Return address checking for buffer overflow exploit

Return address checking for buffer overflow exploit

[markw]

I had an idea, I'm not sure how practical it would be to implement, but....

On functions which allocate arrays on the stack, have the option of 
inserting return address checking. This should be effective because it 
works "pre-exploit."

The idea is that one of the ways an exploit gains control is to send 
data which overwrites the return address and thus transfers control to 
foreign code. On functions with return address checking, this would 
cause the program to assert and fail.

If some compiler flag is set, i.e. --with-return-address-checking
Functions which declare an automatic array will add one more invisible 
automatic variable below all the arrays which contain either a checksum 
or the actual return address. Upon return, this variable is checked with 
what the intended return address is, if they are not the same, then an 
assert error is made and the program exits.

There could be implemented with varying degrees, i.e. every function 
could check the return address or only functions with automatic arrays. 
Pragmas can also be used to force certain functions either to have or 
not have the checking.

#pragma check_return_address
#pragma nocheck_return_address


What do you guys think?