Re: [PATCH v4 1/5] cgroup: subtree_control bypass mode for bypassable controllers

From: Waiman Long
Date: Mon Dec 10 2018 - 14:56:31 EST

Next message: Dave Watson: "[PATCH 00/12] x86/crypto: gcmaes AVX scatter/gather support"
Previous message: David Miller: "Re: [PATCH net-next] ipvlan: Remove a useless comparison"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 11/29/2018 06:18 AM, Dan Carpenter wrote:
> Hi Waiman,
>
> Thank you for the patch! Perhaps something to improve:
>
> url: https://github.com/0day-ci/linux/commits/Waiman-Long/cgroup-Introducing-bypass-mode/20181123-030552
> base: https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git for-next
>
> smatch warnings:
> kernel/cgroup/cgroup.c:4893 css_create() error: we previously assumed 'parent' could be null (see line 4864)
>
> # https://github.com/0day-ci/linux/commit/8b68fd4330e043645667a5d3306398f8f88f9ff2
> git remote add linux-review https://github.com/0day-ci/linux
> git remote update linux-review
> git checkout 8b68fd4330e043645667a5d3306398f8f88f9ff2
> vim +/parent +4893 kernel/cgroup/cgroup.c
>
> a31f2d3ff kernel/cgroup.c Tejun Heo 2012-11-19 4840
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4841 /**
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4842 * css_create - create a cgroup_subsys_state
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4843 * @cgrp: the cgroup new css will be associated with
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4844 * @ss: the subsys of new css
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4845 *
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4846 * Create a new css associated with @cgrp - @ss pair. On success, the new
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4847 * css is online and installed in @cgrp. This function doesn't create the
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4848 * interface files. Returns 0 on success, -errno on failure.
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4849 */
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4850 static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4851 struct cgroup_subsys *ss)
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4852 {
> d51f39b05 kernel/cgroup.c Tejun Heo 2014-05-16 4853 struct cgroup *parent = cgroup_parent(cgrp);
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4854 struct cgroup_subsys_state *parent_css = NULL;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4855 struct cgroup_subsys_state *css;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4856 int err;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4857
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4858 lockdep_assert_held(&cgroup_mutex);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4859
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4860 /*
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4861 * As cgroup may be in bypass mode, need to skip over ancestor
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4862 * cgroups with NULL CSS.
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4863 */
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 @4864 for (; parent && !parent_css; parent = cgroup_parent(parent))
> ^^^^^^^^^^^^^^^^^^^^^
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4865 parent_css = cgroup_css(parent, ss);
>
> When we exit this loop it means either parent is NULL or parent_css
> is non-NULL.

Sorry for the bug. I should have a temp variable to go up
cgroup_parent() iteration chain and leave parent alone. Thanks for
spotting that.

> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4866
> 1fed1b2e3 kernel/cgroup.c Tejun Heo 2014-05-16 4867 css = ss->css_alloc(parent_css);
> e7e15b87f kernel/cgroup.c Tejun Heo 2016-06-21 4868 if (!css)
> e7e15b87f kernel/cgroup.c Tejun Heo 2016-06-21 4869 css = ERR_PTR(-ENOMEM);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4870 if (IS_ERR(css))
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4871 return css;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4872
> 8b68fd433 kernel/cgroup/cgroup.c Waiman Long 2018-11-20 4873 init_and_link_css(css, ss, cgrp, parent_css);
> a2bed8209 kernel/cgroup.c Tejun Heo 2014-05-04 4874
> 2aad2a86f kernel/cgroup.c Tejun Heo 2014-09-24 4875 err = percpu_ref_init(&css->refcnt, css_release, 0, GFP_KERNEL);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4876 if (err)
> 3eb59ec64 kernel/cgroup.c Li Zefan 2014-03-18 4877 goto err_free_css;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4878
> cf780b7dc kernel/cgroup.c Vladimir Davydov 2015-08-03 4879 err = cgroup_idr_alloc(&ss->css_idr, NULL, 2, 0, GFP_KERNEL);
> 15a4c835e kernel/cgroup.c Tejun Heo 2014-05-04 4880 if (err < 0)
> b00c52dae kernel/cgroup.c Wenwei Tao 2016-05-13 4881 goto err_free_css;
> 15a4c835e kernel/cgroup.c Tejun Heo 2014-05-04 4882 css->id = err;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4883
> 15a4c835e kernel/cgroup.c Tejun Heo 2014-05-04 4884 /* @css is ready to be brought online now, make it visible */
> 1fed1b2e3 kernel/cgroup.c Tejun Heo 2014-05-16 4885 list_add_tail_rcu(&css->sibling, &parent_css->children);
> 15a4c835e kernel/cgroup.c Tejun Heo 2014-05-04 4886 cgroup_idr_replace(&ss->css_idr, css, css->id);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4887
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4888 err = online_css(css);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4889 if (err)
> 1fed1b2e3 kernel/cgroup.c Tejun Heo 2014-05-16 4890 goto err_list_del;
> 944196278 kernel/cgroup.c Tejun Heo 2014-03-19 4891
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4892 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
> d51f39b05 kernel/cgroup.c Tejun Heo 2014-05-16 @4893 cgroup_parent(parent)) {
> ^^^^^^
> We dereference parent inside the function, but I don't know for sure
> if this is reachable when "parent" is NULL.
>
> ed3d261b5 kernel/cgroup.c Joe Perches 2014-04-25 4894 pr_warn("%s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4895 current->comm, current->pid, ss->name);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4896 if (!strcmp(ss->name, "memory"))
> ed3d261b5 kernel/cgroup.c Joe Perches 2014-04-25 4897 pr_warn("\"memory\" requires setting use_hierarchy to 1 on the root\n");
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4898 ss->warned_broken_hierarchy = true;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4899 }
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4900
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4901 return css;
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4902
> 1fed1b2e3 kernel/cgroup.c Tejun Heo 2014-05-16 4903 err_list_del:
> 1fed1b2e3 kernel/cgroup.c Tejun Heo 2014-05-16 4904 list_del_rcu(&css->sibling);
> 3eb59ec64 kernel/cgroup.c Li Zefan 2014-03-18 4905 err_free_css:
> 8f53470ba kernel/cgroup/cgroup.c Tejun Heo 2018-04-26 4906 list_del_rcu(&css->rstat_css_node);
> 8f36aaec9 kernel/cgroup/cgroup.c Tejun Heo 2018-03-14 4907 INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
> 8f36aaec9 kernel/cgroup/cgroup.c Tejun Heo 2018-03-14 4908 queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
> 6cd0f5bba kernel/cgroup.c Tejun Heo 2016-03-03 4909 return ERR_PTR(err);
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4910 }
> c81c925ad kernel/cgroup.c Tejun Heo 2013-12-06 4911
>
> ---
> 0-DAY kernel test infrastructure Open Source Technology Center
> https://lists.01.org/pipermail/kbuild-all Intel Corporation

Cheers,
Longman

Next message: Dave Watson: "[PATCH 00/12] x86/crypto: gcmaes AVX scatter/gather support"
Previous message: David Miller: "Re: [PATCH net-next] ipvlan: Remove a useless comparison"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]