xref: /freebsd/sys/netinet/tcp_timer.c (revision f173c2b77e0287807852bb7ba982643f33bdf8a0)
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  * 3. 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/in.h>
59 #include <netinet/in_kdtrace.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.h>
68 #include <netinet/tcp_fsm.h>
69 #include <netinet/tcp_timer.h>
70 #include <netinet/tcp_var.h>
71 #include <netinet/cc/cc.h>
72 #ifdef INET6
73 #include <netinet6/tcp6_var.h>
74 #endif
75 #include <netinet/tcpip.h>
76 #ifdef TCPDEBUG
77 #include <netinet/tcp_debug.h>
78 #endif
79 
80 int    tcp_persmin;
81 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmin, CTLTYPE_INT|CTLFLAG_RW,
82     &tcp_persmin, 0, sysctl_msec_to_ticks, "I", "minimum persistence interval");
83 
84 int    tcp_persmax;
85 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, persmax, CTLTYPE_INT|CTLFLAG_RW,
86     &tcp_persmax, 0, sysctl_msec_to_ticks, "I", "maximum persistence interval");
87 
88 int	tcp_keepinit;
89 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINIT, keepinit, CTLTYPE_INT|CTLFLAG_RW,
90     &tcp_keepinit, 0, sysctl_msec_to_ticks, "I", "time to establish connection");
91 
92 int	tcp_keepidle;
93 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPIDLE, keepidle, CTLTYPE_INT|CTLFLAG_RW,
94     &tcp_keepidle, 0, sysctl_msec_to_ticks, "I", "time before keepalive probes begin");
95 
96 int	tcp_keepintvl;
97 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KEEPINTVL, keepintvl, CTLTYPE_INT|CTLFLAG_RW,
98     &tcp_keepintvl, 0, sysctl_msec_to_ticks, "I", "time between keepalive probes");
99 
100 int	tcp_delacktime;
101 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DELACKTIME, delacktime, CTLTYPE_INT|CTLFLAG_RW,
102     &tcp_delacktime, 0, sysctl_msec_to_ticks, "I",
103     "Time before a delayed ACK is sent");
104 
105 int	tcp_msl;
106 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, msl, CTLTYPE_INT|CTLFLAG_RW,
107     &tcp_msl, 0, sysctl_msec_to_ticks, "I", "Maximum segment lifetime");
108 
109 int	tcp_rexmit_min;
110 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_min, CTLTYPE_INT|CTLFLAG_RW,
111     &tcp_rexmit_min, 0, sysctl_msec_to_ticks, "I",
112     "Minimum Retransmission Timeout");
113 
114 int	tcp_rexmit_slop;
115 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, rexmit_slop, CTLTYPE_INT|CTLFLAG_RW,
116     &tcp_rexmit_slop, 0, sysctl_msec_to_ticks, "I",
117     "Retransmission Timer Slop");
118 
119 static int	always_keepalive = 1;
120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, always_keepalive, CTLFLAG_RW,
121     &always_keepalive , 0, "Assume SO_KEEPALIVE on all TCP connections");
122 
123 int    tcp_fast_finwait2_recycle = 0;
124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fast_finwait2_recycle, CTLFLAG_RW,
125     &tcp_fast_finwait2_recycle, 0,
126     "Recycle closed FIN_WAIT_2 connections faster");
127 
128 int    tcp_finwait2_timeout;
129 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, finwait2_timeout, CTLTYPE_INT|CTLFLAG_RW,
130     &tcp_finwait2_timeout, 0, sysctl_msec_to_ticks, "I", "FIN-WAIT2 timeout");
131 
132 int	tcp_keepcnt = TCPTV_KEEPCNT;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, keepcnt, CTLFLAG_RW, &tcp_keepcnt, 0,
134     "Number of keepalive probes to send");
135 
136 	/* max idle probes */
137 int	tcp_maxpersistidle;
138 
139 static int	tcp_rexmit_drop_options = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmit_drop_options, CTLFLAG_RW,
141     &tcp_rexmit_drop_options, 0,
142     "Drop TCP options from 3rd and later retransmitted SYN");
143 
144 static VNET_DEFINE(int, tcp_pmtud_blackhole_detect);
145 #define	V_tcp_pmtud_blackhole_detect	VNET(tcp_pmtud_blackhole_detect)
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_detection,
147     CTLFLAG_RW|CTLFLAG_VNET,
148     &VNET_NAME(tcp_pmtud_blackhole_detect), 0,
149     "Path MTU Discovery Black Hole Detection Enabled");
150 
151 #ifdef INET
152 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
153 #define	V_tcp_pmtud_blackhole_mss	VNET(tcp_pmtud_blackhole_mss)
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
155     CTLFLAG_RW|CTLFLAG_VNET,
156     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
157     "Path MTU Discovery Black Hole Detection lowered MSS");
158 #endif
159 
160 #ifdef INET6
161 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
162 #define	V_tcp_v6pmtud_blackhole_mss	VNET(tcp_v6pmtud_blackhole_mss)
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 ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
476 	    tp->t_state <= TCPS_CLOSING) {
477 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
478 			goto dropit;
479 		/*
480 		 * Send a packet designed to force a response
481 		 * if the peer is up and reachable:
482 		 * either an ACK if the connection is still alive,
483 		 * or an RST if the peer has closed the connection
484 		 * due to timeout or reboot.
485 		 * Using sequence number tp->snd_una-1
486 		 * causes the transmitted zero-length segment
487 		 * to lie outside the receive window;
488 		 * by the protocol spec, this requires the
489 		 * correspondent TCP to respond.
490 		 */
491 		TCPSTAT_INC(tcps_keepprobe);
492 		t_template = tcpip_maketemplate(inp);
493 		if (t_template) {
494 			tcp_respond(tp, t_template->tt_ipgen,
495 				    &t_template->tt_t, (struct mbuf *)NULL,
496 				    tp->rcv_nxt, tp->snd_una - 1, 0);
497 			free(t_template, M_TEMP);
498 		}
499 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
500 			      tcp_timer_keep, tp);
501 	} else
502 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
503 			      tcp_timer_keep, tp);
504 
505 #ifdef TCPDEBUG
506 	if (inp->inp_socket->so_options & SO_DEBUG)
507 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
508 			  PRU_SLOWTIMO);
509 #endif
510 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
511 	INP_WUNLOCK(inp);
512 	CURVNET_RESTORE();
513 	return;
514 
515 dropit:
516 	TCPSTAT_INC(tcps_keepdrops);
517 
518 	if (tcp_inpinfo_lock_add(inp)) {
519 		tcp_inpinfo_lock_del(inp, tp);
520 		goto out;
521 	}
522 	tp = tcp_drop(tp, ETIMEDOUT);
523 
524 #ifdef TCPDEBUG
525 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
526 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
527 			  PRU_SLOWTIMO);
528 #endif
529 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
530 	tcp_inpinfo_lock_del(inp, tp);
531 out:
532 	CURVNET_RESTORE();
533 }
534 
535 void
536 tcp_timer_persist(void *xtp)
537 {
538 	struct tcpcb *tp = xtp;
539 	struct inpcb *inp;
540 	CURVNET_SET(tp->t_vnet);
541 #ifdef TCPDEBUG
542 	int ostate;
543 
544 	ostate = tp->t_state;
545 #endif
546 	inp = tp->t_inpcb;
547 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
548 	INP_WLOCK(inp);
549 	if (callout_pending(&tp->t_timers->tt_persist) ||
550 	    !callout_active(&tp->t_timers->tt_persist)) {
551 		INP_WUNLOCK(inp);
552 		CURVNET_RESTORE();
553 		return;
554 	}
555 	callout_deactivate(&tp->t_timers->tt_persist);
556 	if ((inp->inp_flags & INP_DROPPED) != 0) {
557 		INP_WUNLOCK(inp);
558 		CURVNET_RESTORE();
559 		return;
560 	}
561 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
562 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
563 	/*
564 	 * Persistence timer into zero window.
565 	 * Force a byte to be output, if possible.
566 	 */
567 	TCPSTAT_INC(tcps_persisttimeo);
568 	/*
569 	 * Hack: if the peer is dead/unreachable, we do not
570 	 * time out if the window is closed.  After a full
571 	 * backoff, drop the connection if the idle time
572 	 * (no responses to probes) reaches the maximum
573 	 * backoff that we would use if retransmitting.
574 	 */
575 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
576 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
577 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
578 		TCPSTAT_INC(tcps_persistdrop);
579 		if (tcp_inpinfo_lock_add(inp)) {
580 			tcp_inpinfo_lock_del(inp, tp);
581 			goto out;
582 		}
583 		tp = tcp_drop(tp, ETIMEDOUT);
584 		tcp_inpinfo_lock_del(inp, tp);
585 		goto out;
586 	}
587 	/*
588 	 * If the user has closed the socket then drop a persisting
589 	 * connection after a much reduced timeout.
590 	 */
591 	if (tp->t_state > TCPS_CLOSE_WAIT &&
592 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
593 		TCPSTAT_INC(tcps_persistdrop);
594 		if (tcp_inpinfo_lock_add(inp)) {
595 			tcp_inpinfo_lock_del(inp, tp);
596 			goto out;
597 		}
598 		tp = tcp_drop(tp, ETIMEDOUT);
599 		tcp_inpinfo_lock_del(inp, tp);
600 		goto out;
601 	}
602 	tcp_setpersist(tp);
603 	tp->t_flags |= TF_FORCEDATA;
604 	(void) tp->t_fb->tfb_tcp_output(tp);
605 	tp->t_flags &= ~TF_FORCEDATA;
606 
607 #ifdef TCPDEBUG
608 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
609 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
610 #endif
611 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
612 	INP_WUNLOCK(inp);
613 out:
614 	CURVNET_RESTORE();
615 }
616 
617 void
618 tcp_timer_rexmt(void * xtp)
619 {
620 	struct tcpcb *tp = xtp;
621 	CURVNET_SET(tp->t_vnet);
622 	int rexmt;
623 	struct inpcb *inp;
624 #ifdef TCPDEBUG
625 	int ostate;
626 
627 	ostate = tp->t_state;
628 #endif
629 	inp = tp->t_inpcb;
630 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
631 	INP_WLOCK(inp);
632 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
633 	    !callout_active(&tp->t_timers->tt_rexmt)) {
634 		INP_WUNLOCK(inp);
635 		CURVNET_RESTORE();
636 		return;
637 	}
638 	callout_deactivate(&tp->t_timers->tt_rexmt);
639 	if ((inp->inp_flags & INP_DROPPED) != 0) {
640 		INP_WUNLOCK(inp);
641 		CURVNET_RESTORE();
642 		return;
643 	}
644 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
645 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
646 	tcp_free_sackholes(tp);
647 	if (tp->t_fb->tfb_tcp_rexmit_tmr) {
648 		/* The stack has a timer action too. */
649 		(*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
650 	}
651 	/*
652 	 * Retransmission timer went off.  Message has not
653 	 * been acked within retransmit interval.  Back off
654 	 * to a longer retransmit interval and retransmit one segment.
655 	 */
656 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
657 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
658 		TCPSTAT_INC(tcps_timeoutdrop);
659 		if (tcp_inpinfo_lock_add(inp)) {
660 			tcp_inpinfo_lock_del(inp, tp);
661 			goto out;
662 		}
663 		tp = tcp_drop(tp, tp->t_softerror ?
664 			      tp->t_softerror : ETIMEDOUT);
665 		tcp_inpinfo_lock_del(inp, tp);
666 		goto out;
667 	}
668 	if (tp->t_state == TCPS_SYN_SENT) {
669 		/*
670 		 * If the SYN was retransmitted, indicate CWND to be
671 		 * limited to 1 segment in cc_conn_init().
672 		 */
673 		tp->snd_cwnd = 1;
674 	} else if (tp->t_rxtshift == 1) {
675 		/*
676 		 * first retransmit; record ssthresh and cwnd so they can
677 		 * be recovered if this turns out to be a "bad" retransmit.
678 		 * A retransmit is considered "bad" if an ACK for this
679 		 * segment is received within RTT/2 interval; the assumption
680 		 * here is that the ACK was already in flight.  See
681 		 * "On Estimating End-to-End Network Path Properties" by
682 		 * Allman and Paxson for more details.
683 		 */
684 		tp->snd_cwnd_prev = tp->snd_cwnd;
685 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
686 		tp->snd_recover_prev = tp->snd_recover;
687 		if (IN_FASTRECOVERY(tp->t_flags))
688 			tp->t_flags |= TF_WASFRECOVERY;
689 		else
690 			tp->t_flags &= ~TF_WASFRECOVERY;
691 		if (IN_CONGRECOVERY(tp->t_flags))
692 			tp->t_flags |= TF_WASCRECOVERY;
693 		else
694 			tp->t_flags &= ~TF_WASCRECOVERY;
695 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
696 		tp->t_flags |= TF_PREVVALID;
697 	} else
698 		tp->t_flags &= ~TF_PREVVALID;
699 	TCPSTAT_INC(tcps_rexmttimeo);
700 	if ((tp->t_state == TCPS_SYN_SENT) ||
701 	    (tp->t_state == TCPS_SYN_RECEIVED))
702 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
703 	else
704 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
705 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
706 		      tp->t_rttmin, TCPTV_REXMTMAX);
707 
708 	/*
709 	 * We enter the path for PLMTUD if connection is established or, if
710 	 * connection is FIN_WAIT_1 status, reason for the last is that if
711 	 * amount of data we send is very small, we could send it in couple of
712 	 * packets and process straight to FIN. In that case we won't catch
713 	 * ESTABLISHED state.
714 	 */
715 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
716 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
717 #ifdef INET6
718 		int isipv6;
719 #endif
720 
721 		/*
722 		 * Idea here is that at each stage of mtu probe (usually, 1448
723 		 * -> 1188 -> 524) should be given 2 chances to recover before
724 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
725 		 *  take care of that.
726 		 */
727 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
728 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
729 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift < 6 &&
730 		    tp->t_rxtshift % 2 == 0)) {
731 			/*
732 			 * Enter Path MTU Black-hole Detection mechanism:
733 			 * - Disable Path MTU Discovery (IP "DF" bit).
734 			 * - Reduce MTU to lower value than what we
735 			 *   negotiated with peer.
736 			 */
737 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) == 0) {
738 				/* Record that we may have found a black hole. */
739 				tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
740 				/* Keep track of previous MSS. */
741 				tp->t_pmtud_saved_maxseg = tp->t_maxseg;
742 			}
743 
744 			/*
745 			 * Reduce the MSS to blackhole value or to the default
746 			 * in an attempt to retransmit.
747 			 */
748 #ifdef INET6
749 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
750 			if (isipv6 &&
751 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
752 				/* Use the sysctl tuneable blackhole MSS. */
753 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
754 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
755 			} else if (isipv6) {
756 				/* Use the default MSS. */
757 				tp->t_maxseg = V_tcp_v6mssdflt;
758 				/*
759 				 * Disable Path MTU Discovery when we switch to
760 				 * minmss.
761 				 */
762 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
763 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
764 			}
765 #endif
766 #if defined(INET6) && defined(INET)
767 			else
768 #endif
769 #ifdef INET
770 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
771 				/* Use the sysctl tuneable blackhole MSS. */
772 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
773 				TCPSTAT_INC(tcps_pmtud_blackhole_activated);
774 			} else {
775 				/* Use the default MSS. */
776 				tp->t_maxseg = V_tcp_mssdflt;
777 				/*
778 				 * Disable Path MTU Discovery when we switch to
779 				 * minmss.
780 				 */
781 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
782 				TCPSTAT_INC(tcps_pmtud_blackhole_activated_min_mss);
783 			}
784 #endif
785 			/*
786 			 * Reset the slow-start flight size
787 			 * as it may depend on the new MSS.
788 			 */
789 			if (CC_ALGO(tp)->conn_init != NULL)
790 				CC_ALGO(tp)->conn_init(tp->ccv);
791 		} else {
792 			/*
793 			 * If further retransmissions are still unsuccessful
794 			 * with a lowered MTU, maybe this isn't a blackhole and
795 			 * we restore the previous MSS and blackhole detection
796 			 * flags.
797 			 * The limit '6' is determined by giving each probe
798 			 * stage (1448, 1188, 524) 2 chances to recover.
799 			 */
800 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
801 			    (tp->t_rxtshift >= 6)) {
802 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
803 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
804 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
805 				TCPSTAT_INC(tcps_pmtud_blackhole_failed);
806 				/*
807 				 * Reset the slow-start flight size as it
808 				 * may depend on the new MSS.
809 				 */
810 				if (CC_ALGO(tp)->conn_init != NULL)
811 					CC_ALGO(tp)->conn_init(tp->ccv);
812 			}
813 		}
814 	}
815 
816 	/*
817 	 * Disable RFC1323 and SACK if we haven't got any response to
818 	 * our third SYN to work-around some broken terminal servers
819 	 * (most of which have hopefully been retired) that have bad VJ
820 	 * header compression code which trashes TCP segments containing
821 	 * unknown-to-them TCP options.
822 	 */
823 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
824 	    (tp->t_rxtshift == 3))
825 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
826 	/*
827 	 * If we backed off this far, notify the L3 protocol that we're having
828 	 * connection problems.
829 	 */
830 	if (tp->t_rxtshift > TCP_RTT_INVALIDATE) {
831 #ifdef INET6
832 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
833 			in6_losing(tp->t_inpcb);
834 		else
835 #endif
836 			in_losing(tp->t_inpcb);
837 	}
838 	tp->snd_nxt = tp->snd_una;
839 	tp->snd_recover = tp->snd_max;
840 	/*
841 	 * Force a segment to be sent.
842 	 */
843 	tp->t_flags |= TF_ACKNOW;
844 	/*
845 	 * If timing a segment in this window, stop the timer.
846 	 */
847 	tp->t_rtttime = 0;
848 
849 	cc_cong_signal(tp, NULL, CC_RTO);
850 
851 	(void) tp->t_fb->tfb_tcp_output(tp);
852 
853 #ifdef TCPDEBUG
854 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
855 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
856 			  PRU_SLOWTIMO);
857 #endif
858 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
859 	INP_WUNLOCK(inp);
860 out:
861 	CURVNET_RESTORE();
862 }
863 
864 void
865 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
866 {
867 	struct callout *t_callout;
868 	timeout_t *f_callout;
869 	struct inpcb *inp = tp->t_inpcb;
870 	int cpu = inp_to_cpuid(inp);
871 
872 #ifdef TCP_OFFLOAD
873 	if (tp->t_flags & TF_TOE)
874 		return;
875 #endif
876 
877 	if (tp->t_timers->tt_flags & TT_STOPPED)
878 		return;
879 
880 	switch (timer_type) {
881 		case TT_DELACK:
882 			t_callout = &tp->t_timers->tt_delack;
883 			f_callout = tcp_timer_delack;
884 			break;
885 		case TT_REXMT:
886 			t_callout = &tp->t_timers->tt_rexmt;
887 			f_callout = tcp_timer_rexmt;
888 			break;
889 		case TT_PERSIST:
890 			t_callout = &tp->t_timers->tt_persist;
891 			f_callout = tcp_timer_persist;
892 			break;
893 		case TT_KEEP:
894 			t_callout = &tp->t_timers->tt_keep;
895 			f_callout = tcp_timer_keep;
896 			break;
897 		case TT_2MSL:
898 			t_callout = &tp->t_timers->tt_2msl;
899 			f_callout = tcp_timer_2msl;
900 			break;
901 		default:
902 			if (tp->t_fb->tfb_tcp_timer_activate) {
903 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
904 				return;
905 			}
906 			panic("tp %p bad timer_type %#x", tp, timer_type);
907 		}
908 	if (delta == 0) {
909 		callout_stop(t_callout);
910 	} else {
911 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
912 	}
913 }
914 
915 int
916 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
917 {
918 	struct callout *t_callout;
919 
920 	switch (timer_type) {
921 		case TT_DELACK:
922 			t_callout = &tp->t_timers->tt_delack;
923 			break;
924 		case TT_REXMT:
925 			t_callout = &tp->t_timers->tt_rexmt;
926 			break;
927 		case TT_PERSIST:
928 			t_callout = &tp->t_timers->tt_persist;
929 			break;
930 		case TT_KEEP:
931 			t_callout = &tp->t_timers->tt_keep;
932 			break;
933 		case TT_2MSL:
934 			t_callout = &tp->t_timers->tt_2msl;
935 			break;
936 		default:
937 			if (tp->t_fb->tfb_tcp_timer_active) {
938 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
939 			}
940 			panic("tp %p bad timer_type %#x", tp, timer_type);
941 		}
942 	return callout_active(t_callout);
943 }
944 
945 void
946 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
947 {
948 	struct callout *t_callout;
949 
950 	tp->t_timers->tt_flags |= TT_STOPPED;
951 	switch (timer_type) {
952 		case TT_DELACK:
953 			t_callout = &tp->t_timers->tt_delack;
954 			break;
955 		case TT_REXMT:
956 			t_callout = &tp->t_timers->tt_rexmt;
957 			break;
958 		case TT_PERSIST:
959 			t_callout = &tp->t_timers->tt_persist;
960 			break;
961 		case TT_KEEP:
962 			t_callout = &tp->t_timers->tt_keep;
963 			break;
964 		case TT_2MSL:
965 			t_callout = &tp->t_timers->tt_2msl;
966 			break;
967 		default:
968 			if (tp->t_fb->tfb_tcp_timer_stop) {
969 				/*
970 				 * XXXrrs we need to look at this with the
971 				 * stop case below (flags).
972 				 */
973 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
974 				return;
975 			}
976 			panic("tp %p bad timer_type %#x", tp, timer_type);
977 		}
978 
979 	if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
980 		/*
981 		 * Can't stop the callout, defer tcpcb actual deletion
982 		 * to the last one. We do this using the async drain
983 		 * function and incrementing the count in
984 		 */
985 		tp->t_timers->tt_draincnt++;
986 	}
987 }
988