Wireshark-dev: Re: [Wireshark-dev] Fuzzing Wireshark with oss-fuzz

From: Guy Harris <guy@xxxxxxxxxxxx>
Date: Wed, 21 Dec 2016 03:19:01 -0800
On Dec 21, 2016, at 2:12 AM, Peter Wu <peter@xxxxxxxxxxxxx> wrote:

> From the efforts that I have seen, Moshe seems to be targeting the
> dissectors functions. Since these may appear over the network, it is
> probably one of the more interesting parts to tackle first.

Then I'm not sure how well

> The second step is to create a fuzzing interface. The fuzzing interface needs to have the following signature:
> 
> extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
>   DoSomethingInterestingWithMyAPI(Data, Size);
>   return 0;  // Non-zero return values are reserved for future use.
> }

matches the way it works.

If you have a program called "rednose", that takes a JPEG image, looks for noses, and colors them red, you could have an API that takes a pointer to an array of bytes containing a JPEG image, and a size_t giving the total number of bytes in the image, and call the nose-detector-and-colorer API (although, given that the argument is a const pointer, it'd have to provide some output buffer).

I guess you *could* have an API that takes a *single* blob of data and hands it to the frame dissector, with enough metadata to have it interpreted as an Ethernet frame, and that would catch *some* issues.  It wouldn't catch any issues that would only show up with a sequence of packets.

Is there some way to fuzz code that's *not* stateless, and that takes a *sequence* of bits of input, in order?