xref: /freebsd/sys/netinet/tcp_subr.c (revision d30a1689f5b37e78ea189232a8b94a7011dc0dc8)
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 #include <sys/priv.h>
66 #include <sys/proc.h>
67 #include <sys/sdt.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/protosw.h>
71 #include <sys/random.h>
72 
73 #include <vm/uma.h>
74 
75 #include <net/route.h>
76 #include <net/route/nhop.h>
77 #include <net/if.h>
78 #include <net/if_var.h>
79 #include <net/vnet.h>
80 
81 #include <netinet/in.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/in_kdtrace.h>
84 #include <netinet/in_pcb.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/in_var.h>
87 #include <netinet/ip.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/ip_var.h>
90 #ifdef INET6
91 #include <netinet/icmp6.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/in6_fib.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet6/nd6.h>
98 #endif
99 
100 #include <netinet/tcp.h>
101 #ifdef INVARIANTS
102 #define TCPSTATES
103 #endif
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 struct inpcb *tcp_mtudisc(struct inpcb *, int);
376 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
377 		    const 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_default_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 void
394 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
395 {
396 	TCPSTAT_INC(tcps_dsack_count);
397 	tp->t_dsack_pack++;
398 	if (tlp == 0) {
399 		if (SEQ_GT(end, start)) {
400 			tp->t_dsack_bytes += (end - start);
401 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
402 		} else {
403 			tp->t_dsack_tlp_bytes += (start - end);
404 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
405 		}
406 	} else {
407 		if (SEQ_GT(end, start)) {
408 			tp->t_dsack_bytes += (end - start);
409 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
410 		} else {
411 			tp->t_dsack_tlp_bytes += (start - end);
412 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
413 		}
414 	}
415 }
416 
417 static struct tcp_function_block *
418 find_tcp_functions_locked(struct tcp_function_set *fs)
419 {
420 	struct tcp_function *f;
421 	struct tcp_function_block *blk=NULL;
422 
423 	TAILQ_FOREACH(f, &t_functions, tf_next) {
424 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
425 			blk = f->tf_fb;
426 			break;
427 		}
428 	}
429 	return(blk);
430 }
431 
432 static struct tcp_function_block *
433 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
434 {
435 	struct tcp_function_block *rblk=NULL;
436 	struct tcp_function *f;
437 
438 	TAILQ_FOREACH(f, &t_functions, tf_next) {
439 		if (f->tf_fb == blk) {
440 			rblk = blk;
441 			if (s) {
442 				*s = f;
443 			}
444 			break;
445 		}
446 	}
447 	return (rblk);
448 }
449 
450 struct tcp_function_block *
451 find_and_ref_tcp_functions(struct tcp_function_set *fs)
452 {
453 	struct tcp_function_block *blk;
454 
455 	rw_rlock(&tcp_function_lock);
456 	blk = find_tcp_functions_locked(fs);
457 	if (blk)
458 		refcount_acquire(&blk->tfb_refcnt);
459 	rw_runlock(&tcp_function_lock);
460 	return(blk);
461 }
462 
463 struct tcp_function_block *
464 find_and_ref_tcp_fb(struct tcp_function_block *blk)
465 {
466 	struct tcp_function_block *rblk;
467 
468 	rw_rlock(&tcp_function_lock);
469 	rblk = find_tcp_fb_locked(blk, NULL);
470 	if (rblk)
471 		refcount_acquire(&rblk->tfb_refcnt);
472 	rw_runlock(&tcp_function_lock);
473 	return(rblk);
474 }
475 
476 /* Find a matching alias for the given tcp_function_block. */
477 int
478 find_tcp_function_alias(struct tcp_function_block *blk,
479     struct tcp_function_set *fs)
480 {
481 	struct tcp_function *f;
482 	int found;
483 
484 	found = 0;
485 	rw_rlock(&tcp_function_lock);
486 	TAILQ_FOREACH(f, &t_functions, tf_next) {
487 		if ((f->tf_fb == blk) &&
488 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
489 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
490 			/* Matching function block with different name. */
491 			strncpy(fs->function_set_name, f->tf_name,
492 			    TCP_FUNCTION_NAME_LEN_MAX);
493 			found = 1;
494 			break;
495 		}
496 	}
497 	/* Null terminate the string appropriately. */
498 	if (found) {
499 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
500 	} else {
501 		fs->function_set_name[0] = '\0';
502 	}
503 	rw_runlock(&tcp_function_lock);
504 	return (found);
505 }
506 
507 static struct tcp_function_block *
508 find_and_ref_tcp_default_fb(void)
509 {
510 	struct tcp_function_block *rblk;
511 
512 	rw_rlock(&tcp_function_lock);
513 	rblk = tcp_func_set_ptr;
514 	refcount_acquire(&rblk->tfb_refcnt);
515 	rw_runlock(&tcp_function_lock);
516 	return (rblk);
517 }
518 
519 void
520 tcp_switch_back_to_default(struct tcpcb *tp)
521 {
522 	struct tcp_function_block *tfb;
523 
524 	KASSERT(tp->t_fb != &tcp_def_funcblk,
525 	    ("%s: called by the built-in default stack", __func__));
526 
527 	/*
528 	 * Release the old stack. This function will either find a new one
529 	 * or panic.
530 	 */
531 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
532 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
533 	refcount_release(&tp->t_fb->tfb_refcnt);
534 
535 	/*
536 	 * Now, we'll find a new function block to use.
537 	 * Start by trying the current user-selected
538 	 * default, unless this stack is the user-selected
539 	 * default.
540 	 */
541 	tfb = find_and_ref_tcp_default_fb();
542 	if (tfb == tp->t_fb) {
543 		refcount_release(&tfb->tfb_refcnt);
544 		tfb = NULL;
545 	}
546 	/* Does the stack accept this connection? */
547 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
548 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
549 		refcount_release(&tfb->tfb_refcnt);
550 		tfb = NULL;
551 	}
552 	/* Try to use that stack. */
553 	if (tfb != NULL) {
554 		/* Initialize the new stack. If it succeeds, we are done. */
555 		tp->t_fb = tfb;
556 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
557 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
558 			return;
559 
560 		/*
561 		 * Initialization failed. Release the reference count on
562 		 * the stack.
563 		 */
564 		refcount_release(&tfb->tfb_refcnt);
565 	}
566 
567 	/*
568 	 * If that wasn't feasible, use the built-in default
569 	 * stack which is not allowed to reject anyone.
570 	 */
571 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
572 	if (tfb == NULL) {
573 		/* there always should be a default */
574 		panic("Can't refer to tcp_def_funcblk");
575 	}
576 	if (tfb->tfb_tcp_handoff_ok != NULL) {
577 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
578 			/* The default stack cannot say no */
579 			panic("Default stack rejects a new session?");
580 		}
581 	}
582 	tp->t_fb = tfb;
583 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
584 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
585 		/* The default stack cannot fail */
586 		panic("Default stack initialization failed");
587 	}
588 }
589 
590 static bool
591 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
592     const struct sockaddr *sa, void *ctx)
593 {
594 	struct ip *iph;
595 #ifdef INET6
596 	struct ip6_hdr *ip6;
597 #endif
598 	struct udphdr *uh;
599 	struct tcphdr *th;
600 	int thlen;
601 	uint16_t port;
602 
603 	TCPSTAT_INC(tcps_tunneled_pkts);
604 	if ((m->m_flags & M_PKTHDR) == 0) {
605 		/* Can't handle one that is not a pkt hdr */
606 		TCPSTAT_INC(tcps_tunneled_errs);
607 		goto out;
608 	}
609 	thlen = sizeof(struct tcphdr);
610 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
611 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
612 		TCPSTAT_INC(tcps_tunneled_errs);
613 		goto out;
614 	}
615 	iph = mtod(m, struct ip *);
616 	uh = (struct udphdr *)((caddr_t)iph + off);
617 	th = (struct tcphdr *)(uh + 1);
618 	thlen = th->th_off << 2;
619 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
620 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
621 		if (m == NULL) {
622 			TCPSTAT_INC(tcps_tunneled_errs);
623 			goto out;
624 		} else {
625 			iph = mtod(m, struct ip *);
626 			uh = (struct udphdr *)((caddr_t)iph + off);
627 			th = (struct tcphdr *)(uh + 1);
628 		}
629 	}
630 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
631 	bcopy(th, uh, m->m_len - off);
632 	m->m_len -= sizeof(struct udphdr);
633 	m->m_pkthdr.len -= sizeof(struct udphdr);
634 	/*
635 	 * We use the same algorithm for
636 	 * both UDP and TCP for c-sum. So
637 	 * the code in tcp_input will skip
638 	 * the checksum. So we do nothing
639 	 * with the flag (m->m_pkthdr.csum_flags).
640 	 */
641 	switch (iph->ip_v) {
642 #ifdef INET
643 	case IPVERSION:
644 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
645 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
646 		break;
647 #endif
648 #ifdef INET6
649 	case IPV6_VERSION >> 4:
650 		ip6 = mtod(m, struct ip6_hdr *);
651 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
652 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
653 		break;
654 #endif
655 	default:
656 		goto out;
657 		break;
658 	}
659 	return (true);
660 out:
661 	m_freem(m);
662 
663 	return (true);
664 }
665 
666 static int
667 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
668 {
669 	int error=ENOENT;
670 	struct tcp_function_set fs;
671 	struct tcp_function_block *blk;
672 
673 	memset(&fs, 0, sizeof(fs));
674 	rw_rlock(&tcp_function_lock);
675 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
676 	if (blk) {
677 		/* Found him */
678 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
679 		fs.pcbcnt = blk->tfb_refcnt;
680 	}
681 	rw_runlock(&tcp_function_lock);
682 	error = sysctl_handle_string(oidp, fs.function_set_name,
683 				     sizeof(fs.function_set_name), req);
684 
685 	/* Check for error or no change */
686 	if (error != 0 || req->newptr == NULL)
687 		return(error);
688 
689 	rw_wlock(&tcp_function_lock);
690 	blk = find_tcp_functions_locked(&fs);
691 	if ((blk == NULL) ||
692 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
693 		error = ENOENT;
694 		goto done;
695 	}
696 	tcp_func_set_ptr = blk;
697 done:
698 	rw_wunlock(&tcp_function_lock);
699 	return (error);
700 }
701 
702 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
703     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
704     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
705     "Set/get the default TCP functions");
706 
707 static int
708 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
709 {
710 	int error, cnt, linesz;
711 	struct tcp_function *f;
712 	char *buffer, *cp;
713 	size_t bufsz, outsz;
714 	bool alias;
715 
716 	cnt = 0;
717 	rw_rlock(&tcp_function_lock);
718 	TAILQ_FOREACH(f, &t_functions, tf_next) {
719 		cnt++;
720 	}
721 	rw_runlock(&tcp_function_lock);
722 
723 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
724 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
725 
726 	error = 0;
727 	cp = buffer;
728 
729 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
730 	    "Alias", "PCB count");
731 	cp += linesz;
732 	bufsz -= linesz;
733 	outsz = linesz;
734 
735 	rw_rlock(&tcp_function_lock);
736 	TAILQ_FOREACH(f, &t_functions, tf_next) {
737 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
738 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
739 		    f->tf_fb->tfb_tcp_block_name,
740 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
741 		    alias ? f->tf_name : "-",
742 		    f->tf_fb->tfb_refcnt);
743 		if (linesz >= bufsz) {
744 			error = EOVERFLOW;
745 			break;
746 		}
747 		cp += linesz;
748 		bufsz -= linesz;
749 		outsz += linesz;
750 	}
751 	rw_runlock(&tcp_function_lock);
752 	if (error == 0)
753 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
754 	free(buffer, M_TEMP);
755 	return (error);
756 }
757 
758 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
759     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
760     NULL, 0, sysctl_net_inet_list_available, "A",
761     "list available TCP Function sets");
762 
763 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
764 
765 #ifdef INET
766 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
767 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
768 #endif
769 #ifdef INET6
770 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
771 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
772 #endif
773 
774 static void
775 tcp_over_udp_stop(void)
776 {
777 	/*
778 	 * This function assumes sysctl caller holds inp_rinfo_lock()
779 	 * for writing!
780 	 */
781 #ifdef INET
782 	if (V_udp4_tun_socket != NULL) {
783 		soclose(V_udp4_tun_socket);
784 		V_udp4_tun_socket = NULL;
785 	}
786 #endif
787 #ifdef INET6
788 	if (V_udp6_tun_socket != NULL) {
789 		soclose(V_udp6_tun_socket);
790 		V_udp6_tun_socket = NULL;
791 	}
792 #endif
793 }
794 
795 static int
796 tcp_over_udp_start(void)
797 {
798 	uint16_t port;
799 	int ret;
800 #ifdef INET
801 	struct sockaddr_in sin;
802 #endif
803 #ifdef INET6
804 	struct sockaddr_in6 sin6;
805 #endif
806 	/*
807 	 * This function assumes sysctl caller holds inp_info_rlock()
808 	 * for writing!
809 	 */
810 	port = V_tcp_udp_tunneling_port;
811 	if (ntohs(port) == 0) {
812 		/* Must have a port set */
813 		return (EINVAL);
814 	}
815 #ifdef INET
816 	if (V_udp4_tun_socket != NULL) {
817 		/* Already running -- must stop first */
818 		return (EALREADY);
819 	}
820 #endif
821 #ifdef INET6
822 	if (V_udp6_tun_socket != NULL) {
823 		/* Already running -- must stop first */
824 		return (EALREADY);
825 	}
826 #endif
827 #ifdef INET
828 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
829 	    SOCK_DGRAM, IPPROTO_UDP,
830 	    curthread->td_ucred, curthread))) {
831 		tcp_over_udp_stop();
832 		return (ret);
833 	}
834 	/* Call the special UDP hook. */
835 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
836 	    tcp_recv_udp_tunneled_packet,
837 	    tcp_ctlinput_viaudp,
838 	    NULL))) {
839 		tcp_over_udp_stop();
840 		return (ret);
841 	}
842 	/* Ok, we have a socket, bind it to the port. */
843 	memset(&sin, 0, sizeof(struct sockaddr_in));
844 	sin.sin_len = sizeof(struct sockaddr_in);
845 	sin.sin_family = AF_INET;
846 	sin.sin_port = htons(port);
847 	if ((ret = sobind(V_udp4_tun_socket,
848 	    (struct sockaddr *)&sin, curthread))) {
849 		tcp_over_udp_stop();
850 		return (ret);
851 	}
852 #endif
853 #ifdef INET6
854 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
855 	    SOCK_DGRAM, IPPROTO_UDP,
856 	    curthread->td_ucred, curthread))) {
857 		tcp_over_udp_stop();
858 		return (ret);
859 	}
860 	/* Call the special UDP hook. */
861 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
862 	    tcp_recv_udp_tunneled_packet,
863 	    tcp6_ctlinput_viaudp,
864 	    NULL))) {
865 		tcp_over_udp_stop();
866 		return (ret);
867 	}
868 	/* Ok, we have a socket, bind it to the port. */
869 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
870 	sin6.sin6_len = sizeof(struct sockaddr_in6);
871 	sin6.sin6_family = AF_INET6;
872 	sin6.sin6_port = htons(port);
873 	if ((ret = sobind(V_udp6_tun_socket,
874 	    (struct sockaddr *)&sin6, curthread))) {
875 		tcp_over_udp_stop();
876 		return (ret);
877 	}
878 #endif
879 	return (0);
880 }
881 
882 static int
883 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
884 {
885 	int error;
886 	uint32_t old, new;
887 
888 	old = V_tcp_udp_tunneling_port;
889 	new = old;
890 	error = sysctl_handle_int(oidp, &new, 0, req);
891 	if ((error == 0) &&
892 	    (req->newptr != NULL)) {
893 		if ((new < TCP_TUNNELING_PORT_MIN) ||
894 		    (new > TCP_TUNNELING_PORT_MAX)) {
895 			error = EINVAL;
896 		} else {
897 			V_tcp_udp_tunneling_port = new;
898 			if (old != 0) {
899 				tcp_over_udp_stop();
900 			}
901 			if (new != 0) {
902 				error = tcp_over_udp_start();
903 			}
904 		}
905 	}
906 	return (error);
907 }
908 
909 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
910     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
911     &VNET_NAME(tcp_udp_tunneling_port),
912     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
913     "Tunneling port for tcp over udp");
914 
915 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
916 
917 static int
918 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
919 {
920 	int error, new;
921 
922 	new = V_tcp_udp_tunneling_overhead;
923 	error = sysctl_handle_int(oidp, &new, 0, req);
924 	if (error == 0 && req->newptr) {
925 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
926 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
927 			error = EINVAL;
928 		else
929 			V_tcp_udp_tunneling_overhead = new;
930 	}
931 	return (error);
932 }
933 
934 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
935     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
936     &VNET_NAME(tcp_udp_tunneling_overhead),
937     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
938     "MSS reduction when using tcp over udp");
939 
940 /*
941  * Exports one (struct tcp_function_info) for each alias/name.
942  */
943 static int
944 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
945 {
946 	int cnt, error;
947 	struct tcp_function *f;
948 	struct tcp_function_info tfi;
949 
950 	/*
951 	 * We don't allow writes.
952 	 */
953 	if (req->newptr != NULL)
954 		return (EINVAL);
955 
956 	/*
957 	 * Wire the old buffer so we can directly copy the functions to
958 	 * user space without dropping the lock.
959 	 */
960 	if (req->oldptr != NULL) {
961 		error = sysctl_wire_old_buffer(req, 0);
962 		if (error)
963 			return (error);
964 	}
965 
966 	/*
967 	 * Walk the list and copy out matching entries. If INVARIANTS
968 	 * is compiled in, also walk the list to verify the length of
969 	 * the list matches what we have recorded.
970 	 */
971 	rw_rlock(&tcp_function_lock);
972 
973 	cnt = 0;
974 #ifndef INVARIANTS
975 	if (req->oldptr == NULL) {
976 		cnt = tcp_fb_cnt;
977 		goto skip_loop;
978 	}
979 #endif
980 	TAILQ_FOREACH(f, &t_functions, tf_next) {
981 #ifdef INVARIANTS
982 		cnt++;
983 #endif
984 		if (req->oldptr != NULL) {
985 			bzero(&tfi, sizeof(tfi));
986 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
987 			tfi.tfi_id = f->tf_fb->tfb_id;
988 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
989 			    sizeof(tfi.tfi_alias));
990 			(void)strlcpy(tfi.tfi_name,
991 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
992 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
993 			/*
994 			 * Don't stop on error, as that is the
995 			 * mechanism we use to accumulate length
996 			 * information if the buffer was too short.
997 			 */
998 		}
999 	}
1000 	KASSERT(cnt == tcp_fb_cnt,
1001 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1002 #ifndef INVARIANTS
1003 skip_loop:
1004 #endif
1005 	rw_runlock(&tcp_function_lock);
1006 	if (req->oldptr == NULL)
1007 		error = SYSCTL_OUT(req, NULL,
1008 		    (cnt + 1) * sizeof(struct tcp_function_info));
1009 
1010 	return (error);
1011 }
1012 
1013 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1014 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1015 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1016 	    "List TCP function block name-to-ID mappings");
1017 
1018 /*
1019  * tfb_tcp_handoff_ok() function for the default stack.
1020  * Note that we'll basically try to take all comers.
1021  */
1022 static int
1023 tcp_default_handoff_ok(struct tcpcb *tp)
1024 {
1025 
1026 	return (0);
1027 }
1028 
1029 /*
1030  * tfb_tcp_fb_init() function for the default stack.
1031  *
1032  * This handles making sure we have appropriate timers set if you are
1033  * transitioning a socket that has some amount of setup done.
1034  *
1035  * The init() fuction from the default can *never* return non-zero i.e.
1036  * it is required to always succeed since it is the stack of last resort!
1037  */
1038 static int
1039 tcp_default_fb_init(struct tcpcb *tp)
1040 {
1041 
1042 	struct socket *so;
1043 
1044 	INP_WLOCK_ASSERT(tp->t_inpcb);
1045 
1046 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1047 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1048 	    tp->t_state));
1049 
1050 	/*
1051 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1052 	 * know what to do for unexpected states (which includes TIME_WAIT).
1053 	 */
1054 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1055 		return (0);
1056 
1057 	/*
1058 	 * Make sure some kind of transmission timer is set if there is
1059 	 * outstanding data.
1060 	 */
1061 	so = tp->t_inpcb->inp_socket;
1062 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1063 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1064 	    tcp_timer_active(tp, TT_PERSIST))) {
1065 		/*
1066 		 * If the session has established and it looks like it should
1067 		 * be in the persist state, set the persist timer. Otherwise,
1068 		 * set the retransmit timer.
1069 		 */
1070 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1071 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1072 		    (int32_t)sbavail(&so->so_snd))
1073 			tcp_setpersist(tp);
1074 		else
1075 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1076 	}
1077 
1078 	/* All non-embryonic sessions get a keepalive timer. */
1079 	if (!tcp_timer_active(tp, TT_KEEP))
1080 		tcp_timer_activate(tp, TT_KEEP,
1081 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1082 		    TP_KEEPINIT(tp));
1083 
1084 	/*
1085 	 * Make sure critical variables are initialized
1086 	 * if transitioning while in Recovery.
1087 	 */
1088 	if IN_FASTRECOVERY(tp->t_flags) {
1089 		if (tp->sackhint.recover_fs == 0)
1090 			tp->sackhint.recover_fs = max(1,
1091 			    tp->snd_nxt - tp->snd_una);
1092 	}
1093 
1094 	return (0);
1095 }
1096 
1097 /*
1098  * tfb_tcp_fb_fini() function for the default stack.
1099  *
1100  * This changes state as necessary (or prudent) to prepare for another stack
1101  * to assume responsibility for the connection.
1102  */
1103 static void
1104 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1105 {
1106 
1107 	INP_WLOCK_ASSERT(tp->t_inpcb);
1108 	return;
1109 }
1110 
1111 /*
1112  * Target size of TCP PCB hash tables. Must be a power of two.
1113  *
1114  * Note that this can be overridden by the kernel environment
1115  * variable net.inet.tcp.tcbhashsize
1116  */
1117 #ifndef TCBHASHSIZE
1118 #define TCBHASHSIZE	0
1119 #endif
1120 
1121 /*
1122  * XXX
1123  * Callouts should be moved into struct tcp directly.  They are currently
1124  * separate because the tcpcb structure is exported to userland for sysctl
1125  * parsing purposes, which do not know about callouts.
1126  */
1127 struct tcpcb_mem {
1128 	struct	tcpcb		tcb;
1129 	struct	tcp_timer	tt;
1130 	struct	cc_var		ccv;
1131 #ifdef TCP_HHOOK
1132 	struct	osd		osd;
1133 #endif
1134 };
1135 
1136 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1137 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1138 
1139 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1140 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1141 
1142 static struct mtx isn_mtx;
1143 
1144 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1145 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1146 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1147 
1148 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1149 
1150 /*
1151  * Take a value and get the next power of 2 that doesn't overflow.
1152  * Used to size the tcp_inpcb hash buckets.
1153  */
1154 static int
1155 maketcp_hashsize(int size)
1156 {
1157 	int hashsize;
1158 
1159 	/*
1160 	 * auto tune.
1161 	 * get the next power of 2 higher than maxsockets.
1162 	 */
1163 	hashsize = 1 << fls(size);
1164 	/* catch overflow, and just go one power of 2 smaller */
1165 	if (hashsize < size) {
1166 		hashsize = 1 << (fls(size) - 1);
1167 	}
1168 	return (hashsize);
1169 }
1170 
1171 static volatile int next_tcp_stack_id = 1;
1172 
1173 /*
1174  * Register a TCP function block with the name provided in the names
1175  * array.  (Note that this function does NOT automatically register
1176  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1177  * explicitly include blk->tfb_tcp_block_name in the list of names if
1178  * you wish to register the stack with that name.)
1179  *
1180  * Either all name registrations will succeed or all will fail.  If
1181  * a name registration fails, the function will update the num_names
1182  * argument to point to the array index of the name that encountered
1183  * the failure.
1184  *
1185  * Returns 0 on success, or an error code on failure.
1186  */
1187 int
1188 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1189     const char *names[], int *num_names)
1190 {
1191 	struct tcp_function *n;
1192 	struct tcp_function_set fs;
1193 	int error, i;
1194 
1195 	KASSERT(names != NULL && *num_names > 0,
1196 	    ("%s: Called with 0-length name list", __func__));
1197 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1198 	KASSERT(rw_initialized(&tcp_function_lock),
1199 	    ("%s: called too early", __func__));
1200 
1201 	if ((blk->tfb_tcp_output == NULL) ||
1202 	    (blk->tfb_tcp_do_segment == NULL) ||
1203 	    (blk->tfb_tcp_ctloutput == NULL) ||
1204 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1205 		/*
1206 		 * These functions are required and you
1207 		 * need a name.
1208 		 */
1209 		*num_names = 0;
1210 		return (EINVAL);
1211 	}
1212 	if (blk->tfb_tcp_timer_stop_all ||
1213 	    blk->tfb_tcp_timer_activate ||
1214 	    blk->tfb_tcp_timer_active ||
1215 	    blk->tfb_tcp_timer_stop) {
1216 		/*
1217 		 * If you define one timer function you
1218 		 * must have them all.
1219 		 */
1220 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1221 		    (blk->tfb_tcp_timer_activate == NULL) ||
1222 		    (blk->tfb_tcp_timer_active == NULL) ||
1223 		    (blk->tfb_tcp_timer_stop == NULL)) {
1224 			*num_names = 0;
1225 			return (EINVAL);
1226 		}
1227 	}
1228 
1229 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1230 		*num_names = 0;
1231 		return (EINVAL);
1232 	}
1233 
1234 	refcount_init(&blk->tfb_refcnt, 0);
1235 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1236 	for (i = 0; i < *num_names; i++) {
1237 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1238 		if (n == NULL) {
1239 			error = ENOMEM;
1240 			goto cleanup;
1241 		}
1242 		n->tf_fb = blk;
1243 
1244 		(void)strlcpy(fs.function_set_name, names[i],
1245 		    sizeof(fs.function_set_name));
1246 		rw_wlock(&tcp_function_lock);
1247 		if (find_tcp_functions_locked(&fs) != NULL) {
1248 			/* Duplicate name space not allowed */
1249 			rw_wunlock(&tcp_function_lock);
1250 			free(n, M_TCPFUNCTIONS);
1251 			error = EALREADY;
1252 			goto cleanup;
1253 		}
1254 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1255 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1256 		tcp_fb_cnt++;
1257 		rw_wunlock(&tcp_function_lock);
1258 	}
1259 	return(0);
1260 
1261 cleanup:
1262 	/*
1263 	 * Deregister the names we just added. Because registration failed
1264 	 * for names[i], we don't need to deregister that name.
1265 	 */
1266 	*num_names = i;
1267 	rw_wlock(&tcp_function_lock);
1268 	while (--i >= 0) {
1269 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1270 			if (!strncmp(n->tf_name, names[i],
1271 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1272 				TAILQ_REMOVE(&t_functions, n, tf_next);
1273 				tcp_fb_cnt--;
1274 				n->tf_fb = NULL;
1275 				free(n, M_TCPFUNCTIONS);
1276 				break;
1277 			}
1278 		}
1279 	}
1280 	rw_wunlock(&tcp_function_lock);
1281 	return (error);
1282 }
1283 
1284 /*
1285  * Register a TCP function block using the name provided in the name
1286  * argument.
1287  *
1288  * Returns 0 on success, or an error code on failure.
1289  */
1290 int
1291 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1292     int wait)
1293 {
1294 	const char *name_list[1];
1295 	int num_names, rv;
1296 
1297 	num_names = 1;
1298 	if (name != NULL)
1299 		name_list[0] = name;
1300 	else
1301 		name_list[0] = blk->tfb_tcp_block_name;
1302 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1303 	return (rv);
1304 }
1305 
1306 /*
1307  * Register a TCP function block using the name defined in
1308  * blk->tfb_tcp_block_name.
1309  *
1310  * Returns 0 on success, or an error code on failure.
1311  */
1312 int
1313 register_tcp_functions(struct tcp_function_block *blk, int wait)
1314 {
1315 
1316 	return (register_tcp_functions_as_name(blk, NULL, wait));
1317 }
1318 
1319 /*
1320  * Deregister all names associated with a function block. This
1321  * functionally removes the function block from use within the system.
1322  *
1323  * When called with a true quiesce argument, mark the function block
1324  * as being removed so no more stacks will use it and determine
1325  * whether the removal would succeed.
1326  *
1327  * When called with a false quiesce argument, actually attempt the
1328  * removal.
1329  *
1330  * When called with a force argument, attempt to switch all TCBs to
1331  * use the default stack instead of returning EBUSY.
1332  *
1333  * Returns 0 on success (or if the removal would succeed, or an error
1334  * code on failure.
1335  */
1336 int
1337 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1338     bool force)
1339 {
1340 	struct tcp_function *f;
1341 
1342 	if (blk == &tcp_def_funcblk) {
1343 		/* You can't un-register the default */
1344 		return (EPERM);
1345 	}
1346 	rw_wlock(&tcp_function_lock);
1347 	if (blk == tcp_func_set_ptr) {
1348 		/* You can't free the current default */
1349 		rw_wunlock(&tcp_function_lock);
1350 		return (EBUSY);
1351 	}
1352 	/* Mark the block so no more stacks can use it. */
1353 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1354 	/*
1355 	 * If TCBs are still attached to the stack, attempt to switch them
1356 	 * to the default stack.
1357 	 */
1358 	if (force && blk->tfb_refcnt) {
1359 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1360 		    INPLOOKUP_WLOCKPCB);
1361 		struct inpcb *inp;
1362 		struct tcpcb *tp;
1363 		VNET_ITERATOR_DECL(vnet_iter);
1364 
1365 		rw_wunlock(&tcp_function_lock);
1366 
1367 		VNET_LIST_RLOCK();
1368 		VNET_FOREACH(vnet_iter) {
1369 			CURVNET_SET(vnet_iter);
1370 			while ((inp = inp_next(&inpi)) != NULL) {
1371 				if (inp->inp_flags & INP_TIMEWAIT)
1372 					continue;
1373 				tp = intotcpcb(inp);
1374 				if (tp == NULL || tp->t_fb != blk)
1375 					continue;
1376 				tcp_switch_back_to_default(tp);
1377 			}
1378 			CURVNET_RESTORE();
1379 		}
1380 		VNET_LIST_RUNLOCK();
1381 
1382 		rw_wlock(&tcp_function_lock);
1383 	}
1384 	if (blk->tfb_refcnt) {
1385 		/* TCBs still attached. */
1386 		rw_wunlock(&tcp_function_lock);
1387 		return (EBUSY);
1388 	}
1389 	if (quiesce) {
1390 		/* Skip removal. */
1391 		rw_wunlock(&tcp_function_lock);
1392 		return (0);
1393 	}
1394 	/* Remove any function names that map to this function block. */
1395 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1396 		TAILQ_REMOVE(&t_functions, f, tf_next);
1397 		tcp_fb_cnt--;
1398 		f->tf_fb = NULL;
1399 		free(f, M_TCPFUNCTIONS);
1400 	}
1401 	rw_wunlock(&tcp_function_lock);
1402 	return (0);
1403 }
1404 
1405 static void
1406 tcp_drain(void)
1407 {
1408 	struct epoch_tracker et;
1409 	VNET_ITERATOR_DECL(vnet_iter);
1410 
1411 	if (!do_tcpdrain)
1412 		return;
1413 
1414 	NET_EPOCH_ENTER(et);
1415 	VNET_LIST_RLOCK_NOSLEEP();
1416 	VNET_FOREACH(vnet_iter) {
1417 		CURVNET_SET(vnet_iter);
1418 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1419 		    INPLOOKUP_WLOCKPCB);
1420 		struct inpcb *inpb;
1421 		struct tcpcb *tcpb;
1422 
1423 	/*
1424 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1425 	 * if there is one...
1426 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1427 	 *      reassembly queue should be flushed, but in a situation
1428 	 *	where we're really low on mbufs, this is potentially
1429 	 *	useful.
1430 	 */
1431 		while ((inpb = inp_next(&inpi)) != NULL) {
1432 			if (inpb->inp_flags & INP_TIMEWAIT)
1433 				continue;
1434 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1435 				tcp_reass_flush(tcpb);
1436 				tcp_clean_sackreport(tcpb);
1437 #ifdef TCP_BLACKBOX
1438 				tcp_log_drain(tcpb);
1439 #endif
1440 #ifdef TCPPCAP
1441 				if (tcp_pcap_aggressive_free) {
1442 					/* Free the TCP PCAP queues. */
1443 					tcp_pcap_drain(&(tcpb->t_inpkts));
1444 					tcp_pcap_drain(&(tcpb->t_outpkts));
1445 				}
1446 #endif
1447 			}
1448 		}
1449 		CURVNET_RESTORE();
1450 	}
1451 	VNET_LIST_RUNLOCK_NOSLEEP();
1452 	NET_EPOCH_EXIT(et);
1453 }
1454 
1455 static void
1456 tcp_vnet_init(void *arg __unused)
1457 {
1458 
1459 #ifdef TCP_HHOOK
1460 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1461 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1462 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1463 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1464 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1465 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1466 #endif
1467 #ifdef STATS
1468 	if (tcp_stats_init())
1469 		printf("%s: WARNING: unable to initialise TCP stats\n",
1470 		    __func__);
1471 #endif
1472 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1473 	    tcp_tcbhashsize);
1474 
1475 	/*
1476 	 * These have to be type stable for the benefit of the timers.
1477 	 */
1478 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1479 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1480 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1481 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1482 
1483 	tcp_tw_init();
1484 	syncache_init();
1485 	tcp_hc_init();
1486 
1487 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1488 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1489 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1490 
1491 	tcp_fastopen_init();
1492 
1493 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1494 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1495 
1496 	V_tcp_msl = TCPTV_MSL;
1497 }
1498 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1499     tcp_vnet_init, NULL);
1500 
1501 static void
1502 tcp_init(void *arg __unused)
1503 {
1504 	const char *tcbhash_tuneable;
1505 	int hashsize;
1506 
1507 	tcp_reass_global_init();
1508 
1509 	/* XXX virtualize those below? */
1510 	tcp_delacktime = TCPTV_DELACK;
1511 	tcp_keepinit = TCPTV_KEEP_INIT;
1512 	tcp_keepidle = TCPTV_KEEP_IDLE;
1513 	tcp_keepintvl = TCPTV_KEEPINTVL;
1514 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1515 	tcp_rexmit_initial = TCPTV_RTOBASE;
1516 	if (tcp_rexmit_initial < 1)
1517 		tcp_rexmit_initial = 1;
1518 	tcp_rexmit_min = TCPTV_MIN;
1519 	if (tcp_rexmit_min < 1)
1520 		tcp_rexmit_min = 1;
1521 	tcp_persmin = TCPTV_PERSMIN;
1522 	tcp_persmax = TCPTV_PERSMAX;
1523 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1524 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1525 
1526 	/* Setup the tcp function block list */
1527 	TAILQ_INIT(&t_functions);
1528 	rw_init(&tcp_function_lock, "tcp_func_lock");
1529 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1530 #ifdef TCP_BLACKBOX
1531 	/* Initialize the TCP logging data. */
1532 	tcp_log_init();
1533 #endif
1534 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1535 
1536 	if (tcp_soreceive_stream) {
1537 #ifdef INET
1538 		tcp_protosw.pr_soreceive = soreceive_stream;
1539 #endif
1540 #ifdef INET6
1541 		tcp6_protosw.pr_soreceive = soreceive_stream;
1542 #endif /* INET6 */
1543 	}
1544 
1545 #ifdef INET6
1546 	max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1547 #else /* INET6 */
1548 	max_protohdr_grow(sizeof(struct tcpiphdr));
1549 #endif /* INET6 */
1550 
1551 	ISN_LOCK_INIT();
1552 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1553 		SHUTDOWN_PRI_DEFAULT);
1554 	EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1555 	EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1556 
1557 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1558 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1559 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1560 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1561 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1562 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1563 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1564 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1565 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1566 #ifdef TCPPCAP
1567 	tcp_pcap_init();
1568 #endif
1569 
1570 	hashsize = TCBHASHSIZE;
1571 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1572 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1573 	if (hashsize == 0) {
1574 		/*
1575 		 * Auto tune the hash size based on maxsockets.
1576 		 * A perfect hash would have a 1:1 mapping
1577 		 * (hashsize = maxsockets) however it's been
1578 		 * suggested that O(2) average is better.
1579 		 */
1580 		hashsize = maketcp_hashsize(maxsockets / 4);
1581 		/*
1582 		 * Our historical default is 512,
1583 		 * do not autotune lower than this.
1584 		 */
1585 		if (hashsize < 512)
1586 			hashsize = 512;
1587 		if (bootverbose)
1588 			printf("%s: %s auto tuned to %d\n", __func__,
1589 			    tcbhash_tuneable, hashsize);
1590 	}
1591 	/*
1592 	 * We require a hashsize to be a power of two.
1593 	 * Previously if it was not a power of two we would just reset it
1594 	 * back to 512, which could be a nasty surprise if you did not notice
1595 	 * the error message.
1596 	 * Instead what we do is clip it to the closest power of two lower
1597 	 * than the specified hash value.
1598 	 */
1599 	if (!powerof2(hashsize)) {
1600 		int oldhashsize = hashsize;
1601 
1602 		hashsize = maketcp_hashsize(hashsize);
1603 		/* prevent absurdly low value */
1604 		if (hashsize < 16)
1605 			hashsize = 16;
1606 		printf("%s: WARNING: TCB hash size not a power of 2, "
1607 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1608 		    hashsize);
1609 	}
1610 	tcp_tcbhashsize = hashsize;
1611 
1612 #ifdef INET
1613 	IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1614 #endif
1615 #ifdef INET6
1616 	IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1617 #endif
1618 }
1619 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1620 
1621 #ifdef VIMAGE
1622 static void
1623 tcp_destroy(void *unused __unused)
1624 {
1625 	int n;
1626 #ifdef TCP_HHOOK
1627 	int error;
1628 #endif
1629 
1630 	/*
1631 	 * All our processes are gone, all our sockets should be cleaned
1632 	 * up, which means, we should be past the tcp_discardcb() calls.
1633 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1634 	 */
1635 	for (;;) {
1636 		INP_INFO_WLOCK(&V_tcbinfo);
1637 		n = V_tcbinfo.ipi_count;
1638 		INP_INFO_WUNLOCK(&V_tcbinfo);
1639 		if (n == 0)
1640 			break;
1641 		pause("tcpdes", hz / 10);
1642 	}
1643 	tcp_hc_destroy();
1644 	syncache_destroy();
1645 	tcp_tw_destroy();
1646 	in_pcbinfo_destroy(&V_tcbinfo);
1647 	/* tcp_discardcb() clears the sack_holes up. */
1648 	uma_zdestroy(V_sack_hole_zone);
1649 	uma_zdestroy(V_tcpcb_zone);
1650 
1651 	/*
1652 	 * Cannot free the zone until all tcpcbs are released as we attach
1653 	 * the allocations to them.
1654 	 */
1655 	tcp_fastopen_destroy();
1656 
1657 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1658 	VNET_PCPUSTAT_FREE(tcpstat);
1659 
1660 #ifdef TCP_HHOOK
1661 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1662 	if (error != 0) {
1663 		printf("%s: WARNING: unable to deregister helper hook "
1664 		    "type=%d, id=%d: error %d returned\n", __func__,
1665 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1666 	}
1667 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1668 	if (error != 0) {
1669 		printf("%s: WARNING: unable to deregister helper hook "
1670 		    "type=%d, id=%d: error %d returned\n", __func__,
1671 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1672 	}
1673 #endif
1674 }
1675 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1676 #endif
1677 
1678 void
1679 tcp_fini(void *xtp)
1680 {
1681 
1682 }
1683 
1684 /*
1685  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1686  * tcp_template used to store this data in mbufs, but we now recopy it out
1687  * of the tcpcb each time to conserve mbufs.
1688  */
1689 void
1690 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1691 {
1692 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1693 
1694 	INP_WLOCK_ASSERT(inp);
1695 
1696 #ifdef INET6
1697 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1698 		struct ip6_hdr *ip6;
1699 
1700 		ip6 = (struct ip6_hdr *)ip_ptr;
1701 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1702 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1703 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1704 			(IPV6_VERSION & IPV6_VERSION_MASK);
1705 		if (port == 0)
1706 			ip6->ip6_nxt = IPPROTO_TCP;
1707 		else
1708 			ip6->ip6_nxt = IPPROTO_UDP;
1709 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1710 		ip6->ip6_src = inp->in6p_laddr;
1711 		ip6->ip6_dst = inp->in6p_faddr;
1712 	}
1713 #endif /* INET6 */
1714 #if defined(INET6) && defined(INET)
1715 	else
1716 #endif
1717 #ifdef INET
1718 	{
1719 		struct ip *ip;
1720 
1721 		ip = (struct ip *)ip_ptr;
1722 		ip->ip_v = IPVERSION;
1723 		ip->ip_hl = 5;
1724 		ip->ip_tos = inp->inp_ip_tos;
1725 		ip->ip_len = 0;
1726 		ip->ip_id = 0;
1727 		ip->ip_off = 0;
1728 		ip->ip_ttl = inp->inp_ip_ttl;
1729 		ip->ip_sum = 0;
1730 		if (port == 0)
1731 			ip->ip_p = IPPROTO_TCP;
1732 		else
1733 			ip->ip_p = IPPROTO_UDP;
1734 		ip->ip_src = inp->inp_laddr;
1735 		ip->ip_dst = inp->inp_faddr;
1736 	}
1737 #endif /* INET */
1738 	th->th_sport = inp->inp_lport;
1739 	th->th_dport = inp->inp_fport;
1740 	th->th_seq = 0;
1741 	th->th_ack = 0;
1742 	th->th_off = 5;
1743 	tcp_set_flags(th, 0);
1744 	th->th_win = 0;
1745 	th->th_urp = 0;
1746 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1747 }
1748 
1749 /*
1750  * Create template to be used to send tcp packets on a connection.
1751  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1752  * use for this function is in keepalives, which use tcp_respond.
1753  */
1754 struct tcptemp *
1755 tcpip_maketemplate(struct inpcb *inp)
1756 {
1757 	struct tcptemp *t;
1758 
1759 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1760 	if (t == NULL)
1761 		return (NULL);
1762 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1763 	return (t);
1764 }
1765 
1766 /*
1767  * Send a single message to the TCP at address specified by
1768  * the given TCP/IP header.  If m == NULL, then we make a copy
1769  * of the tcpiphdr at th and send directly to the addressed host.
1770  * This is used to force keep alive messages out using the TCP
1771  * template for a connection.  If flags are given then we send
1772  * a message back to the TCP which originated the segment th,
1773  * and discard the mbuf containing it and any other attached mbufs.
1774  *
1775  * In any case the ack and sequence number of the transmitted
1776  * segment are as specified by the parameters.
1777  *
1778  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1779  */
1780 void
1781 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1782     tcp_seq ack, tcp_seq seq, int flags)
1783 {
1784 	struct tcpopt to;
1785 	struct inpcb *inp;
1786 	struct ip *ip;
1787 	struct mbuf *optm;
1788 	struct udphdr *uh = NULL;
1789 	struct tcphdr *nth;
1790 	struct tcp_log_buffer *lgb;
1791 	u_char *optp;
1792 #ifdef INET6
1793 	struct ip6_hdr *ip6;
1794 	int isipv6;
1795 #endif /* INET6 */
1796 	int optlen, tlen, win, ulen;
1797 	bool incl_opts;
1798 	uint16_t port;
1799 	int output_ret;
1800 #ifdef INVARIANTS
1801 	int thflags = tcp_get_flags(th);
1802 #endif
1803 
1804 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1805 	NET_EPOCH_ASSERT();
1806 
1807 #ifdef INET6
1808 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1809 	ip6 = ipgen;
1810 #endif /* INET6 */
1811 	ip = ipgen;
1812 
1813 	if (tp != NULL) {
1814 		inp = tp->t_inpcb;
1815 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1816 		INP_LOCK_ASSERT(inp);
1817 	} else
1818 		inp = NULL;
1819 
1820 	if (m != NULL) {
1821 #ifdef INET6
1822 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1823 			port = m->m_pkthdr.tcp_tun_port;
1824 		else
1825 #endif
1826 		if (ip && (ip->ip_p == IPPROTO_UDP))
1827 			port = m->m_pkthdr.tcp_tun_port;
1828 		else
1829 			port = 0;
1830 	} else
1831 		port = tp->t_port;
1832 
1833 	incl_opts = false;
1834 	win = 0;
1835 	if (tp != NULL) {
1836 		if (!(flags & TH_RST)) {
1837 			win = sbspace(&inp->inp_socket->so_rcv);
1838 			if (win > TCP_MAXWIN << tp->rcv_scale)
1839 				win = TCP_MAXWIN << tp->rcv_scale;
1840 		}
1841 		if ((tp->t_flags & TF_NOOPT) == 0)
1842 			incl_opts = true;
1843 	}
1844 	if (m == NULL) {
1845 		m = m_gethdr(M_NOWAIT, MT_DATA);
1846 		if (m == NULL)
1847 			return;
1848 		m->m_data += max_linkhdr;
1849 #ifdef INET6
1850 		if (isipv6) {
1851 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1852 			      sizeof(struct ip6_hdr));
1853 			ip6 = mtod(m, struct ip6_hdr *);
1854 			nth = (struct tcphdr *)(ip6 + 1);
1855 			if (port) {
1856 				/* Insert a UDP header */
1857 				uh = (struct udphdr *)nth;
1858 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1859 				uh->uh_dport = port;
1860 				nth = (struct tcphdr *)(uh + 1);
1861 			}
1862 		} else
1863 #endif /* INET6 */
1864 		{
1865 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1866 			ip = mtod(m, struct ip *);
1867 			nth = (struct tcphdr *)(ip + 1);
1868 			if (port) {
1869 				/* Insert a UDP header */
1870 				uh = (struct udphdr *)nth;
1871 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1872 				uh->uh_dport = port;
1873 				nth = (struct tcphdr *)(uh + 1);
1874 			}
1875 		}
1876 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1877 		flags = TH_ACK;
1878 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1879 		struct mbuf *n;
1880 
1881 		/* Can't reuse 'm', allocate a new mbuf. */
1882 		n = m_gethdr(M_NOWAIT, MT_DATA);
1883 		if (n == NULL) {
1884 			m_freem(m);
1885 			return;
1886 		}
1887 
1888 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1889 			m_freem(m);
1890 			m_freem(n);
1891 			return;
1892 		}
1893 
1894 		n->m_data += max_linkhdr;
1895 		/* m_len is set later */
1896 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1897 #ifdef INET6
1898 		if (isipv6) {
1899 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1900 			      sizeof(struct ip6_hdr));
1901 			ip6 = mtod(n, struct ip6_hdr *);
1902 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1903 			nth = (struct tcphdr *)(ip6 + 1);
1904 			if (port) {
1905 				/* Insert a UDP header */
1906 				uh = (struct udphdr *)nth;
1907 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1908 				uh->uh_dport = port;
1909 				nth = (struct tcphdr *)(uh + 1);
1910 			}
1911 		} else
1912 #endif /* INET6 */
1913 		{
1914 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1915 			ip = mtod(n, struct ip *);
1916 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1917 			nth = (struct tcphdr *)(ip + 1);
1918 			if (port) {
1919 				/* Insert a UDP header */
1920 				uh = (struct udphdr *)nth;
1921 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1922 				uh->uh_dport = port;
1923 				nth = (struct tcphdr *)(uh + 1);
1924 			}
1925 		}
1926 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1927 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1928 		th = nth;
1929 		m_freem(m);
1930 		m = n;
1931 	} else {
1932 		/*
1933 		 *  reuse the mbuf.
1934 		 * XXX MRT We inherit the FIB, which is lucky.
1935 		 */
1936 		m_freem(m->m_next);
1937 		m->m_next = NULL;
1938 		m->m_data = (caddr_t)ipgen;
1939 		/* m_len is set later */
1940 #ifdef INET6
1941 		if (isipv6) {
1942 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1943 			nth = (struct tcphdr *)(ip6 + 1);
1944 		} else
1945 #endif /* INET6 */
1946 		{
1947 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1948 			nth = (struct tcphdr *)(ip + 1);
1949 		}
1950 		if (th != nth) {
1951 			/*
1952 			 * this is usually a case when an extension header
1953 			 * exists between the IPv6 header and the
1954 			 * TCP header.
1955 			 */
1956 			nth->th_sport = th->th_sport;
1957 			nth->th_dport = th->th_dport;
1958 		}
1959 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1960 #undef xchg
1961 	}
1962 	tlen = 0;
1963 #ifdef INET6
1964 	if (isipv6)
1965 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1966 #endif
1967 #if defined(INET) && defined(INET6)
1968 	else
1969 #endif
1970 #ifdef INET
1971 		tlen = sizeof (struct tcpiphdr);
1972 #endif
1973 	if (port)
1974 		tlen += sizeof (struct udphdr);
1975 #ifdef INVARIANTS
1976 	m->m_len = 0;
1977 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1978 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1979 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1980 #endif
1981 	m->m_len = tlen;
1982 	to.to_flags = 0;
1983 	if (incl_opts) {
1984 		/* Make sure we have room. */
1985 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1986 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1987 			if (m->m_next) {
1988 				optp = mtod(m->m_next, u_char *);
1989 				optm = m->m_next;
1990 			} else
1991 				incl_opts = false;
1992 		} else {
1993 			optp = (u_char *) (nth + 1);
1994 			optm = m;
1995 		}
1996 	}
1997 	if (incl_opts) {
1998 		/* Timestamps. */
1999 		if (tp->t_flags & TF_RCVD_TSTMP) {
2000 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
2001 			to.to_tsecr = tp->ts_recent;
2002 			to.to_flags |= TOF_TS;
2003 		}
2004 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2005 		/* TCP-MD5 (RFC2385). */
2006 		if (tp->t_flags & TF_SIGNATURE)
2007 			to.to_flags |= TOF_SIGNATURE;
2008 #endif
2009 		/* Add the options. */
2010 		tlen += optlen = tcp_addoptions(&to, optp);
2011 
2012 		/* Update m_len in the correct mbuf. */
2013 		optm->m_len += optlen;
2014 	} else
2015 		optlen = 0;
2016 #ifdef INET6
2017 	if (isipv6) {
2018 		if (uh) {
2019 			ulen = tlen - sizeof(struct ip6_hdr);
2020 			uh->uh_ulen = htons(ulen);
2021 		}
2022 		ip6->ip6_flow = 0;
2023 		ip6->ip6_vfc = IPV6_VERSION;
2024 		if (port)
2025 			ip6->ip6_nxt = IPPROTO_UDP;
2026 		else
2027 			ip6->ip6_nxt = IPPROTO_TCP;
2028 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
2029 	}
2030 #endif
2031 #if defined(INET) && defined(INET6)
2032 	else
2033 #endif
2034 #ifdef INET
2035 	{
2036 		if (uh) {
2037 			ulen = tlen - sizeof(struct ip);
2038 			uh->uh_ulen = htons(ulen);
2039 		}
2040 		ip->ip_len = htons(tlen);
2041 		ip->ip_ttl = V_ip_defttl;
2042 		if (port) {
2043 			ip->ip_p = IPPROTO_UDP;
2044 		} else {
2045 			ip->ip_p = IPPROTO_TCP;
2046 		}
2047 		if (V_path_mtu_discovery)
2048 			ip->ip_off |= htons(IP_DF);
2049 	}
2050 #endif
2051 	m->m_pkthdr.len = tlen;
2052 	m->m_pkthdr.rcvif = NULL;
2053 #ifdef MAC
2054 	if (inp != NULL) {
2055 		/*
2056 		 * Packet is associated with a socket, so allow the
2057 		 * label of the response to reflect the socket label.
2058 		 */
2059 		INP_LOCK_ASSERT(inp);
2060 		mac_inpcb_create_mbuf(inp, m);
2061 	} else {
2062 		/*
2063 		 * Packet is not associated with a socket, so possibly
2064 		 * update the label in place.
2065 		 */
2066 		mac_netinet_tcp_reply(m);
2067 	}
2068 #endif
2069 	nth->th_seq = htonl(seq);
2070 	nth->th_ack = htonl(ack);
2071 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2072 	tcp_set_flags(nth, flags);
2073 	if (tp != NULL)
2074 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2075 	else
2076 		nth->th_win = htons((u_short)win);
2077 	nth->th_urp = 0;
2078 
2079 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2080 	if (to.to_flags & TOF_SIGNATURE) {
2081 		if (!TCPMD5_ENABLED() ||
2082 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2083 			m_freem(m);
2084 			return;
2085 		}
2086 	}
2087 #endif
2088 
2089 #ifdef INET6
2090 	if (isipv6) {
2091 		if (port) {
2092 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2093 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2094 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2095 			nth->th_sum = 0;
2096 		} else {
2097 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2098 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2099 			nth->th_sum = in6_cksum_pseudo(ip6,
2100 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2101 		}
2102 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2103 		    NULL, NULL);
2104 	}
2105 #endif /* INET6 */
2106 #if defined(INET6) && defined(INET)
2107 	else
2108 #endif
2109 #ifdef INET
2110 	{
2111 		if (port) {
2112 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2113 			    htons(ulen + IPPROTO_UDP));
2114 			m->m_pkthdr.csum_flags = CSUM_UDP;
2115 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2116 			nth->th_sum = 0;
2117 		} else {
2118 			m->m_pkthdr.csum_flags = CSUM_TCP;
2119 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2120 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2121 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2122 		}
2123 	}
2124 #endif /* INET */
2125 #ifdef TCPDEBUG
2126 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2127 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2128 #endif
2129 	TCP_PROBE3(debug__output, tp, th, m);
2130 	if (flags & TH_RST)
2131 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2132 	lgb = NULL;
2133 	if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2134 		if (INP_WLOCKED(inp)) {
2135 			union tcp_log_stackspecific log;
2136 			struct timeval tv;
2137 
2138 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2139 			log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2140 			log.u_bbr.flex8 = 4;
2141 			log.u_bbr.pkts_out = tp->t_maxseg;
2142 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2143 			log.u_bbr.delivered = 0;
2144 			lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2145 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2146 		} else {
2147 			/*
2148 			 * We can not log the packet, since we only own the
2149 			 * read lock, but a write lock is needed. The read lock
2150 			 * is not upgraded to a write lock, since only getting
2151 			 * the read lock was done intentionally to improve the
2152 			 * handling of SYN flooding attacks.
2153 			 * This happens only for pure SYN segments received in
2154 			 * the initial CLOSED state, or received in a more
2155 			 * advanced state than listen and the UDP encapsulation
2156 			 * port is unexpected.
2157 			 * The incoming SYN segments do not really belong to
2158 			 * the TCP connection and the handling does not change
2159 			 * the state of the TCP connection. Therefore, the
2160 			 * sending of the RST segments is not logged. Please
2161 			 * note that also the incoming SYN segments are not
2162 			 * logged.
2163 			 *
2164 			 * The following code ensures that the above description
2165 			 * is and stays correct.
2166 			 */
2167 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2168 			    (tp->t_state == TCPS_CLOSED ||
2169 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2170 			    ("%s: Logging of TCP segment with flags 0x%b and "
2171 			    "UDP encapsulation port %u skipped in state %s",
2172 			    __func__, thflags, PRINT_TH_FLAGS,
2173 			    ntohs(port), tcpstates[tp->t_state]));
2174 		}
2175 	}
2176 
2177 	if (flags & TH_ACK)
2178 		TCPSTAT_INC(tcps_sndacks);
2179 	else if (flags & (TH_SYN|TH_FIN|TH_RST))
2180 		TCPSTAT_INC(tcps_sndctrl);
2181 	TCPSTAT_INC(tcps_sndtotal);
2182 
2183 #ifdef INET6
2184 	if (isipv6) {
2185 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2186 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2187 	}
2188 #endif /* INET6 */
2189 #if defined(INET) && defined(INET6)
2190 	else
2191 #endif
2192 #ifdef INET
2193 	{
2194 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2195 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2196 	}
2197 #endif
2198 	if (lgb != NULL)
2199 		lgb->tlb_errno = output_ret;
2200 }
2201 
2202 /*
2203  * Create a new TCP control block, making an
2204  * empty reassembly queue and hooking it to the argument
2205  * protocol control block.  The `inp' parameter must have
2206  * come from the zone allocator set up in tcp_init().
2207  */
2208 struct tcpcb *
2209 tcp_newtcpcb(struct inpcb *inp)
2210 {
2211 	struct tcpcb_mem *tm;
2212 	struct tcpcb *tp;
2213 #ifdef INET6
2214 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2215 #endif /* INET6 */
2216 
2217 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2218 	if (tm == NULL)
2219 		return (NULL);
2220 	tp = &tm->tcb;
2221 
2222 	/* Initialise cc_var struct for this tcpcb. */
2223 	tp->ccv = &tm->ccv;
2224 	tp->ccv->type = IPPROTO_TCP;
2225 	tp->ccv->ccvc.tcp = tp;
2226 	rw_rlock(&tcp_function_lock);
2227 	tp->t_fb = tcp_func_set_ptr;
2228 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2229 	rw_runlock(&tcp_function_lock);
2230 	/*
2231 	 * Use the current system default CC algorithm.
2232 	 */
2233 	cc_attach(tp, CC_DEFAULT_ALGO());
2234 
2235 	/*
2236 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2237 	 * is called.
2238 	 */
2239 	in_pcbref(inp);	/* Reference for tcpcb */
2240 	tp->t_inpcb = inp;
2241 
2242 	if (CC_ALGO(tp)->cb_init != NULL)
2243 		if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2244 			cc_detach(tp);
2245 			if (tp->t_fb->tfb_tcp_fb_fini)
2246 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2247 			in_pcbrele_wlocked(inp);
2248 			refcount_release(&tp->t_fb->tfb_refcnt);
2249 			uma_zfree(V_tcpcb_zone, tm);
2250 			return (NULL);
2251 		}
2252 
2253 #ifdef TCP_HHOOK
2254 	tp->osd = &tm->osd;
2255 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2256 		if (tp->t_fb->tfb_tcp_fb_fini)
2257 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2258 		in_pcbrele_wlocked(inp);
2259 		refcount_release(&tp->t_fb->tfb_refcnt);
2260 		uma_zfree(V_tcpcb_zone, tm);
2261 		return (NULL);
2262 	}
2263 #endif
2264 
2265 #ifdef VIMAGE
2266 	tp->t_vnet = inp->inp_vnet;
2267 #endif
2268 	tp->t_timers = &tm->tt;
2269 	TAILQ_INIT(&tp->t_segq);
2270 	tp->t_maxseg =
2271 #ifdef INET6
2272 		isipv6 ? V_tcp_v6mssdflt :
2273 #endif /* INET6 */
2274 		V_tcp_mssdflt;
2275 
2276 	/* Set up our timeouts. */
2277 	callout_init(&tp->t_timers->tt_rexmt, 1);
2278 	callout_init(&tp->t_timers->tt_persist, 1);
2279 	callout_init(&tp->t_timers->tt_keep, 1);
2280 	callout_init(&tp->t_timers->tt_2msl, 1);
2281 	callout_init(&tp->t_timers->tt_delack, 1);
2282 
2283 	switch (V_tcp_do_rfc1323) {
2284 		case 0:
2285 			break;
2286 		default:
2287 		case 1:
2288 			tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2289 			break;
2290 		case 2:
2291 			tp->t_flags = TF_REQ_SCALE;
2292 			break;
2293 		case 3:
2294 			tp->t_flags = TF_REQ_TSTMP;
2295 			break;
2296 	}
2297 	if (V_tcp_do_sack)
2298 		tp->t_flags |= TF_SACK_PERMIT;
2299 	TAILQ_INIT(&tp->snd_holes);
2300 
2301 	/*
2302 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2303 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2304 	 * reasonable initial retransmit time.
2305 	 */
2306 	tp->t_srtt = TCPTV_SRTTBASE;
2307 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2308 	tp->t_rttmin = tcp_rexmit_min;
2309 	tp->t_rxtcur = tcp_rexmit_initial;
2310 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2311 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2312 	tp->t_rcvtime = ticks;
2313 	/*
2314 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2315 	 * because the socket may be bound to an IPv6 wildcard address,
2316 	 * which may match an IPv4-mapped IPv6 address.
2317 	 */
2318 	inp->inp_ip_ttl = V_ip_defttl;
2319 	inp->inp_ppcb = tp;
2320 #ifdef TCPPCAP
2321 	/*
2322 	 * Init the TCP PCAP queues.
2323 	 */
2324 	tcp_pcap_tcpcb_init(tp);
2325 #endif
2326 #ifdef TCP_BLACKBOX
2327 	/* Initialize the per-TCPCB log data. */
2328 	tcp_log_tcpcbinit(tp);
2329 #endif
2330 	tp->t_pacing_rate = -1;
2331 	if (tp->t_fb->tfb_tcp_fb_init) {
2332 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2333 			refcount_release(&tp->t_fb->tfb_refcnt);
2334 			in_pcbrele_wlocked(inp);
2335 			uma_zfree(V_tcpcb_zone, tm);
2336 			return (NULL);
2337 		}
2338 	}
2339 #ifdef STATS
2340 	if (V_tcp_perconn_stats_enable == 1)
2341 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2342 #endif
2343 	if (V_tcp_do_lrd)
2344 		tp->t_flags |= TF_LRD;
2345 	return (tp);		/* XXX */
2346 }
2347 
2348 /*
2349  * Drop a TCP connection, reporting
2350  * the specified error.  If connection is synchronized,
2351  * then send a RST to peer.
2352  */
2353 struct tcpcb *
2354 tcp_drop(struct tcpcb *tp, int errno)
2355 {
2356 	struct socket *so = tp->t_inpcb->inp_socket;
2357 
2358 	NET_EPOCH_ASSERT();
2359 	INP_WLOCK_ASSERT(tp->t_inpcb);
2360 
2361 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2362 		tcp_state_change(tp, TCPS_CLOSED);
2363 		/* Don't use tcp_output() here due to possible recursion. */
2364 		(void)tcp_output_nodrop(tp);
2365 		TCPSTAT_INC(tcps_drops);
2366 	} else
2367 		TCPSTAT_INC(tcps_conndrops);
2368 	if (errno == ETIMEDOUT && tp->t_softerror)
2369 		errno = tp->t_softerror;
2370 	so->so_error = errno;
2371 	return (tcp_close(tp));
2372 }
2373 
2374 void
2375 tcp_discardcb(struct tcpcb *tp)
2376 {
2377 	struct inpcb *inp = tp->t_inpcb;
2378 
2379 	INP_WLOCK_ASSERT(inp);
2380 
2381 	/*
2382 	 * Make sure that all of our timers are stopped before we delete the
2383 	 * PCB.
2384 	 *
2385 	 * If stopping a timer fails, we schedule a discard function in same
2386 	 * callout, and the last discard function called will take care of
2387 	 * deleting the tcpcb.
2388 	 */
2389 	tp->t_timers->tt_draincnt = 0;
2390 	tcp_timer_stop(tp, TT_REXMT);
2391 	tcp_timer_stop(tp, TT_PERSIST);
2392 	tcp_timer_stop(tp, TT_KEEP);
2393 	tcp_timer_stop(tp, TT_2MSL);
2394 	tcp_timer_stop(tp, TT_DELACK);
2395 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2396 		/*
2397 		 * Call the stop-all function of the methods,
2398 		 * this function should call the tcp_timer_stop()
2399 		 * method with each of the function specific timeouts.
2400 		 * That stop will be called via the tfb_tcp_timer_stop()
2401 		 * which should use the async drain function of the
2402 		 * callout system (see tcp_var.h).
2403 		 */
2404 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2405 	}
2406 
2407 	/* free the reassembly queue, if any */
2408 	tcp_reass_flush(tp);
2409 
2410 #ifdef TCP_OFFLOAD
2411 	/* Disconnect offload device, if any. */
2412 	if (tp->t_flags & TF_TOE)
2413 		tcp_offload_detach(tp);
2414 #endif
2415 
2416 	tcp_free_sackholes(tp);
2417 
2418 #ifdef TCPPCAP
2419 	/* Free the TCP PCAP queues. */
2420 	tcp_pcap_drain(&(tp->t_inpkts));
2421 	tcp_pcap_drain(&(tp->t_outpkts));
2422 #endif
2423 
2424 	/* Allow the CC algorithm to clean up after itself. */
2425 	if (CC_ALGO(tp)->cb_destroy != NULL)
2426 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2427 	CC_DATA(tp) = NULL;
2428 	/* Detach from the CC algorithm */
2429 	cc_detach(tp);
2430 
2431 #ifdef TCP_HHOOK
2432 	khelp_destroy_osd(tp->osd);
2433 #endif
2434 #ifdef STATS
2435 	stats_blob_destroy(tp->t_stats);
2436 #endif
2437 
2438 	CC_ALGO(tp) = NULL;
2439 	inp->inp_ppcb = NULL;
2440 	if (tp->t_timers->tt_draincnt == 0) {
2441 		bool released __diagused;
2442 
2443 		released = tcp_freecb(tp);
2444 		KASSERT(!released, ("%s: inp %p should not have been released "
2445 		    "here", __func__, inp));
2446 	}
2447 }
2448 
2449 bool
2450 tcp_freecb(struct tcpcb *tp)
2451 {
2452 	struct inpcb *inp = tp->t_inpcb;
2453 	struct socket *so = inp->inp_socket;
2454 #ifdef INET6
2455 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2456 #endif
2457 
2458 	INP_WLOCK_ASSERT(inp);
2459 	MPASS(tp->t_timers->tt_draincnt == 0);
2460 
2461 	/* We own the last reference on tcpcb, let's free it. */
2462 #ifdef TCP_BLACKBOX
2463 	tcp_log_tcpcbfini(tp);
2464 #endif
2465 	TCPSTATES_DEC(tp->t_state);
2466 	if (tp->t_fb->tfb_tcp_fb_fini)
2467 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2468 
2469 	/*
2470 	 * If we got enough samples through the srtt filter,
2471 	 * save the rtt and rttvar in the routing entry.
2472 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2473 	 * 4 samples is enough for the srtt filter to converge
2474 	 * to within enough % of the correct value; fewer samples
2475 	 * and we could save a bogus rtt. The danger is not high
2476 	 * as tcp quickly recovers from everything.
2477 	 * XXX: Works very well but needs some more statistics!
2478 	 *
2479 	 * XXXRRS: Updating must be after the stack fini() since
2480 	 * that may be converting some internal representation of
2481 	 * say srtt etc into the general one used by other stacks.
2482 	 * Lets also at least protect against the so being NULL
2483 	 * as RW stated below.
2484 	 */
2485 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2486 		struct hc_metrics_lite metrics;
2487 		uint32_t ssthresh;
2488 
2489 		bzero(&metrics, sizeof(metrics));
2490 		/*
2491 		 * Update the ssthresh always when the conditions below
2492 		 * are satisfied. This gives us better new start value
2493 		 * for the congestion avoidance for new connections.
2494 		 * ssthresh is only set if packet loss occurred on a session.
2495 		 *
2496 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2497 		 * being torn down.  Ideally this code would not use 'so'.
2498 		 */
2499 		ssthresh = tp->snd_ssthresh;
2500 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2501 			/*
2502 			 * convert the limit from user data bytes to
2503 			 * packets then to packet data bytes.
2504 			 */
2505 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2506 			if (ssthresh < 2)
2507 				ssthresh = 2;
2508 			ssthresh *= (tp->t_maxseg +
2509 #ifdef INET6
2510 			    (isipv6 ? sizeof (struct ip6_hdr) +
2511 			    sizeof (struct tcphdr) :
2512 #endif
2513 			    sizeof (struct tcpiphdr)
2514 #ifdef INET6
2515 			    )
2516 #endif
2517 			    );
2518 		} else
2519 			ssthresh = 0;
2520 		metrics.rmx_ssthresh = ssthresh;
2521 
2522 		metrics.rmx_rtt = tp->t_srtt;
2523 		metrics.rmx_rttvar = tp->t_rttvar;
2524 		metrics.rmx_cwnd = tp->snd_cwnd;
2525 		metrics.rmx_sendpipe = 0;
2526 		metrics.rmx_recvpipe = 0;
2527 
2528 		tcp_hc_update(&inp->inp_inc, &metrics);
2529 	}
2530 
2531 	refcount_release(&tp->t_fb->tfb_refcnt);
2532 	uma_zfree(V_tcpcb_zone, tp);
2533 
2534 	return (in_pcbrele_wlocked(inp));
2535 }
2536 
2537 /*
2538  * Attempt to close a TCP control block, marking it as dropped, and freeing
2539  * the socket if we hold the only reference.
2540  */
2541 struct tcpcb *
2542 tcp_close(struct tcpcb *tp)
2543 {
2544 	struct inpcb *inp = tp->t_inpcb;
2545 	struct socket *so;
2546 
2547 	INP_WLOCK_ASSERT(inp);
2548 
2549 #ifdef TCP_OFFLOAD
2550 	if (tp->t_state == TCPS_LISTEN)
2551 		tcp_offload_listen_stop(tp);
2552 #endif
2553 	/*
2554 	 * This releases the TFO pending counter resource for TFO listen
2555 	 * sockets as well as passively-created TFO sockets that transition
2556 	 * from SYN_RECEIVED to CLOSED.
2557 	 */
2558 	if (tp->t_tfo_pending) {
2559 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2560 		tp->t_tfo_pending = NULL;
2561 	}
2562 #ifdef TCPHPTS
2563 	tcp_hpts_remove(inp);
2564 #endif
2565 	in_pcbdrop(inp);
2566 	TCPSTAT_INC(tcps_closed);
2567 	if (tp->t_state != TCPS_CLOSED)
2568 		tcp_state_change(tp, TCPS_CLOSED);
2569 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2570 	so = inp->inp_socket;
2571 	soisdisconnected(so);
2572 	if (inp->inp_flags & INP_SOCKREF) {
2573 		inp->inp_flags &= ~INP_SOCKREF;
2574 		INP_WUNLOCK(inp);
2575 		sorele(so);
2576 		return (NULL);
2577 	}
2578 	return (tp);
2579 }
2580 
2581 /*
2582  * Notify a tcp user of an asynchronous error;
2583  * store error as soft error, but wake up user
2584  * (for now, won't do anything until can select for soft error).
2585  *
2586  * Do not wake up user since there currently is no mechanism for
2587  * reporting soft errors (yet - a kqueue filter may be added).
2588  */
2589 static struct inpcb *
2590 tcp_notify(struct inpcb *inp, int error)
2591 {
2592 	struct tcpcb *tp;
2593 
2594 	INP_WLOCK_ASSERT(inp);
2595 
2596 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2597 	    (inp->inp_flags & INP_DROPPED))
2598 		return (inp);
2599 
2600 	tp = intotcpcb(inp);
2601 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2602 
2603 	/*
2604 	 * Ignore some errors if we are hooked up.
2605 	 * If connection hasn't completed, has retransmitted several times,
2606 	 * and receives a second error, give up now.  This is better
2607 	 * than waiting a long time to establish a connection that
2608 	 * can never complete.
2609 	 */
2610 	if (tp->t_state == TCPS_ESTABLISHED &&
2611 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2612 	     error == EHOSTDOWN)) {
2613 		if (inp->inp_route.ro_nh) {
2614 			NH_FREE(inp->inp_route.ro_nh);
2615 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2616 		}
2617 		return (inp);
2618 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2619 	    tp->t_softerror) {
2620 		tp = tcp_drop(tp, error);
2621 		if (tp != NULL)
2622 			return (inp);
2623 		else
2624 			return (NULL);
2625 	} else {
2626 		tp->t_softerror = error;
2627 		return (inp);
2628 	}
2629 #if 0
2630 	wakeup( &so->so_timeo);
2631 	sorwakeup(so);
2632 	sowwakeup(so);
2633 #endif
2634 }
2635 
2636 static int
2637 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2638 {
2639 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2640 	    INPLOOKUP_RLOCKPCB);
2641 	struct xinpgen xig;
2642 	struct inpcb *inp;
2643 	int error;
2644 
2645 	if (req->newptr != NULL)
2646 		return (EPERM);
2647 
2648 	if (req->oldptr == NULL) {
2649 		int n;
2650 
2651 		n = V_tcbinfo.ipi_count +
2652 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2653 		n += imax(n / 8, 10);
2654 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2655 		return (0);
2656 	}
2657 
2658 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2659 		return (error);
2660 
2661 	bzero(&xig, sizeof(xig));
2662 	xig.xig_len = sizeof xig;
2663 	xig.xig_count = V_tcbinfo.ipi_count +
2664 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2665 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2666 	xig.xig_sogen = so_gencnt;
2667 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2668 	if (error)
2669 		return (error);
2670 
2671 	error = syncache_pcblist(req);
2672 	if (error)
2673 		return (error);
2674 
2675 	while ((inp = inp_next(&inpi)) != NULL) {
2676 		if (inp->inp_gencnt <= xig.xig_gen) {
2677 			int crerr;
2678 
2679 			/*
2680 			 * XXX: This use of cr_cansee(), introduced with
2681 			 * TCP state changes, is not quite right, but for
2682 			 * now, better than nothing.
2683 			 */
2684 			if (inp->inp_flags & INP_TIMEWAIT) {
2685 				if (intotw(inp) != NULL)
2686 					crerr = cr_cansee(req->td->td_ucred,
2687 					    intotw(inp)->tw_cred);
2688 				else
2689 					crerr = EINVAL;	/* Skip this inp. */
2690 			} else
2691 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2692 			if (crerr == 0) {
2693 				struct xtcpcb xt;
2694 
2695 				tcp_inptoxtp(inp, &xt);
2696 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2697 				if (error) {
2698 					INP_RUNLOCK(inp);
2699 					break;
2700 				} else
2701 					continue;
2702 			}
2703 		}
2704 	}
2705 
2706 	if (!error) {
2707 		/*
2708 		 * Give the user an updated idea of our state.
2709 		 * If the generation differs from what we told
2710 		 * her before, she knows that something happened
2711 		 * while we were processing this request, and it
2712 		 * might be necessary to retry.
2713 		 */
2714 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2715 		xig.xig_sogen = so_gencnt;
2716 		xig.xig_count = V_tcbinfo.ipi_count +
2717 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2718 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2719 	}
2720 
2721 	return (error);
2722 }
2723 
2724 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2725     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2726     NULL, 0, tcp_pcblist, "S,xtcpcb",
2727     "List of active TCP connections");
2728 
2729 #ifdef INET
2730 static int
2731 tcp_getcred(SYSCTL_HANDLER_ARGS)
2732 {
2733 	struct xucred xuc;
2734 	struct sockaddr_in addrs[2];
2735 	struct epoch_tracker et;
2736 	struct inpcb *inp;
2737 	int error;
2738 
2739 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2740 	if (error)
2741 		return (error);
2742 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2743 	if (error)
2744 		return (error);
2745 	NET_EPOCH_ENTER(et);
2746 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2747 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2748 	NET_EPOCH_EXIT(et);
2749 	if (inp != NULL) {
2750 		if (inp->inp_socket == NULL)
2751 			error = ENOENT;
2752 		if (error == 0)
2753 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2754 		if (error == 0)
2755 			cru2x(inp->inp_cred, &xuc);
2756 		INP_RUNLOCK(inp);
2757 	} else
2758 		error = ENOENT;
2759 	if (error == 0)
2760 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2761 	return (error);
2762 }
2763 
2764 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2765     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2766     0, 0, tcp_getcred, "S,xucred",
2767     "Get the xucred of a TCP connection");
2768 #endif /* INET */
2769 
2770 #ifdef INET6
2771 static int
2772 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2773 {
2774 	struct epoch_tracker et;
2775 	struct xucred xuc;
2776 	struct sockaddr_in6 addrs[2];
2777 	struct inpcb *inp;
2778 	int error;
2779 #ifdef INET
2780 	int mapped = 0;
2781 #endif
2782 
2783 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2784 	if (error)
2785 		return (error);
2786 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2787 	if (error)
2788 		return (error);
2789 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2790 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2791 		return (error);
2792 	}
2793 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2794 #ifdef INET
2795 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2796 			mapped = 1;
2797 		else
2798 #endif
2799 			return (EINVAL);
2800 	}
2801 
2802 	NET_EPOCH_ENTER(et);
2803 #ifdef INET
2804 	if (mapped == 1)
2805 		inp = in_pcblookup(&V_tcbinfo,
2806 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2807 			addrs[1].sin6_port,
2808 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2809 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2810 	else
2811 #endif
2812 		inp = in6_pcblookup(&V_tcbinfo,
2813 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2814 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2815 			INPLOOKUP_RLOCKPCB, NULL);
2816 	NET_EPOCH_EXIT(et);
2817 	if (inp != NULL) {
2818 		if (inp->inp_socket == NULL)
2819 			error = ENOENT;
2820 		if (error == 0)
2821 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2822 		if (error == 0)
2823 			cru2x(inp->inp_cred, &xuc);
2824 		INP_RUNLOCK(inp);
2825 	} else
2826 		error = ENOENT;
2827 	if (error == 0)
2828 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2829 	return (error);
2830 }
2831 
2832 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2833     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2834     0, 0, tcp6_getcred, "S,xucred",
2835     "Get the xucred of a TCP6 connection");
2836 #endif /* INET6 */
2837 
2838 #ifdef INET
2839 /* Path MTU to try next when a fragmentation-needed message is received. */
2840 static inline int
2841 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2842 {
2843 	int mtu = ntohs(icp->icmp_nextmtu);
2844 
2845 	/* If no alternative MTU was proposed, try the next smaller one. */
2846 	if (!mtu)
2847 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2848 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2849 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2850 
2851 	return (mtu);
2852 }
2853 
2854 static void
2855 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2856 {
2857 	struct ip *ip = vip;
2858 	struct tcphdr *th;
2859 	struct in_addr faddr;
2860 	struct inpcb *inp;
2861 	struct tcpcb *tp;
2862 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2863 	struct icmp *icp;
2864 	struct in_conninfo inc;
2865 	tcp_seq icmp_tcp_seq;
2866 	int mtu;
2867 
2868 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2869 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2870 		return;
2871 
2872 	if (cmd == PRC_MSGSIZE)
2873 		notify = tcp_mtudisc_notify;
2874 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2875 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2876 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2877 		notify = tcp_drop_syn_sent;
2878 
2879 	/*
2880 	 * Hostdead is ugly because it goes linearly through all PCBs.
2881 	 * XXX: We never get this from ICMP, otherwise it makes an
2882 	 * excellent DoS attack on machines with many connections.
2883 	 */
2884 	else if (cmd == PRC_HOSTDEAD)
2885 		ip = NULL;
2886 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2887 		return;
2888 
2889 	if (ip == NULL) {
2890 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2891 		return;
2892 	}
2893 
2894 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2895 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2896 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2897 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2898 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2899 		/* signal EHOSTDOWN, as it flushes the cached route */
2900 		inp = (*notify)(inp, EHOSTDOWN);
2901 		goto out;
2902 	}
2903 	icmp_tcp_seq = th->th_seq;
2904 	if (inp != NULL)  {
2905 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2906 		    !(inp->inp_flags & INP_DROPPED) &&
2907 		    !(inp->inp_socket == NULL)) {
2908 			tp = intotcpcb(inp);
2909 #ifdef TCP_OFFLOAD
2910 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2911 				/*
2912 				 * MTU discovery for offloaded connections.  Let
2913 				 * the TOE driver verify seq# and process it.
2914 				 */
2915 				mtu = tcp_next_pmtu(icp, ip);
2916 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2917 				goto out;
2918 			}
2919 #endif
2920 			if (tp->t_port != port) {
2921 				goto out;
2922 			}
2923 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2924 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2925 				if (cmd == PRC_MSGSIZE) {
2926 					/*
2927 					 * MTU discovery: we got a needfrag and
2928 					 * will potentially try a lower MTU.
2929 					 */
2930 					mtu = tcp_next_pmtu(icp, ip);
2931 
2932 					/*
2933 					 * Only process the offered MTU if it
2934 					 * is smaller than the current one.
2935 					 */
2936 					if (mtu < tp->t_maxseg +
2937 					    sizeof(struct tcpiphdr)) {
2938 						bzero(&inc, sizeof(inc));
2939 						inc.inc_faddr = faddr;
2940 						inc.inc_fibnum =
2941 						    inp->inp_inc.inc_fibnum;
2942 						tcp_hc_updatemtu(&inc, mtu);
2943 						inp = tcp_mtudisc(inp, mtu);
2944 					}
2945 				} else
2946 					inp = (*notify)(inp,
2947 					    inetctlerrmap[cmd]);
2948 			}
2949 		}
2950 	} else {
2951 		bzero(&inc, sizeof(inc));
2952 		inc.inc_fport = th->th_dport;
2953 		inc.inc_lport = th->th_sport;
2954 		inc.inc_faddr = faddr;
2955 		inc.inc_laddr = ip->ip_src;
2956 		syncache_unreach(&inc, icmp_tcp_seq, port);
2957 	}
2958 out:
2959 	if (inp != NULL)
2960 		INP_WUNLOCK(inp);
2961 }
2962 
2963 void
2964 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2965 {
2966 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2967 }
2968 
2969 void
2970 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2971 {
2972 	/* Its a tunneled TCP over UDP icmp */
2973 	struct ip *outer_ip, *inner_ip;
2974 	struct icmp *icmp;
2975 	struct udphdr *udp;
2976 	struct tcphdr *th, ttemp;
2977 	int i_hlen, o_len;
2978 	uint16_t port;
2979 
2980 	inner_ip = (struct ip *)vip;
2981 	icmp = (struct icmp *)((caddr_t)inner_ip -
2982 	    (sizeof(struct icmp) - sizeof(struct ip)));
2983 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2984 	i_hlen = inner_ip->ip_hl << 2;
2985 	o_len = ntohs(outer_ip->ip_len);
2986 	if (o_len <
2987 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2988 		/* Not enough data present */
2989 		return;
2990 	}
2991 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2992 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2993 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2994 		return;
2995 	}
2996 	port = udp->uh_dport;
2997 	th = (struct tcphdr *)(udp + 1);
2998 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2999 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
3000 	/* Now adjust down the size of the outer IP header */
3001 	o_len -= sizeof(struct udphdr);
3002 	outer_ip->ip_len = htons(o_len);
3003 	/* Now call in to the normal handling code */
3004 	tcp_ctlinput_with_port(cmd, sa, vip, port);
3005 }
3006 #endif /* INET */
3007 
3008 #ifdef INET6
3009 static inline int
3010 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3011 {
3012 	int mtu = ntohl(icmp6->icmp6_mtu);
3013 
3014 	/*
3015 	 * If no alternative MTU was proposed, or the proposed MTU was too
3016 	 * small, set to the min.
3017 	 */
3018 	if (mtu < IPV6_MMTU)
3019 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
3020 	return (mtu);
3021 }
3022 
3023 static void
3024 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3025 {
3026 	struct in6_addr *dst;
3027 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3028 	struct ip6_hdr *ip6;
3029 	struct mbuf *m;
3030 	struct inpcb *inp;
3031 	struct tcpcb *tp;
3032 	struct icmp6_hdr *icmp6;
3033 	struct ip6ctlparam *ip6cp = NULL;
3034 	const struct sockaddr_in6 *sa6_src = NULL;
3035 	struct in_conninfo inc;
3036 	struct tcp_ports {
3037 		uint16_t th_sport;
3038 		uint16_t th_dport;
3039 	} t_ports;
3040 	tcp_seq icmp_tcp_seq;
3041 	unsigned int mtu;
3042 	unsigned int off;
3043 
3044 	if (sa->sa_family != AF_INET6 ||
3045 	    sa->sa_len != sizeof(struct sockaddr_in6))
3046 		return;
3047 
3048 	/* if the parameter is from icmp6, decode it. */
3049 	if (d != NULL) {
3050 		ip6cp = (struct ip6ctlparam *)d;
3051 		icmp6 = ip6cp->ip6c_icmp6;
3052 		m = ip6cp->ip6c_m;
3053 		ip6 = ip6cp->ip6c_ip6;
3054 		off = ip6cp->ip6c_off;
3055 		sa6_src = ip6cp->ip6c_src;
3056 		dst = ip6cp->ip6c_finaldst;
3057 	} else {
3058 		m = NULL;
3059 		ip6 = NULL;
3060 		off = 0;	/* fool gcc */
3061 		sa6_src = &sa6_any;
3062 		dst = NULL;
3063 	}
3064 
3065 	if (cmd == PRC_MSGSIZE)
3066 		notify = tcp_mtudisc_notify;
3067 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3068 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3069 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3070 		notify = tcp_drop_syn_sent;
3071 
3072 	/*
3073 	 * Hostdead is ugly because it goes linearly through all PCBs.
3074 	 * XXX: We never get this from ICMP, otherwise it makes an
3075 	 * excellent DoS attack on machines with many connections.
3076 	 */
3077 	else if (cmd == PRC_HOSTDEAD)
3078 		ip6 = NULL;
3079 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3080 		return;
3081 
3082 	if (ip6 == NULL) {
3083 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3084 			      (const struct sockaddr *)sa6_src,
3085 			      0, cmd, NULL, notify);
3086 		return;
3087 	}
3088 
3089 	/* Check if we can safely get the ports from the tcp hdr */
3090 	if (m == NULL ||
3091 	    (m->m_pkthdr.len <
3092 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3093 		return;
3094 	}
3095 	bzero(&t_ports, sizeof(struct tcp_ports));
3096 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3097 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3098 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3099 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3100 		/* signal EHOSTDOWN, as it flushes the cached route */
3101 		inp = (*notify)(inp, EHOSTDOWN);
3102 		goto out;
3103 	}
3104 	off += sizeof(struct tcp_ports);
3105 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3106 		goto out;
3107 	}
3108 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3109 	if (inp != NULL)  {
3110 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3111 		    !(inp->inp_flags & INP_DROPPED) &&
3112 		    !(inp->inp_socket == NULL)) {
3113 			tp = intotcpcb(inp);
3114 #ifdef TCP_OFFLOAD
3115 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3116 				/* MTU discovery for offloaded connections. */
3117 				mtu = tcp6_next_pmtu(icmp6);
3118 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3119 				goto out;
3120 			}
3121 #endif
3122 			if (tp->t_port != port) {
3123 				goto out;
3124 			}
3125 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3126 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3127 				if (cmd == PRC_MSGSIZE) {
3128 					/*
3129 					 * MTU discovery:
3130 					 * If we got a needfrag set the MTU
3131 					 * in the route to the suggested new
3132 					 * value (if given) and then notify.
3133 					 */
3134 					mtu = tcp6_next_pmtu(icmp6);
3135 
3136 					bzero(&inc, sizeof(inc));
3137 					inc.inc_fibnum = M_GETFIB(m);
3138 					inc.inc_flags |= INC_ISIPV6;
3139 					inc.inc6_faddr = *dst;
3140 					if (in6_setscope(&inc.inc6_faddr,
3141 						m->m_pkthdr.rcvif, NULL))
3142 						goto out;
3143 					/*
3144 					 * Only process the offered MTU if it
3145 					 * is smaller than the current one.
3146 					 */
3147 					if (mtu < tp->t_maxseg +
3148 					    sizeof (struct tcphdr) +
3149 					    sizeof (struct ip6_hdr)) {
3150 						tcp_hc_updatemtu(&inc, mtu);
3151 						tcp_mtudisc(inp, mtu);
3152 						ICMP6STAT_INC(icp6s_pmtuchg);
3153 					}
3154 				} else
3155 					inp = (*notify)(inp,
3156 					    inet6ctlerrmap[cmd]);
3157 			}
3158 		}
3159 	} else {
3160 		bzero(&inc, sizeof(inc));
3161 		inc.inc_fibnum = M_GETFIB(m);
3162 		inc.inc_flags |= INC_ISIPV6;
3163 		inc.inc_fport = t_ports.th_dport;
3164 		inc.inc_lport = t_ports.th_sport;
3165 		inc.inc6_faddr = *dst;
3166 		inc.inc6_laddr = ip6->ip6_src;
3167 		syncache_unreach(&inc, icmp_tcp_seq, port);
3168 	}
3169 out:
3170 	if (inp != NULL)
3171 		INP_WUNLOCK(inp);
3172 }
3173 
3174 void
3175 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3176 {
3177 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3178 }
3179 
3180 void
3181 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3182 {
3183 	struct ip6ctlparam *ip6cp;
3184 	struct mbuf *m;
3185 	struct udphdr *udp;
3186 	uint16_t port;
3187 
3188 	ip6cp = (struct ip6ctlparam *)d;
3189 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3190 	if (m == NULL) {
3191 		return;
3192 	}
3193 	udp = mtod(m, struct udphdr *);
3194 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3195 		return;
3196 	}
3197 	port = udp->uh_dport;
3198 	m_adj(m, sizeof(struct udphdr));
3199 	if ((m->m_flags & M_PKTHDR) == 0) {
3200 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3201 	}
3202 	/* Now call in to the normal handling code */
3203 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3204 }
3205 
3206 #endif /* INET6 */
3207 
3208 static uint32_t
3209 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3210 {
3211 	SIPHASH_CTX ctx;
3212 	uint32_t hash[2];
3213 
3214 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3215 	    ("%s: keylen %u too short ", __func__, len));
3216 	SipHash24_Init(&ctx);
3217 	SipHash_SetKey(&ctx, (uint8_t *)key);
3218 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3219 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3220 	switch (inc->inc_flags & INC_ISIPV6) {
3221 #ifdef INET
3222 	case 0:
3223 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3224 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3225 		break;
3226 #endif
3227 #ifdef INET6
3228 	case INC_ISIPV6:
3229 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3230 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3231 		break;
3232 #endif
3233 	}
3234 	SipHash_Final((uint8_t *)hash, &ctx);
3235 
3236 	return (hash[0] ^ hash[1]);
3237 }
3238 
3239 uint32_t
3240 tcp_new_ts_offset(struct in_conninfo *inc)
3241 {
3242 	struct in_conninfo inc_store, *local_inc;
3243 
3244 	if (!V_tcp_ts_offset_per_conn) {
3245 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3246 		inc_store.inc_lport = 0;
3247 		inc_store.inc_fport = 0;
3248 		local_inc = &inc_store;
3249 	} else {
3250 		local_inc = inc;
3251 	}
3252 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3253 	    sizeof(V_ts_offset_secret)));
3254 }
3255 
3256 /*
3257  * Following is where TCP initial sequence number generation occurs.
3258  *
3259  * There are two places where we must use initial sequence numbers:
3260  * 1.  In SYN-ACK packets.
3261  * 2.  In SYN packets.
3262  *
3263  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3264  * tcp_syncache.c for details.
3265  *
3266  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3267  * depends on this property.  In addition, these ISNs should be
3268  * unguessable so as to prevent connection hijacking.  To satisfy
3269  * the requirements of this situation, the algorithm outlined in
3270  * RFC 1948 is used, with only small modifications.
3271  *
3272  * Implementation details:
3273  *
3274  * Time is based off the system timer, and is corrected so that it
3275  * increases by one megabyte per second.  This allows for proper
3276  * recycling on high speed LANs while still leaving over an hour
3277  * before rollover.
3278  *
3279  * As reading the *exact* system time is too expensive to be done
3280  * whenever setting up a TCP connection, we increment the time
3281  * offset in two ways.  First, a small random positive increment
3282  * is added to isn_offset for each connection that is set up.
3283  * Second, the function tcp_isn_tick fires once per clock tick
3284  * and increments isn_offset as necessary so that sequence numbers
3285  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3286  * random positive increments serve only to ensure that the same
3287  * exact sequence number is never sent out twice (as could otherwise
3288  * happen when a port is recycled in less than the system tick
3289  * interval.)
3290  *
3291  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3292  * between seeding of isn_secret.  This is normally set to zero,
3293  * as reseeding should not be necessary.
3294  *
3295  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3296  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3297  * general, this means holding an exclusive (write) lock.
3298  */
3299 
3300 #define ISN_BYTES_PER_SECOND 1048576
3301 #define ISN_STATIC_INCREMENT 4096
3302 #define ISN_RANDOM_INCREMENT (4096 - 1)
3303 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3304 
3305 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3306 VNET_DEFINE_STATIC(int, isn_last);
3307 VNET_DEFINE_STATIC(int, isn_last_reseed);
3308 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3309 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3310 
3311 #define	V_isn_secret			VNET(isn_secret)
3312 #define	V_isn_last			VNET(isn_last)
3313 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3314 #define	V_isn_offset			VNET(isn_offset)
3315 #define	V_isn_offset_old		VNET(isn_offset_old)
3316 
3317 tcp_seq
3318 tcp_new_isn(struct in_conninfo *inc)
3319 {
3320 	tcp_seq new_isn;
3321 	u_int32_t projected_offset;
3322 
3323 	ISN_LOCK();
3324 	/* Seed if this is the first use, reseed if requested. */
3325 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3326 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3327 		< (u_int)ticks))) {
3328 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3329 		V_isn_last_reseed = ticks;
3330 	}
3331 
3332 	/* Compute the hash and return the ISN. */
3333 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3334 	    sizeof(V_isn_secret));
3335 	V_isn_offset += ISN_STATIC_INCREMENT +
3336 		(arc4random() & ISN_RANDOM_INCREMENT);
3337 	if (ticks != V_isn_last) {
3338 		projected_offset = V_isn_offset_old +
3339 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3340 		if (SEQ_GT(projected_offset, V_isn_offset))
3341 			V_isn_offset = projected_offset;
3342 		V_isn_offset_old = V_isn_offset;
3343 		V_isn_last = ticks;
3344 	}
3345 	new_isn += V_isn_offset;
3346 	ISN_UNLOCK();
3347 	return (new_isn);
3348 }
3349 
3350 /*
3351  * When a specific ICMP unreachable message is received and the
3352  * connection state is SYN-SENT, drop the connection.  This behavior
3353  * is controlled by the icmp_may_rst sysctl.
3354  */
3355 struct inpcb *
3356 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3357 {
3358 	struct tcpcb *tp;
3359 
3360 	NET_EPOCH_ASSERT();
3361 	INP_WLOCK_ASSERT(inp);
3362 
3363 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3364 	    (inp->inp_flags & INP_DROPPED))
3365 		return (inp);
3366 
3367 	tp = intotcpcb(inp);
3368 	if (tp->t_state != TCPS_SYN_SENT)
3369 		return (inp);
3370 
3371 	if (IS_FASTOPEN(tp->t_flags))
3372 		tcp_fastopen_disable_path(tp);
3373 
3374 	tp = tcp_drop(tp, errno);
3375 	if (tp != NULL)
3376 		return (inp);
3377 	else
3378 		return (NULL);
3379 }
3380 
3381 /*
3382  * When `need fragmentation' ICMP is received, update our idea of the MSS
3383  * based on the new value. Also nudge TCP to send something, since we
3384  * know the packet we just sent was dropped.
3385  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3386  */
3387 static struct inpcb *
3388 tcp_mtudisc_notify(struct inpcb *inp, int error)
3389 {
3390 
3391 	return (tcp_mtudisc(inp, -1));
3392 }
3393 
3394 static struct inpcb *
3395 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3396 {
3397 	struct tcpcb *tp;
3398 	struct socket *so;
3399 
3400 	INP_WLOCK_ASSERT(inp);
3401 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3402 	    (inp->inp_flags & INP_DROPPED))
3403 		return (inp);
3404 
3405 	tp = intotcpcb(inp);
3406 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3407 
3408 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3409 
3410 	so = inp->inp_socket;
3411 	SOCKBUF_LOCK(&so->so_snd);
3412 	/* If the mss is larger than the socket buffer, decrease the mss. */
3413 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3414 		tp->t_maxseg = so->so_snd.sb_hiwat;
3415 	SOCKBUF_UNLOCK(&so->so_snd);
3416 
3417 	TCPSTAT_INC(tcps_mturesent);
3418 	tp->t_rtttime = 0;
3419 	tp->snd_nxt = tp->snd_una;
3420 	tcp_free_sackholes(tp);
3421 	tp->snd_recover = tp->snd_max;
3422 	if (tp->t_flags & TF_SACK_PERMIT)
3423 		EXIT_FASTRECOVERY(tp->t_flags);
3424 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3425 		/*
3426 		 * Conceptually the snd_nxt setting
3427 		 * and freeing sack holes should
3428 		 * be done by the default stacks
3429 		 * own tfb_tcp_mtu_chg().
3430 		 */
3431 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3432 	}
3433 	if (tcp_output(tp) < 0)
3434 		return (NULL);
3435 	else
3436 		return (inp);
3437 }
3438 
3439 #ifdef INET
3440 /*
3441  * Look-up the routing entry to the peer of this inpcb.  If no route
3442  * is found and it cannot be allocated, then return 0.  This routine
3443  * is called by TCP routines that access the rmx structure and by
3444  * tcp_mss_update to get the peer/interface MTU.
3445  */
3446 uint32_t
3447 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3448 {
3449 	struct nhop_object *nh;
3450 	struct ifnet *ifp;
3451 	uint32_t maxmtu = 0;
3452 
3453 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3454 
3455 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3456 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3457 		if (nh == NULL)
3458 			return (0);
3459 
3460 		ifp = nh->nh_ifp;
3461 		maxmtu = nh->nh_mtu;
3462 
3463 		/* Report additional interface capabilities. */
3464 		if (cap != NULL) {
3465 			if (ifp->if_capenable & IFCAP_TSO4 &&
3466 			    ifp->if_hwassist & CSUM_TSO) {
3467 				cap->ifcap |= CSUM_TSO;
3468 				cap->tsomax = ifp->if_hw_tsomax;
3469 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3470 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3471 			}
3472 		}
3473 	}
3474 	return (maxmtu);
3475 }
3476 #endif /* INET */
3477 
3478 #ifdef INET6
3479 uint32_t
3480 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3481 {
3482 	struct nhop_object *nh;
3483 	struct in6_addr dst6;
3484 	uint32_t scopeid;
3485 	struct ifnet *ifp;
3486 	uint32_t maxmtu = 0;
3487 
3488 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3489 
3490 	if (inc->inc_flags & INC_IPV6MINMTU)
3491 		return (IPV6_MMTU);
3492 
3493 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3494 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3495 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3496 		if (nh == NULL)
3497 			return (0);
3498 
3499 		ifp = nh->nh_ifp;
3500 		maxmtu = nh->nh_mtu;
3501 
3502 		/* Report additional interface capabilities. */
3503 		if (cap != NULL) {
3504 			if (ifp->if_capenable & IFCAP_TSO6 &&
3505 			    ifp->if_hwassist & CSUM_TSO) {
3506 				cap->ifcap |= CSUM_TSO;
3507 				cap->tsomax = ifp->if_hw_tsomax;
3508 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3509 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3510 			}
3511 		}
3512 	}
3513 
3514 	return (maxmtu);
3515 }
3516 
3517 /*
3518  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3519  *
3520  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3521  * The right place to do that is ip6_setpktopt() that has just been
3522  * executed.  By the way it just filled ip6po_minmtu for us.
3523  */
3524 void
3525 tcp6_use_min_mtu(struct tcpcb *tp)
3526 {
3527 	struct inpcb *inp = tp->t_inpcb;
3528 
3529 	INP_WLOCK_ASSERT(inp);
3530 	/*
3531 	 * In case of the IPV6_USE_MIN_MTU socket
3532 	 * option, the INC_IPV6MINMTU flag to announce
3533 	 * a corresponding MSS during the initial
3534 	 * handshake.  If the TCP connection is not in
3535 	 * the front states, just reduce the MSS being
3536 	 * used.  This avoids the sending of TCP
3537 	 * segments which will be fragmented at the
3538 	 * IPv6 layer.
3539 	 */
3540 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3541 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3542 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3543 		struct ip6_pktopts *opt;
3544 
3545 		opt = inp->in6p_outputopts;
3546 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3547 		    tp->t_maxseg > TCP6_MSS)
3548 			tp->t_maxseg = TCP6_MSS;
3549 	}
3550 }
3551 #endif /* INET6 */
3552 
3553 /*
3554  * Calculate effective SMSS per RFC5681 definition for a given TCP
3555  * connection at its current state, taking into account SACK and etc.
3556  */
3557 u_int
3558 tcp_maxseg(const struct tcpcb *tp)
3559 {
3560 	u_int optlen;
3561 
3562 	if (tp->t_flags & TF_NOOPT)
3563 		return (tp->t_maxseg);
3564 
3565 	/*
3566 	 * Here we have a simplified code from tcp_addoptions(),
3567 	 * without a proper loop, and having most of paddings hardcoded.
3568 	 * We might make mistakes with padding here in some edge cases,
3569 	 * but this is harmless, since result of tcp_maxseg() is used
3570 	 * only in cwnd and ssthresh estimations.
3571 	 */
3572 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3573 		if (tp->t_flags & TF_RCVD_TSTMP)
3574 			optlen = TCPOLEN_TSTAMP_APPA;
3575 		else
3576 			optlen = 0;
3577 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3578 		if (tp->t_flags & TF_SIGNATURE)
3579 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3580 #endif
3581 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3582 			optlen += TCPOLEN_SACKHDR;
3583 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3584 			optlen = PADTCPOLEN(optlen);
3585 		}
3586 	} else {
3587 		if (tp->t_flags & TF_REQ_TSTMP)
3588 			optlen = TCPOLEN_TSTAMP_APPA;
3589 		else
3590 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3591 		if (tp->t_flags & TF_REQ_SCALE)
3592 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3593 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3594 		if (tp->t_flags & TF_SIGNATURE)
3595 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3596 #endif
3597 		if (tp->t_flags & TF_SACK_PERMIT)
3598 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3599 	}
3600 #undef PAD
3601 	optlen = min(optlen, TCP_MAXOLEN);
3602 	return (tp->t_maxseg - optlen);
3603 }
3604 
3605 
3606 u_int
3607 tcp_fixed_maxseg(const struct tcpcb *tp)
3608 {
3609 	int optlen;
3610 
3611 	if (tp->t_flags & TF_NOOPT)
3612 		return (tp->t_maxseg);
3613 
3614 	/*
3615 	 * Here we have a simplified code from tcp_addoptions(),
3616 	 * without a proper loop, and having most of paddings hardcoded.
3617 	 * We only consider fixed options that we would send every
3618 	 * time I.e. SACK is not considered. This is important
3619 	 * for cc modules to figure out what the modulo of the
3620 	 * cwnd should be.
3621 	 */
3622 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3623 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3624 		if (tp->t_flags & TF_RCVD_TSTMP)
3625 			optlen = TCPOLEN_TSTAMP_APPA;
3626 		else
3627 			optlen = 0;
3628 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3629 		if (tp->t_flags & TF_SIGNATURE)
3630 			optlen += PAD(TCPOLEN_SIGNATURE);
3631 #endif
3632 	} else {
3633 		if (tp->t_flags & TF_REQ_TSTMP)
3634 			optlen = TCPOLEN_TSTAMP_APPA;
3635 		else
3636 			optlen = PAD(TCPOLEN_MAXSEG);
3637 		if (tp->t_flags & TF_REQ_SCALE)
3638 			optlen += PAD(TCPOLEN_WINDOW);
3639 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3640 		if (tp->t_flags & TF_SIGNATURE)
3641 			optlen += PAD(TCPOLEN_SIGNATURE);
3642 #endif
3643 		if (tp->t_flags & TF_SACK_PERMIT)
3644 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3645 	}
3646 #undef PAD
3647 	optlen = min(optlen, TCP_MAXOLEN);
3648 	return (tp->t_maxseg - optlen);
3649 }
3650 
3651 
3652 
3653 static int
3654 sysctl_drop(SYSCTL_HANDLER_ARGS)
3655 {
3656 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3657 	struct sockaddr_storage addrs[2];
3658 	struct inpcb *inp;
3659 	struct tcpcb *tp;
3660 	struct tcptw *tw;
3661 #ifdef INET
3662 	struct sockaddr_in *fin = NULL, *lin = NULL;
3663 #endif
3664 	struct epoch_tracker et;
3665 #ifdef INET6
3666 	struct sockaddr_in6 *fin6, *lin6;
3667 #endif
3668 	int error;
3669 
3670 	inp = NULL;
3671 #ifdef INET6
3672 	fin6 = lin6 = NULL;
3673 #endif
3674 	error = 0;
3675 
3676 	if (req->oldptr != NULL || req->oldlen != 0)
3677 		return (EINVAL);
3678 	if (req->newptr == NULL)
3679 		return (EPERM);
3680 	if (req->newlen < sizeof(addrs))
3681 		return (ENOMEM);
3682 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3683 	if (error)
3684 		return (error);
3685 
3686 	switch (addrs[0].ss_family) {
3687 #ifdef INET6
3688 	case AF_INET6:
3689 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3690 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3691 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3692 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3693 			return (EINVAL);
3694 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3695 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3696 				return (EINVAL);
3697 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3698 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3699 #ifdef INET
3700 			fin = (struct sockaddr_in *)&addrs[0];
3701 			lin = (struct sockaddr_in *)&addrs[1];
3702 #endif
3703 			break;
3704 		}
3705 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3706 		if (error)
3707 			return (error);
3708 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3709 		if (error)
3710 			return (error);
3711 		break;
3712 #endif
3713 #ifdef INET
3714 	case AF_INET:
3715 		fin = (struct sockaddr_in *)&addrs[0];
3716 		lin = (struct sockaddr_in *)&addrs[1];
3717 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3718 		    lin->sin_len != sizeof(struct sockaddr_in))
3719 			return (EINVAL);
3720 		break;
3721 #endif
3722 	default:
3723 		return (EINVAL);
3724 	}
3725 	NET_EPOCH_ENTER(et);
3726 	switch (addrs[0].ss_family) {
3727 #ifdef INET6
3728 	case AF_INET6:
3729 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3730 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3731 		    INPLOOKUP_WLOCKPCB, NULL);
3732 		break;
3733 #endif
3734 #ifdef INET
3735 	case AF_INET:
3736 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3737 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3738 		break;
3739 #endif
3740 	}
3741 	if (inp != NULL) {
3742 		if (inp->inp_flags & INP_TIMEWAIT) {
3743 			/*
3744 			 * XXXRW: There currently exists a state where an
3745 			 * inpcb is present, but its timewait state has been
3746 			 * discarded.  For now, don't allow dropping of this
3747 			 * type of inpcb.
3748 			 */
3749 			tw = intotw(inp);
3750 			if (tw != NULL)
3751 				tcp_twclose(tw, 0);
3752 			else
3753 				INP_WUNLOCK(inp);
3754 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3755 		    !SOLISTENING(inp->inp_socket)) {
3756 			tp = intotcpcb(inp);
3757 			tp = tcp_drop(tp, ECONNABORTED);
3758 			if (tp != NULL)
3759 				INP_WUNLOCK(inp);
3760 		} else
3761 			INP_WUNLOCK(inp);
3762 	} else
3763 		error = ESRCH;
3764 	NET_EPOCH_EXIT(et);
3765 	return (error);
3766 }
3767 
3768 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3769     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3770     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3771     "Drop TCP connection");
3772 
3773 static int
3774 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3775 {
3776 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3777 	    &tcp_ctloutput_set));
3778 }
3779 
3780 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3781     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3782     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3783     "Set socket option for TCP endpoint");
3784 
3785 #ifdef KERN_TLS
3786 static int
3787 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3788 {
3789 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3790 	struct sockaddr_storage addrs[2];
3791 	struct inpcb *inp;
3792 #ifdef INET
3793 	struct sockaddr_in *fin = NULL, *lin = NULL;
3794 #endif
3795 	struct epoch_tracker et;
3796 #ifdef INET6
3797 	struct sockaddr_in6 *fin6, *lin6;
3798 #endif
3799 	int error;
3800 
3801 	inp = NULL;
3802 #ifdef INET6
3803 	fin6 = lin6 = NULL;
3804 #endif
3805 	error = 0;
3806 
3807 	if (req->oldptr != NULL || req->oldlen != 0)
3808 		return (EINVAL);
3809 	if (req->newptr == NULL)
3810 		return (EPERM);
3811 	if (req->newlen < sizeof(addrs))
3812 		return (ENOMEM);
3813 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3814 	if (error)
3815 		return (error);
3816 
3817 	switch (addrs[0].ss_family) {
3818 #ifdef INET6
3819 	case AF_INET6:
3820 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3821 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3822 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3823 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3824 			return (EINVAL);
3825 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3826 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3827 				return (EINVAL);
3828 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3829 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3830 #ifdef INET
3831 			fin = (struct sockaddr_in *)&addrs[0];
3832 			lin = (struct sockaddr_in *)&addrs[1];
3833 #endif
3834 			break;
3835 		}
3836 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3837 		if (error)
3838 			return (error);
3839 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3840 		if (error)
3841 			return (error);
3842 		break;
3843 #endif
3844 #ifdef INET
3845 	case AF_INET:
3846 		fin = (struct sockaddr_in *)&addrs[0];
3847 		lin = (struct sockaddr_in *)&addrs[1];
3848 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3849 		    lin->sin_len != sizeof(struct sockaddr_in))
3850 			return (EINVAL);
3851 		break;
3852 #endif
3853 	default:
3854 		return (EINVAL);
3855 	}
3856 	NET_EPOCH_ENTER(et);
3857 	switch (addrs[0].ss_family) {
3858 #ifdef INET6
3859 	case AF_INET6:
3860 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3861 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3862 		    INPLOOKUP_WLOCKPCB, NULL);
3863 		break;
3864 #endif
3865 #ifdef INET
3866 	case AF_INET:
3867 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3868 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3869 		break;
3870 #endif
3871 	}
3872 	NET_EPOCH_EXIT(et);
3873 	if (inp != NULL) {
3874 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3875 		    inp->inp_socket == NULL) {
3876 			error = ECONNRESET;
3877 			INP_WUNLOCK(inp);
3878 		} else {
3879 			struct socket *so;
3880 
3881 			so = inp->inp_socket;
3882 			soref(so);
3883 			error = ktls_set_tx_mode(so,
3884 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3885 			INP_WUNLOCK(inp);
3886 			sorele(so);
3887 		}
3888 	} else
3889 		error = ESRCH;
3890 	return (error);
3891 }
3892 
3893 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3894     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3895     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3896     "Switch TCP connection to SW TLS");
3897 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3898     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3899     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3900     "Switch TCP connection to ifnet TLS");
3901 #endif
3902 
3903 /*
3904  * Generate a standardized TCP log line for use throughout the
3905  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3906  * allow use in the interrupt context.
3907  *
3908  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3909  * NB: The function may return NULL if memory allocation failed.
3910  *
3911  * Due to header inclusion and ordering limitations the struct ip
3912  * and ip6_hdr pointers have to be passed as void pointers.
3913  */
3914 char *
3915 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3916     const void *ip6hdr)
3917 {
3918 
3919 	/* Is logging enabled? */
3920 	if (V_tcp_log_in_vain == 0)
3921 		return (NULL);
3922 
3923 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3924 }
3925 
3926 char *
3927 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3928     const void *ip6hdr)
3929 {
3930 
3931 	/* Is logging enabled? */
3932 	if (tcp_log_debug == 0)
3933 		return (NULL);
3934 
3935 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3936 }
3937 
3938 static char *
3939 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3940     const void *ip6hdr)
3941 {
3942 	char *s, *sp;
3943 	size_t size;
3944 #ifdef INET
3945 	const struct ip *ip = (const struct ip *)ip4hdr;
3946 #endif
3947 #ifdef INET6
3948 	const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3949 #endif /* INET6 */
3950 
3951 	/*
3952 	 * The log line looks like this:
3953 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3954 	 */
3955 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3956 	    sizeof(PRINT_TH_FLAGS) + 1 +
3957 #ifdef INET6
3958 	    2 * INET6_ADDRSTRLEN;
3959 #else
3960 	    2 * INET_ADDRSTRLEN;
3961 #endif /* INET6 */
3962 
3963 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3964 	if (s == NULL)
3965 		return (NULL);
3966 
3967 	strcat(s, "TCP: [");
3968 	sp = s + strlen(s);
3969 
3970 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3971 		inet_ntoa_r(inc->inc_faddr, sp);
3972 		sp = s + strlen(s);
3973 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3974 		sp = s + strlen(s);
3975 		inet_ntoa_r(inc->inc_laddr, sp);
3976 		sp = s + strlen(s);
3977 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3978 #ifdef INET6
3979 	} else if (inc) {
3980 		ip6_sprintf(sp, &inc->inc6_faddr);
3981 		sp = s + strlen(s);
3982 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3983 		sp = s + strlen(s);
3984 		ip6_sprintf(sp, &inc->inc6_laddr);
3985 		sp = s + strlen(s);
3986 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3987 	} else if (ip6 && th) {
3988 		ip6_sprintf(sp, &ip6->ip6_src);
3989 		sp = s + strlen(s);
3990 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3991 		sp = s + strlen(s);
3992 		ip6_sprintf(sp, &ip6->ip6_dst);
3993 		sp = s + strlen(s);
3994 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3995 #endif /* INET6 */
3996 #ifdef INET
3997 	} else if (ip && th) {
3998 		inet_ntoa_r(ip->ip_src, sp);
3999 		sp = s + strlen(s);
4000 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4001 		sp = s + strlen(s);
4002 		inet_ntoa_r(ip->ip_dst, sp);
4003 		sp = s + strlen(s);
4004 		sprintf(sp, "]:%i", ntohs(th->th_dport));
4005 #endif /* INET */
4006 	} else {
4007 		free(s, M_TCPLOG);
4008 		return (NULL);
4009 	}
4010 	sp = s + strlen(s);
4011 	if (th)
4012 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
4013 	if (*(s + size - 1) != '\0')
4014 		panic("%s: string too long", __func__);
4015 	return (s);
4016 }
4017 
4018 /*
4019  * A subroutine which makes it easy to track TCP state changes with DTrace.
4020  * This function shouldn't be called for t_state initializations that don't
4021  * correspond to actual TCP state transitions.
4022  */
4023 void
4024 tcp_state_change(struct tcpcb *tp, int newstate)
4025 {
4026 #if defined(KDTRACE_HOOKS)
4027 	int pstate = tp->t_state;
4028 #endif
4029 
4030 	TCPSTATES_DEC(tp->t_state);
4031 	TCPSTATES_INC(newstate);
4032 	tp->t_state = newstate;
4033 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4034 }
4035 
4036 /*
4037  * Create an external-format (``xtcpcb'') structure using the information in
4038  * the kernel-format tcpcb structure pointed to by tp.  This is done to
4039  * reduce the spew of irrelevant information over this interface, to isolate
4040  * user code from changes in the kernel structure, and potentially to provide
4041  * information-hiding if we decide that some of this information should be
4042  * hidden from users.
4043  */
4044 void
4045 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4046 {
4047 	struct tcpcb *tp = intotcpcb(inp);
4048 	struct tcptw *tw = intotw(inp);
4049 	sbintime_t now;
4050 
4051 	bzero(xt, sizeof(*xt));
4052 	if (inp->inp_flags & INP_TIMEWAIT) {
4053 		xt->t_state = TCPS_TIME_WAIT;
4054 		xt->xt_encaps_port = tw->t_port;
4055 	} else {
4056 		xt->t_state = tp->t_state;
4057 		xt->t_logstate = tp->t_logstate;
4058 		xt->t_flags = tp->t_flags;
4059 		xt->t_sndzerowin = tp->t_sndzerowin;
4060 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4061 		xt->t_rcvoopack = tp->t_rcvoopack;
4062 		xt->t_rcv_wnd = tp->rcv_wnd;
4063 		xt->t_snd_wnd = tp->snd_wnd;
4064 		xt->t_snd_cwnd = tp->snd_cwnd;
4065 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4066 		xt->t_dsack_bytes = tp->t_dsack_bytes;
4067 		xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4068 		xt->t_dsack_pack = tp->t_dsack_pack;
4069 		xt->t_maxseg = tp->t_maxseg;
4070 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4071 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4072 
4073 		now = getsbinuptime();
4074 #define	COPYTIMER(ttt)	do {						\
4075 		if (callout_active(&tp->t_timers->ttt))			\
4076 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4077 			    SBT_1MS;					\
4078 		else							\
4079 			xt->ttt = 0;					\
4080 } while (0)
4081 		COPYTIMER(tt_delack);
4082 		COPYTIMER(tt_rexmt);
4083 		COPYTIMER(tt_persist);
4084 		COPYTIMER(tt_keep);
4085 		COPYTIMER(tt_2msl);
4086 #undef COPYTIMER
4087 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4088 
4089 		xt->xt_encaps_port = tp->t_port;
4090 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4091 		    TCP_FUNCTION_NAME_LEN_MAX);
4092 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4093 		    TCP_CA_NAME_MAX);
4094 #ifdef TCP_BLACKBOX
4095 		(void)tcp_log_get_id(tp, xt->xt_logid);
4096 #endif
4097 	}
4098 
4099 	xt->xt_len = sizeof(struct xtcpcb);
4100 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4101 	if (inp->inp_socket == NULL)
4102 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4103 }
4104 
4105 void
4106 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4107 {
4108 	uint32_t bit, i;
4109 
4110 	if ((tp == NULL) ||
4111 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4112 	    (status == 0)) {
4113 		/* Invalid */
4114 		return;
4115 	}
4116 	if (status > (sizeof(uint32_t) * 8)) {
4117 		/* Should this be a KASSERT? */
4118 		return;
4119 	}
4120 	bit = 1U << (status - 1);
4121 	if (bit & tp->t_end_info_status) {
4122 		/* already logged */
4123 		return;
4124 	}
4125 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4126 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4127 			tp->t_end_info_bytes[i] = status;
4128 			tp->t_end_info_status |= bit;
4129 			break;
4130 		}
4131 	}
4132 }
4133 
4134 int
4135 tcp_can_enable_pacing(void)
4136 {
4137 
4138 	if ((tcp_pacing_limit == -1) ||
4139 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4140 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4141 		shadow_num_connections = number_of_tcp_connections_pacing;
4142 		return (1);
4143 	} else {
4144 		return (0);
4145 	}
4146 }
4147 
4148 static uint8_t tcp_pacing_warning = 0;
4149 
4150 void
4151 tcp_decrement_paced_conn(void)
4152 {
4153 	uint32_t ret;
4154 
4155 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4156 	shadow_num_connections = number_of_tcp_connections_pacing;
4157 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4158 	if (ret == 0) {
4159 		if (tcp_pacing_limit != -1) {
4160 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4161 			tcp_pacing_limit = 0;
4162 		} else if (tcp_pacing_warning == 0) {
4163 			printf("Warning pacing count is invalid, invalid decrement\n");
4164 			tcp_pacing_warning = 1;
4165 		}
4166 	}
4167 }
4168