xref: /freebsd/sys/netinet/tcp_timer.c (revision 09a53ad8f1318c5daae6cfb19d97f4f6459f0013)
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/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 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated);
152 #define	V_tcp_pmtud_blackhole_activated \
153     VNET(tcp_pmtud_blackhole_activated)
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated,
155     CTLFLAG_RD|CTLFLAG_VNET,
156     &VNET_NAME(tcp_pmtud_blackhole_activated), 0,
157     "Path MTU Discovery Black Hole Detection, Activation Count");
158 
159 static VNET_DEFINE(int, tcp_pmtud_blackhole_activated_min_mss);
160 #define	V_tcp_pmtud_blackhole_activated_min_mss \
161     VNET(tcp_pmtud_blackhole_activated_min_mss)
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_activated_min_mss,
163     CTLFLAG_RD|CTLFLAG_VNET,
164     &VNET_NAME(tcp_pmtud_blackhole_activated_min_mss), 0,
165     "Path MTU Discovery Black Hole Detection, Activation Count at min MSS");
166 
167 static VNET_DEFINE(int, tcp_pmtud_blackhole_failed);
168 #define	V_tcp_pmtud_blackhole_failed	VNET(tcp_pmtud_blackhole_failed)
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_failed,
170     CTLFLAG_RD|CTLFLAG_VNET,
171     &VNET_NAME(tcp_pmtud_blackhole_failed), 0,
172     "Path MTU Discovery Black Hole Detection, Failure Count");
173 
174 #ifdef INET
175 static VNET_DEFINE(int, tcp_pmtud_blackhole_mss) = 1200;
176 #define	V_tcp_pmtud_blackhole_mss	VNET(tcp_pmtud_blackhole_mss)
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pmtud_blackhole_mss,
178     CTLFLAG_RW|CTLFLAG_VNET,
179     &VNET_NAME(tcp_pmtud_blackhole_mss), 0,
180     "Path MTU Discovery Black Hole Detection lowered MSS");
181 #endif
182 
183 #ifdef INET6
184 static VNET_DEFINE(int, tcp_v6pmtud_blackhole_mss) = 1220;
185 #define	V_tcp_v6pmtud_blackhole_mss	VNET(tcp_v6pmtud_blackhole_mss)
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, v6pmtud_blackhole_mss,
187     CTLFLAG_RW|CTLFLAG_VNET,
188     &VNET_NAME(tcp_v6pmtud_blackhole_mss), 0,
189     "Path MTU Discovery IPv6 Black Hole Detection lowered MSS");
190 #endif
191 
192 #ifdef	RSS
193 static int	per_cpu_timers = 1;
194 #else
195 static int	per_cpu_timers = 0;
196 #endif
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, per_cpu_timers, CTLFLAG_RW,
198     &per_cpu_timers , 0, "run tcp timers on all cpus");
199 
200 #if 0
201 #define	INP_CPU(inp)	(per_cpu_timers ? (!CPU_ABSENT(((inp)->inp_flowid % (mp_maxid+1))) ? \
202 		((inp)->inp_flowid % (mp_maxid+1)) : curcpu) : 0)
203 #endif
204 
205 /*
206  * Map the given inp to a CPU id.
207  *
208  * This queries RSS if it's compiled in, else it defaults to the current
209  * CPU ID.
210  */
211 static inline int
212 inp_to_cpuid(struct inpcb *inp)
213 {
214 	u_int cpuid;
215 
216 #ifdef	RSS
217 	if (per_cpu_timers) {
218 		cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype);
219 		if (cpuid == NETISR_CPUID_NONE)
220 			return (curcpu);	/* XXX */
221 		else
222 			return (cpuid);
223 	}
224 #else
225 	/* Legacy, pre-RSS behaviour */
226 	if (per_cpu_timers) {
227 		/*
228 		 * We don't have a flowid -> cpuid mapping, so cheat and
229 		 * just map unknown cpuids to curcpu.  Not the best, but
230 		 * apparently better than defaulting to swi 0.
231 		 */
232 		cpuid = inp->inp_flowid % (mp_maxid + 1);
233 		if (! CPU_ABSENT(cpuid))
234 			return (cpuid);
235 		return (curcpu);
236 	}
237 #endif
238 	/* Default for RSS and non-RSS - cpuid 0 */
239 	else {
240 		return (0);
241 	}
242 }
243 
244 /*
245  * Tcp protocol timeout routine called every 500 ms.
246  * Updates timestamps used for TCP
247  * causes finite state machine actions if timers expire.
248  */
249 void
250 tcp_slowtimo(void)
251 {
252 	VNET_ITERATOR_DECL(vnet_iter);
253 
254 	VNET_LIST_RLOCK_NOSLEEP();
255 	VNET_FOREACH(vnet_iter) {
256 		CURVNET_SET(vnet_iter);
257 		(void) tcp_tw_2msl_scan(0);
258 		CURVNET_RESTORE();
259 	}
260 	VNET_LIST_RUNLOCK_NOSLEEP();
261 }
262 
263 int	tcp_syn_backoff[TCP_MAXRXTSHIFT + 1] =
264     { 1, 1, 1, 1, 1, 2, 4, 8, 16, 32, 64, 64, 64 };
265 
266 int	tcp_backoff[TCP_MAXRXTSHIFT + 1] =
267     { 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 512, 512, 512 };
268 
269 static int tcp_totbackoff = 2559;	/* sum of tcp_backoff[] */
270 
271 /*
272  * TCP timer processing.
273  */
274 
275 void
276 tcp_timer_delack(void *xtp)
277 {
278 	struct tcpcb *tp = xtp;
279 	struct inpcb *inp;
280 	CURVNET_SET(tp->t_vnet);
281 
282 	inp = tp->t_inpcb;
283 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
284 	INP_WLOCK(inp);
285 	if (callout_pending(&tp->t_timers->tt_delack) ||
286 	    !callout_active(&tp->t_timers->tt_delack)) {
287 		INP_WUNLOCK(inp);
288 		CURVNET_RESTORE();
289 		return;
290 	}
291 	callout_deactivate(&tp->t_timers->tt_delack);
292 	if ((inp->inp_flags & INP_DROPPED) != 0) {
293 		INP_WUNLOCK(inp);
294 		CURVNET_RESTORE();
295 		return;
296 	}
297 	tp->t_flags |= TF_ACKNOW;
298 	TCPSTAT_INC(tcps_delack);
299 	(void) tp->t_fb->tfb_tcp_output(tp);
300 	INP_WUNLOCK(inp);
301 	CURVNET_RESTORE();
302 }
303 
304 /*
305  * When a timer wants to remove a TCB it must
306  * hold the INP_INFO_RLOCK(). The timer function
307  * should only have grabbed the INP_WLOCK() when
308  * it entered. To safely switch to holding both the
309  * INP_INFO_RLOCK() and the INP_WLOCK() we must first
310  * grab a reference on the inp, which will hold the inp
311  * so that it can't be removed. We then unlock the INP_WLOCK(),
312  * and grab the INP_INFO_RLOCK() lock. Once we have the INP_INFO_RLOCK()
313  * we proceed again to get the INP_WLOCK() (this preserves proper
314  * lock order). After acquiring the INP_WLOCK we must check if someone
315  * else deleted the pcb i.e. the inp_flags check.
316  * If so we return 1 otherwise we return 0.
317  *
318  * No matter what the tcp_inpinfo_lock_add() function
319  * returns the caller must afterwards call tcp_inpinfo_lock_del()
320  * to drop the locks and reference properly.
321  */
322 
323 int
324 tcp_inpinfo_lock_add(struct inpcb *inp)
325 {
326 	in_pcbref(inp);
327 	INP_WUNLOCK(inp);
328 	INP_INFO_RLOCK(&V_tcbinfo);
329 	INP_WLOCK(inp);
330 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
331 		return(1);
332 	}
333 	return(0);
334 
335 }
336 
337 void
338 tcp_inpinfo_lock_del(struct inpcb *inp, struct tcpcb *tp)
339 {
340 	INP_INFO_RUNLOCK(&V_tcbinfo);
341 	if (inp && (tp == NULL)) {
342 		/*
343 		 * If tcp_close/drop() gets called and tp
344 		 * returns NULL, then the function dropped
345 		 * the inp lock, we hold a reference keeping
346 		 * this around, so we must re-aquire the
347 		 * INP_WLOCK() in order to proceed with
348 		 * our dropping the inp reference.
349 		 */
350 		INP_WLOCK(inp);
351 	}
352 	if (inp && in_pcbrele_wlocked(inp) == 0)
353 		INP_WUNLOCK(inp);
354 }
355 
356 void
357 tcp_timer_2msl(void *xtp)
358 {
359 	struct tcpcb *tp = xtp;
360 	struct inpcb *inp;
361 	CURVNET_SET(tp->t_vnet);
362 #ifdef TCPDEBUG
363 	int ostate;
364 
365 	ostate = tp->t_state;
366 #endif
367 	inp = tp->t_inpcb;
368 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
369 	INP_WLOCK(inp);
370 	tcp_free_sackholes(tp);
371 	if (callout_pending(&tp->t_timers->tt_2msl) ||
372 	    !callout_active(&tp->t_timers->tt_2msl)) {
373 		INP_WUNLOCK(tp->t_inpcb);
374 		CURVNET_RESTORE();
375 		return;
376 	}
377 	callout_deactivate(&tp->t_timers->tt_2msl);
378 	if ((inp->inp_flags & INP_DROPPED) != 0) {
379 		INP_WUNLOCK(inp);
380 		CURVNET_RESTORE();
381 		return;
382 	}
383 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
384 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
385 	/*
386 	 * 2 MSL timeout in shutdown went off.  If we're closed but
387 	 * still waiting for peer to close and connection has been idle
388 	 * too long delete connection control block.  Otherwise, check
389 	 * again in a bit.
390 	 *
391 	 * If in TIME_WAIT state just ignore as this timeout is handled in
392 	 * tcp_tw_2msl_scan().
393 	 *
394 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
395 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
396 	 * Ignore fact that there were recent incoming segments.
397 	 */
398 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
399 		INP_WUNLOCK(inp);
400 		CURVNET_RESTORE();
401 		return;
402 	}
403 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
404 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
405 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
406 		TCPSTAT_INC(tcps_finwait2_drops);
407 		if (tcp_inpinfo_lock_add(inp)) {
408 			tcp_inpinfo_lock_del(inp, tp);
409 			goto out;
410 		}
411 		tp = tcp_close(tp);
412 		tcp_inpinfo_lock_del(inp, tp);
413 		goto out;
414 	} else {
415 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
416 			callout_reset(&tp->t_timers->tt_2msl,
417 				      TP_KEEPINTVL(tp), tcp_timer_2msl, tp);
418 		} else {
419 			if (tcp_inpinfo_lock_add(inp)) {
420 				tcp_inpinfo_lock_del(inp, tp);
421 				goto out;
422 			}
423 			tp = tcp_close(tp);
424 			tcp_inpinfo_lock_del(inp, tp);
425 			goto out;
426 		}
427        }
428 
429 #ifdef TCPDEBUG
430 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
431 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
432 			  PRU_SLOWTIMO);
433 #endif
434 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
435 
436 	if (tp != NULL)
437 		INP_WUNLOCK(inp);
438 out:
439 	CURVNET_RESTORE();
440 }
441 
442 void
443 tcp_timer_keep(void *xtp)
444 {
445 	struct tcpcb *tp = xtp;
446 	struct tcptemp *t_template;
447 	struct inpcb *inp;
448 	CURVNET_SET(tp->t_vnet);
449 #ifdef TCPDEBUG
450 	int ostate;
451 
452 	ostate = tp->t_state;
453 #endif
454 	inp = tp->t_inpcb;
455 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
456 	INP_WLOCK(inp);
457 	if (callout_pending(&tp->t_timers->tt_keep) ||
458 	    !callout_active(&tp->t_timers->tt_keep)) {
459 		INP_WUNLOCK(inp);
460 		CURVNET_RESTORE();
461 		return;
462 	}
463 	callout_deactivate(&tp->t_timers->tt_keep);
464 	if ((inp->inp_flags & INP_DROPPED) != 0) {
465 		INP_WUNLOCK(inp);
466 		CURVNET_RESTORE();
467 		return;
468 	}
469 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
470 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
471 	/*
472 	 * Keep-alive timer went off; send something
473 	 * or drop connection if idle for too long.
474 	 */
475 	TCPSTAT_INC(tcps_keeptimeo);
476 	if (tp->t_state < TCPS_ESTABLISHED)
477 		goto dropit;
478 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
479 	    tp->t_state <= TCPS_CLOSING) {
480 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
481 			goto dropit;
482 		/*
483 		 * Send a packet designed to force a response
484 		 * if the peer is up and reachable:
485 		 * either an ACK if the connection is still alive,
486 		 * or an RST if the peer has closed the connection
487 		 * due to timeout or reboot.
488 		 * Using sequence number tp->snd_una-1
489 		 * causes the transmitted zero-length segment
490 		 * to lie outside the receive window;
491 		 * by the protocol spec, this requires the
492 		 * correspondent TCP to respond.
493 		 */
494 		TCPSTAT_INC(tcps_keepprobe);
495 		t_template = tcpip_maketemplate(inp);
496 		if (t_template) {
497 			tcp_respond(tp, t_template->tt_ipgen,
498 				    &t_template->tt_t, (struct mbuf *)NULL,
499 				    tp->rcv_nxt, tp->snd_una - 1, 0);
500 			free(t_template, M_TEMP);
501 		}
502 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
503 			      tcp_timer_keep, tp);
504 	} else
505 		callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
506 			      tcp_timer_keep, tp);
507 
508 #ifdef TCPDEBUG
509 	if (inp->inp_socket->so_options & SO_DEBUG)
510 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
511 			  PRU_SLOWTIMO);
512 #endif
513 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
514 	INP_WUNLOCK(inp);
515 	CURVNET_RESTORE();
516 	return;
517 
518 dropit:
519 	TCPSTAT_INC(tcps_keepdrops);
520 
521 	if (tcp_inpinfo_lock_add(inp)) {
522 		tcp_inpinfo_lock_del(inp, tp);
523 		goto out;
524 	}
525 	tp = tcp_drop(tp, ETIMEDOUT);
526 
527 #ifdef TCPDEBUG
528 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
529 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
530 			  PRU_SLOWTIMO);
531 #endif
532 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
533 	tcp_inpinfo_lock_del(inp, tp);
534 out:
535 	CURVNET_RESTORE();
536 }
537 
538 void
539 tcp_timer_persist(void *xtp)
540 {
541 	struct tcpcb *tp = xtp;
542 	struct inpcb *inp;
543 	CURVNET_SET(tp->t_vnet);
544 #ifdef TCPDEBUG
545 	int ostate;
546 
547 	ostate = tp->t_state;
548 #endif
549 	inp = tp->t_inpcb;
550 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
551 	INP_WLOCK(inp);
552 	if (callout_pending(&tp->t_timers->tt_persist) ||
553 	    !callout_active(&tp->t_timers->tt_persist)) {
554 		INP_WUNLOCK(inp);
555 		CURVNET_RESTORE();
556 		return;
557 	}
558 	callout_deactivate(&tp->t_timers->tt_persist);
559 	if ((inp->inp_flags & INP_DROPPED) != 0) {
560 		INP_WUNLOCK(inp);
561 		CURVNET_RESTORE();
562 		return;
563 	}
564 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
565 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
566 	/*
567 	 * Persistence timer into zero window.
568 	 * Force a byte to be output, if possible.
569 	 */
570 	TCPSTAT_INC(tcps_persisttimeo);
571 	/*
572 	 * Hack: if the peer is dead/unreachable, we do not
573 	 * time out if the window is closed.  After a full
574 	 * backoff, drop the connection if the idle time
575 	 * (no responses to probes) reaches the maximum
576 	 * backoff that we would use if retransmitting.
577 	 */
578 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
579 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
580 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
581 		TCPSTAT_INC(tcps_persistdrop);
582 		if (tcp_inpinfo_lock_add(inp)) {
583 			tcp_inpinfo_lock_del(inp, tp);
584 			goto out;
585 		}
586 		tp = tcp_drop(tp, ETIMEDOUT);
587 		tcp_inpinfo_lock_del(inp, tp);
588 		goto out;
589 	}
590 	/*
591 	 * If the user has closed the socket then drop a persisting
592 	 * connection after a much reduced timeout.
593 	 */
594 	if (tp->t_state > TCPS_CLOSE_WAIT &&
595 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
596 		TCPSTAT_INC(tcps_persistdrop);
597 		if (tcp_inpinfo_lock_add(inp)) {
598 			tcp_inpinfo_lock_del(inp, tp);
599 			goto out;
600 		}
601 		tp = tcp_drop(tp, ETIMEDOUT);
602 		tcp_inpinfo_lock_del(inp, tp);
603 		goto out;
604 	}
605 	tcp_setpersist(tp);
606 	tp->t_flags |= TF_FORCEDATA;
607 	(void) tp->t_fb->tfb_tcp_output(tp);
608 	tp->t_flags &= ~TF_FORCEDATA;
609 
610 #ifdef TCPDEBUG
611 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
612 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
613 #endif
614 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
615 	INP_WUNLOCK(inp);
616 out:
617 	CURVNET_RESTORE();
618 }
619 
620 void
621 tcp_timer_rexmt(void * xtp)
622 {
623 	struct tcpcb *tp = xtp;
624 	CURVNET_SET(tp->t_vnet);
625 	int rexmt;
626 	struct inpcb *inp;
627 #ifdef TCPDEBUG
628 	int ostate;
629 
630 	ostate = tp->t_state;
631 #endif
632 	inp = tp->t_inpcb;
633 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
634 	INP_WLOCK(inp);
635 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
636 	    !callout_active(&tp->t_timers->tt_rexmt)) {
637 		INP_WUNLOCK(inp);
638 		CURVNET_RESTORE();
639 		return;
640 	}
641 	callout_deactivate(&tp->t_timers->tt_rexmt);
642 	if ((inp->inp_flags & INP_DROPPED) != 0) {
643 		INP_WUNLOCK(inp);
644 		CURVNET_RESTORE();
645 		return;
646 	}
647 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
648 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
649 	tcp_free_sackholes(tp);
650 	if (tp->t_fb->tfb_tcp_rexmit_tmr) {
651 		/* The stack has a timer action too. */
652 		(*tp->t_fb->tfb_tcp_rexmit_tmr)(tp);
653 	}
654 	/*
655 	 * Retransmission timer went off.  Message has not
656 	 * been acked within retransmit interval.  Back off
657 	 * to a longer retransmit interval and retransmit one segment.
658 	 */
659 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
660 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
661 		TCPSTAT_INC(tcps_timeoutdrop);
662 		if (tcp_inpinfo_lock_add(inp)) {
663 			tcp_inpinfo_lock_del(inp, tp);
664 			goto out;
665 		}
666 		tp = tcp_drop(tp, tp->t_softerror ?
667 			      tp->t_softerror : ETIMEDOUT);
668 		tcp_inpinfo_lock_del(inp, tp);
669 		goto out;
670 	}
671 	if (tp->t_state == TCPS_SYN_SENT) {
672 		/*
673 		 * If the SYN was retransmitted, indicate CWND to be
674 		 * limited to 1 segment in cc_conn_init().
675 		 */
676 		tp->snd_cwnd = 1;
677 	} else if (tp->t_rxtshift == 1) {
678 		/*
679 		 * first retransmit; record ssthresh and cwnd so they can
680 		 * be recovered if this turns out to be a "bad" retransmit.
681 		 * A retransmit is considered "bad" if an ACK for this
682 		 * segment is received within RTT/2 interval; the assumption
683 		 * here is that the ACK was already in flight.  See
684 		 * "On Estimating End-to-End Network Path Properties" by
685 		 * Allman and Paxson for more details.
686 		 */
687 		tp->snd_cwnd_prev = tp->snd_cwnd;
688 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
689 		tp->snd_recover_prev = tp->snd_recover;
690 		if (IN_FASTRECOVERY(tp->t_flags))
691 			tp->t_flags |= TF_WASFRECOVERY;
692 		else
693 			tp->t_flags &= ~TF_WASFRECOVERY;
694 		if (IN_CONGRECOVERY(tp->t_flags))
695 			tp->t_flags |= TF_WASCRECOVERY;
696 		else
697 			tp->t_flags &= ~TF_WASCRECOVERY;
698 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
699 		tp->t_flags |= TF_PREVVALID;
700 	} else
701 		tp->t_flags &= ~TF_PREVVALID;
702 	TCPSTAT_INC(tcps_rexmttimeo);
703 	if ((tp->t_state == TCPS_SYN_SENT) ||
704 	    (tp->t_state == TCPS_SYN_RECEIVED))
705 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
706 	else
707 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
708 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
709 		      tp->t_rttmin, TCPTV_REXMTMAX);
710 
711 	/*
712 	 * We enter the path for PLMTUD if connection is established or, if
713 	 * connection is FIN_WAIT_1 status, reason for the last is that if
714 	 * amount of data we send is very small, we could send it in couple of
715 	 * packets and process straight to FIN. In that case we won't catch
716 	 * ESTABLISHED state.
717 	 */
718 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
719 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
720 #ifdef INET6
721 		int isipv6;
722 #endif
723 
724 		/*
725 		 * Idea here is that at each stage of mtu probe (usually, 1448
726 		 * -> 1188 -> 524) should be given 2 chances to recover before
727 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
728 		 *  take care of that.
729 		 */
730 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
731 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
732 		    (tp->t_rxtshift >= 2 && 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 			/* Record that we may have found a black hole. */
740 			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
741 
742 			/* Keep track of previous MSS. */
743 			tp->t_pmtud_saved_maxseg = tp->t_maxseg;
744 
745 			/*
746 			 * Reduce the MSS to blackhole value or to the default
747 			 * in an attempt to retransmit.
748 			 */
749 #ifdef INET6
750 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
751 			if (isipv6 &&
752 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
753 				/* Use the sysctl tuneable blackhole MSS. */
754 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
755 				V_tcp_pmtud_blackhole_activated++;
756 			} else if (isipv6) {
757 				/* Use the default MSS. */
758 				tp->t_maxseg = V_tcp_v6mssdflt;
759 				/*
760 				 * Disable Path MTU Discovery when we switch to
761 				 * minmss.
762 				 */
763 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
764 				V_tcp_pmtud_blackhole_activated_min_mss++;
765 			}
766 #endif
767 #if defined(INET6) && defined(INET)
768 			else
769 #endif
770 #ifdef INET
771 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
772 				/* Use the sysctl tuneable blackhole MSS. */
773 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
774 				V_tcp_pmtud_blackhole_activated++;
775 			} else {
776 				/* Use the default MSS. */
777 				tp->t_maxseg = V_tcp_mssdflt;
778 				/*
779 				 * Disable Path MTU Discovery when we switch to
780 				 * minmss.
781 				 */
782 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
783 				V_tcp_pmtud_blackhole_activated_min_mss++;
784 			}
785 #endif
786 			/*
787 			 * Reset the slow-start flight size
788 			 * as it may depend on the new MSS.
789 			 */
790 			if (CC_ALGO(tp)->conn_init != NULL)
791 				CC_ALGO(tp)->conn_init(tp->ccv);
792 		} else {
793 			/*
794 			 * If further retransmissions are still unsuccessful
795 			 * with a lowered MTU, maybe this isn't a blackhole and
796 			 * we restore the previous MSS and blackhole detection
797 			 * flags.
798 			 * The limit '6' is determined by giving each probe
799 			 * stage (1448, 1188, 524) 2 chances to recover.
800 			 */
801 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
802 			    (tp->t_rxtshift > 6)) {
803 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
804 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
805 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
806 				V_tcp_pmtud_blackhole_failed++;
807 				/*
808 				 * Reset the slow-start flight size as it
809 				 * may depend on the new MSS.
810 				 */
811 				if (CC_ALGO(tp)->conn_init != NULL)
812 					CC_ALGO(tp)->conn_init(tp->ccv);
813 			}
814 		}
815 	}
816 
817 	/*
818 	 * Disable RFC1323 and SACK if we haven't got any response to
819 	 * our third SYN to work-around some broken terminal servers
820 	 * (most of which have hopefully been retired) that have bad VJ
821 	 * header compression code which trashes TCP segments containing
822 	 * unknown-to-them TCP options.
823 	 */
824 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
825 	    (tp->t_rxtshift == 3))
826 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
827 	/*
828 	 * If we backed off this far, our srtt estimate is probably bogus.
829 	 * Clobber it so we'll take the next rtt measurement as our srtt;
830 	 * move the current srtt into rttvar to keep the current
831 	 * retransmit times until then.
832 	 */
833 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
834 #ifdef INET6
835 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
836 			in6_losing(tp->t_inpcb);
837 		else
838 #endif
839 			in_losing(tp->t_inpcb);
840 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
841 		tp->t_srtt = 0;
842 	}
843 	tp->snd_nxt = tp->snd_una;
844 	tp->snd_recover = tp->snd_max;
845 	/*
846 	 * Force a segment to be sent.
847 	 */
848 	tp->t_flags |= TF_ACKNOW;
849 	/*
850 	 * If timing a segment in this window, stop the timer.
851 	 */
852 	tp->t_rtttime = 0;
853 
854 	cc_cong_signal(tp, NULL, CC_RTO);
855 
856 	(void) tp->t_fb->tfb_tcp_output(tp);
857 
858 #ifdef TCPDEBUG
859 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
860 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
861 			  PRU_SLOWTIMO);
862 #endif
863 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
864 	INP_WUNLOCK(inp);
865 out:
866 	CURVNET_RESTORE();
867 }
868 
869 void
870 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
871 {
872 	struct callout *t_callout;
873 	timeout_t *f_callout;
874 	struct inpcb *inp = tp->t_inpcb;
875 	int cpu = inp_to_cpuid(inp);
876 
877 #ifdef TCP_OFFLOAD
878 	if (tp->t_flags & TF_TOE)
879 		return;
880 #endif
881 
882 	if (tp->t_timers->tt_flags & TT_STOPPED)
883 		return;
884 
885 	switch (timer_type) {
886 		case TT_DELACK:
887 			t_callout = &tp->t_timers->tt_delack;
888 			f_callout = tcp_timer_delack;
889 			break;
890 		case TT_REXMT:
891 			t_callout = &tp->t_timers->tt_rexmt;
892 			f_callout = tcp_timer_rexmt;
893 			break;
894 		case TT_PERSIST:
895 			t_callout = &tp->t_timers->tt_persist;
896 			f_callout = tcp_timer_persist;
897 			break;
898 		case TT_KEEP:
899 			t_callout = &tp->t_timers->tt_keep;
900 			f_callout = tcp_timer_keep;
901 			break;
902 		case TT_2MSL:
903 			t_callout = &tp->t_timers->tt_2msl;
904 			f_callout = tcp_timer_2msl;
905 			break;
906 		default:
907 			if (tp->t_fb->tfb_tcp_timer_activate) {
908 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
909 				return;
910 			}
911 			panic("tp %p bad timer_type %#x", tp, timer_type);
912 		}
913 	if (delta == 0) {
914 		callout_stop(t_callout);
915 	} else {
916 		callout_reset_on(t_callout, delta, f_callout, tp, cpu);
917 	}
918 }
919 
920 int
921 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
922 {
923 	struct callout *t_callout;
924 
925 	switch (timer_type) {
926 		case TT_DELACK:
927 			t_callout = &tp->t_timers->tt_delack;
928 			break;
929 		case TT_REXMT:
930 			t_callout = &tp->t_timers->tt_rexmt;
931 			break;
932 		case TT_PERSIST:
933 			t_callout = &tp->t_timers->tt_persist;
934 			break;
935 		case TT_KEEP:
936 			t_callout = &tp->t_timers->tt_keep;
937 			break;
938 		case TT_2MSL:
939 			t_callout = &tp->t_timers->tt_2msl;
940 			break;
941 		default:
942 			if (tp->t_fb->tfb_tcp_timer_active) {
943 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
944 			}
945 			panic("tp %p bad timer_type %#x", tp, timer_type);
946 		}
947 	return callout_active(t_callout);
948 }
949 
950 void
951 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
952 {
953 	struct callout *t_callout;
954 
955 	tp->t_timers->tt_flags |= TT_STOPPED;
956 	switch (timer_type) {
957 		case TT_DELACK:
958 			t_callout = &tp->t_timers->tt_delack;
959 			break;
960 		case TT_REXMT:
961 			t_callout = &tp->t_timers->tt_rexmt;
962 			break;
963 		case TT_PERSIST:
964 			t_callout = &tp->t_timers->tt_persist;
965 			break;
966 		case TT_KEEP:
967 			t_callout = &tp->t_timers->tt_keep;
968 			break;
969 		case TT_2MSL:
970 			t_callout = &tp->t_timers->tt_2msl;
971 			break;
972 		default:
973 			if (tp->t_fb->tfb_tcp_timer_stop) {
974 				/*
975 				 * XXXrrs we need to look at this with the
976 				 * stop case below (flags).
977 				 */
978 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
979 				return;
980 			}
981 			panic("tp %p bad timer_type %#x", tp, timer_type);
982 		}
983 
984 	if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
985 		/*
986 		 * Can't stop the callout, defer tcpcb actual deletion
987 		 * to the last one. We do this using the async drain
988 		 * function and incrementing the count in
989 		 */
990 		tp->t_timers->tt_draincnt++;
991 	}
992 }
993 
994 #define	ticks_to_msecs(t)	(1000*(t) / hz)
995 
996 void
997 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
998     struct xtcp_timer *xtimer)
999 {
1000 	sbintime_t now;
1001 
1002 	bzero(xtimer, sizeof(*xtimer));
1003 	if (timer == NULL)
1004 		return;
1005 	now = getsbinuptime();
1006 	if (callout_active(&timer->tt_delack))
1007 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
1008 	if (callout_active(&timer->tt_rexmt))
1009 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
1010 	if (callout_active(&timer->tt_persist))
1011 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
1012 	if (callout_active(&timer->tt_keep))
1013 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
1014 	if (callout_active(&timer->tt_2msl))
1015 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
1016 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
1017 }
1018