xref: /freebsd/sys/netinet/tcp_timer.c (revision 952364486a4b9d135e4b28f7f88a8703a74eae6f)
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/rss_config.h>
55 #include <net/vnet.h>
56 #include <net/netisr.h>
57 
58 #include <netinet/cc.h>
59 #include <netinet/in.h>
60 #include <netinet/in_pcb.h>
61 #include <netinet/in_rss.h>
62 #include <netinet/in_systm.h>
63 #ifdef INET6
64 #include <netinet6/in6_pcb.h>
65 #endif
66 #include <netinet/ip_var.h>
67 #include <netinet/tcp_fsm.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #ifdef INET6
71 #include <netinet6/tcp6_var.h>
72 #endif
73 #include <netinet/tcpip.h>
74 #ifdef TCPDEBUG
75 #include <netinet/tcp_debug.h>
76 #endif
77 
78 int	tcp_keepinit;
79 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
80     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
81 
82 int	tcp_keepidle;
83 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
84     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
85 
86 int	tcp_keepintvl;
87 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
88     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
89 
90 int	tcp_delacktime;
91 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
92     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
93     "Time before a delayed ACK is sent");
94 
95 int	tcp_msl;
96 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
97     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
98 
99 int	tcp_rexmit_min;
100 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
101     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
102     "Minimum Retransmission Timeout");
103 
104 int	tcp_rexmit_slop;
105 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
106     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
107     "Retransmission Timer Slop");
108 
109 static int	always_keepalive = 1;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
111     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
112 
113 int    tcp_fast_finwait2_recycle = 0;
114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
115     &tcp_fast_finwait2_recycle, 0,
116     "Recycle closed FIN_WAIT_2 connections faster");
117 
118 int    tcp_finwait2_timeout;
119 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
120     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
121 
122 int	tcp_keepcnt = TCPTV_KEEPCNT;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
124     "Number of keepalive probes to send");
125 
126 	/* max idle probes */
127 int	tcp_maxpersistidle;
128 
129 static int	tcp_rexmit_drop_options = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
131     &tcp_rexmit_drop_options, 0,
132     "Drop TCP options from 3rd and later retransmitted SYN");
133 
134 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
135 #define	V_tcp_pmtud_blackhole_detect	VNET(tcp_pmtud_blackhole_detect)
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
137     CTLFLAG_RW|CTLFLAG_VNET,
138     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
139     "Path MTU Discovery Black Hole Detection Enabled");
140 
141 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
142 #define	V_tcp_pmtud_blackhole_activated \
143     VNET(tcp_pmtud_blackhole_activated)
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
145     CTLFLAG_RD|CTLFLAG_VNET,
146     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
147     "Path MTU Discovery Black Hole Detection, Activation Count");
148 
149 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
150 #define	V_tcp_pmtud_blackhole_activated_min_mss \
151     VNET(tcp_pmtud_blackhole_activated_min_mss)
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
153     CTLFLAG_RD|CTLFLAG_VNET,
154     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
155     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
156 
157 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
158 #define	V_tcp_pmtud_blackhole_failed	VNET(tcp_pmtud_blackhole_failed)
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
160     CTLFLAG_RD|CTLFLAG_VNET,
161     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
162     "Path MTU Discovery Black Hole Detection, Failure Count");
163 
164 #ifdef INET
165 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
166 #define	V_tcp_pmtud_blackhole_mss	VNET(tcp_pmtud_blackhole_mss)
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
168     CTLFLAG_RW|CTLFLAG_VNET,
169     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
170     "Path MTU Discovery Black Hole Detection lowered MSS");
171 #endif
172 
173 #ifdef INET6
174 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
175 #define	V_tcp_v6pmtud_blackhole_mss	VNET(tcp_v6pmtud_blackhole_mss)
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
177     CTLFLAG_RW|CTLFLAG_VNET,
178     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
179     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
180 #endif
181 
182 #ifdef	RSS
183 static int	per_cpu_timers = 1;
184 #else
185 static int	per_cpu_timers = 0;
186 #endif
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
188     &per_cpu_timers , 0, "run tcp timers on all cpus");
189 
190 #if 0
191 #define	INP_CPU(inp)	(per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
192 		((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
193 #endif
194 
195 /*
196  * Map the given inp to a CPU id.
197  *
198  * This queries RSS if it's compiled in, else it defaults to the current
199  * CPU ID.
200  */
201 static inline int
202 inp_to_cpuid(struct inpcb *inp)
203 {
204 	u_int cpuid;
205 
206 #ifdef	RSS
207 	if (per_cpu_timers) {
208 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
209 		if (cpuid == NETISR_CPUID_NONE)
210 			return (curcpu);	/* XXX */
211 		else
212 			return (cpuid);
213 	}
214 #else
215 	/* Legacy, pre-RSS behaviour */
216 	if (per_cpu_timers) {
217 		/*
218 		 * We don't have a flowid -> cpuid mapping, so cheat and
219 		 * just map unknown cpuids to curcpu.  Not the best, but
220 		 * apparently better than defaulting to swi 0.
221 		 */
222 		cpuid = inp->inp_flowid % (mp_maxid + 1);
223 		if (! CPU_ABSENT(cpuid))
224 			return (cpuid);
225 		return (curcpu);
226 	}
227 #endif
228 	/* Default for RSS and non-RSS - cpuid 0 */
229 	else {
230 		return (0);
231 	}
232 }
233 
234 /*
235  * Tcp protocol timeout routine called every 500 ms.
236  * Updates timestamps used for TCP
237  * causes finite state machine actions if timers expire.
238  */
239 void
240 tcp_slowtimo(void)
241 {
242 	VNET_ITERATOR_DECL(vnet_iter);
243 
244 	VNET_LIST_RLOCK_NOSLEEP();
245 	VNET_FOREACH(vnet_iter) {
246 		CURVNET_SET(vnet_iter);
247 		(void) tcp_tw_2msl_scan(0);
248 		CURVNET_RESTORE();
249 	}
250 	VNET_LIST_RUNLOCK_NOSLEEP();
251 }
252 
253 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
254     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
255 
256 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
257     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
258 
259 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
260 
261 /*
262  * TCP timer processing.
263  */
264 
265 void
266 tcp_timer_delack(void *xtp)
267 {
268 	struct tcpcb *tp = xtp;
269 	struct inpcb *inp;
270 	CURVNET_SET(tp->t_vnet);
271 
272 	inp = tp->t_inpcb;
273 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
274 	INP_WLOCK(inp);
275 	if (callout_pending(&tp->t_timers->tt_delack) ||
276 	    !callout_active(&tp->t_timers->tt_delack)) {
277 		INP_WUNLOCK(inp);
278 		CURVNET_RESTORE();
279 		return;
280 	}
281 	callout_deactivate(&tp->t_timers->tt_delack);
282 	if ((inp->inp_flags & INP_DROPPED) != 0) {
283 		INP_WUNLOCK(inp);
284 		CURVNET_RESTORE();
285 		return;
286 	}
287 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
288 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
289 	KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
290 		("%s: tp %p delack callout should be running", __func__, tp));
291 
292 	tp->t_flags |= TF_ACKNOW;
293 	TCPSTAT_INC(tcps_delack);
294 	(void) tcp_output(tp);
295 	INP_WUNLOCK(inp);
296 	CURVNET_RESTORE();
297 }
298 
299 void
300 tcp_timer_2msl(void *xtp)
301 {
302 	struct tcpcb *tp = xtp;
303 	struct inpcb *inp;
304 	CURVNET_SET(tp->t_vnet);
305 #ifdef TCPDEBUG
306 	int ostate;
307 
308 	ostate = tp->t_state;
309 #endif
310 	INP_INFO_WLOCK(&V_tcbinfo);
311 	inp = tp->t_inpcb;
312 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
313 	INP_WLOCK(inp);
314 	tcp_free_sackholes(tp);
315 	if (callout_pending(&tp->t_timers->tt_2msl) ||
316 	    !callout_active(&tp->t_timers->tt_2msl)) {
317 		INP_WUNLOCK(tp->t_inpcb);
318 		INP_INFO_WUNLOCK(&V_tcbinfo);
319 		CURVNET_RESTORE();
320 		return;
321 	}
322 	callout_deactivate(&tp->t_timers->tt_2msl);
323 	if ((inp->inp_flags & INP_DROPPED) != 0) {
324 		INP_WUNLOCK(inp);
325 		INP_INFO_WUNLOCK(&V_tcbinfo);
326 		CURVNET_RESTORE();
327 		return;
328 	}
329 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
330 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
331 	KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
332 		("%s: tp %p 2msl callout should be running", __func__, tp));
333 	/*
334 	 * 2 MSL timeout in shutdown went off.  If we're closed but
335 	 * still waiting for peer to close and connection has been idle
336 	 * too long, or if 2MSL time is up from TIME_WAIT, delete connection
337 	 * control block.  Otherwise, check again in a bit.
338 	 *
339 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
340 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
341 	 * Ignore fact that there were recent incoming segments.
342 	 */
343 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
344 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
345 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
346 		TCPSTAT_INC(tcps_finwait2_drops);
347 		tp = tcp_close(tp);
348 	} else {
349 		if (tp->t_state != TCPS_TIME_WAIT &&
350 		   ticks - tp->t_rcvtime <= TP_MAXIDLE(tp))
351 		       callout_reset_on(&tp->t_timers->tt_2msl,
352 			   TP_KEEPINTVL(tp), tcp_timer_2msl, tp,
353 			   inp_to_cpuid(inp));
354 	       else
355 		       tp = tcp_close(tp);
356        }
357 
358 #ifdef TCPDEBUG
359 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
360 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
361 			  PRU_SLOWTIMO);
362 #endif
363 	if (tp != NULL)
364 		INP_WUNLOCK(inp);
365 	INP_INFO_WUNLOCK(&V_tcbinfo);
366 	CURVNET_RESTORE();
367 }
368 
369 void
370 tcp_timer_keep(void *xtp)
371 {
372 	struct tcpcb *tp = xtp;
373 	struct tcptemp *t_template;
374 	struct inpcb *inp;
375 	CURVNET_SET(tp->t_vnet);
376 #ifdef TCPDEBUG
377 	int ostate;
378 
379 	ostate = tp->t_state;
380 #endif
381 	INP_INFO_WLOCK(&V_tcbinfo);
382 	inp = tp->t_inpcb;
383 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
384 	INP_WLOCK(inp);
385 	if (callout_pending(&tp->t_timers->tt_keep) ||
386 	    !callout_active(&tp->t_timers->tt_keep)) {
387 		INP_WUNLOCK(inp);
388 		INP_INFO_WUNLOCK(&V_tcbinfo);
389 		CURVNET_RESTORE();
390 		return;
391 	}
392 	callout_deactivate(&tp->t_timers->tt_keep);
393 	if ((inp->inp_flags & INP_DROPPED) != 0) {
394 		INP_WUNLOCK(inp);
395 		INP_INFO_WUNLOCK(&V_tcbinfo);
396 		CURVNET_RESTORE();
397 		return;
398 	}
399 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
400 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
401 	KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
402 		("%s: tp %p keep callout should be running", __func__, tp));
403 	/*
404 	 * Keep-alive timer went off; send something
405 	 * or drop connection if idle for too long.
406 	 */
407 	TCPSTAT_INC(tcps_keeptimeo);
408 	if (tp->t_state < TCPS_ESTABLISHED)
409 		goto dropit;
410 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
411 	    tp->t_state <= TCPS_CLOSING) {
412 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
413 			goto dropit;
414 		/*
415 		 * Send a packet designed to force a response
416 		 * if the peer is up and reachable:
417 		 * either an ACK if the connection is still alive,
418 		 * or an RST if the peer has closed the connection
419 		 * due to timeout or reboot.
420 		 * Using sequence number tp->snd_una-1
421 		 * causes the transmitted zero-length segment
422 		 * to lie outside the receive window;
423 		 * by the protocol spec, this requires the
424 		 * correspondent TCP to respond.
425 		 */
426 		TCPSTAT_INC(tcps_keepprobe);
427 		t_template = tcpip_maketemplate(inp);
428 		if (t_template) {
429 			tcp_respond(tp, t_template->tt_ipgen,
430 				    &t_template->tt_t, (struct mbuf *)NULL,
431 				    tp->rcv_nxt, tp->snd_una - 1, 0);
432 			free(t_template, M_TEMP);
433 		}
434 		callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
435 		    tcp_timer_keep, tp, inp_to_cpuid(inp));
436 	} else
437 		callout_reset_on(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
438 		    tcp_timer_keep, tp, inp_to_cpuid(inp));
439 
440 #ifdef TCPDEBUG
441 	if (inp->inp_socket->so_options & SO_DEBUG)
442 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
443 			  PRU_SLOWTIMO);
444 #endif
445 	INP_WUNLOCK(inp);
446 	INP_INFO_WUNLOCK(&V_tcbinfo);
447 	CURVNET_RESTORE();
448 	return;
449 
450 dropit:
451 	TCPSTAT_INC(tcps_keepdrops);
452 	tp = tcp_drop(tp, ETIMEDOUT);
453 
454 #ifdef TCPDEBUG
455 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
456 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
457 			  PRU_SLOWTIMO);
458 #endif
459 	if (tp != NULL)
460 		INP_WUNLOCK(tp->t_inpcb);
461 	INP_INFO_WUNLOCK(&V_tcbinfo);
462 	CURVNET_RESTORE();
463 }
464 
465 void
466 tcp_timer_persist(void *xtp)
467 {
468 	struct tcpcb *tp = xtp;
469 	struct inpcb *inp;
470 	CURVNET_SET(tp->t_vnet);
471 #ifdef TCPDEBUG
472 	int ostate;
473 
474 	ostate = tp->t_state;
475 #endif
476 	INP_INFO_WLOCK(&V_tcbinfo);
477 	inp = tp->t_inpcb;
478 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
479 	INP_WLOCK(inp);
480 	if (callout_pending(&tp->t_timers->tt_persist) ||
481 	    !callout_active(&tp->t_timers->tt_persist)) {
482 		INP_WUNLOCK(inp);
483 		INP_INFO_WUNLOCK(&V_tcbinfo);
484 		CURVNET_RESTORE();
485 		return;
486 	}
487 	callout_deactivate(&tp->t_timers->tt_persist);
488 	if ((inp->inp_flags & INP_DROPPED) != 0) {
489 		INP_WUNLOCK(inp);
490 		INP_INFO_WUNLOCK(&V_tcbinfo);
491 		CURVNET_RESTORE();
492 		return;
493 	}
494 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
495 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
496 	KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
497 		("%s: tp %p persist callout should be running", __func__, tp));
498 	/*
499 	 * Persistance timer into zero window.
500 	 * Force a byte to be output, if possible.
501 	 */
502 	TCPSTAT_INC(tcps_persisttimeo);
503 	/*
504 	 * Hack: if the peer is dead/unreachable, we do not
505 	 * time out if the window is closed.  After a full
506 	 * backoff, drop the connection if the idle time
507 	 * (no responses to probes) reaches the maximum
508 	 * backoff that we would use if retransmitting.
509 	 */
510 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
511 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
512 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
513 		TCPSTAT_INC(tcps_persistdrop);
514 		tp = tcp_drop(tp, ETIMEDOUT);
515 		goto out;
516 	}
517 	/*
518 	 * If the user has closed the socket then drop a persisting
519 	 * connection after a much reduced timeout.
520 	 */
521 	if (tp->t_state > TCPS_CLOSE_WAIT &&
522 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
523 		TCPSTAT_INC(tcps_persistdrop);
524 		tp = tcp_drop(tp, ETIMEDOUT);
525 		goto out;
526 	}
527 	tcp_setpersist(tp);
528 	tp->t_flags |= TF_FORCEDATA;
529 	(void) tcp_output(tp);
530 	tp->t_flags &= ~TF_FORCEDATA;
531 
532 out:
533 #ifdef TCPDEBUG
534 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
535 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
536 #endif
537 	if (tp != NULL)
538 		INP_WUNLOCK(inp);
539 	INP_INFO_WUNLOCK(&V_tcbinfo);
540 	CURVNET_RESTORE();
541 }
542 
543 void
544 tcp_timer_rexmt(void * xtp)
545 {
546 	struct tcpcb *tp = xtp;
547 	CURVNET_SET(tp->t_vnet);
548 	int rexmt;
549 	int headlocked;
550 	struct inpcb *inp;
551 #ifdef TCPDEBUG
552 	int ostate;
553 
554 	ostate = tp->t_state;
555 #endif
556 
557 	INP_INFO_RLOCK(&V_tcbinfo);
558 	inp = tp->t_inpcb;
559 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
560 	INP_WLOCK(inp);
561 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
562 	    !callout_active(&tp->t_timers->tt_rexmt)) {
563 		INP_WUNLOCK(inp);
564 		INP_INFO_RUNLOCK(&V_tcbinfo);
565 		CURVNET_RESTORE();
566 		return;
567 	}
568 	callout_deactivate(&tp->t_timers->tt_rexmt);
569 	if ((inp->inp_flags & INP_DROPPED) != 0) {
570 		INP_WUNLOCK(inp);
571 		INP_INFO_RUNLOCK(&V_tcbinfo);
572 		CURVNET_RESTORE();
573 		return;
574 	}
575 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
576 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
577 	KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
578 		("%s: tp %p rexmt callout should be running", __func__, tp));
579 	tcp_free_sackholes(tp);
580 	/*
581 	 * Retransmission timer went off.  Message has not
582 	 * been acked within retransmit interval.  Back off
583 	 * to a longer retransmit interval and retransmit one segment.
584 	 */
585 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
586 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
587 		TCPSTAT_INC(tcps_timeoutdrop);
588 		in_pcbref(inp);
589 		INP_INFO_RUNLOCK(&V_tcbinfo);
590 		INP_WUNLOCK(inp);
591 		INP_INFO_WLOCK(&V_tcbinfo);
592 		INP_WLOCK(inp);
593 		if (in_pcbrele_wlocked(inp)) {
594 			INP_INFO_WUNLOCK(&V_tcbinfo);
595 			CURVNET_RESTORE();
596 			return;
597 		}
598 		if (inp->inp_flags & INP_DROPPED) {
599 			INP_WUNLOCK(inp);
600 			INP_INFO_WUNLOCK(&V_tcbinfo);
601 			CURVNET_RESTORE();
602 			return;
603 		}
604 
605 		tp = tcp_drop(tp, tp->t_softerror ?
606 			      tp->t_softerror : ETIMEDOUT);
607 		headlocked = 1;
608 		goto out;
609 	}
610 	INP_INFO_RUNLOCK(&V_tcbinfo);
611 	headlocked = 0;
612 	if (tp->t_state == TCPS_SYN_SENT) {
613 		/*
614 		 * If the SYN was retransmitted, indicate CWND to be
615 		 * limited to 1 segment in cc_conn_init().
616 		 */
617 		tp->snd_cwnd = 1;
618 	} else if (tp->t_rxtshift == 1) {
619 		/*
620 		 * first retransmit; record ssthresh and cwnd so they can
621 		 * be recovered if this turns out to be a "bad" retransmit.
622 		 * A retransmit is considered "bad" if an ACK for this
623 		 * segment is received within RTT/2 interval; the assumption
624 		 * here is that the ACK was already in flight.  See
625 		 * "On Estimating End-to-End Network Path Properties" by
626 		 * Allman and Paxson for more details.
627 		 */
628 		tp->snd_cwnd_prev = tp->snd_cwnd;
629 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
630 		tp->snd_recover_prev = tp->snd_recover;
631 		if (IN_FASTRECOVERY(tp->t_flags))
632 			tp->t_flags |= TF_WASFRECOVERY;
633 		else
634 			tp->t_flags &= ~TF_WASFRECOVERY;
635 		if (IN_CONGRECOVERY(tp->t_flags))
636 			tp->t_flags |= TF_WASCRECOVERY;
637 		else
638 			tp->t_flags &= ~TF_WASCRECOVERY;
639 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
640 		tp->t_flags |= TF_PREVVALID;
641 	} else
642 		tp->t_flags &= ~TF_PREVVALID;
643 	TCPSTAT_INC(tcps_rexmttimeo);
644 	if (tp->t_state == TCPS_SYN_SENT)
645 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
646 	else
647 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
648 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
649 		      tp->t_rttmin, TCPTV_REXMTMAX);
650 
651 	/*
652 	 * We enter the path for PLMTUD if connection is established or, if
653 	 * connection is FIN_WAIT_1 status, reason for the last is that if
654 	 * amount of data we send is very small, we could send it in couple of
655 	 * packets and process straight to FIN. In that case we won't catch
656 	 * ESTABLISHED state.
657 	 */
658 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
659 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
660 		int optlen;
661 #ifdef INET6
662 		int isipv6;
663 #endif
664 
665 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
666 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
667 		    (tp->t_rxtshift <= 2)) {
668 			/*
669 			 * Enter Path MTU Black-hole Detection mechanism:
670 			 * - Disable Path MTU Discovery (IP "DF" bit).
671 			 * - Reduce MTU to lower value than what we
672 			 *   negotiated with peer.
673 			 */
674 			/* Record that we may have found a black hole. */
675 			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
676 
677 			/* Keep track of previous MSS. */
678 			optlen = tp->t_maxopd - tp->t_maxseg;
679 			tp->t_pmtud_saved_maxopd = tp->t_maxopd;
680 
681 			/*
682 			 * Reduce the MSS to blackhole value or to the default
683 			 * in an attempt to retransmit.
684 			 */
685 #ifdef INET6
686 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
687 			if (isipv6 &&
688 			    tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
689 				/* Use the sysctl tuneable blackhole MSS. */
690 				tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
691 				V_tcp_pmtud_blackhole_activated++;
692 			} else if (isipv6) {
693 				/* Use the default MSS. */
694 				tp->t_maxopd = V_tcp_v6mssdflt;
695 				/*
696 				 * Disable Path MTU Discovery when we switch to
697 				 * minmss.
698 				 */
699 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
700 				V_tcp_pmtud_blackhole_activated_min_mss++;
701 			}
702 #endif
703 #if defined(INET6) && defined(INET)
704 			else
705 #endif
706 #ifdef INET
707 			if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
708 				/* Use the sysctl tuneable blackhole MSS. */
709 				tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
710 				V_tcp_pmtud_blackhole_activated++;
711 			} else {
712 				/* Use the default MSS. */
713 				tp->t_maxopd = V_tcp_mssdflt;
714 				/*
715 				 * Disable Path MTU Discovery when we switch to
716 				 * minmss.
717 				 */
718 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
719 				V_tcp_pmtud_blackhole_activated_min_mss++;
720 			}
721 #endif
722 			tp->t_maxseg = tp->t_maxopd - optlen;
723 			/*
724 			 * Reset the slow-start flight size
725 			 * as it may depend on the new MSS.
726 			 */
727 			if (CC_ALGO(tp)->conn_init != NULL)
728 				CC_ALGO(tp)->conn_init(tp->ccv);
729 		} else {
730 			/*
731 			 * If further retransmissions are still unsuccessful
732 			 * with a lowered MTU, maybe this isn't a blackhole and
733 			 * we restore the previous MSS and blackhole detection
734 			 * flags.
735 			 */
736 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
737 			    (tp->t_rxtshift > 4)) {
738 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
739 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
740 				optlen = tp->t_maxopd - tp->t_maxseg;
741 				tp->t_maxopd = tp->t_pmtud_saved_maxopd;
742 				tp->t_maxseg = tp->t_maxopd - optlen;
743 				V_tcp_pmtud_blackhole_failed++;
744 				/*
745 				 * Reset the slow-start flight size as it
746 				 * may depend on the new MSS.
747 				 */
748 				if (CC_ALGO(tp)->conn_init != NULL)
749 					CC_ALGO(tp)->conn_init(tp->ccv);
750 			}
751 		}
752 	}
753 
754 	/*
755 	 * Disable RFC1323 and SACK if we haven't got any response to
756 	 * our third SYN to work-around some broken terminal servers
757 	 * (most of which have hopefully been retired) that have bad VJ
758 	 * header compression code which trashes TCP segments containing
759 	 * unknown-to-them TCP options.
760 	 */
761 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
762 	    (tp->t_rxtshift == 3))
763 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
764 	/*
765 	 * If we backed off this far, our srtt estimate is probably bogus.
766 	 * Clobber it so we'll take the next rtt measurement as our srtt;
767 	 * move the current srtt into rttvar to keep the current
768 	 * retransmit times until then.
769 	 */
770 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
771 #ifdef INET6
772 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
773 			in6_losing(tp->t_inpcb);
774 #endif
775 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
776 		tp->t_srtt = 0;
777 	}
778 	tp->snd_nxt = tp->snd_una;
779 	tp->snd_recover = tp->snd_max;
780 	/*
781 	 * Force a segment to be sent.
782 	 */
783 	tp->t_flags |= TF_ACKNOW;
784 	/*
785 	 * If timing a segment in this window, stop the timer.
786 	 */
787 	tp->t_rtttime = 0;
788 
789 	cc_cong_signal(tp, NULL, CC_RTO);
790 
791 	(void) tcp_output(tp);
792 
793 out:
794 #ifdef TCPDEBUG
795 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
796 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
797 			  PRU_SLOWTIMO);
798 #endif
799 	if (tp != NULL)
800 		INP_WUNLOCK(inp);
801 	if (headlocked)
802 		INP_INFO_WUNLOCK(&V_tcbinfo);
803 	CURVNET_RESTORE();
804 }
805 
806 void
807 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
808 {
809 	struct callout *t_callout;
810 	timeout_t *f_callout;
811 	struct inpcb *inp = tp->t_inpcb;
812 	int cpu = inp_to_cpuid(inp);
813 
814 #ifdef TCP_OFFLOAD
815 	if (tp->t_flags & TF_TOE)
816 		return;
817 #endif
818 
819 	if (tp->t_timers->tt_flags & TT_STOPPED)
820 		return;
821 
822 	switch (timer_type) {
823 		case TT_DELACK:
824 			t_callout = &tp->t_timers->tt_delack;
825 			f_callout = tcp_timer_delack;
826 			break;
827 		case TT_REXMT:
828 			t_callout = &tp->t_timers->tt_rexmt;
829 			f_callout = tcp_timer_rexmt;
830 			break;
831 		case TT_PERSIST:
832 			t_callout = &tp->t_timers->tt_persist;
833 			f_callout = tcp_timer_persist;
834 			break;
835 		case TT_KEEP:
836 			t_callout = &tp->t_timers->tt_keep;
837 			f_callout = tcp_timer_keep;
838 			break;
839 		case TT_2MSL:
840 			t_callout = &tp->t_timers->tt_2msl;
841 			f_callout = tcp_timer_2msl;
842 			break;
843 		default:
844 			panic("tp %p bad timer_type %#x", tp, timer_type);
845 		}
846 	if (delta == 0) {
847 		if ((tp->t_timers->tt_flags & timer_type) &&
848 		    callout_stop(t_callout)) {
849 			tp->t_timers->tt_flags &= ~timer_type;
850 		}
851 	} else {
852 		if ((tp->t_timers->tt_flags & timer_type) == 0) {
853 			tp->t_timers->tt_flags |= timer_type;
854 			callout_reset_on(t_callout, delta, f_callout, tp, cpu);
855 		} else {
856 			/* Reset already running callout on the same CPU. */
857 			callout_reset(t_callout, delta, f_callout, tp);
858 		}
859 	}
860 }
861 
862 int
863 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
864 {
865 	struct callout *t_callout;
866 
867 	switch (timer_type) {
868 		case TT_DELACK:
869 			t_callout = &tp->t_timers->tt_delack;
870 			break;
871 		case TT_REXMT:
872 			t_callout = &tp->t_timers->tt_rexmt;
873 			break;
874 		case TT_PERSIST:
875 			t_callout = &tp->t_timers->tt_persist;
876 			break;
877 		case TT_KEEP:
878 			t_callout = &tp->t_timers->tt_keep;
879 			break;
880 		case TT_2MSL:
881 			t_callout = &tp->t_timers->tt_2msl;
882 			break;
883 		default:
884 			panic("tp %p bad timer_type %#x", tp, timer_type);
885 		}
886 	return callout_active(t_callout);
887 }
888 
889 void
890 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
891 {
892 	struct callout *t_callout;
893 	timeout_t *f_callout;
894 
895 	tp->t_timers->tt_flags |= TT_STOPPED;
896 
897 	switch (timer_type) {
898 		case TT_DELACK:
899 			t_callout = &tp->t_timers->tt_delack;
900 			f_callout = tcp_timer_delack_discard;
901 			break;
902 		case TT_REXMT:
903 			t_callout = &tp->t_timers->tt_rexmt;
904 			f_callout = tcp_timer_rexmt_discard;
905 			break;
906 		case TT_PERSIST:
907 			t_callout = &tp->t_timers->tt_persist;
908 			f_callout = tcp_timer_persist_discard;
909 			break;
910 		case TT_KEEP:
911 			t_callout = &tp->t_timers->tt_keep;
912 			f_callout = tcp_timer_keep_discard;
913 			break;
914 		case TT_2MSL:
915 			t_callout = &tp->t_timers->tt_2msl;
916 			f_callout = tcp_timer_2msl_discard;
917 			break;
918 		default:
919 			panic("tp %p bad timer_type %#x", tp, timer_type);
920 		}
921 
922 	if (tp->t_timers->tt_flags & timer_type) {
923 		if (callout_stop(t_callout)) {
924 			tp->t_timers->tt_flags &= ~timer_type;
925 		} else {
926 			/*
927 			 * Can't stop the callout, defer tcpcb actual deletion
928 			 * to the last tcp timer discard callout.
929 			 * The TT_STOPPED flag will ensure that no tcp timer
930 			 * callouts can be restarted on our behalf, and
931 			 * past this point currently running callouts waiting
932 			 * on inp lock will return right away after the
933 			 * classical check for callout reset/stop events:
934 			 * callout_pending() || !callout_active()
935 			 */
936 			callout_reset(t_callout, 1, f_callout, tp);
937 		}
938 	}
939 }
940 
941 #define	ticks_to_msecs(t)	(1000*(t) / hz)
942 
943 void
944 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
945     struct xtcp_timer *xtimer)
946 {
947 	sbintime_t now;
948 
949 	bzero(xtimer, sizeof(*xtimer));
950 	if (timer == NULL)
951 		return;
952 	now = getsbinuptime();
953 	if (callout_active(&timer->tt_delack))
954 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
955 	if (callout_active(&timer->tt_rexmt))
956 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
957 	if (callout_active(&timer->tt_persist))
958 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
959 	if (callout_active(&timer->tt_keep))
960 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
961 	if (callout_active(&timer->tt_2msl))
962 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
963 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
964 }
965