[Unionfs] unionfs and sys_readahead and squashfs

[Paul Albrecht]

On Fri, 2007-11-09 at 12:39 -0500, Erez Zadok wrote:
> In message <1194629290.3270.2.camel at albrecht.rdi1.com>, Paul Albrecht writes:
> > On Thu, 2007-11-08 at 20:55 -0500, Erez Zadok wrote: 
> > > Paul, I'm happy to report that I wasn't able to reproduce this bug with
> > > unionfs-2.1.8 and 2.6.24-rc1.
> > > 
> > 
> > OK, Thanks for the update. I didn't explain how I got the kernel oops so
> > to make sure the bug got squashed, here's how I triggered the oops using
> > ubuntu 7.10:
> > 
> > echo "/mnt/bin/bash" > readlist
> > mkdir -p /cow
> > mount -t tmpfs tmpfs /cow
> > mount -t unionfs -o dirs=/cow=rw;/=ro unionfs /mnt
> > readahead-list readlist
> 
> Just to confirm, but on my Fedora Core system, there's just a program called
> /usr/bin/readahead: I think that's the same one b/c it also takes an arg
> which is a file list.  Right?
> 

Yes, readahead-list takes a file list as an argument. The list includes
a file in a union mounted filesystem and that triggers the kernel oops.
If you strace the call to readahead-list, you'll see the call the
readahead syscall. When I do the systrace, I get a segfault which
terminates the readahead syscall.

> > > I've offered to help the ubuntu guys move to unionfs-2.x.  I hope now that
> > > their new release is done, they'd have more time to work with me on this
> > > upgrade.
> > 
> > When ubuntu git repos for kernel and module source get upgraded will that be
> > for "gutsy gibbon" or "hardy heron?"
> 
> I don't know.  My understanding was that the apparmor patches are an issue.
> 
> > Paul Albrecht
> 

I think this is really a question for Phillip as I think he is a member
of the canonical kernel team.

Phillip, Any idea? Anything I can do to help?

Paul Albrecht