Skip to contentSkip to conversation

ext4: optimize extent selection for block removal in case of hole punch

PATCH Applied [email protected] 2 participants 6 messages May 24, 2013 – Jun 23, 2013 ext4 file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ext4: optimize extent selection for block removal in case of hole punch

From: Ashish Sangwan <[email protected]>

Both hole punch and truncate use ext4_ext_rm_leaf for removing blocks. Currently we choose the last extent as the starting point for removing blocks: ex = EXT_LAST_EXTENT(eh); This is OK for truncate but for hole punch we can optimize the extent selection as the path is already initialized. We could use this information to select proper starting extent. The code change in this patch will not affect truncate as for truncate path[depth].p_ext will always be NULL.

---
 fs/ext4/extents.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
fs/ext4/extents.c +3 -1
index 3676dae..7c1a5d3 100644
@@ -2490,7 +2490,9 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
 		return -EIO;
 	}
 	/* find where to start removing */
-	ex = EXT_LAST_EXTENT(eh);
+	ex = path[depth].p_ext;
+	if (!ex)
+		ex = EXT_LAST_EXTENT(eh);
 
 	ex_ee_block = le32_to_cpu(ex->ee_block);
 	ex_ee_len = ext4_ext_get_actual_len(ex);
Signature · 1 line
1.7.2.3
maintainer Applied
#1 lore raw
Quoted 11 lines · On Fri, May 24, 2013 at 08:09:17PM +0530, [email protected] wrote:

From: Ashish Sangwan <[email protected]>

Both hole punch and truncate use ext4_ext_rm_leaf for removing blocks. Currently we choose the last extent as the starting point for removing blocks: ex = EXT_LAST_EXTENT(eh); This is OK for truncate but for hole punch we can optimize the extent selection as the path is already initialized. We could use this information to select proper starting extent. The code change in this patch will not affect truncate as for truncate path[depth].p_ext will always be NULL.

Quoted 2 lines

Applied, thanks.

- Ted

#2 lore raw
Quoted 12 lines · Theodore Ts'o
Quoted 11 lines · On Fri, May 24, 2013 at 08:09:17PM +0530, [email protected] wrote:

From: Ashish Sangwan <[email protected]>

Both hole punch and truncate use ext4_ext_rm_leaf for removing blocks. Currently we choose the last extent as the starting point for removing blocks: ex = EXT_LAST_EXTENT(eh); This is OK for truncate but for hole punch we can optimize the extent selection as the path is already initialized. We could use this information to select proper starting extent. The code change in this patch will not affect truncate as for truncate path[depth].p_ext will always be NULL.

Quoted 4 lines
Quoted 2 lines

Applied, thanks.

Hi Ted,

Sorry I cannot see the patch changes in ext4 dev branch.

Regards,
Ashish

Quoted 2 lines

- Ted

maintainer
#3 lore raw
Quoted 13 lines · Ashish Sangwan
Quoted 12 lines · Theodore Ts'o
Quoted 11 lines · On Fri, May 24, 2013 at 08:09:17PM +0530, [email protected] wrote:

From: Ashish Sangwan <[email protected]>

Both hole punch and truncate use ext4_ext_rm_leaf for removing blocks. Currently we choose the last extent as the starting point for removing blocks: ex = EXT_LAST_EXTENT(eh); This is OK for truncate but for hole punch we can optimize the extent selection as the path is already initialized. We could use this information to select proper starting extent. The code change in this patch will not affect truncate as for truncate path[depth].p_ext will always be NULL.

Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

Applied, thanks.

Sorry I cannot see the patch changes in ext4 dev branch.

Sorry, I dropped this patch from the dev branch last night, but I didn't want to send e-mail about it until I had completed enough testing to be sure. It appears that this patch is causing a regression; xfstests generic/269 and generic/279 to fail in the nojournal configuration.

The tests are ones which have multiple fsstress threads racing with dd/ENOSPC hitters, with (#270) and without (#269) quota enabled. It's not at all obvious to me why your particular change would make a difference here, and it may simply be that your optimization is exposing a timing change and is not the root cause of the failure, but I'm going to move this to the unstable portion of the patch series until we do further investigation.

If you could take a look at this, I would appreciate it, but as I said, this may very well turn out not be the fault of your patch.

Regards,

- Ted

#4 lore raw
Quoted 19 lines · Theodore Ts'o
Quoted 18 lines · Ashish Sangwan
Quoted 16 lines · Theodore Ts'o
Quoted 11 lines · On Fri, May 24, 2013 at 08:09:17PM +0530, [email protected] wrote:

From: Ashish Sangwan <[email protected]>

Both hole punch and truncate use ext4_ext_rm_leaf for removing blocks. Currently we choose the last extent as the starting point for removing blocks: ex = EXT_LAST_EXTENT(eh); This is OK for truncate but for hole punch we can optimize the extent selection as the path is already initialized. We could use this information to select proper starting extent. The code change in this patch will not affect truncate as for truncate path[depth].p_ext will always be NULL.

Quoted 2 lines

Applied, thanks.

Quoted 8 lines
Quoted 1 line

Sorry I cannot see the patch changes in ext4 dev branch.

Sorry, I dropped this patch from the dev branch last night, but I didn't want to send e-mail about it until I had completed enough testing to be sure. It appears that this patch is causing a regression; xfstests generic/269 and generic/279 to fail in the nojournal configuration.

Dropping this patch makes sense until root cause is not obvious.

Quoted 11 lines

The tests are ones which have multiple fsstress threads racing with dd/ENOSPC hitters, with (#270) and without (#269) quota enabled. It's not at all obvious to me why your particular change would make a difference here, and it may simply be that your optimization is exposing a timing change and is not the root cause of the failure, but I'm going to move this to the unstable portion of the patch series until we do further investigation.

If you could take a look at this, I would appreciate it, but as I said, this may very well turn out not be the fault of your patch.

Sure, I will try looking into it.

Regards,
Ashish

Quoted 4 lines

Regards,

- Ted

maintainer
#5 lore raw

Lukas's patch: "ext4: only zero partial blocks in ext4_zero_partial_blocks()" appears to address the failures I was seeing with xfstests 269 and 270, so I'm going to add this patch back to the ext4 git tree.

Regards,

- Ted