On Thu, 2012-12-13 at 06:31 -0500, Jeff Layton wrote:
> On Mon, 12 Nov 2012 09:52:55 -0500
> Jeff Layton <jlayton@xxxxxxxxxx> wrote:
>
> > On Sat, 27 Oct 2012 20:19:38 +0800
> > Ricky Ng-Adam <rngadam@xxxxxxxxxxx> wrote:
> >
> > > A one time warning after quite a bit of usage with no other problems:
> > >
> > > ------------[ cut here ]------------
> > > WARNING: at /home/rngadam/lophilo/linux/fs/inode.c:280 drop_nlink+0x54/0x60()
> > > Modules linked in: ipt_REDIRECT xt_tcpudp iptable_nat nf_nat
> > > nf_conntrack_ipv4 nf_defrag_ipv4 nf_conntrack ip_tables x_tables
> > > snd_atmel_ac97c snd_ac97_codec ac97_bus snd_pcm snd_page_alloc
> > > snd_timer snd soundcore lophilo(O) ipv6 autofs4
> > > [<c000dfdc>] (unwind_backtrace+0x0/0xf0) from [<c0017784>]
> > > (warn_slowpath_common+0x4c/0x64)
> > > [<c0017784>] (warn_slowpath_common+0x4c/0x64) from [<c00177b8>]
> > > (warn_slowpath_null+0x1c/0x24)
> > > [<c00177b8>] (warn_slowpath_null+0x1c/0x24) from [<c00a9084>]
> > > (drop_nlink+0x54/0x60)
> > > [<c00a9084>] (drop_nlink+0x54/0x60) from [<c0131268>]
> > > (nfs_dentry_iput+0x38/0x50)
> > > [<c0131268>] (nfs_dentry_iput+0x38/0x50) from [<c00a67b0>] (d_kill+0x9c/0xec)
> > > [<c00a67b0>] (d_kill+0x9c/0xec) from [<c00a6cf0>] (dput+0x78/0x134)
> > > [<c00a6cf0>] (dput+0x78/0x134) from [<c0095128>] (fput+0x194/0x238)
> > > [<c0095128>] (fput+0x194/0x238) from [<c0092380>] (filp_close+0x68/0x80)
> > > [<c0092380>] (filp_close+0x68/0x80) from [<c001ab88>]
> > > (put_files_struct+0xb4/0xd8)
> > > [<c001ab88>] (put_files_struct+0xb4/0xd8) from [<c001ad1c>]
> > > (do_exit+0x134/0x72c)
> > > [<c001ad1c>] (do_exit+0x134/0x72c) from [<c001b59c>] (do_group_exit+0x40/0xac)
> > > [<c001b59c>] (do_group_exit+0x40/0xac) from [<c001b618>]
> > > (__wake_up_parent+0x0/0x18)
> > > ---[ end trace 3bda0ec3b276e040 ]---
> > >
> > > client:
> > >
> > > Linux lophilo 3.4.4+ #34 Sat Oct 6 16:28:57 CST 2012 armv5tejl GNU/Linux
> > >
> > > server:
> > >
> > > Linux rngadam-think 3.5.0-17-generic #28-Ubuntu SMP Tue Oct 9 19:32:08
> > > UTC 2012 i686 i686 i686 GNU/Linux
> > >
> > > # exportfs -v
> > > /home/rngadam/lophilo.nfs
> > > <world>(rw,async,wdelay,no_root_squash,no_subtree_check)
> > > /home/rngadam/lophilo
> > > <world>(rw,wdelay,root_squash,all_squash,no_subtree_check,anonuid=0,anongid=0)
> > >
> > > seems to have occurred while a bunch of faiilng lstat on long paths
> > > were going on:
> > >
> > > stat { [Error: ENOENT, lstat
> > > '/home/lophilo/lophilo/lmc/users/rngadam/lophilojs-example.git/client/code/colorpicker/img/colorpicker/img/colorpicker/img/colorpicker/js/colorpicker/css/colorpicker']
> > > errno: 34,
> > > code: 'ENOENT',
> > > path: '/home/lophilo/lophilo/lmc/users/rngadam/lophilojs-example.git/client/code/colorpicker/img/colorpicker/img/colorpicker/img/colorpicker/js/colorpicker/css/colorpicker'
> > > }
> > >
> >
> > The warning is fairly harmless, but it does look scary. Neil Brown had
> > a patch to fix it, but I don't think Trond ever took it or commented on
> > it.
> >
> > Trond, any thoughts on Neil's one-line patch here?
> >
> > https://lkml.org/lkml/2012/9/25/24
> >
> > (Note that I still have my doubts as to whether CIFS or NFS ought to be
> > manipulating or relying on the i_nlink like this, but the patch looks
> > fairly harmless as an interim fix).
> >
>
> Hi Trond,
>
> I asked about this a while back and never got a response. I have a
> few bugs open against Fedora on this and wouldn't mind laying this to
> rest. Is there any reason not to use the nfs_drop_nlink() helper in
> nfs_dentry_iput()?
You mean aside from the fact that sb->s_remove_count remains a racy
piece of crap that serves no good purpose for NFS, and yet will continue
to give us grief?
I'd prefer to see a solution that gets rid of
drop_nlink()/clear_nlink()/set_nlink() altogether. They all suck...
--
Trond Myklebust
Linux NFS client maintainer
NetApp
Trond.Myklebust@xxxxxxxxxx
www.netapp.com
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
