|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH] x86/p2m: prevent replacing special p2m entries
On 18.11.2026 18:20, Roger Pau Monne wrote:
> The current logic in both p2m_add_foreign() and xenmem_add_to_physmap_one()
> allows replacing special p2m types, that should otherwise only be
> established/cleaned up via special accessors.
>
> When the backing page for a foreign p2m-type mapping only has one reference
> left, the put_page() in p2m_entry_modify() will free the page ahead of it
> being flushed from the p2m and the IOMMU TLBs. This can be triggered with
> IOREQ servers, and requires a domain that has privileges over another one
> (ie: a stubdom). Such a domain would create an IOREQ server, map it using
> XENMEM_acquire_resource (which creates a p2m_map_foreign p2m entry),
> destroy the IOREQ server, and afterwards map something over the IOREQ
> entries using XENMEM_add_to_physmap{,_batch}. At this point the removal
> from the p2m will trigger an assert in p2m_entry_modify() on debug builds,
> or in releases < 4.22 free the page before the TLBs are flushed.
>
> Something similar can also happen with grant mppaings: entries being
Nit: mappings
> replaced using p2m_add_foreign() or xenmem_add_to_physmap_one() will then
> fail when attempting to be cleaned up by replace_grant_host_mapping().
> However that only affects the domain itself, and is no worse than what a
> grant-mapping domain can already do to the grant source.
>
> Note that p2m_add_page() already refuses to replace p2m entries of special
> types.
>
> Fixes: b37d30c8fc52 ("Mapping grant references into HVM guests, take 2")
> Fixes: 0537d246f8db ("mm: add 'is_special_page' inline function")
> Signed-off-by: Roger Pau Monné <roger@xxxxxxxxxxxxxx>
Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx>
However, I thought we had agreed that ...
> --- a/xen/arch/x86/mm/p2m.c
> +++ b/xen/arch/x86/mm/p2m.c
> @@ -1953,6 +1953,13 @@ static int p2m_add_foreign(struct domain *tdom,
> unsigned long fgfn,
>
> /* Remove previously mapped page if it is present. */
> prev_mfn = get_gfn(tdom, gpfn, &p2mt_prev);
> + /* Special p2m slots must be cleaned up via special accessors. */
> + if ( p2m_is_special(p2mt_prev) )
> + {
> + rc = -EPERM;
> + goto put_both;
> + }
... in particular this change comes with a (hopefully small) risk of breaking
existing code which maps one foreign page over a prior such mapping (never
torn down properly). And that you'd mention this aspect in the commit message.
Jan
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |