xref: /freebsd/sys/netinet/tcp_timer.c (revision 715d1396d6233158b94e85414b57520ee3e536cc)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_timer.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
40 #include "opt_rss.h"
41 
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/protosw.h>
48 #include <sys/smp.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/systm.h>
53 
54 #include <net/if.h>
55 #include <net/route.h>
56 #include <net/rss_config.h>
57 #include <net/vnet.h>
58 #include <net/netisr.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_kdtrace.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_rss.h>
64 #include <netinet/in_systm.h>
65 #ifdef INET6
66 #include <netinet6/in6_pcb.h>
67 #endif
68 #include <netinet/ip_var.h>
69 #include <netinet/tcp.h>
70 #include <netinet/tcp_fsm.h>
71 #include <netinet/tcp_log_buf.h>
72 #include <netinet/tcp_timer.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet/cc/cc.h>
75 #ifdef INET6
76 #include <netinet6/tcp6_var.h>
77 #endif
78 #include <netinet/tcpip.h>
79 #ifdef TCPDEBUG
80 #include <netinet/tcp_debug.h>
81 #endif
82 
83 int    tcp_persmin;
84 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
85     &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
86 
87 int    tcp_persmax;
88 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
89     &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
90 
91 int	tcp_keepinit;
92 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
93     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
94 
95 int	tcp_keepidle;
96 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
97     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
98 
99 int	tcp_keepintvl;
100 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
101     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
102 
103 int	tcp_delacktime;
104 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
105     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
106     "Time before a delayed ACK is sent");
107 
108 int	tcp_msl;
109 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
110     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
111 
112 int	tcp_rexmit_min;
113 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
114     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
115     "Minimum Retransmission Timeout");
116 
117 int	tcp_rexmit_slop;
118 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
119     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
120     "Retransmission Timer Slop");
121 
122 int	tcp_always_keepalive = 1;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
124     &tcp_always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
125 
126 int    tcp_fast_finwait2_recycle = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
128     &tcp_fast_finwait2_recycle, 0,
129     "Recycle closed FIN_WAIT_2 connections faster");
130 
131 int    tcp_finwait2_timeout;
132 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
133     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
134 
135 int	tcp_keepcnt = TCPTV_KEEPCNT;
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
137     "Number of keepalive probes to send");
138 
139 	/* max idle probes */
140 int	tcp_maxpersistidle;
141 
142 static int	tcp_rexmit_drop_options = 0;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
144     &tcp_rexmit_drop_options, 0,
145     "Drop TCP options from 3rd and later retransmitted SYN");
146 
147 VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
149     CTLFLAG_RW|CTLFLAG_VNET,
150     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
151     "Path MTU Discovery Black Hole Detection Enabled");
152 
153 #ifdef INET
154 VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
156     CTLFLAG_RW|CTLFLAG_VNET,
157     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
158     "Path MTU Discovery Black Hole Detection lowered MSS");
159 #endif
160 
161 #ifdef INET6
162 VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
164     CTLFLAG_RW|CTLFLAG_VNET,
165     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
166     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
167 #endif
168 
169 #ifdef	RSS
170 static int	per_cpu_timers = 1;
171 #else
172 static int	per_cpu_timers = 0;
173 #endif
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
175     &per_cpu_timers , 0, "run tcp timers on all cpus");
176 
177 #if 0
178 #define	INP_CPU(inp)	(per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
179 		((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
180 #endif
181 
182 /*
183  * Map the given inp to a CPU id.
184  *
185  * This queries RSS if it's compiled in, else it defaults to the current
186  * CPU ID.
187  */
188 static inline int
189 inp_to_cpuid(struct inpcb *inp)
190 {
191 	u_int cpuid;
192 
193 #ifdef	RSS
194 	if (per_cpu_timers) {
195 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
196 		if (cpuid == NETISR_CPUID_NONE)
197 			return (curcpu);	/* XXX */
198 		else
199 			return (cpuid);
200 	}
201 #else
202 	/* Legacy, pre-RSS behaviour */
203 	if (per_cpu_timers) {
204 		/*
205 		 * We don't have a flowid -> cpuid mapping, so cheat and
206 		 * just map unknown cpuids to curcpu.  Not the best, but
207 		 * apparently better than defaulting to swi 0.
208 		 */
209 		cpuid = inp->inp_flowid % (mp_maxid + 1);
210 		if (! CPU_ABSENT(cpuid))
211 			return (cpuid);
212 		return (curcpu);
213 	}
214 #endif
215 	/* Default for RSS and non-RSS - cpuid 0 */
216 	else {
217 		return (0);
218 	}
219 }
220 
221 /*
222  * Tcp protocol timeout routine called every 500 ms.
223  * Updates timestamps used for TCP
224  * causes finite state machine actions if timers expire.
225  */
226 void
227 tcp_slowtimo(void)
228 {
229 	VNET_ITERATOR_DECL(vnet_iter);
230 
231 	VNET_LIST_RLOCK_NOSLEEP();
232 	VNET_FOREACH(vnet_iter) {
233 		CURVNET_SET(vnet_iter);
234 		(void) tcp_tw_2msl_scan(0);
235 		CURVNET_RESTORE();
236 	}
237 	VNET_LIST_RUNLOCK_NOSLEEP();
238 }
239 
240 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
241     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
242 
243 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
244     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
245 
246 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
247 
248 /*
249  * TCP timer processing.
250  */
251 
252 void
253 tcp_timer_delack(void *xtp)
254 {
255 	struct tcpcb *tp = xtp;
256 	struct inpcb *inp;
257 	CURVNET_SET(tp->t_vnet);
258 
259 	inp = tp->t_inpcb;
260 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
261 	INP_WLOCK(inp);
262 	if (callout_pending(&tp->t_timers->tt_delack) ||
263 	    !callout_active(&tp->t_timers->tt_delack)) {
264 		INP_WUNLOCK(inp);
265 		CURVNET_RESTORE();
266 		return;
267 	}
268 	callout_deactivate(&tp->t_timers->tt_delack);
269 	if ((inp->inp_flags & INP_DROPPED) != 0) {
270 		INP_WUNLOCK(inp);
271 		CURVNET_RESTORE();
272 		return;
273 	}
274 	tp->t_flags |= TF_ACKNOW;
275 	TCPSTAT_INC(tcps_delack);
276 	(void) tp->t_fb->tfb_tcp_output(tp);
277 	INP_WUNLOCK(inp);
278 	CURVNET_RESTORE();
279 }
280 
281 /*
282  * When a timer wants to remove a TCB it must
283  * hold the INP_INFO_RLOCK(). The timer function
284  * should only have grabbed the INP_WLOCK() when
285  * it entered. To safely switch to holding both the
286  * INP_INFO_RLOCK() and the INP_WLOCK() we must first
287  * grab a reference on the inp, which will hold the inp
288  * so that it can't be removed. We then unlock the INP_WLOCK(),
289  * and grab the INP_INFO_RLOCK() lock. Once we have the INP_INFO_RLOCK()
290  * we proceed again to get the INP_WLOCK() (this preserves proper
291  * lock order). After acquiring the INP_WLOCK we must check if someone
292  * else deleted the pcb i.e. the inp_flags check.
293  * If so we return 1 otherwise we return 0.
294  *
295  * No matter what the tcp_inpinfo_lock_add() function
296  * returns the caller must afterwards call tcp_inpinfo_lock_del()
297  * to drop the locks and reference properly.
298  */
299 
300 int
301 tcp_inpinfo_lock_add(struct inpcb *inp)
302 {
303 	in_pcbref(inp);
304 	INP_WUNLOCK(inp);
305 	INP_INFO_RLOCK(&V_tcbinfo);
306 	INP_WLOCK(inp);
307 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
308 		return(1);
309 	}
310 	return(0);
311 
312 }
313 
314 void
315 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
316 {
317 	INP_INFO_RUNLOCK(&V_tcbinfo);
318 	if (inp && (tp == NULL)) {
319 		/*
320 		 * If tcp_close/drop() gets called and tp
321 		 * returns NULL, then the function dropped
322 		 * the inp lock, we hold a reference keeping
323 		 * this around, so we must re-aquire the
324 		 * INP_WLOCK() in order to proceed with
325 		 * our dropping the inp reference.
326 		 */
327 		INP_WLOCK(inp);
328 	}
329 	if (inp && in_pcbrele_wlocked(inp) == 0)
330 		INP_WUNLOCK(inp);
331 }
332 
333 void
334 tcp_timer_2msl(void *xtp)
335 {
336 	struct tcpcb *tp = xtp;
337 	struct inpcb *inp;
338 	CURVNET_SET(tp->t_vnet);
339 #ifdef TCPDEBUG
340 	int ostate;
341 
342 	ostate = tp->t_state;
343 #endif
344 	inp = tp->t_inpcb;
345 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
346 	INP_WLOCK(inp);
347 	tcp_free_sackholes(tp);
348 	if (callout_pending(&tp->t_timers->tt_2msl) ||
349 	    !callout_active(&tp->t_timers->tt_2msl)) {
350 		INP_WUNLOCK(tp->t_inpcb);
351 		CURVNET_RESTORE();
352 		return;
353 	}
354 	callout_deactivate(&tp->t_timers->tt_2msl);
355 	if ((inp->inp_flags & INP_DROPPED) != 0) {
356 		INP_WUNLOCK(inp);
357 		CURVNET_RESTORE();
358 		return;
359 	}
360 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
361 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
362 	/*
363 	 * 2 MSL timeout in shutdown went off.  If we're closed but
364 	 * still waiting for peer to close and connection has been idle
365 	 * too long delete connection control block.  Otherwise, check
366 	 * again in a bit.
367 	 *
368 	 * If in TIME_WAIT state just ignore as this timeout is handled in
369 	 * tcp_tw_2msl_scan().
370 	 *
371 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
372 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
373 	 * Ignore fact that there were recent incoming segments.
374 	 */
375 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
376 		INP_WUNLOCK(inp);
377 		CURVNET_RESTORE();
378 		return;
379 	}
380 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
381 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
382 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
383 		TCPSTAT_INC(tcps_finwait2_drops);
384 		if (tcp_inpinfo_lock_add(inp)) {
385 			tcp_inpinfo_lock_del(inp, tp);
386 			goto out;
387 		}
388 		tp = tcp_close(tp);
389 		tcp_inpinfo_lock_del(inp, tp);
390 		goto out;
391 	} else {
392 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
393 			callout_reset(&tp->t_timers->tt_2msl,
394 				      TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
395 		} else {
396 			if (tcp_inpinfo_lock_add(inp)) {
397 				tcp_inpinfo_lock_del(inp, tp);
398 				goto out;
399 			}
400 			tp = tcp_close(tp);
401 			tcp_inpinfo_lock_del(inp, tp);
402 			goto out;
403 		}
404        }
405 
406 #ifdef TCPDEBUG
407 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
408 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
409 			  PRU_SLOWTIMO);
410 #endif
411 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
412 
413 	if (tp != NULL)
414 		INP_WUNLOCK(inp);
415 out:
416 	CURVNET_RESTORE();
417 }
418 
419 void
420 tcp_timer_keep(void *xtp)
421 {
422 	struct tcpcb *tp = xtp;
423 	struct tcptemp *t_template;
424 	struct inpcb *inp;
425 	CURVNET_SET(tp->t_vnet);
426 #ifdef TCPDEBUG
427 	int ostate;
428 
429 	ostate = tp->t_state;
430 #endif
431 	inp = tp->t_inpcb;
432 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
433 	INP_WLOCK(inp);
434 	if (callout_pending(&tp->t_timers->tt_keep) ||
435 	    !callout_active(&tp->t_timers->tt_keep)) {
436 		INP_WUNLOCK(inp);
437 		CURVNET_RESTORE();
438 		return;
439 	}
440 	callout_deactivate(&tp->t_timers->tt_keep);
441 	if ((inp->inp_flags & INP_DROPPED) != 0) {
442 		INP_WUNLOCK(inp);
443 		CURVNET_RESTORE();
444 		return;
445 	}
446 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
447 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
448 
449 	/*
450 	 * Because we don't regularly reset the keepalive callout in
451 	 * the ESTABLISHED state, it may be that we don't actually need
452 	 * to send a keepalive yet. If that occurs, schedule another
453 	 * call for the next time the keepalive timer might expire.
454 	 */
455 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
456 		u_int idletime;
457 
458 		idletime = ticks - tp->t_rcvtime;
459 		if (idletime < TP_KEEPIDLE(tp)) {
460 			callout_reset(&tp->t_timers->tt_keep,
461 			    TP_KEEPIDLE(tp) - idletime, tcp_timer_keep, tp);
462 			INP_WUNLOCK(inp);
463 			CURVNET_RESTORE();
464 			return;
465 		}
466 	}
467 
468 	/*
469 	 * Keep-alive timer went off; send something
470 	 * or drop connection if idle for too long.
471 	 */
472 	TCPSTAT_INC(tcps_keeptimeo);
473 	if (tp->t_state < TCPS_ESTABLISHED)
474 		goto dropit;
475 	if ((tcp_always_keepalive ||
476 	    inp->inp_socket->so_options & SO_KEEPALIVE) &&
477 	    tp->t_state <= TCPS_CLOSING) {
478 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
479 			goto dropit;
480 		/*
481 		 * Send a packet designed to force a response
482 		 * if the peer is up and reachable:
483 		 * either an ACK if the connection is still alive,
484 		 * or an RST if the peer has closed the connection
485 		 * due to timeout or reboot.
486 		 * Using sequence number tp->snd_una-1
487 		 * causes the transmitted zero-length segment
488 		 * to lie outside the receive window;
489 		 * by the protocol spec, this requires the
490 		 * correspondent TCP to respond.
491 		 */
492 		TCPSTAT_INC(tcps_keepprobe);
493 		t_template = tcpip_maketemplate(inp);
494 		if (t_template) {
495 			tcp_respond(tp, t_template->tt_ipgen,
496 				    &t_template->tt_t, (struct mbuf *)NULL,
497 				    tp->rcv_nxt, tp->snd_una - 1, 0);
498 			free(t_template, M_TEMP);
499 		}
500 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
501 			      tcp_timer_keep, tp);
502 	} else
503 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
504 			      tcp_timer_keep, tp);
505 
506 #ifdef TCPDEBUG
507 	if (inp->inp_socket->so_options & SO_DEBUG)
508 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
509 			  PRU_SLOWTIMO);
510 #endif
511 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
512 	INP_WUNLOCK(inp);
513 	CURVNET_RESTORE();
514 	return;
515 
516 dropit:
517 	TCPSTAT_INC(tcps_keepdrops);
518 
519 	if (tcp_inpinfo_lock_add(inp)) {
520 		tcp_inpinfo_lock_del(inp, tp);
521 		goto out;
522 	}
523 	tp = tcp_drop(tp, ETIMEDOUT);
524 
525 #ifdef TCPDEBUG
526 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
527 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
528 			  PRU_SLOWTIMO);
529 #endif
530 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
531 	tcp_inpinfo_lock_del(inp, tp);
532 out:
533 	CURVNET_RESTORE();
534 }
535 
536 void
537 tcp_timer_persist(void *xtp)
538 {
539 	struct tcpcb *tp = xtp;
540 	struct inpcb *inp;
541 	CURVNET_SET(tp->t_vnet);
542 #ifdef TCPDEBUG
543 	int ostate;
544 
545 	ostate = tp->t_state;
546 #endif
547 	inp = tp->t_inpcb;
548 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
549 	INP_WLOCK(inp);
550 	if (callout_pending(&tp->t_timers->tt_persist) ||
551 	    !callout_active(&tp->t_timers->tt_persist)) {
552 		INP_WUNLOCK(inp);
553 		CURVNET_RESTORE();
554 		return;
555 	}
556 	callout_deactivate(&tp->t_timers->tt_persist);
557 	if ((inp->inp_flags & INP_DROPPED) != 0) {
558 		INP_WUNLOCK(inp);
559 		CURVNET_RESTORE();
560 		return;
561 	}
562 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
563 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
564 	/*
565 	 * Persistence timer into zero window.
566 	 * Force a byte to be output, if possible.
567 	 */
568 	TCPSTAT_INC(tcps_persisttimeo);
569 	/*
570 	 * Hack: if the peer is dead/unreachable, we do not
571 	 * time out if the window is closed.  After a full
572 	 * backoff, drop the connection if the idle time
573 	 * (no responses to probes) reaches the maximum
574 	 * backoff that we would use if retransmitting.
575 	 */
576 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
577 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
578 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
579 		TCPSTAT_INC(tcps_persistdrop);
580 		if (tcp_inpinfo_lock_add(inp)) {
581 			tcp_inpinfo_lock_del(inp, tp);
582 			goto out;
583 		}
584 		tp = tcp_drop(tp, ETIMEDOUT);
585 		tcp_inpinfo_lock_del(inp, tp);
586 		goto out;
587 	}
588 	/*
589 	 * If the user has closed the socket then drop a persisting
590 	 * connection after a much reduced timeout.
591 	 */
592 	if (tp->t_state > TCPS_CLOSE_WAIT &&
593 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
594 		TCPSTAT_INC(tcps_persistdrop);
595 		if (tcp_inpinfo_lock_add(inp)) {
596 			tcp_inpinfo_lock_del(inp, tp);
597 			goto out;
598 		}
599 		tp = tcp_drop(tp, ETIMEDOUT);
600 		tcp_inpinfo_lock_del(inp, tp);
601 		goto out;
602 	}
603 	tcp_setpersist(tp);
604 	tp->t_flags |= TF_FORCEDATA;
605 	(void) tp->t_fb->tfb_tcp_output(tp);
606 	tp->t_flags &= ~TF_FORCEDATA;
607 
608 #ifdef TCPDEBUG
609 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
610 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
611 #endif
612 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
613 	INP_WUNLOCK(inp);
614 out:
615 	CURVNET_RESTORE();
616 }
617 
618 void
619 tcp_timer_rexmt(void * xtp)
620 {
621 	struct tcpcb *tp = xtp;
622 	CURVNET_SET(tp->t_vnet);
623 	int rexmt;
624 	struct inpcb *inp;
625 #ifdef TCPDEBUG
626 	int ostate;
627 
628 	ostate = tp->t_state;
629 #endif
630 	inp = tp->t_inpcb;
631 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
632 	INP_WLOCK(inp);
633 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
634 	    !callout_active(&tp->t_timers->tt_rexmt)) {
635 		INP_WUNLOCK(inp);
636 		CURVNET_RESTORE();
637 		return;
638 	}
639 	callout_deactivate(&tp->t_timers->tt_rexmt);
640 	if ((inp->inp_flags & INP_DROPPED) != 0) {
641 		INP_WUNLOCK(inp);
642 		CURVNET_RESTORE();
643 		return;
644 	}
645 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
646 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
647 	tcp_free_sackholes(tp);
648 	TCP_LOG_EVENT(tp, NULL, NULL, NULL, TCP_LOG_RTO, 0, 0, NULL, false);
649 	if (tp->t_fb->tfb_tcp_rexmit_tmr) {
650 		/* The stack has a timer action too. */
651 		(*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
652 	}
653 	/*
654 	 * Retransmission timer went off.  Message has not
655 	 * been acked within retransmit interval.  Back off
656 	 * to a longer retransmit interval and retransmit one segment.
657 	 */
658 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
659 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
660 		TCPSTAT_INC(tcps_timeoutdrop);
661 		if (tcp_inpinfo_lock_add(inp)) {
662 			tcp_inpinfo_lock_del(inp, tp);
663 			goto out;
664 		}
665 		tp = tcp_drop(tp, tp->t_softerror ?
666 			      tp->t_softerror : ETIMEDOUT);
667 		tcp_inpinfo_lock_del(inp, tp);
668 		goto out;
669 	}
670 	if (tp->t_state == TCPS_SYN_SENT) {
671 		/*
672 		 * If the SYN was retransmitted, indicate CWND to be
673 		 * limited to 1 segment in cc_conn_init().
674 		 */
675 		tp->snd_cwnd = 1;
676 	} else if (tp->t_rxtshift == 1) {
677 		/*
678 		 * first retransmit; record ssthresh and cwnd so they can
679 		 * be recovered if this turns out to be a "bad" retransmit.
680 		 * A retransmit is considered "bad" if an ACK for this
681 		 * segment is received within RTT/2 interval; the assumption
682 		 * here is that the ACK was already in flight.  See
683 		 * "On Estimating End-to-End Network Path Properties" by
684 		 * Allman and Paxson for more details.
685 		 */
686 		tp->snd_cwnd_prev = tp->snd_cwnd;
687 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
688 		tp->snd_recover_prev = tp->snd_recover;
689 		if (IN_FASTRECOVERY(tp->t_flags))
690 			tp->t_flags |= TF_WASFRECOVERY;
691 		else
692 			tp->t_flags &= ~TF_WASFRECOVERY;
693 		if (IN_CONGRECOVERY(tp->t_flags))
694 			tp->t_flags |= TF_WASCRECOVERY;
695 		else
696 			tp->t_flags &= ~TF_WASCRECOVERY;
697 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
698 		tp->t_flags |= TF_PREVVALID;
699 	} else
700 		tp->t_flags &= ~TF_PREVVALID;
701 	TCPSTAT_INC(tcps_rexmttimeo);
702 	if ((tp->t_state == TCPS_SYN_SENT) ||
703 	    (tp->t_state == TCPS_SYN_RECEIVED))
704 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
705 	else
706 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
707 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
708 		      tp->t_rttmin, TCPTV_REXMTMAX);
709 
710 	/*
711 	 * We enter the path for PLMTUD if connection is established or, if
712 	 * connection is FIN_WAIT_1 status, reason for the last is that if
713 	 * amount of data we send is very small, we could send it in couple of
714 	 * packets and process straight to FIN. In that case we won't catch
715 	 * ESTABLISHED state.
716 	 */
717 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
718 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
719 #ifdef INET6
720 		int isipv6;
721 #endif
722 
723 		/*
724 		 * Idea here is that at each stage of mtu probe (usually, 1448
725 		 * -> 1188 -> 524) should be given 2 chances to recover before
726 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
727 		 *  take care of that.
728 		 */
729 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
730 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
731 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
732 		    tp->t_rxtshift % 2 == 0)) {
733 			/*
734 			 * Enter Path MTU Black-hole Detection mechanism:
735 			 * - Disable Path MTU Discovery (IP "DF" bit).
736 			 * - Reduce MTU to lower value than what we
737 			 *   negotiated with peer.
738 			 */
739 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
740 				/* Record that we may have found a black hole. */
741 				tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
742 				/* Keep track of previous MSS. */
743 				tp->t_pmtud_saved_maxseg = tp->t_maxseg;
744 			}
745 
746 			/*
747 			 * Reduce the MSS to blackhole value or to the default
748 			 * in an attempt to retransmit.
749 			 */
750 #ifdef INET6
751 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
752 			if (isipv6 &&
753 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
754 				/* Use the sysctl tuneable blackhole MSS. */
755 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
756 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
757 			} else if (isipv6) {
758 				/* Use the default MSS. */
759 				tp->t_maxseg = V_tcp_v6mssdflt;
760 				/*
761 				 * Disable Path MTU Discovery when we switch to
762 				 * minmss.
763 				 */
764 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
765 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
766 			}
767 #endif
768 #if defined(INET6) && defined(INET)
769 			else
770 #endif
771 #ifdef INET
772 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
773 				/* Use the sysctl tuneable blackhole MSS. */
774 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
775 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
776 			} else {
777 				/* Use the default MSS. */
778 				tp->t_maxseg = V_tcp_mssdflt;
779 				/*
780 				 * Disable Path MTU Discovery when we switch to
781 				 * minmss.
782 				 */
783 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
784 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
785 			}
786 #endif
787 			/*
788 			 * Reset the slow-start flight size
789 			 * as it may depend on the new MSS.
790 			 */
791 			if (CC_ALGO(tp)->conn_init != NULL)
792 				CC_ALGO(tp)->conn_init(tp->ccv);
793 		} else {
794 			/*
795 			 * If further retransmissions are still unsuccessful
796 			 * with a lowered MTU, maybe this isn't a blackhole and
797 			 * we restore the previous MSS and blackhole detection
798 			 * flags.
799 			 * The limit '6' is determined by giving each probe
800 			 * stage (1448, 1188, 524) 2 chances to recover.
801 			 */
802 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
803 			    (tp->t_rxtshift >= 6)) {
804 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
805 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
806 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
807 				TCPSTAT_INC(tcps_pmtud_blackhole_failed);
808 				/*
809 				 * Reset the slow-start flight size as it
810 				 * may depend on the new MSS.
811 				 */
812 				if (CC_ALGO(tp)->conn_init != NULL)
813 					CC_ALGO(tp)->conn_init(tp->ccv);
814 			}
815 		}
816 	}
817 
818 	/*
819 	 * Disable RFC1323 and SACK if we haven't got any response to
820 	 * our third SYN to work-around some broken terminal servers
821 	 * (most of which have hopefully been retired) that have bad VJ
822 	 * header compression code which trashes TCP segments containing
823 	 * unknown-to-them TCP options.
824 	 */
825 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
826 	    (tp->t_rxtshift == 3))
827 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
828 	/*
829 	 * If we backed off this far, notify the L3 protocol that we're having
830 	 * connection problems.
831 	 */
832 	if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
833 #ifdef INET6
834 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
835 			in6_losing(tp->t_inpcb);
836 		else
837 #endif
838 			in_losing(tp->t_inpcb);
839 	}
840 	tp->snd_nxt = tp->snd_una;
841 	tp->snd_recover = tp->snd_max;
842 	/*
843 	 * Force a segment to be sent.
844 	 */
845 	tp->t_flags |= TF_ACKNOW;
846 	/*
847 	 * If timing a segment in this window, stop the timer.
848 	 */
849 	tp->t_rtttime = 0;
850 
851 	cc_cong_signal(tp, NULL, CC_RTO);
852 
853 	(void) tp->t_fb->tfb_tcp_output(tp);
854 
855 #ifdef TCPDEBUG
856 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
857 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
858 			  PRU_SLOWTIMO);
859 #endif
860 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
861 	INP_WUNLOCK(inp);
862 out:
863 	CURVNET_RESTORE();
864 }
865 
866 void
867 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
868 {
869 	struct callout *t_callout;
870 	timeout_t *f_callout;
871 	struct inpcb *inp = tp->t_inpcb;
872 	int cpu = inp_to_cpuid(inp);
873 
874 #ifdef TCP_OFFLOAD
875 	if (tp->t_flags & TF_TOE)
876 		return;
877 #endif
878 
879 	if (tp->t_timers->tt_flags & TT_STOPPED)
880 		return;
881 
882 	switch (timer_type) {
883 		case TT_DELACK:
884 			t_callout = &tp->t_timers->tt_delack;
885 			f_callout = tcp_timer_delack;
886 			break;
887 		case TT_REXMT:
888 			t_callout = &tp->t_timers->tt_rexmt;
889 			f_callout = tcp_timer_rexmt;
890 			break;
891 		case TT_PERSIST:
892 			t_callout = &tp->t_timers->tt_persist;
893 			f_callout = tcp_timer_persist;
894 			break;
895 		case TT_KEEP:
896 			t_callout = &tp->t_timers->tt_keep;
897 			f_callout = tcp_timer_keep;
898 			break;
899 		case TT_2MSL:
900 			t_callout = &tp->t_timers->tt_2msl;
901 			f_callout = tcp_timer_2msl;
902 			break;
903 		default:
904 			if (tp->t_fb->tfb_tcp_timer_activate) {
905 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
906 				return;
907 			}
908 			panic("tp %p bad timer_type %#x", tp, timer_type);
909 		}
910 	if (delta == 0) {
911 		callout_stop(t_callout);
912 	} else {
913 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
914 	}
915 }
916 
917 int
918 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
919 {
920 	struct callout *t_callout;
921 
922 	switch (timer_type) {
923 		case TT_DELACK:
924 			t_callout = &tp->t_timers->tt_delack;
925 			break;
926 		case TT_REXMT:
927 			t_callout = &tp->t_timers->tt_rexmt;
928 			break;
929 		case TT_PERSIST:
930 			t_callout = &tp->t_timers->tt_persist;
931 			break;
932 		case TT_KEEP:
933 			t_callout = &tp->t_timers->tt_keep;
934 			break;
935 		case TT_2MSL:
936 			t_callout = &tp->t_timers->tt_2msl;
937 			break;
938 		default:
939 			if (tp->t_fb->tfb_tcp_timer_active) {
940 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
941 			}
942 			panic("tp %p bad timer_type %#x", tp, timer_type);
943 		}
944 	return callout_active(t_callout);
945 }
946 
947 void
948 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
949 {
950 	struct callout *t_callout;
951 
952 	tp->t_timers->tt_flags |= TT_STOPPED;
953 	switch (timer_type) {
954 		case TT_DELACK:
955 			t_callout = &tp->t_timers->tt_delack;
956 			break;
957 		case TT_REXMT:
958 			t_callout = &tp->t_timers->tt_rexmt;
959 			break;
960 		case TT_PERSIST:
961 			t_callout = &tp->t_timers->tt_persist;
962 			break;
963 		case TT_KEEP:
964 			t_callout = &tp->t_timers->tt_keep;
965 			break;
966 		case TT_2MSL:
967 			t_callout = &tp->t_timers->tt_2msl;
968 			break;
969 		default:
970 			if (tp->t_fb->tfb_tcp_timer_stop) {
971 				/*
972 				 * XXXrrs we need to look at this with the
973 				 * stop case below (flags).
974 				 */
975 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
976 				return;
977 			}
978 			panic("tp %p bad timer_type %#x", tp, timer_type);
979 		}
980 
981 	if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
982 		/*
983 		 * Can't stop the callout, defer tcpcb actual deletion
984 		 * to the last one. We do this using the async drain
985 		 * function and incrementing the count in
986 		 */
987 		tp->t_timers->tt_draincnt++;
988 	}
989 }
990