SLUB: Fix memory leak by not reusing cpu_slab
authorChristoph Lameter <clameter@sgi.com>
Mon, 5 Nov 2007 19:31:58 +0000 (11:31 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Mon, 5 Nov 2007 19:37:12 +0000 (11:37 -0800)
Fix the memory leak that may occur when we attempt to reuse a cpu_slab
that was allocated while we reenabled interrupts in order to be able to
grow a slab cache.

The per cpu freelist may contain objects and in that situation we may
overwrite the per cpu freelist pointer loosing objects.  This only
occurs if we find that the concurrently allocated slab fits our
allocation needs.

If we simply always deactivate the slab then the freelist will be
properly reintegrated and the memory leak will go away.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Acked-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slub.c

index bcdb2c8941a5fcb428d4d3d256106f29e921debd..84f59fde1a10d83a78c4cfca6d08f5705da5021e 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1511,26 +1511,8 @@ new_slab:
 
        if (new) {
                c = get_cpu_slab(s, smp_processor_id());
-               if (c->page) {
-                       /*
-                        * Someone else populated the cpu_slab while we
-                        * enabled interrupts, or we have gotten scheduled
-                        * on another cpu. The page may not be on the
-                        * requested node even if __GFP_THISNODE was
-                        * specified. So we need to recheck.
-                        */
-                       if (node_match(c, node)) {
-                               /*
-                                * Current cpuslab is acceptable and we
-                                * want the current one since its cache hot
-                                */
-                               discard_slab(s, new);
-                               slab_lock(c->page);
-                               goto load_freelist;
-                       }
-                       /* New slab does not fit our expectations */
+               if (c->page)
                        flush_slab(s, c);
-               }
                slab_lock(new);
                SetSlabFrozen(new);
                c->page = new;