xref: /freebsd/sys/netinet/tcp_subr.c (revision d34048812292b714a0bf99967270d18fe3097c62)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/eventhandler.h>
46 #ifdef TCP_HHOOK
47 #include <sys/hhook.h>
48 #endif
49 #include <sys/kernel.h>
50 #ifdef TCP_HHOOK
51 #include <sys/khelp.h>
52 #endif
53 #include <sys/sysctl.h>
54 #include <sys/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/refcount.h>
57 #include <sys/mbuf.h>
58 #ifdef INET6
59 #include <sys/domain.h>
60 #endif
61 #include <sys/priv.h>
62 #include <sys/proc.h>
63 #include <sys/sdt.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/random.h>
68 
69 #include <vm/uma.h>
70 
71 #include <net/route.h>
72 #include <net/if.h>
73 #include <net/if_var.h>
74 #include <net/vnet.h>
75 
76 #include <netinet/in.h>
77 #include <netinet/in_fib.h>
78 #include <netinet/in_kdtrace.h>
79 #include <netinet/in_pcb.h>
80 #include <netinet/in_systm.h>
81 #include <netinet/in_var.h>
82 #include <netinet/ip.h>
83 #include <netinet/ip_icmp.h>
84 #include <netinet/ip_var.h>
85 #ifdef INET6
86 #include <netinet/icmp6.h>
87 #include <netinet/ip6.h>
88 #include <netinet6/in6_fib.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #include <netinet6/nd6.h>
93 #endif
94 
95 #include <netinet/tcp.h>
96 #include <netinet/tcp_fsm.h>
97 #include <netinet/tcp_seq.h>
98 #include <netinet/tcp_timer.h>
99 #include <netinet/tcp_var.h>
100 #include <netinet/tcp_log_buf.h>
101 #include <netinet/tcp_syncache.h>
102 #include <netinet/tcp_hpts.h>
103 #include <netinet/cc/cc.h>
104 #ifdef INET6
105 #include <netinet6/tcp6_var.h>
106 #endif
107 #include <netinet/tcpip.h>
108 #include <netinet/tcp_fastopen.h>
109 #ifdef TCPPCAP
110 #include <netinet/tcp_pcap.h>
111 #endif
112 #ifdef TCPDEBUG
113 #include <netinet/tcp_debug.h>
114 #endif
115 #ifdef INET6
116 #include <netinet6/ip6protosw.h>
117 #endif
118 #ifdef TCP_OFFLOAD
119 #include <netinet/tcp_offload.h>
120 #endif
121 
122 #include <netipsec/ipsec_support.h>
123 
124 #include <machine/in_cksum.h>
125 #include <sys/md5.h>
126 
127 #include <security/mac/mac_framework.h>
128 
129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
130 #ifdef INET6
131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
132 #endif
133 
134 struct rwlock tcp_function_lock;
135 
136 static int
137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
138 {
139 	int error, new;
140 
141 	new = V_tcp_mssdflt;
142 	error = sysctl_handle_int(oidp, &new, 0, req);
143 	if (error == 0 && req->newptr) {
144 		if (new < TCP_MINMSS)
145 			error = EINVAL;
146 		else
147 			V_tcp_mssdflt = new;
148 	}
149 	return (error);
150 }
151 
152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
153     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
154     &sysctl_net_inet_tcp_mss_check, "I",
155     "Default TCP Maximum Segment Size");
156 
157 #ifdef INET6
158 static int
159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
160 {
161 	int error, new;
162 
163 	new = V_tcp_v6mssdflt;
164 	error = sysctl_handle_int(oidp, &new, 0, req);
165 	if (error == 0 && req->newptr) {
166 		if (new < TCP_MINMSS)
167 			error = EINVAL;
168 		else
169 			V_tcp_v6mssdflt = new;
170 	}
171 	return (error);
172 }
173 
174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
175     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
176     &sysctl_net_inet_tcp_mss_v6_check, "I",
177    "Default TCP Maximum Segment Size for IPv6");
178 #endif /* INET6 */
179 
180 /*
181  * Minimum MSS we accept and use. This prevents DoS attacks where
182  * we are forced to a ridiculous low MSS like 20 and send hundreds
183  * of packets instead of one. The effect scales with the available
184  * bandwidth and quickly saturates the CPU and network interface
185  * with packet generation and sending. Set to zero to disable MINMSS
186  * checking. This setting prevents us from sending too small packets.
187  */
188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
190      &VNET_NAME(tcp_minmss), 0,
191     "Minimum TCP Maximum Segment Size");
192 
193 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
195     &VNET_NAME(tcp_do_rfc1323), 0,
196     "Enable rfc1323 (high performance TCP) extensions");
197 
198 static int	tcp_log_debug = 0;
199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
200     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
201 
202 static int	tcp_tcbhashsize;
203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
204     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
205 
206 static int	do_tcpdrain = 1;
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
208     "Enable tcp_drain routine for extra help when low on mbufs");
209 
210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
211     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
212 
213 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
214 #define	V_icmp_may_rst			VNET(icmp_may_rst)
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
216     &VNET_NAME(icmp_may_rst), 0,
217     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
218 
219 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
220 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
222     &VNET_NAME(tcp_isn_reseed_interval), 0,
223     "Seconds between reseeding of ISN secret");
224 
225 static int	tcp_soreceive_stream;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
227     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
228 
229 VNET_DEFINE(uma_zone_t, sack_hole_zone);
230 #define	V_sack_hole_zone		VNET(sack_hole_zone)
231 
232 #ifdef TCP_HHOOK
233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
234 #endif
235 
236 #define TS_OFFSET_SECRET_LENGTH 32
237 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
238 #define	V_ts_offset_secret	VNET(ts_offset_secret)
239 
240 static int	tcp_default_fb_init(struct tcpcb *tp);
241 static void	tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
242 static int	tcp_default_handoff_ok(struct tcpcb *tp);
243 static struct inpcb *tcp_notify(struct inpcb *, int);
244 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
245 static void tcp_mtudisc(struct inpcb *, int);
246 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
247 		    void *ip4hdr, const void *ip6hdr);
248 
249 
250 static struct tcp_function_block tcp_def_funcblk = {
251 	.tfb_tcp_block_name = "freebsd",
252 	.tfb_tcp_output = tcp_output,
253 	.tfb_tcp_do_segment = tcp_do_segment,
254 	.tfb_tcp_ctloutput = tcp_default_ctloutput,
255 	.tfb_tcp_handoff_ok = tcp_default_handoff_ok,
256 	.tfb_tcp_fb_init = tcp_default_fb_init,
257 	.tfb_tcp_fb_fini = tcp_default_fb_fini,
258 };
259 
260 int t_functions_inited = 0;
261 static int tcp_fb_cnt = 0;
262 struct tcp_funchead t_functions;
263 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
264 
265 static void
266 init_tcp_functions(void)
267 {
268 	if (t_functions_inited == 0) {
269 		TAILQ_INIT(&t_functions);
270 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
271 		t_functions_inited = 1;
272 	}
273 }
274 
275 static struct tcp_function_block *
276 find_tcp_functions_locked(struct tcp_function_set *fs)
277 {
278 	struct tcp_function *f;
279 	struct tcp_function_block *blk=NULL;
280 
281 	TAILQ_FOREACH(f, &t_functions, tf_next) {
282 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
283 			blk = f->tf_fb;
284 			break;
285 		}
286 	}
287 	return(blk);
288 }
289 
290 static struct tcp_function_block *
291 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
292 {
293 	struct tcp_function_block *rblk=NULL;
294 	struct tcp_function *f;
295 
296 	TAILQ_FOREACH(f, &t_functions, tf_next) {
297 		if (f->tf_fb == blk) {
298 			rblk = blk;
299 			if (s) {
300 				*s = f;
301 			}
302 			break;
303 		}
304 	}
305 	return (rblk);
306 }
307 
308 struct tcp_function_block *
309 find_and_ref_tcp_functions(struct tcp_function_set *fs)
310 {
311 	struct tcp_function_block *blk;
312 
313 	rw_rlock(&tcp_function_lock);
314 	blk = find_tcp_functions_locked(fs);
315 	if (blk)
316 		refcount_acquire(&blk->tfb_refcnt);
317 	rw_runlock(&tcp_function_lock);
318 	return(blk);
319 }
320 
321 struct tcp_function_block *
322 find_and_ref_tcp_fb(struct tcp_function_block *blk)
323 {
324 	struct tcp_function_block *rblk;
325 
326 	rw_rlock(&tcp_function_lock);
327 	rblk = find_tcp_fb_locked(blk, NULL);
328 	if (rblk)
329 		refcount_acquire(&rblk->tfb_refcnt);
330 	rw_runlock(&tcp_function_lock);
331 	return(rblk);
332 }
333 
334 static struct tcp_function_block *
335 find_and_ref_tcp_default_fb(void)
336 {
337 	struct tcp_function_block *rblk;
338 
339 	rw_rlock(&tcp_function_lock);
340 	rblk = tcp_func_set_ptr;
341 	refcount_acquire(&rblk->tfb_refcnt);
342 	rw_runlock(&tcp_function_lock);
343 	return (rblk);
344 }
345 
346 void
347 tcp_switch_back_to_default(struct tcpcb *tp)
348 {
349 	struct tcp_function_block *tfb;
350 
351 	KASSERT(tp->t_fb != &tcp_def_funcblk,
352 	    ("%s: called by the built-in default stack", __func__));
353 
354 	/*
355 	 * Release the old stack. This function will either find a new one
356 	 * or panic.
357 	 */
358 	if (tp->t_fb->tfb_tcp_fb_fini != NULL)
359 		(*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
360 	refcount_release(&tp->t_fb->tfb_refcnt);
361 
362 	/*
363 	 * Now, we'll find a new function block to use.
364 	 * Start by trying the current user-selected
365 	 * default, unless this stack is the user-selected
366 	 * default.
367 	 */
368 	tfb = find_and_ref_tcp_default_fb();
369 	if (tfb == tp->t_fb) {
370 		refcount_release(&tfb->tfb_refcnt);
371 		tfb = NULL;
372 	}
373 	/* Does the stack accept this connection? */
374 	if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
375 	    (*tfb->tfb_tcp_handoff_ok)(tp)) {
376 		refcount_release(&tfb->tfb_refcnt);
377 		tfb = NULL;
378 	}
379 	/* Try to use that stack. */
380 	if (tfb != NULL) {
381 		/* Initialize the new stack. If it succeeds, we are done. */
382 		tp->t_fb = tfb;
383 		if (tp->t_fb->tfb_tcp_fb_init == NULL ||
384 		    (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
385 			return;
386 
387 		/*
388 		 * Initialization failed. Release the reference count on
389 		 * the stack.
390 		 */
391 		refcount_release(&tfb->tfb_refcnt);
392 	}
393 
394 	/*
395 	 * If that wasn't feasible, use the built-in default
396 	 * stack which is not allowed to reject anyone.
397 	 */
398 	tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
399 	if (tfb == NULL) {
400 		/* there always should be a default */
401 		panic("Can't refer to tcp_def_funcblk");
402 	}
403 	if (tfb->tfb_tcp_handoff_ok != NULL) {
404 		if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
405 			/* The default stack cannot say no */
406 			panic("Default stack rejects a new session?");
407 		}
408 	}
409 	tp->t_fb = tfb;
410 	if (tp->t_fb->tfb_tcp_fb_init != NULL &&
411 	    (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
412 		/* The default stack cannot fail */
413 		panic("Default stack initialization failed");
414 	}
415 }
416 
417 static int
418 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
419 {
420 	int error=ENOENT;
421 	struct tcp_function_set fs;
422 	struct tcp_function_block *blk;
423 
424 	memset(&fs, 0, sizeof(fs));
425 	rw_rlock(&tcp_function_lock);
426 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
427 	if (blk) {
428 		/* Found him */
429 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
430 		fs.pcbcnt = blk->tfb_refcnt;
431 	}
432 	rw_runlock(&tcp_function_lock);
433 	error = sysctl_handle_string(oidp, fs.function_set_name,
434 				     sizeof(fs.function_set_name), req);
435 
436 	/* Check for error or no change */
437 	if (error != 0 || req->newptr == NULL)
438 		return(error);
439 
440 	rw_wlock(&tcp_function_lock);
441 	blk = find_tcp_functions_locked(&fs);
442 	if ((blk == NULL) ||
443 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
444 		error = ENOENT;
445 		goto done;
446 	}
447 	tcp_func_set_ptr = blk;
448 done:
449 	rw_wunlock(&tcp_function_lock);
450 	return (error);
451 }
452 
453 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
454 	    CTLTYPE_STRING | CTLFLAG_RW,
455 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
456 	    "Set/get the default TCP functions");
457 
458 static int
459 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
460 {
461 	int error, cnt, linesz;
462 	struct tcp_function *f;
463 	char *buffer, *cp;
464 	size_t bufsz, outsz;
465 	bool alias;
466 
467 	cnt = 0;
468 	rw_rlock(&tcp_function_lock);
469 	TAILQ_FOREACH(f, &t_functions, tf_next) {
470 		cnt++;
471 	}
472 	rw_runlock(&tcp_function_lock);
473 
474 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
475 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
476 
477 	error = 0;
478 	cp = buffer;
479 
480 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
481 	    "Alias", "PCB count");
482 	cp += linesz;
483 	bufsz -= linesz;
484 	outsz = linesz;
485 
486 	rw_rlock(&tcp_function_lock);
487 	TAILQ_FOREACH(f, &t_functions, tf_next) {
488 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
489 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
490 		    f->tf_fb->tfb_tcp_block_name,
491 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
492 		    alias ? f->tf_name : "-",
493 		    f->tf_fb->tfb_refcnt);
494 		if (linesz >= bufsz) {
495 			error = EOVERFLOW;
496 			break;
497 		}
498 		cp += linesz;
499 		bufsz -= linesz;
500 		outsz += linesz;
501 	}
502 	rw_runlock(&tcp_function_lock);
503 	if (error == 0)
504 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
505 	free(buffer, M_TEMP);
506 	return (error);
507 }
508 
509 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
510 	    CTLTYPE_STRING|CTLFLAG_RD,
511 	    NULL, 0, sysctl_net_inet_list_available, "A",
512 	    "list available TCP Function sets");
513 
514 /*
515  * Exports one (struct tcp_function_info) for each alias/name.
516  */
517 static int
518 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
519 {
520 	int cnt, error;
521 	struct tcp_function *f;
522 	struct tcp_function_info tfi;
523 
524 	/*
525 	 * We don't allow writes.
526 	 */
527 	if (req->newptr != NULL)
528 		return (EINVAL);
529 
530 	/*
531 	 * Wire the old buffer so we can directly copy the functions to
532 	 * user space without dropping the lock.
533 	 */
534 	if (req->oldptr != NULL) {
535 		error = sysctl_wire_old_buffer(req, 0);
536 		if (error)
537 			return (error);
538 	}
539 
540 	/*
541 	 * Walk the list and copy out matching entries. If INVARIANTS
542 	 * is compiled in, also walk the list to verify the length of
543 	 * the list matches what we have recorded.
544 	 */
545 	rw_rlock(&tcp_function_lock);
546 
547 	cnt = 0;
548 #ifndef INVARIANTS
549 	if (req->oldptr == NULL) {
550 		cnt = tcp_fb_cnt;
551 		goto skip_loop;
552 	}
553 #endif
554 	TAILQ_FOREACH(f, &t_functions, tf_next) {
555 #ifdef INVARIANTS
556 		cnt++;
557 #endif
558 		if (req->oldptr != NULL) {
559 			bzero(&tfi, sizeof(tfi));
560 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
561 			tfi.tfi_id = f->tf_fb->tfb_id;
562 			(void)strncpy(tfi.tfi_alias, f->tf_name,
563 			    TCP_FUNCTION_NAME_LEN_MAX);
564 			tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
565 			(void)strncpy(tfi.tfi_name,
566 			    f->tf_fb->tfb_tcp_block_name,
567 			    TCP_FUNCTION_NAME_LEN_MAX);
568 			tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
569 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
570 			/*
571 			 * Don't stop on error, as that is the
572 			 * mechanism we use to accumulate length
573 			 * information if the buffer was too short.
574 			 */
575 		}
576 	}
577 	KASSERT(cnt == tcp_fb_cnt,
578 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
579 #ifndef INVARIANTS
580 skip_loop:
581 #endif
582 	rw_runlock(&tcp_function_lock);
583 	if (req->oldptr == NULL)
584 		error = SYSCTL_OUT(req, NULL,
585 		    (cnt + 1) * sizeof(struct tcp_function_info));
586 
587 	return (error);
588 }
589 
590 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
591 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
592 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
593 	    "List TCP function block name-to-ID mappings");
594 
595 /*
596  * tfb_tcp_handoff_ok() function for the default stack.
597  * Note that we'll basically try to take all comers.
598  */
599 static int
600 tcp_default_handoff_ok(struct tcpcb *tp)
601 {
602 
603 	return (0);
604 }
605 
606 /*
607  * tfb_tcp_fb_init() function for the default stack.
608  *
609  * This handles making sure we have appropriate timers set if you are
610  * transitioning a socket that has some amount of setup done.
611  *
612  * The init() fuction from the default can *never* return non-zero i.e.
613  * it is required to always succeed since it is the stack of last resort!
614  */
615 static int
616 tcp_default_fb_init(struct tcpcb *tp)
617 {
618 
619 	struct socket *so;
620 
621 	INP_WLOCK_ASSERT(tp->t_inpcb);
622 
623 	KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
624 	    ("%s: connection %p in unexpected state %d", __func__, tp,
625 	    tp->t_state));
626 
627 	/*
628 	 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
629 	 * know what to do for unexpected states (which includes TIME_WAIT).
630 	 */
631 	if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
632 		return (0);
633 
634 	/*
635 	 * Make sure some kind of transmission timer is set if there is
636 	 * outstanding data.
637 	 */
638 	so = tp->t_inpcb->inp_socket;
639 	if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
640 	    tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
641 	    tcp_timer_active(tp, TT_PERSIST))) {
642 		/*
643 		 * If the session has established and it looks like it should
644 		 * be in the persist state, set the persist timer. Otherwise,
645 		 * set the retransmit timer.
646 		 */
647 		if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
648 		    (int32_t)(tp->snd_nxt - tp->snd_una) <
649 		    (int32_t)sbavail(&so->so_snd))
650 			tcp_setpersist(tp);
651 		else
652 			tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
653 	}
654 
655 	/* All non-embryonic sessions get a keepalive timer. */
656 	if (!tcp_timer_active(tp, TT_KEEP))
657 		tcp_timer_activate(tp, TT_KEEP,
658 		    TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
659 		    TP_KEEPINIT(tp));
660 
661 	return (0);
662 }
663 
664 /*
665  * tfb_tcp_fb_fini() function for the default stack.
666  *
667  * This changes state as necessary (or prudent) to prepare for another stack
668  * to assume responsibility for the connection.
669  */
670 static void
671 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
672 {
673 
674 	INP_WLOCK_ASSERT(tp->t_inpcb);
675 	return;
676 }
677 
678 /*
679  * Target size of TCP PCB hash tables. Must be a power of two.
680  *
681  * Note that this can be overridden by the kernel environment
682  * variable net.inet.tcp.tcbhashsize
683  */
684 #ifndef TCBHASHSIZE
685 #define TCBHASHSIZE	0
686 #endif
687 
688 /*
689  * XXX
690  * Callouts should be moved into struct tcp directly.  They are currently
691  * separate because the tcpcb structure is exported to userland for sysctl
692  * parsing purposes, which do not know about callouts.
693  */
694 struct tcpcb_mem {
695 	struct	tcpcb		tcb;
696 	struct	tcp_timer	tt;
697 	struct	cc_var		ccv;
698 #ifdef TCP_HHOOK
699 	struct	osd		osd;
700 #endif
701 };
702 
703 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
704 #define	V_tcpcb_zone			VNET(tcpcb_zone)
705 
706 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
707 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
708 
709 static struct mtx isn_mtx;
710 
711 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
712 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
713 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
714 
715 /*
716  * TCP initialization.
717  */
718 static void
719 tcp_zone_change(void *tag)
720 {
721 
722 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
723 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
724 	tcp_tw_zone_change();
725 }
726 
727 static int
728 tcp_inpcb_init(void *mem, int size, int flags)
729 {
730 	struct inpcb *inp = mem;
731 
732 	INP_LOCK_INIT(inp, "inp", "tcpinp");
733 	return (0);
734 }
735 
736 /*
737  * Take a value and get the next power of 2 that doesn't overflow.
738  * Used to size the tcp_inpcb hash buckets.
739  */
740 static int
741 maketcp_hashsize(int size)
742 {
743 	int hashsize;
744 
745 	/*
746 	 * auto tune.
747 	 * get the next power of 2 higher than maxsockets.
748 	 */
749 	hashsize = 1 << fls(size);
750 	/* catch overflow, and just go one power of 2 smaller */
751 	if (hashsize < size) {
752 		hashsize = 1 << (fls(size) - 1);
753 	}
754 	return (hashsize);
755 }
756 
757 static volatile int next_tcp_stack_id = 1;
758 
759 /*
760  * Register a TCP function block with the name provided in the names
761  * array.  (Note that this function does NOT automatically register
762  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
763  * explicitly include blk->tfb_tcp_block_name in the list of names if
764  * you wish to register the stack with that name.)
765  *
766  * Either all name registrations will succeed or all will fail.  If
767  * a name registration fails, the function will update the num_names
768  * argument to point to the array index of the name that encountered
769  * the failure.
770  *
771  * Returns 0 on success, or an error code on failure.
772  */
773 int
774 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
775     const char *names[], int *num_names)
776 {
777 	struct tcp_function *n;
778 	struct tcp_function_set fs;
779 	int error, i;
780 
781 	KASSERT(names != NULL && *num_names > 0,
782 	    ("%s: Called with 0-length name list", __func__));
783 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
784 
785 	if (t_functions_inited == 0) {
786 		init_tcp_functions();
787 	}
788 	if ((blk->tfb_tcp_output == NULL) ||
789 	    (blk->tfb_tcp_do_segment == NULL) ||
790 	    (blk->tfb_tcp_ctloutput == NULL) ||
791 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
792 		/*
793 		 * These functions are required and you
794 		 * need a name.
795 		 */
796 		*num_names = 0;
797 		return (EINVAL);
798 	}
799 	if (blk->tfb_tcp_timer_stop_all ||
800 	    blk->tfb_tcp_timer_activate ||
801 	    blk->tfb_tcp_timer_active ||
802 	    blk->tfb_tcp_timer_stop) {
803 		/*
804 		 * If you define one timer function you
805 		 * must have them all.
806 		 */
807 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
808 		    (blk->tfb_tcp_timer_activate == NULL) ||
809 		    (blk->tfb_tcp_timer_active == NULL) ||
810 		    (blk->tfb_tcp_timer_stop == NULL)) {
811 			*num_names = 0;
812 			return (EINVAL);
813 		}
814 	}
815 
816 	refcount_init(&blk->tfb_refcnt, 0);
817 	blk->tfb_flags = 0;
818 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
819 	for (i = 0; i < *num_names; i++) {
820 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
821 		if (n == NULL) {
822 			error = ENOMEM;
823 			goto cleanup;
824 		}
825 		n->tf_fb = blk;
826 
827 		(void)strncpy(fs.function_set_name, names[i],
828 		    TCP_FUNCTION_NAME_LEN_MAX);
829 		fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
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)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
839 		n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
840 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
841 		tcp_fb_cnt++;
842 		rw_wunlock(&tcp_function_lock);
843 	}
844 	return(0);
845 
846 cleanup:
847 	/*
848 	 * Deregister the names we just added. Because registration failed
849 	 * for names[i], we don't need to deregister that name.
850 	 */
851 	*num_names = i;
852 	rw_wlock(&tcp_function_lock);
853 	while (--i >= 0) {
854 		TAILQ_FOREACH(n, &t_functions, tf_next) {
855 			if (!strncmp(n->tf_name, names[i],
856 			    TCP_FUNCTION_NAME_LEN_MAX)) {
857 				TAILQ_REMOVE(&t_functions, n, tf_next);
858 				tcp_fb_cnt--;
859 				n->tf_fb = NULL;
860 				free(n, M_TCPFUNCTIONS);
861 				break;
862 			}
863 		}
864 	}
865 	rw_wunlock(&tcp_function_lock);
866 	return (error);
867 }
868 
869 /*
870  * Register a TCP function block using the name provided in the name
871  * argument.
872  *
873  * Returns 0 on success, or an error code on failure.
874  */
875 int
876 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
877     int wait)
878 {
879 	const char *name_list[1];
880 	int num_names, rv;
881 
882 	num_names = 1;
883 	if (name != NULL)
884 		name_list[0] = name;
885 	else
886 		name_list[0] = blk->tfb_tcp_block_name;
887 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
888 	return (rv);
889 }
890 
891 /*
892  * Register a TCP function block using the name defined in
893  * blk->tfb_tcp_block_name.
894  *
895  * Returns 0 on success, or an error code on failure.
896  */
897 int
898 register_tcp_functions(struct tcp_function_block *blk, int wait)
899 {
900 
901 	return (register_tcp_functions_as_name(blk, NULL, wait));
902 }
903 
904 /*
905  * Deregister all names associated with a function block. This
906  * functionally removes the function block from use within the system.
907  *
908  * When called with a true quiesce argument, mark the function block
909  * as being removed so no more stacks will use it and determine
910  * whether the removal would succeed.
911  *
912  * When called with a false quiesce argument, actually attempt the
913  * removal.
914  *
915  * When called with a force argument, attempt to switch all TCBs to
916  * use the default stack instead of returning EBUSY.
917  *
918  * Returns 0 on success (or if the removal would succeed, or an error
919  * code on failure.
920  */
921 int
922 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
923     bool force)
924 {
925 	struct tcp_function *f;
926 
927 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
928 		/* You can't un-register the default */
929 		return (EPERM);
930 	}
931 	rw_wlock(&tcp_function_lock);
932 	if (blk == tcp_func_set_ptr) {
933 		/* You can't free the current default */
934 		rw_wunlock(&tcp_function_lock);
935 		return (EBUSY);
936 	}
937 	/* Mark the block so no more stacks can use it. */
938 	blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
939 	/*
940 	 * If TCBs are still attached to the stack, attempt to switch them
941 	 * to the default stack.
942 	 */
943 	if (force && blk->tfb_refcnt) {
944 		struct inpcb *inp;
945 		struct tcpcb *tp;
946 		VNET_ITERATOR_DECL(vnet_iter);
947 
948 		rw_wunlock(&tcp_function_lock);
949 
950 		VNET_LIST_RLOCK();
951 		VNET_FOREACH(vnet_iter) {
952 			CURVNET_SET(vnet_iter);
953 			INP_INFO_WLOCK(&V_tcbinfo);
954 			CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
955 				INP_WLOCK(inp);
956 				if (inp->inp_flags & INP_TIMEWAIT) {
957 					INP_WUNLOCK(inp);
958 					continue;
959 				}
960 				tp = intotcpcb(inp);
961 				if (tp == NULL || tp->t_fb != blk) {
962 					INP_WUNLOCK(inp);
963 					continue;
964 				}
965 				tcp_switch_back_to_default(tp);
966 				INP_WUNLOCK(inp);
967 			}
968 			INP_INFO_WUNLOCK(&V_tcbinfo);
969 			CURVNET_RESTORE();
970 		}
971 		VNET_LIST_RUNLOCK();
972 
973 		rw_wlock(&tcp_function_lock);
974 	}
975 	if (blk->tfb_refcnt) {
976 		/* TCBs still attached. */
977 		rw_wunlock(&tcp_function_lock);
978 		return (EBUSY);
979 	}
980 	if (quiesce) {
981 		/* Skip removal. */
982 		rw_wunlock(&tcp_function_lock);
983 		return (0);
984 	}
985 	/* Remove any function names that map to this function block. */
986 	while (find_tcp_fb_locked(blk, &f) != NULL) {
987 		TAILQ_REMOVE(&t_functions, f, tf_next);
988 		tcp_fb_cnt--;
989 		f->tf_fb = NULL;
990 		free(f, M_TCPFUNCTIONS);
991 	}
992 	rw_wunlock(&tcp_function_lock);
993 	return (0);
994 }
995 
996 void
997 tcp_init(void)
998 {
999 	const char *tcbhash_tuneable;
1000 	int hashsize;
1001 
1002 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1003 
1004 #ifdef TCP_HHOOK
1005 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1006 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1007 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1008 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1009 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1010 		printf("%s: WARNING: unable to register helper hook\n", __func__);
1011 #endif
1012 	hashsize = TCBHASHSIZE;
1013 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1014 	if (hashsize == 0) {
1015 		/*
1016 		 * Auto tune the hash size based on maxsockets.
1017 		 * A perfect hash would have a 1:1 mapping
1018 		 * (hashsize = maxsockets) however it's been
1019 		 * suggested that O(2) average is better.
1020 		 */
1021 		hashsize = maketcp_hashsize(maxsockets / 4);
1022 		/*
1023 		 * Our historical default is 512,
1024 		 * do not autotune lower than this.
1025 		 */
1026 		if (hashsize < 512)
1027 			hashsize = 512;
1028 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
1029 			printf("%s: %s auto tuned to %d\n", __func__,
1030 			    tcbhash_tuneable, hashsize);
1031 	}
1032 	/*
1033 	 * We require a hashsize to be a power of two.
1034 	 * Previously if it was not a power of two we would just reset it
1035 	 * back to 512, which could be a nasty surprise if you did not notice
1036 	 * the error message.
1037 	 * Instead what we do is clip it to the closest power of two lower
1038 	 * than the specified hash value.
1039 	 */
1040 	if (!powerof2(hashsize)) {
1041 		int oldhashsize = hashsize;
1042 
1043 		hashsize = maketcp_hashsize(hashsize);
1044 		/* prevent absurdly low value */
1045 		if (hashsize < 16)
1046 			hashsize = 16;
1047 		printf("%s: WARNING: TCB hash size not a power of 2, "
1048 		    "clipped from %d to %d.\n", __func__, oldhashsize,
1049 		    hashsize);
1050 	}
1051 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1052 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1053 
1054 	/*
1055 	 * These have to be type stable for the benefit of the timers.
1056 	 */
1057 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1058 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1059 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
1060 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1061 
1062 	tcp_tw_init();
1063 	syncache_init();
1064 	tcp_hc_init();
1065 
1066 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1067 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1068 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1069 
1070 	tcp_fastopen_init();
1071 
1072 	/* Skip initialization of globals for non-default instances. */
1073 	if (!IS_DEFAULT_VNET(curvnet))
1074 		return;
1075 
1076 	tcp_reass_global_init();
1077 
1078 	/* XXX virtualize those bellow? */
1079 	tcp_delacktime = TCPTV_DELACK;
1080 	tcp_keepinit = TCPTV_KEEP_INIT;
1081 	tcp_keepidle = TCPTV_KEEP_IDLE;
1082 	tcp_keepintvl = TCPTV_KEEPINTVL;
1083 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1084 	tcp_msl = TCPTV_MSL;
1085 	tcp_rexmit_min = TCPTV_MIN;
1086 	if (tcp_rexmit_min < 1)
1087 		tcp_rexmit_min = 1;
1088 	tcp_persmin = TCPTV_PERSMIN;
1089 	tcp_persmax = TCPTV_PERSMAX;
1090 	tcp_rexmit_slop = TCPTV_CPU_VAR;
1091 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1092 	tcp_tcbhashsize = hashsize;
1093 	/* Setup the tcp function block list */
1094 	init_tcp_functions();
1095 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1096 #ifdef TCP_BLACKBOX
1097 	/* Initialize the TCP logging data. */
1098 	tcp_log_init();
1099 #endif
1100 	arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1101 
1102 	if (tcp_soreceive_stream) {
1103 #ifdef INET
1104 		tcp_usrreqs.pru_soreceive = soreceive_stream;
1105 #endif
1106 #ifdef INET6
1107 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
1108 #endif /* INET6 */
1109 	}
1110 
1111 #ifdef INET6
1112 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1113 #else /* INET6 */
1114 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1115 #endif /* INET6 */
1116 	if (max_protohdr < TCP_MINPROTOHDR)
1117 		max_protohdr = TCP_MINPROTOHDR;
1118 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1119 		panic("tcp_init");
1120 #undef TCP_MINPROTOHDR
1121 
1122 	ISN_LOCK_INIT();
1123 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1124 		SHUTDOWN_PRI_DEFAULT);
1125 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1126 		EVENTHANDLER_PRI_ANY);
1127 #ifdef TCPPCAP
1128 	tcp_pcap_init();
1129 #endif
1130 }
1131 
1132 #ifdef VIMAGE
1133 static void
1134 tcp_destroy(void *unused __unused)
1135 {
1136 	int n;
1137 #ifdef TCP_HHOOK
1138 	int error;
1139 #endif
1140 
1141 	/*
1142 	 * All our processes are gone, all our sockets should be cleaned
1143 	 * up, which means, we should be past the tcp_discardcb() calls.
1144 	 * Sleep to let all tcpcb timers really disappear and cleanup.
1145 	 */
1146 	for (;;) {
1147 		INP_LIST_RLOCK(&V_tcbinfo);
1148 		n = V_tcbinfo.ipi_count;
1149 		INP_LIST_RUNLOCK(&V_tcbinfo);
1150 		if (n == 0)
1151 			break;
1152 		pause("tcpdes", hz / 10);
1153 	}
1154 	tcp_hc_destroy();
1155 	syncache_destroy();
1156 	tcp_tw_destroy();
1157 	in_pcbinfo_destroy(&V_tcbinfo);
1158 	/* tcp_discardcb() clears the sack_holes up. */
1159 	uma_zdestroy(V_sack_hole_zone);
1160 	uma_zdestroy(V_tcpcb_zone);
1161 
1162 	/*
1163 	 * Cannot free the zone until all tcpcbs are released as we attach
1164 	 * the allocations to them.
1165 	 */
1166 	tcp_fastopen_destroy();
1167 
1168 #ifdef TCP_HHOOK
1169 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1170 	if (error != 0) {
1171 		printf("%s: WARNING: unable to deregister helper hook "
1172 		    "type=%d, id=%d: error %d returned\n", __func__,
1173 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1174 	}
1175 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1176 	if (error != 0) {
1177 		printf("%s: WARNING: unable to deregister helper hook "
1178 		    "type=%d, id=%d: error %d returned\n", __func__,
1179 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1180 	}
1181 #endif
1182 }
1183 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1184 #endif
1185 
1186 void
1187 tcp_fini(void *xtp)
1188 {
1189 
1190 }
1191 
1192 /*
1193  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1194  * tcp_template used to store this data in mbufs, but we now recopy it out
1195  * of the tcpcb each time to conserve mbufs.
1196  */
1197 void
1198 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
1199 {
1200 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1201 
1202 	INP_WLOCK_ASSERT(inp);
1203 
1204 #ifdef INET6
1205 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1206 		struct ip6_hdr *ip6;
1207 
1208 		ip6 = (struct ip6_hdr *)ip_ptr;
1209 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1210 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
1211 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1212 			(IPV6_VERSION & IPV6_VERSION_MASK);
1213 		ip6->ip6_nxt = IPPROTO_TCP;
1214 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
1215 		ip6->ip6_src = inp->in6p_laddr;
1216 		ip6->ip6_dst = inp->in6p_faddr;
1217 	}
1218 #endif /* INET6 */
1219 #if defined(INET6) && defined(INET)
1220 	else
1221 #endif
1222 #ifdef INET
1223 	{
1224 		struct ip *ip;
1225 
1226 		ip = (struct ip *)ip_ptr;
1227 		ip->ip_v = IPVERSION;
1228 		ip->ip_hl = 5;
1229 		ip->ip_tos = inp->inp_ip_tos;
1230 		ip->ip_len = 0;
1231 		ip->ip_id = 0;
1232 		ip->ip_off = 0;
1233 		ip->ip_ttl = inp->inp_ip_ttl;
1234 		ip->ip_sum = 0;
1235 		ip->ip_p = IPPROTO_TCP;
1236 		ip->ip_src = inp->inp_laddr;
1237 		ip->ip_dst = inp->inp_faddr;
1238 	}
1239 #endif /* INET */
1240 	th->th_sport = inp->inp_lport;
1241 	th->th_dport = inp->inp_fport;
1242 	th->th_seq = 0;
1243 	th->th_ack = 0;
1244 	th->th_x2 = 0;
1245 	th->th_off = 5;
1246 	th->th_flags = 0;
1247 	th->th_win = 0;
1248 	th->th_urp = 0;
1249 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1250 }
1251 
1252 /*
1253  * Create template to be used to send tcp packets on a connection.
1254  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1255  * use for this function is in keepalives, which use tcp_respond.
1256  */
1257 struct tcptemp *
1258 tcpip_maketemplate(struct inpcb *inp)
1259 {
1260 	struct tcptemp *t;
1261 
1262 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1263 	if (t == NULL)
1264 		return (NULL);
1265 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1266 	return (t);
1267 }
1268 
1269 /*
1270  * Send a single message to the TCP at address specified by
1271  * the given TCP/IP header.  If m == NULL, then we make a copy
1272  * of the tcpiphdr at th and send directly to the addressed host.
1273  * This is used to force keep alive messages out using the TCP
1274  * template for a connection.  If flags are given then we send
1275  * a message back to the TCP which originated the segment th,
1276  * and discard the mbuf containing it and any other attached mbufs.
1277  *
1278  * In any case the ack and sequence number of the transmitted
1279  * segment are as specified by the parameters.
1280  *
1281  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1282  */
1283 void
1284 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1285     tcp_seq ack, tcp_seq seq, int flags)
1286 {
1287 	struct tcpopt to;
1288 	struct inpcb *inp;
1289 	struct ip *ip;
1290 	struct mbuf *optm;
1291 	struct tcphdr *nth;
1292 	u_char *optp;
1293 #ifdef INET6
1294 	struct ip6_hdr *ip6;
1295 	int isipv6;
1296 #endif /* INET6 */
1297 	int optlen, tlen, win;
1298 	bool incl_opts;
1299 
1300 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1301 
1302 #ifdef INET6
1303 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1304 	ip6 = ipgen;
1305 #endif /* INET6 */
1306 	ip = ipgen;
1307 
1308 	if (tp != NULL) {
1309 		inp = tp->t_inpcb;
1310 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1311 		INP_WLOCK_ASSERT(inp);
1312 	} else
1313 		inp = NULL;
1314 
1315 	incl_opts = false;
1316 	win = 0;
1317 	if (tp != NULL) {
1318 		if (!(flags & TH_RST)) {
1319 			win = sbspace(&inp->inp_socket->so_rcv);
1320 			if (win > TCP_MAXWIN << tp->rcv_scale)
1321 				win = TCP_MAXWIN << tp->rcv_scale;
1322 		}
1323 		if ((tp->t_flags & TF_NOOPT) == 0)
1324 			incl_opts = true;
1325 	}
1326 	if (m == NULL) {
1327 		m = m_gethdr(M_NOWAIT, MT_DATA);
1328 		if (m == NULL)
1329 			return;
1330 		m->m_data += max_linkhdr;
1331 #ifdef INET6
1332 		if (isipv6) {
1333 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1334 			      sizeof(struct ip6_hdr));
1335 			ip6 = mtod(m, struct ip6_hdr *);
1336 			nth = (struct tcphdr *)(ip6 + 1);
1337 		} else
1338 #endif /* INET6 */
1339 		{
1340 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1341 			ip = mtod(m, struct ip *);
1342 			nth = (struct tcphdr *)(ip + 1);
1343 		}
1344 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1345 		flags = TH_ACK;
1346 	} else if (!M_WRITABLE(m)) {
1347 		struct mbuf *n;
1348 
1349 		/* Can't reuse 'm', allocate a new mbuf. */
1350 		n = m_gethdr(M_NOWAIT, MT_DATA);
1351 		if (n == NULL) {
1352 			m_freem(m);
1353 			return;
1354 		}
1355 
1356 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1357 			m_freem(m);
1358 			m_freem(n);
1359 			return;
1360 		}
1361 
1362 		n->m_data += max_linkhdr;
1363 		/* m_len is set later */
1364 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1365 #ifdef INET6
1366 		if (isipv6) {
1367 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1368 			      sizeof(struct ip6_hdr));
1369 			ip6 = mtod(n, struct ip6_hdr *);
1370 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1371 			nth = (struct tcphdr *)(ip6 + 1);
1372 		} else
1373 #endif /* INET6 */
1374 		{
1375 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1376 			ip = mtod(n, struct ip *);
1377 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1378 			nth = (struct tcphdr *)(ip + 1);
1379 		}
1380 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1381 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1382 		th = nth;
1383 		m_freem(m);
1384 		m = n;
1385 	} else {
1386 		/*
1387 		 *  reuse the mbuf.
1388 		 * XXX MRT We inherit the FIB, which is lucky.
1389 		 */
1390 		m_freem(m->m_next);
1391 		m->m_next = NULL;
1392 		m->m_data = (caddr_t)ipgen;
1393 		/* m_len is set later */
1394 #ifdef INET6
1395 		if (isipv6) {
1396 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1397 			nth = (struct tcphdr *)(ip6 + 1);
1398 		} else
1399 #endif /* INET6 */
1400 		{
1401 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1402 			nth = (struct tcphdr *)(ip + 1);
1403 		}
1404 		if (th != nth) {
1405 			/*
1406 			 * this is usually a case when an extension header
1407 			 * exists between the IPv6 header and the
1408 			 * TCP header.
1409 			 */
1410 			nth->th_sport = th->th_sport;
1411 			nth->th_dport = th->th_dport;
1412 		}
1413 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1414 #undef xchg
1415 	}
1416 	tlen = 0;
1417 #ifdef INET6
1418 	if (isipv6)
1419 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1420 #endif
1421 #if defined(INET) && defined(INET6)
1422 	else
1423 #endif
1424 #ifdef INET
1425 		tlen = sizeof (struct tcpiphdr);
1426 #endif
1427 #ifdef INVARIANTS
1428 	m->m_len = 0;
1429 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1430 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1431 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1432 #endif
1433 	m->m_len = tlen;
1434 	to.to_flags = 0;
1435 	if (incl_opts) {
1436 		/* Make sure we have room. */
1437 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1438 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1439 			if (m->m_next) {
1440 				optp = mtod(m->m_next, u_char *);
1441 				optm = m->m_next;
1442 			} else
1443 				incl_opts = false;
1444 		} else {
1445 			optp = (u_char *) (nth + 1);
1446 			optm = m;
1447 		}
1448 	}
1449 	if (incl_opts) {
1450 		/* Timestamps. */
1451 		if (tp->t_flags & TF_RCVD_TSTMP) {
1452 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1453 			to.to_tsecr = tp->ts_recent;
1454 			to.to_flags |= TOF_TS;
1455 		}
1456 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1457 		/* TCP-MD5 (RFC2385). */
1458 		if (tp->t_flags & TF_SIGNATURE)
1459 			to.to_flags |= TOF_SIGNATURE;
1460 #endif
1461 		/* Add the options. */
1462 		tlen += optlen = tcp_addoptions(&to, optp);
1463 
1464 		/* Update m_len in the correct mbuf. */
1465 		optm->m_len += optlen;
1466 	} else
1467 		optlen = 0;
1468 #ifdef INET6
1469 	if (isipv6) {
1470 		ip6->ip6_flow = 0;
1471 		ip6->ip6_vfc = IPV6_VERSION;
1472 		ip6->ip6_nxt = IPPROTO_TCP;
1473 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1474 	}
1475 #endif
1476 #if defined(INET) && defined(INET6)
1477 	else
1478 #endif
1479 #ifdef INET
1480 	{
1481 		ip->ip_len = htons(tlen);
1482 		ip->ip_ttl = V_ip_defttl;
1483 		if (V_path_mtu_discovery)
1484 			ip->ip_off |= htons(IP_DF);
1485 	}
1486 #endif
1487 	m->m_pkthdr.len = tlen;
1488 	m->m_pkthdr.rcvif = NULL;
1489 #ifdef MAC
1490 	if (inp != NULL) {
1491 		/*
1492 		 * Packet is associated with a socket, so allow the
1493 		 * label of the response to reflect the socket label.
1494 		 */
1495 		INP_WLOCK_ASSERT(inp);
1496 		mac_inpcb_create_mbuf(inp, m);
1497 	} else {
1498 		/*
1499 		 * Packet is not associated with a socket, so possibly
1500 		 * update the label in place.
1501 		 */
1502 		mac_netinet_tcp_reply(m);
1503 	}
1504 #endif
1505 	nth->th_seq = htonl(seq);
1506 	nth->th_ack = htonl(ack);
1507 	nth->th_x2 = 0;
1508 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1509 	nth->th_flags = flags;
1510 	if (tp != NULL)
1511 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1512 	else
1513 		nth->th_win = htons((u_short)win);
1514 	nth->th_urp = 0;
1515 
1516 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1517 	if (to.to_flags & TOF_SIGNATURE) {
1518 		if (!TCPMD5_ENABLED() ||
1519 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1520 			m_freem(m);
1521 			return;
1522 		}
1523 	}
1524 #endif
1525 
1526 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1527 #ifdef INET6
1528 	if (isipv6) {
1529 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1530 		nth->th_sum = in6_cksum_pseudo(ip6,
1531 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1532 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1533 		    NULL, NULL);
1534 	}
1535 #endif /* INET6 */
1536 #if defined(INET6) && defined(INET)
1537 	else
1538 #endif
1539 #ifdef INET
1540 	{
1541 		m->m_pkthdr.csum_flags = CSUM_TCP;
1542 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1543 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1544 	}
1545 #endif /* INET */
1546 #ifdef TCPDEBUG
1547 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1548 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1549 #endif
1550 	TCP_PROBE3(debug__output, tp, th, m);
1551 	if (flags & TH_RST)
1552 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1553 
1554 #ifdef INET6
1555 	if (isipv6) {
1556 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1557 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1558 	}
1559 #endif /* INET6 */
1560 #if defined(INET) && defined(INET6)
1561 	else
1562 #endif
1563 #ifdef INET
1564 	{
1565 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1566 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1567 	}
1568 #endif
1569 }
1570 
1571 /*
1572  * Create a new TCP control block, making an
1573  * empty reassembly queue and hooking it to the argument
1574  * protocol control block.  The `inp' parameter must have
1575  * come from the zone allocator set up in tcp_init().
1576  */
1577 struct tcpcb *
1578 tcp_newtcpcb(struct inpcb *inp)
1579 {
1580 	struct tcpcb_mem *tm;
1581 	struct tcpcb *tp;
1582 #ifdef INET6
1583 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1584 #endif /* INET6 */
1585 
1586 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1587 	if (tm == NULL)
1588 		return (NULL);
1589 	tp = &tm->tcb;
1590 
1591 	/* Initialise cc_var struct for this tcpcb. */
1592 	tp->ccv = &tm->ccv;
1593 	tp->ccv->type = IPPROTO_TCP;
1594 	tp->ccv->ccvc.tcp = tp;
1595 	rw_rlock(&tcp_function_lock);
1596 	tp->t_fb = tcp_func_set_ptr;
1597 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1598 	rw_runlock(&tcp_function_lock);
1599 	/*
1600 	 * Use the current system default CC algorithm.
1601 	 */
1602 	CC_LIST_RLOCK();
1603 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1604 	CC_ALGO(tp) = CC_DEFAULT();
1605 	CC_LIST_RUNLOCK();
1606 
1607 	if (CC_ALGO(tp)->cb_init != NULL)
1608 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1609 			if (tp->t_fb->tfb_tcp_fb_fini)
1610 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1611 			refcount_release(&tp->t_fb->tfb_refcnt);
1612 			uma_zfree(V_tcpcb_zone, tm);
1613 			return (NULL);
1614 		}
1615 
1616 #ifdef TCP_HHOOK
1617 	tp->osd = &tm->osd;
1618 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1619 		if (tp->t_fb->tfb_tcp_fb_fini)
1620 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1621 		refcount_release(&tp->t_fb->tfb_refcnt);
1622 		uma_zfree(V_tcpcb_zone, tm);
1623 		return (NULL);
1624 	}
1625 #endif
1626 
1627 #ifdef VIMAGE
1628 	tp->t_vnet = inp->inp_vnet;
1629 #endif
1630 	tp->t_timers = &tm->tt;
1631 	TAILQ_INIT(&tp->t_segq);
1632 	tp->t_maxseg =
1633 #ifdef INET6
1634 		isipv6 ? V_tcp_v6mssdflt :
1635 #endif /* INET6 */
1636 		V_tcp_mssdflt;
1637 
1638 	/* Set up our timeouts. */
1639 	callout_init(&tp->t_timers->tt_rexmt, 1);
1640 	callout_init(&tp->t_timers->tt_persist, 1);
1641 	callout_init(&tp->t_timers->tt_keep, 1);
1642 	callout_init(&tp->t_timers->tt_2msl, 1);
1643 	callout_init(&tp->t_timers->tt_delack, 1);
1644 
1645 	if (V_tcp_do_rfc1323)
1646 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1647 	if (V_tcp_do_sack)
1648 		tp->t_flags |= TF_SACK_PERMIT;
1649 	TAILQ_INIT(&tp->snd_holes);
1650 	/*
1651 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1652 	 * is called.
1653 	 */
1654 	in_pcbref(inp);	/* Reference for tcpcb */
1655 	tp->t_inpcb = inp;
1656 
1657 	/*
1658 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1659 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1660 	 * reasonable initial retransmit time.
1661 	 */
1662 	tp->t_srtt = TCPTV_SRTTBASE;
1663 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1664 	tp->t_rttmin = tcp_rexmit_min;
1665 	tp->t_rxtcur = TCPTV_RTOBASE;
1666 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1667 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1668 	tp->t_rcvtime = ticks;
1669 	/*
1670 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1671 	 * because the socket may be bound to an IPv6 wildcard address,
1672 	 * which may match an IPv4-mapped IPv6 address.
1673 	 */
1674 	inp->inp_ip_ttl = V_ip_defttl;
1675 	inp->inp_ppcb = tp;
1676 #ifdef TCPPCAP
1677 	/*
1678 	 * Init the TCP PCAP queues.
1679 	 */
1680 	tcp_pcap_tcpcb_init(tp);
1681 #endif
1682 #ifdef TCP_BLACKBOX
1683 	/* Initialize the per-TCPCB log data. */
1684 	tcp_log_tcpcbinit(tp);
1685 #endif
1686 	if (tp->t_fb->tfb_tcp_fb_init) {
1687 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1688 	}
1689 	return (tp);		/* XXX */
1690 }
1691 
1692 /*
1693  * Switch the congestion control algorithm back to NewReno for any active
1694  * control blocks using an algorithm which is about to go away.
1695  * This ensures the CC framework can allow the unload to proceed without leaving
1696  * any dangling pointers which would trigger a panic.
1697  * Returning non-zero would inform the CC framework that something went wrong
1698  * and it would be unsafe to allow the unload to proceed. However, there is no
1699  * way for this to occur with this implementation so we always return zero.
1700  */
1701 int
1702 tcp_ccalgounload(struct cc_algo *unload_algo)
1703 {
1704 	struct cc_algo *tmpalgo;
1705 	struct inpcb *inp;
1706 	struct tcpcb *tp;
1707 	VNET_ITERATOR_DECL(vnet_iter);
1708 
1709 	/*
1710 	 * Check all active control blocks across all network stacks and change
1711 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1712 	 * requires cleanup code to be run, call it.
1713 	 */
1714 	VNET_LIST_RLOCK();
1715 	VNET_FOREACH(vnet_iter) {
1716 		CURVNET_SET(vnet_iter);
1717 		INP_INFO_WLOCK(&V_tcbinfo);
1718 		/*
1719 		 * New connections already part way through being initialised
1720 		 * with the CC algo we're removing will not race with this code
1721 		 * because the INP_INFO_WLOCK is held during initialisation. We
1722 		 * therefore don't enter the loop below until the connection
1723 		 * list has stabilised.
1724 		 */
1725 		CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
1726 			INP_WLOCK(inp);
1727 			/* Important to skip tcptw structs. */
1728 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1729 			    (tp = intotcpcb(inp)) != NULL) {
1730 				/*
1731 				 * By holding INP_WLOCK here, we are assured
1732 				 * that the connection is not currently
1733 				 * executing inside the CC module's functions
1734 				 * i.e. it is safe to make the switch back to
1735 				 * NewReno.
1736 				 */
1737 				if (CC_ALGO(tp) == unload_algo) {
1738 					tmpalgo = CC_ALGO(tp);
1739 					if (tmpalgo->cb_destroy != NULL)
1740 						tmpalgo->cb_destroy(tp->ccv);
1741 					CC_DATA(tp) = NULL;
1742 					/*
1743 					 * NewReno may allocate memory on
1744 					 * demand for certain stateful
1745 					 * configuration as needed, but is
1746 					 * coded to never fail on memory
1747 					 * allocation failure so it is a safe
1748 					 * fallback.
1749 					 */
1750 					CC_ALGO(tp) = &newreno_cc_algo;
1751 				}
1752 			}
1753 			INP_WUNLOCK(inp);
1754 		}
1755 		INP_INFO_WUNLOCK(&V_tcbinfo);
1756 		CURVNET_RESTORE();
1757 	}
1758 	VNET_LIST_RUNLOCK();
1759 
1760 	return (0);
1761 }
1762 
1763 /*
1764  * Drop a TCP connection, reporting
1765  * the specified error.  If connection is synchronized,
1766  * then send a RST to peer.
1767  */
1768 struct tcpcb *
1769 tcp_drop(struct tcpcb *tp, int errno)
1770 {
1771 	struct socket *so = tp->t_inpcb->inp_socket;
1772 
1773 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1774 	INP_WLOCK_ASSERT(tp->t_inpcb);
1775 
1776 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1777 		tcp_state_change(tp, TCPS_CLOSED);
1778 		(void) tp->t_fb->tfb_tcp_output(tp);
1779 		TCPSTAT_INC(tcps_drops);
1780 	} else
1781 		TCPSTAT_INC(tcps_conndrops);
1782 	if (errno == ETIMEDOUT && tp->t_softerror)
1783 		errno = tp->t_softerror;
1784 	so->so_error = errno;
1785 	return (tcp_close(tp));
1786 }
1787 
1788 void
1789 tcp_discardcb(struct tcpcb *tp)
1790 {
1791 	struct inpcb *inp = tp->t_inpcb;
1792 	struct socket *so = inp->inp_socket;
1793 #ifdef INET6
1794 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1795 #endif /* INET6 */
1796 	int released __unused;
1797 
1798 	INP_WLOCK_ASSERT(inp);
1799 
1800 	/*
1801 	 * Make sure that all of our timers are stopped before we delete the
1802 	 * PCB.
1803 	 *
1804 	 * If stopping a timer fails, we schedule a discard function in same
1805 	 * callout, and the last discard function called will take care of
1806 	 * deleting the tcpcb.
1807 	 */
1808 	tp->t_timers->tt_draincnt = 0;
1809 	tcp_timer_stop(tp, TT_REXMT);
1810 	tcp_timer_stop(tp, TT_PERSIST);
1811 	tcp_timer_stop(tp, TT_KEEP);
1812 	tcp_timer_stop(tp, TT_2MSL);
1813 	tcp_timer_stop(tp, TT_DELACK);
1814 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1815 		/*
1816 		 * Call the stop-all function of the methods,
1817 		 * this function should call the tcp_timer_stop()
1818 		 * method with each of the function specific timeouts.
1819 		 * That stop will be called via the tfb_tcp_timer_stop()
1820 		 * which should use the async drain function of the
1821 		 * callout system (see tcp_var.h).
1822 		 */
1823 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1824 	}
1825 
1826 	/*
1827 	 * If we got enough samples through the srtt filter,
1828 	 * save the rtt and rttvar in the routing entry.
1829 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1830 	 * 4 samples is enough for the srtt filter to converge
1831 	 * to within enough % of the correct value; fewer samples
1832 	 * and we could save a bogus rtt. The danger is not high
1833 	 * as tcp quickly recovers from everything.
1834 	 * XXX: Works very well but needs some more statistics!
1835 	 */
1836 	if (tp->t_rttupdated >= 4) {
1837 		struct hc_metrics_lite metrics;
1838 		uint32_t ssthresh;
1839 
1840 		bzero(&metrics, sizeof(metrics));
1841 		/*
1842 		 * Update the ssthresh always when the conditions below
1843 		 * are satisfied. This gives us better new start value
1844 		 * for the congestion avoidance for new connections.
1845 		 * ssthresh is only set if packet loss occurred on a session.
1846 		 *
1847 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1848 		 * being torn down.  Ideally this code would not use 'so'.
1849 		 */
1850 		ssthresh = tp->snd_ssthresh;
1851 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1852 			/*
1853 			 * convert the limit from user data bytes to
1854 			 * packets then to packet data bytes.
1855 			 */
1856 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1857 			if (ssthresh < 2)
1858 				ssthresh = 2;
1859 			ssthresh *= (tp->t_maxseg +
1860 #ifdef INET6
1861 			    (isipv6 ? sizeof (struct ip6_hdr) +
1862 				sizeof (struct tcphdr) :
1863 #endif
1864 				sizeof (struct tcpiphdr)
1865 #ifdef INET6
1866 			    )
1867 #endif
1868 			    );
1869 		} else
1870 			ssthresh = 0;
1871 		metrics.rmx_ssthresh = ssthresh;
1872 
1873 		metrics.rmx_rtt = tp->t_srtt;
1874 		metrics.rmx_rttvar = tp->t_rttvar;
1875 		metrics.rmx_cwnd = tp->snd_cwnd;
1876 		metrics.rmx_sendpipe = 0;
1877 		metrics.rmx_recvpipe = 0;
1878 
1879 		tcp_hc_update(&inp->inp_inc, &metrics);
1880 	}
1881 
1882 	/* free the reassembly queue, if any */
1883 	tcp_reass_flush(tp);
1884 
1885 #ifdef TCP_OFFLOAD
1886 	/* Disconnect offload device, if any. */
1887 	if (tp->t_flags & TF_TOE)
1888 		tcp_offload_detach(tp);
1889 #endif
1890 
1891 	tcp_free_sackholes(tp);
1892 
1893 #ifdef TCPPCAP
1894 	/* Free the TCP PCAP queues. */
1895 	tcp_pcap_drain(&(tp->t_inpkts));
1896 	tcp_pcap_drain(&(tp->t_outpkts));
1897 #endif
1898 
1899 	/* Allow the CC algorithm to clean up after itself. */
1900 	if (CC_ALGO(tp)->cb_destroy != NULL)
1901 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1902 	CC_DATA(tp) = NULL;
1903 
1904 #ifdef TCP_HHOOK
1905 	khelp_destroy_osd(tp->osd);
1906 #endif
1907 
1908 	CC_ALGO(tp) = NULL;
1909 	inp->inp_ppcb = NULL;
1910 	if (tp->t_timers->tt_draincnt == 0) {
1911 		/* We own the last reference on tcpcb, let's free it. */
1912 #ifdef TCP_BLACKBOX
1913 		tcp_log_tcpcbfini(tp);
1914 #endif
1915 		TCPSTATES_DEC(tp->t_state);
1916 		if (tp->t_fb->tfb_tcp_fb_fini)
1917 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1918 		refcount_release(&tp->t_fb->tfb_refcnt);
1919 		tp->t_inpcb = NULL;
1920 		uma_zfree(V_tcpcb_zone, tp);
1921 		released = in_pcbrele_wlocked(inp);
1922 		KASSERT(!released, ("%s: inp %p should not have been released "
1923 			"here", __func__, inp));
1924 	}
1925 }
1926 
1927 void
1928 tcp_timer_discard(void *ptp)
1929 {
1930 	struct inpcb *inp;
1931 	struct tcpcb *tp;
1932 	struct epoch_tracker et;
1933 
1934 	tp = (struct tcpcb *)ptp;
1935 	CURVNET_SET(tp->t_vnet);
1936 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
1937 	inp = tp->t_inpcb;
1938 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1939 		__func__, tp));
1940 	INP_WLOCK(inp);
1941 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1942 		("%s: tcpcb has to be stopped here", __func__));
1943 	tp->t_timers->tt_draincnt--;
1944 	if (tp->t_timers->tt_draincnt == 0) {
1945 		/* We own the last reference on this tcpcb, let's free it. */
1946 #ifdef TCP_BLACKBOX
1947 		tcp_log_tcpcbfini(tp);
1948 #endif
1949 		TCPSTATES_DEC(tp->t_state);
1950 		if (tp->t_fb->tfb_tcp_fb_fini)
1951 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1952 		refcount_release(&tp->t_fb->tfb_refcnt);
1953 		tp->t_inpcb = NULL;
1954 		uma_zfree(V_tcpcb_zone, tp);
1955 		if (in_pcbrele_wlocked(inp)) {
1956 			INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1957 			CURVNET_RESTORE();
1958 			return;
1959 		}
1960 	}
1961 	INP_WUNLOCK(inp);
1962 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
1963 	CURVNET_RESTORE();
1964 }
1965 
1966 /*
1967  * Attempt to close a TCP control block, marking it as dropped, and freeing
1968  * the socket if we hold the only reference.
1969  */
1970 struct tcpcb *
1971 tcp_close(struct tcpcb *tp)
1972 {
1973 	struct inpcb *inp = tp->t_inpcb;
1974 	struct socket *so;
1975 
1976 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1977 	INP_WLOCK_ASSERT(inp);
1978 
1979 #ifdef TCP_OFFLOAD
1980 	if (tp->t_state == TCPS_LISTEN)
1981 		tcp_offload_listen_stop(tp);
1982 #endif
1983 	/*
1984 	 * This releases the TFO pending counter resource for TFO listen
1985 	 * sockets as well as passively-created TFO sockets that transition
1986 	 * from SYN_RECEIVED to CLOSED.
1987 	 */
1988 	if (tp->t_tfo_pending) {
1989 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1990 		tp->t_tfo_pending = NULL;
1991 	}
1992 	in_pcbdrop(inp);
1993 	TCPSTAT_INC(tcps_closed);
1994 	if (tp->t_state != TCPS_CLOSED)
1995 		tcp_state_change(tp, TCPS_CLOSED);
1996 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1997 	so = inp->inp_socket;
1998 	soisdisconnected(so);
1999 	if (inp->inp_flags & INP_SOCKREF) {
2000 		KASSERT(so->so_state & SS_PROTOREF,
2001 		    ("tcp_close: !SS_PROTOREF"));
2002 		inp->inp_flags &= ~INP_SOCKREF;
2003 		INP_WUNLOCK(inp);
2004 		SOCK_LOCK(so);
2005 		so->so_state &= ~SS_PROTOREF;
2006 		sofree(so);
2007 		return (NULL);
2008 	}
2009 	return (tp);
2010 }
2011 
2012 void
2013 tcp_drain(void)
2014 {
2015 	VNET_ITERATOR_DECL(vnet_iter);
2016 
2017 	if (!do_tcpdrain)
2018 		return;
2019 
2020 	VNET_LIST_RLOCK_NOSLEEP();
2021 	VNET_FOREACH(vnet_iter) {
2022 		CURVNET_SET(vnet_iter);
2023 		struct inpcb *inpb;
2024 		struct tcpcb *tcpb;
2025 
2026 	/*
2027 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
2028 	 * if there is one...
2029 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
2030 	 *      reassembly queue should be flushed, but in a situation
2031 	 *	where we're really low on mbufs, this is potentially
2032 	 *	useful.
2033 	 */
2034 		INP_INFO_WLOCK(&V_tcbinfo);
2035 		CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2036 			INP_WLOCK(inpb);
2037 			if (inpb->inp_flags & INP_TIMEWAIT) {
2038 				INP_WUNLOCK(inpb);
2039 				continue;
2040 			}
2041 			if ((tcpb = intotcpcb(inpb)) != NULL) {
2042 				tcp_reass_flush(tcpb);
2043 				tcp_clean_sackreport(tcpb);
2044 #ifdef TCP_BLACKBOX
2045 				tcp_log_drain(tcpb);
2046 #endif
2047 #ifdef TCPPCAP
2048 				if (tcp_pcap_aggressive_free) {
2049 					/* Free the TCP PCAP queues. */
2050 					tcp_pcap_drain(&(tcpb->t_inpkts));
2051 					tcp_pcap_drain(&(tcpb->t_outpkts));
2052 				}
2053 #endif
2054 			}
2055 			INP_WUNLOCK(inpb);
2056 		}
2057 		INP_INFO_WUNLOCK(&V_tcbinfo);
2058 		CURVNET_RESTORE();
2059 	}
2060 	VNET_LIST_RUNLOCK_NOSLEEP();
2061 }
2062 
2063 /*
2064  * Notify a tcp user of an asynchronous error;
2065  * store error as soft error, but wake up user
2066  * (for now, won't do anything until can select for soft error).
2067  *
2068  * Do not wake up user since there currently is no mechanism for
2069  * reporting soft errors (yet - a kqueue filter may be added).
2070  */
2071 static struct inpcb *
2072 tcp_notify(struct inpcb *inp, int error)
2073 {
2074 	struct tcpcb *tp;
2075 
2076 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2077 	INP_WLOCK_ASSERT(inp);
2078 
2079 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2080 	    (inp->inp_flags & INP_DROPPED))
2081 		return (inp);
2082 
2083 	tp = intotcpcb(inp);
2084 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2085 
2086 	/*
2087 	 * Ignore some errors if we are hooked up.
2088 	 * If connection hasn't completed, has retransmitted several times,
2089 	 * and receives a second error, give up now.  This is better
2090 	 * than waiting a long time to establish a connection that
2091 	 * can never complete.
2092 	 */
2093 	if (tp->t_state == TCPS_ESTABLISHED &&
2094 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
2095 	     error == EHOSTDOWN)) {
2096 		if (inp->inp_route.ro_rt) {
2097 			RTFREE(inp->inp_route.ro_rt);
2098 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
2099 		}
2100 		return (inp);
2101 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2102 	    tp->t_softerror) {
2103 		tp = tcp_drop(tp, error);
2104 		if (tp != NULL)
2105 			return (inp);
2106 		else
2107 			return (NULL);
2108 	} else {
2109 		tp->t_softerror = error;
2110 		return (inp);
2111 	}
2112 #if 0
2113 	wakeup( &so->so_timeo);
2114 	sorwakeup(so);
2115 	sowwakeup(so);
2116 #endif
2117 }
2118 
2119 static int
2120 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2121 {
2122 	int error, i, m, n, pcb_count;
2123 	struct inpcb *inp, **inp_list;
2124 	inp_gen_t gencnt;
2125 	struct xinpgen xig;
2126 	struct epoch_tracker et;
2127 
2128 	/*
2129 	 * The process of preparing the TCB list is too time-consuming and
2130 	 * resource-intensive to repeat twice on every request.
2131 	 */
2132 	if (req->oldptr == NULL) {
2133 		n = V_tcbinfo.ipi_count +
2134 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2135 		n += imax(n / 8, 10);
2136 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2137 		return (0);
2138 	}
2139 
2140 	if (req->newptr != NULL)
2141 		return (EPERM);
2142 
2143 	/*
2144 	 * OK, now we're committed to doing something.
2145 	 */
2146 	INP_LIST_RLOCK(&V_tcbinfo);
2147 	gencnt = V_tcbinfo.ipi_gencnt;
2148 	n = V_tcbinfo.ipi_count;
2149 	INP_LIST_RUNLOCK(&V_tcbinfo);
2150 
2151 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2152 
2153 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
2154 		+ (n + m) * sizeof(struct xtcpcb));
2155 	if (error != 0)
2156 		return (error);
2157 
2158 	bzero(&xig, sizeof(xig));
2159 	xig.xig_len = sizeof xig;
2160 	xig.xig_count = n + m;
2161 	xig.xig_gen = gencnt;
2162 	xig.xig_sogen = so_gencnt;
2163 	error = SYSCTL_OUT(req, &xig, sizeof xig);
2164 	if (error)
2165 		return (error);
2166 
2167 	error = syncache_pcblist(req, m, &pcb_count);
2168 	if (error)
2169 		return (error);
2170 
2171 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2172 
2173 	INP_INFO_WLOCK(&V_tcbinfo);
2174 	for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
2175 	    inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) {
2176 		INP_WLOCK(inp);
2177 		if (inp->inp_gencnt <= gencnt) {
2178 			/*
2179 			 * XXX: This use of cr_cansee(), introduced with
2180 			 * TCP state changes, is not quite right, but for
2181 			 * now, better than nothing.
2182 			 */
2183 			if (inp->inp_flags & INP_TIMEWAIT) {
2184 				if (intotw(inp) != NULL)
2185 					error = cr_cansee(req->td->td_ucred,
2186 					    intotw(inp)->tw_cred);
2187 				else
2188 					error = EINVAL;	/* Skip this inp. */
2189 			} else
2190 				error = cr_canseeinpcb(req->td->td_ucred, inp);
2191 			if (error == 0) {
2192 				in_pcbref(inp);
2193 				inp_list[i++] = inp;
2194 			}
2195 		}
2196 		INP_WUNLOCK(inp);
2197 	}
2198 	INP_INFO_WUNLOCK(&V_tcbinfo);
2199 	n = i;
2200 
2201 	error = 0;
2202 	for (i = 0; i < n; i++) {
2203 		inp = inp_list[i];
2204 		INP_RLOCK(inp);
2205 		if (inp->inp_gencnt <= gencnt) {
2206 			struct xtcpcb xt;
2207 
2208 			tcp_inptoxtp(inp, &xt);
2209 			INP_RUNLOCK(inp);
2210 			error = SYSCTL_OUT(req, &xt, sizeof xt);
2211 		} else
2212 			INP_RUNLOCK(inp);
2213 	}
2214 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2215 	for (i = 0; i < n; i++) {
2216 		inp = inp_list[i];
2217 		INP_RLOCK(inp);
2218 		if (!in_pcbrele_rlocked(inp))
2219 			INP_RUNLOCK(inp);
2220 	}
2221 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2222 
2223 	if (!error) {
2224 		/*
2225 		 * Give the user an updated idea of our state.
2226 		 * If the generation differs from what we told
2227 		 * her before, she knows that something happened
2228 		 * while we were processing this request, and it
2229 		 * might be necessary to retry.
2230 		 */
2231 		INP_LIST_RLOCK(&V_tcbinfo);
2232 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
2233 		xig.xig_sogen = so_gencnt;
2234 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
2235 		INP_LIST_RUNLOCK(&V_tcbinfo);
2236 		error = SYSCTL_OUT(req, &xig, sizeof xig);
2237 	}
2238 	free(inp_list, M_TEMP);
2239 	return (error);
2240 }
2241 
2242 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2243     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2244     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2245 
2246 #ifdef INET
2247 static int
2248 tcp_getcred(SYSCTL_HANDLER_ARGS)
2249 {
2250 	struct xucred xuc;
2251 	struct sockaddr_in addrs[2];
2252 	struct inpcb *inp;
2253 	int error;
2254 
2255 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2256 	if (error)
2257 		return (error);
2258 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2259 	if (error)
2260 		return (error);
2261 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2262 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2263 	if (inp != NULL) {
2264 		if (inp->inp_socket == NULL)
2265 			error = ENOENT;
2266 		if (error == 0)
2267 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2268 		if (error == 0)
2269 			cru2x(inp->inp_cred, &xuc);
2270 		INP_RUNLOCK(inp);
2271 	} else
2272 		error = ENOENT;
2273 	if (error == 0)
2274 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2275 	return (error);
2276 }
2277 
2278 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2279     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2280     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2281 #endif /* INET */
2282 
2283 #ifdef INET6
2284 static int
2285 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2286 {
2287 	struct xucred xuc;
2288 	struct sockaddr_in6 addrs[2];
2289 	struct inpcb *inp;
2290 	int error;
2291 #ifdef INET
2292 	int mapped = 0;
2293 #endif
2294 
2295 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2296 	if (error)
2297 		return (error);
2298 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2299 	if (error)
2300 		return (error);
2301 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2302 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2303 		return (error);
2304 	}
2305 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2306 #ifdef INET
2307 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2308 			mapped = 1;
2309 		else
2310 #endif
2311 			return (EINVAL);
2312 	}
2313 
2314 #ifdef INET
2315 	if (mapped == 1)
2316 		inp = in_pcblookup(&V_tcbinfo,
2317 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2318 			addrs[1].sin6_port,
2319 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2320 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2321 	else
2322 #endif
2323 		inp = in6_pcblookup(&V_tcbinfo,
2324 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2325 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2326 			INPLOOKUP_RLOCKPCB, NULL);
2327 	if (inp != NULL) {
2328 		if (inp->inp_socket == NULL)
2329 			error = ENOENT;
2330 		if (error == 0)
2331 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2332 		if (error == 0)
2333 			cru2x(inp->inp_cred, &xuc);
2334 		INP_RUNLOCK(inp);
2335 	} else
2336 		error = ENOENT;
2337 	if (error == 0)
2338 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2339 	return (error);
2340 }
2341 
2342 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2343     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2344     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2345 #endif /* INET6 */
2346 
2347 
2348 #ifdef INET
2349 void
2350 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2351 {
2352 	struct ip *ip = vip;
2353 	struct tcphdr *th;
2354 	struct in_addr faddr;
2355 	struct inpcb *inp;
2356 	struct tcpcb *tp;
2357 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2358 	struct icmp *icp;
2359 	struct in_conninfo inc;
2360 	struct epoch_tracker et;
2361 	tcp_seq icmp_tcp_seq;
2362 	int mtu;
2363 
2364 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2365 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2366 		return;
2367 
2368 	if (cmd == PRC_MSGSIZE)
2369 		notify = tcp_mtudisc_notify;
2370 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2371 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2372 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2373 		notify = tcp_drop_syn_sent;
2374 
2375 	/*
2376 	 * Hostdead is ugly because it goes linearly through all PCBs.
2377 	 * XXX: We never get this from ICMP, otherwise it makes an
2378 	 * excellent DoS attack on machines with many connections.
2379 	 */
2380 	else if (cmd == PRC_HOSTDEAD)
2381 		ip = NULL;
2382 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2383 		return;
2384 
2385 	if (ip == NULL) {
2386 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2387 		return;
2388 	}
2389 
2390 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2391 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2392 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2393 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2394 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2395 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2396 		/* signal EHOSTDOWN, as it flushes the cached route */
2397 		inp = (*notify)(inp, EHOSTDOWN);
2398 		goto out;
2399 	}
2400 	icmp_tcp_seq = th->th_seq;
2401 	if (inp != NULL)  {
2402 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2403 		    !(inp->inp_flags & INP_DROPPED) &&
2404 		    !(inp->inp_socket == NULL)) {
2405 			tp = intotcpcb(inp);
2406 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2407 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2408 				if (cmd == PRC_MSGSIZE) {
2409 					/*
2410 					 * MTU discovery:
2411 					 * If we got a needfrag set the MTU
2412 					 * in the route to the suggested new
2413 					 * value (if given) and then notify.
2414 					 */
2415 					mtu = ntohs(icp->icmp_nextmtu);
2416 					/*
2417 					 * If no alternative MTU was
2418 					 * proposed, try the next smaller
2419 					 * one.
2420 					 */
2421 					if (!mtu)
2422 						mtu = ip_next_mtu(
2423 						    ntohs(ip->ip_len), 1);
2424 					if (mtu < V_tcp_minmss +
2425 					    sizeof(struct tcpiphdr))
2426 						mtu = V_tcp_minmss +
2427 						    sizeof(struct tcpiphdr);
2428 					/*
2429 					 * Only process the offered MTU if it
2430 					 * is smaller than the current one.
2431 					 */
2432 					if (mtu < tp->t_maxseg +
2433 					    sizeof(struct tcpiphdr)) {
2434 						bzero(&inc, sizeof(inc));
2435 						inc.inc_faddr = faddr;
2436 						inc.inc_fibnum =
2437 						    inp->inp_inc.inc_fibnum;
2438 						tcp_hc_updatemtu(&inc, mtu);
2439 						tcp_mtudisc(inp, mtu);
2440 					}
2441 				} else
2442 					inp = (*notify)(inp,
2443 					    inetctlerrmap[cmd]);
2444 			}
2445 		}
2446 	} else {
2447 		bzero(&inc, sizeof(inc));
2448 		inc.inc_fport = th->th_dport;
2449 		inc.inc_lport = th->th_sport;
2450 		inc.inc_faddr = faddr;
2451 		inc.inc_laddr = ip->ip_src;
2452 		syncache_unreach(&inc, icmp_tcp_seq);
2453 	}
2454 out:
2455 	if (inp != NULL)
2456 		INP_WUNLOCK(inp);
2457 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2458 }
2459 #endif /* INET */
2460 
2461 #ifdef INET6
2462 void
2463 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2464 {
2465 	struct in6_addr *dst;
2466 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2467 	struct ip6_hdr *ip6;
2468 	struct mbuf *m;
2469 	struct inpcb *inp;
2470 	struct tcpcb *tp;
2471 	struct icmp6_hdr *icmp6;
2472 	struct ip6ctlparam *ip6cp = NULL;
2473 	const struct sockaddr_in6 *sa6_src = NULL;
2474 	struct in_conninfo inc;
2475 	struct epoch_tracker et;
2476 	struct tcp_ports {
2477 		uint16_t th_sport;
2478 		uint16_t th_dport;
2479 	} t_ports;
2480 	tcp_seq icmp_tcp_seq;
2481 	unsigned int mtu;
2482 	unsigned int off;
2483 
2484 	if (sa->sa_family != AF_INET6 ||
2485 	    sa->sa_len != sizeof(struct sockaddr_in6))
2486 		return;
2487 
2488 	/* if the parameter is from icmp6, decode it. */
2489 	if (d != NULL) {
2490 		ip6cp = (struct ip6ctlparam *)d;
2491 		icmp6 = ip6cp->ip6c_icmp6;
2492 		m = ip6cp->ip6c_m;
2493 		ip6 = ip6cp->ip6c_ip6;
2494 		off = ip6cp->ip6c_off;
2495 		sa6_src = ip6cp->ip6c_src;
2496 		dst = ip6cp->ip6c_finaldst;
2497 	} else {
2498 		m = NULL;
2499 		ip6 = NULL;
2500 		off = 0;	/* fool gcc */
2501 		sa6_src = &sa6_any;
2502 		dst = NULL;
2503 	}
2504 
2505 	if (cmd == PRC_MSGSIZE)
2506 		notify = tcp_mtudisc_notify;
2507 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2508 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2509 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2510 		notify = tcp_drop_syn_sent;
2511 
2512 	/*
2513 	 * Hostdead is ugly because it goes linearly through all PCBs.
2514 	 * XXX: We never get this from ICMP, otherwise it makes an
2515 	 * excellent DoS attack on machines with many connections.
2516 	 */
2517 	else if (cmd == PRC_HOSTDEAD)
2518 		ip6 = NULL;
2519 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2520 		return;
2521 
2522 	if (ip6 == NULL) {
2523 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2524 			      (const struct sockaddr *)sa6_src,
2525 			      0, cmd, NULL, notify);
2526 		return;
2527 	}
2528 
2529 	/* Check if we can safely get the ports from the tcp hdr */
2530 	if (m == NULL ||
2531 	    (m->m_pkthdr.len <
2532 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2533 		return;
2534 	}
2535 	bzero(&t_ports, sizeof(struct tcp_ports));
2536 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2537 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
2538 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2539 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2540 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2541 		/* signal EHOSTDOWN, as it flushes the cached route */
2542 		inp = (*notify)(inp, EHOSTDOWN);
2543 		goto out;
2544 	}
2545 	off += sizeof(struct tcp_ports);
2546 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2547 		goto out;
2548 	}
2549 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2550 	if (inp != NULL)  {
2551 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2552 		    !(inp->inp_flags & INP_DROPPED) &&
2553 		    !(inp->inp_socket == NULL)) {
2554 			tp = intotcpcb(inp);
2555 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2556 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2557 				if (cmd == PRC_MSGSIZE) {
2558 					/*
2559 					 * MTU discovery:
2560 					 * If we got a needfrag set the MTU
2561 					 * in the route to the suggested new
2562 					 * value (if given) and then notify.
2563 					 */
2564 					mtu = ntohl(icmp6->icmp6_mtu);
2565 					/*
2566 					 * If no alternative MTU was
2567 					 * proposed, or the proposed
2568 					 * MTU was too small, set to
2569 					 * the min.
2570 					 */
2571 					if (mtu < IPV6_MMTU)
2572 						mtu = IPV6_MMTU - 8;
2573 					bzero(&inc, sizeof(inc));
2574 					inc.inc_fibnum = M_GETFIB(m);
2575 					inc.inc_flags |= INC_ISIPV6;
2576 					inc.inc6_faddr = *dst;
2577 					if (in6_setscope(&inc.inc6_faddr,
2578 						m->m_pkthdr.rcvif, NULL))
2579 						goto out;
2580 					/*
2581 					 * Only process the offered MTU if it
2582 					 * is smaller than the current one.
2583 					 */
2584 					if (mtu < tp->t_maxseg +
2585 					    sizeof (struct tcphdr) +
2586 					    sizeof (struct ip6_hdr)) {
2587 						tcp_hc_updatemtu(&inc, mtu);
2588 						tcp_mtudisc(inp, mtu);
2589 						ICMP6STAT_INC(icp6s_pmtuchg);
2590 					}
2591 				} else
2592 					inp = (*notify)(inp,
2593 					    inet6ctlerrmap[cmd]);
2594 			}
2595 		}
2596 	} else {
2597 		bzero(&inc, sizeof(inc));
2598 		inc.inc_fibnum = M_GETFIB(m);
2599 		inc.inc_flags |= INC_ISIPV6;
2600 		inc.inc_fport = t_ports.th_dport;
2601 		inc.inc_lport = t_ports.th_sport;
2602 		inc.inc6_faddr = *dst;
2603 		inc.inc6_laddr = ip6->ip6_src;
2604 		syncache_unreach(&inc, icmp_tcp_seq);
2605 	}
2606 out:
2607 	if (inp != NULL)
2608 		INP_WUNLOCK(inp);
2609 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
2610 }
2611 #endif /* INET6 */
2612 
2613 static uint32_t
2614 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
2615 {
2616 	MD5_CTX ctx;
2617 	uint32_t hash[4];
2618 
2619 	MD5Init(&ctx);
2620 	MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2621 	MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2622 	switch (inc->inc_flags & INC_ISIPV6) {
2623 #ifdef INET
2624 	case 0:
2625 		MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2626 		MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2627 		break;
2628 #endif
2629 #ifdef INET6
2630 	case INC_ISIPV6:
2631 		MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2632 		MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2633 		break;
2634 #endif
2635 	}
2636 	MD5Update(&ctx, key, len);
2637 	MD5Final((unsigned char *)hash, &ctx);
2638 
2639 	return (hash[0]);
2640 }
2641 
2642 uint32_t
2643 tcp_new_ts_offset(struct in_conninfo *inc)
2644 {
2645 	return (tcp_keyed_hash(inc, V_ts_offset_secret,
2646 	    sizeof(V_ts_offset_secret)));
2647 }
2648 
2649 /*
2650  * Following is where TCP initial sequence number generation occurs.
2651  *
2652  * There are two places where we must use initial sequence numbers:
2653  * 1.  In SYN-ACK packets.
2654  * 2.  In SYN packets.
2655  *
2656  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2657  * tcp_syncache.c for details.
2658  *
2659  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2660  * depends on this property.  In addition, these ISNs should be
2661  * unguessable so as to prevent connection hijacking.  To satisfy
2662  * the requirements of this situation, the algorithm outlined in
2663  * RFC 1948 is used, with only small modifications.
2664  *
2665  * Implementation details:
2666  *
2667  * Time is based off the system timer, and is corrected so that it
2668  * increases by one megabyte per second.  This allows for proper
2669  * recycling on high speed LANs while still leaving over an hour
2670  * before rollover.
2671  *
2672  * As reading the *exact* system time is too expensive to be done
2673  * whenever setting up a TCP connection, we increment the time
2674  * offset in two ways.  First, a small random positive increment
2675  * is added to isn_offset for each connection that is set up.
2676  * Second, the function tcp_isn_tick fires once per clock tick
2677  * and increments isn_offset as necessary so that sequence numbers
2678  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2679  * random positive increments serve only to ensure that the same
2680  * exact sequence number is never sent out twice (as could otherwise
2681  * happen when a port is recycled in less than the system tick
2682  * interval.)
2683  *
2684  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2685  * between seeding of isn_secret.  This is normally set to zero,
2686  * as reseeding should not be necessary.
2687  *
2688  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2689  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2690  * general, this means holding an exclusive (write) lock.
2691  */
2692 
2693 #define ISN_BYTES_PER_SECOND 1048576
2694 #define ISN_STATIC_INCREMENT 4096
2695 #define ISN_RANDOM_INCREMENT (4096 - 1)
2696 #define ISN_SECRET_LENGTH    32
2697 
2698 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
2699 VNET_DEFINE_STATIC(int, isn_last);
2700 VNET_DEFINE_STATIC(int, isn_last_reseed);
2701 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
2702 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
2703 
2704 #define	V_isn_secret			VNET(isn_secret)
2705 #define	V_isn_last			VNET(isn_last)
2706 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2707 #define	V_isn_offset			VNET(isn_offset)
2708 #define	V_isn_offset_old		VNET(isn_offset_old)
2709 
2710 tcp_seq
2711 tcp_new_isn(struct in_conninfo *inc)
2712 {
2713 	tcp_seq new_isn;
2714 	u_int32_t projected_offset;
2715 
2716 	ISN_LOCK();
2717 	/* Seed if this is the first use, reseed if requested. */
2718 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2719 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2720 		< (u_int)ticks))) {
2721 		arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
2722 		V_isn_last_reseed = ticks;
2723 	}
2724 
2725 	/* Compute the md5 hash and return the ISN. */
2726 	new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
2727 	    sizeof(V_isn_secret));
2728 	V_isn_offset += ISN_STATIC_INCREMENT +
2729 		(arc4random() & ISN_RANDOM_INCREMENT);
2730 	if (ticks != V_isn_last) {
2731 		projected_offset = V_isn_offset_old +
2732 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2733 		if (SEQ_GT(projected_offset, V_isn_offset))
2734 			V_isn_offset = projected_offset;
2735 		V_isn_offset_old = V_isn_offset;
2736 		V_isn_last = ticks;
2737 	}
2738 	new_isn += V_isn_offset;
2739 	ISN_UNLOCK();
2740 	return (new_isn);
2741 }
2742 
2743 /*
2744  * When a specific ICMP unreachable message is received and the
2745  * connection state is SYN-SENT, drop the connection.  This behavior
2746  * is controlled by the icmp_may_rst sysctl.
2747  */
2748 struct inpcb *
2749 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2750 {
2751 	struct tcpcb *tp;
2752 
2753 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2754 	INP_WLOCK_ASSERT(inp);
2755 
2756 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2757 	    (inp->inp_flags & INP_DROPPED))
2758 		return (inp);
2759 
2760 	tp = intotcpcb(inp);
2761 	if (tp->t_state != TCPS_SYN_SENT)
2762 		return (inp);
2763 
2764 	if (IS_FASTOPEN(tp->t_flags))
2765 		tcp_fastopen_disable_path(tp);
2766 
2767 	tp = tcp_drop(tp, errno);
2768 	if (tp != NULL)
2769 		return (inp);
2770 	else
2771 		return (NULL);
2772 }
2773 
2774 /*
2775  * When `need fragmentation' ICMP is received, update our idea of the MSS
2776  * based on the new value. Also nudge TCP to send something, since we
2777  * know the packet we just sent was dropped.
2778  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2779  */
2780 static struct inpcb *
2781 tcp_mtudisc_notify(struct inpcb *inp, int error)
2782 {
2783 
2784 	tcp_mtudisc(inp, -1);
2785 	return (inp);
2786 }
2787 
2788 static void
2789 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2790 {
2791 	struct tcpcb *tp;
2792 	struct socket *so;
2793 
2794 	INP_WLOCK_ASSERT(inp);
2795 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2796 	    (inp->inp_flags & INP_DROPPED))
2797 		return;
2798 
2799 	tp = intotcpcb(inp);
2800 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2801 
2802 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2803 
2804 	so = inp->inp_socket;
2805 	SOCKBUF_LOCK(&so->so_snd);
2806 	/* If the mss is larger than the socket buffer, decrease the mss. */
2807 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2808 		tp->t_maxseg = so->so_snd.sb_hiwat;
2809 	SOCKBUF_UNLOCK(&so->so_snd);
2810 
2811 	TCPSTAT_INC(tcps_mturesent);
2812 	tp->t_rtttime = 0;
2813 	tp->snd_nxt = tp->snd_una;
2814 	tcp_free_sackholes(tp);
2815 	tp->snd_recover = tp->snd_max;
2816 	if (tp->t_flags & TF_SACK_PERMIT)
2817 		EXIT_FASTRECOVERY(tp->t_flags);
2818 	tp->t_fb->tfb_tcp_output(tp);
2819 }
2820 
2821 #ifdef INET
2822 /*
2823  * Look-up the routing entry to the peer of this inpcb.  If no route
2824  * is found and it cannot be allocated, then return 0.  This routine
2825  * is called by TCP routines that access the rmx structure and by
2826  * tcp_mss_update to get the peer/interface MTU.
2827  */
2828 uint32_t
2829 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2830 {
2831 	struct nhop4_extended nh4;
2832 	struct ifnet *ifp;
2833 	uint32_t maxmtu = 0;
2834 
2835 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2836 
2837 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2838 
2839 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2840 		    NHR_REF, 0, &nh4) != 0)
2841 			return (0);
2842 
2843 		ifp = nh4.nh_ifp;
2844 		maxmtu = nh4.nh_mtu;
2845 
2846 		/* Report additional interface capabilities. */
2847 		if (cap != NULL) {
2848 			if (ifp->if_capenable & IFCAP_TSO4 &&
2849 			    ifp->if_hwassist & CSUM_TSO) {
2850 				cap->ifcap |= CSUM_TSO;
2851 				cap->tsomax = ifp->if_hw_tsomax;
2852 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2853 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2854 			}
2855 		}
2856 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2857 	}
2858 	return (maxmtu);
2859 }
2860 #endif /* INET */
2861 
2862 #ifdef INET6
2863 uint32_t
2864 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2865 {
2866 	struct nhop6_extended nh6;
2867 	struct in6_addr dst6;
2868 	uint32_t scopeid;
2869 	struct ifnet *ifp;
2870 	uint32_t maxmtu = 0;
2871 
2872 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2873 
2874 	if (inc->inc_flags & INC_IPV6MINMTU)
2875 		return (IPV6_MMTU);
2876 
2877 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2878 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2879 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2880 		    0, &nh6) != 0)
2881 			return (0);
2882 
2883 		ifp = nh6.nh_ifp;
2884 		maxmtu = nh6.nh_mtu;
2885 
2886 		/* Report additional interface capabilities. */
2887 		if (cap != NULL) {
2888 			if (ifp->if_capenable & IFCAP_TSO6 &&
2889 			    ifp->if_hwassist & CSUM_TSO) {
2890 				cap->ifcap |= CSUM_TSO;
2891 				cap->tsomax = ifp->if_hw_tsomax;
2892 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2893 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2894 			}
2895 		}
2896 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2897 	}
2898 
2899 	return (maxmtu);
2900 }
2901 #endif /* INET6 */
2902 
2903 /*
2904  * Calculate effective SMSS per RFC5681 definition for a given TCP
2905  * connection at its current state, taking into account SACK and etc.
2906  */
2907 u_int
2908 tcp_maxseg(const struct tcpcb *tp)
2909 {
2910 	u_int optlen;
2911 
2912 	if (tp->t_flags & TF_NOOPT)
2913 		return (tp->t_maxseg);
2914 
2915 	/*
2916 	 * Here we have a simplified code from tcp_addoptions(),
2917 	 * without a proper loop, and having most of paddings hardcoded.
2918 	 * We might make mistakes with padding here in some edge cases,
2919 	 * but this is harmless, since result of tcp_maxseg() is used
2920 	 * only in cwnd and ssthresh estimations.
2921 	 */
2922 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2923 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2924 		if (tp->t_flags & TF_RCVD_TSTMP)
2925 			optlen = TCPOLEN_TSTAMP_APPA;
2926 		else
2927 			optlen = 0;
2928 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2929 		if (tp->t_flags & TF_SIGNATURE)
2930 			optlen += PAD(TCPOLEN_SIGNATURE);
2931 #endif
2932 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2933 			optlen += TCPOLEN_SACKHDR;
2934 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2935 			optlen = PAD(optlen);
2936 		}
2937 	} else {
2938 		if (tp->t_flags & TF_REQ_TSTMP)
2939 			optlen = TCPOLEN_TSTAMP_APPA;
2940 		else
2941 			optlen = PAD(TCPOLEN_MAXSEG);
2942 		if (tp->t_flags & TF_REQ_SCALE)
2943 			optlen += PAD(TCPOLEN_WINDOW);
2944 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2945 		if (tp->t_flags & TF_SIGNATURE)
2946 			optlen += PAD(TCPOLEN_SIGNATURE);
2947 #endif
2948 		if (tp->t_flags & TF_SACK_PERMIT)
2949 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2950 	}
2951 #undef PAD
2952 	optlen = min(optlen, TCP_MAXOLEN);
2953 	return (tp->t_maxseg - optlen);
2954 }
2955 
2956 static int
2957 sysctl_drop(SYSCTL_HANDLER_ARGS)
2958 {
2959 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2960 	struct sockaddr_storage addrs[2];
2961 	struct inpcb *inp;
2962 	struct tcpcb *tp;
2963 	struct tcptw *tw;
2964 	struct sockaddr_in *fin, *lin;
2965 	struct epoch_tracker et;
2966 #ifdef INET6
2967 	struct sockaddr_in6 *fin6, *lin6;
2968 #endif
2969 	int error;
2970 
2971 	inp = NULL;
2972 	fin = lin = NULL;
2973 #ifdef INET6
2974 	fin6 = lin6 = NULL;
2975 #endif
2976 	error = 0;
2977 
2978 	if (req->oldptr != NULL || req->oldlen != 0)
2979 		return (EINVAL);
2980 	if (req->newptr == NULL)
2981 		return (EPERM);
2982 	if (req->newlen < sizeof(addrs))
2983 		return (ENOMEM);
2984 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2985 	if (error)
2986 		return (error);
2987 
2988 	switch (addrs[0].ss_family) {
2989 #ifdef INET6
2990 	case AF_INET6:
2991 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2992 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2993 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2994 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2995 			return (EINVAL);
2996 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2997 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2998 				return (EINVAL);
2999 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3000 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3001 			fin = (struct sockaddr_in *)&addrs[0];
3002 			lin = (struct sockaddr_in *)&addrs[1];
3003 			break;
3004 		}
3005 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
3006 		if (error)
3007 			return (error);
3008 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
3009 		if (error)
3010 			return (error);
3011 		break;
3012 #endif
3013 #ifdef INET
3014 	case AF_INET:
3015 		fin = (struct sockaddr_in *)&addrs[0];
3016 		lin = (struct sockaddr_in *)&addrs[1];
3017 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
3018 		    lin->sin_len != sizeof(struct sockaddr_in))
3019 			return (EINVAL);
3020 		break;
3021 #endif
3022 	default:
3023 		return (EINVAL);
3024 	}
3025 	INP_INFO_RLOCK_ET(&V_tcbinfo, et);
3026 	switch (addrs[0].ss_family) {
3027 #ifdef INET6
3028 	case AF_INET6:
3029 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3030 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3031 		    INPLOOKUP_WLOCKPCB, NULL);
3032 		break;
3033 #endif
3034 #ifdef INET
3035 	case AF_INET:
3036 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3037 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3038 		break;
3039 #endif
3040 	}
3041 	if (inp != NULL) {
3042 		if (inp->inp_flags & INP_TIMEWAIT) {
3043 			/*
3044 			 * XXXRW: There currently exists a state where an
3045 			 * inpcb is present, but its timewait state has been
3046 			 * discarded.  For now, don't allow dropping of this
3047 			 * type of inpcb.
3048 			 */
3049 			tw = intotw(inp);
3050 			if (tw != NULL)
3051 				tcp_twclose(tw, 0);
3052 			else
3053 				INP_WUNLOCK(inp);
3054 		} else if (!(inp->inp_flags & INP_DROPPED) &&
3055 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3056 			tp = intotcpcb(inp);
3057 			tp = tcp_drop(tp, ECONNABORTED);
3058 			if (tp != NULL)
3059 				INP_WUNLOCK(inp);
3060 		} else
3061 			INP_WUNLOCK(inp);
3062 	} else
3063 		error = ESRCH;
3064 	INP_INFO_RUNLOCK_ET(&V_tcbinfo, et);
3065 	return (error);
3066 }
3067 
3068 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3069     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
3070     0, sysctl_drop, "", "Drop TCP connection");
3071 
3072 /*
3073  * Generate a standardized TCP log line for use throughout the
3074  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3075  * allow use in the interrupt context.
3076  *
3077  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3078  * NB: The function may return NULL if memory allocation failed.
3079  *
3080  * Due to header inclusion and ordering limitations the struct ip
3081  * and ip6_hdr pointers have to be passed as void pointers.
3082  */
3083 char *
3084 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3085     const void *ip6hdr)
3086 {
3087 
3088 	/* Is logging enabled? */
3089 	if (tcp_log_in_vain == 0)
3090 		return (NULL);
3091 
3092 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3093 }
3094 
3095 char *
3096 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3097     const void *ip6hdr)
3098 {
3099 
3100 	/* Is logging enabled? */
3101 	if (tcp_log_debug == 0)
3102 		return (NULL);
3103 
3104 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3105 }
3106 
3107 static char *
3108 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3109     const void *ip6hdr)
3110 {
3111 	char *s, *sp;
3112 	size_t size;
3113 	struct ip *ip;
3114 #ifdef INET6
3115 	const struct ip6_hdr *ip6;
3116 
3117 	ip6 = (const struct ip6_hdr *)ip6hdr;
3118 #endif /* INET6 */
3119 	ip = (struct ip *)ip4hdr;
3120 
3121 	/*
3122 	 * The log line looks like this:
3123 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3124 	 */
3125 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3126 	    sizeof(PRINT_TH_FLAGS) + 1 +
3127 #ifdef INET6
3128 	    2 * INET6_ADDRSTRLEN;
3129 #else
3130 	    2 * INET_ADDRSTRLEN;
3131 #endif /* INET6 */
3132 
3133 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3134 	if (s == NULL)
3135 		return (NULL);
3136 
3137 	strcat(s, "TCP: [");
3138 	sp = s + strlen(s);
3139 
3140 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3141 		inet_ntoa_r(inc->inc_faddr, sp);
3142 		sp = s + strlen(s);
3143 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3144 		sp = s + strlen(s);
3145 		inet_ntoa_r(inc->inc_laddr, sp);
3146 		sp = s + strlen(s);
3147 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3148 #ifdef INET6
3149 	} else if (inc) {
3150 		ip6_sprintf(sp, &inc->inc6_faddr);
3151 		sp = s + strlen(s);
3152 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3153 		sp = s + strlen(s);
3154 		ip6_sprintf(sp, &inc->inc6_laddr);
3155 		sp = s + strlen(s);
3156 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3157 	} else if (ip6 && th) {
3158 		ip6_sprintf(sp, &ip6->ip6_src);
3159 		sp = s + strlen(s);
3160 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3161 		sp = s + strlen(s);
3162 		ip6_sprintf(sp, &ip6->ip6_dst);
3163 		sp = s + strlen(s);
3164 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3165 #endif /* INET6 */
3166 #ifdef INET
3167 	} else if (ip && th) {
3168 		inet_ntoa_r(ip->ip_src, sp);
3169 		sp = s + strlen(s);
3170 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3171 		sp = s + strlen(s);
3172 		inet_ntoa_r(ip->ip_dst, sp);
3173 		sp = s + strlen(s);
3174 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3175 #endif /* INET */
3176 	} else {
3177 		free(s, M_TCPLOG);
3178 		return (NULL);
3179 	}
3180 	sp = s + strlen(s);
3181 	if (th)
3182 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3183 	if (*(s + size - 1) != '\0')
3184 		panic("%s: string too long", __func__);
3185 	return (s);
3186 }
3187 
3188 /*
3189  * A subroutine which makes it easy to track TCP state changes with DTrace.
3190  * This function shouldn't be called for t_state initializations that don't
3191  * correspond to actual TCP state transitions.
3192  */
3193 void
3194 tcp_state_change(struct tcpcb *tp, int newstate)
3195 {
3196 #if defined(KDTRACE_HOOKS)
3197 	int pstate = tp->t_state;
3198 #endif
3199 
3200 	TCPSTATES_DEC(tp->t_state);
3201 	TCPSTATES_INC(newstate);
3202 	tp->t_state = newstate;
3203 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3204 }
3205 
3206 /*
3207  * Create an external-format (``xtcpcb'') structure using the information in
3208  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3209  * reduce the spew of irrelevant information over this interface, to isolate
3210  * user code from changes in the kernel structure, and potentially to provide
3211  * information-hiding if we decide that some of this information should be
3212  * hidden from users.
3213  */
3214 void
3215 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3216 {
3217 	struct tcpcb *tp = intotcpcb(inp);
3218 	sbintime_t now;
3219 
3220 	bzero(xt, sizeof(*xt));
3221 	if (inp->inp_flags & INP_TIMEWAIT) {
3222 		xt->t_state = TCPS_TIME_WAIT;
3223 	} else {
3224 		xt->t_state = tp->t_state;
3225 		xt->t_logstate = tp->t_logstate;
3226 		xt->t_flags = tp->t_flags;
3227 		xt->t_sndzerowin = tp->t_sndzerowin;
3228 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3229 		xt->t_rcvoopack = tp->t_rcvoopack;
3230 
3231 		now = getsbinuptime();
3232 #define	COPYTIMER(ttt)	do {						\
3233 		if (callout_active(&tp->t_timers->ttt))			\
3234 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
3235 			    SBT_1MS;					\
3236 		else							\
3237 			xt->ttt = 0;					\
3238 } while (0)
3239 		COPYTIMER(tt_delack);
3240 		COPYTIMER(tt_rexmt);
3241 		COPYTIMER(tt_persist);
3242 		COPYTIMER(tt_keep);
3243 		COPYTIMER(tt_2msl);
3244 #undef COPYTIMER
3245 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
3246 
3247 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
3248 		    TCP_FUNCTION_NAME_LEN_MAX);
3249 #ifdef TCP_BLACKBOX
3250 		(void)tcp_log_get_id(tp, xt->xt_logid);
3251 #endif
3252 	}
3253 
3254 	xt->xt_len = sizeof(struct xtcpcb);
3255 	in_pcbtoxinpcb(inp, &xt->xt_inp);
3256 	if (inp->inp_socket == NULL)
3257 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
3258 }
3259