2015-10-01 23:12:29

by Trond Myklebust

Subject: [PATCH] NFS: Fix a write performance regression

If all other conditions in nfs_can_extend_write() are met, and there
are no locks, then we should be able to assume close-to-open semantics
and the ability to extend our write to cover the whole page.

With this patch, the xfstests generic/074 test completes in 232s instead
of >1400s on my test rig.

Fixes: bd61e0a9c852 ("locks: convert posix locks to file_lock_context")
Cc: Jeff Layton <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/write.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 72624dc4a623..1d6f6b43b7f4 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1223,7 +1223,7 @@ static int nfs_can_extend_write(struct file *file, struct page *page, struct ino
return 1;
if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
list_empty_careful(&flctx->flc_posix)))
- return 0;
+ return 1;

/* Check to see if there are whole file write locks */
ret = 0;
--
2.4.3

2015-10-01 23:23:19

by Jeff Layton

Subject: Re: [PATCH] NFS: Fix a write performance regression

On Thu, 1 Oct 2015 19:12:26 -0400
Trond Myklebust <[email protected]> wrote:

> If all other conditions in nfs_can_extend_write() are met, and there
> are no locks, then we should be able to assume close-to-open semantics
> and the ability to extend our write to cover the whole page.
>
> With this patch, the xfstests generic/074 test completes in 232s instead
> of >1400s on my test rig.
>
> Fixes: bd61e0a9c852 ("locks: convert posix locks to file_lock_context")
> Cc: Jeff Layton <[email protected]>
> Signed-off-by: Trond Myklebust <[email protected]>
> ---
> fs/nfs/write.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
> index 72624dc4a623..1d6f6b43b7f4 100644
> --- a/fs/nfs/write.c
> +++ b/fs/nfs/write.c
> @@ -1223,7 +1223,7 @@ static int nfs_can_extend_write(struct file *file, struct page *page, struct ino
> return 1;
> if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
> list_empty_careful(&flctx->flc_posix)))
> - return 0;
> + return 1;
>
> /* Check to see if there are whole file write locks */
> ret = 0;


Ouch -- mea culpa. Actually broken originally in
5263e31e452fb84138b9bee061d5c06c0f359fea, I think...

In any case:

Reviewed-by: Jeff Layton <[email protected]>