xref: /freebsd/sys/netinet/tcp_timer.c (revision 1171c633fb097a19e1da87128604190bc6d27341)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
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.
18  *
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
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
40 #include "opt_rss.h"
41 
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/protosw.h>
48 #include <sys/smp.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
53 
54 #include <net/if.h>
55 #include <net/route.h>
56 #include <net/rss_config.h>
57 #include <net/vnet.h>
58 #include <net/netisr.h>
59 
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>
65 #ifdef INET6
66 #include <netinet6/in6_pcb.h>
67 #endif
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>
76 #ifdef INET6
77 #include <netinet6/tcp6_var.h>
78 #endif
79 #include <netinet/tcpip.h>
80 #ifdef TCPDEBUG
81 #include <netinet/tcp_debug.h>
82 #endif
83 
84 int    tcp_persmin;
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");
87 
88 int    tcp_persmax;
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");
91 
92 int	tcp_keepinit;
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");
95 
96 int	tcp_keepidle;
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");
99 
100 int	tcp_keepintvl;
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");
103 
104 int	tcp_delacktime;
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");
108 
109 int	tcp_msl;
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");
112 
113 int	tcp_rexmit_initial;
114 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_initial, CTLTYPE_INT|CTLFLAG_RW,
115     &tcp_rexmit_initial, 0, sysctl_msec_to_ticks, "I",
116     "Initial Retransmission Timeout");
117 
118 int	tcp_rexmit_min;
119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
120     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
121     "Minimum Retransmission Timeout");
122 
123 int	tcp_rexmit_slop;
124 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
125     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
126     "Retransmission Timer Slop");
127 
128 VNET_DEFINE(int, tcp_always_keepalive) = 1;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_VNET|CTLFLAG_RW,
130     &VNET_NAME(tcp_always_keepalive) , 0,
131     "Assume SO_KEEPALIVE on all TCP connections");
132 
133 int    tcp_fast_finwait2_recycle = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
135     &tcp_fast_finwait2_recycle, 0,
136     "Recycle closed FIN_WAIT_2 connections faster");
137 
138 int    tcp_finwait2_timeout;
139 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
140     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
141 
142 int	tcp_keepcnt = TCPTV_KEEPCNT;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
144     "Number of keepalive probes to send");
145 
146 	/* max idle probes */
147 int	tcp_maxpersistidle;
148 
149 int	tcp_rexmit_drop_options = 0;
150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
151     &tcp_rexmit_drop_options, 0,
152     "Drop TCP options from 3rd and later retransmitted SYN");
153 
154 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
156     CTLFLAG_RW|CTLFLAG_VNET,
157     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
158     "Path MTU Discovery Black Hole Detection Enabled");
159 
160 #ifdef INET
161 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
163     CTLFLAG_RW|CTLFLAG_VNET,
164     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
165     "Path MTU Discovery Black Hole Detection lowered MSS");
166 #endif
167 
168 #ifdef INET6
169 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
171     CTLFLAG_RW|CTLFLAG_VNET,
172     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
173     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
174 #endif
175 
176 #ifdef	RSS
177 static int	per_cpu_timers = 1;
178 #else
179 static int	per_cpu_timers = 0;
180 #endif
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
182     &per_cpu_timers , 0, "run tcp timers on all cpus");
183 
184 /*
185  * Map the given inp to a CPU id.
186  *
187  * This queries RSS if it's compiled in, else it defaults to the current
188  * CPU ID.
189  */
190 inline int
191 inp_to_cpuid(struct inpcb *inp)
192 {
193 	u_int cpuid;
194 
195 #ifdef	RSS
196 	if (per_cpu_timers) {
197 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
198 		if (cpuid == NETISR_CPUID_NONE)
199 			return (curcpu);	/* XXX */
200 		else
201 			return (cpuid);
202 	}
203 #else
204 	/* Legacy, pre-RSS behaviour */
205 	if (per_cpu_timers) {
206 		/*
207 		 * We don't have a flowid -> cpuid mapping, so cheat and
208 		 * just map unknown cpuids to curcpu.  Not the best, but
209 		 * apparently better than defaulting to swi 0.
210 		 */
211 		cpuid = inp->inp_flowid % (mp_maxid + 1);
212 		if (! CPU_ABSENT(cpuid))
213 			return (cpuid);
214 		return (curcpu);
215 	}
216 #endif
217 	/* Default for RSS and non-RSS - cpuid 0 */
218 	else {
219 		return (0);
220 	}
221 }
222 
223 /*
224  * Tcp protocol timeout routine called every 500 ms.
225  * Updates timestamps used for TCP
226  * causes finite state machine actions if timers expire.
227  */
228 void
229 tcp_slowtimo(void)
230 {
231 	VNET_ITERATOR_DECL(vnet_iter);
232 
233 	VNET_LIST_RLOCK_NOSLEEP();
234 	VNET_FOREACH(vnet_iter) {
235 		CURVNET_SET(vnet_iter);
236 		(void) tcp_tw_2msl_scan(0);
237 		CURVNET_RESTORE();
238 	}
239 	VNET_LIST_RUNLOCK_NOSLEEP();
240 }
241 
242 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
243     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
244 
245 int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
246 
247 /*
248  * TCP timer processing.
249  */
250 
251 void
252 tcp_timer_delack(void *xtp)
253 {
254 	struct tcpcb *tp = xtp;
255 	struct inpcb *inp;
256 	CURVNET_SET(tp->t_vnet);
257 
258 	inp = tp->t_inpcb;
259 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
260 	INP_WLOCK(inp);
261 	if (callout_pending(&tp->t_timers->tt_delack) ||
262 	    !callout_active(&tp->t_timers->tt_delack)) {
263 		INP_WUNLOCK(inp);
264 		CURVNET_RESTORE();
265 		return;
266 	}
267 	callout_deactivate(&tp->t_timers->tt_delack);
268 	if ((inp->inp_flags & INP_DROPPED) != 0) {
269 		INP_WUNLOCK(inp);
270 		CURVNET_RESTORE();
271 		return;
272 	}
273 	tp->t_flags |= TF_ACKNOW;
274 	TCPSTAT_INC(tcps_delack);
275 	(void) tp->t_fb->tfb_tcp_output(tp);
276 	INP_WUNLOCK(inp);
277 	CURVNET_RESTORE();
278 }
279 
280 void
281 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
282 {
283 	if (inp && tp != NULL)
284 		INP_WUNLOCK(inp);
285 }
286 
287 void
288 tcp_timer_2msl(void *xtp)
289 {
290 	struct tcpcb *tp = xtp;
291 	struct inpcb *inp;
292 	struct epoch_tracker et;
293 	CURVNET_SET(tp->t_vnet);
294 #ifdef TCPDEBUG
295 	int ostate;
296 
297 	ostate = tp->t_state;
298 #endif
299 	inp = tp->t_inpcb;
300 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
301 	INP_WLOCK(inp);
302 	tcp_free_sackholes(tp);
303 	if (callout_pending(&tp->t_timers->tt_2msl) ||
304 	    !callout_active(&tp->t_timers->tt_2msl)) {
305 		INP_WUNLOCK(tp->t_inpcb);
306 		CURVNET_RESTORE();
307 		return;
308 	}
309 	callout_deactivate(&tp->t_timers->tt_2msl);
310 	if ((inp->inp_flags & INP_DROPPED) != 0) {
311 		INP_WUNLOCK(inp);
312 		CURVNET_RESTORE();
313 		return;
314 	}
315 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
316 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
317 	/*
318 	 * 2 MSL timeout in shutdown went off.  If we're closed but
319 	 * still waiting for peer to close and connection has been idle
320 	 * too long delete connection control block.  Otherwise, check
321 	 * again in a bit.
322 	 *
323 	 * If in TIME_WAIT state just ignore as this timeout is handled in
324 	 * tcp_tw_2msl_scan().
325 	 *
326 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
327 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
328 	 * Ignore fact that there were recent incoming segments.
329 	 */
330 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
331 		INP_WUNLOCK(inp);
332 		CURVNET_RESTORE();
333 		return;
334 	}
335 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
336 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
337 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
338 		TCPSTAT_INC(tcps_finwait2_drops);
339 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
340 			tcp_inpinfo_lock_del(inp, tp);
341 			goto out;
342 		}
343 		NET_EPOCH_ENTER(et);
344 		tp = tcp_close(tp);
345 		NET_EPOCH_EXIT(et);
346 		tcp_inpinfo_lock_del(inp, tp);
347 		goto out;
348 	} else {
349 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
350 			callout_reset(&tp->t_timers->tt_2msl,
351 				      TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
352 		} else {
353 			if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
354 				tcp_inpinfo_lock_del(inp, tp);
355 				goto out;
356 			}
357 			NET_EPOCH_ENTER(et);
358 			tp = tcp_close(tp);
359 			NET_EPOCH_EXIT(et);
360 			tcp_inpinfo_lock_del(inp, tp);
361 			goto out;
362 		}
363 	}
364 
365 #ifdef TCPDEBUG
366 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
367 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
368 			  PRU_SLOWTIMO);
369 #endif
370 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
371 
372 	if (tp != NULL)
373 		INP_WUNLOCK(inp);
374 out:
375 	CURVNET_RESTORE();
376 }
377 
378 void
379 tcp_timer_keep(void *xtp)
380 {
381 	struct tcpcb *tp = xtp;
382 	struct tcptemp *t_template;
383 	struct inpcb *inp;
384 	struct epoch_tracker et;
385 	CURVNET_SET(tp->t_vnet);
386 #ifdef TCPDEBUG
387 	int ostate;
388 
389 	ostate = tp->t_state;
390 #endif
391 	inp = tp->t_inpcb;
392 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
393 	INP_WLOCK(inp);
394 	if (callout_pending(&tp->t_timers->tt_keep) ||
395 	    !callout_active(&tp->t_timers->tt_keep)) {
396 		INP_WUNLOCK(inp);
397 		CURVNET_RESTORE();
398 		return;
399 	}
400 	callout_deactivate(&tp->t_timers->tt_keep);
401 	if ((inp->inp_flags & INP_DROPPED) != 0) {
402 		INP_WUNLOCK(inp);
403 		CURVNET_RESTORE();
404 		return;
405 	}
406 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
407 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
408 
409 	/*
410 	 * Because we don't regularly reset the keepalive callout in
411 	 * the ESTABLISHED state, it may be that we don't actually need
412 	 * to send a keepalive yet. If that occurs, schedule another
413 	 * call for the next time the keepalive timer might expire.
414 	 */
415 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
416 		u_int idletime;
417 
418 		idletime = ticks - tp->t_rcvtime;
419 		if (idletime < TP_KEEPIDLE(tp)) {
420 			callout_reset(&tp->t_timers->tt_keep,
421 			    TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
422 			INP_WUNLOCK(inp);
423 			CURVNET_RESTORE();
424 			return;
425 		}
426 	}
427 
428 	/*
429 	 * Keep-alive timer went off; send something
430 	 * or drop connection if idle for too long.
431 	 */
432 	TCPSTAT_INC(tcps_keeptimeo);
433 	if (tp->t_state < TCPS_ESTABLISHED)
434 		goto dropit;
435 	if ((V_tcp_always_keepalive ||
436 	    inp->inp_socket->so_options & SO_KEEPALIVE) &&
437 	    tp->t_state <= TCPS_CLOSING) {
438 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
439 			goto dropit;
440 		/*
441 		 * Send a packet designed to force a response
442 		 * if the peer is up and reachable:
443 		 * either an ACK if the connection is still alive,
444 		 * or an RST if the peer has closed the connection
445 		 * due to timeout or reboot.
446 		 * Using sequence number tp->snd_una-1
447 		 * causes the transmitted zero-length segment
448 		 * to lie outside the receive window;
449 		 * by the protocol spec, this requires the
450 		 * correspondent TCP to respond.
451 		 */
452 		TCPSTAT_INC(tcps_keepprobe);
453 		t_template = tcpip_maketemplate(inp);
454 		if (t_template) {
455 			tcp_respond(tp, t_template->tt_ipgen,
456 				    &t_template->tt_t, (struct mbuf *)NULL,
457 				    tp->rcv_nxt, tp->snd_una - 1, 0);
458 			free(t_template, M_TEMP);
459 		}
460 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
461 			      tcp_timer_keep, tp);
462 	} else
463 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
464 			      tcp_timer_keep, tp);
465 
466 #ifdef TCPDEBUG
467 	if (inp->inp_socket->so_options & SO_DEBUG)
468 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
469 			  PRU_SLOWTIMO);
470 #endif
471 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
472 	INP_WUNLOCK(inp);
473 	CURVNET_RESTORE();
474 	return;
475 
476 dropit:
477 	TCPSTAT_INC(tcps_keepdrops);
478 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
479 		tcp_inpinfo_lock_del(inp, tp);
480 		goto out;
481 	}
482 	NET_EPOCH_ENTER(et);
483 	tp = tcp_drop(tp, ETIMEDOUT);
484 
485 #ifdef TCPDEBUG
486 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
487 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
488 			  PRU_SLOWTIMO);
489 #endif
490 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
491 	NET_EPOCH_EXIT(et);
492 	tcp_inpinfo_lock_del(inp, tp);
493  out:
494 	CURVNET_RESTORE();
495 }
496 
497 void
498 tcp_timer_persist(void *xtp)
499 {
500 	struct tcpcb *tp = xtp;
501 	struct inpcb *inp;
502 	struct epoch_tracker et;
503 	CURVNET_SET(tp->t_vnet);
504 #ifdef TCPDEBUG
505 	int ostate;
506 
507 	ostate = tp->t_state;
508 #endif
509 	inp = tp->t_inpcb;
510 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
511 	INP_WLOCK(inp);
512 	if (callout_pending(&tp->t_timers->tt_persist) ||
513 	    !callout_active(&tp->t_timers->tt_persist)) {
514 		INP_WUNLOCK(inp);
515 		CURVNET_RESTORE();
516 		return;
517 	}
518 	callout_deactivate(&tp->t_timers->tt_persist);
519 	if ((inp->inp_flags & INP_DROPPED) != 0) {
520 		INP_WUNLOCK(inp);
521 		CURVNET_RESTORE();
522 		return;
523 	}
524 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
525 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
526 	/*
527 	 * Persistence timer into zero window.
528 	 * Force a byte to be output, if possible.
529 	 */
530 	TCPSTAT_INC(tcps_persisttimeo);
531 	/*
532 	 * Hack: if the peer is dead/unreachable, we do not
533 	 * time out if the window is closed.  After a full
534 	 * backoff, drop the connection if the idle time
535 	 * (no responses to probes) reaches the maximum
536 	 * backoff that we would use if retransmitting.
537 	 */
538 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
539 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
540 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
541 		TCPSTAT_INC(tcps_persistdrop);
542 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
543 			tcp_inpinfo_lock_del(inp, tp);
544 			goto out;
545 		}
546 		NET_EPOCH_ENTER(et);
547 		tp = tcp_drop(tp, ETIMEDOUT);
548 		NET_EPOCH_EXIT(et);
549 		tcp_inpinfo_lock_del(inp, tp);
550 		goto out;
551 	}
552 	/*
553 	 * If the user has closed the socket then drop a persisting
554 	 * connection after a much reduced timeout.
555 	 */
556 	if (tp->t_state > TCPS_CLOSE_WAIT &&
557 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
558 		TCPSTAT_INC(tcps_persistdrop);
559 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
560 			tcp_inpinfo_lock_del(inp, tp);
561 			goto out;
562 		}
563 		NET_EPOCH_ENTER(et);
564 		tp = tcp_drop(tp, ETIMEDOUT);
565 		NET_EPOCH_EXIT(et);
566 		tcp_inpinfo_lock_del(inp, tp);
567 		goto out;
568 	}
569 	tcp_setpersist(tp);
570 	tp->t_flags |= TF_FORCEDATA;
571 	(void) tp->t_fb->tfb_tcp_output(tp);
572 	tp->t_flags &= ~TF_FORCEDATA;
573 
574 #ifdef TCPDEBUG
575 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
576 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
577 #endif
578 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
579 	INP_WUNLOCK(inp);
580 out:
581 	CURVNET_RESTORE();
582 }
583 
584 void
585 tcp_timer_rexmt(void * xtp)
586 {
587 	struct tcpcb *tp = xtp;
588 	CURVNET_SET(tp->t_vnet);
589 	int rexmt;
590 	struct inpcb *inp;
591 	struct epoch_tracker et;
592 #ifdef TCPDEBUG
593 	int ostate;
594 
595 	ostate = tp->t_state;
596 #endif
597 	inp = tp->t_inpcb;
598 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
599 	INP_WLOCK(inp);
600 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
601 	    !callout_active(&tp->t_timers->tt_rexmt)) {
602 		INP_WUNLOCK(inp);
603 		CURVNET_RESTORE();
604 		return;
605 	}
606 	callout_deactivate(&tp->t_timers->tt_rexmt);
607 	if ((inp->inp_flags & INP_DROPPED) != 0) {
608 		INP_WUNLOCK(inp);
609 		CURVNET_RESTORE();
610 		return;
611 	}
612 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
613 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
614 	tcp_free_sackholes(tp);
615 	TCP_LOG_EVENT(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0, NULL, false);
616 	if (tp->t_fb->tfb_tcp_rexmit_tmr) {
617 		/* The stack has a timer action too. */
618 		(*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
619 	}
620 	/*
621 	 * Retransmission timer went off.  Message has not
622 	 * been acked within retransmit interval.  Back off
623 	 * to a longer retransmit interval and retransmit one segment.
624 	 */
625 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
626 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
627 		TCPSTAT_INC(tcps_timeoutdrop);
628 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
629 			tcp_inpinfo_lock_del(inp, tp);
630 			goto out;
631 		}
632 		NET_EPOCH_ENTER(et);
633 		tp = tcp_drop(tp, ETIMEDOUT);
634 		NET_EPOCH_EXIT(et);
635 		tcp_inpinfo_lock_del(inp, tp);
636 		goto out;
637 	}
638 	if (tp->t_state == TCPS_SYN_SENT) {
639 		/*
640 		 * If the SYN was retransmitted, indicate CWND to be
641 		 * limited to 1 segment in cc_conn_init().
642 		 */
643 		tp->snd_cwnd = 1;
644 	} else if (tp->t_rxtshift == 1) {
645 		/*
646 		 * first retransmit; record ssthresh and cwnd so they can
647 		 * be recovered if this turns out to be a "bad" retransmit.
648 		 * A retransmit is considered "bad" if an ACK for this
649 		 * segment is received within RTT/2 interval; the assumption
650 		 * here is that the ACK was already in flight.  See
651 		 * "On Estimating End-to-End Network Path Properties" by
652 		 * Allman and Paxson for more details.
653 		 */
654 		tp->snd_cwnd_prev = tp->snd_cwnd;
655 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
656 		tp->snd_recover_prev = tp->snd_recover;
657 		if (IN_FASTRECOVERY(tp->t_flags))
658 			tp->t_flags |= TF_WASFRECOVERY;
659 		else
660 			tp->t_flags &= ~TF_WASFRECOVERY;
661 		if (IN_CONGRECOVERY(tp->t_flags))
662 			tp->t_flags |= TF_WASCRECOVERY;
663 		else
664 			tp->t_flags &= ~TF_WASCRECOVERY;
665 		if ((tp->t_flags & TF_RCVD_TSTMP) == 0)
666 			tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
667 		/* In the event that we've negotiated timestamps
668 		 * badrxtwin will be set to the value that we set
669 		 * the retransmitted packet's to_tsval to by tcp_output
670 		 */
671 		tp->t_flags |= TF_PREVVALID;
672 	} else
673 		tp->t_flags &= ~TF_PREVVALID;
674 	TCPSTAT_INC(tcps_rexmttimeo);
675 	if ((tp->t_state == TCPS_SYN_SENT) ||
676 	    (tp->t_state == TCPS_SYN_RECEIVED))
677 		rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
678 	else
679 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
680 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
681 		      tp->t_rttmin, TCPTV_REXMTMAX);
682 
683 	/*
684 	 * We enter the path for PLMTUD if connection is established or, if
685 	 * connection is FIN_WAIT_1 status, reason for the last is that if
686 	 * amount of data we send is very small, we could send it in couple of
687 	 * packets and process straight to FIN. In that case we won't catch
688 	 * ESTABLISHED state.
689 	 */
690 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
691 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
692 #ifdef INET6
693 		int isipv6;
694 #endif
695 
696 		/*
697 		 * Idea here is that at each stage of mtu probe (usually, 1448
698 		 * -> 1188 -> 524) should be given 2 chances to recover before
699 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
700 		 *  take care of that.
701 		 */
702 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
703 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
704 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
705 		    tp->t_rxtshift % 2 == 0)) {
706 			/*
707 			 * Enter Path MTU Black-hole Detection mechanism:
708 			 * - Disable Path MTU Discovery (IP "DF" bit).
709 			 * - Reduce MTU to lower value than what we
710 			 *   negotiated with peer.
711 			 */
712 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
713 				/* Record that we may have found a black hole. */
714 				tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
715 				/* Keep track of previous MSS. */
716 				tp->t_pmtud_saved_maxseg = tp->t_maxseg;
717 			}
718 
719 			/*
720 			 * Reduce the MSS to blackhole value or to the default
721 			 * in an attempt to retransmit.
722 			 */
723 #ifdef INET6
724 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
725 			if (isipv6 &&
726 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
727 				/* Use the sysctl tuneable blackhole MSS. */
728 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
729 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
730 			} else if (isipv6) {
731 				/* Use the default MSS. */
732 				tp->t_maxseg = V_tcp_v6mssdflt;
733 				/*
734 				 * Disable Path MTU Discovery when we switch to
735 				 * minmss.
736 				 */
737 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
738 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
739 			}
740 #endif
741 #if defined(INET6) && defined(INET)
742 			else
743 #endif
744 #ifdef INET
745 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
746 				/* Use the sysctl tuneable blackhole MSS. */
747 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
748 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
749 			} else {
750 				/* Use the default MSS. */
751 				tp->t_maxseg = V_tcp_mssdflt;
752 				/*
753 				 * Disable Path MTU Discovery when we switch to
754 				 * minmss.
755 				 */
756 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
757 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
758 			}
759 #endif
760 			/*
761 			 * Reset the slow-start flight size
762 			 * as it may depend on the new MSS.
763 			 */
764 			if (CC_ALGO(tp)->conn_init != NULL)
765 				CC_ALGO(tp)->conn_init(tp->ccv);
766 		} else {
767 			/*
768 			 * If further retransmissions are still unsuccessful
769 			 * with a lowered MTU, maybe this isn't a blackhole and
770 			 * we restore the previous MSS and blackhole detection
771 			 * flags.
772 			 * The limit '6' is determined by giving each probe
773 			 * stage (1448, 1188, 524) 2 chances to recover.
774 			 */
775 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
776 			    (tp->t_rxtshift >= 6)) {
777 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
778 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
779 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
780 				TCPSTAT_INC(tcps_pmtud_blackhole_failed);
781 				/*
782 				 * Reset the slow-start flight size as it
783 				 * may depend on the new MSS.
784 				 */
785 				if (CC_ALGO(tp)->conn_init != NULL)
786 					CC_ALGO(tp)->conn_init(tp->ccv);
787 			}
788 		}
789 	}
790 
791 	/*
792 	 * Disable RFC1323 and SACK if we haven't got any response to
793 	 * our third SYN to work-around some broken terminal servers
794 	 * (most of which have hopefully been retired) that have bad VJ
795 	 * header compression code which trashes TCP segments containing
796 	 * unknown-to-them TCP options.
797 	 */
798 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
799 	    (tp->t_rxtshift == 3))
800 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
801 	/*
802 	 * If we backed off this far, notify the L3 protocol that we're having
803 	 * connection problems.
804 	 */
805 	if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
806 #ifdef INET6
807 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
808 			in6_losing(tp->t_inpcb);
809 		else
810 #endif
811 			in_losing(tp->t_inpcb);
812 	}
813 	tp->snd_nxt = tp->snd_una;
814 	tp->snd_recover = tp->snd_max;
815 	/*
816 	 * Force a segment to be sent.
817 	 */
818 	tp->t_flags |= TF_ACKNOW;
819 	/*
820 	 * If timing a segment in this window, stop the timer.
821 	 */
822 	tp->t_rtttime = 0;
823 
824 	cc_cong_signal(tp, NULL, CC_RTO);
825 
826 	(void) tp->t_fb->tfb_tcp_output(tp);
827 
828 #ifdef TCPDEBUG
829 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
830 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
831 			  PRU_SLOWTIMO);
832 #endif
833 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
834 	INP_WUNLOCK(inp);
835 out:
836 	CURVNET_RESTORE();
837 }
838 
839 void
840 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
841 {
842 	struct callout *t_callout;
843 	callout_func_t *f_callout;
844 	struct inpcb *inp = tp->t_inpcb;
845 	int cpu = inp_to_cpuid(inp);
846 
847 #ifdef TCP_OFFLOAD
848 	if (tp->t_flags & TF_TOE)
849 		return;
850 #endif
851 
852 	if (tp->t_timers->tt_flags & TT_STOPPED)
853 		return;
854 
855 	switch (timer_type) {
856 		case TT_DELACK:
857 			t_callout = &tp->t_timers->tt_delack;
858 			f_callout = tcp_timer_delack;
859 			break;
860 		case TT_REXMT:
861 			t_callout = &tp->t_timers->tt_rexmt;
862 			f_callout = tcp_timer_rexmt;
863 			break;
864 		case TT_PERSIST:
865 			t_callout = &tp->t_timers->tt_persist;
866 			f_callout = tcp_timer_persist;
867 			break;
868 		case TT_KEEP:
869 			t_callout = &tp->t_timers->tt_keep;
870 			f_callout = tcp_timer_keep;
871 			break;
872 		case TT_2MSL:
873 			t_callout = &tp->t_timers->tt_2msl;
874 			f_callout = tcp_timer_2msl;
875 			break;
876 		default:
877 			if (tp->t_fb->tfb_tcp_timer_activate) {
878 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
879 				return;
880 			}
881 			panic("tp %p bad timer_type %#x", tp, timer_type);
882 		}
883 	if (delta == 0) {
884 		callout_stop(t_callout);
885 	} else {
886 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
887 	}
888 }
889 
890 int
891 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
892 {
893 	struct callout *t_callout;
894 
895 	switch (timer_type) {
896 		case TT_DELACK:
897 			t_callout = &tp->t_timers->tt_delack;
898 			break;
899 		case TT_REXMT:
900 			t_callout = &tp->t_timers->tt_rexmt;
901 			break;
902 		case TT_PERSIST:
903 			t_callout = &tp->t_timers->tt_persist;
904 			break;
905 		case TT_KEEP:
906 			t_callout = &tp->t_timers->tt_keep;
907 			break;
908 		case TT_2MSL:
909 			t_callout = &tp->t_timers->tt_2msl;
910 			break;
911 		default:
912 			if (tp->t_fb->tfb_tcp_timer_active) {
913 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
914 			}
915 			panic("tp %p bad timer_type %#x", tp, timer_type);
916 		}
917 	return callout_active(t_callout);
918 }
919 
920 /*
921  * Stop the timer from running, and apply a flag
922  * against the timer_flags that will force the
923  * timer never to run. The flag is needed to assure
924  * a race does not leave it running and cause
925  * the timer to possibly restart itself (keep and persist
926  * especially do this).
927  */
928 int
929 tcp_timer_suspend(struct tcpcb *tp, uint32_t timer_type)
930 {
931 	struct callout *t_callout;
932 	uint32_t t_flags;
933 
934 	switch (timer_type) {
935 		case TT_DELACK:
936 			t_flags = TT_DELACK_SUS;
937 			t_callout = &tp->t_timers->tt_delack;
938 			break;
939 		case TT_REXMT:
940 			t_flags = TT_REXMT_SUS;
941 			t_callout = &tp->t_timers->tt_rexmt;
942 			break;
943 		case TT_PERSIST:
944 			t_flags = TT_PERSIST_SUS;
945 			t_callout = &tp->t_timers->tt_persist;
946 			break;
947 		case TT_KEEP:
948 			t_flags = TT_KEEP_SUS;
949 			t_callout = &tp->t_timers->tt_keep;
950 			break;
951 		case TT_2MSL:
952 			t_flags = TT_2MSL_SUS;
953 			t_callout = &tp->t_timers->tt_2msl;
954 			break;
955 		default:
956 			panic("tp:%p bad timer_type 0x%x", tp, timer_type);
957 	}
958 	tp->t_timers->tt_flags |= t_flags;
959 	return (callout_stop(t_callout));
960 }
961 
962 void
963 tcp_timers_unsuspend(struct tcpcb *tp, uint32_t timer_type)
964 {
965 	switch (timer_type) {
966 		case TT_DELACK:
967 			if (tp->t_timers->tt_flags & TT_DELACK_SUS) {
968 				tp->t_timers->tt_flags &= ~TT_DELACK_SUS;
969 				if (tp->t_flags & TF_DELACK) {
970 					/* Delayed ack timer should be up activate a timer */
971 					tp->t_flags &= ~TF_DELACK;
972 					tcp_timer_activate(tp, TT_DELACK,
973 					    tcp_delacktime);
974 				}
975 			}
976 			break;
977 		case TT_REXMT:
978 			if (tp->t_timers->tt_flags & TT_REXMT_SUS) {
979 				tp->t_timers->tt_flags &= ~TT_REXMT_SUS;
980 				if (SEQ_GT(tp->snd_max, tp->snd_una) &&
981 				    (tcp_timer_active((tp), TT_PERSIST) == 0) &&
982 				    tp->snd_wnd) {
983 					/* We have outstanding data activate a timer */
984 					tcp_timer_activate(tp, TT_REXMT,
985                                             tp->t_rxtcur);
986 				}
987 			}
988 			break;
989 		case TT_PERSIST:
990 			if (tp->t_timers->tt_flags & TT_PERSIST_SUS) {
991 				tp->t_timers->tt_flags &= ~TT_PERSIST_SUS;
992 				if (tp->snd_wnd == 0) {
993 					/* Activate the persists timer */
994 					tp->t_rxtshift = 0;
995 					tcp_setpersist(tp);
996 				}
997 			}
998 			break;
999 		case TT_KEEP:
1000 			if (tp->t_timers->tt_flags & TT_KEEP_SUS) {
1001 				tp->t_timers->tt_flags &= ~TT_KEEP_SUS;
1002 				tcp_timer_activate(tp, TT_KEEP,
1003 					    TCPS_HAVEESTABLISHED(tp->t_state) ?
1004 					    TP_KEEPIDLE(tp) : TP_KEEPINIT(tp));
1005 			}
1006 			break;
1007 		case TT_2MSL:
1008 			if (tp->t_timers->tt_flags &= TT_2MSL_SUS) {
1009 				tp->t_timers->tt_flags &= ~TT_2MSL_SUS;
1010 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1011 				    ((tp->t_inpcb->inp_socket == NULL) ||
1012 				     (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE))) {
1013 					/* Star the 2MSL timer */
1014 					tcp_timer_activate(tp, TT_2MSL,
1015 					    (tcp_fast_finwait2_recycle) ?
1016 					    tcp_finwait2_timeout : TP_MAXIDLE(tp));
1017 				}
1018 			}
1019 			break;
1020 		default:
1021 			panic("tp:%p bad timer_type 0x%x", tp, timer_type);
1022 	}
1023 }
1024 
1025 void
1026 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
1027 {
1028 	struct callout *t_callout;
1029 
1030 	tp->t_timers->tt_flags |= TT_STOPPED;
1031 	switch (timer_type) {
1032 		case TT_DELACK:
1033 			t_callout = &tp->t_timers->tt_delack;
1034 			break;
1035 		case TT_REXMT:
1036 			t_callout = &tp->t_timers->tt_rexmt;
1037 			break;
1038 		case TT_PERSIST:
1039 			t_callout = &tp->t_timers->tt_persist;
1040 			break;
1041 		case TT_KEEP:
1042 			t_callout = &tp->t_timers->tt_keep;
1043 			break;
1044 		case TT_2MSL:
1045 			t_callout = &tp->t_timers->tt_2msl;
1046 			break;
1047 		default:
1048 			if (tp->t_fb->tfb_tcp_timer_stop) {
1049 				/*
1050 				 * XXXrrs we need to look at this with the
1051 				 * stop case below (flags).
1052 				 */
1053 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
1054 				return;
1055 			}
1056 			panic("tp %p bad timer_type %#x", tp, timer_type);
1057 		}
1058 
1059 	if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
1060 		/*
1061 		 * Can't stop the callout, defer tcpcb actual deletion
1062 		 * to the last one. We do this using the async drain
1063 		 * function and incrementing the count in
1064 		 */
1065 		tp->t_timers->tt_draincnt++;
1066 	}
1067 }
1068