linux-kernel - Re: [PULL] seccomp update (next)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Sat, 18 Jun 2016 00:02:09 -0700
From:	Kees Cook <keescook@...omium.org>
To:	Andy Lutomirski <luto@...capital.net>
Cc:	James Morris <jmorris@...ei.org>,
	LSM List <linux-security-module@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Andy Lutomirski <luto@...nel.org>
Subject: Re: [PULL] seccomp update (next)

On Fri, Jun 17, 2016 at 11:55 AM, Andy Lutomirski <luto@...capital.net> wrote:
> On Fri, Jun 17, 2016 at 12:15 AM, James Morris <jmorris@...ei.org> wrote:
>> On Tue, 14 Jun 2016, Kees Cook wrote:
>>
>>> Hi,
>>>
>>> Please pull these seccomp changes for next. These have been tested by
>>> myself and Andy, and close a long-standing issue with seccomp where tracers
>>> could change the syscall out from under seccomp.
>>
>> Pulled to security -next.
>
> As a heads up: I think this doesn't quite close the hole on x86.  Consider:
>
> 64-bit task arranges to be traced by a 32-bit task (or presumably a
> 64-bit task that calls ptrace via int80).
>
> Tracer does PTRACE_SYSCALL.
>
> Tracee does a normal syscall.
>
> Tracer writes tracee's orig_ax, thus invoking this thing in
> arch/x86/kernel/ptrace.c:
>
>         if (syscall_get_nr(child, regs) >= 0)
>             child->thread.status |= TS_COMPAT;
>
> Tracer resumes and gets confused.
>
> I think the right fix is to just delete:
>
>         if (syscall_get_nr(child, regs) >= 0)
>             child->thread.status |= TS_COMPAT;
>
> from ptrace.c.   The comment above it is garbage, too.

I'm perfectly happy to see it removed. I can't make sense of the comment. :)

That said, the only confusion I see is pretty minor. The arch is saved
before the tracer could force TS_COMPAT, so nothing confused is handed
to seccomp (the first time). And the syscall will continue to be
looked up on sys_call_table not ia32_sys_call_table.

The only thing I see is if the tracer has also added a
SECCOMP_RET_TRACE filter, after which the recheck will reload all the
seccomp info, including the arch. And at this point, a sensible filter
will reject a non-matching architecture.

Maybe I'm missing something more?

-Kees

-- 
Kees Cook
Chrome OS & Brillo Security