Lines Matching +full:non +full:- +full:urgent
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>
62 * Pasi Sarolahti: F-RTO for dealing with spurious RTOs
100 #define FLAG_DSACKING_ACK 0x800 /* SACK blocks contained D-SACK info */
119 #define REXMIT_NEW 2 /* FRTO-style transmit of unsent/new packets */
127 tp->tcp_clean_acked = cad; in clean_acked_data_enable()
135 tp->tcp_clean_acked = NULL; in clean_acked_data_disable()
149 bool unknown_opt = tcp_sk(sk)->rx_opt.saw_unknown && in bpf_skops_parse_hdr()
163 switch (sk->sk_state) { in bpf_skops_parse_hdr()
217 dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif); in tcp_gro_dev_warn()
218 if (!dev || len >= READ_ONCE(dev->mtu)) in tcp_gro_dev_warn()
220 dev ? dev->name : "Unknown driver"); in tcp_gro_dev_warn()
230 const unsigned int lss = icsk->icsk_ack.last_seg_size; in tcp_measure_rcv_mss()
233 icsk->icsk_ack.last_seg_size = 0; in tcp_measure_rcv_mss()
235 /* skb->len may jitter because of SACKs, even if peer in tcp_measure_rcv_mss()
236 * sends good full-sized frames. in tcp_measure_rcv_mss()
238 len = skb_shinfo(skb)->gso_size ? : skb->len; in tcp_measure_rcv_mss()
239 if (len >= icsk->icsk_ack.rcv_mss) { in tcp_measure_rcv_mss()
244 if (unlikely(len != icsk->icsk_ack.rcv_mss)) { in tcp_measure_rcv_mss()
245 u64 val = (u64)skb->len << TCP_RMEM_TO_WIN_SCALE; in tcp_measure_rcv_mss()
246 u8 old_ratio = tcp_sk(sk)->scaling_ratio; in tcp_measure_rcv_mss()
248 do_div(val, skb->truesize); in tcp_measure_rcv_mss()
249 tcp_sk(sk)->scaling_ratio = val ? val : 1; in tcp_measure_rcv_mss()
251 if (old_ratio != tcp_sk(sk)->scaling_ratio) { in tcp_measure_rcv_mss()
254 val = tcp_win_from_space(sk, sk->sk_rcvbuf); in tcp_measure_rcv_mss()
257 if (tp->window_clamp < tp->rcvq_space.space) in tcp_measure_rcv_mss()
258 tp->rcvq_space.space = tp->window_clamp; in tcp_measure_rcv_mss()
261 icsk->icsk_ack.rcv_mss = min_t(unsigned int, len, in tcp_measure_rcv_mss()
262 tcp_sk(sk)->advmss); in tcp_measure_rcv_mss()
263 /* Account for possibly-removed options */ in tcp_measure_rcv_mss()
264 DO_ONCE_LITE_IF(len > icsk->icsk_ack.rcv_mss + MAX_TCP_OPTION_SPACE, in tcp_measure_rcv_mss()
269 * may be waiting for an ACK if cwnd-bound or using TX zero in tcp_measure_rcv_mss()
272 * reads all of the data and is not ping-pong. If len > MSS in tcp_measure_rcv_mss()
277 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_PSH) in tcp_measure_rcv_mss()
278 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; in tcp_measure_rcv_mss()
285 len += skb->data - skb_transport_header(skb); in tcp_measure_rcv_mss()
290 * to handle super-low mtu links fairly. in tcp_measure_rcv_mss()
298 len -= tcp_sk(sk)->tcp_header_len; in tcp_measure_rcv_mss()
299 icsk->icsk_ack.last_seg_size = len; in tcp_measure_rcv_mss()
301 icsk->icsk_ack.rcv_mss = len; in tcp_measure_rcv_mss()
305 if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED) in tcp_measure_rcv_mss()
306 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2; in tcp_measure_rcv_mss()
307 icsk->icsk_ack.pending |= ICSK_ACK_PUSHED; in tcp_measure_rcv_mss()
314 unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss); in tcp_incr_quickack()
319 if (quickacks > icsk->icsk_ack.quick) in tcp_incr_quickack()
320 icsk->icsk_ack.quick = quickacks; in tcp_incr_quickack()
329 icsk->icsk_ack.ato = TCP_ATO_MIN; in tcp_enter_quickack_mode()
340 return icsk->icsk_ack.dst_quick_ack || in tcp_in_quickack_mode()
341 (icsk->icsk_ack.quick && !inet_csk_in_pingpong_mode(sk)); in tcp_in_quickack_mode()
351 switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) { in tcp_data_ecn_check()
357 if (tp->ecn_flags & TCP_ECN_SEEN) in tcp_data_ecn_check()
364 if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR) && in tcp_data_ecn_check()
368 tp->ecn_flags |= TCP_ECN_DEMAND_CWR; in tcp_data_ecn_check()
376 tp->ecn_flags |= TCP_ECN_SEEN; in tcp_data_ecn_check()
383 tp->ecn_flags |= TCP_ECN_SEEN; in tcp_data_ecn_check()
393 u8 estimate_ecnfield = tp->est_ecnfield; in tcp_accecn_process_option()
404 if (!(flag & FLAG_SLOWPATH) || !tp->rx_opt.accecn) { in tcp_accecn_process_option()
405 if (!tp->saw_accecn_opt) { in tcp_accecn_process_option()
409 if (tp->bytes_sent >= (1 << 23) - 1) { in tcp_accecn_process_option()
418 u8 ecnfield = estimate_ecnfield - 1; in tcp_accecn_process_option()
420 tp->delivered_ecn_bytes[ecnfield] += delivered_bytes; in tcp_accecn_process_option()
426 ptr = skb_transport_header(skb) + tp->rx_opt.accecn; in tcp_accecn_process_option()
427 optlen = ptr[1] - 2; in tcp_accecn_process_option()
433 if (tp->saw_accecn_opt < TCP_ACCECN_OPT_COUNTER_SEEN) { in tcp_accecn_process_option()
434 tp->saw_accecn_opt = tcp_accecn_option_init(skb, in tcp_accecn_process_option()
435 tp->rx_opt.accecn); in tcp_accecn_process_option()
436 if (tp->saw_accecn_opt == TCP_ACCECN_OPT_FAIL_SEEN) in tcp_accecn_process_option()
452 cnt = &tp->delivered_ecn_bytes[ecnfield - 1]; in tcp_accecn_process_option()
460 tp->est_ecnfield = ecnfield; in tcp_accecn_process_option()
468 optlen -= TCPOLEN_ACCECN_PERFIELD; in tcp_accecn_process_option()
472 tp->est_ecnfield = 0; in tcp_accecn_process_option()
479 tp->delivered_ce += ecn_count; in tcp_count_delivered_ce()
486 tp->delivered += delivered; in tcp_count_delivered()
496 u32 old_ceb = tcp_sk(sk)->delivered_ecn_bytes[INET_ECN_CE - 1]; in __tcp_accecn_process()
520 if (tp->received_ce_pending >= TCP_ACCECN_ACE_MAX_DELTA) in __tcp_accecn_process()
521 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; in __tcp_accecn_process()
523 corrected_ace = tcp_accecn_ace(th) - TCP_ACCECN_CEP_INIT_OFFSET; in __tcp_accecn_process()
524 delta = (corrected_ace - tp->delivered_ce) & TCP_ACCECN_CEP_ACE_MASK; in __tcp_accecn_process()
528 safe_delta = delivered_pkts - in __tcp_accecn_process()
529 ((delivered_pkts - delta) & TCP_ACCECN_CEP_ACE_MASK); in __tcp_accecn_process()
532 d_ceb = tp->delivered_ecn_bytes[INET_ECN_CE - 1] - old_ceb; in __tcp_accecn_process()
538 ((1 << inet_csk(sk)->icsk_ca_state) & in __tcp_accecn_process()
553 if (d_ceb > delta * tp->mss_cache) in __tcp_accecn_process()
556 safe_delta * tp->mss_cache >> TCP_ACCECN_SAFETY_SHIFT) in __tcp_accecn_process()
576 if (tp->pred_flags) in tcp_accecn_process()
584 * 1. Tuning sk->sk_sndbuf, when connection enters established state.
590 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; in tcp_sndbuf_expand()
594 /* Worst case is non GSO/TSO : each frame consumes one skb in tcp_sndbuf_expand()
595 * and skb->head is kmalloced using power of two area of memory in tcp_sndbuf_expand()
597 per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) + in tcp_sndbuf_expand()
605 nr_segs = max_t(u32, nr_segs, tp->reordering + 1); in tcp_sndbuf_expand()
611 sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2; in tcp_sndbuf_expand()
614 if (sk->sk_sndbuf < sndmem) in tcp_sndbuf_expand()
615 WRITE_ONCE(sk->sk_sndbuf, in tcp_sndbuf_expand()
616 min(sndmem, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[2]))); in tcp_sndbuf_expand()
622 * forward and advertised in receiver window (tp->rcv_wnd) and
626 * tcp_full_space(), in this case tcp_full_space() - window_clamp
634 * - to enforce header prediction at sender, even when application
636 * - to prevent pruning of receive queue because of misprediction
651 int window = tcp_win_from_space(sk, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])) >> 1; in __tcp_grow_window()
653 while (tp->rcv_ssthresh <= window) { in __tcp_grow_window()
654 if (truesize <= skb->len) in __tcp_grow_window()
655 return 2 * inet_csk(sk)->icsk_ack.rcv_mss; in __tcp_grow_window()
664 * can play nice with us, as sk_buff and skb->head might be either
667 * and if no payload was pulled in skb->head before reaching us.
671 u32 truesize = skb->truesize; in truesize_adjust()
674 truesize -= SKB_TRUESIZE(skb_end_offset(skb)); in truesize_adjust()
676 if (unlikely((int)truesize < (int)skb->len)) in truesize_adjust()
677 truesize = skb->truesize; in truesize_adjust()
688 room = min_t(int, tp->window_clamp, tcp_space(sk)) - tp->rcv_ssthresh; in tcp_grow_window()
701 if (tcp_win_from_space(sk, truesize) <= skb->len) in tcp_grow_window()
702 incr = 2 * tp->advmss; in tcp_grow_window()
707 incr = max_t(int, incr, 2 * skb->len); in tcp_grow_window()
708 tp->rcv_ssthresh += min(room, incr); in tcp_grow_window()
709 inet_csk(sk)->icsk_ack.quick |= 1; in tcp_grow_window()
724 int tcp_app_win = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_app_win); in tcp_init_buffer_space()
728 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) in tcp_init_buffer_space()
732 tp->rcvq_space.time = tp->tcp_mstamp; in tcp_init_buffer_space()
733 tp->rcvq_space.seq = tp->copied_seq; in tcp_init_buffer_space()
737 if (tp->window_clamp >= maxwin) { in tcp_init_buffer_space()
738 WRITE_ONCE(tp->window_clamp, maxwin); in tcp_init_buffer_space()
740 if (tcp_app_win && maxwin > 4 * tp->advmss) in tcp_init_buffer_space()
741 WRITE_ONCE(tp->window_clamp, in tcp_init_buffer_space()
742 max(maxwin - (maxwin >> tcp_app_win), in tcp_init_buffer_space()
743 4 * tp->advmss)); in tcp_init_buffer_space()
748 tp->window_clamp > 2 * tp->advmss && in tcp_init_buffer_space()
749 tp->window_clamp + tp->advmss > maxwin) in tcp_init_buffer_space()
750 WRITE_ONCE(tp->window_clamp, in tcp_init_buffer_space()
751 max(2 * tp->advmss, maxwin - tp->advmss)); in tcp_init_buffer_space()
753 tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp); in tcp_init_buffer_space()
754 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_init_buffer_space()
755 tp->rcvq_space.space = min3(tp->rcv_ssthresh, tp->rcv_wnd, in tcp_init_buffer_space()
756 (u32)TCP_INIT_CWND * tp->advmss); in tcp_init_buffer_space()
767 icsk->icsk_ack.quick = 0; in tcp_clamp_window()
768 rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]); in tcp_clamp_window()
770 if (sk->sk_rcvbuf < rmem2 && in tcp_clamp_window()
771 !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) && in tcp_clamp_window()
774 WRITE_ONCE(sk->sk_rcvbuf, in tcp_clamp_window()
775 min(atomic_read(&sk->sk_rmem_alloc), rmem2)); in tcp_clamp_window()
777 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) in tcp_clamp_window()
778 tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss); in tcp_clamp_window()
791 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); in tcp_initialize_rcv_mss()
793 hint = min(hint, tp->rcv_wnd / 2); in tcp_initialize_rcv_mss()
797 inet_csk(sk)->icsk_ack.rcv_mss = hint; in tcp_initialize_rcv_mss()
804 * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL.
814 u32 new_sample, old_sample = tp->rcv_rtt_est.rtt_us; in tcp_rcv_rtt_update()
820 /* If we sample in larger samples in the non-timestamp in tcp_rcv_rtt_update()
826 * non-timestamp case, we do not smooth things out in tcp_rcv_rtt_update()
833 if (tp->rcv_nxt != tp->copied_seq) in tcp_rcv_rtt_update()
835 new_sample = old_sample - (old_sample >> 3) + sample; in tcp_rcv_rtt_update()
838 tp->rcv_rtt_est.rtt_us = new_sample; in tcp_rcv_rtt_update()
845 if (tp->rcv_rtt_est.time == 0) in tcp_rcv_rtt_measure()
847 if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq)) in tcp_rcv_rtt_measure()
849 delta_us = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcv_rtt_est.time); in tcp_rcv_rtt_measure()
855 tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd; in tcp_rcv_rtt_measure()
856 tp->rcv_rtt_est.time = tp->tcp_mstamp; in tcp_rcv_rtt_measure()
863 delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr; in tcp_rtt_tsopt_us()
864 if (tp->tcp_usec_ts) in tcp_rtt_tsopt_us()
873 return -1; in tcp_rtt_tsopt_us()
881 if (tp->rx_opt.rcv_tsecr == tp->rcv_rtt_last_tsecr) in tcp_rcv_rtt_measure_ts()
883 tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr; in tcp_rcv_rtt_measure_ts()
885 if (TCP_SKB_CB(skb)->end_seq - in tcp_rcv_rtt_measure_ts()
886 TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) { in tcp_rcv_rtt_measure_ts()
901 oldval = tp->rcvq_space.space; in tcp_rcvbuf_grow()
902 tp->rcvq_space.space = newval; in tcp_rcvbuf_grow()
904 if (!READ_ONCE(net->ipv4.sysctl_tcp_moderate_rcvbuf) || in tcp_rcvbuf_grow()
905 (sk->sk_userlocks & SOCK_RCVBUF_LOCK)) in tcp_rcvbuf_grow()
912 grow = (u64)rcvwin * (newval - oldval); in tcp_rcvbuf_grow()
916 if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) in tcp_rcvbuf_grow()
917 rcvwin += TCP_SKB_CB(tp->ooo_last_skb)->end_seq - tp->rcv_nxt; in tcp_rcvbuf_grow()
919 cap = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]); in tcp_rcvbuf_grow()
922 if (rcvbuf > sk->sk_rcvbuf) { in tcp_rcvbuf_grow()
923 WRITE_ONCE(sk->sk_rcvbuf, rcvbuf); in tcp_rcvbuf_grow()
925 WRITE_ONCE(tp->window_clamp, in tcp_rcvbuf_grow()
941 time = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcvq_space.time); in tcp_rcv_space_adjust()
942 if (time < (tp->rcv_rtt_est.rtt_us >> 3) || tp->rcv_rtt_est.rtt_us == 0) in tcp_rcv_space_adjust()
946 copied = tp->copied_seq - tp->rcvq_space.seq; in tcp_rcv_space_adjust()
948 inq = tp->rcv_nxt - tp->copied_seq; in tcp_rcv_space_adjust()
949 copied -= inq; in tcp_rcv_space_adjust()
950 if (copied <= tp->rcvq_space.space) in tcp_rcv_space_adjust()
958 tp->rcvq_space.seq = tp->copied_seq; in tcp_rcv_space_adjust()
959 tp->rcvq_space.time = tp->tcp_mstamp; in tcp_rcv_space_adjust()
967 if (skb->protocol == htons(ETH_P_IPV6)) in tcp_save_lrcv_flowlabel()
968 icsk->icsk_ack.lrcv_flowlabel = ntohl(ip6_flowlabel(ipv6_hdr(skb))); in tcp_save_lrcv_flowlabel()
973 * behavior of the tp->ato delayed ack timeout interval. When a
980 * queue. -DaveM
996 if (!icsk->icsk_ack.ato) { in tcp_event_data_recv()
1001 icsk->icsk_ack.ato = TCP_ATO_MIN; in tcp_event_data_recv()
1003 int m = now - icsk->icsk_ack.lrcvtime; in tcp_event_data_recv()
1007 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2; in tcp_event_data_recv()
1008 } else if (m < icsk->icsk_ack.ato) { in tcp_event_data_recv()
1009 icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m; in tcp_event_data_recv()
1010 if (icsk->icsk_ack.ato > icsk->icsk_rto) in tcp_event_data_recv()
1011 icsk->icsk_ack.ato = icsk->icsk_rto; in tcp_event_data_recv()
1012 } else if (m > icsk->icsk_rto) { in tcp_event_data_recv()
1019 icsk->icsk_ack.lrcvtime = now; in tcp_event_data_recv()
1024 if (skb->len >= 128) in tcp_event_data_recv()
1035 * it up into three procedures. -- erics
1041 u32 srtt = tp->srtt_us; in tcp_rtt_estimator()
1060 m -= (srtt >> 3); /* m is now error in rtt est */ in tcp_rtt_estimator()
1063 m = -m; /* m is now abs(error) */ in tcp_rtt_estimator()
1064 m -= (tp->mdev_us >> 2); /* similar update on mdev */ in tcp_rtt_estimator()
1076 m -= (tp->mdev_us >> 2); /* similar update on mdev */ in tcp_rtt_estimator()
1078 tp->mdev_us += m; /* mdev = 3/4 mdev + 1/4 new */ in tcp_rtt_estimator()
1079 if (tp->mdev_us > tp->mdev_max_us) { in tcp_rtt_estimator()
1080 tp->mdev_max_us = tp->mdev_us; in tcp_rtt_estimator()
1081 if (tp->mdev_max_us > tp->rttvar_us) in tcp_rtt_estimator()
1082 tp->rttvar_us = tp->mdev_max_us; in tcp_rtt_estimator()
1084 if (after(tp->snd_una, tp->rtt_seq)) { in tcp_rtt_estimator()
1085 if (tp->mdev_max_us < tp->rttvar_us) in tcp_rtt_estimator()
1086 tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2; in tcp_rtt_estimator()
1087 tp->rtt_seq = tp->snd_nxt; in tcp_rtt_estimator()
1088 tp->mdev_max_us = tcp_rto_min_us(sk); in tcp_rtt_estimator()
1095 tp->mdev_us = m << 1; /* make sure rto = 3*rtt */ in tcp_rtt_estimator()
1096 tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk)); in tcp_rtt_estimator()
1097 tp->mdev_max_us = tp->rttvar_us; in tcp_rtt_estimator()
1098 tp->rtt_seq = tp->snd_nxt; in tcp_rtt_estimator()
1102 tp->srtt_us = max(1U, srtt); in tcp_rtt_estimator()
1111 rate = (u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3); in tcp_update_pacing_rate()
1117 * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh) in tcp_update_pacing_rate()
1118 * If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching in tcp_update_pacing_rate()
1121 if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2) in tcp_update_pacing_rate()
1122 rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio); in tcp_update_pacing_rate()
1124 rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio); in tcp_update_pacing_rate()
1126 rate *= max(tcp_snd_cwnd(tp), tp->packets_out); in tcp_update_pacing_rate()
1128 if (likely(tp->srtt_us)) in tcp_update_pacing_rate()
1129 do_div(rate, tp->srtt_us); in tcp_update_pacing_rate()
1135 WRITE_ONCE(sk->sk_pacing_rate, in tcp_update_pacing_rate()
1136 min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate))); in tcp_update_pacing_rate()
1152 * is invisible. Actually, Linux-2.4 also generates erratic in tcp_set_rto()
1155 inet_csk(sk)->icsk_rto = __tcp_set_rto(tp); in tcp_set_rto()
1175 return min_t(__u32, cwnd, tp->snd_cwnd_clamp); in tcp_init_cwnd()
1179 /* Timestamps for earliest and latest never-retransmitted segment
1193 /* Take a notice that peer is sending D-SACKs. Skip update of data delivery
1196 * - DSACKed sequence range is larger than maximum receiver's window.
1197 * - Total no. of DSACKed segments exceed the total no. of retransmitted segs.
1207 seq_len = end_seq - start_seq; in tcp_dsack_seen()
1209 if (seq_len > tp->max_window) in tcp_dsack_seen()
1211 if (seq_len > tp->mss_cache) in tcp_dsack_seen()
1212 dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache); in tcp_dsack_seen()
1213 else if (tp->tlp_high_seq && tp->tlp_high_seq == end_seq) in tcp_dsack_seen()
1214 state->flag |= FLAG_DSACK_TLP; in tcp_dsack_seen()
1216 tp->dsack_dups += dup_segs; in tcp_dsack_seen()
1218 if (tp->dsack_dups > tp->total_retrans) in tcp_dsack_seen()
1221 tp->rx_opt.sack_ok |= TCP_DSACK_SEEN; in tcp_dsack_seen()
1226 * is timer-driven, not triggered by RACK). in tcp_dsack_seen()
1228 if (tp->reord_seen && !(state->flag & FLAG_DSACK_TLP)) in tcp_dsack_seen()
1229 tp->rack.dsack_seen = 1; in tcp_dsack_seen()
1231 state->flag |= FLAG_DSACKING_ACK; in tcp_dsack_seen()
1233 state->sack_delivered += dup_segs; in tcp_dsack_seen()
1239 * some lower never-retransmitted sequence ("low_seq"). The maximum reordering
1240 * distance is approximated in full-mss packet distance ("reordering").
1246 const u32 mss = tp->mss_cache; in tcp_check_sack_reordering()
1253 metric = fack - low_seq; in tcp_check_sack_reordering()
1254 if ((metric > tp->reordering * mss) && mss) { in tcp_check_sack_reordering()
1257 tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state, in tcp_check_sack_reordering()
1258 tp->reordering, in tcp_check_sack_reordering()
1260 tp->sacked_out, in tcp_check_sack_reordering()
1261 tp->undo_marker ? tp->undo_retrans : 0); in tcp_check_sack_reordering()
1263 tp->reordering = min_t(u32, (metric + mss - 1) / mss, in tcp_check_sack_reordering()
1264 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering)); in tcp_check_sack_reordering()
1268 tp->reord_seen++; in tcp_check_sack_reordering()
1280 if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) || in tcp_verify_retransmit_hint()
1281 (tp->retransmit_skb_hint && in tcp_verify_retransmit_hint()
1282 before(TCP_SKB_CB(skb)->seq, in tcp_verify_retransmit_hint()
1283 TCP_SKB_CB(tp->retransmit_skb_hint)->seq))) in tcp_verify_retransmit_hint()
1284 tp->retransmit_skb_hint = skb; in tcp_verify_retransmit_hint()
1292 tp->lost += tcp_skb_pcount(skb); in tcp_notify_skb_loss_event()
1297 __u8 sacked = TCP_SKB_CB(skb)->sacked; in tcp_mark_skb_lost()
1307 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS; in tcp_mark_skb_lost()
1308 tp->retrans_out -= tcp_skb_pcount(skb); in tcp_mark_skb_lost()
1314 tp->lost_out += tcp_skb_pcount(skb); in tcp_mark_skb_lost()
1315 TCP_SKB_CB(skb)->sacked |= TCPCB_LOST; in tcp_mark_skb_lost()
1328 * 0 1 - orig segment is in flight.
1329 * S 0 - nothing flies, orig reached receiver.
1330 * L 0 - nothing flies, orig lost by net.
1331 * R 2 - both orig and retransmit are in flight.
1332 * L|R 1 - orig is lost, retransmit is in flight.
1333 * S|R 1 - orig reached receiver, retrans is still in flight.
1335 * but it is equivalent to plain S and code short-circuits it to S.
1336 * L|S is logically invalid, it would mean -1 packet in flight 8))
1346 * 4. D-SACK added new rule: D-SACK changes any tag to S.
1353 * --------------------
1358 * ever retransmitted -> reordering. Alas, we cannot use it
1360 * 2. The last flaw is solved with D-SACK. D-SACK arrives
1361 * for retransmitted and already SACKed segment -> reordering..
1366 * ----------------------
1381 * Implements also blockage to start_seq wrap-around. Problem lies in the
1387 * <- outs wnd -> <- wrapzone ->
1390 * |<------------+------+----- TCP seqno space --------------+---------->|
1391 * ...-- <2^31 ->| |<--------...
1392 * ...---- >2^31 ------>| |<--------...
1397 * snd_nxt wrap -> snd_una region will then become "well defined", i.e.,
1400 * With D-SACK the lower bound is extended to cover sequence space below
1402 * again, D-SACK block must not to go across snd_una (for the same reason as
1404 * ends, TCP might receive valid D-SACKs below that. As long as they reside
1407 * theoretical ones), the D-SACK will nicely cross that boundary due to skb
1410 * the exact amount is rather hard to quantify. However, tp->max_window can
1417 if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq)) in tcp_is_sackblock_valid()
1420 /* Nasty start_seq wrap-around check (see comments above) */ in tcp_is_sackblock_valid()
1421 if (!before(start_seq, tp->snd_nxt)) in tcp_is_sackblock_valid()
1424 /* In outstanding window? ...This is valid exit for D-SACKs too. in tcp_is_sackblock_valid()
1425 * start_seq == snd_una is non-sensical (see comments above) in tcp_is_sackblock_valid()
1427 if (after(start_seq, tp->snd_una)) in tcp_is_sackblock_valid()
1430 if (!is_dsack || !tp->undo_marker) in tcp_is_sackblock_valid()
1433 /* ...Then it's D-SACK, and must reside below snd_una completely */ in tcp_is_sackblock_valid()
1434 if (after(end_seq, tp->snd_una)) in tcp_is_sackblock_valid()
1437 if (!before(start_seq, tp->undo_marker)) in tcp_is_sackblock_valid()
1441 if (!after(end_seq, tp->undo_marker)) in tcp_is_sackblock_valid()
1447 return !before(start_seq, end_seq - tp->max_window); in tcp_is_sackblock_valid()
1459 if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) { in tcp_check_dsack()
1480 /* D-SACK for already forgotten data... Do dumb counting. */ in tcp_check_dsack()
1481 if (tp->undo_marker && tp->undo_retrans > 0 && in tcp_check_dsack()
1483 after(end_seq_0, tp->undo_marker)) in tcp_check_dsack()
1484 tp->undo_retrans = max_t(int, 0, tp->undo_retrans - dup_segs); in tcp_check_dsack()
1505 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && in tcp_match_skb_to_sack()
1506 !before(end_seq, TCP_SKB_CB(skb)->end_seq); in tcp_match_skb_to_sack()
1509 after(TCP_SKB_CB(skb)->end_seq, start_seq)) { in tcp_match_skb_to_sack()
1511 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); in tcp_match_skb_to_sack()
1514 pkt_len = start_seq - TCP_SKB_CB(skb)->seq; in tcp_match_skb_to_sack()
1518 pkt_len = end_seq - TCP_SKB_CB(skb)->seq; in tcp_match_skb_to_sack()
1520 return -EINVAL; in tcp_match_skb_to_sack()
1533 if (pkt_len >= skb->len && !in_sack) in tcp_match_skb_to_sack()
1545 /* Mark the given newly-SACKed range as such, adjusting counters and hints. */
1554 /* Account D-SACK for retransmitted packet. */ in tcp_sacktag_one()
1556 if (tp->undo_marker && tp->undo_retrans > 0 && in tcp_sacktag_one()
1557 after(end_seq, tp->undo_marker)) in tcp_sacktag_one()
1558 tp->undo_retrans = max_t(int, 0, tp->undo_retrans - pcount); in tcp_sacktag_one()
1560 before(start_seq, state->reord)) in tcp_sacktag_one()
1561 state->reord = start_seq; in tcp_sacktag_one()
1565 if (!after(end_seq, tp->snd_una)) in tcp_sacktag_one()
1578 tp->lost_out -= pcount; in tcp_sacktag_one()
1579 tp->retrans_out -= pcount; in tcp_sacktag_one()
1588 before(start_seq, state->reord)) in tcp_sacktag_one()
1589 state->reord = start_seq; in tcp_sacktag_one()
1591 if (!after(end_seq, tp->high_seq)) in tcp_sacktag_one()
1592 state->flag |= FLAG_ORIG_SACK_ACKED; in tcp_sacktag_one()
1593 if (state->first_sackt == 0) in tcp_sacktag_one()
1594 state->first_sackt = xmit_time; in tcp_sacktag_one()
1595 state->last_sackt = xmit_time; in tcp_sacktag_one()
1600 tp->lost_out -= pcount; in tcp_sacktag_one()
1605 state->flag |= FLAG_DATA_SACKED; in tcp_sacktag_one()
1606 tp->sacked_out += pcount; in tcp_sacktag_one()
1607 /* Out-of-order packets delivered */ in tcp_sacktag_one()
1608 state->sack_delivered += pcount; in tcp_sacktag_one()
1609 state->delivered_bytes += plen; in tcp_sacktag_one()
1612 /* D-SACK. We can detect redundant retransmission in S|R and plain R in tcp_sacktag_one()
1618 tp->retrans_out -= pcount; in tcp_sacktag_one()
1624 /* Shift newly-SACKed bytes from this skb to the immediately previous
1625 * already-SACKed sk_buff. Mark the newly-SACKed bytes as such.
1634 u32 start_seq = TCP_SKB_CB(skb)->seq; /* start of newly-SACKed */ in tcp_shifted_skb()
1635 u32 end_seq = start_seq + shifted; /* end of newly-SACKed */ in tcp_shifted_skb()
1645 tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked, in tcp_shifted_skb()
1646 start_seq, end_seq, dup_sack, pcount, skb->len, in tcp_shifted_skb()
1648 tcp_rate_skb_delivered(sk, skb, state->rate); in tcp_shifted_skb()
1650 TCP_SKB_CB(prev)->end_seq += shifted; in tcp_shifted_skb()
1651 TCP_SKB_CB(skb)->seq += shifted; in tcp_shifted_skb()
1655 tcp_skb_pcount_add(skb, -pcount); in tcp_shifted_skb()
1662 if (!TCP_SKB_CB(prev)->tcp_gso_size) in tcp_shifted_skb()
1663 TCP_SKB_CB(prev)->tcp_gso_size = mss; in tcp_shifted_skb()
1667 TCP_SKB_CB(skb)->tcp_gso_size = 0; in tcp_shifted_skb()
1670 TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS); in tcp_shifted_skb()
1672 if (skb->len > 0) { in tcp_shifted_skb()
1678 /* Whole SKB was eaten :-) */ in tcp_shifted_skb()
1680 if (skb == tp->retransmit_skb_hint) in tcp_shifted_skb()
1681 tp->retransmit_skb_hint = prev; in tcp_shifted_skb()
1683 TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; in tcp_shifted_skb()
1684 TCP_SKB_CB(prev)->eor = TCP_SKB_CB(skb)->eor; in tcp_shifted_skb()
1685 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_shifted_skb()
1686 TCP_SKB_CB(prev)->end_seq++; in tcp_shifted_skb()
1692 if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp)) in tcp_shifted_skb()
1693 TCP_SKB_CB(prev)->tx.delivered_mstamp = 0; in tcp_shifted_skb()
1703 * something-or-zero which complicates things
1707 return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb); in tcp_skb_seglen()
1720 * Since TCP_SKB_CB(skb)->tcp_gso_segs is 16 bits, we need in tcp_skb_shift()
1724 if (unlikely(to->len + shiftlen >= 65535 * TCP_MIN_GSO_SIZE)) in tcp_skb_shift()
1748 (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS) in tcp_shift_skb_data()
1753 if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)) in tcp_shift_skb_data()
1761 if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) in tcp_shift_skb_data()
1767 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) && in tcp_shift_skb_data()
1768 !before(end_seq, TCP_SKB_CB(skb)->end_seq); in tcp_shift_skb_data()
1771 len = skb->len; in tcp_shift_skb_data()
1781 if (!after(TCP_SKB_CB(skb)->end_seq, start_seq)) in tcp_shift_skb_data()
1783 /* CHECKME: This is non-MSS split case only?, this will in tcp_shift_skb_data()
1790 in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq); in tcp_shift_skb_data()
1793 * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)), in tcp_shift_skb_data()
1797 * previously. We could try merging non-SACKed ones in tcp_shift_skb_data()
1806 len = end_seq - TCP_SKB_CB(skb)->seq; in tcp_shift_skb_data()
1808 BUG_ON(len > skb->len); in tcp_shift_skb_data()
1832 /* tcp_sacktag_one() won't SACK-tag ranges below snd_una */ in tcp_shift_skb_data()
1833 if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una)) in tcp_shift_skb_data()
1849 ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) || in tcp_shift_skb_data()
1855 len = skb->len; in tcp_shift_skb_data()
1885 /* queue is in-order => we can short-circuit the walk early */ in tcp_sacktag_walk()
1886 if (!before(TCP_SKB_CB(skb)->seq, end_seq)) in tcp_sacktag_walk()
1890 before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) { in tcp_sacktag_walk()
1892 next_dup->start_seq, in tcp_sacktag_walk()
1893 next_dup->end_seq); in tcp_sacktag_walk()
1923 TCP_SKB_CB(skb)->sacked = in tcp_sacktag_walk()
1926 TCP_SKB_CB(skb)->sacked, in tcp_sacktag_walk()
1927 TCP_SKB_CB(skb)->seq, in tcp_sacktag_walk()
1928 TCP_SKB_CB(skb)->end_seq, in tcp_sacktag_walk()
1931 skb->len, in tcp_sacktag_walk()
1933 tcp_rate_skb_delivered(sk, skb, state->rate); in tcp_sacktag_walk()
1934 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_sacktag_walk()
1935 list_del_init(&skb->tcp_tsorted_anchor); in tcp_sacktag_walk()
1937 if (!before(TCP_SKB_CB(skb)->seq, in tcp_sacktag_walk()
1947 struct rb_node *parent, **p = &sk->tcp_rtx_queue.rb_node; in tcp_sacktag_bsearch()
1953 if (before(seq, TCP_SKB_CB(skb)->seq)) { in tcp_sacktag_bsearch()
1954 p = &parent->rb_left; in tcp_sacktag_bsearch()
1957 if (!before(seq, TCP_SKB_CB(skb)->end_seq)) { in tcp_sacktag_bsearch()
1958 p = &parent->rb_right; in tcp_sacktag_bsearch()
1969 if (skb && after(TCP_SKB_CB(skb)->seq, skip_to_seq)) in tcp_sacktag_skip()
1984 if (before(next_dup->start_seq, skip_to_seq)) { in tcp_maybe_skipping_dsack()
1985 skb = tcp_sacktag_skip(skb, sk, next_dup->start_seq); in tcp_maybe_skipping_dsack()
1987 next_dup->start_seq, next_dup->end_seq, in tcp_maybe_skipping_dsack()
1996 return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); in tcp_sack_cache_ok()
2005 TCP_SKB_CB(ack_skb)->sacked); in tcp_sacktag_write_queue()
2010 int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3); in tcp_sacktag_write_queue()
2016 state->flag = 0; in tcp_sacktag_write_queue()
2017 state->reord = tp->snd_nxt; in tcp_sacktag_write_queue()
2019 if (!tp->sacked_out) in tcp_sacktag_write_queue()
2029 if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window)) in tcp_sacktag_write_queue()
2032 if (!tp->packets_out) in tcp_sacktag_write_queue()
2049 if (!tp->undo_marker) in tcp_sacktag_write_queue()
2055 if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) && in tcp_sacktag_write_queue()
2056 !after(sp[used_sacks].end_seq, tp->snd_una)) in tcp_sacktag_write_queue()
2063 first_sack_index = -1; in tcp_sacktag_write_queue()
2070 first_sack_index = -1; in tcp_sacktag_write_queue()
2078 for (i = used_sacks - 1; i > 0; i--) { in tcp_sacktag_write_queue()
2090 state->mss_now = tcp_current_mss(sk); in tcp_sacktag_write_queue()
2094 if (!tp->sacked_out) { in tcp_sacktag_write_queue()
2096 cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache); in tcp_sacktag_write_queue()
2098 cache = tp->recv_sack_cache; in tcp_sacktag_write_queue()
2100 while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq && in tcp_sacktag_write_queue()
2101 !cache->end_seq) in tcp_sacktag_write_queue()
2116 !before(start_seq, cache->end_seq)) in tcp_sacktag_write_queue()
2121 after(end_seq, cache->start_seq)) { in tcp_sacktag_write_queue()
2124 if (before(start_seq, cache->start_seq)) { in tcp_sacktag_write_queue()
2129 cache->start_seq, in tcp_sacktag_write_queue()
2134 if (!after(end_seq, cache->end_seq)) in tcp_sacktag_write_queue()
2139 cache->end_seq); in tcp_sacktag_write_queue()
2142 if (tcp_highest_sack_seq(tp) == cache->end_seq) { in tcp_sacktag_write_queue()
2151 skb = tcp_sacktag_skip(skb, sk, cache->end_seq); in tcp_sacktag_write_queue()
2173 for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) { in tcp_sacktag_write_queue()
2174 tp->recv_sack_cache[i].start_seq = 0; in tcp_sacktag_write_queue()
2175 tp->recv_sack_cache[i].end_seq = 0; in tcp_sacktag_write_queue()
2178 tp->recv_sack_cache[i++] = sp[j]; in tcp_sacktag_write_queue()
2180 if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss || tp->undo_marker) in tcp_sacktag_write_queue()
2181 tcp_check_sack_reordering(sk, state->reord, 0); in tcp_sacktag_write_queue()
2187 WARN_ON((int)tp->sacked_out < 0); in tcp_sacktag_write_queue()
2188 WARN_ON((int)tp->lost_out < 0); in tcp_sacktag_write_queue()
2189 WARN_ON((int)tp->retrans_out < 0); in tcp_sacktag_write_queue()
2192 return state->flag; in tcp_sacktag_write_queue()
2202 holes = max(tp->lost_out, 1U); in tcp_limit_reno_sacked()
2203 holes = min(holes, tp->packets_out); in tcp_limit_reno_sacked()
2205 if ((tp->sacked_out + holes) > tp->packets_out) { in tcp_limit_reno_sacked()
2206 tp->sacked_out = tp->packets_out - holes; in tcp_limit_reno_sacked()
2223 tp->reordering = min_t(u32, tp->packets_out + addend, in tcp_check_reno_reordering()
2224 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering)); in tcp_check_reno_reordering()
2225 tp->reord_seen++; in tcp_check_reno_reordering()
2235 u32 prior_sacked = tp->sacked_out; in tcp_add_reno_sack()
2238 tp->sacked_out += num_dupack; in tcp_add_reno_sack()
2240 delivered = tp->sacked_out - prior_sacked; in tcp_add_reno_sack()
2255 tcp_count_delivered(tp, max_t(int, acked - tp->sacked_out, 1), in tcp_remove_reno_sacks()
2257 if (acked - 1 >= tp->sacked_out) in tcp_remove_reno_sacks()
2258 tp->sacked_out = 0; in tcp_remove_reno_sacks()
2260 tp->sacked_out -= acked - 1; in tcp_remove_reno_sacks()
2268 tp->sacked_out = 0; in tcp_reset_reno_sack()
2273 tp->retrans_out = 0; in tcp_clear_retrans()
2274 tp->lost_out = 0; in tcp_clear_retrans()
2275 tp->undo_marker = 0; in tcp_clear_retrans()
2276 tp->undo_retrans = -1; in tcp_clear_retrans()
2277 tp->sacked_out = 0; in tcp_clear_retrans()
2278 tp->rto_stamp = 0; in tcp_clear_retrans()
2279 tp->total_rto = 0; in tcp_clear_retrans()
2280 tp->total_rto_recoveries = 0; in tcp_clear_retrans()
2281 tp->total_rto_time = 0; in tcp_clear_retrans()
2286 tp->undo_marker = tp->snd_una; in tcp_init_undo()
2290 tp->undo_retrans = tp->retrans_out; in tcp_init_undo()
2292 if (tp->tlp_high_seq && tp->tlp_retrans) in tcp_init_undo()
2293 tp->undo_retrans++; in tcp_init_undo()
2295 if (!tp->undo_retrans) in tcp_init_undo()
2296 tp->undo_retrans = -1; in tcp_init_undo()
2310 is_reneg = head && (TCP_SKB_CB(head)->sacked & TCPCB_SACKED_ACKED); in tcp_timeout_mark_lost()
2313 tp->sacked_out = 0; in tcp_timeout_mark_lost()
2315 tp->is_sack_reneg = 1; in tcp_timeout_mark_lost()
2323 TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED; in tcp_timeout_mark_lost()
2338 bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery; in tcp_enter_loss()
2344 if (icsk->icsk_ca_state <= TCP_CA_Disorder || in tcp_enter_loss()
2345 !after(tp->high_seq, tp->snd_una) || in tcp_enter_loss()
2346 (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) { in tcp_enter_loss()
2347 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_enter_loss()
2348 tp->prior_cwnd = tcp_snd_cwnd(tp); in tcp_enter_loss()
2349 tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk); in tcp_enter_loss()
2354 tp->snd_cwnd_cnt = 0; in tcp_enter_loss()
2355 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_enter_loss()
2358 * suggests that the degree of reordering is over-estimated. in tcp_enter_loss()
2360 reordering = READ_ONCE(net->ipv4.sysctl_tcp_reordering); in tcp_enter_loss()
2361 if (icsk->icsk_ca_state <= TCP_CA_Disorder && in tcp_enter_loss()
2362 tp->sacked_out >= reordering) in tcp_enter_loss()
2363 tp->reordering = min_t(unsigned int, tp->reordering, in tcp_enter_loss()
2367 tp->high_seq = tp->snd_nxt; in tcp_enter_loss()
2368 tp->tlp_high_seq = 0; in tcp_enter_loss()
2371 /* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous in tcp_enter_loss()
2373 * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing in tcp_enter_loss()
2375 tp->frto = READ_ONCE(net->ipv4.sysctl_tcp_frto) && in tcp_enter_loss()
2376 (new_recovery || icsk->icsk_retransmits) && in tcp_enter_loss()
2377 !inet_csk(sk)->icsk_mtup.probe_size; in tcp_enter_loss()
2395 unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4), in tcp_check_sack_reneging()
2406 * --------------------------------------
2415 * "Recovery" CWND was reduced, we are fast-retransmitting.
2419 * - each incoming ACK, if state is not "Open"
2420 * - when arrived ACK is unusual, namely:
2427 * in_flight = packets_out - left_out + retrans_out
2429 * packets_out is SND.NXT-SND.UNA counted in packets.
2458 * (2017-) that checks timing instead of counting DUPACKs.
2484 * timestamps and D-SACKs. When we detect that some segments were
2499 return tp->lost_out != 0; in tcp_time_to_recover()
2504 return tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && in tcp_tsopt_ecr_before()
2505 before(tp->rx_opt.rcv_tsecr, when); in tcp_tsopt_ecr_before()
2514 return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) && in tcp_skb_spurious_retrans()
2515 tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb)); in tcp_skb_spurious_retrans()
2526 if (tp->retrans_stamp) in tcp_packet_delayed()
2527 return tcp_tsopt_ecr_before(tp, tp->retrans_stamp); in tcp_packet_delayed()
2529 /* We set tp->retrans_stamp upon the first retransmission of a loss in tcp_packet_delayed()
2530 * recovery episode, so normally if tp->retrans_stamp is 0 then no in tcp_packet_delayed()
2533 * (tp->retrans_stamp is 0 then apparently a packet was merely delayed, in tcp_packet_delayed()
2535 * clear tp->retrans_stamp, so we check for those exceptions. in tcp_packet_delayed()
2538 /* (1) For non-SACK connections, tcp_is_non_sack_preventing_reopen() in tcp_packet_delayed()
2539 * clears tp->retrans_stamp when snd_una == high_seq. in tcp_packet_delayed()
2541 if (!tcp_is_sack(tp) && !before(tp->snd_una, tp->high_seq)) in tcp_packet_delayed()
2544 /* (2) In TCP_SYN_SENT tcp_clean_rtx_queue() clears tp->retrans_stamp in tcp_packet_delayed()
2548 if (sk->sk_state == TCP_SYN_SENT) in tcp_packet_delayed()
2551 return true; /* tp->retrans_stamp is zero; no retransmit yet */ in tcp_packet_delayed()
2558 * tp->retrans_out, however, that kind of assumptions doesn't consider
2575 if (tp->retrans_out) in tcp_any_retrans_done()
2579 if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS)) in tcp_any_retrans_done()
2587 * retransmits_timed_out() and timestamp-undo are using the correct value.
2592 tcp_sk(sk)->retrans_stamp = 0; in tcp_retrans_stamp_cleanup()
2601 if (sk->sk_family == AF_INET) { in DBGUNDO()
2604 &inet->inet_daddr, ntohs(inet->inet_dport), in DBGUNDO()
2606 tp->snd_ssthresh, tp->prior_ssthresh, in DBGUNDO()
2607 tp->packets_out); in DBGUNDO()
2610 else if (sk->sk_family == AF_INET6) { in DBGUNDO()
2613 &sk->sk_v6_daddr, ntohs(inet->inet_dport), in DBGUNDO()
2615 tp->snd_ssthresh, tp->prior_ssthresh, in DBGUNDO()
2616 tp->packets_out); in DBGUNDO()
2629 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { in tcp_undo_cwnd_reduction()
2630 TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST; in tcp_undo_cwnd_reduction()
2632 tp->lost_out = 0; in tcp_undo_cwnd_reduction()
2636 if (tp->prior_ssthresh) { in tcp_undo_cwnd_reduction()
2639 tcp_snd_cwnd_set(tp, icsk->icsk_ca_ops->undo_cwnd(sk)); in tcp_undo_cwnd_reduction()
2641 if (tp->prior_ssthresh > tp->snd_ssthresh) { in tcp_undo_cwnd_reduction()
2642 tp->snd_ssthresh = tp->prior_ssthresh; in tcp_undo_cwnd_reduction()
2646 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_undo_cwnd_reduction()
2647 tp->undo_marker = 0; in tcp_undo_cwnd_reduction()
2648 tp->rack.advanced = 1; /* Force RACK to re-exam losses */ in tcp_undo_cwnd_reduction()
2653 return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp)); in tcp_may_undo()
2660 if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) { in tcp_is_non_sack_preventing_reopen()
2665 tp->retrans_stamp = 0; in tcp_is_non_sack_preventing_reopen()
2682 DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans"); in tcp_try_undo_recovery()
2684 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss) in tcp_try_undo_recovery()
2690 } else if (tp->rack.reo_wnd_persist) { in tcp_try_undo_recovery()
2691 tp->rack.reo_wnd_persist--; in tcp_try_undo_recovery()
2696 tp->is_sack_reneg = 0; in tcp_try_undo_recovery()
2700 /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
2705 if (tp->undo_marker && !tp->undo_retrans) { in tcp_try_undo_dsack()
2706 tp->rack.reo_wnd_persist = min(TCP_RACK_RECOVERY_THRESH, in tcp_try_undo_dsack()
2707 tp->rack.reo_wnd_persist + 1); in tcp_try_undo_dsack()
2708 DBGUNDO(sk, "D-SACK"); in tcp_try_undo_dsack()
2716 /* Undo during loss recovery after partial ACK or using F-RTO. */
2729 WRITE_ONCE(inet_csk(sk)->icsk_retransmits, 0); in tcp_try_undo_loss()
2734 tp->is_sack_reneg = 0; in tcp_try_undo_loss()
2754 tp->high_seq = tp->snd_nxt; in tcp_init_cwnd_reduction()
2755 tp->tlp_high_seq = 0; in tcp_init_cwnd_reduction()
2756 tp->snd_cwnd_cnt = 0; in tcp_init_cwnd_reduction()
2757 tp->prior_cwnd = tcp_snd_cwnd(tp); in tcp_init_cwnd_reduction()
2758 tp->prr_delivered = 0; in tcp_init_cwnd_reduction()
2759 tp->prr_out = 0; in tcp_init_cwnd_reduction()
2760 tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk); in tcp_init_cwnd_reduction()
2768 int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); in tcp_cwnd_reduction()
2770 if (newly_acked_sacked <= 0 || WARN_ON_ONCE(!tp->prior_cwnd)) in tcp_cwnd_reduction()
2775 tp->prr_delivered += newly_acked_sacked; in tcp_cwnd_reduction()
2777 u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered + in tcp_cwnd_reduction()
2778 tp->prior_cwnd - 1; in tcp_cwnd_reduction()
2779 sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out; in tcp_cwnd_reduction()
2781 sndcnt = max_t(int, tp->prr_delivered - tp->prr_out, in tcp_cwnd_reduction()
2788 sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1)); in tcp_cwnd_reduction()
2796 if (inet_csk(sk)->icsk_ca_ops->cong_control) in tcp_end_cwnd_reduction()
2800 if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH && in tcp_end_cwnd_reduction()
2801 (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) { in tcp_end_cwnd_reduction()
2802 tcp_snd_cwnd_set(tp, tp->snd_ssthresh); in tcp_end_cwnd_reduction()
2803 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_end_cwnd_reduction()
2813 tp->prior_ssthresh = 0; in tcp_enter_cwr()
2814 if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) { in tcp_enter_cwr()
2815 tp->undo_marker = 0; in tcp_enter_cwr()
2830 if (inet_csk(sk)->icsk_ca_state != state) { in tcp_try_keep_open()
2832 tp->high_seq = tp->snd_nxt; in tcp_try_keep_open()
2843 tp->retrans_stamp = 0; in tcp_try_to_open()
2848 if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) { in tcp_try_to_open()
2857 icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1; in tcp_mtup_probe_failed()
2858 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_probe_failed()
2868 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_mtup_probe_success()
2870 val = (u64)tcp_snd_cwnd(tp) * tcp_mss_to_mtu(sk, tp->mss_cache); in tcp_mtup_probe_success()
2871 do_div(val, icsk->icsk_mtup.probe_size); in tcp_mtup_probe_success()
2875 tp->snd_cwnd_cnt = 0; in tcp_mtup_probe_success()
2876 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_mtup_probe_success()
2877 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_mtup_probe_success()
2879 icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size; in tcp_mtup_probe_success()
2880 icsk->icsk_mtup.probe_size = 0; in tcp_mtup_probe_success()
2881 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_mtup_probe_success()
2889 * non-zero value (and may do so in a later calling context due to TSQ), we
2900 if (icsk->icsk_ca_state != TCP_CA_Loss) { in tcp_non_congestion_loss_retransmit()
2901 tp->high_seq = tp->snd_nxt; in tcp_non_congestion_loss_retransmit()
2902 tp->snd_ssthresh = tcp_current_ssthresh(sk); in tcp_non_congestion_loss_retransmit()
2903 tp->prior_ssthresh = 0; in tcp_non_congestion_loss_retransmit()
2904 tp->undo_marker = 0; in tcp_non_congestion_loss_retransmit()
2928 * -1 to mark all frames as lost, otherwise compute the current MSS. in tcp_simple_retransmit()
2930 if (tp->syn_data && sk->sk_state == TCP_SYN_SENT) in tcp_simple_retransmit()
2931 mss = -1; in tcp_simple_retransmit()
2935 skb_rbtree_walk(skb, &sk->tcp_rtx_queue) { in tcp_simple_retransmit()
2940 if (!tp->lost_out) in tcp_simple_retransmit()
2972 tp->prior_ssthresh = 0; in tcp_enter_recovery()
2977 tp->prior_ssthresh = tcp_current_ssthresh(sk); in tcp_enter_recovery()
2985 if (tp->rto_stamp) { in tcp_update_rto_time()
2986 tp->total_rto_time += tcp_time_stamp_ms(tp) - tp->rto_stamp; in tcp_update_rto_time()
2987 tp->rto_stamp = 0; in tcp_update_rto_time()
2998 bool recovered = !before(tp->snd_una, tp->high_seq); in tcp_process_loss()
3000 if ((flag & FLAG_SND_UNA_ADVANCED || rcu_access_pointer(tp->fastopen_rsk)) && in tcp_process_loss()
3004 if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */ in tcp_process_loss()
3006 * lost, i.e., never-retransmitted data are (s)acked. in tcp_process_loss()
3012 if (after(tp->snd_nxt, tp->high_seq)) { in tcp_process_loss()
3014 tp->frto = 0; /* Step 3.a. loss was real */ in tcp_process_loss()
3016 tp->high_seq = tp->snd_nxt; in tcp_process_loss()
3022 after(tcp_wnd_end(tp), tp->snd_nxt)) { in tcp_process_loss()
3026 tp->frto = 0; in tcp_process_loss()
3031 /* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */ in tcp_process_loss()
3036 /* A Reno DUPACK means new data in F-RTO step 2.b above are in tcp_process_loss()
3039 if (after(tp->snd_nxt, tp->high_seq) && num_dupack) in tcp_process_loss()
3052 if (tp->undo_marker && tcp_packet_delayed(tp)) { in tcp_try_undo_partial()
3063 if (tp->retrans_out) in tcp_try_undo_partial()
3067 tp->retrans_stamp = 0; in tcp_try_undo_partial()
3087 u32 prior_retrans = tp->retrans_out; in tcp_identify_packet_loss()
3091 if (prior_retrans > tp->retrans_out) in tcp_identify_packet_loss()
3096 /* Process an event, which can update packets-in-flight not trivially.
3116 if (!tp->packets_out && tp->sacked_out) in tcp_fastretrans_alert()
3117 tp->sacked_out = 0; in tcp_fastretrans_alert()
3122 tp->prior_ssthresh = 0; in tcp_fastretrans_alert()
3133 if (icsk->icsk_ca_state == TCP_CA_Open) { in tcp_fastretrans_alert()
3134 WARN_ON(tp->retrans_out != 0 && !tp->syn_data); in tcp_fastretrans_alert()
3135 tp->retrans_stamp = 0; in tcp_fastretrans_alert()
3136 } else if (!before(tp->snd_una, tp->high_seq)) { in tcp_fastretrans_alert()
3137 switch (icsk->icsk_ca_state) { in tcp_fastretrans_alert()
3141 if (tp->snd_una != tp->high_seq) { in tcp_fastretrans_alert()
3158 switch (icsk->icsk_ca_state) { in tcp_fastretrans_alert()
3170 if (icsk->icsk_ca_state != TCP_CA_Recovery) { in tcp_fastretrans_alert()
3181 if (icsk->icsk_ca_state != TCP_CA_Loss) in tcp_fastretrans_alert()
3184 if (!(icsk->icsk_ca_state == TCP_CA_Open || in tcp_fastretrans_alert()
3196 if (icsk->icsk_ca_state <= TCP_CA_Disorder) in tcp_fastretrans_alert()
3206 if (icsk->icsk_ca_state < TCP_CA_CWR && in tcp_fastretrans_alert()
3207 icsk->icsk_mtup.probe_size && in tcp_fastretrans_alert()
3208 tp->snd_una == tp->mtu_probe.probe_seq_start) { in tcp_fastretrans_alert()
3225 u32 wlen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen) * HZ; in tcp_update_rtt_min()
3235 minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32, in tcp_update_rtt_min()
3245 /* Prefer RTT measured from ACK's timing to TS-ECR. This is because in tcp_ack_update_rtt()
3246 * broken middle-boxes or peers may corrupt TS-ECR fields. But in tcp_ack_update_rtt()
3257 * See draft-ietf-tcplw-high-performance-00, section 3.3. in tcp_ack_update_rtt()
3259 if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp && in tcp_ack_update_rtt()
3260 tp->rx_opt.rcv_tsecr && flag & FLAG_ACKED) in tcp_ack_update_rtt()
3263 rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */ in tcp_ack_update_rtt()
3268 * always taken together with ACK, SACK, or TS-opts. Any negative in tcp_ack_update_rtt()
3276 inet_csk(sk)->icsk_backoff = 0; in tcp_ack_update_rtt()
3284 long rtt_us = -1L; in tcp_synack_rtt_meas()
3286 if (req && !req->num_retrans && tcp_rsk(req)->snt_synack) in tcp_synack_rtt_meas()
3287 rtt_us = tcp_stamp_us_delta(tcp_clock_us(), tcp_rsk(req)->snt_synack); in tcp_synack_rtt_meas()
3289 tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, rtt_us, -1L, rtt_us, &rs); in tcp_synack_rtt_meas()
3297 icsk->icsk_ca_ops->cong_avoid(sk, ack, acked); in tcp_cong_avoid()
3298 tcp_sk(sk)->snd_cwnd_stamp = tcp_jiffies32; in tcp_cong_avoid()
3310 * for SYN-ACK retrans purpose, stay put. in tcp_rearm_rto()
3312 if (rcu_access_pointer(tp->fastopen_rsk)) in tcp_rearm_rto()
3315 if (!tp->packets_out) { in tcp_rearm_rto()
3318 u32 rto = inet_csk(sk)->icsk_rto; in tcp_rearm_rto()
3320 if (icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || in tcp_rearm_rto()
3321 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { in tcp_rearm_rto()
3345 BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una)); in tcp_tso_acked()
3348 if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq)) in tcp_tso_acked()
3350 packets_acked -= tcp_skb_pcount(skb); in tcp_tso_acked()
3354 BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq)); in tcp_tso_acked()
3365 /* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */ in tcp_ack_tstamp()
3366 if (likely(!TCP_SKB_CB(skb)->txstamp_ack)) in tcp_ack_tstamp()
3370 if (!before(shinfo->tskey, prior_snd_una) && in tcp_ack_tstamp()
3371 before(shinfo->tskey, tcp_sk(sk)->snd_una)) { in tcp_ack_tstamp()
3389 u32 prior_sacked = tp->sacked_out; in tcp_clean_rtx_queue()
3390 u32 reord = tp->snd_nxt; /* lowest acked un-retx un-sacked seq */ in tcp_clean_rtx_queue()
3393 long sack_rtt_us = -1L; in tcp_clean_rtx_queue()
3394 long seq_rtt_us = -1L; in tcp_clean_rtx_queue()
3395 long ca_rtt_us = -1L; in tcp_clean_rtx_queue()
3402 for (skb = skb_rb_first(&sk->tcp_rtx_queue); skb; skb = next) { in tcp_clean_rtx_queue()
3404 const u32 start_seq = scb->seq; in tcp_clean_rtx_queue()
3405 u8 sacked = scb->sacked; in tcp_clean_rtx_queue()
3409 if (after(scb->end_seq, tp->snd_una)) { in tcp_clean_rtx_queue()
3411 !after(tp->snd_una, scb->seq)) in tcp_clean_rtx_queue()
3424 tp->retrans_out -= acked_pcount; in tcp_clean_rtx_queue()
3434 if (!after(scb->end_seq, tp->high_seq)) in tcp_clean_rtx_queue()
3439 tp->sacked_out -= acked_pcount; in tcp_clean_rtx_queue()
3441 sack->delivered_bytes -= skb->len; in tcp_clean_rtx_queue()
3445 tcp_rack_advance(tp, sacked, scb->end_seq, in tcp_clean_rtx_queue()
3449 tp->lost_out -= acked_pcount; in tcp_clean_rtx_queue()
3451 tp->packets_out -= acked_pcount; in tcp_clean_rtx_queue()
3453 tcp_rate_skb_delivered(sk, skb, sack->rate); in tcp_clean_rtx_queue()
3462 if (likely(!(scb->tcp_flags & TCPHDR_SYN))) { in tcp_clean_rtx_queue()
3466 tp->retrans_stamp = 0; in tcp_clean_rtx_queue()
3475 if (unlikely(skb == tp->retransmit_skb_hint)) in tcp_clean_rtx_queue()
3476 tp->retransmit_skb_hint = NULL; in tcp_clean_rtx_queue()
3484 if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una))) in tcp_clean_rtx_queue()
3485 tp->snd_up = tp->snd_una; in tcp_clean_rtx_queue()
3489 if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED) in tcp_clean_rtx_queue()
3494 seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt); in tcp_clean_rtx_queue()
3495 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt); in tcp_clean_rtx_queue()
3498 (tp->snd_una - prior_snd_una) < tp->mss_cache && in tcp_clean_rtx_queue()
3499 sack->rate->prior_delivered + 1 == tp->delivered && in tcp_clean_rtx_queue()
3503 * a receiver w/o out-of-order or CE events. in tcp_clean_rtx_queue()
3508 if (sack->first_sackt) { in tcp_clean_rtx_queue()
3509 sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt); in tcp_clean_rtx_queue()
3510 ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->last_sackt); in tcp_clean_rtx_queue()
3513 ca_rtt_us, sack->rate); in tcp_clean_rtx_queue()
3517 if (unlikely(icsk->icsk_mtup.probe_size && in tcp_clean_rtx_queue()
3518 !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) { in tcp_clean_rtx_queue()
3526 * retransmitted, non-SACK case cannot confirm that in tcp_clean_rtx_queue()
3534 /* Non-retransmitted hole got filled? That's reordering */ in tcp_clean_rtx_queue()
3539 sack->delivered_bytes = (skb ? in tcp_clean_rtx_queue()
3540 TCP_SKB_CB(skb)->seq : tp->snd_una) - in tcp_clean_rtx_queue()
3543 sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp, in tcp_clean_rtx_queue()
3545 /* Do not re-arm RTO if the sack RTT is measured from data sent in tcp_clean_rtx_queue()
3552 if (icsk->icsk_ca_ops->pkts_acked) { in tcp_clean_rtx_queue()
3554 .rtt_us = sack->rate->rtt_us }; in tcp_clean_rtx_queue()
3556 sample.in_flight = tp->mss_cache * in tcp_clean_rtx_queue()
3557 (tp->delivered - sack->rate->prior_delivered); in tcp_clean_rtx_queue()
3558 icsk->icsk_ca_ops->pkts_acked(sk, &sample); in tcp_clean_rtx_queue()
3562 WARN_ON((int)tp->sacked_out < 0); in tcp_clean_rtx_queue()
3563 WARN_ON((int)tp->lost_out < 0); in tcp_clean_rtx_queue()
3564 WARN_ON((int)tp->retrans_out < 0); in tcp_clean_rtx_queue()
3565 if (!tp->packets_out && tcp_is_sack(tp)) { in tcp_clean_rtx_queue()
3567 if (tp->lost_out) { in tcp_clean_rtx_queue()
3569 tp->lost_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3570 tp->lost_out = 0; in tcp_clean_rtx_queue()
3572 if (tp->sacked_out) { in tcp_clean_rtx_queue()
3574 tp->sacked_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3575 tp->sacked_out = 0; in tcp_clean_rtx_queue()
3577 if (tp->retrans_out) { in tcp_clean_rtx_queue()
3579 tp->retrans_out, icsk->icsk_ca_state); in tcp_clean_rtx_queue()
3580 tp->retrans_out = 0; in tcp_clean_rtx_queue()
3596 if (!after(TCP_SKB_CB(head)->end_seq, tcp_wnd_end(tp))) { in tcp_ack_probe()
3597 icsk->icsk_backoff = 0; in tcp_ack_probe()
3598 icsk->icsk_probes_tstamp = 0; in tcp_ack_probe()
3614 inet_csk(sk)->icsk_ca_state != TCP_CA_Open; in tcp_ack_is_dubious()
3622 * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/ in tcp_may_raise_cwnd()
3626 if (tcp_sk(sk)->reordering > in tcp_may_raise_cwnd()
3627 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering)) in tcp_may_raise_cwnd()
3643 if (icsk->icsk_ca_ops->cong_control) { in tcp_cong_control()
3644 icsk->icsk_ca_ops->cong_control(sk, ack, flag, rs); in tcp_cong_control()
3650 tcp_cwnd_reduction(sk, acked_sacked, rs->losses, flag); in tcp_cong_control()
3665 return after(ack, tp->snd_una) || in tcp_may_update_window()
3666 after(ack_seq, tp->snd_wl1) || in tcp_may_update_window()
3667 (ack_seq == tp->snd_wl1 && (nwin > tp->snd_wnd || !nwin)); in tcp_may_update_window()
3678 ao = rcu_dereference_protected(tp->ao_info, in tcp_snd_sne_update()
3680 if (ao && ack < tp->snd_una) { in tcp_snd_sne_update()
3681 ao->snd_sne++; in tcp_snd_sne_update()
3682 trace_tcp_ao_snd_sne_update((struct sock *)tp, ao->snd_sne); in tcp_snd_sne_update()
3687 /* If we update tp->snd_una, also update tp->bytes_acked */
3690 u32 delta = ack - tp->snd_una; in tcp_snd_una_update()
3693 tp->bytes_acked += delta; in tcp_snd_una_update()
3695 tp->snd_una = ack; in tcp_snd_una_update()
3706 ao = rcu_dereference_protected(tp->ao_info, in tcp_rcv_sne_update()
3708 if (ao && seq < tp->rcv_nxt) { in tcp_rcv_sne_update()
3709 ao->rcv_sne++; in tcp_rcv_sne_update()
3710 trace_tcp_ao_rcv_sne_update((struct sock *)tp, ao->rcv_sne); in tcp_rcv_sne_update()
3715 /* If we update tp->rcv_nxt, also update tp->bytes_received */
3718 u32 delta = seq - tp->rcv_nxt; in tcp_rcv_nxt_update()
3721 tp->bytes_received += delta; in tcp_rcv_nxt_update()
3723 WRITE_ONCE(tp->rcv_nxt, seq); in tcp_rcv_nxt_update()
3728 * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
3736 u32 nwin = ntohs(tcp_hdr(skb)->window); in tcp_ack_update_window()
3738 if (likely(!tcp_hdr(skb)->syn)) in tcp_ack_update_window()
3739 nwin <<= tp->rx_opt.snd_wscale; in tcp_ack_update_window()
3745 if (tp->snd_wnd != nwin) { in tcp_ack_update_window()
3746 tp->snd_wnd = nwin; in tcp_ack_update_window()
3751 tp->pred_flags = 0; in tcp_ack_update_window()
3757 if (nwin > tp->max_window) { in tcp_ack_update_window()
3758 tp->max_window = nwin; in tcp_ack_update_window()
3759 tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie); in tcp_ack_update_window()
3776 s32 elapsed = (s32)(tcp_jiffies32 - val); in __tcp_oow_rate_limited()
3779 elapsed < READ_ONCE(net->ipv4.sysctl_tcp_invalid_ratelimit)) { in __tcp_oow_rate_limited()
3781 return true; /* rate-limited: don't send yet! */ in __tcp_oow_rate_limited()
3790 return false; /* not rate-limited: go ahead, send dupack now! */ in __tcp_oow_rate_limited()
3793 /* Return true if we're currently rate-limiting out-of-window ACKs and
3794 * thus shouldn't send a dupack right now. We rate-limit dupacks in
3795 * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
3798 * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
3804 if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) && in tcp_oow_rate_limited()
3805 !tcp_hdr(skb)->syn) in tcp_oow_rate_limited()
3817 flags = tcp_accecn_reflector_flags(tp->syn_ect_rcv); in tcp_send_ack_reflect_ect()
3818 __tcp_send_ack(sk, tp->rcv_nxt, flags); in tcp_send_ack_reflect_ect()
3828 /* First check our per-socket dupack rate limit. */ in tcp_send_challenge_ack()
3831 &tp->last_oow_ack_time)) in tcp_send_challenge_ack()
3834 ack_limit = READ_ONCE(net->ipv4.sysctl_tcp_challenge_ack_limit); in tcp_send_challenge_ack()
3838 /* Then check host-wide RFC 5961 rate limit. */ in tcp_send_challenge_ack()
3840 if (now != READ_ONCE(net->ipv4.tcp_challenge_timestamp)) { in tcp_send_challenge_ack()
3843 WRITE_ONCE(net->ipv4.tcp_challenge_timestamp, now); in tcp_send_challenge_ack()
3844 WRITE_ONCE(net->ipv4.tcp_challenge_count, in tcp_send_challenge_ack()
3845 get_random_u32_inclusive(half, ack_limit + half - 1)); in tcp_send_challenge_ack()
3847 count = READ_ONCE(net->ipv4.tcp_challenge_count); in tcp_send_challenge_ack()
3849 WRITE_ONCE(net->ipv4.tcp_challenge_count, count - 1); in tcp_send_challenge_ack()
3858 tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval; in tcp_store_ts_recent()
3859 tp->rx_opt.ts_recent_stamp = ktime_get_seconds(); in tcp_store_ts_recent()
3872 if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) { in tcp_replace_ts_recent()
3875 * for pure ACK frames. -DaveM in tcp_replace_ts_recent()
3880 if (tcp_paws_check(&tp->rx_opt, 0)) { in tcp_replace_ts_recent()
3881 delta = tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent; in tcp_replace_ts_recent()
3896 if (before(ack, tp->tlp_high_seq)) in tcp_process_tlp_ack()
3899 if (!tp->tlp_retrans) { in tcp_process_tlp_ack()
3901 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3904 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3905 } else if (after(ack, tp->tlp_high_seq)) { in tcp_process_tlp_ack()
3918 tp->tlp_high_seq = 0; in tcp_process_tlp_ack()
3926 if (icsk->icsk_ca_ops->in_ack_event) { in tcp_in_ack_event()
3937 icsk->icsk_ca_ops->in_ack_event(sk, ack_ev_flags); in tcp_in_ack_event()
3949 if (rexmit == REXMIT_NONE || sk->sk_state == TCP_SYN_SENT) in tcp_xmit_recovery()
3955 if (after(tp->snd_nxt, tp->high_seq)) in tcp_xmit_recovery()
3957 tp->frto = 0; in tcp_xmit_recovery()
3970 delivered = tp->delivered - prior_delivered; in tcp_newly_delivered()
3989 u32 prior_snd_una = tp->snd_una; in tcp_ack()
3990 bool is_sack_reneg = tp->is_sack_reneg; in tcp_ack()
3991 u32 ack_seq = TCP_SKB_CB(skb)->seq; in tcp_ack()
3992 u32 ack = TCP_SKB_CB(skb)->ack_seq; in tcp_ack()
3994 int prior_packets = tp->packets_out; in tcp_ack()
3995 u32 delivered = tp->delivered; in tcp_ack()
3996 u32 lost = tp->lost; in tcp_ack()
4007 prefetch(sk->tcp_rtx_queue.rb_node); in tcp_ack()
4016 max_window = min_t(u64, tp->max_window, tp->bytes_acked); in tcp_ack()
4018 if (before(ack, prior_snd_una - max_window)) { in tcp_ack()
4021 return -SKB_DROP_REASON_TCP_TOO_OLD_ACK; in tcp_ack()
4029 if (after(ack, tp->snd_nxt)) in tcp_ack()
4030 return -SKB_DROP_REASON_TCP_ACK_UNSENT_DATA; in tcp_ack()
4034 WRITE_ONCE(icsk->icsk_retransmits, 0); in tcp_ack()
4038 if (tp->tcp_clean_acked) in tcp_ack()
4039 tp->tcp_clean_acked(sk, ack); in tcp_ack()
4043 prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una; in tcp_ack()
4050 flag |= tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq); in tcp_ack()
4064 if (ack_seq != TCP_SKB_CB(skb)->end_seq) in tcp_ack()
4071 if (TCP_SKB_CB(skb)->sacked) in tcp_ack()
4088 * with widely-deployed TCP implementations that do this. in tcp_ack()
4095 if (READ_ONCE(sk->sk_err_soft)) in tcp_ack()
4096 WRITE_ONCE(sk->sk_err_soft, 0); in tcp_ack()
4097 WRITE_ONCE(icsk->icsk_probes_out, 0); in tcp_ack()
4098 tp->rcv_tstamp = tcp_jiffies32; in tcp_ack()
4110 tp->delivered - delivered, in tcp_ack()
4116 if (tp->tlp_high_seq) in tcp_ack()
4125 num_dupack = max_t(u16, 1, skb_shinfo(skb)->gso_segs); in tcp_ack()
4140 lost = tp->lost - lost; /* freshly marked lost */ in tcp_ack()
4150 tp->delivered - delivered, in tcp_ack()
4166 if (tp->tlp_high_seq) in tcp_ack()
4174 if (TCP_SKB_CB(skb)->sacked) { in tcp_ack()
4190 /* Valid only in SYN or SYN-ACK with an even length. */ in tcp_parse_fastopen_option()
4196 memcpy(foc->val, cookie, len); in tcp_parse_fastopen_option()
4198 len = -1; in tcp_parse_fastopen_option()
4199 foc->len = len; in tcp_parse_fastopen_option()
4200 foc->exp = exp_opt; in tcp_parse_fastopen_option()
4210 if (th->syn && !(opsize & 1) && in smc_parse_options()
4213 opt_rx->smc_ok = 1; in smc_parse_options()
4227 int length = (th->doff * 4) - sizeof(struct tcphdr); in tcp_parse_mss_option()
4238 length--; in tcp_parse_mss_option()
4257 ptr += opsize - 2; in tcp_parse_mss_option()
4258 length -= opsize; in tcp_parse_mss_option()
4275 int length = (th->doff * 4) - sizeof(struct tcphdr); in tcp_parse_options()
4278 opt_rx->saw_tstamp = 0; in tcp_parse_options()
4279 opt_rx->accecn = 0; in tcp_parse_options()
4280 opt_rx->saw_unknown = 0; in tcp_parse_options()
4290 length--; in tcp_parse_options()
4302 if (opsize == TCPOLEN_MSS && th->syn && !estab) { in tcp_parse_options()
4305 if (opt_rx->user_mss && in tcp_parse_options()
4306 opt_rx->user_mss < in_mss) in tcp_parse_options()
4307 in_mss = opt_rx->user_mss; in tcp_parse_options()
4308 opt_rx->mss_clamp = in_mss; in tcp_parse_options()
4313 if (opsize == TCPOLEN_WINDOW && th->syn && in tcp_parse_options()
4314 !estab && READ_ONCE(net->ipv4.sysctl_tcp_window_scaling)) { in tcp_parse_options()
4316 opt_rx->wscale_ok = 1; in tcp_parse_options()
4324 opt_rx->snd_wscale = snd_wscale; in tcp_parse_options()
4329 ((estab && opt_rx->tstamp_ok) || in tcp_parse_options()
4330 (!estab && READ_ONCE(net->ipv4.sysctl_tcp_timestamps)))) { in tcp_parse_options()
4331 opt_rx->saw_tstamp = 1; in tcp_parse_options()
4332 opt_rx->rcv_tsval = get_unaligned_be32(ptr); in tcp_parse_options()
4333 opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4); in tcp_parse_options()
4337 if (opsize == TCPOLEN_SACK_PERM && th->syn && in tcp_parse_options()
4338 !estab && READ_ONCE(net->ipv4.sysctl_tcp_sack)) { in tcp_parse_options()
4339 opt_rx->sack_ok = TCP_SACK_SEEN; in tcp_parse_options()
4346 !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) && in tcp_parse_options()
4347 opt_rx->sack_ok) { in tcp_parse_options()
4348 TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th; in tcp_parse_options()
4367 opsize - TCPOLEN_FASTOPEN_BASE, in tcp_parse_options()
4368 ptr, th->syn, foc, false); in tcp_parse_options()
4374 opt_rx->accecn = (ptr - 2) - (__u8 *)th; in tcp_parse_options()
4384 tcp_parse_fastopen_option(opsize - in tcp_parse_options()
4386 ptr + 2, th->syn, foc, true); in tcp_parse_options()
4393 opt_rx->saw_unknown = 1; in tcp_parse_options()
4397 opt_rx->saw_unknown = 1; in tcp_parse_options()
4399 ptr += opsize-2; in tcp_parse_options()
4400 length -= opsize; in tcp_parse_options()
4412 tp->rx_opt.saw_tstamp = 1; in tcp_parse_aligned_timestamp()
4414 tp->rx_opt.rcv_tsval = ntohl(*ptr); in tcp_parse_aligned_timestamp()
4417 tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset; in tcp_parse_aligned_timestamp()
4419 tp->rx_opt.rcv_tsecr = 0; in tcp_parse_aligned_timestamp()
4435 if (th->doff == (sizeof(*th) / 4)) { in tcp_fast_parse_options()
4436 tp->rx_opt.saw_tstamp = 0; in tcp_fast_parse_options()
4437 tp->rx_opt.accecn = 0; in tcp_fast_parse_options()
4439 } else if (tp->rx_opt.tstamp_ok && in tcp_fast_parse_options()
4440 th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) { in tcp_fast_parse_options()
4442 tp->rx_opt.accecn = 0; in tcp_fast_parse_options()
4447 tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL); in tcp_fast_parse_options()
4448 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) in tcp_fast_parse_options()
4449 tp->rx_opt.rcv_tsecr -= tp->tsoffset; in tcp_fast_parse_options()
4461 int length = (th->doff << 2) - sizeof(*th); in tcp_do_parse_auth_options()
4480 length--; in tcp_do_parse_auth_options()
4485 return -EINVAL; in tcp_do_parse_auth_options()
4488 return -EINVAL; in tcp_do_parse_auth_options()
4490 return -EEXIST; in tcp_do_parse_auth_options()
4494 return -EINVAL; in tcp_do_parse_auth_options()
4496 return -EEXIST; in tcp_do_parse_auth_options()
4500 ptr += opsize - 2; in tcp_do_parse_auth_options()
4501 length -= opsize; in tcp_do_parse_auth_options()
4508 /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
4539 if (tcp_sk(sk)->tcp_usec_ts) in tcp_tsval_replay()
4540 return inet_csk(sk)->icsk_rto * (USEC_PER_SEC / HZ); in tcp_tsval_replay()
4545 return inet_csk(sk)->icsk_rto * 1200 / HZ; in tcp_tsval_replay()
4554 u32 ack = TCP_SKB_CB(skb)->ack_seq; in tcp_disordered_ack_check()
4555 u32 seq = TCP_SKB_CB(skb)->seq; in tcp_disordered_ack_check()
4558 if (!th->ack || seq != TCP_SKB_CB(skb)->end_seq) in tcp_disordered_ack_check()
4562 if (seq != tp->rcv_nxt) in tcp_disordered_ack_check()
4563 return before(seq, tp->rcv_nxt) ? in tcp_disordered_ack_check()
4568 if (ack != tp->snd_una) in tcp_disordered_ack_check()
4572 if (tcp_may_update_window(tp, ack, seq, ntohs(th->window) << in tcp_disordered_ack_check()
4573 tp->rx_opt.snd_wscale)) in tcp_disordered_ack_check()
4577 if ((s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) > in tcp_disordered_ack_check()
4602 if (before(end_seq, tp->rcv_wup)) in tcp_sequence()
4605 if (after(end_seq, tp->rcv_nxt + tcp_receive_window(tp))) { in tcp_sequence()
4606 if (after(seq, tp->rcv_nxt + tcp_receive_window(tp))) in tcp_sequence()
4610 if (skb_queue_len(&sk->sk_receive_queue)) in tcp_sequence()
4621 WRITE_ONCE(sk->sk_err, err); in tcp_done_with_error()
4646 switch (sk->sk_state) { in tcp_reset()
4666 * If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT
4667 * (and thence onto LAST-ACK and finally, CLOSE, we never enter
4668 * TIME-WAIT)
4670 * If we are in FINWAIT-1, a received FIN indicates simultaneous
4671 * close and we go into CLOSING (and later onto TIME-WAIT)
4673 * If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
4681 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN); in tcp_fin()
4684 switch (sk->sk_state) { in tcp_fin()
4699 /* RFC793: Remain in the LAST-ACK state. */ in tcp_fin()
4711 /* Received a FIN -- send ACK and enter TIME_WAIT. */ in tcp_fin()
4719 pr_err("%s: Impossible, sk->sk_state=%d\n", in tcp_fin()
4720 __func__, sk->sk_state); in tcp_fin()
4724 /* It _is_ possible, that we have something out-of-order _after_ FIN. in tcp_fin()
4727 skb_rbtree_purge(&tp->out_of_order_queue); in tcp_fin()
4729 tcp_sack_reset(&tp->rx_opt); in tcp_fin()
4732 sk->sk_state_change(sk); in tcp_fin()
4735 if (sk->sk_shutdown == SHUTDOWN_MASK || in tcp_fin()
4736 sk->sk_state == TCP_CLOSE) in tcp_fin()
4746 if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) { in tcp_sack_extend()
4747 if (before(seq, sp->start_seq)) in tcp_sack_extend()
4748 sp->start_seq = seq; in tcp_sack_extend()
4749 if (after(end_seq, sp->end_seq)) in tcp_sack_extend()
4750 sp->end_seq = end_seq; in tcp_sack_extend()
4760 if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) { in tcp_dsack_set()
4763 if (before(seq, tp->rcv_nxt)) in tcp_dsack_set()
4770 tp->rx_opt.dsack = 1; in tcp_dsack_set()
4771 tp->duplicate_sack[0].start_seq = seq; in tcp_dsack_set()
4772 tp->duplicate_sack[0].end_seq = end_seq; in tcp_dsack_set()
4780 if (!tp->rx_opt.dsack) in tcp_dsack_extend()
4783 tcp_sack_extend(tp->duplicate_sack, seq, end_seq); in tcp_dsack_extend()
4797 if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss && in tcp_rcv_spurious_retrans()
4798 skb->protocol == htons(ETH_P_IPV6) && in tcp_rcv_spurious_retrans()
4799 (tcp_sk(sk)->inet_conn.icsk_ack.lrcv_flowlabel != in tcp_rcv_spurious_retrans()
4813 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_send_dupack()
4814 before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_send_dupack()
4818 if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) { in tcp_send_dupack()
4819 u32 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_send_dupack()
4822 if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) in tcp_send_dupack()
4823 end_seq = tp->rcv_nxt; in tcp_send_dupack()
4824 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq); in tcp_send_dupack()
4831 /* These routines update the SACK block as out-of-order packets arrive or
4832 * in-order packets close up the sequence space.
4837 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_maybe_coalesce()
4843 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) { in tcp_sack_maybe_coalesce()
4844 if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) { in tcp_sack_maybe_coalesce()
4850 tp->rx_opt.num_sacks--; in tcp_sack_maybe_coalesce()
4851 for (i = this_sack; i < tp->rx_opt.num_sacks; i++) in tcp_sack_maybe_coalesce()
4864 if (!tp->compressed_ack) in tcp_sack_compress_send_ack()
4867 if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1) in tcp_sack_compress_send_ack()
4871 * substract one from tp->compressed_ack to keep in tcp_sack_compress_send_ack()
4875 tp->compressed_ack - 1); in tcp_sack_compress_send_ack()
4877 tp->compressed_ack = 0; in tcp_sack_compress_send_ack()
4890 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_new_ofo_skb()
4891 int cur_sacks = tp->rx_opt.num_sacks; in tcp_sack_new_ofo_skb()
4902 for (; this_sack > 0; this_sack--, sp--) in tcp_sack_new_ofo_skb()
4903 swap(*sp, *(sp - 1)); in tcp_sack_new_ofo_skb()
4920 this_sack--; in tcp_sack_new_ofo_skb()
4921 tp->rx_opt.num_sacks--; in tcp_sack_new_ofo_skb()
4922 sp--; in tcp_sack_new_ofo_skb()
4924 for (; this_sack > 0; this_sack--, sp--) in tcp_sack_new_ofo_skb()
4925 *sp = *(sp - 1); in tcp_sack_new_ofo_skb()
4929 sp->start_seq = seq; in tcp_sack_new_ofo_skb()
4930 sp->end_seq = end_seq; in tcp_sack_new_ofo_skb()
4931 tp->rx_opt.num_sacks++; in tcp_sack_new_ofo_skb()
4938 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_sack_remove()
4939 int num_sacks = tp->rx_opt.num_sacks; in tcp_sack_remove()
4943 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_sack_remove()
4944 tp->rx_opt.num_sacks = 0; in tcp_sack_remove()
4950 if (!before(tp->rcv_nxt, sp->start_seq)) { in tcp_sack_remove()
4954 WARN_ON(before(tp->rcv_nxt, sp->end_seq)); in tcp_sack_remove()
4958 tp->selective_acks[i-1] = tp->selective_acks[i]; in tcp_sack_remove()
4959 num_sacks--; in tcp_sack_remove()
4965 tp->rx_opt.num_sacks = num_sacks; in tcp_sack_remove()
4969 * tcp_try_coalesce - try to merge skb to prior one
4991 if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq) in tcp_try_coalesce()
5000 atomic_add(delta, &sk->sk_rmem_alloc); in tcp_try_coalesce()
5003 TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq; in tcp_try_coalesce()
5004 TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq; in tcp_try_coalesce()
5005 TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags; in tcp_try_coalesce()
5007 if (TCP_SKB_CB(from)->has_rxtstamp) { in tcp_try_coalesce()
5008 TCP_SKB_CB(to)->has_rxtstamp = true; in tcp_try_coalesce()
5009 to->tstamp = from->tstamp; in tcp_try_coalesce()
5010 skb_hwtstamps(to)->hwtstamp = skb_hwtstamps(from)->hwtstamp; in tcp_try_coalesce()
5023 /* In case tcp_drop_reason() is called later, update to->gso_segs */ in tcp_ooo_try_coalesce()
5025 u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) + in tcp_ooo_try_coalesce()
5026 max_t(u16, 1, skb_shinfo(from)->gso_segs); in tcp_ooo_try_coalesce()
5028 skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF); in tcp_ooo_try_coalesce()
5046 __u32 dsack_high = tp->rcv_nxt; in tcp_ofo_queue()
5051 p = rb_first(&tp->out_of_order_queue); in tcp_ofo_queue()
5054 if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) in tcp_ofo_queue()
5057 if (before(TCP_SKB_CB(skb)->seq, dsack_high)) { in tcp_ofo_queue()
5060 if (before(TCP_SKB_CB(skb)->end_seq, dsack_high)) in tcp_ofo_queue()
5061 dsack = TCP_SKB_CB(skb)->end_seq; in tcp_ofo_queue()
5062 tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack); in tcp_ofo_queue()
5065 rb_erase(&skb->rbnode, &tp->out_of_order_queue); in tcp_ofo_queue()
5067 if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) { in tcp_ofo_queue()
5072 tail = skb_peek_tail(&sk->sk_receive_queue); in tcp_ofo_queue()
5074 tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq); in tcp_ofo_queue()
5075 fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN; in tcp_ofo_queue()
5083 /* tcp_fin() purges tp->out_of_order_queue, in tcp_ofo_queue()
5095 * while satisfying sk->sk_rcvbuf limit.
5097 * In theory we should use skb->truesize, but this can cause problems
5099 * When LRO / hw gro is used, the socket might have a high tp->scaling_ratio,
5102 * filling RWIN, but with a high skb->truesize, because most NIC use 4K page
5105 * Note that we use skb->len to decide to accept or drop this packet,
5106 * but sk->sk_rmem_alloc is the sum of all skb->truesize.
5110 unsigned int rmem = atomic_read(&sk->sk_rmem_alloc); in tcp_can_ingest()
5112 return rmem + skb->len <= sk->sk_rcvbuf; in tcp_can_ingest()
5122 return -1; in tcp_try_rmem_schedule()
5126 return -1; in tcp_try_rmem_schedule()
5143 if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) { in tcp_data_queue_ofo()
5145 sk->sk_data_ready(sk); in tcp_data_queue_ofo()
5152 tp->pred_flags = 0; in tcp_data_queue_ofo()
5155 tp->rcv_ooopack += max_t(u16, 1, skb_shinfo(skb)->gso_segs); in tcp_data_queue_ofo()
5157 seq = TCP_SKB_CB(skb)->seq; in tcp_data_queue_ofo()
5158 end_seq = TCP_SKB_CB(skb)->end_seq; in tcp_data_queue_ofo()
5160 p = &tp->out_of_order_queue.rb_node; in tcp_data_queue_ofo()
5161 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_data_queue_ofo()
5164 tp->rx_opt.num_sacks = 1; in tcp_data_queue_ofo()
5165 tp->selective_acks[0].start_seq = seq; in tcp_data_queue_ofo()
5166 tp->selective_acks[0].end_seq = end_seq; in tcp_data_queue_ofo()
5168 rb_link_node(&skb->rbnode, NULL, p); in tcp_data_queue_ofo()
5169 rb_insert_color(&skb->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5170 tp->ooo_last_skb = skb; in tcp_data_queue_ofo()
5177 if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb, in tcp_data_queue_ofo()
5180 /* For non sack flows, do not grow window to force DUPACK in tcp_data_queue_ofo()
5190 if (!before(seq, TCP_SKB_CB(tp->ooo_last_skb)->end_seq)) { in tcp_data_queue_ofo()
5191 parent = &tp->ooo_last_skb->rbnode; in tcp_data_queue_ofo()
5192 p = &parent->rb_right; in tcp_data_queue_ofo()
5201 if (before(seq, TCP_SKB_CB(skb1)->seq)) { in tcp_data_queue_ofo()
5202 p = &parent->rb_left; in tcp_data_queue_ofo()
5205 if (before(seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5206 if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5216 if (after(seq, TCP_SKB_CB(skb1)->seq)) { in tcp_data_queue_ofo()
5218 tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5223 rb_replace_node(&skb1->rbnode, &skb->rbnode, in tcp_data_queue_ofo()
5224 &tp->out_of_order_queue); in tcp_data_queue_ofo()
5226 TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5227 TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5238 p = &parent->rb_right; in tcp_data_queue_ofo()
5242 rb_link_node(&skb->rbnode, parent, p); in tcp_data_queue_ofo()
5243 rb_insert_color(&skb->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5248 if (!after(end_seq, TCP_SKB_CB(skb1)->seq)) in tcp_data_queue_ofo()
5250 if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) { in tcp_data_queue_ofo()
5251 tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5255 rb_erase(&skb1->rbnode, &tp->out_of_order_queue); in tcp_data_queue_ofo()
5256 tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq, in tcp_data_queue_ofo()
5257 TCP_SKB_CB(skb1)->end_seq); in tcp_data_queue_ofo()
5263 tp->ooo_last_skb = skb; in tcp_data_queue_ofo()
5270 /* For non sack flows, do not grow window to force DUPACK in tcp_data_queue_ofo()
5278 /* do not grow rcvbuf for not-yet-accepted or orphaned sockets. */ in tcp_data_queue_ofo()
5279 if (sk->sk_socket) in tcp_data_queue_ofo()
5280 tcp_rcvbuf_grow(sk, tp->rcvq_space.space); in tcp_data_queue_ofo()
5287 struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue); in tcp_queue_rcv()
5292 tcp_rcv_nxt_update(tcp_sk(sk), TCP_SKB_CB(skb)->end_seq); in tcp_queue_rcv()
5303 int err = -ENOMEM; in tcp_send_rcvq()
5316 skb = alloc_skb_with_frags(size - data_len, data_len, in tcp_send_rcvq()
5318 &err, sk->sk_allocation); in tcp_send_rcvq()
5322 skb_put(skb, size - data_len); in tcp_send_rcvq()
5323 skb->data_len = data_len; in tcp_send_rcvq()
5324 skb->len = size; in tcp_send_rcvq()
5326 if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) { in tcp_send_rcvq()
5331 err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size); in tcp_send_rcvq()
5335 TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt; in tcp_send_rcvq()
5336 TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size; in tcp_send_rcvq()
5337 TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1; in tcp_send_rcvq()
5354 if (tcp_epollin_ready(sk, sk->sk_rcvlowat) || sock_flag(sk, SOCK_DONE)) in tcp_data_ready()
5355 sk->sk_data_ready(sk); in tcp_data_ready()
5373 if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) { in tcp_data_queue()
5378 __skb_pull(skb, tcp_hdr(skb)->doff * 4); in tcp_data_queue()
5381 tp->rx_opt.dsack = 0; in tcp_data_queue()
5387 if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) { in tcp_data_queue()
5395 if (!skb->len && in tcp_data_queue()
5396 (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)) in tcp_data_queue()
5406 if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) { in tcp_data_queue()
5408 inet_csk(sk)->icsk_ack.pending |= in tcp_data_queue()
5411 sk->sk_data_ready(sk); in tcp_data_queue()
5413 if (skb_queue_len(&sk->sk_receive_queue) && skb->len) { in tcp_data_queue()
5418 sk_forced_mem_schedule(sk, skb->truesize); in tcp_data_queue()
5422 if (skb->len) in tcp_data_queue()
5424 if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN) in tcp_data_queue()
5427 if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in tcp_data_queue()
5433 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) in tcp_data_queue()
5434 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW; in tcp_data_queue()
5437 if (tp->rx_opt.num_sacks) in tcp_data_queue()
5449 if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) { in tcp_data_queue()
5454 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); in tcp_data_queue()
5465 if (!before(TCP_SKB_CB(skb)->seq, in tcp_data_queue()
5466 tp->rcv_nxt + tcp_receive_window(tp))) { in tcp_data_queue()
5471 if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_data_queue()
5473 tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt); in tcp_data_queue()
5476 * remembering D-SACK for its head made in previous line. in tcp_data_queue()
5492 return !skb_queue_is_last(list, skb) ? skb->next : NULL; in tcp_skb_next()
5506 rb_erase(&skb->rbnode, root); in tcp_collapse_one()
5514 /* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */
5517 struct rb_node **p = &root->rb_node; in tcp_rbtree_insert()
5524 if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq)) in tcp_rbtree_insert()
5525 p = &parent->rb_left; in tcp_rbtree_insert()
5527 p = &parent->rb_right; in tcp_rbtree_insert()
5529 rb_link_node(&skb->rbnode, parent, p); in tcp_rbtree_insert()
5530 rb_insert_color(&skb->rbnode, root); in tcp_rbtree_insert()
5560 if (!before(start, TCP_SKB_CB(skb)->end_seq)) { in tcp_collapse()
5568 * - not SYN/FIN and in tcp_collapse()
5569 * - bloated or contains data before "start" or in tcp_collapse()
5572 if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) && in tcp_collapse()
5573 (tcp_win_from_space(sk, skb->truesize) > skb->len || in tcp_collapse()
5574 before(TCP_SKB_CB(skb)->seq, start))) { in tcp_collapse()
5581 TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) { in tcp_collapse()
5588 start = TCP_SKB_CB(skb)->end_seq; in tcp_collapse()
5591 (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) || in tcp_collapse()
5598 int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start); in tcp_collapse()
5605 memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); in tcp_collapse()
5607 TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start; in tcp_collapse()
5617 int offset = start - TCP_SKB_CB(skb)->seq; in tcp_collapse()
5618 int size = TCP_SKB_CB(skb)->end_seq - start; in tcp_collapse()
5625 TCP_SKB_CB(nskb)->end_seq += size; in tcp_collapse()
5626 copy -= size; in tcp_collapse()
5629 if (!before(start, TCP_SKB_CB(skb)->end_seq)) { in tcp_collapse()
5634 (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) || in tcp_collapse()
5655 skb = skb_rb_first(&tp->out_of_order_queue); in tcp_collapse_ofo_queue()
5658 tp->ooo_last_skb = skb_rb_last(&tp->out_of_order_queue); in tcp_collapse_ofo_queue()
5661 start = TCP_SKB_CB(skb)->seq; in tcp_collapse_ofo_queue()
5662 end = TCP_SKB_CB(skb)->end_seq; in tcp_collapse_ofo_queue()
5663 range_truesize = skb->truesize; in tcp_collapse_ofo_queue()
5672 after(TCP_SKB_CB(skb)->seq, end) || in tcp_collapse_ofo_queue()
5673 before(TCP_SKB_CB(skb)->end_seq, start)) { in tcp_collapse_ofo_queue()
5675 if (range_truesize != head->truesize || in tcp_collapse_ofo_queue()
5676 end - start >= SKB_WITH_OVERHEAD(PAGE_SIZE)) { in tcp_collapse_ofo_queue()
5677 tcp_collapse(sk, NULL, &tp->out_of_order_queue, in tcp_collapse_ofo_queue()
5681 if (sum_tiny > sk->sk_rcvbuf >> 3) in tcp_collapse_ofo_queue()
5687 range_truesize += skb->truesize; in tcp_collapse_ofo_queue()
5688 if (unlikely(before(TCP_SKB_CB(skb)->seq, start))) in tcp_collapse_ofo_queue()
5689 start = TCP_SKB_CB(skb)->seq; in tcp_collapse_ofo_queue()
5690 if (after(TCP_SKB_CB(skb)->end_seq, end)) in tcp_collapse_ofo_queue()
5691 end = TCP_SKB_CB(skb)->end_seq; in tcp_collapse_ofo_queue()
5696 * Clean the out-of-order queue to make room.
5715 if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) in tcp_prune_ofo_queue()
5718 goal = sk->sk_rcvbuf >> 3; in tcp_prune_ofo_queue()
5719 node = &tp->ooo_last_skb->rbnode; in tcp_prune_ofo_queue()
5725 if (after(TCP_SKB_CB(in_skb)->seq, TCP_SKB_CB(skb)->seq)) in tcp_prune_ofo_queue()
5729 rb_erase(node, &tp->out_of_order_queue); in tcp_prune_ofo_queue()
5730 goal -= skb->truesize; in tcp_prune_ofo_queue()
5732 tp->ooo_last_skb = rb_to_skb(prev); in tcp_prune_ofo_queue()
5737 goal = sk->sk_rcvbuf >> 3; in tcp_prune_ofo_queue()
5749 if (tp->rx_opt.sack_ok) in tcp_prune_ofo_queue()
5750 tcp_sack_reset(&tp->rx_opt); in tcp_prune_ofo_queue()
5767 if (!atomic_read(&sk->sk_rmem_alloc)) in tcp_prune_queue()
5768 return -1; in tcp_prune_queue()
5781 if (!skb_queue_empty(&sk->sk_receive_queue)) in tcp_prune_queue()
5782 tcp_collapse(sk, &sk->sk_receive_queue, NULL, in tcp_prune_queue()
5783 skb_peek(&sk->sk_receive_queue), in tcp_prune_queue()
5785 tp->copied_seq, tp->rcv_nxt); in tcp_prune_queue()
5805 tp->pred_flags = 0; in tcp_prune_queue()
5806 return -1; in tcp_prune_queue()
5816 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) in tcp_should_expand_sndbuf()
5828 WRITE_ONCE(sk->sk_sndbuf, unused_mem); in tcp_should_expand_sndbuf()
5850 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_new_space()
5853 INDIRECT_CALL_1(sk->sk_write_space, sk_stream_write_space, sk); in tcp_new_space()
5857 * 1) Freeing skbs in rtx queues (after tp->snd_una has advanced)
5858 * 2) Sent skbs from output queue (and thus advancing tp->snd_nxt)
5861 * 1) Space consumed in output/rtx queues is below sk->sk_sndbuf/2
5862 * 2) notsent amount (tp->write_seq - tp->snd_nxt) became
5870 if (sk->sk_socket && in tcp_check_space()
5871 test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { in tcp_check_space()
5873 if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) in tcp_check_space()
5893 if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss && in __tcp_ack_snd_check()
5899 (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || in __tcp_ack_snd_check()
5900 __tcp_select_window(sk) >= tp->rcv_wnd)) || in __tcp_ack_snd_check()
5903 /* Protocol state mandates a one-time immediate ACK */ in __tcp_ack_snd_check()
5904 inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOW) { in __tcp_ack_snd_check()
5909 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_backlog_ack_defer)) { in __tcp_ack_snd_check()
5910 set_bit(TCP_ACK_DEFERRED, &sk->sk_tsq_flags); in __tcp_ack_snd_check()
5918 if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) { in __tcp_ack_snd_check()
5924 tp->compressed_ack >= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr)) in __tcp_ack_snd_check()
5927 if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) { in __tcp_ack_snd_check()
5928 tp->compressed_ack_rcv_nxt = tp->rcv_nxt; in __tcp_ack_snd_check()
5929 tp->dup_ack_counter = 0; in __tcp_ack_snd_check()
5931 if (tp->dup_ack_counter < TCP_FASTRETRANS_THRESH) { in __tcp_ack_snd_check()
5932 tp->dup_ack_counter++; in __tcp_ack_snd_check()
5935 tp->compressed_ack++; in __tcp_ack_snd_check()
5936 if (hrtimer_is_queued(&tp->compressed_ack_timer)) in __tcp_ack_snd_check()
5941 rtt = tp->rcv_rtt_est.rtt_us; in __tcp_ack_snd_check()
5942 if (tp->srtt_us && tp->srtt_us < rtt) in __tcp_ack_snd_check()
5943 rtt = tp->srtt_us; in __tcp_ack_snd_check()
5946 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns), in __tcp_ack_snd_check()
5949 hrtimer_start_range_ns(&tp->compressed_ack_timer, ns_to_ktime(delay), in __tcp_ack_snd_check()
5950 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_slack_ns), in __tcp_ack_snd_check()
5964 * This routine is only called when we have urgent data
5967 * place. We handle URGent data wrong. We have to - as
5976 u32 ptr = ntohs(th->urg_ptr); in tcp_check_urg()
5978 if (ptr && !READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_stdurg)) in tcp_check_urg()
5979 ptr--; in tcp_check_urg()
5980 ptr += ntohl(th->seq); in tcp_check_urg()
5982 /* Ignore urgent data that we've already seen and read. */ in tcp_check_urg()
5983 if (after(tp->copied_seq, ptr)) in tcp_check_urg()
5996 if (before(ptr, tp->rcv_nxt)) in tcp_check_urg()
5999 /* Do we already have a newer (or duplicate) urgent pointer? */ in tcp_check_urg()
6000 if (tp->urg_data && !after(ptr, tp->urg_seq)) in tcp_check_urg()
6003 /* Tell the world about our new urgent pointer. */ in tcp_check_urg()
6006 /* We may be adding urgent data when the last byte read was in tcp_check_urg()
6007 * urgent. To do this requires some care. We cannot just ignore in tcp_check_urg()
6008 * tp->copied_seq since we would read the last urgent byte again in tcp_check_urg()
6021 if (tp->urg_seq == tp->copied_seq && tp->urg_data && in tcp_check_urg()
6022 !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) { in tcp_check_urg()
6023 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue); in tcp_check_urg()
6024 tp->copied_seq++; in tcp_check_urg()
6025 if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) { in tcp_check_urg()
6026 __skb_unlink(skb, &sk->sk_receive_queue); in tcp_check_urg()
6031 WRITE_ONCE(tp->urg_data, TCP_URG_NOTYET); in tcp_check_urg()
6032 WRITE_ONCE(tp->urg_seq, ptr); in tcp_check_urg()
6035 tp->pred_flags = 0; in tcp_check_urg()
6038 /* This is the 'fast' part of urgent handling. */
6043 /* Check if we get a new urgent pointer - normally not. */ in tcp_urg()
6044 if (unlikely(th->urg)) in tcp_urg()
6047 /* Do we wait for any urgent data? - normally not... */ in tcp_urg()
6048 if (unlikely(tp->urg_data == TCP_URG_NOTYET)) { in tcp_urg()
6049 u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) - in tcp_urg()
6050 th->syn; in tcp_urg()
6052 /* Is the urgent pointer pointing into this packet? */ in tcp_urg()
6053 if (ptr < skb->len) { in tcp_urg()
6057 WRITE_ONCE(tp->urg_data, TCP_URG_VALID | tmp); in tcp_urg()
6059 sk->sk_data_ready(sk); in tcp_urg()
6064 /* Accept RST for rcv_nxt - 1 after a FIN.
6076 return unlikely(TCP_SKB_CB(skb)->seq == (tp->rcv_nxt - 1) && in tcp_reset_check()
6077 (1 << sk->sk_state) & (TCPF_CLOSE_WAIT | TCPF_LAST_ACK | in tcp_reset_check()
6093 !tp->rx_opt.saw_tstamp || in tcp_validate_incoming()
6094 tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW)) in tcp_validate_incoming()
6101 if (th->rst) in tcp_validate_incoming()
6103 if (unlikely(th->syn)) in tcp_validate_incoming()
6116 &tp->last_oow_ack_time)) in tcp_validate_incoming()
6122 reason = tcp_sequence(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq); in tcp_validate_incoming()
6124 /* RFC793, page 37: "In all states except SYN-SENT, all reset in tcp_validate_incoming()
6125 * (RST) segments are validated by checking their SEQ-fields." in tcp_validate_incoming()
6130 if (!th->rst) { in tcp_validate_incoming()
6131 if (th->syn) in tcp_validate_incoming()
6140 &tp->last_oow_ack_time)) in tcp_validate_incoming()
6149 if (th->rst) { in tcp_validate_incoming()
6150 /* RFC 5961 3.2 (extend to match against (RCV.NXT - 1) after a in tcp_validate_incoming()
6152 * If seq num matches RCV.NXT or (RCV.NXT - 1) after a FIN, or in tcp_validate_incoming()
6153 * the right-most SACK block, in tcp_validate_incoming()
6159 if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt || in tcp_validate_incoming()
6163 if (tcp_is_sack(tp) && tp->rx_opt.num_sacks > 0) { in tcp_validate_incoming()
6164 struct tcp_sack_block *sp = &tp->selective_acks[0]; in tcp_validate_incoming()
6168 for (this_sack = 1; this_sack < tp->rx_opt.num_sacks; in tcp_validate_incoming()
6175 if (TCP_SKB_CB(skb)->seq == max_sack) in tcp_validate_incoming()
6179 /* Disable TFO if RST is out-of-order in tcp_validate_incoming()
6183 if (tp->syn_fastopen && !tp->data_segs_in && in tcp_validate_incoming()
6184 sk->sk_state == TCP_ESTABLISHED) in tcp_validate_incoming()
6196 if (th->syn) { in tcp_validate_incoming()
6199 if (tp->rx_opt.accecn && in tcp_validate_incoming()
6200 tp->saw_accecn_opt < TCP_ACCECN_OPT_COUNTER_SEEN) { in tcp_validate_incoming()
6201 u8 saw_opt = tcp_accecn_option_init(skb, tp->rx_opt.accecn); in tcp_validate_incoming()
6207 if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack && in tcp_validate_incoming()
6208 TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq && in tcp_validate_incoming()
6209 TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt && in tcp_validate_incoming()
6210 TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt) in tcp_validate_incoming()
6241 * - A zero window was announced from us - zero window probing
6243 * - Out of order segments arrived.
6244 * - Urgent data is expected.
6245 * - There is no buffer space left
6246 * - Unexpected TCP flags/window values/header lengths are received
6248 * - Data is sent in both directions. Fast path only supports pure senders
6251 * - Unexpected TCP option.
6262 const struct tcphdr *th = (const struct tcphdr *)skb->data; in tcp_rcv_established()
6264 unsigned int len = skb->len; in tcp_rcv_established()
6270 if (unlikely(!rcu_access_pointer(sk->sk_rx_dst))) in tcp_rcv_established()
6271 inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb); in tcp_rcv_established()
6287 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_established()
6288 tp->rx_opt.accecn = 0; in tcp_rcv_established()
6292 * 'S' will always be tp->tcp_header_len >> 2 in tcp_rcv_established()
6299 if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags && in tcp_rcv_established()
6300 TCP_SKB_CB(skb)->seq == tp->rcv_nxt && in tcp_rcv_established()
6301 !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { in tcp_rcv_established()
6302 int tcp_header_len = tp->tcp_header_len; in tcp_rcv_established()
6307 * is automatically equal to th->doff*4 due to pred_flags in tcp_rcv_established()
6317 delta = tp->rx_opt.rcv_tsval - in tcp_rcv_established()
6318 tp->rx_opt.ts_recent; in tcp_rcv_established()
6339 tp->rcv_nxt == tp->rcv_wup) in tcp_rcv_established()
6355 tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr; in tcp_rcv_established()
6369 if (after(TCP_SKB_CB(skb)->end_seq, in tcp_rcv_established()
6370 tp->rcv_nxt + tcp_receive_window(tp))) in tcp_rcv_established()
6373 if ((int)skb->truesize > sk->sk_forward_alloc) in tcp_rcv_established()
6382 tp->rcv_nxt == tp->rcv_wup) in tcp_rcv_established()
6394 len - tcp_header_len); in tcp_rcv_established()
6399 if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) { in tcp_rcv_established()
6406 tcp_update_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_established()
6419 if (len < (th->doff << 2) || tcp_checksum_complete(skb)) in tcp_rcv_established()
6422 if (!th->ack && !th->rst && !th->syn) { in tcp_rcv_established()
6439 reason = -reason; in tcp_rcv_established()
6444 /* Process urgent data. */ in tcp_rcv_established()
6471 icsk->icsk_af_ops->rebuild_header(sk); in tcp_init_transfer()
6475 * Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been in tcp_init_transfer()
6477 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK in tcp_init_transfer()
6480 if (tp->total_retrans > 1 && tp->undo_marker) in tcp_init_transfer()
6484 tp->snd_cwnd_stamp = tcp_jiffies32; in tcp_init_transfer()
6488 if (!icsk->icsk_ca_initialized) in tcp_init_transfer()
6500 icsk->icsk_ack.lrcvtime = tcp_jiffies32; in tcp_finish_connect()
6503 icsk->icsk_af_ops->sk_rx_dst_set(sk, skb); in tcp_finish_connect()
6513 tp->lsndtime = tcp_jiffies32; in tcp_finish_connect()
6518 if (!tp->rx_opt.snd_wscale) in tcp_finish_connect()
6519 __tcp_fast_path_on(tp, tp->snd_wnd); in tcp_finish_connect()
6521 tp->pred_flags = 0; in tcp_finish_connect()
6528 struct sk_buff *data = tp->syn_data ? tcp_rtx_queue_head(sk) : NULL; in tcp_rcv_fastopen_synack()
6529 u16 mss = tp->rx_opt.mss_clamp, try_exp = 0; in tcp_rcv_fastopen_synack()
6532 if (mss == READ_ONCE(tp->rx_opt.user_mss)) { in tcp_rcv_fastopen_synack()
6542 if (!tp->syn_fastopen) { in tcp_rcv_fastopen_synack()
6544 cookie->len = -1; in tcp_rcv_fastopen_synack()
6545 } else if (tp->total_retrans) { in tcp_rcv_fastopen_synack()
6546 /* SYN timed out and the SYN-ACK neither has a cookie nor in tcp_rcv_fastopen_synack()
6549 * SYN-data or the corresponding SYN-ACK was dropped. in tcp_rcv_fastopen_synack()
6551 syn_drop = (cookie->len < 0 && data); in tcp_rcv_fastopen_synack()
6552 } else if (cookie->len < 0 && !tp->syn_data) { in tcp_rcv_fastopen_synack()
6557 try_exp = tp->syn_fastopen_exp ? 2 : 1; in tcp_rcv_fastopen_synack()
6563 if (tp->total_retrans) in tcp_rcv_fastopen_synack()
6564 tp->fastopen_client_fail = TFO_SYN_RETRANSMITTED; in tcp_rcv_fastopen_synack()
6566 tp->fastopen_client_fail = TFO_DATA_NOT_ACKED; in tcp_rcv_fastopen_synack()
6574 tp->syn_data_acked = tp->syn_data; in tcp_rcv_fastopen_synack()
6575 if (tp->syn_data_acked) { in tcp_rcv_fastopen_synack()
6577 /* SYN-data is counted as two separate packets in tcp_ack() */ in tcp_rcv_fastopen_synack()
6578 if (tp->delivered > 1) in tcp_rcv_fastopen_synack()
6579 --tp->delivered; in tcp_rcv_fastopen_synack()
6591 if (tp->syn_smc && !tp->rx_opt.smc_ok) in smc_check_reset_syn()
6592 tp->syn_smc = 0; in smc_check_reset_syn()
6606 syn_stamp = tp->retrans_stamp; in tcp_try_undo_spurious_syn()
6607 if (tp->undo_marker && syn_stamp && tp->rx_opt.saw_tstamp && in tcp_try_undo_spurious_syn()
6608 syn_stamp == tp->rx_opt.rcv_tsecr) in tcp_try_undo_spurious_syn()
6609 tp->undo_marker = 0; in tcp_try_undo_spurious_syn()
6617 struct tcp_fastopen_cookie foc = { .len = -1 }; in tcp_rcv_synsent_state_process()
6618 int saved_clamp = tp->rx_opt.mss_clamp; in tcp_rcv_synsent_state_process()
6622 tcp_parse_options(sock_net(sk), skb, &tp->rx_opt, 0, &foc); in tcp_rcv_synsent_state_process()
6623 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr) in tcp_rcv_synsent_state_process()
6624 tp->rx_opt.rcv_tsecr -= tp->tsoffset; in tcp_rcv_synsent_state_process()
6626 if (th->ack) { in tcp_rcv_synsent_state_process()
6628 * "If the state is SYN-SENT then in tcp_rcv_synsent_state_process()
6635 if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) || in tcp_rcv_synsent_state_process()
6636 after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) { in tcp_rcv_synsent_state_process()
6638 if (icsk->icsk_retransmits == 0) in tcp_rcv_synsent_state_process()
6645 if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr && in tcp_rcv_synsent_state_process()
6646 !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp, in tcp_rcv_synsent_state_process()
6662 if (th->rst) { in tcp_rcv_synsent_state_process()
6674 * --ANK(990513) in tcp_rcv_synsent_state_process()
6676 if (!th->syn) { in tcp_rcv_synsent_state_process()
6689 TCP_SKB_CB(skb)->ip_dsfield); in tcp_rcv_synsent_state_process()
6691 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_synsent_state_process()
6698 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1); in tcp_rcv_synsent_state_process()
6699 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; in tcp_rcv_synsent_state_process()
6704 tp->snd_wnd = ntohs(th->window); in tcp_rcv_synsent_state_process()
6706 if (!tp->rx_opt.wscale_ok) { in tcp_rcv_synsent_state_process()
6707 tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0; in tcp_rcv_synsent_state_process()
6708 WRITE_ONCE(tp->window_clamp, in tcp_rcv_synsent_state_process()
6709 min(tp->window_clamp, 65535U)); in tcp_rcv_synsent_state_process()
6712 if (tp->rx_opt.saw_tstamp) { in tcp_rcv_synsent_state_process()
6713 tp->rx_opt.tstamp_ok = 1; in tcp_rcv_synsent_state_process()
6714 tp->tcp_header_len = in tcp_rcv_synsent_state_process()
6716 tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; in tcp_rcv_synsent_state_process()
6719 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_rcv_synsent_state_process()
6722 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_rcv_synsent_state_process()
6726 * Change state from SYN-SENT only after copied_seq in tcp_rcv_synsent_state_process()
6728 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_synsent_state_process()
6736 fastopen_fail = (tp->syn_fastopen || tp->syn_data) && in tcp_rcv_synsent_state_process()
6740 sk->sk_state_change(sk); in tcp_rcv_synsent_state_process()
6744 return -1; in tcp_rcv_synsent_state_process()
6745 if (sk->sk_write_pending || in tcp_rcv_synsent_state_process()
6746 READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept) || in tcp_rcv_synsent_state_process()
6753 * to stand against the temptation 8) --ANK in tcp_rcv_synsent_state_process()
6762 return -1; in tcp_rcv_synsent_state_process()
6767 if (th->rst) { in tcp_rcv_synsent_state_process()
6778 if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp && in tcp_rcv_synsent_state_process()
6779 tcp_paws_reject(&tp->rx_opt, 0)) { in tcp_rcv_synsent_state_process()
6783 if (th->syn) { in tcp_rcv_synsent_state_process()
6791 ao = rcu_dereference_protected(tp->ao_info, in tcp_rcv_synsent_state_process()
6794 WRITE_ONCE(ao->risn, th->seq); in tcp_rcv_synsent_state_process()
6795 ao->rcv_sne = 0; in tcp_rcv_synsent_state_process()
6800 if (tp->rx_opt.saw_tstamp) { in tcp_rcv_synsent_state_process()
6801 tp->rx_opt.tstamp_ok = 1; in tcp_rcv_synsent_state_process()
6803 tp->tcp_header_len = in tcp_rcv_synsent_state_process()
6806 tp->tcp_header_len = sizeof(struct tcphdr); in tcp_rcv_synsent_state_process()
6809 WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1); in tcp_rcv_synsent_state_process()
6810 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_synsent_state_process()
6811 tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1; in tcp_rcv_synsent_state_process()
6816 tp->snd_wnd = ntohs(th->window); in tcp_rcv_synsent_state_process()
6817 tp->snd_wl1 = TCP_SKB_CB(skb)->seq; in tcp_rcv_synsent_state_process()
6818 tp->max_window = tp->snd_wnd; in tcp_rcv_synsent_state_process()
6823 tcp_sync_mss(sk, icsk->icsk_pmtu_cookie); in tcp_rcv_synsent_state_process()
6839 return -1; in tcp_rcv_synsent_state_process()
6849 tcp_clear_options(&tp->rx_opt); in tcp_rcv_synsent_state_process()
6850 tp->rx_opt.mss_clamp = saved_clamp; in tcp_rcv_synsent_state_process()
6855 tcp_clear_options(&tp->rx_opt); in tcp_rcv_synsent_state_process()
6856 tp->rx_opt.mss_clamp = saved_clamp; in tcp_rcv_synsent_state_process()
6869 if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss && !tp->packets_out) in tcp_rcv_synrecv_state_fastopen()
6873 WRITE_ONCE(inet_csk(sk)->icsk_retransmits, 0); in tcp_rcv_synrecv_state_fastopen()
6887 req = rcu_dereference_protected(tp->fastopen_rsk, in tcp_rcv_synrecv_state_fastopen()
6891 /* Re-arm the timer because data may have been sent out. in tcp_rcv_synrecv_state_fastopen()
6895 * (TFO) - we could try to be more aggressive and in tcp_rcv_synrecv_state_fastopen()
6919 switch (sk->sk_state) { in tcp_rcv_state_process()
6925 if (th->ack) in tcp_rcv_state_process()
6928 if (th->rst) { in tcp_rcv_state_process()
6932 if (th->syn) { in tcp_rcv_state_process()
6933 if (th->fin) { in tcp_rcv_state_process()
6942 icsk->icsk_af_ops->conn_request(sk, skb); in tcp_rcv_state_process()
6953 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_state_process()
6967 tp->rx_opt.saw_tstamp = 0; in tcp_rcv_state_process()
6968 req = rcu_dereference_protected(tp->fastopen_rsk, in tcp_rcv_state_process()
6973 WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV && in tcp_rcv_state_process()
6974 sk->sk_state != TCP_FIN_WAIT1); in tcp_rcv_state_process()
6981 if (!th->ack && !th->rst && !th->syn) { in tcp_rcv_state_process()
6994 if (sk->sk_state == TCP_SYN_RECV) { in tcp_rcv_state_process()
6998 return -reason; in tcp_rcv_state_process()
7003 reason = -reason; in tcp_rcv_state_process()
7008 switch (sk->sk_state) { in tcp_rcv_state_process()
7010 tp->delivered++; /* SYN-ACK delivery isn't tracked in tcp_ack */ in tcp_rcv_state_process()
7011 if (!tp->srtt_us) in tcp_rcv_state_process()
7014 if (tp->rx_opt.tstamp_ok) in tcp_rcv_state_process()
7015 tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; in tcp_rcv_state_process()
7021 tp->retrans_stamp = 0; in tcp_rcv_state_process()
7024 WRITE_ONCE(tp->copied_seq, tp->rcv_nxt); in tcp_rcv_state_process()
7029 sk->sk_state_change(sk); in tcp_rcv_state_process()
7033 * sk->sk_sleep == NULL and sk->sk_socket == NULL. in tcp_rcv_state_process()
7035 if (sk->sk_socket) in tcp_rcv_state_process()
7038 tp->snd_una = TCP_SKB_CB(skb)->ack_seq; in tcp_rcv_state_process()
7039 tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale; in tcp_rcv_state_process()
7040 tcp_init_wl(tp, TCP_SKB_CB(skb)->seq); in tcp_rcv_state_process()
7042 if (!inet_csk(sk)->icsk_ca_ops->cong_control) in tcp_rcv_state_process()
7046 tp->lsndtime = tcp_jiffies32; in tcp_rcv_state_process()
7050 tcp_accecn_third_ack(sk, skb, tp->syn_ect_snt); in tcp_rcv_state_process()
7052 if (sk->sk_shutdown & SEND_SHUTDOWN) in tcp_rcv_state_process()
7063 if (tp->snd_una != tp->write_seq) in tcp_rcv_state_process()
7067 WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | SEND_SHUTDOWN); in tcp_rcv_state_process()
7073 sk->sk_state_change(sk); in tcp_rcv_state_process()
7077 if (READ_ONCE(tp->linger2) < 0) { in tcp_rcv_state_process()
7082 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_rcv_state_process()
7083 after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { in tcp_rcv_state_process()
7085 if (tp->syn_fastopen && th->fin) in tcp_rcv_state_process()
7094 tcp_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN); in tcp_rcv_state_process()
7095 } else if (th->fin || sock_owned_by_user(sk)) { in tcp_rcv_state_process()
7111 if (tp->snd_una == tp->write_seq) { in tcp_rcv_state_process()
7118 if (tp->snd_una == tp->write_seq) { in tcp_rcv_state_process()
7130 switch (sk->sk_state) { in tcp_rcv_state_process()
7134 if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) { in tcp_rcv_state_process()
7149 if (sk->sk_shutdown & RCV_SHUTDOWN) { in tcp_rcv_state_process()
7150 if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq && in tcp_rcv_state_process()
7151 after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) { in tcp_rcv_state_process()
7164 /* tcp_data could move socket to TIME-WAIT */ in tcp_rcv_state_process()
7165 if (sk->sk_state != TCP_CLOSE) { in tcp_rcv_state_process()
7188 &ireq->ir_rmt_addr, port); in pr_drop_req()
7192 &ireq->ir_v6_rmt_addr, port); in pr_drop_req()
7208 * Exception: At least one of the reserved bits of the TCP header (th->res1) is
7220 bool th_ecn = th->ece && th->cwr; in tcp_ecn_create_request()
7225 READ_ONCE(net->ipv4.sysctl_tcp_ecn) >= 3) { in tcp_ecn_create_request()
7226 inet_rsk(req)->ecn_ok = 1; in tcp_ecn_create_request()
7227 tcp_rsk(req)->accecn_ok = 1; in tcp_ecn_create_request()
7228 tcp_rsk(req)->syn_ect_rcv = TCP_SKB_CB(skb)->ip_dsfield & in tcp_ecn_create_request()
7236 ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield); in tcp_ecn_create_request()
7238 ecn_ok = READ_ONCE(net->ipv4.sysctl_tcp_ecn) || ecn_ok_dst; in tcp_ecn_create_request()
7240 if (((!ect || th->res1 || th->ae) && ecn_ok) || in tcp_ecn_create_request()
7244 inet_rsk(req)->ecn_ok = 1; in tcp_ecn_create_request()
7253 req->rsk_rcv_wnd = 0; /* So that tcp_send_synack() knows! */ in tcp_openreq_init()
7254 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; in tcp_openreq_init()
7255 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1; in tcp_openreq_init()
7256 tcp_rsk(req)->snt_synack = 0; in tcp_openreq_init()
7257 tcp_rsk(req)->snt_tsval_first = 0; in tcp_openreq_init()
7258 tcp_rsk(req)->last_oow_ack_time = 0; in tcp_openreq_init()
7259 tcp_rsk(req)->accecn_ok = 0; in tcp_openreq_init()
7260 tcp_rsk(req)->saw_accecn_opt = TCP_ACCECN_OPT_NOT_SEEN; in tcp_openreq_init()
7261 tcp_rsk(req)->accecn_fail_mode = 0; in tcp_openreq_init()
7262 tcp_rsk(req)->syn_ect_rcv = 0; in tcp_openreq_init()
7263 tcp_rsk(req)->syn_ect_snt = 0; in tcp_openreq_init()
7264 req->mss = rx_opt->mss_clamp; in tcp_openreq_init()
7265 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; in tcp_openreq_init()
7266 ireq->tstamp_ok = rx_opt->tstamp_ok; in tcp_openreq_init()
7267 ireq->sack_ok = rx_opt->sack_ok; in tcp_openreq_init()
7268 ireq->snd_wscale = rx_opt->snd_wscale; in tcp_openreq_init()
7269 ireq->wscale_ok = rx_opt->wscale_ok; in tcp_openreq_init()
7270 ireq->acked = 0; in tcp_openreq_init()
7271 ireq->ecn_ok = 0; in tcp_openreq_init()
7272 ireq->ir_rmt_port = tcp_hdr(skb)->source; in tcp_openreq_init()
7273 ireq->ir_num = ntohs(tcp_hdr(skb)->dest); in tcp_openreq_init()
7274 ireq->ir_mark = inet_request_mark(sk, skb); in tcp_openreq_init()
7276 ireq->smc_ok = rx_opt->smc_ok && !(tcp_sk(sk)->smc_hs_congested && in tcp_openreq_init()
7277 tcp_sk(sk)->smc_hs_congested(sk)); in tcp_openreq_init()
7286 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; in tcp_syn_flood_action()
7292 syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies); in tcp_syn_flood_action()
7303 if (syncookies != 2 && !READ_ONCE(queue->synflood_warned)) { in tcp_syn_flood_action()
7304 WRITE_ONCE(queue->synflood_warned, 1); in tcp_syn_flood_action()
7305 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_family == AF_INET6) { in tcp_syn_flood_action()
7308 sk->sk_num, msg); in tcp_syn_flood_action()
7311 proto, &sk->sk_rcv_saddr, in tcp_syn_flood_action()
7312 sk->sk_num, msg); in tcp_syn_flood_action()
7323 if (tcp_sk(sk)->save_syn) { in tcp_reqsk_record_syn()
7329 if (tcp_sk(sk)->save_syn == 2) { /* Save full header. */ in tcp_reqsk_record_syn()
7341 saved_syn->mac_hdrlen = mac_hdrlen; in tcp_reqsk_record_syn()
7342 saved_syn->network_hdrlen = skb_network_header_len(skb); in tcp_reqsk_record_syn()
7343 saved_syn->tcp_hdrlen = tcp_hdrlen(skb); in tcp_reqsk_record_syn()
7344 memcpy(saved_syn->data, base, len); in tcp_reqsk_record_syn()
7345 req->saved_syn = saved_syn; in tcp_reqsk_record_syn()
7360 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) != 2 && in tcp_get_syncookie_mss()
7364 if (!tcp_syn_flood_action(sk, rsk_ops->slab_name)) in tcp_get_syncookie_mss()
7372 mss = tcp_parse_mss_option(th, READ_ONCE(tp->rx_opt.user_mss)); in tcp_get_syncookie_mss()
7374 mss = af_ops->mss_clamp; in tcp_get_syncookie_mss()
7384 struct tcp_fastopen_cookie foc = { .len = -1 }; in tcp_conn_request()
7408 syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies); in tcp_conn_request()
7412 rsk_ops->slab_name); in tcp_conn_request()
7427 req->syncookie = want_cookie; in tcp_conn_request()
7428 tcp_rsk(req)->af_specific = af_ops; in tcp_conn_request()
7429 tcp_rsk(req)->ts_off = 0; in tcp_conn_request()
7430 tcp_rsk(req)->req_usec_ts = false; in tcp_conn_request()
7432 tcp_rsk(req)->is_mptcp = 0; in tcp_conn_request()
7436 tmp_opt.mss_clamp = af_ops->mss_clamp; in tcp_conn_request()
7437 tmp_opt.user_mss = READ_ONCE(tp->rx_opt.user_mss); in tcp_conn_request()
7449 inet_rsk(req)->no_srccheck = inet_test_bit(TRANSPARENT, sk); in tcp_conn_request()
7452 inet_rsk(req)->ir_iif = inet_request_bound_dev_if(sk, skb); in tcp_conn_request()
7454 dst = af_ops->route_req(sk, skb, &fl, req, isn); in tcp_conn_request()
7459 tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst); in tcp_conn_request()
7460 tcp_rsk(req)->ts_off = af_ops->init_ts_off(net, skb); in tcp_conn_request()
7463 int max_syn_backlog = READ_ONCE(net->ipv4.sysctl_max_syn_backlog); in tcp_conn_request()
7467 (max_syn_backlog - inet_csk_reqsk_queue_len(sk) < in tcp_conn_request()
7477 pr_drop_req(req, ntohs(tcp_hdr(skb)->source), in tcp_conn_request()
7478 rsk_ops->family); in tcp_conn_request()
7482 isn = af_ops->init_seq(skb); in tcp_conn_request()
7488 isn = cookie_init_sequence(af_ops, sk, skb, &req->mss); in tcp_conn_request()
7490 inet_rsk(req)->ecn_ok = 0; in tcp_conn_request()
7497 tcp_rsk(req)->used_tcp_ao = true; in tcp_conn_request()
7498 tcp_rsk(req)->ao_rcv_next = aoh->keyid; in tcp_conn_request()
7499 tcp_rsk(req)->ao_keyid = aoh->rnext_keyid; in tcp_conn_request()
7502 tcp_rsk(req)->used_tcp_ao = false; in tcp_conn_request()
7505 tcp_rsk(req)->snt_isn = isn; in tcp_conn_request()
7506 tcp_rsk(req)->txhash = net_tx_rndhash(); in tcp_conn_request()
7507 tcp_rsk(req)->syn_tos = TCP_SKB_CB(skb)->ip_dsfield; in tcp_conn_request()
7515 af_ops->send_synack(fastopen_sk, dst, &fl, req, in tcp_conn_request()
7523 sk->sk_data_ready(sk); in tcp_conn_request()
7527 tcp_rsk(req)->tfo_listener = false; in tcp_conn_request()
7529 req->timeout = tcp_timeout_init((struct sock *)req); in tcp_conn_request()
7531 req->timeout))) { in tcp_conn_request()
7538 af_ops->send_synack(sk, dst, &fl, req, &foc, in tcp_conn_request()