[PATCH] xfs: Don't trim extents for fiemap

[PATCH] xfs: Don't trim extents for fiemap

[Nikolay Borisov]

For file extents xfs currently always calls xfs_bmapi_read with not flags,
meaning extents are going to be truncated to the requested range. This is
differs than what other filesystems do (ext4/btrfs don't trim extents). So
harmonize the behavior across the filesystem by explicitly not trimming
extents when we are called from fiemap code.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 fs/xfs/xfs_iomap.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index f179bdf1644d..129550fbf898 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -1008,7 +1008,8 @@ xfs_file_iomap_begin(
 	end_fsb = XFS_B_TO_FSB(mp, offset + length);
 
 	error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
-			       &nimaps, 0);
+			       &nimaps,
+			       flags & IOMAP_REPORT ? XFS_BMAPI_ENTIRE : 0);
 	if (error)
 		goto out_unlock;
 
-- 
2.7.4

Re: [PATCH] xfs: Don't trim extents for fiemap

[Christoph Hellwig]

On Thu, Nov 23, 2017 at 10:06:09AM +0200, Nikolay Borisov wrote:
> For file extents xfs currently always calls xfs_bmapi_read with not flags,
> meaning extents are going to be truncated to the requested range. This is
> differs than what other filesystems do (ext4/btrfs don't trim extents). So
> harmonize the behavior across the filesystem by explicitly not trimming
> extents when we are called from fiemap code.

Given that the iomap code generally trims extents, can you just try
to always enable XFS_BMAPI_ENTIRE?

Re: [PATCH] xfs: Don't trim extents for fiemap

[Nikolay Borisov]

On 23.11.2017 10:15, Christoph Hellwig wrote:
> On Thu, Nov 23, 2017 at 10:06:09AM +0200, Nikolay Borisov wrote:
>> For file extents xfs currently always calls xfs_bmapi_read with not flags,
>> meaning extents are going to be truncated to the requested range. This is
>> differs than what other filesystems do (ext4/btrfs don't trim extents). So
>> harmonize the behavior across the filesystem by explicitly not trimming
>> extents when we are called from fiemap code.
> 
> Given that the iomap code generally trims extents, can you just try
> to always enable XFS_BMAPI_ENTIRE?

What you are saying is that iomap currently always trims the returned
range from the underlying filesystem so doing trimming in the fs is
redundant and I can just pass XFS_BMAPI_ENTIRE indiscriminately ?

> 

Re: [PATCH] xfs: Don't trim extents for fiemap

[Christoph Hellwig]

On Thu, Nov 23, 2017 at 10:28:39AM +0200, Nikolay Borisov wrote:
> What you are saying is that iomap currently always trims the returned
> range from the underlying filesystem so doing trimming in the fs is
> redundant and I can just pass XFS_BMAPI_ENTIRE indiscriminately ?

I think so.  But the theory needs proper valiation first, so please
do a full test run with that change.

Re: [PATCH] xfs: Don't trim extents for fiemap

[Nikolay Borisov]

On 23.11.2017 10:30, Christoph Hellwig wrote:
> On Thu, Nov 23, 2017 at 10:28:39AM +0200, Nikolay Borisov wrote:
>> What you are saying is that iomap currently always trims the returned
>> range from the underlying filesystem so doing trimming in the fs is
>> redundant and I can just pass XFS_BMAPI_ENTIRE indiscriminately ?
> 
> I think so.  But the theory needs proper valiation first, so please
> do a full test run with that change.
> 

Right, so no new failures introduced by always passing XFS_BMAP_ENTIRE
so I will resend the patch.