For a few years in the 1990s, when PNG was just getting established as a Web image format, I was a developer on the libpng team.
One reason I got involved is that the compression patent on GIFs was a big deal at the time. I had been the maintainer of GIFLIB since 1989; it was on my watch that Marc Andreesen chose that code for use in the first graphics-capable browser in ’94. But I handed that library off to a hacker in Japan who I thought would be less exposed to the vagaries of U.S. IP law. (Years later, after the century had turned and the LZW patents expired, it came back to me.)
Then, sometime within a few years of 1996, I happened to read the PNG standard, and thought the design of the format was very elegant. So I started submitting patches to libpng and ended up writing the support for six of the minor chunk types, as well as implementing the high-level interface to the library that’s now in general use.
As part of my work on PNG, I volunteered to clean up some code that Greg Roelofs had been maintaining and package it for release. This was “gif2png” and it was more or less the project’s official GIF converter.
(Not to be confused, though, with the GIFLIB tools that convert to and from various other graphics formats, which I also maintain. Those had a different origin, and were like libgif itself rather better code.)
gif2pngs’s role then was more important than it later became. ImageMagick already existed, but not in anything like its current form; GIMP had barely launched, and the idea of a universal image converter hadn’t really taken hold yet. The utilities I ship with GIFLIB also had an importance then that they would later lose as ImageMagick’s “convert” became the tool everyone learned to reach for by reflex.
It has to be said that gif2png wasn’t very good code by today’s standards. It had started life in 1995 as a dorm-room project written in journeyman C, with a degree of carelessness about type discipline and bounds checking that was still normal in C code of the time. Neither Greg nor I gave it the thorough rewrite it perhaps should have gotten because, after all, it worked on every well-formed GIF we ever threw at it. And we had larger problems to tackle.
Still, having taken responsibility for it in ’99. I kept it maintained even as it was steadily decreasing in importance. ImageMagick convert(1) had taken over; I got zero bug reports or RFEs for six years between 2003 and 2009.
I did some minor updating in 2010, but more out of completism than anything else; I was convinced that the user constituency for the tool was gone. And that was fine with me – convert(1) had more eyes on it and was almost certainly better code. So gif2png fell to near the bottom of my priority list and stayed there.
A few years after that, fuzzer attacks on programs started to become a serious thing. I got one against GIFLIB, which was issued a CVE and I took very seriously – rogue code execution in a ubiquitous service library is baaaad. A couple of others in GIFLIB’s associated utility programs, which I took much less seriously as I wasn’t convinced anyone still used them at all. You’re going to exploit these…how?
And, recently, two segfaults in gif2png. Which was absolutely at the bottom of my list of security concerns. Standalone program, designed to be used in input files you trust to be reasonably close to well-formed GIFs (there was a ‘recover’ option that could salvage certain malformed ones if you were very lucky). Next to no userbase since around 2003. Again, you’re going to exploit this…how?
Now, I’m no infosec specialist, but there is real-world evidence that I know how to get my priorities right. I’ve led the the NTPsec project for nearly five years now, reworking its code so thoroughly that its size has shrunk by a factor of 4. NTP implementations are a prime attack target because the pre-NTPsec reference version used to be so easy to subvert. And you know what the count of CVEs against our code (as opposed to what we inherited) is?
Zero. Zip. Zilch. Nobody has busted my code or my team’s. Despite half the world’s academics and security auditors running attacks on it. Furthermore, we have a record of generally having plugged about four out of five CVEs in the legacy code by the time they’re issued.
That’s how the security of my code looks when I think it’s worth the effort. For libgif I’ll spend that effort willingly. For the GIFLIB tools around it, less willingly. But for gif2png, that seemed pointless. I was tired of spending effort to deal with the 47,000th CS student thinking “I know! I’ll run a fuzzer on <random -program-x>!” and thinking a crash was a big deal when the program was a superannuated standalone GIF filter that hasn’t seen any serious use since J. Random Student was in diapers.
So two days ago I marked two crashes on malformed input in gif2png won’t-fix, put in in a segfault handler so it would die gracefully no matter what shit you shoved at it, and shipped it…
…only to hear a few ours later, from my friend Perry Metzger, that there was a shitstorm going down on Twitter about how shockingly incompetent this was.
Really? They really thought this program was an attack target, and that you could accomplish anything by running rogue code from inside it?
Narrator voice: No, they didn’t. There are some people for whom any excuse to howl and fling feces will do.
A similar bug in libgif or NTPsec would have been a serious matter. But I’m pretty good at not allowing serious bugs to happen in those. In a quarter century of writing critical service code my CVE count is, I think, two (one long ago in fetchmail) with zero exploits in the wild.
This? This ain’t nothin’. Perry did propose a wildly unlikely scenario in which the gif2png binary somehow got wedged in the middle of somebody’s web framework on a server and allowed to see ill-formed input, allowing a remote exploit, but I don’t believe it.
Alas, if I’ve learned anything about living on the modern Internet it’s that arguing that sort of point with the howler monkeys on Twitter is a waste of time. (Actually, arguing anything with the howler monkeys on Twitter is a waste of time.) Besides, the code may not be an actual security hazard, but it has been kind of embarrassing to drag around ever since I picked it up.
So, rather than patch the C and deal with yet another round of meaningless fuzzer bugs in the future, I’ve rewritten it in Go. Here it is, and now that it’s in a type-safe language with access bounds checking I don’t ever have to worry about that class of problem again.
One good thing may come of this episode (other than lifting code out of C, which is always a plus). I notice that the GIF and PNG libraries in Go are, while serviceable for basic tasks, rather limited. You can convert with them, but you can’t do lossless editing with them. Neither one deserializes the entire ontology of its file format.
As the maintainer of GIFLIB and a past libpng core developer, I don’t know where I’d find a better-qualified person to fix this than me. So now on my to-do list, though not at high priority: push some patches upstream to improve these libraries.