[PATCH] proc: Track /proc/$pid/attr/ opener mm_struct

[Kees Cook]

On Tue, Jun 15, 2021 at 11:19:04AM -0700, Linus Torvalds wrote:
> On Mon, Jun 14, 2021 at 6:55 PM youling 257 <youling257 at gmail.com> wrote:
> >
> > if try to find problem on userspace, i used linux 5.13rc6 on old
> > android 7 cm14.1, not aosp android 11.
> > http://git.osdn.net/view?p=android-x86/system-core.git;a=blob;f=init/service.cpp;h=a5334f447fc2fc34453d2f6a37523bedccadc690;hb=refs/heads/cm-14.1-x86#l457
> >
> >  457         if (!seclabel_.empty()) {
> >  458             if (setexeccon(seclabel_.c_str()) < 0) {
> >  459                 ERROR("cannot setexeccon('%s'): %s\n",
> >  460                       seclabel_.c_str(), strerror(errno));
> >  461                 _exit(127);
> >  462             }
> >  463         }
> 
> I have no idea where the cm14.1 libraries are. Does anybody know where
> the matching source code for setexeccon() would be?
> 
> For me - obviously not on cm14.1 - all "setexeccon()" does is
> 
>    n = openat(AT_FDCWD, "/proc/thread-self/attr/exec", O_RDWR|O_CLOEXEC)
>    write(n, string, len)
>    close(n)
> 
> and if that fails, it would seem to indicate that proc_mem_open()
> failed. Which would be mm_access() failing. But I don't see how that
> can be the case, because mm_access() explicitly allows "mm ==
> current->mm" (which the above clearly should be).

Yeah, that was what I saw too.

> youling, can you double-check with the current -git tree? But as far
> as I can tell, my minimal patch is exactly the same as Kees' patch
> (just smaller and simpler).

FWIW, for that patch:

Acked-by: Kees Cook <keescook at chromium.org>
Cc: stable at vger.kernel.org

> 
> Kees, do you see anything?

No, I haven't been able to reproduce the failure. :(

-- 
Kees Cook