|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH 7/8] x86/mm: handle foreign mappings in p2m_entry_modify
On Thu, Feb 07, 2019 at 05:49:16PM +0000, George Dunlap wrote:
> On 1/30/19 10:36 AM, Roger Pau Monne wrote:
> > diff --git a/xen/include/asm-x86/p2m.h b/xen/include/asm-x86/p2m.h
> > index 834d49d2d4..1cc8acb3fe 100644
> > --- a/xen/include/asm-x86/p2m.h
> > +++ b/xen/include/asm-x86/p2m.h
> > @@ -933,9 +933,12 @@ struct hvm_ioreq_server *p2m_get_ioreq_server(struct
> > domain *d,
> > unsigned int *flags);
> >
> > static inline void p2m_entry_modify(struct p2m_domain *p2m, p2m_type_t nt,
> > - p2m_type_t ot, unsigned int level)
> > + p2m_type_t ot, mfn_t nfn, mfn_t ofn,
> > + unsigned int level)
> > {
> > - if ( level != 1 || nt == ot )
> > + struct page_info *pg;
> > +
> > + if ( level != 1 || (nt == ot && mfn_eq(nfn, ofn)) )
> > return;
>
> Are you sure that foreign mappings (or ioreq server pages, for that
> matter) can never be level > 1?
Not given the current Xen interface, see
XENMEM_add_to_physmap{_batch}. This will have to change if the
interface is expanded to allow 2M or 1G mappings.
> ioreq server pages may be relatively harmless if we get out of sync; but
> the reference count with the foreign mapping is really dangerous if it
> gets screwed up.
>
> I'd be tempted to say that we should BUG_ON(level > 1 && nt == foreign).
Yes, I think that's a sensible safety belt.
>
> >
> > switch ( nt )
> > @@ -948,6 +951,17 @@ static inline void p2m_entry_modify(struct p2m_domain
> > *p2m, p2m_type_t nt,
> > p2m->ioreq.entry_count++;
> > break;
> >
> > + case p2m_map_foreign:
> > + pg = mfn_to_page(nfn);
> > +
> > + if ( !pg || !page_get_owner_and_reference(pg) )
> > + {
> > + ASSERT_UNREACHABLE();
> > + return;
> > + }
>
> Similarly, I'd be tempted to say we should BUG_ON() here instead. If a
> rogue guest can trigger this path, it would be a DoS; but the alternate
> would be to allow the mfn to be put into the p2m table without a
> reference, which could potentially be far worse.
>
> The alternate would be to have this return an error value, which would
> 1) cause the p2m write to fail, and 2) be checked all the way up the chain.
>
> Less worried about the removal side, as if we have BUG_ON's on the
> insertion side, they *really* shouldn't happen.
I would go for the BUG_ON ATM, because it's a simpler solution and
callers of p2m_entry_modify should make sure the operation is correct.
Thanks, Roger.
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |