Smashing the stack for fun and profit

[tptacek]

It's interesting to compare exploit tech from today to how simple things were back in '97. In '97, a complex exploit meant you couldn't have lowercase ASCII equivalents in your shellcode. Now, you scan the target for info leaks that will disclose stack cookies, and spray the heap to increase the odds that you'll hit your exploit in randomized memory, and forge stack frames that will return to mprotect or VirtualProtect, and and and.

So much of "security research" over the last 10 years has basically danced around a fundamental fact: if you lose control of the runtime memory of your program, you lose control of the whole program.

[swolchok]

If you get lucky, your target won't bother with all these new-fangled protections and run on WinXP (no ASLR) and the only thing you'll have to contend with is non-executable stack. Our original exploit for Green Dam used a heap spray only to make writing it easier, and the underground exploit uses a really awful behavior of IE w.r.t. running .NET DLLs -- http://www.milw0rm.com/exploits/8938 . Other than that, there are no defenses to speak of.

[Zarathu]

Well, that's sort of true. You don't necessarily have to keep track of the runtime memory as long as the buffer is large enough.

You could find the address of a JMP %ESP instruction in the victim OS (0x7C8369D8 in XP SP 2) and place it before a NOPsled.

A typical buffer overflow, for example:

[AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA][0x7C8369D8][NOP][shellcode]

You do, of course, need to make sure you don't have any \0 bytes in your shellcode, but that's always been the case.

[mahmud]

A classic. The most referenced underground paper, ever!

112,605 documents found

http://citeseerx.ist.psu.edu/search?q=Smashing+The+Stack+For...

[lg]

Ah, fond memories of my computer security class, where this was assigned reading. We also studied format string vulnerabilities: http://julianor.tripod.com/bc/formatstring-1.2.pdf

[alecco]

If we are going down memory lane, check out JP's "Advanced Doug lea's malloc exploits" http://www.phrack.org/issues.html?issue=61&id=6#article

[scythe]

In one of my freshman CS courses, one of our last assignments was to read this paper (as well as a more approachable tutorial written by our TAs) and carry this out on a toy architecture (LC3) that we had been working with for the whole course. It was probably the most interesting assignment we had that semester.

[Oompa]

I had a similar class. CS2110 at Georgia Tech. I will never forget it :)

[jacquesm]

Funny how the first example contains a subtle bug, the string termination is never done, if the stack wasn't zeroed prior to running that program, you may not only overflow 'buffer' on purpose, you may overflow it a lot further than you'd think by just reading the code.

large_string[255] = 0;

Would solve that.

[wclax04]

I remember reading this paper as an undergrad at Rutgers. We had to do a stack smashing project based on this with different stack protection methods (canaries/ASLR)

This was probably one of the most important things I learned in school.

[haupt]

I remember reading this and finding an error in one of the examples. In addition to learning about the stack, I also realized that superheroes are human, too. <3

[MtL]

Funny, I just re-read that paper yesterday =) Aleph One is a hero in my eyes, he made my childhood much more interesting.

[sown]

For a lot of these tutorials to work, you also have to turn off SELinux or any other stack protection.

[petsos]

This paper was written 7 years before SELinux was merged.

[leif]

I got scut's format string attacks working on a bunch of sort of modern unices (and all the old ones where I had access), but nothing within a year.

[omouse]

And the lesson is, try to avoid using C :)

[tptacek]

And C++.

[ecq]

wow I still remember this. a classic by Aleph One.

Another tutorial pre-dates this by ~1 year. It was written by mudge.

http://biblio.l0t3k.net/b0f/en/howto_write_buffer.txt

[tptacek]

The first published x86 shellcode-style overflow was splitvt, which was announced "officially" just weeks after this tutorial was dated:

http://seclists.org/bugtraq/1995/Dec/2

(My business partner co-authored it).

The first shellcode-style buffer overflow (besides the rtm worm) was Thomas Lopatic's HPUX HTTPd vulnerability from a few months earlier:

http://seclists.org/bugtraq/1995/Feb/109

There was a frantic race to get the first overflow out after 8lgm capped off their run of zero-days with an announcement of a Sendmail 8.6.12 remote that relied on a syslog() overflow (yes, in 1995, syslog(3) had an overflow). I was sitting next to Pieter when he wrote part of the tutorial you linked to. I was pretty young (maybe 19?) but even so, it was a pretty electric time to be involved in security.

[sp332]

(OT) Did you know there's a petition to get mudge to be US Cyberczar? http://www.ipetitions.com/petition/mudge4cyberczar/

[c00p3r]

Oh, that romantic feeling from the boyhood... =)