xref: /freebsd/sys/netinet/tcp_timer.c (revision 4313cc83440a39bdf976f955b1d4d3f3c4d1552f)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
16  *
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
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_tcpdebug.h"
38 #include "opt_rss.h"
39 
40 #include <sys/param.h>
41 #include <sys/kernel.h>
42 #include <sys/lock.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/protosw.h>
46 #include <sys/smp.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 #include <sys/systm.h>
51 
52 #include <net/if.h>
53 #include <net/route.h>
54 #include <net/vnet.h>
55 #include <net/netisr.h>
56 
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>
62 #ifdef INET6
63 #include <netinet6/in6_pcb.h>
64 #endif
65 #include <netinet/ip_var.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_timer.h>
68 #include <netinet/tcp_var.h>
69 #include <netinet/tcpip.h>
70 #ifdef TCPDEBUG
71 #include <netinet/tcp_debug.h>
72 #endif
73 
74 int	tcp_keepinit;
75 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
76     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
77 
78 int	tcp_keepidle;
79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
80     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
81 
82 int	tcp_keepintvl;
83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
84     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
85 
86 int	tcp_delacktime;
87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
88     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
89     "Time before a delayed ACK is sent");
90 
91 int	tcp_msl;
92 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
93     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
94 
95 int	tcp_rexmit_min;
96 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
97     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
98     "Minimum Retransmission Timeout");
99 
100 int	tcp_rexmit_slop;
101 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
102     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
103     "Retransmission Timer Slop");
104 
105 static int	always_keepalive = 1;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
107     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
108 
109 int    tcp_fast_finwait2_recycle = 0;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
111     &tcp_fast_finwait2_recycle, 0,
112     "Recycle closed FIN_WAIT_2 connections faster");
113 
114 int    tcp_finwait2_timeout;
115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
116     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
117 
118 int	tcp_keepcnt = TCPTV_KEEPCNT;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
120     "Number of keepalive probes to send");
121 
122 	/* max idle probes */
123 int	tcp_maxpersistidle;
124 
125 static int	tcp_rexmit_drop_options = 0;
126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
127     &tcp_rexmit_drop_options, 0,
128     "Drop TCP options from 3rd and later retransmitted SYN");
129 
130 static int	per_cpu_timers = 0;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
132     &per_cpu_timers , 0, "run tcp timers on all cpus");
133 
134 #if 0
135 #define	INP_CPU(inp)	(per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
136 		((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
137 #endif
138 
139 /*
140  * Map the given inp to a CPU id.
141  *
142  * This queries RSS if it's compiled in, else it defaults to the current
143  * CPU ID.
144  */
145 static inline int
146 inp_to_cpuid(struct inpcb *inp)
147 {
148 	u_int cpuid;
149 
150 #ifdef	RSS
151 	if (per_cpu_timers) {
152 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
153 		if (cpuid == NETISR_CPUID_NONE)
154 			return (curcpu);	/* XXX */
155 		else
156 			return (cpuid);
157 	}
158 #else
159 	/* Legacy, pre-RSS behaviour */
160 	if (per_cpu_timers) {
161 		/*
162 		 * We don't have a flowid -> cpuid mapping, so cheat and
163 		 * just map unknown cpuids to curcpu.  Not the best, but
164 		 * apparently better than defaulting to swi 0.
165 		 */
166 		cpuid = inp->inp_flowid % (mp_maxid + 1);
167 		if (! CPU_ABSENT(cpuid))
168 			return (cpuid);
169 		return (curcpu);
170 	}
171 #endif
172 	/* Default for RSS and non-RSS - cpuid 0 */
173 	else {
174 		return (0);
175 	}
176 }
177 
178 /*
179  * Tcp protocol timeout routine called every 500 ms.
180  * Updates timestamps used for TCP
181  * causes finite state machine actions if timers expire.
182  */
183 void
184 tcp_slowtimo(void)
185 {
186 	VNET_ITERATOR_DECL(vnet_iter);
187 
188 	VNET_LIST_RLOCK_NOSLEEP();
189 	VNET_FOREACH(vnet_iter) {
190 		CURVNET_SET(vnet_iter);
191 		tcp_tw_2msl_scan();
192 		CURVNET_RESTORE();
193 	}
194 	VNET_LIST_RUNLOCK_NOSLEEP();
195 }
196 
197 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
198     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
199 
200 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
201     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
202 
203 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
204 
205 static int tcp_timer_race;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, timer_race, CTLFLAG_RD, &tcp_timer_race,
207     0, "Count of t_inpcb races on tcp_discardcb");
208 
209 /*
210  * TCP timer processing.
211  */
212 
213 void
214 tcp_timer_delack(void *xtp)
215 {
216 	struct tcpcb *tp = xtp;
217 	struct inpcb *inp;
218 	CURVNET_SET(tp->t_vnet);
219 
220 	inp = tp->t_inpcb;
221 	/*
222 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
223 	 * tear-down mean we need it as a work-around for races between
224 	 * timers and tcp_discardcb().
225 	 *
226 	 * KASSERT(inp != NULL, ("tcp_timer_delack: inp == NULL"));
227 	 */
228 	if (inp == NULL) {
229 		tcp_timer_race++;
230 		CURVNET_RESTORE();
231 		return;
232 	}
233 	INP_WLOCK(inp);
234 	if (callout_pending(&tp->t_timers->tt_delack) ||
235 	    !callout_active(&tp->t_timers->tt_delack)) {
236 		INP_WUNLOCK(inp);
237 		CURVNET_RESTORE();
238 		return;
239 	}
240 	callout_deactivate(&tp->t_timers->tt_delack);
241 	if ((inp->inp_flags & INP_DROPPED) != 0) {
242 		INP_WUNLOCK(inp);
243 		CURVNET_RESTORE();
244 		return;
245 	}
246 
247 	tp->t_flags |= TF_ACKNOW;
248 	TCPSTAT_INC(tcps_delack);
249 	(void) tcp_output(tp);
250 	INP_WUNLOCK(inp);
251 	CURVNET_RESTORE();
252 }
253 
254 void
255 tcp_timer_2msl(void *xtp)
256 {
257 	struct tcpcb *tp = xtp;
258 	struct inpcb *inp;
259 	CURVNET_SET(tp->t_vnet);
260 #ifdef TCPDEBUG
261 	int ostate;
262 
263 	ostate = tp->t_state;
264 #endif
265 	/*
266 	 * XXXRW: Does this actually happen?
267 	 */
268 	INP_INFO_WLOCK(&V_tcbinfo);
269 	inp = tp->t_inpcb;
270 	/*
271 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
272 	 * tear-down mean we need it as a work-around for races between
273 	 * timers and tcp_discardcb().
274 	 *
275 	 * KASSERT(inp != NULL, ("tcp_timer_2msl: inp == NULL"));
276 	 */
277 	if (inp == NULL) {
278 		tcp_timer_race++;
279 		INP_INFO_WUNLOCK(&V_tcbinfo);
280 		CURVNET_RESTORE();
281 		return;
282 	}
283 	INP_WLOCK(inp);
284 	tcp_free_sackholes(tp);
285 	if (callout_pending(&tp->t_timers->tt_2msl) ||
286 	    !callout_active(&tp->t_timers->tt_2msl)) {
287 		INP_WUNLOCK(tp->t_inpcb);
288 		INP_INFO_WUNLOCK(&V_tcbinfo);
289 		CURVNET_RESTORE();
290 		return;
291 	}
292 	callout_deactivate(&tp->t_timers->tt_2msl);
293 	if ((inp->inp_flags & INP_DROPPED) != 0) {
294 		INP_WUNLOCK(inp);
295 		INP_INFO_WUNLOCK(&V_tcbinfo);
296 		CURVNET_RESTORE();
297 		return;
298 	}
299 	/*
300 	 * 2 MSL timeout in shutdown went off.  If we're closed but
301 	 * still waiting for peer to close and connection has been idle
302 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
303 	 * control block.  Otherwise, check again in a bit.
304 	 *
305 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
306 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
307 	 * Ignore fact that there were recent incoming segments.
308 	 */
309 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
310 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
311 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
312 		TCPSTAT_INC(tcps_finwait2_drops);
313 		tp = tcp_close(tp);
314 	} else {
315 		if (tp->t_state != TCPS_TIME_WAIT &&
316 		   ticks - tp->t_rcvtime <= TP_MAXIDLE(tp))
317 		       callout_reset_on(&tp->t_timers->tt_2msl,
318 			   TP_KEEPINTVL(tp), tcp_timer_2msl, tp,
319 			   inp_to_cpuid(inp));
320 	       else
321 		       tp = tcp_close(tp);
322        }
323 
324 #ifdef TCPDEBUG
325 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
326 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
327 			  PRU_SLOWTIMO);
328 #endif
329 	if (tp != NULL)
330 		INP_WUNLOCK(inp);
331 	INP_INFO_WUNLOCK(&V_tcbinfo);
332 	CURVNET_RESTORE();
333 }
334 
335 void
336 tcp_timer_keep(void *xtp)
337 {
338 	struct tcpcb *tp = xtp;
339 	struct tcptemp *t_template;
340 	struct inpcb *inp;
341 	CURVNET_SET(tp->t_vnet);
342 #ifdef TCPDEBUG
343 	int ostate;
344 
345 	ostate = tp->t_state;
346 #endif
347 	INP_INFO_WLOCK(&V_tcbinfo);
348 	inp = tp->t_inpcb;
349 	/*
350 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
351 	 * tear-down mean we need it as a work-around for races between
352 	 * timers and tcp_discardcb().
353 	 *
354 	 * KASSERT(inp != NULL, ("tcp_timer_keep: inp == NULL"));
355 	 */
356 	if (inp == NULL) {
357 		tcp_timer_race++;
358 		INP_INFO_WUNLOCK(&V_tcbinfo);
359 		CURVNET_RESTORE();
360 		return;
361 	}
362 	INP_WLOCK(inp);
363 	if (callout_pending(&tp->t_timers->tt_keep) ||
364 	    !callout_active(&tp->t_timers->tt_keep)) {
365 		INP_WUNLOCK(inp);
366 		INP_INFO_WUNLOCK(&V_tcbinfo);
367 		CURVNET_RESTORE();
368 		return;
369 	}
370 	callout_deactivate(&tp->t_timers->tt_keep);
371 	if ((inp->inp_flags & INP_DROPPED) != 0) {
372 		INP_WUNLOCK(inp);
373 		INP_INFO_WUNLOCK(&V_tcbinfo);
374 		CURVNET_RESTORE();
375 		return;
376 	}
377 	/*
378 	 * Keep-alive timer went off; send something
379 	 * or drop connection if idle for too long.
380 	 */
381 	TCPSTAT_INC(tcps_keeptimeo);
382 	if (tp->t_state < TCPS_ESTABLISHED)
383 		goto dropit;
384 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
385 	    tp->t_state <= TCPS_CLOSING) {
386 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
387 			goto dropit;
388 		/*
389 		 * Send a packet designed to force a response
390 		 * if the peer is up and reachable:
391 		 * either an ACK if the connection is still alive,
392 		 * or an RST if the peer has closed the connection
393 		 * due to timeout or reboot.
394 		 * Using sequence number tp->snd_una-1
395 		 * causes the transmitted zero-length segment
396 		 * to lie outside the receive window;
397 		 * by the protocol spec, this requires the
398 		 * correspondent TCP to respond.
399 		 */
400 		TCPSTAT_INC(tcps_keepprobe);
401 		t_template = tcpip_maketemplate(inp);
402 		if (t_template) {
403 			tcp_respond(tp, t_template->tt_ipgen,
404 				    &t_template->tt_t, (struct mbuf *)NULL,
405 				    tp->rcv_nxt, tp->snd_una - 1, 0);
406 			free(t_template, M_TEMP);
407 		}
408 		callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
409 		    tcp_timer_keep, tp, inp_to_cpuid(inp));
410 	} else
411 		callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
412 		    tcp_timer_keep, tp, inp_to_cpuid(inp));
413 
414 #ifdef TCPDEBUG
415 	if (inp->inp_socket->so_options & SO_DEBUG)
416 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
417 			  PRU_SLOWTIMO);
418 #endif
419 	INP_WUNLOCK(inp);
420 	INP_INFO_WUNLOCK(&V_tcbinfo);
421 	CURVNET_RESTORE();
422 	return;
423 
424 dropit:
425 	TCPSTAT_INC(tcps_keepdrops);
426 	tp = tcp_drop(tp, ETIMEDOUT);
427 
428 #ifdef TCPDEBUG
429 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
430 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
431 			  PRU_SLOWTIMO);
432 #endif
433 	if (tp != NULL)
434 		INP_WUNLOCK(tp->t_inpcb);
435 	INP_INFO_WUNLOCK(&V_tcbinfo);
436 	CURVNET_RESTORE();
437 }
438 
439 void
440 tcp_timer_persist(void *xtp)
441 {
442 	struct tcpcb *tp = xtp;
443 	struct inpcb *inp;
444 	CURVNET_SET(tp->t_vnet);
445 #ifdef TCPDEBUG
446 	int ostate;
447 
448 	ostate = tp->t_state;
449 #endif
450 	INP_INFO_WLOCK(&V_tcbinfo);
451 	inp = tp->t_inpcb;
452 	/*
453 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
454 	 * tear-down mean we need it as a work-around for races between
455 	 * timers and tcp_discardcb().
456 	 *
457 	 * KASSERT(inp != NULL, ("tcp_timer_persist: inp == NULL"));
458 	 */
459 	if (inp == NULL) {
460 		tcp_timer_race++;
461 		INP_INFO_WUNLOCK(&V_tcbinfo);
462 		CURVNET_RESTORE();
463 		return;
464 	}
465 	INP_WLOCK(inp);
466 	if (callout_pending(&tp->t_timers->tt_persist) ||
467 	    !callout_active(&tp->t_timers->tt_persist)) {
468 		INP_WUNLOCK(inp);
469 		INP_INFO_WUNLOCK(&V_tcbinfo);
470 		CURVNET_RESTORE();
471 		return;
472 	}
473 	callout_deactivate(&tp->t_timers->tt_persist);
474 	if ((inp->inp_flags & INP_DROPPED) != 0) {
475 		INP_WUNLOCK(inp);
476 		INP_INFO_WUNLOCK(&V_tcbinfo);
477 		CURVNET_RESTORE();
478 		return;
479 	}
480 	/*
481 	 * Persistance timer into zero window.
482 	 * Force a byte to be output, if possible.
483 	 */
484 	TCPSTAT_INC(tcps_persisttimeo);
485 	/*
486 	 * Hack: if the peer is dead/unreachable, we do not
487 	 * time out if the window is closed.  After a full
488 	 * backoff, drop the connection if the idle time
489 	 * (no responses to probes) reaches the maximum
490 	 * backoff that we would use if retransmitting.
491 	 */
492 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
493 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
494 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
495 		TCPSTAT_INC(tcps_persistdrop);
496 		tp = tcp_drop(tp, ETIMEDOUT);
497 		goto out;
498 	}
499 	/*
500 	 * If the user has closed the socket then drop a persisting
501 	 * connection after a much reduced timeout.
502 	 */
503 	if (tp->t_state > TCPS_CLOSE_WAIT &&
504 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
505 		TCPSTAT_INC(tcps_persistdrop);
506 		tp = tcp_drop(tp, ETIMEDOUT);
507 		goto out;
508 	}
509 	tcp_setpersist(tp);
510 	tp->t_flags |= TF_FORCEDATA;
511 	(void) tcp_output(tp);
512 	tp->t_flags &= ~TF_FORCEDATA;
513 
514 out:
515 #ifdef TCPDEBUG
516 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
517 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
518 #endif
519 	if (tp != NULL)
520 		INP_WUNLOCK(inp);
521 	INP_INFO_WUNLOCK(&V_tcbinfo);
522 	CURVNET_RESTORE();
523 }
524 
525 void
526 tcp_timer_rexmt(void * xtp)
527 {
528 	struct tcpcb *tp = xtp;
529 	CURVNET_SET(tp->t_vnet);
530 	int rexmt;
531 	int headlocked;
532 	struct inpcb *inp;
533 #ifdef TCPDEBUG
534 	int ostate;
535 
536 	ostate = tp->t_state;
537 #endif
538 	INP_INFO_RLOCK(&V_tcbinfo);
539 	inp = tp->t_inpcb;
540 	/*
541 	 * XXXRW: While this assert is in fact correct, bugs in the tcpcb
542 	 * tear-down mean we need it as a work-around for races between
543 	 * timers and tcp_discardcb().
544 	 *
545 	 * KASSERT(inp != NULL, ("tcp_timer_rexmt: inp == NULL"));
546 	 */
547 	if (inp == NULL) {
548 		tcp_timer_race++;
549 		INP_INFO_RUNLOCK(&V_tcbinfo);
550 		CURVNET_RESTORE();
551 		return;
552 	}
553 	INP_WLOCK(inp);
554 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
555 	    !callout_active(&tp->t_timers->tt_rexmt)) {
556 		INP_WUNLOCK(inp);
557 		INP_INFO_RUNLOCK(&V_tcbinfo);
558 		CURVNET_RESTORE();
559 		return;
560 	}
561 	callout_deactivate(&tp->t_timers->tt_rexmt);
562 	if ((inp->inp_flags & INP_DROPPED) != 0) {
563 		INP_WUNLOCK(inp);
564 		INP_INFO_RUNLOCK(&V_tcbinfo);
565 		CURVNET_RESTORE();
566 		return;
567 	}
568 	tcp_free_sackholes(tp);
569 	/*
570 	 * Retransmission timer went off.  Message has not
571 	 * been acked within retransmit interval.  Back off
572 	 * to a longer retransmit interval and retransmit one segment.
573 	 */
574 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
575 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
576 		TCPSTAT_INC(tcps_timeoutdrop);
577 		in_pcbref(inp);
578 		INP_INFO_RUNLOCK(&V_tcbinfo);
579 		INP_WUNLOCK(inp);
580 		INP_INFO_WLOCK(&V_tcbinfo);
581 		INP_WLOCK(inp);
582 		if (in_pcbrele_wlocked(inp)) {
583 			INP_INFO_WUNLOCK(&V_tcbinfo);
584 			CURVNET_RESTORE();
585 			return;
586 		}
587 		if (inp->inp_flags & INP_DROPPED) {
588 			INP_WUNLOCK(inp);
589 			INP_INFO_WUNLOCK(&V_tcbinfo);
590 			CURVNET_RESTORE();
591 			return;
592 		}
593 
594 		tp = tcp_drop(tp, tp->t_softerror ?
595 			      tp->t_softerror : ETIMEDOUT);
596 		headlocked = 1;
597 		goto out;
598 	}
599 	INP_INFO_RUNLOCK(&V_tcbinfo);
600 	headlocked = 0;
601 	if (tp->t_state == TCPS_SYN_SENT) {
602 		/*
603 		 * If the SYN was retransmitted, indicate CWND to be
604 		 * limited to 1 segment in cc_conn_init().
605 		 */
606 		tp->snd_cwnd = 1;
607 	} else if (tp->t_rxtshift == 1) {
608 		/*
609 		 * first retransmit; record ssthresh and cwnd so they can
610 		 * be recovered if this turns out to be a "bad" retransmit.
611 		 * A retransmit is considered "bad" if an ACK for this
612 		 * segment is received within RTT/2 interval; the assumption
613 		 * here is that the ACK was already in flight.  See
614 		 * "On Estimating End-to-End Network Path Properties" by
615 		 * Allman and Paxson for more details.
616 		 */
617 		tp->snd_cwnd_prev = tp->snd_cwnd;
618 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
619 		tp->snd_recover_prev = tp->snd_recover;
620 		if (IN_FASTRECOVERY(tp->t_flags))
621 			tp->t_flags |= TF_WASFRECOVERY;
622 		else
623 			tp->t_flags &= ~TF_WASFRECOVERY;
624 		if (IN_CONGRECOVERY(tp->t_flags))
625 			tp->t_flags |= TF_WASCRECOVERY;
626 		else
627 			tp->t_flags &= ~TF_WASCRECOVERY;
628 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
629 		tp->t_flags |= TF_PREVVALID;
630 	} else
631 		tp->t_flags &= ~TF_PREVVALID;
632 	TCPSTAT_INC(tcps_rexmttimeo);
633 	if (tp->t_state == TCPS_SYN_SENT)
634 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
635 	else
636 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
637 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
638 		      tp->t_rttmin, TCPTV_REXMTMAX);
639 	/*
640 	 * Disable RFC1323 and SACK if we haven't got any response to
641 	 * our third SYN to work-around some broken terminal servers
642 	 * (most of which have hopefully been retired) that have bad VJ
643 	 * header compression code which trashes TCP segments containing
644 	 * unknown-to-them TCP options.
645 	 */
646 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
647 	    (tp->t_rxtshift == 3))
648 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
649 	/*
650 	 * If we backed off this far, our srtt estimate is probably bogus.
651 	 * Clobber it so we'll take the next rtt measurement as our srtt;
652 	 * move the current srtt into rttvar to keep the current
653 	 * retransmit times until then.
654 	 */
655 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
656 #ifdef INET6
657 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
658 			in6_losing(tp->t_inpcb);
659 #endif
660 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
661 		tp->t_srtt = 0;
662 	}
663 	tp->snd_nxt = tp->snd_una;
664 	tp->snd_recover = tp->snd_max;
665 	/*
666 	 * Force a segment to be sent.
667 	 */
668 	tp->t_flags |= TF_ACKNOW;
669 	/*
670 	 * If timing a segment in this window, stop the timer.
671 	 */
672 	tp->t_rtttime = 0;
673 
674 	cc_cong_signal(tp, NULL, CC_RTO);
675 
676 	(void) tcp_output(tp);
677 
678 out:
679 #ifdef TCPDEBUG
680 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
681 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
682 			  PRU_SLOWTIMO);
683 #endif
684 	if (tp != NULL)
685 		INP_WUNLOCK(inp);
686 	if (headlocked)
687 		INP_INFO_WUNLOCK(&V_tcbinfo);
688 	CURVNET_RESTORE();
689 }
690 
691 void
692 tcp_timer_activate(struct tcpcb *tp, int timer_type, u_int delta)
693 {
694 	struct callout *t_callout;
695 	void *f_callout;
696 	struct inpcb *inp = tp->t_inpcb;
697 	int cpu = inp_to_cpuid(inp);
698 
699 #ifdef TCP_OFFLOAD
700 	if (tp->t_flags & TF_TOE)
701 		return;
702 #endif
703 
704 	switch (timer_type) {
705 		case TT_DELACK:
706 			t_callout = &tp->t_timers->tt_delack;
707 			f_callout = tcp_timer_delack;
708 			break;
709 		case TT_REXMT:
710 			t_callout = &tp->t_timers->tt_rexmt;
711 			f_callout = tcp_timer_rexmt;
712 			break;
713 		case TT_PERSIST:
714 			t_callout = &tp->t_timers->tt_persist;
715 			f_callout = tcp_timer_persist;
716 			break;
717 		case TT_KEEP:
718 			t_callout = &tp->t_timers->tt_keep;
719 			f_callout = tcp_timer_keep;
720 			break;
721 		case TT_2MSL:
722 			t_callout = &tp->t_timers->tt_2msl;
723 			f_callout = tcp_timer_2msl;
724 			break;
725 		default:
726 			panic("bad timer_type");
727 		}
728 	if (delta == 0) {
729 		callout_stop(t_callout);
730 	} else {
731 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
732 	}
733 }
734 
735 int
736 tcp_timer_active(struct tcpcb *tp, int timer_type)
737 {
738 	struct callout *t_callout;
739 
740 	switch (timer_type) {
741 		case TT_DELACK:
742 			t_callout = &tp->t_timers->tt_delack;
743 			break;
744 		case TT_REXMT:
745 			t_callout = &tp->t_timers->tt_rexmt;
746 			break;
747 		case TT_PERSIST:
748 			t_callout = &tp->t_timers->tt_persist;
749 			break;
750 		case TT_KEEP:
751 			t_callout = &tp->t_timers->tt_keep;
752 			break;
753 		case TT_2MSL:
754 			t_callout = &tp->t_timers->tt_2msl;
755 			break;
756 		default:
757 			panic("bad timer_type");
758 		}
759 	return callout_active(t_callout);
760 }
761 
762 #define	ticks_to_msecs(t)	(1000*(t) / hz)
763 
764 void
765 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
766     struct xtcp_timer *xtimer)
767 {
768 	sbintime_t now;
769 
770 	bzero(xtimer, sizeof(*xtimer));
771 	if (timer == NULL)
772 		return;
773 	now = getsbinuptime();
774 	if (callout_active(&timer->tt_delack))
775 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
776 	if (callout_active(&timer->tt_rexmt))
777 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
778 	if (callout_active(&timer->tt_persist))
779 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
780 	if (callout_active(&timer->tt_keep))
781 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
782 	if (callout_active(&timer->tt_2msl))
783 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
784 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
785 }
786