cgroup: remove cgroup_pidlist->use_count
authorTejun Heo <tj@kernel.org>
Fri, 29 Nov 2013 15:42:59 +0000 (10:42 -0500)
committerTejun Heo <tj@kernel.org>
Fri, 29 Nov 2013 15:42:59 +0000 (10:42 -0500)
After the recent changes, pidlist ref is held only between
cgroup_pidlist_start() and cgroup_pidlist_stop() during which
cgroup->pidlist_mutex is also held.  IOW, the reference count is
redundant now.  While in use, it's always one and pidlist_mutex is
held - holding the mutex has exactly the same protection.

This patch collapses destroy_dwork queueing into cgroup_pidlist_stop()
so that pidlist_mutex is not released inbetween and drops
pidlist->use_count.

This patch shouldn't introduce any behavior changes.

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

index 671cbde..a245803 100644 (file)
@@ -3456,8 +3456,6 @@ struct cgroup_pidlist {
        pid_t *list;
        /* how many elements the above list has */
        int length;
-       /* how many files are using the current array */
-       int use_count;
        /* each of these stored in a list by its cgroup */
        struct list_head links;
        /* pointer to the cgroup we belong to, for list removal purposes */
@@ -3473,8 +3471,6 @@ struct cgroup_pidlist_open_file {
        struct cgroup_pidlist           *pidlist;
 };
 
-static void cgroup_release_pid_array(struct cgroup_pidlist *l);
-
 /*
  * The following two functions "fix" the issue where there are more pids
  * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
@@ -3524,10 +3520,10 @@ static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
        mutex_lock(&l->owner->pidlist_mutex);
 
        /*
-        * Destroy iff we didn't race with a new user or get queued again.
-        * Queued state won't change as it can only be queued while locked.
+        * Destroy iff we didn't get queued again.  The state won't change
+        * as destroy_dwork can only be queued while locked.
         */
-       if (!l->use_count && !delayed_work_pending(dwork)) {
+       if (!delayed_work_pending(dwork)) {
                list_del(&l->links);
                pidlist_free(l->list);
                put_pid_ns(l->key.ns);
@@ -3775,7 +3771,6 @@ static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
                        return ERR_PTR(ret);
        }
        l = of->pidlist;
-       l->use_count++;
 
        if (pid) {
                int end = l->length;
@@ -3804,9 +3799,11 @@ static void cgroup_pidlist_stop(struct seq_file *s, void *v)
 {
        struct cgroup_pidlist_open_file *of = s->private;
 
-       mutex_unlock(&of->cgrp->pidlist_mutex);
        if (of->pidlist)
-               cgroup_release_pid_array(of->pidlist);
+               mod_delayed_work(cgroup_pidlist_destroy_wq,
+                                &of->pidlist->destroy_dwork,
+                                CGROUP_PIDLIST_DESTROY_DELAY);
+       mutex_unlock(&of->cgrp->pidlist_mutex);
 }
 
 static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
@@ -3844,17 +3841,6 @@ static const struct seq_operations cgroup_pidlist_seq_operations = {
        .show = cgroup_pidlist_show,
 };
 
-static void cgroup_release_pid_array(struct cgroup_pidlist *l)
-{
-       mutex_lock(&l->owner->pidlist_mutex);
-       BUG_ON(!l->use_count);
-       /* if the last user, arm the destroy work */
-       if (!--l->use_count)
-               mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
-                                CGROUP_PIDLIST_DESTROY_DELAY);
-       mutex_unlock(&l->owner->pidlist_mutex);
-}
-
 static const struct file_operations cgroup_pidlist_operations = {
        .read = seq_read,
        .llseek = seq_lseek,