ipv4: Restore old dst_free() behavior.

commit 404e0a8b6a55 (net: ipv4: fix RCU races on dst refcounts) tried
to solve a race but added a problem at device/fib dismantle time :

We really want to call dst_free() as soon as possible, even if sockets
still have dst in their cache.
dst_release() calls in free_fib_info_rcu() are not welcomed.

Root of the problem was that now we also cache output routes (in
nh_rth_output), we must use call_rcu() instead of call_rcu_bh() in
rt_free(), because output route lookups are done in process context.

Based on feedback and initial patch from David Miller (adding another
call_rcu_bh() call in fib, but it appears it was not the right fix)

I left the inet_sk_rx_dst_set() helper and added __rcu attributes
to nh_rth_output and nh_rth_input to better document what is going on in
this code.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 34 insertions, 13 deletions

diff --git a/net/ipv4/fib_semantics.c b/net/ipv4/fib_semantics.c
index e55171f184f..625cf185c48 100644
--- a/net/ipv4/fib_semantics.c
+++ b/net/ipv4/fib_semantics.c
@@ -161,6 +161,21 @@ static void free_nh_exceptions(struct fib_nh *nh)
 	kfree(hash);
 }
 
+static void rt_nexthop_free(struct rtable __rcu **rtp)
+{
+	struct rtable *rt = rcu_dereference_protected(*rtp, 1);
+
+	if (!rt)
+		return;
+
+	/* Not even needed : RCU_INIT_POINTER(*rtp, NULL);
+	 * because we waited an RCU grace period before calling
+	 * free_fib_info_rcu()
+	 */
+
+	dst_free(&rt->dst);
+}
+
 /* Release a nexthop info record */
 static void free_fib_info_rcu(struct rcu_head *head)
 {
@@ -171,10 +186,8 @@ static void free_fib_info_rcu(struct rcu_head *head)
 			dev_put(nexthop_nh->nh_dev);
 		if (nexthop_nh->nh_exceptions)
 			free_nh_exceptions(nexthop_nh);
-		if (nexthop_nh->nh_rth_output)
-			dst_release(&nexthop_nh->nh_rth_output->dst);
-		if (nexthop_nh->nh_rth_input)
-			dst_release(&nexthop_nh->nh_rth_input->dst);
+		rt_nexthop_free(&nexthop_nh->nh_rth_output);
+		rt_nexthop_free(&nexthop_nh->nh_rth_input);
 	} endfor_nexthops(fi);
 
 	release_net(fi->fib_net);
diff --git a/net/ipv4/route.c b/net/ipv4/route.c
index d6eabcfe8a9..2bd10747746 100644
--- a/net/ipv4/route.c
+++ b/net/ipv4/route.c
@@ -1199,23 +1199,31 @@ restart:
 	fnhe->fnhe_stamp = jiffies;
 }
 
+static inline void rt_free(struct rtable *rt)
+{
+	call_rcu(&rt->dst.rcu_head, dst_rcu_free);
+}
+
 static void rt_cache_route(struct fib_nh *nh, struct rtable *rt)
 {
-	struct rtable *orig, *prev, **p = &nh->nh_rth_output;
+	struct rtable *orig, *prev, **p = (struct rtable **)&nh->nh_rth_output;
 
 	if (rt_is_input_route(rt))
-		p = &nh->nh_rth_input;
+		p = (struct rtable **)&nh->nh_rth_input;
 
 	orig = *p;
 
-	rt->dst.flags |= DST_RCU_FREE;
-	dst_hold(&rt->dst);
 	prev = cmpxchg(p, orig, rt);
 	if (prev == orig) {
 		if (orig)
-			dst_release(&orig->dst);
+			rt_free(orig);
 	} else {
-		dst_release(&rt->dst);
+		/* Routes we intend to cache in the FIB nexthop have
+		 * the DST_NOCACHE bit clear.  However, if we are
+		 * unsuccessful at storing this route into the cache
+		 * we really need to set it.
+		 */
+		rt->dst.flags |= DST_NOCACHE;
 	}
 }
 
@@ -1412,7 +1420,7 @@ static int __mkroute_input(struct sk_buff *skb,
 	do_cache = false;
 	if (res->fi) {
 		if (!itag) {
-			rth = FIB_RES_NH(*res).nh_rth_input;
+			rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_input);
 			if (rt_cache_valid(rth)) {
 				skb_dst_set_noref(skb, &rth->dst);
 				goto out;
@@ -1574,7 +1582,7 @@ local_input:
 	do_cache = false;
 	if (res.fi) {
 		if (!itag) {
-			rth = FIB_RES_NH(res).nh_rth_input;
+			rth = rcu_dereference(FIB_RES_NH(res).nh_rth_input);
 			if (rt_cache_valid(rth)) {
 				skb_dst_set_noref(skb, &rth->dst);
 				err = 0;
@@ -1742,7 +1750,7 @@ static struct rtable *__mkroute_output(const struct fib_result *res,
 	if (fi) {
 		fnhe = find_exception(&FIB_RES_NH(*res), fl4->daddr);
 		if (!fnhe) {
-			rth = FIB_RES_NH(*res).nh_rth_output;
+			rth = rcu_dereference(FIB_RES_NH(*res).nh_rth_output);
 			if (rt_cache_valid(rth)) {
 				dst_hold(&rth->dst);
 				return rth;