I noted that DCCP apparently does not update the neighbour tables in the same
way as TCP (or UDP with MSG_CONFIRM).
This patch adds the missing calls to dst_update(). I would appreciate review
and comments in case I missed something.
DCCP does not have the equivalent of tcp_init_metrics()/tcp_update_metrics().
It seems that this functionality would be more in the CCIDs than in the main
DCCP module (for instance, CCID-3 does not understand ssthresh/cwnd and uses
a different RTT sampling algorithm).
This patch is meant for RFC, not currently for submission. It is in the DCCP
test tree, on
git://eden-feed.erg.abdn.ac.uk/dccp_exp [subtree 'dccp']
>>>>>>>>>>>>>>>>>>>>>>>>> Patch <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
dccp: indicate forward progress of connection to lower layer
This patch implements three cases of indicating forward progress:
1. Peer acknowledging receipt of connection initiation.
This corresponds to the cases where tcp_init_metrics() is called:
* receiving Syn-Ack in response to Syn (tcp_rcv_synsent_state_process),
* receiving Ack to finish the handshake (TCP_SYN_RECV -> TCP_ESTABLISHED);
which in DCCP corresponds to
* receiving Response after Request (dccp_rcv_request_sent_state_process),
* moving from RESPOND/PARTOPEN to OPEN when receiving an Ack/DataAck
(RESPOND) or non-Sync/Reset (PARTOPEN).
The latter does no extra rebuild_header() as in tcp_rcv_state_process(),
since in DCCP an active (client) socket can not become a passive one later,
and since dccp_v{4,6}_request_recv_sock() already talk to the routing
system.
2. Peer advancing GAR (greatest acknowledgment number received).
This is comparable to making forward progress in tcp_ack().
3. Peer acknowledging request to terminate the connection.
This corresponds to TCP teardown -
* receiving Ack-of-Fin in FIN_WAIT1/LAST_ACK (tcp_rcv_state_process),
* entering TIME_WAIT (tcp_time_wait),
and is realized in DCCP when receiving
* a Reset packet with code "Closed" (in response to DCCP-Close),
* a Close packet after sending a CloseReq,
* a Close packet after sending a Close (simultaneous close).
Signed-off-by: Gerrit Renker <gerrit@xxxxxxxxxxxxxx>
---
net/dccp/input.c | 11 ++++++++++-
1 file changed, 10 insertions(+), 1 deletion(-)
--- a/net/dccp/input.c
+++ b/net/dccp/input.c
@@ -71,6 +71,7 @@ static int dccp_rcv_close(struct sock *s
/* fall through */
case DCCP_REQUESTING:
case DCCP_ACTIVE_CLOSEREQ:
+ dst_confirm(__sk_dst_get(sk));
dccp_send_reset(sk, DCCP_RESET_CODE_CLOSED);
dccp_done(sk);
break;
@@ -155,6 +156,8 @@ static void dccp_rcv_reset(struct sock *
/* Queue the equivalent of TCP fin so that dccp_recvmsg exits the loop */
dccp_fin(sk, skb);
+ if (dccp_hdr_reset(skb)->dccph_reset_code == DCCP_RESET_CODE_CLOSED)
+ dst_confirm(__sk_dst_get(sk));
if (err && !sock_flag(sk, SOCK_DEAD))
sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
dccp_time_wait(sk, DCCP_TIME_WAIT, 0);
@@ -242,8 +245,10 @@ static int dccp_check_seqno(struct sock
if (dh->dccph_type != DCCP_PKT_SYNC &&
ackno != DCCP_PKT_WITHOUT_ACK_SEQ &&
- after48(ackno, dp->dccps_gar))
+ after48(ackno, dp->dccps_gar)) {
+ dst_confirm(__sk_dst_get(sk));
dp->dccps_gar = ackno;
+ }
} else {
unsigned long now = jiffies;
/*
@@ -475,6 +480,8 @@ static int dccp_rcv_request_sent_state_p
/* Make sure socket is routed, for correct metrics. */
icsk->icsk_af_ops->rebuild_header(sk);
+ dst_confirm(__sk_dst_get(sk));
+
if (!sock_flag(sk, SOCK_DEAD)) {
sk->sk_state_change(sk);
sk_wake_async(sk, SOCK_WAKE_IO, POLL_OUT);
@@ -556,6 +563,8 @@ static int dccp_rcv_respond_partopen_sta
dp->dccps_syn_rtt = dccp_sample_rtt(sk, 10 * delta);
}
+ dst_confirm(__sk_dst_get(sk));
+
dp->dccps_osr = DCCP_SKB_CB(skb)->dccpd_seq;
dccp_set_state(sk, DCCP_OPEN);
--
To unsubscribe from this list: send the line "unsubscribe dccp" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
