xref: /freebsd/sys/netinet/tcp_timer.c (revision 49dae58b287906be26f56ba3e3dc693c3ba8cf37)
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 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
298 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
299 	KASSERT((tp->t_timers->tt_flags & TT_DELACK) != 0,
300 		("%s: tp %p delack callout should be running", __func__, tp));
301 
302 	tp->t_flags |= TF_ACKNOW;
303 	TCPSTAT_INC(tcps_delack);
304 	(void) tp->t_fb->tfb_tcp_output(tp);
305 	INP_WUNLOCK(inp);
306 	CURVNET_RESTORE();
307 }
308 
309 void
310 tcp_timer_2msl(void *xtp)
311 {
312 	struct tcpcb *tp = xtp;
313 	struct inpcb *inp;
314 	CURVNET_SET(tp->t_vnet);
315 #ifdef TCPDEBUG
316 	int ostate;
317 
318 	ostate = tp->t_state;
319 #endif
320 	INP_INFO_RLOCK(&V_tcbinfo);
321 	inp = tp->t_inpcb;
322 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
323 	INP_WLOCK(inp);
324 	tcp_free_sackholes(tp);
325 	if (callout_pending(&tp->t_timers->tt_2msl) ||
326 	    !callout_active(&tp->t_timers->tt_2msl)) {
327 		INP_WUNLOCK(tp->t_inpcb);
328 		INP_INFO_RUNLOCK(&V_tcbinfo);
329 		CURVNET_RESTORE();
330 		return;
331 	}
332 	callout_deactivate(&tp->t_timers->tt_2msl);
333 	if ((inp->inp_flags & INP_DROPPED) != 0) {
334 		INP_WUNLOCK(inp);
335 		INP_INFO_RUNLOCK(&V_tcbinfo);
336 		CURVNET_RESTORE();
337 		return;
338 	}
339 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
340 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
341 	KASSERT((tp->t_timers->tt_flags & TT_2MSL) != 0,
342 		("%s: tp %p 2msl callout should be running", __func__, tp));
343 	/*
344 	 * 2 MSL timeout in shutdown went off.  If we're closed but
345 	 * still waiting for peer to close and connection has been idle
346 	 * too long delete connection control block.  Otherwise, check
347 	 * again in a bit.
348 	 *
349 	 * If in TIME_WAIT state just ignore as this timeout is handled in
350 	 * tcp_tw_2msl_scan().
351 	 *
352 	 * If fastrecycle of FIN_WAIT_2, in FIN_WAIT_2 and receiver has closed,
353 	 * there's no point in hanging onto FIN_WAIT_2 socket. Just close it.
354 	 * Ignore fact that there were recent incoming segments.
355 	 */
356 	if ((inp->inp_flags & INP_TIMEWAIT) != 0) {
357 		INP_WUNLOCK(inp);
358 		INP_INFO_RUNLOCK(&V_tcbinfo);
359 		CURVNET_RESTORE();
360 		return;
361 	}
362 	if (tcp_fast_finwait2_recycle && tp->t_state == TCPS_FIN_WAIT_2 &&
363 	    tp->t_inpcb && tp->t_inpcb->inp_socket &&
364 	    (tp->t_inpcb->inp_socket->so_rcv.sb_state & SBS_CANTRCVMORE)) {
365 		TCPSTAT_INC(tcps_finwait2_drops);
366 		tp = tcp_close(tp);
367 	} else {
368 		if (ticks - tp->t_rcvtime <= TP_MAXIDLE(tp)) {
369 			if (!callout_reset(&tp->t_timers->tt_2msl,
370 			   TP_KEEPINTVL(tp), tcp_timer_2msl, tp)) {
371 				tp->t_timers->tt_flags &= ~TT_2MSL_RST;
372 			}
373 		} else
374 		       tp = tcp_close(tp);
375        }
376 
377 #ifdef TCPDEBUG
378 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
379 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
380 			  PRU_SLOWTIMO);
381 #endif
382 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
383 
384 	if (tp != NULL)
385 		INP_WUNLOCK(inp);
386 	INP_INFO_RUNLOCK(&V_tcbinfo);
387 	CURVNET_RESTORE();
388 }
389 
390 void
391 tcp_timer_keep(void *xtp)
392 {
393 	struct tcpcb *tp = xtp;
394 	struct tcptemp *t_template;
395 	struct inpcb *inp;
396 	CURVNET_SET(tp->t_vnet);
397 #ifdef TCPDEBUG
398 	int ostate;
399 
400 	ostate = tp->t_state;
401 #endif
402 	INP_INFO_RLOCK(&V_tcbinfo);
403 	inp = tp->t_inpcb;
404 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
405 	INP_WLOCK(inp);
406 	if (callout_pending(&tp->t_timers->tt_keep) ||
407 	    !callout_active(&tp->t_timers->tt_keep)) {
408 		INP_WUNLOCK(inp);
409 		INP_INFO_RUNLOCK(&V_tcbinfo);
410 		CURVNET_RESTORE();
411 		return;
412 	}
413 	callout_deactivate(&tp->t_timers->tt_keep);
414 	if ((inp->inp_flags & INP_DROPPED) != 0) {
415 		INP_WUNLOCK(inp);
416 		INP_INFO_RUNLOCK(&V_tcbinfo);
417 		CURVNET_RESTORE();
418 		return;
419 	}
420 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
421 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
422 	KASSERT((tp->t_timers->tt_flags & TT_KEEP) != 0,
423 		("%s: tp %p keep callout should be running", __func__, tp));
424 	/*
425 	 * Keep-alive timer went off; send something
426 	 * or drop connection if idle for too long.
427 	 */
428 	TCPSTAT_INC(tcps_keeptimeo);
429 	if (tp->t_state < TCPS_ESTABLISHED)
430 		goto dropit;
431 	if ((always_keepalive || inp->inp_socket->so_options & SO_KEEPALIVE) &&
432 	    tp->t_state <= TCPS_CLOSING) {
433 		if (ticks - tp->t_rcvtime >= TP_KEEPIDLE(tp) + TP_MAXIDLE(tp))
434 			goto dropit;
435 		/*
436 		 * Send a packet designed to force a response
437 		 * if the peer is up and reachable:
438 		 * either an ACK if the connection is still alive,
439 		 * or an RST if the peer has closed the connection
440 		 * due to timeout or reboot.
441 		 * Using sequence number tp->snd_una-1
442 		 * causes the transmitted zero-length segment
443 		 * to lie outside the receive window;
444 		 * by the protocol spec, this requires the
445 		 * correspondent TCP to respond.
446 		 */
447 		TCPSTAT_INC(tcps_keepprobe);
448 		t_template = tcpip_maketemplate(inp);
449 		if (t_template) {
450 			tcp_respond(tp, t_template->tt_ipgen,
451 				    &t_template->tt_t, (struct mbuf *)NULL,
452 				    tp->rcv_nxt, tp->snd_una - 1, 0);
453 			free(t_template, M_TEMP);
454 		}
455 		if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPINTVL(tp),
456 		    tcp_timer_keep, tp)) {
457 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
458 		}
459 	} else if (!callout_reset(&tp->t_timers->tt_keep, TP_KEEPIDLE(tp),
460 		    tcp_timer_keep, tp)) {
461 			tp->t_timers->tt_flags &= ~TT_KEEP_RST;
462 		}
463 
464 #ifdef TCPDEBUG
465 	if (inp->inp_socket->so_options & SO_DEBUG)
466 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
467 			  PRU_SLOWTIMO);
468 #endif
469 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
470 	INP_WUNLOCK(inp);
471 	INP_INFO_RUNLOCK(&V_tcbinfo);
472 	CURVNET_RESTORE();
473 	return;
474 
475 dropit:
476 	TCPSTAT_INC(tcps_keepdrops);
477 	tp = tcp_drop(tp, ETIMEDOUT);
478 
479 #ifdef TCPDEBUG
480 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
481 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
482 			  PRU_SLOWTIMO);
483 #endif
484 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
485 	if (tp != NULL)
486 		INP_WUNLOCK(tp->t_inpcb);
487 	INP_INFO_RUNLOCK(&V_tcbinfo);
488 	CURVNET_RESTORE();
489 }
490 
491 void
492 tcp_timer_persist(void *xtp)
493 {
494 	struct tcpcb *tp = xtp;
495 	struct inpcb *inp;
496 	CURVNET_SET(tp->t_vnet);
497 #ifdef TCPDEBUG
498 	int ostate;
499 
500 	ostate = tp->t_state;
501 #endif
502 	INP_INFO_RLOCK(&V_tcbinfo);
503 	inp = tp->t_inpcb;
504 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
505 	INP_WLOCK(inp);
506 	if (callout_pending(&tp->t_timers->tt_persist) ||
507 	    !callout_active(&tp->t_timers->tt_persist)) {
508 		INP_WUNLOCK(inp);
509 		INP_INFO_RUNLOCK(&V_tcbinfo);
510 		CURVNET_RESTORE();
511 		return;
512 	}
513 	callout_deactivate(&tp->t_timers->tt_persist);
514 	if ((inp->inp_flags & INP_DROPPED) != 0) {
515 		INP_WUNLOCK(inp);
516 		INP_INFO_RUNLOCK(&V_tcbinfo);
517 		CURVNET_RESTORE();
518 		return;
519 	}
520 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
521 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
522 	KASSERT((tp->t_timers->tt_flags & TT_PERSIST) != 0,
523 		("%s: tp %p persist callout should be running", __func__, tp));
524 	/*
525 	 * Persistance timer into zero window.
526 	 * Force a byte to be output, if possible.
527 	 */
528 	TCPSTAT_INC(tcps_persisttimeo);
529 	/*
530 	 * Hack: if the peer is dead/unreachable, we do not
531 	 * time out if the window is closed.  After a full
532 	 * backoff, drop the connection if the idle time
533 	 * (no responses to probes) reaches the maximum
534 	 * backoff that we would use if retransmitting.
535 	 */
536 	if (tp->t_rxtshift == TCP_MAXRXTSHIFT &&
537 	    (ticks - tp->t_rcvtime >= tcp_maxpersistidle ||
538 	     ticks - tp->t_rcvtime >= TCP_REXMTVAL(tp) * tcp_totbackoff)) {
539 		TCPSTAT_INC(tcps_persistdrop);
540 		tp = tcp_drop(tp, ETIMEDOUT);
541 		goto out;
542 	}
543 	/*
544 	 * If the user has closed the socket then drop a persisting
545 	 * connection after a much reduced timeout.
546 	 */
547 	if (tp->t_state > TCPS_CLOSE_WAIT &&
548 	    (ticks - tp->t_rcvtime) >= TCPTV_PERSMAX) {
549 		TCPSTAT_INC(tcps_persistdrop);
550 		tp = tcp_drop(tp, ETIMEDOUT);
551 		goto out;
552 	}
553 	tcp_setpersist(tp);
554 	tp->t_flags |= TF_FORCEDATA;
555 	(void) tp->t_fb->tfb_tcp_output(tp);
556 	tp->t_flags &= ~TF_FORCEDATA;
557 
558 out:
559 #ifdef TCPDEBUG
560 	if (tp != NULL && tp->t_inpcb->inp_socket->so_options & SO_DEBUG)
561 		tcp_trace(TA_USER, ostate, tp, NULL, NULL, PRU_SLOWTIMO);
562 #endif
563 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
564 	if (tp != NULL)
565 		INP_WUNLOCK(inp);
566 	INP_INFO_RUNLOCK(&V_tcbinfo);
567 	CURVNET_RESTORE();
568 }
569 
570 void
571 tcp_timer_rexmt(void * xtp)
572 {
573 	struct tcpcb *tp = xtp;
574 	CURVNET_SET(tp->t_vnet);
575 	int rexmt;
576 	int headlocked;
577 	struct inpcb *inp;
578 #ifdef TCPDEBUG
579 	int ostate;
580 
581 	ostate = tp->t_state;
582 #endif
583 
584 	INP_INFO_RLOCK(&V_tcbinfo);
585 	inp = tp->t_inpcb;
586 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp));
587 	INP_WLOCK(inp);
588 	if (callout_pending(&tp->t_timers->tt_rexmt) ||
589 	    !callout_active(&tp->t_timers->tt_rexmt)) {
590 		INP_WUNLOCK(inp);
591 		INP_INFO_RUNLOCK(&V_tcbinfo);
592 		CURVNET_RESTORE();
593 		return;
594 	}
595 	callout_deactivate(&tp->t_timers->tt_rexmt);
596 	if ((inp->inp_flags & INP_DROPPED) != 0) {
597 		INP_WUNLOCK(inp);
598 		INP_INFO_RUNLOCK(&V_tcbinfo);
599 		CURVNET_RESTORE();
600 		return;
601 	}
602 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) == 0,
603 		("%s: tp %p tcpcb can't be stopped here", __func__, tp));
604 	KASSERT((tp->t_timers->tt_flags & TT_REXMT) != 0,
605 		("%s: tp %p rexmt callout should be running", __func__, tp));
606 	tcp_free_sackholes(tp);
607 	/*
608 	 * Retransmission timer went off.  Message has not
609 	 * been acked within retransmit interval.  Back off
610 	 * to a longer retransmit interval and retransmit one segment.
611 	 */
612 	if (++tp->t_rxtshift > TCP_MAXRXTSHIFT) {
613 		tp->t_rxtshift = TCP_MAXRXTSHIFT;
614 		TCPSTAT_INC(tcps_timeoutdrop);
615 
616 		tp = tcp_drop(tp, tp->t_softerror ?
617 			      tp->t_softerror : ETIMEDOUT);
618 		headlocked = 1;
619 		goto out;
620 	}
621 	INP_INFO_RUNLOCK(&V_tcbinfo);
622 	headlocked = 0;
623 	if (tp->t_state == TCPS_SYN_SENT) {
624 		/*
625 		 * If the SYN was retransmitted, indicate CWND to be
626 		 * limited to 1 segment in cc_conn_init().
627 		 */
628 		tp->snd_cwnd = 1;
629 	} else if (tp->t_rxtshift == 1) {
630 		/*
631 		 * first retransmit; record ssthresh and cwnd so they can
632 		 * be recovered if this turns out to be a "bad" retransmit.
633 		 * A retransmit is considered "bad" if an ACK for this
634 		 * segment is received within RTT/2 interval; the assumption
635 		 * here is that the ACK was already in flight.  See
636 		 * "On Estimating End-to-End Network Path Properties" by
637 		 * Allman and Paxson for more details.
638 		 */
639 		tp->snd_cwnd_prev = tp->snd_cwnd;
640 		tp->snd_ssthresh_prev = tp->snd_ssthresh;
641 		tp->snd_recover_prev = tp->snd_recover;
642 		if (IN_FASTRECOVERY(tp->t_flags))
643 			tp->t_flags |= TF_WASFRECOVERY;
644 		else
645 			tp->t_flags &= ~TF_WASFRECOVERY;
646 		if (IN_CONGRECOVERY(tp->t_flags))
647 			tp->t_flags |= TF_WASCRECOVERY;
648 		else
649 			tp->t_flags &= ~TF_WASCRECOVERY;
650 		tp->t_badrxtwin = ticks + (tp->t_srtt >> (TCP_RTT_SHIFT + 1));
651 		tp->t_flags |= TF_PREVVALID;
652 	} else
653 		tp->t_flags &= ~TF_PREVVALID;
654 	TCPSTAT_INC(tcps_rexmttimeo);
655 	if ((tp->t_state == TCPS_SYN_SENT) ||
656 	    (tp->t_state == TCPS_SYN_RECEIVED))
657 		rexmt = TCPTV_RTOBASE * tcp_syn_backoff[tp->t_rxtshift];
658 	else
659 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
660 	TCPT_RANGESET(tp->t_rxtcur, rexmt,
661 		      tp->t_rttmin, TCPTV_REXMTMAX);
662 
663 	/*
664 	 * We enter the path for PLMTUD if connection is established or, if
665 	 * connection is FIN_WAIT_1 status, reason for the last is that if
666 	 * amount of data we send is very small, we could send it in couple of
667 	 * packets and process straight to FIN. In that case we won't catch
668 	 * ESTABLISHED state.
669 	 */
670 	if (V_tcp_pmtud_blackhole_detect && (((tp->t_state == TCPS_ESTABLISHED))
671 	    || (tp->t_state == TCPS_FIN_WAIT_1))) {
672 #ifdef INET6
673 		int isipv6;
674 #endif
675 
676 		/*
677 		 * Idea here is that at each stage of mtu probe (usually, 1448
678 		 * -> 1188 -> 524) should be given 2 chances to recover before
679 		 *  further clamping down. 'tp->t_rxtshift % 2 == 0' should
680 		 *  take care of that.
681 		 */
682 		if (((tp->t_flags2 & (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) ==
683 		    (TF2_PLPMTU_PMTUD|TF2_PLPMTU_MAXSEGSNT)) &&
684 		    (tp->t_rxtshift >= 2 && tp->t_rxtshift % 2 == 0)) {
685 			/*
686 			 * Enter Path MTU Black-hole Detection mechanism:
687 			 * - Disable Path MTU Discovery (IP "DF" bit).
688 			 * - Reduce MTU to lower value than what we
689 			 *   negotiated with peer.
690 			 */
691 			/* Record that we may have found a black hole. */
692 			tp->t_flags2 |= TF2_PLPMTU_BLACKHOLE;
693 
694 			/* Keep track of previous MSS. */
695 			tp->t_pmtud_saved_maxseg = tp->t_maxseg;
696 
697 			/*
698 			 * Reduce the MSS to blackhole value or to the default
699 			 * in an attempt to retransmit.
700 			 */
701 #ifdef INET6
702 			isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) ? 1 : 0;
703 			if (isipv6 &&
704 			    tp->t_maxseg > V_tcp_v6pmtud_blackhole_mss) {
705 				/* Use the sysctl tuneable blackhole MSS. */
706 				tp->t_maxseg = V_tcp_v6pmtud_blackhole_mss;
707 				V_tcp_pmtud_blackhole_activated++;
708 			} else if (isipv6) {
709 				/* Use the default MSS. */
710 				tp->t_maxseg = V_tcp_v6mssdflt;
711 				/*
712 				 * Disable Path MTU Discovery when we switch to
713 				 * minmss.
714 				 */
715 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
716 				V_tcp_pmtud_blackhole_activated_min_mss++;
717 			}
718 #endif
719 #if defined(INET6) && defined(INET)
720 			else
721 #endif
722 #ifdef INET
723 			if (tp->t_maxseg > V_tcp_pmtud_blackhole_mss) {
724 				/* Use the sysctl tuneable blackhole MSS. */
725 				tp->t_maxseg = V_tcp_pmtud_blackhole_mss;
726 				V_tcp_pmtud_blackhole_activated++;
727 			} else {
728 				/* Use the default MSS. */
729 				tp->t_maxseg = V_tcp_mssdflt;
730 				/*
731 				 * Disable Path MTU Discovery when we switch to
732 				 * minmss.
733 				 */
734 				tp->t_flags2 &= ~TF2_PLPMTU_PMTUD;
735 				V_tcp_pmtud_blackhole_activated_min_mss++;
736 			}
737 #endif
738 			/*
739 			 * Reset the slow-start flight size
740 			 * as it may depend on the new MSS.
741 			 */
742 			if (CC_ALGO(tp)->conn_init != NULL)
743 				CC_ALGO(tp)->conn_init(tp->ccv);
744 		} else {
745 			/*
746 			 * If further retransmissions are still unsuccessful
747 			 * with a lowered MTU, maybe this isn't a blackhole and
748 			 * we restore the previous MSS and blackhole detection
749 			 * flags.
750 			 * The limit '6' is determined by giving each probe
751 			 * stage (1448, 1188, 524) 2 chances to recover.
752 			 */
753 			if ((tp->t_flags2 & TF2_PLPMTU_BLACKHOLE) &&
754 			    (tp->t_rxtshift > 6)) {
755 				tp->t_flags2 |= TF2_PLPMTU_PMTUD;
756 				tp->t_flags2 &= ~TF2_PLPMTU_BLACKHOLE;
757 				tp->t_maxseg = tp->t_pmtud_saved_maxseg;
758 				V_tcp_pmtud_blackhole_failed++;
759 				/*
760 				 * Reset the slow-start flight size as it
761 				 * may depend on the new MSS.
762 				 */
763 				if (CC_ALGO(tp)->conn_init != NULL)
764 					CC_ALGO(tp)->conn_init(tp->ccv);
765 			}
766 		}
767 	}
768 
769 	/*
770 	 * Disable RFC1323 and SACK if we haven't got any response to
771 	 * our third SYN to work-around some broken terminal servers
772 	 * (most of which have hopefully been retired) that have bad VJ
773 	 * header compression code which trashes TCP segments containing
774 	 * unknown-to-them TCP options.
775 	 */
776 	if (tcp_rexmit_drop_options && (tp->t_state == TCPS_SYN_SENT) &&
777 	    (tp->t_rxtshift == 3))
778 		tp->t_flags &= ~(TF_REQ_SCALE|TF_REQ_TSTMP|TF_SACK_PERMIT);
779 	/*
780 	 * If we backed off this far, our srtt estimate is probably bogus.
781 	 * Clobber it so we'll take the next rtt measurement as our srtt;
782 	 * move the current srtt into rttvar to keep the current
783 	 * retransmit times until then.
784 	 */
785 	if (tp->t_rxtshift > TCP_MAXRXTSHIFT / 4) {
786 #ifdef INET6
787 		if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0)
788 			in6_losing(tp->t_inpcb);
789 		else
790 #endif
791 			in_losing(tp->t_inpcb);
792 		tp->t_rttvar += (tp->t_srtt >> TCP_RTT_SHIFT);
793 		tp->t_srtt = 0;
794 	}
795 	tp->snd_nxt = tp->snd_una;
796 	tp->snd_recover = tp->snd_max;
797 	/*
798 	 * Force a segment to be sent.
799 	 */
800 	tp->t_flags |= TF_ACKNOW;
801 	/*
802 	 * If timing a segment in this window, stop the timer.
803 	 */
804 	tp->t_rtttime = 0;
805 
806 	cc_cong_signal(tp, NULL, CC_RTO);
807 
808 	(void) tp->t_fb->tfb_tcp_output(tp);
809 
810 out:
811 #ifdef TCPDEBUG
812 	if (tp != NULL && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
813 		tcp_trace(TA_USER, ostate, tp, (void *)0, (struct tcphdr *)0,
814 			  PRU_SLOWTIMO);
815 #endif
816 	TCP_PROBE2(debug__user, tp, PRU_SLOWTIMO);
817 	if (tp != NULL)
818 		INP_WUNLOCK(inp);
819 	if (headlocked)
820 		INP_INFO_RUNLOCK(&V_tcbinfo);
821 	CURVNET_RESTORE();
822 }
823 
824 void
825 tcp_timer_activate(struct tcpcb *tp, uint32_t timer_type, u_int delta)
826 {
827 	struct callout *t_callout;
828 	timeout_t *f_callout;
829 	struct inpcb *inp = tp->t_inpcb;
830 	int cpu = inp_to_cpuid(inp);
831 	uint32_t f_reset;
832 
833 #ifdef TCP_OFFLOAD
834 	if (tp->t_flags & TF_TOE)
835 		return;
836 #endif
837 
838 	if (tp->t_timers->tt_flags & TT_STOPPED)
839 		return;
840 
841 	switch (timer_type) {
842 		case TT_DELACK:
843 			t_callout = &tp->t_timers->tt_delack;
844 			f_callout = tcp_timer_delack;
845 			f_reset = TT_DELACK_RST;
846 			break;
847 		case TT_REXMT:
848 			t_callout = &tp->t_timers->tt_rexmt;
849 			f_callout = tcp_timer_rexmt;
850 			f_reset = TT_REXMT_RST;
851 			break;
852 		case TT_PERSIST:
853 			t_callout = &tp->t_timers->tt_persist;
854 			f_callout = tcp_timer_persist;
855 			f_reset = TT_PERSIST_RST;
856 			break;
857 		case TT_KEEP:
858 			t_callout = &tp->t_timers->tt_keep;
859 			f_callout = tcp_timer_keep;
860 			f_reset = TT_KEEP_RST;
861 			break;
862 		case TT_2MSL:
863 			t_callout = &tp->t_timers->tt_2msl;
864 			f_callout = tcp_timer_2msl;
865 			f_reset = TT_2MSL_RST;
866 			break;
867 		default:
868 			if (tp->t_fb->tfb_tcp_timer_activate) {
869 				tp->t_fb->tfb_tcp_timer_activate(tp, timer_type, delta);
870 				return;
871 			}
872 			panic("tp %p bad timer_type %#x", tp, timer_type);
873 		}
874 	if (delta == 0) {
875 		if ((tp->t_timers->tt_flags & timer_type) &&
876 		    (callout_stop(t_callout) > 0) &&
877 		    (tp->t_timers->tt_flags & f_reset)) {
878 			tp->t_timers->tt_flags &= ~(timer_type | f_reset);
879 		}
880 	} else {
881 		if ((tp->t_timers->tt_flags & timer_type) == 0) {
882 			tp->t_timers->tt_flags |= (timer_type | f_reset);
883 			callout_reset_on(t_callout, delta, f_callout, tp, cpu);
884 		} else {
885 			/* Reset already running callout on the same CPU. */
886 			if (!callout_reset(t_callout, delta, f_callout, tp)) {
887 				/*
888 				 * Callout not cancelled, consider it as not
889 				 * properly restarted. */
890 				tp->t_timers->tt_flags &= ~f_reset;
891 			}
892 		}
893 	}
894 }
895 
896 int
897 tcp_timer_active(struct tcpcb *tp, uint32_t timer_type)
898 {
899 	struct callout *t_callout;
900 
901 	switch (timer_type) {
902 		case TT_DELACK:
903 			t_callout = &tp->t_timers->tt_delack;
904 			break;
905 		case TT_REXMT:
906 			t_callout = &tp->t_timers->tt_rexmt;
907 			break;
908 		case TT_PERSIST:
909 			t_callout = &tp->t_timers->tt_persist;
910 			break;
911 		case TT_KEEP:
912 			t_callout = &tp->t_timers->tt_keep;
913 			break;
914 		case TT_2MSL:
915 			t_callout = &tp->t_timers->tt_2msl;
916 			break;
917 		default:
918 			if (tp->t_fb->tfb_tcp_timer_active) {
919 				return(tp->t_fb->tfb_tcp_timer_active(tp, timer_type));
920 			}
921 			panic("tp %p bad timer_type %#x", tp, timer_type);
922 		}
923 	return callout_active(t_callout);
924 }
925 
926 void
927 tcp_timer_stop(struct tcpcb *tp, uint32_t timer_type)
928 {
929 	struct callout *t_callout;
930 	uint32_t f_reset;
931 
932 	tp->t_timers->tt_flags |= TT_STOPPED;
933 
934 	switch (timer_type) {
935 		case TT_DELACK:
936 			t_callout = &tp->t_timers->tt_delack;
937 			f_reset = TT_DELACK_RST;
938 			break;
939 		case TT_REXMT:
940 			t_callout = &tp->t_timers->tt_rexmt;
941 			f_reset = TT_REXMT_RST;
942 			break;
943 		case TT_PERSIST:
944 			t_callout = &tp->t_timers->tt_persist;
945 			f_reset = TT_PERSIST_RST;
946 			break;
947 		case TT_KEEP:
948 			t_callout = &tp->t_timers->tt_keep;
949 			f_reset = TT_KEEP_RST;
950 			break;
951 		case TT_2MSL:
952 			t_callout = &tp->t_timers->tt_2msl;
953 			f_reset = TT_2MSL_RST;
954 			break;
955 		default:
956 			if (tp->t_fb->tfb_tcp_timer_stop) {
957 				/*
958 				 * XXXrrs we need to look at this with the
959 				 * stop case below (flags).
960 				 */
961 				tp->t_fb->tfb_tcp_timer_stop(tp, timer_type);
962 				return;
963 			}
964 			panic("tp %p bad timer_type %#x", tp, timer_type);
965 		}
966 
967 	if (tp->t_timers->tt_flags & timer_type) {
968 		if (callout_async_drain(t_callout, tcp_timer_discard) == 0) {
969 			/*
970 			 * Can't stop the callout, defer tcpcb actual deletion
971 			 * to the last one. We do this using the async drain
972 			 * function and incrementing the count in
973 			 */
974 			tp->t_timers->tt_draincnt++;
975 		}
976 	}
977 }
978 
979 #define	ticks_to_msecs(t)	(1000*(t) / hz)
980 
981 void
982 tcp_timer_to_xtimer(struct tcpcb *tp, struct tcp_timer *timer,
983     struct xtcp_timer *xtimer)
984 {
985 	sbintime_t now;
986 
987 	bzero(xtimer, sizeof(*xtimer));
988 	if (timer == NULL)
989 		return;
990 	now = getsbinuptime();
991 	if (callout_active(&timer->tt_delack))
992 		xtimer->tt_delack = (timer->tt_delack.c_time - now) / SBT_1MS;
993 	if (callout_active(&timer->tt_rexmt))
994 		xtimer->tt_rexmt = (timer->tt_rexmt.c_time - now) / SBT_1MS;
995 	if (callout_active(&timer->tt_persist))
996 		xtimer->tt_persist = (timer->tt_persist.c_time - now) / SBT_1MS;
997 	if (callout_active(&timer->tt_keep))
998 		xtimer->tt_keep = (timer->tt_keep.c_time - now) / SBT_1MS;
999 	if (callout_active(&timer->tt_2msl))
1000 		xtimer->tt_2msl = (timer->tt_2msl.c_time - now) / SBT_1MS;
1001 	xtimer->t_rcvtime = ticks_to_msecs(ticks - tp->t_rcvtime);
1002 }
1003