NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[Erez Zadok]

I get a "permission denied" when trying to mount a localhost nfsv2/3
exported volume, on v2.6.24-rc4-124-gf194d13.  It works w/ nfsv4 mounting.
It worked fine in 2.6.24-rc3.  Here's a sequence of ops I tried:

# mount -t ext2 /dev/hdb1 /n/lower/b0
# exportfs -o no_root_squash,rw localhost:/n/lower/b0
# mount -t nfs -o nfsvers=3 localhost:/n/lower/b0 /mnt

Erez.

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[J. Bruce Fields]

On Thu, Dec 06, 2007 at 09:20:41PM -0500, Erez Zadok wrote:
> I get a "permission denied" when trying to mount a localhost nfsv2/3
> exported volume, on v2.6.24-rc4-124-gf194d13.  It works w/ nfsv4 mounting.
> It worked fine in 2.6.24-rc3.  Here's a sequence of ops I tried:
> 
> # mount -t ext2 /dev/hdb1 /n/lower/b0
> # exportfs -o no_root_squash,rw localhost:/n/lower/b0
> # mount -t nfs -o nfsvers=3 localhost:/n/lower/b0 /mnt

What do you see if you watch the network traffic in ethereal?

--b.

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[Erez Zadok]

In message <20071207025504.GA3070@fieldses.org>, "J. Bruce Fields" writes:
> On Thu, Dec 06, 2007 at 09:20:41PM -0500, Erez Zadok wrote:
> > I get a "permission denied" when trying to mount a localhost nfsv2/3
> > exported volume, on v2.6.24-rc4-124-gf194d13.  It works w/ nfsv4 mounting.
> > It worked fine in 2.6.24-rc3.  Here's a sequence of ops I tried:
> > 
> > # mount -t ext2 /dev/hdb1 /n/lower/b0
> > # exportfs -o no_root_squash,rw localhost:/n/lower/b0
> > # mount -t nfs -o nfsvers=3 localhost:/n/lower/b0 /mnt
> 
> What do you see if you watch the network traffic in ethereal?
> 
> --b.

Bruce, I'm using nfs-utils-1.0.10-14.fc6 on an FC6 system with all latest
FC6 patches.  Using git-bisect I was able to find the patch which broke it:

commit 2b1e300a9dfc3196ccddf6f1d74b91b7af55e416
Author: Eric W. Biederman <ebiederm@xmission.com>
Date:   Sun Dec 2 00:33:17 2007 +1100

    [NETNS]: Fix /proc/net breakage
    
    Well I clearly goofed when I added the initial network namespace support
    for /proc/net.  Currently things work but there are odd details visible to
    user space, even when we have a single network namespace.
    
    Since we do not cache proc_dir_entry dentries at the moment we can just
    modify ->lookup to return a different directory inode depending on the
    network namespace of the process looking at /proc/net, replacing the
    current technique of using a magic and fragile follow_link method.
    
    To accomplish that this patch:
    - introduces a shadow_proc method to allow different dentries to
      be returned from proc_lookup.
    - Removes the old /proc/net follow_link magic
    - Fixes a weakness in our not caching of proc generic dentries.
    
    As shadow_proc uses a task struct to decided which dentry to return we can
    go back later and fix the proc generic caching without modifying any code
    that uses the shadow_proc method.
    
    Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
    Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
    Cc: Pavel Machek <pavel@ucw.cz>
    Cc: Pavel Emelyanov <xemul@openvz.org>
    Cc: "David S. Miller" <davem@davemloft.net>
    Cc: Ingo Molnar <mingo@elte.hu>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

With the above patch, rpc.mountd is unable to open /proc/fs/nfsd/filehandle.
Strace shows:

open("/proc/fs/nfsd/filehandle", O_RDWR|O_LARGEFILE) = -1 ENOENT (No such file or directory)

Without the above patch, /proc/fs/nfsd is populated with a number of files,
including "filehandle".

Erez.

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[J. Bruce Fields]

On Fri, Dec 07, 2007 at 03:00:13PM -0500, Erez Zadok wrote:
> In message <20071207025504.GA3070@fieldses.org>, "J. Bruce Fields" writes:
> > On Thu, Dec 06, 2007 at 09:20:41PM -0500, Erez Zadok wrote:
> > > I get a "permission denied" when trying to mount a localhost nfsv2/3
> > > exported volume, on v2.6.24-rc4-124-gf194d13.  It works w/ nfsv4 mounting.
> > > It worked fine in 2.6.24-rc3.  Here's a sequence of ops I tried:
> > > 
> > > # mount -t ext2 /dev/hdb1 /n/lower/b0
> > > # exportfs -o no_root_squash,rw localhost:/n/lower/b0
> > > # mount -t nfs -o nfsvers=3 localhost:/n/lower/b0 /mnt
> > 
> > What do you see if you watch the network traffic in ethereal?
> > 
> > --b.
> 
> Bruce, I'm using nfs-utils-1.0.10-14.fc6 on an FC6 system with all latest
> FC6 patches.  Using git-bisect I was able to find the patch which broke it:

Wow, thanks for your work finding that.

> 
> commit 2b1e300a9dfc3196ccddf6f1d74b91b7af55e416
> Author: Eric W. Biederman <ebiederm@xmission.com>
> Date:   Sun Dec 2 00:33:17 2007 +1100
> 
>     [NETNS]: Fix /proc/net breakage
>     
>     Well I clearly goofed when I added the initial network namespace support
>     for /proc/net.  Currently things work but there are odd details visible to
>     user space, even when we have a single network namespace.
>     
>     Since we do not cache proc_dir_entry dentries at the moment we can just
>     modify ->lookup to return a different directory inode depending on the
>     network namespace of the process looking at /proc/net, replacing the
>     current technique of using a magic and fragile follow_link method.
>     
>     To accomplish that this patch:
>     - introduces a shadow_proc method to allow different dentries to
>       be returned from proc_lookup.
>     - Removes the old /proc/net follow_link magic
>     - Fixes a weakness in our not caching of proc generic dentries.
>     
>     As shadow_proc uses a task struct to decided which dentry to return we can
>     go back later and fix the proc generic caching without modifying any code
>     that uses the shadow_proc method.
>     
>     Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
>     Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
>     Cc: Pavel Machek <pavel@ucw.cz>
>     Cc: Pavel Emelyanov <xemul@openvz.org>
>     Cc: "David S. Miller" <davem@davemloft.net>
>     Cc: Ingo Molnar <mingo@elte.hu>
>     Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
>     Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
> 
> With the above patch, rpc.mountd is unable to open /proc/fs/nfsd/filehandle.
> Strace shows:
> 
> open("/proc/fs/nfsd/filehandle", O_RDWR|O_LARGEFILE) = -1 ENOENT (No such file or directory)
> 
> Without the above patch, /proc/fs/nfsd is populated with a number of files,
> including "filehandle".

Those files are actually in a separate filesystem (of type "nfsd") which
is supposed to be mounted on /proc/fs/nfsd/.   So that mount must have
failed in the bad case?  It's not immediately obvious to me what this
patch has to do with that.  Hm.

--b.

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[Erez Zadok]

In message <20071207200655.GD2204@fieldses.org>, "J. Bruce Fields" writes:
> On Fri, Dec 07, 2007 at 03:00:13PM -0500, Erez Zadok wrote:
[...]
> Those files are actually in a separate filesystem (of type "nfsd") which
> is supposed to be mounted on /proc/fs/nfsd/.   So that mount must have
> failed in the bad case?  It's not immediately obvious to me what this
> patch has to do with that.  Hm.

Yes, it is indeed a separate mount in both cases, but in the broken case,
/proc/fs/nfsd is empty.

The patch in question introduces a proc ->d_revalidate method which does
this:

static int proc_revalidate_dentry(struct dentry *dentry, struct nameidata *nd)
{
	d_drop(dentry);
	return 0;
}

I'm not sure why it drops the dentry and then returns OK to the VFS; is it
to force the VFS to revalidate the dentry?  In that case, I think it should
return -ESTALE.  I also don't know why /proc needs a ->d_revalidate in the
first place (it was fine up until now).  Perhaps what proc does now is
correct, but its behavior has changed such that nfsd's /proc/fs/nfsd needs
to do something different (like grab an extra dentry ref?).

Anyway, if I comment out the d_drop line in proc_revalidate_dentry, or
remove proc's ->d_revalidate method, nfs exporting works again.

Someone more familiar with this patch and /proc should investigate.  Until
then, nfsv2/3 exporting are broken in 2.6.24-rc4.

> --b.

Cheers,
Erez.

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 1099 bytes --]

On Fri, 07 Dec 2007 15:06:55 EST, "J. Bruce Fields" said:
> On Fri, Dec 07, 2007 at 03:00:13PM -0500, Erez Zadok wrote:

> > commit 2b1e300a9dfc3196ccddf6f1d74b91b7af55e416
> > Author: Eric W. Biederman <ebiederm@xmission.com>
> > Date:   Sun Dec 2 00:33:17 2007 +1100

> Those files are actually in a separate filesystem (of type "nfsd") which
> is supposed to be mounted on /proc/fs/nfsd/.   So that mount must have
> failed in the bad case?  It's not immediately obvious to me what this
> patch has to do with that.  Hm.

For what it's worth, the startup scripts for 'wine' and 'jexec' fail as well,
and at shutdown, I get complaints about being unable to dismount 
/proc/sys/fs/binfmt_misc. 'strace' says:

open("/proc/sys/fs/binfmt_misc/windows", O_WRONLY|O_CREAT|O_TRUNC, 0666) = -1 EN
OENT (No such file or directory)

# grep binfmt /proc/mounts
none /proc/sys/fs/binfmt_misc binfmt_misc rw 0 0
none /proc/sys/fs/binfmt_misc binfmt_misc rw 0 0

I have to wonder if the fact there's 2 of them mounted is related to the
fact that 2 startup scripts tried to touch files under fs/binfmt_misc/ ?


[-- Attachment #2: Type: application/pgp-signature, Size: 226 bytes --]

Re: [NFS] NFSv2/3 broken exporting/mounting (permission denied) in 2.6.24-rc4

[Erez Zadok]

In message <20071207025504.GA3070@fieldses.org>, "J. Bruce Fields" writes:
> On Thu, Dec 06, 2007 at 09:20:41PM -0500, Erez Zadok wrote:
> > I get a "permission denied" when trying to mount a localhost nfsv2/3
> > exported volume, on v2.6.24-rc4-124-gf194d13.  It works w/ nfsv4 mounting.
> > It worked fine in 2.6.24-rc3.  Here's a sequence of ops I tried:
> > 
> > # mount -t ext2 /dev/hdb1 /n/lower/b0
> > # exportfs -o no_root_squash,rw localhost:/n/lower/b0
> > # mount -t nfs -o nfsvers=3 localhost:/n/lower/b0 /mnt
> 
> What do you see if you watch the network traffic in ethereal?
> 
> --b.

Bruce, I'm using nfs-utils-1.0.10-14.fc6 on an FC6 system with all latest
FC6 patches.  Using git-bisect I was able to find the patch which broke it:

commit 2b1e300a9dfc3196ccddf6f1d74b91b7af55e416
Author: Eric W. Biederman <ebiederm@xmission.com>
Date:   Sun Dec 2 00:33:17 2007 +1100

    [NETNS]: Fix /proc/net breakage
    
    Well I clearly goofed when I added the initial network namespace support
    for /proc/net.  Currently things work but there are odd details visible to
    user space, even when we have a single network namespace.
    
    Since we do not cache proc_dir_entry dentries at the moment we can just
    modify ->lookup to return a different directory inode depending on the
    network namespace of the process looking at /proc/net, replacing the
    current technique of using a magic and fragile follow_link method.
    
    To accomplish that this patch:
    - introduces a shadow_proc method to allow different dentries to
      be returned from proc_lookup.
    - Removes the old /proc/net follow_link magic
    - Fixes a weakness in our not caching of proc generic dentries.
    
    As shadow_proc uses a task struct to decided which dentry to return we can
    go back later and fix the proc generic caching without modifying any code
    that uses the shadow_proc method.
    
    Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
    Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
    Cc: Pavel Machek <pavel@ucw.cz>
    Cc: Pavel Emelyanov <xemul@openvz.org>
    Cc: "David S. Miller" <davem@davemloft.net>
    Cc: Ingo Molnar <mingo@elte.hu>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

With the above patch, rpc.mountd is unable to open /proc/fs/nfsd/filehandle.
Strace shows:

open("/proc/fs/nfsd/filehandle", O_RDWR|O_LARGEFILE) = -1 ENOENT (No such file or directory)

Without the above patch, /proc/fs/nfsd is populated with a number of files,
including "filehandle".

Erez.