2015-10-17 12:57:46 +08:00
|
|
|
#include <linux/tcp.h>
|
|
|
|
#include <net/tcp.h>
|
|
|
|
|
2017-01-13 14:11:36 +08:00
|
|
|
int sysctl_tcp_recovery __read_mostly = TCP_RACK_LOSS_DETECTION;
|
2015-10-17 12:57:47 +08:00
|
|
|
|
2017-01-13 14:11:30 +08:00
|
|
|
static void tcp_rack_mark_skb_lost(struct sock *sk, struct sk_buff *skb)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
|
|
|
|
tcp_skb_mark_lost_uncond_verify(tp, skb);
|
|
|
|
if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_RETRANS) {
|
|
|
|
/* Account for retransmits that are lost again */
|
|
|
|
TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;
|
|
|
|
tp->retrans_out -= tcp_skb_pcount(skb);
|
|
|
|
NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSTRETRANSMIT);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-01-13 14:11:34 +08:00
|
|
|
static bool tcp_rack_sent_after(const struct skb_mstamp *t1,
|
|
|
|
const struct skb_mstamp *t2,
|
|
|
|
u32 seq1, u32 seq2)
|
|
|
|
{
|
|
|
|
return skb_mstamp_after(t1, t2) ||
|
|
|
|
(t1->v64 == t2->v64 && after(seq1, seq2));
|
|
|
|
}
|
|
|
|
|
2017-01-13 14:11:36 +08:00
|
|
|
/* RACK loss detection (IETF draft draft-ietf-tcpm-rack-01):
|
|
|
|
*
|
|
|
|
* Marks a packet lost, if some packet sent later has been (s)acked.
|
2015-10-17 12:57:47 +08:00
|
|
|
* The underlying idea is similar to the traditional dupthresh and FACK
|
|
|
|
* but they look at different metrics:
|
|
|
|
*
|
|
|
|
* dupthresh: 3 OOO packets delivered (packet count)
|
|
|
|
* FACK: sequence delta to highest sacked sequence (sequence space)
|
|
|
|
* RACK: sent time delta to the latest delivered packet (time domain)
|
|
|
|
*
|
|
|
|
* The advantage of RACK is it applies to both original and retransmitted
|
|
|
|
* packet and therefore is robust against tail losses. Another advantage
|
|
|
|
* is being more resilient to reordering by simply allowing some
|
|
|
|
* "settling delay", instead of tweaking the dupthresh.
|
|
|
|
*
|
2017-01-13 14:11:36 +08:00
|
|
|
* When tcp_rack_detect_loss() detects some packets are lost and we
|
|
|
|
* are not already in the CA_Recovery state, either tcp_rack_reo_timeout()
|
|
|
|
* or tcp_time_to_recover()'s "Trick#1: the loss is proven" code path will
|
|
|
|
* make us enter the CA_Recovery state.
|
2015-10-17 12:57:47 +08:00
|
|
|
*/
|
2017-01-13 14:11:33 +08:00
|
|
|
static void tcp_rack_detect_loss(struct sock *sk, const struct skb_mstamp *now,
|
|
|
|
u32 *reo_timeout)
|
2015-10-17 12:57:47 +08:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct sk_buff *skb;
|
2017-01-13 14:11:31 +08:00
|
|
|
u32 reo_wnd;
|
2015-10-17 12:57:47 +08:00
|
|
|
|
2017-01-13 14:11:33 +08:00
|
|
|
*reo_timeout = 0;
|
2015-10-17 12:57:47 +08:00
|
|
|
/* To be more reordering resilient, allow min_rtt/4 settling delay
|
|
|
|
* (lower-bounded to 1000uS). We use min_rtt instead of the smoothed
|
|
|
|
* RTT because reordering is often a path property and less related
|
|
|
|
* to queuing or delayed ACKs.
|
|
|
|
*/
|
|
|
|
reo_wnd = 1000;
|
2017-01-13 14:11:36 +08:00
|
|
|
if ((tp->rack.reord || !tp->lost_out) && tcp_min_rtt(tp) != ~0U)
|
2015-10-17 12:57:47 +08:00
|
|
|
reo_wnd = max(tcp_min_rtt(tp) >> 2, reo_wnd);
|
|
|
|
|
|
|
|
tcp_for_write_queue(skb, sk) {
|
|
|
|
struct tcp_skb_cb *scb = TCP_SKB_CB(skb);
|
|
|
|
|
|
|
|
if (skb == tcp_send_head(sk))
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* Skip ones already (s)acked */
|
|
|
|
if (!after(scb->end_seq, tp->snd_una) ||
|
|
|
|
scb->sacked & TCPCB_SACKED_ACKED)
|
|
|
|
continue;
|
|
|
|
|
2017-01-13 14:11:34 +08:00
|
|
|
if (tcp_rack_sent_after(&tp->rack.mstamp, &skb->skb_mstamp,
|
|
|
|
tp->rack.end_seq, scb->end_seq)) {
|
2017-01-13 14:11:32 +08:00
|
|
|
/* Step 3 in draft-cheng-tcpm-rack-00.txt:
|
|
|
|
* A packet is lost if its elapsed time is beyond
|
|
|
|
* the recent RTT plus the reordering window.
|
|
|
|
*/
|
2017-01-13 14:11:33 +08:00
|
|
|
u32 elapsed = skb_mstamp_us_delta(now,
|
|
|
|
&skb->skb_mstamp);
|
|
|
|
s32 remaining = tp->rack.rtt_us + reo_wnd - elapsed;
|
|
|
|
|
|
|
|
if (remaining < 0) {
|
2017-01-13 14:11:32 +08:00
|
|
|
tcp_rack_mark_skb_lost(sk, skb);
|
2017-01-13 14:11:33 +08:00
|
|
|
continue;
|
2017-01-13 14:11:32 +08:00
|
|
|
}
|
2017-01-13 14:11:33 +08:00
|
|
|
|
|
|
|
/* Skip ones marked lost but not yet retransmitted */
|
|
|
|
if ((scb->sacked & TCPCB_LOST) &&
|
|
|
|
!(scb->sacked & TCPCB_SACKED_RETRANS))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* Record maximum wait time (+1 to avoid 0) */
|
|
|
|
*reo_timeout = max_t(u32, *reo_timeout, 1 + remaining);
|
|
|
|
|
2015-10-17 12:57:47 +08:00
|
|
|
} else if (!(scb->sacked & TCPCB_RETRANS)) {
|
|
|
|
/* Original data are sent sequentially so stop early
|
|
|
|
* b/c the rest are all sent after rack_sent
|
|
|
|
*/
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2017-01-13 14:11:31 +08:00
|
|
|
}
|
|
|
|
|
2017-01-13 14:11:32 +08:00
|
|
|
void tcp_rack_mark_lost(struct sock *sk, const struct skb_mstamp *now)
|
2017-01-13 14:11:31 +08:00
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
2017-01-13 14:11:33 +08:00
|
|
|
u32 timeout;
|
2017-01-13 14:11:31 +08:00
|
|
|
|
2017-01-13 14:11:36 +08:00
|
|
|
if (!tp->rack.advanced)
|
2017-01-13 14:11:31 +08:00
|
|
|
return;
|
2017-01-13 14:11:33 +08:00
|
|
|
|
2017-01-13 14:11:31 +08:00
|
|
|
/* Reset the advanced flag to avoid unnecessary queue scanning */
|
|
|
|
tp->rack.advanced = 0;
|
2017-01-13 14:11:33 +08:00
|
|
|
tcp_rack_detect_loss(sk, now, &timeout);
|
|
|
|
if (timeout) {
|
|
|
|
timeout = usecs_to_jiffies(timeout + TCP_REO_TIMEOUT_MIN);
|
|
|
|
inet_csk_reset_xmit_timer(sk, ICSK_TIME_REO_TIMEOUT,
|
|
|
|
timeout, inet_csk(sk)->icsk_rto);
|
|
|
|
}
|
2015-10-17 12:57:47 +08:00
|
|
|
}
|
|
|
|
|
2017-01-13 14:11:32 +08:00
|
|
|
/* Record the most recently (re)sent time among the (s)acked packets
|
|
|
|
* This is "Step 3: Advance RACK.xmit_time and update RACK.RTT" from
|
|
|
|
* draft-cheng-tcpm-rack-00.txt
|
|
|
|
*/
|
2017-01-13 14:11:34 +08:00
|
|
|
void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
|
2017-01-13 14:11:32 +08:00
|
|
|
const struct skb_mstamp *xmit_time,
|
|
|
|
const struct skb_mstamp *ack_time)
|
2015-10-17 12:57:46 +08:00
|
|
|
{
|
2017-01-13 14:11:32 +08:00
|
|
|
u32 rtt_us;
|
|
|
|
|
2015-10-17 12:57:46 +08:00
|
|
|
if (tp->rack.mstamp.v64 &&
|
2017-01-13 14:11:34 +08:00
|
|
|
!tcp_rack_sent_after(xmit_time, &tp->rack.mstamp,
|
|
|
|
end_seq, tp->rack.end_seq))
|
2015-10-17 12:57:46 +08:00
|
|
|
return;
|
|
|
|
|
2017-01-13 14:11:32 +08:00
|
|
|
rtt_us = skb_mstamp_us_delta(ack_time, xmit_time);
|
2015-10-17 12:57:46 +08:00
|
|
|
if (sacked & TCPCB_RETRANS) {
|
|
|
|
/* If the sacked packet was retransmitted, it's ambiguous
|
|
|
|
* whether the retransmission or the original (or the prior
|
|
|
|
* retransmission) was sacked.
|
|
|
|
*
|
|
|
|
* If the original is lost, there is no ambiguity. Otherwise
|
|
|
|
* we assume the original can be delayed up to aRTT + min_rtt.
|
|
|
|
* the aRTT term is bounded by the fast recovery or timeout,
|
|
|
|
* so it's at least one RTT (i.e., retransmission is at least
|
|
|
|
* an RTT later).
|
|
|
|
*/
|
2017-01-13 14:11:32 +08:00
|
|
|
if (rtt_us < tcp_min_rtt(tp))
|
2015-10-17 12:57:46 +08:00
|
|
|
return;
|
|
|
|
}
|
2017-01-13 14:11:32 +08:00
|
|
|
tp->rack.rtt_us = rtt_us;
|
2015-10-17 12:57:46 +08:00
|
|
|
tp->rack.mstamp = *xmit_time;
|
2017-01-13 14:11:34 +08:00
|
|
|
tp->rack.end_seq = end_seq;
|
2015-10-17 12:57:46 +08:00
|
|
|
tp->rack.advanced = 1;
|
|
|
|
}
|
2017-01-13 14:11:33 +08:00
|
|
|
|
|
|
|
/* We have waited long enough to accommodate reordering. Mark the expired
|
|
|
|
* packets lost and retransmit them.
|
|
|
|
*/
|
|
|
|
void tcp_rack_reo_timeout(struct sock *sk)
|
|
|
|
{
|
|
|
|
struct tcp_sock *tp = tcp_sk(sk);
|
|
|
|
struct skb_mstamp now;
|
|
|
|
u32 timeout, prior_inflight;
|
|
|
|
|
|
|
|
skb_mstamp_get(&now);
|
|
|
|
prior_inflight = tcp_packets_in_flight(tp);
|
|
|
|
tcp_rack_detect_loss(sk, &now, &timeout);
|
|
|
|
if (prior_inflight != tcp_packets_in_flight(tp)) {
|
|
|
|
if (inet_csk(sk)->icsk_ca_state != TCP_CA_Recovery) {
|
|
|
|
tcp_enter_recovery(sk, false);
|
|
|
|
if (!inet_csk(sk)->icsk_ca_ops->cong_control)
|
|
|
|
tcp_cwnd_reduction(sk, 1, 0);
|
|
|
|
}
|
|
|
|
tcp_xmit_retransmit_queue(sk);
|
|
|
|
}
|
|
|
|
if (inet_csk(sk)->icsk_pending != ICSK_TIME_RETRANS)
|
|
|
|
tcp_rearm_rto(sk);
|
|
|
|
}
|