From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758049Ab3LFQNU (ORCPT ); Fri, 6 Dec 2013 11:13:20 -0500 Received: from mail-yh0-f54.google.com ([209.85.213.54]:52766 "EHLO mail-yh0-f54.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752498Ab3LFQNS (ORCPT ); Fri, 6 Dec 2013 11:13:18 -0500 Date: Fri, 6 Dec 2013 11:13:12 -0500 From: Tejun Heo To: Vladimir Davydov Cc: Li Zefan , linux-kernel@vger.kernel.org, cgroups@vger.kernel.org, devel@openvz.org Subject: Re: [PATCH cgroup/for-3.13-fixes] cgroup: fix oops in cgroup init failure path Message-ID: <20131206161312.GC13373@htj.dyndns.org> References: <1386234006-1633-1-git-send-email-vdavydov@parallels.com> <20131205211814.GA29598@mtj.dyndns.org> <52A1766F.5090808@parallels.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <52A1766F.5090808@parallels.com> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hello, Vladimir. On Fri, Dec 06, 2013 at 11:02:07AM +0400, Vladimir Davydov wrote: > This patch fixes this bug, but I have a couple of questions regarding it. > > First, cgroup_load_subsys() also calls css_online(), and if it fails, it > calls cgroup_unload_subsys() to rollback. The latter function executes > the following command: > > offline_css(cgroup_css(cgroup_dummy_top, ss)); > > But since we failed to online_css(), cgroup_css() will return NULL > resulting in another oops. I don't think the root css onlining fails for any existing controllers but yeah that looks wrong. Can you please send a patch? > Second, it's not clear to me why we need the CSS_ONLINE flag at all if > we never assign css's that we fail to online to a cgroup. AFAIU we will > never see such css's, because in all places we call offline_css(), > namely cgroup_destroy_locked() (via kill_css()) and > cgroup_unload_subsys(), we use cgroup_css() which will return NULL for them. The whole thing is in flux and will look very different in near future. I actually had patches queued which deal with the issue you spotted but they are being blocked on other changes ATM. So, yeah, there are some spurious stuff now. > Before we get here, we call > > /* each css holds a ref to the cgroup's dentry and the parent css */ > for_each_root_subsys(root, ss) { > struct cgroup_subsys_state *css = css_ar[ss->subsys_id]; > > dget(dentry); > css_get(css->parent); > } > > If we fail to online a css, we will only call > > ss->css_free(css); > > on it skipping css_put() on parent. > > css_put() is called on parent in css_release() on normal destroy path. You're right. I'll revise the patch. Thanks. -- tejun