Upstream bug: 1162782

I've added Jakob to the watch list, I can add others if you ask. TUV 
makes kernel bugs private by default.

Pat

On 11/11/2014 09:52 AM, Jakob Blomer wrote:
> Hi Pat,
>
> I found similar steps with ntfs-3g:
>
> On the server side:
> # Create volume and a file
> dd if=/dev/zero of=ntfs-volume bs=1024 count=$((10*1024))
> mkfs.ntfs -F ntfs-volume
> mkdir /mnt/ntfs
> ntfs-3g ntfs-volume /mnt/ntfs
> echo foo > /mnt/ntfs/foo
>
> # Remount
> umount /mnt/ntfs
> ntfs-3g ntfs-volume /mnt/ntfs
>
> # Export
> echo '/mnt/ntfs *(fsid=1)' > /etc/exports
> exportfs -a
>
>
> On the client side
> mkdir /mnt/nfs
> mount <SERVER>:/mnt/ntfs /mnt/nfs
> find /mnt/nfs -exec cat {} \;  # <-- Server crashes
>
> Cheers,
> Jakob
>
>
> On 11/11/14 16:36, Jakob Blomer wrote:
>> Hi Pat,
>>
>> thanks for the quick reply!  I just created a bugzilla account.
>>
>> The best I can provide quickly uses the cvmfs file system, which is not
>> in the standard repositories but it is a normal fuse module.
>>
>> On the NFS server side:
>>    o) Install cvmfs:
>> yum install
>> https://ecsft.cern.ch/dist/cvmfs/cvmfs-release/cvmfs-release-2-4.el6.noarch.rpm
>> yum install cvmfs
>>
>>    o) Configure and mount cvmfs
>> echo 'CVMFS_HTTP_PROXY=DIRECT' > /etc/cvmfs/default.local
>> echo 'CVMFS_NFS_SOURCE=yes' >> /etc/cvmfs/default.local
>> mkdir /mnt/cvmfs
>> mount -t cvmfs grid.cern.ch /mnt/cvmfs
>>
>>    o) Export via NFS
>> echo '/mnt/cvmfs *(ro,fsid=1)' > /etc/exports
>> exportfs -ra
>>
>>
>> On the NFS client side:
>> mkdir /mnt/nfs
>> mount <SERVER>:/mnt/cvmfs /mnt/nfs
>> ls /mnt/nfs  # Server crashes
>>
>>
>> Let me see if I can produce something similar with a fuse module that
>> comes in the normal repository.
>>
>> Cheers,
>> Jakob
>>
>>
>> On 11/11/14 15:23, Pat Riehecky wrote:
>>> On 11/11/2014 07:50 AM, Jakob Blomer wrote:
>>>> Dear Scientific Linux team,
>>>>
>>>> the latest kernel 2.6.32-504 crashes if a fuse file system is accessed
>>>> from within the kernel space.  As discovered by Simon, this happens for
>>>> instance when a fuse file system is re-exported via NFS.  When it
>>>> happens, the stack trace starts with
>>>>
>>>> BUG: unable to handle kernel NULL pointer dereference at 0000000000000030
>>>> IP: [<ffffffffa0202260>] fuse_dentry_revalidate+0xf0/0x400 [fuse]
>>>>
>>>>
>>>> The behavior was not present with the previous kernel 2.6.32-431.29.2.
>>>> Besides security patches, the latest kernel seems to also backport new
>>>> fuse features which triggers the problem.  As a workaround, we managed
>>>> to avoid the problem with a small kernel patch [1].
>>>>
>>>> Could you report the problem back to Redhat?  If necessary, I can
>>>> provide detailed steps how to reproduce it.
>>>>
>>>> Cheers,
>>>> Jakob
>>>>
>>>>
>>>> [1]
>>>> https://raw.githubusercontent.com/cvmfs/aufs2-standalone/aufs2.1-32-rhel6.6/rhel6.6-fuse.patch
>>>>
>>> If we can get some more detailed replication steps it should help get
>>> the issue clearly explained to upstream.
>>>
>>> Do you have a bugzilla account?
>>>
>>> Pat
>>>


-- 
Pat Riehecky

Scientific Linux developer
http://www.scientificlinux.org/