Is DEP easily evadable?

[John Richard Moser <nigelenki () comcast net>]

I'm no security expert, so bear with me here; I just kind of tripped
over something interesting that I'd like to ask about.

I was blogging about DEP based on MS' technical documentation and came
up with a quick and dirty way to use a buffer overflow (we'll assume no
stackguarding, or that you found a way around it i.e. using a format
string bug) to kick DEP out of the way.  This is pretty much based on
the PaX documentation and justification for mprotect() restrictions.

In a normal shellcode attack, you have a simple process involving:
- Overwrite the return address
- Spew in some executable code
- Let RET take PC to that code and hand you control of the program

With executable protections preventing the stack from being executable,
such as with DEP, a different process (hopefully) occurs:
- Overwrite the return address
- Spew in some executable code
- Segmentation fault (memory not executable)
- Program exits, attack fails

I looked at the fact that Virtual*() functions (similar to mprotect() on
POSIX systems) can be used to make any area of memory executable.  So,
here's my way:
- Overwrite the return address
- Overwrite the stack frame pointer
- Spew in a stack frame with a return pointer to the code you injected
- Spew in some executable code
- Return to a Virtual*() function (don't know nor care which, I'm not
doing the attack)
- Mark the code you injected executable
- Let RET take PC to the code you injected.  You win.

Without mprotect() restrictions OR ASLR, this should be trivial to do,
right?

I can't comment on SafeSEH, although I don't see screwing with exception
handlers as being more interesting to a cracker than injecting your own
code.

-- 
All content of all messages exchanged herein are left in the
Public Domain, unless otherwise explicitly stated.