On Wed, Oct 05, 2016 at 12:18:17PM -0700, John Stultz wrote:
> On Wed, Oct 5, 2016 at 12:10 PM, Dmitry Torokhov
> <[email protected]> wrote:
> > On Mon, Oct 03, 2016 at 09:41:30PM -0700, John Stultz wrote:
> >> +#ifdef CONFIG_CGROUP_NICE_ATTACH
> >> +int cgroup_nice_allow_attach(struct cgroup_taskset *tset)
> >> +{
> >> + const struct cred *cred = current_cred(), *tcred;
> >> + struct task_struct *task;
> >> + struct cgroup_subsys_state *css;
> >> +
> >> + if (capable(CAP_SYS_NICE))
> >> + return 0;
> >> +
> >> + cgroup_taskset_for_each(task, css, tset) {
> >> + tcred = __task_cred(task);
> >
> > __task_cred() requires RCU lock (courtesy Ricky Z).
>
> Again, hopefully this isn't an issue with the new approach, but for
> the short term I'll see if we can get this fixed in the android tree.
>
Actually, it should all be simply removed from there right away, as this
ends up being basically noop (but with all the locking violations and
races):
cgroup_taskset_for_each() needs tasks to be placed on cset->mg_tasks
list, but nobody does this in the ->allow_access() code path, so this
loops always executes exactly 0 times and the whole thing is exactly
equivalent of doing
return capable(CAP_SYS_NICE) ? 0 : -EACESS;
which can be done right in cgroup_procs_write_permission().
Thanks.
--
Dmitry
