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
98 #define FLAG_DSACKING_ACK	0x800 /* SACK blocks contained D-SACK info */
117 #define REXMIT_NEW	2 /* FRTO-style transmit of unsent/new packets */
125 	tp->tcp_clean_acked = cad;  in clean_acked_data_enable()
133 	tp->tcp_clean_acked = NULL;  in clean_acked_data_disable()
147 	bool unknown_opt = tcp_sk(sk)->rx_opt.saw_unknown &&  in bpf_skops_parse_hdr()
161 	switch (sk->sk_state) {  in bpf_skops_parse_hdr()
215 	dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif);  in tcp_gro_dev_warn()
216 	if (!dev || len >= READ_ONCE(dev->mtu))  in tcp_gro_dev_warn()
218 			dev ? dev->name : "Unknown driver");  in tcp_gro_dev_warn()
228 	const unsigned int lss = icsk->icsk_ack.last_seg_size;  in tcp_measure_rcv_mss()
231 	icsk->icsk_ack.last_seg_size = 0;  in tcp_measure_rcv_mss()
233 	/* skb->len may jitter because of SACKs, even if peer  in tcp_measure_rcv_mss()
234 	 * sends good full-sized frames.  in tcp_measure_rcv_mss()
236 	len = skb_shinfo(skb)->gso_size ? : skb->len;  in tcp_measure_rcv_mss()
237 	if (len >= icsk->icsk_ack.rcv_mss) {  in tcp_measure_rcv_mss()
242 		if (unlikely(len != icsk->icsk_ack.rcv_mss)) {  in tcp_measure_rcv_mss()
243 			u64 val = (u64)skb->len << TCP_RMEM_TO_WIN_SCALE;  in tcp_measure_rcv_mss()
244 			u8 old_ratio = tcp_sk(sk)->scaling_ratio;  in tcp_measure_rcv_mss()
246 			do_div(val, skb->truesize);  in tcp_measure_rcv_mss()
247 			tcp_sk(sk)->scaling_ratio = val ? val : 1;  in tcp_measure_rcv_mss()
249 			if (old_ratio != tcp_sk(sk)->scaling_ratio) {  in tcp_measure_rcv_mss()
252 				val = tcp_win_from_space(sk, sk->sk_rcvbuf);  in tcp_measure_rcv_mss()
255 				if (tp->window_clamp < tp->rcvq_space.space)  in tcp_measure_rcv_mss()
256 					tp->rcvq_space.space = tp->window_clamp;  in tcp_measure_rcv_mss()
259 		icsk->icsk_ack.rcv_mss = min_t(unsigned int, len,  in tcp_measure_rcv_mss()
260 					       tcp_sk(sk)->advmss);  in tcp_measure_rcv_mss()
261 		/* Account for possibly-removed options */  in tcp_measure_rcv_mss()
262 		DO_ONCE_LITE_IF(len > icsk->icsk_ack.rcv_mss + MAX_TCP_OPTION_SPACE,  in tcp_measure_rcv_mss()
267 		 * may be waiting for an ACK if cwnd-bound or using TX zero  in tcp_measure_rcv_mss()
270 		 * reads all of the data and is not ping-pong. If len > MSS  in tcp_measure_rcv_mss()
275 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_PSH)  in tcp_measure_rcv_mss()
276 			icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;  in tcp_measure_rcv_mss()
283 		len += skb->data - skb_transport_header(skb);  in tcp_measure_rcv_mss()
288 		     * to handle super-low mtu links fairly.  in tcp_measure_rcv_mss()
296 			len -= tcp_sk(sk)->tcp_header_len;  in tcp_measure_rcv_mss()
297 			icsk->icsk_ack.last_seg_size = len;  in tcp_measure_rcv_mss()
299 				icsk->icsk_ack.rcv_mss = len;  in tcp_measure_rcv_mss()
303 		if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)  in tcp_measure_rcv_mss()
304 			icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2;  in tcp_measure_rcv_mss()
305 		icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;  in tcp_measure_rcv_mss()
312 	unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss);  in tcp_incr_quickack()
317 	if (quickacks > icsk->icsk_ack.quick)  in tcp_incr_quickack()
318 		icsk->icsk_ack.quick = quickacks;  in tcp_incr_quickack()
327 	icsk->icsk_ack.ato = TCP_ATO_MIN;  in tcp_enter_quickack_mode()
338 	return icsk->icsk_ack.dst_quick_ack ||  in tcp_in_quickack_mode()
339 		(icsk->icsk_ack.quick && !inet_csk_in_pingpong_mode(sk));  in tcp_in_quickack_mode()
345 		tp->ecn_flags |= TCP_ECN_QUEUE_CWR;  in tcp_ecn_queue_cwr()
350 	if (tcp_hdr(skb)->cwr) {  in tcp_ecn_accept_cwr()
351 		tcp_sk(sk)->ecn_flags &= ~TCP_ECN_DEMAND_CWR;  in tcp_ecn_accept_cwr()
357 		if (TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq)  in tcp_ecn_accept_cwr()
358 			inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;  in tcp_ecn_accept_cwr()
364 	tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;  in tcp_ecn_withdraw_cwr()
374 	switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) {  in tcp_data_ecn_check()
380 		if (tp->ecn_flags & TCP_ECN_SEEN)  in tcp_data_ecn_check()
387 		if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) {  in tcp_data_ecn_check()
390 			tp->ecn_flags |= TCP_ECN_DEMAND_CWR;  in tcp_data_ecn_check()
392 		tp->ecn_flags |= TCP_ECN_SEEN;  in tcp_data_ecn_check()
397 		tp->ecn_flags |= TCP_ECN_SEEN;  in tcp_data_ecn_check()
404 	if (tcp_ecn_mode_rfc3168(tp) && (!th->ece || th->cwr))  in tcp_ecn_rcv_synack()
410 	if (tcp_ecn_mode_rfc3168(tp) && (!th->ece || !th->cwr))  in tcp_ecn_rcv_syn()
416 	if (th->ece && !th->syn && tcp_ecn_mode_rfc3168(tp))  in tcp_ecn_rcv_ecn_echo()
423 	tp->delivered_ce += ecn_count;  in tcp_count_delivered_ce()
430 	tp->delivered += delivered;  in tcp_count_delivered()
437  * 1. Tuning sk->sk_sndbuf, when connection enters established state.
443 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;  in tcp_sndbuf_expand()
447 	/* Worst case is non GSO/TSO : each frame consumes one skb  in tcp_sndbuf_expand()
448 	 * and skb->head is kmalloced using power of two area of memory  in tcp_sndbuf_expand()
450 	per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) +  in tcp_sndbuf_expand()
458 	nr_segs = max_t(u32, nr_segs, tp->reordering + 1);  in tcp_sndbuf_expand()
464 	sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2;  in tcp_sndbuf_expand()
467 	if (sk->sk_sndbuf < sndmem)  in tcp_sndbuf_expand()
468 		WRITE_ONCE(sk->sk_sndbuf,  in tcp_sndbuf_expand()
469 			   min(sndmem, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[2])));  in tcp_sndbuf_expand()
475  * forward and advertised in receiver window (tp->rcv_wnd) and
479  * tcp_full_space(), in this case tcp_full_space() - window_clamp
487  * - to enforce header prediction at sender, even when application
489  * - to prevent pruning of receive queue because of misprediction
504 	int window = tcp_win_from_space(sk, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])) >> 1;  in __tcp_grow_window()
506 	while (tp->rcv_ssthresh <= window) {  in __tcp_grow_window()
507 		if (truesize <= skb->len)  in __tcp_grow_window()
508 			return 2 * inet_csk(sk)->icsk_ack.rcv_mss;  in __tcp_grow_window()
517  * can play nice with us, as sk_buff and skb->head might be either
520  * and if no payload was pulled in skb->head before reaching us.
524 	u32 truesize = skb->truesize;  in truesize_adjust()
527 		truesize -= SKB_TRUESIZE(skb_end_offset(skb));  in truesize_adjust()
529 		if (unlikely((int)truesize < (int)skb->len))  in truesize_adjust()
530 			truesize = skb->truesize;  in truesize_adjust()
541 	room = min_t(int, tp->window_clamp, tcp_space(sk)) - tp->rcv_ssthresh;  in tcp_grow_window()
554 		if (tcp_win_from_space(sk, truesize) <= skb->len)  in tcp_grow_window()
555 			incr = 2 * tp->advmss;  in tcp_grow_window()
560 			incr = max_t(int, incr, 2 * skb->len);  in tcp_grow_window()
561 			tp->rcv_ssthresh += min(room, incr);  in tcp_grow_window()
562 			inet_csk(sk)->icsk_ack.quick |= 1;  in tcp_grow_window()
577 	int tcp_app_win = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_app_win);  in tcp_init_buffer_space()
581 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK))  in tcp_init_buffer_space()
585 	tp->rcvq_space.time = tp->tcp_mstamp;  in tcp_init_buffer_space()
586 	tp->rcvq_space.seq = tp->copied_seq;  in tcp_init_buffer_space()
590 	if (tp->window_clamp >= maxwin) {  in tcp_init_buffer_space()
591 		WRITE_ONCE(tp->window_clamp, maxwin);  in tcp_init_buffer_space()
593 		if (tcp_app_win && maxwin > 4 * tp->advmss)  in tcp_init_buffer_space()
594 			WRITE_ONCE(tp->window_clamp,  in tcp_init_buffer_space()
595 				   max(maxwin - (maxwin >> tcp_app_win),  in tcp_init_buffer_space()
596 				       4 * tp->advmss));  in tcp_init_buffer_space()
601 	    tp->window_clamp > 2 * tp->advmss &&  in tcp_init_buffer_space()
602 	    tp->window_clamp + tp->advmss > maxwin)  in tcp_init_buffer_space()
603 		WRITE_ONCE(tp->window_clamp,  in tcp_init_buffer_space()
604 			   max(2 * tp->advmss, maxwin - tp->advmss));  in tcp_init_buffer_space()
606 	tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp);  in tcp_init_buffer_space()
607 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_init_buffer_space()
608 	tp->rcvq_space.space = min3(tp->rcv_ssthresh, tp->rcv_wnd,  in tcp_init_buffer_space()
609 				    (u32)TCP_INIT_CWND * tp->advmss);  in tcp_init_buffer_space()
620 	icsk->icsk_ack.quick = 0;  in tcp_clamp_window()
621 	rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]);  in tcp_clamp_window()
623 	if (sk->sk_rcvbuf < rmem2 &&  in tcp_clamp_window()
624 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) &&  in tcp_clamp_window()
627 		WRITE_ONCE(sk->sk_rcvbuf,  in tcp_clamp_window()
628 			   min(atomic_read(&sk->sk_rmem_alloc), rmem2));  in tcp_clamp_window()
630 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)  in tcp_clamp_window()
631 		tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);  in tcp_clamp_window()
644 	unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);  in tcp_initialize_rcv_mss()
646 	hint = min(hint, tp->rcv_wnd / 2);  in tcp_initialize_rcv_mss()
650 	inet_csk(sk)->icsk_ack.rcv_mss = hint;  in tcp_initialize_rcv_mss()
657  * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL.
667 	u32 new_sample = tp->rcv_rtt_est.rtt_us;  in tcp_rcv_rtt_update()
671 		/* If we sample in larger samples in the non-timestamp  in tcp_rcv_rtt_update()
677 		 * non-timestamp case, we do not smooth things out  in tcp_rcv_rtt_update()
682 			m -= (new_sample >> 3);  in tcp_rcv_rtt_update()
694 	tp->rcv_rtt_est.rtt_us = new_sample;  in tcp_rcv_rtt_update()
701 	if (tp->rcv_rtt_est.time == 0)  in tcp_rcv_rtt_measure()
703 	if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq))  in tcp_rcv_rtt_measure()
705 	delta_us = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcv_rtt_est.time);  in tcp_rcv_rtt_measure()
711 	tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd;  in tcp_rcv_rtt_measure()
712 	tp->rcv_rtt_est.time = tp->tcp_mstamp;  in tcp_rcv_rtt_measure()
719 	delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr;  in tcp_rtt_tsopt_us()
720 	if (tp->tcp_usec_ts)  in tcp_rtt_tsopt_us()
729 	return -1;  in tcp_rtt_tsopt_us()
737 	if (tp->rx_opt.rcv_tsecr == tp->rcv_rtt_last_tsecr)  in tcp_rcv_rtt_measure_ts()
739 	tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;  in tcp_rcv_rtt_measure_ts()
741 	if (TCP_SKB_CB(skb)->end_seq -  in tcp_rcv_rtt_measure_ts()
742 	    TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) {  in tcp_rcv_rtt_measure_ts()
763 	time = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcvq_space.time);  in tcp_rcv_space_adjust()
764 	if (time < (tp->rcv_rtt_est.rtt_us >> 3) || tp->rcv_rtt_est.rtt_us == 0)  in tcp_rcv_space_adjust()
768 	copied = tp->copied_seq - tp->rcvq_space.seq;  in tcp_rcv_space_adjust()
769 	if (copied <= tp->rcvq_space.space)  in tcp_rcv_space_adjust()
781 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&  in tcp_rcv_space_adjust()
782 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {  in tcp_rcv_space_adjust()
789 		rcvwin = ((u64)copied << 1) + 16 * tp->advmss;  in tcp_rcv_space_adjust()
792 		grow = rcvwin * (copied - tp->rcvq_space.space);  in tcp_rcv_space_adjust()
793 		do_div(grow, tp->rcvq_space.space);  in tcp_rcv_space_adjust()
797 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));  in tcp_rcv_space_adjust()
798 		if (rcvbuf > sk->sk_rcvbuf) {  in tcp_rcv_space_adjust()
799 			WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);  in tcp_rcv_space_adjust()
802 			WRITE_ONCE(tp->window_clamp,  in tcp_rcv_space_adjust()
806 	tp->rcvq_space.space = copied;  in tcp_rcv_space_adjust()
809 	tp->rcvq_space.seq = tp->copied_seq;  in tcp_rcv_space_adjust()
810 	tp->rcvq_space.time = tp->tcp_mstamp;  in tcp_rcv_space_adjust()
818 	if (skb->protocol == htons(ETH_P_IPV6))  in tcp_save_lrcv_flowlabel()
819 		icsk->icsk_ack.lrcv_flowlabel = ntohl(ip6_flowlabel(ipv6_hdr(skb)));  in tcp_save_lrcv_flowlabel()
824  * behavior of the tp->ato delayed ack timeout interval.  When a
831  * queue.  -DaveM
847 	if (!icsk->icsk_ack.ato) {  in tcp_event_data_recv()
852 		icsk->icsk_ack.ato = TCP_ATO_MIN;  in tcp_event_data_recv()
854 		int m = now - icsk->icsk_ack.lrcvtime;  in tcp_event_data_recv()
858 			icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2;  in tcp_event_data_recv()
859 		} else if (m < icsk->icsk_ack.ato) {  in tcp_event_data_recv()
860 			icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m;  in tcp_event_data_recv()
861 			if (icsk->icsk_ack.ato > icsk->icsk_rto)  in tcp_event_data_recv()
862 				icsk->icsk_ack.ato = icsk->icsk_rto;  in tcp_event_data_recv()
863 		} else if (m > icsk->icsk_rto) {  in tcp_event_data_recv()
870 	icsk->icsk_ack.lrcvtime = now;  in tcp_event_data_recv()
875 	if (skb->len >= 128)  in tcp_event_data_recv()
886  * it up into three procedures. -- erics
892 	u32 srtt = tp->srtt_us;  in tcp_rtt_estimator()
911 		m -= (srtt >> 3);	/* m is now error in rtt est */  in tcp_rtt_estimator()
914 			m = -m;		/* m is now abs(error) */  in tcp_rtt_estimator()
915 			m -= (tp->mdev_us >> 2);   /* similar update on mdev */  in tcp_rtt_estimator()
927 			m -= (tp->mdev_us >> 2);   /* similar update on mdev */  in tcp_rtt_estimator()
929 		tp->mdev_us += m;		/* mdev = 3/4 mdev + 1/4 new */  in tcp_rtt_estimator()
930 		if (tp->mdev_us > tp->mdev_max_us) {  in tcp_rtt_estimator()
931 			tp->mdev_max_us = tp->mdev_us;  in tcp_rtt_estimator()
932 			if (tp->mdev_max_us > tp->rttvar_us)  in tcp_rtt_estimator()
933 				tp->rttvar_us = tp->mdev_max_us;  in tcp_rtt_estimator()
935 		if (after(tp->snd_una, tp->rtt_seq)) {  in tcp_rtt_estimator()
936 			if (tp->mdev_max_us < tp->rttvar_us)  in tcp_rtt_estimator()
937 				tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2;  in tcp_rtt_estimator()
938 			tp->rtt_seq = tp->snd_nxt;  in tcp_rtt_estimator()
939 			tp->mdev_max_us = tcp_rto_min_us(sk);  in tcp_rtt_estimator()
946 		tp->mdev_us = m << 1;	/* make sure rto = 3*rtt */  in tcp_rtt_estimator()
947 		tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk));  in tcp_rtt_estimator()
948 		tp->mdev_max_us = tp->rttvar_us;  in tcp_rtt_estimator()
949 		tp->rtt_seq = tp->snd_nxt;  in tcp_rtt_estimator()
953 	tp->srtt_us = max(1U, srtt);  in tcp_rtt_estimator()
962 	rate = (u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3);  in tcp_update_pacing_rate()
968 	 * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh)  in tcp_update_pacing_rate()
969 	 *	 If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching  in tcp_update_pacing_rate()
972 	if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2)  in tcp_update_pacing_rate()
973 		rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio);  in tcp_update_pacing_rate()
975 		rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio);  in tcp_update_pacing_rate()
977 	rate *= max(tcp_snd_cwnd(tp), tp->packets_out);  in tcp_update_pacing_rate()
979 	if (likely(tp->srtt_us))  in tcp_update_pacing_rate()
980 		do_div(rate, tp->srtt_us);  in tcp_update_pacing_rate()
986 	WRITE_ONCE(sk->sk_pacing_rate,  in tcp_update_pacing_rate()
987 		   min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate)));  in tcp_update_pacing_rate()
1003 	 *    is invisible. Actually, Linux-2.4 also generates erratic  in tcp_set_rto()
1006 	inet_csk(sk)->icsk_rto = __tcp_set_rto(tp);  in tcp_set_rto()
1026 	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);  in tcp_init_cwnd()
1030 	/* Timestamps for earliest and latest never-retransmitted segment
1043 /* Take a notice that peer is sending D-SACKs. Skip update of data delivery
1046  * - DSACKed sequence range is larger than maximum receiver's window.
1047  * - Total no. of DSACKed segments exceed the total no. of retransmitted segs.
1057 	seq_len = end_seq - start_seq;  in tcp_dsack_seen()
1059 	if (seq_len > tp->max_window)  in tcp_dsack_seen()
1061 	if (seq_len > tp->mss_cache)  in tcp_dsack_seen()
1062 		dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache);  in tcp_dsack_seen()
1063 	else if (tp->tlp_high_seq && tp->tlp_high_seq == end_seq)  in tcp_dsack_seen()
1064 		state->flag |= FLAG_DSACK_TLP;  in tcp_dsack_seen()
1066 	tp->dsack_dups += dup_segs;  in tcp_dsack_seen()
1068 	if (tp->dsack_dups > tp->total_retrans)  in tcp_dsack_seen()
1071 	tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;  in tcp_dsack_seen()
1076 	 * is timer-driven, not triggered by RACK).  in tcp_dsack_seen()
1078 	if (tp->reord_seen && !(state->flag & FLAG_DSACK_TLP))  in tcp_dsack_seen()
1079 		tp->rack.dsack_seen = 1;  in tcp_dsack_seen()
1081 	state->flag |= FLAG_DSACKING_ACK;  in tcp_dsack_seen()
1083 	state->sack_delivered += dup_segs;  in tcp_dsack_seen()
1089  * some lower never-retransmitted sequence ("low_seq"). The maximum reordering
1090  * distance is approximated in full-mss packet distance ("reordering").
1096 	const u32 mss = tp->mss_cache;  in tcp_check_sack_reordering()
1103 	metric = fack - low_seq;  in tcp_check_sack_reordering()
1104 	if ((metric > tp->reordering * mss) && mss) {  in tcp_check_sack_reordering()
1107 			 tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state,  in tcp_check_sack_reordering()
1108 			 tp->reordering,  in tcp_check_sack_reordering()
1110 			 tp->sacked_out,  in tcp_check_sack_reordering()
1111 			 tp->undo_marker ? tp->undo_retrans : 0);  in tcp_check_sack_reordering()
1113 		tp->reordering = min_t(u32, (metric + mss - 1) / mss,  in tcp_check_sack_reordering()
1114 				       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering));  in tcp_check_sack_reordering()
1118 	tp->reord_seen++;  in tcp_check_sack_reordering()
1130 	if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||  in tcp_verify_retransmit_hint()
1131 	    (tp->retransmit_skb_hint &&  in tcp_verify_retransmit_hint()
1132 	     before(TCP_SKB_CB(skb)->seq,  in tcp_verify_retransmit_hint()
1133 		    TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))  in tcp_verify_retransmit_hint()
1134 		tp->retransmit_skb_hint = skb;  in tcp_verify_retransmit_hint()
1142 	tp->lost += tcp_skb_pcount(skb);  in tcp_notify_skb_loss_event()
1147 	__u8 sacked = TCP_SKB_CB(skb)->sacked;  in tcp_mark_skb_lost()
1157 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;  in tcp_mark_skb_lost()
1158 			tp->retrans_out -= tcp_skb_pcount(skb);  in tcp_mark_skb_lost()
1164 		tp->lost_out += tcp_skb_pcount(skb);  in tcp_mark_skb_lost()
1165 		TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;  in tcp_mark_skb_lost()
1178  * 0	1		- orig segment is in flight.
1179  * S	0		- nothing flies, orig reached receiver.
1180  * L	0		- nothing flies, orig lost by net.
1181  * R	2		- both orig and retransmit are in flight.
1182  * L|R	1		- orig is lost, retransmit is in flight.
1183  * S|R  1		- orig reached receiver, retrans is still in flight.
1185  *  but it is equivalent to plain S and code short-circuits it to S.
1186  *  L|S is logically invalid, it would mean -1 packet in flight 8))
1196  * 4. D-SACK added new rule: D-SACK changes any tag to S.
1203  * --------------------
1208  *    ever retransmitted -> reordering. Alas, we cannot use it
1210  * 2. The last flaw is solved with D-SACK. D-SACK arrives
1211  *    for retransmitted and already SACKed segment -> reordering..
1216  * ----------------------
1231  * Implements also blockage to start_seq wrap-around. Problem lies in the
1237  *         <- outs wnd ->                          <- wrapzone ->
1240  * |<------------+------+----- TCP seqno space --------------+---------->|
1241  * ...-- <2^31 ->|                                           |<--------...
1242  * ...---- >2^31 ------>|                                    |<--------...
1247  * snd_nxt wrap -> snd_una region will then become "well defined", i.e.,
1250  * With D-SACK the lower bound is extended to cover sequence space below
1252  * again, D-SACK block must not to go across snd_una (for the same reason as
1254  * ends, TCP might receive valid D-SACKs below that. As long as they reside
1257  * theoretical ones), the D-SACK will nicely cross that boundary due to skb
1260  * the exact amount is rather hard to quantify. However, tp->max_window can
1267 	if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq))  in tcp_is_sackblock_valid()
1270 	/* Nasty start_seq wrap-around check (see comments above) */  in tcp_is_sackblock_valid()
1271 	if (!before(start_seq, tp->snd_nxt))  in tcp_is_sackblock_valid()
1274 	/* In outstanding window? ...This is valid exit for D-SACKs too.  in tcp_is_sackblock_valid()
1275 	 * start_seq == snd_una is non-sensical (see comments above)  in tcp_is_sackblock_valid()
1277 	if (after(start_seq, tp->snd_una))  in tcp_is_sackblock_valid()
1280 	if (!is_dsack || !tp->undo_marker)  in tcp_is_sackblock_valid()
1283 	/* ...Then it's D-SACK, and must reside below snd_una completely */  in tcp_is_sackblock_valid()
1284 	if (after(end_seq, tp->snd_una))  in tcp_is_sackblock_valid()
1287 	if (!before(start_seq, tp->undo_marker))  in tcp_is_sackblock_valid()
1291 	if (!after(end_seq, tp->undo_marker))  in tcp_is_sackblock_valid()
1297 	return !before(start_seq, end_seq - tp->max_window);  in tcp_is_sackblock_valid()
1309 	if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) {  in tcp_check_dsack()
1330 	/* D-SACK for already forgotten data... Do dumb counting. */  in tcp_check_dsack()
1331 	if (tp->undo_marker && tp->undo_retrans > 0 &&  in tcp_check_dsack()
1333 	    after(end_seq_0, tp->undo_marker))  in tcp_check_dsack()
1334 		tp->undo_retrans = max_t(int, 0, tp->undo_retrans - dup_segs);  in tcp_check_dsack()
1355 	in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&  in tcp_match_skb_to_sack()
1356 		  !before(end_seq, TCP_SKB_CB(skb)->end_seq);  in tcp_match_skb_to_sack()
1359 	    after(TCP_SKB_CB(skb)->end_seq, start_seq)) {  in tcp_match_skb_to_sack()
1361 		in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);  in tcp_match_skb_to_sack()
1364 			pkt_len = start_seq - TCP_SKB_CB(skb)->seq;  in tcp_match_skb_to_sack()
1368 			pkt_len = end_seq - TCP_SKB_CB(skb)->seq;  in tcp_match_skb_to_sack()
1370 				return -EINVAL;  in tcp_match_skb_to_sack()
1383 		if (pkt_len >= skb->len && !in_sack)  in tcp_match_skb_to_sack()
1395 /* Mark the given newly-SACKed range as such, adjusting counters and hints. */
1404 	/* Account D-SACK for retransmitted packet. */  in tcp_sacktag_one()
1406 		if (tp->undo_marker && tp->undo_retrans > 0 &&  in tcp_sacktag_one()
1407 		    after(end_seq, tp->undo_marker))  in tcp_sacktag_one()
1408 			tp->undo_retrans = max_t(int, 0, tp->undo_retrans - pcount);  in tcp_sacktag_one()
1410 		    before(start_seq, state->reord))  in tcp_sacktag_one()
1411 				state->reord = start_seq;  in tcp_sacktag_one()
1415 	if (!after(end_seq, tp->snd_una))  in tcp_sacktag_one()
1428 				tp->lost_out -= pcount;  in tcp_sacktag_one()
1429 				tp->retrans_out -= pcount;  in tcp_sacktag_one()
1438 				    before(start_seq, state->reord))  in tcp_sacktag_one()
1439 					state->reord = start_seq;  in tcp_sacktag_one()
1441 				if (!after(end_seq, tp->high_seq))  in tcp_sacktag_one()
1442 					state->flag |= FLAG_ORIG_SACK_ACKED;  in tcp_sacktag_one()
1443 				if (state->first_sackt == 0)  in tcp_sacktag_one()
1444 					state->first_sackt = xmit_time;  in tcp_sacktag_one()
1445 				state->last_sackt = xmit_time;  in tcp_sacktag_one()
1450 				tp->lost_out -= pcount;  in tcp_sacktag_one()
1455 		state->flag |= FLAG_DATA_SACKED;  in tcp_sacktag_one()
1456 		tp->sacked_out += pcount;  in tcp_sacktag_one()
1457 		/* Out-of-order packets delivered */  in tcp_sacktag_one()
1458 		state->sack_delivered += pcount;  in tcp_sacktag_one()
1461 		if (tp->lost_skb_hint &&  in tcp_sacktag_one()
1462 		    before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq))  in tcp_sacktag_one()
1463 			tp->lost_cnt_hint += pcount;  in tcp_sacktag_one()
1466 	/* D-SACK. We can detect redundant retransmission in S|R and plain R  in tcp_sacktag_one()
1472 		tp->retrans_out -= pcount;  in tcp_sacktag_one()
1478 /* Shift newly-SACKed bytes from this skb to the immediately previous
1479  * already-SACKed sk_buff. Mark the newly-SACKed bytes as such.
1488 	u32 start_seq = TCP_SKB_CB(skb)->seq;	/* start of newly-SACKed */  in tcp_shifted_skb()
1489 	u32 end_seq = start_seq + shifted;	/* end of newly-SACKed */  in tcp_shifted_skb()
1499 	tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,  in tcp_shifted_skb()
1502 	tcp_rate_skb_delivered(sk, skb, state->rate);  in tcp_shifted_skb()
1504 	if (skb == tp->lost_skb_hint)  in tcp_shifted_skb()
1505 		tp->lost_cnt_hint += pcount;  in tcp_shifted_skb()
1507 	TCP_SKB_CB(prev)->end_seq += shifted;  in tcp_shifted_skb()
1508 	TCP_SKB_CB(skb)->seq += shifted;  in tcp_shifted_skb()
1512 	tcp_skb_pcount_add(skb, -pcount);  in tcp_shifted_skb()
1519 	if (!TCP_SKB_CB(prev)->tcp_gso_size)  in tcp_shifted_skb()
1520 		TCP_SKB_CB(prev)->tcp_gso_size = mss;  in tcp_shifted_skb()
1524 		TCP_SKB_CB(skb)->tcp_gso_size = 0;  in tcp_shifted_skb()
1527 	TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS);  in tcp_shifted_skb()
1529 	if (skb->len > 0) {  in tcp_shifted_skb()
1535 	/* Whole SKB was eaten :-) */  in tcp_shifted_skb()
1537 	if (skb == tp->retransmit_skb_hint)  in tcp_shifted_skb()
1538 		tp->retransmit_skb_hint = prev;  in tcp_shifted_skb()
1539 	if (skb == tp->lost_skb_hint) {  in tcp_shifted_skb()
1540 		tp->lost_skb_hint = prev;  in tcp_shifted_skb()
1541 		tp->lost_cnt_hint -= tcp_skb_pcount(prev);  in tcp_shifted_skb()
1544 	TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;  in tcp_shifted_skb()
1545 	TCP_SKB_CB(prev)->eor = TCP_SKB_CB(skb)->eor;  in tcp_shifted_skb()
1546 	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)  in tcp_shifted_skb()
1547 		TCP_SKB_CB(prev)->end_seq++;  in tcp_shifted_skb()
1553 	if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp))  in tcp_shifted_skb()
1554 		TCP_SKB_CB(prev)->tx.delivered_mstamp = 0;  in tcp_shifted_skb()
1564  * something-or-zero which complicates things
1568 	return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb);  in tcp_skb_seglen()
1581 	 * Since TCP_SKB_CB(skb)->tcp_gso_segs is 16 bits, we need  in tcp_skb_shift()
1585 	if (unlikely(to->len + shiftlen >= 65535 * TCP_MIN_GSO_SIZE))  in tcp_skb_shift()
1609 	    (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS)  in tcp_shift_skb_data()
1614 	if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una))  in tcp_shift_skb_data()
1622 	if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED)  in tcp_shift_skb_data()
1628 	in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&  in tcp_shift_skb_data()
1629 		  !before(end_seq, TCP_SKB_CB(skb)->end_seq);  in tcp_shift_skb_data()
1632 		len = skb->len;  in tcp_shift_skb_data()
1642 		if (!after(TCP_SKB_CB(skb)->end_seq, start_seq))  in tcp_shift_skb_data()
1644 		/* CHECKME: This is non-MSS split case only?, this will  in tcp_shift_skb_data()
1651 		in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);  in tcp_shift_skb_data()
1654 			 * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)),  in tcp_shift_skb_data()
1658 			 * previously. We could try merging non-SACKed ones  in tcp_shift_skb_data()
1667 		len = end_seq - TCP_SKB_CB(skb)->seq;  in tcp_shift_skb_data()
1669 		BUG_ON(len > skb->len);  in tcp_shift_skb_data()
1693 	/* tcp_sacktag_one() won't SACK-tag ranges below snd_una */  in tcp_shift_skb_data()
1694 	if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una))  in tcp_shift_skb_data()
1710 	    ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) ||  in tcp_shift_skb_data()
1716 	len = skb->len;  in tcp_shift_skb_data()
1746 		/* queue is in-order => we can short-circuit the walk early */  in tcp_sacktag_walk()
1747 		if (!before(TCP_SKB_CB(skb)->seq, end_seq))  in tcp_sacktag_walk()
1751 		    before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) {  in tcp_sacktag_walk()
1753 							next_dup->start_seq,  in tcp_sacktag_walk()
1754 							next_dup->end_seq);  in tcp_sacktag_walk()
1784 			TCP_SKB_CB(skb)->sacked =  in tcp_sacktag_walk()
1787 						TCP_SKB_CB(skb)->sacked,  in tcp_sacktag_walk()
1788 						TCP_SKB_CB(skb)->seq,  in tcp_sacktag_walk()
1789 						TCP_SKB_CB(skb)->end_seq,  in tcp_sacktag_walk()
1793 			tcp_rate_skb_delivered(sk, skb, state->rate);  in tcp_sacktag_walk()
1794 			if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_sacktag_walk()
1795 				list_del_init(&skb->tcp_tsorted_anchor);  in tcp_sacktag_walk()
1797 			if (!before(TCP_SKB_CB(skb)->seq,  in tcp_sacktag_walk()
1807 	struct rb_node *parent, **p = &sk->tcp_rtx_queue.rb_node;  in tcp_sacktag_bsearch()
1813 		if (before(seq, TCP_SKB_CB(skb)->seq)) {  in tcp_sacktag_bsearch()
1814 			p = &parent->rb_left;  in tcp_sacktag_bsearch()
1817 		if (!before(seq, TCP_SKB_CB(skb)->end_seq)) {  in tcp_sacktag_bsearch()
1818 			p = &parent->rb_right;  in tcp_sacktag_bsearch()
1829 	if (skb && after(TCP_SKB_CB(skb)->seq, skip_to_seq))  in tcp_sacktag_skip()
1844 	if (before(next_dup->start_seq, skip_to_seq)) {  in tcp_maybe_skipping_dsack()
1845 		skb = tcp_sacktag_skip(skb, sk, next_dup->start_seq);  in tcp_maybe_skipping_dsack()
1847 				       next_dup->start_seq, next_dup->end_seq,  in tcp_maybe_skipping_dsack()
1856 	return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);  in tcp_sack_cache_ok()
1865 				    TCP_SKB_CB(ack_skb)->sacked);  in tcp_sacktag_write_queue()
1870 	int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3);  in tcp_sacktag_write_queue()
1876 	state->flag = 0;  in tcp_sacktag_write_queue()
1877 	state->reord = tp->snd_nxt;  in tcp_sacktag_write_queue()
1879 	if (!tp->sacked_out)  in tcp_sacktag_write_queue()
1889 	if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window))  in tcp_sacktag_write_queue()
1892 	if (!tp->packets_out)  in tcp_sacktag_write_queue()
1909 				if (!tp->undo_marker)  in tcp_sacktag_write_queue()
1915 				if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) &&  in tcp_sacktag_write_queue()
1916 				    !after(sp[used_sacks].end_seq, tp->snd_una))  in tcp_sacktag_write_queue()
1923 				first_sack_index = -1;  in tcp_sacktag_write_queue()
1930 				first_sack_index = -1;  in tcp_sacktag_write_queue()
1938 	for (i = used_sacks - 1; i > 0; i--) {  in tcp_sacktag_write_queue()
1950 	state->mss_now = tcp_current_mss(sk);  in tcp_sacktag_write_queue()
1954 	if (!tp->sacked_out) {  in tcp_sacktag_write_queue()
1956 		cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);  in tcp_sacktag_write_queue()
1958 		cache = tp->recv_sack_cache;  in tcp_sacktag_write_queue()
1960 		while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq &&  in tcp_sacktag_write_queue()
1961 		       !cache->end_seq)  in tcp_sacktag_write_queue()
1976 		       !before(start_seq, cache->end_seq))  in tcp_sacktag_write_queue()
1981 		    after(end_seq, cache->start_seq)) {  in tcp_sacktag_write_queue()
1984 			if (before(start_seq, cache->start_seq)) {  in tcp_sacktag_write_queue()
1989 						       cache->start_seq,  in tcp_sacktag_write_queue()
1994 			if (!after(end_seq, cache->end_seq))  in tcp_sacktag_write_queue()
1999 						       cache->end_seq);  in tcp_sacktag_write_queue()
2002 			if (tcp_highest_sack_seq(tp) == cache->end_seq) {  in tcp_sacktag_write_queue()
2011 			skb = tcp_sacktag_skip(skb, sk, cache->end_seq);  in tcp_sacktag_write_queue()
2033 	for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) {  in tcp_sacktag_write_queue()
2034 		tp->recv_sack_cache[i].start_seq = 0;  in tcp_sacktag_write_queue()
2035 		tp->recv_sack_cache[i].end_seq = 0;  in tcp_sacktag_write_queue()
2038 		tp->recv_sack_cache[i++] = sp[j];  in tcp_sacktag_write_queue()
2040 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss || tp->undo_marker)  in tcp_sacktag_write_queue()
2041 		tcp_check_sack_reordering(sk, state->reord, 0);  in tcp_sacktag_write_queue()
2047 	WARN_ON((int)tp->sacked_out < 0);  in tcp_sacktag_write_queue()
2048 	WARN_ON((int)tp->lost_out < 0);  in tcp_sacktag_write_queue()
2049 	WARN_ON((int)tp->retrans_out < 0);  in tcp_sacktag_write_queue()
2052 	return state->flag;  in tcp_sacktag_write_queue()
2062 	holes = max(tp->lost_out, 1U);  in tcp_limit_reno_sacked()
2063 	holes = min(holes, tp->packets_out);  in tcp_limit_reno_sacked()
2065 	if ((tp->sacked_out + holes) > tp->packets_out) {  in tcp_limit_reno_sacked()
2066 		tp->sacked_out = tp->packets_out - holes;  in tcp_limit_reno_sacked()
2083 	tp->reordering = min_t(u32, tp->packets_out + addend,  in tcp_check_reno_reordering()
2084 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering));  in tcp_check_reno_reordering()
2085 	tp->reord_seen++;  in tcp_check_reno_reordering()
2095 		u32 prior_sacked = tp->sacked_out;  in tcp_add_reno_sack()
2098 		tp->sacked_out += num_dupack;  in tcp_add_reno_sack()
2100 		delivered = tp->sacked_out - prior_sacked;  in tcp_add_reno_sack()
2115 		tcp_count_delivered(tp, max_t(int, acked - tp->sacked_out, 1),  in tcp_remove_reno_sacks()
2117 		if (acked - 1 >= tp->sacked_out)  in tcp_remove_reno_sacks()
2118 			tp->sacked_out = 0;  in tcp_remove_reno_sacks()
2120 			tp->sacked_out -= acked - 1;  in tcp_remove_reno_sacks()
2128 	tp->sacked_out = 0;  in tcp_reset_reno_sack()
2133 	tp->retrans_out = 0;  in tcp_clear_retrans()
2134 	tp->lost_out = 0;  in tcp_clear_retrans()
2135 	tp->undo_marker = 0;  in tcp_clear_retrans()
2136 	tp->undo_retrans = -1;  in tcp_clear_retrans()
2137 	tp->sacked_out = 0;  in tcp_clear_retrans()
2138 	tp->rto_stamp = 0;  in tcp_clear_retrans()
2139 	tp->total_rto = 0;  in tcp_clear_retrans()
2140 	tp->total_rto_recoveries = 0;  in tcp_clear_retrans()
2141 	tp->total_rto_time = 0;  in tcp_clear_retrans()
2146 	tp->undo_marker = tp->snd_una;  in tcp_init_undo()
2150 	tp->undo_retrans = tp->retrans_out;  in tcp_init_undo()
2152 	if (tp->tlp_high_seq && tp->tlp_retrans)  in tcp_init_undo()
2153 		tp->undo_retrans++;  in tcp_init_undo()
2155 	if (!tp->undo_retrans)  in tcp_init_undo()
2156 		tp->undo_retrans = -1;  in tcp_init_undo()
2161 	return READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_recovery) &  in tcp_is_rack()
2176 	is_reneg = head && (TCP_SKB_CB(head)->sacked & TCPCB_SACKED_ACKED);  in tcp_timeout_mark_lost()
2179 		tp->sacked_out = 0;  in tcp_timeout_mark_lost()
2181 		tp->is_sack_reneg = 1;  in tcp_timeout_mark_lost()
2189 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED;  in tcp_timeout_mark_lost()
2205 	bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery;  in tcp_enter_loss()
2211 	if (icsk->icsk_ca_state <= TCP_CA_Disorder ||  in tcp_enter_loss()
2212 	    !after(tp->high_seq, tp->snd_una) ||  in tcp_enter_loss()
2213 	    (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) {  in tcp_enter_loss()
2214 		tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_enter_loss()
2215 		tp->prior_cwnd = tcp_snd_cwnd(tp);  in tcp_enter_loss()
2216 		tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);  in tcp_enter_loss()
2221 	tp->snd_cwnd_cnt   = 0;  in tcp_enter_loss()
2222 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_enter_loss()
2225 	 * suggests that the degree of reordering is over-estimated.  in tcp_enter_loss()
2227 	reordering = READ_ONCE(net->ipv4.sysctl_tcp_reordering);  in tcp_enter_loss()
2228 	if (icsk->icsk_ca_state <= TCP_CA_Disorder &&  in tcp_enter_loss()
2229 	    tp->sacked_out >= reordering)  in tcp_enter_loss()
2230 		tp->reordering = min_t(unsigned int, tp->reordering,  in tcp_enter_loss()
2234 	tp->high_seq = tp->snd_nxt;  in tcp_enter_loss()
2235 	tp->tlp_high_seq = 0;  in tcp_enter_loss()
2238 	/* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous  in tcp_enter_loss()
2240 	 * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing  in tcp_enter_loss()
2242 	tp->frto = READ_ONCE(net->ipv4.sysctl_tcp_frto) &&  in tcp_enter_loss()
2243 		   (new_recovery || icsk->icsk_retransmits) &&  in tcp_enter_loss()
2244 		   !inet_csk(sk)->icsk_mtup.probe_size;  in tcp_enter_loss()
2262 		unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4),  in tcp_check_sack_reneging()
2279  * but when e.g. out-of-window ACKs or packet duplication occurs,
2285 	return tp->sacked_out + 1;  in tcp_dupack_heuristics()
2289  * --------------------------------------
2298  * "Recovery"	CWND was reduced, we are fast-retransmitting.
2302  * - each incoming ACK, if state is not "Open"
2303  * - when arrived ACK is unusual, namely:
2310  *	in_flight = packets_out - left_out + retrans_out
2312  *	packets_out is SND.NXT-SND.UNA counted in packets.
2346  *		RACK (draft-ietf-tcpm-rack-01): it is a newer algorithm
2347  *		(2017-) that checks timing instead of counting DUPACKs.
2373  * timestamps and D-SACKs. When we detect that some segments were
2390 	if (tp->lost_out)  in tcp_time_to_recover()
2393 	/* Not-A-Trick#2 : Classic rule... */  in tcp_time_to_recover()
2394 	if (!tcp_is_rack(sk) && tcp_dupack_heuristics(tp) > tp->reordering)  in tcp_time_to_recover()
2402  * has at least tp->reordering SACKed seqments above it; "packets" refers to
2411 	const u32 loss_high = tp->snd_nxt;  in tcp_mark_head_lost()
2413 	WARN_ON(packets > tp->packets_out);  in tcp_mark_head_lost()
2414 	skb = tp->lost_skb_hint;  in tcp_mark_head_lost()
2417 		if (mark_head && after(TCP_SKB_CB(skb)->seq, tp->snd_una))  in tcp_mark_head_lost()
2419 		cnt = tp->lost_cnt_hint;  in tcp_mark_head_lost()
2428 		tp->lost_skb_hint = skb;  in tcp_mark_head_lost()
2429 		tp->lost_cnt_hint = cnt;  in tcp_mark_head_lost()
2431 		if (after(TCP_SKB_CB(skb)->end_seq, loss_high))  in tcp_mark_head_lost()
2434 		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_mark_head_lost()
2440 		if (!(TCP_SKB_CB(skb)->sacked & TCPCB_LOST))  in tcp_mark_head_lost()
2456 		int sacked_upto = tp->sacked_out - tp->reordering;  in tcp_update_scoreboard()
2466 	return tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&  in tcp_tsopt_ecr_before()
2467 	       before(tp->rx_opt.rcv_tsecr, when);  in tcp_tsopt_ecr_before()
2476 	return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) &&  in tcp_skb_spurious_retrans()
2477 	       tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));  in tcp_skb_spurious_retrans()
2487 	if (tp->retrans_stamp &&  in tcp_packet_delayed()
2488 	    tcp_tsopt_ecr_before(tp, tp->retrans_stamp))  in tcp_packet_delayed()
2496 	if (!tp->retrans_stamp &&	   /* no record of a retransmit/SYN? */  in tcp_packet_delayed()
2497 	    sk->sk_state != TCP_SYN_SENT)  /* not the FLAG_SYN_ACKED case? */  in tcp_packet_delayed()
2507  * tp->retrans_out, however, that kind of assumptions doesn't consider
2524 	if (tp->retrans_out)  in tcp_any_retrans_done()
2528 	if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS))  in tcp_any_retrans_done()
2536  * retransmits_timed_out() and timestamp-undo are using the correct value.
2541 		tcp_sk(sk)->retrans_stamp = 0;  in tcp_retrans_stamp_cleanup()
2550 	if (sk->sk_family == AF_INET) {  in DBGUNDO()
2553 			 &inet->inet_daddr, ntohs(inet->inet_dport),  in DBGUNDO()
2555 			 tp->snd_ssthresh, tp->prior_ssthresh,  in DBGUNDO()
2556 			 tp->packets_out);  in DBGUNDO()
2559 	else if (sk->sk_family == AF_INET6) {  in DBGUNDO()
2562 			 &sk->sk_v6_daddr, ntohs(inet->inet_dport),  in DBGUNDO()
2564 			 tp->snd_ssthresh, tp->prior_ssthresh,  in DBGUNDO()
2565 			 tp->packets_out);  in DBGUNDO()
2578 		skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {  in tcp_undo_cwnd_reduction()
2579 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST;  in tcp_undo_cwnd_reduction()
2581 		tp->lost_out = 0;  in tcp_undo_cwnd_reduction()
2585 	if (tp->prior_ssthresh) {  in tcp_undo_cwnd_reduction()
2588 		tcp_snd_cwnd_set(tp, icsk->icsk_ca_ops->undo_cwnd(sk));  in tcp_undo_cwnd_reduction()
2590 		if (tp->prior_ssthresh > tp->snd_ssthresh) {  in tcp_undo_cwnd_reduction()
2591 			tp->snd_ssthresh = tp->prior_ssthresh;  in tcp_undo_cwnd_reduction()
2595 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_undo_cwnd_reduction()
2596 	tp->undo_marker = 0;  in tcp_undo_cwnd_reduction()
2597 	tp->rack.advanced = 1; /* Force RACK to re-exam losses */  in tcp_undo_cwnd_reduction()
2602 	return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp));  in tcp_may_undo()
2609 	if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) {  in tcp_is_non_sack_preventing_reopen()
2614 			tp->retrans_stamp = 0;  in tcp_is_non_sack_preventing_reopen()
2631 		DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans");  in tcp_try_undo_recovery()
2633 		if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss)  in tcp_try_undo_recovery()
2639 	} else if (tp->rack.reo_wnd_persist) {  in tcp_try_undo_recovery()
2640 		tp->rack.reo_wnd_persist--;  in tcp_try_undo_recovery()
2645 	tp->is_sack_reneg = 0;  in tcp_try_undo_recovery()
2649 /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
2654 	if (tp->undo_marker && !tp->undo_retrans) {  in tcp_try_undo_dsack()
2655 		tp->rack.reo_wnd_persist = min(TCP_RACK_RECOVERY_THRESH,  in tcp_try_undo_dsack()
2656 					       tp->rack.reo_wnd_persist + 1);  in tcp_try_undo_dsack()
2657 		DBGUNDO(sk, "D-SACK");  in tcp_try_undo_dsack()
2665 /* Undo during loss recovery after partial ACK or using F-RTO. */
2678 		inet_csk(sk)->icsk_retransmits = 0;  in tcp_try_undo_loss()
2683 			tp->is_sack_reneg = 0;  in tcp_try_undo_loss()
2703 	tp->high_seq = tp->snd_nxt;  in tcp_init_cwnd_reduction()
2704 	tp->tlp_high_seq = 0;  in tcp_init_cwnd_reduction()
2705 	tp->snd_cwnd_cnt = 0;  in tcp_init_cwnd_reduction()
2706 	tp->prior_cwnd = tcp_snd_cwnd(tp);  in tcp_init_cwnd_reduction()
2707 	tp->prr_delivered = 0;  in tcp_init_cwnd_reduction()
2708 	tp->prr_out = 0;  in tcp_init_cwnd_reduction()
2709 	tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk);  in tcp_init_cwnd_reduction()
2717 	int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp);  in tcp_cwnd_reduction()
2719 	if (newly_acked_sacked <= 0 || WARN_ON_ONCE(!tp->prior_cwnd))  in tcp_cwnd_reduction()
2724 	tp->prr_delivered += newly_acked_sacked;  in tcp_cwnd_reduction()
2726 		u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered +  in tcp_cwnd_reduction()
2727 			       tp->prior_cwnd - 1;  in tcp_cwnd_reduction()
2728 		sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out;  in tcp_cwnd_reduction()
2730 		sndcnt = max_t(int, tp->prr_delivered - tp->prr_out,  in tcp_cwnd_reduction()
2737 	sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1));  in tcp_cwnd_reduction()
2745 	if (inet_csk(sk)->icsk_ca_ops->cong_control)  in tcp_end_cwnd_reduction()
2749 	if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH &&  in tcp_end_cwnd_reduction()
2750 	    (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) {  in tcp_end_cwnd_reduction()
2751 		tcp_snd_cwnd_set(tp, tp->snd_ssthresh);  in tcp_end_cwnd_reduction()
2752 		tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_end_cwnd_reduction()
2762 	tp->prior_ssthresh = 0;  in tcp_enter_cwr()
2763 	if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {  in tcp_enter_cwr()
2764 		tp->undo_marker = 0;  in tcp_enter_cwr()
2779 	if (inet_csk(sk)->icsk_ca_state != state) {  in tcp_try_keep_open()
2781 		tp->high_seq = tp->snd_nxt;  in tcp_try_keep_open()
2792 		tp->retrans_stamp = 0;  in tcp_try_to_open()
2797 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {  in tcp_try_to_open()
2806 	icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1;  in tcp_mtup_probe_failed()
2807 	icsk->icsk_mtup.probe_size = 0;  in tcp_mtup_probe_failed()
2817 	tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_mtup_probe_success()
2819 	val = (u64)tcp_snd_cwnd(tp) * tcp_mss_to_mtu(sk, tp->mss_cache);  in tcp_mtup_probe_success()
2820 	do_div(val, icsk->icsk_mtup.probe_size);  in tcp_mtup_probe_success()
2824 	tp->snd_cwnd_cnt = 0;  in tcp_mtup_probe_success()
2825 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_mtup_probe_success()
2826 	tp->snd_ssthresh = tcp_current_ssthresh(sk);  in tcp_mtup_probe_success()
2828 	icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size;  in tcp_mtup_probe_success()
2829 	icsk->icsk_mtup.probe_size = 0;  in tcp_mtup_probe_success()
2830 	tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_mtup_probe_success()
2838  * non-zero value (and may do so in a later calling context due to TSQ), we
2849 	if (icsk->icsk_ca_state != TCP_CA_Loss) {  in tcp_non_congestion_loss_retransmit()
2850 		tp->high_seq = tp->snd_nxt;  in tcp_non_congestion_loss_retransmit()
2851 		tp->snd_ssthresh = tcp_current_ssthresh(sk);  in tcp_non_congestion_loss_retransmit()
2852 		tp->prior_ssthresh = 0;  in tcp_non_congestion_loss_retransmit()
2853 		tp->undo_marker = 0;  in tcp_non_congestion_loss_retransmit()
2877 	 * -1 to mark all frames as lost, otherwise compute the current MSS.  in tcp_simple_retransmit()
2879 	if (tp->syn_data && sk->sk_state == TCP_SYN_SENT)  in tcp_simple_retransmit()
2880 		mss = -1;  in tcp_simple_retransmit()
2884 	skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {  in tcp_simple_retransmit()
2891 	if (!tp->lost_out)  in tcp_simple_retransmit()
2923 	tp->prior_ssthresh = 0;  in tcp_enter_recovery()
2928 			tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_enter_recovery()
2936 	if (tp->rto_stamp) {  in tcp_update_rto_time()
2937 		tp->total_rto_time += tcp_time_stamp_ms(tp) - tp->rto_stamp;  in tcp_update_rto_time()
2938 		tp->rto_stamp = 0;  in tcp_update_rto_time()
2949 	bool recovered = !before(tp->snd_una, tp->high_seq);  in tcp_process_loss()
2951 	if ((flag & FLAG_SND_UNA_ADVANCED || rcu_access_pointer(tp->fastopen_rsk)) &&  in tcp_process_loss()
2955 	if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */  in tcp_process_loss()
2957 		 * lost, i.e., never-retransmitted data are (s)acked.  in tcp_process_loss()
2963 		if (after(tp->snd_nxt, tp->high_seq)) {  in tcp_process_loss()
2965 				tp->frto = 0; /* Step 3.a. loss was real */  in tcp_process_loss()
2967 			tp->high_seq = tp->snd_nxt;  in tcp_process_loss()
2973 			    after(tcp_wnd_end(tp), tp->snd_nxt)) {  in tcp_process_loss()
2977 			tp->frto = 0;  in tcp_process_loss()
2982 		/* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */  in tcp_process_loss()
2987 		/* A Reno DUPACK means new data in F-RTO step 2.b above are  in tcp_process_loss()
2990 		if (after(tp->snd_nxt, tp->high_seq) && num_dupack)  in tcp_process_loss()
3003 		     tp->snd_una + tp->reordering * tp->mss_cache);  in tcp_force_fast_retransmit()
3012 	if (tp->undo_marker && tcp_packet_delayed(tp)) {  in tcp_try_undo_partial()
3023 		if (tp->retrans_out)  in tcp_try_undo_partial()
3027 			tp->retrans_stamp = 0;  in tcp_try_undo_partial()
3050 		u32 prior_retrans = tp->retrans_out;  in tcp_identify_packet_loss()
3054 		if (prior_retrans > tp->retrans_out)  in tcp_identify_packet_loss()
3059 /* Process an event, which can update packets-in-flight not trivially.
3081 	if (!tp->packets_out && tp->sacked_out)  in tcp_fastretrans_alert()
3082 		tp->sacked_out = 0;  in tcp_fastretrans_alert()
3087 		tp->prior_ssthresh = 0;  in tcp_fastretrans_alert()
3098 	if (icsk->icsk_ca_state == TCP_CA_Open) {  in tcp_fastretrans_alert()
3099 		WARN_ON(tp->retrans_out != 0 && !tp->syn_data);  in tcp_fastretrans_alert()
3100 		tp->retrans_stamp = 0;  in tcp_fastretrans_alert()
3101 	} else if (!before(tp->snd_una, tp->high_seq)) {  in tcp_fastretrans_alert()
3102 		switch (icsk->icsk_ca_state) {  in tcp_fastretrans_alert()
3106 			if (tp->snd_una != tp->high_seq) {  in tcp_fastretrans_alert()
3123 	switch (icsk->icsk_ca_state) {  in tcp_fastretrans_alert()
3135 		if (icsk->icsk_ca_state != TCP_CA_Recovery) {  in tcp_fastretrans_alert()
3146 		if (icsk->icsk_ca_state != TCP_CA_Loss)  in tcp_fastretrans_alert()
3149 		if (!(icsk->icsk_ca_state == TCP_CA_Open ||  in tcp_fastretrans_alert()
3161 		if (icsk->icsk_ca_state <= TCP_CA_Disorder)  in tcp_fastretrans_alert()
3171 		if (icsk->icsk_ca_state < TCP_CA_CWR &&  in tcp_fastretrans_alert()
3172 		    icsk->icsk_mtup.probe_size &&  in tcp_fastretrans_alert()
3173 		    tp->snd_una == tp->mtu_probe.probe_seq_start) {  in tcp_fastretrans_alert()
3193 	u32 wlen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen) * HZ;  in tcp_update_rtt_min()
3203 	minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32,  in tcp_update_rtt_min()
3213 	/* Prefer RTT measured from ACK's timing to TS-ECR. This is because  in tcp_ack_update_rtt()
3214 	 * broken middle-boxes or peers may corrupt TS-ECR fields. But  in tcp_ack_update_rtt()
3225 	 * See draft-ietf-tcplw-high-performance-00, section 3.3.  in tcp_ack_update_rtt()
3227 	if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp &&  in tcp_ack_update_rtt()
3228 	    tp->rx_opt.rcv_tsecr && flag & FLAG_ACKED)  in tcp_ack_update_rtt()
3231 	rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */  in tcp_ack_update_rtt()
3236 	 * always taken together with ACK, SACK, or TS-opts. Any negative  in tcp_ack_update_rtt()
3244 	inet_csk(sk)->icsk_backoff = 0;  in tcp_ack_update_rtt()
3252 	long rtt_us = -1L;  in tcp_synack_rtt_meas()
3254 	if (req && !req->num_retrans && tcp_rsk(req)->snt_synack)  in tcp_synack_rtt_meas()
3255 		rtt_us = tcp_stamp_us_delta(tcp_clock_us(), tcp_rsk(req)->snt_synack);  in tcp_synack_rtt_meas()
3257 	tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, rtt_us, -1L, rtt_us, &rs);  in tcp_synack_rtt_meas()
3265 	icsk->icsk_ca_ops->cong_avoid(sk, ack, acked);  in tcp_cong_avoid()
3266 	tcp_sk(sk)->snd_cwnd_stamp = tcp_jiffies32;  in tcp_cong_avoid()
3278 	 * for SYN-ACK retrans purpose, stay put.  in tcp_rearm_rto()
3280 	if (rcu_access_pointer(tp->fastopen_rsk))  in tcp_rearm_rto()
3283 	if (!tp->packets_out) {  in tcp_rearm_rto()
3286 		u32 rto = inet_csk(sk)->icsk_rto;  in tcp_rearm_rto()
3288 		if (icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||  in tcp_rearm_rto()
3289 		    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {  in tcp_rearm_rto()
3313 	BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una));  in tcp_tso_acked()
3316 	if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))  in tcp_tso_acked()
3318 	packets_acked -= tcp_skb_pcount(skb);  in tcp_tso_acked()
3322 		BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq));  in tcp_tso_acked()
3333 	/* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */  in tcp_ack_tstamp()
3334 	if (likely(!TCP_SKB_CB(skb)->txstamp_ack))  in tcp_ack_tstamp()
3338 	if (!before(shinfo->tskey, prior_snd_una) &&  in tcp_ack_tstamp()
3339 	    before(shinfo->tskey, tcp_sk(sk)->snd_una)) {  in tcp_ack_tstamp()
3357 	u32 prior_sacked = tp->sacked_out;  in tcp_clean_rtx_queue()
3358 	u32 reord = tp->snd_nxt; /* lowest acked un-retx un-sacked seq */  in tcp_clean_rtx_queue()
3361 	long sack_rtt_us = -1L;  in tcp_clean_rtx_queue()
3362 	long seq_rtt_us = -1L;  in tcp_clean_rtx_queue()
3363 	long ca_rtt_us = -1L;  in tcp_clean_rtx_queue()
3370 	for (skb = skb_rb_first(&sk->tcp_rtx_queue); skb; skb = next) {  in tcp_clean_rtx_queue()
3372 		const u32 start_seq = scb->seq;  in tcp_clean_rtx_queue()
3373 		u8 sacked = scb->sacked;  in tcp_clean_rtx_queue()
3377 		if (after(scb->end_seq, tp->snd_una)) {  in tcp_clean_rtx_queue()
3379 			    !after(tp->snd_una, scb->seq))  in tcp_clean_rtx_queue()
3392 				tp->retrans_out -= acked_pcount;  in tcp_clean_rtx_queue()
3402 			if (!after(scb->end_seq, tp->high_seq))  in tcp_clean_rtx_queue()
3407 			tp->sacked_out -= acked_pcount;  in tcp_clean_rtx_queue()
3411 				tcp_rack_advance(tp, sacked, scb->end_seq,  in tcp_clean_rtx_queue()
3415 			tp->lost_out -= acked_pcount;  in tcp_clean_rtx_queue()
3417 		tp->packets_out -= acked_pcount;  in tcp_clean_rtx_queue()
3419 		tcp_rate_skb_delivered(sk, skb, sack->rate);  in tcp_clean_rtx_queue()
3428 		if (likely(!(scb->tcp_flags & TCPHDR_SYN))) {  in tcp_clean_rtx_queue()
3432 			tp->retrans_stamp = 0;  in tcp_clean_rtx_queue()
3441 		if (unlikely(skb == tp->retransmit_skb_hint))  in tcp_clean_rtx_queue()
3442 			tp->retransmit_skb_hint = NULL;  in tcp_clean_rtx_queue()
3443 		if (unlikely(skb == tp->lost_skb_hint))  in tcp_clean_rtx_queue()
3444 			tp->lost_skb_hint = NULL;  in tcp_clean_rtx_queue()
3452 	if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una)))  in tcp_clean_rtx_queue()
3453 		tp->snd_up = tp->snd_una;  in tcp_clean_rtx_queue()
3457 		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_clean_rtx_queue()
3462 		seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt);  in tcp_clean_rtx_queue()
3463 		ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt);  in tcp_clean_rtx_queue()
3466 		    (tp->snd_una - prior_snd_una) < tp->mss_cache &&  in tcp_clean_rtx_queue()
3467 		    sack->rate->prior_delivered + 1 == tp->delivered &&  in tcp_clean_rtx_queue()
3471 			 * a receiver w/o out-of-order or CE events.  in tcp_clean_rtx_queue()
3476 	if (sack->first_sackt) {  in tcp_clean_rtx_queue()
3477 		sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt);  in tcp_clean_rtx_queue()
3478 		ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->last_sackt);  in tcp_clean_rtx_queue()
3481 					ca_rtt_us, sack->rate);  in tcp_clean_rtx_queue()
3485 		if (unlikely(icsk->icsk_mtup.probe_size &&  in tcp_clean_rtx_queue()
3486 			     !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) {  in tcp_clean_rtx_queue()
3494 			 * retransmitted, non-SACK case cannot confirm that  in tcp_clean_rtx_queue()
3504 			/* Non-retransmitted hole got filled? That's reordering */  in tcp_clean_rtx_queue()
3508 			delta = prior_sacked - tp->sacked_out;  in tcp_clean_rtx_queue()
3509 			tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta);  in tcp_clean_rtx_queue()
3512 		   sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp,  in tcp_clean_rtx_queue()
3514 		/* Do not re-arm RTO if the sack RTT is measured from data sent  in tcp_clean_rtx_queue()
3521 	if (icsk->icsk_ca_ops->pkts_acked) {  in tcp_clean_rtx_queue()
3523 					     .rtt_us = sack->rate->rtt_us };  in tcp_clean_rtx_queue()
3525 		sample.in_flight = tp->mss_cache *  in tcp_clean_rtx_queue()
3526 			(tp->delivered - sack->rate->prior_delivered);  in tcp_clean_rtx_queue()
3527 		icsk->icsk_ca_ops->pkts_acked(sk, &sample);  in tcp_clean_rtx_queue()
3531 	WARN_ON((int)tp->sacked_out < 0);  in tcp_clean_rtx_queue()
3532 	WARN_ON((int)tp->lost_out < 0);  in tcp_clean_rtx_queue()
3533 	WARN_ON((int)tp->retrans_out < 0);  in tcp_clean_rtx_queue()
3534 	if (!tp->packets_out && tcp_is_sack(tp)) {  in tcp_clean_rtx_queue()
3536 		if (tp->lost_out) {  in tcp_clean_rtx_queue()
3538 				 tp->lost_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3539 			tp->lost_out = 0;  in tcp_clean_rtx_queue()
3541 		if (tp->sacked_out) {  in tcp_clean_rtx_queue()
3543 				 tp->sacked_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3544 			tp->sacked_out = 0;  in tcp_clean_rtx_queue()
3546 		if (tp->retrans_out) {  in tcp_clean_rtx_queue()
3548 				 tp->retrans_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3549 			tp->retrans_out = 0;  in tcp_clean_rtx_queue()
3565 	if (!after(TCP_SKB_CB(head)->end_seq, tcp_wnd_end(tp))) {  in tcp_ack_probe()
3566 		icsk->icsk_backoff = 0;  in tcp_ack_probe()
3567 		icsk->icsk_probes_tstamp = 0;  in tcp_ack_probe()
3583 		inet_csk(sk)->icsk_ca_state != TCP_CA_Open;  in tcp_ack_is_dubious()
3591 	 * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/  in tcp_may_raise_cwnd()
3595 	if (tcp_sk(sk)->reordering >  in tcp_may_raise_cwnd()
3596 	    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering))  in tcp_may_raise_cwnd()
3612 	if (icsk->icsk_ca_ops->cong_control) {  in tcp_cong_control()
3613 		icsk->icsk_ca_ops->cong_control(sk, ack, flag, rs);  in tcp_cong_control()
3619 		tcp_cwnd_reduction(sk, acked_sacked, rs->losses, flag);  in tcp_cong_control()
3634 	return	after(ack, tp->snd_una) ||  in tcp_may_update_window()
3635 		after(ack_seq, tp->snd_wl1) ||  in tcp_may_update_window()
3636 		(ack_seq == tp->snd_wl1 && (nwin > tp->snd_wnd || !nwin));  in tcp_may_update_window()
3647 	ao = rcu_dereference_protected(tp->ao_info,  in tcp_snd_sne_update()
3649 	if (ao && ack < tp->snd_una) {  in tcp_snd_sne_update()
3650 		ao->snd_sne++;  in tcp_snd_sne_update()
3651 		trace_tcp_ao_snd_sne_update((struct sock *)tp, ao->snd_sne);  in tcp_snd_sne_update()
3656 /* If we update tp->snd_una, also update tp->bytes_acked */
3659 	u32 delta = ack - tp->snd_una;  in tcp_snd_una_update()
3662 	tp->bytes_acked += delta;  in tcp_snd_una_update()
3664 	tp->snd_una = ack;  in tcp_snd_una_update()
3675 	ao = rcu_dereference_protected(tp->ao_info,  in tcp_rcv_sne_update()
3677 	if (ao && seq < tp->rcv_nxt) {  in tcp_rcv_sne_update()
3678 		ao->rcv_sne++;  in tcp_rcv_sne_update()
3679 		trace_tcp_ao_rcv_sne_update((struct sock *)tp, ao->rcv_sne);  in tcp_rcv_sne_update()
3684 /* If we update tp->rcv_nxt, also update tp->bytes_received */
3687 	u32 delta = seq - tp->rcv_nxt;  in tcp_rcv_nxt_update()
3690 	tp->bytes_received += delta;  in tcp_rcv_nxt_update()
3692 	WRITE_ONCE(tp->rcv_nxt, seq);  in tcp_rcv_nxt_update()
3697  * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
3705 	u32 nwin = ntohs(tcp_hdr(skb)->window);  in tcp_ack_update_window()
3707 	if (likely(!tcp_hdr(skb)->syn))  in tcp_ack_update_window()
3708 		nwin <<= tp->rx_opt.snd_wscale;  in tcp_ack_update_window()
3714 		if (tp->snd_wnd != nwin) {  in tcp_ack_update_window()
3715 			tp->snd_wnd = nwin;  in tcp_ack_update_window()
3720 			tp->pred_flags = 0;  in tcp_ack_update_window()
3726 			if (nwin > tp->max_window) {  in tcp_ack_update_window()
3727 				tp->max_window = nwin;  in tcp_ack_update_window()
3728 				tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie);  in tcp_ack_update_window()
3745 		s32 elapsed = (s32)(tcp_jiffies32 - val);  in __tcp_oow_rate_limited()
3748 		    elapsed < READ_ONCE(net->ipv4.sysctl_tcp_invalid_ratelimit)) {  in __tcp_oow_rate_limited()
3750 			return true;	/* rate-limited: don't send yet! */  in __tcp_oow_rate_limited()
3759 	return false;	/* not rate-limited: go ahead, send dupack now! */  in __tcp_oow_rate_limited()
3762 /* Return true if we're currently rate-limiting out-of-window ACKs and
3763  * thus shouldn't send a dupack right now. We rate-limit dupacks in
3764  * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
3767  * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
3773 	if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&  in tcp_oow_rate_limited()
3774 	    !tcp_hdr(skb)->syn)  in tcp_oow_rate_limited()
3787 	/* First check our per-socket dupack rate limit. */  in tcp_send_challenge_ack()
3790 				   &tp->last_oow_ack_time))  in tcp_send_challenge_ack()
3793 	ack_limit = READ_ONCE(net->ipv4.sysctl_tcp_challenge_ack_limit);  in tcp_send_challenge_ack()
3797 	/* Then check host-wide RFC 5961 rate limit. */  in tcp_send_challenge_ack()
3799 	if (now != READ_ONCE(net->ipv4.tcp_challenge_timestamp)) {  in tcp_send_challenge_ack()
3802 		WRITE_ONCE(net->ipv4.tcp_challenge_timestamp, now);  in tcp_send_challenge_ack()
3803 		WRITE_ONCE(net->ipv4.tcp_challenge_count,  in tcp_send_challenge_ack()
3804 			   get_random_u32_inclusive(half, ack_limit + half - 1));  in tcp_send_challenge_ack()
3806 	count = READ_ONCE(net->ipv4.tcp_challenge_count);  in tcp_send_challenge_ack()
3808 		WRITE_ONCE(net->ipv4.tcp_challenge_count, count - 1);  in tcp_send_challenge_ack()
3817 	tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval;  in tcp_store_ts_recent()
3818 	tp->rx_opt.ts_recent_stamp = ktime_get_seconds();  in tcp_store_ts_recent()
3831 	if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) {  in tcp_replace_ts_recent()
3834 		 * for pure ACK frames.  -DaveM  in tcp_replace_ts_recent()
3839 		if (tcp_paws_check(&tp->rx_opt, 0)) {  in tcp_replace_ts_recent()
3840 			delta = tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent;  in tcp_replace_ts_recent()
3849  * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack
3855 	if (before(ack, tp->tlp_high_seq))  in tcp_process_tlp_ack()
3858 	if (!tp->tlp_retrans) {  in tcp_process_tlp_ack()
3860 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3863 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3864 	} else if (after(ack, tp->tlp_high_seq)) {  in tcp_process_tlp_ack()
3877 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3885 	if (icsk->icsk_ca_ops->in_ack_event) {  in tcp_in_ack_event()
3896 		icsk->icsk_ca_ops->in_ack_event(sk, ack_ev_flags);  in tcp_in_ack_event()
3908 	if (rexmit == REXMIT_NONE || sk->sk_state == TCP_SYN_SENT)  in tcp_xmit_recovery()
3914 		if (after(tp->snd_nxt, tp->high_seq))  in tcp_xmit_recovery()
3916 		tp->frto = 0;  in tcp_xmit_recovery()
3928 	delivered = tp->delivered - prior_delivered;  in tcp_newly_delivered()
3943 	u32 prior_snd_una = tp->snd_una;  in tcp_ack()
3944 	bool is_sack_reneg = tp->is_sack_reneg;  in tcp_ack()
3945 	u32 ack_seq = TCP_SKB_CB(skb)->seq;  in tcp_ack()
3946 	u32 ack = TCP_SKB_CB(skb)->ack_seq;  in tcp_ack()
3948 	int prior_packets = tp->packets_out;  in tcp_ack()
3949 	u32 delivered = tp->delivered;  in tcp_ack()
3950 	u32 lost = tp->lost;  in tcp_ack()
3959 	prefetch(sk->tcp_rtx_queue.rb_node);  in tcp_ack()
3968 		max_window = min_t(u64, tp->max_window, tp->bytes_acked);  in tcp_ack()
3970 		if (before(ack, prior_snd_una - max_window)) {  in tcp_ack()
3973 			return -SKB_DROP_REASON_TCP_TOO_OLD_ACK;  in tcp_ack()
3981 	if (after(ack, tp->snd_nxt))  in tcp_ack()
3982 		return -SKB_DROP_REASON_TCP_ACK_UNSENT_DATA;  in tcp_ack()
3986 		icsk->icsk_retransmits = 0;  in tcp_ack()
3990 			if (tp->tcp_clean_acked)  in tcp_ack()
3991 				tp->tcp_clean_acked(sk, ack);  in tcp_ack()
3995 	prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una;  in tcp_ack()
4002 		flag |= tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq);  in tcp_ack()
4016 		if (ack_seq != TCP_SKB_CB(skb)->end_seq)  in tcp_ack()
4023 		if (TCP_SKB_CB(skb)->sacked)  in tcp_ack()
4040 	 * with widely-deployed TCP implementations that do this.  in tcp_ack()
4047 	WRITE_ONCE(sk->sk_err_soft, 0);  in tcp_ack()
4048 	icsk->icsk_probes_out = 0;  in tcp_ack()
4049 	tp->rcv_tstamp = tcp_jiffies32;  in tcp_ack()
4061 	if (tp->tlp_high_seq)  in tcp_ack()
4070 				num_dupack = max_t(u16, 1, skb_shinfo(skb)->gso_segs);  in tcp_ack()
4084 	lost = tp->lost - lost;			/* freshly marked lost */  in tcp_ack()
4105 	if (tp->tlp_high_seq)  in tcp_ack()
4113 	if (TCP_SKB_CB(skb)->sacked) {  in tcp_ack()
4129 	/* Valid only in SYN or SYN-ACK with an even length.  */  in tcp_parse_fastopen_option()
4135 		memcpy(foc->val, cookie, len);  in tcp_parse_fastopen_option()
4137 		len = -1;  in tcp_parse_fastopen_option()
4138 	foc->len = len;  in tcp_parse_fastopen_option()
4139 	foc->exp = exp_opt;  in tcp_parse_fastopen_option()
4149 		if (th->syn && !(opsize & 1) &&  in smc_parse_options()
4152 			opt_rx->smc_ok = 1;  in smc_parse_options()
4166 	int length = (th->doff * 4) - sizeof(struct tcphdr);  in tcp_parse_mss_option()
4177 			length--;  in tcp_parse_mss_option()
4196 			ptr += opsize - 2;  in tcp_parse_mss_option()
4197 			length -= opsize;  in tcp_parse_mss_option()
4214 	int length = (th->doff * 4) - sizeof(struct tcphdr);  in tcp_parse_options()
4217 	opt_rx->saw_tstamp = 0;  in tcp_parse_options()
4218 	opt_rx->saw_unknown = 0;  in tcp_parse_options()
4228 			length--;  in tcp_parse_options()
4240 				if (opsize == TCPOLEN_MSS && th->syn && !estab) {  in tcp_parse_options()
4243 						if (opt_rx->user_mss &&  in tcp_parse_options()
4244 						    opt_rx->user_mss < in_mss)  in tcp_parse_options()
4245 							in_mss = opt_rx->user_mss;  in tcp_parse_options()
4246 						opt_rx->mss_clamp = in_mss;  in tcp_parse_options()
4251 				if (opsize == TCPOLEN_WINDOW && th->syn &&  in tcp_parse_options()
4252 				    !estab && READ_ONCE(net->ipv4.sysctl_tcp_window_scaling)) {  in tcp_parse_options()
4254 					opt_rx->wscale_ok = 1;  in tcp_parse_options()
4262 					opt_rx->snd_wscale = snd_wscale;  in tcp_parse_options()
4267 				    ((estab && opt_rx->tstamp_ok) ||  in tcp_parse_options()
4268 				     (!estab && READ_ONCE(net->ipv4.sysctl_tcp_timestamps)))) {  in tcp_parse_options()
4269 					opt_rx->saw_tstamp = 1;  in tcp_parse_options()
4270 					opt_rx->rcv_tsval = get_unaligned_be32(ptr);  in tcp_parse_options()
4271 					opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4);  in tcp_parse_options()
4275 				if (opsize == TCPOLEN_SACK_PERM && th->syn &&  in tcp_parse_options()
4276 				    !estab && READ_ONCE(net->ipv4.sysctl_tcp_sack)) {  in tcp_parse_options()
4277 					opt_rx->sack_ok = TCP_SACK_SEEN;  in tcp_parse_options()
4284 				   !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) &&  in tcp_parse_options()
4285 				   opt_rx->sack_ok) {  in tcp_parse_options()
4286 					TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th;  in tcp_parse_options()
4305 					opsize - TCPOLEN_FASTOPEN_BASE,  in tcp_parse_options()
4306 					ptr, th->syn, foc, false);  in tcp_parse_options()
4316 					tcp_parse_fastopen_option(opsize -  in tcp_parse_options()
4318 						ptr + 2, th->syn, foc, true);  in tcp_parse_options()
4325 				opt_rx->saw_unknown = 1;  in tcp_parse_options()
4329 				opt_rx->saw_unknown = 1;  in tcp_parse_options()
4331 			ptr += opsize-2;  in tcp_parse_options()
4332 			length -= opsize;  in tcp_parse_options()
4344 		tp->rx_opt.saw_tstamp = 1;  in tcp_parse_aligned_timestamp()
4346 		tp->rx_opt.rcv_tsval = ntohl(*ptr);  in tcp_parse_aligned_timestamp()
4349 			tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset;  in tcp_parse_aligned_timestamp()
4351 			tp->rx_opt.rcv_tsecr = 0;  in tcp_parse_aligned_timestamp()
4367 	if (th->doff == (sizeof(*th) / 4)) {  in tcp_fast_parse_options()
4368 		tp->rx_opt.saw_tstamp = 0;  in tcp_fast_parse_options()
4370 	} else if (tp->rx_opt.tstamp_ok &&  in tcp_fast_parse_options()
4371 		   th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) {  in tcp_fast_parse_options()
4376 	tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL);  in tcp_fast_parse_options()
4377 	if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)  in tcp_fast_parse_options()
4378 		tp->rx_opt.rcv_tsecr -= tp->tsoffset;  in tcp_fast_parse_options()
4390 	int length = (th->doff << 2) - sizeof(*th);  in tcp_do_parse_auth_options()
4409 			length--;  in tcp_do_parse_auth_options()
4414 				return -EINVAL;  in tcp_do_parse_auth_options()
4417 					return -EINVAL;  in tcp_do_parse_auth_options()
4419 					return -EEXIST;  in tcp_do_parse_auth_options()
4423 					return -EINVAL;  in tcp_do_parse_auth_options()
4425 					return -EEXIST;  in tcp_do_parse_auth_options()
4429 		ptr += opsize - 2;  in tcp_do_parse_auth_options()
4430 		length -= opsize;  in tcp_do_parse_auth_options()
4437 /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
4468 	if (tcp_sk(sk)->tcp_usec_ts)  in tcp_tsval_replay()
4469 		return inet_csk(sk)->icsk_rto * (USEC_PER_SEC / HZ);  in tcp_tsval_replay()
4474 	return inet_csk(sk)->icsk_rto * 1200 / HZ;  in tcp_tsval_replay()
4483 	u32 ack = TCP_SKB_CB(skb)->ack_seq;  in tcp_disordered_ack_check()
4484 	u32 seq = TCP_SKB_CB(skb)->seq;  in tcp_disordered_ack_check()
4487 	if (!th->ack || seq != TCP_SKB_CB(skb)->end_seq)  in tcp_disordered_ack_check()
4491 	if (seq != tp->rcv_nxt)  in tcp_disordered_ack_check()
4492 		return before(seq, tp->rcv_nxt) ?  in tcp_disordered_ack_check()
4497 	if (ack != tp->snd_una)  in tcp_disordered_ack_check()
4501 	if (tcp_may_update_window(tp, ack, seq, ntohs(th->window) <<  in tcp_disordered_ack_check()
4502 						tp->rx_opt.snd_wscale))  in tcp_disordered_ack_check()
4506 	if ((s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) >  in tcp_disordered_ack_check()
4529 	if (before(end_seq, tp->rcv_wup))  in tcp_sequence()
4532 	if (after(seq, tp->rcv_nxt + tcp_receive_window(tp)))  in tcp_sequence()
4542 	WRITE_ONCE(sk->sk_err, err);  in tcp_done_with_error()
4567 	switch (sk->sk_state) {  in tcp_reset()
4587  *	If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT
4588  *	(and thence onto LAST-ACK and finally, CLOSE, we never enter
4589  *	TIME-WAIT)
4591  *	If we are in FINWAIT-1, a received FIN indicates simultaneous
4592  *	close and we go into CLOSING (and later onto TIME-WAIT)
4594  *	If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
4602 	WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);  in tcp_fin()
4605 	switch (sk->sk_state) {  in tcp_fin()
4620 		/* RFC793: Remain in the LAST-ACK state. */  in tcp_fin()
4632 		/* Received a FIN -- send ACK and enter TIME_WAIT. */  in tcp_fin()
4640 		pr_err("%s: Impossible, sk->sk_state=%d\n",  in tcp_fin()
4641 		       __func__, sk->sk_state);  in tcp_fin()
4645 	/* It _is_ possible, that we have something out-of-order _after_ FIN.  in tcp_fin()
4648 	skb_rbtree_purge(&tp->out_of_order_queue);  in tcp_fin()
4650 		tcp_sack_reset(&tp->rx_opt);  in tcp_fin()
4653 		sk->sk_state_change(sk);  in tcp_fin()
4656 		if (sk->sk_shutdown == SHUTDOWN_MASK ||  in tcp_fin()
4657 		    sk->sk_state == TCP_CLOSE)  in tcp_fin()
4667 	if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) {  in tcp_sack_extend()
4668 		if (before(seq, sp->start_seq))  in tcp_sack_extend()
4669 			sp->start_seq = seq;  in tcp_sack_extend()
4670 		if (after(end_seq, sp->end_seq))  in tcp_sack_extend()
4671 			sp->end_seq = end_seq;  in tcp_sack_extend()
4681 	if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) {  in tcp_dsack_set()
4684 		if (before(seq, tp->rcv_nxt))  in tcp_dsack_set()
4691 		tp->rx_opt.dsack = 1;  in tcp_dsack_set()
4692 		tp->duplicate_sack[0].start_seq = seq;  in tcp_dsack_set()
4693 		tp->duplicate_sack[0].end_seq = end_seq;  in tcp_dsack_set()
4701 	if (!tp->rx_opt.dsack)  in tcp_dsack_extend()
4704 		tcp_sack_extend(tp->duplicate_sack, seq, end_seq);  in tcp_dsack_extend()
4718 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss &&  in tcp_rcv_spurious_retrans()
4719 	    skb->protocol == htons(ETH_P_IPV6) &&  in tcp_rcv_spurious_retrans()
4720 	    (tcp_sk(sk)->inet_conn.icsk_ack.lrcv_flowlabel !=  in tcp_rcv_spurious_retrans()
4734 	if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_send_dupack()
4735 	    before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_send_dupack()
4739 		if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) {  in tcp_send_dupack()
4740 			u32 end_seq = TCP_SKB_CB(skb)->end_seq;  in tcp_send_dupack()
4743 			if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))  in tcp_send_dupack()
4744 				end_seq = tp->rcv_nxt;  in tcp_send_dupack()
4745 			tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq);  in tcp_send_dupack()
4752 /* These routines update the SACK block as out-of-order packets arrive or
4753  * in-order packets close up the sequence space.
4758 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_maybe_coalesce()
4764 	for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) {  in tcp_sack_maybe_coalesce()
4765 		if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) {  in tcp_sack_maybe_coalesce()
4771 			tp->rx_opt.num_sacks--;  in tcp_sack_maybe_coalesce()
4772 			for (i = this_sack; i < tp->rx_opt.num_sacks; i++)  in tcp_sack_maybe_coalesce()
4785 	if (!tp->compressed_ack)  in tcp_sack_compress_send_ack()
4788 	if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)  in tcp_sack_compress_send_ack()
4792 	 * substract one from tp->compressed_ack to keep  in tcp_sack_compress_send_ack()
4796 		      tp->compressed_ack - 1);  in tcp_sack_compress_send_ack()
4798 	tp->compressed_ack = 0;  in tcp_sack_compress_send_ack()
4811 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_new_ofo_skb()
4812 	int cur_sacks = tp->rx_opt.num_sacks;  in tcp_sack_new_ofo_skb()
4823 			for (; this_sack > 0; this_sack--, sp--)  in tcp_sack_new_ofo_skb()
4824 				swap(*sp, *(sp - 1));  in tcp_sack_new_ofo_skb()
4841 		this_sack--;  in tcp_sack_new_ofo_skb()
4842 		tp->rx_opt.num_sacks--;  in tcp_sack_new_ofo_skb()
4843 		sp--;  in tcp_sack_new_ofo_skb()
4845 	for (; this_sack > 0; this_sack--, sp--)  in tcp_sack_new_ofo_skb()
4846 		*sp = *(sp - 1);  in tcp_sack_new_ofo_skb()
4850 	sp->start_seq = seq;  in tcp_sack_new_ofo_skb()
4851 	sp->end_seq = end_seq;  in tcp_sack_new_ofo_skb()
4852 	tp->rx_opt.num_sacks++;  in tcp_sack_new_ofo_skb()
4859 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_remove()
4860 	int num_sacks = tp->rx_opt.num_sacks;  in tcp_sack_remove()
4864 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_sack_remove()
4865 		tp->rx_opt.num_sacks = 0;  in tcp_sack_remove()
4871 		if (!before(tp->rcv_nxt, sp->start_seq)) {  in tcp_sack_remove()
4875 			WARN_ON(before(tp->rcv_nxt, sp->end_seq));  in tcp_sack_remove()
4879 				tp->selective_acks[i-1] = tp->selective_acks[i];  in tcp_sack_remove()
4880 			num_sacks--;  in tcp_sack_remove()
4886 	tp->rx_opt.num_sacks = num_sacks;  in tcp_sack_remove()
4890  * tcp_try_coalesce - try to merge skb to prior one
4912 	if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq)  in tcp_try_coalesce()
4921 	atomic_add(delta, &sk->sk_rmem_alloc);  in tcp_try_coalesce()
4924 	TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq;  in tcp_try_coalesce()
4925 	TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq;  in tcp_try_coalesce()
4926 	TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags;  in tcp_try_coalesce()
4928 	if (TCP_SKB_CB(from)->has_rxtstamp) {  in tcp_try_coalesce()
4929 		TCP_SKB_CB(to)->has_rxtstamp = true;  in tcp_try_coalesce()
4930 		to->tstamp = from->tstamp;  in tcp_try_coalesce()
4931 		skb_hwtstamps(to)->hwtstamp = skb_hwtstamps(from)->hwtstamp;  in tcp_try_coalesce()
4944 	/* In case tcp_drop_reason() is called later, update to->gso_segs */  in tcp_ooo_try_coalesce()
4946 		u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) +  in tcp_ooo_try_coalesce()
4947 			       max_t(u16, 1, skb_shinfo(from)->gso_segs);  in tcp_ooo_try_coalesce()
4949 		skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF);  in tcp_ooo_try_coalesce()
4967 	__u32 dsack_high = tp->rcv_nxt;  in tcp_ofo_queue()
4972 	p = rb_first(&tp->out_of_order_queue);  in tcp_ofo_queue()
4975 		if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt))  in tcp_ofo_queue()
4978 		if (before(TCP_SKB_CB(skb)->seq, dsack_high)) {  in tcp_ofo_queue()
4980 			if (before(TCP_SKB_CB(skb)->end_seq, dsack_high))  in tcp_ofo_queue()
4981 				dsack_high = TCP_SKB_CB(skb)->end_seq;  in tcp_ofo_queue()
4982 			tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack);  in tcp_ofo_queue()
4985 		rb_erase(&skb->rbnode, &tp->out_of_order_queue);  in tcp_ofo_queue()
4987 		if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) {  in tcp_ofo_queue()
4992 		tail = skb_peek_tail(&sk->sk_receive_queue);  in tcp_ofo_queue()
4994 		tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq);  in tcp_ofo_queue()
4995 		fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;  in tcp_ofo_queue()
5003 			/* tcp_fin() purges tp->out_of_order_queue,  in tcp_ofo_queue()
5017 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||  in tcp_try_rmem_schedule()
5021 			return -1;  in tcp_try_rmem_schedule()
5025 				return -1;  in tcp_try_rmem_schedule()
5042 	if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) {  in tcp_data_queue_ofo()
5044 		sk->sk_data_ready(sk);  in tcp_data_queue_ofo()
5050 	tp->pred_flags = 0;  in tcp_data_queue_ofo()
5053 	tp->rcv_ooopack += max_t(u16, 1, skb_shinfo(skb)->gso_segs);  in tcp_data_queue_ofo()
5055 	seq = TCP_SKB_CB(skb)->seq;  in tcp_data_queue_ofo()
5056 	end_seq = TCP_SKB_CB(skb)->end_seq;  in tcp_data_queue_ofo()
5058 	p = &tp->out_of_order_queue.rb_node;  in tcp_data_queue_ofo()
5059 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_data_queue_ofo()
5062 			tp->rx_opt.num_sacks = 1;  in tcp_data_queue_ofo()
5063 			tp->selective_acks[0].start_seq = seq;  in tcp_data_queue_ofo()
5064 			tp->selective_acks[0].end_seq = end_seq;  in tcp_data_queue_ofo()
5066 		rb_link_node(&skb->rbnode, NULL, p);  in tcp_data_queue_ofo()
5067 		rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5068 		tp->ooo_last_skb = skb;  in tcp_data_queue_ofo()
5075 	if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb,  in tcp_data_queue_ofo()
5078 		/* For non sack flows, do not grow window to force DUPACK  in tcp_data_queue_ofo()
5088 	if (!before(seq, TCP_SKB_CB(tp->ooo_last_skb)->end_seq)) {  in tcp_data_queue_ofo()
5089 		parent = &tp->ooo_last_skb->rbnode;  in tcp_data_queue_ofo()
5090 		p = &parent->rb_right;  in tcp_data_queue_ofo()
5099 		if (before(seq, TCP_SKB_CB(skb1)->seq)) {  in tcp_data_queue_ofo()
5100 			p = &parent->rb_left;  in tcp_data_queue_ofo()
5103 		if (before(seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5104 			if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5114 			if (after(seq, TCP_SKB_CB(skb1)->seq)) {  in tcp_data_queue_ofo()
5116 				tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5121 				rb_replace_node(&skb1->rbnode, &skb->rbnode,  in tcp_data_queue_ofo()
5122 						&tp->out_of_order_queue);  in tcp_data_queue_ofo()
5124 						 TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5125 						 TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5136 		p = &parent->rb_right;  in tcp_data_queue_ofo()
5140 	rb_link_node(&skb->rbnode, parent, p);  in tcp_data_queue_ofo()
5141 	rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5146 		if (!after(end_seq, TCP_SKB_CB(skb1)->seq))  in tcp_data_queue_ofo()
5148 		if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5149 			tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5153 		rb_erase(&skb1->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5154 		tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5155 				 TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5161 		tp->ooo_last_skb = skb;  in tcp_data_queue_ofo()
5168 		/* For non sack flows, do not grow window to force DUPACK  in tcp_data_queue_ofo()
5182 	struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue);  in tcp_queue_rcv()
5187 	tcp_rcv_nxt_update(tcp_sk(sk), TCP_SKB_CB(skb)->end_seq);  in tcp_queue_rcv()
5198 	int err = -ENOMEM;  in tcp_send_rcvq()
5211 	skb = alloc_skb_with_frags(size - data_len, data_len,  in tcp_send_rcvq()
5213 				   &err, sk->sk_allocation);  in tcp_send_rcvq()
5217 	skb_put(skb, size - data_len);  in tcp_send_rcvq()
5218 	skb->data_len = data_len;  in tcp_send_rcvq()
5219 	skb->len = size;  in tcp_send_rcvq()
5221 	if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {  in tcp_send_rcvq()
5226 	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);  in tcp_send_rcvq()
5230 	TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt;  in tcp_send_rcvq()
5231 	TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size;  in tcp_send_rcvq()
5232 	TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1;  in tcp_send_rcvq()
5249 	if (tcp_epollin_ready(sk, sk->sk_rcvlowat) || sock_flag(sk, SOCK_DONE))  in tcp_data_ready()
5250 		sk->sk_data_ready(sk);  in tcp_data_ready()
5268 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {  in tcp_data_queue()
5273 	__skb_pull(skb, tcp_hdr(skb)->doff * 4);  in tcp_data_queue()
5276 	tp->rx_opt.dsack = 0;  in tcp_data_queue()
5282 	if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) {  in tcp_data_queue()
5290 			if (!skb->len &&  in tcp_data_queue()
5291 			    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))  in tcp_data_queue()
5301 		if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {  in tcp_data_queue()
5303 			inet_csk(sk)->icsk_ack.pending |=  in tcp_data_queue()
5306 			sk->sk_data_ready(sk);  in tcp_data_queue()
5308 			if (skb_queue_len(&sk->sk_receive_queue) && skb->len) {  in tcp_data_queue()
5313 			sk_forced_mem_schedule(sk, skb->truesize);  in tcp_data_queue()
5317 		if (skb->len)  in tcp_data_queue()
5319 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)  in tcp_data_queue()
5322 		if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_data_queue()
5328 			if (RB_EMPTY_ROOT(&tp->out_of_order_queue))  in tcp_data_queue()
5329 				inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;  in tcp_data_queue()
5332 		if (tp->rx_opt.num_sacks)  in tcp_data_queue()
5344 	if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) {  in tcp_data_queue()
5349 		tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);  in tcp_data_queue()
5360 	if (!before(TCP_SKB_CB(skb)->seq,  in tcp_data_queue()
5361 		    tp->rcv_nxt + tcp_receive_window(tp))) {  in tcp_data_queue()
5366 	if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_data_queue()
5368 		tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt);  in tcp_data_queue()
5371 		 * remembering D-SACK for its head made in previous line.  in tcp_data_queue()
5387 		return !skb_queue_is_last(list, skb) ? skb->next : NULL;  in tcp_skb_next()
5401 		rb_erase(&skb->rbnode, root);  in tcp_collapse_one()
5409 /* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */
5412 	struct rb_node **p = &root->rb_node;  in tcp_rbtree_insert()
5419 		if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq))  in tcp_rbtree_insert()
5420 			p = &parent->rb_left;  in tcp_rbtree_insert()
5422 			p = &parent->rb_right;  in tcp_rbtree_insert()
5424 	rb_link_node(&skb->rbnode, parent, p);  in tcp_rbtree_insert()
5425 	rb_insert_color(&skb->rbnode, root);  in tcp_rbtree_insert()
5455 		if (!before(start, TCP_SKB_CB(skb)->end_seq)) {  in tcp_collapse()
5463 		 * - not SYN/FIN and  in tcp_collapse()
5464 		 * - bloated or contains data before "start" or  in tcp_collapse()
5467 		if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) &&  in tcp_collapse()
5468 		    (tcp_win_from_space(sk, skb->truesize) > skb->len ||  in tcp_collapse()
5469 		     before(TCP_SKB_CB(skb)->seq, start))) {  in tcp_collapse()
5476 		    TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) {  in tcp_collapse()
5483 		start = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse()
5486 	    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) ||  in tcp_collapse()
5493 		int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start);  in tcp_collapse()
5500 		memcpy(nskb->cb, skb->cb, sizeof(skb->cb));  in tcp_collapse()
5502 		TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start;  in tcp_collapse()
5512 			int offset = start - TCP_SKB_CB(skb)->seq;  in tcp_collapse()
5513 			int size = TCP_SKB_CB(skb)->end_seq - start;  in tcp_collapse()
5520 				TCP_SKB_CB(nskb)->end_seq += size;  in tcp_collapse()
5521 				copy -= size;  in tcp_collapse()
5524 			if (!before(start, TCP_SKB_CB(skb)->end_seq)) {  in tcp_collapse()
5529 				    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) ||  in tcp_collapse()
5550 	skb = skb_rb_first(&tp->out_of_order_queue);  in tcp_collapse_ofo_queue()
5553 		tp->ooo_last_skb = skb_rb_last(&tp->out_of_order_queue);  in tcp_collapse_ofo_queue()
5556 	start = TCP_SKB_CB(skb)->seq;  in tcp_collapse_ofo_queue()
5557 	end = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse_ofo_queue()
5558 	range_truesize = skb->truesize;  in tcp_collapse_ofo_queue()
5567 		    after(TCP_SKB_CB(skb)->seq, end) ||  in tcp_collapse_ofo_queue()
5568 		    before(TCP_SKB_CB(skb)->end_seq, start)) {  in tcp_collapse_ofo_queue()
5570 			if (range_truesize != head->truesize ||  in tcp_collapse_ofo_queue()
5571 			    end - start >= SKB_WITH_OVERHEAD(PAGE_SIZE)) {  in tcp_collapse_ofo_queue()
5572 				tcp_collapse(sk, NULL, &tp->out_of_order_queue,  in tcp_collapse_ofo_queue()
5576 				if (sum_tiny > sk->sk_rcvbuf >> 3)  in tcp_collapse_ofo_queue()
5582 		range_truesize += skb->truesize;  in tcp_collapse_ofo_queue()
5583 		if (unlikely(before(TCP_SKB_CB(skb)->seq, start)))  in tcp_collapse_ofo_queue()
5584 			start = TCP_SKB_CB(skb)->seq;  in tcp_collapse_ofo_queue()
5585 		if (after(TCP_SKB_CB(skb)->end_seq, end))  in tcp_collapse_ofo_queue()
5586 			end = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse_ofo_queue()
5591  * Clean the out-of-order queue to make room.
5610 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue))  in tcp_prune_ofo_queue()
5613 	goal = sk->sk_rcvbuf >> 3;  in tcp_prune_ofo_queue()
5614 	node = &tp->ooo_last_skb->rbnode;  in tcp_prune_ofo_queue()
5620 		if (after(TCP_SKB_CB(in_skb)->seq, TCP_SKB_CB(skb)->seq))  in tcp_prune_ofo_queue()
5624 		rb_erase(node, &tp->out_of_order_queue);  in tcp_prune_ofo_queue()
5625 		goal -= skb->truesize;  in tcp_prune_ofo_queue()
5627 		tp->ooo_last_skb = rb_to_skb(prev);  in tcp_prune_ofo_queue()
5629 			if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&  in tcp_prune_ofo_queue()
5632 			goal = sk->sk_rcvbuf >> 3;  in tcp_prune_ofo_queue()
5644 		if (tp->rx_opt.sack_ok)  in tcp_prune_ofo_queue()
5645 			tcp_sack_reset(&tp->rx_opt);  in tcp_prune_ofo_queue()
5663 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)  in tcp_prune_queue()
5668 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5672 	if (!skb_queue_empty(&sk->sk_receive_queue))  in tcp_prune_queue()
5673 		tcp_collapse(sk, &sk->sk_receive_queue, NULL,  in tcp_prune_queue()
5674 			     skb_peek(&sk->sk_receive_queue),  in tcp_prune_queue()
5676 			     tp->copied_seq, tp->rcv_nxt);  in tcp_prune_queue()
5678 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5686 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5696 	tp->pred_flags = 0;  in tcp_prune_queue()
5697 	return -1;  in tcp_prune_queue()
5707 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)  in tcp_should_expand_sndbuf()
5719 			WRITE_ONCE(sk->sk_sndbuf, unused_mem);  in tcp_should_expand_sndbuf()
5741 		tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_new_space()
5744 	INDIRECT_CALL_1(sk->sk_write_space, sk_stream_write_space, sk);  in tcp_new_space()
5748  * 1) Freeing skbs in rtx queues (after tp->snd_una has advanced)
5749  * 2) Sent skbs from output queue (and thus advancing tp->snd_nxt)
5752  * 1) Space consumed in output/rtx queues is below sk->sk_sndbuf/2
5753  * 2) notsent amount (tp->write_seq - tp->snd_nxt) became
5761 	if (sk->sk_socket &&  in tcp_check_space()
5762 	    test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {  in tcp_check_space()
5764 		if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))  in tcp_check_space()
5784 	if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&  in __tcp_ack_snd_check()
5790 	    (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||  in __tcp_ack_snd_check()
5791 	     __tcp_select_window(sk) >= tp->rcv_wnd)) ||  in __tcp_ack_snd_check()
5794 	    /* Protocol state mandates a one-time immediate ACK */  in __tcp_ack_snd_check()
5795 	    inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOW) {  in __tcp_ack_snd_check()
5800 		    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_backlog_ack_defer)) {  in __tcp_ack_snd_check()
5801 			set_bit(TCP_ACK_DEFERRED, &sk->sk_tsq_flags);  in __tcp_ack_snd_check()
5809 	if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in __tcp_ack_snd_check()
5815 	    tp->compressed_ack >= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr))  in __tcp_ack_snd_check()
5818 	if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) {  in __tcp_ack_snd_check()
5819 		tp->compressed_ack_rcv_nxt = tp->rcv_nxt;  in __tcp_ack_snd_check()
5820 		tp->dup_ack_counter = 0;  in __tcp_ack_snd_check()
5822 	if (tp->dup_ack_counter < TCP_FASTRETRANS_THRESH) {  in __tcp_ack_snd_check()
5823 		tp->dup_ack_counter++;  in __tcp_ack_snd_check()
5826 	tp->compressed_ack++;  in __tcp_ack_snd_check()
5827 	if (hrtimer_is_queued(&tp->compressed_ack_timer))  in __tcp_ack_snd_check()
5832 	rtt = tp->rcv_rtt_est.rtt_us;  in __tcp_ack_snd_check()
5833 	if (tp->srtt_us && tp->srtt_us < rtt)  in __tcp_ack_snd_check()
5834 		rtt = tp->srtt_us;  in __tcp_ack_snd_check()
5837 		      READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns),  in __tcp_ack_snd_check()
5840 	hrtimer_start_range_ns(&tp->compressed_ack_timer, ns_to_ktime(delay),  in __tcp_ack_snd_check()
5841 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_slack_ns),  in __tcp_ack_snd_check()
5855  *	This routine is only called when we have urgent data
5858  *	place. We handle URGent data wrong. We have to - as
5867 	u32 ptr = ntohs(th->urg_ptr);  in tcp_check_urg()
5869 	if (ptr && !READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_stdurg))  in tcp_check_urg()
5870 		ptr--;  in tcp_check_urg()
5871 	ptr += ntohl(th->seq);  in tcp_check_urg()
5873 	/* Ignore urgent data that we've already seen and read. */  in tcp_check_urg()
5874 	if (after(tp->copied_seq, ptr))  in tcp_check_urg()
5887 	if (before(ptr, tp->rcv_nxt))  in tcp_check_urg()
5890 	/* Do we already have a newer (or duplicate) urgent pointer? */  in tcp_check_urg()
5891 	if (tp->urg_data && !after(ptr, tp->urg_seq))  in tcp_check_urg()
5894 	/* Tell the world about our new urgent pointer. */  in tcp_check_urg()
5897 	/* We may be adding urgent data when the last byte read was  in tcp_check_urg()
5898 	 * urgent. To do this requires some care. We cannot just ignore  in tcp_check_urg()
5899 	 * tp->copied_seq since we would read the last urgent byte again  in tcp_check_urg()
5912 	if (tp->urg_seq == tp->copied_seq && tp->urg_data &&  in tcp_check_urg()
5913 	    !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) {  in tcp_check_urg()
5914 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);  in tcp_check_urg()
5915 		tp->copied_seq++;  in tcp_check_urg()
5916 		if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) {  in tcp_check_urg()
5917 			__skb_unlink(skb, &sk->sk_receive_queue);  in tcp_check_urg()
5922 	WRITE_ONCE(tp->urg_data, TCP_URG_NOTYET);  in tcp_check_urg()
5923 	WRITE_ONCE(tp->urg_seq, ptr);  in tcp_check_urg()
5926 	tp->pred_flags = 0;  in tcp_check_urg()
5929 /* This is the 'fast' part of urgent handling. */
5934 	/* Check if we get a new urgent pointer - normally not. */  in tcp_urg()
5935 	if (unlikely(th->urg))  in tcp_urg()
5938 	/* Do we wait for any urgent data? - normally not... */  in tcp_urg()
5939 	if (unlikely(tp->urg_data == TCP_URG_NOTYET)) {  in tcp_urg()
5940 		u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) -  in tcp_urg()
5941 			  th->syn;  in tcp_urg()
5943 		/* Is the urgent pointer pointing into this packet? */  in tcp_urg()
5944 		if (ptr < skb->len) {  in tcp_urg()
5948 			WRITE_ONCE(tp->urg_data, TCP_URG_VALID | tmp);  in tcp_urg()
5950 				sk->sk_data_ready(sk);  in tcp_urg()
5955 /* Accept RST for rcv_nxt - 1 after a FIN.
5967 	return unlikely(TCP_SKB_CB(skb)->seq == (tp->rcv_nxt - 1) &&  in tcp_reset_check()
5968 			(1 << sk->sk_state) & (TCPF_CLOSE_WAIT | TCPF_LAST_ACK |  in tcp_reset_check()
5983 	    !tp->rx_opt.saw_tstamp ||  in tcp_validate_incoming()
5984 	    tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW))  in tcp_validate_incoming()
5991 	if (th->rst)  in tcp_validate_incoming()
5993 	if (unlikely(th->syn))  in tcp_validate_incoming()
6006 				  &tp->last_oow_ack_time))  in tcp_validate_incoming()
6012 	reason = tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);  in tcp_validate_incoming()
6014 		/* RFC793, page 37: "In all states except SYN-SENT, all reset  in tcp_validate_incoming()
6015 		 * (RST) segments are validated by checking their SEQ-fields."  in tcp_validate_incoming()
6020 		if (!th->rst) {  in tcp_validate_incoming()
6021 			if (th->syn)  in tcp_validate_incoming()
6025 						  &tp->last_oow_ack_time))  in tcp_validate_incoming()
6034 	if (th->rst) {  in tcp_validate_incoming()
6035 		/* RFC 5961 3.2 (extend to match against (RCV.NXT - 1) after a  in tcp_validate_incoming()
6037 		 * If seq num matches RCV.NXT or (RCV.NXT - 1) after a FIN, or  in tcp_validate_incoming()
6038 		 * the right-most SACK block,  in tcp_validate_incoming()
6044 		if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt ||  in tcp_validate_incoming()
6048 		if (tcp_is_sack(tp) && tp->rx_opt.num_sacks > 0) {  in tcp_validate_incoming()
6049 			struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_validate_incoming()
6053 			for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;  in tcp_validate_incoming()
6060 			if (TCP_SKB_CB(skb)->seq == max_sack)  in tcp_validate_incoming()
6064 		/* Disable TFO if RST is out-of-order  in tcp_validate_incoming()
6068 		if (tp->syn_fastopen && !tp->data_segs_in &&  in tcp_validate_incoming()
6069 		    sk->sk_state == TCP_ESTABLISHED)  in tcp_validate_incoming()
6081 	if (th->syn) {  in tcp_validate_incoming()
6082 		if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack &&  in tcp_validate_incoming()
6083 		    TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq &&  in tcp_validate_incoming()
6084 		    TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt &&  in tcp_validate_incoming()
6085 		    TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt)  in tcp_validate_incoming()
6116  *	- A zero window was announced from us - zero window probing
6118  *	- Out of order segments arrived.
6119  *	- Urgent data is expected.
6120  *	- There is no buffer space left
6121  *	- Unexpected TCP flags/window values/header lengths are received
6123  *	- Data is sent in both directions. Fast path only supports pure senders
6126  *	- Unexpected TCP option.
6137 	const struct tcphdr *th = (const struct tcphdr *)skb->data;  in tcp_rcv_established()
6139 	unsigned int len = skb->len;  in tcp_rcv_established()
6145 	if (unlikely(!rcu_access_pointer(sk->sk_rx_dst)))  in tcp_rcv_established()
6146 		inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb);  in tcp_rcv_established()
6162 	tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_established()
6166 	 *	'S' will always be tp->tcp_header_len >> 2  in tcp_rcv_established()
6173 	if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags &&  in tcp_rcv_established()
6174 	    TCP_SKB_CB(skb)->seq == tp->rcv_nxt &&  in tcp_rcv_established()
6175 	    !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {  in tcp_rcv_established()
6176 		int tcp_header_len = tp->tcp_header_len;  in tcp_rcv_established()
6181 		 * is automatically equal to th->doff*4 due to pred_flags  in tcp_rcv_established()
6191 			delta = tp->rx_opt.rcv_tsval -  in tcp_rcv_established()
6192 				tp->rx_opt.ts_recent;  in tcp_rcv_established()
6213 				    tp->rcv_nxt == tp->rcv_wup)  in tcp_rcv_established()
6227 				tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;  in tcp_rcv_established()
6241 			if ((int)skb->truesize > sk->sk_forward_alloc)  in tcp_rcv_established()
6250 			    tp->rcv_nxt == tp->rcv_wup)  in tcp_rcv_established()
6265 			if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) {  in tcp_rcv_established()
6272 				tcp_update_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_established()
6285 	if (len < (th->doff << 2) || tcp_checksum_complete(skb))  in tcp_rcv_established()
6288 	if (!th->ack && !th->rst && !th->syn) {  in tcp_rcv_established()
6303 		reason = -reason;  in tcp_rcv_established()
6308 	/* Process urgent data. */  in tcp_rcv_established()
6335 	icsk->icsk_af_ops->rebuild_header(sk);  in tcp_init_transfer()
6339 	 * Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been  in tcp_init_transfer()
6341 	 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK  in tcp_init_transfer()
6344 	if (tp->total_retrans > 1 && tp->undo_marker)  in tcp_init_transfer()
6348 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_init_transfer()
6352 	if (!icsk->icsk_ca_initialized)  in tcp_init_transfer()
6364 	icsk->icsk_ack.lrcvtime = tcp_jiffies32;  in tcp_finish_connect()
6367 		icsk->icsk_af_ops->sk_rx_dst_set(sk, skb);  in tcp_finish_connect()
6377 	tp->lsndtime = tcp_jiffies32;  in tcp_finish_connect()
6382 	if (!tp->rx_opt.snd_wscale)  in tcp_finish_connect()
6383 		__tcp_fast_path_on(tp, tp->snd_wnd);  in tcp_finish_connect()
6385 		tp->pred_flags = 0;  in tcp_finish_connect()
6392 	struct sk_buff *data = tp->syn_data ? tcp_rtx_queue_head(sk) : NULL;  in tcp_rcv_fastopen_synack()
6393 	u16 mss = tp->rx_opt.mss_clamp, try_exp = 0;  in tcp_rcv_fastopen_synack()
6396 	if (mss == tp->rx_opt.user_mss) {  in tcp_rcv_fastopen_synack()
6406 	if (!tp->syn_fastopen) {  in tcp_rcv_fastopen_synack()
6408 		cookie->len = -1;  in tcp_rcv_fastopen_synack()
6409 	} else if (tp->total_retrans) {  in tcp_rcv_fastopen_synack()
6410 		/* SYN timed out and the SYN-ACK neither has a cookie nor  in tcp_rcv_fastopen_synack()
6413 		 * SYN-data or the corresponding SYN-ACK was dropped.  in tcp_rcv_fastopen_synack()
6415 		syn_drop = (cookie->len < 0 && data);  in tcp_rcv_fastopen_synack()
6416 	} else if (cookie->len < 0 && !tp->syn_data) {  in tcp_rcv_fastopen_synack()
6421 		try_exp = tp->syn_fastopen_exp ? 2 : 1;  in tcp_rcv_fastopen_synack()
6427 		if (tp->total_retrans)  in tcp_rcv_fastopen_synack()
6428 			tp->fastopen_client_fail = TFO_SYN_RETRANSMITTED;  in tcp_rcv_fastopen_synack()
6430 			tp->fastopen_client_fail = TFO_DATA_NOT_ACKED;  in tcp_rcv_fastopen_synack()
6438 	tp->syn_data_acked = tp->syn_data;  in tcp_rcv_fastopen_synack()
6439 	if (tp->syn_data_acked) {  in tcp_rcv_fastopen_synack()
6441 		/* SYN-data is counted as two separate packets in tcp_ack() */  in tcp_rcv_fastopen_synack()
6442 		if (tp->delivered > 1)  in tcp_rcv_fastopen_synack()
6443 			--tp->delivered;  in tcp_rcv_fastopen_synack()
6455 		if (tp->syn_smc && !tp->rx_opt.smc_ok)  in smc_check_reset_syn()
6456 			tp->syn_smc = 0;  in smc_check_reset_syn()
6470 	syn_stamp = tp->retrans_stamp;  in tcp_try_undo_spurious_syn()
6471 	if (tp->undo_marker && syn_stamp && tp->rx_opt.saw_tstamp &&  in tcp_try_undo_spurious_syn()
6472 	    syn_stamp == tp->rx_opt.rcv_tsecr)  in tcp_try_undo_spurious_syn()
6473 		tp->undo_marker = 0;  in tcp_try_undo_spurious_syn()
6481 	struct tcp_fastopen_cookie foc = { .len = -1 };  in tcp_rcv_synsent_state_process()
6482 	int saved_clamp = tp->rx_opt.mss_clamp;  in tcp_rcv_synsent_state_process()
6486 	tcp_parse_options(sock_net(sk), skb, &tp->rx_opt, 0, &foc);  in tcp_rcv_synsent_state_process()
6487 	if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)  in tcp_rcv_synsent_state_process()
6488 		tp->rx_opt.rcv_tsecr -= tp->tsoffset;  in tcp_rcv_synsent_state_process()
6490 	if (th->ack) {  in tcp_rcv_synsent_state_process()
6492 		 * "If the state is SYN-SENT then  in tcp_rcv_synsent_state_process()
6499 		if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) ||  in tcp_rcv_synsent_state_process()
6500 		    after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {  in tcp_rcv_synsent_state_process()
6502 			if (icsk->icsk_retransmits == 0)  in tcp_rcv_synsent_state_process()
6509 		if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&  in tcp_rcv_synsent_state_process()
6510 		    !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp,  in tcp_rcv_synsent_state_process()
6526 		if (th->rst) {  in tcp_rcv_synsent_state_process()
6538 		 *                                        --ANK(990513)  in tcp_rcv_synsent_state_process()
6540 		if (!th->syn) {  in tcp_rcv_synsent_state_process()
6553 		tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_synsent_state_process()
6560 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);  in tcp_rcv_synsent_state_process()
6561 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;  in tcp_rcv_synsent_state_process()
6566 		tp->snd_wnd = ntohs(th->window);  in tcp_rcv_synsent_state_process()
6568 		if (!tp->rx_opt.wscale_ok) {  in tcp_rcv_synsent_state_process()
6569 			tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0;  in tcp_rcv_synsent_state_process()
6570 			WRITE_ONCE(tp->window_clamp,  in tcp_rcv_synsent_state_process()
6571 				   min(tp->window_clamp, 65535U));  in tcp_rcv_synsent_state_process()
6574 		if (tp->rx_opt.saw_tstamp) {  in tcp_rcv_synsent_state_process()
6575 			tp->rx_opt.tstamp_ok	   = 1;  in tcp_rcv_synsent_state_process()
6576 			tp->tcp_header_len =  in tcp_rcv_synsent_state_process()
6578 			tp->advmss	    -= TCPOLEN_TSTAMP_ALIGNED;  in tcp_rcv_synsent_state_process()
6581 			tp->tcp_header_len = sizeof(struct tcphdr);  in tcp_rcv_synsent_state_process()
6584 		tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_rcv_synsent_state_process()
6588 		 * Change state from SYN-SENT only after copied_seq  in tcp_rcv_synsent_state_process()
6590 		WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_synsent_state_process()
6598 		fastopen_fail = (tp->syn_fastopen || tp->syn_data) &&  in tcp_rcv_synsent_state_process()
6602 			sk->sk_state_change(sk);  in tcp_rcv_synsent_state_process()
6606 			return -1;  in tcp_rcv_synsent_state_process()
6607 		if (sk->sk_write_pending ||  in tcp_rcv_synsent_state_process()
6608 		    READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept) ||  in tcp_rcv_synsent_state_process()
6615 			 * to stand against the temptation 8)     --ANK  in tcp_rcv_synsent_state_process()
6624 		return -1;  in tcp_rcv_synsent_state_process()
6629 	if (th->rst) {  in tcp_rcv_synsent_state_process()
6640 	if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp &&  in tcp_rcv_synsent_state_process()
6641 	    tcp_paws_reject(&tp->rx_opt, 0)) {  in tcp_rcv_synsent_state_process()
6645 	if (th->syn) {  in tcp_rcv_synsent_state_process()
6653 		ao = rcu_dereference_protected(tp->ao_info,  in tcp_rcv_synsent_state_process()
6656 			WRITE_ONCE(ao->risn, th->seq);  in tcp_rcv_synsent_state_process()
6657 			ao->rcv_sne = 0;  in tcp_rcv_synsent_state_process()
6662 		if (tp->rx_opt.saw_tstamp) {  in tcp_rcv_synsent_state_process()
6663 			tp->rx_opt.tstamp_ok = 1;  in tcp_rcv_synsent_state_process()
6665 			tp->tcp_header_len =  in tcp_rcv_synsent_state_process()
6668 			tp->tcp_header_len = sizeof(struct tcphdr);  in tcp_rcv_synsent_state_process()
6671 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);  in tcp_rcv_synsent_state_process()
6672 		WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_synsent_state_process()
6673 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;  in tcp_rcv_synsent_state_process()
6678 		tp->snd_wnd    = ntohs(th->window);  in tcp_rcv_synsent_state_process()
6679 		tp->snd_wl1    = TCP_SKB_CB(skb)->seq;  in tcp_rcv_synsent_state_process()
6680 		tp->max_window = tp->snd_wnd;  in tcp_rcv_synsent_state_process()
6685 		tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_rcv_synsent_state_process()
6701 		return -1;  in tcp_rcv_synsent_state_process()
6711 	tcp_clear_options(&tp->rx_opt);  in tcp_rcv_synsent_state_process()
6712 	tp->rx_opt.mss_clamp = saved_clamp;  in tcp_rcv_synsent_state_process()
6717 	tcp_clear_options(&tp->rx_opt);  in tcp_rcv_synsent_state_process()
6718 	tp->rx_opt.mss_clamp = saved_clamp;  in tcp_rcv_synsent_state_process()
6731 	if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss && !tp->packets_out)  in tcp_rcv_synrecv_state_fastopen()
6735 	inet_csk(sk)->icsk_retransmits = 0;  in tcp_rcv_synrecv_state_fastopen()
6749 	req = rcu_dereference_protected(tp->fastopen_rsk,  in tcp_rcv_synrecv_state_fastopen()
6753 	/* Re-arm the timer because data may have been sent out.  in tcp_rcv_synrecv_state_fastopen()
6757 	 * (TFO) - we could try to be more aggressive and  in tcp_rcv_synrecv_state_fastopen()
6781 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6787 		if (th->ack)  in tcp_rcv_state_process()
6790 		if (th->rst) {  in tcp_rcv_state_process()
6794 		if (th->syn) {  in tcp_rcv_state_process()
6795 			if (th->fin) {  in tcp_rcv_state_process()
6804 			icsk->icsk_af_ops->conn_request(sk, skb);  in tcp_rcv_state_process()
6815 		tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_state_process()
6829 	tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_state_process()
6830 	req = rcu_dereference_protected(tp->fastopen_rsk,  in tcp_rcv_state_process()
6835 		WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&  in tcp_rcv_state_process()
6836 		    sk->sk_state != TCP_FIN_WAIT1);  in tcp_rcv_state_process()
6843 	if (!th->ack && !th->rst && !th->syn) {  in tcp_rcv_state_process()
6856 		if (sk->sk_state == TCP_SYN_RECV) {  in tcp_rcv_state_process()
6860 			return -reason;  in tcp_rcv_state_process()
6865 			reason = -reason;  in tcp_rcv_state_process()
6870 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6872 		tp->delivered++; /* SYN-ACK delivery isn't tracked in tcp_ack */  in tcp_rcv_state_process()
6873 		if (!tp->srtt_us)  in tcp_rcv_state_process()
6880 			tp->retrans_stamp = 0;  in tcp_rcv_state_process()
6883 			WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_state_process()
6888 		sk->sk_state_change(sk);  in tcp_rcv_state_process()
6892 		 * sk->sk_sleep == NULL and sk->sk_socket == NULL.  in tcp_rcv_state_process()
6894 		if (sk->sk_socket)  in tcp_rcv_state_process()
6897 		tp->snd_una = TCP_SKB_CB(skb)->ack_seq;  in tcp_rcv_state_process()
6898 		tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale;  in tcp_rcv_state_process()
6899 		tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_state_process()
6901 		if (tp->rx_opt.tstamp_ok)  in tcp_rcv_state_process()
6902 			tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;  in tcp_rcv_state_process()
6904 		if (!inet_csk(sk)->icsk_ca_ops->cong_control)  in tcp_rcv_state_process()
6908 		tp->lsndtime = tcp_jiffies32;  in tcp_rcv_state_process()
6912 		if (sk->sk_shutdown & SEND_SHUTDOWN)  in tcp_rcv_state_process()
6922 		if (tp->snd_una != tp->write_seq)  in tcp_rcv_state_process()
6926 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | SEND_SHUTDOWN);  in tcp_rcv_state_process()
6932 			sk->sk_state_change(sk);  in tcp_rcv_state_process()
6936 		if (READ_ONCE(tp->linger2) < 0) {  in tcp_rcv_state_process()
6941 		if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_rcv_state_process()
6942 		    after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {  in tcp_rcv_state_process()
6944 			if (tp->syn_fastopen && th->fin)  in tcp_rcv_state_process()
6953 			tcp_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN);  in tcp_rcv_state_process()
6954 		} else if (th->fin || sock_owned_by_user(sk)) {  in tcp_rcv_state_process()
6970 		if (tp->snd_una == tp->write_seq) {  in tcp_rcv_state_process()
6977 		if (tp->snd_una == tp->write_seq) {  in tcp_rcv_state_process()
6989 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6993 		if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_rcv_state_process()
7008 		if (sk->sk_shutdown & RCV_SHUTDOWN) {  in tcp_rcv_state_process()
7009 			if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_rcv_state_process()
7010 			    after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {  in tcp_rcv_state_process()
7023 	/* tcp_data could move socket to TIME-WAIT */  in tcp_rcv_state_process()
7024 	if (sk->sk_state != TCP_CLOSE) {  in tcp_rcv_state_process()
7047 				    &ireq->ir_rmt_addr, port);  in pr_drop_req()
7051 				    &ireq->ir_v6_rmt_addr, port);  in pr_drop_req()
7067  * Exception: At least one of the reserved bits of the TCP header (th->res1) is
7079 	bool th_ecn = th->ece && th->cwr;  in tcp_ecn_create_request()
7086 	ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield);  in tcp_ecn_create_request()
7088 	ecn_ok = READ_ONCE(net->ipv4.sysctl_tcp_ecn) || ecn_ok_dst;  in tcp_ecn_create_request()
7090 	if (((!ect || th->res1) && ecn_ok) || tcp_ca_needs_ecn(listen_sk) ||  in tcp_ecn_create_request()
7093 		inet_rsk(req)->ecn_ok = 1;  in tcp_ecn_create_request()
7102 	req->rsk_rcv_wnd = 0;		/* So that tcp_send_synack() knows! */  in tcp_openreq_init()
7103 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;  in tcp_openreq_init()
7104 	tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;  in tcp_openreq_init()
7105 	tcp_rsk(req)->snt_synack = 0;  in tcp_openreq_init()
7106 	tcp_rsk(req)->snt_tsval_first = 0;  in tcp_openreq_init()
7107 	tcp_rsk(req)->last_oow_ack_time = 0;  in tcp_openreq_init()
7108 	req->mss = rx_opt->mss_clamp;  in tcp_openreq_init()
7109 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;  in tcp_openreq_init()
7110 	ireq->tstamp_ok = rx_opt->tstamp_ok;  in tcp_openreq_init()
7111 	ireq->sack_ok = rx_opt->sack_ok;  in tcp_openreq_init()
7112 	ireq->snd_wscale = rx_opt->snd_wscale;  in tcp_openreq_init()
7113 	ireq->wscale_ok = rx_opt->wscale_ok;  in tcp_openreq_init()
7114 	ireq->acked = 0;  in tcp_openreq_init()
7115 	ireq->ecn_ok = 0;  in tcp_openreq_init()
7116 	ireq->ir_rmt_port = tcp_hdr(skb)->source;  in tcp_openreq_init()
7117 	ireq->ir_num = ntohs(tcp_hdr(skb)->dest);  in tcp_openreq_init()
7118 	ireq->ir_mark = inet_request_mark(sk, skb);  in tcp_openreq_init()
7120 	ireq->smc_ok = rx_opt->smc_ok && !(tcp_sk(sk)->smc_hs_congested &&  in tcp_openreq_init()
7121 			tcp_sk(sk)->smc_hs_congested(sk));  in tcp_openreq_init()
7130 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;  in tcp_syn_flood_action()
7136 	syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies);  in tcp_syn_flood_action()
7147 	if (!READ_ONCE(queue->synflood_warned) && syncookies != 2 &&  in tcp_syn_flood_action()
7148 	    xchg(&queue->synflood_warned, 1) == 0) {  in tcp_syn_flood_action()
7149 		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_family == AF_INET6) {  in tcp_syn_flood_action()
7152 					sk->sk_num, msg);  in tcp_syn_flood_action()
7155 					proto, &sk->sk_rcv_saddr,  in tcp_syn_flood_action()
7156 					sk->sk_num, msg);  in tcp_syn_flood_action()
7167 	if (tcp_sk(sk)->save_syn) {  in tcp_reqsk_record_syn()
7173 		if (tcp_sk(sk)->save_syn == 2) {  /* Save full header. */  in tcp_reqsk_record_syn()
7185 			saved_syn->mac_hdrlen = mac_hdrlen;  in tcp_reqsk_record_syn()
7186 			saved_syn->network_hdrlen = skb_network_header_len(skb);  in tcp_reqsk_record_syn()
7187 			saved_syn->tcp_hdrlen = tcp_hdrlen(skb);  in tcp_reqsk_record_syn()
7188 			memcpy(saved_syn->data, base, len);  in tcp_reqsk_record_syn()
7189 			req->saved_syn = saved_syn;  in tcp_reqsk_record_syn()
7204 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) != 2 &&  in tcp_get_syncookie_mss()
7208 	if (!tcp_syn_flood_action(sk, rsk_ops->slab_name))  in tcp_get_syncookie_mss()
7216 	mss = tcp_parse_mss_option(th, tp->rx_opt.user_mss);  in tcp_get_syncookie_mss()
7218 		mss = af_ops->mss_clamp;  in tcp_get_syncookie_mss()
7228 	struct tcp_fastopen_cookie foc = { .len = -1 };  in tcp_conn_request()
7252 		syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies);  in tcp_conn_request()
7256 							   rsk_ops->slab_name);  in tcp_conn_request()
7271 	req->syncookie = want_cookie;  in tcp_conn_request()
7272 	tcp_rsk(req)->af_specific = af_ops;  in tcp_conn_request()
7273 	tcp_rsk(req)->ts_off = 0;  in tcp_conn_request()
7274 	tcp_rsk(req)->req_usec_ts = false;  in tcp_conn_request()
7276 	tcp_rsk(req)->is_mptcp = 0;  in tcp_conn_request()
7280 	tmp_opt.mss_clamp = af_ops->mss_clamp;  in tcp_conn_request()
7281 	tmp_opt.user_mss  = tp->rx_opt.user_mss;  in tcp_conn_request()
7293 	inet_rsk(req)->no_srccheck = inet_test_bit(TRANSPARENT, sk);  in tcp_conn_request()
7296 	inet_rsk(req)->ir_iif = inet_request_bound_dev_if(sk, skb);  in tcp_conn_request()
7298 	dst = af_ops->route_req(sk, skb, &fl, req, isn);  in tcp_conn_request()
7303 		tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst);  in tcp_conn_request()
7304 		tcp_rsk(req)->ts_off = af_ops->init_ts_off(net, skb);  in tcp_conn_request()
7307 		int max_syn_backlog = READ_ONCE(net->ipv4.sysctl_max_syn_backlog);  in tcp_conn_request()
7311 		    (max_syn_backlog - inet_csk_reqsk_queue_len(sk) <  in tcp_conn_request()
7321 			pr_drop_req(req, ntohs(tcp_hdr(skb)->source),  in tcp_conn_request()
7322 				    rsk_ops->family);  in tcp_conn_request()
7326 		isn = af_ops->init_seq(skb);  in tcp_conn_request()
7332 		isn = cookie_init_sequence(af_ops, sk, skb, &req->mss);  in tcp_conn_request()
7334 			inet_rsk(req)->ecn_ok = 0;  in tcp_conn_request()
7341 		tcp_rsk(req)->used_tcp_ao = true;  in tcp_conn_request()
7342 		tcp_rsk(req)->ao_rcv_next = aoh->keyid;  in tcp_conn_request()
7343 		tcp_rsk(req)->ao_keyid = aoh->rnext_keyid;  in tcp_conn_request()
7346 		tcp_rsk(req)->used_tcp_ao = false;  in tcp_conn_request()
7349 	tcp_rsk(req)->snt_isn = isn;  in tcp_conn_request()
7350 	tcp_rsk(req)->txhash = net_tx_rndhash();  in tcp_conn_request()
7351 	tcp_rsk(req)->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;  in tcp_conn_request()
7359 		af_ops->send_synack(fastopen_sk, dst, &fl, req,  in tcp_conn_request()
7368 		sk->sk_data_ready(sk);  in tcp_conn_request()
7372 		tcp_rsk(req)->tfo_listener = false;  in tcp_conn_request()
7374 			req->timeout = tcp_timeout_init((struct sock *)req);  in tcp_conn_request()
7376 								    req->timeout))) {  in tcp_conn_request()
7383 		af_ops->send_synack(sk, dst, &fl, req, &foc,  in tcp_conn_request()