On Thu, Jan 12, 2012 at 13:44, Andrew Lutomirski <luto@xxxxxxx> wrote: > On Thu, Jan 12, 2012 at 10:32 AM, Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: >> On Thu, Jan 12, 2012 at 10:18 AM, Andrew Lutomirski <luto@xxxxxxx> wrote: >>> Like this? >>> >>> http://lkml.indiana.edu/hypermail/linux/kernel/1003.3/01225.html >> >> I don't know the execve_nosecurity patches, so the diff makes little >> sense to me, but yeah, I wouldn't expect it to be more than a couple >> of lines. Exactly *how* you set the bit etc is not something I care >> deeply about, prctl seems about as good as anything. >> >>> Note that there's a huge can of worms if execve is allowed but >>> suid/sgid is not: selinux may elevate privileges on exec of pretty >>> much anything. (I think that this is a really awful idea, but it's in >>> the kernel, so we're stuck with it.) >> >> You can do any amount of crazy things with selinux, but the other side >> of the coin is that it would also be trivial to teach selinux about >> this same "restricted environment" bit, and just say that a process >> with that bit set doesn't get to match whatever selinux privilege >> escalation rules.. >> >> I really don't think this is just about "execve cannot do setuid". I >> think it's about the process being marked as restricted. >> >> So in your patch, I think that "PR_RESTRICT_EXEC" bit is wrong. It >> should simply be "PR_RESTRICT_ME", and be done with it, and not try to >> artificially limit it to be some "execve feature", and more think of >> it as a "this is a process that has *no* extra privileges at all, and >> can never get them". > > execve_nosecurity was my attempt to sidestep selinux issues. It's a > different syscall that does all of the non-security-related things > that execve does but does not escalate (or even change) any > privileges. Maybe I'll try to rework that for newer kernels as well. > The idea is that programs that expect to run in sandboxes / chroots / > namespaces / whatever can use it, and older programs that might > malfunction dangerously if the semantics of execve change will just > fail instead. I don't see any issues with SELinux support for this feature. Specifically, when you try to execute something in SELinux, it will first look at the types and try to "execute" (involving a type transition IE: security label change). But if that fails in many cases it may still be allowed to "execute_no_trans" (IE: regular non-privileged exec() without a transition). If you add this feature, it should just disable the normal "execute" with transition path and unconditionally fall back to "execute_no_trans". Likewise, enabling these bits should also disable the "transition" and "dyntransition" process access vectors, and I'm on the fence about whether "setfscreate", etc should be allowed. Cheers, Kyle Moffett -- Curious about my work on the Debian powerpcspe port? I'm keeping a blog here: http://pureperl.blogspot.com/ -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
