xref: /freebsd/sys/netinet/tcp_subr.c (revision d0ba1baed3f6e4936a0c1b89c25f6c59168ef6de)
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_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/eventhandler.h>
46 #ifdef TCP_HHOOK
47 #include <sys/hhook.h>
48 #endif
49 #include <sys/kernel.h>
50 #ifdef TCP_HHOOK
51 #include <sys/khelp.h>
52 #endif
53 #include <sys/sysctl.h>
54 #include <sys/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/refcount.h>
57 #include <sys/mbuf.h>
58 #ifdef INET6
59 #include <sys/domain.h>
60 #endif
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/sdt.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/random.h>
68 
69 #include <vm/uma.h>
70 
71 #include <net/route.h>
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/vnet.h>
75 
76 #include <netinet/in.h>
77 #include <netinet/in_fib.h>
78 #include <netinet/in_kdtrace.h>
79 #include <netinet/in_pcb.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/in_var.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/ip_var.h>
85 #ifdef INET6
86 #include <netinet/icmp6.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_fib.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet6/nd6.h>
93 #endif
94 
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
98 #include <netinet/tcp_timer.h>
99 #include <netinet/tcp_var.h>
100 #include <netinet/tcp_log_buf.h>
101 #include <netinet/tcp_syncache.h>
102 #include <netinet/tcp_hpts.h>
103 #include <netinet/cc/cc.h>
104 #ifdef INET6
105 #include <netinet6/tcp6_var.h>
106 #endif
107 #include <netinet/tcpip.h>
108 #include <netinet/tcp_fastopen.h>
109 #ifdef TCPPCAP
110 #include <netinet/tcp_pcap.h>
111 #endif
112 #ifdef TCPDEBUG
113 #include <netinet/tcp_debug.h>
114 #endif
115 #ifdef INET6
116 #include <netinet6/ip6protosw.h>
117 #endif
118 #ifdef TCP_OFFLOAD
119 #include <netinet/tcp_offload.h>
120 #endif
121 
122 #include <netipsec/ipsec_support.h>
123 
124 #include <machine/in_cksum.h>
125 #include <sys/md5.h>
126 
127 #include <security/mac/mac_framework.h>
128 
129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 #ifdef INET6
131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
132 #endif
133 
134 struct rwlock tcp_function_lock;
135 
136 static int
137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
138 {
139 	int error, new;
140 
141 	new = V_tcp_mssdflt;
142 	error = sysctl_handle_int(oidp, &new, 0, req);
143 	if (error == 0 && req->newptr) {
144 		if (new < TCP_MINMSS)
145 			error = EINVAL;
146 		else
147 			V_tcp_mssdflt = new;
148 	}
149 	return (error);
150 }
151 
152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154     &sysctl_net_inet_tcp_mss_check, "I",
155     "Default TCP Maximum Segment Size");
156 
157 #ifdef INET6
158 static int
159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
160 {
161 	int error, new;
162 
163 	new = V_tcp_v6mssdflt;
164 	error = sysctl_handle_int(oidp, &new, 0, req);
165 	if (error == 0 && req->newptr) {
166 		if (new < TCP_MINMSS)
167 			error = EINVAL;
168 		else
169 			V_tcp_v6mssdflt = new;
170 	}
171 	return (error);
172 }
173 
174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176     &sysctl_net_inet_tcp_mss_v6_check, "I",
177    "Default TCP Maximum Segment Size for IPv6");
178 #endif /* INET6 */
179 
180 /*
181  * Minimum MSS we accept and use. This prevents DoS attacks where
182  * we are forced to a ridiculous low MSS like 20 and send hundreds
183  * of packets instead of one. The effect scales with the available
184  * bandwidth and quickly saturates the CPU and network interface
185  * with packet generation and sending. Set to zero to disable MINMSS
186  * checking. This setting prevents us from sending too small packets.
187  */
188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190      &VNET_NAME(tcp_minmss), 0,
191     "Minimum TCP Maximum Segment Size");
192 
193 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195     &VNET_NAME(tcp_do_rfc1323), 0,
196     "Enable rfc1323 (high performance TCP) extensions");
197 
198 static int	tcp_log_debug = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 
202 static int	tcp_tcbhashsize;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 
206 static int	do_tcpdrain = 1;
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208     "Enable tcp_drain routine for extra help when low on mbufs");
209 
210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 
213 static VNET_DEFINE(int, icmp_may_rst) = 1;
214 #define	V_icmp_may_rst			VNET(icmp_may_rst)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216     &VNET_NAME(icmp_may_rst), 0,
217     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 
219 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
220 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_isn_reseed_interval), 0,
223     "Seconds between reseeding of ISN secret");
224 
225 static int	tcp_soreceive_stream;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 
229 VNET_DEFINE(uma_zone_t, sack_hole_zone);
230 #define	V_sack_hole_zone		VNET(sack_hole_zone)
231 
232 #ifdef TCP_HHOOK
233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
234 #endif
235 
236 static int	tcp_default_fb_init(struct tcpcb *tp);
237 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
238 static int	tcp_default_handoff_ok(struct tcpcb *tp);
239 static struct inpcb *tcp_notify(struct inpcb *, int);
240 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
241 static void tcp_mtudisc(struct inpcb *, int);
242 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
243 		    void *ip4hdr, const void *ip6hdr);
244 
245 
246 static struct tcp_function_block tcp_def_funcblk = {
247 	.tfb_tcp_block_name = "freebsd",
248 	.tfb_tcp_output = tcp_output,
249 	.tfb_tcp_do_segment = tcp_do_segment,
250 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
251 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
252 	.tfb_tcp_fb_init = tcp_default_fb_init,
253 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
254 };
255 
256 int t_functions_inited = 0;
257 static int tcp_fb_cnt = 0;
258 struct tcp_funchead t_functions;
259 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
260 
261 static void
262 init_tcp_functions(void)
263 {
264 	if (t_functions_inited == 0) {
265 		TAILQ_INIT(&t_functions);
266 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
267 		t_functions_inited = 1;
268 	}
269 }
270 
271 static struct tcp_function_block *
272 find_tcp_functions_locked(struct tcp_function_set *fs)
273 {
274 	struct tcp_function *f;
275 	struct tcp_function_block *blk=NULL;
276 
277 	TAILQ_FOREACH(f, &t_functions, tf_next) {
278 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
279 			blk = f->tf_fb;
280 			break;
281 		}
282 	}
283 	return(blk);
284 }
285 
286 static struct tcp_function_block *
287 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
288 {
289 	struct tcp_function_block *rblk=NULL;
290 	struct tcp_function *f;
291 
292 	TAILQ_FOREACH(f, &t_functions, tf_next) {
293 		if (f->tf_fb == blk) {
294 			rblk = blk;
295 			if (s) {
296 				*s = f;
297 			}
298 			break;
299 		}
300 	}
301 	return (rblk);
302 }
303 
304 struct tcp_function_block *
305 find_and_ref_tcp_functions(struct tcp_function_set *fs)
306 {
307 	struct tcp_function_block *blk;
308 
309 	rw_rlock(&tcp_function_lock);
310 	blk = find_tcp_functions_locked(fs);
311 	if (blk)
312 		refcount_acquire(&blk->tfb_refcnt);
313 	rw_runlock(&tcp_function_lock);
314 	return(blk);
315 }
316 
317 struct tcp_function_block *
318 find_and_ref_tcp_fb(struct tcp_function_block *blk)
319 {
320 	struct tcp_function_block *rblk;
321 
322 	rw_rlock(&tcp_function_lock);
323 	rblk = find_tcp_fb_locked(blk, NULL);
324 	if (rblk)
325 		refcount_acquire(&rblk->tfb_refcnt);
326 	rw_runlock(&tcp_function_lock);
327 	return(rblk);
328 }
329 
330 static struct tcp_function_block *
331 find_and_ref_tcp_default_fb(void)
332 {
333 	struct tcp_function_block *rblk;
334 
335 	rw_rlock(&tcp_function_lock);
336 	rblk = tcp_func_set_ptr;
337 	refcount_acquire(&rblk->tfb_refcnt);
338 	rw_runlock(&tcp_function_lock);
339 	return (rblk);
340 }
341 
342 void
343 tcp_switch_back_to_default(struct tcpcb *tp)
344 {
345 	struct tcp_function_block *tfb;
346 
347 	KASSERT(tp->t_fb != &tcp_def_funcblk,
348 	    ("%s: called by the built-in default stack", __func__));
349 
350 	/*
351 	 * Release the old stack. This function will either find a new one
352 	 * or panic.
353 	 */
354 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
355 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
356 	refcount_release(&tp->t_fb->tfb_refcnt);
357 
358 	/*
359 	 * Now, we'll find a new function block to use.
360 	 * Start by trying the current user-selected
361 	 * default, unless this stack is the user-selected
362 	 * default.
363 	 */
364 	tfb = find_and_ref_tcp_default_fb();
365 	if (tfb == tp->t_fb) {
366 		refcount_release(&tfb->tfb_refcnt);
367 		tfb = NULL;
368 	}
369 	/* Does the stack accept this connection? */
370 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
371 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
372 		refcount_release(&tfb->tfb_refcnt);
373 		tfb = NULL;
374 	}
375 	/* Try to use that stack. */
376 	if (tfb != NULL) {
377 		/* Initialize the new stack. If it succeeds, we are done. */
378 		tp->t_fb = tfb;
379 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
380 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
381 			return;
382 
383 		/*
384 		 * Initialization failed. Release the reference count on
385 		 * the stack.
386 		 */
387 		refcount_release(&tfb->tfb_refcnt);
388 	}
389 
390 	/*
391 	 * If that wasn't feasible, use the built-in default
392 	 * stack which is not allowed to reject anyone.
393 	 */
394 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
395 	if (tfb == NULL) {
396 		/* there always should be a default */
397 		panic("Can't refer to tcp_def_funcblk");
398 	}
399 	if (tfb->tfb_tcp_handoff_ok != NULL) {
400 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
401 			/* The default stack cannot say no */
402 			panic("Default stack rejects a new session?");
403 		}
404 	}
405 	tp->t_fb = tfb;
406 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
407 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
408 		/* The default stack cannot fail */
409 		panic("Default stack initialization failed");
410 	}
411 }
412 
413 static int
414 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
415 {
416 	int error=ENOENT;
417 	struct tcp_function_set fs;
418 	struct tcp_function_block *blk;
419 
420 	memset(&fs, 0, sizeof(fs));
421 	rw_rlock(&tcp_function_lock);
422 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
423 	if (blk) {
424 		/* Found him */
425 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
426 		fs.pcbcnt = blk->tfb_refcnt;
427 	}
428 	rw_runlock(&tcp_function_lock);
429 	error = sysctl_handle_string(oidp, fs.function_set_name,
430 				     sizeof(fs.function_set_name), req);
431 
432 	/* Check for error or no change */
433 	if (error != 0 || req->newptr == NULL)
434 		return(error);
435 
436 	rw_wlock(&tcp_function_lock);
437 	blk = find_tcp_functions_locked(&fs);
438 	if ((blk == NULL) ||
439 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
440 		error = ENOENT;
441 		goto done;
442 	}
443 	tcp_func_set_ptr = blk;
444 done:
445 	rw_wunlock(&tcp_function_lock);
446 	return (error);
447 }
448 
449 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
450 	    CTLTYPE_STRING | CTLFLAG_RW,
451 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
452 	    "Set/get the default TCP functions");
453 
454 static int
455 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
456 {
457 	int error, cnt, linesz;
458 	struct tcp_function *f;
459 	char *buffer, *cp;
460 	size_t bufsz, outsz;
461 	bool alias;
462 
463 	cnt = 0;
464 	rw_rlock(&tcp_function_lock);
465 	TAILQ_FOREACH(f, &t_functions, tf_next) {
466 		cnt++;
467 	}
468 	rw_runlock(&tcp_function_lock);
469 
470 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
471 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
472 
473 	error = 0;
474 	cp = buffer;
475 
476 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
477 	    "Alias", "PCB count");
478 	cp += linesz;
479 	bufsz -= linesz;
480 	outsz = linesz;
481 
482 	rw_rlock(&tcp_function_lock);
483 	TAILQ_FOREACH(f, &t_functions, tf_next) {
484 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
485 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
486 		    f->tf_fb->tfb_tcp_block_name,
487 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
488 		    alias ? f->tf_name : "-",
489 		    f->tf_fb->tfb_refcnt);
490 		if (linesz >= bufsz) {
491 			error = EOVERFLOW;
492 			break;
493 		}
494 		cp += linesz;
495 		bufsz -= linesz;
496 		outsz += linesz;
497 	}
498 	rw_runlock(&tcp_function_lock);
499 	if (error == 0)
500 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
501 	free(buffer, M_TEMP);
502 	return (error);
503 }
504 
505 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
506 	    CTLTYPE_STRING|CTLFLAG_RD,
507 	    NULL, 0, sysctl_net_inet_list_available, "A",
508 	    "list available TCP Function sets");
509 
510 /*
511  * Exports one (struct tcp_function_info) for each alias/name.
512  */
513 static int
514 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
515 {
516 	int cnt, error;
517 	struct tcp_function *f;
518 	struct tcp_function_info tfi;
519 
520 	/*
521 	 * We don't allow writes.
522 	 */
523 	if (req->newptr != NULL)
524 		return (EINVAL);
525 
526 	/*
527 	 * Wire the old buffer so we can directly copy the functions to
528 	 * user space without dropping the lock.
529 	 */
530 	if (req->oldptr != NULL) {
531 		error = sysctl_wire_old_buffer(req, 0);
532 		if (error)
533 			return (error);
534 	}
535 
536 	/*
537 	 * Walk the list and copy out matching entries. If INVARIANTS
538 	 * is compiled in, also walk the list to verify the length of
539 	 * the list matches what we have recorded.
540 	 */
541 	rw_rlock(&tcp_function_lock);
542 
543 	cnt = 0;
544 #ifndef INVARIANTS
545 	if (req->oldptr == NULL) {
546 		cnt = tcp_fb_cnt;
547 		goto skip_loop;
548 	}
549 #endif
550 	TAILQ_FOREACH(f, &t_functions, tf_next) {
551 #ifdef INVARIANTS
552 		cnt++;
553 #endif
554 		if (req->oldptr != NULL) {
555 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
556 			tfi.tfi_id = f->tf_fb->tfb_id;
557 			(void)strncpy(tfi.tfi_alias, f->tf_name,
558 			    TCP_FUNCTION_NAME_LEN_MAX);
559 			tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
560 			(void)strncpy(tfi.tfi_name,
561 			    f->tf_fb->tfb_tcp_block_name,
562 			    TCP_FUNCTION_NAME_LEN_MAX);
563 			tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
564 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
565 			/*
566 			 * Don't stop on error, as that is the
567 			 * mechanism we use to accumulate length
568 			 * information if the buffer was too short.
569 			 */
570 		}
571 	}
572 	KASSERT(cnt == tcp_fb_cnt,
573 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
574 #ifndef INVARIANTS
575 skip_loop:
576 #endif
577 	rw_runlock(&tcp_function_lock);
578 	if (req->oldptr == NULL)
579 		error = SYSCTL_OUT(req, NULL,
580 		    (cnt + 1) * sizeof(struct tcp_function_info));
581 
582 	return (error);
583 }
584 
585 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
586 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
587 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
588 	    "List TCP function block name-to-ID mappings");
589 
590 /*
591  * tfb_tcp_handoff_ok() function for the default stack.
592  * Note that we'll basically try to take all comers.
593  */
594 static int
595 tcp_default_handoff_ok(struct tcpcb *tp)
596 {
597 
598 	return (0);
599 }
600 
601 /*
602  * tfb_tcp_fb_init() function for the default stack.
603  *
604  * This handles making sure we have appropriate timers set if you are
605  * transitioning a socket that has some amount of setup done.
606  *
607  * The init() fuction from the default can *never* return non-zero i.e.
608  * it is required to always succeed since it is the stack of last resort!
609  */
610 static int
611 tcp_default_fb_init(struct tcpcb *tp)
612 {
613 
614 	struct socket *so;
615 
616 	INP_WLOCK_ASSERT(tp->t_inpcb);
617 
618 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
619 	    ("%s: connection %p in unexpected state %d", __func__, tp,
620 	    tp->t_state));
621 
622 	/*
623 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
624 	 * know what to do for unexpected states (which includes TIME_WAIT).
625 	 */
626 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
627 		return (0);
628 
629 	/*
630 	 * Make sure some kind of transmission timer is set if there is
631 	 * outstanding data.
632 	 */
633 	so = tp->t_inpcb->inp_socket;
634 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
635 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
636 	    tcp_timer_active(tp, TT_PERSIST))) {
637 		/*
638 		 * If the session has established and it looks like it should
639 		 * be in the persist state, set the persist timer. Otherwise,
640 		 * set the retransmit timer.
641 		 */
642 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
643 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
644 		    (int32_t)sbavail(&so->so_snd))
645 			tcp_setpersist(tp);
646 		else
647 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
648 	}
649 
650 	/* All non-embryonic sessions get a keepalive timer. */
651 	if (!tcp_timer_active(tp, TT_KEEP))
652 		tcp_timer_activate(tp, TT_KEEP,
653 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
654 		    TP_KEEPINIT(tp));
655 
656 	return (0);
657 }
658 
659 /*
660  * tfb_tcp_fb_fini() function for the default stack.
661  *
662  * This changes state as necessary (or prudent) to prepare for another stack
663  * to assume responsibility for the connection.
664  */
665 static void
666 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
667 {
668 
669 	INP_WLOCK_ASSERT(tp->t_inpcb);
670 	return;
671 }
672 
673 /*
674  * Target size of TCP PCB hash tables. Must be a power of two.
675  *
676  * Note that this can be overridden by the kernel environment
677  * variable net.inet.tcp.tcbhashsize
678  */
679 #ifndef TCBHASHSIZE
680 #define TCBHASHSIZE	0
681 #endif
682 
683 /*
684  * XXX
685  * Callouts should be moved into struct tcp directly.  They are currently
686  * separate because the tcpcb structure is exported to userland for sysctl
687  * parsing purposes, which do not know about callouts.
688  */
689 struct tcpcb_mem {
690 	struct	tcpcb		tcb;
691 	struct	tcp_timer	tt;
692 	struct	cc_var		ccv;
693 #ifdef TCP_HHOOK
694 	struct	osd		osd;
695 #endif
696 };
697 
698 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
699 #define	V_tcpcb_zone			VNET(tcpcb_zone)
700 
701 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
702 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
703 
704 static struct mtx isn_mtx;
705 
706 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
707 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
708 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
709 
710 /*
711  * TCP initialization.
712  */
713 static void
714 tcp_zone_change(void *tag)
715 {
716 
717 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
718 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
719 	tcp_tw_zone_change();
720 }
721 
722 static int
723 tcp_inpcb_init(void *mem, int size, int flags)
724 {
725 	struct inpcb *inp = mem;
726 
727 	INP_LOCK_INIT(inp, "inp", "tcpinp");
728 	return (0);
729 }
730 
731 /*
732  * Take a value and get the next power of 2 that doesn't overflow.
733  * Used to size the tcp_inpcb hash buckets.
734  */
735 static int
736 maketcp_hashsize(int size)
737 {
738 	int hashsize;
739 
740 	/*
741 	 * auto tune.
742 	 * get the next power of 2 higher than maxsockets.
743 	 */
744 	hashsize = 1 << fls(size);
745 	/* catch overflow, and just go one power of 2 smaller */
746 	if (hashsize < size) {
747 		hashsize = 1 << (fls(size) - 1);
748 	}
749 	return (hashsize);
750 }
751 
752 static volatile int next_tcp_stack_id = 1;
753 
754 /*
755  * Register a TCP function block with the name provided in the names
756  * array.  (Note that this function does NOT automatically register
757  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
758  * explicitly include blk->tfb_tcp_block_name in the list of names if
759  * you wish to register the stack with that name.)
760  *
761  * Either all name registrations will succeed or all will fail.  If
762  * a name registration fails, the function will update the num_names
763  * argument to point to the array index of the name that encountered
764  * the failure.
765  *
766  * Returns 0 on success, or an error code on failure.
767  */
768 int
769 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
770     const char *names[], int *num_names)
771 {
772 	struct tcp_function *n;
773 	struct tcp_function_set fs;
774 	int error, i;
775 
776 	KASSERT(names != NULL && *num_names > 0,
777 	    ("%s: Called with 0-length name list", __func__));
778 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
779 
780 	if (t_functions_inited == 0) {
781 		init_tcp_functions();
782 	}
783 	if ((blk->tfb_tcp_output == NULL) ||
784 	    (blk->tfb_tcp_do_segment == NULL) ||
785 	    (blk->tfb_tcp_ctloutput == NULL) ||
786 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
787 		/*
788 		 * These functions are required and you
789 		 * need a name.
790 		 */
791 		*num_names = 0;
792 		return (EINVAL);
793 	}
794 	if (blk->tfb_tcp_timer_stop_all ||
795 	    blk->tfb_tcp_timer_activate ||
796 	    blk->tfb_tcp_timer_active ||
797 	    blk->tfb_tcp_timer_stop) {
798 		/*
799 		 * If you define one timer function you
800 		 * must have them all.
801 		 */
802 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
803 		    (blk->tfb_tcp_timer_activate == NULL) ||
804 		    (blk->tfb_tcp_timer_active == NULL) ||
805 		    (blk->tfb_tcp_timer_stop == NULL)) {
806 			*num_names = 0;
807 			return (EINVAL);
808 		}
809 	}
810 
811 	refcount_init(&blk->tfb_refcnt, 0);
812 	blk->tfb_flags = 0;
813 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
814 	for (i = 0; i < *num_names; i++) {
815 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
816 		if (n == NULL) {
817 			error = ENOMEM;
818 			goto cleanup;
819 		}
820 		n->tf_fb = blk;
821 
822 		(void)strncpy(fs.function_set_name, names[i],
823 		    TCP_FUNCTION_NAME_LEN_MAX);
824 		fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
825 		rw_wlock(&tcp_function_lock);
826 		if (find_tcp_functions_locked(&fs) != NULL) {
827 			/* Duplicate name space not allowed */
828 			rw_wunlock(&tcp_function_lock);
829 			free(n, M_TCPFUNCTIONS);
830 			error = EALREADY;
831 			goto cleanup;
832 		}
833 		(void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
834 		n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
835 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
836 		tcp_fb_cnt++;
837 		rw_wunlock(&tcp_function_lock);
838 	}
839 	return(0);
840 
841 cleanup:
842 	/*
843 	 * Deregister the names we just added. Because registration failed
844 	 * for names[i], we don't need to deregister that name.
845 	 */
846 	*num_names = i;
847 	rw_wlock(&tcp_function_lock);
848 	while (--i >= 0) {
849 		TAILQ_FOREACH(n, &t_functions, tf_next) {
850 			if (!strncmp(n->tf_name, names[i],
851 			    TCP_FUNCTION_NAME_LEN_MAX)) {
852 				TAILQ_REMOVE(&t_functions, n, tf_next);
853 				tcp_fb_cnt--;
854 				n->tf_fb = NULL;
855 				free(n, M_TCPFUNCTIONS);
856 				break;
857 			}
858 		}
859 	}
860 	rw_wunlock(&tcp_function_lock);
861 	return (error);
862 }
863 
864 /*
865  * Register a TCP function block using the name provided in the name
866  * argument.
867  *
868  * Returns 0 on success, or an error code on failure.
869  */
870 int
871 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
872     int wait)
873 {
874 	const char *name_list[1];
875 	int num_names, rv;
876 
877 	num_names = 1;
878 	if (name != NULL)
879 		name_list[0] = name;
880 	else
881 		name_list[0] = blk->tfb_tcp_block_name;
882 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
883 	return (rv);
884 }
885 
886 /*
887  * Register a TCP function block using the name defined in
888  * blk->tfb_tcp_block_name.
889  *
890  * Returns 0 on success, or an error code on failure.
891  */
892 int
893 register_tcp_functions(struct tcp_function_block *blk, int wait)
894 {
895 
896 	return (register_tcp_functions_as_name(blk, NULL, wait));
897 }
898 
899 /*
900  * Deregister all names associated with a function block. This
901  * functionally removes the function block from use within the system.
902  *
903  * When called with a true quiesce argument, mark the function block
904  * as being removed so no more stacks will use it and determine
905  * whether the removal would succeed.
906  *
907  * When called with a false quiesce argument, actually attempt the
908  * removal.
909  *
910  * When called with a force argument, attempt to switch all TCBs to
911  * use the default stack instead of returning EBUSY.
912  *
913  * Returns 0 on success (or if the removal would succeed, or an error
914  * code on failure.
915  */
916 int
917 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
918     bool force)
919 {
920 	struct tcp_function *f;
921 
922 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
923 		/* You can't un-register the default */
924 		return (EPERM);
925 	}
926 	rw_wlock(&tcp_function_lock);
927 	if (blk == tcp_func_set_ptr) {
928 		/* You can't free the current default */
929 		rw_wunlock(&tcp_function_lock);
930 		return (EBUSY);
931 	}
932 	/* Mark the block so no more stacks can use it. */
933 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
934 	/*
935 	 * If TCBs are still attached to the stack, attempt to switch them
936 	 * to the default stack.
937 	 */
938 	if (force && blk->tfb_refcnt) {
939 		struct inpcb *inp;
940 		struct tcpcb *tp;
941 		VNET_ITERATOR_DECL(vnet_iter);
942 
943 		rw_wunlock(&tcp_function_lock);
944 
945 		VNET_LIST_RLOCK();
946 		VNET_FOREACH(vnet_iter) {
947 			CURVNET_SET(vnet_iter);
948 			INP_INFO_WLOCK(&V_tcbinfo);
949 			LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
950 				INP_WLOCK(inp);
951 				if (inp->inp_flags & INP_TIMEWAIT) {
952 					INP_WUNLOCK(inp);
953 					continue;
954 				}
955 				tp = intotcpcb(inp);
956 				if (tp == NULL || tp->t_fb != blk) {
957 					INP_WUNLOCK(inp);
958 					continue;
959 				}
960 				tcp_switch_back_to_default(tp);
961 				INP_WUNLOCK(inp);
962 			}
963 			INP_INFO_WUNLOCK(&V_tcbinfo);
964 			CURVNET_RESTORE();
965 		}
966 		VNET_LIST_RUNLOCK();
967 
968 		rw_wlock(&tcp_function_lock);
969 	}
970 	if (blk->tfb_refcnt) {
971 		/* TCBs still attached. */
972 		rw_wunlock(&tcp_function_lock);
973 		return (EBUSY);
974 	}
975 	if (quiesce) {
976 		/* Skip removal. */
977 		rw_wunlock(&tcp_function_lock);
978 		return (0);
979 	}
980 	/* Remove any function names that map to this function block. */
981 	while (find_tcp_fb_locked(blk, &f) != NULL) {
982 		TAILQ_REMOVE(&t_functions, f, tf_next);
983 		tcp_fb_cnt--;
984 		f->tf_fb = NULL;
985 		free(f, M_TCPFUNCTIONS);
986 	}
987 	rw_wunlock(&tcp_function_lock);
988 	return (0);
989 }
990 
991 void
992 tcp_init(void)
993 {
994 	const char *tcbhash_tuneable;
995 	int hashsize;
996 
997 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
998 
999 #ifdef TCP_HHOOK
1000 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1001 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1002 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1003 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1004 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1005 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1006 #endif
1007 	hashsize = TCBHASHSIZE;
1008 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1009 	if (hashsize == 0) {
1010 		/*
1011 		 * Auto tune the hash size based on maxsockets.
1012 		 * A perfect hash would have a 1:1 mapping
1013 		 * (hashsize = maxsockets) however it's been
1014 		 * suggested that O(2) average is better.
1015 		 */
1016 		hashsize = maketcp_hashsize(maxsockets / 4);
1017 		/*
1018 		 * Our historical default is 512,
1019 		 * do not autotune lower than this.
1020 		 */
1021 		if (hashsize < 512)
1022 			hashsize = 512;
1023 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1024 			printf("%s: %s auto tuned to %d\n", __func__,
1025 			    tcbhash_tuneable, hashsize);
1026 	}
1027 	/*
1028 	 * We require a hashsize to be a power of two.
1029 	 * Previously if it was not a power of two we would just reset it
1030 	 * back to 512, which could be a nasty surprise if you did not notice
1031 	 * the error message.
1032 	 * Instead what we do is clip it to the closest power of two lower
1033 	 * than the specified hash value.
1034 	 */
1035 	if (!powerof2(hashsize)) {
1036 		int oldhashsize = hashsize;
1037 
1038 		hashsize = maketcp_hashsize(hashsize);
1039 		/* prevent absurdly low value */
1040 		if (hashsize < 16)
1041 			hashsize = 16;
1042 		printf("%s: WARNING: TCB hash size not a power of 2, "
1043 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1044 		    hashsize);
1045 	}
1046 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1047 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1048 
1049 	/*
1050 	 * These have to be type stable for the benefit of the timers.
1051 	 */
1052 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1053 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1054 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1055 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1056 
1057 	tcp_tw_init();
1058 	syncache_init();
1059 	tcp_hc_init();
1060 
1061 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1062 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1063 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1064 
1065 	tcp_fastopen_init();
1066 
1067 	/* Skip initialization of globals for non-default instances. */
1068 	if (!IS_DEFAULT_VNET(curvnet))
1069 		return;
1070 
1071 	tcp_reass_global_init();
1072 
1073 	/* XXX virtualize those bellow? */
1074 	tcp_delacktime = TCPTV_DELACK;
1075 	tcp_keepinit = TCPTV_KEEP_INIT;
1076 	tcp_keepidle = TCPTV_KEEP_IDLE;
1077 	tcp_keepintvl = TCPTV_KEEPINTVL;
1078 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1079 	tcp_msl = TCPTV_MSL;
1080 	tcp_rexmit_min = TCPTV_MIN;
1081 	if (tcp_rexmit_min < 1)
1082 		tcp_rexmit_min = 1;
1083 	tcp_persmin = TCPTV_PERSMIN;
1084 	tcp_persmax = TCPTV_PERSMAX;
1085 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1086 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1087 	tcp_tcbhashsize = hashsize;
1088 	/* Setup the tcp function block list */
1089 	init_tcp_functions();
1090 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1091 #ifdef TCP_BLACKBOX
1092 	/* Initialize the TCP logging data. */
1093 	tcp_log_init();
1094 #endif
1095 
1096 	if (tcp_soreceive_stream) {
1097 #ifdef INET
1098 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1099 #endif
1100 #ifdef INET6
1101 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1102 #endif /* INET6 */
1103 	}
1104 
1105 #ifdef INET6
1106 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1107 #else /* INET6 */
1108 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1109 #endif /* INET6 */
1110 	if (max_protohdr < TCP_MINPROTOHDR)
1111 		max_protohdr = TCP_MINPROTOHDR;
1112 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1113 		panic("tcp_init");
1114 #undef TCP_MINPROTOHDR
1115 
1116 	ISN_LOCK_INIT();
1117 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1118 		SHUTDOWN_PRI_DEFAULT);
1119 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1120 		EVENTHANDLER_PRI_ANY);
1121 #ifdef TCPPCAP
1122 	tcp_pcap_init();
1123 #endif
1124 }
1125 
1126 #ifdef VIMAGE
1127 static void
1128 tcp_destroy(void *unused __unused)
1129 {
1130 	int n;
1131 #ifdef TCP_HHOOK
1132 	int error;
1133 #endif
1134 
1135 	/*
1136 	 * All our processes are gone, all our sockets should be cleaned
1137 	 * up, which means, we should be past the tcp_discardcb() calls.
1138 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1139 	 */
1140 	for (;;) {
1141 		INP_LIST_RLOCK(&V_tcbinfo);
1142 		n = V_tcbinfo.ipi_count;
1143 		INP_LIST_RUNLOCK(&V_tcbinfo);
1144 		if (n == 0)
1145 			break;
1146 		pause("tcpdes", hz / 10);
1147 	}
1148 	tcp_hc_destroy();
1149 	syncache_destroy();
1150 	tcp_tw_destroy();
1151 	in_pcbinfo_destroy(&V_tcbinfo);
1152 	/* tcp_discardcb() clears the sack_holes up. */
1153 	uma_zdestroy(V_sack_hole_zone);
1154 	uma_zdestroy(V_tcpcb_zone);
1155 
1156 	/*
1157 	 * Cannot free the zone until all tcpcbs are released as we attach
1158 	 * the allocations to them.
1159 	 */
1160 	tcp_fastopen_destroy();
1161 
1162 #ifdef TCP_HHOOK
1163 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1164 	if (error != 0) {
1165 		printf("%s: WARNING: unable to deregister helper hook "
1166 		    "type=%d, id=%d: error %d returned\n", __func__,
1167 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1168 	}
1169 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1170 	if (error != 0) {
1171 		printf("%s: WARNING: unable to deregister helper hook "
1172 		    "type=%d, id=%d: error %d returned\n", __func__,
1173 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1174 	}
1175 #endif
1176 }
1177 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1178 #endif
1179 
1180 void
1181 tcp_fini(void *xtp)
1182 {
1183 
1184 }
1185 
1186 /*
1187  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1188  * tcp_template used to store this data in mbufs, but we now recopy it out
1189  * of the tcpcb each time to conserve mbufs.
1190  */
1191 void
1192 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1193 {
1194 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1195 
1196 	INP_WLOCK_ASSERT(inp);
1197 
1198 #ifdef INET6
1199 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1200 		struct ip6_hdr *ip6;
1201 
1202 		ip6 = (struct ip6_hdr *)ip_ptr;
1203 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1204 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1205 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1206 			(IPV6_VERSION & IPV6_VERSION_MASK);
1207 		ip6->ip6_nxt = IPPROTO_TCP;
1208 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1209 		ip6->ip6_src = inp->in6p_laddr;
1210 		ip6->ip6_dst = inp->in6p_faddr;
1211 	}
1212 #endif /* INET6 */
1213 #if defined(INET6) && defined(INET)
1214 	else
1215 #endif
1216 #ifdef INET
1217 	{
1218 		struct ip *ip;
1219 
1220 		ip = (struct ip *)ip_ptr;
1221 		ip->ip_v = IPVERSION;
1222 		ip->ip_hl = 5;
1223 		ip->ip_tos = inp->inp_ip_tos;
1224 		ip->ip_len = 0;
1225 		ip->ip_id = 0;
1226 		ip->ip_off = 0;
1227 		ip->ip_ttl = inp->inp_ip_ttl;
1228 		ip->ip_sum = 0;
1229 		ip->ip_p = IPPROTO_TCP;
1230 		ip->ip_src = inp->inp_laddr;
1231 		ip->ip_dst = inp->inp_faddr;
1232 	}
1233 #endif /* INET */
1234 	th->th_sport = inp->inp_lport;
1235 	th->th_dport = inp->inp_fport;
1236 	th->th_seq = 0;
1237 	th->th_ack = 0;
1238 	th->th_x2 = 0;
1239 	th->th_off = 5;
1240 	th->th_flags = 0;
1241 	th->th_win = 0;
1242 	th->th_urp = 0;
1243 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1244 }
1245 
1246 /*
1247  * Create template to be used to send tcp packets on a connection.
1248  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1249  * use for this function is in keepalives, which use tcp_respond.
1250  */
1251 struct tcptemp *
1252 tcpip_maketemplate(struct inpcb *inp)
1253 {
1254 	struct tcptemp *t;
1255 
1256 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1257 	if (t == NULL)
1258 		return (NULL);
1259 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1260 	return (t);
1261 }
1262 
1263 /*
1264  * Send a single message to the TCP at address specified by
1265  * the given TCP/IP header.  If m == NULL, then we make a copy
1266  * of the tcpiphdr at th and send directly to the addressed host.
1267  * This is used to force keep alive messages out using the TCP
1268  * template for a connection.  If flags are given then we send
1269  * a message back to the TCP which originated the segment th,
1270  * and discard the mbuf containing it and any other attached mbufs.
1271  *
1272  * In any case the ack and sequence number of the transmitted
1273  * segment are as specified by the parameters.
1274  *
1275  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1276  */
1277 void
1278 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1279     tcp_seq ack, tcp_seq seq, int flags)
1280 {
1281 	struct tcpopt to;
1282 	struct inpcb *inp;
1283 	struct ip *ip;
1284 	struct mbuf *optm;
1285 	struct tcphdr *nth;
1286 	u_char *optp;
1287 #ifdef INET6
1288 	struct ip6_hdr *ip6;
1289 	int isipv6;
1290 #endif /* INET6 */
1291 	int optlen, tlen, win;
1292 	bool incl_opts;
1293 
1294 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1295 
1296 #ifdef INET6
1297 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1298 	ip6 = ipgen;
1299 #endif /* INET6 */
1300 	ip = ipgen;
1301 
1302 	if (tp != NULL) {
1303 		inp = tp->t_inpcb;
1304 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1305 		INP_WLOCK_ASSERT(inp);
1306 	} else
1307 		inp = NULL;
1308 
1309 	incl_opts = false;
1310 	win = 0;
1311 	if (tp != NULL) {
1312 		if (!(flags & TH_RST)) {
1313 			win = sbspace(&inp->inp_socket->so_rcv);
1314 			if (win > TCP_MAXWIN << tp->rcv_scale)
1315 				win = TCP_MAXWIN << tp->rcv_scale;
1316 		}
1317 		if ((tp->t_flags & TF_NOOPT) == 0)
1318 			incl_opts = true;
1319 	}
1320 	if (m == NULL) {
1321 		m = m_gethdr(M_NOWAIT, MT_DATA);
1322 		if (m == NULL)
1323 			return;
1324 		m->m_data += max_linkhdr;
1325 #ifdef INET6
1326 		if (isipv6) {
1327 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1328 			      sizeof(struct ip6_hdr));
1329 			ip6 = mtod(m, struct ip6_hdr *);
1330 			nth = (struct tcphdr *)(ip6 + 1);
1331 		} else
1332 #endif /* INET6 */
1333 		{
1334 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1335 			ip = mtod(m, struct ip *);
1336 			nth = (struct tcphdr *)(ip + 1);
1337 		}
1338 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1339 		flags = TH_ACK;
1340 	} else if (!M_WRITABLE(m)) {
1341 		struct mbuf *n;
1342 
1343 		/* Can't reuse 'm', allocate a new mbuf. */
1344 		n = m_gethdr(M_NOWAIT, MT_DATA);
1345 		if (n == NULL) {
1346 			m_freem(m);
1347 			return;
1348 		}
1349 
1350 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1351 			m_freem(m);
1352 			m_freem(n);
1353 			return;
1354 		}
1355 
1356 		n->m_data += max_linkhdr;
1357 		/* m_len is set later */
1358 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1359 #ifdef INET6
1360 		if (isipv6) {
1361 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1362 			      sizeof(struct ip6_hdr));
1363 			ip6 = mtod(n, struct ip6_hdr *);
1364 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1365 			nth = (struct tcphdr *)(ip6 + 1);
1366 		} else
1367 #endif /* INET6 */
1368 		{
1369 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1370 			ip = mtod(n, struct ip *);
1371 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1372 			nth = (struct tcphdr *)(ip + 1);
1373 		}
1374 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1375 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1376 		th = nth;
1377 		m_freem(m);
1378 		m = n;
1379 	} else {
1380 		/*
1381 		 *  reuse the mbuf.
1382 		 * XXX MRT We inherit the FIB, which is lucky.
1383 		 */
1384 		m_freem(m->m_next);
1385 		m->m_next = NULL;
1386 		m->m_data = (caddr_t)ipgen;
1387 		/* m_len is set later */
1388 #ifdef INET6
1389 		if (isipv6) {
1390 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1391 			nth = (struct tcphdr *)(ip6 + 1);
1392 		} else
1393 #endif /* INET6 */
1394 		{
1395 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1396 			nth = (struct tcphdr *)(ip + 1);
1397 		}
1398 		if (th != nth) {
1399 			/*
1400 			 * this is usually a case when an extension header
1401 			 * exists between the IPv6 header and the
1402 			 * TCP header.
1403 			 */
1404 			nth->th_sport = th->th_sport;
1405 			nth->th_dport = th->th_dport;
1406 		}
1407 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1408 #undef xchg
1409 	}
1410 	tlen = 0;
1411 #ifdef INET6
1412 	if (isipv6)
1413 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1414 #endif
1415 #if defined(INET) && defined(INET6)
1416 	else
1417 #endif
1418 #ifdef INET
1419 		tlen = sizeof (struct tcpiphdr);
1420 #endif
1421 #ifdef INVARIANTS
1422 	m->m_len = 0;
1423 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1424 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1425 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1426 #endif
1427 	m->m_len = tlen;
1428 	to.to_flags = 0;
1429 	if (incl_opts) {
1430 		/* Make sure we have room. */
1431 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1432 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1433 			if (m->m_next) {
1434 				optp = mtod(m->m_next, u_char *);
1435 				optm = m->m_next;
1436 			} else
1437 				incl_opts = false;
1438 		} else {
1439 			optp = (u_char *) (nth + 1);
1440 			optm = m;
1441 		}
1442 	}
1443 	if (incl_opts) {
1444 		/* Timestamps. */
1445 		if (tp->t_flags & TF_RCVD_TSTMP) {
1446 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1447 			to.to_tsecr = tp->ts_recent;
1448 			to.to_flags |= TOF_TS;
1449 		}
1450 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1451 		/* TCP-MD5 (RFC2385). */
1452 		if (tp->t_flags & TF_SIGNATURE)
1453 			to.to_flags |= TOF_SIGNATURE;
1454 #endif
1455 		/* Add the options. */
1456 		tlen += optlen = tcp_addoptions(&to, optp);
1457 
1458 		/* Update m_len in the correct mbuf. */
1459 		optm->m_len += optlen;
1460 	} else
1461 		optlen = 0;
1462 #ifdef INET6
1463 	if (isipv6) {
1464 		ip6->ip6_flow = 0;
1465 		ip6->ip6_vfc = IPV6_VERSION;
1466 		ip6->ip6_nxt = IPPROTO_TCP;
1467 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1468 	}
1469 #endif
1470 #if defined(INET) && defined(INET6)
1471 	else
1472 #endif
1473 #ifdef INET
1474 	{
1475 		ip->ip_len = htons(tlen);
1476 		ip->ip_ttl = V_ip_defttl;
1477 		if (V_path_mtu_discovery)
1478 			ip->ip_off |= htons(IP_DF);
1479 	}
1480 #endif
1481 	m->m_pkthdr.len = tlen;
1482 	m->m_pkthdr.rcvif = NULL;
1483 #ifdef MAC
1484 	if (inp != NULL) {
1485 		/*
1486 		 * Packet is associated with a socket, so allow the
1487 		 * label of the response to reflect the socket label.
1488 		 */
1489 		INP_WLOCK_ASSERT(inp);
1490 		mac_inpcb_create_mbuf(inp, m);
1491 	} else {
1492 		/*
1493 		 * Packet is not associated with a socket, so possibly
1494 		 * update the label in place.
1495 		 */
1496 		mac_netinet_tcp_reply(m);
1497 	}
1498 #endif
1499 	nth->th_seq = htonl(seq);
1500 	nth->th_ack = htonl(ack);
1501 	nth->th_x2 = 0;
1502 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1503 	nth->th_flags = flags;
1504 	if (tp != NULL)
1505 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1506 	else
1507 		nth->th_win = htons((u_short)win);
1508 	nth->th_urp = 0;
1509 
1510 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1511 	if (to.to_flags & TOF_SIGNATURE) {
1512 		if (!TCPMD5_ENABLED() ||
1513 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1514 			m_freem(m);
1515 			return;
1516 		}
1517 	}
1518 #endif
1519 
1520 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1521 #ifdef INET6
1522 	if (isipv6) {
1523 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1524 		nth->th_sum = in6_cksum_pseudo(ip6,
1525 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1526 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1527 		    NULL, NULL);
1528 	}
1529 #endif /* INET6 */
1530 #if defined(INET6) && defined(INET)
1531 	else
1532 #endif
1533 #ifdef INET
1534 	{
1535 		m->m_pkthdr.csum_flags = CSUM_TCP;
1536 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1537 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1538 	}
1539 #endif /* INET */
1540 #ifdef TCPDEBUG
1541 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1542 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1543 #endif
1544 	TCP_PROBE3(debug__output, tp, th, m);
1545 	if (flags & TH_RST)
1546 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1547 
1548 #ifdef INET6
1549 	if (isipv6) {
1550 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1551 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1552 	}
1553 #endif /* INET6 */
1554 #if defined(INET) && defined(INET6)
1555 	else
1556 #endif
1557 #ifdef INET
1558 	{
1559 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1560 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1561 	}
1562 #endif
1563 }
1564 
1565 /*
1566  * Create a new TCP control block, making an
1567  * empty reassembly queue and hooking it to the argument
1568  * protocol control block.  The `inp' parameter must have
1569  * come from the zone allocator set up in tcp_init().
1570  */
1571 struct tcpcb *
1572 tcp_newtcpcb(struct inpcb *inp)
1573 {
1574 	struct tcpcb_mem *tm;
1575 	struct tcpcb *tp;
1576 #ifdef INET6
1577 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1578 #endif /* INET6 */
1579 
1580 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1581 	if (tm == NULL)
1582 		return (NULL);
1583 	tp = &tm->tcb;
1584 
1585 	/* Initialise cc_var struct for this tcpcb. */
1586 	tp->ccv = &tm->ccv;
1587 	tp->ccv->type = IPPROTO_TCP;
1588 	tp->ccv->ccvc.tcp = tp;
1589 	rw_rlock(&tcp_function_lock);
1590 	tp->t_fb = tcp_func_set_ptr;
1591 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1592 	rw_runlock(&tcp_function_lock);
1593 	/*
1594 	 * Use the current system default CC algorithm.
1595 	 */
1596 	CC_LIST_RLOCK();
1597 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1598 	CC_ALGO(tp) = CC_DEFAULT();
1599 	CC_LIST_RUNLOCK();
1600 
1601 	if (CC_ALGO(tp)->cb_init != NULL)
1602 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1603 			if (tp->t_fb->tfb_tcp_fb_fini)
1604 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1605 			refcount_release(&tp->t_fb->tfb_refcnt);
1606 			uma_zfree(V_tcpcb_zone, tm);
1607 			return (NULL);
1608 		}
1609 
1610 #ifdef TCP_HHOOK
1611 	tp->osd = &tm->osd;
1612 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1613 		if (tp->t_fb->tfb_tcp_fb_fini)
1614 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1615 		refcount_release(&tp->t_fb->tfb_refcnt);
1616 		uma_zfree(V_tcpcb_zone, tm);
1617 		return (NULL);
1618 	}
1619 #endif
1620 
1621 #ifdef VIMAGE
1622 	tp->t_vnet = inp->inp_vnet;
1623 #endif
1624 	tp->t_timers = &tm->tt;
1625 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
1626 	tp->t_maxseg =
1627 #ifdef INET6
1628 		isipv6 ? V_tcp_v6mssdflt :
1629 #endif /* INET6 */
1630 		V_tcp_mssdflt;
1631 
1632 	/* Set up our timeouts. */
1633 	callout_init(&tp->t_timers->tt_rexmt, 1);
1634 	callout_init(&tp->t_timers->tt_persist, 1);
1635 	callout_init(&tp->t_timers->tt_keep, 1);
1636 	callout_init(&tp->t_timers->tt_2msl, 1);
1637 	callout_init(&tp->t_timers->tt_delack, 1);
1638 
1639 	if (V_tcp_do_rfc1323)
1640 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1641 	if (V_tcp_do_sack)
1642 		tp->t_flags |= TF_SACK_PERMIT;
1643 	TAILQ_INIT(&tp->snd_holes);
1644 	/*
1645 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1646 	 * is called.
1647 	 */
1648 	in_pcbref(inp);	/* Reference for tcpcb */
1649 	tp->t_inpcb = inp;
1650 
1651 	/*
1652 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1653 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1654 	 * reasonable initial retransmit time.
1655 	 */
1656 	tp->t_srtt = TCPTV_SRTTBASE;
1657 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1658 	tp->t_rttmin = tcp_rexmit_min;
1659 	tp->t_rxtcur = TCPTV_RTOBASE;
1660 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1661 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1662 	tp->t_rcvtime = ticks;
1663 	/*
1664 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1665 	 * because the socket may be bound to an IPv6 wildcard address,
1666 	 * which may match an IPv4-mapped IPv6 address.
1667 	 */
1668 	inp->inp_ip_ttl = V_ip_defttl;
1669 	inp->inp_ppcb = tp;
1670 #ifdef TCPPCAP
1671 	/*
1672 	 * Init the TCP PCAP queues.
1673 	 */
1674 	tcp_pcap_tcpcb_init(tp);
1675 #endif
1676 #ifdef TCP_BLACKBOX
1677 	/* Initialize the per-TCPCB log data. */
1678 	tcp_log_tcpcbinit(tp);
1679 #endif
1680 	if (tp->t_fb->tfb_tcp_fb_init) {
1681 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1682 	}
1683 	return (tp);		/* XXX */
1684 }
1685 
1686 /*
1687  * Switch the congestion control algorithm back to NewReno for any active
1688  * control blocks using an algorithm which is about to go away.
1689  * This ensures the CC framework can allow the unload to proceed without leaving
1690  * any dangling pointers which would trigger a panic.
1691  * Returning non-zero would inform the CC framework that something went wrong
1692  * and it would be unsafe to allow the unload to proceed. However, there is no
1693  * way for this to occur with this implementation so we always return zero.
1694  */
1695 int
1696 tcp_ccalgounload(struct cc_algo *unload_algo)
1697 {
1698 	struct cc_algo *tmpalgo;
1699 	struct inpcb *inp;
1700 	struct tcpcb *tp;
1701 	VNET_ITERATOR_DECL(vnet_iter);
1702 
1703 	/*
1704 	 * Check all active control blocks across all network stacks and change
1705 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1706 	 * requires cleanup code to be run, call it.
1707 	 */
1708 	VNET_LIST_RLOCK();
1709 	VNET_FOREACH(vnet_iter) {
1710 		CURVNET_SET(vnet_iter);
1711 		INP_INFO_WLOCK(&V_tcbinfo);
1712 		/*
1713 		 * New connections already part way through being initialised
1714 		 * with the CC algo we're removing will not race with this code
1715 		 * because the INP_INFO_WLOCK is held during initialisation. We
1716 		 * therefore don't enter the loop below until the connection
1717 		 * list has stabilised.
1718 		 */
1719 		LIST_FOREACH(inp, &V_tcb, inp_list) {
1720 			INP_WLOCK(inp);
1721 			/* Important to skip tcptw structs. */
1722 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1723 			    (tp = intotcpcb(inp)) != NULL) {
1724 				/*
1725 				 * By holding INP_WLOCK here, we are assured
1726 				 * that the connection is not currently
1727 				 * executing inside the CC module's functions
1728 				 * i.e. it is safe to make the switch back to
1729 				 * NewReno.
1730 				 */
1731 				if (CC_ALGO(tp) == unload_algo) {
1732 					tmpalgo = CC_ALGO(tp);
1733 					/* NewReno does not require any init. */
1734 					CC_ALGO(tp) = &newreno_cc_algo;
1735 					if (tmpalgo->cb_destroy != NULL)
1736 						tmpalgo->cb_destroy(tp->ccv);
1737 				}
1738 			}
1739 			INP_WUNLOCK(inp);
1740 		}
1741 		INP_INFO_WUNLOCK(&V_tcbinfo);
1742 		CURVNET_RESTORE();
1743 	}
1744 	VNET_LIST_RUNLOCK();
1745 
1746 	return (0);
1747 }
1748 
1749 /*
1750  * Drop a TCP connection, reporting
1751  * the specified error.  If connection is synchronized,
1752  * then send a RST to peer.
1753  */
1754 struct tcpcb *
1755 tcp_drop(struct tcpcb *tp, int errno)
1756 {
1757 	struct socket *so = tp->t_inpcb->inp_socket;
1758 
1759 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1760 	INP_WLOCK_ASSERT(tp->t_inpcb);
1761 
1762 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1763 		tcp_state_change(tp, TCPS_CLOSED);
1764 		(void) tp->t_fb->tfb_tcp_output(tp);
1765 		TCPSTAT_INC(tcps_drops);
1766 	} else
1767 		TCPSTAT_INC(tcps_conndrops);
1768 	if (errno == ETIMEDOUT && tp->t_softerror)
1769 		errno = tp->t_softerror;
1770 	so->so_error = errno;
1771 	return (tcp_close(tp));
1772 }
1773 
1774 void
1775 tcp_discardcb(struct tcpcb *tp)
1776 {
1777 	struct inpcb *inp = tp->t_inpcb;
1778 	struct socket *so = inp->inp_socket;
1779 #ifdef INET6
1780 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1781 #endif /* INET6 */
1782 	int released __unused;
1783 
1784 	INP_WLOCK_ASSERT(inp);
1785 
1786 	/*
1787 	 * Make sure that all of our timers are stopped before we delete the
1788 	 * PCB.
1789 	 *
1790 	 * If stopping a timer fails, we schedule a discard function in same
1791 	 * callout, and the last discard function called will take care of
1792 	 * deleting the tcpcb.
1793 	 */
1794 	tp->t_timers->tt_draincnt = 0;
1795 	tcp_timer_stop(tp, TT_REXMT);
1796 	tcp_timer_stop(tp, TT_PERSIST);
1797 	tcp_timer_stop(tp, TT_KEEP);
1798 	tcp_timer_stop(tp, TT_2MSL);
1799 	tcp_timer_stop(tp, TT_DELACK);
1800 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1801 		/*
1802 		 * Call the stop-all function of the methods,
1803 		 * this function should call the tcp_timer_stop()
1804 		 * method with each of the function specific timeouts.
1805 		 * That stop will be called via the tfb_tcp_timer_stop()
1806 		 * which should use the async drain function of the
1807 		 * callout system (see tcp_var.h).
1808 		 */
1809 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1810 	}
1811 
1812 	/*
1813 	 * If we got enough samples through the srtt filter,
1814 	 * save the rtt and rttvar in the routing entry.
1815 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1816 	 * 4 samples is enough for the srtt filter to converge
1817 	 * to within enough % of the correct value; fewer samples
1818 	 * and we could save a bogus rtt. The danger is not high
1819 	 * as tcp quickly recovers from everything.
1820 	 * XXX: Works very well but needs some more statistics!
1821 	 */
1822 	if (tp->t_rttupdated >= 4) {
1823 		struct hc_metrics_lite metrics;
1824 		uint32_t ssthresh;
1825 
1826 		bzero(&metrics, sizeof(metrics));
1827 		/*
1828 		 * Update the ssthresh always when the conditions below
1829 		 * are satisfied. This gives us better new start value
1830 		 * for the congestion avoidance for new connections.
1831 		 * ssthresh is only set if packet loss occurred on a session.
1832 		 *
1833 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1834 		 * being torn down.  Ideally this code would not use 'so'.
1835 		 */
1836 		ssthresh = tp->snd_ssthresh;
1837 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1838 			/*
1839 			 * convert the limit from user data bytes to
1840 			 * packets then to packet data bytes.
1841 			 */
1842 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1843 			if (ssthresh < 2)
1844 				ssthresh = 2;
1845 			ssthresh *= (tp->t_maxseg +
1846 #ifdef INET6
1847 			    (isipv6 ? sizeof (struct ip6_hdr) +
1848 				sizeof (struct tcphdr) :
1849 #endif
1850 				sizeof (struct tcpiphdr)
1851 #ifdef INET6
1852 			    )
1853 #endif
1854 			    );
1855 		} else
1856 			ssthresh = 0;
1857 		metrics.rmx_ssthresh = ssthresh;
1858 
1859 		metrics.rmx_rtt = tp->t_srtt;
1860 		metrics.rmx_rttvar = tp->t_rttvar;
1861 		metrics.rmx_cwnd = tp->snd_cwnd;
1862 		metrics.rmx_sendpipe = 0;
1863 		metrics.rmx_recvpipe = 0;
1864 
1865 		tcp_hc_update(&inp->inp_inc, &metrics);
1866 	}
1867 
1868 	/* free the reassembly queue, if any */
1869 	tcp_reass_flush(tp);
1870 
1871 #ifdef TCP_OFFLOAD
1872 	/* Disconnect offload device, if any. */
1873 	if (tp->t_flags & TF_TOE)
1874 		tcp_offload_detach(tp);
1875 #endif
1876 
1877 	tcp_free_sackholes(tp);
1878 
1879 #ifdef TCPPCAP
1880 	/* Free the TCP PCAP queues. */
1881 	tcp_pcap_drain(&(tp->t_inpkts));
1882 	tcp_pcap_drain(&(tp->t_outpkts));
1883 #endif
1884 
1885 	/* Allow the CC algorithm to clean up after itself. */
1886 	if (CC_ALGO(tp)->cb_destroy != NULL)
1887 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1888 
1889 #ifdef TCP_HHOOK
1890 	khelp_destroy_osd(tp->osd);
1891 #endif
1892 
1893 	CC_ALGO(tp) = NULL;
1894 	inp->inp_ppcb = NULL;
1895 	if (tp->t_timers->tt_draincnt == 0) {
1896 		/* We own the last reference on tcpcb, let's free it. */
1897 #ifdef TCP_BLACKBOX
1898 		tcp_log_tcpcbfini(tp);
1899 #endif
1900 		TCPSTATES_DEC(tp->t_state);
1901 		if (tp->t_fb->tfb_tcp_fb_fini)
1902 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1903 		refcount_release(&tp->t_fb->tfb_refcnt);
1904 		tp->t_inpcb = NULL;
1905 		uma_zfree(V_tcpcb_zone, tp);
1906 		released = in_pcbrele_wlocked(inp);
1907 		KASSERT(!released, ("%s: inp %p should not have been released "
1908 			"here", __func__, inp));
1909 	}
1910 }
1911 
1912 void
1913 tcp_timer_discard(void *ptp)
1914 {
1915 	struct inpcb *inp;
1916 	struct tcpcb *tp;
1917 
1918 	tp = (struct tcpcb *)ptp;
1919 	CURVNET_SET(tp->t_vnet);
1920 	INP_INFO_RLOCK(&V_tcbinfo);
1921 	inp = tp->t_inpcb;
1922 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1923 		__func__, tp));
1924 	INP_WLOCK(inp);
1925 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1926 		("%s: tcpcb has to be stopped here", __func__));
1927 	tp->t_timers->tt_draincnt--;
1928 	if (tp->t_timers->tt_draincnt == 0) {
1929 		/* We own the last reference on this tcpcb, let's free it. */
1930 #ifdef TCP_BLACKBOX
1931 		tcp_log_tcpcbfini(tp);
1932 #endif
1933 		TCPSTATES_DEC(tp->t_state);
1934 		if (tp->t_fb->tfb_tcp_fb_fini)
1935 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1936 		refcount_release(&tp->t_fb->tfb_refcnt);
1937 		tp->t_inpcb = NULL;
1938 		uma_zfree(V_tcpcb_zone, tp);
1939 		if (in_pcbrele_wlocked(inp)) {
1940 			INP_INFO_RUNLOCK(&V_tcbinfo);
1941 			CURVNET_RESTORE();
1942 			return;
1943 		}
1944 	}
1945 	INP_WUNLOCK(inp);
1946 	INP_INFO_RUNLOCK(&V_tcbinfo);
1947 	CURVNET_RESTORE();
1948 }
1949 
1950 /*
1951  * Attempt to close a TCP control block, marking it as dropped, and freeing
1952  * the socket if we hold the only reference.
1953  */
1954 struct tcpcb *
1955 tcp_close(struct tcpcb *tp)
1956 {
1957 	struct inpcb *inp = tp->t_inpcb;
1958 	struct socket *so;
1959 
1960 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1961 	INP_WLOCK_ASSERT(inp);
1962 
1963 #ifdef TCP_OFFLOAD
1964 	if (tp->t_state == TCPS_LISTEN)
1965 		tcp_offload_listen_stop(tp);
1966 #endif
1967 	/*
1968 	 * This releases the TFO pending counter resource for TFO listen
1969 	 * sockets as well as passively-created TFO sockets that transition
1970 	 * from SYN_RECEIVED to CLOSED.
1971 	 */
1972 	if (tp->t_tfo_pending) {
1973 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1974 		tp->t_tfo_pending = NULL;
1975 	}
1976 	in_pcbdrop(inp);
1977 	TCPSTAT_INC(tcps_closed);
1978 	if (tp->t_state != TCPS_CLOSED)
1979 		tcp_state_change(tp, TCPS_CLOSED);
1980 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1981 	so = inp->inp_socket;
1982 	soisdisconnected(so);
1983 	if (inp->inp_flags & INP_SOCKREF) {
1984 		KASSERT(so->so_state & SS_PROTOREF,
1985 		    ("tcp_close: !SS_PROTOREF"));
1986 		inp->inp_flags &= ~INP_SOCKREF;
1987 		INP_WUNLOCK(inp);
1988 		SOCK_LOCK(so);
1989 		so->so_state &= ~SS_PROTOREF;
1990 		sofree(so);
1991 		return (NULL);
1992 	}
1993 	return (tp);
1994 }
1995 
1996 void
1997 tcp_drain(void)
1998 {
1999 	VNET_ITERATOR_DECL(vnet_iter);
2000 
2001 	if (!do_tcpdrain)
2002 		return;
2003 
2004 	VNET_LIST_RLOCK_NOSLEEP();
2005 	VNET_FOREACH(vnet_iter) {
2006 		CURVNET_SET(vnet_iter);
2007 		struct inpcb *inpb;
2008 		struct tcpcb *tcpb;
2009 
2010 	/*
2011 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2012 	 * if there is one...
2013 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2014 	 *      reassembly queue should be flushed, but in a situation
2015 	 *	where we're really low on mbufs, this is potentially
2016 	 *	useful.
2017 	 */
2018 		INP_INFO_WLOCK(&V_tcbinfo);
2019 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2020 			if (inpb->inp_flags & INP_TIMEWAIT)
2021 				continue;
2022 			INP_WLOCK(inpb);
2023 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2024 				tcp_reass_flush(tcpb);
2025 				tcp_clean_sackreport(tcpb);
2026 #ifdef TCP_BLACKBOX
2027 				tcp_log_drain(tcpb);
2028 #endif
2029 #ifdef TCPPCAP
2030 				if (tcp_pcap_aggressive_free) {
2031 					/* Free the TCP PCAP queues. */
2032 					tcp_pcap_drain(&(tcpb->t_inpkts));
2033 					tcp_pcap_drain(&(tcpb->t_outpkts));
2034 				}
2035 #endif
2036 			}
2037 			INP_WUNLOCK(inpb);
2038 		}
2039 		INP_INFO_WUNLOCK(&V_tcbinfo);
2040 		CURVNET_RESTORE();
2041 	}
2042 	VNET_LIST_RUNLOCK_NOSLEEP();
2043 }
2044 
2045 /*
2046  * Notify a tcp user of an asynchronous error;
2047  * store error as soft error, but wake up user
2048  * (for now, won't do anything until can select for soft error).
2049  *
2050  * Do not wake up user since there currently is no mechanism for
2051  * reporting soft errors (yet - a kqueue filter may be added).
2052  */
2053 static struct inpcb *
2054 tcp_notify(struct inpcb *inp, int error)
2055 {
2056 	struct tcpcb *tp;
2057 
2058 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2059 	INP_WLOCK_ASSERT(inp);
2060 
2061 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2062 	    (inp->inp_flags & INP_DROPPED))
2063 		return (inp);
2064 
2065 	tp = intotcpcb(inp);
2066 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2067 
2068 	/*
2069 	 * Ignore some errors if we are hooked up.
2070 	 * If connection hasn't completed, has retransmitted several times,
2071 	 * and receives a second error, give up now.  This is better
2072 	 * than waiting a long time to establish a connection that
2073 	 * can never complete.
2074 	 */
2075 	if (tp->t_state == TCPS_ESTABLISHED &&
2076 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2077 	     error == EHOSTDOWN)) {
2078 		if (inp->inp_route.ro_rt) {
2079 			RTFREE(inp->inp_route.ro_rt);
2080 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2081 		}
2082 		return (inp);
2083 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2084 	    tp->t_softerror) {
2085 		tp = tcp_drop(tp, error);
2086 		if (tp != NULL)
2087 			return (inp);
2088 		else
2089 			return (NULL);
2090 	} else {
2091 		tp->t_softerror = error;
2092 		return (inp);
2093 	}
2094 #if 0
2095 	wakeup( &so->so_timeo);
2096 	sorwakeup(so);
2097 	sowwakeup(so);
2098 #endif
2099 }
2100 
2101 static int
2102 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2103 {
2104 	int error, i, m, n, pcb_count;
2105 	struct inpcb *inp, **inp_list;
2106 	inp_gen_t gencnt;
2107 	struct xinpgen xig;
2108 
2109 	/*
2110 	 * The process of preparing the TCB list is too time-consuming and
2111 	 * resource-intensive to repeat twice on every request.
2112 	 */
2113 	if (req->oldptr == NULL) {
2114 		n = V_tcbinfo.ipi_count +
2115 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2116 		n += imax(n / 8, 10);
2117 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2118 		return (0);
2119 	}
2120 
2121 	if (req->newptr != NULL)
2122 		return (EPERM);
2123 
2124 	/*
2125 	 * OK, now we're committed to doing something.
2126 	 */
2127 	INP_LIST_RLOCK(&V_tcbinfo);
2128 	gencnt = V_tcbinfo.ipi_gencnt;
2129 	n = V_tcbinfo.ipi_count;
2130 	INP_LIST_RUNLOCK(&V_tcbinfo);
2131 
2132 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2133 
2134 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2135 		+ (n + m) * sizeof(struct xtcpcb));
2136 	if (error != 0)
2137 		return (error);
2138 
2139 	xig.xig_len = sizeof xig;
2140 	xig.xig_count = n + m;
2141 	xig.xig_gen = gencnt;
2142 	xig.xig_sogen = so_gencnt;
2143 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2144 	if (error)
2145 		return (error);
2146 
2147 	error = syncache_pcblist(req, m, &pcb_count);
2148 	if (error)
2149 		return (error);
2150 
2151 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2152 
2153 	INP_INFO_WLOCK(&V_tcbinfo);
2154 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2155 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
2156 		INP_WLOCK(inp);
2157 		if (inp->inp_gencnt <= gencnt) {
2158 			/*
2159 			 * XXX: This use of cr_cansee(), introduced with
2160 			 * TCP state changes, is not quite right, but for
2161 			 * now, better than nothing.
2162 			 */
2163 			if (inp->inp_flags & INP_TIMEWAIT) {
2164 				if (intotw(inp) != NULL)
2165 					error = cr_cansee(req->td->td_ucred,
2166 					    intotw(inp)->tw_cred);
2167 				else
2168 					error = EINVAL;	/* Skip this inp. */
2169 			} else
2170 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2171 			if (error == 0) {
2172 				in_pcbref(inp);
2173 				inp_list[i++] = inp;
2174 			}
2175 		}
2176 		INP_WUNLOCK(inp);
2177 	}
2178 	INP_INFO_WUNLOCK(&V_tcbinfo);
2179 	n = i;
2180 
2181 	error = 0;
2182 	for (i = 0; i < n; i++) {
2183 		inp = inp_list[i];
2184 		INP_RLOCK(inp);
2185 		if (inp->inp_gencnt <= gencnt) {
2186 			struct xtcpcb xt;
2187 
2188 			tcp_inptoxtp(inp, &xt);
2189 			INP_RUNLOCK(inp);
2190 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2191 		} else
2192 			INP_RUNLOCK(inp);
2193 	}
2194 	INP_INFO_RLOCK(&V_tcbinfo);
2195 	for (i = 0; i < n; i++) {
2196 		inp = inp_list[i];
2197 		INP_RLOCK(inp);
2198 		if (!in_pcbrele_rlocked(inp))
2199 			INP_RUNLOCK(inp);
2200 	}
2201 	INP_INFO_RUNLOCK(&V_tcbinfo);
2202 
2203 	if (!error) {
2204 		/*
2205 		 * Give the user an updated idea of our state.
2206 		 * If the generation differs from what we told
2207 		 * her before, she knows that something happened
2208 		 * while we were processing this request, and it
2209 		 * might be necessary to retry.
2210 		 */
2211 		INP_LIST_RLOCK(&V_tcbinfo);
2212 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2213 		xig.xig_sogen = so_gencnt;
2214 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2215 		INP_LIST_RUNLOCK(&V_tcbinfo);
2216 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2217 	}
2218 	free(inp_list, M_TEMP);
2219 	return (error);
2220 }
2221 
2222 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2223     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2224     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2225 
2226 #ifdef INET
2227 static int
2228 tcp_getcred(SYSCTL_HANDLER_ARGS)
2229 {
2230 	struct xucred xuc;
2231 	struct sockaddr_in addrs[2];
2232 	struct inpcb *inp;
2233 	int error;
2234 
2235 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2236 	if (error)
2237 		return (error);
2238 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2239 	if (error)
2240 		return (error);
2241 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2242 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2243 	if (inp != NULL) {
2244 		if (inp->inp_socket == NULL)
2245 			error = ENOENT;
2246 		if (error == 0)
2247 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2248 		if (error == 0)
2249 			cru2x(inp->inp_cred, &xuc);
2250 		INP_RUNLOCK(inp);
2251 	} else
2252 		error = ENOENT;
2253 	if (error == 0)
2254 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2255 	return (error);
2256 }
2257 
2258 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2259     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2260     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2261 #endif /* INET */
2262 
2263 #ifdef INET6
2264 static int
2265 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2266 {
2267 	struct xucred xuc;
2268 	struct sockaddr_in6 addrs[2];
2269 	struct inpcb *inp;
2270 	int error;
2271 #ifdef INET
2272 	int mapped = 0;
2273 #endif
2274 
2275 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2276 	if (error)
2277 		return (error);
2278 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2279 	if (error)
2280 		return (error);
2281 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2282 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2283 		return (error);
2284 	}
2285 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2286 #ifdef INET
2287 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2288 			mapped = 1;
2289 		else
2290 #endif
2291 			return (EINVAL);
2292 	}
2293 
2294 #ifdef INET
2295 	if (mapped == 1)
2296 		inp = in_pcblookup(&V_tcbinfo,
2297 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2298 			addrs[1].sin6_port,
2299 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2300 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2301 	else
2302 #endif
2303 		inp = in6_pcblookup(&V_tcbinfo,
2304 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2305 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2306 			INPLOOKUP_RLOCKPCB, NULL);
2307 	if (inp != NULL) {
2308 		if (inp->inp_socket == NULL)
2309 			error = ENOENT;
2310 		if (error == 0)
2311 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2312 		if (error == 0)
2313 			cru2x(inp->inp_cred, &xuc);
2314 		INP_RUNLOCK(inp);
2315 	} else
2316 		error = ENOENT;
2317 	if (error == 0)
2318 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2319 	return (error);
2320 }
2321 
2322 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2323     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2324     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2325 #endif /* INET6 */
2326 
2327 
2328 #ifdef INET
2329 void
2330 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2331 {
2332 	struct ip *ip = vip;
2333 	struct tcphdr *th;
2334 	struct in_addr faddr;
2335 	struct inpcb *inp;
2336 	struct tcpcb *tp;
2337 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2338 	struct icmp *icp;
2339 	struct in_conninfo inc;
2340 	tcp_seq icmp_tcp_seq;
2341 	int mtu;
2342 
2343 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2344 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2345 		return;
2346 
2347 	if (cmd == PRC_MSGSIZE)
2348 		notify = tcp_mtudisc_notify;
2349 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2350 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2351 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2352 		notify = tcp_drop_syn_sent;
2353 
2354 	/*
2355 	 * Hostdead is ugly because it goes linearly through all PCBs.
2356 	 * XXX: We never get this from ICMP, otherwise it makes an
2357 	 * excellent DoS attack on machines with many connections.
2358 	 */
2359 	else if (cmd == PRC_HOSTDEAD)
2360 		ip = NULL;
2361 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2362 		return;
2363 
2364 	if (ip == NULL) {
2365 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2366 		return;
2367 	}
2368 
2369 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2370 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2371 	INP_INFO_RLOCK(&V_tcbinfo);
2372 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2373 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2374 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2375 		/* signal EHOSTDOWN, as it flushes the cached route */
2376 		inp = (*notify)(inp, EHOSTDOWN);
2377 		goto out;
2378 	}
2379 	icmp_tcp_seq = th->th_seq;
2380 	if (inp != NULL)  {
2381 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2382 		    !(inp->inp_flags & INP_DROPPED) &&
2383 		    !(inp->inp_socket == NULL)) {
2384 			tp = intotcpcb(inp);
2385 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2386 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2387 				if (cmd == PRC_MSGSIZE) {
2388 					/*
2389 					 * MTU discovery:
2390 					 * If we got a needfrag set the MTU
2391 					 * in the route to the suggested new
2392 					 * value (if given) and then notify.
2393 					 */
2394 					mtu = ntohs(icp->icmp_nextmtu);
2395 					/*
2396 					 * If no alternative MTU was
2397 					 * proposed, try the next smaller
2398 					 * one.
2399 					 */
2400 					if (!mtu)
2401 						mtu = ip_next_mtu(
2402 						    ntohs(ip->ip_len), 1);
2403 					if (mtu < V_tcp_minmss +
2404 					    sizeof(struct tcpiphdr))
2405 						mtu = V_tcp_minmss +
2406 						    sizeof(struct tcpiphdr);
2407 					/*
2408 					 * Only process the offered MTU if it
2409 					 * is smaller than the current one.
2410 					 */
2411 					if (mtu < tp->t_maxseg +
2412 					    sizeof(struct tcpiphdr)) {
2413 						bzero(&inc, sizeof(inc));
2414 						inc.inc_faddr = faddr;
2415 						inc.inc_fibnum =
2416 						    inp->inp_inc.inc_fibnum;
2417 						tcp_hc_updatemtu(&inc, mtu);
2418 						tcp_mtudisc(inp, mtu);
2419 					}
2420 				} else
2421 					inp = (*notify)(inp,
2422 					    inetctlerrmap[cmd]);
2423 			}
2424 		}
2425 	} else {
2426 		bzero(&inc, sizeof(inc));
2427 		inc.inc_fport = th->th_dport;
2428 		inc.inc_lport = th->th_sport;
2429 		inc.inc_faddr = faddr;
2430 		inc.inc_laddr = ip->ip_src;
2431 		syncache_unreach(&inc, icmp_tcp_seq);
2432 	}
2433 out:
2434 	if (inp != NULL)
2435 		INP_WUNLOCK(inp);
2436 	INP_INFO_RUNLOCK(&V_tcbinfo);
2437 }
2438 #endif /* INET */
2439 
2440 #ifdef INET6
2441 void
2442 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2443 {
2444 	struct in6_addr *dst;
2445 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2446 	struct ip6_hdr *ip6;
2447 	struct mbuf *m;
2448 	struct inpcb *inp;
2449 	struct tcpcb *tp;
2450 	struct icmp6_hdr *icmp6;
2451 	struct ip6ctlparam *ip6cp = NULL;
2452 	const struct sockaddr_in6 *sa6_src = NULL;
2453 	struct in_conninfo inc;
2454 	struct tcp_ports {
2455 		uint16_t th_sport;
2456 		uint16_t th_dport;
2457 	} t_ports;
2458 	tcp_seq icmp_tcp_seq;
2459 	unsigned int mtu;
2460 	unsigned int off;
2461 
2462 	if (sa->sa_family != AF_INET6 ||
2463 	    sa->sa_len != sizeof(struct sockaddr_in6))
2464 		return;
2465 
2466 	/* if the parameter is from icmp6, decode it. */
2467 	if (d != NULL) {
2468 		ip6cp = (struct ip6ctlparam *)d;
2469 		icmp6 = ip6cp->ip6c_icmp6;
2470 		m = ip6cp->ip6c_m;
2471 		ip6 = ip6cp->ip6c_ip6;
2472 		off = ip6cp->ip6c_off;
2473 		sa6_src = ip6cp->ip6c_src;
2474 		dst = ip6cp->ip6c_finaldst;
2475 	} else {
2476 		m = NULL;
2477 		ip6 = NULL;
2478 		off = 0;	/* fool gcc */
2479 		sa6_src = &sa6_any;
2480 		dst = NULL;
2481 	}
2482 
2483 	if (cmd == PRC_MSGSIZE)
2484 		notify = tcp_mtudisc_notify;
2485 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2486 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2487 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2488 		notify = tcp_drop_syn_sent;
2489 
2490 	/*
2491 	 * Hostdead is ugly because it goes linearly through all PCBs.
2492 	 * XXX: We never get this from ICMP, otherwise it makes an
2493 	 * excellent DoS attack on machines with many connections.
2494 	 */
2495 	else if (cmd == PRC_HOSTDEAD)
2496 		ip6 = NULL;
2497 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2498 		return;
2499 
2500 	if (ip6 == NULL) {
2501 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2502 			      (const struct sockaddr *)sa6_src,
2503 			      0, cmd, NULL, notify);
2504 		return;
2505 	}
2506 
2507 	/* Check if we can safely get the ports from the tcp hdr */
2508 	if (m == NULL ||
2509 	    (m->m_pkthdr.len <
2510 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2511 		return;
2512 	}
2513 	bzero(&t_ports, sizeof(struct tcp_ports));
2514 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2515 	INP_INFO_RLOCK(&V_tcbinfo);
2516 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2517 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2518 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2519 		/* signal EHOSTDOWN, as it flushes the cached route */
2520 		inp = (*notify)(inp, EHOSTDOWN);
2521 		goto out;
2522 	}
2523 	off += sizeof(struct tcp_ports);
2524 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2525 		goto out;
2526 	}
2527 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2528 	if (inp != NULL)  {
2529 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2530 		    !(inp->inp_flags & INP_DROPPED) &&
2531 		    !(inp->inp_socket == NULL)) {
2532 			tp = intotcpcb(inp);
2533 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2534 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2535 				if (cmd == PRC_MSGSIZE) {
2536 					/*
2537 					 * MTU discovery:
2538 					 * If we got a needfrag set the MTU
2539 					 * in the route to the suggested new
2540 					 * value (if given) and then notify.
2541 					 */
2542 					mtu = ntohl(icmp6->icmp6_mtu);
2543 					/*
2544 					 * If no alternative MTU was
2545 					 * proposed, or the proposed
2546 					 * MTU was too small, set to
2547 					 * the min.
2548 					 */
2549 					if (mtu < IPV6_MMTU)
2550 						mtu = IPV6_MMTU - 8;
2551 					bzero(&inc, sizeof(inc));
2552 					inc.inc_fibnum = M_GETFIB(m);
2553 					inc.inc_flags |= INC_ISIPV6;
2554 					inc.inc6_faddr = *dst;
2555 					if (in6_setscope(&inc.inc6_faddr,
2556 						m->m_pkthdr.rcvif, NULL))
2557 						goto out;
2558 					/*
2559 					 * Only process the offered MTU if it
2560 					 * is smaller than the current one.
2561 					 */
2562 					if (mtu < tp->t_maxseg +
2563 					    sizeof (struct tcphdr) +
2564 					    sizeof (struct ip6_hdr)) {
2565 						tcp_hc_updatemtu(&inc, mtu);
2566 						tcp_mtudisc(inp, mtu);
2567 						ICMP6STAT_INC(icp6s_pmtuchg);
2568 					}
2569 				} else
2570 					inp = (*notify)(inp,
2571 					    inet6ctlerrmap[cmd]);
2572 			}
2573 		}
2574 	} else {
2575 		bzero(&inc, sizeof(inc));
2576 		inc.inc_fibnum = M_GETFIB(m);
2577 		inc.inc_flags |= INC_ISIPV6;
2578 		inc.inc_fport = t_ports.th_dport;
2579 		inc.inc_lport = t_ports.th_sport;
2580 		inc.inc6_faddr = *dst;
2581 		inc.inc6_laddr = ip6->ip6_src;
2582 		syncache_unreach(&inc, icmp_tcp_seq);
2583 	}
2584 out:
2585 	if (inp != NULL)
2586 		INP_WUNLOCK(inp);
2587 	INP_INFO_RUNLOCK(&V_tcbinfo);
2588 }
2589 #endif /* INET6 */
2590 
2591 
2592 /*
2593  * Following is where TCP initial sequence number generation occurs.
2594  *
2595  * There are two places where we must use initial sequence numbers:
2596  * 1.  In SYN-ACK packets.
2597  * 2.  In SYN packets.
2598  *
2599  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2600  * tcp_syncache.c for details.
2601  *
2602  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2603  * depends on this property.  In addition, these ISNs should be
2604  * unguessable so as to prevent connection hijacking.  To satisfy
2605  * the requirements of this situation, the algorithm outlined in
2606  * RFC 1948 is used, with only small modifications.
2607  *
2608  * Implementation details:
2609  *
2610  * Time is based off the system timer, and is corrected so that it
2611  * increases by one megabyte per second.  This allows for proper
2612  * recycling on high speed LANs while still leaving over an hour
2613  * before rollover.
2614  *
2615  * As reading the *exact* system time is too expensive to be done
2616  * whenever setting up a TCP connection, we increment the time
2617  * offset in two ways.  First, a small random positive increment
2618  * is added to isn_offset for each connection that is set up.
2619  * Second, the function tcp_isn_tick fires once per clock tick
2620  * and increments isn_offset as necessary so that sequence numbers
2621  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2622  * random positive increments serve only to ensure that the same
2623  * exact sequence number is never sent out twice (as could otherwise
2624  * happen when a port is recycled in less than the system tick
2625  * interval.)
2626  *
2627  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2628  * between seeding of isn_secret.  This is normally set to zero,
2629  * as reseeding should not be necessary.
2630  *
2631  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2632  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2633  * general, this means holding an exclusive (write) lock.
2634  */
2635 
2636 #define ISN_BYTES_PER_SECOND 1048576
2637 #define ISN_STATIC_INCREMENT 4096
2638 #define ISN_RANDOM_INCREMENT (4096 - 1)
2639 
2640 static VNET_DEFINE(u_char, isn_secret[32]);
2641 static VNET_DEFINE(int, isn_last);
2642 static VNET_DEFINE(int, isn_last_reseed);
2643 static VNET_DEFINE(u_int32_t, isn_offset);
2644 static VNET_DEFINE(u_int32_t, isn_offset_old);
2645 
2646 #define	V_isn_secret			VNET(isn_secret)
2647 #define	V_isn_last			VNET(isn_last)
2648 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2649 #define	V_isn_offset			VNET(isn_offset)
2650 #define	V_isn_offset_old		VNET(isn_offset_old)
2651 
2652 tcp_seq
2653 tcp_new_isn(struct tcpcb *tp)
2654 {
2655 	MD5_CTX isn_ctx;
2656 	u_int32_t md5_buffer[4];
2657 	tcp_seq new_isn;
2658 	u_int32_t projected_offset;
2659 
2660 	INP_WLOCK_ASSERT(tp->t_inpcb);
2661 
2662 	ISN_LOCK();
2663 	/* Seed if this is the first use, reseed if requested. */
2664 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2665 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2666 		< (u_int)ticks))) {
2667 		read_random(&V_isn_secret, sizeof(V_isn_secret));
2668 		V_isn_last_reseed = ticks;
2669 	}
2670 
2671 	/* Compute the md5 hash and return the ISN. */
2672 	MD5Init(&isn_ctx);
2673 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2674 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2675 #ifdef INET6
2676 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2677 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2678 			  sizeof(struct in6_addr));
2679 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2680 			  sizeof(struct in6_addr));
2681 	} else
2682 #endif
2683 	{
2684 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2685 			  sizeof(struct in_addr));
2686 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2687 			  sizeof(struct in_addr));
2688 	}
2689 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2690 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
2691 	new_isn = (tcp_seq) md5_buffer[0];
2692 	V_isn_offset += ISN_STATIC_INCREMENT +
2693 		(arc4random() & ISN_RANDOM_INCREMENT);
2694 	if (ticks != V_isn_last) {
2695 		projected_offset = V_isn_offset_old +
2696 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2697 		if (SEQ_GT(projected_offset, V_isn_offset))
2698 			V_isn_offset = projected_offset;
2699 		V_isn_offset_old = V_isn_offset;
2700 		V_isn_last = ticks;
2701 	}
2702 	new_isn += V_isn_offset;
2703 	ISN_UNLOCK();
2704 	return (new_isn);
2705 }
2706 
2707 /*
2708  * When a specific ICMP unreachable message is received and the
2709  * connection state is SYN-SENT, drop the connection.  This behavior
2710  * is controlled by the icmp_may_rst sysctl.
2711  */
2712 struct inpcb *
2713 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2714 {
2715 	struct tcpcb *tp;
2716 
2717 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2718 	INP_WLOCK_ASSERT(inp);
2719 
2720 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2721 	    (inp->inp_flags & INP_DROPPED))
2722 		return (inp);
2723 
2724 	tp = intotcpcb(inp);
2725 	if (tp->t_state != TCPS_SYN_SENT)
2726 		return (inp);
2727 
2728 	if (IS_FASTOPEN(tp->t_flags))
2729 		tcp_fastopen_disable_path(tp);
2730 
2731 	tp = tcp_drop(tp, errno);
2732 	if (tp != NULL)
2733 		return (inp);
2734 	else
2735 		return (NULL);
2736 }
2737 
2738 /*
2739  * When `need fragmentation' ICMP is received, update our idea of the MSS
2740  * based on the new value. Also nudge TCP to send something, since we
2741  * know the packet we just sent was dropped.
2742  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2743  */
2744 static struct inpcb *
2745 tcp_mtudisc_notify(struct inpcb *inp, int error)
2746 {
2747 
2748 	tcp_mtudisc(inp, -1);
2749 	return (inp);
2750 }
2751 
2752 static void
2753 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2754 {
2755 	struct tcpcb *tp;
2756 	struct socket *so;
2757 
2758 	INP_WLOCK_ASSERT(inp);
2759 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2760 	    (inp->inp_flags & INP_DROPPED))
2761 		return;
2762 
2763 	tp = intotcpcb(inp);
2764 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2765 
2766 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2767 
2768 	so = inp->inp_socket;
2769 	SOCKBUF_LOCK(&so->so_snd);
2770 	/* If the mss is larger than the socket buffer, decrease the mss. */
2771 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2772 		tp->t_maxseg = so->so_snd.sb_hiwat;
2773 	SOCKBUF_UNLOCK(&so->so_snd);
2774 
2775 	TCPSTAT_INC(tcps_mturesent);
2776 	tp->t_rtttime = 0;
2777 	tp->snd_nxt = tp->snd_una;
2778 	tcp_free_sackholes(tp);
2779 	tp->snd_recover = tp->snd_max;
2780 	if (tp->t_flags & TF_SACK_PERMIT)
2781 		EXIT_FASTRECOVERY(tp->t_flags);
2782 	tp->t_fb->tfb_tcp_output(tp);
2783 }
2784 
2785 #ifdef INET
2786 /*
2787  * Look-up the routing entry to the peer of this inpcb.  If no route
2788  * is found and it cannot be allocated, then return 0.  This routine
2789  * is called by TCP routines that access the rmx structure and by
2790  * tcp_mss_update to get the peer/interface MTU.
2791  */
2792 uint32_t
2793 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2794 {
2795 	struct nhop4_extended nh4;
2796 	struct ifnet *ifp;
2797 	uint32_t maxmtu = 0;
2798 
2799 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2800 
2801 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2802 
2803 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2804 		    NHR_REF, 0, &nh4) != 0)
2805 			return (0);
2806 
2807 		ifp = nh4.nh_ifp;
2808 		maxmtu = nh4.nh_mtu;
2809 
2810 		/* Report additional interface capabilities. */
2811 		if (cap != NULL) {
2812 			if (ifp->if_capenable & IFCAP_TSO4 &&
2813 			    ifp->if_hwassist & CSUM_TSO) {
2814 				cap->ifcap |= CSUM_TSO;
2815 				cap->tsomax = ifp->if_hw_tsomax;
2816 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2817 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2818 			}
2819 		}
2820 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2821 	}
2822 	return (maxmtu);
2823 }
2824 #endif /* INET */
2825 
2826 #ifdef INET6
2827 uint32_t
2828 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2829 {
2830 	struct nhop6_extended nh6;
2831 	struct in6_addr dst6;
2832 	uint32_t scopeid;
2833 	struct ifnet *ifp;
2834 	uint32_t maxmtu = 0;
2835 
2836 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2837 
2838 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2839 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2840 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2841 		    0, &nh6) != 0)
2842 			return (0);
2843 
2844 		ifp = nh6.nh_ifp;
2845 		maxmtu = nh6.nh_mtu;
2846 
2847 		/* Report additional interface capabilities. */
2848 		if (cap != NULL) {
2849 			if (ifp->if_capenable & IFCAP_TSO6 &&
2850 			    ifp->if_hwassist & CSUM_TSO) {
2851 				cap->ifcap |= CSUM_TSO;
2852 				cap->tsomax = ifp->if_hw_tsomax;
2853 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2854 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2855 			}
2856 		}
2857 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2858 	}
2859 
2860 	return (maxmtu);
2861 }
2862 #endif /* INET6 */
2863 
2864 /*
2865  * Calculate effective SMSS per RFC5681 definition for a given TCP
2866  * connection at its current state, taking into account SACK and etc.
2867  */
2868 u_int
2869 tcp_maxseg(const struct tcpcb *tp)
2870 {
2871 	u_int optlen;
2872 
2873 	if (tp->t_flags & TF_NOOPT)
2874 		return (tp->t_maxseg);
2875 
2876 	/*
2877 	 * Here we have a simplified code from tcp_addoptions(),
2878 	 * without a proper loop, and having most of paddings hardcoded.
2879 	 * We might make mistakes with padding here in some edge cases,
2880 	 * but this is harmless, since result of tcp_maxseg() is used
2881 	 * only in cwnd and ssthresh estimations.
2882 	 */
2883 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2884 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2885 		if (tp->t_flags & TF_RCVD_TSTMP)
2886 			optlen = TCPOLEN_TSTAMP_APPA;
2887 		else
2888 			optlen = 0;
2889 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2890 		if (tp->t_flags & TF_SIGNATURE)
2891 			optlen += PAD(TCPOLEN_SIGNATURE);
2892 #endif
2893 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2894 			optlen += TCPOLEN_SACKHDR;
2895 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2896 			optlen = PAD(optlen);
2897 		}
2898 	} else {
2899 		if (tp->t_flags & TF_REQ_TSTMP)
2900 			optlen = TCPOLEN_TSTAMP_APPA;
2901 		else
2902 			optlen = PAD(TCPOLEN_MAXSEG);
2903 		if (tp->t_flags & TF_REQ_SCALE)
2904 			optlen += PAD(TCPOLEN_WINDOW);
2905 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2906 		if (tp->t_flags & TF_SIGNATURE)
2907 			optlen += PAD(TCPOLEN_SIGNATURE);
2908 #endif
2909 		if (tp->t_flags & TF_SACK_PERMIT)
2910 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2911 	}
2912 #undef PAD
2913 	optlen = min(optlen, TCP_MAXOLEN);
2914 	return (tp->t_maxseg - optlen);
2915 }
2916 
2917 static int
2918 sysctl_drop(SYSCTL_HANDLER_ARGS)
2919 {
2920 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2921 	struct sockaddr_storage addrs[2];
2922 	struct inpcb *inp;
2923 	struct tcpcb *tp;
2924 	struct tcptw *tw;
2925 	struct sockaddr_in *fin, *lin;
2926 #ifdef INET6
2927 	struct sockaddr_in6 *fin6, *lin6;
2928 #endif
2929 	int error;
2930 
2931 	inp = NULL;
2932 	fin = lin = NULL;
2933 #ifdef INET6
2934 	fin6 = lin6 = NULL;
2935 #endif
2936 	error = 0;
2937 
2938 	if (req->oldptr != NULL || req->oldlen != 0)
2939 		return (EINVAL);
2940 	if (req->newptr == NULL)
2941 		return (EPERM);
2942 	if (req->newlen < sizeof(addrs))
2943 		return (ENOMEM);
2944 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2945 	if (error)
2946 		return (error);
2947 
2948 	switch (addrs[0].ss_family) {
2949 #ifdef INET6
2950 	case AF_INET6:
2951 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2952 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2953 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2954 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2955 			return (EINVAL);
2956 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2957 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2958 				return (EINVAL);
2959 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2960 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2961 			fin = (struct sockaddr_in *)&addrs[0];
2962 			lin = (struct sockaddr_in *)&addrs[1];
2963 			break;
2964 		}
2965 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2966 		if (error)
2967 			return (error);
2968 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2969 		if (error)
2970 			return (error);
2971 		break;
2972 #endif
2973 #ifdef INET
2974 	case AF_INET:
2975 		fin = (struct sockaddr_in *)&addrs[0];
2976 		lin = (struct sockaddr_in *)&addrs[1];
2977 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2978 		    lin->sin_len != sizeof(struct sockaddr_in))
2979 			return (EINVAL);
2980 		break;
2981 #endif
2982 	default:
2983 		return (EINVAL);
2984 	}
2985 	INP_INFO_RLOCK(&V_tcbinfo);
2986 	switch (addrs[0].ss_family) {
2987 #ifdef INET6
2988 	case AF_INET6:
2989 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2990 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2991 		    INPLOOKUP_WLOCKPCB, NULL);
2992 		break;
2993 #endif
2994 #ifdef INET
2995 	case AF_INET:
2996 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2997 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2998 		break;
2999 #endif
3000 	}
3001 	if (inp != NULL) {
3002 		if (inp->inp_flags & INP_TIMEWAIT) {
3003 			/*
3004 			 * XXXRW: There currently exists a state where an
3005 			 * inpcb is present, but its timewait state has been
3006 			 * discarded.  For now, don't allow dropping of this
3007 			 * type of inpcb.
3008 			 */
3009 			tw = intotw(inp);
3010 			if (tw != NULL)
3011 				tcp_twclose(tw, 0);
3012 			else
3013 				INP_WUNLOCK(inp);
3014 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3015 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3016 			tp = intotcpcb(inp);
3017 			tp = tcp_drop(tp, ECONNABORTED);
3018 			if (tp != NULL)
3019 				INP_WUNLOCK(inp);
3020 		} else
3021 			INP_WUNLOCK(inp);
3022 	} else
3023 		error = ESRCH;
3024 	INP_INFO_RUNLOCK(&V_tcbinfo);
3025 	return (error);
3026 }
3027 
3028 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3029     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3030     0, sysctl_drop, "", "Drop TCP connection");
3031 
3032 /*
3033  * Generate a standardized TCP log line for use throughout the
3034  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3035  * allow use in the interrupt context.
3036  *
3037  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3038  * NB: The function may return NULL if memory allocation failed.
3039  *
3040  * Due to header inclusion and ordering limitations the struct ip
3041  * and ip6_hdr pointers have to be passed as void pointers.
3042  */
3043 char *
3044 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3045     const void *ip6hdr)
3046 {
3047 
3048 	/* Is logging enabled? */
3049 	if (tcp_log_in_vain == 0)
3050 		return (NULL);
3051 
3052 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3053 }
3054 
3055 char *
3056 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3057     const void *ip6hdr)
3058 {
3059 
3060 	/* Is logging enabled? */
3061 	if (tcp_log_debug == 0)
3062 		return (NULL);
3063 
3064 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3065 }
3066 
3067 static char *
3068 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3069     const void *ip6hdr)
3070 {
3071 	char *s, *sp;
3072 	size_t size;
3073 	struct ip *ip;
3074 #ifdef INET6
3075 	const struct ip6_hdr *ip6;
3076 
3077 	ip6 = (const struct ip6_hdr *)ip6hdr;
3078 #endif /* INET6 */
3079 	ip = (struct ip *)ip4hdr;
3080 
3081 	/*
3082 	 * The log line looks like this:
3083 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3084 	 */
3085 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3086 	    sizeof(PRINT_TH_FLAGS) + 1 +
3087 #ifdef INET6
3088 	    2 * INET6_ADDRSTRLEN;
3089 #else
3090 	    2 * INET_ADDRSTRLEN;
3091 #endif /* INET6 */
3092 
3093 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3094 	if (s == NULL)
3095 		return (NULL);
3096 
3097 	strcat(s, "TCP: [");
3098 	sp = s + strlen(s);
3099 
3100 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3101 		inet_ntoa_r(inc->inc_faddr, sp);
3102 		sp = s + strlen(s);
3103 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3104 		sp = s + strlen(s);
3105 		inet_ntoa_r(inc->inc_laddr, sp);
3106 		sp = s + strlen(s);
3107 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3108 #ifdef INET6
3109 	} else if (inc) {
3110 		ip6_sprintf(sp, &inc->inc6_faddr);
3111 		sp = s + strlen(s);
3112 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3113 		sp = s + strlen(s);
3114 		ip6_sprintf(sp, &inc->inc6_laddr);
3115 		sp = s + strlen(s);
3116 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3117 	} else if (ip6 && th) {
3118 		ip6_sprintf(sp, &ip6->ip6_src);
3119 		sp = s + strlen(s);
3120 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3121 		sp = s + strlen(s);
3122 		ip6_sprintf(sp, &ip6->ip6_dst);
3123 		sp = s + strlen(s);
3124 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3125 #endif /* INET6 */
3126 #ifdef INET
3127 	} else if (ip && th) {
3128 		inet_ntoa_r(ip->ip_src, sp);
3129 		sp = s + strlen(s);
3130 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3131 		sp = s + strlen(s);
3132 		inet_ntoa_r(ip->ip_dst, sp);
3133 		sp = s + strlen(s);
3134 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3135 #endif /* INET */
3136 	} else {
3137 		free(s, M_TCPLOG);
3138 		return (NULL);
3139 	}
3140 	sp = s + strlen(s);
3141 	if (th)
3142 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3143 	if (*(s + size - 1) != '\0')
3144 		panic("%s: string too long", __func__);
3145 	return (s);
3146 }
3147 
3148 /*
3149  * A subroutine which makes it easy to track TCP state changes with DTrace.
3150  * This function shouldn't be called for t_state initializations that don't
3151  * correspond to actual TCP state transitions.
3152  */
3153 void
3154 tcp_state_change(struct tcpcb *tp, int newstate)
3155 {
3156 #if defined(KDTRACE_HOOKS)
3157 	int pstate = tp->t_state;
3158 #endif
3159 
3160 	TCPSTATES_DEC(tp->t_state);
3161 	TCPSTATES_INC(newstate);
3162 	tp->t_state = newstate;
3163 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3164 }
3165 
3166 /*
3167  * Create an external-format (``xtcpcb'') structure using the information in
3168  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3169  * reduce the spew of irrelevant information over this interface, to isolate
3170  * user code from changes in the kernel structure, and potentially to provide
3171  * information-hiding if we decide that some of this information should be
3172  * hidden from users.
3173  */
3174 void
3175 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3176 {
3177 	struct tcpcb *tp = intotcpcb(inp);
3178 	sbintime_t now;
3179 
3180 	if (inp->inp_flags & INP_TIMEWAIT) {
3181 		bzero(xt, sizeof(struct xtcpcb));
3182 		xt->t_state = TCPS_TIME_WAIT;
3183 	} else {
3184 		xt->t_state = tp->t_state;
3185 		xt->t_logstate = tp->t_logstate;
3186 		xt->t_flags = tp->t_flags;
3187 		xt->t_sndzerowin = tp->t_sndzerowin;
3188 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3189 		xt->t_rcvoopack = tp->t_rcvoopack;
3190 
3191 		now = getsbinuptime();
3192 #define	COPYTIMER(ttt)	do {						\
3193 		if (callout_active(&tp->t_timers->ttt))			\
3194 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3195 			    SBT_1MS;					\
3196 		else							\
3197 			xt->ttt = 0;					\
3198 } while (0)
3199 		COPYTIMER(tt_delack);
3200 		COPYTIMER(tt_rexmt);
3201 		COPYTIMER(tt_persist);
3202 		COPYTIMER(tt_keep);
3203 		COPYTIMER(tt_2msl);
3204 #undef COPYTIMER
3205 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3206 
3207 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3208 		    TCP_FUNCTION_NAME_LEN_MAX);
3209 		bzero(xt->xt_logid, TCP_LOG_ID_LEN);
3210 #ifdef TCP_BLACKBOX
3211 		(void)tcp_log_get_id(tp, xt->xt_logid);
3212 #endif
3213 	}
3214 
3215 	xt->xt_len = sizeof(struct xtcpcb);
3216 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3217 	if (inp->inp_socket == NULL)
3218 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3219 }
3220