Re: [gentoo-hardened] ssp random bytes solution

> > 1. don't initialize __guard[] to all 0s, on some compilers it will force
> >    it into the .data section (instead of .bss), it's a slight waste...
>
> I have read the stack value can be used here with zero preformance loss. Just
> to set initial values.

'stack value' = ?

anyway, what i meant was that if you have a variable that you'd initialize
as 0, don't initialize it at all, it will then go to the .bss section
(= takes up no space in the file itself).

> > 2. you have a nice buffer overflow on 64 bit archs where sizeof long = 8
> >    but in your loop you access twice as many elements (you divide by 4
> >    instead of sizeof long).
>
> The /4 is to prevent a bug. Only multiples of 4 and 16 can be used, its a
> double check.

ok, let's forget about the overflow, it would occur only if sizeof(int)>4
and i think no 64 bit arch defines it that way (but someone should double
check).

anyway, taking into account my 3rd observation as well, you should do
something like this:

unsigned long __guard;

void __guard_setup (void)
{
       int mib[3];
       size_t len;

       if (__guard != 0UL)
               return;

       /* Random is another depth in Linux, hence an array of 3. */
       mib[0] = CTL_KERN;
       mib[1] = KERN_RANDOM;
       mib[2] = RANDOM_ERANDOM;

       len = sizeof(unsigned long);
       if (-1 == __sysctl(mib, 3, &__guard, &len, NULL, 0))
               /* If sysctl was unsuccessful, use the "terminator canary". */
               __guard = 0xFF0A0D00UL;
}



--
gentoo-hardened@g.o mailing list