net: rfs: add hash collision detection

Receive Flow Steering is a nice solution but suffers from
hash collisions when a mix of connected and unconnected traffic
is received on the host, when flow hash table is populated.

Also, clearing flow in inet_release() makes RFS not very good
for short lived flows, as many packets can follow close().
(FIN , ACK packets, ...)

This patch extends the information stored into global hash table
to not only include cpu number, but upper part of the hash value.

I use a 32bit value, and dynamically split it in two parts.

For host with less than 64 possible cpus, this gives 6 bits for the
cpu number, and 26 (32-6) bits for the upper part of the hash.

Since hash bucket selection use low order bits of the hash, we have
a full hash match, if /proc/sys/net/core/rps_sock_flow_entries is big
enough.

If the hash found in flow table does not match, we fallback to RPS (if
it is enabled for the rxqueue).

This means that a packet for an non connected flow can avoid the
IPI through a unrelated/victim CPU.

This also means we no longer have to clear the table at socket
close time, and this helps short lived flows performance.

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

3 files changed, 28 insertions, 24 deletions

diff --git a/net/core/dev.c b/net/core/dev.c
index a3a96ffc67f4..8be38675e1a8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3030,6 +3030,8 @@ static inline void ____napi_schedule(struct softnet_data *sd,
 /* One global table that all flow-based protocols share. */
 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
 EXPORT_SYMBOL(rps_sock_flow_table);
+u32 rps_cpu_mask __read_mostly;
+EXPORT_SYMBOL(rps_cpu_mask);
 
 struct static_key rps_needed __read_mostly;
 
@@ -3086,16 +3088,17 @@ set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		       struct rps_dev_flow **rflowp)
 {
-	struct netdev_rx_queue *rxqueue;
-	struct rps_map *map;
+	const struct rps_sock_flow_table *sock_flow_table;
+	struct netdev_rx_queue *rxqueue = dev->_rx;
 	struct rps_dev_flow_table *flow_table;
-	struct rps_sock_flow_table *sock_flow_table;
+	struct rps_map *map;
 	int cpu = -1;
-	u16 tcpu;
+	u32 tcpu;
 	u32 hash;
 
 	if (skb_rx_queue_recorded(skb)) {
 		u16 index = skb_get_rx_queue(skb);
+
 		if (unlikely(index >= dev->real_num_rx_queues)) {
 			WARN_ONCE(dev->real_num_rx_queues > 1,
 				  "%s received packet on queue %u, but number "
@@ -3103,39 +3106,40 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 				  dev->name, index, dev->real_num_rx_queues);
 			goto done;
 		}
-		rxqueue = dev->_rx + index;
-	} else
-		rxqueue = dev->_rx;
+		rxqueue += index;
+	}
 
+	/* Avoid computing hash if RFS/RPS is not active for this rxqueue */
+
+	flow_table = rcu_dereference(rxqueue->rps_flow_table);
 	map = rcu_dereference(rxqueue->rps_map);
-	if (map) {
-		if (map->len == 1 &&
-		    !rcu_access_pointer(rxqueue->rps_flow_table)) {
-			tcpu = map->cpus[0];
-			if (cpu_online(tcpu))
-				cpu = tcpu;
-			goto done;
-		}
-	} else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
+	if (!flow_table && !map)
 		goto done;
-	}
 
 	skb_reset_network_header(skb);
 	hash = skb_get_hash(skb);
 	if (!hash)
 		goto done;
 
-	flow_table = rcu_dereference(rxqueue->rps_flow_table);
 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
 	if (flow_table && sock_flow_table) {
-		u16 next_cpu;
 		struct rps_dev_flow *rflow;
+		u32 next_cpu;
+		u32 ident;
+
+		/* First check into global flow table if there is a match */
+		ident = sock_flow_table->ents[hash & sock_flow_table->mask];
+		if ((ident ^ hash) & ~rps_cpu_mask)
+			goto try_rps;
 
+		next_cpu = ident & rps_cpu_mask;
+
+		/* OK, now we know there is a match,
+		 * we can look at the local (per receive queue) flow table
+		 */
 		rflow = &flow_table->flows[hash & flow_table->mask];
 		tcpu = rflow->cpu;
 
-		next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
-
 		/*
 		 * If the desired CPU (where last recvmsg was done) is
 		 * different from current CPU (one in the rx-queue flow
@@ -3162,6 +3166,8 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
 		}
 	}
 
+try_rps:
+
 	if (map) {
 		tcpu = map->cpus[reciprocal_scale(hash, map->len)];
 		if (cpu_online(tcpu)) {
diff --git a/net/core/sysctl_net_core.c b/net/core/sysctl_net_core.c
index fde21d19e61b..7a31be5e361f 100644
--- a/net/core/sysctl_net_core.c
+++ b/net/core/sysctl_net_core.c
@@ -65,7 +65,7 @@ static int rps_sock_flow_sysctl(struct ctl_table *table, int write,
 					mutex_unlock(&sock_flow_mutex);
 					return -ENOMEM;
 				}
-
+				rps_cpu_mask = roundup_pow_of_two(nr_cpu_ids) - 1;
 				sock_table->mask = size - 1;
 			} else
 				sock_table = orig_sock_table;
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index a44773c8346c..d2e49baaff63 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -395,8 +395,6 @@ int inet_release(struct socket *sock)
 	if (sk) {
 		long timeout;
 
-		sock_rps_reset_flow(sk);
-
 		/* Applications forget to leave groups before exiting */
 		ip_mc_drop_socket(sk);