mm: fix rss count leakage during migration
authorKonstantin Khlebnikov <khlebnikov@openvz.org>
Fri, 20 Jan 2012 22:34:24 +0000 (14:34 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 23 Jan 2012 16:38:49 +0000 (08:38 -0800)
Memory migration fills a pte with a migration entry and it doesn't
update the rss counters.  Then it replaces the migration entry with the
new page (or the old one if migration failed).  But between these two
passes this pte can be unmaped, or a task can fork a child and it will
get a copy of this migration entry.  Nobody accounts for this in the rss
counters.

This patch properly adjust rss counters for migration entries in
zap_pte_range() and copy_one_pte().  Thus we avoid extra atomic
operations on the migration fast-path.

Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memory.c

index 5e30583c2605d7b791d907ae8a5e07be2d9ad5b6..fa2f04e0337c437e739e75a683ff5b80a070a95a 100644 (file)
@@ -878,15 +878,24 @@ copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
                        }
                        if (likely(!non_swap_entry(entry)))
                                rss[MM_SWAPENTS]++;
-                       else if (is_write_migration_entry(entry) &&
-                                       is_cow_mapping(vm_flags)) {
-                               /*
-                                * COW mappings require pages in both parent
-                                * and child to be set to read.
-                                */
-                               make_migration_entry_read(&entry);
-                               pte = swp_entry_to_pte(entry);
-                               set_pte_at(src_mm, addr, src_pte, pte);
+                       else if (is_migration_entry(entry)) {
+                               page = migration_entry_to_page(entry);
+
+                               if (PageAnon(page))
+                                       rss[MM_ANONPAGES]++;
+                               else
+                                       rss[MM_FILEPAGES]++;
+
+                               if (is_write_migration_entry(entry) &&
+                                   is_cow_mapping(vm_flags)) {
+                                       /*
+                                        * COW mappings require pages in both
+                                        * parent and child to be set to read.
+                                        */
+                                       make_migration_entry_read(&entry);
+                                       pte = swp_entry_to_pte(entry);
+                                       set_pte_at(src_mm, addr, src_pte, pte);
+                               }
                        }
                }
                goto out_set_pte;
@@ -1191,6 +1200,16 @@ again:
 
                        if (!non_swap_entry(entry))
                                rss[MM_SWAPENTS]--;
+                       else if (is_migration_entry(entry)) {
+                               struct page *page;
+
+                               page = migration_entry_to_page(entry);
+
+                               if (PageAnon(page))
+                                       rss[MM_ANONPAGES]--;
+                               else
+                                       rss[MM_FILEPAGES]--;
+                       }
                        if (unlikely(!free_swap_and_cache(entry)))
                                print_bad_pte(vma, addr, ptent, NULL);
                }