Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 10 Jul 2017 11:42:53 +0200
From: "Dr. Thomas Orgis" <>
Subject: Re: mpg123: global buffer overflow in III_i_stereo

Am Mon, 10 Jul 2017 09:13:07 +0000
schrieb "Agostino Sarubbo" <>:

> Description:
> mpg123 is a fast console MPEG Audio Player and decoder library.
> The complete ASan output of the issue:
> # mpg123-mpg123 -t $FILE
> ==10588==ERROR: AddressSanitizer: global-buffer-overflow on address 0x7f01025c5cbc at pc 0x7f010229bfe3 bp 0x7ffc988ac5b0 sp 0x7ffc988ac5a8
> READ of size 4 at 0x7f01025c5cbc thread T0

I would like to point out that this only crashes with the
AddressSanitizer. The normal effect is some bad values from adjacent
data being used in numerical computation. No segfault or the like, no
jumps depending on the values.

> Affected version:
> 1.25.0

1.25.1, too

> Fixed version:
> N/A

To be released, I hoped to also fix another lurking issue that in practice
is hidden by padding of a buffer for alignment. If I don't manage that
today, I guess I have to push out 1.25.2 as-is.

> Commit fix:
> N/A

It is fixed with subversion revision 4275.

Can this info be updated? Agostino: You might want to jump to 4278,
too, as that fixes that other left shift of -1 complaint you had.

> CVE:
> CVE-2017-11126

Is this really worth a CVE, though? So far I was only able to see a
crash triggered by the AddressSanitizer. Never from a normal build. So
every build of mpg123 in the wild, except for extremely hardened
distros that build everything with GCC's sanitizers enabled for daily
use, is not affected. Are people running binaries in production with
the sanitizers on?

Thing about this specific kind of input: This is a fuzzed stream that
triggers a code path for MP3 with intensity stereo. These are really
hard to find and I was not able to produce a valid file (using the FhG
mp3enc demo, the only encoder that does i-stereo to my knowledge) that
also triggers this specific array access. So I cannot even say if this
bug causes bad output from valid input as I never observed the valid
input! For this fuzzed input data, the produced output data (sound)
will not be nice, but this is the case with and without the fix.

It's hard to see a security issue here, although of course there
technically is a buffer overflow. That is why I wanted to take some
time to possibly dig out a file that tells me that mpg123 is now
producing correct output. So far, the overflow is averted, but if we do
something sensible now cannot be tested in lack of valid test data that
uses 5 bits for scale factors with intensity stereo.

Alrighty then,


PS: I will do some fuzzing myself to weed out any lurking issues of this
type. Found a fuzzer in my mailbox, from over 10 years ago, when this
was a thing before;-)

Dr. Thomas Orgis
Universit├Ąt Hamburg
RRZ / Basis-Infrastruktur / HPC
Schl├╝terstr. 70
20146 Hamburg
Tel.: 040/42838 8826
Fax: 040/428 38 6270

Download attachment "smime.p7s" of type "application/pkcs7-signature" (4967 bytes)

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.