1 // SPDX-License-Identifier: GPL-2.0-only
12  *		Corey Minyard <wf-[email protected]>
13  *		Florian La Roche, <[email protected]-sb.de>
34 	user_timeout = READ_ONCE(icsk->icsk_user_timeout);  in tcp_clamp_rto_to_user_timeout()
36 		return icsk->icsk_rto;  in tcp_clamp_rto_to_user_timeout()
38 	elapsed = tcp_time_stamp_ts(tp) - tp->retrans_stamp;  in tcp_clamp_rto_to_user_timeout()
39 	if (tp->tcp_usec_ts)  in tcp_clamp_rto_to_user_timeout()
42 	remaining = user_timeout - elapsed;  in tcp_clamp_rto_to_user_timeout()
46 	return min_t(u32, icsk->icsk_rto, msecs_to_jiffies(remaining));  in tcp_clamp_rto_to_user_timeout()
55 	user_timeout = READ_ONCE(icsk->icsk_user_timeout);  in tcp_clamp_probe0_to_user_timeout()
56 	if (!user_timeout || !icsk->icsk_probes_tstamp)  in tcp_clamp_probe0_to_user_timeout()
59 	elapsed = tcp_jiffies32 - icsk->icsk_probes_tstamp;  in tcp_clamp_probe0_to_user_timeout()
62 	remaining = msecs_to_jiffies(user_timeout) - elapsed;  in tcp_clamp_probe0_to_user_timeout()
69  *  tcp_write_err() - close socket and save error info
77 	tcp_done_with_error(sk, READ_ONCE(sk->sk_err_soft) ? : ETIMEDOUT);  in tcp_write_err()
82  *  tcp_out_of_resources() - Close socket if out of resources
112 	if ((s32)(tcp_jiffies32 - tp->lsndtime) > 2*TCP_RTO_MAX || !do_reset)  in tcp_out_of_resources()
116 	if (READ_ONCE(sk->sk_err_soft))  in tcp_out_of_resources()
122 		if ((s32)(tcp_jiffies32 - tp->lsndtime) <= TCP_TIMEWAIT_LEN ||  in tcp_out_of_resources()
124 		    (!tp->snd_wnd && !tp->packets_out))  in tcp_out_of_resources()
144  *  tcp_orphan_retries() - Returns maximal number of retries on an orphaned socket
150 	int retries = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_orphan_retries); /* May be zero. */  in tcp_orphan_retries()  local
153 	if (READ_ONCE(sk->sk_err_soft) && !alive)  in tcp_orphan_retries()
154 		retries = 0;  in tcp_orphan_retries()
157 	 * number of retries. 8 corresponds to >100 seconds with minimal  in tcp_orphan_retries()
159 	if (retries == 0 && alive)  in tcp_orphan_retries()
160 		retries = 8;  in tcp_orphan_retries()
161 	return retries;  in tcp_orphan_retries()
170 	if (!READ_ONCE(net->ipv4.sysctl_tcp_mtu_probing))  in tcp_mtu_probing()
173 	if (!icsk->icsk_mtup.enabled) {  in tcp_mtu_probing()
174 		icsk->icsk_mtup.enabled = 1;  in tcp_mtu_probing()
175 		icsk->icsk_mtup.probe_timestamp = tcp_jiffies32;  in tcp_mtu_probing()
177 		mss = tcp_mtu_to_mss(sk, icsk->icsk_mtup.search_low) >> 1;  in tcp_mtu_probing()
178 		mss = min(READ_ONCE(net->ipv4.sysctl_tcp_base_mss), mss);  in tcp_mtu_probing()
179 		mss = max(mss, READ_ONCE(net->ipv4.sysctl_tcp_mtu_probe_floor));  in tcp_mtu_probing()
180 		mss = max(mss, READ_ONCE(net->ipv4.sysctl_tcp_min_snd_mss));  in tcp_mtu_probing()
181 		icsk->icsk_mtup.search_low = tcp_mss_to_mtu(sk, mss);  in tcp_mtu_probing()
183 	tcp_sync_mss(sk, icsk->icsk_pmtu_cookie);  in tcp_mtu_probing()
194 		timeout = ((2 << boundary) - 1) * rto_base;  in tcp_model_timeout()
196 		timeout = ((2 << linear_backoff_thresh) - 1) * rto_base +  in tcp_model_timeout()
197 			(boundary - linear_backoff_thresh) * TCP_RTO_MAX;  in tcp_model_timeout()
201  *  retransmits_timed_out() - returns true if this connection has timed out
203  *  @boundary: max number of retransmissions
210  * after "boundary" unsuccessful, exponentially backed-off
220 	if (!inet_csk(sk)->icsk_retransmits)  in retransmits_timed_out()
223 	start_ts = tp->retrans_stamp;  in retransmits_timed_out()
227 		if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))  in retransmits_timed_out()
232 	if (tp->tcp_usec_ts) {  in retransmits_timed_out()
234 		delta = tp->tcp_mstamp - start_ts + jiffies_to_usecs(1);  in retransmits_timed_out()
235 		return (s32)(delta - timeout * USEC_PER_MSEC) >= 0;  in retransmits_timed_out()
237 	return (s32)(tcp_time_stamp_ts(tp) - start_ts - timeout) >= 0;  in retransmits_timed_out()
249 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {  in tcp_write_timeout()
250 		if (icsk->icsk_retransmits)  in tcp_write_timeout()
253 		retry_until = READ_ONCE(icsk->icsk_syn_retries) ? :  in tcp_write_timeout()
254 			READ_ONCE(net->ipv4.sysctl_tcp_syn_retries);  in tcp_write_timeout()
257 		if (sk->sk_state == TCP_SYN_SENT)  in tcp_write_timeout()
258 			max_retransmits += READ_ONCE(net->ipv4.sysctl_tcp_syn_linear_timeouts);  in tcp_write_timeout()
260 		expired = icsk->icsk_retransmits >= max_retransmits;  in tcp_write_timeout()
262 		if (retransmits_timed_out(sk, READ_ONCE(net->ipv4.sysctl_tcp_retries1), 0)) {  in tcp_write_timeout()
269 		retry_until = READ_ONCE(net->ipv4.sysctl_tcp_retries2);  in tcp_write_timeout()
271 			const bool alive = icsk->icsk_rto < TCP_RTO_MAX;  in tcp_write_timeout()
283 						READ_ONCE(icsk->icsk_user_timeout));  in tcp_write_timeout()
289 				  icsk->icsk_retransmits,  in tcp_write_timeout()
290 				  icsk->icsk_rto, (int)expired);  in tcp_write_timeout()
299 		tp->timeout_rehash++;  in tcp_write_timeout()
312 	if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))  in tcp_delack_timer_handler()
316 	if (tp->compressed_ack) {  in tcp_delack_timer_handler()
322 	if (!(icsk->icsk_ack.pending & ICSK_ACK_TIMER))  in tcp_delack_timer_handler()
325 	if (time_after(icsk->icsk_ack.timeout, jiffies)) {  in tcp_delack_timer_handler()
326 		sk_reset_timer(sk, &icsk->icsk_delack_timer, icsk->icsk_ack.timeout);  in tcp_delack_timer_handler()
329 	icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;  in tcp_delack_timer_handler()
334 			icsk->icsk_ack.ato = min_t(u32, icsk->icsk_ack.ato << 1, icsk->icsk_rto);  in tcp_delack_timer_handler()
340 			icsk->icsk_ack.ato      = TCP_ATO_MIN;  in tcp_delack_timer_handler()
350  *  tcp_delack_timer() - The TCP delayed ACK timeout handler
362 	struct sock *sk = &icsk->icsk_inet.sk;  in tcp_delack_timer()
368 	if (!(smp_load_acquire(&icsk->icsk_ack.pending) & ICSK_ACK_TIMER) &&  in tcp_delack_timer()
369 	    !READ_ONCE(tcp_sk(sk)->compressed_ack))  in tcp_delack_timer()
378 		if (!test_and_set_bit(TCP_DELACK_TIMER_DEFERRED, &sk->sk_tsq_flags))  in tcp_delack_timer()
393 	if (tp->packets_out || !skb) {  in tcp_probe_timer()
394 		icsk->icsk_probes_out = 0;  in tcp_probe_timer()
395 		icsk->icsk_probes_tstamp = 0;  in tcp_probe_timer()
407 	if (!icsk->icsk_probes_tstamp) {  in tcp_probe_timer()
408 		icsk->icsk_probes_tstamp = tcp_jiffies32;  in tcp_probe_timer()
410 		u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);  in tcp_probe_timer()
413 		    (s32)(tcp_jiffies32 - icsk->icsk_probes_tstamp) >=  in tcp_probe_timer()
417 	max_probes = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_retries2);  in tcp_probe_timer()
422 		if (!alive && icsk->icsk_backoff >= max_probes)  in tcp_probe_timer()
428 	if (icsk->icsk_probes_out >= max_probes) {  in tcp_probe_timer()
441 	if (!icsk->icsk_retransmits) {  in tcp_update_rto_stats()
442 		tp->total_rto_recoveries++;  in tcp_update_rto_stats()
443 		tp->rto_stamp = tcp_time_stamp_ms(tp);  in tcp_update_rto_stats()
445 	icsk->icsk_retransmits++;  in tcp_update_rto_stats()
446 	tp->total_rto++;  in tcp_update_rto_stats()
459 	req->rsk_ops->syn_ack_timeout(req);  in tcp_fastopen_synack_timer()
464 	max_retries = READ_ONCE(icsk->icsk_syn_retries) ? :  in tcp_fastopen_synack_timer()
465 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_synack_retries) + 1;  in tcp_fastopen_synack_timer()
467 	if (req->num_timeout >= max_retries) {  in tcp_fastopen_synack_timer()
472 	if (icsk->icsk_retransmits == 1)  in tcp_fastopen_synack_timer()
474 	/* XXX (TFO) - Unlike regular SYN-ACK retransmit, we ignore error  in tcp_fastopen_synack_timer()
480 	req->num_timeout++;  in tcp_fastopen_synack_timer()
482 	if (!tp->retrans_stamp)  in tcp_fastopen_synack_timer()
483 		tp->retrans_stamp = tcp_time_stamp_ts(tp);  in tcp_fastopen_synack_timer()
485 			  req->timeout << req->num_timeout, TCP_RTO_MAX);  in tcp_fastopen_synack_timer()
493 	u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);  in tcp_rtx_probe0_timed_out()
508 	 * and tp->rcv_tstamp might very well have been written recently.  in tcp_rtx_probe0_timed_out()
511 	rcv_delta = icsk->icsk_timeout - tp->rcv_tstamp;  in tcp_rtx_probe0_timed_out()
519  *  tcp_retransmit_timer() - The TCP retransmit timeout handler
537 	req = rcu_dereference_protected(tp->fastopen_rsk,  in tcp_retransmit_timer()
540 		WARN_ON_ONCE(sk->sk_state != TCP_SYN_RECV &&  in tcp_retransmit_timer()
541 			     sk->sk_state != TCP_FIN_WAIT1);  in tcp_retransmit_timer()
543 		/* Before we receive ACK to our SYN-ACK don't retransmit  in tcp_retransmit_timer()
549 	if (!tp->packets_out)  in tcp_retransmit_timer()
556 	if (!tp->snd_wnd && !sock_flag(sk, SOCK_DEAD) &&  in tcp_retransmit_timer()
557 	    !((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV))) {  in tcp_retransmit_timer()
566 		rtx_delta = tcp_time_stamp_ts(tp) - (tp->retrans_stamp ?:   in tcp_retransmit_timer()
567 				tcp_skb_timestamp_ts(tp->tcp_usec_ts, skb));  in tcp_retransmit_timer()
568 		if (tp->tcp_usec_ts)  in tcp_retransmit_timer()
571 		if (sk->sk_family == AF_INET) {  in tcp_retransmit_timer()
572 …	net_dbg_ratelimited("Probing zero-window on %pI4:%u/%u, seq=%u:%u, recv %ums ago, lasting %ums\n",  in tcp_retransmit_timer()
573 				&inet->inet_daddr, ntohs(inet->inet_dport),  in tcp_retransmit_timer()
574 				inet->inet_num, tp->snd_una, tp->snd_nxt,  in tcp_retransmit_timer()
575 				jiffies_to_msecs(jiffies - tp->rcv_tstamp),  in tcp_retransmit_timer()
579 		else if (sk->sk_family == AF_INET6) {  in tcp_retransmit_timer()
580 …	net_dbg_ratelimited("Probing zero-window on %pI6:%u/%u, seq=%u:%u, recv %ums ago, lasting %ums\n",  in tcp_retransmit_timer()
581 				&sk->sk_v6_daddr, ntohs(inet->inet_dport),  in tcp_retransmit_timer()
582 				inet->inet_num, tp->snd_una, tp->snd_nxt,  in tcp_retransmit_timer()
583 				jiffies_to_msecs(jiffies - tp->rcv_tstamp),  in tcp_retransmit_timer()
601 	if (icsk->icsk_retransmits == 0) {  in tcp_retransmit_timer()
604 		if (icsk->icsk_ca_state == TCP_CA_Recovery) {  in tcp_retransmit_timer()
609 		} else if (icsk->icsk_ca_state == TCP_CA_Loss) {  in tcp_retransmit_timer()
611 		} else if ((icsk->icsk_ca_state == TCP_CA_Disorder) ||  in tcp_retransmit_timer()
612 			   tp->sacked_out) {  in tcp_retransmit_timer()
659 	 * linear-timeout retransmissions into a black hole  in tcp_retransmit_timer()
661 	if (sk->sk_state == TCP_ESTABLISHED &&  in tcp_retransmit_timer()
662 	    (tp->thin_lto || READ_ONCE(net->ipv4.sysctl_tcp_thin_linear_timeouts)) &&  in tcp_retransmit_timer()
664 	    icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {  in tcp_retransmit_timer()
665 		icsk->icsk_backoff = 0;  in tcp_retransmit_timer()
666 		icsk->icsk_rto = clamp(__tcp_set_rto(tp),  in tcp_retransmit_timer()
669 	} else if (sk->sk_state != TCP_SYN_SENT ||  in tcp_retransmit_timer()
670 		   tp->total_rto >  in tcp_retransmit_timer()
671 		   READ_ONCE(net->ipv4.sysctl_tcp_syn_linear_timeouts)) {  in tcp_retransmit_timer()
675 		icsk->icsk_backoff++;  in tcp_retransmit_timer()
676 		icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);  in tcp_retransmit_timer()
680 	if (retransmits_timed_out(sk, READ_ONCE(net->ipv4.sysctl_tcp_retries1) + 1, 0))  in tcp_retransmit_timer()
686 /* Called with bottom-half processing disabled.
693 	if (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN)) ||  in tcp_write_timer_handler()
694 	    !icsk->icsk_pending)  in tcp_write_timer_handler()
697 	if (time_after(icsk->icsk_timeout, jiffies)) {  in tcp_write_timer_handler()
698 		sk_reset_timer(sk, &icsk->icsk_retransmit_timer, icsk->icsk_timeout);  in tcp_write_timer_handler()
703 	event = icsk->icsk_pending;  in tcp_write_timer_handler()
713 		smp_store_release(&icsk->icsk_pending, 0);  in tcp_write_timer_handler()
717 		smp_store_release(&icsk->icsk_pending, 0);  in tcp_write_timer_handler()
727 	struct sock *sk = &icsk->icsk_inet.sk;  in tcp_write_timer()
730 	if (!smp_load_acquire(&icsk->icsk_pending))  in tcp_write_timer()
738 		if (!test_and_set_bit(TCP_WRITE_TIMER_DEFERRED, &sk->sk_tsq_flags))  in tcp_write_timer()
748 	struct net *net = read_pnet(&inet_rsk(req)->ireq_net);  in tcp_syn_ack_timeout()
756 	if ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))  in tcp_set_keepalive()
782 	if (sk->sk_state == TCP_LISTEN) {  in tcp_keepalive_timer()
788 	if (sk->sk_state == TCP_FIN_WAIT2 && sock_flag(sk, SOCK_DEAD)) {  in tcp_keepalive_timer()
789 		if (READ_ONCE(tp->linger2) >= 0) {  in tcp_keepalive_timer()
790 			const int tmo = tcp_fin_time(sk) - TCP_TIMEWAIT_LEN;  in tcp_keepalive_timer()
802 	    ((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)))  in tcp_keepalive_timer()
808 	if (tp->packets_out || !tcp_write_queue_empty(sk))  in tcp_keepalive_timer()
814 		u32 user_timeout = READ_ONCE(icsk->icsk_user_timeout);  in tcp_keepalive_timer()
821 		    icsk->icsk_probes_out > 0) ||  in tcp_keepalive_timer()
823 		    icsk->icsk_probes_out >= keepalive_probes(tp))) {  in tcp_keepalive_timer()
830 			icsk->icsk_probes_out++;  in tcp_keepalive_timer()
839 		/* It is tp->rcv_tstamp + keepalive_time_when(tp) */  in tcp_keepalive_timer()
840 		elapsed = keepalive_time_when(tp) - elapsed;  in tcp_keepalive_timer()
862 		if (tp->compressed_ack) {  in tcp_compressed_ack_kick()
864 			 * subtract one from tp->compressed_ack to keep  in tcp_compressed_ack_kick()
867 			tp->compressed_ack--;  in tcp_compressed_ack_kick()
873 				      &sk->sk_tsq_flags))  in tcp_compressed_ack_kick()
887 	hrtimer_init(&tcp_sk(sk)->pacing_timer, CLOCK_MONOTONIC,  in tcp_init_xmit_timers()
889 	tcp_sk(sk)->pacing_timer.function = tcp_pace_kick;  in tcp_init_xmit_timers()
891 	hrtimer_init(&tcp_sk(sk)->compressed_ack_timer, CLOCK_MONOTONIC,  in tcp_init_xmit_timers()
893 	tcp_sk(sk)->compressed_ack_timer.function = tcp_compressed_ack_kick;  in tcp_init_xmit_timers()