ipv6: tcp: send consistent autoflowlabel in SYN_RECV state
authorEric Dumazet <edumazet@google.com>
Wed, 31 Aug 2022 20:37:29 +0000 (13:37 -0700)
committerJakub Kicinski <kuba@kernel.org>
Fri, 2 Sep 2022 03:57:03 +0000 (20:57 -0700)
This is a followup of commit c67b85558ff2 ("ipv6: tcp: send consistent
autoflowlabel in TIME_WAIT state"), but for SYN_RECV state.

In some cases, TCP sends a challenge ACK on behalf of a SYN_RECV request.
WHen this happens, we want to use the flow label that was used when
the prior SYNACK packet was sent, instead of another one.

After his patch, following packetdrill passes:

    0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
   +0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
   +0 bind(3, ..., ...) = 0
   +0 listen(3, 1) = 0

  +.2 < S 0:0(0) win 32792 <mss 1000,sackOK,nop,nop,nop,wscale 7>
   +0 > (flowlabel 0x11) S. 0:0(0) ack 1 <...>
// Test if a challenge ack is properly sent (same flowlabel than prior SYNACK)
   +.01 < . 4000000000:4000000000(0) ack 1 win 320
   +0  > (flowlabel 0x11) . 1:1(0) ack 1

Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://lore.kernel.org/r/20220831203729.458000-1-eric.dumazet@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/ipv6/tcp_ipv6.c

index ff5c4fc..3501349 100644 (file)
@@ -858,7 +858,7 @@ const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
                                 u32 ack, u32 win, u32 tsval, u32 tsecr,
                                 int oif, struct tcp_md5sig_key *key, int rst,
-                                u8 tclass, __be32 label, u32 priority)
+                                u8 tclass, __be32 label, u32 priority, u32 txhash)
 {
        const struct tcphdr *th = tcp_hdr(skb);
        struct tcphdr *t1;
@@ -949,16 +949,16 @@ static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32
        }
 
        if (sk) {
-               if (sk->sk_state == TCP_TIME_WAIT) {
+               if (sk->sk_state == TCP_TIME_WAIT)
                        mark = inet_twsk(sk)->tw_mark;
-                       /* autoflowlabel relies on buff->hash */
-                       skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
-                                    PKT_HASH_TYPE_L4);
-               } else {
+               else
                        mark = sk->sk_mark;
-               }
                skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
        }
+       if (txhash) {
+               /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
+               skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
+       }
        fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
        fl6.fl6_dport = t1->dest;
        fl6.fl6_sport = t1->source;
@@ -1085,7 +1085,7 @@ static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
        }
 
        tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
-                            ipv6_get_dsfield(ipv6h), label, priority);
+                            ipv6_get_dsfield(ipv6h), label, priority, 0);
 
 #ifdef CONFIG_TCP_MD5SIG
 out:
@@ -1096,10 +1096,10 @@ out:
 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
                            u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
                            struct tcp_md5sig_key *key, u8 tclass,
-                           __be32 label, u32 priority)
+                           __be32 label, u32 priority, u32 txhash)
 {
        tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
-                            tclass, label, priority);
+                            tclass, label, priority, txhash);
 }
 
 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
@@ -1111,7 +1111,8 @@ static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
                        tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
                        tcp_time_stamp_raw() + tcptw->tw_ts_offset,
                        tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
-                       tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
+                       tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
+                       tw->tw_txhash);
 
        inet_twsk_put(tw);
 }
@@ -1138,7 +1139,8 @@ static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
                        tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
                        req->ts_recent, sk->sk_bound_dev_if,
                        tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
-                       ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority);
+                       ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority,
+                       tcp_rsk(req)->txhash);
 }