xref: /freebsd/sys/netinet/tcp_subr.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
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 		    const 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 bool
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 (true);
663 out:
664 	m_freem(m);
665 
666 	return (true);
667 }
668 
669 static int
670 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
671 {
672 	int error=ENOENT;
673 	struct tcp_function_set fs;
674 	struct tcp_function_block *blk;
675 
676 	memset(&fs, 0, sizeof(fs));
677 	rw_rlock(&tcp_function_lock);
678 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
679 	if (blk) {
680 		/* Found him */
681 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
682 		fs.pcbcnt = blk->tfb_refcnt;
683 	}
684 	rw_runlock(&tcp_function_lock);
685 	error = sysctl_handle_string(oidp, fs.function_set_name,
686 				     sizeof(fs.function_set_name), req);
687 
688 	/* Check for error or no change */
689 	if (error != 0 || req->newptr == NULL)
690 		return(error);
691 
692 	rw_wlock(&tcp_function_lock);
693 	blk = find_tcp_functions_locked(&fs);
694 	if ((blk == NULL) ||
695 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
696 		error = ENOENT;
697 		goto done;
698 	}
699 	tcp_func_set_ptr = blk;
700 done:
701 	rw_wunlock(&tcp_function_lock);
702 	return (error);
703 }
704 
705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
706     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
707     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
708     "Set/get the default TCP functions");
709 
710 static int
711 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
712 {
713 	int error, cnt, linesz;
714 	struct tcp_function *f;
715 	char *buffer, *cp;
716 	size_t bufsz, outsz;
717 	bool alias;
718 
719 	cnt = 0;
720 	rw_rlock(&tcp_function_lock);
721 	TAILQ_FOREACH(f, &t_functions, tf_next) {
722 		cnt++;
723 	}
724 	rw_runlock(&tcp_function_lock);
725 
726 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
727 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
728 
729 	error = 0;
730 	cp = buffer;
731 
732 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
733 	    "Alias", "PCB count");
734 	cp += linesz;
735 	bufsz -= linesz;
736 	outsz = linesz;
737 
738 	rw_rlock(&tcp_function_lock);
739 	TAILQ_FOREACH(f, &t_functions, tf_next) {
740 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
741 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
742 		    f->tf_fb->tfb_tcp_block_name,
743 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
744 		    alias ? f->tf_name : "-",
745 		    f->tf_fb->tfb_refcnt);
746 		if (linesz >= bufsz) {
747 			error = EOVERFLOW;
748 			break;
749 		}
750 		cp += linesz;
751 		bufsz -= linesz;
752 		outsz += linesz;
753 	}
754 	rw_runlock(&tcp_function_lock);
755 	if (error == 0)
756 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
757 	free(buffer, M_TEMP);
758 	return (error);
759 }
760 
761 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
762     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
763     NULL, 0, sysctl_net_inet_list_available, "A",
764     "list available TCP Function sets");
765 
766 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
767 
768 #ifdef INET
769 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
770 #define	V_udp4_tun_socket	VNET(udp4_tun_socket)
771 #endif
772 #ifdef INET6
773 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
774 #define	V_udp6_tun_socket	VNET(udp6_tun_socket)
775 #endif
776 
777 static void
778 tcp_over_udp_stop(void)
779 {
780 	/*
781 	 * This function assumes sysctl caller holds inp_rinfo_lock()
782 	 * for writing!
783 	 */
784 #ifdef INET
785 	if (V_udp4_tun_socket != NULL) {
786 		soclose(V_udp4_tun_socket);
787 		V_udp4_tun_socket = NULL;
788 	}
789 #endif
790 #ifdef INET6
791 	if (V_udp6_tun_socket != NULL) {
792 		soclose(V_udp6_tun_socket);
793 		V_udp6_tun_socket = NULL;
794 	}
795 #endif
796 }
797 
798 static int
799 tcp_over_udp_start(void)
800 {
801 	uint16_t port;
802 	int ret;
803 #ifdef INET
804 	struct sockaddr_in sin;
805 #endif
806 #ifdef INET6
807 	struct sockaddr_in6 sin6;
808 #endif
809 	/*
810 	 * This function assumes sysctl caller holds inp_info_rlock()
811 	 * for writing!
812 	 */
813 	port = V_tcp_udp_tunneling_port;
814 	if (ntohs(port) == 0) {
815 		/* Must have a port set */
816 		return (EINVAL);
817 	}
818 #ifdef INET
819 	if (V_udp4_tun_socket != NULL) {
820 		/* Already running -- must stop first */
821 		return (EALREADY);
822 	}
823 #endif
824 #ifdef INET6
825 	if (V_udp6_tun_socket != NULL) {
826 		/* Already running -- must stop first */
827 		return (EALREADY);
828 	}
829 #endif
830 #ifdef INET
831 	if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
832 	    SOCK_DGRAM, IPPROTO_UDP,
833 	    curthread->td_ucred, curthread))) {
834 		tcp_over_udp_stop();
835 		return (ret);
836 	}
837 	/* Call the special UDP hook. */
838 	if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
839 	    tcp_recv_udp_tunneled_packet,
840 	    tcp_ctlinput_viaudp,
841 	    NULL))) {
842 		tcp_over_udp_stop();
843 		return (ret);
844 	}
845 	/* Ok, we have a socket, bind it to the port. */
846 	memset(&sin, 0, sizeof(struct sockaddr_in));
847 	sin.sin_len = sizeof(struct sockaddr_in);
848 	sin.sin_family = AF_INET;
849 	sin.sin_port = htons(port);
850 	if ((ret = sobind(V_udp4_tun_socket,
851 	    (struct sockaddr *)&sin, curthread))) {
852 		tcp_over_udp_stop();
853 		return (ret);
854 	}
855 #endif
856 #ifdef INET6
857 	if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
858 	    SOCK_DGRAM, IPPROTO_UDP,
859 	    curthread->td_ucred, curthread))) {
860 		tcp_over_udp_stop();
861 		return (ret);
862 	}
863 	/* Call the special UDP hook. */
864 	if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
865 	    tcp_recv_udp_tunneled_packet,
866 	    tcp6_ctlinput_viaudp,
867 	    NULL))) {
868 		tcp_over_udp_stop();
869 		return (ret);
870 	}
871 	/* Ok, we have a socket, bind it to the port. */
872 	memset(&sin6, 0, sizeof(struct sockaddr_in6));
873 	sin6.sin6_len = sizeof(struct sockaddr_in6);
874 	sin6.sin6_family = AF_INET6;
875 	sin6.sin6_port = htons(port);
876 	if ((ret = sobind(V_udp6_tun_socket,
877 	    (struct sockaddr *)&sin6, curthread))) {
878 		tcp_over_udp_stop();
879 		return (ret);
880 	}
881 #endif
882 	return (0);
883 }
884 
885 static int
886 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
887 {
888 	int error;
889 	uint32_t old, new;
890 
891 	old = V_tcp_udp_tunneling_port;
892 	new = old;
893 	error = sysctl_handle_int(oidp, &new, 0, req);
894 	if ((error == 0) &&
895 	    (req->newptr != NULL)) {
896 		if ((new < TCP_TUNNELING_PORT_MIN) ||
897 		    (new > TCP_TUNNELING_PORT_MAX)) {
898 			error = EINVAL;
899 		} else {
900 			V_tcp_udp_tunneling_port = new;
901 			if (old != 0) {
902 				tcp_over_udp_stop();
903 			}
904 			if (new != 0) {
905 				error = tcp_over_udp_start();
906 			}
907 		}
908 	}
909 	return (error);
910 }
911 
912 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
913     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
914     &VNET_NAME(tcp_udp_tunneling_port),
915     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
916     "Tunneling port for tcp over udp");
917 
918 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
919 
920 static int
921 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
922 {
923 	int error, new;
924 
925 	new = V_tcp_udp_tunneling_overhead;
926 	error = sysctl_handle_int(oidp, &new, 0, req);
927 	if (error == 0 && req->newptr) {
928 		if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
929 		    (new > TCP_TUNNELING_OVERHEAD_MAX))
930 			error = EINVAL;
931 		else
932 			V_tcp_udp_tunneling_overhead = new;
933 	}
934 	return (error);
935 }
936 
937 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
938     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
939     &VNET_NAME(tcp_udp_tunneling_overhead),
940     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
941     "MSS reduction when using tcp over udp");
942 
943 /*
944  * Exports one (struct tcp_function_info) for each alias/name.
945  */
946 static int
947 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
948 {
949 	int cnt, error;
950 	struct tcp_function *f;
951 	struct tcp_function_info tfi;
952 
953 	/*
954 	 * We don't allow writes.
955 	 */
956 	if (req->newptr != NULL)
957 		return (EINVAL);
958 
959 	/*
960 	 * Wire the old buffer so we can directly copy the functions to
961 	 * user space without dropping the lock.
962 	 */
963 	if (req->oldptr != NULL) {
964 		error = sysctl_wire_old_buffer(req, 0);
965 		if (error)
966 			return (error);
967 	}
968 
969 	/*
970 	 * Walk the list and copy out matching entries. If INVARIANTS
971 	 * is compiled in, also walk the list to verify the length of
972 	 * the list matches what we have recorded.
973 	 */
974 	rw_rlock(&tcp_function_lock);
975 
976 	cnt = 0;
977 #ifndef INVARIANTS
978 	if (req->oldptr == NULL) {
979 		cnt = tcp_fb_cnt;
980 		goto skip_loop;
981 	}
982 #endif
983 	TAILQ_FOREACH(f, &t_functions, tf_next) {
984 #ifdef INVARIANTS
985 		cnt++;
986 #endif
987 		if (req->oldptr != NULL) {
988 			bzero(&tfi, sizeof(tfi));
989 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
990 			tfi.tfi_id = f->tf_fb->tfb_id;
991 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
992 			    sizeof(tfi.tfi_alias));
993 			(void)strlcpy(tfi.tfi_name,
994 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
995 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
996 			/*
997 			 * Don't stop on error, as that is the
998 			 * mechanism we use to accumulate length
999 			 * information if the buffer was too short.
1000 			 */
1001 		}
1002 	}
1003 	KASSERT(cnt == tcp_fb_cnt,
1004 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
1005 #ifndef INVARIANTS
1006 skip_loop:
1007 #endif
1008 	rw_runlock(&tcp_function_lock);
1009 	if (req->oldptr == NULL)
1010 		error = SYSCTL_OUT(req, NULL,
1011 		    (cnt + 1) * sizeof(struct tcp_function_info));
1012 
1013 	return (error);
1014 }
1015 
1016 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
1017 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
1018 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
1019 	    "List TCP function block name-to-ID mappings");
1020 
1021 /*
1022  * tfb_tcp_handoff_ok() function for the default stack.
1023  * Note that we'll basically try to take all comers.
1024  */
1025 static int
1026 tcp_default_handoff_ok(struct tcpcb *tp)
1027 {
1028 
1029 	return (0);
1030 }
1031 
1032 /*
1033  * tfb_tcp_fb_init() function for the default stack.
1034  *
1035  * This handles making sure we have appropriate timers set if you are
1036  * transitioning a socket that has some amount of setup done.
1037  *
1038  * The init() fuction from the default can *never* return non-zero i.e.
1039  * it is required to always succeed since it is the stack of last resort!
1040  */
1041 static int
1042 tcp_default_fb_init(struct tcpcb *tp)
1043 {
1044 
1045 	struct socket *so;
1046 
1047 	INP_WLOCK_ASSERT(tp->t_inpcb);
1048 
1049 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
1050 	    ("%s: connection %p in unexpected state %d", __func__, tp,
1051 	    tp->t_state));
1052 
1053 	/*
1054 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
1055 	 * know what to do for unexpected states (which includes TIME_WAIT).
1056 	 */
1057 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
1058 		return (0);
1059 
1060 	/*
1061 	 * Make sure some kind of transmission timer is set if there is
1062 	 * outstanding data.
1063 	 */
1064 	so = tp->t_inpcb->inp_socket;
1065 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
1066 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1067 	    tcp_timer_active(tp, TT_PERSIST))) {
1068 		/*
1069 		 * If the session has established and it looks like it should
1070 		 * be in the persist state, set the persist timer. Otherwise,
1071 		 * set the retransmit timer.
1072 		 */
1073 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1074 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
1075 		    (int32_t)sbavail(&so->so_snd))
1076 			tcp_setpersist(tp);
1077 		else
1078 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1079 	}
1080 
1081 	/* All non-embryonic sessions get a keepalive timer. */
1082 	if (!tcp_timer_active(tp, TT_KEEP))
1083 		tcp_timer_activate(tp, TT_KEEP,
1084 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1085 		    TP_KEEPINIT(tp));
1086 
1087 	/*
1088 	 * Make sure critical variables are initialized
1089 	 * if transitioning while in Recovery.
1090 	 */
1091 	if IN_FASTRECOVERY(tp->t_flags) {
1092 		if (tp->sackhint.recover_fs == 0)
1093 			tp->sackhint.recover_fs = max(1,
1094 			    tp->snd_nxt - tp->snd_una);
1095 	}
1096 
1097 	return (0);
1098 }
1099 
1100 /*
1101  * tfb_tcp_fb_fini() function for the default stack.
1102  *
1103  * This changes state as necessary (or prudent) to prepare for another stack
1104  * to assume responsibility for the connection.
1105  */
1106 static void
1107 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1108 {
1109 
1110 	INP_WLOCK_ASSERT(tp->t_inpcb);
1111 	return;
1112 }
1113 
1114 /*
1115  * Target size of TCP PCB hash tables. Must be a power of two.
1116  *
1117  * Note that this can be overridden by the kernel environment
1118  * variable net.inet.tcp.tcbhashsize
1119  */
1120 #ifndef TCBHASHSIZE
1121 #define TCBHASHSIZE	0
1122 #endif
1123 
1124 /*
1125  * XXX
1126  * Callouts should be moved into struct tcp directly.  They are currently
1127  * separate because the tcpcb structure is exported to userland for sysctl
1128  * parsing purposes, which do not know about callouts.
1129  */
1130 struct tcpcb_mem {
1131 	struct	tcpcb		tcb;
1132 	struct	tcp_timer	tt;
1133 	struct	cc_var		ccv;
1134 #ifdef TCP_HHOOK
1135 	struct	osd		osd;
1136 #endif
1137 };
1138 
1139 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1140 #define	V_tcpcb_zone			VNET(tcpcb_zone)
1141 
1142 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1143 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1144 
1145 static struct mtx isn_mtx;
1146 
1147 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1148 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
1149 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
1150 
1151 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash");
1152 
1153 /*
1154  * Take a value and get the next power of 2 that doesn't overflow.
1155  * Used to size the tcp_inpcb hash buckets.
1156  */
1157 static int
1158 maketcp_hashsize(int size)
1159 {
1160 	int hashsize;
1161 
1162 	/*
1163 	 * auto tune.
1164 	 * get the next power of 2 higher than maxsockets.
1165 	 */
1166 	hashsize = 1 << fls(size);
1167 	/* catch overflow, and just go one power of 2 smaller */
1168 	if (hashsize < size) {
1169 		hashsize = 1 << (fls(size) - 1);
1170 	}
1171 	return (hashsize);
1172 }
1173 
1174 static volatile int next_tcp_stack_id = 1;
1175 
1176 /*
1177  * Register a TCP function block with the name provided in the names
1178  * array.  (Note that this function does NOT automatically register
1179  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1180  * explicitly include blk->tfb_tcp_block_name in the list of names if
1181  * you wish to register the stack with that name.)
1182  *
1183  * Either all name registrations will succeed or all will fail.  If
1184  * a name registration fails, the function will update the num_names
1185  * argument to point to the array index of the name that encountered
1186  * the failure.
1187  *
1188  * Returns 0 on success, or an error code on failure.
1189  */
1190 int
1191 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1192     const char *names[], int *num_names)
1193 {
1194 	struct tcp_function *n;
1195 	struct tcp_function_set fs;
1196 	int error, i;
1197 
1198 	KASSERT(names != NULL && *num_names > 0,
1199 	    ("%s: Called with 0-length name list", __func__));
1200 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1201 	KASSERT(rw_initialized(&tcp_function_lock),
1202 	    ("%s: called too early", __func__));
1203 
1204 	if ((blk->tfb_tcp_output == NULL) ||
1205 	    (blk->tfb_tcp_do_segment == NULL) ||
1206 	    (blk->tfb_tcp_ctloutput == NULL) ||
1207 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
1208 		/*
1209 		 * These functions are required and you
1210 		 * need a name.
1211 		 */
1212 		*num_names = 0;
1213 		return (EINVAL);
1214 	}
1215 	if (blk->tfb_tcp_timer_stop_all ||
1216 	    blk->tfb_tcp_timer_activate ||
1217 	    blk->tfb_tcp_timer_active ||
1218 	    blk->tfb_tcp_timer_stop) {
1219 		/*
1220 		 * If you define one timer function you
1221 		 * must have them all.
1222 		 */
1223 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1224 		    (blk->tfb_tcp_timer_activate == NULL) ||
1225 		    (blk->tfb_tcp_timer_active == NULL) ||
1226 		    (blk->tfb_tcp_timer_stop == NULL)) {
1227 			*num_names = 0;
1228 			return (EINVAL);
1229 		}
1230 	}
1231 
1232 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1233 		*num_names = 0;
1234 		return (EINVAL);
1235 	}
1236 
1237 	refcount_init(&blk->tfb_refcnt, 0);
1238 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1239 	for (i = 0; i < *num_names; i++) {
1240 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1241 		if (n == NULL) {
1242 			error = ENOMEM;
1243 			goto cleanup;
1244 		}
1245 		n->tf_fb = blk;
1246 
1247 		(void)strlcpy(fs.function_set_name, names[i],
1248 		    sizeof(fs.function_set_name));
1249 		rw_wlock(&tcp_function_lock);
1250 		if (find_tcp_functions_locked(&fs) != NULL) {
1251 			/* Duplicate name space not allowed */
1252 			rw_wunlock(&tcp_function_lock);
1253 			free(n, M_TCPFUNCTIONS);
1254 			error = EALREADY;
1255 			goto cleanup;
1256 		}
1257 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1258 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1259 		tcp_fb_cnt++;
1260 		rw_wunlock(&tcp_function_lock);
1261 	}
1262 	return(0);
1263 
1264 cleanup:
1265 	/*
1266 	 * Deregister the names we just added. Because registration failed
1267 	 * for names[i], we don't need to deregister that name.
1268 	 */
1269 	*num_names = i;
1270 	rw_wlock(&tcp_function_lock);
1271 	while (--i >= 0) {
1272 		TAILQ_FOREACH(n, &t_functions, tf_next) {
1273 			if (!strncmp(n->tf_name, names[i],
1274 			    TCP_FUNCTION_NAME_LEN_MAX)) {
1275 				TAILQ_REMOVE(&t_functions, n, tf_next);
1276 				tcp_fb_cnt--;
1277 				n->tf_fb = NULL;
1278 				free(n, M_TCPFUNCTIONS);
1279 				break;
1280 			}
1281 		}
1282 	}
1283 	rw_wunlock(&tcp_function_lock);
1284 	return (error);
1285 }
1286 
1287 /*
1288  * Register a TCP function block using the name provided in the name
1289  * argument.
1290  *
1291  * Returns 0 on success, or an error code on failure.
1292  */
1293 int
1294 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1295     int wait)
1296 {
1297 	const char *name_list[1];
1298 	int num_names, rv;
1299 
1300 	num_names = 1;
1301 	if (name != NULL)
1302 		name_list[0] = name;
1303 	else
1304 		name_list[0] = blk->tfb_tcp_block_name;
1305 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1306 	return (rv);
1307 }
1308 
1309 /*
1310  * Register a TCP function block using the name defined in
1311  * blk->tfb_tcp_block_name.
1312  *
1313  * Returns 0 on success, or an error code on failure.
1314  */
1315 int
1316 register_tcp_functions(struct tcp_function_block *blk, int wait)
1317 {
1318 
1319 	return (register_tcp_functions_as_name(blk, NULL, wait));
1320 }
1321 
1322 /*
1323  * Deregister all names associated with a function block. This
1324  * functionally removes the function block from use within the system.
1325  *
1326  * When called with a true quiesce argument, mark the function block
1327  * as being removed so no more stacks will use it and determine
1328  * whether the removal would succeed.
1329  *
1330  * When called with a false quiesce argument, actually attempt the
1331  * removal.
1332  *
1333  * When called with a force argument, attempt to switch all TCBs to
1334  * use the default stack instead of returning EBUSY.
1335  *
1336  * Returns 0 on success (or if the removal would succeed, or an error
1337  * code on failure.
1338  */
1339 int
1340 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1341     bool force)
1342 {
1343 	struct tcp_function *f;
1344 
1345 	if (blk == &tcp_def_funcblk) {
1346 		/* You can't un-register the default */
1347 		return (EPERM);
1348 	}
1349 	rw_wlock(&tcp_function_lock);
1350 	if (blk == tcp_func_set_ptr) {
1351 		/* You can't free the current default */
1352 		rw_wunlock(&tcp_function_lock);
1353 		return (EBUSY);
1354 	}
1355 	/* Mark the block so no more stacks can use it. */
1356 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1357 	/*
1358 	 * If TCBs are still attached to the stack, attempt to switch them
1359 	 * to the default stack.
1360 	 */
1361 	if (force && blk->tfb_refcnt) {
1362 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
1363 		    INPLOOKUP_WLOCKPCB);
1364 		struct inpcb *inp;
1365 		struct tcpcb *tp;
1366 		VNET_ITERATOR_DECL(vnet_iter);
1367 
1368 		rw_wunlock(&tcp_function_lock);
1369 
1370 		VNET_LIST_RLOCK();
1371 		VNET_FOREACH(vnet_iter) {
1372 			CURVNET_SET(vnet_iter);
1373 			while ((inp = inp_next(&inpi)) != NULL) {
1374 				if (inp->inp_flags & INP_TIMEWAIT)
1375 					continue;
1376 				tp = intotcpcb(inp);
1377 				if (tp == NULL || tp->t_fb != blk)
1378 					continue;
1379 				tcp_switch_back_to_default(tp);
1380 			}
1381 			CURVNET_RESTORE();
1382 		}
1383 		VNET_LIST_RUNLOCK();
1384 
1385 		rw_wlock(&tcp_function_lock);
1386 	}
1387 	if (blk->tfb_refcnt) {
1388 		/* TCBs still attached. */
1389 		rw_wunlock(&tcp_function_lock);
1390 		return (EBUSY);
1391 	}
1392 	if (quiesce) {
1393 		/* Skip removal. */
1394 		rw_wunlock(&tcp_function_lock);
1395 		return (0);
1396 	}
1397 	/* Remove any function names that map to this function block. */
1398 	while (find_tcp_fb_locked(blk, &f) != NULL) {
1399 		TAILQ_REMOVE(&t_functions, f, tf_next);
1400 		tcp_fb_cnt--;
1401 		f->tf_fb = NULL;
1402 		free(f, M_TCPFUNCTIONS);
1403 	}
1404 	rw_wunlock(&tcp_function_lock);
1405 	return (0);
1406 }
1407 
1408 static void
1409 tcp_vnet_init(void *arg __unused)
1410 {
1411 
1412 #ifdef TCP_HHOOK
1413 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1414 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1415 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1416 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1417 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1418 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1419 #endif
1420 #ifdef STATS
1421 	if (tcp_stats_init())
1422 		printf("%s: WARNING: unable to initialise TCP stats\n",
1423 		    __func__);
1424 #endif
1425 	in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize,
1426 	    tcp_tcbhashsize);
1427 
1428 	/*
1429 	 * These have to be type stable for the benefit of the timers.
1430 	 */
1431 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1432 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1433 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1434 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1435 
1436 	tcp_tw_init();
1437 	syncache_init();
1438 	tcp_hc_init();
1439 
1440 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1441 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1442 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1443 
1444 	tcp_fastopen_init();
1445 
1446 	COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
1447 	VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
1448 
1449 	V_tcp_msl = TCPTV_MSL;
1450 }
1451 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH,
1452     tcp_vnet_init, NULL);
1453 
1454 static void
1455 tcp_init(void *arg __unused)
1456 {
1457 	const char *tcbhash_tuneable;
1458 	int hashsize;
1459 
1460 	tcp_reass_global_init();
1461 
1462 	/* XXX virtualize those below? */
1463 	tcp_delacktime = TCPTV_DELACK;
1464 	tcp_keepinit = TCPTV_KEEP_INIT;
1465 	tcp_keepidle = TCPTV_KEEP_IDLE;
1466 	tcp_keepintvl = TCPTV_KEEPINTVL;
1467 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1468 	tcp_rexmit_initial = TCPTV_RTOBASE;
1469 	if (tcp_rexmit_initial < 1)
1470 		tcp_rexmit_initial = 1;
1471 	tcp_rexmit_min = TCPTV_MIN;
1472 	if (tcp_rexmit_min < 1)
1473 		tcp_rexmit_min = 1;
1474 	tcp_persmin = TCPTV_PERSMIN;
1475 	tcp_persmax = TCPTV_PERSMAX;
1476 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1477 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1478 
1479 	/* Setup the tcp function block list */
1480 	TAILQ_INIT(&t_functions);
1481 	rw_init(&tcp_function_lock, "tcp_func_lock");
1482 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1483 #ifdef TCP_BLACKBOX
1484 	/* Initialize the TCP logging data. */
1485 	tcp_log_init();
1486 #endif
1487 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1488 
1489 	if (tcp_soreceive_stream) {
1490 #ifdef INET
1491 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1492 #endif
1493 #ifdef INET6
1494 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1495 #endif /* INET6 */
1496 	}
1497 
1498 #ifdef INET6
1499 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1500 #else /* INET6 */
1501 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1502 #endif /* INET6 */
1503 	if (max_protohdr < TCP_MINPROTOHDR)
1504 		max_protohdr = TCP_MINPROTOHDR;
1505 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1506 		panic("tcp_init");
1507 #undef TCP_MINPROTOHDR
1508 
1509 	ISN_LOCK_INIT();
1510 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1511 		SHUTDOWN_PRI_DEFAULT);
1512 
1513 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1514 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1515 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1516 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1517 	tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1518 	tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1519 	tcp_comp_total = counter_u64_alloc(M_WAITOK);
1520 	tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1521 	tcp_bad_csums = counter_u64_alloc(M_WAITOK);
1522 #ifdef TCPPCAP
1523 	tcp_pcap_init();
1524 #endif
1525 
1526 	hashsize = TCBHASHSIZE;
1527 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1528 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1529 	if (hashsize == 0) {
1530 		/*
1531 		 * Auto tune the hash size based on maxsockets.
1532 		 * A perfect hash would have a 1:1 mapping
1533 		 * (hashsize = maxsockets) however it's been
1534 		 * suggested that O(2) average is better.
1535 		 */
1536 		hashsize = maketcp_hashsize(maxsockets / 4);
1537 		/*
1538 		 * Our historical default is 512,
1539 		 * do not autotune lower than this.
1540 		 */
1541 		if (hashsize < 512)
1542 			hashsize = 512;
1543 		if (bootverbose)
1544 			printf("%s: %s auto tuned to %d\n", __func__,
1545 			    tcbhash_tuneable, hashsize);
1546 	}
1547 	/*
1548 	 * We require a hashsize to be a power of two.
1549 	 * Previously if it was not a power of two we would just reset it
1550 	 * back to 512, which could be a nasty surprise if you did not notice
1551 	 * the error message.
1552 	 * Instead what we do is clip it to the closest power of two lower
1553 	 * than the specified hash value.
1554 	 */
1555 	if (!powerof2(hashsize)) {
1556 		int oldhashsize = hashsize;
1557 
1558 		hashsize = maketcp_hashsize(hashsize);
1559 		/* prevent absurdly low value */
1560 		if (hashsize < 16)
1561 			hashsize = 16;
1562 		printf("%s: WARNING: TCB hash size not a power of 2, "
1563 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1564 		    hashsize);
1565 	}
1566 	tcp_tcbhashsize = hashsize;
1567 }
1568 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL);
1569 
1570 #ifdef VIMAGE
1571 static void
1572 tcp_destroy(void *unused __unused)
1573 {
1574 	int n;
1575 #ifdef TCP_HHOOK
1576 	int error;
1577 #endif
1578 
1579 	/*
1580 	 * All our processes are gone, all our sockets should be cleaned
1581 	 * up, which means, we should be past the tcp_discardcb() calls.
1582 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1583 	 */
1584 	for (;;) {
1585 		INP_INFO_WLOCK(&V_tcbinfo);
1586 		n = V_tcbinfo.ipi_count;
1587 		INP_INFO_WUNLOCK(&V_tcbinfo);
1588 		if (n == 0)
1589 			break;
1590 		pause("tcpdes", hz / 10);
1591 	}
1592 	tcp_hc_destroy();
1593 	syncache_destroy();
1594 	tcp_tw_destroy();
1595 	in_pcbinfo_destroy(&V_tcbinfo);
1596 	/* tcp_discardcb() clears the sack_holes up. */
1597 	uma_zdestroy(V_sack_hole_zone);
1598 	uma_zdestroy(V_tcpcb_zone);
1599 
1600 	/*
1601 	 * Cannot free the zone until all tcpcbs are released as we attach
1602 	 * the allocations to them.
1603 	 */
1604 	tcp_fastopen_destroy();
1605 
1606 	COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
1607 	VNET_PCPUSTAT_FREE(tcpstat);
1608 
1609 #ifdef TCP_HHOOK
1610 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1611 	if (error != 0) {
1612 		printf("%s: WARNING: unable to deregister helper hook "
1613 		    "type=%d, id=%d: error %d returned\n", __func__,
1614 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1615 	}
1616 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1617 	if (error != 0) {
1618 		printf("%s: WARNING: unable to deregister helper hook "
1619 		    "type=%d, id=%d: error %d returned\n", __func__,
1620 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1621 	}
1622 #endif
1623 }
1624 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1625 #endif
1626 
1627 void
1628 tcp_fini(void *xtp)
1629 {
1630 
1631 }
1632 
1633 /*
1634  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1635  * tcp_template used to store this data in mbufs, but we now recopy it out
1636  * of the tcpcb each time to conserve mbufs.
1637  */
1638 void
1639 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1640 {
1641 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1642 
1643 	INP_WLOCK_ASSERT(inp);
1644 
1645 #ifdef INET6
1646 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1647 		struct ip6_hdr *ip6;
1648 
1649 		ip6 = (struct ip6_hdr *)ip_ptr;
1650 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1651 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1652 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1653 			(IPV6_VERSION & IPV6_VERSION_MASK);
1654 		if (port == 0)
1655 			ip6->ip6_nxt = IPPROTO_TCP;
1656 		else
1657 			ip6->ip6_nxt = IPPROTO_UDP;
1658 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1659 		ip6->ip6_src = inp->in6p_laddr;
1660 		ip6->ip6_dst = inp->in6p_faddr;
1661 	}
1662 #endif /* INET6 */
1663 #if defined(INET6) && defined(INET)
1664 	else
1665 #endif
1666 #ifdef INET
1667 	{
1668 		struct ip *ip;
1669 
1670 		ip = (struct ip *)ip_ptr;
1671 		ip->ip_v = IPVERSION;
1672 		ip->ip_hl = 5;
1673 		ip->ip_tos = inp->inp_ip_tos;
1674 		ip->ip_len = 0;
1675 		ip->ip_id = 0;
1676 		ip->ip_off = 0;
1677 		ip->ip_ttl = inp->inp_ip_ttl;
1678 		ip->ip_sum = 0;
1679 		if (port == 0)
1680 			ip->ip_p = IPPROTO_TCP;
1681 		else
1682 			ip->ip_p = IPPROTO_UDP;
1683 		ip->ip_src = inp->inp_laddr;
1684 		ip->ip_dst = inp->inp_faddr;
1685 	}
1686 #endif /* INET */
1687 	th->th_sport = inp->inp_lport;
1688 	th->th_dport = inp->inp_fport;
1689 	th->th_seq = 0;
1690 	th->th_ack = 0;
1691 	th->th_off = 5;
1692 	tcp_set_flags(th, 0);
1693 	th->th_win = 0;
1694 	th->th_urp = 0;
1695 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1696 }
1697 
1698 /*
1699  * Create template to be used to send tcp packets on a connection.
1700  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1701  * use for this function is in keepalives, which use tcp_respond.
1702  */
1703 struct tcptemp *
1704 tcpip_maketemplate(struct inpcb *inp)
1705 {
1706 	struct tcptemp *t;
1707 
1708 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1709 	if (t == NULL)
1710 		return (NULL);
1711 	tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1712 	return (t);
1713 }
1714 
1715 /*
1716  * Send a single message to the TCP at address specified by
1717  * the given TCP/IP header.  If m == NULL, then we make a copy
1718  * of the tcpiphdr at th and send directly to the addressed host.
1719  * This is used to force keep alive messages out using the TCP
1720  * template for a connection.  If flags are given then we send
1721  * a message back to the TCP which originated the segment th,
1722  * and discard the mbuf containing it and any other attached mbufs.
1723  *
1724  * In any case the ack and sequence number of the transmitted
1725  * segment are as specified by the parameters.
1726  *
1727  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1728  */
1729 void
1730 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1731     tcp_seq ack, tcp_seq seq, int flags)
1732 {
1733 	struct tcpopt to;
1734 	struct inpcb *inp;
1735 	struct ip *ip;
1736 	struct mbuf *optm;
1737 	struct udphdr *uh = NULL;
1738 	struct tcphdr *nth;
1739 	struct tcp_log_buffer *lgb;
1740 	u_char *optp;
1741 #ifdef INET6
1742 	struct ip6_hdr *ip6;
1743 	int isipv6;
1744 #endif /* INET6 */
1745 	int optlen, tlen, win, ulen;
1746 	bool incl_opts;
1747 	uint16_t port;
1748 	int output_ret;
1749 #ifdef INVARIANTS
1750 	int thflags = tcp_get_flags(th);
1751 #endif
1752 
1753 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1754 	NET_EPOCH_ASSERT();
1755 
1756 #ifdef INET6
1757 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1758 	ip6 = ipgen;
1759 #endif /* INET6 */
1760 	ip = ipgen;
1761 
1762 	if (tp != NULL) {
1763 		inp = tp->t_inpcb;
1764 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1765 		INP_LOCK_ASSERT(inp);
1766 	} else
1767 		inp = NULL;
1768 
1769 	if (m != NULL) {
1770 #ifdef INET6
1771 		if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1772 			port = m->m_pkthdr.tcp_tun_port;
1773 		else
1774 #endif
1775 		if (ip && (ip->ip_p == IPPROTO_UDP))
1776 			port = m->m_pkthdr.tcp_tun_port;
1777 		else
1778 			port = 0;
1779 	} else
1780 		port = tp->t_port;
1781 
1782 	incl_opts = false;
1783 	win = 0;
1784 	if (tp != NULL) {
1785 		if (!(flags & TH_RST)) {
1786 			win = sbspace(&inp->inp_socket->so_rcv);
1787 			if (win > TCP_MAXWIN << tp->rcv_scale)
1788 				win = TCP_MAXWIN << tp->rcv_scale;
1789 		}
1790 		if ((tp->t_flags & TF_NOOPT) == 0)
1791 			incl_opts = true;
1792 	}
1793 	if (m == NULL) {
1794 		m = m_gethdr(M_NOWAIT, MT_DATA);
1795 		if (m == NULL)
1796 			return;
1797 		m->m_data += max_linkhdr;
1798 #ifdef INET6
1799 		if (isipv6) {
1800 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1801 			      sizeof(struct ip6_hdr));
1802 			ip6 = mtod(m, struct ip6_hdr *);
1803 			nth = (struct tcphdr *)(ip6 + 1);
1804 			if (port) {
1805 				/* Insert a UDP header */
1806 				uh = (struct udphdr *)nth;
1807 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1808 				uh->uh_dport = port;
1809 				nth = (struct tcphdr *)(uh + 1);
1810 			}
1811 		} else
1812 #endif /* INET6 */
1813 		{
1814 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1815 			ip = mtod(m, struct ip *);
1816 			nth = (struct tcphdr *)(ip + 1);
1817 			if (port) {
1818 				/* Insert a UDP header */
1819 				uh = (struct udphdr *)nth;
1820 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1821 				uh->uh_dport = port;
1822 				nth = (struct tcphdr *)(uh + 1);
1823 			}
1824 		}
1825 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1826 		flags = TH_ACK;
1827 	} else if ((!M_WRITABLE(m)) || (port != 0)) {
1828 		struct mbuf *n;
1829 
1830 		/* Can't reuse 'm', allocate a new mbuf. */
1831 		n = m_gethdr(M_NOWAIT, MT_DATA);
1832 		if (n == NULL) {
1833 			m_freem(m);
1834 			return;
1835 		}
1836 
1837 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1838 			m_freem(m);
1839 			m_freem(n);
1840 			return;
1841 		}
1842 
1843 		n->m_data += max_linkhdr;
1844 		/* m_len is set later */
1845 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1846 #ifdef INET6
1847 		if (isipv6) {
1848 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1849 			      sizeof(struct ip6_hdr));
1850 			ip6 = mtod(n, struct ip6_hdr *);
1851 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1852 			nth = (struct tcphdr *)(ip6 + 1);
1853 			if (port) {
1854 				/* Insert a UDP header */
1855 				uh = (struct udphdr *)nth;
1856 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1857 				uh->uh_dport = port;
1858 				nth = (struct tcphdr *)(uh + 1);
1859 			}
1860 		} else
1861 #endif /* INET6 */
1862 		{
1863 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1864 			ip = mtod(n, struct ip *);
1865 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1866 			nth = (struct tcphdr *)(ip + 1);
1867 			if (port) {
1868 				/* Insert a UDP header */
1869 				uh = (struct udphdr *)nth;
1870 				uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1871 				uh->uh_dport = port;
1872 				nth = (struct tcphdr *)(uh + 1);
1873 			}
1874 		}
1875 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1876 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1877 		th = nth;
1878 		m_freem(m);
1879 		m = n;
1880 	} else {
1881 		/*
1882 		 *  reuse the mbuf.
1883 		 * XXX MRT We inherit the FIB, which is lucky.
1884 		 */
1885 		m_freem(m->m_next);
1886 		m->m_next = NULL;
1887 		m->m_data = (caddr_t)ipgen;
1888 		/* m_len is set later */
1889 #ifdef INET6
1890 		if (isipv6) {
1891 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1892 			nth = (struct tcphdr *)(ip6 + 1);
1893 		} else
1894 #endif /* INET6 */
1895 		{
1896 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1897 			nth = (struct tcphdr *)(ip + 1);
1898 		}
1899 		if (th != nth) {
1900 			/*
1901 			 * this is usually a case when an extension header
1902 			 * exists between the IPv6 header and the
1903 			 * TCP header.
1904 			 */
1905 			nth->th_sport = th->th_sport;
1906 			nth->th_dport = th->th_dport;
1907 		}
1908 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1909 #undef xchg
1910 	}
1911 	tlen = 0;
1912 #ifdef INET6
1913 	if (isipv6)
1914 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1915 #endif
1916 #if defined(INET) && defined(INET6)
1917 	else
1918 #endif
1919 #ifdef INET
1920 		tlen = sizeof (struct tcpiphdr);
1921 #endif
1922 	if (port)
1923 		tlen += sizeof (struct udphdr);
1924 #ifdef INVARIANTS
1925 	m->m_len = 0;
1926 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1927 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1928 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1929 #endif
1930 	m->m_len = tlen;
1931 	to.to_flags = 0;
1932 	if (incl_opts) {
1933 		/* Make sure we have room. */
1934 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1935 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1936 			if (m->m_next) {
1937 				optp = mtod(m->m_next, u_char *);
1938 				optm = m->m_next;
1939 			} else
1940 				incl_opts = false;
1941 		} else {
1942 			optp = (u_char *) (nth + 1);
1943 			optm = m;
1944 		}
1945 	}
1946 	if (incl_opts) {
1947 		/* Timestamps. */
1948 		if (tp->t_flags & TF_RCVD_TSTMP) {
1949 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1950 			to.to_tsecr = tp->ts_recent;
1951 			to.to_flags |= TOF_TS;
1952 		}
1953 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1954 		/* TCP-MD5 (RFC2385). */
1955 		if (tp->t_flags & TF_SIGNATURE)
1956 			to.to_flags |= TOF_SIGNATURE;
1957 #endif
1958 		/* Add the options. */
1959 		tlen += optlen = tcp_addoptions(&to, optp);
1960 
1961 		/* Update m_len in the correct mbuf. */
1962 		optm->m_len += optlen;
1963 	} else
1964 		optlen = 0;
1965 #ifdef INET6
1966 	if (isipv6) {
1967 		if (uh) {
1968 			ulen = tlen - sizeof(struct ip6_hdr);
1969 			uh->uh_ulen = htons(ulen);
1970 		}
1971 		ip6->ip6_flow = 0;
1972 		ip6->ip6_vfc = IPV6_VERSION;
1973 		if (port)
1974 			ip6->ip6_nxt = IPPROTO_UDP;
1975 		else
1976 			ip6->ip6_nxt = IPPROTO_TCP;
1977 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1978 	}
1979 #endif
1980 #if defined(INET) && defined(INET6)
1981 	else
1982 #endif
1983 #ifdef INET
1984 	{
1985 		if (uh) {
1986 			ulen = tlen - sizeof(struct ip);
1987 			uh->uh_ulen = htons(ulen);
1988 		}
1989 		ip->ip_len = htons(tlen);
1990 		ip->ip_ttl = V_ip_defttl;
1991 		if (port) {
1992 			ip->ip_p = IPPROTO_UDP;
1993 		} else {
1994 			ip->ip_p = IPPROTO_TCP;
1995 		}
1996 		if (V_path_mtu_discovery)
1997 			ip->ip_off |= htons(IP_DF);
1998 	}
1999 #endif
2000 	m->m_pkthdr.len = tlen;
2001 	m->m_pkthdr.rcvif = NULL;
2002 #ifdef MAC
2003 	if (inp != NULL) {
2004 		/*
2005 		 * Packet is associated with a socket, so allow the
2006 		 * label of the response to reflect the socket label.
2007 		 */
2008 		INP_LOCK_ASSERT(inp);
2009 		mac_inpcb_create_mbuf(inp, m);
2010 	} else {
2011 		/*
2012 		 * Packet is not associated with a socket, so possibly
2013 		 * update the label in place.
2014 		 */
2015 		mac_netinet_tcp_reply(m);
2016 	}
2017 #endif
2018 	nth->th_seq = htonl(seq);
2019 	nth->th_ack = htonl(ack);
2020 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
2021 	tcp_set_flags(nth, flags);
2022 	if (tp != NULL)
2023 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
2024 	else
2025 		nth->th_win = htons((u_short)win);
2026 	nth->th_urp = 0;
2027 
2028 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2029 	if (to.to_flags & TOF_SIGNATURE) {
2030 		if (!TCPMD5_ENABLED() ||
2031 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
2032 			m_freem(m);
2033 			return;
2034 		}
2035 	}
2036 #endif
2037 
2038 #ifdef INET6
2039 	if (isipv6) {
2040 		if (port) {
2041 			m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
2042 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2043 			uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
2044 			nth->th_sum = 0;
2045 		} else {
2046 			m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
2047 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2048 			nth->th_sum = in6_cksum_pseudo(ip6,
2049 			    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
2050 		}
2051 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
2052 		    NULL, NULL);
2053 	}
2054 #endif /* INET6 */
2055 #if defined(INET6) && defined(INET)
2056 	else
2057 #endif
2058 #ifdef INET
2059 	{
2060 		if (port) {
2061 			uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2062 			    htons(ulen + IPPROTO_UDP));
2063 			m->m_pkthdr.csum_flags = CSUM_UDP;
2064 			m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2065 			nth->th_sum = 0;
2066 		} else {
2067 			m->m_pkthdr.csum_flags = CSUM_TCP;
2068 			m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2069 			nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2070 			    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2071 		}
2072 	}
2073 #endif /* INET */
2074 #ifdef TCPDEBUG
2075 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2076 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2077 #endif
2078 	TCP_PROBE3(debug__output, tp, th, m);
2079 	if (flags & TH_RST)
2080 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2081 	lgb = NULL;
2082 	if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) {
2083 		if (INP_WLOCKED(inp)) {
2084 			union tcp_log_stackspecific log;
2085 			struct timeval tv;
2086 
2087 			memset(&log.u_bbr, 0, sizeof(log.u_bbr));
2088 			log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts;
2089 			log.u_bbr.flex8 = 4;
2090 			log.u_bbr.pkts_out = tp->t_maxseg;
2091 			log.u_bbr.timeStamp = tcp_get_usecs(&tv);
2092 			log.u_bbr.delivered = 0;
2093 			lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT,
2094 			    ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv);
2095 		} else {
2096 			/*
2097 			 * We can not log the packet, since we only own the
2098 			 * read lock, but a write lock is needed. The read lock
2099 			 * is not upgraded to a write lock, since only getting
2100 			 * the read lock was done intentionally to improve the
2101 			 * handling of SYN flooding attacks.
2102 			 * This happens only for pure SYN segments received in
2103 			 * the initial CLOSED state, or received in a more
2104 			 * advanced state than listen and the UDP encapsulation
2105 			 * port is unexpected.
2106 			 * The incoming SYN segments do not really belong to
2107 			 * the TCP connection and the handling does not change
2108 			 * the state of the TCP connection. Therefore, the
2109 			 * sending of the RST segments is not logged. Please
2110 			 * note that also the incoming SYN segments are not
2111 			 * logged.
2112 			 *
2113 			 * The following code ensures that the above description
2114 			 * is and stays correct.
2115 			 */
2116 			KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN &&
2117 			    (tp->t_state == TCPS_CLOSED ||
2118 			    (tp->t_state > TCPS_LISTEN && tp->t_port != port)),
2119 			    ("%s: Logging of TCP segment with flags 0x%b and "
2120 			    "UDP encapsulation port %u skipped in state %s",
2121 			    __func__, thflags, PRINT_TH_FLAGS,
2122 			    ntohs(port), tcpstates[tp->t_state]));
2123 		}
2124 	}
2125 
2126 #ifdef INET6
2127 	if (isipv6) {
2128 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2129 		output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2130 	}
2131 #endif /* INET6 */
2132 #if defined(INET) && defined(INET6)
2133 	else
2134 #endif
2135 #ifdef INET
2136 	{
2137 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2138 		output_ret = ip_output(m, NULL, NULL, 0, NULL, inp);
2139 	}
2140 #endif
2141 	if (lgb != NULL)
2142 		lgb->tlb_errno = output_ret;
2143 }
2144 
2145 /*
2146  * Create a new TCP control block, making an
2147  * empty reassembly queue and hooking it to the argument
2148  * protocol control block.  The `inp' parameter must have
2149  * come from the zone allocator set up in tcp_init().
2150  */
2151 struct tcpcb *
2152 tcp_newtcpcb(struct inpcb *inp)
2153 {
2154 	struct tcpcb_mem *tm;
2155 	struct tcpcb *tp;
2156 #ifdef INET6
2157 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2158 #endif /* INET6 */
2159 
2160 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2161 	if (tm == NULL)
2162 		return (NULL);
2163 	tp = &tm->tcb;
2164 
2165 	/* Initialise cc_var struct for this tcpcb. */
2166 	tp->ccv = &tm->ccv;
2167 	tp->ccv->type = IPPROTO_TCP;
2168 	tp->ccv->ccvc.tcp = tp;
2169 	rw_rlock(&tcp_function_lock);
2170 	tp->t_fb = tcp_func_set_ptr;
2171 	refcount_acquire(&tp->t_fb->tfb_refcnt);
2172 	rw_runlock(&tcp_function_lock);
2173 	/*
2174 	 * Use the current system default CC algorithm.
2175 	 */
2176 	cc_attach(tp, CC_DEFAULT_ALGO());
2177 
2178 	/*
2179 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2180 	 * is called.
2181 	 */
2182 	in_pcbref(inp);	/* Reference for tcpcb */
2183 	tp->t_inpcb = inp;
2184 
2185 	if (CC_ALGO(tp)->cb_init != NULL)
2186 		if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) {
2187 			cc_detach(tp);
2188 			if (tp->t_fb->tfb_tcp_fb_fini)
2189 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2190 			in_pcbrele_wlocked(inp);
2191 			refcount_release(&tp->t_fb->tfb_refcnt);
2192 			uma_zfree(V_tcpcb_zone, tm);
2193 			return (NULL);
2194 		}
2195 
2196 #ifdef TCP_HHOOK
2197 	tp->osd = &tm->osd;
2198 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2199 		if (tp->t_fb->tfb_tcp_fb_fini)
2200 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2201 		in_pcbrele_wlocked(inp);
2202 		refcount_release(&tp->t_fb->tfb_refcnt);
2203 		uma_zfree(V_tcpcb_zone, tm);
2204 		return (NULL);
2205 	}
2206 #endif
2207 
2208 #ifdef VIMAGE
2209 	tp->t_vnet = inp->inp_vnet;
2210 #endif
2211 	tp->t_timers = &tm->tt;
2212 	TAILQ_INIT(&tp->t_segq);
2213 	tp->t_maxseg =
2214 #ifdef INET6
2215 		isipv6 ? V_tcp_v6mssdflt :
2216 #endif /* INET6 */
2217 		V_tcp_mssdflt;
2218 
2219 	/* Set up our timeouts. */
2220 	callout_init(&tp->t_timers->tt_rexmt, 1);
2221 	callout_init(&tp->t_timers->tt_persist, 1);
2222 	callout_init(&tp->t_timers->tt_keep, 1);
2223 	callout_init(&tp->t_timers->tt_2msl, 1);
2224 	callout_init(&tp->t_timers->tt_delack, 1);
2225 
2226 	if (V_tcp_do_rfc1323)
2227 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2228 	if (V_tcp_do_sack)
2229 		tp->t_flags |= TF_SACK_PERMIT;
2230 	TAILQ_INIT(&tp->snd_holes);
2231 
2232 	/*
2233 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2234 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2235 	 * reasonable initial retransmit time.
2236 	 */
2237 	tp->t_srtt = TCPTV_SRTTBASE;
2238 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2239 	tp->t_rttmin = tcp_rexmit_min;
2240 	tp->t_rxtcur = tcp_rexmit_initial;
2241 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2242 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2243 	tp->t_rcvtime = ticks;
2244 	/*
2245 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2246 	 * because the socket may be bound to an IPv6 wildcard address,
2247 	 * which may match an IPv4-mapped IPv6 address.
2248 	 */
2249 	inp->inp_ip_ttl = V_ip_defttl;
2250 	inp->inp_ppcb = tp;
2251 #ifdef TCPPCAP
2252 	/*
2253 	 * Init the TCP PCAP queues.
2254 	 */
2255 	tcp_pcap_tcpcb_init(tp);
2256 #endif
2257 #ifdef TCP_BLACKBOX
2258 	/* Initialize the per-TCPCB log data. */
2259 	tcp_log_tcpcbinit(tp);
2260 #endif
2261 	tp->t_pacing_rate = -1;
2262 	if (tp->t_fb->tfb_tcp_fb_init) {
2263 		if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2264 			refcount_release(&tp->t_fb->tfb_refcnt);
2265 			in_pcbrele_wlocked(inp);
2266 			uma_zfree(V_tcpcb_zone, tm);
2267 			return (NULL);
2268 		}
2269 	}
2270 #ifdef STATS
2271 	if (V_tcp_perconn_stats_enable == 1)
2272 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2273 #endif
2274 	if (V_tcp_do_lrd)
2275 		tp->t_flags |= TF_LRD;
2276 	return (tp);		/* XXX */
2277 }
2278 
2279 /*
2280  * Drop a TCP connection, reporting
2281  * the specified error.  If connection is synchronized,
2282  * then send a RST to peer.
2283  */
2284 struct tcpcb *
2285 tcp_drop(struct tcpcb *tp, int errno)
2286 {
2287 	struct socket *so = tp->t_inpcb->inp_socket;
2288 
2289 	NET_EPOCH_ASSERT();
2290 	INP_WLOCK_ASSERT(tp->t_inpcb);
2291 
2292 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
2293 		tcp_state_change(tp, TCPS_CLOSED);
2294 		/* Don't use tcp_output() here due to possible recursion. */
2295 		(void)tcp_output_nodrop(tp);
2296 		TCPSTAT_INC(tcps_drops);
2297 	} else
2298 		TCPSTAT_INC(tcps_conndrops);
2299 	if (errno == ETIMEDOUT && tp->t_softerror)
2300 		errno = tp->t_softerror;
2301 	so->so_error = errno;
2302 	return (tcp_close(tp));
2303 }
2304 
2305 void
2306 tcp_discardcb(struct tcpcb *tp)
2307 {
2308 	struct inpcb *inp = tp->t_inpcb;
2309 
2310 	INP_WLOCK_ASSERT(inp);
2311 
2312 	/*
2313 	 * Make sure that all of our timers are stopped before we delete the
2314 	 * PCB.
2315 	 *
2316 	 * If stopping a timer fails, we schedule a discard function in same
2317 	 * callout, and the last discard function called will take care of
2318 	 * deleting the tcpcb.
2319 	 */
2320 	tp->t_timers->tt_draincnt = 0;
2321 	tcp_timer_stop(tp, TT_REXMT);
2322 	tcp_timer_stop(tp, TT_PERSIST);
2323 	tcp_timer_stop(tp, TT_KEEP);
2324 	tcp_timer_stop(tp, TT_2MSL);
2325 	tcp_timer_stop(tp, TT_DELACK);
2326 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
2327 		/*
2328 		 * Call the stop-all function of the methods,
2329 		 * this function should call the tcp_timer_stop()
2330 		 * method with each of the function specific timeouts.
2331 		 * That stop will be called via the tfb_tcp_timer_stop()
2332 		 * which should use the async drain function of the
2333 		 * callout system (see tcp_var.h).
2334 		 */
2335 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
2336 	}
2337 
2338 	/* free the reassembly queue, if any */
2339 	tcp_reass_flush(tp);
2340 
2341 #ifdef TCP_OFFLOAD
2342 	/* Disconnect offload device, if any. */
2343 	if (tp->t_flags & TF_TOE)
2344 		tcp_offload_detach(tp);
2345 #endif
2346 
2347 	tcp_free_sackholes(tp);
2348 
2349 #ifdef TCPPCAP
2350 	/* Free the TCP PCAP queues. */
2351 	tcp_pcap_drain(&(tp->t_inpkts));
2352 	tcp_pcap_drain(&(tp->t_outpkts));
2353 #endif
2354 
2355 	/* Allow the CC algorithm to clean up after itself. */
2356 	if (CC_ALGO(tp)->cb_destroy != NULL)
2357 		CC_ALGO(tp)->cb_destroy(tp->ccv);
2358 	CC_DATA(tp) = NULL;
2359 	/* Detach from the CC algorithm */
2360 	cc_detach(tp);
2361 
2362 #ifdef TCP_HHOOK
2363 	khelp_destroy_osd(tp->osd);
2364 #endif
2365 #ifdef STATS
2366 	stats_blob_destroy(tp->t_stats);
2367 #endif
2368 
2369 	CC_ALGO(tp) = NULL;
2370 	inp->inp_ppcb = NULL;
2371 	if (tp->t_timers->tt_draincnt == 0) {
2372 		bool released __diagused;
2373 
2374 		released = tcp_freecb(tp);
2375 		KASSERT(!released, ("%s: inp %p should not have been released "
2376 		    "here", __func__, inp));
2377 	}
2378 }
2379 
2380 bool
2381 tcp_freecb(struct tcpcb *tp)
2382 {
2383 	struct inpcb *inp = tp->t_inpcb;
2384 	struct socket *so = inp->inp_socket;
2385 #ifdef INET6
2386 	bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2387 #endif
2388 
2389 	INP_WLOCK_ASSERT(inp);
2390 	MPASS(tp->t_timers->tt_draincnt == 0);
2391 
2392 	/* We own the last reference on tcpcb, let's free it. */
2393 #ifdef TCP_BLACKBOX
2394 	tcp_log_tcpcbfini(tp);
2395 #endif
2396 	TCPSTATES_DEC(tp->t_state);
2397 	if (tp->t_fb->tfb_tcp_fb_fini)
2398 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2399 
2400 	/*
2401 	 * If we got enough samples through the srtt filter,
2402 	 * save the rtt and rttvar in the routing entry.
2403 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
2404 	 * 4 samples is enough for the srtt filter to converge
2405 	 * to within enough % of the correct value; fewer samples
2406 	 * and we could save a bogus rtt. The danger is not high
2407 	 * as tcp quickly recovers from everything.
2408 	 * XXX: Works very well but needs some more statistics!
2409 	 *
2410 	 * XXXRRS: Updating must be after the stack fini() since
2411 	 * that may be converting some internal representation of
2412 	 * say srtt etc into the general one used by other stacks.
2413 	 * Lets also at least protect against the so being NULL
2414 	 * as RW stated below.
2415 	 */
2416 	if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2417 		struct hc_metrics_lite metrics;
2418 		uint32_t ssthresh;
2419 
2420 		bzero(&metrics, sizeof(metrics));
2421 		/*
2422 		 * Update the ssthresh always when the conditions below
2423 		 * are satisfied. This gives us better new start value
2424 		 * for the congestion avoidance for new connections.
2425 		 * ssthresh is only set if packet loss occurred on a session.
2426 		 *
2427 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2428 		 * being torn down.  Ideally this code would not use 'so'.
2429 		 */
2430 		ssthresh = tp->snd_ssthresh;
2431 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2432 			/*
2433 			 * convert the limit from user data bytes to
2434 			 * packets then to packet data bytes.
2435 			 */
2436 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2437 			if (ssthresh < 2)
2438 				ssthresh = 2;
2439 			ssthresh *= (tp->t_maxseg +
2440 #ifdef INET6
2441 			    (isipv6 ? sizeof (struct ip6_hdr) +
2442 			    sizeof (struct tcphdr) :
2443 #endif
2444 			    sizeof (struct tcpiphdr)
2445 #ifdef INET6
2446 			    )
2447 #endif
2448 			    );
2449 		} else
2450 			ssthresh = 0;
2451 		metrics.rmx_ssthresh = ssthresh;
2452 
2453 		metrics.rmx_rtt = tp->t_srtt;
2454 		metrics.rmx_rttvar = tp->t_rttvar;
2455 		metrics.rmx_cwnd = tp->snd_cwnd;
2456 		metrics.rmx_sendpipe = 0;
2457 		metrics.rmx_recvpipe = 0;
2458 
2459 		tcp_hc_update(&inp->inp_inc, &metrics);
2460 	}
2461 
2462 	refcount_release(&tp->t_fb->tfb_refcnt);
2463 	uma_zfree(V_tcpcb_zone, tp);
2464 
2465 	return (in_pcbrele_wlocked(inp));
2466 }
2467 
2468 /*
2469  * Attempt to close a TCP control block, marking it as dropped, and freeing
2470  * the socket if we hold the only reference.
2471  */
2472 struct tcpcb *
2473 tcp_close(struct tcpcb *tp)
2474 {
2475 	struct inpcb *inp = tp->t_inpcb;
2476 	struct socket *so;
2477 
2478 	INP_WLOCK_ASSERT(inp);
2479 
2480 #ifdef TCP_OFFLOAD
2481 	if (tp->t_state == TCPS_LISTEN)
2482 		tcp_offload_listen_stop(tp);
2483 #endif
2484 	/*
2485 	 * This releases the TFO pending counter resource for TFO listen
2486 	 * sockets as well as passively-created TFO sockets that transition
2487 	 * from SYN_RECEIVED to CLOSED.
2488 	 */
2489 	if (tp->t_tfo_pending) {
2490 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2491 		tp->t_tfo_pending = NULL;
2492 	}
2493 #ifdef TCPHPTS
2494 	tcp_hpts_remove(inp);
2495 #endif
2496 	in_pcbdrop(inp);
2497 	TCPSTAT_INC(tcps_closed);
2498 	if (tp->t_state != TCPS_CLOSED)
2499 		tcp_state_change(tp, TCPS_CLOSED);
2500 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2501 	so = inp->inp_socket;
2502 	soisdisconnected(so);
2503 	if (inp->inp_flags & INP_SOCKREF) {
2504 		inp->inp_flags &= ~INP_SOCKREF;
2505 		INP_WUNLOCK(inp);
2506 		sorele(so);
2507 		return (NULL);
2508 	}
2509 	return (tp);
2510 }
2511 
2512 void
2513 tcp_drain(void)
2514 {
2515 	VNET_ITERATOR_DECL(vnet_iter);
2516 
2517 	if (!do_tcpdrain)
2518 		return;
2519 
2520 	VNET_LIST_RLOCK_NOSLEEP();
2521 	VNET_FOREACH(vnet_iter) {
2522 		CURVNET_SET(vnet_iter);
2523 		struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2524 		    INPLOOKUP_WLOCKPCB);
2525 		struct inpcb *inpb;
2526 		struct tcpcb *tcpb;
2527 
2528 	/*
2529 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2530 	 * if there is one...
2531 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2532 	 *      reassembly queue should be flushed, but in a situation
2533 	 *	where we're really low on mbufs, this is potentially
2534 	 *	useful.
2535 	 */
2536 		while ((inpb = inp_next(&inpi)) != NULL) {
2537 			if (inpb->inp_flags & INP_TIMEWAIT)
2538 				continue;
2539 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2540 				tcp_reass_flush(tcpb);
2541 				tcp_clean_sackreport(tcpb);
2542 #ifdef TCP_BLACKBOX
2543 				tcp_log_drain(tcpb);
2544 #endif
2545 #ifdef TCPPCAP
2546 				if (tcp_pcap_aggressive_free) {
2547 					/* Free the TCP PCAP queues. */
2548 					tcp_pcap_drain(&(tcpb->t_inpkts));
2549 					tcp_pcap_drain(&(tcpb->t_outpkts));
2550 				}
2551 #endif
2552 			}
2553 		}
2554 		CURVNET_RESTORE();
2555 	}
2556 	VNET_LIST_RUNLOCK_NOSLEEP();
2557 }
2558 
2559 /*
2560  * Notify a tcp user of an asynchronous error;
2561  * store error as soft error, but wake up user
2562  * (for now, won't do anything until can select for soft error).
2563  *
2564  * Do not wake up user since there currently is no mechanism for
2565  * reporting soft errors (yet - a kqueue filter may be added).
2566  */
2567 static struct inpcb *
2568 tcp_notify(struct inpcb *inp, int error)
2569 {
2570 	struct tcpcb *tp;
2571 
2572 	INP_WLOCK_ASSERT(inp);
2573 
2574 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2575 	    (inp->inp_flags & INP_DROPPED))
2576 		return (inp);
2577 
2578 	tp = intotcpcb(inp);
2579 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2580 
2581 	/*
2582 	 * Ignore some errors if we are hooked up.
2583 	 * If connection hasn't completed, has retransmitted several times,
2584 	 * and receives a second error, give up now.  This is better
2585 	 * than waiting a long time to establish a connection that
2586 	 * can never complete.
2587 	 */
2588 	if (tp->t_state == TCPS_ESTABLISHED &&
2589 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2590 	     error == EHOSTDOWN)) {
2591 		if (inp->inp_route.ro_nh) {
2592 			NH_FREE(inp->inp_route.ro_nh);
2593 			inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2594 		}
2595 		return (inp);
2596 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2597 	    tp->t_softerror) {
2598 		tp = tcp_drop(tp, error);
2599 		if (tp != NULL)
2600 			return (inp);
2601 		else
2602 			return (NULL);
2603 	} else {
2604 		tp->t_softerror = error;
2605 		return (inp);
2606 	}
2607 #if 0
2608 	wakeup( &so->so_timeo);
2609 	sorwakeup(so);
2610 	sowwakeup(so);
2611 #endif
2612 }
2613 
2614 static int
2615 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2616 {
2617 	struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo,
2618 	    INPLOOKUP_RLOCKPCB);
2619 	struct xinpgen xig;
2620 	struct inpcb *inp;
2621 	int error;
2622 
2623 	if (req->newptr != NULL)
2624 		return (EPERM);
2625 
2626 	if (req->oldptr == NULL) {
2627 		int n;
2628 
2629 		n = V_tcbinfo.ipi_count +
2630 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2631 		n += imax(n / 8, 10);
2632 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2633 		return (0);
2634 	}
2635 
2636 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2637 		return (error);
2638 
2639 	bzero(&xig, sizeof(xig));
2640 	xig.xig_len = sizeof xig;
2641 	xig.xig_count = V_tcbinfo.ipi_count +
2642 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2643 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2644 	xig.xig_sogen = so_gencnt;
2645 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2646 	if (error)
2647 		return (error);
2648 
2649 	error = syncache_pcblist(req);
2650 	if (error)
2651 		return (error);
2652 
2653 	while ((inp = inp_next(&inpi)) != NULL) {
2654 		if (inp->inp_gencnt <= xig.xig_gen) {
2655 			int crerr;
2656 
2657 			/*
2658 			 * XXX: This use of cr_cansee(), introduced with
2659 			 * TCP state changes, is not quite right, but for
2660 			 * now, better than nothing.
2661 			 */
2662 			if (inp->inp_flags & INP_TIMEWAIT) {
2663 				if (intotw(inp) != NULL)
2664 					crerr = cr_cansee(req->td->td_ucred,
2665 					    intotw(inp)->tw_cred);
2666 				else
2667 					crerr = EINVAL;	/* Skip this inp. */
2668 			} else
2669 				crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2670 			if (crerr == 0) {
2671 				struct xtcpcb xt;
2672 
2673 				tcp_inptoxtp(inp, &xt);
2674 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2675 				if (error) {
2676 					INP_RUNLOCK(inp);
2677 					break;
2678 				} else
2679 					continue;
2680 			}
2681 		}
2682 	}
2683 
2684 	if (!error) {
2685 		/*
2686 		 * Give the user an updated idea of our state.
2687 		 * If the generation differs from what we told
2688 		 * her before, she knows that something happened
2689 		 * while we were processing this request, and it
2690 		 * might be necessary to retry.
2691 		 */
2692 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2693 		xig.xig_sogen = so_gencnt;
2694 		xig.xig_count = V_tcbinfo.ipi_count +
2695 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2696 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2697 	}
2698 
2699 	return (error);
2700 }
2701 
2702 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2703     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2704     NULL, 0, tcp_pcblist, "S,xtcpcb",
2705     "List of active TCP connections");
2706 
2707 #ifdef INET
2708 static int
2709 tcp_getcred(SYSCTL_HANDLER_ARGS)
2710 {
2711 	struct xucred xuc;
2712 	struct sockaddr_in addrs[2];
2713 	struct epoch_tracker et;
2714 	struct inpcb *inp;
2715 	int error;
2716 
2717 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2718 	if (error)
2719 		return (error);
2720 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2721 	if (error)
2722 		return (error);
2723 	NET_EPOCH_ENTER(et);
2724 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2725 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2726 	NET_EPOCH_EXIT(et);
2727 	if (inp != NULL) {
2728 		if (inp->inp_socket == NULL)
2729 			error = ENOENT;
2730 		if (error == 0)
2731 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2732 		if (error == 0)
2733 			cru2x(inp->inp_cred, &xuc);
2734 		INP_RUNLOCK(inp);
2735 	} else
2736 		error = ENOENT;
2737 	if (error == 0)
2738 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2739 	return (error);
2740 }
2741 
2742 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2743     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2744     0, 0, tcp_getcred, "S,xucred",
2745     "Get the xucred of a TCP connection");
2746 #endif /* INET */
2747 
2748 #ifdef INET6
2749 static int
2750 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2751 {
2752 	struct epoch_tracker et;
2753 	struct xucred xuc;
2754 	struct sockaddr_in6 addrs[2];
2755 	struct inpcb *inp;
2756 	int error;
2757 #ifdef INET
2758 	int mapped = 0;
2759 #endif
2760 
2761 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2762 	if (error)
2763 		return (error);
2764 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2765 	if (error)
2766 		return (error);
2767 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2768 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2769 		return (error);
2770 	}
2771 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2772 #ifdef INET
2773 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2774 			mapped = 1;
2775 		else
2776 #endif
2777 			return (EINVAL);
2778 	}
2779 
2780 	NET_EPOCH_ENTER(et);
2781 #ifdef INET
2782 	if (mapped == 1)
2783 		inp = in_pcblookup(&V_tcbinfo,
2784 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2785 			addrs[1].sin6_port,
2786 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2787 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2788 	else
2789 #endif
2790 		inp = in6_pcblookup(&V_tcbinfo,
2791 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2792 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2793 			INPLOOKUP_RLOCKPCB, NULL);
2794 	NET_EPOCH_EXIT(et);
2795 	if (inp != NULL) {
2796 		if (inp->inp_socket == NULL)
2797 			error = ENOENT;
2798 		if (error == 0)
2799 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2800 		if (error == 0)
2801 			cru2x(inp->inp_cred, &xuc);
2802 		INP_RUNLOCK(inp);
2803 	} else
2804 		error = ENOENT;
2805 	if (error == 0)
2806 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2807 	return (error);
2808 }
2809 
2810 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2811     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2812     0, 0, tcp6_getcred, "S,xucred",
2813     "Get the xucred of a TCP6 connection");
2814 #endif /* INET6 */
2815 
2816 #ifdef INET
2817 /* Path MTU to try next when a fragmentation-needed message is received. */
2818 static inline int
2819 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2820 {
2821 	int mtu = ntohs(icp->icmp_nextmtu);
2822 
2823 	/* If no alternative MTU was proposed, try the next smaller one. */
2824 	if (!mtu)
2825 		mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2826 	if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2827 		mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2828 
2829 	return (mtu);
2830 }
2831 
2832 static void
2833 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2834 {
2835 	struct ip *ip = vip;
2836 	struct tcphdr *th;
2837 	struct in_addr faddr;
2838 	struct inpcb *inp;
2839 	struct tcpcb *tp;
2840 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2841 	struct icmp *icp;
2842 	struct in_conninfo inc;
2843 	tcp_seq icmp_tcp_seq;
2844 	int mtu;
2845 
2846 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2847 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2848 		return;
2849 
2850 	if (cmd == PRC_MSGSIZE)
2851 		notify = tcp_mtudisc_notify;
2852 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2853 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2854 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2855 		notify = tcp_drop_syn_sent;
2856 
2857 	/*
2858 	 * Hostdead is ugly because it goes linearly through all PCBs.
2859 	 * XXX: We never get this from ICMP, otherwise it makes an
2860 	 * excellent DoS attack on machines with many connections.
2861 	 */
2862 	else if (cmd == PRC_HOSTDEAD)
2863 		ip = NULL;
2864 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2865 		return;
2866 
2867 	if (ip == NULL) {
2868 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2869 		return;
2870 	}
2871 
2872 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2873 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2874 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2875 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2876 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2877 		/* signal EHOSTDOWN, as it flushes the cached route */
2878 		inp = (*notify)(inp, EHOSTDOWN);
2879 		goto out;
2880 	}
2881 	icmp_tcp_seq = th->th_seq;
2882 	if (inp != NULL)  {
2883 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2884 		    !(inp->inp_flags & INP_DROPPED) &&
2885 		    !(inp->inp_socket == NULL)) {
2886 			tp = intotcpcb(inp);
2887 #ifdef TCP_OFFLOAD
2888 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2889 				/*
2890 				 * MTU discovery for offloaded connections.  Let
2891 				 * the TOE driver verify seq# and process it.
2892 				 */
2893 				mtu = tcp_next_pmtu(icp, ip);
2894 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2895 				goto out;
2896 			}
2897 #endif
2898 			if (tp->t_port != port) {
2899 				goto out;
2900 			}
2901 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2902 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2903 				if (cmd == PRC_MSGSIZE) {
2904 					/*
2905 					 * MTU discovery: we got a needfrag and
2906 					 * will potentially try a lower MTU.
2907 					 */
2908 					mtu = tcp_next_pmtu(icp, ip);
2909 
2910 					/*
2911 					 * Only process the offered MTU if it
2912 					 * is smaller than the current one.
2913 					 */
2914 					if (mtu < tp->t_maxseg +
2915 					    sizeof(struct tcpiphdr)) {
2916 						bzero(&inc, sizeof(inc));
2917 						inc.inc_faddr = faddr;
2918 						inc.inc_fibnum =
2919 						    inp->inp_inc.inc_fibnum;
2920 						tcp_hc_updatemtu(&inc, mtu);
2921 						inp = tcp_mtudisc(inp, mtu);
2922 					}
2923 				} else
2924 					inp = (*notify)(inp,
2925 					    inetctlerrmap[cmd]);
2926 			}
2927 		}
2928 	} else {
2929 		bzero(&inc, sizeof(inc));
2930 		inc.inc_fport = th->th_dport;
2931 		inc.inc_lport = th->th_sport;
2932 		inc.inc_faddr = faddr;
2933 		inc.inc_laddr = ip->ip_src;
2934 		syncache_unreach(&inc, icmp_tcp_seq, port);
2935 	}
2936 out:
2937 	if (inp != NULL)
2938 		INP_WUNLOCK(inp);
2939 }
2940 
2941 void
2942 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2943 {
2944 	tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2945 }
2946 
2947 void
2948 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2949 {
2950 	/* Its a tunneled TCP over UDP icmp */
2951 	struct ip *outer_ip, *inner_ip;
2952 	struct icmp *icmp;
2953 	struct udphdr *udp;
2954 	struct tcphdr *th, ttemp;
2955 	int i_hlen, o_len;
2956 	uint16_t port;
2957 
2958 	inner_ip = (struct ip *)vip;
2959 	icmp = (struct icmp *)((caddr_t)inner_ip -
2960 	    (sizeof(struct icmp) - sizeof(struct ip)));
2961 	outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2962 	i_hlen = inner_ip->ip_hl << 2;
2963 	o_len = ntohs(outer_ip->ip_len);
2964 	if (o_len <
2965 	    (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2966 		/* Not enough data present */
2967 		return;
2968 	}
2969 	/* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2970 	udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2971 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2972 		return;
2973 	}
2974 	port = udp->uh_dport;
2975 	th = (struct tcphdr *)(udp + 1);
2976 	memcpy(&ttemp, th, sizeof(struct tcphdr));
2977 	memcpy(udp, &ttemp, sizeof(struct tcphdr));
2978 	/* Now adjust down the size of the outer IP header */
2979 	o_len -= sizeof(struct udphdr);
2980 	outer_ip->ip_len = htons(o_len);
2981 	/* Now call in to the normal handling code */
2982 	tcp_ctlinput_with_port(cmd, sa, vip, port);
2983 }
2984 #endif /* INET */
2985 
2986 #ifdef INET6
2987 static inline int
2988 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2989 {
2990 	int mtu = ntohl(icmp6->icmp6_mtu);
2991 
2992 	/*
2993 	 * If no alternative MTU was proposed, or the proposed MTU was too
2994 	 * small, set to the min.
2995 	 */
2996 	if (mtu < IPV6_MMTU)
2997 		mtu = IPV6_MMTU - 8;	/* XXXNP: what is the adjustment for? */
2998 	return (mtu);
2999 }
3000 
3001 static void
3002 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3003 {
3004 	struct in6_addr *dst;
3005 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3006 	struct ip6_hdr *ip6;
3007 	struct mbuf *m;
3008 	struct inpcb *inp;
3009 	struct tcpcb *tp;
3010 	struct icmp6_hdr *icmp6;
3011 	struct ip6ctlparam *ip6cp = NULL;
3012 	const struct sockaddr_in6 *sa6_src = NULL;
3013 	struct in_conninfo inc;
3014 	struct tcp_ports {
3015 		uint16_t th_sport;
3016 		uint16_t th_dport;
3017 	} t_ports;
3018 	tcp_seq icmp_tcp_seq;
3019 	unsigned int mtu;
3020 	unsigned int off;
3021 
3022 	if (sa->sa_family != AF_INET6 ||
3023 	    sa->sa_len != sizeof(struct sockaddr_in6))
3024 		return;
3025 
3026 	/* if the parameter is from icmp6, decode it. */
3027 	if (d != NULL) {
3028 		ip6cp = (struct ip6ctlparam *)d;
3029 		icmp6 = ip6cp->ip6c_icmp6;
3030 		m = ip6cp->ip6c_m;
3031 		ip6 = ip6cp->ip6c_ip6;
3032 		off = ip6cp->ip6c_off;
3033 		sa6_src = ip6cp->ip6c_src;
3034 		dst = ip6cp->ip6c_finaldst;
3035 	} else {
3036 		m = NULL;
3037 		ip6 = NULL;
3038 		off = 0;	/* fool gcc */
3039 		sa6_src = &sa6_any;
3040 		dst = NULL;
3041 	}
3042 
3043 	if (cmd == PRC_MSGSIZE)
3044 		notify = tcp_mtudisc_notify;
3045 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3046 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3047 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3048 		notify = tcp_drop_syn_sent;
3049 
3050 	/*
3051 	 * Hostdead is ugly because it goes linearly through all PCBs.
3052 	 * XXX: We never get this from ICMP, otherwise it makes an
3053 	 * excellent DoS attack on machines with many connections.
3054 	 */
3055 	else if (cmd == PRC_HOSTDEAD)
3056 		ip6 = NULL;
3057 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3058 		return;
3059 
3060 	if (ip6 == NULL) {
3061 		in6_pcbnotify(&V_tcbinfo, sa, 0,
3062 			      (const struct sockaddr *)sa6_src,
3063 			      0, cmd, NULL, notify);
3064 		return;
3065 	}
3066 
3067 	/* Check if we can safely get the ports from the tcp hdr */
3068 	if (m == NULL ||
3069 	    (m->m_pkthdr.len <
3070 		(int32_t) (off + sizeof(struct tcp_ports)))) {
3071 		return;
3072 	}
3073 	bzero(&t_ports, sizeof(struct tcp_ports));
3074 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3075 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3076 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3077 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3078 		/* signal EHOSTDOWN, as it flushes the cached route */
3079 		inp = (*notify)(inp, EHOSTDOWN);
3080 		goto out;
3081 	}
3082 	off += sizeof(struct tcp_ports);
3083 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3084 		goto out;
3085 	}
3086 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3087 	if (inp != NULL)  {
3088 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
3089 		    !(inp->inp_flags & INP_DROPPED) &&
3090 		    !(inp->inp_socket == NULL)) {
3091 			tp = intotcpcb(inp);
3092 #ifdef TCP_OFFLOAD
3093 			if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3094 				/* MTU discovery for offloaded connections. */
3095 				mtu = tcp6_next_pmtu(icmp6);
3096 				tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3097 				goto out;
3098 			}
3099 #endif
3100 			if (tp->t_port != port) {
3101 				goto out;
3102 			}
3103 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3104 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3105 				if (cmd == PRC_MSGSIZE) {
3106 					/*
3107 					 * MTU discovery:
3108 					 * If we got a needfrag set the MTU
3109 					 * in the route to the suggested new
3110 					 * value (if given) and then notify.
3111 					 */
3112 					mtu = tcp6_next_pmtu(icmp6);
3113 
3114 					bzero(&inc, sizeof(inc));
3115 					inc.inc_fibnum = M_GETFIB(m);
3116 					inc.inc_flags |= INC_ISIPV6;
3117 					inc.inc6_faddr = *dst;
3118 					if (in6_setscope(&inc.inc6_faddr,
3119 						m->m_pkthdr.rcvif, NULL))
3120 						goto out;
3121 					/*
3122 					 * Only process the offered MTU if it
3123 					 * is smaller than the current one.
3124 					 */
3125 					if (mtu < tp->t_maxseg +
3126 					    sizeof (struct tcphdr) +
3127 					    sizeof (struct ip6_hdr)) {
3128 						tcp_hc_updatemtu(&inc, mtu);
3129 						tcp_mtudisc(inp, mtu);
3130 						ICMP6STAT_INC(icp6s_pmtuchg);
3131 					}
3132 				} else
3133 					inp = (*notify)(inp,
3134 					    inet6ctlerrmap[cmd]);
3135 			}
3136 		}
3137 	} else {
3138 		bzero(&inc, sizeof(inc));
3139 		inc.inc_fibnum = M_GETFIB(m);
3140 		inc.inc_flags |= INC_ISIPV6;
3141 		inc.inc_fport = t_ports.th_dport;
3142 		inc.inc_lport = t_ports.th_sport;
3143 		inc.inc6_faddr = *dst;
3144 		inc.inc6_laddr = ip6->ip6_src;
3145 		syncache_unreach(&inc, icmp_tcp_seq, port);
3146 	}
3147 out:
3148 	if (inp != NULL)
3149 		INP_WUNLOCK(inp);
3150 }
3151 
3152 void
3153 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3154 {
3155 	tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3156 }
3157 
3158 void
3159 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3160 {
3161 	struct ip6ctlparam *ip6cp;
3162 	struct mbuf *m;
3163 	struct udphdr *udp;
3164 	uint16_t port;
3165 
3166 	ip6cp = (struct ip6ctlparam *)d;
3167 	m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3168 	if (m == NULL) {
3169 		return;
3170 	}
3171 	udp = mtod(m, struct udphdr *);
3172 	if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3173 		return;
3174 	}
3175 	port = udp->uh_dport;
3176 	m_adj(m, sizeof(struct udphdr));
3177 	if ((m->m_flags & M_PKTHDR) == 0) {
3178 		ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3179 	}
3180 	/* Now call in to the normal handling code */
3181 	tcp6_ctlinput_with_port(cmd, sa, d, port);
3182 }
3183 
3184 #endif /* INET6 */
3185 
3186 static uint32_t
3187 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3188 {
3189 	SIPHASH_CTX ctx;
3190 	uint32_t hash[2];
3191 
3192 	KASSERT(len >= SIPHASH_KEY_LENGTH,
3193 	    ("%s: keylen %u too short ", __func__, len));
3194 	SipHash24_Init(&ctx);
3195 	SipHash_SetKey(&ctx, (uint8_t *)key);
3196 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3197 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3198 	switch (inc->inc_flags & INC_ISIPV6) {
3199 #ifdef INET
3200 	case 0:
3201 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3202 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3203 		break;
3204 #endif
3205 #ifdef INET6
3206 	case INC_ISIPV6:
3207 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3208 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3209 		break;
3210 #endif
3211 	}
3212 	SipHash_Final((uint8_t *)hash, &ctx);
3213 
3214 	return (hash[0] ^ hash[1]);
3215 }
3216 
3217 uint32_t
3218 tcp_new_ts_offset(struct in_conninfo *inc)
3219 {
3220 	struct in_conninfo inc_store, *local_inc;
3221 
3222 	if (!V_tcp_ts_offset_per_conn) {
3223 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3224 		inc_store.inc_lport = 0;
3225 		inc_store.inc_fport = 0;
3226 		local_inc = &inc_store;
3227 	} else {
3228 		local_inc = inc;
3229 	}
3230 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3231 	    sizeof(V_ts_offset_secret)));
3232 }
3233 
3234 /*
3235  * Following is where TCP initial sequence number generation occurs.
3236  *
3237  * There are two places where we must use initial sequence numbers:
3238  * 1.  In SYN-ACK packets.
3239  * 2.  In SYN packets.
3240  *
3241  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3242  * tcp_syncache.c for details.
3243  *
3244  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3245  * depends on this property.  In addition, these ISNs should be
3246  * unguessable so as to prevent connection hijacking.  To satisfy
3247  * the requirements of this situation, the algorithm outlined in
3248  * RFC 1948 is used, with only small modifications.
3249  *
3250  * Implementation details:
3251  *
3252  * Time is based off the system timer, and is corrected so that it
3253  * increases by one megabyte per second.  This allows for proper
3254  * recycling on high speed LANs while still leaving over an hour
3255  * before rollover.
3256  *
3257  * As reading the *exact* system time is too expensive to be done
3258  * whenever setting up a TCP connection, we increment the time
3259  * offset in two ways.  First, a small random positive increment
3260  * is added to isn_offset for each connection that is set up.
3261  * Second, the function tcp_isn_tick fires once per clock tick
3262  * and increments isn_offset as necessary so that sequence numbers
3263  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3264  * random positive increments serve only to ensure that the same
3265  * exact sequence number is never sent out twice (as could otherwise
3266  * happen when a port is recycled in less than the system tick
3267  * interval.)
3268  *
3269  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3270  * between seeding of isn_secret.  This is normally set to zero,
3271  * as reseeding should not be necessary.
3272  *
3273  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3274  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3275  * general, this means holding an exclusive (write) lock.
3276  */
3277 
3278 #define ISN_BYTES_PER_SECOND 1048576
3279 #define ISN_STATIC_INCREMENT 4096
3280 #define ISN_RANDOM_INCREMENT (4096 - 1)
3281 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3282 
3283 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3284 VNET_DEFINE_STATIC(int, isn_last);
3285 VNET_DEFINE_STATIC(int, isn_last_reseed);
3286 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3287 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3288 
3289 #define	V_isn_secret			VNET(isn_secret)
3290 #define	V_isn_last			VNET(isn_last)
3291 #define	V_isn_last_reseed		VNET(isn_last_reseed)
3292 #define	V_isn_offset			VNET(isn_offset)
3293 #define	V_isn_offset_old		VNET(isn_offset_old)
3294 
3295 tcp_seq
3296 tcp_new_isn(struct in_conninfo *inc)
3297 {
3298 	tcp_seq new_isn;
3299 	u_int32_t projected_offset;
3300 
3301 	ISN_LOCK();
3302 	/* Seed if this is the first use, reseed if requested. */
3303 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3304 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3305 		< (u_int)ticks))) {
3306 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3307 		V_isn_last_reseed = ticks;
3308 	}
3309 
3310 	/* Compute the hash and return the ISN. */
3311 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3312 	    sizeof(V_isn_secret));
3313 	V_isn_offset += ISN_STATIC_INCREMENT +
3314 		(arc4random() & ISN_RANDOM_INCREMENT);
3315 	if (ticks != V_isn_last) {
3316 		projected_offset = V_isn_offset_old +
3317 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3318 		if (SEQ_GT(projected_offset, V_isn_offset))
3319 			V_isn_offset = projected_offset;
3320 		V_isn_offset_old = V_isn_offset;
3321 		V_isn_last = ticks;
3322 	}
3323 	new_isn += V_isn_offset;
3324 	ISN_UNLOCK();
3325 	return (new_isn);
3326 }
3327 
3328 /*
3329  * When a specific ICMP unreachable message is received and the
3330  * connection state is SYN-SENT, drop the connection.  This behavior
3331  * is controlled by the icmp_may_rst sysctl.
3332  */
3333 struct inpcb *
3334 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3335 {
3336 	struct tcpcb *tp;
3337 
3338 	NET_EPOCH_ASSERT();
3339 	INP_WLOCK_ASSERT(inp);
3340 
3341 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3342 	    (inp->inp_flags & INP_DROPPED))
3343 		return (inp);
3344 
3345 	tp = intotcpcb(inp);
3346 	if (tp->t_state != TCPS_SYN_SENT)
3347 		return (inp);
3348 
3349 	if (IS_FASTOPEN(tp->t_flags))
3350 		tcp_fastopen_disable_path(tp);
3351 
3352 	tp = tcp_drop(tp, errno);
3353 	if (tp != NULL)
3354 		return (inp);
3355 	else
3356 		return (NULL);
3357 }
3358 
3359 /*
3360  * When `need fragmentation' ICMP is received, update our idea of the MSS
3361  * based on the new value. Also nudge TCP to send something, since we
3362  * know the packet we just sent was dropped.
3363  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3364  */
3365 static struct inpcb *
3366 tcp_mtudisc_notify(struct inpcb *inp, int error)
3367 {
3368 
3369 	return (tcp_mtudisc(inp, -1));
3370 }
3371 
3372 static struct inpcb *
3373 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3374 {
3375 	struct tcpcb *tp;
3376 	struct socket *so;
3377 
3378 	INP_WLOCK_ASSERT(inp);
3379 	if ((inp->inp_flags & INP_TIMEWAIT) ||
3380 	    (inp->inp_flags & INP_DROPPED))
3381 		return (inp);
3382 
3383 	tp = intotcpcb(inp);
3384 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3385 
3386 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3387 
3388 	so = inp->inp_socket;
3389 	SOCKBUF_LOCK(&so->so_snd);
3390 	/* If the mss is larger than the socket buffer, decrease the mss. */
3391 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
3392 		tp->t_maxseg = so->so_snd.sb_hiwat;
3393 	SOCKBUF_UNLOCK(&so->so_snd);
3394 
3395 	TCPSTAT_INC(tcps_mturesent);
3396 	tp->t_rtttime = 0;
3397 	tp->snd_nxt = tp->snd_una;
3398 	tcp_free_sackholes(tp);
3399 	tp->snd_recover = tp->snd_max;
3400 	if (tp->t_flags & TF_SACK_PERMIT)
3401 		EXIT_FASTRECOVERY(tp->t_flags);
3402 	if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3403 		/*
3404 		 * Conceptually the snd_nxt setting
3405 		 * and freeing sack holes should
3406 		 * be done by the default stacks
3407 		 * own tfb_tcp_mtu_chg().
3408 		 */
3409 		tp->t_fb->tfb_tcp_mtu_chg(tp);
3410 	}
3411 	if (tcp_output(tp) < 0)
3412 		return (NULL);
3413 	else
3414 		return (inp);
3415 }
3416 
3417 #ifdef INET
3418 /*
3419  * Look-up the routing entry to the peer of this inpcb.  If no route
3420  * is found and it cannot be allocated, then return 0.  This routine
3421  * is called by TCP routines that access the rmx structure and by
3422  * tcp_mss_update to get the peer/interface MTU.
3423  */
3424 uint32_t
3425 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3426 {
3427 	struct nhop_object *nh;
3428 	struct ifnet *ifp;
3429 	uint32_t maxmtu = 0;
3430 
3431 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3432 
3433 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
3434 		nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3435 		if (nh == NULL)
3436 			return (0);
3437 
3438 		ifp = nh->nh_ifp;
3439 		maxmtu = nh->nh_mtu;
3440 
3441 		/* Report additional interface capabilities. */
3442 		if (cap != NULL) {
3443 			if (ifp->if_capenable & IFCAP_TSO4 &&
3444 			    ifp->if_hwassist & CSUM_TSO) {
3445 				cap->ifcap |= CSUM_TSO;
3446 				cap->tsomax = ifp->if_hw_tsomax;
3447 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3448 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3449 			}
3450 		}
3451 	}
3452 	return (maxmtu);
3453 }
3454 #endif /* INET */
3455 
3456 #ifdef INET6
3457 uint32_t
3458 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3459 {
3460 	struct nhop_object *nh;
3461 	struct in6_addr dst6;
3462 	uint32_t scopeid;
3463 	struct ifnet *ifp;
3464 	uint32_t maxmtu = 0;
3465 
3466 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3467 
3468 	if (inc->inc_flags & INC_IPV6MINMTU)
3469 		return (IPV6_MMTU);
3470 
3471 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3472 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3473 		nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3474 		if (nh == NULL)
3475 			return (0);
3476 
3477 		ifp = nh->nh_ifp;
3478 		maxmtu = nh->nh_mtu;
3479 
3480 		/* Report additional interface capabilities. */
3481 		if (cap != NULL) {
3482 			if (ifp->if_capenable & IFCAP_TSO6 &&
3483 			    ifp->if_hwassist & CSUM_TSO) {
3484 				cap->ifcap |= CSUM_TSO;
3485 				cap->tsomax = ifp->if_hw_tsomax;
3486 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3487 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3488 			}
3489 		}
3490 	}
3491 
3492 	return (maxmtu);
3493 }
3494 
3495 /*
3496  * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack.
3497  *
3498  * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag.
3499  * The right place to do that is ip6_setpktopt() that has just been
3500  * executed.  By the way it just filled ip6po_minmtu for us.
3501  */
3502 void
3503 tcp6_use_min_mtu(struct tcpcb *tp)
3504 {
3505 	struct inpcb *inp = tp->t_inpcb;
3506 
3507 	INP_WLOCK_ASSERT(inp);
3508 	/*
3509 	 * In case of the IPV6_USE_MIN_MTU socket
3510 	 * option, the INC_IPV6MINMTU flag to announce
3511 	 * a corresponding MSS during the initial
3512 	 * handshake.  If the TCP connection is not in
3513 	 * the front states, just reduce the MSS being
3514 	 * used.  This avoids the sending of TCP
3515 	 * segments which will be fragmented at the
3516 	 * IPv6 layer.
3517 	 */
3518 	inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
3519 	if ((tp->t_state >= TCPS_SYN_SENT) &&
3520 	    (inp->inp_inc.inc_flags & INC_ISIPV6)) {
3521 		struct ip6_pktopts *opt;
3522 
3523 		opt = inp->in6p_outputopts;
3524 		if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL &&
3525 		    tp->t_maxseg > TCP6_MSS)
3526 			tp->t_maxseg = TCP6_MSS;
3527 	}
3528 }
3529 #endif /* INET6 */
3530 
3531 /*
3532  * Calculate effective SMSS per RFC5681 definition for a given TCP
3533  * connection at its current state, taking into account SACK and etc.
3534  */
3535 u_int
3536 tcp_maxseg(const struct tcpcb *tp)
3537 {
3538 	u_int optlen;
3539 
3540 	if (tp->t_flags & TF_NOOPT)
3541 		return (tp->t_maxseg);
3542 
3543 	/*
3544 	 * Here we have a simplified code from tcp_addoptions(),
3545 	 * without a proper loop, and having most of paddings hardcoded.
3546 	 * We might make mistakes with padding here in some edge cases,
3547 	 * but this is harmless, since result of tcp_maxseg() is used
3548 	 * only in cwnd and ssthresh estimations.
3549 	 */
3550 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3551 		if (tp->t_flags & TF_RCVD_TSTMP)
3552 			optlen = TCPOLEN_TSTAMP_APPA;
3553 		else
3554 			optlen = 0;
3555 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3556 		if (tp->t_flags & TF_SIGNATURE)
3557 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3558 #endif
3559 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3560 			optlen += TCPOLEN_SACKHDR;
3561 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3562 			optlen = PADTCPOLEN(optlen);
3563 		}
3564 	} else {
3565 		if (tp->t_flags & TF_REQ_TSTMP)
3566 			optlen = TCPOLEN_TSTAMP_APPA;
3567 		else
3568 			optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3569 		if (tp->t_flags & TF_REQ_SCALE)
3570 			optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3571 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3572 		if (tp->t_flags & TF_SIGNATURE)
3573 			optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3574 #endif
3575 		if (tp->t_flags & TF_SACK_PERMIT)
3576 			optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3577 	}
3578 #undef PAD
3579 	optlen = min(optlen, TCP_MAXOLEN);
3580 	return (tp->t_maxseg - optlen);
3581 }
3582 
3583 
3584 u_int
3585 tcp_fixed_maxseg(const struct tcpcb *tp)
3586 {
3587 	int optlen;
3588 
3589 	if (tp->t_flags & TF_NOOPT)
3590 		return (tp->t_maxseg);
3591 
3592 	/*
3593 	 * Here we have a simplified code from tcp_addoptions(),
3594 	 * without a proper loop, and having most of paddings hardcoded.
3595 	 * We only consider fixed options that we would send every
3596 	 * time I.e. SACK is not considered. This is important
3597 	 * for cc modules to figure out what the modulo of the
3598 	 * cwnd should be.
3599 	 */
3600 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
3601 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3602 		if (tp->t_flags & TF_RCVD_TSTMP)
3603 			optlen = TCPOLEN_TSTAMP_APPA;
3604 		else
3605 			optlen = 0;
3606 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3607 		if (tp->t_flags & TF_SIGNATURE)
3608 			optlen += PAD(TCPOLEN_SIGNATURE);
3609 #endif
3610 	} else {
3611 		if (tp->t_flags & TF_REQ_TSTMP)
3612 			optlen = TCPOLEN_TSTAMP_APPA;
3613 		else
3614 			optlen = PAD(TCPOLEN_MAXSEG);
3615 		if (tp->t_flags & TF_REQ_SCALE)
3616 			optlen += PAD(TCPOLEN_WINDOW);
3617 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3618 		if (tp->t_flags & TF_SIGNATURE)
3619 			optlen += PAD(TCPOLEN_SIGNATURE);
3620 #endif
3621 		if (tp->t_flags & TF_SACK_PERMIT)
3622 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
3623 	}
3624 #undef PAD
3625 	optlen = min(optlen, TCP_MAXOLEN);
3626 	return (tp->t_maxseg - optlen);
3627 }
3628 
3629 
3630 
3631 static int
3632 sysctl_drop(SYSCTL_HANDLER_ARGS)
3633 {
3634 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3635 	struct sockaddr_storage addrs[2];
3636 	struct inpcb *inp;
3637 	struct tcpcb *tp;
3638 	struct tcptw *tw;
3639 #ifdef INET
3640 	struct sockaddr_in *fin = NULL, *lin = NULL;
3641 #endif
3642 	struct epoch_tracker et;
3643 #ifdef INET6
3644 	struct sockaddr_in6 *fin6, *lin6;
3645 #endif
3646 	int error;
3647 
3648 	inp = NULL;
3649 #ifdef INET6
3650 	fin6 = lin6 = NULL;
3651 #endif
3652 	error = 0;
3653 
3654 	if (req->oldptr != NULL || req->oldlen != 0)
3655 		return (EINVAL);
3656 	if (req->newptr == NULL)
3657 		return (EPERM);
3658 	if (req->newlen < sizeof(addrs))
3659 		return (ENOMEM);
3660 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3661 	if (error)
3662 		return (error);
3663 
3664 	switch (addrs[0].ss_family) {
3665 #ifdef INET6
3666 	case AF_INET6:
3667 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3668 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3669 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3670 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3671 			return (EINVAL);
3672 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3673 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3674 				return (EINVAL);
3675 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3676 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3677 #ifdef INET
3678 			fin = (struct sockaddr_in *)&addrs[0];
3679 			lin = (struct sockaddr_in *)&addrs[1];
3680 #endif
3681 			break;
3682 		}
3683 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3684 		if (error)
3685 			return (error);
3686 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3687 		if (error)
3688 			return (error);
3689 		break;
3690 #endif
3691 #ifdef INET
3692 	case AF_INET:
3693 		fin = (struct sockaddr_in *)&addrs[0];
3694 		lin = (struct sockaddr_in *)&addrs[1];
3695 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3696 		    lin->sin_len != sizeof(struct sockaddr_in))
3697 			return (EINVAL);
3698 		break;
3699 #endif
3700 	default:
3701 		return (EINVAL);
3702 	}
3703 	NET_EPOCH_ENTER(et);
3704 	switch (addrs[0].ss_family) {
3705 #ifdef INET6
3706 	case AF_INET6:
3707 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3708 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3709 		    INPLOOKUP_WLOCKPCB, NULL);
3710 		break;
3711 #endif
3712 #ifdef INET
3713 	case AF_INET:
3714 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3715 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3716 		break;
3717 #endif
3718 	}
3719 	if (inp != NULL) {
3720 		if (inp->inp_flags & INP_TIMEWAIT) {
3721 			/*
3722 			 * XXXRW: There currently exists a state where an
3723 			 * inpcb is present, but its timewait state has been
3724 			 * discarded.  For now, don't allow dropping of this
3725 			 * type of inpcb.
3726 			 */
3727 			tw = intotw(inp);
3728 			if (tw != NULL)
3729 				tcp_twclose(tw, 0);
3730 			else
3731 				INP_WUNLOCK(inp);
3732 		} else if ((inp->inp_flags & INP_DROPPED) == 0 &&
3733 		    !SOLISTENING(inp->inp_socket)) {
3734 			tp = intotcpcb(inp);
3735 			tp = tcp_drop(tp, ECONNABORTED);
3736 			if (tp != NULL)
3737 				INP_WUNLOCK(inp);
3738 		} else
3739 			INP_WUNLOCK(inp);
3740 	} else
3741 		error = ESRCH;
3742 	NET_EPOCH_EXIT(et);
3743 	return (error);
3744 }
3745 
3746 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3747     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3748     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3749     "Drop TCP connection");
3750 
3751 static int
3752 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS)
3753 {
3754 	return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo,
3755 	    &tcp_ctloutput_set));
3756 }
3757 
3758 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt,
3759     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3760     CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "",
3761     "Set socket option for TCP endpoint");
3762 
3763 #ifdef KERN_TLS
3764 static int
3765 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3766 {
3767 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3768 	struct sockaddr_storage addrs[2];
3769 	struct inpcb *inp;
3770 #ifdef INET
3771 	struct sockaddr_in *fin = NULL, *lin = NULL;
3772 #endif
3773 	struct epoch_tracker et;
3774 #ifdef INET6
3775 	struct sockaddr_in6 *fin6, *lin6;
3776 #endif
3777 	int error;
3778 
3779 	inp = NULL;
3780 #ifdef INET6
3781 	fin6 = lin6 = NULL;
3782 #endif
3783 	error = 0;
3784 
3785 	if (req->oldptr != NULL || req->oldlen != 0)
3786 		return (EINVAL);
3787 	if (req->newptr == NULL)
3788 		return (EPERM);
3789 	if (req->newlen < sizeof(addrs))
3790 		return (ENOMEM);
3791 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3792 	if (error)
3793 		return (error);
3794 
3795 	switch (addrs[0].ss_family) {
3796 #ifdef INET6
3797 	case AF_INET6:
3798 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3799 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3800 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3801 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3802 			return (EINVAL);
3803 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3804 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3805 				return (EINVAL);
3806 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3807 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3808 #ifdef INET
3809 			fin = (struct sockaddr_in *)&addrs[0];
3810 			lin = (struct sockaddr_in *)&addrs[1];
3811 #endif
3812 			break;
3813 		}
3814 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3815 		if (error)
3816 			return (error);
3817 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3818 		if (error)
3819 			return (error);
3820 		break;
3821 #endif
3822 #ifdef INET
3823 	case AF_INET:
3824 		fin = (struct sockaddr_in *)&addrs[0];
3825 		lin = (struct sockaddr_in *)&addrs[1];
3826 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3827 		    lin->sin_len != sizeof(struct sockaddr_in))
3828 			return (EINVAL);
3829 		break;
3830 #endif
3831 	default:
3832 		return (EINVAL);
3833 	}
3834 	NET_EPOCH_ENTER(et);
3835 	switch (addrs[0].ss_family) {
3836 #ifdef INET6
3837 	case AF_INET6:
3838 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3839 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3840 		    INPLOOKUP_WLOCKPCB, NULL);
3841 		break;
3842 #endif
3843 #ifdef INET
3844 	case AF_INET:
3845 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3846 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3847 		break;
3848 #endif
3849 	}
3850 	NET_EPOCH_EXIT(et);
3851 	if (inp != NULL) {
3852 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3853 		    inp->inp_socket == NULL) {
3854 			error = ECONNRESET;
3855 			INP_WUNLOCK(inp);
3856 		} else {
3857 			struct socket *so;
3858 
3859 			so = inp->inp_socket;
3860 			soref(so);
3861 			error = ktls_set_tx_mode(so,
3862 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3863 			INP_WUNLOCK(inp);
3864 			sorele(so);
3865 		}
3866 	} else
3867 		error = ESRCH;
3868 	return (error);
3869 }
3870 
3871 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3872     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3873     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3874     "Switch TCP connection to SW TLS");
3875 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3876     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3877     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3878     "Switch TCP connection to ifnet TLS");
3879 #endif
3880 
3881 /*
3882  * Generate a standardized TCP log line for use throughout the
3883  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3884  * allow use in the interrupt context.
3885  *
3886  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3887  * NB: The function may return NULL if memory allocation failed.
3888  *
3889  * Due to header inclusion and ordering limitations the struct ip
3890  * and ip6_hdr pointers have to be passed as void pointers.
3891  */
3892 char *
3893 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3894     const void *ip6hdr)
3895 {
3896 
3897 	/* Is logging enabled? */
3898 	if (V_tcp_log_in_vain == 0)
3899 		return (NULL);
3900 
3901 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3902 }
3903 
3904 char *
3905 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3906     const void *ip6hdr)
3907 {
3908 
3909 	/* Is logging enabled? */
3910 	if (tcp_log_debug == 0)
3911 		return (NULL);
3912 
3913 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3914 }
3915 
3916 static char *
3917 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr,
3918     const void *ip6hdr)
3919 {
3920 	char *s, *sp;
3921 	size_t size;
3922 #ifdef INET
3923 	const struct ip *ip = (const struct ip *)ip4hdr;
3924 #endif
3925 #ifdef INET6
3926 	const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr;
3927 #endif /* INET6 */
3928 
3929 	/*
3930 	 * The log line looks like this:
3931 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3932 	 */
3933 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3934 	    sizeof(PRINT_TH_FLAGS) + 1 +
3935 #ifdef INET6
3936 	    2 * INET6_ADDRSTRLEN;
3937 #else
3938 	    2 * INET_ADDRSTRLEN;
3939 #endif /* INET6 */
3940 
3941 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3942 	if (s == NULL)
3943 		return (NULL);
3944 
3945 	strcat(s, "TCP: [");
3946 	sp = s + strlen(s);
3947 
3948 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3949 		inet_ntoa_r(inc->inc_faddr, sp);
3950 		sp = s + strlen(s);
3951 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3952 		sp = s + strlen(s);
3953 		inet_ntoa_r(inc->inc_laddr, sp);
3954 		sp = s + strlen(s);
3955 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3956 #ifdef INET6
3957 	} else if (inc) {
3958 		ip6_sprintf(sp, &inc->inc6_faddr);
3959 		sp = s + strlen(s);
3960 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3961 		sp = s + strlen(s);
3962 		ip6_sprintf(sp, &inc->inc6_laddr);
3963 		sp = s + strlen(s);
3964 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3965 	} else if (ip6 && th) {
3966 		ip6_sprintf(sp, &ip6->ip6_src);
3967 		sp = s + strlen(s);
3968 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3969 		sp = s + strlen(s);
3970 		ip6_sprintf(sp, &ip6->ip6_dst);
3971 		sp = s + strlen(s);
3972 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3973 #endif /* INET6 */
3974 #ifdef INET
3975 	} else if (ip && th) {
3976 		inet_ntoa_r(ip->ip_src, sp);
3977 		sp = s + strlen(s);
3978 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3979 		sp = s + strlen(s);
3980 		inet_ntoa_r(ip->ip_dst, sp);
3981 		sp = s + strlen(s);
3982 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3983 #endif /* INET */
3984 	} else {
3985 		free(s, M_TCPLOG);
3986 		return (NULL);
3987 	}
3988 	sp = s + strlen(s);
3989 	if (th)
3990 		sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS);
3991 	if (*(s + size - 1) != '\0')
3992 		panic("%s: string too long", __func__);
3993 	return (s);
3994 }
3995 
3996 /*
3997  * A subroutine which makes it easy to track TCP state changes with DTrace.
3998  * This function shouldn't be called for t_state initializations that don't
3999  * correspond to actual TCP state transitions.
4000  */
4001 void
4002 tcp_state_change(struct tcpcb *tp, int newstate)
4003 {
4004 #if defined(KDTRACE_HOOKS)
4005 	int pstate = tp->t_state;
4006 #endif
4007 
4008 	TCPSTATES_DEC(tp->t_state);
4009 	TCPSTATES_INC(newstate);
4010 	tp->t_state = newstate;
4011 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
4012 }
4013 
4014 /*
4015  * Create an external-format (``xtcpcb'') structure using the information in
4016  * the kernel-format tcpcb structure pointed to by tp.  This is done to
4017  * reduce the spew of irrelevant information over this interface, to isolate
4018  * user code from changes in the kernel structure, and potentially to provide
4019  * information-hiding if we decide that some of this information should be
4020  * hidden from users.
4021  */
4022 void
4023 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
4024 {
4025 	struct tcpcb *tp = intotcpcb(inp);
4026 	struct tcptw *tw = intotw(inp);
4027 	sbintime_t now;
4028 
4029 	bzero(xt, sizeof(*xt));
4030 	if (inp->inp_flags & INP_TIMEWAIT) {
4031 		xt->t_state = TCPS_TIME_WAIT;
4032 		xt->xt_encaps_port = tw->t_port;
4033 	} else {
4034 		xt->t_state = tp->t_state;
4035 		xt->t_logstate = tp->t_logstate;
4036 		xt->t_flags = tp->t_flags;
4037 		xt->t_sndzerowin = tp->t_sndzerowin;
4038 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
4039 		xt->t_rcvoopack = tp->t_rcvoopack;
4040 		xt->t_rcv_wnd = tp->rcv_wnd;
4041 		xt->t_snd_wnd = tp->snd_wnd;
4042 		xt->t_snd_cwnd = tp->snd_cwnd;
4043 		xt->t_snd_ssthresh = tp->snd_ssthresh;
4044 		xt->t_dsack_bytes = tp->t_dsack_bytes;
4045 		xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes;
4046 		xt->t_dsack_pack = tp->t_dsack_pack;
4047 		xt->t_maxseg = tp->t_maxseg;
4048 		xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4049 			     (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4050 
4051 		now = getsbinuptime();
4052 #define	COPYTIMER(ttt)	do {						\
4053 		if (callout_active(&tp->t_timers->ttt))			\
4054 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
4055 			    SBT_1MS;					\
4056 		else							\
4057 			xt->ttt = 0;					\
4058 } while (0)
4059 		COPYTIMER(tt_delack);
4060 		COPYTIMER(tt_rexmt);
4061 		COPYTIMER(tt_persist);
4062 		COPYTIMER(tt_keep);
4063 		COPYTIMER(tt_2msl);
4064 #undef COPYTIMER
4065 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4066 
4067 		xt->xt_encaps_port = tp->t_port;
4068 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4069 		    TCP_FUNCTION_NAME_LEN_MAX);
4070 		bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4071 		    TCP_CA_NAME_MAX);
4072 #ifdef TCP_BLACKBOX
4073 		(void)tcp_log_get_id(tp, xt->xt_logid);
4074 #endif
4075 	}
4076 
4077 	xt->xt_len = sizeof(struct xtcpcb);
4078 	in_pcbtoxinpcb(inp, &xt->xt_inp);
4079 	if (inp->inp_socket == NULL)
4080 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4081 }
4082 
4083 void
4084 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4085 {
4086 	uint32_t bit, i;
4087 
4088 	if ((tp == NULL) ||
4089 	    (status > TCP_EI_STATUS_MAX_VALUE) ||
4090 	    (status == 0)) {
4091 		/* Invalid */
4092 		return;
4093 	}
4094 	if (status > (sizeof(uint32_t) * 8)) {
4095 		/* Should this be a KASSERT? */
4096 		return;
4097 	}
4098 	bit = 1U << (status - 1);
4099 	if (bit & tp->t_end_info_status) {
4100 		/* already logged */
4101 		return;
4102 	}
4103 	for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4104 		if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4105 			tp->t_end_info_bytes[i] = status;
4106 			tp->t_end_info_status |= bit;
4107 			break;
4108 		}
4109 	}
4110 }
4111 
4112 int
4113 tcp_can_enable_pacing(void)
4114 {
4115 
4116 	if ((tcp_pacing_limit == -1) ||
4117 	    (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4118 		atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4119 		shadow_num_connections = number_of_tcp_connections_pacing;
4120 		return (1);
4121 	} else {
4122 		return (0);
4123 	}
4124 }
4125 
4126 static uint8_t tcp_pacing_warning = 0;
4127 
4128 void
4129 tcp_decrement_paced_conn(void)
4130 {
4131 	uint32_t ret;
4132 
4133 	ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4134 	shadow_num_connections = number_of_tcp_connections_pacing;
4135 	KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4136 	if (ret == 0) {
4137 		if (tcp_pacing_limit != -1) {
4138 			printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4139 			tcp_pacing_limit = 0;
4140 		} else if (tcp_pacing_warning == 0) {
4141 			printf("Warning pacing count is invalid, invalid decrement\n");
4142 			tcp_pacing_warning = 1;
4143 		}
4144 	}
4145 }
4146