xref: /freebsd/sys/netinet/tcp_timer.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
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 int	tcp_always_keepalive = 1;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
130     &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
131 
132 int    tcp_fast_finwait2_recycle = 0;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
134     &tcp_fast_finwait2_recycle, 0,
135     "Recycle closed FIN_WAIT_2 connections faster");
136 
137 int    tcp_finwait2_timeout;
138 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
139     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
140 
141 int	tcp_keepcnt = TCPTV_KEEPCNT;
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
143     "Number of keepalive probes to send");
144 
145 	/* max idle probes */
146 int	tcp_maxpersistidle;
147 
148 int	tcp_rexmit_drop_options = 0;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
150     &tcp_rexmit_drop_options, 0,
151     "Drop TCP options from 3rd and later retransmitted SYN");
152 
153 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
155     CTLFLAG_RW|CTLFLAG_VNET,
156     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
157     "Path MTU Discovery Black Hole Detection Enabled");
158 
159 #ifdef INET
160 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
162     CTLFLAG_RW|CTLFLAG_VNET,
163     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
164     "Path MTU Discovery Black Hole Detection lowered MSS");
165 #endif
166 
167 #ifdef INET6
168 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
170     CTLFLAG_RW|CTLFLAG_VNET,
171     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
172     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
173 #endif
174 
175 #ifdef	RSS
176 static int	per_cpu_timers = 1;
177 #else
178 static int	per_cpu_timers = 0;
179 #endif
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
181     &per_cpu_timers , 0, "run tcp timers on all cpus");
182 
183 /*
184  * Map the given inp to a CPU id.
185  *
186  * This queries RSS if it's compiled in, else it defaults to the current
187  * CPU ID.
188  */
189 inline int
190 inp_to_cpuid(struct inpcb *inp)
191 {
192 	u_int cpuid;
193 
194 #ifdef	RSS
195 	if (per_cpu_timers) {
196 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
197 		if (cpuid == NETISR_CPUID_NONE)
198 			return (curcpu);	/* XXX */
199 		else
200 			return (cpuid);
201 	}
202 #else
203 	/* Legacy, pre-RSS behaviour */
204 	if (per_cpu_timers) {
205 		/*
206 		 * We don't have a flowid -> cpuid mapping, so cheat and
207 		 * just map unknown cpuids to curcpu.  Not the best, but
208 		 * apparently better than defaulting to swi 0.
209 		 */
210 		cpuid = inp->inp_flowid % (mp_maxid + 1);
211 		if (! CPU_ABSENT(cpuid))
212 			return (cpuid);
213 		return (curcpu);
214 	}
215 #endif
216 	/* Default for RSS and non-RSS - cpuid 0 */
217 	else {
218 		return (0);
219 	}
220 }
221 
222 /*
223  * Tcp protocol timeout routine called every 500 ms.
224  * Updates timestamps used for TCP
225  * causes finite state machine actions if timers expire.
226  */
227 void
228 tcp_slowtimo(void)
229 {
230 	VNET_ITERATOR_DECL(vnet_iter);
231 
232 	VNET_LIST_RLOCK_NOSLEEP();
233 	VNET_FOREACH(vnet_iter) {
234 		CURVNET_SET(vnet_iter);
235 		(void) tcp_tw_2msl_scan(0);
236 		CURVNET_RESTORE();
237 	}
238 	VNET_LIST_RUNLOCK_NOSLEEP();
239 }
240 
241 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
242     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
243 
244 int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
245 
246 /*
247  * TCP timer processing.
248  */
249 
250 void
251 tcp_timer_delack(void *xtp)
252 {
253 	struct tcpcb *tp = xtp;
254 	struct inpcb *inp;
255 	CURVNET_SET(tp->t_vnet);
256 
257 	inp = tp->t_inpcb;
258 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
259 	INP_WLOCK(inp);
260 	if (callout_pending(&tp->t_timers->tt_delack) ||
261 	    !callout_active(&tp->t_timers->tt_delack)) {
262 		INP_WUNLOCK(inp);
263 		CURVNET_RESTORE();
264 		return;
265 	}
266 	callout_deactivate(&tp->t_timers->tt_delack);
267 	if ((inp->inp_flags & INP_DROPPED) != 0) {
268 		INP_WUNLOCK(inp);
269 		CURVNET_RESTORE();
270 		return;
271 	}
272 	tp->t_flags |= TF_ACKNOW;
273 	TCPSTAT_INC(tcps_delack);
274 	(void) tp->t_fb->tfb_tcp_output(tp);
275 	INP_WUNLOCK(inp);
276 	CURVNET_RESTORE();
277 }
278 
279 void
280 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
281 {
282 	if (inp && tp != NULL)
283 		INP_WUNLOCK(inp);
284 }
285 
286 void
287 tcp_timer_2msl(void *xtp)
288 {
289 	struct tcpcb *tp = xtp;
290 	struct inpcb *inp;
291 	struct epoch_tracker et;
292 	CURVNET_SET(tp->t_vnet);
293 #ifdef TCPDEBUG
294 	int ostate;
295 
296 	ostate = tp->t_state;
297 #endif
298 	inp = tp->t_inpcb;
299 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
300 	INP_WLOCK(inp);
301 	tcp_free_sackholes(tp);
302 	if (callout_pending(&tp->t_timers->tt_2msl) ||
303 	    !callout_active(&tp->t_timers->tt_2msl)) {
304 		INP_WUNLOCK(tp->t_inpcb);
305 		CURVNET_RESTORE();
306 		return;
307 	}
308 	callout_deactivate(&tp->t_timers->tt_2msl);
309 	if ((inp->inp_flags & INP_DROPPED) != 0) {
310 		INP_WUNLOCK(inp);
311 		CURVNET_RESTORE();
312 		return;
313 	}
314 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
315 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
316 	/*
317 	 * 2 MSL timeout in shutdown went off.  If we're closed but
318 	 * still waiting for peer to close and connection has been idle
319 	 * too long delete connection control block.  Otherwise, check
320 	 * again in a bit.
321 	 *
322 	 * If in TIME_WAIT state just ignore as this timeout is handled in
323 	 * tcp_tw_2msl_scan().
324 	 *
325 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
326 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
327 	 * Ignore fact that there were recent incoming segments.
328 	 */
329 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
330 		INP_WUNLOCK(inp);
331 		CURVNET_RESTORE();
332 		return;
333 	}
334 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
335 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
336 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
337 		TCPSTAT_INC(tcps_finwait2_drops);
338 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
339 			tcp_inpinfo_lock_del(inp, tp);
340 			goto out;
341 		}
342 		INP_INFO_RLOCK_ET(&V_tcbinfo, et);
343 		tp = tcp_close(tp);
344 		INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
345 		tcp_inpinfo_lock_del(inp, tp);
346 		goto out;
347 	} else {
348 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
349 			callout_reset(&tp->t_timers->tt_2msl,
350 				      TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
351 		} else {
352 			if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
353 				tcp_inpinfo_lock_del(inp, tp);
354 				goto out;
355 			}
356 			INP_INFO_RLOCK_ET(&V_tcbinfo, et);
357 			tp = tcp_close(tp);
358 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
359 			tcp_inpinfo_lock_del(inp, tp);
360 			goto out;
361 		}
362 	}
363 
364 #ifdef TCPDEBUG
365 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
366 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
367 			  PRU_SLOWTIMO);
368 #endif
369 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
370 
371 	if (tp != NULL)
372 		INP_WUNLOCK(inp);
373 out:
374 	CURVNET_RESTORE();
375 }
376 
377 void
378 tcp_timer_keep(void *xtp)
379 {
380 	struct tcpcb *tp = xtp;
381 	struct tcptemp *t_template;
382 	struct inpcb *inp;
383 	struct epoch_tracker et;
384 	CURVNET_SET(tp->t_vnet);
385 #ifdef TCPDEBUG
386 	int ostate;
387 
388 	ostate = tp->t_state;
389 #endif
390 	inp = tp->t_inpcb;
391 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
392 	INP_WLOCK(inp);
393 	if (callout_pending(&tp->t_timers->tt_keep) ||
394 	    !callout_active(&tp->t_timers->tt_keep)) {
395 		INP_WUNLOCK(inp);
396 		CURVNET_RESTORE();
397 		return;
398 	}
399 	callout_deactivate(&tp->t_timers->tt_keep);
400 	if ((inp->inp_flags & INP_DROPPED) != 0) {
401 		INP_WUNLOCK(inp);
402 		CURVNET_RESTORE();
403 		return;
404 	}
405 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
406 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
407 
408 	/*
409 	 * Because we don't regularly reset the keepalive callout in
410 	 * the ESTABLISHED state, it may be that we don't actually need
411 	 * to send a keepalive yet. If that occurs, schedule another
412 	 * call for the next time the keepalive timer might expire.
413 	 */
414 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
415 		u_int idletime;
416 
417 		idletime = ticks - tp->t_rcvtime;
418 		if (idletime < TP_KEEPIDLE(tp)) {
419 			callout_reset(&tp->t_timers->tt_keep,
420 			    TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
421 			INP_WUNLOCK(inp);
422 			CURVNET_RESTORE();
423 			return;
424 		}
425 	}
426 
427 	/*
428 	 * Keep-alive timer went off; send something
429 	 * or drop connection if idle for too long.
430 	 */
431 	TCPSTAT_INC(tcps_keeptimeo);
432 	if (tp->t_state < TCPS_ESTABLISHED)
433 		goto dropit;
434 	if ((tcp_always_keepalive ||
435 	    inp->inp_socket->so_options & SO_KEEPALIVE) &&
436 	    tp->t_state <= TCPS_CLOSING) {
437 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
438 			goto dropit;
439 		/*
440 		 * Send a packet designed to force a response
441 		 * if the peer is up and reachable:
442 		 * either an ACK if the connection is still alive,
443 		 * or an RST if the peer has closed the connection
444 		 * due to timeout or reboot.
445 		 * Using sequence number tp->snd_una-1
446 		 * causes the transmitted zero-length segment
447 		 * to lie outside the receive window;
448 		 * by the protocol spec, this requires the
449 		 * correspondent TCP to respond.
450 		 */
451 		TCPSTAT_INC(tcps_keepprobe);
452 		t_template = tcpip_maketemplate(inp);
453 		if (t_template) {
454 			tcp_respond(tp, t_template->tt_ipgen,
455 				    &t_template->tt_t, (struct mbuf *)NULL,
456 				    tp->rcv_nxt, tp->snd_una - 1, 0);
457 			free(t_template, M_TEMP);
458 		}
459 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
460 			      tcp_timer_keep, tp);
461 	} else
462 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
463 			      tcp_timer_keep, tp);
464 
465 #ifdef TCPDEBUG
466 	if (inp->inp_socket->so_options & SO_DEBUG)
467 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
468 			  PRU_SLOWTIMO);
469 #endif
470 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
471 	INP_WUNLOCK(inp);
472 	CURVNET_RESTORE();
473 	return;
474 
475 dropit:
476 	TCPSTAT_INC(tcps_keepdrops);
477 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
478 		tcp_inpinfo_lock_del(inp, tp);
479 		goto out;
480 	}
481 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
482 	tp = tcp_drop(tp, ETIMEDOUT);
483 
484 #ifdef TCPDEBUG
485 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
486 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
487 			  PRU_SLOWTIMO);
488 #endif
489 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
490 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
491 	tcp_inpinfo_lock_del(inp, tp);
492  out:
493 	CURVNET_RESTORE();
494 }
495 
496 void
497 tcp_timer_persist(void *xtp)
498 {
499 	struct tcpcb *tp = xtp;
500 	struct inpcb *inp;
501 	struct epoch_tracker et;
502 	CURVNET_SET(tp->t_vnet);
503 #ifdef TCPDEBUG
504 	int ostate;
505 
506 	ostate = tp->t_state;
507 #endif
508 	inp = tp->t_inpcb;
509 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
510 	INP_WLOCK(inp);
511 	if (callout_pending(&tp->t_timers->tt_persist) ||
512 	    !callout_active(&tp->t_timers->tt_persist)) {
513 		INP_WUNLOCK(inp);
514 		CURVNET_RESTORE();
515 		return;
516 	}
517 	callout_deactivate(&tp->t_timers->tt_persist);
518 	if ((inp->inp_flags & INP_DROPPED) != 0) {
519 		INP_WUNLOCK(inp);
520 		CURVNET_RESTORE();
521 		return;
522 	}
523 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
524 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
525 	/*
526 	 * Persistence timer into zero window.
527 	 * Force a byte to be output, if possible.
528 	 */
529 	TCPSTAT_INC(tcps_persisttimeo);
530 	/*
531 	 * Hack: if the peer is dead/unreachable, we do not
532 	 * time out if the window is closed.  After a full
533 	 * backoff, drop the connection if the idle time
534 	 * (no responses to probes) reaches the maximum
535 	 * backoff that we would use if retransmitting.
536 	 */
537 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
538 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
539 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
540 		TCPSTAT_INC(tcps_persistdrop);
541 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
542 			tcp_inpinfo_lock_del(inp, tp);
543 			goto out;
544 		}
545 		INP_INFO_RLOCK_ET(&V_tcbinfo, et);
546 		tp = tcp_drop(tp, ETIMEDOUT);
547 		INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
548 		tcp_inpinfo_lock_del(inp, tp);
549 		goto out;
550 	}
551 	/*
552 	 * If the user has closed the socket then drop a persisting
553 	 * connection after a much reduced timeout.
554 	 */
555 	if (tp->t_state > TCPS_CLOSE_WAIT &&
556 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
557 		TCPSTAT_INC(tcps_persistdrop);
558 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
559 			tcp_inpinfo_lock_del(inp, tp);
560 			goto out;
561 		}
562 		INP_INFO_RLOCK_ET(&V_tcbinfo, et);
563 		tp = tcp_drop(tp, ETIMEDOUT);
564 		INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
565 		tcp_inpinfo_lock_del(inp, tp);
566 		goto out;
567 	}
568 	tcp_setpersist(tp);
569 	tp->t_flags |= TF_FORCEDATA;
570 	(void) tp->t_fb->tfb_tcp_output(tp);
571 	tp->t_flags &= ~TF_FORCEDATA;
572 
573 #ifdef TCPDEBUG
574 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
575 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
576 #endif
577 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
578 	INP_WUNLOCK(inp);
579 out:
580 	CURVNET_RESTORE();
581 }
582 
583 void
584 tcp_timer_rexmt(void * xtp)
585 {
586 	struct tcpcb *tp = xtp;
587 	CURVNET_SET(tp->t_vnet);
588 	int rexmt;
589 	struct inpcb *inp;
590 	struct epoch_tracker et;
591 #ifdef TCPDEBUG
592 	int ostate;
593 
594 	ostate = tp->t_state;
595 #endif
596 	inp = tp->t_inpcb;
597 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
598 	INP_WLOCK(inp);
599 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
600 	    !callout_active(&tp->t_timers->tt_rexmt)) {
601 		INP_WUNLOCK(inp);
602 		CURVNET_RESTORE();
603 		return;
604 	}
605 	callout_deactivate(&tp->t_timers->tt_rexmt);
606 	if ((inp->inp_flags & INP_DROPPED) != 0) {
607 		INP_WUNLOCK(inp);
608 		CURVNET_RESTORE();
609 		return;
610 	}
611 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
612 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
613 	tcp_free_sackholes(tp);
614 	TCP_LOG_EVENT(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0, NULL, false);
615 	if (tp->t_fb->tfb_tcp_rexmit_tmr) {
616 		/* The stack has a timer action too. */
617 		(*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
618 	}
619 	/*
620 	 * Retransmission timer went off.  Message has not
621 	 * been acked within retransmit interval.  Back off
622 	 * to a longer retransmit interval and retransmit one segment.
623 	 */
624 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
625 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
626 		TCPSTAT_INC(tcps_timeoutdrop);
627 		if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
628 			tcp_inpinfo_lock_del(inp, tp);
629 			goto out;
630 		}
631 		INP_INFO_RLOCK_ET(&V_tcbinfo, et);
632 		tp = tcp_drop(tp, ETIMEDOUT);
633 		INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
634 		tcp_inpinfo_lock_del(inp, tp);
635 		goto out;
636 	}
637 	if (tp->t_state == TCPS_SYN_SENT) {
638 		/*
639 		 * If the SYN was retransmitted, indicate CWND to be
640 		 * limited to 1 segment in cc_conn_init().
641 		 */
642 		tp->snd_cwnd = 1;
643 	} else if (tp->t_rxtshift == 1) {
644 		/*
645 		 * first retransmit; record ssthresh and cwnd so they can
646 		 * be recovered if this turns out to be a "bad" retransmit.
647 		 * A retransmit is considered "bad" if an ACK for this
648 		 * segment is received within RTT/2 interval; the assumption
649 		 * here is that the ACK was already in flight.  See
650 		 * "On Estimating End-to-End Network Path Properties" by
651 		 * Allman and Paxson for more details.
652 		 */
653 		tp->snd_cwnd_prev = tp->snd_cwnd;
654 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
655 		tp->snd_recover_prev = tp->snd_recover;
656 		if (IN_FASTRECOVERY(tp->t_flags))
657 			tp->t_flags |= TF_WASFRECOVERY;
658 		else
659 			tp->t_flags &= ~TF_WASFRECOVERY;
660 		if (IN_CONGRECOVERY(tp->t_flags))
661 			tp->t_flags |= TF_WASCRECOVERY;
662 		else
663 			tp->t_flags &= ~TF_WASCRECOVERY;
664 		if ((tp->t_flags & TF_RCVD_TSTMP) == 0)
665 			tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
666 		/* In the event that we've negotiated timestamps
667 		 * badrxtwin will be set to the value that we set
668 		 * the retransmitted packet's to_tsval to by tcp_output
669 		 */
670 		tp->t_flags |= TF_PREVVALID;
671 	} else
672 		tp->t_flags &= ~TF_PREVVALID;
673 	TCPSTAT_INC(tcps_rexmttimeo);
674 	if ((tp->t_state == TCPS_SYN_SENT) ||
675 	    (tp->t_state == TCPS_SYN_RECEIVED))
676 		rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
677 	else
678 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
679 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
680 		      tp->t_rttmin, TCPTV_REXMTMAX);
681 
682 	/*
683 	 * We enter the path for PLMTUD if connection is established or, if
684 	 * connection is FIN_WAIT_1 status, reason for the last is that if
685 	 * amount of data we send is very small, we could send it in couple of
686 	 * packets and process straight to FIN. In that case we won't catch
687 	 * ESTABLISHED state.
688 	 */
689 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
690 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
691 #ifdef INET6
692 		int isipv6;
693 #endif
694 
695 		/*
696 		 * Idea here is that at each stage of mtu probe (usually, 1448
697 		 * -> 1188 -> 524) should be given 2 chances to recover before
698 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
699 		 *  take care of that.
700 		 */
701 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
702 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
703 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
704 		    tp->t_rxtshift % 2 == 0)) {
705 			/*
706 			 * Enter Path MTU Black-hole Detection mechanism:
707 			 * - Disable Path MTU Discovery (IP "DF" bit).
708 			 * - Reduce MTU to lower value than what we
709 			 *   negotiated with peer.
710 			 */
711 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
712 				/* Record that we may have found a black hole. */
713 				tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
714 				/* Keep track of previous MSS. */
715 				tp->t_pmtud_saved_maxseg = tp->t_maxseg;
716 			}
717 
718 			/*
719 			 * Reduce the MSS to blackhole value or to the default
720 			 * in an attempt to retransmit.
721 			 */
722 #ifdef INET6
723 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
724 			if (isipv6 &&
725 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
726 				/* Use the sysctl tuneable blackhole MSS. */
727 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
728 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
729 			} else if (isipv6) {
730 				/* Use the default MSS. */
731 				tp->t_maxseg = V_tcp_v6mssdflt;
732 				/*
733 				 * Disable Path MTU Discovery when we switch to
734 				 * minmss.
735 				 */
736 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
737 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
738 			}
739 #endif
740 #if defined(INET6) && defined(INET)
741 			else
742 #endif
743 #ifdef INET
744 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
745 				/* Use the sysctl tuneable blackhole MSS. */
746 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
747 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
748 			} else {
749 				/* Use the default MSS. */
750 				tp->t_maxseg = V_tcp_mssdflt;
751 				/*
752 				 * Disable Path MTU Discovery when we switch to
753 				 * minmss.
754 				 */
755 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
756 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
757 			}
758 #endif
759 			/*
760 			 * Reset the slow-start flight size
761 			 * as it may depend on the new MSS.
762 			 */
763 			if (CC_ALGO(tp)->conn_init != NULL)
764 				CC_ALGO(tp)->conn_init(tp->ccv);
765 		} else {
766 			/*
767 			 * If further retransmissions are still unsuccessful
768 			 * with a lowered MTU, maybe this isn't a blackhole and
769 			 * we restore the previous MSS and blackhole detection
770 			 * flags.
771 			 * The limit '6' is determined by giving each probe
772 			 * stage (1448, 1188, 524) 2 chances to recover.
773 			 */
774 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
775 			    (tp->t_rxtshift >= 6)) {
776 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
777 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
778 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
779 				TCPSTAT_INC(tcps_pmtud_blackhole_failed);
780 				/*
781 				 * Reset the slow-start flight size as it
782 				 * may depend on the new MSS.
783 				 */
784 				if (CC_ALGO(tp)->conn_init != NULL)
785 					CC_ALGO(tp)->conn_init(tp->ccv);
786 			}
787 		}
788 	}
789 
790 	/*
791 	 * Disable RFC1323 and SACK if we haven't got any response to
792 	 * our third SYN to work-around some broken terminal servers
793 	 * (most of which have hopefully been retired) that have bad VJ
794 	 * header compression code which trashes TCP segments containing
795 	 * unknown-to-them TCP options.
796 	 */
797 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
798 	    (tp->t_rxtshift == 3))
799 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
800 	/*
801 	 * If we backed off this far, notify the L3 protocol that we're having
802 	 * connection problems.
803 	 */
804 	if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
805 #ifdef INET6
806 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
807 			in6_losing(tp->t_inpcb);
808 		else
809 #endif
810 			in_losing(tp->t_inpcb);
811 	}
812 	tp->snd_nxt = tp->snd_una;
813 	tp->snd_recover = tp->snd_max;
814 	/*
815 	 * Force a segment to be sent.
816 	 */
817 	tp->t_flags |= TF_ACKNOW;
818 	/*
819 	 * If timing a segment in this window, stop the timer.
820 	 */
821 	tp->t_rtttime = 0;
822 
823 	cc_cong_signal(tp, NULL, CC_RTO);
824 
825 	(void) tp->t_fb->tfb_tcp_output(tp);
826 
827 #ifdef TCPDEBUG
828 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
829 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
830 			  PRU_SLOWTIMO);
831 #endif
832 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
833 	INP_WUNLOCK(inp);
834 out:
835 	CURVNET_RESTORE();
836 }
837 
838 void
839 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
840 {
841 	struct callout *t_callout;
842 	timeout_t *f_callout;
843 	struct inpcb *inp = tp->t_inpcb;
844 	int cpu = inp_to_cpuid(inp);
845 
846 #ifdef TCP_OFFLOAD
847 	if (tp->t_flags & TF_TOE)
848 		return;
849 #endif
850 
851 	if (tp->t_timers->tt_flags & TT_STOPPED)
852 		return;
853 
854 	switch (timer_type) {
855 		case TT_DELACK:
856 			t_callout = &tp->t_timers->tt_delack;
857 			f_callout = tcp_timer_delack;
858 			break;
859 		case TT_REXMT:
860 			t_callout = &tp->t_timers->tt_rexmt;
861 			f_callout = tcp_timer_rexmt;
862 			break;
863 		case TT_PERSIST:
864 			t_callout = &tp->t_timers->tt_persist;
865 			f_callout = tcp_timer_persist;
866 			break;
867 		case TT_KEEP:
868 			t_callout = &tp->t_timers->tt_keep;
869 			f_callout = tcp_timer_keep;
870 			break;
871 		case TT_2MSL:
872 			t_callout = &tp->t_timers->tt_2msl;
873 			f_callout = tcp_timer_2msl;
874 			break;
875 		default:
876 			if (tp->t_fb->tfb_tcp_timer_activate) {
877 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
878 				return;
879 			}
880 			panic("tp %p bad timer_type %#x", tp, timer_type);
881 		}
882 	if (delta == 0) {
883 		callout_stop(t_callout);
884 	} else {
885 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
886 	}
887 }
888 
889 int
890 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
891 {
892 	struct callout *t_callout;
893 
894 	switch (timer_type) {
895 		case TT_DELACK:
896 			t_callout = &tp->t_timers->tt_delack;
897 			break;
898 		case TT_REXMT:
899 			t_callout = &tp->t_timers->tt_rexmt;
900 			break;
901 		case TT_PERSIST:
902 			t_callout = &tp->t_timers->tt_persist;
903 			break;
904 		case TT_KEEP:
905 			t_callout = &tp->t_timers->tt_keep;
906 			break;
907 		case TT_2MSL:
908 			t_callout = &tp->t_timers->tt_2msl;
909 			break;
910 		default:
911 			if (tp->t_fb->tfb_tcp_timer_active) {
912 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
913 			}
914 			panic("tp %p bad timer_type %#x", tp, timer_type);
915 		}
916 	return callout_active(t_callout);
917 }
918 
919 /*
920  * Stop the timer from running, and apply a flag
921  * against the timer_flags that will force the
922  * timer never to run. The flag is needed to assure
923  * a race does not leave it running and cause
924  * the timer to possibly restart itself (keep and persist
925  * especially do this).
926  */
927 int
928 tcp_timer_suspend(struct tcpcb *tp, uint32_t timer_type)
929 {
930 	struct callout *t_callout;
931 	uint32_t t_flags;
932 
933 	switch (timer_type) {
934 		case TT_DELACK:
935 			t_flags = TT_DELACK_SUS;
936 			t_callout = &tp->t_timers->tt_delack;
937 			break;
938 		case TT_REXMT:
939 			t_flags = TT_REXMT_SUS;
940 			t_callout = &tp->t_timers->tt_rexmt;
941 			break;
942 		case TT_PERSIST:
943 			t_flags = TT_PERSIST_SUS;
944 			t_callout = &tp->t_timers->tt_persist;
945 			break;
946 		case TT_KEEP:
947 			t_flags = TT_KEEP_SUS;
948 			t_callout = &tp->t_timers->tt_keep;
949 			break;
950 		case TT_2MSL:
951 			t_flags = TT_2MSL_SUS;
952 			t_callout = &tp->t_timers->tt_2msl;
953 			break;
954 		default:
955 			panic("tp:%p bad timer_type 0x%x", tp, timer_type);
956 	}
957 	tp->t_timers->tt_flags |= t_flags;
958 	return (callout_stop(t_callout));
959 }
960 
961 void
962 tcp_timers_unsuspend(struct tcpcb *tp, uint32_t timer_type)
963 {
964 	switch (timer_type) {
965 		case TT_DELACK:
966 			if (tp->t_timers->tt_flags & TT_DELACK_SUS) {
967 				tp->t_timers->tt_flags &= ~TT_DELACK_SUS;
968 				if (tp->t_flags & TF_DELACK) {
969 					/* Delayed ack timer should be up activate a timer */
970 					tp->t_flags &= ~TF_DELACK;
971 					tcp_timer_activate(tp, TT_DELACK,
972 					    tcp_delacktime);
973 				}
974 			}
975 			break;
976 		case TT_REXMT:
977 			if (tp->t_timers->tt_flags & TT_REXMT_SUS) {
978 				tp->t_timers->tt_flags &= ~TT_REXMT_SUS;
979 				if (SEQ_GT(tp->snd_max, tp->snd_una) &&
980 				    (tcp_timer_active((tp), TT_PERSIST) == 0) &&
981 				    tp->snd_wnd) {
982 					/* We have outstanding data activate a timer */
983 					tcp_timer_activate(tp, TT_REXMT,
984                                             tp->t_rxtcur);
985 				}
986 			}
987 			break;
988 		case TT_PERSIST:
989 			if (tp->t_timers->tt_flags & TT_PERSIST_SUS) {
990 				tp->t_timers->tt_flags &= ~TT_PERSIST_SUS;
991 				if (tp->snd_wnd == 0) {
992 					/* Activate the persists timer */
993 					tp->t_rxtshift = 0;
994 					tcp_setpersist(tp);
995 				}
996 			}
997 			break;
998 		case TT_KEEP:
999 			if (tp->t_timers->tt_flags & TT_KEEP_SUS) {
1000 				tp->t_timers->tt_flags &= ~TT_KEEP_SUS;
1001 				tcp_timer_activate(tp, TT_KEEP,
1002 					    TCPS_HAVEESTABLISHED(tp->t_state) ?
1003 					    TP_KEEPIDLE(tp) : TP_KEEPINIT(tp));
1004 			}
1005 			break;
1006 		case TT_2MSL:
1007 			if (tp->t_timers->tt_flags &= TT_2MSL_SUS) {
1008 				tp->t_timers->tt_flags &= ~TT_2MSL_SUS;
1009 				if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1010 				    ((tp->t_inpcb->inp_socket == NULL) ||
1011 				     (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE))) {
1012 					/* Star the 2MSL timer */
1013 					tcp_timer_activate(tp, TT_2MSL,
1014 					    (tcp_fast_finwait2_recycle) ?
1015 					    tcp_finwait2_timeout : TP_MAXIDLE(tp));
1016 				}
1017 			}
1018 			break;
1019 		default:
1020 			panic("tp:%p bad timer_type 0x%x", tp, timer_type);
1021 	}
1022 }
1023 
1024 void
1025 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
1026 {
1027 	struct callout *t_callout;
1028 
1029 	tp->t_timers->tt_flags |= TT_STOPPED;
1030 	switch (timer_type) {
1031 		case TT_DELACK:
1032 			t_callout = &tp->t_timers->tt_delack;
1033 			break;
1034 		case TT_REXMT:
1035 			t_callout = &tp->t_timers->tt_rexmt;
1036 			break;
1037 		case TT_PERSIST:
1038 			t_callout = &tp->t_timers->tt_persist;
1039 			break;
1040 		case TT_KEEP:
1041 			t_callout = &tp->t_timers->tt_keep;
1042 			break;
1043 		case TT_2MSL:
1044 			t_callout = &tp->t_timers->tt_2msl;
1045 			break;
1046 		default:
1047 			if (tp->t_fb->tfb_tcp_timer_stop) {
1048 				/*
1049 				 * XXXrrs we need to look at this with the
1050 				 * stop case below (flags).
1051 				 */
1052 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
1053 				return;
1054 			}
1055 			panic("tp %p bad timer_type %#x", tp, timer_type);
1056 		}
1057 
1058 	if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
1059 		/*
1060 		 * Can't stop the callout, defer tcpcb actual deletion
1061 		 * to the last one. We do this using the async drain
1062 		 * function and incrementing the count in
1063 		 */
1064 		tp->t_timers->tt_draincnt++;
1065 	}
1066 }
1067