tcp/dccp: install syn_recv requests into ehash table
In this patch, we insert request sockets into TCP/DCCP
regular ehash table (where ESTABLISHED and TIMEWAIT sockets
are) instead of using the per listener hash table.
ACK packets find SYN_RECV pseudo sockets without having
to find and lock the listener.
In nominal conditions, this halves pressure on listener lock.
Note that this will allow for SO_REUSEPORT refinements,
so that we can select a listener using cpu/numa affinities instead
of the prior 'consistent hash', since only SYN packets will
apply this selection logic.
We will shrink listen_sock in the following patch to ease
code review.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Ying Cai <ycai@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/net/dccp/ipv6.c b/net/dccp/ipv6.c
index e8753aa..1361a3f 100644
--- a/net/dccp/ipv6.c
+++ b/net/dccp/ipv6.c
@@ -290,37 +290,6 @@
.syn_ack_timeout = dccp_syn_ack_timeout,
};
-static struct sock *dccp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
-{
- const struct dccp_hdr *dh = dccp_hdr(skb);
- const struct ipv6hdr *iph = ipv6_hdr(skb);
- struct request_sock *req;
- struct sock *nsk;
-
- req = inet6_csk_search_req(sk, dh->dccph_sport, &iph->saddr,
- &iph->daddr, inet6_iif(skb));
- if (req) {
- nsk = dccp_check_req(sk, skb, req);
- if (!nsk)
- reqsk_put(req);
- return nsk;
- }
- nsk = __inet6_lookup_established(sock_net(sk), &dccp_hashinfo,
- &iph->saddr, dh->dccph_sport,
- &iph->daddr, ntohs(dh->dccph_dport),
- inet6_iif(skb));
- if (nsk != NULL) {
- if (nsk->sk_state != DCCP_TIME_WAIT) {
- bh_lock_sock(nsk);
- return nsk;
- }
- inet_twsk_put(inet_twsk(nsk));
- return NULL;
- }
-
- return sk;
-}
-
static int dccp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
{
struct request_sock *req;
@@ -398,7 +367,7 @@
if (dccp_v6_send_response(sk, req))
goto drop_and_free;
- inet6_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
+ inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
return 0;
drop_and_free:
@@ -641,24 +610,6 @@
* NOTE: the check for the packet types is done in
* dccp_rcv_state_process
*/
- if (sk->sk_state == DCCP_LISTEN) {
- struct sock *nsk = dccp_v6_hnd_req(sk, skb);
-
- if (nsk == NULL)
- goto discard;
- /*
- * Queue it on the new socket if the new socket is active,
- * otherwise we just shortcircuit this and continue with
- * the new socket..
- */
- if (nsk != sk) {
- if (dccp_child_process(sk, nsk, skb))
- goto reset;
- if (opt_skb != NULL)
- __kfree_skb(opt_skb);
- return 0;
- }
- }
if (dccp_rcv_state_process(sk, skb, dccp_hdr(skb), skb->len))
goto reset;
@@ -732,6 +683,27 @@
goto no_dccp_socket;
}
+ if (sk->sk_state == DCCP_NEW_SYN_RECV) {
+ struct request_sock *req = inet_reqsk(sk);
+ struct sock *nsk = NULL;
+
+ sk = req->rsk_listener;
+ if (sk->sk_state == DCCP_LISTEN)
+ nsk = dccp_check_req(sk, skb, req);
+ if (!nsk) {
+ reqsk_put(req);
+ goto discard_it;
+ }
+ if (nsk == sk) {
+ sock_hold(sk);
+ reqsk_put(req);
+ } else if (dccp_child_process(sk, nsk, skb)) {
+ dccp_v6_ctl_send_reset(sk, skb);
+ goto discard_it;
+ } else {
+ return 0;
+ }
+ }
/*
* RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
* o if MinCsCov = 0, only packets with CsCov = 0 are accepted