2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * @(#)tcp_timer.c 8.2 (Berkeley) 5/24/95
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
37 #include "opt_tcpdebug.h"
40 #include <sys/param.h>
41 #include <sys/kernel.h>
44 #include <sys/mutex.h>
45 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
53 #include <net/route.h>
55 #include <net/netisr.h>
57 #include <netinet/cc.h>
58 #include <netinet/in.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_fsm.h>
67 #include <netinet/tcp_timer.h>
68 #include <netinet/tcp_var.h>
70 #include <netinet6/tcp6_var.h>
72 #include <netinet/tcpip.h>
74 #include <netinet/tcp_debug.h>
78 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
79 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
82 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
83 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
86 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
87 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
90 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
91 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
92 "Time before a delayed ACK is sent");
95 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
96 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
99 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
100 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
101 "Minimum Retransmission Timeout");
104 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
105 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
106 "Retransmission Timer Slop");
108 static int always_keepalive = 1;
109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
110 &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
112 int tcp_fast_finwait2_recycle = 0;
113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
114 &tcp_fast_finwait2_recycle, 0,
115 "Recycle closed FIN_WAIT_2 connections faster");
117 int tcp_finwait2_timeout;
118 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
119 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
121 int tcp_keepcnt = TCPTV_KEEPCNT;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
123 "Number of keepalive probes to send");
125 /* max idle probes */
126 int tcp_maxpersistidle;
128 static int tcp_rexmit_drop_options = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
130 &tcp_rexmit_drop_options, 0,
131 "Drop TCP options from 3rd and later retransmitted SYN");
133 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
134 #define V_tcp_pmtud_blackhole_detect VNET(tcp_pmtud_blackhole_detect)
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
136 CTLFLAG_RW|CTLFLAG_VNET,
137 &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
138 "Path MTU Discovery Black Hole Detection Enabled");
140 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
141 #define V_tcp_pmtud_blackhole_activated \
142 VNET(tcp_pmtud_blackhole_activated)
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
144 CTLFLAG_RD|CTLFLAG_VNET,
145 &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
146 "Path MTU Discovery Black Hole Detection, Activation Count");
148 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
149 #define V_tcp_pmtud_blackhole_activated_min_mss \
150 VNET(tcp_pmtud_blackhole_activated_min_mss)
151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
152 CTLFLAG_RD|CTLFLAG_VNET,
153 &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
154 "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
156 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
157 #define V_tcp_pmtud_blackhole_failed VNET(tcp_pmtud_blackhole_failed)
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
159 CTLFLAG_RD|CTLFLAG_VNET,
160 &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
161 "Path MTU Discovery Black Hole Detection, Failure Count");
164 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
165 #define V_tcp_pmtud_blackhole_mss VNET(tcp_pmtud_blackhole_mss)
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
167 CTLFLAG_RW|CTLFLAG_VNET,
168 &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
169 "Path MTU Discovery Black Hole Detection lowered MSS");
173 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
174 #define V_tcp_v6pmtud_blackhole_mss VNET(tcp_v6pmtud_blackhole_mss)
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
176 CTLFLAG_RW|CTLFLAG_VNET,
177 &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
178 "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
182 static int per_cpu_timers = 1;
184 static int per_cpu_timers = 0;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
187 &per_cpu_timers , 0, "run tcp timers on all cpus");
190 #define INP_CPU(inp) (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
191 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
195 * Map the given inp to a CPU id.
197 * This queries RSS if it's compiled in, else it defaults to the current
201 inp_to_cpuid(struct inpcb *inp)
206 if (per_cpu_timers) {
207 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
208 if (cpuid == NETISR_CPUID_NONE)
209 return (curcpu); /* XXX */
214 /* Legacy, pre-RSS behaviour */
215 if (per_cpu_timers) {
217 * We don't have a flowid -> cpuid mapping, so cheat and
218 * just map unknown cpuids to curcpu. Not the best, but
219 * apparently better than defaulting to swi 0.
221 cpuid = inp->inp_flowid % (mp_maxid + 1);
222 if (! CPU_ABSENT(cpuid))
227 /* Default for RSS and non-RSS - cpuid 0 */
234 * Tcp protocol timeout routine called every 500 ms.
235 * Updates timestamps used for TCP
236 * causes finite state machine actions if timers expire.
241 VNET_ITERATOR_DECL(vnet_iter);
243 VNET_LIST_RLOCK_NOSLEEP();
244 VNET_FOREACH(vnet_iter) {
245 CURVNET_SET(vnet_iter);
246 (void) tcp_tw_2msl_scan(0);
249 VNET_LIST_RUNLOCK_NOSLEEP();
252 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
253 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
255 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
256 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
258 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
260 static int tcp_timer_race;
261 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
262 0, "Count of t_inpcb races on tcp_discardcb");
265 * TCP timer processing.
269 tcp_timer_delack(void *xtp)
271 struct tcpcb *tp = xtp;
273 CURVNET_SET(tp->t_vnet);
277 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
278 * tear-down mean we need it as a work-around for races between
279 * timers and tcp_discardcb().
281 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
289 if (callout_pending(&tp->t_timers->tt_delack) ||
290 !callout_active(&tp->t_timers->tt_delack)) {
295 callout_deactivate(&tp->t_timers->tt_delack);
296 if ((inp->inp_flags & INP_DROPPED) != 0) {
302 tp->t_flags |= TF_ACKNOW;
303 TCPSTAT_INC(tcps_delack);
304 (void) tcp_output(tp);
310 tcp_timer_2msl(void *xtp)
312 struct tcpcb *tp = xtp;
314 CURVNET_SET(tp->t_vnet);
318 ostate = tp->t_state;
321 * XXXRW: Does this actually happen?
323 INP_INFO_WLOCK(&V_tcbinfo);
326 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
327 * tear-down mean we need it as a work-around for races between
328 * timers and tcp_discardcb().
330 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
334 INP_INFO_WUNLOCK(&V_tcbinfo);
339 tcp_free_sackholes(tp);
340 if (callout_pending(&tp->t_timers->tt_2msl) ||
341 !callout_active(&tp->t_timers->tt_2msl)) {
342 INP_WUNLOCK(tp->t_inpcb);
343 INP_INFO_WUNLOCK(&V_tcbinfo);
347 callout_deactivate(&tp->t_timers->tt_2msl);
348 if ((inp->inp_flags & INP_DROPPED) != 0) {
350 INP_INFO_WUNLOCK(&V_tcbinfo);
355 * 2 MSL timeout in shutdown went off. If we're closed but
356 * still waiting for peer to close and connection has been idle
357 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
358 * control block. Otherwise, check again in a bit.
360 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
361 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
362 * Ignore fact that there were recent incoming segments.
364 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
365 tp->t_inpcb && tp->t_inpcb->inp_socket &&
366 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
367 TCPSTAT_INC(tcps_finwait2_drops);
370 if (tp->t_state != TCPS_TIME_WAIT &&
371 ticks - tp->t_rcvtime <= TP_MAXIDLE(tp))
372 callout_reset_on(&tp->t_timers->tt_2msl,
373 TP_KEEPINTVL(tp), tcp_timer_2msl, tp,
380 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
381 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
386 INP_INFO_WUNLOCK(&V_tcbinfo);
391 tcp_timer_keep(void *xtp)
393 struct tcpcb *tp = xtp;
394 struct tcptemp *t_template;
396 CURVNET_SET(tp->t_vnet);
400 ostate = tp->t_state;
402 INP_INFO_WLOCK(&V_tcbinfo);
405 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
406 * tear-down mean we need it as a work-around for races between
407 * timers and tcp_discardcb().
409 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
413 INP_INFO_WUNLOCK(&V_tcbinfo);
418 if (callout_pending(&tp->t_timers->tt_keep) ||
419 !callout_active(&tp->t_timers->tt_keep)) {
421 INP_INFO_WUNLOCK(&V_tcbinfo);
425 callout_deactivate(&tp->t_timers->tt_keep);
426 if ((inp->inp_flags & INP_DROPPED) != 0) {
428 INP_INFO_WUNLOCK(&V_tcbinfo);
433 * Keep-alive timer went off; send something
434 * or drop connection if idle for too long.
436 TCPSTAT_INC(tcps_keeptimeo);
437 if (tp->t_state < TCPS_ESTABLISHED)
439 if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
440 tp->t_state <= TCPS_CLOSING) {
441 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
444 * Send a packet designed to force a response
445 * if the peer is up and reachable:
446 * either an ACK if the connection is still alive,
447 * or an RST if the peer has closed the connection
448 * due to timeout or reboot.
449 * Using sequence number tp->snd_una-1
450 * causes the transmitted zero-length segment
451 * to lie outside the receive window;
452 * by the protocol spec, this requires the
453 * correspondent TCP to respond.
455 TCPSTAT_INC(tcps_keepprobe);
456 t_template = tcpip_maketemplate(inp);
458 tcp_respond(tp, t_template->tt_ipgen,
459 &t_template->tt_t, (struct mbuf *)NULL,
460 tp->rcv_nxt, tp->snd_una - 1, 0);
461 free(t_template, M_TEMP);
463 callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
464 tcp_timer_keep, tp, inp_to_cpuid(inp));
466 callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
467 tcp_timer_keep, tp, inp_to_cpuid(inp));
470 if (inp->inp_socket->so_options & SO_DEBUG)
471 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
475 INP_INFO_WUNLOCK(&V_tcbinfo);
480 TCPSTAT_INC(tcps_keepdrops);
481 tp = tcp_drop(tp, ETIMEDOUT);
484 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
485 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
489 INP_WUNLOCK(tp->t_inpcb);
490 INP_INFO_WUNLOCK(&V_tcbinfo);
495 tcp_timer_persist(void *xtp)
497 struct tcpcb *tp = xtp;
499 CURVNET_SET(tp->t_vnet);
503 ostate = tp->t_state;
505 INP_INFO_WLOCK(&V_tcbinfo);
508 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
509 * tear-down mean we need it as a work-around for races between
510 * timers and tcp_discardcb().
512 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
516 INP_INFO_WUNLOCK(&V_tcbinfo);
521 if (callout_pending(&tp->t_timers->tt_persist) ||
522 !callout_active(&tp->t_timers->tt_persist)) {
524 INP_INFO_WUNLOCK(&V_tcbinfo);
528 callout_deactivate(&tp->t_timers->tt_persist);
529 if ((inp->inp_flags & INP_DROPPED) != 0) {
531 INP_INFO_WUNLOCK(&V_tcbinfo);
536 * Persistance timer into zero window.
537 * Force a byte to be output, if possible.
539 TCPSTAT_INC(tcps_persisttimeo);
541 * Hack: if the peer is dead/unreachable, we do not
542 * time out if the window is closed. After a full
543 * backoff, drop the connection if the idle time
544 * (no responses to probes) reaches the maximum
545 * backoff that we would use if retransmitting.
547 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
548 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
549 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
550 TCPSTAT_INC(tcps_persistdrop);
551 tp = tcp_drop(tp, ETIMEDOUT);
555 * If the user has closed the socket then drop a persisting
556 * connection after a much reduced timeout.
558 if (tp->t_state > TCPS_CLOSE_WAIT &&
559 (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
560 TCPSTAT_INC(tcps_persistdrop);
561 tp = tcp_drop(tp, ETIMEDOUT);
565 tp->t_flags |= TF_FORCEDATA;
566 (void) tcp_output(tp);
567 tp->t_flags &= ~TF_FORCEDATA;
571 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
572 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
576 INP_INFO_WUNLOCK(&V_tcbinfo);
581 tcp_timer_rexmt(void * xtp)
583 struct tcpcb *tp = xtp;
584 CURVNET_SET(tp->t_vnet);
591 ostate = tp->t_state;
594 INP_INFO_RLOCK(&V_tcbinfo);
597 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
598 * tear-down mean we need it as a work-around for races between
599 * timers and tcp_discardcb().
601 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
605 INP_INFO_RUNLOCK(&V_tcbinfo);
610 if (callout_pending(&tp->t_timers->tt_rexmt) ||
611 !callout_active(&tp->t_timers->tt_rexmt)) {
613 INP_INFO_RUNLOCK(&V_tcbinfo);
617 callout_deactivate(&tp->t_timers->tt_rexmt);
618 if ((inp->inp_flags & INP_DROPPED) != 0) {
620 INP_INFO_RUNLOCK(&V_tcbinfo);
624 tcp_free_sackholes(tp);
626 * Retransmission timer went off. Message has not
627 * been acked within retransmit interval. Back off
628 * to a longer retransmit interval and retransmit one segment.
630 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
631 tp->t_rxtshift = TCP_MAXRXTSHIFT;
632 TCPSTAT_INC(tcps_timeoutdrop);
634 INP_INFO_RUNLOCK(&V_tcbinfo);
636 INP_INFO_WLOCK(&V_tcbinfo);
638 if (in_pcbrele_wlocked(inp)) {
639 INP_INFO_WUNLOCK(&V_tcbinfo);
643 if (inp->inp_flags & INP_DROPPED) {
645 INP_INFO_WUNLOCK(&V_tcbinfo);
650 tp = tcp_drop(tp, tp->t_softerror ?
651 tp->t_softerror : ETIMEDOUT);
655 INP_INFO_RUNLOCK(&V_tcbinfo);
657 if (tp->t_state == TCPS_SYN_SENT) {
659 * If the SYN was retransmitted, indicate CWND to be
660 * limited to 1 segment in cc_conn_init().
663 } else if (tp->t_rxtshift == 1) {
665 * first retransmit; record ssthresh and cwnd so they can
666 * be recovered if this turns out to be a "bad" retransmit.
667 * A retransmit is considered "bad" if an ACK for this
668 * segment is received within RTT/2 interval; the assumption
669 * here is that the ACK was already in flight. See
670 * "On Estimating End-to-End Network Path Properties" by
671 * Allman and Paxson for more details.
673 tp->snd_cwnd_prev = tp->snd_cwnd;
674 tp->snd_ssthresh_prev = tp->snd_ssthresh;
675 tp->snd_recover_prev = tp->snd_recover;
676 if (IN_FASTRECOVERY(tp->t_flags))
677 tp->t_flags |= TF_WASFRECOVERY;
679 tp->t_flags &= ~TF_WASFRECOVERY;
680 if (IN_CONGRECOVERY(tp->t_flags))
681 tp->t_flags |= TF_WASCRECOVERY;
683 tp->t_flags &= ~TF_WASCRECOVERY;
684 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
685 tp->t_flags |= TF_PREVVALID;
687 tp->t_flags &= ~TF_PREVVALID;
688 TCPSTAT_INC(tcps_rexmttimeo);
689 if (tp->t_state == TCPS_SYN_SENT)
690 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
692 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
693 TCPT_RANGESET(tp->t_rxtcur, rexmt,
694 tp->t_rttmin, TCPTV_REXMTMAX);
697 * We enter the path for PLMTUD if connection is established or, if
698 * connection is FIN_WAIT_1 status, reason for the last is that if
699 * amount of data we send is very small, we could send it in couple of
700 * packets and process straight to FIN. In that case we won't catch
703 if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
704 || (tp->t_state == TCPS_FIN_WAIT_1))) {
710 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
711 (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
712 (tp->t_rxtshift <= 2)) {
714 * Enter Path MTU Black-hole Detection mechanism:
715 * - Disable Path MTU Discovery (IP "DF" bit).
716 * - Reduce MTU to lower value than what we
717 * negotiated with peer.
719 /* Record that we may have found a black hole. */
720 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
722 /* Keep track of previous MSS. */
723 optlen = tp->t_maxopd - tp->t_maxseg;
724 tp->t_pmtud_saved_maxopd = tp->t_maxopd;
727 * Reduce the MSS to blackhole value or to the default
728 * in an attempt to retransmit.
731 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
733 tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
734 /* Use the sysctl tuneable blackhole MSS. */
735 tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
736 V_tcp_pmtud_blackhole_activated++;
738 /* Use the default MSS. */
739 tp->t_maxopd = V_tcp_v6mssdflt;
741 * Disable Path MTU Discovery when we switch to
744 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
745 V_tcp_pmtud_blackhole_activated_min_mss++;
748 #if defined(INET6) && defined(INET)
752 if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
753 /* Use the sysctl tuneable blackhole MSS. */
754 tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
755 V_tcp_pmtud_blackhole_activated++;
757 /* Use the default MSS. */
758 tp->t_maxopd = V_tcp_mssdflt;
760 * Disable Path MTU Discovery when we switch to
763 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
764 V_tcp_pmtud_blackhole_activated_min_mss++;
767 tp->t_maxseg = tp->t_maxopd - optlen;
769 * Reset the slow-start flight size
770 * as it may depend on the new MSS.
772 if (CC_ALGO(tp)->conn_init != NULL)
773 CC_ALGO(tp)->conn_init(tp->ccv);
776 * If further retransmissions are still unsuccessful
777 * with a lowered MTU, maybe this isn't a blackhole and
778 * we restore the previous MSS and blackhole detection
781 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
782 (tp->t_rxtshift > 4)) {
783 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
784 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
785 optlen = tp->t_maxopd - tp->t_maxseg;
786 tp->t_maxopd = tp->t_pmtud_saved_maxopd;
787 tp->t_maxseg = tp->t_maxopd - optlen;
788 V_tcp_pmtud_blackhole_failed++;
790 * Reset the slow-start flight size as it
791 * may depend on the new MSS.
793 if (CC_ALGO(tp)->conn_init != NULL)
794 CC_ALGO(tp)->conn_init(tp->ccv);
800 * Disable RFC1323 and SACK if we haven't got any response to
801 * our third SYN to work-around some broken terminal servers
802 * (most of which have hopefully been retired) that have bad VJ
803 * header compression code which trashes TCP segments containing
804 * unknown-to-them TCP options.
806 if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
807 (tp->t_rxtshift == 3))
808 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
810 * If we backed off this far, our srtt estimate is probably bogus.
811 * Clobber it so we'll take the next rtt measurement as our srtt;
812 * move the current srtt into rttvar to keep the current
813 * retransmit times until then.
815 if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
817 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
818 in6_losing(tp->t_inpcb);
820 tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
823 tp->snd_nxt = tp->snd_una;
824 tp->snd_recover = tp->snd_max;
826 * Force a segment to be sent.
828 tp->t_flags |= TF_ACKNOW;
830 * If timing a segment in this window, stop the timer.
834 cc_cong_signal(tp, NULL, CC_RTO);
836 (void) tcp_output(tp);
840 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
841 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
847 INP_INFO_WUNLOCK(&V_tcbinfo);
852 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
854 struct callout *t_callout;
856 struct inpcb *inp = tp->t_inpcb;
857 int cpu = inp_to_cpuid(inp);
860 if (tp->t_flags & TF_TOE)
864 switch (timer_type) {
866 t_callout = &tp->t_timers->tt_delack;
867 f_callout = tcp_timer_delack;
870 t_callout = &tp->t_timers->tt_rexmt;
871 f_callout = tcp_timer_rexmt;
874 t_callout = &tp->t_timers->tt_persist;
875 f_callout = tcp_timer_persist;
878 t_callout = &tp->t_timers->tt_keep;
879 f_callout = tcp_timer_keep;
882 t_callout = &tp->t_timers->tt_2msl;
883 f_callout = tcp_timer_2msl;
886 panic("bad timer_type");
889 callout_stop(t_callout);
891 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
896 tcp_timer_active(struct tcpcb *tp, int timer_type)
898 struct callout *t_callout;
900 switch (timer_type) {
902 t_callout = &tp->t_timers->tt_delack;
905 t_callout = &tp->t_timers->tt_rexmt;
908 t_callout = &tp->t_timers->tt_persist;
911 t_callout = &tp->t_timers->tt_keep;
914 t_callout = &tp->t_timers->tt_2msl;
917 panic("bad timer_type");
919 return callout_active(t_callout);
922 #define ticks_to_msecs(t) (1000*(t) / hz)
925 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
926 struct xtcp_timer *xtimer)
930 bzero(xtimer, sizeof(*xtimer));
933 now = getsbinuptime();
934 if (callout_active(&timer->tt_delack))
935 xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
936 if (callout_active(&timer->tt_rexmt))
937 xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
938 if (callout_active(&timer->tt_persist))
939 xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
940 if (callout_active(&timer->tt_keep))
941 xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
942 if (callout_active(&timer->tt_2msl))
943 xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
944 xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);