Re: [PATCH 6/6] xfs: refactor xfs_file_fallocate

On Wed, Aug 21, 2024 at 02:57:56PM +0200, Christoph Hellwig wrote:

On Wed, Aug 21, 2024 at 08:44:31AM -0400, Brian Foster wrote:

quoted

quoted

+	error = xfs_reflink_unshare(XFS_I(inode), offset, len);
+	if (error)
+		return error;
+

Doesn't unshare imply alloc?

Yes, ooks like that got lost and no test noticed it

quoted

quoted

-	if (xfs_file_sync_writes(file))
+	if (!error && xfs_file_sync_writes(file))
 		error = xfs_log_force_inode(ip);

I'd think if you hit -ENOSPC or something after doing a partial alloc to
a sync inode, you'd still want to flush the changes that were made..?

Persistence behavior on error is always undefined.  And that's also
what the current code does, as it jumps past the log force from all
error exits.

Ok, if this preserves existing behavior then I'm not too worried about
it. Thanks.

Brian