From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752699Ab1K1LBH (ORCPT ); Mon, 28 Nov 2011 06:01:07 -0500 Received: from mx2.parallels.com ([64.131.90.16]:33259 "EHLO mx2.parallels.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752143Ab1K1LBF (ORCPT ); Mon, 28 Nov 2011 06:01:05 -0500 Message-ID: <4ED369C8.9090101@parallels.com> Date: Mon, 28 Nov 2011 09:00:24 -0200 From: Glauber Costa User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:7.0) Gecko/20110927 Thunderbird/7.0 MIME-Version: 1.0 To: KAMEZAWA Hiroyuki CC: , , , , , , , , , , , , Subject: Re: [PATCH v6 10/10] Disable task moving when using kernel memory accounting References: <1322242696-27682-1-git-send-email-glommer@parallels.com> <1322242696-27682-11-git-send-email-glommer@parallels.com> <20111128133203.2d52ee28.kamezawa.hiroyu@jp.fujitsu.com> In-Reply-To: <20111128133203.2d52ee28.kamezawa.hiroyu@jp.fujitsu.com> Content-Type: text/plain; charset="ISO-8859-1"; format=flowed Content-Transfer-Encoding: 7bit X-Originating-IP: [201.82.130.234] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 11/28/2011 02:32 AM, KAMEZAWA Hiroyuki wrote: > On Fri, 25 Nov 2011 15:38:16 -0200 > Glauber Costa wrote: > >> Since this code is still experimental, we are leaving the exact >> details of how to move tasks between cgroups when kernel memory >> accounting is used as future work. >> >> For now, we simply disallow movement if there are any pending >> accounted memory. >> >> Signed-off-by: Glauber Costa >> CC: Hiroyouki Kamezawa >> --- >> mm/memcontrol.c | 23 ++++++++++++++++++++++- >> 1 files changed, 22 insertions(+), 1 deletions(-) >> >> diff --git a/mm/memcontrol.c b/mm/memcontrol.c >> index 2df5d3c..ab7e57b 100644 >> --- a/mm/memcontrol.c >> +++ b/mm/memcontrol.c >> @@ -5451,10 +5451,19 @@ static int mem_cgroup_can_attach(struct cgroup_subsys *ss, >> { >> int ret = 0; >> struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); >> + struct mem_cgroup *from = mem_cgroup_from_task(p); >> + >> +#if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM)&& defined(CONFIG_INET) >> + if (from != mem&& !mem_cgroup_is_root(from)&& >> + res_counter_read_u64(&from->tcp_mem.tcp_memory_allocated, RES_USAGE)) { >> + printk(KERN_WARNING "Can't move tasks between cgroups: " >> + "Kernel memory held. task: %s\n", p->comm); >> + return 1; >> + } >> +#endif > > Hmm, the kernel memory is not guaranteed as being held by the 'task' ? > > How about > "Now, moving task between cgroup is disallowed while the source cgroup > containes kmem reference." ? > > Hmm.. we need to fix this task-move/rmdir issue before production use. > > > Thanks, > -Kame > Hi Kame, Let me tell you the direction I am going wrt task movement: The only reasons I haven't included so far, is that I believe it needs more testing, and as you know, I am right now more interested in getting past the initial barriers for inclusion. I am committed to fix anything that needs to be fixed - stylish or non-stylish before we remove the experimental flag. So what I intend to do, is to basically * lock the task, * scan through its file descriptors list, * identify which of them are sockets, * cast them to struct sock *, * see if it has a cgrp associated * see if cgrp == from At this point we can decrement sockets allocated by 1 in from, and memory_allocated by sk_forward_alloc (increasing by equal quantities in the destination cgroup) I belive it will work.