[0/3] nfsd: fully close all files in the nfsd threads

Message

NeilBrown Dec. 8, 2023, 3:27 a.m. UTC

This is a new version of my patches to address a rare problem with nfsd
closing files faster than __fput() can complete the close in a different
thread.

This time I'm simply switching to __fput_sync().  I cannot see any
reason that this would be a problem, but if any else does and can show
me what I'm missing, I'd appreciate it.

Thanks,
NeilBrown


 [PATCH 1/3] nfsd: use __fput_sync() to avoid delayed closing of
 [PATCH 2/3] nfsd: Don't leave work of closing files to a work queue.
 [PATCH 3/3] VFS: don't export flush_delayed_fput().

Comments

Jeffrey Layton Dec. 8, 2023, 11:40 a.m. UTC | #1

On Fri, 2023-12-08 at 14:27 +1100, NeilBrown wrote:
> This is a new version of my patches to address a rare problem with nfsd
> closing files faster than __fput() can complete the close in a different
> thread.
> 
> This time I'm simply switching to __fput_sync().  I cannot see any
> reason that this would be a problem, but if any else does and can show
> me what I'm missing, I'd appreciate it.
> 
> Thanks,
> NeilBrown
> 
> 
>  [PATCH 1/3] nfsd: use __fput_sync() to avoid delayed closing of
>  [PATCH 2/3] nfsd: Don't leave work of closing files to a work queue.
>  [PATCH 3/3] VFS: don't export flush_delayed_fput().

Looks good to me.

Reviewed-by: Jeff Layton <[e-mail]>

Jens Axboe Dec. 8, 2023, 2:33 p.m. UTC | #2

On 12/7/23 8:27 PM, NeilBrown wrote:
> This is a new version of my patches to address a rare problem with nfsd
> closing files faster than __fput() can complete the close in a different
> thread.
> 
> This time I'm simply switching to __fput_sync().  I cannot see any
> reason that this would be a problem, but if any else does and can show
> me what I'm missing, I'd appreciate it.

Much better than the previous attempts, imho.

Reviewed-by: Jens Axboe <[e-mail]>