xref: /freebsd/sys/netinet/tcp_timewait.c (revision 817420dc8eac7df799c78f5309b75092b7f7cd40)
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 <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/proc.h>
53 #include <sys/socket.h>
54 #include <sys/socketvar.h>
55 #include <sys/protosw.h>
56 
57 #include <vm/vm_zone.h>
58 
59 #include <net/route.h>
60 #include <net/if.h>
61 
62 #define _IP_VHL
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #ifdef INET6
67 #include <netinet/ip6.h>
68 #endif
69 #include <netinet/in_pcb.h>
70 #ifdef INET6
71 #include <netinet6/in6_pcb.h>
72 #endif
73 #include <netinet/in_var.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #endif
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #ifdef INET6
84 #include <netinet6/tcp6_var.h>
85 #endif
86 #include <netinet/tcpip.h>
87 #ifdef TCPDEBUG
88 #include <netinet/tcp_debug.h>
89 #endif
90 #include <netinet6/ip6protosw.h>
91 
92 #ifdef IPSEC
93 #include <netinet6/ipsec.h>
94 #ifdef INET6
95 #include <netinet6/ipsec6.h>
96 #endif
97 #endif /*IPSEC*/
98 
99 #include <machine/in_cksum.h>
100 
101 int 	tcp_mssdflt = TCP_MSS;
102 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
103     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
104 
105 #ifdef INET6
106 int	tcp_v6mssdflt = TCP6_MSS;
107 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
108 	CTLFLAG_RW, &tcp_v6mssdflt , 0,
109 	"Default TCP Maximum Segment Size for IPv6");
110 #endif
111 
112 #if 0
113 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
114 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
115     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
116 #endif
117 
118 static int	tcp_do_rfc1323 = 1;
119 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
120     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
121 
122 static int	tcp_do_rfc1644 = 0;
123 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
124     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
125 
126 static int	tcp_tcbhashsize = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
128      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
129 
130 static int	do_tcpdrain = 1;
131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
132      "Enable tcp_drain routine for extra help when low on mbufs");
133 
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
135     &tcbinfo.ipi_count, 0, "Number of active PCBs");
136 
137 static void	tcp_cleartaocache __P((void));
138 static void	tcp_notify __P((struct inpcb *, int));
139 
140 /*
141  * Target size of TCP PCB hash tables. Must be a power of two.
142  *
143  * Note that this can be overridden by the kernel environment
144  * variable net.inet.tcp.tcbhashsize
145  */
146 #ifndef TCBHASHSIZE
147 #define TCBHASHSIZE	512
148 #endif
149 
150 /*
151  * This is the actual shape of what we allocate using the zone
152  * allocator.  Doing it this way allows us to protect both structures
153  * using the same generation count, and also eliminates the overhead
154  * of allocating tcpcbs separately.  By hiding the structure here,
155  * we avoid changing most of the rest of the code (although it needs
156  * to be changed, eventually, for greater efficiency).
157  */
158 #define	ALIGNMENT	32
159 #define	ALIGNM1		(ALIGNMENT - 1)
160 struct	inp_tp {
161 	union {
162 		struct	inpcb inp;
163 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
164 	} inp_tp_u;
165 	struct	tcpcb tcb;
166 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
167 	struct	callout inp_tp_delack;
168 };
169 #undef ALIGNMENT
170 #undef ALIGNM1
171 
172 /*
173  * Tcp initialization
174  */
175 void
176 tcp_init()
177 {
178 	int hashsize;
179 
180 	tcp_iss = arc4random();	/* wrong, but better than a constant */
181 	tcp_ccgen = 1;
182 	tcp_cleartaocache();
183 
184 	tcp_delacktime = TCPTV_DELACK;
185 	tcp_keepinit = TCPTV_KEEP_INIT;
186 	tcp_keepidle = TCPTV_KEEP_IDLE;
187 	tcp_keepintvl = TCPTV_KEEPINTVL;
188 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
189 	tcp_msl = TCPTV_MSL;
190 
191 	LIST_INIT(&tcb);
192 	tcbinfo.listhead = &tcb;
193 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
194 	if (!powerof2(hashsize)) {
195 		printf("WARNING: TCB hash size not a power of 2\n");
196 		hashsize = 512; /* safe default */
197 	}
198 	tcp_tcbhashsize = hashsize;
199 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
200 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
201 					&tcbinfo.porthashmask);
202 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
203 				 ZONE_INTERRUPT, 0);
204 #ifdef INET6
205 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
206 #else /* INET6 */
207 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
208 #endif /* INET6 */
209 	if (max_protohdr < TCP_MINPROTOHDR)
210 		max_protohdr = TCP_MINPROTOHDR;
211 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
212 		panic("tcp_init");
213 #undef TCP_MINPROTOHDR
214 }
215 
216 /*
217  * Create template to be used to send tcp packets on a connection.
218  * Call after host entry created, allocates an mbuf and fills
219  * in a skeletal tcp/ip header, minimizing the amount of work
220  * necessary when the connection is used.
221  */
222 struct tcptemp *
223 tcp_template(tp)
224 	struct tcpcb *tp;
225 {
226 	register struct inpcb *inp = tp->t_inpcb;
227 	register struct mbuf *m;
228 	register struct tcptemp *n;
229 
230 	if ((n = tp->t_template) == 0) {
231 		m = m_get(M_DONTWAIT, MT_HEADER);
232 		if (m == NULL)
233 			return (0);
234 		m->m_len = sizeof (struct tcptemp);
235 		n = mtod(m, struct tcptemp *);
236 	}
237 #ifdef INET6
238 	if ((inp->inp_vflag & INP_IPV6) != 0) {
239 		register struct ip6_hdr *ip6;
240 
241 		ip6 = (struct ip6_hdr *)n->tt_ipgen;
242 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
243 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
244 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
245 			(IPV6_VERSION & IPV6_VERSION_MASK);
246 		ip6->ip6_nxt = IPPROTO_TCP;
247 		ip6->ip6_plen = sizeof(struct tcphdr);
248 		ip6->ip6_src = inp->in6p_laddr;
249 		ip6->ip6_dst = inp->in6p_faddr;
250 		n->tt_t.th_sum = 0;
251 	} else
252 #endif
253       {
254 	struct ip *ip = (struct ip *)n->tt_ipgen;
255 
256 	bzero(ip, sizeof(struct ip));		/* XXX overkill? */
257 	ip->ip_vhl = IP_VHL_BORING;
258 	ip->ip_p = IPPROTO_TCP;
259 	ip->ip_src = inp->inp_laddr;
260 	ip->ip_dst = inp->inp_faddr;
261 	n->tt_t.th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
262 	    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
263       }
264 	n->tt_t.th_sport = inp->inp_lport;
265 	n->tt_t.th_dport = inp->inp_fport;
266 	n->tt_t.th_seq = 0;
267 	n->tt_t.th_ack = 0;
268 	n->tt_t.th_x2 = 0;
269 	n->tt_t.th_off = 5;
270 	n->tt_t.th_flags = 0;
271 	n->tt_t.th_win = 0;
272 	n->tt_t.th_urp = 0;
273 	return (n);
274 }
275 
276 /*
277  * Send a single message to the TCP at address specified by
278  * the given TCP/IP header.  If m == 0, then we make a copy
279  * of the tcpiphdr at ti and send directly to the addressed host.
280  * This is used to force keep alive messages out using the TCP
281  * template for a connection tp->t_template.  If flags are given
282  * then we send a message back to the TCP which originated the
283  * segment ti, and discard the mbuf containing it and any other
284  * attached mbufs.
285  *
286  * In any case the ack and sequence number of the transmitted
287  * segment are as specified by the parameters.
288  *
289  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
290  */
291 void
292 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
293 	struct tcpcb *tp;
294 	void *ipgen;
295 	register struct tcphdr *th;
296 	register struct mbuf *m;
297 	tcp_seq ack, seq;
298 	int flags;
299 {
300 	register int tlen;
301 	int win = 0;
302 	struct route *ro = 0;
303 	struct route sro;
304 	struct ip *ip;
305 	struct tcphdr *nth;
306 #ifdef INET6
307 	struct route_in6 *ro6 = 0;
308 	struct route_in6 sro6;
309 	struct ip6_hdr *ip6;
310 	int isipv6;
311 #endif /* INET6 */
312 	int ipflags = 0;
313 
314 #ifdef INET6
315 	isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
316 	ip6 = ipgen;
317 #endif /* INET6 */
318 	ip = ipgen;
319 
320 	if (tp) {
321 		if (!(flags & TH_RST)) {
322 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
323 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
324 				win = (long)TCP_MAXWIN << tp->rcv_scale;
325 		}
326 #ifdef INET6
327 		if (isipv6)
328 			ro6 = &tp->t_inpcb->in6p_route;
329 		else
330 #endif /* INET6 */
331 		ro = &tp->t_inpcb->inp_route;
332 	} else {
333 #ifdef INET6
334 		if (isipv6) {
335 			ro6 = &sro6;
336 			bzero(ro6, sizeof *ro6);
337 		} else
338 #endif /* INET6 */
339 	      {
340 		ro = &sro;
341 		bzero(ro, sizeof *ro);
342 	      }
343 	}
344 	if (m == 0) {
345 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
346 		if (m == NULL)
347 			return;
348 #ifdef TCP_COMPAT_42
349 		tlen = 1;
350 #else
351 		tlen = 0;
352 #endif
353 		m->m_data += max_linkhdr;
354 #ifdef INET6
355 		if (isipv6) {
356 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
357 			      sizeof(struct ip6_hdr));
358 			ip6 = mtod(m, struct ip6_hdr *);
359 			nth = (struct tcphdr *)(ip6 + 1);
360 		} else
361 #endif /* INET6 */
362 	      {
363 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
364 		ip = mtod(m, struct ip *);
365 		nth = (struct tcphdr *)(ip + 1);
366 	      }
367 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
368 		flags = TH_ACK;
369 	} else {
370 		m_freem(m->m_next);
371 		m->m_next = 0;
372 		m->m_data = (caddr_t)ipgen;
373 		/* m_len is set later */
374 		tlen = 0;
375 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
376 #ifdef INET6
377 		if (isipv6) {
378 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
379 			nth = (struct tcphdr *)(ip6 + 1);
380 		} else
381 #endif /* INET6 */
382 	      {
383 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
384 		nth = (struct tcphdr *)(ip + 1);
385 	      }
386 		if (th != nth) {
387 			/*
388 			 * this is usually a case when an extension header
389 			 * exists between the IPv6 header and the
390 			 * TCP header.
391 			 */
392 			nth->th_sport = th->th_sport;
393 			nth->th_dport = th->th_dport;
394 		}
395 		xchg(nth->th_dport, nth->th_sport, n_short);
396 #undef xchg
397 	}
398 #ifdef INET6
399 	if (isipv6) {
400 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
401 						tlen));
402 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
403 	} else
404 #endif
405       {
406 	tlen += sizeof (struct tcpiphdr);
407 	ip->ip_len = tlen;
408 	ip->ip_ttl = ip_defttl;
409       }
410 	m->m_len = tlen;
411 	m->m_pkthdr.len = tlen;
412 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
413 	nth->th_seq = htonl(seq);
414 	nth->th_ack = htonl(ack);
415 	nth->th_x2 = 0;
416 	nth->th_off = sizeof (struct tcphdr) >> 2;
417 	nth->th_flags = flags;
418 	if (tp)
419 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
420 	else
421 		nth->th_win = htons((u_short)win);
422 	nth->th_urp = 0;
423 #ifdef INET6
424 	if (isipv6) {
425 		nth->th_sum = 0;
426 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
427 					sizeof(struct ip6_hdr),
428 					tlen - sizeof(struct ip6_hdr));
429 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
430 					       ro6 && ro6->ro_rt ?
431 					       ro6->ro_rt->rt_ifp :
432 					       NULL);
433 	} else
434 #endif /* INET6 */
435       {
436         nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
437 	    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
438         m->m_pkthdr.csum_flags = CSUM_TCP;
439         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
440       }
441 #ifdef TCPDEBUG
442 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
443 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
444 #endif
445 #ifdef IPSEC
446 	ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL);
447 #endif
448 #ifdef INET6
449 	if (isipv6) {
450 		(void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
451 		if (ro6 == &sro6 && ro6->ro_rt) {
452 			RTFREE(ro6->ro_rt);
453 			ro6->ro_rt = NULL;
454 		}
455 	} else
456 #endif /* INET6 */
457       {
458 	(void) ip_output(m, NULL, ro, ipflags, NULL);
459 	if (ro == &sro && ro->ro_rt) {
460 		RTFREE(ro->ro_rt);
461 		ro->ro_rt = NULL;
462 	}
463       }
464 }
465 
466 /*
467  * Create a new TCP control block, making an
468  * empty reassembly queue and hooking it to the argument
469  * protocol control block.  The `inp' parameter must have
470  * come from the zone allocator set up in tcp_init().
471  */
472 struct tcpcb *
473 tcp_newtcpcb(inp)
474 	struct inpcb *inp;
475 {
476 	struct inp_tp *it;
477 	register struct tcpcb *tp;
478 #ifdef INET6
479 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
480 #endif /* INET6 */
481 
482 	it = (struct inp_tp *)inp;
483 	tp = &it->tcb;
484 	bzero((char *) tp, sizeof(struct tcpcb));
485 	LIST_INIT(&tp->t_segq);
486 	tp->t_maxseg = tp->t_maxopd =
487 #ifdef INET6
488 		isipv6 ? tcp_v6mssdflt :
489 #endif /* INET6 */
490 		tcp_mssdflt;
491 
492 	/* Set up our timeouts. */
493 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt);
494 	callout_init(tp->tt_persist = &it->inp_tp_persist);
495 	callout_init(tp->tt_keep = &it->inp_tp_keep);
496 	callout_init(tp->tt_2msl = &it->inp_tp_2msl);
497 	callout_init(tp->tt_delack = &it->inp_tp_delack);
498 
499 	if (tcp_do_rfc1323)
500 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
501 	if (tcp_do_rfc1644)
502 		tp->t_flags |= TF_REQ_CC;
503 	tp->t_inpcb = inp;	/* XXX */
504 	/*
505 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
506 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
507 	 * reasonable initial retransmit time.
508 	 */
509 	tp->t_srtt = TCPTV_SRTTBASE;
510 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
511 	tp->t_rttmin = TCPTV_MIN;
512 	tp->t_rxtcur = TCPTV_RTOBASE;
513 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
514 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
515 	tp->t_rcvtime = ticks;
516         /*
517 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
518 	 * because the socket may be bound to an IPv6 wildcard address,
519 	 * which may match an IPv4-mapped IPv6 address.
520 	 */
521 	inp->inp_ip_ttl = ip_defttl;
522 	inp->inp_ppcb = (caddr_t)tp;
523 	return (tp);		/* XXX */
524 }
525 
526 /*
527  * Drop a TCP connection, reporting
528  * the specified error.  If connection is synchronized,
529  * then send a RST to peer.
530  */
531 struct tcpcb *
532 tcp_drop(tp, errno)
533 	register struct tcpcb *tp;
534 	int errno;
535 {
536 	struct socket *so = tp->t_inpcb->inp_socket;
537 
538 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
539 		tp->t_state = TCPS_CLOSED;
540 		(void) tcp_output(tp);
541 		tcpstat.tcps_drops++;
542 	} else
543 		tcpstat.tcps_conndrops++;
544 	if (errno == ETIMEDOUT && tp->t_softerror)
545 		errno = tp->t_softerror;
546 	so->so_error = errno;
547 	return (tcp_close(tp));
548 }
549 
550 /*
551  * Close a TCP control block:
552  *	discard all space held by the tcp
553  *	discard internet protocol block
554  *	wake up any sleepers
555  */
556 struct tcpcb *
557 tcp_close(tp)
558 	register struct tcpcb *tp;
559 {
560 	register struct tseg_qent *q;
561 	struct inpcb *inp = tp->t_inpcb;
562 	struct socket *so = inp->inp_socket;
563 #ifdef INET6
564 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
565 #endif /* INET6 */
566 	register struct rtentry *rt;
567 	int dosavessthresh;
568 
569 	/*
570 	 * Make sure that all of our timers are stopped before we
571 	 * delete the PCB.
572 	 */
573 	callout_stop(tp->tt_rexmt);
574 	callout_stop(tp->tt_persist);
575 	callout_stop(tp->tt_keep);
576 	callout_stop(tp->tt_2msl);
577 	callout_stop(tp->tt_delack);
578 
579 	/*
580 	 * If we got enough samples through the srtt filter,
581 	 * save the rtt and rttvar in the routing entry.
582 	 * 'Enough' is arbitrarily defined as the 16 samples.
583 	 * 16 samples is enough for the srtt filter to converge
584 	 * to within 5% of the correct value; fewer samples and
585 	 * we could save a very bogus rtt.
586 	 *
587 	 * Don't update the default route's characteristics and don't
588 	 * update anything that the user "locked".
589 	 */
590 	if (tp->t_rttupdated >= 16) {
591 		register u_long i = 0;
592 #ifdef INET6
593 		if (isipv6) {
594 			struct sockaddr_in6 *sin6;
595 
596 			if ((rt = inp->in6p_route.ro_rt) == NULL)
597 				goto no_valid_rt;
598 			sin6 = (struct sockaddr_in6 *)rt_key(rt);
599 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
600 				goto no_valid_rt;
601 		}
602 		else
603 #endif /* INET6 */
604 		if ((rt = inp->inp_route.ro_rt) == NULL ||
605 		    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
606 		    == INADDR_ANY)
607 			goto no_valid_rt;
608 
609 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
610 			i = tp->t_srtt *
611 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
612 			if (rt->rt_rmx.rmx_rtt && i)
613 				/*
614 				 * filter this update to half the old & half
615 				 * the new values, converting scale.
616 				 * See route.h and tcp_var.h for a
617 				 * description of the scaling constants.
618 				 */
619 				rt->rt_rmx.rmx_rtt =
620 				    (rt->rt_rmx.rmx_rtt + i) / 2;
621 			else
622 				rt->rt_rmx.rmx_rtt = i;
623 			tcpstat.tcps_cachedrtt++;
624 		}
625 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
626 			i = tp->t_rttvar *
627 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
628 			if (rt->rt_rmx.rmx_rttvar && i)
629 				rt->rt_rmx.rmx_rttvar =
630 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
631 			else
632 				rt->rt_rmx.rmx_rttvar = i;
633 			tcpstat.tcps_cachedrttvar++;
634 		}
635 		/*
636 		 * The old comment here said:
637 		 * update the pipelimit (ssthresh) if it has been updated
638 		 * already or if a pipesize was specified & the threshhold
639 		 * got below half the pipesize.  I.e., wait for bad news
640 		 * before we start updating, then update on both good
641 		 * and bad news.
642 		 *
643 		 * But we want to save the ssthresh even if no pipesize is
644 		 * specified explicitly in the route, because such
645 		 * connections still have an implicit pipesize specified
646 		 * by the global tcp_sendspace.  In the absence of a reliable
647 		 * way to calculate the pipesize, it will have to do.
648 		 */
649 		i = tp->snd_ssthresh;
650 		if (rt->rt_rmx.rmx_sendpipe != 0)
651 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
652 		else
653 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
654 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
655 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
656 		    || dosavessthresh) {
657 			/*
658 			 * convert the limit from user data bytes to
659 			 * packets then to packet data bytes.
660 			 */
661 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
662 			if (i < 2)
663 				i = 2;
664 			i *= (u_long)(tp->t_maxseg +
665 #ifdef INET6
666 				      (isipv6 ? sizeof (struct ip6_hdr) +
667 					       sizeof (struct tcphdr) :
668 #endif
669 				       sizeof (struct tcpiphdr)
670 #ifdef INET6
671 				       )
672 #endif
673 				      );
674 			if (rt->rt_rmx.rmx_ssthresh)
675 				rt->rt_rmx.rmx_ssthresh =
676 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
677 			else
678 				rt->rt_rmx.rmx_ssthresh = i;
679 			tcpstat.tcps_cachedssthresh++;
680 		}
681 	}
682 	rt = inp->inp_route.ro_rt;
683 	if (rt) {
684 		/*
685 		 * mark route for deletion if no information is
686 		 * cached.
687 		 */
688 		if ((tp->t_flags & TF_LQ_OVERFLOW) &&
689 		    ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){
690 			if (rt->rt_rmx.rmx_rtt == 0)
691 				rt->rt_flags |= RTF_DELCLONE;
692 		}
693 	}
694     no_valid_rt:
695 	/* free the reassembly queue, if any */
696 	while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
697 		LIST_REMOVE(q, tqe_q);
698 		m_freem(q->tqe_m);
699 		FREE(q, M_TSEGQ);
700 	}
701 	if (tp->t_template)
702 		(void) m_free(dtom(tp->t_template));
703 	inp->inp_ppcb = NULL;
704 	soisdisconnected(so);
705 #ifdef INET6
706 	if (INP_CHECK_SOCKAF(so, AF_INET6))
707 		in6_pcbdetach(inp);
708 	else
709 #endif /* INET6 */
710 	in_pcbdetach(inp);
711 	tcpstat.tcps_closed++;
712 	return ((struct tcpcb *)0);
713 }
714 
715 void
716 tcp_drain()
717 {
718 	if (do_tcpdrain)
719 	{
720 		struct inpcb *inpb;
721 		struct tcpcb *tcpb;
722 		struct tseg_qent *te;
723 
724 	/*
725 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
726 	 * if there is one...
727 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
728 	 *      reassembly queue should be flushed, but in a situation
729 	 * 	where we're really low on mbufs, this is potentially
730 	 *  	usefull.
731 	 */
732 		for (inpb = tcbinfo.listhead->lh_first; inpb;
733 	    		inpb = inpb->inp_list.le_next) {
734 				if ((tcpb = intotcpcb(inpb))) {
735 					while ((te = LIST_FIRST(&tcpb->t_segq))
736 					       != NULL) {
737 					LIST_REMOVE(te, tqe_q);
738 					m_freem(te->tqe_m);
739 					FREE(te, M_TSEGQ);
740 				}
741 			}
742 		}
743 
744 	}
745 }
746 
747 /*
748  * Notify a tcp user of an asynchronous error;
749  * store error as soft error, but wake up user
750  * (for now, won't do anything until can select for soft error).
751  */
752 static void
753 tcp_notify(inp, error)
754 	struct inpcb *inp;
755 	int error;
756 {
757 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
758 	register struct socket *so = inp->inp_socket;
759 
760 	/*
761 	 * Ignore some errors if we are hooked up.
762 	 * If connection hasn't completed, has retransmitted several times,
763 	 * and receives a second error, give up now.  This is better
764 	 * than waiting a long time to establish a connection that
765 	 * can never complete.
766 	 */
767 	if (tp->t_state == TCPS_ESTABLISHED &&
768 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
769 	      error == EHOSTDOWN)) {
770 		return;
771 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
772 	    tp->t_softerror)
773 		so->so_error = error;
774 	else
775 		tp->t_softerror = error;
776 	wakeup((caddr_t) &so->so_timeo);
777 	sorwakeup(so);
778 	sowwakeup(so);
779 }
780 
781 static int
782 tcp_pcblist(SYSCTL_HANDLER_ARGS)
783 {
784 	int error, i, n, s;
785 	struct inpcb *inp, **inp_list;
786 	inp_gen_t gencnt;
787 	struct xinpgen xig;
788 
789 	/*
790 	 * The process of preparing the TCB list is too time-consuming and
791 	 * resource-intensive to repeat twice on every request.
792 	 */
793 	if (req->oldptr == 0) {
794 		n = tcbinfo.ipi_count;
795 		req->oldidx = 2 * (sizeof xig)
796 			+ (n + n/8) * sizeof(struct xtcpcb);
797 		return 0;
798 	}
799 
800 	if (req->newptr != 0)
801 		return EPERM;
802 
803 	/*
804 	 * OK, now we're committed to doing something.
805 	 */
806 	s = splnet();
807 	gencnt = tcbinfo.ipi_gencnt;
808 	n = tcbinfo.ipi_count;
809 	splx(s);
810 
811 	xig.xig_len = sizeof xig;
812 	xig.xig_count = n;
813 	xig.xig_gen = gencnt;
814 	xig.xig_sogen = so_gencnt;
815 	error = SYSCTL_OUT(req, &xig, sizeof xig);
816 	if (error)
817 		return error;
818 
819 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
820 	if (inp_list == 0)
821 		return ENOMEM;
822 
823 	s = splnet();
824 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
825 	     inp = inp->inp_list.le_next) {
826 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
827 			inp_list[i++] = inp;
828 	}
829 	splx(s);
830 	n = i;
831 
832 	error = 0;
833 	for (i = 0; i < n; i++) {
834 		inp = inp_list[i];
835 		if (inp->inp_gencnt <= gencnt) {
836 			struct xtcpcb xt;
837 			caddr_t inp_ppcb;
838 			xt.xt_len = sizeof xt;
839 			/* XXX should avoid extra copy */
840 			bcopy(inp, &xt.xt_inp, sizeof *inp);
841 			inp_ppcb = inp->inp_ppcb;
842 			if (inp_ppcb != NULL)
843 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
844 			else
845 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
846 			if (inp->inp_socket)
847 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
848 			error = SYSCTL_OUT(req, &xt, sizeof xt);
849 		}
850 	}
851 	if (!error) {
852 		/*
853 		 * Give the user an updated idea of our state.
854 		 * If the generation differs from what we told
855 		 * her before, she knows that something happened
856 		 * while we were processing this request, and it
857 		 * might be necessary to retry.
858 		 */
859 		s = splnet();
860 		xig.xig_gen = tcbinfo.ipi_gencnt;
861 		xig.xig_sogen = so_gencnt;
862 		xig.xig_count = tcbinfo.ipi_count;
863 		splx(s);
864 		error = SYSCTL_OUT(req, &xig, sizeof xig);
865 	}
866 	free(inp_list, M_TEMP);
867 	return error;
868 }
869 
870 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
871 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
872 
873 static int
874 tcp_getcred(SYSCTL_HANDLER_ARGS)
875 {
876 	struct sockaddr_in addrs[2];
877 	struct inpcb *inp;
878 	int error, s;
879 
880 	error = suser(req->p);
881 	if (error)
882 		return (error);
883 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
884 	if (error)
885 		return (error);
886 	s = splnet();
887 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
888 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
889 	if (inp == NULL || inp->inp_socket == NULL) {
890 		error = ENOENT;
891 		goto out;
892 	}
893 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
894 out:
895 	splx(s);
896 	return (error);
897 }
898 
899 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
900     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
901 
902 #ifdef INET6
903 static int
904 tcp6_getcred(SYSCTL_HANDLER_ARGS)
905 {
906 	struct sockaddr_in6 addrs[2];
907 	struct inpcb *inp;
908 	int error, s, mapped = 0;
909 
910 	error = suser(req->p);
911 	if (error)
912 		return (error);
913 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
914 	if (error)
915 		return (error);
916 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
917 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
918 			mapped = 1;
919 		else
920 			return (EINVAL);
921 	}
922 	s = splnet();
923 	if (mapped == 1)
924 		inp = in_pcblookup_hash(&tcbinfo,
925 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
926 			addrs[1].sin6_port,
927 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
928 			addrs[0].sin6_port,
929 			0, NULL);
930 	else
931 		inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
932 				 addrs[1].sin6_port,
933 				 &addrs[0].sin6_addr, addrs[0].sin6_port,
934 				 0, NULL);
935 	if (inp == NULL || inp->inp_socket == NULL) {
936 		error = ENOENT;
937 		goto out;
938 	}
939 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred,
940 			   sizeof(struct ucred));
941 out:
942 	splx(s);
943 	return (error);
944 }
945 
946 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
947 	    0, 0,
948 	    tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
949 #endif
950 
951 
952 void
953 tcp_ctlinput(cmd, sa, vip)
954 	int cmd;
955 	struct sockaddr *sa;
956 	void *vip;
957 {
958 	register struct ip *ip = vip;
959 	register struct tcphdr *th;
960 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
961 
962 	if (cmd == PRC_QUENCH)
963 		notify = tcp_quench;
964 	else if (cmd == PRC_MSGSIZE)
965 		notify = tcp_mtudisc;
966 	else if (!PRC_IS_REDIRECT(cmd) &&
967 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
968 		return;
969 	if (ip) {
970 		th = (struct tcphdr *)((caddr_t)ip
971 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
972 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
973 			cmd, notify);
974 	} else
975 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
976 }
977 
978 #ifdef INET6
979 void
980 tcp6_ctlinput(cmd, sa, d)
981 	int cmd;
982 	struct sockaddr *sa;
983 	void *d;
984 {
985 	register struct tcphdr *thp;
986 	struct tcphdr th;
987 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
988 	struct sockaddr_in6 sa6;
989 	struct ip6_hdr *ip6;
990 	struct mbuf *m;
991 	int off;
992 
993 	if (sa->sa_family != AF_INET6 ||
994 	    sa->sa_len != sizeof(struct sockaddr_in6))
995 		return;
996 
997 	if (cmd == PRC_QUENCH)
998 		notify = tcp_quench;
999 	else if (cmd == PRC_MSGSIZE)
1000 		notify = tcp_mtudisc;
1001 	else if (!PRC_IS_REDIRECT(cmd) &&
1002 		 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1003 		return;
1004 
1005 	/* if the parameter is from icmp6, decode it. */
1006 	if (d != NULL) {
1007 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1008 		m = ip6cp->ip6c_m;
1009 		ip6 = ip6cp->ip6c_ip6;
1010 		off = ip6cp->ip6c_off;
1011 	} else {
1012 		m = NULL;
1013 		ip6 = NULL;
1014 		off = 0;	/* fool gcc */
1015 	}
1016 
1017 	/*
1018 	 * Translate addresses into internal form.
1019 	 * Sa check if it is AF_INET6 is done at the top of this funciton.
1020 	 */
1021 	sa6 = *(struct sockaddr_in6 *)sa;
1022 	if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) != 0 && m != NULL &&
1023 	    m->m_pkthdr.rcvif != NULL)
1024 		sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1025 
1026 	if (ip6) {
1027 		/*
1028 		 * XXX: We assume that when IPV6 is non NULL,
1029 		 * M and OFF are valid.
1030 		 */
1031 		struct in6_addr s;
1032 
1033 		/* translate addresses into internal form */
1034 		memcpy(&s, &ip6->ip6_src, sizeof(s));
1035 		if (IN6_IS_ADDR_LINKLOCAL(&s) != 0 && m != NULL &&
1036 		    m->m_pkthdr.rcvif != NULL)
1037 			s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1038 
1039 		/* check if we can safely examine src and dst ports */
1040 		if (m->m_pkthdr.len < off + sizeof(th))
1041 			return;
1042 
1043 		if (m->m_len < off + sizeof(th)) {
1044 			/*
1045 			 * this should be rare case
1046 			 * because now MINCLSIZE is "(MHLEN + 1)",
1047 			 * so we compromise on this copy...
1048 			 */
1049 			m_copydata(m, off, sizeof(th), (caddr_t)&th);
1050 			thp = &th;
1051 		} else
1052 			thp = (struct tcphdr *)(mtod(m, caddr_t) + off);
1053 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, thp->th_dport,
1054 			      &s, thp->th_sport, cmd, notify);
1055 	} else
1056 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, 0, &zeroin6_addr,
1057 			      0, cmd, notify);
1058 }
1059 #endif /* INET6 */
1060 
1061 /*
1062  * When a source quench is received, close congestion window
1063  * to one segment.  We will gradually open it again as we proceed.
1064  */
1065 void
1066 tcp_quench(inp, errno)
1067 	struct inpcb *inp;
1068 	int errno;
1069 {
1070 	struct tcpcb *tp = intotcpcb(inp);
1071 
1072 	if (tp)
1073 		tp->snd_cwnd = tp->t_maxseg;
1074 }
1075 
1076 /*
1077  * When `need fragmentation' ICMP is received, update our idea of the MSS
1078  * based on the new value in the route.  Also nudge TCP to send something,
1079  * since we know the packet we just sent was dropped.
1080  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1081  */
1082 void
1083 tcp_mtudisc(inp, errno)
1084 	struct inpcb *inp;
1085 	int errno;
1086 {
1087 	struct tcpcb *tp = intotcpcb(inp);
1088 	struct rtentry *rt;
1089 	struct rmxp_tao *taop;
1090 	struct socket *so = inp->inp_socket;
1091 	int offered;
1092 	int mss;
1093 #ifdef INET6
1094 	int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1095 #endif /* INET6 */
1096 
1097 	if (tp) {
1098 #ifdef INET6
1099 		if (isipv6)
1100 			rt = tcp_rtlookup6(inp);
1101 		else
1102 #endif /* INET6 */
1103 		rt = tcp_rtlookup(inp);
1104 		if (!rt || !rt->rt_rmx.rmx_mtu) {
1105 			tp->t_maxopd = tp->t_maxseg =
1106 #ifdef INET6
1107 				isipv6 ? tcp_v6mssdflt :
1108 #endif /* INET6 */
1109 				tcp_mssdflt;
1110 			return;
1111 		}
1112 		taop = rmx_taop(rt->rt_rmx);
1113 		offered = taop->tao_mssopt;
1114 		mss = rt->rt_rmx.rmx_mtu -
1115 #ifdef INET6
1116 			(isipv6 ?
1117 			 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1118 #endif /* INET6 */
1119 			 sizeof(struct tcpiphdr)
1120 #ifdef INET6
1121 			 )
1122 #endif /* INET6 */
1123 			;
1124 
1125 		if (offered)
1126 			mss = min(mss, offered);
1127 		/*
1128 		 * XXX - The above conditional probably violates the TCP
1129 		 * spec.  The problem is that, since we don't know the
1130 		 * other end's MSS, we are supposed to use a conservative
1131 		 * default.  But, if we do that, then MTU discovery will
1132 		 * never actually take place, because the conservative
1133 		 * default is much less than the MTUs typically seen
1134 		 * on the Internet today.  For the moment, we'll sweep
1135 		 * this under the carpet.
1136 		 *
1137 		 * The conservative default might not actually be a problem
1138 		 * if the only case this occurs is when sending an initial
1139 		 * SYN with options and data to a host we've never talked
1140 		 * to before.  Then, they will reply with an MSS value which
1141 		 * will get recorded and the new parameters should get
1142 		 * recomputed.  For Further Study.
1143 		 */
1144 		if (tp->t_maxopd <= mss)
1145 			return;
1146 		tp->t_maxopd = mss;
1147 
1148 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1149 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1150 			mss -= TCPOLEN_TSTAMP_APPA;
1151 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1152 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1153 			mss -= TCPOLEN_CC_APPA;
1154 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1155 		if (mss > MCLBYTES)
1156 			mss &= ~(MCLBYTES-1);
1157 #else
1158 		if (mss > MCLBYTES)
1159 			mss = mss / MCLBYTES * MCLBYTES;
1160 #endif
1161 		if (so->so_snd.sb_hiwat < mss)
1162 			mss = so->so_snd.sb_hiwat;
1163 
1164 		tp->t_maxseg = mss;
1165 
1166 		tcpstat.tcps_mturesent++;
1167 		tp->t_rtttime = 0;
1168 		tp->snd_nxt = tp->snd_una;
1169 		tcp_output(tp);
1170 	}
1171 }
1172 
1173 /*
1174  * Look-up the routing entry to the peer of this inpcb.  If no route
1175  * is found and it cannot be allocated the return NULL.  This routine
1176  * is called by TCP routines that access the rmx structure and by tcp_mss
1177  * to get the interface MTU.
1178  */
1179 struct rtentry *
1180 tcp_rtlookup(inp)
1181 	struct inpcb *inp;
1182 {
1183 	struct route *ro;
1184 	struct rtentry *rt;
1185 
1186 	ro = &inp->inp_route;
1187 	rt = ro->ro_rt;
1188 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1189 		/* No route yet, so try to acquire one */
1190 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1191 			ro->ro_dst.sa_family = AF_INET;
1192 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1193 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1194 				inp->inp_faddr;
1195 			rtalloc(ro);
1196 			rt = ro->ro_rt;
1197 		}
1198 	}
1199 	return rt;
1200 }
1201 
1202 #ifdef INET6
1203 struct rtentry *
1204 tcp_rtlookup6(inp)
1205 	struct inpcb *inp;
1206 {
1207 	struct route_in6 *ro6;
1208 	struct rtentry *rt;
1209 
1210 	ro6 = &inp->in6p_route;
1211 	rt = ro6->ro_rt;
1212 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1213 		/* No route yet, so try to acquire one */
1214 		if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1215 			ro6->ro_dst.sin6_family = AF_INET6;
1216 			ro6->ro_dst.sin6_len = sizeof(ro6->ro_dst);
1217 			ro6->ro_dst.sin6_addr = inp->in6p_faddr;
1218 			rtalloc((struct route *)ro6);
1219 			rt = ro6->ro_rt;
1220 		}
1221 	}
1222 	return rt;
1223 }
1224 #endif /* INET6 */
1225 
1226 #ifdef IPSEC
1227 /* compute ESP/AH header size for TCP, including outer IP header. */
1228 size_t
1229 ipsec_hdrsiz_tcp(tp)
1230 	struct tcpcb *tp;
1231 {
1232 	struct inpcb *inp;
1233 	struct mbuf *m;
1234 	size_t hdrsiz;
1235 	struct ip *ip;
1236 #ifdef INET6
1237 	struct ip6_hdr *ip6;
1238 #endif /* INET6 */
1239 	struct tcphdr *th;
1240 
1241 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1242 		return 0;
1243 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1244 	if (!m)
1245 		return 0;
1246 
1247 #ifdef INET6
1248 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1249 		ip6 = mtod(m, struct ip6_hdr *);
1250 		th = (struct tcphdr *)(ip6 + 1);
1251 		m->m_pkthdr.len = m->m_len =
1252 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1253 		bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip6,
1254 		      sizeof(struct ip6_hdr));
1255 		bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1256 		      sizeof(struct tcphdr));
1257 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1258 	} else
1259 #endif /* INET6 */
1260       {
1261 	ip = mtod(m, struct ip *);
1262 	th = (struct tcphdr *)(ip + 1);
1263 	m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1264 	bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip,
1265 	      sizeof(struct ip));
1266 	bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1267 	      sizeof(struct tcphdr));
1268 	hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1269       }
1270 
1271 	m_free(m);
1272 	return hdrsiz;
1273 }
1274 #endif /*IPSEC*/
1275 
1276 /*
1277  * Return a pointer to the cached information about the remote host.
1278  * The cached information is stored in the protocol specific part of
1279  * the route metrics.
1280  */
1281 struct rmxp_tao *
1282 tcp_gettaocache(inp)
1283 	struct inpcb *inp;
1284 {
1285 	struct rtentry *rt;
1286 
1287 #ifdef INET6
1288 	if ((inp->inp_vflag & INP_IPV6) != 0)
1289 		rt = tcp_rtlookup6(inp);
1290 	else
1291 #endif /* INET6 */
1292 	rt = tcp_rtlookup(inp);
1293 
1294 	/* Make sure this is a host route and is up. */
1295 	if (rt == NULL ||
1296 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1297 		return NULL;
1298 
1299 	return rmx_taop(rt->rt_rmx);
1300 }
1301 
1302 /*
1303  * Clear all the TAO cache entries, called from tcp_init.
1304  *
1305  * XXX
1306  * This routine is just an empty one, because we assume that the routing
1307  * routing tables are initialized at the same time when TCP, so there is
1308  * nothing in the cache left over.
1309  */
1310 static void
1311 tcp_cleartaocache()
1312 {
1313 }
1314