Re: [PATCH v6 04/12] Add memory slot versioning and use it toprovide fast guest write interface

[Gleb Natapov]

On Wed, Oct 06, 2010 at 11:38:47AM -0300, Marcelo Tosatti wrote:
> On Wed, Oct 06, 2010 at 01:14:17PM +0200, Gleb Natapov wrote:
> > > > +int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
> > > > +			      gpa_t gpa)
> > > > +{
> > > > +	struct kvm_memslots *slots = kvm_memslots(kvm);
> > > > +	int offset = offset_in_page(gpa);
> > > > +	gfn_t gfn = gpa >> PAGE_SHIFT;
> > > > +
> > > > +	ghc->gpa = gpa;
> > > > +	ghc->generation = slots->generation;
> 
> kvm->memslots can change here.
> 
> > > > +	ghc->memslot = gfn_to_memslot(kvm, gfn);
> > > > +	ghc->hva = gfn_to_hva(kvm, gfn);
> 
> And if so, gfn_to_memslot / gfn_to_hva will use new memslots pointer.
> 
> Should dereference all values from one copy of kvm->memslots pointer.
>  
Ah, I see now. Thanks! Will fix.

> > > > +	if (!kvm_is_error_hva(ghc->hva))
> > > > +		ghc->hva += offset;
> > > > +	else
> > > > +		return -EFAULT;
> > > > +
> > > > +	return 0;
> > > > +}
> > > 
> > > Should use a unique kvm_memslots structure for the cache entry, since it
> > > can change in between (use gfn_to_hva_memslot, etc on "slots" pointer).
> > > 
> > I do not understand what do you mean here. kvm_memslots structure itself
> > is not cached only various translation that use it are cached. Translation
> > result are never used if kvm_memslots was changed.
> 
> > > Also should zap any cached entries on overflow, otherwise malicious
> > > userspace could make use of stale slots:
> > > 
> > There is only one cached entry at each given time. User who wants to
> > write into guest memory often defines gfn_to_hva_cache variable
> > somewhere. Init it with kvm_gfn_to_hva_cache_init() and then calls
> > kvm_write_guest_cached() on it. If there was no slot changes in between
> > cached translation are used. Otherwise cache is recalculated.
> 
> Malicious userspace can cause entry to be cached, ioctl
> SET_USER_MEMORY_REGION 2^32 times, generation number will match,
> mark_page_dirty_in_slot will be called with pointer to freed memory.
> 
Hmm. To zap all cached entires on overflow we need to track them. If we
will track then we can zap them on each slot update and drop "generation"
entirely.

--
			Gleb.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/