Re: [PATCH 1/1] psi: remove 500ms min window size limitation for triggers

[Suren Baghdasaryan]

On Thu, Mar 2, 2023 at 7:30 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Wed, Mar 01, 2023 at 12:48:38PM -0800, Suren Baghdasaryan wrote:
> > On Wed, Mar 1, 2023 at 12:07 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> > >
> > > On Wed, Mar 01, 2023 at 11:34:03AM -0800, Suren Baghdasaryan wrote:
> > > > Current 500ms min window size for psi triggers limits polling interval
> > > > to 50ms to prevent polling threads from using too much cpu bandwidth by
> > > > polling too frequently. However the number of cgroups with triggers is
> > > > unlimited, so this protection can be defeated by creating multiple
> > > > cgroups with psi triggers (triggers in each cgroup are served by a single
> > > > "psimon" kernel thread).
> > > > Instead of limiting min polling period, which also limits the latency of
> > > > psi events, it's better to limit psi trigger creation to authorized users
> > > > only, like we do for system-wide psi triggers (/proc/pressure/* files can
> > > > be written only by processes with CAP_SYS_RESOURCE capability). This also
> > > > makes access rules for cgroup psi files consistent with system-wide ones.
> > > > Add a CAP_SYS_RESOURCE capability check for cgroup psi file writers and
> > > > remove the psi window min size limitation.
> > > >
> > > > Suggested-by: Sudarshan Rajagopalan <quic_sudaraja@xxxxxxxxxxx>
> > > > Link: https://lore.kernel.org/all/cover.1676067791.git.quic_sudaraja@xxxxxxxxxxx/
> > > > Signed-off-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> > > > ---
> > > >  kernel/cgroup/cgroup.c | 10 ++++++++++
> > > >  kernel/sched/psi.c     |  4 +---
> > > >  2 files changed, 11 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
> > > > index 935e8121b21e..b600a6baaeca 100644
> > > > --- a/kernel/cgroup/cgroup.c
> > > > +++ b/kernel/cgroup/cgroup.c
> > > > @@ -3867,6 +3867,12 @@ static __poll_t cgroup_pressure_poll(struct kernfs_open_file *of,
> > > >       return psi_trigger_poll(&ctx->psi.trigger, of->file, pt);
> > > >  }
> > > >
> > > > +static int cgroup_pressure_open(struct kernfs_open_file *of)
> > > > +{
> > > > +     return (of->file->f_mode & FMODE_WRITE && !capable(CAP_SYS_RESOURCE)) ?
> > > > +             -EPERM : 0;
> > > > +}
> > >
> > > I agree with the change, but it's a bit unfortunate that this check is
> > > duplicated between system and cgroup.
> > >
> > > What do you think about psi_trigger_create() taking the file and
> > > checking FMODE_WRITE and CAP_SYS_RESOURCE against file->f_cred?
> >
> > That's definitely doable and we don't even need to pass file to
> > psi_trigger_create() since it's called only when we write to the file.
> > However by moving the capability check into psi_trigger_create() we
> > also postpone the check until write() instead of failing early in
> > open(). I always assumed failing early is preferable but if
> > consolidating the code here makes more sense then I can make the
> > switch. Please let me know if you still prefer to move the check.
>
> Just for context, a person on our team is working on allowing
> unprivileged polls with windows that are multiples of 2s, which can be
> triggered from the regular aggregator threads. This should be useful
> for container delegation, and also for the desktop monitor app usecase
> that Chris Down brought up some time ago. At that point, everybody can
> open the file for write, and permissions are checked against the
> trigger parameters.
>
> So I don't think it's a big deal to check this particular permission
> at write time. But if you prefer we can also merge your patch as-is
> and do the refactor as part of the other series.

Let's roll this check without additional changes and then consolidate
the checking inside psi_trigger_create() in a separate patch. If
anybody objects to the late permission check we will just revert that
last change without affecting anything else.

>
> Your call. In either case, please feel free to add
>
> Acked-by: Johannes Weiner <hannes@xxxxxxxxxxx>

Thanks! Will post the final patch with Ack's later today. Originally
it was purely cgroup-related change but now it's more of a PSI change.
Therefore Peter's tree will probably be the right place for it.