xref: /freebsd/sys/netinet/tcp_subr.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
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 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_ipsec.h"
38 #include "opt_kern_tls.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/arb.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #ifdef TCP_HHOOK
46 #include <sys/hhook.h>
47 #endif
48 #include <sys/kernel.h>
49 #ifdef TCP_HHOOK
50 #include <sys/khelp.h>
51 #endif
52 #ifdef KERN_TLS
53 #include <sys/ktls.h>
54 #endif
55 #include <sys/qmath.h>
56 #include <sys/stats.h>
57 #include <sys/sysctl.h>
58 #include <sys/jail.h>
59 #include <sys/malloc.h>
60 #include <sys/refcount.h>
61 #include <sys/mbuf.h>
62 #include <sys/priv.h>
63 #include <sys/proc.h>
64 #include <sys/sdt.h>
65 #include <sys/socket.h>
66 #include <sys/socketvar.h>
67 #include <sys/protosw.h>
68 #include <sys/random.h>
69 
70 #include <vm/uma.h>
71 
72 #include <net/route.h>
73 #include <net/route/nhop.h>
74 #include <net/if.h>
75 #include <net/if_var.h>
76 #include <net/if_private.h>
77 #include <net/vnet.h>
78 
79 #include <netinet/in.h>
80 #include <netinet/in_fib.h>
81 #include <netinet/in_kdtrace.h>
82 #include <netinet/in_pcb.h>
83 #include <netinet/in_systm.h>
84 #include <netinet/in_var.h>
85 #include <netinet/ip.h>
86 #include <netinet/ip_icmp.h>
87 #include <netinet/ip_var.h>
88 #ifdef INET6
89 #include <netinet/icmp6.h>
90 #include <netinet/ip6.h>
91 #include <netinet6/in6_fib.h>
92 #include <netinet6/in6_pcb.h>
93 #include <netinet6/ip6_var.h>
94 #include <netinet6/scope6_var.h>
95 #include <netinet6/nd6.h>
96 #endif
97 
98 #include <netinet/tcp.h>
99 #ifdef INVARIANTS
100 #define TCPSTATES
101 #endif
102 #include <netinet/tcp_fsm.h>
103 #include <netinet/tcp_seq.h>
104 #include <netinet/tcp_timer.h>
105 #include <netinet/tcp_var.h>
106 #include <netinet/tcp_ecn.h>
107 #include <netinet/tcp_log_buf.h>
108 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_hpts.h>
110 #include <netinet/tcp_lro.h>
111 #include <netinet/cc/cc.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/tcp_fastopen.h>
114 #include <netinet/tcp_accounting.h>
115 #ifdef TCPPCAP
116 #include <netinet/tcp_pcap.h>
117 #endif
118 #ifdef TCP_OFFLOAD
119 #include <netinet/tcp_offload.h>
120 #endif
121 #include <netinet/udp.h>
122 #include <netinet/udp_var.h>
123 #ifdef INET6
124 #include <netinet6/tcp6_var.h>
125 #endif
126 
127 #include <netipsec/ipsec_support.h>
128 
129 #include <machine/in_cksum.h>
130 #include <crypto/siphash/siphash.h>
131 
132 #include <security/mac/mac_framework.h>
133 
134 #ifdef INET6
135 static ip6proto_ctlinput_t tcp6_ctlinput;
136 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
137 #endif
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 TCP_SAD_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_sad_limit = 10000;
155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, limit,
156     CTLFLAG_RW,
157     &tcp_sad_limit, 10000,
158     "If SaD is enabled, what is the limit to sendmap entries (0 = unlimited)?");
159 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
161     CTLFLAG_RW,
162     &tcp_sack_to_ack_thresh, 700,
163     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
164 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
166     CTLFLAG_RW,
167     &tcp_sack_to_move_thresh, 600,
168     "Percentage of sack moves we must see above (10.1 percent is 101)");
169 int32_t tcp_restoral_thresh = 450;	/* 45 % (sack:2:ack -25%) (mv:ratio -15%) **/
170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
171     CTLFLAG_RW,
172     &tcp_restoral_thresh, 450,
173     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
174 int32_t tcp_sad_decay_val = 800;
175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
176     CTLFLAG_RW,
177     &tcp_sad_decay_val, 800,
178     "The decay percentage (10.1 percent equals 101 )");
179 int32_t tcp_map_minimum = 500;
180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
181     CTLFLAG_RW,
182     &tcp_map_minimum, 500,
183     "Number of Map enteries before we start detection");
184 int32_t tcp_sad_pacing_interval = 2000;
185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
186     CTLFLAG_RW,
187     &tcp_sad_pacing_interval, 2000,
188     "What is the minimum pacing interval for a classified attacker?");
189 
190 int32_t tcp_sad_low_pps = 100;
191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
192     CTLFLAG_RW,
193     &tcp_sad_low_pps, 100,
194     "What is the input pps that below which we do not decay?");
195 #endif
196 uint32_t tcp_ack_war_time_window = 1000;
197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
198     CTLFLAG_RW,
199     &tcp_ack_war_time_window, 1000,
200    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
201 uint32_t tcp_ack_war_cnt = 5;
202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
203     CTLFLAG_RW,
204     &tcp_ack_war_cnt, 5,
205    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
206 
207 struct rwlock tcp_function_lock;
208 
209 static int
210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
211 {
212 	int error, new;
213 
214 	new = V_tcp_mssdflt;
215 	error = sysctl_handle_int(oidp, &new, 0, req);
216 	if (error == 0 && req->newptr) {
217 		if (new < TCP_MINMSS)
218 			error = EINVAL;
219 		else
220 			V_tcp_mssdflt = new;
221 	}
222 	return (error);
223 }
224 
225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
226     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
227     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
228     "Default TCP Maximum Segment Size");
229 
230 #ifdef INET6
231 static int
232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
233 {
234 	int error, new;
235 
236 	new = V_tcp_v6mssdflt;
237 	error = sysctl_handle_int(oidp, &new, 0, req);
238 	if (error == 0 && req->newptr) {
239 		if (new < TCP_MINMSS)
240 			error = EINVAL;
241 		else
242 			V_tcp_v6mssdflt = new;
243 	}
244 	return (error);
245 }
246 
247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
248     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
249     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
250    "Default TCP Maximum Segment Size for IPv6");
251 #endif /* INET6 */
252 
253 /*
254  * Minimum MSS we accept and use. This prevents DoS attacks where
255  * we are forced to a ridiculous low MSS like 20 and send hundreds
256  * of packets instead of one. The effect scales with the available
257  * bandwidth and quickly saturates the CPU and network interface
258  * with packet generation and sending. Set to zero to disable MINMSS
259  * checking. This setting prevents us from sending too small packets.
260  */
261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
263      &VNET_NAME(tcp_minmss), 0,
264     "Minimum TCP Maximum Segment Size");
265 
266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
268     &VNET_NAME(tcp_do_rfc1323), 0,
269     "Enable rfc1323 (high performance TCP) extensions");
270 
271 /*
272  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
273  * Some stacks negotiate TS, but never send them after connection setup. Some
274  * stacks negotiate TS, but don't send them when sending keep-alive segments.
275  * These include modern widely deployed TCP stacks.
276  * Therefore tolerating violations for now...
277  */
278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
280     &VNET_NAME(tcp_tolerate_missing_ts), 0,
281     "Tolerate missing TCP timestamps");
282 
283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
285     &VNET_NAME(tcp_ts_offset_per_conn), 0,
286     "Initialize TCP timestamps per connection instead of per host pair");
287 
288 /* How many connections are pacing */
289 static volatile uint32_t number_of_tcp_connections_pacing = 0;
290 static uint32_t shadow_num_connections = 0;
291 static counter_u64_t tcp_pacing_failures;
292 
293 static int tcp_pacing_limit = 10000;
294 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
295     &tcp_pacing_limit, 1000,
296     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
297 
298 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
299     &shadow_num_connections, 0, "Number of TCP connections being paced");
300 
301 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD,
302     &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit");
303 
304 static int	tcp_log_debug = 0;
305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
306     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
307 
308 static int	tcp_tcbhashsize;
309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
310     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
311 
312 static int	do_tcpdrain = 1;
313 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
314     "Enable tcp_drain routine for extra help when low on mbufs");
315 
316 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
317     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
318 
319 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
320 #define	V_icmp_may_rst			VNET(icmp_may_rst)
321 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
322     &VNET_NAME(icmp_may_rst), 0,
323     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
324 
325 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
326 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
327 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
328     &VNET_NAME(tcp_isn_reseed_interval), 0,
329     "Seconds between reseeding of ISN secret");
330 
331 static int	tcp_soreceive_stream;
332 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
333     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
334 
335 VNET_DEFINE(uma_zone_t, sack_hole_zone);
336 #define	V_sack_hole_zone		VNET(sack_hole_zone)
337 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
338 static int
339 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
340 {
341 	int error;
342 	uint32_t new;
343 
344 	new = V_tcp_map_entries_limit;
345 	error = sysctl_handle_int(oidp, &new, 0, req);
346 	if (error == 0 && req->newptr) {
347 		/* only allow "0" and value > minimum */
348 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
349 			error = EINVAL;
350 		else
351 			V_tcp_map_entries_limit = new;
352 	}
353 	return (error);
354 }
355 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
356     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
357     &VNET_NAME(tcp_map_entries_limit), 0,
358     &sysctl_net_inet_tcp_map_limit_check, "IU",
359     "Total sendmap entries limit");
360 
361 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
362 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
363      &VNET_NAME(tcp_map_split_limit), 0,
364     "Total sendmap split entries limit");
365 
366 #ifdef TCP_HHOOK
367 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
368 #endif
369 
370 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
371 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
372 #define	V_ts_offset_secret	VNET(ts_offset_secret)
373 
374 static int	tcp_default_fb_init(struct tcpcb *tp, void **ptr);
375 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
376 static int	tcp_default_handoff_ok(struct tcpcb *tp);
377 static struct inpcb *tcp_notify(struct inpcb *, int);
378 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
379 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
380 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
381 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
382 		    const void *ip4hdr, const void *ip6hdr);
383 static void	tcp_default_switch_failed(struct tcpcb *tp);
384 static ipproto_ctlinput_t	tcp_ctlinput;
385 static udp_tun_icmp_t		tcp_ctlinput_viaudp;
386 
387 static struct tcp_function_block tcp_def_funcblk = {
388 	.tfb_tcp_block_name = "freebsd",
389 	.tfb_tcp_output = tcp_default_output,
390 	.tfb_tcp_do_segment = tcp_do_segment,
391 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
392 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
393 	.tfb_tcp_fb_init = tcp_default_fb_init,
394 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
395 	.tfb_switch_failed = tcp_default_switch_failed,
396 };
397 
398 static int tcp_fb_cnt = 0;
399 struct tcp_funchead t_functions;
400 VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk;
401 #define	V_tcp_func_set_ptr VNET(tcp_func_set_ptr)
402 
403 void
404 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
405 {
406 	TCPSTAT_INC(tcps_dsack_count);
407 	tp->t_dsack_pack++;
408 	if (tlp == 0) {
409 		if (SEQ_GT(end, start)) {
410 			tp->t_dsack_bytes += (end - start);
411 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
412 		} else {
413 			tp->t_dsack_tlp_bytes += (start - end);
414 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
415 		}
416 	} else {
417 		if (SEQ_GT(end, start)) {
418 			tp->t_dsack_bytes += (end - start);
419 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
420 		} else {
421 			tp->t_dsack_tlp_bytes += (start - end);
422 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
423 		}
424 	}
425 }
426 
427 static struct tcp_function_block *
428 find_tcp_functions_locked(struct tcp_function_set *fs)
429 {
430 	struct tcp_function *f;
431 	struct tcp_function_block *blk=NULL;
432 
433 	TAILQ_FOREACH(f, &t_functions, tf_next) {
434 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
435 			blk = f->tf_fb;
436 			break;
437 		}
438 	}
439 	return(blk);
440 }
441 
442 static struct tcp_function_block *
443 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
444 {
445 	struct tcp_function_block *rblk=NULL;
446 	struct tcp_function *f;
447 
448 	TAILQ_FOREACH(f, &t_functions, tf_next) {
449 		if (f->tf_fb == blk) {
450 			rblk = blk;
451 			if (s) {
452 				*s = f;
453 			}
454 			break;
455 		}
456 	}
457 	return (rblk);
458 }
459 
460 struct tcp_function_block *
461 find_and_ref_tcp_functions(struct tcp_function_set *fs)
462 {
463 	struct tcp_function_block *blk;
464 
465 	rw_rlock(&tcp_function_lock);
466 	blk = find_tcp_functions_locked(fs);
467 	if (blk)
468 		refcount_acquire(&blk->tfb_refcnt);
469 	rw_runlock(&tcp_function_lock);
470 	return(blk);
471 }
472 
473 struct tcp_function_block *
474 find_and_ref_tcp_fb(struct tcp_function_block *blk)
475 {
476 	struct tcp_function_block *rblk;
477 
478 	rw_rlock(&tcp_function_lock);
479 	rblk = find_tcp_fb_locked(blk, NULL);
480 	if (rblk)
481 		refcount_acquire(&rblk->tfb_refcnt);
482 	rw_runlock(&tcp_function_lock);
483 	return(rblk);
484 }
485 
486 /* Find a matching alias for the given tcp_function_block. */
487 int
488 find_tcp_function_alias(struct tcp_function_block *blk,
489     struct tcp_function_set *fs)
490 {
491 	struct tcp_function *f;
492 	int found;
493 
494 	found = 0;
495 	rw_rlock(&tcp_function_lock);
496 	TAILQ_FOREACH(f, &t_functions, tf_next) {
497 		if ((f->tf_fb == blk) &&
498 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
499 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
500 			/* Matching function block with different name. */
501 			strncpy(fs->function_set_name, f->tf_name,
502 			    TCP_FUNCTION_NAME_LEN_MAX);
503 			found = 1;
504 			break;
505 		}
506 	}
507 	/* Null terminate the string appropriately. */
508 	if (found) {
509 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
510 	} else {
511 		fs->function_set_name[0] = '\0';
512 	}
513 	rw_runlock(&tcp_function_lock);
514 	return (found);
515 }
516 
517 static struct tcp_function_block *
518 find_and_ref_tcp_default_fb(void)
519 {
520 	struct tcp_function_block *rblk;
521 
522 	rw_rlock(&tcp_function_lock);
523 	rblk = V_tcp_func_set_ptr;
524 	refcount_acquire(&rblk->tfb_refcnt);
525 	rw_runlock(&tcp_function_lock);
526 	return (rblk);
527 }
528 
529 void
530 tcp_switch_back_to_default(struct tcpcb *tp)
531 {
532 	struct tcp_function_block *tfb;
533 	void *ptr = NULL;
534 
535 	KASSERT(tp->t_fb != &tcp_def_funcblk,
536 	    ("%s: called by the built-in default stack", __func__));
537 
538 	/*
539 	 * Now, we'll find a new function block to use.
540 	 * Start by trying the current user-selected
541 	 * default, unless this stack is the user-selected
542 	 * default.
543 	 */
544 	tfb = find_and_ref_tcp_default_fb();
545 	if (tfb == tp->t_fb) {
546 		refcount_release(&tfb->tfb_refcnt);
547 		tfb = NULL;
548 	}
549 	/* Does the stack accept this connection? */
550 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 		refcount_release(&tfb->tfb_refcnt);
553 		tfb = NULL;
554 	}
555 	/* Try to use that stack. */
556 	if (tfb != NULL) {
557 		/* Initialize the new stack. If it succeeds, we are done. */
558 		if (tfb->tfb_tcp_fb_init == NULL ||
559 		    (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) {
560 			/* Release the old stack */
561 			if (tp->t_fb->tfb_tcp_fb_fini != NULL)
562 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
563 			refcount_release(&tp->t_fb->tfb_refcnt);
564 			/* Now set in all the pointers */
565 			tp->t_fb = tfb;
566 			tp->t_fb_ptr = ptr;
567 			return;
568 		}
569 		/*
570 		 * Initialization failed. Release the reference count on
571 		 * the looked up default stack.
572 		 */
573 		refcount_release(&tfb->tfb_refcnt);
574 	}
575 
576 	/*
577 	 * If that wasn't feasible, use the built-in default
578 	 * stack which is not allowed to reject anyone.
579 	 */
580 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
581 	if (tfb == NULL) {
582 		/* there always should be a default */
583 		panic("Can't refer to tcp_def_funcblk");
584 	}
585 	if (tfb->tfb_tcp_handoff_ok != NULL) {
586 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
587 			/* The default stack cannot say no */
588 			panic("Default stack rejects a new session?");
589 		}
590 	}
591 	if (tfb->tfb_tcp_fb_init != NULL &&
592 	    (*tfb->tfb_tcp_fb_init)(tp, &ptr)) {
593 		/* The default stack cannot fail */
594 		panic("Default stack initialization failed");
595 	}
596 	/* Now release the old stack */
597 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
598 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
599 	refcount_release(&tp->t_fb->tfb_refcnt);
600 	/* And set in the pointers to the new */
601 	tp->t_fb = tfb;
602 	tp->t_fb_ptr = ptr;
603 }
604 
605 static bool
606 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
607     const struct sockaddr *sa, void *ctx)
608 {
609 	struct ip *iph;
610 #ifdef INET6
611 	struct ip6_hdr *ip6;
612 #endif
613 	struct udphdr *uh;
614 	struct tcphdr *th;
615 	int thlen;
616 	uint16_t port;
617 
618 	TCPSTAT_INC(tcps_tunneled_pkts);
619 	if ((m->m_flags & M_PKTHDR) == 0) {
620 		/* Can't handle one that is not a pkt hdr */
621 		TCPSTAT_INC(tcps_tunneled_errs);
622 		goto out;
623 	}
624 	thlen = sizeof(struct tcphdr);
625 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
626 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
627 		TCPSTAT_INC(tcps_tunneled_errs);
628 		goto out;
629 	}
630 	iph = mtod(m, struct ip *);
631 	uh = (struct udphdr *)((caddr_t)iph + off);
632 	th = (struct tcphdr *)(uh + 1);
633 	thlen = th->th_off << 2;
634 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
635 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
636 		if (m == NULL) {
637 			TCPSTAT_INC(tcps_tunneled_errs);
638 			goto out;
639 		} else {
640 			iph = mtod(m, struct ip *);
641 			uh = (struct udphdr *)((caddr_t)iph + off);
642 			th = (struct tcphdr *)(uh + 1);
643 		}
644 	}
645 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
646 	bcopy(th, uh, m->m_len - off);
647 	m->m_len -= sizeof(struct udphdr);
648 	m->m_pkthdr.len -= sizeof(struct udphdr);
649 	/*
650 	 * We use the same algorithm for
651 	 * both UDP and TCP for c-sum. So
652 	 * the code in tcp_input will skip
653 	 * the checksum. So we do nothing
654 	 * with the flag (m->m_pkthdr.csum_flags).
655 	 */
656 	switch (iph->ip_v) {
657 #ifdef INET
658 	case IPVERSION:
659 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
660 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
661 		break;
662 #endif
663 #ifdef INET6
664 	case IPV6_VERSION >> 4:
665 		ip6 = mtod(m, struct ip6_hdr *);
666 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
667 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
668 		break;
669 #endif
670 	default:
671 		goto out;
672 		break;
673 	}
674 	return (true);
675 out:
676 	m_freem(m);
677 
678 	return (true);
679 }
680 
681 static int
682 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
683 {
684 	int error=ENOENT;
685 	struct tcp_function_set fs;
686 	struct tcp_function_block *blk;
687 
688 	memset(&fs, 0, sizeof(fs));
689 	rw_rlock(&tcp_function_lock);
690 	blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL);
691 	if (blk) {
692 		/* Found him */
693 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
694 		fs.pcbcnt = blk->tfb_refcnt;
695 	}
696 	rw_runlock(&tcp_function_lock);
697 	error = sysctl_handle_string(oidp, fs.function_set_name,
698 				     sizeof(fs.function_set_name), req);
699 
700 	/* Check for error or no change */
701 	if (error != 0 || req->newptr == NULL)
702 		return(error);
703 
704 	rw_wlock(&tcp_function_lock);
705 	blk = find_tcp_functions_locked(&fs);
706 	if ((blk == NULL) ||
707 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
708 		error = ENOENT;
709 		goto done;
710 	}
711 	V_tcp_func_set_ptr = blk;
712 done:
713 	rw_wunlock(&tcp_function_lock);
714 	return (error);
715 }
716 
717 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
718     CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
719     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
720     "Set/get the default TCP functions");
721 
722 static int
723 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
724 {
725 	int error, cnt, linesz;
726 	struct tcp_function *f;
727 	char *buffer, *cp;
728 	size_t bufsz, outsz;
729 	bool alias;
730 
731 	cnt = 0;
732 	rw_rlock(&tcp_function_lock);
733 	TAILQ_FOREACH(f, &t_functions, tf_next) {
734 		cnt++;
735 	}
736 	rw_runlock(&tcp_function_lock);
737 
738 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
739 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
740 
741 	error = 0;
742 	cp = buffer;
743 
744 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
745 	    "Alias", "PCB count");
746 	cp += linesz;
747 	bufsz -= linesz;
748 	outsz = linesz;
749 
750 	rw_rlock(&tcp_function_lock);
751 	TAILQ_FOREACH(f, &t_functions, tf_next) {
752 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
753 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
754 		    f->tf_fb->tfb_tcp_block_name,
755 		    (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ',
756 		    alias ? f->tf_name : "-",
757 		    f->tf_fb->tfb_refcnt);
758 		if (linesz >= bufsz) {
759 			error = EOVERFLOW;
760 			break;
761 		}
762 		cp += linesz;
763 		bufsz -= linesz;
764 		outsz += linesz;
765 	}
766 	rw_runlock(&tcp_function_lock);
767 	if (error == 0)
768 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
769 	free(buffer, M_TEMP);
770 	return (error);
771 }
772 
773 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
774     CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
775     NULL, 0, sysctl_net_inet_list_available, "A",
776     "list available TCP Function sets");
777 
778 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
779 
780 #ifdef INET
781 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
782 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
783 #endif
784 #ifdef INET6
785 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
786 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
787 #endif
788 
789 static struct sx tcpoudp_lock;
790 
791 static void
792 tcp_over_udp_stop(void)
793 {
794 
795 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
796 
797 #ifdef INET
798 	if (V_udp4_tun_socket != NULL) {
799 		soclose(V_udp4_tun_socket);
800 		V_udp4_tun_socket = NULL;
801 	}
802 #endif
803 #ifdef INET6
804 	if (V_udp6_tun_socket != NULL) {
805 		soclose(V_udp6_tun_socket);
806 		V_udp6_tun_socket = NULL;
807 	}
808 #endif
809 }
810 
811 static int
812 tcp_over_udp_start(void)
813 {
814 	uint16_t port;
815 	int ret;
816 #ifdef INET
817 	struct sockaddr_in sin;
818 #endif
819 #ifdef INET6
820 	struct sockaddr_in6 sin6;
821 #endif
822 
823 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
824 
825 	port = V_tcp_udp_tunneling_port;
826 	if (ntohs(port) == 0) {
827 		/* Must have a port set */
828 		return (EINVAL);
829 	}
830 #ifdef INET
831 	if (V_udp4_tun_socket != NULL) {
832 		/* Already running -- must stop first */
833 		return (EALREADY);
834 	}
835 #endif
836 #ifdef INET6
837 	if (V_udp6_tun_socket != NULL) {
838 		/* Already running -- must stop first */
839 		return (EALREADY);
840 	}
841 #endif
842 #ifdef INET
843 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
844 	    SOCK_DGRAM, IPPROTO_UDP,
845 	    curthread->td_ucred, curthread))) {
846 		tcp_over_udp_stop();
847 		return (ret);
848 	}
849 	/* Call the special UDP hook. */
850 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
851 	    tcp_recv_udp_tunneled_packet,
852 	    tcp_ctlinput_viaudp,
853 	    NULL))) {
854 		tcp_over_udp_stop();
855 		return (ret);
856 	}
857 	/* Ok, we have a socket, bind it to the port. */
858 	memset(&sin, 0, sizeof(struct sockaddr_in));
859 	sin.sin_len = sizeof(struct sockaddr_in);
860 	sin.sin_family = AF_INET;
861 	sin.sin_port = htons(port);
862 	if ((ret = sobind(V_udp4_tun_socket,
863 	    (struct sockaddr *)&sin, curthread))) {
864 		tcp_over_udp_stop();
865 		return (ret);
866 	}
867 #endif
868 #ifdef INET6
869 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
870 	    SOCK_DGRAM, IPPROTO_UDP,
871 	    curthread->td_ucred, curthread))) {
872 		tcp_over_udp_stop();
873 		return (ret);
874 	}
875 	/* Call the special UDP hook. */
876 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
877 	    tcp_recv_udp_tunneled_packet,
878 	    tcp6_ctlinput_viaudp,
879 	    NULL))) {
880 		tcp_over_udp_stop();
881 		return (ret);
882 	}
883 	/* Ok, we have a socket, bind it to the port. */
884 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
885 	sin6.sin6_len = sizeof(struct sockaddr_in6);
886 	sin6.sin6_family = AF_INET6;
887 	sin6.sin6_port = htons(port);
888 	if ((ret = sobind(V_udp6_tun_socket,
889 	    (struct sockaddr *)&sin6, curthread))) {
890 		tcp_over_udp_stop();
891 		return (ret);
892 	}
893 #endif
894 	return (0);
895 }
896 
897 static int
898 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
899 {
900 	int error;
901 	uint32_t old, new;
902 
903 	old = V_tcp_udp_tunneling_port;
904 	new = old;
905 	error = sysctl_handle_int(oidp, &new, 0, req);
906 	if ((error == 0) &&
907 	    (req->newptr != NULL)) {
908 		if ((new < TCP_TUNNELING_PORT_MIN) ||
909 		    (new > TCP_TUNNELING_PORT_MAX)) {
910 			error = EINVAL;
911 		} else {
912 			sx_xlock(&tcpoudp_lock);
913 			V_tcp_udp_tunneling_port = new;
914 			if (old != 0) {
915 				tcp_over_udp_stop();
916 			}
917 			if (new != 0) {
918 				error = tcp_over_udp_start();
919 				if (error != 0) {
920 					V_tcp_udp_tunneling_port = 0;
921 				}
922 			}
923 			sx_xunlock(&tcpoudp_lock);
924 		}
925 	}
926 	return (error);
927 }
928 
929 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
930     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
931     &VNET_NAME(tcp_udp_tunneling_port),
932     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
933     "Tunneling port for tcp over udp");
934 
935 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
936 
937 static int
938 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
939 {
940 	int error, new;
941 
942 	new = V_tcp_udp_tunneling_overhead;
943 	error = sysctl_handle_int(oidp, &new, 0, req);
944 	if (error == 0 && req->newptr) {
945 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
946 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
947 			error = EINVAL;
948 		else
949 			V_tcp_udp_tunneling_overhead = new;
950 	}
951 	return (error);
952 }
953 
954 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
955     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
956     &VNET_NAME(tcp_udp_tunneling_overhead),
957     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
958     "MSS reduction when using tcp over udp");
959 
960 /*
961  * Exports one (struct tcp_function_info) for each alias/name.
962  */
963 static int
964 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
965 {
966 	int cnt, error;
967 	struct tcp_function *f;
968 	struct tcp_function_info tfi;
969 
970 	/*
971 	 * We don't allow writes.
972 	 */
973 	if (req->newptr != NULL)
974 		return (EINVAL);
975 
976 	/*
977 	 * Wire the old buffer so we can directly copy the functions to
978 	 * user space without dropping the lock.
979 	 */
980 	if (req->oldptr != NULL) {
981 		error = sysctl_wire_old_buffer(req, 0);
982 		if (error)
983 			return (error);
984 	}
985 
986 	/*
987 	 * Walk the list and copy out matching entries. If INVARIANTS
988 	 * is compiled in, also walk the list to verify the length of
989 	 * the list matches what we have recorded.
990 	 */
991 	rw_rlock(&tcp_function_lock);
992 
993 	cnt = 0;
994 #ifndef INVARIANTS
995 	if (req->oldptr == NULL) {
996 		cnt = tcp_fb_cnt;
997 		goto skip_loop;
998 	}
999 #endif
1000 	TAILQ_FOREACH(f, &t_functions, tf_next) {
1001 #ifdef INVARIANTS
1002 		cnt++;
1003 #endif
1004 		if (req->oldptr != NULL) {
1005 			bzero(&tfi, sizeof(tfi));
1006 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
1007 			tfi.tfi_id = f->tf_fb->tfb_id;
1008 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
1009 			    sizeof(tfi.tfi_alias));
1010 			(void)strlcpy(tfi.tfi_name,
1011 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
1012 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
1013 			/*
1014 			 * Don't stop on error, as that is the
1015 			 * mechanism we use to accumulate length
1016 			 * information if the buffer was too short.
1017 			 */
1018 		}
1019 	}
1020 	KASSERT(cnt == tcp_fb_cnt,
1021 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1022 #ifndef INVARIANTS
1023 skip_loop:
1024 #endif
1025 	rw_runlock(&tcp_function_lock);
1026 	if (req->oldptr == NULL)
1027 		error = SYSCTL_OUT(req, NULL,
1028 		    (cnt + 1) * sizeof(struct tcp_function_info));
1029 
1030 	return (error);
1031 }
1032 
1033 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1034 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1035 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1036 	    "List TCP function block name-to-ID mappings");
1037 
1038 /*
1039  * tfb_tcp_handoff_ok() function for the default stack.
1040  * Note that we'll basically try to take all comers.
1041  */
1042 static int
1043 tcp_default_handoff_ok(struct tcpcb *tp)
1044 {
1045 
1046 	return (0);
1047 }
1048 
1049 /*
1050  * tfb_tcp_fb_init() function for the default stack.
1051  *
1052  * This handles making sure we have appropriate timers set if you are
1053  * transitioning a socket that has some amount of setup done.
1054  *
1055  * The init() fuction from the default can *never* return non-zero i.e.
1056  * it is required to always succeed since it is the stack of last resort!
1057  */
1058 static int
1059 tcp_default_fb_init(struct tcpcb *tp, void **ptr)
1060 {
1061 	struct socket *so = tptosocket(tp);
1062 	int rexmt;
1063 
1064 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1065 	/* We don't use the pointer */
1066 	*ptr = NULL;
1067 
1068 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1069 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1070 	    tp->t_state));
1071 
1072 	/* Make sure we get no interesting mbuf queuing behavior */
1073 	/* All mbuf queue/ack compress flags should be off */
1074 	tcp_lro_features_off(tp);
1075 
1076 	/* Cancel the GP measurement in progress */
1077 	tp->t_flags &= ~TF_GPUTINPROG;
1078 	/* Validate the timers are not in usec, if they are convert */
1079 	tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
1080 	if ((tp->t_state == TCPS_SYN_SENT) ||
1081 	    (tp->t_state == TCPS_SYN_RECEIVED))
1082 		rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
1083 	else
1084 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
1085 	if (tp->t_rxtshift == 0)
1086 		tp->t_rxtcur = rexmt;
1087 	else
1088 		TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX);
1089 
1090 	/*
1091 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1092 	 * know what to do for unexpected states (which includes TIME_WAIT).
1093 	 */
1094 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1095 		return (0);
1096 
1097 	/*
1098 	 * Make sure some kind of transmission timer is set if there is
1099 	 * outstanding data.
1100 	 */
1101 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1102 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1103 	    tcp_timer_active(tp, TT_PERSIST))) {
1104 		/*
1105 		 * If the session has established and it looks like it should
1106 		 * be in the persist state, set the persist timer. Otherwise,
1107 		 * set the retransmit timer.
1108 		 */
1109 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1110 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1111 		    (int32_t)sbavail(&so->so_snd))
1112 			tcp_setpersist(tp);
1113 		else
1114 			tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
1115 	}
1116 
1117 	/* All non-embryonic sessions get a keepalive timer. */
1118 	if (!tcp_timer_active(tp, TT_KEEP))
1119 		tcp_timer_activate(tp, TT_KEEP,
1120 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1121 		    TP_KEEPINIT(tp));
1122 
1123 	/*
1124 	 * Make sure critical variables are initialized
1125 	 * if transitioning while in Recovery.
1126 	 */
1127 	if IN_FASTRECOVERY(tp->t_flags) {
1128 		if (tp->sackhint.recover_fs == 0)
1129 			tp->sackhint.recover_fs = max(1,
1130 			    tp->snd_nxt - tp->snd_una);
1131 	}
1132 
1133 	return (0);
1134 }
1135 
1136 /*
1137  * tfb_tcp_fb_fini() function for the default stack.
1138  *
1139  * This changes state as necessary (or prudent) to prepare for another stack
1140  * to assume responsibility for the connection.
1141  */
1142 static void
1143 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1144 {
1145 
1146 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1147 
1148 #ifdef TCP_BLACKBOX
1149 	tcp_log_flowend(tp);
1150 #endif
1151 	tp->t_acktime = 0;
1152 	return;
1153 }
1154 
1155 /*
1156  * Target size of TCP PCB hash tables. Must be a power of two.
1157  *
1158  * Note that this can be overridden by the kernel environment
1159  * variable net.inet.tcp.tcbhashsize
1160  */
1161 #ifndef TCBHASHSIZE
1162 #define TCBHASHSIZE	0
1163 #endif
1164 
1165 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1166 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1167 
1168 static struct mtx isn_mtx;
1169 
1170 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1171 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1172 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1173 
1174 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1175 
1176 /*
1177  * Take a value and get the next power of 2 that doesn't overflow.
1178  * Used to size the tcp_inpcb hash buckets.
1179  */
1180 static int
1181 maketcp_hashsize(int size)
1182 {
1183 	int hashsize;
1184 
1185 	/*
1186 	 * auto tune.
1187 	 * get the next power of 2 higher than maxsockets.
1188 	 */
1189 	hashsize = 1 << fls(size);
1190 	/* catch overflow, and just go one power of 2 smaller */
1191 	if (hashsize < size) {
1192 		hashsize = 1 << (fls(size) - 1);
1193 	}
1194 	return (hashsize);
1195 }
1196 
1197 static volatile int next_tcp_stack_id = 1;
1198 
1199 /*
1200  * Register a TCP function block with the name provided in the names
1201  * array.  (Note that this function does NOT automatically register
1202  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1203  * explicitly include blk->tfb_tcp_block_name in the list of names if
1204  * you wish to register the stack with that name.)
1205  *
1206  * Either all name registrations will succeed or all will fail.  If
1207  * a name registration fails, the function will update the num_names
1208  * argument to point to the array index of the name that encountered
1209  * the failure.
1210  *
1211  * Returns 0 on success, or an error code on failure.
1212  */
1213 int
1214 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1215     const char *names[], int *num_names)
1216 {
1217 	struct tcp_function *n;
1218 	struct tcp_function_set fs;
1219 	int error, i;
1220 
1221 	KASSERT(names != NULL && *num_names > 0,
1222 	    ("%s: Called with 0-length name list", __func__));
1223 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1224 	KASSERT(rw_initialized(&tcp_function_lock),
1225 	    ("%s: called too early", __func__));
1226 
1227 	if ((blk->tfb_tcp_output == NULL) ||
1228 	    (blk->tfb_tcp_do_segment == NULL) ||
1229 	    (blk->tfb_tcp_ctloutput == NULL) ||
1230 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1231 		/*
1232 		 * These functions are required and you
1233 		 * need a name.
1234 		 */
1235 		*num_names = 0;
1236 		return (EINVAL);
1237 	}
1238 
1239 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1240 		*num_names = 0;
1241 		return (EINVAL);
1242 	}
1243 
1244 	refcount_init(&blk->tfb_refcnt, 0);
1245 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1246 	for (i = 0; i < *num_names; i++) {
1247 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1248 		if (n == NULL) {
1249 			error = ENOMEM;
1250 			goto cleanup;
1251 		}
1252 		n->tf_fb = blk;
1253 
1254 		(void)strlcpy(fs.function_set_name, names[i],
1255 		    sizeof(fs.function_set_name));
1256 		rw_wlock(&tcp_function_lock);
1257 		if (find_tcp_functions_locked(&fs) != NULL) {
1258 			/* Duplicate name space not allowed */
1259 			rw_wunlock(&tcp_function_lock);
1260 			free(n, M_TCPFUNCTIONS);
1261 			error = EALREADY;
1262 			goto cleanup;
1263 		}
1264 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1265 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1266 		tcp_fb_cnt++;
1267 		rw_wunlock(&tcp_function_lock);
1268 	}
1269 	return(0);
1270 
1271 cleanup:
1272 	/*
1273 	 * Deregister the names we just added. Because registration failed
1274 	 * for names[i], we don't need to deregister that name.
1275 	 */
1276 	*num_names = i;
1277 	rw_wlock(&tcp_function_lock);
1278 	while (--i >= 0) {
1279 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1280 			if (!strncmp(n->tf_name, names[i],
1281 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1282 				TAILQ_REMOVE(&t_functions, n, tf_next);
1283 				tcp_fb_cnt--;
1284 				n->tf_fb = NULL;
1285 				free(n, M_TCPFUNCTIONS);
1286 				break;
1287 			}
1288 		}
1289 	}
1290 	rw_wunlock(&tcp_function_lock);
1291 	return (error);
1292 }
1293 
1294 /*
1295  * Register a TCP function block using the name provided in the name
1296  * argument.
1297  *
1298  * Returns 0 on success, or an error code on failure.
1299  */
1300 int
1301 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1302     int wait)
1303 {
1304 	const char *name_list[1];
1305 	int num_names, rv;
1306 
1307 	num_names = 1;
1308 	if (name != NULL)
1309 		name_list[0] = name;
1310 	else
1311 		name_list[0] = blk->tfb_tcp_block_name;
1312 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1313 	return (rv);
1314 }
1315 
1316 /*
1317  * Register a TCP function block using the name defined in
1318  * blk->tfb_tcp_block_name.
1319  *
1320  * Returns 0 on success, or an error code on failure.
1321  */
1322 int
1323 register_tcp_functions(struct tcp_function_block *blk, int wait)
1324 {
1325 
1326 	return (register_tcp_functions_as_name(blk, NULL, wait));
1327 }
1328 
1329 /*
1330  * Deregister all names associated with a function block. This
1331  * functionally removes the function block from use within the system.
1332  *
1333  * When called with a true quiesce argument, mark the function block
1334  * as being removed so no more stacks will use it and determine
1335  * whether the removal would succeed.
1336  *
1337  * When called with a false quiesce argument, actually attempt the
1338  * removal.
1339  *
1340  * When called with a force argument, attempt to switch all TCBs to
1341  * use the default stack instead of returning EBUSY.
1342  *
1343  * Returns 0 on success (or if the removal would succeed), or an error
1344  * code on failure.
1345  */
1346 int
1347 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1348     bool force)
1349 {
1350 	struct tcp_function *f;
1351 	VNET_ITERATOR_DECL(vnet_iter);
1352 
1353 	if (blk == &tcp_def_funcblk) {
1354 		/* You can't un-register the default */
1355 		return (EPERM);
1356 	}
1357 	rw_wlock(&tcp_function_lock);
1358 	VNET_LIST_RLOCK_NOSLEEP();
1359 	VNET_FOREACH(vnet_iter) {
1360 		CURVNET_SET(vnet_iter);
1361 		if (blk == V_tcp_func_set_ptr) {
1362 			/* You can't free the current default in some vnet. */
1363 			CURVNET_RESTORE();
1364 			VNET_LIST_RUNLOCK_NOSLEEP();
1365 			rw_wunlock(&tcp_function_lock);
1366 			return (EBUSY);
1367 		}
1368 		CURVNET_RESTORE();
1369 	}
1370 	VNET_LIST_RUNLOCK_NOSLEEP();
1371 	/* Mark the block so no more stacks can use it. */
1372 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1373 	/*
1374 	 * If TCBs are still attached to the stack, attempt to switch them
1375 	 * to the default stack.
1376 	 */
1377 	if (force && blk->tfb_refcnt) {
1378 		struct inpcb *inp;
1379 		struct tcpcb *tp;
1380 		VNET_ITERATOR_DECL(vnet_iter);
1381 
1382 		rw_wunlock(&tcp_function_lock);
1383 
1384 		VNET_LIST_RLOCK();
1385 		VNET_FOREACH(vnet_iter) {
1386 			CURVNET_SET(vnet_iter);
1387 			struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1388 			    INPLOOKUP_WLOCKPCB);
1389 
1390 			while ((inp = inp_next(&inpi)) != NULL) {
1391 				tp = intotcpcb(inp);
1392 				if (tp == NULL || tp->t_fb != blk)
1393 					continue;
1394 				tcp_switch_back_to_default(tp);
1395 			}
1396 			CURVNET_RESTORE();
1397 		}
1398 		VNET_LIST_RUNLOCK();
1399 
1400 		rw_wlock(&tcp_function_lock);
1401 	}
1402 	if (blk->tfb_refcnt) {
1403 		/* TCBs still attached. */
1404 		rw_wunlock(&tcp_function_lock);
1405 		return (EBUSY);
1406 	}
1407 	if (quiesce) {
1408 		/* Skip removal. */
1409 		rw_wunlock(&tcp_function_lock);
1410 		return (0);
1411 	}
1412 	/* Remove any function names that map to this function block. */
1413 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1414 		TAILQ_REMOVE(&t_functions, f, tf_next);
1415 		tcp_fb_cnt--;
1416 		f->tf_fb = NULL;
1417 		free(f, M_TCPFUNCTIONS);
1418 	}
1419 	rw_wunlock(&tcp_function_lock);
1420 	return (0);
1421 }
1422 
1423 static void
1424 tcp_drain(void)
1425 {
1426 	struct epoch_tracker et;
1427 	VNET_ITERATOR_DECL(vnet_iter);
1428 
1429 	if (!do_tcpdrain)
1430 		return;
1431 
1432 	NET_EPOCH_ENTER(et);
1433 	VNET_LIST_RLOCK_NOSLEEP();
1434 	VNET_FOREACH(vnet_iter) {
1435 		CURVNET_SET(vnet_iter);
1436 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1437 		    INPLOOKUP_WLOCKPCB);
1438 		struct inpcb *inpb;
1439 		struct tcpcb *tcpb;
1440 
1441 	/*
1442 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1443 	 * if there is one...
1444 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1445 	 *      reassembly queue should be flushed, but in a situation
1446 	 *	where we're really low on mbufs, this is potentially
1447 	 *	useful.
1448 	 */
1449 		while ((inpb = inp_next(&inpi)) != NULL) {
1450 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1451 				tcp_reass_flush(tcpb);
1452 				tcp_clean_sackreport(tcpb);
1453 #ifdef TCP_BLACKBOX
1454 				tcp_log_drain(tcpb);
1455 #endif
1456 #ifdef TCPPCAP
1457 				if (tcp_pcap_aggressive_free) {
1458 					/* Free the TCP PCAP queues. */
1459 					tcp_pcap_drain(&(tcpb->t_inpkts));
1460 					tcp_pcap_drain(&(tcpb->t_outpkts));
1461 				}
1462 #endif
1463 			}
1464 		}
1465 		CURVNET_RESTORE();
1466 	}
1467 	VNET_LIST_RUNLOCK_NOSLEEP();
1468 	NET_EPOCH_EXIT(et);
1469 }
1470 
1471 static void
1472 tcp_vnet_init(void *arg __unused)
1473 {
1474 
1475 #ifdef TCP_HHOOK
1476 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1477 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1478 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1479 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1480 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1481 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1482 #endif
1483 #ifdef STATS
1484 	if (tcp_stats_init())
1485 		printf("%s: WARNING: unable to initialise TCP stats\n",
1486 		    __func__);
1487 #endif
1488 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1489 	    tcp_tcbhashsize);
1490 
1491 	syncache_init();
1492 	tcp_hc_init();
1493 
1494 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1495 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1496 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1497 
1498 	tcp_fastopen_init();
1499 
1500 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1501 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1502 
1503 	V_tcp_msl = TCPTV_MSL;
1504 }
1505 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1506     tcp_vnet_init, NULL);
1507 
1508 static void
1509 tcp_init(void *arg __unused)
1510 {
1511 	const char *tcbhash_tuneable;
1512 	int hashsize;
1513 
1514 	tcp_reass_global_init();
1515 
1516 	/* XXX virtualize those below? */
1517 	tcp_delacktime = TCPTV_DELACK;
1518 	tcp_keepinit = TCPTV_KEEP_INIT;
1519 	tcp_keepidle = TCPTV_KEEP_IDLE;
1520 	tcp_keepintvl = TCPTV_KEEPINTVL;
1521 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1522 	tcp_rexmit_initial = TCPTV_RTOBASE;
1523 	if (tcp_rexmit_initial < 1)
1524 		tcp_rexmit_initial = 1;
1525 	tcp_rexmit_min = TCPTV_MIN;
1526 	if (tcp_rexmit_min < 1)
1527 		tcp_rexmit_min = 1;
1528 	tcp_persmin = TCPTV_PERSMIN;
1529 	tcp_persmax = TCPTV_PERSMAX;
1530 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1531 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1532 
1533 	/* Setup the tcp function block list */
1534 	TAILQ_INIT(&t_functions);
1535 	rw_init(&tcp_function_lock, "tcp_func_lock");
1536 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1537 	sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1538 #ifdef TCP_BLACKBOX
1539 	/* Initialize the TCP logging data. */
1540 	tcp_log_init();
1541 #endif
1542 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1543 
1544 	if (tcp_soreceive_stream) {
1545 #ifdef INET
1546 		tcp_protosw.pr_soreceive = soreceive_stream;
1547 #endif
1548 #ifdef INET6
1549 		tcp6_protosw.pr_soreceive = soreceive_stream;
1550 #endif /* INET6 */
1551 	}
1552 
1553 #ifdef INET6
1554 	max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1555 #else /* INET6 */
1556 	max_protohdr_grow(sizeof(struct tcpiphdr));
1557 #endif /* INET6 */
1558 
1559 	ISN_LOCK_INIT();
1560 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1561 		SHUTDOWN_PRI_DEFAULT);
1562 	EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1563 	EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1564 
1565 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1566 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1567 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1568 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1569 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1570 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1571 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1572 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1573 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1574 	tcp_pacing_failures = counter_u64_alloc(M_WAITOK);
1575 #ifdef TCPPCAP
1576 	tcp_pcap_init();
1577 #endif
1578 
1579 	hashsize = TCBHASHSIZE;
1580 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1581 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1582 	if (hashsize == 0) {
1583 		/*
1584 		 * Auto tune the hash size based on maxsockets.
1585 		 * A perfect hash would have a 1:1 mapping
1586 		 * (hashsize = maxsockets) however it's been
1587 		 * suggested that O(2) average is better.
1588 		 */
1589 		hashsize = maketcp_hashsize(maxsockets / 4);
1590 		/*
1591 		 * Our historical default is 512,
1592 		 * do not autotune lower than this.
1593 		 */
1594 		if (hashsize < 512)
1595 			hashsize = 512;
1596 		if (bootverbose)
1597 			printf("%s: %s auto tuned to %d\n", __func__,
1598 			    tcbhash_tuneable, hashsize);
1599 	}
1600 	/*
1601 	 * We require a hashsize to be a power of two.
1602 	 * Previously if it was not a power of two we would just reset it
1603 	 * back to 512, which could be a nasty surprise if you did not notice
1604 	 * the error message.
1605 	 * Instead what we do is clip it to the closest power of two lower
1606 	 * than the specified hash value.
1607 	 */
1608 	if (!powerof2(hashsize)) {
1609 		int oldhashsize = hashsize;
1610 
1611 		hashsize = maketcp_hashsize(hashsize);
1612 		/* prevent absurdly low value */
1613 		if (hashsize < 16)
1614 			hashsize = 16;
1615 		printf("%s: WARNING: TCB hash size not a power of 2, "
1616 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1617 		    hashsize);
1618 	}
1619 	tcp_tcbhashsize = hashsize;
1620 
1621 #ifdef INET
1622 	IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1623 #endif
1624 #ifdef INET6
1625 	IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1626 #endif
1627 }
1628 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1629 
1630 #ifdef VIMAGE
1631 static void
1632 tcp_destroy(void *unused __unused)
1633 {
1634 	int n;
1635 #ifdef TCP_HHOOK
1636 	int error;
1637 #endif
1638 
1639 	/*
1640 	 * All our processes are gone, all our sockets should be cleaned
1641 	 * up, which means, we should be past the tcp_discardcb() calls.
1642 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1643 	 */
1644 	for (;;) {
1645 		INP_INFO_WLOCK(&V_tcbinfo);
1646 		n = V_tcbinfo.ipi_count;
1647 		INP_INFO_WUNLOCK(&V_tcbinfo);
1648 		if (n == 0)
1649 			break;
1650 		pause("tcpdes", hz / 10);
1651 	}
1652 	tcp_hc_destroy();
1653 	syncache_destroy();
1654 	in_pcbinfo_destroy(&V_tcbinfo);
1655 	/* tcp_discardcb() clears the sack_holes up. */
1656 	uma_zdestroy(V_sack_hole_zone);
1657 
1658 	/*
1659 	 * Cannot free the zone until all tcpcbs are released as we attach
1660 	 * the allocations to them.
1661 	 */
1662 	tcp_fastopen_destroy();
1663 
1664 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1665 	VNET_PCPUSTAT_FREE(tcpstat);
1666 
1667 #ifdef TCP_HHOOK
1668 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1669 	if (error != 0) {
1670 		printf("%s: WARNING: unable to deregister helper hook "
1671 		    "type=%d, id=%d: error %d returned\n", __func__,
1672 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1673 	}
1674 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1675 	if (error != 0) {
1676 		printf("%s: WARNING: unable to deregister helper hook "
1677 		    "type=%d, id=%d: error %d returned\n", __func__,
1678 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1679 	}
1680 #endif
1681 }
1682 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1683 #endif
1684 
1685 void
1686 tcp_fini(void *xtp)
1687 {
1688 
1689 }
1690 
1691 /*
1692  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1693  * tcp_template used to store this data in mbufs, but we now recopy it out
1694  * of the tcpcb each time to conserve mbufs.
1695  */
1696 void
1697 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1698 {
1699 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1700 
1701 	INP_WLOCK_ASSERT(inp);
1702 
1703 #ifdef INET6
1704 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1705 		struct ip6_hdr *ip6;
1706 
1707 		ip6 = (struct ip6_hdr *)ip_ptr;
1708 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1709 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1710 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1711 			(IPV6_VERSION & IPV6_VERSION_MASK);
1712 		if (port == 0)
1713 			ip6->ip6_nxt = IPPROTO_TCP;
1714 		else
1715 			ip6->ip6_nxt = IPPROTO_UDP;
1716 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1717 		ip6->ip6_src = inp->in6p_laddr;
1718 		ip6->ip6_dst = inp->in6p_faddr;
1719 	}
1720 #endif /* INET6 */
1721 #if defined(INET6) && defined(INET)
1722 	else
1723 #endif
1724 #ifdef INET
1725 	{
1726 		struct ip *ip;
1727 
1728 		ip = (struct ip *)ip_ptr;
1729 		ip->ip_v = IPVERSION;
1730 		ip->ip_hl = 5;
1731 		ip->ip_tos = inp->inp_ip_tos;
1732 		ip->ip_len = 0;
1733 		ip->ip_id = 0;
1734 		ip->ip_off = 0;
1735 		ip->ip_ttl = inp->inp_ip_ttl;
1736 		ip->ip_sum = 0;
1737 		if (port == 0)
1738 			ip->ip_p = IPPROTO_TCP;
1739 		else
1740 			ip->ip_p = IPPROTO_UDP;
1741 		ip->ip_src = inp->inp_laddr;
1742 		ip->ip_dst = inp->inp_faddr;
1743 	}
1744 #endif /* INET */
1745 	th->th_sport = inp->inp_lport;
1746 	th->th_dport = inp->inp_fport;
1747 	th->th_seq = 0;
1748 	th->th_ack = 0;
1749 	th->th_off = 5;
1750 	tcp_set_flags(th, 0);
1751 	th->th_win = 0;
1752 	th->th_urp = 0;
1753 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1754 }
1755 
1756 /*
1757  * Create template to be used to send tcp packets on a connection.
1758  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1759  * use for this function is in keepalives, which use tcp_respond.
1760  */
1761 struct tcptemp *
1762 tcpip_maketemplate(struct inpcb *inp)
1763 {
1764 	struct tcptemp *t;
1765 
1766 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1767 	if (t == NULL)
1768 		return (NULL);
1769 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1770 	return (t);
1771 }
1772 
1773 /*
1774  * Send a single message to the TCP at address specified by
1775  * the given TCP/IP header.  If m == NULL, then we make a copy
1776  * of the tcpiphdr at th and send directly to the addressed host.
1777  * This is used to force keep alive messages out using the TCP
1778  * template for a connection.  If flags are given then we send
1779  * a message back to the TCP which originated the segment th,
1780  * and discard the mbuf containing it and any other attached mbufs.
1781  *
1782  * In any case the ack and sequence number of the transmitted
1783  * segment are as specified by the parameters.
1784  *
1785  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1786  */
1787 void
1788 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1789     tcp_seq ack, tcp_seq seq, uint16_t flags)
1790 {
1791 	struct tcpopt to;
1792 	struct inpcb *inp;
1793 	struct ip *ip;
1794 	struct mbuf *optm;
1795 	struct udphdr *uh = NULL;
1796 	struct tcphdr *nth;
1797 	struct tcp_log_buffer *lgb;
1798 	u_char *optp;
1799 #ifdef INET6
1800 	struct ip6_hdr *ip6;
1801 	int isipv6;
1802 #endif /* INET6 */
1803 	int optlen, tlen, win, ulen;
1804 	int ect = 0;
1805 	bool incl_opts;
1806 	uint16_t port;
1807 	int output_ret;
1808 #ifdef INVARIANTS
1809 	int thflags = tcp_get_flags(th);
1810 #endif
1811 
1812 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1813 	NET_EPOCH_ASSERT();
1814 
1815 #ifdef INET6
1816 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1817 	ip6 = ipgen;
1818 #endif /* INET6 */
1819 	ip = ipgen;
1820 
1821 	if (tp != NULL) {
1822 		inp = tptoinpcb(tp);
1823 		INP_LOCK_ASSERT(inp);
1824 	} else
1825 		inp = NULL;
1826 
1827 	if (m != NULL) {
1828 #ifdef INET6
1829 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1830 			port = m->m_pkthdr.tcp_tun_port;
1831 		else
1832 #endif
1833 		if (ip && (ip->ip_p == IPPROTO_UDP))
1834 			port = m->m_pkthdr.tcp_tun_port;
1835 		else
1836 			port = 0;
1837 	} else
1838 		port = tp->t_port;
1839 
1840 	incl_opts = false;
1841 	win = 0;
1842 	if (tp != NULL) {
1843 		if (!(flags & TH_RST)) {
1844 			win = sbspace(&inp->inp_socket->so_rcv);
1845 			if (win > TCP_MAXWIN << tp->rcv_scale)
1846 				win = TCP_MAXWIN << tp->rcv_scale;
1847 		}
1848 		if ((tp->t_flags & TF_NOOPT) == 0)
1849 			incl_opts = true;
1850 	}
1851 	if (m == NULL) {
1852 		m = m_gethdr(M_NOWAIT, MT_DATA);
1853 		if (m == NULL)
1854 			return;
1855 		m->m_data += max_linkhdr;
1856 #ifdef INET6
1857 		if (isipv6) {
1858 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1859 			      sizeof(struct ip6_hdr));
1860 			ip6 = mtod(m, struct ip6_hdr *);
1861 			nth = (struct tcphdr *)(ip6 + 1);
1862 			if (port) {
1863 				/* Insert a UDP header */
1864 				uh = (struct udphdr *)nth;
1865 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1866 				uh->uh_dport = port;
1867 				nth = (struct tcphdr *)(uh + 1);
1868 			}
1869 		} else
1870 #endif /* INET6 */
1871 		{
1872 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1873 			ip = mtod(m, struct ip *);
1874 			nth = (struct tcphdr *)(ip + 1);
1875 			if (port) {
1876 				/* Insert a UDP header */
1877 				uh = (struct udphdr *)nth;
1878 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1879 				uh->uh_dport = port;
1880 				nth = (struct tcphdr *)(uh + 1);
1881 			}
1882 		}
1883 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1884 		flags = TH_ACK;
1885 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1886 		struct mbuf *n;
1887 
1888 		/* Can't reuse 'm', allocate a new mbuf. */
1889 		n = m_gethdr(M_NOWAIT, MT_DATA);
1890 		if (n == NULL) {
1891 			m_freem(m);
1892 			return;
1893 		}
1894 
1895 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1896 			m_freem(m);
1897 			m_freem(n);
1898 			return;
1899 		}
1900 
1901 		n->m_data += max_linkhdr;
1902 		/* m_len is set later */
1903 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1904 #ifdef INET6
1905 		if (isipv6) {
1906 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1907 			      sizeof(struct ip6_hdr));
1908 			ip6 = mtod(n, struct ip6_hdr *);
1909 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1910 			nth = (struct tcphdr *)(ip6 + 1);
1911 			if (port) {
1912 				/* Insert a UDP header */
1913 				uh = (struct udphdr *)nth;
1914 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1915 				uh->uh_dport = port;
1916 				nth = (struct tcphdr *)(uh + 1);
1917 			}
1918 		} else
1919 #endif /* INET6 */
1920 		{
1921 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1922 			ip = mtod(n, struct ip *);
1923 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1924 			nth = (struct tcphdr *)(ip + 1);
1925 			if (port) {
1926 				/* Insert a UDP header */
1927 				uh = (struct udphdr *)nth;
1928 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1929 				uh->uh_dport = port;
1930 				nth = (struct tcphdr *)(uh + 1);
1931 			}
1932 		}
1933 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1934 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1935 		th = nth;
1936 		m_freem(m);
1937 		m = n;
1938 	} else {
1939 		/*
1940 		 *  reuse the mbuf.
1941 		 * XXX MRT We inherit the FIB, which is lucky.
1942 		 */
1943 		m_freem(m->m_next);
1944 		m->m_next = NULL;
1945 		m->m_data = (caddr_t)ipgen;
1946 		/* clear any receive flags for proper bpf timestamping */
1947 		m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO);
1948 		/* m_len is set later */
1949 #ifdef INET6
1950 		if (isipv6) {
1951 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1952 			nth = (struct tcphdr *)(ip6 + 1);
1953 		} else
1954 #endif /* INET6 */
1955 		{
1956 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1957 			nth = (struct tcphdr *)(ip + 1);
1958 		}
1959 		if (th != nth) {
1960 			/*
1961 			 * this is usually a case when an extension header
1962 			 * exists between the IPv6 header and the
1963 			 * TCP header.
1964 			 */
1965 			nth->th_sport = th->th_sport;
1966 			nth->th_dport = th->th_dport;
1967 		}
1968 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1969 #undef xchg
1970 	}
1971 	tlen = 0;
1972 #ifdef INET6
1973 	if (isipv6)
1974 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1975 #endif
1976 #if defined(INET) && defined(INET6)
1977 	else
1978 #endif
1979 #ifdef INET
1980 		tlen = sizeof (struct tcpiphdr);
1981 #endif
1982 	if (port)
1983 		tlen += sizeof (struct udphdr);
1984 #ifdef INVARIANTS
1985 	m->m_len = 0;
1986 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1987 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1988 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1989 #endif
1990 	m->m_len = tlen;
1991 	to.to_flags = 0;
1992 	if (incl_opts) {
1993 		ect = tcp_ecn_output_established(tp, &flags, 0, false);
1994 		/* Make sure we have room. */
1995 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1996 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1997 			if (m->m_next) {
1998 				optp = mtod(m->m_next, u_char *);
1999 				optm = m->m_next;
2000 			} else
2001 				incl_opts = false;
2002 		} else {
2003 			optp = (u_char *) (nth + 1);
2004 			optm = m;
2005 		}
2006 	}
2007 	if (incl_opts) {
2008 		/* Timestamps. */
2009 		if (tp->t_flags & TF_RCVD_TSTMP) {
2010 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
2011 			to.to_tsecr = tp->ts_recent;
2012 			to.to_flags |= TOF_TS;
2013 		}
2014 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2015 		/* TCP-MD5 (RFC2385). */
2016 		if (tp->t_flags & TF_SIGNATURE)
2017 			to.to_flags |= TOF_SIGNATURE;
2018 #endif
2019 		/* Add the options. */
2020 		tlen += optlen = tcp_addoptions(&to, optp);
2021 
2022 		/* Update m_len in the correct mbuf. */
2023 		optm->m_len += optlen;
2024 	} else
2025 		optlen = 0;
2026 #ifdef INET6
2027 	if (isipv6) {
2028 		if (uh) {
2029 			ulen = tlen - sizeof(struct ip6_hdr);
2030 			uh->uh_ulen = htons(ulen);
2031 		}
2032 		ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN);
2033 		ip6->ip6_vfc = IPV6_VERSION;
2034 		if (port)
2035 			ip6->ip6_nxt = IPPROTO_UDP;
2036 		else
2037 			ip6->ip6_nxt = IPPROTO_TCP;
2038 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
2039 	}
2040 #endif
2041 #if defined(INET) && defined(INET6)
2042 	else
2043 #endif
2044 #ifdef INET
2045 	{
2046 		if (uh) {
2047 			ulen = tlen - sizeof(struct ip);
2048 			uh->uh_ulen = htons(ulen);
2049 		}
2050 		ip->ip_len = htons(tlen);
2051 		if (inp != NULL) {
2052 			ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK;
2053 			ip->ip_ttl = inp->inp_ip_ttl;
2054 		} else {
2055 			ip->ip_tos = 0;
2056 			ip->ip_ttl = V_ip_defttl;
2057 		}
2058 		ip->ip_tos |= ect;
2059 		if (port) {
2060 			ip->ip_p = IPPROTO_UDP;
2061 		} else {
2062 			ip->ip_p = IPPROTO_TCP;
2063 		}
2064 		if (V_path_mtu_discovery)
2065 			ip->ip_off |= htons(IP_DF);
2066 	}
2067 #endif
2068 	m->m_pkthdr.len = tlen;
2069 	m->m_pkthdr.rcvif = NULL;
2070 #ifdef MAC
2071 	if (inp != NULL) {
2072 		/*
2073 		 * Packet is associated with a socket, so allow the
2074 		 * label of the response to reflect the socket label.
2075 		 */
2076 		INP_LOCK_ASSERT(inp);
2077 		mac_inpcb_create_mbuf(inp, m);
2078 	} else {
2079 		/*
2080 		 * Packet is not associated with a socket, so possibly
2081 		 * update the label in place.
2082 		 */
2083 		mac_netinet_tcp_reply(m);
2084 	}
2085 #endif
2086 	nth->th_seq = htonl(seq);
2087 	nth->th_ack = htonl(ack);
2088 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2089 	tcp_set_flags(nth, flags);
2090 	if (tp && (flags & TH_RST)) {
2091 		/* Log the reset */
2092 		tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2093 	}
2094 	if (tp != NULL)
2095 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2096 	else
2097 		nth->th_win = htons((u_short)win);
2098 	nth->th_urp = 0;
2099 
2100 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2101 	if (to.to_flags & TOF_SIGNATURE) {
2102 		if (!TCPMD5_ENABLED() ||
2103 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2104 			m_freem(m);
2105 			return;
2106 		}
2107 	}
2108 #endif
2109 
2110 #ifdef INET6
2111 	if (isipv6) {
2112 		if (port) {
2113 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2114 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2115 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2116 			nth->th_sum = 0;
2117 		} else {
2118 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2119 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2120 			nth->th_sum = in6_cksum_pseudo(ip6,
2121 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2122 		}
2123 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2124 	}
2125 #endif /* INET6 */
2126 #if defined(INET6) && defined(INET)
2127 	else
2128 #endif
2129 #ifdef INET
2130 	{
2131 		if (port) {
2132 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2133 			    htons(ulen + IPPROTO_UDP));
2134 			m->m_pkthdr.csum_flags = CSUM_UDP;
2135 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2136 			nth->th_sum = 0;
2137 		} else {
2138 			m->m_pkthdr.csum_flags = CSUM_TCP;
2139 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2140 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2141 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2142 		}
2143 	}
2144 #endif /* INET */
2145 	TCP_PROBE3(debug__output, tp, th, m);
2146 	if (flags & TH_RST)
2147 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2148 	lgb = NULL;
2149 	if ((tp != NULL) && tcp_bblogging_on(tp)) {
2150 		if (INP_WLOCKED(inp)) {
2151 			union tcp_log_stackspecific log;
2152 			struct timeval tv;
2153 
2154 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2155 			log.u_bbr.inhpts = tcp_in_hpts(tp);
2156 			log.u_bbr.flex8 = 4;
2157 			log.u_bbr.pkts_out = tp->t_maxseg;
2158 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2159 			log.u_bbr.delivered = 0;
2160 			lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT,
2161 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2162 		} else {
2163 			/*
2164 			 * We can not log the packet, since we only own the
2165 			 * read lock, but a write lock is needed. The read lock
2166 			 * is not upgraded to a write lock, since only getting
2167 			 * the read lock was done intentionally to improve the
2168 			 * handling of SYN flooding attacks.
2169 			 * This happens only for pure SYN segments received in
2170 			 * the initial CLOSED state, or received in a more
2171 			 * advanced state than listen and the UDP encapsulation
2172 			 * port is unexpected.
2173 			 * The incoming SYN segments do not really belong to
2174 			 * the TCP connection and the handling does not change
2175 			 * the state of the TCP connection. Therefore, the
2176 			 * sending of the RST segments is not logged. Please
2177 			 * note that also the incoming SYN segments are not
2178 			 * logged.
2179 			 *
2180 			 * The following code ensures that the above description
2181 			 * is and stays correct.
2182 			 */
2183 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2184 			    (tp->t_state == TCPS_CLOSED ||
2185 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2186 			    ("%s: Logging of TCP segment with flags 0x%b and "
2187 			    "UDP encapsulation port %u skipped in state %s",
2188 			    __func__, thflags, PRINT_TH_FLAGS,
2189 			    ntohs(port), tcpstates[tp->t_state]));
2190 		}
2191 	}
2192 
2193 	if (flags & TH_ACK)
2194 		TCPSTAT_INC(tcps_sndacks);
2195 	else if (flags & (TH_SYN|TH_FIN|TH_RST))
2196 		TCPSTAT_INC(tcps_sndctrl);
2197 	TCPSTAT_INC(tcps_sndtotal);
2198 
2199 #ifdef INET6
2200 	if (isipv6) {
2201 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2202 		output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL,
2203 		    NULL, 0, NULL, NULL, inp);
2204 	}
2205 #endif /* INET6 */
2206 #if defined(INET) && defined(INET6)
2207 	else
2208 #endif
2209 #ifdef INET
2210 	{
2211 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2212 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2213 	}
2214 #endif
2215 	if (lgb != NULL)
2216 		lgb->tlb_errno = output_ret;
2217 }
2218 
2219 /*
2220  * Create a new TCP control block, making an empty reassembly queue and hooking
2221  * it to the argument protocol control block.  The `inp' parameter must have
2222  * come from the zone allocator set up by tcpcbstor declaration.
2223  */
2224 struct tcpcb *
2225 tcp_newtcpcb(struct inpcb *inp)
2226 {
2227 	struct tcpcb *tp = intotcpcb(inp);
2228 #ifdef INET6
2229 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2230 #endif /* INET6 */
2231 
2232 	/*
2233 	 * Historically allocation was done with M_ZERO.  There is a lot of
2234 	 * code that rely on that.  For now take safe approach and zero whole
2235 	 * tcpcb.  This definitely can be optimized.
2236 	 */
2237 	bzero(&tp->t_start_zero, t_zero_size);
2238 
2239 	/* Initialise cc_var struct for this tcpcb. */
2240 	tp->t_ccv.type = IPPROTO_TCP;
2241 	tp->t_ccv.ccvc.tcp = tp;
2242 	rw_rlock(&tcp_function_lock);
2243 	tp->t_fb = V_tcp_func_set_ptr;
2244 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2245 	rw_runlock(&tcp_function_lock);
2246 	/*
2247 	 * Use the current system default CC algorithm.
2248 	 */
2249 	cc_attach(tp, CC_DEFAULT_ALGO());
2250 
2251 	if (CC_ALGO(tp)->cb_init != NULL)
2252 		if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2253 			cc_detach(tp);
2254 			if (tp->t_fb->tfb_tcp_fb_fini)
2255 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2256 			refcount_release(&tp->t_fb->tfb_refcnt);
2257 			return (NULL);
2258 		}
2259 
2260 #ifdef TCP_HHOOK
2261 	if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2262 		if (tp->t_fb->tfb_tcp_fb_fini)
2263 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2264 		refcount_release(&tp->t_fb->tfb_refcnt);
2265 		return (NULL);
2266 	}
2267 #endif
2268 
2269 	TAILQ_INIT(&tp->t_segq);
2270 	STAILQ_INIT(&tp->t_inqueue);
2271 	tp->t_maxseg =
2272 #ifdef INET6
2273 		isipv6 ? V_tcp_v6mssdflt :
2274 #endif /* INET6 */
2275 		V_tcp_mssdflt;
2276 
2277 	/* All mbuf queue/ack compress flags should be off */
2278 	tcp_lro_features_off(tp);
2279 
2280 	callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2281 	for (int i = 0; i < TT_N; i++)
2282 		tp->t_timers[i] = SBT_MAX;
2283 
2284 	switch (V_tcp_do_rfc1323) {
2285 		case 0:
2286 			break;
2287 		default:
2288 		case 1:
2289 			tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2290 			break;
2291 		case 2:
2292 			tp->t_flags = TF_REQ_SCALE;
2293 			break;
2294 		case 3:
2295 			tp->t_flags = TF_REQ_TSTMP;
2296 			break;
2297 	}
2298 	if (V_tcp_do_sack)
2299 		tp->t_flags |= TF_SACK_PERMIT;
2300 	TAILQ_INIT(&tp->snd_holes);
2301 
2302 	/*
2303 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2304 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2305 	 * reasonable initial retransmit time.
2306 	 */
2307 	tp->t_srtt = TCPTV_SRTTBASE;
2308 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2309 	tp->t_rttmin = tcp_rexmit_min;
2310 	tp->t_rxtcur = tcp_rexmit_initial;
2311 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2312 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2313 	tp->t_rcvtime = ticks;
2314 	/* We always start with ticks granularity */
2315 	tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
2316 	/*
2317 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2318 	 * because the socket may be bound to an IPv6 wildcard address,
2319 	 * which may match an IPv4-mapped IPv6 address.
2320 	 */
2321 	inp->inp_ip_ttl = V_ip_defttl;
2322 #ifdef TCPHPTS
2323 	tcp_hpts_init(tp);
2324 #endif
2325 #ifdef TCPPCAP
2326 	/*
2327 	 * Init the TCP PCAP queues.
2328 	 */
2329 	tcp_pcap_tcpcb_init(tp);
2330 #endif
2331 #ifdef TCP_BLACKBOX
2332 	/* Initialize the per-TCPCB log data. */
2333 	tcp_log_tcpcbinit(tp);
2334 #endif
2335 	tp->t_pacing_rate = -1;
2336 	if (tp->t_fb->tfb_tcp_fb_init) {
2337 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) {
2338 			refcount_release(&tp->t_fb->tfb_refcnt);
2339 			return (NULL);
2340 		}
2341 	}
2342 #ifdef STATS
2343 	if (V_tcp_perconn_stats_enable == 1)
2344 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2345 #endif
2346 	if (V_tcp_do_lrd)
2347 		tp->t_flags |= TF_LRD;
2348 
2349 	return (tp);
2350 }
2351 
2352 /*
2353  * Drop a TCP connection, reporting
2354  * the specified error.  If connection is synchronized,
2355  * then send a RST to peer.
2356  */
2357 struct tcpcb *
2358 tcp_drop(struct tcpcb *tp, int errno)
2359 {
2360 	struct socket *so = tptosocket(tp);
2361 
2362 	NET_EPOCH_ASSERT();
2363 	INP_WLOCK_ASSERT(tptoinpcb(tp));
2364 
2365 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2366 		tcp_state_change(tp, TCPS_CLOSED);
2367 		/* Don't use tcp_output() here due to possible recursion. */
2368 		(void)tcp_output_nodrop(tp);
2369 		TCPSTAT_INC(tcps_drops);
2370 	} else
2371 		TCPSTAT_INC(tcps_conndrops);
2372 	if (errno == ETIMEDOUT && tp->t_softerror)
2373 		errno = tp->t_softerror;
2374 	so->so_error = errno;
2375 	return (tcp_close(tp));
2376 }
2377 
2378 void
2379 tcp_discardcb(struct tcpcb *tp)
2380 {
2381 	struct inpcb *inp = tptoinpcb(tp);
2382 	struct socket *so = tptosocket(tp);
2383 	struct mbuf *m;
2384 #ifdef INET6
2385 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2386 #endif
2387 
2388 	INP_WLOCK_ASSERT(inp);
2389 
2390 	tcp_timer_stop(tp);
2391 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2392 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2393 	}
2394 
2395 	/* free the reassembly queue, if any */
2396 	tcp_reass_flush(tp);
2397 
2398 #ifdef TCP_OFFLOAD
2399 	/* Disconnect offload device, if any. */
2400 	if (tp->t_flags & TF_TOE)
2401 		tcp_offload_detach(tp);
2402 #endif
2403 
2404 	tcp_free_sackholes(tp);
2405 
2406 #ifdef TCPPCAP
2407 	/* Free the TCP PCAP queues. */
2408 	tcp_pcap_drain(&(tp->t_inpkts));
2409 	tcp_pcap_drain(&(tp->t_outpkts));
2410 #endif
2411 
2412 	/* Allow the CC algorithm to clean up after itself. */
2413 	if (CC_ALGO(tp)->cb_destroy != NULL)
2414 		CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2415 	CC_DATA(tp) = NULL;
2416 	/* Detach from the CC algorithm */
2417 	cc_detach(tp);
2418 
2419 #ifdef TCP_HHOOK
2420 	khelp_destroy_osd(&tp->t_osd);
2421 #endif
2422 #ifdef STATS
2423 	stats_blob_destroy(tp->t_stats);
2424 #endif
2425 
2426 	CC_ALGO(tp) = NULL;
2427 	if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) {
2428 		struct mbuf *prev;
2429 
2430 		STAILQ_INIT(&tp->t_inqueue);
2431 		STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev)
2432 			m_freem(m);
2433 	}
2434 	TCPSTATES_DEC(tp->t_state);
2435 
2436 	if (tp->t_fb->tfb_tcp_fb_fini)
2437 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2438 	MPASS(!tcp_in_hpts(tp));
2439 #ifdef TCP_BLACKBOX
2440 	tcp_log_tcpcbfini(tp);
2441 #endif
2442 
2443 	/*
2444 	 * If we got enough samples through the srtt filter,
2445 	 * save the rtt and rttvar in the routing entry.
2446 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2447 	 * 4 samples is enough for the srtt filter to converge
2448 	 * to within enough % of the correct value; fewer samples
2449 	 * and we could save a bogus rtt. The danger is not high
2450 	 * as tcp quickly recovers from everything.
2451 	 * XXX: Works very well but needs some more statistics!
2452 	 *
2453 	 * XXXRRS: Updating must be after the stack fini() since
2454 	 * that may be converting some internal representation of
2455 	 * say srtt etc into the general one used by other stacks.
2456 	 * Lets also at least protect against the so being NULL
2457 	 * as RW stated below.
2458 	 */
2459 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2460 		struct hc_metrics_lite metrics;
2461 		uint32_t ssthresh;
2462 
2463 		bzero(&metrics, sizeof(metrics));
2464 		/*
2465 		 * Update the ssthresh always when the conditions below
2466 		 * are satisfied. This gives us better new start value
2467 		 * for the congestion avoidance for new connections.
2468 		 * ssthresh is only set if packet loss occurred on a session.
2469 		 *
2470 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2471 		 * being torn down.  Ideally this code would not use 'so'.
2472 		 */
2473 		ssthresh = tp->snd_ssthresh;
2474 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2475 			/*
2476 			 * convert the limit from user data bytes to
2477 			 * packets then to packet data bytes.
2478 			 */
2479 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2480 			if (ssthresh < 2)
2481 				ssthresh = 2;
2482 			ssthresh *= (tp->t_maxseg +
2483 #ifdef INET6
2484 			    (isipv6 ? sizeof (struct ip6_hdr) +
2485 			    sizeof (struct tcphdr) :
2486 #endif
2487 			    sizeof (struct tcpiphdr)
2488 #ifdef INET6
2489 			    )
2490 #endif
2491 			    );
2492 		} else
2493 			ssthresh = 0;
2494 		metrics.rmx_ssthresh = ssthresh;
2495 
2496 		metrics.rmx_rtt = tp->t_srtt;
2497 		metrics.rmx_rttvar = tp->t_rttvar;
2498 		metrics.rmx_cwnd = tp->snd_cwnd;
2499 		metrics.rmx_sendpipe = 0;
2500 		metrics.rmx_recvpipe = 0;
2501 
2502 		tcp_hc_update(&inp->inp_inc, &metrics);
2503 	}
2504 
2505 	refcount_release(&tp->t_fb->tfb_refcnt);
2506 }
2507 
2508 /*
2509  * Attempt to close a TCP control block, marking it as dropped, and freeing
2510  * the socket if we hold the only reference.
2511  */
2512 struct tcpcb *
2513 tcp_close(struct tcpcb *tp)
2514 {
2515 	struct inpcb *inp = tptoinpcb(tp);
2516 	struct socket *so = tptosocket(tp);
2517 
2518 	INP_WLOCK_ASSERT(inp);
2519 
2520 #ifdef TCP_OFFLOAD
2521 	if (tp->t_state == TCPS_LISTEN)
2522 		tcp_offload_listen_stop(tp);
2523 #endif
2524 	/*
2525 	 * This releases the TFO pending counter resource for TFO listen
2526 	 * sockets as well as passively-created TFO sockets that transition
2527 	 * from SYN_RECEIVED to CLOSED.
2528 	 */
2529 	if (tp->t_tfo_pending) {
2530 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2531 		tp->t_tfo_pending = NULL;
2532 	}
2533 #ifdef TCPHPTS
2534 	tcp_hpts_remove(tp);
2535 #endif
2536 	in_pcbdrop(inp);
2537 	TCPSTAT_INC(tcps_closed);
2538 	if (tp->t_state != TCPS_CLOSED)
2539 		tcp_state_change(tp, TCPS_CLOSED);
2540 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2541 	soisdisconnected(so);
2542 	if (inp->inp_flags & INP_SOCKREF) {
2543 		inp->inp_flags &= ~INP_SOCKREF;
2544 		INP_WUNLOCK(inp);
2545 		sorele(so);
2546 		return (NULL);
2547 	}
2548 	return (tp);
2549 }
2550 
2551 /*
2552  * Notify a tcp user of an asynchronous error;
2553  * store error as soft error, but wake up user
2554  * (for now, won't do anything until can select for soft error).
2555  *
2556  * Do not wake up user since there currently is no mechanism for
2557  * reporting soft errors (yet - a kqueue filter may be added).
2558  */
2559 static struct inpcb *
2560 tcp_notify(struct inpcb *inp, int error)
2561 {
2562 	struct tcpcb *tp;
2563 
2564 	INP_WLOCK_ASSERT(inp);
2565 
2566 	tp = intotcpcb(inp);
2567 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2568 
2569 	/*
2570 	 * Ignore some errors if we are hooked up.
2571 	 * If connection hasn't completed, has retransmitted several times,
2572 	 * and receives a second error, give up now.  This is better
2573 	 * than waiting a long time to establish a connection that
2574 	 * can never complete.
2575 	 */
2576 	if (tp->t_state == TCPS_ESTABLISHED &&
2577 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2578 	     error == EHOSTDOWN)) {
2579 		if (inp->inp_route.ro_nh) {
2580 			NH_FREE(inp->inp_route.ro_nh);
2581 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2582 		}
2583 		return (inp);
2584 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2585 	    tp->t_softerror) {
2586 		tp = tcp_drop(tp, error);
2587 		if (tp != NULL)
2588 			return (inp);
2589 		else
2590 			return (NULL);
2591 	} else {
2592 		tp->t_softerror = error;
2593 		return (inp);
2594 	}
2595 #if 0
2596 	wakeup( &so->so_timeo);
2597 	sorwakeup(so);
2598 	sowwakeup(so);
2599 #endif
2600 }
2601 
2602 static int
2603 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2604 {
2605 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2606 	    INPLOOKUP_RLOCKPCB);
2607 	struct xinpgen xig;
2608 	struct inpcb *inp;
2609 	int error;
2610 
2611 	if (req->newptr != NULL)
2612 		return (EPERM);
2613 
2614 	if (req->oldptr == NULL) {
2615 		int n;
2616 
2617 		n = V_tcbinfo.ipi_count +
2618 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2619 		n += imax(n / 8, 10);
2620 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2621 		return (0);
2622 	}
2623 
2624 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2625 		return (error);
2626 
2627 	bzero(&xig, sizeof(xig));
2628 	xig.xig_len = sizeof xig;
2629 	xig.xig_count = V_tcbinfo.ipi_count +
2630 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2631 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2632 	xig.xig_sogen = so_gencnt;
2633 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2634 	if (error)
2635 		return (error);
2636 
2637 	error = syncache_pcblist(req);
2638 	if (error)
2639 		return (error);
2640 
2641 	while ((inp = inp_next(&inpi)) != NULL) {
2642 		if (inp->inp_gencnt <= xig.xig_gen &&
2643 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2644 			struct xtcpcb xt;
2645 
2646 			tcp_inptoxtp(inp, &xt);
2647 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2648 			if (error) {
2649 				INP_RUNLOCK(inp);
2650 				break;
2651 			} else
2652 				continue;
2653 		}
2654 	}
2655 
2656 	if (!error) {
2657 		/*
2658 		 * Give the user an updated idea of our state.
2659 		 * If the generation differs from what we told
2660 		 * her before, she knows that something happened
2661 		 * while we were processing this request, and it
2662 		 * might be necessary to retry.
2663 		 */
2664 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2665 		xig.xig_sogen = so_gencnt;
2666 		xig.xig_count = V_tcbinfo.ipi_count +
2667 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2668 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2669 	}
2670 
2671 	return (error);
2672 }
2673 
2674 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2675     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2676     NULL, 0, tcp_pcblist, "S,xtcpcb",
2677     "List of active TCP connections");
2678 
2679 #ifdef INET
2680 static int
2681 tcp_getcred(SYSCTL_HANDLER_ARGS)
2682 {
2683 	struct xucred xuc;
2684 	struct sockaddr_in addrs[2];
2685 	struct epoch_tracker et;
2686 	struct inpcb *inp;
2687 	int error;
2688 
2689 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2690 	if (error)
2691 		return (error);
2692 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2693 	if (error)
2694 		return (error);
2695 	NET_EPOCH_ENTER(et);
2696 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2697 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2698 	NET_EPOCH_EXIT(et);
2699 	if (inp != NULL) {
2700 		if (error == 0)
2701 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2702 		if (error == 0)
2703 			cru2x(inp->inp_cred, &xuc);
2704 		INP_RUNLOCK(inp);
2705 	} else
2706 		error = ENOENT;
2707 	if (error == 0)
2708 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2709 	return (error);
2710 }
2711 
2712 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2713     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2714     0, 0, tcp_getcred, "S,xucred",
2715     "Get the xucred of a TCP connection");
2716 #endif /* INET */
2717 
2718 #ifdef INET6
2719 static int
2720 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2721 {
2722 	struct epoch_tracker et;
2723 	struct xucred xuc;
2724 	struct sockaddr_in6 addrs[2];
2725 	struct inpcb *inp;
2726 	int error;
2727 #ifdef INET
2728 	int mapped = 0;
2729 #endif
2730 
2731 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2732 	if (error)
2733 		return (error);
2734 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2735 	if (error)
2736 		return (error);
2737 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2738 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2739 		return (error);
2740 	}
2741 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2742 #ifdef INET
2743 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2744 			mapped = 1;
2745 		else
2746 #endif
2747 			return (EINVAL);
2748 	}
2749 
2750 	NET_EPOCH_ENTER(et);
2751 #ifdef INET
2752 	if (mapped == 1)
2753 		inp = in_pcblookup(&V_tcbinfo,
2754 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2755 			addrs[1].sin6_port,
2756 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2757 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2758 	else
2759 #endif
2760 		inp = in6_pcblookup(&V_tcbinfo,
2761 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2762 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2763 			INPLOOKUP_RLOCKPCB, NULL);
2764 	NET_EPOCH_EXIT(et);
2765 	if (inp != NULL) {
2766 		if (error == 0)
2767 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2768 		if (error == 0)
2769 			cru2x(inp->inp_cred, &xuc);
2770 		INP_RUNLOCK(inp);
2771 	} else
2772 		error = ENOENT;
2773 	if (error == 0)
2774 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2775 	return (error);
2776 }
2777 
2778 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2779     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2780     0, 0, tcp6_getcred, "S,xucred",
2781     "Get the xucred of a TCP6 connection");
2782 #endif /* INET6 */
2783 
2784 #ifdef INET
2785 /* Path MTU to try next when a fragmentation-needed message is received. */
2786 static inline int
2787 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2788 {
2789 	int mtu = ntohs(icp->icmp_nextmtu);
2790 
2791 	/* If no alternative MTU was proposed, try the next smaller one. */
2792 	if (!mtu)
2793 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2794 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2795 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2796 
2797 	return (mtu);
2798 }
2799 
2800 static void
2801 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2802 {
2803 	struct ip *ip;
2804 	struct tcphdr *th;
2805 	struct inpcb *inp;
2806 	struct tcpcb *tp;
2807 	struct inpcb *(*notify)(struct inpcb *, int);
2808 	struct in_conninfo inc;
2809 	tcp_seq icmp_tcp_seq;
2810 	int errno, mtu;
2811 
2812 	errno = icmp_errmap(icp);
2813 	switch (errno) {
2814 	case 0:
2815 		return;
2816 	case EMSGSIZE:
2817 		notify = tcp_mtudisc_notify;
2818 		break;
2819 	case ECONNREFUSED:
2820 		if (V_icmp_may_rst)
2821 			notify = tcp_drop_syn_sent;
2822 		else
2823 			notify = tcp_notify;
2824 		break;
2825 	case EHOSTUNREACH:
2826 		if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2827 			notify = tcp_drop_syn_sent;
2828 		else
2829 			notify = tcp_notify;
2830 		break;
2831 	default:
2832 		notify = tcp_notify;
2833 	}
2834 
2835 	ip = &icp->icmp_ip;
2836 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2837 	icmp_tcp_seq = th->th_seq;
2838 	inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2839 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2840 	if (inp != NULL)  {
2841 		tp = intotcpcb(inp);
2842 #ifdef TCP_OFFLOAD
2843 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2844 			/*
2845 			 * MTU discovery for offloaded connections.  Let
2846 			 * the TOE driver verify seq# and process it.
2847 			 */
2848 			mtu = tcp_next_pmtu(icp, ip);
2849 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2850 			goto out;
2851 		}
2852 #endif
2853 		if (tp->t_port != port)
2854 			goto out;
2855 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2856 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2857 			if (errno == EMSGSIZE) {
2858 				/*
2859 				 * MTU discovery: we got a needfrag and
2860 				 * will potentially try a lower MTU.
2861 				 */
2862 				mtu = tcp_next_pmtu(icp, ip);
2863 
2864 				/*
2865 				 * Only process the offered MTU if it
2866 				 * is smaller than the current one.
2867 				 */
2868 				if (mtu < tp->t_maxseg +
2869 				    sizeof(struct tcpiphdr)) {
2870 					bzero(&inc, sizeof(inc));
2871 					inc.inc_faddr = ip->ip_dst;
2872 					inc.inc_fibnum =
2873 					    inp->inp_inc.inc_fibnum;
2874 					tcp_hc_updatemtu(&inc, mtu);
2875 					inp = tcp_mtudisc(inp, mtu);
2876 				}
2877 			} else
2878 				inp = (*notify)(inp, errno);
2879 		}
2880 	} else {
2881 		bzero(&inc, sizeof(inc));
2882 		inc.inc_fport = th->th_dport;
2883 		inc.inc_lport = th->th_sport;
2884 		inc.inc_faddr = ip->ip_dst;
2885 		inc.inc_laddr = ip->ip_src;
2886 		syncache_unreach(&inc, icmp_tcp_seq, port);
2887 	}
2888 out:
2889 	if (inp != NULL)
2890 		INP_WUNLOCK(inp);
2891 }
2892 
2893 static void
2894 tcp_ctlinput(struct icmp *icmp)
2895 {
2896 	tcp_ctlinput_with_port(icmp, htons(0));
2897 }
2898 
2899 static void
2900 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2901 {
2902 	/* Its a tunneled TCP over UDP icmp */
2903 	struct icmp *icmp = param.icmp;
2904 	struct ip *outer_ip, *inner_ip;
2905 	struct udphdr *udp;
2906 	struct tcphdr *th, ttemp;
2907 	int i_hlen, o_len;
2908 	uint16_t port;
2909 
2910 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2911 	inner_ip = &icmp->icmp_ip;
2912 	i_hlen = inner_ip->ip_hl << 2;
2913 	o_len = ntohs(outer_ip->ip_len);
2914 	if (o_len <
2915 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2916 		/* Not enough data present */
2917 		return;
2918 	}
2919 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2920 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2921 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2922 		return;
2923 	}
2924 	port = udp->uh_dport;
2925 	th = (struct tcphdr *)(udp + 1);
2926 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2927 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2928 	/* Now adjust down the size of the outer IP header */
2929 	o_len -= sizeof(struct udphdr);
2930 	outer_ip->ip_len = htons(o_len);
2931 	/* Now call in to the normal handling code */
2932 	tcp_ctlinput_with_port(icmp, port);
2933 }
2934 #endif /* INET */
2935 
2936 #ifdef INET6
2937 static inline int
2938 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2939 {
2940 	int mtu = ntohl(icmp6->icmp6_mtu);
2941 
2942 	/*
2943 	 * If no alternative MTU was proposed, or the proposed MTU was too
2944 	 * small, set to the min.
2945 	 */
2946 	if (mtu < IPV6_MMTU)
2947 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
2948 	return (mtu);
2949 }
2950 
2951 static void
2952 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2953 {
2954 	struct in6_addr *dst;
2955 	struct inpcb *(*notify)(struct inpcb *, int);
2956 	struct ip6_hdr *ip6;
2957 	struct mbuf *m;
2958 	struct inpcb *inp;
2959 	struct tcpcb *tp;
2960 	struct icmp6_hdr *icmp6;
2961 	struct in_conninfo inc;
2962 	struct tcp_ports {
2963 		uint16_t th_sport;
2964 		uint16_t th_dport;
2965 	} t_ports;
2966 	tcp_seq icmp_tcp_seq;
2967 	unsigned int mtu;
2968 	unsigned int off;
2969 	int errno;
2970 
2971 	icmp6 = ip6cp->ip6c_icmp6;
2972 	m = ip6cp->ip6c_m;
2973 	ip6 = ip6cp->ip6c_ip6;
2974 	off = ip6cp->ip6c_off;
2975 	dst = &ip6cp->ip6c_finaldst->sin6_addr;
2976 
2977 	errno = icmp6_errmap(icmp6);
2978 	switch (errno) {
2979 	case 0:
2980 		return;
2981 	case EMSGSIZE:
2982 		notify = tcp_mtudisc_notify;
2983 		break;
2984 	case ECONNREFUSED:
2985 		if (V_icmp_may_rst)
2986 			notify = tcp_drop_syn_sent;
2987 		else
2988 			notify = tcp_notify;
2989 		break;
2990 	case EHOSTUNREACH:
2991 		/*
2992 		 * There are only four ICMPs that may reset connection:
2993 		 * - administratively prohibited
2994 		 * - port unreachable
2995 		 * - time exceeded in transit
2996 		 * - unknown next header
2997 		 */
2998 		if (V_icmp_may_rst &&
2999 		    ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
3000 		     (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
3001 		      icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
3002 		    (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
3003 		      icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
3004 		    (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
3005 		      icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
3006 			notify = tcp_drop_syn_sent;
3007 		else
3008 			notify = tcp_notify;
3009 		break;
3010 	default:
3011 		notify = tcp_notify;
3012 	}
3013 
3014 	/* Check if we can safely get the ports from the tcp hdr */
3015 	if (m == NULL ||
3016 	    (m->m_pkthdr.len <
3017 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3018 		return;
3019 	}
3020 	bzero(&t_ports, sizeof(struct tcp_ports));
3021 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3022 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3023 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3024 	off += sizeof(struct tcp_ports);
3025 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3026 		goto out;
3027 	}
3028 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3029 	if (inp != NULL)  {
3030 		tp = intotcpcb(inp);
3031 #ifdef TCP_OFFLOAD
3032 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
3033 			/* MTU discovery for offloaded connections. */
3034 			mtu = tcp6_next_pmtu(icmp6);
3035 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3036 			goto out;
3037 		}
3038 #endif
3039 		if (tp->t_port != port)
3040 			goto out;
3041 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3042 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3043 			if (errno == EMSGSIZE) {
3044 				/*
3045 				 * MTU discovery:
3046 				 * If we got a needfrag set the MTU
3047 				 * in the route to the suggested new
3048 				 * value (if given) and then notify.
3049 				 */
3050 				mtu = tcp6_next_pmtu(icmp6);
3051 
3052 				bzero(&inc, sizeof(inc));
3053 				inc.inc_fibnum = M_GETFIB(m);
3054 				inc.inc_flags |= INC_ISIPV6;
3055 				inc.inc6_faddr = *dst;
3056 				if (in6_setscope(&inc.inc6_faddr,
3057 					m->m_pkthdr.rcvif, NULL))
3058 					goto out;
3059 				/*
3060 				 * Only process the offered MTU if it
3061 				 * is smaller than the current one.
3062 				 */
3063 				if (mtu < tp->t_maxseg +
3064 				    sizeof (struct tcphdr) +
3065 				    sizeof (struct ip6_hdr)) {
3066 					tcp_hc_updatemtu(&inc, mtu);
3067 					tcp_mtudisc(inp, mtu);
3068 					ICMP6STAT_INC(icp6s_pmtuchg);
3069 				}
3070 			} else
3071 				inp = (*notify)(inp, errno);
3072 		}
3073 	} else {
3074 		bzero(&inc, sizeof(inc));
3075 		inc.inc_fibnum = M_GETFIB(m);
3076 		inc.inc_flags |= INC_ISIPV6;
3077 		inc.inc_fport = t_ports.th_dport;
3078 		inc.inc_lport = t_ports.th_sport;
3079 		inc.inc6_faddr = *dst;
3080 		inc.inc6_laddr = ip6->ip6_src;
3081 		syncache_unreach(&inc, icmp_tcp_seq, port);
3082 	}
3083 out:
3084 	if (inp != NULL)
3085 		INP_WUNLOCK(inp);
3086 }
3087 
3088 static void
3089 tcp6_ctlinput(struct ip6ctlparam *ctl)
3090 {
3091 	tcp6_ctlinput_with_port(ctl, htons(0));
3092 }
3093 
3094 static void
3095 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3096 {
3097 	struct ip6ctlparam *ip6cp = param.ip6cp;
3098 	struct mbuf *m;
3099 	struct udphdr *udp;
3100 	uint16_t port;
3101 
3102 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3103 	if (m == NULL) {
3104 		return;
3105 	}
3106 	udp = mtod(m, struct udphdr *);
3107 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3108 		return;
3109 	}
3110 	port = udp->uh_dport;
3111 	m_adj(m, sizeof(struct udphdr));
3112 	if ((m->m_flags & M_PKTHDR) == 0) {
3113 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3114 	}
3115 	/* Now call in to the normal handling code */
3116 	tcp6_ctlinput_with_port(ip6cp, port);
3117 }
3118 
3119 #endif /* INET6 */
3120 
3121 static uint32_t
3122 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3123 {
3124 	SIPHASH_CTX ctx;
3125 	uint32_t hash[2];
3126 
3127 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3128 	    ("%s: keylen %u too short ", __func__, len));
3129 	SipHash24_Init(&ctx);
3130 	SipHash_SetKey(&ctx, (uint8_t *)key);
3131 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3132 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3133 	switch (inc->inc_flags & INC_ISIPV6) {
3134 #ifdef INET
3135 	case 0:
3136 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3137 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3138 		break;
3139 #endif
3140 #ifdef INET6
3141 	case INC_ISIPV6:
3142 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3143 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3144 		break;
3145 #endif
3146 	}
3147 	SipHash_Final((uint8_t *)hash, &ctx);
3148 
3149 	return (hash[0] ^ hash[1]);
3150 }
3151 
3152 uint32_t
3153 tcp_new_ts_offset(struct in_conninfo *inc)
3154 {
3155 	struct in_conninfo inc_store, *local_inc;
3156 
3157 	if (!V_tcp_ts_offset_per_conn) {
3158 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3159 		inc_store.inc_lport = 0;
3160 		inc_store.inc_fport = 0;
3161 		local_inc = &inc_store;
3162 	} else {
3163 		local_inc = inc;
3164 	}
3165 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3166 	    sizeof(V_ts_offset_secret)));
3167 }
3168 
3169 /*
3170  * Following is where TCP initial sequence number generation occurs.
3171  *
3172  * There are two places where we must use initial sequence numbers:
3173  * 1.  In SYN-ACK packets.
3174  * 2.  In SYN packets.
3175  *
3176  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3177  * tcp_syncache.c for details.
3178  *
3179  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3180  * depends on this property.  In addition, these ISNs should be
3181  * unguessable so as to prevent connection hijacking.  To satisfy
3182  * the requirements of this situation, the algorithm outlined in
3183  * RFC 1948 is used, with only small modifications.
3184  *
3185  * Implementation details:
3186  *
3187  * Time is based off the system timer, and is corrected so that it
3188  * increases by one megabyte per second.  This allows for proper
3189  * recycling on high speed LANs while still leaving over an hour
3190  * before rollover.
3191  *
3192  * As reading the *exact* system time is too expensive to be done
3193  * whenever setting up a TCP connection, we increment the time
3194  * offset in two ways.  First, a small random positive increment
3195  * is added to isn_offset for each connection that is set up.
3196  * Second, the function tcp_isn_tick fires once per clock tick
3197  * and increments isn_offset as necessary so that sequence numbers
3198  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3199  * random positive increments serve only to ensure that the same
3200  * exact sequence number is never sent out twice (as could otherwise
3201  * happen when a port is recycled in less than the system tick
3202  * interval.)
3203  *
3204  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3205  * between seeding of isn_secret.  This is normally set to zero,
3206  * as reseeding should not be necessary.
3207  *
3208  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3209  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3210  * general, this means holding an exclusive (write) lock.
3211  */
3212 
3213 #define ISN_BYTES_PER_SECOND 1048576
3214 #define ISN_STATIC_INCREMENT 4096
3215 #define ISN_RANDOM_INCREMENT (4096 - 1)
3216 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3217 
3218 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3219 VNET_DEFINE_STATIC(int, isn_last);
3220 VNET_DEFINE_STATIC(int, isn_last_reseed);
3221 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3222 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3223 
3224 #define	V_isn_secret			VNET(isn_secret)
3225 #define	V_isn_last			VNET(isn_last)
3226 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3227 #define	V_isn_offset			VNET(isn_offset)
3228 #define	V_isn_offset_old		VNET(isn_offset_old)
3229 
3230 tcp_seq
3231 tcp_new_isn(struct in_conninfo *inc)
3232 {
3233 	tcp_seq new_isn;
3234 	u_int32_t projected_offset;
3235 
3236 	ISN_LOCK();
3237 	/* Seed if this is the first use, reseed if requested. */
3238 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3239 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3240 		< (u_int)ticks))) {
3241 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3242 		V_isn_last_reseed = ticks;
3243 	}
3244 
3245 	/* Compute the hash and return the ISN. */
3246 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3247 	    sizeof(V_isn_secret));
3248 	V_isn_offset += ISN_STATIC_INCREMENT +
3249 		(arc4random() & ISN_RANDOM_INCREMENT);
3250 	if (ticks != V_isn_last) {
3251 		projected_offset = V_isn_offset_old +
3252 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3253 		if (SEQ_GT(projected_offset, V_isn_offset))
3254 			V_isn_offset = projected_offset;
3255 		V_isn_offset_old = V_isn_offset;
3256 		V_isn_last = ticks;
3257 	}
3258 	new_isn += V_isn_offset;
3259 	ISN_UNLOCK();
3260 	return (new_isn);
3261 }
3262 
3263 /*
3264  * When a specific ICMP unreachable message is received and the
3265  * connection state is SYN-SENT, drop the connection.  This behavior
3266  * is controlled by the icmp_may_rst sysctl.
3267  */
3268 static struct inpcb *
3269 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3270 {
3271 	struct tcpcb *tp;
3272 
3273 	NET_EPOCH_ASSERT();
3274 	INP_WLOCK_ASSERT(inp);
3275 
3276 	tp = intotcpcb(inp);
3277 	if (tp->t_state != TCPS_SYN_SENT)
3278 		return (inp);
3279 
3280 	if (IS_FASTOPEN(tp->t_flags))
3281 		tcp_fastopen_disable_path(tp);
3282 
3283 	tp = tcp_drop(tp, errno);
3284 	if (tp != NULL)
3285 		return (inp);
3286 	else
3287 		return (NULL);
3288 }
3289 
3290 /*
3291  * When `need fragmentation' ICMP is received, update our idea of the MSS
3292  * based on the new value. Also nudge TCP to send something, since we
3293  * know the packet we just sent was dropped.
3294  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3295  */
3296 static struct inpcb *
3297 tcp_mtudisc_notify(struct inpcb *inp, int error)
3298 {
3299 
3300 	return (tcp_mtudisc(inp, -1));
3301 }
3302 
3303 static struct inpcb *
3304 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3305 {
3306 	struct tcpcb *tp;
3307 	struct socket *so;
3308 
3309 	INP_WLOCK_ASSERT(inp);
3310 
3311 	tp = intotcpcb(inp);
3312 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3313 
3314 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3315 
3316 	so = inp->inp_socket;
3317 	SOCKBUF_LOCK(&so->so_snd);
3318 	/* If the mss is larger than the socket buffer, decrease the mss. */
3319 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3320 		tp->t_maxseg = so->so_snd.sb_hiwat;
3321 	SOCKBUF_UNLOCK(&so->so_snd);
3322 
3323 	TCPSTAT_INC(tcps_mturesent);
3324 	tp->t_rtttime = 0;
3325 	tp->snd_nxt = tp->snd_una;
3326 	tcp_free_sackholes(tp);
3327 	tp->snd_recover = tp->snd_max;
3328 	if (tp->t_flags & TF_SACK_PERMIT)
3329 		EXIT_FASTRECOVERY(tp->t_flags);
3330 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3331 		/*
3332 		 * Conceptually the snd_nxt setting
3333 		 * and freeing sack holes should
3334 		 * be done by the default stacks
3335 		 * own tfb_tcp_mtu_chg().
3336 		 */
3337 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3338 	}
3339 	if (tcp_output(tp) < 0)
3340 		return (NULL);
3341 	else
3342 		return (inp);
3343 }
3344 
3345 #ifdef INET
3346 /*
3347  * Look-up the routing entry to the peer of this inpcb.  If no route
3348  * is found and it cannot be allocated, then return 0.  This routine
3349  * is called by TCP routines that access the rmx structure and by
3350  * tcp_mss_update to get the peer/interface MTU.
3351  */
3352 uint32_t
3353 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3354 {
3355 	struct nhop_object *nh;
3356 	struct ifnet *ifp;
3357 	uint32_t maxmtu = 0;
3358 
3359 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3360 
3361 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3362 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3363 		if (nh == NULL)
3364 			return (0);
3365 
3366 		ifp = nh->nh_ifp;
3367 		maxmtu = nh->nh_mtu;
3368 
3369 		/* Report additional interface capabilities. */
3370 		if (cap != NULL) {
3371 			if (ifp->if_capenable & IFCAP_TSO4 &&
3372 			    ifp->if_hwassist & CSUM_TSO) {
3373 				cap->ifcap |= CSUM_TSO;
3374 				cap->tsomax = ifp->if_hw_tsomax;
3375 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3376 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3377 			}
3378 		}
3379 	}
3380 	return (maxmtu);
3381 }
3382 #endif /* INET */
3383 
3384 #ifdef INET6
3385 uint32_t
3386 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3387 {
3388 	struct nhop_object *nh;
3389 	struct in6_addr dst6;
3390 	uint32_t scopeid;
3391 	struct ifnet *ifp;
3392 	uint32_t maxmtu = 0;
3393 
3394 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3395 
3396 	if (inc->inc_flags & INC_IPV6MINMTU)
3397 		return (IPV6_MMTU);
3398 
3399 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3400 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3401 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3402 		if (nh == NULL)
3403 			return (0);
3404 
3405 		ifp = nh->nh_ifp;
3406 		maxmtu = nh->nh_mtu;
3407 
3408 		/* Report additional interface capabilities. */
3409 		if (cap != NULL) {
3410 			if (ifp->if_capenable & IFCAP_TSO6 &&
3411 			    ifp->if_hwassist & CSUM_TSO) {
3412 				cap->ifcap |= CSUM_TSO;
3413 				cap->tsomax = ifp->if_hw_tsomax;
3414 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3415 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3416 			}
3417 		}
3418 	}
3419 
3420 	return (maxmtu);
3421 }
3422 
3423 /*
3424  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3425  *
3426  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3427  * The right place to do that is ip6_setpktopt() that has just been
3428  * executed.  By the way it just filled ip6po_minmtu for us.
3429  */
3430 void
3431 tcp6_use_min_mtu(struct tcpcb *tp)
3432 {
3433 	struct inpcb *inp = tptoinpcb(tp);
3434 
3435 	INP_WLOCK_ASSERT(inp);
3436 	/*
3437 	 * In case of the IPV6_USE_MIN_MTU socket
3438 	 * option, the INC_IPV6MINMTU flag to announce
3439 	 * a corresponding MSS during the initial
3440 	 * handshake.  If the TCP connection is not in
3441 	 * the front states, just reduce the MSS being
3442 	 * used.  This avoids the sending of TCP
3443 	 * segments which will be fragmented at the
3444 	 * IPv6 layer.
3445 	 */
3446 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3447 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3448 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3449 		struct ip6_pktopts *opt;
3450 
3451 		opt = inp->in6p_outputopts;
3452 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3453 		    tp->t_maxseg > TCP6_MSS)
3454 			tp->t_maxseg = TCP6_MSS;
3455 	}
3456 }
3457 #endif /* INET6 */
3458 
3459 /*
3460  * Calculate effective SMSS per RFC5681 definition for a given TCP
3461  * connection at its current state, taking into account SACK and etc.
3462  */
3463 u_int
3464 tcp_maxseg(const struct tcpcb *tp)
3465 {
3466 	u_int optlen;
3467 
3468 	if (tp->t_flags & TF_NOOPT)
3469 		return (tp->t_maxseg);
3470 
3471 	/*
3472 	 * Here we have a simplified code from tcp_addoptions(),
3473 	 * without a proper loop, and having most of paddings hardcoded.
3474 	 * We might make mistakes with padding here in some edge cases,
3475 	 * but this is harmless, since result of tcp_maxseg() is used
3476 	 * only in cwnd and ssthresh estimations.
3477 	 */
3478 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3479 		if (tp->t_flags & TF_RCVD_TSTMP)
3480 			optlen = TCPOLEN_TSTAMP_APPA;
3481 		else
3482 			optlen = 0;
3483 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3484 		if (tp->t_flags & TF_SIGNATURE)
3485 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3486 #endif
3487 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3488 			optlen += TCPOLEN_SACKHDR;
3489 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3490 			optlen = PADTCPOLEN(optlen);
3491 		}
3492 	} else {
3493 		if (tp->t_flags & TF_REQ_TSTMP)
3494 			optlen = TCPOLEN_TSTAMP_APPA;
3495 		else
3496 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3497 		if (tp->t_flags & TF_REQ_SCALE)
3498 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3499 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3500 		if (tp->t_flags & TF_SIGNATURE)
3501 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3502 #endif
3503 		if (tp->t_flags & TF_SACK_PERMIT)
3504 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3505 	}
3506 #undef PAD
3507 	optlen = min(optlen, TCP_MAXOLEN);
3508 	return (tp->t_maxseg - optlen);
3509 }
3510 
3511 
3512 u_int
3513 tcp_fixed_maxseg(const struct tcpcb *tp)
3514 {
3515 	int optlen;
3516 
3517 	if (tp->t_flags & TF_NOOPT)
3518 		return (tp->t_maxseg);
3519 
3520 	/*
3521 	 * Here we have a simplified code from tcp_addoptions(),
3522 	 * without a proper loop, and having most of paddings hardcoded.
3523 	 * We only consider fixed options that we would send every
3524 	 * time I.e. SACK is not considered. This is important
3525 	 * for cc modules to figure out what the modulo of the
3526 	 * cwnd should be.
3527 	 */
3528 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3529 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3530 		if (tp->t_flags & TF_RCVD_TSTMP)
3531 			optlen = TCPOLEN_TSTAMP_APPA;
3532 		else
3533 			optlen = 0;
3534 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3535 		if (tp->t_flags & TF_SIGNATURE)
3536 			optlen += PAD(TCPOLEN_SIGNATURE);
3537 #endif
3538 	} else {
3539 		if (tp->t_flags & TF_REQ_TSTMP)
3540 			optlen = TCPOLEN_TSTAMP_APPA;
3541 		else
3542 			optlen = PAD(TCPOLEN_MAXSEG);
3543 		if (tp->t_flags & TF_REQ_SCALE)
3544 			optlen += PAD(TCPOLEN_WINDOW);
3545 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3546 		if (tp->t_flags & TF_SIGNATURE)
3547 			optlen += PAD(TCPOLEN_SIGNATURE);
3548 #endif
3549 		if (tp->t_flags & TF_SACK_PERMIT)
3550 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3551 	}
3552 #undef PAD
3553 	optlen = min(optlen, TCP_MAXOLEN);
3554 	return (tp->t_maxseg - optlen);
3555 }
3556 
3557 
3558 
3559 static int
3560 sysctl_drop(SYSCTL_HANDLER_ARGS)
3561 {
3562 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3563 	struct sockaddr_storage addrs[2];
3564 	struct inpcb *inp;
3565 	struct tcpcb *tp;
3566 #ifdef INET
3567 	struct sockaddr_in *fin = NULL, *lin = NULL;
3568 #endif
3569 	struct epoch_tracker et;
3570 #ifdef INET6
3571 	struct sockaddr_in6 *fin6, *lin6;
3572 #endif
3573 	int error;
3574 
3575 	inp = NULL;
3576 #ifdef INET6
3577 	fin6 = lin6 = NULL;
3578 #endif
3579 	error = 0;
3580 
3581 	if (req->oldptr != NULL || req->oldlen != 0)
3582 		return (EINVAL);
3583 	if (req->newptr == NULL)
3584 		return (EPERM);
3585 	if (req->newlen < sizeof(addrs))
3586 		return (ENOMEM);
3587 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3588 	if (error)
3589 		return (error);
3590 
3591 	switch (addrs[0].ss_family) {
3592 #ifdef INET6
3593 	case AF_INET6:
3594 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3595 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3596 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3597 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3598 			return (EINVAL);
3599 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3600 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3601 				return (EINVAL);
3602 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3603 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3604 #ifdef INET
3605 			fin = (struct sockaddr_in *)&addrs[0];
3606 			lin = (struct sockaddr_in *)&addrs[1];
3607 #endif
3608 			break;
3609 		}
3610 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3611 		if (error)
3612 			return (error);
3613 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3614 		if (error)
3615 			return (error);
3616 		break;
3617 #endif
3618 #ifdef INET
3619 	case AF_INET:
3620 		fin = (struct sockaddr_in *)&addrs[0];
3621 		lin = (struct sockaddr_in *)&addrs[1];
3622 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3623 		    lin->sin_len != sizeof(struct sockaddr_in))
3624 			return (EINVAL);
3625 		break;
3626 #endif
3627 	default:
3628 		return (EINVAL);
3629 	}
3630 	NET_EPOCH_ENTER(et);
3631 	switch (addrs[0].ss_family) {
3632 #ifdef INET6
3633 	case AF_INET6:
3634 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3635 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3636 		    INPLOOKUP_WLOCKPCB, NULL);
3637 		break;
3638 #endif
3639 #ifdef INET
3640 	case AF_INET:
3641 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3642 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3643 		break;
3644 #endif
3645 	}
3646 	if (inp != NULL) {
3647 		if (!SOLISTENING(inp->inp_socket)) {
3648 			tp = intotcpcb(inp);
3649 			tp = tcp_drop(tp, ECONNABORTED);
3650 			if (tp != NULL)
3651 				INP_WUNLOCK(inp);
3652 		} else
3653 			INP_WUNLOCK(inp);
3654 	} else
3655 		error = ESRCH;
3656 	NET_EPOCH_EXIT(et);
3657 	return (error);
3658 }
3659 
3660 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3661     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3662     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3663     "Drop TCP connection");
3664 
3665 static int
3666 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3667 {
3668 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3669 	    &tcp_ctloutput_set));
3670 }
3671 
3672 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3673     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3674     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3675     "Set socket option for TCP endpoint");
3676 
3677 #ifdef KERN_TLS
3678 static int
3679 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3680 {
3681 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3682 	struct sockaddr_storage addrs[2];
3683 	struct inpcb *inp;
3684 #ifdef INET
3685 	struct sockaddr_in *fin = NULL, *lin = NULL;
3686 #endif
3687 	struct epoch_tracker et;
3688 #ifdef INET6
3689 	struct sockaddr_in6 *fin6, *lin6;
3690 #endif
3691 	int error;
3692 
3693 	inp = NULL;
3694 #ifdef INET6
3695 	fin6 = lin6 = NULL;
3696 #endif
3697 	error = 0;
3698 
3699 	if (req->oldptr != NULL || req->oldlen != 0)
3700 		return (EINVAL);
3701 	if (req->newptr == NULL)
3702 		return (EPERM);
3703 	if (req->newlen < sizeof(addrs))
3704 		return (ENOMEM);
3705 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3706 	if (error)
3707 		return (error);
3708 
3709 	switch (addrs[0].ss_family) {
3710 #ifdef INET6
3711 	case AF_INET6:
3712 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3713 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3714 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3715 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3716 			return (EINVAL);
3717 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3718 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3719 				return (EINVAL);
3720 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3721 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3722 #ifdef INET
3723 			fin = (struct sockaddr_in *)&addrs[0];
3724 			lin = (struct sockaddr_in *)&addrs[1];
3725 #endif
3726 			break;
3727 		}
3728 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3729 		if (error)
3730 			return (error);
3731 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3732 		if (error)
3733 			return (error);
3734 		break;
3735 #endif
3736 #ifdef INET
3737 	case AF_INET:
3738 		fin = (struct sockaddr_in *)&addrs[0];
3739 		lin = (struct sockaddr_in *)&addrs[1];
3740 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3741 		    lin->sin_len != sizeof(struct sockaddr_in))
3742 			return (EINVAL);
3743 		break;
3744 #endif
3745 	default:
3746 		return (EINVAL);
3747 	}
3748 	NET_EPOCH_ENTER(et);
3749 	switch (addrs[0].ss_family) {
3750 #ifdef INET6
3751 	case AF_INET6:
3752 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3753 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3754 		    INPLOOKUP_WLOCKPCB, NULL);
3755 		break;
3756 #endif
3757 #ifdef INET
3758 	case AF_INET:
3759 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3760 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3761 		break;
3762 #endif
3763 	}
3764 	NET_EPOCH_EXIT(et);
3765 	if (inp != NULL) {
3766 		struct socket *so;
3767 
3768 		so = inp->inp_socket;
3769 		soref(so);
3770 		error = ktls_set_tx_mode(so,
3771 		    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3772 		INP_WUNLOCK(inp);
3773 		sorele(so);
3774 	} else
3775 		error = ESRCH;
3776 	return (error);
3777 }
3778 
3779 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3780     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3781     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3782     "Switch TCP connection to SW TLS");
3783 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3784     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3785     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3786     "Switch TCP connection to ifnet TLS");
3787 #endif
3788 
3789 /*
3790  * Generate a standardized TCP log line for use throughout the
3791  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3792  * allow use in the interrupt context.
3793  *
3794  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3795  * NB: The function may return NULL if memory allocation failed.
3796  *
3797  * Due to header inclusion and ordering limitations the struct ip
3798  * and ip6_hdr pointers have to be passed as void pointers.
3799  */
3800 char *
3801 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3802     const void *ip6hdr)
3803 {
3804 
3805 	/* Is logging enabled? */
3806 	if (V_tcp_log_in_vain == 0)
3807 		return (NULL);
3808 
3809 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3810 }
3811 
3812 char *
3813 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3814     const void *ip6hdr)
3815 {
3816 
3817 	/* Is logging enabled? */
3818 	if (tcp_log_debug == 0)
3819 		return (NULL);
3820 
3821 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3822 }
3823 
3824 static char *
3825 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3826     const void *ip6hdr)
3827 {
3828 	char *s, *sp;
3829 	size_t size;
3830 #ifdef INET
3831 	const struct ip *ip = (const struct ip *)ip4hdr;
3832 #endif
3833 #ifdef INET6
3834 	const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3835 #endif /* INET6 */
3836 
3837 	/*
3838 	 * The log line looks like this:
3839 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3840 	 */
3841 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3842 	    sizeof(PRINT_TH_FLAGS) + 1 +
3843 #ifdef INET6
3844 	    2 * INET6_ADDRSTRLEN;
3845 #else
3846 	    2 * INET_ADDRSTRLEN;
3847 #endif /* INET6 */
3848 
3849 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3850 	if (s == NULL)
3851 		return (NULL);
3852 
3853 	strcat(s, "TCP: [");
3854 	sp = s + strlen(s);
3855 
3856 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3857 		inet_ntoa_r(inc->inc_faddr, sp);
3858 		sp = s + strlen(s);
3859 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3860 		sp = s + strlen(s);
3861 		inet_ntoa_r(inc->inc_laddr, sp);
3862 		sp = s + strlen(s);
3863 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3864 #ifdef INET6
3865 	} else if (inc) {
3866 		ip6_sprintf(sp, &inc->inc6_faddr);
3867 		sp = s + strlen(s);
3868 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3869 		sp = s + strlen(s);
3870 		ip6_sprintf(sp, &inc->inc6_laddr);
3871 		sp = s + strlen(s);
3872 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3873 	} else if (ip6 && th) {
3874 		ip6_sprintf(sp, &ip6->ip6_src);
3875 		sp = s + strlen(s);
3876 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3877 		sp = s + strlen(s);
3878 		ip6_sprintf(sp, &ip6->ip6_dst);
3879 		sp = s + strlen(s);
3880 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3881 #endif /* INET6 */
3882 #ifdef INET
3883 	} else if (ip && th) {
3884 		inet_ntoa_r(ip->ip_src, sp);
3885 		sp = s + strlen(s);
3886 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3887 		sp = s + strlen(s);
3888 		inet_ntoa_r(ip->ip_dst, sp);
3889 		sp = s + strlen(s);
3890 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3891 #endif /* INET */
3892 	} else {
3893 		free(s, M_TCPLOG);
3894 		return (NULL);
3895 	}
3896 	sp = s + strlen(s);
3897 	if (th)
3898 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3899 	if (*(s + size - 1) != '\0')
3900 		panic("%s: string too long", __func__);
3901 	return (s);
3902 }
3903 
3904 /*
3905  * A subroutine which makes it easy to track TCP state changes with DTrace.
3906  * This function shouldn't be called for t_state initializations that don't
3907  * correspond to actual TCP state transitions.
3908  */
3909 void
3910 tcp_state_change(struct tcpcb *tp, int newstate)
3911 {
3912 #if defined(KDTRACE_HOOKS)
3913 	int pstate = tp->t_state;
3914 #endif
3915 
3916 	TCPSTATES_DEC(tp->t_state);
3917 	TCPSTATES_INC(newstate);
3918 	tp->t_state = newstate;
3919 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3920 }
3921 
3922 /*
3923  * Create an external-format (``xtcpcb'') structure using the information in
3924  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3925  * reduce the spew of irrelevant information over this interface, to isolate
3926  * user code from changes in the kernel structure, and potentially to provide
3927  * information-hiding if we decide that some of this information should be
3928  * hidden from users.
3929  */
3930 void
3931 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3932 {
3933 	struct tcpcb *tp = intotcpcb(inp);
3934 	sbintime_t now;
3935 
3936 	bzero(xt, sizeof(*xt));
3937 	xt->t_state = tp->t_state;
3938 	xt->t_logstate = tcp_get_bblog_state(tp);
3939 	xt->t_flags = tp->t_flags;
3940 	xt->t_sndzerowin = tp->t_sndzerowin;
3941 	xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3942 	xt->t_rcvoopack = tp->t_rcvoopack;
3943 	xt->t_rcv_wnd = tp->rcv_wnd;
3944 	xt->t_snd_wnd = tp->snd_wnd;
3945 	xt->t_snd_cwnd = tp->snd_cwnd;
3946 	xt->t_snd_ssthresh = tp->snd_ssthresh;
3947 	xt->t_dsack_bytes = tp->t_dsack_bytes;
3948 	xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3949 	xt->t_dsack_pack = tp->t_dsack_pack;
3950 	xt->t_maxseg = tp->t_maxseg;
3951 	xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3952 		     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3953 
3954 	now = getsbinuptime();
3955 #define	COPYTIMER(which,where)	do {					\
3956 	if (tp->t_timers[which] != SBT_MAX)				\
3957 		xt->where = (tp->t_timers[which] - now) / SBT_1MS;	\
3958 	else								\
3959 		xt->where = 0;						\
3960 } while (0)
3961 	COPYTIMER(TT_DELACK, tt_delack);
3962 	COPYTIMER(TT_REXMT, tt_rexmt);
3963 	COPYTIMER(TT_PERSIST, tt_persist);
3964 	COPYTIMER(TT_KEEP, tt_keep);
3965 	COPYTIMER(TT_2MSL, tt_2msl);
3966 #undef COPYTIMER
3967 	xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3968 
3969 	xt->xt_encaps_port = tp->t_port;
3970 	bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3971 	    TCP_FUNCTION_NAME_LEN_MAX);
3972 	bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3973 #ifdef TCP_BLACKBOX
3974 	(void)tcp_log_get_id(tp, xt->xt_logid);
3975 #endif
3976 
3977 	xt->xt_len = sizeof(struct xtcpcb);
3978 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3979 	/*
3980 	 * TCP doesn't use inp_ppcb pointer, we embed inpcb into tcpcb.
3981 	 * Fixup the pointer that in_pcbtoxinpcb() has set.  When printing
3982 	 * TCP netstat(1) used to use this pointer, so this fixup needs to
3983 	 * stay for stable/14.
3984 	 */
3985 	xt->xt_inp.inp_ppcb = (uintptr_t)tp;
3986 }
3987 
3988 void
3989 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3990 {
3991 	uint32_t bit, i;
3992 
3993 	if ((tp == NULL) ||
3994 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
3995 	    (status == 0)) {
3996 		/* Invalid */
3997 		return;
3998 	}
3999 	if (status > (sizeof(uint32_t) * 8)) {
4000 		/* Should this be a KASSERT? */
4001 		return;
4002 	}
4003 	bit = 1U << (status - 1);
4004 	if (bit & tp->t_end_info_status) {
4005 		/* already logged */
4006 		return;
4007 	}
4008 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4009 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4010 			tp->t_end_info_bytes[i] = status;
4011 			tp->t_end_info_status |= bit;
4012 			break;
4013 		}
4014 	}
4015 }
4016 
4017 int
4018 tcp_can_enable_pacing(void)
4019 {
4020 
4021 	if ((tcp_pacing_limit == -1) ||
4022 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4023 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4024 		shadow_num_connections = number_of_tcp_connections_pacing;
4025 		return (1);
4026 	} else {
4027 		counter_u64_add(tcp_pacing_failures, 1);
4028 		return (0);
4029 	}
4030 }
4031 
4032 static uint8_t tcp_pacing_warning = 0;
4033 
4034 void
4035 tcp_decrement_paced_conn(void)
4036 {
4037 	uint32_t ret;
4038 
4039 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4040 	shadow_num_connections = number_of_tcp_connections_pacing;
4041 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4042 	if (ret == 0) {
4043 		if (tcp_pacing_limit != -1) {
4044 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4045 			tcp_pacing_limit = 0;
4046 		} else if (tcp_pacing_warning == 0) {
4047 			printf("Warning pacing count is invalid, invalid decrement\n");
4048 			tcp_pacing_warning = 1;
4049 		}
4050 	}
4051 }
4052 
4053 static void
4054 tcp_default_switch_failed(struct tcpcb *tp)
4055 {
4056 	/*
4057 	 * If a switch fails we only need to
4058 	 * care about two things:
4059 	 * a) The t_flags2
4060 	 * and
4061 	 * b) The timer granularity.
4062 	 * Timeouts, at least for now, don't use the
4063 	 * old callout system in the other stacks so
4064 	 * those are hopefully safe.
4065 	 */
4066 	tcp_lro_features_off(tp);
4067 	tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
4068 }
4069 
4070 #ifdef TCP_ACCOUNTING
4071 int
4072 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss)
4073 {
4074 	if (SEQ_LT(th->th_ack, tp->snd_una)) {
4075 		/* Do we have a SACK? */
4076 		if (to->to_flags & TOF_SACK) {
4077 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4078 				tp->tcp_cnt_counters[ACK_SACK]++;
4079 			}
4080 			return (ACK_SACK);
4081 		} else {
4082 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4083 				tp->tcp_cnt_counters[ACK_BEHIND]++;
4084 			}
4085 			return (ACK_BEHIND);
4086 		}
4087 	} else if (th->th_ack == tp->snd_una) {
4088 		/* Do we have a SACK? */
4089 		if (to->to_flags & TOF_SACK) {
4090 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4091 				tp->tcp_cnt_counters[ACK_SACK]++;
4092 			}
4093 			return (ACK_SACK);
4094 		} else if (tiwin != tp->snd_wnd) {
4095 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4096 				tp->tcp_cnt_counters[ACK_RWND]++;
4097 			}
4098 			return (ACK_RWND);
4099 		} else {
4100 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4101 				tp->tcp_cnt_counters[ACK_DUPACK]++;
4102 			}
4103 			return (ACK_DUPACK);
4104 		}
4105 	} else {
4106 		if (!SEQ_GT(th->th_ack, tp->snd_max)) {
4107 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4108 				tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss);
4109 			}
4110 		}
4111 		if (to->to_flags & TOF_SACK) {
4112 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4113 				tp->tcp_cnt_counters[ACK_CUMACK_SACK]++;
4114 			}
4115 			return (ACK_CUMACK_SACK);
4116 		} else {
4117 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4118 				tp->tcp_cnt_counters[ACK_CUMACK]++;
4119 			}
4120 			return (ACK_CUMACK);
4121 		}
4122 	}
4123 }
4124 #endif
4125 
4126 void
4127 tcp_change_time_units(struct tcpcb *tp, int granularity)
4128 {
4129 	if (tp->t_tmr_granularity == granularity) {
4130 		/* We are there */
4131 		return;
4132 	}
4133 	if (granularity == TCP_TMR_GRANULARITY_USEC) {
4134 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS),
4135 			("Granularity is not TICKS its %u in tp:%p",
4136 			 tp->t_tmr_granularity, tp));
4137 		tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow);
4138 		if (tp->t_srtt > 1) {
4139 			uint32_t val, frac;
4140 
4141 			val = tp->t_srtt >> TCP_RTT_SHIFT;
4142 			frac = tp->t_srtt & 0x1f;
4143 			tp->t_srtt = TICKS_2_USEC(val);
4144 			/*
4145 			 * frac is the fractional part of the srtt (if any)
4146 			 * but its in ticks and every bit represents
4147 			 * 1/32nd of a hz.
4148 			 */
4149 			if (frac) {
4150 				if (hz == 1000) {
4151 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4152 				} else {
4153 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4154 				}
4155 				tp->t_srtt += frac;
4156 			}
4157 		}
4158 		if (tp->t_rttvar) {
4159 			uint32_t val, frac;
4160 
4161 			val = tp->t_rttvar >> TCP_RTTVAR_SHIFT;
4162 			frac = tp->t_rttvar & 0x1f;
4163 			tp->t_rttvar = TICKS_2_USEC(val);
4164 			/*
4165 			 * frac is the fractional part of the srtt (if any)
4166 			 * but its in ticks and every bit represents
4167 			 * 1/32nd of a hz.
4168 			 */
4169 			if (frac) {
4170 				if (hz == 1000) {
4171 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4172 				} else {
4173 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4174 				}
4175 				tp->t_rttvar += frac;
4176 			}
4177 		}
4178 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC;
4179 	} else if (granularity == TCP_TMR_GRANULARITY_TICKS) {
4180 		/* Convert back to ticks, with  */
4181 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC),
4182 			("Granularity is not USEC its %u in tp:%p",
4183 			 tp->t_tmr_granularity, tp));
4184 		if (tp->t_srtt > 1) {
4185 			uint32_t val, frac;
4186 
4187 			val = USEC_2_TICKS(tp->t_srtt);
4188 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4189 			tp->t_srtt = val << TCP_RTT_SHIFT;
4190 			/*
4191 			 * frac is the fractional part here is left
4192 			 * over from converting to hz and shifting.
4193 			 * We need to convert this to the 5 bit
4194 			 * remainder.
4195 			 */
4196 			if (frac) {
4197 				if (hz == 1000) {
4198 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4199 				} else {
4200 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4201 				}
4202 				tp->t_srtt += frac;
4203 			}
4204 		}
4205 		if (tp->t_rttvar) {
4206 			uint32_t val, frac;
4207 
4208 			val = USEC_2_TICKS(tp->t_rttvar);
4209 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4210 			tp->t_rttvar = val <<  TCP_RTTVAR_SHIFT;
4211 			/*
4212 			 * frac is the fractional part here is left
4213 			 * over from converting to hz and shifting.
4214 			 * We need to convert this to the 5 bit
4215 			 * remainder.
4216 			 */
4217 			if (frac) {
4218 				if (hz == 1000) {
4219 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4220 				} else {
4221 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4222 				}
4223 				tp->t_rttvar += frac;
4224 			}
4225 		}
4226 		tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow);
4227 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
4228 	}
4229 #ifdef INVARIANTS
4230 	else {
4231 		panic("Unknown granularity:%d tp:%p",
4232 		      granularity, tp);
4233 	}
4234 #endif
4235 }
4236 
4237 void
4238 tcp_handle_orphaned_packets(struct tcpcb *tp)
4239 {
4240 	struct mbuf *save, *m, *prev;
4241 	/*
4242 	 * Called when a stack switch is occuring from the fini()
4243 	 * of the old stack. We assue the init() as already been
4244 	 * run of the new stack and it has set the t_flags2 to
4245 	 * what it supports. This function will then deal with any
4246 	 * differences i.e. cleanup packets that maybe queued that
4247 	 * the newstack does not support.
4248 	 */
4249 
4250 	if (tp->t_flags2 & TF2_MBUF_L_ACKS)
4251 		return;
4252 	if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 &&
4253 	    !STAILQ_EMPTY(&tp->t_inqueue)) {
4254 		/*
4255 		 * It is unsafe to process the packets since a
4256 		 * reset may be lurking in them (its rare but it
4257 		 * can occur). If we were to find a RST, then we
4258 		 * would end up dropping the connection and the
4259 		 * INP lock, so when we return the caller (tcp_usrreq)
4260 		 * will blow up when it trys to unlock the inp.
4261 		 * This new stack does not do any fancy LRO features
4262 		 * so all we can do is toss the packets.
4263 		 */
4264 		m = STAILQ_FIRST(&tp->t_inqueue);
4265 		STAILQ_INIT(&tp->t_inqueue);
4266 		STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save)
4267 			m_freem(m);
4268 	} else {
4269 		/*
4270 		 * Here we have a stack that does mbuf queuing but
4271 		 * does not support compressed ack's. We must
4272 		 * walk all the mbufs and discard any compressed acks.
4273 		 */
4274 		STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) {
4275 			if (m->m_flags & M_ACKCMP) {
4276 				if (m == STAILQ_FIRST(&tp->t_inqueue))
4277 					STAILQ_REMOVE_HEAD(&tp->t_inqueue,
4278 					    m_stailqpkt);
4279 				else
4280 					STAILQ_REMOVE_AFTER(&tp->t_inqueue,
4281 					    prev, m_stailqpkt);
4282 				m_freem(m);
4283 			} else
4284 				prev = m;
4285 		}
4286 	}
4287 }
4288 
4289 #ifdef TCP_REQUEST_TRK
4290 uint32_t
4291 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes)
4292 {
4293 #ifdef KERN_TLS
4294 	struct ktls_session *tls;
4295 	uint32_t rec_oh, records;
4296 
4297 	tls = so->so_snd.sb_tls_info;
4298 	if (tls == NULL)
4299 	    return (0);
4300 
4301 	rec_oh = tls->params.tls_hlen + tls->params.tls_tlen;
4302 	records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len);
4303 	return (records * rec_oh);
4304 #else
4305 	return (0);
4306 #endif
4307 }
4308 
4309 extern uint32_t tcp_stale_entry_time;
4310 uint32_t tcp_stale_entry_time = 250000;
4311 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW,
4312     &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out");
4313 
4314 void
4315 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req,
4316     uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes)
4317 {
4318 	if (tcp_bblogging_on(tp)) {
4319 		union tcp_log_stackspecific log;
4320 		struct timeval tv;
4321 
4322 		memset(&log.u_bbr, 0, sizeof(log.u_bbr));
4323 #ifdef TCPHPTS
4324 		log.u_bbr.inhpts = tcp_in_hpts(tp);
4325 #endif
4326 		log.u_bbr.flex8 = val;
4327 		log.u_bbr.rttProp = req->timestamp;
4328 		log.u_bbr.delRate = req->start;
4329 		log.u_bbr.cur_del_rate = req->end;
4330 		log.u_bbr.flex1 = req->start_seq;
4331 		log.u_bbr.flex2 = req->end_seq;
4332 		log.u_bbr.flex3 = req->flags;
4333 		log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff);
4334 		log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff);
4335 		log.u_bbr.flex7 = slot;
4336 		log.u_bbr.bw_inuse = offset;
4337 		/* nbytes = flex6 | epoch */
4338 		log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff);
4339 		log.u_bbr.epoch = (nbytes & 0x00000000ffffffff);
4340 		/* cspr =  lt_epoch | pkts_out */
4341 		log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff);
4342 		log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff);
4343 		log.u_bbr.applimited = tp->t_tcpreq_closed;
4344 		log.u_bbr.applimited <<= 8;
4345 		log.u_bbr.applimited |= tp->t_tcpreq_open;
4346 		log.u_bbr.applimited <<= 8;
4347 		log.u_bbr.applimited |= tp->t_tcpreq_req;
4348 		log.u_bbr.timeStamp = tcp_get_usecs(&tv);
4349 		TCP_LOG_EVENTP(tp, NULL,
4350 		    &tptosocket(tp)->so_rcv,
4351 		    &tptosocket(tp)->so_snd,
4352 		    TCP_LOG_REQ_T, 0,
4353 		    0, &log, false, &tv);
4354 	}
4355 }
4356 
4357 void
4358 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent)
4359 {
4360 	if (tp->t_tcpreq_req > 0)
4361 		tp->t_tcpreq_req--;
4362 	if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4363 		if (tp->t_tcpreq_open > 0)
4364 			tp->t_tcpreq_open--;
4365 	} else {
4366 		if (tp->t_tcpreq_closed > 0)
4367 			tp->t_tcpreq_closed--;
4368 	}
4369 	ent->flags = TCP_TRK_TRACK_FLG_EMPTY;
4370 }
4371 
4372 static void
4373 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest)
4374 {
4375 	struct tcp_sendfile_track *ent;
4376 	uint64_t time_delta, oldest_delta;
4377 	int i, oldest, oldest_set = 0, cnt_rm = 0;
4378 
4379 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4380 		ent = &tp->t_tcpreq_info[i];
4381 		if (ent->flags != TCP_TRK_TRACK_FLG_USED) {
4382 			/*
4383 			 * We only care about closed end ranges
4384 			 * that are allocated and have no sendfile
4385 			 * ever touching them. They would be in
4386 			 * state USED.
4387 			 */
4388 			continue;
4389 		}
4390 		if (ts >= ent->localtime)
4391 			time_delta = ts - ent->localtime;
4392 		else
4393 			time_delta = 0;
4394 		if (time_delta &&
4395 		    ((oldest_delta < time_delta) || (oldest_set == 0))) {
4396 			oldest_set = 1;
4397 			oldest = i;
4398 			oldest_delta = time_delta;
4399 		}
4400 		if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) {
4401 			/*
4402 			 * No sendfile in a our time-limit
4403 			 * time to purge it.
4404 			 */
4405 			cnt_rm++;
4406 			tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4407 					      time_delta, 0);
4408 			tcp_req_free_a_slot(tp, ent);
4409 		}
4410 	}
4411 	if ((cnt_rm == 0) && rm_oldest && oldest_set) {
4412 		ent = &tp->t_tcpreq_info[oldest];
4413 		tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4414 				      oldest_delta, 1);
4415 		tcp_req_free_a_slot(tp, ent);
4416 	}
4417 }
4418 
4419 int
4420 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point)
4421 {
4422 	int i, ret=0;
4423 	struct tcp_sendfile_track *ent;
4424 
4425 	/* Clean up any old closed end requests that are now completed */
4426 	if (tp->t_tcpreq_req == 0)
4427 		return(0);
4428 	if (tp->t_tcpreq_closed == 0)
4429 		return(0);
4430 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4431 		ent = &tp->t_tcpreq_info[i];
4432 		/* Skip empty ones */
4433 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4434 			continue;
4435 		/* Skip open ones */
4436 		if (ent->flags & TCP_TRK_TRACK_FLG_OPEN)
4437 			continue;
4438 		if (SEQ_GEQ(ack_point, ent->end_seq)) {
4439 			/* We are past it -- free it */
4440 			tcp_req_log_req_info(tp, ent,
4441 					      i, TCP_TRK_REQ_LOG_FREED, 0, 0);
4442 			tcp_req_free_a_slot(tp, ent);
4443 			ret++;
4444 		}
4445 	}
4446 	return (ret);
4447 }
4448 
4449 int
4450 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point)
4451 {
4452 	if (tp->t_tcpreq_req == 0)
4453 		return(-1);
4454 	if (tp->t_tcpreq_closed == 0)
4455 		return(-1);
4456 	if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4457 		return(-1);
4458 	if (SEQ_GEQ(ack_point, ent->end_seq)) {
4459 		return (1);
4460 	}
4461 	return (0);
4462 }
4463 
4464 struct tcp_sendfile_track *
4465 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip)
4466 {
4467 	/*
4468 	 * Given an ack point (th_ack) walk through our entries and
4469 	 * return the first one found that th_ack goes past the
4470 	 * end_seq.
4471 	 */
4472 	struct tcp_sendfile_track *ent;
4473 	int i;
4474 
4475 	if (tp->t_tcpreq_req == 0) {
4476 		/* none open */
4477 		return (NULL);
4478 	}
4479 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4480 		ent = &tp->t_tcpreq_info[i];
4481 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4482 			continue;
4483 		if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) {
4484 			if (SEQ_GEQ(th_ack, ent->end_seq)) {
4485 				*ip = i;
4486 				return (ent);
4487 			}
4488 		}
4489 	}
4490 	return (NULL);
4491 }
4492 
4493 struct tcp_sendfile_track *
4494 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq)
4495 {
4496 	struct tcp_sendfile_track *ent;
4497 	int i;
4498 
4499 	if (tp->t_tcpreq_req == 0) {
4500 		/* none open */
4501 		return (NULL);
4502 	}
4503 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4504 		ent = &tp->t_tcpreq_info[i];
4505 		tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH,
4506 				      (uint64_t)seq, 0);
4507 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4508 			continue;
4509 		}
4510 		if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4511 			/*
4512 			 * An open end request only needs to
4513 			 * match the beginning seq or be
4514 			 * all we have (once we keep going on
4515 			 * a open end request we may have a seq
4516 			 * wrap).
4517 			 */
4518 			if ((SEQ_GEQ(seq, ent->start_seq)) ||
4519 			    (tp->t_tcpreq_closed == 0))
4520 				return (ent);
4521 		} else {
4522 			/*
4523 			 * For this one we need to
4524 			 * be a bit more careful if its
4525 			 * completed at least.
4526 			 */
4527 			if ((SEQ_GEQ(seq, ent->start_seq)) &&
4528 			    (SEQ_LT(seq, ent->end_seq))) {
4529 				return (ent);
4530 			}
4531 		}
4532 	}
4533 	return (NULL);
4534 }
4535 
4536 /* Should this be in its own file tcp_req.c ? */
4537 struct tcp_sendfile_track *
4538 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups)
4539 {
4540 	struct tcp_sendfile_track *fil;
4541 	int i, allocated;
4542 
4543 	/* In case the stack does not check for completions do so now */
4544 	tcp_req_check_for_comp(tp, tp->snd_una);
4545 	/* Check for stale entries */
4546 	if (tp->t_tcpreq_req)
4547 		tcp_req_check_for_stale_entries(tp, ts,
4548 		    (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ));
4549 	/* Check to see if this is a duplicate of one not started */
4550 	if (tp->t_tcpreq_req) {
4551 		for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4552 			fil = &tp->t_tcpreq_info[i];
4553 			if (fil->flags != TCP_TRK_TRACK_FLG_USED)
4554 				continue;
4555 			if ((fil->timestamp == req->timestamp) &&
4556 			    (fil->start == req->start) &&
4557 			    ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) ||
4558 			     (fil->end == req->end))) {
4559 				/*
4560 				 * We already have this request
4561 				 * and it has not been started with sendfile.
4562 				 * This probably means the user was returned
4563 				 * a 4xx of some sort and its going to age
4564 				 * out, lets not duplicate it.
4565 				 */
4566 				return(fil);
4567 			}
4568 		}
4569 	}
4570 	/* Ok if there is no room at the inn we are in trouble */
4571 	if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) {
4572 		tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL);
4573 		for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4574 			tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i],
4575 			    i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0);
4576 		}
4577 		return (NULL);
4578 	}
4579 	for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4580 		fil = &tp->t_tcpreq_info[i];
4581 		if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4582 			allocated = 1;
4583 			fil->flags = TCP_TRK_TRACK_FLG_USED;
4584 			fil->timestamp = req->timestamp;
4585 			fil->localtime = ts;
4586 			fil->start = req->start;
4587 			if (req->flags & TCP_LOG_HTTPD_RANGE_END) {
4588 				fil->end = req->end;
4589 			} else {
4590 				fil->end = 0;
4591 				fil->flags |= TCP_TRK_TRACK_FLG_OPEN;
4592 			}
4593 			/*
4594 			 * We can set the min boundaries to the TCP Sequence space,
4595 			 * but it might be found to be further up when sendfile
4596 			 * actually runs on this range (if it ever does).
4597 			 */
4598 			fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc;
4599 			fil->start_seq = tp->snd_una +
4600 			    tptosocket(tp)->so_snd.sb_ccc;
4601 			fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start)));
4602 			if (tptosocket(tp)->so_snd.sb_tls_info) {
4603 				/*
4604 				 * This session is doing TLS. Take a swag guess
4605 				 * at the overhead.
4606 				 */
4607 				fil->end_seq += tcp_estimate_tls_overhead(
4608 				    tptosocket(tp), (fil->end - fil->start));
4609 			}
4610 			tp->t_tcpreq_req++;
4611 			if (fil->flags & TCP_TRK_TRACK_FLG_OPEN)
4612 				tp->t_tcpreq_open++;
4613 			else
4614 				tp->t_tcpreq_closed++;
4615 			tcp_req_log_req_info(tp, fil, i,
4616 			    TCP_TRK_REQ_LOG_NEW, 0, 0);
4617 			break;
4618 		} else
4619 			fil = NULL;
4620 	}
4621 	return (fil);
4622 }
4623 
4624 void
4625 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts)
4626 {
4627 	(void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1);
4628 }
4629 #endif
4630 
4631 void
4632 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err)
4633 {
4634 	if (tcp_bblogging_on(tp)) {
4635 		struct tcp_log_buffer *l;
4636 
4637 		l = tcp_log_event(tp, NULL,
4638 		        &tptosocket(tp)->so_rcv,
4639 		        &tptosocket(tp)->so_snd,
4640 		        TCP_LOG_SOCKET_OPT,
4641 		        err, 0, NULL, 1,
4642 		        NULL, NULL, 0, NULL);
4643 		if (l) {
4644 			l->tlb_flex1 = option_num;
4645 			l->tlb_flex2 = option_val;
4646 		}
4647 	}
4648 }
4649 
4650 uint32_t
4651 tcp_get_srtt(struct tcpcb *tp, int granularity)
4652 {
4653 	uint32_t srtt;
4654 
4655 	KASSERT(granularity == TCP_TMR_GRANULARITY_USEC ||
4656 	    granularity == TCP_TMR_GRANULARITY_TICKS,
4657 	    ("%s: called with unexpected granularity %d", __func__,
4658 	    granularity));
4659 
4660 	srtt = tp->t_srtt;
4661 
4662 	/*
4663 	 * We only support two granularities. If the stored granularity
4664 	 * does not match the granularity requested by the caller,
4665 	 * convert the stored value to the requested unit of granularity.
4666 	 */
4667 	if (tp->t_tmr_granularity != granularity) {
4668 		if (granularity == TCP_TMR_GRANULARITY_USEC)
4669 			srtt = TICKS_2_USEC(srtt);
4670 		else
4671 			srtt = USEC_2_TICKS(srtt);
4672 	}
4673 
4674 	/*
4675 	 * If the srtt is stored with ticks granularity, we need to
4676 	 * unshift to get the actual value. We do this after the
4677 	 * conversion above (if one was necessary) in order to maximize
4678 	 * precision.
4679 	 */
4680 	if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS)
4681 		srtt = srtt >> TCP_RTT_SHIFT;
4682 
4683 	return (srtt);
4684 }
4685