sched/cgroup: Fix/cleanup cgroup teardown/init
authorPeter Zijlstra <peterz@infradead.org>
Wed, 16 Mar 2016 15:22:45 +0000 (16:22 +0100)
committerIngo Molnar <mingo@kernel.org>
Mon, 21 Mar 2016 09:49:23 +0000 (10:49 +0100)
The CPU controller hasn't kept up with the various changes in the whole
cgroup initialization / destruction sequence, and commit:

  2e91fa7f6d45 ("cgroup: keep zombies associated with their original cgroups")

caused it to explode.

The reason for this is that zombies do not inhibit css_offline() from
being called, but do stall css_released(). Now we tear down the cfs_rq
structures on css_offline() but zombies can run after that, leading to
use-after-free issues.

The solution is to move the tear-down to css_released(), which
guarantees nobody (including no zombies) is still using our cgroup.

Furthermore, a few simple cleanups are possible too. There doesn't
appear to be any point to us using css_online() (anymore?) so fold that
in css_alloc().

And since cgroup code guarantees an RCU grace period between
css_released() and css_free() we can forgo using call_rcu() and free the
stuff immediately.

Suggested-by: Tejun Heo <tj@kernel.org>
Reported-by: Kazuki Yamaguchi <k@rhe.jp>
Reported-by: Niklas Cassel <niklas.cassel@axis.com>
Tested-by: Niklas Cassel <niklas.cassel@axis.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 2e91fa7f6d45 ("cgroup: keep zombies associated with their original cgroups")
Link: http://lkml.kernel.org/r/20160316152245.GY6344@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/core.c

index 4d872e19244b1a002ecf8271cae64825764c932b..2a87bdde8d4eb9c73afa9649b658479379c3117e 100644 (file)
@@ -7536,7 +7536,7 @@ void set_curr_task(int cpu, struct task_struct *p)
 /* task_group_lock serializes the addition/removal of task groups */
 static DEFINE_SPINLOCK(task_group_lock);
 
-static void free_sched_group(struct task_group *tg)
+static void sched_free_group(struct task_group *tg)
 {
        free_fair_sched_group(tg);
        free_rt_sched_group(tg);
@@ -7562,7 +7562,7 @@ struct task_group *sched_create_group(struct task_group *parent)
        return tg;
 
 err:
-       free_sched_group(tg);
+       sched_free_group(tg);
        return ERR_PTR(-ENOMEM);
 }
 
@@ -7582,17 +7582,16 @@ void sched_online_group(struct task_group *tg, struct task_group *parent)
 }
 
 /* rcu callback to free various structures associated with a task group */
-static void free_sched_group_rcu(struct rcu_head *rhp)
+static void sched_free_group_rcu(struct rcu_head *rhp)
 {
        /* now it should be safe to free those cfs_rqs */
-       free_sched_group(container_of(rhp, struct task_group, rcu));
+       sched_free_group(container_of(rhp, struct task_group, rcu));
 }
 
-/* Destroy runqueue etc associated with a task group */
 void sched_destroy_group(struct task_group *tg)
 {
        /* wait for possible concurrent references to cfs_rqs complete */
-       call_rcu(&tg->rcu, free_sched_group_rcu);
+       call_rcu(&tg->rcu, sched_free_group_rcu);
 }
 
 void sched_offline_group(struct task_group *tg)
@@ -8051,31 +8050,26 @@ cpu_cgroup_css_alloc(struct cgroup_subsys_state *parent_css)
        if (IS_ERR(tg))
                return ERR_PTR(-ENOMEM);
 
+       sched_online_group(tg, parent);
+
        return &tg->css;
 }
 
-static int cpu_cgroup_css_online(struct cgroup_subsys_state *css)
+static void cpu_cgroup_css_released(struct cgroup_subsys_state *css)
 {
        struct task_group *tg = css_tg(css);
-       struct task_group *parent = css_tg(css->parent);
 
-       if (parent)
-               sched_online_group(tg, parent);
-       return 0;
+       sched_offline_group(tg);
 }
 
 static void cpu_cgroup_css_free(struct cgroup_subsys_state *css)
 {
        struct task_group *tg = css_tg(css);
 
-       sched_destroy_group(tg);
-}
-
-static void cpu_cgroup_css_offline(struct cgroup_subsys_state *css)
-{
-       struct task_group *tg = css_tg(css);
-
-       sched_offline_group(tg);
+       /*
+        * Relies on the RCU grace period between css_released() and this.
+        */
+       sched_free_group(tg);
 }
 
 static void cpu_cgroup_fork(struct task_struct *task)
@@ -8435,9 +8429,8 @@ static struct cftype cpu_files[] = {
 
 struct cgroup_subsys cpu_cgrp_subsys = {
        .css_alloc      = cpu_cgroup_css_alloc,
+       .css_released   = cpu_cgroup_css_released,
        .css_free       = cpu_cgroup_css_free,
-       .css_online     = cpu_cgroup_css_online,
-       .css_offline    = cpu_cgroup_css_offline,
        .fork           = cpu_cgroup_fork,
        .can_attach     = cpu_cgroup_can_attach,
        .attach         = cpu_cgroup_attach,