xref: /freebsd/sys/netinet/tcp_subr.c (revision 23f282aa31e9b6fceacd449020e936e98d6f2298)
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  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
34  * $FreeBSD$
35  */
36 
37 #include "opt_compat.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 #include "opt_tcpdebug.h"
41 
42 #include <stddef.h>
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/callout.h>
46 #include <sys/kernel.h>
47 #include <sys/sysctl.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #ifdef INET6
51 #include <sys/domain.h>
52 #endif
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 
58 #include <vm/vm_zone.h>
59 
60 #include <net/route.h>
61 #include <net/if.h>
62 
63 #define _IP_VHL
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #endif
79 #include <netinet/tcp.h>
80 #include <netinet/tcp_fsm.h>
81 #include <netinet/tcp_seq.h>
82 #include <netinet/tcp_timer.h>
83 #include <netinet/tcp_var.h>
84 #ifdef INET6
85 #include <netinet6/tcp6_var.h>
86 #endif
87 #include <netinet/tcpip.h>
88 #ifdef TCPDEBUG
89 #include <netinet/tcp_debug.h>
90 #endif
91 #include <netinet6/ip6protosw.h>
92 
93 #ifdef IPSEC
94 #include <netinet6/ipsec.h>
95 #endif /*IPSEC*/
96 
97 #include <machine/in_cksum.h>
98 
99 int 	tcp_mssdflt = TCP_MSS;
100 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
101     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
102 
103 #ifdef INET6
104 int	tcp_v6mssdflt = TCP6_MSS;
105 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
106 	CTLFLAG_RW, &tcp_v6mssdflt , 0,
107 	"Default TCP Maximum Segment Size for IPv6");
108 #endif
109 
110 #if 0
111 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
112 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
113     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
114 #endif
115 
116 static int	tcp_do_rfc1323 = 1;
117 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
118     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
119 
120 static int	tcp_do_rfc1644 = 0;
121 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
122     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
123 
124 static int	tcp_tcbhashsize = 0;
125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
126      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
127 
128 static int	do_tcpdrain = 1;
129 SYSCTL_INT(_debug, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
130      "Enable non Net3 compliant tcp_drain");
131 
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
133     &tcbinfo.ipi_count, 0, "Number of active PCBs");
134 
135 static void	tcp_cleartaocache __P((void));
136 static void	tcp_notify __P((struct inpcb *, int));
137 
138 /*
139  * Target size of TCP PCB hash tables. Must be a power of two.
140  *
141  * Note that this can be overridden by the kernel environment
142  * variable net.inet.tcp.tcbhashsize
143  */
144 #ifndef TCBHASHSIZE
145 #define TCBHASHSIZE	512
146 #endif
147 
148 /*
149  * This is the actual shape of what we allocate using the zone
150  * allocator.  Doing it this way allows us to protect both structures
151  * using the same generation count, and also eliminates the overhead
152  * of allocating tcpcbs separately.  By hiding the structure here,
153  * we avoid changing most of the rest of the code (although it needs
154  * to be changed, eventually, for greater efficiency).
155  */
156 #define	ALIGNMENT	32
157 #define	ALIGNM1		(ALIGNMENT - 1)
158 struct	inp_tp {
159 	union {
160 		struct	inpcb inp;
161 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
162 	} inp_tp_u;
163 	struct	tcpcb tcb;
164 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
165 	struct	callout inp_tp_delack;
166 };
167 #undef ALIGNMENT
168 #undef ALIGNM1
169 
170 /*
171  * Tcp initialization
172  */
173 void
174 tcp_init()
175 {
176 	int hashsize;
177 
178 	tcp_iss = random();	/* wrong, but better than a constant */
179 	tcp_ccgen = 1;
180 	tcp_cleartaocache();
181 
182 	tcp_delacktime = TCPTV_DELACK;
183 	tcp_keepinit = TCPTV_KEEP_INIT;
184 	tcp_keepidle = TCPTV_KEEP_IDLE;
185 	tcp_keepintvl = TCPTV_KEEPINTVL;
186 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
187 	tcp_msl = TCPTV_MSL;
188 
189 	LIST_INIT(&tcb);
190 	tcbinfo.listhead = &tcb;
191 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
192 	if (!powerof2(hashsize)) {
193 		printf("WARNING: TCB hash size not a power of 2\n");
194 		hashsize = 512; /* safe default */
195 	}
196 	tcp_tcbhashsize = hashsize;
197 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
198 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
199 					&tcbinfo.porthashmask);
200 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
201 				 ZONE_INTERRUPT, 0);
202 #ifdef INET6
203 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
204 #else /* INET6 */
205 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
206 #endif /* INET6 */
207 	if (max_protohdr < TCP_MINPROTOHDR)
208 		max_protohdr = TCP_MINPROTOHDR;
209 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
210 		panic("tcp_init");
211 #undef TCP_MINPROTOHDR
212 }
213 
214 /*
215  * Create template to be used to send tcp packets on a connection.
216  * Call after host entry created, allocates an mbuf and fills
217  * in a skeletal tcp/ip header, minimizing the amount of work
218  * necessary when the connection is used.
219  */
220 struct tcptemp *
221 tcp_template(tp)
222 	struct tcpcb *tp;
223 {
224 	register struct inpcb *inp = tp->t_inpcb;
225 	register struct mbuf *m;
226 	register struct tcptemp *n;
227 
228 	if ((n = tp->t_template) == 0) {
229 		m = m_get(M_DONTWAIT, MT_HEADER);
230 		if (m == NULL)
231 			return (0);
232 		m->m_len = sizeof (struct tcptemp);
233 		n = mtod(m, struct tcptemp *);
234 	}
235 #ifdef INET6
236 	if ((inp->inp_vflag & INP_IPV6) != 0) {
237 		register struct ip6_hdr *ip6;
238 
239 		ip6 = (struct ip6_hdr *)n->tt_ipgen;
240 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
241 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
242 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
243 			(IPV6_VERSION & IPV6_VERSION_MASK);
244 		ip6->ip6_nxt = IPPROTO_TCP;
245 		ip6->ip6_plen = sizeof(struct tcphdr);
246 		ip6->ip6_src = inp->in6p_laddr;
247 		ip6->ip6_dst = inp->in6p_faddr;
248 		n->tt_t.th_sum = 0;
249 	} else
250 #endif
251       {
252 	struct ip *ip = (struct ip *)n->tt_ipgen;
253 
254 	bzero(ip, sizeof(struct ip));		/* XXX overkill? */
255 	ip->ip_vhl = IP_VHL_BORING;
256 	ip->ip_p = IPPROTO_TCP;
257 	ip->ip_src = inp->inp_laddr;
258 	ip->ip_dst = inp->inp_faddr;
259 	n->tt_t.th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
260 	    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
261       }
262 	n->tt_t.th_sport = inp->inp_lport;
263 	n->tt_t.th_dport = inp->inp_fport;
264 	n->tt_t.th_seq = 0;
265 	n->tt_t.th_ack = 0;
266 	n->tt_t.th_x2 = 0;
267 	n->tt_t.th_off = 5;
268 	n->tt_t.th_flags = 0;
269 	n->tt_t.th_win = 0;
270 	n->tt_t.th_urp = 0;
271 	return (n);
272 }
273 
274 /*
275  * Send a single message to the TCP at address specified by
276  * the given TCP/IP header.  If m == 0, then we make a copy
277  * of the tcpiphdr at ti and send directly to the addressed host.
278  * This is used to force keep alive messages out using the TCP
279  * template for a connection tp->t_template.  If flags are given
280  * then we send a message back to the TCP which originated the
281  * segment ti, and discard the mbuf containing it and any other
282  * attached mbufs.
283  *
284  * In any case the ack and sequence number of the transmitted
285  * segment are as specified by the parameters.
286  *
287  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
288  */
289 void
290 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
291 	struct tcpcb *tp;
292 	void *ipgen;
293 	register struct tcphdr *th;
294 	register struct mbuf *m;
295 	tcp_seq ack, seq;
296 	int flags;
297 {
298 	register int tlen;
299 	int win = 0;
300 	struct route *ro = 0;
301 	struct route sro;
302 	struct ip *ip;
303 	struct tcphdr *nth;
304 #ifdef INET6
305 	struct route_in6 *ro6 = 0;
306 	struct route_in6 sro6;
307 	struct ip6_hdr *ip6;
308 	int isipv6;
309 #endif /* INET6 */
310 	int ipflags = 0;
311 
312 #ifdef INET6
313 	isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
314 	ip6 = ipgen;
315 #endif /* INET6 */
316 	ip = ipgen;
317 
318 	if (tp) {
319 		if (!(flags & TH_RST)) {
320 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
321 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
322 				win = (long)TCP_MAXWIN << tp->rcv_scale;
323 		}
324 #ifdef INET6
325 		if (isipv6)
326 			ro6 = &tp->t_inpcb->in6p_route;
327 		else
328 #endif /* INET6 */
329 		ro = &tp->t_inpcb->inp_route;
330 	} else {
331 #ifdef INET6
332 		if (isipv6) {
333 			ro6 = &sro6;
334 			bzero(ro6, sizeof *ro6);
335 		} else
336 #endif /* INET6 */
337 	      {
338 		ro = &sro;
339 		bzero(ro, sizeof *ro);
340 	      }
341 	}
342 	if (m == 0) {
343 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
344 		if (m == NULL)
345 			return;
346 #ifdef TCP_COMPAT_42
347 		tlen = 1;
348 #else
349 		tlen = 0;
350 #endif
351 		m->m_data += max_linkhdr;
352 #ifdef INET6
353 		if (isipv6) {
354 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
355 			      sizeof(struct ip6_hdr));
356 			ip6 = mtod(m, struct ip6_hdr *);
357 			nth = (struct tcphdr *)(ip6 + 1);
358 		} else
359 #endif /* INET6 */
360 	      {
361 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
362 		ip = mtod(m, struct ip *);
363 		nth = (struct tcphdr *)(ip + 1);
364 	      }
365 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
366 		flags = TH_ACK;
367 	} else {
368 		m_freem(m->m_next);
369 		m->m_next = 0;
370 		m->m_data = (caddr_t)ipgen;
371 		/* m_len is set later */
372 		tlen = 0;
373 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
374 #ifdef INET6
375 		if (isipv6) {
376 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
377 			nth = (struct tcphdr *)(ip6 + 1);
378 		} else
379 #endif /* INET6 */
380 	      {
381 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
382 		nth = (struct tcphdr *)(ip + 1);
383 	      }
384 		if (th != nth) {
385 			/*
386 			 * this is usually a case when an extension header
387 			 * exists between the IPv6 header and the
388 			 * TCP header.
389 			 */
390 			nth->th_sport = th->th_sport;
391 			nth->th_dport = th->th_dport;
392 		}
393 		xchg(nth->th_dport, nth->th_sport, n_short);
394 #undef xchg
395 	}
396 #ifdef INET6
397 	if (isipv6) {
398 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
399 						tlen));
400 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
401 	} else
402 #endif
403       {
404 	tlen += sizeof (struct tcpiphdr);
405 	ip->ip_len = tlen;
406 	ip->ip_ttl = ip_defttl;
407       }
408 	m->m_len = tlen;
409 	m->m_pkthdr.len = tlen;
410 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
411 	nth->th_seq = htonl(seq);
412 	nth->th_ack = htonl(ack);
413 	nth->th_x2 = 0;
414 	nth->th_off = sizeof (struct tcphdr) >> 2;
415 	nth->th_flags = flags;
416 	if (tp)
417 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
418 	else
419 		nth->th_win = htons((u_short)win);
420 	nth->th_urp = 0;
421 #ifdef INET6
422 	if (isipv6) {
423 		nth->th_sum = 0;
424 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
425 					sizeof(struct ip6_hdr),
426 					tlen - sizeof(struct ip6_hdr));
427 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
428 					       ro6 && ro6->ro_rt ?
429 					       ro6->ro_rt->rt_ifp :
430 					       NULL);
431 	} else
432 #endif /* INET6 */
433       {
434         nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
435 	    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
436         m->m_pkthdr.csum_flags = CSUM_TCP;
437         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
438       }
439 #ifdef TCPDEBUG
440 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
441 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
442 #endif
443 #ifdef IPSEC
444 	if (tp != NULL) {
445 		m->m_pkthdr.rcvif = (struct ifnet *)tp->t_inpcb->inp_socket;
446 		ipflags |=
447 #ifdef INET6
448 			isipv6 ? IPV6_SOCKINMRCVIF :
449 #endif
450 			IP_SOCKINMRCVIF;
451 	}
452 #endif
453 #ifdef INET6
454 	if (isipv6) {
455 		(void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
456 		if (ro6 == &sro6 && ro6->ro_rt) {
457 			RTFREE(ro6->ro_rt);
458 			ro6->ro_rt = NULL;
459 		}
460 	} else
461 #endif /* INET6 */
462       {
463 	(void) ip_output(m, NULL, ro, ipflags, NULL);
464 	if (ro == &sro && ro->ro_rt) {
465 		RTFREE(ro->ro_rt);
466 		ro->ro_rt = NULL;
467 	}
468       }
469 }
470 
471 /*
472  * Create a new TCP control block, making an
473  * empty reassembly queue and hooking it to the argument
474  * protocol control block.  The `inp' parameter must have
475  * come from the zone allocator set up in tcp_init().
476  */
477 struct tcpcb *
478 tcp_newtcpcb(inp)
479 	struct inpcb *inp;
480 {
481 	struct inp_tp *it;
482 	register struct tcpcb *tp;
483 #ifdef INET6
484 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
485 #endif /* INET6 */
486 
487 	it = (struct inp_tp *)inp;
488 	tp = &it->tcb;
489 	bzero((char *) tp, sizeof(struct tcpcb));
490 	LIST_INIT(&tp->t_segq);
491 	tp->t_maxseg = tp->t_maxopd =
492 #ifdef INET6
493 		isipv6 ? tcp_v6mssdflt :
494 #endif /* INET6 */
495 		tcp_mssdflt;
496 
497 	/* Set up our timeouts. */
498 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
499 	callout_init(tp->tt_persist = &it->inp_tp_persist);
500 	callout_init(tp->tt_keep = &it->inp_tp_keep);
501 	callout_init(tp->tt_2msl = &it->inp_tp_2msl);
502 	callout_init(tp->tt_delack = &it->inp_tp_delack);
503 
504 	if (tcp_do_rfc1323)
505 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
506 	if (tcp_do_rfc1644)
507 		tp->t_flags |= TF_REQ_CC;
508 	tp->t_inpcb = inp;	/* XXX */
509 	/*
510 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
511 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
512 	 * reasonable initial retransmit time.
513 	 */
514 	tp->t_srtt = TCPTV_SRTTBASE;
515 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
516 	tp->t_rttmin = TCPTV_MIN;
517 	tp->t_rxtcur = TCPTV_RTOBASE;
518 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
519 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
520 	tp->t_rcvtime = ticks;
521         /*
522 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
523 	 * because the socket may be bound to an IPv6 wildcard address,
524 	 * which may match an IPv4-mapped IPv6 address.
525 	 */
526 	inp->inp_ip_ttl = ip_defttl;
527 	inp->inp_ppcb = (caddr_t)tp;
528 	return (tp);		/* XXX */
529 }
530 
531 /*
532  * Drop a TCP connection, reporting
533  * the specified error.  If connection is synchronized,
534  * then send a RST to peer.
535  */
536 struct tcpcb *
537 tcp_drop(tp, errno)
538 	register struct tcpcb *tp;
539 	int errno;
540 {
541 	struct socket *so = tp->t_inpcb->inp_socket;
542 
543 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
544 		tp->t_state = TCPS_CLOSED;
545 		(void) tcp_output(tp);
546 		tcpstat.tcps_drops++;
547 	} else
548 		tcpstat.tcps_conndrops++;
549 	if (errno == ETIMEDOUT && tp->t_softerror)
550 		errno = tp->t_softerror;
551 	so->so_error = errno;
552 	return (tcp_close(tp));
553 }
554 
555 /*
556  * Close a TCP control block:
557  *	discard all space held by the tcp
558  *	discard internet protocol block
559  *	wake up any sleepers
560  */
561 struct tcpcb *
562 tcp_close(tp)
563 	register struct tcpcb *tp;
564 {
565 	register struct tseg_qent *q;
566 	struct inpcb *inp = tp->t_inpcb;
567 	struct socket *so = inp->inp_socket;
568 #ifdef INET6
569 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
570 #endif /* INET6 */
571 	register struct rtentry *rt;
572 	int dosavessthresh;
573 
574 	/*
575 	 * Make sure that all of our timers are stopped before we
576 	 * delete the PCB.
577 	 */
578 	callout_stop(tp->tt_rexmt);
579 	callout_stop(tp->tt_persist);
580 	callout_stop(tp->tt_keep);
581 	callout_stop(tp->tt_2msl);
582 	callout_stop(tp->tt_delack);
583 
584 	/*
585 	 * If we got enough samples through the srtt filter,
586 	 * save the rtt and rttvar in the routing entry.
587 	 * 'Enough' is arbitrarily defined as the 16 samples.
588 	 * 16 samples is enough for the srtt filter to converge
589 	 * to within 5% of the correct value; fewer samples and
590 	 * we could save a very bogus rtt.
591 	 *
592 	 * Don't update the default route's characteristics and don't
593 	 * update anything that the user "locked".
594 	 */
595 	if (tp->t_rttupdated >= 16) {
596 		register u_long i = 0;
597 #ifdef INET6
598 		if (isipv6) {
599 			struct sockaddr_in6 *sin6;
600 
601 			if ((rt = inp->in6p_route.ro_rt) == NULL)
602 				goto no_valid_rt;
603 			sin6 = (struct sockaddr_in6 *)rt_key(rt);
604 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
605 				goto no_valid_rt;
606 		}
607 		else
608 #endif /* INET6 */
609 		if ((rt = inp->inp_route.ro_rt) == NULL ||
610 		    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
611 		    == INADDR_ANY)
612 			goto no_valid_rt;
613 
614 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
615 			i = tp->t_srtt *
616 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
617 			if (rt->rt_rmx.rmx_rtt && i)
618 				/*
619 				 * filter this update to half the old & half
620 				 * the new values, converting scale.
621 				 * See route.h and tcp_var.h for a
622 				 * description of the scaling constants.
623 				 */
624 				rt->rt_rmx.rmx_rtt =
625 				    (rt->rt_rmx.rmx_rtt + i) / 2;
626 			else
627 				rt->rt_rmx.rmx_rtt = i;
628 			tcpstat.tcps_cachedrtt++;
629 		}
630 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
631 			i = tp->t_rttvar *
632 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
633 			if (rt->rt_rmx.rmx_rttvar && i)
634 				rt->rt_rmx.rmx_rttvar =
635 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
636 			else
637 				rt->rt_rmx.rmx_rttvar = i;
638 			tcpstat.tcps_cachedrttvar++;
639 		}
640 		/*
641 		 * The old comment here said:
642 		 * update the pipelimit (ssthresh) if it has been updated
643 		 * already or if a pipesize was specified & the threshhold
644 		 * got below half the pipesize.  I.e., wait for bad news
645 		 * before we start updating, then update on both good
646 		 * and bad news.
647 		 *
648 		 * But we want to save the ssthresh even if no pipesize is
649 		 * specified explicitly in the route, because such
650 		 * connections still have an implicit pipesize specified
651 		 * by the global tcp_sendspace.  In the absence of a reliable
652 		 * way to calculate the pipesize, it will have to do.
653 		 */
654 		i = tp->snd_ssthresh;
655 		if (rt->rt_rmx.rmx_sendpipe != 0)
656 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
657 		else
658 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
659 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
660 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
661 		    || dosavessthresh) {
662 			/*
663 			 * convert the limit from user data bytes to
664 			 * packets then to packet data bytes.
665 			 */
666 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
667 			if (i < 2)
668 				i = 2;
669 			i *= (u_long)(tp->t_maxseg +
670 #ifdef INET6
671 				      (isipv6 ? sizeof (struct ip6_hdr) +
672 					       sizeof (struct tcphdr) :
673 #endif
674 				       sizeof (struct tcpiphdr)
675 #ifdef INET6
676 				       )
677 #endif
678 				      );
679 			if (rt->rt_rmx.rmx_ssthresh)
680 				rt->rt_rmx.rmx_ssthresh =
681 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
682 			else
683 				rt->rt_rmx.rmx_ssthresh = i;
684 			tcpstat.tcps_cachedssthresh++;
685 		}
686 	}
687     no_valid_rt:
688 	/* free the reassembly queue, if any */
689 	while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
690 		LIST_REMOVE(q, tqe_q);
691 		m_freem(q->tqe_m);
692 		FREE(q, M_TSEGQ);
693 	}
694 	if (tp->t_template)
695 		(void) m_free(dtom(tp->t_template));
696 	inp->inp_ppcb = NULL;
697 	soisdisconnected(so);
698 #ifdef INET6
699 	if (INP_CHECK_SOCKAF(so, AF_INET6))
700 		in6_pcbdetach(inp);
701 	else
702 #endif /* INET6 */
703 	in_pcbdetach(inp);
704 	tcpstat.tcps_closed++;
705 	return ((struct tcpcb *)0);
706 }
707 
708 void
709 tcp_drain()
710 {
711 	if (do_tcpdrain)
712 	{
713 		struct inpcb *inpb;
714 		struct tcpcb *tcpb;
715 		struct tseg_qent *te;
716 
717 	/*
718 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
719 	 * if there is one...
720 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
721 	 *      reassembly queue should be flushed, but in a situation
722 	 * 	where we're really low on mbufs, this is potentially
723 	 *  	usefull.
724 	 */
725 		for (inpb = tcbinfo.listhead->lh_first; inpb;
726 	    		inpb = inpb->inp_list.le_next) {
727 				if ((tcpb = intotcpcb(inpb))) {
728 					while ((te = LIST_FIRST(&tcpb->t_segq))
729 					       != NULL) {
730 					LIST_REMOVE(te, tqe_q);
731 					m_freem(te->tqe_m);
732 					FREE(te, M_TSEGQ);
733 				}
734 			}
735 		}
736 
737 	}
738 }
739 
740 /*
741  * Notify a tcp user of an asynchronous error;
742  * store error as soft error, but wake up user
743  * (for now, won't do anything until can select for soft error).
744  */
745 static void
746 tcp_notify(inp, error)
747 	struct inpcb *inp;
748 	int error;
749 {
750 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
751 	register struct socket *so = inp->inp_socket;
752 
753 	/*
754 	 * Ignore some errors if we are hooked up.
755 	 * If connection hasn't completed, has retransmitted several times,
756 	 * and receives a second error, give up now.  This is better
757 	 * than waiting a long time to establish a connection that
758 	 * can never complete.
759 	 */
760 	if (tp->t_state == TCPS_ESTABLISHED &&
761 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
762 	      error == EHOSTDOWN)) {
763 		return;
764 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
765 	    tp->t_softerror)
766 		so->so_error = error;
767 	else
768 		tp->t_softerror = error;
769 	wakeup((caddr_t) &so->so_timeo);
770 	sorwakeup(so);
771 	sowwakeup(so);
772 }
773 
774 static int
775 tcp_pcblist SYSCTL_HANDLER_ARGS
776 {
777 	int error, i, n, s;
778 	struct inpcb *inp, **inp_list;
779 	inp_gen_t gencnt;
780 	struct xinpgen xig;
781 
782 	/*
783 	 * The process of preparing the TCB list is too time-consuming and
784 	 * resource-intensive to repeat twice on every request.
785 	 */
786 	if (req->oldptr == 0) {
787 		n = tcbinfo.ipi_count;
788 		req->oldidx = 2 * (sizeof xig)
789 			+ (n + n/8) * sizeof(struct xtcpcb);
790 		return 0;
791 	}
792 
793 	if (req->newptr != 0)
794 		return EPERM;
795 
796 	/*
797 	 * OK, now we're committed to doing something.
798 	 */
799 	s = splnet();
800 	gencnt = tcbinfo.ipi_gencnt;
801 	n = tcbinfo.ipi_count;
802 	splx(s);
803 
804 	xig.xig_len = sizeof xig;
805 	xig.xig_count = n;
806 	xig.xig_gen = gencnt;
807 	xig.xig_sogen = so_gencnt;
808 	error = SYSCTL_OUT(req, &xig, sizeof xig);
809 	if (error)
810 		return error;
811 
812 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
813 	if (inp_list == 0)
814 		return ENOMEM;
815 
816 	s = splnet();
817 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
818 	     inp = inp->inp_list.le_next) {
819 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
820 			inp_list[i++] = inp;
821 	}
822 	splx(s);
823 	n = i;
824 
825 	error = 0;
826 	for (i = 0; i < n; i++) {
827 		inp = inp_list[i];
828 		if (inp->inp_gencnt <= gencnt) {
829 			struct xtcpcb xt;
830 			caddr_t inp_ppcb;
831 			xt.xt_len = sizeof xt;
832 			/* XXX should avoid extra copy */
833 			bcopy(inp, &xt.xt_inp, sizeof *inp);
834 			inp_ppcb = inp->inp_ppcb;
835 			if (inp_ppcb != NULL)
836 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
837 			else
838 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
839 			if (inp->inp_socket)
840 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
841 			error = SYSCTL_OUT(req, &xt, sizeof xt);
842 		}
843 	}
844 	if (!error) {
845 		/*
846 		 * Give the user an updated idea of our state.
847 		 * If the generation differs from what we told
848 		 * her before, she knows that something happened
849 		 * while we were processing this request, and it
850 		 * might be necessary to retry.
851 		 */
852 		s = splnet();
853 		xig.xig_gen = tcbinfo.ipi_gencnt;
854 		xig.xig_sogen = so_gencnt;
855 		xig.xig_count = tcbinfo.ipi_count;
856 		splx(s);
857 		error = SYSCTL_OUT(req, &xig, sizeof xig);
858 	}
859 	free(inp_list, M_TEMP);
860 	return error;
861 }
862 
863 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
864 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
865 
866 static int
867 tcp_getcred SYSCTL_HANDLER_ARGS
868 {
869 	struct sockaddr_in addrs[2];
870 	struct inpcb *inp;
871 	int error, s;
872 
873 	error = suser(req->p);
874 	if (error)
875 		return (error);
876 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
877 	if (error)
878 		return (error);
879 	s = splnet();
880 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
881 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
882 	if (inp == NULL || inp->inp_socket == NULL) {
883 		error = ENOENT;
884 		goto out;
885 	}
886 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
887 out:
888 	splx(s);
889 	return (error);
890 }
891 
892 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
893     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
894 
895 #ifdef INET6
896 static int
897 tcp6_getcred SYSCTL_HANDLER_ARGS
898 {
899 	struct sockaddr_in6 addrs[2];
900 	struct inpcb *inp;
901 	int error, s, mapped = 0;
902 
903 	error = suser(req->p);
904 	if (error)
905 		return (error);
906 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
907 	if (error)
908 		return (error);
909 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
910 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
911 			mapped = 1;
912 		else
913 			return (EINVAL);
914 	}
915 	s = splnet();
916 	if (mapped == 1)
917 		inp = in_pcblookup_hash(&tcbinfo,
918 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
919 			addrs[1].sin6_port,
920 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
921 			addrs[0].sin6_port,
922 			0, NULL);
923 	else
924 		inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
925 				 addrs[1].sin6_port,
926 				 &addrs[0].sin6_addr, addrs[0].sin6_port,
927 				 0, NULL);
928 	if (inp == NULL || inp->inp_socket == NULL) {
929 		error = ENOENT;
930 		goto out;
931 	}
932 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred,
933 			   sizeof(struct ucred));
934 out:
935 	splx(s);
936 	return (error);
937 }
938 
939 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
940 	    0, 0,
941 	    tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
942 #endif
943 
944 
945 void
946 tcp_ctlinput(cmd, sa, vip)
947 	int cmd;
948 	struct sockaddr *sa;
949 	void *vip;
950 {
951 	register struct ip *ip = vip;
952 	register struct tcphdr *th;
953 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
954 
955 	if (cmd == PRC_QUENCH)
956 		notify = tcp_quench;
957 	else if (cmd == PRC_MSGSIZE)
958 		notify = tcp_mtudisc;
959 	else if (!PRC_IS_REDIRECT(cmd) &&
960 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
961 		return;
962 	if (ip) {
963 		th = (struct tcphdr *)((caddr_t)ip
964 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
965 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
966 			cmd, notify);
967 	} else
968 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
969 }
970 
971 #ifdef INET6
972 void
973 tcp6_ctlinput(cmd, sa, d)
974 	int cmd;
975 	struct sockaddr *sa;
976 	void *d;
977 {
978 	register struct tcphdr *thp;
979 	struct tcphdr th;
980 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
981 	struct sockaddr_in6 sa6;
982 	struct ip6_hdr *ip6;
983 	struct mbuf *m;
984 	int off;
985 
986 	if (sa->sa_family != AF_INET6 ||
987 	    sa->sa_len != sizeof(struct sockaddr_in6))
988 		return;
989 
990 	if (cmd == PRC_QUENCH)
991 		notify = tcp_quench;
992 	else if (cmd == PRC_MSGSIZE)
993 		notify = tcp_mtudisc;
994 	else if (!PRC_IS_REDIRECT(cmd) &&
995 		 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
996 		return;
997 
998 	/* if the parameter is from icmp6, decode it. */
999 	if (d != NULL) {
1000 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1001 		m = ip6cp->ip6c_m;
1002 		ip6 = ip6cp->ip6c_ip6;
1003 		off = ip6cp->ip6c_off;
1004 	} else {
1005 		m = NULL;
1006 		ip6 = NULL;
1007 	}
1008 
1009 	/*
1010 	 * Translate addresses into internal form.
1011 	 * Sa check if it is AF_INET6 is done at the top of this funciton.
1012 	 */
1013 	sa6 = *(struct sockaddr_in6 *)sa;
1014 	if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) != 0 && m != NULL &&
1015 	    m->m_pkthdr.rcvif != NULL)
1016 		sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1017 
1018 	if (ip6) {
1019 		/*
1020 		 * XXX: We assume that when IPV6 is non NULL,
1021 		 * M and OFF are valid.
1022 		 */
1023 		struct in6_addr s;
1024 
1025 		/* translate addresses into internal form */
1026 		memcpy(&s, &ip6->ip6_src, sizeof(s));
1027 		if (IN6_IS_ADDR_LINKLOCAL(&s) != 0 && m != NULL &&
1028 		    m->m_pkthdr.rcvif != NULL)
1029 			s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1030 
1031 		if (m->m_len < off + sizeof(*thp)) {
1032 			/*
1033 			 * this should be rare case
1034 			 * because now MINCLSIZE is "(MHLEN + 1)",
1035 			 * so we compromise on this copy...
1036 			 */
1037 			m_copydata(m, off, sizeof(th), (caddr_t)&th);
1038 			thp = &th;
1039 		} else
1040 			thp = (struct tcphdr *)(mtod(m, caddr_t) + off);
1041 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, thp->th_dport,
1042 			      &s, thp->th_sport, cmd, notify);
1043 	} else
1044 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, 0, &zeroin6_addr,
1045 			      0, cmd, notify);
1046 }
1047 #endif /* INET6 */
1048 
1049 /*
1050  * When a source quench is received, close congestion window
1051  * to one segment.  We will gradually open it again as we proceed.
1052  */
1053 void
1054 tcp_quench(inp, errno)
1055 	struct inpcb *inp;
1056 	int errno;
1057 {
1058 	struct tcpcb *tp = intotcpcb(inp);
1059 
1060 	if (tp)
1061 		tp->snd_cwnd = tp->t_maxseg;
1062 }
1063 
1064 /*
1065  * When `need fragmentation' ICMP is received, update our idea of the MSS
1066  * based on the new value in the route.  Also nudge TCP to send something,
1067  * since we know the packet we just sent was dropped.
1068  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1069  */
1070 void
1071 tcp_mtudisc(inp, errno)
1072 	struct inpcb *inp;
1073 	int errno;
1074 {
1075 	struct tcpcb *tp = intotcpcb(inp);
1076 	struct rtentry *rt;
1077 	struct rmxp_tao *taop;
1078 	struct socket *so = inp->inp_socket;
1079 	int offered;
1080 	int mss;
1081 #ifdef INET6
1082 	int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1083 #endif /* INET6 */
1084 
1085 	if (tp) {
1086 #ifdef INET6
1087 		if (isipv6)
1088 			rt = tcp_rtlookup6(inp);
1089 		else
1090 #endif /* INET6 */
1091 		rt = tcp_rtlookup(inp);
1092 		if (!rt || !rt->rt_rmx.rmx_mtu) {
1093 			tp->t_maxopd = tp->t_maxseg =
1094 #ifdef INET6
1095 				isipv6 ? tcp_v6mssdflt :
1096 #endif /* INET6 */
1097 				tcp_mssdflt;
1098 			return;
1099 		}
1100 		taop = rmx_taop(rt->rt_rmx);
1101 		offered = taop->tao_mssopt;
1102 		mss = rt->rt_rmx.rmx_mtu -
1103 #ifdef INET6
1104 			(isipv6 ?
1105 			 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1106 #endif /* INET6 */
1107 			 sizeof(struct tcpiphdr)
1108 #ifdef INET6
1109 			 )
1110 #endif /* INET6 */
1111 			;
1112 
1113 		if (offered)
1114 			mss = min(mss, offered);
1115 		/*
1116 		 * XXX - The above conditional probably violates the TCP
1117 		 * spec.  The problem is that, since we don't know the
1118 		 * other end's MSS, we are supposed to use a conservative
1119 		 * default.  But, if we do that, then MTU discovery will
1120 		 * never actually take place, because the conservative
1121 		 * default is much less than the MTUs typically seen
1122 		 * on the Internet today.  For the moment, we'll sweep
1123 		 * this under the carpet.
1124 		 *
1125 		 * The conservative default might not actually be a problem
1126 		 * if the only case this occurs is when sending an initial
1127 		 * SYN with options and data to a host we've never talked
1128 		 * to before.  Then, they will reply with an MSS value which
1129 		 * will get recorded and the new parameters should get
1130 		 * recomputed.  For Further Study.
1131 		 */
1132 		if (tp->t_maxopd <= mss)
1133 			return;
1134 		tp->t_maxopd = mss;
1135 
1136 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1137 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1138 			mss -= TCPOLEN_TSTAMP_APPA;
1139 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1140 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1141 			mss -= TCPOLEN_CC_APPA;
1142 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1143 		if (mss > MCLBYTES)
1144 			mss &= ~(MCLBYTES-1);
1145 #else
1146 		if (mss > MCLBYTES)
1147 			mss = mss / MCLBYTES * MCLBYTES;
1148 #endif
1149 		if (so->so_snd.sb_hiwat < mss)
1150 			mss = so->so_snd.sb_hiwat;
1151 
1152 		tp->t_maxseg = mss;
1153 
1154 		tcpstat.tcps_mturesent++;
1155 		tp->t_rtttime = 0;
1156 		tp->snd_nxt = tp->snd_una;
1157 		tcp_output(tp);
1158 	}
1159 }
1160 
1161 /*
1162  * Look-up the routing entry to the peer of this inpcb.  If no route
1163  * is found and it cannot be allocated the return NULL.  This routine
1164  * is called by TCP routines that access the rmx structure and by tcp_mss
1165  * to get the interface MTU.
1166  */
1167 struct rtentry *
1168 tcp_rtlookup(inp)
1169 	struct inpcb *inp;
1170 {
1171 	struct route *ro;
1172 	struct rtentry *rt;
1173 
1174 	ro = &inp->inp_route;
1175 	rt = ro->ro_rt;
1176 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1177 		/* No route yet, so try to acquire one */
1178 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1179 			ro->ro_dst.sa_family = AF_INET;
1180 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1181 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1182 				inp->inp_faddr;
1183 			rtalloc(ro);
1184 			rt = ro->ro_rt;
1185 		}
1186 	}
1187 	return rt;
1188 }
1189 
1190 #ifdef INET6
1191 struct rtentry *
1192 tcp_rtlookup6(inp)
1193 	struct inpcb *inp;
1194 {
1195 	struct route_in6 *ro6;
1196 	struct rtentry *rt;
1197 
1198 	ro6 = &inp->in6p_route;
1199 	rt = ro6->ro_rt;
1200 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1201 		/* No route yet, so try to acquire one */
1202 		if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1203 			ro6->ro_dst.sin6_family = AF_INET6;
1204 			ro6->ro_dst.sin6_len = sizeof(ro6->ro_dst);
1205 			ro6->ro_dst.sin6_addr = inp->in6p_faddr;
1206 			rtalloc((struct route *)ro6);
1207 			rt = ro6->ro_rt;
1208 		}
1209 	}
1210 	return rt;
1211 }
1212 #endif /* INET6 */
1213 
1214 #ifdef IPSEC
1215 /* compute ESP/AH header size for TCP, including outer IP header. */
1216 size_t
1217 ipsec_hdrsiz_tcp(tp)
1218 	struct tcpcb *tp;
1219 {
1220 	struct inpcb *inp;
1221 	struct mbuf *m;
1222 	size_t hdrsiz;
1223 	struct ip *ip;
1224 #ifdef INET6
1225 	struct ip6_hdr *ip6;
1226 #endif /* INET6 */
1227 	struct tcphdr *th;
1228 
1229 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1230 		return 0;
1231 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1232 	if (!m)
1233 		return 0;
1234 
1235 #ifdef INET6
1236 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1237 		ip6 = mtod(m, struct ip6_hdr *);
1238 		th = (struct tcphdr *)(ip6 + 1);
1239 		m->m_pkthdr.len = m->m_len =
1240 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1241 		bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip6,
1242 		      sizeof(struct ip6_hdr));
1243 		bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1244 		      sizeof(struct tcphdr));
1245 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1246 	} else
1247 #endif /* INET6 */
1248       {
1249 	ip = mtod(m, struct ip *);
1250 	th = (struct tcphdr *)(ip + 1);
1251 	m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1252 	bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip,
1253 	      sizeof(struct ip));
1254 	bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1255 	      sizeof(struct tcphdr));
1256 	hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1257       }
1258 
1259 	m_free(m);
1260 	return hdrsiz;
1261 }
1262 #endif /*IPSEC*/
1263 
1264 /*
1265  * Return a pointer to the cached information about the remote host.
1266  * The cached information is stored in the protocol specific part of
1267  * the route metrics.
1268  */
1269 struct rmxp_tao *
1270 tcp_gettaocache(inp)
1271 	struct inpcb *inp;
1272 {
1273 	struct rtentry *rt;
1274 
1275 #ifdef INET6
1276 	if ((inp->inp_vflag & INP_IPV6) != 0)
1277 		rt = tcp_rtlookup6(inp);
1278 	else
1279 #endif /* INET6 */
1280 	rt = tcp_rtlookup(inp);
1281 
1282 	/* Make sure this is a host route and is up. */
1283 	if (rt == NULL ||
1284 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1285 		return NULL;
1286 
1287 	return rmx_taop(rt->rt_rmx);
1288 }
1289 
1290 /*
1291  * Clear all the TAO cache entries, called from tcp_init.
1292  *
1293  * XXX
1294  * This routine is just an empty one, because we assume that the routing
1295  * routing tables are initialized at the same time when TCP, so there is
1296  * nothing in the cache left over.
1297  */
1298 static void
1299 tcp_cleartaocache()
1300 {
1301 }
1302