tcp/dccp: do not touch listener sk_refcnt under synflood
When a SYNFLOOD targets a non SO_REUSEPORT listener, multiple
cpus contend on sk->sk_refcnt and sk->sk_wmem_alloc changes.
By letting listeners use SOCK_RCU_FREE infrastructure,
we can relax TCP_LISTEN lookup rules and avoid touching sk_refcnt
Note that we still use SLAB_DESTROY_BY_RCU rules for other sockets,
only listeners are impacted by this change.
Peak performance under SYNFLOOD is increased by ~33% :
On my test machine, I could process 3.2 Mpps instead of 2.4 Mpps
Most consuming functions are now skb_set_owner_w() and sock_wfree()
contending on sk->sk_wmem_alloc when cooking SYNACK and freeing them.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/dccp/ipv4.c b/net/dccp/ipv4.c
index 9c67a96..6438c5a 100644
--- a/net/dccp/ipv4.c
+++ b/net/dccp/ipv4.c
@@ -764,6 +764,7 @@
{
const struct dccp_hdr *dh;
const struct iphdr *iph;
+ bool refcounted;
struct sock *sk;
int min_cov;
@@ -801,7 +802,7 @@
lookup:
sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
- dh->dccph_sport, dh->dccph_dport);
+ dh->dccph_sport, dh->dccph_dport, &refcounted);
if (!sk) {
dccp_pr_debug("failed to look up flow ID in table and "
"get corresponding socket\n");
@@ -830,6 +831,7 @@
goto lookup;
}
sock_hold(sk);
+ refcounted = true;
nsk = dccp_check_req(sk, skb, req);
if (!nsk) {
reqsk_put(req);
@@ -886,7 +888,8 @@
return 0;
discard_and_relse:
- sock_put(sk);
+ if (refcounted)
+ sock_put(sk);
goto discard_it;
}
