From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-pf0-f199.google.com (mail-pf0-f199.google.com [209.85.192.199]) by kanga.kvack.org (Postfix) with ESMTP id 449596B0003 for ; Tue, 10 Apr 2018 11:18:23 -0400 (EDT) Received: by mail-pf0-f199.google.com with SMTP id b16so7071568pfi.5 for ; Tue, 10 Apr 2018 08:18:23 -0700 (PDT) Received: from mail.kernel.org (mail.kernel.org. [198.145.29.99]) by mx.google.com with ESMTPS id b5-v6si2722280ple.584.2018.04.10.08.18.22 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Tue, 10 Apr 2018 08:18:22 -0700 (PDT) Date: Tue, 10 Apr 2018 08:18:20 -0700 From: Jaegeuk Kim Subject: Re: [PATCH 2/2] page cache: Mask off unwanted GFP flags Message-ID: <20180410151820.GA69325@jaegeuk-macbookpro.roam.corp.google.com> References: <20180410125351.15837-1-willy@infradead.org> <20180410125351.15837-2-willy@infradead.org> <20180410134545.GA35354@rodete-laptop-imager.corp.google.com> <20180410140223.GE22118@bombadil.infradead.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180410140223.GE22118@bombadil.infradead.org> Sender: owner-linux-mm@kvack.org List-ID: To: Matthew Wilcox Cc: Minchan Kim , linux-mm@kvack.org, Matthew Wilcox , Christoph Lameter , Pekka Enberg , David Rientjes , Joonsoo Kim , Andrew Morton , linux-kernel@vger.kernel.org, Jan Kara , Jeff Layton , Mel Gorman , stable@vger.kernel.org On 04/10, Matthew Wilcox wrote: > On Tue, Apr 10, 2018 at 10:45:45PM +0900, Minchan Kim wrote: > > On Tue, Apr 10, 2018 at 05:53:51AM -0700, Matthew Wilcox wrote: > > > From: Matthew Wilcox > > > > > > The page cache has used the mapping's GFP flags for allocating > > > radix tree nodes for a long time. It took care to always mask off the > > > __GFP_HIGHMEM flag, and masked off other flags in other paths, but the > > > __GFP_ZERO flag was still able to sneak through. The __GFP_DMA and > > > __GFP_DMA32 flags would also have been able to sneak through if they > > > were ever used. Fix them all by using GFP_RECLAIM_MASK at the innermost > > > location, and remove it from earlier in the callchain. > > > > > > Fixes: 19f99cee206c ("f2fs: add core inode operations") > > > > Why this patch fix 19f99cee206c instead of 449dd6984d0e? > > F2FS doesn't have any problem before introducing 449dd6984d0e? > > Well, there's the problem. This bug is the combination of three different > things: > > 1. The working set code relying on list_empty. > 2. The page cache not filtering out the bad flags. > 3. F2FS specifying a flag nobody had ever specified before. > > So what single patch does this patch fix? I don't think it really matters. Hope there'd be someone who does care about patch description though, IMHO, this fixes the MM regression introduced by: 449dd6984d0e ("mm: keep page cache radix tree nodes in check") merged in v3.15, 2014. 19f99cee206c ("f2fs: add core inode operations) merged in v3.8, 2012, just revealed this out. In fact, I've never hit this bug in old kernels. >>From the user viewpoint, may I suggest to describe what kind of symptom we're able to see due to this bug? Something like: [ 7858.792946] [] __list_del_entry+0x30/0xd0 [ 7858.792951] [] list_lru_del+0xac/0x1ac [ 7858.792957] [] page_cache_tree_insert+0xd8/0x110 [ 7858.792962] [] __add_to_page_cache_locked+0xf8/0x4e0 [ 7858.792967] [] add_to_page_cache_lru+0x50/0x1ac [ 7858.792972] [] pagecache_get_page+0x468/0x57c [ 7858.792979] [] __get_node_page+0x84/0x764 [ 7858.792986] [] f2fs_iget+0x264/0xdc8 [ 7858.792991] [] f2fs_lookup+0x3b4/0x660 [ 7858.792998] [] lookup_slow+0x1e4/0x348 [ 7858.793003] [] walk_component+0x21c/0x320 [ 7858.793008] [] path_lookupat+0x90/0x1bc [ 7858.793013] [] filename_lookup+0x8c/0x1a0 [ 7858.793018] [] vfs_fstatat+0x84/0x10c [ 7858.793023] [] SyS_newfstatat+0x28/0x64 Thanks,