From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754491AbdKAJjY (ORCPT ); Wed, 1 Nov 2017 05:39:24 -0400 Received: from szxga06-in.huawei.com ([45.249.212.32]:40269 "EHLO huawei.com" rhost-flags-OK-FAIL-OK-FAIL) by vger.kernel.org with ESMTP id S1754462AbdKAJjW (ORCPT ); Wed, 1 Nov 2017 05:39:22 -0400 Subject: Re: [PATCH RFC v2 1/4] mm/mempolicy: Fix get_nodes() mask miscalculation To: Vlastimil Babka , , , , , , References: <1509099265-30868-1-git-send-email-xieyisheng1@huawei.com> <1509099265-30868-2-git-send-email-xieyisheng1@huawei.com> <922a4767-9eed-40aa-c437-6f6fcdcab150@suse.cz> CC: , , , , Andi Kleen From: Yisheng Xie Message-ID: Date: Wed, 1 Nov 2017 17:37:36 +0800 User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.1.0 MIME-Version: 1.0 In-Reply-To: <922a4767-9eed-40aa-c437-6f6fcdcab150@suse.cz> Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit X-Originating-IP: [10.177.29.40] X-CFilter-Loop: Reflected Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Vlastimil, Thanks for comment! On 2017/10/31 16:34, Vlastimil Babka wrote: > On 10/27/2017 12:14 PM, Yisheng Xie wrote: >> It appears there is a nodemask miscalculation in the get_nodes() >> function in mm/mempolicy.c. This bug has two effects: >> >> 1. It is impossible to specify a length 1 nodemask. >> 2. It is impossible to specify a nodemask containing the last node. > > This should be more specific, which syscalls are you talking about? > I assume it's set_mempolicy() and mbind() and it's the same issue that > was discussed at https://marc.info/?l=linux-mm&m=150732591909576&w=2 ? I just missed this thread, sorry about that. Not only set_mempolicy() and mbind(), but migrate_pages() also suffers this problem. Maybe related manpage should documented this as your mentioned below. Thanks Yisheng Xie > >> Brent have submmit a patch before v2.6.12, however, Andi revert his >> changed for ABI problem. I just resent this patch as RFC, for do not >> clear about what's the problem Andi have met. > > You should have CC'd Andi. As was discussed in the other thread, this > would make existing programs potentially unsafe, so we can't change it. > Instead it should be documented. > >> As manpage of set_mempolicy, If the value of maxnode is zero, the >> nodemask argument is ignored. but we should not ignore the nodemask >> when maxnode is 1. >>