memcg: fix css_id() RCU locking for real
authorKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Tue, 11 May 2010 21:06:58 +0000 (14:06 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 12 May 2010 00:33:42 +0000 (17:33 -0700)
Commit ad4ba375373937817404fd92239ef4cadbded23b ("memcg: css_id() must be
called under rcu_read_lock()") modifies memcontol.c for fixing RCU check
message.  But Andrew Morton pointed out that the fix doesn't seems sane
and it was just for hidining lockdep messages.

This is a patch for do proper things.  Checking again, all places,
accessing without rcu_read_lock, that commit fixies was intentional....
all callers of css_id() has reference count on it.  So, it's not necessary
to be under rcu_read_lock().

Considering again, we can use rcu_dereference_check for css_id().  We know
css->id is valid if css->refcnt > 0.  (css->id never changes and freed
after css->refcnt going to be 0.)

This patch makes use of rcu_dereference_check() in css_id/depth and remove
unnecessary rcu-read-lock added by the commit.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/cgroup.c
mm/memcontrol.c

index 3a53c77..6db8b7f 100644 (file)
@@ -4435,7 +4435,15 @@ __setup("cgroup_disable=", cgroup_disable);
  */
 unsigned short css_id(struct cgroup_subsys_state *css)
 {
-       struct css_id *cssid = rcu_dereference(css->id);
+       struct css_id *cssid;
+
+       /*
+        * This css_id() can return correct value when somone has refcnt
+        * on this or this is under rcu_read_lock(). Once css->id is allocated,
+        * it's unchanged until freed.
+        */
+       cssid = rcu_dereference_check(css->id,
+                       rcu_read_lock_held() || atomic_read(&css->refcnt));
 
        if (cssid)
                return cssid->id;
@@ -4445,7 +4453,10 @@ EXPORT_SYMBOL_GPL(css_id);
 
 unsigned short css_depth(struct cgroup_subsys_state *css)
 {
-       struct css_id *cssid = rcu_dereference(css->id);
+       struct css_id *cssid;
+
+       cssid = rcu_dereference_check(css->id,
+                       rcu_read_lock_held() || atomic_read(&css->refcnt));
 
        if (cssid)
                return cssid->depth;
index 0f711c2..595d03f 100644 (file)
@@ -2314,9 +2314,7 @@ mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout)
 
        /* record memcg information */
        if (do_swap_account && swapout && memcg) {
-               rcu_read_lock();
                swap_cgroup_record(ent, css_id(&memcg->css));
-               rcu_read_unlock();
                mem_cgroup_get(memcg);
        }
        if (swapout && memcg)
@@ -2373,10 +2371,8 @@ static int mem_cgroup_move_swap_account(swp_entry_t entry,
 {
        unsigned short old_id, new_id;
 
-       rcu_read_lock();
        old_id = css_id(&from->css);
        new_id = css_id(&to->css);
-       rcu_read_unlock();
 
        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
                mem_cgroup_swap_statistics(from, false);
@@ -4044,16 +4040,11 @@ static int is_target_pte_for_mc(struct vm_area_struct *vma,
                        put_page(page);
        }
        /* throught */
-       if (ent.val && do_swap_account && !ret) {
-               unsigned short id;
-               rcu_read_lock();
-               id = css_id(&mc.from->css);
-               rcu_read_unlock();
-               if (id == lookup_swap_cgroup(ent)) {
-                       ret = MC_TARGET_SWAP;
-                       if (target)
-                               target->ent = ent;
-               }
+       if (ent.val && do_swap_account && !ret &&
+                       css_id(&mc.from->css) == lookup_swap_cgroup(ent)) {
+               ret = MC_TARGET_SWAP;
+               if (target)
+                       target->ent = ent;
        }
        return ret;
 }