Lines Matching +full:wakeup +full:- +full:rtt +full:- +full:timer
1 // SPDX-License-Identifier: GPL-2.0
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
27 * Better retransmit timer handling.
44 * Andrey Savochkin: Fix RTT measurements in the presence of
62 * Pasi Sarolahti: F-RTO for dealing with spurious RTOs
98 #define FLAG_DSACKING_ACK 0x800 /* SACK blocks contained D-SACK info */
99 #define FLAG_SET_XMIT_TIMER 0x1000 /* Set TLP or RTO timer */
116 #define REXMIT_NEW 2 /* FRTO-style transmit of unsent/new packets */
124 icsk->icsk_clean_acked = cad; in clean_acked_data_enable()
132 icsk->icsk_clean_acked = NULL; in clean_acked_data_disable()
146 bool unknown_opt = tcp_sk(sk)->rx_opt.saw_unknown && in bpf_skops_parse_hdr()
160 switch (sk->sk_state) { in bpf_skops_parse_hdr()
212 dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif); in tcp_gro_dev_warn()
213 if (!dev || len >= READ_ONCE(dev->mtu)) in tcp_gro_dev_warn()
215 dev ? dev->name : "Unknown driver"); in tcp_gro_dev_warn()
225 const unsigned int lss = icsk->icsk_ack.last_seg_size; in tcp_measure_rcv_mss()
228 icsk->icsk_ack.last_seg_size = 0; in tcp_measure_rcv_mss()
230 /* skb->len may jitter because of SACKs, even if peer in tcp_measure_rcv_mss()
231 * sends good full-sized frames. in tcp_measure_rcv_mss()
233 len = skb_shinfo(skb)->gso_size ? : skb->len; in tcp_measure_rcv_mss()
234 if (len >= icsk->icsk_ack.rcv_mss) { in tcp_measure_rcv_mss()
239 if (unlikely(len != icsk->icsk_ack.rcv_mss)) { in tcp_measure_rcv_mss()
240 u64 val = (u64)skb->len << TCP_RMEM_TO_WIN_SCALE; in tcp_measure_rcv_mss()
241 u8 old_ratio = tcp_sk(sk)->scaling_ratio; in tcp_measure_rcv_mss()
243 do_div(val, skb->truesize); in tcp_measure_rcv_mss()
244 tcp_sk(sk)->scaling_ratio = val ? val : 1; in tcp_measure_rcv_mss()
246 if (old_ratio != tcp_sk(sk)->scaling_ratio) { in tcp_measure_rcv_mss()
249 val = tcp_win_from_space(sk, sk->sk_rcvbuf); in tcp_measure_rcv_mss()
252 if (tp->window_clamp < tp->rcvq_space.space) in tcp_measure_rcv_mss()
253 tp->rcvq_space.space = tp->window_clamp; in tcp_measure_rcv_mss()
256 icsk->icsk_ack.rcv_mss = min_t(unsigned int, len, in tcp_measure_rcv_mss()
257 tcp_sk(sk)->advmss); in tcp_measure_rcv_mss()
258 /* Account for possibly-removed options */ in tcp_measure_rcv_mss()
259 DO_ONCE_LITE_IF(len > icsk->icsk_ack.rcv_mss + MAX_TCP_OPTION_SPACE, in tcp_measure_rcv_mss()
264 * may be waiting for an ACK if cwnd-bound or using TX zero in tcp_measure_rcv_mss()
267 * reads all of the data and is not ping-pong. If len > MSS in tcp_measure_rcv_mss()
272 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_PSH) in tcp_measure_rcv_mss()
273 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; in tcp_measure_rcv_mss()
280 len += skb->data - skb_transport_header(skb); in tcp_measure_rcv_mss()
285 * to handle super-low mtu links fairly. in tcp_measure_rcv_mss()
293 len -= tcp_sk(sk)->tcp_header_len; in tcp_measure_rcv_mss()
294 icsk->icsk_ack.last_seg_size = len; in tcp_measure_rcv_mss()
296 icsk->icsk_ack.rcv_mss = len; in tcp_measure_rcv_mss()
300 if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED) in tcp_measure_rcv_mss()
301 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2; in tcp_measure_rcv_mss()
302 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; in tcp_measure_rcv_mss()
309 unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss); in tcp_incr_quickack()
314 if (quickacks > icsk->icsk_ack.quick) in tcp_incr_quickack()
315 icsk->icsk_ack.quick = quickacks; in tcp_incr_quickack()
324 icsk->icsk_ack.ato = TCP_ATO_MIN; in tcp_enter_quickack_mode()
337 (icsk->icsk_ack.quick && !inet_csk_in_pingpong_mode(sk)); in tcp_in_quickack_mode()
342 if (tp->ecn_flags & TCP_ECN_OK) in tcp_ecn_queue_cwr()
343 tp->ecn_flags |= TCP_ECN_QUEUE_CWR; in tcp_ecn_queue_cwr()
348 if (tcp_hdr(skb)->cwr) { in tcp_ecn_accept_cwr()
349 tcp_sk(sk)->ecn_flags &= ~TCP_ECN_DEMAND_CWR; in tcp_ecn_accept_cwr()
355 if (TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) in tcp_ecn_accept_cwr()
356 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; in tcp_ecn_accept_cwr()
362 tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR; in tcp_ecn_withdraw_cwr()
369 switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) { in __tcp_ecn_check_ce()
375 if (tp->ecn_flags & TCP_ECN_SEEN) in __tcp_ecn_check_ce()
382 if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) { in __tcp_ecn_check_ce()
385 tp->ecn_flags |= TCP_ECN_DEMAND_CWR; in __tcp_ecn_check_ce()
387 tp->ecn_flags |= TCP_ECN_SEEN; in __tcp_ecn_check_ce()
392 tp->ecn_flags |= TCP_ECN_SEEN; in __tcp_ecn_check_ce()
399 if (tcp_sk(sk)->ecn_flags & TCP_ECN_OK) in tcp_ecn_check_ce()
405 if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr)) in tcp_ecn_rcv_synack()
406 tp->ecn_flags &= ~TCP_ECN_OK; in tcp_ecn_rcv_synack()
411 if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr)) in tcp_ecn_rcv_syn()
412 tp->ecn_flags &= ~TCP_ECN_OK; in tcp_ecn_rcv_syn()
417 if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK)) in tcp_ecn_rcv_ecn_echo()
424 * 1. Tuning sk->sk_sndbuf, when connection enters established state.
430 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_sndbuf_expand()
435 * and skb->head is kmalloced using power of two area of memory in tcp_sndbuf_expand()
437 per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + in tcp_sndbuf_expand()
445 nr_segs = max_t(u32, nr_segs, tp->reordering + 1); in tcp_sndbuf_expand()
451 sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2; in tcp_sndbuf_expand()
454 if (sk->sk_sndbuf < sndmem) in tcp_sndbuf_expand()
455 WRITE_ONCE(sk->sk_sndbuf, in tcp_sndbuf_expand()
456 min(sndmem, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[2]))); in tcp_sndbuf_expand()
462 * forward and advertised in receiver window (tp->rcv_wnd) and
466 * tcp_full_space(), in this case tcp_full_space() - window_clamp
474 * - to enforce header prediction at sender, even when application
476 * - to prevent pruning of receive queue because of misprediction
491 int window = tcp_win_from_space(sk, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])) >> 1; in __tcp_grow_window()
493 while (tp->rcv_ssthresh <= window) { in __tcp_grow_window()
494 if (truesize <= skb->len) in __tcp_grow_window()
495 return 2 * inet_csk(sk)->icsk_ack.rcv_mss; in __tcp_grow_window()
504 * can play nice with us, as sk_buff and skb->head might be either
507 * and if no payload was pulled in skb->head before reaching us.
511 u32 truesize = skb->truesize; in truesize_adjust()
514 truesize -= SKB_TRUESIZE(skb_end_offset(skb)); in truesize_adjust()
516 if (unlikely((int)truesize < (int)skb->len)) in truesize_adjust()
517 truesize = skb->truesize; in truesize_adjust()
528 room = min_t(int, tp->window_clamp, tcp_space(sk)) - tp->rcv_ssthresh; in tcp_grow_window()
541 if (tcp_win_from_space(sk, truesize) <= skb->len) in tcp_grow_window()
542 incr = 2 * tp->advmss; in tcp_grow_window()
547 incr = max_t(int, incr, 2 * skb->len); in tcp_grow_window()
548 tp->rcv_ssthresh += min(room, incr); in tcp_grow_window()
549 inet_csk(sk)->icsk_ack.quick |= 1; in tcp_grow_window()
564 int tcp_app_win = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_app_win); in tcp_init_buffer_space()
568 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) in tcp_init_buffer_space()
572 tp->rcvq_space.time = tp->tcp_mstamp; in tcp_init_buffer_space()
573 tp->rcvq_space.seq = tp->copied_seq; in tcp_init_buffer_space()
577 if (tp->window_clamp >= maxwin) { in tcp_init_buffer_space()
578 WRITE_ONCE(tp->window_clamp, maxwin); in tcp_init_buffer_space()
580 if (tcp_app_win && maxwin > 4 * tp->advmss) in tcp_init_buffer_space()
581 WRITE_ONCE(tp->window_clamp, in tcp_init_buffer_space()
582 max(maxwin - (maxwin >> tcp_app_win), in tcp_init_buffer_space()
583 4 * tp->advmss)); in tcp_init_buffer_space()
588 tp->window_clamp > 2 * tp->advmss && in tcp_init_buffer_space()
589 tp->window_clamp + tp->advmss > maxwin) in tcp_init_buffer_space()
590 WRITE_ONCE(tp->window_clamp, in tcp_init_buffer_space()
591 max(2 * tp->advmss, maxwin - tp->advmss)); in tcp_init_buffer_space()
593 tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp); in tcp_init_buffer_space()
594 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_init_buffer_space()
595 tp->rcvq_space.space = min3(tp->rcv_ssthresh, tp->rcv_wnd, in tcp_init_buffer_space()
596 (u32)TCP_INIT_CWND * tp->advmss); in tcp_init_buffer_space()
607 icsk->icsk_ack.quick = 0; in tcp_clamp_window()
608 rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]); in tcp_clamp_window()
610 if (sk->sk_rcvbuf < rmem2 && in tcp_clamp_window()
611 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) && in tcp_clamp_window()
614 WRITE_ONCE(sk->sk_rcvbuf, in tcp_clamp_window()
615 min(atomic_read(&sk->sk_rmem_alloc), rmem2)); in tcp_clamp_window()
617 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) in tcp_clamp_window()
618 tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss); in tcp_clamp_window()
631 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); in tcp_initialize_rcv_mss()
633 hint = min(hint, tp->rcv_wnd / 2); in tcp_initialize_rcv_mss()
637 inet_csk(sk)->icsk_ack.rcv_mss = hint; in tcp_initialize_rcv_mss()
643 * The algorithm for RTT estimation w/o timestamps is based on
644 * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL.
654 u32 new_sample = tp->rcv_rtt_est.rtt_us; in tcp_rcv_rtt_update()
658 /* If we sample in larger samples in the non-timestamp in tcp_rcv_rtt_update()
659 * case, we could grossly overestimate the RTT especially in tcp_rcv_rtt_update()
664 * non-timestamp case, we do not smooth things out in tcp_rcv_rtt_update()
669 m -= (new_sample >> 3); in tcp_rcv_rtt_update()
681 tp->rcv_rtt_est.rtt_us = new_sample; in tcp_rcv_rtt_update()
688 if (tp->rcv_rtt_est.time == 0) in tcp_rcv_rtt_measure()
690 if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq)) in tcp_rcv_rtt_measure()
692 delta_us = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcv_rtt_est.time); in tcp_rcv_rtt_measure()
698 tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd; in tcp_rcv_rtt_measure()
699 tp->rcv_rtt_est.time = tp->tcp_mstamp; in tcp_rcv_rtt_measure()
706 delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr; in tcp_rtt_tsopt_us()
707 if (tp->tcp_usec_ts) in tcp_rtt_tsopt_us()
716 return -1; in tcp_rtt_tsopt_us()
724 if (tp->rx_opt.rcv_tsecr == tp->rcv_rtt_last_tsecr) in tcp_rcv_rtt_measure_ts()
726 tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr; in tcp_rcv_rtt_measure_ts()
728 if (TCP_SKB_CB(skb)->end_seq - in tcp_rcv_rtt_measure_ts()
729 TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) { in tcp_rcv_rtt_measure_ts()
750 time = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcvq_space.time); in tcp_rcv_space_adjust()
751 if (time < (tp->rcv_rtt_est.rtt_us >> 3) || tp->rcv_rtt_est.rtt_us == 0) in tcp_rcv_space_adjust()
754 /* Number of bytes copied to user in last RTT */ in tcp_rcv_space_adjust()
755 copied = tp->copied_seq - tp->rcvq_space.seq; in tcp_rcv_space_adjust()
756 if (copied <= tp->rcvq_space.space) in tcp_rcv_space_adjust()
760 * copied = bytes received in previous RTT, our base window in tcp_rcv_space_adjust()
763 * every RTT, we need a 4x factor, because the ACK we are sending in tcp_rcv_space_adjust()
764 * now is for the next RTT, not the current one : in tcp_rcv_space_adjust()
765 * <prev RTT . ><current RTT .. ><next RTT .... > in tcp_rcv_space_adjust()
768 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) && in tcp_rcv_space_adjust()
769 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) { in tcp_rcv_space_adjust()
776 rcvwin = ((u64)copied << 1) + 16 * tp->advmss; in tcp_rcv_space_adjust()
779 grow = rcvwin * (copied - tp->rcvq_space.space); in tcp_rcv_space_adjust()
780 do_div(grow, tp->rcvq_space.space); in tcp_rcv_space_adjust()
784 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])); in tcp_rcv_space_adjust()
785 if (rcvbuf > sk->sk_rcvbuf) { in tcp_rcv_space_adjust()
786 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf); in tcp_rcv_space_adjust()
789 WRITE_ONCE(tp->window_clamp, in tcp_rcv_space_adjust()
793 tp->rcvq_space.space = copied; in tcp_rcv_space_adjust()
796 tp->rcvq_space.seq = tp->copied_seq; in tcp_rcv_space_adjust()
797 tp->rcvq_space.time = tp->tcp_mstamp; in tcp_rcv_space_adjust()
805 if (skb->protocol == htons(ETH_P_IPV6)) in tcp_save_lrcv_flowlabel()
806 icsk->icsk_ack.lrcv_flowlabel = ntohl(ip6_flowlabel(ipv6_hdr(skb))); in tcp_save_lrcv_flowlabel()
811 * behavior of the tp->ato delayed ack timeout interval. When a
818 * queue. -DaveM
834 if (!icsk->icsk_ack.ato) { in tcp_event_data_recv()
839 icsk->icsk_ack.ato = TCP_ATO_MIN; in tcp_event_data_recv()
841 int m = now - icsk->icsk_ack.lrcvtime; in tcp_event_data_recv()
845 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2; in tcp_event_data_recv()
846 } else if (m < icsk->icsk_ack.ato) { in tcp_event_data_recv()
847 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m; in tcp_event_data_recv()
848 if (icsk->icsk_ack.ato > icsk->icsk_rto) in tcp_event_data_recv()
849 icsk->icsk_ack.ato = icsk->icsk_rto; in tcp_event_data_recv()
850 } else if (m > icsk->icsk_rto) { in tcp_event_data_recv()
857 icsk->icsk_ack.lrcvtime = now; in tcp_event_data_recv()
862 if (skb->len >= 128) in tcp_event_data_recv()
866 /* Called to compute a smoothed rtt estimate. The data fed to this
873 * it up into three procedures. -- erics
878 long m = mrtt_us; /* RTT */ in tcp_rtt_estimator()
879 u32 srtt = tp->srtt_us; in tcp_rtt_estimator()
882 * article in SIGCOMM '88. Note that rtt and mdev in tcp_rtt_estimator()
883 * are scaled versions of rtt and mean deviation. in tcp_rtt_estimator()
888 * RTO = rtt + 4 * mdev in tcp_rtt_estimator()
898 m -= (srtt >> 3); /* m is now error in rtt est */ in tcp_rtt_estimator()
899 srtt += m; /* rtt = 7/8 rtt + 1/8 new */ in tcp_rtt_estimator()
901 m = -m; /* m is now abs(error) */ in tcp_rtt_estimator()
902 m -= (tp->mdev_us >> 2); /* similar update on mdev */ in tcp_rtt_estimator()
904 * Eifel blocks mdev updates when rtt decreases. in tcp_rtt_estimator()
914 m -= (tp->mdev_us >> 2); /* similar update on mdev */ in tcp_rtt_estimator()
916 tp->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */ in tcp_rtt_estimator()
917 if (tp->mdev_us > tp->mdev_max_us) { in tcp_rtt_estimator()
918 tp->mdev_max_us = tp->mdev_us; in tcp_rtt_estimator()
919 if (tp->mdev_max_us > tp->rttvar_us) in tcp_rtt_estimator()
920 tp->rttvar_us = tp->mdev_max_us; in tcp_rtt_estimator()
922 if (after(tp->snd_una, tp->rtt_seq)) { in tcp_rtt_estimator()
923 if (tp->mdev_max_us < tp->rttvar_us) in tcp_rtt_estimator()
924 tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2; in tcp_rtt_estimator()
925 tp->rtt_seq = tp->snd_nxt; in tcp_rtt_estimator()
926 tp->mdev_max_us = tcp_rto_min_us(sk); in tcp_rtt_estimator()
932 srtt = m << 3; /* take the measured time to be rtt */ in tcp_rtt_estimator()
933 tp->mdev_us = m << 1; /* make sure rto = 3*rtt */ in tcp_rtt_estimator()
934 tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk)); in tcp_rtt_estimator()
935 tp->mdev_max_us = tp->rttvar_us; in tcp_rtt_estimator()
936 tp->rtt_seq = tp->snd_nxt; in tcp_rtt_estimator()
940 tp->srtt_us = max(1U, srtt); in tcp_rtt_estimator()
949 rate = (u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3); in tcp_update_pacing_rate()
955 * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh) in tcp_update_pacing_rate()
956 * If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching in tcp_update_pacing_rate()
959 if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2) in tcp_update_pacing_rate()
960 rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio); in tcp_update_pacing_rate()
962 rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio); in tcp_update_pacing_rate()
964 rate *= max(tcp_snd_cwnd(tp), tp->packets_out); in tcp_update_pacing_rate()
966 if (likely(tp->srtt_us)) in tcp_update_pacing_rate()
967 do_div(rate, tp->srtt_us); in tcp_update_pacing_rate()
973 WRITE_ONCE(sk->sk_pacing_rate, in tcp_update_pacing_rate()
974 min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate))); in tcp_update_pacing_rate()
986 * 1. If rtt variance happened to be less 50msec, it is hallucination. in tcp_set_rto()
990 * is invisible. Actually, Linux-2.4 also generates erratic in tcp_set_rto()
993 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp); in tcp_set_rto()
1013 return min_t(__u32, cwnd, tp->snd_cwnd_clamp); in tcp_init_cwnd()
1017 /* Timestamps for earliest and latest never-retransmitted segment
1018 * that was SACKed. RTO needs the earliest RTT to stay conservative,
1030 /* Take a notice that peer is sending D-SACKs. Skip update of data delivery
1033 * - DSACKed sequence range is larger than maximum receiver's window.
1034 * - Total no. of DSACKed segments exceed the total no. of retransmitted segs.
1044 seq_len = end_seq - start_seq; in tcp_dsack_seen()
1046 if (seq_len > tp->max_window) in tcp_dsack_seen()
1048 if (seq_len > tp->mss_cache) in tcp_dsack_seen()
1049 dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache); in tcp_dsack_seen()
1050 else if (tp->tlp_high_seq && tp->tlp_high_seq == end_seq) in tcp_dsack_seen()
1051 state->flag |= FLAG_DSACK_TLP; in tcp_dsack_seen()
1053 tp->dsack_dups += dup_segs; in tcp_dsack_seen()
1055 if (tp->dsack_dups > tp->total_retrans) in tcp_dsack_seen()
1058 tp->rx_opt.sack_ok |= TCP_DSACK_SEEN; in tcp_dsack_seen()
1063 * is timer-driven, not triggered by RACK). in tcp_dsack_seen()
1065 if (tp->reord_seen && !(state->flag & FLAG_DSACK_TLP)) in tcp_dsack_seen()
1066 tp->rack.dsack_seen = 1; in tcp_dsack_seen()
1068 state->flag |= FLAG_DSACKING_ACK; in tcp_dsack_seen()
1070 state->sack_delivered += dup_segs; in tcp_dsack_seen()
1076 * some lower never-retransmitted sequence ("low_seq"). The maximum reordering
1077 * distance is approximated in full-mss packet distance ("reordering").
1083 const u32 mss = tp->mss_cache; in tcp_check_sack_reordering()
1090 metric = fack - low_seq; in tcp_check_sack_reordering()
1091 if ((metric > tp->reordering * mss) && mss) { in tcp_check_sack_reordering()
1094 tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state, in tcp_check_sack_reordering()
1095 tp->reordering, in tcp_check_sack_reordering()
1097 tp->sacked_out, in tcp_check_sack_reordering()
1098 tp->undo_marker ? tp->undo_retrans : 0); in tcp_check_sack_reordering()
1100 tp->reordering = min_t(u32, (metric + mss - 1) / mss, in tcp_check_sack_reordering()
1101 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering)); in tcp_check_sack_reordering()
1105 tp->reord_seen++; in tcp_check_sack_reordering()
1117 if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) || in tcp_verify_retransmit_hint()
1118 (tp->retransmit_skb_hint && in tcp_verify_retransmit_hint()
1119 before(TCP_SKB_CB(skb)->seq, in tcp_verify_retransmit_hint()
1120 TCP_SKB_CB(tp->retransmit_skb_hint)->seq))) in tcp_verify_retransmit_hint()
1121 tp->retransmit_skb_hint = skb; in tcp_verify_retransmit_hint()
1129 tp->lost += tcp_skb_pcount(skb); in tcp_notify_skb_loss_event()
1134 __u8 sacked = TCP_SKB_CB(skb)->sacked; in tcp_mark_skb_lost()
1144 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; in tcp_mark_skb_lost()
1145 tp->retrans_out -= tcp_skb_pcount(skb); in tcp_mark_skb_lost()
1151 tp->lost_out += tcp_skb_pcount(skb); in tcp_mark_skb_lost()
1152 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; in tcp_mark_skb_lost()
1161 tp->delivered += delivered; in tcp_count_delivered()
1163 tp->delivered_ce += delivered; in tcp_count_delivered()
1174 * 0 1 - orig segment is in flight.
1175 * S 0 - nothing flies, orig reached receiver.
1176 * L 0 - nothing flies, orig lost by net.
1177 * R 2 - both orig and retransmit are in flight.
1178 * L|R 1 - orig is lost, retransmit is in flight.
1179 * S|R 1 - orig reached receiver, retrans is still in flight.
1181 * but it is equivalent to plain S and code short-circuits it to S.
1182 * L|S is logically invalid, it would mean -1 packet in flight 8))
1192 * 4. D-SACK added new rule: D-SACK changes any tag to S.
1199 * --------------------
1204 * ever retransmitted -> reordering. Alas, we cannot use it
1206 * 2. The last flaw is solved with D-SACK. D-SACK arrives
1207 * for retransmitted and already SACKed segment -> reordering..
1212 * ----------------------
1227 * Implements also blockage to start_seq wrap-around. Problem lies in the
1233 * <- outs wnd -> <- wrapzone ->
1236 * |<------------+------+----- TCP seqno space --------------+---------->|
1237 * ...-- <2^31 ->| |<--------...
1238 * ...---- >2^31 ------>| |<--------...
1243 * snd_nxt wrap -> snd_una region will then become "well defined", i.e.,
1246 * With D-SACK the lower bound is extended to cover sequence space below
1248 * again, D-SACK block must not to go across snd_una (for the same reason as
1250 * ends, TCP might receive valid D-SACKs below that. As long as they reside
1253 * theoretical ones), the D-SACK will nicely cross that boundary due to skb
1256 * the exact amount is rather hard to quantify. However, tp->max_window can
1263 if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq)) in tcp_is_sackblock_valid()
1266 /* Nasty start_seq wrap-around check (see comments above) */ in tcp_is_sackblock_valid()
1267 if (!before(start_seq, tp->snd_nxt)) in tcp_is_sackblock_valid()
1270 /* In outstanding window? ...This is valid exit for D-SACKs too. in tcp_is_sackblock_valid()
1271 * start_seq == snd_una is non-sensical (see comments above) in tcp_is_sackblock_valid()
1273 if (after(start_seq, tp->snd_una)) in tcp_is_sackblock_valid()
1276 if (!is_dsack || !tp->undo_marker) in tcp_is_sackblock_valid()
1279 /* ...Then it's D-SACK, and must reside below snd_una completely */ in tcp_is_sackblock_valid()
1280 if (after(end_seq, tp->snd_una)) in tcp_is_sackblock_valid()
1283 if (!before(start_seq, tp->undo_marker)) in tcp_is_sackblock_valid()
1287 if (!after(end_seq, tp->undo_marker)) in tcp_is_sackblock_valid()
1293 return !before(start_seq, end_seq - tp->max_window); in tcp_is_sackblock_valid()
1305 if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) { in tcp_check_dsack()
1326 /* D-SACK for already forgotten data... Do dumb counting. */ in tcp_check_dsack()
1327 if (tp->undo_marker && tp->undo_retrans > 0 && in tcp_check_dsack()
1329 after(end_seq_0, tp->undo_marker)) in tcp_check_dsack()
1330 tp->undo_retrans = max_t(int, 0, tp->undo_retrans - dup_segs); in tcp_check_dsack()
1351 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && in tcp_match_skb_to_sack()
1352 !before(end_seq, TCP_SKB_CB(skb)->end_seq); in tcp_match_skb_to_sack()
1355 after(TCP_SKB_CB(skb)->end_seq, start_seq)) { in tcp_match_skb_to_sack()
1357 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); in tcp_match_skb_to_sack()
1360 pkt_len = start_seq - TCP_SKB_CB(skb)->seq; in tcp_match_skb_to_sack()
1364 pkt_len = end_seq - TCP_SKB_CB(skb)->seq; in tcp_match_skb_to_sack()
1366 return -EINVAL; in tcp_match_skb_to_sack()
1379 if (pkt_len >= skb->len && !in_sack) in tcp_match_skb_to_sack()
1391 /* Mark the given newly-SACKed range as such, adjusting counters and hints. */
1400 /* Account D-SACK for retransmitted packet. */ in tcp_sacktag_one()
1402 if (tp->undo_marker && tp->undo_retrans > 0 && in tcp_sacktag_one()
1403 after(end_seq, tp->undo_marker)) in tcp_sacktag_one()
1404 tp->undo_retrans = max_t(int, 0, tp->undo_retrans - pcount); in tcp_sacktag_one()
1406 before(start_seq, state->reord)) in tcp_sacktag_one()
1407 state->reord = start_seq; in tcp_sacktag_one()
1411 if (!after(end_seq, tp->snd_una)) in tcp_sacktag_one()
1424 tp->lost_out -= pcount; in tcp_sacktag_one()
1425 tp->retrans_out -= pcount; in tcp_sacktag_one()
1434 before(start_seq, state->reord)) in tcp_sacktag_one()
1435 state->reord = start_seq; in tcp_sacktag_one()
1437 if (!after(end_seq, tp->high_seq)) in tcp_sacktag_one()
1438 state->flag |= FLAG_ORIG_SACK_ACKED; in tcp_sacktag_one()
1439 if (state->first_sackt == 0) in tcp_sacktag_one()
1440 state->first_sackt = xmit_time; in tcp_sacktag_one()
1441 state->last_sackt = xmit_time; in tcp_sacktag_one()
1446 tp->lost_out -= pcount; in tcp_sacktag_one()
1451 state->flag |= FLAG_DATA_SACKED; in tcp_sacktag_one()
1452 tp->sacked_out += pcount; in tcp_sacktag_one()
1453 /* Out-of-order packets delivered */ in tcp_sacktag_one()
1454 state->sack_delivered += pcount; in tcp_sacktag_one()
1457 if (tp->lost_skb_hint && in tcp_sacktag_one()
1458 before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq)) in tcp_sacktag_one()
1459 tp->lost_cnt_hint += pcount; in tcp_sacktag_one()
1462 /* D-SACK. We can detect redundant retransmission in S|R and plain R in tcp_sacktag_one()
1468 tp->retrans_out -= pcount; in tcp_sacktag_one()
1474 /* Shift newly-SACKed bytes from this skb to the immediately previous
1475 * already-SACKed sk_buff. Mark the newly-SACKed bytes as such.
1484 u32 start_seq = TCP_SKB_CB(skb)->seq; /* start of newly-SACKed */ in tcp_shifted_skb()
1485 u32 end_seq = start_seq + shifted; /* end of newly-SACKed */ in tcp_shifted_skb()
1495 tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked, in tcp_shifted_skb()
1498 tcp_rate_skb_delivered(sk, skb, state->rate); in tcp_shifted_skb()
1500 if (skb == tp->lost_skb_hint) in tcp_shifted_skb()
1501 tp->lost_cnt_hint += pcount; in tcp_shifted_skb()
1503 TCP_SKB_CB(prev)->end_seq += shifted; in tcp_shifted_skb()
1504 TCP_SKB_CB(skb)->seq += shifted; in tcp_shifted_skb()
1508 tcp_skb_pcount_add(skb, -pcount); in tcp_shifted_skb()
1515 if (!TCP_SKB_CB(prev)->tcp_gso_size) in tcp_shifted_skb()
1516 TCP_SKB_CB(prev)->tcp_gso_size = mss; in tcp_shifted_skb()
1520 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_shifted_skb()
1523 TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS); in tcp_shifted_skb()
1525 if (skb->len > 0) { in tcp_shifted_skb()
1531 /* Whole SKB was eaten :-) */ in tcp_shifted_skb()
1533 if (skb == tp->retransmit_skb_hint) in tcp_shifted_skb()
1534 tp->retransmit_skb_hint = prev; in tcp_shifted_skb()
1535 if (skb == tp->lost_skb_hint) { in tcp_shifted_skb()
1536 tp->lost_skb_hint = prev; in tcp_shifted_skb()
1537 tp->lost_cnt_hint -= tcp_skb_pcount(prev); in tcp_shifted_skb()
1540 TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; in tcp_shifted_skb()
1541 TCP_SKB_CB(prev)->eor = TCP_SKB_CB(skb)->eor; in tcp_shifted_skb()
1542 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_shifted_skb()
1543 TCP_SKB_CB(prev)->end_seq++; in tcp_shifted_skb()
1549 if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp)) in tcp_shifted_skb()
1550 TCP_SKB_CB(prev)->tx.delivered_mstamp = 0; in tcp_shifted_skb()
1560 * something-or-zero which complicates things
1564 return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb); in tcp_skb_seglen()
1577 * Since TCP_SKB_CB(skb)->tcp_gso_segs is 16 bits, we need in tcp_skb_shift()
1581 if (unlikely(to->len + shiftlen >= 65535 * TCP_MIN_GSO_SIZE)) in tcp_skb_shift()
1605 (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS) in tcp_shift_skb_data()
1610 if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) in tcp_shift_skb_data()
1618 if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) in tcp_shift_skb_data()
1624 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && in tcp_shift_skb_data()
1625 !before(end_seq, TCP_SKB_CB(skb)->end_seq); in tcp_shift_skb_data()
1628 len = skb->len; in tcp_shift_skb_data()
1638 if (!after(TCP_SKB_CB(skb)->end_seq, start_seq)) in tcp_shift_skb_data()
1640 /* CHECKME: This is non-MSS split case only?, this will in tcp_shift_skb_data()
1647 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); in tcp_shift_skb_data()
1650 * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)), in tcp_shift_skb_data()
1654 * previously. We could try merging non-SACKed ones in tcp_shift_skb_data()
1663 len = end_seq - TCP_SKB_CB(skb)->seq; in tcp_shift_skb_data()
1665 BUG_ON(len > skb->len); in tcp_shift_skb_data()
1689 /* tcp_sacktag_one() won't SACK-tag ranges below snd_una */ in tcp_shift_skb_data()
1690 if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una)) in tcp_shift_skb_data()
1706 ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) || in tcp_shift_skb_data()
1712 len = skb->len; in tcp_shift_skb_data()
1742 /* queue is in-order => we can short-circuit the walk early */ in tcp_sacktag_walk()
1743 if (!before(TCP_SKB_CB(skb)->seq, end_seq)) in tcp_sacktag_walk()
1747 before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) { in tcp_sacktag_walk()
1749 next_dup->start_seq, in tcp_sacktag_walk()
1750 next_dup->end_seq); in tcp_sacktag_walk()
1780 TCP_SKB_CB(skb)->sacked = in tcp_sacktag_walk()
1783 TCP_SKB_CB(skb)->sacked, in tcp_sacktag_walk()
1784 TCP_SKB_CB(skb)->seq, in tcp_sacktag_walk()
1785 TCP_SKB_CB(skb)->end_seq, in tcp_sacktag_walk()
1789 tcp_rate_skb_delivered(sk, skb, state->rate); in tcp_sacktag_walk()
1790 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_sacktag_walk()
1791 list_del_init(&skb->tcp_tsorted_anchor); in tcp_sacktag_walk()
1793 if (!before(TCP_SKB_CB(skb)->seq, in tcp_sacktag_walk()
1803 struct rb_node *parent, **p = &sk->tcp_rtx_queue.rb_node; in tcp_sacktag_bsearch()
1809 if (before(seq, TCP_SKB_CB(skb)->seq)) { in tcp_sacktag_bsearch()
1810 p = &parent->rb_left; in tcp_sacktag_bsearch()
1813 if (!before(seq, TCP_SKB_CB(skb)->end_seq)) { in tcp_sacktag_bsearch()
1814 p = &parent->rb_right; in tcp_sacktag_bsearch()
1825 if (skb && after(TCP_SKB_CB(skb)->seq, skip_to_seq)) in tcp_sacktag_skip()
1840 if (before(next_dup->start_seq, skip_to_seq)) { in tcp_maybe_skipping_dsack()
1841 skb = tcp_sacktag_skip(skb, sk, next_dup->start_seq); in tcp_maybe_skipping_dsack()
1843 next_dup->start_seq, next_dup->end_seq, in tcp_maybe_skipping_dsack()
1852 return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); in tcp_sack_cache_ok()
1861 TCP_SKB_CB(ack_skb)->sacked); in tcp_sacktag_write_queue()
1866 int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3); in tcp_sacktag_write_queue()
1872 state->flag = 0; in tcp_sacktag_write_queue()
1873 state->reord = tp->snd_nxt; in tcp_sacktag_write_queue()
1875 if (!tp->sacked_out) in tcp_sacktag_write_queue()
1885 if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window)) in tcp_sacktag_write_queue()
1888 if (!tp->packets_out) in tcp_sacktag_write_queue()
1905 if (!tp->undo_marker) in tcp_sacktag_write_queue()
1911 if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) && in tcp_sacktag_write_queue()
1912 !after(sp[used_sacks].end_seq, tp->snd_una)) in tcp_sacktag_write_queue()
1919 first_sack_index = -1; in tcp_sacktag_write_queue()
1926 first_sack_index = -1; in tcp_sacktag_write_queue()
1934 for (i = used_sacks - 1; i > 0; i--) { in tcp_sacktag_write_queue()
1946 state->mss_now = tcp_current_mss(sk); in tcp_sacktag_write_queue()
1950 if (!tp->sacked_out) { in tcp_sacktag_write_queue()
1952 cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); in tcp_sacktag_write_queue()
1954 cache = tp->recv_sack_cache; in tcp_sacktag_write_queue()
1956 while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq && in tcp_sacktag_write_queue()
1957 !cache->end_seq) in tcp_sacktag_write_queue()
1972 !before(start_seq, cache->end_seq)) in tcp_sacktag_write_queue()
1977 after(end_seq, cache->start_seq)) { in tcp_sacktag_write_queue()
1980 if (before(start_seq, cache->start_seq)) { in tcp_sacktag_write_queue()
1985 cache->start_seq, in tcp_sacktag_write_queue()
1990 if (!after(end_seq, cache->end_seq)) in tcp_sacktag_write_queue()
1995 cache->end_seq); in tcp_sacktag_write_queue()
1998 if (tcp_highest_sack_seq(tp) == cache->end_seq) { in tcp_sacktag_write_queue()
2007 skb = tcp_sacktag_skip(skb, sk, cache->end_seq); in tcp_sacktag_write_queue()
2029 for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) { in tcp_sacktag_write_queue()
2030 tp->recv_sack_cache[i].start_seq = 0; in tcp_sacktag_write_queue()
2031 tp->recv_sack_cache[i].end_seq = 0; in tcp_sacktag_write_queue()
2034 tp->recv_sack_cache[i++] = sp[j]; in tcp_sacktag_write_queue()
2036 if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss || tp->undo_marker) in tcp_sacktag_write_queue()
2037 tcp_check_sack_reordering(sk, state->reord, 0); in tcp_sacktag_write_queue()
2043 WARN_ON((int)tp->sacked_out < 0); in tcp_sacktag_write_queue()
2044 WARN_ON((int)tp->lost_out < 0); in tcp_sacktag_write_queue()
2045 WARN_ON((int)tp->retrans_out < 0); in tcp_sacktag_write_queue()
2048 return state->flag; in tcp_sacktag_write_queue()
2058 holes = max(tp->lost_out, 1U); in tcp_limit_reno_sacked()
2059 holes = min(holes, tp->packets_out); in tcp_limit_reno_sacked()
2061 if ((tp->sacked_out + holes) > tp->packets_out) { in tcp_limit_reno_sacked()
2062 tp->sacked_out = tp->packets_out - holes; in tcp_limit_reno_sacked()
2079 tp->reordering = min_t(u32, tp->packets_out + addend, in tcp_check_reno_reordering()
2080 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering)); in tcp_check_reno_reordering()
2081 tp->reord_seen++; in tcp_check_reno_reordering()
2091 u32 prior_sacked = tp->sacked_out; in tcp_add_reno_sack()
2094 tp->sacked_out += num_dupack; in tcp_add_reno_sack()
2096 delivered = tp->sacked_out - prior_sacked; in tcp_add_reno_sack()
2111 tcp_count_delivered(tp, max_t(int, acked - tp->sacked_out, 1), in tcp_remove_reno_sacks()
2113 if (acked - 1 >= tp->sacked_out) in tcp_remove_reno_sacks()
2114 tp->sacked_out = 0; in tcp_remove_reno_sacks()
2116 tp->sacked_out -= acked - 1; in tcp_remove_reno_sacks()
2124 tp->sacked_out = 0; in tcp_reset_reno_sack()
2129 tp->retrans_out = 0; in tcp_clear_retrans()
2130 tp->lost_out = 0; in tcp_clear_retrans()
2131 tp->undo_marker = 0; in tcp_clear_retrans()
2132 tp->undo_retrans = -1; in tcp_clear_retrans()
2133 tp->sacked_out = 0; in tcp_clear_retrans()
2134 tp->rto_stamp = 0; in tcp_clear_retrans()
2135 tp->total_rto = 0; in tcp_clear_retrans()
2136 tp->total_rto_recoveries = 0; in tcp_clear_retrans()
2137 tp->total_rto_time = 0; in tcp_clear_retrans()
2142 tp->undo_marker = tp->snd_una; in tcp_init_undo()
2146 tp->undo_retrans = tp->retrans_out; in tcp_init_undo()
2148 if (tp->tlp_high_seq && tp->tlp_retrans) in tcp_init_undo()
2149 tp->undo_retrans++; in tcp_init_undo()
2151 if (!tp->undo_retrans) in tcp_init_undo()
2152 tp->undo_retrans = -1; in tcp_init_undo()
2157 return READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_recovery) & in tcp_is_rack()
2172 is_reneg = head && (TCP_SKB_CB(head)->sacked & TCPCB_SACKED_ACKED); in tcp_timeout_mark_lost()
2175 tp->sacked_out = 0; in tcp_timeout_mark_lost()
2177 tp->is_sack_reneg = 1; in tcp_timeout_mark_lost()
2185 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED; in tcp_timeout_mark_lost()
2201 bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery; in tcp_enter_loss()
2207 if (icsk->icsk_ca_state <= TCP_CA_Disorder || in tcp_enter_loss()
2208 !after(tp->high_seq, tp->snd_una) || in tcp_enter_loss()
2209 (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) { in tcp_enter_loss()
2210 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_enter_loss()
2211 tp->prior_cwnd = tcp_snd_cwnd(tp); in tcp_enter_loss()
2212 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk); in tcp_enter_loss()
2217 tp->snd_cwnd_cnt = 0; in tcp_enter_loss()
2218 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_enter_loss()
2221 * suggests that the degree of reordering is over-estimated. in tcp_enter_loss()
2223 reordering = READ_ONCE(net->ipv4.sysctl_tcp_reordering); in tcp_enter_loss()
2224 if (icsk->icsk_ca_state <= TCP_CA_Disorder && in tcp_enter_loss()
2225 tp->sacked_out >= reordering) in tcp_enter_loss()
2226 tp->reordering = min_t(unsigned int, tp->reordering, in tcp_enter_loss()
2230 tp->high_seq = tp->snd_nxt; in tcp_enter_loss()
2231 tp->tlp_high_seq = 0; in tcp_enter_loss()
2234 /* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous in tcp_enter_loss()
2236 * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing in tcp_enter_loss()
2238 tp->frto = READ_ONCE(net->ipv4.sysctl_tcp_frto) && in tcp_enter_loss()
2239 (new_recovery || icsk->icsk_retransmits) && in tcp_enter_loss()
2240 !inet_csk(sk)->icsk_mtup.probe_size; in tcp_enter_loss()
2249 * little time (max(RTT/2, 10ms)) to send us some more ACKs that will
2258 unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4), in tcp_check_sack_reneging()
2276 * but when e.g. out-of-window ACKs or packet duplication occurs,
2282 return tp->sacked_out + 1; in tcp_dupack_heuristics()
2286 * --------------------------------------
2295 * "Recovery" CWND was reduced, we are fast-retransmitting.
2299 * - each incoming ACK, if state is not "Open"
2300 * - when arrived ACK is unusual, namely:
2307 * in_flight = packets_out - left_out + retrans_out
2309 * packets_out is SND.NXT-SND.UNA counted in packets.
2343 * RACK (draft-ietf-tcpm-rack-01): it is a newer algorithm
2344 * (2017-) that checks timing instead of counting DUPACKs.
2346 * after RTT + reordering_window, where both metrics are
2370 * timestamps and D-SACKs. When we detect that some segments were
2387 if (tp->lost_out) in tcp_time_to_recover()
2390 /* Not-A-Trick#2 : Classic rule... */ in tcp_time_to_recover()
2391 if (!tcp_is_rack(sk) && tcp_dupack_heuristics(tp) > tp->reordering) in tcp_time_to_recover()
2399 * has at least tp->reordering SACKed seqments above it; "packets" refers to
2408 const u32 loss_high = tp->snd_nxt; in tcp_mark_head_lost()
2410 WARN_ON(packets > tp->packets_out); in tcp_mark_head_lost()
2411 skb = tp->lost_skb_hint; in tcp_mark_head_lost()
2414 if (mark_head && after(TCP_SKB_CB(skb)->seq, tp->snd_una)) in tcp_mark_head_lost()
2416 cnt = tp->lost_cnt_hint; in tcp_mark_head_lost()
2425 tp->lost_skb_hint = skb; in tcp_mark_head_lost()
2426 tp->lost_cnt_hint = cnt; in tcp_mark_head_lost()
2428 if (after(TCP_SKB_CB(skb)->end_seq, loss_high)) in tcp_mark_head_lost()
2431 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_mark_head_lost()
2437 if (!(TCP_SKB_CB(skb)->sacked & TCPCB_LOST)) in tcp_mark_head_lost()
2453 int sacked_upto = tp->sacked_out - tp->reordering; in tcp_update_scoreboard()
2463 return tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && in tcp_tsopt_ecr_before()
2464 before(tp->rx_opt.rcv_tsecr, when); in tcp_tsopt_ecr_before()
2473 return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) && in tcp_skb_spurious_retrans()
2474 tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb)); in tcp_skb_spurious_retrans()
2484 if (tp->retrans_stamp && in tcp_packet_delayed()
2485 tcp_tsopt_ecr_before(tp, tp->retrans_stamp)) in tcp_packet_delayed()
2493 if (!tp->retrans_stamp && /* no record of a retransmit/SYN? */ in tcp_packet_delayed()
2494 sk->sk_state != TCP_SYN_SENT) /* not the FLAG_SYN_ACKED case? */ in tcp_packet_delayed()
2504 * tp->retrans_out, however, that kind of assumptions doesn't consider
2521 if (tp->retrans_out) in tcp_any_retrans_done()
2525 if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS)) in tcp_any_retrans_done()
2533 * retransmits_timed_out() and timestamp-undo are using the correct value.
2538 tcp_sk(sk)->retrans_stamp = 0; in tcp_retrans_stamp_cleanup()
2547 if (sk->sk_family == AF_INET) { in DBGUNDO()
2550 &inet->inet_daddr, ntohs(inet->inet_dport), in DBGUNDO()
2552 tp->snd_ssthresh, tp->prior_ssthresh, in DBGUNDO()
2553 tp->packets_out); in DBGUNDO()
2556 else if (sk->sk_family == AF_INET6) { in DBGUNDO()
2559 &sk->sk_v6_daddr, ntohs(inet->inet_dport), in DBGUNDO()
2561 tp->snd_ssthresh, tp->prior_ssthresh, in DBGUNDO()
2562 tp->packets_out); in DBGUNDO()
2575 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { in tcp_undo_cwnd_reduction()
2576 TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST; in tcp_undo_cwnd_reduction()
2578 tp->lost_out = 0; in tcp_undo_cwnd_reduction()
2582 if (tp->prior_ssthresh) { in tcp_undo_cwnd_reduction()
2585 tcp_snd_cwnd_set(tp, icsk->icsk_ca_ops->undo_cwnd(sk)); in tcp_undo_cwnd_reduction()
2587 if (tp->prior_ssthresh > tp->snd_ssthresh) { in tcp_undo_cwnd_reduction()
2588 tp->snd_ssthresh = tp->prior_ssthresh; in tcp_undo_cwnd_reduction()
2592 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_undo_cwnd_reduction()
2593 tp->undo_marker = 0; in tcp_undo_cwnd_reduction()
2594 tp->rack.advanced = 1; /* Force RACK to re-exam losses */ in tcp_undo_cwnd_reduction()
2599 return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp)); in tcp_may_undo()
2606 if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) { in tcp_is_non_sack_preventing_reopen()
2611 tp->retrans_stamp = 0; in tcp_is_non_sack_preventing_reopen()
2628 DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans"); in tcp_try_undo_recovery()
2630 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss) in tcp_try_undo_recovery()
2636 } else if (tp->rack.reo_wnd_persist) { in tcp_try_undo_recovery()
2637 tp->rack.reo_wnd_persist--; in tcp_try_undo_recovery()
2642 tp->is_sack_reneg = 0; in tcp_try_undo_recovery()
2646 /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
2651 if (tp->undo_marker && !tp->undo_retrans) { in tcp_try_undo_dsack()
2652 tp->rack.reo_wnd_persist = min(TCP_RACK_RECOVERY_THRESH, in tcp_try_undo_dsack()
2653 tp->rack.reo_wnd_persist + 1); in tcp_try_undo_dsack()
2654 DBGUNDO(sk, "D-SACK"); in tcp_try_undo_dsack()
2662 /* Undo during loss recovery after partial ACK or using F-RTO. */
2675 inet_csk(sk)->icsk_retransmits = 0; in tcp_try_undo_loss()
2680 tp->is_sack_reneg = 0; in tcp_try_undo_loss()
2691 * cwnd reductions across a full RTT.
2700 tp->high_seq = tp->snd_nxt; in tcp_init_cwnd_reduction()
2701 tp->tlp_high_seq = 0; in tcp_init_cwnd_reduction()
2702 tp->snd_cwnd_cnt = 0; in tcp_init_cwnd_reduction()
2703 tp->prior_cwnd = tcp_snd_cwnd(tp); in tcp_init_cwnd_reduction()
2704 tp->prr_delivered = 0; in tcp_init_cwnd_reduction()
2705 tp->prr_out = 0; in tcp_init_cwnd_reduction()
2706 tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk); in tcp_init_cwnd_reduction()
2714 int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); in tcp_cwnd_reduction()
2716 if (newly_acked_sacked <= 0 || WARN_ON_ONCE(!tp->prior_cwnd)) in tcp_cwnd_reduction()
2719 tp->prr_delivered += newly_acked_sacked; in tcp_cwnd_reduction()
2721 u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered + in tcp_cwnd_reduction()
2722 tp->prior_cwnd - 1; in tcp_cwnd_reduction()
2723 sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out; in tcp_cwnd_reduction()
2725 sndcnt = max_t(int, tp->prr_delivered - tp->prr_out, in tcp_cwnd_reduction()
2732 sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1)); in tcp_cwnd_reduction()
2740 if (inet_csk(sk)->icsk_ca_ops->cong_control) in tcp_end_cwnd_reduction()
2744 if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH && in tcp_end_cwnd_reduction()
2745 (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) { in tcp_end_cwnd_reduction()
2746 tcp_snd_cwnd_set(tp, tp->snd_ssthresh); in tcp_end_cwnd_reduction()
2747 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_end_cwnd_reduction()
2757 tp->prior_ssthresh = 0; in tcp_enter_cwr()
2758 if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { in tcp_enter_cwr()
2759 tp->undo_marker = 0; in tcp_enter_cwr()
2774 if (inet_csk(sk)->icsk_ca_state != state) { in tcp_try_keep_open()
2776 tp->high_seq = tp->snd_nxt; in tcp_try_keep_open()
2787 tp->retrans_stamp = 0; in tcp_try_to_open()
2792 if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) { in tcp_try_to_open()
2801 icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1; in tcp_mtup_probe_failed()
2802 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_probe_failed()
2812 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_mtup_probe_success()
2814 val = (u64)tcp_snd_cwnd(tp) * tcp_mss_to_mtu(sk, tp->mss_cache); in tcp_mtup_probe_success()
2815 do_div(val, icsk->icsk_mtup.probe_size); in tcp_mtup_probe_success()
2819 tp->snd_cwnd_cnt = 0; in tcp_mtup_probe_success()
2820 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_mtup_probe_success()
2821 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_mtup_probe_success()
2823 icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size; in tcp_mtup_probe_success()
2824 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_probe_success()
2825 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_mtup_probe_success()
2833 * non-zero value (and may do so in a later calling context due to TSQ), we
2844 if (icsk->icsk_ca_state != TCP_CA_Loss) { in tcp_non_congestion_loss_retransmit()
2845 tp->high_seq = tp->snd_nxt; in tcp_non_congestion_loss_retransmit()
2846 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_non_congestion_loss_retransmit()
2847 tp->prior_ssthresh = 0; in tcp_non_congestion_loss_retransmit()
2848 tp->undo_marker = 0; in tcp_non_congestion_loss_retransmit()
2872 * -1 to mark all frames as lost, otherwise compute the current MSS. in tcp_simple_retransmit()
2874 if (tp->syn_data && sk->sk_state == TCP_SYN_SENT) in tcp_simple_retransmit()
2875 mss = -1; in tcp_simple_retransmit()
2879 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { in tcp_simple_retransmit()
2886 if (!tp->lost_out) in tcp_simple_retransmit()
2918 tp->prior_ssthresh = 0; in tcp_enter_recovery()
2923 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_enter_recovery()
2931 if (tp->rto_stamp) { in tcp_update_rto_time()
2932 tp->total_rto_time += tcp_time_stamp_ms(tp) - tp->rto_stamp; in tcp_update_rto_time()
2933 tp->rto_stamp = 0; in tcp_update_rto_time()
2944 bool recovered = !before(tp->snd_una, tp->high_seq); in tcp_process_loss()
2946 if ((flag & FLAG_SND_UNA_ADVANCED || rcu_access_pointer(tp->fastopen_rsk)) && in tcp_process_loss()
2950 if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */ in tcp_process_loss()
2952 * lost, i.e., never-retransmitted data are (s)acked. in tcp_process_loss()
2958 if (after(tp->snd_nxt, tp->high_seq)) { in tcp_process_loss()
2960 tp->frto = 0; /* Step 3.a. loss was real */ in tcp_process_loss()
2962 tp->high_seq = tp->snd_nxt; in tcp_process_loss()
2968 after(tcp_wnd_end(tp), tp->snd_nxt)) { in tcp_process_loss()
2972 tp->frto = 0; in tcp_process_loss()
2977 /* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */ in tcp_process_loss()
2982 /* A Reno DUPACK means new data in F-RTO step 2.b above are in tcp_process_loss()
2985 if (after(tp->snd_nxt, tp->high_seq) && num_dupack) in tcp_process_loss()
2998 tp->snd_una + tp->reordering * tp->mss_cache); in tcp_force_fast_retransmit()
3007 if (tp->undo_marker && tcp_packet_delayed(tp)) { in tcp_try_undo_partial()
3018 if (tp->retrans_out) in tcp_try_undo_partial()
3022 tp->retrans_stamp = 0; in tcp_try_undo_partial()
3045 u32 prior_retrans = tp->retrans_out; in tcp_identify_packet_loss()
3049 if (prior_retrans > tp->retrans_out) in tcp_identify_packet_loss()
3054 /* Process an event, which can update packets-in-flight not trivially.
3076 if (!tp->packets_out && tp->sacked_out) in tcp_fastretrans_alert()
3077 tp->sacked_out = 0; in tcp_fastretrans_alert()
3082 tp->prior_ssthresh = 0; in tcp_fastretrans_alert()
3093 if (icsk->icsk_ca_state == TCP_CA_Open) { in tcp_fastretrans_alert()
3094 WARN_ON(tp->retrans_out != 0 && !tp->syn_data); in tcp_fastretrans_alert()
3095 tp->retrans_stamp = 0; in tcp_fastretrans_alert()
3096 } else if (!before(tp->snd_una, tp->high_seq)) { in tcp_fastretrans_alert()
3097 switch (icsk->icsk_ca_state) { in tcp_fastretrans_alert()
3101 if (tp->snd_una != tp->high_seq) { in tcp_fastretrans_alert()
3118 switch (icsk->icsk_ca_state) { in tcp_fastretrans_alert()
3130 if (icsk->icsk_ca_state != TCP_CA_Recovery) { in tcp_fastretrans_alert()
3141 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_fastretrans_alert()
3144 if (!(icsk->icsk_ca_state == TCP_CA_Open || in tcp_fastretrans_alert()
3156 if (icsk->icsk_ca_state <= TCP_CA_Disorder) in tcp_fastretrans_alert()
3166 if (icsk->icsk_ca_state < TCP_CA_CWR && in tcp_fastretrans_alert()
3167 icsk->icsk_mtup.probe_size && in tcp_fastretrans_alert()
3168 tp->snd_una == tp->mtu_probe.probe_seq_start) { in tcp_fastretrans_alert()
3188 u32 wlen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen) * HZ; in tcp_update_rtt_min()
3198 minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32, in tcp_update_rtt_min()
3208 /* Prefer RTT measured from ACK's timing to TS-ECR. This is because in tcp_ack_update_rtt()
3209 * broken middle-boxes or peers may corrupt TS-ECR fields. But in tcp_ack_update_rtt()
3210 * Karn's algorithm forbids taking RTT if some retransmitted data in tcp_ack_update_rtt()
3217 * update the averaged RTT measurement only if the segment in tcp_ack_update_rtt()
3220 * See draft-ietf-tcplw-high-performance-00, section 3.3. in tcp_ack_update_rtt()
3222 if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && in tcp_ack_update_rtt()
3223 tp->rx_opt.rcv_tsecr && flag & FLAG_ACKED) in tcp_ack_update_rtt()
3226 rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */ in tcp_ack_update_rtt()
3231 * always taken together with ACK, SACK, or TS-opts. Any negative in tcp_ack_update_rtt()
3238 /* RFC6298: only reset backoff on valid RTT measurement. */ in tcp_ack_update_rtt()
3239 inet_csk(sk)->icsk_backoff = 0; in tcp_ack_update_rtt()
3247 long rtt_us = -1L; in tcp_synack_rtt_meas()
3249 if (req && !req->num_retrans && tcp_rsk(req)->snt_synack) in tcp_synack_rtt_meas()
3250 rtt_us = tcp_stamp_us_delta(tcp_clock_us(), tcp_rsk(req)->snt_synack); in tcp_synack_rtt_meas()
3252 tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, rtt_us, -1L, rtt_us, &rs); in tcp_synack_rtt_meas()
3260 icsk->icsk_ca_ops->cong_avoid(sk, ack, acked); in tcp_cong_avoid()
3261 tcp_sk(sk)->snd_cwnd_stamp = tcp_jiffies32; in tcp_cong_avoid()
3264 /* Restart timer after forward progress on connection.
3265 * RFC2988 recommends to restart timer to now+rto.
3272 /* If the retrans timer is currently being used by Fast Open in tcp_rearm_rto()
3273 * for SYN-ACK retrans purpose, stay put. in tcp_rearm_rto()
3275 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_rearm_rto()
3278 if (!tp->packets_out) { in tcp_rearm_rto()
3281 u32 rto = inet_csk(sk)->icsk_rto; in tcp_rearm_rto()
3283 if (icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || in tcp_rearm_rto()
3284 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { in tcp_rearm_rto()
3287 * when the retrans timer fires and is rescheduled. in tcp_rearm_rto()
3309 BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)); in tcp_tso_acked()
3312 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in tcp_tso_acked()
3314 packets_acked -= tcp_skb_pcount(skb); in tcp_tso_acked()
3318 BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)); in tcp_tso_acked()
3329 /* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */ in tcp_ack_tstamp()
3330 if (likely(!TCP_SKB_CB(skb)->txstamp_ack)) in tcp_ack_tstamp()
3334 if (!before(shinfo->tskey, prior_snd_una) && in tcp_ack_tstamp()
3335 before(shinfo->tskey, tcp_sk(sk)->snd_una)) { in tcp_ack_tstamp()
3353 u32 prior_sacked = tp->sacked_out; in tcp_clean_rtx_queue()
3354 u32 reord = tp->snd_nxt; /* lowest acked un-retx un-sacked seq */ in tcp_clean_rtx_queue()
3357 long sack_rtt_us = -1L; in tcp_clean_rtx_queue()
3358 long seq_rtt_us = -1L; in tcp_clean_rtx_queue()
3359 long ca_rtt_us = -1L; in tcp_clean_rtx_queue()
3366 for (skb = skb_rb_first(&sk->tcp_rtx_queue); skb; skb = next) { in tcp_clean_rtx_queue()
3368 const u32 start_seq = scb->seq; in tcp_clean_rtx_queue()
3369 u8 sacked = scb->sacked; in tcp_clean_rtx_queue()
3373 if (after(scb->end_seq, tp->snd_una)) { in tcp_clean_rtx_queue()
3375 !after(tp->snd_una, scb->seq)) in tcp_clean_rtx_queue()
3388 tp->retrans_out -= acked_pcount; in tcp_clean_rtx_queue()
3398 if (!after(scb->end_seq, tp->high_seq)) in tcp_clean_rtx_queue()
3403 tp->sacked_out -= acked_pcount; in tcp_clean_rtx_queue()
3407 tcp_rack_advance(tp, sacked, scb->end_seq, in tcp_clean_rtx_queue()
3411 tp->lost_out -= acked_pcount; in tcp_clean_rtx_queue()
3413 tp->packets_out -= acked_pcount; in tcp_clean_rtx_queue()
3415 tcp_rate_skb_delivered(sk, skb, sack->rate); in tcp_clean_rtx_queue()
3424 if (likely(!(scb->tcp_flags & TCPHDR_SYN))) { in tcp_clean_rtx_queue()
3428 tp->retrans_stamp = 0; in tcp_clean_rtx_queue()
3437 if (unlikely(skb == tp->retransmit_skb_hint)) in tcp_clean_rtx_queue()
3438 tp->retransmit_skb_hint = NULL; in tcp_clean_rtx_queue()
3439 if (unlikely(skb == tp->lost_skb_hint)) in tcp_clean_rtx_queue()
3440 tp->lost_skb_hint = NULL; in tcp_clean_rtx_queue()
3448 if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una))) in tcp_clean_rtx_queue()
3449 tp->snd_up = tp->snd_una; in tcp_clean_rtx_queue()
3453 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_clean_rtx_queue()
3458 seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt); in tcp_clean_rtx_queue()
3459 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt); in tcp_clean_rtx_queue()
3462 (tp->snd_una - prior_snd_una) < tp->mss_cache && in tcp_clean_rtx_queue()
3463 sack->rate->prior_delivered + 1 == tp->delivered && in tcp_clean_rtx_queue()
3467 * a receiver w/o out-of-order or CE events. in tcp_clean_rtx_queue()
3472 if (sack->first_sackt) { in tcp_clean_rtx_queue()
3473 sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt); in tcp_clean_rtx_queue()
3474 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->last_sackt); in tcp_clean_rtx_queue()
3477 ca_rtt_us, sack->rate); in tcp_clean_rtx_queue()
3480 flag |= FLAG_SET_XMIT_TIMER; /* set TLP or RTO timer */ in tcp_clean_rtx_queue()
3481 if (unlikely(icsk->icsk_mtup.probe_size && in tcp_clean_rtx_queue()
3482 !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) { in tcp_clean_rtx_queue()
3490 * retransmitted, non-SACK case cannot confirm that in tcp_clean_rtx_queue()
3500 /* Non-retransmitted hole got filled? That's reordering */ in tcp_clean_rtx_queue()
3504 delta = prior_sacked - tp->sacked_out; in tcp_clean_rtx_queue()
3505 tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta); in tcp_clean_rtx_queue()
3508 sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp, in tcp_clean_rtx_queue()
3510 /* Do not re-arm RTO if the sack RTT is measured from data sent in tcp_clean_rtx_queue()
3514 flag |= FLAG_SET_XMIT_TIMER; /* set TLP or RTO timer */ in tcp_clean_rtx_queue()
3517 if (icsk->icsk_ca_ops->pkts_acked) { in tcp_clean_rtx_queue()
3519 .rtt_us = sack->rate->rtt_us }; in tcp_clean_rtx_queue()
3521 sample.in_flight = tp->mss_cache * in tcp_clean_rtx_queue()
3522 (tp->delivered - sack->rate->prior_delivered); in tcp_clean_rtx_queue()
3523 icsk->icsk_ca_ops->pkts_acked(sk, &sample); in tcp_clean_rtx_queue()
3527 WARN_ON((int)tp->sacked_out < 0); in tcp_clean_rtx_queue()
3528 WARN_ON((int)tp->lost_out < 0); in tcp_clean_rtx_queue()
3529 WARN_ON((int)tp->retrans_out < 0); in tcp_clean_rtx_queue()
3530 if (!tp->packets_out && tcp_is_sack(tp)) { in tcp_clean_rtx_queue()
3532 if (tp->lost_out) { in tcp_clean_rtx_queue()
3534 tp->lost_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3535 tp->lost_out = 0; in tcp_clean_rtx_queue()
3537 if (tp->sacked_out) { in tcp_clean_rtx_queue()
3539 tp->sacked_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3540 tp->sacked_out = 0; in tcp_clean_rtx_queue()
3542 if (tp->retrans_out) { in tcp_clean_rtx_queue()
3544 tp->retrans_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3545 tp->retrans_out = 0; in tcp_clean_rtx_queue()
3561 if (!after(TCP_SKB_CB(head)->end_seq, tcp_wnd_end(tp))) { in tcp_ack_probe()
3562 icsk->icsk_backoff = 0; in tcp_ack_probe()
3563 icsk->icsk_probes_tstamp = 0; in tcp_ack_probe()
3579 inet_csk(sk)->icsk_ca_state != TCP_CA_Open; in tcp_ack_is_dubious()
3587 * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/ in tcp_may_raise_cwnd()
3591 if (tcp_sk(sk)->reordering > in tcp_may_raise_cwnd()
3592 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering)) in tcp_may_raise_cwnd()
3608 if (icsk->icsk_ca_ops->cong_control) { in tcp_cong_control()
3609 icsk->icsk_ca_ops->cong_control(sk, ack, flag, rs); in tcp_cong_control()
3615 tcp_cwnd_reduction(sk, acked_sacked, rs->losses, flag); in tcp_cong_control()
3630 return after(ack, tp->snd_una) || in tcp_may_update_window()
3631 after(ack_seq, tp->snd_wl1) || in tcp_may_update_window()
3632 (ack_seq == tp->snd_wl1 && (nwin > tp->snd_wnd || !nwin)); in tcp_may_update_window()
3643 ao = rcu_dereference_protected(tp->ao_info, in tcp_snd_sne_update()
3645 if (ao && ack < tp->snd_una) { in tcp_snd_sne_update()
3646 ao->snd_sne++; in tcp_snd_sne_update()
3647 trace_tcp_ao_snd_sne_update((struct sock *)tp, ao->snd_sne); in tcp_snd_sne_update()
3652 /* If we update tp->snd_una, also update tp->bytes_acked */
3655 u32 delta = ack - tp->snd_una; in tcp_snd_una_update()
3658 tp->bytes_acked += delta; in tcp_snd_una_update()
3660 tp->snd_una = ack; in tcp_snd_una_update()
3671 ao = rcu_dereference_protected(tp->ao_info, in tcp_rcv_sne_update()
3673 if (ao && seq < tp->rcv_nxt) { in tcp_rcv_sne_update()
3674 ao->rcv_sne++; in tcp_rcv_sne_update()
3675 trace_tcp_ao_rcv_sne_update((struct sock *)tp, ao->rcv_sne); in tcp_rcv_sne_update()
3680 /* If we update tp->rcv_nxt, also update tp->bytes_received */
3683 u32 delta = seq - tp->rcv_nxt; in tcp_rcv_nxt_update()
3686 tp->bytes_received += delta; in tcp_rcv_nxt_update()
3688 WRITE_ONCE(tp->rcv_nxt, seq); in tcp_rcv_nxt_update()
3693 * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
3701 u32 nwin = ntohs(tcp_hdr(skb)->window); in tcp_ack_update_window()
3703 if (likely(!tcp_hdr(skb)->syn)) in tcp_ack_update_window()
3704 nwin <<= tp->rx_opt.snd_wscale; in tcp_ack_update_window()
3710 if (tp->snd_wnd != nwin) { in tcp_ack_update_window()
3711 tp->snd_wnd = nwin; in tcp_ack_update_window()
3716 tp->pred_flags = 0; in tcp_ack_update_window()
3722 if (nwin > tp->max_window) { in tcp_ack_update_window()
3723 tp->max_window = nwin; in tcp_ack_update_window()
3724 tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie); in tcp_ack_update_window()
3741 s32 elapsed = (s32)(tcp_jiffies32 - val); in __tcp_oow_rate_limited()
3744 elapsed < READ_ONCE(net->ipv4.sysctl_tcp_invalid_ratelimit)) { in __tcp_oow_rate_limited()
3746 return true; /* rate-limited: don't send yet! */ in __tcp_oow_rate_limited()
3755 return false; /* not rate-limited: go ahead, send dupack now! */ in __tcp_oow_rate_limited()
3758 /* Return true if we're currently rate-limiting out-of-window ACKs and
3759 * thus shouldn't send a dupack right now. We rate-limit dupacks in
3760 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
3763 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
3769 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) && in tcp_oow_rate_limited()
3770 !tcp_hdr(skb)->syn) in tcp_oow_rate_limited()
3783 /* First check our per-socket dupack rate limit. */ in tcp_send_challenge_ack()
3786 &tp->last_oow_ack_time)) in tcp_send_challenge_ack()
3789 ack_limit = READ_ONCE(net->ipv4.sysctl_tcp_challenge_ack_limit); in tcp_send_challenge_ack()
3793 /* Then check host-wide RFC 5961 rate limit. */ in tcp_send_challenge_ack()
3795 if (now != READ_ONCE(net->ipv4.tcp_challenge_timestamp)) { in tcp_send_challenge_ack()
3798 WRITE_ONCE(net->ipv4.tcp_challenge_timestamp, now); in tcp_send_challenge_ack()
3799 WRITE_ONCE(net->ipv4.tcp_challenge_count, in tcp_send_challenge_ack()
3800 get_random_u32_inclusive(half, ack_limit + half - 1)); in tcp_send_challenge_ack()
3802 count = READ_ONCE(net->ipv4.tcp_challenge_count); in tcp_send_challenge_ack()
3804 WRITE_ONCE(net->ipv4.tcp_challenge_count, count - 1); in tcp_send_challenge_ack()
3813 tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval; in tcp_store_ts_recent()
3814 tp->rx_opt.ts_recent_stamp = ktime_get_seconds(); in tcp_store_ts_recent()
3819 if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) { in tcp_replace_ts_recent()
3822 * for pure ACK frames. -DaveM in tcp_replace_ts_recent()
3827 if (tcp_paws_check(&tp->rx_opt, 0)) in tcp_replace_ts_recent()
3833 * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack
3839 if (before(ack, tp->tlp_high_seq)) in tcp_process_tlp_ack()
3842 if (!tp->tlp_retrans) { in tcp_process_tlp_ack()
3844 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3847 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3848 } else if (after(ack, tp->tlp_high_seq)) { in tcp_process_tlp_ack()
3861 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3869 if (icsk->icsk_ca_ops->in_ack_event) in tcp_in_ack_event()
3870 icsk->icsk_ca_ops->in_ack_event(sk, flags); in tcp_in_ack_event()
3881 if (rexmit == REXMIT_NONE || sk->sk_state == TCP_SYN_SENT) in tcp_xmit_recovery()
3887 if (after(tp->snd_nxt, tp->high_seq)) in tcp_xmit_recovery()
3889 tp->frto = 0; in tcp_xmit_recovery()
3901 delivered = tp->delivered - prior_delivered; in tcp_newly_delivered()
3916 u32 prior_snd_una = tp->snd_una; in tcp_ack()
3917 bool is_sack_reneg = tp->is_sack_reneg; in tcp_ack()
3918 u32 ack_seq = TCP_SKB_CB(skb)->seq; in tcp_ack()
3919 u32 ack = TCP_SKB_CB(skb)->ack_seq; in tcp_ack()
3921 int prior_packets = tp->packets_out; in tcp_ack()
3922 u32 delivered = tp->delivered; in tcp_ack()
3923 u32 lost = tp->lost; in tcp_ack()
3932 prefetch(sk->tcp_rtx_queue.rb_node); in tcp_ack()
3941 max_window = min_t(u64, tp->max_window, tp->bytes_acked); in tcp_ack()
3943 if (before(ack, prior_snd_una - max_window)) { in tcp_ack()
3946 return -SKB_DROP_REASON_TCP_TOO_OLD_ACK; in tcp_ack()
3954 if (after(ack, tp->snd_nxt)) in tcp_ack()
3955 return -SKB_DROP_REASON_TCP_ACK_UNSENT_DATA; in tcp_ack()
3959 icsk->icsk_retransmits = 0; in tcp_ack()
3963 if (icsk->icsk_clean_acked) in tcp_ack()
3964 icsk->icsk_clean_acked(sk, ack); in tcp_ack()
3968 prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una; in tcp_ack()
3975 tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq); in tcp_ack()
3993 if (ack_seq != TCP_SKB_CB(skb)->end_seq) in tcp_ack()
4000 if (TCP_SKB_CB(skb)->sacked) in tcp_ack()
4024 * with widely-deployed TCP implementations that do this. in tcp_ack()
4031 WRITE_ONCE(sk->sk_err_soft, 0); in tcp_ack()
4032 icsk->icsk_probes_out = 0; in tcp_ack()
4033 tp->rcv_tstamp = tcp_jiffies32; in tcp_ack()
4043 if (tp->tlp_high_seq) in tcp_ack()
4052 num_dupack = max_t(u16, 1, skb_shinfo(skb)->gso_segs); in tcp_ack()
4058 /* If needed, reset TLP/RTO timer when RACK doesn't set. */ in tcp_ack()
4066 lost = tp->lost - lost; /* freshly marked lost */ in tcp_ack()
4086 if (tp->tlp_high_seq) in tcp_ack()
4094 if (TCP_SKB_CB(skb)->sacked) { in tcp_ack()
4110 /* Valid only in SYN or SYN-ACK with an even length. */ in tcp_parse_fastopen_option()
4116 memcpy(foc->val, cookie, len); in tcp_parse_fastopen_option()
4118 len = -1; in tcp_parse_fastopen_option()
4119 foc->len = len; in tcp_parse_fastopen_option()
4120 foc->exp = exp_opt; in tcp_parse_fastopen_option()
4130 if (th->syn && !(opsize & 1) && in smc_parse_options()
4133 opt_rx->smc_ok = 1; in smc_parse_options()
4147 int length = (th->doff * 4) - sizeof(struct tcphdr); in tcp_parse_mss_option()
4158 length--; in tcp_parse_mss_option()
4177 ptr += opsize - 2; in tcp_parse_mss_option()
4178 length -= opsize; in tcp_parse_mss_option()
4196 int length = (th->doff * 4) - sizeof(struct tcphdr); in tcp_parse_options()
4199 opt_rx->saw_tstamp = 0; in tcp_parse_options()
4200 opt_rx->saw_unknown = 0; in tcp_parse_options()
4210 length--; in tcp_parse_options()
4222 if (opsize == TCPOLEN_MSS && th->syn && !estab) { in tcp_parse_options()
4225 if (opt_rx->user_mss && in tcp_parse_options()
4226 opt_rx->user_mss < in_mss) in tcp_parse_options()
4227 in_mss = opt_rx->user_mss; in tcp_parse_options()
4228 opt_rx->mss_clamp = in_mss; in tcp_parse_options()
4233 if (opsize == TCPOLEN_WINDOW && th->syn && in tcp_parse_options()
4234 !estab && READ_ONCE(net->ipv4.sysctl_tcp_window_scaling)) { in tcp_parse_options()
4236 opt_rx->wscale_ok = 1; in tcp_parse_options()
4244 opt_rx->snd_wscale = snd_wscale; in tcp_parse_options()
4249 ((estab && opt_rx->tstamp_ok) || in tcp_parse_options()
4250 (!estab && READ_ONCE(net->ipv4.sysctl_tcp_timestamps)))) { in tcp_parse_options()
4251 opt_rx->saw_tstamp = 1; in tcp_parse_options()
4252 opt_rx->rcv_tsval = get_unaligned_be32(ptr); in tcp_parse_options()
4253 opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4); in tcp_parse_options()
4257 if (opsize == TCPOLEN_SACK_PERM && th->syn && in tcp_parse_options()
4258 !estab && READ_ONCE(net->ipv4.sysctl_tcp_sack)) { in tcp_parse_options()
4259 opt_rx->sack_ok = TCP_SACK_SEEN; in tcp_parse_options()
4266 !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) && in tcp_parse_options()
4267 opt_rx->sack_ok) { in tcp_parse_options()
4268 TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th; in tcp_parse_options()
4287 opsize - TCPOLEN_FASTOPEN_BASE, in tcp_parse_options()
4288 ptr, th->syn, foc, false); in tcp_parse_options()
4298 tcp_parse_fastopen_option(opsize - in tcp_parse_options()
4300 ptr + 2, th->syn, foc, true); in tcp_parse_options()
4307 opt_rx->saw_unknown = 1; in tcp_parse_options()
4311 opt_rx->saw_unknown = 1; in tcp_parse_options()
4313 ptr += opsize-2; in tcp_parse_options()
4314 length -= opsize; in tcp_parse_options()
4326 tp->rx_opt.saw_tstamp = 1; in tcp_parse_aligned_timestamp()
4328 tp->rx_opt.rcv_tsval = ntohl(*ptr); in tcp_parse_aligned_timestamp()
4331 tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset; in tcp_parse_aligned_timestamp()
4333 tp->rx_opt.rcv_tsecr = 0; in tcp_parse_aligned_timestamp()
4349 if (th->doff == (sizeof(*th) / 4)) { in tcp_fast_parse_options()
4350 tp->rx_opt.saw_tstamp = 0; in tcp_fast_parse_options()
4352 } else if (tp->rx_opt.tstamp_ok && in tcp_fast_parse_options()
4353 th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) { in tcp_fast_parse_options()
4358 tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL); in tcp_fast_parse_options()
4359 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) in tcp_fast_parse_options()
4360 tp->rx_opt.rcv_tsecr -= tp->tsoffset; in tcp_fast_parse_options()
4372 int length = (th->doff << 2) - sizeof(*th); in tcp_do_parse_auth_options()
4391 length--; in tcp_do_parse_auth_options()
4396 return -EINVAL; in tcp_do_parse_auth_options()
4399 return -EINVAL; in tcp_do_parse_auth_options()
4401 return -EEXIST; in tcp_do_parse_auth_options()
4405 return -EINVAL; in tcp_do_parse_auth_options()
4407 return -EEXIST; in tcp_do_parse_auth_options()
4411 ptr += opsize - 2; in tcp_do_parse_auth_options()
4412 length -= opsize; in tcp_do_parse_auth_options()
4419 /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
4450 if (tcp_sk(sk)->tcp_usec_ts) in tcp_tsval_replay()
4451 return inet_csk(sk)->icsk_rto * (USEC_PER_SEC / HZ); in tcp_tsval_replay()
4456 return inet_csk(sk)->icsk_rto * 1200 / HZ; in tcp_tsval_replay()
4465 u32 ack = TCP_SKB_CB(skb)->ack_seq; in tcp_disordered_ack_check()
4466 u32 seq = TCP_SKB_CB(skb)->seq; in tcp_disordered_ack_check()
4469 if (!th->ack || seq != TCP_SKB_CB(skb)->end_seq) in tcp_disordered_ack_check()
4473 if (seq != tp->rcv_nxt) in tcp_disordered_ack_check()
4474 return before(seq, tp->rcv_nxt) ? in tcp_disordered_ack_check()
4479 if (ack != tp->snd_una) in tcp_disordered_ack_check()
4483 if (tcp_may_update_window(tp, ack, seq, ntohs(th->window) << in tcp_disordered_ack_check()
4484 tp->rx_opt.snd_wscale)) in tcp_disordered_ack_check()
4488 if ((s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) > in tcp_disordered_ack_check()
4511 if (before(end_seq, tp->rcv_wup)) in tcp_sequence()
4514 if (after(seq, tp->rcv_nxt + tcp_receive_window(tp))) in tcp_sequence()
4524 WRITE_ONCE(sk->sk_err, err); in tcp_done_with_error()
4549 switch (sk->sk_state) { in tcp_reset()
4569 * If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT
4570 * (and thence onto LAST-ACK and finally, CLOSE, we never enter
4571 * TIME-WAIT)
4573 * If we are in FINWAIT-1, a received FIN indicates simultaneous
4574 * close and we go into CLOSING (and later onto TIME-WAIT)
4576 * If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
4584 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN); in tcp_fin()
4587 switch (sk->sk_state) { in tcp_fin()
4602 /* RFC793: Remain in the LAST-ACK state. */ in tcp_fin()
4614 /* Received a FIN -- send ACK and enter TIME_WAIT. */ in tcp_fin()
4622 pr_err("%s: Impossible, sk->sk_state=%d\n", in tcp_fin()
4623 __func__, sk->sk_state); in tcp_fin()
4627 /* It _is_ possible, that we have something out-of-order _after_ FIN. in tcp_fin()
4630 skb_rbtree_purge(&tp->out_of_order_queue); in tcp_fin()
4632 tcp_sack_reset(&tp->rx_opt); in tcp_fin()
4635 sk->sk_state_change(sk); in tcp_fin()
4638 if (sk->sk_shutdown == SHUTDOWN_MASK || in tcp_fin()
4639 sk->sk_state == TCP_CLOSE) in tcp_fin()
4649 if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { in tcp_sack_extend()
4650 if (before(seq, sp->start_seq)) in tcp_sack_extend()
4651 sp->start_seq = seq; in tcp_sack_extend()
4652 if (after(end_seq, sp->end_seq)) in tcp_sack_extend()
4653 sp->end_seq = end_seq; in tcp_sack_extend()
4663 if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) { in tcp_dsack_set()
4666 if (before(seq, tp->rcv_nxt)) in tcp_dsack_set()
4673 tp->rx_opt.dsack = 1; in tcp_dsack_set()
4674 tp->duplicate_sack[0].start_seq = seq; in tcp_dsack_set()
4675 tp->duplicate_sack[0].end_seq = end_seq; in tcp_dsack_set()
4683 if (!tp->rx_opt.dsack) in tcp_dsack_extend()
4686 tcp_sack_extend(tp->duplicate_sack, seq, end_seq); in tcp_dsack_extend()
4700 if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss && in tcp_rcv_spurious_retrans()
4701 skb->protocol == htons(ETH_P_IPV6) && in tcp_rcv_spurious_retrans()
4702 (tcp_sk(sk)->inet_conn.icsk_ack.lrcv_flowlabel != in tcp_rcv_spurious_retrans()
4716 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_send_dupack()
4717 before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_send_dupack()
4721 if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) { in tcp_send_dupack()
4722 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_send_dupack()
4725 if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) in tcp_send_dupack()
4726 end_seq = tp->rcv_nxt; in tcp_send_dupack()
4727 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq); in tcp_send_dupack()
4734 /* These routines update the SACK block as out-of-order packets arrive or
4735 * in-order packets close up the sequence space.
4740 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_maybe_coalesce()
4746 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) { in tcp_sack_maybe_coalesce()
4747 if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) { in tcp_sack_maybe_coalesce()
4753 tp->rx_opt.num_sacks--; in tcp_sack_maybe_coalesce()
4754 for (i = this_sack; i < tp->rx_opt.num_sacks; i++) in tcp_sack_maybe_coalesce()
4767 if (!tp->compressed_ack) in tcp_sack_compress_send_ack()
4770 if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1) in tcp_sack_compress_send_ack()
4774 * substract one from tp->compressed_ack to keep in tcp_sack_compress_send_ack()
4778 tp->compressed_ack - 1); in tcp_sack_compress_send_ack()
4780 tp->compressed_ack = 0; in tcp_sack_compress_send_ack()
4793 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_new_ofo_skb()
4794 int cur_sacks = tp->rx_opt.num_sacks; in tcp_sack_new_ofo_skb()
4805 for (; this_sack > 0; this_sack--, sp--) in tcp_sack_new_ofo_skb()
4806 swap(*sp, *(sp - 1)); in tcp_sack_new_ofo_skb()
4823 this_sack--; in tcp_sack_new_ofo_skb()
4824 tp->rx_opt.num_sacks--; in tcp_sack_new_ofo_skb()
4825 sp--; in tcp_sack_new_ofo_skb()
4827 for (; this_sack > 0; this_sack--, sp--) in tcp_sack_new_ofo_skb()
4828 *sp = *(sp - 1); in tcp_sack_new_ofo_skb()
4832 sp->start_seq = seq; in tcp_sack_new_ofo_skb()
4833 sp->end_seq = end_seq; in tcp_sack_new_ofo_skb()
4834 tp->rx_opt.num_sacks++; in tcp_sack_new_ofo_skb()
4841 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_remove()
4842 int num_sacks = tp->rx_opt.num_sacks; in tcp_sack_remove()
4846 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_sack_remove()
4847 tp->rx_opt.num_sacks = 0; in tcp_sack_remove()
4853 if (!before(tp->rcv_nxt, sp->start_seq)) { in tcp_sack_remove()
4857 WARN_ON(before(tp->rcv_nxt, sp->end_seq)); in tcp_sack_remove()
4861 tp->selective_acks[i-1] = tp->selective_acks[i]; in tcp_sack_remove()
4862 num_sacks--; in tcp_sack_remove()
4868 tp->rx_opt.num_sacks = num_sacks; in tcp_sack_remove()
4872 * tcp_try_coalesce - try to merge skb to prior one
4894 if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq) in tcp_try_coalesce()
4903 atomic_add(delta, &sk->sk_rmem_alloc); in tcp_try_coalesce()
4906 TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq; in tcp_try_coalesce()
4907 TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq; in tcp_try_coalesce()
4908 TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags; in tcp_try_coalesce()
4910 if (TCP_SKB_CB(from)->has_rxtstamp) { in tcp_try_coalesce()
4911 TCP_SKB_CB(to)->has_rxtstamp = true; in tcp_try_coalesce()
4912 to->tstamp = from->tstamp; in tcp_try_coalesce()
4913 skb_hwtstamps(to)->hwtstamp = skb_hwtstamps(from)->hwtstamp; in tcp_try_coalesce()
4926 /* In case tcp_drop_reason() is called later, update to->gso_segs */ in tcp_ooo_try_coalesce()
4928 u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) + in tcp_ooo_try_coalesce()
4929 max_t(u16, 1, skb_shinfo(from)->gso_segs); in tcp_ooo_try_coalesce()
4931 skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF); in tcp_ooo_try_coalesce()
4949 __u32 dsack_high = tp->rcv_nxt; in tcp_ofo_queue()
4954 p = rb_first(&tp->out_of_order_queue); in tcp_ofo_queue()
4957 if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) in tcp_ofo_queue()
4960 if (before(TCP_SKB_CB(skb)->seq, dsack_high)) { in tcp_ofo_queue()
4962 if (before(TCP_SKB_CB(skb)->end_seq, dsack_high)) in tcp_ofo_queue()
4963 dsack_high = TCP_SKB_CB(skb)->end_seq; in tcp_ofo_queue()
4964 tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack); in tcp_ofo_queue()
4967 rb_erase(&skb->rbnode, &tp->out_of_order_queue); in tcp_ofo_queue()
4969 if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) { in tcp_ofo_queue()
4974 tail = skb_peek_tail(&sk->sk_receive_queue); in tcp_ofo_queue()
4976 tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq); in tcp_ofo_queue()
4977 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; in tcp_ofo_queue()
4985 /* tcp_fin() purges tp->out_of_order_queue, in tcp_ofo_queue()
4999 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf || in tcp_try_rmem_schedule()
5003 return -1; in tcp_try_rmem_schedule()
5007 return -1; in tcp_try_rmem_schedule()
5024 if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) { in tcp_data_queue_ofo()
5026 sk->sk_data_ready(sk); in tcp_data_queue_ofo()
5032 tp->pred_flags = 0; in tcp_data_queue_ofo()
5035 tp->rcv_ooopack += max_t(u16, 1, skb_shinfo(skb)->gso_segs); in tcp_data_queue_ofo()
5037 seq = TCP_SKB_CB(skb)->seq; in tcp_data_queue_ofo()
5038 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_data_queue_ofo()
5040 p = &tp->out_of_order_queue.rb_node; in tcp_data_queue_ofo()
5041 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_data_queue_ofo()
5044 tp->rx_opt.num_sacks = 1; in tcp_data_queue_ofo()
5045 tp->selective_acks[0].start_seq = seq; in tcp_data_queue_ofo()
5046 tp->selective_acks[0].end_seq = end_seq; in tcp_data_queue_ofo()
5048 rb_link_node(&skb->rbnode, NULL, p); in tcp_data_queue_ofo()
5049 rb_insert_color(&skb->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5050 tp->ooo_last_skb = skb; in tcp_data_queue_ofo()
5057 if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb, in tcp_data_queue_ofo()
5070 if (!before(seq, TCP_SKB_CB(tp->ooo_last_skb)->end_seq)) { in tcp_data_queue_ofo()
5071 parent = &tp->ooo_last_skb->rbnode; in tcp_data_queue_ofo()
5072 p = &parent->rb_right; in tcp_data_queue_ofo()
5081 if (before(seq, TCP_SKB_CB(skb1)->seq)) { in tcp_data_queue_ofo()
5082 p = &parent->rb_left; in tcp_data_queue_ofo()
5085 if (before(seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5086 if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5096 if (after(seq, TCP_SKB_CB(skb1)->seq)) { in tcp_data_queue_ofo()
5098 tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5103 rb_replace_node(&skb1->rbnode, &skb->rbnode, in tcp_data_queue_ofo()
5104 &tp->out_of_order_queue); in tcp_data_queue_ofo()
5106 TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5107 TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5118 p = &parent->rb_right; in tcp_data_queue_ofo()
5122 rb_link_node(&skb->rbnode, parent, p); in tcp_data_queue_ofo()
5123 rb_insert_color(&skb->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5128 if (!after(end_seq, TCP_SKB_CB(skb1)->seq)) in tcp_data_queue_ofo()
5130 if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5131 tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5135 rb_erase(&skb1->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5136 tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5137 TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5143 tp->ooo_last_skb = skb; in tcp_data_queue_ofo()
5164 struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue); in tcp_queue_rcv()
5169 tcp_rcv_nxt_update(tcp_sk(sk), TCP_SKB_CB(skb)->end_seq); in tcp_queue_rcv()
5180 int err = -ENOMEM; in tcp_send_rcvq()
5193 skb = alloc_skb_with_frags(size - data_len, data_len, in tcp_send_rcvq()
5195 &err, sk->sk_allocation); in tcp_send_rcvq()
5199 skb_put(skb, size - data_len); in tcp_send_rcvq()
5200 skb->data_len = data_len; in tcp_send_rcvq()
5201 skb->len = size; in tcp_send_rcvq()
5203 if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) { in tcp_send_rcvq()
5208 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size); in tcp_send_rcvq()
5212 TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt; in tcp_send_rcvq()
5213 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size; in tcp_send_rcvq()
5214 TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1; in tcp_send_rcvq()
5231 if (tcp_epollin_ready(sk, sk->sk_rcvlowat) || sock_flag(sk, SOCK_DONE)) in tcp_data_ready()
5232 sk->sk_data_ready(sk); in tcp_data_ready()
5250 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { in tcp_data_queue()
5255 __skb_pull(skb, tcp_hdr(skb)->doff * 4); in tcp_data_queue()
5258 tp->rx_opt.dsack = 0; in tcp_data_queue()
5264 if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) { in tcp_data_queue()
5272 if (!skb->len && in tcp_data_queue()
5273 (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_data_queue()
5283 if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) { in tcp_data_queue()
5285 inet_csk(sk)->icsk_ack.pending |= in tcp_data_queue()
5288 sk->sk_data_ready(sk); in tcp_data_queue()
5290 if (skb_queue_len(&sk->sk_receive_queue) && skb->len) { in tcp_data_queue()
5295 sk_forced_mem_schedule(sk, skb->truesize); in tcp_data_queue()
5299 if (skb->len) in tcp_data_queue()
5301 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_data_queue()
5304 if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_data_queue()
5310 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) in tcp_data_queue()
5311 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; in tcp_data_queue()
5314 if (tp->rx_opt.num_sacks) in tcp_data_queue()
5326 if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { in tcp_data_queue()
5331 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); in tcp_data_queue()
5342 if (!before(TCP_SKB_CB(skb)->seq, in tcp_data_queue()
5343 tp->rcv_nxt + tcp_receive_window(tp))) { in tcp_data_queue()
5348 if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_data_queue()
5350 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt); in tcp_data_queue()
5353 * remembering D-SACK for its head made in previous line. in tcp_data_queue()
5369 return !skb_queue_is_last(list, skb) ? skb->next : NULL; in tcp_skb_next()
5383 rb_erase(&skb->rbnode, root); in tcp_collapse_one()
5391 /* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */
5394 struct rb_node **p = &root->rb_node; in tcp_rbtree_insert()
5401 if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq)) in tcp_rbtree_insert()
5402 p = &parent->rb_left; in tcp_rbtree_insert()
5404 p = &parent->rb_right; in tcp_rbtree_insert()
5406 rb_link_node(&skb->rbnode, parent, p); in tcp_rbtree_insert()
5407 rb_insert_color(&skb->rbnode, root); in tcp_rbtree_insert()
5437 if (!before(start, TCP_SKB_CB(skb)->end_seq)) { in tcp_collapse()
5445 * - not SYN/FIN and in tcp_collapse()
5446 * - bloated or contains data before "start" or in tcp_collapse()
5449 if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) && in tcp_collapse()
5450 (tcp_win_from_space(sk, skb->truesize) > skb->len || in tcp_collapse()
5451 before(TCP_SKB_CB(skb)->seq, start))) { in tcp_collapse()
5458 TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) { in tcp_collapse()
5465 start = TCP_SKB_CB(skb)->end_seq; in tcp_collapse()
5468 (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) || in tcp_collapse()
5475 int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start); in tcp_collapse()
5482 memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); in tcp_collapse()
5484 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start; in tcp_collapse()
5494 int offset = start - TCP_SKB_CB(skb)->seq; in tcp_collapse()
5495 int size = TCP_SKB_CB(skb)->end_seq - start; in tcp_collapse()
5502 TCP_SKB_CB(nskb)->end_seq += size; in tcp_collapse()
5503 copy -= size; in tcp_collapse()
5506 if (!before(start, TCP_SKB_CB(skb)->end_seq)) { in tcp_collapse()
5511 (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) || in tcp_collapse()
5532 skb = skb_rb_first(&tp->out_of_order_queue); in tcp_collapse_ofo_queue()
5535 tp->ooo_last_skb = skb_rb_last(&tp->out_of_order_queue); in tcp_collapse_ofo_queue()
5538 start = TCP_SKB_CB(skb)->seq; in tcp_collapse_ofo_queue()
5539 end = TCP_SKB_CB(skb)->end_seq; in tcp_collapse_ofo_queue()
5540 range_truesize = skb->truesize; in tcp_collapse_ofo_queue()
5549 after(TCP_SKB_CB(skb)->seq, end) || in tcp_collapse_ofo_queue()
5550 before(TCP_SKB_CB(skb)->end_seq, start)) { in tcp_collapse_ofo_queue()
5552 if (range_truesize != head->truesize || in tcp_collapse_ofo_queue()
5553 end - start >= SKB_WITH_OVERHEAD(PAGE_SIZE)) { in tcp_collapse_ofo_queue()
5554 tcp_collapse(sk, NULL, &tp->out_of_order_queue, in tcp_collapse_ofo_queue()
5558 if (sum_tiny > sk->sk_rcvbuf >> 3) in tcp_collapse_ofo_queue()
5564 range_truesize += skb->truesize; in tcp_collapse_ofo_queue()
5565 if (unlikely(before(TCP_SKB_CB(skb)->seq, start))) in tcp_collapse_ofo_queue()
5566 start = TCP_SKB_CB(skb)->seq; in tcp_collapse_ofo_queue()
5567 if (after(TCP_SKB_CB(skb)->end_seq, end)) in tcp_collapse_ofo_queue()
5568 end = TCP_SKB_CB(skb)->end_seq; in tcp_collapse_ofo_queue()
5573 * Clean the out-of-order queue to make room.
5592 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) in tcp_prune_ofo_queue()
5595 goal = sk->sk_rcvbuf >> 3; in tcp_prune_ofo_queue()
5596 node = &tp->ooo_last_skb->rbnode; in tcp_prune_ofo_queue()
5602 if (after(TCP_SKB_CB(in_skb)->seq, TCP_SKB_CB(skb)->seq)) in tcp_prune_ofo_queue()
5606 rb_erase(node, &tp->out_of_order_queue); in tcp_prune_ofo_queue()
5607 goal -= skb->truesize; in tcp_prune_ofo_queue()
5609 tp->ooo_last_skb = rb_to_skb(prev); in tcp_prune_ofo_queue()
5611 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf && in tcp_prune_ofo_queue()
5614 goal = sk->sk_rcvbuf >> 3; in tcp_prune_ofo_queue()
5626 if (tp->rx_opt.sack_ok) in tcp_prune_ofo_queue()
5627 tcp_sack_reset(&tp->rx_opt); in tcp_prune_ofo_queue()
5645 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) in tcp_prune_queue()
5650 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) in tcp_prune_queue()
5654 if (!skb_queue_empty(&sk->sk_receive_queue)) in tcp_prune_queue()
5655 tcp_collapse(sk, &sk->sk_receive_queue, NULL, in tcp_prune_queue()
5656 skb_peek(&sk->sk_receive_queue), in tcp_prune_queue()
5658 tp->copied_seq, tp->rcv_nxt); in tcp_prune_queue()
5660 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) in tcp_prune_queue()
5668 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf) in tcp_prune_queue()
5678 tp->pred_flags = 0; in tcp_prune_queue()
5679 return -1; in tcp_prune_queue()
5689 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) in tcp_should_expand_sndbuf()
5701 WRITE_ONCE(sk->sk_sndbuf, unused_mem); in tcp_should_expand_sndbuf()
5723 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_new_space()
5726 INDIRECT_CALL_1(sk->sk_write_space, sk_stream_write_space, sk); in tcp_new_space()
5730 * 1) Freeing skbs in rtx queues (after tp->snd_una has advanced)
5731 * 2) Sent skbs from output queue (and thus advancing tp->snd_nxt)
5734 * 1) Space consumed in output/rtx queues is below sk->sk_sndbuf/2
5735 * 2) notsent amount (tp->write_seq - tp->snd_nxt) became
5743 if (sk->sk_socket && in tcp_check_space()
5744 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_check_space()
5746 if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) in tcp_check_space()
5763 unsigned long rtt, delay; in __tcp_ack_snd_check() local
5766 if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss && in __tcp_ack_snd_check()
5772 (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || in __tcp_ack_snd_check()
5773 __tcp_select_window(sk) >= tp->rcv_wnd)) || in __tcp_ack_snd_check()
5776 /* Protocol state mandates a one-time immediate ACK */ in __tcp_ack_snd_check()
5777 inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOW) { in __tcp_ack_snd_check()
5782 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_backlog_ack_defer)) { in __tcp_ack_snd_check()
5783 set_bit(TCP_ACK_DEFERRED, &sk->sk_tsq_flags); in __tcp_ack_snd_check()
5791 if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in __tcp_ack_snd_check()
5797 tp->compressed_ack >= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr)) in __tcp_ack_snd_check()
5800 if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) { in __tcp_ack_snd_check()
5801 tp->compressed_ack_rcv_nxt = tp->rcv_nxt; in __tcp_ack_snd_check()
5802 tp->dup_ack_counter = 0; in __tcp_ack_snd_check()
5804 if (tp->dup_ack_counter < TCP_FASTRETRANS_THRESH) { in __tcp_ack_snd_check()
5805 tp->dup_ack_counter++; in __tcp_ack_snd_check()
5808 tp->compressed_ack++; in __tcp_ack_snd_check()
5809 if (hrtimer_is_queued(&tp->compressed_ack_timer)) in __tcp_ack_snd_check()
5812 /* compress ack timer : 5 % of rtt, but no more than tcp_comp_sack_delay_ns */ in __tcp_ack_snd_check()
5814 rtt = tp->rcv_rtt_est.rtt_us; in __tcp_ack_snd_check()
5815 if (tp->srtt_us && tp->srtt_us < rtt) in __tcp_ack_snd_check()
5816 rtt = tp->srtt_us; in __tcp_ack_snd_check()
5819 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns), in __tcp_ack_snd_check()
5820 rtt * (NSEC_PER_USEC >> 3)/20); in __tcp_ack_snd_check()
5822 hrtimer_start_range_ns(&tp->compressed_ack_timer, ns_to_ktime(delay), in __tcp_ack_snd_check()
5823 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_slack_ns), in __tcp_ack_snd_check()
5840 * place. We handle URGent data wrong. We have to - as
5849 u32 ptr = ntohs(th->urg_ptr); in tcp_check_urg()
5851 if (ptr && !READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_stdurg)) in tcp_check_urg()
5852 ptr--; in tcp_check_urg()
5853 ptr += ntohl(th->seq); in tcp_check_urg()
5856 if (after(tp->copied_seq, ptr)) in tcp_check_urg()
5869 if (before(ptr, tp->rcv_nxt)) in tcp_check_urg()
5873 if (tp->urg_data && !after(ptr, tp->urg_seq)) in tcp_check_urg()
5881 * tp->copied_seq since we would read the last urgent byte again in tcp_check_urg()
5894 if (tp->urg_seq == tp->copied_seq && tp->urg_data && in tcp_check_urg()
5895 !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) { in tcp_check_urg()
5896 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); in tcp_check_urg()
5897 tp->copied_seq++; in tcp_check_urg()
5898 if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) { in tcp_check_urg()
5899 __skb_unlink(skb, &sk->sk_receive_queue); in tcp_check_urg()
5904 WRITE_ONCE(tp->urg_data, TCP_URG_NOTYET); in tcp_check_urg()
5905 WRITE_ONCE(tp->urg_seq, ptr); in tcp_check_urg()
5908 tp->pred_flags = 0; in tcp_check_urg()
5916 /* Check if we get a new urgent pointer - normally not. */ in tcp_urg()
5917 if (unlikely(th->urg)) in tcp_urg()
5920 /* Do we wait for any urgent data? - normally not... */ in tcp_urg()
5921 if (unlikely(tp->urg_data == TCP_URG_NOTYET)) { in tcp_urg()
5922 u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) - in tcp_urg()
5923 th->syn; in tcp_urg()
5926 if (ptr < skb->len) { in tcp_urg()
5930 WRITE_ONCE(tp->urg_data, TCP_URG_VALID | tmp); in tcp_urg()
5932 sk->sk_data_ready(sk); in tcp_urg()
5937 /* Accept RST for rcv_nxt - 1 after a FIN.
5949 return unlikely(TCP_SKB_CB(skb)->seq == (tp->rcv_nxt - 1) && in tcp_reset_check()
5950 (1 << sk->sk_state) & (TCPF_CLOSE_WAIT | TCPF_LAST_ACK | in tcp_reset_check()
5965 !tp->rx_opt.saw_tstamp || in tcp_validate_incoming()
5966 tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW)) in tcp_validate_incoming()
5973 if (th->rst) in tcp_validate_incoming()
5975 if (unlikely(th->syn)) in tcp_validate_incoming()
5988 &tp->last_oow_ack_time)) in tcp_validate_incoming()
5994 reason = tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); in tcp_validate_incoming()
5996 /* RFC793, page 37: "In all states except SYN-SENT, all reset in tcp_validate_incoming()
5997 * (RST) segments are validated by checking their SEQ-fields." in tcp_validate_incoming()
6002 if (!th->rst) { in tcp_validate_incoming()
6003 if (th->syn) in tcp_validate_incoming()
6007 &tp->last_oow_ack_time)) in tcp_validate_incoming()
6016 if (th->rst) { in tcp_validate_incoming()
6017 /* RFC 5961 3.2 (extend to match against (RCV.NXT - 1) after a in tcp_validate_incoming()
6019 * If seq num matches RCV.NXT or (RCV.NXT - 1) after a FIN, or in tcp_validate_incoming()
6020 * the right-most SACK block, in tcp_validate_incoming()
6026 if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt || in tcp_validate_incoming()
6030 if (tcp_is_sack(tp) && tp->rx_opt.num_sacks > 0) { in tcp_validate_incoming()
6031 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_validate_incoming()
6035 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks; in tcp_validate_incoming()
6042 if (TCP_SKB_CB(skb)->seq == max_sack) in tcp_validate_incoming()
6046 /* Disable TFO if RST is out-of-order in tcp_validate_incoming()
6050 if (tp->syn_fastopen && !tp->data_segs_in && in tcp_validate_incoming()
6051 sk->sk_state == TCP_ESTABLISHED) in tcp_validate_incoming()
6063 if (th->syn) { in tcp_validate_incoming()
6064 if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack && in tcp_validate_incoming()
6065 TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq && in tcp_validate_incoming()
6066 TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt && in tcp_validate_incoming()
6067 TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt) in tcp_validate_incoming()
6098 * - A zero window was announced from us - zero window probing
6100 * - Out of order segments arrived.
6101 * - Urgent data is expected.
6102 * - There is no buffer space left
6103 * - Unexpected TCP flags/window values/header lengths are received
6105 * - Data is sent in both directions. Fast path only supports pure senders
6108 * - Unexpected TCP option.
6119 const struct tcphdr *th = (const struct tcphdr *)skb->data; in tcp_rcv_established()
6121 unsigned int len = skb->len; in tcp_rcv_established()
6127 if (unlikely(!rcu_access_pointer(sk->sk_rx_dst))) in tcp_rcv_established()
6128 inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb); in tcp_rcv_established()
6144 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_established()
6148 * 'S' will always be tp->tcp_header_len >> 2 in tcp_rcv_established()
6155 if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags && in tcp_rcv_established()
6156 TCP_SKB_CB(skb)->seq == tp->rcv_nxt && in tcp_rcv_established()
6157 !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { in tcp_rcv_established()
6158 int tcp_header_len = tp->tcp_header_len; in tcp_rcv_established()
6161 * is automatically equal to th->doff*4 due to pred_flags in tcp_rcv_established()
6172 if ((s32)(tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent) < 0) in tcp_rcv_established()
6191 tp->rcv_nxt == tp->rcv_wup) in tcp_rcv_established()
6204 tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr; in tcp_rcv_established()
6218 if ((int)skb->truesize > sk->sk_forward_alloc) in tcp_rcv_established()
6227 tp->rcv_nxt == tp->rcv_wup) in tcp_rcv_established()
6241 if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) { in tcp_rcv_established()
6248 tcp_update_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_established()
6261 if (len < (th->doff << 2) || tcp_checksum_complete(skb)) in tcp_rcv_established()
6264 if (!th->ack && !th->rst && !th->syn) { in tcp_rcv_established()
6279 reason = -reason; in tcp_rcv_established()
6311 icsk->icsk_af_ops->rebuild_header(sk); in tcp_init_transfer()
6315 * Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been in tcp_init_transfer()
6317 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK in tcp_init_transfer()
6320 if (tp->total_retrans > 1 && tp->undo_marker) in tcp_init_transfer()
6324 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_init_transfer()
6328 if (!icsk->icsk_ca_initialized) in tcp_init_transfer()
6340 icsk->icsk_ack.lrcvtime = tcp_jiffies32; in tcp_finish_connect()
6343 icsk->icsk_af_ops->sk_rx_dst_set(sk, skb); in tcp_finish_connect()
6353 tp->lsndtime = tcp_jiffies32; in tcp_finish_connect()
6358 if (!tp->rx_opt.snd_wscale) in tcp_finish_connect()
6359 __tcp_fast_path_on(tp, tp->snd_wnd); in tcp_finish_connect()
6361 tp->pred_flags = 0; in tcp_finish_connect()
6368 struct sk_buff *data = tp->syn_data ? tcp_rtx_queue_head(sk) : NULL; in tcp_rcv_fastopen_synack()
6369 u16 mss = tp->rx_opt.mss_clamp, try_exp = 0; in tcp_rcv_fastopen_synack()
6372 if (mss == tp->rx_opt.user_mss) { in tcp_rcv_fastopen_synack()
6382 if (!tp->syn_fastopen) { in tcp_rcv_fastopen_synack()
6384 cookie->len = -1; in tcp_rcv_fastopen_synack()
6385 } else if (tp->total_retrans) { in tcp_rcv_fastopen_synack()
6386 /* SYN timed out and the SYN-ACK neither has a cookie nor in tcp_rcv_fastopen_synack()
6389 * SYN-data or the corresponding SYN-ACK was dropped. in tcp_rcv_fastopen_synack()
6391 syn_drop = (cookie->len < 0 && data); in tcp_rcv_fastopen_synack()
6392 } else if (cookie->len < 0 && !tp->syn_data) { in tcp_rcv_fastopen_synack()
6397 try_exp = tp->syn_fastopen_exp ? 2 : 1; in tcp_rcv_fastopen_synack()
6403 if (tp->total_retrans) in tcp_rcv_fastopen_synack()
6404 tp->fastopen_client_fail = TFO_SYN_RETRANSMITTED; in tcp_rcv_fastopen_synack()
6406 tp->fastopen_client_fail = TFO_DATA_NOT_ACKED; in tcp_rcv_fastopen_synack()
6414 tp->syn_data_acked = tp->syn_data; in tcp_rcv_fastopen_synack()
6415 if (tp->syn_data_acked) { in tcp_rcv_fastopen_synack()
6417 /* SYN-data is counted as two separate packets in tcp_ack() */ in tcp_rcv_fastopen_synack()
6418 if (tp->delivered > 1) in tcp_rcv_fastopen_synack()
6419 --tp->delivered; in tcp_rcv_fastopen_synack()
6431 if (tp->syn_smc && !tp->rx_opt.smc_ok) in smc_check_reset_syn()
6432 tp->syn_smc = 0; in smc_check_reset_syn()
6446 syn_stamp = tp->retrans_stamp; in tcp_try_undo_spurious_syn()
6447 if (tp->undo_marker && syn_stamp && tp->rx_opt.saw_tstamp && in tcp_try_undo_spurious_syn()
6448 syn_stamp == tp->rx_opt.rcv_tsecr) in tcp_try_undo_spurious_syn()
6449 tp->undo_marker = 0; in tcp_try_undo_spurious_syn()
6457 struct tcp_fastopen_cookie foc = { .len = -1 }; in tcp_rcv_synsent_state_process()
6458 int saved_clamp = tp->rx_opt.mss_clamp; in tcp_rcv_synsent_state_process()
6462 tcp_parse_options(sock_net(sk), skb, &tp->rx_opt, 0, &foc); in tcp_rcv_synsent_state_process()
6463 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) in tcp_rcv_synsent_state_process()
6464 tp->rx_opt.rcv_tsecr -= tp->tsoffset; in tcp_rcv_synsent_state_process()
6466 if (th->ack) { in tcp_rcv_synsent_state_process()
6468 * "If the state is SYN-SENT then in tcp_rcv_synsent_state_process()
6475 if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) || in tcp_rcv_synsent_state_process()
6476 after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { in tcp_rcv_synsent_state_process()
6478 if (icsk->icsk_retransmits == 0) in tcp_rcv_synsent_state_process()
6486 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && in tcp_rcv_synsent_state_process()
6487 !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp, in tcp_rcv_synsent_state_process()
6503 if (th->rst) { in tcp_rcv_synsent_state_process()
6515 * --ANK(990513) in tcp_rcv_synsent_state_process()
6517 if (!th->syn) { in tcp_rcv_synsent_state_process()
6530 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_synsent_state_process()
6537 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1); in tcp_rcv_synsent_state_process()
6538 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; in tcp_rcv_synsent_state_process()
6543 tp->snd_wnd = ntohs(th->window); in tcp_rcv_synsent_state_process()
6545 if (!tp->rx_opt.wscale_ok) { in tcp_rcv_synsent_state_process()
6546 tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0; in tcp_rcv_synsent_state_process()
6547 WRITE_ONCE(tp->window_clamp, in tcp_rcv_synsent_state_process()
6548 min(tp->window_clamp, 65535U)); in tcp_rcv_synsent_state_process()
6551 if (tp->rx_opt.saw_tstamp) { in tcp_rcv_synsent_state_process()
6552 tp->rx_opt.tstamp_ok = 1; in tcp_rcv_synsent_state_process()
6553 tp->tcp_header_len = in tcp_rcv_synsent_state_process()
6555 tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; in tcp_rcv_synsent_state_process()
6558 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_rcv_synsent_state_process()
6561 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_rcv_synsent_state_process()
6565 * Change state from SYN-SENT only after copied_seq in tcp_rcv_synsent_state_process()
6567 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_synsent_state_process()
6575 fastopen_fail = (tp->syn_fastopen || tp->syn_data) && in tcp_rcv_synsent_state_process()
6579 sk->sk_state_change(sk); in tcp_rcv_synsent_state_process()
6583 return -1; in tcp_rcv_synsent_state_process()
6584 if (sk->sk_write_pending || in tcp_rcv_synsent_state_process()
6585 READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept) || in tcp_rcv_synsent_state_process()
6592 * to stand against the temptation 8) --ANK in tcp_rcv_synsent_state_process()
6601 return -1; in tcp_rcv_synsent_state_process()
6606 if (th->rst) { in tcp_rcv_synsent_state_process()
6617 if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp && in tcp_rcv_synsent_state_process()
6618 tcp_paws_reject(&tp->rx_opt, 0)) { in tcp_rcv_synsent_state_process()
6622 if (th->syn) { in tcp_rcv_synsent_state_process()
6630 ao = rcu_dereference_protected(tp->ao_info, in tcp_rcv_synsent_state_process()
6633 WRITE_ONCE(ao->risn, th->seq); in tcp_rcv_synsent_state_process()
6634 ao->rcv_sne = 0; in tcp_rcv_synsent_state_process()
6639 if (tp->rx_opt.saw_tstamp) { in tcp_rcv_synsent_state_process()
6640 tp->rx_opt.tstamp_ok = 1; in tcp_rcv_synsent_state_process()
6642 tp->tcp_header_len = in tcp_rcv_synsent_state_process()
6645 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_rcv_synsent_state_process()
6648 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1); in tcp_rcv_synsent_state_process()
6649 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_synsent_state_process()
6650 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; in tcp_rcv_synsent_state_process()
6655 tp->snd_wnd = ntohs(th->window); in tcp_rcv_synsent_state_process()
6656 tp->snd_wl1 = TCP_SKB_CB(skb)->seq; in tcp_rcv_synsent_state_process()
6657 tp->max_window = tp->snd_wnd; in tcp_rcv_synsent_state_process()
6662 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_rcv_synsent_state_process()
6678 return -1; in tcp_rcv_synsent_state_process()
6688 tcp_clear_options(&tp->rx_opt); in tcp_rcv_synsent_state_process()
6689 tp->rx_opt.mss_clamp = saved_clamp; in tcp_rcv_synsent_state_process()
6694 tcp_clear_options(&tp->rx_opt); in tcp_rcv_synsent_state_process()
6695 tp->rx_opt.mss_clamp = saved_clamp; in tcp_rcv_synsent_state_process()
6708 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss && !tp->packets_out) in tcp_rcv_synrecv_state_fastopen()
6712 inet_csk(sk)->icsk_retransmits = 0; in tcp_rcv_synrecv_state_fastopen()
6726 req = rcu_dereference_protected(tp->fastopen_rsk, in tcp_rcv_synrecv_state_fastopen()
6730 /* Re-arm the timer because data may have been sent out. in tcp_rcv_synrecv_state_fastopen()
6734 * (TFO) - we could try to be more aggressive and in tcp_rcv_synrecv_state_fastopen()
6758 switch (sk->sk_state) { in tcp_rcv_state_process()
6764 if (th->ack) in tcp_rcv_state_process()
6767 if (th->rst) { in tcp_rcv_state_process()
6771 if (th->syn) { in tcp_rcv_state_process()
6772 if (th->fin) { in tcp_rcv_state_process()
6781 icsk->icsk_af_ops->conn_request(sk, skb); in tcp_rcv_state_process()
6792 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_state_process()
6806 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_state_process()
6807 req = rcu_dereference_protected(tp->fastopen_rsk, in tcp_rcv_state_process()
6812 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && in tcp_rcv_state_process()
6813 sk->sk_state != TCP_FIN_WAIT1); in tcp_rcv_state_process()
6821 if (!th->ack && !th->rst && !th->syn) { in tcp_rcv_state_process()
6834 if (sk->sk_state == TCP_SYN_RECV) { in tcp_rcv_state_process()
6838 return -reason; in tcp_rcv_state_process()
6843 reason = -reason; in tcp_rcv_state_process()
6848 switch (sk->sk_state) { in tcp_rcv_state_process()
6850 tp->delivered++; /* SYN-ACK delivery isn't tracked in tcp_ack */ in tcp_rcv_state_process()
6851 if (!tp->srtt_us) in tcp_rcv_state_process()
6858 tp->retrans_stamp = 0; in tcp_rcv_state_process()
6861 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_state_process()
6866 sk->sk_state_change(sk); in tcp_rcv_state_process()
6868 /* Note, that this wakeup is only for marginal crossed SYN case. in tcp_rcv_state_process()
6870 * sk->sk_sleep == NULL and sk->sk_socket == NULL. in tcp_rcv_state_process()
6872 if (sk->sk_socket) in tcp_rcv_state_process()
6875 tp->snd_una = TCP_SKB_CB(skb)->ack_seq; in tcp_rcv_state_process()
6876 tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale; in tcp_rcv_state_process()
6877 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_state_process()
6879 if (tp->rx_opt.tstamp_ok) in tcp_rcv_state_process()
6880 tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; in tcp_rcv_state_process()
6882 if (!inet_csk(sk)->icsk_ca_ops->cong_control) in tcp_rcv_state_process()
6886 tp->lsndtime = tcp_jiffies32; in tcp_rcv_state_process()
6890 if (sk->sk_shutdown & SEND_SHUTDOWN) in tcp_rcv_state_process()
6900 if (tp->snd_una != tp->write_seq) in tcp_rcv_state_process()
6904 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | SEND_SHUTDOWN); in tcp_rcv_state_process()
6910 sk->sk_state_change(sk); in tcp_rcv_state_process()
6914 if (READ_ONCE(tp->linger2) < 0) { in tcp_rcv_state_process()
6919 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_rcv_state_process()
6920 after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { in tcp_rcv_state_process()
6922 if (tp->syn_fastopen && th->fin) in tcp_rcv_state_process()
6931 inet_csk_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN); in tcp_rcv_state_process()
6932 } else if (th->fin || sock_owned_by_user(sk)) { in tcp_rcv_state_process()
6948 if (tp->snd_una == tp->write_seq) { in tcp_rcv_state_process()
6955 if (tp->snd_una == tp->write_seq) { in tcp_rcv_state_process()
6967 switch (sk->sk_state) { in tcp_rcv_state_process()
6971 if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_rcv_state_process()
6986 if (sk->sk_shutdown & RCV_SHUTDOWN) { in tcp_rcv_state_process()
6987 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_rcv_state_process()
6988 after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { in tcp_rcv_state_process()
7001 /* tcp_data could move socket to TIME-WAIT */ in tcp_rcv_state_process()
7002 if (sk->sk_state != TCP_CLOSE) { in tcp_rcv_state_process()
7025 &ireq->ir_rmt_addr, port); in pr_drop_req()
7029 &ireq->ir_v6_rmt_addr, port); in pr_drop_req()
7045 * Exception: At least one of the reserved bits of the TCP header (th->res1) is
7057 bool th_ecn = th->ece && th->cwr; in tcp_ecn_create_request()
7064 ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield); in tcp_ecn_create_request()
7066 ecn_ok = READ_ONCE(net->ipv4.sysctl_tcp_ecn) || ecn_ok_dst; in tcp_ecn_create_request()
7068 if (((!ect || th->res1) && ecn_ok) || tcp_ca_needs_ecn(listen_sk) || in tcp_ecn_create_request()
7071 inet_rsk(req)->ecn_ok = 1; in tcp_ecn_create_request()
7080 req->rsk_rcv_wnd = 0; /* So that tcp_send_synack() knows! */ in tcp_openreq_init()
7081 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; in tcp_openreq_init()
7082 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; in tcp_openreq_init()
7083 tcp_rsk(req)->snt_synack = 0; in tcp_openreq_init()
7084 tcp_rsk(req)->last_oow_ack_time = 0; in tcp_openreq_init()
7085 req->mss = rx_opt->mss_clamp; in tcp_openreq_init()
7086 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; in tcp_openreq_init()
7087 ireq->tstamp_ok = rx_opt->tstamp_ok; in tcp_openreq_init()
7088 ireq->sack_ok = rx_opt->sack_ok; in tcp_openreq_init()
7089 ireq->snd_wscale = rx_opt->snd_wscale; in tcp_openreq_init()
7090 ireq->wscale_ok = rx_opt->wscale_ok; in tcp_openreq_init()
7091 ireq->acked = 0; in tcp_openreq_init()
7092 ireq->ecn_ok = 0; in tcp_openreq_init()
7093 ireq->ir_rmt_port = tcp_hdr(skb)->source; in tcp_openreq_init()
7094 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); in tcp_openreq_init()
7095 ireq->ir_mark = inet_request_mark(sk, skb); in tcp_openreq_init()
7097 ireq->smc_ok = rx_opt->smc_ok && !(tcp_sk(sk)->smc_hs_congested && in tcp_openreq_init()
7098 tcp_sk(sk)->smc_hs_congested(sk)); in tcp_openreq_init()
7107 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; in tcp_syn_flood_action()
7113 syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies); in tcp_syn_flood_action()
7124 if (!READ_ONCE(queue->synflood_warned) && syncookies != 2 && in tcp_syn_flood_action()
7125 xchg(&queue->synflood_warned, 1) == 0) { in tcp_syn_flood_action()
7126 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_family == AF_INET6) { in tcp_syn_flood_action()
7129 sk->sk_num, msg); in tcp_syn_flood_action()
7132 proto, &sk->sk_rcv_saddr, in tcp_syn_flood_action()
7133 sk->sk_num, msg); in tcp_syn_flood_action()
7144 if (tcp_sk(sk)->save_syn) { in tcp_reqsk_record_syn()
7150 if (tcp_sk(sk)->save_syn == 2) { /* Save full header. */ in tcp_reqsk_record_syn()
7162 saved_syn->mac_hdrlen = mac_hdrlen; in tcp_reqsk_record_syn()
7163 saved_syn->network_hdrlen = skb_network_header_len(skb); in tcp_reqsk_record_syn()
7164 saved_syn->tcp_hdrlen = tcp_hdrlen(skb); in tcp_reqsk_record_syn()
7165 memcpy(saved_syn->data, base, len); in tcp_reqsk_record_syn()
7166 req->saved_syn = saved_syn; in tcp_reqsk_record_syn()
7181 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) != 2 && in tcp_get_syncookie_mss()
7185 if (!tcp_syn_flood_action(sk, rsk_ops->slab_name)) in tcp_get_syncookie_mss()
7193 mss = tcp_parse_mss_option(th, tp->rx_opt.user_mss); in tcp_get_syncookie_mss()
7195 mss = af_ops->mss_clamp; in tcp_get_syncookie_mss()
7205 struct tcp_fastopen_cookie foc = { .len = -1 }; in tcp_conn_request()
7229 syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies); in tcp_conn_request()
7233 rsk_ops->slab_name); in tcp_conn_request()
7248 req->syncookie = want_cookie; in tcp_conn_request()
7249 tcp_rsk(req)->af_specific = af_ops; in tcp_conn_request()
7250 tcp_rsk(req)->ts_off = 0; in tcp_conn_request()
7251 tcp_rsk(req)->req_usec_ts = false; in tcp_conn_request()
7253 tcp_rsk(req)->is_mptcp = 0; in tcp_conn_request()
7257 tmp_opt.mss_clamp = af_ops->mss_clamp; in tcp_conn_request()
7258 tmp_opt.user_mss = tp->rx_opt.user_mss; in tcp_conn_request()
7270 inet_rsk(req)->no_srccheck = inet_test_bit(TRANSPARENT, sk); in tcp_conn_request()
7273 inet_rsk(req)->ir_iif = inet_request_bound_dev_if(sk, skb); in tcp_conn_request()
7275 dst = af_ops->route_req(sk, skb, &fl, req, isn); in tcp_conn_request()
7280 tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst); in tcp_conn_request()
7281 tcp_rsk(req)->ts_off = af_ops->init_ts_off(net, skb); in tcp_conn_request()
7284 int max_syn_backlog = READ_ONCE(net->ipv4.sysctl_max_syn_backlog); in tcp_conn_request()
7288 (max_syn_backlog - inet_csk_reqsk_queue_len(sk) < in tcp_conn_request()
7298 pr_drop_req(req, ntohs(tcp_hdr(skb)->source), in tcp_conn_request()
7299 rsk_ops->family); in tcp_conn_request()
7303 isn = af_ops->init_seq(skb); in tcp_conn_request()
7309 isn = cookie_init_sequence(af_ops, sk, skb, &req->mss); in tcp_conn_request()
7311 inet_rsk(req)->ecn_ok = 0; in tcp_conn_request()
7318 tcp_rsk(req)->used_tcp_ao = true; in tcp_conn_request()
7319 tcp_rsk(req)->ao_rcv_next = aoh->keyid; in tcp_conn_request()
7320 tcp_rsk(req)->ao_keyid = aoh->rnext_keyid; in tcp_conn_request()
7323 tcp_rsk(req)->used_tcp_ao = false; in tcp_conn_request()
7326 tcp_rsk(req)->snt_isn = isn; in tcp_conn_request()
7327 tcp_rsk(req)->txhash = net_tx_rndhash(); in tcp_conn_request()
7328 tcp_rsk(req)->syn_tos = TCP_SKB_CB(skb)->ip_dsfield; in tcp_conn_request()
7336 af_ops->send_synack(fastopen_sk, dst, &fl, req, in tcp_conn_request()
7345 sk->sk_data_ready(sk); in tcp_conn_request()
7349 tcp_rsk(req)->tfo_listener = false; in tcp_conn_request()
7351 req->timeout = tcp_timeout_init((struct sock *)req); in tcp_conn_request()
7353 req->timeout))) { in tcp_conn_request()
7360 af_ops->send_synack(sk, dst, &fl, req, &foc, in tcp_conn_request()