[PATCH net-next 0/2] ipvlan: multicast delivery changes

[PATCH net-next 0/2] ipvlan: multicast delivery changes

[Eric Dumazet]

As we did recently for macvlan, this series adds some relief
when ipvlan is under multicast storms.

Eric Dumazet (2):
  ipvlan: ipvlan_handle_mode_l2() refactoring
  ipvlan: avoid spinlock contention in ipvlan_multicast_enqueue()

 drivers/net/ipvlan/ipvlan_core.c | 42 +++++++++++++++++---------------
 1 file changed, 23 insertions(+), 19 deletions(-)

-- 
2.53.0.1213.gd9a14994de-goog

[PATCH net-next 1/2] ipvlan: ipvlan_handle_mode_l2() refactoring

[Eric Dumazet]

Reduce indentation level, and add a likely() clause
as we expect to process more unicast packets than multicast ones.

No functional change, this eases the following patch review.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 drivers/net/ipvlan/ipvlan_core.c | 38 +++++++++++++++-----------------
 1 file changed, 18 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ipvlan/ipvlan_core.c b/drivers/net/ipvlan/ipvlan_core.c
index 0b493a8aa33857d531329e8eaef6b25a5c6f572d..214fd299a5aa6e40579aa2dbcb178b5474b561a4 100644
--- a/drivers/net/ipvlan/ipvlan_core.c
+++ b/drivers/net/ipvlan/ipvlan_core.c
@@ -744,34 +744,32 @@ static rx_handler_result_t ipvlan_handle_mode_l3(struct sk_buff **pskb,
 static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 						 struct ipvl_port *port)
 {
-	struct sk_buff *skb = *pskb;
+	struct sk_buff *nskb, *skb = *pskb;
 	struct ethhdr *eth = eth_hdr(skb);
-	rx_handler_result_t ret = RX_HANDLER_PASS;
 
 	if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
 		return RX_HANDLER_PASS;
 
-	if (is_multicast_ether_addr(eth->h_dest)) {
-		if (ipvlan_external_frame(skb, port)) {
-			struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
+	/* Perform like l3 mode for non-multicast packet */
+	if (likely(!is_multicast_ether_addr(eth->h_dest)))
+		return ipvlan_handle_mode_l3(pskb, port);
 
-			/* External frames are queued for device local
-			 * distribution, but a copy is given to master
-			 * straight away to avoid sending duplicates later
-			 * when work-queue processes this frame. This is
-			 * achieved by returning RX_HANDLER_PASS.
-			 */
-			if (nskb) {
-				ipvlan_skb_crossing_ns(nskb, NULL);
-				ipvlan_multicast_enqueue(port, nskb, false);
-			}
-		}
-	} else {
-		/* Perform like l3 mode for non-multicast packet */
-		ret = ipvlan_handle_mode_l3(pskb, port);
+	/* External frames are queued for device local
+	 * distribution, but a copy is given to master
+	 * straight away to avoid sending duplicates later
+	 * when work-queue processes this frame.
+	 * This is achieved by returning RX_HANDLER_PASS.
+	 */
+	if (!ipvlan_external_frame(skb, port))
+		return RX_HANDLER_PASS;
+
+	nskb = skb_clone(skb, GFP_ATOMIC);
+	if (nskb) {
+		ipvlan_skb_crossing_ns(nskb, NULL);
+		ipvlan_multicast_enqueue(port, nskb, false);
 	}
 
-	return ret;
+	return RX_HANDLER_PASS;
 }
 
 rx_handler_result_t ipvlan_handle_frame(struct sk_buff **pskb)
-- 
2.53.0.1213.gd9a14994de-goog

[PATCH net-next 2/2] ipvlan: avoid spinlock contention in ipvlan_multicast_enqueue()

[Eric Dumazet]

Under high stress, we spend a lot of time cloning skbs,
then acquiring a spinlock, then freeing the clone because
the queue is full.

Add a shortcut to avoid these costs under pressure, as we did
in macvlan with commit 0d5dc1d7aad1 ("macvlan: avoid spinlock
contention in macvlan_broadcast_enqueue()")

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 drivers/net/ipvlan/ipvlan_core.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/net/ipvlan/ipvlan_core.c b/drivers/net/ipvlan/ipvlan_core.c
index 214fd299a5aa6e40579aa2dbcb178b5474b561a4..1be8620ad3971d281fb36fd0770efd67b566ae60 100644
--- a/drivers/net/ipvlan/ipvlan_core.c
+++ b/drivers/net/ipvlan/ipvlan_core.c
@@ -763,10 +763,16 @@ static rx_handler_result_t ipvlan_handle_mode_l2(struct sk_buff **pskb,
 	if (!ipvlan_external_frame(skb, port))
 		return RX_HANDLER_PASS;
 
-	nskb = skb_clone(skb, GFP_ATOMIC);
+	if (skb_queue_len_lockless(&port->backlog) >= IPVLAN_QBACKLOG_LIMIT)
+		nskb = NULL;
+	else
+		nskb = skb_clone(skb, GFP_ATOMIC);
+
 	if (nskb) {
 		ipvlan_skb_crossing_ns(nskb, NULL);
 		ipvlan_multicast_enqueue(port, nskb, false);
+	} else {
+		dev_core_stats_rx_dropped_inc(skb->dev);
 	}
 
 	return RX_HANDLER_PASS;
-- 
2.53.0.1213.gd9a14994de-goog