On Fri, Nov 07, 2014 at 04:32:49PM +0100, Hanno Böck wrote:
> Also see attachmend, output from american fuzzy lop with latest git
> code and your two patches. 9 crashes, 10 hangs.

Thanks. One of those pointed out that my overflow check for hash section
sizes was bogus. Fixed version attached. The others seem to be because
handle_versym didn't initialize its vernames and filenames. Then when
an ELF file didn't set them we did check they were not set (NULL), but
that check failed, because the elements still contained random data.
The second second patch fixes that.

I have pushed all three fuzz-robustify patches to master.

Note that the testcases you say are hanging are just really, realy slow.
Because of very large input values they try to process a lot of elements,
but eventually they will finish. We still might to sanity check some of
those excessively large input values, but they don't lead to hangs or
crashes. Just very long runtimes.

Cheers,

Mark