WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()

WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()

[Jeff Layton]

Over the last several year or so, we've gotten several reports of this sort
of warning message popping up:

[34011.669912] ------------[ cut here ]------------
[34011.670473] WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()
[34011.671198] Hardware name: Bochs
[34011.671564] Root dentry has weird name <>  vfsmnt: fs:nfs4
[34011.672195] Modules linked in: nfsv4 auth_rpcgss nfs lockd sunrpc dns_resolver fscache kvm_amd kvm microcode virtio_net virtio_balloon i2c_piix4 cirrus drm_kms_helper ttm virtio_blk drm i2c_core
[34011.674464] Pid: 7470, comm: ls Tainted: G        W    3.7.0-6.fc19.x86_64 #1
[34011.675417] Call Trace:
[34011.675712]  [<ffffffff8106892f>] warn_slowpath_common+0x7f/0xc0
[34011.676404]  [<ffffffff81068a26>] warn_slowpath_fmt+0x46/0x50
[34011.677058]  [<ffffffff8109eb24>] ? lg_local_lock+0x44/0x80
[34011.677675]  [<ffffffff811e8f18>] ? prepend_path+0x38/0x200
[34011.678315]  [<ffffffff811e90cb>] prepend_path+0x1eb/0x200
[34011.678926]  [<ffffffff811e913c>] path_with_deleted+0x5c/0xa0
[34011.679578]  [<ffffffff811e976f>] d_path+0xbf/0x100
[34011.680139]  [<ffffffff81244bb5>] proc_pid_readlink+0x95/0x100
[34011.680783]  [<ffffffff811d8509>] sys_readlinkat+0xa9/0xc0
[34011.681418]  [<ffffffff81104a5c>] ? __audit_syscall_entry+0xcc/0x300
[34011.682144]  [<ffffffff811d853b>] sys_readlink+0x1b/0x20
[34011.682740]  [<ffffffff816fbd19>] system_call_fastpath+0x16/0x1b
[34011.683425] ---[ end trace 5e8c789abf4f6e53 ]---

This warning was added in commit 98dc568bc.

I've found a way to trigger this at will:

Mount up a nfs mount and cd into it from a shell. Do a lazy umount of
the mount, and then do this from the shell:

    $ ls -l /proc/self/cwd

That will cause that warning message to fire. The mnt will be detached
and will look like the global root, but the d_name.name will be a
zero-length string.

I'm pretty sure there are other ways to trigger this too -- maybe
involving races with shrinkable mounts (not sure on this).

My initial thought was to add "(unreachable)" to the string, a'la
d_path_with_unreachable(), but the changelog on commit 7b2a69ba70 made
me think that that's not such a good idea.

Given that this is fairly easy to reproduce, is there any point in
keeping this warning around or should we just remove it?
-- 
Jeff Layton <jlayton@redhat.com>

[PATCH] vfs: remove warning about "weird" root dentry name from prepend_path()

[Jeff Layton]

We've had several reports of this warning over the last year or so:

[34011.669912] ------------[ cut here ]------------
[34011.670473] WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()
[34011.671198] Hardware name: Bochs
[34011.671564] Root dentry has weird name <>  vfsmnt: fs:nfs4
[34011.672195] Modules linked in: nfsv4 auth_rpcgss nfs lockd sunrpc dns_resolver fscache kvm_amd kvm microcode virtio_net virtio_balloon i2c_piix4 cirrus drm_kms_helper ttm virtio_blk drm i2c_core
[34011.674464] Pid: 7470, comm: ls Tainted: G        W    3.7.0-6.fc19.x86_64 #1
[34011.675417] Call Trace:
[34011.675712]  [<ffffffff8106892f>] warn_slowpath_common+0x7f/0xc0
[34011.676404]  [<ffffffff81068a26>] warn_slowpath_fmt+0x46/0x50
[34011.677058]  [<ffffffff8109eb24>] ? lg_local_lock+0x44/0x80
[34011.677675]  [<ffffffff811e8f18>] ? prepend_path+0x38/0x200
[34011.678315]  [<ffffffff811e90cb>] prepend_path+0x1eb/0x200
[34011.678926]  [<ffffffff811e913c>] path_with_deleted+0x5c/0xa0
[34011.679578]  [<ffffffff811e976f>] d_path+0xbf/0x100
[34011.680139]  [<ffffffff81244bb5>] proc_pid_readlink+0x95/0x100
[34011.680783]  [<ffffffff811d8509>] sys_readlinkat+0xa9/0xc0
[34011.681418]  [<ffffffff81104a5c>] ? __audit_syscall_entry+0xcc/0x300
[34011.682144]  [<ffffffff811d853b>] sys_readlink+0x1b/0x20
[34011.682740]  [<ffffffff816fbd19>] system_call_fastpath+0x16/0x1b
[34011.683425] ---[ end trace 5e8c789abf4f6e53 ]---

I'm able to reproduce this warning at will by mounting up a NFS
filesystem, cd'ing into it. If I then lazy unmount the fs and then stat
/proc/pid/cwd for the process sitting in the detached mount, this
warning will fire.

Given that this is fairly easy to reproduce without doing anything too
exotic, I don't see much value in keeping this warning around.

Cc: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
---
 fs/dcache.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/fs/dcache.c b/fs/dcache.c
index 19153a0..08a10a7 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -2587,15 +2587,6 @@ out:
 	return error;
 
 global_root:
-	/*
-	 * Filesystems needing to implement special "root names"
-	 * should do so with ->d_dname()
-	 */
-	if (IS_ROOT(dentry) &&
-	    (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
-		WARN(1, "Root dentry has weird name <%.*s>\n",
-		     (int) dentry->d_name.len, dentry->d_name.name);
-	}
 	if (!slash)
 		error = prepend(buffer, buflen, "/", 1);
 	if (!error)
-- 
1.7.11.7

Re: [PATCH] vfs: remove warning about "weird" root dentry name from prepend_path()

[Miklos Szeredi]

On Mon, Jan 14, 2013 at 3:54 PM, Jeff Layton <jlayton@redhat.com> wrote:
> We've had several reports of this warning over the last year or so:
>
> [34011.669912] ------------[ cut here ]------------
> [34011.670473] WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()
> [34011.671198] Hardware name: Bochs
> [34011.671564] Root dentry has weird name <>  vfsmnt: fs:nfs4

[...]

> I'm able to reproduce this warning at will by mounting up a NFS
> filesystem, cd'ing into it. If I then lazy unmount the fs and then stat
> /proc/pid/cwd for the process sitting in the detached mount, this
> warning will fire.
>
> Given that this is fairly easy to reproduce without doing anything too
> exotic, I don't see much value in keeping this warning around.

The value of the warning is in preventing filesystems from doing weird
and exotic things without thinking about it.

The obvious question here is what is NFS4 doing differently from other
filesystems to trigger this warning and why.

Thanks,
Miklos

>
> Cc: Miklos Szeredi <miklos@szeredi.hu>
> Signed-off-by: Jeff Layton <jlayton@redhat.com>
> ---
>  fs/dcache.c | 9 ---------
>  1 file changed, 9 deletions(-)
>
> diff --git a/fs/dcache.c b/fs/dcache.c
> index 19153a0..08a10a7 100644
> --- a/fs/dcache.c
> +++ b/fs/dcache.c
> @@ -2587,15 +2587,6 @@ out:
>         return error;
>
>  global_root:
> -       /*
> -        * Filesystems needing to implement special "root names"
> -        * should do so with ->d_dname()
> -        */
> -       if (IS_ROOT(dentry) &&
> -           (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
> -               WARN(1, "Root dentry has weird name <%.*s>\n",
> -                    (int) dentry->d_name.len, dentry->d_name.name);
> -       }
>         if (!slash)
>                 error = prepend(buffer, buflen, "/", 1);
>         if (!error)
> --
> 1.7.11.7
>

Re: [PATCH] vfs: remove warning about "weird" root dentry name from prepend_path()

[Jeff Layton]

On Mon, 14 Jan 2013 14:24:58 -0500
Jeff Layton <jlayton@redhat.com> wrote:

> On Mon, 14 Jan 2013 16:50:20 +0100
> Miklos Szeredi <miklos@szeredi.hu> wrote:
> 
> > On Mon, Jan 14, 2013 at 3:54 PM, Jeff Layton <jlayton@redhat.com> wrote:
> > > We've had several reports of this warning over the last year or so:
> > >
> > > [34011.669912] ------------[ cut here ]------------
> > > [34011.670473] WARNING: at fs/dcache.c:2630 prepend_path+0x1eb/0x200()
> > > [34011.671198] Hardware name: Bochs
> > > [34011.671564] Root dentry has weird name <>  vfsmnt: fs:nfs4
> > 
> > [...]
> > 
> > > I'm able to reproduce this warning at will by mounting up a NFS
> > > filesystem, cd'ing into it. If I then lazy unmount the fs and then stat
> > > /proc/pid/cwd for the process sitting in the detached mount, this
> > > warning will fire.
> > >
> > > Given that this is fairly easy to reproduce without doing anything too
> > > exotic, I don't see much value in keeping this warning around.
> > 
> > The value of the warning is in preventing filesystems from doing weird
> > and exotic things without thinking about it.
> > 
> > The obvious question here is what is NFS4 doing differently from other
> > filesystems to trigger this warning and why.
> > 
> > Thanks,
> > Miklos
> > 
> > >
> > > Cc: Miklos Szeredi <miklos@szeredi.hu>
> > > Signed-off-by: Jeff Layton <jlayton@redhat.com>
> > > ---
> > >  fs/dcache.c | 9 ---------
> > >  1 file changed, 9 deletions(-)
> > >
> > > diff --git a/fs/dcache.c b/fs/dcache.c
> > > index 19153a0..08a10a7 100644
> > > --- a/fs/dcache.c
> > > +++ b/fs/dcache.c
> > > @@ -2587,15 +2587,6 @@ out:
> > >         return error;
> > >
> > >  global_root:
> > > -       /*
> > > -        * Filesystems needing to implement special "root names"
> > > -        * should do so with ->d_dname()
> > > -        */
> > > -       if (IS_ROOT(dentry) &&
> > > -           (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
> > > -               WARN(1, "Root dentry has weird name <%.*s>\n",
> > > -                    (int) dentry->d_name.len, dentry->d_name.name);
> > > -       }
> > >         if (!slash)
> > >                 error = prepend(buffer, buflen, "/", 1);
> > >         if (!error)
> > > --
> > > 1.7.11.7
> > >
> 
> In 3.7-ish kernels, the mnt_root of a NFSv4 mount pointed to a dentry
> with a zero-length name. In 3.8-ish kernels that has changed -- the
> dentry is called '/'. So yeah, it looks like this patch isn't
> necessary.
> 
> The question at this point is why and how that changed. I'm guessing
> I'll have to bisect that down to figure it out. Stay tuned...
> 
> Thanks,

Ok, found it:

commit b911a6bdeef5848c468597d040e3407e0aee04ce
Author: NeilBrown <neilb@suse.de>
Date:   Thu Nov 8 16:09:37 2012 -0800

    vfs: d_obtain_alias() needs to use "/" as default name.
    
Looks like Neil Brown already fixed this the right way. My patch can be
dropped. Sorry for the noise...

Thanks,
-- 
Jeff Layton <jlayton@redhat.com>