xref: /freebsd/sys/netinet/tcp_subr.c (revision b07549f3c15637af6064d2f52b572f239f5ba870)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/arb.h>
45 #include <sys/callout.h>
46 #include <sys/eventhandler.h>
47 #ifdef TCP_HHOOK
48 #include <sys/hhook.h>
49 #endif
50 #include <sys/kernel.h>
51 #ifdef TCP_HHOOK
52 #include <sys/khelp.h>
53 #endif
54 #ifdef KERN_TLS
55 #include <sys/ktls.h>
56 #endif
57 #include <sys/qmath.h>
58 #include <sys/stats.h>
59 #include <sys/sysctl.h>
60 #include <sys/jail.h>
61 #include <sys/malloc.h>
62 #include <sys/refcount.h>
63 #include <sys/mbuf.h>
64 #include <sys/priv.h>
65 #include <sys/proc.h>
66 #include <sys/sdt.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/protosw.h>
70 #include <sys/random.h>
71 
72 #include <vm/uma.h>
73 
74 #include <net/route.h>
75 #include <net/route/nhop.h>
76 #include <net/if.h>
77 #include <net/if_var.h>
78 #include <net/if_private.h>
79 #include <net/vnet.h>
80 
81 #include <netinet/in.h>
82 #include <netinet/in_fib.h>
83 #include <netinet/in_kdtrace.h>
84 #include <netinet/in_pcb.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/in_var.h>
87 #include <netinet/ip.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/ip_var.h>
90 #ifdef INET6
91 #include <netinet/icmp6.h>
92 #include <netinet/ip6.h>
93 #include <netinet6/in6_fib.h>
94 #include <netinet6/in6_pcb.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet6/nd6.h>
98 #endif
99 
100 #include <netinet/tcp.h>
101 #ifdef INVARIANTS
102 #define TCPSTATES
103 #endif
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet/tcp_ecn.h>
109 #include <netinet/tcp_log_buf.h>
110 #include <netinet/tcp_syncache.h>
111 #include <netinet/tcp_hpts.h>
112 #include <netinet/tcp_lro.h>
113 #include <netinet/cc/cc.h>
114 #include <netinet/tcpip.h>
115 #include <netinet/tcp_fastopen.h>
116 #include <netinet/tcp_accounting.h>
117 #ifdef TCPPCAP
118 #include <netinet/tcp_pcap.h>
119 #endif
120 #ifdef TCP_OFFLOAD
121 #include <netinet/tcp_offload.h>
122 #endif
123 #include <netinet/udp.h>
124 #include <netinet/udp_var.h>
125 #ifdef INET6
126 #include <netinet6/tcp6_var.h>
127 #endif
128 
129 #include <netipsec/ipsec_support.h>
130 
131 #include <machine/in_cksum.h>
132 #include <crypto/siphash/siphash.h>
133 
134 #include <security/mac/mac_framework.h>
135 
136 #ifdef INET6
137 static ip6proto_ctlinput_t tcp6_ctlinput;
138 static udp_tun_icmp_t tcp6_ctlinput_viaudp;
139 #endif
140 
141 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
142 #ifdef INET6
143 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
144 #endif
145 
146 #ifdef TCP_SAD_DETECTION
147 /*  Sack attack detection thresholds and such */
148 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
149     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
150     "Sack Attack detection thresholds");
151 int32_t tcp_force_detection = 0;
152 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
153     CTLFLAG_RW,
154     &tcp_force_detection, 0,
155     "Do we force detection even if the INP has it off?");
156 int32_t tcp_sad_limit = 10000;
157 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, limit,
158     CTLFLAG_RW,
159     &tcp_sad_limit, 10000,
160     "If SaD is enabled, what is the limit to sendmap entries (0 = unlimited)?");
161 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
162 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
163     CTLFLAG_RW,
164     &tcp_sack_to_ack_thresh, 700,
165     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
166 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
167 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
168     CTLFLAG_RW,
169     &tcp_sack_to_move_thresh, 600,
170     "Percentage of sack moves we must see above (10.1 percent is 101)");
171 int32_t tcp_restoral_thresh = 450;	/* 45 % (sack:2:ack -25%) (mv:ratio -15%) **/
172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
173     CTLFLAG_RW,
174     &tcp_restoral_thresh, 450,
175     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
176 int32_t tcp_sad_decay_val = 800;
177 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
178     CTLFLAG_RW,
179     &tcp_sad_decay_val, 800,
180     "The decay percentage (10.1 percent equals 101 )");
181 int32_t tcp_map_minimum = 500;
182 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
183     CTLFLAG_RW,
184     &tcp_map_minimum, 500,
185     "Number of Map enteries before we start detection");
186 int32_t tcp_sad_pacing_interval = 2000;
187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
188     CTLFLAG_RW,
189     &tcp_sad_pacing_interval, 2000,
190     "What is the minimum pacing interval for a classified attacker?");
191 
192 int32_t tcp_sad_low_pps = 100;
193 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
194     CTLFLAG_RW,
195     &tcp_sad_low_pps, 100,
196     "What is the input pps that below which we do not decay?");
197 #endif
198 uint32_t tcp_ack_war_time_window = 1000;
199 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
200     CTLFLAG_RW,
201     &tcp_ack_war_time_window, 1000,
202    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
203 uint32_t tcp_ack_war_cnt = 5;
204 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
205     CTLFLAG_RW,
206     &tcp_ack_war_cnt, 5,
207    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
208 
209 struct rwlock tcp_function_lock;
210 
211 static int
212 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
213 {
214 	int error, new;
215 
216 	new = V_tcp_mssdflt;
217 	error = sysctl_handle_int(oidp, &new, 0, req);
218 	if (error == 0 && req->newptr) {
219 		if (new < TCP_MINMSS)
220 			error = EINVAL;
221 		else
222 			V_tcp_mssdflt = new;
223 	}
224 	return (error);
225 }
226 
227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
228     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
229     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
230     "Default TCP Maximum Segment Size");
231 
232 #ifdef INET6
233 static int
234 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
235 {
236 	int error, new;
237 
238 	new = V_tcp_v6mssdflt;
239 	error = sysctl_handle_int(oidp, &new, 0, req);
240 	if (error == 0 && req->newptr) {
241 		if (new < TCP_MINMSS)
242 			error = EINVAL;
243 		else
244 			V_tcp_v6mssdflt = new;
245 	}
246 	return (error);
247 }
248 
249 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
250     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
251     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
252    "Default TCP Maximum Segment Size for IPv6");
253 #endif /* INET6 */
254 
255 /*
256  * Minimum MSS we accept and use. This prevents DoS attacks where
257  * we are forced to a ridiculous low MSS like 20 and send hundreds
258  * of packets instead of one. The effect scales with the available
259  * bandwidth and quickly saturates the CPU and network interface
260  * with packet generation and sending. Set to zero to disable MINMSS
261  * checking. This setting prevents us from sending too small packets.
262  */
263 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
265      &VNET_NAME(tcp_minmss), 0,
266     "Minimum TCP Maximum Segment Size");
267 
268 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
269 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
270     &VNET_NAME(tcp_do_rfc1323), 0,
271     "Enable rfc1323 (high performance TCP) extensions");
272 
273 /*
274  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
275  * Some stacks negotiate TS, but never send them after connection setup. Some
276  * stacks negotiate TS, but don't send them when sending keep-alive segments.
277  * These include modern widely deployed TCP stacks.
278  * Therefore tolerating violations for now...
279  */
280 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
282     &VNET_NAME(tcp_tolerate_missing_ts), 0,
283     "Tolerate missing TCP timestamps");
284 
285 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
287     &VNET_NAME(tcp_ts_offset_per_conn), 0,
288     "Initialize TCP timestamps per connection instead of per host pair");
289 
290 /* How many connections are pacing */
291 static volatile uint32_t number_of_tcp_connections_pacing = 0;
292 static uint32_t shadow_num_connections = 0;
293 static counter_u64_t tcp_pacing_failures;
294 
295 static int tcp_pacing_limit = 10000;
296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
297     &tcp_pacing_limit, 1000,
298     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
299 
300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
301     &shadow_num_connections, 0, "Number of TCP connections being paced");
302 
303 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD,
304     &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit");
305 
306 static int	tcp_log_debug = 0;
307 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
308     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
309 
310 static int	tcp_tcbhashsize;
311 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
312     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
313 
314 static int	do_tcpdrain = 1;
315 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
316     "Enable tcp_drain routine for extra help when low on mbufs");
317 
318 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
319     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
320 
321 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
322 #define	V_icmp_may_rst			VNET(icmp_may_rst)
323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
324     &VNET_NAME(icmp_may_rst), 0,
325     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
326 
327 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
328 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
329 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
330     &VNET_NAME(tcp_isn_reseed_interval), 0,
331     "Seconds between reseeding of ISN secret");
332 
333 static int	tcp_soreceive_stream;
334 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
335     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
336 
337 VNET_DEFINE(uma_zone_t, sack_hole_zone);
338 #define	V_sack_hole_zone		VNET(sack_hole_zone)
339 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
340 static int
341 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
342 {
343 	int error;
344 	uint32_t new;
345 
346 	new = V_tcp_map_entries_limit;
347 	error = sysctl_handle_int(oidp, &new, 0, req);
348 	if (error == 0 && req->newptr) {
349 		/* only allow "0" and value > minimum */
350 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
351 			error = EINVAL;
352 		else
353 			V_tcp_map_entries_limit = new;
354 	}
355 	return (error);
356 }
357 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
358     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
359     &VNET_NAME(tcp_map_entries_limit), 0,
360     &sysctl_net_inet_tcp_map_limit_check, "IU",
361     "Total sendmap entries limit");
362 
363 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
364 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
365      &VNET_NAME(tcp_map_split_limit), 0,
366     "Total sendmap split entries limit");
367 
368 #ifdef TCP_HHOOK
369 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
370 #endif
371 
372 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
373 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
374 #define	V_ts_offset_secret	VNET(ts_offset_secret)
375 
376 static int	tcp_default_fb_init(struct tcpcb *tp, void **ptr);
377 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
378 static int	tcp_default_handoff_ok(struct tcpcb *tp);
379 static struct inpcb *tcp_notify(struct inpcb *, int);
380 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
381 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
382 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
383 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
384 		    const void *ip4hdr, const void *ip6hdr);
385 static void	tcp_default_switch_failed(struct tcpcb *tp);
386 static ipproto_ctlinput_t	tcp_ctlinput;
387 static udp_tun_icmp_t		tcp_ctlinput_viaudp;
388 
389 static struct tcp_function_block tcp_def_funcblk = {
390 	.tfb_tcp_block_name = "freebsd",
391 	.tfb_tcp_output = tcp_default_output,
392 	.tfb_tcp_do_segment = tcp_do_segment,
393 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
394 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
395 	.tfb_tcp_fb_init = tcp_default_fb_init,
396 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
397 	.tfb_switch_failed = tcp_default_switch_failed,
398 };
399 
400 static int tcp_fb_cnt = 0;
401 struct tcp_funchead t_functions;
402 VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk;
403 #define	V_tcp_func_set_ptr VNET(tcp_func_set_ptr)
404 
405 void
406 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
407 {
408 	TCPSTAT_INC(tcps_dsack_count);
409 	tp->t_dsack_pack++;
410 	if (tlp == 0) {
411 		if (SEQ_GT(end, start)) {
412 			tp->t_dsack_bytes += (end - start);
413 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
414 		} else {
415 			tp->t_dsack_tlp_bytes += (start - end);
416 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
417 		}
418 	} else {
419 		if (SEQ_GT(end, start)) {
420 			tp->t_dsack_bytes += (end - start);
421 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
422 		} else {
423 			tp->t_dsack_tlp_bytes += (start - end);
424 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
425 		}
426 	}
427 }
428 
429 static struct tcp_function_block *
430 find_tcp_functions_locked(struct tcp_function_set *fs)
431 {
432 	struct tcp_function *f;
433 	struct tcp_function_block *blk=NULL;
434 
435 	TAILQ_FOREACH(f, &t_functions, tf_next) {
436 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
437 			blk = f->tf_fb;
438 			break;
439 		}
440 	}
441 	return(blk);
442 }
443 
444 static struct tcp_function_block *
445 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
446 {
447 	struct tcp_function_block *rblk=NULL;
448 	struct tcp_function *f;
449 
450 	TAILQ_FOREACH(f, &t_functions, tf_next) {
451 		if (f->tf_fb == blk) {
452 			rblk = blk;
453 			if (s) {
454 				*s = f;
455 			}
456 			break;
457 		}
458 	}
459 	return (rblk);
460 }
461 
462 struct tcp_function_block *
463 find_and_ref_tcp_functions(struct tcp_function_set *fs)
464 {
465 	struct tcp_function_block *blk;
466 
467 	rw_rlock(&tcp_function_lock);
468 	blk = find_tcp_functions_locked(fs);
469 	if (blk)
470 		refcount_acquire(&blk->tfb_refcnt);
471 	rw_runlock(&tcp_function_lock);
472 	return(blk);
473 }
474 
475 struct tcp_function_block *
476 find_and_ref_tcp_fb(struct tcp_function_block *blk)
477 {
478 	struct tcp_function_block *rblk;
479 
480 	rw_rlock(&tcp_function_lock);
481 	rblk = find_tcp_fb_locked(blk, NULL);
482 	if (rblk)
483 		refcount_acquire(&rblk->tfb_refcnt);
484 	rw_runlock(&tcp_function_lock);
485 	return(rblk);
486 }
487 
488 /* Find a matching alias for the given tcp_function_block. */
489 int
490 find_tcp_function_alias(struct tcp_function_block *blk,
491     struct tcp_function_set *fs)
492 {
493 	struct tcp_function *f;
494 	int found;
495 
496 	found = 0;
497 	rw_rlock(&tcp_function_lock);
498 	TAILQ_FOREACH(f, &t_functions, tf_next) {
499 		if ((f->tf_fb == blk) &&
500 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
501 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
502 			/* Matching function block with different name. */
503 			strncpy(fs->function_set_name, f->tf_name,
504 			    TCP_FUNCTION_NAME_LEN_MAX);
505 			found = 1;
506 			break;
507 		}
508 	}
509 	/* Null terminate the string appropriately. */
510 	if (found) {
511 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
512 	} else {
513 		fs->function_set_name[0] = '\0';
514 	}
515 	rw_runlock(&tcp_function_lock);
516 	return (found);
517 }
518 
519 static struct tcp_function_block *
520 find_and_ref_tcp_default_fb(void)
521 {
522 	struct tcp_function_block *rblk;
523 
524 	rw_rlock(&tcp_function_lock);
525 	rblk = V_tcp_func_set_ptr;
526 	refcount_acquire(&rblk->tfb_refcnt);
527 	rw_runlock(&tcp_function_lock);
528 	return (rblk);
529 }
530 
531 void
532 tcp_switch_back_to_default(struct tcpcb *tp)
533 {
534 	struct tcp_function_block *tfb;
535 	void *ptr = NULL;
536 
537 	KASSERT(tp->t_fb != &tcp_def_funcblk,
538 	    ("%s: called by the built-in default stack", __func__));
539 
540 	/*
541 	 * Now, we'll find a new function block to use.
542 	 * Start by trying the current user-selected
543 	 * default, unless this stack is the user-selected
544 	 * default.
545 	 */
546 	tfb = find_and_ref_tcp_default_fb();
547 	if (tfb == tp->t_fb) {
548 		refcount_release(&tfb->tfb_refcnt);
549 		tfb = NULL;
550 	}
551 	/* Does the stack accept this connection? */
552 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
553 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
554 		refcount_release(&tfb->tfb_refcnt);
555 		tfb = NULL;
556 	}
557 	/* Try to use that stack. */
558 	if (tfb != NULL) {
559 		/* Initialize the new stack. If it succeeds, we are done. */
560 		if (tfb->tfb_tcp_fb_init == NULL ||
561 		    (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) {
562 			/* Release the old stack */
563 			if (tp->t_fb->tfb_tcp_fb_fini != NULL)
564 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
565 			refcount_release(&tp->t_fb->tfb_refcnt);
566 			/* Now set in all the pointers */
567 			tp->t_fb = tfb;
568 			tp->t_fb_ptr = ptr;
569 			return;
570 		}
571 		/*
572 		 * Initialization failed. Release the reference count on
573 		 * the looked up default stack.
574 		 */
575 		refcount_release(&tfb->tfb_refcnt);
576 	}
577 
578 	/*
579 	 * If that wasn't feasible, use the built-in default
580 	 * stack which is not allowed to reject anyone.
581 	 */
582 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
583 	if (tfb == NULL) {
584 		/* there always should be a default */
585 		panic("Can't refer to tcp_def_funcblk");
586 	}
587 	if (tfb->tfb_tcp_handoff_ok != NULL) {
588 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
589 			/* The default stack cannot say no */
590 			panic("Default stack rejects a new session?");
591 		}
592 	}
593 	if (tfb->tfb_tcp_fb_init != NULL &&
594 	    (*tfb->tfb_tcp_fb_init)(tp, &ptr)) {
595 		/* The default stack cannot fail */
596 		panic("Default stack initialization failed");
597 	}
598 	/* Now release the old stack */
599 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
600 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
601 	refcount_release(&tp->t_fb->tfb_refcnt);
602 	/* And set in the pointers to the new */
603 	tp->t_fb = tfb;
604 	tp->t_fb_ptr = ptr;
605 }
606 
607 static bool
608 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
609     const struct sockaddr *sa, void *ctx)
610 {
611 	struct ip *iph;
612 #ifdef INET6
613 	struct ip6_hdr *ip6;
614 #endif
615 	struct udphdr *uh;
616 	struct tcphdr *th;
617 	int thlen;
618 	uint16_t port;
619 
620 	TCPSTAT_INC(tcps_tunneled_pkts);
621 	if ((m->m_flags & M_PKTHDR) == 0) {
622 		/* Can't handle one that is not a pkt hdr */
623 		TCPSTAT_INC(tcps_tunneled_errs);
624 		goto out;
625 	}
626 	thlen = sizeof(struct tcphdr);
627 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
628 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
629 		TCPSTAT_INC(tcps_tunneled_errs);
630 		goto out;
631 	}
632 	iph = mtod(m, struct ip *);
633 	uh = (struct udphdr *)((caddr_t)iph + off);
634 	th = (struct tcphdr *)(uh + 1);
635 	thlen = th->th_off << 2;
636 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
637 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
638 		if (m == NULL) {
639 			TCPSTAT_INC(tcps_tunneled_errs);
640 			goto out;
641 		} else {
642 			iph = mtod(m, struct ip *);
643 			uh = (struct udphdr *)((caddr_t)iph + off);
644 			th = (struct tcphdr *)(uh + 1);
645 		}
646 	}
647 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
648 	bcopy(th, uh, m->m_len - off);
649 	m->m_len -= sizeof(struct udphdr);
650 	m->m_pkthdr.len -= sizeof(struct udphdr);
651 	/*
652 	 * We use the same algorithm for
653 	 * both UDP and TCP for c-sum. So
654 	 * the code in tcp_input will skip
655 	 * the checksum. So we do nothing
656 	 * with the flag (m->m_pkthdr.csum_flags).
657 	 */
658 	switch (iph->ip_v) {
659 #ifdef INET
660 	case IPVERSION:
661 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
662 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
663 		break;
664 #endif
665 #ifdef INET6
666 	case IPV6_VERSION >> 4:
667 		ip6 = mtod(m, struct ip6_hdr *);
668 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
669 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
670 		break;
671 #endif
672 	default:
673 		goto out;
674 		break;
675 	}
676 	return (true);
677 out:
678 	m_freem(m);
679 
680 	return (true);
681 }
682 
683 static int
684 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
685 {
686 	int error=ENOENT;
687 	struct tcp_function_set fs;
688 	struct tcp_function_block *blk;
689 
690 	memset(&fs, 0, sizeof(fs));
691 	rw_rlock(&tcp_function_lock);
692 	blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL);
693 	if (blk) {
694 		/* Found him */
695 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
696 		fs.pcbcnt = blk->tfb_refcnt;
697 	}
698 	rw_runlock(&tcp_function_lock);
699 	error = sysctl_handle_string(oidp, fs.function_set_name,
700 				     sizeof(fs.function_set_name), req);
701 
702 	/* Check for error or no change */
703 	if (error != 0 || req->newptr == NULL)
704 		return(error);
705 
706 	rw_wlock(&tcp_function_lock);
707 	blk = find_tcp_functions_locked(&fs);
708 	if ((blk == NULL) ||
709 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
710 		error = ENOENT;
711 		goto done;
712 	}
713 	V_tcp_func_set_ptr = blk;
714 done:
715 	rw_wunlock(&tcp_function_lock);
716 	return (error);
717 }
718 
719 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
720     CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
721     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
722     "Set/get the default TCP functions");
723 
724 static int
725 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
726 {
727 	int error, cnt, linesz;
728 	struct tcp_function *f;
729 	char *buffer, *cp;
730 	size_t bufsz, outsz;
731 	bool alias;
732 
733 	cnt = 0;
734 	rw_rlock(&tcp_function_lock);
735 	TAILQ_FOREACH(f, &t_functions, tf_next) {
736 		cnt++;
737 	}
738 	rw_runlock(&tcp_function_lock);
739 
740 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
741 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
742 
743 	error = 0;
744 	cp = buffer;
745 
746 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
747 	    "Alias", "PCB count");
748 	cp += linesz;
749 	bufsz -= linesz;
750 	outsz = linesz;
751 
752 	rw_rlock(&tcp_function_lock);
753 	TAILQ_FOREACH(f, &t_functions, tf_next) {
754 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
755 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
756 		    f->tf_fb->tfb_tcp_block_name,
757 		    (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ',
758 		    alias ? f->tf_name : "-",
759 		    f->tf_fb->tfb_refcnt);
760 		if (linesz >= bufsz) {
761 			error = EOVERFLOW;
762 			break;
763 		}
764 		cp += linesz;
765 		bufsz -= linesz;
766 		outsz += linesz;
767 	}
768 	rw_runlock(&tcp_function_lock);
769 	if (error == 0)
770 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
771 	free(buffer, M_TEMP);
772 	return (error);
773 }
774 
775 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
776     CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
777     NULL, 0, sysctl_net_inet_list_available, "A",
778     "list available TCP Function sets");
779 
780 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
781 
782 #ifdef INET
783 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
784 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
785 #endif
786 #ifdef INET6
787 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
788 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
789 #endif
790 
791 static struct sx tcpoudp_lock;
792 
793 static void
794 tcp_over_udp_stop(void)
795 {
796 
797 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
798 
799 #ifdef INET
800 	if (V_udp4_tun_socket != NULL) {
801 		soclose(V_udp4_tun_socket);
802 		V_udp4_tun_socket = NULL;
803 	}
804 #endif
805 #ifdef INET6
806 	if (V_udp6_tun_socket != NULL) {
807 		soclose(V_udp6_tun_socket);
808 		V_udp6_tun_socket = NULL;
809 	}
810 #endif
811 }
812 
813 static int
814 tcp_over_udp_start(void)
815 {
816 	uint16_t port;
817 	int ret;
818 #ifdef INET
819 	struct sockaddr_in sin;
820 #endif
821 #ifdef INET6
822 	struct sockaddr_in6 sin6;
823 #endif
824 
825 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
826 
827 	port = V_tcp_udp_tunneling_port;
828 	if (ntohs(port) == 0) {
829 		/* Must have a port set */
830 		return (EINVAL);
831 	}
832 #ifdef INET
833 	if (V_udp4_tun_socket != NULL) {
834 		/* Already running -- must stop first */
835 		return (EALREADY);
836 	}
837 #endif
838 #ifdef INET6
839 	if (V_udp6_tun_socket != NULL) {
840 		/* Already running -- must stop first */
841 		return (EALREADY);
842 	}
843 #endif
844 #ifdef INET
845 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
846 	    SOCK_DGRAM, IPPROTO_UDP,
847 	    curthread->td_ucred, curthread))) {
848 		tcp_over_udp_stop();
849 		return (ret);
850 	}
851 	/* Call the special UDP hook. */
852 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
853 	    tcp_recv_udp_tunneled_packet,
854 	    tcp_ctlinput_viaudp,
855 	    NULL))) {
856 		tcp_over_udp_stop();
857 		return (ret);
858 	}
859 	/* Ok, we have a socket, bind it to the port. */
860 	memset(&sin, 0, sizeof(struct sockaddr_in));
861 	sin.sin_len = sizeof(struct sockaddr_in);
862 	sin.sin_family = AF_INET;
863 	sin.sin_port = htons(port);
864 	if ((ret = sobind(V_udp4_tun_socket,
865 	    (struct sockaddr *)&sin, curthread))) {
866 		tcp_over_udp_stop();
867 		return (ret);
868 	}
869 #endif
870 #ifdef INET6
871 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
872 	    SOCK_DGRAM, IPPROTO_UDP,
873 	    curthread->td_ucred, curthread))) {
874 		tcp_over_udp_stop();
875 		return (ret);
876 	}
877 	/* Call the special UDP hook. */
878 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
879 	    tcp_recv_udp_tunneled_packet,
880 	    tcp6_ctlinput_viaudp,
881 	    NULL))) {
882 		tcp_over_udp_stop();
883 		return (ret);
884 	}
885 	/* Ok, we have a socket, bind it to the port. */
886 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
887 	sin6.sin6_len = sizeof(struct sockaddr_in6);
888 	sin6.sin6_family = AF_INET6;
889 	sin6.sin6_port = htons(port);
890 	if ((ret = sobind(V_udp6_tun_socket,
891 	    (struct sockaddr *)&sin6, curthread))) {
892 		tcp_over_udp_stop();
893 		return (ret);
894 	}
895 #endif
896 	return (0);
897 }
898 
899 static int
900 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
901 {
902 	int error;
903 	uint32_t old, new;
904 
905 	old = V_tcp_udp_tunneling_port;
906 	new = old;
907 	error = sysctl_handle_int(oidp, &new, 0, req);
908 	if ((error == 0) &&
909 	    (req->newptr != NULL)) {
910 		if ((new < TCP_TUNNELING_PORT_MIN) ||
911 		    (new > TCP_TUNNELING_PORT_MAX)) {
912 			error = EINVAL;
913 		} else {
914 			sx_xlock(&tcpoudp_lock);
915 			V_tcp_udp_tunneling_port = new;
916 			if (old != 0) {
917 				tcp_over_udp_stop();
918 			}
919 			if (new != 0) {
920 				error = tcp_over_udp_start();
921 				if (error != 0) {
922 					V_tcp_udp_tunneling_port = 0;
923 				}
924 			}
925 			sx_xunlock(&tcpoudp_lock);
926 		}
927 	}
928 	return (error);
929 }
930 
931 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
932     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
933     &VNET_NAME(tcp_udp_tunneling_port),
934     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
935     "Tunneling port for tcp over udp");
936 
937 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
938 
939 static int
940 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
941 {
942 	int error, new;
943 
944 	new = V_tcp_udp_tunneling_overhead;
945 	error = sysctl_handle_int(oidp, &new, 0, req);
946 	if (error == 0 && req->newptr) {
947 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
948 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
949 			error = EINVAL;
950 		else
951 			V_tcp_udp_tunneling_overhead = new;
952 	}
953 	return (error);
954 }
955 
956 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
957     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
958     &VNET_NAME(tcp_udp_tunneling_overhead),
959     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
960     "MSS reduction when using tcp over udp");
961 
962 /*
963  * Exports one (struct tcp_function_info) for each alias/name.
964  */
965 static int
966 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
967 {
968 	int cnt, error;
969 	struct tcp_function *f;
970 	struct tcp_function_info tfi;
971 
972 	/*
973 	 * We don't allow writes.
974 	 */
975 	if (req->newptr != NULL)
976 		return (EINVAL);
977 
978 	/*
979 	 * Wire the old buffer so we can directly copy the functions to
980 	 * user space without dropping the lock.
981 	 */
982 	if (req->oldptr != NULL) {
983 		error = sysctl_wire_old_buffer(req, 0);
984 		if (error)
985 			return (error);
986 	}
987 
988 	/*
989 	 * Walk the list and copy out matching entries. If INVARIANTS
990 	 * is compiled in, also walk the list to verify the length of
991 	 * the list matches what we have recorded.
992 	 */
993 	rw_rlock(&tcp_function_lock);
994 
995 	cnt = 0;
996 #ifndef INVARIANTS
997 	if (req->oldptr == NULL) {
998 		cnt = tcp_fb_cnt;
999 		goto skip_loop;
1000 	}
1001 #endif
1002 	TAILQ_FOREACH(f, &t_functions, tf_next) {
1003 #ifdef INVARIANTS
1004 		cnt++;
1005 #endif
1006 		if (req->oldptr != NULL) {
1007 			bzero(&tfi, sizeof(tfi));
1008 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
1009 			tfi.tfi_id = f->tf_fb->tfb_id;
1010 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
1011 			    sizeof(tfi.tfi_alias));
1012 			(void)strlcpy(tfi.tfi_name,
1013 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
1014 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
1015 			/*
1016 			 * Don't stop on error, as that is the
1017 			 * mechanism we use to accumulate length
1018 			 * information if the buffer was too short.
1019 			 */
1020 		}
1021 	}
1022 	KASSERT(cnt == tcp_fb_cnt,
1023 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1024 #ifndef INVARIANTS
1025 skip_loop:
1026 #endif
1027 	rw_runlock(&tcp_function_lock);
1028 	if (req->oldptr == NULL)
1029 		error = SYSCTL_OUT(req, NULL,
1030 		    (cnt + 1) * sizeof(struct tcp_function_info));
1031 
1032 	return (error);
1033 }
1034 
1035 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1036 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1037 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1038 	    "List TCP function block name-to-ID mappings");
1039 
1040 /*
1041  * tfb_tcp_handoff_ok() function for the default stack.
1042  * Note that we'll basically try to take all comers.
1043  */
1044 static int
1045 tcp_default_handoff_ok(struct tcpcb *tp)
1046 {
1047 
1048 	return (0);
1049 }
1050 
1051 /*
1052  * tfb_tcp_fb_init() function for the default stack.
1053  *
1054  * This handles making sure we have appropriate timers set if you are
1055  * transitioning a socket that has some amount of setup done.
1056  *
1057  * The init() fuction from the default can *never* return non-zero i.e.
1058  * it is required to always succeed since it is the stack of last resort!
1059  */
1060 static int
1061 tcp_default_fb_init(struct tcpcb *tp, void **ptr)
1062 {
1063 	struct socket *so = tptosocket(tp);
1064 	int rexmt;
1065 
1066 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1067 	/* We don't use the pointer */
1068 	*ptr = NULL;
1069 
1070 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1071 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1072 	    tp->t_state));
1073 
1074 	/* Make sure we get no interesting mbuf queuing behavior */
1075 	/* All mbuf queue/ack compress flags should be off */
1076 	tcp_lro_features_off(tp);
1077 
1078 	/* Cancel the GP measurement in progress */
1079 	tp->t_flags &= ~TF_GPUTINPROG;
1080 	/* Validate the timers are not in usec, if they are convert */
1081 	tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
1082 	if ((tp->t_state == TCPS_SYN_SENT) ||
1083 	    (tp->t_state == TCPS_SYN_RECEIVED))
1084 		rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
1085 	else
1086 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
1087 	if (tp->t_rxtshift == 0)
1088 		tp->t_rxtcur = rexmt;
1089 	else
1090 		TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX);
1091 
1092 	/*
1093 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1094 	 * know what to do for unexpected states (which includes TIME_WAIT).
1095 	 */
1096 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1097 		return (0);
1098 
1099 	/*
1100 	 * Make sure some kind of transmission timer is set if there is
1101 	 * outstanding data.
1102 	 */
1103 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1104 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1105 	    tcp_timer_active(tp, TT_PERSIST))) {
1106 		/*
1107 		 * If the session has established and it looks like it should
1108 		 * be in the persist state, set the persist timer. Otherwise,
1109 		 * set the retransmit timer.
1110 		 */
1111 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1112 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1113 		    (int32_t)sbavail(&so->so_snd))
1114 			tcp_setpersist(tp);
1115 		else
1116 			tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
1117 	}
1118 
1119 	/* All non-embryonic sessions get a keepalive timer. */
1120 	if (!tcp_timer_active(tp, TT_KEEP))
1121 		tcp_timer_activate(tp, TT_KEEP,
1122 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1123 		    TP_KEEPINIT(tp));
1124 
1125 	/*
1126 	 * Make sure critical variables are initialized
1127 	 * if transitioning while in Recovery.
1128 	 */
1129 	if IN_FASTRECOVERY(tp->t_flags) {
1130 		if (tp->sackhint.recover_fs == 0)
1131 			tp->sackhint.recover_fs = max(1,
1132 			    tp->snd_nxt - tp->snd_una);
1133 	}
1134 
1135 	return (0);
1136 }
1137 
1138 /*
1139  * tfb_tcp_fb_fini() function for the default stack.
1140  *
1141  * This changes state as necessary (or prudent) to prepare for another stack
1142  * to assume responsibility for the connection.
1143  */
1144 static void
1145 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1146 {
1147 
1148 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1149 
1150 #ifdef TCP_BLACKBOX
1151 	tcp_log_flowend(tp);
1152 #endif
1153 	tp->t_acktime = 0;
1154 	return;
1155 }
1156 
1157 /*
1158  * Target size of TCP PCB hash tables. Must be a power of two.
1159  *
1160  * Note that this can be overridden by the kernel environment
1161  * variable net.inet.tcp.tcbhashsize
1162  */
1163 #ifndef TCBHASHSIZE
1164 #define TCBHASHSIZE	0
1165 #endif
1166 
1167 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1168 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1169 
1170 static struct mtx isn_mtx;
1171 
1172 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1173 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1174 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1175 
1176 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1177 
1178 /*
1179  * Take a value and get the next power of 2 that doesn't overflow.
1180  * Used to size the tcp_inpcb hash buckets.
1181  */
1182 static int
1183 maketcp_hashsize(int size)
1184 {
1185 	int hashsize;
1186 
1187 	/*
1188 	 * auto tune.
1189 	 * get the next power of 2 higher than maxsockets.
1190 	 */
1191 	hashsize = 1 << fls(size);
1192 	/* catch overflow, and just go one power of 2 smaller */
1193 	if (hashsize < size) {
1194 		hashsize = 1 << (fls(size) - 1);
1195 	}
1196 	return (hashsize);
1197 }
1198 
1199 static volatile int next_tcp_stack_id = 1;
1200 
1201 /*
1202  * Register a TCP function block with the name provided in the names
1203  * array.  (Note that this function does NOT automatically register
1204  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1205  * explicitly include blk->tfb_tcp_block_name in the list of names if
1206  * you wish to register the stack with that name.)
1207  *
1208  * Either all name registrations will succeed or all will fail.  If
1209  * a name registration fails, the function will update the num_names
1210  * argument to point to the array index of the name that encountered
1211  * the failure.
1212  *
1213  * Returns 0 on success, or an error code on failure.
1214  */
1215 int
1216 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1217     const char *names[], int *num_names)
1218 {
1219 	struct tcp_function *n;
1220 	struct tcp_function_set fs;
1221 	int error, i;
1222 
1223 	KASSERT(names != NULL && *num_names > 0,
1224 	    ("%s: Called with 0-length name list", __func__));
1225 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1226 	KASSERT(rw_initialized(&tcp_function_lock),
1227 	    ("%s: called too early", __func__));
1228 
1229 	if ((blk->tfb_tcp_output == NULL) ||
1230 	    (blk->tfb_tcp_do_segment == NULL) ||
1231 	    (blk->tfb_tcp_ctloutput == NULL) ||
1232 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1233 		/*
1234 		 * These functions are required and you
1235 		 * need a name.
1236 		 */
1237 		*num_names = 0;
1238 		return (EINVAL);
1239 	}
1240 
1241 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1242 		*num_names = 0;
1243 		return (EINVAL);
1244 	}
1245 
1246 	refcount_init(&blk->tfb_refcnt, 0);
1247 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1248 	for (i = 0; i < *num_names; i++) {
1249 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1250 		if (n == NULL) {
1251 			error = ENOMEM;
1252 			goto cleanup;
1253 		}
1254 		n->tf_fb = blk;
1255 
1256 		(void)strlcpy(fs.function_set_name, names[i],
1257 		    sizeof(fs.function_set_name));
1258 		rw_wlock(&tcp_function_lock);
1259 		if (find_tcp_functions_locked(&fs) != NULL) {
1260 			/* Duplicate name space not allowed */
1261 			rw_wunlock(&tcp_function_lock);
1262 			free(n, M_TCPFUNCTIONS);
1263 			error = EALREADY;
1264 			goto cleanup;
1265 		}
1266 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1267 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1268 		tcp_fb_cnt++;
1269 		rw_wunlock(&tcp_function_lock);
1270 	}
1271 	return(0);
1272 
1273 cleanup:
1274 	/*
1275 	 * Deregister the names we just added. Because registration failed
1276 	 * for names[i], we don't need to deregister that name.
1277 	 */
1278 	*num_names = i;
1279 	rw_wlock(&tcp_function_lock);
1280 	while (--i >= 0) {
1281 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1282 			if (!strncmp(n->tf_name, names[i],
1283 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1284 				TAILQ_REMOVE(&t_functions, n, tf_next);
1285 				tcp_fb_cnt--;
1286 				n->tf_fb = NULL;
1287 				free(n, M_TCPFUNCTIONS);
1288 				break;
1289 			}
1290 		}
1291 	}
1292 	rw_wunlock(&tcp_function_lock);
1293 	return (error);
1294 }
1295 
1296 /*
1297  * Register a TCP function block using the name provided in the name
1298  * argument.
1299  *
1300  * Returns 0 on success, or an error code on failure.
1301  */
1302 int
1303 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1304     int wait)
1305 {
1306 	const char *name_list[1];
1307 	int num_names, rv;
1308 
1309 	num_names = 1;
1310 	if (name != NULL)
1311 		name_list[0] = name;
1312 	else
1313 		name_list[0] = blk->tfb_tcp_block_name;
1314 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1315 	return (rv);
1316 }
1317 
1318 /*
1319  * Register a TCP function block using the name defined in
1320  * blk->tfb_tcp_block_name.
1321  *
1322  * Returns 0 on success, or an error code on failure.
1323  */
1324 int
1325 register_tcp_functions(struct tcp_function_block *blk, int wait)
1326 {
1327 
1328 	return (register_tcp_functions_as_name(blk, NULL, wait));
1329 }
1330 
1331 /*
1332  * Deregister all names associated with a function block. This
1333  * functionally removes the function block from use within the system.
1334  *
1335  * When called with a true quiesce argument, mark the function block
1336  * as being removed so no more stacks will use it and determine
1337  * whether the removal would succeed.
1338  *
1339  * When called with a false quiesce argument, actually attempt the
1340  * removal.
1341  *
1342  * When called with a force argument, attempt to switch all TCBs to
1343  * use the default stack instead of returning EBUSY.
1344  *
1345  * Returns 0 on success (or if the removal would succeed), or an error
1346  * code on failure.
1347  */
1348 int
1349 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1350     bool force)
1351 {
1352 	struct tcp_function *f;
1353 	VNET_ITERATOR_DECL(vnet_iter);
1354 
1355 	if (blk == &tcp_def_funcblk) {
1356 		/* You can't un-register the default */
1357 		return (EPERM);
1358 	}
1359 	rw_wlock(&tcp_function_lock);
1360 	VNET_LIST_RLOCK_NOSLEEP();
1361 	VNET_FOREACH(vnet_iter) {
1362 		CURVNET_SET(vnet_iter);
1363 		if (blk == V_tcp_func_set_ptr) {
1364 			/* You can't free the current default in some vnet. */
1365 			CURVNET_RESTORE();
1366 			VNET_LIST_RUNLOCK_NOSLEEP();
1367 			rw_wunlock(&tcp_function_lock);
1368 			return (EBUSY);
1369 		}
1370 		CURVNET_RESTORE();
1371 	}
1372 	VNET_LIST_RUNLOCK_NOSLEEP();
1373 	/* Mark the block so no more stacks can use it. */
1374 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1375 	/*
1376 	 * If TCBs are still attached to the stack, attempt to switch them
1377 	 * to the default stack.
1378 	 */
1379 	if (force && blk->tfb_refcnt) {
1380 		struct inpcb *inp;
1381 		struct tcpcb *tp;
1382 		VNET_ITERATOR_DECL(vnet_iter);
1383 
1384 		rw_wunlock(&tcp_function_lock);
1385 
1386 		VNET_LIST_RLOCK();
1387 		VNET_FOREACH(vnet_iter) {
1388 			CURVNET_SET(vnet_iter);
1389 			struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1390 			    INPLOOKUP_WLOCKPCB);
1391 
1392 			while ((inp = inp_next(&inpi)) != NULL) {
1393 				tp = intotcpcb(inp);
1394 				if (tp == NULL || tp->t_fb != blk)
1395 					continue;
1396 				tcp_switch_back_to_default(tp);
1397 			}
1398 			CURVNET_RESTORE();
1399 		}
1400 		VNET_LIST_RUNLOCK();
1401 
1402 		rw_wlock(&tcp_function_lock);
1403 	}
1404 	if (blk->tfb_refcnt) {
1405 		/* TCBs still attached. */
1406 		rw_wunlock(&tcp_function_lock);
1407 		return (EBUSY);
1408 	}
1409 	if (quiesce) {
1410 		/* Skip removal. */
1411 		rw_wunlock(&tcp_function_lock);
1412 		return (0);
1413 	}
1414 	/* Remove any function names that map to this function block. */
1415 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1416 		TAILQ_REMOVE(&t_functions, f, tf_next);
1417 		tcp_fb_cnt--;
1418 		f->tf_fb = NULL;
1419 		free(f, M_TCPFUNCTIONS);
1420 	}
1421 	rw_wunlock(&tcp_function_lock);
1422 	return (0);
1423 }
1424 
1425 static void
1426 tcp_drain(void)
1427 {
1428 	struct epoch_tracker et;
1429 	VNET_ITERATOR_DECL(vnet_iter);
1430 
1431 	if (!do_tcpdrain)
1432 		return;
1433 
1434 	NET_EPOCH_ENTER(et);
1435 	VNET_LIST_RLOCK_NOSLEEP();
1436 	VNET_FOREACH(vnet_iter) {
1437 		CURVNET_SET(vnet_iter);
1438 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1439 		    INPLOOKUP_WLOCKPCB);
1440 		struct inpcb *inpb;
1441 		struct tcpcb *tcpb;
1442 
1443 	/*
1444 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1445 	 * if there is one...
1446 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1447 	 *      reassembly queue should be flushed, but in a situation
1448 	 *	where we're really low on mbufs, this is potentially
1449 	 *	useful.
1450 	 */
1451 		while ((inpb = inp_next(&inpi)) != NULL) {
1452 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1453 				tcp_reass_flush(tcpb);
1454 				tcp_clean_sackreport(tcpb);
1455 #ifdef TCP_BLACKBOX
1456 				tcp_log_drain(tcpb);
1457 #endif
1458 #ifdef TCPPCAP
1459 				if (tcp_pcap_aggressive_free) {
1460 					/* Free the TCP PCAP queues. */
1461 					tcp_pcap_drain(&(tcpb->t_inpkts));
1462 					tcp_pcap_drain(&(tcpb->t_outpkts));
1463 				}
1464 #endif
1465 			}
1466 		}
1467 		CURVNET_RESTORE();
1468 	}
1469 	VNET_LIST_RUNLOCK_NOSLEEP();
1470 	NET_EPOCH_EXIT(et);
1471 }
1472 
1473 static void
1474 tcp_vnet_init(void *arg __unused)
1475 {
1476 
1477 #ifdef TCP_HHOOK
1478 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1479 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1480 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1481 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1482 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1483 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1484 #endif
1485 #ifdef STATS
1486 	if (tcp_stats_init())
1487 		printf("%s: WARNING: unable to initialise TCP stats\n",
1488 		    __func__);
1489 #endif
1490 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1491 	    tcp_tcbhashsize);
1492 
1493 	syncache_init();
1494 	tcp_hc_init();
1495 
1496 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1497 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1498 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1499 
1500 	tcp_fastopen_init();
1501 
1502 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1503 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1504 
1505 	V_tcp_msl = TCPTV_MSL;
1506 }
1507 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1508     tcp_vnet_init, NULL);
1509 
1510 static void
1511 tcp_init(void *arg __unused)
1512 {
1513 	const char *tcbhash_tuneable;
1514 	int hashsize;
1515 
1516 	tcp_reass_global_init();
1517 
1518 	/* XXX virtualize those below? */
1519 	tcp_delacktime = TCPTV_DELACK;
1520 	tcp_keepinit = TCPTV_KEEP_INIT;
1521 	tcp_keepidle = TCPTV_KEEP_IDLE;
1522 	tcp_keepintvl = TCPTV_KEEPINTVL;
1523 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1524 	tcp_rexmit_initial = TCPTV_RTOBASE;
1525 	if (tcp_rexmit_initial < 1)
1526 		tcp_rexmit_initial = 1;
1527 	tcp_rexmit_min = TCPTV_MIN;
1528 	if (tcp_rexmit_min < 1)
1529 		tcp_rexmit_min = 1;
1530 	tcp_persmin = TCPTV_PERSMIN;
1531 	tcp_persmax = TCPTV_PERSMAX;
1532 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1533 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1534 
1535 	/* Setup the tcp function block list */
1536 	TAILQ_INIT(&t_functions);
1537 	rw_init(&tcp_function_lock, "tcp_func_lock");
1538 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1539 	sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1540 #ifdef TCP_BLACKBOX
1541 	/* Initialize the TCP logging data. */
1542 	tcp_log_init();
1543 #endif
1544 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1545 
1546 	if (tcp_soreceive_stream) {
1547 #ifdef INET
1548 		tcp_protosw.pr_soreceive = soreceive_stream;
1549 #endif
1550 #ifdef INET6
1551 		tcp6_protosw.pr_soreceive = soreceive_stream;
1552 #endif /* INET6 */
1553 	}
1554 
1555 #ifdef INET6
1556 	max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1557 #else /* INET6 */
1558 	max_protohdr_grow(sizeof(struct tcpiphdr));
1559 #endif /* INET6 */
1560 
1561 	ISN_LOCK_INIT();
1562 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1563 		SHUTDOWN_PRI_DEFAULT);
1564 	EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1565 	EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1566 
1567 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1568 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1569 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1570 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1571 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1572 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1573 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1574 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1575 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1576 	tcp_pacing_failures = counter_u64_alloc(M_WAITOK);
1577 #ifdef TCPPCAP
1578 	tcp_pcap_init();
1579 #endif
1580 
1581 	hashsize = TCBHASHSIZE;
1582 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1583 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1584 	if (hashsize == 0) {
1585 		/*
1586 		 * Auto tune the hash size based on maxsockets.
1587 		 * A perfect hash would have a 1:1 mapping
1588 		 * (hashsize = maxsockets) however it's been
1589 		 * suggested that O(2) average is better.
1590 		 */
1591 		hashsize = maketcp_hashsize(maxsockets / 4);
1592 		/*
1593 		 * Our historical default is 512,
1594 		 * do not autotune lower than this.
1595 		 */
1596 		if (hashsize < 512)
1597 			hashsize = 512;
1598 		if (bootverbose)
1599 			printf("%s: %s auto tuned to %d\n", __func__,
1600 			    tcbhash_tuneable, hashsize);
1601 	}
1602 	/*
1603 	 * We require a hashsize to be a power of two.
1604 	 * Previously if it was not a power of two we would just reset it
1605 	 * back to 512, which could be a nasty surprise if you did not notice
1606 	 * the error message.
1607 	 * Instead what we do is clip it to the closest power of two lower
1608 	 * than the specified hash value.
1609 	 */
1610 	if (!powerof2(hashsize)) {
1611 		int oldhashsize = hashsize;
1612 
1613 		hashsize = maketcp_hashsize(hashsize);
1614 		/* prevent absurdly low value */
1615 		if (hashsize < 16)
1616 			hashsize = 16;
1617 		printf("%s: WARNING: TCB hash size not a power of 2, "
1618 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1619 		    hashsize);
1620 	}
1621 	tcp_tcbhashsize = hashsize;
1622 
1623 #ifdef INET
1624 	IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1625 #endif
1626 #ifdef INET6
1627 	IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1628 #endif
1629 }
1630 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1631 
1632 #ifdef VIMAGE
1633 static void
1634 tcp_destroy(void *unused __unused)
1635 {
1636 	int n;
1637 #ifdef TCP_HHOOK
1638 	int error;
1639 #endif
1640 
1641 	/*
1642 	 * All our processes are gone, all our sockets should be cleaned
1643 	 * up, which means, we should be past the tcp_discardcb() calls.
1644 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1645 	 */
1646 	for (;;) {
1647 		INP_INFO_WLOCK(&V_tcbinfo);
1648 		n = V_tcbinfo.ipi_count;
1649 		INP_INFO_WUNLOCK(&V_tcbinfo);
1650 		if (n == 0)
1651 			break;
1652 		pause("tcpdes", hz / 10);
1653 	}
1654 	tcp_hc_destroy();
1655 	syncache_destroy();
1656 	in_pcbinfo_destroy(&V_tcbinfo);
1657 	/* tcp_discardcb() clears the sack_holes up. */
1658 	uma_zdestroy(V_sack_hole_zone);
1659 
1660 	/*
1661 	 * Cannot free the zone until all tcpcbs are released as we attach
1662 	 * the allocations to them.
1663 	 */
1664 	tcp_fastopen_destroy();
1665 
1666 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1667 	VNET_PCPUSTAT_FREE(tcpstat);
1668 
1669 #ifdef TCP_HHOOK
1670 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1671 	if (error != 0) {
1672 		printf("%s: WARNING: unable to deregister helper hook "
1673 		    "type=%d, id=%d: error %d returned\n", __func__,
1674 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1675 	}
1676 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1677 	if (error != 0) {
1678 		printf("%s: WARNING: unable to deregister helper hook "
1679 		    "type=%d, id=%d: error %d returned\n", __func__,
1680 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1681 	}
1682 #endif
1683 }
1684 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1685 #endif
1686 
1687 void
1688 tcp_fini(void *xtp)
1689 {
1690 
1691 }
1692 
1693 /*
1694  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1695  * tcp_template used to store this data in mbufs, but we now recopy it out
1696  * of the tcpcb each time to conserve mbufs.
1697  */
1698 void
1699 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1700 {
1701 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1702 
1703 	INP_WLOCK_ASSERT(inp);
1704 
1705 #ifdef INET6
1706 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1707 		struct ip6_hdr *ip6;
1708 
1709 		ip6 = (struct ip6_hdr *)ip_ptr;
1710 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1711 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1712 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1713 			(IPV6_VERSION & IPV6_VERSION_MASK);
1714 		if (port == 0)
1715 			ip6->ip6_nxt = IPPROTO_TCP;
1716 		else
1717 			ip6->ip6_nxt = IPPROTO_UDP;
1718 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1719 		ip6->ip6_src = inp->in6p_laddr;
1720 		ip6->ip6_dst = inp->in6p_faddr;
1721 	}
1722 #endif /* INET6 */
1723 #if defined(INET6) && defined(INET)
1724 	else
1725 #endif
1726 #ifdef INET
1727 	{
1728 		struct ip *ip;
1729 
1730 		ip = (struct ip *)ip_ptr;
1731 		ip->ip_v = IPVERSION;
1732 		ip->ip_hl = 5;
1733 		ip->ip_tos = inp->inp_ip_tos;
1734 		ip->ip_len = 0;
1735 		ip->ip_id = 0;
1736 		ip->ip_off = 0;
1737 		ip->ip_ttl = inp->inp_ip_ttl;
1738 		ip->ip_sum = 0;
1739 		if (port == 0)
1740 			ip->ip_p = IPPROTO_TCP;
1741 		else
1742 			ip->ip_p = IPPROTO_UDP;
1743 		ip->ip_src = inp->inp_laddr;
1744 		ip->ip_dst = inp->inp_faddr;
1745 	}
1746 #endif /* INET */
1747 	th->th_sport = inp->inp_lport;
1748 	th->th_dport = inp->inp_fport;
1749 	th->th_seq = 0;
1750 	th->th_ack = 0;
1751 	th->th_off = 5;
1752 	tcp_set_flags(th, 0);
1753 	th->th_win = 0;
1754 	th->th_urp = 0;
1755 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1756 }
1757 
1758 /*
1759  * Create template to be used to send tcp packets on a connection.
1760  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1761  * use for this function is in keepalives, which use tcp_respond.
1762  */
1763 struct tcptemp *
1764 tcpip_maketemplate(struct inpcb *inp)
1765 {
1766 	struct tcptemp *t;
1767 
1768 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1769 	if (t == NULL)
1770 		return (NULL);
1771 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1772 	return (t);
1773 }
1774 
1775 /*
1776  * Send a single message to the TCP at address specified by
1777  * the given TCP/IP header.  If m == NULL, then we make a copy
1778  * of the tcpiphdr at th and send directly to the addressed host.
1779  * This is used to force keep alive messages out using the TCP
1780  * template for a connection.  If flags are given then we send
1781  * a message back to the TCP which originated the segment th,
1782  * and discard the mbuf containing it and any other attached mbufs.
1783  *
1784  * In any case the ack and sequence number of the transmitted
1785  * segment are as specified by the parameters.
1786  *
1787  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1788  */
1789 void
1790 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1791     tcp_seq ack, tcp_seq seq, uint16_t flags)
1792 {
1793 	struct tcpopt to;
1794 	struct inpcb *inp;
1795 	struct ip *ip;
1796 	struct mbuf *optm;
1797 	struct udphdr *uh = NULL;
1798 	struct tcphdr *nth;
1799 	struct tcp_log_buffer *lgb;
1800 	u_char *optp;
1801 #ifdef INET6
1802 	struct ip6_hdr *ip6;
1803 	int isipv6;
1804 #endif /* INET6 */
1805 	int optlen, tlen, win, ulen;
1806 	int ect = 0;
1807 	bool incl_opts;
1808 	uint16_t port;
1809 	int output_ret;
1810 #ifdef INVARIANTS
1811 	int thflags = tcp_get_flags(th);
1812 #endif
1813 
1814 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1815 	NET_EPOCH_ASSERT();
1816 
1817 #ifdef INET6
1818 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1819 	ip6 = ipgen;
1820 #endif /* INET6 */
1821 	ip = ipgen;
1822 
1823 	if (tp != NULL) {
1824 		inp = tptoinpcb(tp);
1825 		INP_LOCK_ASSERT(inp);
1826 	} else
1827 		inp = NULL;
1828 
1829 	if (m != NULL) {
1830 #ifdef INET6
1831 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1832 			port = m->m_pkthdr.tcp_tun_port;
1833 		else
1834 #endif
1835 		if (ip && (ip->ip_p == IPPROTO_UDP))
1836 			port = m->m_pkthdr.tcp_tun_port;
1837 		else
1838 			port = 0;
1839 	} else
1840 		port = tp->t_port;
1841 
1842 	incl_opts = false;
1843 	win = 0;
1844 	if (tp != NULL) {
1845 		if (!(flags & TH_RST)) {
1846 			win = sbspace(&inp->inp_socket->so_rcv);
1847 			if (win > TCP_MAXWIN << tp->rcv_scale)
1848 				win = TCP_MAXWIN << tp->rcv_scale;
1849 		}
1850 		if ((tp->t_flags & TF_NOOPT) == 0)
1851 			incl_opts = true;
1852 	}
1853 	if (m == NULL) {
1854 		m = m_gethdr(M_NOWAIT, MT_DATA);
1855 		if (m == NULL)
1856 			return;
1857 		m->m_data += max_linkhdr;
1858 #ifdef INET6
1859 		if (isipv6) {
1860 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1861 			      sizeof(struct ip6_hdr));
1862 			ip6 = mtod(m, struct ip6_hdr *);
1863 			nth = (struct tcphdr *)(ip6 + 1);
1864 			if (port) {
1865 				/* Insert a UDP header */
1866 				uh = (struct udphdr *)nth;
1867 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1868 				uh->uh_dport = port;
1869 				nth = (struct tcphdr *)(uh + 1);
1870 			}
1871 		} else
1872 #endif /* INET6 */
1873 		{
1874 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1875 			ip = mtod(m, struct ip *);
1876 			nth = (struct tcphdr *)(ip + 1);
1877 			if (port) {
1878 				/* Insert a UDP header */
1879 				uh = (struct udphdr *)nth;
1880 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1881 				uh->uh_dport = port;
1882 				nth = (struct tcphdr *)(uh + 1);
1883 			}
1884 		}
1885 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1886 		flags = TH_ACK;
1887 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1888 		struct mbuf *n;
1889 
1890 		/* Can't reuse 'm', allocate a new mbuf. */
1891 		n = m_gethdr(M_NOWAIT, MT_DATA);
1892 		if (n == NULL) {
1893 			m_freem(m);
1894 			return;
1895 		}
1896 
1897 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1898 			m_freem(m);
1899 			m_freem(n);
1900 			return;
1901 		}
1902 
1903 		n->m_data += max_linkhdr;
1904 		/* m_len is set later */
1905 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1906 #ifdef INET6
1907 		if (isipv6) {
1908 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1909 			      sizeof(struct ip6_hdr));
1910 			ip6 = mtod(n, struct ip6_hdr *);
1911 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1912 			nth = (struct tcphdr *)(ip6 + 1);
1913 			if (port) {
1914 				/* Insert a UDP header */
1915 				uh = (struct udphdr *)nth;
1916 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1917 				uh->uh_dport = port;
1918 				nth = (struct tcphdr *)(uh + 1);
1919 			}
1920 		} else
1921 #endif /* INET6 */
1922 		{
1923 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1924 			ip = mtod(n, struct ip *);
1925 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1926 			nth = (struct tcphdr *)(ip + 1);
1927 			if (port) {
1928 				/* Insert a UDP header */
1929 				uh = (struct udphdr *)nth;
1930 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1931 				uh->uh_dport = port;
1932 				nth = (struct tcphdr *)(uh + 1);
1933 			}
1934 		}
1935 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1936 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1937 		th = nth;
1938 		m_freem(m);
1939 		m = n;
1940 	} else {
1941 		/*
1942 		 *  reuse the mbuf.
1943 		 * XXX MRT We inherit the FIB, which is lucky.
1944 		 */
1945 		m_freem(m->m_next);
1946 		m->m_next = NULL;
1947 		m->m_data = (caddr_t)ipgen;
1948 		/* clear any receive flags for proper bpf timestamping */
1949 		m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO);
1950 		/* m_len is set later */
1951 #ifdef INET6
1952 		if (isipv6) {
1953 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1954 			nth = (struct tcphdr *)(ip6 + 1);
1955 		} else
1956 #endif /* INET6 */
1957 		{
1958 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1959 			nth = (struct tcphdr *)(ip + 1);
1960 		}
1961 		if (th != nth) {
1962 			/*
1963 			 * this is usually a case when an extension header
1964 			 * exists between the IPv6 header and the
1965 			 * TCP header.
1966 			 */
1967 			nth->th_sport = th->th_sport;
1968 			nth->th_dport = th->th_dport;
1969 		}
1970 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1971 #undef xchg
1972 	}
1973 	tlen = 0;
1974 #ifdef INET6
1975 	if (isipv6)
1976 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1977 #endif
1978 #if defined(INET) && defined(INET6)
1979 	else
1980 #endif
1981 #ifdef INET
1982 		tlen = sizeof (struct tcpiphdr);
1983 #endif
1984 	if (port)
1985 		tlen += sizeof (struct udphdr);
1986 #ifdef INVARIANTS
1987 	m->m_len = 0;
1988 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1989 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1990 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1991 #endif
1992 	m->m_len = tlen;
1993 	to.to_flags = 0;
1994 	if (incl_opts) {
1995 		ect = tcp_ecn_output_established(tp, &flags, 0, false);
1996 		/* Make sure we have room. */
1997 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1998 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1999 			if (m->m_next) {
2000 				optp = mtod(m->m_next, u_char *);
2001 				optm = m->m_next;
2002 			} else
2003 				incl_opts = false;
2004 		} else {
2005 			optp = (u_char *) (nth + 1);
2006 			optm = m;
2007 		}
2008 	}
2009 	if (incl_opts) {
2010 		/* Timestamps. */
2011 		if (tp->t_flags & TF_RCVD_TSTMP) {
2012 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
2013 			to.to_tsecr = tp->ts_recent;
2014 			to.to_flags |= TOF_TS;
2015 		}
2016 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2017 		/* TCP-MD5 (RFC2385). */
2018 		if (tp->t_flags & TF_SIGNATURE)
2019 			to.to_flags |= TOF_SIGNATURE;
2020 #endif
2021 		/* Add the options. */
2022 		tlen += optlen = tcp_addoptions(&to, optp);
2023 
2024 		/* Update m_len in the correct mbuf. */
2025 		optm->m_len += optlen;
2026 	} else
2027 		optlen = 0;
2028 #ifdef INET6
2029 	if (isipv6) {
2030 		if (uh) {
2031 			ulen = tlen - sizeof(struct ip6_hdr);
2032 			uh->uh_ulen = htons(ulen);
2033 		}
2034 		ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN);
2035 		ip6->ip6_vfc = IPV6_VERSION;
2036 		if (port)
2037 			ip6->ip6_nxt = IPPROTO_UDP;
2038 		else
2039 			ip6->ip6_nxt = IPPROTO_TCP;
2040 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
2041 	}
2042 #endif
2043 #if defined(INET) && defined(INET6)
2044 	else
2045 #endif
2046 #ifdef INET
2047 	{
2048 		if (uh) {
2049 			ulen = tlen - sizeof(struct ip);
2050 			uh->uh_ulen = htons(ulen);
2051 		}
2052 		ip->ip_len = htons(tlen);
2053 		if (inp != NULL) {
2054 			ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK;
2055 			ip->ip_ttl = inp->inp_ip_ttl;
2056 		} else {
2057 			ip->ip_tos = 0;
2058 			ip->ip_ttl = V_ip_defttl;
2059 		}
2060 		ip->ip_tos |= ect;
2061 		if (port) {
2062 			ip->ip_p = IPPROTO_UDP;
2063 		} else {
2064 			ip->ip_p = IPPROTO_TCP;
2065 		}
2066 		if (V_path_mtu_discovery)
2067 			ip->ip_off |= htons(IP_DF);
2068 	}
2069 #endif
2070 	m->m_pkthdr.len = tlen;
2071 	m->m_pkthdr.rcvif = NULL;
2072 #ifdef MAC
2073 	if (inp != NULL) {
2074 		/*
2075 		 * Packet is associated with a socket, so allow the
2076 		 * label of the response to reflect the socket label.
2077 		 */
2078 		INP_LOCK_ASSERT(inp);
2079 		mac_inpcb_create_mbuf(inp, m);
2080 	} else {
2081 		/*
2082 		 * Packet is not associated with a socket, so possibly
2083 		 * update the label in place.
2084 		 */
2085 		mac_netinet_tcp_reply(m);
2086 	}
2087 #endif
2088 	nth->th_seq = htonl(seq);
2089 	nth->th_ack = htonl(ack);
2090 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2091 	tcp_set_flags(nth, flags);
2092 	if (tp && (flags & TH_RST)) {
2093 		/* Log the reset */
2094 		tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2095 	}
2096 	if (tp != NULL)
2097 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2098 	else
2099 		nth->th_win = htons((u_short)win);
2100 	nth->th_urp = 0;
2101 
2102 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2103 	if (to.to_flags & TOF_SIGNATURE) {
2104 		if (!TCPMD5_ENABLED() ||
2105 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2106 			m_freem(m);
2107 			return;
2108 		}
2109 	}
2110 #endif
2111 
2112 #ifdef INET6
2113 	if (isipv6) {
2114 		if (port) {
2115 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2116 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2117 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2118 			nth->th_sum = 0;
2119 		} else {
2120 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2121 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2122 			nth->th_sum = in6_cksum_pseudo(ip6,
2123 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2124 		}
2125 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2126 	}
2127 #endif /* INET6 */
2128 #if defined(INET6) && defined(INET)
2129 	else
2130 #endif
2131 #ifdef INET
2132 	{
2133 		if (port) {
2134 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2135 			    htons(ulen + IPPROTO_UDP));
2136 			m->m_pkthdr.csum_flags = CSUM_UDP;
2137 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2138 			nth->th_sum = 0;
2139 		} else {
2140 			m->m_pkthdr.csum_flags = CSUM_TCP;
2141 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2142 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2143 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2144 		}
2145 	}
2146 #endif /* INET */
2147 	TCP_PROBE3(debug__output, tp, th, m);
2148 	if (flags & TH_RST)
2149 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2150 	lgb = NULL;
2151 	if ((tp != NULL) && tcp_bblogging_on(tp)) {
2152 		if (INP_WLOCKED(inp)) {
2153 			union tcp_log_stackspecific log;
2154 			struct timeval tv;
2155 
2156 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2157 			log.u_bbr.inhpts = tcp_in_hpts(tp);
2158 			log.u_bbr.flex8 = 4;
2159 			log.u_bbr.pkts_out = tp->t_maxseg;
2160 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2161 			log.u_bbr.delivered = 0;
2162 			lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT,
2163 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2164 		} else {
2165 			/*
2166 			 * We can not log the packet, since we only own the
2167 			 * read lock, but a write lock is needed. The read lock
2168 			 * is not upgraded to a write lock, since only getting
2169 			 * the read lock was done intentionally to improve the
2170 			 * handling of SYN flooding attacks.
2171 			 * This happens only for pure SYN segments received in
2172 			 * the initial CLOSED state, or received in a more
2173 			 * advanced state than listen and the UDP encapsulation
2174 			 * port is unexpected.
2175 			 * The incoming SYN segments do not really belong to
2176 			 * the TCP connection and the handling does not change
2177 			 * the state of the TCP connection. Therefore, the
2178 			 * sending of the RST segments is not logged. Please
2179 			 * note that also the incoming SYN segments are not
2180 			 * logged.
2181 			 *
2182 			 * The following code ensures that the above description
2183 			 * is and stays correct.
2184 			 */
2185 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2186 			    (tp->t_state == TCPS_CLOSED ||
2187 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2188 			    ("%s: Logging of TCP segment with flags 0x%b and "
2189 			    "UDP encapsulation port %u skipped in state %s",
2190 			    __func__, thflags, PRINT_TH_FLAGS,
2191 			    ntohs(port), tcpstates[tp->t_state]));
2192 		}
2193 	}
2194 
2195 	if (flags & TH_ACK)
2196 		TCPSTAT_INC(tcps_sndacks);
2197 	else if (flags & (TH_SYN|TH_FIN|TH_RST))
2198 		TCPSTAT_INC(tcps_sndctrl);
2199 	TCPSTAT_INC(tcps_sndtotal);
2200 
2201 #ifdef INET6
2202 	if (isipv6) {
2203 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2204 		output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL,
2205 		    NULL, 0, NULL, NULL, inp);
2206 	}
2207 #endif /* INET6 */
2208 #if defined(INET) && defined(INET6)
2209 	else
2210 #endif
2211 #ifdef INET
2212 	{
2213 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2214 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2215 	}
2216 #endif
2217 	if (lgb != NULL)
2218 		lgb->tlb_errno = output_ret;
2219 }
2220 
2221 /*
2222  * Create a new TCP control block, making an empty reassembly queue and hooking
2223  * it to the argument protocol control block.  The `inp' parameter must have
2224  * come from the zone allocator set up by tcpcbstor declaration.
2225  */
2226 struct tcpcb *
2227 tcp_newtcpcb(struct inpcb *inp)
2228 {
2229 	struct tcpcb *tp = intotcpcb(inp);
2230 #ifdef INET6
2231 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2232 #endif /* INET6 */
2233 
2234 	/*
2235 	 * Historically allocation was done with M_ZERO.  There is a lot of
2236 	 * code that rely on that.  For now take safe approach and zero whole
2237 	 * tcpcb.  This definitely can be optimized.
2238 	 */
2239 	bzero(&tp->t_start_zero, t_zero_size);
2240 
2241 	/* Initialise cc_var struct for this tcpcb. */
2242 	tp->t_ccv.type = IPPROTO_TCP;
2243 	tp->t_ccv.ccvc.tcp = tp;
2244 	rw_rlock(&tcp_function_lock);
2245 	tp->t_fb = V_tcp_func_set_ptr;
2246 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2247 	rw_runlock(&tcp_function_lock);
2248 	/*
2249 	 * Use the current system default CC algorithm.
2250 	 */
2251 	cc_attach(tp, CC_DEFAULT_ALGO());
2252 
2253 	if (CC_ALGO(tp)->cb_init != NULL)
2254 		if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2255 			cc_detach(tp);
2256 			if (tp->t_fb->tfb_tcp_fb_fini)
2257 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2258 			refcount_release(&tp->t_fb->tfb_refcnt);
2259 			return (NULL);
2260 		}
2261 
2262 #ifdef TCP_HHOOK
2263 	if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2264 		if (tp->t_fb->tfb_tcp_fb_fini)
2265 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2266 		refcount_release(&tp->t_fb->tfb_refcnt);
2267 		return (NULL);
2268 	}
2269 #endif
2270 
2271 	TAILQ_INIT(&tp->t_segq);
2272 	STAILQ_INIT(&tp->t_inqueue);
2273 	tp->t_maxseg =
2274 #ifdef INET6
2275 		isipv6 ? V_tcp_v6mssdflt :
2276 #endif /* INET6 */
2277 		V_tcp_mssdflt;
2278 
2279 	/* All mbuf queue/ack compress flags should be off */
2280 	tcp_lro_features_off(tp);
2281 
2282 	callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2283 	for (int i = 0; i < TT_N; i++)
2284 		tp->t_timers[i] = SBT_MAX;
2285 
2286 	switch (V_tcp_do_rfc1323) {
2287 		case 0:
2288 			break;
2289 		default:
2290 		case 1:
2291 			tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2292 			break;
2293 		case 2:
2294 			tp->t_flags = TF_REQ_SCALE;
2295 			break;
2296 		case 3:
2297 			tp->t_flags = TF_REQ_TSTMP;
2298 			break;
2299 	}
2300 	if (V_tcp_do_sack)
2301 		tp->t_flags |= TF_SACK_PERMIT;
2302 	TAILQ_INIT(&tp->snd_holes);
2303 
2304 	/*
2305 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2306 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2307 	 * reasonable initial retransmit time.
2308 	 */
2309 	tp->t_srtt = TCPTV_SRTTBASE;
2310 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2311 	tp->t_rttmin = tcp_rexmit_min;
2312 	tp->t_rxtcur = tcp_rexmit_initial;
2313 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2314 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2315 	tp->t_rcvtime = ticks;
2316 	/* We always start with ticks granularity */
2317 	tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
2318 	/*
2319 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2320 	 * because the socket may be bound to an IPv6 wildcard address,
2321 	 * which may match an IPv4-mapped IPv6 address.
2322 	 */
2323 	inp->inp_ip_ttl = V_ip_defttl;
2324 #ifdef TCPHPTS
2325 	tcp_hpts_init(tp);
2326 #endif
2327 #ifdef TCPPCAP
2328 	/*
2329 	 * Init the TCP PCAP queues.
2330 	 */
2331 	tcp_pcap_tcpcb_init(tp);
2332 #endif
2333 #ifdef TCP_BLACKBOX
2334 	/* Initialize the per-TCPCB log data. */
2335 	tcp_log_tcpcbinit(tp);
2336 #endif
2337 	tp->t_pacing_rate = -1;
2338 	if (tp->t_fb->tfb_tcp_fb_init) {
2339 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) {
2340 			refcount_release(&tp->t_fb->tfb_refcnt);
2341 			return (NULL);
2342 		}
2343 	}
2344 #ifdef STATS
2345 	if (V_tcp_perconn_stats_enable == 1)
2346 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2347 #endif
2348 	if (V_tcp_do_lrd)
2349 		tp->t_flags |= TF_LRD;
2350 
2351 	return (tp);
2352 }
2353 
2354 /*
2355  * Drop a TCP connection, reporting
2356  * the specified error.  If connection is synchronized,
2357  * then send a RST to peer.
2358  */
2359 struct tcpcb *
2360 tcp_drop(struct tcpcb *tp, int errno)
2361 {
2362 	struct socket *so = tptosocket(tp);
2363 
2364 	NET_EPOCH_ASSERT();
2365 	INP_WLOCK_ASSERT(tptoinpcb(tp));
2366 
2367 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2368 		tcp_state_change(tp, TCPS_CLOSED);
2369 		/* Don't use tcp_output() here due to possible recursion. */
2370 		(void)tcp_output_nodrop(tp);
2371 		TCPSTAT_INC(tcps_drops);
2372 	} else
2373 		TCPSTAT_INC(tcps_conndrops);
2374 	if (errno == ETIMEDOUT && tp->t_softerror)
2375 		errno = tp->t_softerror;
2376 	so->so_error = errno;
2377 	return (tcp_close(tp));
2378 }
2379 
2380 void
2381 tcp_discardcb(struct tcpcb *tp)
2382 {
2383 	struct inpcb *inp = tptoinpcb(tp);
2384 	struct socket *so = tptosocket(tp);
2385 	struct mbuf *m;
2386 #ifdef INET6
2387 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2388 #endif
2389 
2390 	INP_WLOCK_ASSERT(inp);
2391 
2392 	tcp_timer_stop(tp);
2393 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2394 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2395 	}
2396 
2397 	/* free the reassembly queue, if any */
2398 	tcp_reass_flush(tp);
2399 
2400 #ifdef TCP_OFFLOAD
2401 	/* Disconnect offload device, if any. */
2402 	if (tp->t_flags & TF_TOE)
2403 		tcp_offload_detach(tp);
2404 #endif
2405 
2406 	tcp_free_sackholes(tp);
2407 
2408 #ifdef TCPPCAP
2409 	/* Free the TCP PCAP queues. */
2410 	tcp_pcap_drain(&(tp->t_inpkts));
2411 	tcp_pcap_drain(&(tp->t_outpkts));
2412 #endif
2413 
2414 	/* Allow the CC algorithm to clean up after itself. */
2415 	if (CC_ALGO(tp)->cb_destroy != NULL)
2416 		CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2417 	CC_DATA(tp) = NULL;
2418 	/* Detach from the CC algorithm */
2419 	cc_detach(tp);
2420 
2421 #ifdef TCP_HHOOK
2422 	khelp_destroy_osd(&tp->t_osd);
2423 #endif
2424 #ifdef STATS
2425 	stats_blob_destroy(tp->t_stats);
2426 #endif
2427 
2428 	CC_ALGO(tp) = NULL;
2429 	if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) {
2430 		struct mbuf *prev;
2431 
2432 		STAILQ_INIT(&tp->t_inqueue);
2433 		STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev)
2434 			m_freem(m);
2435 	}
2436 	TCPSTATES_DEC(tp->t_state);
2437 
2438 	if (tp->t_fb->tfb_tcp_fb_fini)
2439 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2440 	MPASS(!tcp_in_hpts(tp));
2441 #ifdef TCP_BLACKBOX
2442 	tcp_log_tcpcbfini(tp);
2443 #endif
2444 
2445 	/*
2446 	 * If we got enough samples through the srtt filter,
2447 	 * save the rtt and rttvar in the routing entry.
2448 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2449 	 * 4 samples is enough for the srtt filter to converge
2450 	 * to within enough % of the correct value; fewer samples
2451 	 * and we could save a bogus rtt. The danger is not high
2452 	 * as tcp quickly recovers from everything.
2453 	 * XXX: Works very well but needs some more statistics!
2454 	 *
2455 	 * XXXRRS: Updating must be after the stack fini() since
2456 	 * that may be converting some internal representation of
2457 	 * say srtt etc into the general one used by other stacks.
2458 	 * Lets also at least protect against the so being NULL
2459 	 * as RW stated below.
2460 	 */
2461 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2462 		struct hc_metrics_lite metrics;
2463 		uint32_t ssthresh;
2464 
2465 		bzero(&metrics, sizeof(metrics));
2466 		/*
2467 		 * Update the ssthresh always when the conditions below
2468 		 * are satisfied. This gives us better new start value
2469 		 * for the congestion avoidance for new connections.
2470 		 * ssthresh is only set if packet loss occurred on a session.
2471 		 *
2472 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2473 		 * being torn down.  Ideally this code would not use 'so'.
2474 		 */
2475 		ssthresh = tp->snd_ssthresh;
2476 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2477 			/*
2478 			 * convert the limit from user data bytes to
2479 			 * packets then to packet data bytes.
2480 			 */
2481 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2482 			if (ssthresh < 2)
2483 				ssthresh = 2;
2484 			ssthresh *= (tp->t_maxseg +
2485 #ifdef INET6
2486 			    (isipv6 ? sizeof (struct ip6_hdr) +
2487 			    sizeof (struct tcphdr) :
2488 #endif
2489 			    sizeof (struct tcpiphdr)
2490 #ifdef INET6
2491 			    )
2492 #endif
2493 			    );
2494 		} else
2495 			ssthresh = 0;
2496 		metrics.rmx_ssthresh = ssthresh;
2497 
2498 		metrics.rmx_rtt = tp->t_srtt;
2499 		metrics.rmx_rttvar = tp->t_rttvar;
2500 		metrics.rmx_cwnd = tp->snd_cwnd;
2501 		metrics.rmx_sendpipe = 0;
2502 		metrics.rmx_recvpipe = 0;
2503 
2504 		tcp_hc_update(&inp->inp_inc, &metrics);
2505 	}
2506 
2507 	refcount_release(&tp->t_fb->tfb_refcnt);
2508 }
2509 
2510 /*
2511  * Attempt to close a TCP control block, marking it as dropped, and freeing
2512  * the socket if we hold the only reference.
2513  */
2514 struct tcpcb *
2515 tcp_close(struct tcpcb *tp)
2516 {
2517 	struct inpcb *inp = tptoinpcb(tp);
2518 	struct socket *so = tptosocket(tp);
2519 
2520 	INP_WLOCK_ASSERT(inp);
2521 
2522 #ifdef TCP_OFFLOAD
2523 	if (tp->t_state == TCPS_LISTEN)
2524 		tcp_offload_listen_stop(tp);
2525 #endif
2526 	/*
2527 	 * This releases the TFO pending counter resource for TFO listen
2528 	 * sockets as well as passively-created TFO sockets that transition
2529 	 * from SYN_RECEIVED to CLOSED.
2530 	 */
2531 	if (tp->t_tfo_pending) {
2532 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2533 		tp->t_tfo_pending = NULL;
2534 	}
2535 #ifdef TCPHPTS
2536 	tcp_hpts_remove(tp);
2537 #endif
2538 	in_pcbdrop(inp);
2539 	TCPSTAT_INC(tcps_closed);
2540 	if (tp->t_state != TCPS_CLOSED)
2541 		tcp_state_change(tp, TCPS_CLOSED);
2542 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2543 	soisdisconnected(so);
2544 	if (inp->inp_flags & INP_SOCKREF) {
2545 		inp->inp_flags &= ~INP_SOCKREF;
2546 		INP_WUNLOCK(inp);
2547 		sorele(so);
2548 		return (NULL);
2549 	}
2550 	return (tp);
2551 }
2552 
2553 /*
2554  * Notify a tcp user of an asynchronous error;
2555  * store error as soft error, but wake up user
2556  * (for now, won't do anything until can select for soft error).
2557  *
2558  * Do not wake up user since there currently is no mechanism for
2559  * reporting soft errors (yet - a kqueue filter may be added).
2560  */
2561 static struct inpcb *
2562 tcp_notify(struct inpcb *inp, int error)
2563 {
2564 	struct tcpcb *tp;
2565 
2566 	INP_WLOCK_ASSERT(inp);
2567 
2568 	tp = intotcpcb(inp);
2569 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2570 
2571 	/*
2572 	 * Ignore some errors if we are hooked up.
2573 	 * If connection hasn't completed, has retransmitted several times,
2574 	 * and receives a second error, give up now.  This is better
2575 	 * than waiting a long time to establish a connection that
2576 	 * can never complete.
2577 	 */
2578 	if (tp->t_state == TCPS_ESTABLISHED &&
2579 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2580 	     error == EHOSTDOWN)) {
2581 		if (inp->inp_route.ro_nh) {
2582 			NH_FREE(inp->inp_route.ro_nh);
2583 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2584 		}
2585 		return (inp);
2586 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2587 	    tp->t_softerror) {
2588 		tp = tcp_drop(tp, error);
2589 		if (tp != NULL)
2590 			return (inp);
2591 		else
2592 			return (NULL);
2593 	} else {
2594 		tp->t_softerror = error;
2595 		return (inp);
2596 	}
2597 #if 0
2598 	wakeup( &so->so_timeo);
2599 	sorwakeup(so);
2600 	sowwakeup(so);
2601 #endif
2602 }
2603 
2604 static int
2605 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2606 {
2607 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2608 	    INPLOOKUP_RLOCKPCB);
2609 	struct xinpgen xig;
2610 	struct inpcb *inp;
2611 	int error;
2612 
2613 	if (req->newptr != NULL)
2614 		return (EPERM);
2615 
2616 	if (req->oldptr == NULL) {
2617 		int n;
2618 
2619 		n = V_tcbinfo.ipi_count +
2620 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2621 		n += imax(n / 8, 10);
2622 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2623 		return (0);
2624 	}
2625 
2626 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2627 		return (error);
2628 
2629 	bzero(&xig, sizeof(xig));
2630 	xig.xig_len = sizeof xig;
2631 	xig.xig_count = V_tcbinfo.ipi_count +
2632 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2633 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2634 	xig.xig_sogen = so_gencnt;
2635 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2636 	if (error)
2637 		return (error);
2638 
2639 	error = syncache_pcblist(req);
2640 	if (error)
2641 		return (error);
2642 
2643 	while ((inp = inp_next(&inpi)) != NULL) {
2644 		if (inp->inp_gencnt <= xig.xig_gen &&
2645 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2646 			struct xtcpcb xt;
2647 
2648 			tcp_inptoxtp(inp, &xt);
2649 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2650 			if (error) {
2651 				INP_RUNLOCK(inp);
2652 				break;
2653 			} else
2654 				continue;
2655 		}
2656 	}
2657 
2658 	if (!error) {
2659 		/*
2660 		 * Give the user an updated idea of our state.
2661 		 * If the generation differs from what we told
2662 		 * her before, she knows that something happened
2663 		 * while we were processing this request, and it
2664 		 * might be necessary to retry.
2665 		 */
2666 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2667 		xig.xig_sogen = so_gencnt;
2668 		xig.xig_count = V_tcbinfo.ipi_count +
2669 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2670 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2671 	}
2672 
2673 	return (error);
2674 }
2675 
2676 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2677     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2678     NULL, 0, tcp_pcblist, "S,xtcpcb",
2679     "List of active TCP connections");
2680 
2681 #ifdef INET
2682 static int
2683 tcp_getcred(SYSCTL_HANDLER_ARGS)
2684 {
2685 	struct xucred xuc;
2686 	struct sockaddr_in addrs[2];
2687 	struct epoch_tracker et;
2688 	struct inpcb *inp;
2689 	int error;
2690 
2691 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2692 	if (error)
2693 		return (error);
2694 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2695 	if (error)
2696 		return (error);
2697 	NET_EPOCH_ENTER(et);
2698 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2699 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2700 	NET_EPOCH_EXIT(et);
2701 	if (inp != NULL) {
2702 		if (error == 0)
2703 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2704 		if (error == 0)
2705 			cru2x(inp->inp_cred, &xuc);
2706 		INP_RUNLOCK(inp);
2707 	} else
2708 		error = ENOENT;
2709 	if (error == 0)
2710 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2711 	return (error);
2712 }
2713 
2714 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2715     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2716     0, 0, tcp_getcred, "S,xucred",
2717     "Get the xucred of a TCP connection");
2718 #endif /* INET */
2719 
2720 #ifdef INET6
2721 static int
2722 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2723 {
2724 	struct epoch_tracker et;
2725 	struct xucred xuc;
2726 	struct sockaddr_in6 addrs[2];
2727 	struct inpcb *inp;
2728 	int error;
2729 #ifdef INET
2730 	int mapped = 0;
2731 #endif
2732 
2733 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2734 	if (error)
2735 		return (error);
2736 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2737 	if (error)
2738 		return (error);
2739 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2740 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2741 		return (error);
2742 	}
2743 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2744 #ifdef INET
2745 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2746 			mapped = 1;
2747 		else
2748 #endif
2749 			return (EINVAL);
2750 	}
2751 
2752 	NET_EPOCH_ENTER(et);
2753 #ifdef INET
2754 	if (mapped == 1)
2755 		inp = in_pcblookup(&V_tcbinfo,
2756 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2757 			addrs[1].sin6_port,
2758 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2759 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2760 	else
2761 #endif
2762 		inp = in6_pcblookup(&V_tcbinfo,
2763 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2764 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2765 			INPLOOKUP_RLOCKPCB, NULL);
2766 	NET_EPOCH_EXIT(et);
2767 	if (inp != NULL) {
2768 		if (error == 0)
2769 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2770 		if (error == 0)
2771 			cru2x(inp->inp_cred, &xuc);
2772 		INP_RUNLOCK(inp);
2773 	} else
2774 		error = ENOENT;
2775 	if (error == 0)
2776 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2777 	return (error);
2778 }
2779 
2780 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2781     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2782     0, 0, tcp6_getcred, "S,xucred",
2783     "Get the xucred of a TCP6 connection");
2784 #endif /* INET6 */
2785 
2786 #ifdef INET
2787 /* Path MTU to try next when a fragmentation-needed message is received. */
2788 static inline int
2789 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2790 {
2791 	int mtu = ntohs(icp->icmp_nextmtu);
2792 
2793 	/* If no alternative MTU was proposed, try the next smaller one. */
2794 	if (!mtu)
2795 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2796 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2797 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2798 
2799 	return (mtu);
2800 }
2801 
2802 static void
2803 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2804 {
2805 	struct ip *ip;
2806 	struct tcphdr *th;
2807 	struct inpcb *inp;
2808 	struct tcpcb *tp;
2809 	struct inpcb *(*notify)(struct inpcb *, int);
2810 	struct in_conninfo inc;
2811 	tcp_seq icmp_tcp_seq;
2812 	int errno, mtu;
2813 
2814 	errno = icmp_errmap(icp);
2815 	switch (errno) {
2816 	case 0:
2817 		return;
2818 	case EMSGSIZE:
2819 		notify = tcp_mtudisc_notify;
2820 		break;
2821 	case ECONNREFUSED:
2822 		if (V_icmp_may_rst)
2823 			notify = tcp_drop_syn_sent;
2824 		else
2825 			notify = tcp_notify;
2826 		break;
2827 	case EHOSTUNREACH:
2828 		if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2829 			notify = tcp_drop_syn_sent;
2830 		else
2831 			notify = tcp_notify;
2832 		break;
2833 	default:
2834 		notify = tcp_notify;
2835 	}
2836 
2837 	ip = &icp->icmp_ip;
2838 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2839 	icmp_tcp_seq = th->th_seq;
2840 	inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2841 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2842 	if (inp != NULL)  {
2843 		tp = intotcpcb(inp);
2844 #ifdef TCP_OFFLOAD
2845 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2846 			/*
2847 			 * MTU discovery for offloaded connections.  Let
2848 			 * the TOE driver verify seq# and process it.
2849 			 */
2850 			mtu = tcp_next_pmtu(icp, ip);
2851 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2852 			goto out;
2853 		}
2854 #endif
2855 		if (tp->t_port != port)
2856 			goto out;
2857 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2858 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2859 			if (errno == EMSGSIZE) {
2860 				/*
2861 				 * MTU discovery: we got a needfrag and
2862 				 * will potentially try a lower MTU.
2863 				 */
2864 				mtu = tcp_next_pmtu(icp, ip);
2865 
2866 				/*
2867 				 * Only process the offered MTU if it
2868 				 * is smaller than the current one.
2869 				 */
2870 				if (mtu < tp->t_maxseg +
2871 				    sizeof(struct tcpiphdr)) {
2872 					bzero(&inc, sizeof(inc));
2873 					inc.inc_faddr = ip->ip_dst;
2874 					inc.inc_fibnum =
2875 					    inp->inp_inc.inc_fibnum;
2876 					tcp_hc_updatemtu(&inc, mtu);
2877 					inp = tcp_mtudisc(inp, mtu);
2878 				}
2879 			} else
2880 				inp = (*notify)(inp, errno);
2881 		}
2882 	} else {
2883 		bzero(&inc, sizeof(inc));
2884 		inc.inc_fport = th->th_dport;
2885 		inc.inc_lport = th->th_sport;
2886 		inc.inc_faddr = ip->ip_dst;
2887 		inc.inc_laddr = ip->ip_src;
2888 		syncache_unreach(&inc, icmp_tcp_seq, port);
2889 	}
2890 out:
2891 	if (inp != NULL)
2892 		INP_WUNLOCK(inp);
2893 }
2894 
2895 static void
2896 tcp_ctlinput(struct icmp *icmp)
2897 {
2898 	tcp_ctlinput_with_port(icmp, htons(0));
2899 }
2900 
2901 static void
2902 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2903 {
2904 	/* Its a tunneled TCP over UDP icmp */
2905 	struct icmp *icmp = param.icmp;
2906 	struct ip *outer_ip, *inner_ip;
2907 	struct udphdr *udp;
2908 	struct tcphdr *th, ttemp;
2909 	int i_hlen, o_len;
2910 	uint16_t port;
2911 
2912 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2913 	inner_ip = &icmp->icmp_ip;
2914 	i_hlen = inner_ip->ip_hl << 2;
2915 	o_len = ntohs(outer_ip->ip_len);
2916 	if (o_len <
2917 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2918 		/* Not enough data present */
2919 		return;
2920 	}
2921 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2922 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2923 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2924 		return;
2925 	}
2926 	port = udp->uh_dport;
2927 	th = (struct tcphdr *)(udp + 1);
2928 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2929 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2930 	/* Now adjust down the size of the outer IP header */
2931 	o_len -= sizeof(struct udphdr);
2932 	outer_ip->ip_len = htons(o_len);
2933 	/* Now call in to the normal handling code */
2934 	tcp_ctlinput_with_port(icmp, port);
2935 }
2936 #endif /* INET */
2937 
2938 #ifdef INET6
2939 static inline int
2940 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2941 {
2942 	int mtu = ntohl(icmp6->icmp6_mtu);
2943 
2944 	/*
2945 	 * If no alternative MTU was proposed, or the proposed MTU was too
2946 	 * small, set to the min.
2947 	 */
2948 	if (mtu < IPV6_MMTU)
2949 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
2950 	return (mtu);
2951 }
2952 
2953 static void
2954 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2955 {
2956 	struct in6_addr *dst;
2957 	struct inpcb *(*notify)(struct inpcb *, int);
2958 	struct ip6_hdr *ip6;
2959 	struct mbuf *m;
2960 	struct inpcb *inp;
2961 	struct tcpcb *tp;
2962 	struct icmp6_hdr *icmp6;
2963 	struct in_conninfo inc;
2964 	struct tcp_ports {
2965 		uint16_t th_sport;
2966 		uint16_t th_dport;
2967 	} t_ports;
2968 	tcp_seq icmp_tcp_seq;
2969 	unsigned int mtu;
2970 	unsigned int off;
2971 	int errno;
2972 
2973 	icmp6 = ip6cp->ip6c_icmp6;
2974 	m = ip6cp->ip6c_m;
2975 	ip6 = ip6cp->ip6c_ip6;
2976 	off = ip6cp->ip6c_off;
2977 	dst = &ip6cp->ip6c_finaldst->sin6_addr;
2978 
2979 	errno = icmp6_errmap(icmp6);
2980 	switch (errno) {
2981 	case 0:
2982 		return;
2983 	case EMSGSIZE:
2984 		notify = tcp_mtudisc_notify;
2985 		break;
2986 	case ECONNREFUSED:
2987 		if (V_icmp_may_rst)
2988 			notify = tcp_drop_syn_sent;
2989 		else
2990 			notify = tcp_notify;
2991 		break;
2992 	case EHOSTUNREACH:
2993 		/*
2994 		 * There are only four ICMPs that may reset connection:
2995 		 * - administratively prohibited
2996 		 * - port unreachable
2997 		 * - time exceeded in transit
2998 		 * - unknown next header
2999 		 */
3000 		if (V_icmp_may_rst &&
3001 		    ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
3002 		     (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
3003 		      icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
3004 		    (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
3005 		      icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
3006 		    (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
3007 		      icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
3008 			notify = tcp_drop_syn_sent;
3009 		else
3010 			notify = tcp_notify;
3011 		break;
3012 	default:
3013 		notify = tcp_notify;
3014 	}
3015 
3016 	/* Check if we can safely get the ports from the tcp hdr */
3017 	if (m == NULL ||
3018 	    (m->m_pkthdr.len <
3019 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3020 		return;
3021 	}
3022 	bzero(&t_ports, sizeof(struct tcp_ports));
3023 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3024 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3025 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3026 	off += sizeof(struct tcp_ports);
3027 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3028 		goto out;
3029 	}
3030 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3031 	if (inp != NULL)  {
3032 		tp = intotcpcb(inp);
3033 #ifdef TCP_OFFLOAD
3034 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
3035 			/* MTU discovery for offloaded connections. */
3036 			mtu = tcp6_next_pmtu(icmp6);
3037 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3038 			goto out;
3039 		}
3040 #endif
3041 		if (tp->t_port != port)
3042 			goto out;
3043 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3044 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3045 			if (errno == EMSGSIZE) {
3046 				/*
3047 				 * MTU discovery:
3048 				 * If we got a needfrag set the MTU
3049 				 * in the route to the suggested new
3050 				 * value (if given) and then notify.
3051 				 */
3052 				mtu = tcp6_next_pmtu(icmp6);
3053 
3054 				bzero(&inc, sizeof(inc));
3055 				inc.inc_fibnum = M_GETFIB(m);
3056 				inc.inc_flags |= INC_ISIPV6;
3057 				inc.inc6_faddr = *dst;
3058 				if (in6_setscope(&inc.inc6_faddr,
3059 					m->m_pkthdr.rcvif, NULL))
3060 					goto out;
3061 				/*
3062 				 * Only process the offered MTU if it
3063 				 * is smaller than the current one.
3064 				 */
3065 				if (mtu < tp->t_maxseg +
3066 				    sizeof (struct tcphdr) +
3067 				    sizeof (struct ip6_hdr)) {
3068 					tcp_hc_updatemtu(&inc, mtu);
3069 					tcp_mtudisc(inp, mtu);
3070 					ICMP6STAT_INC(icp6s_pmtuchg);
3071 				}
3072 			} else
3073 				inp = (*notify)(inp, errno);
3074 		}
3075 	} else {
3076 		bzero(&inc, sizeof(inc));
3077 		inc.inc_fibnum = M_GETFIB(m);
3078 		inc.inc_flags |= INC_ISIPV6;
3079 		inc.inc_fport = t_ports.th_dport;
3080 		inc.inc_lport = t_ports.th_sport;
3081 		inc.inc6_faddr = *dst;
3082 		inc.inc6_laddr = ip6->ip6_src;
3083 		syncache_unreach(&inc, icmp_tcp_seq, port);
3084 	}
3085 out:
3086 	if (inp != NULL)
3087 		INP_WUNLOCK(inp);
3088 }
3089 
3090 static void
3091 tcp6_ctlinput(struct ip6ctlparam *ctl)
3092 {
3093 	tcp6_ctlinput_with_port(ctl, htons(0));
3094 }
3095 
3096 static void
3097 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3098 {
3099 	struct ip6ctlparam *ip6cp = param.ip6cp;
3100 	struct mbuf *m;
3101 	struct udphdr *udp;
3102 	uint16_t port;
3103 
3104 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3105 	if (m == NULL) {
3106 		return;
3107 	}
3108 	udp = mtod(m, struct udphdr *);
3109 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3110 		return;
3111 	}
3112 	port = udp->uh_dport;
3113 	m_adj(m, sizeof(struct udphdr));
3114 	if ((m->m_flags & M_PKTHDR) == 0) {
3115 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3116 	}
3117 	/* Now call in to the normal handling code */
3118 	tcp6_ctlinput_with_port(ip6cp, port);
3119 }
3120 
3121 #endif /* INET6 */
3122 
3123 static uint32_t
3124 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3125 {
3126 	SIPHASH_CTX ctx;
3127 	uint32_t hash[2];
3128 
3129 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3130 	    ("%s: keylen %u too short ", __func__, len));
3131 	SipHash24_Init(&ctx);
3132 	SipHash_SetKey(&ctx, (uint8_t *)key);
3133 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3134 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3135 	switch (inc->inc_flags & INC_ISIPV6) {
3136 #ifdef INET
3137 	case 0:
3138 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3139 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3140 		break;
3141 #endif
3142 #ifdef INET6
3143 	case INC_ISIPV6:
3144 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3145 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3146 		break;
3147 #endif
3148 	}
3149 	SipHash_Final((uint8_t *)hash, &ctx);
3150 
3151 	return (hash[0] ^ hash[1]);
3152 }
3153 
3154 uint32_t
3155 tcp_new_ts_offset(struct in_conninfo *inc)
3156 {
3157 	struct in_conninfo inc_store, *local_inc;
3158 
3159 	if (!V_tcp_ts_offset_per_conn) {
3160 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3161 		inc_store.inc_lport = 0;
3162 		inc_store.inc_fport = 0;
3163 		local_inc = &inc_store;
3164 	} else {
3165 		local_inc = inc;
3166 	}
3167 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3168 	    sizeof(V_ts_offset_secret)));
3169 }
3170 
3171 /*
3172  * Following is where TCP initial sequence number generation occurs.
3173  *
3174  * There are two places where we must use initial sequence numbers:
3175  * 1.  In SYN-ACK packets.
3176  * 2.  In SYN packets.
3177  *
3178  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3179  * tcp_syncache.c for details.
3180  *
3181  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3182  * depends on this property.  In addition, these ISNs should be
3183  * unguessable so as to prevent connection hijacking.  To satisfy
3184  * the requirements of this situation, the algorithm outlined in
3185  * RFC 1948 is used, with only small modifications.
3186  *
3187  * Implementation details:
3188  *
3189  * Time is based off the system timer, and is corrected so that it
3190  * increases by one megabyte per second.  This allows for proper
3191  * recycling on high speed LANs while still leaving over an hour
3192  * before rollover.
3193  *
3194  * As reading the *exact* system time is too expensive to be done
3195  * whenever setting up a TCP connection, we increment the time
3196  * offset in two ways.  First, a small random positive increment
3197  * is added to isn_offset for each connection that is set up.
3198  * Second, the function tcp_isn_tick fires once per clock tick
3199  * and increments isn_offset as necessary so that sequence numbers
3200  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3201  * random positive increments serve only to ensure that the same
3202  * exact sequence number is never sent out twice (as could otherwise
3203  * happen when a port is recycled in less than the system tick
3204  * interval.)
3205  *
3206  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3207  * between seeding of isn_secret.  This is normally set to zero,
3208  * as reseeding should not be necessary.
3209  *
3210  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3211  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3212  * general, this means holding an exclusive (write) lock.
3213  */
3214 
3215 #define ISN_BYTES_PER_SECOND 1048576
3216 #define ISN_STATIC_INCREMENT 4096
3217 #define ISN_RANDOM_INCREMENT (4096 - 1)
3218 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3219 
3220 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3221 VNET_DEFINE_STATIC(int, isn_last);
3222 VNET_DEFINE_STATIC(int, isn_last_reseed);
3223 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3224 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3225 
3226 #define	V_isn_secret			VNET(isn_secret)
3227 #define	V_isn_last			VNET(isn_last)
3228 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3229 #define	V_isn_offset			VNET(isn_offset)
3230 #define	V_isn_offset_old		VNET(isn_offset_old)
3231 
3232 tcp_seq
3233 tcp_new_isn(struct in_conninfo *inc)
3234 {
3235 	tcp_seq new_isn;
3236 	u_int32_t projected_offset;
3237 
3238 	ISN_LOCK();
3239 	/* Seed if this is the first use, reseed if requested. */
3240 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3241 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3242 		< (u_int)ticks))) {
3243 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3244 		V_isn_last_reseed = ticks;
3245 	}
3246 
3247 	/* Compute the hash and return the ISN. */
3248 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3249 	    sizeof(V_isn_secret));
3250 	V_isn_offset += ISN_STATIC_INCREMENT +
3251 		(arc4random() & ISN_RANDOM_INCREMENT);
3252 	if (ticks != V_isn_last) {
3253 		projected_offset = V_isn_offset_old +
3254 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3255 		if (SEQ_GT(projected_offset, V_isn_offset))
3256 			V_isn_offset = projected_offset;
3257 		V_isn_offset_old = V_isn_offset;
3258 		V_isn_last = ticks;
3259 	}
3260 	new_isn += V_isn_offset;
3261 	ISN_UNLOCK();
3262 	return (new_isn);
3263 }
3264 
3265 /*
3266  * When a specific ICMP unreachable message is received and the
3267  * connection state is SYN-SENT, drop the connection.  This behavior
3268  * is controlled by the icmp_may_rst sysctl.
3269  */
3270 static struct inpcb *
3271 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3272 {
3273 	struct tcpcb *tp;
3274 
3275 	NET_EPOCH_ASSERT();
3276 	INP_WLOCK_ASSERT(inp);
3277 
3278 	tp = intotcpcb(inp);
3279 	if (tp->t_state != TCPS_SYN_SENT)
3280 		return (inp);
3281 
3282 	if (IS_FASTOPEN(tp->t_flags))
3283 		tcp_fastopen_disable_path(tp);
3284 
3285 	tp = tcp_drop(tp, errno);
3286 	if (tp != NULL)
3287 		return (inp);
3288 	else
3289 		return (NULL);
3290 }
3291 
3292 /*
3293  * When `need fragmentation' ICMP is received, update our idea of the MSS
3294  * based on the new value. Also nudge TCP to send something, since we
3295  * know the packet we just sent was dropped.
3296  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3297  */
3298 static struct inpcb *
3299 tcp_mtudisc_notify(struct inpcb *inp, int error)
3300 {
3301 
3302 	return (tcp_mtudisc(inp, -1));
3303 }
3304 
3305 static struct inpcb *
3306 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3307 {
3308 	struct tcpcb *tp;
3309 	struct socket *so;
3310 
3311 	INP_WLOCK_ASSERT(inp);
3312 
3313 	tp = intotcpcb(inp);
3314 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3315 
3316 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3317 
3318 	so = inp->inp_socket;
3319 	SOCKBUF_LOCK(&so->so_snd);
3320 	/* If the mss is larger than the socket buffer, decrease the mss. */
3321 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3322 		tp->t_maxseg = so->so_snd.sb_hiwat;
3323 	SOCKBUF_UNLOCK(&so->so_snd);
3324 
3325 	TCPSTAT_INC(tcps_mturesent);
3326 	tp->t_rtttime = 0;
3327 	tp->snd_nxt = tp->snd_una;
3328 	tcp_free_sackholes(tp);
3329 	tp->snd_recover = tp->snd_max;
3330 	if (tp->t_flags & TF_SACK_PERMIT)
3331 		EXIT_FASTRECOVERY(tp->t_flags);
3332 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3333 		/*
3334 		 * Conceptually the snd_nxt setting
3335 		 * and freeing sack holes should
3336 		 * be done by the default stacks
3337 		 * own tfb_tcp_mtu_chg().
3338 		 */
3339 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3340 	}
3341 	if (tcp_output(tp) < 0)
3342 		return (NULL);
3343 	else
3344 		return (inp);
3345 }
3346 
3347 #ifdef INET
3348 /*
3349  * Look-up the routing entry to the peer of this inpcb.  If no route
3350  * is found and it cannot be allocated, then return 0.  This routine
3351  * is called by TCP routines that access the rmx structure and by
3352  * tcp_mss_update to get the peer/interface MTU.
3353  */
3354 uint32_t
3355 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3356 {
3357 	struct nhop_object *nh;
3358 	struct ifnet *ifp;
3359 	uint32_t maxmtu = 0;
3360 
3361 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3362 
3363 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3364 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3365 		if (nh == NULL)
3366 			return (0);
3367 
3368 		ifp = nh->nh_ifp;
3369 		maxmtu = nh->nh_mtu;
3370 
3371 		/* Report additional interface capabilities. */
3372 		if (cap != NULL) {
3373 			if (ifp->if_capenable & IFCAP_TSO4 &&
3374 			    ifp->if_hwassist & CSUM_TSO) {
3375 				cap->ifcap |= CSUM_TSO;
3376 				cap->tsomax = ifp->if_hw_tsomax;
3377 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3378 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3379 			}
3380 		}
3381 	}
3382 	return (maxmtu);
3383 }
3384 #endif /* INET */
3385 
3386 #ifdef INET6
3387 uint32_t
3388 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3389 {
3390 	struct nhop_object *nh;
3391 	struct in6_addr dst6;
3392 	uint32_t scopeid;
3393 	struct ifnet *ifp;
3394 	uint32_t maxmtu = 0;
3395 
3396 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3397 
3398 	if (inc->inc_flags & INC_IPV6MINMTU)
3399 		return (IPV6_MMTU);
3400 
3401 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3402 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3403 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3404 		if (nh == NULL)
3405 			return (0);
3406 
3407 		ifp = nh->nh_ifp;
3408 		maxmtu = nh->nh_mtu;
3409 
3410 		/* Report additional interface capabilities. */
3411 		if (cap != NULL) {
3412 			if (ifp->if_capenable & IFCAP_TSO6 &&
3413 			    ifp->if_hwassist & CSUM_TSO) {
3414 				cap->ifcap |= CSUM_TSO;
3415 				cap->tsomax = ifp->if_hw_tsomax;
3416 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3417 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3418 			}
3419 		}
3420 	}
3421 
3422 	return (maxmtu);
3423 }
3424 
3425 /*
3426  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3427  *
3428  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3429  * The right place to do that is ip6_setpktopt() that has just been
3430  * executed.  By the way it just filled ip6po_minmtu for us.
3431  */
3432 void
3433 tcp6_use_min_mtu(struct tcpcb *tp)
3434 {
3435 	struct inpcb *inp = tptoinpcb(tp);
3436 
3437 	INP_WLOCK_ASSERT(inp);
3438 	/*
3439 	 * In case of the IPV6_USE_MIN_MTU socket
3440 	 * option, the INC_IPV6MINMTU flag to announce
3441 	 * a corresponding MSS during the initial
3442 	 * handshake.  If the TCP connection is not in
3443 	 * the front states, just reduce the MSS being
3444 	 * used.  This avoids the sending of TCP
3445 	 * segments which will be fragmented at the
3446 	 * IPv6 layer.
3447 	 */
3448 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3449 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3450 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3451 		struct ip6_pktopts *opt;
3452 
3453 		opt = inp->in6p_outputopts;
3454 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3455 		    tp->t_maxseg > TCP6_MSS)
3456 			tp->t_maxseg = TCP6_MSS;
3457 	}
3458 }
3459 #endif /* INET6 */
3460 
3461 /*
3462  * Calculate effective SMSS per RFC5681 definition for a given TCP
3463  * connection at its current state, taking into account SACK and etc.
3464  */
3465 u_int
3466 tcp_maxseg(const struct tcpcb *tp)
3467 {
3468 	u_int optlen;
3469 
3470 	if (tp->t_flags & TF_NOOPT)
3471 		return (tp->t_maxseg);
3472 
3473 	/*
3474 	 * Here we have a simplified code from tcp_addoptions(),
3475 	 * without a proper loop, and having most of paddings hardcoded.
3476 	 * We might make mistakes with padding here in some edge cases,
3477 	 * but this is harmless, since result of tcp_maxseg() is used
3478 	 * only in cwnd and ssthresh estimations.
3479 	 */
3480 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3481 		if (tp->t_flags & TF_RCVD_TSTMP)
3482 			optlen = TCPOLEN_TSTAMP_APPA;
3483 		else
3484 			optlen = 0;
3485 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3486 		if (tp->t_flags & TF_SIGNATURE)
3487 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3488 #endif
3489 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3490 			optlen += TCPOLEN_SACKHDR;
3491 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3492 			optlen = PADTCPOLEN(optlen);
3493 		}
3494 	} else {
3495 		if (tp->t_flags & TF_REQ_TSTMP)
3496 			optlen = TCPOLEN_TSTAMP_APPA;
3497 		else
3498 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3499 		if (tp->t_flags & TF_REQ_SCALE)
3500 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3502 		if (tp->t_flags & TF_SIGNATURE)
3503 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3504 #endif
3505 		if (tp->t_flags & TF_SACK_PERMIT)
3506 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3507 	}
3508 #undef PAD
3509 	optlen = min(optlen, TCP_MAXOLEN);
3510 	return (tp->t_maxseg - optlen);
3511 }
3512 
3513 
3514 u_int
3515 tcp_fixed_maxseg(const struct tcpcb *tp)
3516 {
3517 	int optlen;
3518 
3519 	if (tp->t_flags & TF_NOOPT)
3520 		return (tp->t_maxseg);
3521 
3522 	/*
3523 	 * Here we have a simplified code from tcp_addoptions(),
3524 	 * without a proper loop, and having most of paddings hardcoded.
3525 	 * We only consider fixed options that we would send every
3526 	 * time I.e. SACK is not considered. This is important
3527 	 * for cc modules to figure out what the modulo of the
3528 	 * cwnd should be.
3529 	 */
3530 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3531 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3532 		if (tp->t_flags & TF_RCVD_TSTMP)
3533 			optlen = TCPOLEN_TSTAMP_APPA;
3534 		else
3535 			optlen = 0;
3536 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3537 		if (tp->t_flags & TF_SIGNATURE)
3538 			optlen += PAD(TCPOLEN_SIGNATURE);
3539 #endif
3540 	} else {
3541 		if (tp->t_flags & TF_REQ_TSTMP)
3542 			optlen = TCPOLEN_TSTAMP_APPA;
3543 		else
3544 			optlen = PAD(TCPOLEN_MAXSEG);
3545 		if (tp->t_flags & TF_REQ_SCALE)
3546 			optlen += PAD(TCPOLEN_WINDOW);
3547 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3548 		if (tp->t_flags & TF_SIGNATURE)
3549 			optlen += PAD(TCPOLEN_SIGNATURE);
3550 #endif
3551 		if (tp->t_flags & TF_SACK_PERMIT)
3552 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3553 	}
3554 #undef PAD
3555 	optlen = min(optlen, TCP_MAXOLEN);
3556 	return (tp->t_maxseg - optlen);
3557 }
3558 
3559 
3560 
3561 static int
3562 sysctl_drop(SYSCTL_HANDLER_ARGS)
3563 {
3564 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3565 	struct sockaddr_storage addrs[2];
3566 	struct inpcb *inp;
3567 	struct tcpcb *tp;
3568 #ifdef INET
3569 	struct sockaddr_in *fin = NULL, *lin = NULL;
3570 #endif
3571 	struct epoch_tracker et;
3572 #ifdef INET6
3573 	struct sockaddr_in6 *fin6, *lin6;
3574 #endif
3575 	int error;
3576 
3577 	inp = NULL;
3578 #ifdef INET6
3579 	fin6 = lin6 = NULL;
3580 #endif
3581 	error = 0;
3582 
3583 	if (req->oldptr != NULL || req->oldlen != 0)
3584 		return (EINVAL);
3585 	if (req->newptr == NULL)
3586 		return (EPERM);
3587 	if (req->newlen < sizeof(addrs))
3588 		return (ENOMEM);
3589 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3590 	if (error)
3591 		return (error);
3592 
3593 	switch (addrs[0].ss_family) {
3594 #ifdef INET6
3595 	case AF_INET6:
3596 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3597 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3598 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3599 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3600 			return (EINVAL);
3601 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3602 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3603 				return (EINVAL);
3604 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3605 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3606 #ifdef INET
3607 			fin = (struct sockaddr_in *)&addrs[0];
3608 			lin = (struct sockaddr_in *)&addrs[1];
3609 #endif
3610 			break;
3611 		}
3612 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3613 		if (error)
3614 			return (error);
3615 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3616 		if (error)
3617 			return (error);
3618 		break;
3619 #endif
3620 #ifdef INET
3621 	case AF_INET:
3622 		fin = (struct sockaddr_in *)&addrs[0];
3623 		lin = (struct sockaddr_in *)&addrs[1];
3624 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3625 		    lin->sin_len != sizeof(struct sockaddr_in))
3626 			return (EINVAL);
3627 		break;
3628 #endif
3629 	default:
3630 		return (EINVAL);
3631 	}
3632 	NET_EPOCH_ENTER(et);
3633 	switch (addrs[0].ss_family) {
3634 #ifdef INET6
3635 	case AF_INET6:
3636 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3637 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3638 		    INPLOOKUP_WLOCKPCB, NULL);
3639 		break;
3640 #endif
3641 #ifdef INET
3642 	case AF_INET:
3643 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3644 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3645 		break;
3646 #endif
3647 	}
3648 	if (inp != NULL) {
3649 		if (!SOLISTENING(inp->inp_socket)) {
3650 			tp = intotcpcb(inp);
3651 			tp = tcp_drop(tp, ECONNABORTED);
3652 			if (tp != NULL)
3653 				INP_WUNLOCK(inp);
3654 		} else
3655 			INP_WUNLOCK(inp);
3656 	} else
3657 		error = ESRCH;
3658 	NET_EPOCH_EXIT(et);
3659 	return (error);
3660 }
3661 
3662 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3663     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3664     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3665     "Drop TCP connection");
3666 
3667 static int
3668 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3669 {
3670 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3671 	    &tcp_ctloutput_set));
3672 }
3673 
3674 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3675     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3676     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3677     "Set socket option for TCP endpoint");
3678 
3679 #ifdef KERN_TLS
3680 static int
3681 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3682 {
3683 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3684 	struct sockaddr_storage addrs[2];
3685 	struct inpcb *inp;
3686 #ifdef INET
3687 	struct sockaddr_in *fin = NULL, *lin = NULL;
3688 #endif
3689 	struct epoch_tracker et;
3690 #ifdef INET6
3691 	struct sockaddr_in6 *fin6, *lin6;
3692 #endif
3693 	int error;
3694 
3695 	inp = NULL;
3696 #ifdef INET6
3697 	fin6 = lin6 = NULL;
3698 #endif
3699 	error = 0;
3700 
3701 	if (req->oldptr != NULL || req->oldlen != 0)
3702 		return (EINVAL);
3703 	if (req->newptr == NULL)
3704 		return (EPERM);
3705 	if (req->newlen < sizeof(addrs))
3706 		return (ENOMEM);
3707 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3708 	if (error)
3709 		return (error);
3710 
3711 	switch (addrs[0].ss_family) {
3712 #ifdef INET6
3713 	case AF_INET6:
3714 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3715 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3716 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3717 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3718 			return (EINVAL);
3719 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3720 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3721 				return (EINVAL);
3722 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3723 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3724 #ifdef INET
3725 			fin = (struct sockaddr_in *)&addrs[0];
3726 			lin = (struct sockaddr_in *)&addrs[1];
3727 #endif
3728 			break;
3729 		}
3730 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3731 		if (error)
3732 			return (error);
3733 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3734 		if (error)
3735 			return (error);
3736 		break;
3737 #endif
3738 #ifdef INET
3739 	case AF_INET:
3740 		fin = (struct sockaddr_in *)&addrs[0];
3741 		lin = (struct sockaddr_in *)&addrs[1];
3742 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3743 		    lin->sin_len != sizeof(struct sockaddr_in))
3744 			return (EINVAL);
3745 		break;
3746 #endif
3747 	default:
3748 		return (EINVAL);
3749 	}
3750 	NET_EPOCH_ENTER(et);
3751 	switch (addrs[0].ss_family) {
3752 #ifdef INET6
3753 	case AF_INET6:
3754 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3755 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3756 		    INPLOOKUP_WLOCKPCB, NULL);
3757 		break;
3758 #endif
3759 #ifdef INET
3760 	case AF_INET:
3761 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3762 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3763 		break;
3764 #endif
3765 	}
3766 	NET_EPOCH_EXIT(et);
3767 	if (inp != NULL) {
3768 		struct socket *so;
3769 
3770 		so = inp->inp_socket;
3771 		soref(so);
3772 		error = ktls_set_tx_mode(so,
3773 		    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3774 		INP_WUNLOCK(inp);
3775 		sorele(so);
3776 	} else
3777 		error = ESRCH;
3778 	return (error);
3779 }
3780 
3781 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3782     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3783     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3784     "Switch TCP connection to SW TLS");
3785 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3786     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3787     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3788     "Switch TCP connection to ifnet TLS");
3789 #endif
3790 
3791 /*
3792  * Generate a standardized TCP log line for use throughout the
3793  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3794  * allow use in the interrupt context.
3795  *
3796  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3797  * NB: The function may return NULL if memory allocation failed.
3798  *
3799  * Due to header inclusion and ordering limitations the struct ip
3800  * and ip6_hdr pointers have to be passed as void pointers.
3801  */
3802 char *
3803 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3804     const void *ip6hdr)
3805 {
3806 
3807 	/* Is logging enabled? */
3808 	if (V_tcp_log_in_vain == 0)
3809 		return (NULL);
3810 
3811 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3812 }
3813 
3814 char *
3815 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3816     const void *ip6hdr)
3817 {
3818 
3819 	/* Is logging enabled? */
3820 	if (tcp_log_debug == 0)
3821 		return (NULL);
3822 
3823 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3824 }
3825 
3826 static char *
3827 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3828     const void *ip6hdr)
3829 {
3830 	char *s, *sp;
3831 	size_t size;
3832 #ifdef INET
3833 	const struct ip *ip = (const struct ip *)ip4hdr;
3834 #endif
3835 #ifdef INET6
3836 	const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3837 #endif /* INET6 */
3838 
3839 	/*
3840 	 * The log line looks like this:
3841 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3842 	 */
3843 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3844 	    sizeof(PRINT_TH_FLAGS) + 1 +
3845 #ifdef INET6
3846 	    2 * INET6_ADDRSTRLEN;
3847 #else
3848 	    2 * INET_ADDRSTRLEN;
3849 #endif /* INET6 */
3850 
3851 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3852 	if (s == NULL)
3853 		return (NULL);
3854 
3855 	strcat(s, "TCP: [");
3856 	sp = s + strlen(s);
3857 
3858 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3859 		inet_ntoa_r(inc->inc_faddr, sp);
3860 		sp = s + strlen(s);
3861 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3862 		sp = s + strlen(s);
3863 		inet_ntoa_r(inc->inc_laddr, sp);
3864 		sp = s + strlen(s);
3865 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3866 #ifdef INET6
3867 	} else if (inc) {
3868 		ip6_sprintf(sp, &inc->inc6_faddr);
3869 		sp = s + strlen(s);
3870 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3871 		sp = s + strlen(s);
3872 		ip6_sprintf(sp, &inc->inc6_laddr);
3873 		sp = s + strlen(s);
3874 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3875 	} else if (ip6 && th) {
3876 		ip6_sprintf(sp, &ip6->ip6_src);
3877 		sp = s + strlen(s);
3878 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3879 		sp = s + strlen(s);
3880 		ip6_sprintf(sp, &ip6->ip6_dst);
3881 		sp = s + strlen(s);
3882 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3883 #endif /* INET6 */
3884 #ifdef INET
3885 	} else if (ip && th) {
3886 		inet_ntoa_r(ip->ip_src, sp);
3887 		sp = s + strlen(s);
3888 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3889 		sp = s + strlen(s);
3890 		inet_ntoa_r(ip->ip_dst, sp);
3891 		sp = s + strlen(s);
3892 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3893 #endif /* INET */
3894 	} else {
3895 		free(s, M_TCPLOG);
3896 		return (NULL);
3897 	}
3898 	sp = s + strlen(s);
3899 	if (th)
3900 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3901 	if (*(s + size - 1) != '\0')
3902 		panic("%s: string too long", __func__);
3903 	return (s);
3904 }
3905 
3906 /*
3907  * A subroutine which makes it easy to track TCP state changes with DTrace.
3908  * This function shouldn't be called for t_state initializations that don't
3909  * correspond to actual TCP state transitions.
3910  */
3911 void
3912 tcp_state_change(struct tcpcb *tp, int newstate)
3913 {
3914 #if defined(KDTRACE_HOOKS)
3915 	int pstate = tp->t_state;
3916 #endif
3917 
3918 	TCPSTATES_DEC(tp->t_state);
3919 	TCPSTATES_INC(newstate);
3920 	tp->t_state = newstate;
3921 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3922 }
3923 
3924 /*
3925  * Create an external-format (``xtcpcb'') structure using the information in
3926  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3927  * reduce the spew of irrelevant information over this interface, to isolate
3928  * user code from changes in the kernel structure, and potentially to provide
3929  * information-hiding if we decide that some of this information should be
3930  * hidden from users.
3931  */
3932 void
3933 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3934 {
3935 	struct tcpcb *tp = intotcpcb(inp);
3936 	sbintime_t now;
3937 
3938 	bzero(xt, sizeof(*xt));
3939 	xt->t_state = tp->t_state;
3940 	xt->t_logstate = tcp_get_bblog_state(tp);
3941 	xt->t_flags = tp->t_flags;
3942 	xt->t_sndzerowin = tp->t_sndzerowin;
3943 	xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3944 	xt->t_rcvoopack = tp->t_rcvoopack;
3945 	xt->t_rcv_wnd = tp->rcv_wnd;
3946 	xt->t_snd_wnd = tp->snd_wnd;
3947 	xt->t_snd_cwnd = tp->snd_cwnd;
3948 	xt->t_snd_ssthresh = tp->snd_ssthresh;
3949 	xt->t_dsack_bytes = tp->t_dsack_bytes;
3950 	xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3951 	xt->t_dsack_pack = tp->t_dsack_pack;
3952 	xt->t_maxseg = tp->t_maxseg;
3953 	xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3954 		     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3955 
3956 	now = getsbinuptime();
3957 #define	COPYTIMER(which,where)	do {					\
3958 	if (tp->t_timers[which] != SBT_MAX)				\
3959 		xt->where = (tp->t_timers[which] - now) / SBT_1MS;	\
3960 	else								\
3961 		xt->where = 0;						\
3962 } while (0)
3963 	COPYTIMER(TT_DELACK, tt_delack);
3964 	COPYTIMER(TT_REXMT, tt_rexmt);
3965 	COPYTIMER(TT_PERSIST, tt_persist);
3966 	COPYTIMER(TT_KEEP, tt_keep);
3967 	COPYTIMER(TT_2MSL, tt_2msl);
3968 #undef COPYTIMER
3969 	xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3970 
3971 	xt->xt_encaps_port = tp->t_port;
3972 	bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3973 	    TCP_FUNCTION_NAME_LEN_MAX);
3974 	bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3975 #ifdef TCP_BLACKBOX
3976 	(void)tcp_log_get_id(tp, xt->xt_logid);
3977 #endif
3978 
3979 	xt->xt_len = sizeof(struct xtcpcb);
3980 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3981 	/*
3982 	 * TCP doesn't use inp_ppcb pointer, we embed inpcb into tcpcb.
3983 	 * Fixup the pointer that in_pcbtoxinpcb() has set.  When printing
3984 	 * TCP netstat(1) used to use this pointer, so this fixup needs to
3985 	 * stay for stable/14.
3986 	 */
3987 	xt->xt_inp.inp_ppcb = (uintptr_t)tp;
3988 }
3989 
3990 void
3991 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3992 {
3993 	uint32_t bit, i;
3994 
3995 	if ((tp == NULL) ||
3996 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
3997 	    (status == 0)) {
3998 		/* Invalid */
3999 		return;
4000 	}
4001 	if (status > (sizeof(uint32_t) * 8)) {
4002 		/* Should this be a KASSERT? */
4003 		return;
4004 	}
4005 	bit = 1U << (status - 1);
4006 	if (bit & tp->t_end_info_status) {
4007 		/* already logged */
4008 		return;
4009 	}
4010 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4011 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4012 			tp->t_end_info_bytes[i] = status;
4013 			tp->t_end_info_status |= bit;
4014 			break;
4015 		}
4016 	}
4017 }
4018 
4019 int
4020 tcp_can_enable_pacing(void)
4021 {
4022 
4023 	if ((tcp_pacing_limit == -1) ||
4024 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4025 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4026 		shadow_num_connections = number_of_tcp_connections_pacing;
4027 		return (1);
4028 	} else {
4029 		counter_u64_add(tcp_pacing_failures, 1);
4030 		return (0);
4031 	}
4032 }
4033 
4034 static uint8_t tcp_pacing_warning = 0;
4035 
4036 void
4037 tcp_decrement_paced_conn(void)
4038 {
4039 	uint32_t ret;
4040 
4041 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4042 	shadow_num_connections = number_of_tcp_connections_pacing;
4043 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4044 	if (ret == 0) {
4045 		if (tcp_pacing_limit != -1) {
4046 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4047 			tcp_pacing_limit = 0;
4048 		} else if (tcp_pacing_warning == 0) {
4049 			printf("Warning pacing count is invalid, invalid decrement\n");
4050 			tcp_pacing_warning = 1;
4051 		}
4052 	}
4053 }
4054 
4055 static void
4056 tcp_default_switch_failed(struct tcpcb *tp)
4057 {
4058 	/*
4059 	 * If a switch fails we only need to
4060 	 * care about two things:
4061 	 * a) The t_flags2
4062 	 * and
4063 	 * b) The timer granularity.
4064 	 * Timeouts, at least for now, don't use the
4065 	 * old callout system in the other stacks so
4066 	 * those are hopefully safe.
4067 	 */
4068 	tcp_lro_features_off(tp);
4069 	tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
4070 }
4071 
4072 #ifdef TCP_ACCOUNTING
4073 int
4074 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss)
4075 {
4076 	if (SEQ_LT(th->th_ack, tp->snd_una)) {
4077 		/* Do we have a SACK? */
4078 		if (to->to_flags & TOF_SACK) {
4079 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4080 				tp->tcp_cnt_counters[ACK_SACK]++;
4081 			}
4082 			return (ACK_SACK);
4083 		} else {
4084 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4085 				tp->tcp_cnt_counters[ACK_BEHIND]++;
4086 			}
4087 			return (ACK_BEHIND);
4088 		}
4089 	} else if (th->th_ack == tp->snd_una) {
4090 		/* Do we have a SACK? */
4091 		if (to->to_flags & TOF_SACK) {
4092 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4093 				tp->tcp_cnt_counters[ACK_SACK]++;
4094 			}
4095 			return (ACK_SACK);
4096 		} else if (tiwin != tp->snd_wnd) {
4097 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4098 				tp->tcp_cnt_counters[ACK_RWND]++;
4099 			}
4100 			return (ACK_RWND);
4101 		} else {
4102 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4103 				tp->tcp_cnt_counters[ACK_DUPACK]++;
4104 			}
4105 			return (ACK_DUPACK);
4106 		}
4107 	} else {
4108 		if (!SEQ_GT(th->th_ack, tp->snd_max)) {
4109 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4110 				tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss);
4111 			}
4112 		}
4113 		if (to->to_flags & TOF_SACK) {
4114 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4115 				tp->tcp_cnt_counters[ACK_CUMACK_SACK]++;
4116 			}
4117 			return (ACK_CUMACK_SACK);
4118 		} else {
4119 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4120 				tp->tcp_cnt_counters[ACK_CUMACK]++;
4121 			}
4122 			return (ACK_CUMACK);
4123 		}
4124 	}
4125 }
4126 #endif
4127 
4128 void
4129 tcp_change_time_units(struct tcpcb *tp, int granularity)
4130 {
4131 	if (tp->t_tmr_granularity == granularity) {
4132 		/* We are there */
4133 		return;
4134 	}
4135 	if (granularity == TCP_TMR_GRANULARITY_USEC) {
4136 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS),
4137 			("Granularity is not TICKS its %u in tp:%p",
4138 			 tp->t_tmr_granularity, tp));
4139 		tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow);
4140 		if (tp->t_srtt > 1) {
4141 			uint32_t val, frac;
4142 
4143 			val = tp->t_srtt >> TCP_RTT_SHIFT;
4144 			frac = tp->t_srtt & 0x1f;
4145 			tp->t_srtt = TICKS_2_USEC(val);
4146 			/*
4147 			 * frac is the fractional part of the srtt (if any)
4148 			 * but its in ticks and every bit represents
4149 			 * 1/32nd of a hz.
4150 			 */
4151 			if (frac) {
4152 				if (hz == 1000) {
4153 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4154 				} else {
4155 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4156 				}
4157 				tp->t_srtt += frac;
4158 			}
4159 		}
4160 		if (tp->t_rttvar) {
4161 			uint32_t val, frac;
4162 
4163 			val = tp->t_rttvar >> TCP_RTTVAR_SHIFT;
4164 			frac = tp->t_rttvar & 0x1f;
4165 			tp->t_rttvar = TICKS_2_USEC(val);
4166 			/*
4167 			 * frac is the fractional part of the srtt (if any)
4168 			 * but its in ticks and every bit represents
4169 			 * 1/32nd of a hz.
4170 			 */
4171 			if (frac) {
4172 				if (hz == 1000) {
4173 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4174 				} else {
4175 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4176 				}
4177 				tp->t_rttvar += frac;
4178 			}
4179 		}
4180 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC;
4181 	} else if (granularity == TCP_TMR_GRANULARITY_TICKS) {
4182 		/* Convert back to ticks, with  */
4183 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC),
4184 			("Granularity is not USEC its %u in tp:%p",
4185 			 tp->t_tmr_granularity, tp));
4186 		if (tp->t_srtt > 1) {
4187 			uint32_t val, frac;
4188 
4189 			val = USEC_2_TICKS(tp->t_srtt);
4190 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4191 			tp->t_srtt = val << TCP_RTT_SHIFT;
4192 			/*
4193 			 * frac is the fractional part here is left
4194 			 * over from converting to hz and shifting.
4195 			 * We need to convert this to the 5 bit
4196 			 * remainder.
4197 			 */
4198 			if (frac) {
4199 				if (hz == 1000) {
4200 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4201 				} else {
4202 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4203 				}
4204 				tp->t_srtt += frac;
4205 			}
4206 		}
4207 		if (tp->t_rttvar) {
4208 			uint32_t val, frac;
4209 
4210 			val = USEC_2_TICKS(tp->t_rttvar);
4211 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4212 			tp->t_rttvar = val <<  TCP_RTTVAR_SHIFT;
4213 			/*
4214 			 * frac is the fractional part here is left
4215 			 * over from converting to hz and shifting.
4216 			 * We need to convert this to the 5 bit
4217 			 * remainder.
4218 			 */
4219 			if (frac) {
4220 				if (hz == 1000) {
4221 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4222 				} else {
4223 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4224 				}
4225 				tp->t_rttvar += frac;
4226 			}
4227 		}
4228 		tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow);
4229 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
4230 	}
4231 #ifdef INVARIANTS
4232 	else {
4233 		panic("Unknown granularity:%d tp:%p",
4234 		      granularity, tp);
4235 	}
4236 #endif
4237 }
4238 
4239 void
4240 tcp_handle_orphaned_packets(struct tcpcb *tp)
4241 {
4242 	struct mbuf *save, *m, *prev;
4243 	/*
4244 	 * Called when a stack switch is occuring from the fini()
4245 	 * of the old stack. We assue the init() as already been
4246 	 * run of the new stack and it has set the t_flags2 to
4247 	 * what it supports. This function will then deal with any
4248 	 * differences i.e. cleanup packets that maybe queued that
4249 	 * the newstack does not support.
4250 	 */
4251 
4252 	if (tp->t_flags2 & TF2_MBUF_L_ACKS)
4253 		return;
4254 	if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 &&
4255 	    !STAILQ_EMPTY(&tp->t_inqueue)) {
4256 		/*
4257 		 * It is unsafe to process the packets since a
4258 		 * reset may be lurking in them (its rare but it
4259 		 * can occur). If we were to find a RST, then we
4260 		 * would end up dropping the connection and the
4261 		 * INP lock, so when we return the caller (tcp_usrreq)
4262 		 * will blow up when it trys to unlock the inp.
4263 		 * This new stack does not do any fancy LRO features
4264 		 * so all we can do is toss the packets.
4265 		 */
4266 		m = STAILQ_FIRST(&tp->t_inqueue);
4267 		STAILQ_INIT(&tp->t_inqueue);
4268 		STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save)
4269 			m_freem(m);
4270 	} else {
4271 		/*
4272 		 * Here we have a stack that does mbuf queuing but
4273 		 * does not support compressed ack's. We must
4274 		 * walk all the mbufs and discard any compressed acks.
4275 		 */
4276 		STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) {
4277 			if (m->m_flags & M_ACKCMP) {
4278 				if (m == STAILQ_FIRST(&tp->t_inqueue))
4279 					STAILQ_REMOVE_HEAD(&tp->t_inqueue,
4280 					    m_stailqpkt);
4281 				else
4282 					STAILQ_REMOVE_AFTER(&tp->t_inqueue,
4283 					    prev, m_stailqpkt);
4284 				m_freem(m);
4285 			} else
4286 				prev = m;
4287 		}
4288 	}
4289 }
4290 
4291 #ifdef TCP_REQUEST_TRK
4292 uint32_t
4293 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes)
4294 {
4295 #ifdef KERN_TLS
4296 	struct ktls_session *tls;
4297 	uint32_t rec_oh, records;
4298 
4299 	tls = so->so_snd.sb_tls_info;
4300 	if (tls == NULL)
4301 	    return (0);
4302 
4303 	rec_oh = tls->params.tls_hlen + tls->params.tls_tlen;
4304 	records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len);
4305 	return (records * rec_oh);
4306 #else
4307 	return (0);
4308 #endif
4309 }
4310 
4311 extern uint32_t tcp_stale_entry_time;
4312 uint32_t tcp_stale_entry_time = 250000;
4313 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW,
4314     &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out");
4315 
4316 void
4317 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req,
4318     uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes)
4319 {
4320 	if (tcp_bblogging_on(tp)) {
4321 		union tcp_log_stackspecific log;
4322 		struct timeval tv;
4323 
4324 		memset(&log.u_bbr, 0, sizeof(log.u_bbr));
4325 #ifdef TCPHPTS
4326 		log.u_bbr.inhpts = tcp_in_hpts(tp);
4327 #endif
4328 		log.u_bbr.flex8 = val;
4329 		log.u_bbr.rttProp = req->timestamp;
4330 		log.u_bbr.delRate = req->start;
4331 		log.u_bbr.cur_del_rate = req->end;
4332 		log.u_bbr.flex1 = req->start_seq;
4333 		log.u_bbr.flex2 = req->end_seq;
4334 		log.u_bbr.flex3 = req->flags;
4335 		log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff);
4336 		log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff);
4337 		log.u_bbr.flex7 = slot;
4338 		log.u_bbr.bw_inuse = offset;
4339 		/* nbytes = flex6 | epoch */
4340 		log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff);
4341 		log.u_bbr.epoch = (nbytes & 0x00000000ffffffff);
4342 		/* cspr =  lt_epoch | pkts_out */
4343 		log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff);
4344 		log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff);
4345 		log.u_bbr.applimited = tp->t_tcpreq_closed;
4346 		log.u_bbr.applimited <<= 8;
4347 		log.u_bbr.applimited |= tp->t_tcpreq_open;
4348 		log.u_bbr.applimited <<= 8;
4349 		log.u_bbr.applimited |= tp->t_tcpreq_req;
4350 		log.u_bbr.timeStamp = tcp_get_usecs(&tv);
4351 		TCP_LOG_EVENTP(tp, NULL,
4352 		    &tptosocket(tp)->so_rcv,
4353 		    &tptosocket(tp)->so_snd,
4354 		    TCP_LOG_REQ_T, 0,
4355 		    0, &log, false, &tv);
4356 	}
4357 }
4358 
4359 void
4360 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent)
4361 {
4362 	if (tp->t_tcpreq_req > 0)
4363 		tp->t_tcpreq_req--;
4364 	if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4365 		if (tp->t_tcpreq_open > 0)
4366 			tp->t_tcpreq_open--;
4367 	} else {
4368 		if (tp->t_tcpreq_closed > 0)
4369 			tp->t_tcpreq_closed--;
4370 	}
4371 	ent->flags = TCP_TRK_TRACK_FLG_EMPTY;
4372 }
4373 
4374 static void
4375 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest)
4376 {
4377 	struct tcp_sendfile_track *ent;
4378 	uint64_t time_delta, oldest_delta;
4379 	int i, oldest, oldest_set = 0, cnt_rm = 0;
4380 
4381 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4382 		ent = &tp->t_tcpreq_info[i];
4383 		if (ent->flags != TCP_TRK_TRACK_FLG_USED) {
4384 			/*
4385 			 * We only care about closed end ranges
4386 			 * that are allocated and have no sendfile
4387 			 * ever touching them. They would be in
4388 			 * state USED.
4389 			 */
4390 			continue;
4391 		}
4392 		if (ts >= ent->localtime)
4393 			time_delta = ts - ent->localtime;
4394 		else
4395 			time_delta = 0;
4396 		if (time_delta &&
4397 		    ((oldest_delta < time_delta) || (oldest_set == 0))) {
4398 			oldest_set = 1;
4399 			oldest = i;
4400 			oldest_delta = time_delta;
4401 		}
4402 		if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) {
4403 			/*
4404 			 * No sendfile in a our time-limit
4405 			 * time to purge it.
4406 			 */
4407 			cnt_rm++;
4408 			tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4409 					      time_delta, 0);
4410 			tcp_req_free_a_slot(tp, ent);
4411 		}
4412 	}
4413 	if ((cnt_rm == 0) && rm_oldest && oldest_set) {
4414 		ent = &tp->t_tcpreq_info[oldest];
4415 		tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE,
4416 				      oldest_delta, 1);
4417 		tcp_req_free_a_slot(tp, ent);
4418 	}
4419 }
4420 
4421 int
4422 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point)
4423 {
4424 	int i, ret=0;
4425 	struct tcp_sendfile_track *ent;
4426 
4427 	/* Clean up any old closed end requests that are now completed */
4428 	if (tp->t_tcpreq_req == 0)
4429 		return(0);
4430 	if (tp->t_tcpreq_closed == 0)
4431 		return(0);
4432 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4433 		ent = &tp->t_tcpreq_info[i];
4434 		/* Skip empty ones */
4435 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4436 			continue;
4437 		/* Skip open ones */
4438 		if (ent->flags & TCP_TRK_TRACK_FLG_OPEN)
4439 			continue;
4440 		if (SEQ_GEQ(ack_point, ent->end_seq)) {
4441 			/* We are past it -- free it */
4442 			tcp_req_log_req_info(tp, ent,
4443 					      i, TCP_TRK_REQ_LOG_FREED, 0, 0);
4444 			tcp_req_free_a_slot(tp, ent);
4445 			ret++;
4446 		}
4447 	}
4448 	return (ret);
4449 }
4450 
4451 int
4452 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point)
4453 {
4454 	if (tp->t_tcpreq_req == 0)
4455 		return(-1);
4456 	if (tp->t_tcpreq_closed == 0)
4457 		return(-1);
4458 	if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4459 		return(-1);
4460 	if (SEQ_GEQ(ack_point, ent->end_seq)) {
4461 		return (1);
4462 	}
4463 	return (0);
4464 }
4465 
4466 struct tcp_sendfile_track *
4467 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip)
4468 {
4469 	/*
4470 	 * Given an ack point (th_ack) walk through our entries and
4471 	 * return the first one found that th_ack goes past the
4472 	 * end_seq.
4473 	 */
4474 	struct tcp_sendfile_track *ent;
4475 	int i;
4476 
4477 	if (tp->t_tcpreq_req == 0) {
4478 		/* none open */
4479 		return (NULL);
4480 	}
4481 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4482 		ent = &tp->t_tcpreq_info[i];
4483 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY)
4484 			continue;
4485 		if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) {
4486 			if (SEQ_GEQ(th_ack, ent->end_seq)) {
4487 				*ip = i;
4488 				return (ent);
4489 			}
4490 		}
4491 	}
4492 	return (NULL);
4493 }
4494 
4495 struct tcp_sendfile_track *
4496 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq)
4497 {
4498 	struct tcp_sendfile_track *ent;
4499 	int i;
4500 
4501 	if (tp->t_tcpreq_req == 0) {
4502 		/* none open */
4503 		return (NULL);
4504 	}
4505 	for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4506 		ent = &tp->t_tcpreq_info[i];
4507 		tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH,
4508 				      (uint64_t)seq, 0);
4509 		if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4510 			continue;
4511 		}
4512 		if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) {
4513 			/*
4514 			 * An open end request only needs to
4515 			 * match the beginning seq or be
4516 			 * all we have (once we keep going on
4517 			 * a open end request we may have a seq
4518 			 * wrap).
4519 			 */
4520 			if ((SEQ_GEQ(seq, ent->start_seq)) ||
4521 			    (tp->t_tcpreq_closed == 0))
4522 				return (ent);
4523 		} else {
4524 			/*
4525 			 * For this one we need to
4526 			 * be a bit more careful if its
4527 			 * completed at least.
4528 			 */
4529 			if ((SEQ_GEQ(seq, ent->start_seq)) &&
4530 			    (SEQ_LT(seq, ent->end_seq))) {
4531 				return (ent);
4532 			}
4533 		}
4534 	}
4535 	return (NULL);
4536 }
4537 
4538 /* Should this be in its own file tcp_req.c ? */
4539 struct tcp_sendfile_track *
4540 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups)
4541 {
4542 	struct tcp_sendfile_track *fil;
4543 	int i, allocated;
4544 
4545 	/* In case the stack does not check for completions do so now */
4546 	tcp_req_check_for_comp(tp, tp->snd_una);
4547 	/* Check for stale entries */
4548 	if (tp->t_tcpreq_req)
4549 		tcp_req_check_for_stale_entries(tp, ts,
4550 		    (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ));
4551 	/* Check to see if this is a duplicate of one not started */
4552 	if (tp->t_tcpreq_req) {
4553 		for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4554 			fil = &tp->t_tcpreq_info[i];
4555 			if (fil->flags != TCP_TRK_TRACK_FLG_USED)
4556 				continue;
4557 			if ((fil->timestamp == req->timestamp) &&
4558 			    (fil->start == req->start) &&
4559 			    ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) ||
4560 			     (fil->end == req->end))) {
4561 				/*
4562 				 * We already have this request
4563 				 * and it has not been started with sendfile.
4564 				 * This probably means the user was returned
4565 				 * a 4xx of some sort and its going to age
4566 				 * out, lets not duplicate it.
4567 				 */
4568 				return(fil);
4569 			}
4570 		}
4571 	}
4572 	/* Ok if there is no room at the inn we are in trouble */
4573 	if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) {
4574 		tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL);
4575 		for(i = 0; i < MAX_TCP_TRK_REQ; i++) {
4576 			tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i],
4577 			    i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0);
4578 		}
4579 		return (NULL);
4580 	}
4581 	for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) {
4582 		fil = &tp->t_tcpreq_info[i];
4583 		if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) {
4584 			allocated = 1;
4585 			fil->flags = TCP_TRK_TRACK_FLG_USED;
4586 			fil->timestamp = req->timestamp;
4587 			fil->localtime = ts;
4588 			fil->start = req->start;
4589 			if (req->flags & TCP_LOG_HTTPD_RANGE_END) {
4590 				fil->end = req->end;
4591 			} else {
4592 				fil->end = 0;
4593 				fil->flags |= TCP_TRK_TRACK_FLG_OPEN;
4594 			}
4595 			/*
4596 			 * We can set the min boundaries to the TCP Sequence space,
4597 			 * but it might be found to be further up when sendfile
4598 			 * actually runs on this range (if it ever does).
4599 			 */
4600 			fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc;
4601 			fil->start_seq = tp->snd_una +
4602 			    tptosocket(tp)->so_snd.sb_ccc;
4603 			fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start)));
4604 			if (tptosocket(tp)->so_snd.sb_tls_info) {
4605 				/*
4606 				 * This session is doing TLS. Take a swag guess
4607 				 * at the overhead.
4608 				 */
4609 				fil->end_seq += tcp_estimate_tls_overhead(
4610 				    tptosocket(tp), (fil->end - fil->start));
4611 			}
4612 			tp->t_tcpreq_req++;
4613 			if (fil->flags & TCP_TRK_TRACK_FLG_OPEN)
4614 				tp->t_tcpreq_open++;
4615 			else
4616 				tp->t_tcpreq_closed++;
4617 			tcp_req_log_req_info(tp, fil, i,
4618 			    TCP_TRK_REQ_LOG_NEW, 0, 0);
4619 			break;
4620 		} else
4621 			fil = NULL;
4622 	}
4623 	return (fil);
4624 }
4625 
4626 void
4627 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts)
4628 {
4629 	(void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1);
4630 }
4631 #endif
4632 
4633 void
4634 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err)
4635 {
4636 	if (tcp_bblogging_on(tp)) {
4637 		struct tcp_log_buffer *l;
4638 
4639 		l = tcp_log_event(tp, NULL,
4640 		        &tptosocket(tp)->so_rcv,
4641 		        &tptosocket(tp)->so_snd,
4642 		        TCP_LOG_SOCKET_OPT,
4643 		        err, 0, NULL, 1,
4644 		        NULL, NULL, 0, NULL);
4645 		if (l) {
4646 			l->tlb_flex1 = option_num;
4647 			l->tlb_flex2 = option_val;
4648 		}
4649 	}
4650 }
4651 
4652 uint32_t
4653 tcp_get_srtt(struct tcpcb *tp, int granularity)
4654 {
4655 	uint32_t srtt;
4656 
4657 	KASSERT(granularity == TCP_TMR_GRANULARITY_USEC ||
4658 	    granularity == TCP_TMR_GRANULARITY_TICKS,
4659 	    ("%s: called with unexpected granularity %d", __func__,
4660 	    granularity));
4661 
4662 	srtt = tp->t_srtt;
4663 
4664 	/*
4665 	 * We only support two granularities. If the stored granularity
4666 	 * does not match the granularity requested by the caller,
4667 	 * convert the stored value to the requested unit of granularity.
4668 	 */
4669 	if (tp->t_tmr_granularity != granularity) {
4670 		if (granularity == TCP_TMR_GRANULARITY_USEC)
4671 			srtt = TICKS_2_USEC(srtt);
4672 		else
4673 			srtt = USEC_2_TICKS(srtt);
4674 	}
4675 
4676 	/*
4677 	 * If the srtt is stored with ticks granularity, we need to
4678 	 * unshift to get the actual value. We do this after the
4679 	 * conversion above (if one was necessary) in order to maximize
4680 	 * precision.
4681 	 */
4682 	if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS)
4683 		srtt = srtt >> TCP_RTT_SHIFT;
4684 
4685 	return (srtt);
4686 }
4687