX_READ,
 	OPT_ERR
 };
@@ -270,6 +271,7 @@ static match_table_t tokens = {
 	{OPT_DEFAULT_PERMISSIONS,	"default_permissions"},
 	{OPT_ALLOW_OTHER,		"allow_other"},
 	{OPT_KERNEL_CACHE,		"kernel_cache"},
+	{OPT_DIRECT_IO,			"direct_io"},
 	{OPT_MAX_READ,			"max_read=%u"},
 	{OPT_ERR,			NULL}
 };
@@ -329,6 +331,10 @@ static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
 			d->flags |= FUSE_KERNEL_CACHE;
 			break;
 
+		case OPT_DIRECT_IO:
+			d->flags |= FUSE_DIRECT_IO;
+			break;
+
 		case OPT_MAX_READ:
 			if (match_int(&args[0], &value))
 				return 0;
@@ -359,6 +365,8 @@ static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 		seq_puts(m, ",allow_other");
 	if (fc->flags & FUSE_KERNEL_CACHE)
 		seq_puts(m, ",kernel_cache");
+	if (fc->flags & FUSE_DIRECT_IO)
+		seq_puts(m, ",direct_io");
 	if (fc->max_read != ~0)
 		seq_printf(m, ",max_read=%u", fc->max_read);
 	return 0;
@@ -489,6 +497,7 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 	fc->max_read = d.max_read;
 	if (fc->max_read / PAGE_CACHE_SIZE < fc->bdi.ra_pages)
 		fc->bdi.ra_pages = fc->max_read / PAGE_CACHE_SIZE;
+	fc->max_write = FUSE_MAX_IN / 2;
 
 	err = -ENOMEM;
 	root = get_root_inode(sb, d.rootmode);