xref: /freebsd/sys/netinet/tcp_subr.c (revision 7cd22ac43418da08448d0bab1009ff3cbda85120)
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/route/nhop.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
93 #ifdef INET6
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
101 #endif
102 
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_syncache.h>
110 #include <netinet/tcp_hpts.h>
111 #include <netinet/cc/cc.h>
112 #ifdef INET6
113 #include <netinet6/tcp6_var.h>
114 #endif
115 #include <netinet/tcpip.h>
116 #include <netinet/tcp_fastopen.h>
117 #ifdef TCPPCAP
118 #include <netinet/tcp_pcap.h>
119 #endif
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif
123 #ifdef INET6
124 #include <netinet6/ip6protosw.h>
125 #endif
126 #ifdef TCP_OFFLOAD
127 #include <netinet/tcp_offload.h>
128 #endif
129 #include <netinet/udp.h>
130 #include <netinet/udp_var.h>
131 
132 #include <netipsec/ipsec_support.h>
133 
134 #include <machine/in_cksum.h>
135 #include <crypto/siphash/siphash.h>
136 
137 #include <security/mac/mac_framework.h>
138 
139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
140 #ifdef INET6
141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
142 #endif
143 
144 #ifdef NETFLIX_EXP_DETECTION
145 /*  Sack attack detection thresholds and such */
146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
147     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
148     "Sack Attack detection thresholds");
149 int32_t tcp_force_detection = 0;
150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
151     CTLFLAG_RW,
152     &tcp_force_detection, 0,
153     "Do we force detection even if the INP has it off?");
154 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
156     CTLFLAG_RW,
157     &tcp_sack_to_ack_thresh, 700,
158     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
159 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
161     CTLFLAG_RW,
162     &tcp_sack_to_move_thresh, 600,
163     "Percentage of sack moves we must see above (10.1 percent is 101)");
164 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
166     CTLFLAG_RW,
167     &tcp_restoral_thresh, 550,
168     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
169 int32_t tcp_sad_decay_val = 800;
170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
171     CTLFLAG_RW,
172     &tcp_sad_decay_val, 800,
173     "The decay percentage (10.1 percent equals 101 )");
174 int32_t tcp_map_minimum = 500;
175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
176     CTLFLAG_RW,
177     &tcp_map_minimum, 500,
178     "Number of Map enteries before we start detection");
179 int32_t tcp_attack_on_turns_on_logging = 0;
180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
181     CTLFLAG_RW,
182     &tcp_attack_on_turns_on_logging, 0,
183    "When we have a positive hit on attack, do we turn on logging?");
184 int32_t tcp_sad_pacing_interval = 2000;
185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
186     CTLFLAG_RW,
187     &tcp_sad_pacing_interval, 2000,
188     "What is the minimum pacing interval for a classified attacker?");
189 
190 int32_t tcp_sad_low_pps = 100;
191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
192     CTLFLAG_RW,
193     &tcp_sad_low_pps, 100,
194     "What is the input pps that below which we do not decay?");
195 #endif
196 uint32_t tcp_ack_war_time_window = 1000;
197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
198     CTLFLAG_RW,
199     &tcp_ack_war_time_window, 1000,
200    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
201 uint32_t tcp_ack_war_cnt = 5;
202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
203     CTLFLAG_RW,
204     &tcp_ack_war_cnt, 5,
205    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
206 
207 struct rwlock tcp_function_lock;
208 
209 static int
210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
211 {
212 	int error, new;
213 
214 	new = V_tcp_mssdflt;
215 	error = sysctl_handle_int(oidp, &new, 0, req);
216 	if (error == 0 && req->newptr) {
217 		if (new < TCP_MINMSS)
218 			error = EINVAL;
219 		else
220 			V_tcp_mssdflt = new;
221 	}
222 	return (error);
223 }
224 
225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
226     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
227     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
228     "Default TCP Maximum Segment Size");
229 
230 #ifdef INET6
231 static int
232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
233 {
234 	int error, new;
235 
236 	new = V_tcp_v6mssdflt;
237 	error = sysctl_handle_int(oidp, &new, 0, req);
238 	if (error == 0 && req->newptr) {
239 		if (new < TCP_MINMSS)
240 			error = EINVAL;
241 		else
242 			V_tcp_v6mssdflt = new;
243 	}
244 	return (error);
245 }
246 
247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
248     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
250    "Default TCP Maximum Segment Size for IPv6");
251 #endif /* INET6 */
252 
253 /*
254  * Minimum MSS we accept and use. This prevents DoS attacks where
255  * we are forced to a ridiculous low MSS like 20 and send hundreds
256  * of packets instead of one. The effect scales with the available
257  * bandwidth and quickly saturates the CPU and network interface
258  * with packet generation and sending. Set to zero to disable MINMSS
259  * checking. This setting prevents us from sending too small packets.
260  */
261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
263      &VNET_NAME(tcp_minmss), 0,
264     "Minimum TCP Maximum Segment Size");
265 
266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
268     &VNET_NAME(tcp_do_rfc1323), 0,
269     "Enable rfc1323 (high performance TCP) extensions");
270 
271 /*
272  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
273  * Some stacks negotiate TS, but never send them after connection setup. Some
274  * stacks negotiate TS, but don't send them when sending keep-alive segments.
275  * These include modern widely deployed TCP stacks.
276  * Therefore tolerating violations for now...
277  */
278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
280     &VNET_NAME(tcp_tolerate_missing_ts), 0,
281     "Tolerate missing TCP timestamps");
282 
283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
285     &VNET_NAME(tcp_ts_offset_per_conn), 0,
286     "Initialize TCP timestamps per connection instead of per host pair");
287 
288 /* How many connections are pacing */
289 static volatile uint32_t number_of_tcp_connections_pacing = 0;
290 static uint32_t shadow_num_connections = 0;
291 
292 static int tcp_pacing_limit = 10000;
293 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
294     &tcp_pacing_limit, 1000,
295     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
296 
297 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
298     &shadow_num_connections, 0, "Number of TCP connections being paced");
299 
300 static int	tcp_log_debug = 0;
301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
302     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
303 
304 static int	tcp_tcbhashsize;
305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
306     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
307 
308 static int	do_tcpdrain = 1;
309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
310     "Enable tcp_drain routine for extra help when low on mbufs");
311 
312 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
313     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
314 
315 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
316 #define	V_icmp_may_rst			VNET(icmp_may_rst)
317 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
318     &VNET_NAME(icmp_may_rst), 0,
319     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
320 
321 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
322 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
324     &VNET_NAME(tcp_isn_reseed_interval), 0,
325     "Seconds between reseeding of ISN secret");
326 
327 static int	tcp_soreceive_stream;
328 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
329     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
330 
331 VNET_DEFINE(uma_zone_t, sack_hole_zone);
332 #define	V_sack_hole_zone		VNET(sack_hole_zone)
333 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
334 static int
335 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
336 {
337 	int error;
338 	uint32_t new;
339 
340 	new = V_tcp_map_entries_limit;
341 	error = sysctl_handle_int(oidp, &new, 0, req);
342 	if (error == 0 && req->newptr) {
343 		/* only allow "0" and value > minimum */
344 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
345 			error = EINVAL;
346 		else
347 			V_tcp_map_entries_limit = new;
348 	}
349 	return (error);
350 }
351 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
352     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
353     &VNET_NAME(tcp_map_entries_limit), 0,
354     &sysctl_net_inet_tcp_map_limit_check, "IU",
355     "Total sendmap entries limit");
356 
357 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
358 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
359      &VNET_NAME(tcp_map_split_limit), 0,
360     "Total sendmap split entries limit");
361 
362 #ifdef TCP_HHOOK
363 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
364 #endif
365 
366 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
367 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
368 #define	V_ts_offset_secret	VNET(ts_offset_secret)
369 
370 static int	tcp_default_fb_init(struct tcpcb *tp);
371 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
372 static int	tcp_default_handoff_ok(struct tcpcb *tp);
373 static struct inpcb *tcp_notify(struct inpcb *, int);
374 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
375 static void tcp_mtudisc(struct inpcb *, int);
376 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
377 		    void *ip4hdr, const void *ip6hdr);
378 
379 static struct tcp_function_block tcp_def_funcblk = {
380 	.tfb_tcp_block_name = "freebsd",
381 	.tfb_tcp_output = tcp_output,
382 	.tfb_tcp_do_segment = tcp_do_segment,
383 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
384 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
385 	.tfb_tcp_fb_init = tcp_default_fb_init,
386 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
387 };
388 
389 static int tcp_fb_cnt = 0;
390 struct tcp_funchead t_functions;
391 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
392 
393 static struct tcp_function_block *
394 find_tcp_functions_locked(struct tcp_function_set *fs)
395 {
396 	struct tcp_function *f;
397 	struct tcp_function_block *blk=NULL;
398 
399 	TAILQ_FOREACH(f, &t_functions, tf_next) {
400 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
401 			blk = f->tf_fb;
402 			break;
403 		}
404 	}
405 	return(blk);
406 }
407 
408 static struct tcp_function_block *
409 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
410 {
411 	struct tcp_function_block *rblk=NULL;
412 	struct tcp_function *f;
413 
414 	TAILQ_FOREACH(f, &t_functions, tf_next) {
415 		if (f->tf_fb == blk) {
416 			rblk = blk;
417 			if (s) {
418 				*s = f;
419 			}
420 			break;
421 		}
422 	}
423 	return (rblk);
424 }
425 
426 struct tcp_function_block *
427 find_and_ref_tcp_functions(struct tcp_function_set *fs)
428 {
429 	struct tcp_function_block *blk;
430 
431 	rw_rlock(&tcp_function_lock);
432 	blk = find_tcp_functions_locked(fs);
433 	if (blk)
434 		refcount_acquire(&blk->tfb_refcnt);
435 	rw_runlock(&tcp_function_lock);
436 	return(blk);
437 }
438 
439 struct tcp_function_block *
440 find_and_ref_tcp_fb(struct tcp_function_block *blk)
441 {
442 	struct tcp_function_block *rblk;
443 
444 	rw_rlock(&tcp_function_lock);
445 	rblk = find_tcp_fb_locked(blk, NULL);
446 	if (rblk)
447 		refcount_acquire(&rblk->tfb_refcnt);
448 	rw_runlock(&tcp_function_lock);
449 	return(rblk);
450 }
451 
452 static struct tcp_function_block *
453 find_and_ref_tcp_default_fb(void)
454 {
455 	struct tcp_function_block *rblk;
456 
457 	rw_rlock(&tcp_function_lock);
458 	rblk = tcp_func_set_ptr;
459 	refcount_acquire(&rblk->tfb_refcnt);
460 	rw_runlock(&tcp_function_lock);
461 	return (rblk);
462 }
463 
464 void
465 tcp_switch_back_to_default(struct tcpcb *tp)
466 {
467 	struct tcp_function_block *tfb;
468 
469 	KASSERT(tp->t_fb != &tcp_def_funcblk,
470 	    ("%s: called by the built-in default stack", __func__));
471 
472 	/*
473 	 * Release the old stack. This function will either find a new one
474 	 * or panic.
475 	 */
476 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
477 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
478 	refcount_release(&tp->t_fb->tfb_refcnt);
479 
480 	/*
481 	 * Now, we'll find a new function block to use.
482 	 * Start by trying the current user-selected
483 	 * default, unless this stack is the user-selected
484 	 * default.
485 	 */
486 	tfb = find_and_ref_tcp_default_fb();
487 	if (tfb == tp->t_fb) {
488 		refcount_release(&tfb->tfb_refcnt);
489 		tfb = NULL;
490 	}
491 	/* Does the stack accept this connection? */
492 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
493 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
494 		refcount_release(&tfb->tfb_refcnt);
495 		tfb = NULL;
496 	}
497 	/* Try to use that stack. */
498 	if (tfb != NULL) {
499 		/* Initialize the new stack. If it succeeds, we are done. */
500 		tp->t_fb = tfb;
501 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
502 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
503 			return;
504 
505 		/*
506 		 * Initialization failed. Release the reference count on
507 		 * the stack.
508 		 */
509 		refcount_release(&tfb->tfb_refcnt);
510 	}
511 
512 	/*
513 	 * If that wasn't feasible, use the built-in default
514 	 * stack which is not allowed to reject anyone.
515 	 */
516 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
517 	if (tfb == NULL) {
518 		/* there always should be a default */
519 		panic("Can't refer to tcp_def_funcblk");
520 	}
521 	if (tfb->tfb_tcp_handoff_ok != NULL) {
522 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
523 			/* The default stack cannot say no */
524 			panic("Default stack rejects a new session?");
525 		}
526 	}
527 	tp->t_fb = tfb;
528 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
529 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
530 		/* The default stack cannot fail */
531 		panic("Default stack initialization failed");
532 	}
533 }
534 
535 static void
536 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
537     const struct sockaddr *sa, void *ctx)
538 {
539 	struct ip *iph;
540 #ifdef INET6
541 	struct ip6_hdr *ip6;
542 #endif
543 	struct udphdr *uh;
544 	struct tcphdr *th;
545 	int thlen;
546 	uint16_t port;
547 
548 	TCPSTAT_INC(tcps_tunneled_pkts);
549 	if ((m->m_flags & M_PKTHDR) == 0) {
550 		/* Can't handle one that is not a pkt hdr */
551 		TCPSTAT_INC(tcps_tunneled_errs);
552 		goto out;
553 	}
554 	thlen = sizeof(struct tcphdr);
555 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
556 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
557 		TCPSTAT_INC(tcps_tunneled_errs);
558 		goto out;
559 	}
560 	iph = mtod(m, struct ip *);
561 	uh = (struct udphdr *)((caddr_t)iph + off);
562 	th = (struct tcphdr *)(uh + 1);
563 	thlen = th->th_off << 2;
564 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
565 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
566 		if (m == NULL) {
567 			TCPSTAT_INC(tcps_tunneled_errs);
568 			goto out;
569 		} else {
570 			iph = mtod(m, struct ip *);
571 			uh = (struct udphdr *)((caddr_t)iph + off);
572 			th = (struct tcphdr *)(uh + 1);
573 		}
574 	}
575 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
576 	bcopy(th, uh, m->m_len - off);
577 	m->m_len -= sizeof(struct udphdr);
578 	m->m_pkthdr.len -= sizeof(struct udphdr);
579 	/*
580 	 * We use the same algorithm for
581 	 * both UDP and TCP for c-sum. So
582 	 * the code in tcp_input will skip
583 	 * the checksum. So we do nothing
584 	 * with the flag (m->m_pkthdr.csum_flags).
585 	 */
586 	switch (iph->ip_v) {
587 #ifdef INET
588 	case IPVERSION:
589 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
590 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
591 		break;
592 #endif
593 #ifdef INET6
594 	case IPV6_VERSION >> 4:
595 		ip6 = mtod(m, struct ip6_hdr *);
596 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
597 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
598 		break;
599 #endif
600 	default:
601 		goto out;
602 		break;
603 	}
604 	return;
605 out:
606 	m_freem(m);
607 }
608 
609 static int
610 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
611 {
612 	int error=ENOENT;
613 	struct tcp_function_set fs;
614 	struct tcp_function_block *blk;
615 
616 	memset(&fs, 0, sizeof(fs));
617 	rw_rlock(&tcp_function_lock);
618 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
619 	if (blk) {
620 		/* Found him */
621 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
622 		fs.pcbcnt = blk->tfb_refcnt;
623 	}
624 	rw_runlock(&tcp_function_lock);
625 	error = sysctl_handle_string(oidp, fs.function_set_name,
626 				     sizeof(fs.function_set_name), req);
627 
628 	/* Check for error or no change */
629 	if (error != 0 || req->newptr == NULL)
630 		return(error);
631 
632 	rw_wlock(&tcp_function_lock);
633 	blk = find_tcp_functions_locked(&fs);
634 	if ((blk == NULL) ||
635 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
636 		error = ENOENT;
637 		goto done;
638 	}
639 	tcp_func_set_ptr = blk;
640 done:
641 	rw_wunlock(&tcp_function_lock);
642 	return (error);
643 }
644 
645 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
646     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
647     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
648     "Set/get the default TCP functions");
649 
650 static int
651 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
652 {
653 	int error, cnt, linesz;
654 	struct tcp_function *f;
655 	char *buffer, *cp;
656 	size_t bufsz, outsz;
657 	bool alias;
658 
659 	cnt = 0;
660 	rw_rlock(&tcp_function_lock);
661 	TAILQ_FOREACH(f, &t_functions, tf_next) {
662 		cnt++;
663 	}
664 	rw_runlock(&tcp_function_lock);
665 
666 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
667 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
668 
669 	error = 0;
670 	cp = buffer;
671 
672 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
673 	    "Alias", "PCB count");
674 	cp += linesz;
675 	bufsz -= linesz;
676 	outsz = linesz;
677 
678 	rw_rlock(&tcp_function_lock);
679 	TAILQ_FOREACH(f, &t_functions, tf_next) {
680 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
681 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
682 		    f->tf_fb->tfb_tcp_block_name,
683 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
684 		    alias ? f->tf_name : "-",
685 		    f->tf_fb->tfb_refcnt);
686 		if (linesz >= bufsz) {
687 			error = EOVERFLOW;
688 			break;
689 		}
690 		cp += linesz;
691 		bufsz -= linesz;
692 		outsz += linesz;
693 	}
694 	rw_runlock(&tcp_function_lock);
695 	if (error == 0)
696 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
697 	free(buffer, M_TEMP);
698 	return (error);
699 }
700 
701 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
702     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
703     NULL, 0, sysctl_net_inet_list_available, "A",
704     "list available TCP Function sets");
705 
706 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
707 
708 #ifdef INET
709 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
710 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
711 #endif
712 #ifdef INET6
713 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
714 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
715 #endif
716 
717 static void
718 tcp_over_udp_stop(void)
719 {
720 	/*
721 	 * This function assumes sysctl caller holds inp_rinfo_lock()
722 	 * for writting!
723 	 */
724 #ifdef INET
725 	if (V_udp4_tun_socket != NULL) {
726 		soclose(V_udp4_tun_socket);
727 		V_udp4_tun_socket = NULL;
728 	}
729 #endif
730 #ifdef INET6
731 	if (V_udp6_tun_socket != NULL) {
732 		soclose(V_udp6_tun_socket);
733 		V_udp6_tun_socket = NULL;
734 	}
735 #endif
736 }
737 
738 static int
739 tcp_over_udp_start(void)
740 {
741 	uint16_t port;
742 	int ret;
743 #ifdef INET
744 	struct sockaddr_in sin;
745 #endif
746 #ifdef INET6
747 	struct sockaddr_in6 sin6;
748 #endif
749 	/*
750 	 * This function assumes sysctl caller holds inp_info_rlock()
751 	 * for writting!
752 	 */
753 	port = V_tcp_udp_tunneling_port;
754 	if (ntohs(port) == 0) {
755 		/* Must have a port set */
756 		return (EINVAL);
757 	}
758 #ifdef INET
759 	if (V_udp4_tun_socket != NULL) {
760 		/* Already running -- must stop first */
761 		return (EALREADY);
762 	}
763 #endif
764 #ifdef INET6
765 	if (V_udp6_tun_socket != NULL) {
766 		/* Already running -- must stop first */
767 		return (EALREADY);
768 	}
769 #endif
770 #ifdef INET
771 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
772 	    SOCK_DGRAM, IPPROTO_UDP,
773 	    curthread->td_ucred, curthread))) {
774 		tcp_over_udp_stop();
775 		return (ret);
776 	}
777 	/* Call the special UDP hook. */
778 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
779 	    tcp_recv_udp_tunneled_packet,
780 	    tcp_ctlinput_viaudp,
781 	    NULL))) {
782 		tcp_over_udp_stop();
783 		return (ret);
784 	}
785 	/* Ok, we have a socket, bind it to the port. */
786 	memset(&sin, 0, sizeof(struct sockaddr_in));
787 	sin.sin_len = sizeof(struct sockaddr_in);
788 	sin.sin_family = AF_INET;
789 	sin.sin_port = htons(port);
790 	if ((ret = sobind(V_udp4_tun_socket,
791 	    (struct sockaddr *)&sin, curthread))) {
792 		tcp_over_udp_stop();
793 		return (ret);
794 	}
795 #endif
796 #ifdef INET6
797 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
798 	    SOCK_DGRAM, IPPROTO_UDP,
799 	    curthread->td_ucred, curthread))) {
800 		tcp_over_udp_stop();
801 		return (ret);
802 	}
803 	/* Call the special UDP hook. */
804 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
805 	    tcp_recv_udp_tunneled_packet,
806 	    tcp6_ctlinput_viaudp,
807 	    NULL))) {
808 		tcp_over_udp_stop();
809 		return (ret);
810 	}
811 	/* Ok, we have a socket, bind it to the port. */
812 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
813 	sin6.sin6_len = sizeof(struct sockaddr_in6);
814 	sin6.sin6_family = AF_INET6;
815 	sin6.sin6_port = htons(port);
816 	if ((ret = sobind(V_udp6_tun_socket,
817 	    (struct sockaddr *)&sin6, curthread))) {
818 		tcp_over_udp_stop();
819 		return (ret);
820 	}
821 #endif
822 	return (0);
823 }
824 
825 static int
826 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
827 {
828 	int error;
829 	uint32_t old, new;
830 
831 	old = V_tcp_udp_tunneling_port;
832 	new = old;
833 	error = sysctl_handle_int(oidp, &new, 0, req);
834 	if ((error == 0) &&
835 	    (req->newptr != NULL)) {
836 		if ((new < TCP_TUNNELING_PORT_MIN) ||
837 		    (new > TCP_TUNNELING_PORT_MAX)) {
838 			error = EINVAL;
839 		} else {
840 			V_tcp_udp_tunneling_port = new;
841 			if (old != 0) {
842 				tcp_over_udp_stop();
843 			}
844 			if (new != 0) {
845 				error = tcp_over_udp_start();
846 			}
847 		}
848 	}
849 	return (error);
850 }
851 
852 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
853     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
854     &VNET_NAME(tcp_udp_tunneling_port),
855     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
856     "Tunneling port for tcp over udp");
857 
858 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
859 
860 static int
861 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
862 {
863 	int error, new;
864 
865 	new = V_tcp_udp_tunneling_overhead;
866 	error = sysctl_handle_int(oidp, &new, 0, req);
867 	if (error == 0 && req->newptr) {
868 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
869 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
870 			error = EINVAL;
871 		else
872 			V_tcp_udp_tunneling_overhead = new;
873 	}
874 	return (error);
875 }
876 
877 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
878     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
879     &VNET_NAME(tcp_udp_tunneling_overhead),
880     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
881     "MSS reduction when using tcp over udp");
882 
883 /*
884  * Exports one (struct tcp_function_info) for each alias/name.
885  */
886 static int
887 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
888 {
889 	int cnt, error;
890 	struct tcp_function *f;
891 	struct tcp_function_info tfi;
892 
893 	/*
894 	 * We don't allow writes.
895 	 */
896 	if (req->newptr != NULL)
897 		return (EINVAL);
898 
899 	/*
900 	 * Wire the old buffer so we can directly copy the functions to
901 	 * user space without dropping the lock.
902 	 */
903 	if (req->oldptr != NULL) {
904 		error = sysctl_wire_old_buffer(req, 0);
905 		if (error)
906 			return (error);
907 	}
908 
909 	/*
910 	 * Walk the list and copy out matching entries. If INVARIANTS
911 	 * is compiled in, also walk the list to verify the length of
912 	 * the list matches what we have recorded.
913 	 */
914 	rw_rlock(&tcp_function_lock);
915 
916 	cnt = 0;
917 #ifndef INVARIANTS
918 	if (req->oldptr == NULL) {
919 		cnt = tcp_fb_cnt;
920 		goto skip_loop;
921 	}
922 #endif
923 	TAILQ_FOREACH(f, &t_functions, tf_next) {
924 #ifdef INVARIANTS
925 		cnt++;
926 #endif
927 		if (req->oldptr != NULL) {
928 			bzero(&tfi, sizeof(tfi));
929 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
930 			tfi.tfi_id = f->tf_fb->tfb_id;
931 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
932 			    sizeof(tfi.tfi_alias));
933 			(void)strlcpy(tfi.tfi_name,
934 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
935 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
936 			/*
937 			 * Don't stop on error, as that is the
938 			 * mechanism we use to accumulate length
939 			 * information if the buffer was too short.
940 			 */
941 		}
942 	}
943 	KASSERT(cnt == tcp_fb_cnt,
944 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
945 #ifndef INVARIANTS
946 skip_loop:
947 #endif
948 	rw_runlock(&tcp_function_lock);
949 	if (req->oldptr == NULL)
950 		error = SYSCTL_OUT(req, NULL,
951 		    (cnt + 1) * sizeof(struct tcp_function_info));
952 
953 	return (error);
954 }
955 
956 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
957 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
958 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
959 	    "List TCP function block name-to-ID mappings");
960 
961 /*
962  * tfb_tcp_handoff_ok() function for the default stack.
963  * Note that we'll basically try to take all comers.
964  */
965 static int
966 tcp_default_handoff_ok(struct tcpcb *tp)
967 {
968 
969 	return (0);
970 }
971 
972 /*
973  * tfb_tcp_fb_init() function for the default stack.
974  *
975  * This handles making sure we have appropriate timers set if you are
976  * transitioning a socket that has some amount of setup done.
977  *
978  * The init() fuction from the default can *never* return non-zero i.e.
979  * it is required to always succeed since it is the stack of last resort!
980  */
981 static int
982 tcp_default_fb_init(struct tcpcb *tp)
983 {
984 
985 	struct socket *so;
986 
987 	INP_WLOCK_ASSERT(tp->t_inpcb);
988 
989 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
990 	    ("%s: connection %p in unexpected state %d", __func__, tp,
991 	    tp->t_state));
992 
993 	/*
994 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
995 	 * know what to do for unexpected states (which includes TIME_WAIT).
996 	 */
997 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
998 		return (0);
999 
1000 	/*
1001 	 * Make sure some kind of transmission timer is set if there is
1002 	 * outstanding data.
1003 	 */
1004 	so = tp->t_inpcb->inp_socket;
1005 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1006 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1007 	    tcp_timer_active(tp, TT_PERSIST))) {
1008 		/*
1009 		 * If the session has established and it looks like it should
1010 		 * be in the persist state, set the persist timer. Otherwise,
1011 		 * set the retransmit timer.
1012 		 */
1013 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1014 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1015 		    (int32_t)sbavail(&so->so_snd))
1016 			tcp_setpersist(tp);
1017 		else
1018 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1019 	}
1020 
1021 	/* All non-embryonic sessions get a keepalive timer. */
1022 	if (!tcp_timer_active(tp, TT_KEEP))
1023 		tcp_timer_activate(tp, TT_KEEP,
1024 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1025 		    TP_KEEPINIT(tp));
1026 
1027 	/*
1028 	 * Make sure critical variables are initialized
1029 	 * if transitioning while in Recovery.
1030 	 */
1031 	if IN_FASTRECOVERY(tp->t_flags) {
1032 		if (tp->sackhint.recover_fs == 0)
1033 			tp->sackhint.recover_fs = max(1,
1034 			    tp->snd_nxt - tp->snd_una);
1035 	}
1036 
1037 	return (0);
1038 }
1039 
1040 /*
1041  * tfb_tcp_fb_fini() function for the default stack.
1042  *
1043  * This changes state as necessary (or prudent) to prepare for another stack
1044  * to assume responsibility for the connection.
1045  */
1046 static void
1047 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1048 {
1049 
1050 	INP_WLOCK_ASSERT(tp->t_inpcb);
1051 	return;
1052 }
1053 
1054 /*
1055  * Target size of TCP PCB hash tables. Must be a power of two.
1056  *
1057  * Note that this can be overridden by the kernel environment
1058  * variable net.inet.tcp.tcbhashsize
1059  */
1060 #ifndef TCBHASHSIZE
1061 #define TCBHASHSIZE	0
1062 #endif
1063 
1064 /*
1065  * XXX
1066  * Callouts should be moved into struct tcp directly.  They are currently
1067  * separate because the tcpcb structure is exported to userland for sysctl
1068  * parsing purposes, which do not know about callouts.
1069  */
1070 struct tcpcb_mem {
1071 	struct	tcpcb		tcb;
1072 	struct	tcp_timer	tt;
1073 	struct	cc_var		ccv;
1074 #ifdef TCP_HHOOK
1075 	struct	osd		osd;
1076 #endif
1077 };
1078 
1079 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1080 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1081 
1082 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1083 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1084 
1085 static struct mtx isn_mtx;
1086 
1087 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1088 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1089 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1090 
1091 /*
1092  * TCP initialization.
1093  */
1094 static void
1095 tcp_zone_change(void *tag)
1096 {
1097 
1098 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
1099 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1100 	tcp_tw_zone_change();
1101 }
1102 
1103 static int
1104 tcp_inpcb_init(void *mem, int size, int flags)
1105 {
1106 	struct inpcb *inp = mem;
1107 
1108 	INP_LOCK_INIT(inp, "inp", "tcpinp");
1109 	return (0);
1110 }
1111 
1112 /*
1113  * Take a value and get the next power of 2 that doesn't overflow.
1114  * Used to size the tcp_inpcb hash buckets.
1115  */
1116 static int
1117 maketcp_hashsize(int size)
1118 {
1119 	int hashsize;
1120 
1121 	/*
1122 	 * auto tune.
1123 	 * get the next power of 2 higher than maxsockets.
1124 	 */
1125 	hashsize = 1 << fls(size);
1126 	/* catch overflow, and just go one power of 2 smaller */
1127 	if (hashsize < size) {
1128 		hashsize = 1 << (fls(size) - 1);
1129 	}
1130 	return (hashsize);
1131 }
1132 
1133 static volatile int next_tcp_stack_id = 1;
1134 
1135 /*
1136  * Register a TCP function block with the name provided in the names
1137  * array.  (Note that this function does NOT automatically register
1138  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1139  * explicitly include blk->tfb_tcp_block_name in the list of names if
1140  * you wish to register the stack with that name.)
1141  *
1142  * Either all name registrations will succeed or all will fail.  If
1143  * a name registration fails, the function will update the num_names
1144  * argument to point to the array index of the name that encountered
1145  * the failure.
1146  *
1147  * Returns 0 on success, or an error code on failure.
1148  */
1149 int
1150 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1151     const char *names[], int *num_names)
1152 {
1153 	struct tcp_function *n;
1154 	struct tcp_function_set fs;
1155 	int error, i;
1156 
1157 	KASSERT(names != NULL && *num_names > 0,
1158 	    ("%s: Called with 0-length name list", __func__));
1159 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1160 	KASSERT(rw_initialized(&tcp_function_lock),
1161 	    ("%s: called too early", __func__));
1162 
1163 	if ((blk->tfb_tcp_output == NULL) ||
1164 	    (blk->tfb_tcp_do_segment == NULL) ||
1165 	    (blk->tfb_tcp_ctloutput == NULL) ||
1166 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1167 		/*
1168 		 * These functions are required and you
1169 		 * need a name.
1170 		 */
1171 		*num_names = 0;
1172 		return (EINVAL);
1173 	}
1174 	if (blk->tfb_tcp_timer_stop_all ||
1175 	    blk->tfb_tcp_timer_activate ||
1176 	    blk->tfb_tcp_timer_active ||
1177 	    blk->tfb_tcp_timer_stop) {
1178 		/*
1179 		 * If you define one timer function you
1180 		 * must have them all.
1181 		 */
1182 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1183 		    (blk->tfb_tcp_timer_activate == NULL) ||
1184 		    (blk->tfb_tcp_timer_active == NULL) ||
1185 		    (blk->tfb_tcp_timer_stop == NULL)) {
1186 			*num_names = 0;
1187 			return (EINVAL);
1188 		}
1189 	}
1190 
1191 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1192 		*num_names = 0;
1193 		return (EINVAL);
1194 	}
1195 
1196 	refcount_init(&blk->tfb_refcnt, 0);
1197 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1198 	for (i = 0; i < *num_names; i++) {
1199 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1200 		if (n == NULL) {
1201 			error = ENOMEM;
1202 			goto cleanup;
1203 		}
1204 		n->tf_fb = blk;
1205 
1206 		(void)strlcpy(fs.function_set_name, names[i],
1207 		    sizeof(fs.function_set_name));
1208 		rw_wlock(&tcp_function_lock);
1209 		if (find_tcp_functions_locked(&fs) != NULL) {
1210 			/* Duplicate name space not allowed */
1211 			rw_wunlock(&tcp_function_lock);
1212 			free(n, M_TCPFUNCTIONS);
1213 			error = EALREADY;
1214 			goto cleanup;
1215 		}
1216 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1217 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1218 		tcp_fb_cnt++;
1219 		rw_wunlock(&tcp_function_lock);
1220 	}
1221 	return(0);
1222 
1223 cleanup:
1224 	/*
1225 	 * Deregister the names we just added. Because registration failed
1226 	 * for names[i], we don't need to deregister that name.
1227 	 */
1228 	*num_names = i;
1229 	rw_wlock(&tcp_function_lock);
1230 	while (--i >= 0) {
1231 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1232 			if (!strncmp(n->tf_name, names[i],
1233 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1234 				TAILQ_REMOVE(&t_functions, n, tf_next);
1235 				tcp_fb_cnt--;
1236 				n->tf_fb = NULL;
1237 				free(n, M_TCPFUNCTIONS);
1238 				break;
1239 			}
1240 		}
1241 	}
1242 	rw_wunlock(&tcp_function_lock);
1243 	return (error);
1244 }
1245 
1246 /*
1247  * Register a TCP function block using the name provided in the name
1248  * argument.
1249  *
1250  * Returns 0 on success, or an error code on failure.
1251  */
1252 int
1253 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1254     int wait)
1255 {
1256 	const char *name_list[1];
1257 	int num_names, rv;
1258 
1259 	num_names = 1;
1260 	if (name != NULL)
1261 		name_list[0] = name;
1262 	else
1263 		name_list[0] = blk->tfb_tcp_block_name;
1264 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1265 	return (rv);
1266 }
1267 
1268 /*
1269  * Register a TCP function block using the name defined in
1270  * blk->tfb_tcp_block_name.
1271  *
1272  * Returns 0 on success, or an error code on failure.
1273  */
1274 int
1275 register_tcp_functions(struct tcp_function_block *blk, int wait)
1276 {
1277 
1278 	return (register_tcp_functions_as_name(blk, NULL, wait));
1279 }
1280 
1281 /*
1282  * Deregister all names associated with a function block. This
1283  * functionally removes the function block from use within the system.
1284  *
1285  * When called with a true quiesce argument, mark the function block
1286  * as being removed so no more stacks will use it and determine
1287  * whether the removal would succeed.
1288  *
1289  * When called with a false quiesce argument, actually attempt the
1290  * removal.
1291  *
1292  * When called with a force argument, attempt to switch all TCBs to
1293  * use the default stack instead of returning EBUSY.
1294  *
1295  * Returns 0 on success (or if the removal would succeed, or an error
1296  * code on failure.
1297  */
1298 int
1299 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1300     bool force)
1301 {
1302 	struct tcp_function *f;
1303 
1304 	if (blk == &tcp_def_funcblk) {
1305 		/* You can't un-register the default */
1306 		return (EPERM);
1307 	}
1308 	rw_wlock(&tcp_function_lock);
1309 	if (blk == tcp_func_set_ptr) {
1310 		/* You can't free the current default */
1311 		rw_wunlock(&tcp_function_lock);
1312 		return (EBUSY);
1313 	}
1314 	/* Mark the block so no more stacks can use it. */
1315 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1316 	/*
1317 	 * If TCBs are still attached to the stack, attempt to switch them
1318 	 * to the default stack.
1319 	 */
1320 	if (force && blk->tfb_refcnt) {
1321 		struct inpcb *inp;
1322 		struct tcpcb *tp;
1323 		VNET_ITERATOR_DECL(vnet_iter);
1324 
1325 		rw_wunlock(&tcp_function_lock);
1326 
1327 		VNET_LIST_RLOCK();
1328 		VNET_FOREACH(vnet_iter) {
1329 			CURVNET_SET(vnet_iter);
1330 			INP_INFO_WLOCK(&V_tcbinfo);
1331 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1332 				INP_WLOCK(inp);
1333 				if (inp->inp_flags & INP_TIMEWAIT) {
1334 					INP_WUNLOCK(inp);
1335 					continue;
1336 				}
1337 				tp = intotcpcb(inp);
1338 				if (tp == NULL || tp->t_fb != blk) {
1339 					INP_WUNLOCK(inp);
1340 					continue;
1341 				}
1342 				tcp_switch_back_to_default(tp);
1343 				INP_WUNLOCK(inp);
1344 			}
1345 			INP_INFO_WUNLOCK(&V_tcbinfo);
1346 			CURVNET_RESTORE();
1347 		}
1348 		VNET_LIST_RUNLOCK();
1349 
1350 		rw_wlock(&tcp_function_lock);
1351 	}
1352 	if (blk->tfb_refcnt) {
1353 		/* TCBs still attached. */
1354 		rw_wunlock(&tcp_function_lock);
1355 		return (EBUSY);
1356 	}
1357 	if (quiesce) {
1358 		/* Skip removal. */
1359 		rw_wunlock(&tcp_function_lock);
1360 		return (0);
1361 	}
1362 	/* Remove any function names that map to this function block. */
1363 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1364 		TAILQ_REMOVE(&t_functions, f, tf_next);
1365 		tcp_fb_cnt--;
1366 		f->tf_fb = NULL;
1367 		free(f, M_TCPFUNCTIONS);
1368 	}
1369 	rw_wunlock(&tcp_function_lock);
1370 	return (0);
1371 }
1372 
1373 void
1374 tcp_init(void)
1375 {
1376 	const char *tcbhash_tuneable;
1377 	int hashsize;
1378 
1379 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1380 
1381 #ifdef TCP_HHOOK
1382 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1383 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1384 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1385 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1386 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1387 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1388 #endif
1389 #ifdef STATS
1390 	if (tcp_stats_init())
1391 		printf("%s: WARNING: unable to initialise TCP stats\n",
1392 		    __func__);
1393 #endif
1394 	hashsize = TCBHASHSIZE;
1395 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1396 	if (hashsize == 0) {
1397 		/*
1398 		 * Auto tune the hash size based on maxsockets.
1399 		 * A perfect hash would have a 1:1 mapping
1400 		 * (hashsize = maxsockets) however it's been
1401 		 * suggested that O(2) average is better.
1402 		 */
1403 		hashsize = maketcp_hashsize(maxsockets / 4);
1404 		/*
1405 		 * Our historical default is 512,
1406 		 * do not autotune lower than this.
1407 		 */
1408 		if (hashsize < 512)
1409 			hashsize = 512;
1410 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1411 			printf("%s: %s auto tuned to %d\n", __func__,
1412 			    tcbhash_tuneable, hashsize);
1413 	}
1414 	/*
1415 	 * We require a hashsize to be a power of two.
1416 	 * Previously if it was not a power of two we would just reset it
1417 	 * back to 512, which could be a nasty surprise if you did not notice
1418 	 * the error message.
1419 	 * Instead what we do is clip it to the closest power of two lower
1420 	 * than the specified hash value.
1421 	 */
1422 	if (!powerof2(hashsize)) {
1423 		int oldhashsize = hashsize;
1424 
1425 		hashsize = maketcp_hashsize(hashsize);
1426 		/* prevent absurdly low value */
1427 		if (hashsize < 16)
1428 			hashsize = 16;
1429 		printf("%s: WARNING: TCB hash size not a power of 2, "
1430 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1431 		    hashsize);
1432 	}
1433 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1434 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1435 
1436 	/*
1437 	 * These have to be type stable for the benefit of the timers.
1438 	 */
1439 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1440 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1441 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1442 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1443 
1444 	tcp_tw_init();
1445 	syncache_init();
1446 	tcp_hc_init();
1447 
1448 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1449 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1450 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1451 
1452 	tcp_fastopen_init();
1453 
1454 	/* Skip initialization of globals for non-default instances. */
1455 	if (!IS_DEFAULT_VNET(curvnet))
1456 		return;
1457 
1458 	tcp_reass_global_init();
1459 
1460 	/* XXX virtualize those bellow? */
1461 	tcp_delacktime = TCPTV_DELACK;
1462 	tcp_keepinit = TCPTV_KEEP_INIT;
1463 	tcp_keepidle = TCPTV_KEEP_IDLE;
1464 	tcp_keepintvl = TCPTV_KEEPINTVL;
1465 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1466 	tcp_msl = TCPTV_MSL;
1467 	tcp_rexmit_initial = TCPTV_RTOBASE;
1468 	if (tcp_rexmit_initial < 1)
1469 		tcp_rexmit_initial = 1;
1470 	tcp_rexmit_min = TCPTV_MIN;
1471 	if (tcp_rexmit_min < 1)
1472 		tcp_rexmit_min = 1;
1473 	tcp_persmin = TCPTV_PERSMIN;
1474 	tcp_persmax = TCPTV_PERSMAX;
1475 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1476 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1477 	tcp_tcbhashsize = hashsize;
1478 
1479 	/* Setup the tcp function block list */
1480 	TAILQ_INIT(&t_functions);
1481 	rw_init(&tcp_function_lock, "tcp_func_lock");
1482 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1483 #ifdef TCP_BLACKBOX
1484 	/* Initialize the TCP logging data. */
1485 	tcp_log_init();
1486 #endif
1487 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1488 
1489 	if (tcp_soreceive_stream) {
1490 #ifdef INET
1491 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1492 #endif
1493 #ifdef INET6
1494 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1495 #endif /* INET6 */
1496 	}
1497 
1498 #ifdef INET6
1499 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1500 #else /* INET6 */
1501 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1502 #endif /* INET6 */
1503 	if (max_protohdr < TCP_MINPROTOHDR)
1504 		max_protohdr = TCP_MINPROTOHDR;
1505 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1506 		panic("tcp_init");
1507 #undef TCP_MINPROTOHDR
1508 
1509 	ISN_LOCK_INIT();
1510 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1511 		SHUTDOWN_PRI_DEFAULT);
1512 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1513 		EVENTHANDLER_PRI_ANY);
1514 
1515 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1516 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1517 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1518 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1519 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1520 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1521 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1522 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1523 #ifdef TCPPCAP
1524 	tcp_pcap_init();
1525 #endif
1526 }
1527 
1528 #ifdef VIMAGE
1529 static void
1530 tcp_destroy(void *unused __unused)
1531 {
1532 	int n;
1533 #ifdef TCP_HHOOK
1534 	int error;
1535 #endif
1536 
1537 	/*
1538 	 * All our processes are gone, all our sockets should be cleaned
1539 	 * up, which means, we should be past the tcp_discardcb() calls.
1540 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1541 	 */
1542 	for (;;) {
1543 		INP_LIST_RLOCK(&V_tcbinfo);
1544 		n = V_tcbinfo.ipi_count;
1545 		INP_LIST_RUNLOCK(&V_tcbinfo);
1546 		if (n == 0)
1547 			break;
1548 		pause("tcpdes", hz / 10);
1549 	}
1550 	tcp_hc_destroy();
1551 	syncache_destroy();
1552 	tcp_tw_destroy();
1553 	in_pcbinfo_destroy(&V_tcbinfo);
1554 	/* tcp_discardcb() clears the sack_holes up. */
1555 	uma_zdestroy(V_sack_hole_zone);
1556 	uma_zdestroy(V_tcpcb_zone);
1557 
1558 	/*
1559 	 * Cannot free the zone until all tcpcbs are released as we attach
1560 	 * the allocations to them.
1561 	 */
1562 	tcp_fastopen_destroy();
1563 
1564 #ifdef TCP_HHOOK
1565 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1566 	if (error != 0) {
1567 		printf("%s: WARNING: unable to deregister helper hook "
1568 		    "type=%d, id=%d: error %d returned\n", __func__,
1569 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1570 	}
1571 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1572 	if (error != 0) {
1573 		printf("%s: WARNING: unable to deregister helper hook "
1574 		    "type=%d, id=%d: error %d returned\n", __func__,
1575 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1576 	}
1577 #endif
1578 }
1579 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1580 #endif
1581 
1582 void
1583 tcp_fini(void *xtp)
1584 {
1585 
1586 }
1587 
1588 /*
1589  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1590  * tcp_template used to store this data in mbufs, but we now recopy it out
1591  * of the tcpcb each time to conserve mbufs.
1592  */
1593 void
1594 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1595 {
1596 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1597 
1598 	INP_WLOCK_ASSERT(inp);
1599 
1600 #ifdef INET6
1601 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1602 		struct ip6_hdr *ip6;
1603 
1604 		ip6 = (struct ip6_hdr *)ip_ptr;
1605 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1606 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1607 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1608 			(IPV6_VERSION & IPV6_VERSION_MASK);
1609 		if (port == 0)
1610 			ip6->ip6_nxt = IPPROTO_TCP;
1611 		else
1612 			ip6->ip6_nxt = IPPROTO_UDP;
1613 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1614 		ip6->ip6_src = inp->in6p_laddr;
1615 		ip6->ip6_dst = inp->in6p_faddr;
1616 	}
1617 #endif /* INET6 */
1618 #if defined(INET6) && defined(INET)
1619 	else
1620 #endif
1621 #ifdef INET
1622 	{
1623 		struct ip *ip;
1624 
1625 		ip = (struct ip *)ip_ptr;
1626 		ip->ip_v = IPVERSION;
1627 		ip->ip_hl = 5;
1628 		ip->ip_tos = inp->inp_ip_tos;
1629 		ip->ip_len = 0;
1630 		ip->ip_id = 0;
1631 		ip->ip_off = 0;
1632 		ip->ip_ttl = inp->inp_ip_ttl;
1633 		ip->ip_sum = 0;
1634 		if (port == 0)
1635 			ip->ip_p = IPPROTO_TCP;
1636 		else
1637 			ip->ip_p = IPPROTO_UDP;
1638 		ip->ip_src = inp->inp_laddr;
1639 		ip->ip_dst = inp->inp_faddr;
1640 	}
1641 #endif /* INET */
1642 	th->th_sport = inp->inp_lport;
1643 	th->th_dport = inp->inp_fport;
1644 	th->th_seq = 0;
1645 	th->th_ack = 0;
1646 	th->th_x2 = 0;
1647 	th->th_off = 5;
1648 	th->th_flags = 0;
1649 	th->th_win = 0;
1650 	th->th_urp = 0;
1651 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1652 }
1653 
1654 /*
1655  * Create template to be used to send tcp packets on a connection.
1656  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1657  * use for this function is in keepalives, which use tcp_respond.
1658  */
1659 struct tcptemp *
1660 tcpip_maketemplate(struct inpcb *inp)
1661 {
1662 	struct tcptemp *t;
1663 
1664 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1665 	if (t == NULL)
1666 		return (NULL);
1667 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1668 	return (t);
1669 }
1670 
1671 /*
1672  * Send a single message to the TCP at address specified by
1673  * the given TCP/IP header.  If m == NULL, then we make a copy
1674  * of the tcpiphdr at th and send directly to the addressed host.
1675  * This is used to force keep alive messages out using the TCP
1676  * template for a connection.  If flags are given then we send
1677  * a message back to the TCP which originated the segment th,
1678  * and discard the mbuf containing it and any other attached mbufs.
1679  *
1680  * In any case the ack and sequence number of the transmitted
1681  * segment are as specified by the parameters.
1682  *
1683  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1684  */
1685 void
1686 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1687     tcp_seq ack, tcp_seq seq, int flags)
1688 {
1689 	struct tcpopt to;
1690 	struct inpcb *inp;
1691 	struct ip *ip;
1692 	struct mbuf *optm;
1693 	struct udphdr *uh = NULL;
1694 	struct tcphdr *nth;
1695 	u_char *optp;
1696 #ifdef INET6
1697 	struct ip6_hdr *ip6;
1698 	int isipv6;
1699 #endif /* INET6 */
1700 	int optlen, tlen, win, ulen;
1701 	bool incl_opts;
1702 	uint16_t port;
1703 
1704 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1705 	NET_EPOCH_ASSERT();
1706 
1707 #ifdef INET6
1708 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1709 	ip6 = ipgen;
1710 #endif /* INET6 */
1711 	ip = ipgen;
1712 
1713 	if (tp != NULL) {
1714 		inp = tp->t_inpcb;
1715 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1716 		INP_LOCK_ASSERT(inp);
1717 	} else
1718 		inp = NULL;
1719 
1720 	if (m != NULL) {
1721 #ifdef INET6
1722 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1723 			port = m->m_pkthdr.tcp_tun_port;
1724 		else
1725 #endif
1726 		if (ip && (ip->ip_p == IPPROTO_UDP))
1727 			port = m->m_pkthdr.tcp_tun_port;
1728 		else
1729 			port = 0;
1730 	} else
1731 		port = tp->t_port;
1732 
1733 	incl_opts = false;
1734 	win = 0;
1735 	if (tp != NULL) {
1736 		if (!(flags & TH_RST)) {
1737 			win = sbspace(&inp->inp_socket->so_rcv);
1738 			if (win > TCP_MAXWIN << tp->rcv_scale)
1739 				win = TCP_MAXWIN << tp->rcv_scale;
1740 		}
1741 		if ((tp->t_flags & TF_NOOPT) == 0)
1742 			incl_opts = true;
1743 	}
1744 	if (m == NULL) {
1745 		m = m_gethdr(M_NOWAIT, MT_DATA);
1746 		if (m == NULL)
1747 			return;
1748 		m->m_data += max_linkhdr;
1749 #ifdef INET6
1750 		if (isipv6) {
1751 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1752 			      sizeof(struct ip6_hdr));
1753 			ip6 = mtod(m, struct ip6_hdr *);
1754 			nth = (struct tcphdr *)(ip6 + 1);
1755 			if (port) {
1756 				/* Insert a UDP header */
1757 				uh = (struct udphdr *)nth;
1758 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1759 				uh->uh_dport = port;
1760 				nth = (struct tcphdr *)(uh + 1);
1761 			}
1762 		} else
1763 #endif /* INET6 */
1764 		{
1765 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1766 			ip = mtod(m, struct ip *);
1767 			nth = (struct tcphdr *)(ip + 1);
1768 			if (port) {
1769 				/* Insert a UDP header */
1770 				uh = (struct udphdr *)nth;
1771 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1772 				uh->uh_dport = port;
1773 				nth = (struct tcphdr *)(uh + 1);
1774 			}
1775 		}
1776 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1777 		flags = TH_ACK;
1778 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1779 		struct mbuf *n;
1780 
1781 		/* Can't reuse 'm', allocate a new mbuf. */
1782 		n = m_gethdr(M_NOWAIT, MT_DATA);
1783 		if (n == NULL) {
1784 			m_freem(m);
1785 			return;
1786 		}
1787 
1788 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1789 			m_freem(m);
1790 			m_freem(n);
1791 			return;
1792 		}
1793 
1794 		n->m_data += max_linkhdr;
1795 		/* m_len is set later */
1796 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1797 #ifdef INET6
1798 		if (isipv6) {
1799 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1800 			      sizeof(struct ip6_hdr));
1801 			ip6 = mtod(n, struct ip6_hdr *);
1802 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1803 			nth = (struct tcphdr *)(ip6 + 1);
1804 			if (port) {
1805 				/* Insert a UDP header */
1806 				uh = (struct udphdr *)nth;
1807 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1808 				uh->uh_dport = port;
1809 				nth = (struct tcphdr *)(uh + 1);
1810 			}
1811 		} else
1812 #endif /* INET6 */
1813 		{
1814 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1815 			ip = mtod(n, struct ip *);
1816 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1817 			nth = (struct tcphdr *)(ip + 1);
1818 			if (port) {
1819 				/* Insert a UDP header */
1820 				uh = (struct udphdr *)nth;
1821 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1822 				uh->uh_dport = port;
1823 				nth = (struct tcphdr *)(uh + 1);
1824 			}
1825 		}
1826 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1827 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1828 		th = nth;
1829 		m_freem(m);
1830 		m = n;
1831 	} else {
1832 		/*
1833 		 *  reuse the mbuf.
1834 		 * XXX MRT We inherit the FIB, which is lucky.
1835 		 */
1836 		m_freem(m->m_next);
1837 		m->m_next = NULL;
1838 		m->m_data = (caddr_t)ipgen;
1839 		/* m_len is set later */
1840 #ifdef INET6
1841 		if (isipv6) {
1842 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1843 			nth = (struct tcphdr *)(ip6 + 1);
1844 		} else
1845 #endif /* INET6 */
1846 		{
1847 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1848 			nth = (struct tcphdr *)(ip + 1);
1849 		}
1850 		if (th != nth) {
1851 			/*
1852 			 * this is usually a case when an extension header
1853 			 * exists between the IPv6 header and the
1854 			 * TCP header.
1855 			 */
1856 			nth->th_sport = th->th_sport;
1857 			nth->th_dport = th->th_dport;
1858 		}
1859 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1860 #undef xchg
1861 	}
1862 	tlen = 0;
1863 #ifdef INET6
1864 	if (isipv6)
1865 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1866 #endif
1867 #if defined(INET) && defined(INET6)
1868 	else
1869 #endif
1870 #ifdef INET
1871 		tlen = sizeof (struct tcpiphdr);
1872 #endif
1873 	if (port)
1874 		tlen += sizeof (struct udphdr);
1875 #ifdef INVARIANTS
1876 	m->m_len = 0;
1877 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1878 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1879 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1880 #endif
1881 	m->m_len = tlen;
1882 	to.to_flags = 0;
1883 	if (incl_opts) {
1884 		/* Make sure we have room. */
1885 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1886 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1887 			if (m->m_next) {
1888 				optp = mtod(m->m_next, u_char *);
1889 				optm = m->m_next;
1890 			} else
1891 				incl_opts = false;
1892 		} else {
1893 			optp = (u_char *) (nth + 1);
1894 			optm = m;
1895 		}
1896 	}
1897 	if (incl_opts) {
1898 		/* Timestamps. */
1899 		if (tp->t_flags & TF_RCVD_TSTMP) {
1900 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1901 			to.to_tsecr = tp->ts_recent;
1902 			to.to_flags |= TOF_TS;
1903 		}
1904 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1905 		/* TCP-MD5 (RFC2385). */
1906 		if (tp->t_flags & TF_SIGNATURE)
1907 			to.to_flags |= TOF_SIGNATURE;
1908 #endif
1909 		/* Add the options. */
1910 		tlen += optlen = tcp_addoptions(&to, optp);
1911 
1912 		/* Update m_len in the correct mbuf. */
1913 		optm->m_len += optlen;
1914 	} else
1915 		optlen = 0;
1916 #ifdef INET6
1917 	if (isipv6) {
1918 		if (uh) {
1919 			ulen = tlen - sizeof(struct ip6_hdr);
1920 			uh->uh_ulen = htons(ulen);
1921 		}
1922 		ip6->ip6_flow = 0;
1923 		ip6->ip6_vfc = IPV6_VERSION;
1924 		if (port)
1925 			ip6->ip6_nxt = IPPROTO_UDP;
1926 		else
1927 			ip6->ip6_nxt = IPPROTO_TCP;
1928 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1929 	}
1930 #endif
1931 #if defined(INET) && defined(INET6)
1932 	else
1933 #endif
1934 #ifdef INET
1935 	{
1936 		if (uh) {
1937 			ulen = tlen - sizeof(struct ip);
1938 			uh->uh_ulen = htons(ulen);
1939 		}
1940 		ip->ip_len = htons(tlen);
1941 		ip->ip_ttl = V_ip_defttl;
1942 		if (port) {
1943 			ip->ip_p = IPPROTO_UDP;
1944 		} else {
1945 			ip->ip_p = IPPROTO_TCP;
1946 		}
1947 		if (V_path_mtu_discovery)
1948 			ip->ip_off |= htons(IP_DF);
1949 	}
1950 #endif
1951 	m->m_pkthdr.len = tlen;
1952 	m->m_pkthdr.rcvif = NULL;
1953 #ifdef MAC
1954 	if (inp != NULL) {
1955 		/*
1956 		 * Packet is associated with a socket, so allow the
1957 		 * label of the response to reflect the socket label.
1958 		 */
1959 		INP_LOCK_ASSERT(inp);
1960 		mac_inpcb_create_mbuf(inp, m);
1961 	} else {
1962 		/*
1963 		 * Packet is not associated with a socket, so possibly
1964 		 * update the label in place.
1965 		 */
1966 		mac_netinet_tcp_reply(m);
1967 	}
1968 #endif
1969 	nth->th_seq = htonl(seq);
1970 	nth->th_ack = htonl(ack);
1971 	nth->th_x2 = 0;
1972 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1973 	nth->th_flags = flags;
1974 	if (tp != NULL)
1975 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1976 	else
1977 		nth->th_win = htons((u_short)win);
1978 	nth->th_urp = 0;
1979 
1980 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1981 	if (to.to_flags & TOF_SIGNATURE) {
1982 		if (!TCPMD5_ENABLED() ||
1983 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1984 			m_freem(m);
1985 			return;
1986 		}
1987 	}
1988 #endif
1989 
1990 #ifdef INET6
1991 	if (isipv6) {
1992 		if (port) {
1993 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
1994 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1995 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
1996 			nth->th_sum = 0;
1997 		} else {
1998 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1999 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2000 			nth->th_sum = in6_cksum_pseudo(ip6,
2001 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2002 		}
2003 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2004 		    NULL, NULL);
2005 	}
2006 #endif /* INET6 */
2007 #if defined(INET6) && defined(INET)
2008 	else
2009 #endif
2010 #ifdef INET
2011 	{
2012 		if (port) {
2013 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2014 			    htons(ulen + IPPROTO_UDP));
2015 			m->m_pkthdr.csum_flags = CSUM_UDP;
2016 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2017 			nth->th_sum = 0;
2018 		} else {
2019 			m->m_pkthdr.csum_flags = CSUM_TCP;
2020 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2021 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2022 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2023 		}
2024 	}
2025 #endif /* INET */
2026 #ifdef TCPDEBUG
2027 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2028 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2029 #endif
2030 	TCP_PROBE3(debug__output, tp, th, m);
2031 	if (flags & TH_RST)
2032 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2033 
2034 #ifdef INET6
2035 	if (isipv6) {
2036 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2037 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2038 	}
2039 #endif /* INET6 */
2040 #if defined(INET) && defined(INET6)
2041 	else
2042 #endif
2043 #ifdef INET
2044 	{
2045 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2046 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
2047 	}
2048 #endif
2049 }
2050 
2051 /*
2052  * Create a new TCP control block, making an
2053  * empty reassembly queue and hooking it to the argument
2054  * protocol control block.  The `inp' parameter must have
2055  * come from the zone allocator set up in tcp_init().
2056  */
2057 struct tcpcb *
2058 tcp_newtcpcb(struct inpcb *inp)
2059 {
2060 	struct tcpcb_mem *tm;
2061 	struct tcpcb *tp;
2062 #ifdef INET6
2063 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2064 #endif /* INET6 */
2065 
2066 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2067 	if (tm == NULL)
2068 		return (NULL);
2069 	tp = &tm->tcb;
2070 
2071 	/* Initialise cc_var struct for this tcpcb. */
2072 	tp->ccv = &tm->ccv;
2073 	tp->ccv->type = IPPROTO_TCP;
2074 	tp->ccv->ccvc.tcp = tp;
2075 	rw_rlock(&tcp_function_lock);
2076 	tp->t_fb = tcp_func_set_ptr;
2077 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2078 	rw_runlock(&tcp_function_lock);
2079 	/*
2080 	 * Use the current system default CC algorithm.
2081 	 */
2082 	CC_LIST_RLOCK();
2083 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2084 	CC_ALGO(tp) = CC_DEFAULT();
2085 	CC_LIST_RUNLOCK();
2086 	/*
2087 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2088 	 * is called.
2089 	 */
2090 	in_pcbref(inp);	/* Reference for tcpcb */
2091 	tp->t_inpcb = inp;
2092 
2093 	if (CC_ALGO(tp)->cb_init != NULL)
2094 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
2095 			if (tp->t_fb->tfb_tcp_fb_fini)
2096 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2097 			in_pcbrele_wlocked(inp);
2098 			refcount_release(&tp->t_fb->tfb_refcnt);
2099 			uma_zfree(V_tcpcb_zone, tm);
2100 			return (NULL);
2101 		}
2102 
2103 #ifdef TCP_HHOOK
2104 	tp->osd = &tm->osd;
2105 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2106 		if (tp->t_fb->tfb_tcp_fb_fini)
2107 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2108 		in_pcbrele_wlocked(inp);
2109 		refcount_release(&tp->t_fb->tfb_refcnt);
2110 		uma_zfree(V_tcpcb_zone, tm);
2111 		return (NULL);
2112 	}
2113 #endif
2114 
2115 #ifdef VIMAGE
2116 	tp->t_vnet = inp->inp_vnet;
2117 #endif
2118 	tp->t_timers = &tm->tt;
2119 	TAILQ_INIT(&tp->t_segq);
2120 	tp->t_maxseg =
2121 #ifdef INET6
2122 		isipv6 ? V_tcp_v6mssdflt :
2123 #endif /* INET6 */
2124 		V_tcp_mssdflt;
2125 
2126 	/* Set up our timeouts. */
2127 	callout_init(&tp->t_timers->tt_rexmt, 1);
2128 	callout_init(&tp->t_timers->tt_persist, 1);
2129 	callout_init(&tp->t_timers->tt_keep, 1);
2130 	callout_init(&tp->t_timers->tt_2msl, 1);
2131 	callout_init(&tp->t_timers->tt_delack, 1);
2132 
2133 	if (V_tcp_do_rfc1323)
2134 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2135 	if (V_tcp_do_sack)
2136 		tp->t_flags |= TF_SACK_PERMIT;
2137 	TAILQ_INIT(&tp->snd_holes);
2138 
2139 	/*
2140 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2141 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2142 	 * reasonable initial retransmit time.
2143 	 */
2144 	tp->t_srtt = TCPTV_SRTTBASE;
2145 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2146 	tp->t_rttmin = tcp_rexmit_min;
2147 	tp->t_rxtcur = tcp_rexmit_initial;
2148 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2149 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2150 	tp->t_rcvtime = ticks;
2151 	/*
2152 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2153 	 * because the socket may be bound to an IPv6 wildcard address,
2154 	 * which may match an IPv4-mapped IPv6 address.
2155 	 */
2156 	inp->inp_ip_ttl = V_ip_defttl;
2157 	inp->inp_ppcb = tp;
2158 #ifdef TCPPCAP
2159 	/*
2160 	 * Init the TCP PCAP queues.
2161 	 */
2162 	tcp_pcap_tcpcb_init(tp);
2163 #endif
2164 #ifdef TCP_BLACKBOX
2165 	/* Initialize the per-TCPCB log data. */
2166 	tcp_log_tcpcbinit(tp);
2167 #endif
2168 	tp->t_pacing_rate = -1;
2169 	if (tp->t_fb->tfb_tcp_fb_init) {
2170 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2171 			refcount_release(&tp->t_fb->tfb_refcnt);
2172 			in_pcbrele_wlocked(inp);
2173 			uma_zfree(V_tcpcb_zone, tm);
2174 			return (NULL);
2175 		}
2176 	}
2177 #ifdef STATS
2178 	if (V_tcp_perconn_stats_enable == 1)
2179 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2180 #endif
2181 	if (V_tcp_do_lrd)
2182 		tp->t_flags |= TF_LRD;
2183 	return (tp);		/* XXX */
2184 }
2185 
2186 /*
2187  * Switch the congestion control algorithm back to NewReno for any active
2188  * control blocks using an algorithm which is about to go away.
2189  * This ensures the CC framework can allow the unload to proceed without leaving
2190  * any dangling pointers which would trigger a panic.
2191  * Returning non-zero would inform the CC framework that something went wrong
2192  * and it would be unsafe to allow the unload to proceed. However, there is no
2193  * way for this to occur with this implementation so we always return zero.
2194  */
2195 int
2196 tcp_ccalgounload(struct cc_algo *unload_algo)
2197 {
2198 	struct cc_algo *tmpalgo;
2199 	struct inpcb *inp;
2200 	struct tcpcb *tp;
2201 	VNET_ITERATOR_DECL(vnet_iter);
2202 
2203 	/*
2204 	 * Check all active control blocks across all network stacks and change
2205 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
2206 	 * requires cleanup code to be run, call it.
2207 	 */
2208 	VNET_LIST_RLOCK();
2209 	VNET_FOREACH(vnet_iter) {
2210 		CURVNET_SET(vnet_iter);
2211 		INP_INFO_WLOCK(&V_tcbinfo);
2212 		/*
2213 		 * New connections already part way through being initialised
2214 		 * with the CC algo we're removing will not race with this code
2215 		 * because the INP_INFO_WLOCK is held during initialisation. We
2216 		 * therefore don't enter the loop below until the connection
2217 		 * list has stabilised.
2218 		 */
2219 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
2220 			INP_WLOCK(inp);
2221 			/* Important to skip tcptw structs. */
2222 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
2223 			    (tp = intotcpcb(inp)) != NULL) {
2224 				/*
2225 				 * By holding INP_WLOCK here, we are assured
2226 				 * that the connection is not currently
2227 				 * executing inside the CC module's functions
2228 				 * i.e. it is safe to make the switch back to
2229 				 * NewReno.
2230 				 */
2231 				if (CC_ALGO(tp) == unload_algo) {
2232 					tmpalgo = CC_ALGO(tp);
2233 					if (tmpalgo->cb_destroy != NULL)
2234 						tmpalgo->cb_destroy(tp->ccv);
2235 					CC_DATA(tp) = NULL;
2236 					/*
2237 					 * NewReno may allocate memory on
2238 					 * demand for certain stateful
2239 					 * configuration as needed, but is
2240 					 * coded to never fail on memory
2241 					 * allocation failure so it is a safe
2242 					 * fallback.
2243 					 */
2244 					CC_ALGO(tp) = &newreno_cc_algo;
2245 				}
2246 			}
2247 			INP_WUNLOCK(inp);
2248 		}
2249 		INP_INFO_WUNLOCK(&V_tcbinfo);
2250 		CURVNET_RESTORE();
2251 	}
2252 	VNET_LIST_RUNLOCK();
2253 
2254 	return (0);
2255 }
2256 
2257 /*
2258  * Drop a TCP connection, reporting
2259  * the specified error.  If connection is synchronized,
2260  * then send a RST to peer.
2261  */
2262 struct tcpcb *
2263 tcp_drop(struct tcpcb *tp, int errno)
2264 {
2265 	struct socket *so = tp->t_inpcb->inp_socket;
2266 
2267 	NET_EPOCH_ASSERT();
2268 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2269 	INP_WLOCK_ASSERT(tp->t_inpcb);
2270 
2271 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2272 		tcp_state_change(tp, TCPS_CLOSED);
2273 		(void) tp->t_fb->tfb_tcp_output(tp);
2274 		TCPSTAT_INC(tcps_drops);
2275 	} else
2276 		TCPSTAT_INC(tcps_conndrops);
2277 	if (errno == ETIMEDOUT && tp->t_softerror)
2278 		errno = tp->t_softerror;
2279 	so->so_error = errno;
2280 	return (tcp_close(tp));
2281 }
2282 
2283 void
2284 tcp_discardcb(struct tcpcb *tp)
2285 {
2286 	struct inpcb *inp = tp->t_inpcb;
2287 	struct socket *so = inp->inp_socket;
2288 #ifdef INET6
2289 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2290 #endif /* INET6 */
2291 	int released __unused;
2292 
2293 	INP_WLOCK_ASSERT(inp);
2294 
2295 	/*
2296 	 * Make sure that all of our timers are stopped before we delete the
2297 	 * PCB.
2298 	 *
2299 	 * If stopping a timer fails, we schedule a discard function in same
2300 	 * callout, and the last discard function called will take care of
2301 	 * deleting the tcpcb.
2302 	 */
2303 	tp->t_timers->tt_draincnt = 0;
2304 	tcp_timer_stop(tp, TT_REXMT);
2305 	tcp_timer_stop(tp, TT_PERSIST);
2306 	tcp_timer_stop(tp, TT_KEEP);
2307 	tcp_timer_stop(tp, TT_2MSL);
2308 	tcp_timer_stop(tp, TT_DELACK);
2309 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2310 		/*
2311 		 * Call the stop-all function of the methods,
2312 		 * this function should call the tcp_timer_stop()
2313 		 * method with each of the function specific timeouts.
2314 		 * That stop will be called via the tfb_tcp_timer_stop()
2315 		 * which should use the async drain function of the
2316 		 * callout system (see tcp_var.h).
2317 		 */
2318 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2319 	}
2320 
2321 	/* free the reassembly queue, if any */
2322 	tcp_reass_flush(tp);
2323 
2324 #ifdef TCP_OFFLOAD
2325 	/* Disconnect offload device, if any. */
2326 	if (tp->t_flags & TF_TOE)
2327 		tcp_offload_detach(tp);
2328 #endif
2329 
2330 	tcp_free_sackholes(tp);
2331 
2332 #ifdef TCPPCAP
2333 	/* Free the TCP PCAP queues. */
2334 	tcp_pcap_drain(&(tp->t_inpkts));
2335 	tcp_pcap_drain(&(tp->t_outpkts));
2336 #endif
2337 
2338 	/* Allow the CC algorithm to clean up after itself. */
2339 	if (CC_ALGO(tp)->cb_destroy != NULL)
2340 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2341 	CC_DATA(tp) = NULL;
2342 
2343 #ifdef TCP_HHOOK
2344 	khelp_destroy_osd(tp->osd);
2345 #endif
2346 #ifdef STATS
2347 	stats_blob_destroy(tp->t_stats);
2348 #endif
2349 
2350 	CC_ALGO(tp) = NULL;
2351 	inp->inp_ppcb = NULL;
2352 	if (tp->t_timers->tt_draincnt == 0) {
2353 		/* We own the last reference on tcpcb, let's free it. */
2354 #ifdef TCP_BLACKBOX
2355 		tcp_log_tcpcbfini(tp);
2356 #endif
2357 		TCPSTATES_DEC(tp->t_state);
2358 		if (tp->t_fb->tfb_tcp_fb_fini)
2359 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2360 
2361 		/*
2362 		 * If we got enough samples through the srtt filter,
2363 		 * save the rtt and rttvar in the routing entry.
2364 		 * 'Enough' is arbitrarily defined as 4 rtt samples.
2365 		 * 4 samples is enough for the srtt filter to converge
2366 		 * to within enough % of the correct value; fewer samples
2367 		 * and we could save a bogus rtt. The danger is not high
2368 		 * as tcp quickly recovers from everything.
2369 		 * XXX: Works very well but needs some more statistics!
2370 		 *
2371 		 * XXXRRS: Updating must be after the stack fini() since
2372 		 * that may be converting some internal representation of
2373 		 * say srtt etc into the general one used by other stacks.
2374 		 * Lets also at least protect against the so being NULL
2375 		 * as RW stated below.
2376 		 */
2377 		if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2378 			struct hc_metrics_lite metrics;
2379 			uint32_t ssthresh;
2380 
2381 			bzero(&metrics, sizeof(metrics));
2382 			/*
2383 			 * Update the ssthresh always when the conditions below
2384 			 * are satisfied. This gives us better new start value
2385 			 * for the congestion avoidance for new connections.
2386 			 * ssthresh is only set if packet loss occurred on a session.
2387 			 *
2388 			 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2389 			 * being torn down.  Ideally this code would not use 'so'.
2390 			 */
2391 			ssthresh = tp->snd_ssthresh;
2392 			if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2393 				/*
2394 				 * convert the limit from user data bytes to
2395 				 * packets then to packet data bytes.
2396 				 */
2397 				ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2398 				if (ssthresh < 2)
2399 					ssthresh = 2;
2400 				ssthresh *= (tp->t_maxseg +
2401 #ifdef INET6
2402 					     (isipv6 ? sizeof (struct ip6_hdr) +
2403 					      sizeof (struct tcphdr) :
2404 #endif
2405 					      sizeof (struct tcpiphdr)
2406 #ifdef INET6
2407 						     )
2408 #endif
2409 					);
2410 			} else
2411 				ssthresh = 0;
2412 			metrics.rmx_ssthresh = ssthresh;
2413 
2414 			metrics.rmx_rtt = tp->t_srtt;
2415 			metrics.rmx_rttvar = tp->t_rttvar;
2416 			metrics.rmx_cwnd = tp->snd_cwnd;
2417 			metrics.rmx_sendpipe = 0;
2418 			metrics.rmx_recvpipe = 0;
2419 
2420 			tcp_hc_update(&inp->inp_inc, &metrics);
2421 		}
2422 		refcount_release(&tp->t_fb->tfb_refcnt);
2423 		tp->t_inpcb = NULL;
2424 		uma_zfree(V_tcpcb_zone, tp);
2425 		released = in_pcbrele_wlocked(inp);
2426 		KASSERT(!released, ("%s: inp %p should not have been released "
2427 			"here", __func__, inp));
2428 	}
2429 }
2430 
2431 void
2432 tcp_timer_discard(void *ptp)
2433 {
2434 	struct inpcb *inp;
2435 	struct tcpcb *tp;
2436 	struct epoch_tracker et;
2437 
2438 	tp = (struct tcpcb *)ptp;
2439 	CURVNET_SET(tp->t_vnet);
2440 	NET_EPOCH_ENTER(et);
2441 	inp = tp->t_inpcb;
2442 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2443 		__func__, tp));
2444 	INP_WLOCK(inp);
2445 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2446 		("%s: tcpcb has to be stopped here", __func__));
2447 	tp->t_timers->tt_draincnt--;
2448 	if (tp->t_timers->tt_draincnt == 0) {
2449 		/* We own the last reference on this tcpcb, let's free it. */
2450 #ifdef TCP_BLACKBOX
2451 		tcp_log_tcpcbfini(tp);
2452 #endif
2453 		TCPSTATES_DEC(tp->t_state);
2454 		if (tp->t_fb->tfb_tcp_fb_fini)
2455 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2456 		refcount_release(&tp->t_fb->tfb_refcnt);
2457 		tp->t_inpcb = NULL;
2458 		uma_zfree(V_tcpcb_zone, tp);
2459 		if (in_pcbrele_wlocked(inp)) {
2460 			NET_EPOCH_EXIT(et);
2461 			CURVNET_RESTORE();
2462 			return;
2463 		}
2464 	}
2465 	INP_WUNLOCK(inp);
2466 	NET_EPOCH_EXIT(et);
2467 	CURVNET_RESTORE();
2468 }
2469 
2470 /*
2471  * Attempt to close a TCP control block, marking it as dropped, and freeing
2472  * the socket if we hold the only reference.
2473  */
2474 struct tcpcb *
2475 tcp_close(struct tcpcb *tp)
2476 {
2477 	struct inpcb *inp = tp->t_inpcb;
2478 	struct socket *so;
2479 
2480 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2481 	INP_WLOCK_ASSERT(inp);
2482 
2483 #ifdef TCP_OFFLOAD
2484 	if (tp->t_state == TCPS_LISTEN)
2485 		tcp_offload_listen_stop(tp);
2486 #endif
2487 	/*
2488 	 * This releases the TFO pending counter resource for TFO listen
2489 	 * sockets as well as passively-created TFO sockets that transition
2490 	 * from SYN_RECEIVED to CLOSED.
2491 	 */
2492 	if (tp->t_tfo_pending) {
2493 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2494 		tp->t_tfo_pending = NULL;
2495 	}
2496 	in_pcbdrop(inp);
2497 	TCPSTAT_INC(tcps_closed);
2498 	if (tp->t_state != TCPS_CLOSED)
2499 		tcp_state_change(tp, TCPS_CLOSED);
2500 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2501 	so = inp->inp_socket;
2502 	soisdisconnected(so);
2503 	if (inp->inp_flags & INP_SOCKREF) {
2504 		KASSERT(so->so_state & SS_PROTOREF,
2505 		    ("tcp_close: !SS_PROTOREF"));
2506 		inp->inp_flags &= ~INP_SOCKREF;
2507 		INP_WUNLOCK(inp);
2508 		SOCK_LOCK(so);
2509 		so->so_state &= ~SS_PROTOREF;
2510 		sofree(so);
2511 		return (NULL);
2512 	}
2513 	return (tp);
2514 }
2515 
2516 void
2517 tcp_drain(void)
2518 {
2519 	VNET_ITERATOR_DECL(vnet_iter);
2520 
2521 	if (!do_tcpdrain)
2522 		return;
2523 
2524 	VNET_LIST_RLOCK_NOSLEEP();
2525 	VNET_FOREACH(vnet_iter) {
2526 		CURVNET_SET(vnet_iter);
2527 		struct inpcb *inpb;
2528 		struct tcpcb *tcpb;
2529 
2530 	/*
2531 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2532 	 * if there is one...
2533 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2534 	 *      reassembly queue should be flushed, but in a situation
2535 	 *	where we're really low on mbufs, this is potentially
2536 	 *	useful.
2537 	 */
2538 		INP_INFO_WLOCK(&V_tcbinfo);
2539 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2540 			INP_WLOCK(inpb);
2541 			if (inpb->inp_flags & INP_TIMEWAIT) {
2542 				INP_WUNLOCK(inpb);
2543 				continue;
2544 			}
2545 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2546 				tcp_reass_flush(tcpb);
2547 				tcp_clean_sackreport(tcpb);
2548 #ifdef TCP_BLACKBOX
2549 				tcp_log_drain(tcpb);
2550 #endif
2551 #ifdef TCPPCAP
2552 				if (tcp_pcap_aggressive_free) {
2553 					/* Free the TCP PCAP queues. */
2554 					tcp_pcap_drain(&(tcpb->t_inpkts));
2555 					tcp_pcap_drain(&(tcpb->t_outpkts));
2556 				}
2557 #endif
2558 			}
2559 			INP_WUNLOCK(inpb);
2560 		}
2561 		INP_INFO_WUNLOCK(&V_tcbinfo);
2562 		CURVNET_RESTORE();
2563 	}
2564 	VNET_LIST_RUNLOCK_NOSLEEP();
2565 }
2566 
2567 /*
2568  * Notify a tcp user of an asynchronous error;
2569  * store error as soft error, but wake up user
2570  * (for now, won't do anything until can select for soft error).
2571  *
2572  * Do not wake up user since there currently is no mechanism for
2573  * reporting soft errors (yet - a kqueue filter may be added).
2574  */
2575 static struct inpcb *
2576 tcp_notify(struct inpcb *inp, int error)
2577 {
2578 	struct tcpcb *tp;
2579 
2580 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2581 	INP_WLOCK_ASSERT(inp);
2582 
2583 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2584 	    (inp->inp_flags & INP_DROPPED))
2585 		return (inp);
2586 
2587 	tp = intotcpcb(inp);
2588 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2589 
2590 	/*
2591 	 * Ignore some errors if we are hooked up.
2592 	 * If connection hasn't completed, has retransmitted several times,
2593 	 * and receives a second error, give up now.  This is better
2594 	 * than waiting a long time to establish a connection that
2595 	 * can never complete.
2596 	 */
2597 	if (tp->t_state == TCPS_ESTABLISHED &&
2598 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2599 	     error == EHOSTDOWN)) {
2600 		if (inp->inp_route.ro_nh) {
2601 			NH_FREE(inp->inp_route.ro_nh);
2602 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2603 		}
2604 		return (inp);
2605 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2606 	    tp->t_softerror) {
2607 		tp = tcp_drop(tp, error);
2608 		if (tp != NULL)
2609 			return (inp);
2610 		else
2611 			return (NULL);
2612 	} else {
2613 		tp->t_softerror = error;
2614 		return (inp);
2615 	}
2616 #if 0
2617 	wakeup( &so->so_timeo);
2618 	sorwakeup(so);
2619 	sowwakeup(so);
2620 #endif
2621 }
2622 
2623 static int
2624 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2625 {
2626 	struct epoch_tracker et;
2627 	struct inpcb *inp;
2628 	struct xinpgen xig;
2629 	int error;
2630 
2631 	if (req->newptr != NULL)
2632 		return (EPERM);
2633 
2634 	if (req->oldptr == NULL) {
2635 		int n;
2636 
2637 		n = V_tcbinfo.ipi_count +
2638 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2639 		n += imax(n / 8, 10);
2640 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2641 		return (0);
2642 	}
2643 
2644 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2645 		return (error);
2646 
2647 	bzero(&xig, sizeof(xig));
2648 	xig.xig_len = sizeof xig;
2649 	xig.xig_count = V_tcbinfo.ipi_count +
2650 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2651 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2652 	xig.xig_sogen = so_gencnt;
2653 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2654 	if (error)
2655 		return (error);
2656 
2657 	error = syncache_pcblist(req);
2658 	if (error)
2659 		return (error);
2660 
2661 	NET_EPOCH_ENTER(et);
2662 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2663 	    inp != NULL;
2664 	    inp = CK_LIST_NEXT(inp, inp_list)) {
2665 		INP_RLOCK(inp);
2666 		if (inp->inp_gencnt <= xig.xig_gen) {
2667 			int crerr;
2668 
2669 			/*
2670 			 * XXX: This use of cr_cansee(), introduced with
2671 			 * TCP state changes, is not quite right, but for
2672 			 * now, better than nothing.
2673 			 */
2674 			if (inp->inp_flags & INP_TIMEWAIT) {
2675 				if (intotw(inp) != NULL)
2676 					crerr = cr_cansee(req->td->td_ucred,
2677 					    intotw(inp)->tw_cred);
2678 				else
2679 					crerr = EINVAL;	/* Skip this inp. */
2680 			} else
2681 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2682 			if (crerr == 0) {
2683 				struct xtcpcb xt;
2684 
2685 				tcp_inptoxtp(inp, &xt);
2686 				INP_RUNLOCK(inp);
2687 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2688 				if (error)
2689 					break;
2690 				else
2691 					continue;
2692 			}
2693 		}
2694 		INP_RUNLOCK(inp);
2695 	}
2696 	NET_EPOCH_EXIT(et);
2697 
2698 	if (!error) {
2699 		/*
2700 		 * Give the user an updated idea of our state.
2701 		 * If the generation differs from what we told
2702 		 * her before, she knows that something happened
2703 		 * while we were processing this request, and it
2704 		 * might be necessary to retry.
2705 		 */
2706 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2707 		xig.xig_sogen = so_gencnt;
2708 		xig.xig_count = V_tcbinfo.ipi_count +
2709 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2710 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2711 	}
2712 
2713 	return (error);
2714 }
2715 
2716 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2717     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2718     NULL, 0, tcp_pcblist, "S,xtcpcb",
2719     "List of active TCP connections");
2720 
2721 #ifdef INET
2722 static int
2723 tcp_getcred(SYSCTL_HANDLER_ARGS)
2724 {
2725 	struct xucred xuc;
2726 	struct sockaddr_in addrs[2];
2727 	struct epoch_tracker et;
2728 	struct inpcb *inp;
2729 	int error;
2730 
2731 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2732 	if (error)
2733 		return (error);
2734 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2735 	if (error)
2736 		return (error);
2737 	NET_EPOCH_ENTER(et);
2738 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2739 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2740 	NET_EPOCH_EXIT(et);
2741 	if (inp != NULL) {
2742 		if (inp->inp_socket == NULL)
2743 			error = ENOENT;
2744 		if (error == 0)
2745 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2746 		if (error == 0)
2747 			cru2x(inp->inp_cred, &xuc);
2748 		INP_RUNLOCK(inp);
2749 	} else
2750 		error = ENOENT;
2751 	if (error == 0)
2752 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2753 	return (error);
2754 }
2755 
2756 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2757     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2758     0, 0, tcp_getcred, "S,xucred",
2759     "Get the xucred of a TCP connection");
2760 #endif /* INET */
2761 
2762 #ifdef INET6
2763 static int
2764 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2765 {
2766 	struct epoch_tracker et;
2767 	struct xucred xuc;
2768 	struct sockaddr_in6 addrs[2];
2769 	struct inpcb *inp;
2770 	int error;
2771 #ifdef INET
2772 	int mapped = 0;
2773 #endif
2774 
2775 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2776 	if (error)
2777 		return (error);
2778 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2779 	if (error)
2780 		return (error);
2781 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2782 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2783 		return (error);
2784 	}
2785 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2786 #ifdef INET
2787 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2788 			mapped = 1;
2789 		else
2790 #endif
2791 			return (EINVAL);
2792 	}
2793 
2794 	NET_EPOCH_ENTER(et);
2795 #ifdef INET
2796 	if (mapped == 1)
2797 		inp = in_pcblookup(&V_tcbinfo,
2798 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2799 			addrs[1].sin6_port,
2800 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2801 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2802 	else
2803 #endif
2804 		inp = in6_pcblookup(&V_tcbinfo,
2805 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2806 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2807 			INPLOOKUP_RLOCKPCB, NULL);
2808 	NET_EPOCH_EXIT(et);
2809 	if (inp != NULL) {
2810 		if (inp->inp_socket == NULL)
2811 			error = ENOENT;
2812 		if (error == 0)
2813 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2814 		if (error == 0)
2815 			cru2x(inp->inp_cred, &xuc);
2816 		INP_RUNLOCK(inp);
2817 	} else
2818 		error = ENOENT;
2819 	if (error == 0)
2820 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2821 	return (error);
2822 }
2823 
2824 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2825     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2826     0, 0, tcp6_getcred, "S,xucred",
2827     "Get the xucred of a TCP6 connection");
2828 #endif /* INET6 */
2829 
2830 #ifdef INET
2831 /* Path MTU to try next when a fragmentation-needed message is received. */
2832 static inline int
2833 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2834 {
2835 	int mtu = ntohs(icp->icmp_nextmtu);
2836 
2837 	/* If no alternative MTU was proposed, try the next smaller one. */
2838 	if (!mtu)
2839 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2840 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2841 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2842 
2843 	return (mtu);
2844 }
2845 
2846 static void
2847 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2848 {
2849 	struct ip *ip = vip;
2850 	struct tcphdr *th;
2851 	struct in_addr faddr;
2852 	struct inpcb *inp;
2853 	struct tcpcb *tp;
2854 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2855 	struct icmp *icp;
2856 	struct in_conninfo inc;
2857 	tcp_seq icmp_tcp_seq;
2858 	int mtu;
2859 
2860 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2861 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2862 		return;
2863 
2864 	if (cmd == PRC_MSGSIZE)
2865 		notify = tcp_mtudisc_notify;
2866 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2867 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2868 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2869 		notify = tcp_drop_syn_sent;
2870 
2871 	/*
2872 	 * Hostdead is ugly because it goes linearly through all PCBs.
2873 	 * XXX: We never get this from ICMP, otherwise it makes an
2874 	 * excellent DoS attack on machines with many connections.
2875 	 */
2876 	else if (cmd == PRC_HOSTDEAD)
2877 		ip = NULL;
2878 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2879 		return;
2880 
2881 	if (ip == NULL) {
2882 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2883 		return;
2884 	}
2885 
2886 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2887 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2888 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2889 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2890 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2891 		/* signal EHOSTDOWN, as it flushes the cached route */
2892 		inp = (*notify)(inp, EHOSTDOWN);
2893 		goto out;
2894 	}
2895 	icmp_tcp_seq = th->th_seq;
2896 	if (inp != NULL)  {
2897 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2898 		    !(inp->inp_flags & INP_DROPPED) &&
2899 		    !(inp->inp_socket == NULL)) {
2900 			tp = intotcpcb(inp);
2901 #ifdef TCP_OFFLOAD
2902 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2903 				/*
2904 				 * MTU discovery for offloaded connections.  Let
2905 				 * the TOE driver verify seq# and process it.
2906 				 */
2907 				mtu = tcp_next_pmtu(icp, ip);
2908 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2909 				goto out;
2910 			}
2911 #endif
2912 			if (tp->t_port != port) {
2913 				goto out;
2914 			}
2915 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2916 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2917 				if (cmd == PRC_MSGSIZE) {
2918 					/*
2919 					 * MTU discovery: we got a needfrag and
2920 					 * will potentially try a lower MTU.
2921 					 */
2922 					mtu = tcp_next_pmtu(icp, ip);
2923 
2924 					/*
2925 					 * Only process the offered MTU if it
2926 					 * is smaller than the current one.
2927 					 */
2928 					if (mtu < tp->t_maxseg +
2929 					    sizeof(struct tcpiphdr)) {
2930 						bzero(&inc, sizeof(inc));
2931 						inc.inc_faddr = faddr;
2932 						inc.inc_fibnum =
2933 						    inp->inp_inc.inc_fibnum;
2934 						tcp_hc_updatemtu(&inc, mtu);
2935 						tcp_mtudisc(inp, mtu);
2936 					}
2937 				} else
2938 					inp = (*notify)(inp,
2939 					    inetctlerrmap[cmd]);
2940 			}
2941 		}
2942 	} else {
2943 		bzero(&inc, sizeof(inc));
2944 		inc.inc_fport = th->th_dport;
2945 		inc.inc_lport = th->th_sport;
2946 		inc.inc_faddr = faddr;
2947 		inc.inc_laddr = ip->ip_src;
2948 		syncache_unreach(&inc, icmp_tcp_seq, port);
2949 	}
2950 out:
2951 	if (inp != NULL)
2952 		INP_WUNLOCK(inp);
2953 }
2954 
2955 void
2956 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2957 {
2958 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2959 }
2960 
2961 void
2962 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2963 {
2964 	/* Its a tunneled TCP over UDP icmp */
2965 	struct ip *outer_ip, *inner_ip;
2966 	struct icmp *icmp;
2967 	struct udphdr *udp;
2968 	struct tcphdr *th, ttemp;
2969 	int i_hlen, o_len;
2970 	uint16_t port;
2971 
2972 	inner_ip = (struct ip *)vip;
2973 	icmp = (struct icmp *)((caddr_t)inner_ip -
2974 	    (sizeof(struct icmp) - sizeof(struct ip)));
2975 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2976 	i_hlen = inner_ip->ip_hl << 2;
2977 	o_len = ntohs(outer_ip->ip_len);
2978 	if (o_len <
2979 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2980 		/* Not enough data present */
2981 		return;
2982 	}
2983 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2984 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2985 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2986 		return;
2987 	}
2988 	port = udp->uh_dport;
2989 	th = (struct tcphdr *)(udp + 1);
2990 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2991 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2992 	/* Now adjust down the size of the outer IP header */
2993 	o_len -= sizeof(struct udphdr);
2994 	outer_ip->ip_len = htons(o_len);
2995 	/* Now call in to the normal handling code */
2996 	tcp_ctlinput_with_port(cmd, sa, vip, port);
2997 }
2998 #endif /* INET */
2999 
3000 #ifdef INET6
3001 static inline int
3002 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3003 {
3004 	int mtu = ntohl(icmp6->icmp6_mtu);
3005 
3006 	/*
3007 	 * If no alternative MTU was proposed, or the proposed MTU was too
3008 	 * small, set to the min.
3009 	 */
3010 	if (mtu < IPV6_MMTU)
3011 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
3012 	return (mtu);
3013 }
3014 
3015 static void
3016 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3017 {
3018 	struct in6_addr *dst;
3019 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3020 	struct ip6_hdr *ip6;
3021 	struct mbuf *m;
3022 	struct inpcb *inp;
3023 	struct tcpcb *tp;
3024 	struct icmp6_hdr *icmp6;
3025 	struct ip6ctlparam *ip6cp = NULL;
3026 	const struct sockaddr_in6 *sa6_src = NULL;
3027 	struct in_conninfo inc;
3028 	struct tcp_ports {
3029 		uint16_t th_sport;
3030 		uint16_t th_dport;
3031 	} t_ports;
3032 	tcp_seq icmp_tcp_seq;
3033 	unsigned int mtu;
3034 	unsigned int off;
3035 
3036 	if (sa->sa_family != AF_INET6 ||
3037 	    sa->sa_len != sizeof(struct sockaddr_in6))
3038 		return;
3039 
3040 	/* if the parameter is from icmp6, decode it. */
3041 	if (d != NULL) {
3042 		ip6cp = (struct ip6ctlparam *)d;
3043 		icmp6 = ip6cp->ip6c_icmp6;
3044 		m = ip6cp->ip6c_m;
3045 		ip6 = ip6cp->ip6c_ip6;
3046 		off = ip6cp->ip6c_off;
3047 		sa6_src = ip6cp->ip6c_src;
3048 		dst = ip6cp->ip6c_finaldst;
3049 	} else {
3050 		m = NULL;
3051 		ip6 = NULL;
3052 		off = 0;	/* fool gcc */
3053 		sa6_src = &sa6_any;
3054 		dst = NULL;
3055 	}
3056 
3057 	if (cmd == PRC_MSGSIZE)
3058 		notify = tcp_mtudisc_notify;
3059 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3060 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3061 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3062 		notify = tcp_drop_syn_sent;
3063 
3064 	/*
3065 	 * Hostdead is ugly because it goes linearly through all PCBs.
3066 	 * XXX: We never get this from ICMP, otherwise it makes an
3067 	 * excellent DoS attack on machines with many connections.
3068 	 */
3069 	else if (cmd == PRC_HOSTDEAD)
3070 		ip6 = NULL;
3071 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3072 		return;
3073 
3074 	if (ip6 == NULL) {
3075 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3076 			      (const struct sockaddr *)sa6_src,
3077 			      0, cmd, NULL, notify);
3078 		return;
3079 	}
3080 
3081 	/* Check if we can safely get the ports from the tcp hdr */
3082 	if (m == NULL ||
3083 	    (m->m_pkthdr.len <
3084 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3085 		return;
3086 	}
3087 	bzero(&t_ports, sizeof(struct tcp_ports));
3088 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3089 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3090 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3091 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3092 		/* signal EHOSTDOWN, as it flushes the cached route */
3093 		inp = (*notify)(inp, EHOSTDOWN);
3094 		goto out;
3095 	}
3096 	off += sizeof(struct tcp_ports);
3097 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3098 		goto out;
3099 	}
3100 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3101 	if (inp != NULL)  {
3102 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3103 		    !(inp->inp_flags & INP_DROPPED) &&
3104 		    !(inp->inp_socket == NULL)) {
3105 			tp = intotcpcb(inp);
3106 #ifdef TCP_OFFLOAD
3107 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3108 				/* MTU discovery for offloaded connections. */
3109 				mtu = tcp6_next_pmtu(icmp6);
3110 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3111 				goto out;
3112 			}
3113 #endif
3114 			if (tp->t_port != port) {
3115 				goto out;
3116 			}
3117 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3118 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3119 				if (cmd == PRC_MSGSIZE) {
3120 					/*
3121 					 * MTU discovery:
3122 					 * If we got a needfrag set the MTU
3123 					 * in the route to the suggested new
3124 					 * value (if given) and then notify.
3125 					 */
3126 					mtu = tcp6_next_pmtu(icmp6);
3127 
3128 					bzero(&inc, sizeof(inc));
3129 					inc.inc_fibnum = M_GETFIB(m);
3130 					inc.inc_flags |= INC_ISIPV6;
3131 					inc.inc6_faddr = *dst;
3132 					if (in6_setscope(&inc.inc6_faddr,
3133 						m->m_pkthdr.rcvif, NULL))
3134 						goto out;
3135 					/*
3136 					 * Only process the offered MTU if it
3137 					 * is smaller than the current one.
3138 					 */
3139 					if (mtu < tp->t_maxseg +
3140 					    sizeof (struct tcphdr) +
3141 					    sizeof (struct ip6_hdr)) {
3142 						tcp_hc_updatemtu(&inc, mtu);
3143 						tcp_mtudisc(inp, mtu);
3144 						ICMP6STAT_INC(icp6s_pmtuchg);
3145 					}
3146 				} else
3147 					inp = (*notify)(inp,
3148 					    inet6ctlerrmap[cmd]);
3149 			}
3150 		}
3151 	} else {
3152 		bzero(&inc, sizeof(inc));
3153 		inc.inc_fibnum = M_GETFIB(m);
3154 		inc.inc_flags |= INC_ISIPV6;
3155 		inc.inc_fport = t_ports.th_dport;
3156 		inc.inc_lport = t_ports.th_sport;
3157 		inc.inc6_faddr = *dst;
3158 		inc.inc6_laddr = ip6->ip6_src;
3159 		syncache_unreach(&inc, icmp_tcp_seq, port);
3160 	}
3161 out:
3162 	if (inp != NULL)
3163 		INP_WUNLOCK(inp);
3164 }
3165 
3166 void
3167 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3168 {
3169 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3170 }
3171 
3172 void
3173 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3174 {
3175 	struct ip6ctlparam *ip6cp;
3176 	struct mbuf *m;
3177 	struct udphdr *udp;
3178 	uint16_t port;
3179 
3180 	ip6cp = (struct ip6ctlparam *)d;
3181 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3182 	if (m == NULL) {
3183 		return;
3184 	}
3185 	udp = mtod(m, struct udphdr *);
3186 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3187 		return;
3188 	}
3189 	port = udp->uh_dport;
3190 	m_adj(m, sizeof(struct udphdr));
3191 	if ((m->m_flags & M_PKTHDR) == 0) {
3192 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3193 	}
3194 	/* Now call in to the normal handling code */
3195 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3196 }
3197 
3198 #endif /* INET6 */
3199 
3200 static uint32_t
3201 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3202 {
3203 	SIPHASH_CTX ctx;
3204 	uint32_t hash[2];
3205 
3206 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3207 	    ("%s: keylen %u too short ", __func__, len));
3208 	SipHash24_Init(&ctx);
3209 	SipHash_SetKey(&ctx, (uint8_t *)key);
3210 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3211 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3212 	switch (inc->inc_flags & INC_ISIPV6) {
3213 #ifdef INET
3214 	case 0:
3215 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3216 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3217 		break;
3218 #endif
3219 #ifdef INET6
3220 	case INC_ISIPV6:
3221 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3222 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3223 		break;
3224 #endif
3225 	}
3226 	SipHash_Final((uint8_t *)hash, &ctx);
3227 
3228 	return (hash[0] ^ hash[1]);
3229 }
3230 
3231 uint32_t
3232 tcp_new_ts_offset(struct in_conninfo *inc)
3233 {
3234 	struct in_conninfo inc_store, *local_inc;
3235 
3236 	if (!V_tcp_ts_offset_per_conn) {
3237 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3238 		inc_store.inc_lport = 0;
3239 		inc_store.inc_fport = 0;
3240 		local_inc = &inc_store;
3241 	} else {
3242 		local_inc = inc;
3243 	}
3244 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3245 	    sizeof(V_ts_offset_secret)));
3246 }
3247 
3248 /*
3249  * Following is where TCP initial sequence number generation occurs.
3250  *
3251  * There are two places where we must use initial sequence numbers:
3252  * 1.  In SYN-ACK packets.
3253  * 2.  In SYN packets.
3254  *
3255  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3256  * tcp_syncache.c for details.
3257  *
3258  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3259  * depends on this property.  In addition, these ISNs should be
3260  * unguessable so as to prevent connection hijacking.  To satisfy
3261  * the requirements of this situation, the algorithm outlined in
3262  * RFC 1948 is used, with only small modifications.
3263  *
3264  * Implementation details:
3265  *
3266  * Time is based off the system timer, and is corrected so that it
3267  * increases by one megabyte per second.  This allows for proper
3268  * recycling on high speed LANs while still leaving over an hour
3269  * before rollover.
3270  *
3271  * As reading the *exact* system time is too expensive to be done
3272  * whenever setting up a TCP connection, we increment the time
3273  * offset in two ways.  First, a small random positive increment
3274  * is added to isn_offset for each connection that is set up.
3275  * Second, the function tcp_isn_tick fires once per clock tick
3276  * and increments isn_offset as necessary so that sequence numbers
3277  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3278  * random positive increments serve only to ensure that the same
3279  * exact sequence number is never sent out twice (as could otherwise
3280  * happen when a port is recycled in less than the system tick
3281  * interval.)
3282  *
3283  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3284  * between seeding of isn_secret.  This is normally set to zero,
3285  * as reseeding should not be necessary.
3286  *
3287  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3288  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3289  * general, this means holding an exclusive (write) lock.
3290  */
3291 
3292 #define ISN_BYTES_PER_SECOND 1048576
3293 #define ISN_STATIC_INCREMENT 4096
3294 #define ISN_RANDOM_INCREMENT (4096 - 1)
3295 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3296 
3297 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3298 VNET_DEFINE_STATIC(int, isn_last);
3299 VNET_DEFINE_STATIC(int, isn_last_reseed);
3300 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3301 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3302 
3303 #define	V_isn_secret			VNET(isn_secret)
3304 #define	V_isn_last			VNET(isn_last)
3305 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3306 #define	V_isn_offset			VNET(isn_offset)
3307 #define	V_isn_offset_old		VNET(isn_offset_old)
3308 
3309 tcp_seq
3310 tcp_new_isn(struct in_conninfo *inc)
3311 {
3312 	tcp_seq new_isn;
3313 	u_int32_t projected_offset;
3314 
3315 	ISN_LOCK();
3316 	/* Seed if this is the first use, reseed if requested. */
3317 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3318 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3319 		< (u_int)ticks))) {
3320 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3321 		V_isn_last_reseed = ticks;
3322 	}
3323 
3324 	/* Compute the hash and return the ISN. */
3325 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3326 	    sizeof(V_isn_secret));
3327 	V_isn_offset += ISN_STATIC_INCREMENT +
3328 		(arc4random() & ISN_RANDOM_INCREMENT);
3329 	if (ticks != V_isn_last) {
3330 		projected_offset = V_isn_offset_old +
3331 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3332 		if (SEQ_GT(projected_offset, V_isn_offset))
3333 			V_isn_offset = projected_offset;
3334 		V_isn_offset_old = V_isn_offset;
3335 		V_isn_last = ticks;
3336 	}
3337 	new_isn += V_isn_offset;
3338 	ISN_UNLOCK();
3339 	return (new_isn);
3340 }
3341 
3342 /*
3343  * When a specific ICMP unreachable message is received and the
3344  * connection state is SYN-SENT, drop the connection.  This behavior
3345  * is controlled by the icmp_may_rst sysctl.
3346  */
3347 struct inpcb *
3348 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3349 {
3350 	struct tcpcb *tp;
3351 
3352 	NET_EPOCH_ASSERT();
3353 	INP_WLOCK_ASSERT(inp);
3354 
3355 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3356 	    (inp->inp_flags & INP_DROPPED))
3357 		return (inp);
3358 
3359 	tp = intotcpcb(inp);
3360 	if (tp->t_state != TCPS_SYN_SENT)
3361 		return (inp);
3362 
3363 	if (IS_FASTOPEN(tp->t_flags))
3364 		tcp_fastopen_disable_path(tp);
3365 
3366 	tp = tcp_drop(tp, errno);
3367 	if (tp != NULL)
3368 		return (inp);
3369 	else
3370 		return (NULL);
3371 }
3372 
3373 /*
3374  * When `need fragmentation' ICMP is received, update our idea of the MSS
3375  * based on the new value. Also nudge TCP to send something, since we
3376  * know the packet we just sent was dropped.
3377  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3378  */
3379 static struct inpcb *
3380 tcp_mtudisc_notify(struct inpcb *inp, int error)
3381 {
3382 
3383 	tcp_mtudisc(inp, -1);
3384 	return (inp);
3385 }
3386 
3387 static void
3388 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3389 {
3390 	struct tcpcb *tp;
3391 	struct socket *so;
3392 
3393 	INP_WLOCK_ASSERT(inp);
3394 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3395 	    (inp->inp_flags & INP_DROPPED))
3396 		return;
3397 
3398 	tp = intotcpcb(inp);
3399 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3400 
3401 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3402 
3403 	so = inp->inp_socket;
3404 	SOCKBUF_LOCK(&so->so_snd);
3405 	/* If the mss is larger than the socket buffer, decrease the mss. */
3406 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3407 		tp->t_maxseg = so->so_snd.sb_hiwat;
3408 	SOCKBUF_UNLOCK(&so->so_snd);
3409 
3410 	TCPSTAT_INC(tcps_mturesent);
3411 	tp->t_rtttime = 0;
3412 	tp->snd_nxt = tp->snd_una;
3413 	tcp_free_sackholes(tp);
3414 	tp->snd_recover = tp->snd_max;
3415 	if (tp->t_flags & TF_SACK_PERMIT)
3416 		EXIT_FASTRECOVERY(tp->t_flags);
3417 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3418 		/*
3419 		 * Conceptually the snd_nxt setting
3420 		 * and freeing sack holes should
3421 		 * be done by the default stacks
3422 		 * own tfb_tcp_mtu_chg().
3423 		 */
3424 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3425 	}
3426 	tp->t_fb->tfb_tcp_output(tp);
3427 }
3428 
3429 #ifdef INET
3430 /*
3431  * Look-up the routing entry to the peer of this inpcb.  If no route
3432  * is found and it cannot be allocated, then return 0.  This routine
3433  * is called by TCP routines that access the rmx structure and by
3434  * tcp_mss_update to get the peer/interface MTU.
3435  */
3436 uint32_t
3437 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3438 {
3439 	struct nhop_object *nh;
3440 	struct ifnet *ifp;
3441 	uint32_t maxmtu = 0;
3442 
3443 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3444 
3445 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3446 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3447 		if (nh == NULL)
3448 			return (0);
3449 
3450 		ifp = nh->nh_ifp;
3451 		maxmtu = nh->nh_mtu;
3452 
3453 		/* Report additional interface capabilities. */
3454 		if (cap != NULL) {
3455 			if (ifp->if_capenable & IFCAP_TSO4 &&
3456 			    ifp->if_hwassist & CSUM_TSO) {
3457 				cap->ifcap |= CSUM_TSO;
3458 				cap->tsomax = ifp->if_hw_tsomax;
3459 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3460 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3461 			}
3462 		}
3463 	}
3464 	return (maxmtu);
3465 }
3466 #endif /* INET */
3467 
3468 #ifdef INET6
3469 uint32_t
3470 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3471 {
3472 	struct nhop_object *nh;
3473 	struct in6_addr dst6;
3474 	uint32_t scopeid;
3475 	struct ifnet *ifp;
3476 	uint32_t maxmtu = 0;
3477 
3478 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3479 
3480 	if (inc->inc_flags & INC_IPV6MINMTU)
3481 		return (IPV6_MMTU);
3482 
3483 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3484 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3485 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3486 		if (nh == NULL)
3487 			return (0);
3488 
3489 		ifp = nh->nh_ifp;
3490 		maxmtu = nh->nh_mtu;
3491 
3492 		/* Report additional interface capabilities. */
3493 		if (cap != NULL) {
3494 			if (ifp->if_capenable & IFCAP_TSO6 &&
3495 			    ifp->if_hwassist & CSUM_TSO) {
3496 				cap->ifcap |= CSUM_TSO;
3497 				cap->tsomax = ifp->if_hw_tsomax;
3498 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3499 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3500 			}
3501 		}
3502 	}
3503 
3504 	return (maxmtu);
3505 }
3506 #endif /* INET6 */
3507 
3508 /*
3509  * Calculate effective SMSS per RFC5681 definition for a given TCP
3510  * connection at its current state, taking into account SACK and etc.
3511  */
3512 u_int
3513 tcp_maxseg(const struct tcpcb *tp)
3514 {
3515 	u_int optlen;
3516 
3517 	if (tp->t_flags & TF_NOOPT)
3518 		return (tp->t_maxseg);
3519 
3520 	/*
3521 	 * Here we have a simplified code from tcp_addoptions(),
3522 	 * without a proper loop, and having most of paddings hardcoded.
3523 	 * We might make mistakes with padding here in some edge cases,
3524 	 * but this is harmless, since result of tcp_maxseg() is used
3525 	 * only in cwnd and ssthresh estimations.
3526 	 */
3527 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3528 		if (tp->t_flags & TF_RCVD_TSTMP)
3529 			optlen = TCPOLEN_TSTAMP_APPA;
3530 		else
3531 			optlen = 0;
3532 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3533 		if (tp->t_flags & TF_SIGNATURE)
3534 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3535 #endif
3536 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3537 			optlen += TCPOLEN_SACKHDR;
3538 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3539 			optlen = PADTCPOLEN(optlen);
3540 		}
3541 	} else {
3542 		if (tp->t_flags & TF_REQ_TSTMP)
3543 			optlen = TCPOLEN_TSTAMP_APPA;
3544 		else
3545 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3546 		if (tp->t_flags & TF_REQ_SCALE)
3547 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3548 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3549 		if (tp->t_flags & TF_SIGNATURE)
3550 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3551 #endif
3552 		if (tp->t_flags & TF_SACK_PERMIT)
3553 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3554 	}
3555 #undef PAD
3556 	optlen = min(optlen, TCP_MAXOLEN);
3557 	return (tp->t_maxseg - optlen);
3558 }
3559 
3560 
3561 u_int
3562 tcp_fixed_maxseg(const struct tcpcb *tp)
3563 {
3564 	int optlen;
3565 
3566 	if (tp->t_flags & TF_NOOPT)
3567 		return (tp->t_maxseg);
3568 
3569 	/*
3570 	 * Here we have a simplified code from tcp_addoptions(),
3571 	 * without a proper loop, and having most of paddings hardcoded.
3572 	 * We only consider fixed options that we would send every
3573 	 * time I.e. SACK is not considered. This is important
3574 	 * for cc modules to figure out what the modulo of the
3575 	 * cwnd should be.
3576 	 */
3577 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3578 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3579 		if (tp->t_flags & TF_RCVD_TSTMP)
3580 			optlen = TCPOLEN_TSTAMP_APPA;
3581 		else
3582 			optlen = 0;
3583 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3584 		if (tp->t_flags & TF_SIGNATURE)
3585 			optlen += PAD(TCPOLEN_SIGNATURE);
3586 #endif
3587 	} else {
3588 		if (tp->t_flags & TF_REQ_TSTMP)
3589 			optlen = TCPOLEN_TSTAMP_APPA;
3590 		else
3591 			optlen = PAD(TCPOLEN_MAXSEG);
3592 		if (tp->t_flags & TF_REQ_SCALE)
3593 			optlen += PAD(TCPOLEN_WINDOW);
3594 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3595 		if (tp->t_flags & TF_SIGNATURE)
3596 			optlen += PAD(TCPOLEN_SIGNATURE);
3597 #endif
3598 		if (tp->t_flags & TF_SACK_PERMIT)
3599 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3600 	}
3601 #undef PAD
3602 	optlen = min(optlen, TCP_MAXOLEN);
3603 	return (tp->t_maxseg - optlen);
3604 }
3605 
3606 
3607 
3608 static int
3609 sysctl_drop(SYSCTL_HANDLER_ARGS)
3610 {
3611 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3612 	struct sockaddr_storage addrs[2];
3613 	struct inpcb *inp;
3614 	struct tcpcb *tp;
3615 	struct tcptw *tw;
3616 	struct sockaddr_in *fin, *lin;
3617 	struct epoch_tracker et;
3618 #ifdef INET6
3619 	struct sockaddr_in6 *fin6, *lin6;
3620 #endif
3621 	int error;
3622 
3623 	inp = NULL;
3624 	fin = lin = NULL;
3625 #ifdef INET6
3626 	fin6 = lin6 = NULL;
3627 #endif
3628 	error = 0;
3629 
3630 	if (req->oldptr != NULL || req->oldlen != 0)
3631 		return (EINVAL);
3632 	if (req->newptr == NULL)
3633 		return (EPERM);
3634 	if (req->newlen < sizeof(addrs))
3635 		return (ENOMEM);
3636 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3637 	if (error)
3638 		return (error);
3639 
3640 	switch (addrs[0].ss_family) {
3641 #ifdef INET6
3642 	case AF_INET6:
3643 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3644 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3645 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3646 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3647 			return (EINVAL);
3648 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3649 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3650 				return (EINVAL);
3651 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3652 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3653 			fin = (struct sockaddr_in *)&addrs[0];
3654 			lin = (struct sockaddr_in *)&addrs[1];
3655 			break;
3656 		}
3657 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3658 		if (error)
3659 			return (error);
3660 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3661 		if (error)
3662 			return (error);
3663 		break;
3664 #endif
3665 #ifdef INET
3666 	case AF_INET:
3667 		fin = (struct sockaddr_in *)&addrs[0];
3668 		lin = (struct sockaddr_in *)&addrs[1];
3669 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3670 		    lin->sin_len != sizeof(struct sockaddr_in))
3671 			return (EINVAL);
3672 		break;
3673 #endif
3674 	default:
3675 		return (EINVAL);
3676 	}
3677 	NET_EPOCH_ENTER(et);
3678 	switch (addrs[0].ss_family) {
3679 #ifdef INET6
3680 	case AF_INET6:
3681 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3682 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3683 		    INPLOOKUP_WLOCKPCB, NULL);
3684 		break;
3685 #endif
3686 #ifdef INET
3687 	case AF_INET:
3688 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3689 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3690 		break;
3691 #endif
3692 	}
3693 	if (inp != NULL) {
3694 		if (inp->inp_flags & INP_TIMEWAIT) {
3695 			/*
3696 			 * XXXRW: There currently exists a state where an
3697 			 * inpcb is present, but its timewait state has been
3698 			 * discarded.  For now, don't allow dropping of this
3699 			 * type of inpcb.
3700 			 */
3701 			tw = intotw(inp);
3702 			if (tw != NULL)
3703 				tcp_twclose(tw, 0);
3704 			else
3705 				INP_WUNLOCK(inp);
3706 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3707 		    !SOLISTENING(inp->inp_socket)) {
3708 			tp = intotcpcb(inp);
3709 			tp = tcp_drop(tp, ECONNABORTED);
3710 			if (tp != NULL)
3711 				INP_WUNLOCK(inp);
3712 		} else
3713 			INP_WUNLOCK(inp);
3714 	} else
3715 		error = ESRCH;
3716 	NET_EPOCH_EXIT(et);
3717 	return (error);
3718 }
3719 
3720 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3721     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3722     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3723     "Drop TCP connection");
3724 
3725 #ifdef KERN_TLS
3726 static int
3727 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3728 {
3729 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3730 	struct sockaddr_storage addrs[2];
3731 	struct inpcb *inp;
3732 	struct sockaddr_in *fin, *lin;
3733 	struct epoch_tracker et;
3734 #ifdef INET6
3735 	struct sockaddr_in6 *fin6, *lin6;
3736 #endif
3737 	int error;
3738 
3739 	inp = NULL;
3740 	fin = lin = NULL;
3741 #ifdef INET6
3742 	fin6 = lin6 = NULL;
3743 #endif
3744 	error = 0;
3745 
3746 	if (req->oldptr != NULL || req->oldlen != 0)
3747 		return (EINVAL);
3748 	if (req->newptr == NULL)
3749 		return (EPERM);
3750 	if (req->newlen < sizeof(addrs))
3751 		return (ENOMEM);
3752 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3753 	if (error)
3754 		return (error);
3755 
3756 	switch (addrs[0].ss_family) {
3757 #ifdef INET6
3758 	case AF_INET6:
3759 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3760 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3761 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3762 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3763 			return (EINVAL);
3764 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3765 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3766 				return (EINVAL);
3767 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3768 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3769 			fin = (struct sockaddr_in *)&addrs[0];
3770 			lin = (struct sockaddr_in *)&addrs[1];
3771 			break;
3772 		}
3773 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3774 		if (error)
3775 			return (error);
3776 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3777 		if (error)
3778 			return (error);
3779 		break;
3780 #endif
3781 #ifdef INET
3782 	case AF_INET:
3783 		fin = (struct sockaddr_in *)&addrs[0];
3784 		lin = (struct sockaddr_in *)&addrs[1];
3785 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3786 		    lin->sin_len != sizeof(struct sockaddr_in))
3787 			return (EINVAL);
3788 		break;
3789 #endif
3790 	default:
3791 		return (EINVAL);
3792 	}
3793 	NET_EPOCH_ENTER(et);
3794 	switch (addrs[0].ss_family) {
3795 #ifdef INET6
3796 	case AF_INET6:
3797 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3798 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3799 		    INPLOOKUP_WLOCKPCB, NULL);
3800 		break;
3801 #endif
3802 #ifdef INET
3803 	case AF_INET:
3804 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3805 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3806 		break;
3807 #endif
3808 	}
3809 	NET_EPOCH_EXIT(et);
3810 	if (inp != NULL) {
3811 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3812 		    inp->inp_socket == NULL) {
3813 			error = ECONNRESET;
3814 			INP_WUNLOCK(inp);
3815 		} else {
3816 			struct socket *so;
3817 
3818 			so = inp->inp_socket;
3819 			soref(so);
3820 			error = ktls_set_tx_mode(so,
3821 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3822 			INP_WUNLOCK(inp);
3823 			SOCK_LOCK(so);
3824 			sorele(so);
3825 		}
3826 	} else
3827 		error = ESRCH;
3828 	return (error);
3829 }
3830 
3831 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3832     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3833     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3834     "Switch TCP connection to SW TLS");
3835 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3836     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3837     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3838     "Switch TCP connection to ifnet TLS");
3839 #endif
3840 
3841 /*
3842  * Generate a standardized TCP log line for use throughout the
3843  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3844  * allow use in the interrupt context.
3845  *
3846  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3847  * NB: The function may return NULL if memory allocation failed.
3848  *
3849  * Due to header inclusion and ordering limitations the struct ip
3850  * and ip6_hdr pointers have to be passed as void pointers.
3851  */
3852 char *
3853 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3854     const void *ip6hdr)
3855 {
3856 
3857 	/* Is logging enabled? */
3858 	if (V_tcp_log_in_vain == 0)
3859 		return (NULL);
3860 
3861 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3862 }
3863 
3864 char *
3865 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3866     const void *ip6hdr)
3867 {
3868 
3869 	/* Is logging enabled? */
3870 	if (tcp_log_debug == 0)
3871 		return (NULL);
3872 
3873 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3874 }
3875 
3876 static char *
3877 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3878     const void *ip6hdr)
3879 {
3880 	char *s, *sp;
3881 	size_t size;
3882 	struct ip *ip;
3883 #ifdef INET6
3884 	const struct ip6_hdr *ip6;
3885 
3886 	ip6 = (const struct ip6_hdr *)ip6hdr;
3887 #endif /* INET6 */
3888 	ip = (struct ip *)ip4hdr;
3889 
3890 	/*
3891 	 * The log line looks like this:
3892 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3893 	 */
3894 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3895 	    sizeof(PRINT_TH_FLAGS) + 1 +
3896 #ifdef INET6
3897 	    2 * INET6_ADDRSTRLEN;
3898 #else
3899 	    2 * INET_ADDRSTRLEN;
3900 #endif /* INET6 */
3901 
3902 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3903 	if (s == NULL)
3904 		return (NULL);
3905 
3906 	strcat(s, "TCP: [");
3907 	sp = s + strlen(s);
3908 
3909 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3910 		inet_ntoa_r(inc->inc_faddr, sp);
3911 		sp = s + strlen(s);
3912 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3913 		sp = s + strlen(s);
3914 		inet_ntoa_r(inc->inc_laddr, sp);
3915 		sp = s + strlen(s);
3916 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3917 #ifdef INET6
3918 	} else if (inc) {
3919 		ip6_sprintf(sp, &inc->inc6_faddr);
3920 		sp = s + strlen(s);
3921 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3922 		sp = s + strlen(s);
3923 		ip6_sprintf(sp, &inc->inc6_laddr);
3924 		sp = s + strlen(s);
3925 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3926 	} else if (ip6 && th) {
3927 		ip6_sprintf(sp, &ip6->ip6_src);
3928 		sp = s + strlen(s);
3929 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3930 		sp = s + strlen(s);
3931 		ip6_sprintf(sp, &ip6->ip6_dst);
3932 		sp = s + strlen(s);
3933 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3934 #endif /* INET6 */
3935 #ifdef INET
3936 	} else if (ip && th) {
3937 		inet_ntoa_r(ip->ip_src, sp);
3938 		sp = s + strlen(s);
3939 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3940 		sp = s + strlen(s);
3941 		inet_ntoa_r(ip->ip_dst, sp);
3942 		sp = s + strlen(s);
3943 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3944 #endif /* INET */
3945 	} else {
3946 		free(s, M_TCPLOG);
3947 		return (NULL);
3948 	}
3949 	sp = s + strlen(s);
3950 	if (th)
3951 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3952 	if (*(s + size - 1) != '\0')
3953 		panic("%s: string too long", __func__);
3954 	return (s);
3955 }
3956 
3957 /*
3958  * A subroutine which makes it easy to track TCP state changes with DTrace.
3959  * This function shouldn't be called for t_state initializations that don't
3960  * correspond to actual TCP state transitions.
3961  */
3962 void
3963 tcp_state_change(struct tcpcb *tp, int newstate)
3964 {
3965 #if defined(KDTRACE_HOOKS)
3966 	int pstate = tp->t_state;
3967 #endif
3968 
3969 	TCPSTATES_DEC(tp->t_state);
3970 	TCPSTATES_INC(newstate);
3971 	tp->t_state = newstate;
3972 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3973 }
3974 
3975 /*
3976  * Create an external-format (``xtcpcb'') structure using the information in
3977  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3978  * reduce the spew of irrelevant information over this interface, to isolate
3979  * user code from changes in the kernel structure, and potentially to provide
3980  * information-hiding if we decide that some of this information should be
3981  * hidden from users.
3982  */
3983 void
3984 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3985 {
3986 	struct tcpcb *tp = intotcpcb(inp);
3987 	struct tcptw *tw = intotw(inp);
3988 	sbintime_t now;
3989 
3990 	bzero(xt, sizeof(*xt));
3991 	if (inp->inp_flags & INP_TIMEWAIT) {
3992 		xt->t_state = TCPS_TIME_WAIT;
3993 		xt->xt_encaps_port = tw->t_port;
3994 	} else {
3995 		xt->t_state = tp->t_state;
3996 		xt->t_logstate = tp->t_logstate;
3997 		xt->t_flags = tp->t_flags;
3998 		xt->t_sndzerowin = tp->t_sndzerowin;
3999 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4000 		xt->t_rcvoopack = tp->t_rcvoopack;
4001 		xt->t_rcv_wnd = tp->rcv_wnd;
4002 		xt->t_snd_wnd = tp->snd_wnd;
4003 		xt->t_snd_cwnd = tp->snd_cwnd;
4004 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4005 		xt->t_maxseg = tp->t_maxseg;
4006 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4007 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4008 
4009 		now = getsbinuptime();
4010 #define	COPYTIMER(ttt)	do {						\
4011 		if (callout_active(&tp->t_timers->ttt))			\
4012 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4013 			    SBT_1MS;					\
4014 		else							\
4015 			xt->ttt = 0;					\
4016 } while (0)
4017 		COPYTIMER(tt_delack);
4018 		COPYTIMER(tt_rexmt);
4019 		COPYTIMER(tt_persist);
4020 		COPYTIMER(tt_keep);
4021 		COPYTIMER(tt_2msl);
4022 #undef COPYTIMER
4023 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4024 
4025 		xt->xt_encaps_port = tp->t_port;
4026 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4027 		    TCP_FUNCTION_NAME_LEN_MAX);
4028 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4029 		    TCP_CA_NAME_MAX);
4030 #ifdef TCP_BLACKBOX
4031 		(void)tcp_log_get_id(tp, xt->xt_logid);
4032 #endif
4033 	}
4034 
4035 	xt->xt_len = sizeof(struct xtcpcb);
4036 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4037 	if (inp->inp_socket == NULL)
4038 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4039 }
4040 
4041 void
4042 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4043 {
4044 	uint32_t bit, i;
4045 
4046 	if ((tp == NULL) ||
4047 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4048 	    (status == 0)) {
4049 		/* Invalid */
4050 		return;
4051 	}
4052 	if (status > (sizeof(uint32_t) * 8)) {
4053 		/* Should this be a KASSERT? */
4054 		return;
4055 	}
4056 	bit = 1U << (status - 1);
4057 	if (bit & tp->t_end_info_status) {
4058 		/* already logged */
4059 		return;
4060 	}
4061 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4062 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4063 			tp->t_end_info_bytes[i] = status;
4064 			tp->t_end_info_status |= bit;
4065 			break;
4066 		}
4067 	}
4068 }
4069 
4070 int
4071 tcp_can_enable_pacing(void)
4072 {
4073 
4074 	if ((tcp_pacing_limit == -1) ||
4075 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4076 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4077 		shadow_num_connections = number_of_tcp_connections_pacing;
4078 		return (1);
4079 	} else {
4080 		return (0);
4081 	}
4082 }
4083 
4084 static uint8_t tcp_pacing_warning = 0;
4085 
4086 void
4087 tcp_decrement_paced_conn(void)
4088 {
4089 	uint32_t ret;
4090 
4091 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4092 	shadow_num_connections = number_of_tcp_connections_pacing;
4093 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4094 	if (ret == 0) {
4095 		if (tcp_pacing_limit != -1) {
4096 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4097 			tcp_pacing_limit = 0;
4098 		} else if (tcp_pacing_warning == 0) {
4099 			printf("Warning pacing count is invalid, invalid decrement\n");
4100 			tcp_pacing_warning = 1;
4101 		}
4102 	}
4103 }
4104