Merge branch 'net-sched-race-fix'

Cong Wang says:

====================
net_sched: close the race between call_rcu() and cleanup_net()

This patchset tries to fix the race between call_rcu() and
cleanup_net() again. Without holding the netns refcnt the
tc_action_net_exit() in netns workqueue could be called before
filter destroy works in tc filter workqueue. This patchset
moves the netns refcnt from tc actions to tcf_exts, without
breaking per-netns tc actions.

Patch 1 reverts the previous fix, patch 2 introduces two new
API's to help to address the bug and the rest patches switch
to the new API's. Please see each patch for details.

I was not able to reproduce this bug, but now after adding
some delay in filter destroy work I manage to trigger the
crash. After this patchset, the crash is not reproducible
any more and the debugging printk's show the order is expected
too.
====================

Fixes: ddf97ccdd7cb ("net_sched: add network namespace support for tc actions")
Reported-by: Lucas Bates <lucasb@mojatatu.com>
Cc: Lucas Bates <lucasb@mojatatu.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 18 insertions, 6 deletions

diff --git a/net/sched/cls_flow.c b/net/sched/cls_flow.c
index 67f3a2af6aab..85f765cff697 100644
--- a/net/sched/cls_flow.c
+++ b/net/sched/cls_flow.c
@@ -372,15 +372,21 @@ static const struct nla_policy flow_policy[TCA_FLOW_MAX + 1] = {
 	[TCA_FLOW_PERTURB]	= { .type = NLA_U32 },
 };
 
-static void flow_destroy_filter_work(struct work_struct *work)
+static void __flow_destroy_filter(struct flow_filter *f)
 {
-	struct flow_filter *f = container_of(work, struct flow_filter, work);
-
-	rtnl_lock();
 	del_timer_sync(&f->perturb_timer);
 	tcf_exts_destroy(&f->exts);
 	tcf_em_tree_destroy(&f->ematches);
+	tcf_exts_put_net(&f->exts);
 	kfree(f);
+}
+
+static void flow_destroy_filter_work(struct work_struct *work)
+{
+	struct flow_filter *f = container_of(work, struct flow_filter, work);
+
+	rtnl_lock();
+	__flow_destroy_filter(f);
 	rtnl_unlock();
 }
 
@@ -552,8 +558,10 @@ static int flow_change(struct net *net, struct sk_buff *in_skb,
 
 	*arg = fnew;
 
-	if (fold)
+	if (fold) {
+		tcf_exts_get_net(&fold->exts);
 		call_rcu(&fold->rcu, flow_destroy_filter);
+	}
 	return 0;
 
 err2:
@@ -570,6 +578,7 @@ static int flow_delete(struct tcf_proto *tp, void *arg, bool *last)
 	struct flow_filter *f = arg;
 
 	list_del_rcu(&f->list);
+	tcf_exts_get_net(&f->exts);
 	call_rcu(&f->rcu, flow_destroy_filter);
 	*last = list_empty(&head->filters);
 	return 0;
@@ -594,7 +603,10 @@ static void flow_destroy(struct tcf_proto *tp)
 
 	list_for_each_entry_safe(f, next, &head->filters, list) {
 		list_del_rcu(&f->list);
-		call_rcu(&f->rcu, flow_destroy_filter);
+		if (tcf_exts_get_net(&f->exts))
+			call_rcu(&f->rcu, flow_destroy_filter);
+		else
+			__flow_destroy_filter(f);
 	}
 	kfree_rcu(head, rcu);
 }