Merge branch 'tcp'
Yuchung Cheng says:
====================
This patch series improve RTT sampling in three ways:
1. Sample RTT during fast recovery and reordering events.
2. Favor ack-based RTT to timestamps because of broken TS ECR fields
3. Consolidate the RTT measurement logic.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/include/net/tcp.h b/include/net/tcp.h
index d198005..c586847 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -591,7 +591,6 @@
extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
extern int tcp_mss_to_mtu(struct sock *sk, int mss);
extern void tcp_mtup_init(struct sock *sk);
-extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
extern void tcp_init_buffer_space(struct sock *sk);
static inline void tcp_bound_rto(const struct sock *sk)
@@ -1094,15 +1093,6 @@
ireq->loc_port = tcp_hdr(skb)->dest;
}
-/* Compute time elapsed between SYNACK and the ACK completing 3WHS */
-static inline void tcp_synack_rtt_meas(struct sock *sk,
- struct request_sock *req)
-{
- if (tcp_rsk(req)->snt_synack)
- tcp_valid_rtt_meas(sk,
- tcp_time_stamp - tcp_rsk(req)->snt_synack);
-}
-
extern void tcp_enter_memory_pressure(struct sock *sk);
static inline int keepalive_intvl_when(const struct tcp_sock *tp)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 28af45a..b61274b 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -1048,6 +1048,7 @@
int reord;
int fack_count;
int flag;
+ s32 rtt; /* RTT measured by SACKing never-retransmitted data */
};
/* Check if skb is fully within the SACK block. In presence of GSO skbs,
@@ -1108,7 +1109,7 @@
static u8 tcp_sacktag_one(struct sock *sk,
struct tcp_sacktag_state *state, u8 sacked,
u32 start_seq, u32 end_seq,
- bool dup_sack, int pcount)
+ int dup_sack, int pcount, u32 xmit_time)
{
struct tcp_sock *tp = tcp_sk(sk);
int fack_count = state->fack_count;
@@ -1148,6 +1149,9 @@
state->reord);
if (!after(end_seq, tp->high_seq))
state->flag |= FLAG_ORIG_SACK_ACKED;
+ /* Pick the earliest sequence sacked for RTT */
+ if (state->rtt < 0)
+ state->rtt = tcp_time_stamp - xmit_time;
}
if (sacked & TCPCB_LOST) {
@@ -1205,7 +1209,8 @@
* tcp_highest_sack_seq() when skb is highest_sack.
*/
tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,
- start_seq, end_seq, dup_sack, pcount);
+ start_seq, end_seq, dup_sack, pcount,
+ TCP_SKB_CB(skb)->when);
if (skb == tp->lost_skb_hint)
tp->lost_cnt_hint += pcount;
@@ -1479,7 +1484,8 @@
TCP_SKB_CB(skb)->seq,
TCP_SKB_CB(skb)->end_seq,
dup_sack,
- tcp_skb_pcount(skb));
+ tcp_skb_pcount(skb),
+ TCP_SKB_CB(skb)->when);
if (!before(TCP_SKB_CB(skb)->seq,
tcp_highest_sack_seq(tp)))
@@ -1536,7 +1542,7 @@
static int
tcp_sacktag_write_queue(struct sock *sk, const struct sk_buff *ack_skb,
- u32 prior_snd_una)
+ u32 prior_snd_una, s32 *sack_rtt)
{
struct tcp_sock *tp = tcp_sk(sk);
const unsigned char *ptr = (skb_transport_header(ack_skb) +
@@ -1554,6 +1560,7 @@
state.flag = 0;
state.reord = tp->packets_out;
+ state.rtt = -1;
if (!tp->sacked_out) {
if (WARN_ON(tp->fackets_out))
@@ -1737,6 +1744,7 @@
WARN_ON((int)tp->retrans_out < 0);
WARN_ON((int)tcp_packets_in_flight(tp) < 0);
#endif
+ *sack_rtt = state.rtt;
return state.flag;
}
@@ -2792,65 +2800,51 @@
tcp_xmit_retransmit_queue(sk);
}
-void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
+static inline bool tcp_ack_update_rtt(struct sock *sk, const int flag,
+ s32 seq_rtt, s32 sack_rtt)
{
- tcp_rtt_estimator(sk, seq_rtt);
- tcp_set_rto(sk);
- inet_csk(sk)->icsk_backoff = 0;
-}
-EXPORT_SYMBOL(tcp_valid_rtt_meas);
+ const struct tcp_sock *tp = tcp_sk(sk);
-/* Read draft-ietf-tcplw-high-performance before mucking
- * with this code. (Supersedes RFC1323)
- */
-static void tcp_ack_saw_tstamp(struct sock *sk, int flag)
-{
+ /* Prefer RTT measured from ACK's timing to TS-ECR. This is because
+ * broken middle-boxes or peers may corrupt TS-ECR fields. But
+ * Karn's algorithm forbids taking RTT if some retransmitted data
+ * is acked (RFC6298).
+ */
+ if (flag & FLAG_RETRANS_DATA_ACKED)
+ seq_rtt = -1;
+
+ if (seq_rtt < 0)
+ seq_rtt = sack_rtt;
+
/* RTTM Rule: A TSecr value received in a segment is used to
* update the averaged RTT measurement only if the segment
* acknowledges some new data, i.e., only if it advances the
* left edge of the send window.
- *
* See draft-ietf-tcplw-high-performance-00, section 3.3.
- * 1998/04/10 Andrey V. Savochkin <saw@msu.ru>
- *
- * Changed: reset backoff as soon as we see the first valid sample.
- * If we do not, we get strongly overestimated rto. With timestamps
- * samples are accepted even from very old segments: f.e., when rtt=1
- * increases to 8, we retransmit 5 times and after 8 seconds delayed
- * answer arrives rto becomes 120 seconds! If at least one of segments
- * in window is lost... Voila. --ANK (010210)
*/
+ if (seq_rtt < 0 && tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
+ seq_rtt = tcp_time_stamp - tp->rx_opt.rcv_tsecr;
+
+ if (seq_rtt < 0)
+ return false;
+
+ tcp_rtt_estimator(sk, seq_rtt);
+ tcp_set_rto(sk);
+
+ /* RFC6298: only reset backoff on valid RTT measurement. */
+ inet_csk(sk)->icsk_backoff = 0;
+ return true;
+}
+
+/* Compute time elapsed between (last) SYNACK and the ACK completing 3WHS. */
+static void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req)
+{
struct tcp_sock *tp = tcp_sk(sk);
+ s32 seq_rtt = -1;
- tcp_valid_rtt_meas(sk, tcp_time_stamp - tp->rx_opt.rcv_tsecr);
-}
-
-static void tcp_ack_no_tstamp(struct sock *sk, u32 seq_rtt, int flag)
-{
- /* We don't have a timestamp. Can only use
- * packets that are not retransmitted to determine
- * rtt estimates. Also, we must not reset the
- * backoff for rto until we get a non-retransmitted
- * packet. This allows us to deal with a situation
- * where the network delay has increased suddenly.
- * I.e. Karn's algorithm. (SIGCOMM '87, p5.)
- */
-
- if (flag & FLAG_RETRANS_DATA_ACKED)
- return;
-
- tcp_valid_rtt_meas(sk, seq_rtt);
-}
-
-static inline void tcp_ack_update_rtt(struct sock *sk, const int flag,
- const s32 seq_rtt)
-{
- const struct tcp_sock *tp = tcp_sk(sk);
- /* Note that peer MAY send zero echo. In this case it is ignored. (rfc1323) */
- if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)
- tcp_ack_saw_tstamp(sk, flag);
- else if (seq_rtt >= 0)
- tcp_ack_no_tstamp(sk, seq_rtt, flag);
+ if (tp->lsndtime && !tp->total_retrans)
+ seq_rtt = tcp_time_stamp - tp->lsndtime;
+ tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, seq_rtt, -1);
}
static void tcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
@@ -2939,7 +2933,7 @@
* arrived at the other end.
*/
static int tcp_clean_rtx_queue(struct sock *sk, int prior_fackets,
- u32 prior_snd_una)
+ u32 prior_snd_una, s32 sack_rtt)
{
struct tcp_sock *tp = tcp_sk(sk);
const struct inet_connection_sock *icsk = inet_csk(sk);
@@ -2978,8 +2972,6 @@
if (sacked & TCPCB_SACKED_RETRANS)
tp->retrans_out -= acked_pcount;
flag |= FLAG_RETRANS_DATA_ACKED;
- ca_seq_rtt = -1;
- seq_rtt = -1;
} else {
ca_seq_rtt = now - scb->when;
last_ackt = skb->tstamp;
@@ -3031,6 +3023,10 @@
if (skb && (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED))
flag |= FLAG_SACK_RENEGING;
+ if (tcp_ack_update_rtt(sk, flag, seq_rtt, sack_rtt) ||
+ (flag & FLAG_ACKED))
+ tcp_rearm_rto(sk);
+
if (flag & FLAG_ACKED) {
const struct tcp_congestion_ops *ca_ops
= inet_csk(sk)->icsk_ca_ops;
@@ -3040,9 +3036,6 @@
tcp_mtup_probe_success(sk);
}
- tcp_ack_update_rtt(sk, flag, seq_rtt);
- tcp_rearm_rto(sk);
-
if (tcp_is_reno(tp)) {
tcp_remove_reno_sacks(sk, pkts_acked);
} else {
@@ -3274,6 +3267,7 @@
int prior_packets = tp->packets_out;
const int prior_unsacked = tp->packets_out - tp->sacked_out;
int acked = 0; /* Number of packets newly acked */
+ s32 sack_rtt = -1;
/* If the ack is older than previous acks
* then we can probably ignore it.
@@ -3330,7 +3324,8 @@
flag |= tcp_ack_update_window(sk, skb, ack, ack_seq);
if (TCP_SKB_CB(skb)->sacked)
- flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una);
+ flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
+ &sack_rtt);
if (TCP_ECN_rcv_ecn_echo(tp, tcp_hdr(skb)))
flag |= FLAG_ECE;
@@ -3349,7 +3344,7 @@
/* See if we can take anything off of the retransmit queue. */
acked = tp->packets_out;
- flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una);
+ flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una, sack_rtt);
acked -= tp->packets_out;
if (tcp_ack_is_dubious(sk, flag)) {
@@ -3402,7 +3397,8 @@
* If data was DSACKed, see if we can undo a cwnd reduction.
*/
if (TCP_SKB_CB(skb)->sacked) {
- flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una);
+ flag |= tcp_sacktag_write_queue(sk, skb, prior_snd_una,
+ &sack_rtt);
tcp_fastretrans_alert(sk, acked, prior_unsacked,
is_dupack, flag);
}
@@ -5624,9 +5620,7 @@
* so release it.
*/
if (req) {
- tcp_synack_rtt_meas(sk, req);
tp->total_retrans = req->num_retrans;
-
reqsk_fastopen_remove(sk, req, false);
} else {
/* Make sure socket is routed, for correct metrics. */
@@ -5651,6 +5645,7 @@
tp->snd_una = TCP_SKB_CB(skb)->ack_seq;
tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale;
tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);
+ tcp_synack_rtt_meas(sk, req);
if (tp->rx_opt.tstamp_ok)
tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index b299da5f..2e3f129 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -1671,8 +1671,6 @@
newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
tcp_initialize_rcv_mss(newsk);
- tcp_synack_rtt_meas(newsk, req);
- newtp->total_retrans = req->num_retrans;
#ifdef CONFIG_TCP_MD5SIG
/* Copy over the MD5 key from the original socket */
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index ab1c086..58a3e69 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -411,6 +411,8 @@
newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
tcp_enable_early_retrans(newtp);
newtp->tlp_high_seq = 0;
+ newtp->lsndtime = treq->snt_synack;
+ newtp->total_retrans = req->num_retrans;
/* So many TCP implementations out there (incorrectly) count the
* initial SYN frame in their delayed-ACK and congestion control
@@ -666,12 +668,6 @@
if (!(flg & TCP_FLAG_ACK))
return NULL;
- /* Got ACK for our SYNACK, so update baseline for SYNACK RTT sample. */
- if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
- tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
- else if (req->num_retrans) /* don't take RTT sample if retrans && ~TS */
- tcp_rsk(req)->snt_synack = 0;
-
/* For Fast Open no more processing is needed (sk is the
* child socket).
*/
diff --git a/net/ipv6/tcp_ipv6.c b/net/ipv6/tcp_ipv6.c
index 6e1649d..80fe69e 100644
--- a/net/ipv6/tcp_ipv6.c
+++ b/net/ipv6/tcp_ipv6.c
@@ -1237,8 +1237,6 @@
newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
tcp_initialize_rcv_mss(newsk);
- tcp_synack_rtt_meas(newsk, req);
- newtp->total_retrans = req->num_retrans;
newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
newinet->inet_rcv_saddr = LOOPBACK4_IPV6;