xref: /freebsd/sys/netinet/tcp_subr.c (revision 69718b786d3943ea9a99eeeb5f5f6162f11c78b7)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/eventhandler.h>
45 #ifdef TCP_HHOOK
46 #include <sys/hhook.h>
47 #endif
48 #include <sys/kernel.h>
49 #ifdef TCP_HHOOK
50 #include <sys/khelp.h>
51 #endif
52 #include <sys/sysctl.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
56 #include <sys/mbuf.h>
57 #ifdef INET6
58 #include <sys/domain.h>
59 #endif
60 #include <sys/priv.h>
61 #include <sys/proc.h>
62 #include <sys/sdt.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/protosw.h>
66 #include <sys/random.h>
67 
68 #include <vm/uma.h>
69 
70 #include <net/route.h>
71 #include <net/if.h>
72 #include <net/if_var.h>
73 #include <net/vnet.h>
74 
75 #include <netinet/in.h>
76 #include <netinet/in_fib.h>
77 #include <netinet/in_kdtrace.h>
78 #include <netinet/in_pcb.h>
79 #include <netinet/in_systm.h>
80 #include <netinet/in_var.h>
81 #include <netinet/ip.h>
82 #include <netinet/ip_icmp.h>
83 #include <netinet/ip_var.h>
84 #ifdef INET6
85 #include <netinet/icmp6.h>
86 #include <netinet/ip6.h>
87 #include <netinet6/in6_fib.h>
88 #include <netinet6/in6_pcb.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/scope6_var.h>
91 #include <netinet6/nd6.h>
92 #endif
93 
94 #ifdef TCP_RFC7413
95 #include <netinet/tcp_fastopen.h>
96 #endif
97 #include <netinet/tcp.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 #include <netinet/tcp_syncache.h>
103 #include <netinet/cc/cc.h>
104 #ifdef INET6
105 #include <netinet6/tcp6_var.h>
106 #endif
107 #include <netinet/tcpip.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 #ifdef IPSEC
122 #include <netipsec/ipsec.h>
123 #include <netipsec/xform.h>
124 #ifdef INET6
125 #include <netipsec/ipsec6.h>
126 #endif
127 #include <netipsec/key.h>
128 #include <sys/syslog.h>
129 #endif /*IPSEC*/
130 
131 #include <machine/in_cksum.h>
132 #include <sys/md5.h>
133 
134 #include <security/mac/mac_framework.h>
135 
136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
137 #ifdef INET6
138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
139 #endif
140 
141 struct rwlock tcp_function_lock;
142 
143 static int
144 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
145 {
146 	int error, new;
147 
148 	new = V_tcp_mssdflt;
149 	error = sysctl_handle_int(oidp, &new, 0, req);
150 	if (error == 0 && req->newptr) {
151 		if (new < TCP_MINMSS)
152 			error = EINVAL;
153 		else
154 			V_tcp_mssdflt = new;
155 	}
156 	return (error);
157 }
158 
159 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
160     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
161     &sysctl_net_inet_tcp_mss_check, "I",
162     "Default TCP Maximum Segment Size");
163 
164 #ifdef INET6
165 static int
166 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
167 {
168 	int error, new;
169 
170 	new = V_tcp_v6mssdflt;
171 	error = sysctl_handle_int(oidp, &new, 0, req);
172 	if (error == 0 && req->newptr) {
173 		if (new < TCP_MINMSS)
174 			error = EINVAL;
175 		else
176 			V_tcp_v6mssdflt = new;
177 	}
178 	return (error);
179 }
180 
181 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
182     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
183     &sysctl_net_inet_tcp_mss_v6_check, "I",
184    "Default TCP Maximum Segment Size for IPv6");
185 #endif /* INET6 */
186 
187 /*
188  * Minimum MSS we accept and use. This prevents DoS attacks where
189  * we are forced to a ridiculous low MSS like 20 and send hundreds
190  * of packets instead of one. The effect scales with the available
191  * bandwidth and quickly saturates the CPU and network interface
192  * with packet generation and sending. Set to zero to disable MINMSS
193  * checking. This setting prevents us from sending too small packets.
194  */
195 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
197      &VNET_NAME(tcp_minmss), 0,
198     "Minimum TCP Maximum Segment Size");
199 
200 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
201 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
202     &VNET_NAME(tcp_do_rfc1323), 0,
203     "Enable rfc1323 (high performance TCP) extensions");
204 
205 static int	tcp_log_debug = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
207     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
208 
209 static int	tcp_tcbhashsize;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
211     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
212 
213 static int	do_tcpdrain = 1;
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
215     "Enable tcp_drain routine for extra help when low on mbufs");
216 
217 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
218     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
219 
220 static VNET_DEFINE(int, icmp_may_rst) = 1;
221 #define	V_icmp_may_rst			VNET(icmp_may_rst)
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
223     &VNET_NAME(icmp_may_rst), 0,
224     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
225 
226 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
227 #define	V_tcp_isn_reseed_interval	VNET(tcp_isn_reseed_interval)
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
229     &VNET_NAME(tcp_isn_reseed_interval), 0,
230     "Seconds between reseeding of ISN secret");
231 
232 static int	tcp_soreceive_stream;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
234     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
235 
236 #ifdef TCP_SIGNATURE
237 static int	tcp_sig_checksigs = 1;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
239     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
240 #endif
241 
242 VNET_DEFINE(uma_zone_t, sack_hole_zone);
243 #define	V_sack_hole_zone		VNET(sack_hole_zone)
244 
245 #ifdef TCP_HHOOK
246 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
247 #endif
248 
249 static struct inpcb *tcp_notify(struct inpcb *, int);
250 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
251 static void tcp_mtudisc(struct inpcb *, int);
252 static char *	tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
253 		    void *ip4hdr, const void *ip6hdr);
254 
255 
256 static struct tcp_function_block tcp_def_funcblk = {
257 	"default",
258 	tcp_output,
259 	tcp_do_segment,
260 	tcp_default_ctloutput,
261 	NULL,
262 	NULL,
263 	NULL,
264 	NULL,
265 	NULL,
266 	NULL,
267 	0,
268 	0
269 };
270 
271 int t_functions_inited = 0;
272 struct tcp_funchead t_functions;
273 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
274 
275 static void
276 init_tcp_functions(void)
277 {
278 	if (t_functions_inited == 0) {
279 		TAILQ_INIT(&t_functions);
280 		rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
281 		t_functions_inited = 1;
282 	}
283 }
284 
285 static struct tcp_function_block *
286 find_tcp_functions_locked(struct tcp_function_set *fs)
287 {
288 	struct tcp_function *f;
289 	struct tcp_function_block *blk=NULL;
290 
291 	TAILQ_FOREACH(f, &t_functions, tf_next) {
292 		if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
293 			blk = f->tf_fb;
294 			break;
295 		}
296 	}
297 	return(blk);
298 }
299 
300 static struct tcp_function_block *
301 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
302 {
303 	struct tcp_function_block *rblk=NULL;
304 	struct tcp_function *f;
305 
306 	TAILQ_FOREACH(f, &t_functions, tf_next) {
307 		if (f->tf_fb == blk) {
308 			rblk = blk;
309 			if (s) {
310 				*s = f;
311 			}
312 			break;
313 		}
314 	}
315 	return (rblk);
316 }
317 
318 struct tcp_function_block *
319 find_and_ref_tcp_functions(struct tcp_function_set *fs)
320 {
321 	struct tcp_function_block *blk;
322 
323 	rw_rlock(&tcp_function_lock);
324 	blk = find_tcp_functions_locked(fs);
325 	if (blk)
326 		refcount_acquire(&blk->tfb_refcnt);
327 	rw_runlock(&tcp_function_lock);
328 	return(blk);
329 }
330 
331 struct tcp_function_block *
332 find_and_ref_tcp_fb(struct tcp_function_block *blk)
333 {
334 	struct tcp_function_block *rblk;
335 
336 	rw_rlock(&tcp_function_lock);
337 	rblk = find_tcp_fb_locked(blk, NULL);
338 	if (rblk)
339 		refcount_acquire(&rblk->tfb_refcnt);
340 	rw_runlock(&tcp_function_lock);
341 	return(rblk);
342 }
343 
344 
345 static int
346 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
347 {
348 	int error=ENOENT;
349 	struct tcp_function_set fs;
350 	struct tcp_function_block *blk;
351 
352 	memset(&fs, 0, sizeof(fs));
353 	rw_rlock(&tcp_function_lock);
354 	blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
355 	if (blk) {
356 		/* Found him */
357 		strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
358 		fs.pcbcnt = blk->tfb_refcnt;
359 	}
360 	rw_runlock(&tcp_function_lock);
361 	error = sysctl_handle_string(oidp, fs.function_set_name,
362 				     sizeof(fs.function_set_name), req);
363 
364 	/* Check for error or no change */
365 	if (error != 0 || req->newptr == NULL)
366 		return(error);
367 
368 	rw_wlock(&tcp_function_lock);
369 	blk = find_tcp_functions_locked(&fs);
370 	if ((blk == NULL) ||
371 	    (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
372 		error = ENOENT;
373 		goto done;
374 	}
375 	tcp_func_set_ptr = blk;
376 done:
377 	rw_wunlock(&tcp_function_lock);
378 	return (error);
379 }
380 
381 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
382 	    CTLTYPE_STRING | CTLFLAG_RW,
383 	    NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
384 	    "Set/get the default TCP functions");
385 
386 static int
387 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
388 {
389 	int error, cnt, linesz;
390 	struct tcp_function *f;
391 	char *buffer, *cp;
392 	size_t bufsz, outsz;
393 
394 	cnt = 0;
395 	rw_rlock(&tcp_function_lock);
396 	TAILQ_FOREACH(f, &t_functions, tf_next) {
397 		cnt++;
398 	}
399 	rw_runlock(&tcp_function_lock);
400 
401 	bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
402 	buffer = malloc(bufsz, M_TEMP, M_WAITOK);
403 
404 	error = 0;
405 	cp = buffer;
406 
407 	linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
408 	cp += linesz;
409 	bufsz -= linesz;
410 	outsz = linesz;
411 
412 	rw_rlock(&tcp_function_lock);
413 	TAILQ_FOREACH(f, &t_functions, tf_next) {
414 		linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
415 		    f->tf_fb->tfb_tcp_block_name,
416 		    (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
417 		    f->tf_fb->tfb_refcnt);
418 		if (linesz >= bufsz) {
419 			error = EOVERFLOW;
420 			break;
421 		}
422 		cp += linesz;
423 		bufsz -= linesz;
424 		outsz += linesz;
425 	}
426 	rw_runlock(&tcp_function_lock);
427 	if (error == 0)
428 		error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
429 	free(buffer, M_TEMP);
430 	return (error);
431 }
432 
433 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
434 	    CTLTYPE_STRING|CTLFLAG_RD,
435 	    NULL, 0, sysctl_net_inet_list_available, "A",
436 	    "list available TCP Function sets");
437 
438 /*
439  * Target size of TCP PCB hash tables. Must be a power of two.
440  *
441  * Note that this can be overridden by the kernel environment
442  * variable net.inet.tcp.tcbhashsize
443  */
444 #ifndef TCBHASHSIZE
445 #define TCBHASHSIZE	0
446 #endif
447 
448 /*
449  * XXX
450  * Callouts should be moved into struct tcp directly.  They are currently
451  * separate because the tcpcb structure is exported to userland for sysctl
452  * parsing purposes, which do not know about callouts.
453  */
454 struct tcpcb_mem {
455 	struct	tcpcb		tcb;
456 	struct	tcp_timer	tt;
457 	struct	cc_var		ccv;
458 #ifdef TCP_HHOOK
459 	struct	osd		osd;
460 #endif
461 };
462 
463 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
464 #define	V_tcpcb_zone			VNET(tcpcb_zone)
465 
466 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
467 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
468 
469 static struct mtx isn_mtx;
470 
471 #define	ISN_LOCK_INIT()	mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
472 #define	ISN_LOCK()	mtx_lock(&isn_mtx)
473 #define	ISN_UNLOCK()	mtx_unlock(&isn_mtx)
474 
475 /*
476  * TCP initialization.
477  */
478 static void
479 tcp_zone_change(void *tag)
480 {
481 
482 	uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
483 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
484 	tcp_tw_zone_change();
485 }
486 
487 static int
488 tcp_inpcb_init(void *mem, int size, int flags)
489 {
490 	struct inpcb *inp = mem;
491 
492 	INP_LOCK_INIT(inp, "inp", "tcpinp");
493 	return (0);
494 }
495 
496 /*
497  * Take a value and get the next power of 2 that doesn't overflow.
498  * Used to size the tcp_inpcb hash buckets.
499  */
500 static int
501 maketcp_hashsize(int size)
502 {
503 	int hashsize;
504 
505 	/*
506 	 * auto tune.
507 	 * get the next power of 2 higher than maxsockets.
508 	 */
509 	hashsize = 1 << fls(size);
510 	/* catch overflow, and just go one power of 2 smaller */
511 	if (hashsize < size) {
512 		hashsize = 1 << (fls(size) - 1);
513 	}
514 	return (hashsize);
515 }
516 
517 int
518 register_tcp_functions(struct tcp_function_block *blk, int wait)
519 {
520 	struct tcp_function_block *lblk;
521 	struct tcp_function *n;
522 	struct tcp_function_set fs;
523 
524 	if (t_functions_inited == 0) {
525 		init_tcp_functions();
526 	}
527 	if ((blk->tfb_tcp_output == NULL) ||
528 	    (blk->tfb_tcp_do_segment == NULL) ||
529 	    (blk->tfb_tcp_ctloutput == NULL) ||
530 	    (strlen(blk->tfb_tcp_block_name) == 0)) {
531 		/*
532 		 * These functions are required and you
533 		 * need a name.
534 		 */
535 		return (EINVAL);
536 	}
537 	if (blk->tfb_tcp_timer_stop_all ||
538 	    blk->tfb_tcp_timer_activate ||
539 	    blk->tfb_tcp_timer_active ||
540 	    blk->tfb_tcp_timer_stop) {
541 		/*
542 		 * If you define one timer function you
543 		 * must have them all.
544 		 */
545 		if ((blk->tfb_tcp_timer_stop_all == NULL) ||
546 		    (blk->tfb_tcp_timer_activate == NULL) ||
547 		    (blk->tfb_tcp_timer_active == NULL) ||
548 		    (blk->tfb_tcp_timer_stop == NULL)) {
549 			return (EINVAL);
550 		}
551 	}
552 	n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
553 	if (n == NULL) {
554 		return (ENOMEM);
555 	}
556 	n->tf_fb = blk;
557 	strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
558 	rw_wlock(&tcp_function_lock);
559 	lblk = find_tcp_functions_locked(&fs);
560 	if (lblk) {
561 		/* Duplicate name space not allowed */
562 		rw_wunlock(&tcp_function_lock);
563 		free(n, M_TCPFUNCTIONS);
564 		return (EALREADY);
565 	}
566 	refcount_init(&blk->tfb_refcnt, 0);
567 	blk->tfb_flags = 0;
568 	TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
569 	rw_wunlock(&tcp_function_lock);
570 	return(0);
571 }
572 
573 int
574 deregister_tcp_functions(struct tcp_function_block *blk)
575 {
576 	struct tcp_function_block *lblk;
577 	struct tcp_function *f;
578 	int error=ENOENT;
579 
580 	if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
581 		/* You can't un-register the default */
582 		return (EPERM);
583 	}
584 	rw_wlock(&tcp_function_lock);
585 	if (blk == tcp_func_set_ptr) {
586 		/* You can't free the current default */
587 		rw_wunlock(&tcp_function_lock);
588 		return (EBUSY);
589 	}
590 	if (blk->tfb_refcnt) {
591 		/* Still tcb attached, mark it. */
592 		blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
593 		rw_wunlock(&tcp_function_lock);
594 		return (EBUSY);
595 	}
596 	lblk = find_tcp_fb_locked(blk, &f);
597 	if (lblk) {
598 		/* Found */
599 		TAILQ_REMOVE(&t_functions, f, tf_next);
600 		f->tf_fb = NULL;
601 		free(f, M_TCPFUNCTIONS);
602 		error = 0;
603 	}
604 	rw_wunlock(&tcp_function_lock);
605 	return (error);
606 }
607 
608 void
609 tcp_init(void)
610 {
611 	const char *tcbhash_tuneable;
612 	int hashsize;
613 
614 	tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
615 
616 #ifdef TCP_HHOOK
617 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
618 	    &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
619 		printf("%s: WARNING: unable to register helper hook\n", __func__);
620 	if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
621 	    &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
622 		printf("%s: WARNING: unable to register helper hook\n", __func__);
623 #endif
624 	hashsize = TCBHASHSIZE;
625 	TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
626 	if (hashsize == 0) {
627 		/*
628 		 * Auto tune the hash size based on maxsockets.
629 		 * A perfect hash would have a 1:1 mapping
630 		 * (hashsize = maxsockets) however it's been
631 		 * suggested that O(2) average is better.
632 		 */
633 		hashsize = maketcp_hashsize(maxsockets / 4);
634 		/*
635 		 * Our historical default is 512,
636 		 * do not autotune lower than this.
637 		 */
638 		if (hashsize < 512)
639 			hashsize = 512;
640 		if (bootverbose && IS_DEFAULT_VNET(curvnet))
641 			printf("%s: %s auto tuned to %d\n", __func__,
642 			    tcbhash_tuneable, hashsize);
643 	}
644 	/*
645 	 * We require a hashsize to be a power of two.
646 	 * Previously if it was not a power of two we would just reset it
647 	 * back to 512, which could be a nasty surprise if you did not notice
648 	 * the error message.
649 	 * Instead what we do is clip it to the closest power of two lower
650 	 * than the specified hash value.
651 	 */
652 	if (!powerof2(hashsize)) {
653 		int oldhashsize = hashsize;
654 
655 		hashsize = maketcp_hashsize(hashsize);
656 		/* prevent absurdly low value */
657 		if (hashsize < 16)
658 			hashsize = 16;
659 		printf("%s: WARNING: TCB hash size not a power of 2, "
660 		    "clipped from %d to %d.\n", __func__, oldhashsize,
661 		    hashsize);
662 	}
663 	in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
664 	    "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
665 
666 	/*
667 	 * These have to be type stable for the benefit of the timers.
668 	 */
669 	V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
670 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
671 	uma_zone_set_max(V_tcpcb_zone, maxsockets);
672 	uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
673 
674 	tcp_tw_init();
675 	syncache_init();
676 	tcp_hc_init();
677 
678 	TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
679 	V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
680 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
681 
682 	/* Skip initialization of globals for non-default instances. */
683 	if (!IS_DEFAULT_VNET(curvnet))
684 		return;
685 
686 	tcp_reass_global_init();
687 
688 	/* XXX virtualize those bellow? */
689 	tcp_delacktime = TCPTV_DELACK;
690 	tcp_keepinit = TCPTV_KEEP_INIT;
691 	tcp_keepidle = TCPTV_KEEP_IDLE;
692 	tcp_keepintvl = TCPTV_KEEPINTVL;
693 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
694 	tcp_msl = TCPTV_MSL;
695 	tcp_rexmit_min = TCPTV_MIN;
696 	if (tcp_rexmit_min < 1)
697 		tcp_rexmit_min = 1;
698 	tcp_persmin = TCPTV_PERSMIN;
699 	tcp_persmax = TCPTV_PERSMAX;
700 	tcp_rexmit_slop = TCPTV_CPU_VAR;
701 	tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
702 	tcp_tcbhashsize = hashsize;
703 	/* Setup the tcp function block list */
704 	init_tcp_functions();
705 	register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
706 
707 	if (tcp_soreceive_stream) {
708 #ifdef INET
709 		tcp_usrreqs.pru_soreceive = soreceive_stream;
710 #endif
711 #ifdef INET6
712 		tcp6_usrreqs.pru_soreceive = soreceive_stream;
713 #endif /* INET6 */
714 	}
715 
716 #ifdef INET6
717 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
718 #else /* INET6 */
719 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
720 #endif /* INET6 */
721 	if (max_protohdr < TCP_MINPROTOHDR)
722 		max_protohdr = TCP_MINPROTOHDR;
723 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
724 		panic("tcp_init");
725 #undef TCP_MINPROTOHDR
726 
727 	ISN_LOCK_INIT();
728 	EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
729 		SHUTDOWN_PRI_DEFAULT);
730 	EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
731 		EVENTHANDLER_PRI_ANY);
732 #ifdef TCPPCAP
733 	tcp_pcap_init();
734 #endif
735 
736 #ifdef TCP_RFC7413
737 	tcp_fastopen_init();
738 #endif
739 }
740 
741 #ifdef VIMAGE
742 static void
743 tcp_destroy(void *unused __unused)
744 {
745 	int n;
746 #ifdef TCP_HHOOK
747 	int error;
748 #endif
749 
750 	/*
751 	 * All our processes are gone, all our sockets should be cleaned
752 	 * up, which means, we should be past the tcp_discardcb() calls.
753 	 * Sleep to let all tcpcb timers really disappear and cleanup.
754 	 */
755 	for (;;) {
756 		INP_LIST_RLOCK(&V_tcbinfo);
757 		n = V_tcbinfo.ipi_count;
758 		INP_LIST_RUNLOCK(&V_tcbinfo);
759 		if (n == 0)
760 			break;
761 		pause("tcpdes", hz / 10);
762 	}
763 	tcp_hc_destroy();
764 	syncache_destroy();
765 	tcp_tw_destroy();
766 	in_pcbinfo_destroy(&V_tcbinfo);
767 	/* tcp_discardcb() clears the sack_holes up. */
768 	uma_zdestroy(V_sack_hole_zone);
769 	uma_zdestroy(V_tcpcb_zone);
770 
771 #ifdef TCP_RFC7413
772 	/*
773 	 * Cannot free the zone until all tcpcbs are released as we attach
774 	 * the allocations to them.
775 	 */
776 	tcp_fastopen_destroy();
777 #endif
778 
779 #ifdef TCP_HHOOK
780 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
781 	if (error != 0) {
782 		printf("%s: WARNING: unable to deregister helper hook "
783 		    "type=%d, id=%d: error %d returned\n", __func__,
784 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
785 	}
786 	error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
787 	if (error != 0) {
788 		printf("%s: WARNING: unable to deregister helper hook "
789 		    "type=%d, id=%d: error %d returned\n", __func__,
790 		    HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
791 	}
792 #endif
793 }
794 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
795 #endif
796 
797 void
798 tcp_fini(void *xtp)
799 {
800 
801 }
802 
803 /*
804  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
805  * tcp_template used to store this data in mbufs, but we now recopy it out
806  * of the tcpcb each time to conserve mbufs.
807  */
808 void
809 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
810 {
811 	struct tcphdr *th = (struct tcphdr *)tcp_ptr;
812 
813 	INP_WLOCK_ASSERT(inp);
814 
815 #ifdef INET6
816 	if ((inp->inp_vflag & INP_IPV6) != 0) {
817 		struct ip6_hdr *ip6;
818 
819 		ip6 = (struct ip6_hdr *)ip_ptr;
820 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
821 			(inp->inp_flow & IPV6_FLOWINFO_MASK);
822 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
823 			(IPV6_VERSION & IPV6_VERSION_MASK);
824 		ip6->ip6_nxt = IPPROTO_TCP;
825 		ip6->ip6_plen = htons(sizeof(struct tcphdr));
826 		ip6->ip6_src = inp->in6p_laddr;
827 		ip6->ip6_dst = inp->in6p_faddr;
828 	}
829 #endif /* INET6 */
830 #if defined(INET6) && defined(INET)
831 	else
832 #endif
833 #ifdef INET
834 	{
835 		struct ip *ip;
836 
837 		ip = (struct ip *)ip_ptr;
838 		ip->ip_v = IPVERSION;
839 		ip->ip_hl = 5;
840 		ip->ip_tos = inp->inp_ip_tos;
841 		ip->ip_len = 0;
842 		ip->ip_id = 0;
843 		ip->ip_off = 0;
844 		ip->ip_ttl = inp->inp_ip_ttl;
845 		ip->ip_sum = 0;
846 		ip->ip_p = IPPROTO_TCP;
847 		ip->ip_src = inp->inp_laddr;
848 		ip->ip_dst = inp->inp_faddr;
849 	}
850 #endif /* INET */
851 	th->th_sport = inp->inp_lport;
852 	th->th_dport = inp->inp_fport;
853 	th->th_seq = 0;
854 	th->th_ack = 0;
855 	th->th_x2 = 0;
856 	th->th_off = 5;
857 	th->th_flags = 0;
858 	th->th_win = 0;
859 	th->th_urp = 0;
860 	th->th_sum = 0;		/* in_pseudo() is called later for ipv4 */
861 }
862 
863 /*
864  * Create template to be used to send tcp packets on a connection.
865  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
866  * use for this function is in keepalives, which use tcp_respond.
867  */
868 struct tcptemp *
869 tcpip_maketemplate(struct inpcb *inp)
870 {
871 	struct tcptemp *t;
872 
873 	t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
874 	if (t == NULL)
875 		return (NULL);
876 	tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
877 	return (t);
878 }
879 
880 /*
881  * Send a single message to the TCP at address specified by
882  * the given TCP/IP header.  If m == NULL, then we make a copy
883  * of the tcpiphdr at th and send directly to the addressed host.
884  * This is used to force keep alive messages out using the TCP
885  * template for a connection.  If flags are given then we send
886  * a message back to the TCP which originated the segment th,
887  * and discard the mbuf containing it and any other attached mbufs.
888  *
889  * In any case the ack and sequence number of the transmitted
890  * segment are as specified by the parameters.
891  *
892  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
893  */
894 void
895 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
896     tcp_seq ack, tcp_seq seq, int flags)
897 {
898 	struct tcpopt to;
899 	struct inpcb *inp;
900 	struct ip *ip;
901 	struct mbuf *optm;
902 	struct tcphdr *nth;
903 	u_char *optp;
904 #ifdef INET6
905 	struct ip6_hdr *ip6;
906 	int isipv6;
907 #endif /* INET6 */
908 	int optlen, tlen, win;
909 	bool incl_opts;
910 
911 	KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
912 
913 #ifdef INET6
914 	isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
915 	ip6 = ipgen;
916 #endif /* INET6 */
917 	ip = ipgen;
918 
919 	if (tp != NULL) {
920 		inp = tp->t_inpcb;
921 		KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
922 		INP_WLOCK_ASSERT(inp);
923 	} else
924 		inp = NULL;
925 
926 	incl_opts = false;
927 	win = 0;
928 	if (tp != NULL) {
929 		if (!(flags & TH_RST)) {
930 			win = sbspace(&inp->inp_socket->so_rcv);
931 			if (win > TCP_MAXWIN << tp->rcv_scale)
932 				win = TCP_MAXWIN << tp->rcv_scale;
933 		}
934 		if ((tp->t_flags & TF_NOOPT) == 0)
935 			incl_opts = true;
936 	}
937 	if (m == NULL) {
938 		m = m_gethdr(M_NOWAIT, MT_DATA);
939 		if (m == NULL)
940 			return;
941 		m->m_data += max_linkhdr;
942 #ifdef INET6
943 		if (isipv6) {
944 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
945 			      sizeof(struct ip6_hdr));
946 			ip6 = mtod(m, struct ip6_hdr *);
947 			nth = (struct tcphdr *)(ip6 + 1);
948 		} else
949 #endif /* INET6 */
950 		{
951 			bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
952 			ip = mtod(m, struct ip *);
953 			nth = (struct tcphdr *)(ip + 1);
954 		}
955 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
956 		flags = TH_ACK;
957 	} else if (!M_WRITABLE(m)) {
958 		struct mbuf *n;
959 
960 		/* Can't reuse 'm', allocate a new mbuf. */
961 		n = m_gethdr(M_NOWAIT, MT_DATA);
962 		if (n == NULL) {
963 			m_freem(m);
964 			return;
965 		}
966 
967 		if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
968 			m_freem(m);
969 			m_freem(n);
970 			return;
971 		}
972 
973 		n->m_data += max_linkhdr;
974 		/* m_len is set later */
975 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
976 #ifdef INET6
977 		if (isipv6) {
978 			bcopy((caddr_t)ip6, mtod(n, caddr_t),
979 			      sizeof(struct ip6_hdr));
980 			ip6 = mtod(n, struct ip6_hdr *);
981 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
982 			nth = (struct tcphdr *)(ip6 + 1);
983 		} else
984 #endif /* INET6 */
985 		{
986 			bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
987 			ip = mtod(n, struct ip *);
988 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
989 			nth = (struct tcphdr *)(ip + 1);
990 		}
991 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
992 		xchg(nth->th_dport, nth->th_sport, uint16_t);
993 		th = nth;
994 		m_freem(m);
995 		m = n;
996 	} else {
997 		/*
998 		 *  reuse the mbuf.
999 		 * XXX MRT We inherit the FIB, which is lucky.
1000 		 */
1001 		m_freem(m->m_next);
1002 		m->m_next = NULL;
1003 		m->m_data = (caddr_t)ipgen;
1004 		/* m_len is set later */
1005 #ifdef INET6
1006 		if (isipv6) {
1007 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1008 			nth = (struct tcphdr *)(ip6 + 1);
1009 		} else
1010 #endif /* INET6 */
1011 		{
1012 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1013 			nth = (struct tcphdr *)(ip + 1);
1014 		}
1015 		if (th != nth) {
1016 			/*
1017 			 * this is usually a case when an extension header
1018 			 * exists between the IPv6 header and the
1019 			 * TCP header.
1020 			 */
1021 			nth->th_sport = th->th_sport;
1022 			nth->th_dport = th->th_dport;
1023 		}
1024 		xchg(nth->th_dport, nth->th_sport, uint16_t);
1025 #undef xchg
1026 	}
1027 	tlen = 0;
1028 #ifdef INET6
1029 	if (isipv6)
1030 		tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1031 #endif
1032 #if defined(INET) && defined(INET6)
1033 	else
1034 #endif
1035 #ifdef INET
1036 		tlen = sizeof (struct tcpiphdr);
1037 #endif
1038 #ifdef INVARIANTS
1039 	m->m_len = 0;
1040 	KASSERT(M_TRAILINGSPACE(m) >= tlen,
1041 	    ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1042 	    m, tlen, (long)M_TRAILINGSPACE(m)));
1043 #endif
1044 	m->m_len = tlen;
1045 	to.to_flags = 0;
1046 	if (incl_opts) {
1047 		/* Make sure we have room. */
1048 		if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1049 			m->m_next = m_get(M_NOWAIT, MT_DATA);
1050 			if (m->m_next) {
1051 				optp = mtod(m->m_next, u_char *);
1052 				optm = m->m_next;
1053 			} else
1054 				incl_opts = false;
1055 		} else {
1056 			optp = (u_char *) (nth + 1);
1057 			optm = m;
1058 		}
1059 	}
1060 	if (incl_opts) {
1061 		/* Timestamps. */
1062 		if (tp->t_flags & TF_RCVD_TSTMP) {
1063 			to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1064 			to.to_tsecr = tp->ts_recent;
1065 			to.to_flags |= TOF_TS;
1066 		}
1067 #ifdef TCP_SIGNATURE
1068 		/* TCP-MD5 (RFC2385). */
1069 		if (tp->t_flags & TF_SIGNATURE)
1070 			to.to_flags |= TOF_SIGNATURE;
1071 #endif
1072 
1073 		/* Add the options. */
1074 		tlen += optlen = tcp_addoptions(&to, optp);
1075 
1076 		/* Update m_len in the correct mbuf. */
1077 		optm->m_len += optlen;
1078 	} else
1079 		optlen = 0;
1080 #ifdef INET6
1081 	if (isipv6) {
1082 		ip6->ip6_flow = 0;
1083 		ip6->ip6_vfc = IPV6_VERSION;
1084 		ip6->ip6_nxt = IPPROTO_TCP;
1085 		ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1086 	}
1087 #endif
1088 #if defined(INET) && defined(INET6)
1089 	else
1090 #endif
1091 #ifdef INET
1092 	{
1093 		ip->ip_len = htons(tlen);
1094 		ip->ip_ttl = V_ip_defttl;
1095 		if (V_path_mtu_discovery)
1096 			ip->ip_off |= htons(IP_DF);
1097 	}
1098 #endif
1099 	m->m_pkthdr.len = tlen;
1100 	m->m_pkthdr.rcvif = NULL;
1101 #ifdef MAC
1102 	if (inp != NULL) {
1103 		/*
1104 		 * Packet is associated with a socket, so allow the
1105 		 * label of the response to reflect the socket label.
1106 		 */
1107 		INP_WLOCK_ASSERT(inp);
1108 		mac_inpcb_create_mbuf(inp, m);
1109 	} else {
1110 		/*
1111 		 * Packet is not associated with a socket, so possibly
1112 		 * update the label in place.
1113 		 */
1114 		mac_netinet_tcp_reply(m);
1115 	}
1116 #endif
1117 	nth->th_seq = htonl(seq);
1118 	nth->th_ack = htonl(ack);
1119 	nth->th_x2 = 0;
1120 	nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1121 	nth->th_flags = flags;
1122 	if (tp != NULL)
1123 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1124 	else
1125 		nth->th_win = htons((u_short)win);
1126 	nth->th_urp = 0;
1127 
1128 #ifdef TCP_SIGNATURE
1129 	if (to.to_flags & TOF_SIGNATURE) {
1130 		tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1131 		    IPSEC_DIR_OUTBOUND);
1132 	}
1133 #endif
1134 
1135 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1136 #ifdef INET6
1137 	if (isipv6) {
1138 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1139 		nth->th_sum = in6_cksum_pseudo(ip6,
1140 		    tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1141 		ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1142 		    NULL, NULL);
1143 	}
1144 #endif /* INET6 */
1145 #if defined(INET6) && defined(INET)
1146 	else
1147 #endif
1148 #ifdef INET
1149 	{
1150 		m->m_pkthdr.csum_flags = CSUM_TCP;
1151 		nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1152 		    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1153 	}
1154 #endif /* INET */
1155 #ifdef TCPDEBUG
1156 	if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1157 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1158 #endif
1159 	TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1160 	if (flags & TH_RST)
1161 		TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1162 		    tp, nth);
1163 
1164 	TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1165 #ifdef INET6
1166 	if (isipv6)
1167 		(void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1168 #endif /* INET6 */
1169 #if defined(INET) && defined(INET6)
1170 	else
1171 #endif
1172 #ifdef INET
1173 		(void) ip_output(m, NULL, NULL, 0, NULL, inp);
1174 #endif
1175 }
1176 
1177 /*
1178  * Create a new TCP control block, making an
1179  * empty reassembly queue and hooking it to the argument
1180  * protocol control block.  The `inp' parameter must have
1181  * come from the zone allocator set up in tcp_init().
1182  */
1183 struct tcpcb *
1184 tcp_newtcpcb(struct inpcb *inp)
1185 {
1186 	struct tcpcb_mem *tm;
1187 	struct tcpcb *tp;
1188 #ifdef INET6
1189 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1190 #endif /* INET6 */
1191 
1192 	tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1193 	if (tm == NULL)
1194 		return (NULL);
1195 	tp = &tm->tcb;
1196 
1197 	/* Initialise cc_var struct for this tcpcb. */
1198 	tp->ccv = &tm->ccv;
1199 	tp->ccv->type = IPPROTO_TCP;
1200 	tp->ccv->ccvc.tcp = tp;
1201 	rw_rlock(&tcp_function_lock);
1202 	tp->t_fb = tcp_func_set_ptr;
1203 	refcount_acquire(&tp->t_fb->tfb_refcnt);
1204 	rw_runlock(&tcp_function_lock);
1205 	/*
1206 	 * Use the current system default CC algorithm.
1207 	 */
1208 	CC_LIST_RLOCK();
1209 	KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1210 	CC_ALGO(tp) = CC_DEFAULT();
1211 	CC_LIST_RUNLOCK();
1212 
1213 	if (CC_ALGO(tp)->cb_init != NULL)
1214 		if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1215 			if (tp->t_fb->tfb_tcp_fb_fini)
1216 				(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1217 			refcount_release(&tp->t_fb->tfb_refcnt);
1218 			uma_zfree(V_tcpcb_zone, tm);
1219 			return (NULL);
1220 		}
1221 
1222 #ifdef TCP_HHOOK
1223 	tp->osd = &tm->osd;
1224 	if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1225 		if (tp->t_fb->tfb_tcp_fb_fini)
1226 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1227 		refcount_release(&tp->t_fb->tfb_refcnt);
1228 		uma_zfree(V_tcpcb_zone, tm);
1229 		return (NULL);
1230 	}
1231 #endif
1232 
1233 #ifdef VIMAGE
1234 	tp->t_vnet = inp->inp_vnet;
1235 #endif
1236 	tp->t_timers = &tm->tt;
1237 	/*	LIST_INIT(&tp->t_segq); */	/* XXX covered by M_ZERO */
1238 	tp->t_maxseg =
1239 #ifdef INET6
1240 		isipv6 ? V_tcp_v6mssdflt :
1241 #endif /* INET6 */
1242 		V_tcp_mssdflt;
1243 
1244 	/* Set up our timeouts. */
1245 	callout_init(&tp->t_timers->tt_rexmt, 1);
1246 	callout_init(&tp->t_timers->tt_persist, 1);
1247 	callout_init(&tp->t_timers->tt_keep, 1);
1248 	callout_init(&tp->t_timers->tt_2msl, 1);
1249 	callout_init(&tp->t_timers->tt_delack, 1);
1250 
1251 	if (V_tcp_do_rfc1323)
1252 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1253 	if (V_tcp_do_sack)
1254 		tp->t_flags |= TF_SACK_PERMIT;
1255 	TAILQ_INIT(&tp->snd_holes);
1256 	/*
1257 	 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1258 	 * is called.
1259 	 */
1260 	in_pcbref(inp);	/* Reference for tcpcb */
1261 	tp->t_inpcb = inp;
1262 
1263 	/*
1264 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1265 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1266 	 * reasonable initial retransmit time.
1267 	 */
1268 	tp->t_srtt = TCPTV_SRTTBASE;
1269 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1270 	tp->t_rttmin = tcp_rexmit_min;
1271 	tp->t_rxtcur = TCPTV_RTOBASE;
1272 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1273 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1274 	tp->t_rcvtime = ticks;
1275 	/*
1276 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1277 	 * because the socket may be bound to an IPv6 wildcard address,
1278 	 * which may match an IPv4-mapped IPv6 address.
1279 	 */
1280 	inp->inp_ip_ttl = V_ip_defttl;
1281 	inp->inp_ppcb = tp;
1282 #ifdef TCPPCAP
1283 	/*
1284 	 * Init the TCP PCAP queues.
1285 	 */
1286 	tcp_pcap_tcpcb_init(tp);
1287 #endif
1288 	if (tp->t_fb->tfb_tcp_fb_init) {
1289 		(*tp->t_fb->tfb_tcp_fb_init)(tp);
1290 	}
1291 	return (tp);		/* XXX */
1292 }
1293 
1294 /*
1295  * Switch the congestion control algorithm back to NewReno for any active
1296  * control blocks using an algorithm which is about to go away.
1297  * This ensures the CC framework can allow the unload to proceed without leaving
1298  * any dangling pointers which would trigger a panic.
1299  * Returning non-zero would inform the CC framework that something went wrong
1300  * and it would be unsafe to allow the unload to proceed. However, there is no
1301  * way for this to occur with this implementation so we always return zero.
1302  */
1303 int
1304 tcp_ccalgounload(struct cc_algo *unload_algo)
1305 {
1306 	struct cc_algo *tmpalgo;
1307 	struct inpcb *inp;
1308 	struct tcpcb *tp;
1309 	VNET_ITERATOR_DECL(vnet_iter);
1310 
1311 	/*
1312 	 * Check all active control blocks across all network stacks and change
1313 	 * any that are using "unload_algo" back to NewReno. If "unload_algo"
1314 	 * requires cleanup code to be run, call it.
1315 	 */
1316 	VNET_LIST_RLOCK();
1317 	VNET_FOREACH(vnet_iter) {
1318 		CURVNET_SET(vnet_iter);
1319 		INP_INFO_WLOCK(&V_tcbinfo);
1320 		/*
1321 		 * New connections already part way through being initialised
1322 		 * with the CC algo we're removing will not race with this code
1323 		 * because the INP_INFO_WLOCK is held during initialisation. We
1324 		 * therefore don't enter the loop below until the connection
1325 		 * list has stabilised.
1326 		 */
1327 		LIST_FOREACH(inp, &V_tcb, inp_list) {
1328 			INP_WLOCK(inp);
1329 			/* Important to skip tcptw structs. */
1330 			if (!(inp->inp_flags & INP_TIMEWAIT) &&
1331 			    (tp = intotcpcb(inp)) != NULL) {
1332 				/*
1333 				 * By holding INP_WLOCK here, we are assured
1334 				 * that the connection is not currently
1335 				 * executing inside the CC module's functions
1336 				 * i.e. it is safe to make the switch back to
1337 				 * NewReno.
1338 				 */
1339 				if (CC_ALGO(tp) == unload_algo) {
1340 					tmpalgo = CC_ALGO(tp);
1341 					/* NewReno does not require any init. */
1342 					CC_ALGO(tp) = &newreno_cc_algo;
1343 					if (tmpalgo->cb_destroy != NULL)
1344 						tmpalgo->cb_destroy(tp->ccv);
1345 				}
1346 			}
1347 			INP_WUNLOCK(inp);
1348 		}
1349 		INP_INFO_WUNLOCK(&V_tcbinfo);
1350 		CURVNET_RESTORE();
1351 	}
1352 	VNET_LIST_RUNLOCK();
1353 
1354 	return (0);
1355 }
1356 
1357 /*
1358  * Drop a TCP connection, reporting
1359  * the specified error.  If connection is synchronized,
1360  * then send a RST to peer.
1361  */
1362 struct tcpcb *
1363 tcp_drop(struct tcpcb *tp, int errno)
1364 {
1365 	struct socket *so = tp->t_inpcb->inp_socket;
1366 
1367 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1368 	INP_WLOCK_ASSERT(tp->t_inpcb);
1369 
1370 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
1371 		tcp_state_change(tp, TCPS_CLOSED);
1372 		(void) tp->t_fb->tfb_tcp_output(tp);
1373 		TCPSTAT_INC(tcps_drops);
1374 	} else
1375 		TCPSTAT_INC(tcps_conndrops);
1376 	if (errno == ETIMEDOUT && tp->t_softerror)
1377 		errno = tp->t_softerror;
1378 	so->so_error = errno;
1379 	return (tcp_close(tp));
1380 }
1381 
1382 void
1383 tcp_discardcb(struct tcpcb *tp)
1384 {
1385 	struct inpcb *inp = tp->t_inpcb;
1386 	struct socket *so = inp->inp_socket;
1387 #ifdef INET6
1388 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1389 #endif /* INET6 */
1390 	int released;
1391 
1392 	INP_WLOCK_ASSERT(inp);
1393 
1394 	/*
1395 	 * Make sure that all of our timers are stopped before we delete the
1396 	 * PCB.
1397 	 *
1398 	 * If stopping a timer fails, we schedule a discard function in same
1399 	 * callout, and the last discard function called will take care of
1400 	 * deleting the tcpcb.
1401 	 */
1402 	tp->t_timers->tt_draincnt = 0;
1403 	tcp_timer_stop(tp, TT_REXMT);
1404 	tcp_timer_stop(tp, TT_PERSIST);
1405 	tcp_timer_stop(tp, TT_KEEP);
1406 	tcp_timer_stop(tp, TT_2MSL);
1407 	tcp_timer_stop(tp, TT_DELACK);
1408 	if (tp->t_fb->tfb_tcp_timer_stop_all) {
1409 		/*
1410 		 * Call the stop-all function of the methods,
1411 		 * this function should call the tcp_timer_stop()
1412 		 * method with each of the function specific timeouts.
1413 		 * That stop will be called via the tfb_tcp_timer_stop()
1414 		 * which should use the async drain function of the
1415 		 * callout system (see tcp_var.h).
1416 		 */
1417 		tp->t_fb->tfb_tcp_timer_stop_all(tp);
1418 	}
1419 
1420 	/*
1421 	 * If we got enough samples through the srtt filter,
1422 	 * save the rtt and rttvar in the routing entry.
1423 	 * 'Enough' is arbitrarily defined as 4 rtt samples.
1424 	 * 4 samples is enough for the srtt filter to converge
1425 	 * to within enough % of the correct value; fewer samples
1426 	 * and we could save a bogus rtt. The danger is not high
1427 	 * as tcp quickly recovers from everything.
1428 	 * XXX: Works very well but needs some more statistics!
1429 	 */
1430 	if (tp->t_rttupdated >= 4) {
1431 		struct hc_metrics_lite metrics;
1432 		uint32_t ssthresh;
1433 
1434 		bzero(&metrics, sizeof(metrics));
1435 		/*
1436 		 * Update the ssthresh always when the conditions below
1437 		 * are satisfied. This gives us better new start value
1438 		 * for the congestion avoidance for new connections.
1439 		 * ssthresh is only set if packet loss occurred on a session.
1440 		 *
1441 		 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1442 		 * being torn down.  Ideally this code would not use 'so'.
1443 		 */
1444 		ssthresh = tp->snd_ssthresh;
1445 		if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1446 			/*
1447 			 * convert the limit from user data bytes to
1448 			 * packets then to packet data bytes.
1449 			 */
1450 			ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1451 			if (ssthresh < 2)
1452 				ssthresh = 2;
1453 			ssthresh *= (tp->t_maxseg +
1454 #ifdef INET6
1455 			    (isipv6 ? sizeof (struct ip6_hdr) +
1456 				sizeof (struct tcphdr) :
1457 #endif
1458 				sizeof (struct tcpiphdr)
1459 #ifdef INET6
1460 			    )
1461 #endif
1462 			    );
1463 		} else
1464 			ssthresh = 0;
1465 		metrics.rmx_ssthresh = ssthresh;
1466 
1467 		metrics.rmx_rtt = tp->t_srtt;
1468 		metrics.rmx_rttvar = tp->t_rttvar;
1469 		metrics.rmx_cwnd = tp->snd_cwnd;
1470 		metrics.rmx_sendpipe = 0;
1471 		metrics.rmx_recvpipe = 0;
1472 
1473 		tcp_hc_update(&inp->inp_inc, &metrics);
1474 	}
1475 
1476 	/* free the reassembly queue, if any */
1477 	tcp_reass_flush(tp);
1478 
1479 #ifdef TCP_OFFLOAD
1480 	/* Disconnect offload device, if any. */
1481 	if (tp->t_flags & TF_TOE)
1482 		tcp_offload_detach(tp);
1483 #endif
1484 
1485 	tcp_free_sackholes(tp);
1486 
1487 #ifdef TCPPCAP
1488 	/* Free the TCP PCAP queues. */
1489 	tcp_pcap_drain(&(tp->t_inpkts));
1490 	tcp_pcap_drain(&(tp->t_outpkts));
1491 #endif
1492 
1493 	/* Allow the CC algorithm to clean up after itself. */
1494 	if (CC_ALGO(tp)->cb_destroy != NULL)
1495 		CC_ALGO(tp)->cb_destroy(tp->ccv);
1496 
1497 #ifdef TCP_HHOOK
1498 	khelp_destroy_osd(tp->osd);
1499 #endif
1500 
1501 	CC_ALGO(tp) = NULL;
1502 	inp->inp_ppcb = NULL;
1503 	if (tp->t_timers->tt_draincnt == 0) {
1504 		/* We own the last reference on tcpcb, let's free it. */
1505 		TCPSTATES_DEC(tp->t_state);
1506 		if (tp->t_fb->tfb_tcp_fb_fini)
1507 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1508 		refcount_release(&tp->t_fb->tfb_refcnt);
1509 		tp->t_inpcb = NULL;
1510 		uma_zfree(V_tcpcb_zone, tp);
1511 		released = in_pcbrele_wlocked(inp);
1512 		KASSERT(!released, ("%s: inp %p should not have been released "
1513 			"here", __func__, inp));
1514 	}
1515 }
1516 
1517 void
1518 tcp_timer_discard(void *ptp)
1519 {
1520 	struct inpcb *inp;
1521 	struct tcpcb *tp;
1522 
1523 	tp = (struct tcpcb *)ptp;
1524 	CURVNET_SET(tp->t_vnet);
1525 	INP_INFO_RLOCK(&V_tcbinfo);
1526 	inp = tp->t_inpcb;
1527 	KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1528 		__func__, tp));
1529 	INP_WLOCK(inp);
1530 	KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1531 		("%s: tcpcb has to be stopped here", __func__));
1532 	tp->t_timers->tt_draincnt--;
1533 	if (tp->t_timers->tt_draincnt == 0) {
1534 		/* We own the last reference on this tcpcb, let's free it. */
1535 		TCPSTATES_DEC(tp->t_state);
1536 		if (tp->t_fb->tfb_tcp_fb_fini)
1537 			(*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1538 		refcount_release(&tp->t_fb->tfb_refcnt);
1539 		tp->t_inpcb = NULL;
1540 		uma_zfree(V_tcpcb_zone, tp);
1541 		if (in_pcbrele_wlocked(inp)) {
1542 			INP_INFO_RUNLOCK(&V_tcbinfo);
1543 			CURVNET_RESTORE();
1544 			return;
1545 		}
1546 	}
1547 	INP_WUNLOCK(inp);
1548 	INP_INFO_RUNLOCK(&V_tcbinfo);
1549 	CURVNET_RESTORE();
1550 }
1551 
1552 /*
1553  * Attempt to close a TCP control block, marking it as dropped, and freeing
1554  * the socket if we hold the only reference.
1555  */
1556 struct tcpcb *
1557 tcp_close(struct tcpcb *tp)
1558 {
1559 	struct inpcb *inp = tp->t_inpcb;
1560 	struct socket *so;
1561 
1562 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1563 	INP_WLOCK_ASSERT(inp);
1564 
1565 #ifdef TCP_OFFLOAD
1566 	if (tp->t_state == TCPS_LISTEN)
1567 		tcp_offload_listen_stop(tp);
1568 #endif
1569 #ifdef TCP_RFC7413
1570 	/*
1571 	 * This releases the TFO pending counter resource for TFO listen
1572 	 * sockets as well as passively-created TFO sockets that transition
1573 	 * from SYN_RECEIVED to CLOSED.
1574 	 */
1575 	if (tp->t_tfo_pending) {
1576 		tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1577 		tp->t_tfo_pending = NULL;
1578 	}
1579 #endif
1580 	in_pcbdrop(inp);
1581 	TCPSTAT_INC(tcps_closed);
1582 	if (tp->t_state != TCPS_CLOSED)
1583 		tcp_state_change(tp, TCPS_CLOSED);
1584 	KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1585 	so = inp->inp_socket;
1586 	soisdisconnected(so);
1587 	if (inp->inp_flags & INP_SOCKREF) {
1588 		KASSERT(so->so_state & SS_PROTOREF,
1589 		    ("tcp_close: !SS_PROTOREF"));
1590 		inp->inp_flags &= ~INP_SOCKREF;
1591 		INP_WUNLOCK(inp);
1592 		ACCEPT_LOCK();
1593 		SOCK_LOCK(so);
1594 		so->so_state &= ~SS_PROTOREF;
1595 		sofree(so);
1596 		return (NULL);
1597 	}
1598 	return (tp);
1599 }
1600 
1601 void
1602 tcp_drain(void)
1603 {
1604 	VNET_ITERATOR_DECL(vnet_iter);
1605 
1606 	if (!do_tcpdrain)
1607 		return;
1608 
1609 	VNET_LIST_RLOCK_NOSLEEP();
1610 	VNET_FOREACH(vnet_iter) {
1611 		CURVNET_SET(vnet_iter);
1612 		struct inpcb *inpb;
1613 		struct tcpcb *tcpb;
1614 
1615 	/*
1616 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
1617 	 * if there is one...
1618 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
1619 	 *      reassembly queue should be flushed, but in a situation
1620 	 *	where we're really low on mbufs, this is potentially
1621 	 *	useful.
1622 	 */
1623 		INP_INFO_WLOCK(&V_tcbinfo);
1624 		LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1625 			if (inpb->inp_flags & INP_TIMEWAIT)
1626 				continue;
1627 			INP_WLOCK(inpb);
1628 			if ((tcpb = intotcpcb(inpb)) != NULL) {
1629 				tcp_reass_flush(tcpb);
1630 				tcp_clean_sackreport(tcpb);
1631 #ifdef TCPPCAP
1632 				if (tcp_pcap_aggressive_free) {
1633 					/* Free the TCP PCAP queues. */
1634 					tcp_pcap_drain(&(tcpb->t_inpkts));
1635 					tcp_pcap_drain(&(tcpb->t_outpkts));
1636 				}
1637 #endif
1638 			}
1639 			INP_WUNLOCK(inpb);
1640 		}
1641 		INP_INFO_WUNLOCK(&V_tcbinfo);
1642 		CURVNET_RESTORE();
1643 	}
1644 	VNET_LIST_RUNLOCK_NOSLEEP();
1645 }
1646 
1647 /*
1648  * Notify a tcp user of an asynchronous error;
1649  * store error as soft error, but wake up user
1650  * (for now, won't do anything until can select for soft error).
1651  *
1652  * Do not wake up user since there currently is no mechanism for
1653  * reporting soft errors (yet - a kqueue filter may be added).
1654  */
1655 static struct inpcb *
1656 tcp_notify(struct inpcb *inp, int error)
1657 {
1658 	struct tcpcb *tp;
1659 
1660 	INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1661 	INP_WLOCK_ASSERT(inp);
1662 
1663 	if ((inp->inp_flags & INP_TIMEWAIT) ||
1664 	    (inp->inp_flags & INP_DROPPED))
1665 		return (inp);
1666 
1667 	tp = intotcpcb(inp);
1668 	KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1669 
1670 	/*
1671 	 * Ignore some errors if we are hooked up.
1672 	 * If connection hasn't completed, has retransmitted several times,
1673 	 * and receives a second error, give up now.  This is better
1674 	 * than waiting a long time to establish a connection that
1675 	 * can never complete.
1676 	 */
1677 	if (tp->t_state == TCPS_ESTABLISHED &&
1678 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1679 	     error == EHOSTDOWN)) {
1680 		if (inp->inp_route.ro_rt) {
1681 			RTFREE(inp->inp_route.ro_rt);
1682 			inp->inp_route.ro_rt = (struct rtentry *)NULL;
1683 		}
1684 		return (inp);
1685 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1686 	    tp->t_softerror) {
1687 		tp = tcp_drop(tp, error);
1688 		if (tp != NULL)
1689 			return (inp);
1690 		else
1691 			return (NULL);
1692 	} else {
1693 		tp->t_softerror = error;
1694 		return (inp);
1695 	}
1696 #if 0
1697 	wakeup( &so->so_timeo);
1698 	sorwakeup(so);
1699 	sowwakeup(so);
1700 #endif
1701 }
1702 
1703 static int
1704 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1705 {
1706 	int error, i, m, n, pcb_count;
1707 	struct inpcb *inp, **inp_list;
1708 	inp_gen_t gencnt;
1709 	struct xinpgen xig;
1710 
1711 	/*
1712 	 * The process of preparing the TCB list is too time-consuming and
1713 	 * resource-intensive to repeat twice on every request.
1714 	 */
1715 	if (req->oldptr == NULL) {
1716 		n = V_tcbinfo.ipi_count +
1717 		    counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1718 		n += imax(n / 8, 10);
1719 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1720 		return (0);
1721 	}
1722 
1723 	if (req->newptr != NULL)
1724 		return (EPERM);
1725 
1726 	/*
1727 	 * OK, now we're committed to doing something.
1728 	 */
1729 	INP_LIST_RLOCK(&V_tcbinfo);
1730 	gencnt = V_tcbinfo.ipi_gencnt;
1731 	n = V_tcbinfo.ipi_count;
1732 	INP_LIST_RUNLOCK(&V_tcbinfo);
1733 
1734 	m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1735 
1736 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1737 		+ (n + m) * sizeof(struct xtcpcb));
1738 	if (error != 0)
1739 		return (error);
1740 
1741 	xig.xig_len = sizeof xig;
1742 	xig.xig_count = n + m;
1743 	xig.xig_gen = gencnt;
1744 	xig.xig_sogen = so_gencnt;
1745 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1746 	if (error)
1747 		return (error);
1748 
1749 	error = syncache_pcblist(req, m, &pcb_count);
1750 	if (error)
1751 		return (error);
1752 
1753 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1754 
1755 	INP_INFO_WLOCK(&V_tcbinfo);
1756 	for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1757 	    inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1758 		INP_WLOCK(inp);
1759 		if (inp->inp_gencnt <= gencnt) {
1760 			/*
1761 			 * XXX: This use of cr_cansee(), introduced with
1762 			 * TCP state changes, is not quite right, but for
1763 			 * now, better than nothing.
1764 			 */
1765 			if (inp->inp_flags & INP_TIMEWAIT) {
1766 				if (intotw(inp) != NULL)
1767 					error = cr_cansee(req->td->td_ucred,
1768 					    intotw(inp)->tw_cred);
1769 				else
1770 					error = EINVAL;	/* Skip this inp. */
1771 			} else
1772 				error = cr_canseeinpcb(req->td->td_ucred, inp);
1773 			if (error == 0) {
1774 				in_pcbref(inp);
1775 				inp_list[i++] = inp;
1776 			}
1777 		}
1778 		INP_WUNLOCK(inp);
1779 	}
1780 	INP_INFO_WUNLOCK(&V_tcbinfo);
1781 	n = i;
1782 
1783 	error = 0;
1784 	for (i = 0; i < n; i++) {
1785 		inp = inp_list[i];
1786 		INP_RLOCK(inp);
1787 		if (inp->inp_gencnt <= gencnt) {
1788 			struct xtcpcb xt;
1789 			void *inp_ppcb;
1790 
1791 			bzero(&xt, sizeof(xt));
1792 			xt.xt_len = sizeof xt;
1793 			/* XXX should avoid extra copy */
1794 			bcopy(inp, &xt.xt_inp, sizeof *inp);
1795 			inp_ppcb = inp->inp_ppcb;
1796 			if (inp_ppcb == NULL)
1797 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1798 			else if (inp->inp_flags & INP_TIMEWAIT) {
1799 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1800 				xt.xt_tp.t_state = TCPS_TIME_WAIT;
1801 			} else {
1802 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1803 				if (xt.xt_tp.t_timers)
1804 					tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1805 			}
1806 			if (inp->inp_socket != NULL)
1807 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
1808 			else {
1809 				bzero(&xt.xt_socket, sizeof xt.xt_socket);
1810 				xt.xt_socket.xso_protocol = IPPROTO_TCP;
1811 			}
1812 			xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1813 			INP_RUNLOCK(inp);
1814 			error = SYSCTL_OUT(req, &xt, sizeof xt);
1815 		} else
1816 			INP_RUNLOCK(inp);
1817 	}
1818 	INP_INFO_RLOCK(&V_tcbinfo);
1819 	for (i = 0; i < n; i++) {
1820 		inp = inp_list[i];
1821 		INP_RLOCK(inp);
1822 		if (!in_pcbrele_rlocked(inp))
1823 			INP_RUNLOCK(inp);
1824 	}
1825 	INP_INFO_RUNLOCK(&V_tcbinfo);
1826 
1827 	if (!error) {
1828 		/*
1829 		 * Give the user an updated idea of our state.
1830 		 * If the generation differs from what we told
1831 		 * her before, she knows that something happened
1832 		 * while we were processing this request, and it
1833 		 * might be necessary to retry.
1834 		 */
1835 		INP_LIST_RLOCK(&V_tcbinfo);
1836 		xig.xig_gen = V_tcbinfo.ipi_gencnt;
1837 		xig.xig_sogen = so_gencnt;
1838 		xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1839 		INP_LIST_RUNLOCK(&V_tcbinfo);
1840 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1841 	}
1842 	free(inp_list, M_TEMP);
1843 	return (error);
1844 }
1845 
1846 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1847     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1848     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1849 
1850 #ifdef INET
1851 static int
1852 tcp_getcred(SYSCTL_HANDLER_ARGS)
1853 {
1854 	struct xucred xuc;
1855 	struct sockaddr_in addrs[2];
1856 	struct inpcb *inp;
1857 	int error;
1858 
1859 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1860 	if (error)
1861 		return (error);
1862 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1863 	if (error)
1864 		return (error);
1865 	inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1866 	    addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1867 	if (inp != NULL) {
1868 		if (inp->inp_socket == NULL)
1869 			error = ENOENT;
1870 		if (error == 0)
1871 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1872 		if (error == 0)
1873 			cru2x(inp->inp_cred, &xuc);
1874 		INP_RUNLOCK(inp);
1875 	} else
1876 		error = ENOENT;
1877 	if (error == 0)
1878 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1879 	return (error);
1880 }
1881 
1882 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1883     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1884     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1885 #endif /* INET */
1886 
1887 #ifdef INET6
1888 static int
1889 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1890 {
1891 	struct xucred xuc;
1892 	struct sockaddr_in6 addrs[2];
1893 	struct inpcb *inp;
1894 	int error;
1895 #ifdef INET
1896 	int mapped = 0;
1897 #endif
1898 
1899 	error = priv_check(req->td, PRIV_NETINET_GETCRED);
1900 	if (error)
1901 		return (error);
1902 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
1903 	if (error)
1904 		return (error);
1905 	if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1906 	    (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1907 		return (error);
1908 	}
1909 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1910 #ifdef INET
1911 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1912 			mapped = 1;
1913 		else
1914 #endif
1915 			return (EINVAL);
1916 	}
1917 
1918 #ifdef INET
1919 	if (mapped == 1)
1920 		inp = in_pcblookup(&V_tcbinfo,
1921 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1922 			addrs[1].sin6_port,
1923 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1924 			addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1925 	else
1926 #endif
1927 		inp = in6_pcblookup(&V_tcbinfo,
1928 			&addrs[1].sin6_addr, addrs[1].sin6_port,
1929 			&addrs[0].sin6_addr, addrs[0].sin6_port,
1930 			INPLOOKUP_RLOCKPCB, NULL);
1931 	if (inp != NULL) {
1932 		if (inp->inp_socket == NULL)
1933 			error = ENOENT;
1934 		if (error == 0)
1935 			error = cr_canseeinpcb(req->td->td_ucred, inp);
1936 		if (error == 0)
1937 			cru2x(inp->inp_cred, &xuc);
1938 		INP_RUNLOCK(inp);
1939 	} else
1940 		error = ENOENT;
1941 	if (error == 0)
1942 		error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1943 	return (error);
1944 }
1945 
1946 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1947     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1948     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1949 #endif /* INET6 */
1950 
1951 
1952 #ifdef INET
1953 void
1954 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1955 {
1956 	struct ip *ip = vip;
1957 	struct tcphdr *th;
1958 	struct in_addr faddr;
1959 	struct inpcb *inp;
1960 	struct tcpcb *tp;
1961 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1962 	struct icmp *icp;
1963 	struct in_conninfo inc;
1964 	tcp_seq icmp_tcp_seq;
1965 	int mtu;
1966 
1967 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1968 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1969 		return;
1970 
1971 	if (cmd == PRC_MSGSIZE)
1972 		notify = tcp_mtudisc_notify;
1973 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1974 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1975 		cmd == PRC_TIMXCEED_INTRANS) && ip)
1976 		notify = tcp_drop_syn_sent;
1977 
1978 	/*
1979 	 * Hostdead is ugly because it goes linearly through all PCBs.
1980 	 * XXX: We never get this from ICMP, otherwise it makes an
1981 	 * excellent DoS attack on machines with many connections.
1982 	 */
1983 	else if (cmd == PRC_HOSTDEAD)
1984 		ip = NULL;
1985 	else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1986 		return;
1987 
1988 	if (ip == NULL) {
1989 		in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1990 		return;
1991 	}
1992 
1993 	icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1994 	th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1995 	INP_INFO_RLOCK(&V_tcbinfo);
1996 	inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1997 	    th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1998 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1999 		/* signal EHOSTDOWN, as it flushes the cached route */
2000 		inp = (*notify)(inp, EHOSTDOWN);
2001 		if (inp != NULL)
2002 			INP_WUNLOCK(inp);
2003 	} else if (inp != NULL)  {
2004 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2005 		    !(inp->inp_flags & INP_DROPPED) &&
2006 		    !(inp->inp_socket == NULL)) {
2007 			icmp_tcp_seq = ntohl(th->th_seq);
2008 			tp = intotcpcb(inp);
2009 			if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2010 			    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2011 				if (cmd == PRC_MSGSIZE) {
2012 					/*
2013 					 * MTU discovery:
2014 					 * If we got a needfrag set the MTU
2015 					 * in the route to the suggested new
2016 					 * value (if given) and then notify.
2017 					 */
2018 				    	mtu = ntohs(icp->icmp_nextmtu);
2019 					/*
2020 					 * If no alternative MTU was
2021 					 * proposed, try the next smaller
2022 					 * one.
2023 					 */
2024 					if (!mtu)
2025 						mtu = ip_next_mtu(
2026 						    ntohs(ip->ip_len), 1);
2027 					if (mtu < V_tcp_minmss +
2028 					    sizeof(struct tcpiphdr))
2029 						mtu = V_tcp_minmss +
2030 						    sizeof(struct tcpiphdr);
2031 					/*
2032 					 * Only process the offered MTU if it
2033 					 * is smaller than the current one.
2034 					 */
2035 					if (mtu < tp->t_maxseg +
2036 					    sizeof(struct tcpiphdr)) {
2037 						bzero(&inc, sizeof(inc));
2038 						inc.inc_faddr = faddr;
2039 						inc.inc_fibnum =
2040 						    inp->inp_inc.inc_fibnum;
2041 						tcp_hc_updatemtu(&inc, mtu);
2042 						tcp_mtudisc(inp, mtu);
2043 					}
2044 				} else
2045 					inp = (*notify)(inp,
2046 					    inetctlerrmap[cmd]);
2047 			}
2048 		}
2049 		if (inp != NULL)
2050 			INP_WUNLOCK(inp);
2051 	} else {
2052 		bzero(&inc, sizeof(inc));
2053 		inc.inc_fport = th->th_dport;
2054 		inc.inc_lport = th->th_sport;
2055 		inc.inc_faddr = faddr;
2056 		inc.inc_laddr = ip->ip_src;
2057 		syncache_unreach(&inc, th);
2058 	}
2059 	INP_INFO_RUNLOCK(&V_tcbinfo);
2060 }
2061 #endif /* INET */
2062 
2063 #ifdef INET6
2064 void
2065 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2066 {
2067 	struct in6_addr *dst;
2068 	struct tcphdr *th;
2069 	struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2070 	struct ip6_hdr *ip6;
2071 	struct mbuf *m;
2072 	struct inpcb *inp;
2073 	struct tcpcb *tp;
2074 	struct icmp6_hdr *icmp6;
2075 	struct ip6ctlparam *ip6cp = NULL;
2076 	const struct sockaddr_in6 *sa6_src = NULL;
2077 	struct in_conninfo inc;
2078 	tcp_seq icmp_tcp_seq;
2079 	unsigned int mtu;
2080 	unsigned int off;
2081 
2082 
2083 	if (sa->sa_family != AF_INET6 ||
2084 	    sa->sa_len != sizeof(struct sockaddr_in6))
2085 		return;
2086 
2087 	/* if the parameter is from icmp6, decode it. */
2088 	if (d != NULL) {
2089 		ip6cp = (struct ip6ctlparam *)d;
2090 		icmp6 = ip6cp->ip6c_icmp6;
2091 		m = ip6cp->ip6c_m;
2092 		ip6 = ip6cp->ip6c_ip6;
2093 		off = ip6cp->ip6c_off;
2094 		sa6_src = ip6cp->ip6c_src;
2095 		dst = ip6cp->ip6c_finaldst;
2096 	} else {
2097 		m = NULL;
2098 		ip6 = NULL;
2099 		off = 0;	/* fool gcc */
2100 		sa6_src = &sa6_any;
2101 		dst = NULL;
2102 	}
2103 
2104 	if (cmd == PRC_MSGSIZE)
2105 		notify = tcp_mtudisc_notify;
2106 	else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2107 		cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2108 		cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2109 		notify = tcp_drop_syn_sent;
2110 
2111 	/*
2112 	 * Hostdead is ugly because it goes linearly through all PCBs.
2113 	 * XXX: We never get this from ICMP, otherwise it makes an
2114 	 * excellent DoS attack on machines with many connections.
2115 	 */
2116 	else if (cmd == PRC_HOSTDEAD)
2117 		ip6 = NULL;
2118 	else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2119 		return;
2120 
2121 	if (ip6 == NULL) {
2122 		in6_pcbnotify(&V_tcbinfo, sa, 0,
2123 			      (const struct sockaddr *)sa6_src,
2124 			      0, cmd, NULL, notify);
2125 		return;
2126 	}
2127 
2128 	/* Check if we can safely get the ports from the tcp hdr */
2129 	if (m == NULL ||
2130 	    (m->m_pkthdr.len <
2131 		(int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2132 		return;
2133 	}
2134 
2135 	th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2136 	INP_INFO_RLOCK(&V_tcbinfo);
2137 	inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2138 	    &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2139 	if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2140 		/* signal EHOSTDOWN, as it flushes the cached route */
2141 		inp = (*notify)(inp, EHOSTDOWN);
2142 		if (inp != NULL)
2143 			INP_WUNLOCK(inp);
2144 	} else if (inp != NULL)  {
2145 		if (!(inp->inp_flags & INP_TIMEWAIT) &&
2146 		    !(inp->inp_flags & INP_DROPPED) &&
2147 		    !(inp->inp_socket == NULL)) {
2148 			icmp_tcp_seq = ntohl(th->th_seq);
2149 			tp = intotcpcb(inp);
2150 			if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2151 			    SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2152 				if (cmd == PRC_MSGSIZE) {
2153 					/*
2154 					 * MTU discovery:
2155 					 * If we got a needfrag set the MTU
2156 					 * in the route to the suggested new
2157 					 * value (if given) and then notify.
2158 					 */
2159 					mtu = ntohl(icmp6->icmp6_mtu);
2160 					/*
2161 					 * If no alternative MTU was
2162 					 * proposed, or the proposed
2163 					 * MTU was too small, set to
2164 					 * the min.
2165 					 */
2166 					if (mtu < IPV6_MMTU)
2167 						mtu = IPV6_MMTU - 8;
2168 
2169 
2170 					bzero(&inc, sizeof(inc));
2171 					inc.inc_fibnum = M_GETFIB(m);
2172 					inc.inc_flags |= INC_ISIPV6;
2173 					inc.inc6_faddr = *dst;
2174 					if (in6_setscope(&inc.inc6_faddr,
2175 						m->m_pkthdr.rcvif, NULL))
2176 						goto unlock_inp;
2177 
2178 					/*
2179 					 * Only process the offered MTU if it
2180 					 * is smaller than the current one.
2181 					 */
2182 					if (mtu < tp->t_maxseg +
2183 					    (sizeof (*th) + sizeof (*ip6))) {
2184 						tcp_hc_updatemtu(&inc, mtu);
2185 						tcp_mtudisc(inp, mtu);
2186 						ICMP6STAT_INC(icp6s_pmtuchg);
2187 					}
2188 				} else
2189 					inp = (*notify)(inp,
2190 					    inet6ctlerrmap[cmd]);
2191 			}
2192 		}
2193 unlock_inp:
2194 		if (inp != NULL)
2195 			INP_WUNLOCK(inp);
2196 	} else {
2197 		bzero(&inc, sizeof(inc));
2198 		inc.inc_fibnum = M_GETFIB(m);
2199 		inc.inc_flags |= INC_ISIPV6;
2200 		inc.inc_fport = th->th_dport;
2201 		inc.inc_lport = th->th_sport;
2202 		inc.inc6_faddr = *dst;
2203 		inc.inc6_laddr = ip6->ip6_src;
2204 		syncache_unreach(&inc, th);
2205 	}
2206 	INP_INFO_RUNLOCK(&V_tcbinfo);
2207 }
2208 #endif /* INET6 */
2209 
2210 
2211 /*
2212  * Following is where TCP initial sequence number generation occurs.
2213  *
2214  * There are two places where we must use initial sequence numbers:
2215  * 1.  In SYN-ACK packets.
2216  * 2.  In SYN packets.
2217  *
2218  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2219  * tcp_syncache.c for details.
2220  *
2221  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2222  * depends on this property.  In addition, these ISNs should be
2223  * unguessable so as to prevent connection hijacking.  To satisfy
2224  * the requirements of this situation, the algorithm outlined in
2225  * RFC 1948 is used, with only small modifications.
2226  *
2227  * Implementation details:
2228  *
2229  * Time is based off the system timer, and is corrected so that it
2230  * increases by one megabyte per second.  This allows for proper
2231  * recycling on high speed LANs while still leaving over an hour
2232  * before rollover.
2233  *
2234  * As reading the *exact* system time is too expensive to be done
2235  * whenever setting up a TCP connection, we increment the time
2236  * offset in two ways.  First, a small random positive increment
2237  * is added to isn_offset for each connection that is set up.
2238  * Second, the function tcp_isn_tick fires once per clock tick
2239  * and increments isn_offset as necessary so that sequence numbers
2240  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2241  * random positive increments serve only to ensure that the same
2242  * exact sequence number is never sent out twice (as could otherwise
2243  * happen when a port is recycled in less than the system tick
2244  * interval.)
2245  *
2246  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2247  * between seeding of isn_secret.  This is normally set to zero,
2248  * as reseeding should not be necessary.
2249  *
2250  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2251  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2252  * general, this means holding an exclusive (write) lock.
2253  */
2254 
2255 #define ISN_BYTES_PER_SECOND 1048576
2256 #define ISN_STATIC_INCREMENT 4096
2257 #define ISN_RANDOM_INCREMENT (4096 - 1)
2258 
2259 static VNET_DEFINE(u_char, isn_secret[32]);
2260 static VNET_DEFINE(int, isn_last);
2261 static VNET_DEFINE(int, isn_last_reseed);
2262 static VNET_DEFINE(u_int32_t, isn_offset);
2263 static VNET_DEFINE(u_int32_t, isn_offset_old);
2264 
2265 #define	V_isn_secret			VNET(isn_secret)
2266 #define	V_isn_last			VNET(isn_last)
2267 #define	V_isn_last_reseed		VNET(isn_last_reseed)
2268 #define	V_isn_offset			VNET(isn_offset)
2269 #define	V_isn_offset_old		VNET(isn_offset_old)
2270 
2271 tcp_seq
2272 tcp_new_isn(struct tcpcb *tp)
2273 {
2274 	MD5_CTX isn_ctx;
2275 	u_int32_t md5_buffer[4];
2276 	tcp_seq new_isn;
2277 	u_int32_t projected_offset;
2278 
2279 	INP_WLOCK_ASSERT(tp->t_inpcb);
2280 
2281 	ISN_LOCK();
2282 	/* Seed if this is the first use, reseed if requested. */
2283 	if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2284 	     (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2285 		< (u_int)ticks))) {
2286 		read_random(&V_isn_secret, sizeof(V_isn_secret));
2287 		V_isn_last_reseed = ticks;
2288 	}
2289 
2290 	/* Compute the md5 hash and return the ISN. */
2291 	MD5Init(&isn_ctx);
2292 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2293 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2294 #ifdef INET6
2295 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2296 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2297 			  sizeof(struct in6_addr));
2298 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2299 			  sizeof(struct in6_addr));
2300 	} else
2301 #endif
2302 	{
2303 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2304 			  sizeof(struct in_addr));
2305 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2306 			  sizeof(struct in_addr));
2307 	}
2308 	MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2309 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
2310 	new_isn = (tcp_seq) md5_buffer[0];
2311 	V_isn_offset += ISN_STATIC_INCREMENT +
2312 		(arc4random() & ISN_RANDOM_INCREMENT);
2313 	if (ticks != V_isn_last) {
2314 		projected_offset = V_isn_offset_old +
2315 		    ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2316 		if (SEQ_GT(projected_offset, V_isn_offset))
2317 			V_isn_offset = projected_offset;
2318 		V_isn_offset_old = V_isn_offset;
2319 		V_isn_last = ticks;
2320 	}
2321 	new_isn += V_isn_offset;
2322 	ISN_UNLOCK();
2323 	return (new_isn);
2324 }
2325 
2326 /*
2327  * When a specific ICMP unreachable message is received and the
2328  * connection state is SYN-SENT, drop the connection.  This behavior
2329  * is controlled by the icmp_may_rst sysctl.
2330  */
2331 struct inpcb *
2332 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2333 {
2334 	struct tcpcb *tp;
2335 
2336 	INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2337 	INP_WLOCK_ASSERT(inp);
2338 
2339 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2340 	    (inp->inp_flags & INP_DROPPED))
2341 		return (inp);
2342 
2343 	tp = intotcpcb(inp);
2344 	if (tp->t_state != TCPS_SYN_SENT)
2345 		return (inp);
2346 
2347 	tp = tcp_drop(tp, errno);
2348 	if (tp != NULL)
2349 		return (inp);
2350 	else
2351 		return (NULL);
2352 }
2353 
2354 /*
2355  * When `need fragmentation' ICMP is received, update our idea of the MSS
2356  * based on the new value. Also nudge TCP to send something, since we
2357  * know the packet we just sent was dropped.
2358  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2359  */
2360 static struct inpcb *
2361 tcp_mtudisc_notify(struct inpcb *inp, int error)
2362 {
2363 
2364 	tcp_mtudisc(inp, -1);
2365 	return (inp);
2366 }
2367 
2368 static void
2369 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2370 {
2371 	struct tcpcb *tp;
2372 	struct socket *so;
2373 
2374 	INP_WLOCK_ASSERT(inp);
2375 	if ((inp->inp_flags & INP_TIMEWAIT) ||
2376 	    (inp->inp_flags & INP_DROPPED))
2377 		return;
2378 
2379 	tp = intotcpcb(inp);
2380 	KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2381 
2382 	tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2383 
2384 	so = inp->inp_socket;
2385 	SOCKBUF_LOCK(&so->so_snd);
2386 	/* If the mss is larger than the socket buffer, decrease the mss. */
2387 	if (so->so_snd.sb_hiwat < tp->t_maxseg)
2388 		tp->t_maxseg = so->so_snd.sb_hiwat;
2389 	SOCKBUF_UNLOCK(&so->so_snd);
2390 
2391 	TCPSTAT_INC(tcps_mturesent);
2392 	tp->t_rtttime = 0;
2393 	tp->snd_nxt = tp->snd_una;
2394 	tcp_free_sackholes(tp);
2395 	tp->snd_recover = tp->snd_max;
2396 	if (tp->t_flags & TF_SACK_PERMIT)
2397 		EXIT_FASTRECOVERY(tp->t_flags);
2398 	tp->t_fb->tfb_tcp_output(tp);
2399 }
2400 
2401 #ifdef INET
2402 /*
2403  * Look-up the routing entry to the peer of this inpcb.  If no route
2404  * is found and it cannot be allocated, then return 0.  This routine
2405  * is called by TCP routines that access the rmx structure and by
2406  * tcp_mss_update to get the peer/interface MTU.
2407  */
2408 uint32_t
2409 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2410 {
2411 	struct nhop4_extended nh4;
2412 	struct ifnet *ifp;
2413 	uint32_t maxmtu = 0;
2414 
2415 	KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2416 
2417 	if (inc->inc_faddr.s_addr != INADDR_ANY) {
2418 
2419 		if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2420 		    NHR_REF, 0, &nh4) != 0)
2421 			return (0);
2422 
2423 		ifp = nh4.nh_ifp;
2424 		maxmtu = nh4.nh_mtu;
2425 
2426 		/* Report additional interface capabilities. */
2427 		if (cap != NULL) {
2428 			if (ifp->if_capenable & IFCAP_TSO4 &&
2429 			    ifp->if_hwassist & CSUM_TSO) {
2430 				cap->ifcap |= CSUM_TSO;
2431 				cap->tsomax = ifp->if_hw_tsomax;
2432 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2433 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2434 			}
2435 		}
2436 		fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2437 	}
2438 	return (maxmtu);
2439 }
2440 #endif /* INET */
2441 
2442 #ifdef INET6
2443 uint32_t
2444 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2445 {
2446 	struct nhop6_extended nh6;
2447 	struct in6_addr dst6;
2448 	uint32_t scopeid;
2449 	struct ifnet *ifp;
2450 	uint32_t maxmtu = 0;
2451 
2452 	KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2453 
2454 	if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2455 		in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2456 		if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2457 		    0, &nh6) != 0)
2458 			return (0);
2459 
2460 		ifp = nh6.nh_ifp;
2461 		maxmtu = nh6.nh_mtu;
2462 
2463 		/* Report additional interface capabilities. */
2464 		if (cap != NULL) {
2465 			if (ifp->if_capenable & IFCAP_TSO6 &&
2466 			    ifp->if_hwassist & CSUM_TSO) {
2467 				cap->ifcap |= CSUM_TSO;
2468 				cap->tsomax = ifp->if_hw_tsomax;
2469 				cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2470 				cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2471 			}
2472 		}
2473 		fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2474 	}
2475 
2476 	return (maxmtu);
2477 }
2478 #endif /* INET6 */
2479 
2480 /*
2481  * Calculate effective SMSS per RFC5681 definition for a given TCP
2482  * connection at its current state, taking into account SACK and etc.
2483  */
2484 u_int
2485 tcp_maxseg(const struct tcpcb *tp)
2486 {
2487 	u_int optlen;
2488 
2489 	if (tp->t_flags & TF_NOOPT)
2490 		return (tp->t_maxseg);
2491 
2492 	/*
2493 	 * Here we have a simplified code from tcp_addoptions(),
2494 	 * without a proper loop, and having most of paddings hardcoded.
2495 	 * We might make mistakes with padding here in some edge cases,
2496 	 * but this is harmless, since result of tcp_maxseg() is used
2497 	 * only in cwnd and ssthresh estimations.
2498 	 */
2499 #define	PAD(len)	((((len) / 4) + !!((len) % 4)) * 4)
2500 	if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2501 		if (tp->t_flags & TF_RCVD_TSTMP)
2502 			optlen = TCPOLEN_TSTAMP_APPA;
2503 		else
2504 			optlen = 0;
2505 #ifdef TCP_SIGNATURE
2506 		if (tp->t_flags & TF_SIGNATURE)
2507 			optlen += PAD(TCPOLEN_SIGNATURE);
2508 #endif
2509 		if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2510 			optlen += TCPOLEN_SACKHDR;
2511 			optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2512 			optlen = PAD(optlen);
2513 		}
2514 	} else {
2515 		if (tp->t_flags & TF_REQ_TSTMP)
2516 			optlen = TCPOLEN_TSTAMP_APPA;
2517 		else
2518 			optlen = PAD(TCPOLEN_MAXSEG);
2519 		if (tp->t_flags & TF_REQ_SCALE)
2520 			optlen += PAD(TCPOLEN_WINDOW);
2521 #ifdef TCP_SIGNATURE
2522 		if (tp->t_flags & TF_SIGNATURE)
2523 			optlen += PAD(TCPOLEN_SIGNATURE);
2524 #endif
2525 		if (tp->t_flags & TF_SACK_PERMIT)
2526 			optlen += PAD(TCPOLEN_SACK_PERMITTED);
2527 	}
2528 #undef PAD
2529 	optlen = min(optlen, TCP_MAXOLEN);
2530 	return (tp->t_maxseg - optlen);
2531 }
2532 
2533 #ifdef IPSEC
2534 /* compute ESP/AH header size for TCP, including outer IP header. */
2535 size_t
2536 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2537 {
2538 	struct inpcb *inp;
2539 	struct mbuf *m;
2540 	size_t hdrsiz;
2541 	struct ip *ip;
2542 #ifdef INET6
2543 	struct ip6_hdr *ip6;
2544 #endif
2545 	struct tcphdr *th;
2546 
2547 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2548 		(!key_havesp(IPSEC_DIR_OUTBOUND)))
2549 		return (0);
2550 	m = m_gethdr(M_NOWAIT, MT_DATA);
2551 	if (!m)
2552 		return (0);
2553 
2554 #ifdef INET6
2555 	if ((inp->inp_vflag & INP_IPV6) != 0) {
2556 		ip6 = mtod(m, struct ip6_hdr *);
2557 		th = (struct tcphdr *)(ip6 + 1);
2558 		m->m_pkthdr.len = m->m_len =
2559 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2560 		tcpip_fillheaders(inp, ip6, th);
2561 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2562 	} else
2563 #endif /* INET6 */
2564 	{
2565 		ip = mtod(m, struct ip *);
2566 		th = (struct tcphdr *)(ip + 1);
2567 		m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2568 		tcpip_fillheaders(inp, ip, th);
2569 		hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2570 	}
2571 
2572 	m_free(m);
2573 	return (hdrsiz);
2574 }
2575 #endif /* IPSEC */
2576 
2577 #ifdef TCP_SIGNATURE
2578 /*
2579  * Callback function invoked by m_apply() to digest TCP segment data
2580  * contained within an mbuf chain.
2581  */
2582 static int
2583 tcp_signature_apply(void *fstate, void *data, u_int len)
2584 {
2585 
2586 	MD5Update(fstate, (u_char *)data, len);
2587 	return (0);
2588 }
2589 
2590 /*
2591  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2592  * search with the destination IP address, and a 'magic SPI' to be
2593  * determined by the application. This is hardcoded elsewhere to 1179
2594 */
2595 struct secasvar *
2596 tcp_get_sav(struct mbuf *m, u_int direction)
2597 {
2598 	union sockaddr_union dst;
2599 	struct secasvar *sav;
2600 	struct ip *ip;
2601 #ifdef INET6
2602 	struct ip6_hdr *ip6;
2603 	char ip6buf[INET6_ADDRSTRLEN];
2604 #endif
2605 
2606 	/* Extract the destination from the IP header in the mbuf. */
2607 	bzero(&dst, sizeof(union sockaddr_union));
2608 	ip = mtod(m, struct ip *);
2609 #ifdef INET6
2610 	ip6 = NULL;	/* Make the compiler happy. */
2611 #endif
2612 	switch (ip->ip_v) {
2613 #ifdef INET
2614 	case IPVERSION:
2615 		dst.sa.sa_len = sizeof(struct sockaddr_in);
2616 		dst.sa.sa_family = AF_INET;
2617 		dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2618 		    ip->ip_src : ip->ip_dst;
2619 		break;
2620 #endif
2621 #ifdef INET6
2622 	case (IPV6_VERSION >> 4):
2623 		ip6 = mtod(m, struct ip6_hdr *);
2624 		dst.sa.sa_len = sizeof(struct sockaddr_in6);
2625 		dst.sa.sa_family = AF_INET6;
2626 		dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2627 		    ip6->ip6_src : ip6->ip6_dst;
2628 		break;
2629 #endif
2630 	default:
2631 		return (NULL);
2632 		/* NOTREACHED */
2633 		break;
2634 	}
2635 
2636 	/* Look up an SADB entry which matches the address of the peer. */
2637 	sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2638 	if (sav == NULL) {
2639 		ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2640 		    (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2641 #ifdef INET6
2642 			(ip->ip_v == (IPV6_VERSION >> 4)) ?
2643 			    ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2644 #endif
2645 			"(unsupported)"));
2646 	}
2647 
2648 	return (sav);
2649 }
2650 
2651 /*
2652  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2653  *
2654  * Parameters:
2655  * m		pointer to head of mbuf chain
2656  * len		length of TCP segment data, excluding options
2657  * optlen	length of TCP segment options
2658  * buf		pointer to storage for computed MD5 digest
2659  * sav		pointer to security assosiation
2660  *
2661  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2662  * When called from tcp_input(), we can be sure that th_sum has been
2663  * zeroed out and verified already.
2664  *
2665  * Releases reference to SADB key before return.
2666  *
2667  * Return 0 if successful, otherwise return -1.
2668  *
2669  */
2670 int
2671 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2672     u_char *buf, struct secasvar *sav)
2673 {
2674 #ifdef INET
2675 	struct ippseudo ippseudo;
2676 #endif
2677 	MD5_CTX ctx;
2678 	int doff;
2679 	struct ip *ip;
2680 #ifdef INET
2681 	struct ipovly *ipovly;
2682 #endif
2683 	struct tcphdr *th;
2684 #ifdef INET6
2685 	struct ip6_hdr *ip6;
2686 	struct in6_addr in6;
2687 	uint32_t plen;
2688 	uint16_t nhdr;
2689 #endif
2690 	u_short savecsum;
2691 
2692 	KASSERT(m != NULL, ("NULL mbuf chain"));
2693 	KASSERT(buf != NULL, ("NULL signature pointer"));
2694 
2695 	/* Extract the destination from the IP header in the mbuf. */
2696 	ip = mtod(m, struct ip *);
2697 #ifdef INET6
2698 	ip6 = NULL;	/* Make the compiler happy. */
2699 #endif
2700 
2701 	MD5Init(&ctx);
2702 	/*
2703 	 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2704 	 *
2705 	 * XXX The ippseudo header MUST be digested in network byte order,
2706 	 * or else we'll fail the regression test. Assume all fields we've
2707 	 * been doing arithmetic on have been in host byte order.
2708 	 * XXX One cannot depend on ipovly->ih_len here. When called from
2709 	 * tcp_output(), the underlying ip_len member has not yet been set.
2710 	 */
2711 	switch (ip->ip_v) {
2712 #ifdef INET
2713 	case IPVERSION:
2714 		ipovly = (struct ipovly *)ip;
2715 		ippseudo.ippseudo_src = ipovly->ih_src;
2716 		ippseudo.ippseudo_dst = ipovly->ih_dst;
2717 		ippseudo.ippseudo_pad = 0;
2718 		ippseudo.ippseudo_p = IPPROTO_TCP;
2719 		ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2720 		    optlen);
2721 		MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2722 
2723 		th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2724 		doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2725 		break;
2726 #endif
2727 #ifdef INET6
2728 	/*
2729 	 * RFC 2385, 2.0  Proposal
2730 	 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2731 	 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2732 	 * extended next header value (to form 32 bits), and 32-bit segment
2733 	 * length.
2734 	 * Note: Upper-Layer Packet Length comes before Next Header.
2735 	 */
2736 	case (IPV6_VERSION >> 4):
2737 		ip6 = mtod(m, struct ip6_hdr *);
2738 		in6 = ip6->ip6_src;
2739 		in6_clearscope(&in6);
2740 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2741 		in6 = ip6->ip6_dst;
2742 		in6_clearscope(&in6);
2743 		MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2744 		plen = htonl(len + sizeof(struct tcphdr) + optlen);
2745 		MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2746 		nhdr = 0;
2747 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2748 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2749 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2750 		nhdr = IPPROTO_TCP;
2751 		MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2752 
2753 		th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2754 		doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2755 		break;
2756 #endif
2757 	default:
2758 		KEY_FREESAV(&sav);
2759 		return (-1);
2760 		/* NOTREACHED */
2761 		break;
2762 	}
2763 
2764 
2765 	/*
2766 	 * Step 2: Update MD5 hash with TCP header, excluding options.
2767 	 * The TCP checksum must be set to zero.
2768 	 */
2769 	savecsum = th->th_sum;
2770 	th->th_sum = 0;
2771 	MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2772 	th->th_sum = savecsum;
2773 
2774 	/*
2775 	 * Step 3: Update MD5 hash with TCP segment data.
2776 	 *         Use m_apply() to avoid an early m_pullup().
2777 	 */
2778 	if (len > 0)
2779 		m_apply(m, doff, len, tcp_signature_apply, &ctx);
2780 
2781 	/*
2782 	 * Step 4: Update MD5 hash with shared secret.
2783 	 */
2784 	MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2785 	MD5Final(buf, &ctx);
2786 
2787 	key_sa_recordxfer(sav, m);
2788 	KEY_FREESAV(&sav);
2789 	return (0);
2790 }
2791 
2792 /*
2793  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2794  *
2795  * Return 0 if successful, otherwise return -1.
2796  */
2797 int
2798 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2799     u_char *buf, u_int direction)
2800 {
2801 	struct secasvar *sav;
2802 
2803 	if ((sav = tcp_get_sav(m, direction)) == NULL)
2804 		return (-1);
2805 
2806 	return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2807 }
2808 
2809 /*
2810  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2811  *
2812  * Parameters:
2813  * m		pointer to head of mbuf chain
2814  * len		length of TCP segment data, excluding options
2815  * optlen	length of TCP segment options
2816  * buf		pointer to storage for computed MD5 digest
2817  * direction	direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2818  *
2819  * Return 1 if successful, otherwise return 0.
2820  */
2821 int
2822 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2823     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2824 {
2825 	char tmpdigest[TCP_SIGLEN];
2826 
2827 	if (tcp_sig_checksigs == 0)
2828 		return (1);
2829 	if ((tcpbflag & TF_SIGNATURE) == 0) {
2830 		if ((to->to_flags & TOF_SIGNATURE) != 0) {
2831 
2832 			/*
2833 			 * If this socket is not expecting signature but
2834 			 * the segment contains signature just fail.
2835 			 */
2836 			TCPSTAT_INC(tcps_sig_err_sigopt);
2837 			TCPSTAT_INC(tcps_sig_rcvbadsig);
2838 			return (0);
2839 		}
2840 
2841 		/* Signature is not expected, and not present in segment. */
2842 		return (1);
2843 	}
2844 
2845 	/*
2846 	 * If this socket is expecting signature but the segment does not
2847 	 * contain any just fail.
2848 	 */
2849 	if ((to->to_flags & TOF_SIGNATURE) == 0) {
2850 		TCPSTAT_INC(tcps_sig_err_nosigopt);
2851 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2852 		return (0);
2853 	}
2854 	if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2855 	    IPSEC_DIR_INBOUND) == -1) {
2856 		TCPSTAT_INC(tcps_sig_err_buildsig);
2857 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2858 		return (0);
2859 	}
2860 
2861 	if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2862 		TCPSTAT_INC(tcps_sig_rcvbadsig);
2863 		return (0);
2864 	}
2865 	TCPSTAT_INC(tcps_sig_rcvgoodsig);
2866 	return (1);
2867 }
2868 #endif /* TCP_SIGNATURE */
2869 
2870 static int
2871 sysctl_drop(SYSCTL_HANDLER_ARGS)
2872 {
2873 	/* addrs[0] is a foreign socket, addrs[1] is a local one. */
2874 	struct sockaddr_storage addrs[2];
2875 	struct inpcb *inp;
2876 	struct tcpcb *tp;
2877 	struct tcptw *tw;
2878 	struct sockaddr_in *fin, *lin;
2879 #ifdef INET6
2880 	struct sockaddr_in6 *fin6, *lin6;
2881 #endif
2882 	int error;
2883 
2884 	inp = NULL;
2885 	fin = lin = NULL;
2886 #ifdef INET6
2887 	fin6 = lin6 = NULL;
2888 #endif
2889 	error = 0;
2890 
2891 	if (req->oldptr != NULL || req->oldlen != 0)
2892 		return (EINVAL);
2893 	if (req->newptr == NULL)
2894 		return (EPERM);
2895 	if (req->newlen < sizeof(addrs))
2896 		return (ENOMEM);
2897 	error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2898 	if (error)
2899 		return (error);
2900 
2901 	switch (addrs[0].ss_family) {
2902 #ifdef INET6
2903 	case AF_INET6:
2904 		fin6 = (struct sockaddr_in6 *)&addrs[0];
2905 		lin6 = (struct sockaddr_in6 *)&addrs[1];
2906 		if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2907 		    lin6->sin6_len != sizeof(struct sockaddr_in6))
2908 			return (EINVAL);
2909 		if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2910 			if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2911 				return (EINVAL);
2912 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2913 			in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2914 			fin = (struct sockaddr_in *)&addrs[0];
2915 			lin = (struct sockaddr_in *)&addrs[1];
2916 			break;
2917 		}
2918 		error = sa6_embedscope(fin6, V_ip6_use_defzone);
2919 		if (error)
2920 			return (error);
2921 		error = sa6_embedscope(lin6, V_ip6_use_defzone);
2922 		if (error)
2923 			return (error);
2924 		break;
2925 #endif
2926 #ifdef INET
2927 	case AF_INET:
2928 		fin = (struct sockaddr_in *)&addrs[0];
2929 		lin = (struct sockaddr_in *)&addrs[1];
2930 		if (fin->sin_len != sizeof(struct sockaddr_in) ||
2931 		    lin->sin_len != sizeof(struct sockaddr_in))
2932 			return (EINVAL);
2933 		break;
2934 #endif
2935 	default:
2936 		return (EINVAL);
2937 	}
2938 	INP_INFO_RLOCK(&V_tcbinfo);
2939 	switch (addrs[0].ss_family) {
2940 #ifdef INET6
2941 	case AF_INET6:
2942 		inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2943 		    fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2944 		    INPLOOKUP_WLOCKPCB, NULL);
2945 		break;
2946 #endif
2947 #ifdef INET
2948 	case AF_INET:
2949 		inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2950 		    lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2951 		break;
2952 #endif
2953 	}
2954 	if (inp != NULL) {
2955 		if (inp->inp_flags & INP_TIMEWAIT) {
2956 			/*
2957 			 * XXXRW: There currently exists a state where an
2958 			 * inpcb is present, but its timewait state has been
2959 			 * discarded.  For now, don't allow dropping of this
2960 			 * type of inpcb.
2961 			 */
2962 			tw = intotw(inp);
2963 			if (tw != NULL)
2964 				tcp_twclose(tw, 0);
2965 			else
2966 				INP_WUNLOCK(inp);
2967 		} else if (!(inp->inp_flags & INP_DROPPED) &&
2968 			   !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2969 			tp = intotcpcb(inp);
2970 			tp = tcp_drop(tp, ECONNABORTED);
2971 			if (tp != NULL)
2972 				INP_WUNLOCK(inp);
2973 		} else
2974 			INP_WUNLOCK(inp);
2975 	} else
2976 		error = ESRCH;
2977 	INP_INFO_RUNLOCK(&V_tcbinfo);
2978 	return (error);
2979 }
2980 
2981 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2982     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2983     0, sysctl_drop, "", "Drop TCP connection");
2984 
2985 /*
2986  * Generate a standardized TCP log line for use throughout the
2987  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2988  * allow use in the interrupt context.
2989  *
2990  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2991  * NB: The function may return NULL if memory allocation failed.
2992  *
2993  * Due to header inclusion and ordering limitations the struct ip
2994  * and ip6_hdr pointers have to be passed as void pointers.
2995  */
2996 char *
2997 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2998     const void *ip6hdr)
2999 {
3000 
3001 	/* Is logging enabled? */
3002 	if (tcp_log_in_vain == 0)
3003 		return (NULL);
3004 
3005 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3006 }
3007 
3008 char *
3009 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3010     const void *ip6hdr)
3011 {
3012 
3013 	/* Is logging enabled? */
3014 	if (tcp_log_debug == 0)
3015 		return (NULL);
3016 
3017 	return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3018 }
3019 
3020 static char *
3021 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3022     const void *ip6hdr)
3023 {
3024 	char *s, *sp;
3025 	size_t size;
3026 	struct ip *ip;
3027 #ifdef INET6
3028 	const struct ip6_hdr *ip6;
3029 
3030 	ip6 = (const struct ip6_hdr *)ip6hdr;
3031 #endif /* INET6 */
3032 	ip = (struct ip *)ip4hdr;
3033 
3034 	/*
3035 	 * The log line looks like this:
3036 	 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3037 	 */
3038 	size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3039 	    sizeof(PRINT_TH_FLAGS) + 1 +
3040 #ifdef INET6
3041 	    2 * INET6_ADDRSTRLEN;
3042 #else
3043 	    2 * INET_ADDRSTRLEN;
3044 #endif /* INET6 */
3045 
3046 	s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3047 	if (s == NULL)
3048 		return (NULL);
3049 
3050 	strcat(s, "TCP: [");
3051 	sp = s + strlen(s);
3052 
3053 	if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3054 		inet_ntoa_r(inc->inc_faddr, sp);
3055 		sp = s + strlen(s);
3056 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3057 		sp = s + strlen(s);
3058 		inet_ntoa_r(inc->inc_laddr, sp);
3059 		sp = s + strlen(s);
3060 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3061 #ifdef INET6
3062 	} else if (inc) {
3063 		ip6_sprintf(sp, &inc->inc6_faddr);
3064 		sp = s + strlen(s);
3065 		sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3066 		sp = s + strlen(s);
3067 		ip6_sprintf(sp, &inc->inc6_laddr);
3068 		sp = s + strlen(s);
3069 		sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3070 	} else if (ip6 && th) {
3071 		ip6_sprintf(sp, &ip6->ip6_src);
3072 		sp = s + strlen(s);
3073 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3074 		sp = s + strlen(s);
3075 		ip6_sprintf(sp, &ip6->ip6_dst);
3076 		sp = s + strlen(s);
3077 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3078 #endif /* INET6 */
3079 #ifdef INET
3080 	} else if (ip && th) {
3081 		inet_ntoa_r(ip->ip_src, sp);
3082 		sp = s + strlen(s);
3083 		sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3084 		sp = s + strlen(s);
3085 		inet_ntoa_r(ip->ip_dst, sp);
3086 		sp = s + strlen(s);
3087 		sprintf(sp, "]:%i", ntohs(th->th_dport));
3088 #endif /* INET */
3089 	} else {
3090 		free(s, M_TCPLOG);
3091 		return (NULL);
3092 	}
3093 	sp = s + strlen(s);
3094 	if (th)
3095 		sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3096 	if (*(s + size - 1) != '\0')
3097 		panic("%s: string too long", __func__);
3098 	return (s);
3099 }
3100 
3101 /*
3102  * A subroutine which makes it easy to track TCP state changes with DTrace.
3103  * This function shouldn't be called for t_state initializations that don't
3104  * correspond to actual TCP state transitions.
3105  */
3106 void
3107 tcp_state_change(struct tcpcb *tp, int newstate)
3108 {
3109 #if defined(KDTRACE_HOOKS)
3110 	int pstate = tp->t_state;
3111 #endif
3112 
3113 	TCPSTATES_DEC(tp->t_state);
3114 	TCPSTATES_INC(newstate);
3115 	tp->t_state = newstate;
3116 	TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3117 }
3118