xref: /freebsd/sys/netinet/tcp_subr.c (revision 994297b01b98816bea1abf45ae4bac1bc69ee7a0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/route/nhop.h>
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/vnet.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_fib.h>
86 #include <netinet/in_kdtrace.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_systm.h>
89 #include <netinet/in_var.h>
90 #include <netinet/ip.h>
91 #include <netinet/ip_icmp.h>
92 #include <netinet/ip_var.h>
93 #ifdef INET6
94 #include <netinet/icmp6.h>
95 #include <netinet/ip6.h>
96 #include <netinet6/in6_fib.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/ip6_var.h>
99 #include <netinet6/scope6_var.h>
100 #include <netinet6/nd6.h>
101 #endif
102 
103 #include <netinet/tcp.h>
104 #ifdef INVARIANTS
105 #define TCPSTATES
106 #endif
107 #include <netinet/tcp_fsm.h>
108 #include <netinet/tcp_seq.h>
109 #include <netinet/tcp_timer.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet/tcp_log_buf.h>
112 #include <netinet/tcp_syncache.h>
113 #include <netinet/tcp_hpts.h>
114 #include <netinet/cc/cc.h>
115 #ifdef INET6
116 #include <netinet6/tcp6_var.h>
117 #endif
118 #include <netinet/tcpip.h>
119 #include <netinet/tcp_fastopen.h>
120 #ifdef TCPPCAP
121 #include <netinet/tcp_pcap.h>
122 #endif
123 #ifdef TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126 #ifdef INET6
127 #include <netinet6/ip6protosw.h>
128 #endif
129 #ifdef TCP_OFFLOAD
130 #include <netinet/tcp_offload.h>
131 #endif
132 #include <netinet/udp.h>
133 #include <netinet/udp_var.h>
134 
135 #include <netipsec/ipsec_support.h>
136 
137 #include <machine/in_cksum.h>
138 #include <crypto/siphash/siphash.h>
139 
140 #include <security/mac/mac_framework.h>
141 
142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
143 #ifdef INET6
144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
145 #endif
146 
147 #ifdef NETFLIX_EXP_DETECTION
148 /*  Sack attack detection thresholds and such */
149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
150     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
151     "Sack Attack detection thresholds");
152 int32_t tcp_force_detection = 0;
153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
154     CTLFLAG_RW,
155     &tcp_force_detection, 0,
156     "Do we force detection even if the INP has it off?");
157 int32_t tcp_sack_to_ack_thresh = 700;	/* 70 % */
158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
159     CTLFLAG_RW,
160     &tcp_sack_to_ack_thresh, 700,
161     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
162 int32_t tcp_sack_to_move_thresh = 600;	/* 60 % */
163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
164     CTLFLAG_RW,
165     &tcp_sack_to_move_thresh, 600,
166     "Percentage of sack moves we must see above (10.1 percent is 101)");
167 int32_t tcp_restoral_thresh = 650;	/* 65 % (sack:2:ack -5%) */
168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
169     CTLFLAG_RW,
170     &tcp_restoral_thresh, 550,
171     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
172 int32_t tcp_sad_decay_val = 800;
173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
174     CTLFLAG_RW,
175     &tcp_sad_decay_val, 800,
176     "The decay percentage (10.1 percent equals 101 )");
177 int32_t tcp_map_minimum = 500;
178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
179     CTLFLAG_RW,
180     &tcp_map_minimum, 500,
181     "Number of Map enteries before we start detection");
182 int32_t tcp_attack_on_turns_on_logging = 0;
183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
184     CTLFLAG_RW,
185     &tcp_attack_on_turns_on_logging, 0,
186    "When we have a positive hit on attack, do we turn on logging?");
187 int32_t tcp_sad_pacing_interval = 2000;
188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
189     CTLFLAG_RW,
190     &tcp_sad_pacing_interval, 2000,
191     "What is the minimum pacing interval for a classified attacker?");
192 
193 int32_t tcp_sad_low_pps = 100;
194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
195     CTLFLAG_RW,
196     &tcp_sad_low_pps, 100,
197     "What is the input pps that below which we do not decay?");
198 #endif
199 uint32_t tcp_ack_war_time_window = 1000;
200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
201     CTLFLAG_RW,
202     &tcp_ack_war_time_window, 1000,
203    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
204 uint32_t tcp_ack_war_cnt = 5;
205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
206     CTLFLAG_RW,
207     &tcp_ack_war_cnt, 5,
208    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
209 
210 struct rwlock tcp_function_lock;
211 
212 static int
213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
214 {
215 	int error, new;
216 
217 	new = V_tcp_mssdflt;
218 	error = sysctl_handle_int(oidp, &new, 0, req);
219 	if (error == 0 && req->newptr) {
220 		if (new < TCP_MINMSS)
221 			error = EINVAL;
222 		else
223 			V_tcp_mssdflt = new;
224 	}
225 	return (error);
226 }
227 
228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
229     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
230     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
231     "Default TCP Maximum Segment Size");
232 
233 #ifdef INET6
234 static int
235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
236 {
237 	int error, new;
238 
239 	new = V_tcp_v6mssdflt;
240 	error = sysctl_handle_int(oidp, &new, 0, req);
241 	if (error == 0 && req->newptr) {
242 		if (new < TCP_MINMSS)
243 			error = EINVAL;
244 		else
245 			V_tcp_v6mssdflt = new;
246 	}
247 	return (error);
248 }
249 
250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
251     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
252     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
253    "Default TCP Maximum Segment Size for IPv6");
254 #endif /* INET6 */
255 
256 /*
257  * Minimum MSS we accept and use. This prevents DoS attacks where
258  * we are forced to a ridiculous low MSS like 20 and send hundreds
259  * of packets instead of one. The effect scales with the available
260  * bandwidth and quickly saturates the CPU and network interface
261  * with packet generation and sending. Set to zero to disable MINMSS
262  * checking. This setting prevents us from sending too small packets.
263  */
264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
266      &VNET_NAME(tcp_minmss), 0,
267     "Minimum TCP Maximum Segment Size");
268 
269 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
271     &VNET_NAME(tcp_do_rfc1323), 0,
272     "Enable rfc1323 (high performance TCP) extensions");
273 
274 /*
275  * As of June 2021, several TCP stacks violate RFC 7323 from September 2014.
276  * Some stacks negotiate TS, but never send them after connection setup. Some
277  * stacks negotiate TS, but don't send them when sending keep-alive segments.
278  * These include modern widely deployed TCP stacks.
279  * Therefore tolerating violations for now...
280  */
281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1;
282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
283     &VNET_NAME(tcp_tolerate_missing_ts), 0,
284     "Tolerate missing TCP timestamps");
285 
286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
288     &VNET_NAME(tcp_ts_offset_per_conn), 0,
289     "Initialize TCP timestamps per connection instead of per host pair");
290 
291 /* How many connections are pacing */
292 static volatile uint32_t number_of_tcp_connections_pacing = 0;
293 static uint32_t shadow_num_connections = 0;
294 
295 static int tcp_pacing_limit = 10000;
296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
297     &tcp_pacing_limit, 1000,
298     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
299 
300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
301     &shadow_num_connections, 0, "Number of TCP connections being paced");
302 
303 static int	tcp_log_debug = 0;
304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
305     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
306 
307 static int	tcp_tcbhashsize;
308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
309     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
310 
311 static int	do_tcpdrain = 1;
312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
313     "Enable tcp_drain routine for extra help when low on mbufs");
314 
315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
316     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
317 
318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
319 #define	V_icmp_may_rst			VNET(icmp_may_rst)
320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
321     &VNET_NAME(icmp_may_rst), 0,
322     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
323 
324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
325 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
327     &VNET_NAME(tcp_isn_reseed_interval), 0,
328     "Seconds between reseeding of ISN secret");
329 
330 static int	tcp_soreceive_stream;
331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
332     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
333 
334 VNET_DEFINE(uma_zone_t, sack_hole_zone);
335 #define	V_sack_hole_zone		VNET(sack_hole_zone)
336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;	/* unlimited */
337 static int
338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
339 {
340 	int error;
341 	uint32_t new;
342 
343 	new = V_tcp_map_entries_limit;
344 	error = sysctl_handle_int(oidp, &new, 0, req);
345 	if (error == 0 && req->newptr) {
346 		/* only allow "0" and value > minimum */
347 		if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
348 			error = EINVAL;
349 		else
350 			V_tcp_map_entries_limit = new;
351 	}
352 	return (error);
353 }
354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
355     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
356     &VNET_NAME(tcp_map_entries_limit), 0,
357     &sysctl_net_inet_tcp_map_limit_check, "IU",
358     "Total sendmap entries limit");
359 
360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0;	/* unlimited */
361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
362      &VNET_NAME(tcp_map_split_limit), 0,
363     "Total sendmap split entries limit");
364 
365 #ifdef TCP_HHOOK
366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
367 #endif
368 
369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
371 #define	V_ts_offset_secret	VNET(ts_offset_secret)
372 
373 static int	tcp_default_fb_init(struct tcpcb *tp);
374 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
375 static int	tcp_default_handoff_ok(struct tcpcb *tp);
376 static struct inpcb *tcp_notify(struct inpcb *, int);
377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
378 static struct inpcb *tcp_mtudisc(struct inpcb *, int);
379 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
380 		    void *ip4hdr, const void *ip6hdr);
381 
382 static struct tcp_function_block tcp_def_funcblk = {
383 	.tfb_tcp_block_name = "freebsd",
384 	.tfb_tcp_output = tcp_default_output,
385 	.tfb_tcp_do_segment = tcp_do_segment,
386 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
387 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
388 	.tfb_tcp_fb_init = tcp_default_fb_init,
389 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
390 };
391 
392 static int tcp_fb_cnt = 0;
393 struct tcp_funchead t_functions;
394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
395 
396 void
397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp)
398 {
399 	TCPSTAT_INC(tcps_dsack_count);
400 	tp->t_dsack_pack++;
401 	if (tlp == 0) {
402 		if (SEQ_GT(end, start)) {
403 			tp->t_dsack_bytes += (end - start);
404 			TCPSTAT_ADD(tcps_dsack_bytes, (end - start));
405 		} else {
406 			tp->t_dsack_tlp_bytes += (start - end);
407 			TCPSTAT_ADD(tcps_dsack_bytes, (start - end));
408 		}
409 	} else {
410 		if (SEQ_GT(end, start)) {
411 			tp->t_dsack_bytes += (end - start);
412 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start));
413 		} else {
414 			tp->t_dsack_tlp_bytes += (start - end);
415 			TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end));
416 		}
417 	}
418 }
419 
420 static struct tcp_function_block *
421 find_tcp_functions_locked(struct tcp_function_set *fs)
422 {
423 	struct tcp_function *f;
424 	struct tcp_function_block *blk=NULL;
425 
426 	TAILQ_FOREACH(f, &t_functions, tf_next) {
427 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
428 			blk = f->tf_fb;
429 			break;
430 		}
431 	}
432 	return(blk);
433 }
434 
435 static struct tcp_function_block *
436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
437 {
438 	struct tcp_function_block *rblk=NULL;
439 	struct tcp_function *f;
440 
441 	TAILQ_FOREACH(f, &t_functions, tf_next) {
442 		if (f->tf_fb == blk) {
443 			rblk = blk;
444 			if (s) {
445 				*s = f;
446 			}
447 			break;
448 		}
449 	}
450 	return (rblk);
451 }
452 
453 struct tcp_function_block *
454 find_and_ref_tcp_functions(struct tcp_function_set *fs)
455 {
456 	struct tcp_function_block *blk;
457 
458 	rw_rlock(&tcp_function_lock);
459 	blk = find_tcp_functions_locked(fs);
460 	if (blk)
461 		refcount_acquire(&blk->tfb_refcnt);
462 	rw_runlock(&tcp_function_lock);
463 	return(blk);
464 }
465 
466 struct tcp_function_block *
467 find_and_ref_tcp_fb(struct tcp_function_block *blk)
468 {
469 	struct tcp_function_block *rblk;
470 
471 	rw_rlock(&tcp_function_lock);
472 	rblk = find_tcp_fb_locked(blk, NULL);
473 	if (rblk)
474 		refcount_acquire(&rblk->tfb_refcnt);
475 	rw_runlock(&tcp_function_lock);
476 	return(rblk);
477 }
478 
479 /* Find a matching alias for the given tcp_function_block. */
480 int
481 find_tcp_function_alias(struct tcp_function_block *blk,
482     struct tcp_function_set *fs)
483 {
484 	struct tcp_function *f;
485 	int found;
486 
487 	found = 0;
488 	rw_rlock(&tcp_function_lock);
489 	TAILQ_FOREACH(f, &t_functions, tf_next) {
490 		if ((f->tf_fb == blk) &&
491 		    (strncmp(f->tf_name, blk->tfb_tcp_block_name,
492 		        TCP_FUNCTION_NAME_LEN_MAX) != 0)) {
493 			/* Matching function block with different name. */
494 			strncpy(fs->function_set_name, f->tf_name,
495 			    TCP_FUNCTION_NAME_LEN_MAX);
496 			found = 1;
497 			break;
498 		}
499 	}
500 	/* Null terminate the string appropriately. */
501 	if (found) {
502 		fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
503 	} else {
504 		fs->function_set_name[0] = '\0';
505 	}
506 	rw_runlock(&tcp_function_lock);
507 	return (found);
508 }
509 
510 static struct tcp_function_block *
511 find_and_ref_tcp_default_fb(void)
512 {
513 	struct tcp_function_block *rblk;
514 
515 	rw_rlock(&tcp_function_lock);
516 	rblk = tcp_func_set_ptr;
517 	refcount_acquire(&rblk->tfb_refcnt);
518 	rw_runlock(&tcp_function_lock);
519 	return (rblk);
520 }
521 
522 void
523 tcp_switch_back_to_default(struct tcpcb *tp)
524 {
525 	struct tcp_function_block *tfb;
526 
527 	KASSERT(tp->t_fb != &tcp_def_funcblk,
528 	    ("%s: called by the built-in default stack", __func__));
529 
530 	/*
531 	 * Release the old stack. This function will either find a new one
532 	 * or panic.
533 	 */
534 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
535 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
536 	refcount_release(&tp->t_fb->tfb_refcnt);
537 
538 	/*
539 	 * Now, we'll find a new function block to use.
540 	 * Start by trying the current user-selected
541 	 * default, unless this stack is the user-selected
542 	 * default.
543 	 */
544 	tfb = find_and_ref_tcp_default_fb();
545 	if (tfb == tp->t_fb) {
546 		refcount_release(&tfb->tfb_refcnt);
547 		tfb = NULL;
548 	}
549 	/* Does the stack accept this connection? */
550 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
551 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
552 		refcount_release(&tfb->tfb_refcnt);
553 		tfb = NULL;
554 	}
555 	/* Try to use that stack. */
556 	if (tfb != NULL) {
557 		/* Initialize the new stack. If it succeeds, we are done. */
558 		tp->t_fb = tfb;
559 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
560 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
561 			return;
562 
563 		/*
564 		 * Initialization failed. Release the reference count on
565 		 * the stack.
566 		 */
567 		refcount_release(&tfb->tfb_refcnt);
568 	}
569 
570 	/*
571 	 * If that wasn't feasible, use the built-in default
572 	 * stack which is not allowed to reject anyone.
573 	 */
574 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
575 	if (tfb == NULL) {
576 		/* there always should be a default */
577 		panic("Can't refer to tcp_def_funcblk");
578 	}
579 	if (tfb->tfb_tcp_handoff_ok != NULL) {
580 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
581 			/* The default stack cannot say no */
582 			panic("Default stack rejects a new session?");
583 		}
584 	}
585 	tp->t_fb = tfb;
586 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
587 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
588 		/* The default stack cannot fail */
589 		panic("Default stack initialization failed");
590 	}
591 }
592 
593 static void
594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
595     const struct sockaddr *sa, void *ctx)
596 {
597 	struct ip *iph;
598 #ifdef INET6
599 	struct ip6_hdr *ip6;
600 #endif
601 	struct udphdr *uh;
602 	struct tcphdr *th;
603 	int thlen;
604 	uint16_t port;
605 
606 	TCPSTAT_INC(tcps_tunneled_pkts);
607 	if ((m->m_flags & M_PKTHDR) == 0) {
608 		/* Can't handle one that is not a pkt hdr */
609 		TCPSTAT_INC(tcps_tunneled_errs);
610 		goto out;
611 	}
612 	thlen = sizeof(struct tcphdr);
613 	if (m->m_len < off + sizeof(struct udphdr) + thlen &&
614 	    (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
615 		TCPSTAT_INC(tcps_tunneled_errs);
616 		goto out;
617 	}
618 	iph = mtod(m, struct ip *);
619 	uh = (struct udphdr *)((caddr_t)iph + off);
620 	th = (struct tcphdr *)(uh + 1);
621 	thlen = th->th_off << 2;
622 	if (m->m_len < off + sizeof(struct udphdr) + thlen) {
623 		m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
624 		if (m == NULL) {
625 			TCPSTAT_INC(tcps_tunneled_errs);
626 			goto out;
627 		} else {
628 			iph = mtod(m, struct ip *);
629 			uh = (struct udphdr *)((caddr_t)iph + off);
630 			th = (struct tcphdr *)(uh + 1);
631 		}
632 	}
633 	m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
634 	bcopy(th, uh, m->m_len - off);
635 	m->m_len -= sizeof(struct udphdr);
636 	m->m_pkthdr.len -= sizeof(struct udphdr);
637 	/*
638 	 * We use the same algorithm for
639 	 * both UDP and TCP for c-sum. So
640 	 * the code in tcp_input will skip
641 	 * the checksum. So we do nothing
642 	 * with the flag (m->m_pkthdr.csum_flags).
643 	 */
644 	switch (iph->ip_v) {
645 #ifdef INET
646 	case IPVERSION:
647 		iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
648 		tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
649 		break;
650 #endif
651 #ifdef INET6
652 	case IPV6_VERSION >> 4:
653 		ip6 = mtod(m, struct ip6_hdr *);
654 		ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
655 		tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
656 		break;
657 #endif
658 	default:
659 		goto out;
660 		break;
661 	}
662 	return;
663 out:
664 	m_freem(m);
665 }
666 
667 static int
668 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
669 {
670 	int error=ENOENT;
671 	struct tcp_function_set fs;
672 	struct tcp_function_block *blk;
673 
674 	memset(&fs, 0, sizeof(fs));
675 	rw_rlock(&tcp_function_lock);
676 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
677 	if (blk) {
678 		/* Found him */
679 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
680 		fs.pcbcnt = blk->tfb_refcnt;
681 	}
682 	rw_runlock(&tcp_function_lock);
683 	error = sysctl_handle_string(oidp, fs.function_set_name,
684 				     sizeof(fs.function_set_name), req);
685 
686 	/* Check for error or no change */
687 	if (error != 0 || req->newptr == NULL)
688 		return(error);
689 
690 	rw_wlock(&tcp_function_lock);
691 	blk = find_tcp_functions_locked(&fs);
692 	if ((blk == NULL) ||
693 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
694 		error = ENOENT;
695 		goto done;
696 	}
697 	tcp_func_set_ptr = blk;
698 done:
699 	rw_wunlock(&tcp_function_lock);
700 	return (error);
701 }
702 
703 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
704     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
705     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
706     "Set/get the default TCP functions");
707 
708 static int
709 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
710 {
711 	int error, cnt, linesz;
712 	struct tcp_function *f;
713 	char *buffer, *cp;
714 	size_t bufsz, outsz;
715 	bool alias;
716 
717 	cnt = 0;
718 	rw_rlock(&tcp_function_lock);
719 	TAILQ_FOREACH(f, &t_functions, tf_next) {
720 		cnt++;
721 	}
722 	rw_runlock(&tcp_function_lock);
723 
724 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
725 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
726 
727 	error = 0;
728 	cp = buffer;
729 
730 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
731 	    "Alias", "PCB count");
732 	cp += linesz;
733 	bufsz -= linesz;
734 	outsz = linesz;
735 
736 	rw_rlock(&tcp_function_lock);
737 	TAILQ_FOREACH(f, &t_functions, tf_next) {
738 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
739 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
740 		    f->tf_fb->tfb_tcp_block_name,
741 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
742 		    alias ? f->tf_name : "-",
743 		    f->tf_fb->tfb_refcnt);
744 		if (linesz >= bufsz) {
745 			error = EOVERFLOW;
746 			break;
747 		}
748 		cp += linesz;
749 		bufsz -= linesz;
750 		outsz += linesz;
751 	}
752 	rw_runlock(&tcp_function_lock);
753 	if (error == 0)
754 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
755 	free(buffer, M_TEMP);
756 	return (error);
757 }
758 
759 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
760     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
761     NULL, 0, sysctl_net_inet_list_available, "A",
762     "list available TCP Function sets");
763 
764 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
765 
766 #ifdef INET
767 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
768 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
769 #endif
770 #ifdef INET6
771 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
772 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
773 #endif
774 
775 static void
776 tcp_over_udp_stop(void)
777 {
778 	/*
779 	 * This function assumes sysctl caller holds inp_rinfo_lock()
780 	 * for writing!
781 	 */
782 #ifdef INET
783 	if (V_udp4_tun_socket != NULL) {
784 		soclose(V_udp4_tun_socket);
785 		V_udp4_tun_socket = NULL;
786 	}
787 #endif
788 #ifdef INET6
789 	if (V_udp6_tun_socket != NULL) {
790 		soclose(V_udp6_tun_socket);
791 		V_udp6_tun_socket = NULL;
792 	}
793 #endif
794 }
795 
796 static int
797 tcp_over_udp_start(void)
798 {
799 	uint16_t port;
800 	int ret;
801 #ifdef INET
802 	struct sockaddr_in sin;
803 #endif
804 #ifdef INET6
805 	struct sockaddr_in6 sin6;
806 #endif
807 	/*
808 	 * This function assumes sysctl caller holds inp_info_rlock()
809 	 * for writing!
810 	 */
811 	port = V_tcp_udp_tunneling_port;
812 	if (ntohs(port) == 0) {
813 		/* Must have a port set */
814 		return (EINVAL);
815 	}
816 #ifdef INET
817 	if (V_udp4_tun_socket != NULL) {
818 		/* Already running -- must stop first */
819 		return (EALREADY);
820 	}
821 #endif
822 #ifdef INET6
823 	if (V_udp6_tun_socket != NULL) {
824 		/* Already running -- must stop first */
825 		return (EALREADY);
826 	}
827 #endif
828 #ifdef INET
829 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
830 	    SOCK_DGRAM, IPPROTO_UDP,
831 	    curthread->td_ucred, curthread))) {
832 		tcp_over_udp_stop();
833 		return (ret);
834 	}
835 	/* Call the special UDP hook. */
836 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
837 	    tcp_recv_udp_tunneled_packet,
838 	    tcp_ctlinput_viaudp,
839 	    NULL))) {
840 		tcp_over_udp_stop();
841 		return (ret);
842 	}
843 	/* Ok, we have a socket, bind it to the port. */
844 	memset(&sin, 0, sizeof(struct sockaddr_in));
845 	sin.sin_len = sizeof(struct sockaddr_in);
846 	sin.sin_family = AF_INET;
847 	sin.sin_port = htons(port);
848 	if ((ret = sobind(V_udp4_tun_socket,
849 	    (struct sockaddr *)&sin, curthread))) {
850 		tcp_over_udp_stop();
851 		return (ret);
852 	}
853 #endif
854 #ifdef INET6
855 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
856 	    SOCK_DGRAM, IPPROTO_UDP,
857 	    curthread->td_ucred, curthread))) {
858 		tcp_over_udp_stop();
859 		return (ret);
860 	}
861 	/* Call the special UDP hook. */
862 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
863 	    tcp_recv_udp_tunneled_packet,
864 	    tcp6_ctlinput_viaudp,
865 	    NULL))) {
866 		tcp_over_udp_stop();
867 		return (ret);
868 	}
869 	/* Ok, we have a socket, bind it to the port. */
870 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
871 	sin6.sin6_len = sizeof(struct sockaddr_in6);
872 	sin6.sin6_family = AF_INET6;
873 	sin6.sin6_port = htons(port);
874 	if ((ret = sobind(V_udp6_tun_socket,
875 	    (struct sockaddr *)&sin6, curthread))) {
876 		tcp_over_udp_stop();
877 		return (ret);
878 	}
879 #endif
880 	return (0);
881 }
882 
883 static int
884 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
885 {
886 	int error;
887 	uint32_t old, new;
888 
889 	old = V_tcp_udp_tunneling_port;
890 	new = old;
891 	error = sysctl_handle_int(oidp, &new, 0, req);
892 	if ((error == 0) &&
893 	    (req->newptr != NULL)) {
894 		if ((new < TCP_TUNNELING_PORT_MIN) ||
895 		    (new > TCP_TUNNELING_PORT_MAX)) {
896 			error = EINVAL;
897 		} else {
898 			V_tcp_udp_tunneling_port = new;
899 			if (old != 0) {
900 				tcp_over_udp_stop();
901 			}
902 			if (new != 0) {
903 				error = tcp_over_udp_start();
904 			}
905 		}
906 	}
907 	return (error);
908 }
909 
910 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
911     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
912     &VNET_NAME(tcp_udp_tunneling_port),
913     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
914     "Tunneling port for tcp over udp");
915 
916 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
917 
918 static int
919 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
920 {
921 	int error, new;
922 
923 	new = V_tcp_udp_tunneling_overhead;
924 	error = sysctl_handle_int(oidp, &new, 0, req);
925 	if (error == 0 && req->newptr) {
926 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
927 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
928 			error = EINVAL;
929 		else
930 			V_tcp_udp_tunneling_overhead = new;
931 	}
932 	return (error);
933 }
934 
935 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
936     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
937     &VNET_NAME(tcp_udp_tunneling_overhead),
938     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
939     "MSS reduction when using tcp over udp");
940 
941 /*
942  * Exports one (struct tcp_function_info) for each alias/name.
943  */
944 static int
945 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
946 {
947 	int cnt, error;
948 	struct tcp_function *f;
949 	struct tcp_function_info tfi;
950 
951 	/*
952 	 * We don't allow writes.
953 	 */
954 	if (req->newptr != NULL)
955 		return (EINVAL);
956 
957 	/*
958 	 * Wire the old buffer so we can directly copy the functions to
959 	 * user space without dropping the lock.
960 	 */
961 	if (req->oldptr != NULL) {
962 		error = sysctl_wire_old_buffer(req, 0);
963 		if (error)
964 			return (error);
965 	}
966 
967 	/*
968 	 * Walk the list and copy out matching entries. If INVARIANTS
969 	 * is compiled in, also walk the list to verify the length of
970 	 * the list matches what we have recorded.
971 	 */
972 	rw_rlock(&tcp_function_lock);
973 
974 	cnt = 0;
975 #ifndef INVARIANTS
976 	if (req->oldptr == NULL) {
977 		cnt = tcp_fb_cnt;
978 		goto skip_loop;
979 	}
980 #endif
981 	TAILQ_FOREACH(f, &t_functions, tf_next) {
982 #ifdef INVARIANTS
983 		cnt++;
984 #endif
985 		if (req->oldptr != NULL) {
986 			bzero(&tfi, sizeof(tfi));
987 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
988 			tfi.tfi_id = f->tf_fb->tfb_id;
989 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
990 			    sizeof(tfi.tfi_alias));
991 			(void)strlcpy(tfi.tfi_name,
992 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
993 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
994 			/*
995 			 * Don't stop on error, as that is the
996 			 * mechanism we use to accumulate length
997 			 * information if the buffer was too short.
998 			 */
999 		}
1000 	}
1001 	KASSERT(cnt == tcp_fb_cnt,
1002 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1003 #ifndef INVARIANTS
1004 skip_loop:
1005 #endif
1006 	rw_runlock(&tcp_function_lock);
1007 	if (req->oldptr == NULL)
1008 		error = SYSCTL_OUT(req, NULL,
1009 		    (cnt + 1) * sizeof(struct tcp_function_info));
1010 
1011 	return (error);
1012 }
1013 
1014 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1015 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1016 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1017 	    "List TCP function block name-to-ID mappings");
1018 
1019 /*
1020  * tfb_tcp_handoff_ok() function for the default stack.
1021  * Note that we'll basically try to take all comers.
1022  */
1023 static int
1024 tcp_default_handoff_ok(struct tcpcb *tp)
1025 {
1026 
1027 	return (0);
1028 }
1029 
1030 /*
1031  * tfb_tcp_fb_init() function for the default stack.
1032  *
1033  * This handles making sure we have appropriate timers set if you are
1034  * transitioning a socket that has some amount of setup done.
1035  *
1036  * The init() fuction from the default can *never* return non-zero i.e.
1037  * it is required to always succeed since it is the stack of last resort!
1038  */
1039 static int
1040 tcp_default_fb_init(struct tcpcb *tp)
1041 {
1042 
1043 	struct socket *so;
1044 
1045 	INP_WLOCK_ASSERT(tp->t_inpcb);
1046 
1047 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1048 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1049 	    tp->t_state));
1050 
1051 	/*
1052 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1053 	 * know what to do for unexpected states (which includes TIME_WAIT).
1054 	 */
1055 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1056 		return (0);
1057 
1058 	/*
1059 	 * Make sure some kind of transmission timer is set if there is
1060 	 * outstanding data.
1061 	 */
1062 	so = tp->t_inpcb->inp_socket;
1063 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1064 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1065 	    tcp_timer_active(tp, TT_PERSIST))) {
1066 		/*
1067 		 * If the session has established and it looks like it should
1068 		 * be in the persist state, set the persist timer. Otherwise,
1069 		 * set the retransmit timer.
1070 		 */
1071 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1072 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1073 		    (int32_t)sbavail(&so->so_snd))
1074 			tcp_setpersist(tp);
1075 		else
1076 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1077 	}
1078 
1079 	/* All non-embryonic sessions get a keepalive timer. */
1080 	if (!tcp_timer_active(tp, TT_KEEP))
1081 		tcp_timer_activate(tp, TT_KEEP,
1082 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1083 		    TP_KEEPINIT(tp));
1084 
1085 	/*
1086 	 * Make sure critical variables are initialized
1087 	 * if transitioning while in Recovery.
1088 	 */
1089 	if IN_FASTRECOVERY(tp->t_flags) {
1090 		if (tp->sackhint.recover_fs == 0)
1091 			tp->sackhint.recover_fs = max(1,
1092 			    tp->snd_nxt - tp->snd_una);
1093 	}
1094 
1095 	return (0);
1096 }
1097 
1098 /*
1099  * tfb_tcp_fb_fini() function for the default stack.
1100  *
1101  * This changes state as necessary (or prudent) to prepare for another stack
1102  * to assume responsibility for the connection.
1103  */
1104 static void
1105 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1106 {
1107 
1108 	INP_WLOCK_ASSERT(tp->t_inpcb);
1109 	return;
1110 }
1111 
1112 /*
1113  * Target size of TCP PCB hash tables. Must be a power of two.
1114  *
1115  * Note that this can be overridden by the kernel environment
1116  * variable net.inet.tcp.tcbhashsize
1117  */
1118 #ifndef TCBHASHSIZE
1119 #define TCBHASHSIZE	0
1120 #endif
1121 
1122 /*
1123  * XXX
1124  * Callouts should be moved into struct tcp directly.  They are currently
1125  * separate because the tcpcb structure is exported to userland for sysctl
1126  * parsing purposes, which do not know about callouts.
1127  */
1128 struct tcpcb_mem {
1129 	struct	tcpcb		tcb;
1130 	struct	tcp_timer	tt;
1131 	struct	cc_var		ccv;
1132 #ifdef TCP_HHOOK
1133 	struct	osd		osd;
1134 #endif
1135 };
1136 
1137 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1138 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1139 
1140 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1141 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1142 
1143 static struct mtx isn_mtx;
1144 
1145 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1146 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1147 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1148 
1149 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1150 
1151 /*
1152  * Take a value and get the next power of 2 that doesn't overflow.
1153  * Used to size the tcp_inpcb hash buckets.
1154  */
1155 static int
1156 maketcp_hashsize(int size)
1157 {
1158 	int hashsize;
1159 
1160 	/*
1161 	 * auto tune.
1162 	 * get the next power of 2 higher than maxsockets.
1163 	 */
1164 	hashsize = 1 << fls(size);
1165 	/* catch overflow, and just go one power of 2 smaller */
1166 	if (hashsize < size) {
1167 		hashsize = 1 << (fls(size) - 1);
1168 	}
1169 	return (hashsize);
1170 }
1171 
1172 static volatile int next_tcp_stack_id = 1;
1173 
1174 /*
1175  * Register a TCP function block with the name provided in the names
1176  * array.  (Note that this function does NOT automatically register
1177  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1178  * explicitly include blk->tfb_tcp_block_name in the list of names if
1179  * you wish to register the stack with that name.)
1180  *
1181  * Either all name registrations will succeed or all will fail.  If
1182  * a name registration fails, the function will update the num_names
1183  * argument to point to the array index of the name that encountered
1184  * the failure.
1185  *
1186  * Returns 0 on success, or an error code on failure.
1187  */
1188 int
1189 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1190     const char *names[], int *num_names)
1191 {
1192 	struct tcp_function *n;
1193 	struct tcp_function_set fs;
1194 	int error, i;
1195 
1196 	KASSERT(names != NULL && *num_names > 0,
1197 	    ("%s: Called with 0-length name list", __func__));
1198 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1199 	KASSERT(rw_initialized(&tcp_function_lock),
1200 	    ("%s: called too early", __func__));
1201 
1202 	if ((blk->tfb_tcp_output == NULL) ||
1203 	    (blk->tfb_tcp_do_segment == NULL) ||
1204 	    (blk->tfb_tcp_ctloutput == NULL) ||
1205 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1206 		/*
1207 		 * These functions are required and you
1208 		 * need a name.
1209 		 */
1210 		*num_names = 0;
1211 		return (EINVAL);
1212 	}
1213 	if (blk->tfb_tcp_timer_stop_all ||
1214 	    blk->tfb_tcp_timer_activate ||
1215 	    blk->tfb_tcp_timer_active ||
1216 	    blk->tfb_tcp_timer_stop) {
1217 		/*
1218 		 * If you define one timer function you
1219 		 * must have them all.
1220 		 */
1221 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1222 		    (blk->tfb_tcp_timer_activate == NULL) ||
1223 		    (blk->tfb_tcp_timer_active == NULL) ||
1224 		    (blk->tfb_tcp_timer_stop == NULL)) {
1225 			*num_names = 0;
1226 			return (EINVAL);
1227 		}
1228 	}
1229 
1230 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1231 		*num_names = 0;
1232 		return (EINVAL);
1233 	}
1234 
1235 	refcount_init(&blk->tfb_refcnt, 0);
1236 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1237 	for (i = 0; i < *num_names; i++) {
1238 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1239 		if (n == NULL) {
1240 			error = ENOMEM;
1241 			goto cleanup;
1242 		}
1243 		n->tf_fb = blk;
1244 
1245 		(void)strlcpy(fs.function_set_name, names[i],
1246 		    sizeof(fs.function_set_name));
1247 		rw_wlock(&tcp_function_lock);
1248 		if (find_tcp_functions_locked(&fs) != NULL) {
1249 			/* Duplicate name space not allowed */
1250 			rw_wunlock(&tcp_function_lock);
1251 			free(n, M_TCPFUNCTIONS);
1252 			error = EALREADY;
1253 			goto cleanup;
1254 		}
1255 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1256 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1257 		tcp_fb_cnt++;
1258 		rw_wunlock(&tcp_function_lock);
1259 	}
1260 	return(0);
1261 
1262 cleanup:
1263 	/*
1264 	 * Deregister the names we just added. Because registration failed
1265 	 * for names[i], we don't need to deregister that name.
1266 	 */
1267 	*num_names = i;
1268 	rw_wlock(&tcp_function_lock);
1269 	while (--i >= 0) {
1270 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1271 			if (!strncmp(n->tf_name, names[i],
1272 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1273 				TAILQ_REMOVE(&t_functions, n, tf_next);
1274 				tcp_fb_cnt--;
1275 				n->tf_fb = NULL;
1276 				free(n, M_TCPFUNCTIONS);
1277 				break;
1278 			}
1279 		}
1280 	}
1281 	rw_wunlock(&tcp_function_lock);
1282 	return (error);
1283 }
1284 
1285 /*
1286  * Register a TCP function block using the name provided in the name
1287  * argument.
1288  *
1289  * Returns 0 on success, or an error code on failure.
1290  */
1291 int
1292 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1293     int wait)
1294 {
1295 	const char *name_list[1];
1296 	int num_names, rv;
1297 
1298 	num_names = 1;
1299 	if (name != NULL)
1300 		name_list[0] = name;
1301 	else
1302 		name_list[0] = blk->tfb_tcp_block_name;
1303 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1304 	return (rv);
1305 }
1306 
1307 /*
1308  * Register a TCP function block using the name defined in
1309  * blk->tfb_tcp_block_name.
1310  *
1311  * Returns 0 on success, or an error code on failure.
1312  */
1313 int
1314 register_tcp_functions(struct tcp_function_block *blk, int wait)
1315 {
1316 
1317 	return (register_tcp_functions_as_name(blk, NULL, wait));
1318 }
1319 
1320 /*
1321  * Deregister all names associated with a function block. This
1322  * functionally removes the function block from use within the system.
1323  *
1324  * When called with a true quiesce argument, mark the function block
1325  * as being removed so no more stacks will use it and determine
1326  * whether the removal would succeed.
1327  *
1328  * When called with a false quiesce argument, actually attempt the
1329  * removal.
1330  *
1331  * When called with a force argument, attempt to switch all TCBs to
1332  * use the default stack instead of returning EBUSY.
1333  *
1334  * Returns 0 on success (or if the removal would succeed, or an error
1335  * code on failure.
1336  */
1337 int
1338 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1339     bool force)
1340 {
1341 	struct tcp_function *f;
1342 
1343 	if (blk == &tcp_def_funcblk) {
1344 		/* You can't un-register the default */
1345 		return (EPERM);
1346 	}
1347 	rw_wlock(&tcp_function_lock);
1348 	if (blk == tcp_func_set_ptr) {
1349 		/* You can't free the current default */
1350 		rw_wunlock(&tcp_function_lock);
1351 		return (EBUSY);
1352 	}
1353 	/* Mark the block so no more stacks can use it. */
1354 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1355 	/*
1356 	 * If TCBs are still attached to the stack, attempt to switch them
1357 	 * to the default stack.
1358 	 */
1359 	if (force && blk->tfb_refcnt) {
1360 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1361 		    INPLOOKUP_WLOCKPCB);
1362 		struct inpcb *inp;
1363 		struct tcpcb *tp;
1364 		VNET_ITERATOR_DECL(vnet_iter);
1365 
1366 		rw_wunlock(&tcp_function_lock);
1367 
1368 		VNET_LIST_RLOCK();
1369 		VNET_FOREACH(vnet_iter) {
1370 			CURVNET_SET(vnet_iter);
1371 			while ((inp = inp_next(&inpi)) != NULL) {
1372 				if (inp->inp_flags & INP_TIMEWAIT)
1373 					continue;
1374 				tp = intotcpcb(inp);
1375 				if (tp == NULL || tp->t_fb != blk)
1376 					continue;
1377 				tcp_switch_back_to_default(tp);
1378 			}
1379 			CURVNET_RESTORE();
1380 		}
1381 		VNET_LIST_RUNLOCK();
1382 
1383 		rw_wlock(&tcp_function_lock);
1384 	}
1385 	if (blk->tfb_refcnt) {
1386 		/* TCBs still attached. */
1387 		rw_wunlock(&tcp_function_lock);
1388 		return (EBUSY);
1389 	}
1390 	if (quiesce) {
1391 		/* Skip removal. */
1392 		rw_wunlock(&tcp_function_lock);
1393 		return (0);
1394 	}
1395 	/* Remove any function names that map to this function block. */
1396 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1397 		TAILQ_REMOVE(&t_functions, f, tf_next);
1398 		tcp_fb_cnt--;
1399 		f->tf_fb = NULL;
1400 		free(f, M_TCPFUNCTIONS);
1401 	}
1402 	rw_wunlock(&tcp_function_lock);
1403 	return (0);
1404 }
1405 
1406 static void
1407 tcp_vnet_init(void *arg __unused)
1408 {
1409 
1410 #ifdef TCP_HHOOK
1411 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1412 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1413 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1414 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1415 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1416 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1417 #endif
1418 #ifdef STATS
1419 	if (tcp_stats_init())
1420 		printf("%s: WARNING: unable to initialise TCP stats\n",
1421 		    __func__);
1422 #endif
1423 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1424 	    tcp_tcbhashsize);
1425 
1426 	/*
1427 	 * These have to be type stable for the benefit of the timers.
1428 	 */
1429 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1430 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1431 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1432 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1433 
1434 	tcp_tw_init();
1435 	syncache_init();
1436 	tcp_hc_init();
1437 
1438 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1439 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1440 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1441 
1442 	tcp_fastopen_init();
1443 
1444 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1445 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1446 
1447 	V_tcp_msl = TCPTV_MSL;
1448 }
1449 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1450     tcp_vnet_init, NULL);
1451 
1452 static void
1453 tcp_init(void *arg __unused)
1454 {
1455 	const char *tcbhash_tuneable;
1456 	int hashsize;
1457 
1458 	tcp_reass_global_init();
1459 
1460 	/* XXX virtualize those below? */
1461 	tcp_delacktime = TCPTV_DELACK;
1462 	tcp_keepinit = TCPTV_KEEP_INIT;
1463 	tcp_keepidle = TCPTV_KEEP_IDLE;
1464 	tcp_keepintvl = TCPTV_KEEPINTVL;
1465 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1466 	tcp_rexmit_initial = TCPTV_RTOBASE;
1467 	if (tcp_rexmit_initial < 1)
1468 		tcp_rexmit_initial = 1;
1469 	tcp_rexmit_min = TCPTV_MIN;
1470 	if (tcp_rexmit_min < 1)
1471 		tcp_rexmit_min = 1;
1472 	tcp_persmin = TCPTV_PERSMIN;
1473 	tcp_persmax = TCPTV_PERSMAX;
1474 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1475 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1476 
1477 	/* Setup the tcp function block list */
1478 	TAILQ_INIT(&t_functions);
1479 	rw_init(&tcp_function_lock, "tcp_func_lock");
1480 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1481 #ifdef TCP_BLACKBOX
1482 	/* Initialize the TCP logging data. */
1483 	tcp_log_init();
1484 #endif
1485 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1486 
1487 	if (tcp_soreceive_stream) {
1488 #ifdef INET
1489 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1490 #endif
1491 #ifdef INET6
1492 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1493 #endif /* INET6 */
1494 	}
1495 
1496 #ifdef INET6
1497 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1498 #else /* INET6 */
1499 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1500 #endif /* INET6 */
1501 	if (max_protohdr < TCP_MINPROTOHDR)
1502 		max_protohdr = TCP_MINPROTOHDR;
1503 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1504 		panic("tcp_init");
1505 #undef TCP_MINPROTOHDR
1506 
1507 	ISN_LOCK_INIT();
1508 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1509 		SHUTDOWN_PRI_DEFAULT);
1510 
1511 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1512 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1513 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1514 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1515 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1516 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1517 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1518 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1519 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1520 #ifdef TCPPCAP
1521 	tcp_pcap_init();
1522 #endif
1523 
1524 	hashsize = TCBHASHSIZE;
1525 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1526 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1527 	if (hashsize == 0) {
1528 		/*
1529 		 * Auto tune the hash size based on maxsockets.
1530 		 * A perfect hash would have a 1:1 mapping
1531 		 * (hashsize = maxsockets) however it's been
1532 		 * suggested that O(2) average is better.
1533 		 */
1534 		hashsize = maketcp_hashsize(maxsockets / 4);
1535 		/*
1536 		 * Our historical default is 512,
1537 		 * do not autotune lower than this.
1538 		 */
1539 		if (hashsize < 512)
1540 			hashsize = 512;
1541 		if (bootverbose)
1542 			printf("%s: %s auto tuned to %d\n", __func__,
1543 			    tcbhash_tuneable, hashsize);
1544 	}
1545 	/*
1546 	 * We require a hashsize to be a power of two.
1547 	 * Previously if it was not a power of two we would just reset it
1548 	 * back to 512, which could be a nasty surprise if you did not notice
1549 	 * the error message.
1550 	 * Instead what we do is clip it to the closest power of two lower
1551 	 * than the specified hash value.
1552 	 */
1553 	if (!powerof2(hashsize)) {
1554 		int oldhashsize = hashsize;
1555 
1556 		hashsize = maketcp_hashsize(hashsize);
1557 		/* prevent absurdly low value */
1558 		if (hashsize < 16)
1559 			hashsize = 16;
1560 		printf("%s: WARNING: TCB hash size not a power of 2, "
1561 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1562 		    hashsize);
1563 	}
1564 	tcp_tcbhashsize = hashsize;
1565 }
1566 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1567 
1568 #ifdef VIMAGE
1569 static void
1570 tcp_destroy(void *unused __unused)
1571 {
1572 	int n;
1573 #ifdef TCP_HHOOK
1574 	int error;
1575 #endif
1576 
1577 	/*
1578 	 * All our processes are gone, all our sockets should be cleaned
1579 	 * up, which means, we should be past the tcp_discardcb() calls.
1580 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1581 	 */
1582 	for (;;) {
1583 		INP_INFO_WLOCK(&V_tcbinfo);
1584 		n = V_tcbinfo.ipi_count;
1585 		INP_INFO_WUNLOCK(&V_tcbinfo);
1586 		if (n == 0)
1587 			break;
1588 		pause("tcpdes", hz / 10);
1589 	}
1590 	tcp_hc_destroy();
1591 	syncache_destroy();
1592 	tcp_tw_destroy();
1593 	in_pcbinfo_destroy(&V_tcbinfo);
1594 	/* tcp_discardcb() clears the sack_holes up. */
1595 	uma_zdestroy(V_sack_hole_zone);
1596 	uma_zdestroy(V_tcpcb_zone);
1597 
1598 	/*
1599 	 * Cannot free the zone until all tcpcbs are released as we attach
1600 	 * the allocations to them.
1601 	 */
1602 	tcp_fastopen_destroy();
1603 
1604 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1605 	VNET_PCPUSTAT_FREE(tcpstat);
1606 
1607 #ifdef TCP_HHOOK
1608 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1609 	if (error != 0) {
1610 		printf("%s: WARNING: unable to deregister helper hook "
1611 		    "type=%d, id=%d: error %d returned\n", __func__,
1612 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1613 	}
1614 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1615 	if (error != 0) {
1616 		printf("%s: WARNING: unable to deregister helper hook "
1617 		    "type=%d, id=%d: error %d returned\n", __func__,
1618 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1619 	}
1620 #endif
1621 }
1622 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1623 #endif
1624 
1625 void
1626 tcp_fini(void *xtp)
1627 {
1628 
1629 }
1630 
1631 /*
1632  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1633  * tcp_template used to store this data in mbufs, but we now recopy it out
1634  * of the tcpcb each time to conserve mbufs.
1635  */
1636 void
1637 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1638 {
1639 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1640 
1641 	INP_WLOCK_ASSERT(inp);
1642 
1643 #ifdef INET6
1644 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1645 		struct ip6_hdr *ip6;
1646 
1647 		ip6 = (struct ip6_hdr *)ip_ptr;
1648 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1649 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1650 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1651 			(IPV6_VERSION & IPV6_VERSION_MASK);
1652 		if (port == 0)
1653 			ip6->ip6_nxt = IPPROTO_TCP;
1654 		else
1655 			ip6->ip6_nxt = IPPROTO_UDP;
1656 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1657 		ip6->ip6_src = inp->in6p_laddr;
1658 		ip6->ip6_dst = inp->in6p_faddr;
1659 	}
1660 #endif /* INET6 */
1661 #if defined(INET6) && defined(INET)
1662 	else
1663 #endif
1664 #ifdef INET
1665 	{
1666 		struct ip *ip;
1667 
1668 		ip = (struct ip *)ip_ptr;
1669 		ip->ip_v = IPVERSION;
1670 		ip->ip_hl = 5;
1671 		ip->ip_tos = inp->inp_ip_tos;
1672 		ip->ip_len = 0;
1673 		ip->ip_id = 0;
1674 		ip->ip_off = 0;
1675 		ip->ip_ttl = inp->inp_ip_ttl;
1676 		ip->ip_sum = 0;
1677 		if (port == 0)
1678 			ip->ip_p = IPPROTO_TCP;
1679 		else
1680 			ip->ip_p = IPPROTO_UDP;
1681 		ip->ip_src = inp->inp_laddr;
1682 		ip->ip_dst = inp->inp_faddr;
1683 	}
1684 #endif /* INET */
1685 	th->th_sport = inp->inp_lport;
1686 	th->th_dport = inp->inp_fport;
1687 	th->th_seq = 0;
1688 	th->th_ack = 0;
1689 	th->th_off = 5;
1690 	tcp_set_flags(th, 0);
1691 	th->th_win = 0;
1692 	th->th_urp = 0;
1693 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1694 }
1695 
1696 /*
1697  * Create template to be used to send tcp packets on a connection.
1698  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1699  * use for this function is in keepalives, which use tcp_respond.
1700  */
1701 struct tcptemp *
1702 tcpip_maketemplate(struct inpcb *inp)
1703 {
1704 	struct tcptemp *t;
1705 
1706 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1707 	if (t == NULL)
1708 		return (NULL);
1709 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1710 	return (t);
1711 }
1712 
1713 /*
1714  * Send a single message to the TCP at address specified by
1715  * the given TCP/IP header.  If m == NULL, then we make a copy
1716  * of the tcpiphdr at th and send directly to the addressed host.
1717  * This is used to force keep alive messages out using the TCP
1718  * template for a connection.  If flags are given then we send
1719  * a message back to the TCP which originated the segment th,
1720  * and discard the mbuf containing it and any other attached mbufs.
1721  *
1722  * In any case the ack and sequence number of the transmitted
1723  * segment are as specified by the parameters.
1724  *
1725  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1726  */
1727 void
1728 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1729     tcp_seq ack, tcp_seq seq, int flags)
1730 {
1731 	struct tcpopt to;
1732 	struct inpcb *inp;
1733 	struct ip *ip;
1734 	struct mbuf *optm;
1735 	struct udphdr *uh = NULL;
1736 	struct tcphdr *nth;
1737 	struct tcp_log_buffer *lgb;
1738 	u_char *optp;
1739 #ifdef INET6
1740 	struct ip6_hdr *ip6;
1741 	int isipv6;
1742 #endif /* INET6 */
1743 	int optlen, tlen, win, ulen;
1744 	bool incl_opts;
1745 	uint16_t port;
1746 	int output_ret;
1747 #ifdef INVARIANTS
1748 	int thflags = tcp_get_flags(th);
1749 #endif
1750 
1751 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1752 	NET_EPOCH_ASSERT();
1753 
1754 #ifdef INET6
1755 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1756 	ip6 = ipgen;
1757 #endif /* INET6 */
1758 	ip = ipgen;
1759 
1760 	if (tp != NULL) {
1761 		inp = tp->t_inpcb;
1762 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1763 		INP_LOCK_ASSERT(inp);
1764 	} else
1765 		inp = NULL;
1766 
1767 	if (m != NULL) {
1768 #ifdef INET6
1769 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1770 			port = m->m_pkthdr.tcp_tun_port;
1771 		else
1772 #endif
1773 		if (ip && (ip->ip_p == IPPROTO_UDP))
1774 			port = m->m_pkthdr.tcp_tun_port;
1775 		else
1776 			port = 0;
1777 	} else
1778 		port = tp->t_port;
1779 
1780 	incl_opts = false;
1781 	win = 0;
1782 	if (tp != NULL) {
1783 		if (!(flags & TH_RST)) {
1784 			win = sbspace(&inp->inp_socket->so_rcv);
1785 			if (win > TCP_MAXWIN << tp->rcv_scale)
1786 				win = TCP_MAXWIN << tp->rcv_scale;
1787 		}
1788 		if ((tp->t_flags & TF_NOOPT) == 0)
1789 			incl_opts = true;
1790 	}
1791 	if (m == NULL) {
1792 		m = m_gethdr(M_NOWAIT, MT_DATA);
1793 		if (m == NULL)
1794 			return;
1795 		m->m_data += max_linkhdr;
1796 #ifdef INET6
1797 		if (isipv6) {
1798 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1799 			      sizeof(struct ip6_hdr));
1800 			ip6 = mtod(m, struct ip6_hdr *);
1801 			nth = (struct tcphdr *)(ip6 + 1);
1802 			if (port) {
1803 				/* Insert a UDP header */
1804 				uh = (struct udphdr *)nth;
1805 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1806 				uh->uh_dport = port;
1807 				nth = (struct tcphdr *)(uh + 1);
1808 			}
1809 		} else
1810 #endif /* INET6 */
1811 		{
1812 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1813 			ip = mtod(m, struct ip *);
1814 			nth = (struct tcphdr *)(ip + 1);
1815 			if (port) {
1816 				/* Insert a UDP header */
1817 				uh = (struct udphdr *)nth;
1818 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1819 				uh->uh_dport = port;
1820 				nth = (struct tcphdr *)(uh + 1);
1821 			}
1822 		}
1823 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1824 		flags = TH_ACK;
1825 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1826 		struct mbuf *n;
1827 
1828 		/* Can't reuse 'm', allocate a new mbuf. */
1829 		n = m_gethdr(M_NOWAIT, MT_DATA);
1830 		if (n == NULL) {
1831 			m_freem(m);
1832 			return;
1833 		}
1834 
1835 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1836 			m_freem(m);
1837 			m_freem(n);
1838 			return;
1839 		}
1840 
1841 		n->m_data += max_linkhdr;
1842 		/* m_len is set later */
1843 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1844 #ifdef INET6
1845 		if (isipv6) {
1846 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1847 			      sizeof(struct ip6_hdr));
1848 			ip6 = mtod(n, struct ip6_hdr *);
1849 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1850 			nth = (struct tcphdr *)(ip6 + 1);
1851 			if (port) {
1852 				/* Insert a UDP header */
1853 				uh = (struct udphdr *)nth;
1854 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1855 				uh->uh_dport = port;
1856 				nth = (struct tcphdr *)(uh + 1);
1857 			}
1858 		} else
1859 #endif /* INET6 */
1860 		{
1861 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1862 			ip = mtod(n, struct ip *);
1863 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1864 			nth = (struct tcphdr *)(ip + 1);
1865 			if (port) {
1866 				/* Insert a UDP header */
1867 				uh = (struct udphdr *)nth;
1868 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1869 				uh->uh_dport = port;
1870 				nth = (struct tcphdr *)(uh + 1);
1871 			}
1872 		}
1873 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1874 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1875 		th = nth;
1876 		m_freem(m);
1877 		m = n;
1878 	} else {
1879 		/*
1880 		 *  reuse the mbuf.
1881 		 * XXX MRT We inherit the FIB, which is lucky.
1882 		 */
1883 		m_freem(m->m_next);
1884 		m->m_next = NULL;
1885 		m->m_data = (caddr_t)ipgen;
1886 		/* m_len is set later */
1887 #ifdef INET6
1888 		if (isipv6) {
1889 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1890 			nth = (struct tcphdr *)(ip6 + 1);
1891 		} else
1892 #endif /* INET6 */
1893 		{
1894 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1895 			nth = (struct tcphdr *)(ip + 1);
1896 		}
1897 		if (th != nth) {
1898 			/*
1899 			 * this is usually a case when an extension header
1900 			 * exists between the IPv6 header and the
1901 			 * TCP header.
1902 			 */
1903 			nth->th_sport = th->th_sport;
1904 			nth->th_dport = th->th_dport;
1905 		}
1906 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1907 #undef xchg
1908 	}
1909 	tlen = 0;
1910 #ifdef INET6
1911 	if (isipv6)
1912 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1913 #endif
1914 #if defined(INET) && defined(INET6)
1915 	else
1916 #endif
1917 #ifdef INET
1918 		tlen = sizeof (struct tcpiphdr);
1919 #endif
1920 	if (port)
1921 		tlen += sizeof (struct udphdr);
1922 #ifdef INVARIANTS
1923 	m->m_len = 0;
1924 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1925 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1926 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1927 #endif
1928 	m->m_len = tlen;
1929 	to.to_flags = 0;
1930 	if (incl_opts) {
1931 		/* Make sure we have room. */
1932 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1933 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1934 			if (m->m_next) {
1935 				optp = mtod(m->m_next, u_char *);
1936 				optm = m->m_next;
1937 			} else
1938 				incl_opts = false;
1939 		} else {
1940 			optp = (u_char *) (nth + 1);
1941 			optm = m;
1942 		}
1943 	}
1944 	if (incl_opts) {
1945 		/* Timestamps. */
1946 		if (tp->t_flags & TF_RCVD_TSTMP) {
1947 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1948 			to.to_tsecr = tp->ts_recent;
1949 			to.to_flags |= TOF_TS;
1950 		}
1951 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1952 		/* TCP-MD5 (RFC2385). */
1953 		if (tp->t_flags & TF_SIGNATURE)
1954 			to.to_flags |= TOF_SIGNATURE;
1955 #endif
1956 		/* Add the options. */
1957 		tlen += optlen = tcp_addoptions(&to, optp);
1958 
1959 		/* Update m_len in the correct mbuf. */
1960 		optm->m_len += optlen;
1961 	} else
1962 		optlen = 0;
1963 #ifdef INET6
1964 	if (isipv6) {
1965 		if (uh) {
1966 			ulen = tlen - sizeof(struct ip6_hdr);
1967 			uh->uh_ulen = htons(ulen);
1968 		}
1969 		ip6->ip6_flow = 0;
1970 		ip6->ip6_vfc = IPV6_VERSION;
1971 		if (port)
1972 			ip6->ip6_nxt = IPPROTO_UDP;
1973 		else
1974 			ip6->ip6_nxt = IPPROTO_TCP;
1975 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1976 	}
1977 #endif
1978 #if defined(INET) && defined(INET6)
1979 	else
1980 #endif
1981 #ifdef INET
1982 	{
1983 		if (uh) {
1984 			ulen = tlen - sizeof(struct ip);
1985 			uh->uh_ulen = htons(ulen);
1986 		}
1987 		ip->ip_len = htons(tlen);
1988 		ip->ip_ttl = V_ip_defttl;
1989 		if (port) {
1990 			ip->ip_p = IPPROTO_UDP;
1991 		} else {
1992 			ip->ip_p = IPPROTO_TCP;
1993 		}
1994 		if (V_path_mtu_discovery)
1995 			ip->ip_off |= htons(IP_DF);
1996 	}
1997 #endif
1998 	m->m_pkthdr.len = tlen;
1999 	m->m_pkthdr.rcvif = NULL;
2000 #ifdef MAC
2001 	if (inp != NULL) {
2002 		/*
2003 		 * Packet is associated with a socket, so allow the
2004 		 * label of the response to reflect the socket label.
2005 		 */
2006 		INP_LOCK_ASSERT(inp);
2007 		mac_inpcb_create_mbuf(inp, m);
2008 	} else {
2009 		/*
2010 		 * Packet is not associated with a socket, so possibly
2011 		 * update the label in place.
2012 		 */
2013 		mac_netinet_tcp_reply(m);
2014 	}
2015 #endif
2016 	nth->th_seq = htonl(seq);
2017 	nth->th_ack = htonl(ack);
2018 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2019 	tcp_set_flags(nth, flags);
2020 	if (tp != NULL)
2021 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2022 	else
2023 		nth->th_win = htons((u_short)win);
2024 	nth->th_urp = 0;
2025 
2026 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2027 	if (to.to_flags & TOF_SIGNATURE) {
2028 		if (!TCPMD5_ENABLED() ||
2029 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2030 			m_freem(m);
2031 			return;
2032 		}
2033 	}
2034 #endif
2035 
2036 #ifdef INET6
2037 	if (isipv6) {
2038 		if (port) {
2039 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2040 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2041 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2042 			nth->th_sum = 0;
2043 		} else {
2044 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2045 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2046 			nth->th_sum = in6_cksum_pseudo(ip6,
2047 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2048 		}
2049 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2050 		    NULL, NULL);
2051 	}
2052 #endif /* INET6 */
2053 #if defined(INET6) && defined(INET)
2054 	else
2055 #endif
2056 #ifdef INET
2057 	{
2058 		if (port) {
2059 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2060 			    htons(ulen + IPPROTO_UDP));
2061 			m->m_pkthdr.csum_flags = CSUM_UDP;
2062 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2063 			nth->th_sum = 0;
2064 		} else {
2065 			m->m_pkthdr.csum_flags = CSUM_TCP;
2066 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2067 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2068 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2069 		}
2070 	}
2071 #endif /* INET */
2072 #ifdef TCPDEBUG
2073 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2074 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2075 #endif
2076 	TCP_PROBE3(debug__output, tp, th, m);
2077 	if (flags & TH_RST)
2078 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2079 	lgb = NULL;
2080 	if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2081 		if (INP_WLOCKED(inp)) {
2082 			union tcp_log_stackspecific log;
2083 			struct timeval tv;
2084 
2085 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2086 			log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2087 			log.u_bbr.flex8 = 4;
2088 			log.u_bbr.pkts_out = tp->t_maxseg;
2089 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2090 			log.u_bbr.delivered = 0;
2091 			lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2092 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2093 		} else {
2094 			/*
2095 			 * We can not log the packet, since we only own the
2096 			 * read lock, but a write lock is needed. The read lock
2097 			 * is not upgraded to a write lock, since only getting
2098 			 * the read lock was done intentionally to improve the
2099 			 * handling of SYN flooding attacks.
2100 			 * This happens only for pure SYN segments received in
2101 			 * the initial CLOSED state, or received in a more
2102 			 * advanced state than listen and the UDP encapsulation
2103 			 * port is unexpected.
2104 			 * The incoming SYN segments do not really belong to
2105 			 * the TCP connection and the handling does not change
2106 			 * the state of the TCP connection. Therefore, the
2107 			 * sending of the RST segments is not logged. Please
2108 			 * note that also the incoming SYN segments are not
2109 			 * logged.
2110 			 *
2111 			 * The following code ensures that the above description
2112 			 * is and stays correct.
2113 			 */
2114 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2115 			    (tp->t_state == TCPS_CLOSED ||
2116 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2117 			    ("%s: Logging of TCP segment with flags 0x%b and "
2118 			    "UDP encapsulation port %u skipped in state %s",
2119 			    __func__, thflags, PRINT_TH_FLAGS,
2120 			    ntohs(port), tcpstates[tp->t_state]));
2121 		}
2122 	}
2123 
2124 #ifdef INET6
2125 	if (isipv6) {
2126 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2127 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2128 	}
2129 #endif /* INET6 */
2130 #if defined(INET) && defined(INET6)
2131 	else
2132 #endif
2133 #ifdef INET
2134 	{
2135 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2136 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2137 	}
2138 #endif
2139 	if (lgb != NULL)
2140 		lgb->tlb_errno = output_ret;
2141 }
2142 
2143 /*
2144  * Create a new TCP control block, making an
2145  * empty reassembly queue and hooking it to the argument
2146  * protocol control block.  The `inp' parameter must have
2147  * come from the zone allocator set up in tcp_init().
2148  */
2149 struct tcpcb *
2150 tcp_newtcpcb(struct inpcb *inp)
2151 {
2152 	struct tcpcb_mem *tm;
2153 	struct tcpcb *tp;
2154 #ifdef INET6
2155 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2156 #endif /* INET6 */
2157 
2158 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2159 	if (tm == NULL)
2160 		return (NULL);
2161 	tp = &tm->tcb;
2162 
2163 	/* Initialise cc_var struct for this tcpcb. */
2164 	tp->ccv = &tm->ccv;
2165 	tp->ccv->type = IPPROTO_TCP;
2166 	tp->ccv->ccvc.tcp = tp;
2167 	rw_rlock(&tcp_function_lock);
2168 	tp->t_fb = tcp_func_set_ptr;
2169 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2170 	rw_runlock(&tcp_function_lock);
2171 	/*
2172 	 * Use the current system default CC algorithm.
2173 	 */
2174 	cc_attach(tp, CC_DEFAULT_ALGO());
2175 
2176 	/*
2177 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2178 	 * is called.
2179 	 */
2180 	in_pcbref(inp);	/* Reference for tcpcb */
2181 	tp->t_inpcb = inp;
2182 
2183 	if (CC_ALGO(tp)->cb_init != NULL)
2184 		if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2185 			cc_detach(tp);
2186 			if (tp->t_fb->tfb_tcp_fb_fini)
2187 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2188 			in_pcbrele_wlocked(inp);
2189 			refcount_release(&tp->t_fb->tfb_refcnt);
2190 			uma_zfree(V_tcpcb_zone, tm);
2191 			return (NULL);
2192 		}
2193 
2194 #ifdef TCP_HHOOK
2195 	tp->osd = &tm->osd;
2196 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2197 		if (tp->t_fb->tfb_tcp_fb_fini)
2198 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2199 		in_pcbrele_wlocked(inp);
2200 		refcount_release(&tp->t_fb->tfb_refcnt);
2201 		uma_zfree(V_tcpcb_zone, tm);
2202 		return (NULL);
2203 	}
2204 #endif
2205 
2206 #ifdef VIMAGE
2207 	tp->t_vnet = inp->inp_vnet;
2208 #endif
2209 	tp->t_timers = &tm->tt;
2210 	TAILQ_INIT(&tp->t_segq);
2211 	tp->t_maxseg =
2212 #ifdef INET6
2213 		isipv6 ? V_tcp_v6mssdflt :
2214 #endif /* INET6 */
2215 		V_tcp_mssdflt;
2216 
2217 	/* Set up our timeouts. */
2218 	callout_init(&tp->t_timers->tt_rexmt, 1);
2219 	callout_init(&tp->t_timers->tt_persist, 1);
2220 	callout_init(&tp->t_timers->tt_keep, 1);
2221 	callout_init(&tp->t_timers->tt_2msl, 1);
2222 	callout_init(&tp->t_timers->tt_delack, 1);
2223 
2224 	if (V_tcp_do_rfc1323)
2225 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2226 	if (V_tcp_do_sack)
2227 		tp->t_flags |= TF_SACK_PERMIT;
2228 	TAILQ_INIT(&tp->snd_holes);
2229 
2230 	/*
2231 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2232 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2233 	 * reasonable initial retransmit time.
2234 	 */
2235 	tp->t_srtt = TCPTV_SRTTBASE;
2236 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2237 	tp->t_rttmin = tcp_rexmit_min;
2238 	tp->t_rxtcur = tcp_rexmit_initial;
2239 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2240 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2241 	tp->t_rcvtime = ticks;
2242 	/*
2243 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2244 	 * because the socket may be bound to an IPv6 wildcard address,
2245 	 * which may match an IPv4-mapped IPv6 address.
2246 	 */
2247 	inp->inp_ip_ttl = V_ip_defttl;
2248 	inp->inp_ppcb = tp;
2249 #ifdef TCPPCAP
2250 	/*
2251 	 * Init the TCP PCAP queues.
2252 	 */
2253 	tcp_pcap_tcpcb_init(tp);
2254 #endif
2255 #ifdef TCP_BLACKBOX
2256 	/* Initialize the per-TCPCB log data. */
2257 	tcp_log_tcpcbinit(tp);
2258 #endif
2259 	tp->t_pacing_rate = -1;
2260 	if (tp->t_fb->tfb_tcp_fb_init) {
2261 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2262 			refcount_release(&tp->t_fb->tfb_refcnt);
2263 			in_pcbrele_wlocked(inp);
2264 			uma_zfree(V_tcpcb_zone, tm);
2265 			return (NULL);
2266 		}
2267 	}
2268 #ifdef STATS
2269 	if (V_tcp_perconn_stats_enable == 1)
2270 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2271 #endif
2272 	if (V_tcp_do_lrd)
2273 		tp->t_flags |= TF_LRD;
2274 	return (tp);		/* XXX */
2275 }
2276 
2277 /*
2278  * Drop a TCP connection, reporting
2279  * the specified error.  If connection is synchronized,
2280  * then send a RST to peer.
2281  */
2282 struct tcpcb *
2283 tcp_drop(struct tcpcb *tp, int errno)
2284 {
2285 	struct socket *so = tp->t_inpcb->inp_socket;
2286 
2287 	NET_EPOCH_ASSERT();
2288 	INP_WLOCK_ASSERT(tp->t_inpcb);
2289 
2290 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2291 		tcp_state_change(tp, TCPS_CLOSED);
2292 		/* Don't use tcp_output() here due to possible recursion. */
2293 		(void)tcp_output_nodrop(tp);
2294 		TCPSTAT_INC(tcps_drops);
2295 	} else
2296 		TCPSTAT_INC(tcps_conndrops);
2297 	if (errno == ETIMEDOUT && tp->t_softerror)
2298 		errno = tp->t_softerror;
2299 	so->so_error = errno;
2300 	return (tcp_close(tp));
2301 }
2302 
2303 void
2304 tcp_discardcb(struct tcpcb *tp)
2305 {
2306 	struct inpcb *inp = tp->t_inpcb;
2307 
2308 	INP_WLOCK_ASSERT(inp);
2309 
2310 	/*
2311 	 * Make sure that all of our timers are stopped before we delete the
2312 	 * PCB.
2313 	 *
2314 	 * If stopping a timer fails, we schedule a discard function in same
2315 	 * callout, and the last discard function called will take care of
2316 	 * deleting the tcpcb.
2317 	 */
2318 	tp->t_timers->tt_draincnt = 0;
2319 	tcp_timer_stop(tp, TT_REXMT);
2320 	tcp_timer_stop(tp, TT_PERSIST);
2321 	tcp_timer_stop(tp, TT_KEEP);
2322 	tcp_timer_stop(tp, TT_2MSL);
2323 	tcp_timer_stop(tp, TT_DELACK);
2324 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2325 		/*
2326 		 * Call the stop-all function of the methods,
2327 		 * this function should call the tcp_timer_stop()
2328 		 * method with each of the function specific timeouts.
2329 		 * That stop will be called via the tfb_tcp_timer_stop()
2330 		 * which should use the async drain function of the
2331 		 * callout system (see tcp_var.h).
2332 		 */
2333 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2334 	}
2335 
2336 	/* free the reassembly queue, if any */
2337 	tcp_reass_flush(tp);
2338 
2339 #ifdef TCP_OFFLOAD
2340 	/* Disconnect offload device, if any. */
2341 	if (tp->t_flags & TF_TOE)
2342 		tcp_offload_detach(tp);
2343 #endif
2344 
2345 	tcp_free_sackholes(tp);
2346 
2347 #ifdef TCPPCAP
2348 	/* Free the TCP PCAP queues. */
2349 	tcp_pcap_drain(&(tp->t_inpkts));
2350 	tcp_pcap_drain(&(tp->t_outpkts));
2351 #endif
2352 
2353 	/* Allow the CC algorithm to clean up after itself. */
2354 	if (CC_ALGO(tp)->cb_destroy != NULL)
2355 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2356 	CC_DATA(tp) = NULL;
2357 	/* Detach from the CC algorithm */
2358 	cc_detach(tp);
2359 
2360 #ifdef TCP_HHOOK
2361 	khelp_destroy_osd(tp->osd);
2362 #endif
2363 #ifdef STATS
2364 	stats_blob_destroy(tp->t_stats);
2365 #endif
2366 
2367 	CC_ALGO(tp) = NULL;
2368 	inp->inp_ppcb = NULL;
2369 	if (tp->t_timers->tt_draincnt == 0) {
2370 		bool released __diagused;
2371 
2372 		released = tcp_freecb(tp);
2373 		KASSERT(!released, ("%s: inp %p should not have been released "
2374 		    "here", __func__, inp));
2375 	}
2376 }
2377 
2378 bool
2379 tcp_freecb(struct tcpcb *tp)
2380 {
2381 	struct inpcb *inp = tp->t_inpcb;
2382 	struct socket *so = inp->inp_socket;
2383 #ifdef INET6
2384 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2385 #endif
2386 
2387 	INP_WLOCK_ASSERT(inp);
2388 	MPASS(tp->t_timers->tt_draincnt == 0);
2389 
2390 	/* We own the last reference on tcpcb, let's free it. */
2391 #ifdef TCP_BLACKBOX
2392 	tcp_log_tcpcbfini(tp);
2393 #endif
2394 	TCPSTATES_DEC(tp->t_state);
2395 	if (tp->t_fb->tfb_tcp_fb_fini)
2396 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2397 
2398 	/*
2399 	 * If we got enough samples through the srtt filter,
2400 	 * save the rtt and rttvar in the routing entry.
2401 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2402 	 * 4 samples is enough for the srtt filter to converge
2403 	 * to within enough % of the correct value; fewer samples
2404 	 * and we could save a bogus rtt. The danger is not high
2405 	 * as tcp quickly recovers from everything.
2406 	 * XXX: Works very well but needs some more statistics!
2407 	 *
2408 	 * XXXRRS: Updating must be after the stack fini() since
2409 	 * that may be converting some internal representation of
2410 	 * say srtt etc into the general one used by other stacks.
2411 	 * Lets also at least protect against the so being NULL
2412 	 * as RW stated below.
2413 	 */
2414 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2415 		struct hc_metrics_lite metrics;
2416 		uint32_t ssthresh;
2417 
2418 		bzero(&metrics, sizeof(metrics));
2419 		/*
2420 		 * Update the ssthresh always when the conditions below
2421 		 * are satisfied. This gives us better new start value
2422 		 * for the congestion avoidance for new connections.
2423 		 * ssthresh is only set if packet loss occurred on a session.
2424 		 *
2425 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2426 		 * being torn down.  Ideally this code would not use 'so'.
2427 		 */
2428 		ssthresh = tp->snd_ssthresh;
2429 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2430 			/*
2431 			 * convert the limit from user data bytes to
2432 			 * packets then to packet data bytes.
2433 			 */
2434 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2435 			if (ssthresh < 2)
2436 				ssthresh = 2;
2437 			ssthresh *= (tp->t_maxseg +
2438 #ifdef INET6
2439 			    (isipv6 ? sizeof (struct ip6_hdr) +
2440 			    sizeof (struct tcphdr) :
2441 #endif
2442 			    sizeof (struct tcpiphdr)
2443 #ifdef INET6
2444 			    )
2445 #endif
2446 			    );
2447 		} else
2448 			ssthresh = 0;
2449 		metrics.rmx_ssthresh = ssthresh;
2450 
2451 		metrics.rmx_rtt = tp->t_srtt;
2452 		metrics.rmx_rttvar = tp->t_rttvar;
2453 		metrics.rmx_cwnd = tp->snd_cwnd;
2454 		metrics.rmx_sendpipe = 0;
2455 		metrics.rmx_recvpipe = 0;
2456 
2457 		tcp_hc_update(&inp->inp_inc, &metrics);
2458 	}
2459 
2460 	refcount_release(&tp->t_fb->tfb_refcnt);
2461 	uma_zfree(V_tcpcb_zone, tp);
2462 
2463 	return (in_pcbrele_wlocked(inp));
2464 }
2465 
2466 /*
2467  * Attempt to close a TCP control block, marking it as dropped, and freeing
2468  * the socket if we hold the only reference.
2469  */
2470 struct tcpcb *
2471 tcp_close(struct tcpcb *tp)
2472 {
2473 	struct inpcb *inp = tp->t_inpcb;
2474 	struct socket *so;
2475 
2476 	INP_WLOCK_ASSERT(inp);
2477 
2478 #ifdef TCP_OFFLOAD
2479 	if (tp->t_state == TCPS_LISTEN)
2480 		tcp_offload_listen_stop(tp);
2481 #endif
2482 	/*
2483 	 * This releases the TFO pending counter resource for TFO listen
2484 	 * sockets as well as passively-created TFO sockets that transition
2485 	 * from SYN_RECEIVED to CLOSED.
2486 	 */
2487 	if (tp->t_tfo_pending) {
2488 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2489 		tp->t_tfo_pending = NULL;
2490 	}
2491 #ifdef TCPHPTS
2492 	tcp_hpts_remove(inp);
2493 #endif
2494 	in_pcbdrop(inp);
2495 	TCPSTAT_INC(tcps_closed);
2496 	if (tp->t_state != TCPS_CLOSED)
2497 		tcp_state_change(tp, TCPS_CLOSED);
2498 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2499 	so = inp->inp_socket;
2500 	soisdisconnected(so);
2501 	if (inp->inp_flags & INP_SOCKREF) {
2502 		KASSERT(so->so_state & SS_PROTOREF,
2503 		    ("tcp_close: !SS_PROTOREF"));
2504 		inp->inp_flags &= ~INP_SOCKREF;
2505 		INP_WUNLOCK(inp);
2506 		SOCK_LOCK(so);
2507 		so->so_state &= ~SS_PROTOREF;
2508 		sofree(so);
2509 		return (NULL);
2510 	}
2511 	return (tp);
2512 }
2513 
2514 void
2515 tcp_drain(void)
2516 {
2517 	VNET_ITERATOR_DECL(vnet_iter);
2518 
2519 	if (!do_tcpdrain)
2520 		return;
2521 
2522 	VNET_LIST_RLOCK_NOSLEEP();
2523 	VNET_FOREACH(vnet_iter) {
2524 		CURVNET_SET(vnet_iter);
2525 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2526 		    INPLOOKUP_WLOCKPCB);
2527 		struct inpcb *inpb;
2528 		struct tcpcb *tcpb;
2529 
2530 	/*
2531 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2532 	 * if there is one...
2533 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2534 	 *      reassembly queue should be flushed, but in a situation
2535 	 *	where we're really low on mbufs, this is potentially
2536 	 *	useful.
2537 	 */
2538 		while ((inpb = inp_next(&inpi)) != NULL) {
2539 			if (inpb->inp_flags & INP_TIMEWAIT)
2540 				continue;
2541 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2542 				tcp_reass_flush(tcpb);
2543 				tcp_clean_sackreport(tcpb);
2544 #ifdef TCP_BLACKBOX
2545 				tcp_log_drain(tcpb);
2546 #endif
2547 #ifdef TCPPCAP
2548 				if (tcp_pcap_aggressive_free) {
2549 					/* Free the TCP PCAP queues. */
2550 					tcp_pcap_drain(&(tcpb->t_inpkts));
2551 					tcp_pcap_drain(&(tcpb->t_outpkts));
2552 				}
2553 #endif
2554 			}
2555 		}
2556 		CURVNET_RESTORE();
2557 	}
2558 	VNET_LIST_RUNLOCK_NOSLEEP();
2559 }
2560 
2561 /*
2562  * Notify a tcp user of an asynchronous error;
2563  * store error as soft error, but wake up user
2564  * (for now, won't do anything until can select for soft error).
2565  *
2566  * Do not wake up user since there currently is no mechanism for
2567  * reporting soft errors (yet - a kqueue filter may be added).
2568  */
2569 static struct inpcb *
2570 tcp_notify(struct inpcb *inp, int error)
2571 {
2572 	struct tcpcb *tp;
2573 
2574 	INP_WLOCK_ASSERT(inp);
2575 
2576 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2577 	    (inp->inp_flags & INP_DROPPED))
2578 		return (inp);
2579 
2580 	tp = intotcpcb(inp);
2581 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2582 
2583 	/*
2584 	 * Ignore some errors if we are hooked up.
2585 	 * If connection hasn't completed, has retransmitted several times,
2586 	 * and receives a second error, give up now.  This is better
2587 	 * than waiting a long time to establish a connection that
2588 	 * can never complete.
2589 	 */
2590 	if (tp->t_state == TCPS_ESTABLISHED &&
2591 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2592 	     error == EHOSTDOWN)) {
2593 		if (inp->inp_route.ro_nh) {
2594 			NH_FREE(inp->inp_route.ro_nh);
2595 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2596 		}
2597 		return (inp);
2598 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2599 	    tp->t_softerror) {
2600 		tp = tcp_drop(tp, error);
2601 		if (tp != NULL)
2602 			return (inp);
2603 		else
2604 			return (NULL);
2605 	} else {
2606 		tp->t_softerror = error;
2607 		return (inp);
2608 	}
2609 #if 0
2610 	wakeup( &so->so_timeo);
2611 	sorwakeup(so);
2612 	sowwakeup(so);
2613 #endif
2614 }
2615 
2616 static int
2617 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2618 {
2619 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2620 	    INPLOOKUP_RLOCKPCB);
2621 	struct xinpgen xig;
2622 	struct inpcb *inp;
2623 	int error;
2624 
2625 	if (req->newptr != NULL)
2626 		return (EPERM);
2627 
2628 	if (req->oldptr == NULL) {
2629 		int n;
2630 
2631 		n = V_tcbinfo.ipi_count +
2632 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2633 		n += imax(n / 8, 10);
2634 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2635 		return (0);
2636 	}
2637 
2638 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2639 		return (error);
2640 
2641 	bzero(&xig, sizeof(xig));
2642 	xig.xig_len = sizeof xig;
2643 	xig.xig_count = V_tcbinfo.ipi_count +
2644 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2645 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2646 	xig.xig_sogen = so_gencnt;
2647 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2648 	if (error)
2649 		return (error);
2650 
2651 	error = syncache_pcblist(req);
2652 	if (error)
2653 		return (error);
2654 
2655 	while ((inp = inp_next(&inpi)) != NULL) {
2656 		if (inp->inp_gencnt <= xig.xig_gen) {
2657 			int crerr;
2658 
2659 			/*
2660 			 * XXX: This use of cr_cansee(), introduced with
2661 			 * TCP state changes, is not quite right, but for
2662 			 * now, better than nothing.
2663 			 */
2664 			if (inp->inp_flags & INP_TIMEWAIT) {
2665 				if (intotw(inp) != NULL)
2666 					crerr = cr_cansee(req->td->td_ucred,
2667 					    intotw(inp)->tw_cred);
2668 				else
2669 					crerr = EINVAL;	/* Skip this inp. */
2670 			} else
2671 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2672 			if (crerr == 0) {
2673 				struct xtcpcb xt;
2674 
2675 				tcp_inptoxtp(inp, &xt);
2676 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2677 				if (error) {
2678 					INP_RUNLOCK(inp);
2679 					break;
2680 				} else
2681 					continue;
2682 			}
2683 		}
2684 	}
2685 
2686 	if (!error) {
2687 		/*
2688 		 * Give the user an updated idea of our state.
2689 		 * If the generation differs from what we told
2690 		 * her before, she knows that something happened
2691 		 * while we were processing this request, and it
2692 		 * might be necessary to retry.
2693 		 */
2694 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2695 		xig.xig_sogen = so_gencnt;
2696 		xig.xig_count = V_tcbinfo.ipi_count +
2697 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2698 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2699 	}
2700 
2701 	return (error);
2702 }
2703 
2704 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2705     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2706     NULL, 0, tcp_pcblist, "S,xtcpcb",
2707     "List of active TCP connections");
2708 
2709 #ifdef INET
2710 static int
2711 tcp_getcred(SYSCTL_HANDLER_ARGS)
2712 {
2713 	struct xucred xuc;
2714 	struct sockaddr_in addrs[2];
2715 	struct epoch_tracker et;
2716 	struct inpcb *inp;
2717 	int error;
2718 
2719 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2720 	if (error)
2721 		return (error);
2722 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2723 	if (error)
2724 		return (error);
2725 	NET_EPOCH_ENTER(et);
2726 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2727 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2728 	NET_EPOCH_EXIT(et);
2729 	if (inp != NULL) {
2730 		if (inp->inp_socket == NULL)
2731 			error = ENOENT;
2732 		if (error == 0)
2733 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2734 		if (error == 0)
2735 			cru2x(inp->inp_cred, &xuc);
2736 		INP_RUNLOCK(inp);
2737 	} else
2738 		error = ENOENT;
2739 	if (error == 0)
2740 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2741 	return (error);
2742 }
2743 
2744 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2745     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2746     0, 0, tcp_getcred, "S,xucred",
2747     "Get the xucred of a TCP connection");
2748 #endif /* INET */
2749 
2750 #ifdef INET6
2751 static int
2752 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2753 {
2754 	struct epoch_tracker et;
2755 	struct xucred xuc;
2756 	struct sockaddr_in6 addrs[2];
2757 	struct inpcb *inp;
2758 	int error;
2759 #ifdef INET
2760 	int mapped = 0;
2761 #endif
2762 
2763 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2764 	if (error)
2765 		return (error);
2766 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2767 	if (error)
2768 		return (error);
2769 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2770 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2771 		return (error);
2772 	}
2773 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2774 #ifdef INET
2775 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2776 			mapped = 1;
2777 		else
2778 #endif
2779 			return (EINVAL);
2780 	}
2781 
2782 	NET_EPOCH_ENTER(et);
2783 #ifdef INET
2784 	if (mapped == 1)
2785 		inp = in_pcblookup(&V_tcbinfo,
2786 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2787 			addrs[1].sin6_port,
2788 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2789 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2790 	else
2791 #endif
2792 		inp = in6_pcblookup(&V_tcbinfo,
2793 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2794 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2795 			INPLOOKUP_RLOCKPCB, NULL);
2796 	NET_EPOCH_EXIT(et);
2797 	if (inp != NULL) {
2798 		if (inp->inp_socket == NULL)
2799 			error = ENOENT;
2800 		if (error == 0)
2801 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2802 		if (error == 0)
2803 			cru2x(inp->inp_cred, &xuc);
2804 		INP_RUNLOCK(inp);
2805 	} else
2806 		error = ENOENT;
2807 	if (error == 0)
2808 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2809 	return (error);
2810 }
2811 
2812 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2813     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2814     0, 0, tcp6_getcred, "S,xucred",
2815     "Get the xucred of a TCP6 connection");
2816 #endif /* INET6 */
2817 
2818 #ifdef INET
2819 /* Path MTU to try next when a fragmentation-needed message is received. */
2820 static inline int
2821 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2822 {
2823 	int mtu = ntohs(icp->icmp_nextmtu);
2824 
2825 	/* If no alternative MTU was proposed, try the next smaller one. */
2826 	if (!mtu)
2827 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2828 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2829 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2830 
2831 	return (mtu);
2832 }
2833 
2834 static void
2835 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2836 {
2837 	struct ip *ip = vip;
2838 	struct tcphdr *th;
2839 	struct in_addr faddr;
2840 	struct inpcb *inp;
2841 	struct tcpcb *tp;
2842 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2843 	struct icmp *icp;
2844 	struct in_conninfo inc;
2845 	tcp_seq icmp_tcp_seq;
2846 	int mtu;
2847 
2848 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2849 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2850 		return;
2851 
2852 	if (cmd == PRC_MSGSIZE)
2853 		notify = tcp_mtudisc_notify;
2854 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2855 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2856 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2857 		notify = tcp_drop_syn_sent;
2858 
2859 	/*
2860 	 * Hostdead is ugly because it goes linearly through all PCBs.
2861 	 * XXX: We never get this from ICMP, otherwise it makes an
2862 	 * excellent DoS attack on machines with many connections.
2863 	 */
2864 	else if (cmd == PRC_HOSTDEAD)
2865 		ip = NULL;
2866 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2867 		return;
2868 
2869 	if (ip == NULL) {
2870 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2871 		return;
2872 	}
2873 
2874 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2875 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2876 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2877 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2878 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2879 		/* signal EHOSTDOWN, as it flushes the cached route */
2880 		inp = (*notify)(inp, EHOSTDOWN);
2881 		goto out;
2882 	}
2883 	icmp_tcp_seq = th->th_seq;
2884 	if (inp != NULL)  {
2885 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2886 		    !(inp->inp_flags & INP_DROPPED) &&
2887 		    !(inp->inp_socket == NULL)) {
2888 			tp = intotcpcb(inp);
2889 #ifdef TCP_OFFLOAD
2890 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2891 				/*
2892 				 * MTU discovery for offloaded connections.  Let
2893 				 * the TOE driver verify seq# and process it.
2894 				 */
2895 				mtu = tcp_next_pmtu(icp, ip);
2896 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2897 				goto out;
2898 			}
2899 #endif
2900 			if (tp->t_port != port) {
2901 				goto out;
2902 			}
2903 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2904 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2905 				if (cmd == PRC_MSGSIZE) {
2906 					/*
2907 					 * MTU discovery: we got a needfrag and
2908 					 * will potentially try a lower MTU.
2909 					 */
2910 					mtu = tcp_next_pmtu(icp, ip);
2911 
2912 					/*
2913 					 * Only process the offered MTU if it
2914 					 * is smaller than the current one.
2915 					 */
2916 					if (mtu < tp->t_maxseg +
2917 					    sizeof(struct tcpiphdr)) {
2918 						bzero(&inc, sizeof(inc));
2919 						inc.inc_faddr = faddr;
2920 						inc.inc_fibnum =
2921 						    inp->inp_inc.inc_fibnum;
2922 						tcp_hc_updatemtu(&inc, mtu);
2923 						inp = tcp_mtudisc(inp, mtu);
2924 					}
2925 				} else
2926 					inp = (*notify)(inp,
2927 					    inetctlerrmap[cmd]);
2928 			}
2929 		}
2930 	} else {
2931 		bzero(&inc, sizeof(inc));
2932 		inc.inc_fport = th->th_dport;
2933 		inc.inc_lport = th->th_sport;
2934 		inc.inc_faddr = faddr;
2935 		inc.inc_laddr = ip->ip_src;
2936 		syncache_unreach(&inc, icmp_tcp_seq, port);
2937 	}
2938 out:
2939 	if (inp != NULL)
2940 		INP_WUNLOCK(inp);
2941 }
2942 
2943 void
2944 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2945 {
2946 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2947 }
2948 
2949 void
2950 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2951 {
2952 	/* Its a tunneled TCP over UDP icmp */
2953 	struct ip *outer_ip, *inner_ip;
2954 	struct icmp *icmp;
2955 	struct udphdr *udp;
2956 	struct tcphdr *th, ttemp;
2957 	int i_hlen, o_len;
2958 	uint16_t port;
2959 
2960 	inner_ip = (struct ip *)vip;
2961 	icmp = (struct icmp *)((caddr_t)inner_ip -
2962 	    (sizeof(struct icmp) - sizeof(struct ip)));
2963 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2964 	i_hlen = inner_ip->ip_hl << 2;
2965 	o_len = ntohs(outer_ip->ip_len);
2966 	if (o_len <
2967 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2968 		/* Not enough data present */
2969 		return;
2970 	}
2971 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2972 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2973 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2974 		return;
2975 	}
2976 	port = udp->uh_dport;
2977 	th = (struct tcphdr *)(udp + 1);
2978 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2979 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2980 	/* Now adjust down the size of the outer IP header */
2981 	o_len -= sizeof(struct udphdr);
2982 	outer_ip->ip_len = htons(o_len);
2983 	/* Now call in to the normal handling code */
2984 	tcp_ctlinput_with_port(cmd, sa, vip, port);
2985 }
2986 #endif /* INET */
2987 
2988 #ifdef INET6
2989 static inline int
2990 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2991 {
2992 	int mtu = ntohl(icmp6->icmp6_mtu);
2993 
2994 	/*
2995 	 * If no alternative MTU was proposed, or the proposed MTU was too
2996 	 * small, set to the min.
2997 	 */
2998 	if (mtu < IPV6_MMTU)
2999 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
3000 	return (mtu);
3001 }
3002 
3003 static void
3004 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3005 {
3006 	struct in6_addr *dst;
3007 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3008 	struct ip6_hdr *ip6;
3009 	struct mbuf *m;
3010 	struct inpcb *inp;
3011 	struct tcpcb *tp;
3012 	struct icmp6_hdr *icmp6;
3013 	struct ip6ctlparam *ip6cp = NULL;
3014 	const struct sockaddr_in6 *sa6_src = NULL;
3015 	struct in_conninfo inc;
3016 	struct tcp_ports {
3017 		uint16_t th_sport;
3018 		uint16_t th_dport;
3019 	} t_ports;
3020 	tcp_seq icmp_tcp_seq;
3021 	unsigned int mtu;
3022 	unsigned int off;
3023 
3024 	if (sa->sa_family != AF_INET6 ||
3025 	    sa->sa_len != sizeof(struct sockaddr_in6))
3026 		return;
3027 
3028 	/* if the parameter is from icmp6, decode it. */
3029 	if (d != NULL) {
3030 		ip6cp = (struct ip6ctlparam *)d;
3031 		icmp6 = ip6cp->ip6c_icmp6;
3032 		m = ip6cp->ip6c_m;
3033 		ip6 = ip6cp->ip6c_ip6;
3034 		off = ip6cp->ip6c_off;
3035 		sa6_src = ip6cp->ip6c_src;
3036 		dst = ip6cp->ip6c_finaldst;
3037 	} else {
3038 		m = NULL;
3039 		ip6 = NULL;
3040 		off = 0;	/* fool gcc */
3041 		sa6_src = &sa6_any;
3042 		dst = NULL;
3043 	}
3044 
3045 	if (cmd == PRC_MSGSIZE)
3046 		notify = tcp_mtudisc_notify;
3047 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3048 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3049 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3050 		notify = tcp_drop_syn_sent;
3051 
3052 	/*
3053 	 * Hostdead is ugly because it goes linearly through all PCBs.
3054 	 * XXX: We never get this from ICMP, otherwise it makes an
3055 	 * excellent DoS attack on machines with many connections.
3056 	 */
3057 	else if (cmd == PRC_HOSTDEAD)
3058 		ip6 = NULL;
3059 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3060 		return;
3061 
3062 	if (ip6 == NULL) {
3063 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3064 			      (const struct sockaddr *)sa6_src,
3065 			      0, cmd, NULL, notify);
3066 		return;
3067 	}
3068 
3069 	/* Check if we can safely get the ports from the tcp hdr */
3070 	if (m == NULL ||
3071 	    (m->m_pkthdr.len <
3072 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3073 		return;
3074 	}
3075 	bzero(&t_ports, sizeof(struct tcp_ports));
3076 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3077 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3078 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3079 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3080 		/* signal EHOSTDOWN, as it flushes the cached route */
3081 		inp = (*notify)(inp, EHOSTDOWN);
3082 		goto out;
3083 	}
3084 	off += sizeof(struct tcp_ports);
3085 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3086 		goto out;
3087 	}
3088 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3089 	if (inp != NULL)  {
3090 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3091 		    !(inp->inp_flags & INP_DROPPED) &&
3092 		    !(inp->inp_socket == NULL)) {
3093 			tp = intotcpcb(inp);
3094 #ifdef TCP_OFFLOAD
3095 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3096 				/* MTU discovery for offloaded connections. */
3097 				mtu = tcp6_next_pmtu(icmp6);
3098 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3099 				goto out;
3100 			}
3101 #endif
3102 			if (tp->t_port != port) {
3103 				goto out;
3104 			}
3105 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3106 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3107 				if (cmd == PRC_MSGSIZE) {
3108 					/*
3109 					 * MTU discovery:
3110 					 * If we got a needfrag set the MTU
3111 					 * in the route to the suggested new
3112 					 * value (if given) and then notify.
3113 					 */
3114 					mtu = tcp6_next_pmtu(icmp6);
3115 
3116 					bzero(&inc, sizeof(inc));
3117 					inc.inc_fibnum = M_GETFIB(m);
3118 					inc.inc_flags |= INC_ISIPV6;
3119 					inc.inc6_faddr = *dst;
3120 					if (in6_setscope(&inc.inc6_faddr,
3121 						m->m_pkthdr.rcvif, NULL))
3122 						goto out;
3123 					/*
3124 					 * Only process the offered MTU if it
3125 					 * is smaller than the current one.
3126 					 */
3127 					if (mtu < tp->t_maxseg +
3128 					    sizeof (struct tcphdr) +
3129 					    sizeof (struct ip6_hdr)) {
3130 						tcp_hc_updatemtu(&inc, mtu);
3131 						tcp_mtudisc(inp, mtu);
3132 						ICMP6STAT_INC(icp6s_pmtuchg);
3133 					}
3134 				} else
3135 					inp = (*notify)(inp,
3136 					    inet6ctlerrmap[cmd]);
3137 			}
3138 		}
3139 	} else {
3140 		bzero(&inc, sizeof(inc));
3141 		inc.inc_fibnum = M_GETFIB(m);
3142 		inc.inc_flags |= INC_ISIPV6;
3143 		inc.inc_fport = t_ports.th_dport;
3144 		inc.inc_lport = t_ports.th_sport;
3145 		inc.inc6_faddr = *dst;
3146 		inc.inc6_laddr = ip6->ip6_src;
3147 		syncache_unreach(&inc, icmp_tcp_seq, port);
3148 	}
3149 out:
3150 	if (inp != NULL)
3151 		INP_WUNLOCK(inp);
3152 }
3153 
3154 void
3155 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3156 {
3157 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3158 }
3159 
3160 void
3161 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3162 {
3163 	struct ip6ctlparam *ip6cp;
3164 	struct mbuf *m;
3165 	struct udphdr *udp;
3166 	uint16_t port;
3167 
3168 	ip6cp = (struct ip6ctlparam *)d;
3169 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3170 	if (m == NULL) {
3171 		return;
3172 	}
3173 	udp = mtod(m, struct udphdr *);
3174 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3175 		return;
3176 	}
3177 	port = udp->uh_dport;
3178 	m_adj(m, sizeof(struct udphdr));
3179 	if ((m->m_flags & M_PKTHDR) == 0) {
3180 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3181 	}
3182 	/* Now call in to the normal handling code */
3183 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3184 }
3185 
3186 #endif /* INET6 */
3187 
3188 static uint32_t
3189 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3190 {
3191 	SIPHASH_CTX ctx;
3192 	uint32_t hash[2];
3193 
3194 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3195 	    ("%s: keylen %u too short ", __func__, len));
3196 	SipHash24_Init(&ctx);
3197 	SipHash_SetKey(&ctx, (uint8_t *)key);
3198 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3199 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3200 	switch (inc->inc_flags & INC_ISIPV6) {
3201 #ifdef INET
3202 	case 0:
3203 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3204 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3205 		break;
3206 #endif
3207 #ifdef INET6
3208 	case INC_ISIPV6:
3209 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3210 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3211 		break;
3212 #endif
3213 	}
3214 	SipHash_Final((uint8_t *)hash, &ctx);
3215 
3216 	return (hash[0] ^ hash[1]);
3217 }
3218 
3219 uint32_t
3220 tcp_new_ts_offset(struct in_conninfo *inc)
3221 {
3222 	struct in_conninfo inc_store, *local_inc;
3223 
3224 	if (!V_tcp_ts_offset_per_conn) {
3225 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3226 		inc_store.inc_lport = 0;
3227 		inc_store.inc_fport = 0;
3228 		local_inc = &inc_store;
3229 	} else {
3230 		local_inc = inc;
3231 	}
3232 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3233 	    sizeof(V_ts_offset_secret)));
3234 }
3235 
3236 /*
3237  * Following is where TCP initial sequence number generation occurs.
3238  *
3239  * There are two places where we must use initial sequence numbers:
3240  * 1.  In SYN-ACK packets.
3241  * 2.  In SYN packets.
3242  *
3243  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3244  * tcp_syncache.c for details.
3245  *
3246  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3247  * depends on this property.  In addition, these ISNs should be
3248  * unguessable so as to prevent connection hijacking.  To satisfy
3249  * the requirements of this situation, the algorithm outlined in
3250  * RFC 1948 is used, with only small modifications.
3251  *
3252  * Implementation details:
3253  *
3254  * Time is based off the system timer, and is corrected so that it
3255  * increases by one megabyte per second.  This allows for proper
3256  * recycling on high speed LANs while still leaving over an hour
3257  * before rollover.
3258  *
3259  * As reading the *exact* system time is too expensive to be done
3260  * whenever setting up a TCP connection, we increment the time
3261  * offset in two ways.  First, a small random positive increment
3262  * is added to isn_offset for each connection that is set up.
3263  * Second, the function tcp_isn_tick fires once per clock tick
3264  * and increments isn_offset as necessary so that sequence numbers
3265  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3266  * random positive increments serve only to ensure that the same
3267  * exact sequence number is never sent out twice (as could otherwise
3268  * happen when a port is recycled in less than the system tick
3269  * interval.)
3270  *
3271  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3272  * between seeding of isn_secret.  This is normally set to zero,
3273  * as reseeding should not be necessary.
3274  *
3275  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3276  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3277  * general, this means holding an exclusive (write) lock.
3278  */
3279 
3280 #define ISN_BYTES_PER_SECOND 1048576
3281 #define ISN_STATIC_INCREMENT 4096
3282 #define ISN_RANDOM_INCREMENT (4096 - 1)
3283 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3284 
3285 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3286 VNET_DEFINE_STATIC(int, isn_last);
3287 VNET_DEFINE_STATIC(int, isn_last_reseed);
3288 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3289 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3290 
3291 #define	V_isn_secret			VNET(isn_secret)
3292 #define	V_isn_last			VNET(isn_last)
3293 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3294 #define	V_isn_offset			VNET(isn_offset)
3295 #define	V_isn_offset_old		VNET(isn_offset_old)
3296 
3297 tcp_seq
3298 tcp_new_isn(struct in_conninfo *inc)
3299 {
3300 	tcp_seq new_isn;
3301 	u_int32_t projected_offset;
3302 
3303 	ISN_LOCK();
3304 	/* Seed if this is the first use, reseed if requested. */
3305 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3306 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3307 		< (u_int)ticks))) {
3308 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3309 		V_isn_last_reseed = ticks;
3310 	}
3311 
3312 	/* Compute the hash and return the ISN. */
3313 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3314 	    sizeof(V_isn_secret));
3315 	V_isn_offset += ISN_STATIC_INCREMENT +
3316 		(arc4random() & ISN_RANDOM_INCREMENT);
3317 	if (ticks != V_isn_last) {
3318 		projected_offset = V_isn_offset_old +
3319 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3320 		if (SEQ_GT(projected_offset, V_isn_offset))
3321 			V_isn_offset = projected_offset;
3322 		V_isn_offset_old = V_isn_offset;
3323 		V_isn_last = ticks;
3324 	}
3325 	new_isn += V_isn_offset;
3326 	ISN_UNLOCK();
3327 	return (new_isn);
3328 }
3329 
3330 /*
3331  * When a specific ICMP unreachable message is received and the
3332  * connection state is SYN-SENT, drop the connection.  This behavior
3333  * is controlled by the icmp_may_rst sysctl.
3334  */
3335 struct inpcb *
3336 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3337 {
3338 	struct tcpcb *tp;
3339 
3340 	NET_EPOCH_ASSERT();
3341 	INP_WLOCK_ASSERT(inp);
3342 
3343 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3344 	    (inp->inp_flags & INP_DROPPED))
3345 		return (inp);
3346 
3347 	tp = intotcpcb(inp);
3348 	if (tp->t_state != TCPS_SYN_SENT)
3349 		return (inp);
3350 
3351 	if (IS_FASTOPEN(tp->t_flags))
3352 		tcp_fastopen_disable_path(tp);
3353 
3354 	tp = tcp_drop(tp, errno);
3355 	if (tp != NULL)
3356 		return (inp);
3357 	else
3358 		return (NULL);
3359 }
3360 
3361 /*
3362  * When `need fragmentation' ICMP is received, update our idea of the MSS
3363  * based on the new value. Also nudge TCP to send something, since we
3364  * know the packet we just sent was dropped.
3365  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3366  */
3367 static struct inpcb *
3368 tcp_mtudisc_notify(struct inpcb *inp, int error)
3369 {
3370 
3371 	return (tcp_mtudisc(inp, -1));
3372 }
3373 
3374 static struct inpcb *
3375 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3376 {
3377 	struct tcpcb *tp;
3378 	struct socket *so;
3379 
3380 	INP_WLOCK_ASSERT(inp);
3381 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3382 	    (inp->inp_flags & INP_DROPPED))
3383 		return (inp);
3384 
3385 	tp = intotcpcb(inp);
3386 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3387 
3388 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3389 
3390 	so = inp->inp_socket;
3391 	SOCKBUF_LOCK(&so->so_snd);
3392 	/* If the mss is larger than the socket buffer, decrease the mss. */
3393 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3394 		tp->t_maxseg = so->so_snd.sb_hiwat;
3395 	SOCKBUF_UNLOCK(&so->so_snd);
3396 
3397 	TCPSTAT_INC(tcps_mturesent);
3398 	tp->t_rtttime = 0;
3399 	tp->snd_nxt = tp->snd_una;
3400 	tcp_free_sackholes(tp);
3401 	tp->snd_recover = tp->snd_max;
3402 	if (tp->t_flags & TF_SACK_PERMIT)
3403 		EXIT_FASTRECOVERY(tp->t_flags);
3404 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3405 		/*
3406 		 * Conceptually the snd_nxt setting
3407 		 * and freeing sack holes should
3408 		 * be done by the default stacks
3409 		 * own tfb_tcp_mtu_chg().
3410 		 */
3411 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3412 	}
3413 	if (tcp_output(tp) < 0)
3414 		return (NULL);
3415 	else
3416 		return (inp);
3417 }
3418 
3419 #ifdef INET
3420 /*
3421  * Look-up the routing entry to the peer of this inpcb.  If no route
3422  * is found and it cannot be allocated, then return 0.  This routine
3423  * is called by TCP routines that access the rmx structure and by
3424  * tcp_mss_update to get the peer/interface MTU.
3425  */
3426 uint32_t
3427 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3428 {
3429 	struct nhop_object *nh;
3430 	struct ifnet *ifp;
3431 	uint32_t maxmtu = 0;
3432 
3433 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3434 
3435 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3436 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3437 		if (nh == NULL)
3438 			return (0);
3439 
3440 		ifp = nh->nh_ifp;
3441 		maxmtu = nh->nh_mtu;
3442 
3443 		/* Report additional interface capabilities. */
3444 		if (cap != NULL) {
3445 			if (ifp->if_capenable & IFCAP_TSO4 &&
3446 			    ifp->if_hwassist & CSUM_TSO) {
3447 				cap->ifcap |= CSUM_TSO;
3448 				cap->tsomax = ifp->if_hw_tsomax;
3449 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3450 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3451 			}
3452 		}
3453 	}
3454 	return (maxmtu);
3455 }
3456 #endif /* INET */
3457 
3458 #ifdef INET6
3459 uint32_t
3460 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3461 {
3462 	struct nhop_object *nh;
3463 	struct in6_addr dst6;
3464 	uint32_t scopeid;
3465 	struct ifnet *ifp;
3466 	uint32_t maxmtu = 0;
3467 
3468 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3469 
3470 	if (inc->inc_flags & INC_IPV6MINMTU)
3471 		return (IPV6_MMTU);
3472 
3473 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3474 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3475 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3476 		if (nh == NULL)
3477 			return (0);
3478 
3479 		ifp = nh->nh_ifp;
3480 		maxmtu = nh->nh_mtu;
3481 
3482 		/* Report additional interface capabilities. */
3483 		if (cap != NULL) {
3484 			if (ifp->if_capenable & IFCAP_TSO6 &&
3485 			    ifp->if_hwassist & CSUM_TSO) {
3486 				cap->ifcap |= CSUM_TSO;
3487 				cap->tsomax = ifp->if_hw_tsomax;
3488 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3489 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3490 			}
3491 		}
3492 	}
3493 
3494 	return (maxmtu);
3495 }
3496 
3497 /*
3498  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3499  *
3500  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3501  * The right place to do that is ip6_setpktopt() that has just been
3502  * executed.  By the way it just filled ip6po_minmtu for us.
3503  */
3504 void
3505 tcp6_use_min_mtu(struct tcpcb *tp)
3506 {
3507 	struct inpcb *inp = tp->t_inpcb;
3508 
3509 	INP_WLOCK_ASSERT(inp);
3510 	/*
3511 	 * In case of the IPV6_USE_MIN_MTU socket
3512 	 * option, the INC_IPV6MINMTU flag to announce
3513 	 * a corresponding MSS during the initial
3514 	 * handshake.  If the TCP connection is not in
3515 	 * the front states, just reduce the MSS being
3516 	 * used.  This avoids the sending of TCP
3517 	 * segments which will be fragmented at the
3518 	 * IPv6 layer.
3519 	 */
3520 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3521 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3522 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3523 		struct ip6_pktopts *opt;
3524 
3525 		opt = inp->in6p_outputopts;
3526 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3527 		    tp->t_maxseg > TCP6_MSS)
3528 			tp->t_maxseg = TCP6_MSS;
3529 	}
3530 }
3531 #endif /* INET6 */
3532 
3533 /*
3534  * Calculate effective SMSS per RFC5681 definition for a given TCP
3535  * connection at its current state, taking into account SACK and etc.
3536  */
3537 u_int
3538 tcp_maxseg(const struct tcpcb *tp)
3539 {
3540 	u_int optlen;
3541 
3542 	if (tp->t_flags & TF_NOOPT)
3543 		return (tp->t_maxseg);
3544 
3545 	/*
3546 	 * Here we have a simplified code from tcp_addoptions(),
3547 	 * without a proper loop, and having most of paddings hardcoded.
3548 	 * We might make mistakes with padding here in some edge cases,
3549 	 * but this is harmless, since result of tcp_maxseg() is used
3550 	 * only in cwnd and ssthresh estimations.
3551 	 */
3552 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3553 		if (tp->t_flags & TF_RCVD_TSTMP)
3554 			optlen = TCPOLEN_TSTAMP_APPA;
3555 		else
3556 			optlen = 0;
3557 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3558 		if (tp->t_flags & TF_SIGNATURE)
3559 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3560 #endif
3561 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3562 			optlen += TCPOLEN_SACKHDR;
3563 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3564 			optlen = PADTCPOLEN(optlen);
3565 		}
3566 	} else {
3567 		if (tp->t_flags & TF_REQ_TSTMP)
3568 			optlen = TCPOLEN_TSTAMP_APPA;
3569 		else
3570 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3571 		if (tp->t_flags & TF_REQ_SCALE)
3572 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3573 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3574 		if (tp->t_flags & TF_SIGNATURE)
3575 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3576 #endif
3577 		if (tp->t_flags & TF_SACK_PERMIT)
3578 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3579 	}
3580 #undef PAD
3581 	optlen = min(optlen, TCP_MAXOLEN);
3582 	return (tp->t_maxseg - optlen);
3583 }
3584 
3585 
3586 u_int
3587 tcp_fixed_maxseg(const struct tcpcb *tp)
3588 {
3589 	int optlen;
3590 
3591 	if (tp->t_flags & TF_NOOPT)
3592 		return (tp->t_maxseg);
3593 
3594 	/*
3595 	 * Here we have a simplified code from tcp_addoptions(),
3596 	 * without a proper loop, and having most of paddings hardcoded.
3597 	 * We only consider fixed options that we would send every
3598 	 * time I.e. SACK is not considered. This is important
3599 	 * for cc modules to figure out what the modulo of the
3600 	 * cwnd should be.
3601 	 */
3602 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3603 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3604 		if (tp->t_flags & TF_RCVD_TSTMP)
3605 			optlen = TCPOLEN_TSTAMP_APPA;
3606 		else
3607 			optlen = 0;
3608 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3609 		if (tp->t_flags & TF_SIGNATURE)
3610 			optlen += PAD(TCPOLEN_SIGNATURE);
3611 #endif
3612 	} else {
3613 		if (tp->t_flags & TF_REQ_TSTMP)
3614 			optlen = TCPOLEN_TSTAMP_APPA;
3615 		else
3616 			optlen = PAD(TCPOLEN_MAXSEG);
3617 		if (tp->t_flags & TF_REQ_SCALE)
3618 			optlen += PAD(TCPOLEN_WINDOW);
3619 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3620 		if (tp->t_flags & TF_SIGNATURE)
3621 			optlen += PAD(TCPOLEN_SIGNATURE);
3622 #endif
3623 		if (tp->t_flags & TF_SACK_PERMIT)
3624 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3625 	}
3626 #undef PAD
3627 	optlen = min(optlen, TCP_MAXOLEN);
3628 	return (tp->t_maxseg - optlen);
3629 }
3630 
3631 
3632 
3633 static int
3634 sysctl_drop(SYSCTL_HANDLER_ARGS)
3635 {
3636 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3637 	struct sockaddr_storage addrs[2];
3638 	struct inpcb *inp;
3639 	struct tcpcb *tp;
3640 	struct tcptw *tw;
3641 	struct sockaddr_in *fin, *lin;
3642 	struct epoch_tracker et;
3643 #ifdef INET6
3644 	struct sockaddr_in6 *fin6, *lin6;
3645 #endif
3646 	int error;
3647 
3648 	inp = NULL;
3649 	fin = lin = 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 			fin = (struct sockaddr_in *)&addrs[0];
3679 			lin = (struct sockaddr_in *)&addrs[1];
3680 			break;
3681 		}
3682 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3683 		if (error)
3684 			return (error);
3685 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3686 		if (error)
3687 			return (error);
3688 		break;
3689 #endif
3690 #ifdef INET
3691 	case AF_INET:
3692 		fin = (struct sockaddr_in *)&addrs[0];
3693 		lin = (struct sockaddr_in *)&addrs[1];
3694 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3695 		    lin->sin_len != sizeof(struct sockaddr_in))
3696 			return (EINVAL);
3697 		break;
3698 #endif
3699 	default:
3700 		return (EINVAL);
3701 	}
3702 	NET_EPOCH_ENTER(et);
3703 	switch (addrs[0].ss_family) {
3704 #ifdef INET6
3705 	case AF_INET6:
3706 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3707 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3708 		    INPLOOKUP_WLOCKPCB, NULL);
3709 		break;
3710 #endif
3711 #ifdef INET
3712 	case AF_INET:
3713 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3714 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3715 		break;
3716 #endif
3717 	}
3718 	if (inp != NULL) {
3719 		if (inp->inp_flags & INP_TIMEWAIT) {
3720 			/*
3721 			 * XXXRW: There currently exists a state where an
3722 			 * inpcb is present, but its timewait state has been
3723 			 * discarded.  For now, don't allow dropping of this
3724 			 * type of inpcb.
3725 			 */
3726 			tw = intotw(inp);
3727 			if (tw != NULL)
3728 				tcp_twclose(tw, 0);
3729 			else
3730 				INP_WUNLOCK(inp);
3731 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3732 		    !SOLISTENING(inp->inp_socket)) {
3733 			tp = intotcpcb(inp);
3734 			tp = tcp_drop(tp, ECONNABORTED);
3735 			if (tp != NULL)
3736 				INP_WUNLOCK(inp);
3737 		} else
3738 			INP_WUNLOCK(inp);
3739 	} else
3740 		error = ESRCH;
3741 	NET_EPOCH_EXIT(et);
3742 	return (error);
3743 }
3744 
3745 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3746     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3747     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3748     "Drop TCP connection");
3749 
3750 static int
3751 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3752 {
3753 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3754 	    &tcp_ctloutput_set));
3755 }
3756 
3757 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3758     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3759     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3760     "Set socket option for TCP endpoint");
3761 
3762 #ifdef KERN_TLS
3763 static int
3764 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3765 {
3766 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3767 	struct sockaddr_storage addrs[2];
3768 	struct inpcb *inp;
3769 	struct sockaddr_in *fin, *lin;
3770 	struct epoch_tracker et;
3771 #ifdef INET6
3772 	struct sockaddr_in6 *fin6, *lin6;
3773 #endif
3774 	int error;
3775 
3776 	inp = NULL;
3777 	fin = lin = NULL;
3778 #ifdef INET6
3779 	fin6 = lin6 = NULL;
3780 #endif
3781 	error = 0;
3782 
3783 	if (req->oldptr != NULL || req->oldlen != 0)
3784 		return (EINVAL);
3785 	if (req->newptr == NULL)
3786 		return (EPERM);
3787 	if (req->newlen < sizeof(addrs))
3788 		return (ENOMEM);
3789 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3790 	if (error)
3791 		return (error);
3792 
3793 	switch (addrs[0].ss_family) {
3794 #ifdef INET6
3795 	case AF_INET6:
3796 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3797 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3798 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3799 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3800 			return (EINVAL);
3801 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3802 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3803 				return (EINVAL);
3804 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3805 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3806 			fin = (struct sockaddr_in *)&addrs[0];
3807 			lin = (struct sockaddr_in *)&addrs[1];
3808 			break;
3809 		}
3810 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3811 		if (error)
3812 			return (error);
3813 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3814 		if (error)
3815 			return (error);
3816 		break;
3817 #endif
3818 #ifdef INET
3819 	case AF_INET:
3820 		fin = (struct sockaddr_in *)&addrs[0];
3821 		lin = (struct sockaddr_in *)&addrs[1];
3822 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3823 		    lin->sin_len != sizeof(struct sockaddr_in))
3824 			return (EINVAL);
3825 		break;
3826 #endif
3827 	default:
3828 		return (EINVAL);
3829 	}
3830 	NET_EPOCH_ENTER(et);
3831 	switch (addrs[0].ss_family) {
3832 #ifdef INET6
3833 	case AF_INET6:
3834 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3835 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3836 		    INPLOOKUP_WLOCKPCB, NULL);
3837 		break;
3838 #endif
3839 #ifdef INET
3840 	case AF_INET:
3841 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3842 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3843 		break;
3844 #endif
3845 	}
3846 	NET_EPOCH_EXIT(et);
3847 	if (inp != NULL) {
3848 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3849 		    inp->inp_socket == NULL) {
3850 			error = ECONNRESET;
3851 			INP_WUNLOCK(inp);
3852 		} else {
3853 			struct socket *so;
3854 
3855 			so = inp->inp_socket;
3856 			soref(so);
3857 			error = ktls_set_tx_mode(so,
3858 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3859 			INP_WUNLOCK(inp);
3860 			sorele(so);
3861 		}
3862 	} else
3863 		error = ESRCH;
3864 	return (error);
3865 }
3866 
3867 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3868     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3869     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3870     "Switch TCP connection to SW TLS");
3871 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3872     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3873     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3874     "Switch TCP connection to ifnet TLS");
3875 #endif
3876 
3877 /*
3878  * Generate a standardized TCP log line for use throughout the
3879  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3880  * allow use in the interrupt context.
3881  *
3882  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3883  * NB: The function may return NULL if memory allocation failed.
3884  *
3885  * Due to header inclusion and ordering limitations the struct ip
3886  * and ip6_hdr pointers have to be passed as void pointers.
3887  */
3888 char *
3889 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3890     const void *ip6hdr)
3891 {
3892 
3893 	/* Is logging enabled? */
3894 	if (V_tcp_log_in_vain == 0)
3895 		return (NULL);
3896 
3897 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3898 }
3899 
3900 char *
3901 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3902     const void *ip6hdr)
3903 {
3904 
3905 	/* Is logging enabled? */
3906 	if (tcp_log_debug == 0)
3907 		return (NULL);
3908 
3909 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3910 }
3911 
3912 static char *
3913 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3914     const void *ip6hdr)
3915 {
3916 	char *s, *sp;
3917 	size_t size;
3918 	struct ip *ip;
3919 #ifdef INET6
3920 	const struct ip6_hdr *ip6;
3921 
3922 	ip6 = (const struct ip6_hdr *)ip6hdr;
3923 #endif /* INET6 */
3924 	ip = (struct ip *)ip4hdr;
3925 
3926 	/*
3927 	 * The log line looks like this:
3928 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3929 	 */
3930 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3931 	    sizeof(PRINT_TH_FLAGS) + 1 +
3932 #ifdef INET6
3933 	    2 * INET6_ADDRSTRLEN;
3934 #else
3935 	    2 * INET_ADDRSTRLEN;
3936 #endif /* INET6 */
3937 
3938 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3939 	if (s == NULL)
3940 		return (NULL);
3941 
3942 	strcat(s, "TCP: [");
3943 	sp = s + strlen(s);
3944 
3945 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3946 		inet_ntoa_r(inc->inc_faddr, sp);
3947 		sp = s + strlen(s);
3948 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3949 		sp = s + strlen(s);
3950 		inet_ntoa_r(inc->inc_laddr, sp);
3951 		sp = s + strlen(s);
3952 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3953 #ifdef INET6
3954 	} else if (inc) {
3955 		ip6_sprintf(sp, &inc->inc6_faddr);
3956 		sp = s + strlen(s);
3957 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3958 		sp = s + strlen(s);
3959 		ip6_sprintf(sp, &inc->inc6_laddr);
3960 		sp = s + strlen(s);
3961 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3962 	} else if (ip6 && th) {
3963 		ip6_sprintf(sp, &ip6->ip6_src);
3964 		sp = s + strlen(s);
3965 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3966 		sp = s + strlen(s);
3967 		ip6_sprintf(sp, &ip6->ip6_dst);
3968 		sp = s + strlen(s);
3969 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3970 #endif /* INET6 */
3971 #ifdef INET
3972 	} else if (ip && th) {
3973 		inet_ntoa_r(ip->ip_src, sp);
3974 		sp = s + strlen(s);
3975 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3976 		sp = s + strlen(s);
3977 		inet_ntoa_r(ip->ip_dst, sp);
3978 		sp = s + strlen(s);
3979 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3980 #endif /* INET */
3981 	} else {
3982 		free(s, M_TCPLOG);
3983 		return (NULL);
3984 	}
3985 	sp = s + strlen(s);
3986 	if (th)
3987 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3988 	if (*(s + size - 1) != '\0')
3989 		panic("%s: string too long", __func__);
3990 	return (s);
3991 }
3992 
3993 /*
3994  * A subroutine which makes it easy to track TCP state changes with DTrace.
3995  * This function shouldn't be called for t_state initializations that don't
3996  * correspond to actual TCP state transitions.
3997  */
3998 void
3999 tcp_state_change(struct tcpcb *tp, int newstate)
4000 {
4001 #if defined(KDTRACE_HOOKS)
4002 	int pstate = tp->t_state;
4003 #endif
4004 
4005 	TCPSTATES_DEC(tp->t_state);
4006 	TCPSTATES_INC(newstate);
4007 	tp->t_state = newstate;
4008 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4009 }
4010 
4011 /*
4012  * Create an external-format (``xtcpcb'') structure using the information in
4013  * the kernel-format tcpcb structure pointed to by tp.  This is done to
4014  * reduce the spew of irrelevant information over this interface, to isolate
4015  * user code from changes in the kernel structure, and potentially to provide
4016  * information-hiding if we decide that some of this information should be
4017  * hidden from users.
4018  */
4019 void
4020 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4021 {
4022 	struct tcpcb *tp = intotcpcb(inp);
4023 	struct tcptw *tw = intotw(inp);
4024 	sbintime_t now;
4025 
4026 	bzero(xt, sizeof(*xt));
4027 	if (inp->inp_flags & INP_TIMEWAIT) {
4028 		xt->t_state = TCPS_TIME_WAIT;
4029 		xt->xt_encaps_port = tw->t_port;
4030 	} else {
4031 		xt->t_state = tp->t_state;
4032 		xt->t_logstate = tp->t_logstate;
4033 		xt->t_flags = tp->t_flags;
4034 		xt->t_sndzerowin = tp->t_sndzerowin;
4035 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4036 		xt->t_rcvoopack = tp->t_rcvoopack;
4037 		xt->t_rcv_wnd = tp->rcv_wnd;
4038 		xt->t_snd_wnd = tp->snd_wnd;
4039 		xt->t_snd_cwnd = tp->snd_cwnd;
4040 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4041 		xt->t_dsack_bytes = tp->t_dsack_bytes;
4042 		xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4043 		xt->t_dsack_pack = tp->t_dsack_pack;
4044 		xt->t_maxseg = tp->t_maxseg;
4045 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4046 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4047 
4048 		now = getsbinuptime();
4049 #define	COPYTIMER(ttt)	do {						\
4050 		if (callout_active(&tp->t_timers->ttt))			\
4051 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4052 			    SBT_1MS;					\
4053 		else							\
4054 			xt->ttt = 0;					\
4055 } while (0)
4056 		COPYTIMER(tt_delack);
4057 		COPYTIMER(tt_rexmt);
4058 		COPYTIMER(tt_persist);
4059 		COPYTIMER(tt_keep);
4060 		COPYTIMER(tt_2msl);
4061 #undef COPYTIMER
4062 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4063 
4064 		xt->xt_encaps_port = tp->t_port;
4065 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4066 		    TCP_FUNCTION_NAME_LEN_MAX);
4067 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4068 		    TCP_CA_NAME_MAX);
4069 #ifdef TCP_BLACKBOX
4070 		(void)tcp_log_get_id(tp, xt->xt_logid);
4071 #endif
4072 	}
4073 
4074 	xt->xt_len = sizeof(struct xtcpcb);
4075 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4076 	if (inp->inp_socket == NULL)
4077 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4078 }
4079 
4080 void
4081 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4082 {
4083 	uint32_t bit, i;
4084 
4085 	if ((tp == NULL) ||
4086 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4087 	    (status == 0)) {
4088 		/* Invalid */
4089 		return;
4090 	}
4091 	if (status > (sizeof(uint32_t) * 8)) {
4092 		/* Should this be a KASSERT? */
4093 		return;
4094 	}
4095 	bit = 1U << (status - 1);
4096 	if (bit & tp->t_end_info_status) {
4097 		/* already logged */
4098 		return;
4099 	}
4100 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4101 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4102 			tp->t_end_info_bytes[i] = status;
4103 			tp->t_end_info_status |= bit;
4104 			break;
4105 		}
4106 	}
4107 }
4108 
4109 int
4110 tcp_can_enable_pacing(void)
4111 {
4112 
4113 	if ((tcp_pacing_limit == -1) ||
4114 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4115 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4116 		shadow_num_connections = number_of_tcp_connections_pacing;
4117 		return (1);
4118 	} else {
4119 		return (0);
4120 	}
4121 }
4122 
4123 static uint8_t tcp_pacing_warning = 0;
4124 
4125 void
4126 tcp_decrement_paced_conn(void)
4127 {
4128 	uint32_t ret;
4129 
4130 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4131 	shadow_num_connections = number_of_tcp_connections_pacing;
4132 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4133 	if (ret == 0) {
4134 		if (tcp_pacing_limit != -1) {
4135 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4136 			tcp_pacing_limit = 0;
4137 		} else if (tcp_pacing_warning == 0) {
4138 			printf("Warning pacing count is invalid, invalid decrement\n");
4139 			tcp_pacing_warning = 1;
4140 		}
4141 	}
4142 }
4143