Re: ext3 issue on 3.6.1

[NeilBrown]

On Fri, 19 Oct 2012 00:08:09 +0200 Jan Kara <jack@xxxxxxx> wrote:

> On Thu 18-10-12 23:40:25, Paul Bolle wrote:
> > On Thu, 2012-10-18 at 23:23 +0200, Jan Kara wrote:
> > > On Fri 12-10-12 14:57:55, Fabio Coatti wrote:
> > > > [13031.051521] ------------[ cut here ]------------
> > > > [13031.051576] WARNING: at fs/inode.c:280 drop_nlink+0x1b/0x35()
> > > > [13031.051624] Hardware name: ProLiant BL465c G7
> > > > [13031.051668] Pid: 3344, comm: php Tainted: G        W
> > > > 3.6.1-1000hz-preempt #2
> > > > [13031.051746] Call Trace:
> > > > [13031.051787]  [<ffffffff810578c4>] ? warn_slowpath_common+0x73/0x87
> > > > [13031.051837]  [<ffffffff810ec628>] ? drop_nlink+0x1b/0x35
> > > > [13031.051885]  [<ffffffff8118ad51>] ? nfs_dentry_iput+0x33/0x49
> > > > [13031.051934]  [<ffffffff810ea920>] ? d_kill+0xe8/0x108
> > > > [13031.051980]  [<ffffffff810eb001>] ? dput+0x147/0x154
> > > > [13031.052027]  [<ffffffff810d9e46>] ? __fput+0x19a/0x1b2
> > > > [13031.052073]  [<ffffffff8106bdf0>] ? task_work_run+0x4c/0x60
> > > > [13031.052123]  [<ffffffff815ff5e8>] ? int_signal+0x12/0x17
> > > > [13031.052169] ---[ end trace e60232a455c8e2dd ]---
> > >   And this seems unrelated - likely an NFS problem... Let's sort this out
> > > if you still see it after ext3 issue is solved.
> > 
> > Looks rather similar too https://lkml.org/lkml/2012/8/29/165 , doesn't
> > it?
>   Yup. I wonder why that patch didn't get merged. Neil?
> 
> 								Honza

Don't know.  Maybe I slipped under Trond's radar some how.

Trond:  can you comment on and hopefully apply this patch?

Subject of original email was "WARNING: at fs/inode.c:280 drop_nlink+0x31/0x33()


Thanks,
NeilBrown

------
Subject: NFS: avoid warning from nfs_drop_nlink

If you remove a file which is open, NFS will 'silly-rename' it to a
hidden file.
If you then remove that hidden file, and then close the open file,
then nfs_dentry_iput will perform an extra drop_nlink().
Since 3.3-rc1, this has produced a warning.
The simplest way to suppress it is to use "nfs_drop_nlink" which
checks for i_nlink being zero.

Reviewed-by: Jeff Layton <jlayton@xxxxxxxxxx>
Signed-off-by: NeilBrown <neilb@xxxxxxx>

diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c
index 627f108..268af03 100644
--- a/fs/nfs/dir.c
+++ b/fs/nfs/dir.c
@@ -1174,7 +1174,7 @@ static void nfs_dentry_iput(struct dentry *dentry,
struct inode *inode) NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
 
 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
-		drop_nlink(inode);
+		nfs_drop_nlink(inode);
 		nfs_complete_unlink(dentry, inode);
 	}
 	iput(inode);

Attachment: signature.asc
Description: PGP signature