net_sched: get rid of tcfa_rcu
authorCong Wang <xiyou.wangcong@gmail.com>
Mon, 11 Sep 2017 23:33:30 +0000 (16:33 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 13 Sep 2017 03:41:02 +0000 (20:41 -0700)
gen estimator has been rewritten in commit 1c0d32fde5bd
("net_sched: gen_estimator: complete rewrite of rate estimators"),
the caller is no longer needed to wait for a grace period.
So this patch gets rid of it.

This also completely closes a race condition between action free
path and filter chain add/remove path for the following patch.
Because otherwise the nested RCU callback can't be caught by
rcu_barrier().

Please see also the comments in code.

Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/act_api.h
net/sched/act_api.c

index 8f3d5d8b5ae0af88384730a7f5a5e6e3a50d7e03..b944e0eb93be367944673939a943b05c3473f2e9 100644 (file)
@@ -34,7 +34,6 @@ struct tc_action {
        struct gnet_stats_queue         tcfa_qstats;
        struct net_rate_estimator __rcu *tcfa_rate_est;
        spinlock_t                      tcfa_lock;
-       struct rcu_head                 tcfa_rcu;
        struct gnet_stats_basic_cpu __percpu *cpu_bstats;
        struct gnet_stats_queue __percpu *cpu_qstats;
        struct tc_cookie        *act_cookie;
@@ -50,7 +49,6 @@ struct tc_action {
 #define tcf_qstats     common.tcfa_qstats
 #define tcf_rate_est   common.tcfa_rate_est
 #define tcf_lock       common.tcfa_lock
-#define tcf_rcu                common.tcfa_rcu
 
 /* Update lastuse only if needed, to avoid dirtying a cache line.
  * We use a temp variable to avoid fetching jiffies twice.
index a306974e2fb4bfe866628d84b8c4aa7ae4fddb8c..fcd7dc7b807ad1d45cb671d989e72fa24a10921b 100644 (file)
@@ -53,10 +53,13 @@ static void tcf_action_goto_chain_exec(const struct tc_action *a,
        res->goto_tp = rcu_dereference_bh(chain->filter_chain);
 }
 
-static void free_tcf(struct rcu_head *head)
+/* XXX: For standalone actions, we don't need a RCU grace period either, because
+ * actions are always connected to filters and filters are already destroyed in
+ * RCU callbacks, so after a RCU grace period actions are already disconnected
+ * from filters. Readers later can not find us.
+ */
+static void free_tcf(struct tc_action *p)
 {
-       struct tc_action *p = container_of(head, struct tc_action, tcfa_rcu);
-
        free_percpu(p->cpu_bstats);
        free_percpu(p->cpu_qstats);
 
@@ -76,11 +79,7 @@ static void tcf_idr_remove(struct tcf_idrinfo *idrinfo, struct tc_action *p)
        idr_remove_ext(&idrinfo->action_idr, p->tcfa_index);
        spin_unlock_bh(&idrinfo->lock);
        gen_kill_estimator(&p->tcfa_rate_est);
-       /*
-        * gen_estimator est_timer() might access p->tcfa_lock
-        * or bstats, wait a RCU grace period before freeing p
-        */
-       call_rcu(&p->tcfa_rcu, free_tcf);
+       free_tcf(p);
 }
 
 int __tcf_idr_release(struct tc_action *p, bool bind, bool strict)
@@ -259,7 +258,7 @@ void tcf_idr_cleanup(struct tc_action *a, struct nlattr *est)
 {
        if (est)
                gen_kill_estimator(&a->tcfa_rate_est);
-       call_rcu(&a->tcfa_rcu, free_tcf);
+       free_tcf(a);
 }
 EXPORT_SYMBOL(tcf_idr_cleanup);