[mapserver-dev] Fuzzing MapServer

Even Rouault even.rouault at spatialys.com
Fri Jun 11 09:25:08 PDT 2021


I've a prototype of a fuzzing target that can run under ossfuzz now 
working - locally - with the tooling they provide to run locally. I 
discovered that they now document how to make it work with shared 
which made things a bit easier.

The current fuzzer basically runs msCGIDispatchRequest() on a mapfile 
used by WFS tests, with a request that is built from the fuzzed input 
buffer. So it simulates the effect of feeding a (not-so-)random 
QUERY_STRING, which is of course the more obvious attack vector. I've 
confirmed the fuzzer works since in one minute it found a memleak :-) 
Additional fuzzers could potentially be written, like fuzzing the 
mapfile itself, or using a larger set of mapfiles than just the one I 
picked up.

So the question now is: do we want to submit this to ossfuzz for 
inclusion in their nightly builds & CI ? (issues go public 90 days after 
being spotted, or 30 days after beeing fixed)

Note: I do *not* volunteer to fix all bugs it will report.

If we go to apply for inclusion in ossfuzz, I'll need a list of email 
address (that are associated with a gmail account) from developers/PSC 
members interested in accessing the (private) reports.


Le 15/04/2021 à 21:20, Even Rouault a écrit :
> Le 15/04/2021 à 19:28, Steve Lime a écrit :
>> I hear what you're saying from a release standpoint. I guess I could 
>> have said "initiate a fuzzing effort" as part of the 8.0 release. I 
>> like your idea to concentrate on the query string, that represents a 
>> pretty big surface depending what the fixed mapfile contains. With 
>> oss-fuzz there's a time limit on certain types of bugs before 
>> public disclosure, correct?
> Details at 
> https://google.github.io/oss-fuzz/getting-started/bug-disclosure-guidelines/ 
> . They don't make differences between type of bugs.
>> That's a bit worrisome if you got slammed and nobody was available to 
>> address bugs.
> We might also want to decide what to do if bugs impacting security are 
> uncovered (might be hard to decide what is exploitable. a double-free 
> can in some circumstances be exploited, but I doubt any of us as the 
> expertise to evaluate that)
>> Are there alternatives to oss-fuzz that could be considered (Seth 
>> referenced one of them)?
>> Funding would be great although our only source of $'s at the moment 
>> is the OSGeo project budget which is really small and partially 
>> committed to the TravisCI subscription.
> For the mapserver (non-doc) repo, we're close to be ready to unplug 
> Travis-CI now we have a github action job. The remaining thing would 
> be to add coveralls support 
> (https://github.com/MapServer/MapServer/issues/6299 created as remainder)
My software is free, but my time generally not.

More information about the mapserver-dev mailing list