xref: /freebsd/sys/netinet/tcp_subr.c (revision 3a3deb00a5e449c9478156b162dfa10ec82a2a3f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_kern_tls.h"
41 #include "opt_tcpdebug.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/arb.h>
46 #include <sys/callout.h>
47 #include <sys/eventhandler.h>
48 #ifdef TCP_HHOOK
49 #include <sys/hhook.h>
50 #endif
51 #include <sys/kernel.h>
52 #ifdef TCP_HHOOK
53 #include <sys/khelp.h>
54 #endif
55 #ifdef KERN_TLS
56 #include <sys/ktls.h>
57 #endif
58 #include <sys/qmath.h>
59 #include <sys/stats.h>
60 #include <sys/sysctl.h>
61 #include <sys/jail.h>
62 #include <sys/malloc.h>
63 #include <sys/refcount.h>
64 #include <sys/mbuf.h>
65 #ifdef INET6
66 #include <sys/domain.h>
67 #endif
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/sdt.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/protosw.h>
74 #include <sys/random.h>
75 
76 #include <vm/uma.h>
77 
78 #include <net/route.h>
79 #include <net/if.h>
80 #include <net/if_var.h>
81 #include <net/vnet.h>
82 
83 #include <netinet/in.h>
84 #include <netinet/in_fib.h>
85 #include <netinet/in_kdtrace.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/in_systm.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip.h>
90 #include <netinet/ip_icmp.h>
91 #include <netinet/ip_var.h>
92 #ifdef INET6
93 #include <netinet/icmp6.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/in6_fib.h>
96 #include <netinet6/in6_pcb.h>
97 #include <netinet6/ip6_var.h>
98 #include <netinet6/scope6_var.h>
99 #include <netinet6/nd6.h>
100 #endif
101 
102 #include <netinet/tcp.h>
103 #include <netinet/tcp_fsm.h>
104 #include <netinet/tcp_seq.h>
105 #include <netinet/tcp_timer.h>
106 #include <netinet/tcp_var.h>
107 #include <netinet/tcp_log_buf.h>
108 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_hpts.h>
110 #include <netinet/cc/cc.h>
111 #ifdef INET6
112 #include <netinet6/tcp6_var.h>
113 #endif
114 #include <netinet/tcpip.h>
115 #include <netinet/tcp_fastopen.h>
116 #ifdef TCPPCAP
117 #include <netinet/tcp_pcap.h>
118 #endif
119 #ifdef TCPDEBUG
120 #include <netinet/tcp_debug.h>
121 #endif
122 #ifdef INET6
123 #include <netinet6/ip6protosw.h>
124 #endif
125 #ifdef TCP_OFFLOAD
126 #include <netinet/tcp_offload.h>
127 #endif
128 
129 #include <netipsec/ipsec_support.h>
130 
131 #include <machine/in_cksum.h>
132 #include <crypto/siphash/siphash.h>
133 
134 #include <security/mac/mac_framework.h>
135 
136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
137 #ifdef INET6
138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
139 #endif
140 
141 struct rwlock tcp_function_lock;
142 
143 static int
144 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
145 {
146 	int error, new;
147 
148 	new = V_tcp_mssdflt;
149 	error = sysctl_handle_int(oidp, &new, 0, req);
150 	if (error == 0 && req->newptr) {
151 		if (new < TCP_MINMSS)
152 			error = EINVAL;
153 		else
154 			V_tcp_mssdflt = new;
155 	}
156 	return (error);
157 }
158 
159 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
160     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
161     &sysctl_net_inet_tcp_mss_check, "I",
162     "Default TCP Maximum Segment Size");
163 
164 #ifdef INET6
165 static int
166 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
167 {
168 	int error, new;
169 
170 	new = V_tcp_v6mssdflt;
171 	error = sysctl_handle_int(oidp, &new, 0, req);
172 	if (error == 0 && req->newptr) {
173 		if (new < TCP_MINMSS)
174 			error = EINVAL;
175 		else
176 			V_tcp_v6mssdflt = new;
177 	}
178 	return (error);
179 }
180 
181 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
182     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
183     &sysctl_net_inet_tcp_mss_v6_check, "I",
184    "Default TCP Maximum Segment Size for IPv6");
185 #endif /* INET6 */
186 
187 /*
188  * Minimum MSS we accept and use. This prevents DoS attacks where
189  * we are forced to a ridiculous low MSS like 20 and send hundreds
190  * of packets instead of one. The effect scales with the available
191  * bandwidth and quickly saturates the CPU and network interface
192  * with packet generation and sending. Set to zero to disable MINMSS
193  * checking. This setting prevents us from sending too small packets.
194  */
195 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
197      &VNET_NAME(tcp_minmss), 0,
198     "Minimum TCP Maximum Segment Size");
199 
200 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
201 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
202     &VNET_NAME(tcp_do_rfc1323), 0,
203     "Enable rfc1323 (high performance TCP) extensions");
204 
205 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
207     &VNET_NAME(tcp_ts_offset_per_conn), 0,
208     "Initialize TCP timestamps per connection instead of per host pair");
209 
210 static int	tcp_log_debug = 0;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
212     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
213 
214 static int	tcp_tcbhashsize;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
216     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
217 
218 static int	do_tcpdrain = 1;
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
220     "Enable tcp_drain routine for extra help when low on mbufs");
221 
222 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
223     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
224 
225 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
226 #define	V_icmp_may_rst			VNET(icmp_may_rst)
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
228     &VNET_NAME(icmp_may_rst), 0,
229     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
230 
231 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
232 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
234     &VNET_NAME(tcp_isn_reseed_interval), 0,
235     "Seconds between reseeding of ISN secret");
236 
237 static int	tcp_soreceive_stream;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
239     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
240 
241 VNET_DEFINE(uma_zone_t, sack_hole_zone);
242 #define	V_sack_hole_zone		VNET(sack_hole_zone)
243 
244 #ifdef TCP_HHOOK
245 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
246 #endif
247 
248 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
249 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
250 #define	V_ts_offset_secret	VNET(ts_offset_secret)
251 
252 static int	tcp_default_fb_init(struct tcpcb *tp);
253 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
254 static int	tcp_default_handoff_ok(struct tcpcb *tp);
255 static struct inpcb *tcp_notify(struct inpcb *, int);
256 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
257 static void tcp_mtudisc(struct inpcb *, int);
258 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
259 		    void *ip4hdr, const void *ip6hdr);
260 
261 
262 static struct tcp_function_block tcp_def_funcblk = {
263 	.tfb_tcp_block_name = "freebsd",
264 	.tfb_tcp_output = tcp_output,
265 	.tfb_tcp_do_segment = tcp_do_segment,
266 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
267 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
268 	.tfb_tcp_fb_init = tcp_default_fb_init,
269 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
270 };
271 
272 static int tcp_fb_cnt = 0;
273 struct tcp_funchead t_functions;
274 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
275 
276 static struct tcp_function_block *
277 find_tcp_functions_locked(struct tcp_function_set *fs)
278 {
279 	struct tcp_function *f;
280 	struct tcp_function_block *blk=NULL;
281 
282 	TAILQ_FOREACH(f, &t_functions, tf_next) {
283 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
284 			blk = f->tf_fb;
285 			break;
286 		}
287 	}
288 	return(blk);
289 }
290 
291 static struct tcp_function_block *
292 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
293 {
294 	struct tcp_function_block *rblk=NULL;
295 	struct tcp_function *f;
296 
297 	TAILQ_FOREACH(f, &t_functions, tf_next) {
298 		if (f->tf_fb == blk) {
299 			rblk = blk;
300 			if (s) {
301 				*s = f;
302 			}
303 			break;
304 		}
305 	}
306 	return (rblk);
307 }
308 
309 struct tcp_function_block *
310 find_and_ref_tcp_functions(struct tcp_function_set *fs)
311 {
312 	struct tcp_function_block *blk;
313 
314 	rw_rlock(&tcp_function_lock);
315 	blk = find_tcp_functions_locked(fs);
316 	if (blk)
317 		refcount_acquire(&blk->tfb_refcnt);
318 	rw_runlock(&tcp_function_lock);
319 	return(blk);
320 }
321 
322 struct tcp_function_block *
323 find_and_ref_tcp_fb(struct tcp_function_block *blk)
324 {
325 	struct tcp_function_block *rblk;
326 
327 	rw_rlock(&tcp_function_lock);
328 	rblk = find_tcp_fb_locked(blk, NULL);
329 	if (rblk)
330 		refcount_acquire(&rblk->tfb_refcnt);
331 	rw_runlock(&tcp_function_lock);
332 	return(rblk);
333 }
334 
335 static struct tcp_function_block *
336 find_and_ref_tcp_default_fb(void)
337 {
338 	struct tcp_function_block *rblk;
339 
340 	rw_rlock(&tcp_function_lock);
341 	rblk = tcp_func_set_ptr;
342 	refcount_acquire(&rblk->tfb_refcnt);
343 	rw_runlock(&tcp_function_lock);
344 	return (rblk);
345 }
346 
347 void
348 tcp_switch_back_to_default(struct tcpcb *tp)
349 {
350 	struct tcp_function_block *tfb;
351 
352 	KASSERT(tp->t_fb != &tcp_def_funcblk,
353 	    ("%s: called by the built-in default stack", __func__));
354 
355 	/*
356 	 * Release the old stack. This function will either find a new one
357 	 * or panic.
358 	 */
359 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
360 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
361 	refcount_release(&tp->t_fb->tfb_refcnt);
362 
363 	/*
364 	 * Now, we'll find a new function block to use.
365 	 * Start by trying the current user-selected
366 	 * default, unless this stack is the user-selected
367 	 * default.
368 	 */
369 	tfb = find_and_ref_tcp_default_fb();
370 	if (tfb == tp->t_fb) {
371 		refcount_release(&tfb->tfb_refcnt);
372 		tfb = NULL;
373 	}
374 	/* Does the stack accept this connection? */
375 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
376 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
377 		refcount_release(&tfb->tfb_refcnt);
378 		tfb = NULL;
379 	}
380 	/* Try to use that stack. */
381 	if (tfb != NULL) {
382 		/* Initialize the new stack. If it succeeds, we are done. */
383 		tp->t_fb = tfb;
384 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
385 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
386 			return;
387 
388 		/*
389 		 * Initialization failed. Release the reference count on
390 		 * the stack.
391 		 */
392 		refcount_release(&tfb->tfb_refcnt);
393 	}
394 
395 	/*
396 	 * If that wasn't feasible, use the built-in default
397 	 * stack which is not allowed to reject anyone.
398 	 */
399 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
400 	if (tfb == NULL) {
401 		/* there always should be a default */
402 		panic("Can't refer to tcp_def_funcblk");
403 	}
404 	if (tfb->tfb_tcp_handoff_ok != NULL) {
405 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
406 			/* The default stack cannot say no */
407 			panic("Default stack rejects a new session?");
408 		}
409 	}
410 	tp->t_fb = tfb;
411 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
412 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
413 		/* The default stack cannot fail */
414 		panic("Default stack initialization failed");
415 	}
416 }
417 
418 static int
419 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
420 {
421 	int error=ENOENT;
422 	struct tcp_function_set fs;
423 	struct tcp_function_block *blk;
424 
425 	memset(&fs, 0, sizeof(fs));
426 	rw_rlock(&tcp_function_lock);
427 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
428 	if (blk) {
429 		/* Found him */
430 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
431 		fs.pcbcnt = blk->tfb_refcnt;
432 	}
433 	rw_runlock(&tcp_function_lock);
434 	error = sysctl_handle_string(oidp, fs.function_set_name,
435 				     sizeof(fs.function_set_name), req);
436 
437 	/* Check for error or no change */
438 	if (error != 0 || req->newptr == NULL)
439 		return(error);
440 
441 	rw_wlock(&tcp_function_lock);
442 	blk = find_tcp_functions_locked(&fs);
443 	if ((blk == NULL) ||
444 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
445 		error = ENOENT;
446 		goto done;
447 	}
448 	tcp_func_set_ptr = blk;
449 done:
450 	rw_wunlock(&tcp_function_lock);
451 	return (error);
452 }
453 
454 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
455 	    CTLTYPE_STRING | CTLFLAG_RW,
456 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
457 	    "Set/get the default TCP functions");
458 
459 static int
460 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
461 {
462 	int error, cnt, linesz;
463 	struct tcp_function *f;
464 	char *buffer, *cp;
465 	size_t bufsz, outsz;
466 	bool alias;
467 
468 	cnt = 0;
469 	rw_rlock(&tcp_function_lock);
470 	TAILQ_FOREACH(f, &t_functions, tf_next) {
471 		cnt++;
472 	}
473 	rw_runlock(&tcp_function_lock);
474 
475 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
476 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
477 
478 	error = 0;
479 	cp = buffer;
480 
481 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
482 	    "Alias", "PCB count");
483 	cp += linesz;
484 	bufsz -= linesz;
485 	outsz = linesz;
486 
487 	rw_rlock(&tcp_function_lock);
488 	TAILQ_FOREACH(f, &t_functions, tf_next) {
489 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
490 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
491 		    f->tf_fb->tfb_tcp_block_name,
492 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
493 		    alias ? f->tf_name : "-",
494 		    f->tf_fb->tfb_refcnt);
495 		if (linesz >= bufsz) {
496 			error = EOVERFLOW;
497 			break;
498 		}
499 		cp += linesz;
500 		bufsz -= linesz;
501 		outsz += linesz;
502 	}
503 	rw_runlock(&tcp_function_lock);
504 	if (error == 0)
505 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
506 	free(buffer, M_TEMP);
507 	return (error);
508 }
509 
510 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
511 	    CTLTYPE_STRING|CTLFLAG_RD,
512 	    NULL, 0, sysctl_net_inet_list_available, "A",
513 	    "list available TCP Function sets");
514 
515 /*
516  * Exports one (struct tcp_function_info) for each alias/name.
517  */
518 static int
519 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
520 {
521 	int cnt, error;
522 	struct tcp_function *f;
523 	struct tcp_function_info tfi;
524 
525 	/*
526 	 * We don't allow writes.
527 	 */
528 	if (req->newptr != NULL)
529 		return (EINVAL);
530 
531 	/*
532 	 * Wire the old buffer so we can directly copy the functions to
533 	 * user space without dropping the lock.
534 	 */
535 	if (req->oldptr != NULL) {
536 		error = sysctl_wire_old_buffer(req, 0);
537 		if (error)
538 			return (error);
539 	}
540 
541 	/*
542 	 * Walk the list and copy out matching entries. If INVARIANTS
543 	 * is compiled in, also walk the list to verify the length of
544 	 * the list matches what we have recorded.
545 	 */
546 	rw_rlock(&tcp_function_lock);
547 
548 	cnt = 0;
549 #ifndef INVARIANTS
550 	if (req->oldptr == NULL) {
551 		cnt = tcp_fb_cnt;
552 		goto skip_loop;
553 	}
554 #endif
555 	TAILQ_FOREACH(f, &t_functions, tf_next) {
556 #ifdef INVARIANTS
557 		cnt++;
558 #endif
559 		if (req->oldptr != NULL) {
560 			bzero(&tfi, sizeof(tfi));
561 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
562 			tfi.tfi_id = f->tf_fb->tfb_id;
563 			(void)strlcpy(tfi.tfi_alias, f->tf_name,
564 			    sizeof(tfi.tfi_alias));
565 			(void)strlcpy(tfi.tfi_name,
566 			    f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
567 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
568 			/*
569 			 * Don't stop on error, as that is the
570 			 * mechanism we use to accumulate length
571 			 * information if the buffer was too short.
572 			 */
573 		}
574 	}
575 	KASSERT(cnt == tcp_fb_cnt,
576 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
577 #ifndef INVARIANTS
578 skip_loop:
579 #endif
580 	rw_runlock(&tcp_function_lock);
581 	if (req->oldptr == NULL)
582 		error = SYSCTL_OUT(req, NULL,
583 		    (cnt + 1) * sizeof(struct tcp_function_info));
584 
585 	return (error);
586 }
587 
588 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
589 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
590 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
591 	    "List TCP function block name-to-ID mappings");
592 
593 /*
594  * tfb_tcp_handoff_ok() function for the default stack.
595  * Note that we'll basically try to take all comers.
596  */
597 static int
598 tcp_default_handoff_ok(struct tcpcb *tp)
599 {
600 
601 	return (0);
602 }
603 
604 /*
605  * tfb_tcp_fb_init() function for the default stack.
606  *
607  * This handles making sure we have appropriate timers set if you are
608  * transitioning a socket that has some amount of setup done.
609  *
610  * The init() fuction from the default can *never* return non-zero i.e.
611  * it is required to always succeed since it is the stack of last resort!
612  */
613 static int
614 tcp_default_fb_init(struct tcpcb *tp)
615 {
616 
617 	struct socket *so;
618 
619 	INP_WLOCK_ASSERT(tp->t_inpcb);
620 
621 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
622 	    ("%s: connection %p in unexpected state %d", __func__, tp,
623 	    tp->t_state));
624 
625 	/*
626 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
627 	 * know what to do for unexpected states (which includes TIME_WAIT).
628 	 */
629 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
630 		return (0);
631 
632 	/*
633 	 * Make sure some kind of transmission timer is set if there is
634 	 * outstanding data.
635 	 */
636 	so = tp->t_inpcb->inp_socket;
637 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
638 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
639 	    tcp_timer_active(tp, TT_PERSIST))) {
640 		/*
641 		 * If the session has established and it looks like it should
642 		 * be in the persist state, set the persist timer. Otherwise,
643 		 * set the retransmit timer.
644 		 */
645 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
646 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
647 		    (int32_t)sbavail(&so->so_snd))
648 			tcp_setpersist(tp);
649 		else
650 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
651 	}
652 
653 	/* All non-embryonic sessions get a keepalive timer. */
654 	if (!tcp_timer_active(tp, TT_KEEP))
655 		tcp_timer_activate(tp, TT_KEEP,
656 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
657 		    TP_KEEPINIT(tp));
658 
659 	return (0);
660 }
661 
662 /*
663  * tfb_tcp_fb_fini() function for the default stack.
664  *
665  * This changes state as necessary (or prudent) to prepare for another stack
666  * to assume responsibility for the connection.
667  */
668 static void
669 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
670 {
671 
672 	INP_WLOCK_ASSERT(tp->t_inpcb);
673 	return;
674 }
675 
676 /*
677  * Target size of TCP PCB hash tables. Must be a power of two.
678  *
679  * Note that this can be overridden by the kernel environment
680  * variable net.inet.tcp.tcbhashsize
681  */
682 #ifndef TCBHASHSIZE
683 #define TCBHASHSIZE	0
684 #endif
685 
686 /*
687  * XXX
688  * Callouts should be moved into struct tcp directly.  They are currently
689  * separate because the tcpcb structure is exported to userland for sysctl
690  * parsing purposes, which do not know about callouts.
691  */
692 struct tcpcb_mem {
693 	struct	tcpcb		tcb;
694 	struct	tcp_timer	tt;
695 	struct	cc_var		ccv;
696 #ifdef TCP_HHOOK
697 	struct	osd		osd;
698 #endif
699 };
700 
701 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
702 #define	V_tcpcb_zone			VNET(tcpcb_zone)
703 
704 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
705 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
706 
707 static struct mtx isn_mtx;
708 
709 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
710 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
711 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
712 
713 /*
714  * TCP initialization.
715  */
716 static void
717 tcp_zone_change(void *tag)
718 {
719 
720 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
721 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
722 	tcp_tw_zone_change();
723 }
724 
725 static int
726 tcp_inpcb_init(void *mem, int size, int flags)
727 {
728 	struct inpcb *inp = mem;
729 
730 	INP_LOCK_INIT(inp, "inp", "tcpinp");
731 	return (0);
732 }
733 
734 /*
735  * Take a value and get the next power of 2 that doesn't overflow.
736  * Used to size the tcp_inpcb hash buckets.
737  */
738 static int
739 maketcp_hashsize(int size)
740 {
741 	int hashsize;
742 
743 	/*
744 	 * auto tune.
745 	 * get the next power of 2 higher than maxsockets.
746 	 */
747 	hashsize = 1 << fls(size);
748 	/* catch overflow, and just go one power of 2 smaller */
749 	if (hashsize < size) {
750 		hashsize = 1 << (fls(size) - 1);
751 	}
752 	return (hashsize);
753 }
754 
755 static volatile int next_tcp_stack_id = 1;
756 
757 /*
758  * Register a TCP function block with the name provided in the names
759  * array.  (Note that this function does NOT automatically register
760  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
761  * explicitly include blk->tfb_tcp_block_name in the list of names if
762  * you wish to register the stack with that name.)
763  *
764  * Either all name registrations will succeed or all will fail.  If
765  * a name registration fails, the function will update the num_names
766  * argument to point to the array index of the name that encountered
767  * the failure.
768  *
769  * Returns 0 on success, or an error code on failure.
770  */
771 int
772 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
773     const char *names[], int *num_names)
774 {
775 	struct tcp_function *n;
776 	struct tcp_function_set fs;
777 	int error, i;
778 
779 	KASSERT(names != NULL && *num_names > 0,
780 	    ("%s: Called with 0-length name list", __func__));
781 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
782 	KASSERT(rw_initialized(&tcp_function_lock),
783 	    ("%s: called too early", __func__));
784 
785 	if ((blk->tfb_tcp_output == NULL) ||
786 	    (blk->tfb_tcp_do_segment == NULL) ||
787 	    (blk->tfb_tcp_ctloutput == NULL) ||
788 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
789 		/*
790 		 * These functions are required and you
791 		 * need a name.
792 		 */
793 		*num_names = 0;
794 		return (EINVAL);
795 	}
796 	if (blk->tfb_tcp_timer_stop_all ||
797 	    blk->tfb_tcp_timer_activate ||
798 	    blk->tfb_tcp_timer_active ||
799 	    blk->tfb_tcp_timer_stop) {
800 		/*
801 		 * If you define one timer function you
802 		 * must have them all.
803 		 */
804 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
805 		    (blk->tfb_tcp_timer_activate == NULL) ||
806 		    (blk->tfb_tcp_timer_active == NULL) ||
807 		    (blk->tfb_tcp_timer_stop == NULL)) {
808 			*num_names = 0;
809 			return (EINVAL);
810 		}
811 	}
812 
813 	if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
814 		*num_names = 0;
815 		return (EINVAL);
816 	}
817 
818 	refcount_init(&blk->tfb_refcnt, 0);
819 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
820 	for (i = 0; i < *num_names; i++) {
821 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
822 		if (n == NULL) {
823 			error = ENOMEM;
824 			goto cleanup;
825 		}
826 		n->tf_fb = blk;
827 
828 		(void)strlcpy(fs.function_set_name, names[i],
829 		    sizeof(fs.function_set_name));
830 		rw_wlock(&tcp_function_lock);
831 		if (find_tcp_functions_locked(&fs) != NULL) {
832 			/* Duplicate name space not allowed */
833 			rw_wunlock(&tcp_function_lock);
834 			free(n, M_TCPFUNCTIONS);
835 			error = EALREADY;
836 			goto cleanup;
837 		}
838 		(void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
839 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
840 		tcp_fb_cnt++;
841 		rw_wunlock(&tcp_function_lock);
842 	}
843 	return(0);
844 
845 cleanup:
846 	/*
847 	 * Deregister the names we just added. Because registration failed
848 	 * for names[i], we don't need to deregister that name.
849 	 */
850 	*num_names = i;
851 	rw_wlock(&tcp_function_lock);
852 	while (--i >= 0) {
853 		TAILQ_FOREACH(n, &t_functions, tf_next) {
854 			if (!strncmp(n->tf_name, names[i],
855 			    TCP_FUNCTION_NAME_LEN_MAX)) {
856 				TAILQ_REMOVE(&t_functions, n, tf_next);
857 				tcp_fb_cnt--;
858 				n->tf_fb = NULL;
859 				free(n, M_TCPFUNCTIONS);
860 				break;
861 			}
862 		}
863 	}
864 	rw_wunlock(&tcp_function_lock);
865 	return (error);
866 }
867 
868 /*
869  * Register a TCP function block using the name provided in the name
870  * argument.
871  *
872  * Returns 0 on success, or an error code on failure.
873  */
874 int
875 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
876     int wait)
877 {
878 	const char *name_list[1];
879 	int num_names, rv;
880 
881 	num_names = 1;
882 	if (name != NULL)
883 		name_list[0] = name;
884 	else
885 		name_list[0] = blk->tfb_tcp_block_name;
886 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
887 	return (rv);
888 }
889 
890 /*
891  * Register a TCP function block using the name defined in
892  * blk->tfb_tcp_block_name.
893  *
894  * Returns 0 on success, or an error code on failure.
895  */
896 int
897 register_tcp_functions(struct tcp_function_block *blk, int wait)
898 {
899 
900 	return (register_tcp_functions_as_name(blk, NULL, wait));
901 }
902 
903 /*
904  * Deregister all names associated with a function block. This
905  * functionally removes the function block from use within the system.
906  *
907  * When called with a true quiesce argument, mark the function block
908  * as being removed so no more stacks will use it and determine
909  * whether the removal would succeed.
910  *
911  * When called with a false quiesce argument, actually attempt the
912  * removal.
913  *
914  * When called with a force argument, attempt to switch all TCBs to
915  * use the default stack instead of returning EBUSY.
916  *
917  * Returns 0 on success (or if the removal would succeed, or an error
918  * code on failure.
919  */
920 int
921 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
922     bool force)
923 {
924 	struct tcp_function *f;
925 
926 	if (blk == &tcp_def_funcblk) {
927 		/* You can't un-register the default */
928 		return (EPERM);
929 	}
930 	rw_wlock(&tcp_function_lock);
931 	if (blk == tcp_func_set_ptr) {
932 		/* You can't free the current default */
933 		rw_wunlock(&tcp_function_lock);
934 		return (EBUSY);
935 	}
936 	/* Mark the block so no more stacks can use it. */
937 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
938 	/*
939 	 * If TCBs are still attached to the stack, attempt to switch them
940 	 * to the default stack.
941 	 */
942 	if (force && blk->tfb_refcnt) {
943 		struct inpcb *inp;
944 		struct tcpcb *tp;
945 		VNET_ITERATOR_DECL(vnet_iter);
946 
947 		rw_wunlock(&tcp_function_lock);
948 
949 		VNET_LIST_RLOCK();
950 		VNET_FOREACH(vnet_iter) {
951 			CURVNET_SET(vnet_iter);
952 			INP_INFO_WLOCK(&V_tcbinfo);
953 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
954 				INP_WLOCK(inp);
955 				if (inp->inp_flags & INP_TIMEWAIT) {
956 					INP_WUNLOCK(inp);
957 					continue;
958 				}
959 				tp = intotcpcb(inp);
960 				if (tp == NULL || tp->t_fb != blk) {
961 					INP_WUNLOCK(inp);
962 					continue;
963 				}
964 				tcp_switch_back_to_default(tp);
965 				INP_WUNLOCK(inp);
966 			}
967 			INP_INFO_WUNLOCK(&V_tcbinfo);
968 			CURVNET_RESTORE();
969 		}
970 		VNET_LIST_RUNLOCK();
971 
972 		rw_wlock(&tcp_function_lock);
973 	}
974 	if (blk->tfb_refcnt) {
975 		/* TCBs still attached. */
976 		rw_wunlock(&tcp_function_lock);
977 		return (EBUSY);
978 	}
979 	if (quiesce) {
980 		/* Skip removal. */
981 		rw_wunlock(&tcp_function_lock);
982 		return (0);
983 	}
984 	/* Remove any function names that map to this function block. */
985 	while (find_tcp_fb_locked(blk, &f) != NULL) {
986 		TAILQ_REMOVE(&t_functions, f, tf_next);
987 		tcp_fb_cnt--;
988 		f->tf_fb = NULL;
989 		free(f, M_TCPFUNCTIONS);
990 	}
991 	rw_wunlock(&tcp_function_lock);
992 	return (0);
993 }
994 
995 void
996 tcp_init(void)
997 {
998 	const char *tcbhash_tuneable;
999 	int hashsize;
1000 
1001 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1002 
1003 #ifdef TCP_HHOOK
1004 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1005 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1006 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1007 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1008 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1009 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1010 #endif
1011 #ifdef STATS
1012 	if (tcp_stats_init())
1013 		printf("%s: WARNING: unable to initialise TCP stats\n",
1014 		    __func__);
1015 #endif
1016 	hashsize = TCBHASHSIZE;
1017 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1018 	if (hashsize == 0) {
1019 		/*
1020 		 * Auto tune the hash size based on maxsockets.
1021 		 * A perfect hash would have a 1:1 mapping
1022 		 * (hashsize = maxsockets) however it's been
1023 		 * suggested that O(2) average is better.
1024 		 */
1025 		hashsize = maketcp_hashsize(maxsockets / 4);
1026 		/*
1027 		 * Our historical default is 512,
1028 		 * do not autotune lower than this.
1029 		 */
1030 		if (hashsize < 512)
1031 			hashsize = 512;
1032 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1033 			printf("%s: %s auto tuned to %d\n", __func__,
1034 			    tcbhash_tuneable, hashsize);
1035 	}
1036 	/*
1037 	 * We require a hashsize to be a power of two.
1038 	 * Previously if it was not a power of two we would just reset it
1039 	 * back to 512, which could be a nasty surprise if you did not notice
1040 	 * the error message.
1041 	 * Instead what we do is clip it to the closest power of two lower
1042 	 * than the specified hash value.
1043 	 */
1044 	if (!powerof2(hashsize)) {
1045 		int oldhashsize = hashsize;
1046 
1047 		hashsize = maketcp_hashsize(hashsize);
1048 		/* prevent absurdly low value */
1049 		if (hashsize < 16)
1050 			hashsize = 16;
1051 		printf("%s: WARNING: TCB hash size not a power of 2, "
1052 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1053 		    hashsize);
1054 	}
1055 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1056 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1057 
1058 	/*
1059 	 * These have to be type stable for the benefit of the timers.
1060 	 */
1061 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1062 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1063 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1064 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1065 
1066 	tcp_tw_init();
1067 	syncache_init();
1068 	tcp_hc_init();
1069 
1070 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1071 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1072 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1073 
1074 	tcp_fastopen_init();
1075 
1076 	/* Skip initialization of globals for non-default instances. */
1077 	if (!IS_DEFAULT_VNET(curvnet))
1078 		return;
1079 
1080 	tcp_reass_global_init();
1081 
1082 	/* XXX virtualize those bellow? */
1083 	tcp_delacktime = TCPTV_DELACK;
1084 	tcp_keepinit = TCPTV_KEEP_INIT;
1085 	tcp_keepidle = TCPTV_KEEP_IDLE;
1086 	tcp_keepintvl = TCPTV_KEEPINTVL;
1087 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1088 	tcp_msl = TCPTV_MSL;
1089 	tcp_rexmit_initial = TCPTV_RTOBASE;
1090 	if (tcp_rexmit_initial < 1)
1091 		tcp_rexmit_initial = 1;
1092 	tcp_rexmit_min = TCPTV_MIN;
1093 	if (tcp_rexmit_min < 1)
1094 		tcp_rexmit_min = 1;
1095 	tcp_persmin = TCPTV_PERSMIN;
1096 	tcp_persmax = TCPTV_PERSMAX;
1097 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1098 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1099 	tcp_tcbhashsize = hashsize;
1100 
1101 	/* Setup the tcp function block list */
1102 	TAILQ_INIT(&t_functions);
1103 	rw_init(&tcp_function_lock, "tcp_func_lock");
1104 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1105 #ifdef TCP_BLACKBOX
1106 	/* Initialize the TCP logging data. */
1107 	tcp_log_init();
1108 #endif
1109 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1110 
1111 	if (tcp_soreceive_stream) {
1112 #ifdef INET
1113 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1114 #endif
1115 #ifdef INET6
1116 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1117 #endif /* INET6 */
1118 	}
1119 
1120 #ifdef INET6
1121 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1122 #else /* INET6 */
1123 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1124 #endif /* INET6 */
1125 	if (max_protohdr < TCP_MINPROTOHDR)
1126 		max_protohdr = TCP_MINPROTOHDR;
1127 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1128 		panic("tcp_init");
1129 #undef TCP_MINPROTOHDR
1130 
1131 	ISN_LOCK_INIT();
1132 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1133 		SHUTDOWN_PRI_DEFAULT);
1134 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1135 		EVENTHANDLER_PRI_ANY);
1136 
1137 	tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1138 	tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1139 	tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1140 	tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK);
1141 	tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1142 	tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK);
1143 #ifdef TCPPCAP
1144 	tcp_pcap_init();
1145 #endif
1146 }
1147 
1148 #ifdef VIMAGE
1149 static void
1150 tcp_destroy(void *unused __unused)
1151 {
1152 	int n;
1153 #ifdef TCP_HHOOK
1154 	int error;
1155 #endif
1156 
1157 	/*
1158 	 * All our processes are gone, all our sockets should be cleaned
1159 	 * up, which means, we should be past the tcp_discardcb() calls.
1160 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1161 	 */
1162 	for (;;) {
1163 		INP_LIST_RLOCK(&V_tcbinfo);
1164 		n = V_tcbinfo.ipi_count;
1165 		INP_LIST_RUNLOCK(&V_tcbinfo);
1166 		if (n == 0)
1167 			break;
1168 		pause("tcpdes", hz / 10);
1169 	}
1170 	tcp_hc_destroy();
1171 	syncache_destroy();
1172 	tcp_tw_destroy();
1173 	in_pcbinfo_destroy(&V_tcbinfo);
1174 	/* tcp_discardcb() clears the sack_holes up. */
1175 	uma_zdestroy(V_sack_hole_zone);
1176 	uma_zdestroy(V_tcpcb_zone);
1177 
1178 	/*
1179 	 * Cannot free the zone until all tcpcbs are released as we attach
1180 	 * the allocations to them.
1181 	 */
1182 	tcp_fastopen_destroy();
1183 
1184 #ifdef TCP_HHOOK
1185 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1186 	if (error != 0) {
1187 		printf("%s: WARNING: unable to deregister helper hook "
1188 		    "type=%d, id=%d: error %d returned\n", __func__,
1189 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1190 	}
1191 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1192 	if (error != 0) {
1193 		printf("%s: WARNING: unable to deregister helper hook "
1194 		    "type=%d, id=%d: error %d returned\n", __func__,
1195 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1196 	}
1197 #endif
1198 }
1199 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1200 #endif
1201 
1202 void
1203 tcp_fini(void *xtp)
1204 {
1205 
1206 }
1207 
1208 /*
1209  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1210  * tcp_template used to store this data in mbufs, but we now recopy it out
1211  * of the tcpcb each time to conserve mbufs.
1212  */
1213 void
1214 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1215 {
1216 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1217 
1218 	INP_WLOCK_ASSERT(inp);
1219 
1220 #ifdef INET6
1221 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1222 		struct ip6_hdr *ip6;
1223 
1224 		ip6 = (struct ip6_hdr *)ip_ptr;
1225 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1226 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1227 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1228 			(IPV6_VERSION & IPV6_VERSION_MASK);
1229 		ip6->ip6_nxt = IPPROTO_TCP;
1230 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1231 		ip6->ip6_src = inp->in6p_laddr;
1232 		ip6->ip6_dst = inp->in6p_faddr;
1233 	}
1234 #endif /* INET6 */
1235 #if defined(INET6) && defined(INET)
1236 	else
1237 #endif
1238 #ifdef INET
1239 	{
1240 		struct ip *ip;
1241 
1242 		ip = (struct ip *)ip_ptr;
1243 		ip->ip_v = IPVERSION;
1244 		ip->ip_hl = 5;
1245 		ip->ip_tos = inp->inp_ip_tos;
1246 		ip->ip_len = 0;
1247 		ip->ip_id = 0;
1248 		ip->ip_off = 0;
1249 		ip->ip_ttl = inp->inp_ip_ttl;
1250 		ip->ip_sum = 0;
1251 		ip->ip_p = IPPROTO_TCP;
1252 		ip->ip_src = inp->inp_laddr;
1253 		ip->ip_dst = inp->inp_faddr;
1254 	}
1255 #endif /* INET */
1256 	th->th_sport = inp->inp_lport;
1257 	th->th_dport = inp->inp_fport;
1258 	th->th_seq = 0;
1259 	th->th_ack = 0;
1260 	th->th_x2 = 0;
1261 	th->th_off = 5;
1262 	th->th_flags = 0;
1263 	th->th_win = 0;
1264 	th->th_urp = 0;
1265 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1266 }
1267 
1268 /*
1269  * Create template to be used to send tcp packets on a connection.
1270  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1271  * use for this function is in keepalives, which use tcp_respond.
1272  */
1273 struct tcptemp *
1274 tcpip_maketemplate(struct inpcb *inp)
1275 {
1276 	struct tcptemp *t;
1277 
1278 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1279 	if (t == NULL)
1280 		return (NULL);
1281 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1282 	return (t);
1283 }
1284 
1285 /*
1286  * Send a single message to the TCP at address specified by
1287  * the given TCP/IP header.  If m == NULL, then we make a copy
1288  * of the tcpiphdr at th and send directly to the addressed host.
1289  * This is used to force keep alive messages out using the TCP
1290  * template for a connection.  If flags are given then we send
1291  * a message back to the TCP which originated the segment th,
1292  * and discard the mbuf containing it and any other attached mbufs.
1293  *
1294  * In any case the ack and sequence number of the transmitted
1295  * segment are as specified by the parameters.
1296  *
1297  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1298  */
1299 void
1300 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1301     tcp_seq ack, tcp_seq seq, int flags)
1302 {
1303 	struct tcpopt to;
1304 	struct inpcb *inp;
1305 	struct ip *ip;
1306 	struct mbuf *optm;
1307 	struct tcphdr *nth;
1308 	u_char *optp;
1309 #ifdef INET6
1310 	struct ip6_hdr *ip6;
1311 	int isipv6;
1312 #endif /* INET6 */
1313 	int optlen, tlen, win;
1314 	bool incl_opts;
1315 
1316 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1317 
1318 #ifdef INET6
1319 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1320 	ip6 = ipgen;
1321 #endif /* INET6 */
1322 	ip = ipgen;
1323 
1324 	if (tp != NULL) {
1325 		inp = tp->t_inpcb;
1326 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1327 		INP_WLOCK_ASSERT(inp);
1328 	} else
1329 		inp = NULL;
1330 
1331 	incl_opts = false;
1332 	win = 0;
1333 	if (tp != NULL) {
1334 		if (!(flags & TH_RST)) {
1335 			win = sbspace(&inp->inp_socket->so_rcv);
1336 			if (win > TCP_MAXWIN << tp->rcv_scale)
1337 				win = TCP_MAXWIN << tp->rcv_scale;
1338 		}
1339 		if ((tp->t_flags & TF_NOOPT) == 0)
1340 			incl_opts = true;
1341 	}
1342 	if (m == NULL) {
1343 		m = m_gethdr(M_NOWAIT, MT_DATA);
1344 		if (m == NULL)
1345 			return;
1346 		m->m_data += max_linkhdr;
1347 #ifdef INET6
1348 		if (isipv6) {
1349 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1350 			      sizeof(struct ip6_hdr));
1351 			ip6 = mtod(m, struct ip6_hdr *);
1352 			nth = (struct tcphdr *)(ip6 + 1);
1353 		} else
1354 #endif /* INET6 */
1355 		{
1356 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1357 			ip = mtod(m, struct ip *);
1358 			nth = (struct tcphdr *)(ip + 1);
1359 		}
1360 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1361 		flags = TH_ACK;
1362 	} else if (!M_WRITABLE(m)) {
1363 		struct mbuf *n;
1364 
1365 		/* Can't reuse 'm', allocate a new mbuf. */
1366 		n = m_gethdr(M_NOWAIT, MT_DATA);
1367 		if (n == NULL) {
1368 			m_freem(m);
1369 			return;
1370 		}
1371 
1372 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1373 			m_freem(m);
1374 			m_freem(n);
1375 			return;
1376 		}
1377 
1378 		n->m_data += max_linkhdr;
1379 		/* m_len is set later */
1380 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1381 #ifdef INET6
1382 		if (isipv6) {
1383 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1384 			      sizeof(struct ip6_hdr));
1385 			ip6 = mtod(n, struct ip6_hdr *);
1386 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1387 			nth = (struct tcphdr *)(ip6 + 1);
1388 		} else
1389 #endif /* INET6 */
1390 		{
1391 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1392 			ip = mtod(n, struct ip *);
1393 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1394 			nth = (struct tcphdr *)(ip + 1);
1395 		}
1396 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1397 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1398 		th = nth;
1399 		m_freem(m);
1400 		m = n;
1401 	} else {
1402 		/*
1403 		 *  reuse the mbuf.
1404 		 * XXX MRT We inherit the FIB, which is lucky.
1405 		 */
1406 		m_freem(m->m_next);
1407 		m->m_next = NULL;
1408 		m->m_data = (caddr_t)ipgen;
1409 		/* m_len is set later */
1410 #ifdef INET6
1411 		if (isipv6) {
1412 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1413 			nth = (struct tcphdr *)(ip6 + 1);
1414 		} else
1415 #endif /* INET6 */
1416 		{
1417 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1418 			nth = (struct tcphdr *)(ip + 1);
1419 		}
1420 		if (th != nth) {
1421 			/*
1422 			 * this is usually a case when an extension header
1423 			 * exists between the IPv6 header and the
1424 			 * TCP header.
1425 			 */
1426 			nth->th_sport = th->th_sport;
1427 			nth->th_dport = th->th_dport;
1428 		}
1429 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1430 #undef xchg
1431 	}
1432 	tlen = 0;
1433 #ifdef INET6
1434 	if (isipv6)
1435 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1436 #endif
1437 #if defined(INET) && defined(INET6)
1438 	else
1439 #endif
1440 #ifdef INET
1441 		tlen = sizeof (struct tcpiphdr);
1442 #endif
1443 #ifdef INVARIANTS
1444 	m->m_len = 0;
1445 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1446 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1447 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1448 #endif
1449 	m->m_len = tlen;
1450 	to.to_flags = 0;
1451 	if (incl_opts) {
1452 		/* Make sure we have room. */
1453 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1454 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1455 			if (m->m_next) {
1456 				optp = mtod(m->m_next, u_char *);
1457 				optm = m->m_next;
1458 			} else
1459 				incl_opts = false;
1460 		} else {
1461 			optp = (u_char *) (nth + 1);
1462 			optm = m;
1463 		}
1464 	}
1465 	if (incl_opts) {
1466 		/* Timestamps. */
1467 		if (tp->t_flags & TF_RCVD_TSTMP) {
1468 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1469 			to.to_tsecr = tp->ts_recent;
1470 			to.to_flags |= TOF_TS;
1471 		}
1472 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1473 		/* TCP-MD5 (RFC2385). */
1474 		if (tp->t_flags & TF_SIGNATURE)
1475 			to.to_flags |= TOF_SIGNATURE;
1476 #endif
1477 		/* Add the options. */
1478 		tlen += optlen = tcp_addoptions(&to, optp);
1479 
1480 		/* Update m_len in the correct mbuf. */
1481 		optm->m_len += optlen;
1482 	} else
1483 		optlen = 0;
1484 #ifdef INET6
1485 	if (isipv6) {
1486 		ip6->ip6_flow = 0;
1487 		ip6->ip6_vfc = IPV6_VERSION;
1488 		ip6->ip6_nxt = IPPROTO_TCP;
1489 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1490 	}
1491 #endif
1492 #if defined(INET) && defined(INET6)
1493 	else
1494 #endif
1495 #ifdef INET
1496 	{
1497 		ip->ip_len = htons(tlen);
1498 		ip->ip_ttl = V_ip_defttl;
1499 		if (V_path_mtu_discovery)
1500 			ip->ip_off |= htons(IP_DF);
1501 	}
1502 #endif
1503 	m->m_pkthdr.len = tlen;
1504 	m->m_pkthdr.rcvif = NULL;
1505 #ifdef MAC
1506 	if (inp != NULL) {
1507 		/*
1508 		 * Packet is associated with a socket, so allow the
1509 		 * label of the response to reflect the socket label.
1510 		 */
1511 		INP_WLOCK_ASSERT(inp);
1512 		mac_inpcb_create_mbuf(inp, m);
1513 	} else {
1514 		/*
1515 		 * Packet is not associated with a socket, so possibly
1516 		 * update the label in place.
1517 		 */
1518 		mac_netinet_tcp_reply(m);
1519 	}
1520 #endif
1521 	nth->th_seq = htonl(seq);
1522 	nth->th_ack = htonl(ack);
1523 	nth->th_x2 = 0;
1524 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1525 	nth->th_flags = flags;
1526 	if (tp != NULL)
1527 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1528 	else
1529 		nth->th_win = htons((u_short)win);
1530 	nth->th_urp = 0;
1531 
1532 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1533 	if (to.to_flags & TOF_SIGNATURE) {
1534 		if (!TCPMD5_ENABLED() ||
1535 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1536 			m_freem(m);
1537 			return;
1538 		}
1539 	}
1540 #endif
1541 
1542 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1543 #ifdef INET6
1544 	if (isipv6) {
1545 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1546 		nth->th_sum = in6_cksum_pseudo(ip6,
1547 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1548 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1549 		    NULL, NULL);
1550 	}
1551 #endif /* INET6 */
1552 #if defined(INET6) && defined(INET)
1553 	else
1554 #endif
1555 #ifdef INET
1556 	{
1557 		m->m_pkthdr.csum_flags = CSUM_TCP;
1558 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1559 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1560 	}
1561 #endif /* INET */
1562 #ifdef TCPDEBUG
1563 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1564 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1565 #endif
1566 	TCP_PROBE3(debug__output, tp, th, m);
1567 	if (flags & TH_RST)
1568 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1569 
1570 #ifdef INET6
1571 	if (isipv6) {
1572 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1573 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1574 	}
1575 #endif /* INET6 */
1576 #if defined(INET) && defined(INET6)
1577 	else
1578 #endif
1579 #ifdef INET
1580 	{
1581 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1582 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1583 	}
1584 #endif
1585 }
1586 
1587 /*
1588  * Create a new TCP control block, making an
1589  * empty reassembly queue and hooking it to the argument
1590  * protocol control block.  The `inp' parameter must have
1591  * come from the zone allocator set up in tcp_init().
1592  */
1593 struct tcpcb *
1594 tcp_newtcpcb(struct inpcb *inp)
1595 {
1596 	struct tcpcb_mem *tm;
1597 	struct tcpcb *tp;
1598 #ifdef INET6
1599 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1600 #endif /* INET6 */
1601 
1602 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1603 	if (tm == NULL)
1604 		return (NULL);
1605 	tp = &tm->tcb;
1606 
1607 	/* Initialise cc_var struct for this tcpcb. */
1608 	tp->ccv = &tm->ccv;
1609 	tp->ccv->type = IPPROTO_TCP;
1610 	tp->ccv->ccvc.tcp = tp;
1611 	rw_rlock(&tcp_function_lock);
1612 	tp->t_fb = tcp_func_set_ptr;
1613 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1614 	rw_runlock(&tcp_function_lock);
1615 	/*
1616 	 * Use the current system default CC algorithm.
1617 	 */
1618 	CC_LIST_RLOCK();
1619 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1620 	CC_ALGO(tp) = CC_DEFAULT();
1621 	CC_LIST_RUNLOCK();
1622 
1623 	if (CC_ALGO(tp)->cb_init != NULL)
1624 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1625 			if (tp->t_fb->tfb_tcp_fb_fini)
1626 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1627 			refcount_release(&tp->t_fb->tfb_refcnt);
1628 			uma_zfree(V_tcpcb_zone, tm);
1629 			return (NULL);
1630 		}
1631 
1632 #ifdef TCP_HHOOK
1633 	tp->osd = &tm->osd;
1634 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1635 		if (tp->t_fb->tfb_tcp_fb_fini)
1636 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1637 		refcount_release(&tp->t_fb->tfb_refcnt);
1638 		uma_zfree(V_tcpcb_zone, tm);
1639 		return (NULL);
1640 	}
1641 #endif
1642 
1643 #ifdef VIMAGE
1644 	tp->t_vnet = inp->inp_vnet;
1645 #endif
1646 	tp->t_timers = &tm->tt;
1647 	TAILQ_INIT(&tp->t_segq);
1648 	tp->t_maxseg =
1649 #ifdef INET6
1650 		isipv6 ? V_tcp_v6mssdflt :
1651 #endif /* INET6 */
1652 		V_tcp_mssdflt;
1653 
1654 	/* Set up our timeouts. */
1655 	callout_init(&tp->t_timers->tt_rexmt, 1);
1656 	callout_init(&tp->t_timers->tt_persist, 1);
1657 	callout_init(&tp->t_timers->tt_keep, 1);
1658 	callout_init(&tp->t_timers->tt_2msl, 1);
1659 	callout_init(&tp->t_timers->tt_delack, 1);
1660 
1661 	if (V_tcp_do_rfc1323)
1662 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1663 	if (V_tcp_do_sack)
1664 		tp->t_flags |= TF_SACK_PERMIT;
1665 	TAILQ_INIT(&tp->snd_holes);
1666 	/*
1667 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1668 	 * is called.
1669 	 */
1670 	in_pcbref(inp);	/* Reference for tcpcb */
1671 	tp->t_inpcb = inp;
1672 
1673 	/*
1674 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1675 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1676 	 * reasonable initial retransmit time.
1677 	 */
1678 	tp->t_srtt = TCPTV_SRTTBASE;
1679 	tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1680 	tp->t_rttmin = tcp_rexmit_min;
1681 	tp->t_rxtcur = tcp_rexmit_initial;
1682 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1683 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1684 	tp->t_rcvtime = ticks;
1685 	/*
1686 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1687 	 * because the socket may be bound to an IPv6 wildcard address,
1688 	 * which may match an IPv4-mapped IPv6 address.
1689 	 */
1690 	inp->inp_ip_ttl = V_ip_defttl;
1691 	inp->inp_ppcb = tp;
1692 #ifdef TCPPCAP
1693 	/*
1694 	 * Init the TCP PCAP queues.
1695 	 */
1696 	tcp_pcap_tcpcb_init(tp);
1697 #endif
1698 #ifdef TCP_BLACKBOX
1699 	/* Initialize the per-TCPCB log data. */
1700 	tcp_log_tcpcbinit(tp);
1701 #endif
1702 	if (tp->t_fb->tfb_tcp_fb_init) {
1703 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1704 	}
1705 #ifdef STATS
1706 	if (V_tcp_perconn_stats_enable == 1)
1707 		tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
1708 #endif
1709 	return (tp);		/* XXX */
1710 }
1711 
1712 /*
1713  * Switch the congestion control algorithm back to NewReno for any active
1714  * control blocks using an algorithm which is about to go away.
1715  * This ensures the CC framework can allow the unload to proceed without leaving
1716  * any dangling pointers which would trigger a panic.
1717  * Returning non-zero would inform the CC framework that something went wrong
1718  * and it would be unsafe to allow the unload to proceed. However, there is no
1719  * way for this to occur with this implementation so we always return zero.
1720  */
1721 int
1722 tcp_ccalgounload(struct cc_algo *unload_algo)
1723 {
1724 	struct cc_algo *tmpalgo;
1725 	struct inpcb *inp;
1726 	struct tcpcb *tp;
1727 	VNET_ITERATOR_DECL(vnet_iter);
1728 
1729 	/*
1730 	 * Check all active control blocks across all network stacks and change
1731 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1732 	 * requires cleanup code to be run, call it.
1733 	 */
1734 	VNET_LIST_RLOCK();
1735 	VNET_FOREACH(vnet_iter) {
1736 		CURVNET_SET(vnet_iter);
1737 		INP_INFO_WLOCK(&V_tcbinfo);
1738 		/*
1739 		 * New connections already part way through being initialised
1740 		 * with the CC algo we're removing will not race with this code
1741 		 * because the INP_INFO_WLOCK is held during initialisation. We
1742 		 * therefore don't enter the loop below until the connection
1743 		 * list has stabilised.
1744 		 */
1745 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1746 			INP_WLOCK(inp);
1747 			/* Important to skip tcptw structs. */
1748 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1749 			    (tp = intotcpcb(inp)) != NULL) {
1750 				/*
1751 				 * By holding INP_WLOCK here, we are assured
1752 				 * that the connection is not currently
1753 				 * executing inside the CC module's functions
1754 				 * i.e. it is safe to make the switch back to
1755 				 * NewReno.
1756 				 */
1757 				if (CC_ALGO(tp) == unload_algo) {
1758 					tmpalgo = CC_ALGO(tp);
1759 					if (tmpalgo->cb_destroy != NULL)
1760 						tmpalgo->cb_destroy(tp->ccv);
1761 					CC_DATA(tp) = NULL;
1762 					/*
1763 					 * NewReno may allocate memory on
1764 					 * demand for certain stateful
1765 					 * configuration as needed, but is
1766 					 * coded to never fail on memory
1767 					 * allocation failure so it is a safe
1768 					 * fallback.
1769 					 */
1770 					CC_ALGO(tp) = &newreno_cc_algo;
1771 				}
1772 			}
1773 			INP_WUNLOCK(inp);
1774 		}
1775 		INP_INFO_WUNLOCK(&V_tcbinfo);
1776 		CURVNET_RESTORE();
1777 	}
1778 	VNET_LIST_RUNLOCK();
1779 
1780 	return (0);
1781 }
1782 
1783 /*
1784  * Drop a TCP connection, reporting
1785  * the specified error.  If connection is synchronized,
1786  * then send a RST to peer.
1787  */
1788 struct tcpcb *
1789 tcp_drop(struct tcpcb *tp, int errno)
1790 {
1791 	struct socket *so = tp->t_inpcb->inp_socket;
1792 
1793 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1794 	INP_WLOCK_ASSERT(tp->t_inpcb);
1795 
1796 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1797 		tcp_state_change(tp, TCPS_CLOSED);
1798 		(void) tp->t_fb->tfb_tcp_output(tp);
1799 		TCPSTAT_INC(tcps_drops);
1800 	} else
1801 		TCPSTAT_INC(tcps_conndrops);
1802 	if (errno == ETIMEDOUT && tp->t_softerror)
1803 		errno = tp->t_softerror;
1804 	so->so_error = errno;
1805 	return (tcp_close(tp));
1806 }
1807 
1808 void
1809 tcp_discardcb(struct tcpcb *tp)
1810 {
1811 	struct inpcb *inp = tp->t_inpcb;
1812 	struct socket *so = inp->inp_socket;
1813 #ifdef INET6
1814 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1815 #endif /* INET6 */
1816 	int released __unused;
1817 
1818 	INP_WLOCK_ASSERT(inp);
1819 
1820 	/*
1821 	 * Make sure that all of our timers are stopped before we delete the
1822 	 * PCB.
1823 	 *
1824 	 * If stopping a timer fails, we schedule a discard function in same
1825 	 * callout, and the last discard function called will take care of
1826 	 * deleting the tcpcb.
1827 	 */
1828 	tp->t_timers->tt_draincnt = 0;
1829 	tcp_timer_stop(tp, TT_REXMT);
1830 	tcp_timer_stop(tp, TT_PERSIST);
1831 	tcp_timer_stop(tp, TT_KEEP);
1832 	tcp_timer_stop(tp, TT_2MSL);
1833 	tcp_timer_stop(tp, TT_DELACK);
1834 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1835 		/*
1836 		 * Call the stop-all function of the methods,
1837 		 * this function should call the tcp_timer_stop()
1838 		 * method with each of the function specific timeouts.
1839 		 * That stop will be called via the tfb_tcp_timer_stop()
1840 		 * which should use the async drain function of the
1841 		 * callout system (see tcp_var.h).
1842 		 */
1843 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1844 	}
1845 
1846 	/*
1847 	 * If we got enough samples through the srtt filter,
1848 	 * save the rtt and rttvar in the routing entry.
1849 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1850 	 * 4 samples is enough for the srtt filter to converge
1851 	 * to within enough % of the correct value; fewer samples
1852 	 * and we could save a bogus rtt. The danger is not high
1853 	 * as tcp quickly recovers from everything.
1854 	 * XXX: Works very well but needs some more statistics!
1855 	 */
1856 	if (tp->t_rttupdated >= 4) {
1857 		struct hc_metrics_lite metrics;
1858 		uint32_t ssthresh;
1859 
1860 		bzero(&metrics, sizeof(metrics));
1861 		/*
1862 		 * Update the ssthresh always when the conditions below
1863 		 * are satisfied. This gives us better new start value
1864 		 * for the congestion avoidance for new connections.
1865 		 * ssthresh is only set if packet loss occurred on a session.
1866 		 *
1867 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1868 		 * being torn down.  Ideally this code would not use 'so'.
1869 		 */
1870 		ssthresh = tp->snd_ssthresh;
1871 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1872 			/*
1873 			 * convert the limit from user data bytes to
1874 			 * packets then to packet data bytes.
1875 			 */
1876 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1877 			if (ssthresh < 2)
1878 				ssthresh = 2;
1879 			ssthresh *= (tp->t_maxseg +
1880 #ifdef INET6
1881 			    (isipv6 ? sizeof (struct ip6_hdr) +
1882 				sizeof (struct tcphdr) :
1883 #endif
1884 				sizeof (struct tcpiphdr)
1885 #ifdef INET6
1886 			    )
1887 #endif
1888 			    );
1889 		} else
1890 			ssthresh = 0;
1891 		metrics.rmx_ssthresh = ssthresh;
1892 
1893 		metrics.rmx_rtt = tp->t_srtt;
1894 		metrics.rmx_rttvar = tp->t_rttvar;
1895 		metrics.rmx_cwnd = tp->snd_cwnd;
1896 		metrics.rmx_sendpipe = 0;
1897 		metrics.rmx_recvpipe = 0;
1898 
1899 		tcp_hc_update(&inp->inp_inc, &metrics);
1900 	}
1901 
1902 	/* free the reassembly queue, if any */
1903 	tcp_reass_flush(tp);
1904 
1905 #ifdef TCP_OFFLOAD
1906 	/* Disconnect offload device, if any. */
1907 	if (tp->t_flags & TF_TOE)
1908 		tcp_offload_detach(tp);
1909 #endif
1910 
1911 	tcp_free_sackholes(tp);
1912 
1913 #ifdef TCPPCAP
1914 	/* Free the TCP PCAP queues. */
1915 	tcp_pcap_drain(&(tp->t_inpkts));
1916 	tcp_pcap_drain(&(tp->t_outpkts));
1917 #endif
1918 
1919 	/* Allow the CC algorithm to clean up after itself. */
1920 	if (CC_ALGO(tp)->cb_destroy != NULL)
1921 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1922 	CC_DATA(tp) = NULL;
1923 
1924 #ifdef TCP_HHOOK
1925 	khelp_destroy_osd(tp->osd);
1926 #endif
1927 #ifdef STATS
1928 	stats_blob_destroy(tp->t_stats);
1929 #endif
1930 
1931 	CC_ALGO(tp) = NULL;
1932 	inp->inp_ppcb = NULL;
1933 	if (tp->t_timers->tt_draincnt == 0) {
1934 		/* We own the last reference on tcpcb, let's free it. */
1935 #ifdef TCP_BLACKBOX
1936 		tcp_log_tcpcbfini(tp);
1937 #endif
1938 		TCPSTATES_DEC(tp->t_state);
1939 		if (tp->t_fb->tfb_tcp_fb_fini)
1940 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1941 		refcount_release(&tp->t_fb->tfb_refcnt);
1942 		tp->t_inpcb = NULL;
1943 		uma_zfree(V_tcpcb_zone, tp);
1944 		released = in_pcbrele_wlocked(inp);
1945 		KASSERT(!released, ("%s: inp %p should not have been released "
1946 			"here", __func__, inp));
1947 	}
1948 }
1949 
1950 void
1951 tcp_timer_discard(void *ptp)
1952 {
1953 	struct inpcb *inp;
1954 	struct tcpcb *tp;
1955 	struct epoch_tracker et;
1956 
1957 	tp = (struct tcpcb *)ptp;
1958 	CURVNET_SET(tp->t_vnet);
1959 	NET_EPOCH_ENTER(et);
1960 	inp = tp->t_inpcb;
1961 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1962 		__func__, tp));
1963 	INP_WLOCK(inp);
1964 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1965 		("%s: tcpcb has to be stopped here", __func__));
1966 	tp->t_timers->tt_draincnt--;
1967 	if (tp->t_timers->tt_draincnt == 0) {
1968 		/* We own the last reference on this tcpcb, let's free it. */
1969 #ifdef TCP_BLACKBOX
1970 		tcp_log_tcpcbfini(tp);
1971 #endif
1972 		TCPSTATES_DEC(tp->t_state);
1973 		if (tp->t_fb->tfb_tcp_fb_fini)
1974 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1975 		refcount_release(&tp->t_fb->tfb_refcnt);
1976 		tp->t_inpcb = NULL;
1977 		uma_zfree(V_tcpcb_zone, tp);
1978 		if (in_pcbrele_wlocked(inp)) {
1979 			NET_EPOCH_EXIT(et);
1980 			CURVNET_RESTORE();
1981 			return;
1982 		}
1983 	}
1984 	INP_WUNLOCK(inp);
1985 	NET_EPOCH_EXIT(et);
1986 	CURVNET_RESTORE();
1987 }
1988 
1989 /*
1990  * Attempt to close a TCP control block, marking it as dropped, and freeing
1991  * the socket if we hold the only reference.
1992  */
1993 struct tcpcb *
1994 tcp_close(struct tcpcb *tp)
1995 {
1996 	struct inpcb *inp = tp->t_inpcb;
1997 	struct socket *so;
1998 
1999 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2000 	INP_WLOCK_ASSERT(inp);
2001 
2002 #ifdef TCP_OFFLOAD
2003 	if (tp->t_state == TCPS_LISTEN)
2004 		tcp_offload_listen_stop(tp);
2005 #endif
2006 	/*
2007 	 * This releases the TFO pending counter resource for TFO listen
2008 	 * sockets as well as passively-created TFO sockets that transition
2009 	 * from SYN_RECEIVED to CLOSED.
2010 	 */
2011 	if (tp->t_tfo_pending) {
2012 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2013 		tp->t_tfo_pending = NULL;
2014 	}
2015 	in_pcbdrop(inp);
2016 	TCPSTAT_INC(tcps_closed);
2017 	if (tp->t_state != TCPS_CLOSED)
2018 		tcp_state_change(tp, TCPS_CLOSED);
2019 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2020 	so = inp->inp_socket;
2021 	soisdisconnected(so);
2022 	if (inp->inp_flags & INP_SOCKREF) {
2023 		KASSERT(so->so_state & SS_PROTOREF,
2024 		    ("tcp_close: !SS_PROTOREF"));
2025 		inp->inp_flags &= ~INP_SOCKREF;
2026 		INP_WUNLOCK(inp);
2027 		SOCK_LOCK(so);
2028 		so->so_state &= ~SS_PROTOREF;
2029 		sofree(so);
2030 		return (NULL);
2031 	}
2032 	return (tp);
2033 }
2034 
2035 void
2036 tcp_drain(void)
2037 {
2038 	VNET_ITERATOR_DECL(vnet_iter);
2039 
2040 	if (!do_tcpdrain)
2041 		return;
2042 
2043 	VNET_LIST_RLOCK_NOSLEEP();
2044 	VNET_FOREACH(vnet_iter) {
2045 		CURVNET_SET(vnet_iter);
2046 		struct inpcb *inpb;
2047 		struct tcpcb *tcpb;
2048 
2049 	/*
2050 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2051 	 * if there is one...
2052 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2053 	 *      reassembly queue should be flushed, but in a situation
2054 	 *	where we're really low on mbufs, this is potentially
2055 	 *	useful.
2056 	 */
2057 		INP_INFO_WLOCK(&V_tcbinfo);
2058 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2059 			INP_WLOCK(inpb);
2060 			if (inpb->inp_flags & INP_TIMEWAIT) {
2061 				INP_WUNLOCK(inpb);
2062 				continue;
2063 			}
2064 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2065 				tcp_reass_flush(tcpb);
2066 				tcp_clean_sackreport(tcpb);
2067 #ifdef TCP_BLACKBOX
2068 				tcp_log_drain(tcpb);
2069 #endif
2070 #ifdef TCPPCAP
2071 				if (tcp_pcap_aggressive_free) {
2072 					/* Free the TCP PCAP queues. */
2073 					tcp_pcap_drain(&(tcpb->t_inpkts));
2074 					tcp_pcap_drain(&(tcpb->t_outpkts));
2075 				}
2076 #endif
2077 			}
2078 			INP_WUNLOCK(inpb);
2079 		}
2080 		INP_INFO_WUNLOCK(&V_tcbinfo);
2081 		CURVNET_RESTORE();
2082 	}
2083 	VNET_LIST_RUNLOCK_NOSLEEP();
2084 }
2085 
2086 /*
2087  * Notify a tcp user of an asynchronous error;
2088  * store error as soft error, but wake up user
2089  * (for now, won't do anything until can select for soft error).
2090  *
2091  * Do not wake up user since there currently is no mechanism for
2092  * reporting soft errors (yet - a kqueue filter may be added).
2093  */
2094 static struct inpcb *
2095 tcp_notify(struct inpcb *inp, int error)
2096 {
2097 	struct tcpcb *tp;
2098 
2099 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2100 	INP_WLOCK_ASSERT(inp);
2101 
2102 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2103 	    (inp->inp_flags & INP_DROPPED))
2104 		return (inp);
2105 
2106 	tp = intotcpcb(inp);
2107 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2108 
2109 	/*
2110 	 * Ignore some errors if we are hooked up.
2111 	 * If connection hasn't completed, has retransmitted several times,
2112 	 * and receives a second error, give up now.  This is better
2113 	 * than waiting a long time to establish a connection that
2114 	 * can never complete.
2115 	 */
2116 	if (tp->t_state == TCPS_ESTABLISHED &&
2117 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2118 	     error == EHOSTDOWN)) {
2119 		if (inp->inp_route.ro_rt) {
2120 			RTFREE(inp->inp_route.ro_rt);
2121 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2122 		}
2123 		return (inp);
2124 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2125 	    tp->t_softerror) {
2126 		tp = tcp_drop(tp, error);
2127 		if (tp != NULL)
2128 			return (inp);
2129 		else
2130 			return (NULL);
2131 	} else {
2132 		tp->t_softerror = error;
2133 		return (inp);
2134 	}
2135 #if 0
2136 	wakeup( &so->so_timeo);
2137 	sorwakeup(so);
2138 	sowwakeup(so);
2139 #endif
2140 }
2141 
2142 static int
2143 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2144 {
2145 	struct epoch_tracker et;
2146 	struct inpcb *inp;
2147 	struct xinpgen xig;
2148 	int error;
2149 
2150 	if (req->newptr != NULL)
2151 		return (EPERM);
2152 
2153 	if (req->oldptr == NULL) {
2154 		int n;
2155 
2156 		n = V_tcbinfo.ipi_count +
2157 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2158 		n += imax(n / 8, 10);
2159 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2160 		return (0);
2161 	}
2162 
2163 	if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2164 		return (error);
2165 
2166 	bzero(&xig, sizeof(xig));
2167 	xig.xig_len = sizeof xig;
2168 	xig.xig_count = V_tcbinfo.ipi_count +
2169 	    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2170 	xig.xig_gen = V_tcbinfo.ipi_gencnt;
2171 	xig.xig_sogen = so_gencnt;
2172 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2173 	if (error)
2174 		return (error);
2175 
2176 	error = syncache_pcblist(req);
2177 	if (error)
2178 		return (error);
2179 
2180 	NET_EPOCH_ENTER(et);
2181 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2182 	    inp != NULL;
2183 	    inp = CK_LIST_NEXT(inp, inp_list)) {
2184 		INP_RLOCK(inp);
2185 		if (inp->inp_gencnt <= xig.xig_gen) {
2186 			/*
2187 			 * XXX: This use of cr_cansee(), introduced with
2188 			 * TCP state changes, is not quite right, but for
2189 			 * now, better than nothing.
2190 			 */
2191 			if (inp->inp_flags & INP_TIMEWAIT) {
2192 				if (intotw(inp) != NULL)
2193 					error = cr_cansee(req->td->td_ucred,
2194 					    intotw(inp)->tw_cred);
2195 				else
2196 					error = EINVAL;	/* Skip this inp. */
2197 			} else
2198 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2199 			if (error == 0) {
2200 				struct xtcpcb xt;
2201 
2202 				tcp_inptoxtp(inp, &xt);
2203 				INP_RUNLOCK(inp);
2204 				error = SYSCTL_OUT(req, &xt, sizeof xt);
2205 				if (error)
2206 					break;
2207 				else
2208 					continue;
2209 			}
2210 		}
2211 		INP_RUNLOCK(inp);
2212 	}
2213 	NET_EPOCH_EXIT(et);
2214 
2215 	if (!error) {
2216 		/*
2217 		 * Give the user an updated idea of our state.
2218 		 * If the generation differs from what we told
2219 		 * her before, she knows that something happened
2220 		 * while we were processing this request, and it
2221 		 * might be necessary to retry.
2222 		 */
2223 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2224 		xig.xig_sogen = so_gencnt;
2225 		xig.xig_count = V_tcbinfo.ipi_count +
2226 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2227 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2228 	}
2229 
2230 	return (error);
2231 }
2232 
2233 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2234     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2235     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2236 
2237 #ifdef INET
2238 static int
2239 tcp_getcred(SYSCTL_HANDLER_ARGS)
2240 {
2241 	struct xucred xuc;
2242 	struct sockaddr_in addrs[2];
2243 	struct epoch_tracker et;
2244 	struct inpcb *inp;
2245 	int error;
2246 
2247 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2248 	if (error)
2249 		return (error);
2250 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2251 	if (error)
2252 		return (error);
2253 	NET_EPOCH_ENTER(et);
2254 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2255 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2256 	NET_EPOCH_EXIT(et);
2257 	if (inp != NULL) {
2258 		if (inp->inp_socket == NULL)
2259 			error = ENOENT;
2260 		if (error == 0)
2261 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2262 		if (error == 0)
2263 			cru2x(inp->inp_cred, &xuc);
2264 		INP_RUNLOCK(inp);
2265 	} else
2266 		error = ENOENT;
2267 	if (error == 0)
2268 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2269 	return (error);
2270 }
2271 
2272 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2273     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2274     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2275 #endif /* INET */
2276 
2277 #ifdef INET6
2278 static int
2279 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2280 {
2281 	struct epoch_tracker et;
2282 	struct xucred xuc;
2283 	struct sockaddr_in6 addrs[2];
2284 	struct inpcb *inp;
2285 	int error;
2286 #ifdef INET
2287 	int mapped = 0;
2288 #endif
2289 
2290 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2291 	if (error)
2292 		return (error);
2293 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2294 	if (error)
2295 		return (error);
2296 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2297 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2298 		return (error);
2299 	}
2300 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2301 #ifdef INET
2302 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2303 			mapped = 1;
2304 		else
2305 #endif
2306 			return (EINVAL);
2307 	}
2308 
2309 	NET_EPOCH_ENTER(et);
2310 #ifdef INET
2311 	if (mapped == 1)
2312 		inp = in_pcblookup(&V_tcbinfo,
2313 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2314 			addrs[1].sin6_port,
2315 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2316 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2317 	else
2318 #endif
2319 		inp = in6_pcblookup(&V_tcbinfo,
2320 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2321 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2322 			INPLOOKUP_RLOCKPCB, NULL);
2323 	NET_EPOCH_EXIT(et);
2324 	if (inp != NULL) {
2325 		if (inp->inp_socket == NULL)
2326 			error = ENOENT;
2327 		if (error == 0)
2328 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2329 		if (error == 0)
2330 			cru2x(inp->inp_cred, &xuc);
2331 		INP_RUNLOCK(inp);
2332 	} else
2333 		error = ENOENT;
2334 	if (error == 0)
2335 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2336 	return (error);
2337 }
2338 
2339 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2340     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2341     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2342 #endif /* INET6 */
2343 
2344 
2345 #ifdef INET
2346 void
2347 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2348 {
2349 	struct ip *ip = vip;
2350 	struct tcphdr *th;
2351 	struct in_addr faddr;
2352 	struct inpcb *inp;
2353 	struct tcpcb *tp;
2354 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2355 	struct icmp *icp;
2356 	struct in_conninfo inc;
2357 	tcp_seq icmp_tcp_seq;
2358 	int mtu;
2359 
2360 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2361 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2362 		return;
2363 
2364 	if (cmd == PRC_MSGSIZE)
2365 		notify = tcp_mtudisc_notify;
2366 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2367 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2368 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2369 		notify = tcp_drop_syn_sent;
2370 
2371 	/*
2372 	 * Hostdead is ugly because it goes linearly through all PCBs.
2373 	 * XXX: We never get this from ICMP, otherwise it makes an
2374 	 * excellent DoS attack on machines with many connections.
2375 	 */
2376 	else if (cmd == PRC_HOSTDEAD)
2377 		ip = NULL;
2378 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2379 		return;
2380 
2381 	if (ip == NULL) {
2382 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2383 		return;
2384 	}
2385 
2386 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2387 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2388 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2389 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2390 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2391 		/* signal EHOSTDOWN, as it flushes the cached route */
2392 		inp = (*notify)(inp, EHOSTDOWN);
2393 		goto out;
2394 	}
2395 	icmp_tcp_seq = th->th_seq;
2396 	if (inp != NULL)  {
2397 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2398 		    !(inp->inp_flags & INP_DROPPED) &&
2399 		    !(inp->inp_socket == NULL)) {
2400 			tp = intotcpcb(inp);
2401 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2402 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2403 				if (cmd == PRC_MSGSIZE) {
2404 					/*
2405 					 * MTU discovery:
2406 					 * If we got a needfrag set the MTU
2407 					 * in the route to the suggested new
2408 					 * value (if given) and then notify.
2409 					 */
2410 					mtu = ntohs(icp->icmp_nextmtu);
2411 					/*
2412 					 * If no alternative MTU was
2413 					 * proposed, try the next smaller
2414 					 * one.
2415 					 */
2416 					if (!mtu)
2417 						mtu = ip_next_mtu(
2418 						    ntohs(ip->ip_len), 1);
2419 					if (mtu < V_tcp_minmss +
2420 					    sizeof(struct tcpiphdr))
2421 						mtu = V_tcp_minmss +
2422 						    sizeof(struct tcpiphdr);
2423 					/*
2424 					 * Only process the offered MTU if it
2425 					 * is smaller than the current one.
2426 					 */
2427 					if (mtu < tp->t_maxseg +
2428 					    sizeof(struct tcpiphdr)) {
2429 						bzero(&inc, sizeof(inc));
2430 						inc.inc_faddr = faddr;
2431 						inc.inc_fibnum =
2432 						    inp->inp_inc.inc_fibnum;
2433 						tcp_hc_updatemtu(&inc, mtu);
2434 						tcp_mtudisc(inp, mtu);
2435 					}
2436 				} else
2437 					inp = (*notify)(inp,
2438 					    inetctlerrmap[cmd]);
2439 			}
2440 		}
2441 	} else {
2442 		bzero(&inc, sizeof(inc));
2443 		inc.inc_fport = th->th_dport;
2444 		inc.inc_lport = th->th_sport;
2445 		inc.inc_faddr = faddr;
2446 		inc.inc_laddr = ip->ip_src;
2447 		syncache_unreach(&inc, icmp_tcp_seq);
2448 	}
2449 out:
2450 	if (inp != NULL)
2451 		INP_WUNLOCK(inp);
2452 }
2453 #endif /* INET */
2454 
2455 #ifdef INET6
2456 void
2457 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2458 {
2459 	struct in6_addr *dst;
2460 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2461 	struct ip6_hdr *ip6;
2462 	struct mbuf *m;
2463 	struct inpcb *inp;
2464 	struct tcpcb *tp;
2465 	struct icmp6_hdr *icmp6;
2466 	struct ip6ctlparam *ip6cp = NULL;
2467 	const struct sockaddr_in6 *sa6_src = NULL;
2468 	struct in_conninfo inc;
2469 	struct tcp_ports {
2470 		uint16_t th_sport;
2471 		uint16_t th_dport;
2472 	} t_ports;
2473 	tcp_seq icmp_tcp_seq;
2474 	unsigned int mtu;
2475 	unsigned int off;
2476 
2477 	if (sa->sa_family != AF_INET6 ||
2478 	    sa->sa_len != sizeof(struct sockaddr_in6))
2479 		return;
2480 
2481 	/* if the parameter is from icmp6, decode it. */
2482 	if (d != NULL) {
2483 		ip6cp = (struct ip6ctlparam *)d;
2484 		icmp6 = ip6cp->ip6c_icmp6;
2485 		m = ip6cp->ip6c_m;
2486 		ip6 = ip6cp->ip6c_ip6;
2487 		off = ip6cp->ip6c_off;
2488 		sa6_src = ip6cp->ip6c_src;
2489 		dst = ip6cp->ip6c_finaldst;
2490 	} else {
2491 		m = NULL;
2492 		ip6 = NULL;
2493 		off = 0;	/* fool gcc */
2494 		sa6_src = &sa6_any;
2495 		dst = NULL;
2496 	}
2497 
2498 	if (cmd == PRC_MSGSIZE)
2499 		notify = tcp_mtudisc_notify;
2500 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2501 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2502 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2503 		notify = tcp_drop_syn_sent;
2504 
2505 	/*
2506 	 * Hostdead is ugly because it goes linearly through all PCBs.
2507 	 * XXX: We never get this from ICMP, otherwise it makes an
2508 	 * excellent DoS attack on machines with many connections.
2509 	 */
2510 	else if (cmd == PRC_HOSTDEAD)
2511 		ip6 = NULL;
2512 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2513 		return;
2514 
2515 	if (ip6 == NULL) {
2516 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2517 			      (const struct sockaddr *)sa6_src,
2518 			      0, cmd, NULL, notify);
2519 		return;
2520 	}
2521 
2522 	/* Check if we can safely get the ports from the tcp hdr */
2523 	if (m == NULL ||
2524 	    (m->m_pkthdr.len <
2525 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2526 		return;
2527 	}
2528 	bzero(&t_ports, sizeof(struct tcp_ports));
2529 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2530 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2531 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2532 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2533 		/* signal EHOSTDOWN, as it flushes the cached route */
2534 		inp = (*notify)(inp, EHOSTDOWN);
2535 		goto out;
2536 	}
2537 	off += sizeof(struct tcp_ports);
2538 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2539 		goto out;
2540 	}
2541 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2542 	if (inp != NULL)  {
2543 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2544 		    !(inp->inp_flags & INP_DROPPED) &&
2545 		    !(inp->inp_socket == NULL)) {
2546 			tp = intotcpcb(inp);
2547 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2548 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2549 				if (cmd == PRC_MSGSIZE) {
2550 					/*
2551 					 * MTU discovery:
2552 					 * If we got a needfrag set the MTU
2553 					 * in the route to the suggested new
2554 					 * value (if given) and then notify.
2555 					 */
2556 					mtu = ntohl(icmp6->icmp6_mtu);
2557 					/*
2558 					 * If no alternative MTU was
2559 					 * proposed, or the proposed
2560 					 * MTU was too small, set to
2561 					 * the min.
2562 					 */
2563 					if (mtu < IPV6_MMTU)
2564 						mtu = IPV6_MMTU - 8;
2565 					bzero(&inc, sizeof(inc));
2566 					inc.inc_fibnum = M_GETFIB(m);
2567 					inc.inc_flags |= INC_ISIPV6;
2568 					inc.inc6_faddr = *dst;
2569 					if (in6_setscope(&inc.inc6_faddr,
2570 						m->m_pkthdr.rcvif, NULL))
2571 						goto out;
2572 					/*
2573 					 * Only process the offered MTU if it
2574 					 * is smaller than the current one.
2575 					 */
2576 					if (mtu < tp->t_maxseg +
2577 					    sizeof (struct tcphdr) +
2578 					    sizeof (struct ip6_hdr)) {
2579 						tcp_hc_updatemtu(&inc, mtu);
2580 						tcp_mtudisc(inp, mtu);
2581 						ICMP6STAT_INC(icp6s_pmtuchg);
2582 					}
2583 				} else
2584 					inp = (*notify)(inp,
2585 					    inet6ctlerrmap[cmd]);
2586 			}
2587 		}
2588 	} else {
2589 		bzero(&inc, sizeof(inc));
2590 		inc.inc_fibnum = M_GETFIB(m);
2591 		inc.inc_flags |= INC_ISIPV6;
2592 		inc.inc_fport = t_ports.th_dport;
2593 		inc.inc_lport = t_ports.th_sport;
2594 		inc.inc6_faddr = *dst;
2595 		inc.inc6_laddr = ip6->ip6_src;
2596 		syncache_unreach(&inc, icmp_tcp_seq);
2597 	}
2598 out:
2599 	if (inp != NULL)
2600 		INP_WUNLOCK(inp);
2601 }
2602 #endif /* INET6 */
2603 
2604 static uint32_t
2605 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2606 {
2607 	SIPHASH_CTX ctx;
2608 	uint32_t hash[2];
2609 
2610 	KASSERT(len >= SIPHASH_KEY_LENGTH,
2611 	    ("%s: keylen %u too short ", __func__, len));
2612 	SipHash24_Init(&ctx);
2613 	SipHash_SetKey(&ctx, (uint8_t *)key);
2614 	SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2615 	SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2616 	switch (inc->inc_flags & INC_ISIPV6) {
2617 #ifdef INET
2618 	case 0:
2619 		SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2620 		SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2621 		break;
2622 #endif
2623 #ifdef INET6
2624 	case INC_ISIPV6:
2625 		SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2626 		SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2627 		break;
2628 #endif
2629 	}
2630 	SipHash_Final((uint8_t *)hash, &ctx);
2631 
2632 	return (hash[0] ^ hash[1]);
2633 }
2634 
2635 uint32_t
2636 tcp_new_ts_offset(struct in_conninfo *inc)
2637 {
2638 	struct in_conninfo inc_store, *local_inc;
2639 
2640 	if (!V_tcp_ts_offset_per_conn) {
2641 		memcpy(&inc_store, inc, sizeof(struct in_conninfo));
2642 		inc_store.inc_lport = 0;
2643 		inc_store.inc_fport = 0;
2644 		local_inc = &inc_store;
2645 	} else {
2646 		local_inc = inc;
2647 	}
2648 	return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
2649 	    sizeof(V_ts_offset_secret)));
2650 }
2651 
2652 /*
2653  * Following is where TCP initial sequence number generation occurs.
2654  *
2655  * There are two places where we must use initial sequence numbers:
2656  * 1.  In SYN-ACK packets.
2657  * 2.  In SYN packets.
2658  *
2659  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2660  * tcp_syncache.c for details.
2661  *
2662  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2663  * depends on this property.  In addition, these ISNs should be
2664  * unguessable so as to prevent connection hijacking.  To satisfy
2665  * the requirements of this situation, the algorithm outlined in
2666  * RFC 1948 is used, with only small modifications.
2667  *
2668  * Implementation details:
2669  *
2670  * Time is based off the system timer, and is corrected so that it
2671  * increases by one megabyte per second.  This allows for proper
2672  * recycling on high speed LANs while still leaving over an hour
2673  * before rollover.
2674  *
2675  * As reading the *exact* system time is too expensive to be done
2676  * whenever setting up a TCP connection, we increment the time
2677  * offset in two ways.  First, a small random positive increment
2678  * is added to isn_offset for each connection that is set up.
2679  * Second, the function tcp_isn_tick fires once per clock tick
2680  * and increments isn_offset as necessary so that sequence numbers
2681  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2682  * random positive increments serve only to ensure that the same
2683  * exact sequence number is never sent out twice (as could otherwise
2684  * happen when a port is recycled in less than the system tick
2685  * interval.)
2686  *
2687  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2688  * between seeding of isn_secret.  This is normally set to zero,
2689  * as reseeding should not be necessary.
2690  *
2691  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2692  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2693  * general, this means holding an exclusive (write) lock.
2694  */
2695 
2696 #define ISN_BYTES_PER_SECOND 1048576
2697 #define ISN_STATIC_INCREMENT 4096
2698 #define ISN_RANDOM_INCREMENT (4096 - 1)
2699 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
2700 
2701 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2702 VNET_DEFINE_STATIC(int, isn_last);
2703 VNET_DEFINE_STATIC(int, isn_last_reseed);
2704 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2705 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2706 
2707 #define	V_isn_secret			VNET(isn_secret)
2708 #define	V_isn_last			VNET(isn_last)
2709 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2710 #define	V_isn_offset			VNET(isn_offset)
2711 #define	V_isn_offset_old		VNET(isn_offset_old)
2712 
2713 tcp_seq
2714 tcp_new_isn(struct in_conninfo *inc)
2715 {
2716 	tcp_seq new_isn;
2717 	u_int32_t projected_offset;
2718 
2719 	ISN_LOCK();
2720 	/* Seed if this is the first use, reseed if requested. */
2721 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2722 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2723 		< (u_int)ticks))) {
2724 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2725 		V_isn_last_reseed = ticks;
2726 	}
2727 
2728 	/* Compute the hash and return the ISN. */
2729 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2730 	    sizeof(V_isn_secret));
2731 	V_isn_offset += ISN_STATIC_INCREMENT +
2732 		(arc4random() & ISN_RANDOM_INCREMENT);
2733 	if (ticks != V_isn_last) {
2734 		projected_offset = V_isn_offset_old +
2735 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2736 		if (SEQ_GT(projected_offset, V_isn_offset))
2737 			V_isn_offset = projected_offset;
2738 		V_isn_offset_old = V_isn_offset;
2739 		V_isn_last = ticks;
2740 	}
2741 	new_isn += V_isn_offset;
2742 	ISN_UNLOCK();
2743 	return (new_isn);
2744 }
2745 
2746 /*
2747  * When a specific ICMP unreachable message is received and the
2748  * connection state is SYN-SENT, drop the connection.  This behavior
2749  * is controlled by the icmp_may_rst sysctl.
2750  */
2751 struct inpcb *
2752 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2753 {
2754 	struct tcpcb *tp;
2755 
2756 	NET_EPOCH_ASSERT();
2757 	INP_WLOCK_ASSERT(inp);
2758 
2759 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2760 	    (inp->inp_flags & INP_DROPPED))
2761 		return (inp);
2762 
2763 	tp = intotcpcb(inp);
2764 	if (tp->t_state != TCPS_SYN_SENT)
2765 		return (inp);
2766 
2767 	if (IS_FASTOPEN(tp->t_flags))
2768 		tcp_fastopen_disable_path(tp);
2769 
2770 	tp = tcp_drop(tp, errno);
2771 	if (tp != NULL)
2772 		return (inp);
2773 	else
2774 		return (NULL);
2775 }
2776 
2777 /*
2778  * When `need fragmentation' ICMP is received, update our idea of the MSS
2779  * based on the new value. Also nudge TCP to send something, since we
2780  * know the packet we just sent was dropped.
2781  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2782  */
2783 static struct inpcb *
2784 tcp_mtudisc_notify(struct inpcb *inp, int error)
2785 {
2786 
2787 	tcp_mtudisc(inp, -1);
2788 	return (inp);
2789 }
2790 
2791 static void
2792 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2793 {
2794 	struct tcpcb *tp;
2795 	struct socket *so;
2796 
2797 	INP_WLOCK_ASSERT(inp);
2798 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2799 	    (inp->inp_flags & INP_DROPPED))
2800 		return;
2801 
2802 	tp = intotcpcb(inp);
2803 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2804 
2805 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2806 
2807 	so = inp->inp_socket;
2808 	SOCKBUF_LOCK(&so->so_snd);
2809 	/* If the mss is larger than the socket buffer, decrease the mss. */
2810 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2811 		tp->t_maxseg = so->so_snd.sb_hiwat;
2812 	SOCKBUF_UNLOCK(&so->so_snd);
2813 
2814 	TCPSTAT_INC(tcps_mturesent);
2815 	tp->t_rtttime = 0;
2816 	tp->snd_nxt = tp->snd_una;
2817 	tcp_free_sackholes(tp);
2818 	tp->snd_recover = tp->snd_max;
2819 	if (tp->t_flags & TF_SACK_PERMIT)
2820 		EXIT_FASTRECOVERY(tp->t_flags);
2821 	tp->t_fb->tfb_tcp_output(tp);
2822 }
2823 
2824 #ifdef INET
2825 /*
2826  * Look-up the routing entry to the peer of this inpcb.  If no route
2827  * is found and it cannot be allocated, then return 0.  This routine
2828  * is called by TCP routines that access the rmx structure and by
2829  * tcp_mss_update to get the peer/interface MTU.
2830  */
2831 uint32_t
2832 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2833 {
2834 	struct nhop4_extended nh4;
2835 	struct ifnet *ifp;
2836 	uint32_t maxmtu = 0;
2837 
2838 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2839 
2840 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2841 
2842 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2843 		    NHR_REF, 0, &nh4) != 0)
2844 			return (0);
2845 
2846 		ifp = nh4.nh_ifp;
2847 		maxmtu = nh4.nh_mtu;
2848 
2849 		/* Report additional interface capabilities. */
2850 		if (cap != NULL) {
2851 			if (ifp->if_capenable & IFCAP_TSO4 &&
2852 			    ifp->if_hwassist & CSUM_TSO) {
2853 				cap->ifcap |= CSUM_TSO;
2854 				cap->tsomax = ifp->if_hw_tsomax;
2855 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2856 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2857 			}
2858 		}
2859 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2860 	}
2861 	return (maxmtu);
2862 }
2863 #endif /* INET */
2864 
2865 #ifdef INET6
2866 uint32_t
2867 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2868 {
2869 	struct nhop6_extended nh6;
2870 	struct in6_addr dst6;
2871 	uint32_t scopeid;
2872 	struct ifnet *ifp;
2873 	uint32_t maxmtu = 0;
2874 
2875 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2876 
2877 	if (inc->inc_flags & INC_IPV6MINMTU)
2878 		return (IPV6_MMTU);
2879 
2880 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2881 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2882 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2883 		    0, &nh6) != 0)
2884 			return (0);
2885 
2886 		ifp = nh6.nh_ifp;
2887 		maxmtu = nh6.nh_mtu;
2888 
2889 		/* Report additional interface capabilities. */
2890 		if (cap != NULL) {
2891 			if (ifp->if_capenable & IFCAP_TSO6 &&
2892 			    ifp->if_hwassist & CSUM_TSO) {
2893 				cap->ifcap |= CSUM_TSO;
2894 				cap->tsomax = ifp->if_hw_tsomax;
2895 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2896 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2897 			}
2898 		}
2899 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2900 	}
2901 
2902 	return (maxmtu);
2903 }
2904 #endif /* INET6 */
2905 
2906 /*
2907  * Calculate effective SMSS per RFC5681 definition for a given TCP
2908  * connection at its current state, taking into account SACK and etc.
2909  */
2910 u_int
2911 tcp_maxseg(const struct tcpcb *tp)
2912 {
2913 	u_int optlen;
2914 
2915 	if (tp->t_flags & TF_NOOPT)
2916 		return (tp->t_maxseg);
2917 
2918 	/*
2919 	 * Here we have a simplified code from tcp_addoptions(),
2920 	 * without a proper loop, and having most of paddings hardcoded.
2921 	 * We might make mistakes with padding here in some edge cases,
2922 	 * but this is harmless, since result of tcp_maxseg() is used
2923 	 * only in cwnd and ssthresh estimations.
2924 	 */
2925 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2926 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2927 		if (tp->t_flags & TF_RCVD_TSTMP)
2928 			optlen = TCPOLEN_TSTAMP_APPA;
2929 		else
2930 			optlen = 0;
2931 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2932 		if (tp->t_flags & TF_SIGNATURE)
2933 			optlen += PAD(TCPOLEN_SIGNATURE);
2934 #endif
2935 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2936 			optlen += TCPOLEN_SACKHDR;
2937 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2938 			optlen = PAD(optlen);
2939 		}
2940 	} else {
2941 		if (tp->t_flags & TF_REQ_TSTMP)
2942 			optlen = TCPOLEN_TSTAMP_APPA;
2943 		else
2944 			optlen = PAD(TCPOLEN_MAXSEG);
2945 		if (tp->t_flags & TF_REQ_SCALE)
2946 			optlen += PAD(TCPOLEN_WINDOW);
2947 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2948 		if (tp->t_flags & TF_SIGNATURE)
2949 			optlen += PAD(TCPOLEN_SIGNATURE);
2950 #endif
2951 		if (tp->t_flags & TF_SACK_PERMIT)
2952 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2953 	}
2954 #undef PAD
2955 	optlen = min(optlen, TCP_MAXOLEN);
2956 	return (tp->t_maxseg - optlen);
2957 }
2958 
2959 static int
2960 sysctl_drop(SYSCTL_HANDLER_ARGS)
2961 {
2962 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2963 	struct sockaddr_storage addrs[2];
2964 	struct inpcb *inp;
2965 	struct tcpcb *tp;
2966 	struct tcptw *tw;
2967 	struct sockaddr_in *fin, *lin;
2968 	struct epoch_tracker et;
2969 #ifdef INET6
2970 	struct sockaddr_in6 *fin6, *lin6;
2971 #endif
2972 	int error;
2973 
2974 	inp = NULL;
2975 	fin = lin = NULL;
2976 #ifdef INET6
2977 	fin6 = lin6 = NULL;
2978 #endif
2979 	error = 0;
2980 
2981 	if (req->oldptr != NULL || req->oldlen != 0)
2982 		return (EINVAL);
2983 	if (req->newptr == NULL)
2984 		return (EPERM);
2985 	if (req->newlen < sizeof(addrs))
2986 		return (ENOMEM);
2987 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2988 	if (error)
2989 		return (error);
2990 
2991 	switch (addrs[0].ss_family) {
2992 #ifdef INET6
2993 	case AF_INET6:
2994 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2995 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2996 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2997 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2998 			return (EINVAL);
2999 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3000 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3001 				return (EINVAL);
3002 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3003 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3004 			fin = (struct sockaddr_in *)&addrs[0];
3005 			lin = (struct sockaddr_in *)&addrs[1];
3006 			break;
3007 		}
3008 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3009 		if (error)
3010 			return (error);
3011 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3012 		if (error)
3013 			return (error);
3014 		break;
3015 #endif
3016 #ifdef INET
3017 	case AF_INET:
3018 		fin = (struct sockaddr_in *)&addrs[0];
3019 		lin = (struct sockaddr_in *)&addrs[1];
3020 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3021 		    lin->sin_len != sizeof(struct sockaddr_in))
3022 			return (EINVAL);
3023 		break;
3024 #endif
3025 	default:
3026 		return (EINVAL);
3027 	}
3028 	NET_EPOCH_ENTER(et);
3029 	switch (addrs[0].ss_family) {
3030 #ifdef INET6
3031 	case AF_INET6:
3032 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3033 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3034 		    INPLOOKUP_WLOCKPCB, NULL);
3035 		break;
3036 #endif
3037 #ifdef INET
3038 	case AF_INET:
3039 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3040 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3041 		break;
3042 #endif
3043 	}
3044 	if (inp != NULL) {
3045 		if (inp->inp_flags & INP_TIMEWAIT) {
3046 			/*
3047 			 * XXXRW: There currently exists a state where an
3048 			 * inpcb is present, but its timewait state has been
3049 			 * discarded.  For now, don't allow dropping of this
3050 			 * type of inpcb.
3051 			 */
3052 			tw = intotw(inp);
3053 			if (tw != NULL)
3054 				tcp_twclose(tw, 0);
3055 			else
3056 				INP_WUNLOCK(inp);
3057 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3058 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3059 			tp = intotcpcb(inp);
3060 			tp = tcp_drop(tp, ECONNABORTED);
3061 			if (tp != NULL)
3062 				INP_WUNLOCK(inp);
3063 		} else
3064 			INP_WUNLOCK(inp);
3065 	} else
3066 		error = ESRCH;
3067 	NET_EPOCH_EXIT(et);
3068 	return (error);
3069 }
3070 
3071 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3072     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3073     0, sysctl_drop, "", "Drop TCP connection");
3074 
3075 #ifdef KERN_TLS
3076 static int
3077 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3078 {
3079 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
3080 	struct sockaddr_storage addrs[2];
3081 	struct inpcb *inp;
3082 	struct sockaddr_in *fin, *lin;
3083 	struct epoch_tracker et;
3084 #ifdef INET6
3085 	struct sockaddr_in6 *fin6, *lin6;
3086 #endif
3087 	int error;
3088 
3089 	inp = NULL;
3090 	fin = lin = NULL;
3091 #ifdef INET6
3092 	fin6 = lin6 = NULL;
3093 #endif
3094 	error = 0;
3095 
3096 	if (req->oldptr != NULL || req->oldlen != 0)
3097 		return (EINVAL);
3098 	if (req->newptr == NULL)
3099 		return (EPERM);
3100 	if (req->newlen < sizeof(addrs))
3101 		return (ENOMEM);
3102 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3103 	if (error)
3104 		return (error);
3105 
3106 	switch (addrs[0].ss_family) {
3107 #ifdef INET6
3108 	case AF_INET6:
3109 		fin6 = (struct sockaddr_in6 *)&addrs[0];
3110 		lin6 = (struct sockaddr_in6 *)&addrs[1];
3111 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3112 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
3113 			return (EINVAL);
3114 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3115 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3116 				return (EINVAL);
3117 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3118 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3119 			fin = (struct sockaddr_in *)&addrs[0];
3120 			lin = (struct sockaddr_in *)&addrs[1];
3121 			break;
3122 		}
3123 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3124 		if (error)
3125 			return (error);
3126 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3127 		if (error)
3128 			return (error);
3129 		break;
3130 #endif
3131 #ifdef INET
3132 	case AF_INET:
3133 		fin = (struct sockaddr_in *)&addrs[0];
3134 		lin = (struct sockaddr_in *)&addrs[1];
3135 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3136 		    lin->sin_len != sizeof(struct sockaddr_in))
3137 			return (EINVAL);
3138 		break;
3139 #endif
3140 	default:
3141 		return (EINVAL);
3142 	}
3143 	NET_EPOCH_ENTER(et);
3144 	switch (addrs[0].ss_family) {
3145 #ifdef INET6
3146 	case AF_INET6:
3147 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3148 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3149 		    INPLOOKUP_WLOCKPCB, NULL);
3150 		break;
3151 #endif
3152 #ifdef INET
3153 	case AF_INET:
3154 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3155 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3156 		break;
3157 #endif
3158 	}
3159 	NET_EPOCH_EXIT(et);
3160 	if (inp != NULL) {
3161 		if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3162 		    inp->inp_socket == NULL) {
3163 			error = ECONNRESET;
3164 			INP_WUNLOCK(inp);
3165 		} else {
3166 			struct socket *so;
3167 
3168 			so = inp->inp_socket;
3169 			soref(so);
3170 			error = ktls_set_tx_mode(so,
3171 			    arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3172 			INP_WUNLOCK(inp);
3173 			SOCK_LOCK(so);
3174 			sorele(so);
3175 		}
3176 	} else
3177 		error = ESRCH;
3178 	return (error);
3179 }
3180 
3181 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3182     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3183     0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS");
3184 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3185     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3186     1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS");
3187 #endif
3188 
3189 /*
3190  * Generate a standardized TCP log line for use throughout the
3191  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3192  * allow use in the interrupt context.
3193  *
3194  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3195  * NB: The function may return NULL if memory allocation failed.
3196  *
3197  * Due to header inclusion and ordering limitations the struct ip
3198  * and ip6_hdr pointers have to be passed as void pointers.
3199  */
3200 char *
3201 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3202     const void *ip6hdr)
3203 {
3204 
3205 	/* Is logging enabled? */
3206 	if (tcp_log_in_vain == 0)
3207 		return (NULL);
3208 
3209 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3210 }
3211 
3212 char *
3213 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3214     const void *ip6hdr)
3215 {
3216 
3217 	/* Is logging enabled? */
3218 	if (tcp_log_debug == 0)
3219 		return (NULL);
3220 
3221 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3222 }
3223 
3224 static char *
3225 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3226     const void *ip6hdr)
3227 {
3228 	char *s, *sp;
3229 	size_t size;
3230 	struct ip *ip;
3231 #ifdef INET6
3232 	const struct ip6_hdr *ip6;
3233 
3234 	ip6 = (const struct ip6_hdr *)ip6hdr;
3235 #endif /* INET6 */
3236 	ip = (struct ip *)ip4hdr;
3237 
3238 	/*
3239 	 * The log line looks like this:
3240 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3241 	 */
3242 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3243 	    sizeof(PRINT_TH_FLAGS) + 1 +
3244 #ifdef INET6
3245 	    2 * INET6_ADDRSTRLEN;
3246 #else
3247 	    2 * INET_ADDRSTRLEN;
3248 #endif /* INET6 */
3249 
3250 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3251 	if (s == NULL)
3252 		return (NULL);
3253 
3254 	strcat(s, "TCP: [");
3255 	sp = s + strlen(s);
3256 
3257 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3258 		inet_ntoa_r(inc->inc_faddr, sp);
3259 		sp = s + strlen(s);
3260 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3261 		sp = s + strlen(s);
3262 		inet_ntoa_r(inc->inc_laddr, sp);
3263 		sp = s + strlen(s);
3264 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3265 #ifdef INET6
3266 	} else if (inc) {
3267 		ip6_sprintf(sp, &inc->inc6_faddr);
3268 		sp = s + strlen(s);
3269 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3270 		sp = s + strlen(s);
3271 		ip6_sprintf(sp, &inc->inc6_laddr);
3272 		sp = s + strlen(s);
3273 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3274 	} else if (ip6 && th) {
3275 		ip6_sprintf(sp, &ip6->ip6_src);
3276 		sp = s + strlen(s);
3277 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3278 		sp = s + strlen(s);
3279 		ip6_sprintf(sp, &ip6->ip6_dst);
3280 		sp = s + strlen(s);
3281 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3282 #endif /* INET6 */
3283 #ifdef INET
3284 	} else if (ip && th) {
3285 		inet_ntoa_r(ip->ip_src, sp);
3286 		sp = s + strlen(s);
3287 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3288 		sp = s + strlen(s);
3289 		inet_ntoa_r(ip->ip_dst, sp);
3290 		sp = s + strlen(s);
3291 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3292 #endif /* INET */
3293 	} else {
3294 		free(s, M_TCPLOG);
3295 		return (NULL);
3296 	}
3297 	sp = s + strlen(s);
3298 	if (th)
3299 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3300 	if (*(s + size - 1) != '\0')
3301 		panic("%s: string too long", __func__);
3302 	return (s);
3303 }
3304 
3305 /*
3306  * A subroutine which makes it easy to track TCP state changes with DTrace.
3307  * This function shouldn't be called for t_state initializations that don't
3308  * correspond to actual TCP state transitions.
3309  */
3310 void
3311 tcp_state_change(struct tcpcb *tp, int newstate)
3312 {
3313 #if defined(KDTRACE_HOOKS)
3314 	int pstate = tp->t_state;
3315 #endif
3316 
3317 	TCPSTATES_DEC(tp->t_state);
3318 	TCPSTATES_INC(newstate);
3319 	tp->t_state = newstate;
3320 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3321 }
3322 
3323 /*
3324  * Create an external-format (``xtcpcb'') structure using the information in
3325  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3326  * reduce the spew of irrelevant information over this interface, to isolate
3327  * user code from changes in the kernel structure, and potentially to provide
3328  * information-hiding if we decide that some of this information should be
3329  * hidden from users.
3330  */
3331 void
3332 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3333 {
3334 	struct tcpcb *tp = intotcpcb(inp);
3335 	sbintime_t now;
3336 
3337 	bzero(xt, sizeof(*xt));
3338 	if (inp->inp_flags & INP_TIMEWAIT) {
3339 		xt->t_state = TCPS_TIME_WAIT;
3340 	} else {
3341 		xt->t_state = tp->t_state;
3342 		xt->t_logstate = tp->t_logstate;
3343 		xt->t_flags = tp->t_flags;
3344 		xt->t_sndzerowin = tp->t_sndzerowin;
3345 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3346 		xt->t_rcvoopack = tp->t_rcvoopack;
3347 
3348 		now = getsbinuptime();
3349 #define	COPYTIMER(ttt)	do {						\
3350 		if (callout_active(&tp->t_timers->ttt))			\
3351 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3352 			    SBT_1MS;					\
3353 		else							\
3354 			xt->ttt = 0;					\
3355 } while (0)
3356 		COPYTIMER(tt_delack);
3357 		COPYTIMER(tt_rexmt);
3358 		COPYTIMER(tt_persist);
3359 		COPYTIMER(tt_keep);
3360 		COPYTIMER(tt_2msl);
3361 #undef COPYTIMER
3362 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3363 
3364 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3365 		    TCP_FUNCTION_NAME_LEN_MAX);
3366 #ifdef TCP_BLACKBOX
3367 		(void)tcp_log_get_id(tp, xt->xt_logid);
3368 #endif
3369 	}
3370 
3371 	xt->xt_len = sizeof(struct xtcpcb);
3372 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3373 	if (inp->inp_socket == NULL)
3374 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3375 }
3376