Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753234AbbBKPln (ORCPT ); Wed, 11 Feb 2015 10:41:43 -0500 Received: from mx2.parallels.com ([199.115.105.18]:57118 "EHLO mx2.parallels.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752233AbbBKPlm (ORCPT ); Wed, 11 Feb 2015 10:41:42 -0500 Date: Wed, 11 Feb 2015 18:41:28 +0300 From: Vladimir Davydov To: Christoph Lameter CC: Andrew Morton , , , Huang Ying , "Pekka Enberg" , David Rientjes , "Joonsoo Kim" Subject: Re: [PATCH -mm] slub: kmem_cache_shrink: init discard list after freeing slabs Message-ID: <20150211154128.GA26049@esperanza> References: <1423627463.5968.99.camel@intel.com> <1423642582-23553-1-git-send-email-vdavydov@parallels.com> MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Disposition: inline In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 4446 Lines: 105 On Wed, Feb 11, 2015 at 09:00:39AM -0600, Christoph Lameter wrote: > Hmmmm... Thinking about this some more. It may be better to initialize the > list head at the beginning of the loop? > > Also the promote array should also be initialized in the loop right? I do initialize promote lists in the loop using list_splice_init, but yeah, initializing them in the beginning of the loop would look more readable indeed. The updated patch is below. Thanks! --- From: Vladimir Davydov Subject: [PATCH] slub: kmem_cache_shrink: fix crash due to uninitialized discard list Currently, the discard list is only initialized at the beginning of the function. As a result, if there are > 1 nodes, we can get use-after-free while processing the second or higher node: WARNING: CPU: 60 PID: 1 at lib/list_debug.c:29 __list_add+0x3c/0xa9() list_add corruption. next->prev should be prev (ffff881ff0a6bb98), but was ffffea007ff57020. (next=ffffea007fbf7320). Modules linked in: CPU: 60 PID: 1 Comm: swapper/0 Not tainted 3.19.0-rc7-next-20150203-gb50cadf #2178 Hardware name: Intel Corporation BRICKLAND/BRICKLAND, BIOS BIVTSDP1.86B.0038.R02.1307231126 07/23/2013 0000000000000009 ffff881ff0a6ba88 ffffffff81c2e096 ffffffff810e2d03 ffff881ff0a6bad8 ffff881ff0a6bac8 ffffffff8108b320 ffff881ff0a6bb18 ffffffff8154bbc7 ffff881ff0a6bb98 ffffea007fbf7320 ffffea00ffc3c220 Call Trace: [] dump_stack+0x4c/0x65 [] ? console_unlock+0x398/0x3c7 [] warn_slowpath_common+0xa1/0xbb [] ? __list_add+0x3c/0xa9 [] warn_slowpath_fmt+0x46/0x48 [] __list_add+0x3c/0xa9 [] __kmem_cache_shrink+0x12b/0x24c [] kmem_cache_shrink+0x26/0x38 [] acpi_os_purge_cache+0xe/0x12 [] acpi_purge_cached_objects+0x32/0x7a [] acpi_initialize_objects+0x17e/0x1ae [] ? acpi_sleep_proc_init+0x2a/0x2a [] acpi_init+0x92/0x25e [] ? do_one_initcall+0x90/0x17f [] ? kfree+0x1fc/0x2d5 [] ? acpi_sleep_proc_init+0x2a/0x2a [] do_one_initcall+0xed/0x17f [] kernel_init_freeable+0x1f0/0x278 [] ? rest_init+0x13e/0x13e [] kernel_init+0xe/0xda [] ret_from_fork+0x7c/0xb0 [] ? rest_init+0x13e/0x13e Fix this by initializing the discard list at each iteration of the for_each_kmem_cache_node loop. Also, move promote lists initialization to the beginning of the loop to conform. fixes: slub-never-fail-to-shrink-cache Signed-off-by: Vladimir Davydov Reported-by: Huang Ying Cc: Christoph Lameter Cc: Pekka Enberg Cc: David Rientjes Cc: Joonsoo Kim diff --git a/mm/slub.c b/mm/slub.c index 0909e13cf708..6832c4eab104 100644 --- a/mm/slub.c +++ b/mm/slub.c @@ -3437,7 +3437,7 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) struct kmem_cache_node *n; struct page *page; struct page *t; - LIST_HEAD(discard); + struct list_head discard; struct list_head promote[SHRINK_PROMOTE_MAX]; unsigned long flags; int ret = 0; @@ -3457,11 +3457,12 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) kick_all_cpus_sync(); } - for (i = 0; i < SHRINK_PROMOTE_MAX; i++) - INIT_LIST_HEAD(promote + i); - flush_all(s); for_each_kmem_cache_node(s, node, n) { + INIT_LIST_HEAD(&discard); + for (i = 0; i < SHRINK_PROMOTE_MAX; i++) + INIT_LIST_HEAD(promote + i); + spin_lock_irqsave(&n->list_lock, flags); /* @@ -3491,7 +3492,7 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) * partial list. */ for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) - list_splice_init(promote + i, &n->partial); + list_splice(promote + i, &n->partial); spin_unlock_irqrestore(&n->list_lock, flags); -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/