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>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
42 #include <sys/param.h>
43 #include <sys/kernel.h>
46 #include <sys/mutex.h>
47 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
55 #include <net/route.h>
56 #include <net/rss_config.h>
58 #include <net/netisr.h>
60 #include <netinet/in.h>
61 #include <netinet/in_kdtrace.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_rss.h>
64 #include <netinet/in_systm.h>
66 #include <netinet6/in6_pcb.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/tcp.h>
70 #include <netinet/tcp_fsm.h>
71 #include <netinet/tcp_log_buf.h>
72 #include <netinet/tcp_timer.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet/tcp_seq.h>
75 #include <netinet/cc/cc.h>
77 #include <netinet6/tcp6_var.h>
79 #include <netinet/tcpip.h>
81 #include <netinet/tcp_debug.h>
85 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
86 &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
89 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
90 &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
93 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
94 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
97 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
98 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
101 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
102 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
105 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
106 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
107 "Time before a delayed ACK is sent");
110 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
111 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
114 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
115 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
116 "Minimum Retransmission Timeout");
119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
120 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
121 "Retransmission Timer Slop");
123 int tcp_always_keepalive = 1;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
125 &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
127 int tcp_fast_finwait2_recycle = 0;
128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
129 &tcp_fast_finwait2_recycle, 0,
130 "Recycle closed FIN_WAIT_2 connections faster");
132 int tcp_finwait2_timeout;
133 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
134 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
136 int tcp_keepcnt = TCPTV_KEEPCNT;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
138 "Number of keepalive probes to send");
140 /* max idle probes */
141 int tcp_maxpersistidle;
143 int tcp_rexmit_drop_options = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
145 &tcp_rexmit_drop_options, 0,
146 "Drop TCP options from 3rd and later retransmitted SYN");
148 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
150 CTLFLAG_RW|CTLFLAG_VNET,
151 &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
152 "Path MTU Discovery Black Hole Detection Enabled");
155 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
157 CTLFLAG_RW|CTLFLAG_VNET,
158 &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
159 "Path MTU Discovery Black Hole Detection lowered MSS");
163 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
165 CTLFLAG_RW|CTLFLAG_VNET,
166 &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
167 "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
171 static int per_cpu_timers = 1;
173 static int per_cpu_timers = 0;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
176 &per_cpu_timers , 0, "run tcp timers on all cpus");
179 * Map the given inp to a CPU id.
181 * This queries RSS if it's compiled in, else it defaults to the current
185 inp_to_cpuid(struct inpcb *inp)
190 if (per_cpu_timers) {
191 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
192 if (cpuid == NETISR_CPUID_NONE)
193 return (curcpu); /* XXX */
198 /* Legacy, pre-RSS behaviour */
199 if (per_cpu_timers) {
201 * We don't have a flowid -> cpuid mapping, so cheat and
202 * just map unknown cpuids to curcpu. Not the best, but
203 * apparently better than defaulting to swi 0.
205 cpuid = inp->inp_flowid % (mp_maxid + 1);
206 if (! CPU_ABSENT(cpuid))
211 /* Default for RSS and non-RSS - cpuid 0 */
218 * Tcp protocol timeout routine called every 500 ms.
219 * Updates timestamps used for TCP
220 * causes finite state machine actions if timers expire.
225 VNET_ITERATOR_DECL(vnet_iter);
227 VNET_LIST_RLOCK_NOSLEEP();
228 VNET_FOREACH(vnet_iter) {
229 CURVNET_SET(vnet_iter);
230 (void) tcp_tw_2msl_scan(0);
233 VNET_LIST_RUNLOCK_NOSLEEP();
236 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
237 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
239 int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
242 * TCP timer processing.
246 tcp_timer_delack(void *xtp)
248 struct tcpcb *tp = xtp;
250 CURVNET_SET(tp->t_vnet);
253 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
255 if (callout_pending(&tp->t_timers->tt_delack) ||
256 !callout_active(&tp->t_timers->tt_delack)) {
261 callout_deactivate(&tp->t_timers->tt_delack);
262 if ((inp->inp_flags & INP_DROPPED) != 0) {
267 tp->t_flags |= TF_ACKNOW;
268 TCPSTAT_INC(tcps_delack);
269 (void) tp->t_fb->tfb_tcp_output(tp);
275 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
277 if (inp && tp != NULL)
282 tcp_timer_2msl(void *xtp)
284 struct tcpcb *tp = xtp;
286 struct epoch_tracker et;
287 CURVNET_SET(tp->t_vnet);
291 ostate = tp->t_state;
294 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
296 tcp_free_sackholes(tp);
297 if (callout_pending(&tp->t_timers->tt_2msl) ||
298 !callout_active(&tp->t_timers->tt_2msl)) {
299 INP_WUNLOCK(tp->t_inpcb);
303 callout_deactivate(&tp->t_timers->tt_2msl);
304 if ((inp->inp_flags & INP_DROPPED) != 0) {
309 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
310 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
312 * 2 MSL timeout in shutdown went off. If we're closed but
313 * still waiting for peer to close and connection has been idle
314 * too long delete connection control block. Otherwise, check
317 * If in TIME_WAIT state just ignore as this timeout is handled in
318 * tcp_tw_2msl_scan().
320 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
321 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
322 * Ignore fact that there were recent incoming segments.
324 if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
329 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
330 tp->t_inpcb && tp->t_inpcb->inp_socket &&
331 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
332 TCPSTAT_INC(tcps_finwait2_drops);
333 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
334 tcp_inpinfo_lock_del(inp, tp);
337 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
339 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
340 tcp_inpinfo_lock_del(inp, tp);
343 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
344 callout_reset(&tp->t_timers->tt_2msl,
345 TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
347 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
348 tcp_inpinfo_lock_del(inp, tp);
351 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
353 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
354 tcp_inpinfo_lock_del(inp, tp);
360 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
361 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
364 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
373 tcp_timer_keep(void *xtp)
375 struct tcpcb *tp = xtp;
376 struct tcptemp *t_template;
378 struct epoch_tracker et;
379 CURVNET_SET(tp->t_vnet);
383 ostate = tp->t_state;
386 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
388 if (callout_pending(&tp->t_timers->tt_keep) ||
389 !callout_active(&tp->t_timers->tt_keep)) {
394 callout_deactivate(&tp->t_timers->tt_keep);
395 if ((inp->inp_flags & INP_DROPPED) != 0) {
400 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
401 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
404 * Because we don't regularly reset the keepalive callout in
405 * the ESTABLISHED state, it may be that we don't actually need
406 * to send a keepalive yet. If that occurs, schedule another
407 * call for the next time the keepalive timer might expire.
409 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
412 idletime = ticks - tp->t_rcvtime;
413 if (idletime < TP_KEEPIDLE(tp)) {
414 callout_reset(&tp->t_timers->tt_keep,
415 TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
423 * Keep-alive timer went off; send something
424 * or drop connection if idle for too long.
426 TCPSTAT_INC(tcps_keeptimeo);
427 if (tp->t_state < TCPS_ESTABLISHED)
429 if ((tcp_always_keepalive ||
430 inp->inp_socket->so_options & SO_KEEPALIVE) &&
431 tp->t_state <= TCPS_CLOSING) {
432 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
435 * Send a packet designed to force a response
436 * if the peer is up and reachable:
437 * either an ACK if the connection is still alive,
438 * or an RST if the peer has closed the connection
439 * due to timeout or reboot.
440 * Using sequence number tp->snd_una-1
441 * causes the transmitted zero-length segment
442 * to lie outside the receive window;
443 * by the protocol spec, this requires the
444 * correspondent TCP to respond.
446 TCPSTAT_INC(tcps_keepprobe);
447 t_template = tcpip_maketemplate(inp);
449 tcp_respond(tp, t_template->tt_ipgen,
450 &t_template->tt_t, (struct mbuf *)NULL,
451 tp->rcv_nxt, tp->snd_una - 1, 0);
452 free(t_template, M_TEMP);
454 callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
457 callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
461 if (inp->inp_socket->so_options & SO_DEBUG)
462 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
465 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
471 TCPSTAT_INC(tcps_keepdrops);
472 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
473 tcp_inpinfo_lock_del(inp, tp);
476 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
477 tp = tcp_drop(tp, ETIMEDOUT);
480 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
481 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
484 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
485 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
486 tcp_inpinfo_lock_del(inp, tp);
492 tcp_timer_persist(void *xtp)
494 struct tcpcb *tp = xtp;
496 struct epoch_tracker et;
497 CURVNET_SET(tp->t_vnet);
501 ostate = tp->t_state;
504 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
506 if (callout_pending(&tp->t_timers->tt_persist) ||
507 !callout_active(&tp->t_timers->tt_persist)) {
512 callout_deactivate(&tp->t_timers->tt_persist);
513 if ((inp->inp_flags & INP_DROPPED) != 0) {
518 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
519 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
521 * Persistence timer into zero window.
522 * Force a byte to be output, if possible.
524 TCPSTAT_INC(tcps_persisttimeo);
526 * Hack: if the peer is dead/unreachable, we do not
527 * time out if the window is closed. After a full
528 * backoff, drop the connection if the idle time
529 * (no responses to probes) reaches the maximum
530 * backoff that we would use if retransmitting.
532 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
533 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
534 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
535 TCPSTAT_INC(tcps_persistdrop);
536 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
537 tcp_inpinfo_lock_del(inp, tp);
540 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
541 tp = tcp_drop(tp, ETIMEDOUT);
542 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
543 tcp_inpinfo_lock_del(inp, tp);
547 * If the user has closed the socket then drop a persisting
548 * connection after a much reduced timeout.
550 if (tp->t_state > TCPS_CLOSE_WAIT &&
551 (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
552 TCPSTAT_INC(tcps_persistdrop);
553 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
554 tcp_inpinfo_lock_del(inp, tp);
557 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
558 tp = tcp_drop(tp, ETIMEDOUT);
559 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
560 tcp_inpinfo_lock_del(inp, tp);
564 tp->t_flags |= TF_FORCEDATA;
565 (void) tp->t_fb->tfb_tcp_output(tp);
566 tp->t_flags &= ~TF_FORCEDATA;
569 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
570 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
572 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
579 tcp_timer_rexmt(void * xtp)
581 struct tcpcb *tp = xtp;
582 CURVNET_SET(tp->t_vnet);
585 struct epoch_tracker et;
589 ostate = tp->t_state;
592 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
594 if (callout_pending(&tp->t_timers->tt_rexmt) ||
595 !callout_active(&tp->t_timers->tt_rexmt)) {
600 callout_deactivate(&tp->t_timers->tt_rexmt);
601 if ((inp->inp_flags & INP_DROPPED) != 0) {
606 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
607 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
608 tcp_free_sackholes(tp);
609 TCP_LOG_EVENT(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0, NULL, false);
610 if (tp->t_fb->tfb_tcp_rexmit_tmr) {
611 /* The stack has a timer action too. */
612 (*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
615 * Retransmission timer went off. Message has not
616 * been acked within retransmit interval. Back off
617 * to a longer retransmit interval and retransmit one segment.
619 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
620 tp->t_rxtshift = TCP_MAXRXTSHIFT;
621 TCPSTAT_INC(tcps_timeoutdrop);
622 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
623 tcp_inpinfo_lock_del(inp, tp);
626 INP_INFO_RLOCK_ET(&V_tcbinfo, et);
627 tp = tcp_drop(tp, ETIMEDOUT);
628 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
629 tcp_inpinfo_lock_del(inp, tp);
632 if (tp->t_state == TCPS_SYN_SENT) {
634 * If the SYN was retransmitted, indicate CWND to be
635 * limited to 1 segment in cc_conn_init().
638 } else if (tp->t_rxtshift == 1) {
640 * first retransmit; record ssthresh and cwnd so they can
641 * be recovered if this turns out to be a "bad" retransmit.
642 * A retransmit is considered "bad" if an ACK for this
643 * segment is received within RTT/2 interval; the assumption
644 * here is that the ACK was already in flight. See
645 * "On Estimating End-to-End Network Path Properties" by
646 * Allman and Paxson for more details.
648 tp->snd_cwnd_prev = tp->snd_cwnd;
649 tp->snd_ssthresh_prev = tp->snd_ssthresh;
650 tp->snd_recover_prev = tp->snd_recover;
651 if (IN_FASTRECOVERY(tp->t_flags))
652 tp->t_flags |= TF_WASFRECOVERY;
654 tp->t_flags &= ~TF_WASFRECOVERY;
655 if (IN_CONGRECOVERY(tp->t_flags))
656 tp->t_flags |= TF_WASCRECOVERY;
658 tp->t_flags &= ~TF_WASCRECOVERY;
659 if ((tp->t_flags & TF_RCVD_TSTMP) == 0)
660 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
661 /* In the event that we've negotiated timestamps
662 * badrxtwin will be set to the value that we set
663 * the retransmitted packet's to_tsval to by tcp_output
665 tp->t_flags |= TF_PREVVALID;
667 tp->t_flags &= ~TF_PREVVALID;
668 TCPSTAT_INC(tcps_rexmttimeo);
669 if ((tp->t_state == TCPS_SYN_SENT) ||
670 (tp->t_state == TCPS_SYN_RECEIVED))
671 rexmt = TCPTV_RTOBASE * tcp_backoff[tp->t_rxtshift];
673 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
674 TCPT_RANGESET(tp->t_rxtcur, rexmt,
675 tp->t_rttmin, TCPTV_REXMTMAX);
678 * We enter the path for PLMTUD if connection is established or, if
679 * connection is FIN_WAIT_1 status, reason for the last is that if
680 * amount of data we send is very small, we could send it in couple of
681 * packets and process straight to FIN. In that case we won't catch
684 if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
685 || (tp->t_state == TCPS_FIN_WAIT_1))) {
691 * Idea here is that at each stage of mtu probe (usually, 1448
692 * -> 1188 -> 524) should be given 2 chances to recover before
693 * further clamping down. 'tp->t_rxtshift % 2 == 0' should
696 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
697 (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
698 (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
699 tp->t_rxtshift % 2 == 0)) {
701 * Enter Path MTU Black-hole Detection mechanism:
702 * - Disable Path MTU Discovery (IP "DF" bit).
703 * - Reduce MTU to lower value than what we
704 * negotiated with peer.
706 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
707 /* Record that we may have found a black hole. */
708 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
709 /* Keep track of previous MSS. */
710 tp->t_pmtud_saved_maxseg = tp->t_maxseg;
714 * Reduce the MSS to blackhole value or to the default
715 * in an attempt to retransmit.
718 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
720 tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
721 /* Use the sysctl tuneable blackhole MSS. */
722 tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
723 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
725 /* Use the default MSS. */
726 tp->t_maxseg = V_tcp_v6mssdflt;
728 * Disable Path MTU Discovery when we switch to
731 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
732 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
735 #if defined(INET6) && defined(INET)
739 if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
740 /* Use the sysctl tuneable blackhole MSS. */
741 tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
742 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
744 /* Use the default MSS. */
745 tp->t_maxseg = V_tcp_mssdflt;
747 * Disable Path MTU Discovery when we switch to
750 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
751 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
755 * Reset the slow-start flight size
756 * as it may depend on the new MSS.
758 if (CC_ALGO(tp)->conn_init != NULL)
759 CC_ALGO(tp)->conn_init(tp->ccv);
762 * If further retransmissions are still unsuccessful
763 * with a lowered MTU, maybe this isn't a blackhole and
764 * we restore the previous MSS and blackhole detection
766 * The limit '6' is determined by giving each probe
767 * stage (1448, 1188, 524) 2 chances to recover.
769 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
770 (tp->t_rxtshift >= 6)) {
771 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
772 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
773 tp->t_maxseg = tp->t_pmtud_saved_maxseg;
774 TCPSTAT_INC(tcps_pmtud_blackhole_failed);
776 * Reset the slow-start flight size as it
777 * may depend on the new MSS.
779 if (CC_ALGO(tp)->conn_init != NULL)
780 CC_ALGO(tp)->conn_init(tp->ccv);
786 * Disable RFC1323 and SACK if we haven't got any response to
787 * our third SYN to work-around some broken terminal servers
788 * (most of which have hopefully been retired) that have bad VJ
789 * header compression code which trashes TCP segments containing
790 * unknown-to-them TCP options.
792 if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
793 (tp->t_rxtshift == 3))
794 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
796 * If we backed off this far, notify the L3 protocol that we're having
797 * connection problems.
799 if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
801 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
802 in6_losing(tp->t_inpcb);
805 in_losing(tp->t_inpcb);
807 tp->snd_nxt = tp->snd_una;
808 tp->snd_recover = tp->snd_max;
810 * Force a segment to be sent.
812 tp->t_flags |= TF_ACKNOW;
814 * If timing a segment in this window, stop the timer.
818 cc_cong_signal(tp, NULL, CC_RTO);
820 (void) tp->t_fb->tfb_tcp_output(tp);
823 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
824 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
827 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
834 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
836 struct callout *t_callout;
837 timeout_t *f_callout;
838 struct inpcb *inp = tp->t_inpcb;
839 int cpu = inp_to_cpuid(inp);
842 if (tp->t_flags & TF_TOE)
846 if (tp->t_timers->tt_flags & TT_STOPPED)
849 switch (timer_type) {
851 t_callout = &tp->t_timers->tt_delack;
852 f_callout = tcp_timer_delack;
855 t_callout = &tp->t_timers->tt_rexmt;
856 f_callout = tcp_timer_rexmt;
859 t_callout = &tp->t_timers->tt_persist;
860 f_callout = tcp_timer_persist;
863 t_callout = &tp->t_timers->tt_keep;
864 f_callout = tcp_timer_keep;
867 t_callout = &tp->t_timers->tt_2msl;
868 f_callout = tcp_timer_2msl;
871 if (tp->t_fb->tfb_tcp_timer_activate) {
872 tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
875 panic("tp %p bad timer_type %#x", tp, timer_type);
878 callout_stop(t_callout);
880 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
885 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
887 struct callout *t_callout;
889 switch (timer_type) {
891 t_callout = &tp->t_timers->tt_delack;
894 t_callout = &tp->t_timers->tt_rexmt;
897 t_callout = &tp->t_timers->tt_persist;
900 t_callout = &tp->t_timers->tt_keep;
903 t_callout = &tp->t_timers->tt_2msl;
906 if (tp->t_fb->tfb_tcp_timer_active) {
907 return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
909 panic("tp %p bad timer_type %#x", tp, timer_type);
911 return callout_active(t_callout);
915 * Stop the timer from running, and apply a flag
916 * against the timer_flags that will force the
917 * timer never to run. The flag is needed to assure
918 * a race does not leave it running and cause
919 * the timer to possibly restart itself (keep and persist
920 * especially do this).
923 tcp_timer_suspend(struct tcpcb *tp, uint32_t timer_type)
925 struct callout *t_callout;
928 switch (timer_type) {
930 t_flags = TT_DELACK_SUS;
931 t_callout = &tp->t_timers->tt_delack;
934 t_flags = TT_REXMT_SUS;
935 t_callout = &tp->t_timers->tt_rexmt;
938 t_flags = TT_PERSIST_SUS;
939 t_callout = &tp->t_timers->tt_persist;
942 t_flags = TT_KEEP_SUS;
943 t_callout = &tp->t_timers->tt_keep;
946 t_flags = TT_2MSL_SUS;
947 t_callout = &tp->t_timers->tt_2msl;
950 panic("tp:%p bad timer_type 0x%x", tp, timer_type);
952 tp->t_timers->tt_flags |= t_flags;
953 return (callout_stop(t_callout));
957 tcp_timers_unsuspend(struct tcpcb *tp, uint32_t timer_type)
959 switch (timer_type) {
961 if (tp->t_timers->tt_flags & TT_DELACK_SUS) {
962 tp->t_timers->tt_flags &= ~TT_DELACK_SUS;
963 if (tp->t_flags & TF_DELACK) {
964 /* Delayed ack timer should be up activate a timer */
965 tp->t_flags &= ~TF_DELACK;
966 tcp_timer_activate(tp, TT_DELACK,
972 if (tp->t_timers->tt_flags & TT_REXMT_SUS) {
973 tp->t_timers->tt_flags &= ~TT_REXMT_SUS;
974 if (SEQ_GT(tp->snd_max, tp->snd_una) &&
975 (tcp_timer_active((tp), TT_PERSIST) == 0) &&
977 /* We have outstanding data activate a timer */
978 tcp_timer_activate(tp, TT_REXMT,
984 if (tp->t_timers->tt_flags & TT_PERSIST_SUS) {
985 tp->t_timers->tt_flags &= ~TT_PERSIST_SUS;
986 if (tp->snd_wnd == 0) {
987 /* Activate the persists timer */
994 if (tp->t_timers->tt_flags & TT_KEEP_SUS) {
995 tp->t_timers->tt_flags &= ~TT_KEEP_SUS;
996 tcp_timer_activate(tp, TT_KEEP,
997 TCPS_HAVEESTABLISHED(tp->t_state) ?
998 TP_KEEPIDLE(tp) : TP_KEEPINIT(tp));
1002 if (tp->t_timers->tt_flags &= TT_2MSL_SUS) {
1003 tp->t_timers->tt_flags &= ~TT_2MSL_SUS;
1004 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1005 ((tp->t_inpcb->inp_socket == NULL) ||
1006 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE))) {
1007 /* Star the 2MSL timer */
1008 tcp_timer_activate(tp, TT_2MSL,
1009 (tcp_fast_finwait2_recycle) ?
1010 tcp_finwait2_timeout : TP_MAXIDLE(tp));
1015 panic("tp:%p bad timer_type 0x%x", tp, timer_type);
1020 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
1022 struct callout *t_callout;
1024 tp->t_timers->tt_flags |= TT_STOPPED;
1025 switch (timer_type) {
1027 t_callout = &tp->t_timers->tt_delack;
1030 t_callout = &tp->t_timers->tt_rexmt;
1033 t_callout = &tp->t_timers->tt_persist;
1036 t_callout = &tp->t_timers->tt_keep;
1039 t_callout = &tp->t_timers->tt_2msl;
1042 if (tp->t_fb->tfb_tcp_timer_stop) {
1044 * XXXrrs we need to look at this with the
1045 * stop case below (flags).
1047 tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
1050 panic("tp %p bad timer_type %#x", tp, timer_type);
1053 if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
1055 * Can't stop the callout, defer tcpcb actual deletion
1056 * to the last one. We do this using the async drain
1057 * function and incrementing the count in
1059 tp->t_timers->tt_draincnt++;