[PATCH v11 11/11] seccomp: implement SECCOMP_FILTER_FLAG_TSYNC

[Kees Cook]

On Thu, Jul 17, 2014 at 8:45 AM, Kees Cook <keescook at chromium.org> wrote:
> On Thu, Jul 17, 2014 at 8:04 AM, David Drysdale <drysdale at google.com> wrote:
>> On Wed, Jul 16, 2014 at 10:50 PM, Kees Cook <keescook at chromium.org> wrote:
>>> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
>>> index 9065d2c79c56..2125b83ccfd4 100644
>>> +/**
>>> + * seccomp_can_sync_threads: checks if all threads can be synchronized
>>> + *
>>> + * Expects sighand and cred_guard_mutex locks to be held.
>>> + *
>>> + * Returns 0 on success, -ve on error, or the pid of a thread which was
>>> + * either not in the correct seccomp mode or it did not have an ancestral
>>> + * seccomp filter.
>>> + */
>>> +static inline pid_t seccomp_can_sync_threads(void)
>>> +{
>>> +       struct task_struct *thread, *caller;
>>> +
>>> +       BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
>>> +       BUG_ON(!spin_is_locked(&current->sighand->siglock));
>>> +
>>> +       if (current->seccomp.mode != SECCOMP_MODE_FILTER)
>>> +               return -EACCES;
>>
>> Quick question -- is it possible to apply the first filter and also synchronize
>> it across threads in the same operation?  If so, does this arm also need to
>> cope with seccomp.mode being SECCOMP_MODE_DISABLED?
>>
>> [seccomp_set_mode_filter() looks to call this via seccomp_attach_filter()
>> before it does seccomp_assign_mode()]
>
> I don't entirely understand what you're asking. The threads gain the
> filter and the mode before the current thread may gain the mode (if
> it's the first time this has been called). Due to all the locks,
> though, this isn't a problem. Is there a situation you see where there
> might be a problem?

Just to follow up for posterity on lkml: the problem was that mode was
being set in "current" _after_ sync, so the mode check in can_sync
would fail if "current" was not yet in filter mode. (i.e. the first
attached filter could not have the TSYNC flag.) This check was
redundant with the attach_filter entry point checks, and protected
nothing, so it has been removed and a new test added to the seccomp
regression test suite. :)

I sent it as a new patch on top of v11, instead of respinning
everything as v12. If that's not preferred, I can send v12 with this
fix incorporated.

Thanks!

-Kees

-- 
Kees Cook
Chrome OS Security