2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
34 #include <sys/cdefs.h>
36 #include "opt_inet6.h"
39 #include <sys/param.h>
40 #include <sys/kernel.h>
43 #include <sys/mutex.h>
44 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/sysctl.h>
49 #include <sys/systm.h>
52 #include <net/route.h>
53 #include <net/rss_config.h>
55 #include <net/netisr.h>
57 #include <netinet/in.h>
58 #include <netinet/in_kdtrace.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_rss.h>
61 #include <netinet/in_systm.h>
63 #include <netinet6/in6_pcb.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/tcp.h>
67 #include <netinet/tcp_fsm.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcp_log_buf.h>
71 #include <netinet/tcp_seq.h>
72 #include <netinet/cc/cc.h>
74 #include <netinet6/tcp6_var.h>
76 #include <netinet/tcpip.h>
79 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin,
80 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
81 &tcp_persmin, 0, sysctl_msec_to_ticks, "I",
82 "minimum persistence interval");
85 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax,
86 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
87 &tcp_persmax, 0, sysctl_msec_to_ticks, "I",
88 "maximum persistence interval");
91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit,
92 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
93 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I",
94 "time to establish connection");
97 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle,
98 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
99 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I",
100 "time before keepalive probes begin");
103 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl,
104 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
105 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I",
106 "time between keepalive probes");
109 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime,
110 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
111 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
112 "Time before a delayed ACK is sent");
114 VNET_DEFINE(int, tcp_msl);
115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl,
116 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_VNET,
117 &VNET_NAME(tcp_msl), 0, sysctl_msec_to_ticks, "I",
118 "Maximum segment lifetime");
120 int tcp_rexmit_initial;
121 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_initial,
122 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
123 &tcp_rexmit_initial, 0, sysctl_msec_to_ticks, "I",
124 "Initial Retransmission Timeout");
127 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min,
128 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
129 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
130 "Minimum Retransmission Timeout");
133 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop,
134 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
135 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
136 "Retransmission Timer Slop");
138 VNET_DEFINE(int, tcp_always_keepalive) = 1;
139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_VNET|CTLFLAG_RW,
140 &VNET_NAME(tcp_always_keepalive) , 0,
141 "Assume SO_KEEPALIVE on all TCP connections");
143 int tcp_fast_finwait2_recycle = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
145 &tcp_fast_finwait2_recycle, 0,
146 "Recycle closed FIN_WAIT_2 connections faster");
148 int tcp_finwait2_timeout;
149 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout,
150 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
151 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I",
152 "FIN-WAIT2 timeout");
154 int tcp_keepcnt = TCPTV_KEEPCNT;
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
156 "Number of keepalive probes to send");
158 /* max idle probes */
159 int tcp_maxpersistidle;
161 int tcp_rexmit_drop_options = 0;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
163 &tcp_rexmit_drop_options, 0,
164 "Drop TCP options from 3rd and later retransmitted SYN");
166 int tcp_maxunacktime = TCPTV_MAXUNACKTIME;
167 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxunacktime,
168 CTLTYPE_INT|CTLFLAG_RW | CTLFLAG_NEEDGIANT,
169 &tcp_maxunacktime, 0, sysctl_msec_to_ticks, "I",
170 "Maximum time (in ms) that a session can linger without making progress");
172 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
174 CTLFLAG_RW|CTLFLAG_VNET,
175 &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
176 "Path MTU Discovery Black Hole Detection Enabled");
179 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
181 CTLFLAG_RW|CTLFLAG_VNET,
182 &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
183 "Path MTU Discovery Black Hole Detection lowered MSS");
187 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
189 CTLFLAG_RW|CTLFLAG_VNET,
190 &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
191 "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
195 static int per_cpu_timers = 1;
197 static int per_cpu_timers = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
200 &per_cpu_timers , 0, "run tcp timers on all cpus");
203 sysctl_net_inet_tcp_retries(SYSCTL_HANDLER_ARGS)
208 error = sysctl_handle_int(oidp, &new, 0, req);
209 if (error == 0 && req->newptr) {
210 if ((new < 1) || (new > TCP_MAXRXTSHIFT))
218 VNET_DEFINE(int, tcp_retries) = TCP_MAXRXTSHIFT;
219 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, retries,
220 CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_retries), 0, sysctl_net_inet_tcp_retries, "I",
222 "maximum number of consecutive timer based retransmissions");
225 * Map the given inp to a CPU id.
227 * This queries RSS if it's compiled in, else it defaults to the current
231 inp_to_cpuid(struct inpcb *inp)
235 if (per_cpu_timers) {
237 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
238 if (cpuid == NETISR_CPUID_NONE)
239 return (curcpu); /* XXX */
244 * We don't have a flowid -> cpuid mapping, so cheat and
245 * just map unknown cpuids to curcpu. Not the best, but
246 * apparently better than defaulting to swi 0.
248 cpuid = inp->inp_flowid % (mp_maxid + 1);
249 if (! CPU_ABSENT(cpuid))
257 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
258 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
260 int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
263 * TCP timer processing.
265 * Each connection has 5 timers associated with it, which can be scheduled
266 * simultaneously. They all are serviced by one callout tcp_timer_enter().
267 * This function executes the next timer via tcp_timersw[] vector. Each
268 * timer is supposed to return 'true' unless the connection was destroyed.
269 * In the former case tcp_timer_enter() will schedule callout for next timer.
272 typedef bool tcp_timer_t(struct tcpcb *);
273 static tcp_timer_t tcp_timer_delack;
274 static tcp_timer_t tcp_timer_2msl;
275 static tcp_timer_t tcp_timer_keep;
276 static tcp_timer_t tcp_timer_persist;
277 static tcp_timer_t tcp_timer_rexmt;
279 static tcp_timer_t * const tcp_timersw[TT_N] = {
280 [TT_DELACK] = tcp_timer_delack,
281 [TT_REXMT] = tcp_timer_rexmt,
282 [TT_PERSIST] = tcp_timer_persist,
283 [TT_KEEP] = tcp_timer_keep,
284 [TT_2MSL] = tcp_timer_2msl,
288 * tcp_output_locked() s a timer specific variation of call to tcp_output(),
289 * see tcp_var.h for the rest. It handles drop request from advanced stacks,
290 * but keeps tcpcb locked unless tcp_drop() destroyed it.
291 * Returns true if tcpcb is valid and locked.
294 tcp_output_locked(struct tcpcb *tp)
298 INP_WLOCK_ASSERT(tptoinpcb(tp));
300 if ((rv = tp->t_fb->tfb_tcp_output(tp)) < 0) {
301 KASSERT(tp->t_fb->tfb_flags & TCP_FUNC_OUTPUT_CANDROP,
302 ("TCP stack %s requested tcp_drop(%p)",
303 tp->t_fb->tfb_tcp_block_name, tp));
304 tp = tcp_drop(tp, rv);
311 tcp_timer_delack(struct tcpcb *tp)
313 struct epoch_tracker et;
314 #if defined(INVARIANTS) || defined(VIMAGE)
315 struct inpcb *inp = tptoinpcb(tp);
319 INP_WLOCK_ASSERT(inp);
321 CURVNET_SET(inp->inp_vnet);
322 tp->t_flags |= TF_ACKNOW;
323 TCPSTAT_INC(tcps_delack);
325 rv = tcp_output_locked(tp);
333 tcp_timer_2msl(struct tcpcb *tp)
335 struct inpcb *inp = tptoinpcb(tp);
338 INP_WLOCK_ASSERT(inp);
340 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
341 CURVNET_SET(inp->inp_vnet);
342 tcp_log_end_status(tp, TCP_EI_STATUS_2MSL);
343 tcp_free_sackholes(tp);
345 * 2 MSL timeout in shutdown went off. If we're closed but
346 * still waiting for peer to close and connection has been idle
347 * too long delete connection control block. Otherwise, check
350 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
351 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
352 * Ignore fact that there were recent incoming segments.
354 * XXXGL: check if inp_socket shall always be !NULL here?
356 if (tp->t_state == TCPS_TIME_WAIT) {
358 } else if (tp->t_state == TCPS_FIN_WAIT_2 &&
359 tcp_fast_finwait2_recycle && inp->inp_socket &&
360 (inp->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
361 TCPSTAT_INC(tcps_finwait2_drops);
364 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp))
365 tcp_timer_activate(tp, TT_2MSL, TP_KEEPINTVL(tp));
370 struct epoch_tracker et;
382 tcp_timer_keep(struct tcpcb *tp)
384 struct epoch_tracker et;
385 struct inpcb *inp = tptoinpcb(tp);
386 struct tcptemp *t_template;
388 INP_WLOCK_ASSERT(inp);
390 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
391 CURVNET_SET(inp->inp_vnet);
393 * Because we don't regularly reset the keepalive callout in
394 * the ESTABLISHED state, it may be that we don't actually need
395 * to send a keepalive yet. If that occurs, schedule another
396 * call for the next time the keepalive timer might expire.
398 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
401 idletime = ticks - tp->t_rcvtime;
402 if (idletime < TP_KEEPIDLE(tp)) {
403 tcp_timer_activate(tp, TT_KEEP,
404 TP_KEEPIDLE(tp) - idletime);
411 * Keep-alive timer went off; send something
412 * or drop connection if idle for too long.
414 TCPSTAT_INC(tcps_keeptimeo);
415 if (tp->t_state < TCPS_ESTABLISHED)
417 if ((V_tcp_always_keepalive ||
418 inp->inp_socket->so_options & SO_KEEPALIVE) &&
419 tp->t_state <= TCPS_CLOSING) {
420 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
423 * Send a packet designed to force a response
424 * if the peer is up and reachable:
425 * either an ACK if the connection is still alive,
426 * or an RST if the peer has closed the connection
427 * due to timeout or reboot.
428 * Using sequence number tp->snd_una-1
429 * causes the transmitted zero-length segment
430 * to lie outside the receive window;
431 * by the protocol spec, this requires the
432 * correspondent TCP to respond.
434 TCPSTAT_INC(tcps_keepprobe);
435 t_template = tcpip_maketemplate(inp);
438 tcp_respond(tp, t_template->tt_ipgen,
439 &t_template->tt_t, (struct mbuf *)NULL,
440 tp->rcv_nxt, tp->snd_una - 1, 0);
442 free(t_template, M_TEMP);
444 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINTVL(tp));
446 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
452 TCPSTAT_INC(tcps_keepdrops);
454 tcp_log_end_status(tp, TCP_EI_STATUS_KEEP_MAX);
455 tp = tcp_drop(tp, ETIMEDOUT);
463 * Has this session exceeded the maximum time without seeing a substantive
464 * acknowledgement? If so, return true; otherwise false.
467 tcp_maxunacktime_check(struct tcpcb *tp)
470 /* Are we tracking this timer for this session? */
471 if (TP_MAXUNACKTIME(tp) == 0)
474 /* Do we have a current measurement. */
475 if (tp->t_acktime == 0)
478 /* Are we within the acceptable range? */
479 if (TSTMP_GT(TP_MAXUNACKTIME(tp) + tp->t_acktime, (u_int)ticks))
482 /* We exceeded the timer. */
483 TCPSTAT_INC(tcps_progdrops);
488 tcp_timer_persist(struct tcpcb *tp)
490 struct epoch_tracker et;
491 #if defined(INVARIANTS) || defined(VIMAGE)
492 struct inpcb *inp = tptoinpcb(tp);
496 INP_WLOCK_ASSERT(inp);
498 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
499 CURVNET_SET(inp->inp_vnet);
501 * Persistence timer into zero window.
502 * Force a byte to be output, if possible.
504 TCPSTAT_INC(tcps_persisttimeo);
506 * Hack: if the peer is dead/unreachable, we do not
507 * time out if the window is closed. After a full
508 * backoff, drop the connection if the idle time
509 * (no responses to probes) reaches the maximum
510 * backoff that we would use if retransmitting.
511 * Also, drop the connection if we haven't been making
514 progdrop = tcp_maxunacktime_check(tp);
515 if (progdrop || (tp->t_rxtshift >= V_tcp_retries &&
516 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
517 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff))) {
519 TCPSTAT_INC(tcps_persistdrop);
520 tcp_log_end_status(tp, TCP_EI_STATUS_PERSIST_MAX);
524 * If the user has closed the socket then drop a persisting
525 * connection after a much reduced timeout.
527 if (tp->t_state > TCPS_CLOSE_WAIT &&
528 (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
529 TCPSTAT_INC(tcps_persistdrop);
530 tcp_log_end_status(tp, TCP_EI_STATUS_PERSIST_MAX);
534 tp->t_flags |= TF_FORCEDATA;
536 if ((rv = tcp_output_locked(tp)))
537 tp->t_flags &= ~TF_FORCEDATA;
545 tp = tcp_drop(tp, ETIMEDOUT);
553 tcp_timer_rexmt(struct tcpcb *tp)
555 struct epoch_tracker et;
556 struct inpcb *inp = tptoinpcb(tp);
560 INP_WLOCK_ASSERT(inp);
562 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
563 CURVNET_SET(inp->inp_vnet);
564 tcp_free_sackholes(tp);
565 if (tp->t_fb->tfb_tcp_rexmit_tmr) {
566 /* The stack has a timer action too. */
567 (*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
570 * Retransmission timer went off. Message has not
571 * been acked within retransmit interval. Back off
572 * to a longer retransmit interval and retransmit one segment.
574 * If we've either exceeded the maximum number of retransmissions,
575 * or we've gone long enough without making progress, then drop
578 if (++tp->t_rxtshift > V_tcp_retries || tcp_maxunacktime_check(tp)) {
579 if (tp->t_rxtshift > V_tcp_retries)
580 TCPSTAT_INC(tcps_timeoutdrop);
581 tp->t_rxtshift = V_tcp_retries;
582 tcp_log_end_status(tp, TCP_EI_STATUS_RETRAN);
584 tp = tcp_drop(tp, ETIMEDOUT);
590 if (tp->t_state == TCPS_SYN_SENT) {
592 * If the SYN was retransmitted, indicate CWND to be
593 * limited to 1 segment in cc_conn_init().
596 } else if (tp->t_rxtshift == 1) {
598 * first retransmit; record ssthresh and cwnd so they can
599 * be recovered if this turns out to be a "bad" retransmit.
600 * A retransmit is considered "bad" if an ACK for this
601 * segment is received within RTT/2 interval; the assumption
602 * here is that the ACK was already in flight. See
603 * "On Estimating End-to-End Network Path Properties" by
604 * Allman and Paxson for more details.
606 tp->snd_cwnd_prev = tp->snd_cwnd;
607 tp->snd_ssthresh_prev = tp->snd_ssthresh;
608 tp->snd_recover_prev = tp->snd_recover;
609 if (IN_FASTRECOVERY(tp->t_flags))
610 tp->t_flags |= TF_WASFRECOVERY;
612 tp->t_flags &= ~TF_WASFRECOVERY;
613 if (IN_CONGRECOVERY(tp->t_flags))
614 tp->t_flags |= TF_WASCRECOVERY;
616 tp->t_flags &= ~TF_WASCRECOVERY;
617 if ((tp->t_flags & TF_RCVD_TSTMP) == 0)
618 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
619 /* In the event that we've negotiated timestamps
620 * badrxtwin will be set to the value that we set
621 * the retransmitted packet's to_tsval to by tcp_output
623 tp->t_flags |= TF_PREVVALID;
625 tp->t_flags &= ~TF_PREVVALID;
626 TCPSTAT_INC(tcps_rexmttimeo);
627 if ((tp->t_state == TCPS_SYN_SENT) ||
628 (tp->t_state == TCPS_SYN_RECEIVED))
629 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
631 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
632 TCPT_RANGESET(tp->t_rxtcur, rexmt,
633 tp->t_rttmin, TCPTV_REXMTMAX);
636 * We enter the path for PLMTUD if connection is established or, if
637 * connection is FIN_WAIT_1 status, reason for the last is that if
638 * amount of data we send is very small, we could send it in couple of
639 * packets and process straight to FIN. In that case we won't catch
643 isipv6 = (inp->inp_vflag & INP_IPV6) ? true : false;
647 if (((V_tcp_pmtud_blackhole_detect == 1) ||
648 (V_tcp_pmtud_blackhole_detect == 2 && !isipv6) ||
649 (V_tcp_pmtud_blackhole_detect == 3 && isipv6)) &&
650 ((tp->t_state == TCPS_ESTABLISHED) ||
651 (tp->t_state == TCPS_FIN_WAIT_1))) {
652 if (tp->t_rxtshift == 1) {
654 * We enter blackhole detection after the first
655 * unsuccessful timer based retransmission.
656 * Then we reduce up to two times the MSS, each
657 * candidate giving two tries of retransmissions.
658 * But we give a candidate only two tries, if it
659 * actually reduces the MSS.
661 tp->t_blackhole_enter = 2;
662 tp->t_blackhole_exit = tp->t_blackhole_enter;
665 if (tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss)
666 tp->t_blackhole_exit += 2;
667 if (tp->t_maxseg > V_tcp_v6mssdflt &&
668 V_tcp_v6pmtud_blackhole_mss > V_tcp_v6mssdflt)
669 tp->t_blackhole_exit += 2;
673 if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss)
674 tp->t_blackhole_exit += 2;
675 if (tp->t_maxseg > V_tcp_mssdflt &&
676 V_tcp_pmtud_blackhole_mss > V_tcp_mssdflt)
677 tp->t_blackhole_exit += 2;
681 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
682 (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
683 (tp->t_rxtshift >= tp->t_blackhole_enter &&
684 tp->t_rxtshift < tp->t_blackhole_exit &&
685 (tp->t_rxtshift - tp->t_blackhole_enter) % 2 == 0)) {
687 * Enter Path MTU Black-hole Detection mechanism:
688 * - Disable Path MTU Discovery (IP "DF" bit).
689 * - Reduce MTU to lower value than what we
690 * negotiated with peer.
692 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
693 /* Record that we may have found a black hole. */
694 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
695 /* Keep track of previous MSS. */
696 tp->t_pmtud_saved_maxseg = tp->t_maxseg;
700 * Reduce the MSS to blackhole value or to the default
701 * in an attempt to retransmit.
705 tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss &&
706 V_tcp_v6pmtud_blackhole_mss > V_tcp_v6mssdflt) {
707 /* Use the sysctl tuneable blackhole MSS. */
708 tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
709 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
711 /* Use the default MSS. */
712 tp->t_maxseg = V_tcp_v6mssdflt;
714 * Disable Path MTU Discovery when we switch to
717 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
718 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
721 #if defined(INET6) && defined(INET)
725 if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss &&
726 V_tcp_pmtud_blackhole_mss > V_tcp_mssdflt) {
727 /* Use the sysctl tuneable blackhole MSS. */
728 tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
729 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
731 /* Use the default MSS. */
732 tp->t_maxseg = V_tcp_mssdflt;
734 * Disable Path MTU Discovery when we switch to
737 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
738 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
742 * Reset the slow-start flight size
743 * as it may depend on the new MSS.
745 if (CC_ALGO(tp)->conn_init != NULL)
746 CC_ALGO(tp)->conn_init(&tp->t_ccv);
749 * If further retransmissions are still unsuccessful
750 * with a lowered MTU, maybe this isn't a blackhole and
751 * we restore the previous MSS and blackhole detection
754 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
755 (tp->t_rxtshift >= tp->t_blackhole_exit)) {
756 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
757 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
758 tp->t_maxseg = tp->t_pmtud_saved_maxseg;
759 TCPSTAT_INC(tcps_pmtud_blackhole_failed);
761 * Reset the slow-start flight size as it
762 * may depend on the new MSS.
764 if (CC_ALGO(tp)->conn_init != NULL)
765 CC_ALGO(tp)->conn_init(&tp->t_ccv);
771 * Disable RFC1323 and SACK if we haven't got any response to
772 * our third SYN to work-around some broken terminal servers
773 * (most of which have hopefully been retired) that have bad VJ
774 * header compression code which trashes TCP segments containing
775 * unknown-to-them TCP options.
777 if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
778 (tp->t_rxtshift == 3))
779 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
781 * If we backed off this far, notify the L3 protocol that we're having
782 * connection problems.
784 if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
786 if ((inp->inp_vflag & INP_IPV6) != 0)
792 tp->snd_nxt = tp->snd_una;
793 tp->snd_recover = tp->snd_max;
795 * Force a segment to be sent.
797 tp->t_flags |= TF_ACKNOW;
799 * If timing a segment in this window, stop the timer.
803 cc_cong_signal(tp, NULL, CC_RTO);
805 rv = tcp_output_locked(tp);
813 tcp_bblog_timer(struct tcpcb *tp, tt_which which, tt_what what, uint32_t ticks)
815 struct tcp_log_buffer *lgb;
818 INP_WLOCK_ASSERT(tptoinpcb(tp));
819 if (tcp_bblogging_on(tp))
820 lgb = tcp_log_event(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0,
821 NULL, false, NULL, NULL, 0, NULL);
825 lgb->tlb_flex1 = (what << 8) | which;
826 if (what == TT_STARTING) {
827 /* Convert ticks to ms and store it in tlb_flex2. */
829 lgb->tlb_flex2 = ticks;
831 ms = (((uint64_t)ticks * 1000) + (hz - 1)) / hz;
833 lgb->tlb_flex2 = UINT32_MAX;
835 lgb->tlb_flex2 = (uint32_t)ms;
841 static inline tt_which
842 tcp_timer_next(struct tcpcb *tp, sbintime_t *precision)
845 sbintime_t after, before;
847 for (i = 0, rv = TT_N, after = before = SBT_MAX; i < TT_N; i++) {
848 if (tp->t_timers[i] < after) {
849 after = tp->t_timers[i];
852 before = MIN(before, tp->t_timers[i] + tp->t_precisions[i]);
854 if (precision != NULL)
855 *precision = before - after;
861 tcp_timer_enter(void *xtp)
863 struct tcpcb *tp = xtp;
864 struct inpcb *inp = tptoinpcb(tp);
865 sbintime_t precision;
869 INP_WLOCK_ASSERT(inp);
870 MPASS((curthread->td_pflags & TDP_INTCPCALLOUT) == 0);
872 curthread->td_pflags |= TDP_INTCPCALLOUT;
874 which = tcp_timer_next(tp, NULL);
876 tp->t_timers[which] = SBT_MAX;
877 tp->t_precisions[which] = 0;
879 tcp_bblog_timer(tp, which, TT_PROCESSING, 0);
880 tp_valid = tcp_timersw[which](tp);
882 tcp_bblog_timer(tp, which, TT_PROCESSED, 0);
883 if ((which = tcp_timer_next(tp, &precision)) != TT_N) {
884 callout_reset_sbt_on(&tp->t_callout,
885 tp->t_timers[which], precision, tcp_timer_enter,
886 tp, inp_to_cpuid(inp), C_ABSOLUTE);
891 curthread->td_pflags &= ~TDP_INTCPCALLOUT;
895 * Activate or stop (delta == 0) a TCP timer.
898 tcp_timer_activate(struct tcpcb *tp, tt_which which, u_int delta)
900 struct inpcb *inp = tptoinpcb(tp);
901 sbintime_t precision;
905 if (tp->t_flags & TF_TOE)
909 INP_WLOCK_ASSERT(inp);
913 callout_when(tick_sbt * delta, 0, C_HARDCLOCK,
914 &tp->t_timers[which], &tp->t_precisions[which]);
917 tp->t_timers[which] = SBT_MAX;
919 tcp_bblog_timer(tp, which, what, delta);
921 if ((which = tcp_timer_next(tp, &precision)) != TT_N)
922 callout_reset_sbt_on(&tp->t_callout, tp->t_timers[which],
923 precision, tcp_timer_enter, tp, inp_to_cpuid(inp),
926 callout_stop(&tp->t_callout);
930 tcp_timer_active(struct tcpcb *tp, tt_which which)
933 INP_WLOCK_ASSERT(tptoinpcb(tp));
935 return (tp->t_timers[which] != SBT_MAX);
939 * Stop all timers associated with tcpcb.
941 * Called only on tcpcb destruction. The tcpcb shall already be dropped from
942 * the pcb lookup database and socket is not losing the last reference.
944 * XXXGL: unfortunately our callout(9) is not able to fully stop a locked
945 * callout even when only two threads are involved: the callout itself and the
946 * thread that does callout_stop(). See where softclock_call_cc() swaps the
947 * callwheel lock to callout lock and then checks cc_exec_cancel(). This is
948 * the race window. If it happens, the tcp_timer_enter() won't be executed,
949 * however pcb lock will be locked and released, hence we can't free memory.
950 * Until callout(9) is improved, just keep retrying. In my profiling I've seen
951 * such event happening less than 1 time per hour with 20-30 Gbit/s of traffic.
954 tcp_timer_stop(struct tcpcb *tp)
956 struct inpcb *inp = tptoinpcb(tp);
958 INP_WLOCK_ASSERT(inp);
960 if (curthread->td_pflags & TDP_INTCPCALLOUT) {
961 int stopped __diagused;
963 stopped = callout_stop(&tp->t_callout);
965 } else while(__predict_false(callout_stop(&tp->t_callout) == 0)) {
967 kern_yield(PRI_UNCHANGED);