cgroup: move css_set_rwsem locking outside of cgroup_task_migrate()
authorTejun Heo <tj@kernel.org>
Thu, 13 Feb 2014 11:58:41 +0000 (06:58 -0500)
committerTejun Heo <tj@kernel.org>
Thu, 13 Feb 2014 11:58:41 +0000 (06:58 -0500)
Instead of repeatedly locking and unlocking css_set_rwsem inside
cgroup_task_migrate(), update cgroup_attach_task() to grab it outside
of the loop and update cgroup_task_migrate() to use
put_css_set_locked().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Li Zefan <lizefan@huawei.com>
kernel/cgroup.c

index 17b10b8efbcf563d290a7d27bdf6706fc9d469a4..704c590a81d767af579c27758fb378cae18eaf46 100644 (file)
@@ -1715,10 +1715,13 @@ int cgroup_taskset_size(struct cgroup_taskset *tset)
 EXPORT_SYMBOL_GPL(cgroup_taskset_size);
 
 
-/*
+/**
  * cgroup_task_migrate - move a task from one cgroup to another.
+ * @old_cgrp; the cgroup @tsk is being migrated from
+ * @tsk: the task being migrated
+ * @new_cset: the new css_set @tsk is being attached to
  *
- * Must be called with cgroup_mutex and threadgroup locked.
+ * Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
  */
 static void cgroup_task_migrate(struct cgroup *old_cgrp,
                                struct task_struct *tsk,
@@ -1726,6 +1729,9 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
 {
        struct css_set *old_cset;
 
+       lockdep_assert_held(&cgroup_mutex);
+       lockdep_assert_held(&css_set_rwsem);
+
        /*
         * We are synchronized through threadgroup_lock() against PF_EXITING
         * setting such that we can't race against cgroup_exit() changing the
@@ -1738,9 +1744,7 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
        rcu_assign_pointer(tsk->cgroups, new_cset);
        task_unlock(tsk);
 
-       down_write(&css_set_rwsem);
        list_move(&tsk->cg_list, &new_cset->tasks);
-       up_write(&css_set_rwsem);
 
        /*
         * We just gained a reference on old_cset by taking it from the
@@ -1748,7 +1752,7 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
         * we're safe to drop it here; it will be freed under RCU.
         */
        set_bit(CGRP_RELEASABLE, &old_cgrp->flags);
-       put_css_set(old_cset, false);
+       put_css_set_locked(old_cset, false);
 }
 
 /**
@@ -1871,10 +1875,12 @@ static int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk,
         * proceed to move all tasks to the new cgroup.  There are no
         * failure cases after here, so this is the commit point.
         */
+       down_write(&css_set_rwsem);
        for (i = 0; i < group_size; i++) {
                tc = flex_array_get(group, i);
                cgroup_task_migrate(tc->cgrp, tc->task, tc->cset);
        }
+       up_write(&css_set_rwsem);
        /* nothing is sensitive to fork() after this point. */
 
        /*