page allocator: sanity check order in the page allocator slow path

Callers may speculatively call different allocators in order of preference
trying to allocate a buffer of a given size.  The order needed to allocate
this may be larger than what the page allocator can normally handle.
While the allocator mostly does the right thing, it should not direct
reclaim or wakeup kswapd with a bogus order.  This patch sanity checks the
order in the slow path and returns NULL if it is too large.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Dave Hansen <dave@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c

@@ -1446,9 +1446,6 @@ get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
 	int zlc_active = 0;		/* set if using zonelist_cache */
 	int did_zlc_setup = 0;		/* just call zlc_setup() one time */
 
-	if (WARN_ON_ONCE(order >= MAX_ORDER))
-		return NULL;
-
 	classzone_idx = zone_idx(preferred_zone);
 zonelist_scan:
 	/*
@@ -1706,6 +1703,15 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
 	unsigned long did_some_progress;
 	struct task_struct *p = current;
 
+	/*
+	 * In the slowpath, we sanity check order to avoid ever trying to
+	 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
+	 * be using allocators in order of preference for an area that is
+	 * too large.
+	 */
+	if (WARN_ON_ONCE(order >= MAX_ORDER))
+		return NULL;
+
 	/*
 	 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
 	 * __GFP_NOWARN set) should not cause reclaim since the subsystem