Re: [PATCH 1/2] KVM: arm64: Disable interrupts while walking userspace PTs

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, 13 Mar 2023 15:53:37 +0000,
Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
> 
> +David
> 
> On Mon, Mar 13, 2023, Marc Zyngier wrote:
> > We walk the userspace PTs to discover what mapping size was
> > used there. However, this can race against the userspace tables
> > being freed, and we end-up in the weeds.
> > 
> > Thankfully, the mm code is being generous and will IPI us when
> > doing so. So let's implement our part of the bargain and disable
> > interrupts around the walk. This ensures that nothing terrible
> > happens during that time.
> > 
> > We still need to handle the removal of the page tables before
> > the walk. For that, allow get_user_mapping_size() to return an
> > error, and make sure this error can be propagated all the way
> > to the the exit handler.
> > 
> > Signed-off-by: Marc Zyngier <maz@xxxxxxxxxx>
> > Cc: stable@xxxxxxxxxxxxxxx
> > ---
> >  arch/arm64/kvm/mmu.c | 35 ++++++++++++++++++++++++++++-------
> >  1 file changed, 28 insertions(+), 7 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> > index 7113587222ff..d7b8b25942df 100644
> > --- a/arch/arm64/kvm/mmu.c
> > +++ b/arch/arm64/kvm/mmu.c
> > @@ -666,14 +666,23 @@ static int get_user_mapping_size(struct kvm *kvm, u64 addr)
> >  				   CONFIG_PGTABLE_LEVELS),
> >  		.mm_ops		= &kvm_user_mm_ops,
> >  	};
> > +	unsigned long flags;
> >  	kvm_pte_t pte = 0;	/* Keep GCC quiet... */
> >  	u32 level = ~0;
> >  	int ret;
> >  
> > +	/*
> > +	 * Disable IRQs so that we hazard against a concurrent
> > +	 * teardown of the userspace page tables (which relies on
> > +	 * IPI-ing threads).
> > +	 */
> > +	local_irq_save(flags);
> >  	ret = kvm_pgtable_get_leaf(&pgt, addr, &pte, &level);
> > -	VM_BUG_ON(ret);
> > -	VM_BUG_ON(level >= KVM_PGTABLE_MAX_LEVELS);
> > -	VM_BUG_ON(!(pte & PTE_VALID));
> > +	local_irq_restore(flags);
> > +
> > +	/* Oops, the userspace PTs are gone... */
> > +	if (ret || level >= KVM_PGTABLE_MAX_LEVELS || !(pte & PTE_VALID))
> > +		return -EFAULT;
> 
> I don't think this should return -EFAULT all the way out to userspace.  Unless
> arm64 differs from x86 in terms of how the userspace page tables are managed, not
> having a valid translation _right now_ doesn't mean that one can't be created in
> the future, e.g. by way of a subsequent hva_to_pfn().

I probably took an overly restrictive approach of only looking at the
issue at hand, where exit_mmap() had already torn down the userspace
PTs. But I guess there are other ways for this scenario to happen,
none of which deserve -EFAULT indeed.

> FWIW, the approach x86 takes is to install a 4KiB (smallest granuale) translation,
> which is safe since there _was_ a valid translation when mmu_lock was acquired and
> mmu_invalidate_retry() was checked.  It's the primary MMU's responsibility to ensure
> all secondary MMUs are purged before freeing memory, i.e. worst case should be that
> KVMs stage-2 translation will be immediately zapped via mmu_notifier.

I'd rather avoid extra work. At this stage, we might as well return
-EAGAIN and replay the fault. We already do that in a number of racy
cases, so it fits in the infrastructure.

> KVM ARM also has a bug that might be related: the mmu_seq snapshot needs to be
> taken _before_ mmap_read_unlock(), otherwise vma_shift may be stale by the time
> it's consumed.  I believe David is going to submit a patch (I found and "reported"
> the bug when doing an internal review of "common MMU" stuff).

Huh, that's interesting. David, please post this at your earliest
convenience. I'd rather squash these all in one go.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux