StackGuard Mechanism: Emsi's Vulnerability

A significant security vulnerability has been discovered by Mariusz Woloszyn <emsi@it.pl> that permits attackers to perpetrate successful attacks against StackGuarded programs under particular circumstances. Woloszyn is preparing a Phrack article describing this vulnerability, which we summarize here. StackGuard 1.21 effectively protects against this vulnerability. The Immunix team would like to thank Mariusz for kindly notifying us first about this vulnerability, and allowing us the time to develop and distribute a defense.

The Problem

Consider this vulnerable code:

foo(char * arg) {
    char *    p = arg;    // a vulnerable pointer
    char a[25];    // the buffer that makes the pointer vulnerable

    gets(a);    // using gets() makes you vulnerable
    gets(p);    // this is the good part
}

In attacking this code, the attacker first overflows the buffer a[] with a goal of changing the value of the char * p pointer. Specifically, the attacker can cause the p pointer to point anywhere in memory, but especially at a return address record in an activation record. When the program then takes input and stores it where p points, the input data is stored where the attacker said to store it.

The above attack is effective against the Random and Terminator Canary mechanisms because those methods assume that the attack is linear, i.e. that an attacker seeking to corrupt the return address must necessarily use a string operation that overflows an automatic buffer on the stack, moving up memory through the canary word, and only then reach the return address entry. The above attack form, however, allows the attacker to synthesize a pointer to arbitrary space, including pointing directly at the return address, bypassing canary protection.

The Solution: The XOR Random Canary

StackGuard 1.21 introduces a new canary defense mechanism: the XOR Random canary. Like the random canary mechanism, we choose a vector of 128 random canary words at exec() time, but we also XOR the canary with the return address word, so that the return address is bound to the random canary value. The exact procedure is as follows:

Setting up an activation record: when calling a function

push the return address
look up the random canary word for the current function
XOR the random canary word with the return address
store the result immediately below the return address in the activation record

Tearing down an activation record: when returning from a function

fetch the canary word from memory
XOR the memory canary word with the return address on the stack
compare the result with the random canary word associated with the current function

The result of this method is that we have the same protection as with the classic Random canary, and also the property that the attacker cannot modify the return address without invalidating the canary word.

Availability

StackGuard 1.21 has been made available. We have done partial testing with this compiler, using it to build many programs common in Linux distributions, and have not observed any problems. However, we have not yet done a complete build of an entire Linux distribution, so this compiler should be considered beta for now.