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/cc/cc.h>
76 #include <netinet6/tcp6_var.h>
78 #include <netinet/tcpip.h>
80 #include <netinet/tcp_debug.h>
84 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
85 &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
88 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
89 &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
92 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
93 &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
96 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
97 &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
100 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
101 &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
104 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
105 &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
106 "Time before a delayed ACK is sent");
109 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
110 &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
113 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
114 &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
115 "Minimum Retransmission Timeout");
118 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
119 &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
120 "Retransmission Timer Slop");
122 int tcp_always_keepalive = 1;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
124 &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
126 int tcp_fast_finwait2_recycle = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
128 &tcp_fast_finwait2_recycle, 0,
129 "Recycle closed FIN_WAIT_2 connections faster");
131 int tcp_finwait2_timeout;
132 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
133 &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
135 int tcp_keepcnt = TCPTV_KEEPCNT;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
137 "Number of keepalive probes to send");
139 /* max idle probes */
140 int tcp_maxpersistidle;
142 static int tcp_rexmit_drop_options = 0;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
144 &tcp_rexmit_drop_options, 0,
145 "Drop TCP options from 3rd and later retransmitted SYN");
147 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
149 CTLFLAG_RW|CTLFLAG_VNET,
150 &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
151 "Path MTU Discovery Black Hole Detection Enabled");
154 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
156 CTLFLAG_RW|CTLFLAG_VNET,
157 &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
158 "Path MTU Discovery Black Hole Detection lowered MSS");
162 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
164 CTLFLAG_RW|CTLFLAG_VNET,
165 &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
166 "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
170 static int per_cpu_timers = 1;
172 static int per_cpu_timers = 0;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
175 &per_cpu_timers , 0, "run tcp timers on all cpus");
178 #define INP_CPU(inp) (per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
179 ((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
183 * Map the given inp to a CPU id.
185 * This queries RSS if it's compiled in, else it defaults to the current
189 inp_to_cpuid(struct inpcb *inp)
194 if (per_cpu_timers) {
195 cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
196 if (cpuid == NETISR_CPUID_NONE)
197 return (curcpu); /* XXX */
202 /* Legacy, pre-RSS behaviour */
203 if (per_cpu_timers) {
205 * We don't have a flowid -> cpuid mapping, so cheat and
206 * just map unknown cpuids to curcpu. Not the best, but
207 * apparently better than defaulting to swi 0.
209 cpuid = inp->inp_flowid % (mp_maxid + 1);
210 if (! CPU_ABSENT(cpuid))
215 /* Default for RSS and non-RSS - cpuid 0 */
222 * Tcp protocol timeout routine called every 500 ms.
223 * Updates timestamps used for TCP
224 * causes finite state machine actions if timers expire.
229 VNET_ITERATOR_DECL(vnet_iter);
231 VNET_LIST_RLOCK_NOSLEEP();
232 VNET_FOREACH(vnet_iter) {
233 CURVNET_SET(vnet_iter);
234 (void) tcp_tw_2msl_scan(0);
237 VNET_LIST_RUNLOCK_NOSLEEP();
240 int tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
241 { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
243 int tcp_backoff[TCP_MAXRXTSHIFT + 1] =
244 { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
246 static int tcp_totbackoff = 2559; /* sum of tcp_backoff[] */
249 * TCP timer processing.
253 tcp_timer_delack(void *xtp)
255 struct tcpcb *tp = xtp;
257 CURVNET_SET(tp->t_vnet);
260 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
262 if (callout_pending(&tp->t_timers->tt_delack) ||
263 !callout_active(&tp->t_timers->tt_delack)) {
268 callout_deactivate(&tp->t_timers->tt_delack);
269 if ((inp->inp_flags & INP_DROPPED) != 0) {
274 tp->t_flags |= TF_ACKNOW;
275 TCPSTAT_INC(tcps_delack);
276 (void) tp->t_fb->tfb_tcp_output(tp);
282 * When a timer wants to remove a TCB it must
283 * hold the INP_INFO_RLOCK(). The timer function
284 * should only have grabbed the INP_WLOCK() when
285 * it entered. To safely switch to holding both the
286 * INP_INFO_RLOCK() and the INP_WLOCK() we must first
287 * grab a reference on the inp, which will hold the inp
288 * so that it can't be removed. We then unlock the INP_WLOCK(),
289 * and grab the INP_INFO_RLOCK() lock. Once we have the INP_INFO_RLOCK()
290 * we proceed again to get the INP_WLOCK() (this preserves proper
291 * lock order). After acquiring the INP_WLOCK we must check if someone
292 * else deleted the pcb i.e. the inp_flags check.
293 * If so we return 1 otherwise we return 0.
295 * No matter what the tcp_inpinfo_lock_add() function
296 * returns the caller must afterwards call tcp_inpinfo_lock_del()
297 * to drop the locks and reference properly.
301 tcp_inpinfo_lock_add(struct inpcb *inp)
305 INP_INFO_RLOCK(&V_tcbinfo);
307 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
315 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
317 INP_INFO_RUNLOCK(&V_tcbinfo);
318 if (inp && (tp == NULL)) {
320 * If tcp_close/drop() gets called and tp
321 * returns NULL, then the function dropped
322 * the inp lock, we hold a reference keeping
323 * this around, so we must re-aquire the
324 * INP_WLOCK() in order to proceed with
325 * our dropping the inp reference.
329 if (inp && in_pcbrele_wlocked(inp) == 0)
334 tcp_timer_2msl(void *xtp)
336 struct tcpcb *tp = xtp;
338 CURVNET_SET(tp->t_vnet);
342 ostate = tp->t_state;
345 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
347 tcp_free_sackholes(tp);
348 if (callout_pending(&tp->t_timers->tt_2msl) ||
349 !callout_active(&tp->t_timers->tt_2msl)) {
350 INP_WUNLOCK(tp->t_inpcb);
354 callout_deactivate(&tp->t_timers->tt_2msl);
355 if ((inp->inp_flags & INP_DROPPED) != 0) {
360 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
361 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
363 * 2 MSL timeout in shutdown went off. If we're closed but
364 * still waiting for peer to close and connection has been idle
365 * too long delete connection control block. Otherwise, check
368 * If in TIME_WAIT state just ignore as this timeout is handled in
369 * tcp_tw_2msl_scan().
371 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
372 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
373 * Ignore fact that there were recent incoming segments.
375 if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
380 if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
381 tp->t_inpcb && tp->t_inpcb->inp_socket &&
382 (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
383 TCPSTAT_INC(tcps_finwait2_drops);
384 if (tcp_inpinfo_lock_add(inp)) {
385 tcp_inpinfo_lock_del(inp, tp);
389 tcp_inpinfo_lock_del(inp, tp);
392 if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
393 callout_reset(&tp->t_timers->tt_2msl,
394 TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
396 if (tcp_inpinfo_lock_add(inp)) {
397 tcp_inpinfo_lock_del(inp, tp);
401 tcp_inpinfo_lock_del(inp, tp);
407 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
408 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
411 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
420 tcp_timer_keep(void *xtp)
422 struct tcpcb *tp = xtp;
423 struct tcptemp *t_template;
425 CURVNET_SET(tp->t_vnet);
429 ostate = tp->t_state;
432 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
434 if (callout_pending(&tp->t_timers->tt_keep) ||
435 !callout_active(&tp->t_timers->tt_keep)) {
440 callout_deactivate(&tp->t_timers->tt_keep);
441 if ((inp->inp_flags & INP_DROPPED) != 0) {
446 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
447 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
450 * Because we don't regularly reset the keepalive callout in
451 * the ESTABLISHED state, it may be that we don't actually need
452 * to send a keepalive yet. If that occurs, schedule another
453 * call for the next time the keepalive timer might expire.
455 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
458 idletime = ticks - tp->t_rcvtime;
459 if (idletime < TP_KEEPIDLE(tp)) {
460 callout_reset(&tp->t_timers->tt_keep,
461 TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
469 * Keep-alive timer went off; send something
470 * or drop connection if idle for too long.
472 TCPSTAT_INC(tcps_keeptimeo);
473 if (tp->t_state < TCPS_ESTABLISHED)
475 if ((tcp_always_keepalive ||
476 inp->inp_socket->so_options & SO_KEEPALIVE) &&
477 tp->t_state <= TCPS_CLOSING) {
478 if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
481 * Send a packet designed to force a response
482 * if the peer is up and reachable:
483 * either an ACK if the connection is still alive,
484 * or an RST if the peer has closed the connection
485 * due to timeout or reboot.
486 * Using sequence number tp->snd_una-1
487 * causes the transmitted zero-length segment
488 * to lie outside the receive window;
489 * by the protocol spec, this requires the
490 * correspondent TCP to respond.
492 TCPSTAT_INC(tcps_keepprobe);
493 t_template = tcpip_maketemplate(inp);
495 tcp_respond(tp, t_template->tt_ipgen,
496 &t_template->tt_t, (struct mbuf *)NULL,
497 tp->rcv_nxt, tp->snd_una - 1, 0);
498 free(t_template, M_TEMP);
500 callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
503 callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
507 if (inp->inp_socket->so_options & SO_DEBUG)
508 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
511 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
517 TCPSTAT_INC(tcps_keepdrops);
519 if (tcp_inpinfo_lock_add(inp)) {
520 tcp_inpinfo_lock_del(inp, tp);
523 tp = tcp_drop(tp, ETIMEDOUT);
526 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
527 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
530 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
531 tcp_inpinfo_lock_del(inp, tp);
537 tcp_timer_persist(void *xtp)
539 struct tcpcb *tp = xtp;
541 CURVNET_SET(tp->t_vnet);
545 ostate = tp->t_state;
548 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
550 if (callout_pending(&tp->t_timers->tt_persist) ||
551 !callout_active(&tp->t_timers->tt_persist)) {
556 callout_deactivate(&tp->t_timers->tt_persist);
557 if ((inp->inp_flags & INP_DROPPED) != 0) {
562 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
563 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
565 * Persistence timer into zero window.
566 * Force a byte to be output, if possible.
568 TCPSTAT_INC(tcps_persisttimeo);
570 * Hack: if the peer is dead/unreachable, we do not
571 * time out if the window is closed. After a full
572 * backoff, drop the connection if the idle time
573 * (no responses to probes) reaches the maximum
574 * backoff that we would use if retransmitting.
576 if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
577 (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
578 ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
579 TCPSTAT_INC(tcps_persistdrop);
580 if (tcp_inpinfo_lock_add(inp)) {
581 tcp_inpinfo_lock_del(inp, tp);
584 tp = tcp_drop(tp, ETIMEDOUT);
585 tcp_inpinfo_lock_del(inp, tp);
589 * If the user has closed the socket then drop a persisting
590 * connection after a much reduced timeout.
592 if (tp->t_state > TCPS_CLOSE_WAIT &&
593 (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
594 TCPSTAT_INC(tcps_persistdrop);
595 if (tcp_inpinfo_lock_add(inp)) {
596 tcp_inpinfo_lock_del(inp, tp);
599 tp = tcp_drop(tp, ETIMEDOUT);
600 tcp_inpinfo_lock_del(inp, tp);
604 tp->t_flags |= TF_FORCEDATA;
605 (void) tp->t_fb->tfb_tcp_output(tp);
606 tp->t_flags &= ~TF_FORCEDATA;
609 if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
610 tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
612 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
619 tcp_timer_rexmt(void * xtp)
621 struct tcpcb *tp = xtp;
622 CURVNET_SET(tp->t_vnet);
628 ostate = tp->t_state;
631 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
633 if (callout_pending(&tp->t_timers->tt_rexmt) ||
634 !callout_active(&tp->t_timers->tt_rexmt)) {
639 callout_deactivate(&tp->t_timers->tt_rexmt);
640 if ((inp->inp_flags & INP_DROPPED) != 0) {
645 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
646 ("%s: tp %p tcpcb can't be stopped here", __func__, tp));
647 tcp_free_sackholes(tp);
648 TCP_LOG_EVENT(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0, NULL, false);
649 if (tp->t_fb->tfb_tcp_rexmit_tmr) {
650 /* The stack has a timer action too. */
651 (*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
654 * Retransmission timer went off. Message has not
655 * been acked within retransmit interval. Back off
656 * to a longer retransmit interval and retransmit one segment.
658 if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
659 tp->t_rxtshift = TCP_MAXRXTSHIFT;
660 TCPSTAT_INC(tcps_timeoutdrop);
661 if (tcp_inpinfo_lock_add(inp)) {
662 tcp_inpinfo_lock_del(inp, tp);
665 tp = tcp_drop(tp, tp->t_softerror ?
666 tp->t_softerror : ETIMEDOUT);
667 tcp_inpinfo_lock_del(inp, tp);
670 if (tp->t_state == TCPS_SYN_SENT) {
672 * If the SYN was retransmitted, indicate CWND to be
673 * limited to 1 segment in cc_conn_init().
676 } else if (tp->t_rxtshift == 1) {
678 * first retransmit; record ssthresh and cwnd so they can
679 * be recovered if this turns out to be a "bad" retransmit.
680 * A retransmit is considered "bad" if an ACK for this
681 * segment is received within RTT/2 interval; the assumption
682 * here is that the ACK was already in flight. See
683 * "On Estimating End-to-End Network Path Properties" by
684 * Allman and Paxson for more details.
686 tp->snd_cwnd_prev = tp->snd_cwnd;
687 tp->snd_ssthresh_prev = tp->snd_ssthresh;
688 tp->snd_recover_prev = tp->snd_recover;
689 if (IN_FASTRECOVERY(tp->t_flags))
690 tp->t_flags |= TF_WASFRECOVERY;
692 tp->t_flags &= ~TF_WASFRECOVERY;
693 if (IN_CONGRECOVERY(tp->t_flags))
694 tp->t_flags |= TF_WASCRECOVERY;
696 tp->t_flags &= ~TF_WASCRECOVERY;
697 tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
698 tp->t_flags |= TF_PREVVALID;
700 tp->t_flags &= ~TF_PREVVALID;
701 TCPSTAT_INC(tcps_rexmttimeo);
702 if ((tp->t_state == TCPS_SYN_SENT) ||
703 (tp->t_state == TCPS_SYN_RECEIVED))
704 rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
706 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
707 TCPT_RANGESET(tp->t_rxtcur, rexmt,
708 tp->t_rttmin, TCPTV_REXMTMAX);
711 * We enter the path for PLMTUD if connection is established or, if
712 * connection is FIN_WAIT_1 status, reason for the last is that if
713 * amount of data we send is very small, we could send it in couple of
714 * packets and process straight to FIN. In that case we won't catch
717 if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
718 || (tp->t_state == TCPS_FIN_WAIT_1))) {
724 * Idea here is that at each stage of mtu probe (usually, 1448
725 * -> 1188 -> 524) should be given 2 chances to recover before
726 * further clamping down. 'tp->t_rxtshift % 2 == 0' should
729 if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
730 (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
731 (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
732 tp->t_rxtshift % 2 == 0)) {
734 * Enter Path MTU Black-hole Detection mechanism:
735 * - Disable Path MTU Discovery (IP "DF" bit).
736 * - Reduce MTU to lower value than what we
737 * negotiated with peer.
739 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
740 /* Record that we may have found a black hole. */
741 tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
742 /* Keep track of previous MSS. */
743 tp->t_pmtud_saved_maxseg = tp->t_maxseg;
747 * Reduce the MSS to blackhole value or to the default
748 * in an attempt to retransmit.
751 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
753 tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
754 /* Use the sysctl tuneable blackhole MSS. */
755 tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
756 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
758 /* Use the default MSS. */
759 tp->t_maxseg = V_tcp_v6mssdflt;
761 * Disable Path MTU Discovery when we switch to
764 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
765 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
768 #if defined(INET6) && defined(INET)
772 if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
773 /* Use the sysctl tuneable blackhole MSS. */
774 tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
775 TCPSTAT_INC(tcps_pmtud_blackhole_activated);
777 /* Use the default MSS. */
778 tp->t_maxseg = V_tcp_mssdflt;
780 * Disable Path MTU Discovery when we switch to
783 tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
784 TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
788 * Reset the slow-start flight size
789 * as it may depend on the new MSS.
791 if (CC_ALGO(tp)->conn_init != NULL)
792 CC_ALGO(tp)->conn_init(tp->ccv);
795 * If further retransmissions are still unsuccessful
796 * with a lowered MTU, maybe this isn't a blackhole and
797 * we restore the previous MSS and blackhole detection
799 * The limit '6' is determined by giving each probe
800 * stage (1448, 1188, 524) 2 chances to recover.
802 if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
803 (tp->t_rxtshift >= 6)) {
804 tp->t_flags2 |= TF2_PLPMTU_PMTUD;
805 tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
806 tp->t_maxseg = tp->t_pmtud_saved_maxseg;
807 TCPSTAT_INC(tcps_pmtud_blackhole_failed);
809 * Reset the slow-start flight size as it
810 * may depend on the new MSS.
812 if (CC_ALGO(tp)->conn_init != NULL)
813 CC_ALGO(tp)->conn_init(tp->ccv);
819 * Disable RFC1323 and SACK if we haven't got any response to
820 * our third SYN to work-around some broken terminal servers
821 * (most of which have hopefully been retired) that have bad VJ
822 * header compression code which trashes TCP segments containing
823 * unknown-to-them TCP options.
825 if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
826 (tp->t_rxtshift == 3))
827 tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
829 * If we backed off this far, notify the L3 protocol that we're having
830 * connection problems.
832 if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
834 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
835 in6_losing(tp->t_inpcb);
838 in_losing(tp->t_inpcb);
840 tp->snd_nxt = tp->snd_una;
841 tp->snd_recover = tp->snd_max;
843 * Force a segment to be sent.
845 tp->t_flags |= TF_ACKNOW;
847 * If timing a segment in this window, stop the timer.
851 cc_cong_signal(tp, NULL, CC_RTO);
853 (void) tp->t_fb->tfb_tcp_output(tp);
856 if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
857 tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
860 TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
867 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
869 struct callout *t_callout;
870 timeout_t *f_callout;
871 struct inpcb *inp = tp->t_inpcb;
872 int cpu = inp_to_cpuid(inp);
875 if (tp->t_flags & TF_TOE)
879 if (tp->t_timers->tt_flags & TT_STOPPED)
882 switch (timer_type) {
884 t_callout = &tp->t_timers->tt_delack;
885 f_callout = tcp_timer_delack;
888 t_callout = &tp->t_timers->tt_rexmt;
889 f_callout = tcp_timer_rexmt;
892 t_callout = &tp->t_timers->tt_persist;
893 f_callout = tcp_timer_persist;
896 t_callout = &tp->t_timers->tt_keep;
897 f_callout = tcp_timer_keep;
900 t_callout = &tp->t_timers->tt_2msl;
901 f_callout = tcp_timer_2msl;
904 if (tp->t_fb->tfb_tcp_timer_activate) {
905 tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
908 panic("tp %p bad timer_type %#x", tp, timer_type);
911 callout_stop(t_callout);
913 callout_reset_on(t_callout, delta, f_callout, tp, cpu);
918 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
920 struct callout *t_callout;
922 switch (timer_type) {
924 t_callout = &tp->t_timers->tt_delack;
927 t_callout = &tp->t_timers->tt_rexmt;
930 t_callout = &tp->t_timers->tt_persist;
933 t_callout = &tp->t_timers->tt_keep;
936 t_callout = &tp->t_timers->tt_2msl;
939 if (tp->t_fb->tfb_tcp_timer_active) {
940 return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
942 panic("tp %p bad timer_type %#x", tp, timer_type);
944 return callout_active(t_callout);
948 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
950 struct callout *t_callout;
952 tp->t_timers->tt_flags |= TT_STOPPED;
953 switch (timer_type) {
955 t_callout = &tp->t_timers->tt_delack;
958 t_callout = &tp->t_timers->tt_rexmt;
961 t_callout = &tp->t_timers->tt_persist;
964 t_callout = &tp->t_timers->tt_keep;
967 t_callout = &tp->t_timers->tt_2msl;
970 if (tp->t_fb->tfb_tcp_timer_stop) {
972 * XXXrrs we need to look at this with the
973 * stop case below (flags).
975 tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
978 panic("tp %p bad timer_type %#x", tp, timer_type);
981 if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
983 * Can't stop the callout, defer tcpcb actual deletion
984 * to the last one. We do this using the async drain
985 * function and incrementing the count in
987 tp->t_timers->tt_draincnt++;