4 ms·
Why buffer overflow exploitation took so long to mature
- drawkbox 17y agoDamn interesting I might say. It piqued my interest and is truly strange in that the waves of buffer overflow exploits took time to become mainstream. I have seen it with other stuff like good mobile, killer apps, standards etc. Where ideas are talked about for a long time then finally they get written, then packaged and products, then mass adoption after a certain time. I would argue is that when it became a game, people played, competition stifled... The article states “Smashing the Stack for Fun and Profit“ was released and then many examples spawned, which further spawned more and more ad infinitum. I mean which one piques your interest: - COMPUTER SECURITY TECHNOLOGY PLANNING STUDY OR - “Smashing the Stack for Fun and Profit“ Game it...
- viraptor 17y ago> took time to become mainstream One more important event is not mentioned. Internet became a mainstream thing in middle 90' - that's more or less when the flood of buffer overflows and format string exploits really started. Before that, network apps were mostly created and used by people who "knew stuff" in general...
- barrkel 17y agoYou've pin-pointed it. Without a commonly available transmission vector, buffer overflow exploits were the hard way to get around - and if you weren't coming from across a network, you likely already had full control over the local machine, as it was probably running DOS, Windows 3.11, Win 95, etc. Much easier was boot sector viruses, particularly since PCs by default tried to boot off any floppy left in the drive when turned on; that, and infected executables.
- tptacek 17y agoNo. Attacks against machines running Windows weren't common until the second half of the '90s. Far and away the most popular targets in the earlier '90s were NFS servers. And nobody attacked machines by trying to infect them with boot sector malware.
- barrkel 17y agoIt depends on what you mean by "attack"; and I include DOS and Win 3.11. In the early 90s, the most common viruses I saw were infected executables (i.e. modified programs that when run, searched for other executables to infect) and boot sector viruses, often both in one. I worked part-time as a PC technician in the early 90s, building, upgrading and repairing PCs. I was there :)
- InclinedPlane 17y agoMore so, the internet became so popular that significant portions of the IP v4 address space was in use. This is important because it means that a very simple piece of worm code which merely probes random IP addresses attempting to infect other vulnerable machines will have a non-trivially high success rate. High enough in fact to be able to spread exponentially through the internet until nearly all vulnerable hosts have been infected over a matter of mere days. This combination: massive vulnerabilities in popular operating systems due to a failure to take the risk of exploitation seriously; large numbers of computers operated by unsophisticated users directly on the internet (plugged directly into a cable or dsl modem, for example) with no firewall or NAT bridge protecting any of their services; and an address space full of vulnerable systems, led to a perfect storm for the spread of computer worms.
- tptacek 17y agoThe first genuinely successful worm didn't strike until 8 years after the first modern overflow exploit. So I don't think that's a compelling argument.
- NateLawson 17y agoI disagree. The Internet had grown enough by 1988 to host the Internet worm, which exploited a buffer overflow among other things. So that doesn't explain this 7-year gap. The best explanation is that there were plenty of other Unix holes (race conditions, rhosts, etc.) to exploit that didn't require inventing an entirely new technique. The second part of the article goes into this in much more depth (HN link). http://news.ycombinator.com/item?id=1321562 http://news.ycombinator.com/item?id=1321562
- alanh 17y agoInteresting, but I don’t think the question raised in the title was actually answered — he just retold the history.
- InclinedPlane 17y agoIt's a series of articles on the topic, this is just the first.
- NateLawson 17y agoCorrect, here is the second one which goes more in-depth: http://news.ycombinator.com/item?id=1321562 http://news.ycombinator.com/item?id=1321562
- tptacek 17y agoI think "Smashing The Stack" is a very important milestone, but I don't think it's the tipping point for buffer overflows. I think that'd be Sendmail 8.6.12 and 8lgm's syslog(3) overflow exploit. For the first half of the 1990s, Sendmail was to Internet security what Wordpress is now: the dominant attack vector. Sendmail zero days were the bearer bonds of #hack. Whenever Allman's team fixed a flaw, the current version of Sendmail became one of the hottest targets on the net. 8lgm had been coming off a string of something like 12 high-quality advisories. An unpublished 8lgm advisory in 8.6.12 was a big deal indeed. 8lgm basically dared the Internet to reproduce their finding; they said "it's an overflow in vasprintf in syslog, but we're not showing the exploit". For the next 8 months or so, everyone was madly trying to cobble together stack overflow exploits. I remember sitting in someone's apartment at a party watching Mudge walk us through some example shellcode. I started Matasano with one of the guys who published the first x86 stack overflow. My point is, well before Smashing the Stack, buffer overflows were a very big deal. Let me head off a likely follow-on argument: there's no "Smashing the Stack"-a-like for heap overflow exploits; the techniques to exploit heap overflows are as folkloric as stack overflows were prior to Elias' article. That is to say, it's out there, it's not hard to find (especially if you can read and play with exploit code), but there's no solid cookbook. And yet there's a thriving ecology of heap exploits out there. So it's hard to argue that the '90s would have played out that differently without Elias' article. This does nothing to answer the core question: why so many years between rtm's worm and Lopatic's HTTPD exploit?
- NateLawson 17y agoIn the second part, I cover this very briefly. Basically, there were so many other holes to exploit from 1988-1995 that you didn't need to learn a whole new exploitation technique. Also note that it took a few years for buffer overflow exploitation techniques developed on Linux to hit Windows -- the systems were far enough apart that it took a little while to port. So an entirely new exploitation technique (not just new platform) would require much more work, and thus be a larger barrier to entry. Also, Lopatic was a creative thinker. Sometimes it just takes being shown something is possible to challenge many others to try to replicate it.