Re: [gentoo-hardened] SSP + setjmp() = badness?

[Ned Ludd]

On Sun, 2006-08-06 at 13:17 +0200, Peter S. Mazinger wrote:
> On Mon, 31 Jul 2006, Ned Ludd wrote:
>
> > On Mon, 2006-07-31 at 20:33 +0200, pageexec@××××××××.hu wrote:
> > > On 31 Jul 2006 at 11:28, Ned Ludd wrote:
> > > > gcc-4.x and hardened are not in the works..
> > > > We are undecided at this time how much we wish to pursue that avenue.
> > >
> > > which part of hardened gcc is in question? i'd think that at least
> > > PIE/RELRO/BIND_NOW are as easy to support as in 3.x.
> >
> > Yes these would be pretty trivial to do.
> >
> > > ssp may or may
> > > not be a good idea given how new the 4.x series is, but as Mike said,
> > > at least there's an eager upstream to fix any issues.
>
> toolchain.eclass misses the needed support for gcc-4.1 like ssp
>
> > I think the major problem we are facing here is how to cleanly upgrade
> > from 3.x to 4.x. symbol names have changed. And using the stub/aliases
> > method Peter used in uClibc svn allows the __guard to be overwritten.
>
> How can __guard be overwritten, it is even marked with attribute_relro?

We tested and talked about this before peter.. 
None the less here is a simple testcase for you.

extern unsigned long __guard;

void fail() {
        __guard = 0;
}

int main() {
        puts("This should fail");
        fail();
        puts("This should not be reached");
        return 0;
}


./guardtest
This should fail
main: stack smashing attack in function fail()
Aborted

You results should be similar.


-- 
Ned Ludd <solar@g.o>
Gentoo Linux

-- 
gentoo-hardened@g.o mailing list