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 */
116 #define REXMIT_NEW	2 /* FRTO-style transmit of unsent/new packets */
124 	icsk->icsk_clean_acked = cad;  in clean_acked_data_enable()
132 	icsk->icsk_clean_acked = NULL;  in clean_acked_data_disable()
146 	bool unknown_opt = tcp_sk(sk)->rx_opt.saw_unknown &&  in bpf_skops_parse_hdr()
160 	switch (sk->sk_state) {  in bpf_skops_parse_hdr()
212 	dev = dev_get_by_index_rcu(sock_net(sk), skb->skb_iif);  in tcp_gro_dev_warn()
213 	if (!dev || len >= READ_ONCE(dev->mtu))  in tcp_gro_dev_warn()
215 			dev ? dev->name : "Unknown driver");  in tcp_gro_dev_warn()
219 /* Adapt the MSS value used to make delayed ack decision to the
225 	const unsigned int lss = icsk->icsk_ack.last_seg_size;  in tcp_measure_rcv_mss()
228 	icsk->icsk_ack.last_seg_size = 0;  in tcp_measure_rcv_mss()
230 	/* skb->len may jitter because of SACKs, even if peer  in tcp_measure_rcv_mss()
231 	 * sends good full-sized frames.  in tcp_measure_rcv_mss()
233 	len = skb_shinfo(skb)->gso_size ? : skb->len;  in tcp_measure_rcv_mss()
234 	if (len >= icsk->icsk_ack.rcv_mss) {  in tcp_measure_rcv_mss()
239 		if (unlikely(len != icsk->icsk_ack.rcv_mss)) {  in tcp_measure_rcv_mss()
240 			u64 val = (u64)skb->len << TCP_RMEM_TO_WIN_SCALE;  in tcp_measure_rcv_mss()
241 			u8 old_ratio = tcp_sk(sk)->scaling_ratio;  in tcp_measure_rcv_mss()
243 			do_div(val, skb->truesize);  in tcp_measure_rcv_mss()
244 			tcp_sk(sk)->scaling_ratio = val ? val : 1;  in tcp_measure_rcv_mss()
246 			if (old_ratio != tcp_sk(sk)->scaling_ratio)  in tcp_measure_rcv_mss()
247 				WRITE_ONCE(tcp_sk(sk)->window_clamp,  in tcp_measure_rcv_mss()
248 					   tcp_win_from_space(sk, sk->sk_rcvbuf));  in tcp_measure_rcv_mss()
250 		icsk->icsk_ack.rcv_mss = min_t(unsigned int, len,  in tcp_measure_rcv_mss()
251 					       tcp_sk(sk)->advmss);  in tcp_measure_rcv_mss()
252 		/* Account for possibly-removed options */  in tcp_measure_rcv_mss()
253 		DO_ONCE_LITE_IF(len > icsk->icsk_ack.rcv_mss + MAX_TCP_OPTION_SPACE,  in tcp_measure_rcv_mss()
258 		 * may be waiting for an ACK if cwnd-bound or using TX zero  in tcp_measure_rcv_mss()
261 		 * reads all of the data and is not ping-pong. If len > MSS  in tcp_measure_rcv_mss()
266 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_PSH)  in tcp_measure_rcv_mss()
267 			icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;  in tcp_measure_rcv_mss()
274 		len += skb->data - skb_transport_header(skb);  in tcp_measure_rcv_mss()
279 		     * to handle super-low mtu links fairly.  in tcp_measure_rcv_mss()
287 			len -= tcp_sk(sk)->tcp_header_len;  in tcp_measure_rcv_mss()
288 			icsk->icsk_ack.last_seg_size = len;  in tcp_measure_rcv_mss()
290 				icsk->icsk_ack.rcv_mss = len;  in tcp_measure_rcv_mss()
294 		if (icsk->icsk_ack.pending & ICSK_ACK_PUSHED)  in tcp_measure_rcv_mss()
295 			icsk->icsk_ack.pending |= ICSK_ACK_PUSHED2;  in tcp_measure_rcv_mss()
296 		icsk->icsk_ack.pending |= ICSK_ACK_PUSHED;  in tcp_measure_rcv_mss()
303 	unsigned int quickacks = tcp_sk(sk)->rcv_wnd / (2 * icsk->icsk_ack.rcv_mss);  in tcp_incr_quickack()
308 	if (quickacks > icsk->icsk_ack.quick)  in tcp_incr_quickack()
309 		icsk->icsk_ack.quick = quickacks;  in tcp_incr_quickack()
318 	icsk->icsk_ack.ato = TCP_ATO_MIN;  in tcp_enter_quickack_mode()
331 		(icsk->icsk_ack.quick && !inet_csk_in_pingpong_mode(sk));  in tcp_in_quickack_mode()
336 	if (tp->ecn_flags & TCP_ECN_OK)  in tcp_ecn_queue_cwr()
337 		tp->ecn_flags |= TCP_ECN_QUEUE_CWR;  in tcp_ecn_queue_cwr()
342 	if (tcp_hdr(skb)->cwr) {  in tcp_ecn_accept_cwr()
343 		tcp_sk(sk)->ecn_flags &= ~TCP_ECN_DEMAND_CWR;  in tcp_ecn_accept_cwr()
349 		if (TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq)  in tcp_ecn_accept_cwr()
350 			inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;  in tcp_ecn_accept_cwr()
356 	tp->ecn_flags &= ~TCP_ECN_QUEUE_CWR;  in tcp_ecn_withdraw_cwr()
363 	switch (TCP_SKB_CB(skb)->ip_dsfield & INET_ECN_MASK) {  in __tcp_ecn_check_ce()
369 		if (tp->ecn_flags & TCP_ECN_SEEN)  in __tcp_ecn_check_ce()
376 		if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) {  in __tcp_ecn_check_ce()
379 			tp->ecn_flags |= TCP_ECN_DEMAND_CWR;  in __tcp_ecn_check_ce()
381 		tp->ecn_flags |= TCP_ECN_SEEN;  in __tcp_ecn_check_ce()
386 		tp->ecn_flags |= TCP_ECN_SEEN;  in __tcp_ecn_check_ce()
393 	if (tcp_sk(sk)->ecn_flags & TCP_ECN_OK)  in tcp_ecn_check_ce()
399 	if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || th->cwr))  in tcp_ecn_rcv_synack()
400 		tp->ecn_flags &= ~TCP_ECN_OK;  in tcp_ecn_rcv_synack()
405 	if ((tp->ecn_flags & TCP_ECN_OK) && (!th->ece || !th->cwr))  in tcp_ecn_rcv_syn()
406 		tp->ecn_flags &= ~TCP_ECN_OK;  in tcp_ecn_rcv_syn()
411 	if (th->ece && !th->syn && (tp->ecn_flags & TCP_ECN_OK))  in tcp_ecn_rcv_ecn_echo()
418  * 1. Tuning sk->sk_sndbuf, when connection enters established state.
424 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;  in tcp_sndbuf_expand()
429 	 * and skb->head is kmalloced using power of two area of memory  in tcp_sndbuf_expand()
431 	per_mss = max_t(u32, tp->rx_opt.mss_clamp, tp->mss_cache) +  in tcp_sndbuf_expand()
439 	nr_segs = max_t(u32, nr_segs, tp->reordering + 1);  in tcp_sndbuf_expand()
442 	 * Cubic needs 1.7 factor, rounded to 2 to include  in tcp_sndbuf_expand()
445 	sndmem = ca_ops->sndbuf_expand ? ca_ops->sndbuf_expand(sk) : 2;  in tcp_sndbuf_expand()
448 	if (sk->sk_sndbuf < sndmem)  in tcp_sndbuf_expand()
449 		WRITE_ONCE(sk->sk_sndbuf,  in tcp_sndbuf_expand()
450 			   min(sndmem, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_wmem[2])));  in tcp_sndbuf_expand()
456  * forward and advertised in receiver window (tp->rcv_wnd) and
460  * tcp_full_space(), in this case tcp_full_space() - window_clamp
468  * - to enforce header prediction at sender, even when application
470  * - to prevent pruning of receive queue because of misprediction
485 	int window = tcp_win_from_space(sk, READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2])) >> 1;  in __tcp_grow_window()
487 	while (tp->rcv_ssthresh <= window) {  in __tcp_grow_window()
488 		if (truesize <= skb->len)  in __tcp_grow_window()
489 			return 2 * inet_csk(sk)->icsk_ack.rcv_mss;  in __tcp_grow_window()
498  * can play nice with us, as sk_buff and skb->head might be either
501  * and if no payload was pulled in skb->head before reaching us.
505 	u32 truesize = skb->truesize;  in truesize_adjust()
508 		truesize -= SKB_TRUESIZE(skb_end_offset(skb));  in truesize_adjust()
510 		if (unlikely((int)truesize < (int)skb->len))  in truesize_adjust()
511 			truesize = skb->truesize;  in truesize_adjust()
522 	room = min_t(int, tp->window_clamp, tcp_space(sk)) - tp->rcv_ssthresh;  in tcp_grow_window()
535 		if (tcp_win_from_space(sk, truesize) <= skb->len)  in tcp_grow_window()
536 			incr = 2 * tp->advmss;  in tcp_grow_window()
541 			incr = max_t(int, incr, 2 * skb->len);  in tcp_grow_window()
542 			tp->rcv_ssthresh += min(room, incr);  in tcp_grow_window()
543 			inet_csk(sk)->icsk_ack.quick |= 1;  in tcp_grow_window()
558 	int tcp_app_win = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_app_win);  in tcp_init_buffer_space()
562 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK))  in tcp_init_buffer_space()
566 	tp->rcvq_space.time = tp->tcp_mstamp;  in tcp_init_buffer_space()
567 	tp->rcvq_space.seq = tp->copied_seq;  in tcp_init_buffer_space()
571 	if (tp->window_clamp >= maxwin) {  in tcp_init_buffer_space()
572 		WRITE_ONCE(tp->window_clamp, maxwin);  in tcp_init_buffer_space()
574 		if (tcp_app_win && maxwin > 4 * tp->advmss)  in tcp_init_buffer_space()
575 			WRITE_ONCE(tp->window_clamp,  in tcp_init_buffer_space()
576 				   max(maxwin - (maxwin >> tcp_app_win),  in tcp_init_buffer_space()
577 				       4 * tp->advmss));  in tcp_init_buffer_space()
582 	    tp->window_clamp > 2 * tp->advmss &&  in tcp_init_buffer_space()
583 	    tp->window_clamp + tp->advmss > maxwin)  in tcp_init_buffer_space()
584 		WRITE_ONCE(tp->window_clamp,  in tcp_init_buffer_space()
585 			   max(2 * tp->advmss, maxwin - tp->advmss));  in tcp_init_buffer_space()
587 	tp->rcv_ssthresh = min(tp->rcv_ssthresh, tp->window_clamp);  in tcp_init_buffer_space()
588 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_init_buffer_space()
589 	tp->rcvq_space.space = min3(tp->rcv_ssthresh, tp->rcv_wnd,  in tcp_init_buffer_space()
590 				    (u32)TCP_INIT_CWND * tp->advmss);  in tcp_init_buffer_space()
601 	icsk->icsk_ack.quick = 0;  in tcp_clamp_window()
602 	rmem2 = READ_ONCE(net->ipv4.sysctl_tcp_rmem[2]);  in tcp_clamp_window()
604 	if (sk->sk_rcvbuf < rmem2 &&  in tcp_clamp_window()
605 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK) &&  in tcp_clamp_window()
608 		WRITE_ONCE(sk->sk_rcvbuf,  in tcp_clamp_window()
609 			   min(atomic_read(&sk->sk_rmem_alloc), rmem2));  in tcp_clamp_window()
611 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf)  in tcp_clamp_window()
612 		tp->rcv_ssthresh = min(tp->window_clamp, 2U * tp->advmss);  in tcp_clamp_window()
615 /* Initialize RCV_MSS value.
625 	unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);  in tcp_initialize_rcv_mss()
627 	hint = min(hint, tp->rcv_wnd / 2);  in tcp_initialize_rcv_mss()
631 	inet_csk(sk)->icsk_ack.rcv_mss = hint;  in tcp_initialize_rcv_mss()
638  * Dynamic Right-Sizing (DRS) by Wu Feng and Mike Fisk of LANL.
648 	u32 new_sample = tp->rcv_rtt_est.rtt_us;  in tcp_rcv_rtt_update()
652 		/* If we sample in larger samples in the non-timestamp  in tcp_rcv_rtt_update()
658 		 * non-timestamp case, we do not smooth things out  in tcp_rcv_rtt_update()
663 			m -= (new_sample >> 3);  in tcp_rcv_rtt_update()
675 	tp->rcv_rtt_est.rtt_us = new_sample;  in tcp_rcv_rtt_update()
682 	if (tp->rcv_rtt_est.time == 0)  in tcp_rcv_rtt_measure()
684 	if (before(tp->rcv_nxt, tp->rcv_rtt_est.seq))  in tcp_rcv_rtt_measure()
686 	delta_us = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcv_rtt_est.time);  in tcp_rcv_rtt_measure()
692 	tp->rcv_rtt_est.seq = tp->rcv_nxt + tp->rcv_wnd;  in tcp_rcv_rtt_measure()
693 	tp->rcv_rtt_est.time = tp->tcp_mstamp;  in tcp_rcv_rtt_measure()
700 	delta = tcp_time_stamp_ts(tp) - tp->rx_opt.rcv_tsecr;  in tcp_rtt_tsopt_us()
701 	if (tp->tcp_usec_ts)  in tcp_rtt_tsopt_us()
710 	return -1;  in tcp_rtt_tsopt_us()
718 	if (tp->rx_opt.rcv_tsecr == tp->rcv_rtt_last_tsecr)  in tcp_rcv_rtt_measure_ts()
720 	tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;  in tcp_rcv_rtt_measure_ts()
722 	if (TCP_SKB_CB(skb)->end_seq -  in tcp_rcv_rtt_measure_ts()
723 	    TCP_SKB_CB(skb)->seq >= inet_csk(sk)->icsk_ack.rcv_mss) {  in tcp_rcv_rtt_measure_ts()
744 	time = tcp_stamp_us_delta(tp->tcp_mstamp, tp->rcvq_space.time);  in tcp_rcv_space_adjust()
745 	if (time < (tp->rcv_rtt_est.rtt_us >> 3) || tp->rcv_rtt_est.rtt_us == 0)  in tcp_rcv_space_adjust()
749 	copied = tp->copied_seq - tp->rcvq_space.seq;  in tcp_rcv_space_adjust()
750 	if (copied <= tp->rcvq_space.space)  in tcp_rcv_space_adjust()
755 	 * To cope with packet losses, we need a 2x factor  in tcp_rcv_space_adjust()
757 	 * every RTT, we need a 4x factor, because the ACK we are sending  in tcp_rcv_space_adjust()
762 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_moderate_rcvbuf) &&  in tcp_rcv_space_adjust()
763 	    !(sk->sk_userlocks & SOCK_RCVBUF_LOCK)) {  in tcp_rcv_space_adjust()
770 		rcvwin = ((u64)copied << 1) + 16 * tp->advmss;  in tcp_rcv_space_adjust()
773 		grow = rcvwin * (copied - tp->rcvq_space.space);  in tcp_rcv_space_adjust()
774 		do_div(grow, tp->rcvq_space.space);  in tcp_rcv_space_adjust()
778 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rmem[2]));  in tcp_rcv_space_adjust()
779 		if (rcvbuf > sk->sk_rcvbuf) {  in tcp_rcv_space_adjust()
780 			WRITE_ONCE(sk->sk_rcvbuf, rcvbuf);  in tcp_rcv_space_adjust()
783 			WRITE_ONCE(tp->window_clamp,  in tcp_rcv_space_adjust()
787 	tp->rcvq_space.space = copied;  in tcp_rcv_space_adjust()
790 	tp->rcvq_space.seq = tp->copied_seq;  in tcp_rcv_space_adjust()
791 	tp->rcvq_space.time = tp->tcp_mstamp;  in tcp_rcv_space_adjust()
799 	if (skb->protocol == htons(ETH_P_IPV6))  in tcp_save_lrcv_flowlabel()
800 		icsk->icsk_ack.lrcv_flowlabel = ntohl(ip6_flowlabel(ipv6_hdr(skb)));  in tcp_save_lrcv_flowlabel()
805  * behavior of the tp->ato delayed ack timeout interval.  When a
812  * queue.  -DaveM
828 	if (!icsk->icsk_ack.ato) {  in tcp_event_data_recv()
833 		icsk->icsk_ack.ato = TCP_ATO_MIN;  in tcp_event_data_recv()
835 		int m = now - icsk->icsk_ack.lrcvtime;  in tcp_event_data_recv()
839 			icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + TCP_ATO_MIN / 2;  in tcp_event_data_recv()
840 		} else if (m < icsk->icsk_ack.ato) {  in tcp_event_data_recv()
841 			icsk->icsk_ack.ato = (icsk->icsk_ack.ato >> 1) + m;  in tcp_event_data_recv()
842 			if (icsk->icsk_ack.ato > icsk->icsk_rto)  in tcp_event_data_recv()
843 				icsk->icsk_ack.ato = icsk->icsk_rto;  in tcp_event_data_recv()
844 		} else if (m > icsk->icsk_rto) {  in tcp_event_data_recv()
851 	icsk->icsk_ack.lrcvtime = now;  in tcp_event_data_recv()
856 	if (skb->len >= 128)  in tcp_event_data_recv()
867  * it up into three procedures. -- erics
873 	u32 srtt = tp->srtt_us;  in tcp_rtt_estimator()
881 	 *	On a 1990 paper the rto value is changed to:  in tcp_rtt_estimator()
887 	 * etc. I guess in BSD RTO takes ONE value, so that it is absolutely  in tcp_rtt_estimator()
892 		m -= (srtt >> 3);	/* m is now error in rtt est */  in tcp_rtt_estimator()
895 			m = -m;		/* m is now abs(error) */  in tcp_rtt_estimator()
896 			m -= (tp->mdev_us >> 2);   /* similar update on mdev */  in tcp_rtt_estimator()
908 			m -= (tp->mdev_us >> 2);   /* similar update on mdev */  in tcp_rtt_estimator()
910 		tp->mdev_us += m;		/* mdev = 3/4 mdev + 1/4 new */  in tcp_rtt_estimator()
911 		if (tp->mdev_us > tp->mdev_max_us) {  in tcp_rtt_estimator()
912 			tp->mdev_max_us = tp->mdev_us;  in tcp_rtt_estimator()
913 			if (tp->mdev_max_us > tp->rttvar_us)  in tcp_rtt_estimator()
914 				tp->rttvar_us = tp->mdev_max_us;  in tcp_rtt_estimator()
916 		if (after(tp->snd_una, tp->rtt_seq)) {  in tcp_rtt_estimator()
917 			if (tp->mdev_max_us < tp->rttvar_us)  in tcp_rtt_estimator()
918 				tp->rttvar_us -= (tp->rttvar_us - tp->mdev_max_us) >> 2;  in tcp_rtt_estimator()
919 			tp->rtt_seq = tp->snd_nxt;  in tcp_rtt_estimator()
920 			tp->mdev_max_us = tcp_rto_min_us(sk);  in tcp_rtt_estimator()
927 		tp->mdev_us = m << 1;	/* make sure rto = 3*rtt */  in tcp_rtt_estimator()
928 		tp->rttvar_us = max(tp->mdev_us, tcp_rto_min_us(sk));  in tcp_rtt_estimator()
929 		tp->mdev_max_us = tp->rttvar_us;  in tcp_rtt_estimator()
930 		tp->rtt_seq = tp->snd_nxt;  in tcp_rtt_estimator()
934 	tp->srtt_us = max(1U, srtt);  in tcp_rtt_estimator()
943 	rate = (u64)tp->mss_cache * ((USEC_PER_SEC / 100) << 3);  in tcp_update_pacing_rate()
949 	 * [1] : Normal Slow Start condition is (tp->snd_cwnd < tp->snd_ssthresh)  in tcp_update_pacing_rate()
950 	 *	 If snd_cwnd >= (tp->snd_ssthresh / 2), we are approaching  in tcp_update_pacing_rate()
953 	if (tcp_snd_cwnd(tp) < tp->snd_ssthresh / 2)  in tcp_update_pacing_rate()
954 		rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ss_ratio);  in tcp_update_pacing_rate()
956 		rate *= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pacing_ca_ratio);  in tcp_update_pacing_rate()
958 	rate *= max(tcp_snd_cwnd(tp), tp->packets_out);  in tcp_update_pacing_rate()
960 	if (likely(tp->srtt_us))  in tcp_update_pacing_rate()
961 		do_div(rate, tp->srtt_us);  in tcp_update_pacing_rate()
967 	WRITE_ONCE(sk->sk_pacing_rate,  in tcp_update_pacing_rate()
968 		   min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate)));  in tcp_update_pacing_rate()
984 	 *    is invisible. Actually, Linux-2.4 also generates erratic  in tcp_set_rto()
987 	inet_csk(sk)->icsk_rto = __tcp_set_rto(tp);  in tcp_set_rto()
1007 	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);  in tcp_init_cwnd()
1011 	/* Timestamps for earliest and latest never-retransmitted segment
1024 /* Take a notice that peer is sending D-SACKs. Skip update of data delivery
1027  * - DSACKed sequence range is larger than maximum receiver's window.
1028  * - Total no. of DSACKed segments exceed the total no. of retransmitted segs.
1038 	seq_len = end_seq - start_seq;  in tcp_dsack_seen()
1040 	if (seq_len > tp->max_window)  in tcp_dsack_seen()
1042 	if (seq_len > tp->mss_cache)  in tcp_dsack_seen()
1043 		dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache);  in tcp_dsack_seen()
1044 	else if (tp->tlp_high_seq && tp->tlp_high_seq == end_seq)  in tcp_dsack_seen()
1045 		state->flag |= FLAG_DSACK_TLP;  in tcp_dsack_seen()
1047 	tp->dsack_dups += dup_segs;  in tcp_dsack_seen()
1049 	if (tp->dsack_dups > tp->total_retrans)  in tcp_dsack_seen()
1052 	tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;  in tcp_dsack_seen()
1057 	 * is timer-driven, not triggered by RACK).  in tcp_dsack_seen()
1059 	if (tp->reord_seen && !(state->flag & FLAG_DSACK_TLP))  in tcp_dsack_seen()
1060 		tp->rack.dsack_seen = 1;  in tcp_dsack_seen()
1062 	state->flag |= FLAG_DSACKING_ACK;  in tcp_dsack_seen()
1064 	state->sack_delivered += dup_segs;  in tcp_dsack_seen()
1070  * some lower never-retransmitted sequence ("low_seq"). The maximum reordering
1071  * distance is approximated in full-mss packet distance ("reordering").
1077 	const u32 mss = tp->mss_cache;  in tcp_check_sack_reordering()
1084 	metric = fack - low_seq;  in tcp_check_sack_reordering()
1085 	if ((metric > tp->reordering * mss) && mss) {  in tcp_check_sack_reordering()
1088 			 tp->rx_opt.sack_ok, inet_csk(sk)->icsk_ca_state,  in tcp_check_sack_reordering()
1089 			 tp->reordering,  in tcp_check_sack_reordering()
1091 			 tp->sacked_out,  in tcp_check_sack_reordering()
1092 			 tp->undo_marker ? tp->undo_retrans : 0);  in tcp_check_sack_reordering()
1094 		tp->reordering = min_t(u32, (metric + mss - 1) / mss,  in tcp_check_sack_reordering()
1095 				       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering));  in tcp_check_sack_reordering()
1099 	tp->reord_seen++;  in tcp_check_sack_reordering()
1111 	if ((!tp->retransmit_skb_hint && tp->retrans_out >= tp->lost_out) ||  in tcp_verify_retransmit_hint()
1112 	    (tp->retransmit_skb_hint &&  in tcp_verify_retransmit_hint()
1113 	     before(TCP_SKB_CB(skb)->seq,  in tcp_verify_retransmit_hint()
1114 		    TCP_SKB_CB(tp->retransmit_skb_hint)->seq)))  in tcp_verify_retransmit_hint()
1115 		tp->retransmit_skb_hint = skb;  in tcp_verify_retransmit_hint()
1123 	tp->lost += tcp_skb_pcount(skb);  in tcp_notify_skb_loss_event()
1128 	__u8 sacked = TCP_SKB_CB(skb)->sacked;  in tcp_mark_skb_lost()
1138 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_RETRANS;  in tcp_mark_skb_lost()
1139 			tp->retrans_out -= tcp_skb_pcount(skb);  in tcp_mark_skb_lost()
1145 		tp->lost_out += tcp_skb_pcount(skb);  in tcp_mark_skb_lost()
1146 		TCP_SKB_CB(skb)->sacked |= TCPCB_LOST;  in tcp_mark_skb_lost()
1155 	tp->delivered += delivered;  in tcp_count_delivered()
1157 		tp->delivered_ce += delivered;  in tcp_count_delivered()
1168  * 0	1		- orig segment is in flight.
1169  * S	0		- nothing flies, orig reached receiver.
1170  * L	0		- nothing flies, orig lost by net.
1171  * R	2		- both orig and retransmit are in flight.
1172  * L|R	1		- orig is lost, retransmit is in flight.
1173  * S|R  1		- orig reached receiver, retrans is still in flight.
1175  *  but it is equivalent to plain S and code short-circuits it to S.
1176  *  L|S is logically invalid, it would mean -1 packet in flight 8))
1186  * 4. D-SACK added new rule: D-SACK changes any tag to S.
1193  * --------------------
1198  *    ever retransmitted -> reordering. Alas, we cannot use it
1200  * 2. The last flaw is solved with D-SACK. D-SACK arrives
1201  *    for retransmitted and already SACKed segment -> reordering..
1206  * ----------------------
1221  * Implements also blockage to start_seq wrap-around. Problem lies in the
1227  *         <- outs wnd ->                          <- wrapzone ->
1230  * |<------------+------+----- TCP seqno space --------------+---------->|
1231  * ...-- <2^31 ->|                                           |<--------...
1232  * ...---- >2^31 ------>|                                    |<--------...
1237  * snd_nxt wrap -> snd_una region will then become "well defined", i.e.,
1238  * equal to the ideal case (infinite seqno space without wrap caused issues).
1240  * With D-SACK the lower bound is extended to cover sequence space below
1242  * again, D-SACK block must not to go across snd_una (for the same reason as
1244  * ends, TCP might receive valid D-SACKs below that. As long as they reside
1247  * theoretical ones), the D-SACK will nicely cross that boundary due to skb
1250  * the exact amount is rather hard to quantify. However, tp->max_window can
1257 	if (after(end_seq, tp->snd_nxt) || !before(start_seq, end_seq))  in tcp_is_sackblock_valid()
1260 	/* Nasty start_seq wrap-around check (see comments above) */  in tcp_is_sackblock_valid()
1261 	if (!before(start_seq, tp->snd_nxt))  in tcp_is_sackblock_valid()
1264 	/* In outstanding window? ...This is valid exit for D-SACKs too.  in tcp_is_sackblock_valid()
1265 	 * start_seq == snd_una is non-sensical (see comments above)  in tcp_is_sackblock_valid()
1267 	if (after(start_seq, tp->snd_una))  in tcp_is_sackblock_valid()
1270 	if (!is_dsack || !tp->undo_marker)  in tcp_is_sackblock_valid()
1273 	/* ...Then it's D-SACK, and must reside below snd_una completely */  in tcp_is_sackblock_valid()
1274 	if (after(end_seq, tp->snd_una))  in tcp_is_sackblock_valid()
1277 	if (!before(start_seq, tp->undo_marker))  in tcp_is_sackblock_valid()
1281 	if (!after(end_seq, tp->undo_marker))  in tcp_is_sackblock_valid()
1287 	return !before(start_seq, end_seq - tp->max_window);  in tcp_is_sackblock_valid()
1299 	if (before(start_seq_0, TCP_SKB_CB(ack_skb)->ack_seq)) {  in tcp_check_dsack()
1320 	/* D-SACK for already forgotten data... Do dumb counting. */  in tcp_check_dsack()
1321 	if (tp->undo_marker && tp->undo_retrans > 0 &&  in tcp_check_dsack()
1323 	    after(end_seq_0, tp->undo_marker))  in tcp_check_dsack()
1324 		tp->undo_retrans = max_t(int, 0, tp->undo_retrans - dup_segs);  in tcp_check_dsack()
1345 	in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&  in tcp_match_skb_to_sack()
1346 		  !before(end_seq, TCP_SKB_CB(skb)->end_seq);  in tcp_match_skb_to_sack()
1349 	    after(TCP_SKB_CB(skb)->end_seq, start_seq)) {  in tcp_match_skb_to_sack()
1351 		in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);  in tcp_match_skb_to_sack()
1354 			pkt_len = start_seq - TCP_SKB_CB(skb)->seq;  in tcp_match_skb_to_sack()
1358 			pkt_len = end_seq - TCP_SKB_CB(skb)->seq;  in tcp_match_skb_to_sack()
1360 				return -EINVAL;  in tcp_match_skb_to_sack()
1373 		if (pkt_len >= skb->len && !in_sack)  in tcp_match_skb_to_sack()
1385 /* Mark the given newly-SACKed range as such, adjusting counters and hints. */
1394 	/* Account D-SACK for retransmitted packet. */  in tcp_sacktag_one()
1396 		if (tp->undo_marker && tp->undo_retrans > 0 &&  in tcp_sacktag_one()
1397 		    after(end_seq, tp->undo_marker))  in tcp_sacktag_one()
1398 			tp->undo_retrans = max_t(int, 0, tp->undo_retrans - pcount);  in tcp_sacktag_one()
1400 		    before(start_seq, state->reord))  in tcp_sacktag_one()
1401 				state->reord = start_seq;  in tcp_sacktag_one()
1405 	if (!after(end_seq, tp->snd_una))  in tcp_sacktag_one()
1418 				tp->lost_out -= pcount;  in tcp_sacktag_one()
1419 				tp->retrans_out -= pcount;  in tcp_sacktag_one()
1428 				    before(start_seq, state->reord))  in tcp_sacktag_one()
1429 					state->reord = start_seq;  in tcp_sacktag_one()
1431 				if (!after(end_seq, tp->high_seq))  in tcp_sacktag_one()
1432 					state->flag |= FLAG_ORIG_SACK_ACKED;  in tcp_sacktag_one()
1433 				if (state->first_sackt == 0)  in tcp_sacktag_one()
1434 					state->first_sackt = xmit_time;  in tcp_sacktag_one()
1435 				state->last_sackt = xmit_time;  in tcp_sacktag_one()
1440 				tp->lost_out -= pcount;  in tcp_sacktag_one()
1445 		state->flag |= FLAG_DATA_SACKED;  in tcp_sacktag_one()
1446 		tp->sacked_out += pcount;  in tcp_sacktag_one()
1447 		/* Out-of-order packets delivered */  in tcp_sacktag_one()
1448 		state->sack_delivered += pcount;  in tcp_sacktag_one()
1451 		if (tp->lost_skb_hint &&  in tcp_sacktag_one()
1452 		    before(start_seq, TCP_SKB_CB(tp->lost_skb_hint)->seq))  in tcp_sacktag_one()
1453 			tp->lost_cnt_hint += pcount;  in tcp_sacktag_one()
1456 	/* D-SACK. We can detect redundant retransmission in S|R and plain R  in tcp_sacktag_one()
1462 		tp->retrans_out -= pcount;  in tcp_sacktag_one()
1468 /* Shift newly-SACKed bytes from this skb to the immediately previous
1469  * already-SACKed sk_buff. Mark the newly-SACKed bytes as such.
1478 	u32 start_seq = TCP_SKB_CB(skb)->seq;	/* start of newly-SACKed */  in tcp_shifted_skb()
1479 	u32 end_seq = start_seq + shifted;	/* end of newly-SACKed */  in tcp_shifted_skb()
1484 	 * range but discard the return value since prev is already  in tcp_shifted_skb()
1489 	tcp_sacktag_one(sk, state, TCP_SKB_CB(skb)->sacked,  in tcp_shifted_skb()
1492 	tcp_rate_skb_delivered(sk, skb, state->rate);  in tcp_shifted_skb()
1494 	if (skb == tp->lost_skb_hint)  in tcp_shifted_skb()
1495 		tp->lost_cnt_hint += pcount;  in tcp_shifted_skb()
1497 	TCP_SKB_CB(prev)->end_seq += shifted;  in tcp_shifted_skb()
1498 	TCP_SKB_CB(skb)->seq += shifted;  in tcp_shifted_skb()
1502 	tcp_skb_pcount_add(skb, -pcount);  in tcp_shifted_skb()
1509 	if (!TCP_SKB_CB(prev)->tcp_gso_size)  in tcp_shifted_skb()
1510 		TCP_SKB_CB(prev)->tcp_gso_size = mss;  in tcp_shifted_skb()
1514 		TCP_SKB_CB(skb)->tcp_gso_size = 0;  in tcp_shifted_skb()
1517 	TCP_SKB_CB(prev)->sacked |= (TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS);  in tcp_shifted_skb()
1519 	if (skb->len > 0) {  in tcp_shifted_skb()
1525 	/* Whole SKB was eaten :-) */  in tcp_shifted_skb()
1527 	if (skb == tp->retransmit_skb_hint)  in tcp_shifted_skb()
1528 		tp->retransmit_skb_hint = prev;  in tcp_shifted_skb()
1529 	if (skb == tp->lost_skb_hint) {  in tcp_shifted_skb()
1530 		tp->lost_skb_hint = prev;  in tcp_shifted_skb()
1531 		tp->lost_cnt_hint -= tcp_skb_pcount(prev);  in tcp_shifted_skb()
1534 	TCP_SKB_CB(prev)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;  in tcp_shifted_skb()
1535 	TCP_SKB_CB(prev)->eor = TCP_SKB_CB(skb)->eor;  in tcp_shifted_skb()
1536 	if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)  in tcp_shifted_skb()
1537 		TCP_SKB_CB(prev)->end_seq++;  in tcp_shifted_skb()
1543 	if (unlikely(TCP_SKB_CB(prev)->tx.delivered_mstamp))  in tcp_shifted_skb()
1544 		TCP_SKB_CB(prev)->tx.delivered_mstamp = 0;  in tcp_shifted_skb()
1554  * something-or-zero which complicates things
1558 	return tcp_skb_pcount(skb) == 1 ? skb->len : tcp_skb_mss(skb);  in tcp_skb_seglen()
1571 	 * Since TCP_SKB_CB(skb)->tcp_gso_segs is 16 bits, we need  in tcp_skb_shift()
1575 	if (unlikely(to->len + shiftlen >= 65535 * TCP_MIN_GSO_SIZE))  in tcp_skb_shift()
1599 	    (TCP_SKB_CB(skb)->sacked & (TCPCB_LOST|TCPCB_SACKED_RETRANS)) == TCPCB_SACKED_RETRANS)  in tcp_shift_skb_data()
1604 	if (!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una))  in tcp_shift_skb_data()
1612 	if ((TCP_SKB_CB(prev)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED)  in tcp_shift_skb_data()
1618 	in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq) &&  in tcp_shift_skb_data()
1619 		  !before(end_seq, TCP_SKB_CB(skb)->end_seq);  in tcp_shift_skb_data()
1622 		len = skb->len;  in tcp_shift_skb_data()
1632 		if (!after(TCP_SKB_CB(skb)->end_seq, start_seq))  in tcp_shift_skb_data()
1634 		/* CHECKME: This is non-MSS split case only?, this will  in tcp_shift_skb_data()
1641 		in_sack = !after(start_seq, TCP_SKB_CB(skb)->seq);  in tcp_shift_skb_data()
1644 			 * if (!after(TCP_SKB_CB(skb)->end_seq, end_seq)),  in tcp_shift_skb_data()
1648 			 * previously. We could try merging non-SACKed ones  in tcp_shift_skb_data()
1657 		len = end_seq - TCP_SKB_CB(skb)->seq;  in tcp_shift_skb_data()
1659 		BUG_ON(len > skb->len);  in tcp_shift_skb_data()
1683 	/* tcp_sacktag_one() won't SACK-tag ranges below snd_una */  in tcp_shift_skb_data()
1684 	if (!after(TCP_SKB_CB(skb)->seq + len, tp->snd_una))  in tcp_shift_skb_data()
1700 	    ((TCP_SKB_CB(skb)->sacked & TCPCB_TAGBITS) != TCPCB_SACKED_ACKED) ||  in tcp_shift_skb_data()
1706 	len = skb->len;  in tcp_shift_skb_data()
1736 		/* queue is in-order => we can short-circuit the walk early */  in tcp_sacktag_walk()
1737 		if (!before(TCP_SKB_CB(skb)->seq, end_seq))  in tcp_sacktag_walk()
1741 		    before(TCP_SKB_CB(skb)->seq, next_dup->end_seq)) {  in tcp_sacktag_walk()
1743 							next_dup->start_seq,  in tcp_sacktag_walk()
1744 							next_dup->end_seq);  in tcp_sacktag_walk()
1774 			TCP_SKB_CB(skb)->sacked =  in tcp_sacktag_walk()
1777 						TCP_SKB_CB(skb)->sacked,  in tcp_sacktag_walk()
1778 						TCP_SKB_CB(skb)->seq,  in tcp_sacktag_walk()
1779 						TCP_SKB_CB(skb)->end_seq,  in tcp_sacktag_walk()
1783 			tcp_rate_skb_delivered(sk, skb, state->rate);  in tcp_sacktag_walk()
1784 			if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_sacktag_walk()
1785 				list_del_init(&skb->tcp_tsorted_anchor);  in tcp_sacktag_walk()
1787 			if (!before(TCP_SKB_CB(skb)->seq,  in tcp_sacktag_walk()
1797 	struct rb_node *parent, **p = &sk->tcp_rtx_queue.rb_node;  in tcp_sacktag_bsearch()
1803 		if (before(seq, TCP_SKB_CB(skb)->seq)) {  in tcp_sacktag_bsearch()
1804 			p = &parent->rb_left;  in tcp_sacktag_bsearch()
1807 		if (!before(seq, TCP_SKB_CB(skb)->end_seq)) {  in tcp_sacktag_bsearch()
1808 			p = &parent->rb_right;  in tcp_sacktag_bsearch()
1819 	if (skb && after(TCP_SKB_CB(skb)->seq, skip_to_seq))  in tcp_sacktag_skip()
1834 	if (before(next_dup->start_seq, skip_to_seq)) {  in tcp_maybe_skipping_dsack()
1835 		skb = tcp_sacktag_skip(skb, sk, next_dup->start_seq);  in tcp_maybe_skipping_dsack()
1837 				       next_dup->start_seq, next_dup->end_seq,  in tcp_maybe_skipping_dsack()
1846 	return cache < tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);  in tcp_sack_cache_ok()
1855 				    TCP_SKB_CB(ack_skb)->sacked);  in tcp_sacktag_write_queue()
1860 	int num_sacks = min(TCP_NUM_SACKS, (ptr[1] - TCPOLEN_SACK_BASE) >> 3);  in tcp_sacktag_write_queue()
1866 	state->flag = 0;  in tcp_sacktag_write_queue()
1867 	state->reord = tp->snd_nxt;  in tcp_sacktag_write_queue()
1869 	if (!tp->sacked_out)  in tcp_sacktag_write_queue()
1879 	if (before(TCP_SKB_CB(ack_skb)->ack_seq, prior_snd_una - tp->max_window))  in tcp_sacktag_write_queue()
1882 	if (!tp->packets_out)  in tcp_sacktag_write_queue()
1899 				if (!tp->undo_marker)  in tcp_sacktag_write_queue()
1905 				if ((TCP_SKB_CB(ack_skb)->ack_seq != tp->snd_una) &&  in tcp_sacktag_write_queue()
1906 				    !after(sp[used_sacks].end_seq, tp->snd_una))  in tcp_sacktag_write_queue()
1913 				first_sack_index = -1;  in tcp_sacktag_write_queue()
1920 				first_sack_index = -1;  in tcp_sacktag_write_queue()
1928 	for (i = used_sacks - 1; i > 0; i--) {  in tcp_sacktag_write_queue()
1940 	state->mss_now = tcp_current_mss(sk);  in tcp_sacktag_write_queue()
1944 	if (!tp->sacked_out) {  in tcp_sacktag_write_queue()
1946 		cache = tp->recv_sack_cache + ARRAY_SIZE(tp->recv_sack_cache);  in tcp_sacktag_write_queue()
1948 		cache = tp->recv_sack_cache;  in tcp_sacktag_write_queue()
1950 		while (tcp_sack_cache_ok(tp, cache) && !cache->start_seq &&  in tcp_sacktag_write_queue()
1951 		       !cache->end_seq)  in tcp_sacktag_write_queue()
1966 		       !before(start_seq, cache->end_seq))  in tcp_sacktag_write_queue()
1971 		    after(end_seq, cache->start_seq)) {  in tcp_sacktag_write_queue()
1974 			if (before(start_seq, cache->start_seq)) {  in tcp_sacktag_write_queue()
1979 						       cache->start_seq,  in tcp_sacktag_write_queue()
1984 			if (!after(end_seq, cache->end_seq))  in tcp_sacktag_write_queue()
1989 						       cache->end_seq);  in tcp_sacktag_write_queue()
1992 			if (tcp_highest_sack_seq(tp) == cache->end_seq) {  in tcp_sacktag_write_queue()
2001 			skb = tcp_sacktag_skip(skb, sk, cache->end_seq);  in tcp_sacktag_write_queue()
2023 	for (i = 0; i < ARRAY_SIZE(tp->recv_sack_cache) - used_sacks; i++) {  in tcp_sacktag_write_queue()
2024 		tp->recv_sack_cache[i].start_seq = 0;  in tcp_sacktag_write_queue()
2025 		tp->recv_sack_cache[i].end_seq = 0;  in tcp_sacktag_write_queue()
2028 		tp->recv_sack_cache[i++] = sp[j];  in tcp_sacktag_write_queue()
2030 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss || tp->undo_marker)  in tcp_sacktag_write_queue()
2031 		tcp_check_sack_reordering(sk, state->reord, 0);  in tcp_sacktag_write_queue()
2037 	WARN_ON((int)tp->sacked_out < 0);  in tcp_sacktag_write_queue()
2038 	WARN_ON((int)tp->lost_out < 0);  in tcp_sacktag_write_queue()
2039 	WARN_ON((int)tp->retrans_out < 0);  in tcp_sacktag_write_queue()
2042 	return state->flag;  in tcp_sacktag_write_queue()
2052 	holes = max(tp->lost_out, 1U);  in tcp_limit_reno_sacked()
2053 	holes = min(holes, tp->packets_out);  in tcp_limit_reno_sacked()
2055 	if ((tp->sacked_out + holes) > tp->packets_out) {  in tcp_limit_reno_sacked()
2056 		tp->sacked_out = tp->packets_out - holes;  in tcp_limit_reno_sacked()
2073 	tp->reordering = min_t(u32, tp->packets_out + addend,  in tcp_check_reno_reordering()
2074 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_max_reordering));  in tcp_check_reno_reordering()
2075 	tp->reord_seen++;  in tcp_check_reno_reordering()
2085 		u32 prior_sacked = tp->sacked_out;  in tcp_add_reno_sack()
2088 		tp->sacked_out += num_dupack;  in tcp_add_reno_sack()
2090 		delivered = tp->sacked_out - prior_sacked;  in tcp_add_reno_sack()
2105 		tcp_count_delivered(tp, max_t(int, acked - tp->sacked_out, 1),  in tcp_remove_reno_sacks()
2107 		if (acked - 1 >= tp->sacked_out)  in tcp_remove_reno_sacks()
2108 			tp->sacked_out = 0;  in tcp_remove_reno_sacks()
2110 			tp->sacked_out -= acked - 1;  in tcp_remove_reno_sacks()
2118 	tp->sacked_out = 0;  in tcp_reset_reno_sack()
2123 	tp->retrans_out = 0;  in tcp_clear_retrans()
2124 	tp->lost_out = 0;  in tcp_clear_retrans()
2125 	tp->undo_marker = 0;  in tcp_clear_retrans()
2126 	tp->undo_retrans = -1;  in tcp_clear_retrans()
2127 	tp->sacked_out = 0;  in tcp_clear_retrans()
2128 	tp->rto_stamp = 0;  in tcp_clear_retrans()
2129 	tp->total_rto = 0;  in tcp_clear_retrans()
2130 	tp->total_rto_recoveries = 0;  in tcp_clear_retrans()
2131 	tp->total_rto_time = 0;  in tcp_clear_retrans()
2136 	tp->undo_marker = tp->snd_una;  in tcp_init_undo()
2140 	tp->undo_retrans = tp->retrans_out;  in tcp_init_undo()
2142 	if (tp->tlp_high_seq && tp->tlp_retrans)  in tcp_init_undo()
2143 		tp->undo_retrans++;  in tcp_init_undo()
2145 	if (!tp->undo_retrans)  in tcp_init_undo()
2146 		tp->undo_retrans = -1;  in tcp_init_undo()
2151 	return READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_recovery) &  in tcp_is_rack()
2166 	is_reneg = head && (TCP_SKB_CB(head)->sacked & TCPCB_SACKED_ACKED);  in tcp_timeout_mark_lost()
2169 		tp->sacked_out = 0;  in tcp_timeout_mark_lost()
2171 		tp->is_sack_reneg = 1;  in tcp_timeout_mark_lost()
2179 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_SACKED_ACKED;  in tcp_timeout_mark_lost()
2195 	bool new_recovery = icsk->icsk_ca_state < TCP_CA_Recovery;  in tcp_enter_loss()
2201 	if (icsk->icsk_ca_state <= TCP_CA_Disorder ||  in tcp_enter_loss()
2202 	    !after(tp->high_seq, tp->snd_una) ||  in tcp_enter_loss()
2203 	    (icsk->icsk_ca_state == TCP_CA_Loss && !icsk->icsk_retransmits)) {  in tcp_enter_loss()
2204 		tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_enter_loss()
2205 		tp->prior_cwnd = tcp_snd_cwnd(tp);  in tcp_enter_loss()
2206 		tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);  in tcp_enter_loss()
2211 	tp->snd_cwnd_cnt   = 0;  in tcp_enter_loss()
2212 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_enter_loss()
2215 	 * suggests that the degree of reordering is over-estimated.  in tcp_enter_loss()
2217 	reordering = READ_ONCE(net->ipv4.sysctl_tcp_reordering);  in tcp_enter_loss()
2218 	if (icsk->icsk_ca_state <= TCP_CA_Disorder &&  in tcp_enter_loss()
2219 	    tp->sacked_out >= reordering)  in tcp_enter_loss()
2220 		tp->reordering = min_t(unsigned int, tp->reordering,  in tcp_enter_loss()
2224 	tp->high_seq = tp->snd_nxt;  in tcp_enter_loss()
2225 	tp->tlp_high_seq = 0;  in tcp_enter_loss()
2228 	/* F-RTO RFC5682 sec 3.1 step 1: retransmit SND.UNA if no previous  in tcp_enter_loss()
2230 	 * the same SND.UNA (sec 3.2). Disable F-RTO on path MTU probing  in tcp_enter_loss()
2232 	tp->frto = READ_ONCE(net->ipv4.sysctl_tcp_frto) &&  in tcp_enter_loss()
2233 		   (new_recovery || icsk->icsk_retransmits) &&  in tcp_enter_loss()
2234 		   !inet_csk(sk)->icsk_mtup.probe_size;  in tcp_enter_loss()
2252 		unsigned long delay = max(usecs_to_jiffies(tp->srtt_us >> 4),  in tcp_check_sack_reneging()
2270  * but when e.g. out-of-window ACKs or packet duplication occurs,
2276 	return tp->sacked_out + 1;  in tcp_dupack_heuristics()
2280  * --------------------------------------
2289  * "Recovery"	CWND was reduced, we are fast-retransmitting.
2293  * - each incoming ACK, if state is not "Open"
2294  * - when arrived ACK is unusual, namely:
2301  *	in_flight = packets_out - left_out + retrans_out
2303  *	packets_out is SND.NXT-SND.UNA counted in packets.
2337  *		RACK (draft-ietf-tcpm-rack-01): it is a newer algorithm
2338  *		(2017-) that checks timing instead of counting DUPACKs.
2364  * timestamps and D-SACKs. When we detect that some segments were
2381 	if (tp->lost_out)  in tcp_time_to_recover()
2384 	/* Not-A-Trick#2 : Classic rule... */  in tcp_time_to_recover()
2385 	if (!tcp_is_rack(sk) && tcp_dupack_heuristics(tp) > tp->reordering)  in tcp_time_to_recover()
2393  * has at least tp->reordering SACKed seqments above it; "packets" refers to
2402 	const u32 loss_high = tp->snd_nxt;  in tcp_mark_head_lost()
2404 	WARN_ON(packets > tp->packets_out);  in tcp_mark_head_lost()
2405 	skb = tp->lost_skb_hint;  in tcp_mark_head_lost()
2408 		if (mark_head && after(TCP_SKB_CB(skb)->seq, tp->snd_una))  in tcp_mark_head_lost()
2410 		cnt = tp->lost_cnt_hint;  in tcp_mark_head_lost()
2419 		tp->lost_skb_hint = skb;  in tcp_mark_head_lost()
2420 		tp->lost_cnt_hint = cnt;  in tcp_mark_head_lost()
2422 		if (after(TCP_SKB_CB(skb)->end_seq, loss_high))  in tcp_mark_head_lost()
2425 		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_mark_head_lost()
2431 		if (!(TCP_SKB_CB(skb)->sacked & TCPCB_LOST))  in tcp_mark_head_lost()
2447 		int sacked_upto = tp->sacked_out - tp->reordering;  in tcp_update_scoreboard()
2457 	return tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&  in tcp_tsopt_ecr_before()
2458 	       before(tp->rx_opt.rcv_tsecr, when);  in tcp_tsopt_ecr_before()
2467 	return (TCP_SKB_CB(skb)->sacked & TCPCB_RETRANS) &&  in tcp_skb_spurious_retrans()
2468 	       tcp_tsopt_ecr_before(tp, tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));  in tcp_skb_spurious_retrans()
2478 	if (tp->retrans_stamp &&  in tcp_packet_delayed()
2479 	    tcp_tsopt_ecr_before(tp, tp->retrans_stamp))  in tcp_packet_delayed()
2487 	if (!tp->retrans_stamp &&	   /* no record of a retransmit/SYN? */  in tcp_packet_delayed()
2488 	    sk->sk_state != TCP_SYN_SENT)  /* not the FLAG_SYN_ACKED case? */  in tcp_packet_delayed()
2498  * tp->retrans_out, however, that kind of assumptions doesn't consider
2515 	if (tp->retrans_out)  in tcp_any_retrans_done()
2519 	if (unlikely(skb && TCP_SKB_CB(skb)->sacked & TCPCB_EVER_RETRANS))  in tcp_any_retrans_done()
2527  * retransmits_timed_out() and timestamp-undo are using the correct value.
2532 		tcp_sk(sk)->retrans_stamp = 0;  in tcp_retrans_stamp_cleanup()
2541 	if (sk->sk_family == AF_INET) {  in DBGUNDO()
2544 			 &inet->inet_daddr, ntohs(inet->inet_dport),  in DBGUNDO()
2546 			 tp->snd_ssthresh, tp->prior_ssthresh,  in DBGUNDO()
2547 			 tp->packets_out);  in DBGUNDO()
2550 	else if (sk->sk_family == AF_INET6) {  in DBGUNDO()
2553 			 &sk->sk_v6_daddr, ntohs(inet->inet_dport),  in DBGUNDO()
2555 			 tp->snd_ssthresh, tp->prior_ssthresh,  in DBGUNDO()
2556 			 tp->packets_out);  in DBGUNDO()
2569 		skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {  in tcp_undo_cwnd_reduction()
2570 			TCP_SKB_CB(skb)->sacked &= ~TCPCB_LOST;  in tcp_undo_cwnd_reduction()
2572 		tp->lost_out = 0;  in tcp_undo_cwnd_reduction()
2576 	if (tp->prior_ssthresh) {  in tcp_undo_cwnd_reduction()
2579 		tcp_snd_cwnd_set(tp, icsk->icsk_ca_ops->undo_cwnd(sk));  in tcp_undo_cwnd_reduction()
2581 		if (tp->prior_ssthresh > tp->snd_ssthresh) {  in tcp_undo_cwnd_reduction()
2582 			tp->snd_ssthresh = tp->prior_ssthresh;  in tcp_undo_cwnd_reduction()
2586 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_undo_cwnd_reduction()
2587 	tp->undo_marker = 0;  in tcp_undo_cwnd_reduction()
2588 	tp->rack.advanced = 1; /* Force RACK to re-exam losses */  in tcp_undo_cwnd_reduction()
2593 	return tp->undo_marker && (!tp->undo_retrans || tcp_packet_delayed(tp));  in tcp_may_undo()
2600 	if (tp->snd_una == tp->high_seq && tcp_is_reno(tp)) {  in tcp_is_non_sack_preventing_reopen()
2605 			tp->retrans_stamp = 0;  in tcp_is_non_sack_preventing_reopen()
2622 		DBGUNDO(sk, inet_csk(sk)->icsk_ca_state == TCP_CA_Loss ? "loss" : "retrans");  in tcp_try_undo_recovery()
2624 		if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss)  in tcp_try_undo_recovery()
2630 	} else if (tp->rack.reo_wnd_persist) {  in tcp_try_undo_recovery()
2631 		tp->rack.reo_wnd_persist--;  in tcp_try_undo_recovery()
2636 	tp->is_sack_reneg = 0;  in tcp_try_undo_recovery()
2640 /* Try to undo cwnd reduction, because D-SACKs acked all retransmitted data */
2645 	if (tp->undo_marker && !tp->undo_retrans) {  in tcp_try_undo_dsack()
2646 		tp->rack.reo_wnd_persist = min(TCP_RACK_RECOVERY_THRESH,  in tcp_try_undo_dsack()
2647 					       tp->rack.reo_wnd_persist + 1);  in tcp_try_undo_dsack()
2648 		DBGUNDO(sk, "D-SACK");  in tcp_try_undo_dsack()
2656 /* Undo during loss recovery after partial ACK or using F-RTO. */
2669 		inet_csk(sk)->icsk_retransmits = 0;  in tcp_try_undo_loss()
2674 			tp->is_sack_reneg = 0;  in tcp_try_undo_loss()
2694 	tp->high_seq = tp->snd_nxt;  in tcp_init_cwnd_reduction()
2695 	tp->tlp_high_seq = 0;  in tcp_init_cwnd_reduction()
2696 	tp->snd_cwnd_cnt = 0;  in tcp_init_cwnd_reduction()
2697 	tp->prior_cwnd = tcp_snd_cwnd(tp);  in tcp_init_cwnd_reduction()
2698 	tp->prr_delivered = 0;  in tcp_init_cwnd_reduction()
2699 	tp->prr_out = 0;  in tcp_init_cwnd_reduction()
2700 	tp->snd_ssthresh = inet_csk(sk)->icsk_ca_ops->ssthresh(sk);  in tcp_init_cwnd_reduction()
2708 	int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp);  in tcp_cwnd_reduction()
2710 	if (newly_acked_sacked <= 0 || WARN_ON_ONCE(!tp->prior_cwnd))  in tcp_cwnd_reduction()
2713 	tp->prr_delivered += newly_acked_sacked;  in tcp_cwnd_reduction()
2715 		u64 dividend = (u64)tp->snd_ssthresh * tp->prr_delivered +  in tcp_cwnd_reduction()
2716 			       tp->prior_cwnd - 1;  in tcp_cwnd_reduction()
2717 		sndcnt = div_u64(dividend, tp->prior_cwnd) - tp->prr_out;  in tcp_cwnd_reduction()
2719 		sndcnt = max_t(int, tp->prr_delivered - tp->prr_out,  in tcp_cwnd_reduction()
2726 	sndcnt = max(sndcnt, (tp->prr_out ? 0 : 1));  in tcp_cwnd_reduction()
2734 	if (inet_csk(sk)->icsk_ca_ops->cong_control)  in tcp_end_cwnd_reduction()
2738 	if (tp->snd_ssthresh < TCP_INFINITE_SSTHRESH &&  in tcp_end_cwnd_reduction()
2739 	    (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR || tp->undo_marker)) {  in tcp_end_cwnd_reduction()
2740 		tcp_snd_cwnd_set(tp, tp->snd_ssthresh);  in tcp_end_cwnd_reduction()
2741 		tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_end_cwnd_reduction()
2751 	tp->prior_ssthresh = 0;  in tcp_enter_cwr()
2752 	if (inet_csk(sk)->icsk_ca_state < TCP_CA_CWR) {  in tcp_enter_cwr()
2753 		tp->undo_marker = 0;  in tcp_enter_cwr()
2768 	if (inet_csk(sk)->icsk_ca_state != state) {  in tcp_try_keep_open()
2770 		tp->high_seq = tp->snd_nxt;  in tcp_try_keep_open()
2781 		tp->retrans_stamp = 0;  in tcp_try_to_open()
2786 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {  in tcp_try_to_open()
2795 	icsk->icsk_mtup.search_high = icsk->icsk_mtup.probe_size - 1;  in tcp_mtup_probe_failed()
2796 	icsk->icsk_mtup.probe_size = 0;  in tcp_mtup_probe_failed()
2806 	tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_mtup_probe_success()
2808 	val = (u64)tcp_snd_cwnd(tp) * tcp_mss_to_mtu(sk, tp->mss_cache);  in tcp_mtup_probe_success()
2809 	do_div(val, icsk->icsk_mtup.probe_size);  in tcp_mtup_probe_success()
2813 	tp->snd_cwnd_cnt = 0;  in tcp_mtup_probe_success()
2814 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_mtup_probe_success()
2815 	tp->snd_ssthresh = tcp_current_ssthresh(sk);  in tcp_mtup_probe_success()
2817 	icsk->icsk_mtup.search_low = icsk->icsk_mtup.probe_size;  in tcp_mtup_probe_success()
2818 	icsk->icsk_mtup.probe_size = 0;  in tcp_mtup_probe_success()
2819 	tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_mtup_probe_success()
2827  * non-zero value (and may do so in a later calling context due to TSQ), we
2838 	if (icsk->icsk_ca_state != TCP_CA_Loss) {  in tcp_non_congestion_loss_retransmit()
2839 		tp->high_seq = tp->snd_nxt;  in tcp_non_congestion_loss_retransmit()
2840 		tp->snd_ssthresh = tcp_current_ssthresh(sk);  in tcp_non_congestion_loss_retransmit()
2841 		tp->prior_ssthresh = 0;  in tcp_non_congestion_loss_retransmit()
2842 		tp->undo_marker = 0;  in tcp_non_congestion_loss_retransmit()
2866 	 * -1 to mark all frames as lost, otherwise compute the current MSS.  in tcp_simple_retransmit()
2868 	if (tp->syn_data && sk->sk_state == TCP_SYN_SENT)  in tcp_simple_retransmit()
2869 		mss = -1;  in tcp_simple_retransmit()
2873 	skb_rbtree_walk(skb, &sk->tcp_rtx_queue) {  in tcp_simple_retransmit()
2880 	if (!tp->lost_out)  in tcp_simple_retransmit()
2912 	tp->prior_ssthresh = 0;  in tcp_enter_recovery()
2917 			tp->prior_ssthresh = tcp_current_ssthresh(sk);  in tcp_enter_recovery()
2925 	if (tp->rto_stamp) {  in tcp_update_rto_time()
2926 		tp->total_rto_time += tcp_time_stamp_ms(tp) - tp->rto_stamp;  in tcp_update_rto_time()
2927 		tp->rto_stamp = 0;  in tcp_update_rto_time()
2938 	bool recovered = !before(tp->snd_una, tp->high_seq);  in tcp_process_loss()
2940 	if ((flag & FLAG_SND_UNA_ADVANCED || rcu_access_pointer(tp->fastopen_rsk)) &&  in tcp_process_loss()
2944 	if (tp->frto) { /* F-RTO RFC5682 sec 3.1 (sack enhanced version). */  in tcp_process_loss()
2946 		 * lost, i.e., never-retransmitted data are (s)acked.  in tcp_process_loss()
2952 		if (after(tp->snd_nxt, tp->high_seq)) {  in tcp_process_loss()
2954 				tp->frto = 0; /* Step 3.a. loss was real */  in tcp_process_loss()
2956 			tp->high_seq = tp->snd_nxt;  in tcp_process_loss()
2962 			    after(tcp_wnd_end(tp), tp->snd_nxt)) {  in tcp_process_loss()
2966 			tp->frto = 0;  in tcp_process_loss()
2971 		/* F-RTO RFC5682 sec 3.1 step 2.a and 1st part of step 3.a */  in tcp_process_loss()
2976 		/* A Reno DUPACK means new data in F-RTO step 2.b above are  in tcp_process_loss()
2979 		if (after(tp->snd_nxt, tp->high_seq) && num_dupack)  in tcp_process_loss()
2992 		     tp->snd_una + tp->reordering * tp->mss_cache);  in tcp_force_fast_retransmit()
3001 	if (tp->undo_marker && tcp_packet_delayed(tp)) {  in tcp_try_undo_partial()
3012 		if (tp->retrans_out)  in tcp_try_undo_partial()
3016 			tp->retrans_stamp = 0;  in tcp_try_undo_partial()
3039 		u32 prior_retrans = tp->retrans_out;  in tcp_identify_packet_loss()
3043 		if (prior_retrans > tp->retrans_out)  in tcp_identify_packet_loss()
3048 /* Process an event, which can update packets-in-flight not trivially.
3070 	if (!tp->packets_out && tp->sacked_out)  in tcp_fastretrans_alert()
3071 		tp->sacked_out = 0;  in tcp_fastretrans_alert()
3076 		tp->prior_ssthresh = 0;  in tcp_fastretrans_alert()
3087 	if (icsk->icsk_ca_state == TCP_CA_Open) {  in tcp_fastretrans_alert()
3088 		WARN_ON(tp->retrans_out != 0 && !tp->syn_data);  in tcp_fastretrans_alert()
3089 		tp->retrans_stamp = 0;  in tcp_fastretrans_alert()
3090 	} else if (!before(tp->snd_una, tp->high_seq)) {  in tcp_fastretrans_alert()
3091 		switch (icsk->icsk_ca_state) {  in tcp_fastretrans_alert()
3095 			if (tp->snd_una != tp->high_seq) {  in tcp_fastretrans_alert()
3112 	switch (icsk->icsk_ca_state) {  in tcp_fastretrans_alert()
3124 		if (icsk->icsk_ca_state != TCP_CA_Recovery) {  in tcp_fastretrans_alert()
3135 		if (icsk->icsk_ca_state != TCP_CA_Loss)  in tcp_fastretrans_alert()
3138 		if (!(icsk->icsk_ca_state == TCP_CA_Open ||  in tcp_fastretrans_alert()
3150 		if (icsk->icsk_ca_state <= TCP_CA_Disorder)  in tcp_fastretrans_alert()
3160 		if (icsk->icsk_ca_state < TCP_CA_CWR &&  in tcp_fastretrans_alert()
3161 		    icsk->icsk_mtup.probe_size &&  in tcp_fastretrans_alert()
3162 		    tp->snd_una == tp->mtu_probe.probe_seq_start) {  in tcp_fastretrans_alert()
3182 	u32 wlen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen) * HZ;  in tcp_update_rtt_min()
3192 	minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32,  in tcp_update_rtt_min()
3202 	/* Prefer RTT measured from ACK's timing to TS-ECR. This is because  in tcp_ack_update_rtt()
3203 	 * broken middle-boxes or peers may corrupt TS-ECR fields. But  in tcp_ack_update_rtt()
3210 	/* RTTM Rule: A TSecr value received in a segment is used to  in tcp_ack_update_rtt()
3214 	 * See draft-ietf-tcplw-high-performance-00, section 3.3.  in tcp_ack_update_rtt()
3216 	if (seq_rtt_us < 0 && tp->rx_opt.saw_tstamp &&  in tcp_ack_update_rtt()
3217 	    tp->rx_opt.rcv_tsecr && flag & FLAG_ACKED)  in tcp_ack_update_rtt()
3220 	rs->rtt_us = ca_rtt_us; /* RTT of last (S)ACKed packet (or -1) */  in tcp_ack_update_rtt()
3225 	 * always taken together with ACK, SACK, or TS-opts. Any negative  in tcp_ack_update_rtt()
3233 	inet_csk(sk)->icsk_backoff = 0;  in tcp_ack_update_rtt()
3241 	long rtt_us = -1L;  in tcp_synack_rtt_meas()
3243 	if (req && !req->num_retrans && tcp_rsk(req)->snt_synack)  in tcp_synack_rtt_meas()
3244 		rtt_us = tcp_stamp_us_delta(tcp_clock_us(), tcp_rsk(req)->snt_synack);  in tcp_synack_rtt_meas()
3246 	tcp_ack_update_rtt(sk, FLAG_SYN_ACKED, rtt_us, -1L, rtt_us, &rs);  in tcp_synack_rtt_meas()
3254 	icsk->icsk_ca_ops->cong_avoid(sk, ack, acked);  in tcp_cong_avoid()
3255 	tcp_sk(sk)->snd_cwnd_stamp = tcp_jiffies32;  in tcp_cong_avoid()
3267 	 * for SYN-ACK retrans purpose, stay put.  in tcp_rearm_rto()
3269 	if (rcu_access_pointer(tp->fastopen_rsk))  in tcp_rearm_rto()
3272 	if (!tp->packets_out) {  in tcp_rearm_rto()
3275 		u32 rto = inet_csk(sk)->icsk_rto;  in tcp_rearm_rto()
3277 		if (icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||  in tcp_rearm_rto()
3278 		    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {  in tcp_rearm_rto()
3303 	BUG_ON(!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una));  in tcp_tso_acked()
3306 	if (tcp_trim_head(sk, skb, tp->snd_una - TCP_SKB_CB(skb)->seq))  in tcp_tso_acked()
3308 	packets_acked -= tcp_skb_pcount(skb);  in tcp_tso_acked()
3312 		BUG_ON(!before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq));  in tcp_tso_acked()
3323 	/* Avoid cache line misses to get skb_shinfo() and shinfo->tx_flags */  in tcp_ack_tstamp()
3324 	if (likely(!TCP_SKB_CB(skb)->txstamp_ack))  in tcp_ack_tstamp()
3328 	if (!before(shinfo->tskey, prior_snd_una) &&  in tcp_ack_tstamp()
3329 	    before(shinfo->tskey, tcp_sk(sk)->snd_una)) {  in tcp_ack_tstamp()
3347 	u32 prior_sacked = tp->sacked_out;  in tcp_clean_rtx_queue()
3348 	u32 reord = tp->snd_nxt; /* lowest acked un-retx un-sacked seq */  in tcp_clean_rtx_queue()
3351 	long sack_rtt_us = -1L;  in tcp_clean_rtx_queue()
3352 	long seq_rtt_us = -1L;  in tcp_clean_rtx_queue()
3353 	long ca_rtt_us = -1L;  in tcp_clean_rtx_queue()
3360 	for (skb = skb_rb_first(&sk->tcp_rtx_queue); skb; skb = next) {  in tcp_clean_rtx_queue()
3362 		const u32 start_seq = scb->seq;  in tcp_clean_rtx_queue()
3363 		u8 sacked = scb->sacked;  in tcp_clean_rtx_queue()
3367 		if (after(scb->end_seq, tp->snd_una)) {  in tcp_clean_rtx_queue()
3369 			    !after(tp->snd_una, scb->seq))  in tcp_clean_rtx_queue()
3382 				tp->retrans_out -= acked_pcount;  in tcp_clean_rtx_queue()
3392 			if (!after(scb->end_seq, tp->high_seq))  in tcp_clean_rtx_queue()
3397 			tp->sacked_out -= acked_pcount;  in tcp_clean_rtx_queue()
3401 				tcp_rack_advance(tp, sacked, scb->end_seq,  in tcp_clean_rtx_queue()
3405 			tp->lost_out -= acked_pcount;  in tcp_clean_rtx_queue()
3407 		tp->packets_out -= acked_pcount;  in tcp_clean_rtx_queue()
3409 		tcp_rate_skb_delivered(sk, skb, sack->rate);  in tcp_clean_rtx_queue()
3418 		if (likely(!(scb->tcp_flags & TCPHDR_SYN))) {  in tcp_clean_rtx_queue()
3422 			tp->retrans_stamp = 0;  in tcp_clean_rtx_queue()
3431 		if (unlikely(skb == tp->retransmit_skb_hint))  in tcp_clean_rtx_queue()
3432 			tp->retransmit_skb_hint = NULL;  in tcp_clean_rtx_queue()
3433 		if (unlikely(skb == tp->lost_skb_hint))  in tcp_clean_rtx_queue()
3434 			tp->lost_skb_hint = NULL;  in tcp_clean_rtx_queue()
3442 	if (likely(between(tp->snd_up, prior_snd_una, tp->snd_una)))  in tcp_clean_rtx_queue()
3443 		tp->snd_up = tp->snd_una;  in tcp_clean_rtx_queue()
3447 		if (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)  in tcp_clean_rtx_queue()
3452 		seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt);  in tcp_clean_rtx_queue()
3453 		ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt);  in tcp_clean_rtx_queue()
3456 		    (tp->snd_una - prior_snd_una) < tp->mss_cache &&  in tcp_clean_rtx_queue()
3457 		    sack->rate->prior_delivered + 1 == tp->delivered &&  in tcp_clean_rtx_queue()
3461 			 * a receiver w/o out-of-order or CE events.  in tcp_clean_rtx_queue()
3466 	if (sack->first_sackt) {  in tcp_clean_rtx_queue()
3467 		sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt);  in tcp_clean_rtx_queue()
3468 		ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->last_sackt);  in tcp_clean_rtx_queue()
3471 					ca_rtt_us, sack->rate);  in tcp_clean_rtx_queue()
3475 		if (unlikely(icsk->icsk_mtup.probe_size &&  in tcp_clean_rtx_queue()
3476 			     !after(tp->mtu_probe.probe_seq_end, tp->snd_una))) {  in tcp_clean_rtx_queue()
3484 			 * retransmitted, non-SACK case cannot confirm that  in tcp_clean_rtx_queue()
3494 			/* Non-retransmitted hole got filled? That's reordering */  in tcp_clean_rtx_queue()
3498 			delta = prior_sacked - tp->sacked_out;  in tcp_clean_rtx_queue()
3499 			tp->lost_cnt_hint -= min(tp->lost_cnt_hint, delta);  in tcp_clean_rtx_queue()
3502 		   sack_rtt_us > tcp_stamp_us_delta(tp->tcp_mstamp,  in tcp_clean_rtx_queue()
3504 		/* Do not re-arm RTO if the sack RTT is measured from data sent  in tcp_clean_rtx_queue()
3511 	if (icsk->icsk_ca_ops->pkts_acked) {  in tcp_clean_rtx_queue()
3513 					     .rtt_us = sack->rate->rtt_us };  in tcp_clean_rtx_queue()
3515 		sample.in_flight = tp->mss_cache *  in tcp_clean_rtx_queue()
3516 			(tp->delivered - sack->rate->prior_delivered);  in tcp_clean_rtx_queue()
3517 		icsk->icsk_ca_ops->pkts_acked(sk, &sample);  in tcp_clean_rtx_queue()
3521 	WARN_ON((int)tp->sacked_out < 0);  in tcp_clean_rtx_queue()
3522 	WARN_ON((int)tp->lost_out < 0);  in tcp_clean_rtx_queue()
3523 	WARN_ON((int)tp->retrans_out < 0);  in tcp_clean_rtx_queue()
3524 	if (!tp->packets_out && tcp_is_sack(tp)) {  in tcp_clean_rtx_queue()
3526 		if (tp->lost_out) {  in tcp_clean_rtx_queue()
3528 				 tp->lost_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3529 			tp->lost_out = 0;  in tcp_clean_rtx_queue()
3531 		if (tp->sacked_out) {  in tcp_clean_rtx_queue()
3533 				 tp->sacked_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3534 			tp->sacked_out = 0;  in tcp_clean_rtx_queue()
3536 		if (tp->retrans_out) {  in tcp_clean_rtx_queue()
3538 				 tp->retrans_out, icsk->icsk_ca_state);  in tcp_clean_rtx_queue()
3539 			tp->retrans_out = 0;  in tcp_clean_rtx_queue()
3555 	if (!after(TCP_SKB_CB(head)->end_seq, tcp_wnd_end(tp))) {  in tcp_ack_probe()
3556 		icsk->icsk_backoff = 0;  in tcp_ack_probe()
3557 		icsk->icsk_probes_tstamp = 0;  in tcp_ack_probe()
3573 		inet_csk(sk)->icsk_ca_state != TCP_CA_Open;  in tcp_ack_is_dubious()
3581 	 * and only grow cwnd on in-order delivery (RFC5681). A stretched ACK w/  in tcp_may_raise_cwnd()
3585 	if (tcp_sk(sk)->reordering >  in tcp_may_raise_cwnd()
3586 	    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reordering))  in tcp_may_raise_cwnd()
3602 	if (icsk->icsk_ca_ops->cong_control) {  in tcp_cong_control()
3603 		icsk->icsk_ca_ops->cong_control(sk, ack, flag, rs);  in tcp_cong_control()
3609 		tcp_cwnd_reduction(sk, acked_sacked, rs->losses, flag);  in tcp_cong_control()
3624 	return	after(ack, tp->snd_una) ||  in tcp_may_update_window()
3625 		after(ack_seq, tp->snd_wl1) ||  in tcp_may_update_window()
3626 		(ack_seq == tp->snd_wl1 && (nwin > tp->snd_wnd || !nwin));  in tcp_may_update_window()
3637 	ao = rcu_dereference_protected(tp->ao_info,  in tcp_snd_sne_update()
3639 	if (ao && ack < tp->snd_una) {  in tcp_snd_sne_update()
3640 		ao->snd_sne++;  in tcp_snd_sne_update()
3641 		trace_tcp_ao_snd_sne_update((struct sock *)tp, ao->snd_sne);  in tcp_snd_sne_update()
3646 /* If we update tp->snd_una, also update tp->bytes_acked */
3649 	u32 delta = ack - tp->snd_una;  in tcp_snd_una_update()
3652 	tp->bytes_acked += delta;  in tcp_snd_una_update()
3654 	tp->snd_una = ack;  in tcp_snd_una_update()
3665 	ao = rcu_dereference_protected(tp->ao_info,  in tcp_rcv_sne_update()
3667 	if (ao && seq < tp->rcv_nxt) {  in tcp_rcv_sne_update()
3668 		ao->rcv_sne++;  in tcp_rcv_sne_update()
3669 		trace_tcp_ao_rcv_sne_update((struct sock *)tp, ao->rcv_sne);  in tcp_rcv_sne_update()
3674 /* If we update tp->rcv_nxt, also update tp->bytes_received */
3677 	u32 delta = seq - tp->rcv_nxt;  in tcp_rcv_nxt_update()
3680 	tp->bytes_received += delta;  in tcp_rcv_nxt_update()
3682 	WRITE_ONCE(tp->rcv_nxt, seq);  in tcp_rcv_nxt_update()
3687  * Window update algorithm, described in RFC793/RFC1122 (used in linux-2.2
3695 	u32 nwin = ntohs(tcp_hdr(skb)->window);  in tcp_ack_update_window()
3697 	if (likely(!tcp_hdr(skb)->syn))  in tcp_ack_update_window()
3698 		nwin <<= tp->rx_opt.snd_wscale;  in tcp_ack_update_window()
3704 		if (tp->snd_wnd != nwin) {  in tcp_ack_update_window()
3705 			tp->snd_wnd = nwin;  in tcp_ack_update_window()
3710 			tp->pred_flags = 0;  in tcp_ack_update_window()
3716 			if (nwin > tp->max_window) {  in tcp_ack_update_window()
3717 				tp->max_window = nwin;  in tcp_ack_update_window()
3718 				tcp_sync_mss(sk, inet_csk(sk)->icsk_pmtu_cookie);  in tcp_ack_update_window()
3735 		s32 elapsed = (s32)(tcp_jiffies32 - val);  in __tcp_oow_rate_limited()
3738 		    elapsed < READ_ONCE(net->ipv4.sysctl_tcp_invalid_ratelimit)) {  in __tcp_oow_rate_limited()
3740 			return true;	/* rate-limited: don't send yet! */  in __tcp_oow_rate_limited()
3749 	return false;	/* not rate-limited: go ahead, send dupack now! */  in __tcp_oow_rate_limited()
3752 /* Return true if we're currently rate-limiting out-of-window ACKs and
3753  * thus shouldn't send a dupack right now. We rate-limit dupacks in
3754  * response to out-of-window SYNs or ACKs to mitigate ACK loops or DoS
3757  * endpoint is sending out-of-window SYNs or pure ACKs at a high rate.
3763 	if ((TCP_SKB_CB(skb)->seq != TCP_SKB_CB(skb)->end_seq) &&  in tcp_oow_rate_limited()
3764 	    !tcp_hdr(skb)->syn)  in tcp_oow_rate_limited()
3777 	/* First check our per-socket dupack rate limit. */  in tcp_send_challenge_ack()
3780 				   &tp->last_oow_ack_time))  in tcp_send_challenge_ack()
3783 	ack_limit = READ_ONCE(net->ipv4.sysctl_tcp_challenge_ack_limit);  in tcp_send_challenge_ack()
3787 	/* Then check host-wide RFC 5961 rate limit. */  in tcp_send_challenge_ack()
3789 	if (now != READ_ONCE(net->ipv4.tcp_challenge_timestamp)) {  in tcp_send_challenge_ack()
3792 		WRITE_ONCE(net->ipv4.tcp_challenge_timestamp, now);  in tcp_send_challenge_ack()
3793 		WRITE_ONCE(net->ipv4.tcp_challenge_count,  in tcp_send_challenge_ack()
3794 			   get_random_u32_inclusive(half, ack_limit + half - 1));  in tcp_send_challenge_ack()
3796 	count = READ_ONCE(net->ipv4.tcp_challenge_count);  in tcp_send_challenge_ack()
3798 		WRITE_ONCE(net->ipv4.tcp_challenge_count, count - 1);  in tcp_send_challenge_ack()
3807 	tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval;  in tcp_store_ts_recent()
3808 	tp->rx_opt.ts_recent_stamp = ktime_get_seconds();  in tcp_store_ts_recent()
3813 	if (tp->rx_opt.saw_tstamp && !after(seq, tp->rcv_wup)) {  in tcp_replace_ts_recent()
3816 		 * for pure ACK frames.  -DaveM  in tcp_replace_ts_recent()
3821 		if (tcp_paws_check(&tp->rx_opt, 0))  in tcp_replace_ts_recent()
3827  * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack
3833 	if (before(ack, tp->tlp_high_seq))  in tcp_process_tlp_ack()
3836 	if (!tp->tlp_retrans) {  in tcp_process_tlp_ack()
3838 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3841 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3842 	} else if (after(ack, tp->tlp_high_seq)) {  in tcp_process_tlp_ack()
3855 		tp->tlp_high_seq = 0;  in tcp_process_tlp_ack()
3863 	if (icsk->icsk_ca_ops->in_ack_event)  in tcp_in_ack_event()
3864 		icsk->icsk_ca_ops->in_ack_event(sk, flags);  in tcp_in_ack_event()
3875 	if (rexmit == REXMIT_NONE || sk->sk_state == TCP_SYN_SENT)  in tcp_xmit_recovery()
3881 		if (after(tp->snd_nxt, tp->high_seq))  in tcp_xmit_recovery()
3883 		tp->frto = 0;  in tcp_xmit_recovery()
3895 	delivered = tp->delivered - prior_delivered;  in tcp_newly_delivered()
3910 	u32 prior_snd_una = tp->snd_una;  in tcp_ack()
3911 	bool is_sack_reneg = tp->is_sack_reneg;  in tcp_ack()
3912 	u32 ack_seq = TCP_SKB_CB(skb)->seq;  in tcp_ack()
3913 	u32 ack = TCP_SKB_CB(skb)->ack_seq;  in tcp_ack()
3915 	int prior_packets = tp->packets_out;  in tcp_ack()
3916 	u32 delivered = tp->delivered;  in tcp_ack()
3917 	u32 lost = tp->lost;  in tcp_ack()
3926 	prefetch(sk->tcp_rtx_queue.rb_node);  in tcp_ack()
3935 		max_window = min_t(u64, tp->max_window, tp->bytes_acked);  in tcp_ack()
3937 		if (before(ack, prior_snd_una - max_window)) {  in tcp_ack()
3940 			return -SKB_DROP_REASON_TCP_TOO_OLD_ACK;  in tcp_ack()
3948 	if (after(ack, tp->snd_nxt))  in tcp_ack()
3949 		return -SKB_DROP_REASON_TCP_ACK_UNSENT_DATA;  in tcp_ack()
3953 		icsk->icsk_retransmits = 0;  in tcp_ack()
3957 			if (icsk->icsk_clean_acked)  in tcp_ack()
3958 				icsk->icsk_clean_acked(sk, ack);  in tcp_ack()
3962 	prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una;  in tcp_ack()
3969 		tcp_replace_ts_recent(tp, TCP_SKB_CB(skb)->seq);  in tcp_ack()
3987 		if (ack_seq != TCP_SKB_CB(skb)->end_seq)  in tcp_ack()
3994 		if (TCP_SKB_CB(skb)->sacked)  in tcp_ack()
4018 	 * with widely-deployed TCP implementations that do this.  in tcp_ack()
4025 	WRITE_ONCE(sk->sk_err_soft, 0);  in tcp_ack()
4026 	icsk->icsk_probes_out = 0;  in tcp_ack()
4027 	tp->rcv_tstamp = tcp_jiffies32;  in tcp_ack()
4037 	if (tp->tlp_high_seq)  in tcp_ack()
4046 				num_dupack = max_t(u16, 1, skb_shinfo(skb)->gso_segs);  in tcp_ack()
4060 	lost = tp->lost - lost;			/* freshly marked lost */  in tcp_ack()
4080 	if (tp->tlp_high_seq)  in tcp_ack()
4088 	if (TCP_SKB_CB(skb)->sacked) {  in tcp_ack()
4104 	/* Valid only in SYN or SYN-ACK with an even length.  */  in tcp_parse_fastopen_option()
4110 		memcpy(foc->val, cookie, len);  in tcp_parse_fastopen_option()
4112 		len = -1;  in tcp_parse_fastopen_option()
4113 	foc->len = len;  in tcp_parse_fastopen_option()
4114 	foc->exp = exp_opt;  in tcp_parse_fastopen_option()
4124 		if (th->syn && !(opsize & 1) &&  in smc_parse_options()
4127 			opt_rx->smc_ok = 1;  in smc_parse_options()
4136  * value on success.
4141 	int length = (th->doff * 4) - sizeof(struct tcphdr);  in tcp_parse_mss_option()
4152 			length--;  in tcp_parse_mss_option()
4171 			ptr += opsize - 2;  in tcp_parse_mss_option()
4172 			length -= opsize;  in tcp_parse_mss_option()
4190 	int length = (th->doff * 4) - sizeof(struct tcphdr);  in tcp_parse_options()
4193 	opt_rx->saw_tstamp = 0;  in tcp_parse_options()
4194 	opt_rx->saw_unknown = 0;  in tcp_parse_options()
4204 			length--;  in tcp_parse_options()
4216 				if (opsize == TCPOLEN_MSS && th->syn && !estab) {  in tcp_parse_options()
4219 						if (opt_rx->user_mss &&  in tcp_parse_options()
4220 						    opt_rx->user_mss < in_mss)  in tcp_parse_options()
4221 							in_mss = opt_rx->user_mss;  in tcp_parse_options()
4222 						opt_rx->mss_clamp = in_mss;  in tcp_parse_options()
4227 				if (opsize == TCPOLEN_WINDOW && th->syn &&  in tcp_parse_options()
4228 				    !estab && READ_ONCE(net->ipv4.sysctl_tcp_window_scaling)) {  in tcp_parse_options()
4230 					opt_rx->wscale_ok = 1;  in tcp_parse_options()
4232 						net_info_ratelimited("%s: Illegal window scaling value %d > %u received\n",  in tcp_parse_options()
4238 					opt_rx->snd_wscale = snd_wscale;  in tcp_parse_options()
4243 				    ((estab && opt_rx->tstamp_ok) ||  in tcp_parse_options()
4244 				     (!estab && READ_ONCE(net->ipv4.sysctl_tcp_timestamps)))) {  in tcp_parse_options()
4245 					opt_rx->saw_tstamp = 1;  in tcp_parse_options()
4246 					opt_rx->rcv_tsval = get_unaligned_be32(ptr);  in tcp_parse_options()
4247 					opt_rx->rcv_tsecr = get_unaligned_be32(ptr + 4);  in tcp_parse_options()
4251 				if (opsize == TCPOLEN_SACK_PERM && th->syn &&  in tcp_parse_options()
4252 				    !estab && READ_ONCE(net->ipv4.sysctl_tcp_sack)) {  in tcp_parse_options()
4253 					opt_rx->sack_ok = TCP_SACK_SEEN;  in tcp_parse_options()
4260 				   !((opsize - TCPOLEN_SACK_BASE) % TCPOLEN_SACK_PERBLOCK) &&  in tcp_parse_options()
4261 				   opt_rx->sack_ok) {  in tcp_parse_options()
4262 					TCP_SKB_CB(skb)->sacked = (ptr - 2) - (unsigned char *)th;  in tcp_parse_options()
4281 					opsize - TCPOLEN_FASTOPEN_BASE,  in tcp_parse_options()
4282 					ptr, th->syn, foc, false);  in tcp_parse_options()
4292 					tcp_parse_fastopen_option(opsize -  in tcp_parse_options()
4294 						ptr + 2, th->syn, foc, true);  in tcp_parse_options()
4301 				opt_rx->saw_unknown = 1;  in tcp_parse_options()
4305 				opt_rx->saw_unknown = 1;  in tcp_parse_options()
4307 			ptr += opsize-2;  in tcp_parse_options()
4308 			length -= opsize;  in tcp_parse_options()
4320 		tp->rx_opt.saw_tstamp = 1;  in tcp_parse_aligned_timestamp()
4322 		tp->rx_opt.rcv_tsval = ntohl(*ptr);  in tcp_parse_aligned_timestamp()
4325 			tp->rx_opt.rcv_tsecr = ntohl(*ptr) - tp->tsoffset;  in tcp_parse_aligned_timestamp()
4327 			tp->rx_opt.rcv_tsecr = 0;  in tcp_parse_aligned_timestamp()
4343 	if (th->doff == (sizeof(*th) / 4)) {  in tcp_fast_parse_options()
4344 		tp->rx_opt.saw_tstamp = 0;  in tcp_fast_parse_options()
4346 	} else if (tp->rx_opt.tstamp_ok &&  in tcp_fast_parse_options()
4347 		   th->doff == ((sizeof(*th) + TCPOLEN_TSTAMP_ALIGNED) / 4)) {  in tcp_fast_parse_options()
4352 	tcp_parse_options(net, skb, &tp->rx_opt, 1, NULL);  in tcp_fast_parse_options()
4353 	if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)  in tcp_fast_parse_options()
4354 		tp->rx_opt.rcv_tsecr -= tp->tsoffset;  in tcp_fast_parse_options()
4366 	int length = (th->doff << 2) - sizeof(*th);  in tcp_do_parse_auth_options()
4385 			length--;  in tcp_do_parse_auth_options()
4390 				return -EINVAL;  in tcp_do_parse_auth_options()
4393 					return -EINVAL;  in tcp_do_parse_auth_options()
4395 					return -EEXIST;  in tcp_do_parse_auth_options()
4399 					return -EINVAL;  in tcp_do_parse_auth_options()
4401 					return -EEXIST;  in tcp_do_parse_auth_options()
4405 		ptr += opsize - 2;  in tcp_do_parse_auth_options()
4406 		length -= opsize;  in tcp_do_parse_auth_options()
4413 /* Sorry, PAWS as specified is broken wrt. pure-ACKs -DaveM
4444 	if (tcp_sk(sk)->tcp_usec_ts)  in tcp_tsval_replay()
4445 		return inet_csk(sk)->icsk_rto * (USEC_PER_SEC / HZ);  in tcp_tsval_replay()
4450 	return inet_csk(sk)->icsk_rto * 1200 / HZ;  in tcp_tsval_replay()
4457 	u32 seq = TCP_SKB_CB(skb)->seq;  in tcp_disordered_ack()
4458 	u32 ack = TCP_SKB_CB(skb)->ack_seq;  in tcp_disordered_ack()
4461 		(th->ack && seq == TCP_SKB_CB(skb)->end_seq && seq == tp->rcv_nxt) &&  in tcp_disordered_ack()
4464 		ack == tp->snd_una &&  in tcp_disordered_ack()
4467 		!tcp_may_update_window(tp, ack, seq, ntohs(th->window) << tp->rx_opt.snd_wscale) &&  in tcp_disordered_ack()
4470 		(s32)(tp->rx_opt.ts_recent - tp->rx_opt.rcv_tsval) <=  in tcp_disordered_ack()
4479 	return !tcp_paws_check(&tp->rx_opt, TCP_PAWS_WINDOW) &&  in tcp_paws_discard()
4499 	if (before(end_seq, tp->rcv_wup))  in tcp_sequence()
4502 	if (after(seq, tp->rcv_nxt + tcp_receive_window(tp)))  in tcp_sequence()
4512 	WRITE_ONCE(sk->sk_err, err);  in tcp_done_with_error()
4531 	 * so just ignore the return value of mptcp_incoming_options().  in tcp_reset()
4537 	switch (sk->sk_state) {  in tcp_reset()
4557  *	If we are ESTABLISHED, a received fin moves us to CLOSE-WAIT
4558  *	(and thence onto LAST-ACK and finally, CLOSE, we never enter
4559  *	TIME-WAIT)
4561  *	If we are in FINWAIT-1, a received FIN indicates simultaneous
4562  *	close and we go into CLOSING (and later onto TIME-WAIT)
4564  *	If we are in FINWAIT-2, a received FIN moves us to TIME-WAIT.
4572 	WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | RCV_SHUTDOWN);  in tcp_fin()
4575 	switch (sk->sk_state) {  in tcp_fin()
4590 		/* RFC793: Remain in the LAST-ACK state. */  in tcp_fin()
4602 		/* Received a FIN -- send ACK and enter TIME_WAIT. */  in tcp_fin()
4610 		pr_err("%s: Impossible, sk->sk_state=%d\n",  in tcp_fin()
4611 		       __func__, sk->sk_state);  in tcp_fin()
4615 	/* It _is_ possible, that we have something out-of-order _after_ FIN.  in tcp_fin()
4618 	skb_rbtree_purge(&tp->out_of_order_queue);  in tcp_fin()
4620 		tcp_sack_reset(&tp->rx_opt);  in tcp_fin()
4623 		sk->sk_state_change(sk);  in tcp_fin()
4626 		if (sk->sk_shutdown == SHUTDOWN_MASK ||  in tcp_fin()
4627 		    sk->sk_state == TCP_CLOSE)  in tcp_fin()
4637 	if (!after(seq, sp->end_seq) && !after(sp->start_seq, end_seq)) {  in tcp_sack_extend()
4638 		if (before(seq, sp->start_seq))  in tcp_sack_extend()
4639 			sp->start_seq = seq;  in tcp_sack_extend()
4640 		if (after(end_seq, sp->end_seq))  in tcp_sack_extend()
4641 			sp->end_seq = end_seq;  in tcp_sack_extend()
4651 	if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) {  in tcp_dsack_set()
4654 		if (before(seq, tp->rcv_nxt))  in tcp_dsack_set()
4661 		tp->rx_opt.dsack = 1;  in tcp_dsack_set()
4662 		tp->duplicate_sack[0].start_seq = seq;  in tcp_dsack_set()
4663 		tp->duplicate_sack[0].end_seq = end_seq;  in tcp_dsack_set()
4671 	if (!tp->rx_opt.dsack)  in tcp_dsack_extend()
4674 		tcp_sack_extend(tp->duplicate_sack, seq, end_seq);  in tcp_dsack_extend()
4688 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_Loss &&  in tcp_rcv_spurious_retrans()
4689 	    skb->protocol == htons(ETH_P_IPV6) &&  in tcp_rcv_spurious_retrans()
4690 	    (tcp_sk(sk)->inet_conn.icsk_ack.lrcv_flowlabel !=  in tcp_rcv_spurious_retrans()
4704 	if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_send_dupack()
4705 	    before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_send_dupack()
4709 		if (tcp_is_sack(tp) && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_dsack)) {  in tcp_send_dupack()
4710 			u32 end_seq = TCP_SKB_CB(skb)->end_seq;  in tcp_send_dupack()
4713 			if (after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))  in tcp_send_dupack()
4714 				end_seq = tp->rcv_nxt;  in tcp_send_dupack()
4715 			tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, end_seq);  in tcp_send_dupack()
4722 /* These routines update the SACK block as out-of-order packets arrive or
4723  * in-order packets close up the sequence space.
4728 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_maybe_coalesce()
4734 	for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;) {  in tcp_sack_maybe_coalesce()
4735 		if (tcp_sack_extend(sp, swalk->start_seq, swalk->end_seq)) {  in tcp_sack_maybe_coalesce()
4741 			tp->rx_opt.num_sacks--;  in tcp_sack_maybe_coalesce()
4742 			for (i = this_sack; i < tp->rx_opt.num_sacks; i++)  in tcp_sack_maybe_coalesce()
4755 	if (!tp->compressed_ack)  in tcp_sack_compress_send_ack()
4758 	if (hrtimer_try_to_cancel(&tp->compressed_ack_timer) == 1)  in tcp_sack_compress_send_ack()
4762 	 * substract one from tp->compressed_ack to keep  in tcp_sack_compress_send_ack()
4766 		      tp->compressed_ack - 1);  in tcp_sack_compress_send_ack()
4768 	tp->compressed_ack = 0;  in tcp_sack_compress_send_ack()
4781 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_new_ofo_skb()
4782 	int cur_sacks = tp->rx_opt.num_sacks;  in tcp_sack_new_ofo_skb()
4793 			for (; this_sack > 0; this_sack--, sp--)  in tcp_sack_new_ofo_skb()
4794 				swap(*sp, *(sp - 1));  in tcp_sack_new_ofo_skb()
4811 		this_sack--;  in tcp_sack_new_ofo_skb()
4812 		tp->rx_opt.num_sacks--;  in tcp_sack_new_ofo_skb()
4813 		sp--;  in tcp_sack_new_ofo_skb()
4815 	for (; this_sack > 0; this_sack--, sp--)  in tcp_sack_new_ofo_skb()
4816 		*sp = *(sp - 1);  in tcp_sack_new_ofo_skb()
4820 	sp->start_seq = seq;  in tcp_sack_new_ofo_skb()
4821 	sp->end_seq = end_seq;  in tcp_sack_new_ofo_skb()
4822 	tp->rx_opt.num_sacks++;  in tcp_sack_new_ofo_skb()
4829 	struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_sack_remove()
4830 	int num_sacks = tp->rx_opt.num_sacks;  in tcp_sack_remove()
4834 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_sack_remove()
4835 		tp->rx_opt.num_sacks = 0;  in tcp_sack_remove()
4841 		if (!before(tp->rcv_nxt, sp->start_seq)) {  in tcp_sack_remove()
4845 			WARN_ON(before(tp->rcv_nxt, sp->end_seq));  in tcp_sack_remove()
4849 				tp->selective_acks[i-1] = tp->selective_acks[i];  in tcp_sack_remove()
4850 			num_sacks--;  in tcp_sack_remove()
4856 	tp->rx_opt.num_sacks = num_sacks;  in tcp_sack_remove()
4860  * tcp_try_coalesce - try to merge skb to prior one
4882 	if (TCP_SKB_CB(from)->seq != TCP_SKB_CB(to)->end_seq)  in tcp_try_coalesce()
4891 	atomic_add(delta, &sk->sk_rmem_alloc);  in tcp_try_coalesce()
4894 	TCP_SKB_CB(to)->end_seq = TCP_SKB_CB(from)->end_seq;  in tcp_try_coalesce()
4895 	TCP_SKB_CB(to)->ack_seq = TCP_SKB_CB(from)->ack_seq;  in tcp_try_coalesce()
4896 	TCP_SKB_CB(to)->tcp_flags |= TCP_SKB_CB(from)->tcp_flags;  in tcp_try_coalesce()
4898 	if (TCP_SKB_CB(from)->has_rxtstamp) {  in tcp_try_coalesce()
4899 		TCP_SKB_CB(to)->has_rxtstamp = true;  in tcp_try_coalesce()
4900 		to->tstamp = from->tstamp;  in tcp_try_coalesce()
4901 		skb_hwtstamps(to)->hwtstamp = skb_hwtstamps(from)->hwtstamp;  in tcp_try_coalesce()
4914 	/* In case tcp_drop_reason() is called later, update to->gso_segs */  in tcp_ooo_try_coalesce()
4916 		u32 gso_segs = max_t(u16, 1, skb_shinfo(to)->gso_segs) +  in tcp_ooo_try_coalesce()
4917 			       max_t(u16, 1, skb_shinfo(from)->gso_segs);  in tcp_ooo_try_coalesce()
4919 		skb_shinfo(to)->gso_segs = min_t(u32, gso_segs, 0xFFFF);  in tcp_ooo_try_coalesce()
4937 	__u32 dsack_high = tp->rcv_nxt;  in tcp_ofo_queue()
4942 	p = rb_first(&tp->out_of_order_queue);  in tcp_ofo_queue()
4945 		if (after(TCP_SKB_CB(skb)->seq, tp->rcv_nxt))  in tcp_ofo_queue()
4948 		if (before(TCP_SKB_CB(skb)->seq, dsack_high)) {  in tcp_ofo_queue()
4950 			if (before(TCP_SKB_CB(skb)->end_seq, dsack_high))  in tcp_ofo_queue()
4951 				dsack_high = TCP_SKB_CB(skb)->end_seq;  in tcp_ofo_queue()
4952 			tcp_dsack_extend(sk, TCP_SKB_CB(skb)->seq, dsack);  in tcp_ofo_queue()
4955 		rb_erase(&skb->rbnode, &tp->out_of_order_queue);  in tcp_ofo_queue()
4957 		if (unlikely(!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt))) {  in tcp_ofo_queue()
4962 		tail = skb_peek_tail(&sk->sk_receive_queue);  in tcp_ofo_queue()
4964 		tcp_rcv_nxt_update(tp, TCP_SKB_CB(skb)->end_seq);  in tcp_ofo_queue()
4965 		fin = TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN;  in tcp_ofo_queue()
4967 			__skb_queue_tail(&sk->sk_receive_queue, skb);  in tcp_ofo_queue()
4973 			/* tcp_fin() purges tp->out_of_order_queue,  in tcp_ofo_queue()
4987 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||  in tcp_try_rmem_schedule()
4991 			return -1;  in tcp_try_rmem_schedule()
4995 				return -1;  in tcp_try_rmem_schedule()
5012 	if (unlikely(tcp_try_rmem_schedule(sk, skb, skb->truesize))) {  in tcp_data_queue_ofo()
5014 		sk->sk_data_ready(sk);  in tcp_data_queue_ofo()
5020 	tp->pred_flags = 0;  in tcp_data_queue_ofo()
5023 	tp->rcv_ooopack += max_t(u16, 1, skb_shinfo(skb)->gso_segs);  in tcp_data_queue_ofo()
5025 	seq = TCP_SKB_CB(skb)->seq;  in tcp_data_queue_ofo()
5026 	end_seq = TCP_SKB_CB(skb)->end_seq;  in tcp_data_queue_ofo()
5028 	p = &tp->out_of_order_queue.rb_node;  in tcp_data_queue_ofo()
5029 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_data_queue_ofo()
5032 			tp->rx_opt.num_sacks = 1;  in tcp_data_queue_ofo()
5033 			tp->selective_acks[0].start_seq = seq;  in tcp_data_queue_ofo()
5034 			tp->selective_acks[0].end_seq = end_seq;  in tcp_data_queue_ofo()
5036 		rb_link_node(&skb->rbnode, NULL, p);  in tcp_data_queue_ofo()
5037 		rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5038 		tp->ooo_last_skb = skb;  in tcp_data_queue_ofo()
5045 	if (tcp_ooo_try_coalesce(sk, tp->ooo_last_skb,  in tcp_data_queue_ofo()
5058 	if (!before(seq, TCP_SKB_CB(tp->ooo_last_skb)->end_seq)) {  in tcp_data_queue_ofo()
5059 		parent = &tp->ooo_last_skb->rbnode;  in tcp_data_queue_ofo()
5060 		p = &parent->rb_right;  in tcp_data_queue_ofo()
5069 		if (before(seq, TCP_SKB_CB(skb1)->seq)) {  in tcp_data_queue_ofo()
5070 			p = &parent->rb_left;  in tcp_data_queue_ofo()
5073 		if (before(seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5074 			if (!after(end_seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5084 			if (after(seq, TCP_SKB_CB(skb1)->seq)) {  in tcp_data_queue_ofo()
5086 				tcp_dsack_set(sk, seq, TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5091 				rb_replace_node(&skb1->rbnode, &skb->rbnode,  in tcp_data_queue_ofo()
5092 						&tp->out_of_order_queue);  in tcp_data_queue_ofo()
5094 						 TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5095 						 TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5106 		p = &parent->rb_right;  in tcp_data_queue_ofo()
5110 	rb_link_node(&skb->rbnode, parent, p);  in tcp_data_queue_ofo()
5111 	rb_insert_color(&skb->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5116 		if (!after(end_seq, TCP_SKB_CB(skb1)->seq))  in tcp_data_queue_ofo()
5118 		if (before(end_seq, TCP_SKB_CB(skb1)->end_seq)) {  in tcp_data_queue_ofo()
5119 			tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5123 		rb_erase(&skb1->rbnode, &tp->out_of_order_queue);  in tcp_data_queue_ofo()
5124 		tcp_dsack_extend(sk, TCP_SKB_CB(skb1)->seq,  in tcp_data_queue_ofo()
5125 				 TCP_SKB_CB(skb1)->end_seq);  in tcp_data_queue_ofo()
5131 		tp->ooo_last_skb = skb;  in tcp_data_queue_ofo()
5152 	struct sk_buff *tail = skb_peek_tail(&sk->sk_receive_queue);  in tcp_queue_rcv()
5157 	tcp_rcv_nxt_update(tcp_sk(sk), TCP_SKB_CB(skb)->end_seq);  in tcp_queue_rcv()
5159 		__skb_queue_tail(&sk->sk_receive_queue, skb);  in tcp_queue_rcv()
5168 	int err = -ENOMEM;  in tcp_send_rcvq()
5181 	skb = alloc_skb_with_frags(size - data_len, data_len,  in tcp_send_rcvq()
5183 				   &err, sk->sk_allocation);  in tcp_send_rcvq()
5187 	skb_put(skb, size - data_len);  in tcp_send_rcvq()
5188 	skb->data_len = data_len;  in tcp_send_rcvq()
5189 	skb->len = size;  in tcp_send_rcvq()
5191 	if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {  in tcp_send_rcvq()
5196 	err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, size);  in tcp_send_rcvq()
5200 	TCP_SKB_CB(skb)->seq = tcp_sk(sk)->rcv_nxt;  in tcp_send_rcvq()
5201 	TCP_SKB_CB(skb)->end_seq = TCP_SKB_CB(skb)->seq + size;  in tcp_send_rcvq()
5202 	TCP_SKB_CB(skb)->ack_seq = tcp_sk(sk)->snd_una - 1;  in tcp_send_rcvq()
5219 	if (tcp_epollin_ready(sk, sk->sk_rcvlowat) || sock_flag(sk, SOCK_DONE))  in tcp_data_ready()
5220 		sk->sk_data_ready(sk);  in tcp_data_ready()
5238 	if (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq) {  in tcp_data_queue()
5243 	__skb_pull(skb, tcp_hdr(skb)->doff * 4);  in tcp_data_queue()
5246 	tp->rx_opt.dsack = 0;  in tcp_data_queue()
5252 	if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt) {  in tcp_data_queue()
5260 			if (!skb->len &&  in tcp_data_queue()
5261 			    (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN))  in tcp_data_queue()
5271 		if (tcp_try_rmem_schedule(sk, skb, skb->truesize)) {  in tcp_data_queue()
5273 			inet_csk(sk)->icsk_ack.pending |=  in tcp_data_queue()
5276 			sk->sk_data_ready(sk);  in tcp_data_queue()
5278 			if (skb_queue_len(&sk->sk_receive_queue) && skb->len) {  in tcp_data_queue()
5283 			sk_forced_mem_schedule(sk, skb->truesize);  in tcp_data_queue()
5287 		if (skb->len)  in tcp_data_queue()
5289 		if (TCP_SKB_CB(skb)->tcp_flags & TCPHDR_FIN)  in tcp_data_queue()
5292 		if (!RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in tcp_data_queue()
5298 			if (RB_EMPTY_ROOT(&tp->out_of_order_queue))  in tcp_data_queue()
5299 				inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_NOW;  in tcp_data_queue()
5302 		if (tp->rx_opt.num_sacks)  in tcp_data_queue()
5314 	if (!after(TCP_SKB_CB(skb)->end_seq, tp->rcv_nxt)) {  in tcp_data_queue()
5319 		tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);  in tcp_data_queue()
5330 	if (!before(TCP_SKB_CB(skb)->seq,  in tcp_data_queue()
5331 		    tp->rcv_nxt + tcp_receive_window(tp))) {  in tcp_data_queue()
5336 	if (before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_data_queue()
5338 		tcp_dsack_set(sk, TCP_SKB_CB(skb)->seq, tp->rcv_nxt);  in tcp_data_queue()
5341 		 * remembering D-SACK for its head made in previous line.  in tcp_data_queue()
5357 		return !skb_queue_is_last(list, skb) ? skb->next : NULL;  in tcp_skb_next()
5371 		rb_erase(&skb->rbnode, root);  in tcp_collapse_one()
5379 /* Insert skb into rb tree, ordered by TCP_SKB_CB(skb)->seq */
5382 	struct rb_node **p = &root->rb_node;  in tcp_rbtree_insert()
5389 		if (before(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb1)->seq))  in tcp_rbtree_insert()
5390 			p = &parent->rb_left;  in tcp_rbtree_insert()
5392 			p = &parent->rb_right;  in tcp_rbtree_insert()
5394 	rb_link_node(&skb->rbnode, parent, p);  in tcp_rbtree_insert()
5395 	rb_insert_color(&skb->rbnode, root);  in tcp_rbtree_insert()
5425 		if (!before(start, TCP_SKB_CB(skb)->end_seq)) {  in tcp_collapse()
5433 		 * - not SYN/FIN and  in tcp_collapse()
5434 		 * - bloated or contains data before "start" or  in tcp_collapse()
5437 		if (!(TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) &&  in tcp_collapse()
5438 		    (tcp_win_from_space(sk, skb->truesize) > skb->len ||  in tcp_collapse()
5439 		     before(TCP_SKB_CB(skb)->seq, start))) {  in tcp_collapse()
5446 		    TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(n)->seq) {  in tcp_collapse()
5453 		start = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse()
5456 	    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) ||  in tcp_collapse()
5463 		int copy = min_t(int, SKB_MAX_ORDER(0, 0), end - start);  in tcp_collapse()
5470 		memcpy(nskb->cb, skb->cb, sizeof(skb->cb));  in tcp_collapse()
5472 		TCP_SKB_CB(nskb)->seq = TCP_SKB_CB(nskb)->end_seq = start;  in tcp_collapse()
5482 			int offset = start - TCP_SKB_CB(skb)->seq;  in tcp_collapse()
5483 			int size = TCP_SKB_CB(skb)->end_seq - start;  in tcp_collapse()
5490 				TCP_SKB_CB(nskb)->end_seq += size;  in tcp_collapse()
5491 				copy -= size;  in tcp_collapse()
5494 			if (!before(start, TCP_SKB_CB(skb)->end_seq)) {  in tcp_collapse()
5499 				    (TCP_SKB_CB(skb)->tcp_flags & (TCPHDR_SYN | TCPHDR_FIN)) ||  in tcp_collapse()
5520 	skb = skb_rb_first(&tp->out_of_order_queue);  in tcp_collapse_ofo_queue()
5523 		tp->ooo_last_skb = skb_rb_last(&tp->out_of_order_queue);  in tcp_collapse_ofo_queue()
5526 	start = TCP_SKB_CB(skb)->seq;  in tcp_collapse_ofo_queue()
5527 	end = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse_ofo_queue()
5528 	range_truesize = skb->truesize;  in tcp_collapse_ofo_queue()
5537 		    after(TCP_SKB_CB(skb)->seq, end) ||  in tcp_collapse_ofo_queue()
5538 		    before(TCP_SKB_CB(skb)->end_seq, start)) {  in tcp_collapse_ofo_queue()
5540 			if (range_truesize != head->truesize ||  in tcp_collapse_ofo_queue()
5541 			    end - start >= SKB_WITH_OVERHEAD(PAGE_SIZE)) {  in tcp_collapse_ofo_queue()
5542 				tcp_collapse(sk, NULL, &tp->out_of_order_queue,  in tcp_collapse_ofo_queue()
5546 				if (sum_tiny > sk->sk_rcvbuf >> 3)  in tcp_collapse_ofo_queue()
5552 		range_truesize += skb->truesize;  in tcp_collapse_ofo_queue()
5553 		if (unlikely(before(TCP_SKB_CB(skb)->seq, start)))  in tcp_collapse_ofo_queue()
5554 			start = TCP_SKB_CB(skb)->seq;  in tcp_collapse_ofo_queue()
5555 		if (after(TCP_SKB_CB(skb)->end_seq, end))  in tcp_collapse_ofo_queue()
5556 			end = TCP_SKB_CB(skb)->end_seq;  in tcp_collapse_ofo_queue()
5561  * Clean the out-of-order queue to make room.
5580 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue))  in tcp_prune_ofo_queue()
5583 	goal = sk->sk_rcvbuf >> 3;  in tcp_prune_ofo_queue()
5584 	node = &tp->ooo_last_skb->rbnode;  in tcp_prune_ofo_queue()
5590 		if (after(TCP_SKB_CB(in_skb)->seq, TCP_SKB_CB(skb)->seq))  in tcp_prune_ofo_queue()
5594 		rb_erase(node, &tp->out_of_order_queue);  in tcp_prune_ofo_queue()
5595 		goal -= skb->truesize;  in tcp_prune_ofo_queue()
5597 		tp->ooo_last_skb = rb_to_skb(prev);  in tcp_prune_ofo_queue()
5599 			if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&  in tcp_prune_ofo_queue()
5602 			goal = sk->sk_rcvbuf >> 3;  in tcp_prune_ofo_queue()
5614 		if (tp->rx_opt.sack_ok)  in tcp_prune_ofo_queue()
5615 			tcp_sack_reset(&tp->rx_opt);  in tcp_prune_ofo_queue()
5633 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)  in tcp_prune_queue()
5638 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5642 	if (!skb_queue_empty(&sk->sk_receive_queue))  in tcp_prune_queue()
5643 		tcp_collapse(sk, &sk->sk_receive_queue, NULL,  in tcp_prune_queue()
5644 			     skb_peek(&sk->sk_receive_queue),  in tcp_prune_queue()
5646 			     tp->copied_seq, tp->rcv_nxt);  in tcp_prune_queue()
5648 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5656 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf)  in tcp_prune_queue()
5666 	tp->pred_flags = 0;  in tcp_prune_queue()
5667 	return -1;  in tcp_prune_queue()
5677 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)  in tcp_should_expand_sndbuf()
5689 			WRITE_ONCE(sk->sk_sndbuf, unused_mem);  in tcp_should_expand_sndbuf()
5711 		tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_new_space()
5714 	INDIRECT_CALL_1(sk->sk_write_space, sk_stream_write_space, sk);  in tcp_new_space()
5718  * 1) Freeing skbs in rtx queues (after tp->snd_una has advanced)
5719  * 2) Sent skbs from output queue (and thus advancing tp->snd_nxt)
5722  * 1) Space consumed in output/rtx queues is below sk->sk_sndbuf/2
5723  * 2) notsent amount (tp->write_seq - tp->snd_nxt) became
5731 	if (sk->sk_socket &&  in tcp_check_space()
5732 	    test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {  in tcp_check_space()
5734 		if (!test_bit(SOCK_NOSPACE, &sk->sk_socket->flags))  in tcp_check_space()
5754 	if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss &&  in __tcp_ack_snd_check()
5760 	    (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat ||  in __tcp_ack_snd_check()
5761 	     __tcp_select_window(sk) >= tp->rcv_wnd)) ||  in __tcp_ack_snd_check()
5764 	    /* Protocol state mandates a one-time immediate ACK */  in __tcp_ack_snd_check()
5765 	    inet_csk(sk)->icsk_ack.pending & ICSK_ACK_NOW) {  in __tcp_ack_snd_check()
5770 		    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_backlog_ack_defer)) {  in __tcp_ack_snd_check()
5771 			set_bit(TCP_ACK_DEFERRED, &sk->sk_tsq_flags);  in __tcp_ack_snd_check()
5779 	if (!ofo_possible || RB_EMPTY_ROOT(&tp->out_of_order_queue)) {  in __tcp_ack_snd_check()
5785 	    tp->compressed_ack >= READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_nr))  in __tcp_ack_snd_check()
5788 	if (tp->compressed_ack_rcv_nxt != tp->rcv_nxt) {  in __tcp_ack_snd_check()
5789 		tp->compressed_ack_rcv_nxt = tp->rcv_nxt;  in __tcp_ack_snd_check()
5790 		tp->dup_ack_counter = 0;  in __tcp_ack_snd_check()
5792 	if (tp->dup_ack_counter < TCP_FASTRETRANS_THRESH) {  in __tcp_ack_snd_check()
5793 		tp->dup_ack_counter++;  in __tcp_ack_snd_check()
5796 	tp->compressed_ack++;  in __tcp_ack_snd_check()
5797 	if (hrtimer_is_queued(&tp->compressed_ack_timer))  in __tcp_ack_snd_check()
5802 	rtt = tp->rcv_rtt_est.rtt_us;  in __tcp_ack_snd_check()
5803 	if (tp->srtt_us && tp->srtt_us < rtt)  in __tcp_ack_snd_check()
5804 		rtt = tp->srtt_us;  in __tcp_ack_snd_check()
5807 		      READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_delay_ns),  in __tcp_ack_snd_check()
5810 	hrtimer_start_range_ns(&tp->compressed_ack_timer, ns_to_ktime(delay),  in __tcp_ack_snd_check()
5811 			       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_comp_sack_slack_ns),  in __tcp_ack_snd_check()
5828  *	place. We handle URGent data wrong. We have to - as
5837 	u32 ptr = ntohs(th->urg_ptr);  in tcp_check_urg()
5839 	if (ptr && !READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_stdurg))  in tcp_check_urg()
5840 		ptr--;  in tcp_check_urg()
5841 	ptr += ntohl(th->seq);  in tcp_check_urg()
5844 	if (after(tp->copied_seq, ptr))  in tcp_check_urg()
5857 	if (before(ptr, tp->rcv_nxt))  in tcp_check_urg()
5861 	if (tp->urg_data && !after(ptr, tp->urg_seq))  in tcp_check_urg()
5869 	 * tp->copied_seq since we would read the last urgent byte again  in tcp_check_urg()
5882 	if (tp->urg_seq == tp->copied_seq && tp->urg_data &&  in tcp_check_urg()
5883 	    !sock_flag(sk, SOCK_URGINLINE) && tp->copied_seq != tp->rcv_nxt) {  in tcp_check_urg()
5884 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);  in tcp_check_urg()
5885 		tp->copied_seq++;  in tcp_check_urg()
5886 		if (skb && !before(tp->copied_seq, TCP_SKB_CB(skb)->end_seq)) {  in tcp_check_urg()
5887 			__skb_unlink(skb, &sk->sk_receive_queue);  in tcp_check_urg()
5892 	WRITE_ONCE(tp->urg_data, TCP_URG_NOTYET);  in tcp_check_urg()
5893 	WRITE_ONCE(tp->urg_seq, ptr);  in tcp_check_urg()
5896 	tp->pred_flags = 0;  in tcp_check_urg()
5904 	/* Check if we get a new urgent pointer - normally not. */  in tcp_urg()
5905 	if (unlikely(th->urg))  in tcp_urg()
5908 	/* Do we wait for any urgent data? - normally not... */  in tcp_urg()
5909 	if (unlikely(tp->urg_data == TCP_URG_NOTYET)) {  in tcp_urg()
5910 		u32 ptr = tp->urg_seq - ntohl(th->seq) + (th->doff * 4) -  in tcp_urg()
5911 			  th->syn;  in tcp_urg()
5914 		if (ptr < skb->len) {  in tcp_urg()
5918 			WRITE_ONCE(tp->urg_data, TCP_URG_VALID | tmp);  in tcp_urg()
5920 				sk->sk_data_ready(sk);  in tcp_urg()
5925 /* Accept RST for rcv_nxt - 1 after a FIN.
5937 	return unlikely(TCP_SKB_CB(skb)->seq == (tp->rcv_nxt - 1) &&  in tcp_reset_check()
5938 			(1 << sk->sk_state) & (TCPF_CLOSE_WAIT | TCPF_LAST_ACK |  in tcp_reset_check()
5953 	    tp->rx_opt.saw_tstamp &&  in tcp_validate_incoming()
5955 		if (!th->rst) {  in tcp_validate_incoming()
5956 			if (unlikely(th->syn))  in tcp_validate_incoming()
5961 						  &tp->last_oow_ack_time))  in tcp_validate_incoming()
5970 	reason = tcp_sequence(tp, TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq);  in tcp_validate_incoming()
5972 		/* RFC793, page 37: "In all states except SYN-SENT, all reset  in tcp_validate_incoming()
5973 		 * (RST) segments are validated by checking their SEQ-fields."  in tcp_validate_incoming()
5978 		if (!th->rst) {  in tcp_validate_incoming()
5979 			if (th->syn)  in tcp_validate_incoming()
5983 						  &tp->last_oow_ack_time))  in tcp_validate_incoming()
5992 	if (th->rst) {  in tcp_validate_incoming()
5993 		/* RFC 5961 3.2 (extend to match against (RCV.NXT - 1) after a  in tcp_validate_incoming()
5995 		 * If seq num matches RCV.NXT or (RCV.NXT - 1) after a FIN, or  in tcp_validate_incoming()
5996 		 * the right-most SACK block,  in tcp_validate_incoming()
6002 		if (TCP_SKB_CB(skb)->seq == tp->rcv_nxt ||  in tcp_validate_incoming()
6006 		if (tcp_is_sack(tp) && tp->rx_opt.num_sacks > 0) {  in tcp_validate_incoming()
6007 			struct tcp_sack_block *sp = &tp->selective_acks[0];  in tcp_validate_incoming()
6011 			for (this_sack = 1; this_sack < tp->rx_opt.num_sacks;  in tcp_validate_incoming()
6018 			if (TCP_SKB_CB(skb)->seq == max_sack)  in tcp_validate_incoming()
6022 		/* Disable TFO if RST is out-of-order  in tcp_validate_incoming()
6026 		if (tp->syn_fastopen && !tp->data_segs_in &&  in tcp_validate_incoming()
6027 		    sk->sk_state == TCP_ESTABLISHED)  in tcp_validate_incoming()
6039 	if (th->syn) {  in tcp_validate_incoming()
6040 		if (sk->sk_state == TCP_SYN_RECV && sk->sk_socket && th->ack &&  in tcp_validate_incoming()
6041 		    TCP_SKB_CB(skb)->seq + 1 == TCP_SKB_CB(skb)->end_seq &&  in tcp_validate_incoming()
6042 		    TCP_SKB_CB(skb)->seq + 1 == tp->rcv_nxt &&  in tcp_validate_incoming()
6043 		    TCP_SKB_CB(skb)->ack_seq == tp->snd_nxt)  in tcp_validate_incoming()
6074  *	- A zero window was announced from us - zero window probing
6076  *	- Out of order segments arrived.
6077  *	- Urgent data is expected.
6078  *	- There is no buffer space left
6079  *	- Unexpected TCP flags/window values/header lengths are received
6081  *	- Data is sent in both directions. Fast path only supports pure senders
6083  *	  value must stay constant)
6084  *	- Unexpected TCP option.
6095 	const struct tcphdr *th = (const struct tcphdr *)skb->data;  in tcp_rcv_established()
6097 	unsigned int len = skb->len;  in tcp_rcv_established()
6103 	if (unlikely(!rcu_access_pointer(sk->sk_rx_dst)))  in tcp_rcv_established()
6104 		inet_csk(sk)->icsk_af_ops->sk_rx_dst_set(sk, skb);  in tcp_rcv_established()
6120 	tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_established()
6124 	 *	'S' will always be tp->tcp_header_len >> 2  in tcp_rcv_established()
6131 	if ((tcp_flag_word(th) & TCP_HP_BITS) == tp->pred_flags &&  in tcp_rcv_established()
6132 	    TCP_SKB_CB(skb)->seq == tp->rcv_nxt &&  in tcp_rcv_established()
6133 	    !after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {  in tcp_rcv_established()
6134 		int tcp_header_len = tp->tcp_header_len;  in tcp_rcv_established()
6137 		 * is automatically equal to th->doff*4 due to pred_flags  in tcp_rcv_established()
6148 			if ((s32)(tp->rx_opt.rcv_tsval - tp->rx_opt.ts_recent) < 0)  in tcp_rcv_established()
6167 				    tp->rcv_nxt == tp->rcv_wup)  in tcp_rcv_established()
6180 				tp->rcv_rtt_last_tsecr = tp->rx_opt.rcv_tsecr;  in tcp_rcv_established()
6194 			if ((int)skb->truesize > sk->sk_forward_alloc)  in tcp_rcv_established()
6203 			    tp->rcv_nxt == tp->rcv_wup)  in tcp_rcv_established()
6217 			if (TCP_SKB_CB(skb)->ack_seq != tp->snd_una) {  in tcp_rcv_established()
6224 				tcp_update_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_established()
6237 	if (len < (th->doff << 2) || tcp_checksum_complete(skb))  in tcp_rcv_established()
6240 	if (!th->ack && !th->rst && !th->syn) {  in tcp_rcv_established()
6255 		reason = -reason;  in tcp_rcv_established()
6287 	icsk->icsk_af_ops->rebuild_header(sk);  in tcp_init_transfer()
6291 	 * Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been  in tcp_init_transfer()
6293 	 * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK  in tcp_init_transfer()
6296 	if (tp->total_retrans > 1 && tp->undo_marker)  in tcp_init_transfer()
6300 	tp->snd_cwnd_stamp = tcp_jiffies32;  in tcp_init_transfer()
6304 	if (!icsk->icsk_ca_initialized)  in tcp_init_transfer()
6316 	icsk->icsk_ack.lrcvtime = tcp_jiffies32;  in tcp_finish_connect()
6319 		icsk->icsk_af_ops->sk_rx_dst_set(sk, skb);  in tcp_finish_connect()
6329 	tp->lsndtime = tcp_jiffies32;  in tcp_finish_connect()
6334 	if (!tp->rx_opt.snd_wscale)  in tcp_finish_connect()
6335 		__tcp_fast_path_on(tp, tp->snd_wnd);  in tcp_finish_connect()
6337 		tp->pred_flags = 0;  in tcp_finish_connect()
6344 	struct sk_buff *data = tp->syn_data ? tcp_rtx_queue_head(sk) : NULL;  in tcp_rcv_fastopen_synack()
6345 	u16 mss = tp->rx_opt.mss_clamp, try_exp = 0;  in tcp_rcv_fastopen_synack()
6348 	if (mss == tp->rx_opt.user_mss) {  in tcp_rcv_fastopen_synack()
6351 		/* Get original SYNACK MSS value if user MSS sets mss_clamp */  in tcp_rcv_fastopen_synack()
6358 	if (!tp->syn_fastopen) {  in tcp_rcv_fastopen_synack()
6360 		cookie->len = -1;  in tcp_rcv_fastopen_synack()
6361 	} else if (tp->total_retrans) {  in tcp_rcv_fastopen_synack()
6362 		/* SYN timed out and the SYN-ACK neither has a cookie nor  in tcp_rcv_fastopen_synack()
6365 		 * SYN-data or the corresponding SYN-ACK was dropped.  in tcp_rcv_fastopen_synack()
6367 		syn_drop = (cookie->len < 0 && data);  in tcp_rcv_fastopen_synack()
6368 	} else if (cookie->len < 0 && !tp->syn_data) {  in tcp_rcv_fastopen_synack()
6373 		try_exp = tp->syn_fastopen_exp ? 2 : 1;  in tcp_rcv_fastopen_synack()
6379 		if (tp->total_retrans)  in tcp_rcv_fastopen_synack()
6380 			tp->fastopen_client_fail = TFO_SYN_RETRANSMITTED;  in tcp_rcv_fastopen_synack()
6382 			tp->fastopen_client_fail = TFO_DATA_NOT_ACKED;  in tcp_rcv_fastopen_synack()
6390 	tp->syn_data_acked = tp->syn_data;  in tcp_rcv_fastopen_synack()
6391 	if (tp->syn_data_acked) {  in tcp_rcv_fastopen_synack()
6393 		/* SYN-data is counted as two separate packets in tcp_ack() */  in tcp_rcv_fastopen_synack()
6394 		if (tp->delivered > 1)  in tcp_rcv_fastopen_synack()
6395 			--tp->delivered;  in tcp_rcv_fastopen_synack()
6407 		if (tp->syn_smc && !tp->rx_opt.smc_ok)  in smc_check_reset_syn()
6408 			tp->syn_smc = 0;  in smc_check_reset_syn()
6419 	 * spurious if the ACK's timestamp option echo value matches the  in tcp_try_undo_spurious_syn()
6422 	syn_stamp = tp->retrans_stamp;  in tcp_try_undo_spurious_syn()
6423 	if (tp->undo_marker && syn_stamp && tp->rx_opt.saw_tstamp &&  in tcp_try_undo_spurious_syn()
6424 	    syn_stamp == tp->rx_opt.rcv_tsecr)  in tcp_try_undo_spurious_syn()
6425 		tp->undo_marker = 0;  in tcp_try_undo_spurious_syn()
6433 	struct tcp_fastopen_cookie foc = { .len = -1 };  in tcp_rcv_synsent_state_process()
6434 	int saved_clamp = tp->rx_opt.mss_clamp;  in tcp_rcv_synsent_state_process()
6438 	tcp_parse_options(sock_net(sk), skb, &tp->rx_opt, 0, &foc);  in tcp_rcv_synsent_state_process()
6439 	if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr)  in tcp_rcv_synsent_state_process()
6440 		tp->rx_opt.rcv_tsecr -= tp->tsoffset;  in tcp_rcv_synsent_state_process()
6442 	if (th->ack) {  in tcp_rcv_synsent_state_process()
6444 		 * "If the state is SYN-SENT then  in tcp_rcv_synsent_state_process()
6451 		if (!after(TCP_SKB_CB(skb)->ack_seq, tp->snd_una) ||  in tcp_rcv_synsent_state_process()
6452 		    after(TCP_SKB_CB(skb)->ack_seq, tp->snd_nxt)) {  in tcp_rcv_synsent_state_process()
6454 			if (icsk->icsk_retransmits == 0)  in tcp_rcv_synsent_state_process()
6462 		if (tp->rx_opt.saw_tstamp && tp->rx_opt.rcv_tsecr &&  in tcp_rcv_synsent_state_process()
6463 		    !between(tp->rx_opt.rcv_tsecr, tp->retrans_stamp,  in tcp_rcv_synsent_state_process()
6479 		if (th->rst) {  in tcp_rcv_synsent_state_process()
6491 		 *                                        --ANK(990513)  in tcp_rcv_synsent_state_process()
6493 		if (!th->syn) {  in tcp_rcv_synsent_state_process()
6506 		tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_synsent_state_process()
6513 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);  in tcp_rcv_synsent_state_process()
6514 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;  in tcp_rcv_synsent_state_process()
6519 		tp->snd_wnd = ntohs(th->window);  in tcp_rcv_synsent_state_process()
6521 		if (!tp->rx_opt.wscale_ok) {  in tcp_rcv_synsent_state_process()
6522 			tp->rx_opt.snd_wscale = tp->rx_opt.rcv_wscale = 0;  in tcp_rcv_synsent_state_process()
6523 			WRITE_ONCE(tp->window_clamp,  in tcp_rcv_synsent_state_process()
6524 				   min(tp->window_clamp, 65535U));  in tcp_rcv_synsent_state_process()
6527 		if (tp->rx_opt.saw_tstamp) {  in tcp_rcv_synsent_state_process()
6528 			tp->rx_opt.tstamp_ok	   = 1;  in tcp_rcv_synsent_state_process()
6529 			tp->tcp_header_len =  in tcp_rcv_synsent_state_process()
6531 			tp->advmss	    -= TCPOLEN_TSTAMP_ALIGNED;  in tcp_rcv_synsent_state_process()
6534 			tp->tcp_header_len = sizeof(struct tcphdr);  in tcp_rcv_synsent_state_process()
6537 		tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_rcv_synsent_state_process()
6541 		 * Change state from SYN-SENT only after copied_seq  in tcp_rcv_synsent_state_process()
6543 		WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_synsent_state_process()
6551 		fastopen_fail = (tp->syn_fastopen || tp->syn_data) &&  in tcp_rcv_synsent_state_process()
6555 			sk->sk_state_change(sk);  in tcp_rcv_synsent_state_process()
6559 			return -1;  in tcp_rcv_synsent_state_process()
6560 		if (sk->sk_write_pending ||  in tcp_rcv_synsent_state_process()
6561 		    READ_ONCE(icsk->icsk_accept_queue.rskq_defer_accept) ||  in tcp_rcv_synsent_state_process()
6568 			 * to stand against the temptation 8)     --ANK  in tcp_rcv_synsent_state_process()
6577 		return -1;  in tcp_rcv_synsent_state_process()
6582 	if (th->rst) {  in tcp_rcv_synsent_state_process()
6593 	if (tp->rx_opt.ts_recent_stamp && tp->rx_opt.saw_tstamp &&  in tcp_rcv_synsent_state_process()
6594 	    tcp_paws_reject(&tp->rx_opt, 0)) {  in tcp_rcv_synsent_state_process()
6598 	if (th->syn) {  in tcp_rcv_synsent_state_process()
6606 		ao = rcu_dereference_protected(tp->ao_info,  in tcp_rcv_synsent_state_process()
6609 			WRITE_ONCE(ao->risn, th->seq);  in tcp_rcv_synsent_state_process()
6610 			ao->rcv_sne = 0;  in tcp_rcv_synsent_state_process()
6615 		if (tp->rx_opt.saw_tstamp) {  in tcp_rcv_synsent_state_process()
6616 			tp->rx_opt.tstamp_ok = 1;  in tcp_rcv_synsent_state_process()
6618 			tp->tcp_header_len =  in tcp_rcv_synsent_state_process()
6621 			tp->tcp_header_len = sizeof(struct tcphdr);  in tcp_rcv_synsent_state_process()
6624 		WRITE_ONCE(tp->rcv_nxt, TCP_SKB_CB(skb)->seq + 1);  in tcp_rcv_synsent_state_process()
6625 		WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_synsent_state_process()
6626 		tp->rcv_wup = TCP_SKB_CB(skb)->seq + 1;  in tcp_rcv_synsent_state_process()
6631 		tp->snd_wnd    = ntohs(th->window);  in tcp_rcv_synsent_state_process()
6632 		tp->snd_wl1    = TCP_SKB_CB(skb)->seq;  in tcp_rcv_synsent_state_process()
6633 		tp->max_window = tp->snd_wnd;  in tcp_rcv_synsent_state_process()
6638 		tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_rcv_synsent_state_process()
6654 		return -1;  in tcp_rcv_synsent_state_process()
6664 	tcp_clear_options(&tp->rx_opt);  in tcp_rcv_synsent_state_process()
6665 	tp->rx_opt.mss_clamp = saved_clamp;  in tcp_rcv_synsent_state_process()
6670 	tcp_clear_options(&tp->rx_opt);  in tcp_rcv_synsent_state_process()
6671 	tp->rx_opt.mss_clamp = saved_clamp;  in tcp_rcv_synsent_state_process()
6684 	if (inet_csk(sk)->icsk_ca_state == TCP_CA_Loss && !tp->packets_out)  in tcp_rcv_synrecv_state_fastopen()
6688 	inet_csk(sk)->icsk_retransmits = 0;  in tcp_rcv_synrecv_state_fastopen()
6702 	req = rcu_dereference_protected(tp->fastopen_rsk,  in tcp_rcv_synrecv_state_fastopen()
6706 	/* Re-arm the timer because data may have been sent out.  in tcp_rcv_synrecv_state_fastopen()
6710 	 * (TFO) - we could try to be more aggressive and  in tcp_rcv_synrecv_state_fastopen()
6734 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6740 		if (th->ack)  in tcp_rcv_state_process()
6743 		if (th->rst) {  in tcp_rcv_state_process()
6747 		if (th->syn) {  in tcp_rcv_state_process()
6748 			if (th->fin) {  in tcp_rcv_state_process()
6757 			icsk->icsk_af_ops->conn_request(sk, skb);  in tcp_rcv_state_process()
6768 		tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_state_process()
6782 	tp->rx_opt.saw_tstamp = 0;  in tcp_rcv_state_process()
6783 	req = rcu_dereference_protected(tp->fastopen_rsk,  in tcp_rcv_state_process()
6788 		WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&  in tcp_rcv_state_process()
6789 		    sk->sk_state != TCP_FIN_WAIT1);  in tcp_rcv_state_process()
6797 	if (!th->ack && !th->rst && !th->syn) {  in tcp_rcv_state_process()
6810 		if (sk->sk_state == TCP_SYN_RECV) {  in tcp_rcv_state_process()
6814 			return -reason;  in tcp_rcv_state_process()
6819 			reason = -reason;  in tcp_rcv_state_process()
6824 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6826 		tp->delivered++; /* SYN-ACK delivery isn't tracked in tcp_ack */  in tcp_rcv_state_process()
6827 		if (!tp->srtt_us)  in tcp_rcv_state_process()
6834 			tp->retrans_stamp = 0;  in tcp_rcv_state_process()
6837 			WRITE_ONCE(tp->copied_seq, tp->rcv_nxt);  in tcp_rcv_state_process()
6842 		sk->sk_state_change(sk);  in tcp_rcv_state_process()
6846 		 * sk->sk_sleep == NULL and sk->sk_socket == NULL.  in tcp_rcv_state_process()
6848 		if (sk->sk_socket)  in tcp_rcv_state_process()
6851 		tp->snd_una = TCP_SKB_CB(skb)->ack_seq;  in tcp_rcv_state_process()
6852 		tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale;  in tcp_rcv_state_process()
6853 		tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);  in tcp_rcv_state_process()
6855 		if (tp->rx_opt.tstamp_ok)  in tcp_rcv_state_process()
6856 			tp->advmss -= TCPOLEN_TSTAMP_ALIGNED;  in tcp_rcv_state_process()
6858 		if (!inet_csk(sk)->icsk_ca_ops->cong_control)  in tcp_rcv_state_process()
6862 		tp->lsndtime = tcp_jiffies32;  in tcp_rcv_state_process()
6866 		if (sk->sk_shutdown & SEND_SHUTDOWN)  in tcp_rcv_state_process()
6876 		if (tp->snd_una != tp->write_seq)  in tcp_rcv_state_process()
6880 		WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | SEND_SHUTDOWN);  in tcp_rcv_state_process()
6886 			sk->sk_state_change(sk);  in tcp_rcv_state_process()
6890 		if (READ_ONCE(tp->linger2) < 0) {  in tcp_rcv_state_process()
6895 		if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_rcv_state_process()
6896 		    after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {  in tcp_rcv_state_process()
6898 			if (tp->syn_fastopen && th->fin)  in tcp_rcv_state_process()
6907 			inet_csk_reset_keepalive_timer(sk, tmo - TCP_TIMEWAIT_LEN);  in tcp_rcv_state_process()
6908 		} else if (th->fin || sock_owned_by_user(sk)) {  in tcp_rcv_state_process()
6924 		if (tp->snd_una == tp->write_seq) {  in tcp_rcv_state_process()
6931 		if (tp->snd_una == tp->write_seq) {  in tcp_rcv_state_process()
6943 	switch (sk->sk_state) {  in tcp_rcv_state_process()
6947 		if (!before(TCP_SKB_CB(skb)->seq, tp->rcv_nxt)) {  in tcp_rcv_state_process()
6962 		if (sk->sk_shutdown & RCV_SHUTDOWN) {  in tcp_rcv_state_process()
6963 			if (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq &&  in tcp_rcv_state_process()
6964 			    after(TCP_SKB_CB(skb)->end_seq - th->fin, tp->rcv_nxt)) {  in tcp_rcv_state_process()
6977 	/* tcp_data could move socket to TIME-WAIT */  in tcp_rcv_state_process()
6978 	if (sk->sk_state != TCP_CLOSE) {  in tcp_rcv_state_process()
7001 				    &ireq->ir_rmt_addr, port);  in pr_drop_req()
7005 				    &ireq->ir_v6_rmt_addr, port);  in pr_drop_req()
7021  * Exception: At least one of the reserved bits of the TCP header (th->res1) is
7033 	bool th_ecn = th->ece && th->cwr;  in tcp_ecn_create_request()
7040 	ect = !INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield);  in tcp_ecn_create_request()
7042 	ecn_ok = READ_ONCE(net->ipv4.sysctl_tcp_ecn) || ecn_ok_dst;  in tcp_ecn_create_request()
7044 	if (((!ect || th->res1) && ecn_ok) || tcp_ca_needs_ecn(listen_sk) ||  in tcp_ecn_create_request()
7047 		inet_rsk(req)->ecn_ok = 1;  in tcp_ecn_create_request()
7056 	req->rsk_rcv_wnd = 0;		/* So that tcp_send_synack() knows! */  in tcp_openreq_init()
7057 	tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;  in tcp_openreq_init()
7058 	tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;  in tcp_openreq_init()
7059 	tcp_rsk(req)->snt_synack = 0;  in tcp_openreq_init()
7060 	tcp_rsk(req)->last_oow_ack_time = 0;  in tcp_openreq_init()
7061 	req->mss = rx_opt->mss_clamp;  in tcp_openreq_init()
7062 	req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;  in tcp_openreq_init()
7063 	ireq->tstamp_ok = rx_opt->tstamp_ok;  in tcp_openreq_init()
7064 	ireq->sack_ok = rx_opt->sack_ok;  in tcp_openreq_init()
7065 	ireq->snd_wscale = rx_opt->snd_wscale;  in tcp_openreq_init()
7066 	ireq->wscale_ok = rx_opt->wscale_ok;  in tcp_openreq_init()
7067 	ireq->acked = 0;  in tcp_openreq_init()
7068 	ireq->ecn_ok = 0;  in tcp_openreq_init()
7069 	ireq->ir_rmt_port = tcp_hdr(skb)->source;  in tcp_openreq_init()
7070 	ireq->ir_num = ntohs(tcp_hdr(skb)->dest);  in tcp_openreq_init()
7071 	ireq->ir_mark = inet_request_mark(sk, skb);  in tcp_openreq_init()
7073 	ireq->smc_ok = rx_opt->smc_ok && !(tcp_sk(sk)->smc_hs_congested &&  in tcp_openreq_init()
7074 			tcp_sk(sk)->smc_hs_congested(sk));  in tcp_openreq_init()
7083 	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;  in tcp_syn_flood_action()
7089 	syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies);  in tcp_syn_flood_action()
7100 	if (!READ_ONCE(queue->synflood_warned) && syncookies != 2 &&  in tcp_syn_flood_action()
7101 	    xchg(&queue->synflood_warned, 1) == 0) {  in tcp_syn_flood_action()
7102 		if (IS_ENABLED(CONFIG_IPV6) && sk->sk_family == AF_INET6) {  in tcp_syn_flood_action()
7105 					sk->sk_num, msg);  in tcp_syn_flood_action()
7108 					proto, &sk->sk_rcv_saddr,  in tcp_syn_flood_action()
7109 					sk->sk_num, msg);  in tcp_syn_flood_action()
7120 	if (tcp_sk(sk)->save_syn) {  in tcp_reqsk_record_syn()
7126 		if (tcp_sk(sk)->save_syn == 2) {  /* Save full header. */  in tcp_reqsk_record_syn()
7138 			saved_syn->mac_hdrlen = mac_hdrlen;  in tcp_reqsk_record_syn()
7139 			saved_syn->network_hdrlen = skb_network_header_len(skb);  in tcp_reqsk_record_syn()
7140 			saved_syn->tcp_hdrlen = tcp_hdrlen(skb);  in tcp_reqsk_record_syn()
7141 			memcpy(saved_syn->data, base, len);  in tcp_reqsk_record_syn()
7142 			req->saved_syn = saved_syn;  in tcp_reqsk_record_syn()
7147 /* If a SYN cookie is required and supported, returns a clamped MSS value to be
7157 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) != 2 &&  in tcp_get_syncookie_mss()
7161 	if (!tcp_syn_flood_action(sk, rsk_ops->slab_name))  in tcp_get_syncookie_mss()
7169 	mss = tcp_parse_mss_option(th, tp->rx_opt.user_mss);  in tcp_get_syncookie_mss()
7171 		mss = af_ops->mss_clamp;  in tcp_get_syncookie_mss()
7181 	struct tcp_fastopen_cookie foc = { .len = -1 };  in tcp_conn_request()
7205 		syncookies = READ_ONCE(net->ipv4.sysctl_tcp_syncookies);  in tcp_conn_request()
7209 							   rsk_ops->slab_name);  in tcp_conn_request()
7224 	req->syncookie = want_cookie;  in tcp_conn_request()
7225 	tcp_rsk(req)->af_specific = af_ops;  in tcp_conn_request()
7226 	tcp_rsk(req)->ts_off = 0;  in tcp_conn_request()
7227 	tcp_rsk(req)->req_usec_ts = false;  in tcp_conn_request()
7229 	tcp_rsk(req)->is_mptcp = 0;  in tcp_conn_request()
7233 	tmp_opt.mss_clamp = af_ops->mss_clamp;  in tcp_conn_request()
7234 	tmp_opt.user_mss  = tp->rx_opt.user_mss;  in tcp_conn_request()
7246 	inet_rsk(req)->no_srccheck = inet_test_bit(TRANSPARENT, sk);  in tcp_conn_request()
7249 	inet_rsk(req)->ir_iif = inet_request_bound_dev_if(sk, skb);  in tcp_conn_request()
7251 	dst = af_ops->route_req(sk, skb, &fl, req, isn);  in tcp_conn_request()
7256 		tcp_rsk(req)->req_usec_ts = dst_tcp_usec_ts(dst);  in tcp_conn_request()
7257 		tcp_rsk(req)->ts_off = af_ops->init_ts_off(net, skb);  in tcp_conn_request()
7260 		int max_syn_backlog = READ_ONCE(net->ipv4.sysctl_max_syn_backlog);  in tcp_conn_request()
7264 		    (max_syn_backlog - inet_csk_reqsk_queue_len(sk) <  in tcp_conn_request()
7274 			pr_drop_req(req, ntohs(tcp_hdr(skb)->source),  in tcp_conn_request()
7275 				    rsk_ops->family);  in tcp_conn_request()
7279 		isn = af_ops->init_seq(skb);  in tcp_conn_request()
7285 		isn = cookie_init_sequence(af_ops, sk, skb, &req->mss);  in tcp_conn_request()
7287 			inet_rsk(req)->ecn_ok = 0;  in tcp_conn_request()
7294 		tcp_rsk(req)->used_tcp_ao = true;  in tcp_conn_request()
7295 		tcp_rsk(req)->ao_rcv_next = aoh->keyid;  in tcp_conn_request()
7296 		tcp_rsk(req)->ao_keyid = aoh->rnext_keyid;  in tcp_conn_request()
7299 		tcp_rsk(req)->used_tcp_ao = false;  in tcp_conn_request()
7302 	tcp_rsk(req)->snt_isn = isn;  in tcp_conn_request()
7303 	tcp_rsk(req)->txhash = net_tx_rndhash();  in tcp_conn_request()
7304 	tcp_rsk(req)->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;  in tcp_conn_request()
7312 		af_ops->send_synack(fastopen_sk, dst, &fl, req,  in tcp_conn_request()
7321 		sk->sk_data_ready(sk);  in tcp_conn_request()
7325 		tcp_rsk(req)->tfo_listener = false;  in tcp_conn_request()
7327 			req->timeout = tcp_timeout_init((struct sock *)req);  in tcp_conn_request()
7329 								    req->timeout))) {  in tcp_conn_request()
7335 		af_ops->send_synack(sk, dst, &fl, req, &foc,  in tcp_conn_request()