[Unionfs] BUG: unable to handle kernel NULL pointer dereference with unionfs and kernel 2.6.21.5

[Erez Zadok]

I took out lkml from the cc list; I don't think it's a generic kernel bug.
At least from your oopses it seems more related to unionfs (but I first have
to reproduce it and then investigate).

In message <200707061047.04222.rene.mayrhofer at gibraltar.at>, Rene Mayrhofer writes:

> Hi lists(s),
> 
> [Please CC me in replies to LKML, I am not subscribed to this list.]
> [Sorry for the cross-post, but I'm not sure that this is only a unionfs issue, 
> so any hints are appreciated.]
> 
> I have a problem with unionfs 2.6.21.5-u3 more or less crashing my system
> upon the first write attempt (to a tmpfs file system mounted over a
> read-only tree from a squashfs image that is mounted loopback as the root
> filesystem).  Attached is the "dmesg" output from bootup to the BUG (which
> happens immediately). After that, only a hard reboot will make anything
> happen anymore...
> 
> The kernel is based on a vanilla 2.6.21.5 from k.o with patches in the 
> following order:
> - pax-linux-2.6.21.5-test9.patch
> - squashfs 3.2-r2: linux-2.6.20/squashfs3.2-patch
> - linux-2.6.21-imq2.diff
> - linux-2.6.21.5-u3.diff.gz (unionfs)
> - netfilter-layer7-v2.9.tar.gz
> - routes-2.6.21-15.diff
> and a very few select pom-ng patches.
> 
> I don't think that the IMQ, routes, layer7, or pom-ng patches can have 
> anything to do with it, and the squashfs patch has been working for ages on 
> different kernel versions. So my only suspicion so far is a conflict between 
> the PAX patch and unionfs. Btw, the same combination of patches worked nicely 
> when based on 2.6.20.x with the respective older patch versions.
> 
> I'd be happy about any hints on solving this.
> 
> Thanks,
> Rene

Rene, I'd like to ask you the following:

1. open a bugzilla report in <https://bugzilla.filesystems.org/>.

2. upload the URLs to the patches above, or the patches themselves, so I can
   build a kernel just as yours.

3. Give me your precise configuration.  Preferably, step by step
   instructions how you build your tmpfs, squashfs image, what you mount,
   loopback, and in what order.  The more info, the better.

I'll try to reproduce your bug on my end here, but the above info will go a
long way towards reproducibility (which is half of the battle to fix any bug
:-)

Thanks,
Erez.