From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751438AbbCHB1r (ORCPT ); Sat, 7 Mar 2015 20:27:47 -0500 Received: from mail-ig0-f179.google.com ([209.85.213.179]:39217 "EHLO mail-ig0-f179.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751024AbbCHB1p (ORCPT ); Sat, 7 Mar 2015 20:27:45 -0500 Date: Sat, 7 Mar 2015 17:27:43 -0800 (PST) From: David Rientjes X-X-Sender: rientjes@chino.kir.corp.google.com To: Jens Axboe cc: Lars Ellenberg , linux-kernel@vger.kernel.org, drbd-user@lists.linbit.com Subject: Re: [patch 1/2] block, drbd: fix drbd_req_new() initialization In-Reply-To: <54FB9FE2.5040707@fb.com> Message-ID: References: <54FB98F3.2050308@fb.com> <54FB9FE2.5040707@fb.com> User-Agent: Alpine 2.10 (DEB 1266 2009-07-14) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sat, 7 Mar 2015, Jens Axboe wrote: > >>> mempool_alloc() does not support __GFP_ZERO since elements may come from > >>> memory that has already been released by mempool_free(). > >>> > >>> Remove __GFP_ZERO from mempool_alloc() in drbd_req_new() and properly > >>> initialize it to 0. > >> > >> You should add it to mempool instead, avoid having this issue show up for > >> other folks as well. It'd be trivial to do. Normal ->alloc() should honor > >> __GFP_ZERO, just do the same manually for removing an item from the internal > >> pool. > >> > > > > Umm, it's not trivial to do and wouldn't make sense to do it. Mempools > > Uhm, it would make sense, though. > Disagree, I don't think we should extend mempool to know the element size, modify every user of mempool to pass it in, and keep it consistent with mempool_alloc_t for the benefit of __GFP_ZERO for this one buggy caller. Most users don't need __GFP_ZERO and just overwrite the entire element after mempool_alloc() and it would be an unnecessary overhead to even check for the bit set. So it wouldn't make sense in terms of performance or maintainability. > > don't know the element size, in other words it wouldn't know the length to > > memset() to 0 for mempool_alloc(). It shouldn't be modified to know the > > element size since elements are allocated by the implementation of > > mempool_alloc_t and they could easily become inconsistent. This patch is > > what you want to merge, really. > > > > I forgot we don't have the size in there. Then I would suggest adding a > WARN_ON() for __GFP_ZERO being set in mempool_alloc(), at the very least. > There is, it's a VM_WARN_ON_ONCE() that will show up if you configure CONFIG_DEBUG_VM.