xref: /freebsd/sys/netinet/tcp_subr.c (revision ce9de47260d4edc963a94140789e4a52642c28e6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/route/nhop.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
93 #ifdef INET6
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
101 #endif
102 
103 #include <netinet/tcp.h>
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_log_buf.h>
109 #include <netinet/tcp_syncache.h>
110 #include <netinet/tcp_hpts.h>
111 #include <netinet/cc/cc.h>
112 #ifdef INET6
113 #include <netinet6/tcp6_var.h>
114 #endif
115 #include <netinet/tcpip.h>
116 #include <netinet/tcp_fastopen.h>
117 #ifdef TCPPCAP
118 #include <netinet/tcp_pcap.h>
119 #endif
120 #ifdef TCPDEBUG
121 #include <netinet/tcp_debug.h>
122 #endif
123 #ifdef INET6
124 #include <netinet6/ip6protosw.h>
125 #endif
126 #ifdef TCP_OFFLOAD
127 #include <netinet/tcp_offload.h>
128 #endif
129 #include <netinet/udp.h>
130 #include <netinet/udp_var.h>
131 
132 #include <netipsec/ipsec_support.h>
133 
134 #include <machine/in_cksum.h>
135 #include <crypto/siphash/siphash.h>
136 
137 #include <security/mac/mac_framework.h>
138 
139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
140 #ifdef INET6
141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
142 #endif
143 
144 #ifdef NETFLIX_EXP_DETECTION
145 /*  Sack attack detection thresholds and such */
146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
147     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
148     "Sack Attack detection thresholds");
149 int32_t tcp_force_detection = 0;
150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
151     CTLFLAG_RW,
152     &tcp_force_detection, 0,
153     "Do we force detection even if the INP has it off?");
154 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
156     CTLFLAG_RW,
157     &tcp_sack_to_ack_thresh, 700,
158     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
159 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
161     CTLFLAG_RW,
162     &tcp_sack_to_move_thresh, 600,
163     "Percentage of sack moves we must see above (10.1 percent is 101)");
164 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
166     CTLFLAG_RW,
167     &tcp_restoral_thresh, 550,
168     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
169 int32_t tcp_sad_decay_val = 800;
170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
171     CTLFLAG_RW,
172     &tcp_sad_decay_val, 800,
173     "The decay percentage (10.1 percent equals 101 )");
174 int32_t tcp_map_minimum = 500;
175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
176     CTLFLAG_RW,
177     &tcp_map_minimum, 500,
178     "Number of Map enteries before we start detection");
179 int32_t tcp_attack_on_turns_on_logging = 0;
180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
181     CTLFLAG_RW,
182     &tcp_attack_on_turns_on_logging, 0,
183    "When we have a positive hit on attack, do we turn on logging?");
184 int32_t tcp_sad_pacing_interval = 2000;
185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
186     CTLFLAG_RW,
187     &tcp_sad_pacing_interval, 2000,
188     "What is the minimum pacing interval for a classified attacker?");
189 
190 int32_t tcp_sad_low_pps = 100;
191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
192     CTLFLAG_RW,
193     &tcp_sad_low_pps, 100,
194     "What is the input pps that below which we do not decay?");
195 #endif
196 uint32_t tcp_ack_war_time_window = 1000;
197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
198     CTLFLAG_RW,
199     &tcp_ack_war_time_window, 1000,
200    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
201 uint32_t tcp_ack_war_cnt = 5;
202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
203     CTLFLAG_RW,
204     &tcp_ack_war_cnt, 5,
205    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
206 
207 struct rwlock tcp_function_lock;
208 
209 static int
210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
211 {
212 	int error, new;
213 
214 	new = V_tcp_mssdflt;
215 	error = sysctl_handle_int(oidp, &new, 0, req);
216 	if (error == 0 && req->newptr) {
217 		if (new < TCP_MINMSS)
218 			error = EINVAL;
219 		else
220 			V_tcp_mssdflt = new;
221 	}
222 	return (error);
223 }
224 
225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
226     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
227     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
228     "Default TCP Maximum Segment Size");
229 
230 #ifdef INET6
231 static int
232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
233 {
234 	int error, new;
235 
236 	new = V_tcp_v6mssdflt;
237 	error = sysctl_handle_int(oidp, &new, 0, req);
238 	if (error == 0 && req->newptr) {
239 		if (new < TCP_MINMSS)
240 			error = EINVAL;
241 		else
242 			V_tcp_v6mssdflt = new;
243 	}
244 	return (error);
245 }
246 
247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
248     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
250    "Default TCP Maximum Segment Size for IPv6");
251 #endif /* INET6 */
252 
253 /*
254  * Minimum MSS we accept and use. This prevents DoS attacks where
255  * we are forced to a ridiculous low MSS like 20 and send hundreds
256  * of packets instead of one. The effect scales with the available
257  * bandwidth and quickly saturates the CPU and network interface
258  * with packet generation and sending. Set to zero to disable MINMSS
259  * checking. This setting prevents us from sending too small packets.
260  */
261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
263      &VNET_NAME(tcp_minmss), 0,
264     "Minimum TCP Maximum Segment Size");
265 
266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
268     &VNET_NAME(tcp_do_rfc1323), 0,
269     "Enable rfc1323 (high performance TCP) extensions");
270 
271 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 0;
272 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
273     &VNET_NAME(tcp_tolerate_missing_ts), 0,
274     "Tolerate missing TCP timestamps");
275 
276 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
277 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
278     &VNET_NAME(tcp_ts_offset_per_conn), 0,
279     "Initialize TCP timestamps per connection instead of per host pair");
280 
281 /* How many connections are pacing */
282 static volatile uint32_t number_of_tcp_connections_pacing = 0;
283 static uint32_t shadow_num_connections = 0;
284 
285 static int tcp_pacing_limit = 10000;
286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
287     &tcp_pacing_limit, 1000,
288     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
289 
290 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
291     &shadow_num_connections, 0, "Number of TCP connections being paced");
292 
293 static int	tcp_log_debug = 0;
294 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
295     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
296 
297 static int	tcp_tcbhashsize;
298 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
299     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
300 
301 static int	do_tcpdrain = 1;
302 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
303     "Enable tcp_drain routine for extra help when low on mbufs");
304 
305 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
306     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
307 
308 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
309 #define	V_icmp_may_rst			VNET(icmp_may_rst)
310 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
311     &VNET_NAME(icmp_may_rst), 0,
312     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
313 
314 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
315 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
316 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
317     &VNET_NAME(tcp_isn_reseed_interval), 0,
318     "Seconds between reseeding of ISN secret");
319 
320 static int	tcp_soreceive_stream;
321 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
322     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
323 
324 VNET_DEFINE(uma_zone_t, sack_hole_zone);
325 #define	V_sack_hole_zone		VNET(sack_hole_zone)
326 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
327 static int
328 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
329 {
330 	int error;
331 	uint32_t new;
332 
333 	new = V_tcp_map_entries_limit;
334 	error = sysctl_handle_int(oidp, &new, 0, req);
335 	if (error == 0 && req->newptr) {
336 		/* only allow "0" and value > minimum */
337 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
338 			error = EINVAL;
339 		else
340 			V_tcp_map_entries_limit = new;
341 	}
342 	return (error);
343 }
344 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
345     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
346     &VNET_NAME(tcp_map_entries_limit), 0,
347     &sysctl_net_inet_tcp_map_limit_check, "IU",
348     "Total sendmap entries limit");
349 
350 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
351 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
352      &VNET_NAME(tcp_map_split_limit), 0,
353     "Total sendmap split entries limit");
354 
355 #ifdef TCP_HHOOK
356 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
357 #endif
358 
359 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
360 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
361 #define	V_ts_offset_secret	VNET(ts_offset_secret)
362 
363 static int	tcp_default_fb_init(struct tcpcb *tp);
364 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
365 static int	tcp_default_handoff_ok(struct tcpcb *tp);
366 static struct inpcb *tcp_notify(struct inpcb *, int);
367 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
368 static void tcp_mtudisc(struct inpcb *, int);
369 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
370 		    void *ip4hdr, const void *ip6hdr);
371 
372 static struct tcp_function_block tcp_def_funcblk = {
373 	.tfb_tcp_block_name = "freebsd",
374 	.tfb_tcp_output = tcp_output,
375 	.tfb_tcp_do_segment = tcp_do_segment,
376 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
377 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
378 	.tfb_tcp_fb_init = tcp_default_fb_init,
379 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
380 };
381 
382 static int tcp_fb_cnt = 0;
383 struct tcp_funchead t_functions;
384 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
385 
386 static struct tcp_function_block *
387 find_tcp_functions_locked(struct tcp_function_set *fs)
388 {
389 	struct tcp_function *f;
390 	struct tcp_function_block *blk=NULL;
391 
392 	TAILQ_FOREACH(f, &t_functions, tf_next) {
393 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
394 			blk = f->tf_fb;
395 			break;
396 		}
397 	}
398 	return(blk);
399 }
400 
401 static struct tcp_function_block *
402 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
403 {
404 	struct tcp_function_block *rblk=NULL;
405 	struct tcp_function *f;
406 
407 	TAILQ_FOREACH(f, &t_functions, tf_next) {
408 		if (f->tf_fb == blk) {
409 			rblk = blk;
410 			if (s) {
411 				*s = f;
412 			}
413 			break;
414 		}
415 	}
416 	return (rblk);
417 }
418 
419 struct tcp_function_block *
420 find_and_ref_tcp_functions(struct tcp_function_set *fs)
421 {
422 	struct tcp_function_block *blk;
423 
424 	rw_rlock(&tcp_function_lock);
425 	blk = find_tcp_functions_locked(fs);
426 	if (blk)
427 		refcount_acquire(&blk->tfb_refcnt);
428 	rw_runlock(&tcp_function_lock);
429 	return(blk);
430 }
431 
432 struct tcp_function_block *
433 find_and_ref_tcp_fb(struct tcp_function_block *blk)
434 {
435 	struct tcp_function_block *rblk;
436 
437 	rw_rlock(&tcp_function_lock);
438 	rblk = find_tcp_fb_locked(blk, NULL);
439 	if (rblk)
440 		refcount_acquire(&rblk->tfb_refcnt);
441 	rw_runlock(&tcp_function_lock);
442 	return(rblk);
443 }
444 
445 static struct tcp_function_block *
446 find_and_ref_tcp_default_fb(void)
447 {
448 	struct tcp_function_block *rblk;
449 
450 	rw_rlock(&tcp_function_lock);
451 	rblk = tcp_func_set_ptr;
452 	refcount_acquire(&rblk->tfb_refcnt);
453 	rw_runlock(&tcp_function_lock);
454 	return (rblk);
455 }
456 
457 void
458 tcp_switch_back_to_default(struct tcpcb *tp)
459 {
460 	struct tcp_function_block *tfb;
461 
462 	KASSERT(tp->t_fb != &tcp_def_funcblk,
463 	    ("%s: called by the built-in default stack", __func__));
464 
465 	/*
466 	 * Release the old stack. This function will either find a new one
467 	 * or panic.
468 	 */
469 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
470 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
471 	refcount_release(&tp->t_fb->tfb_refcnt);
472 
473 	/*
474 	 * Now, we'll find a new function block to use.
475 	 * Start by trying the current user-selected
476 	 * default, unless this stack is the user-selected
477 	 * default.
478 	 */
479 	tfb = find_and_ref_tcp_default_fb();
480 	if (tfb == tp->t_fb) {
481 		refcount_release(&tfb->tfb_refcnt);
482 		tfb = NULL;
483 	}
484 	/* Does the stack accept this connection? */
485 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
486 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
487 		refcount_release(&tfb->tfb_refcnt);
488 		tfb = NULL;
489 	}
490 	/* Try to use that stack. */
491 	if (tfb != NULL) {
492 		/* Initialize the new stack. If it succeeds, we are done. */
493 		tp->t_fb = tfb;
494 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
495 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
496 			return;
497 
498 		/*
499 		 * Initialization failed. Release the reference count on
500 		 * the stack.
501 		 */
502 		refcount_release(&tfb->tfb_refcnt);
503 	}
504 
505 	/*
506 	 * If that wasn't feasible, use the built-in default
507 	 * stack which is not allowed to reject anyone.
508 	 */
509 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
510 	if (tfb == NULL) {
511 		/* there always should be a default */
512 		panic("Can't refer to tcp_def_funcblk");
513 	}
514 	if (tfb->tfb_tcp_handoff_ok != NULL) {
515 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
516 			/* The default stack cannot say no */
517 			panic("Default stack rejects a new session?");
518 		}
519 	}
520 	tp->t_fb = tfb;
521 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
522 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
523 		/* The default stack cannot fail */
524 		panic("Default stack initialization failed");
525 	}
526 }
527 
528 static void
529 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
530     const struct sockaddr *sa, void *ctx)
531 {
532 	struct ip *iph;
533 #ifdef INET6
534 	struct ip6_hdr *ip6;
535 #endif
536 	struct udphdr *uh;
537 	struct tcphdr *th;
538 	int thlen;
539 	uint16_t port;
540 
541 	TCPSTAT_INC(tcps_tunneled_pkts);
542 	if ((m->m_flags & M_PKTHDR) == 0) {
543 		/* Can't handle one that is not a pkt hdr */
544 		TCPSTAT_INC(tcps_tunneled_errs);
545 		goto out;
546 	}
547 	thlen = sizeof(struct tcphdr);
548 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
549 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
550 		TCPSTAT_INC(tcps_tunneled_errs);
551 		goto out;
552 	}
553 	iph = mtod(m, struct ip *);
554 	uh = (struct udphdr *)((caddr_t)iph + off);
555 	th = (struct tcphdr *)(uh + 1);
556 	thlen = th->th_off << 2;
557 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
558 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
559 		if (m == NULL) {
560 			TCPSTAT_INC(tcps_tunneled_errs);
561 			goto out;
562 		} else {
563 			iph = mtod(m, struct ip *);
564 			uh = (struct udphdr *)((caddr_t)iph + off);
565 			th = (struct tcphdr *)(uh + 1);
566 		}
567 	}
568 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
569 	bcopy(th, uh, m->m_len - off);
570 	m->m_len -= sizeof(struct udphdr);
571 	m->m_pkthdr.len -= sizeof(struct udphdr);
572 	/*
573 	 * We use the same algorithm for
574 	 * both UDP and TCP for c-sum. So
575 	 * the code in tcp_input will skip
576 	 * the checksum. So we do nothing
577 	 * with the flag (m->m_pkthdr.csum_flags).
578 	 */
579 	switch (iph->ip_v) {
580 #ifdef INET
581 	case IPVERSION:
582 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
583 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
584 		break;
585 #endif
586 #ifdef INET6
587 	case IPV6_VERSION >> 4:
588 		ip6 = mtod(m, struct ip6_hdr *);
589 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
590 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
591 		break;
592 #endif
593 	default:
594 		goto out;
595 		break;
596 	}
597 	return;
598 out:
599 	m_freem(m);
600 }
601 
602 static int
603 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
604 {
605 	int error=ENOENT;
606 	struct tcp_function_set fs;
607 	struct tcp_function_block *blk;
608 
609 	memset(&fs, 0, sizeof(fs));
610 	rw_rlock(&tcp_function_lock);
611 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
612 	if (blk) {
613 		/* Found him */
614 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
615 		fs.pcbcnt = blk->tfb_refcnt;
616 	}
617 	rw_runlock(&tcp_function_lock);
618 	error = sysctl_handle_string(oidp, fs.function_set_name,
619 				     sizeof(fs.function_set_name), req);
620 
621 	/* Check for error or no change */
622 	if (error != 0 || req->newptr == NULL)
623 		return(error);
624 
625 	rw_wlock(&tcp_function_lock);
626 	blk = find_tcp_functions_locked(&fs);
627 	if ((blk == NULL) ||
628 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
629 		error = ENOENT;
630 		goto done;
631 	}
632 	tcp_func_set_ptr = blk;
633 done:
634 	rw_wunlock(&tcp_function_lock);
635 	return (error);
636 }
637 
638 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
639     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
640     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
641     "Set/get the default TCP functions");
642 
643 static int
644 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
645 {
646 	int error, cnt, linesz;
647 	struct tcp_function *f;
648 	char *buffer, *cp;
649 	size_t bufsz, outsz;
650 	bool alias;
651 
652 	cnt = 0;
653 	rw_rlock(&tcp_function_lock);
654 	TAILQ_FOREACH(f, &t_functions, tf_next) {
655 		cnt++;
656 	}
657 	rw_runlock(&tcp_function_lock);
658 
659 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
660 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
661 
662 	error = 0;
663 	cp = buffer;
664 
665 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
666 	    "Alias", "PCB count");
667 	cp += linesz;
668 	bufsz -= linesz;
669 	outsz = linesz;
670 
671 	rw_rlock(&tcp_function_lock);
672 	TAILQ_FOREACH(f, &t_functions, tf_next) {
673 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
674 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
675 		    f->tf_fb->tfb_tcp_block_name,
676 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
677 		    alias ? f->tf_name : "-",
678 		    f->tf_fb->tfb_refcnt);
679 		if (linesz >= bufsz) {
680 			error = EOVERFLOW;
681 			break;
682 		}
683 		cp += linesz;
684 		bufsz -= linesz;
685 		outsz += linesz;
686 	}
687 	rw_runlock(&tcp_function_lock);
688 	if (error == 0)
689 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
690 	free(buffer, M_TEMP);
691 	return (error);
692 }
693 
694 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
695     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
696     NULL, 0, sysctl_net_inet_list_available, "A",
697     "list available TCP Function sets");
698 
699 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
700 
701 #ifdef INET
702 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
703 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
704 #endif
705 #ifdef INET6
706 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
707 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
708 #endif
709 
710 static void
711 tcp_over_udp_stop(void)
712 {
713 	/*
714 	 * This function assumes sysctl caller holds inp_rinfo_lock()
715 	 * for writting!
716 	 */
717 #ifdef INET
718 	if (V_udp4_tun_socket != NULL) {
719 		soclose(V_udp4_tun_socket);
720 		V_udp4_tun_socket = NULL;
721 	}
722 #endif
723 #ifdef INET6
724 	if (V_udp6_tun_socket != NULL) {
725 		soclose(V_udp6_tun_socket);
726 		V_udp6_tun_socket = NULL;
727 	}
728 #endif
729 }
730 
731 static int
732 tcp_over_udp_start(void)
733 {
734 	uint16_t port;
735 	int ret;
736 #ifdef INET
737 	struct sockaddr_in sin;
738 #endif
739 #ifdef INET6
740 	struct sockaddr_in6 sin6;
741 #endif
742 	/*
743 	 * This function assumes sysctl caller holds inp_info_rlock()
744 	 * for writting!
745 	 */
746 	port = V_tcp_udp_tunneling_port;
747 	if (ntohs(port) == 0) {
748 		/* Must have a port set */
749 		return (EINVAL);
750 	}
751 #ifdef INET
752 	if (V_udp4_tun_socket != NULL) {
753 		/* Already running -- must stop first */
754 		return (EALREADY);
755 	}
756 #endif
757 #ifdef INET6
758 	if (V_udp6_tun_socket != NULL) {
759 		/* Already running -- must stop first */
760 		return (EALREADY);
761 	}
762 #endif
763 #ifdef INET
764 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
765 	    SOCK_DGRAM, IPPROTO_UDP,
766 	    curthread->td_ucred, curthread))) {
767 		tcp_over_udp_stop();
768 		return (ret);
769 	}
770 	/* Call the special UDP hook. */
771 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
772 	    tcp_recv_udp_tunneled_packet,
773 	    tcp_ctlinput_viaudp,
774 	    NULL))) {
775 		tcp_over_udp_stop();
776 		return (ret);
777 	}
778 	/* Ok, we have a socket, bind it to the port. */
779 	memset(&sin, 0, sizeof(struct sockaddr_in));
780 	sin.sin_len = sizeof(struct sockaddr_in);
781 	sin.sin_family = AF_INET;
782 	sin.sin_port = htons(port);
783 	if ((ret = sobind(V_udp4_tun_socket,
784 	    (struct sockaddr *)&sin, curthread))) {
785 		tcp_over_udp_stop();
786 		return (ret);
787 	}
788 #endif
789 #ifdef INET6
790 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
791 	    SOCK_DGRAM, IPPROTO_UDP,
792 	    curthread->td_ucred, curthread))) {
793 		tcp_over_udp_stop();
794 		return (ret);
795 	}
796 	/* Call the special UDP hook. */
797 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
798 	    tcp_recv_udp_tunneled_packet,
799 	    tcp6_ctlinput_viaudp,
800 	    NULL))) {
801 		tcp_over_udp_stop();
802 		return (ret);
803 	}
804 	/* Ok, we have a socket, bind it to the port. */
805 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
806 	sin6.sin6_len = sizeof(struct sockaddr_in6);
807 	sin6.sin6_family = AF_INET6;
808 	sin6.sin6_port = htons(port);
809 	if ((ret = sobind(V_udp6_tun_socket,
810 	    (struct sockaddr *)&sin6, curthread))) {
811 		tcp_over_udp_stop();
812 		return (ret);
813 	}
814 #endif
815 	return (0);
816 }
817 
818 static int
819 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
820 {
821 	int error;
822 	uint32_t old, new;
823 
824 	old = V_tcp_udp_tunneling_port;
825 	new = old;
826 	error = sysctl_handle_int(oidp, &new, 0, req);
827 	if ((error == 0) &&
828 	    (req->newptr != NULL)) {
829 		if ((new < TCP_TUNNELING_PORT_MIN) ||
830 		    (new > TCP_TUNNELING_PORT_MAX)) {
831 			error = EINVAL;
832 		} else {
833 			V_tcp_udp_tunneling_port = new;
834 			if (old != 0) {
835 				tcp_over_udp_stop();
836 			}
837 			if (new != 0) {
838 				error = tcp_over_udp_start();
839 			}
840 		}
841 	}
842 	return (error);
843 }
844 
845 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
846     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
847     &VNET_NAME(tcp_udp_tunneling_port),
848     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
849     "Tunneling port for tcp over udp");
850 
851 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
852 
853 static int
854 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
855 {
856 	int error, new;
857 
858 	new = V_tcp_udp_tunneling_overhead;
859 	error = sysctl_handle_int(oidp, &new, 0, req);
860 	if (error == 0 && req->newptr) {
861 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
862 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
863 			error = EINVAL;
864 		else
865 			V_tcp_udp_tunneling_overhead = new;
866 	}
867 	return (error);
868 }
869 
870 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
871     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
872     &VNET_NAME(tcp_udp_tunneling_overhead),
873     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
874     "MSS reduction when using tcp over udp");
875 
876 /*
877  * Exports one (struct tcp_function_info) for each alias/name.
878  */
879 static int
880 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
881 {
882 	int cnt, error;
883 	struct tcp_function *f;
884 	struct tcp_function_info tfi;
885 
886 	/*
887 	 * We don't allow writes.
888 	 */
889 	if (req->newptr != NULL)
890 		return (EINVAL);
891 
892 	/*
893 	 * Wire the old buffer so we can directly copy the functions to
894 	 * user space without dropping the lock.
895 	 */
896 	if (req->oldptr != NULL) {
897 		error = sysctl_wire_old_buffer(req, 0);
898 		if (error)
899 			return (error);
900 	}
901 
902 	/*
903 	 * Walk the list and copy out matching entries. If INVARIANTS
904 	 * is compiled in, also walk the list to verify the length of
905 	 * the list matches what we have recorded.
906 	 */
907 	rw_rlock(&tcp_function_lock);
908 
909 	cnt = 0;
910 #ifndef INVARIANTS
911 	if (req->oldptr == NULL) {
912 		cnt = tcp_fb_cnt;
913 		goto skip_loop;
914 	}
915 #endif
916 	TAILQ_FOREACH(f, &t_functions, tf_next) {
917 #ifdef INVARIANTS
918 		cnt++;
919 #endif
920 		if (req->oldptr != NULL) {
921 			bzero(&tfi, sizeof(tfi));
922 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
923 			tfi.tfi_id = f->tf_fb->tfb_id;
924 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
925 			    sizeof(tfi.tfi_alias));
926 			(void)strlcpy(tfi.tfi_name,
927 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
928 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
929 			/*
930 			 * Don't stop on error, as that is the
931 			 * mechanism we use to accumulate length
932 			 * information if the buffer was too short.
933 			 */
934 		}
935 	}
936 	KASSERT(cnt == tcp_fb_cnt,
937 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
938 #ifndef INVARIANTS
939 skip_loop:
940 #endif
941 	rw_runlock(&tcp_function_lock);
942 	if (req->oldptr == NULL)
943 		error = SYSCTL_OUT(req, NULL,
944 		    (cnt + 1) * sizeof(struct tcp_function_info));
945 
946 	return (error);
947 }
948 
949 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
950 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
951 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
952 	    "List TCP function block name-to-ID mappings");
953 
954 /*
955  * tfb_tcp_handoff_ok() function for the default stack.
956  * Note that we'll basically try to take all comers.
957  */
958 static int
959 tcp_default_handoff_ok(struct tcpcb *tp)
960 {
961 
962 	return (0);
963 }
964 
965 /*
966  * tfb_tcp_fb_init() function for the default stack.
967  *
968  * This handles making sure we have appropriate timers set if you are
969  * transitioning a socket that has some amount of setup done.
970  *
971  * The init() fuction from the default can *never* return non-zero i.e.
972  * it is required to always succeed since it is the stack of last resort!
973  */
974 static int
975 tcp_default_fb_init(struct tcpcb *tp)
976 {
977 
978 	struct socket *so;
979 
980 	INP_WLOCK_ASSERT(tp->t_inpcb);
981 
982 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
983 	    ("%s: connection %p in unexpected state %d", __func__, tp,
984 	    tp->t_state));
985 
986 	/*
987 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
988 	 * know what to do for unexpected states (which includes TIME_WAIT).
989 	 */
990 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
991 		return (0);
992 
993 	/*
994 	 * Make sure some kind of transmission timer is set if there is
995 	 * outstanding data.
996 	 */
997 	so = tp->t_inpcb->inp_socket;
998 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
999 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1000 	    tcp_timer_active(tp, TT_PERSIST))) {
1001 		/*
1002 		 * If the session has established and it looks like it should
1003 		 * be in the persist state, set the persist timer. Otherwise,
1004 		 * set the retransmit timer.
1005 		 */
1006 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1007 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1008 		    (int32_t)sbavail(&so->so_snd))
1009 			tcp_setpersist(tp);
1010 		else
1011 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1012 	}
1013 
1014 	/* All non-embryonic sessions get a keepalive timer. */
1015 	if (!tcp_timer_active(tp, TT_KEEP))
1016 		tcp_timer_activate(tp, TT_KEEP,
1017 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1018 		    TP_KEEPINIT(tp));
1019 
1020 	/*
1021 	 * Make sure critical variables are initialized
1022 	 * if transitioning while in Recovery.
1023 	 */
1024 	if IN_FASTRECOVERY(tp->t_flags) {
1025 		if (tp->sackhint.recover_fs == 0)
1026 			tp->sackhint.recover_fs = max(1,
1027 			    tp->snd_nxt - tp->snd_una);
1028 	}
1029 
1030 	return (0);
1031 }
1032 
1033 /*
1034  * tfb_tcp_fb_fini() function for the default stack.
1035  *
1036  * This changes state as necessary (or prudent) to prepare for another stack
1037  * to assume responsibility for the connection.
1038  */
1039 static void
1040 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1041 {
1042 
1043 	INP_WLOCK_ASSERT(tp->t_inpcb);
1044 	return;
1045 }
1046 
1047 /*
1048  * Target size of TCP PCB hash tables. Must be a power of two.
1049  *
1050  * Note that this can be overridden by the kernel environment
1051  * variable net.inet.tcp.tcbhashsize
1052  */
1053 #ifndef TCBHASHSIZE
1054 #define TCBHASHSIZE	0
1055 #endif
1056 
1057 /*
1058  * XXX
1059  * Callouts should be moved into struct tcp directly.  They are currently
1060  * separate because the tcpcb structure is exported to userland for sysctl
1061  * parsing purposes, which do not know about callouts.
1062  */
1063 struct tcpcb_mem {
1064 	struct	tcpcb		tcb;
1065 	struct	tcp_timer	tt;
1066 	struct	cc_var		ccv;
1067 #ifdef TCP_HHOOK
1068 	struct	osd		osd;
1069 #endif
1070 };
1071 
1072 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1073 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1074 
1075 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1076 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1077 
1078 static struct mtx isn_mtx;
1079 
1080 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1081 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1082 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1083 
1084 /*
1085  * TCP initialization.
1086  */
1087 static void
1088 tcp_zone_change(void *tag)
1089 {
1090 
1091 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
1092 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1093 	tcp_tw_zone_change();
1094 }
1095 
1096 static int
1097 tcp_inpcb_init(void *mem, int size, int flags)
1098 {
1099 	struct inpcb *inp = mem;
1100 
1101 	INP_LOCK_INIT(inp, "inp", "tcpinp");
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Take a value and get the next power of 2 that doesn't overflow.
1107  * Used to size the tcp_inpcb hash buckets.
1108  */
1109 static int
1110 maketcp_hashsize(int size)
1111 {
1112 	int hashsize;
1113 
1114 	/*
1115 	 * auto tune.
1116 	 * get the next power of 2 higher than maxsockets.
1117 	 */
1118 	hashsize = 1 << fls(size);
1119 	/* catch overflow, and just go one power of 2 smaller */
1120 	if (hashsize < size) {
1121 		hashsize = 1 << (fls(size) - 1);
1122 	}
1123 	return (hashsize);
1124 }
1125 
1126 static volatile int next_tcp_stack_id = 1;
1127 
1128 /*
1129  * Register a TCP function block with the name provided in the names
1130  * array.  (Note that this function does NOT automatically register
1131  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1132  * explicitly include blk->tfb_tcp_block_name in the list of names if
1133  * you wish to register the stack with that name.)
1134  *
1135  * Either all name registrations will succeed or all will fail.  If
1136  * a name registration fails, the function will update the num_names
1137  * argument to point to the array index of the name that encountered
1138  * the failure.
1139  *
1140  * Returns 0 on success, or an error code on failure.
1141  */
1142 int
1143 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1144     const char *names[], int *num_names)
1145 {
1146 	struct tcp_function *n;
1147 	struct tcp_function_set fs;
1148 	int error, i;
1149 
1150 	KASSERT(names != NULL && *num_names > 0,
1151 	    ("%s: Called with 0-length name list", __func__));
1152 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1153 	KASSERT(rw_initialized(&tcp_function_lock),
1154 	    ("%s: called too early", __func__));
1155 
1156 	if ((blk->tfb_tcp_output == NULL) ||
1157 	    (blk->tfb_tcp_do_segment == NULL) ||
1158 	    (blk->tfb_tcp_ctloutput == NULL) ||
1159 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1160 		/*
1161 		 * These functions are required and you
1162 		 * need a name.
1163 		 */
1164 		*num_names = 0;
1165 		return (EINVAL);
1166 	}
1167 	if (blk->tfb_tcp_timer_stop_all ||
1168 	    blk->tfb_tcp_timer_activate ||
1169 	    blk->tfb_tcp_timer_active ||
1170 	    blk->tfb_tcp_timer_stop) {
1171 		/*
1172 		 * If you define one timer function you
1173 		 * must have them all.
1174 		 */
1175 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1176 		    (blk->tfb_tcp_timer_activate == NULL) ||
1177 		    (blk->tfb_tcp_timer_active == NULL) ||
1178 		    (blk->tfb_tcp_timer_stop == NULL)) {
1179 			*num_names = 0;
1180 			return (EINVAL);
1181 		}
1182 	}
1183 
1184 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1185 		*num_names = 0;
1186 		return (EINVAL);
1187 	}
1188 
1189 	refcount_init(&blk->tfb_refcnt, 0);
1190 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1191 	for (i = 0; i < *num_names; i++) {
1192 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1193 		if (n == NULL) {
1194 			error = ENOMEM;
1195 			goto cleanup;
1196 		}
1197 		n->tf_fb = blk;
1198 
1199 		(void)strlcpy(fs.function_set_name, names[i],
1200 		    sizeof(fs.function_set_name));
1201 		rw_wlock(&tcp_function_lock);
1202 		if (find_tcp_functions_locked(&fs) != NULL) {
1203 			/* Duplicate name space not allowed */
1204 			rw_wunlock(&tcp_function_lock);
1205 			free(n, M_TCPFUNCTIONS);
1206 			error = EALREADY;
1207 			goto cleanup;
1208 		}
1209 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1210 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1211 		tcp_fb_cnt++;
1212 		rw_wunlock(&tcp_function_lock);
1213 	}
1214 	return(0);
1215 
1216 cleanup:
1217 	/*
1218 	 * Deregister the names we just added. Because registration failed
1219 	 * for names[i], we don't need to deregister that name.
1220 	 */
1221 	*num_names = i;
1222 	rw_wlock(&tcp_function_lock);
1223 	while (--i >= 0) {
1224 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1225 			if (!strncmp(n->tf_name, names[i],
1226 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1227 				TAILQ_REMOVE(&t_functions, n, tf_next);
1228 				tcp_fb_cnt--;
1229 				n->tf_fb = NULL;
1230 				free(n, M_TCPFUNCTIONS);
1231 				break;
1232 			}
1233 		}
1234 	}
1235 	rw_wunlock(&tcp_function_lock);
1236 	return (error);
1237 }
1238 
1239 /*
1240  * Register a TCP function block using the name provided in the name
1241  * argument.
1242  *
1243  * Returns 0 on success, or an error code on failure.
1244  */
1245 int
1246 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1247     int wait)
1248 {
1249 	const char *name_list[1];
1250 	int num_names, rv;
1251 
1252 	num_names = 1;
1253 	if (name != NULL)
1254 		name_list[0] = name;
1255 	else
1256 		name_list[0] = blk->tfb_tcp_block_name;
1257 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1258 	return (rv);
1259 }
1260 
1261 /*
1262  * Register a TCP function block using the name defined in
1263  * blk->tfb_tcp_block_name.
1264  *
1265  * Returns 0 on success, or an error code on failure.
1266  */
1267 int
1268 register_tcp_functions(struct tcp_function_block *blk, int wait)
1269 {
1270 
1271 	return (register_tcp_functions_as_name(blk, NULL, wait));
1272 }
1273 
1274 /*
1275  * Deregister all names associated with a function block. This
1276  * functionally removes the function block from use within the system.
1277  *
1278  * When called with a true quiesce argument, mark the function block
1279  * as being removed so no more stacks will use it and determine
1280  * whether the removal would succeed.
1281  *
1282  * When called with a false quiesce argument, actually attempt the
1283  * removal.
1284  *
1285  * When called with a force argument, attempt to switch all TCBs to
1286  * use the default stack instead of returning EBUSY.
1287  *
1288  * Returns 0 on success (or if the removal would succeed, or an error
1289  * code on failure.
1290  */
1291 int
1292 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1293     bool force)
1294 {
1295 	struct tcp_function *f;
1296 
1297 	if (blk == &tcp_def_funcblk) {
1298 		/* You can't un-register the default */
1299 		return (EPERM);
1300 	}
1301 	rw_wlock(&tcp_function_lock);
1302 	if (blk == tcp_func_set_ptr) {
1303 		/* You can't free the current default */
1304 		rw_wunlock(&tcp_function_lock);
1305 		return (EBUSY);
1306 	}
1307 	/* Mark the block so no more stacks can use it. */
1308 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1309 	/*
1310 	 * If TCBs are still attached to the stack, attempt to switch them
1311 	 * to the default stack.
1312 	 */
1313 	if (force && blk->tfb_refcnt) {
1314 		struct inpcb *inp;
1315 		struct tcpcb *tp;
1316 		VNET_ITERATOR_DECL(vnet_iter);
1317 
1318 		rw_wunlock(&tcp_function_lock);
1319 
1320 		VNET_LIST_RLOCK();
1321 		VNET_FOREACH(vnet_iter) {
1322 			CURVNET_SET(vnet_iter);
1323 			INP_INFO_WLOCK(&V_tcbinfo);
1324 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1325 				INP_WLOCK(inp);
1326 				if (inp->inp_flags & INP_TIMEWAIT) {
1327 					INP_WUNLOCK(inp);
1328 					continue;
1329 				}
1330 				tp = intotcpcb(inp);
1331 				if (tp == NULL || tp->t_fb != blk) {
1332 					INP_WUNLOCK(inp);
1333 					continue;
1334 				}
1335 				tcp_switch_back_to_default(tp);
1336 				INP_WUNLOCK(inp);
1337 			}
1338 			INP_INFO_WUNLOCK(&V_tcbinfo);
1339 			CURVNET_RESTORE();
1340 		}
1341 		VNET_LIST_RUNLOCK();
1342 
1343 		rw_wlock(&tcp_function_lock);
1344 	}
1345 	if (blk->tfb_refcnt) {
1346 		/* TCBs still attached. */
1347 		rw_wunlock(&tcp_function_lock);
1348 		return (EBUSY);
1349 	}
1350 	if (quiesce) {
1351 		/* Skip removal. */
1352 		rw_wunlock(&tcp_function_lock);
1353 		return (0);
1354 	}
1355 	/* Remove any function names that map to this function block. */
1356 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1357 		TAILQ_REMOVE(&t_functions, f, tf_next);
1358 		tcp_fb_cnt--;
1359 		f->tf_fb = NULL;
1360 		free(f, M_TCPFUNCTIONS);
1361 	}
1362 	rw_wunlock(&tcp_function_lock);
1363 	return (0);
1364 }
1365 
1366 void
1367 tcp_init(void)
1368 {
1369 	const char *tcbhash_tuneable;
1370 	int hashsize;
1371 
1372 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1373 
1374 #ifdef TCP_HHOOK
1375 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1376 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1377 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1378 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1379 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1380 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1381 #endif
1382 #ifdef STATS
1383 	if (tcp_stats_init())
1384 		printf("%s: WARNING: unable to initialise TCP stats\n",
1385 		    __func__);
1386 #endif
1387 	hashsize = TCBHASHSIZE;
1388 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1389 	if (hashsize == 0) {
1390 		/*
1391 		 * Auto tune the hash size based on maxsockets.
1392 		 * A perfect hash would have a 1:1 mapping
1393 		 * (hashsize = maxsockets) however it's been
1394 		 * suggested that O(2) average is better.
1395 		 */
1396 		hashsize = maketcp_hashsize(maxsockets / 4);
1397 		/*
1398 		 * Our historical default is 512,
1399 		 * do not autotune lower than this.
1400 		 */
1401 		if (hashsize < 512)
1402 			hashsize = 512;
1403 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1404 			printf("%s: %s auto tuned to %d\n", __func__,
1405 			    tcbhash_tuneable, hashsize);
1406 	}
1407 	/*
1408 	 * We require a hashsize to be a power of two.
1409 	 * Previously if it was not a power of two we would just reset it
1410 	 * back to 512, which could be a nasty surprise if you did not notice
1411 	 * the error message.
1412 	 * Instead what we do is clip it to the closest power of two lower
1413 	 * than the specified hash value.
1414 	 */
1415 	if (!powerof2(hashsize)) {
1416 		int oldhashsize = hashsize;
1417 
1418 		hashsize = maketcp_hashsize(hashsize);
1419 		/* prevent absurdly low value */
1420 		if (hashsize < 16)
1421 			hashsize = 16;
1422 		printf("%s: WARNING: TCB hash size not a power of 2, "
1423 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1424 		    hashsize);
1425 	}
1426 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1427 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1428 
1429 	/*
1430 	 * These have to be type stable for the benefit of the timers.
1431 	 */
1432 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1433 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1434 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1435 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1436 
1437 	tcp_tw_init();
1438 	syncache_init();
1439 	tcp_hc_init();
1440 
1441 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1442 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1443 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1444 
1445 	tcp_fastopen_init();
1446 
1447 	/* Skip initialization of globals for non-default instances. */
1448 	if (!IS_DEFAULT_VNET(curvnet))
1449 		return;
1450 
1451 	tcp_reass_global_init();
1452 
1453 	/* XXX virtualize those bellow? */
1454 	tcp_delacktime = TCPTV_DELACK;
1455 	tcp_keepinit = TCPTV_KEEP_INIT;
1456 	tcp_keepidle = TCPTV_KEEP_IDLE;
1457 	tcp_keepintvl = TCPTV_KEEPINTVL;
1458 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1459 	tcp_msl = TCPTV_MSL;
1460 	tcp_rexmit_initial = TCPTV_RTOBASE;
1461 	if (tcp_rexmit_initial < 1)
1462 		tcp_rexmit_initial = 1;
1463 	tcp_rexmit_min = TCPTV_MIN;
1464 	if (tcp_rexmit_min < 1)
1465 		tcp_rexmit_min = 1;
1466 	tcp_persmin = TCPTV_PERSMIN;
1467 	tcp_persmax = TCPTV_PERSMAX;
1468 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1469 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1470 	tcp_tcbhashsize = hashsize;
1471 
1472 	/* Setup the tcp function block list */
1473 	TAILQ_INIT(&t_functions);
1474 	rw_init(&tcp_function_lock, "tcp_func_lock");
1475 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1476 #ifdef TCP_BLACKBOX
1477 	/* Initialize the TCP logging data. */
1478 	tcp_log_init();
1479 #endif
1480 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1481 
1482 	if (tcp_soreceive_stream) {
1483 #ifdef INET
1484 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1485 #endif
1486 #ifdef INET6
1487 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1488 #endif /* INET6 */
1489 	}
1490 
1491 #ifdef INET6
1492 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1493 #else /* INET6 */
1494 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1495 #endif /* INET6 */
1496 	if (max_protohdr < TCP_MINPROTOHDR)
1497 		max_protohdr = TCP_MINPROTOHDR;
1498 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1499 		panic("tcp_init");
1500 #undef TCP_MINPROTOHDR
1501 
1502 	ISN_LOCK_INIT();
1503 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1504 		SHUTDOWN_PRI_DEFAULT);
1505 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1506 		EVENTHANDLER_PRI_ANY);
1507 
1508 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1509 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1510 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1511 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1512 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1513 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1514 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1515 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1516 #ifdef TCPPCAP
1517 	tcp_pcap_init();
1518 #endif
1519 }
1520 
1521 #ifdef VIMAGE
1522 static void
1523 tcp_destroy(void *unused __unused)
1524 {
1525 	int n;
1526 #ifdef TCP_HHOOK
1527 	int error;
1528 #endif
1529 
1530 	/*
1531 	 * All our processes are gone, all our sockets should be cleaned
1532 	 * up, which means, we should be past the tcp_discardcb() calls.
1533 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1534 	 */
1535 	for (;;) {
1536 		INP_LIST_RLOCK(&V_tcbinfo);
1537 		n = V_tcbinfo.ipi_count;
1538 		INP_LIST_RUNLOCK(&V_tcbinfo);
1539 		if (n == 0)
1540 			break;
1541 		pause("tcpdes", hz / 10);
1542 	}
1543 	tcp_hc_destroy();
1544 	syncache_destroy();
1545 	tcp_tw_destroy();
1546 	in_pcbinfo_destroy(&V_tcbinfo);
1547 	/* tcp_discardcb() clears the sack_holes up. */
1548 	uma_zdestroy(V_sack_hole_zone);
1549 	uma_zdestroy(V_tcpcb_zone);
1550 
1551 	/*
1552 	 * Cannot free the zone until all tcpcbs are released as we attach
1553 	 * the allocations to them.
1554 	 */
1555 	tcp_fastopen_destroy();
1556 
1557 #ifdef TCP_HHOOK
1558 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1559 	if (error != 0) {
1560 		printf("%s: WARNING: unable to deregister helper hook "
1561 		    "type=%d, id=%d: error %d returned\n", __func__,
1562 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1563 	}
1564 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1565 	if (error != 0) {
1566 		printf("%s: WARNING: unable to deregister helper hook "
1567 		    "type=%d, id=%d: error %d returned\n", __func__,
1568 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1569 	}
1570 #endif
1571 }
1572 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1573 #endif
1574 
1575 void
1576 tcp_fini(void *xtp)
1577 {
1578 
1579 }
1580 
1581 /*
1582  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1583  * tcp_template used to store this data in mbufs, but we now recopy it out
1584  * of the tcpcb each time to conserve mbufs.
1585  */
1586 void
1587 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1588 {
1589 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1590 
1591 	INP_WLOCK_ASSERT(inp);
1592 
1593 #ifdef INET6
1594 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1595 		struct ip6_hdr *ip6;
1596 
1597 		ip6 = (struct ip6_hdr *)ip_ptr;
1598 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1599 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1600 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1601 			(IPV6_VERSION & IPV6_VERSION_MASK);
1602 		if (port == 0)
1603 			ip6->ip6_nxt = IPPROTO_TCP;
1604 		else
1605 			ip6->ip6_nxt = IPPROTO_UDP;
1606 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1607 		ip6->ip6_src = inp->in6p_laddr;
1608 		ip6->ip6_dst = inp->in6p_faddr;
1609 	}
1610 #endif /* INET6 */
1611 #if defined(INET6) && defined(INET)
1612 	else
1613 #endif
1614 #ifdef INET
1615 	{
1616 		struct ip *ip;
1617 
1618 		ip = (struct ip *)ip_ptr;
1619 		ip->ip_v = IPVERSION;
1620 		ip->ip_hl = 5;
1621 		ip->ip_tos = inp->inp_ip_tos;
1622 		ip->ip_len = 0;
1623 		ip->ip_id = 0;
1624 		ip->ip_off = 0;
1625 		ip->ip_ttl = inp->inp_ip_ttl;
1626 		ip->ip_sum = 0;
1627 		if (port == 0)
1628 			ip->ip_p = IPPROTO_TCP;
1629 		else
1630 			ip->ip_p = IPPROTO_UDP;
1631 		ip->ip_src = inp->inp_laddr;
1632 		ip->ip_dst = inp->inp_faddr;
1633 	}
1634 #endif /* INET */
1635 	th->th_sport = inp->inp_lport;
1636 	th->th_dport = inp->inp_fport;
1637 	th->th_seq = 0;
1638 	th->th_ack = 0;
1639 	th->th_x2 = 0;
1640 	th->th_off = 5;
1641 	th->th_flags = 0;
1642 	th->th_win = 0;
1643 	th->th_urp = 0;
1644 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1645 }
1646 
1647 /*
1648  * Create template to be used to send tcp packets on a connection.
1649  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1650  * use for this function is in keepalives, which use tcp_respond.
1651  */
1652 struct tcptemp *
1653 tcpip_maketemplate(struct inpcb *inp)
1654 {
1655 	struct tcptemp *t;
1656 
1657 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1658 	if (t == NULL)
1659 		return (NULL);
1660 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1661 	return (t);
1662 }
1663 
1664 /*
1665  * Send a single message to the TCP at address specified by
1666  * the given TCP/IP header.  If m == NULL, then we make a copy
1667  * of the tcpiphdr at th and send directly to the addressed host.
1668  * This is used to force keep alive messages out using the TCP
1669  * template for a connection.  If flags are given then we send
1670  * a message back to the TCP which originated the segment th,
1671  * and discard the mbuf containing it and any other attached mbufs.
1672  *
1673  * In any case the ack and sequence number of the transmitted
1674  * segment are as specified by the parameters.
1675  *
1676  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1677  */
1678 void
1679 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1680     tcp_seq ack, tcp_seq seq, int flags)
1681 {
1682 	struct tcpopt to;
1683 	struct inpcb *inp;
1684 	struct ip *ip;
1685 	struct mbuf *optm;
1686 	struct udphdr *uh = NULL;
1687 	struct tcphdr *nth;
1688 	u_char *optp;
1689 #ifdef INET6
1690 	struct ip6_hdr *ip6;
1691 	int isipv6;
1692 #endif /* INET6 */
1693 	int optlen, tlen, win, ulen;
1694 	bool incl_opts;
1695 	uint16_t port;
1696 
1697 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1698 	NET_EPOCH_ASSERT();
1699 
1700 #ifdef INET6
1701 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1702 	ip6 = ipgen;
1703 #endif /* INET6 */
1704 	ip = ipgen;
1705 
1706 	if (tp != NULL) {
1707 		inp = tp->t_inpcb;
1708 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1709 		INP_LOCK_ASSERT(inp);
1710 	} else
1711 		inp = NULL;
1712 
1713 	if (m != NULL) {
1714 #ifdef INET6
1715 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1716 			port = m->m_pkthdr.tcp_tun_port;
1717 		else
1718 #endif
1719 		if (ip && (ip->ip_p == IPPROTO_UDP))
1720 			port = m->m_pkthdr.tcp_tun_port;
1721 		else
1722 			port = 0;
1723 	} else
1724 		port = tp->t_port;
1725 
1726 	incl_opts = false;
1727 	win = 0;
1728 	if (tp != NULL) {
1729 		if (!(flags & TH_RST)) {
1730 			win = sbspace(&inp->inp_socket->so_rcv);
1731 			if (win > TCP_MAXWIN << tp->rcv_scale)
1732 				win = TCP_MAXWIN << tp->rcv_scale;
1733 		}
1734 		if ((tp->t_flags & TF_NOOPT) == 0)
1735 			incl_opts = true;
1736 	}
1737 	if (m == NULL) {
1738 		m = m_gethdr(M_NOWAIT, MT_DATA);
1739 		if (m == NULL)
1740 			return;
1741 		m->m_data += max_linkhdr;
1742 #ifdef INET6
1743 		if (isipv6) {
1744 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1745 			      sizeof(struct ip6_hdr));
1746 			ip6 = mtod(m, struct ip6_hdr *);
1747 			nth = (struct tcphdr *)(ip6 + 1);
1748 			if (port) {
1749 				/* Insert a UDP header */
1750 				uh = (struct udphdr *)nth;
1751 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1752 				uh->uh_dport = port;
1753 				nth = (struct tcphdr *)(uh + 1);
1754 			}
1755 		} else
1756 #endif /* INET6 */
1757 		{
1758 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1759 			ip = mtod(m, struct ip *);
1760 			nth = (struct tcphdr *)(ip + 1);
1761 			if (port) {
1762 				/* Insert a UDP header */
1763 				uh = (struct udphdr *)nth;
1764 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1765 				uh->uh_dport = port;
1766 				nth = (struct tcphdr *)(uh + 1);
1767 			}
1768 		}
1769 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1770 		flags = TH_ACK;
1771 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1772 		struct mbuf *n;
1773 
1774 		/* Can't reuse 'm', allocate a new mbuf. */
1775 		n = m_gethdr(M_NOWAIT, MT_DATA);
1776 		if (n == NULL) {
1777 			m_freem(m);
1778 			return;
1779 		}
1780 
1781 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1782 			m_freem(m);
1783 			m_freem(n);
1784 			return;
1785 		}
1786 
1787 		n->m_data += max_linkhdr;
1788 		/* m_len is set later */
1789 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1790 #ifdef INET6
1791 		if (isipv6) {
1792 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1793 			      sizeof(struct ip6_hdr));
1794 			ip6 = mtod(n, struct ip6_hdr *);
1795 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1796 			nth = (struct tcphdr *)(ip6 + 1);
1797 			if (port) {
1798 				/* Insert a UDP header */
1799 				uh = (struct udphdr *)nth;
1800 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1801 				uh->uh_dport = port;
1802 				nth = (struct tcphdr *)(uh + 1);
1803 			}
1804 		} else
1805 #endif /* INET6 */
1806 		{
1807 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1808 			ip = mtod(n, struct ip *);
1809 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1810 			nth = (struct tcphdr *)(ip + 1);
1811 			if (port) {
1812 				/* Insert a UDP header */
1813 				uh = (struct udphdr *)nth;
1814 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1815 				uh->uh_dport = port;
1816 				nth = (struct tcphdr *)(uh + 1);
1817 			}
1818 		}
1819 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1820 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1821 		th = nth;
1822 		m_freem(m);
1823 		m = n;
1824 	} else {
1825 		/*
1826 		 *  reuse the mbuf.
1827 		 * XXX MRT We inherit the FIB, which is lucky.
1828 		 */
1829 		m_freem(m->m_next);
1830 		m->m_next = NULL;
1831 		m->m_data = (caddr_t)ipgen;
1832 		/* m_len is set later */
1833 #ifdef INET6
1834 		if (isipv6) {
1835 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1836 			nth = (struct tcphdr *)(ip6 + 1);
1837 		} else
1838 #endif /* INET6 */
1839 		{
1840 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1841 			nth = (struct tcphdr *)(ip + 1);
1842 		}
1843 		if (th != nth) {
1844 			/*
1845 			 * this is usually a case when an extension header
1846 			 * exists between the IPv6 header and the
1847 			 * TCP header.
1848 			 */
1849 			nth->th_sport = th->th_sport;
1850 			nth->th_dport = th->th_dport;
1851 		}
1852 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1853 #undef xchg
1854 	}
1855 	tlen = 0;
1856 #ifdef INET6
1857 	if (isipv6)
1858 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1859 #endif
1860 #if defined(INET) && defined(INET6)
1861 	else
1862 #endif
1863 #ifdef INET
1864 		tlen = sizeof (struct tcpiphdr);
1865 #endif
1866 	if (port)
1867 		tlen += sizeof (struct udphdr);
1868 #ifdef INVARIANTS
1869 	m->m_len = 0;
1870 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1871 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1872 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1873 #endif
1874 	m->m_len = tlen;
1875 	to.to_flags = 0;
1876 	if (incl_opts) {
1877 		/* Make sure we have room. */
1878 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1879 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1880 			if (m->m_next) {
1881 				optp = mtod(m->m_next, u_char *);
1882 				optm = m->m_next;
1883 			} else
1884 				incl_opts = false;
1885 		} else {
1886 			optp = (u_char *) (nth + 1);
1887 			optm = m;
1888 		}
1889 	}
1890 	if (incl_opts) {
1891 		/* Timestamps. */
1892 		if (tp->t_flags & TF_RCVD_TSTMP) {
1893 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1894 			to.to_tsecr = tp->ts_recent;
1895 			to.to_flags |= TOF_TS;
1896 		}
1897 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1898 		/* TCP-MD5 (RFC2385). */
1899 		if (tp->t_flags & TF_SIGNATURE)
1900 			to.to_flags |= TOF_SIGNATURE;
1901 #endif
1902 		/* Add the options. */
1903 		tlen += optlen = tcp_addoptions(&to, optp);
1904 
1905 		/* Update m_len in the correct mbuf. */
1906 		optm->m_len += optlen;
1907 	} else
1908 		optlen = 0;
1909 #ifdef INET6
1910 	if (isipv6) {
1911 		if (uh) {
1912 			ulen = tlen - sizeof(struct ip6_hdr);
1913 			uh->uh_ulen = htons(ulen);
1914 		}
1915 		ip6->ip6_flow = 0;
1916 		ip6->ip6_vfc = IPV6_VERSION;
1917 		if (port)
1918 			ip6->ip6_nxt = IPPROTO_UDP;
1919 		else
1920 			ip6->ip6_nxt = IPPROTO_TCP;
1921 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1922 	}
1923 #endif
1924 #if defined(INET) && defined(INET6)
1925 	else
1926 #endif
1927 #ifdef INET
1928 	{
1929 		if (uh) {
1930 			ulen = tlen - sizeof(struct ip);
1931 			uh->uh_ulen = htons(ulen);
1932 		}
1933 		ip->ip_len = htons(tlen);
1934 		ip->ip_ttl = V_ip_defttl;
1935 		if (port) {
1936 			ip->ip_p = IPPROTO_UDP;
1937 		} else {
1938 			ip->ip_p = IPPROTO_TCP;
1939 		}
1940 		if (V_path_mtu_discovery)
1941 			ip->ip_off |= htons(IP_DF);
1942 	}
1943 #endif
1944 	m->m_pkthdr.len = tlen;
1945 	m->m_pkthdr.rcvif = NULL;
1946 #ifdef MAC
1947 	if (inp != NULL) {
1948 		/*
1949 		 * Packet is associated with a socket, so allow the
1950 		 * label of the response to reflect the socket label.
1951 		 */
1952 		INP_LOCK_ASSERT(inp);
1953 		mac_inpcb_create_mbuf(inp, m);
1954 	} else {
1955 		/*
1956 		 * Packet is not associated with a socket, so possibly
1957 		 * update the label in place.
1958 		 */
1959 		mac_netinet_tcp_reply(m);
1960 	}
1961 #endif
1962 	nth->th_seq = htonl(seq);
1963 	nth->th_ack = htonl(ack);
1964 	nth->th_x2 = 0;
1965 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1966 	nth->th_flags = flags;
1967 	if (tp != NULL)
1968 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1969 	else
1970 		nth->th_win = htons((u_short)win);
1971 	nth->th_urp = 0;
1972 
1973 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1974 	if (to.to_flags & TOF_SIGNATURE) {
1975 		if (!TCPMD5_ENABLED() ||
1976 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1977 			m_freem(m);
1978 			return;
1979 		}
1980 	}
1981 #endif
1982 
1983 #ifdef INET6
1984 	if (isipv6) {
1985 		if (port) {
1986 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
1987 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1988 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
1989 			nth->th_sum = 0;
1990 		} else {
1991 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1992 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1993 			nth->th_sum = in6_cksum_pseudo(ip6,
1994 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1995 		}
1996 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1997 		    NULL, NULL);
1998 	}
1999 #endif /* INET6 */
2000 #if defined(INET6) && defined(INET)
2001 	else
2002 #endif
2003 #ifdef INET
2004 	{
2005 		if (port) {
2006 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2007 			    htons(ulen + IPPROTO_UDP));
2008 			m->m_pkthdr.csum_flags = CSUM_UDP;
2009 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2010 			nth->th_sum = 0;
2011 		} else {
2012 			m->m_pkthdr.csum_flags = CSUM_TCP;
2013 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2014 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2015 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2016 		}
2017 	}
2018 #endif /* INET */
2019 #ifdef TCPDEBUG
2020 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2021 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2022 #endif
2023 	TCP_PROBE3(debug__output, tp, th, m);
2024 	if (flags & TH_RST)
2025 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2026 
2027 #ifdef INET6
2028 	if (isipv6) {
2029 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2030 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2031 	}
2032 #endif /* INET6 */
2033 #if defined(INET) && defined(INET6)
2034 	else
2035 #endif
2036 #ifdef INET
2037 	{
2038 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2039 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
2040 	}
2041 #endif
2042 }
2043 
2044 /*
2045  * Create a new TCP control block, making an
2046  * empty reassembly queue and hooking it to the argument
2047  * protocol control block.  The `inp' parameter must have
2048  * come from the zone allocator set up in tcp_init().
2049  */
2050 struct tcpcb *
2051 tcp_newtcpcb(struct inpcb *inp)
2052 {
2053 	struct tcpcb_mem *tm;
2054 	struct tcpcb *tp;
2055 #ifdef INET6
2056 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2057 #endif /* INET6 */
2058 
2059 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2060 	if (tm == NULL)
2061 		return (NULL);
2062 	tp = &tm->tcb;
2063 
2064 	/* Initialise cc_var struct for this tcpcb. */
2065 	tp->ccv = &tm->ccv;
2066 	tp->ccv->type = IPPROTO_TCP;
2067 	tp->ccv->ccvc.tcp = tp;
2068 	rw_rlock(&tcp_function_lock);
2069 	tp->t_fb = tcp_func_set_ptr;
2070 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2071 	rw_runlock(&tcp_function_lock);
2072 	/*
2073 	 * Use the current system default CC algorithm.
2074 	 */
2075 	CC_LIST_RLOCK();
2076 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2077 	CC_ALGO(tp) = CC_DEFAULT();
2078 	CC_LIST_RUNLOCK();
2079 	/*
2080 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2081 	 * is called.
2082 	 */
2083 	in_pcbref(inp);	/* Reference for tcpcb */
2084 	tp->t_inpcb = inp;
2085 
2086 	if (CC_ALGO(tp)->cb_init != NULL)
2087 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
2088 			if (tp->t_fb->tfb_tcp_fb_fini)
2089 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2090 			in_pcbrele_wlocked(inp);
2091 			refcount_release(&tp->t_fb->tfb_refcnt);
2092 			uma_zfree(V_tcpcb_zone, tm);
2093 			return (NULL);
2094 		}
2095 
2096 #ifdef TCP_HHOOK
2097 	tp->osd = &tm->osd;
2098 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2099 		if (tp->t_fb->tfb_tcp_fb_fini)
2100 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2101 		in_pcbrele_wlocked(inp);
2102 		refcount_release(&tp->t_fb->tfb_refcnt);
2103 		uma_zfree(V_tcpcb_zone, tm);
2104 		return (NULL);
2105 	}
2106 #endif
2107 
2108 #ifdef VIMAGE
2109 	tp->t_vnet = inp->inp_vnet;
2110 #endif
2111 	tp->t_timers = &tm->tt;
2112 	TAILQ_INIT(&tp->t_segq);
2113 	tp->t_maxseg =
2114 #ifdef INET6
2115 		isipv6 ? V_tcp_v6mssdflt :
2116 #endif /* INET6 */
2117 		V_tcp_mssdflt;
2118 
2119 	/* Set up our timeouts. */
2120 	callout_init(&tp->t_timers->tt_rexmt, 1);
2121 	callout_init(&tp->t_timers->tt_persist, 1);
2122 	callout_init(&tp->t_timers->tt_keep, 1);
2123 	callout_init(&tp->t_timers->tt_2msl, 1);
2124 	callout_init(&tp->t_timers->tt_delack, 1);
2125 
2126 	if (V_tcp_do_rfc1323)
2127 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2128 	if (V_tcp_do_sack)
2129 		tp->t_flags |= TF_SACK_PERMIT;
2130 	TAILQ_INIT(&tp->snd_holes);
2131 
2132 	/*
2133 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2134 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2135 	 * reasonable initial retransmit time.
2136 	 */
2137 	tp->t_srtt = TCPTV_SRTTBASE;
2138 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2139 	tp->t_rttmin = tcp_rexmit_min;
2140 	tp->t_rxtcur = tcp_rexmit_initial;
2141 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2142 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2143 	tp->t_rcvtime = ticks;
2144 	/*
2145 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2146 	 * because the socket may be bound to an IPv6 wildcard address,
2147 	 * which may match an IPv4-mapped IPv6 address.
2148 	 */
2149 	inp->inp_ip_ttl = V_ip_defttl;
2150 	inp->inp_ppcb = tp;
2151 #ifdef TCPPCAP
2152 	/*
2153 	 * Init the TCP PCAP queues.
2154 	 */
2155 	tcp_pcap_tcpcb_init(tp);
2156 #endif
2157 #ifdef TCP_BLACKBOX
2158 	/* Initialize the per-TCPCB log data. */
2159 	tcp_log_tcpcbinit(tp);
2160 #endif
2161 	tp->t_pacing_rate = -1;
2162 	if (tp->t_fb->tfb_tcp_fb_init) {
2163 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2164 			refcount_release(&tp->t_fb->tfb_refcnt);
2165 			in_pcbrele_wlocked(inp);
2166 			uma_zfree(V_tcpcb_zone, tm);
2167 			return (NULL);
2168 		}
2169 	}
2170 #ifdef STATS
2171 	if (V_tcp_perconn_stats_enable == 1)
2172 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2173 #endif
2174 	return (tp);		/* XXX */
2175 }
2176 
2177 /*
2178  * Switch the congestion control algorithm back to NewReno for any active
2179  * control blocks using an algorithm which is about to go away.
2180  * This ensures the CC framework can allow the unload to proceed without leaving
2181  * any dangling pointers which would trigger a panic.
2182  * Returning non-zero would inform the CC framework that something went wrong
2183  * and it would be unsafe to allow the unload to proceed. However, there is no
2184  * way for this to occur with this implementation so we always return zero.
2185  */
2186 int
2187 tcp_ccalgounload(struct cc_algo *unload_algo)
2188 {
2189 	struct cc_algo *tmpalgo;
2190 	struct inpcb *inp;
2191 	struct tcpcb *tp;
2192 	VNET_ITERATOR_DECL(vnet_iter);
2193 
2194 	/*
2195 	 * Check all active control blocks across all network stacks and change
2196 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
2197 	 * requires cleanup code to be run, call it.
2198 	 */
2199 	VNET_LIST_RLOCK();
2200 	VNET_FOREACH(vnet_iter) {
2201 		CURVNET_SET(vnet_iter);
2202 		INP_INFO_WLOCK(&V_tcbinfo);
2203 		/*
2204 		 * New connections already part way through being initialised
2205 		 * with the CC algo we're removing will not race with this code
2206 		 * because the INP_INFO_WLOCK is held during initialisation. We
2207 		 * therefore don't enter the loop below until the connection
2208 		 * list has stabilised.
2209 		 */
2210 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
2211 			INP_WLOCK(inp);
2212 			/* Important to skip tcptw structs. */
2213 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
2214 			    (tp = intotcpcb(inp)) != NULL) {
2215 				/*
2216 				 * By holding INP_WLOCK here, we are assured
2217 				 * that the connection is not currently
2218 				 * executing inside the CC module's functions
2219 				 * i.e. it is safe to make the switch back to
2220 				 * NewReno.
2221 				 */
2222 				if (CC_ALGO(tp) == unload_algo) {
2223 					tmpalgo = CC_ALGO(tp);
2224 					if (tmpalgo->cb_destroy != NULL)
2225 						tmpalgo->cb_destroy(tp->ccv);
2226 					CC_DATA(tp) = NULL;
2227 					/*
2228 					 * NewReno may allocate memory on
2229 					 * demand for certain stateful
2230 					 * configuration as needed, but is
2231 					 * coded to never fail on memory
2232 					 * allocation failure so it is a safe
2233 					 * fallback.
2234 					 */
2235 					CC_ALGO(tp) = &newreno_cc_algo;
2236 				}
2237 			}
2238 			INP_WUNLOCK(inp);
2239 		}
2240 		INP_INFO_WUNLOCK(&V_tcbinfo);
2241 		CURVNET_RESTORE();
2242 	}
2243 	VNET_LIST_RUNLOCK();
2244 
2245 	return (0);
2246 }
2247 
2248 /*
2249  * Drop a TCP connection, reporting
2250  * the specified error.  If connection is synchronized,
2251  * then send a RST to peer.
2252  */
2253 struct tcpcb *
2254 tcp_drop(struct tcpcb *tp, int errno)
2255 {
2256 	struct socket *so = tp->t_inpcb->inp_socket;
2257 
2258 	NET_EPOCH_ASSERT();
2259 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2260 	INP_WLOCK_ASSERT(tp->t_inpcb);
2261 
2262 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2263 		tcp_state_change(tp, TCPS_CLOSED);
2264 		(void) tp->t_fb->tfb_tcp_output(tp);
2265 		TCPSTAT_INC(tcps_drops);
2266 	} else
2267 		TCPSTAT_INC(tcps_conndrops);
2268 	if (errno == ETIMEDOUT && tp->t_softerror)
2269 		errno = tp->t_softerror;
2270 	so->so_error = errno;
2271 	return (tcp_close(tp));
2272 }
2273 
2274 void
2275 tcp_discardcb(struct tcpcb *tp)
2276 {
2277 	struct inpcb *inp = tp->t_inpcb;
2278 	struct socket *so = inp->inp_socket;
2279 #ifdef INET6
2280 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2281 #endif /* INET6 */
2282 	int released __unused;
2283 
2284 	INP_WLOCK_ASSERT(inp);
2285 
2286 	/*
2287 	 * Make sure that all of our timers are stopped before we delete the
2288 	 * PCB.
2289 	 *
2290 	 * If stopping a timer fails, we schedule a discard function in same
2291 	 * callout, and the last discard function called will take care of
2292 	 * deleting the tcpcb.
2293 	 */
2294 	tp->t_timers->tt_draincnt = 0;
2295 	tcp_timer_stop(tp, TT_REXMT);
2296 	tcp_timer_stop(tp, TT_PERSIST);
2297 	tcp_timer_stop(tp, TT_KEEP);
2298 	tcp_timer_stop(tp, TT_2MSL);
2299 	tcp_timer_stop(tp, TT_DELACK);
2300 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2301 		/*
2302 		 * Call the stop-all function of the methods,
2303 		 * this function should call the tcp_timer_stop()
2304 		 * method with each of the function specific timeouts.
2305 		 * That stop will be called via the tfb_tcp_timer_stop()
2306 		 * which should use the async drain function of the
2307 		 * callout system (see tcp_var.h).
2308 		 */
2309 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2310 	}
2311 
2312 	/*
2313 	 * If we got enough samples through the srtt filter,
2314 	 * save the rtt and rttvar in the routing entry.
2315 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2316 	 * 4 samples is enough for the srtt filter to converge
2317 	 * to within enough % of the correct value; fewer samples
2318 	 * and we could save a bogus rtt. The danger is not high
2319 	 * as tcp quickly recovers from everything.
2320 	 * XXX: Works very well but needs some more statistics!
2321 	 */
2322 	if (tp->t_rttupdated >= 4) {
2323 		struct hc_metrics_lite metrics;
2324 		uint32_t ssthresh;
2325 
2326 		bzero(&metrics, sizeof(metrics));
2327 		/*
2328 		 * Update the ssthresh always when the conditions below
2329 		 * are satisfied. This gives us better new start value
2330 		 * for the congestion avoidance for new connections.
2331 		 * ssthresh is only set if packet loss occurred on a session.
2332 		 *
2333 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2334 		 * being torn down.  Ideally this code would not use 'so'.
2335 		 */
2336 		ssthresh = tp->snd_ssthresh;
2337 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2338 			/*
2339 			 * convert the limit from user data bytes to
2340 			 * packets then to packet data bytes.
2341 			 */
2342 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2343 			if (ssthresh < 2)
2344 				ssthresh = 2;
2345 			ssthresh *= (tp->t_maxseg +
2346 #ifdef INET6
2347 			    (isipv6 ? sizeof (struct ip6_hdr) +
2348 				sizeof (struct tcphdr) :
2349 #endif
2350 				sizeof (struct tcpiphdr)
2351 #ifdef INET6
2352 			    )
2353 #endif
2354 			    );
2355 		} else
2356 			ssthresh = 0;
2357 		metrics.rmx_ssthresh = ssthresh;
2358 
2359 		metrics.rmx_rtt = tp->t_srtt;
2360 		metrics.rmx_rttvar = tp->t_rttvar;
2361 		metrics.rmx_cwnd = tp->snd_cwnd;
2362 		metrics.rmx_sendpipe = 0;
2363 		metrics.rmx_recvpipe = 0;
2364 
2365 		tcp_hc_update(&inp->inp_inc, &metrics);
2366 	}
2367 
2368 	/* free the reassembly queue, if any */
2369 	tcp_reass_flush(tp);
2370 
2371 #ifdef TCP_OFFLOAD
2372 	/* Disconnect offload device, if any. */
2373 	if (tp->t_flags & TF_TOE)
2374 		tcp_offload_detach(tp);
2375 #endif
2376 
2377 	tcp_free_sackholes(tp);
2378 
2379 #ifdef TCPPCAP
2380 	/* Free the TCP PCAP queues. */
2381 	tcp_pcap_drain(&(tp->t_inpkts));
2382 	tcp_pcap_drain(&(tp->t_outpkts));
2383 #endif
2384 
2385 	/* Allow the CC algorithm to clean up after itself. */
2386 	if (CC_ALGO(tp)->cb_destroy != NULL)
2387 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2388 	CC_DATA(tp) = NULL;
2389 
2390 #ifdef TCP_HHOOK
2391 	khelp_destroy_osd(tp->osd);
2392 #endif
2393 #ifdef STATS
2394 	stats_blob_destroy(tp->t_stats);
2395 #endif
2396 
2397 	CC_ALGO(tp) = NULL;
2398 	inp->inp_ppcb = NULL;
2399 	if (tp->t_timers->tt_draincnt == 0) {
2400 		/* We own the last reference on tcpcb, let's free it. */
2401 #ifdef TCP_BLACKBOX
2402 		tcp_log_tcpcbfini(tp);
2403 #endif
2404 		TCPSTATES_DEC(tp->t_state);
2405 		if (tp->t_fb->tfb_tcp_fb_fini)
2406 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2407 		refcount_release(&tp->t_fb->tfb_refcnt);
2408 		tp->t_inpcb = NULL;
2409 		uma_zfree(V_tcpcb_zone, tp);
2410 		released = in_pcbrele_wlocked(inp);
2411 		KASSERT(!released, ("%s: inp %p should not have been released "
2412 			"here", __func__, inp));
2413 	}
2414 }
2415 
2416 void
2417 tcp_timer_discard(void *ptp)
2418 {
2419 	struct inpcb *inp;
2420 	struct tcpcb *tp;
2421 	struct epoch_tracker et;
2422 
2423 	tp = (struct tcpcb *)ptp;
2424 	CURVNET_SET(tp->t_vnet);
2425 	NET_EPOCH_ENTER(et);
2426 	inp = tp->t_inpcb;
2427 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2428 		__func__, tp));
2429 	INP_WLOCK(inp);
2430 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2431 		("%s: tcpcb has to be stopped here", __func__));
2432 	tp->t_timers->tt_draincnt--;
2433 	if (tp->t_timers->tt_draincnt == 0) {
2434 		/* We own the last reference on this tcpcb, let's free it. */
2435 #ifdef TCP_BLACKBOX
2436 		tcp_log_tcpcbfini(tp);
2437 #endif
2438 		TCPSTATES_DEC(tp->t_state);
2439 		if (tp->t_fb->tfb_tcp_fb_fini)
2440 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2441 		refcount_release(&tp->t_fb->tfb_refcnt);
2442 		tp->t_inpcb = NULL;
2443 		uma_zfree(V_tcpcb_zone, tp);
2444 		if (in_pcbrele_wlocked(inp)) {
2445 			NET_EPOCH_EXIT(et);
2446 			CURVNET_RESTORE();
2447 			return;
2448 		}
2449 	}
2450 	INP_WUNLOCK(inp);
2451 	NET_EPOCH_EXIT(et);
2452 	CURVNET_RESTORE();
2453 }
2454 
2455 /*
2456  * Attempt to close a TCP control block, marking it as dropped, and freeing
2457  * the socket if we hold the only reference.
2458  */
2459 struct tcpcb *
2460 tcp_close(struct tcpcb *tp)
2461 {
2462 	struct inpcb *inp = tp->t_inpcb;
2463 	struct socket *so;
2464 
2465 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2466 	INP_WLOCK_ASSERT(inp);
2467 
2468 #ifdef TCP_OFFLOAD
2469 	if (tp->t_state == TCPS_LISTEN)
2470 		tcp_offload_listen_stop(tp);
2471 #endif
2472 	/*
2473 	 * This releases the TFO pending counter resource for TFO listen
2474 	 * sockets as well as passively-created TFO sockets that transition
2475 	 * from SYN_RECEIVED to CLOSED.
2476 	 */
2477 	if (tp->t_tfo_pending) {
2478 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2479 		tp->t_tfo_pending = NULL;
2480 	}
2481 	in_pcbdrop(inp);
2482 	TCPSTAT_INC(tcps_closed);
2483 	if (tp->t_state != TCPS_CLOSED)
2484 		tcp_state_change(tp, TCPS_CLOSED);
2485 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2486 	so = inp->inp_socket;
2487 	soisdisconnected(so);
2488 	if (inp->inp_flags & INP_SOCKREF) {
2489 		KASSERT(so->so_state & SS_PROTOREF,
2490 		    ("tcp_close: !SS_PROTOREF"));
2491 		inp->inp_flags &= ~INP_SOCKREF;
2492 		INP_WUNLOCK(inp);
2493 		SOCK_LOCK(so);
2494 		so->so_state &= ~SS_PROTOREF;
2495 		sofree(so);
2496 		return (NULL);
2497 	}
2498 	return (tp);
2499 }
2500 
2501 void
2502 tcp_drain(void)
2503 {
2504 	VNET_ITERATOR_DECL(vnet_iter);
2505 
2506 	if (!do_tcpdrain)
2507 		return;
2508 
2509 	VNET_LIST_RLOCK_NOSLEEP();
2510 	VNET_FOREACH(vnet_iter) {
2511 		CURVNET_SET(vnet_iter);
2512 		struct inpcb *inpb;
2513 		struct tcpcb *tcpb;
2514 
2515 	/*
2516 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2517 	 * if there is one...
2518 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2519 	 *      reassembly queue should be flushed, but in a situation
2520 	 *	where we're really low on mbufs, this is potentially
2521 	 *	useful.
2522 	 */
2523 		INP_INFO_WLOCK(&V_tcbinfo);
2524 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2525 			INP_WLOCK(inpb);
2526 			if (inpb->inp_flags & INP_TIMEWAIT) {
2527 				INP_WUNLOCK(inpb);
2528 				continue;
2529 			}
2530 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2531 				tcp_reass_flush(tcpb);
2532 				tcp_clean_sackreport(tcpb);
2533 #ifdef TCP_BLACKBOX
2534 				tcp_log_drain(tcpb);
2535 #endif
2536 #ifdef TCPPCAP
2537 				if (tcp_pcap_aggressive_free) {
2538 					/* Free the TCP PCAP queues. */
2539 					tcp_pcap_drain(&(tcpb->t_inpkts));
2540 					tcp_pcap_drain(&(tcpb->t_outpkts));
2541 				}
2542 #endif
2543 			}
2544 			INP_WUNLOCK(inpb);
2545 		}
2546 		INP_INFO_WUNLOCK(&V_tcbinfo);
2547 		CURVNET_RESTORE();
2548 	}
2549 	VNET_LIST_RUNLOCK_NOSLEEP();
2550 }
2551 
2552 /*
2553  * Notify a tcp user of an asynchronous error;
2554  * store error as soft error, but wake up user
2555  * (for now, won't do anything until can select for soft error).
2556  *
2557  * Do not wake up user since there currently is no mechanism for
2558  * reporting soft errors (yet - a kqueue filter may be added).
2559  */
2560 static struct inpcb *
2561 tcp_notify(struct inpcb *inp, int error)
2562 {
2563 	struct tcpcb *tp;
2564 
2565 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2566 	INP_WLOCK_ASSERT(inp);
2567 
2568 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2569 	    (inp->inp_flags & INP_DROPPED))
2570 		return (inp);
2571 
2572 	tp = intotcpcb(inp);
2573 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2574 
2575 	/*
2576 	 * Ignore some errors if we are hooked up.
2577 	 * If connection hasn't completed, has retransmitted several times,
2578 	 * and receives a second error, give up now.  This is better
2579 	 * than waiting a long time to establish a connection that
2580 	 * can never complete.
2581 	 */
2582 	if (tp->t_state == TCPS_ESTABLISHED &&
2583 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2584 	     error == EHOSTDOWN)) {
2585 		if (inp->inp_route.ro_nh) {
2586 			NH_FREE(inp->inp_route.ro_nh);
2587 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2588 		}
2589 		return (inp);
2590 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2591 	    tp->t_softerror) {
2592 		tp = tcp_drop(tp, error);
2593 		if (tp != NULL)
2594 			return (inp);
2595 		else
2596 			return (NULL);
2597 	} else {
2598 		tp->t_softerror = error;
2599 		return (inp);
2600 	}
2601 #if 0
2602 	wakeup( &so->so_timeo);
2603 	sorwakeup(so);
2604 	sowwakeup(so);
2605 #endif
2606 }
2607 
2608 static int
2609 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2610 {
2611 	struct epoch_tracker et;
2612 	struct inpcb *inp;
2613 	struct xinpgen xig;
2614 	int error;
2615 
2616 	if (req->newptr != NULL)
2617 		return (EPERM);
2618 
2619 	if (req->oldptr == NULL) {
2620 		int n;
2621 
2622 		n = V_tcbinfo.ipi_count +
2623 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2624 		n += imax(n / 8, 10);
2625 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2626 		return (0);
2627 	}
2628 
2629 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2630 		return (error);
2631 
2632 	bzero(&xig, sizeof(xig));
2633 	xig.xig_len = sizeof xig;
2634 	xig.xig_count = V_tcbinfo.ipi_count +
2635 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2636 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2637 	xig.xig_sogen = so_gencnt;
2638 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2639 	if (error)
2640 		return (error);
2641 
2642 	error = syncache_pcblist(req);
2643 	if (error)
2644 		return (error);
2645 
2646 	NET_EPOCH_ENTER(et);
2647 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2648 	    inp != NULL;
2649 	    inp = CK_LIST_NEXT(inp, inp_list)) {
2650 		INP_RLOCK(inp);
2651 		if (inp->inp_gencnt <= xig.xig_gen) {
2652 			int crerr;
2653 
2654 			/*
2655 			 * XXX: This use of cr_cansee(), introduced with
2656 			 * TCP state changes, is not quite right, but for
2657 			 * now, better than nothing.
2658 			 */
2659 			if (inp->inp_flags & INP_TIMEWAIT) {
2660 				if (intotw(inp) != NULL)
2661 					crerr = cr_cansee(req->td->td_ucred,
2662 					    intotw(inp)->tw_cred);
2663 				else
2664 					crerr = EINVAL;	/* Skip this inp. */
2665 			} else
2666 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2667 			if (crerr == 0) {
2668 				struct xtcpcb xt;
2669 
2670 				tcp_inptoxtp(inp, &xt);
2671 				INP_RUNLOCK(inp);
2672 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2673 				if (error)
2674 					break;
2675 				else
2676 					continue;
2677 			}
2678 		}
2679 		INP_RUNLOCK(inp);
2680 	}
2681 	NET_EPOCH_EXIT(et);
2682 
2683 	if (!error) {
2684 		/*
2685 		 * Give the user an updated idea of our state.
2686 		 * If the generation differs from what we told
2687 		 * her before, she knows that something happened
2688 		 * while we were processing this request, and it
2689 		 * might be necessary to retry.
2690 		 */
2691 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2692 		xig.xig_sogen = so_gencnt;
2693 		xig.xig_count = V_tcbinfo.ipi_count +
2694 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2695 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2696 	}
2697 
2698 	return (error);
2699 }
2700 
2701 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2702     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2703     NULL, 0, tcp_pcblist, "S,xtcpcb",
2704     "List of active TCP connections");
2705 
2706 #ifdef INET
2707 static int
2708 tcp_getcred(SYSCTL_HANDLER_ARGS)
2709 {
2710 	struct xucred xuc;
2711 	struct sockaddr_in addrs[2];
2712 	struct epoch_tracker et;
2713 	struct inpcb *inp;
2714 	int error;
2715 
2716 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2717 	if (error)
2718 		return (error);
2719 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2720 	if (error)
2721 		return (error);
2722 	NET_EPOCH_ENTER(et);
2723 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2724 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2725 	NET_EPOCH_EXIT(et);
2726 	if (inp != NULL) {
2727 		if (inp->inp_socket == NULL)
2728 			error = ENOENT;
2729 		if (error == 0)
2730 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2731 		if (error == 0)
2732 			cru2x(inp->inp_cred, &xuc);
2733 		INP_RUNLOCK(inp);
2734 	} else
2735 		error = ENOENT;
2736 	if (error == 0)
2737 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2738 	return (error);
2739 }
2740 
2741 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2742     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2743     0, 0, tcp_getcred, "S,xucred",
2744     "Get the xucred of a TCP connection");
2745 #endif /* INET */
2746 
2747 #ifdef INET6
2748 static int
2749 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2750 {
2751 	struct epoch_tracker et;
2752 	struct xucred xuc;
2753 	struct sockaddr_in6 addrs[2];
2754 	struct inpcb *inp;
2755 	int error;
2756 #ifdef INET
2757 	int mapped = 0;
2758 #endif
2759 
2760 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2761 	if (error)
2762 		return (error);
2763 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2764 	if (error)
2765 		return (error);
2766 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2767 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2768 		return (error);
2769 	}
2770 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2771 #ifdef INET
2772 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2773 			mapped = 1;
2774 		else
2775 #endif
2776 			return (EINVAL);
2777 	}
2778 
2779 	NET_EPOCH_ENTER(et);
2780 #ifdef INET
2781 	if (mapped == 1)
2782 		inp = in_pcblookup(&V_tcbinfo,
2783 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2784 			addrs[1].sin6_port,
2785 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2786 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2787 	else
2788 #endif
2789 		inp = in6_pcblookup(&V_tcbinfo,
2790 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2791 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2792 			INPLOOKUP_RLOCKPCB, NULL);
2793 	NET_EPOCH_EXIT(et);
2794 	if (inp != NULL) {
2795 		if (inp->inp_socket == NULL)
2796 			error = ENOENT;
2797 		if (error == 0)
2798 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2799 		if (error == 0)
2800 			cru2x(inp->inp_cred, &xuc);
2801 		INP_RUNLOCK(inp);
2802 	} else
2803 		error = ENOENT;
2804 	if (error == 0)
2805 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2806 	return (error);
2807 }
2808 
2809 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2810     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2811     0, 0, tcp6_getcred, "S,xucred",
2812     "Get the xucred of a TCP6 connection");
2813 #endif /* INET6 */
2814 
2815 #ifdef INET
2816 /* Path MTU to try next when a fragmentation-needed message is received. */
2817 static inline int
2818 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2819 {
2820 	int mtu = ntohs(icp->icmp_nextmtu);
2821 
2822 	/* If no alternative MTU was proposed, try the next smaller one. */
2823 	if (!mtu)
2824 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2825 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2826 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2827 
2828 	return (mtu);
2829 }
2830 
2831 static void
2832 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2833 {
2834 	struct ip *ip = vip;
2835 	struct tcphdr *th;
2836 	struct in_addr faddr;
2837 	struct inpcb *inp;
2838 	struct tcpcb *tp;
2839 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2840 	struct icmp *icp;
2841 	struct in_conninfo inc;
2842 	tcp_seq icmp_tcp_seq;
2843 	int mtu;
2844 
2845 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2846 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2847 		return;
2848 
2849 	if (cmd == PRC_MSGSIZE)
2850 		notify = tcp_mtudisc_notify;
2851 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2852 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2853 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2854 		notify = tcp_drop_syn_sent;
2855 
2856 	/*
2857 	 * Hostdead is ugly because it goes linearly through all PCBs.
2858 	 * XXX: We never get this from ICMP, otherwise it makes an
2859 	 * excellent DoS attack on machines with many connections.
2860 	 */
2861 	else if (cmd == PRC_HOSTDEAD)
2862 		ip = NULL;
2863 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2864 		return;
2865 
2866 	if (ip == NULL) {
2867 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2868 		return;
2869 	}
2870 
2871 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2872 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2873 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2874 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2875 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2876 		/* signal EHOSTDOWN, as it flushes the cached route */
2877 		inp = (*notify)(inp, EHOSTDOWN);
2878 		goto out;
2879 	}
2880 	icmp_tcp_seq = th->th_seq;
2881 	if (inp != NULL)  {
2882 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2883 		    !(inp->inp_flags & INP_DROPPED) &&
2884 		    !(inp->inp_socket == NULL)) {
2885 			tp = intotcpcb(inp);
2886 #ifdef TCP_OFFLOAD
2887 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2888 				/*
2889 				 * MTU discovery for offloaded connections.  Let
2890 				 * the TOE driver verify seq# and process it.
2891 				 */
2892 				mtu = tcp_next_pmtu(icp, ip);
2893 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2894 				goto out;
2895 			}
2896 #endif
2897 			if (tp->t_port != port) {
2898 				goto out;
2899 			}
2900 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2901 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2902 				if (cmd == PRC_MSGSIZE) {
2903 					/*
2904 					 * MTU discovery: we got a needfrag and
2905 					 * will potentially try a lower MTU.
2906 					 */
2907 					mtu = tcp_next_pmtu(icp, ip);
2908 
2909 					/*
2910 					 * Only process the offered MTU if it
2911 					 * is smaller than the current one.
2912 					 */
2913 					if (mtu < tp->t_maxseg +
2914 					    sizeof(struct tcpiphdr)) {
2915 						bzero(&inc, sizeof(inc));
2916 						inc.inc_faddr = faddr;
2917 						inc.inc_fibnum =
2918 						    inp->inp_inc.inc_fibnum;
2919 						tcp_hc_updatemtu(&inc, mtu);
2920 						tcp_mtudisc(inp, mtu);
2921 					}
2922 				} else
2923 					inp = (*notify)(inp,
2924 					    inetctlerrmap[cmd]);
2925 			}
2926 		}
2927 	} else {
2928 		bzero(&inc, sizeof(inc));
2929 		inc.inc_fport = th->th_dport;
2930 		inc.inc_lport = th->th_sport;
2931 		inc.inc_faddr = faddr;
2932 		inc.inc_laddr = ip->ip_src;
2933 		syncache_unreach(&inc, icmp_tcp_seq, port);
2934 	}
2935 out:
2936 	if (inp != NULL)
2937 		INP_WUNLOCK(inp);
2938 }
2939 
2940 void
2941 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2942 {
2943 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2944 }
2945 
2946 void
2947 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2948 {
2949 	/* Its a tunneled TCP over UDP icmp */
2950 	struct ip *outer_ip, *inner_ip;
2951 	struct icmp *icmp;
2952 	struct udphdr *udp;
2953 	struct tcphdr *th, ttemp;
2954 	int i_hlen, o_len;
2955 	uint16_t port;
2956 
2957 	inner_ip = (struct ip *)vip;
2958 	icmp = (struct icmp *)((caddr_t)inner_ip -
2959 	    (sizeof(struct icmp) - sizeof(struct ip)));
2960 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2961 	i_hlen = inner_ip->ip_hl << 2;
2962 	o_len = ntohs(outer_ip->ip_len);
2963 	if (o_len <
2964 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2965 		/* Not enough data present */
2966 		return;
2967 	}
2968 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2969 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2970 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2971 		return;
2972 	}
2973 	port = udp->uh_dport;
2974 	th = (struct tcphdr *)(udp + 1);
2975 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2976 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2977 	/* Now adjust down the size of the outer IP header */
2978 	o_len -= sizeof(struct udphdr);
2979 	outer_ip->ip_len = htons(o_len);
2980 	/* Now call in to the normal handling code */
2981 	tcp_ctlinput_with_port(cmd, sa, vip, port);
2982 }
2983 #endif /* INET */
2984 
2985 #ifdef INET6
2986 static inline int
2987 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2988 {
2989 	int mtu = ntohl(icmp6->icmp6_mtu);
2990 
2991 	/*
2992 	 * If no alternative MTU was proposed, or the proposed MTU was too
2993 	 * small, set to the min.
2994 	 */
2995 	if (mtu < IPV6_MMTU)
2996 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
2997 	return (mtu);
2998 }
2999 
3000 static void
3001 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3002 {
3003 	struct in6_addr *dst;
3004 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3005 	struct ip6_hdr *ip6;
3006 	struct mbuf *m;
3007 	struct inpcb *inp;
3008 	struct tcpcb *tp;
3009 	struct icmp6_hdr *icmp6;
3010 	struct ip6ctlparam *ip6cp = NULL;
3011 	const struct sockaddr_in6 *sa6_src = NULL;
3012 	struct in_conninfo inc;
3013 	struct tcp_ports {
3014 		uint16_t th_sport;
3015 		uint16_t th_dport;
3016 	} t_ports;
3017 	tcp_seq icmp_tcp_seq;
3018 	unsigned int mtu;
3019 	unsigned int off;
3020 
3021 	if (sa->sa_family != AF_INET6 ||
3022 	    sa->sa_len != sizeof(struct sockaddr_in6))
3023 		return;
3024 
3025 	/* if the parameter is from icmp6, decode it. */
3026 	if (d != NULL) {
3027 		ip6cp = (struct ip6ctlparam *)d;
3028 		icmp6 = ip6cp->ip6c_icmp6;
3029 		m = ip6cp->ip6c_m;
3030 		ip6 = ip6cp->ip6c_ip6;
3031 		off = ip6cp->ip6c_off;
3032 		sa6_src = ip6cp->ip6c_src;
3033 		dst = ip6cp->ip6c_finaldst;
3034 	} else {
3035 		m = NULL;
3036 		ip6 = NULL;
3037 		off = 0;	/* fool gcc */
3038 		sa6_src = &sa6_any;
3039 		dst = NULL;
3040 	}
3041 
3042 	if (cmd == PRC_MSGSIZE)
3043 		notify = tcp_mtudisc_notify;
3044 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3045 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3046 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3047 		notify = tcp_drop_syn_sent;
3048 
3049 	/*
3050 	 * Hostdead is ugly because it goes linearly through all PCBs.
3051 	 * XXX: We never get this from ICMP, otherwise it makes an
3052 	 * excellent DoS attack on machines with many connections.
3053 	 */
3054 	else if (cmd == PRC_HOSTDEAD)
3055 		ip6 = NULL;
3056 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3057 		return;
3058 
3059 	if (ip6 == NULL) {
3060 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3061 			      (const struct sockaddr *)sa6_src,
3062 			      0, cmd, NULL, notify);
3063 		return;
3064 	}
3065 
3066 	/* Check if we can safely get the ports from the tcp hdr */
3067 	if (m == NULL ||
3068 	    (m->m_pkthdr.len <
3069 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3070 		return;
3071 	}
3072 	bzero(&t_ports, sizeof(struct tcp_ports));
3073 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3074 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3075 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3076 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3077 		/* signal EHOSTDOWN, as it flushes the cached route */
3078 		inp = (*notify)(inp, EHOSTDOWN);
3079 		goto out;
3080 	}
3081 	off += sizeof(struct tcp_ports);
3082 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3083 		goto out;
3084 	}
3085 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3086 	if (inp != NULL)  {
3087 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3088 		    !(inp->inp_flags & INP_DROPPED) &&
3089 		    !(inp->inp_socket == NULL)) {
3090 			tp = intotcpcb(inp);
3091 #ifdef TCP_OFFLOAD
3092 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3093 				/* MTU discovery for offloaded connections. */
3094 				mtu = tcp6_next_pmtu(icmp6);
3095 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3096 				goto out;
3097 			}
3098 #endif
3099 			if (tp->t_port != port) {
3100 				goto out;
3101 			}
3102 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3103 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3104 				if (cmd == PRC_MSGSIZE) {
3105 					/*
3106 					 * MTU discovery:
3107 					 * If we got a needfrag set the MTU
3108 					 * in the route to the suggested new
3109 					 * value (if given) and then notify.
3110 					 */
3111 					mtu = tcp6_next_pmtu(icmp6);
3112 
3113 					bzero(&inc, sizeof(inc));
3114 					inc.inc_fibnum = M_GETFIB(m);
3115 					inc.inc_flags |= INC_ISIPV6;
3116 					inc.inc6_faddr = *dst;
3117 					if (in6_setscope(&inc.inc6_faddr,
3118 						m->m_pkthdr.rcvif, NULL))
3119 						goto out;
3120 					/*
3121 					 * Only process the offered MTU if it
3122 					 * is smaller than the current one.
3123 					 */
3124 					if (mtu < tp->t_maxseg +
3125 					    sizeof (struct tcphdr) +
3126 					    sizeof (struct ip6_hdr)) {
3127 						tcp_hc_updatemtu(&inc, mtu);
3128 						tcp_mtudisc(inp, mtu);
3129 						ICMP6STAT_INC(icp6s_pmtuchg);
3130 					}
3131 				} else
3132 					inp = (*notify)(inp,
3133 					    inet6ctlerrmap[cmd]);
3134 			}
3135 		}
3136 	} else {
3137 		bzero(&inc, sizeof(inc));
3138 		inc.inc_fibnum = M_GETFIB(m);
3139 		inc.inc_flags |= INC_ISIPV6;
3140 		inc.inc_fport = t_ports.th_dport;
3141 		inc.inc_lport = t_ports.th_sport;
3142 		inc.inc6_faddr = *dst;
3143 		inc.inc6_laddr = ip6->ip6_src;
3144 		syncache_unreach(&inc, icmp_tcp_seq, port);
3145 	}
3146 out:
3147 	if (inp != NULL)
3148 		INP_WUNLOCK(inp);
3149 }
3150 
3151 void
3152 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3153 {
3154 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3155 }
3156 
3157 void
3158 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3159 {
3160 	struct ip6ctlparam *ip6cp;
3161 	struct mbuf *m;
3162 	struct udphdr *udp;
3163 	uint16_t port;
3164 
3165 	ip6cp = (struct ip6ctlparam *)d;
3166 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3167 	if (m == NULL) {
3168 		return;
3169 	}
3170 	udp = mtod(m, struct udphdr *);
3171 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3172 		return;
3173 	}
3174 	port = udp->uh_dport;
3175 	m_adj(m, sizeof(struct udphdr));
3176 	if ((m->m_flags & M_PKTHDR) == 0) {
3177 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3178 	}
3179 	/* Now call in to the normal handling code */
3180 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3181 }
3182 
3183 #endif /* INET6 */
3184 
3185 static uint32_t
3186 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3187 {
3188 	SIPHASH_CTX ctx;
3189 	uint32_t hash[2];
3190 
3191 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3192 	    ("%s: keylen %u too short ", __func__, len));
3193 	SipHash24_Init(&ctx);
3194 	SipHash_SetKey(&ctx, (uint8_t *)key);
3195 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3196 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3197 	switch (inc->inc_flags & INC_ISIPV6) {
3198 #ifdef INET
3199 	case 0:
3200 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3201 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3202 		break;
3203 #endif
3204 #ifdef INET6
3205 	case INC_ISIPV6:
3206 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3207 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3208 		break;
3209 #endif
3210 	}
3211 	SipHash_Final((uint8_t *)hash, &ctx);
3212 
3213 	return (hash[0] ^ hash[1]);
3214 }
3215 
3216 uint32_t
3217 tcp_new_ts_offset(struct in_conninfo *inc)
3218 {
3219 	struct in_conninfo inc_store, *local_inc;
3220 
3221 	if (!V_tcp_ts_offset_per_conn) {
3222 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3223 		inc_store.inc_lport = 0;
3224 		inc_store.inc_fport = 0;
3225 		local_inc = &inc_store;
3226 	} else {
3227 		local_inc = inc;
3228 	}
3229 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3230 	    sizeof(V_ts_offset_secret)));
3231 }
3232 
3233 /*
3234  * Following is where TCP initial sequence number generation occurs.
3235  *
3236  * There are two places where we must use initial sequence numbers:
3237  * 1.  In SYN-ACK packets.
3238  * 2.  In SYN packets.
3239  *
3240  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3241  * tcp_syncache.c for details.
3242  *
3243  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3244  * depends on this property.  In addition, these ISNs should be
3245  * unguessable so as to prevent connection hijacking.  To satisfy
3246  * the requirements of this situation, the algorithm outlined in
3247  * RFC 1948 is used, with only small modifications.
3248  *
3249  * Implementation details:
3250  *
3251  * Time is based off the system timer, and is corrected so that it
3252  * increases by one megabyte per second.  This allows for proper
3253  * recycling on high speed LANs while still leaving over an hour
3254  * before rollover.
3255  *
3256  * As reading the *exact* system time is too expensive to be done
3257  * whenever setting up a TCP connection, we increment the time
3258  * offset in two ways.  First, a small random positive increment
3259  * is added to isn_offset for each connection that is set up.
3260  * Second, the function tcp_isn_tick fires once per clock tick
3261  * and increments isn_offset as necessary so that sequence numbers
3262  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3263  * random positive increments serve only to ensure that the same
3264  * exact sequence number is never sent out twice (as could otherwise
3265  * happen when a port is recycled in less than the system tick
3266  * interval.)
3267  *
3268  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3269  * between seeding of isn_secret.  This is normally set to zero,
3270  * as reseeding should not be necessary.
3271  *
3272  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3273  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3274  * general, this means holding an exclusive (write) lock.
3275  */
3276 
3277 #define ISN_BYTES_PER_SECOND 1048576
3278 #define ISN_STATIC_INCREMENT 4096
3279 #define ISN_RANDOM_INCREMENT (4096 - 1)
3280 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3281 
3282 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3283 VNET_DEFINE_STATIC(int, isn_last);
3284 VNET_DEFINE_STATIC(int, isn_last_reseed);
3285 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3286 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3287 
3288 #define	V_isn_secret			VNET(isn_secret)
3289 #define	V_isn_last			VNET(isn_last)
3290 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3291 #define	V_isn_offset			VNET(isn_offset)
3292 #define	V_isn_offset_old		VNET(isn_offset_old)
3293 
3294 tcp_seq
3295 tcp_new_isn(struct in_conninfo *inc)
3296 {
3297 	tcp_seq new_isn;
3298 	u_int32_t projected_offset;
3299 
3300 	ISN_LOCK();
3301 	/* Seed if this is the first use, reseed if requested. */
3302 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3303 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3304 		< (u_int)ticks))) {
3305 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3306 		V_isn_last_reseed = ticks;
3307 	}
3308 
3309 	/* Compute the hash and return the ISN. */
3310 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3311 	    sizeof(V_isn_secret));
3312 	V_isn_offset += ISN_STATIC_INCREMENT +
3313 		(arc4random() & ISN_RANDOM_INCREMENT);
3314 	if (ticks != V_isn_last) {
3315 		projected_offset = V_isn_offset_old +
3316 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3317 		if (SEQ_GT(projected_offset, V_isn_offset))
3318 			V_isn_offset = projected_offset;
3319 		V_isn_offset_old = V_isn_offset;
3320 		V_isn_last = ticks;
3321 	}
3322 	new_isn += V_isn_offset;
3323 	ISN_UNLOCK();
3324 	return (new_isn);
3325 }
3326 
3327 /*
3328  * When a specific ICMP unreachable message is received and the
3329  * connection state is SYN-SENT, drop the connection.  This behavior
3330  * is controlled by the icmp_may_rst sysctl.
3331  */
3332 struct inpcb *
3333 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3334 {
3335 	struct tcpcb *tp;
3336 
3337 	NET_EPOCH_ASSERT();
3338 	INP_WLOCK_ASSERT(inp);
3339 
3340 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3341 	    (inp->inp_flags & INP_DROPPED))
3342 		return (inp);
3343 
3344 	tp = intotcpcb(inp);
3345 	if (tp->t_state != TCPS_SYN_SENT)
3346 		return (inp);
3347 
3348 	if (IS_FASTOPEN(tp->t_flags))
3349 		tcp_fastopen_disable_path(tp);
3350 
3351 	tp = tcp_drop(tp, errno);
3352 	if (tp != NULL)
3353 		return (inp);
3354 	else
3355 		return (NULL);
3356 }
3357 
3358 /*
3359  * When `need fragmentation' ICMP is received, update our idea of the MSS
3360  * based on the new value. Also nudge TCP to send something, since we
3361  * know the packet we just sent was dropped.
3362  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3363  */
3364 static struct inpcb *
3365 tcp_mtudisc_notify(struct inpcb *inp, int error)
3366 {
3367 
3368 	tcp_mtudisc(inp, -1);
3369 	return (inp);
3370 }
3371 
3372 static void
3373 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3374 {
3375 	struct tcpcb *tp;
3376 	struct socket *so;
3377 
3378 	INP_WLOCK_ASSERT(inp);
3379 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3380 	    (inp->inp_flags & INP_DROPPED))
3381 		return;
3382 
3383 	tp = intotcpcb(inp);
3384 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3385 
3386 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3387 
3388 	so = inp->inp_socket;
3389 	SOCKBUF_LOCK(&so->so_snd);
3390 	/* If the mss is larger than the socket buffer, decrease the mss. */
3391 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3392 		tp->t_maxseg = so->so_snd.sb_hiwat;
3393 	SOCKBUF_UNLOCK(&so->so_snd);
3394 
3395 	TCPSTAT_INC(tcps_mturesent);
3396 	tp->t_rtttime = 0;
3397 	tp->snd_nxt = tp->snd_una;
3398 	tcp_free_sackholes(tp);
3399 	tp->snd_recover = tp->snd_max;
3400 	if (tp->t_flags & TF_SACK_PERMIT)
3401 		EXIT_FASTRECOVERY(tp->t_flags);
3402 	tp->t_fb->tfb_tcp_output(tp);
3403 }
3404 
3405 #ifdef INET
3406 /*
3407  * Look-up the routing entry to the peer of this inpcb.  If no route
3408  * is found and it cannot be allocated, then return 0.  This routine
3409  * is called by TCP routines that access the rmx structure and by
3410  * tcp_mss_update to get the peer/interface MTU.
3411  */
3412 uint32_t
3413 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3414 {
3415 	struct nhop_object *nh;
3416 	struct ifnet *ifp;
3417 	uint32_t maxmtu = 0;
3418 
3419 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3420 
3421 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3422 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3423 		if (nh == NULL)
3424 			return (0);
3425 
3426 		ifp = nh->nh_ifp;
3427 		maxmtu = nh->nh_mtu;
3428 
3429 		/* Report additional interface capabilities. */
3430 		if (cap != NULL) {
3431 			if (ifp->if_capenable & IFCAP_TSO4 &&
3432 			    ifp->if_hwassist & CSUM_TSO) {
3433 				cap->ifcap |= CSUM_TSO;
3434 				cap->tsomax = ifp->if_hw_tsomax;
3435 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3436 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3437 			}
3438 		}
3439 	}
3440 	return (maxmtu);
3441 }
3442 #endif /* INET */
3443 
3444 #ifdef INET6
3445 uint32_t
3446 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3447 {
3448 	struct nhop_object *nh;
3449 	struct in6_addr dst6;
3450 	uint32_t scopeid;
3451 	struct ifnet *ifp;
3452 	uint32_t maxmtu = 0;
3453 
3454 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3455 
3456 	if (inc->inc_flags & INC_IPV6MINMTU)
3457 		return (IPV6_MMTU);
3458 
3459 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3460 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3461 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3462 		if (nh == NULL)
3463 			return (0);
3464 
3465 		ifp = nh->nh_ifp;
3466 		maxmtu = nh->nh_mtu;
3467 
3468 		/* Report additional interface capabilities. */
3469 		if (cap != NULL) {
3470 			if (ifp->if_capenable & IFCAP_TSO6 &&
3471 			    ifp->if_hwassist & CSUM_TSO) {
3472 				cap->ifcap |= CSUM_TSO;
3473 				cap->tsomax = ifp->if_hw_tsomax;
3474 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3475 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3476 			}
3477 		}
3478 	}
3479 
3480 	return (maxmtu);
3481 }
3482 #endif /* INET6 */
3483 
3484 /*
3485  * Calculate effective SMSS per RFC5681 definition for a given TCP
3486  * connection at its current state, taking into account SACK and etc.
3487  */
3488 u_int
3489 tcp_maxseg(const struct tcpcb *tp)
3490 {
3491 	u_int optlen;
3492 
3493 	if (tp->t_flags & TF_NOOPT)
3494 		return (tp->t_maxseg);
3495 
3496 	/*
3497 	 * Here we have a simplified code from tcp_addoptions(),
3498 	 * without a proper loop, and having most of paddings hardcoded.
3499 	 * We might make mistakes with padding here in some edge cases,
3500 	 * but this is harmless, since result of tcp_maxseg() is used
3501 	 * only in cwnd and ssthresh estimations.
3502 	 */
3503 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3504 		if (tp->t_flags & TF_RCVD_TSTMP)
3505 			optlen = TCPOLEN_TSTAMP_APPA;
3506 		else
3507 			optlen = 0;
3508 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3509 		if (tp->t_flags & TF_SIGNATURE)
3510 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3511 #endif
3512 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3513 			optlen += TCPOLEN_SACKHDR;
3514 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3515 			optlen = PADTCPOLEN(optlen);
3516 		}
3517 	} else {
3518 		if (tp->t_flags & TF_REQ_TSTMP)
3519 			optlen = TCPOLEN_TSTAMP_APPA;
3520 		else
3521 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3522 		if (tp->t_flags & TF_REQ_SCALE)
3523 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3524 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3525 		if (tp->t_flags & TF_SIGNATURE)
3526 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3527 #endif
3528 		if (tp->t_flags & TF_SACK_PERMIT)
3529 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3530 	}
3531 #undef PAD
3532 	optlen = min(optlen, TCP_MAXOLEN);
3533 	return (tp->t_maxseg - optlen);
3534 }
3535 
3536 
3537 u_int
3538 tcp_fixed_maxseg(const struct tcpcb *tp)
3539 {
3540 	int optlen;
3541 
3542 	if (tp->t_flags & TF_NOOPT)
3543 		return (tp->t_maxseg);
3544 
3545 	/*
3546 	 * Here we have a simplified code from tcp_addoptions(),
3547 	 * without a proper loop, and having most of paddings hardcoded.
3548 	 * We only consider fixed options that we would send every
3549 	 * time I.e. SACK is not considered. This is important
3550 	 * for cc modules to figure out what the modulo of the
3551 	 * cwnd should be.
3552 	 */
3553 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3554 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3555 		if (tp->t_flags & TF_RCVD_TSTMP)
3556 			optlen = TCPOLEN_TSTAMP_APPA;
3557 		else
3558 			optlen = 0;
3559 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3560 		if (tp->t_flags & TF_SIGNATURE)
3561 			optlen += PAD(TCPOLEN_SIGNATURE);
3562 #endif
3563 	} else {
3564 		if (tp->t_flags & TF_REQ_TSTMP)
3565 			optlen = TCPOLEN_TSTAMP_APPA;
3566 		else
3567 			optlen = PAD(TCPOLEN_MAXSEG);
3568 		if (tp->t_flags & TF_REQ_SCALE)
3569 			optlen += PAD(TCPOLEN_WINDOW);
3570 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3571 		if (tp->t_flags & TF_SIGNATURE)
3572 			optlen += PAD(TCPOLEN_SIGNATURE);
3573 #endif
3574 		if (tp->t_flags & TF_SACK_PERMIT)
3575 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3576 	}
3577 #undef PAD
3578 	optlen = min(optlen, TCP_MAXOLEN);
3579 	return (tp->t_maxseg - optlen);
3580 }
3581 
3582 
3583 
3584 static int
3585 sysctl_drop(SYSCTL_HANDLER_ARGS)
3586 {
3587 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3588 	struct sockaddr_storage addrs[2];
3589 	struct inpcb *inp;
3590 	struct tcpcb *tp;
3591 	struct tcptw *tw;
3592 	struct sockaddr_in *fin, *lin;
3593 	struct epoch_tracker et;
3594 #ifdef INET6
3595 	struct sockaddr_in6 *fin6, *lin6;
3596 #endif
3597 	int error;
3598 
3599 	inp = NULL;
3600 	fin = lin = NULL;
3601 #ifdef INET6
3602 	fin6 = lin6 = NULL;
3603 #endif
3604 	error = 0;
3605 
3606 	if (req->oldptr != NULL || req->oldlen != 0)
3607 		return (EINVAL);
3608 	if (req->newptr == NULL)
3609 		return (EPERM);
3610 	if (req->newlen < sizeof(addrs))
3611 		return (ENOMEM);
3612 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3613 	if (error)
3614 		return (error);
3615 
3616 	switch (addrs[0].ss_family) {
3617 #ifdef INET6
3618 	case AF_INET6:
3619 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3620 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3621 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3622 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3623 			return (EINVAL);
3624 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3625 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3626 				return (EINVAL);
3627 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3628 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3629 			fin = (struct sockaddr_in *)&addrs[0];
3630 			lin = (struct sockaddr_in *)&addrs[1];
3631 			break;
3632 		}
3633 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3634 		if (error)
3635 			return (error);
3636 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3637 		if (error)
3638 			return (error);
3639 		break;
3640 #endif
3641 #ifdef INET
3642 	case AF_INET:
3643 		fin = (struct sockaddr_in *)&addrs[0];
3644 		lin = (struct sockaddr_in *)&addrs[1];
3645 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3646 		    lin->sin_len != sizeof(struct sockaddr_in))
3647 			return (EINVAL);
3648 		break;
3649 #endif
3650 	default:
3651 		return (EINVAL);
3652 	}
3653 	NET_EPOCH_ENTER(et);
3654 	switch (addrs[0].ss_family) {
3655 #ifdef INET6
3656 	case AF_INET6:
3657 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3658 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3659 		    INPLOOKUP_WLOCKPCB, NULL);
3660 		break;
3661 #endif
3662 #ifdef INET
3663 	case AF_INET:
3664 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3665 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3666 		break;
3667 #endif
3668 	}
3669 	if (inp != NULL) {
3670 		if (inp->inp_flags & INP_TIMEWAIT) {
3671 			/*
3672 			 * XXXRW: There currently exists a state where an
3673 			 * inpcb is present, but its timewait state has been
3674 			 * discarded.  For now, don't allow dropping of this
3675 			 * type of inpcb.
3676 			 */
3677 			tw = intotw(inp);
3678 			if (tw != NULL)
3679 				tcp_twclose(tw, 0);
3680 			else
3681 				INP_WUNLOCK(inp);
3682 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3683 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3684 			tp = intotcpcb(inp);
3685 			tp = tcp_drop(tp, ECONNABORTED);
3686 			if (tp != NULL)
3687 				INP_WUNLOCK(inp);
3688 		} else
3689 			INP_WUNLOCK(inp);
3690 	} else
3691 		error = ESRCH;
3692 	NET_EPOCH_EXIT(et);
3693 	return (error);
3694 }
3695 
3696 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3697     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3698     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3699     "Drop TCP connection");
3700 
3701 #ifdef KERN_TLS
3702 static int
3703 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3704 {
3705 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3706 	struct sockaddr_storage addrs[2];
3707 	struct inpcb *inp;
3708 	struct sockaddr_in *fin, *lin;
3709 	struct epoch_tracker et;
3710 #ifdef INET6
3711 	struct sockaddr_in6 *fin6, *lin6;
3712 #endif
3713 	int error;
3714 
3715 	inp = NULL;
3716 	fin = lin = NULL;
3717 #ifdef INET6
3718 	fin6 = lin6 = NULL;
3719 #endif
3720 	error = 0;
3721 
3722 	if (req->oldptr != NULL || req->oldlen != 0)
3723 		return (EINVAL);
3724 	if (req->newptr == NULL)
3725 		return (EPERM);
3726 	if (req->newlen < sizeof(addrs))
3727 		return (ENOMEM);
3728 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3729 	if (error)
3730 		return (error);
3731 
3732 	switch (addrs[0].ss_family) {
3733 #ifdef INET6
3734 	case AF_INET6:
3735 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3736 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3737 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3738 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3739 			return (EINVAL);
3740 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3741 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3742 				return (EINVAL);
3743 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3744 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3745 			fin = (struct sockaddr_in *)&addrs[0];
3746 			lin = (struct sockaddr_in *)&addrs[1];
3747 			break;
3748 		}
3749 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3750 		if (error)
3751 			return (error);
3752 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3753 		if (error)
3754 			return (error);
3755 		break;
3756 #endif
3757 #ifdef INET
3758 	case AF_INET:
3759 		fin = (struct sockaddr_in *)&addrs[0];
3760 		lin = (struct sockaddr_in *)&addrs[1];
3761 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3762 		    lin->sin_len != sizeof(struct sockaddr_in))
3763 			return (EINVAL);
3764 		break;
3765 #endif
3766 	default:
3767 		return (EINVAL);
3768 	}
3769 	NET_EPOCH_ENTER(et);
3770 	switch (addrs[0].ss_family) {
3771 #ifdef INET6
3772 	case AF_INET6:
3773 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3774 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3775 		    INPLOOKUP_WLOCKPCB, NULL);
3776 		break;
3777 #endif
3778 #ifdef INET
3779 	case AF_INET:
3780 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3781 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3782 		break;
3783 #endif
3784 	}
3785 	NET_EPOCH_EXIT(et);
3786 	if (inp != NULL) {
3787 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3788 		    inp->inp_socket == NULL) {
3789 			error = ECONNRESET;
3790 			INP_WUNLOCK(inp);
3791 		} else {
3792 			struct socket *so;
3793 
3794 			so = inp->inp_socket;
3795 			soref(so);
3796 			error = ktls_set_tx_mode(so,
3797 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3798 			INP_WUNLOCK(inp);
3799 			SOCK_LOCK(so);
3800 			sorele(so);
3801 		}
3802 	} else
3803 		error = ESRCH;
3804 	return (error);
3805 }
3806 
3807 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3808     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3809     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3810     "Switch TCP connection to SW TLS");
3811 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3812     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3813     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3814     "Switch TCP connection to ifnet TLS");
3815 #endif
3816 
3817 /*
3818  * Generate a standardized TCP log line for use throughout the
3819  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3820  * allow use in the interrupt context.
3821  *
3822  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3823  * NB: The function may return NULL if memory allocation failed.
3824  *
3825  * Due to header inclusion and ordering limitations the struct ip
3826  * and ip6_hdr pointers have to be passed as void pointers.
3827  */
3828 char *
3829 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3830     const void *ip6hdr)
3831 {
3832 
3833 	/* Is logging enabled? */
3834 	if (V_tcp_log_in_vain == 0)
3835 		return (NULL);
3836 
3837 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3838 }
3839 
3840 char *
3841 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3842     const void *ip6hdr)
3843 {
3844 
3845 	/* Is logging enabled? */
3846 	if (tcp_log_debug == 0)
3847 		return (NULL);
3848 
3849 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3850 }
3851 
3852 static char *
3853 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3854     const void *ip6hdr)
3855 {
3856 	char *s, *sp;
3857 	size_t size;
3858 	struct ip *ip;
3859 #ifdef INET6
3860 	const struct ip6_hdr *ip6;
3861 
3862 	ip6 = (const struct ip6_hdr *)ip6hdr;
3863 #endif /* INET6 */
3864 	ip = (struct ip *)ip4hdr;
3865 
3866 	/*
3867 	 * The log line looks like this:
3868 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3869 	 */
3870 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3871 	    sizeof(PRINT_TH_FLAGS) + 1 +
3872 #ifdef INET6
3873 	    2 * INET6_ADDRSTRLEN;
3874 #else
3875 	    2 * INET_ADDRSTRLEN;
3876 #endif /* INET6 */
3877 
3878 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3879 	if (s == NULL)
3880 		return (NULL);
3881 
3882 	strcat(s, "TCP: [");
3883 	sp = s + strlen(s);
3884 
3885 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3886 		inet_ntoa_r(inc->inc_faddr, sp);
3887 		sp = s + strlen(s);
3888 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3889 		sp = s + strlen(s);
3890 		inet_ntoa_r(inc->inc_laddr, sp);
3891 		sp = s + strlen(s);
3892 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3893 #ifdef INET6
3894 	} else if (inc) {
3895 		ip6_sprintf(sp, &inc->inc6_faddr);
3896 		sp = s + strlen(s);
3897 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3898 		sp = s + strlen(s);
3899 		ip6_sprintf(sp, &inc->inc6_laddr);
3900 		sp = s + strlen(s);
3901 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3902 	} else if (ip6 && th) {
3903 		ip6_sprintf(sp, &ip6->ip6_src);
3904 		sp = s + strlen(s);
3905 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3906 		sp = s + strlen(s);
3907 		ip6_sprintf(sp, &ip6->ip6_dst);
3908 		sp = s + strlen(s);
3909 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3910 #endif /* INET6 */
3911 #ifdef INET
3912 	} else if (ip && th) {
3913 		inet_ntoa_r(ip->ip_src, sp);
3914 		sp = s + strlen(s);
3915 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3916 		sp = s + strlen(s);
3917 		inet_ntoa_r(ip->ip_dst, sp);
3918 		sp = s + strlen(s);
3919 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3920 #endif /* INET */
3921 	} else {
3922 		free(s, M_TCPLOG);
3923 		return (NULL);
3924 	}
3925 	sp = s + strlen(s);
3926 	if (th)
3927 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3928 	if (*(s + size - 1) != '\0')
3929 		panic("%s: string too long", __func__);
3930 	return (s);
3931 }
3932 
3933 /*
3934  * A subroutine which makes it easy to track TCP state changes with DTrace.
3935  * This function shouldn't be called for t_state initializations that don't
3936  * correspond to actual TCP state transitions.
3937  */
3938 void
3939 tcp_state_change(struct tcpcb *tp, int newstate)
3940 {
3941 #if defined(KDTRACE_HOOKS)
3942 	int pstate = tp->t_state;
3943 #endif
3944 
3945 	TCPSTATES_DEC(tp->t_state);
3946 	TCPSTATES_INC(newstate);
3947 	tp->t_state = newstate;
3948 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3949 }
3950 
3951 /*
3952  * Create an external-format (``xtcpcb'') structure using the information in
3953  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3954  * reduce the spew of irrelevant information over this interface, to isolate
3955  * user code from changes in the kernel structure, and potentially to provide
3956  * information-hiding if we decide that some of this information should be
3957  * hidden from users.
3958  */
3959 void
3960 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3961 {
3962 	struct tcpcb *tp = intotcpcb(inp);
3963 	struct tcptw *tw = intotw(inp);
3964 	sbintime_t now;
3965 
3966 	bzero(xt, sizeof(*xt));
3967 	if (inp->inp_flags & INP_TIMEWAIT) {
3968 		xt->t_state = TCPS_TIME_WAIT;
3969 		xt->xt_encaps_port = tw->t_port;
3970 	} else {
3971 		xt->t_state = tp->t_state;
3972 		xt->t_logstate = tp->t_logstate;
3973 		xt->t_flags = tp->t_flags;
3974 		xt->t_sndzerowin = tp->t_sndzerowin;
3975 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3976 		xt->t_rcvoopack = tp->t_rcvoopack;
3977 		xt->t_rcv_wnd = tp->rcv_wnd;
3978 		xt->t_snd_wnd = tp->snd_wnd;
3979 		xt->t_snd_cwnd = tp->snd_cwnd;
3980 		xt->t_snd_ssthresh = tp->snd_ssthresh;
3981 		xt->t_maxseg = tp->t_maxseg;
3982 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3983 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3984 
3985 		now = getsbinuptime();
3986 #define	COPYTIMER(ttt)	do {						\
3987 		if (callout_active(&tp->t_timers->ttt))			\
3988 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3989 			    SBT_1MS;					\
3990 		else							\
3991 			xt->ttt = 0;					\
3992 } while (0)
3993 		COPYTIMER(tt_delack);
3994 		COPYTIMER(tt_rexmt);
3995 		COPYTIMER(tt_persist);
3996 		COPYTIMER(tt_keep);
3997 		COPYTIMER(tt_2msl);
3998 #undef COPYTIMER
3999 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4000 
4001 		xt->xt_encaps_port = tp->t_port;
4002 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4003 		    TCP_FUNCTION_NAME_LEN_MAX);
4004 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4005 		    TCP_CA_NAME_MAX);
4006 #ifdef TCP_BLACKBOX
4007 		(void)tcp_log_get_id(tp, xt->xt_logid);
4008 #endif
4009 	}
4010 
4011 	xt->xt_len = sizeof(struct xtcpcb);
4012 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4013 	if (inp->inp_socket == NULL)
4014 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4015 }
4016 
4017 void
4018 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4019 {
4020 	uint32_t bit, i;
4021 
4022 	if ((tp == NULL) ||
4023 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4024 	    (status == 0)) {
4025 		/* Invalid */
4026 		return;
4027 	}
4028 	if (status > (sizeof(uint32_t) * 8)) {
4029 		/* Should this be a KASSERT? */
4030 		return;
4031 	}
4032 	bit = 1U << (status - 1);
4033 	if (bit & tp->t_end_info_status) {
4034 		/* already logged */
4035 		return;
4036 	}
4037 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4038 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4039 			tp->t_end_info_bytes[i] = status;
4040 			tp->t_end_info_status |= bit;
4041 			break;
4042 		}
4043 	}
4044 }
4045 
4046 int
4047 tcp_can_enable_pacing(void)
4048 {
4049 
4050 	if ((tcp_pacing_limit == -1) ||
4051 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4052 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4053 		shadow_num_connections = number_of_tcp_connections_pacing;
4054 		return (1);
4055 	} else {
4056 		return (0);
4057 	}
4058 }
4059 
4060 static uint8_t tcp_pacing_warning = 0;
4061 
4062 void
4063 tcp_decrement_paced_conn(void)
4064 {
4065 	uint32_t ret;
4066 
4067 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4068 	shadow_num_connections = number_of_tcp_connections_pacing;
4069 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4070 	if (ret == 0) {
4071 		if (tcp_pacing_limit != -1) {
4072 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4073 			tcp_pacing_limit = 0;
4074 		} else if (tcp_pacing_warning == 0) {
4075 			printf("Warning pacing count is invalid, invalid decrement\n");
4076 			tcp_pacing_warning = 1;
4077 		}
4078 	}
4079 }
4080