xref: /freebsd/sys/netinet/tcp_timer.c (revision ca2e4ecd7395ba655ab4bebe7262a06e634216ce)
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 			if (!callout_reset(&tp->t_timers->tt_2msl,
352 			   TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
353 				tp->t_timers->tt_flags &= ~TT_2MSL_RST;
354 			}
355 		} else
356 		       tp = tcp_close(tp);
357        }
358 
359 #ifdef TCPDEBUG
360 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
361 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
362 			  PRU_SLOWTIMO);
363 #endif
364 	if (tp != NULL)
365 		INP_WUNLOCK(inp);
366 	INP_INFO_WUNLOCK(&V_tcbinfo);
367 	CURVNET_RESTORE();
368 }
369 
370 void
371 tcp_timer_keep(void *xtp)
372 {
373 	struct tcpcb *tp = xtp;
374 	struct tcptemp *t_template;
375 	struct inpcb *inp;
376 	CURVNET_SET(tp->t_vnet);
377 #ifdef TCPDEBUG
378 	int ostate;
379 
380 	ostate = tp->t_state;
381 #endif
382 	INP_INFO_WLOCK(&V_tcbinfo);
383 	inp = tp->t_inpcb;
384 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
385 	INP_WLOCK(inp);
386 	if (callout_pending(&tp->t_timers->tt_keep) ||
387 	    !callout_active(&tp->t_timers->tt_keep)) {
388 		INP_WUNLOCK(inp);
389 		INP_INFO_WUNLOCK(&V_tcbinfo);
390 		CURVNET_RESTORE();
391 		return;
392 	}
393 	callout_deactivate(&tp->t_timers->tt_keep);
394 	if ((inp->inp_flags & INP_DROPPED) != 0) {
395 		INP_WUNLOCK(inp);
396 		INP_INFO_WUNLOCK(&V_tcbinfo);
397 		CURVNET_RESTORE();
398 		return;
399 	}
400 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
401 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
402 	KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
403 		("%s: tp %p keep callout should be running", __func__, tp));
404 	/*
405 	 * Keep-alive timer went off; send something
406 	 * or drop connection if idle for too long.
407 	 */
408 	TCPSTAT_INC(tcps_keeptimeo);
409 	if (tp->t_state < TCPS_ESTABLISHED)
410 		goto dropit;
411 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
412 	    tp->t_state <= TCPS_CLOSING) {
413 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
414 			goto dropit;
415 		/*
416 		 * Send a packet designed to force a response
417 		 * if the peer is up and reachable:
418 		 * either an ACK if the connection is still alive,
419 		 * or an RST if the peer has closed the connection
420 		 * due to timeout or reboot.
421 		 * Using sequence number tp->snd_una-1
422 		 * causes the transmitted zero-length segment
423 		 * to lie outside the receive window;
424 		 * by the protocol spec, this requires the
425 		 * correspondent TCP to respond.
426 		 */
427 		TCPSTAT_INC(tcps_keepprobe);
428 		t_template = tcpip_maketemplate(inp);
429 		if (t_template) {
430 			tcp_respond(tp, t_template->tt_ipgen,
431 				    &t_template->tt_t, (struct mbuf *)NULL,
432 				    tp->rcv_nxt, tp->snd_una - 1, 0);
433 			free(t_template, M_TEMP);
434 		}
435 		if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
436 		    tcp_timer_keep, tp)) {
437 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
438 		}
439 	} else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
440 		    tcp_timer_keep, tp)) {
441 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
442 		}
443 
444 #ifdef TCPDEBUG
445 	if (inp->inp_socket->so_options & SO_DEBUG)
446 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
447 			  PRU_SLOWTIMO);
448 #endif
449 	INP_WUNLOCK(inp);
450 	INP_INFO_WUNLOCK(&V_tcbinfo);
451 	CURVNET_RESTORE();
452 	return;
453 
454 dropit:
455 	TCPSTAT_INC(tcps_keepdrops);
456 	tp = tcp_drop(tp, ETIMEDOUT);
457 
458 #ifdef TCPDEBUG
459 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
460 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
461 			  PRU_SLOWTIMO);
462 #endif
463 	if (tp != NULL)
464 		INP_WUNLOCK(tp->t_inpcb);
465 	INP_INFO_WUNLOCK(&V_tcbinfo);
466 	CURVNET_RESTORE();
467 }
468 
469 void
470 tcp_timer_persist(void *xtp)
471 {
472 	struct tcpcb *tp = xtp;
473 	struct inpcb *inp;
474 	CURVNET_SET(tp->t_vnet);
475 #ifdef TCPDEBUG
476 	int ostate;
477 
478 	ostate = tp->t_state;
479 #endif
480 	INP_INFO_WLOCK(&V_tcbinfo);
481 	inp = tp->t_inpcb;
482 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
483 	INP_WLOCK(inp);
484 	if (callout_pending(&tp->t_timers->tt_persist) ||
485 	    !callout_active(&tp->t_timers->tt_persist)) {
486 		INP_WUNLOCK(inp);
487 		INP_INFO_WUNLOCK(&V_tcbinfo);
488 		CURVNET_RESTORE();
489 		return;
490 	}
491 	callout_deactivate(&tp->t_timers->tt_persist);
492 	if ((inp->inp_flags & INP_DROPPED) != 0) {
493 		INP_WUNLOCK(inp);
494 		INP_INFO_WUNLOCK(&V_tcbinfo);
495 		CURVNET_RESTORE();
496 		return;
497 	}
498 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
499 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
500 	KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
501 		("%s: tp %p persist callout should be running", __func__, tp));
502 	/*
503 	 * Persistance timer into zero window.
504 	 * Force a byte to be output, if possible.
505 	 */
506 	TCPSTAT_INC(tcps_persisttimeo);
507 	/*
508 	 * Hack: if the peer is dead/unreachable, we do not
509 	 * time out if the window is closed.  After a full
510 	 * backoff, drop the connection if the idle time
511 	 * (no responses to probes) reaches the maximum
512 	 * backoff that we would use if retransmitting.
513 	 */
514 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
515 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
516 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
517 		TCPSTAT_INC(tcps_persistdrop);
518 		tp = tcp_drop(tp, ETIMEDOUT);
519 		goto out;
520 	}
521 	/*
522 	 * If the user has closed the socket then drop a persisting
523 	 * connection after a much reduced timeout.
524 	 */
525 	if (tp->t_state > TCPS_CLOSE_WAIT &&
526 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
527 		TCPSTAT_INC(tcps_persistdrop);
528 		tp = tcp_drop(tp, ETIMEDOUT);
529 		goto out;
530 	}
531 	tcp_setpersist(tp);
532 	tp->t_flags |= TF_FORCEDATA;
533 	(void) tcp_output(tp);
534 	tp->t_flags &= ~TF_FORCEDATA;
535 
536 out:
537 #ifdef TCPDEBUG
538 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
539 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
540 #endif
541 	if (tp != NULL)
542 		INP_WUNLOCK(inp);
543 	INP_INFO_WUNLOCK(&V_tcbinfo);
544 	CURVNET_RESTORE();
545 }
546 
547 void
548 tcp_timer_rexmt(void * xtp)
549 {
550 	struct tcpcb *tp = xtp;
551 	CURVNET_SET(tp->t_vnet);
552 	int rexmt;
553 	int headlocked;
554 	struct inpcb *inp;
555 #ifdef TCPDEBUG
556 	int ostate;
557 
558 	ostate = tp->t_state;
559 #endif
560 
561 	INP_INFO_RLOCK(&V_tcbinfo);
562 	inp = tp->t_inpcb;
563 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
564 	INP_WLOCK(inp);
565 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
566 	    !callout_active(&tp->t_timers->tt_rexmt)) {
567 		INP_WUNLOCK(inp);
568 		INP_INFO_RUNLOCK(&V_tcbinfo);
569 		CURVNET_RESTORE();
570 		return;
571 	}
572 	callout_deactivate(&tp->t_timers->tt_rexmt);
573 	if ((inp->inp_flags & INP_DROPPED) != 0) {
574 		INP_WUNLOCK(inp);
575 		INP_INFO_RUNLOCK(&V_tcbinfo);
576 		CURVNET_RESTORE();
577 		return;
578 	}
579 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
580 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
581 	KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
582 		("%s: tp %p rexmt callout should be running", __func__, tp));
583 	tcp_free_sackholes(tp);
584 	/*
585 	 * Retransmission timer went off.  Message has not
586 	 * been acked within retransmit interval.  Back off
587 	 * to a longer retransmit interval and retransmit one segment.
588 	 */
589 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
590 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
591 		TCPSTAT_INC(tcps_timeoutdrop);
592 		in_pcbref(inp);
593 		INP_INFO_RUNLOCK(&V_tcbinfo);
594 		INP_WUNLOCK(inp);
595 		INP_INFO_WLOCK(&V_tcbinfo);
596 		INP_WLOCK(inp);
597 		if (in_pcbrele_wlocked(inp)) {
598 			INP_INFO_WUNLOCK(&V_tcbinfo);
599 			CURVNET_RESTORE();
600 			return;
601 		}
602 		if (inp->inp_flags & INP_DROPPED) {
603 			INP_WUNLOCK(inp);
604 			INP_INFO_WUNLOCK(&V_tcbinfo);
605 			CURVNET_RESTORE();
606 			return;
607 		}
608 
609 		tp = tcp_drop(tp, tp->t_softerror ?
610 			      tp->t_softerror : ETIMEDOUT);
611 		headlocked = 1;
612 		goto out;
613 	}
614 	INP_INFO_RUNLOCK(&V_tcbinfo);
615 	headlocked = 0;
616 	if (tp->t_state == TCPS_SYN_SENT) {
617 		/*
618 		 * If the SYN was retransmitted, indicate CWND to be
619 		 * limited to 1 segment in cc_conn_init().
620 		 */
621 		tp->snd_cwnd = 1;
622 	} else if (tp->t_rxtshift == 1) {
623 		/*
624 		 * first retransmit; record ssthresh and cwnd so they can
625 		 * be recovered if this turns out to be a "bad" retransmit.
626 		 * A retransmit is considered "bad" if an ACK for this
627 		 * segment is received within RTT/2 interval; the assumption
628 		 * here is that the ACK was already in flight.  See
629 		 * "On Estimating End-to-End Network Path Properties" by
630 		 * Allman and Paxson for more details.
631 		 */
632 		tp->snd_cwnd_prev = tp->snd_cwnd;
633 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
634 		tp->snd_recover_prev = tp->snd_recover;
635 		if (IN_FASTRECOVERY(tp->t_flags))
636 			tp->t_flags |= TF_WASFRECOVERY;
637 		else
638 			tp->t_flags &= ~TF_WASFRECOVERY;
639 		if (IN_CONGRECOVERY(tp->t_flags))
640 			tp->t_flags |= TF_WASCRECOVERY;
641 		else
642 			tp->t_flags &= ~TF_WASCRECOVERY;
643 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
644 		tp->t_flags |= TF_PREVVALID;
645 	} else
646 		tp->t_flags &= ~TF_PREVVALID;
647 	TCPSTAT_INC(tcps_rexmttimeo);
648 	if (tp->t_state == TCPS_SYN_SENT)
649 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
650 	else
651 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
652 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
653 		      tp->t_rttmin, TCPTV_REXMTMAX);
654 
655 	/*
656 	 * We enter the path for PLMTUD if connection is established or, if
657 	 * connection is FIN_WAIT_1 status, reason for the last is that if
658 	 * amount of data we send is very small, we could send it in couple of
659 	 * packets and process straight to FIN. In that case we won't catch
660 	 * ESTABLISHED state.
661 	 */
662 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
663 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
664 		int optlen;
665 #ifdef INET6
666 		int isipv6;
667 #endif
668 
669 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
670 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
671 		    (tp->t_rxtshift <= 2)) {
672 			/*
673 			 * Enter Path MTU Black-hole Detection mechanism:
674 			 * - Disable Path MTU Discovery (IP "DF" bit).
675 			 * - Reduce MTU to lower value than what we
676 			 *   negotiated with peer.
677 			 */
678 			/* Record that we may have found a black hole. */
679 			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
680 
681 			/* Keep track of previous MSS. */
682 			optlen = tp->t_maxopd - tp->t_maxseg;
683 			tp->t_pmtud_saved_maxopd = tp->t_maxopd;
684 
685 			/*
686 			 * Reduce the MSS to blackhole value or to the default
687 			 * in an attempt to retransmit.
688 			 */
689 #ifdef INET6
690 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
691 			if (isipv6 &&
692 			    tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
693 				/* Use the sysctl tuneable blackhole MSS. */
694 				tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
695 				V_tcp_pmtud_blackhole_activated++;
696 			} else if (isipv6) {
697 				/* Use the default MSS. */
698 				tp->t_maxopd = V_tcp_v6mssdflt;
699 				/*
700 				 * Disable Path MTU Discovery when we switch to
701 				 * minmss.
702 				 */
703 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
704 				V_tcp_pmtud_blackhole_activated_min_mss++;
705 			}
706 #endif
707 #if defined(INET6) && defined(INET)
708 			else
709 #endif
710 #ifdef INET
711 			if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
712 				/* Use the sysctl tuneable blackhole MSS. */
713 				tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
714 				V_tcp_pmtud_blackhole_activated++;
715 			} else {
716 				/* Use the default MSS. */
717 				tp->t_maxopd = V_tcp_mssdflt;
718 				/*
719 				 * Disable Path MTU Discovery when we switch to
720 				 * minmss.
721 				 */
722 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
723 				V_tcp_pmtud_blackhole_activated_min_mss++;
724 			}
725 #endif
726 			tp->t_maxseg = tp->t_maxopd - optlen;
727 			/*
728 			 * Reset the slow-start flight size
729 			 * as it may depend on the new MSS.
730 			 */
731 			if (CC_ALGO(tp)->conn_init != NULL)
732 				CC_ALGO(tp)->conn_init(tp->ccv);
733 		} else {
734 			/*
735 			 * If further retransmissions are still unsuccessful
736 			 * with a lowered MTU, maybe this isn't a blackhole and
737 			 * we restore the previous MSS and blackhole detection
738 			 * flags.
739 			 */
740 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
741 			    (tp->t_rxtshift > 4)) {
742 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
743 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
744 				optlen = tp->t_maxopd - tp->t_maxseg;
745 				tp->t_maxopd = tp->t_pmtud_saved_maxopd;
746 				tp->t_maxseg = tp->t_maxopd - optlen;
747 				V_tcp_pmtud_blackhole_failed++;
748 				/*
749 				 * Reset the slow-start flight size as it
750 				 * may depend on the new MSS.
751 				 */
752 				if (CC_ALGO(tp)->conn_init != NULL)
753 					CC_ALGO(tp)->conn_init(tp->ccv);
754 			}
755 		}
756 	}
757 
758 	/*
759 	 * Disable RFC1323 and SACK if we haven't got any response to
760 	 * our third SYN to work-around some broken terminal servers
761 	 * (most of which have hopefully been retired) that have bad VJ
762 	 * header compression code which trashes TCP segments containing
763 	 * unknown-to-them TCP options.
764 	 */
765 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
766 	    (tp->t_rxtshift == 3))
767 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
768 	/*
769 	 * If we backed off this far, our srtt estimate is probably bogus.
770 	 * Clobber it so we'll take the next rtt measurement as our srtt;
771 	 * move the current srtt into rttvar to keep the current
772 	 * retransmit times until then.
773 	 */
774 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
775 #ifdef INET6
776 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
777 			in6_losing(tp->t_inpcb);
778 #endif
779 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
780 		tp->t_srtt = 0;
781 	}
782 	tp->snd_nxt = tp->snd_una;
783 	tp->snd_recover = tp->snd_max;
784 	/*
785 	 * Force a segment to be sent.
786 	 */
787 	tp->t_flags |= TF_ACKNOW;
788 	/*
789 	 * If timing a segment in this window, stop the timer.
790 	 */
791 	tp->t_rtttime = 0;
792 
793 	cc_cong_signal(tp, NULL, CC_RTO);
794 
795 	(void) tcp_output(tp);
796 
797 out:
798 #ifdef TCPDEBUG
799 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
800 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
801 			  PRU_SLOWTIMO);
802 #endif
803 	if (tp != NULL)
804 		INP_WUNLOCK(inp);
805 	if (headlocked)
806 		INP_INFO_WUNLOCK(&V_tcbinfo);
807 	CURVNET_RESTORE();
808 }
809 
810 void
811 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
812 {
813 	struct callout *t_callout;
814 	timeout_t *f_callout;
815 	struct inpcb *inp = tp->t_inpcb;
816 	int cpu = inp_to_cpuid(inp);
817 	uint32_t f_reset;
818 
819 #ifdef TCP_OFFLOAD
820 	if (tp->t_flags & TF_TOE)
821 		return;
822 #endif
823 
824 	if (tp->t_timers->tt_flags & TT_STOPPED)
825 		return;
826 
827 	switch (timer_type) {
828 		case TT_DELACK:
829 			t_callout = &tp->t_timers->tt_delack;
830 			f_callout = tcp_timer_delack;
831 			f_reset = TT_DELACK_RST;
832 			break;
833 		case TT_REXMT:
834 			t_callout = &tp->t_timers->tt_rexmt;
835 			f_callout = tcp_timer_rexmt;
836 			f_reset = TT_REXMT_RST;
837 			break;
838 		case TT_PERSIST:
839 			t_callout = &tp->t_timers->tt_persist;
840 			f_callout = tcp_timer_persist;
841 			f_reset = TT_PERSIST_RST;
842 			break;
843 		case TT_KEEP:
844 			t_callout = &tp->t_timers->tt_keep;
845 			f_callout = tcp_timer_keep;
846 			f_reset = TT_KEEP_RST;
847 			break;
848 		case TT_2MSL:
849 			t_callout = &tp->t_timers->tt_2msl;
850 			f_callout = tcp_timer_2msl;
851 			f_reset = TT_2MSL_RST;
852 			break;
853 		default:
854 			panic("tp %p bad timer_type %#x", tp, timer_type);
855 		}
856 	if (delta == 0) {
857 		if ((tp->t_timers->tt_flags & timer_type) &&
858 		    callout_stop(t_callout) &&
859 		    (tp->t_timers->tt_flags & f_reset)) {
860 			tp->t_timers->tt_flags &= ~(timer_type | f_reset);
861 		}
862 	} else {
863 		if ((tp->t_timers->tt_flags & timer_type) == 0) {
864 			tp->t_timers->tt_flags |= (timer_type | f_reset);
865 			callout_reset_on(t_callout, delta, f_callout, tp, cpu);
866 		} else {
867 			/* Reset already running callout on the same CPU. */
868 			if (!callout_reset(t_callout, delta, f_callout, tp)) {
869 				/*
870 				 * Callout not cancelled, consider it as not
871 				 * properly restarted. */
872 				tp->t_timers->tt_flags &= ~f_reset;
873 			}
874 		}
875 	}
876 }
877 
878 int
879 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
880 {
881 	struct callout *t_callout;
882 
883 	switch (timer_type) {
884 		case TT_DELACK:
885 			t_callout = &tp->t_timers->tt_delack;
886 			break;
887 		case TT_REXMT:
888 			t_callout = &tp->t_timers->tt_rexmt;
889 			break;
890 		case TT_PERSIST:
891 			t_callout = &tp->t_timers->tt_persist;
892 			break;
893 		case TT_KEEP:
894 			t_callout = &tp->t_timers->tt_keep;
895 			break;
896 		case TT_2MSL:
897 			t_callout = &tp->t_timers->tt_2msl;
898 			break;
899 		default:
900 			panic("tp %p bad timer_type %#x", tp, timer_type);
901 		}
902 	return callout_active(t_callout);
903 }
904 
905 void
906 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
907 {
908 	struct callout *t_callout;
909 	timeout_t *f_callout;
910 	uint32_t f_reset;
911 
912 	tp->t_timers->tt_flags |= TT_STOPPED;
913 
914 	switch (timer_type) {
915 		case TT_DELACK:
916 			t_callout = &tp->t_timers->tt_delack;
917 			f_callout = tcp_timer_delack_discard;
918 			f_reset = TT_DELACK_RST;
919 			break;
920 		case TT_REXMT:
921 			t_callout = &tp->t_timers->tt_rexmt;
922 			f_callout = tcp_timer_rexmt_discard;
923 			f_reset = TT_REXMT_RST;
924 			break;
925 		case TT_PERSIST:
926 			t_callout = &tp->t_timers->tt_persist;
927 			f_callout = tcp_timer_persist_discard;
928 			f_reset = TT_PERSIST_RST;
929 			break;
930 		case TT_KEEP:
931 			t_callout = &tp->t_timers->tt_keep;
932 			f_callout = tcp_timer_keep_discard;
933 			f_reset = TT_KEEP_RST;
934 			break;
935 		case TT_2MSL:
936 			t_callout = &tp->t_timers->tt_2msl;
937 			f_callout = tcp_timer_2msl_discard;
938 			f_reset = TT_2MSL_RST;
939 			break;
940 		default:
941 			panic("tp %p bad timer_type %#x", tp, timer_type);
942 		}
943 
944 	if (tp->t_timers->tt_flags & timer_type) {
945 		if (callout_stop(t_callout) &&
946 		    (tp->t_timers->tt_flags & f_reset)) {
947 			tp->t_timers->tt_flags &= ~(timer_type | f_reset);
948 		} else {
949 			/*
950 			 * Can't stop the callout, defer tcpcb actual deletion
951 			 * to the last tcp timer discard callout.
952 			 * The TT_STOPPED flag will ensure that no tcp timer
953 			 * callouts can be restarted on our behalf, and
954 			 * past this point currently running callouts waiting
955 			 * on inp lock will return right away after the
956 			 * classical check for callout reset/stop events:
957 			 * callout_pending() || !callout_active()
958 			 */
959 			callout_reset(t_callout, 1, f_callout, tp);
960 		}
961 	}
962 }
963 
964 #define	ticks_to_msecs(t)	(1000*(t) / hz)
965 
966 void
967 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
968     struct xtcp_timer *xtimer)
969 {
970 	sbintime_t now;
971 
972 	bzero(xtimer, sizeof(*xtimer));
973 	if (timer == NULL)
974 		return;
975 	now = getsbinuptime();
976 	if (callout_active(&timer->tt_delack))
977 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
978 	if (callout_active(&timer->tt_rexmt))
979 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
980 	if (callout_active(&timer->tt_persist))
981 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
982 	if (callout_active(&timer->tt_keep))
983 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
984 	if (callout_active(&timer->tt_2msl))
985 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
986 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
987 }
988