xref: /freebsd/sys/netinet/tcp_subr.c (revision d012836fb61654942c9d573c8e0f9def598d4ae2)
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 = 650;	/* 65 % (sack:2:ack -5%) */
172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
173     CTLFLAG_RW,
174     &tcp_restoral_thresh, 550,
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 
294 static int tcp_pacing_limit = 10000;
295 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
296     &tcp_pacing_limit, 1000,
297     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
298 
299 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
300     &shadow_num_connections, 0, "Number of TCP connections being paced");
301 
302 static int	tcp_log_debug = 0;
303 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
304     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
305 
306 static int	tcp_tcbhashsize;
307 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
308     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
309 
310 static int	do_tcpdrain = 1;
311 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
312     "Enable tcp_drain routine for extra help when low on mbufs");
313 
314 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
315     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
316 
317 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
318 #define	V_icmp_may_rst			VNET(icmp_may_rst)
319 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
320     &VNET_NAME(icmp_may_rst), 0,
321     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
322 
323 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
324 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
325 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
326     &VNET_NAME(tcp_isn_reseed_interval), 0,
327     "Seconds between reseeding of ISN secret");
328 
329 static int	tcp_soreceive_stream;
330 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
331     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
332 
333 VNET_DEFINE(uma_zone_t, sack_hole_zone);
334 #define	V_sack_hole_zone		VNET(sack_hole_zone)
335 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
336 static int
337 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
338 {
339 	int error;
340 	uint32_t new;
341 
342 	new = V_tcp_map_entries_limit;
343 	error = sysctl_handle_int(oidp, &new, 0, req);
344 	if (error == 0 && req->newptr) {
345 		/* only allow "0" and value > minimum */
346 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
347 			error = EINVAL;
348 		else
349 			V_tcp_map_entries_limit = new;
350 	}
351 	return (error);
352 }
353 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
354     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
355     &VNET_NAME(tcp_map_entries_limit), 0,
356     &sysctl_net_inet_tcp_map_limit_check, "IU",
357     "Total sendmap entries limit");
358 
359 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
360 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
361      &VNET_NAME(tcp_map_split_limit), 0,
362     "Total sendmap split entries limit");
363 
364 #ifdef TCP_HHOOK
365 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
366 #endif
367 
368 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
369 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
370 #define	V_ts_offset_secret	VNET(ts_offset_secret)
371 
372 static int	tcp_default_fb_init(struct tcpcb *tp, void **ptr);
373 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
374 static int	tcp_default_handoff_ok(struct tcpcb *tp);
375 static struct inpcb *tcp_notify(struct inpcb *, int);
376 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
377 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
378 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int);
379 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
380 		    const void *ip4hdr, const void *ip6hdr);
381 static ipproto_ctlinput_t	tcp_ctlinput;
382 static udp_tun_icmp_t		tcp_ctlinput_viaudp;
383 
384 static struct tcp_function_block tcp_def_funcblk = {
385 	.tfb_tcp_block_name = "freebsd",
386 	.tfb_tcp_output = tcp_default_output,
387 	.tfb_tcp_do_segment = tcp_do_segment,
388 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
389 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
390 	.tfb_tcp_fb_init = tcp_default_fb_init,
391 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
392 };
393 
394 static int tcp_fb_cnt = 0;
395 struct tcp_funchead t_functions;
396 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
397 
398 void
399 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
400 {
401 	TCPSTAT_INC(tcps_dsack_count);
402 	tp->t_dsack_pack++;
403 	if (tlp == 0) {
404 		if (SEQ_GT(end, start)) {
405 			tp->t_dsack_bytes += (end - start);
406 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
407 		} else {
408 			tp->t_dsack_tlp_bytes += (start - end);
409 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
410 		}
411 	} else {
412 		if (SEQ_GT(end, start)) {
413 			tp->t_dsack_bytes += (end - start);
414 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
415 		} else {
416 			tp->t_dsack_tlp_bytes += (start - end);
417 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
418 		}
419 	}
420 }
421 
422 static struct tcp_function_block *
423 find_tcp_functions_locked(struct tcp_function_set *fs)
424 {
425 	struct tcp_function *f;
426 	struct tcp_function_block *blk=NULL;
427 
428 	TAILQ_FOREACH(f, &t_functions, tf_next) {
429 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
430 			blk = f->tf_fb;
431 			break;
432 		}
433 	}
434 	return(blk);
435 }
436 
437 static struct tcp_function_block *
438 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
439 {
440 	struct tcp_function_block *rblk=NULL;
441 	struct tcp_function *f;
442 
443 	TAILQ_FOREACH(f, &t_functions, tf_next) {
444 		if (f->tf_fb == blk) {
445 			rblk = blk;
446 			if (s) {
447 				*s = f;
448 			}
449 			break;
450 		}
451 	}
452 	return (rblk);
453 }
454 
455 struct tcp_function_block *
456 find_and_ref_tcp_functions(struct tcp_function_set *fs)
457 {
458 	struct tcp_function_block *blk;
459 
460 	rw_rlock(&tcp_function_lock);
461 	blk = find_tcp_functions_locked(fs);
462 	if (blk)
463 		refcount_acquire(&blk->tfb_refcnt);
464 	rw_runlock(&tcp_function_lock);
465 	return(blk);
466 }
467 
468 struct tcp_function_block *
469 find_and_ref_tcp_fb(struct tcp_function_block *blk)
470 {
471 	struct tcp_function_block *rblk;
472 
473 	rw_rlock(&tcp_function_lock);
474 	rblk = find_tcp_fb_locked(blk, NULL);
475 	if (rblk)
476 		refcount_acquire(&rblk->tfb_refcnt);
477 	rw_runlock(&tcp_function_lock);
478 	return(rblk);
479 }
480 
481 /* Find a matching alias for the given tcp_function_block. */
482 int
483 find_tcp_function_alias(struct tcp_function_block *blk,
484     struct tcp_function_set *fs)
485 {
486 	struct tcp_function *f;
487 	int found;
488 
489 	found = 0;
490 	rw_rlock(&tcp_function_lock);
491 	TAILQ_FOREACH(f, &t_functions, tf_next) {
492 		if ((f->tf_fb == blk) &&
493 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
494 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
495 			/* Matching function block with different name. */
496 			strncpy(fs->function_set_name, f->tf_name,
497 			    TCP_FUNCTION_NAME_LEN_MAX);
498 			found = 1;
499 			break;
500 		}
501 	}
502 	/* Null terminate the string appropriately. */
503 	if (found) {
504 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
505 	} else {
506 		fs->function_set_name[0] = '\0';
507 	}
508 	rw_runlock(&tcp_function_lock);
509 	return (found);
510 }
511 
512 static struct tcp_function_block *
513 find_and_ref_tcp_default_fb(void)
514 {
515 	struct tcp_function_block *rblk;
516 
517 	rw_rlock(&tcp_function_lock);
518 	rblk = tcp_func_set_ptr;
519 	refcount_acquire(&rblk->tfb_refcnt);
520 	rw_runlock(&tcp_function_lock);
521 	return (rblk);
522 }
523 
524 void
525 tcp_switch_back_to_default(struct tcpcb *tp)
526 {
527 	struct tcp_function_block *tfb;
528 	void *ptr = NULL;
529 
530 	KASSERT(tp->t_fb != &tcp_def_funcblk,
531 	    ("%s: called by the built-in default stack", __func__));
532 
533 	/*
534 	 * Now, we'll find a new function block to use.
535 	 * Start by trying the current user-selected
536 	 * default, unless this stack is the user-selected
537 	 * default.
538 	 */
539 	tfb = find_and_ref_tcp_default_fb();
540 	if (tfb == tp->t_fb) {
541 		refcount_release(&tfb->tfb_refcnt);
542 		tfb = NULL;
543 	}
544 	/* Does the stack accept this connection? */
545 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
546 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
547 		refcount_release(&tfb->tfb_refcnt);
548 		tfb = NULL;
549 	}
550 	/* Try to use that stack. */
551 	if (tfb != NULL) {
552 		/* Initialize the new stack. If it succeeds, we are done. */
553 		if (tfb->tfb_tcp_fb_init == NULL ||
554 		    (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) {
555 			/* Release the old stack */
556 			if (tp->t_fb->tfb_tcp_fb_fini != NULL)
557 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
558 			refcount_release(&tp->t_fb->tfb_refcnt);
559 			/* Now set in all the pointers */
560 			tp->t_fb = tfb;
561 			tp->t_fb_ptr = ptr;
562 			return;
563 		}
564 		/*
565 		 * Initialization failed. Release the reference count on
566 		 * the looked up default stack.
567 		 */
568 		refcount_release(&tfb->tfb_refcnt);
569 	}
570 
571 	/*
572 	 * If that wasn't feasible, use the built-in default
573 	 * stack which is not allowed to reject anyone.
574 	 */
575 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
576 	if (tfb == NULL) {
577 		/* there always should be a default */
578 		panic("Can't refer to tcp_def_funcblk");
579 	}
580 	if (tfb->tfb_tcp_handoff_ok != NULL) {
581 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
582 			/* The default stack cannot say no */
583 			panic("Default stack rejects a new session?");
584 		}
585 	}
586 	if (tfb->tfb_tcp_fb_init != NULL &&
587 	    (*tfb->tfb_tcp_fb_init)(tp, &ptr)) {
588 		/* The default stack cannot fail */
589 		panic("Default stack initialization failed");
590 	}
591 	/* Now release the old stack */
592 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
593 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
594 	refcount_release(&tp->t_fb->tfb_refcnt);
595 	/* And set in the pointers to the new */
596 	tp->t_fb = tfb;
597 	tp->t_fb_ptr = ptr;
598 }
599 
600 static bool
601 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
602     const struct sockaddr *sa, void *ctx)
603 {
604 	struct ip *iph;
605 #ifdef INET6
606 	struct ip6_hdr *ip6;
607 #endif
608 	struct udphdr *uh;
609 	struct tcphdr *th;
610 	int thlen;
611 	uint16_t port;
612 
613 	TCPSTAT_INC(tcps_tunneled_pkts);
614 	if ((m->m_flags & M_PKTHDR) == 0) {
615 		/* Can't handle one that is not a pkt hdr */
616 		TCPSTAT_INC(tcps_tunneled_errs);
617 		goto out;
618 	}
619 	thlen = sizeof(struct tcphdr);
620 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
621 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
622 		TCPSTAT_INC(tcps_tunneled_errs);
623 		goto out;
624 	}
625 	iph = mtod(m, struct ip *);
626 	uh = (struct udphdr *)((caddr_t)iph + off);
627 	th = (struct tcphdr *)(uh + 1);
628 	thlen = th->th_off << 2;
629 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
630 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
631 		if (m == NULL) {
632 			TCPSTAT_INC(tcps_tunneled_errs);
633 			goto out;
634 		} else {
635 			iph = mtod(m, struct ip *);
636 			uh = (struct udphdr *)((caddr_t)iph + off);
637 			th = (struct tcphdr *)(uh + 1);
638 		}
639 	}
640 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
641 	bcopy(th, uh, m->m_len - off);
642 	m->m_len -= sizeof(struct udphdr);
643 	m->m_pkthdr.len -= sizeof(struct udphdr);
644 	/*
645 	 * We use the same algorithm for
646 	 * both UDP and TCP for c-sum. So
647 	 * the code in tcp_input will skip
648 	 * the checksum. So we do nothing
649 	 * with the flag (m->m_pkthdr.csum_flags).
650 	 */
651 	switch (iph->ip_v) {
652 #ifdef INET
653 	case IPVERSION:
654 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
655 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
656 		break;
657 #endif
658 #ifdef INET6
659 	case IPV6_VERSION >> 4:
660 		ip6 = mtod(m, struct ip6_hdr *);
661 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
662 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
663 		break;
664 #endif
665 	default:
666 		goto out;
667 		break;
668 	}
669 	return (true);
670 out:
671 	m_freem(m);
672 
673 	return (true);
674 }
675 
676 static int
677 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
678 {
679 	int error=ENOENT;
680 	struct tcp_function_set fs;
681 	struct tcp_function_block *blk;
682 
683 	memset(&fs, 0, sizeof(fs));
684 	rw_rlock(&tcp_function_lock);
685 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
686 	if (blk) {
687 		/* Found him */
688 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
689 		fs.pcbcnt = blk->tfb_refcnt;
690 	}
691 	rw_runlock(&tcp_function_lock);
692 	error = sysctl_handle_string(oidp, fs.function_set_name,
693 				     sizeof(fs.function_set_name), req);
694 
695 	/* Check for error or no change */
696 	if (error != 0 || req->newptr == NULL)
697 		return(error);
698 
699 	rw_wlock(&tcp_function_lock);
700 	blk = find_tcp_functions_locked(&fs);
701 	if ((blk == NULL) ||
702 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
703 		error = ENOENT;
704 		goto done;
705 	}
706 	tcp_func_set_ptr = blk;
707 done:
708 	rw_wunlock(&tcp_function_lock);
709 	return (error);
710 }
711 
712 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
713     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
714     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
715     "Set/get the default TCP functions");
716 
717 static int
718 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
719 {
720 	int error, cnt, linesz;
721 	struct tcp_function *f;
722 	char *buffer, *cp;
723 	size_t bufsz, outsz;
724 	bool alias;
725 
726 	cnt = 0;
727 	rw_rlock(&tcp_function_lock);
728 	TAILQ_FOREACH(f, &t_functions, tf_next) {
729 		cnt++;
730 	}
731 	rw_runlock(&tcp_function_lock);
732 
733 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
734 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
735 
736 	error = 0;
737 	cp = buffer;
738 
739 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
740 	    "Alias", "PCB count");
741 	cp += linesz;
742 	bufsz -= linesz;
743 	outsz = linesz;
744 
745 	rw_rlock(&tcp_function_lock);
746 	TAILQ_FOREACH(f, &t_functions, tf_next) {
747 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
748 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
749 		    f->tf_fb->tfb_tcp_block_name,
750 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
751 		    alias ? f->tf_name : "-",
752 		    f->tf_fb->tfb_refcnt);
753 		if (linesz >= bufsz) {
754 			error = EOVERFLOW;
755 			break;
756 		}
757 		cp += linesz;
758 		bufsz -= linesz;
759 		outsz += linesz;
760 	}
761 	rw_runlock(&tcp_function_lock);
762 	if (error == 0)
763 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
764 	free(buffer, M_TEMP);
765 	return (error);
766 }
767 
768 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
769     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
770     NULL, 0, sysctl_net_inet_list_available, "A",
771     "list available TCP Function sets");
772 
773 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
774 
775 #ifdef INET
776 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
777 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
778 #endif
779 #ifdef INET6
780 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
781 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
782 #endif
783 
784 static struct sx tcpoudp_lock;
785 
786 static void
787 tcp_over_udp_stop(void)
788 {
789 
790 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
791 
792 #ifdef INET
793 	if (V_udp4_tun_socket != NULL) {
794 		soclose(V_udp4_tun_socket);
795 		V_udp4_tun_socket = NULL;
796 	}
797 #endif
798 #ifdef INET6
799 	if (V_udp6_tun_socket != NULL) {
800 		soclose(V_udp6_tun_socket);
801 		V_udp6_tun_socket = NULL;
802 	}
803 #endif
804 }
805 
806 static int
807 tcp_over_udp_start(void)
808 {
809 	uint16_t port;
810 	int ret;
811 #ifdef INET
812 	struct sockaddr_in sin;
813 #endif
814 #ifdef INET6
815 	struct sockaddr_in6 sin6;
816 #endif
817 
818 	sx_assert(&tcpoudp_lock, SA_XLOCKED);
819 
820 	port = V_tcp_udp_tunneling_port;
821 	if (ntohs(port) == 0) {
822 		/* Must have a port set */
823 		return (EINVAL);
824 	}
825 #ifdef INET
826 	if (V_udp4_tun_socket != NULL) {
827 		/* Already running -- must stop first */
828 		return (EALREADY);
829 	}
830 #endif
831 #ifdef INET6
832 	if (V_udp6_tun_socket != NULL) {
833 		/* Already running -- must stop first */
834 		return (EALREADY);
835 	}
836 #endif
837 #ifdef INET
838 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
839 	    SOCK_DGRAM, IPPROTO_UDP,
840 	    curthread->td_ucred, curthread))) {
841 		tcp_over_udp_stop();
842 		return (ret);
843 	}
844 	/* Call the special UDP hook. */
845 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
846 	    tcp_recv_udp_tunneled_packet,
847 	    tcp_ctlinput_viaudp,
848 	    NULL))) {
849 		tcp_over_udp_stop();
850 		return (ret);
851 	}
852 	/* Ok, we have a socket, bind it to the port. */
853 	memset(&sin, 0, sizeof(struct sockaddr_in));
854 	sin.sin_len = sizeof(struct sockaddr_in);
855 	sin.sin_family = AF_INET;
856 	sin.sin_port = htons(port);
857 	if ((ret = sobind(V_udp4_tun_socket,
858 	    (struct sockaddr *)&sin, curthread))) {
859 		tcp_over_udp_stop();
860 		return (ret);
861 	}
862 #endif
863 #ifdef INET6
864 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
865 	    SOCK_DGRAM, IPPROTO_UDP,
866 	    curthread->td_ucred, curthread))) {
867 		tcp_over_udp_stop();
868 		return (ret);
869 	}
870 	/* Call the special UDP hook. */
871 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
872 	    tcp_recv_udp_tunneled_packet,
873 	    tcp6_ctlinput_viaudp,
874 	    NULL))) {
875 		tcp_over_udp_stop();
876 		return (ret);
877 	}
878 	/* Ok, we have a socket, bind it to the port. */
879 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
880 	sin6.sin6_len = sizeof(struct sockaddr_in6);
881 	sin6.sin6_family = AF_INET6;
882 	sin6.sin6_port = htons(port);
883 	if ((ret = sobind(V_udp6_tun_socket,
884 	    (struct sockaddr *)&sin6, curthread))) {
885 		tcp_over_udp_stop();
886 		return (ret);
887 	}
888 #endif
889 	return (0);
890 }
891 
892 static int
893 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
894 {
895 	int error;
896 	uint32_t old, new;
897 
898 	old = V_tcp_udp_tunneling_port;
899 	new = old;
900 	error = sysctl_handle_int(oidp, &new, 0, req);
901 	if ((error == 0) &&
902 	    (req->newptr != NULL)) {
903 		if ((new < TCP_TUNNELING_PORT_MIN) ||
904 		    (new > TCP_TUNNELING_PORT_MAX)) {
905 			error = EINVAL;
906 		} else {
907 			sx_xlock(&tcpoudp_lock);
908 			V_tcp_udp_tunneling_port = new;
909 			if (old != 0) {
910 				tcp_over_udp_stop();
911 			}
912 			if (new != 0) {
913 				error = tcp_over_udp_start();
914 				if (error != 0) {
915 					V_tcp_udp_tunneling_port = 0;
916 				}
917 			}
918 			sx_xunlock(&tcpoudp_lock);
919 		}
920 	}
921 	return (error);
922 }
923 
924 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
925     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
926     &VNET_NAME(tcp_udp_tunneling_port),
927     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
928     "Tunneling port for tcp over udp");
929 
930 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
931 
932 static int
933 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
934 {
935 	int error, new;
936 
937 	new = V_tcp_udp_tunneling_overhead;
938 	error = sysctl_handle_int(oidp, &new, 0, req);
939 	if (error == 0 && req->newptr) {
940 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
941 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
942 			error = EINVAL;
943 		else
944 			V_tcp_udp_tunneling_overhead = new;
945 	}
946 	return (error);
947 }
948 
949 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
950     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
951     &VNET_NAME(tcp_udp_tunneling_overhead),
952     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
953     "MSS reduction when using tcp over udp");
954 
955 /*
956  * Exports one (struct tcp_function_info) for each alias/name.
957  */
958 static int
959 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
960 {
961 	int cnt, error;
962 	struct tcp_function *f;
963 	struct tcp_function_info tfi;
964 
965 	/*
966 	 * We don't allow writes.
967 	 */
968 	if (req->newptr != NULL)
969 		return (EINVAL);
970 
971 	/*
972 	 * Wire the old buffer so we can directly copy the functions to
973 	 * user space without dropping the lock.
974 	 */
975 	if (req->oldptr != NULL) {
976 		error = sysctl_wire_old_buffer(req, 0);
977 		if (error)
978 			return (error);
979 	}
980 
981 	/*
982 	 * Walk the list and copy out matching entries. If INVARIANTS
983 	 * is compiled in, also walk the list to verify the length of
984 	 * the list matches what we have recorded.
985 	 */
986 	rw_rlock(&tcp_function_lock);
987 
988 	cnt = 0;
989 #ifndef INVARIANTS
990 	if (req->oldptr == NULL) {
991 		cnt = tcp_fb_cnt;
992 		goto skip_loop;
993 	}
994 #endif
995 	TAILQ_FOREACH(f, &t_functions, tf_next) {
996 #ifdef INVARIANTS
997 		cnt++;
998 #endif
999 		if (req->oldptr != NULL) {
1000 			bzero(&tfi, sizeof(tfi));
1001 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
1002 			tfi.tfi_id = f->tf_fb->tfb_id;
1003 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
1004 			    sizeof(tfi.tfi_alias));
1005 			(void)strlcpy(tfi.tfi_name,
1006 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
1007 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
1008 			/*
1009 			 * Don't stop on error, as that is the
1010 			 * mechanism we use to accumulate length
1011 			 * information if the buffer was too short.
1012 			 */
1013 		}
1014 	}
1015 	KASSERT(cnt == tcp_fb_cnt,
1016 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1017 #ifndef INVARIANTS
1018 skip_loop:
1019 #endif
1020 	rw_runlock(&tcp_function_lock);
1021 	if (req->oldptr == NULL)
1022 		error = SYSCTL_OUT(req, NULL,
1023 		    (cnt + 1) * sizeof(struct tcp_function_info));
1024 
1025 	return (error);
1026 }
1027 
1028 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1029 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1030 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1031 	    "List TCP function block name-to-ID mappings");
1032 
1033 /*
1034  * tfb_tcp_handoff_ok() function for the default stack.
1035  * Note that we'll basically try to take all comers.
1036  */
1037 static int
1038 tcp_default_handoff_ok(struct tcpcb *tp)
1039 {
1040 
1041 	return (0);
1042 }
1043 
1044 /*
1045  * tfb_tcp_fb_init() function for the default stack.
1046  *
1047  * This handles making sure we have appropriate timers set if you are
1048  * transitioning a socket that has some amount of setup done.
1049  *
1050  * The init() fuction from the default can *never* return non-zero i.e.
1051  * it is required to always succeed since it is the stack of last resort!
1052  */
1053 static int
1054 tcp_default_fb_init(struct tcpcb *tp, void **ptr)
1055 {
1056 	struct socket *so = tptosocket(tp);
1057 	int rexmt;
1058 
1059 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1060 	/* We don't use the pointer */
1061 	*ptr = NULL;
1062 
1063 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1064 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1065 	    tp->t_state));
1066 
1067 	/* Make sure we get no interesting mbuf queuing behavior */
1068 	/* All mbuf queue/ack compress flags should be off */
1069 	tcp_lro_features_off(tptoinpcb(tp));
1070 
1071 	/* Cancel the GP measurement in progress */
1072 	tp->t_flags &= ~TF_GPUTINPROG;
1073 	/* Validate the timers are not in usec, if they are convert */
1074 	tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS);
1075 	if ((tp->t_state == TCPS_SYN_SENT) ||
1076 	    (tp->t_state == TCPS_SYN_RECEIVED))
1077 		rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift];
1078 	else
1079 		rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift];
1080 	if (tp->t_rxtshift == 0)
1081 		tp->t_rxtcur = rexmt;
1082 	else
1083 		TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX);
1084 
1085 	/*
1086 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1087 	 * know what to do for unexpected states (which includes TIME_WAIT).
1088 	 */
1089 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1090 		return (0);
1091 
1092 	/*
1093 	 * Make sure some kind of transmission timer is set if there is
1094 	 * outstanding data.
1095 	 */
1096 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1097 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1098 	    tcp_timer_active(tp, TT_PERSIST))) {
1099 		/*
1100 		 * If the session has established and it looks like it should
1101 		 * be in the persist state, set the persist timer. Otherwise,
1102 		 * set the retransmit timer.
1103 		 */
1104 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1105 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1106 		    (int32_t)sbavail(&so->so_snd))
1107 			tcp_setpersist(tp);
1108 		else
1109 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1110 	}
1111 
1112 	/* All non-embryonic sessions get a keepalive timer. */
1113 	if (!tcp_timer_active(tp, TT_KEEP))
1114 		tcp_timer_activate(tp, TT_KEEP,
1115 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1116 		    TP_KEEPINIT(tp));
1117 
1118 	/*
1119 	 * Make sure critical variables are initialized
1120 	 * if transitioning while in Recovery.
1121 	 */
1122 	if IN_FASTRECOVERY(tp->t_flags) {
1123 		if (tp->sackhint.recover_fs == 0)
1124 			tp->sackhint.recover_fs = max(1,
1125 			    tp->snd_nxt - tp->snd_una);
1126 	}
1127 
1128 	return (0);
1129 }
1130 
1131 /*
1132  * tfb_tcp_fb_fini() function for the default stack.
1133  *
1134  * This changes state as necessary (or prudent) to prepare for another stack
1135  * to assume responsibility for the connection.
1136  */
1137 static void
1138 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1139 {
1140 
1141 	INP_WLOCK_ASSERT(tptoinpcb(tp));
1142 }
1143 
1144 /*
1145  * Target size of TCP PCB hash tables. Must be a power of two.
1146  *
1147  * Note that this can be overridden by the kernel environment
1148  * variable net.inet.tcp.tcbhashsize
1149  */
1150 #ifndef TCBHASHSIZE
1151 #define TCBHASHSIZE	0
1152 #endif
1153 
1154 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1155 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1156 
1157 static struct mtx isn_mtx;
1158 
1159 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1160 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1161 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1162 
1163 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1164 
1165 /*
1166  * Take a value and get the next power of 2 that doesn't overflow.
1167  * Used to size the tcp_inpcb hash buckets.
1168  */
1169 static int
1170 maketcp_hashsize(int size)
1171 {
1172 	int hashsize;
1173 
1174 	/*
1175 	 * auto tune.
1176 	 * get the next power of 2 higher than maxsockets.
1177 	 */
1178 	hashsize = 1 << fls(size);
1179 	/* catch overflow, and just go one power of 2 smaller */
1180 	if (hashsize < size) {
1181 		hashsize = 1 << (fls(size) - 1);
1182 	}
1183 	return (hashsize);
1184 }
1185 
1186 static volatile int next_tcp_stack_id = 1;
1187 
1188 /*
1189  * Register a TCP function block with the name provided in the names
1190  * array.  (Note that this function does NOT automatically register
1191  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1192  * explicitly include blk->tfb_tcp_block_name in the list of names if
1193  * you wish to register the stack with that name.)
1194  *
1195  * Either all name registrations will succeed or all will fail.  If
1196  * a name registration fails, the function will update the num_names
1197  * argument to point to the array index of the name that encountered
1198  * the failure.
1199  *
1200  * Returns 0 on success, or an error code on failure.
1201  */
1202 int
1203 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1204     const char *names[], int *num_names)
1205 {
1206 	struct tcp_function *n;
1207 	struct tcp_function_set fs;
1208 	int error, i;
1209 
1210 	KASSERT(names != NULL && *num_names > 0,
1211 	    ("%s: Called with 0-length name list", __func__));
1212 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1213 	KASSERT(rw_initialized(&tcp_function_lock),
1214 	    ("%s: called too early", __func__));
1215 
1216 	if ((blk->tfb_tcp_output == NULL) ||
1217 	    (blk->tfb_tcp_do_segment == NULL) ||
1218 	    (blk->tfb_tcp_ctloutput == NULL) ||
1219 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1220 		/*
1221 		 * These functions are required and you
1222 		 * need a name.
1223 		 */
1224 		*num_names = 0;
1225 		return (EINVAL);
1226 	}
1227 
1228 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1229 		*num_names = 0;
1230 		return (EINVAL);
1231 	}
1232 
1233 	refcount_init(&blk->tfb_refcnt, 0);
1234 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1235 	for (i = 0; i < *num_names; i++) {
1236 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1237 		if (n == NULL) {
1238 			error = ENOMEM;
1239 			goto cleanup;
1240 		}
1241 		n->tf_fb = blk;
1242 
1243 		(void)strlcpy(fs.function_set_name, names[i],
1244 		    sizeof(fs.function_set_name));
1245 		rw_wlock(&tcp_function_lock);
1246 		if (find_tcp_functions_locked(&fs) != NULL) {
1247 			/* Duplicate name space not allowed */
1248 			rw_wunlock(&tcp_function_lock);
1249 			free(n, M_TCPFUNCTIONS);
1250 			error = EALREADY;
1251 			goto cleanup;
1252 		}
1253 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1254 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1255 		tcp_fb_cnt++;
1256 		rw_wunlock(&tcp_function_lock);
1257 	}
1258 	return(0);
1259 
1260 cleanup:
1261 	/*
1262 	 * Deregister the names we just added. Because registration failed
1263 	 * for names[i], we don't need to deregister that name.
1264 	 */
1265 	*num_names = i;
1266 	rw_wlock(&tcp_function_lock);
1267 	while (--i >= 0) {
1268 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1269 			if (!strncmp(n->tf_name, names[i],
1270 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1271 				TAILQ_REMOVE(&t_functions, n, tf_next);
1272 				tcp_fb_cnt--;
1273 				n->tf_fb = NULL;
1274 				free(n, M_TCPFUNCTIONS);
1275 				break;
1276 			}
1277 		}
1278 	}
1279 	rw_wunlock(&tcp_function_lock);
1280 	return (error);
1281 }
1282 
1283 /*
1284  * Register a TCP function block using the name provided in the name
1285  * argument.
1286  *
1287  * Returns 0 on success, or an error code on failure.
1288  */
1289 int
1290 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1291     int wait)
1292 {
1293 	const char *name_list[1];
1294 	int num_names, rv;
1295 
1296 	num_names = 1;
1297 	if (name != NULL)
1298 		name_list[0] = name;
1299 	else
1300 		name_list[0] = blk->tfb_tcp_block_name;
1301 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1302 	return (rv);
1303 }
1304 
1305 /*
1306  * Register a TCP function block using the name defined in
1307  * blk->tfb_tcp_block_name.
1308  *
1309  * Returns 0 on success, or an error code on failure.
1310  */
1311 int
1312 register_tcp_functions(struct tcp_function_block *blk, int wait)
1313 {
1314 
1315 	return (register_tcp_functions_as_name(blk, NULL, wait));
1316 }
1317 
1318 /*
1319  * Deregister all names associated with a function block. This
1320  * functionally removes the function block from use within the system.
1321  *
1322  * When called with a true quiesce argument, mark the function block
1323  * as being removed so no more stacks will use it and determine
1324  * whether the removal would succeed.
1325  *
1326  * When called with a false quiesce argument, actually attempt the
1327  * removal.
1328  *
1329  * When called with a force argument, attempt to switch all TCBs to
1330  * use the default stack instead of returning EBUSY.
1331  *
1332  * Returns 0 on success (or if the removal would succeed, or an error
1333  * code on failure.
1334  */
1335 int
1336 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1337     bool force)
1338 {
1339 	struct tcp_function *f;
1340 
1341 	if (blk == &tcp_def_funcblk) {
1342 		/* You can't un-register the default */
1343 		return (EPERM);
1344 	}
1345 	rw_wlock(&tcp_function_lock);
1346 	if (blk == tcp_func_set_ptr) {
1347 		/* You can't free the current default */
1348 		rw_wunlock(&tcp_function_lock);
1349 		return (EBUSY);
1350 	}
1351 	/* Mark the block so no more stacks can use it. */
1352 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1353 	/*
1354 	 * If TCBs are still attached to the stack, attempt to switch them
1355 	 * to the default stack.
1356 	 */
1357 	if (force && blk->tfb_refcnt) {
1358 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1359 		    INPLOOKUP_WLOCKPCB);
1360 		struct inpcb *inp;
1361 		struct tcpcb *tp;
1362 		VNET_ITERATOR_DECL(vnet_iter);
1363 
1364 		rw_wunlock(&tcp_function_lock);
1365 
1366 		VNET_LIST_RLOCK();
1367 		VNET_FOREACH(vnet_iter) {
1368 			CURVNET_SET(vnet_iter);
1369 			while ((inp = inp_next(&inpi)) != NULL) {
1370 				tp = intotcpcb(inp);
1371 				if (tp == NULL || tp->t_fb != blk)
1372 					continue;
1373 				tcp_switch_back_to_default(tp);
1374 			}
1375 			CURVNET_RESTORE();
1376 		}
1377 		VNET_LIST_RUNLOCK();
1378 
1379 		rw_wlock(&tcp_function_lock);
1380 	}
1381 	if (blk->tfb_refcnt) {
1382 		/* TCBs still attached. */
1383 		rw_wunlock(&tcp_function_lock);
1384 		return (EBUSY);
1385 	}
1386 	if (quiesce) {
1387 		/* Skip removal. */
1388 		rw_wunlock(&tcp_function_lock);
1389 		return (0);
1390 	}
1391 	/* Remove any function names that map to this function block. */
1392 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1393 		TAILQ_REMOVE(&t_functions, f, tf_next);
1394 		tcp_fb_cnt--;
1395 		f->tf_fb = NULL;
1396 		free(f, M_TCPFUNCTIONS);
1397 	}
1398 	rw_wunlock(&tcp_function_lock);
1399 	return (0);
1400 }
1401 
1402 static void
1403 tcp_drain(void)
1404 {
1405 	struct epoch_tracker et;
1406 	VNET_ITERATOR_DECL(vnet_iter);
1407 
1408 	if (!do_tcpdrain)
1409 		return;
1410 
1411 	NET_EPOCH_ENTER(et);
1412 	VNET_LIST_RLOCK_NOSLEEP();
1413 	VNET_FOREACH(vnet_iter) {
1414 		CURVNET_SET(vnet_iter);
1415 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1416 		    INPLOOKUP_WLOCKPCB);
1417 		struct inpcb *inpb;
1418 		struct tcpcb *tcpb;
1419 
1420 	/*
1421 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1422 	 * if there is one...
1423 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1424 	 *      reassembly queue should be flushed, but in a situation
1425 	 *	where we're really low on mbufs, this is potentially
1426 	 *	useful.
1427 	 */
1428 		while ((inpb = inp_next(&inpi)) != NULL) {
1429 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1430 				tcp_reass_flush(tcpb);
1431 				tcp_clean_sackreport(tcpb);
1432 #ifdef TCP_BLACKBOX
1433 				tcp_log_drain(tcpb);
1434 #endif
1435 #ifdef TCPPCAP
1436 				if (tcp_pcap_aggressive_free) {
1437 					/* Free the TCP PCAP queues. */
1438 					tcp_pcap_drain(&(tcpb->t_inpkts));
1439 					tcp_pcap_drain(&(tcpb->t_outpkts));
1440 				}
1441 #endif
1442 			}
1443 		}
1444 		CURVNET_RESTORE();
1445 	}
1446 	VNET_LIST_RUNLOCK_NOSLEEP();
1447 	NET_EPOCH_EXIT(et);
1448 }
1449 
1450 static void
1451 tcp_vnet_init(void *arg __unused)
1452 {
1453 
1454 #ifdef TCP_HHOOK
1455 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1456 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1457 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1458 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1459 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1460 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1461 #endif
1462 #ifdef STATS
1463 	if (tcp_stats_init())
1464 		printf("%s: WARNING: unable to initialise TCP stats\n",
1465 		    __func__);
1466 #endif
1467 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1468 	    tcp_tcbhashsize);
1469 
1470 	syncache_init();
1471 	tcp_hc_init();
1472 
1473 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1474 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1475 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1476 
1477 	tcp_fastopen_init();
1478 
1479 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1480 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1481 
1482 	V_tcp_msl = TCPTV_MSL;
1483 }
1484 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1485     tcp_vnet_init, NULL);
1486 
1487 static void
1488 tcp_init(void *arg __unused)
1489 {
1490 	const char *tcbhash_tuneable;
1491 	int hashsize;
1492 
1493 	tcp_reass_global_init();
1494 
1495 	/* XXX virtualize those below? */
1496 	tcp_delacktime = TCPTV_DELACK;
1497 	tcp_keepinit = TCPTV_KEEP_INIT;
1498 	tcp_keepidle = TCPTV_KEEP_IDLE;
1499 	tcp_keepintvl = TCPTV_KEEPINTVL;
1500 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1501 	tcp_rexmit_initial = TCPTV_RTOBASE;
1502 	if (tcp_rexmit_initial < 1)
1503 		tcp_rexmit_initial = 1;
1504 	tcp_rexmit_min = TCPTV_MIN;
1505 	if (tcp_rexmit_min < 1)
1506 		tcp_rexmit_min = 1;
1507 	tcp_persmin = TCPTV_PERSMIN;
1508 	tcp_persmax = TCPTV_PERSMAX;
1509 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1510 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1511 
1512 	/* Setup the tcp function block list */
1513 	TAILQ_INIT(&t_functions);
1514 	rw_init(&tcp_function_lock, "tcp_func_lock");
1515 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1516 	sx_init(&tcpoudp_lock, "TCP over UDP configuration");
1517 #ifdef TCP_BLACKBOX
1518 	/* Initialize the TCP logging data. */
1519 	tcp_log_init();
1520 #endif
1521 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1522 
1523 	if (tcp_soreceive_stream) {
1524 #ifdef INET
1525 		tcp_protosw.pr_soreceive = soreceive_stream;
1526 #endif
1527 #ifdef INET6
1528 		tcp6_protosw.pr_soreceive = soreceive_stream;
1529 #endif /* INET6 */
1530 	}
1531 
1532 #ifdef INET6
1533 	max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
1534 #else /* INET6 */
1535 	max_protohdr_grow(sizeof(struct tcpiphdr));
1536 #endif /* INET6 */
1537 
1538 	ISN_LOCK_INIT();
1539 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1540 		SHUTDOWN_PRI_DEFAULT);
1541 	EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1542 	EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT);
1543 
1544 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1545 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1546 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1547 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1548 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1549 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1550 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1551 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1552 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1553 #ifdef TCPPCAP
1554 	tcp_pcap_init();
1555 #endif
1556 
1557 	hashsize = TCBHASHSIZE;
1558 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1559 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1560 	if (hashsize == 0) {
1561 		/*
1562 		 * Auto tune the hash size based on maxsockets.
1563 		 * A perfect hash would have a 1:1 mapping
1564 		 * (hashsize = maxsockets) however it's been
1565 		 * suggested that O(2) average is better.
1566 		 */
1567 		hashsize = maketcp_hashsize(maxsockets / 4);
1568 		/*
1569 		 * Our historical default is 512,
1570 		 * do not autotune lower than this.
1571 		 */
1572 		if (hashsize < 512)
1573 			hashsize = 512;
1574 		if (bootverbose)
1575 			printf("%s: %s auto tuned to %d\n", __func__,
1576 			    tcbhash_tuneable, hashsize);
1577 	}
1578 	/*
1579 	 * We require a hashsize to be a power of two.
1580 	 * Previously if it was not a power of two we would just reset it
1581 	 * back to 512, which could be a nasty surprise if you did not notice
1582 	 * the error message.
1583 	 * Instead what we do is clip it to the closest power of two lower
1584 	 * than the specified hash value.
1585 	 */
1586 	if (!powerof2(hashsize)) {
1587 		int oldhashsize = hashsize;
1588 
1589 		hashsize = maketcp_hashsize(hashsize);
1590 		/* prevent absurdly low value */
1591 		if (hashsize < 16)
1592 			hashsize = 16;
1593 		printf("%s: WARNING: TCB hash size not a power of 2, "
1594 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1595 		    hashsize);
1596 	}
1597 	tcp_tcbhashsize = hashsize;
1598 
1599 #ifdef INET
1600 	IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput);
1601 #endif
1602 #ifdef INET6
1603 	IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput);
1604 #endif
1605 }
1606 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1607 
1608 #ifdef VIMAGE
1609 static void
1610 tcp_destroy(void *unused __unused)
1611 {
1612 	int n;
1613 #ifdef TCP_HHOOK
1614 	int error;
1615 #endif
1616 
1617 	/*
1618 	 * All our processes are gone, all our sockets should be cleaned
1619 	 * up, which means, we should be past the tcp_discardcb() calls.
1620 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1621 	 */
1622 	for (;;) {
1623 		INP_INFO_WLOCK(&V_tcbinfo);
1624 		n = V_tcbinfo.ipi_count;
1625 		INP_INFO_WUNLOCK(&V_tcbinfo);
1626 		if (n == 0)
1627 			break;
1628 		pause("tcpdes", hz / 10);
1629 	}
1630 	tcp_hc_destroy();
1631 	syncache_destroy();
1632 	in_pcbinfo_destroy(&V_tcbinfo);
1633 	/* tcp_discardcb() clears the sack_holes up. */
1634 	uma_zdestroy(V_sack_hole_zone);
1635 
1636 	/*
1637 	 * Cannot free the zone until all tcpcbs are released as we attach
1638 	 * the allocations to them.
1639 	 */
1640 	tcp_fastopen_destroy();
1641 
1642 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1643 	VNET_PCPUSTAT_FREE(tcpstat);
1644 
1645 #ifdef TCP_HHOOK
1646 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1647 	if (error != 0) {
1648 		printf("%s: WARNING: unable to deregister helper hook "
1649 		    "type=%d, id=%d: error %d returned\n", __func__,
1650 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1651 	}
1652 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1653 	if (error != 0) {
1654 		printf("%s: WARNING: unable to deregister helper hook "
1655 		    "type=%d, id=%d: error %d returned\n", __func__,
1656 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1657 	}
1658 #endif
1659 }
1660 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1661 #endif
1662 
1663 void
1664 tcp_fini(void *xtp)
1665 {
1666 
1667 }
1668 
1669 /*
1670  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1671  * tcp_template used to store this data in mbufs, but we now recopy it out
1672  * of the tcpcb each time to conserve mbufs.
1673  */
1674 void
1675 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1676 {
1677 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1678 
1679 	INP_WLOCK_ASSERT(inp);
1680 
1681 #ifdef INET6
1682 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1683 		struct ip6_hdr *ip6;
1684 
1685 		ip6 = (struct ip6_hdr *)ip_ptr;
1686 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1687 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1688 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1689 			(IPV6_VERSION & IPV6_VERSION_MASK);
1690 		if (port == 0)
1691 			ip6->ip6_nxt = IPPROTO_TCP;
1692 		else
1693 			ip6->ip6_nxt = IPPROTO_UDP;
1694 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1695 		ip6->ip6_src = inp->in6p_laddr;
1696 		ip6->ip6_dst = inp->in6p_faddr;
1697 	}
1698 #endif /* INET6 */
1699 #if defined(INET6) && defined(INET)
1700 	else
1701 #endif
1702 #ifdef INET
1703 	{
1704 		struct ip *ip;
1705 
1706 		ip = (struct ip *)ip_ptr;
1707 		ip->ip_v = IPVERSION;
1708 		ip->ip_hl = 5;
1709 		ip->ip_tos = inp->inp_ip_tos;
1710 		ip->ip_len = 0;
1711 		ip->ip_id = 0;
1712 		ip->ip_off = 0;
1713 		ip->ip_ttl = inp->inp_ip_ttl;
1714 		ip->ip_sum = 0;
1715 		if (port == 0)
1716 			ip->ip_p = IPPROTO_TCP;
1717 		else
1718 			ip->ip_p = IPPROTO_UDP;
1719 		ip->ip_src = inp->inp_laddr;
1720 		ip->ip_dst = inp->inp_faddr;
1721 	}
1722 #endif /* INET */
1723 	th->th_sport = inp->inp_lport;
1724 	th->th_dport = inp->inp_fport;
1725 	th->th_seq = 0;
1726 	th->th_ack = 0;
1727 	th->th_off = 5;
1728 	tcp_set_flags(th, 0);
1729 	th->th_win = 0;
1730 	th->th_urp = 0;
1731 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1732 }
1733 
1734 /*
1735  * Create template to be used to send tcp packets on a connection.
1736  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1737  * use for this function is in keepalives, which use tcp_respond.
1738  */
1739 struct tcptemp *
1740 tcpip_maketemplate(struct inpcb *inp)
1741 {
1742 	struct tcptemp *t;
1743 
1744 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1745 	if (t == NULL)
1746 		return (NULL);
1747 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1748 	return (t);
1749 }
1750 
1751 /*
1752  * Send a single message to the TCP at address specified by
1753  * the given TCP/IP header.  If m == NULL, then we make a copy
1754  * of the tcpiphdr at th and send directly to the addressed host.
1755  * This is used to force keep alive messages out using the TCP
1756  * template for a connection.  If flags are given then we send
1757  * a message back to the TCP which originated the segment th,
1758  * and discard the mbuf containing it and any other attached mbufs.
1759  *
1760  * In any case the ack and sequence number of the transmitted
1761  * segment are as specified by the parameters.
1762  *
1763  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1764  */
1765 void
1766 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1767     tcp_seq ack, tcp_seq seq, uint16_t flags)
1768 {
1769 	struct tcpopt to;
1770 	struct inpcb *inp;
1771 	struct ip *ip;
1772 	struct mbuf *optm;
1773 	struct udphdr *uh = NULL;
1774 	struct tcphdr *nth;
1775 	struct tcp_log_buffer *lgb;
1776 	u_char *optp;
1777 #ifdef INET6
1778 	struct ip6_hdr *ip6;
1779 	int isipv6;
1780 #endif /* INET6 */
1781 	int optlen, tlen, win, ulen;
1782 	int ect = 0;
1783 	bool incl_opts;
1784 	uint16_t port;
1785 	int output_ret;
1786 #ifdef INVARIANTS
1787 	int thflags = tcp_get_flags(th);
1788 #endif
1789 
1790 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1791 	NET_EPOCH_ASSERT();
1792 
1793 #ifdef INET6
1794 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1795 	ip6 = ipgen;
1796 #endif /* INET6 */
1797 	ip = ipgen;
1798 
1799 	if (tp != NULL) {
1800 		inp = tptoinpcb(tp);
1801 		INP_LOCK_ASSERT(inp);
1802 	} else
1803 		inp = NULL;
1804 
1805 	if (m != NULL) {
1806 #ifdef INET6
1807 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1808 			port = m->m_pkthdr.tcp_tun_port;
1809 		else
1810 #endif
1811 		if (ip && (ip->ip_p == IPPROTO_UDP))
1812 			port = m->m_pkthdr.tcp_tun_port;
1813 		else
1814 			port = 0;
1815 	} else
1816 		port = tp->t_port;
1817 
1818 	incl_opts = false;
1819 	win = 0;
1820 	if (tp != NULL) {
1821 		if (!(flags & TH_RST)) {
1822 			win = sbspace(&inp->inp_socket->so_rcv);
1823 			if (win > TCP_MAXWIN << tp->rcv_scale)
1824 				win = TCP_MAXWIN << tp->rcv_scale;
1825 		}
1826 		if ((tp->t_flags & TF_NOOPT) == 0)
1827 			incl_opts = true;
1828 	}
1829 	if (m == NULL) {
1830 		m = m_gethdr(M_NOWAIT, MT_DATA);
1831 		if (m == NULL)
1832 			return;
1833 		m->m_data += max_linkhdr;
1834 #ifdef INET6
1835 		if (isipv6) {
1836 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1837 			      sizeof(struct ip6_hdr));
1838 			ip6 = mtod(m, struct ip6_hdr *);
1839 			nth = (struct tcphdr *)(ip6 + 1);
1840 			if (port) {
1841 				/* Insert a UDP header */
1842 				uh = (struct udphdr *)nth;
1843 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1844 				uh->uh_dport = port;
1845 				nth = (struct tcphdr *)(uh + 1);
1846 			}
1847 		} else
1848 #endif /* INET6 */
1849 		{
1850 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1851 			ip = mtod(m, struct ip *);
1852 			nth = (struct tcphdr *)(ip + 1);
1853 			if (port) {
1854 				/* Insert a UDP header */
1855 				uh = (struct udphdr *)nth;
1856 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1857 				uh->uh_dport = port;
1858 				nth = (struct tcphdr *)(uh + 1);
1859 			}
1860 		}
1861 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1862 		flags = TH_ACK;
1863 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1864 		struct mbuf *n;
1865 
1866 		/* Can't reuse 'm', allocate a new mbuf. */
1867 		n = m_gethdr(M_NOWAIT, MT_DATA);
1868 		if (n == NULL) {
1869 			m_freem(m);
1870 			return;
1871 		}
1872 
1873 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1874 			m_freem(m);
1875 			m_freem(n);
1876 			return;
1877 		}
1878 
1879 		n->m_data += max_linkhdr;
1880 		/* m_len is set later */
1881 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1882 #ifdef INET6
1883 		if (isipv6) {
1884 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1885 			      sizeof(struct ip6_hdr));
1886 			ip6 = mtod(n, struct ip6_hdr *);
1887 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1888 			nth = (struct tcphdr *)(ip6 + 1);
1889 			if (port) {
1890 				/* Insert a UDP header */
1891 				uh = (struct udphdr *)nth;
1892 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1893 				uh->uh_dport = port;
1894 				nth = (struct tcphdr *)(uh + 1);
1895 			}
1896 		} else
1897 #endif /* INET6 */
1898 		{
1899 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1900 			ip = mtod(n, struct ip *);
1901 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1902 			nth = (struct tcphdr *)(ip + 1);
1903 			if (port) {
1904 				/* Insert a UDP header */
1905 				uh = (struct udphdr *)nth;
1906 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1907 				uh->uh_dport = port;
1908 				nth = (struct tcphdr *)(uh + 1);
1909 			}
1910 		}
1911 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1912 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1913 		th = nth;
1914 		m_freem(m);
1915 		m = n;
1916 	} else {
1917 		/*
1918 		 *  reuse the mbuf.
1919 		 * XXX MRT We inherit the FIB, which is lucky.
1920 		 */
1921 		m_freem(m->m_next);
1922 		m->m_next = NULL;
1923 		m->m_data = (caddr_t)ipgen;
1924 		/* m_len is set later */
1925 #ifdef INET6
1926 		if (isipv6) {
1927 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1928 			nth = (struct tcphdr *)(ip6 + 1);
1929 		} else
1930 #endif /* INET6 */
1931 		{
1932 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1933 			nth = (struct tcphdr *)(ip + 1);
1934 		}
1935 		if (th != nth) {
1936 			/*
1937 			 * this is usually a case when an extension header
1938 			 * exists between the IPv6 header and the
1939 			 * TCP header.
1940 			 */
1941 			nth->th_sport = th->th_sport;
1942 			nth->th_dport = th->th_dport;
1943 		}
1944 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1945 #undef xchg
1946 	}
1947 	tlen = 0;
1948 #ifdef INET6
1949 	if (isipv6)
1950 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1951 #endif
1952 #if defined(INET) && defined(INET6)
1953 	else
1954 #endif
1955 #ifdef INET
1956 		tlen = sizeof (struct tcpiphdr);
1957 #endif
1958 	if (port)
1959 		tlen += sizeof (struct udphdr);
1960 #ifdef INVARIANTS
1961 	m->m_len = 0;
1962 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1963 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1964 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1965 #endif
1966 	m->m_len = tlen;
1967 	to.to_flags = 0;
1968 	if (incl_opts) {
1969 		ect = tcp_ecn_output_established(tp, &flags, 0, false);
1970 		/* Make sure we have room. */
1971 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1972 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1973 			if (m->m_next) {
1974 				optp = mtod(m->m_next, u_char *);
1975 				optm = m->m_next;
1976 			} else
1977 				incl_opts = false;
1978 		} else {
1979 			optp = (u_char *) (nth + 1);
1980 			optm = m;
1981 		}
1982 	}
1983 	if (incl_opts) {
1984 		/* Timestamps. */
1985 		if (tp->t_flags & TF_RCVD_TSTMP) {
1986 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1987 			to.to_tsecr = tp->ts_recent;
1988 			to.to_flags |= TOF_TS;
1989 		}
1990 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1991 		/* TCP-MD5 (RFC2385). */
1992 		if (tp->t_flags & TF_SIGNATURE)
1993 			to.to_flags |= TOF_SIGNATURE;
1994 #endif
1995 		/* Add the options. */
1996 		tlen += optlen = tcp_addoptions(&to, optp);
1997 
1998 		/* Update m_len in the correct mbuf. */
1999 		optm->m_len += optlen;
2000 	} else
2001 		optlen = 0;
2002 #ifdef INET6
2003 	if (isipv6) {
2004 		if (uh) {
2005 			ulen = tlen - sizeof(struct ip6_hdr);
2006 			uh->uh_ulen = htons(ulen);
2007 		}
2008 		ip6->ip6_flow = htonl(ect << 20);
2009 		ip6->ip6_vfc = IPV6_VERSION;
2010 		if (port)
2011 			ip6->ip6_nxt = IPPROTO_UDP;
2012 		else
2013 			ip6->ip6_nxt = IPPROTO_TCP;
2014 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
2015 	}
2016 #endif
2017 #if defined(INET) && defined(INET6)
2018 	else
2019 #endif
2020 #ifdef INET
2021 	{
2022 		if (uh) {
2023 			ulen = tlen - sizeof(struct ip);
2024 			uh->uh_ulen = htons(ulen);
2025 		}
2026 		ip->ip_tos = ect;
2027 		ip->ip_len = htons(tlen);
2028 		ip->ip_ttl = V_ip_defttl;
2029 		if (port) {
2030 			ip->ip_p = IPPROTO_UDP;
2031 		} else {
2032 			ip->ip_p = IPPROTO_TCP;
2033 		}
2034 		if (V_path_mtu_discovery)
2035 			ip->ip_off |= htons(IP_DF);
2036 	}
2037 #endif
2038 	m->m_pkthdr.len = tlen;
2039 	m->m_pkthdr.rcvif = NULL;
2040 #ifdef MAC
2041 	if (inp != NULL) {
2042 		/*
2043 		 * Packet is associated with a socket, so allow the
2044 		 * label of the response to reflect the socket label.
2045 		 */
2046 		INP_LOCK_ASSERT(inp);
2047 		mac_inpcb_create_mbuf(inp, m);
2048 	} else {
2049 		/*
2050 		 * Packet is not associated with a socket, so possibly
2051 		 * update the label in place.
2052 		 */
2053 		mac_netinet_tcp_reply(m);
2054 	}
2055 #endif
2056 	nth->th_seq = htonl(seq);
2057 	nth->th_ack = htonl(ack);
2058 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2059 	tcp_set_flags(nth, flags);
2060 	if (tp != NULL)
2061 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2062 	else
2063 		nth->th_win = htons((u_short)win);
2064 	nth->th_urp = 0;
2065 
2066 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2067 	if (to.to_flags & TOF_SIGNATURE) {
2068 		if (!TCPMD5_ENABLED() ||
2069 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2070 			m_freem(m);
2071 			return;
2072 		}
2073 	}
2074 #endif
2075 
2076 #ifdef INET6
2077 	if (isipv6) {
2078 		if (port) {
2079 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2080 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2081 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2082 			nth->th_sum = 0;
2083 		} else {
2084 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2085 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2086 			nth->th_sum = in6_cksum_pseudo(ip6,
2087 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2088 		}
2089 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
2090 	}
2091 #endif /* INET6 */
2092 #if defined(INET6) && defined(INET)
2093 	else
2094 #endif
2095 #ifdef INET
2096 	{
2097 		if (port) {
2098 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2099 			    htons(ulen + IPPROTO_UDP));
2100 			m->m_pkthdr.csum_flags = CSUM_UDP;
2101 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2102 			nth->th_sum = 0;
2103 		} else {
2104 			m->m_pkthdr.csum_flags = CSUM_TCP;
2105 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2106 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2107 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2108 		}
2109 	}
2110 #endif /* INET */
2111 	TCP_PROBE3(debug__output, tp, th, m);
2112 	if (flags & TH_RST)
2113 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2114 	lgb = NULL;
2115 	if ((tp != NULL) && tcp_bblogging_on(tp)) {
2116 		if (INP_WLOCKED(inp)) {
2117 			union tcp_log_stackspecific log;
2118 			struct timeval tv;
2119 
2120 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2121 			log.u_bbr.inhpts = inp->inp_in_hpts;
2122 			log.u_bbr.flex8 = 4;
2123 			log.u_bbr.pkts_out = tp->t_maxseg;
2124 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2125 			log.u_bbr.delivered = 0;
2126 			lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT,
2127 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2128 		} else {
2129 			/*
2130 			 * We can not log the packet, since we only own the
2131 			 * read lock, but a write lock is needed. The read lock
2132 			 * is not upgraded to a write lock, since only getting
2133 			 * the read lock was done intentionally to improve the
2134 			 * handling of SYN flooding attacks.
2135 			 * This happens only for pure SYN segments received in
2136 			 * the initial CLOSED state, or received in a more
2137 			 * advanced state than listen and the UDP encapsulation
2138 			 * port is unexpected.
2139 			 * The incoming SYN segments do not really belong to
2140 			 * the TCP connection and the handling does not change
2141 			 * the state of the TCP connection. Therefore, the
2142 			 * sending of the RST segments is not logged. Please
2143 			 * note that also the incoming SYN segments are not
2144 			 * logged.
2145 			 *
2146 			 * The following code ensures that the above description
2147 			 * is and stays correct.
2148 			 */
2149 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2150 			    (tp->t_state == TCPS_CLOSED ||
2151 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2152 			    ("%s: Logging of TCP segment with flags 0x%b and "
2153 			    "UDP encapsulation port %u skipped in state %s",
2154 			    __func__, thflags, PRINT_TH_FLAGS,
2155 			    ntohs(port), tcpstates[tp->t_state]));
2156 		}
2157 	}
2158 
2159 	if (flags & TH_ACK)
2160 		TCPSTAT_INC(tcps_sndacks);
2161 	else if (flags & (TH_SYN|TH_FIN|TH_RST))
2162 		TCPSTAT_INC(tcps_sndctrl);
2163 	TCPSTAT_INC(tcps_sndtotal);
2164 
2165 #ifdef INET6
2166 	if (isipv6) {
2167 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2168 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2169 	}
2170 #endif /* INET6 */
2171 #if defined(INET) && defined(INET6)
2172 	else
2173 #endif
2174 #ifdef INET
2175 	{
2176 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2177 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2178 	}
2179 #endif
2180 	if (lgb != NULL)
2181 		lgb->tlb_errno = output_ret;
2182 }
2183 
2184 /*
2185  * Create a new TCP control block, making an empty reassembly queue and hooking
2186  * it to the argument protocol control block.  The `inp' parameter must have
2187  * come from the zone allocator set up by tcpcbstor declaration.
2188  */
2189 struct tcpcb *
2190 tcp_newtcpcb(struct inpcb *inp)
2191 {
2192 	struct tcpcb *tp = intotcpcb(inp);
2193 #ifdef INET6
2194 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2195 #endif /* INET6 */
2196 
2197 	/*
2198 	 * Historically allocation was done with M_ZERO.  There is a lot of
2199 	 * code that rely on that.  For now take safe approach and zero whole
2200 	 * tcpcb.  This definitely can be optimized.
2201 	 */
2202 	bzero(&tp->t_start_zero, t_zero_size);
2203 
2204 	/* Initialise cc_var struct for this tcpcb. */
2205 	tp->t_ccv.type = IPPROTO_TCP;
2206 	tp->t_ccv.ccvc.tcp = tp;
2207 	rw_rlock(&tcp_function_lock);
2208 	tp->t_fb = tcp_func_set_ptr;
2209 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2210 	rw_runlock(&tcp_function_lock);
2211 	/*
2212 	 * Use the current system default CC algorithm.
2213 	 */
2214 	cc_attach(tp, CC_DEFAULT_ALGO());
2215 
2216 	if (CC_ALGO(tp)->cb_init != NULL)
2217 		if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) {
2218 			cc_detach(tp);
2219 			if (tp->t_fb->tfb_tcp_fb_fini)
2220 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2221 			refcount_release(&tp->t_fb->tfb_refcnt);
2222 			return (NULL);
2223 		}
2224 
2225 #ifdef TCP_HHOOK
2226 	if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) {
2227 		if (tp->t_fb->tfb_tcp_fb_fini)
2228 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2229 		refcount_release(&tp->t_fb->tfb_refcnt);
2230 		return (NULL);
2231 	}
2232 #endif
2233 
2234 	TAILQ_INIT(&tp->t_segq);
2235 	tp->t_maxseg =
2236 #ifdef INET6
2237 		isipv6 ? V_tcp_v6mssdflt :
2238 #endif /* INET6 */
2239 		V_tcp_mssdflt;
2240 
2241 	callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED);
2242 	for (int i = 0; i < TT_N; i++)
2243 		tp->t_timers[i] = SBT_MAX;
2244 
2245 	switch (V_tcp_do_rfc1323) {
2246 		case 0:
2247 			break;
2248 		default:
2249 		case 1:
2250 			tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2251 			break;
2252 		case 2:
2253 			tp->t_flags = TF_REQ_SCALE;
2254 			break;
2255 		case 3:
2256 			tp->t_flags = TF_REQ_TSTMP;
2257 			break;
2258 	}
2259 	if (V_tcp_do_sack)
2260 		tp->t_flags |= TF_SACK_PERMIT;
2261 	TAILQ_INIT(&tp->snd_holes);
2262 
2263 	/*
2264 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2265 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2266 	 * reasonable initial retransmit time.
2267 	 */
2268 	tp->t_srtt = TCPTV_SRTTBASE;
2269 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2270 	tp->t_rttmin = tcp_rexmit_min;
2271 	tp->t_rxtcur = tcp_rexmit_initial;
2272 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2273 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2274 	tp->t_rcvtime = ticks;
2275 	/* We always start with ticks granularity */
2276 	tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
2277 	/*
2278 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2279 	 * because the socket may be bound to an IPv6 wildcard address,
2280 	 * which may match an IPv4-mapped IPv6 address.
2281 	 */
2282 	inp->inp_ip_ttl = V_ip_defttl;
2283 #ifdef TCPHPTS
2284 	/*
2285 	 * If using hpts lets drop a random number in so
2286 	 * not all new connections fall on the same CPU.
2287 	 */
2288 	inp->inp_hpts_cpu = hpts_random_cpu(inp);
2289 #endif
2290 #ifdef TCPPCAP
2291 	/*
2292 	 * Init the TCP PCAP queues.
2293 	 */
2294 	tcp_pcap_tcpcb_init(tp);
2295 #endif
2296 #ifdef TCP_BLACKBOX
2297 	/* Initialize the per-TCPCB log data. */
2298 	tcp_log_tcpcbinit(tp);
2299 #endif
2300 	tp->t_pacing_rate = -1;
2301 	if (tp->t_fb->tfb_tcp_fb_init) {
2302 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) {
2303 			refcount_release(&tp->t_fb->tfb_refcnt);
2304 			return (NULL);
2305 		}
2306 	}
2307 #ifdef STATS
2308 	if (V_tcp_perconn_stats_enable == 1)
2309 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2310 #endif
2311 	if (V_tcp_do_lrd)
2312 		tp->t_flags |= TF_LRD;
2313 
2314 	return (tp);
2315 }
2316 
2317 /*
2318  * Drop a TCP connection, reporting
2319  * the specified error.  If connection is synchronized,
2320  * then send a RST to peer.
2321  */
2322 struct tcpcb *
2323 tcp_drop(struct tcpcb *tp, int errno)
2324 {
2325 	struct socket *so = tptosocket(tp);
2326 
2327 	NET_EPOCH_ASSERT();
2328 	INP_WLOCK_ASSERT(tptoinpcb(tp));
2329 
2330 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2331 		tcp_state_change(tp, TCPS_CLOSED);
2332 		/* Don't use tcp_output() here due to possible recursion. */
2333 		(void)tcp_output_nodrop(tp);
2334 		TCPSTAT_INC(tcps_drops);
2335 	} else
2336 		TCPSTAT_INC(tcps_conndrops);
2337 	if (errno == ETIMEDOUT && tp->t_softerror)
2338 		errno = tp->t_softerror;
2339 	so->so_error = errno;
2340 	return (tcp_close(tp));
2341 }
2342 
2343 void
2344 tcp_discardcb(struct tcpcb *tp)
2345 {
2346 	struct inpcb *inp = tptoinpcb(tp);
2347 	struct socket *so = tptosocket(tp);
2348 #ifdef INET6
2349 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2350 #endif
2351 
2352 	INP_WLOCK_ASSERT(inp);
2353 
2354 	tcp_timer_stop(tp);
2355 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2356 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2357 	}
2358 
2359 	/* free the reassembly queue, if any */
2360 	tcp_reass_flush(tp);
2361 
2362 #ifdef TCP_OFFLOAD
2363 	/* Disconnect offload device, if any. */
2364 	if (tp->t_flags & TF_TOE)
2365 		tcp_offload_detach(tp);
2366 #endif
2367 
2368 	tcp_free_sackholes(tp);
2369 
2370 #ifdef TCPPCAP
2371 	/* Free the TCP PCAP queues. */
2372 	tcp_pcap_drain(&(tp->t_inpkts));
2373 	tcp_pcap_drain(&(tp->t_outpkts));
2374 #endif
2375 
2376 	/* Allow the CC algorithm to clean up after itself. */
2377 	if (CC_ALGO(tp)->cb_destroy != NULL)
2378 		CC_ALGO(tp)->cb_destroy(&tp->t_ccv);
2379 	CC_DATA(tp) = NULL;
2380 	/* Detach from the CC algorithm */
2381 	cc_detach(tp);
2382 
2383 #ifdef TCP_HHOOK
2384 	khelp_destroy_osd(&tp->t_osd);
2385 #endif
2386 #ifdef STATS
2387 	stats_blob_destroy(tp->t_stats);
2388 #endif
2389 
2390 	CC_ALGO(tp) = NULL;
2391 
2392 #ifdef TCP_BLACKBOX
2393 	tcp_log_tcpcbfini(tp);
2394 #endif
2395 	TCPSTATES_DEC(tp->t_state);
2396 	if (tp->t_fb->tfb_tcp_fb_fini)
2397 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2398 
2399 	/*
2400 	 * If we got enough samples through the srtt filter,
2401 	 * save the rtt and rttvar in the routing entry.
2402 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2403 	 * 4 samples is enough for the srtt filter to converge
2404 	 * to within enough % of the correct value; fewer samples
2405 	 * and we could save a bogus rtt. The danger is not high
2406 	 * as tcp quickly recovers from everything.
2407 	 * XXX: Works very well but needs some more statistics!
2408 	 *
2409 	 * XXXRRS: Updating must be after the stack fini() since
2410 	 * that may be converting some internal representation of
2411 	 * say srtt etc into the general one used by other stacks.
2412 	 * Lets also at least protect against the so being NULL
2413 	 * as RW stated below.
2414 	 */
2415 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2416 		struct hc_metrics_lite metrics;
2417 		uint32_t ssthresh;
2418 
2419 		bzero(&metrics, sizeof(metrics));
2420 		/*
2421 		 * Update the ssthresh always when the conditions below
2422 		 * are satisfied. This gives us better new start value
2423 		 * for the congestion avoidance for new connections.
2424 		 * ssthresh is only set if packet loss occurred on a session.
2425 		 *
2426 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2427 		 * being torn down.  Ideally this code would not use 'so'.
2428 		 */
2429 		ssthresh = tp->snd_ssthresh;
2430 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2431 			/*
2432 			 * convert the limit from user data bytes to
2433 			 * packets then to packet data bytes.
2434 			 */
2435 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2436 			if (ssthresh < 2)
2437 				ssthresh = 2;
2438 			ssthresh *= (tp->t_maxseg +
2439 #ifdef INET6
2440 			    (isipv6 ? sizeof (struct ip6_hdr) +
2441 			    sizeof (struct tcphdr) :
2442 #endif
2443 			    sizeof (struct tcpiphdr)
2444 #ifdef INET6
2445 			    )
2446 #endif
2447 			    );
2448 		} else
2449 			ssthresh = 0;
2450 		metrics.rmx_ssthresh = ssthresh;
2451 
2452 		metrics.rmx_rtt = tp->t_srtt;
2453 		metrics.rmx_rttvar = tp->t_rttvar;
2454 		metrics.rmx_cwnd = tp->snd_cwnd;
2455 		metrics.rmx_sendpipe = 0;
2456 		metrics.rmx_recvpipe = 0;
2457 
2458 		tcp_hc_update(&inp->inp_inc, &metrics);
2459 	}
2460 
2461 	refcount_release(&tp->t_fb->tfb_refcnt);
2462 }
2463 
2464 /*
2465  * Attempt to close a TCP control block, marking it as dropped, and freeing
2466  * the socket if we hold the only reference.
2467  */
2468 struct tcpcb *
2469 tcp_close(struct tcpcb *tp)
2470 {
2471 	struct inpcb *inp = tptoinpcb(tp);
2472 	struct socket *so = tptosocket(tp);
2473 
2474 	INP_WLOCK_ASSERT(inp);
2475 
2476 #ifdef TCP_OFFLOAD
2477 	if (tp->t_state == TCPS_LISTEN)
2478 		tcp_offload_listen_stop(tp);
2479 #endif
2480 	/*
2481 	 * This releases the TFO pending counter resource for TFO listen
2482 	 * sockets as well as passively-created TFO sockets that transition
2483 	 * from SYN_RECEIVED to CLOSED.
2484 	 */
2485 	if (tp->t_tfo_pending) {
2486 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2487 		tp->t_tfo_pending = NULL;
2488 	}
2489 #ifdef TCPHPTS
2490 	tcp_hpts_remove(inp);
2491 #endif
2492 	in_pcbdrop(inp);
2493 	TCPSTAT_INC(tcps_closed);
2494 	if (tp->t_state != TCPS_CLOSED)
2495 		tcp_state_change(tp, TCPS_CLOSED);
2496 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2497 	soisdisconnected(so);
2498 	if (inp->inp_flags & INP_SOCKREF) {
2499 		inp->inp_flags &= ~INP_SOCKREF;
2500 		INP_WUNLOCK(inp);
2501 		sorele(so);
2502 		return (NULL);
2503 	}
2504 	return (tp);
2505 }
2506 
2507 /*
2508  * Notify a tcp user of an asynchronous error;
2509  * store error as soft error, but wake up user
2510  * (for now, won't do anything until can select for soft error).
2511  *
2512  * Do not wake up user since there currently is no mechanism for
2513  * reporting soft errors (yet - a kqueue filter may be added).
2514  */
2515 static struct inpcb *
2516 tcp_notify(struct inpcb *inp, int error)
2517 {
2518 	struct tcpcb *tp;
2519 
2520 	INP_WLOCK_ASSERT(inp);
2521 
2522 	tp = intotcpcb(inp);
2523 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2524 
2525 	/*
2526 	 * Ignore some errors if we are hooked up.
2527 	 * If connection hasn't completed, has retransmitted several times,
2528 	 * and receives a second error, give up now.  This is better
2529 	 * than waiting a long time to establish a connection that
2530 	 * can never complete.
2531 	 */
2532 	if (tp->t_state == TCPS_ESTABLISHED &&
2533 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2534 	     error == EHOSTDOWN)) {
2535 		if (inp->inp_route.ro_nh) {
2536 			NH_FREE(inp->inp_route.ro_nh);
2537 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2538 		}
2539 		return (inp);
2540 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2541 	    tp->t_softerror) {
2542 		tp = tcp_drop(tp, error);
2543 		if (tp != NULL)
2544 			return (inp);
2545 		else
2546 			return (NULL);
2547 	} else {
2548 		tp->t_softerror = error;
2549 		return (inp);
2550 	}
2551 #if 0
2552 	wakeup( &so->so_timeo);
2553 	sorwakeup(so);
2554 	sowwakeup(so);
2555 #endif
2556 }
2557 
2558 static int
2559 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2560 {
2561 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2562 	    INPLOOKUP_RLOCKPCB);
2563 	struct xinpgen xig;
2564 	struct inpcb *inp;
2565 	int error;
2566 
2567 	if (req->newptr != NULL)
2568 		return (EPERM);
2569 
2570 	if (req->oldptr == NULL) {
2571 		int n;
2572 
2573 		n = V_tcbinfo.ipi_count +
2574 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2575 		n += imax(n / 8, 10);
2576 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2577 		return (0);
2578 	}
2579 
2580 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2581 		return (error);
2582 
2583 	bzero(&xig, sizeof(xig));
2584 	xig.xig_len = sizeof xig;
2585 	xig.xig_count = V_tcbinfo.ipi_count +
2586 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2587 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2588 	xig.xig_sogen = so_gencnt;
2589 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2590 	if (error)
2591 		return (error);
2592 
2593 	error = syncache_pcblist(req);
2594 	if (error)
2595 		return (error);
2596 
2597 	while ((inp = inp_next(&inpi)) != NULL) {
2598 		if (inp->inp_gencnt <= xig.xig_gen &&
2599 		    cr_canseeinpcb(req->td->td_ucred, inp) == 0) {
2600 			struct xtcpcb xt;
2601 
2602 			tcp_inptoxtp(inp, &xt);
2603 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2604 			if (error) {
2605 				INP_RUNLOCK(inp);
2606 				break;
2607 			} else
2608 				continue;
2609 		}
2610 	}
2611 
2612 	if (!error) {
2613 		/*
2614 		 * Give the user an updated idea of our state.
2615 		 * If the generation differs from what we told
2616 		 * her before, she knows that something happened
2617 		 * while we were processing this request, and it
2618 		 * might be necessary to retry.
2619 		 */
2620 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2621 		xig.xig_sogen = so_gencnt;
2622 		xig.xig_count = V_tcbinfo.ipi_count +
2623 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2624 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2625 	}
2626 
2627 	return (error);
2628 }
2629 
2630 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2631     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2632     NULL, 0, tcp_pcblist, "S,xtcpcb",
2633     "List of active TCP connections");
2634 
2635 #ifdef INET
2636 static int
2637 tcp_getcred(SYSCTL_HANDLER_ARGS)
2638 {
2639 	struct xucred xuc;
2640 	struct sockaddr_in addrs[2];
2641 	struct epoch_tracker et;
2642 	struct inpcb *inp;
2643 	int error;
2644 
2645 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2646 	if (error)
2647 		return (error);
2648 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2649 	if (error)
2650 		return (error);
2651 	NET_EPOCH_ENTER(et);
2652 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2653 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2654 	NET_EPOCH_EXIT(et);
2655 	if (inp != NULL) {
2656 		if (error == 0)
2657 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2658 		if (error == 0)
2659 			cru2x(inp->inp_cred, &xuc);
2660 		INP_RUNLOCK(inp);
2661 	} else
2662 		error = ENOENT;
2663 	if (error == 0)
2664 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2665 	return (error);
2666 }
2667 
2668 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2669     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2670     0, 0, tcp_getcred, "S,xucred",
2671     "Get the xucred of a TCP connection");
2672 #endif /* INET */
2673 
2674 #ifdef INET6
2675 static int
2676 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2677 {
2678 	struct epoch_tracker et;
2679 	struct xucred xuc;
2680 	struct sockaddr_in6 addrs[2];
2681 	struct inpcb *inp;
2682 	int error;
2683 #ifdef INET
2684 	int mapped = 0;
2685 #endif
2686 
2687 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2688 	if (error)
2689 		return (error);
2690 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2691 	if (error)
2692 		return (error);
2693 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2694 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2695 		return (error);
2696 	}
2697 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2698 #ifdef INET
2699 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2700 			mapped = 1;
2701 		else
2702 #endif
2703 			return (EINVAL);
2704 	}
2705 
2706 	NET_EPOCH_ENTER(et);
2707 #ifdef INET
2708 	if (mapped == 1)
2709 		inp = in_pcblookup(&V_tcbinfo,
2710 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2711 			addrs[1].sin6_port,
2712 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2713 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2714 	else
2715 #endif
2716 		inp = in6_pcblookup(&V_tcbinfo,
2717 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2718 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2719 			INPLOOKUP_RLOCKPCB, NULL);
2720 	NET_EPOCH_EXIT(et);
2721 	if (inp != NULL) {
2722 		if (error == 0)
2723 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2724 		if (error == 0)
2725 			cru2x(inp->inp_cred, &xuc);
2726 		INP_RUNLOCK(inp);
2727 	} else
2728 		error = ENOENT;
2729 	if (error == 0)
2730 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2731 	return (error);
2732 }
2733 
2734 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2735     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2736     0, 0, tcp6_getcred, "S,xucred",
2737     "Get the xucred of a TCP6 connection");
2738 #endif /* INET6 */
2739 
2740 #ifdef INET
2741 /* Path MTU to try next when a fragmentation-needed message is received. */
2742 static inline int
2743 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2744 {
2745 	int mtu = ntohs(icp->icmp_nextmtu);
2746 
2747 	/* If no alternative MTU was proposed, try the next smaller one. */
2748 	if (!mtu)
2749 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2750 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2751 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2752 
2753 	return (mtu);
2754 }
2755 
2756 static void
2757 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port)
2758 {
2759 	struct ip *ip;
2760 	struct tcphdr *th;
2761 	struct inpcb *inp;
2762 	struct tcpcb *tp;
2763 	struct inpcb *(*notify)(struct inpcb *, int);
2764 	struct in_conninfo inc;
2765 	tcp_seq icmp_tcp_seq;
2766 	int errno, mtu;
2767 
2768 	errno = icmp_errmap(icp);
2769 	switch (errno) {
2770 	case 0:
2771 		return;
2772 	case EMSGSIZE:
2773 		notify = tcp_mtudisc_notify;
2774 		break;
2775 	case ECONNREFUSED:
2776 		if (V_icmp_may_rst)
2777 			notify = tcp_drop_syn_sent;
2778 		else
2779 			notify = tcp_notify;
2780 		break;
2781 	case EHOSTUNREACH:
2782 		if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED)
2783 			notify = tcp_drop_syn_sent;
2784 		else
2785 			notify = tcp_notify;
2786 		break;
2787 	default:
2788 		notify = tcp_notify;
2789 	}
2790 
2791 	ip = &icp->icmp_ip;
2792 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2793 	icmp_tcp_seq = th->th_seq;
2794 	inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src,
2795 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2796 	if (inp != NULL)  {
2797 		tp = intotcpcb(inp);
2798 #ifdef TCP_OFFLOAD
2799 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2800 			/*
2801 			 * MTU discovery for offloaded connections.  Let
2802 			 * the TOE driver verify seq# and process it.
2803 			 */
2804 			mtu = tcp_next_pmtu(icp, ip);
2805 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2806 			goto out;
2807 		}
2808 #endif
2809 		if (tp->t_port != port)
2810 			goto out;
2811 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2812 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2813 			if (errno == EMSGSIZE) {
2814 				/*
2815 				 * MTU discovery: we got a needfrag and
2816 				 * will potentially try a lower MTU.
2817 				 */
2818 				mtu = tcp_next_pmtu(icp, ip);
2819 
2820 				/*
2821 				 * Only process the offered MTU if it
2822 				 * is smaller than the current one.
2823 				 */
2824 				if (mtu < tp->t_maxseg +
2825 				    sizeof(struct tcpiphdr)) {
2826 					bzero(&inc, sizeof(inc));
2827 					inc.inc_faddr = ip->ip_dst;
2828 					inc.inc_fibnum =
2829 					    inp->inp_inc.inc_fibnum;
2830 					tcp_hc_updatemtu(&inc, mtu);
2831 					inp = tcp_mtudisc(inp, mtu);
2832 				}
2833 			} else
2834 				inp = (*notify)(inp, errno);
2835 		}
2836 	} else {
2837 		bzero(&inc, sizeof(inc));
2838 		inc.inc_fport = th->th_dport;
2839 		inc.inc_lport = th->th_sport;
2840 		inc.inc_faddr = ip->ip_dst;
2841 		inc.inc_laddr = ip->ip_src;
2842 		syncache_unreach(&inc, icmp_tcp_seq, port);
2843 	}
2844 out:
2845 	if (inp != NULL)
2846 		INP_WUNLOCK(inp);
2847 }
2848 
2849 static void
2850 tcp_ctlinput(struct icmp *icmp)
2851 {
2852 	tcp_ctlinput_with_port(icmp, htons(0));
2853 }
2854 
2855 static void
2856 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param)
2857 {
2858 	/* Its a tunneled TCP over UDP icmp */
2859 	struct icmp *icmp = param.icmp;
2860 	struct ip *outer_ip, *inner_ip;
2861 	struct udphdr *udp;
2862 	struct tcphdr *th, ttemp;
2863 	int i_hlen, o_len;
2864 	uint16_t port;
2865 
2866 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2867 	inner_ip = &icmp->icmp_ip;
2868 	i_hlen = inner_ip->ip_hl << 2;
2869 	o_len = ntohs(outer_ip->ip_len);
2870 	if (o_len <
2871 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2872 		/* Not enough data present */
2873 		return;
2874 	}
2875 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2876 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2877 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2878 		return;
2879 	}
2880 	port = udp->uh_dport;
2881 	th = (struct tcphdr *)(udp + 1);
2882 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2883 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2884 	/* Now adjust down the size of the outer IP header */
2885 	o_len -= sizeof(struct udphdr);
2886 	outer_ip->ip_len = htons(o_len);
2887 	/* Now call in to the normal handling code */
2888 	tcp_ctlinput_with_port(icmp, port);
2889 }
2890 #endif /* INET */
2891 
2892 #ifdef INET6
2893 static inline int
2894 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2895 {
2896 	int mtu = ntohl(icmp6->icmp6_mtu);
2897 
2898 	/*
2899 	 * If no alternative MTU was proposed, or the proposed MTU was too
2900 	 * small, set to the min.
2901 	 */
2902 	if (mtu < IPV6_MMTU)
2903 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
2904 	return (mtu);
2905 }
2906 
2907 static void
2908 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port)
2909 {
2910 	struct in6_addr *dst;
2911 	struct inpcb *(*notify)(struct inpcb *, int);
2912 	struct ip6_hdr *ip6;
2913 	struct mbuf *m;
2914 	struct inpcb *inp;
2915 	struct tcpcb *tp;
2916 	struct icmp6_hdr *icmp6;
2917 	struct in_conninfo inc;
2918 	struct tcp_ports {
2919 		uint16_t th_sport;
2920 		uint16_t th_dport;
2921 	} t_ports;
2922 	tcp_seq icmp_tcp_seq;
2923 	unsigned int mtu;
2924 	unsigned int off;
2925 	int errno;
2926 
2927 	icmp6 = ip6cp->ip6c_icmp6;
2928 	m = ip6cp->ip6c_m;
2929 	ip6 = ip6cp->ip6c_ip6;
2930 	off = ip6cp->ip6c_off;
2931 	dst = &ip6cp->ip6c_finaldst->sin6_addr;
2932 
2933 	errno = icmp6_errmap(icmp6);
2934 	switch (errno) {
2935 	case 0:
2936 		return;
2937 	case EMSGSIZE:
2938 		notify = tcp_mtudisc_notify;
2939 		break;
2940 	case ECONNREFUSED:
2941 		if (V_icmp_may_rst)
2942 			notify = tcp_drop_syn_sent;
2943 		else
2944 			notify = tcp_notify;
2945 		break;
2946 	case EHOSTUNREACH:
2947 		/*
2948 		 * There are only four ICMPs that may reset connection:
2949 		 * - administratively prohibited
2950 		 * - port unreachable
2951 		 * - time exceeded in transit
2952 		 * - unknown next header
2953 		 */
2954 		if (V_icmp_may_rst &&
2955 		    ((icmp6->icmp6_type == ICMP6_DST_UNREACH &&
2956 		     (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN ||
2957 		      icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) ||
2958 		    (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED &&
2959 		      icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) ||
2960 		    (icmp6->icmp6_type == ICMP6_PARAM_PROB &&
2961 		      icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER)))
2962 			notify = tcp_drop_syn_sent;
2963 		else
2964 			notify = tcp_notify;
2965 		break;
2966 	default:
2967 		notify = tcp_notify;
2968 	}
2969 
2970 	/* Check if we can safely get the ports from the tcp hdr */
2971 	if (m == NULL ||
2972 	    (m->m_pkthdr.len <
2973 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2974 		return;
2975 	}
2976 	bzero(&t_ports, sizeof(struct tcp_ports));
2977 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2978 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2979 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2980 	off += sizeof(struct tcp_ports);
2981 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2982 		goto out;
2983 	}
2984 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2985 	if (inp != NULL)  {
2986 		tp = intotcpcb(inp);
2987 #ifdef TCP_OFFLOAD
2988 		if (tp->t_flags & TF_TOE && errno == EMSGSIZE) {
2989 			/* MTU discovery for offloaded connections. */
2990 			mtu = tcp6_next_pmtu(icmp6);
2991 			tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2992 			goto out;
2993 		}
2994 #endif
2995 		if (tp->t_port != port)
2996 			goto out;
2997 		if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2998 		    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2999 			if (errno == EMSGSIZE) {
3000 				/*
3001 				 * MTU discovery:
3002 				 * If we got a needfrag set the MTU
3003 				 * in the route to the suggested new
3004 				 * value (if given) and then notify.
3005 				 */
3006 				mtu = tcp6_next_pmtu(icmp6);
3007 
3008 				bzero(&inc, sizeof(inc));
3009 				inc.inc_fibnum = M_GETFIB(m);
3010 				inc.inc_flags |= INC_ISIPV6;
3011 				inc.inc6_faddr = *dst;
3012 				if (in6_setscope(&inc.inc6_faddr,
3013 					m->m_pkthdr.rcvif, NULL))
3014 					goto out;
3015 				/*
3016 				 * Only process the offered MTU if it
3017 				 * is smaller than the current one.
3018 				 */
3019 				if (mtu < tp->t_maxseg +
3020 				    sizeof (struct tcphdr) +
3021 				    sizeof (struct ip6_hdr)) {
3022 					tcp_hc_updatemtu(&inc, mtu);
3023 					tcp_mtudisc(inp, mtu);
3024 					ICMP6STAT_INC(icp6s_pmtuchg);
3025 				}
3026 			} else
3027 				inp = (*notify)(inp, errno);
3028 		}
3029 	} else {
3030 		bzero(&inc, sizeof(inc));
3031 		inc.inc_fibnum = M_GETFIB(m);
3032 		inc.inc_flags |= INC_ISIPV6;
3033 		inc.inc_fport = t_ports.th_dport;
3034 		inc.inc_lport = t_ports.th_sport;
3035 		inc.inc6_faddr = *dst;
3036 		inc.inc6_laddr = ip6->ip6_src;
3037 		syncache_unreach(&inc, icmp_tcp_seq, port);
3038 	}
3039 out:
3040 	if (inp != NULL)
3041 		INP_WUNLOCK(inp);
3042 }
3043 
3044 static void
3045 tcp6_ctlinput(struct ip6ctlparam *ctl)
3046 {
3047 	tcp6_ctlinput_with_port(ctl, htons(0));
3048 }
3049 
3050 static void
3051 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param)
3052 {
3053 	struct ip6ctlparam *ip6cp = param.ip6cp;
3054 	struct mbuf *m;
3055 	struct udphdr *udp;
3056 	uint16_t port;
3057 
3058 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3059 	if (m == NULL) {
3060 		return;
3061 	}
3062 	udp = mtod(m, struct udphdr *);
3063 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3064 		return;
3065 	}
3066 	port = udp->uh_dport;
3067 	m_adj(m, sizeof(struct udphdr));
3068 	if ((m->m_flags & M_PKTHDR) == 0) {
3069 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3070 	}
3071 	/* Now call in to the normal handling code */
3072 	tcp6_ctlinput_with_port(ip6cp, port);
3073 }
3074 
3075 #endif /* INET6 */
3076 
3077 static uint32_t
3078 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3079 {
3080 	SIPHASH_CTX ctx;
3081 	uint32_t hash[2];
3082 
3083 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3084 	    ("%s: keylen %u too short ", __func__, len));
3085 	SipHash24_Init(&ctx);
3086 	SipHash_SetKey(&ctx, (uint8_t *)key);
3087 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3088 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3089 	switch (inc->inc_flags & INC_ISIPV6) {
3090 #ifdef INET
3091 	case 0:
3092 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3093 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3094 		break;
3095 #endif
3096 #ifdef INET6
3097 	case INC_ISIPV6:
3098 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3099 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3100 		break;
3101 #endif
3102 	}
3103 	SipHash_Final((uint8_t *)hash, &ctx);
3104 
3105 	return (hash[0] ^ hash[1]);
3106 }
3107 
3108 uint32_t
3109 tcp_new_ts_offset(struct in_conninfo *inc)
3110 {
3111 	struct in_conninfo inc_store, *local_inc;
3112 
3113 	if (!V_tcp_ts_offset_per_conn) {
3114 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3115 		inc_store.inc_lport = 0;
3116 		inc_store.inc_fport = 0;
3117 		local_inc = &inc_store;
3118 	} else {
3119 		local_inc = inc;
3120 	}
3121 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3122 	    sizeof(V_ts_offset_secret)));
3123 }
3124 
3125 /*
3126  * Following is where TCP initial sequence number generation occurs.
3127  *
3128  * There are two places where we must use initial sequence numbers:
3129  * 1.  In SYN-ACK packets.
3130  * 2.  In SYN packets.
3131  *
3132  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3133  * tcp_syncache.c for details.
3134  *
3135  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3136  * depends on this property.  In addition, these ISNs should be
3137  * unguessable so as to prevent connection hijacking.  To satisfy
3138  * the requirements of this situation, the algorithm outlined in
3139  * RFC 1948 is used, with only small modifications.
3140  *
3141  * Implementation details:
3142  *
3143  * Time is based off the system timer, and is corrected so that it
3144  * increases by one megabyte per second.  This allows for proper
3145  * recycling on high speed LANs while still leaving over an hour
3146  * before rollover.
3147  *
3148  * As reading the *exact* system time is too expensive to be done
3149  * whenever setting up a TCP connection, we increment the time
3150  * offset in two ways.  First, a small random positive increment
3151  * is added to isn_offset for each connection that is set up.
3152  * Second, the function tcp_isn_tick fires once per clock tick
3153  * and increments isn_offset as necessary so that sequence numbers
3154  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3155  * random positive increments serve only to ensure that the same
3156  * exact sequence number is never sent out twice (as could otherwise
3157  * happen when a port is recycled in less than the system tick
3158  * interval.)
3159  *
3160  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3161  * between seeding of isn_secret.  This is normally set to zero,
3162  * as reseeding should not be necessary.
3163  *
3164  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3165  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3166  * general, this means holding an exclusive (write) lock.
3167  */
3168 
3169 #define ISN_BYTES_PER_SECOND 1048576
3170 #define ISN_STATIC_INCREMENT 4096
3171 #define ISN_RANDOM_INCREMENT (4096 - 1)
3172 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3173 
3174 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3175 VNET_DEFINE_STATIC(int, isn_last);
3176 VNET_DEFINE_STATIC(int, isn_last_reseed);
3177 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3178 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3179 
3180 #define	V_isn_secret			VNET(isn_secret)
3181 #define	V_isn_last			VNET(isn_last)
3182 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3183 #define	V_isn_offset			VNET(isn_offset)
3184 #define	V_isn_offset_old		VNET(isn_offset_old)
3185 
3186 tcp_seq
3187 tcp_new_isn(struct in_conninfo *inc)
3188 {
3189 	tcp_seq new_isn;
3190 	u_int32_t projected_offset;
3191 
3192 	ISN_LOCK();
3193 	/* Seed if this is the first use, reseed if requested. */
3194 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3195 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3196 		< (u_int)ticks))) {
3197 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3198 		V_isn_last_reseed = ticks;
3199 	}
3200 
3201 	/* Compute the hash and return the ISN. */
3202 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3203 	    sizeof(V_isn_secret));
3204 	V_isn_offset += ISN_STATIC_INCREMENT +
3205 		(arc4random() & ISN_RANDOM_INCREMENT);
3206 	if (ticks != V_isn_last) {
3207 		projected_offset = V_isn_offset_old +
3208 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3209 		if (SEQ_GT(projected_offset, V_isn_offset))
3210 			V_isn_offset = projected_offset;
3211 		V_isn_offset_old = V_isn_offset;
3212 		V_isn_last = ticks;
3213 	}
3214 	new_isn += V_isn_offset;
3215 	ISN_UNLOCK();
3216 	return (new_isn);
3217 }
3218 
3219 /*
3220  * When a specific ICMP unreachable message is received and the
3221  * connection state is SYN-SENT, drop the connection.  This behavior
3222  * is controlled by the icmp_may_rst sysctl.
3223  */
3224 static struct inpcb *
3225 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3226 {
3227 	struct tcpcb *tp;
3228 
3229 	NET_EPOCH_ASSERT();
3230 	INP_WLOCK_ASSERT(inp);
3231 
3232 	tp = intotcpcb(inp);
3233 	if (tp->t_state != TCPS_SYN_SENT)
3234 		return (inp);
3235 
3236 	if (IS_FASTOPEN(tp->t_flags))
3237 		tcp_fastopen_disable_path(tp);
3238 
3239 	tp = tcp_drop(tp, errno);
3240 	if (tp != NULL)
3241 		return (inp);
3242 	else
3243 		return (NULL);
3244 }
3245 
3246 /*
3247  * When `need fragmentation' ICMP is received, update our idea of the MSS
3248  * based on the new value. Also nudge TCP to send something, since we
3249  * know the packet we just sent was dropped.
3250  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3251  */
3252 static struct inpcb *
3253 tcp_mtudisc_notify(struct inpcb *inp, int error)
3254 {
3255 
3256 	return (tcp_mtudisc(inp, -1));
3257 }
3258 
3259 static struct inpcb *
3260 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3261 {
3262 	struct tcpcb *tp;
3263 	struct socket *so;
3264 
3265 	INP_WLOCK_ASSERT(inp);
3266 
3267 	tp = intotcpcb(inp);
3268 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3269 
3270 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3271 
3272 	so = inp->inp_socket;
3273 	SOCKBUF_LOCK(&so->so_snd);
3274 	/* If the mss is larger than the socket buffer, decrease the mss. */
3275 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3276 		tp->t_maxseg = so->so_snd.sb_hiwat;
3277 	SOCKBUF_UNLOCK(&so->so_snd);
3278 
3279 	TCPSTAT_INC(tcps_mturesent);
3280 	tp->t_rtttime = 0;
3281 	tp->snd_nxt = tp->snd_una;
3282 	tcp_free_sackholes(tp);
3283 	tp->snd_recover = tp->snd_max;
3284 	if (tp->t_flags & TF_SACK_PERMIT)
3285 		EXIT_FASTRECOVERY(tp->t_flags);
3286 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3287 		/*
3288 		 * Conceptually the snd_nxt setting
3289 		 * and freeing sack holes should
3290 		 * be done by the default stacks
3291 		 * own tfb_tcp_mtu_chg().
3292 		 */
3293 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3294 	}
3295 	if (tcp_output(tp) < 0)
3296 		return (NULL);
3297 	else
3298 		return (inp);
3299 }
3300 
3301 #ifdef INET
3302 /*
3303  * Look-up the routing entry to the peer of this inpcb.  If no route
3304  * is found and it cannot be allocated, then return 0.  This routine
3305  * is called by TCP routines that access the rmx structure and by
3306  * tcp_mss_update to get the peer/interface MTU.
3307  */
3308 uint32_t
3309 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3310 {
3311 	struct nhop_object *nh;
3312 	struct ifnet *ifp;
3313 	uint32_t maxmtu = 0;
3314 
3315 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3316 
3317 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3318 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3319 		if (nh == NULL)
3320 			return (0);
3321 
3322 		ifp = nh->nh_ifp;
3323 		maxmtu = nh->nh_mtu;
3324 
3325 		/* Report additional interface capabilities. */
3326 		if (cap != NULL) {
3327 			if (ifp->if_capenable & IFCAP_TSO4 &&
3328 			    ifp->if_hwassist & CSUM_TSO) {
3329 				cap->ifcap |= CSUM_TSO;
3330 				cap->tsomax = ifp->if_hw_tsomax;
3331 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3332 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3333 			}
3334 		}
3335 	}
3336 	return (maxmtu);
3337 }
3338 #endif /* INET */
3339 
3340 #ifdef INET6
3341 uint32_t
3342 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3343 {
3344 	struct nhop_object *nh;
3345 	struct in6_addr dst6;
3346 	uint32_t scopeid;
3347 	struct ifnet *ifp;
3348 	uint32_t maxmtu = 0;
3349 
3350 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3351 
3352 	if (inc->inc_flags & INC_IPV6MINMTU)
3353 		return (IPV6_MMTU);
3354 
3355 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3356 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3357 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3358 		if (nh == NULL)
3359 			return (0);
3360 
3361 		ifp = nh->nh_ifp;
3362 		maxmtu = nh->nh_mtu;
3363 
3364 		/* Report additional interface capabilities. */
3365 		if (cap != NULL) {
3366 			if (ifp->if_capenable & IFCAP_TSO6 &&
3367 			    ifp->if_hwassist & CSUM_TSO) {
3368 				cap->ifcap |= CSUM_TSO;
3369 				cap->tsomax = ifp->if_hw_tsomax;
3370 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3371 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3372 			}
3373 		}
3374 	}
3375 
3376 	return (maxmtu);
3377 }
3378 
3379 /*
3380  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3381  *
3382  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3383  * The right place to do that is ip6_setpktopt() that has just been
3384  * executed.  By the way it just filled ip6po_minmtu for us.
3385  */
3386 void
3387 tcp6_use_min_mtu(struct tcpcb *tp)
3388 {
3389 	struct inpcb *inp = tptoinpcb(tp);
3390 
3391 	INP_WLOCK_ASSERT(inp);
3392 	/*
3393 	 * In case of the IPV6_USE_MIN_MTU socket
3394 	 * option, the INC_IPV6MINMTU flag to announce
3395 	 * a corresponding MSS during the initial
3396 	 * handshake.  If the TCP connection is not in
3397 	 * the front states, just reduce the MSS being
3398 	 * used.  This avoids the sending of TCP
3399 	 * segments which will be fragmented at the
3400 	 * IPv6 layer.
3401 	 */
3402 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3403 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3404 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3405 		struct ip6_pktopts *opt;
3406 
3407 		opt = inp->in6p_outputopts;
3408 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3409 		    tp->t_maxseg > TCP6_MSS)
3410 			tp->t_maxseg = TCP6_MSS;
3411 	}
3412 }
3413 #endif /* INET6 */
3414 
3415 /*
3416  * Calculate effective SMSS per RFC5681 definition for a given TCP
3417  * connection at its current state, taking into account SACK and etc.
3418  */
3419 u_int
3420 tcp_maxseg(const struct tcpcb *tp)
3421 {
3422 	u_int optlen;
3423 
3424 	if (tp->t_flags & TF_NOOPT)
3425 		return (tp->t_maxseg);
3426 
3427 	/*
3428 	 * Here we have a simplified code from tcp_addoptions(),
3429 	 * without a proper loop, and having most of paddings hardcoded.
3430 	 * We might make mistakes with padding here in some edge cases,
3431 	 * but this is harmless, since result of tcp_maxseg() is used
3432 	 * only in cwnd and ssthresh estimations.
3433 	 */
3434 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3435 		if (tp->t_flags & TF_RCVD_TSTMP)
3436 			optlen = TCPOLEN_TSTAMP_APPA;
3437 		else
3438 			optlen = 0;
3439 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3440 		if (tp->t_flags & TF_SIGNATURE)
3441 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3442 #endif
3443 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3444 			optlen += TCPOLEN_SACKHDR;
3445 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3446 			optlen = PADTCPOLEN(optlen);
3447 		}
3448 	} else {
3449 		if (tp->t_flags & TF_REQ_TSTMP)
3450 			optlen = TCPOLEN_TSTAMP_APPA;
3451 		else
3452 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3453 		if (tp->t_flags & TF_REQ_SCALE)
3454 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3455 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3456 		if (tp->t_flags & TF_SIGNATURE)
3457 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3458 #endif
3459 		if (tp->t_flags & TF_SACK_PERMIT)
3460 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3461 	}
3462 #undef PAD
3463 	optlen = min(optlen, TCP_MAXOLEN);
3464 	return (tp->t_maxseg - optlen);
3465 }
3466 
3467 
3468 u_int
3469 tcp_fixed_maxseg(const struct tcpcb *tp)
3470 {
3471 	int optlen;
3472 
3473 	if (tp->t_flags & TF_NOOPT)
3474 		return (tp->t_maxseg);
3475 
3476 	/*
3477 	 * Here we have a simplified code from tcp_addoptions(),
3478 	 * without a proper loop, and having most of paddings hardcoded.
3479 	 * We only consider fixed options that we would send every
3480 	 * time I.e. SACK is not considered. This is important
3481 	 * for cc modules to figure out what the modulo of the
3482 	 * cwnd should be.
3483 	 */
3484 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3485 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3486 		if (tp->t_flags & TF_RCVD_TSTMP)
3487 			optlen = TCPOLEN_TSTAMP_APPA;
3488 		else
3489 			optlen = 0;
3490 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3491 		if (tp->t_flags & TF_SIGNATURE)
3492 			optlen += PAD(TCPOLEN_SIGNATURE);
3493 #endif
3494 	} else {
3495 		if (tp->t_flags & TF_REQ_TSTMP)
3496 			optlen = TCPOLEN_TSTAMP_APPA;
3497 		else
3498 			optlen = PAD(TCPOLEN_MAXSEG);
3499 		if (tp->t_flags & TF_REQ_SCALE)
3500 			optlen += PAD(TCPOLEN_WINDOW);
3501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3502 		if (tp->t_flags & TF_SIGNATURE)
3503 			optlen += PAD(TCPOLEN_SIGNATURE);
3504 #endif
3505 		if (tp->t_flags & TF_SACK_PERMIT)
3506 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3507 	}
3508 #undef PAD
3509 	optlen = min(optlen, TCP_MAXOLEN);
3510 	return (tp->t_maxseg - optlen);
3511 }
3512 
3513 
3514 
3515 static int
3516 sysctl_drop(SYSCTL_HANDLER_ARGS)
3517 {
3518 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3519 	struct sockaddr_storage addrs[2];
3520 	struct inpcb *inp;
3521 	struct tcpcb *tp;
3522 #ifdef INET
3523 	struct sockaddr_in *fin = NULL, *lin = NULL;
3524 #endif
3525 	struct epoch_tracker et;
3526 #ifdef INET6
3527 	struct sockaddr_in6 *fin6, *lin6;
3528 #endif
3529 	int error;
3530 
3531 	inp = NULL;
3532 #ifdef INET6
3533 	fin6 = lin6 = NULL;
3534 #endif
3535 	error = 0;
3536 
3537 	if (req->oldptr != NULL || req->oldlen != 0)
3538 		return (EINVAL);
3539 	if (req->newptr == NULL)
3540 		return (EPERM);
3541 	if (req->newlen < sizeof(addrs))
3542 		return (ENOMEM);
3543 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3544 	if (error)
3545 		return (error);
3546 
3547 	switch (addrs[0].ss_family) {
3548 #ifdef INET6
3549 	case AF_INET6:
3550 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3551 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3552 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3553 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3554 			return (EINVAL);
3555 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3556 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3557 				return (EINVAL);
3558 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3559 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3560 #ifdef INET
3561 			fin = (struct sockaddr_in *)&addrs[0];
3562 			lin = (struct sockaddr_in *)&addrs[1];
3563 #endif
3564 			break;
3565 		}
3566 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3567 		if (error)
3568 			return (error);
3569 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3570 		if (error)
3571 			return (error);
3572 		break;
3573 #endif
3574 #ifdef INET
3575 	case AF_INET:
3576 		fin = (struct sockaddr_in *)&addrs[0];
3577 		lin = (struct sockaddr_in *)&addrs[1];
3578 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3579 		    lin->sin_len != sizeof(struct sockaddr_in))
3580 			return (EINVAL);
3581 		break;
3582 #endif
3583 	default:
3584 		return (EINVAL);
3585 	}
3586 	NET_EPOCH_ENTER(et);
3587 	switch (addrs[0].ss_family) {
3588 #ifdef INET6
3589 	case AF_INET6:
3590 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3591 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3592 		    INPLOOKUP_WLOCKPCB, NULL);
3593 		break;
3594 #endif
3595 #ifdef INET
3596 	case AF_INET:
3597 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3598 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3599 		break;
3600 #endif
3601 	}
3602 	if (inp != NULL) {
3603 		if (!SOLISTENING(inp->inp_socket)) {
3604 			tp = intotcpcb(inp);
3605 			tp = tcp_drop(tp, ECONNABORTED);
3606 			if (tp != NULL)
3607 				INP_WUNLOCK(inp);
3608 		} else
3609 			INP_WUNLOCK(inp);
3610 	} else
3611 		error = ESRCH;
3612 	NET_EPOCH_EXIT(et);
3613 	return (error);
3614 }
3615 
3616 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3617     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3618     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3619     "Drop TCP connection");
3620 
3621 static int
3622 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3623 {
3624 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3625 	    &tcp_ctloutput_set));
3626 }
3627 
3628 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3629     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3630     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3631     "Set socket option for TCP endpoint");
3632 
3633 #ifdef KERN_TLS
3634 static int
3635 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3636 {
3637 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3638 	struct sockaddr_storage addrs[2];
3639 	struct inpcb *inp;
3640 #ifdef INET
3641 	struct sockaddr_in *fin = NULL, *lin = NULL;
3642 #endif
3643 	struct epoch_tracker et;
3644 #ifdef INET6
3645 	struct sockaddr_in6 *fin6, *lin6;
3646 #endif
3647 	int error;
3648 
3649 	inp = NULL;
3650 #ifdef INET6
3651 	fin6 = lin6 = NULL;
3652 #endif
3653 	error = 0;
3654 
3655 	if (req->oldptr != NULL || req->oldlen != 0)
3656 		return (EINVAL);
3657 	if (req->newptr == NULL)
3658 		return (EPERM);
3659 	if (req->newlen < sizeof(addrs))
3660 		return (ENOMEM);
3661 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3662 	if (error)
3663 		return (error);
3664 
3665 	switch (addrs[0].ss_family) {
3666 #ifdef INET6
3667 	case AF_INET6:
3668 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3669 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3670 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3671 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3672 			return (EINVAL);
3673 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3674 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3675 				return (EINVAL);
3676 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3677 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3678 #ifdef INET
3679 			fin = (struct sockaddr_in *)&addrs[0];
3680 			lin = (struct sockaddr_in *)&addrs[1];
3681 #endif
3682 			break;
3683 		}
3684 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3685 		if (error)
3686 			return (error);
3687 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3688 		if (error)
3689 			return (error);
3690 		break;
3691 #endif
3692 #ifdef INET
3693 	case AF_INET:
3694 		fin = (struct sockaddr_in *)&addrs[0];
3695 		lin = (struct sockaddr_in *)&addrs[1];
3696 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3697 		    lin->sin_len != sizeof(struct sockaddr_in))
3698 			return (EINVAL);
3699 		break;
3700 #endif
3701 	default:
3702 		return (EINVAL);
3703 	}
3704 	NET_EPOCH_ENTER(et);
3705 	switch (addrs[0].ss_family) {
3706 #ifdef INET6
3707 	case AF_INET6:
3708 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3709 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3710 		    INPLOOKUP_WLOCKPCB, NULL);
3711 		break;
3712 #endif
3713 #ifdef INET
3714 	case AF_INET:
3715 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3716 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3717 		break;
3718 #endif
3719 	}
3720 	NET_EPOCH_EXIT(et);
3721 	if (inp != NULL) {
3722 		struct socket *so;
3723 
3724 		so = inp->inp_socket;
3725 		soref(so);
3726 		error = ktls_set_tx_mode(so,
3727 		    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3728 		INP_WUNLOCK(inp);
3729 		sorele(so);
3730 	} else
3731 		error = ESRCH;
3732 	return (error);
3733 }
3734 
3735 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3736     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3737     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3738     "Switch TCP connection to SW TLS");
3739 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3740     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3741     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3742     "Switch TCP connection to ifnet TLS");
3743 #endif
3744 
3745 /*
3746  * Generate a standardized TCP log line for use throughout the
3747  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3748  * allow use in the interrupt context.
3749  *
3750  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3751  * NB: The function may return NULL if memory allocation failed.
3752  *
3753  * Due to header inclusion and ordering limitations the struct ip
3754  * and ip6_hdr pointers have to be passed as void pointers.
3755  */
3756 char *
3757 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3758     const void *ip6hdr)
3759 {
3760 
3761 	/* Is logging enabled? */
3762 	if (V_tcp_log_in_vain == 0)
3763 		return (NULL);
3764 
3765 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3766 }
3767 
3768 char *
3769 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3770     const void *ip6hdr)
3771 {
3772 
3773 	/* Is logging enabled? */
3774 	if (tcp_log_debug == 0)
3775 		return (NULL);
3776 
3777 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3778 }
3779 
3780 static char *
3781 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3782     const void *ip6hdr)
3783 {
3784 	char *s, *sp;
3785 	size_t size;
3786 #ifdef INET
3787 	const struct ip *ip = (const struct ip *)ip4hdr;
3788 #endif
3789 #ifdef INET6
3790 	const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3791 #endif /* INET6 */
3792 
3793 	/*
3794 	 * The log line looks like this:
3795 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3796 	 */
3797 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3798 	    sizeof(PRINT_TH_FLAGS) + 1 +
3799 #ifdef INET6
3800 	    2 * INET6_ADDRSTRLEN;
3801 #else
3802 	    2 * INET_ADDRSTRLEN;
3803 #endif /* INET6 */
3804 
3805 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3806 	if (s == NULL)
3807 		return (NULL);
3808 
3809 	strcat(s, "TCP: [");
3810 	sp = s + strlen(s);
3811 
3812 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3813 		inet_ntoa_r(inc->inc_faddr, sp);
3814 		sp = s + strlen(s);
3815 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3816 		sp = s + strlen(s);
3817 		inet_ntoa_r(inc->inc_laddr, sp);
3818 		sp = s + strlen(s);
3819 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3820 #ifdef INET6
3821 	} else if (inc) {
3822 		ip6_sprintf(sp, &inc->inc6_faddr);
3823 		sp = s + strlen(s);
3824 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3825 		sp = s + strlen(s);
3826 		ip6_sprintf(sp, &inc->inc6_laddr);
3827 		sp = s + strlen(s);
3828 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3829 	} else if (ip6 && th) {
3830 		ip6_sprintf(sp, &ip6->ip6_src);
3831 		sp = s + strlen(s);
3832 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3833 		sp = s + strlen(s);
3834 		ip6_sprintf(sp, &ip6->ip6_dst);
3835 		sp = s + strlen(s);
3836 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3837 #endif /* INET6 */
3838 #ifdef INET
3839 	} else if (ip && th) {
3840 		inet_ntoa_r(ip->ip_src, sp);
3841 		sp = s + strlen(s);
3842 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3843 		sp = s + strlen(s);
3844 		inet_ntoa_r(ip->ip_dst, sp);
3845 		sp = s + strlen(s);
3846 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3847 #endif /* INET */
3848 	} else {
3849 		free(s, M_TCPLOG);
3850 		return (NULL);
3851 	}
3852 	sp = s + strlen(s);
3853 	if (th)
3854 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3855 	if (*(s + size - 1) != '\0')
3856 		panic("%s: string too long", __func__);
3857 	return (s);
3858 }
3859 
3860 /*
3861  * A subroutine which makes it easy to track TCP state changes with DTrace.
3862  * This function shouldn't be called for t_state initializations that don't
3863  * correspond to actual TCP state transitions.
3864  */
3865 void
3866 tcp_state_change(struct tcpcb *tp, int newstate)
3867 {
3868 #if defined(KDTRACE_HOOKS)
3869 	int pstate = tp->t_state;
3870 #endif
3871 
3872 	TCPSTATES_DEC(tp->t_state);
3873 	TCPSTATES_INC(newstate);
3874 	tp->t_state = newstate;
3875 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3876 }
3877 
3878 /*
3879  * Create an external-format (``xtcpcb'') structure using the information in
3880  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3881  * reduce the spew of irrelevant information over this interface, to isolate
3882  * user code from changes in the kernel structure, and potentially to provide
3883  * information-hiding if we decide that some of this information should be
3884  * hidden from users.
3885  */
3886 void
3887 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3888 {
3889 	struct tcpcb *tp = intotcpcb(inp);
3890 	sbintime_t now;
3891 
3892 	bzero(xt, sizeof(*xt));
3893 	xt->t_state = tp->t_state;
3894 	xt->t_logstate = tcp_get_bblog_state(tp);
3895 	xt->t_flags = tp->t_flags;
3896 	xt->t_sndzerowin = tp->t_sndzerowin;
3897 	xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3898 	xt->t_rcvoopack = tp->t_rcvoopack;
3899 	xt->t_rcv_wnd = tp->rcv_wnd;
3900 	xt->t_snd_wnd = tp->snd_wnd;
3901 	xt->t_snd_cwnd = tp->snd_cwnd;
3902 	xt->t_snd_ssthresh = tp->snd_ssthresh;
3903 	xt->t_dsack_bytes = tp->t_dsack_bytes;
3904 	xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
3905 	xt->t_dsack_pack = tp->t_dsack_pack;
3906 	xt->t_maxseg = tp->t_maxseg;
3907 	xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
3908 		     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
3909 
3910 	now = getsbinuptime();
3911 #define	COPYTIMER(which,where)	do {					\
3912 	if (tp->t_timers[which] != SBT_MAX)				\
3913 		xt->where = (tp->t_timers[which] - now) / SBT_1MS;	\
3914 	else								\
3915 		xt->where = 0;						\
3916 } while (0)
3917 	COPYTIMER(TT_DELACK, tt_delack);
3918 	COPYTIMER(TT_REXMT, tt_rexmt);
3919 	COPYTIMER(TT_PERSIST, tt_persist);
3920 	COPYTIMER(TT_KEEP, tt_keep);
3921 	COPYTIMER(TT_2MSL, tt_2msl);
3922 #undef COPYTIMER
3923 	xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3924 
3925 	xt->xt_encaps_port = tp->t_port;
3926 	bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3927 	    TCP_FUNCTION_NAME_LEN_MAX);
3928 	bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX);
3929 #ifdef TCP_BLACKBOX
3930 	(void)tcp_log_get_id(tp, xt->xt_logid);
3931 #endif
3932 
3933 	xt->xt_len = sizeof(struct xtcpcb);
3934 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3935 }
3936 
3937 void
3938 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
3939 {
3940 	uint32_t bit, i;
3941 
3942 	if ((tp == NULL) ||
3943 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
3944 	    (status == 0)) {
3945 		/* Invalid */
3946 		return;
3947 	}
3948 	if (status > (sizeof(uint32_t) * 8)) {
3949 		/* Should this be a KASSERT? */
3950 		return;
3951 	}
3952 	bit = 1U << (status - 1);
3953 	if (bit & tp->t_end_info_status) {
3954 		/* already logged */
3955 		return;
3956 	}
3957 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
3958 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
3959 			tp->t_end_info_bytes[i] = status;
3960 			tp->t_end_info_status |= bit;
3961 			break;
3962 		}
3963 	}
3964 }
3965 
3966 int
3967 tcp_can_enable_pacing(void)
3968 {
3969 
3970 	if ((tcp_pacing_limit == -1) ||
3971 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
3972 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
3973 		shadow_num_connections = number_of_tcp_connections_pacing;
3974 		return (1);
3975 	} else {
3976 		return (0);
3977 	}
3978 }
3979 
3980 static uint8_t tcp_pacing_warning = 0;
3981 
3982 void
3983 tcp_decrement_paced_conn(void)
3984 {
3985 	uint32_t ret;
3986 
3987 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
3988 	shadow_num_connections = number_of_tcp_connections_pacing;
3989 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
3990 	if (ret == 0) {
3991 		if (tcp_pacing_limit != -1) {
3992 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
3993 			tcp_pacing_limit = 0;
3994 		} else if (tcp_pacing_warning == 0) {
3995 			printf("Warning pacing count is invalid, invalid decrement\n");
3996 			tcp_pacing_warning = 1;
3997 		}
3998 	}
3999 }
4000 
4001 #ifdef TCP_ACCOUNTING
4002 int
4003 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss)
4004 {
4005 	if (SEQ_LT(th->th_ack, tp->snd_una)) {
4006 		/* Do we have a SACK? */
4007 		if (to->to_flags & TOF_SACK) {
4008 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4009 				tp->tcp_cnt_counters[ACK_SACK]++;
4010 			}
4011 			return (ACK_SACK);
4012 		} else {
4013 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4014 				tp->tcp_cnt_counters[ACK_BEHIND]++;
4015 			}
4016 			return (ACK_BEHIND);
4017 		}
4018 	} else if (th->th_ack == tp->snd_una) {
4019 		/* Do we have a SACK? */
4020 		if (to->to_flags & TOF_SACK) {
4021 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4022 				tp->tcp_cnt_counters[ACK_SACK]++;
4023 			}
4024 			return (ACK_SACK);
4025 		} else if (tiwin != tp->snd_wnd) {
4026 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4027 				tp->tcp_cnt_counters[ACK_RWND]++;
4028 			}
4029 			return (ACK_RWND);
4030 		} else {
4031 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4032 				tp->tcp_cnt_counters[ACK_DUPACK]++;
4033 			}
4034 			return (ACK_DUPACK);
4035 		}
4036 	} else {
4037 		if (!SEQ_GT(th->th_ack, tp->snd_max)) {
4038 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4039 				tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss);
4040 			}
4041 		}
4042 		if (to->to_flags & TOF_SACK) {
4043 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4044 				tp->tcp_cnt_counters[ACK_CUMACK_SACK]++;
4045 			}
4046 			return (ACK_CUMACK_SACK);
4047 		} else {
4048 			if (tp->t_flags2 & TF2_TCP_ACCOUNTING) {
4049 				tp->tcp_cnt_counters[ACK_CUMACK]++;
4050 			}
4051 			return (ACK_CUMACK);
4052 		}
4053 	}
4054 }
4055 #endif
4056 
4057 void
4058 tcp_change_time_units(struct tcpcb *tp, int granularity)
4059 {
4060 	if (tp->t_tmr_granularity == granularity) {
4061 		/* We are there */
4062 		return;
4063 	}
4064 	if (granularity == TCP_TMR_GRANULARITY_USEC) {
4065 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS),
4066 			("Granularity is not TICKS its %u in tp:%p",
4067 			 tp->t_tmr_granularity, tp));
4068 		tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow);
4069 		if (tp->t_srtt > 1) {
4070 			uint32_t val, frac;
4071 
4072 			val = tp->t_srtt >> TCP_RTT_SHIFT;
4073 			frac = tp->t_srtt & 0x1f;
4074 			tp->t_srtt = TICKS_2_USEC(val);
4075 			/*
4076 			 * frac is the fractional part of the srtt (if any)
4077 			 * but its in ticks and every bit represents
4078 			 * 1/32nd of a hz.
4079 			 */
4080 			if (frac) {
4081 				if (hz == 1000) {
4082 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4083 				} else {
4084 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4085 				}
4086 				tp->t_srtt += frac;
4087 			}
4088 		}
4089 		if (tp->t_rttvar) {
4090 			uint32_t val, frac;
4091 
4092 			val = tp->t_rttvar >> TCP_RTTVAR_SHIFT;
4093 			frac = tp->t_rttvar & 0x1f;
4094 			tp->t_rttvar = TICKS_2_USEC(val);
4095 			/*
4096 			 * frac is the fractional part of the srtt (if any)
4097 			 * but its in ticks and every bit represents
4098 			 * 1/32nd of a hz.
4099 			 */
4100 			if (frac) {
4101 				if (hz == 1000) {
4102 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE);
4103 				} else {
4104 					frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE));
4105 				}
4106 				tp->t_rttvar += frac;
4107 			}
4108 		}
4109 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC;
4110 	} else if (granularity == TCP_TMR_GRANULARITY_TICKS) {
4111 		/* Convert back to ticks, with  */
4112 		KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC),
4113 			("Granularity is not USEC its %u in tp:%p",
4114 			 tp->t_tmr_granularity, tp));
4115 		if (tp->t_srtt > 1) {
4116 			uint32_t val, frac;
4117 
4118 			val = USEC_2_TICKS(tp->t_srtt);
4119 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4120 			tp->t_srtt = val << TCP_RTT_SHIFT;
4121 			/*
4122 			 * frac is the fractional part here is left
4123 			 * over from converting to hz and shifting.
4124 			 * We need to convert this to the 5 bit
4125 			 * remainder.
4126 			 */
4127 			if (frac) {
4128 				if (hz == 1000) {
4129 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4130 				} else {
4131 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4132 				}
4133 				tp->t_srtt += frac;
4134 			}
4135 		}
4136 		if (tp->t_rttvar) {
4137 			uint32_t val, frac;
4138 
4139 			val = USEC_2_TICKS(tp->t_rttvar);
4140 			frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz);
4141 			tp->t_rttvar = val <<  TCP_RTTVAR_SHIFT;
4142 			/*
4143 			 * frac is the fractional part here is left
4144 			 * over from converting to hz and shifting.
4145 			 * We need to convert this to the 5 bit
4146 			 * remainder.
4147 			 */
4148 			if (frac) {
4149 				if (hz == 1000) {
4150 					frac = (((uint64_t)frac *  (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC);
4151 				} else {
4152 					frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC);
4153 				}
4154 				tp->t_rttvar += frac;
4155 			}
4156 		}
4157 		tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow);
4158 		tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS;
4159 	}
4160 #ifdef INVARIANTS
4161 	else {
4162 		panic("Unknown granularity:%d tp:%p",
4163 		      granularity, tp);
4164 	}
4165 #endif
4166 }
4167 
4168 void
4169 tcp_handle_orphaned_packets(struct tcpcb *tp)
4170 {
4171 	struct mbuf *save, *m, *prev;
4172 	/*
4173 	 * Called when a stack switch is occuring from the fini()
4174 	 * of the old stack. We assue the init() as already been
4175 	 * run of the new stack and it has set the inp_flags2 to
4176 	 * what it supports. This function will then deal with any
4177 	 * differences i.e. cleanup packets that maybe queued that
4178 	 * the newstack does not support.
4179 	 */
4180 
4181 	if (tptoinpcb(tp)->inp_flags2 & INP_MBUF_L_ACKS)
4182 		return;
4183 	if ((tptoinpcb(tp)->inp_flags2 & INP_SUPPORTS_MBUFQ) == 0) {
4184 		/*
4185 		 * It is unsafe to process the packets since a
4186 		 * reset may be lurking in them (its rare but it
4187 		 * can occur). If we were to find a RST, then we
4188 		 * would end up dropping the connection and the
4189 		 * INP lock, so when we return the caller (tcp_usrreq)
4190 		 * will blow up when it trys to unlock the inp.
4191 		 * This new stack does not do any fancy LRO features
4192 		 * so all we can do is toss the packets.
4193 		 */
4194 		m = tp->t_in_pkt;
4195 		tp->t_in_pkt = NULL;
4196 		tp->t_tail_pkt = NULL;
4197 		while (m) {
4198 			save = m->m_nextpkt;
4199 			m->m_nextpkt = NULL;
4200 			m_freem(m);
4201 			m = save;
4202 		}
4203 	} else {
4204 		/*
4205 		 * Here we have a stack that does mbuf queuing but
4206 		 * does not support compressed ack's. We must
4207 		 * walk all the mbufs and discard any compressed acks.
4208 		 */
4209 		m = tp->t_in_pkt;
4210 		prev = NULL;
4211 		while (m) {
4212 			if (m->m_flags & M_ACKCMP) {
4213 				/* We must toss this packet */
4214 				if (tp->t_tail_pkt == m)
4215 					tp->t_tail_pkt = prev;
4216 				if (prev)
4217 					prev->m_nextpkt = m->m_nextpkt;
4218 				else
4219 					tp->t_in_pkt =  m->m_nextpkt;
4220 				m->m_nextpkt = NULL;
4221 				m_freem(m);
4222 				/* move forward */
4223 				if (prev)
4224 					m = prev->m_nextpkt;
4225 				else
4226 					m = tp->t_in_pkt;
4227 			} else {
4228 				/* this one is ok */
4229 				prev = m;
4230 				m = m->m_nextpkt;
4231 			}
4232 		}
4233 	}
4234 }
4235 
4236 #ifdef TCP_REQUEST_TRK
4237 uint32_t
4238 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes)
4239 {
4240 #ifdef KERN_TLS
4241 	struct ktls_session *tls;
4242 	uint32_t rec_oh, records;
4243 
4244 	tls = so->so_snd.sb_tls_info;
4245 	if (tls == NULL)
4246 	    return (0);
4247 
4248 	rec_oh = tls->params.tls_hlen + tls->params.tls_tlen;
4249 	records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len);
4250 	return (records * rec_oh);
4251 #else
4252 	return (0);
4253 #endif
4254 }
4255 
4256 extern uint32_t tcp_stale_entry_time;
4257 uint32_t tcp_stale_entry_time = 250000;
4258 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW,
4259     &tcp_stale_entry_time, 250000, "Time that a http entry without a sendfile ages out");
4260 
4261 void
4262 tcp_http_log_req_info(struct tcpcb *tp, struct http_sendfile_track *http,
4263     uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes)
4264 {
4265 	if (tcp_bblogging_on(tp)) {
4266 		union tcp_log_stackspecific log;
4267 		struct timeval tv;
4268 
4269 		memset(&log.u_bbr, 0, sizeof(log.u_bbr));
4270 #ifdef TCPHPTS
4271 		log.u_bbr.inhpts = tcp_in_hpts(tptoinpcb(tp));
4272 #endif
4273 		log.u_bbr.flex8 = val;
4274 		log.u_bbr.rttProp = http->timestamp;
4275 		log.u_bbr.delRate = http->start;
4276 		log.u_bbr.cur_del_rate = http->end;
4277 		log.u_bbr.flex1 = http->start_seq;
4278 		log.u_bbr.flex2 = http->end_seq;
4279 		log.u_bbr.flex3 = http->flags;
4280 		log.u_bbr.flex4 = ((http->localtime >> 32) & 0x00000000ffffffff);
4281 		log.u_bbr.flex5 = (http->localtime & 0x00000000ffffffff);
4282 		log.u_bbr.flex7 = slot;
4283 		log.u_bbr.bw_inuse = offset;
4284 		/* nbytes = flex6 | epoch */
4285 		log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff);
4286 		log.u_bbr.epoch = (nbytes & 0x00000000ffffffff);
4287 		/* cspr =  lt_epoch | pkts_out */
4288 		log.u_bbr.lt_epoch = ((http->cspr >> 32) & 0x00000000ffffffff);
4289 		log.u_bbr.pkts_out |= (http->cspr & 0x00000000ffffffff);
4290 		log.u_bbr.applimited = tp->t_http_closed;
4291 		log.u_bbr.applimited <<= 8;
4292 		log.u_bbr.applimited |= tp->t_http_open;
4293 		log.u_bbr.applimited <<= 8;
4294 		log.u_bbr.applimited |= tp->t_http_req;
4295 		log.u_bbr.timeStamp = tcp_get_usecs(&tv);
4296 		TCP_LOG_EVENTP(tp, NULL,
4297 		    &tptosocket(tp)->so_rcv,
4298 		    &tptosocket(tp)->so_snd,
4299 		    TCP_LOG_HTTP_T, 0,
4300 		    0, &log, false, &tv);
4301 	}
4302 }
4303 
4304 void
4305 tcp_http_free_a_slot(struct tcpcb *tp, struct http_sendfile_track *ent)
4306 {
4307 	if (tp->t_http_req > 0)
4308 		tp->t_http_req--;
4309 	if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN) {
4310 		if (tp->t_http_open > 0)
4311 			tp->t_http_open--;
4312 	} else {
4313 		if (tp->t_http_closed > 0)
4314 			tp->t_http_closed--;
4315 	}
4316 	ent->flags = TCP_HTTP_TRACK_FLG_EMPTY;
4317 }
4318 
4319 static void
4320 tcp_http_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest)
4321 {
4322 	struct http_sendfile_track *ent;
4323 	uint64_t time_delta, oldest_delta;
4324 	int i, oldest, oldest_set = 0, cnt_rm = 0;
4325 
4326 	for(i = 0; i < MAX_TCP_HTTP_REQ; i++) {
4327 		ent = &tp->t_http_info[i];
4328 		if (ent->flags != TCP_HTTP_TRACK_FLG_USED) {
4329 			/*
4330 			 * We only care about closed end ranges
4331 			 * that are allocated and have no sendfile
4332 			 * ever touching them. They would be in
4333 			 * state USED.
4334 			 */
4335 			continue;
4336 		}
4337 		if (ts >= ent->localtime)
4338 			time_delta = ts - ent->localtime;
4339 		else
4340 			time_delta = 0;
4341 		if (time_delta &&
4342 		    ((oldest_delta < time_delta) || (oldest_set == 0))) {
4343 			oldest_set = 1;
4344 			oldest = i;
4345 			oldest_delta = time_delta;
4346 		}
4347 		if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) {
4348 			/*
4349 			 * No sendfile in a our time-limit
4350 			 * time to purge it.
4351 			 */
4352 			cnt_rm++;
4353 			tcp_http_log_req_info(tp, &tp->t_http_info[i], i, TCP_HTTP_REQ_LOG_STALE,
4354 					      time_delta, 0);
4355 			tcp_http_free_a_slot(tp, ent);
4356 		}
4357 	}
4358 	if ((cnt_rm == 0) && rm_oldest && oldest_set) {
4359 		ent = &tp->t_http_info[oldest];
4360 		tcp_http_log_req_info(tp, &tp->t_http_info[i], i, TCP_HTTP_REQ_LOG_STALE,
4361 				      oldest_delta, 1);
4362 		tcp_http_free_a_slot(tp, ent);
4363 	}
4364 }
4365 
4366 int
4367 tcp_http_check_for_comp(struct tcpcb *tp, tcp_seq ack_point)
4368 {
4369 	int i, ret=0;
4370 	struct http_sendfile_track *ent;
4371 
4372 	/* Clean up any old closed end requests that are now completed */
4373 	if (tp->t_http_req == 0)
4374 		return(0);
4375 	if (tp->t_http_closed == 0)
4376 		return(0);
4377 	for(i = 0; i < MAX_TCP_HTTP_REQ; i++) {
4378 		ent = &tp->t_http_info[i];
4379 		/* Skip empty ones */
4380 		if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY)
4381 			continue;
4382 		/* Skip open ones */
4383 		if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN)
4384 			continue;
4385 		if (SEQ_GEQ(ack_point, ent->end_seq)) {
4386 			/* We are past it -- free it */
4387 			tcp_http_log_req_info(tp, ent,
4388 					      i, TCP_HTTP_REQ_LOG_FREED, 0, 0);
4389 			tcp_http_free_a_slot(tp, ent);
4390 			ret++;
4391 		}
4392 	}
4393 	return (ret);
4394 }
4395 
4396 int
4397 tcp_http_is_entry_comp(struct tcpcb *tp, struct http_sendfile_track *ent, tcp_seq ack_point)
4398 {
4399 	if (tp->t_http_req == 0)
4400 		return(-1);
4401 	if (tp->t_http_closed == 0)
4402 		return(-1);
4403 	if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY)
4404 		return(-1);
4405 	if (SEQ_GEQ(ack_point, ent->end_seq)) {
4406 		return (1);
4407 	}
4408 	return (0);
4409 }
4410 
4411 struct http_sendfile_track *
4412 tcp_http_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip)
4413 {
4414 	/*
4415 	 * Given an ack point (th_ack) walk through our entries and
4416 	 * return the first one found that th_ack goes past the
4417 	 * end_seq.
4418 	 */
4419 	struct http_sendfile_track *ent;
4420 	int i;
4421 
4422 	if (tp->t_http_req == 0) {
4423 		/* none open */
4424 		return (NULL);
4425 	}
4426 	for(i = 0; i < MAX_TCP_HTTP_REQ; i++) {
4427 		ent = &tp->t_http_info[i];
4428 		if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY)
4429 			continue;
4430 		if ((ent->flags & TCP_HTTP_TRACK_FLG_OPEN) == 0) {
4431 			if (SEQ_GEQ(th_ack, ent->end_seq)) {
4432 				*ip = i;
4433 				return (ent);
4434 			}
4435 		}
4436 	}
4437 	return (NULL);
4438 }
4439 
4440 struct http_sendfile_track *
4441 tcp_http_find_req_for_seq(struct tcpcb *tp, tcp_seq seq)
4442 {
4443 	struct http_sendfile_track *ent;
4444 	int i;
4445 
4446 	if (tp->t_http_req == 0) {
4447 		/* none open */
4448 		return (NULL);
4449 	}
4450 	for(i = 0; i < MAX_TCP_HTTP_REQ; i++) {
4451 		ent = &tp->t_http_info[i];
4452 		tcp_http_log_req_info(tp, ent, i, TCP_HTTP_REQ_LOG_SEARCH,
4453 				      (uint64_t)seq, 0);
4454 		if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY) {
4455 			continue;
4456 		}
4457 		if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN) {
4458 			/*
4459 			 * An open end request only needs to
4460 			 * match the beginning seq or be
4461 			 * all we have (once we keep going on
4462 			 * a open end request we may have a seq
4463 			 * wrap).
4464 			 */
4465 			if ((SEQ_GEQ(seq, ent->start_seq)) ||
4466 			    (tp->t_http_closed == 0))
4467 				return (ent);
4468 		} else {
4469 			/*
4470 			 * For this one we need to
4471 			 * be a bit more careful if its
4472 			 * completed at least.
4473 			 */
4474 			if ((SEQ_GEQ(seq, ent->start_seq)) &&
4475 			    (SEQ_LT(seq, ent->end_seq))) {
4476 				return (ent);
4477 			}
4478 		}
4479 	}
4480 	return (NULL);
4481 }
4482 
4483 /* Should this be in its own file tcp_http.c ? */
4484 struct http_sendfile_track *
4485 tcp_http_alloc_req_full(struct tcpcb *tp, struct http_req *req, uint64_t ts, int rec_dups)
4486 {
4487 	struct http_sendfile_track *fil;
4488 	int i, allocated;
4489 
4490 	/* In case the stack does not check for completions do so now */
4491 	tcp_http_check_for_comp(tp, tp->snd_una);
4492 	/* Check for stale entries */
4493 	if (tp->t_http_req)
4494 		tcp_http_check_for_stale_entries(tp, ts,
4495 		    (tp->t_http_req >= MAX_TCP_HTTP_REQ));
4496 	/* Check to see if this is a duplicate of one not started */
4497 	if (tp->t_http_req) {
4498 		for(i = 0, allocated = 0; i < MAX_TCP_HTTP_REQ; i++) {
4499 			fil = &tp->t_http_info[i];
4500 			if (fil->flags != TCP_HTTP_TRACK_FLG_USED)
4501 				continue;
4502 			if ((fil->timestamp == req->timestamp) &&
4503 			    (fil->start == req->start) &&
4504 			    ((fil->flags & TCP_HTTP_TRACK_FLG_OPEN) ||
4505 			     (fil->end == req->end))) {
4506 				/*
4507 				 * We already have this request
4508 				 * and it has not been started with sendfile.
4509 				 * This probably means the user was returned
4510 				 * a 4xx of some sort and its going to age
4511 				 * out, lets not duplicate it.
4512 				 */
4513 				return(fil);
4514 			}
4515 		}
4516 	}
4517 	/* Ok if there is no room at the inn we are in trouble */
4518 	if (tp->t_http_req >= MAX_TCP_HTTP_REQ) {
4519 		tcp_trace_point(tp, TCP_TP_HTTP_LOG_FAIL);
4520 		for(i = 0; i < MAX_TCP_HTTP_REQ; i++) {
4521 			tcp_http_log_req_info(tp, &tp->t_http_info[i],
4522 			    i, TCP_HTTP_REQ_LOG_ALLOCFAIL, 0, 0);
4523 		}
4524 		return (NULL);
4525 	}
4526 	for(i = 0, allocated = 0; i < MAX_TCP_HTTP_REQ; i++) {
4527 		fil = &tp->t_http_info[i];
4528 		if (fil->flags == TCP_HTTP_TRACK_FLG_EMPTY) {
4529 			allocated = 1;
4530 			fil->flags = TCP_HTTP_TRACK_FLG_USED;
4531 			fil->timestamp = req->timestamp;
4532 			fil->localtime = ts;
4533 			fil->start = req->start;
4534 			if (req->flags & TCP_LOG_HTTPD_RANGE_END) {
4535 				fil->end = req->end;
4536 			} else {
4537 				fil->end = 0;
4538 				fil->flags |= TCP_HTTP_TRACK_FLG_OPEN;
4539 			}
4540 			/*
4541 			 * We can set the min boundaries to the TCP Sequence space,
4542 			 * but it might be found to be further up when sendfile
4543 			 * actually runs on this range (if it ever does).
4544 			 */
4545 			fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc;
4546 			fil->start_seq = tp->snd_una +
4547 			    tptosocket(tp)->so_snd.sb_ccc;
4548 			fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start)));
4549 			if (tptosocket(tp)->so_snd.sb_tls_info) {
4550 				/*
4551 				 * This session is doing TLS. Take a swag guess
4552 				 * at the overhead.
4553 				 */
4554 				fil->end_seq += tcp_estimate_tls_overhead(
4555 				    tptosocket(tp), (fil->end - fil->start));
4556 			}
4557 			tp->t_http_req++;
4558 			if (fil->flags & TCP_HTTP_TRACK_FLG_OPEN)
4559 				tp->t_http_open++;
4560 			else
4561 				tp->t_http_closed++;
4562 			tcp_http_log_req_info(tp, fil, i,
4563 			    TCP_HTTP_REQ_LOG_NEW, 0, 0);
4564 			break;
4565 		} else
4566 			fil = NULL;
4567 	}
4568 	return (fil);
4569 }
4570 
4571 void
4572 tcp_http_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts)
4573 {
4574 	(void)tcp_http_alloc_req_full(tp, &user->http_req, ts, 1);
4575 }
4576 #endif
4577 
4578 void
4579 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err)
4580 {
4581 	if (tcp_bblogging_on(tp)) {
4582 		struct tcp_log_buffer *l;
4583 
4584 		l = tcp_log_event(tp, NULL,
4585 		        &tptosocket(tp)->so_rcv,
4586 		        &tptosocket(tp)->so_snd,
4587 		        TCP_LOG_SOCKET_OPT,
4588 		        err, 0, NULL, 1,
4589 		        NULL, NULL, 0, NULL);
4590 		if (l) {
4591 			l->tlb_flex1 = option_num;
4592 			l->tlb_flex2 = option_val;
4593 		}
4594 	}
4595 }
4596