The current code changes txhash (flowlables) on every retransmitted
SYN/ACK, but only after the 2nd retransmitted SYN and only after
tcp_retries1 RTO retransmits.
With this patch:
1) txhash is changed with every SYN retransmits
2) txhash is changed with every RTO.
The result is that we can start re-routing around failed (or very
congested paths) as soon as possible. Otherwise application health
checks may fail and the connection may be terminated before we start
to change txhash.
v4: Removed sysctl, txhash is changed for all RTOs
v3: Removed text saying default value of sysctl is 0 (it is 100)
v2: Added sysctl documentation and cleaned code
Tested with packetdrill tests
Signed-off-by: Lawrence Brakmo <bra...@fb.com>
---
net/ipv4/tcp_timer.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index f712b41..3ea1cf8 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -192,6 +192,8 @@ static int tcp_write_timeout(struct sock *sk)
if (tp->syn_data && icsk->icsk_retransmits == 1)
NET_INC_STATS(sock_net(sk),
LINUX_MIB_TCPFASTOPENACTIVEFAIL);
+ } else if (!tp->syn_data && !tp->syn_fastopen) {
+ sk_rethink_txhash(sk);
}
retry_until = icsk->icsk_syn_retries ? :
net->ipv4.sysctl_tcp_syn_retries;
syn_set = true;
@@ -213,6 +215,8 @@ static int tcp_write_timeout(struct sock *sk)
tcp_mtu_probing(icsk, sk);
dst_negative_advice(sk);
+ } else {
+ sk_rethink_txhash(sk);
}
retry_until = net->ipv4.sysctl_tcp_retries2;
--
2.9.3
