xref: /freebsd/sys/netinet/tcp_subr.c (revision ce6a89e27cd190313be39bb479880aeda4778436)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/if.h>
80 #include <net/if_var.h>
81 #include <net/vnet.h>
82 
83 #include <netinet/in.h>
84 #include <netinet/in_fib.h>
85 #include <netinet/in_kdtrace.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/ip_icmp.h>
91 #include <netinet/ip_var.h>
92 #ifdef INET6
93 #include <netinet/icmp6.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/in6_fib.h>
96 #include <netinet6/in6_pcb.h>
97 #include <netinet6/ip6_var.h>
98 #include <netinet6/scope6_var.h>
99 #include <netinet6/nd6.h>
100 #endif
101 
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcp_log_buf.h>
108 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_hpts.h>
110 #include <netinet/cc/cc.h>
111 #ifdef INET6
112 #include <netinet6/tcp6_var.h>
113 #endif
114 #include <netinet/tcpip.h>
115 #include <netinet/tcp_fastopen.h>
116 #ifdef TCPPCAP
117 #include <netinet/tcp_pcap.h>
118 #endif
119 #ifdef TCPDEBUG
120 #include <netinet/tcp_debug.h>
121 #endif
122 #ifdef INET6
123 #include <netinet6/ip6protosw.h>
124 #endif
125 #ifdef TCP_OFFLOAD
126 #include <netinet/tcp_offload.h>
127 #endif
128 
129 #include <netipsec/ipsec_support.h>
130 
131 #include <machine/in_cksum.h>
132 #include <crypto/siphash/siphash.h>
133 
134 #include <security/mac/mac_framework.h>
135 
136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
137 #ifdef INET6
138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
139 #endif
140 
141 #ifdef NETFLIX_EXP_DETECTION
142 /*  Sack attack detection thresholds and such */
143 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
144     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
145     "Sack Attack detection thresholds");
146 int32_t tcp_force_detection = 0;
147 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
148     CTLFLAG_RW,
149     &tcp_force_detection, 0,
150     "Do we force detection even if the INP has it off?");
151 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
152 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
153     CTLFLAG_RW,
154     &tcp_sack_to_ack_thresh, 700,
155     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
156 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
157 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
158     CTLFLAG_RW,
159     &tcp_sack_to_move_thresh, 600,
160     "Percentage of sack moves we must see above (10.1 percent is 101)");
161 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
162 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
163     CTLFLAG_RW,
164     &tcp_restoral_thresh, 550,
165     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
166 int32_t tcp_sad_decay_val = 800;
167 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
168     CTLFLAG_RW,
169     &tcp_sad_decay_val, 800,
170     "The decay percentage (10.1 percent equals 101 )");
171 int32_t tcp_map_minimum = 500;
172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
173     CTLFLAG_RW,
174     &tcp_map_minimum, 500,
175     "Number of Map enteries before we start detection");
176 int32_t tcp_attack_on_turns_on_logging = 0;
177 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
178     CTLFLAG_RW,
179     &tcp_attack_on_turns_on_logging, 0,
180    "When we have a positive hit on attack, do we turn on logging?");
181 int32_t tcp_sad_pacing_interval = 2000;
182 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
183     CTLFLAG_RW,
184     &tcp_sad_pacing_interval, 2000,
185     "What is the minimum pacing interval for a classified attacker?");
186 
187 int32_t tcp_sad_low_pps = 100;
188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
189     CTLFLAG_RW,
190     &tcp_sad_low_pps, 100,
191     "What is the input pps that below which we do not decay?");
192 #endif
193 
194 struct rwlock tcp_function_lock;
195 
196 static int
197 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
198 {
199 	int error, new;
200 
201 	new = V_tcp_mssdflt;
202 	error = sysctl_handle_int(oidp, &new, 0, req);
203 	if (error == 0 && req->newptr) {
204 		if (new < TCP_MINMSS)
205 			error = EINVAL;
206 		else
207 			V_tcp_mssdflt = new;
208 	}
209 	return (error);
210 }
211 
212 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
213     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
214     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
215     "Default TCP Maximum Segment Size");
216 
217 #ifdef INET6
218 static int
219 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
220 {
221 	int error, new;
222 
223 	new = V_tcp_v6mssdflt;
224 	error = sysctl_handle_int(oidp, &new, 0, req);
225 	if (error == 0 && req->newptr) {
226 		if (new < TCP_MINMSS)
227 			error = EINVAL;
228 		else
229 			V_tcp_v6mssdflt = new;
230 	}
231 	return (error);
232 }
233 
234 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
235     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
236     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
237    "Default TCP Maximum Segment Size for IPv6");
238 #endif /* INET6 */
239 
240 /*
241  * Minimum MSS we accept and use. This prevents DoS attacks where
242  * we are forced to a ridiculous low MSS like 20 and send hundreds
243  * of packets instead of one. The effect scales with the available
244  * bandwidth and quickly saturates the CPU and network interface
245  * with packet generation and sending. Set to zero to disable MINMSS
246  * checking. This setting prevents us from sending too small packets.
247  */
248 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
249 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
250      &VNET_NAME(tcp_minmss), 0,
251     "Minimum TCP Maximum Segment Size");
252 
253 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
254 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
255     &VNET_NAME(tcp_do_rfc1323), 0,
256     "Enable rfc1323 (high performance TCP) extensions");
257 
258 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
259 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
260     &VNET_NAME(tcp_ts_offset_per_conn), 0,
261     "Initialize TCP timestamps per connection instead of per host pair");
262 
263 static int	tcp_log_debug = 0;
264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
265     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
266 
267 static int	tcp_tcbhashsize;
268 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
269     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
270 
271 static int	do_tcpdrain = 1;
272 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
273     "Enable tcp_drain routine for extra help when low on mbufs");
274 
275 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
276     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
277 
278 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
279 #define	V_icmp_may_rst			VNET(icmp_may_rst)
280 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
281     &VNET_NAME(icmp_may_rst), 0,
282     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
283 
284 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
285 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
287     &VNET_NAME(tcp_isn_reseed_interval), 0,
288     "Seconds between reseeding of ISN secret");
289 
290 static int	tcp_soreceive_stream;
291 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
292     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
293 
294 VNET_DEFINE(uma_zone_t, sack_hole_zone);
295 #define	V_sack_hole_zone		VNET(sack_hole_zone)
296 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
297 static int
298 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
299 {
300 	int error;
301 	uint32_t new;
302 
303 	new = V_tcp_map_entries_limit;
304 	error = sysctl_handle_int(oidp, &new, 0, req);
305 	if (error == 0 && req->newptr) {
306 		/* only allow "0" and value > minimum */
307 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
308 			error = EINVAL;
309 		else
310 			V_tcp_map_entries_limit = new;
311 	}
312 	return (error);
313 }
314 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
315     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
316     &VNET_NAME(tcp_map_entries_limit), 0,
317     &sysctl_net_inet_tcp_map_limit_check, "IU",
318     "Total sendmap entries limit");
319 
320 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
321 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
322      &VNET_NAME(tcp_map_split_limit), 0,
323     "Total sendmap split entries limit");
324 
325 #ifdef TCP_HHOOK
326 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
327 #endif
328 
329 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
330 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
331 #define	V_ts_offset_secret	VNET(ts_offset_secret)
332 
333 static int	tcp_default_fb_init(struct tcpcb *tp);
334 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
335 static int	tcp_default_handoff_ok(struct tcpcb *tp);
336 static struct inpcb *tcp_notify(struct inpcb *, int);
337 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
338 static void tcp_mtudisc(struct inpcb *, int);
339 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
340 		    void *ip4hdr, const void *ip6hdr);
341 
342 
343 static struct tcp_function_block tcp_def_funcblk = {
344 	.tfb_tcp_block_name = "freebsd",
345 	.tfb_tcp_output = tcp_output,
346 	.tfb_tcp_do_segment = tcp_do_segment,
347 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
348 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
349 	.tfb_tcp_fb_init = tcp_default_fb_init,
350 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
351 };
352 
353 static int tcp_fb_cnt = 0;
354 struct tcp_funchead t_functions;
355 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
356 
357 static struct tcp_function_block *
358 find_tcp_functions_locked(struct tcp_function_set *fs)
359 {
360 	struct tcp_function *f;
361 	struct tcp_function_block *blk=NULL;
362 
363 	TAILQ_FOREACH(f, &t_functions, tf_next) {
364 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
365 			blk = f->tf_fb;
366 			break;
367 		}
368 	}
369 	return(blk);
370 }
371 
372 static struct tcp_function_block *
373 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
374 {
375 	struct tcp_function_block *rblk=NULL;
376 	struct tcp_function *f;
377 
378 	TAILQ_FOREACH(f, &t_functions, tf_next) {
379 		if (f->tf_fb == blk) {
380 			rblk = blk;
381 			if (s) {
382 				*s = f;
383 			}
384 			break;
385 		}
386 	}
387 	return (rblk);
388 }
389 
390 struct tcp_function_block *
391 find_and_ref_tcp_functions(struct tcp_function_set *fs)
392 {
393 	struct tcp_function_block *blk;
394 
395 	rw_rlock(&tcp_function_lock);
396 	blk = find_tcp_functions_locked(fs);
397 	if (blk)
398 		refcount_acquire(&blk->tfb_refcnt);
399 	rw_runlock(&tcp_function_lock);
400 	return(blk);
401 }
402 
403 struct tcp_function_block *
404 find_and_ref_tcp_fb(struct tcp_function_block *blk)
405 {
406 	struct tcp_function_block *rblk;
407 
408 	rw_rlock(&tcp_function_lock);
409 	rblk = find_tcp_fb_locked(blk, NULL);
410 	if (rblk)
411 		refcount_acquire(&rblk->tfb_refcnt);
412 	rw_runlock(&tcp_function_lock);
413 	return(rblk);
414 }
415 
416 static struct tcp_function_block *
417 find_and_ref_tcp_default_fb(void)
418 {
419 	struct tcp_function_block *rblk;
420 
421 	rw_rlock(&tcp_function_lock);
422 	rblk = tcp_func_set_ptr;
423 	refcount_acquire(&rblk->tfb_refcnt);
424 	rw_runlock(&tcp_function_lock);
425 	return (rblk);
426 }
427 
428 void
429 tcp_switch_back_to_default(struct tcpcb *tp)
430 {
431 	struct tcp_function_block *tfb;
432 
433 	KASSERT(tp->t_fb != &tcp_def_funcblk,
434 	    ("%s: called by the built-in default stack", __func__));
435 
436 	/*
437 	 * Release the old stack. This function will either find a new one
438 	 * or panic.
439 	 */
440 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
441 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
442 	refcount_release(&tp->t_fb->tfb_refcnt);
443 
444 	/*
445 	 * Now, we'll find a new function block to use.
446 	 * Start by trying the current user-selected
447 	 * default, unless this stack is the user-selected
448 	 * default.
449 	 */
450 	tfb = find_and_ref_tcp_default_fb();
451 	if (tfb == tp->t_fb) {
452 		refcount_release(&tfb->tfb_refcnt);
453 		tfb = NULL;
454 	}
455 	/* Does the stack accept this connection? */
456 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
457 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
458 		refcount_release(&tfb->tfb_refcnt);
459 		tfb = NULL;
460 	}
461 	/* Try to use that stack. */
462 	if (tfb != NULL) {
463 		/* Initialize the new stack. If it succeeds, we are done. */
464 		tp->t_fb = tfb;
465 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
466 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
467 			return;
468 
469 		/*
470 		 * Initialization failed. Release the reference count on
471 		 * the stack.
472 		 */
473 		refcount_release(&tfb->tfb_refcnt);
474 	}
475 
476 	/*
477 	 * If that wasn't feasible, use the built-in default
478 	 * stack which is not allowed to reject anyone.
479 	 */
480 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
481 	if (tfb == NULL) {
482 		/* there always should be a default */
483 		panic("Can't refer to tcp_def_funcblk");
484 	}
485 	if (tfb->tfb_tcp_handoff_ok != NULL) {
486 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
487 			/* The default stack cannot say no */
488 			panic("Default stack rejects a new session?");
489 		}
490 	}
491 	tp->t_fb = tfb;
492 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
493 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
494 		/* The default stack cannot fail */
495 		panic("Default stack initialization failed");
496 	}
497 }
498 
499 static int
500 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
501 {
502 	int error=ENOENT;
503 	struct tcp_function_set fs;
504 	struct tcp_function_block *blk;
505 
506 	memset(&fs, 0, sizeof(fs));
507 	rw_rlock(&tcp_function_lock);
508 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
509 	if (blk) {
510 		/* Found him */
511 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
512 		fs.pcbcnt = blk->tfb_refcnt;
513 	}
514 	rw_runlock(&tcp_function_lock);
515 	error = sysctl_handle_string(oidp, fs.function_set_name,
516 				     sizeof(fs.function_set_name), req);
517 
518 	/* Check for error or no change */
519 	if (error != 0 || req->newptr == NULL)
520 		return(error);
521 
522 	rw_wlock(&tcp_function_lock);
523 	blk = find_tcp_functions_locked(&fs);
524 	if ((blk == NULL) ||
525 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
526 		error = ENOENT;
527 		goto done;
528 	}
529 	tcp_func_set_ptr = blk;
530 done:
531 	rw_wunlock(&tcp_function_lock);
532 	return (error);
533 }
534 
535 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
536     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
537     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
538     "Set/get the default TCP functions");
539 
540 static int
541 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
542 {
543 	int error, cnt, linesz;
544 	struct tcp_function *f;
545 	char *buffer, *cp;
546 	size_t bufsz, outsz;
547 	bool alias;
548 
549 	cnt = 0;
550 	rw_rlock(&tcp_function_lock);
551 	TAILQ_FOREACH(f, &t_functions, tf_next) {
552 		cnt++;
553 	}
554 	rw_runlock(&tcp_function_lock);
555 
556 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
557 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
558 
559 	error = 0;
560 	cp = buffer;
561 
562 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
563 	    "Alias", "PCB count");
564 	cp += linesz;
565 	bufsz -= linesz;
566 	outsz = linesz;
567 
568 	rw_rlock(&tcp_function_lock);
569 	TAILQ_FOREACH(f, &t_functions, tf_next) {
570 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
571 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
572 		    f->tf_fb->tfb_tcp_block_name,
573 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
574 		    alias ? f->tf_name : "-",
575 		    f->tf_fb->tfb_refcnt);
576 		if (linesz >= bufsz) {
577 			error = EOVERFLOW;
578 			break;
579 		}
580 		cp += linesz;
581 		bufsz -= linesz;
582 		outsz += linesz;
583 	}
584 	rw_runlock(&tcp_function_lock);
585 	if (error == 0)
586 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
587 	free(buffer, M_TEMP);
588 	return (error);
589 }
590 
591 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
592     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
593     NULL, 0, sysctl_net_inet_list_available, "A",
594     "list available TCP Function sets");
595 
596 /*
597  * Exports one (struct tcp_function_info) for each alias/name.
598  */
599 static int
600 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
601 {
602 	int cnt, error;
603 	struct tcp_function *f;
604 	struct tcp_function_info tfi;
605 
606 	/*
607 	 * We don't allow writes.
608 	 */
609 	if (req->newptr != NULL)
610 		return (EINVAL);
611 
612 	/*
613 	 * Wire the old buffer so we can directly copy the functions to
614 	 * user space without dropping the lock.
615 	 */
616 	if (req->oldptr != NULL) {
617 		error = sysctl_wire_old_buffer(req, 0);
618 		if (error)
619 			return (error);
620 	}
621 
622 	/*
623 	 * Walk the list and copy out matching entries. If INVARIANTS
624 	 * is compiled in, also walk the list to verify the length of
625 	 * the list matches what we have recorded.
626 	 */
627 	rw_rlock(&tcp_function_lock);
628 
629 	cnt = 0;
630 #ifndef INVARIANTS
631 	if (req->oldptr == NULL) {
632 		cnt = tcp_fb_cnt;
633 		goto skip_loop;
634 	}
635 #endif
636 	TAILQ_FOREACH(f, &t_functions, tf_next) {
637 #ifdef INVARIANTS
638 		cnt++;
639 #endif
640 		if (req->oldptr != NULL) {
641 			bzero(&tfi, sizeof(tfi));
642 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
643 			tfi.tfi_id = f->tf_fb->tfb_id;
644 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
645 			    sizeof(tfi.tfi_alias));
646 			(void)strlcpy(tfi.tfi_name,
647 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
648 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
649 			/*
650 			 * Don't stop on error, as that is the
651 			 * mechanism we use to accumulate length
652 			 * information if the buffer was too short.
653 			 */
654 		}
655 	}
656 	KASSERT(cnt == tcp_fb_cnt,
657 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
658 #ifndef INVARIANTS
659 skip_loop:
660 #endif
661 	rw_runlock(&tcp_function_lock);
662 	if (req->oldptr == NULL)
663 		error = SYSCTL_OUT(req, NULL,
664 		    (cnt + 1) * sizeof(struct tcp_function_info));
665 
666 	return (error);
667 }
668 
669 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
670 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
671 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
672 	    "List TCP function block name-to-ID mappings");
673 
674 /*
675  * tfb_tcp_handoff_ok() function for the default stack.
676  * Note that we'll basically try to take all comers.
677  */
678 static int
679 tcp_default_handoff_ok(struct tcpcb *tp)
680 {
681 
682 	return (0);
683 }
684 
685 /*
686  * tfb_tcp_fb_init() function for the default stack.
687  *
688  * This handles making sure we have appropriate timers set if you are
689  * transitioning a socket that has some amount of setup done.
690  *
691  * The init() fuction from the default can *never* return non-zero i.e.
692  * it is required to always succeed since it is the stack of last resort!
693  */
694 static int
695 tcp_default_fb_init(struct tcpcb *tp)
696 {
697 
698 	struct socket *so;
699 
700 	INP_WLOCK_ASSERT(tp->t_inpcb);
701 
702 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
703 	    ("%s: connection %p in unexpected state %d", __func__, tp,
704 	    tp->t_state));
705 
706 	/*
707 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
708 	 * know what to do for unexpected states (which includes TIME_WAIT).
709 	 */
710 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
711 		return (0);
712 
713 	/*
714 	 * Make sure some kind of transmission timer is set if there is
715 	 * outstanding data.
716 	 */
717 	so = tp->t_inpcb->inp_socket;
718 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
719 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
720 	    tcp_timer_active(tp, TT_PERSIST))) {
721 		/*
722 		 * If the session has established and it looks like it should
723 		 * be in the persist state, set the persist timer. Otherwise,
724 		 * set the retransmit timer.
725 		 */
726 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
727 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
728 		    (int32_t)sbavail(&so->so_snd))
729 			tcp_setpersist(tp);
730 		else
731 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
732 	}
733 
734 	/* All non-embryonic sessions get a keepalive timer. */
735 	if (!tcp_timer_active(tp, TT_KEEP))
736 		tcp_timer_activate(tp, TT_KEEP,
737 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
738 		    TP_KEEPINIT(tp));
739 
740 	return (0);
741 }
742 
743 /*
744  * tfb_tcp_fb_fini() function for the default stack.
745  *
746  * This changes state as necessary (or prudent) to prepare for another stack
747  * to assume responsibility for the connection.
748  */
749 static void
750 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
751 {
752 
753 	INP_WLOCK_ASSERT(tp->t_inpcb);
754 	return;
755 }
756 
757 /*
758  * Target size of TCP PCB hash tables. Must be a power of two.
759  *
760  * Note that this can be overridden by the kernel environment
761  * variable net.inet.tcp.tcbhashsize
762  */
763 #ifndef TCBHASHSIZE
764 #define TCBHASHSIZE	0
765 #endif
766 
767 /*
768  * XXX
769  * Callouts should be moved into struct tcp directly.  They are currently
770  * separate because the tcpcb structure is exported to userland for sysctl
771  * parsing purposes, which do not know about callouts.
772  */
773 struct tcpcb_mem {
774 	struct	tcpcb		tcb;
775 	struct	tcp_timer	tt;
776 	struct	cc_var		ccv;
777 #ifdef TCP_HHOOK
778 	struct	osd		osd;
779 #endif
780 };
781 
782 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
783 #define	V_tcpcb_zone			VNET(tcpcb_zone)
784 
785 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
786 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
787 
788 static struct mtx isn_mtx;
789 
790 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
791 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
792 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
793 
794 /*
795  * TCP initialization.
796  */
797 static void
798 tcp_zone_change(void *tag)
799 {
800 
801 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
802 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
803 	tcp_tw_zone_change();
804 }
805 
806 static int
807 tcp_inpcb_init(void *mem, int size, int flags)
808 {
809 	struct inpcb *inp = mem;
810 
811 	INP_LOCK_INIT(inp, "inp", "tcpinp");
812 	return (0);
813 }
814 
815 /*
816  * Take a value and get the next power of 2 that doesn't overflow.
817  * Used to size the tcp_inpcb hash buckets.
818  */
819 static int
820 maketcp_hashsize(int size)
821 {
822 	int hashsize;
823 
824 	/*
825 	 * auto tune.
826 	 * get the next power of 2 higher than maxsockets.
827 	 */
828 	hashsize = 1 << fls(size);
829 	/* catch overflow, and just go one power of 2 smaller */
830 	if (hashsize < size) {
831 		hashsize = 1 << (fls(size) - 1);
832 	}
833 	return (hashsize);
834 }
835 
836 static volatile int next_tcp_stack_id = 1;
837 
838 /*
839  * Register a TCP function block with the name provided in the names
840  * array.  (Note that this function does NOT automatically register
841  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
842  * explicitly include blk->tfb_tcp_block_name in the list of names if
843  * you wish to register the stack with that name.)
844  *
845  * Either all name registrations will succeed or all will fail.  If
846  * a name registration fails, the function will update the num_names
847  * argument to point to the array index of the name that encountered
848  * the failure.
849  *
850  * Returns 0 on success, or an error code on failure.
851  */
852 int
853 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
854     const char *names[], int *num_names)
855 {
856 	struct tcp_function *n;
857 	struct tcp_function_set fs;
858 	int error, i;
859 
860 	KASSERT(names != NULL && *num_names > 0,
861 	    ("%s: Called with 0-length name list", __func__));
862 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
863 	KASSERT(rw_initialized(&tcp_function_lock),
864 	    ("%s: called too early", __func__));
865 
866 	if ((blk->tfb_tcp_output == NULL) ||
867 	    (blk->tfb_tcp_do_segment == NULL) ||
868 	    (blk->tfb_tcp_ctloutput == NULL) ||
869 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
870 		/*
871 		 * These functions are required and you
872 		 * need a name.
873 		 */
874 		*num_names = 0;
875 		return (EINVAL);
876 	}
877 	if (blk->tfb_tcp_timer_stop_all ||
878 	    blk->tfb_tcp_timer_activate ||
879 	    blk->tfb_tcp_timer_active ||
880 	    blk->tfb_tcp_timer_stop) {
881 		/*
882 		 * If you define one timer function you
883 		 * must have them all.
884 		 */
885 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
886 		    (blk->tfb_tcp_timer_activate == NULL) ||
887 		    (blk->tfb_tcp_timer_active == NULL) ||
888 		    (blk->tfb_tcp_timer_stop == NULL)) {
889 			*num_names = 0;
890 			return (EINVAL);
891 		}
892 	}
893 
894 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
895 		*num_names = 0;
896 		return (EINVAL);
897 	}
898 
899 	refcount_init(&blk->tfb_refcnt, 0);
900 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
901 	for (i = 0; i < *num_names; i++) {
902 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
903 		if (n == NULL) {
904 			error = ENOMEM;
905 			goto cleanup;
906 		}
907 		n->tf_fb = blk;
908 
909 		(void)strlcpy(fs.function_set_name, names[i],
910 		    sizeof(fs.function_set_name));
911 		rw_wlock(&tcp_function_lock);
912 		if (find_tcp_functions_locked(&fs) != NULL) {
913 			/* Duplicate name space not allowed */
914 			rw_wunlock(&tcp_function_lock);
915 			free(n, M_TCPFUNCTIONS);
916 			error = EALREADY;
917 			goto cleanup;
918 		}
919 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
920 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
921 		tcp_fb_cnt++;
922 		rw_wunlock(&tcp_function_lock);
923 	}
924 	return(0);
925 
926 cleanup:
927 	/*
928 	 * Deregister the names we just added. Because registration failed
929 	 * for names[i], we don't need to deregister that name.
930 	 */
931 	*num_names = i;
932 	rw_wlock(&tcp_function_lock);
933 	while (--i >= 0) {
934 		TAILQ_FOREACH(n, &t_functions, tf_next) {
935 			if (!strncmp(n->tf_name, names[i],
936 			    TCP_FUNCTION_NAME_LEN_MAX)) {
937 				TAILQ_REMOVE(&t_functions, n, tf_next);
938 				tcp_fb_cnt--;
939 				n->tf_fb = NULL;
940 				free(n, M_TCPFUNCTIONS);
941 				break;
942 			}
943 		}
944 	}
945 	rw_wunlock(&tcp_function_lock);
946 	return (error);
947 }
948 
949 /*
950  * Register a TCP function block using the name provided in the name
951  * argument.
952  *
953  * Returns 0 on success, or an error code on failure.
954  */
955 int
956 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
957     int wait)
958 {
959 	const char *name_list[1];
960 	int num_names, rv;
961 
962 	num_names = 1;
963 	if (name != NULL)
964 		name_list[0] = name;
965 	else
966 		name_list[0] = blk->tfb_tcp_block_name;
967 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
968 	return (rv);
969 }
970 
971 /*
972  * Register a TCP function block using the name defined in
973  * blk->tfb_tcp_block_name.
974  *
975  * Returns 0 on success, or an error code on failure.
976  */
977 int
978 register_tcp_functions(struct tcp_function_block *blk, int wait)
979 {
980 
981 	return (register_tcp_functions_as_name(blk, NULL, wait));
982 }
983 
984 /*
985  * Deregister all names associated with a function block. This
986  * functionally removes the function block from use within the system.
987  *
988  * When called with a true quiesce argument, mark the function block
989  * as being removed so no more stacks will use it and determine
990  * whether the removal would succeed.
991  *
992  * When called with a false quiesce argument, actually attempt the
993  * removal.
994  *
995  * When called with a force argument, attempt to switch all TCBs to
996  * use the default stack instead of returning EBUSY.
997  *
998  * Returns 0 on success (or if the removal would succeed, or an error
999  * code on failure.
1000  */
1001 int
1002 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1003     bool force)
1004 {
1005 	struct tcp_function *f;
1006 
1007 	if (blk == &tcp_def_funcblk) {
1008 		/* You can't un-register the default */
1009 		return (EPERM);
1010 	}
1011 	rw_wlock(&tcp_function_lock);
1012 	if (blk == tcp_func_set_ptr) {
1013 		/* You can't free the current default */
1014 		rw_wunlock(&tcp_function_lock);
1015 		return (EBUSY);
1016 	}
1017 	/* Mark the block so no more stacks can use it. */
1018 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1019 	/*
1020 	 * If TCBs are still attached to the stack, attempt to switch them
1021 	 * to the default stack.
1022 	 */
1023 	if (force && blk->tfb_refcnt) {
1024 		struct inpcb *inp;
1025 		struct tcpcb *tp;
1026 		VNET_ITERATOR_DECL(vnet_iter);
1027 
1028 		rw_wunlock(&tcp_function_lock);
1029 
1030 		VNET_LIST_RLOCK();
1031 		VNET_FOREACH(vnet_iter) {
1032 			CURVNET_SET(vnet_iter);
1033 			INP_INFO_WLOCK(&V_tcbinfo);
1034 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1035 				INP_WLOCK(inp);
1036 				if (inp->inp_flags & INP_TIMEWAIT) {
1037 					INP_WUNLOCK(inp);
1038 					continue;
1039 				}
1040 				tp = intotcpcb(inp);
1041 				if (tp == NULL || tp->t_fb != blk) {
1042 					INP_WUNLOCK(inp);
1043 					continue;
1044 				}
1045 				tcp_switch_back_to_default(tp);
1046 				INP_WUNLOCK(inp);
1047 			}
1048 			INP_INFO_WUNLOCK(&V_tcbinfo);
1049 			CURVNET_RESTORE();
1050 		}
1051 		VNET_LIST_RUNLOCK();
1052 
1053 		rw_wlock(&tcp_function_lock);
1054 	}
1055 	if (blk->tfb_refcnt) {
1056 		/* TCBs still attached. */
1057 		rw_wunlock(&tcp_function_lock);
1058 		return (EBUSY);
1059 	}
1060 	if (quiesce) {
1061 		/* Skip removal. */
1062 		rw_wunlock(&tcp_function_lock);
1063 		return (0);
1064 	}
1065 	/* Remove any function names that map to this function block. */
1066 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1067 		TAILQ_REMOVE(&t_functions, f, tf_next);
1068 		tcp_fb_cnt--;
1069 		f->tf_fb = NULL;
1070 		free(f, M_TCPFUNCTIONS);
1071 	}
1072 	rw_wunlock(&tcp_function_lock);
1073 	return (0);
1074 }
1075 
1076 void
1077 tcp_init(void)
1078 {
1079 	const char *tcbhash_tuneable;
1080 	int hashsize;
1081 
1082 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1083 
1084 #ifdef TCP_HHOOK
1085 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1086 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1087 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1088 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1089 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1090 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1091 #endif
1092 #ifdef STATS
1093 	if (tcp_stats_init())
1094 		printf("%s: WARNING: unable to initialise TCP stats\n",
1095 		    __func__);
1096 #endif
1097 	hashsize = TCBHASHSIZE;
1098 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1099 	if (hashsize == 0) {
1100 		/*
1101 		 * Auto tune the hash size based on maxsockets.
1102 		 * A perfect hash would have a 1:1 mapping
1103 		 * (hashsize = maxsockets) however it's been
1104 		 * suggested that O(2) average is better.
1105 		 */
1106 		hashsize = maketcp_hashsize(maxsockets / 4);
1107 		/*
1108 		 * Our historical default is 512,
1109 		 * do not autotune lower than this.
1110 		 */
1111 		if (hashsize < 512)
1112 			hashsize = 512;
1113 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1114 			printf("%s: %s auto tuned to %d\n", __func__,
1115 			    tcbhash_tuneable, hashsize);
1116 	}
1117 	/*
1118 	 * We require a hashsize to be a power of two.
1119 	 * Previously if it was not a power of two we would just reset it
1120 	 * back to 512, which could be a nasty surprise if you did not notice
1121 	 * the error message.
1122 	 * Instead what we do is clip it to the closest power of two lower
1123 	 * than the specified hash value.
1124 	 */
1125 	if (!powerof2(hashsize)) {
1126 		int oldhashsize = hashsize;
1127 
1128 		hashsize = maketcp_hashsize(hashsize);
1129 		/* prevent absurdly low value */
1130 		if (hashsize < 16)
1131 			hashsize = 16;
1132 		printf("%s: WARNING: TCB hash size not a power of 2, "
1133 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1134 		    hashsize);
1135 	}
1136 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1137 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1138 
1139 	/*
1140 	 * These have to be type stable for the benefit of the timers.
1141 	 */
1142 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1143 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1144 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1145 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1146 
1147 	tcp_tw_init();
1148 	syncache_init();
1149 	tcp_hc_init();
1150 
1151 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1152 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1153 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1154 
1155 	tcp_fastopen_init();
1156 
1157 	/* Skip initialization of globals for non-default instances. */
1158 	if (!IS_DEFAULT_VNET(curvnet))
1159 		return;
1160 
1161 	tcp_reass_global_init();
1162 
1163 	/* XXX virtualize those bellow? */
1164 	tcp_delacktime = TCPTV_DELACK;
1165 	tcp_keepinit = TCPTV_KEEP_INIT;
1166 	tcp_keepidle = TCPTV_KEEP_IDLE;
1167 	tcp_keepintvl = TCPTV_KEEPINTVL;
1168 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1169 	tcp_msl = TCPTV_MSL;
1170 	tcp_rexmit_initial = TCPTV_RTOBASE;
1171 	if (tcp_rexmit_initial < 1)
1172 		tcp_rexmit_initial = 1;
1173 	tcp_rexmit_min = TCPTV_MIN;
1174 	if (tcp_rexmit_min < 1)
1175 		tcp_rexmit_min = 1;
1176 	tcp_persmin = TCPTV_PERSMIN;
1177 	tcp_persmax = TCPTV_PERSMAX;
1178 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1179 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1180 	tcp_tcbhashsize = hashsize;
1181 
1182 	/* Setup the tcp function block list */
1183 	TAILQ_INIT(&t_functions);
1184 	rw_init(&tcp_function_lock, "tcp_func_lock");
1185 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1186 #ifdef TCP_BLACKBOX
1187 	/* Initialize the TCP logging data. */
1188 	tcp_log_init();
1189 #endif
1190 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1191 
1192 	if (tcp_soreceive_stream) {
1193 #ifdef INET
1194 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1195 #endif
1196 #ifdef INET6
1197 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1198 #endif /* INET6 */
1199 	}
1200 
1201 #ifdef INET6
1202 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1203 #else /* INET6 */
1204 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1205 #endif /* INET6 */
1206 	if (max_protohdr < TCP_MINPROTOHDR)
1207 		max_protohdr = TCP_MINPROTOHDR;
1208 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1209 		panic("tcp_init");
1210 #undef TCP_MINPROTOHDR
1211 
1212 	ISN_LOCK_INIT();
1213 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1214 		SHUTDOWN_PRI_DEFAULT);
1215 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1216 		EVENTHANDLER_PRI_ANY);
1217 
1218 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1219 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1220 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1221 	tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK);
1222 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1223 	tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK);
1224 #ifdef TCPPCAP
1225 	tcp_pcap_init();
1226 #endif
1227 }
1228 
1229 #ifdef VIMAGE
1230 static void
1231 tcp_destroy(void *unused __unused)
1232 {
1233 	int n;
1234 #ifdef TCP_HHOOK
1235 	int error;
1236 #endif
1237 
1238 	/*
1239 	 * All our processes are gone, all our sockets should be cleaned
1240 	 * up, which means, we should be past the tcp_discardcb() calls.
1241 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1242 	 */
1243 	for (;;) {
1244 		INP_LIST_RLOCK(&V_tcbinfo);
1245 		n = V_tcbinfo.ipi_count;
1246 		INP_LIST_RUNLOCK(&V_tcbinfo);
1247 		if (n == 0)
1248 			break;
1249 		pause("tcpdes", hz / 10);
1250 	}
1251 	tcp_hc_destroy();
1252 	syncache_destroy();
1253 	tcp_tw_destroy();
1254 	in_pcbinfo_destroy(&V_tcbinfo);
1255 	/* tcp_discardcb() clears the sack_holes up. */
1256 	uma_zdestroy(V_sack_hole_zone);
1257 	uma_zdestroy(V_tcpcb_zone);
1258 
1259 	/*
1260 	 * Cannot free the zone until all tcpcbs are released as we attach
1261 	 * the allocations to them.
1262 	 */
1263 	tcp_fastopen_destroy();
1264 
1265 #ifdef TCP_HHOOK
1266 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1267 	if (error != 0) {
1268 		printf("%s: WARNING: unable to deregister helper hook "
1269 		    "type=%d, id=%d: error %d returned\n", __func__,
1270 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1271 	}
1272 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1273 	if (error != 0) {
1274 		printf("%s: WARNING: unable to deregister helper hook "
1275 		    "type=%d, id=%d: error %d returned\n", __func__,
1276 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1277 	}
1278 #endif
1279 }
1280 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1281 #endif
1282 
1283 void
1284 tcp_fini(void *xtp)
1285 {
1286 
1287 }
1288 
1289 /*
1290  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1291  * tcp_template used to store this data in mbufs, but we now recopy it out
1292  * of the tcpcb each time to conserve mbufs.
1293  */
1294 void
1295 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1296 {
1297 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1298 
1299 	INP_WLOCK_ASSERT(inp);
1300 
1301 #ifdef INET6
1302 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1303 		struct ip6_hdr *ip6;
1304 
1305 		ip6 = (struct ip6_hdr *)ip_ptr;
1306 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1307 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1308 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1309 			(IPV6_VERSION & IPV6_VERSION_MASK);
1310 		ip6->ip6_nxt = IPPROTO_TCP;
1311 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1312 		ip6->ip6_src = inp->in6p_laddr;
1313 		ip6->ip6_dst = inp->in6p_faddr;
1314 	}
1315 #endif /* INET6 */
1316 #if defined(INET6) && defined(INET)
1317 	else
1318 #endif
1319 #ifdef INET
1320 	{
1321 		struct ip *ip;
1322 
1323 		ip = (struct ip *)ip_ptr;
1324 		ip->ip_v = IPVERSION;
1325 		ip->ip_hl = 5;
1326 		ip->ip_tos = inp->inp_ip_tos;
1327 		ip->ip_len = 0;
1328 		ip->ip_id = 0;
1329 		ip->ip_off = 0;
1330 		ip->ip_ttl = inp->inp_ip_ttl;
1331 		ip->ip_sum = 0;
1332 		ip->ip_p = IPPROTO_TCP;
1333 		ip->ip_src = inp->inp_laddr;
1334 		ip->ip_dst = inp->inp_faddr;
1335 	}
1336 #endif /* INET */
1337 	th->th_sport = inp->inp_lport;
1338 	th->th_dport = inp->inp_fport;
1339 	th->th_seq = 0;
1340 	th->th_ack = 0;
1341 	th->th_x2 = 0;
1342 	th->th_off = 5;
1343 	th->th_flags = 0;
1344 	th->th_win = 0;
1345 	th->th_urp = 0;
1346 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1347 }
1348 
1349 /*
1350  * Create template to be used to send tcp packets on a connection.
1351  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1352  * use for this function is in keepalives, which use tcp_respond.
1353  */
1354 struct tcptemp *
1355 tcpip_maketemplate(struct inpcb *inp)
1356 {
1357 	struct tcptemp *t;
1358 
1359 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1360 	if (t == NULL)
1361 		return (NULL);
1362 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1363 	return (t);
1364 }
1365 
1366 /*
1367  * Send a single message to the TCP at address specified by
1368  * the given TCP/IP header.  If m == NULL, then we make a copy
1369  * of the tcpiphdr at th and send directly to the addressed host.
1370  * This is used to force keep alive messages out using the TCP
1371  * template for a connection.  If flags are given then we send
1372  * a message back to the TCP which originated the segment th,
1373  * and discard the mbuf containing it and any other attached mbufs.
1374  *
1375  * In any case the ack and sequence number of the transmitted
1376  * segment are as specified by the parameters.
1377  *
1378  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1379  */
1380 void
1381 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1382     tcp_seq ack, tcp_seq seq, int flags)
1383 {
1384 	struct tcpopt to;
1385 	struct inpcb *inp;
1386 	struct ip *ip;
1387 	struct mbuf *optm;
1388 	struct tcphdr *nth;
1389 	u_char *optp;
1390 #ifdef INET6
1391 	struct ip6_hdr *ip6;
1392 	int isipv6;
1393 #endif /* INET6 */
1394 	int optlen, tlen, win;
1395 	bool incl_opts;
1396 
1397 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1398 	NET_EPOCH_ASSERT();
1399 
1400 #ifdef INET6
1401 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1402 	ip6 = ipgen;
1403 #endif /* INET6 */
1404 	ip = ipgen;
1405 
1406 	if (tp != NULL) {
1407 		inp = tp->t_inpcb;
1408 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1409 		INP_WLOCK_ASSERT(inp);
1410 	} else
1411 		inp = NULL;
1412 
1413 	incl_opts = false;
1414 	win = 0;
1415 	if (tp != NULL) {
1416 		if (!(flags & TH_RST)) {
1417 			win = sbspace(&inp->inp_socket->so_rcv);
1418 			if (win > TCP_MAXWIN << tp->rcv_scale)
1419 				win = TCP_MAXWIN << tp->rcv_scale;
1420 		}
1421 		if ((tp->t_flags & TF_NOOPT) == 0)
1422 			incl_opts = true;
1423 	}
1424 	if (m == NULL) {
1425 		m = m_gethdr(M_NOWAIT, MT_DATA);
1426 		if (m == NULL)
1427 			return;
1428 		m->m_data += max_linkhdr;
1429 #ifdef INET6
1430 		if (isipv6) {
1431 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1432 			      sizeof(struct ip6_hdr));
1433 			ip6 = mtod(m, struct ip6_hdr *);
1434 			nth = (struct tcphdr *)(ip6 + 1);
1435 		} else
1436 #endif /* INET6 */
1437 		{
1438 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1439 			ip = mtod(m, struct ip *);
1440 			nth = (struct tcphdr *)(ip + 1);
1441 		}
1442 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1443 		flags = TH_ACK;
1444 	} else if (!M_WRITABLE(m)) {
1445 		struct mbuf *n;
1446 
1447 		/* Can't reuse 'm', allocate a new mbuf. */
1448 		n = m_gethdr(M_NOWAIT, MT_DATA);
1449 		if (n == NULL) {
1450 			m_freem(m);
1451 			return;
1452 		}
1453 
1454 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1455 			m_freem(m);
1456 			m_freem(n);
1457 			return;
1458 		}
1459 
1460 		n->m_data += max_linkhdr;
1461 		/* m_len is set later */
1462 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1463 #ifdef INET6
1464 		if (isipv6) {
1465 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1466 			      sizeof(struct ip6_hdr));
1467 			ip6 = mtod(n, struct ip6_hdr *);
1468 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1469 			nth = (struct tcphdr *)(ip6 + 1);
1470 		} else
1471 #endif /* INET6 */
1472 		{
1473 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1474 			ip = mtod(n, struct ip *);
1475 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1476 			nth = (struct tcphdr *)(ip + 1);
1477 		}
1478 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1479 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1480 		th = nth;
1481 		m_freem(m);
1482 		m = n;
1483 	} else {
1484 		/*
1485 		 *  reuse the mbuf.
1486 		 * XXX MRT We inherit the FIB, which is lucky.
1487 		 */
1488 		m_freem(m->m_next);
1489 		m->m_next = NULL;
1490 		m->m_data = (caddr_t)ipgen;
1491 		/* m_len is set later */
1492 #ifdef INET6
1493 		if (isipv6) {
1494 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1495 			nth = (struct tcphdr *)(ip6 + 1);
1496 		} else
1497 #endif /* INET6 */
1498 		{
1499 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1500 			nth = (struct tcphdr *)(ip + 1);
1501 		}
1502 		if (th != nth) {
1503 			/*
1504 			 * this is usually a case when an extension header
1505 			 * exists between the IPv6 header and the
1506 			 * TCP header.
1507 			 */
1508 			nth->th_sport = th->th_sport;
1509 			nth->th_dport = th->th_dport;
1510 		}
1511 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1512 #undef xchg
1513 	}
1514 	tlen = 0;
1515 #ifdef INET6
1516 	if (isipv6)
1517 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1518 #endif
1519 #if defined(INET) && defined(INET6)
1520 	else
1521 #endif
1522 #ifdef INET
1523 		tlen = sizeof (struct tcpiphdr);
1524 #endif
1525 #ifdef INVARIANTS
1526 	m->m_len = 0;
1527 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1528 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1529 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1530 #endif
1531 	m->m_len = tlen;
1532 	to.to_flags = 0;
1533 	if (incl_opts) {
1534 		/* Make sure we have room. */
1535 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1536 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1537 			if (m->m_next) {
1538 				optp = mtod(m->m_next, u_char *);
1539 				optm = m->m_next;
1540 			} else
1541 				incl_opts = false;
1542 		} else {
1543 			optp = (u_char *) (nth + 1);
1544 			optm = m;
1545 		}
1546 	}
1547 	if (incl_opts) {
1548 		/* Timestamps. */
1549 		if (tp->t_flags & TF_RCVD_TSTMP) {
1550 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1551 			to.to_tsecr = tp->ts_recent;
1552 			to.to_flags |= TOF_TS;
1553 		}
1554 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1555 		/* TCP-MD5 (RFC2385). */
1556 		if (tp->t_flags & TF_SIGNATURE)
1557 			to.to_flags |= TOF_SIGNATURE;
1558 #endif
1559 		/* Add the options. */
1560 		tlen += optlen = tcp_addoptions(&to, optp);
1561 
1562 		/* Update m_len in the correct mbuf. */
1563 		optm->m_len += optlen;
1564 	} else
1565 		optlen = 0;
1566 #ifdef INET6
1567 	if (isipv6) {
1568 		ip6->ip6_flow = 0;
1569 		ip6->ip6_vfc = IPV6_VERSION;
1570 		ip6->ip6_nxt = IPPROTO_TCP;
1571 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1572 	}
1573 #endif
1574 #if defined(INET) && defined(INET6)
1575 	else
1576 #endif
1577 #ifdef INET
1578 	{
1579 		ip->ip_len = htons(tlen);
1580 		ip->ip_ttl = V_ip_defttl;
1581 		if (V_path_mtu_discovery)
1582 			ip->ip_off |= htons(IP_DF);
1583 	}
1584 #endif
1585 	m->m_pkthdr.len = tlen;
1586 	m->m_pkthdr.rcvif = NULL;
1587 #ifdef MAC
1588 	if (inp != NULL) {
1589 		/*
1590 		 * Packet is associated with a socket, so allow the
1591 		 * label of the response to reflect the socket label.
1592 		 */
1593 		INP_WLOCK_ASSERT(inp);
1594 		mac_inpcb_create_mbuf(inp, m);
1595 	} else {
1596 		/*
1597 		 * Packet is not associated with a socket, so possibly
1598 		 * update the label in place.
1599 		 */
1600 		mac_netinet_tcp_reply(m);
1601 	}
1602 #endif
1603 	nth->th_seq = htonl(seq);
1604 	nth->th_ack = htonl(ack);
1605 	nth->th_x2 = 0;
1606 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1607 	nth->th_flags = flags;
1608 	if (tp != NULL)
1609 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1610 	else
1611 		nth->th_win = htons((u_short)win);
1612 	nth->th_urp = 0;
1613 
1614 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1615 	if (to.to_flags & TOF_SIGNATURE) {
1616 		if (!TCPMD5_ENABLED() ||
1617 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1618 			m_freem(m);
1619 			return;
1620 		}
1621 	}
1622 #endif
1623 
1624 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1625 #ifdef INET6
1626 	if (isipv6) {
1627 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1628 		nth->th_sum = in6_cksum_pseudo(ip6,
1629 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1630 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1631 		    NULL, NULL);
1632 	}
1633 #endif /* INET6 */
1634 #if defined(INET6) && defined(INET)
1635 	else
1636 #endif
1637 #ifdef INET
1638 	{
1639 		m->m_pkthdr.csum_flags = CSUM_TCP;
1640 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1641 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1642 	}
1643 #endif /* INET */
1644 #ifdef TCPDEBUG
1645 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1646 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1647 #endif
1648 	TCP_PROBE3(debug__output, tp, th, m);
1649 	if (flags & TH_RST)
1650 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1651 
1652 #ifdef INET6
1653 	if (isipv6) {
1654 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1655 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1656 	}
1657 #endif /* INET6 */
1658 #if defined(INET) && defined(INET6)
1659 	else
1660 #endif
1661 #ifdef INET
1662 	{
1663 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1664 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1665 	}
1666 #endif
1667 }
1668 
1669 /*
1670  * Create a new TCP control block, making an
1671  * empty reassembly queue and hooking it to the argument
1672  * protocol control block.  The `inp' parameter must have
1673  * come from the zone allocator set up in tcp_init().
1674  */
1675 struct tcpcb *
1676 tcp_newtcpcb(struct inpcb *inp)
1677 {
1678 	struct tcpcb_mem *tm;
1679 	struct tcpcb *tp;
1680 #ifdef INET6
1681 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1682 #endif /* INET6 */
1683 
1684 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1685 	if (tm == NULL)
1686 		return (NULL);
1687 	tp = &tm->tcb;
1688 
1689 	/* Initialise cc_var struct for this tcpcb. */
1690 	tp->ccv = &tm->ccv;
1691 	tp->ccv->type = IPPROTO_TCP;
1692 	tp->ccv->ccvc.tcp = tp;
1693 	rw_rlock(&tcp_function_lock);
1694 	tp->t_fb = tcp_func_set_ptr;
1695 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1696 	rw_runlock(&tcp_function_lock);
1697 	/*
1698 	 * Use the current system default CC algorithm.
1699 	 */
1700 	CC_LIST_RLOCK();
1701 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1702 	CC_ALGO(tp) = CC_DEFAULT();
1703 	CC_LIST_RUNLOCK();
1704 
1705 	if (CC_ALGO(tp)->cb_init != NULL)
1706 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1707 			if (tp->t_fb->tfb_tcp_fb_fini)
1708 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1709 			refcount_release(&tp->t_fb->tfb_refcnt);
1710 			uma_zfree(V_tcpcb_zone, tm);
1711 			return (NULL);
1712 		}
1713 
1714 #ifdef TCP_HHOOK
1715 	tp->osd = &tm->osd;
1716 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1717 		if (tp->t_fb->tfb_tcp_fb_fini)
1718 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1719 		refcount_release(&tp->t_fb->tfb_refcnt);
1720 		uma_zfree(V_tcpcb_zone, tm);
1721 		return (NULL);
1722 	}
1723 #endif
1724 
1725 #ifdef VIMAGE
1726 	tp->t_vnet = inp->inp_vnet;
1727 #endif
1728 	tp->t_timers = &tm->tt;
1729 	TAILQ_INIT(&tp->t_segq);
1730 	tp->t_maxseg =
1731 #ifdef INET6
1732 		isipv6 ? V_tcp_v6mssdflt :
1733 #endif /* INET6 */
1734 		V_tcp_mssdflt;
1735 
1736 	/* Set up our timeouts. */
1737 	callout_init(&tp->t_timers->tt_rexmt, 1);
1738 	callout_init(&tp->t_timers->tt_persist, 1);
1739 	callout_init(&tp->t_timers->tt_keep, 1);
1740 	callout_init(&tp->t_timers->tt_2msl, 1);
1741 	callout_init(&tp->t_timers->tt_delack, 1);
1742 
1743 	if (V_tcp_do_rfc1323)
1744 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1745 	if (V_tcp_do_sack)
1746 		tp->t_flags |= TF_SACK_PERMIT;
1747 	TAILQ_INIT(&tp->snd_holes);
1748 	/*
1749 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1750 	 * is called.
1751 	 */
1752 	in_pcbref(inp);	/* Reference for tcpcb */
1753 	tp->t_inpcb = inp;
1754 
1755 	/*
1756 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1757 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1758 	 * reasonable initial retransmit time.
1759 	 */
1760 	tp->t_srtt = TCPTV_SRTTBASE;
1761 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1762 	tp->t_rttmin = tcp_rexmit_min;
1763 	tp->t_rxtcur = tcp_rexmit_initial;
1764 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1765 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1766 	tp->t_rcvtime = ticks;
1767 	/*
1768 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1769 	 * because the socket may be bound to an IPv6 wildcard address,
1770 	 * which may match an IPv4-mapped IPv6 address.
1771 	 */
1772 	inp->inp_ip_ttl = V_ip_defttl;
1773 	inp->inp_ppcb = tp;
1774 #ifdef TCPPCAP
1775 	/*
1776 	 * Init the TCP PCAP queues.
1777 	 */
1778 	tcp_pcap_tcpcb_init(tp);
1779 #endif
1780 #ifdef TCP_BLACKBOX
1781 	/* Initialize the per-TCPCB log data. */
1782 	tcp_log_tcpcbinit(tp);
1783 #endif
1784 	if (tp->t_fb->tfb_tcp_fb_init) {
1785 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1786 	}
1787 #ifdef STATS
1788 	if (V_tcp_perconn_stats_enable == 1)
1789 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
1790 #endif
1791 	return (tp);		/* XXX */
1792 }
1793 
1794 /*
1795  * Switch the congestion control algorithm back to NewReno for any active
1796  * control blocks using an algorithm which is about to go away.
1797  * This ensures the CC framework can allow the unload to proceed without leaving
1798  * any dangling pointers which would trigger a panic.
1799  * Returning non-zero would inform the CC framework that something went wrong
1800  * and it would be unsafe to allow the unload to proceed. However, there is no
1801  * way for this to occur with this implementation so we always return zero.
1802  */
1803 int
1804 tcp_ccalgounload(struct cc_algo *unload_algo)
1805 {
1806 	struct cc_algo *tmpalgo;
1807 	struct inpcb *inp;
1808 	struct tcpcb *tp;
1809 	VNET_ITERATOR_DECL(vnet_iter);
1810 
1811 	/*
1812 	 * Check all active control blocks across all network stacks and change
1813 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1814 	 * requires cleanup code to be run, call it.
1815 	 */
1816 	VNET_LIST_RLOCK();
1817 	VNET_FOREACH(vnet_iter) {
1818 		CURVNET_SET(vnet_iter);
1819 		INP_INFO_WLOCK(&V_tcbinfo);
1820 		/*
1821 		 * New connections already part way through being initialised
1822 		 * with the CC algo we're removing will not race with this code
1823 		 * because the INP_INFO_WLOCK is held during initialisation. We
1824 		 * therefore don't enter the loop below until the connection
1825 		 * list has stabilised.
1826 		 */
1827 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1828 			INP_WLOCK(inp);
1829 			/* Important to skip tcptw structs. */
1830 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1831 			    (tp = intotcpcb(inp)) != NULL) {
1832 				/*
1833 				 * By holding INP_WLOCK here, we are assured
1834 				 * that the connection is not currently
1835 				 * executing inside the CC module's functions
1836 				 * i.e. it is safe to make the switch back to
1837 				 * NewReno.
1838 				 */
1839 				if (CC_ALGO(tp) == unload_algo) {
1840 					tmpalgo = CC_ALGO(tp);
1841 					if (tmpalgo->cb_destroy != NULL)
1842 						tmpalgo->cb_destroy(tp->ccv);
1843 					CC_DATA(tp) = NULL;
1844 					/*
1845 					 * NewReno may allocate memory on
1846 					 * demand for certain stateful
1847 					 * configuration as needed, but is
1848 					 * coded to never fail on memory
1849 					 * allocation failure so it is a safe
1850 					 * fallback.
1851 					 */
1852 					CC_ALGO(tp) = &newreno_cc_algo;
1853 				}
1854 			}
1855 			INP_WUNLOCK(inp);
1856 		}
1857 		INP_INFO_WUNLOCK(&V_tcbinfo);
1858 		CURVNET_RESTORE();
1859 	}
1860 	VNET_LIST_RUNLOCK();
1861 
1862 	return (0);
1863 }
1864 
1865 /*
1866  * Drop a TCP connection, reporting
1867  * the specified error.  If connection is synchronized,
1868  * then send a RST to peer.
1869  */
1870 struct tcpcb *
1871 tcp_drop(struct tcpcb *tp, int errno)
1872 {
1873 	struct socket *so = tp->t_inpcb->inp_socket;
1874 
1875 	NET_EPOCH_ASSERT();
1876 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1877 	INP_WLOCK_ASSERT(tp->t_inpcb);
1878 
1879 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1880 		tcp_state_change(tp, TCPS_CLOSED);
1881 		(void) tp->t_fb->tfb_tcp_output(tp);
1882 		TCPSTAT_INC(tcps_drops);
1883 	} else
1884 		TCPSTAT_INC(tcps_conndrops);
1885 	if (errno == ETIMEDOUT && tp->t_softerror)
1886 		errno = tp->t_softerror;
1887 	so->so_error = errno;
1888 	return (tcp_close(tp));
1889 }
1890 
1891 void
1892 tcp_discardcb(struct tcpcb *tp)
1893 {
1894 	struct inpcb *inp = tp->t_inpcb;
1895 	struct socket *so = inp->inp_socket;
1896 #ifdef INET6
1897 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1898 #endif /* INET6 */
1899 	int released __unused;
1900 
1901 	INP_WLOCK_ASSERT(inp);
1902 
1903 	/*
1904 	 * Make sure that all of our timers are stopped before we delete the
1905 	 * PCB.
1906 	 *
1907 	 * If stopping a timer fails, we schedule a discard function in same
1908 	 * callout, and the last discard function called will take care of
1909 	 * deleting the tcpcb.
1910 	 */
1911 	tp->t_timers->tt_draincnt = 0;
1912 	tcp_timer_stop(tp, TT_REXMT);
1913 	tcp_timer_stop(tp, TT_PERSIST);
1914 	tcp_timer_stop(tp, TT_KEEP);
1915 	tcp_timer_stop(tp, TT_2MSL);
1916 	tcp_timer_stop(tp, TT_DELACK);
1917 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1918 		/*
1919 		 * Call the stop-all function of the methods,
1920 		 * this function should call the tcp_timer_stop()
1921 		 * method with each of the function specific timeouts.
1922 		 * That stop will be called via the tfb_tcp_timer_stop()
1923 		 * which should use the async drain function of the
1924 		 * callout system (see tcp_var.h).
1925 		 */
1926 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1927 	}
1928 
1929 	/*
1930 	 * If we got enough samples through the srtt filter,
1931 	 * save the rtt and rttvar in the routing entry.
1932 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1933 	 * 4 samples is enough for the srtt filter to converge
1934 	 * to within enough % of the correct value; fewer samples
1935 	 * and we could save a bogus rtt. The danger is not high
1936 	 * as tcp quickly recovers from everything.
1937 	 * XXX: Works very well but needs some more statistics!
1938 	 */
1939 	if (tp->t_rttupdated >= 4) {
1940 		struct hc_metrics_lite metrics;
1941 		uint32_t ssthresh;
1942 
1943 		bzero(&metrics, sizeof(metrics));
1944 		/*
1945 		 * Update the ssthresh always when the conditions below
1946 		 * are satisfied. This gives us better new start value
1947 		 * for the congestion avoidance for new connections.
1948 		 * ssthresh is only set if packet loss occurred on a session.
1949 		 *
1950 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1951 		 * being torn down.  Ideally this code would not use 'so'.
1952 		 */
1953 		ssthresh = tp->snd_ssthresh;
1954 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1955 			/*
1956 			 * convert the limit from user data bytes to
1957 			 * packets then to packet data bytes.
1958 			 */
1959 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1960 			if (ssthresh < 2)
1961 				ssthresh = 2;
1962 			ssthresh *= (tp->t_maxseg +
1963 #ifdef INET6
1964 			    (isipv6 ? sizeof (struct ip6_hdr) +
1965 				sizeof (struct tcphdr) :
1966 #endif
1967 				sizeof (struct tcpiphdr)
1968 #ifdef INET6
1969 			    )
1970 #endif
1971 			    );
1972 		} else
1973 			ssthresh = 0;
1974 		metrics.rmx_ssthresh = ssthresh;
1975 
1976 		metrics.rmx_rtt = tp->t_srtt;
1977 		metrics.rmx_rttvar = tp->t_rttvar;
1978 		metrics.rmx_cwnd = tp->snd_cwnd;
1979 		metrics.rmx_sendpipe = 0;
1980 		metrics.rmx_recvpipe = 0;
1981 
1982 		tcp_hc_update(&inp->inp_inc, &metrics);
1983 	}
1984 
1985 	/* free the reassembly queue, if any */
1986 	tcp_reass_flush(tp);
1987 
1988 #ifdef TCP_OFFLOAD
1989 	/* Disconnect offload device, if any. */
1990 	if (tp->t_flags & TF_TOE)
1991 		tcp_offload_detach(tp);
1992 #endif
1993 
1994 	tcp_free_sackholes(tp);
1995 
1996 #ifdef TCPPCAP
1997 	/* Free the TCP PCAP queues. */
1998 	tcp_pcap_drain(&(tp->t_inpkts));
1999 	tcp_pcap_drain(&(tp->t_outpkts));
2000 #endif
2001 
2002 	/* Allow the CC algorithm to clean up after itself. */
2003 	if (CC_ALGO(tp)->cb_destroy != NULL)
2004 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2005 	CC_DATA(tp) = NULL;
2006 
2007 #ifdef TCP_HHOOK
2008 	khelp_destroy_osd(tp->osd);
2009 #endif
2010 #ifdef STATS
2011 	stats_blob_destroy(tp->t_stats);
2012 #endif
2013 
2014 	CC_ALGO(tp) = NULL;
2015 	inp->inp_ppcb = NULL;
2016 	if (tp->t_timers->tt_draincnt == 0) {
2017 		/* We own the last reference on tcpcb, let's free it. */
2018 #ifdef TCP_BLACKBOX
2019 		tcp_log_tcpcbfini(tp);
2020 #endif
2021 		TCPSTATES_DEC(tp->t_state);
2022 		if (tp->t_fb->tfb_tcp_fb_fini)
2023 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2024 		refcount_release(&tp->t_fb->tfb_refcnt);
2025 		tp->t_inpcb = NULL;
2026 		uma_zfree(V_tcpcb_zone, tp);
2027 		released = in_pcbrele_wlocked(inp);
2028 		KASSERT(!released, ("%s: inp %p should not have been released "
2029 			"here", __func__, inp));
2030 	}
2031 }
2032 
2033 void
2034 tcp_timer_discard(void *ptp)
2035 {
2036 	struct inpcb *inp;
2037 	struct tcpcb *tp;
2038 	struct epoch_tracker et;
2039 
2040 	tp = (struct tcpcb *)ptp;
2041 	CURVNET_SET(tp->t_vnet);
2042 	NET_EPOCH_ENTER(et);
2043 	inp = tp->t_inpcb;
2044 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2045 		__func__, tp));
2046 	INP_WLOCK(inp);
2047 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2048 		("%s: tcpcb has to be stopped here", __func__));
2049 	tp->t_timers->tt_draincnt--;
2050 	if (tp->t_timers->tt_draincnt == 0) {
2051 		/* We own the last reference on this tcpcb, let's free it. */
2052 #ifdef TCP_BLACKBOX
2053 		tcp_log_tcpcbfini(tp);
2054 #endif
2055 		TCPSTATES_DEC(tp->t_state);
2056 		if (tp->t_fb->tfb_tcp_fb_fini)
2057 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2058 		refcount_release(&tp->t_fb->tfb_refcnt);
2059 		tp->t_inpcb = NULL;
2060 		uma_zfree(V_tcpcb_zone, tp);
2061 		if (in_pcbrele_wlocked(inp)) {
2062 			NET_EPOCH_EXIT(et);
2063 			CURVNET_RESTORE();
2064 			return;
2065 		}
2066 	}
2067 	INP_WUNLOCK(inp);
2068 	NET_EPOCH_EXIT(et);
2069 	CURVNET_RESTORE();
2070 }
2071 
2072 /*
2073  * Attempt to close a TCP control block, marking it as dropped, and freeing
2074  * the socket if we hold the only reference.
2075  */
2076 struct tcpcb *
2077 tcp_close(struct tcpcb *tp)
2078 {
2079 	struct inpcb *inp = tp->t_inpcb;
2080 	struct socket *so;
2081 
2082 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2083 	INP_WLOCK_ASSERT(inp);
2084 
2085 #ifdef TCP_OFFLOAD
2086 	if (tp->t_state == TCPS_LISTEN)
2087 		tcp_offload_listen_stop(tp);
2088 #endif
2089 	/*
2090 	 * This releases the TFO pending counter resource for TFO listen
2091 	 * sockets as well as passively-created TFO sockets that transition
2092 	 * from SYN_RECEIVED to CLOSED.
2093 	 */
2094 	if (tp->t_tfo_pending) {
2095 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2096 		tp->t_tfo_pending = NULL;
2097 	}
2098 	in_pcbdrop(inp);
2099 	TCPSTAT_INC(tcps_closed);
2100 	if (tp->t_state != TCPS_CLOSED)
2101 		tcp_state_change(tp, TCPS_CLOSED);
2102 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2103 	so = inp->inp_socket;
2104 	soisdisconnected(so);
2105 	if (inp->inp_flags & INP_SOCKREF) {
2106 		KASSERT(so->so_state & SS_PROTOREF,
2107 		    ("tcp_close: !SS_PROTOREF"));
2108 		inp->inp_flags &= ~INP_SOCKREF;
2109 		INP_WUNLOCK(inp);
2110 		SOCK_LOCK(so);
2111 		so->so_state &= ~SS_PROTOREF;
2112 		sofree(so);
2113 		return (NULL);
2114 	}
2115 	return (tp);
2116 }
2117 
2118 void
2119 tcp_drain(void)
2120 {
2121 	VNET_ITERATOR_DECL(vnet_iter);
2122 
2123 	if (!do_tcpdrain)
2124 		return;
2125 
2126 	VNET_LIST_RLOCK_NOSLEEP();
2127 	VNET_FOREACH(vnet_iter) {
2128 		CURVNET_SET(vnet_iter);
2129 		struct inpcb *inpb;
2130 		struct tcpcb *tcpb;
2131 
2132 	/*
2133 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2134 	 * if there is one...
2135 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2136 	 *      reassembly queue should be flushed, but in a situation
2137 	 *	where we're really low on mbufs, this is potentially
2138 	 *	useful.
2139 	 */
2140 		INP_INFO_WLOCK(&V_tcbinfo);
2141 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2142 			INP_WLOCK(inpb);
2143 			if (inpb->inp_flags & INP_TIMEWAIT) {
2144 				INP_WUNLOCK(inpb);
2145 				continue;
2146 			}
2147 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2148 				tcp_reass_flush(tcpb);
2149 				tcp_clean_sackreport(tcpb);
2150 #ifdef TCP_BLACKBOX
2151 				tcp_log_drain(tcpb);
2152 #endif
2153 #ifdef TCPPCAP
2154 				if (tcp_pcap_aggressive_free) {
2155 					/* Free the TCP PCAP queues. */
2156 					tcp_pcap_drain(&(tcpb->t_inpkts));
2157 					tcp_pcap_drain(&(tcpb->t_outpkts));
2158 				}
2159 #endif
2160 			}
2161 			INP_WUNLOCK(inpb);
2162 		}
2163 		INP_INFO_WUNLOCK(&V_tcbinfo);
2164 		CURVNET_RESTORE();
2165 	}
2166 	VNET_LIST_RUNLOCK_NOSLEEP();
2167 }
2168 
2169 /*
2170  * Notify a tcp user of an asynchronous error;
2171  * store error as soft error, but wake up user
2172  * (for now, won't do anything until can select for soft error).
2173  *
2174  * Do not wake up user since there currently is no mechanism for
2175  * reporting soft errors (yet - a kqueue filter may be added).
2176  */
2177 static struct inpcb *
2178 tcp_notify(struct inpcb *inp, int error)
2179 {
2180 	struct tcpcb *tp;
2181 
2182 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2183 	INP_WLOCK_ASSERT(inp);
2184 
2185 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2186 	    (inp->inp_flags & INP_DROPPED))
2187 		return (inp);
2188 
2189 	tp = intotcpcb(inp);
2190 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2191 
2192 	/*
2193 	 * Ignore some errors if we are hooked up.
2194 	 * If connection hasn't completed, has retransmitted several times,
2195 	 * and receives a second error, give up now.  This is better
2196 	 * than waiting a long time to establish a connection that
2197 	 * can never complete.
2198 	 */
2199 	if (tp->t_state == TCPS_ESTABLISHED &&
2200 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2201 	     error == EHOSTDOWN)) {
2202 		if (inp->inp_route.ro_rt) {
2203 			RTFREE(inp->inp_route.ro_rt);
2204 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2205 		}
2206 		return (inp);
2207 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2208 	    tp->t_softerror) {
2209 		tp = tcp_drop(tp, error);
2210 		if (tp != NULL)
2211 			return (inp);
2212 		else
2213 			return (NULL);
2214 	} else {
2215 		tp->t_softerror = error;
2216 		return (inp);
2217 	}
2218 #if 0
2219 	wakeup( &so->so_timeo);
2220 	sorwakeup(so);
2221 	sowwakeup(so);
2222 #endif
2223 }
2224 
2225 static int
2226 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2227 {
2228 	struct epoch_tracker et;
2229 	struct inpcb *inp;
2230 	struct xinpgen xig;
2231 	int error;
2232 
2233 	if (req->newptr != NULL)
2234 		return (EPERM);
2235 
2236 	if (req->oldptr == NULL) {
2237 		int n;
2238 
2239 		n = V_tcbinfo.ipi_count +
2240 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2241 		n += imax(n / 8, 10);
2242 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2243 		return (0);
2244 	}
2245 
2246 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2247 		return (error);
2248 
2249 	bzero(&xig, sizeof(xig));
2250 	xig.xig_len = sizeof xig;
2251 	xig.xig_count = V_tcbinfo.ipi_count +
2252 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2253 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2254 	xig.xig_sogen = so_gencnt;
2255 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2256 	if (error)
2257 		return (error);
2258 
2259 	error = syncache_pcblist(req);
2260 	if (error)
2261 		return (error);
2262 
2263 	NET_EPOCH_ENTER(et);
2264 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2265 	    inp != NULL;
2266 	    inp = CK_LIST_NEXT(inp, inp_list)) {
2267 		INP_RLOCK(inp);
2268 		if (inp->inp_gencnt <= xig.xig_gen) {
2269 			int crerr;
2270 
2271 			/*
2272 			 * XXX: This use of cr_cansee(), introduced with
2273 			 * TCP state changes, is not quite right, but for
2274 			 * now, better than nothing.
2275 			 */
2276 			if (inp->inp_flags & INP_TIMEWAIT) {
2277 				if (intotw(inp) != NULL)
2278 					crerr = cr_cansee(req->td->td_ucred,
2279 					    intotw(inp)->tw_cred);
2280 				else
2281 					crerr = EINVAL;	/* Skip this inp. */
2282 			} else
2283 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2284 			if (crerr == 0) {
2285 				struct xtcpcb xt;
2286 
2287 				tcp_inptoxtp(inp, &xt);
2288 				INP_RUNLOCK(inp);
2289 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2290 				if (error)
2291 					break;
2292 				else
2293 					continue;
2294 			}
2295 		}
2296 		INP_RUNLOCK(inp);
2297 	}
2298 	NET_EPOCH_EXIT(et);
2299 
2300 	if (!error) {
2301 		/*
2302 		 * Give the user an updated idea of our state.
2303 		 * If the generation differs from what we told
2304 		 * her before, she knows that something happened
2305 		 * while we were processing this request, and it
2306 		 * might be necessary to retry.
2307 		 */
2308 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2309 		xig.xig_sogen = so_gencnt;
2310 		xig.xig_count = V_tcbinfo.ipi_count +
2311 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2312 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2313 	}
2314 
2315 	return (error);
2316 }
2317 
2318 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2319     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2320     NULL, 0, tcp_pcblist, "S,xtcpcb",
2321     "List of active TCP connections");
2322 
2323 #ifdef INET
2324 static int
2325 tcp_getcred(SYSCTL_HANDLER_ARGS)
2326 {
2327 	struct xucred xuc;
2328 	struct sockaddr_in addrs[2];
2329 	struct epoch_tracker et;
2330 	struct inpcb *inp;
2331 	int error;
2332 
2333 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2334 	if (error)
2335 		return (error);
2336 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2337 	if (error)
2338 		return (error);
2339 	NET_EPOCH_ENTER(et);
2340 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2341 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2342 	NET_EPOCH_EXIT(et);
2343 	if (inp != NULL) {
2344 		if (inp->inp_socket == NULL)
2345 			error = ENOENT;
2346 		if (error == 0)
2347 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2348 		if (error == 0)
2349 			cru2x(inp->inp_cred, &xuc);
2350 		INP_RUNLOCK(inp);
2351 	} else
2352 		error = ENOENT;
2353 	if (error == 0)
2354 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2355 	return (error);
2356 }
2357 
2358 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2359     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2360     0, 0, tcp_getcred, "S,xucred",
2361     "Get the xucred of a TCP connection");
2362 #endif /* INET */
2363 
2364 #ifdef INET6
2365 static int
2366 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2367 {
2368 	struct epoch_tracker et;
2369 	struct xucred xuc;
2370 	struct sockaddr_in6 addrs[2];
2371 	struct inpcb *inp;
2372 	int error;
2373 #ifdef INET
2374 	int mapped = 0;
2375 #endif
2376 
2377 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2378 	if (error)
2379 		return (error);
2380 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2381 	if (error)
2382 		return (error);
2383 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2384 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2385 		return (error);
2386 	}
2387 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2388 #ifdef INET
2389 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2390 			mapped = 1;
2391 		else
2392 #endif
2393 			return (EINVAL);
2394 	}
2395 
2396 	NET_EPOCH_ENTER(et);
2397 #ifdef INET
2398 	if (mapped == 1)
2399 		inp = in_pcblookup(&V_tcbinfo,
2400 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2401 			addrs[1].sin6_port,
2402 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2403 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2404 	else
2405 #endif
2406 		inp = in6_pcblookup(&V_tcbinfo,
2407 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2408 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2409 			INPLOOKUP_RLOCKPCB, NULL);
2410 	NET_EPOCH_EXIT(et);
2411 	if (inp != NULL) {
2412 		if (inp->inp_socket == NULL)
2413 			error = ENOENT;
2414 		if (error == 0)
2415 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2416 		if (error == 0)
2417 			cru2x(inp->inp_cred, &xuc);
2418 		INP_RUNLOCK(inp);
2419 	} else
2420 		error = ENOENT;
2421 	if (error == 0)
2422 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2423 	return (error);
2424 }
2425 
2426 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2427     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2428     0, 0, tcp6_getcred, "S,xucred",
2429     "Get the xucred of a TCP6 connection");
2430 #endif /* INET6 */
2431 
2432 
2433 #ifdef INET
2434 void
2435 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2436 {
2437 	struct ip *ip = vip;
2438 	struct tcphdr *th;
2439 	struct in_addr faddr;
2440 	struct inpcb *inp;
2441 	struct tcpcb *tp;
2442 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2443 	struct icmp *icp;
2444 	struct in_conninfo inc;
2445 	tcp_seq icmp_tcp_seq;
2446 	int mtu;
2447 
2448 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2449 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2450 		return;
2451 
2452 	if (cmd == PRC_MSGSIZE)
2453 		notify = tcp_mtudisc_notify;
2454 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2455 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2456 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2457 		notify = tcp_drop_syn_sent;
2458 
2459 	/*
2460 	 * Hostdead is ugly because it goes linearly through all PCBs.
2461 	 * XXX: We never get this from ICMP, otherwise it makes an
2462 	 * excellent DoS attack on machines with many connections.
2463 	 */
2464 	else if (cmd == PRC_HOSTDEAD)
2465 		ip = NULL;
2466 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2467 		return;
2468 
2469 	if (ip == NULL) {
2470 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2471 		return;
2472 	}
2473 
2474 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2475 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2476 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2477 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2478 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2479 		/* signal EHOSTDOWN, as it flushes the cached route */
2480 		inp = (*notify)(inp, EHOSTDOWN);
2481 		goto out;
2482 	}
2483 	icmp_tcp_seq = th->th_seq;
2484 	if (inp != NULL)  {
2485 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2486 		    !(inp->inp_flags & INP_DROPPED) &&
2487 		    !(inp->inp_socket == NULL)) {
2488 			tp = intotcpcb(inp);
2489 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2490 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2491 				if (cmd == PRC_MSGSIZE) {
2492 					/*
2493 					 * MTU discovery:
2494 					 * If we got a needfrag set the MTU
2495 					 * in the route to the suggested new
2496 					 * value (if given) and then notify.
2497 					 */
2498 					mtu = ntohs(icp->icmp_nextmtu);
2499 					/*
2500 					 * If no alternative MTU was
2501 					 * proposed, try the next smaller
2502 					 * one.
2503 					 */
2504 					if (!mtu)
2505 						mtu = ip_next_mtu(
2506 						    ntohs(ip->ip_len), 1);
2507 					if (mtu < V_tcp_minmss +
2508 					    sizeof(struct tcpiphdr))
2509 						mtu = V_tcp_minmss +
2510 						    sizeof(struct tcpiphdr);
2511 					/*
2512 					 * Only process the offered MTU if it
2513 					 * is smaller than the current one.
2514 					 */
2515 					if (mtu < tp->t_maxseg +
2516 					    sizeof(struct tcpiphdr)) {
2517 						bzero(&inc, sizeof(inc));
2518 						inc.inc_faddr = faddr;
2519 						inc.inc_fibnum =
2520 						    inp->inp_inc.inc_fibnum;
2521 						tcp_hc_updatemtu(&inc, mtu);
2522 						tcp_mtudisc(inp, mtu);
2523 					}
2524 				} else
2525 					inp = (*notify)(inp,
2526 					    inetctlerrmap[cmd]);
2527 			}
2528 		}
2529 	} else {
2530 		bzero(&inc, sizeof(inc));
2531 		inc.inc_fport = th->th_dport;
2532 		inc.inc_lport = th->th_sport;
2533 		inc.inc_faddr = faddr;
2534 		inc.inc_laddr = ip->ip_src;
2535 		syncache_unreach(&inc, icmp_tcp_seq);
2536 	}
2537 out:
2538 	if (inp != NULL)
2539 		INP_WUNLOCK(inp);
2540 }
2541 #endif /* INET */
2542 
2543 #ifdef INET6
2544 void
2545 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2546 {
2547 	struct in6_addr *dst;
2548 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2549 	struct ip6_hdr *ip6;
2550 	struct mbuf *m;
2551 	struct inpcb *inp;
2552 	struct tcpcb *tp;
2553 	struct icmp6_hdr *icmp6;
2554 	struct ip6ctlparam *ip6cp = NULL;
2555 	const struct sockaddr_in6 *sa6_src = NULL;
2556 	struct in_conninfo inc;
2557 	struct tcp_ports {
2558 		uint16_t th_sport;
2559 		uint16_t th_dport;
2560 	} t_ports;
2561 	tcp_seq icmp_tcp_seq;
2562 	unsigned int mtu;
2563 	unsigned int off;
2564 
2565 	if (sa->sa_family != AF_INET6 ||
2566 	    sa->sa_len != sizeof(struct sockaddr_in6))
2567 		return;
2568 
2569 	/* if the parameter is from icmp6, decode it. */
2570 	if (d != NULL) {
2571 		ip6cp = (struct ip6ctlparam *)d;
2572 		icmp6 = ip6cp->ip6c_icmp6;
2573 		m = ip6cp->ip6c_m;
2574 		ip6 = ip6cp->ip6c_ip6;
2575 		off = ip6cp->ip6c_off;
2576 		sa6_src = ip6cp->ip6c_src;
2577 		dst = ip6cp->ip6c_finaldst;
2578 	} else {
2579 		m = NULL;
2580 		ip6 = NULL;
2581 		off = 0;	/* fool gcc */
2582 		sa6_src = &sa6_any;
2583 		dst = NULL;
2584 	}
2585 
2586 	if (cmd == PRC_MSGSIZE)
2587 		notify = tcp_mtudisc_notify;
2588 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2589 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2590 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2591 		notify = tcp_drop_syn_sent;
2592 
2593 	/*
2594 	 * Hostdead is ugly because it goes linearly through all PCBs.
2595 	 * XXX: We never get this from ICMP, otherwise it makes an
2596 	 * excellent DoS attack on machines with many connections.
2597 	 */
2598 	else if (cmd == PRC_HOSTDEAD)
2599 		ip6 = NULL;
2600 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2601 		return;
2602 
2603 	if (ip6 == NULL) {
2604 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2605 			      (const struct sockaddr *)sa6_src,
2606 			      0, cmd, NULL, notify);
2607 		return;
2608 	}
2609 
2610 	/* Check if we can safely get the ports from the tcp hdr */
2611 	if (m == NULL ||
2612 	    (m->m_pkthdr.len <
2613 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2614 		return;
2615 	}
2616 	bzero(&t_ports, sizeof(struct tcp_ports));
2617 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2618 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2619 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2620 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2621 		/* signal EHOSTDOWN, as it flushes the cached route */
2622 		inp = (*notify)(inp, EHOSTDOWN);
2623 		goto out;
2624 	}
2625 	off += sizeof(struct tcp_ports);
2626 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2627 		goto out;
2628 	}
2629 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2630 	if (inp != NULL)  {
2631 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2632 		    !(inp->inp_flags & INP_DROPPED) &&
2633 		    !(inp->inp_socket == NULL)) {
2634 			tp = intotcpcb(inp);
2635 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2636 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2637 				if (cmd == PRC_MSGSIZE) {
2638 					/*
2639 					 * MTU discovery:
2640 					 * If we got a needfrag set the MTU
2641 					 * in the route to the suggested new
2642 					 * value (if given) and then notify.
2643 					 */
2644 					mtu = ntohl(icmp6->icmp6_mtu);
2645 					/*
2646 					 * If no alternative MTU was
2647 					 * proposed, or the proposed
2648 					 * MTU was too small, set to
2649 					 * the min.
2650 					 */
2651 					if (mtu < IPV6_MMTU)
2652 						mtu = IPV6_MMTU - 8;
2653 					bzero(&inc, sizeof(inc));
2654 					inc.inc_fibnum = M_GETFIB(m);
2655 					inc.inc_flags |= INC_ISIPV6;
2656 					inc.inc6_faddr = *dst;
2657 					if (in6_setscope(&inc.inc6_faddr,
2658 						m->m_pkthdr.rcvif, NULL))
2659 						goto out;
2660 					/*
2661 					 * Only process the offered MTU if it
2662 					 * is smaller than the current one.
2663 					 */
2664 					if (mtu < tp->t_maxseg +
2665 					    sizeof (struct tcphdr) +
2666 					    sizeof (struct ip6_hdr)) {
2667 						tcp_hc_updatemtu(&inc, mtu);
2668 						tcp_mtudisc(inp, mtu);
2669 						ICMP6STAT_INC(icp6s_pmtuchg);
2670 					}
2671 				} else
2672 					inp = (*notify)(inp,
2673 					    inet6ctlerrmap[cmd]);
2674 			}
2675 		}
2676 	} else {
2677 		bzero(&inc, sizeof(inc));
2678 		inc.inc_fibnum = M_GETFIB(m);
2679 		inc.inc_flags |= INC_ISIPV6;
2680 		inc.inc_fport = t_ports.th_dport;
2681 		inc.inc_lport = t_ports.th_sport;
2682 		inc.inc6_faddr = *dst;
2683 		inc.inc6_laddr = ip6->ip6_src;
2684 		syncache_unreach(&inc, icmp_tcp_seq);
2685 	}
2686 out:
2687 	if (inp != NULL)
2688 		INP_WUNLOCK(inp);
2689 }
2690 #endif /* INET6 */
2691 
2692 static uint32_t
2693 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2694 {
2695 	SIPHASH_CTX ctx;
2696 	uint32_t hash[2];
2697 
2698 	KASSERT(len >= SIPHASH_KEY_LENGTH,
2699 	    ("%s: keylen %u too short ", __func__, len));
2700 	SipHash24_Init(&ctx);
2701 	SipHash_SetKey(&ctx, (uint8_t *)key);
2702 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2703 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2704 	switch (inc->inc_flags & INC_ISIPV6) {
2705 #ifdef INET
2706 	case 0:
2707 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2708 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2709 		break;
2710 #endif
2711 #ifdef INET6
2712 	case INC_ISIPV6:
2713 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2714 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2715 		break;
2716 #endif
2717 	}
2718 	SipHash_Final((uint8_t *)hash, &ctx);
2719 
2720 	return (hash[0] ^ hash[1]);
2721 }
2722 
2723 uint32_t
2724 tcp_new_ts_offset(struct in_conninfo *inc)
2725 {
2726 	struct in_conninfo inc_store, *local_inc;
2727 
2728 	if (!V_tcp_ts_offset_per_conn) {
2729 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2730 		inc_store.inc_lport = 0;
2731 		inc_store.inc_fport = 0;
2732 		local_inc = &inc_store;
2733 	} else {
2734 		local_inc = inc;
2735 	}
2736 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2737 	    sizeof(V_ts_offset_secret)));
2738 }
2739 
2740 /*
2741  * Following is where TCP initial sequence number generation occurs.
2742  *
2743  * There are two places where we must use initial sequence numbers:
2744  * 1.  In SYN-ACK packets.
2745  * 2.  In SYN packets.
2746  *
2747  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2748  * tcp_syncache.c for details.
2749  *
2750  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2751  * depends on this property.  In addition, these ISNs should be
2752  * unguessable so as to prevent connection hijacking.  To satisfy
2753  * the requirements of this situation, the algorithm outlined in
2754  * RFC 1948 is used, with only small modifications.
2755  *
2756  * Implementation details:
2757  *
2758  * Time is based off the system timer, and is corrected so that it
2759  * increases by one megabyte per second.  This allows for proper
2760  * recycling on high speed LANs while still leaving over an hour
2761  * before rollover.
2762  *
2763  * As reading the *exact* system time is too expensive to be done
2764  * whenever setting up a TCP connection, we increment the time
2765  * offset in two ways.  First, a small random positive increment
2766  * is added to isn_offset for each connection that is set up.
2767  * Second, the function tcp_isn_tick fires once per clock tick
2768  * and increments isn_offset as necessary so that sequence numbers
2769  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2770  * random positive increments serve only to ensure that the same
2771  * exact sequence number is never sent out twice (as could otherwise
2772  * happen when a port is recycled in less than the system tick
2773  * interval.)
2774  *
2775  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2776  * between seeding of isn_secret.  This is normally set to zero,
2777  * as reseeding should not be necessary.
2778  *
2779  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2780  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2781  * general, this means holding an exclusive (write) lock.
2782  */
2783 
2784 #define ISN_BYTES_PER_SECOND 1048576
2785 #define ISN_STATIC_INCREMENT 4096
2786 #define ISN_RANDOM_INCREMENT (4096 - 1)
2787 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
2788 
2789 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2790 VNET_DEFINE_STATIC(int, isn_last);
2791 VNET_DEFINE_STATIC(int, isn_last_reseed);
2792 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2793 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2794 
2795 #define	V_isn_secret			VNET(isn_secret)
2796 #define	V_isn_last			VNET(isn_last)
2797 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2798 #define	V_isn_offset			VNET(isn_offset)
2799 #define	V_isn_offset_old		VNET(isn_offset_old)
2800 
2801 tcp_seq
2802 tcp_new_isn(struct in_conninfo *inc)
2803 {
2804 	tcp_seq new_isn;
2805 	u_int32_t projected_offset;
2806 
2807 	ISN_LOCK();
2808 	/* Seed if this is the first use, reseed if requested. */
2809 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2810 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2811 		< (u_int)ticks))) {
2812 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2813 		V_isn_last_reseed = ticks;
2814 	}
2815 
2816 	/* Compute the hash and return the ISN. */
2817 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2818 	    sizeof(V_isn_secret));
2819 	V_isn_offset += ISN_STATIC_INCREMENT +
2820 		(arc4random() & ISN_RANDOM_INCREMENT);
2821 	if (ticks != V_isn_last) {
2822 		projected_offset = V_isn_offset_old +
2823 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2824 		if (SEQ_GT(projected_offset, V_isn_offset))
2825 			V_isn_offset = projected_offset;
2826 		V_isn_offset_old = V_isn_offset;
2827 		V_isn_last = ticks;
2828 	}
2829 	new_isn += V_isn_offset;
2830 	ISN_UNLOCK();
2831 	return (new_isn);
2832 }
2833 
2834 /*
2835  * When a specific ICMP unreachable message is received and the
2836  * connection state is SYN-SENT, drop the connection.  This behavior
2837  * is controlled by the icmp_may_rst sysctl.
2838  */
2839 struct inpcb *
2840 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2841 {
2842 	struct tcpcb *tp;
2843 
2844 	NET_EPOCH_ASSERT();
2845 	INP_WLOCK_ASSERT(inp);
2846 
2847 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2848 	    (inp->inp_flags & INP_DROPPED))
2849 		return (inp);
2850 
2851 	tp = intotcpcb(inp);
2852 	if (tp->t_state != TCPS_SYN_SENT)
2853 		return (inp);
2854 
2855 	if (IS_FASTOPEN(tp->t_flags))
2856 		tcp_fastopen_disable_path(tp);
2857 
2858 	tp = tcp_drop(tp, errno);
2859 	if (tp != NULL)
2860 		return (inp);
2861 	else
2862 		return (NULL);
2863 }
2864 
2865 /*
2866  * When `need fragmentation' ICMP is received, update our idea of the MSS
2867  * based on the new value. Also nudge TCP to send something, since we
2868  * know the packet we just sent was dropped.
2869  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2870  */
2871 static struct inpcb *
2872 tcp_mtudisc_notify(struct inpcb *inp, int error)
2873 {
2874 
2875 	tcp_mtudisc(inp, -1);
2876 	return (inp);
2877 }
2878 
2879 static void
2880 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2881 {
2882 	struct tcpcb *tp;
2883 	struct socket *so;
2884 
2885 	INP_WLOCK_ASSERT(inp);
2886 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2887 	    (inp->inp_flags & INP_DROPPED))
2888 		return;
2889 
2890 	tp = intotcpcb(inp);
2891 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2892 
2893 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2894 
2895 	so = inp->inp_socket;
2896 	SOCKBUF_LOCK(&so->so_snd);
2897 	/* If the mss is larger than the socket buffer, decrease the mss. */
2898 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2899 		tp->t_maxseg = so->so_snd.sb_hiwat;
2900 	SOCKBUF_UNLOCK(&so->so_snd);
2901 
2902 	TCPSTAT_INC(tcps_mturesent);
2903 	tp->t_rtttime = 0;
2904 	tp->snd_nxt = tp->snd_una;
2905 	tcp_free_sackholes(tp);
2906 	tp->snd_recover = tp->snd_max;
2907 	if (tp->t_flags & TF_SACK_PERMIT)
2908 		EXIT_FASTRECOVERY(tp->t_flags);
2909 	tp->t_fb->tfb_tcp_output(tp);
2910 }
2911 
2912 #ifdef INET
2913 /*
2914  * Look-up the routing entry to the peer of this inpcb.  If no route
2915  * is found and it cannot be allocated, then return 0.  This routine
2916  * is called by TCP routines that access the rmx structure and by
2917  * tcp_mss_update to get the peer/interface MTU.
2918  */
2919 uint32_t
2920 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2921 {
2922 	struct nhop4_extended nh4;
2923 	struct ifnet *ifp;
2924 	uint32_t maxmtu = 0;
2925 
2926 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2927 
2928 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2929 
2930 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2931 		    NHR_REF, 0, &nh4) != 0)
2932 			return (0);
2933 
2934 		ifp = nh4.nh_ifp;
2935 		maxmtu = nh4.nh_mtu;
2936 
2937 		/* Report additional interface capabilities. */
2938 		if (cap != NULL) {
2939 			if (ifp->if_capenable & IFCAP_TSO4 &&
2940 			    ifp->if_hwassist & CSUM_TSO) {
2941 				cap->ifcap |= CSUM_TSO;
2942 				cap->tsomax = ifp->if_hw_tsomax;
2943 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2944 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2945 			}
2946 		}
2947 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2948 	}
2949 	return (maxmtu);
2950 }
2951 #endif /* INET */
2952 
2953 #ifdef INET6
2954 uint32_t
2955 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2956 {
2957 	struct nhop6_extended nh6;
2958 	struct in6_addr dst6;
2959 	uint32_t scopeid;
2960 	struct ifnet *ifp;
2961 	uint32_t maxmtu = 0;
2962 
2963 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2964 
2965 	if (inc->inc_flags & INC_IPV6MINMTU)
2966 		return (IPV6_MMTU);
2967 
2968 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2969 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2970 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2971 		    0, &nh6) != 0)
2972 			return (0);
2973 
2974 		ifp = nh6.nh_ifp;
2975 		maxmtu = nh6.nh_mtu;
2976 
2977 		/* Report additional interface capabilities. */
2978 		if (cap != NULL) {
2979 			if (ifp->if_capenable & IFCAP_TSO6 &&
2980 			    ifp->if_hwassist & CSUM_TSO) {
2981 				cap->ifcap |= CSUM_TSO;
2982 				cap->tsomax = ifp->if_hw_tsomax;
2983 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2984 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2985 			}
2986 		}
2987 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2988 	}
2989 
2990 	return (maxmtu);
2991 }
2992 #endif /* INET6 */
2993 
2994 /*
2995  * Calculate effective SMSS per RFC5681 definition for a given TCP
2996  * connection at its current state, taking into account SACK and etc.
2997  */
2998 u_int
2999 tcp_maxseg(const struct tcpcb *tp)
3000 {
3001 	u_int optlen;
3002 
3003 	if (tp->t_flags & TF_NOOPT)
3004 		return (tp->t_maxseg);
3005 
3006 	/*
3007 	 * Here we have a simplified code from tcp_addoptions(),
3008 	 * without a proper loop, and having most of paddings hardcoded.
3009 	 * We might make mistakes with padding here in some edge cases,
3010 	 * but this is harmless, since result of tcp_maxseg() is used
3011 	 * only in cwnd and ssthresh estimations.
3012 	 */
3013 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3014 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3015 		if (tp->t_flags & TF_RCVD_TSTMP)
3016 			optlen = TCPOLEN_TSTAMP_APPA;
3017 		else
3018 			optlen = 0;
3019 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3020 		if (tp->t_flags & TF_SIGNATURE)
3021 			optlen += PAD(TCPOLEN_SIGNATURE);
3022 #endif
3023 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3024 			optlen += TCPOLEN_SACKHDR;
3025 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3026 			optlen = PAD(optlen);
3027 		}
3028 	} else {
3029 		if (tp->t_flags & TF_REQ_TSTMP)
3030 			optlen = TCPOLEN_TSTAMP_APPA;
3031 		else
3032 			optlen = PAD(TCPOLEN_MAXSEG);
3033 		if (tp->t_flags & TF_REQ_SCALE)
3034 			optlen += PAD(TCPOLEN_WINDOW);
3035 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3036 		if (tp->t_flags & TF_SIGNATURE)
3037 			optlen += PAD(TCPOLEN_SIGNATURE);
3038 #endif
3039 		if (tp->t_flags & TF_SACK_PERMIT)
3040 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3041 	}
3042 #undef PAD
3043 	optlen = min(optlen, TCP_MAXOLEN);
3044 	return (tp->t_maxseg - optlen);
3045 }
3046 
3047 static int
3048 sysctl_drop(SYSCTL_HANDLER_ARGS)
3049 {
3050 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3051 	struct sockaddr_storage addrs[2];
3052 	struct inpcb *inp;
3053 	struct tcpcb *tp;
3054 	struct tcptw *tw;
3055 	struct sockaddr_in *fin, *lin;
3056 	struct epoch_tracker et;
3057 #ifdef INET6
3058 	struct sockaddr_in6 *fin6, *lin6;
3059 #endif
3060 	int error;
3061 
3062 	inp = NULL;
3063 	fin = lin = NULL;
3064 #ifdef INET6
3065 	fin6 = lin6 = NULL;
3066 #endif
3067 	error = 0;
3068 
3069 	if (req->oldptr != NULL || req->oldlen != 0)
3070 		return (EINVAL);
3071 	if (req->newptr == NULL)
3072 		return (EPERM);
3073 	if (req->newlen < sizeof(addrs))
3074 		return (ENOMEM);
3075 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3076 	if (error)
3077 		return (error);
3078 
3079 	switch (addrs[0].ss_family) {
3080 #ifdef INET6
3081 	case AF_INET6:
3082 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3083 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3084 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3085 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3086 			return (EINVAL);
3087 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3088 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3089 				return (EINVAL);
3090 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3091 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3092 			fin = (struct sockaddr_in *)&addrs[0];
3093 			lin = (struct sockaddr_in *)&addrs[1];
3094 			break;
3095 		}
3096 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3097 		if (error)
3098 			return (error);
3099 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3100 		if (error)
3101 			return (error);
3102 		break;
3103 #endif
3104 #ifdef INET
3105 	case AF_INET:
3106 		fin = (struct sockaddr_in *)&addrs[0];
3107 		lin = (struct sockaddr_in *)&addrs[1];
3108 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3109 		    lin->sin_len != sizeof(struct sockaddr_in))
3110 			return (EINVAL);
3111 		break;
3112 #endif
3113 	default:
3114 		return (EINVAL);
3115 	}
3116 	NET_EPOCH_ENTER(et);
3117 	switch (addrs[0].ss_family) {
3118 #ifdef INET6
3119 	case AF_INET6:
3120 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3121 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3122 		    INPLOOKUP_WLOCKPCB, NULL);
3123 		break;
3124 #endif
3125 #ifdef INET
3126 	case AF_INET:
3127 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3128 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3129 		break;
3130 #endif
3131 	}
3132 	if (inp != NULL) {
3133 		if (inp->inp_flags & INP_TIMEWAIT) {
3134 			/*
3135 			 * XXXRW: There currently exists a state where an
3136 			 * inpcb is present, but its timewait state has been
3137 			 * discarded.  For now, don't allow dropping of this
3138 			 * type of inpcb.
3139 			 */
3140 			tw = intotw(inp);
3141 			if (tw != NULL)
3142 				tcp_twclose(tw, 0);
3143 			else
3144 				INP_WUNLOCK(inp);
3145 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3146 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3147 			tp = intotcpcb(inp);
3148 			tp = tcp_drop(tp, ECONNABORTED);
3149 			if (tp != NULL)
3150 				INP_WUNLOCK(inp);
3151 		} else
3152 			INP_WUNLOCK(inp);
3153 	} else
3154 		error = ESRCH;
3155 	NET_EPOCH_EXIT(et);
3156 	return (error);
3157 }
3158 
3159 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3160     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3161     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3162     "Drop TCP connection");
3163 
3164 #ifdef KERN_TLS
3165 static int
3166 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3167 {
3168 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3169 	struct sockaddr_storage addrs[2];
3170 	struct inpcb *inp;
3171 	struct sockaddr_in *fin, *lin;
3172 	struct epoch_tracker et;
3173 #ifdef INET6
3174 	struct sockaddr_in6 *fin6, *lin6;
3175 #endif
3176 	int error;
3177 
3178 	inp = NULL;
3179 	fin = lin = NULL;
3180 #ifdef INET6
3181 	fin6 = lin6 = NULL;
3182 #endif
3183 	error = 0;
3184 
3185 	if (req->oldptr != NULL || req->oldlen != 0)
3186 		return (EINVAL);
3187 	if (req->newptr == NULL)
3188 		return (EPERM);
3189 	if (req->newlen < sizeof(addrs))
3190 		return (ENOMEM);
3191 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3192 	if (error)
3193 		return (error);
3194 
3195 	switch (addrs[0].ss_family) {
3196 #ifdef INET6
3197 	case AF_INET6:
3198 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3199 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3200 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3201 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3202 			return (EINVAL);
3203 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3204 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3205 				return (EINVAL);
3206 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3207 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3208 			fin = (struct sockaddr_in *)&addrs[0];
3209 			lin = (struct sockaddr_in *)&addrs[1];
3210 			break;
3211 		}
3212 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3213 		if (error)
3214 			return (error);
3215 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3216 		if (error)
3217 			return (error);
3218 		break;
3219 #endif
3220 #ifdef INET
3221 	case AF_INET:
3222 		fin = (struct sockaddr_in *)&addrs[0];
3223 		lin = (struct sockaddr_in *)&addrs[1];
3224 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3225 		    lin->sin_len != sizeof(struct sockaddr_in))
3226 			return (EINVAL);
3227 		break;
3228 #endif
3229 	default:
3230 		return (EINVAL);
3231 	}
3232 	NET_EPOCH_ENTER(et);
3233 	switch (addrs[0].ss_family) {
3234 #ifdef INET6
3235 	case AF_INET6:
3236 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3237 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3238 		    INPLOOKUP_WLOCKPCB, NULL);
3239 		break;
3240 #endif
3241 #ifdef INET
3242 	case AF_INET:
3243 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3244 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3245 		break;
3246 #endif
3247 	}
3248 	NET_EPOCH_EXIT(et);
3249 	if (inp != NULL) {
3250 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3251 		    inp->inp_socket == NULL) {
3252 			error = ECONNRESET;
3253 			INP_WUNLOCK(inp);
3254 		} else {
3255 			struct socket *so;
3256 
3257 			so = inp->inp_socket;
3258 			soref(so);
3259 			error = ktls_set_tx_mode(so,
3260 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3261 			INP_WUNLOCK(inp);
3262 			SOCK_LOCK(so);
3263 			sorele(so);
3264 		}
3265 	} else
3266 		error = ESRCH;
3267 	return (error);
3268 }
3269 
3270 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3271     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3272     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3273     "Switch TCP connection to SW TLS");
3274 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3275     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3276     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3277     "Switch TCP connection to ifnet TLS");
3278 #endif
3279 
3280 /*
3281  * Generate a standardized TCP log line for use throughout the
3282  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3283  * allow use in the interrupt context.
3284  *
3285  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3286  * NB: The function may return NULL if memory allocation failed.
3287  *
3288  * Due to header inclusion and ordering limitations the struct ip
3289  * and ip6_hdr pointers have to be passed as void pointers.
3290  */
3291 char *
3292 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3293     const void *ip6hdr)
3294 {
3295 
3296 	/* Is logging enabled? */
3297 	if (V_tcp_log_in_vain == 0)
3298 		return (NULL);
3299 
3300 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3301 }
3302 
3303 char *
3304 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3305     const void *ip6hdr)
3306 {
3307 
3308 	/* Is logging enabled? */
3309 	if (tcp_log_debug == 0)
3310 		return (NULL);
3311 
3312 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3313 }
3314 
3315 static char *
3316 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3317     const void *ip6hdr)
3318 {
3319 	char *s, *sp;
3320 	size_t size;
3321 	struct ip *ip;
3322 #ifdef INET6
3323 	const struct ip6_hdr *ip6;
3324 
3325 	ip6 = (const struct ip6_hdr *)ip6hdr;
3326 #endif /* INET6 */
3327 	ip = (struct ip *)ip4hdr;
3328 
3329 	/*
3330 	 * The log line looks like this:
3331 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3332 	 */
3333 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3334 	    sizeof(PRINT_TH_FLAGS) + 1 +
3335 #ifdef INET6
3336 	    2 * INET6_ADDRSTRLEN;
3337 #else
3338 	    2 * INET_ADDRSTRLEN;
3339 #endif /* INET6 */
3340 
3341 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3342 	if (s == NULL)
3343 		return (NULL);
3344 
3345 	strcat(s, "TCP: [");
3346 	sp = s + strlen(s);
3347 
3348 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3349 		inet_ntoa_r(inc->inc_faddr, sp);
3350 		sp = s + strlen(s);
3351 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3352 		sp = s + strlen(s);
3353 		inet_ntoa_r(inc->inc_laddr, sp);
3354 		sp = s + strlen(s);
3355 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3356 #ifdef INET6
3357 	} else if (inc) {
3358 		ip6_sprintf(sp, &inc->inc6_faddr);
3359 		sp = s + strlen(s);
3360 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3361 		sp = s + strlen(s);
3362 		ip6_sprintf(sp, &inc->inc6_laddr);
3363 		sp = s + strlen(s);
3364 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3365 	} else if (ip6 && th) {
3366 		ip6_sprintf(sp, &ip6->ip6_src);
3367 		sp = s + strlen(s);
3368 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3369 		sp = s + strlen(s);
3370 		ip6_sprintf(sp, &ip6->ip6_dst);
3371 		sp = s + strlen(s);
3372 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3373 #endif /* INET6 */
3374 #ifdef INET
3375 	} else if (ip && th) {
3376 		inet_ntoa_r(ip->ip_src, sp);
3377 		sp = s + strlen(s);
3378 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3379 		sp = s + strlen(s);
3380 		inet_ntoa_r(ip->ip_dst, sp);
3381 		sp = s + strlen(s);
3382 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3383 #endif /* INET */
3384 	} else {
3385 		free(s, M_TCPLOG);
3386 		return (NULL);
3387 	}
3388 	sp = s + strlen(s);
3389 	if (th)
3390 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3391 	if (*(s + size - 1) != '\0')
3392 		panic("%s: string too long", __func__);
3393 	return (s);
3394 }
3395 
3396 /*
3397  * A subroutine which makes it easy to track TCP state changes with DTrace.
3398  * This function shouldn't be called for t_state initializations that don't
3399  * correspond to actual TCP state transitions.
3400  */
3401 void
3402 tcp_state_change(struct tcpcb *tp, int newstate)
3403 {
3404 #if defined(KDTRACE_HOOKS)
3405 	int pstate = tp->t_state;
3406 #endif
3407 
3408 	TCPSTATES_DEC(tp->t_state);
3409 	TCPSTATES_INC(newstate);
3410 	tp->t_state = newstate;
3411 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3412 }
3413 
3414 /*
3415  * Create an external-format (``xtcpcb'') structure using the information in
3416  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3417  * reduce the spew of irrelevant information over this interface, to isolate
3418  * user code from changes in the kernel structure, and potentially to provide
3419  * information-hiding if we decide that some of this information should be
3420  * hidden from users.
3421  */
3422 void
3423 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3424 {
3425 	struct tcpcb *tp = intotcpcb(inp);
3426 	sbintime_t now;
3427 
3428 	bzero(xt, sizeof(*xt));
3429 	if (inp->inp_flags & INP_TIMEWAIT) {
3430 		xt->t_state = TCPS_TIME_WAIT;
3431 	} else {
3432 		xt->t_state = tp->t_state;
3433 		xt->t_logstate = tp->t_logstate;
3434 		xt->t_flags = tp->t_flags;
3435 		xt->t_sndzerowin = tp->t_sndzerowin;
3436 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3437 		xt->t_rcvoopack = tp->t_rcvoopack;
3438 
3439 		now = getsbinuptime();
3440 #define	COPYTIMER(ttt)	do {						\
3441 		if (callout_active(&tp->t_timers->ttt))			\
3442 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3443 			    SBT_1MS;					\
3444 		else							\
3445 			xt->ttt = 0;					\
3446 } while (0)
3447 		COPYTIMER(tt_delack);
3448 		COPYTIMER(tt_rexmt);
3449 		COPYTIMER(tt_persist);
3450 		COPYTIMER(tt_keep);
3451 		COPYTIMER(tt_2msl);
3452 #undef COPYTIMER
3453 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3454 
3455 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3456 		    TCP_FUNCTION_NAME_LEN_MAX);
3457 #ifdef TCP_BLACKBOX
3458 		(void)tcp_log_get_id(tp, xt->xt_logid);
3459 #endif
3460 	}
3461 
3462 	xt->xt_len = sizeof(struct xtcpcb);
3463 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3464 	if (inp->inp_socket == NULL)
3465 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3466 }
3467