cgroup_freezer: add ->post_create() and ->pre_destroy() and track online state
authorTejun Heo <tj@kernel.org>
Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)
committerTejun Heo <tj@kernel.org>
Fri, 9 Nov 2012 17:12:30 +0000 (09:12 -0800)
A cgroup is online and visible to iteration between ->post_create()
and ->pre_destroy().  This patch introduces CGROUP_FREEZER_ONLINE and
toggles it from the newly added freezer_post_create() and
freezer_pre_destroy() while holding freezer->lock such that a
cgroup_freezer can be reilably distinguished to be online.  This will
be used by full hierarchy support.

ONLINE test is added to freezer_apply_state() but it currently doesn't
make any difference as freezer_write() can only be called for an
online cgroup.

Adjusting system_freezing_cnt on destruction is moved from
freezer_destroy() to the new freezer_pre_destroy() for consistency.

This patch doesn't introduce any noticeable behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
kernel/cgroup_freezer.c

index b8ad93c6f5f6d7d48f2ec9e18ac383a0821450b5..4f12d317c4c37897bd3940dab0c59d5c10569239 100644 (file)
@@ -23,6 +23,7 @@
 #include <linux/seq_file.h>
 
 enum freezer_state_flags {
+       CGROUP_FREEZER_ONLINE   = (1 << 0), /* freezer is fully online */
        CGROUP_FREEZING_SELF    = (1 << 1), /* this freezer is freezing */
        CGROUP_FREEZING_PARENT  = (1 << 2), /* the parent freezer is freezing */
        CGROUP_FROZEN           = (1 << 3), /* this and its descendants frozen */
@@ -98,13 +99,45 @@ static struct cgroup_subsys_state *freezer_create(struct cgroup *cgroup)
        return &freezer->css;
 }
 
-static void freezer_destroy(struct cgroup *cgroup)
+/**
+ * freezer_post_create - commit creation of a freezer cgroup
+ * @cgroup: cgroup being created
+ *
+ * We're committing to creation of @cgroup.  Mark it online.
+ */
+static void freezer_post_create(struct cgroup *cgroup)
 {
        struct freezer *freezer = cgroup_freezer(cgroup);
 
+       spin_lock_irq(&freezer->lock);
+       freezer->state |= CGROUP_FREEZER_ONLINE;
+       spin_unlock_irq(&freezer->lock);
+}
+
+/**
+ * freezer_pre_destroy - initiate destruction of @cgroup
+ * @cgroup: cgroup being destroyed
+ *
+ * @cgroup is going away.  Mark it dead and decrement system_freezing_count
+ * if it was holding one.
+ */
+static void freezer_pre_destroy(struct cgroup *cgroup)
+{
+       struct freezer *freezer = cgroup_freezer(cgroup);
+
+       spin_lock_irq(&freezer->lock);
+
        if (freezer->state & CGROUP_FREEZING)
                atomic_dec(&system_freezing_cnt);
-       kfree(freezer);
+
+       freezer->state = 0;
+
+       spin_unlock_irq(&freezer->lock);
+}
+
+static void freezer_destroy(struct cgroup *cgroup)
+{
+       kfree(cgroup_freezer(cgroup));
 }
 
 /*
@@ -260,6 +293,9 @@ static void freezer_apply_state(struct freezer *freezer, bool freeze,
        /* also synchronizes against task migration, see freezer_attach() */
        lockdep_assert_held(&freezer->lock);
 
+       if (!(freezer->state & CGROUP_FREEZER_ONLINE))
+               return;
+
        if (freeze) {
                if (!(freezer->state & CGROUP_FREEZING))
                        atomic_inc(&system_freezing_cnt);
@@ -347,6 +383,8 @@ static struct cftype files[] = {
 struct cgroup_subsys freezer_subsys = {
        .name           = "freezer",
        .create         = freezer_create,
+       .post_create    = freezer_post_create,
+       .pre_destroy    = freezer_pre_destroy,
        .destroy        = freezer_destroy,
        .subsys_id      = freezer_subsys_id,
        .attach         = freezer_attach,