net: No dst refcounting in ip_queue_xmit()

TCP outgoing packets can avoid two atomic ops, and dirtying
of previously higly contended cache line using new refdst
infrastructure.

Note 1: loopback device excluded because of !IFF_XMIT_DST_RELEASE
Note 2: UDP packets dsts are built before ip_queue_xmit().

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

1 files changed, 7 insertions, 2 deletions

diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 252897443ef9..9a4a6c96cb0d 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -318,10 +318,12 @@ int ip_queue_xmit(struct sk_buff *skb)
         struct ip_options *opt = inet->opt;
         struct rtable *rt;
         struct iphdr *iph;
+        int res;
 
         /* Skip all of this if the packet is already routed,
          * f.e. by something like SCTP.
          */
+        rcu_read_lock();
         rt = skb_rtable(skb);
         if (rt != NULL)
                 goto packet_routed;
@@ -359,7 +361,7 @@ int ip_queue_xmit(struct sk_buff *skb)
                 }
                 sk_setup_caps(sk, &rt->u.dst);
         }
-        skb_dst_set(skb, dst_clone(&rt->u.dst));
+        skb_dst_set_noref(skb, &rt->u.dst);
 
 packet_routed:
         if (opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway)
@@ -391,9 +393,12 @@ packet_routed:
         skb->priority = sk->sk_priority;
         skb->mark = sk->sk_mark;
 
-        return ip_local_out(skb);
+        res = ip_local_out(skb);
+        rcu_read_unlock();
+        return res;
 
 no_route:
+        rcu_read_unlock();
         IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
         kfree_skb(skb);
         return -EHOSTUNREACH;