crash while checking MD5sums on jigdo include list

[Colin Watson]

The build machine behind cdimage.ubuntu.com was recently upgraded to edgy/amd64. While testing the new system, I found that mkisofs segfaulted while trying to build Ubuntu alternate install CD images. Investigation showed that this is due to incorrect linked list code used while constructing the JTE include and exclude lists for jigdo file generation. This is unlikely to affect normal users of mkisofs, but affects those trying to build Debian or Ubuntu CD images. I would like to propose this for a stable release update in order to fix Ubuntu's production CD image building system without the need for a local build of mkisofs.

This is also Debian bug 411873, and I've sent a patch there to fix the linked list generation. I have applied this patch in cdrkit 9:1.1.6-1ubuntu1 in gutsy. Corresponding patches against edgy and feisty are attached here.

Since the crash is rather dependent on precise details of stack layout, it's not necessarily easy to reproduce directly. However, valgrind picks it up with no trouble. Steps to reproduce on an edgy system with mkisofs and valgrind installed (for feisty and gutsy, substitute "genisoimage" for "mkisofs" throughout:

  mkdir -p /tmp/mkisofs-test
  cd /tmp/mkisofs-test
  mkdir -p t/pool
  dd if=/dev/zero of=t/file bs=1024 count=100
  touch md5
  valgrind mkisofs -o t.iso -jigdo-jigdo t.jigdo -jigdo-template t.template -md5-list md5 -jigdo-force-md5 /pool/ t

The output will contain an error similar to the following (exact numbers are unimportant):

  ==8624== Conditional jump or move depends on uninitialised value(s)
  ==8624== at 0x42BB38: (within /usr/bin/mkisofs)
  ==8624== by 0x40E7D6: (within /usr/bin/mkisofs)
  ==8624== by 0x40EBD9: (within /usr/bin/mkisofs)
  ==8624== by 0x40460E: main (in /usr/bin/mkisofs)
  ==8624== by 0x4C570C3: (below main) (in /lib/libc-2.4.so)

The patch affects only JTE code, so (barring regressions due simply to rebuilding cdrtools), other uses of mkisofs will not be affected. Typical uses of mkisofs to build simple ISO images from a tree of files on disk do not use JTE; it is an advanced feature for use by distributors of heavily-downloaded ISO images.

[Colin Watson (cjwatson)]

cdrkit (9:1.1.6-1ubuntu1) gutsy; urgency=low

  * Always initialise next pointer of jigdo include and exclude list
    structures (LP: #130376).

 -- Colin Watson <email address hidden> Sat, 04 Aug 2007 16:16:40 +0100

[Colin Watson (cjwatson)]

(I'm suggesting both Edgy and Feisty SRUs as, while the problem we have in production is currently on Edgy, we might in the future upgrade to Feisty and I'd rather have the security of knowing that we won't have to work around this problem there if we do.)

[Martin Pitt (pitti)]

Thanks, Colin! The patch makes perfect sense. Edgy's patch is fine as it is. Can you please do the XSBC-Original-Maintainer: dance for feisty?

Please go ahead and upload.

[Colin Watson (cjwatson)]

Both uploaded, including the Maintainer change for feisty-proposed.

[Martin Pitt (pitti)]

cdrkit accepted into feisty-proposed, please go ahead with QA testing.

[Martin Pitt (pitti)]

cdrtools accepted into edgy-proposed, please go ahead with QA testing.

[Colin Watson (cjwatson)]

cdrkit (9:1.1.2-1ubuntu1) feisty-proposed; urgency=low

  * Always initialise next pointer of jigdo include and exclude list
    structures (LP: #130376).

 -- Colin Watson <email address hidden> Fri, 10 Aug 2007 10:56:23 +0100

[Colin Watson (cjwatson)]

cdrtools (4:2.01+01a03-5ubuntu3) edgy-proposed; urgency=low

  * debian/patches/34_JTE.dpatch:
    - Always initialise next pointer of jigdo include and exclude list
      structures (LP: #130376).

 -- Colin Watson <email address hidden> Sat, 04 Aug 2007 16:18:54 +0100

[Colin Watson (cjwatson)]

I've asked the Canonical sysadmins to upgrade mkisofs on our build system to this version so that we can test it there.

[Colin Watson (cjwatson)]

ii mkisofs 2.01+01a03-5ubuntu4 Creates ISO-9660 CD-ROM filesystem images

We used this for the 7.10 release builds on antimony ... I'd say that's verified!

[Martin Pitt (pitti)]

Copied to -updates.