xfs
[Top] [All Lists]

Re: [PATCH] xfs: use s_umount sema in xfs_sync_worker

To: Mark Tinguely <tinguely@xxxxxxx>
Subject: Re: [PATCH] xfs: use s_umount sema in xfs_sync_worker
From: Ben Myers <bpm@xxxxxxx>
Date: Tue, 15 May 2012 14:06:54 -0500
Cc: xfs@xxxxxxxxxxx
In-reply-to: <4FB2A0D6.3050409@xxxxxxx>
References: <20120323174327.GU7762@xxxxxxx> <20120514203449.GE16099@xxxxxxx> <4FB2A0D6.3050409@xxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
On Tue, May 15, 2012 at 01:30:46PM -0500, Mark Tinguely wrote:
> On 05/14/12 15:34, Ben Myers wrote:
> 
> >Protect xfs_sync_worker by using the s_umount semaphore at the read level to
> >provide exclusion with unmount while work is progressing.
> >
> 
> >-                    xfs_log_force(mp, 0);
> >+    if (down_read_trylock(&mp->m_super->s_umount)) {
> 
> Okay, the s_umount has down_write() performed upon it when it is
> created and does not do a up_write until the filesystem mount is
> complete.
> 
> This test is just like the one in grab_super_passive(). The if
> statements have to be separated to know to do the corresponding
> up_read().
> 
> >+            if (!(mp->m_flags&  XFS_MOUNT_RDONLY)) {
> ...
> >+            up_read(&mp->m_super->s_umount);
> 
> Looks good.
> 
> Reveiewed-by: Mark Tinguely <tinguely@xxxxxxx>

Thanks Mark.  xfstests ran in a loop overnight and is still going, so I'm
satisfied.

Signed-off-by: Ben Myers <bpm@xxxxxxx>

<Prev in Thread] Current Thread [Next in Thread>