ipv4: udp: Optimise multicast reception

UDP multicast rx path is a bit complex and can hold a spinlock
for a long time.

Using a small (32 or 64 entries) stack of socket pointers can help
to perform expensive operations (skb_clone(), udp_queue_rcv_skb())
outside of the lock, in most cases.

It's also a base for a future RCU conversion of multicast recption.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Lucian Adrian Grijincu <lgrijincu@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index dd7f3d2..9d9072c 100644 (file)
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1329,49 +1329,73 @@ drop:
        return -1;
 }
 
+
+static void flush_stack(struct sock **stack, unsigned int count,
+                       struct sk_buff *skb, unsigned int final)
+{
+       unsigned int i;
+       struct sk_buff *skb1 = NULL;
+
+       for (i = 0; i < count; i++) {
+               if (likely(skb1 == NULL))
+                       skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
+
+               if (skb1 && udp_queue_rcv_skb(stack[i], skb1) <= 0)
+                       skb1 = NULL;
+       }
+       if (unlikely(skb1))
+               kfree_skb(skb1);
+}
+
 /*
  *     Multicasts and broadcasts go to each listener.
  *
- *     Note: called only from the BH handler context,
- *     so we don't need to lock the hashes.
+ *     Note: called only from the BH handler context.
  */
 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
                                    struct udphdr  *uh,
                                    __be32 saddr, __be32 daddr,
                                    struct udp_table *udptable)
 {
-       struct sock *sk;
+       struct sock *sk, *stack[256 / sizeof(struct sock *)];
        struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
        int dif;
+       unsigned int i, count = 0;
 
        spin_lock(&hslot->lock);
        sk = sk_nulls_head(&hslot->head);
        dif = skb->dev->ifindex;
        sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
-       if (sk) {
-               struct sock *sknext = NULL;
-
-               do {
-                       struct sk_buff *skb1 = skb;
-
-                       sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
-                                                  daddr, uh->source, saddr,
-                                                  dif);
-                       if (sknext)
-                               skb1 = skb_clone(skb, GFP_ATOMIC);
-
-                       if (skb1) {
-                               int ret = udp_queue_rcv_skb(sk, skb1);
-                               if (ret > 0)
-                                       /* we should probably re-process instead
-                                        * of dropping packets here. */
-                                       kfree_skb(skb1);
-                       }
-                       sk = sknext;
-               } while (sknext);
-       } else
-               consume_skb(skb);
+       while (sk) {
+               stack[count++] = sk;
+               sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
+                                      daddr, uh->source, saddr, dif);
+               if (unlikely(count == ARRAY_SIZE(stack))) {
+                       if (!sk)
+                               break;
+                       flush_stack(stack, count, skb, ~0);
+                       count = 0;
+               }
+       }
+       /*
+        * before releasing chain lock, we must take a reference on sockets
+        */
+       for (i = 0; i < count; i++)
+               sock_hold(stack[i]);
+
        spin_unlock(&hslot->lock);
+
+       /*
+        * do the slow work with no lock held
+        */
+       if (count) {
+               flush_stack(stack, count, skb, count - 1);
+
+               for (i = 0; i < count; i++)
+                       sock_put(stack[i]);
+       } else {
+               kfree_skb(skb);
+       }
        return 0;
 }