x86/physmap: Prevent incorrect updates of m2p mappings

In certain conditions, such as low memory, set_p2m_entry() can fail.
Currently, the p2m and m2p tables will get out of sync because we still
update the m2p table after the p2m update has failed.

If that happens, subsequent guest-invoked memory operations can cause
BUG()s and ASSERT()s to kill Xen.

This is fixed by only updating the m2p table iff the p2m was
successfully updated.

This is a security problem, XSA-22 / CVE-2012-4537.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
Acked-by: Ian Jackson <ian.jackson@eu.citrix.com>
Committed-by: Ian Jackson <ian.jackson@eu.citrix.com>

diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
index 9857b5b2dec9a191867196635ce14d36cb01a27b..e35194216c0c8f1c4f6e1f5c595b3b9070f91b9c 100644 (file)
--- a/xen/arch/x86/mm/p2m.c
+++ b/xen/arch/x86/mm/p2m.c
@@ -633,7 +633,10 @@ guest_physmap_add_entry(struct domain *d, unsigned long gfn,
     if ( mfn_valid(_mfn(mfn)) ) 
     {
         if ( !set_p2m_entry(p2m, gfn, _mfn(mfn), page_order, t, p2m->default_access) )
+        {
             rc = -EINVAL;
+            goto out; /* Failed to update p2m, bail without updating m2p. */
+        }
         if ( !p2m_is_grant(t) )
         {
             for ( i = 0; i < (1UL << page_order); i++ )
@@ -656,6 +659,7 @@ guest_physmap_add_entry(struct domain *d, unsigned long gfn,
         }
     }
 
+out:
     p2m_unlock(p2m);
 
     return rc;