From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-oi0-f71.google.com (mail-oi0-f71.google.com [209.85.218.71]) by kanga.kvack.org (Postfix) with ESMTP id 4C4146B0038 for ; Mon, 3 Apr 2017 08:36:51 -0400 (EDT) Received: by mail-oi0-f71.google.com with SMTP id e188so101370614oif.18 for ; Mon, 03 Apr 2017 05:36:51 -0700 (PDT) Received: from EUR03-VE1-obe.outbound.protection.outlook.com (mail-eopbgr50112.outbound.protection.outlook.com. [40.107.5.112]) by mx.google.com with ESMTPS id 34si6507231otr.209.2017.04.03.05.36.50 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-SHA bits=128/128); Mon, 03 Apr 2017 05:36:50 -0700 (PDT) Subject: Re: [PATCH] mm/zswap: fix potential deadlock in zswap_frontswap_store() References: <20170331153009.11397-1-aryabinin@virtuozzo.com> <20170403084729.GG24661@dhcp22.suse.cz> From: Andrey Ryabinin Message-ID: Date: Mon, 3 Apr 2017 15:38:08 +0300 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset="windows-1252" Content-Transfer-Encoding: 7bit Sender: owner-linux-mm@kvack.org List-ID: To: Michal Hocko , Shakeel Butt Cc: Seth Jennings , Dan Streetman , Linux MM , LKML , Andrew Morton , Vlastimil Babka On 04/03/2017 03:37 PM, Andrey Ryabinin wrote: > > > On 04/03/2017 11:47 AM, Michal Hocko wrote: >> On Fri 31-03-17 10:00:30, Shakeel Butt wrote: >>> On Fri, Mar 31, 2017 at 8:30 AM, Andrey Ryabinin >>> wrote: >>>> zswap_frontswap_store() is called during memory reclaim from >>>> __frontswap_store() from swap_writepage() from shrink_page_list(). >>>> This may happen in NOFS context, thus zswap shouldn't use __GFP_FS, >>>> otherwise we may renter into fs code and deadlock. >>>> zswap_frontswap_store() also shouldn't use __GFP_IO to avoid recursion >>>> into itself. >>>> >>> >>> Is it possible to enter fs code (or IO) from zswap_frontswap_store() >>> other than recursive memory reclaim? However recursive memory reclaim >>> is protected through PF_MEMALLOC task flag. The change seems fine but >>> IMHO reasoning needs an update. Adding Michal for expert opinion. >> >> Yes this is true. > > Actually, no. I think we have a bug in allocator which may lead to recursive direct reclaim. > > E.g. for costly order allocations (or order > 0 && ac->migratetype != MIGRATE_MOVABLE) > with __GFP_NOMEMALLOC (gfp_pfmemalloc_allowed() returns false) > __alloc_pages_slowpath() may call __alloc_pages_direct_compact() and unconditionally clear PF_MEMALLOC: > > __alloc_pages_direct_compact(): > ... > current->flags |= PF_MEMALLOC; > *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac, > prio); > current->flags &= ~PF_MEMALLOC; > > > > And later in __alloc_pages_slowpath(): > > /* Avoid recursion of direct reclaim */ > if (current->flags & PF_MEMALLOC) <=== false > goto nopage; > > /* Try direct reclaim and then allocating */ > page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac, > &did_some_progress); > Seems it was broken by a8161d1ed6098506303c65b3701dedba876df42a Author: Vlastimil Babka Date: Thu Jul 28 15:49:19 2016 -0700 mm, page_alloc: restructure direct compaction handling in slowpath -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org