On Sunday 23 February 2003 12:38 pm, Crazy Einstein wrote:
> /*
> \ PoC local exploit for zlib <= 1.1.4
> / just for fun..not for root :)
> > / Usage: gcc -o zlib zlib.c -lz
> > / by CrZ [crazy_einstein (at) yahoo (dot) com [email concealed]] lbyte
> [lbyte.void.ru]
> */

Ok, one simple proof of concept is enough. A second potentially
dangerous one (even for fun)...time to address this. ;)

Attached below is a patch RK and I whipped up yesterday, after I
caught wind of this problem sometime in the afternoon. It adds
extra code to properly gather the vsnprintf() return code if
available, and some ./configure checks to automatically set
macro definitions when it detects the requisite features. zlib
will still build if the host doesn't have the requisite
functions for full security, but ./configure will tell you about
how far you're bending over. The patch went through two
revisions to get to this level of completeness; it works as it
should on Linux==2.4.18/glibc>=2.2.5 but has not been tested on
other platforms.

RK and I both considered just completely dropping the vulnerable
codepaths; environments where zlib would have to fall back to
these codepaths honestly just don't deserve breathing rights.
But...I figure a fix isn't truly robust unless the fixed product
will still build on all the systems where it would build before.
At least now zlib builds secure-where-possible, instead of
broken-by-default.

During zlib ./configure, you should now see the following lines:

Checking whether to use vsnprintf() or snprintf()... using vsnprintf()
Checking for vsnprintf() in stdio.h... Yes.
Checking for return value of vsnprintf()... Yes.

> #include <zlib.h>
> #include <errno.h>
> #include <stdio.h>
<snip harmless but potentially wicked Proof-of-Concept code>
>
>
> [crz@blacksand crz]$ gcc -o zlib zlib.c -lz
> [crz@blacksand crz]$ ./zlib
> [>] exploiting...
> [>] xret = 0xbffff8f0
> sh-2.05b$ exit
> exit
> [crz@blacksand crz]$

On vulnerable system:

[ kelledin@valhalla ~ ] # gcc -o zlibexp zlibexp.c -lz
[ kelledin@valhalla ~ ] # ./zlibexp
[>] exploiting...
[>] xret = 0xbffffaf0
sh-2.05a$ exit
exit
[ kelledin@valhalla ~ ] #

On patched system:

[ kelledin@valhalla /usr/src ] # ./zlibexp
[>] exploiting...
[>] xret = 0xbffffb50
>Sent!..
gzprintf -> 0
gzclose -> 0 [1]
[ kelledin@valhalla /usr/src ] #

The vulnerability consists of a buffer overflow and a
string-format vulnerability (in case something feeds '("Hello%c
there\n", '\0')' to gzprintf). Both should be fixed by the
patch below. How exploitable is this? Well, not very. The
gzprintf() function is seldom used, even on a fully loaded
system, so a would-be 0wner would likely have to code his own
app and trick the 0wnee into running it. I've got reliable
anecdotal evidence that ImageMagick calls gzprintf(), though I
haven't checked for myself.

--
Kelledin
"If a server crashes in a server farm and no one pings it, does
it still cost four figures to fix?"

[ reply ]