Skip to contentSkip to conversation

do_memory_op: cleanup if copy_to_guest fails

PATCH In review Olaf Hering 2 participants 3 messages Dec 16, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 do_memory_op: cleanup if copy_to_guest fails

Undo the page allocation in the ulikely event the copy_to_guest fails.

---

I have not exercised this code path, it was found during code inspection in 4.0

 xen/common/memory.c |    9 +++++++++
 1 file changed, 9 insertions(+)
xen/common/memory.c +9 -0
@@ -82,7 +82,10 @@ static void increase_reservation(struct
         {
             mfn = page_to_mfn(page);
             if ( unlikely(__copy_to_guest_offset(a->extent_list, i, &mfn, 1)) )
+            {
+                free_domheap_pages(page, a->extent_order);
                 goto out;
+            }
         }
     }
 
@@ -144,7 +147,13 @@ static void populate_physmap(struct memo
 
                 /* Inform the domain of the new page's machine address. */ 
                 if ( unlikely(__copy_to_guest_offset(a->extent_list, i, &mfn, 1)) )
+                {
+                    for ( j = 0; j < (1 << a->extent_order); j++ )
+                        set_gpfn_from_mfn(mfn + j, INVALID_M2P_ENTRY);
+                    guest_physmap_remove_page(d, gpfn, mfn, a->extent_order);
+                    free_domheap_pages(page, a->extent_order);
                     goto out;
+                }
             }
         }
     }
#1 lore raw
Quoted 1 line · Olaf Hering

Undo the page allocation in the ulikely event the copy_to_guest fails.

You can't really clean up in this case. Once a page has been alloc'ed to a domain, it can immediately see it and map it. Trying to then free_domheap_page() ignoring the current page reference count is actually introducing a bug.

Leaving a bit of a mess on failed copy_to_guest is okay imo, as that is usually a pretty fatal sign anyway. If there were good reason for cleaning up better, we should at least be doing if (test_and_clear(PGC_allocated)) put_page().

-- Keir

Quoted 44 lines

---

I have not exercised this code path, it was found during code inspection in 4.0

 xen/common/memory.c |    9 +++++++++
 1 file changed, 9 insertions(+)
xen/common/memory.c +9 -0
@@ -82,7 +82,10 @@ static void increase_reservation(struct
         {
             mfn = page_to_mfn(page);
             if ( unlikely(__copy_to_guest_offset(a->extent_list, i, &mfn, 1))
 )
+            {
+                free_domheap_pages(page, a->extent_order);
                 goto out;
+            }
         }
     }
@@ -144,7 +147,13 @@ static void populate_physmap(struct memo
 
                 /* Inform the domain of the new page's machine address. */
                 if ( unlikely(__copy_to_guest_offset(a->extent_list, i, &mfn,
 1)) )
+                {
+                    for ( j = 0; j < (1 << a->extent_order); j++ )
+                        set_gpfn_from_mfn(mfn + j, INVALID_M2P_ENTRY);
+                    guest_physmap_remove_page(d, gpfn, mfn, a->extent_order);
+                    free_domheap_pages(page, a->extent_order);
                     goto out;
+                }
             }
         }
     }

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel

#2 lore raw
Quoted 4 lines · Keir Fraser

Leaving a bit of a mess on failed copy_to_guest is okay imo, as that is usually a pretty fatal sign anyway. If there were good reason for cleaning up better, we should at least be doing if (test_and_clear(PGC_allocated)) put_page().

Yes, it would leak just a page and from what I have seen, the Linux drivers will BUG() most of the time anyway if the hypercall fails. So I will drop this from my series.

Olaf