xref: /freebsd/sys/netinet/tcp_timer.c (revision 93e779a26c651610ac6e7986d67ecc9ed2cadcbf)
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_kdtrace.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/in_rss.h>
63 #include <netinet/in_systm.h>
64 #ifdef INET6
65 #include <netinet6/in6_pcb.h>
66 #endif
67 #include <netinet/ip_var.h>
68 #include <netinet/tcp_fsm.h>
69 #include <netinet/tcp_timer.h>
70 #include <netinet/tcp_var.h>
71 #ifdef INET6
72 #include <netinet6/tcp6_var.h>
73 #endif
74 #include <netinet/tcpip.h>
75 #ifdef TCPDEBUG
76 #include <netinet/tcp_debug.h>
77 #endif
78 
79 int	tcp_keepinit;
80 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
81     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
82 
83 int	tcp_keepidle;
84 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
85     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
86 
87 int	tcp_keepintvl;
88 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
89     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
90 
91 int	tcp_delacktime;
92 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
93     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
94     "Time before a delayed ACK is sent");
95 
96 int	tcp_msl;
97 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
98     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
99 
100 int	tcp_rexmit_min;
101 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
102     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
103     "Minimum Retransmission Timeout");
104 
105 int	tcp_rexmit_slop;
106 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
107     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
108     "Retransmission Timer Slop");
109 
110 static int	always_keepalive = 1;
111 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
112     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
113 
114 int    tcp_fast_finwait2_recycle = 0;
115 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
116     &tcp_fast_finwait2_recycle, 0,
117     "Recycle closed FIN_WAIT_2 connections faster");
118 
119 int    tcp_finwait2_timeout;
120 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
121     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
122 
123 int	tcp_keepcnt = TCPTV_KEEPCNT;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
125     "Number of keepalive probes to send");
126 
127 	/* max idle probes */
128 int	tcp_maxpersistidle;
129 
130 static int	tcp_rexmit_drop_options = 0;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
132     &tcp_rexmit_drop_options, 0,
133     "Drop TCP options from 3rd and later retransmitted SYN");
134 
135 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
136 #define	V_tcp_pmtud_blackhole_detect	VNET(tcp_pmtud_blackhole_detect)
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
138     CTLFLAG_RW|CTLFLAG_VNET,
139     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
140     "Path MTU Discovery Black Hole Detection Enabled");
141 
142 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
143 #define	V_tcp_pmtud_blackhole_activated \
144     VNET(tcp_pmtud_blackhole_activated)
145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
146     CTLFLAG_RD|CTLFLAG_VNET,
147     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
148     "Path MTU Discovery Black Hole Detection, Activation Count");
149 
150 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
151 #define	V_tcp_pmtud_blackhole_activated_min_mss \
152     VNET(tcp_pmtud_blackhole_activated_min_mss)
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
154     CTLFLAG_RD|CTLFLAG_VNET,
155     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
156     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
157 
158 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
159 #define	V_tcp_pmtud_blackhole_failed	VNET(tcp_pmtud_blackhole_failed)
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
161     CTLFLAG_RD|CTLFLAG_VNET,
162     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
163     "Path MTU Discovery Black Hole Detection, Failure Count");
164 
165 #ifdef INET
166 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
167 #define	V_tcp_pmtud_blackhole_mss	VNET(tcp_pmtud_blackhole_mss)
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
169     CTLFLAG_RW|CTLFLAG_VNET,
170     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
171     "Path MTU Discovery Black Hole Detection lowered MSS");
172 #endif
173 
174 #ifdef INET6
175 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
176 #define	V_tcp_v6pmtud_blackhole_mss	VNET(tcp_v6pmtud_blackhole_mss)
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
178     CTLFLAG_RW|CTLFLAG_VNET,
179     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
180     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
181 #endif
182 
183 #ifdef	RSS
184 static int	per_cpu_timers = 1;
185 #else
186 static int	per_cpu_timers = 0;
187 #endif
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
189     &per_cpu_timers , 0, "run tcp timers on all cpus");
190 
191 #if 0
192 #define	INP_CPU(inp)	(per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
193 		((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
194 #endif
195 
196 /*
197  * Map the given inp to a CPU id.
198  *
199  * This queries RSS if it's compiled in, else it defaults to the current
200  * CPU ID.
201  */
202 static inline int
203 inp_to_cpuid(struct inpcb *inp)
204 {
205 	u_int cpuid;
206 
207 #ifdef	RSS
208 	if (per_cpu_timers) {
209 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
210 		if (cpuid == NETISR_CPUID_NONE)
211 			return (curcpu);	/* XXX */
212 		else
213 			return (cpuid);
214 	}
215 #else
216 	/* Legacy, pre-RSS behaviour */
217 	if (per_cpu_timers) {
218 		/*
219 		 * We don't have a flowid -> cpuid mapping, so cheat and
220 		 * just map unknown cpuids to curcpu.  Not the best, but
221 		 * apparently better than defaulting to swi 0.
222 		 */
223 		cpuid = inp->inp_flowid % (mp_maxid + 1);
224 		if (! CPU_ABSENT(cpuid))
225 			return (cpuid);
226 		return (curcpu);
227 	}
228 #endif
229 	/* Default for RSS and non-RSS - cpuid 0 */
230 	else {
231 		return (0);
232 	}
233 }
234 
235 /*
236  * Tcp protocol timeout routine called every 500 ms.
237  * Updates timestamps used for TCP
238  * causes finite state machine actions if timers expire.
239  */
240 void
241 tcp_slowtimo(void)
242 {
243 	VNET_ITERATOR_DECL(vnet_iter);
244 
245 	VNET_LIST_RLOCK_NOSLEEP();
246 	VNET_FOREACH(vnet_iter) {
247 		CURVNET_SET(vnet_iter);
248 		(void) tcp_tw_2msl_scan(0);
249 		CURVNET_RESTORE();
250 	}
251 	VNET_LIST_RUNLOCK_NOSLEEP();
252 }
253 
254 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
255     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
256 
257 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
258     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
259 
260 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
261 
262 /*
263  * TCP timer processing.
264  */
265 
266 void
267 tcp_timer_delack(void *xtp)
268 {
269 	struct tcpcb *tp = xtp;
270 	struct inpcb *inp;
271 	CURVNET_SET(tp->t_vnet);
272 
273 	inp = tp->t_inpcb;
274 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
275 	INP_WLOCK(inp);
276 	if (callout_pending(&tp->t_timers->tt_delack) ||
277 	    !callout_active(&tp->t_timers->tt_delack)) {
278 		INP_WUNLOCK(inp);
279 		CURVNET_RESTORE();
280 		return;
281 	}
282 	callout_deactivate(&tp->t_timers->tt_delack);
283 	if ((inp->inp_flags & INP_DROPPED) != 0) {
284 		INP_WUNLOCK(inp);
285 		CURVNET_RESTORE();
286 		return;
287 	}
288 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
289 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
290 	KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
291 		("%s: tp %p delack callout should be running", __func__, tp));
292 
293 	tp->t_flags |= TF_ACKNOW;
294 	TCPSTAT_INC(tcps_delack);
295 	(void) tcp_output(tp);
296 	INP_WUNLOCK(inp);
297 	CURVNET_RESTORE();
298 }
299 
300 void
301 tcp_timer_2msl(void *xtp)
302 {
303 	struct tcpcb *tp = xtp;
304 	struct inpcb *inp;
305 	CURVNET_SET(tp->t_vnet);
306 #ifdef TCPDEBUG
307 	int ostate;
308 
309 	ostate = tp->t_state;
310 #endif
311 	INP_INFO_RLOCK(&V_tcbinfo);
312 	inp = tp->t_inpcb;
313 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
314 	INP_WLOCK(inp);
315 	tcp_free_sackholes(tp);
316 	if (callout_pending(&tp->t_timers->tt_2msl) ||
317 	    !callout_active(&tp->t_timers->tt_2msl)) {
318 		INP_WUNLOCK(tp->t_inpcb);
319 		INP_INFO_RUNLOCK(&V_tcbinfo);
320 		CURVNET_RESTORE();
321 		return;
322 	}
323 	callout_deactivate(&tp->t_timers->tt_2msl);
324 	if ((inp->inp_flags & INP_DROPPED) != 0) {
325 		INP_WUNLOCK(inp);
326 		INP_INFO_RUNLOCK(&V_tcbinfo);
327 		CURVNET_RESTORE();
328 		return;
329 	}
330 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
331 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
332 	KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
333 		("%s: tp %p 2msl callout should be running", __func__, tp));
334 	/*
335 	 * 2 MSL timeout in shutdown went off.  If we're closed but
336 	 * still waiting for peer to close and connection has been idle
337 	 * too long delete connection control block.  Otherwise, check
338 	 * again in a bit.
339 	 *
340 	 * If in TIME_WAIT state just ignore as this timeout is handled in
341 	 * tcp_tw_2msl_scan().
342 	 *
343 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
344 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
345 	 * Ignore fact that there were recent incoming segments.
346 	 */
347 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
348 		INP_WUNLOCK(inp);
349 		INP_INFO_RUNLOCK(&V_tcbinfo);
350 		CURVNET_RESTORE();
351 		return;
352 	}
353 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
354 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
355 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
356 		TCPSTAT_INC(tcps_finwait2_drops);
357 		tp = tcp_close(tp);
358 	} else {
359 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
360 			if (!callout_reset(&tp->t_timers->tt_2msl,
361 			   TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
362 				tp->t_timers->tt_flags &= ~TT_2MSL_RST;
363 			}
364 		} else
365 		       tp = tcp_close(tp);
366        }
367 
368 #ifdef TCPDEBUG
369 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
370 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
371 			  PRU_SLOWTIMO);
372 #endif
373 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
374 
375 	if (tp != NULL)
376 		INP_WUNLOCK(inp);
377 	INP_INFO_RUNLOCK(&V_tcbinfo);
378 	CURVNET_RESTORE();
379 }
380 
381 void
382 tcp_timer_keep(void *xtp)
383 {
384 	struct tcpcb *tp = xtp;
385 	struct tcptemp *t_template;
386 	struct inpcb *inp;
387 	CURVNET_SET(tp->t_vnet);
388 #ifdef TCPDEBUG
389 	int ostate;
390 
391 	ostate = tp->t_state;
392 #endif
393 	INP_INFO_RLOCK(&V_tcbinfo);
394 	inp = tp->t_inpcb;
395 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
396 	INP_WLOCK(inp);
397 	if (callout_pending(&tp->t_timers->tt_keep) ||
398 	    !callout_active(&tp->t_timers->tt_keep)) {
399 		INP_WUNLOCK(inp);
400 		INP_INFO_RUNLOCK(&V_tcbinfo);
401 		CURVNET_RESTORE();
402 		return;
403 	}
404 	callout_deactivate(&tp->t_timers->tt_keep);
405 	if ((inp->inp_flags & INP_DROPPED) != 0) {
406 		INP_WUNLOCK(inp);
407 		INP_INFO_RUNLOCK(&V_tcbinfo);
408 		CURVNET_RESTORE();
409 		return;
410 	}
411 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
412 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
413 	KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
414 		("%s: tp %p keep callout should be running", __func__, tp));
415 	/*
416 	 * Keep-alive timer went off; send something
417 	 * or drop connection if idle for too long.
418 	 */
419 	TCPSTAT_INC(tcps_keeptimeo);
420 	if (tp->t_state < TCPS_ESTABLISHED)
421 		goto dropit;
422 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
423 	    tp->t_state <= TCPS_CLOSING) {
424 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
425 			goto dropit;
426 		/*
427 		 * Send a packet designed to force a response
428 		 * if the peer is up and reachable:
429 		 * either an ACK if the connection is still alive,
430 		 * or an RST if the peer has closed the connection
431 		 * due to timeout or reboot.
432 		 * Using sequence number tp->snd_una-1
433 		 * causes the transmitted zero-length segment
434 		 * to lie outside the receive window;
435 		 * by the protocol spec, this requires the
436 		 * correspondent TCP to respond.
437 		 */
438 		TCPSTAT_INC(tcps_keepprobe);
439 		t_template = tcpip_maketemplate(inp);
440 		if (t_template) {
441 			tcp_respond(tp, t_template->tt_ipgen,
442 				    &t_template->tt_t, (struct mbuf *)NULL,
443 				    tp->rcv_nxt, tp->snd_una - 1, 0);
444 			free(t_template, M_TEMP);
445 		}
446 		if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
447 		    tcp_timer_keep, tp)) {
448 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
449 		}
450 	} else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
451 		    tcp_timer_keep, tp)) {
452 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
453 		}
454 
455 #ifdef TCPDEBUG
456 	if (inp->inp_socket->so_options & SO_DEBUG)
457 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
458 			  PRU_SLOWTIMO);
459 #endif
460 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
461 	INP_WUNLOCK(inp);
462 	INP_INFO_RUNLOCK(&V_tcbinfo);
463 	CURVNET_RESTORE();
464 	return;
465 
466 dropit:
467 	TCPSTAT_INC(tcps_keepdrops);
468 	tp = tcp_drop(tp, ETIMEDOUT);
469 
470 #ifdef TCPDEBUG
471 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
472 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
473 			  PRU_SLOWTIMO);
474 #endif
475 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
476 	if (tp != NULL)
477 		INP_WUNLOCK(tp->t_inpcb);
478 	INP_INFO_RUNLOCK(&V_tcbinfo);
479 	CURVNET_RESTORE();
480 }
481 
482 void
483 tcp_timer_persist(void *xtp)
484 {
485 	struct tcpcb *tp = xtp;
486 	struct inpcb *inp;
487 	CURVNET_SET(tp->t_vnet);
488 #ifdef TCPDEBUG
489 	int ostate;
490 
491 	ostate = tp->t_state;
492 #endif
493 	INP_INFO_RLOCK(&V_tcbinfo);
494 	inp = tp->t_inpcb;
495 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
496 	INP_WLOCK(inp);
497 	if (callout_pending(&tp->t_timers->tt_persist) ||
498 	    !callout_active(&tp->t_timers->tt_persist)) {
499 		INP_WUNLOCK(inp);
500 		INP_INFO_RUNLOCK(&V_tcbinfo);
501 		CURVNET_RESTORE();
502 		return;
503 	}
504 	callout_deactivate(&tp->t_timers->tt_persist);
505 	if ((inp->inp_flags & INP_DROPPED) != 0) {
506 		INP_WUNLOCK(inp);
507 		INP_INFO_RUNLOCK(&V_tcbinfo);
508 		CURVNET_RESTORE();
509 		return;
510 	}
511 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
512 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
513 	KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
514 		("%s: tp %p persist callout should be running", __func__, tp));
515 	/*
516 	 * Persistance timer into zero window.
517 	 * Force a byte to be output, if possible.
518 	 */
519 	TCPSTAT_INC(tcps_persisttimeo);
520 	/*
521 	 * Hack: if the peer is dead/unreachable, we do not
522 	 * time out if the window is closed.  After a full
523 	 * backoff, drop the connection if the idle time
524 	 * (no responses to probes) reaches the maximum
525 	 * backoff that we would use if retransmitting.
526 	 */
527 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
528 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
529 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
530 		TCPSTAT_INC(tcps_persistdrop);
531 		tp = tcp_drop(tp, ETIMEDOUT);
532 		goto out;
533 	}
534 	/*
535 	 * If the user has closed the socket then drop a persisting
536 	 * connection after a much reduced timeout.
537 	 */
538 	if (tp->t_state > TCPS_CLOSE_WAIT &&
539 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
540 		TCPSTAT_INC(tcps_persistdrop);
541 		tp = tcp_drop(tp, ETIMEDOUT);
542 		goto out;
543 	}
544 	tcp_setpersist(tp);
545 	tp->t_flags |= TF_FORCEDATA;
546 	(void) tcp_output(tp);
547 	tp->t_flags &= ~TF_FORCEDATA;
548 
549 out:
550 #ifdef TCPDEBUG
551 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
552 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
553 #endif
554 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
555 	if (tp != NULL)
556 		INP_WUNLOCK(inp);
557 	INP_INFO_RUNLOCK(&V_tcbinfo);
558 	CURVNET_RESTORE();
559 }
560 
561 void
562 tcp_timer_rexmt(void * xtp)
563 {
564 	struct tcpcb *tp = xtp;
565 	CURVNET_SET(tp->t_vnet);
566 	int rexmt;
567 	int headlocked;
568 	struct inpcb *inp;
569 #ifdef TCPDEBUG
570 	int ostate;
571 
572 	ostate = tp->t_state;
573 #endif
574 
575 	INP_INFO_RLOCK(&V_tcbinfo);
576 	inp = tp->t_inpcb;
577 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
578 	INP_WLOCK(inp);
579 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
580 	    !callout_active(&tp->t_timers->tt_rexmt)) {
581 		INP_WUNLOCK(inp);
582 		INP_INFO_RUNLOCK(&V_tcbinfo);
583 		CURVNET_RESTORE();
584 		return;
585 	}
586 	callout_deactivate(&tp->t_timers->tt_rexmt);
587 	if ((inp->inp_flags & INP_DROPPED) != 0) {
588 		INP_WUNLOCK(inp);
589 		INP_INFO_RUNLOCK(&V_tcbinfo);
590 		CURVNET_RESTORE();
591 		return;
592 	}
593 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
594 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
595 	KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
596 		("%s: tp %p rexmt callout should be running", __func__, tp));
597 	tcp_free_sackholes(tp);
598 	/*
599 	 * Retransmission timer went off.  Message has not
600 	 * been acked within retransmit interval.  Back off
601 	 * to a longer retransmit interval and retransmit one segment.
602 	 */
603 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
604 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
605 		TCPSTAT_INC(tcps_timeoutdrop);
606 
607 		tp = tcp_drop(tp, tp->t_softerror ?
608 			      tp->t_softerror : ETIMEDOUT);
609 		headlocked = 1;
610 		goto out;
611 	}
612 	INP_INFO_RUNLOCK(&V_tcbinfo);
613 	headlocked = 0;
614 	if (tp->t_state == TCPS_SYN_SENT) {
615 		/*
616 		 * If the SYN was retransmitted, indicate CWND to be
617 		 * limited to 1 segment in cc_conn_init().
618 		 */
619 		tp->snd_cwnd = 1;
620 	} else if (tp->t_rxtshift == 1) {
621 		/*
622 		 * first retransmit; record ssthresh and cwnd so they can
623 		 * be recovered if this turns out to be a "bad" retransmit.
624 		 * A retransmit is considered "bad" if an ACK for this
625 		 * segment is received within RTT/2 interval; the assumption
626 		 * here is that the ACK was already in flight.  See
627 		 * "On Estimating End-to-End Network Path Properties" by
628 		 * Allman and Paxson for more details.
629 		 */
630 		tp->snd_cwnd_prev = tp->snd_cwnd;
631 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
632 		tp->snd_recover_prev = tp->snd_recover;
633 		if (IN_FASTRECOVERY(tp->t_flags))
634 			tp->t_flags |= TF_WASFRECOVERY;
635 		else
636 			tp->t_flags &= ~TF_WASFRECOVERY;
637 		if (IN_CONGRECOVERY(tp->t_flags))
638 			tp->t_flags |= TF_WASCRECOVERY;
639 		else
640 			tp->t_flags &= ~TF_WASCRECOVERY;
641 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
642 		tp->t_flags |= TF_PREVVALID;
643 	} else
644 		tp->t_flags &= ~TF_PREVVALID;
645 	TCPSTAT_INC(tcps_rexmttimeo);
646 	if (tp->t_state == TCPS_SYN_SENT)
647 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
648 	else
649 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
650 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
651 		      tp->t_rttmin, TCPTV_REXMTMAX);
652 
653 	/*
654 	 * We enter the path for PLMTUD if connection is established or, if
655 	 * connection is FIN_WAIT_1 status, reason for the last is that if
656 	 * amount of data we send is very small, we could send it in couple of
657 	 * packets and process straight to FIN. In that case we won't catch
658 	 * ESTABLISHED state.
659 	 */
660 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
661 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
662 		int optlen;
663 #ifdef INET6
664 		int isipv6;
665 #endif
666 
667 		/*
668 		 * Idea here is that at each stage of mtu probe (usually, 1448
669 		 * -> 1188 -> 524) should be given 2 chances to recover before
670 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
671 		 *  take care of that.
672 		 */
673 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
674 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
675 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
676 			/*
677 			 * Enter Path MTU Black-hole Detection mechanism:
678 			 * - Disable Path MTU Discovery (IP "DF" bit).
679 			 * - Reduce MTU to lower value than what we
680 			 *   negotiated with peer.
681 			 */
682 			/* Record that we may have found a black hole. */
683 			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
684 
685 			/* Keep track of previous MSS. */
686 			optlen = tp->t_maxopd - tp->t_maxseg;
687 			tp->t_pmtud_saved_maxopd = tp->t_maxopd;
688 
689 			/*
690 			 * Reduce the MSS to blackhole value or to the default
691 			 * in an attempt to retransmit.
692 			 */
693 #ifdef INET6
694 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
695 			if (isipv6 &&
696 			    tp->t_maxopd > V_tcp_v6pmtud_blackhole_mss) {
697 				/* Use the sysctl tuneable blackhole MSS. */
698 				tp->t_maxopd = V_tcp_v6pmtud_blackhole_mss;
699 				V_tcp_pmtud_blackhole_activated++;
700 			} else if (isipv6) {
701 				/* Use the default MSS. */
702 				tp->t_maxopd = V_tcp_v6mssdflt;
703 				/*
704 				 * Disable Path MTU Discovery when we switch to
705 				 * minmss.
706 				 */
707 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
708 				V_tcp_pmtud_blackhole_activated_min_mss++;
709 			}
710 #endif
711 #if defined(INET6) && defined(INET)
712 			else
713 #endif
714 #ifdef INET
715 			if (tp->t_maxopd > V_tcp_pmtud_blackhole_mss) {
716 				/* Use the sysctl tuneable blackhole MSS. */
717 				tp->t_maxopd = V_tcp_pmtud_blackhole_mss;
718 				V_tcp_pmtud_blackhole_activated++;
719 			} else {
720 				/* Use the default MSS. */
721 				tp->t_maxopd = V_tcp_mssdflt;
722 				/*
723 				 * Disable Path MTU Discovery when we switch to
724 				 * minmss.
725 				 */
726 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
727 				V_tcp_pmtud_blackhole_activated_min_mss++;
728 			}
729 #endif
730 			tp->t_maxseg = tp->t_maxopd - optlen;
731 			/*
732 			 * Reset the slow-start flight size
733 			 * as it may depend on the new MSS.
734 			 */
735 			if (CC_ALGO(tp)->conn_init != NULL)
736 				CC_ALGO(tp)->conn_init(tp->ccv);
737 		} else {
738 			/*
739 			 * If further retransmissions are still unsuccessful
740 			 * with a lowered MTU, maybe this isn't a blackhole and
741 			 * we restore the previous MSS and blackhole detection
742 			 * flags.
743 			 * The limit '6' is determined by giving each probe
744 			 * stage (1448, 1188, 524) 2 chances to recover.
745 			 */
746 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
747 			    (tp->t_rxtshift > 6)) {
748 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
749 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
750 				optlen = tp->t_maxopd - tp->t_maxseg;
751 				tp->t_maxopd = tp->t_pmtud_saved_maxopd;
752 				tp->t_maxseg = tp->t_maxopd - optlen;
753 				V_tcp_pmtud_blackhole_failed++;
754 				/*
755 				 * Reset the slow-start flight size as it
756 				 * may depend on the new MSS.
757 				 */
758 				if (CC_ALGO(tp)->conn_init != NULL)
759 					CC_ALGO(tp)->conn_init(tp->ccv);
760 			}
761 		}
762 	}
763 
764 	/*
765 	 * Disable RFC1323 and SACK if we haven't got any response to
766 	 * our third SYN to work-around some broken terminal servers
767 	 * (most of which have hopefully been retired) that have bad VJ
768 	 * header compression code which trashes TCP segments containing
769 	 * unknown-to-them TCP options.
770 	 */
771 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
772 	    (tp->t_rxtshift == 3))
773 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
774 	/*
775 	 * If we backed off this far, our srtt estimate is probably bogus.
776 	 * Clobber it so we'll take the next rtt measurement as our srtt;
777 	 * move the current srtt into rttvar to keep the current
778 	 * retransmit times until then.
779 	 */
780 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
781 #ifdef INET6
782 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
783 			in6_losing(tp->t_inpcb);
784 #endif
785 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
786 		tp->t_srtt = 0;
787 	}
788 	tp->snd_nxt = tp->snd_una;
789 	tp->snd_recover = tp->snd_max;
790 	/*
791 	 * Force a segment to be sent.
792 	 */
793 	tp->t_flags |= TF_ACKNOW;
794 	/*
795 	 * If timing a segment in this window, stop the timer.
796 	 */
797 	tp->t_rtttime = 0;
798 
799 	cc_cong_signal(tp, NULL, CC_RTO);
800 
801 	(void) tcp_output(tp);
802 
803 out:
804 #ifdef TCPDEBUG
805 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
806 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
807 			  PRU_SLOWTIMO);
808 #endif
809 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
810 	if (tp != NULL)
811 		INP_WUNLOCK(inp);
812 	if (headlocked)
813 		INP_INFO_RUNLOCK(&V_tcbinfo);
814 	CURVNET_RESTORE();
815 }
816 
817 void
818 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
819 {
820 	struct callout *t_callout;
821 	timeout_t *f_callout;
822 	struct inpcb *inp = tp->t_inpcb;
823 	int cpu = inp_to_cpuid(inp);
824 	uint32_t f_reset;
825 
826 #ifdef TCP_OFFLOAD
827 	if (tp->t_flags & TF_TOE)
828 		return;
829 #endif
830 
831 	if (tp->t_timers->tt_flags & TT_STOPPED)
832 		return;
833 
834 	switch (timer_type) {
835 		case TT_DELACK:
836 			t_callout = &tp->t_timers->tt_delack;
837 			f_callout = tcp_timer_delack;
838 			f_reset = TT_DELACK_RST;
839 			break;
840 		case TT_REXMT:
841 			t_callout = &tp->t_timers->tt_rexmt;
842 			f_callout = tcp_timer_rexmt;
843 			f_reset = TT_REXMT_RST;
844 			break;
845 		case TT_PERSIST:
846 			t_callout = &tp->t_timers->tt_persist;
847 			f_callout = tcp_timer_persist;
848 			f_reset = TT_PERSIST_RST;
849 			break;
850 		case TT_KEEP:
851 			t_callout = &tp->t_timers->tt_keep;
852 			f_callout = tcp_timer_keep;
853 			f_reset = TT_KEEP_RST;
854 			break;
855 		case TT_2MSL:
856 			t_callout = &tp->t_timers->tt_2msl;
857 			f_callout = tcp_timer_2msl;
858 			f_reset = TT_2MSL_RST;
859 			break;
860 		default:
861 			panic("tp %p bad timer_type %#x", tp, timer_type);
862 		}
863 	if (delta == 0) {
864 		if ((tp->t_timers->tt_flags & timer_type) &&
865 		    (callout_stop(t_callout) > 0) &&
866 		    (tp->t_timers->tt_flags & f_reset)) {
867 			tp->t_timers->tt_flags &= ~(timer_type | f_reset);
868 		}
869 	} else {
870 		if ((tp->t_timers->tt_flags & timer_type) == 0) {
871 			tp->t_timers->tt_flags |= (timer_type | f_reset);
872 			callout_reset_on(t_callout, delta, f_callout, tp, cpu);
873 		} else {
874 			/* Reset already running callout on the same CPU. */
875 			if (!callout_reset(t_callout, delta, f_callout, tp)) {
876 				/*
877 				 * Callout not cancelled, consider it as not
878 				 * properly restarted. */
879 				tp->t_timers->tt_flags &= ~f_reset;
880 			}
881 		}
882 	}
883 }
884 
885 int
886 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
887 {
888 	struct callout *t_callout;
889 
890 	switch (timer_type) {
891 		case TT_DELACK:
892 			t_callout = &tp->t_timers->tt_delack;
893 			break;
894 		case TT_REXMT:
895 			t_callout = &tp->t_timers->tt_rexmt;
896 			break;
897 		case TT_PERSIST:
898 			t_callout = &tp->t_timers->tt_persist;
899 			break;
900 		case TT_KEEP:
901 			t_callout = &tp->t_timers->tt_keep;
902 			break;
903 		case TT_2MSL:
904 			t_callout = &tp->t_timers->tt_2msl;
905 			break;
906 		default:
907 			panic("tp %p bad timer_type %#x", tp, timer_type);
908 		}
909 	return callout_active(t_callout);
910 }
911 
912 void
913 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
914 {
915 	struct callout *t_callout;
916 	timeout_t *f_callout;
917 	uint32_t f_reset;
918 
919 	tp->t_timers->tt_flags |= TT_STOPPED;
920 
921 	switch (timer_type) {
922 		case TT_DELACK:
923 			t_callout = &tp->t_timers->tt_delack;
924 			f_callout = tcp_timer_delack_discard;
925 			f_reset = TT_DELACK_RST;
926 			break;
927 		case TT_REXMT:
928 			t_callout = &tp->t_timers->tt_rexmt;
929 			f_callout = tcp_timer_rexmt_discard;
930 			f_reset = TT_REXMT_RST;
931 			break;
932 		case TT_PERSIST:
933 			t_callout = &tp->t_timers->tt_persist;
934 			f_callout = tcp_timer_persist_discard;
935 			f_reset = TT_PERSIST_RST;
936 			break;
937 		case TT_KEEP:
938 			t_callout = &tp->t_timers->tt_keep;
939 			f_callout = tcp_timer_keep_discard;
940 			f_reset = TT_KEEP_RST;
941 			break;
942 		case TT_2MSL:
943 			t_callout = &tp->t_timers->tt_2msl;
944 			f_callout = tcp_timer_2msl_discard;
945 			f_reset = TT_2MSL_RST;
946 			break;
947 		default:
948 			panic("tp %p bad timer_type %#x", tp, timer_type);
949 		}
950 
951 	if (tp->t_timers->tt_flags & timer_type) {
952 		if ((callout_stop(t_callout) > 0) &&
953 		    (tp->t_timers->tt_flags & f_reset)) {
954 			tp->t_timers->tt_flags &= ~(timer_type | f_reset);
955 		} else {
956 			/*
957 			 * Can't stop the callout, defer tcpcb actual deletion
958 			 * to the last tcp timer discard callout.
959 			 * The TT_STOPPED flag will ensure that no tcp timer
960 			 * callouts can be restarted on our behalf, and
961 			 * past this point currently running callouts waiting
962 			 * on inp lock will return right away after the
963 			 * classical check for callout reset/stop events:
964 			 * callout_pending() || !callout_active()
965 			 */
966 			callout_reset(t_callout, 1, f_callout, tp);
967 		}
968 	}
969 }
970 
971 #define	ticks_to_msecs(t)	(1000*(t) / hz)
972 
973 void
974 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
975     struct xtcp_timer *xtimer)
976 {
977 	sbintime_t now;
978 
979 	bzero(xtimer, sizeof(*xtimer));
980 	if (timer == NULL)
981 		return;
982 	now = getsbinuptime();
983 	if (callout_active(&timer->tt_delack))
984 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
985 	if (callout_active(&timer->tt_rexmt))
986 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
987 	if (callout_active(&timer->tt_persist))
988 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
989 	if (callout_active(&timer->tt_keep))
990 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
991 	if (callout_active(&timer->tt_2msl))
992 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
993 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
994 }
995