Commit bf7ba8ee authored by Josef Bacik's avatar Josef Bacik Committed by David Sterba

btrfs: fix deadlock with fsync+fiemap+transaction commit

We are hitting the following deadlock in production occasionally

Task 1		Task 2		Task 3		Task 4		Task 5
		fsync(A)
		 start trans
						start commit
				falloc(A)
				 lock 5m-10m
				 start trans
				  wait for commit
fiemap(A)
 lock 0-10m
  wait for 5m-10m
   (have 0-5m locked)

		 have btrfs_need_log_full_commit
		  !full_sync
		  wait_ordered_extents
								finish_ordered_io(A)
								lock 0-5m
								DEADLOCK

We have an existing dependency of file extent lock -> transaction.
However in fsync if we tried to do the fast logging, but then had to
fall back to committing the transaction, we will be forced to call
btrfs_wait_ordered_range() to make sure all of our extents are updated.

This creates a dependency of transaction -> file extent lock, because
btrfs_finish_ordered_io() will need to take the file extent lock in
order to run the ordered extents.

Fix this by stopping the transaction if we have to do the full commit
and we attempted to do the fast logging.  Then attach to the transaction
and commit it if we need to.

CC: stable@vger.kernel.org # 5.15+
Reviewed-by: default avatarFilipe Manana <fdmanana@suse.com>
Signed-off-by: default avatarJosef Bacik <josef@toxicpanda.com>
Signed-off-by: default avatarDavid Sterba <dsterba@suse.com>
parent 97e86631
...@@ -2322,7 +2322,12 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) ...@@ -2322,7 +2322,12 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
*/ */
btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP); btrfs_inode_unlock(inode, BTRFS_ILOCK_MMAP);
if (ret != BTRFS_NO_LOG_SYNC) { if (ret == BTRFS_NO_LOG_SYNC) {
ret = btrfs_end_transaction(trans);
goto out;
}
/* We successfully logged the inode, attempt to sync the log. */
if (!ret) { if (!ret) {
ret = btrfs_sync_log(trans, root, &ctx); ret = btrfs_sync_log(trans, root, &ctx);
if (!ret) { if (!ret) {
...@@ -2330,17 +2335,49 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) ...@@ -2330,17 +2335,49 @@ int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
goto out; goto out;
} }
} }
/*
* At this point we need to commit the transaction because we had
* btrfs_need_log_full_commit() or some other error.
*
* If we didn't do a full sync we have to stop the trans handle, wait on
* the ordered extents, start it again and commit the transaction. If
* we attempt to wait on the ordered extents here we could deadlock with
* something like fallocate() that is holding the extent lock trying to
* start a transaction while some other thread is trying to commit the
* transaction while we (fsync) are currently holding the transaction
* open.
*/
if (!full_sync) { if (!full_sync) {
ret = btrfs_end_transaction(trans);
if (ret)
goto out;
ret = btrfs_wait_ordered_range(inode, start, len); ret = btrfs_wait_ordered_range(inode, start, len);
if (ret) { if (ret)
btrfs_end_transaction(trans); goto out;
/*
* This is safe to use here because we're only interested in
* making sure the transaction that had the ordered extents is
* committed. We aren't waiting on anything past this point,
* we're purely getting the transaction and committing it.
*/
trans = btrfs_attach_transaction_barrier(root);
if (IS_ERR(trans)) {
ret = PTR_ERR(trans);
/*
* We committed the transaction and there's no currently
* running transaction, this means everything we care
* about made it to disk and we are done.
*/
if (ret == -ENOENT)
ret = 0;
goto out; goto out;
} }
} }
ret = btrfs_commit_transaction(trans); ret = btrfs_commit_transaction(trans);
} else {
ret = btrfs_end_transaction(trans);
}
out: out:
ASSERT(list_empty(&ctx.list)); ASSERT(list_empty(&ctx.list));
err = file_check_and_advance_wb_err(file); err = file_check_and_advance_wb_err(file);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment