x86: don't allow page_unlock() to drop the last type reference

Only _put_page_type() does the necessary cleanup, and hence not all
domain pages can be released during guest cleanup (leaving around
zombie domains) if we get this wrong.

This is XSA-242.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
master commit: 6410733a8a0dff2fe581338ff631670cf91889db
master date: 2017-10-12 14:49:46 +0200

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 16fc39c1cf6425973b294166c65fb72c0f008922..a3e42f721aaf0c62616b529a1708828ffaea5884 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -1739,7 +1739,11 @@ void page_unlock(struct page_info *page)
 
     do {
         x = y;
+        ASSERT((x & PGT_count_mask) && (x & PGT_locked));
+
         nx = x - (1 | PGT_locked);
+        /* We must not drop the last reference here. */
+        ASSERT(nx & PGT_count_mask);
     } while ( (y = cmpxchg(&page->u.inuse.type_info, x, nx)) != x );
 }
 
@@ -2421,6 +2425,17 @@ static int _put_page_type(struct page_info *page, bool_t preemptible,
                    (page->count_info & PGC_page_table)) )
                 page_set_tlbflush_timestamp(page);
         }
+        else if ( unlikely((nx & (PGT_locked | PGT_count_mask)) ==
+                           (PGT_locked | 1)) )
+        {
+            /*
+             * We must not drop the second to last reference when the page is
+             * locked, as page_unlock() doesn't do any cleanup of the type.
+             */
+            cpu_relax();
+            y = page->u.inuse.type_info;
+            continue;
+        }
 
         if ( likely((y = cmpxchg(&page->u.inuse.type_info, x, nx)) == x) )
             break;