ext4: simplify io_end handling for AIO DIO

When mapping blocks for direct IO, we allocate io_end structure before
mapping blocks and store pointer to it in the inode. This creates a
requirement that any AIO DIO using io_end must be protected by i_mutex.
This created problems in the past with dioread_nolock mode which was
corrupting io_end pointers. Also io_end is allocated unnecessarily in
case where we don't need to convert any extents (which is a common case
for example when overwriting file).

We fix the problem by allocating io_end only once we return unwritten
extent from block mapping function for AIO DIO (so we can save some
pointless io_end allocations) and we pass pointer to it in bh->b_private
which generic DIO code later passes to our end IO callback. That way we
remove any need for global pointer to io_end structure and thus fix the
races.

The downside of this change is that the checking for unwritten IO in
flight in ext4_extents_can_be_merged() is more racy since we now
increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after dropping
i_data_sem. However the check has been racy already before because
ext4_writepages() already increment i_unwritten after dropping
i_data_sem and reserved blocks save us from hitting ENOSPC in the worst
case.

Signed-off-by: Jan Kara <jack@suse.cz>

fs/ext4/ext4.h

@@ -1513,16 +1513,6 @@ static inline void ext4_set_io_unwritten_flag(struct inode *inode,
 	}
 }
 
-static inline ext4_io_end_t *ext4_inode_aio(struct inode *inode)
-{
-	return inode->i_private;
-}
-
-static inline void ext4_inode_aio_set(struct inode *inode, ext4_io_end_t *io)
-{
-	inode->i_private = io;
-}
-
 /*
  * Inode dynamic state flags
  */

fs/ext4/extents.c

@@ -1736,6 +1736,12 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
 	 */
 	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
 		return 0;
+	/*
+	 * The check for IO to unwritten extent is somewhat racy as we
+	 * increment i_unwritten / set EXT4_STATE_DIO_UNWRITTEN only after
+	 * dropping i_data_sem. But reserved blocks should save us in that
+	 * case.
+	 */
 	if (ext4_ext_is_unwritten(ex1) &&
 	    (ext4_test_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN) ||
 	     atomic_read(&EXT4_I(inode)->i_unwritten) ||
@@ -4007,7 +4013,6 @@ ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
 	struct ext4_ext_path *path = *ppath;
 	int ret = 0;
 	int err = 0;
-	ext4_io_end_t *io = ext4_inode_aio(inode);
 
 	ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
 		  "block %llu, max_blocks %u, flags %x, allocated %u\n",
@@ -4030,15 +4035,6 @@ ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
 					 flags | EXT4_GET_BLOCKS_CONVERT);
 		if (ret <= 0)
 			goto out;
-		/*
-		 * Flag the inode(non aio case) or end_io struct (aio case)
-		 * that this IO needs to conversion to written when IO is
-		 * completed
-		 */
-		if (io)
-			ext4_set_io_unwritten_flag(inode, io);
-		else
-			ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 		map->m_flags |= EXT4_MAP_UNWRITTEN;
 		goto out;
 	}
@@ -4283,9 +4279,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
 	unsigned int allocated = 0, offset = 0;
 	unsigned int allocated_clusters = 0;
 	struct ext4_allocation_request ar;
-	ext4_io_end_t *io = ext4_inode_aio(inode);
 	ext4_lblk_t cluster_offset;
-	int set_unwritten = 0;
 	bool map_from_cluster = false;
 
 	ext_debug("blocks %u/%u requested for inode %lu\n",
@@ -4482,15 +4476,6 @@ got_allocated_blocks:
 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
 		ext4_ext_mark_unwritten(&newex);
 		map->m_flags |= EXT4_MAP_UNWRITTEN;
-		/*
-		 * io_end structure was created for every IO write to an
-		 * unwritten extent. To avoid unnecessary conversion,
-		 * here we flag the IO that really needs the conversion.
-		 * For non asycn direct IO case, flag the inode state
-		 * that we need to perform conversion when IO is done.
-		 */
-		if (flags & EXT4_GET_BLOCKS_PRE_IO)
-			set_unwritten = 1;
 	}
 
 	err = 0;
@@ -4501,14 +4486,6 @@ got_allocated_blocks:
 		err = ext4_ext_insert_extent(handle, inode, &path,
 					     &newex, flags);
 
-	if (!err && set_unwritten) {
-		if (io)
-			ext4_set_io_unwritten_flag(inode, io);
-		else
-			ext4_set_inode_state(inode,
-					     EXT4_STATE_DIO_UNWRITTEN);
-	}
-
 	if (err && free_on_err) {
 		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
 			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;

fs/ext4/inode.c

@@ -797,18 +797,16 @@ int ext4_dio_get_block(struct inode *inode, sector_t iblock,
 }
 
 /*
- * Get block function for DIO writes when we create unwritten extent if
+ * Get block function for AIO DIO writes when we create unwritten extent if
  * blocks are not allocated yet. The extent will be converted to written
  * after IO is complete.
  */
-static int ext4_dio_get_block_unwritten(struct inode *inode, sector_t iblock,
-			struct buffer_head *bh_result, int create)
+static int ext4_dio_get_block_unwritten_async(struct inode *inode,
+		sector_t iblock, struct buffer_head *bh_result,	int create)
 {
 	handle_t *handle;
 	int ret;
 
-	ext4_debug("ext4_dio_get_block_unwritten: inode %lu, create flag %d\n",
-		   inode->i_ino, create);
 	/* We don't expect handle for direct IO */
 	WARN_ON_ONCE(ext4_journal_current_handle());
 
@@ -818,16 +816,62 @@ static int ext4_dio_get_block_unwritten(struct inode *inode, sector_t iblock,
 	ret = _ext4_get_block(inode, iblock, bh_result,
 			      EXT4_GET_BLOCKS_IO_CREATE_EXT);
 	ext4_journal_stop(handle);
-	if (!ret && buffer_unwritten(bh_result)) {
-		ext4_io_end_t *io_end = ext4_inode_aio(inode);
 
+	/*
+	 * When doing DIO using unwritten extents, we need io_end to convert
+	 * unwritten extents to written on IO completion. We allocate io_end
+	 * once we spot unwritten extent and store it in b_private. Generic
+	 * DIO code keeps b_private set and furthermore passes the value to
+	 * our completion callback in 'private' argument.
+	 */
+	if (!ret && buffer_unwritten(bh_result)) {
+		if (!bh_result->b_private) {
+			ext4_io_end_t *io_end;
+
+			io_end = ext4_init_io_end(inode, GFP_KERNEL);
+			if (!io_end)
+				return -ENOMEM;
+			bh_result->b_private = io_end;
+			ext4_set_io_unwritten_flag(inode, io_end);
+		}
 		set_buffer_defer_completion(bh_result);
-		WARN_ON_ONCE(io_end && !(io_end->flag & EXT4_IO_END_UNWRITTEN));
 	}
 
 	return ret;
 }
 
+/*
+ * Get block function for non-AIO DIO writes when we create unwritten extent if
+ * blocks are not allocated yet. The extent will be converted to written
+ * after IO is complete from ext4_ext_direct_IO() function.
+ */
+static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
+		sector_t iblock, struct buffer_head *bh_result,	int create)
+{
+	handle_t *handle;
+	int ret;
+
+	/* We don't expect handle for direct IO */
+	WARN_ON_ONCE(ext4_journal_current_handle());
+
+	handle = start_dio_trans(inode, bh_result);
+	if (IS_ERR(handle))
+		return PTR_ERR(handle);
+	ret = _ext4_get_block(inode, iblock, bh_result,
+			      EXT4_GET_BLOCKS_IO_CREATE_EXT);
+	ext4_journal_stop(handle);
+
+	/*
+	 * Mark inode as having pending DIO writes to unwritten extents.
+	 * ext4_ext_direct_IO() checks this flag and converts extents to
+	 * written.
+	 */
+	if (!ret && buffer_unwritten(bh_result))
+		ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
+
+	return ret;
+}
+
 static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
 		   struct buffer_head *bh_result, int create)
 {
@@ -3243,7 +3287,7 @@ out:
 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 			    ssize_t size, void *private)
 {
-        ext4_io_end_t *io_end = iocb->private;
+        ext4_io_end_t *io_end = private;
 
 	/* if not async direct IO just return */
 	if (!io_end)
@@ -3251,10 +3295,8 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 
 	ext_debug("ext4_end_io_dio(): io_end 0x%p "
 		  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
- 		  iocb->private, io_end->inode->i_ino, iocb, offset,
-		  size);
+		  io_end, io_end->inode->i_ino, iocb, offset, size);
 
-	iocb->private = NULL;
 	io_end->offset = offset;
 	io_end->size = size;
 	ext4_put_io_end(io_end);
@@ -3290,7 +3332,6 @@ static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	get_block_t *get_block_func = NULL;
 	int dio_flags = 0;
 	loff_t final_size = offset + count;
-	ext4_io_end_t *io_end = NULL;
 
 	/* Use the old path for reads and writes beyond i_size. */
 	if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
@@ -3315,16 +3356,17 @@ static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	/*
 	 * We could direct write to holes and fallocate.
 	 *
-	 * Allocated blocks to fill the hole are marked as
-	 * unwritten to prevent parallel buffered read to expose
-	 * the stale data before DIO complete the data IO.
+	 * Allocated blocks to fill the hole are marked as unwritten to prevent
+	 * parallel buffered read to expose the stale data before DIO complete
+	 * the data IO.
 	 *
-	 * As to previously fallocated extents, ext4 get_block will
-	 * just simply mark the buffer mapped but still keep the
-	 * extents unwritten.
+	 * As to previously fallocated extents, ext4 get_block will just simply
+	 * mark the buffer mapped but still keep the extents unwritten.
 	 *
-	 * For non AIO case, we will convert those unwritten extents
-	 * to written after return back from blockdev_direct_IO.
+	 * For non AIO case, we will convert those unwritten extents to written
+	 * after return back from blockdev_direct_IO. That way we save us from
+	 * allocating io_end structure and also the overhead of offloading
+	 * the extent convertion to a workqueue.
 	 *
 	 * For async DIO, the conversion needs to be deferred when the
 	 * IO is completed. The ext4 end_io callback function will be
@@ -3332,30 +3374,13 @@ static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	 * case, we allocate an io_end structure to hook to the iocb.
 	 */
 	iocb->private = NULL;
-	if (overwrite) {
+	if (overwrite)
 		get_block_func = ext4_dio_get_block_overwrite;
+	else if (is_sync_kiocb(iocb)) {
+		get_block_func = ext4_dio_get_block_unwritten_sync;
+		dio_flags = DIO_LOCKING;
 	} else {
-		ext4_inode_aio_set(inode, NULL);
-		if (!is_sync_kiocb(iocb)) {
-			io_end = ext4_init_io_end(inode, GFP_NOFS);
-			if (!io_end) {
-				ret = -ENOMEM;
-				goto retake_lock;
-			}
-			/*
-			 * Grab reference for DIO. Will be dropped in
-			 * ext4_end_io_dio()
-			 */
-			iocb->private = ext4_get_io_end(io_end);
-			/*
-			 * we save the io structure for current async direct
-			 * IO, so that later ext4_map_blocks() could flag the
-			 * io structure whether there is a unwritten extents
-			 * needs to be converted when IO is completed.
-			 */
-			ext4_inode_aio_set(inode, io_end);
-		}
-		get_block_func = ext4_dio_get_block_unwritten;
+		get_block_func = ext4_dio_get_block_unwritten_async;
 		dio_flags = DIO_LOCKING;
 	}
 #ifdef CONFIG_EXT4_FS_ENCRYPTION
@@ -3370,27 +3395,6 @@ static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 					   get_block_func,
 					   ext4_end_io_dio, NULL, dio_flags);
 
-	/*
-	 * Put our reference to io_end. This can free the io_end structure e.g.
-	 * in sync IO case or in case of error. It can even perform extent
-	 * conversion if all bios we submitted finished before we got here.
-	 * Note that in that case iocb->private can be already set to NULL
-	 * here.
-	 */
-	if (io_end) {
-		ext4_inode_aio_set(inode, NULL);
-		ext4_put_io_end(io_end);
-		/*
-		 * When no IO was submitted ext4_end_io_dio() was not
-		 * called so we have to put iocb's reference.
-		 */
-		if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
-			WARN_ON(iocb->private != io_end);
-			WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
-			ext4_put_io_end(io_end);
-			iocb->private = NULL;
-		}
-	}
 	if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
 						EXT4_STATE_DIO_UNWRITTEN)) {
 		int err;
@@ -3405,7 +3409,6 @@ static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 		ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
 	}
 
-retake_lock:
 	if (iov_iter_rw(iter) == WRITE)
 		inode_dio_end(inode);
 	/* take i_mutex locking again if we do a ovewrite dio */