Re: [bug #23029] autogsdoc buffer overflow in gnustep-base 1.14.2 and 1.14.3 on Rhel 5

[Gareth Armstrong]

Hello again Richard,

I confirm that the fix you implemented works on a Fedora 8 x86_64 
platform.  I will test it out on Rhel5 i386 and x86_64 as well.  Will 
this fix make it into the 1.14.x series of gnustep-base? It would be 
very nice to have.

Many thanks again,

Gareth

[gareth@localhost] ~/WORK/OCEK/gnustep
$ autogsdoc
2008-06-10 14:51:38.724 autogsdoc[19561] No .h, .m, .c, .gsdoc, or .html 
filename arguments found ... giving up


Richard Frith-Macdonald wrote:
> Update of bug #23029 (project gnustep):
>
>                   Status:               Need Info => Fixed
>              Open/Closed:                    Open => In Test
>
>     _______________________________________________________
>
> Follow-up Comment #3:
>
> Thanks.
> This section:
>
> #5 0x00000039388e90bb in __realpath_chk (buf=0x66c8 <Address 0x66c8 out
> of bounds>,
> resolved=0x66c8 <Address 0x66c8 out of bounds>, resolvedlen=6) at
> realpath_chk.c:30
> #6 0x00002aaaaacd96af in -[NSString stringByResolvingSymlinksInPath]
> (self=0x6854a0,
>
> Tells me that the problem is detected in the libc realpath() function.
>
> Now, as far as I can see, the only way this can have a problem is if the
> output buffer supplied to the function is not large enough to hold the
> expanded path.
> The code was defaulting to using 1024 if MAX_PATH was not defined, so I've
> changed it to refrain from using realpath() in that situation.
>
> Please could you update using the code from subversion, and see if this fixes
> the problem and let me know (you can email direct to richard at
> tiptree.demon.co.uk if you are still having problems with email filtering).
>   
--