Bug#867579: libopenmpt: Security updates libopenmpt-0.2.7386-beta20.3-p10 available

[James Cowgill]

Control: severity -1 grave
Control: tags -1 fixed-upstream

Hi,

On 07/07/17 15:41, Jörn Heusipp wrote:
> Source: libopenmpt
> Version: 0.2.7386~beta20.3-3
> Severity: important
> Tags: upstream
> 
> Dear Maintainer,
> 
> A couple of security-related fixes have been released upstream as
> version 0.2.7386-beta20.3-p10. See
> https://lib.openmpt.org/libopenmpt/md_announce-2017-07-07.html .
> 
> p10 fixes a heap buffer overflow which allows an attacker to write
> arbitrary data to an arbitrarily choosen offset. It can be triggered
> with a maliciously modified PSM file. This needs to be fixed ASAP via
> a security update in Stretch. The bug happens due to 2 samples in a
> PSM file using the same sample slot in libopenmpt, whereby the second
> sample uses an invalid offset inside the file. That way, the second
> sample did not re-allocate (via
> sampleHeader.GetSampleFormat().ReadSample(Samples[smp], file); deeper
> down the call chain in SampleIO.cpp:73) the sample buffer itself but
> only set the sample size metadata
> (sampleHeader.ConvertToMPT(Samples[smp]);, ultimately at
> Load_psm.cpp:1054). Later, as a loading post-processing step,
> Sndfile.cpp:411 calls PrecomputeLoops() which writes a couple of
> samples before and after the actual sample data (the amount is
> statically known (InterpolationMaxLookahead) and accounted for when
> allocating the sample buffer). However, due to the sample buffer and
> sample length mismatch caused by the bug, this can write extrapolated
> sample data to an arbitary location offset from the first sample's
> buffer (PrecomputeLoopsImpl<T>() in modsmp_ctrl.cpp:263).

Firstly, sorry it's taken some time for me to get around to this. Since
this bug had the potential for remote code execution and looked pretty
serious, I requested a CVE number for it and it has been assigned
CVE-2017-11311.

> p8 is an out-of-bounds read directly after a heap-allocated allocated
> buffer. It is difficult to trigger in practice because std::vector
> does grow its buffer exponentially.

OK this should be fixed as well.

> p9 fixes another potential race condition due to the use of non
> thread-safe <time.h> functions. As discussed previously in
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=864195#67 , this
> again can at worst cause wrong data to be returned for date metadata
> in libopenmpt. However, please note that the same, now rewritten code
> path, could also trigger an assertion failure in glibc under memory
> pressure (which probably is a glibc bug, see
> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=867283 ), thereby
> causing the application to crash.

Again, I'm not sure if it is worth fixing this in stretch - it modifies
quite a bit of code. The glibc bug is important, but I'm not sure it
should be worked around in libopenmpt. It's also mitigated by the fact
that on Linux, if you're suffering from memory pressure, something is
probably about to be killed by the OOM killer anyway.

Thanks,
James

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: OpenPGP digital signature
URL: <http://lists.alioth.debian.org/pipermail/pkg-multimedia-maintainers/attachments/20170714/b0a43fc4/attachment.sig>