ext4: async direct IO for holes and fallocate support

For async direct IO that covers holes or fallocate, the end_io
callback function now queued the convertion work on workqueue but
don't flush the work rightaway as it might take too long to afford.

But when fsync is called after all the data is completed, user expects
the metadata also being updated before fsync returns.

Thus we need to flush the conversion work when fsync() is called.
This patch keep track of a listed of completed async direct io that
has a work queued on workqueue.  When fsync() is called, it will go
through the list and do the conversion.

Signed-off-by: Mingming Cao <cmm@us.ibm.com>

5 files changed, 234 insertions, 41 deletions

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index ccb4dbf359c..b491576e11c 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -127,10 +127,11 @@ struct mpage_da_data {
 	int pages_written;
 	int retval;
 };
-
+#define	DIO_AIO_UNWRITTEN	0x1
 typedef struct ext4_io_end {
+	struct list_head	list;		/* per-file finished AIO list */
 	struct inode		*inode;		/* file being written to */
-	unsigned int		flag;		/* sync IO or AIO */
+	unsigned int		flag;		/* unwritten or not */
 	int			error;		/* I/O error code */
 	ext4_lblk_t		offset;		/* offset in the file */
 	size_t			size;		/* size of the extent */
@@ -690,6 +691,11 @@ struct ext4_inode_info {
 	__u16 i_extra_isize;
 
 	spinlock_t i_block_reservation_lock;
+
+	/* completed async DIOs that might need unwritten extents handling */
+	struct list_head i_aio_dio_complete_list;
+	/* current io_end structure for async DIO write*/
+	ext4_io_end_t *cur_aio_dio;
 };
 
 /*
@@ -1419,7 +1425,7 @@ extern int ext4_block_truncate_page(handle_t *handle,
 		struct address_space *mapping, loff_t from);
 extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf);
 extern qsize_t ext4_get_reserved_space(struct inode *inode);
-
+extern int flush_aio_dio_completed_IO(struct inode *inode);
 /* ioctl.c */
 extern long ext4_ioctl(struct file *, unsigned int, unsigned long);
 extern long ext4_compat_ioctl(struct file *, unsigned int, unsigned long);
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index a38e651c004..10a63096a95 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -3033,6 +3033,7 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
 {
 	int ret = 0;
 	int err = 0;
+	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
 
 	ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical"
 		  "block %llu, max_blocks %u, flags %d, allocated %u",
@@ -3045,6 +3046,9 @@ ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode,
 		ret = ext4_split_unwritten_extents(handle,
 						inode, path, iblock,
 						max_blocks, flags);
+		/* flag the io_end struct that we need convert when IO done */
+		if (io)
+			io->flag = DIO_AIO_UNWRITTEN;
 		goto out;
 	}
 	/* DIO end_io complete, convert the filled extent to written */
@@ -3130,6 +3134,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
 	int err = 0, depth, ret, cache_type;
 	unsigned int allocated = 0;
 	struct ext4_allocation_request ar;
+	ext4_io_end_t *io = EXT4_I(inode)->cur_aio_dio;
 
 	__clear_bit(BH_New, &bh_result->b_state);
 	ext_debug("blocks %u/%u requested for inode %lu\n",
@@ -3279,8 +3284,20 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
 	/* try to insert new extent into found leaf and return */
 	ext4_ext_store_pblock(&newex, newblock);
 	newex.ee_len = cpu_to_le16(ar.len);
-	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT)  /* Mark uninitialized */
+	/* Mark uninitialized */
+	if (flags & EXT4_GET_BLOCKS_UNINIT_EXT){
 		ext4_ext_mark_uninitialized(&newex);
+		/*
+		 * io_end structure was created for every async
+		 * direct IO write to the middle of the file.
+		 * To avoid unecessary convertion for every aio dio rewrite
+		 * to the mid of file, here we flag the IO that is really
+		 * need the convertion.
+		 *
+		 */
+		if (io && flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT)
+			io->flag = DIO_AIO_UNWRITTEN;
+	}
 	err = ext4_ext_insert_extent(handle, inode, path, &newex, flags);
 	if (err) {
 		/* free data blocks we just allocated */
diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 07475740b51..2b1531266ee 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -44,6 +44,8 @@
  *
  * What we do is just kick off a commit and wait on it.  This will snapshot the
  * inode to disk.
+ *
+ * i_mutex lock is held when entering and exiting this function
  */
 
 int ext4_sync_file(struct file *file, struct dentry *dentry, int datasync)
@@ -56,6 +58,9 @@ int ext4_sync_file(struct file *file, struct dentry *dentry, int datasync)
 
 	trace_ext4_sync_file(file, dentry, datasync);
 
+	ret = flush_aio_dio_completed_IO(inode);
+	if (ret < 0)
+		goto out;
 	/*
 	 * data=writeback:
 	 *  The caller's filemap_fdatawrite()/wait will sync the data.
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 5633af6a704..118e16ca91d 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3451,6 +3451,8 @@ static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
 	unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
 	int dio_credits;
 
+	ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
+		   inode->i_ino, create);
 	/*
 	 * DIO VFS code passes create = 0 flag for write to
 	 * the middle of file. It does this to avoid block
@@ -3491,55 +3493,152 @@ out:
 	return ret;
 }
 
-#define		DIO_AIO		0x1
-
 static void ext4_free_io_end(ext4_io_end_t *io)
 {
+	BUG_ON(!io);
+	iput(io->inode);
 	kfree(io);
 }
+static void dump_aio_dio_list(struct inode * inode)
+{
+#ifdef	EXT4_DEBUG
+	struct list_head *cur, *before, *after;
+	ext4_io_end_t *io, *io0, *io1;
+
+	if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
+		ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
+		return;
+	}
+
+	ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
+	list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
+		cur = &io->list;
+		before = cur->prev;
+		io0 = container_of(before, ext4_io_end_t, list);
+		after = cur->next;
+		io1 = container_of(after, ext4_io_end_t, list);
+
+		ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
+			    io, inode->i_ino, io0, io1);
+	}
+#endif
+}
 
 /*
- * IO write completion for unwritten extents.
- *
  * check a range of space and convert unwritten extents to written.
  */
-static void ext4_end_dio_unwritten(struct work_struct *work)
+static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
 {
-	ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
 	struct inode *inode = io->inode;
 	loff_t offset = io->offset;
 	size_t size = io->size;
 	int ret = 0;
-	int aio = io->flag & DIO_AIO;
 
-	if (aio)
-		mutex_lock(&inode->i_mutex);
+	ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
+		   "list->prev 0x%p\n",
+	           io, inode->i_ino, io->list.next, io->list.prev);
+
+	if (list_empty(&io->list))
+		return ret;
+
+	if (io->flag != DIO_AIO_UNWRITTEN)
+		return ret;
+
 	if (offset + size <= i_size_read(inode))
 		ret = ext4_convert_unwritten_extents(inode, offset, size);
 
-	if (ret < 0)
+	if (ret < 0) {
 		printk(KERN_EMERG "%s: failed to convert unwritten"
-			"extents to written extents, error is %d\n",
-                       __func__, ret);
+			"extents to written extents, error is %d"
+			" io is still on inode %lu aio dio list\n",
+                       __func__, ret, inode->i_ino);
+		return ret;
+	}
+
+	/* clear the DIO AIO unwritten flag */
+	io->flag = 0;
+	return ret;
+}
+/*
+ * work on completed aio dio IO, to convert unwritten extents to extents
+ */
+static void ext4_end_aio_dio_work(struct work_struct *work)
+{
+	ext4_io_end_t *io  = container_of(work, ext4_io_end_t, work);
+	struct inode *inode = io->inode;
+	int ret = 0;
+
+	mutex_lock(&inode->i_mutex);
+	ret = ext4_end_aio_dio_nolock(io);
+	if (ret >= 0) {
+		if (!list_empty(&io->list))
+			list_del_init(&io->list);
+		ext4_free_io_end(io);
+	}
+	mutex_unlock(&inode->i_mutex);
+}
+/*
+ * This function is called from ext4_sync_file().
+ *
+ * When AIO DIO IO is completed, the work to convert unwritten
+ * extents to written is queued on workqueue but may not get immediately
+ * scheduled. When fsync is called, we need to ensure the
+ * conversion is complete before fsync returns.
+ * The inode keeps track of a list of completed AIO from DIO path
+ * that might needs to do the conversion. This function walks through
+ * the list and convert the related unwritten extents to written.
+ */
+int flush_aio_dio_completed_IO(struct inode *inode)
+{
+	ext4_io_end_t *io;
+	int ret = 0;
+	int ret2 = 0;
 
-	ext4_free_io_end(io);
-	if (aio)
-		mutex_unlock(&inode->i_mutex);
+	if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
+		return ret;
+
+	dump_aio_dio_list(inode);
+	while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
+		io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
+				ext4_io_end_t, list);
+		/*
+		 * Calling ext4_end_aio_dio_nolock() to convert completed
+		 * IO to written.
+		 *
+		 * When ext4_sync_file() is called, run_queue() may already
+		 * about to flush the work corresponding to this io structure.
+		 * It will be upset if it founds the io structure related
+		 * to the work-to-be schedule is freed.
+		 *
+		 * Thus we need to keep the io structure still valid here after
+		 * convertion finished. The io structure has a flag to
+		 * avoid double converting from both fsync and background work
+		 * queue work.
+		 */
+		ret = ext4_end_aio_dio_nolock(io);
+		if (ret < 0)
+			ret2 = ret;
+		else
+			list_del_init(&io->list);
+	}
+	return (ret2 < 0) ? ret2 : 0;
 }
 
-static ext4_io_end_t *ext4_init_io_end (struct inode *inode, unsigned int flag)
+static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
 {
 	ext4_io_end_t *io = NULL;
 
 	io = kmalloc(sizeof(*io), GFP_NOFS);
 
 	if (io) {
+		igrab(inode);
 		io->inode = inode;
-		io->flag = flag;
+		io->flag = 0;
 		io->offset = 0;
 		io->size = 0;
 		io->error = 0;
-		INIT_WORK(&io->work, ext4_end_dio_unwritten);
+		INIT_WORK(&io->work, ext4_end_aio_dio_work);
+		INIT_LIST_HEAD(&io->list);
 	}
 
 	return io;
@@ -3551,19 +3650,31 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
         ext4_io_end_t *io_end = iocb->private;
 	struct workqueue_struct *wq;
 
-	/* if not hole or unwritten extents, just simple return */
-	if (!io_end || !size || !iocb->private)
+	ext_debug("ext4_end_io_dio(): io_end 0x%p"
+		  "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
+ 		  iocb->private, io_end->inode->i_ino, iocb, offset,
+		  size);
+	/* if not async direct IO or dio with 0 bytes write, just return */
+	if (!io_end || !size)
 		return;
+
+	/* if not aio dio with unwritten extents, just free io and return */
+	if (io_end->flag != DIO_AIO_UNWRITTEN){
+		ext4_free_io_end(io_end);
+		iocb->private = NULL;
+		return;
+	}
+
 	io_end->offset = offset;
 	io_end->size = size;
 	wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
 
-	/* We need to convert unwritten extents to written */
+	/* queue the work to convert unwritten extents to written */
 	queue_work(wq, &io_end->work);
 
-        if (is_sync_kiocb(iocb))
-		flush_workqueue(wq);
-
+	/* Add the io_end to per-inode completed aio dio list*/
+	list_add_tail(&io_end->list,
+		 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
 	iocb->private = NULL;
 }
 /*
@@ -3575,8 +3686,10 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
  * If those blocks were preallocated, we mark sure they are splited, but
  * still keep the range to write as unintialized.
  *
- * When end_io call back function called at the last IO complete time,
- * those extents will be converted to written extents.
+ * The unwrritten extents will be converted to written when DIO is completed.
+ * For async direct IO, since the IO may still pending when return, we
+ * set up an end_io call back function, which will do the convertion
+ * when async direct IO completed.
  *
  * If the O_DIRECT write will extend the file then add this inode to the
  * orphan list.  So recovery will truncate it back to the original size
@@ -3595,28 +3708,76 @@ static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
 	loff_t final_size = offset + count;
 	if (rw == WRITE && final_size <= inode->i_size) {
 		/*
- 		 * For DIO we fallocate blocks for holes, we fallocate blocks
- 		 * The fallocated extent for hole is marked as uninitialized
+ 		 * We could direct write to holes and fallocate.
+		 *
+ 		 * Allocated blocks to fill the hole are marked as uninitialized
  		 * to prevent paralel buffered read to expose the stale data
  		 * before DIO complete the data IO.
- 		 * as for previously fallocated extents, ext4 get_block
+		 *
+ 		 * As to previously fallocated extents, ext4 get_block
  		 * will just simply mark the buffer mapped but still
  		 * keep the extents uninitialized.
  		 *
- 		 * At the end of IO, the ext4 end_io callback function
- 		 * will convert those unwritten extents to written,
- 		 *
+		 * for non AIO case, we will convert those unwritten extents
+		 * to written after return back from blockdev_direct_IO.
+		 *
+		 * for async DIO, the conversion needs to be defered when
+		 * the IO is completed. The ext4 end_io callback function
+		 * will be called to take care of the conversion work.
+		 * Here for async case, we allocate an io_end structure to
+		 * hook to the iocb.
  		 */
-		iocb->private = ext4_init_io_end(inode, !is_sync_kiocb(iocb));
-		if (!iocb->private)
-			return -ENOMEM;
+		iocb->private = NULL;
+		EXT4_I(inode)->cur_aio_dio = NULL;
+		if (!is_sync_kiocb(iocb)) {
+			iocb->private = ext4_init_io_end(inode);
+			if (!iocb->private)
+				return -ENOMEM;
+			/*
+			 * we save the io structure for current async
+			 * direct IO, so that later ext4_get_blocks()
+			 * could flag the io structure whether there
+			 * is a unwritten extents needs to be converted
+			 * when IO is completed.
+			 */
+			EXT4_I(inode)->cur_aio_dio = iocb->private;
+		}
+
 		ret = blockdev_direct_IO(rw, iocb, inode,
 					 inode->i_sb->s_bdev, iov,
 					 offset, nr_segs,
 					 ext4_get_block_dio_write,
 					 ext4_end_io_dio);
+		if (iocb->private)
+			EXT4_I(inode)->cur_aio_dio = NULL;
+		/*
+		 * The io_end structure takes a reference to the inode,
+		 * that structure needs to be destroyed and the
+		 * reference to the inode need to be dropped, when IO is
+		 * complete, even with 0 byte write, or failed.
+		 *
+		 * In the successful AIO DIO case, the io_end structure will be
+		 * desctroyed and the reference to the inode will be dropped
+		 * after the end_io call back function is called.
+		 *
+		 * In the case there is 0 byte write, or error case, since
+		 * VFS direct IO won't invoke the end_io call back function,
+		 * we need to free the end_io structure here.
+		 */
+		if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
+			ext4_free_io_end(iocb->private);
+			iocb->private = NULL;
+		} else if (ret > 0)
+			/*
+			 * for non AIO case, since the IO is already
+			 * completed, we could do the convertion right here
+			 */
+			ret = ext4_convert_unwritten_extents(inode,
+								offset, ret);
 		return ret;
 	}
+
+	/* for write the the end of file case, we fall back to old way */
 	return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
 }
 
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 1a03ea98fdd..f095c60b569 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -687,6 +687,8 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
 	ei->i_allocated_meta_blocks = 0;
 	ei->i_delalloc_reserved_flag = 0;
 	spin_lock_init(&(ei->i_block_reservation_lock));
+	INIT_LIST_HEAD(&ei->i_aio_dio_complete_list);
+	ei->cur_aio_dio = NULL;
 
 	return &ei->vfs_inode;
 }
@@ -3375,11 +3377,13 @@ static int ext4_sync_fs(struct super_block *sb, int wait)
 {
 	int ret = 0;
 	tid_t target;
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
 
 	trace_ext4_sync_fs(sb, wait);
-	if (jbd2_journal_start_commit(EXT4_SB(sb)->s_journal, &target)) {
+	flush_workqueue(sbi->dio_unwritten_wq);
+	if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
 		if (wait)
-			jbd2_log_wait_commit(EXT4_SB(sb)->s_journal, target);
+			jbd2_log_wait_commit(sbi->s_journal, target);
 	}
 	return ret;
 }