xref: /freebsd/sys/netinet/tcp_subr.c (revision 43a5ec4eb41567cc92586503212743d89686d78f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/route/nhop.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
93 #ifdef INET6
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
101 #endif
102 
103 #include <netinet/tcp.h>
104 #ifdef INVARIANTS
105 #define TCPSTATES
106 #endif
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_seq.h>
109 #include <netinet/tcp_timer.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet/tcp_log_buf.h>
112 #include <netinet/tcp_syncache.h>
113 #include <netinet/tcp_hpts.h>
114 #include <netinet/cc/cc.h>
115 #ifdef INET6
116 #include <netinet6/tcp6_var.h>
117 #endif
118 #include <netinet/tcpip.h>
119 #include <netinet/tcp_fastopen.h>
120 #ifdef TCPPCAP
121 #include <netinet/tcp_pcap.h>
122 #endif
123 #ifdef TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126 #ifdef INET6
127 #include <netinet6/ip6protosw.h>
128 #endif
129 #ifdef TCP_OFFLOAD
130 #include <netinet/tcp_offload.h>
131 #endif
132 #include <netinet/udp.h>
133 #include <netinet/udp_var.h>
134 
135 #include <netipsec/ipsec_support.h>
136 
137 #include <machine/in_cksum.h>
138 #include <crypto/siphash/siphash.h>
139 
140 #include <security/mac/mac_framework.h>
141 
142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
143 #ifdef INET6
144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
145 #endif
146 
147 #ifdef NETFLIX_EXP_DETECTION
148 /*  Sack attack detection thresholds and such */
149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
150     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
151     "Sack Attack detection thresholds");
152 int32_t tcp_force_detection = 0;
153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
154     CTLFLAG_RW,
155     &tcp_force_detection, 0,
156     "Do we force detection even if the INP has it off?");
157 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
159     CTLFLAG_RW,
160     &tcp_sack_to_ack_thresh, 700,
161     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
162 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
164     CTLFLAG_RW,
165     &tcp_sack_to_move_thresh, 600,
166     "Percentage of sack moves we must see above (10.1 percent is 101)");
167 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
169     CTLFLAG_RW,
170     &tcp_restoral_thresh, 550,
171     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
172 int32_t tcp_sad_decay_val = 800;
173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
174     CTLFLAG_RW,
175     &tcp_sad_decay_val, 800,
176     "The decay percentage (10.1 percent equals 101 )");
177 int32_t tcp_map_minimum = 500;
178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
179     CTLFLAG_RW,
180     &tcp_map_minimum, 500,
181     "Number of Map enteries before we start detection");
182 int32_t tcp_attack_on_turns_on_logging = 0;
183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
184     CTLFLAG_RW,
185     &tcp_attack_on_turns_on_logging, 0,
186    "When we have a positive hit on attack, do we turn on logging?");
187 int32_t tcp_sad_pacing_interval = 2000;
188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
189     CTLFLAG_RW,
190     &tcp_sad_pacing_interval, 2000,
191     "What is the minimum pacing interval for a classified attacker?");
192 
193 int32_t tcp_sad_low_pps = 100;
194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
195     CTLFLAG_RW,
196     &tcp_sad_low_pps, 100,
197     "What is the input pps that below which we do not decay?");
198 #endif
199 uint32_t tcp_ack_war_time_window = 1000;
200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
201     CTLFLAG_RW,
202     &tcp_ack_war_time_window, 1000,
203    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
204 uint32_t tcp_ack_war_cnt = 5;
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
206     CTLFLAG_RW,
207     &tcp_ack_war_cnt, 5,
208    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
209 
210 struct rwlock tcp_function_lock;
211 
212 static int
213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
214 {
215 	int error, new;
216 
217 	new = V_tcp_mssdflt;
218 	error = sysctl_handle_int(oidp, &new, 0, req);
219 	if (error == 0 && req->newptr) {
220 		if (new < TCP_MINMSS)
221 			error = EINVAL;
222 		else
223 			V_tcp_mssdflt = new;
224 	}
225 	return (error);
226 }
227 
228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
229     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
230     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
231     "Default TCP Maximum Segment Size");
232 
233 #ifdef INET6
234 static int
235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
236 {
237 	int error, new;
238 
239 	new = V_tcp_v6mssdflt;
240 	error = sysctl_handle_int(oidp, &new, 0, req);
241 	if (error == 0 && req->newptr) {
242 		if (new < TCP_MINMSS)
243 			error = EINVAL;
244 		else
245 			V_tcp_v6mssdflt = new;
246 	}
247 	return (error);
248 }
249 
250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
251     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
252     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
253    "Default TCP Maximum Segment Size for IPv6");
254 #endif /* INET6 */
255 
256 /*
257  * Minimum MSS we accept and use. This prevents DoS attacks where
258  * we are forced to a ridiculous low MSS like 20 and send hundreds
259  * of packets instead of one. The effect scales with the available
260  * bandwidth and quickly saturates the CPU and network interface
261  * with packet generation and sending. Set to zero to disable MINMSS
262  * checking. This setting prevents us from sending too small packets.
263  */
264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
266      &VNET_NAME(tcp_minmss), 0,
267     "Minimum TCP Maximum Segment Size");
268 
269 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
271     &VNET_NAME(tcp_do_rfc1323), 0,
272     "Enable rfc1323 (high performance TCP) extensions");
273 
274 /*
275  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
276  * Some stacks negotiate TS, but never send them after connection setup. Some
277  * stacks negotiate TS, but don't send them when sending keep-alive segments.
278  * These include modern widely deployed TCP stacks.
279  * Therefore tolerating violations for now...
280  */
281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
283     &VNET_NAME(tcp_tolerate_missing_ts), 0,
284     "Tolerate missing TCP timestamps");
285 
286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
288     &VNET_NAME(tcp_ts_offset_per_conn), 0,
289     "Initialize TCP timestamps per connection instead of per host pair");
290 
291 /* How many connections are pacing */
292 static volatile uint32_t number_of_tcp_connections_pacing = 0;
293 static uint32_t shadow_num_connections = 0;
294 
295 static int tcp_pacing_limit = 10000;
296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
297     &tcp_pacing_limit, 1000,
298     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
299 
300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
301     &shadow_num_connections, 0, "Number of TCP connections being paced");
302 
303 static int	tcp_log_debug = 0;
304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
305     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
306 
307 static int	tcp_tcbhashsize;
308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
309     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
310 
311 static int	do_tcpdrain = 1;
312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
313     "Enable tcp_drain routine for extra help when low on mbufs");
314 
315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
316     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
317 
318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
319 #define	V_icmp_may_rst			VNET(icmp_may_rst)
320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
321     &VNET_NAME(icmp_may_rst), 0,
322     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
323 
324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
325 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
327     &VNET_NAME(tcp_isn_reseed_interval), 0,
328     "Seconds between reseeding of ISN secret");
329 
330 static int	tcp_soreceive_stream;
331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
332     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
333 
334 VNET_DEFINE(uma_zone_t, sack_hole_zone);
335 #define	V_sack_hole_zone		VNET(sack_hole_zone)
336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
337 static int
338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
339 {
340 	int error;
341 	uint32_t new;
342 
343 	new = V_tcp_map_entries_limit;
344 	error = sysctl_handle_int(oidp, &new, 0, req);
345 	if (error == 0 && req->newptr) {
346 		/* only allow "0" and value > minimum */
347 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
348 			error = EINVAL;
349 		else
350 			V_tcp_map_entries_limit = new;
351 	}
352 	return (error);
353 }
354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
355     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
356     &VNET_NAME(tcp_map_entries_limit), 0,
357     &sysctl_net_inet_tcp_map_limit_check, "IU",
358     "Total sendmap entries limit");
359 
360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
362      &VNET_NAME(tcp_map_split_limit), 0,
363     "Total sendmap split entries limit");
364 
365 #ifdef TCP_HHOOK
366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
367 #endif
368 
369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
371 #define	V_ts_offset_secret	VNET(ts_offset_secret)
372 
373 static int	tcp_default_fb_init(struct tcpcb *tp);
374 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
375 static int	tcp_default_handoff_ok(struct tcpcb *tp);
376 static struct inpcb *tcp_notify(struct inpcb *, int);
377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
378 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
379 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
380 		    void *ip4hdr, const void *ip6hdr);
381 
382 static struct tcp_function_block tcp_def_funcblk = {
383 	.tfb_tcp_block_name = "freebsd",
384 	.tfb_tcp_output = tcp_default_output,
385 	.tfb_tcp_do_segment = tcp_do_segment,
386 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
387 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
388 	.tfb_tcp_fb_init = tcp_default_fb_init,
389 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
390 };
391 
392 static int tcp_fb_cnt = 0;
393 struct tcp_funchead t_functions;
394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
395 
396 void
397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
398 {
399 	TCPSTAT_INC(tcps_dsack_count);
400 	tp->t_dsack_pack++;
401 	if (tlp == 0) {
402 		if (SEQ_GT(end, start)) {
403 			tp->t_dsack_bytes += (end - start);
404 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
405 		} else {
406 			tp->t_dsack_tlp_bytes += (start - end);
407 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
408 		}
409 	} else {
410 		if (SEQ_GT(end, start)) {
411 			tp->t_dsack_bytes += (end - start);
412 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
413 		} else {
414 			tp->t_dsack_tlp_bytes += (start - end);
415 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
416 		}
417 	}
418 }
419 
420 static struct tcp_function_block *
421 find_tcp_functions_locked(struct tcp_function_set *fs)
422 {
423 	struct tcp_function *f;
424 	struct tcp_function_block *blk=NULL;
425 
426 	TAILQ_FOREACH(f, &t_functions, tf_next) {
427 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
428 			blk = f->tf_fb;
429 			break;
430 		}
431 	}
432 	return(blk);
433 }
434 
435 static struct tcp_function_block *
436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
437 {
438 	struct tcp_function_block *rblk=NULL;
439 	struct tcp_function *f;
440 
441 	TAILQ_FOREACH(f, &t_functions, tf_next) {
442 		if (f->tf_fb == blk) {
443 			rblk = blk;
444 			if (s) {
445 				*s = f;
446 			}
447 			break;
448 		}
449 	}
450 	return (rblk);
451 }
452 
453 struct tcp_function_block *
454 find_and_ref_tcp_functions(struct tcp_function_set *fs)
455 {
456 	struct tcp_function_block *blk;
457 
458 	rw_rlock(&tcp_function_lock);
459 	blk = find_tcp_functions_locked(fs);
460 	if (blk)
461 		refcount_acquire(&blk->tfb_refcnt);
462 	rw_runlock(&tcp_function_lock);
463 	return(blk);
464 }
465 
466 struct tcp_function_block *
467 find_and_ref_tcp_fb(struct tcp_function_block *blk)
468 {
469 	struct tcp_function_block *rblk;
470 
471 	rw_rlock(&tcp_function_lock);
472 	rblk = find_tcp_fb_locked(blk, NULL);
473 	if (rblk)
474 		refcount_acquire(&rblk->tfb_refcnt);
475 	rw_runlock(&tcp_function_lock);
476 	return(rblk);
477 }
478 
479 /* Find a matching alias for the given tcp_function_block. */
480 int
481 find_tcp_function_alias(struct tcp_function_block *blk,
482     struct tcp_function_set *fs)
483 {
484 	struct tcp_function *f;
485 	int found;
486 
487 	found = 0;
488 	rw_rlock(&tcp_function_lock);
489 	TAILQ_FOREACH(f, &t_functions, tf_next) {
490 		if ((f->tf_fb == blk) &&
491 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
492 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
493 			/* Matching function block with different name. */
494 			strncpy(fs->function_set_name, f->tf_name,
495 			    TCP_FUNCTION_NAME_LEN_MAX);
496 			found = 1;
497 			break;
498 		}
499 	}
500 	/* Null terminate the string appropriately. */
501 	if (found) {
502 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
503 	} else {
504 		fs->function_set_name[0] = '\0';
505 	}
506 	rw_runlock(&tcp_function_lock);
507 	return (found);
508 }
509 
510 static struct tcp_function_block *
511 find_and_ref_tcp_default_fb(void)
512 {
513 	struct tcp_function_block *rblk;
514 
515 	rw_rlock(&tcp_function_lock);
516 	rblk = tcp_func_set_ptr;
517 	refcount_acquire(&rblk->tfb_refcnt);
518 	rw_runlock(&tcp_function_lock);
519 	return (rblk);
520 }
521 
522 void
523 tcp_switch_back_to_default(struct tcpcb *tp)
524 {
525 	struct tcp_function_block *tfb;
526 
527 	KASSERT(tp->t_fb != &tcp_def_funcblk,
528 	    ("%s: called by the built-in default stack", __func__));
529 
530 	/*
531 	 * Release the old stack. This function will either find a new one
532 	 * or panic.
533 	 */
534 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
535 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
536 	refcount_release(&tp->t_fb->tfb_refcnt);
537 
538 	/*
539 	 * Now, we'll find a new function block to use.
540 	 * Start by trying the current user-selected
541 	 * default, unless this stack is the user-selected
542 	 * default.
543 	 */
544 	tfb = find_and_ref_tcp_default_fb();
545 	if (tfb == tp->t_fb) {
546 		refcount_release(&tfb->tfb_refcnt);
547 		tfb = NULL;
548 	}
549 	/* Does the stack accept this connection? */
550 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 		refcount_release(&tfb->tfb_refcnt);
553 		tfb = NULL;
554 	}
555 	/* Try to use that stack. */
556 	if (tfb != NULL) {
557 		/* Initialize the new stack. If it succeeds, we are done. */
558 		tp->t_fb = tfb;
559 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
560 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
561 			return;
562 
563 		/*
564 		 * Initialization failed. Release the reference count on
565 		 * the stack.
566 		 */
567 		refcount_release(&tfb->tfb_refcnt);
568 	}
569 
570 	/*
571 	 * If that wasn't feasible, use the built-in default
572 	 * stack which is not allowed to reject anyone.
573 	 */
574 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
575 	if (tfb == NULL) {
576 		/* there always should be a default */
577 		panic("Can't refer to tcp_def_funcblk");
578 	}
579 	if (tfb->tfb_tcp_handoff_ok != NULL) {
580 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
581 			/* The default stack cannot say no */
582 			panic("Default stack rejects a new session?");
583 		}
584 	}
585 	tp->t_fb = tfb;
586 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
587 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
588 		/* The default stack cannot fail */
589 		panic("Default stack initialization failed");
590 	}
591 }
592 
593 static void
594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
595     const struct sockaddr *sa, void *ctx)
596 {
597 	struct ip *iph;
598 #ifdef INET6
599 	struct ip6_hdr *ip6;
600 #endif
601 	struct udphdr *uh;
602 	struct tcphdr *th;
603 	int thlen;
604 	uint16_t port;
605 
606 	TCPSTAT_INC(tcps_tunneled_pkts);
607 	if ((m->m_flags & M_PKTHDR) == 0) {
608 		/* Can't handle one that is not a pkt hdr */
609 		TCPSTAT_INC(tcps_tunneled_errs);
610 		goto out;
611 	}
612 	thlen = sizeof(struct tcphdr);
613 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
614 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
615 		TCPSTAT_INC(tcps_tunneled_errs);
616 		goto out;
617 	}
618 	iph = mtod(m, struct ip *);
619 	uh = (struct udphdr *)((caddr_t)iph + off);
620 	th = (struct tcphdr *)(uh + 1);
621 	thlen = th->th_off << 2;
622 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
623 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
624 		if (m == NULL) {
625 			TCPSTAT_INC(tcps_tunneled_errs);
626 			goto out;
627 		} else {
628 			iph = mtod(m, struct ip *);
629 			uh = (struct udphdr *)((caddr_t)iph + off);
630 			th = (struct tcphdr *)(uh + 1);
631 		}
632 	}
633 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
634 	bcopy(th, uh, m->m_len - off);
635 	m->m_len -= sizeof(struct udphdr);
636 	m->m_pkthdr.len -= sizeof(struct udphdr);
637 	/*
638 	 * We use the same algorithm for
639 	 * both UDP and TCP for c-sum. So
640 	 * the code in tcp_input will skip
641 	 * the checksum. So we do nothing
642 	 * with the flag (m->m_pkthdr.csum_flags).
643 	 */
644 	switch (iph->ip_v) {
645 #ifdef INET
646 	case IPVERSION:
647 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
648 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
649 		break;
650 #endif
651 #ifdef INET6
652 	case IPV6_VERSION >> 4:
653 		ip6 = mtod(m, struct ip6_hdr *);
654 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
655 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
656 		break;
657 #endif
658 	default:
659 		goto out;
660 		break;
661 	}
662 	return;
663 out:
664 	m_freem(m);
665 }
666 
667 static int
668 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
669 {
670 	int error=ENOENT;
671 	struct tcp_function_set fs;
672 	struct tcp_function_block *blk;
673 
674 	memset(&fs, 0, sizeof(fs));
675 	rw_rlock(&tcp_function_lock);
676 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
677 	if (blk) {
678 		/* Found him */
679 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
680 		fs.pcbcnt = blk->tfb_refcnt;
681 	}
682 	rw_runlock(&tcp_function_lock);
683 	error = sysctl_handle_string(oidp, fs.function_set_name,
684 				     sizeof(fs.function_set_name), req);
685 
686 	/* Check for error or no change */
687 	if (error != 0 || req->newptr == NULL)
688 		return(error);
689 
690 	rw_wlock(&tcp_function_lock);
691 	blk = find_tcp_functions_locked(&fs);
692 	if ((blk == NULL) ||
693 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
694 		error = ENOENT;
695 		goto done;
696 	}
697 	tcp_func_set_ptr = blk;
698 done:
699 	rw_wunlock(&tcp_function_lock);
700 	return (error);
701 }
702 
703 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
704     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
705     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
706     "Set/get the default TCP functions");
707 
708 static int
709 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
710 {
711 	int error, cnt, linesz;
712 	struct tcp_function *f;
713 	char *buffer, *cp;
714 	size_t bufsz, outsz;
715 	bool alias;
716 
717 	cnt = 0;
718 	rw_rlock(&tcp_function_lock);
719 	TAILQ_FOREACH(f, &t_functions, tf_next) {
720 		cnt++;
721 	}
722 	rw_runlock(&tcp_function_lock);
723 
724 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
725 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
726 
727 	error = 0;
728 	cp = buffer;
729 
730 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
731 	    "Alias", "PCB count");
732 	cp += linesz;
733 	bufsz -= linesz;
734 	outsz = linesz;
735 
736 	rw_rlock(&tcp_function_lock);
737 	TAILQ_FOREACH(f, &t_functions, tf_next) {
738 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
739 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
740 		    f->tf_fb->tfb_tcp_block_name,
741 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
742 		    alias ? f->tf_name : "-",
743 		    f->tf_fb->tfb_refcnt);
744 		if (linesz >= bufsz) {
745 			error = EOVERFLOW;
746 			break;
747 		}
748 		cp += linesz;
749 		bufsz -= linesz;
750 		outsz += linesz;
751 	}
752 	rw_runlock(&tcp_function_lock);
753 	if (error == 0)
754 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
755 	free(buffer, M_TEMP);
756 	return (error);
757 }
758 
759 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
760     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
761     NULL, 0, sysctl_net_inet_list_available, "A",
762     "list available TCP Function sets");
763 
764 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
765 
766 #ifdef INET
767 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
768 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
769 #endif
770 #ifdef INET6
771 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
772 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
773 #endif
774 
775 static void
776 tcp_over_udp_stop(void)
777 {
778 	/*
779 	 * This function assumes sysctl caller holds inp_rinfo_lock()
780 	 * for writing!
781 	 */
782 #ifdef INET
783 	if (V_udp4_tun_socket != NULL) {
784 		soclose(V_udp4_tun_socket);
785 		V_udp4_tun_socket = NULL;
786 	}
787 #endif
788 #ifdef INET6
789 	if (V_udp6_tun_socket != NULL) {
790 		soclose(V_udp6_tun_socket);
791 		V_udp6_tun_socket = NULL;
792 	}
793 #endif
794 }
795 
796 static int
797 tcp_over_udp_start(void)
798 {
799 	uint16_t port;
800 	int ret;
801 #ifdef INET
802 	struct sockaddr_in sin;
803 #endif
804 #ifdef INET6
805 	struct sockaddr_in6 sin6;
806 #endif
807 	/*
808 	 * This function assumes sysctl caller holds inp_info_rlock()
809 	 * for writing!
810 	 */
811 	port = V_tcp_udp_tunneling_port;
812 	if (ntohs(port) == 0) {
813 		/* Must have a port set */
814 		return (EINVAL);
815 	}
816 #ifdef INET
817 	if (V_udp4_tun_socket != NULL) {
818 		/* Already running -- must stop first */
819 		return (EALREADY);
820 	}
821 #endif
822 #ifdef INET6
823 	if (V_udp6_tun_socket != NULL) {
824 		/* Already running -- must stop first */
825 		return (EALREADY);
826 	}
827 #endif
828 #ifdef INET
829 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
830 	    SOCK_DGRAM, IPPROTO_UDP,
831 	    curthread->td_ucred, curthread))) {
832 		tcp_over_udp_stop();
833 		return (ret);
834 	}
835 	/* Call the special UDP hook. */
836 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
837 	    tcp_recv_udp_tunneled_packet,
838 	    tcp_ctlinput_viaudp,
839 	    NULL))) {
840 		tcp_over_udp_stop();
841 		return (ret);
842 	}
843 	/* Ok, we have a socket, bind it to the port. */
844 	memset(&sin, 0, sizeof(struct sockaddr_in));
845 	sin.sin_len = sizeof(struct sockaddr_in);
846 	sin.sin_family = AF_INET;
847 	sin.sin_port = htons(port);
848 	if ((ret = sobind(V_udp4_tun_socket,
849 	    (struct sockaddr *)&sin, curthread))) {
850 		tcp_over_udp_stop();
851 		return (ret);
852 	}
853 #endif
854 #ifdef INET6
855 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
856 	    SOCK_DGRAM, IPPROTO_UDP,
857 	    curthread->td_ucred, curthread))) {
858 		tcp_over_udp_stop();
859 		return (ret);
860 	}
861 	/* Call the special UDP hook. */
862 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
863 	    tcp_recv_udp_tunneled_packet,
864 	    tcp6_ctlinput_viaudp,
865 	    NULL))) {
866 		tcp_over_udp_stop();
867 		return (ret);
868 	}
869 	/* Ok, we have a socket, bind it to the port. */
870 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
871 	sin6.sin6_len = sizeof(struct sockaddr_in6);
872 	sin6.sin6_family = AF_INET6;
873 	sin6.sin6_port = htons(port);
874 	if ((ret = sobind(V_udp6_tun_socket,
875 	    (struct sockaddr *)&sin6, curthread))) {
876 		tcp_over_udp_stop();
877 		return (ret);
878 	}
879 #endif
880 	return (0);
881 }
882 
883 static int
884 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
885 {
886 	int error;
887 	uint32_t old, new;
888 
889 	old = V_tcp_udp_tunneling_port;
890 	new = old;
891 	error = sysctl_handle_int(oidp, &new, 0, req);
892 	if ((error == 0) &&
893 	    (req->newptr != NULL)) {
894 		if ((new < TCP_TUNNELING_PORT_MIN) ||
895 		    (new > TCP_TUNNELING_PORT_MAX)) {
896 			error = EINVAL;
897 		} else {
898 			V_tcp_udp_tunneling_port = new;
899 			if (old != 0) {
900 				tcp_over_udp_stop();
901 			}
902 			if (new != 0) {
903 				error = tcp_over_udp_start();
904 			}
905 		}
906 	}
907 	return (error);
908 }
909 
910 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
911     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
912     &VNET_NAME(tcp_udp_tunneling_port),
913     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
914     "Tunneling port for tcp over udp");
915 
916 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
917 
918 static int
919 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
920 {
921 	int error, new;
922 
923 	new = V_tcp_udp_tunneling_overhead;
924 	error = sysctl_handle_int(oidp, &new, 0, req);
925 	if (error == 0 && req->newptr) {
926 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
927 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
928 			error = EINVAL;
929 		else
930 			V_tcp_udp_tunneling_overhead = new;
931 	}
932 	return (error);
933 }
934 
935 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
936     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
937     &VNET_NAME(tcp_udp_tunneling_overhead),
938     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
939     "MSS reduction when using tcp over udp");
940 
941 /*
942  * Exports one (struct tcp_function_info) for each alias/name.
943  */
944 static int
945 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
946 {
947 	int cnt, error;
948 	struct tcp_function *f;
949 	struct tcp_function_info tfi;
950 
951 	/*
952 	 * We don't allow writes.
953 	 */
954 	if (req->newptr != NULL)
955 		return (EINVAL);
956 
957 	/*
958 	 * Wire the old buffer so we can directly copy the functions to
959 	 * user space without dropping the lock.
960 	 */
961 	if (req->oldptr != NULL) {
962 		error = sysctl_wire_old_buffer(req, 0);
963 		if (error)
964 			return (error);
965 	}
966 
967 	/*
968 	 * Walk the list and copy out matching entries. If INVARIANTS
969 	 * is compiled in, also walk the list to verify the length of
970 	 * the list matches what we have recorded.
971 	 */
972 	rw_rlock(&tcp_function_lock);
973 
974 	cnt = 0;
975 #ifndef INVARIANTS
976 	if (req->oldptr == NULL) {
977 		cnt = tcp_fb_cnt;
978 		goto skip_loop;
979 	}
980 #endif
981 	TAILQ_FOREACH(f, &t_functions, tf_next) {
982 #ifdef INVARIANTS
983 		cnt++;
984 #endif
985 		if (req->oldptr != NULL) {
986 			bzero(&tfi, sizeof(tfi));
987 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
988 			tfi.tfi_id = f->tf_fb->tfb_id;
989 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
990 			    sizeof(tfi.tfi_alias));
991 			(void)strlcpy(tfi.tfi_name,
992 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
993 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
994 			/*
995 			 * Don't stop on error, as that is the
996 			 * mechanism we use to accumulate length
997 			 * information if the buffer was too short.
998 			 */
999 		}
1000 	}
1001 	KASSERT(cnt == tcp_fb_cnt,
1002 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1003 #ifndef INVARIANTS
1004 skip_loop:
1005 #endif
1006 	rw_runlock(&tcp_function_lock);
1007 	if (req->oldptr == NULL)
1008 		error = SYSCTL_OUT(req, NULL,
1009 		    (cnt + 1) * sizeof(struct tcp_function_info));
1010 
1011 	return (error);
1012 }
1013 
1014 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1015 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1016 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1017 	    "List TCP function block name-to-ID mappings");
1018 
1019 /*
1020  * tfb_tcp_handoff_ok() function for the default stack.
1021  * Note that we'll basically try to take all comers.
1022  */
1023 static int
1024 tcp_default_handoff_ok(struct tcpcb *tp)
1025 {
1026 
1027 	return (0);
1028 }
1029 
1030 /*
1031  * tfb_tcp_fb_init() function for the default stack.
1032  *
1033  * This handles making sure we have appropriate timers set if you are
1034  * transitioning a socket that has some amount of setup done.
1035  *
1036  * The init() fuction from the default can *never* return non-zero i.e.
1037  * it is required to always succeed since it is the stack of last resort!
1038  */
1039 static int
1040 tcp_default_fb_init(struct tcpcb *tp)
1041 {
1042 
1043 	struct socket *so;
1044 
1045 	INP_WLOCK_ASSERT(tp->t_inpcb);
1046 
1047 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1048 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1049 	    tp->t_state));
1050 
1051 	/*
1052 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1053 	 * know what to do for unexpected states (which includes TIME_WAIT).
1054 	 */
1055 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1056 		return (0);
1057 
1058 	/*
1059 	 * Make sure some kind of transmission timer is set if there is
1060 	 * outstanding data.
1061 	 */
1062 	so = tp->t_inpcb->inp_socket;
1063 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1064 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1065 	    tcp_timer_active(tp, TT_PERSIST))) {
1066 		/*
1067 		 * If the session has established and it looks like it should
1068 		 * be in the persist state, set the persist timer. Otherwise,
1069 		 * set the retransmit timer.
1070 		 */
1071 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1072 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1073 		    (int32_t)sbavail(&so->so_snd))
1074 			tcp_setpersist(tp);
1075 		else
1076 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1077 	}
1078 
1079 	/* All non-embryonic sessions get a keepalive timer. */
1080 	if (!tcp_timer_active(tp, TT_KEEP))
1081 		tcp_timer_activate(tp, TT_KEEP,
1082 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1083 		    TP_KEEPINIT(tp));
1084 
1085 	/*
1086 	 * Make sure critical variables are initialized
1087 	 * if transitioning while in Recovery.
1088 	 */
1089 	if IN_FASTRECOVERY(tp->t_flags) {
1090 		if (tp->sackhint.recover_fs == 0)
1091 			tp->sackhint.recover_fs = max(1,
1092 			    tp->snd_nxt - tp->snd_una);
1093 	}
1094 
1095 	return (0);
1096 }
1097 
1098 /*
1099  * tfb_tcp_fb_fini() function for the default stack.
1100  *
1101  * This changes state as necessary (or prudent) to prepare for another stack
1102  * to assume responsibility for the connection.
1103  */
1104 static void
1105 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1106 {
1107 
1108 	INP_WLOCK_ASSERT(tp->t_inpcb);
1109 	return;
1110 }
1111 
1112 /*
1113  * Target size of TCP PCB hash tables. Must be a power of two.
1114  *
1115  * Note that this can be overridden by the kernel environment
1116  * variable net.inet.tcp.tcbhashsize
1117  */
1118 #ifndef TCBHASHSIZE
1119 #define TCBHASHSIZE	0
1120 #endif
1121 
1122 /*
1123  * XXX
1124  * Callouts should be moved into struct tcp directly.  They are currently
1125  * separate because the tcpcb structure is exported to userland for sysctl
1126  * parsing purposes, which do not know about callouts.
1127  */
1128 struct tcpcb_mem {
1129 	struct	tcpcb		tcb;
1130 	struct	tcp_timer	tt;
1131 	struct	cc_var		ccv;
1132 #ifdef TCP_HHOOK
1133 	struct	osd		osd;
1134 #endif
1135 };
1136 
1137 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1138 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1139 
1140 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1141 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1142 
1143 static struct mtx isn_mtx;
1144 
1145 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1146 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1147 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1148 
1149 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1150 
1151 /*
1152  * Take a value and get the next power of 2 that doesn't overflow.
1153  * Used to size the tcp_inpcb hash buckets.
1154  */
1155 static int
1156 maketcp_hashsize(int size)
1157 {
1158 	int hashsize;
1159 
1160 	/*
1161 	 * auto tune.
1162 	 * get the next power of 2 higher than maxsockets.
1163 	 */
1164 	hashsize = 1 << fls(size);
1165 	/* catch overflow, and just go one power of 2 smaller */
1166 	if (hashsize < size) {
1167 		hashsize = 1 << (fls(size) - 1);
1168 	}
1169 	return (hashsize);
1170 }
1171 
1172 static volatile int next_tcp_stack_id = 1;
1173 
1174 /*
1175  * Register a TCP function block with the name provided in the names
1176  * array.  (Note that this function does NOT automatically register
1177  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1178  * explicitly include blk->tfb_tcp_block_name in the list of names if
1179  * you wish to register the stack with that name.)
1180  *
1181  * Either all name registrations will succeed or all will fail.  If
1182  * a name registration fails, the function will update the num_names
1183  * argument to point to the array index of the name that encountered
1184  * the failure.
1185  *
1186  * Returns 0 on success, or an error code on failure.
1187  */
1188 int
1189 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1190     const char *names[], int *num_names)
1191 {
1192 	struct tcp_function *n;
1193 	struct tcp_function_set fs;
1194 	int error, i;
1195 
1196 	KASSERT(names != NULL && *num_names > 0,
1197 	    ("%s: Called with 0-length name list", __func__));
1198 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1199 	KASSERT(rw_initialized(&tcp_function_lock),
1200 	    ("%s: called too early", __func__));
1201 
1202 	if ((blk->tfb_tcp_output == NULL) ||
1203 	    (blk->tfb_tcp_do_segment == NULL) ||
1204 	    (blk->tfb_tcp_ctloutput == NULL) ||
1205 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1206 		/*
1207 		 * These functions are required and you
1208 		 * need a name.
1209 		 */
1210 		*num_names = 0;
1211 		return (EINVAL);
1212 	}
1213 	if (blk->tfb_tcp_timer_stop_all ||
1214 	    blk->tfb_tcp_timer_activate ||
1215 	    blk->tfb_tcp_timer_active ||
1216 	    blk->tfb_tcp_timer_stop) {
1217 		/*
1218 		 * If you define one timer function you
1219 		 * must have them all.
1220 		 */
1221 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1222 		    (blk->tfb_tcp_timer_activate == NULL) ||
1223 		    (blk->tfb_tcp_timer_active == NULL) ||
1224 		    (blk->tfb_tcp_timer_stop == NULL)) {
1225 			*num_names = 0;
1226 			return (EINVAL);
1227 		}
1228 	}
1229 
1230 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1231 		*num_names = 0;
1232 		return (EINVAL);
1233 	}
1234 
1235 	refcount_init(&blk->tfb_refcnt, 0);
1236 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1237 	for (i = 0; i < *num_names; i++) {
1238 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1239 		if (n == NULL) {
1240 			error = ENOMEM;
1241 			goto cleanup;
1242 		}
1243 		n->tf_fb = blk;
1244 
1245 		(void)strlcpy(fs.function_set_name, names[i],
1246 		    sizeof(fs.function_set_name));
1247 		rw_wlock(&tcp_function_lock);
1248 		if (find_tcp_functions_locked(&fs) != NULL) {
1249 			/* Duplicate name space not allowed */
1250 			rw_wunlock(&tcp_function_lock);
1251 			free(n, M_TCPFUNCTIONS);
1252 			error = EALREADY;
1253 			goto cleanup;
1254 		}
1255 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1256 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1257 		tcp_fb_cnt++;
1258 		rw_wunlock(&tcp_function_lock);
1259 	}
1260 	return(0);
1261 
1262 cleanup:
1263 	/*
1264 	 * Deregister the names we just added. Because registration failed
1265 	 * for names[i], we don't need to deregister that name.
1266 	 */
1267 	*num_names = i;
1268 	rw_wlock(&tcp_function_lock);
1269 	while (--i >= 0) {
1270 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1271 			if (!strncmp(n->tf_name, names[i],
1272 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1273 				TAILQ_REMOVE(&t_functions, n, tf_next);
1274 				tcp_fb_cnt--;
1275 				n->tf_fb = NULL;
1276 				free(n, M_TCPFUNCTIONS);
1277 				break;
1278 			}
1279 		}
1280 	}
1281 	rw_wunlock(&tcp_function_lock);
1282 	return (error);
1283 }
1284 
1285 /*
1286  * Register a TCP function block using the name provided in the name
1287  * argument.
1288  *
1289  * Returns 0 on success, or an error code on failure.
1290  */
1291 int
1292 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1293     int wait)
1294 {
1295 	const char *name_list[1];
1296 	int num_names, rv;
1297 
1298 	num_names = 1;
1299 	if (name != NULL)
1300 		name_list[0] = name;
1301 	else
1302 		name_list[0] = blk->tfb_tcp_block_name;
1303 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1304 	return (rv);
1305 }
1306 
1307 /*
1308  * Register a TCP function block using the name defined in
1309  * blk->tfb_tcp_block_name.
1310  *
1311  * Returns 0 on success, or an error code on failure.
1312  */
1313 int
1314 register_tcp_functions(struct tcp_function_block *blk, int wait)
1315 {
1316 
1317 	return (register_tcp_functions_as_name(blk, NULL, wait));
1318 }
1319 
1320 /*
1321  * Deregister all names associated with a function block. This
1322  * functionally removes the function block from use within the system.
1323  *
1324  * When called with a true quiesce argument, mark the function block
1325  * as being removed so no more stacks will use it and determine
1326  * whether the removal would succeed.
1327  *
1328  * When called with a false quiesce argument, actually attempt the
1329  * removal.
1330  *
1331  * When called with a force argument, attempt to switch all TCBs to
1332  * use the default stack instead of returning EBUSY.
1333  *
1334  * Returns 0 on success (or if the removal would succeed, or an error
1335  * code on failure.
1336  */
1337 int
1338 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1339     bool force)
1340 {
1341 	struct tcp_function *f;
1342 
1343 	if (blk == &tcp_def_funcblk) {
1344 		/* You can't un-register the default */
1345 		return (EPERM);
1346 	}
1347 	rw_wlock(&tcp_function_lock);
1348 	if (blk == tcp_func_set_ptr) {
1349 		/* You can't free the current default */
1350 		rw_wunlock(&tcp_function_lock);
1351 		return (EBUSY);
1352 	}
1353 	/* Mark the block so no more stacks can use it. */
1354 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1355 	/*
1356 	 * If TCBs are still attached to the stack, attempt to switch them
1357 	 * to the default stack.
1358 	 */
1359 	if (force && blk->tfb_refcnt) {
1360 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1361 		    INPLOOKUP_WLOCKPCB);
1362 		struct inpcb *inp;
1363 		struct tcpcb *tp;
1364 		VNET_ITERATOR_DECL(vnet_iter);
1365 
1366 		rw_wunlock(&tcp_function_lock);
1367 
1368 		VNET_LIST_RLOCK();
1369 		VNET_FOREACH(vnet_iter) {
1370 			CURVNET_SET(vnet_iter);
1371 			while ((inp = inp_next(&inpi)) != NULL) {
1372 				if (inp->inp_flags & INP_TIMEWAIT)
1373 					continue;
1374 				tp = intotcpcb(inp);
1375 				if (tp == NULL || tp->t_fb != blk)
1376 					continue;
1377 				tcp_switch_back_to_default(tp);
1378 			}
1379 			CURVNET_RESTORE();
1380 		}
1381 		VNET_LIST_RUNLOCK();
1382 
1383 		rw_wlock(&tcp_function_lock);
1384 	}
1385 	if (blk->tfb_refcnt) {
1386 		/* TCBs still attached. */
1387 		rw_wunlock(&tcp_function_lock);
1388 		return (EBUSY);
1389 	}
1390 	if (quiesce) {
1391 		/* Skip removal. */
1392 		rw_wunlock(&tcp_function_lock);
1393 		return (0);
1394 	}
1395 	/* Remove any function names that map to this function block. */
1396 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1397 		TAILQ_REMOVE(&t_functions, f, tf_next);
1398 		tcp_fb_cnt--;
1399 		f->tf_fb = NULL;
1400 		free(f, M_TCPFUNCTIONS);
1401 	}
1402 	rw_wunlock(&tcp_function_lock);
1403 	return (0);
1404 }
1405 
1406 static void
1407 tcp_vnet_init(void *arg __unused)
1408 {
1409 
1410 #ifdef TCP_HHOOK
1411 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1412 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1413 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1414 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1415 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1416 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1417 #endif
1418 #ifdef STATS
1419 	if (tcp_stats_init())
1420 		printf("%s: WARNING: unable to initialise TCP stats\n",
1421 		    __func__);
1422 #endif
1423 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1424 	    tcp_tcbhashsize);
1425 
1426 	/*
1427 	 * These have to be type stable for the benefit of the timers.
1428 	 */
1429 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1430 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1431 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1432 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1433 
1434 	tcp_tw_init();
1435 	syncache_init();
1436 	tcp_hc_init();
1437 
1438 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1439 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1440 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1441 
1442 	tcp_fastopen_init();
1443 
1444 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1445 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1446 
1447 	V_tcp_msl = TCPTV_MSL;
1448 }
1449 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1450     tcp_vnet_init, NULL);
1451 
1452 static void
1453 tcp_init(void *arg __unused)
1454 {
1455 	const char *tcbhash_tuneable;
1456 	int hashsize;
1457 
1458 	tcp_reass_global_init();
1459 
1460 	/* XXX virtualize those below? */
1461 	tcp_delacktime = TCPTV_DELACK;
1462 	tcp_keepinit = TCPTV_KEEP_INIT;
1463 	tcp_keepidle = TCPTV_KEEP_IDLE;
1464 	tcp_keepintvl = TCPTV_KEEPINTVL;
1465 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1466 	tcp_rexmit_initial = TCPTV_RTOBASE;
1467 	if (tcp_rexmit_initial < 1)
1468 		tcp_rexmit_initial = 1;
1469 	tcp_rexmit_min = TCPTV_MIN;
1470 	if (tcp_rexmit_min < 1)
1471 		tcp_rexmit_min = 1;
1472 	tcp_persmin = TCPTV_PERSMIN;
1473 	tcp_persmax = TCPTV_PERSMAX;
1474 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1475 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1476 
1477 	/* Setup the tcp function block list */
1478 	TAILQ_INIT(&t_functions);
1479 	rw_init(&tcp_function_lock, "tcp_func_lock");
1480 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1481 #ifdef TCP_BLACKBOX
1482 	/* Initialize the TCP logging data. */
1483 	tcp_log_init();
1484 #endif
1485 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1486 
1487 	if (tcp_soreceive_stream) {
1488 #ifdef INET
1489 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1490 #endif
1491 #ifdef INET6
1492 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1493 #endif /* INET6 */
1494 	}
1495 
1496 #ifdef INET6
1497 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1498 #else /* INET6 */
1499 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1500 #endif /* INET6 */
1501 	if (max_protohdr < TCP_MINPROTOHDR)
1502 		max_protohdr = TCP_MINPROTOHDR;
1503 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1504 		panic("tcp_init");
1505 #undef TCP_MINPROTOHDR
1506 
1507 	ISN_LOCK_INIT();
1508 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1509 		SHUTDOWN_PRI_DEFAULT);
1510 
1511 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1512 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1513 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1514 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1515 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1516 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1517 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1518 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1519 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1520 #ifdef TCPPCAP
1521 	tcp_pcap_init();
1522 #endif
1523 
1524 	hashsize = TCBHASHSIZE;
1525 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1526 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1527 	if (hashsize == 0) {
1528 		/*
1529 		 * Auto tune the hash size based on maxsockets.
1530 		 * A perfect hash would have a 1:1 mapping
1531 		 * (hashsize = maxsockets) however it's been
1532 		 * suggested that O(2) average is better.
1533 		 */
1534 		hashsize = maketcp_hashsize(maxsockets / 4);
1535 		/*
1536 		 * Our historical default is 512,
1537 		 * do not autotune lower than this.
1538 		 */
1539 		if (hashsize < 512)
1540 			hashsize = 512;
1541 		if (bootverbose)
1542 			printf("%s: %s auto tuned to %d\n", __func__,
1543 			    tcbhash_tuneable, hashsize);
1544 	}
1545 	/*
1546 	 * We require a hashsize to be a power of two.
1547 	 * Previously if it was not a power of two we would just reset it
1548 	 * back to 512, which could be a nasty surprise if you did not notice
1549 	 * the error message.
1550 	 * Instead what we do is clip it to the closest power of two lower
1551 	 * than the specified hash value.
1552 	 */
1553 	if (!powerof2(hashsize)) {
1554 		int oldhashsize = hashsize;
1555 
1556 		hashsize = maketcp_hashsize(hashsize);
1557 		/* prevent absurdly low value */
1558 		if (hashsize < 16)
1559 			hashsize = 16;
1560 		printf("%s: WARNING: TCB hash size not a power of 2, "
1561 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1562 		    hashsize);
1563 	}
1564 	tcp_tcbhashsize = hashsize;
1565 }
1566 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1567 
1568 #ifdef VIMAGE
1569 static void
1570 tcp_destroy(void *unused __unused)
1571 {
1572 	int n;
1573 #ifdef TCP_HHOOK
1574 	int error;
1575 #endif
1576 
1577 	/*
1578 	 * All our processes are gone, all our sockets should be cleaned
1579 	 * up, which means, we should be past the tcp_discardcb() calls.
1580 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1581 	 */
1582 	for (;;) {
1583 		INP_INFO_WLOCK(&V_tcbinfo);
1584 		n = V_tcbinfo.ipi_count;
1585 		INP_INFO_WUNLOCK(&V_tcbinfo);
1586 		if (n == 0)
1587 			break;
1588 		pause("tcpdes", hz / 10);
1589 	}
1590 	tcp_hc_destroy();
1591 	syncache_destroy();
1592 	tcp_tw_destroy();
1593 	in_pcbinfo_destroy(&V_tcbinfo);
1594 	/* tcp_discardcb() clears the sack_holes up. */
1595 	uma_zdestroy(V_sack_hole_zone);
1596 	uma_zdestroy(V_tcpcb_zone);
1597 
1598 	/*
1599 	 * Cannot free the zone until all tcpcbs are released as we attach
1600 	 * the allocations to them.
1601 	 */
1602 	tcp_fastopen_destroy();
1603 
1604 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1605 	VNET_PCPUSTAT_FREE(tcpstat);
1606 
1607 #ifdef TCP_HHOOK
1608 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1609 	if (error != 0) {
1610 		printf("%s: WARNING: unable to deregister helper hook "
1611 		    "type=%d, id=%d: error %d returned\n", __func__,
1612 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1613 	}
1614 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1615 	if (error != 0) {
1616 		printf("%s: WARNING: unable to deregister helper hook "
1617 		    "type=%d, id=%d: error %d returned\n", __func__,
1618 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1619 	}
1620 #endif
1621 }
1622 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1623 #endif
1624 
1625 void
1626 tcp_fini(void *xtp)
1627 {
1628 
1629 }
1630 
1631 /*
1632  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1633  * tcp_template used to store this data in mbufs, but we now recopy it out
1634  * of the tcpcb each time to conserve mbufs.
1635  */
1636 void
1637 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1638 {
1639 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1640 
1641 	INP_WLOCK_ASSERT(inp);
1642 
1643 #ifdef INET6
1644 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1645 		struct ip6_hdr *ip6;
1646 
1647 		ip6 = (struct ip6_hdr *)ip_ptr;
1648 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1649 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1650 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1651 			(IPV6_VERSION & IPV6_VERSION_MASK);
1652 		if (port == 0)
1653 			ip6->ip6_nxt = IPPROTO_TCP;
1654 		else
1655 			ip6->ip6_nxt = IPPROTO_UDP;
1656 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1657 		ip6->ip6_src = inp->in6p_laddr;
1658 		ip6->ip6_dst = inp->in6p_faddr;
1659 	}
1660 #endif /* INET6 */
1661 #if defined(INET6) && defined(INET)
1662 	else
1663 #endif
1664 #ifdef INET
1665 	{
1666 		struct ip *ip;
1667 
1668 		ip = (struct ip *)ip_ptr;
1669 		ip->ip_v = IPVERSION;
1670 		ip->ip_hl = 5;
1671 		ip->ip_tos = inp->inp_ip_tos;
1672 		ip->ip_len = 0;
1673 		ip->ip_id = 0;
1674 		ip->ip_off = 0;
1675 		ip->ip_ttl = inp->inp_ip_ttl;
1676 		ip->ip_sum = 0;
1677 		if (port == 0)
1678 			ip->ip_p = IPPROTO_TCP;
1679 		else
1680 			ip->ip_p = IPPROTO_UDP;
1681 		ip->ip_src = inp->inp_laddr;
1682 		ip->ip_dst = inp->inp_faddr;
1683 	}
1684 #endif /* INET */
1685 	th->th_sport = inp->inp_lport;
1686 	th->th_dport = inp->inp_fport;
1687 	th->th_seq = 0;
1688 	th->th_ack = 0;
1689 	th->th_x2 = 0;
1690 	th->th_off = 5;
1691 	th->th_flags = 0;
1692 	th->th_win = 0;
1693 	th->th_urp = 0;
1694 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1695 }
1696 
1697 /*
1698  * Create template to be used to send tcp packets on a connection.
1699  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1700  * use for this function is in keepalives, which use tcp_respond.
1701  */
1702 struct tcptemp *
1703 tcpip_maketemplate(struct inpcb *inp)
1704 {
1705 	struct tcptemp *t;
1706 
1707 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1708 	if (t == NULL)
1709 		return (NULL);
1710 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1711 	return (t);
1712 }
1713 
1714 /*
1715  * Send a single message to the TCP at address specified by
1716  * the given TCP/IP header.  If m == NULL, then we make a copy
1717  * of the tcpiphdr at th and send directly to the addressed host.
1718  * This is used to force keep alive messages out using the TCP
1719  * template for a connection.  If flags are given then we send
1720  * a message back to the TCP which originated the segment th,
1721  * and discard the mbuf containing it and any other attached mbufs.
1722  *
1723  * In any case the ack and sequence number of the transmitted
1724  * segment are as specified by the parameters.
1725  *
1726  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1727  */
1728 void
1729 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1730     tcp_seq ack, tcp_seq seq, int flags)
1731 {
1732 	struct tcpopt to;
1733 	struct inpcb *inp;
1734 	struct ip *ip;
1735 	struct mbuf *optm;
1736 	struct udphdr *uh = NULL;
1737 	struct tcphdr *nth;
1738 	struct tcp_log_buffer *lgb;
1739 	u_char *optp;
1740 #ifdef INET6
1741 	struct ip6_hdr *ip6;
1742 	int isipv6;
1743 #endif /* INET6 */
1744 	int optlen, tlen, win, ulen;
1745 	bool incl_opts;
1746 	uint16_t port;
1747 	int output_ret;
1748 #ifdef INVARIANTS
1749 	int thflags = th->th_flags;
1750 #endif
1751 
1752 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1753 	NET_EPOCH_ASSERT();
1754 
1755 #ifdef INET6
1756 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1757 	ip6 = ipgen;
1758 #endif /* INET6 */
1759 	ip = ipgen;
1760 
1761 	if (tp != NULL) {
1762 		inp = tp->t_inpcb;
1763 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1764 		INP_LOCK_ASSERT(inp);
1765 	} else
1766 		inp = NULL;
1767 
1768 	if (m != NULL) {
1769 #ifdef INET6
1770 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1771 			port = m->m_pkthdr.tcp_tun_port;
1772 		else
1773 #endif
1774 		if (ip && (ip->ip_p == IPPROTO_UDP))
1775 			port = m->m_pkthdr.tcp_tun_port;
1776 		else
1777 			port = 0;
1778 	} else
1779 		port = tp->t_port;
1780 
1781 	incl_opts = false;
1782 	win = 0;
1783 	if (tp != NULL) {
1784 		if (!(flags & TH_RST)) {
1785 			win = sbspace(&inp->inp_socket->so_rcv);
1786 			if (win > TCP_MAXWIN << tp->rcv_scale)
1787 				win = TCP_MAXWIN << tp->rcv_scale;
1788 		}
1789 		if ((tp->t_flags & TF_NOOPT) == 0)
1790 			incl_opts = true;
1791 	}
1792 	if (m == NULL) {
1793 		m = m_gethdr(M_NOWAIT, MT_DATA);
1794 		if (m == NULL)
1795 			return;
1796 		m->m_data += max_linkhdr;
1797 #ifdef INET6
1798 		if (isipv6) {
1799 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1800 			      sizeof(struct ip6_hdr));
1801 			ip6 = mtod(m, struct ip6_hdr *);
1802 			nth = (struct tcphdr *)(ip6 + 1);
1803 			if (port) {
1804 				/* Insert a UDP header */
1805 				uh = (struct udphdr *)nth;
1806 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1807 				uh->uh_dport = port;
1808 				nth = (struct tcphdr *)(uh + 1);
1809 			}
1810 		} else
1811 #endif /* INET6 */
1812 		{
1813 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1814 			ip = mtod(m, struct ip *);
1815 			nth = (struct tcphdr *)(ip + 1);
1816 			if (port) {
1817 				/* Insert a UDP header */
1818 				uh = (struct udphdr *)nth;
1819 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1820 				uh->uh_dport = port;
1821 				nth = (struct tcphdr *)(uh + 1);
1822 			}
1823 		}
1824 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1825 		flags = TH_ACK;
1826 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1827 		struct mbuf *n;
1828 
1829 		/* Can't reuse 'm', allocate a new mbuf. */
1830 		n = m_gethdr(M_NOWAIT, MT_DATA);
1831 		if (n == NULL) {
1832 			m_freem(m);
1833 			return;
1834 		}
1835 
1836 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1837 			m_freem(m);
1838 			m_freem(n);
1839 			return;
1840 		}
1841 
1842 		n->m_data += max_linkhdr;
1843 		/* m_len is set later */
1844 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1845 #ifdef INET6
1846 		if (isipv6) {
1847 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1848 			      sizeof(struct ip6_hdr));
1849 			ip6 = mtod(n, struct ip6_hdr *);
1850 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1851 			nth = (struct tcphdr *)(ip6 + 1);
1852 			if (port) {
1853 				/* Insert a UDP header */
1854 				uh = (struct udphdr *)nth;
1855 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1856 				uh->uh_dport = port;
1857 				nth = (struct tcphdr *)(uh + 1);
1858 			}
1859 		} else
1860 #endif /* INET6 */
1861 		{
1862 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1863 			ip = mtod(n, struct ip *);
1864 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1865 			nth = (struct tcphdr *)(ip + 1);
1866 			if (port) {
1867 				/* Insert a UDP header */
1868 				uh = (struct udphdr *)nth;
1869 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1870 				uh->uh_dport = port;
1871 				nth = (struct tcphdr *)(uh + 1);
1872 			}
1873 		}
1874 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1875 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1876 		th = nth;
1877 		m_freem(m);
1878 		m = n;
1879 	} else {
1880 		/*
1881 		 *  reuse the mbuf.
1882 		 * XXX MRT We inherit the FIB, which is lucky.
1883 		 */
1884 		m_freem(m->m_next);
1885 		m->m_next = NULL;
1886 		m->m_data = (caddr_t)ipgen;
1887 		/* m_len is set later */
1888 #ifdef INET6
1889 		if (isipv6) {
1890 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1891 			nth = (struct tcphdr *)(ip6 + 1);
1892 		} else
1893 #endif /* INET6 */
1894 		{
1895 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1896 			nth = (struct tcphdr *)(ip + 1);
1897 		}
1898 		if (th != nth) {
1899 			/*
1900 			 * this is usually a case when an extension header
1901 			 * exists between the IPv6 header and the
1902 			 * TCP header.
1903 			 */
1904 			nth->th_sport = th->th_sport;
1905 			nth->th_dport = th->th_dport;
1906 		}
1907 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1908 #undef xchg
1909 	}
1910 	tlen = 0;
1911 #ifdef INET6
1912 	if (isipv6)
1913 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1914 #endif
1915 #if defined(INET) && defined(INET6)
1916 	else
1917 #endif
1918 #ifdef INET
1919 		tlen = sizeof (struct tcpiphdr);
1920 #endif
1921 	if (port)
1922 		tlen += sizeof (struct udphdr);
1923 #ifdef INVARIANTS
1924 	m->m_len = 0;
1925 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1926 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1927 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1928 #endif
1929 	m->m_len = tlen;
1930 	to.to_flags = 0;
1931 	if (incl_opts) {
1932 		/* Make sure we have room. */
1933 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1934 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1935 			if (m->m_next) {
1936 				optp = mtod(m->m_next, u_char *);
1937 				optm = m->m_next;
1938 			} else
1939 				incl_opts = false;
1940 		} else {
1941 			optp = (u_char *) (nth + 1);
1942 			optm = m;
1943 		}
1944 	}
1945 	if (incl_opts) {
1946 		/* Timestamps. */
1947 		if (tp->t_flags & TF_RCVD_TSTMP) {
1948 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1949 			to.to_tsecr = tp->ts_recent;
1950 			to.to_flags |= TOF_TS;
1951 		}
1952 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1953 		/* TCP-MD5 (RFC2385). */
1954 		if (tp->t_flags & TF_SIGNATURE)
1955 			to.to_flags |= TOF_SIGNATURE;
1956 #endif
1957 		/* Add the options. */
1958 		tlen += optlen = tcp_addoptions(&to, optp);
1959 
1960 		/* Update m_len in the correct mbuf. */
1961 		optm->m_len += optlen;
1962 	} else
1963 		optlen = 0;
1964 #ifdef INET6
1965 	if (isipv6) {
1966 		if (uh) {
1967 			ulen = tlen - sizeof(struct ip6_hdr);
1968 			uh->uh_ulen = htons(ulen);
1969 		}
1970 		ip6->ip6_flow = 0;
1971 		ip6->ip6_vfc = IPV6_VERSION;
1972 		if (port)
1973 			ip6->ip6_nxt = IPPROTO_UDP;
1974 		else
1975 			ip6->ip6_nxt = IPPROTO_TCP;
1976 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1977 	}
1978 #endif
1979 #if defined(INET) && defined(INET6)
1980 	else
1981 #endif
1982 #ifdef INET
1983 	{
1984 		if (uh) {
1985 			ulen = tlen - sizeof(struct ip);
1986 			uh->uh_ulen = htons(ulen);
1987 		}
1988 		ip->ip_len = htons(tlen);
1989 		ip->ip_ttl = V_ip_defttl;
1990 		if (port) {
1991 			ip->ip_p = IPPROTO_UDP;
1992 		} else {
1993 			ip->ip_p = IPPROTO_TCP;
1994 		}
1995 		if (V_path_mtu_discovery)
1996 			ip->ip_off |= htons(IP_DF);
1997 	}
1998 #endif
1999 	m->m_pkthdr.len = tlen;
2000 	m->m_pkthdr.rcvif = NULL;
2001 #ifdef MAC
2002 	if (inp != NULL) {
2003 		/*
2004 		 * Packet is associated with a socket, so allow the
2005 		 * label of the response to reflect the socket label.
2006 		 */
2007 		INP_LOCK_ASSERT(inp);
2008 		mac_inpcb_create_mbuf(inp, m);
2009 	} else {
2010 		/*
2011 		 * Packet is not associated with a socket, so possibly
2012 		 * update the label in place.
2013 		 */
2014 		mac_netinet_tcp_reply(m);
2015 	}
2016 #endif
2017 	nth->th_seq = htonl(seq);
2018 	nth->th_ack = htonl(ack);
2019 	nth->th_x2 = 0;
2020 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2021 	nth->th_flags = flags;
2022 	if (tp != NULL)
2023 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2024 	else
2025 		nth->th_win = htons((u_short)win);
2026 	nth->th_urp = 0;
2027 
2028 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2029 	if (to.to_flags & TOF_SIGNATURE) {
2030 		if (!TCPMD5_ENABLED() ||
2031 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2032 			m_freem(m);
2033 			return;
2034 		}
2035 	}
2036 #endif
2037 
2038 #ifdef INET6
2039 	if (isipv6) {
2040 		if (port) {
2041 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2042 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2043 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2044 			nth->th_sum = 0;
2045 		} else {
2046 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2047 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2048 			nth->th_sum = in6_cksum_pseudo(ip6,
2049 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2050 		}
2051 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2052 		    NULL, NULL);
2053 	}
2054 #endif /* INET6 */
2055 #if defined(INET6) && defined(INET)
2056 	else
2057 #endif
2058 #ifdef INET
2059 	{
2060 		if (port) {
2061 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2062 			    htons(ulen + IPPROTO_UDP));
2063 			m->m_pkthdr.csum_flags = CSUM_UDP;
2064 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2065 			nth->th_sum = 0;
2066 		} else {
2067 			m->m_pkthdr.csum_flags = CSUM_TCP;
2068 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2069 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2070 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2071 		}
2072 	}
2073 #endif /* INET */
2074 #ifdef TCPDEBUG
2075 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2076 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2077 #endif
2078 	TCP_PROBE3(debug__output, tp, th, m);
2079 	if (flags & TH_RST)
2080 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2081 	lgb = NULL;
2082 	if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2083 		if (INP_WLOCKED(inp)) {
2084 			union tcp_log_stackspecific log;
2085 			struct timeval tv;
2086 
2087 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2088 			log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2089 			log.u_bbr.flex8 = 4;
2090 			log.u_bbr.pkts_out = tp->t_maxseg;
2091 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2092 			log.u_bbr.delivered = 0;
2093 			lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2094 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2095 		} else {
2096 			/*
2097 			 * We can not log the packet, since we only own the
2098 			 * read lock, but a write lock is needed. The read lock
2099 			 * is not upgraded to a write lock, since only getting
2100 			 * the read lock was done intentionally to improve the
2101 			 * handling of SYN flooding attacks.
2102 			 * This happens only for pure SYN segments received in
2103 			 * the initial CLOSED state, or received in a more
2104 			 * advanced state than listen and the UDP encapsulation
2105 			 * port is unexpected.
2106 			 * The incoming SYN segments do not really belong to
2107 			 * the TCP connection and the handling does not change
2108 			 * the state of the TCP connection. Therefore, the
2109 			 * sending of the RST segments is not logged. Please
2110 			 * note that also the incoming SYN segments are not
2111 			 * logged.
2112 			 *
2113 			 * The following code ensures that the above description
2114 			 * is and stays correct.
2115 			 */
2116 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2117 			    (tp->t_state == TCPS_CLOSED ||
2118 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2119 			    ("%s: Logging of TCP segment with flags 0x%b and "
2120 			    "UDP encapsulation port %u skipped in state %s",
2121 			    __func__, thflags, PRINT_TH_FLAGS,
2122 			    ntohs(port), tcpstates[tp->t_state]));
2123 		}
2124 	}
2125 
2126 #ifdef INET6
2127 	if (isipv6) {
2128 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2129 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2130 	}
2131 #endif /* INET6 */
2132 #if defined(INET) && defined(INET6)
2133 	else
2134 #endif
2135 #ifdef INET
2136 	{
2137 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2138 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2139 	}
2140 #endif
2141 	if (lgb != NULL)
2142 		lgb->tlb_errno = output_ret;
2143 }
2144 
2145 /*
2146  * Create a new TCP control block, making an
2147  * empty reassembly queue and hooking it to the argument
2148  * protocol control block.  The `inp' parameter must have
2149  * come from the zone allocator set up in tcp_init().
2150  */
2151 struct tcpcb *
2152 tcp_newtcpcb(struct inpcb *inp)
2153 {
2154 	struct tcpcb_mem *tm;
2155 	struct tcpcb *tp;
2156 #ifdef INET6
2157 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2158 #endif /* INET6 */
2159 
2160 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2161 	if (tm == NULL)
2162 		return (NULL);
2163 	tp = &tm->tcb;
2164 
2165 	/* Initialise cc_var struct for this tcpcb. */
2166 	tp->ccv = &tm->ccv;
2167 	tp->ccv->type = IPPROTO_TCP;
2168 	tp->ccv->ccvc.tcp = tp;
2169 	rw_rlock(&tcp_function_lock);
2170 	tp->t_fb = tcp_func_set_ptr;
2171 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2172 	rw_runlock(&tcp_function_lock);
2173 	/*
2174 	 * Use the current system default CC algorithm.
2175 	 */
2176 	CC_LIST_RLOCK();
2177 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2178 	CC_ALGO(tp) = CC_DEFAULT_ALGO();
2179 	CC_LIST_RUNLOCK();
2180 
2181 	/*
2182 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2183 	 * is called.
2184 	 */
2185 	in_pcbref(inp);	/* Reference for tcpcb */
2186 	tp->t_inpcb = inp;
2187 
2188 	if (CC_ALGO(tp)->cb_init != NULL)
2189 		if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2190 			if (tp->t_fb->tfb_tcp_fb_fini)
2191 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2192 			in_pcbrele_wlocked(inp);
2193 			refcount_release(&tp->t_fb->tfb_refcnt);
2194 			uma_zfree(V_tcpcb_zone, tm);
2195 			return (NULL);
2196 		}
2197 
2198 #ifdef TCP_HHOOK
2199 	tp->osd = &tm->osd;
2200 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2201 		if (tp->t_fb->tfb_tcp_fb_fini)
2202 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2203 		in_pcbrele_wlocked(inp);
2204 		refcount_release(&tp->t_fb->tfb_refcnt);
2205 		uma_zfree(V_tcpcb_zone, tm);
2206 		return (NULL);
2207 	}
2208 #endif
2209 
2210 #ifdef VIMAGE
2211 	tp->t_vnet = inp->inp_vnet;
2212 #endif
2213 	tp->t_timers = &tm->tt;
2214 	TAILQ_INIT(&tp->t_segq);
2215 	tp->t_maxseg =
2216 #ifdef INET6
2217 		isipv6 ? V_tcp_v6mssdflt :
2218 #endif /* INET6 */
2219 		V_tcp_mssdflt;
2220 
2221 	/* Set up our timeouts. */
2222 	callout_init(&tp->t_timers->tt_rexmt, 1);
2223 	callout_init(&tp->t_timers->tt_persist, 1);
2224 	callout_init(&tp->t_timers->tt_keep, 1);
2225 	callout_init(&tp->t_timers->tt_2msl, 1);
2226 	callout_init(&tp->t_timers->tt_delack, 1);
2227 
2228 	if (V_tcp_do_rfc1323)
2229 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2230 	if (V_tcp_do_sack)
2231 		tp->t_flags |= TF_SACK_PERMIT;
2232 	TAILQ_INIT(&tp->snd_holes);
2233 
2234 	/*
2235 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2236 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2237 	 * reasonable initial retransmit time.
2238 	 */
2239 	tp->t_srtt = TCPTV_SRTTBASE;
2240 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2241 	tp->t_rttmin = tcp_rexmit_min;
2242 	tp->t_rxtcur = tcp_rexmit_initial;
2243 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2244 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2245 	tp->t_rcvtime = ticks;
2246 	/*
2247 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2248 	 * because the socket may be bound to an IPv6 wildcard address,
2249 	 * which may match an IPv4-mapped IPv6 address.
2250 	 */
2251 	inp->inp_ip_ttl = V_ip_defttl;
2252 	inp->inp_ppcb = tp;
2253 #ifdef TCPPCAP
2254 	/*
2255 	 * Init the TCP PCAP queues.
2256 	 */
2257 	tcp_pcap_tcpcb_init(tp);
2258 #endif
2259 #ifdef TCP_BLACKBOX
2260 	/* Initialize the per-TCPCB log data. */
2261 	tcp_log_tcpcbinit(tp);
2262 #endif
2263 	tp->t_pacing_rate = -1;
2264 	if (tp->t_fb->tfb_tcp_fb_init) {
2265 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2266 			refcount_release(&tp->t_fb->tfb_refcnt);
2267 			in_pcbrele_wlocked(inp);
2268 			uma_zfree(V_tcpcb_zone, tm);
2269 			return (NULL);
2270 		}
2271 	}
2272 #ifdef STATS
2273 	if (V_tcp_perconn_stats_enable == 1)
2274 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2275 #endif
2276 	if (V_tcp_do_lrd)
2277 		tp->t_flags |= TF_LRD;
2278 	return (tp);		/* XXX */
2279 }
2280 
2281 /*
2282  * Switch the congestion control algorithm back to Vnet default for any active
2283  * control blocks using an algorithm which is about to go away. If the algorithm
2284  * has a cb_init function and it fails (no memory) then the operation fails and
2285  * the unload will not succeed.
2286  *
2287  */
2288 int
2289 tcp_ccalgounload(struct cc_algo *unload_algo)
2290 {
2291 	struct cc_algo *oldalgo, *newalgo;
2292 	struct inpcb *inp;
2293 	struct tcpcb *tp;
2294 	VNET_ITERATOR_DECL(vnet_iter);
2295 
2296 	/*
2297 	 * Check all active control blocks across all network stacks and change
2298 	 * any that are using "unload_algo" back to its default. If "unload_algo"
2299 	 * requires cleanup code to be run, call it.
2300 	 */
2301 	VNET_LIST_RLOCK();
2302 	VNET_FOREACH(vnet_iter) {
2303 		CURVNET_SET(vnet_iter);
2304 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2305 		    INPLOOKUP_WLOCKPCB);
2306 		/*
2307 		 * XXXGL: would new accept(2)d connections use algo being
2308 		 * unloaded?
2309 		 */
2310 		newalgo = CC_DEFAULT_ALGO();
2311 		while ((inp = inp_next(&inpi)) != NULL) {
2312 			/* Important to skip tcptw structs. */
2313 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
2314 			    (tp = intotcpcb(inp)) != NULL) {
2315 				/*
2316 				 * By holding INP_WLOCK here, we are assured
2317 				 * that the connection is not currently
2318 				 * executing inside the CC module's functions.
2319 				 * We attempt to switch to the Vnets default,
2320 				 * if the init fails then we fail the whole
2321 				 * operation and the module unload will fail.
2322 				 */
2323 				if (CC_ALGO(tp) == unload_algo) {
2324 					struct cc_var cc_mem;
2325 					int err;
2326 
2327 					oldalgo = CC_ALGO(tp);
2328 					memset(&cc_mem, 0, sizeof(cc_mem));
2329 					cc_mem.ccvc.tcp = tp;
2330 					if (newalgo->cb_init == NULL) {
2331 						/*
2332 						 * No init we can skip the
2333 						 * dance around a possible failure.
2334 						 */
2335 						CC_DATA(tp) = NULL;
2336 						goto proceed;
2337 					}
2338 					err = (newalgo->cb_init)(&cc_mem, NULL);
2339 					if (err) {
2340 						/*
2341 						 * Presumably no memory the caller will
2342 						 * need to try again.
2343 						 */
2344 						INP_WUNLOCK(inp);
2345 						CURVNET_RESTORE();
2346 						VNET_LIST_RUNLOCK();
2347 						return (err);
2348 					}
2349 proceed:
2350 					if (oldalgo->cb_destroy != NULL)
2351 						oldalgo->cb_destroy(tp->ccv);
2352 					CC_ALGO(tp) = newalgo;
2353 					memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var));
2354 					if (TCPS_HAVEESTABLISHED(tp->t_state) &&
2355 					    (CC_ALGO(tp)->conn_init != NULL)) {
2356 						/* Yep run the connection init for the new CC */
2357 						CC_ALGO(tp)->conn_init(tp->ccv);
2358 					}
2359 				}
2360 			}
2361 		}
2362 		CURVNET_RESTORE();
2363 	}
2364 	VNET_LIST_RUNLOCK();
2365 	return (0);
2366 }
2367 
2368 /*
2369  * Drop a TCP connection, reporting
2370  * the specified error.  If connection is synchronized,
2371  * then send a RST to peer.
2372  */
2373 struct tcpcb *
2374 tcp_drop(struct tcpcb *tp, int errno)
2375 {
2376 	struct socket *so = tp->t_inpcb->inp_socket;
2377 
2378 	NET_EPOCH_ASSERT();
2379 	INP_WLOCK_ASSERT(tp->t_inpcb);
2380 
2381 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2382 		tcp_state_change(tp, TCPS_CLOSED);
2383 		/* Don't use tcp_output() here due to possible recursion. */
2384 		(void)tcp_output_nodrop(tp);
2385 		TCPSTAT_INC(tcps_drops);
2386 	} else
2387 		TCPSTAT_INC(tcps_conndrops);
2388 	if (errno == ETIMEDOUT && tp->t_softerror)
2389 		errno = tp->t_softerror;
2390 	so->so_error = errno;
2391 	return (tcp_close(tp));
2392 }
2393 
2394 void
2395 tcp_discardcb(struct tcpcb *tp)
2396 {
2397 	struct inpcb *inp = tp->t_inpcb;
2398 
2399 	INP_WLOCK_ASSERT(inp);
2400 
2401 	/*
2402 	 * Make sure that all of our timers are stopped before we delete the
2403 	 * PCB.
2404 	 *
2405 	 * If stopping a timer fails, we schedule a discard function in same
2406 	 * callout, and the last discard function called will take care of
2407 	 * deleting the tcpcb.
2408 	 */
2409 	tp->t_timers->tt_draincnt = 0;
2410 	tcp_timer_stop(tp, TT_REXMT);
2411 	tcp_timer_stop(tp, TT_PERSIST);
2412 	tcp_timer_stop(tp, TT_KEEP);
2413 	tcp_timer_stop(tp, TT_2MSL);
2414 	tcp_timer_stop(tp, TT_DELACK);
2415 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2416 		/*
2417 		 * Call the stop-all function of the methods,
2418 		 * this function should call the tcp_timer_stop()
2419 		 * method with each of the function specific timeouts.
2420 		 * That stop will be called via the tfb_tcp_timer_stop()
2421 		 * which should use the async drain function of the
2422 		 * callout system (see tcp_var.h).
2423 		 */
2424 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2425 	}
2426 
2427 	/* free the reassembly queue, if any */
2428 	tcp_reass_flush(tp);
2429 
2430 #ifdef TCP_OFFLOAD
2431 	/* Disconnect offload device, if any. */
2432 	if (tp->t_flags & TF_TOE)
2433 		tcp_offload_detach(tp);
2434 #endif
2435 
2436 	tcp_free_sackholes(tp);
2437 
2438 #ifdef TCPPCAP
2439 	/* Free the TCP PCAP queues. */
2440 	tcp_pcap_drain(&(tp->t_inpkts));
2441 	tcp_pcap_drain(&(tp->t_outpkts));
2442 #endif
2443 
2444 	/* Allow the CC algorithm to clean up after itself. */
2445 	if (CC_ALGO(tp)->cb_destroy != NULL)
2446 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2447 	CC_DATA(tp) = NULL;
2448 
2449 #ifdef TCP_HHOOK
2450 	khelp_destroy_osd(tp->osd);
2451 #endif
2452 #ifdef STATS
2453 	stats_blob_destroy(tp->t_stats);
2454 #endif
2455 
2456 	CC_ALGO(tp) = NULL;
2457 	inp->inp_ppcb = NULL;
2458 	if (tp->t_timers->tt_draincnt == 0) {
2459 		bool released __diagused;
2460 
2461 		released = tcp_freecb(tp);
2462 		KASSERT(!released, ("%s: inp %p should not have been released "
2463 		    "here", __func__, inp));
2464 	}
2465 }
2466 
2467 bool
2468 tcp_freecb(struct tcpcb *tp)
2469 {
2470 	struct inpcb *inp = tp->t_inpcb;
2471 	struct socket *so = inp->inp_socket;
2472 #ifdef INET6
2473 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2474 #endif
2475 
2476 	INP_WLOCK_ASSERT(inp);
2477 	MPASS(tp->t_timers->tt_draincnt == 0);
2478 
2479 	/* We own the last reference on tcpcb, let's free it. */
2480 #ifdef TCP_BLACKBOX
2481 	tcp_log_tcpcbfini(tp);
2482 #endif
2483 	TCPSTATES_DEC(tp->t_state);
2484 	if (tp->t_fb->tfb_tcp_fb_fini)
2485 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2486 
2487 	/*
2488 	 * If we got enough samples through the srtt filter,
2489 	 * save the rtt and rttvar in the routing entry.
2490 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2491 	 * 4 samples is enough for the srtt filter to converge
2492 	 * to within enough % of the correct value; fewer samples
2493 	 * and we could save a bogus rtt. The danger is not high
2494 	 * as tcp quickly recovers from everything.
2495 	 * XXX: Works very well but needs some more statistics!
2496 	 *
2497 	 * XXXRRS: Updating must be after the stack fini() since
2498 	 * that may be converting some internal representation of
2499 	 * say srtt etc into the general one used by other stacks.
2500 	 * Lets also at least protect against the so being NULL
2501 	 * as RW stated below.
2502 	 */
2503 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2504 		struct hc_metrics_lite metrics;
2505 		uint32_t ssthresh;
2506 
2507 		bzero(&metrics, sizeof(metrics));
2508 		/*
2509 		 * Update the ssthresh always when the conditions below
2510 		 * are satisfied. This gives us better new start value
2511 		 * for the congestion avoidance for new connections.
2512 		 * ssthresh is only set if packet loss occurred on a session.
2513 		 *
2514 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2515 		 * being torn down.  Ideally this code would not use 'so'.
2516 		 */
2517 		ssthresh = tp->snd_ssthresh;
2518 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2519 			/*
2520 			 * convert the limit from user data bytes to
2521 			 * packets then to packet data bytes.
2522 			 */
2523 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2524 			if (ssthresh < 2)
2525 				ssthresh = 2;
2526 			ssthresh *= (tp->t_maxseg +
2527 #ifdef INET6
2528 			    (isipv6 ? sizeof (struct ip6_hdr) +
2529 			    sizeof (struct tcphdr) :
2530 #endif
2531 			    sizeof (struct tcpiphdr)
2532 #ifdef INET6
2533 			    )
2534 #endif
2535 			    );
2536 		} else
2537 			ssthresh = 0;
2538 		metrics.rmx_ssthresh = ssthresh;
2539 
2540 		metrics.rmx_rtt = tp->t_srtt;
2541 		metrics.rmx_rttvar = tp->t_rttvar;
2542 		metrics.rmx_cwnd = tp->snd_cwnd;
2543 		metrics.rmx_sendpipe = 0;
2544 		metrics.rmx_recvpipe = 0;
2545 
2546 		tcp_hc_update(&inp->inp_inc, &metrics);
2547 	}
2548 
2549 	refcount_release(&tp->t_fb->tfb_refcnt);
2550 	uma_zfree(V_tcpcb_zone, tp);
2551 
2552 	return (in_pcbrele_wlocked(inp));
2553 }
2554 
2555 /*
2556  * Attempt to close a TCP control block, marking it as dropped, and freeing
2557  * the socket if we hold the only reference.
2558  */
2559 struct tcpcb *
2560 tcp_close(struct tcpcb *tp)
2561 {
2562 	struct inpcb *inp = tp->t_inpcb;
2563 	struct socket *so;
2564 
2565 	INP_WLOCK_ASSERT(inp);
2566 
2567 #ifdef TCP_OFFLOAD
2568 	if (tp->t_state == TCPS_LISTEN)
2569 		tcp_offload_listen_stop(tp);
2570 #endif
2571 	/*
2572 	 * This releases the TFO pending counter resource for TFO listen
2573 	 * sockets as well as passively-created TFO sockets that transition
2574 	 * from SYN_RECEIVED to CLOSED.
2575 	 */
2576 	if (tp->t_tfo_pending) {
2577 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2578 		tp->t_tfo_pending = NULL;
2579 	}
2580 #ifdef TCPHPTS
2581 	tcp_hpts_remove(inp);
2582 #endif
2583 	in_pcbdrop(inp);
2584 	TCPSTAT_INC(tcps_closed);
2585 	if (tp->t_state != TCPS_CLOSED)
2586 		tcp_state_change(tp, TCPS_CLOSED);
2587 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2588 	so = inp->inp_socket;
2589 	soisdisconnected(so);
2590 	if (inp->inp_flags & INP_SOCKREF) {
2591 		KASSERT(so->so_state & SS_PROTOREF,
2592 		    ("tcp_close: !SS_PROTOREF"));
2593 		inp->inp_flags &= ~INP_SOCKREF;
2594 		INP_WUNLOCK(inp);
2595 		SOCK_LOCK(so);
2596 		so->so_state &= ~SS_PROTOREF;
2597 		sofree(so);
2598 		return (NULL);
2599 	}
2600 	return (tp);
2601 }
2602 
2603 void
2604 tcp_drain(void)
2605 {
2606 	VNET_ITERATOR_DECL(vnet_iter);
2607 
2608 	if (!do_tcpdrain)
2609 		return;
2610 
2611 	VNET_LIST_RLOCK_NOSLEEP();
2612 	VNET_FOREACH(vnet_iter) {
2613 		CURVNET_SET(vnet_iter);
2614 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2615 		    INPLOOKUP_WLOCKPCB);
2616 		struct inpcb *inpb;
2617 		struct tcpcb *tcpb;
2618 
2619 	/*
2620 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2621 	 * if there is one...
2622 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2623 	 *      reassembly queue should be flushed, but in a situation
2624 	 *	where we're really low on mbufs, this is potentially
2625 	 *	useful.
2626 	 */
2627 		while ((inpb = inp_next(&inpi)) != NULL) {
2628 			if (inpb->inp_flags & INP_TIMEWAIT)
2629 				continue;
2630 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2631 				tcp_reass_flush(tcpb);
2632 				tcp_clean_sackreport(tcpb);
2633 #ifdef TCP_BLACKBOX
2634 				tcp_log_drain(tcpb);
2635 #endif
2636 #ifdef TCPPCAP
2637 				if (tcp_pcap_aggressive_free) {
2638 					/* Free the TCP PCAP queues. */
2639 					tcp_pcap_drain(&(tcpb->t_inpkts));
2640 					tcp_pcap_drain(&(tcpb->t_outpkts));
2641 				}
2642 #endif
2643 			}
2644 		}
2645 		CURVNET_RESTORE();
2646 	}
2647 	VNET_LIST_RUNLOCK_NOSLEEP();
2648 }
2649 
2650 /*
2651  * Notify a tcp user of an asynchronous error;
2652  * store error as soft error, but wake up user
2653  * (for now, won't do anything until can select for soft error).
2654  *
2655  * Do not wake up user since there currently is no mechanism for
2656  * reporting soft errors (yet - a kqueue filter may be added).
2657  */
2658 static struct inpcb *
2659 tcp_notify(struct inpcb *inp, int error)
2660 {
2661 	struct tcpcb *tp;
2662 
2663 	INP_WLOCK_ASSERT(inp);
2664 
2665 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2666 	    (inp->inp_flags & INP_DROPPED))
2667 		return (inp);
2668 
2669 	tp = intotcpcb(inp);
2670 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2671 
2672 	/*
2673 	 * Ignore some errors if we are hooked up.
2674 	 * If connection hasn't completed, has retransmitted several times,
2675 	 * and receives a second error, give up now.  This is better
2676 	 * than waiting a long time to establish a connection that
2677 	 * can never complete.
2678 	 */
2679 	if (tp->t_state == TCPS_ESTABLISHED &&
2680 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2681 	     error == EHOSTDOWN)) {
2682 		if (inp->inp_route.ro_nh) {
2683 			NH_FREE(inp->inp_route.ro_nh);
2684 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2685 		}
2686 		return (inp);
2687 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2688 	    tp->t_softerror) {
2689 		tp = tcp_drop(tp, error);
2690 		if (tp != NULL)
2691 			return (inp);
2692 		else
2693 			return (NULL);
2694 	} else {
2695 		tp->t_softerror = error;
2696 		return (inp);
2697 	}
2698 #if 0
2699 	wakeup( &so->so_timeo);
2700 	sorwakeup(so);
2701 	sowwakeup(so);
2702 #endif
2703 }
2704 
2705 static int
2706 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2707 {
2708 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2709 	    INPLOOKUP_RLOCKPCB);
2710 	struct xinpgen xig;
2711 	struct inpcb *inp;
2712 	int error;
2713 
2714 	if (req->newptr != NULL)
2715 		return (EPERM);
2716 
2717 	if (req->oldptr == NULL) {
2718 		int n;
2719 
2720 		n = V_tcbinfo.ipi_count +
2721 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2722 		n += imax(n / 8, 10);
2723 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2724 		return (0);
2725 	}
2726 
2727 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2728 		return (error);
2729 
2730 	bzero(&xig, sizeof(xig));
2731 	xig.xig_len = sizeof xig;
2732 	xig.xig_count = V_tcbinfo.ipi_count +
2733 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2734 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2735 	xig.xig_sogen = so_gencnt;
2736 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2737 	if (error)
2738 		return (error);
2739 
2740 	error = syncache_pcblist(req);
2741 	if (error)
2742 		return (error);
2743 
2744 	while ((inp = inp_next(&inpi)) != NULL) {
2745 		if (inp->inp_gencnt <= xig.xig_gen) {
2746 			int crerr;
2747 
2748 			/*
2749 			 * XXX: This use of cr_cansee(), introduced with
2750 			 * TCP state changes, is not quite right, but for
2751 			 * now, better than nothing.
2752 			 */
2753 			if (inp->inp_flags & INP_TIMEWAIT) {
2754 				if (intotw(inp) != NULL)
2755 					crerr = cr_cansee(req->td->td_ucred,
2756 					    intotw(inp)->tw_cred);
2757 				else
2758 					crerr = EINVAL;	/* Skip this inp. */
2759 			} else
2760 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2761 			if (crerr == 0) {
2762 				struct xtcpcb xt;
2763 
2764 				tcp_inptoxtp(inp, &xt);
2765 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2766 				if (error) {
2767 					INP_RUNLOCK(inp);
2768 					break;
2769 				} else
2770 					continue;
2771 			}
2772 		}
2773 	}
2774 
2775 	if (!error) {
2776 		/*
2777 		 * Give the user an updated idea of our state.
2778 		 * If the generation differs from what we told
2779 		 * her before, she knows that something happened
2780 		 * while we were processing this request, and it
2781 		 * might be necessary to retry.
2782 		 */
2783 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2784 		xig.xig_sogen = so_gencnt;
2785 		xig.xig_count = V_tcbinfo.ipi_count +
2786 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2787 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2788 	}
2789 
2790 	return (error);
2791 }
2792 
2793 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2794     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2795     NULL, 0, tcp_pcblist, "S,xtcpcb",
2796     "List of active TCP connections");
2797 
2798 #ifdef INET
2799 static int
2800 tcp_getcred(SYSCTL_HANDLER_ARGS)
2801 {
2802 	struct xucred xuc;
2803 	struct sockaddr_in addrs[2];
2804 	struct epoch_tracker et;
2805 	struct inpcb *inp;
2806 	int error;
2807 
2808 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2809 	if (error)
2810 		return (error);
2811 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2812 	if (error)
2813 		return (error);
2814 	NET_EPOCH_ENTER(et);
2815 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2816 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2817 	NET_EPOCH_EXIT(et);
2818 	if (inp != NULL) {
2819 		if (inp->inp_socket == NULL)
2820 			error = ENOENT;
2821 		if (error == 0)
2822 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2823 		if (error == 0)
2824 			cru2x(inp->inp_cred, &xuc);
2825 		INP_RUNLOCK(inp);
2826 	} else
2827 		error = ENOENT;
2828 	if (error == 0)
2829 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2830 	return (error);
2831 }
2832 
2833 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2834     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2835     0, 0, tcp_getcred, "S,xucred",
2836     "Get the xucred of a TCP connection");
2837 #endif /* INET */
2838 
2839 #ifdef INET6
2840 static int
2841 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2842 {
2843 	struct epoch_tracker et;
2844 	struct xucred xuc;
2845 	struct sockaddr_in6 addrs[2];
2846 	struct inpcb *inp;
2847 	int error;
2848 #ifdef INET
2849 	int mapped = 0;
2850 #endif
2851 
2852 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2853 	if (error)
2854 		return (error);
2855 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2856 	if (error)
2857 		return (error);
2858 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2859 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2860 		return (error);
2861 	}
2862 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2863 #ifdef INET
2864 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2865 			mapped = 1;
2866 		else
2867 #endif
2868 			return (EINVAL);
2869 	}
2870 
2871 	NET_EPOCH_ENTER(et);
2872 #ifdef INET
2873 	if (mapped == 1)
2874 		inp = in_pcblookup(&V_tcbinfo,
2875 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2876 			addrs[1].sin6_port,
2877 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2878 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2879 	else
2880 #endif
2881 		inp = in6_pcblookup(&V_tcbinfo,
2882 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2883 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2884 			INPLOOKUP_RLOCKPCB, NULL);
2885 	NET_EPOCH_EXIT(et);
2886 	if (inp != NULL) {
2887 		if (inp->inp_socket == NULL)
2888 			error = ENOENT;
2889 		if (error == 0)
2890 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2891 		if (error == 0)
2892 			cru2x(inp->inp_cred, &xuc);
2893 		INP_RUNLOCK(inp);
2894 	} else
2895 		error = ENOENT;
2896 	if (error == 0)
2897 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2898 	return (error);
2899 }
2900 
2901 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2902     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2903     0, 0, tcp6_getcred, "S,xucred",
2904     "Get the xucred of a TCP6 connection");
2905 #endif /* INET6 */
2906 
2907 #ifdef INET
2908 /* Path MTU to try next when a fragmentation-needed message is received. */
2909 static inline int
2910 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2911 {
2912 	int mtu = ntohs(icp->icmp_nextmtu);
2913 
2914 	/* If no alternative MTU was proposed, try the next smaller one. */
2915 	if (!mtu)
2916 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2917 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2918 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2919 
2920 	return (mtu);
2921 }
2922 
2923 static void
2924 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2925 {
2926 	struct ip *ip = vip;
2927 	struct tcphdr *th;
2928 	struct in_addr faddr;
2929 	struct inpcb *inp;
2930 	struct tcpcb *tp;
2931 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2932 	struct icmp *icp;
2933 	struct in_conninfo inc;
2934 	tcp_seq icmp_tcp_seq;
2935 	int mtu;
2936 
2937 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2938 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2939 		return;
2940 
2941 	if (cmd == PRC_MSGSIZE)
2942 		notify = tcp_mtudisc_notify;
2943 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2944 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2945 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2946 		notify = tcp_drop_syn_sent;
2947 
2948 	/*
2949 	 * Hostdead is ugly because it goes linearly through all PCBs.
2950 	 * XXX: We never get this from ICMP, otherwise it makes an
2951 	 * excellent DoS attack on machines with many connections.
2952 	 */
2953 	else if (cmd == PRC_HOSTDEAD)
2954 		ip = NULL;
2955 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2956 		return;
2957 
2958 	if (ip == NULL) {
2959 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2960 		return;
2961 	}
2962 
2963 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2964 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2965 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2966 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2967 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2968 		/* signal EHOSTDOWN, as it flushes the cached route */
2969 		inp = (*notify)(inp, EHOSTDOWN);
2970 		goto out;
2971 	}
2972 	icmp_tcp_seq = th->th_seq;
2973 	if (inp != NULL)  {
2974 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2975 		    !(inp->inp_flags & INP_DROPPED) &&
2976 		    !(inp->inp_socket == NULL)) {
2977 			tp = intotcpcb(inp);
2978 #ifdef TCP_OFFLOAD
2979 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2980 				/*
2981 				 * MTU discovery for offloaded connections.  Let
2982 				 * the TOE driver verify seq# and process it.
2983 				 */
2984 				mtu = tcp_next_pmtu(icp, ip);
2985 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2986 				goto out;
2987 			}
2988 #endif
2989 			if (tp->t_port != port) {
2990 				goto out;
2991 			}
2992 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2993 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2994 				if (cmd == PRC_MSGSIZE) {
2995 					/*
2996 					 * MTU discovery: we got a needfrag and
2997 					 * will potentially try a lower MTU.
2998 					 */
2999 					mtu = tcp_next_pmtu(icp, ip);
3000 
3001 					/*
3002 					 * Only process the offered MTU if it
3003 					 * is smaller than the current one.
3004 					 */
3005 					if (mtu < tp->t_maxseg +
3006 					    sizeof(struct tcpiphdr)) {
3007 						bzero(&inc, sizeof(inc));
3008 						inc.inc_faddr = faddr;
3009 						inc.inc_fibnum =
3010 						    inp->inp_inc.inc_fibnum;
3011 						tcp_hc_updatemtu(&inc, mtu);
3012 						inp = tcp_mtudisc(inp, mtu);
3013 					}
3014 				} else
3015 					inp = (*notify)(inp,
3016 					    inetctlerrmap[cmd]);
3017 			}
3018 		}
3019 	} else {
3020 		bzero(&inc, sizeof(inc));
3021 		inc.inc_fport = th->th_dport;
3022 		inc.inc_lport = th->th_sport;
3023 		inc.inc_faddr = faddr;
3024 		inc.inc_laddr = ip->ip_src;
3025 		syncache_unreach(&inc, icmp_tcp_seq, port);
3026 	}
3027 out:
3028 	if (inp != NULL)
3029 		INP_WUNLOCK(inp);
3030 }
3031 
3032 void
3033 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
3034 {
3035 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
3036 }
3037 
3038 void
3039 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
3040 {
3041 	/* Its a tunneled TCP over UDP icmp */
3042 	struct ip *outer_ip, *inner_ip;
3043 	struct icmp *icmp;
3044 	struct udphdr *udp;
3045 	struct tcphdr *th, ttemp;
3046 	int i_hlen, o_len;
3047 	uint16_t port;
3048 
3049 	inner_ip = (struct ip *)vip;
3050 	icmp = (struct icmp *)((caddr_t)inner_ip -
3051 	    (sizeof(struct icmp) - sizeof(struct ip)));
3052 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
3053 	i_hlen = inner_ip->ip_hl << 2;
3054 	o_len = ntohs(outer_ip->ip_len);
3055 	if (o_len <
3056 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
3057 		/* Not enough data present */
3058 		return;
3059 	}
3060 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
3061 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
3062 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3063 		return;
3064 	}
3065 	port = udp->uh_dport;
3066 	th = (struct tcphdr *)(udp + 1);
3067 	memcpy(&ttemp, th, sizeof(struct tcphdr));
3068 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
3069 	/* Now adjust down the size of the outer IP header */
3070 	o_len -= sizeof(struct udphdr);
3071 	outer_ip->ip_len = htons(o_len);
3072 	/* Now call in to the normal handling code */
3073 	tcp_ctlinput_with_port(cmd, sa, vip, port);
3074 }
3075 #endif /* INET */
3076 
3077 #ifdef INET6
3078 static inline int
3079 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
3080 {
3081 	int mtu = ntohl(icmp6->icmp6_mtu);
3082 
3083 	/*
3084 	 * If no alternative MTU was proposed, or the proposed MTU was too
3085 	 * small, set to the min.
3086 	 */
3087 	if (mtu < IPV6_MMTU)
3088 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
3089 	return (mtu);
3090 }
3091 
3092 static void
3093 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3094 {
3095 	struct in6_addr *dst;
3096 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3097 	struct ip6_hdr *ip6;
3098 	struct mbuf *m;
3099 	struct inpcb *inp;
3100 	struct tcpcb *tp;
3101 	struct icmp6_hdr *icmp6;
3102 	struct ip6ctlparam *ip6cp = NULL;
3103 	const struct sockaddr_in6 *sa6_src = NULL;
3104 	struct in_conninfo inc;
3105 	struct tcp_ports {
3106 		uint16_t th_sport;
3107 		uint16_t th_dport;
3108 	} t_ports;
3109 	tcp_seq icmp_tcp_seq;
3110 	unsigned int mtu;
3111 	unsigned int off;
3112 
3113 	if (sa->sa_family != AF_INET6 ||
3114 	    sa->sa_len != sizeof(struct sockaddr_in6))
3115 		return;
3116 
3117 	/* if the parameter is from icmp6, decode it. */
3118 	if (d != NULL) {
3119 		ip6cp = (struct ip6ctlparam *)d;
3120 		icmp6 = ip6cp->ip6c_icmp6;
3121 		m = ip6cp->ip6c_m;
3122 		ip6 = ip6cp->ip6c_ip6;
3123 		off = ip6cp->ip6c_off;
3124 		sa6_src = ip6cp->ip6c_src;
3125 		dst = ip6cp->ip6c_finaldst;
3126 	} else {
3127 		m = NULL;
3128 		ip6 = NULL;
3129 		off = 0;	/* fool gcc */
3130 		sa6_src = &sa6_any;
3131 		dst = NULL;
3132 	}
3133 
3134 	if (cmd == PRC_MSGSIZE)
3135 		notify = tcp_mtudisc_notify;
3136 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3137 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3138 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3139 		notify = tcp_drop_syn_sent;
3140 
3141 	/*
3142 	 * Hostdead is ugly because it goes linearly through all PCBs.
3143 	 * XXX: We never get this from ICMP, otherwise it makes an
3144 	 * excellent DoS attack on machines with many connections.
3145 	 */
3146 	else if (cmd == PRC_HOSTDEAD)
3147 		ip6 = NULL;
3148 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3149 		return;
3150 
3151 	if (ip6 == NULL) {
3152 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3153 			      (const struct sockaddr *)sa6_src,
3154 			      0, cmd, NULL, notify);
3155 		return;
3156 	}
3157 
3158 	/* Check if we can safely get the ports from the tcp hdr */
3159 	if (m == NULL ||
3160 	    (m->m_pkthdr.len <
3161 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3162 		return;
3163 	}
3164 	bzero(&t_ports, sizeof(struct tcp_ports));
3165 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3166 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3167 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3168 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3169 		/* signal EHOSTDOWN, as it flushes the cached route */
3170 		inp = (*notify)(inp, EHOSTDOWN);
3171 		goto out;
3172 	}
3173 	off += sizeof(struct tcp_ports);
3174 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3175 		goto out;
3176 	}
3177 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3178 	if (inp != NULL)  {
3179 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3180 		    !(inp->inp_flags & INP_DROPPED) &&
3181 		    !(inp->inp_socket == NULL)) {
3182 			tp = intotcpcb(inp);
3183 #ifdef TCP_OFFLOAD
3184 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3185 				/* MTU discovery for offloaded connections. */
3186 				mtu = tcp6_next_pmtu(icmp6);
3187 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3188 				goto out;
3189 			}
3190 #endif
3191 			if (tp->t_port != port) {
3192 				goto out;
3193 			}
3194 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3195 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3196 				if (cmd == PRC_MSGSIZE) {
3197 					/*
3198 					 * MTU discovery:
3199 					 * If we got a needfrag set the MTU
3200 					 * in the route to the suggested new
3201 					 * value (if given) and then notify.
3202 					 */
3203 					mtu = tcp6_next_pmtu(icmp6);
3204 
3205 					bzero(&inc, sizeof(inc));
3206 					inc.inc_fibnum = M_GETFIB(m);
3207 					inc.inc_flags |= INC_ISIPV6;
3208 					inc.inc6_faddr = *dst;
3209 					if (in6_setscope(&inc.inc6_faddr,
3210 						m->m_pkthdr.rcvif, NULL))
3211 						goto out;
3212 					/*
3213 					 * Only process the offered MTU if it
3214 					 * is smaller than the current one.
3215 					 */
3216 					if (mtu < tp->t_maxseg +
3217 					    sizeof (struct tcphdr) +
3218 					    sizeof (struct ip6_hdr)) {
3219 						tcp_hc_updatemtu(&inc, mtu);
3220 						tcp_mtudisc(inp, mtu);
3221 						ICMP6STAT_INC(icp6s_pmtuchg);
3222 					}
3223 				} else
3224 					inp = (*notify)(inp,
3225 					    inet6ctlerrmap[cmd]);
3226 			}
3227 		}
3228 	} else {
3229 		bzero(&inc, sizeof(inc));
3230 		inc.inc_fibnum = M_GETFIB(m);
3231 		inc.inc_flags |= INC_ISIPV6;
3232 		inc.inc_fport = t_ports.th_dport;
3233 		inc.inc_lport = t_ports.th_sport;
3234 		inc.inc6_faddr = *dst;
3235 		inc.inc6_laddr = ip6->ip6_src;
3236 		syncache_unreach(&inc, icmp_tcp_seq, port);
3237 	}
3238 out:
3239 	if (inp != NULL)
3240 		INP_WUNLOCK(inp);
3241 }
3242 
3243 void
3244 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3245 {
3246 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3247 }
3248 
3249 void
3250 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3251 {
3252 	struct ip6ctlparam *ip6cp;
3253 	struct mbuf *m;
3254 	struct udphdr *udp;
3255 	uint16_t port;
3256 
3257 	ip6cp = (struct ip6ctlparam *)d;
3258 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3259 	if (m == NULL) {
3260 		return;
3261 	}
3262 	udp = mtod(m, struct udphdr *);
3263 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3264 		return;
3265 	}
3266 	port = udp->uh_dport;
3267 	m_adj(m, sizeof(struct udphdr));
3268 	if ((m->m_flags & M_PKTHDR) == 0) {
3269 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3270 	}
3271 	/* Now call in to the normal handling code */
3272 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3273 }
3274 
3275 #endif /* INET6 */
3276 
3277 static uint32_t
3278 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3279 {
3280 	SIPHASH_CTX ctx;
3281 	uint32_t hash[2];
3282 
3283 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3284 	    ("%s: keylen %u too short ", __func__, len));
3285 	SipHash24_Init(&ctx);
3286 	SipHash_SetKey(&ctx, (uint8_t *)key);
3287 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3288 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3289 	switch (inc->inc_flags & INC_ISIPV6) {
3290 #ifdef INET
3291 	case 0:
3292 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3293 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3294 		break;
3295 #endif
3296 #ifdef INET6
3297 	case INC_ISIPV6:
3298 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3299 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3300 		break;
3301 #endif
3302 	}
3303 	SipHash_Final((uint8_t *)hash, &ctx);
3304 
3305 	return (hash[0] ^ hash[1]);
3306 }
3307 
3308 uint32_t
3309 tcp_new_ts_offset(struct in_conninfo *inc)
3310 {
3311 	struct in_conninfo inc_store, *local_inc;
3312 
3313 	if (!V_tcp_ts_offset_per_conn) {
3314 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3315 		inc_store.inc_lport = 0;
3316 		inc_store.inc_fport = 0;
3317 		local_inc = &inc_store;
3318 	} else {
3319 		local_inc = inc;
3320 	}
3321 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3322 	    sizeof(V_ts_offset_secret)));
3323 }
3324 
3325 /*
3326  * Following is where TCP initial sequence number generation occurs.
3327  *
3328  * There are two places where we must use initial sequence numbers:
3329  * 1.  In SYN-ACK packets.
3330  * 2.  In SYN packets.
3331  *
3332  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3333  * tcp_syncache.c for details.
3334  *
3335  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3336  * depends on this property.  In addition, these ISNs should be
3337  * unguessable so as to prevent connection hijacking.  To satisfy
3338  * the requirements of this situation, the algorithm outlined in
3339  * RFC 1948 is used, with only small modifications.
3340  *
3341  * Implementation details:
3342  *
3343  * Time is based off the system timer, and is corrected so that it
3344  * increases by one megabyte per second.  This allows for proper
3345  * recycling on high speed LANs while still leaving over an hour
3346  * before rollover.
3347  *
3348  * As reading the *exact* system time is too expensive to be done
3349  * whenever setting up a TCP connection, we increment the time
3350  * offset in two ways.  First, a small random positive increment
3351  * is added to isn_offset for each connection that is set up.
3352  * Second, the function tcp_isn_tick fires once per clock tick
3353  * and increments isn_offset as necessary so that sequence numbers
3354  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3355  * random positive increments serve only to ensure that the same
3356  * exact sequence number is never sent out twice (as could otherwise
3357  * happen when a port is recycled in less than the system tick
3358  * interval.)
3359  *
3360  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3361  * between seeding of isn_secret.  This is normally set to zero,
3362  * as reseeding should not be necessary.
3363  *
3364  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3365  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3366  * general, this means holding an exclusive (write) lock.
3367  */
3368 
3369 #define ISN_BYTES_PER_SECOND 1048576
3370 #define ISN_STATIC_INCREMENT 4096
3371 #define ISN_RANDOM_INCREMENT (4096 - 1)
3372 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3373 
3374 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3375 VNET_DEFINE_STATIC(int, isn_last);
3376 VNET_DEFINE_STATIC(int, isn_last_reseed);
3377 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3378 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3379 
3380 #define	V_isn_secret			VNET(isn_secret)
3381 #define	V_isn_last			VNET(isn_last)
3382 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3383 #define	V_isn_offset			VNET(isn_offset)
3384 #define	V_isn_offset_old		VNET(isn_offset_old)
3385 
3386 tcp_seq
3387 tcp_new_isn(struct in_conninfo *inc)
3388 {
3389 	tcp_seq new_isn;
3390 	u_int32_t projected_offset;
3391 
3392 	ISN_LOCK();
3393 	/* Seed if this is the first use, reseed if requested. */
3394 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3395 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3396 		< (u_int)ticks))) {
3397 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3398 		V_isn_last_reseed = ticks;
3399 	}
3400 
3401 	/* Compute the hash and return the ISN. */
3402 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3403 	    sizeof(V_isn_secret));
3404 	V_isn_offset += ISN_STATIC_INCREMENT +
3405 		(arc4random() & ISN_RANDOM_INCREMENT);
3406 	if (ticks != V_isn_last) {
3407 		projected_offset = V_isn_offset_old +
3408 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3409 		if (SEQ_GT(projected_offset, V_isn_offset))
3410 			V_isn_offset = projected_offset;
3411 		V_isn_offset_old = V_isn_offset;
3412 		V_isn_last = ticks;
3413 	}
3414 	new_isn += V_isn_offset;
3415 	ISN_UNLOCK();
3416 	return (new_isn);
3417 }
3418 
3419 /*
3420  * When a specific ICMP unreachable message is received and the
3421  * connection state is SYN-SENT, drop the connection.  This behavior
3422  * is controlled by the icmp_may_rst sysctl.
3423  */
3424 struct inpcb *
3425 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3426 {
3427 	struct tcpcb *tp;
3428 
3429 	NET_EPOCH_ASSERT();
3430 	INP_WLOCK_ASSERT(inp);
3431 
3432 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3433 	    (inp->inp_flags & INP_DROPPED))
3434 		return (inp);
3435 
3436 	tp = intotcpcb(inp);
3437 	if (tp->t_state != TCPS_SYN_SENT)
3438 		return (inp);
3439 
3440 	if (IS_FASTOPEN(tp->t_flags))
3441 		tcp_fastopen_disable_path(tp);
3442 
3443 	tp = tcp_drop(tp, errno);
3444 	if (tp != NULL)
3445 		return (inp);
3446 	else
3447 		return (NULL);
3448 }
3449 
3450 /*
3451  * When `need fragmentation' ICMP is received, update our idea of the MSS
3452  * based on the new value. Also nudge TCP to send something, since we
3453  * know the packet we just sent was dropped.
3454  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3455  */
3456 static struct inpcb *
3457 tcp_mtudisc_notify(struct inpcb *inp, int error)
3458 {
3459 
3460 	return (tcp_mtudisc(inp, -1));
3461 }
3462 
3463 static struct inpcb *
3464 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3465 {
3466 	struct tcpcb *tp;
3467 	struct socket *so;
3468 
3469 	INP_WLOCK_ASSERT(inp);
3470 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3471 	    (inp->inp_flags & INP_DROPPED))
3472 		return (inp);
3473 
3474 	tp = intotcpcb(inp);
3475 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3476 
3477 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3478 
3479 	so = inp->inp_socket;
3480 	SOCKBUF_LOCK(&so->so_snd);
3481 	/* If the mss is larger than the socket buffer, decrease the mss. */
3482 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3483 		tp->t_maxseg = so->so_snd.sb_hiwat;
3484 	SOCKBUF_UNLOCK(&so->so_snd);
3485 
3486 	TCPSTAT_INC(tcps_mturesent);
3487 	tp->t_rtttime = 0;
3488 	tp->snd_nxt = tp->snd_una;
3489 	tcp_free_sackholes(tp);
3490 	tp->snd_recover = tp->snd_max;
3491 	if (tp->t_flags & TF_SACK_PERMIT)
3492 		EXIT_FASTRECOVERY(tp->t_flags);
3493 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3494 		/*
3495 		 * Conceptually the snd_nxt setting
3496 		 * and freeing sack holes should
3497 		 * be done by the default stacks
3498 		 * own tfb_tcp_mtu_chg().
3499 		 */
3500 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3501 	}
3502 	if (tcp_output(tp) < 0)
3503 		return (NULL);
3504 	else
3505 		return (inp);
3506 }
3507 
3508 #ifdef INET
3509 /*
3510  * Look-up the routing entry to the peer of this inpcb.  If no route
3511  * is found and it cannot be allocated, then return 0.  This routine
3512  * is called by TCP routines that access the rmx structure and by
3513  * tcp_mss_update to get the peer/interface MTU.
3514  */
3515 uint32_t
3516 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3517 {
3518 	struct nhop_object *nh;
3519 	struct ifnet *ifp;
3520 	uint32_t maxmtu = 0;
3521 
3522 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3523 
3524 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3525 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3526 		if (nh == NULL)
3527 			return (0);
3528 
3529 		ifp = nh->nh_ifp;
3530 		maxmtu = nh->nh_mtu;
3531 
3532 		/* Report additional interface capabilities. */
3533 		if (cap != NULL) {
3534 			if (ifp->if_capenable & IFCAP_TSO4 &&
3535 			    ifp->if_hwassist & CSUM_TSO) {
3536 				cap->ifcap |= CSUM_TSO;
3537 				cap->tsomax = ifp->if_hw_tsomax;
3538 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3539 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3540 			}
3541 		}
3542 	}
3543 	return (maxmtu);
3544 }
3545 #endif /* INET */
3546 
3547 #ifdef INET6
3548 uint32_t
3549 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3550 {
3551 	struct nhop_object *nh;
3552 	struct in6_addr dst6;
3553 	uint32_t scopeid;
3554 	struct ifnet *ifp;
3555 	uint32_t maxmtu = 0;
3556 
3557 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3558 
3559 	if (inc->inc_flags & INC_IPV6MINMTU)
3560 		return (IPV6_MMTU);
3561 
3562 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3563 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3564 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3565 		if (nh == NULL)
3566 			return (0);
3567 
3568 		ifp = nh->nh_ifp;
3569 		maxmtu = nh->nh_mtu;
3570 
3571 		/* Report additional interface capabilities. */
3572 		if (cap != NULL) {
3573 			if (ifp->if_capenable & IFCAP_TSO6 &&
3574 			    ifp->if_hwassist & CSUM_TSO) {
3575 				cap->ifcap |= CSUM_TSO;
3576 				cap->tsomax = ifp->if_hw_tsomax;
3577 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3578 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3579 			}
3580 		}
3581 	}
3582 
3583 	return (maxmtu);
3584 }
3585 
3586 /*
3587  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3588  *
3589  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3590  * The right place to do that is ip6_setpktopt() that has just been
3591  * executed.  By the way it just filled ip6po_minmtu for us.
3592  */
3593 void
3594 tcp6_use_min_mtu(struct tcpcb *tp)
3595 {
3596 	struct inpcb *inp = tp->t_inpcb;
3597 
3598 	INP_WLOCK_ASSERT(inp);
3599 	/*
3600 	 * In case of the IPV6_USE_MIN_MTU socket
3601 	 * option, the INC_IPV6MINMTU flag to announce
3602 	 * a corresponding MSS during the initial
3603 	 * handshake.  If the TCP connection is not in
3604 	 * the front states, just reduce the MSS being
3605 	 * used.  This avoids the sending of TCP
3606 	 * segments which will be fragmented at the
3607 	 * IPv6 layer.
3608 	 */
3609 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3610 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3611 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3612 		struct ip6_pktopts *opt;
3613 
3614 		opt = inp->in6p_outputopts;
3615 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3616 		    tp->t_maxseg > TCP6_MSS)
3617 			tp->t_maxseg = TCP6_MSS;
3618 	}
3619 }
3620 #endif /* INET6 */
3621 
3622 /*
3623  * Calculate effective SMSS per RFC5681 definition for a given TCP
3624  * connection at its current state, taking into account SACK and etc.
3625  */
3626 u_int
3627 tcp_maxseg(const struct tcpcb *tp)
3628 {
3629 	u_int optlen;
3630 
3631 	if (tp->t_flags & TF_NOOPT)
3632 		return (tp->t_maxseg);
3633 
3634 	/*
3635 	 * Here we have a simplified code from tcp_addoptions(),
3636 	 * without a proper loop, and having most of paddings hardcoded.
3637 	 * We might make mistakes with padding here in some edge cases,
3638 	 * but this is harmless, since result of tcp_maxseg() is used
3639 	 * only in cwnd and ssthresh estimations.
3640 	 */
3641 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3642 		if (tp->t_flags & TF_RCVD_TSTMP)
3643 			optlen = TCPOLEN_TSTAMP_APPA;
3644 		else
3645 			optlen = 0;
3646 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3647 		if (tp->t_flags & TF_SIGNATURE)
3648 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3649 #endif
3650 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3651 			optlen += TCPOLEN_SACKHDR;
3652 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3653 			optlen = PADTCPOLEN(optlen);
3654 		}
3655 	} else {
3656 		if (tp->t_flags & TF_REQ_TSTMP)
3657 			optlen = TCPOLEN_TSTAMP_APPA;
3658 		else
3659 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3660 		if (tp->t_flags & TF_REQ_SCALE)
3661 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3662 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3663 		if (tp->t_flags & TF_SIGNATURE)
3664 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3665 #endif
3666 		if (tp->t_flags & TF_SACK_PERMIT)
3667 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3668 	}
3669 #undef PAD
3670 	optlen = min(optlen, TCP_MAXOLEN);
3671 	return (tp->t_maxseg - optlen);
3672 }
3673 
3674 
3675 u_int
3676 tcp_fixed_maxseg(const struct tcpcb *tp)
3677 {
3678 	int optlen;
3679 
3680 	if (tp->t_flags & TF_NOOPT)
3681 		return (tp->t_maxseg);
3682 
3683 	/*
3684 	 * Here we have a simplified code from tcp_addoptions(),
3685 	 * without a proper loop, and having most of paddings hardcoded.
3686 	 * We only consider fixed options that we would send every
3687 	 * time I.e. SACK is not considered. This is important
3688 	 * for cc modules to figure out what the modulo of the
3689 	 * cwnd should be.
3690 	 */
3691 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3692 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3693 		if (tp->t_flags & TF_RCVD_TSTMP)
3694 			optlen = TCPOLEN_TSTAMP_APPA;
3695 		else
3696 			optlen = 0;
3697 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3698 		if (tp->t_flags & TF_SIGNATURE)
3699 			optlen += PAD(TCPOLEN_SIGNATURE);
3700 #endif
3701 	} else {
3702 		if (tp->t_flags & TF_REQ_TSTMP)
3703 			optlen = TCPOLEN_TSTAMP_APPA;
3704 		else
3705 			optlen = PAD(TCPOLEN_MAXSEG);
3706 		if (tp->t_flags & TF_REQ_SCALE)
3707 			optlen += PAD(TCPOLEN_WINDOW);
3708 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3709 		if (tp->t_flags & TF_SIGNATURE)
3710 			optlen += PAD(TCPOLEN_SIGNATURE);
3711 #endif
3712 		if (tp->t_flags & TF_SACK_PERMIT)
3713 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3714 	}
3715 #undef PAD
3716 	optlen = min(optlen, TCP_MAXOLEN);
3717 	return (tp->t_maxseg - optlen);
3718 }
3719 
3720 
3721 
3722 static int
3723 sysctl_drop(SYSCTL_HANDLER_ARGS)
3724 {
3725 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3726 	struct sockaddr_storage addrs[2];
3727 	struct inpcb *inp;
3728 	struct tcpcb *tp;
3729 	struct tcptw *tw;
3730 	struct sockaddr_in *fin, *lin;
3731 	struct epoch_tracker et;
3732 #ifdef INET6
3733 	struct sockaddr_in6 *fin6, *lin6;
3734 #endif
3735 	int error;
3736 
3737 	inp = NULL;
3738 	fin = lin = NULL;
3739 #ifdef INET6
3740 	fin6 = lin6 = NULL;
3741 #endif
3742 	error = 0;
3743 
3744 	if (req->oldptr != NULL || req->oldlen != 0)
3745 		return (EINVAL);
3746 	if (req->newptr == NULL)
3747 		return (EPERM);
3748 	if (req->newlen < sizeof(addrs))
3749 		return (ENOMEM);
3750 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3751 	if (error)
3752 		return (error);
3753 
3754 	switch (addrs[0].ss_family) {
3755 #ifdef INET6
3756 	case AF_INET6:
3757 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3758 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3759 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3760 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3761 			return (EINVAL);
3762 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3763 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3764 				return (EINVAL);
3765 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3766 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3767 			fin = (struct sockaddr_in *)&addrs[0];
3768 			lin = (struct sockaddr_in *)&addrs[1];
3769 			break;
3770 		}
3771 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3772 		if (error)
3773 			return (error);
3774 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3775 		if (error)
3776 			return (error);
3777 		break;
3778 #endif
3779 #ifdef INET
3780 	case AF_INET:
3781 		fin = (struct sockaddr_in *)&addrs[0];
3782 		lin = (struct sockaddr_in *)&addrs[1];
3783 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3784 		    lin->sin_len != sizeof(struct sockaddr_in))
3785 			return (EINVAL);
3786 		break;
3787 #endif
3788 	default:
3789 		return (EINVAL);
3790 	}
3791 	NET_EPOCH_ENTER(et);
3792 	switch (addrs[0].ss_family) {
3793 #ifdef INET6
3794 	case AF_INET6:
3795 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3796 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3797 		    INPLOOKUP_WLOCKPCB, NULL);
3798 		break;
3799 #endif
3800 #ifdef INET
3801 	case AF_INET:
3802 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3803 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3804 		break;
3805 #endif
3806 	}
3807 	if (inp != NULL) {
3808 		if (inp->inp_flags & INP_TIMEWAIT) {
3809 			/*
3810 			 * XXXRW: There currently exists a state where an
3811 			 * inpcb is present, but its timewait state has been
3812 			 * discarded.  For now, don't allow dropping of this
3813 			 * type of inpcb.
3814 			 */
3815 			tw = intotw(inp);
3816 			if (tw != NULL)
3817 				tcp_twclose(tw, 0);
3818 			else
3819 				INP_WUNLOCK(inp);
3820 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3821 		    !SOLISTENING(inp->inp_socket)) {
3822 			tp = intotcpcb(inp);
3823 			tp = tcp_drop(tp, ECONNABORTED);
3824 			if (tp != NULL)
3825 				INP_WUNLOCK(inp);
3826 		} else
3827 			INP_WUNLOCK(inp);
3828 	} else
3829 		error = ESRCH;
3830 	NET_EPOCH_EXIT(et);
3831 	return (error);
3832 }
3833 
3834 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3835     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3836     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3837     "Drop TCP connection");
3838 
3839 #ifdef KERN_TLS
3840 static int
3841 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3842 {
3843 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3844 	struct sockaddr_storage addrs[2];
3845 	struct inpcb *inp;
3846 	struct sockaddr_in *fin, *lin;
3847 	struct epoch_tracker et;
3848 #ifdef INET6
3849 	struct sockaddr_in6 *fin6, *lin6;
3850 #endif
3851 	int error;
3852 
3853 	inp = NULL;
3854 	fin = lin = NULL;
3855 #ifdef INET6
3856 	fin6 = lin6 = NULL;
3857 #endif
3858 	error = 0;
3859 
3860 	if (req->oldptr != NULL || req->oldlen != 0)
3861 		return (EINVAL);
3862 	if (req->newptr == NULL)
3863 		return (EPERM);
3864 	if (req->newlen < sizeof(addrs))
3865 		return (ENOMEM);
3866 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3867 	if (error)
3868 		return (error);
3869 
3870 	switch (addrs[0].ss_family) {
3871 #ifdef INET6
3872 	case AF_INET6:
3873 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3874 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3875 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3876 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3877 			return (EINVAL);
3878 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3879 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3880 				return (EINVAL);
3881 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3882 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3883 			fin = (struct sockaddr_in *)&addrs[0];
3884 			lin = (struct sockaddr_in *)&addrs[1];
3885 			break;
3886 		}
3887 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3888 		if (error)
3889 			return (error);
3890 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3891 		if (error)
3892 			return (error);
3893 		break;
3894 #endif
3895 #ifdef INET
3896 	case AF_INET:
3897 		fin = (struct sockaddr_in *)&addrs[0];
3898 		lin = (struct sockaddr_in *)&addrs[1];
3899 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3900 		    lin->sin_len != sizeof(struct sockaddr_in))
3901 			return (EINVAL);
3902 		break;
3903 #endif
3904 	default:
3905 		return (EINVAL);
3906 	}
3907 	NET_EPOCH_ENTER(et);
3908 	switch (addrs[0].ss_family) {
3909 #ifdef INET6
3910 	case AF_INET6:
3911 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3912 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3913 		    INPLOOKUP_WLOCKPCB, NULL);
3914 		break;
3915 #endif
3916 #ifdef INET
3917 	case AF_INET:
3918 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3919 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3920 		break;
3921 #endif
3922 	}
3923 	NET_EPOCH_EXIT(et);
3924 	if (inp != NULL) {
3925 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3926 		    inp->inp_socket == NULL) {
3927 			error = ECONNRESET;
3928 			INP_WUNLOCK(inp);
3929 		} else {
3930 			struct socket *so;
3931 
3932 			so = inp->inp_socket;
3933 			soref(so);
3934 			error = ktls_set_tx_mode(so,
3935 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3936 			INP_WUNLOCK(inp);
3937 			sorele(so);
3938 		}
3939 	} else
3940 		error = ESRCH;
3941 	return (error);
3942 }
3943 
3944 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3945     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3946     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3947     "Switch TCP connection to SW TLS");
3948 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3949     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3950     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3951     "Switch TCP connection to ifnet TLS");
3952 #endif
3953 
3954 /*
3955  * Generate a standardized TCP log line for use throughout the
3956  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3957  * allow use in the interrupt context.
3958  *
3959  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3960  * NB: The function may return NULL if memory allocation failed.
3961  *
3962  * Due to header inclusion and ordering limitations the struct ip
3963  * and ip6_hdr pointers have to be passed as void pointers.
3964  */
3965 char *
3966 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3967     const void *ip6hdr)
3968 {
3969 
3970 	/* Is logging enabled? */
3971 	if (V_tcp_log_in_vain == 0)
3972 		return (NULL);
3973 
3974 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3975 }
3976 
3977 char *
3978 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3979     const void *ip6hdr)
3980 {
3981 
3982 	/* Is logging enabled? */
3983 	if (tcp_log_debug == 0)
3984 		return (NULL);
3985 
3986 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3987 }
3988 
3989 static char *
3990 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3991     const void *ip6hdr)
3992 {
3993 	char *s, *sp;
3994 	size_t size;
3995 	struct ip *ip;
3996 #ifdef INET6
3997 	const struct ip6_hdr *ip6;
3998 
3999 	ip6 = (const struct ip6_hdr *)ip6hdr;
4000 #endif /* INET6 */
4001 	ip = (struct ip *)ip4hdr;
4002 
4003 	/*
4004 	 * The log line looks like this:
4005 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
4006 	 */
4007 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
4008 	    sizeof(PRINT_TH_FLAGS) + 1 +
4009 #ifdef INET6
4010 	    2 * INET6_ADDRSTRLEN;
4011 #else
4012 	    2 * INET_ADDRSTRLEN;
4013 #endif /* INET6 */
4014 
4015 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
4016 	if (s == NULL)
4017 		return (NULL);
4018 
4019 	strcat(s, "TCP: [");
4020 	sp = s + strlen(s);
4021 
4022 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
4023 		inet_ntoa_r(inc->inc_faddr, sp);
4024 		sp = s + strlen(s);
4025 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4026 		sp = s + strlen(s);
4027 		inet_ntoa_r(inc->inc_laddr, sp);
4028 		sp = s + strlen(s);
4029 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4030 #ifdef INET6
4031 	} else if (inc) {
4032 		ip6_sprintf(sp, &inc->inc6_faddr);
4033 		sp = s + strlen(s);
4034 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
4035 		sp = s + strlen(s);
4036 		ip6_sprintf(sp, &inc->inc6_laddr);
4037 		sp = s + strlen(s);
4038 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
4039 	} else if (ip6 && th) {
4040 		ip6_sprintf(sp, &ip6->ip6_src);
4041 		sp = s + strlen(s);
4042 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4043 		sp = s + strlen(s);
4044 		ip6_sprintf(sp, &ip6->ip6_dst);
4045 		sp = s + strlen(s);
4046 		sprintf(sp, "]:%i", ntohs(th->th_dport));
4047 #endif /* INET6 */
4048 #ifdef INET
4049 	} else if (ip && th) {
4050 		inet_ntoa_r(ip->ip_src, sp);
4051 		sp = s + strlen(s);
4052 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
4053 		sp = s + strlen(s);
4054 		inet_ntoa_r(ip->ip_dst, sp);
4055 		sp = s + strlen(s);
4056 		sprintf(sp, "]:%i", ntohs(th->th_dport));
4057 #endif /* INET */
4058 	} else {
4059 		free(s, M_TCPLOG);
4060 		return (NULL);
4061 	}
4062 	sp = s + strlen(s);
4063 	if (th)
4064 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
4065 	if (*(s + size - 1) != '\0')
4066 		panic("%s: string too long", __func__);
4067 	return (s);
4068 }
4069 
4070 /*
4071  * A subroutine which makes it easy to track TCP state changes with DTrace.
4072  * This function shouldn't be called for t_state initializations that don't
4073  * correspond to actual TCP state transitions.
4074  */
4075 void
4076 tcp_state_change(struct tcpcb *tp, int newstate)
4077 {
4078 #if defined(KDTRACE_HOOKS)
4079 	int pstate = tp->t_state;
4080 #endif
4081 
4082 	TCPSTATES_DEC(tp->t_state);
4083 	TCPSTATES_INC(newstate);
4084 	tp->t_state = newstate;
4085 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4086 }
4087 
4088 /*
4089  * Create an external-format (``xtcpcb'') structure using the information in
4090  * the kernel-format tcpcb structure pointed to by tp.  This is done to
4091  * reduce the spew of irrelevant information over this interface, to isolate
4092  * user code from changes in the kernel structure, and potentially to provide
4093  * information-hiding if we decide that some of this information should be
4094  * hidden from users.
4095  */
4096 void
4097 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4098 {
4099 	struct tcpcb *tp = intotcpcb(inp);
4100 	struct tcptw *tw = intotw(inp);
4101 	sbintime_t now;
4102 
4103 	bzero(xt, sizeof(*xt));
4104 	if (inp->inp_flags & INP_TIMEWAIT) {
4105 		xt->t_state = TCPS_TIME_WAIT;
4106 		xt->xt_encaps_port = tw->t_port;
4107 	} else {
4108 		xt->t_state = tp->t_state;
4109 		xt->t_logstate = tp->t_logstate;
4110 		xt->t_flags = tp->t_flags;
4111 		xt->t_sndzerowin = tp->t_sndzerowin;
4112 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4113 		xt->t_rcvoopack = tp->t_rcvoopack;
4114 		xt->t_rcv_wnd = tp->rcv_wnd;
4115 		xt->t_snd_wnd = tp->snd_wnd;
4116 		xt->t_snd_cwnd = tp->snd_cwnd;
4117 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4118 		xt->t_dsack_bytes = tp->t_dsack_bytes;
4119 		xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4120 		xt->t_dsack_pack = tp->t_dsack_pack;
4121 		xt->t_maxseg = tp->t_maxseg;
4122 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4123 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4124 
4125 		now = getsbinuptime();
4126 #define	COPYTIMER(ttt)	do {						\
4127 		if (callout_active(&tp->t_timers->ttt))			\
4128 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4129 			    SBT_1MS;					\
4130 		else							\
4131 			xt->ttt = 0;					\
4132 } while (0)
4133 		COPYTIMER(tt_delack);
4134 		COPYTIMER(tt_rexmt);
4135 		COPYTIMER(tt_persist);
4136 		COPYTIMER(tt_keep);
4137 		COPYTIMER(tt_2msl);
4138 #undef COPYTIMER
4139 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4140 
4141 		xt->xt_encaps_port = tp->t_port;
4142 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4143 		    TCP_FUNCTION_NAME_LEN_MAX);
4144 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4145 		    TCP_CA_NAME_MAX);
4146 #ifdef TCP_BLACKBOX
4147 		(void)tcp_log_get_id(tp, xt->xt_logid);
4148 #endif
4149 	}
4150 
4151 	xt->xt_len = sizeof(struct xtcpcb);
4152 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4153 	if (inp->inp_socket == NULL)
4154 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4155 }
4156 
4157 void
4158 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4159 {
4160 	uint32_t bit, i;
4161 
4162 	if ((tp == NULL) ||
4163 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4164 	    (status == 0)) {
4165 		/* Invalid */
4166 		return;
4167 	}
4168 	if (status > (sizeof(uint32_t) * 8)) {
4169 		/* Should this be a KASSERT? */
4170 		return;
4171 	}
4172 	bit = 1U << (status - 1);
4173 	if (bit & tp->t_end_info_status) {
4174 		/* already logged */
4175 		return;
4176 	}
4177 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4178 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4179 			tp->t_end_info_bytes[i] = status;
4180 			tp->t_end_info_status |= bit;
4181 			break;
4182 		}
4183 	}
4184 }
4185 
4186 int
4187 tcp_can_enable_pacing(void)
4188 {
4189 
4190 	if ((tcp_pacing_limit == -1) ||
4191 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4192 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4193 		shadow_num_connections = number_of_tcp_connections_pacing;
4194 		return (1);
4195 	} else {
4196 		return (0);
4197 	}
4198 }
4199 
4200 static uint8_t tcp_pacing_warning = 0;
4201 
4202 void
4203 tcp_decrement_paced_conn(void)
4204 {
4205 	uint32_t ret;
4206 
4207 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4208 	shadow_num_connections = number_of_tcp_connections_pacing;
4209 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4210 	if (ret == 0) {
4211 		if (tcp_pacing_limit != -1) {
4212 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4213 			tcp_pacing_limit = 0;
4214 		} else if (tcp_pacing_warning == 0) {
4215 			printf("Warning pacing count is invalid, invalid decrement\n");
4216 			tcp_pacing_warning = 1;
4217 		}
4218 	}
4219 }
4220