linux-kernel - Re: [patch 2/2] x86/kvm: Sanitize kvm_async_pf_task

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]

Message-ID: <CALCETrXpW5TYRNu2hMXt=fGC8EOh7WVqffCzS5GrwApC1inTzw@mail.gmail.com>
Date:   Sat, 7 Mar 2020 08:06:52 -0800
From:   Andy Lutomirski <luto@...nel.org>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     Andy Lutomirski <luto@...nel.org>,
        LKML <linux-kernel@...r.kernel.org>, X86 ML <x86@...nel.org>,
        Paolo Bonzini <pbonzini@...hat.com>, KVM <kvm@...r.kernel.org>,
        "Paul E. McKenney" <paulmck@...nel.org>
Subject: Re: [patch 2/2] x86/kvm: Sanitize kvm_async_pf_task_wait()

On Sat, Mar 7, 2020 at 7:52 AM Thomas Gleixner <tglx@...utronix.de> wrote:
>
> Andy Lutomirski <luto@...nel.org> writes:
> > On Sat, Mar 7, 2020 at 2:01 AM Thomas Gleixner <tglx@...utronix.de> wrote:
> >> > What’s the local_irq_disable() here for? I would believe a
> >> > lockdep_assert_irqs_disabled() somewhere in here would make sense.
> >> > (Yes, I see you copied this from the old code. It’s still nonsense.)
> >>
> >> native_safe_halt() does:
> >>
> >>          STI
> >>          HLT
> >>
> >> So the irq disable is required as the loop should exit with interrupts
> >> disabled.
> >
> > Oops, should have looked at what native_safe_halt() does.
> >
> >>
> >> > I also find it truly bizarre that hlt actually works in this context.
> >> > Does KVM in fact wake a HLTed guest that HLTed with IRQs off when a
> >> > pending async pf is satisfied?  This would make sense if the wake
> >> > event were an interrupt, but it’s not according to Paolo.
> >>
> >> See above. safe halt enables interrupts, which means IF == 1 and the
> >> host sanity check for IF == 1 is satisfied.
> >>
> >> In fact, if e.g. some regular interrupt arrives before the page becomes
> >> available and the guest entered the halt loop because the fault happened
> >> inside a RCU read side critical section with preemption enabled, then
> >> the interrupt might wake up another task, set need resched and this
> >> other task can run.
> >
> > Now I'm confused again.  Your patch is very careful not to schedule if
> > we're in an RCU read-side critical section, but the regular preemption
> > code (preempt_schedule_irq, etc) seems to be willing to schedule
> > inside an RCU read-side critical section.  Why is the latter okay but
> > not the async pf case?
>
> Preemption is fine, but voluntary schedule not. voluntary schedule might
> end up in idle if this is the last runnable task.

I guess something horrible happens if we try to go idle while in an
RCU read-side critical section.  Like perhaps it's considered a grace
period.  Hmm.

>
> > Ignoring that, this still seems racy:
> >
> > STI
> > nested #PF telling us to wake up
> > #PF returns
> > HLT
>
> You will say Ooops, should have looked .... when I tell you that the
> above cannot happen. From the SDM:
>
>   If IF = 0, maskable hardware interrupts remain inhibited on the
>   instruction boundary following an execution of STI.
>
> Otherwise safe_halt would not work at all :)

Ooops, should have looked. :)

> > Sadly, wrmsr to turn off async pf will inject wakeups even if IF == 0.
>
> WHAT? That's fundamentally broken. Can you point me to the code in
> question?

I think Paolo said so in a different thread, but I can't Let me see if
I can find it:

kvm_pv_enable_async_pf()
  kvm_clear_async_pf_completion_queue()

but that doesn't actually seem to send #PF.  So maybe I'm wrong.

I will admit that, even after reading the host code a few times, I'm
also not convinced that wakeups don't get swallowed on occasion if
they would have been delivered at times when it's illegal.