Home page logo

oss-sec logo oss-sec mailing list archives

Re: CVE Request -- libsndfile -- Integer overflow by processing certain PAF files
From: Erik de Castro Lopo <erikd () mega-nerd com>
Date: Fri, 15 Jul 2011 10:47:44 +1000

Please CC me on all mails regarding this bug. I am not on the list
where Dan Rosenberg wrote:

On Thu, Jul 14, 2011 at 2:49 AM, Erik de Castro Lopo
<erikd (AT) mega-nerd (DOT) com> wrote:
Jan Lieskovsky wrote:

* *an integer overflow, leading to heap-based buffer overflow flaw was
found in the way libsndfile, library for reading and writing of sound
files, processed certain PARIS Audio Format (PAF) audio files with
crafted count of channels in the PAF file header. A remote attacker
could provided a specially-crafted PAF audio file, which once opened by
a local, unsuspecting user in an application, linked against libsndfile,
could lead to that particular application crash (denial of service),

I agree with everything up to here.

or, potentially arbitrary code execution with the privileges of the
user running the application.

but this is rubbish. The heap gets overwritten with zeros which would
certainly lead to the application segfaulting. However, there is
no way for arbitrary code to be executed on amy sane OS with proper
memory protection.

This is not a sound assumption. Any sort of partially controlled heap
corruption, even if the data that's being written isn't controllable
by an attacker, should be considered potentially exploitable. Modern
heap exploitation is alive and well - it's worth pointing out that a
recent remote vulnerability in Microsoft IIS FTPD that allowed for a
heap overflow of strictly 0xff bytes was shown to be exploitable,
contradicting Microsoft's claims that it could only cause denial of

The code which caused the heap overflow was this:

    memset (ppaf24->samples, 0, ppaf24->samplesperblock * ppaf24->channels) ;

where it was the ppaf24->channels value that was not validated (and
ppaf24->samplesperblock is always 10). In future versions of libsndfile
ppaf24->samplesperblock will be replaced by a compile time constant

That means that the heap is overwritten in blocks that are a multiple
of 10 bytes which makes it significatly more difficult to exploit.

Think about partially overwriting certain elements of heap
metadata, or even heap data, with zeroes. Suppose an application with
heavy function pointer usage was linked against libsndfile, and this
overflow allowed overwriting the least significant bytes of a function
pointer with zeroes and ultimately allowed for controlling execution

For this instance of heap overflow (overwritten in multiples of 10 bytes
with the base being 4 byte aligned), its only possible to zero the lowest
2 bytes of a function pointer (assuming a little endian machine) if it
happens to lie in exactly the right place.

In terms of ease of exploitation, this one has to be in the very difficult

It's better to be safe than sorry.

That's why I rushed out a new release. I do take this seriously, but
I do not like to see the threat exaggerated beyond reason.

Erik de Castro Lopo

  By Date           By Thread  

Current thread:
[ Nmap | Sec Tools | Mailing Lists | Site News | About/Contact | Advertising | Privacy ]