xref: /freebsd/sys/netinet/tcp_subr.c (revision 0629b152762b06325dd75a41bcb0a2789514141b)
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/cc/cc.h>
103 #ifdef INET6
104 #include <netinet6/tcp6_var.h>
105 #endif
106 #include <netinet/tcpip.h>
107 #include <netinet/tcp_fastopen.h>
108 #ifdef TCPPCAP
109 #include <netinet/tcp_pcap.h>
110 #endif
111 #ifdef TCPDEBUG
112 #include <netinet/tcp_debug.h>
113 #endif
114 #ifdef INET6
115 #include <netinet6/ip6protosw.h>
116 #endif
117 #ifdef TCP_OFFLOAD
118 #include <netinet/tcp_offload.h>
119 #endif
120 
121 #include <netipsec/ipsec_support.h>
122 
123 #include <machine/in_cksum.h>
124 #include <sys/md5.h>
125 
126 #include <security/mac/mac_framework.h>
127 
128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
129 #ifdef INET6
130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
131 #endif
132 
133 struct rwlock tcp_function_lock;
134 
135 static int
136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
137 {
138 	int error, new;
139 
140 	new = V_tcp_mssdflt;
141 	error = sysctl_handle_int(oidp, &new, 0, req);
142 	if (error == 0 && req->newptr) {
143 		if (new < TCP_MINMSS)
144 			error = EINVAL;
145 		else
146 			V_tcp_mssdflt = new;
147 	}
148 	return (error);
149 }
150 
151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
152     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
153     &sysctl_net_inet_tcp_mss_check, "I",
154     "Default TCP Maximum Segment Size");
155 
156 #ifdef INET6
157 static int
158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
159 {
160 	int error, new;
161 
162 	new = V_tcp_v6mssdflt;
163 	error = sysctl_handle_int(oidp, &new, 0, req);
164 	if (error == 0 && req->newptr) {
165 		if (new < TCP_MINMSS)
166 			error = EINVAL;
167 		else
168 			V_tcp_v6mssdflt = new;
169 	}
170 	return (error);
171 }
172 
173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
174     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
175     &sysctl_net_inet_tcp_mss_v6_check, "I",
176    "Default TCP Maximum Segment Size for IPv6");
177 #endif /* INET6 */
178 
179 /*
180  * Minimum MSS we accept and use. This prevents DoS attacks where
181  * we are forced to a ridiculous low MSS like 20 and send hundreds
182  * of packets instead of one. The effect scales with the available
183  * bandwidth and quickly saturates the CPU and network interface
184  * with packet generation and sending. Set to zero to disable MINMSS
185  * checking. This setting prevents us from sending too small packets.
186  */
187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
189      &VNET_NAME(tcp_minmss), 0,
190     "Minimum TCP Maximum Segment Size");
191 
192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
194     &VNET_NAME(tcp_do_rfc1323), 0,
195     "Enable rfc1323 (high performance TCP) extensions");
196 
197 static int	tcp_log_debug = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
199     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
200 
201 static int	tcp_tcbhashsize;
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
203     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
204 
205 static int	do_tcpdrain = 1;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
207     "Enable tcp_drain routine for extra help when low on mbufs");
208 
209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
210     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
211 
212 static VNET_DEFINE(int, icmp_may_rst) = 1;
213 #define	V_icmp_may_rst			VNET(icmp_may_rst)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
215     &VNET_NAME(icmp_may_rst), 0,
216     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
217 
218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
219 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
221     &VNET_NAME(tcp_isn_reseed_interval), 0,
222     "Seconds between reseeding of ISN secret");
223 
224 static int	tcp_soreceive_stream;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
226     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
227 
228 VNET_DEFINE(uma_zone_t, sack_hole_zone);
229 #define	V_sack_hole_zone		VNET(sack_hole_zone)
230 
231 #ifdef TCP_HHOOK
232 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
233 #endif
234 
235 static struct inpcb *tcp_notify(struct inpcb *, int);
236 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
237 static void tcp_mtudisc(struct inpcb *, int);
238 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
239 		    void *ip4hdr, const void *ip6hdr);
240 
241 
242 static struct tcp_function_block tcp_def_funcblk = {
243 	"default",
244 	tcp_output,
245 	tcp_do_segment,
246 	tcp_default_ctloutput,
247 	NULL,
248 	NULL,
249 	NULL,
250 	NULL,
251 	NULL,
252 	NULL,
253 	0,
254 	0
255 };
256 
257 int t_functions_inited = 0;
258 static int tcp_fb_cnt = 0;
259 struct tcp_funchead t_functions;
260 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
261 
262 static void
263 init_tcp_functions(void)
264 {
265 	if (t_functions_inited == 0) {
266 		TAILQ_INIT(&t_functions);
267 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
268 		t_functions_inited = 1;
269 	}
270 }
271 
272 static struct tcp_function_block *
273 find_tcp_functions_locked(struct tcp_function_set *fs)
274 {
275 	struct tcp_function *f;
276 	struct tcp_function_block *blk=NULL;
277 
278 	TAILQ_FOREACH(f, &t_functions, tf_next) {
279 		if (strcmp(f->tf_name, fs->function_set_name) == 0) {
280 			blk = f->tf_fb;
281 			break;
282 		}
283 	}
284 	return(blk);
285 }
286 
287 static struct tcp_function_block *
288 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
289 {
290 	struct tcp_function_block *rblk=NULL;
291 	struct tcp_function *f;
292 
293 	TAILQ_FOREACH(f, &t_functions, tf_next) {
294 		if (f->tf_fb == blk) {
295 			rblk = blk;
296 			if (s) {
297 				*s = f;
298 			}
299 			break;
300 		}
301 	}
302 	return (rblk);
303 }
304 
305 struct tcp_function_block *
306 find_and_ref_tcp_functions(struct tcp_function_set *fs)
307 {
308 	struct tcp_function_block *blk;
309 
310 	rw_rlock(&tcp_function_lock);
311 	blk = find_tcp_functions_locked(fs);
312 	if (blk)
313 		refcount_acquire(&blk->tfb_refcnt);
314 	rw_runlock(&tcp_function_lock);
315 	return(blk);
316 }
317 
318 struct tcp_function_block *
319 find_and_ref_tcp_fb(struct tcp_function_block *blk)
320 {
321 	struct tcp_function_block *rblk;
322 
323 	rw_rlock(&tcp_function_lock);
324 	rblk = find_tcp_fb_locked(blk, NULL);
325 	if (rblk)
326 		refcount_acquire(&rblk->tfb_refcnt);
327 	rw_runlock(&tcp_function_lock);
328 	return(rblk);
329 }
330 
331 
332 static int
333 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
334 {
335 	int error=ENOENT;
336 	struct tcp_function_set fs;
337 	struct tcp_function_block *blk;
338 
339 	memset(&fs, 0, sizeof(fs));
340 	rw_rlock(&tcp_function_lock);
341 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
342 	if (blk) {
343 		/* Found him */
344 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
345 		fs.pcbcnt = blk->tfb_refcnt;
346 	}
347 	rw_runlock(&tcp_function_lock);
348 	error = sysctl_handle_string(oidp, fs.function_set_name,
349 				     sizeof(fs.function_set_name), req);
350 
351 	/* Check for error or no change */
352 	if (error != 0 || req->newptr == NULL)
353 		return(error);
354 
355 	rw_wlock(&tcp_function_lock);
356 	blk = find_tcp_functions_locked(&fs);
357 	if ((blk == NULL) ||
358 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
359 		error = ENOENT;
360 		goto done;
361 	}
362 	tcp_func_set_ptr = blk;
363 done:
364 	rw_wunlock(&tcp_function_lock);
365 	return (error);
366 }
367 
368 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
369 	    CTLTYPE_STRING | CTLFLAG_RW,
370 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
371 	    "Set/get the default TCP functions");
372 
373 static int
374 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
375 {
376 	int error, cnt, linesz;
377 	struct tcp_function *f;
378 	char *buffer, *cp;
379 	size_t bufsz, outsz;
380 	bool alias;
381 
382 	cnt = 0;
383 	rw_rlock(&tcp_function_lock);
384 	TAILQ_FOREACH(f, &t_functions, tf_next) {
385 		cnt++;
386 	}
387 	rw_runlock(&tcp_function_lock);
388 
389 	bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
390 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
391 
392 	error = 0;
393 	cp = buffer;
394 
395 	linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
396 	    "Alias", "PCB count");
397 	cp += linesz;
398 	bufsz -= linesz;
399 	outsz = linesz;
400 
401 	rw_rlock(&tcp_function_lock);
402 	TAILQ_FOREACH(f, &t_functions, tf_next) {
403 		alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
404 		linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
405 		    f->tf_fb->tfb_tcp_block_name,
406 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
407 		    alias ? f->tf_name : "-",
408 		    f->tf_fb->tfb_refcnt);
409 		if (linesz >= bufsz) {
410 			error = EOVERFLOW;
411 			break;
412 		}
413 		cp += linesz;
414 		bufsz -= linesz;
415 		outsz += linesz;
416 	}
417 	rw_runlock(&tcp_function_lock);
418 	if (error == 0)
419 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
420 	free(buffer, M_TEMP);
421 	return (error);
422 }
423 
424 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
425 	    CTLTYPE_STRING|CTLFLAG_RD,
426 	    NULL, 0, sysctl_net_inet_list_available, "A",
427 	    "list available TCP Function sets");
428 
429 /*
430  * Exports one (struct tcp_function_info) for each alias/name.
431  */
432 static int
433 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
434 {
435 	int cnt, error;
436 	struct tcp_function *f;
437 	struct tcp_function_info tfi;
438 
439 	/*
440 	 * We don't allow writes.
441 	 */
442 	if (req->newptr != NULL)
443 		return (EINVAL);
444 
445 	/*
446 	 * Wire the old buffer so we can directly copy the functions to
447 	 * user space without dropping the lock.
448 	 */
449 	if (req->oldptr != NULL) {
450 		error = sysctl_wire_old_buffer(req, 0);
451 		if (error)
452 			return (error);
453 	}
454 
455 	/*
456 	 * Walk the list and copy out matching entries. If INVARIANTS
457 	 * is compiled in, also walk the list to verify the length of
458 	 * the list matches what we have recorded.
459 	 */
460 	rw_rlock(&tcp_function_lock);
461 #ifdef INVARIANTS
462 	cnt = 0;
463 #else
464 	if (req->oldptr == NULL) {
465 		cnt = tcp_fb_cnt;
466 		goto skip_loop;
467 	}
468 #endif
469 	TAILQ_FOREACH(f, &t_functions, tf_next) {
470 #ifdef INVARIANTS
471 		cnt++;
472 #endif
473 		if (req->oldptr != NULL) {
474 			tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
475 			tfi.tfi_id = f->tf_fb->tfb_id;
476 			(void)strncpy(tfi.tfi_alias, f->tf_name,
477 			    TCP_FUNCTION_NAME_LEN_MAX);
478 			tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
479 			(void)strncpy(tfi.tfi_name,
480 			    f->tf_fb->tfb_tcp_block_name,
481 			    TCP_FUNCTION_NAME_LEN_MAX);
482 			tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
483 			error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
484 			/*
485 			 * Don't stop on error, as that is the
486 			 * mechanism we use to accumulate length
487 			 * information if the buffer was too short.
488 			 */
489 		}
490 	}
491 	KASSERT(cnt == tcp_fb_cnt,
492 	    ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
493 #ifndef INVARIANTS
494 skip_loop:
495 #endif
496 	rw_runlock(&tcp_function_lock);
497 	if (req->oldptr == NULL)
498 		error = SYSCTL_OUT(req, NULL,
499 		    (cnt + 1) * sizeof(struct tcp_function_info));
500 
501 	return (error);
502 }
503 
504 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
505 	    CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
506 	    NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
507 	    "List TCP function block name-to-ID mappings");
508 
509 /*
510  * Target size of TCP PCB hash tables. Must be a power of two.
511  *
512  * Note that this can be overridden by the kernel environment
513  * variable net.inet.tcp.tcbhashsize
514  */
515 #ifndef TCBHASHSIZE
516 #define TCBHASHSIZE	0
517 #endif
518 
519 /*
520  * XXX
521  * Callouts should be moved into struct tcp directly.  They are currently
522  * separate because the tcpcb structure is exported to userland for sysctl
523  * parsing purposes, which do not know about callouts.
524  */
525 struct tcpcb_mem {
526 	struct	tcpcb		tcb;
527 	struct	tcp_timer	tt;
528 	struct	cc_var		ccv;
529 #ifdef TCP_HHOOK
530 	struct	osd		osd;
531 #endif
532 };
533 
534 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
535 #define	V_tcpcb_zone			VNET(tcpcb_zone)
536 
537 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
538 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
539 
540 static struct mtx isn_mtx;
541 
542 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
543 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
544 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
545 
546 /*
547  * TCP initialization.
548  */
549 static void
550 tcp_zone_change(void *tag)
551 {
552 
553 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
554 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
555 	tcp_tw_zone_change();
556 }
557 
558 static int
559 tcp_inpcb_init(void *mem, int size, int flags)
560 {
561 	struct inpcb *inp = mem;
562 
563 	INP_LOCK_INIT(inp, "inp", "tcpinp");
564 	return (0);
565 }
566 
567 /*
568  * Take a value and get the next power of 2 that doesn't overflow.
569  * Used to size the tcp_inpcb hash buckets.
570  */
571 static int
572 maketcp_hashsize(int size)
573 {
574 	int hashsize;
575 
576 	/*
577 	 * auto tune.
578 	 * get the next power of 2 higher than maxsockets.
579 	 */
580 	hashsize = 1 << fls(size);
581 	/* catch overflow, and just go one power of 2 smaller */
582 	if (hashsize < size) {
583 		hashsize = 1 << (fls(size) - 1);
584 	}
585 	return (hashsize);
586 }
587 
588 static volatile int next_tcp_stack_id = 1;
589 
590 /*
591  * Register a TCP function block with the name provided in the names
592  * array.  (Note that this function does NOT automatically register
593  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
594  * explicitly include blk->tfb_tcp_block_name in the list of names if
595  * you wish to register the stack with that name.)
596  *
597  * Either all name registrations will succeed or all will fail.  If
598  * a name registration fails, the function will update the num_names
599  * argument to point to the array index of the name that encountered
600  * the failure.
601  *
602  * Returns 0 on success, or an error code on failure.
603  */
604 int
605 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
606     const char *names[], int *num_names)
607 {
608 	struct tcp_function *n;
609 	struct tcp_function_set fs;
610 	int error, i;
611 
612 	KASSERT(names != NULL && *num_names > 0,
613 	    ("%s: Called with 0-length name list", __func__));
614 	KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
615 
616 	if (t_functions_inited == 0) {
617 		init_tcp_functions();
618 	}
619 	if ((blk->tfb_tcp_output == NULL) ||
620 	    (blk->tfb_tcp_do_segment == NULL) ||
621 	    (blk->tfb_tcp_ctloutput == NULL) ||
622 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
623 		/*
624 		 * These functions are required and you
625 		 * need a name.
626 		 */
627 		*num_names = 0;
628 		return (EINVAL);
629 	}
630 	if (blk->tfb_tcp_timer_stop_all ||
631 	    blk->tfb_tcp_timer_activate ||
632 	    blk->tfb_tcp_timer_active ||
633 	    blk->tfb_tcp_timer_stop) {
634 		/*
635 		 * If you define one timer function you
636 		 * must have them all.
637 		 */
638 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
639 		    (blk->tfb_tcp_timer_activate == NULL) ||
640 		    (blk->tfb_tcp_timer_active == NULL) ||
641 		    (blk->tfb_tcp_timer_stop == NULL)) {
642 			*num_names = 0;
643 			return (EINVAL);
644 		}
645 	}
646 
647 	refcount_init(&blk->tfb_refcnt, 0);
648 	blk->tfb_flags = 0;
649 	blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
650 	for (i = 0; i < *num_names; i++) {
651 		n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
652 		if (n == NULL) {
653 			error = ENOMEM;
654 			goto cleanup;
655 		}
656 		n->tf_fb = blk;
657 
658 		(void)strncpy(fs.function_set_name, names[i],
659 		    TCP_FUNCTION_NAME_LEN_MAX);
660 		fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
661 		rw_wlock(&tcp_function_lock);
662 		if (find_tcp_functions_locked(&fs) != NULL) {
663 			/* Duplicate name space not allowed */
664 			rw_wunlock(&tcp_function_lock);
665 			free(n, M_TCPFUNCTIONS);
666 			error = EALREADY;
667 			goto cleanup;
668 		}
669 		(void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
670 		n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
671 		TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
672 		tcp_fb_cnt++;
673 		rw_wunlock(&tcp_function_lock);
674 	}
675 	return(0);
676 
677 cleanup:
678 	/*
679 	 * Deregister the names we just added. Because registration failed
680 	 * for names[i], we don't need to deregister that name.
681 	 */
682 	*num_names = i;
683 	rw_wlock(&tcp_function_lock);
684 	while (--i >= 0) {
685 		TAILQ_FOREACH(n, &t_functions, tf_next) {
686 			if (!strncmp(n->tf_name, names[i],
687 			    TCP_FUNCTION_NAME_LEN_MAX)) {
688 				TAILQ_REMOVE(&t_functions, n, tf_next);
689 				tcp_fb_cnt--;
690 				n->tf_fb = NULL;
691 				free(n, M_TCPFUNCTIONS);
692 				break;
693 			}
694 		}
695 	}
696 	rw_wunlock(&tcp_function_lock);
697 	return (error);
698 }
699 
700 /*
701  * Register a TCP function block using the name provided in the name
702  * argument.
703  *
704  * Returns 0 on success, or an error code on failure.
705  */
706 int
707 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
708     int wait)
709 {
710 	const char *name_list[1];
711 	int num_names, rv;
712 
713 	num_names = 1;
714 	if (name != NULL)
715 		name_list[0] = name;
716 	else
717 		name_list[0] = blk->tfb_tcp_block_name;
718 	rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
719 	return (rv);
720 }
721 
722 /*
723  * Register a TCP function block using the name defined in
724  * blk->tfb_tcp_block_name.
725  *
726  * Returns 0 on success, or an error code on failure.
727  */
728 int
729 register_tcp_functions(struct tcp_function_block *blk, int wait)
730 {
731 
732 	return (register_tcp_functions_as_name(blk, NULL, wait));
733 }
734 
735 int
736 deregister_tcp_functions(struct tcp_function_block *blk)
737 {
738 	struct tcp_function *f;
739 	int error=ENOENT;
740 
741 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
742 		/* You can't un-register the default */
743 		return (EPERM);
744 	}
745 	rw_wlock(&tcp_function_lock);
746 	if (blk == tcp_func_set_ptr) {
747 		/* You can't free the current default */
748 		rw_wunlock(&tcp_function_lock);
749 		return (EBUSY);
750 	}
751 	if (blk->tfb_refcnt) {
752 		/* Still tcb attached, mark it. */
753 		blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
754 		rw_wunlock(&tcp_function_lock);
755 		return (EBUSY);
756 	}
757 	while (find_tcp_fb_locked(blk, &f) != NULL) {
758 		/* Found */
759 		TAILQ_REMOVE(&t_functions, f, tf_next);
760 		tcp_fb_cnt--;
761 		f->tf_fb = NULL;
762 		free(f, M_TCPFUNCTIONS);
763 		error = 0;
764 	}
765 	rw_wunlock(&tcp_function_lock);
766 	return (error);
767 }
768 
769 void
770 tcp_init(void)
771 {
772 	const char *tcbhash_tuneable;
773 	int hashsize;
774 
775 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
776 
777 #ifdef TCP_HHOOK
778 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
779 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
780 		printf("%s: WARNING: unable to register helper hook\n", __func__);
781 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
782 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
783 		printf("%s: WARNING: unable to register helper hook\n", __func__);
784 #endif
785 	hashsize = TCBHASHSIZE;
786 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
787 	if (hashsize == 0) {
788 		/*
789 		 * Auto tune the hash size based on maxsockets.
790 		 * A perfect hash would have a 1:1 mapping
791 		 * (hashsize = maxsockets) however it's been
792 		 * suggested that O(2) average is better.
793 		 */
794 		hashsize = maketcp_hashsize(maxsockets / 4);
795 		/*
796 		 * Our historical default is 512,
797 		 * do not autotune lower than this.
798 		 */
799 		if (hashsize < 512)
800 			hashsize = 512;
801 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
802 			printf("%s: %s auto tuned to %d\n", __func__,
803 			    tcbhash_tuneable, hashsize);
804 	}
805 	/*
806 	 * We require a hashsize to be a power of two.
807 	 * Previously if it was not a power of two we would just reset it
808 	 * back to 512, which could be a nasty surprise if you did not notice
809 	 * the error message.
810 	 * Instead what we do is clip it to the closest power of two lower
811 	 * than the specified hash value.
812 	 */
813 	if (!powerof2(hashsize)) {
814 		int oldhashsize = hashsize;
815 
816 		hashsize = maketcp_hashsize(hashsize);
817 		/* prevent absurdly low value */
818 		if (hashsize < 16)
819 			hashsize = 16;
820 		printf("%s: WARNING: TCB hash size not a power of 2, "
821 		    "clipped from %d to %d.\n", __func__, oldhashsize,
822 		    hashsize);
823 	}
824 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
825 	    "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
826 
827 	/*
828 	 * These have to be type stable for the benefit of the timers.
829 	 */
830 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
831 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
832 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
833 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
834 
835 	tcp_tw_init();
836 	syncache_init();
837 	tcp_hc_init();
838 
839 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
840 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
841 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
842 
843 	tcp_fastopen_init();
844 
845 	/* Skip initialization of globals for non-default instances. */
846 	if (!IS_DEFAULT_VNET(curvnet))
847 		return;
848 
849 	tcp_reass_global_init();
850 
851 	/* XXX virtualize those bellow? */
852 	tcp_delacktime = TCPTV_DELACK;
853 	tcp_keepinit = TCPTV_KEEP_INIT;
854 	tcp_keepidle = TCPTV_KEEP_IDLE;
855 	tcp_keepintvl = TCPTV_KEEPINTVL;
856 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
857 	tcp_msl = TCPTV_MSL;
858 	tcp_rexmit_min = TCPTV_MIN;
859 	if (tcp_rexmit_min < 1)
860 		tcp_rexmit_min = 1;
861 	tcp_persmin = TCPTV_PERSMIN;
862 	tcp_persmax = TCPTV_PERSMAX;
863 	tcp_rexmit_slop = TCPTV_CPU_VAR;
864 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
865 	tcp_tcbhashsize = hashsize;
866 	/* Setup the tcp function block list */
867 	init_tcp_functions();
868 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
869 #ifdef TCP_BLACKBOX
870 	/* Initialize the TCP logging data. */
871 	tcp_log_init();
872 #endif
873 
874 	if (tcp_soreceive_stream) {
875 #ifdef INET
876 		tcp_usrreqs.pru_soreceive = soreceive_stream;
877 #endif
878 #ifdef INET6
879 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
880 #endif /* INET6 */
881 	}
882 
883 #ifdef INET6
884 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
885 #else /* INET6 */
886 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
887 #endif /* INET6 */
888 	if (max_protohdr < TCP_MINPROTOHDR)
889 		max_protohdr = TCP_MINPROTOHDR;
890 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
891 		panic("tcp_init");
892 #undef TCP_MINPROTOHDR
893 
894 	ISN_LOCK_INIT();
895 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
896 		SHUTDOWN_PRI_DEFAULT);
897 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
898 		EVENTHANDLER_PRI_ANY);
899 #ifdef TCPPCAP
900 	tcp_pcap_init();
901 #endif
902 }
903 
904 #ifdef VIMAGE
905 static void
906 tcp_destroy(void *unused __unused)
907 {
908 	int n;
909 #ifdef TCP_HHOOK
910 	int error;
911 #endif
912 
913 	/*
914 	 * All our processes are gone, all our sockets should be cleaned
915 	 * up, which means, we should be past the tcp_discardcb() calls.
916 	 * Sleep to let all tcpcb timers really disappear and cleanup.
917 	 */
918 	for (;;) {
919 		INP_LIST_RLOCK(&V_tcbinfo);
920 		n = V_tcbinfo.ipi_count;
921 		INP_LIST_RUNLOCK(&V_tcbinfo);
922 		if (n == 0)
923 			break;
924 		pause("tcpdes", hz / 10);
925 	}
926 	tcp_hc_destroy();
927 	syncache_destroy();
928 	tcp_tw_destroy();
929 	in_pcbinfo_destroy(&V_tcbinfo);
930 	/* tcp_discardcb() clears the sack_holes up. */
931 	uma_zdestroy(V_sack_hole_zone);
932 	uma_zdestroy(V_tcpcb_zone);
933 
934 	/*
935 	 * Cannot free the zone until all tcpcbs are released as we attach
936 	 * the allocations to them.
937 	 */
938 	tcp_fastopen_destroy();
939 
940 #ifdef TCP_HHOOK
941 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
942 	if (error != 0) {
943 		printf("%s: WARNING: unable to deregister helper hook "
944 		    "type=%d, id=%d: error %d returned\n", __func__,
945 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
946 	}
947 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
948 	if (error != 0) {
949 		printf("%s: WARNING: unable to deregister helper hook "
950 		    "type=%d, id=%d: error %d returned\n", __func__,
951 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
952 	}
953 #endif
954 }
955 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
956 #endif
957 
958 void
959 tcp_fini(void *xtp)
960 {
961 
962 }
963 
964 /*
965  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
966  * tcp_template used to store this data in mbufs, but we now recopy it out
967  * of the tcpcb each time to conserve mbufs.
968  */
969 void
970 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
971 {
972 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
973 
974 	INP_WLOCK_ASSERT(inp);
975 
976 #ifdef INET6
977 	if ((inp->inp_vflag & INP_IPV6) != 0) {
978 		struct ip6_hdr *ip6;
979 
980 		ip6 = (struct ip6_hdr *)ip_ptr;
981 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
982 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
983 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
984 			(IPV6_VERSION & IPV6_VERSION_MASK);
985 		ip6->ip6_nxt = IPPROTO_TCP;
986 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
987 		ip6->ip6_src = inp->in6p_laddr;
988 		ip6->ip6_dst = inp->in6p_faddr;
989 	}
990 #endif /* INET6 */
991 #if defined(INET6) && defined(INET)
992 	else
993 #endif
994 #ifdef INET
995 	{
996 		struct ip *ip;
997 
998 		ip = (struct ip *)ip_ptr;
999 		ip->ip_v = IPVERSION;
1000 		ip->ip_hl = 5;
1001 		ip->ip_tos = inp->inp_ip_tos;
1002 		ip->ip_len = 0;
1003 		ip->ip_id = 0;
1004 		ip->ip_off = 0;
1005 		ip->ip_ttl = inp->inp_ip_ttl;
1006 		ip->ip_sum = 0;
1007 		ip->ip_p = IPPROTO_TCP;
1008 		ip->ip_src = inp->inp_laddr;
1009 		ip->ip_dst = inp->inp_faddr;
1010 	}
1011 #endif /* INET */
1012 	th->th_sport = inp->inp_lport;
1013 	th->th_dport = inp->inp_fport;
1014 	th->th_seq = 0;
1015 	th->th_ack = 0;
1016 	th->th_x2 = 0;
1017 	th->th_off = 5;
1018 	th->th_flags = 0;
1019 	th->th_win = 0;
1020 	th->th_urp = 0;
1021 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
1022 }
1023 
1024 /*
1025  * Create template to be used to send tcp packets on a connection.
1026  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1027  * use for this function is in keepalives, which use tcp_respond.
1028  */
1029 struct tcptemp *
1030 tcpip_maketemplate(struct inpcb *inp)
1031 {
1032 	struct tcptemp *t;
1033 
1034 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1035 	if (t == NULL)
1036 		return (NULL);
1037 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1038 	return (t);
1039 }
1040 
1041 /*
1042  * Send a single message to the TCP at address specified by
1043  * the given TCP/IP header.  If m == NULL, then we make a copy
1044  * of the tcpiphdr at th and send directly to the addressed host.
1045  * This is used to force keep alive messages out using the TCP
1046  * template for a connection.  If flags are given then we send
1047  * a message back to the TCP which originated the segment th,
1048  * and discard the mbuf containing it and any other attached mbufs.
1049  *
1050  * In any case the ack and sequence number of the transmitted
1051  * segment are as specified by the parameters.
1052  *
1053  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1054  */
1055 void
1056 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1057     tcp_seq ack, tcp_seq seq, int flags)
1058 {
1059 	struct tcpopt to;
1060 	struct inpcb *inp;
1061 	struct ip *ip;
1062 	struct mbuf *optm;
1063 	struct tcphdr *nth;
1064 	u_char *optp;
1065 #ifdef INET6
1066 	struct ip6_hdr *ip6;
1067 	int isipv6;
1068 #endif /* INET6 */
1069 	int optlen, tlen, win;
1070 	bool incl_opts;
1071 
1072 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1073 
1074 #ifdef INET6
1075 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1076 	ip6 = ipgen;
1077 #endif /* INET6 */
1078 	ip = ipgen;
1079 
1080 	if (tp != NULL) {
1081 		inp = tp->t_inpcb;
1082 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1083 		INP_WLOCK_ASSERT(inp);
1084 	} else
1085 		inp = NULL;
1086 
1087 	incl_opts = false;
1088 	win = 0;
1089 	if (tp != NULL) {
1090 		if (!(flags & TH_RST)) {
1091 			win = sbspace(&inp->inp_socket->so_rcv);
1092 			if (win > TCP_MAXWIN << tp->rcv_scale)
1093 				win = TCP_MAXWIN << tp->rcv_scale;
1094 		}
1095 		if ((tp->t_flags & TF_NOOPT) == 0)
1096 			incl_opts = true;
1097 	}
1098 	if (m == NULL) {
1099 		m = m_gethdr(M_NOWAIT, MT_DATA);
1100 		if (m == NULL)
1101 			return;
1102 		m->m_data += max_linkhdr;
1103 #ifdef INET6
1104 		if (isipv6) {
1105 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
1106 			      sizeof(struct ip6_hdr));
1107 			ip6 = mtod(m, struct ip6_hdr *);
1108 			nth = (struct tcphdr *)(ip6 + 1);
1109 		} else
1110 #endif /* INET6 */
1111 		{
1112 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1113 			ip = mtod(m, struct ip *);
1114 			nth = (struct tcphdr *)(ip + 1);
1115 		}
1116 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1117 		flags = TH_ACK;
1118 	} else if (!M_WRITABLE(m)) {
1119 		struct mbuf *n;
1120 
1121 		/* Can't reuse 'm', allocate a new mbuf. */
1122 		n = m_gethdr(M_NOWAIT, MT_DATA);
1123 		if (n == NULL) {
1124 			m_freem(m);
1125 			return;
1126 		}
1127 
1128 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1129 			m_freem(m);
1130 			m_freem(n);
1131 			return;
1132 		}
1133 
1134 		n->m_data += max_linkhdr;
1135 		/* m_len is set later */
1136 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1137 #ifdef INET6
1138 		if (isipv6) {
1139 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
1140 			      sizeof(struct ip6_hdr));
1141 			ip6 = mtod(n, struct ip6_hdr *);
1142 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1143 			nth = (struct tcphdr *)(ip6 + 1);
1144 		} else
1145 #endif /* INET6 */
1146 		{
1147 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1148 			ip = mtod(n, struct ip *);
1149 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1150 			nth = (struct tcphdr *)(ip + 1);
1151 		}
1152 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1153 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1154 		th = nth;
1155 		m_freem(m);
1156 		m = n;
1157 	} else {
1158 		/*
1159 		 *  reuse the mbuf.
1160 		 * XXX MRT We inherit the FIB, which is lucky.
1161 		 */
1162 		m_freem(m->m_next);
1163 		m->m_next = NULL;
1164 		m->m_data = (caddr_t)ipgen;
1165 		/* m_len is set later */
1166 #ifdef INET6
1167 		if (isipv6) {
1168 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1169 			nth = (struct tcphdr *)(ip6 + 1);
1170 		} else
1171 #endif /* INET6 */
1172 		{
1173 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1174 			nth = (struct tcphdr *)(ip + 1);
1175 		}
1176 		if (th != nth) {
1177 			/*
1178 			 * this is usually a case when an extension header
1179 			 * exists between the IPv6 header and the
1180 			 * TCP header.
1181 			 */
1182 			nth->th_sport = th->th_sport;
1183 			nth->th_dport = th->th_dport;
1184 		}
1185 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1186 #undef xchg
1187 	}
1188 	tlen = 0;
1189 #ifdef INET6
1190 	if (isipv6)
1191 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1192 #endif
1193 #if defined(INET) && defined(INET6)
1194 	else
1195 #endif
1196 #ifdef INET
1197 		tlen = sizeof (struct tcpiphdr);
1198 #endif
1199 #ifdef INVARIANTS
1200 	m->m_len = 0;
1201 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1202 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1203 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1204 #endif
1205 	m->m_len = tlen;
1206 	to.to_flags = 0;
1207 	if (incl_opts) {
1208 		/* Make sure we have room. */
1209 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1210 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1211 			if (m->m_next) {
1212 				optp = mtod(m->m_next, u_char *);
1213 				optm = m->m_next;
1214 			} else
1215 				incl_opts = false;
1216 		} else {
1217 			optp = (u_char *) (nth + 1);
1218 			optm = m;
1219 		}
1220 	}
1221 	if (incl_opts) {
1222 		/* Timestamps. */
1223 		if (tp->t_flags & TF_RCVD_TSTMP) {
1224 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1225 			to.to_tsecr = tp->ts_recent;
1226 			to.to_flags |= TOF_TS;
1227 		}
1228 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1229 		/* TCP-MD5 (RFC2385). */
1230 		if (tp->t_flags & TF_SIGNATURE)
1231 			to.to_flags |= TOF_SIGNATURE;
1232 #endif
1233 		/* Add the options. */
1234 		tlen += optlen = tcp_addoptions(&to, optp);
1235 
1236 		/* Update m_len in the correct mbuf. */
1237 		optm->m_len += optlen;
1238 	} else
1239 		optlen = 0;
1240 #ifdef INET6
1241 	if (isipv6) {
1242 		ip6->ip6_flow = 0;
1243 		ip6->ip6_vfc = IPV6_VERSION;
1244 		ip6->ip6_nxt = IPPROTO_TCP;
1245 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1246 	}
1247 #endif
1248 #if defined(INET) && defined(INET6)
1249 	else
1250 #endif
1251 #ifdef INET
1252 	{
1253 		ip->ip_len = htons(tlen);
1254 		ip->ip_ttl = V_ip_defttl;
1255 		if (V_path_mtu_discovery)
1256 			ip->ip_off |= htons(IP_DF);
1257 	}
1258 #endif
1259 	m->m_pkthdr.len = tlen;
1260 	m->m_pkthdr.rcvif = NULL;
1261 #ifdef MAC
1262 	if (inp != NULL) {
1263 		/*
1264 		 * Packet is associated with a socket, so allow the
1265 		 * label of the response to reflect the socket label.
1266 		 */
1267 		INP_WLOCK_ASSERT(inp);
1268 		mac_inpcb_create_mbuf(inp, m);
1269 	} else {
1270 		/*
1271 		 * Packet is not associated with a socket, so possibly
1272 		 * update the label in place.
1273 		 */
1274 		mac_netinet_tcp_reply(m);
1275 	}
1276 #endif
1277 	nth->th_seq = htonl(seq);
1278 	nth->th_ack = htonl(ack);
1279 	nth->th_x2 = 0;
1280 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1281 	nth->th_flags = flags;
1282 	if (tp != NULL)
1283 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1284 	else
1285 		nth->th_win = htons((u_short)win);
1286 	nth->th_urp = 0;
1287 
1288 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1289 	if (to.to_flags & TOF_SIGNATURE) {
1290 		if (!TCPMD5_ENABLED() ||
1291 		    TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1292 			m_freem(m);
1293 			return;
1294 		}
1295 	}
1296 #endif
1297 
1298 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1299 #ifdef INET6
1300 	if (isipv6) {
1301 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1302 		nth->th_sum = in6_cksum_pseudo(ip6,
1303 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1304 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1305 		    NULL, NULL);
1306 	}
1307 #endif /* INET6 */
1308 #if defined(INET6) && defined(INET)
1309 	else
1310 #endif
1311 #ifdef INET
1312 	{
1313 		m->m_pkthdr.csum_flags = CSUM_TCP;
1314 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1315 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1316 	}
1317 #endif /* INET */
1318 #ifdef TCPDEBUG
1319 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1320 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1321 #endif
1322 	TCP_PROBE3(debug__output, tp, th, m);
1323 	if (flags & TH_RST)
1324 		TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1325 
1326 #ifdef INET6
1327 	if (isipv6) {
1328 		TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1329 		(void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1330 	}
1331 #endif /* INET6 */
1332 #if defined(INET) && defined(INET6)
1333 	else
1334 #endif
1335 #ifdef INET
1336 	{
1337 		TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1338 		(void)ip_output(m, NULL, NULL, 0, NULL, inp);
1339 	}
1340 #endif
1341 }
1342 
1343 /*
1344  * Create a new TCP control block, making an
1345  * empty reassembly queue and hooking it to the argument
1346  * protocol control block.  The `inp' parameter must have
1347  * come from the zone allocator set up in tcp_init().
1348  */
1349 struct tcpcb *
1350 tcp_newtcpcb(struct inpcb *inp)
1351 {
1352 	struct tcpcb_mem *tm;
1353 	struct tcpcb *tp;
1354 #ifdef INET6
1355 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1356 #endif /* INET6 */
1357 
1358 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1359 	if (tm == NULL)
1360 		return (NULL);
1361 	tp = &tm->tcb;
1362 
1363 	/* Initialise cc_var struct for this tcpcb. */
1364 	tp->ccv = &tm->ccv;
1365 	tp->ccv->type = IPPROTO_TCP;
1366 	tp->ccv->ccvc.tcp = tp;
1367 	rw_rlock(&tcp_function_lock);
1368 	tp->t_fb = tcp_func_set_ptr;
1369 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1370 	rw_runlock(&tcp_function_lock);
1371 	/*
1372 	 * Use the current system default CC algorithm.
1373 	 */
1374 	CC_LIST_RLOCK();
1375 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1376 	CC_ALGO(tp) = CC_DEFAULT();
1377 	CC_LIST_RUNLOCK();
1378 
1379 	if (CC_ALGO(tp)->cb_init != NULL)
1380 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1381 			if (tp->t_fb->tfb_tcp_fb_fini)
1382 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1383 			refcount_release(&tp->t_fb->tfb_refcnt);
1384 			uma_zfree(V_tcpcb_zone, tm);
1385 			return (NULL);
1386 		}
1387 
1388 #ifdef TCP_HHOOK
1389 	tp->osd = &tm->osd;
1390 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1391 		if (tp->t_fb->tfb_tcp_fb_fini)
1392 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1393 		refcount_release(&tp->t_fb->tfb_refcnt);
1394 		uma_zfree(V_tcpcb_zone, tm);
1395 		return (NULL);
1396 	}
1397 #endif
1398 
1399 #ifdef VIMAGE
1400 	tp->t_vnet = inp->inp_vnet;
1401 #endif
1402 	tp->t_timers = &tm->tt;
1403 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
1404 	tp->t_maxseg =
1405 #ifdef INET6
1406 		isipv6 ? V_tcp_v6mssdflt :
1407 #endif /* INET6 */
1408 		V_tcp_mssdflt;
1409 
1410 	/* Set up our timeouts. */
1411 	callout_init(&tp->t_timers->tt_rexmt, 1);
1412 	callout_init(&tp->t_timers->tt_persist, 1);
1413 	callout_init(&tp->t_timers->tt_keep, 1);
1414 	callout_init(&tp->t_timers->tt_2msl, 1);
1415 	callout_init(&tp->t_timers->tt_delack, 1);
1416 
1417 	if (V_tcp_do_rfc1323)
1418 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1419 	if (V_tcp_do_sack)
1420 		tp->t_flags |= TF_SACK_PERMIT;
1421 	TAILQ_INIT(&tp->snd_holes);
1422 	/*
1423 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1424 	 * is called.
1425 	 */
1426 	in_pcbref(inp);	/* Reference for tcpcb */
1427 	tp->t_inpcb = inp;
1428 
1429 	/*
1430 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1431 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1432 	 * reasonable initial retransmit time.
1433 	 */
1434 	tp->t_srtt = TCPTV_SRTTBASE;
1435 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1436 	tp->t_rttmin = tcp_rexmit_min;
1437 	tp->t_rxtcur = TCPTV_RTOBASE;
1438 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1439 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1440 	tp->t_rcvtime = ticks;
1441 	/*
1442 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1443 	 * because the socket may be bound to an IPv6 wildcard address,
1444 	 * which may match an IPv4-mapped IPv6 address.
1445 	 */
1446 	inp->inp_ip_ttl = V_ip_defttl;
1447 	inp->inp_ppcb = tp;
1448 #ifdef TCPPCAP
1449 	/*
1450 	 * Init the TCP PCAP queues.
1451 	 */
1452 	tcp_pcap_tcpcb_init(tp);
1453 #endif
1454 #ifdef TCP_BLACKBOX
1455 	/* Initialize the per-TCPCB log data. */
1456 	tcp_log_tcpcbinit(tp);
1457 #endif
1458 	if (tp->t_fb->tfb_tcp_fb_init) {
1459 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1460 	}
1461 	return (tp);		/* XXX */
1462 }
1463 
1464 /*
1465  * Switch the congestion control algorithm back to NewReno for any active
1466  * control blocks using an algorithm which is about to go away.
1467  * This ensures the CC framework can allow the unload to proceed without leaving
1468  * any dangling pointers which would trigger a panic.
1469  * Returning non-zero would inform the CC framework that something went wrong
1470  * and it would be unsafe to allow the unload to proceed. However, there is no
1471  * way for this to occur with this implementation so we always return zero.
1472  */
1473 int
1474 tcp_ccalgounload(struct cc_algo *unload_algo)
1475 {
1476 	struct cc_algo *tmpalgo;
1477 	struct inpcb *inp;
1478 	struct tcpcb *tp;
1479 	VNET_ITERATOR_DECL(vnet_iter);
1480 
1481 	/*
1482 	 * Check all active control blocks across all network stacks and change
1483 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1484 	 * requires cleanup code to be run, call it.
1485 	 */
1486 	VNET_LIST_RLOCK();
1487 	VNET_FOREACH(vnet_iter) {
1488 		CURVNET_SET(vnet_iter);
1489 		INP_INFO_WLOCK(&V_tcbinfo);
1490 		/*
1491 		 * New connections already part way through being initialised
1492 		 * with the CC algo we're removing will not race with this code
1493 		 * because the INP_INFO_WLOCK is held during initialisation. We
1494 		 * therefore don't enter the loop below until the connection
1495 		 * list has stabilised.
1496 		 */
1497 		LIST_FOREACH(inp, &V_tcb, inp_list) {
1498 			INP_WLOCK(inp);
1499 			/* Important to skip tcptw structs. */
1500 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1501 			    (tp = intotcpcb(inp)) != NULL) {
1502 				/*
1503 				 * By holding INP_WLOCK here, we are assured
1504 				 * that the connection is not currently
1505 				 * executing inside the CC module's functions
1506 				 * i.e. it is safe to make the switch back to
1507 				 * NewReno.
1508 				 */
1509 				if (CC_ALGO(tp) == unload_algo) {
1510 					tmpalgo = CC_ALGO(tp);
1511 					/* NewReno does not require any init. */
1512 					CC_ALGO(tp) = &newreno_cc_algo;
1513 					if (tmpalgo->cb_destroy != NULL)
1514 						tmpalgo->cb_destroy(tp->ccv);
1515 				}
1516 			}
1517 			INP_WUNLOCK(inp);
1518 		}
1519 		INP_INFO_WUNLOCK(&V_tcbinfo);
1520 		CURVNET_RESTORE();
1521 	}
1522 	VNET_LIST_RUNLOCK();
1523 
1524 	return (0);
1525 }
1526 
1527 /*
1528  * Drop a TCP connection, reporting
1529  * the specified error.  If connection is synchronized,
1530  * then send a RST to peer.
1531  */
1532 struct tcpcb *
1533 tcp_drop(struct tcpcb *tp, int errno)
1534 {
1535 	struct socket *so = tp->t_inpcb->inp_socket;
1536 
1537 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1538 	INP_WLOCK_ASSERT(tp->t_inpcb);
1539 
1540 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1541 		tcp_state_change(tp, TCPS_CLOSED);
1542 		(void) tp->t_fb->tfb_tcp_output(tp);
1543 		TCPSTAT_INC(tcps_drops);
1544 	} else
1545 		TCPSTAT_INC(tcps_conndrops);
1546 	if (errno == ETIMEDOUT && tp->t_softerror)
1547 		errno = tp->t_softerror;
1548 	so->so_error = errno;
1549 	return (tcp_close(tp));
1550 }
1551 
1552 void
1553 tcp_discardcb(struct tcpcb *tp)
1554 {
1555 	struct inpcb *inp = tp->t_inpcb;
1556 	struct socket *so = inp->inp_socket;
1557 #ifdef INET6
1558 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1559 #endif /* INET6 */
1560 	int released;
1561 
1562 	INP_WLOCK_ASSERT(inp);
1563 
1564 	/*
1565 	 * Make sure that all of our timers are stopped before we delete the
1566 	 * PCB.
1567 	 *
1568 	 * If stopping a timer fails, we schedule a discard function in same
1569 	 * callout, and the last discard function called will take care of
1570 	 * deleting the tcpcb.
1571 	 */
1572 	tp->t_timers->tt_draincnt = 0;
1573 	tcp_timer_stop(tp, TT_REXMT);
1574 	tcp_timer_stop(tp, TT_PERSIST);
1575 	tcp_timer_stop(tp, TT_KEEP);
1576 	tcp_timer_stop(tp, TT_2MSL);
1577 	tcp_timer_stop(tp, TT_DELACK);
1578 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1579 		/*
1580 		 * Call the stop-all function of the methods,
1581 		 * this function should call the tcp_timer_stop()
1582 		 * method with each of the function specific timeouts.
1583 		 * That stop will be called via the tfb_tcp_timer_stop()
1584 		 * which should use the async drain function of the
1585 		 * callout system (see tcp_var.h).
1586 		 */
1587 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1588 	}
1589 
1590 	/*
1591 	 * If we got enough samples through the srtt filter,
1592 	 * save the rtt and rttvar in the routing entry.
1593 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1594 	 * 4 samples is enough for the srtt filter to converge
1595 	 * to within enough % of the correct value; fewer samples
1596 	 * and we could save a bogus rtt. The danger is not high
1597 	 * as tcp quickly recovers from everything.
1598 	 * XXX: Works very well but needs some more statistics!
1599 	 */
1600 	if (tp->t_rttupdated >= 4) {
1601 		struct hc_metrics_lite metrics;
1602 		uint32_t ssthresh;
1603 
1604 		bzero(&metrics, sizeof(metrics));
1605 		/*
1606 		 * Update the ssthresh always when the conditions below
1607 		 * are satisfied. This gives us better new start value
1608 		 * for the congestion avoidance for new connections.
1609 		 * ssthresh is only set if packet loss occurred on a session.
1610 		 *
1611 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1612 		 * being torn down.  Ideally this code would not use 'so'.
1613 		 */
1614 		ssthresh = tp->snd_ssthresh;
1615 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1616 			/*
1617 			 * convert the limit from user data bytes to
1618 			 * packets then to packet data bytes.
1619 			 */
1620 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1621 			if (ssthresh < 2)
1622 				ssthresh = 2;
1623 			ssthresh *= (tp->t_maxseg +
1624 #ifdef INET6
1625 			    (isipv6 ? sizeof (struct ip6_hdr) +
1626 				sizeof (struct tcphdr) :
1627 #endif
1628 				sizeof (struct tcpiphdr)
1629 #ifdef INET6
1630 			    )
1631 #endif
1632 			    );
1633 		} else
1634 			ssthresh = 0;
1635 		metrics.rmx_ssthresh = ssthresh;
1636 
1637 		metrics.rmx_rtt = tp->t_srtt;
1638 		metrics.rmx_rttvar = tp->t_rttvar;
1639 		metrics.rmx_cwnd = tp->snd_cwnd;
1640 		metrics.rmx_sendpipe = 0;
1641 		metrics.rmx_recvpipe = 0;
1642 
1643 		tcp_hc_update(&inp->inp_inc, &metrics);
1644 	}
1645 
1646 	/* free the reassembly queue, if any */
1647 	tcp_reass_flush(tp);
1648 
1649 #ifdef TCP_OFFLOAD
1650 	/* Disconnect offload device, if any. */
1651 	if (tp->t_flags & TF_TOE)
1652 		tcp_offload_detach(tp);
1653 #endif
1654 
1655 	tcp_free_sackholes(tp);
1656 
1657 #ifdef TCPPCAP
1658 	/* Free the TCP PCAP queues. */
1659 	tcp_pcap_drain(&(tp->t_inpkts));
1660 	tcp_pcap_drain(&(tp->t_outpkts));
1661 #endif
1662 
1663 	/* Allow the CC algorithm to clean up after itself. */
1664 	if (CC_ALGO(tp)->cb_destroy != NULL)
1665 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1666 
1667 #ifdef TCP_HHOOK
1668 	khelp_destroy_osd(tp->osd);
1669 #endif
1670 
1671 	CC_ALGO(tp) = NULL;
1672 	inp->inp_ppcb = NULL;
1673 	if (tp->t_timers->tt_draincnt == 0) {
1674 		/* We own the last reference on tcpcb, let's free it. */
1675 #ifdef TCP_BLACKBOX
1676 		tcp_log_tcpcbfini(tp);
1677 #endif
1678 		TCPSTATES_DEC(tp->t_state);
1679 		if (tp->t_fb->tfb_tcp_fb_fini)
1680 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1681 		refcount_release(&tp->t_fb->tfb_refcnt);
1682 		tp->t_inpcb = NULL;
1683 		uma_zfree(V_tcpcb_zone, tp);
1684 		released = in_pcbrele_wlocked(inp);
1685 		KASSERT(!released, ("%s: inp %p should not have been released "
1686 			"here", __func__, inp));
1687 	}
1688 }
1689 
1690 void
1691 tcp_timer_discard(void *ptp)
1692 {
1693 	struct inpcb *inp;
1694 	struct tcpcb *tp;
1695 
1696 	tp = (struct tcpcb *)ptp;
1697 	CURVNET_SET(tp->t_vnet);
1698 	INP_INFO_RLOCK(&V_tcbinfo);
1699 	inp = tp->t_inpcb;
1700 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1701 		__func__, tp));
1702 	INP_WLOCK(inp);
1703 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1704 		("%s: tcpcb has to be stopped here", __func__));
1705 	tp->t_timers->tt_draincnt--;
1706 	if (tp->t_timers->tt_draincnt == 0) {
1707 		/* We own the last reference on this tcpcb, let's free it. */
1708 #ifdef TCP_BLACKBOX
1709 		tcp_log_tcpcbfini(tp);
1710 #endif
1711 		TCPSTATES_DEC(tp->t_state);
1712 		if (tp->t_fb->tfb_tcp_fb_fini)
1713 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1714 		refcount_release(&tp->t_fb->tfb_refcnt);
1715 		tp->t_inpcb = NULL;
1716 		uma_zfree(V_tcpcb_zone, tp);
1717 		if (in_pcbrele_wlocked(inp)) {
1718 			INP_INFO_RUNLOCK(&V_tcbinfo);
1719 			CURVNET_RESTORE();
1720 			return;
1721 		}
1722 	}
1723 	INP_WUNLOCK(inp);
1724 	INP_INFO_RUNLOCK(&V_tcbinfo);
1725 	CURVNET_RESTORE();
1726 }
1727 
1728 /*
1729  * Attempt to close a TCP control block, marking it as dropped, and freeing
1730  * the socket if we hold the only reference.
1731  */
1732 struct tcpcb *
1733 tcp_close(struct tcpcb *tp)
1734 {
1735 	struct inpcb *inp = tp->t_inpcb;
1736 	struct socket *so;
1737 
1738 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1739 	INP_WLOCK_ASSERT(inp);
1740 
1741 #ifdef TCP_OFFLOAD
1742 	if (tp->t_state == TCPS_LISTEN)
1743 		tcp_offload_listen_stop(tp);
1744 #endif
1745 	/*
1746 	 * This releases the TFO pending counter resource for TFO listen
1747 	 * sockets as well as passively-created TFO sockets that transition
1748 	 * from SYN_RECEIVED to CLOSED.
1749 	 */
1750 	if (tp->t_tfo_pending) {
1751 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1752 		tp->t_tfo_pending = NULL;
1753 	}
1754 	in_pcbdrop(inp);
1755 	TCPSTAT_INC(tcps_closed);
1756 	if (tp->t_state != TCPS_CLOSED)
1757 		tcp_state_change(tp, TCPS_CLOSED);
1758 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1759 	so = inp->inp_socket;
1760 	soisdisconnected(so);
1761 	if (inp->inp_flags & INP_SOCKREF) {
1762 		KASSERT(so->so_state & SS_PROTOREF,
1763 		    ("tcp_close: !SS_PROTOREF"));
1764 		inp->inp_flags &= ~INP_SOCKREF;
1765 		INP_WUNLOCK(inp);
1766 		SOCK_LOCK(so);
1767 		so->so_state &= ~SS_PROTOREF;
1768 		sofree(so);
1769 		return (NULL);
1770 	}
1771 	return (tp);
1772 }
1773 
1774 void
1775 tcp_drain(void)
1776 {
1777 	VNET_ITERATOR_DECL(vnet_iter);
1778 
1779 	if (!do_tcpdrain)
1780 		return;
1781 
1782 	VNET_LIST_RLOCK_NOSLEEP();
1783 	VNET_FOREACH(vnet_iter) {
1784 		CURVNET_SET(vnet_iter);
1785 		struct inpcb *inpb;
1786 		struct tcpcb *tcpb;
1787 
1788 	/*
1789 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1790 	 * if there is one...
1791 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1792 	 *      reassembly queue should be flushed, but in a situation
1793 	 *	where we're really low on mbufs, this is potentially
1794 	 *	useful.
1795 	 */
1796 		INP_INFO_WLOCK(&V_tcbinfo);
1797 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1798 			if (inpb->inp_flags & INP_TIMEWAIT)
1799 				continue;
1800 			INP_WLOCK(inpb);
1801 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1802 				tcp_reass_flush(tcpb);
1803 				tcp_clean_sackreport(tcpb);
1804 #ifdef TCP_BLACKBOX
1805 				tcp_log_drain(tcpb);
1806 #endif
1807 #ifdef TCPPCAP
1808 				if (tcp_pcap_aggressive_free) {
1809 					/* Free the TCP PCAP queues. */
1810 					tcp_pcap_drain(&(tcpb->t_inpkts));
1811 					tcp_pcap_drain(&(tcpb->t_outpkts));
1812 				}
1813 #endif
1814 			}
1815 			INP_WUNLOCK(inpb);
1816 		}
1817 		INP_INFO_WUNLOCK(&V_tcbinfo);
1818 		CURVNET_RESTORE();
1819 	}
1820 	VNET_LIST_RUNLOCK_NOSLEEP();
1821 }
1822 
1823 /*
1824  * Notify a tcp user of an asynchronous error;
1825  * store error as soft error, but wake up user
1826  * (for now, won't do anything until can select for soft error).
1827  *
1828  * Do not wake up user since there currently is no mechanism for
1829  * reporting soft errors (yet - a kqueue filter may be added).
1830  */
1831 static struct inpcb *
1832 tcp_notify(struct inpcb *inp, int error)
1833 {
1834 	struct tcpcb *tp;
1835 
1836 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1837 	INP_WLOCK_ASSERT(inp);
1838 
1839 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1840 	    (inp->inp_flags & INP_DROPPED))
1841 		return (inp);
1842 
1843 	tp = intotcpcb(inp);
1844 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1845 
1846 	/*
1847 	 * Ignore some errors if we are hooked up.
1848 	 * If connection hasn't completed, has retransmitted several times,
1849 	 * and receives a second error, give up now.  This is better
1850 	 * than waiting a long time to establish a connection that
1851 	 * can never complete.
1852 	 */
1853 	if (tp->t_state == TCPS_ESTABLISHED &&
1854 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1855 	     error == EHOSTDOWN)) {
1856 		if (inp->inp_route.ro_rt) {
1857 			RTFREE(inp->inp_route.ro_rt);
1858 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
1859 		}
1860 		return (inp);
1861 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1862 	    tp->t_softerror) {
1863 		tp = tcp_drop(tp, error);
1864 		if (tp != NULL)
1865 			return (inp);
1866 		else
1867 			return (NULL);
1868 	} else {
1869 		tp->t_softerror = error;
1870 		return (inp);
1871 	}
1872 #if 0
1873 	wakeup( &so->so_timeo);
1874 	sorwakeup(so);
1875 	sowwakeup(so);
1876 #endif
1877 }
1878 
1879 static int
1880 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1881 {
1882 	int error, i, m, n, pcb_count;
1883 	struct inpcb *inp, **inp_list;
1884 	inp_gen_t gencnt;
1885 	struct xinpgen xig;
1886 
1887 	/*
1888 	 * The process of preparing the TCB list is too time-consuming and
1889 	 * resource-intensive to repeat twice on every request.
1890 	 */
1891 	if (req->oldptr == NULL) {
1892 		n = V_tcbinfo.ipi_count +
1893 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1894 		n += imax(n / 8, 10);
1895 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1896 		return (0);
1897 	}
1898 
1899 	if (req->newptr != NULL)
1900 		return (EPERM);
1901 
1902 	/*
1903 	 * OK, now we're committed to doing something.
1904 	 */
1905 	INP_LIST_RLOCK(&V_tcbinfo);
1906 	gencnt = V_tcbinfo.ipi_gencnt;
1907 	n = V_tcbinfo.ipi_count;
1908 	INP_LIST_RUNLOCK(&V_tcbinfo);
1909 
1910 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1911 
1912 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1913 		+ (n + m) * sizeof(struct xtcpcb));
1914 	if (error != 0)
1915 		return (error);
1916 
1917 	xig.xig_len = sizeof xig;
1918 	xig.xig_count = n + m;
1919 	xig.xig_gen = gencnt;
1920 	xig.xig_sogen = so_gencnt;
1921 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1922 	if (error)
1923 		return (error);
1924 
1925 	error = syncache_pcblist(req, m, &pcb_count);
1926 	if (error)
1927 		return (error);
1928 
1929 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1930 
1931 	INP_INFO_WLOCK(&V_tcbinfo);
1932 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1933 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1934 		INP_WLOCK(inp);
1935 		if (inp->inp_gencnt <= gencnt) {
1936 			/*
1937 			 * XXX: This use of cr_cansee(), introduced with
1938 			 * TCP state changes, is not quite right, but for
1939 			 * now, better than nothing.
1940 			 */
1941 			if (inp->inp_flags & INP_TIMEWAIT) {
1942 				if (intotw(inp) != NULL)
1943 					error = cr_cansee(req->td->td_ucred,
1944 					    intotw(inp)->tw_cred);
1945 				else
1946 					error = EINVAL;	/* Skip this inp. */
1947 			} else
1948 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1949 			if (error == 0) {
1950 				in_pcbref(inp);
1951 				inp_list[i++] = inp;
1952 			}
1953 		}
1954 		INP_WUNLOCK(inp);
1955 	}
1956 	INP_INFO_WUNLOCK(&V_tcbinfo);
1957 	n = i;
1958 
1959 	error = 0;
1960 	for (i = 0; i < n; i++) {
1961 		inp = inp_list[i];
1962 		INP_RLOCK(inp);
1963 		if (inp->inp_gencnt <= gencnt) {
1964 			struct xtcpcb xt;
1965 
1966 			tcp_inptoxtp(inp, &xt);
1967 			INP_RUNLOCK(inp);
1968 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1969 		} else
1970 			INP_RUNLOCK(inp);
1971 	}
1972 	INP_INFO_RLOCK(&V_tcbinfo);
1973 	for (i = 0; i < n; i++) {
1974 		inp = inp_list[i];
1975 		INP_RLOCK(inp);
1976 		if (!in_pcbrele_rlocked(inp))
1977 			INP_RUNLOCK(inp);
1978 	}
1979 	INP_INFO_RUNLOCK(&V_tcbinfo);
1980 
1981 	if (!error) {
1982 		/*
1983 		 * Give the user an updated idea of our state.
1984 		 * If the generation differs from what we told
1985 		 * her before, she knows that something happened
1986 		 * while we were processing this request, and it
1987 		 * might be necessary to retry.
1988 		 */
1989 		INP_LIST_RLOCK(&V_tcbinfo);
1990 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1991 		xig.xig_sogen = so_gencnt;
1992 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1993 		INP_LIST_RUNLOCK(&V_tcbinfo);
1994 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1995 	}
1996 	free(inp_list, M_TEMP);
1997 	return (error);
1998 }
1999 
2000 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2001     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
2002     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2003 
2004 #ifdef INET
2005 static int
2006 tcp_getcred(SYSCTL_HANDLER_ARGS)
2007 {
2008 	struct xucred xuc;
2009 	struct sockaddr_in addrs[2];
2010 	struct inpcb *inp;
2011 	int error;
2012 
2013 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2014 	if (error)
2015 		return (error);
2016 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2017 	if (error)
2018 		return (error);
2019 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2020 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2021 	if (inp != NULL) {
2022 		if (inp->inp_socket == NULL)
2023 			error = ENOENT;
2024 		if (error == 0)
2025 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2026 		if (error == 0)
2027 			cru2x(inp->inp_cred, &xuc);
2028 		INP_RUNLOCK(inp);
2029 	} else
2030 		error = ENOENT;
2031 	if (error == 0)
2032 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2033 	return (error);
2034 }
2035 
2036 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2037     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2038     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
2039 #endif /* INET */
2040 
2041 #ifdef INET6
2042 static int
2043 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2044 {
2045 	struct xucred xuc;
2046 	struct sockaddr_in6 addrs[2];
2047 	struct inpcb *inp;
2048 	int error;
2049 #ifdef INET
2050 	int mapped = 0;
2051 #endif
2052 
2053 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
2054 	if (error)
2055 		return (error);
2056 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
2057 	if (error)
2058 		return (error);
2059 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2060 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2061 		return (error);
2062 	}
2063 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2064 #ifdef INET
2065 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2066 			mapped = 1;
2067 		else
2068 #endif
2069 			return (EINVAL);
2070 	}
2071 
2072 #ifdef INET
2073 	if (mapped == 1)
2074 		inp = in_pcblookup(&V_tcbinfo,
2075 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2076 			addrs[1].sin6_port,
2077 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2078 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2079 	else
2080 #endif
2081 		inp = in6_pcblookup(&V_tcbinfo,
2082 			&addrs[1].sin6_addr, addrs[1].sin6_port,
2083 			&addrs[0].sin6_addr, addrs[0].sin6_port,
2084 			INPLOOKUP_RLOCKPCB, NULL);
2085 	if (inp != NULL) {
2086 		if (inp->inp_socket == NULL)
2087 			error = ENOENT;
2088 		if (error == 0)
2089 			error = cr_canseeinpcb(req->td->td_ucred, inp);
2090 		if (error == 0)
2091 			cru2x(inp->inp_cred, &xuc);
2092 		INP_RUNLOCK(inp);
2093 	} else
2094 		error = ENOENT;
2095 	if (error == 0)
2096 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2097 	return (error);
2098 }
2099 
2100 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2101     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2102     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2103 #endif /* INET6 */
2104 
2105 
2106 #ifdef INET
2107 void
2108 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2109 {
2110 	struct ip *ip = vip;
2111 	struct tcphdr *th;
2112 	struct in_addr faddr;
2113 	struct inpcb *inp;
2114 	struct tcpcb *tp;
2115 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2116 	struct icmp *icp;
2117 	struct in_conninfo inc;
2118 	tcp_seq icmp_tcp_seq;
2119 	int mtu;
2120 
2121 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
2122 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2123 		return;
2124 
2125 	if (cmd == PRC_MSGSIZE)
2126 		notify = tcp_mtudisc_notify;
2127 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2128 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2129 		cmd == PRC_TIMXCEED_INTRANS) && ip)
2130 		notify = tcp_drop_syn_sent;
2131 
2132 	/*
2133 	 * Hostdead is ugly because it goes linearly through all PCBs.
2134 	 * XXX: We never get this from ICMP, otherwise it makes an
2135 	 * excellent DoS attack on machines with many connections.
2136 	 */
2137 	else if (cmd == PRC_HOSTDEAD)
2138 		ip = NULL;
2139 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2140 		return;
2141 
2142 	if (ip == NULL) {
2143 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2144 		return;
2145 	}
2146 
2147 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2148 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2149 	INP_INFO_RLOCK(&V_tcbinfo);
2150 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2151 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2152 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2153 		/* signal EHOSTDOWN, as it flushes the cached route */
2154 		inp = (*notify)(inp, EHOSTDOWN);
2155 		goto out;
2156 	}
2157 	icmp_tcp_seq = th->th_seq;
2158 	if (inp != NULL)  {
2159 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2160 		    !(inp->inp_flags & INP_DROPPED) &&
2161 		    !(inp->inp_socket == NULL)) {
2162 			tp = intotcpcb(inp);
2163 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2164 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2165 				if (cmd == PRC_MSGSIZE) {
2166 					/*
2167 					 * MTU discovery:
2168 					 * If we got a needfrag set the MTU
2169 					 * in the route to the suggested new
2170 					 * value (if given) and then notify.
2171 					 */
2172 					mtu = ntohs(icp->icmp_nextmtu);
2173 					/*
2174 					 * If no alternative MTU was
2175 					 * proposed, try the next smaller
2176 					 * one.
2177 					 */
2178 					if (!mtu)
2179 						mtu = ip_next_mtu(
2180 						    ntohs(ip->ip_len), 1);
2181 					if (mtu < V_tcp_minmss +
2182 					    sizeof(struct tcpiphdr))
2183 						mtu = V_tcp_minmss +
2184 						    sizeof(struct tcpiphdr);
2185 					/*
2186 					 * Only process the offered MTU if it
2187 					 * is smaller than the current one.
2188 					 */
2189 					if (mtu < tp->t_maxseg +
2190 					    sizeof(struct tcpiphdr)) {
2191 						bzero(&inc, sizeof(inc));
2192 						inc.inc_faddr = faddr;
2193 						inc.inc_fibnum =
2194 						    inp->inp_inc.inc_fibnum;
2195 						tcp_hc_updatemtu(&inc, mtu);
2196 						tcp_mtudisc(inp, mtu);
2197 					}
2198 				} else
2199 					inp = (*notify)(inp,
2200 					    inetctlerrmap[cmd]);
2201 			}
2202 		}
2203 	} else {
2204 		bzero(&inc, sizeof(inc));
2205 		inc.inc_fport = th->th_dport;
2206 		inc.inc_lport = th->th_sport;
2207 		inc.inc_faddr = faddr;
2208 		inc.inc_laddr = ip->ip_src;
2209 		syncache_unreach(&inc, icmp_tcp_seq);
2210 	}
2211 out:
2212 	if (inp != NULL)
2213 		INP_WUNLOCK(inp);
2214 	INP_INFO_RUNLOCK(&V_tcbinfo);
2215 }
2216 #endif /* INET */
2217 
2218 #ifdef INET6
2219 void
2220 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2221 {
2222 	struct in6_addr *dst;
2223 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2224 	struct ip6_hdr *ip6;
2225 	struct mbuf *m;
2226 	struct inpcb *inp;
2227 	struct tcpcb *tp;
2228 	struct icmp6_hdr *icmp6;
2229 	struct ip6ctlparam *ip6cp = NULL;
2230 	const struct sockaddr_in6 *sa6_src = NULL;
2231 	struct in_conninfo inc;
2232 	struct tcp_ports {
2233 		uint16_t th_sport;
2234 		uint16_t th_dport;
2235 	} t_ports;
2236 	tcp_seq icmp_tcp_seq;
2237 	unsigned int mtu;
2238 	unsigned int off;
2239 
2240 	if (sa->sa_family != AF_INET6 ||
2241 	    sa->sa_len != sizeof(struct sockaddr_in6))
2242 		return;
2243 
2244 	/* if the parameter is from icmp6, decode it. */
2245 	if (d != NULL) {
2246 		ip6cp = (struct ip6ctlparam *)d;
2247 		icmp6 = ip6cp->ip6c_icmp6;
2248 		m = ip6cp->ip6c_m;
2249 		ip6 = ip6cp->ip6c_ip6;
2250 		off = ip6cp->ip6c_off;
2251 		sa6_src = ip6cp->ip6c_src;
2252 		dst = ip6cp->ip6c_finaldst;
2253 	} else {
2254 		m = NULL;
2255 		ip6 = NULL;
2256 		off = 0;	/* fool gcc */
2257 		sa6_src = &sa6_any;
2258 		dst = NULL;
2259 	}
2260 
2261 	if (cmd == PRC_MSGSIZE)
2262 		notify = tcp_mtudisc_notify;
2263 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2264 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2265 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2266 		notify = tcp_drop_syn_sent;
2267 
2268 	/*
2269 	 * Hostdead is ugly because it goes linearly through all PCBs.
2270 	 * XXX: We never get this from ICMP, otherwise it makes an
2271 	 * excellent DoS attack on machines with many connections.
2272 	 */
2273 	else if (cmd == PRC_HOSTDEAD)
2274 		ip6 = NULL;
2275 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2276 		return;
2277 
2278 	if (ip6 == NULL) {
2279 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2280 			      (const struct sockaddr *)sa6_src,
2281 			      0, cmd, NULL, notify);
2282 		return;
2283 	}
2284 
2285 	/* Check if we can safely get the ports from the tcp hdr */
2286 	if (m == NULL ||
2287 	    (m->m_pkthdr.len <
2288 		(int32_t) (off + sizeof(struct tcp_ports)))) {
2289 		return;
2290 	}
2291 	bzero(&t_ports, sizeof(struct tcp_ports));
2292 	m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2293 	INP_INFO_RLOCK(&V_tcbinfo);
2294 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2295 	    &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2296 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2297 		/* signal EHOSTDOWN, as it flushes the cached route */
2298 		inp = (*notify)(inp, EHOSTDOWN);
2299 		goto out;
2300 	}
2301 	off += sizeof(struct tcp_ports);
2302 	if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2303 		goto out;
2304 	}
2305 	m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2306 	if (inp != NULL)  {
2307 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2308 		    !(inp->inp_flags & INP_DROPPED) &&
2309 		    !(inp->inp_socket == NULL)) {
2310 			tp = intotcpcb(inp);
2311 			if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2312 			    SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2313 				if (cmd == PRC_MSGSIZE) {
2314 					/*
2315 					 * MTU discovery:
2316 					 * If we got a needfrag set the MTU
2317 					 * in the route to the suggested new
2318 					 * value (if given) and then notify.
2319 					 */
2320 					mtu = ntohl(icmp6->icmp6_mtu);
2321 					/*
2322 					 * If no alternative MTU was
2323 					 * proposed, or the proposed
2324 					 * MTU was too small, set to
2325 					 * the min.
2326 					 */
2327 					if (mtu < IPV6_MMTU)
2328 						mtu = IPV6_MMTU - 8;
2329 					bzero(&inc, sizeof(inc));
2330 					inc.inc_fibnum = M_GETFIB(m);
2331 					inc.inc_flags |= INC_ISIPV6;
2332 					inc.inc6_faddr = *dst;
2333 					if (in6_setscope(&inc.inc6_faddr,
2334 						m->m_pkthdr.rcvif, NULL))
2335 						goto out;
2336 					/*
2337 					 * Only process the offered MTU if it
2338 					 * is smaller than the current one.
2339 					 */
2340 					if (mtu < tp->t_maxseg +
2341 					    sizeof (struct tcphdr) +
2342 					    sizeof (struct ip6_hdr)) {
2343 						tcp_hc_updatemtu(&inc, mtu);
2344 						tcp_mtudisc(inp, mtu);
2345 						ICMP6STAT_INC(icp6s_pmtuchg);
2346 					}
2347 				} else
2348 					inp = (*notify)(inp,
2349 					    inet6ctlerrmap[cmd]);
2350 			}
2351 		}
2352 	} else {
2353 		bzero(&inc, sizeof(inc));
2354 		inc.inc_fibnum = M_GETFIB(m);
2355 		inc.inc_flags |= INC_ISIPV6;
2356 		inc.inc_fport = t_ports.th_dport;
2357 		inc.inc_lport = t_ports.th_sport;
2358 		inc.inc6_faddr = *dst;
2359 		inc.inc6_laddr = ip6->ip6_src;
2360 		syncache_unreach(&inc, icmp_tcp_seq);
2361 	}
2362 out:
2363 	if (inp != NULL)
2364 		INP_WUNLOCK(inp);
2365 	INP_INFO_RUNLOCK(&V_tcbinfo);
2366 }
2367 #endif /* INET6 */
2368 
2369 
2370 /*
2371  * Following is where TCP initial sequence number generation occurs.
2372  *
2373  * There are two places where we must use initial sequence numbers:
2374  * 1.  In SYN-ACK packets.
2375  * 2.  In SYN packets.
2376  *
2377  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2378  * tcp_syncache.c for details.
2379  *
2380  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2381  * depends on this property.  In addition, these ISNs should be
2382  * unguessable so as to prevent connection hijacking.  To satisfy
2383  * the requirements of this situation, the algorithm outlined in
2384  * RFC 1948 is used, with only small modifications.
2385  *
2386  * Implementation details:
2387  *
2388  * Time is based off the system timer, and is corrected so that it
2389  * increases by one megabyte per second.  This allows for proper
2390  * recycling on high speed LANs while still leaving over an hour
2391  * before rollover.
2392  *
2393  * As reading the *exact* system time is too expensive to be done
2394  * whenever setting up a TCP connection, we increment the time
2395  * offset in two ways.  First, a small random positive increment
2396  * is added to isn_offset for each connection that is set up.
2397  * Second, the function tcp_isn_tick fires once per clock tick
2398  * and increments isn_offset as necessary so that sequence numbers
2399  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2400  * random positive increments serve only to ensure that the same
2401  * exact sequence number is never sent out twice (as could otherwise
2402  * happen when a port is recycled in less than the system tick
2403  * interval.)
2404  *
2405  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2406  * between seeding of isn_secret.  This is normally set to zero,
2407  * as reseeding should not be necessary.
2408  *
2409  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2410  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2411  * general, this means holding an exclusive (write) lock.
2412  */
2413 
2414 #define ISN_BYTES_PER_SECOND 1048576
2415 #define ISN_STATIC_INCREMENT 4096
2416 #define ISN_RANDOM_INCREMENT (4096 - 1)
2417 
2418 static VNET_DEFINE(u_char, isn_secret[32]);
2419 static VNET_DEFINE(int, isn_last);
2420 static VNET_DEFINE(int, isn_last_reseed);
2421 static VNET_DEFINE(u_int32_t, isn_offset);
2422 static VNET_DEFINE(u_int32_t, isn_offset_old);
2423 
2424 #define	V_isn_secret			VNET(isn_secret)
2425 #define	V_isn_last			VNET(isn_last)
2426 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2427 #define	V_isn_offset			VNET(isn_offset)
2428 #define	V_isn_offset_old		VNET(isn_offset_old)
2429 
2430 tcp_seq
2431 tcp_new_isn(struct tcpcb *tp)
2432 {
2433 	MD5_CTX isn_ctx;
2434 	u_int32_t md5_buffer[4];
2435 	tcp_seq new_isn;
2436 	u_int32_t projected_offset;
2437 
2438 	INP_WLOCK_ASSERT(tp->t_inpcb);
2439 
2440 	ISN_LOCK();
2441 	/* Seed if this is the first use, reseed if requested. */
2442 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2443 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2444 		< (u_int)ticks))) {
2445 		read_random(&V_isn_secret, sizeof(V_isn_secret));
2446 		V_isn_last_reseed = ticks;
2447 	}
2448 
2449 	/* Compute the md5 hash and return the ISN. */
2450 	MD5Init(&isn_ctx);
2451 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2452 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2453 #ifdef INET6
2454 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2455 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2456 			  sizeof(struct in6_addr));
2457 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2458 			  sizeof(struct in6_addr));
2459 	} else
2460 #endif
2461 	{
2462 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2463 			  sizeof(struct in_addr));
2464 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2465 			  sizeof(struct in_addr));
2466 	}
2467 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2468 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
2469 	new_isn = (tcp_seq) md5_buffer[0];
2470 	V_isn_offset += ISN_STATIC_INCREMENT +
2471 		(arc4random() & ISN_RANDOM_INCREMENT);
2472 	if (ticks != V_isn_last) {
2473 		projected_offset = V_isn_offset_old +
2474 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2475 		if (SEQ_GT(projected_offset, V_isn_offset))
2476 			V_isn_offset = projected_offset;
2477 		V_isn_offset_old = V_isn_offset;
2478 		V_isn_last = ticks;
2479 	}
2480 	new_isn += V_isn_offset;
2481 	ISN_UNLOCK();
2482 	return (new_isn);
2483 }
2484 
2485 /*
2486  * When a specific ICMP unreachable message is received and the
2487  * connection state is SYN-SENT, drop the connection.  This behavior
2488  * is controlled by the icmp_may_rst sysctl.
2489  */
2490 struct inpcb *
2491 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2492 {
2493 	struct tcpcb *tp;
2494 
2495 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2496 	INP_WLOCK_ASSERT(inp);
2497 
2498 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2499 	    (inp->inp_flags & INP_DROPPED))
2500 		return (inp);
2501 
2502 	tp = intotcpcb(inp);
2503 	if (tp->t_state != TCPS_SYN_SENT)
2504 		return (inp);
2505 
2506 	if (IS_FASTOPEN(tp->t_flags))
2507 		tcp_fastopen_disable_path(tp);
2508 
2509 	tp = tcp_drop(tp, errno);
2510 	if (tp != NULL)
2511 		return (inp);
2512 	else
2513 		return (NULL);
2514 }
2515 
2516 /*
2517  * When `need fragmentation' ICMP is received, update our idea of the MSS
2518  * based on the new value. Also nudge TCP to send something, since we
2519  * know the packet we just sent was dropped.
2520  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2521  */
2522 static struct inpcb *
2523 tcp_mtudisc_notify(struct inpcb *inp, int error)
2524 {
2525 
2526 	tcp_mtudisc(inp, -1);
2527 	return (inp);
2528 }
2529 
2530 static void
2531 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2532 {
2533 	struct tcpcb *tp;
2534 	struct socket *so;
2535 
2536 	INP_WLOCK_ASSERT(inp);
2537 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2538 	    (inp->inp_flags & INP_DROPPED))
2539 		return;
2540 
2541 	tp = intotcpcb(inp);
2542 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2543 
2544 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2545 
2546 	so = inp->inp_socket;
2547 	SOCKBUF_LOCK(&so->so_snd);
2548 	/* If the mss is larger than the socket buffer, decrease the mss. */
2549 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2550 		tp->t_maxseg = so->so_snd.sb_hiwat;
2551 	SOCKBUF_UNLOCK(&so->so_snd);
2552 
2553 	TCPSTAT_INC(tcps_mturesent);
2554 	tp->t_rtttime = 0;
2555 	tp->snd_nxt = tp->snd_una;
2556 	tcp_free_sackholes(tp);
2557 	tp->snd_recover = tp->snd_max;
2558 	if (tp->t_flags & TF_SACK_PERMIT)
2559 		EXIT_FASTRECOVERY(tp->t_flags);
2560 	tp->t_fb->tfb_tcp_output(tp);
2561 }
2562 
2563 #ifdef INET
2564 /*
2565  * Look-up the routing entry to the peer of this inpcb.  If no route
2566  * is found and it cannot be allocated, then return 0.  This routine
2567  * is called by TCP routines that access the rmx structure and by
2568  * tcp_mss_update to get the peer/interface MTU.
2569  */
2570 uint32_t
2571 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2572 {
2573 	struct nhop4_extended nh4;
2574 	struct ifnet *ifp;
2575 	uint32_t maxmtu = 0;
2576 
2577 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2578 
2579 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2580 
2581 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2582 		    NHR_REF, 0, &nh4) != 0)
2583 			return (0);
2584 
2585 		ifp = nh4.nh_ifp;
2586 		maxmtu = nh4.nh_mtu;
2587 
2588 		/* Report additional interface capabilities. */
2589 		if (cap != NULL) {
2590 			if (ifp->if_capenable & IFCAP_TSO4 &&
2591 			    ifp->if_hwassist & CSUM_TSO) {
2592 				cap->ifcap |= CSUM_TSO;
2593 				cap->tsomax = ifp->if_hw_tsomax;
2594 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2595 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2596 			}
2597 		}
2598 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2599 	}
2600 	return (maxmtu);
2601 }
2602 #endif /* INET */
2603 
2604 #ifdef INET6
2605 uint32_t
2606 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2607 {
2608 	struct nhop6_extended nh6;
2609 	struct in6_addr dst6;
2610 	uint32_t scopeid;
2611 	struct ifnet *ifp;
2612 	uint32_t maxmtu = 0;
2613 
2614 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2615 
2616 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2617 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2618 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2619 		    0, &nh6) != 0)
2620 			return (0);
2621 
2622 		ifp = nh6.nh_ifp;
2623 		maxmtu = nh6.nh_mtu;
2624 
2625 		/* Report additional interface capabilities. */
2626 		if (cap != NULL) {
2627 			if (ifp->if_capenable & IFCAP_TSO6 &&
2628 			    ifp->if_hwassist & CSUM_TSO) {
2629 				cap->ifcap |= CSUM_TSO;
2630 				cap->tsomax = ifp->if_hw_tsomax;
2631 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2632 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2633 			}
2634 		}
2635 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2636 	}
2637 
2638 	return (maxmtu);
2639 }
2640 #endif /* INET6 */
2641 
2642 /*
2643  * Calculate effective SMSS per RFC5681 definition for a given TCP
2644  * connection at its current state, taking into account SACK and etc.
2645  */
2646 u_int
2647 tcp_maxseg(const struct tcpcb *tp)
2648 {
2649 	u_int optlen;
2650 
2651 	if (tp->t_flags & TF_NOOPT)
2652 		return (tp->t_maxseg);
2653 
2654 	/*
2655 	 * Here we have a simplified code from tcp_addoptions(),
2656 	 * without a proper loop, and having most of paddings hardcoded.
2657 	 * We might make mistakes with padding here in some edge cases,
2658 	 * but this is harmless, since result of tcp_maxseg() is used
2659 	 * only in cwnd and ssthresh estimations.
2660 	 */
2661 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2662 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2663 		if (tp->t_flags & TF_RCVD_TSTMP)
2664 			optlen = TCPOLEN_TSTAMP_APPA;
2665 		else
2666 			optlen = 0;
2667 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2668 		if (tp->t_flags & TF_SIGNATURE)
2669 			optlen += PAD(TCPOLEN_SIGNATURE);
2670 #endif
2671 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2672 			optlen += TCPOLEN_SACKHDR;
2673 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2674 			optlen = PAD(optlen);
2675 		}
2676 	} else {
2677 		if (tp->t_flags & TF_REQ_TSTMP)
2678 			optlen = TCPOLEN_TSTAMP_APPA;
2679 		else
2680 			optlen = PAD(TCPOLEN_MAXSEG);
2681 		if (tp->t_flags & TF_REQ_SCALE)
2682 			optlen += PAD(TCPOLEN_WINDOW);
2683 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2684 		if (tp->t_flags & TF_SIGNATURE)
2685 			optlen += PAD(TCPOLEN_SIGNATURE);
2686 #endif
2687 		if (tp->t_flags & TF_SACK_PERMIT)
2688 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2689 	}
2690 #undef PAD
2691 	optlen = min(optlen, TCP_MAXOLEN);
2692 	return (tp->t_maxseg - optlen);
2693 }
2694 
2695 static int
2696 sysctl_drop(SYSCTL_HANDLER_ARGS)
2697 {
2698 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2699 	struct sockaddr_storage addrs[2];
2700 	struct inpcb *inp;
2701 	struct tcpcb *tp;
2702 	struct tcptw *tw;
2703 	struct sockaddr_in *fin, *lin;
2704 #ifdef INET6
2705 	struct sockaddr_in6 *fin6, *lin6;
2706 #endif
2707 	int error;
2708 
2709 	inp = NULL;
2710 	fin = lin = NULL;
2711 #ifdef INET6
2712 	fin6 = lin6 = NULL;
2713 #endif
2714 	error = 0;
2715 
2716 	if (req->oldptr != NULL || req->oldlen != 0)
2717 		return (EINVAL);
2718 	if (req->newptr == NULL)
2719 		return (EPERM);
2720 	if (req->newlen < sizeof(addrs))
2721 		return (ENOMEM);
2722 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2723 	if (error)
2724 		return (error);
2725 
2726 	switch (addrs[0].ss_family) {
2727 #ifdef INET6
2728 	case AF_INET6:
2729 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2730 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2731 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2732 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2733 			return (EINVAL);
2734 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2735 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2736 				return (EINVAL);
2737 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2738 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2739 			fin = (struct sockaddr_in *)&addrs[0];
2740 			lin = (struct sockaddr_in *)&addrs[1];
2741 			break;
2742 		}
2743 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2744 		if (error)
2745 			return (error);
2746 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2747 		if (error)
2748 			return (error);
2749 		break;
2750 #endif
2751 #ifdef INET
2752 	case AF_INET:
2753 		fin = (struct sockaddr_in *)&addrs[0];
2754 		lin = (struct sockaddr_in *)&addrs[1];
2755 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2756 		    lin->sin_len != sizeof(struct sockaddr_in))
2757 			return (EINVAL);
2758 		break;
2759 #endif
2760 	default:
2761 		return (EINVAL);
2762 	}
2763 	INP_INFO_RLOCK(&V_tcbinfo);
2764 	switch (addrs[0].ss_family) {
2765 #ifdef INET6
2766 	case AF_INET6:
2767 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2768 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2769 		    INPLOOKUP_WLOCKPCB, NULL);
2770 		break;
2771 #endif
2772 #ifdef INET
2773 	case AF_INET:
2774 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2775 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2776 		break;
2777 #endif
2778 	}
2779 	if (inp != NULL) {
2780 		if (inp->inp_flags & INP_TIMEWAIT) {
2781 			/*
2782 			 * XXXRW: There currently exists a state where an
2783 			 * inpcb is present, but its timewait state has been
2784 			 * discarded.  For now, don't allow dropping of this
2785 			 * type of inpcb.
2786 			 */
2787 			tw = intotw(inp);
2788 			if (tw != NULL)
2789 				tcp_twclose(tw, 0);
2790 			else
2791 				INP_WUNLOCK(inp);
2792 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2793 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2794 			tp = intotcpcb(inp);
2795 			tp = tcp_drop(tp, ECONNABORTED);
2796 			if (tp != NULL)
2797 				INP_WUNLOCK(inp);
2798 		} else
2799 			INP_WUNLOCK(inp);
2800 	} else
2801 		error = ESRCH;
2802 	INP_INFO_RUNLOCK(&V_tcbinfo);
2803 	return (error);
2804 }
2805 
2806 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2807     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2808     0, sysctl_drop, "", "Drop TCP connection");
2809 
2810 /*
2811  * Generate a standardized TCP log line for use throughout the
2812  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2813  * allow use in the interrupt context.
2814  *
2815  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2816  * NB: The function may return NULL if memory allocation failed.
2817  *
2818  * Due to header inclusion and ordering limitations the struct ip
2819  * and ip6_hdr pointers have to be passed as void pointers.
2820  */
2821 char *
2822 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2823     const void *ip6hdr)
2824 {
2825 
2826 	/* Is logging enabled? */
2827 	if (tcp_log_in_vain == 0)
2828 		return (NULL);
2829 
2830 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2831 }
2832 
2833 char *
2834 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2835     const void *ip6hdr)
2836 {
2837 
2838 	/* Is logging enabled? */
2839 	if (tcp_log_debug == 0)
2840 		return (NULL);
2841 
2842 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2843 }
2844 
2845 static char *
2846 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2847     const void *ip6hdr)
2848 {
2849 	char *s, *sp;
2850 	size_t size;
2851 	struct ip *ip;
2852 #ifdef INET6
2853 	const struct ip6_hdr *ip6;
2854 
2855 	ip6 = (const struct ip6_hdr *)ip6hdr;
2856 #endif /* INET6 */
2857 	ip = (struct ip *)ip4hdr;
2858 
2859 	/*
2860 	 * The log line looks like this:
2861 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2862 	 */
2863 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2864 	    sizeof(PRINT_TH_FLAGS) + 1 +
2865 #ifdef INET6
2866 	    2 * INET6_ADDRSTRLEN;
2867 #else
2868 	    2 * INET_ADDRSTRLEN;
2869 #endif /* INET6 */
2870 
2871 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2872 	if (s == NULL)
2873 		return (NULL);
2874 
2875 	strcat(s, "TCP: [");
2876 	sp = s + strlen(s);
2877 
2878 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2879 		inet_ntoa_r(inc->inc_faddr, sp);
2880 		sp = s + strlen(s);
2881 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2882 		sp = s + strlen(s);
2883 		inet_ntoa_r(inc->inc_laddr, sp);
2884 		sp = s + strlen(s);
2885 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2886 #ifdef INET6
2887 	} else if (inc) {
2888 		ip6_sprintf(sp, &inc->inc6_faddr);
2889 		sp = s + strlen(s);
2890 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2891 		sp = s + strlen(s);
2892 		ip6_sprintf(sp, &inc->inc6_laddr);
2893 		sp = s + strlen(s);
2894 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2895 	} else if (ip6 && th) {
2896 		ip6_sprintf(sp, &ip6->ip6_src);
2897 		sp = s + strlen(s);
2898 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2899 		sp = s + strlen(s);
2900 		ip6_sprintf(sp, &ip6->ip6_dst);
2901 		sp = s + strlen(s);
2902 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2903 #endif /* INET6 */
2904 #ifdef INET
2905 	} else if (ip && th) {
2906 		inet_ntoa_r(ip->ip_src, sp);
2907 		sp = s + strlen(s);
2908 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2909 		sp = s + strlen(s);
2910 		inet_ntoa_r(ip->ip_dst, sp);
2911 		sp = s + strlen(s);
2912 		sprintf(sp, "]:%i", ntohs(th->th_dport));
2913 #endif /* INET */
2914 	} else {
2915 		free(s, M_TCPLOG);
2916 		return (NULL);
2917 	}
2918 	sp = s + strlen(s);
2919 	if (th)
2920 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2921 	if (*(s + size - 1) != '\0')
2922 		panic("%s: string too long", __func__);
2923 	return (s);
2924 }
2925 
2926 /*
2927  * A subroutine which makes it easy to track TCP state changes with DTrace.
2928  * This function shouldn't be called for t_state initializations that don't
2929  * correspond to actual TCP state transitions.
2930  */
2931 void
2932 tcp_state_change(struct tcpcb *tp, int newstate)
2933 {
2934 #if defined(KDTRACE_HOOKS)
2935 	int pstate = tp->t_state;
2936 #endif
2937 
2938 	TCPSTATES_DEC(tp->t_state);
2939 	TCPSTATES_INC(newstate);
2940 	tp->t_state = newstate;
2941 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2942 }
2943 
2944 /*
2945  * Create an external-format (``xtcpcb'') structure using the information in
2946  * the kernel-format tcpcb structure pointed to by tp.  This is done to
2947  * reduce the spew of irrelevant information over this interface, to isolate
2948  * user code from changes in the kernel structure, and potentially to provide
2949  * information-hiding if we decide that some of this information should be
2950  * hidden from users.
2951  */
2952 void
2953 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2954 {
2955 	struct tcpcb *tp = intotcpcb(inp);
2956 	sbintime_t now;
2957 
2958 	if (inp->inp_flags & INP_TIMEWAIT) {
2959 		bzero(xt, sizeof(struct xtcpcb));
2960 		xt->t_state = TCPS_TIME_WAIT;
2961 	} else {
2962 		xt->t_state = tp->t_state;
2963 		xt->t_logstate = tp->t_logstate;
2964 		xt->t_flags = tp->t_flags;
2965 		xt->t_sndzerowin = tp->t_sndzerowin;
2966 		xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2967 		xt->t_rcvoopack = tp->t_rcvoopack;
2968 
2969 		now = getsbinuptime();
2970 #define	COPYTIMER(ttt)	do {						\
2971 		if (callout_active(&tp->t_timers->ttt))			\
2972 			xt->ttt = (tp->t_timers->ttt.c_time - now) /	\
2973 			    SBT_1MS;					\
2974 		else							\
2975 			xt->ttt = 0;					\
2976 } while (0)
2977 		COPYTIMER(tt_delack);
2978 		COPYTIMER(tt_rexmt);
2979 		COPYTIMER(tt_persist);
2980 		COPYTIMER(tt_keep);
2981 		COPYTIMER(tt_2msl);
2982 #undef COPYTIMER
2983 		xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2984 
2985 		bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2986 		    TCP_FUNCTION_NAME_LEN_MAX);
2987 		bzero(xt->xt_logid, TCP_LOG_ID_LEN);
2988 #ifdef TCP_BLACKBOX
2989 		(void)tcp_log_get_id(tp, xt->xt_logid);
2990 #endif
2991 	}
2992 
2993 	xt->xt_len = sizeof(struct xtcpcb);
2994 	in_pcbtoxinpcb(inp, &xt->xt_inp);
2995 	if (inp->inp_socket == NULL)
2996 		xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
2997 }
2998