Re: [PATCH v2 2/3] sched: Skip priority checks with SCHED_FLAG_KEEP_PARAMS

[Quentin Perret]

On Thursday 10 Jun 2021 at 21:15:45 (+0200), Peter Zijlstra wrote:
> On Thu, Jun 10, 2021 at 03:13:05PM +0000, Quentin Perret wrote:
> > SCHED_FLAG_KEEP_PARAMS can be passed to sched_setattr to specify that
> > the call must not touch scheduling parameters (nice or priority). This
> > is particularly handy for uclamp when used in conjunction with
> > SCHED_FLAG_KEEP_POLICY as that allows to issue a syscall that only
> > impacts uclamp values.
> > 
> > However, sched_setattr always checks whether the priorities and nice
> > values passed in sched_attr are valid first, even if those never get
> > used down the line. This is useless at best since userspace can
> > trivially bypass this check to set the uclamp values by specifying low
> > priorities. However, it is cumbersome to do so as there is no single
> > expression of this that skips both RT and CFS checks at once. As such,
> > userspace needs to query the task policy first with e.g. sched_getattr
> > and then set sched_attr.sched_priority accordingly. This is racy and
> > slower than a single call.
> > 
> > As the priority and nice checks are useless when SCHED_FLAG_KEEP_PARAMS
> > is specified, simply inherit them in this case to match the policy
> > inheritance of SCHED_FLAG_KEEP_POLICY.
> > 
> > Reported-by: Wei Wang <wvw@xxxxxxxxxx>
> > Signed-off-by: Quentin Perret <qperret@xxxxxxxxxx>
> > ---
> >  kernel/sched/core.c | 4 ++++
> >  1 file changed, 4 insertions(+)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 3b213402798e..1d4aedbbcf96 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -6585,6 +6585,10 @@ SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr,
> >  	rcu_read_unlock();
> >  
> >  	if (likely(p)) {
> > +		if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) {
> > +			attr.sched_priority = p->rt_priority;
> > +			attr.sched_nice = task_nice(p);
> > +		}
> >  		retval = sched_setattr(p, &attr);
> >  		put_task_struct(p);
> >  	}
> 
> I don't like this much... afaict the KEEP_PARAMS clause in
> __setscheduler() also covers the DL params, and you 'forgot' to copy
> those.
>
> Can't we short circuit the validation logic?

I think we can but I didn't like the look of it, because we end up
sprinkling checks all over the place. KEEP_PARAMS doesn't imply
KEEP_POLICY IIUC, and the policy and params checks are all mixed up.

But maybe that wants fixing too? I guess it could make sense to switch
policies without touching the params in some cases (e.g switching
between FIFO and RR, or BATCH and NORMAL), but I'm not sure what that
would mean for cross-sched_class transitions.