xref: /freebsd/sys/netinet/tcp_subr.c (revision c68159a6d8eede11766cf13896d0f7670dbd51aa)
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 /*
138  * Treat ICMP administratively prohibited like a TCP RST
139  * as required by rfc1122 section 3.2.2.1
140  */
141 
142 static int	icmp_admin_prohib_like_rst = 1;
143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_admin_prohib_like_rst, CTLFLAG_RW,
144 	&icmp_admin_prohib_like_rst, 0,
145 	"Treat ICMP administratively prohibited messages like TCP RST, rfc1122 section 3.2.2.1");
146 
147 /*
148  * When icmp_admin_prohib_like_rst is enabled, only act on
149  * sessions in SYN-SENT state
150  */
151 
152 static int	icmp_like_rst_syn_sent_only = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_like_rst_syn_sent_only, CTLFLAG_RW,
154 	&icmp_like_rst_syn_sent_only, 0,
155 	"When icmp_admin_prohib_like_rst is enabled, only act on sessions in SYN-SENT state");
156 
157 static void	tcp_cleartaocache __P((void));
158 static void	tcp_notify __P((struct inpcb *, int));
159 
160 /*
161  * Target size of TCP PCB hash tables. Must be a power of two.
162  *
163  * Note that this can be overridden by the kernel environment
164  * variable net.inet.tcp.tcbhashsize
165  */
166 #ifndef TCBHASHSIZE
167 #define TCBHASHSIZE	512
168 #endif
169 
170 /*
171  * This is the actual shape of what we allocate using the zone
172  * allocator.  Doing it this way allows us to protect both structures
173  * using the same generation count, and also eliminates the overhead
174  * of allocating tcpcbs separately.  By hiding the structure here,
175  * we avoid changing most of the rest of the code (although it needs
176  * to be changed, eventually, for greater efficiency).
177  */
178 #define	ALIGNMENT	32
179 #define	ALIGNM1		(ALIGNMENT - 1)
180 struct	inp_tp {
181 	union {
182 		struct	inpcb inp;
183 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
184 	} inp_tp_u;
185 	struct	tcpcb tcb;
186 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
187 	struct	callout inp_tp_delack;
188 };
189 #undef ALIGNMENT
190 #undef ALIGNM1
191 
192 /*
193  * Tcp initialization
194  */
195 void
196 tcp_init()
197 {
198 	int hashsize;
199 
200 	tcp_iss = arc4random();	/* wrong, but better than a constant */
201 	tcp_ccgen = 1;
202 	tcp_cleartaocache();
203 
204 	tcp_delacktime = TCPTV_DELACK;
205 	tcp_keepinit = TCPTV_KEEP_INIT;
206 	tcp_keepidle = TCPTV_KEEP_IDLE;
207 	tcp_keepintvl = TCPTV_KEEPINTVL;
208 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
209 	tcp_msl = TCPTV_MSL;
210 
211 	LIST_INIT(&tcb);
212 	tcbinfo.listhead = &tcb;
213 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", TCBHASHSIZE, hashsize);
214 	if (!powerof2(hashsize)) {
215 		printf("WARNING: TCB hash size not a power of 2\n");
216 		hashsize = 512; /* safe default */
217 	}
218 	tcp_tcbhashsize = hashsize;
219 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
220 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
221 					&tcbinfo.porthashmask);
222 	tcbinfo.ipi_zone = zinit("tcpcb", sizeof(struct inp_tp), maxsockets,
223 				 ZONE_INTERRUPT, 0);
224 #ifdef INET6
225 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
226 #else /* INET6 */
227 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
228 #endif /* INET6 */
229 	if (max_protohdr < TCP_MINPROTOHDR)
230 		max_protohdr = TCP_MINPROTOHDR;
231 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
232 		panic("tcp_init");
233 #undef TCP_MINPROTOHDR
234 }
235 
236 /*
237  * Create template to be used to send tcp packets on a connection.
238  * Call after host entry created, allocates an mbuf and fills
239  * in a skeletal tcp/ip header, minimizing the amount of work
240  * necessary when the connection is used.
241  */
242 struct tcptemp *
243 tcp_template(tp)
244 	struct tcpcb *tp;
245 {
246 	register struct inpcb *inp = tp->t_inpcb;
247 	register struct mbuf *m;
248 	register struct tcptemp *n;
249 
250 	if ((n = tp->t_template) == 0) {
251 		m = m_get(M_DONTWAIT, MT_HEADER);
252 		if (m == NULL)
253 			return (0);
254 		m->m_len = sizeof (struct tcptemp);
255 		n = mtod(m, struct tcptemp *);
256 	}
257 #ifdef INET6
258 	if ((inp->inp_vflag & INP_IPV6) != 0) {
259 		register struct ip6_hdr *ip6;
260 
261 		ip6 = (struct ip6_hdr *)n->tt_ipgen;
262 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
263 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
264 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
265 			(IPV6_VERSION & IPV6_VERSION_MASK);
266 		ip6->ip6_nxt = IPPROTO_TCP;
267 		ip6->ip6_plen = sizeof(struct tcphdr);
268 		ip6->ip6_src = inp->in6p_laddr;
269 		ip6->ip6_dst = inp->in6p_faddr;
270 		n->tt_t.th_sum = 0;
271 	} else
272 #endif
273       {
274 	struct ip *ip = (struct ip *)n->tt_ipgen;
275 
276 	bzero(ip, sizeof(struct ip));		/* XXX overkill? */
277 	ip->ip_vhl = IP_VHL_BORING;
278 	ip->ip_p = IPPROTO_TCP;
279 	ip->ip_src = inp->inp_laddr;
280 	ip->ip_dst = inp->inp_faddr;
281 	n->tt_t.th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
282 	    htons(sizeof(struct tcphdr) + IPPROTO_TCP));
283       }
284 	n->tt_t.th_sport = inp->inp_lport;
285 	n->tt_t.th_dport = inp->inp_fport;
286 	n->tt_t.th_seq = 0;
287 	n->tt_t.th_ack = 0;
288 	n->tt_t.th_x2 = 0;
289 	n->tt_t.th_off = 5;
290 	n->tt_t.th_flags = 0;
291 	n->tt_t.th_win = 0;
292 	n->tt_t.th_urp = 0;
293 	return (n);
294 }
295 
296 /*
297  * Send a single message to the TCP at address specified by
298  * the given TCP/IP header.  If m == 0, then we make a copy
299  * of the tcpiphdr at ti and send directly to the addressed host.
300  * This is used to force keep alive messages out using the TCP
301  * template for a connection tp->t_template.  If flags are given
302  * then we send a message back to the TCP which originated the
303  * segment ti, and discard the mbuf containing it and any other
304  * attached mbufs.
305  *
306  * In any case the ack and sequence number of the transmitted
307  * segment are as specified by the parameters.
308  *
309  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
310  */
311 void
312 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
313 	struct tcpcb *tp;
314 	void *ipgen;
315 	register struct tcphdr *th;
316 	register struct mbuf *m;
317 	tcp_seq ack, seq;
318 	int flags;
319 {
320 	register int tlen;
321 	int win = 0;
322 	struct route *ro = 0;
323 	struct route sro;
324 	struct ip *ip;
325 	struct tcphdr *nth;
326 #ifdef INET6
327 	struct route_in6 *ro6 = 0;
328 	struct route_in6 sro6;
329 	struct ip6_hdr *ip6;
330 	int isipv6;
331 #endif /* INET6 */
332 	int ipflags = 0;
333 
334 #ifdef INET6
335 	isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
336 	ip6 = ipgen;
337 #endif /* INET6 */
338 	ip = ipgen;
339 
340 	if (tp) {
341 		if (!(flags & TH_RST)) {
342 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
343 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
344 				win = (long)TCP_MAXWIN << tp->rcv_scale;
345 		}
346 #ifdef INET6
347 		if (isipv6)
348 			ro6 = &tp->t_inpcb->in6p_route;
349 		else
350 #endif /* INET6 */
351 		ro = &tp->t_inpcb->inp_route;
352 	} else {
353 #ifdef INET6
354 		if (isipv6) {
355 			ro6 = &sro6;
356 			bzero(ro6, sizeof *ro6);
357 		} else
358 #endif /* INET6 */
359 	      {
360 		ro = &sro;
361 		bzero(ro, sizeof *ro);
362 	      }
363 	}
364 	if (m == 0) {
365 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
366 		if (m == NULL)
367 			return;
368 #ifdef TCP_COMPAT_42
369 		tlen = 1;
370 #else
371 		tlen = 0;
372 #endif
373 		m->m_data += max_linkhdr;
374 #ifdef INET6
375 		if (isipv6) {
376 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
377 			      sizeof(struct ip6_hdr));
378 			ip6 = mtod(m, struct ip6_hdr *);
379 			nth = (struct tcphdr *)(ip6 + 1);
380 		} else
381 #endif /* INET6 */
382 	      {
383 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
384 		ip = mtod(m, struct ip *);
385 		nth = (struct tcphdr *)(ip + 1);
386 	      }
387 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
388 		flags = TH_ACK;
389 	} else {
390 		m_freem(m->m_next);
391 		m->m_next = 0;
392 		m->m_data = (caddr_t)ipgen;
393 		/* m_len is set later */
394 		tlen = 0;
395 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
396 #ifdef INET6
397 		if (isipv6) {
398 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
399 			nth = (struct tcphdr *)(ip6 + 1);
400 		} else
401 #endif /* INET6 */
402 	      {
403 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
404 		nth = (struct tcphdr *)(ip + 1);
405 	      }
406 		if (th != nth) {
407 			/*
408 			 * this is usually a case when an extension header
409 			 * exists between the IPv6 header and the
410 			 * TCP header.
411 			 */
412 			nth->th_sport = th->th_sport;
413 			nth->th_dport = th->th_dport;
414 		}
415 		xchg(nth->th_dport, nth->th_sport, n_short);
416 #undef xchg
417 	}
418 #ifdef INET6
419 	if (isipv6) {
420 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
421 						tlen));
422 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
423 	} else
424 #endif
425       {
426 	tlen += sizeof (struct tcpiphdr);
427 	ip->ip_len = tlen;
428 	ip->ip_ttl = ip_defttl;
429       }
430 	m->m_len = tlen;
431 	m->m_pkthdr.len = tlen;
432 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
433 	nth->th_seq = htonl(seq);
434 	nth->th_ack = htonl(ack);
435 	nth->th_x2 = 0;
436 	nth->th_off = sizeof (struct tcphdr) >> 2;
437 	nth->th_flags = flags;
438 	if (tp)
439 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
440 	else
441 		nth->th_win = htons((u_short)win);
442 	nth->th_urp = 0;
443 #ifdef INET6
444 	if (isipv6) {
445 		nth->th_sum = 0;
446 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
447 					sizeof(struct ip6_hdr),
448 					tlen - sizeof(struct ip6_hdr));
449 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
450 					       ro6 && ro6->ro_rt ?
451 					       ro6->ro_rt->rt_ifp :
452 					       NULL);
453 	} else
454 #endif /* INET6 */
455       {
456         nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
457 	    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
458         m->m_pkthdr.csum_flags = CSUM_TCP;
459         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
460       }
461 #ifdef TCPDEBUG
462 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
463 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
464 #endif
465 #ifdef IPSEC
466 	ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL);
467 #endif
468 #ifdef INET6
469 	if (isipv6) {
470 		(void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
471 		if (ro6 == &sro6 && ro6->ro_rt) {
472 			RTFREE(ro6->ro_rt);
473 			ro6->ro_rt = NULL;
474 		}
475 	} else
476 #endif /* INET6 */
477       {
478 	(void) ip_output(m, NULL, ro, ipflags, NULL);
479 	if (ro == &sro && ro->ro_rt) {
480 		RTFREE(ro->ro_rt);
481 		ro->ro_rt = NULL;
482 	}
483       }
484 }
485 
486 /*
487  * Create a new TCP control block, making an
488  * empty reassembly queue and hooking it to the argument
489  * protocol control block.  The `inp' parameter must have
490  * come from the zone allocator set up in tcp_init().
491  */
492 struct tcpcb *
493 tcp_newtcpcb(inp)
494 	struct inpcb *inp;
495 {
496 	struct inp_tp *it;
497 	register struct tcpcb *tp;
498 #ifdef INET6
499 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
500 #endif /* INET6 */
501 
502 	it = (struct inp_tp *)inp;
503 	tp = &it->tcb;
504 	bzero((char *) tp, sizeof(struct tcpcb));
505 	LIST_INIT(&tp->t_segq);
506 	tp->t_maxseg = tp->t_maxopd =
507 #ifdef INET6
508 		isipv6 ? tcp_v6mssdflt :
509 #endif /* INET6 */
510 		tcp_mssdflt;
511 
512 	/* Set up our timeouts. */
513 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0);
514 	callout_init(tp->tt_persist = &it->inp_tp_persist, 0);
515 	callout_init(tp->tt_keep = &it->inp_tp_keep, 0);
516 	callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0);
517 	callout_init(tp->tt_delack = &it->inp_tp_delack, 0);
518 
519 	if (tcp_do_rfc1323)
520 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
521 	if (tcp_do_rfc1644)
522 		tp->t_flags |= TF_REQ_CC;
523 	tp->t_inpcb = inp;	/* XXX */
524 	/*
525 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
526 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
527 	 * reasonable initial retransmit time.
528 	 */
529 	tp->t_srtt = TCPTV_SRTTBASE;
530 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
531 	tp->t_rttmin = TCPTV_MIN;
532 	tp->t_rxtcur = TCPTV_RTOBASE;
533 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
534 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
535 	tp->t_rcvtime = ticks;
536         /*
537 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
538 	 * because the socket may be bound to an IPv6 wildcard address,
539 	 * which may match an IPv4-mapped IPv6 address.
540 	 */
541 	inp->inp_ip_ttl = ip_defttl;
542 	inp->inp_ppcb = (caddr_t)tp;
543 	return (tp);		/* XXX */
544 }
545 
546 /*
547  * Drop a TCP connection, reporting
548  * the specified error.  If connection is synchronized,
549  * then send a RST to peer.
550  */
551 struct tcpcb *
552 tcp_drop(tp, errno)
553 	register struct tcpcb *tp;
554 	int errno;
555 {
556 	struct socket *so = tp->t_inpcb->inp_socket;
557 
558 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
559 		tp->t_state = TCPS_CLOSED;
560 		(void) tcp_output(tp);
561 		tcpstat.tcps_drops++;
562 	} else
563 		tcpstat.tcps_conndrops++;
564 	if (errno == ETIMEDOUT && tp->t_softerror)
565 		errno = tp->t_softerror;
566 	so->so_error = errno;
567 	return (tcp_close(tp));
568 }
569 
570 /*
571  * Close a TCP control block:
572  *	discard all space held by the tcp
573  *	discard internet protocol block
574  *	wake up any sleepers
575  */
576 struct tcpcb *
577 tcp_close(tp)
578 	register struct tcpcb *tp;
579 {
580 	register struct tseg_qent *q;
581 	struct inpcb *inp = tp->t_inpcb;
582 	struct socket *so = inp->inp_socket;
583 #ifdef INET6
584 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
585 #endif /* INET6 */
586 	register struct rtentry *rt;
587 	int dosavessthresh;
588 
589 	/*
590 	 * Make sure that all of our timers are stopped before we
591 	 * delete the PCB.
592 	 */
593 	callout_stop(tp->tt_rexmt);
594 	callout_stop(tp->tt_persist);
595 	callout_stop(tp->tt_keep);
596 	callout_stop(tp->tt_2msl);
597 	callout_stop(tp->tt_delack);
598 
599 	/*
600 	 * If we got enough samples through the srtt filter,
601 	 * save the rtt and rttvar in the routing entry.
602 	 * 'Enough' is arbitrarily defined as the 16 samples.
603 	 * 16 samples is enough for the srtt filter to converge
604 	 * to within 5% of the correct value; fewer samples and
605 	 * we could save a very bogus rtt.
606 	 *
607 	 * Don't update the default route's characteristics and don't
608 	 * update anything that the user "locked".
609 	 */
610 	if (tp->t_rttupdated >= 16) {
611 		register u_long i = 0;
612 #ifdef INET6
613 		if (isipv6) {
614 			struct sockaddr_in6 *sin6;
615 
616 			if ((rt = inp->in6p_route.ro_rt) == NULL)
617 				goto no_valid_rt;
618 			sin6 = (struct sockaddr_in6 *)rt_key(rt);
619 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
620 				goto no_valid_rt;
621 		}
622 		else
623 #endif /* INET6 */
624 		if ((rt = inp->inp_route.ro_rt) == NULL ||
625 		    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
626 		    == INADDR_ANY)
627 			goto no_valid_rt;
628 
629 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
630 			i = tp->t_srtt *
631 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
632 			if (rt->rt_rmx.rmx_rtt && i)
633 				/*
634 				 * filter this update to half the old & half
635 				 * the new values, converting scale.
636 				 * See route.h and tcp_var.h for a
637 				 * description of the scaling constants.
638 				 */
639 				rt->rt_rmx.rmx_rtt =
640 				    (rt->rt_rmx.rmx_rtt + i) / 2;
641 			else
642 				rt->rt_rmx.rmx_rtt = i;
643 			tcpstat.tcps_cachedrtt++;
644 		}
645 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
646 			i = tp->t_rttvar *
647 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
648 			if (rt->rt_rmx.rmx_rttvar && i)
649 				rt->rt_rmx.rmx_rttvar =
650 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
651 			else
652 				rt->rt_rmx.rmx_rttvar = i;
653 			tcpstat.tcps_cachedrttvar++;
654 		}
655 		/*
656 		 * The old comment here said:
657 		 * update the pipelimit (ssthresh) if it has been updated
658 		 * already or if a pipesize was specified & the threshhold
659 		 * got below half the pipesize.  I.e., wait for bad news
660 		 * before we start updating, then update on both good
661 		 * and bad news.
662 		 *
663 		 * But we want to save the ssthresh even if no pipesize is
664 		 * specified explicitly in the route, because such
665 		 * connections still have an implicit pipesize specified
666 		 * by the global tcp_sendspace.  In the absence of a reliable
667 		 * way to calculate the pipesize, it will have to do.
668 		 */
669 		i = tp->snd_ssthresh;
670 		if (rt->rt_rmx.rmx_sendpipe != 0)
671 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
672 		else
673 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
674 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
675 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
676 		    || dosavessthresh) {
677 			/*
678 			 * convert the limit from user data bytes to
679 			 * packets then to packet data bytes.
680 			 */
681 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
682 			if (i < 2)
683 				i = 2;
684 			i *= (u_long)(tp->t_maxseg +
685 #ifdef INET6
686 				      (isipv6 ? sizeof (struct ip6_hdr) +
687 					       sizeof (struct tcphdr) :
688 #endif
689 				       sizeof (struct tcpiphdr)
690 #ifdef INET6
691 				       )
692 #endif
693 				      );
694 			if (rt->rt_rmx.rmx_ssthresh)
695 				rt->rt_rmx.rmx_ssthresh =
696 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
697 			else
698 				rt->rt_rmx.rmx_ssthresh = i;
699 			tcpstat.tcps_cachedssthresh++;
700 		}
701 	}
702 	rt = inp->inp_route.ro_rt;
703 	if (rt) {
704 		/*
705 		 * mark route for deletion if no information is
706 		 * cached.
707 		 */
708 		if ((tp->t_flags & TF_LQ_OVERFLOW) &&
709 		    ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0)){
710 			if (rt->rt_rmx.rmx_rtt == 0)
711 				rt->rt_flags |= RTF_DELCLONE;
712 		}
713 	}
714     no_valid_rt:
715 	/* free the reassembly queue, if any */
716 	while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
717 		LIST_REMOVE(q, tqe_q);
718 		m_freem(q->tqe_m);
719 		FREE(q, M_TSEGQ);
720 	}
721 	if (tp->t_template)
722 		(void) m_free(dtom(tp->t_template));
723 	inp->inp_ppcb = NULL;
724 	soisdisconnected(so);
725 #ifdef INET6
726 	if (INP_CHECK_SOCKAF(so, AF_INET6))
727 		in6_pcbdetach(inp);
728 	else
729 #endif /* INET6 */
730 	in_pcbdetach(inp);
731 	tcpstat.tcps_closed++;
732 	return ((struct tcpcb *)0);
733 }
734 
735 void
736 tcp_drain()
737 {
738 	if (do_tcpdrain)
739 	{
740 		struct inpcb *inpb;
741 		struct tcpcb *tcpb;
742 		struct tseg_qent *te;
743 
744 	/*
745 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
746 	 * if there is one...
747 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
748 	 *      reassembly queue should be flushed, but in a situation
749 	 * 	where we're really low on mbufs, this is potentially
750 	 *  	usefull.
751 	 */
752 		for (inpb = tcbinfo.listhead->lh_first; inpb;
753 	    		inpb = inpb->inp_list.le_next) {
754 				if ((tcpb = intotcpcb(inpb))) {
755 					while ((te = LIST_FIRST(&tcpb->t_segq))
756 					       != NULL) {
757 					LIST_REMOVE(te, tqe_q);
758 					m_freem(te->tqe_m);
759 					FREE(te, M_TSEGQ);
760 				}
761 			}
762 		}
763 
764 	}
765 }
766 
767 /*
768  * Notify a tcp user of an asynchronous error;
769  * store error as soft error, but wake up user
770  * (for now, won't do anything until can select for soft error).
771  */
772 static void
773 tcp_notify(inp, error)
774 	struct inpcb *inp;
775 	int error;
776 {
777 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
778 	register struct socket *so = inp->inp_socket;
779 
780 	/*
781 	 * Ignore some errors if we are hooked up.
782 	 * If connection hasn't completed, has retransmitted several times,
783 	 * and receives a second error, give up now.  This is better
784 	 * than waiting a long time to establish a connection that
785 	 * can never complete.
786 	 */
787 	if (tp->t_state == TCPS_ESTABLISHED &&
788 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
789 	      error == EHOSTDOWN)) {
790 		return;
791 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
792 	    tp->t_softerror)
793 		so->so_error = error;
794 	else
795 		tp->t_softerror = error;
796 	wakeup((caddr_t) &so->so_timeo);
797 	sorwakeup(so);
798 	sowwakeup(so);
799 }
800 
801 static int
802 tcp_pcblist(SYSCTL_HANDLER_ARGS)
803 {
804 	int error, i, n, s;
805 	struct inpcb *inp, **inp_list;
806 	inp_gen_t gencnt;
807 	struct xinpgen xig;
808 
809 	/*
810 	 * The process of preparing the TCB list is too time-consuming and
811 	 * resource-intensive to repeat twice on every request.
812 	 */
813 	if (req->oldptr == 0) {
814 		n = tcbinfo.ipi_count;
815 		req->oldidx = 2 * (sizeof xig)
816 			+ (n + n/8) * sizeof(struct xtcpcb);
817 		return 0;
818 	}
819 
820 	if (req->newptr != 0)
821 		return EPERM;
822 
823 	/*
824 	 * OK, now we're committed to doing something.
825 	 */
826 	s = splnet();
827 	gencnt = tcbinfo.ipi_gencnt;
828 	n = tcbinfo.ipi_count;
829 	splx(s);
830 
831 	xig.xig_len = sizeof xig;
832 	xig.xig_count = n;
833 	xig.xig_gen = gencnt;
834 	xig.xig_sogen = so_gencnt;
835 	error = SYSCTL_OUT(req, &xig, sizeof xig);
836 	if (error)
837 		return error;
838 
839 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
840 	if (inp_list == 0)
841 		return ENOMEM;
842 
843 	s = splnet();
844 	for (inp = tcbinfo.listhead->lh_first, i = 0; inp && i < n;
845 	     inp = inp->inp_list.le_next) {
846 		if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
847 			inp_list[i++] = inp;
848 	}
849 	splx(s);
850 	n = i;
851 
852 	error = 0;
853 	for (i = 0; i < n; i++) {
854 		inp = inp_list[i];
855 		if (inp->inp_gencnt <= gencnt) {
856 			struct xtcpcb xt;
857 			caddr_t inp_ppcb;
858 			xt.xt_len = sizeof xt;
859 			/* XXX should avoid extra copy */
860 			bcopy(inp, &xt.xt_inp, sizeof *inp);
861 			inp_ppcb = inp->inp_ppcb;
862 			if (inp_ppcb != NULL)
863 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
864 			else
865 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
866 			if (inp->inp_socket)
867 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
868 			error = SYSCTL_OUT(req, &xt, sizeof xt);
869 		}
870 	}
871 	if (!error) {
872 		/*
873 		 * Give the user an updated idea of our state.
874 		 * If the generation differs from what we told
875 		 * her before, she knows that something happened
876 		 * while we were processing this request, and it
877 		 * might be necessary to retry.
878 		 */
879 		s = splnet();
880 		xig.xig_gen = tcbinfo.ipi_gencnt;
881 		xig.xig_sogen = so_gencnt;
882 		xig.xig_count = tcbinfo.ipi_count;
883 		splx(s);
884 		error = SYSCTL_OUT(req, &xig, sizeof xig);
885 	}
886 	free(inp_list, M_TEMP);
887 	return error;
888 }
889 
890 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
891 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
892 
893 static int
894 tcp_getcred(SYSCTL_HANDLER_ARGS)
895 {
896 	struct sockaddr_in addrs[2];
897 	struct inpcb *inp;
898 	int error, s;
899 
900 	error = suser(req->p);
901 	if (error)
902 		return (error);
903 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
904 	if (error)
905 		return (error);
906 	s = splnet();
907 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
908 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
909 	if (inp == NULL || inp->inp_socket == NULL) {
910 		error = ENOENT;
911 		goto out;
912 	}
913 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
914 out:
915 	splx(s);
916 	return (error);
917 }
918 
919 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
920     0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
921 
922 #ifdef INET6
923 static int
924 tcp6_getcred(SYSCTL_HANDLER_ARGS)
925 {
926 	struct sockaddr_in6 addrs[2];
927 	struct inpcb *inp;
928 	int error, s, mapped = 0;
929 
930 	error = suser(req->p);
931 	if (error)
932 		return (error);
933 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
934 	if (error)
935 		return (error);
936 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
937 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
938 			mapped = 1;
939 		else
940 			return (EINVAL);
941 	}
942 	s = splnet();
943 	if (mapped == 1)
944 		inp = in_pcblookup_hash(&tcbinfo,
945 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
946 			addrs[1].sin6_port,
947 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
948 			addrs[0].sin6_port,
949 			0, NULL);
950 	else
951 		inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
952 				 addrs[1].sin6_port,
953 				 &addrs[0].sin6_addr, addrs[0].sin6_port,
954 				 0, NULL);
955 	if (inp == NULL || inp->inp_socket == NULL) {
956 		error = ENOENT;
957 		goto out;
958 	}
959 	error = SYSCTL_OUT(req, inp->inp_socket->so_cred,
960 			   sizeof(struct ucred));
961 out:
962 	splx(s);
963 	return (error);
964 }
965 
966 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW,
967 	    0, 0,
968 	    tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
969 #endif
970 
971 
972 void
973 tcp_ctlinput(cmd, sa, vip)
974 	int cmd;
975 	struct sockaddr *sa;
976 	void *vip;
977 {
978 	register struct ip *ip = vip;
979 	register struct tcphdr *th;
980 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
981 	tcp_seq tcp_sequence = 0;
982 	int tcp_seq_check = 0;
983 
984 	if (cmd == PRC_QUENCH)
985 		notify = tcp_quench;
986 	else if ((icmp_admin_prohib_like_rst == 1) && (cmd == PRC_UNREACH_PORT) &&
987 			(ip) && ((IP_VHL_HL(ip->ip_vhl) << 2) == sizeof(struct ip))) {
988 		/*
989 		 * Only go here if the length of the IP header in the ICMP packet
990 		 * is 20 bytes, that is it doesn't have options, if it does have
991 		 * options, we will not have the first 8 bytes of the TCP header,
992 		 * and thus we cannot match against TCP source/destination port
993 		 * numbers and TCP sequence number.
994 		 */
995 		tcp_seq_check = 1;
996 		notify = tcp_drop_syn_sent;
997 	} else if (cmd == PRC_MSGSIZE)
998 		notify = tcp_mtudisc;
999 	else if (!PRC_IS_REDIRECT(cmd) &&
1000 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
1001 		return;
1002 	if (ip) {
1003 		th = (struct tcphdr *)((caddr_t)ip
1004 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
1005 		if (tcp_seq_check == 1)
1006 			tcp_sequence = ntohl(th->th_seq);
1007 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
1008 			cmd, notify, tcp_sequence, tcp_seq_check);
1009 	} else
1010 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify, 0, 0);
1011 }
1012 
1013 #ifdef INET6
1014 void
1015 tcp6_ctlinput(cmd, sa, d)
1016 	int cmd;
1017 	struct sockaddr *sa;
1018 	void *d;
1019 {
1020 	register struct tcphdr *thp;
1021 	struct tcphdr th;
1022 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
1023 	struct sockaddr_in6 sa6;
1024 	struct ip6_hdr *ip6;
1025 	struct mbuf *m;
1026 	int off;
1027 
1028 	if (sa->sa_family != AF_INET6 ||
1029 	    sa->sa_len != sizeof(struct sockaddr_in6))
1030 		return;
1031 
1032 	if (cmd == PRC_QUENCH)
1033 		notify = tcp_quench;
1034 	else if (cmd == PRC_MSGSIZE)
1035 		notify = tcp_mtudisc;
1036 	else if (!PRC_IS_REDIRECT(cmd) &&
1037 		 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1038 		return;
1039 
1040 	/* if the parameter is from icmp6, decode it. */
1041 	if (d != NULL) {
1042 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1043 		m = ip6cp->ip6c_m;
1044 		ip6 = ip6cp->ip6c_ip6;
1045 		off = ip6cp->ip6c_off;
1046 	} else {
1047 		m = NULL;
1048 		ip6 = NULL;
1049 		off = 0;	/* fool gcc */
1050 	}
1051 
1052 	/*
1053 	 * Translate addresses into internal form.
1054 	 * Sa check if it is AF_INET6 is done at the top of this funciton.
1055 	 */
1056 	sa6 = *(struct sockaddr_in6 *)sa;
1057 	if (IN6_IS_ADDR_LINKLOCAL(&sa6.sin6_addr) != 0 && m != NULL &&
1058 	    m->m_pkthdr.rcvif != NULL)
1059 		sa6.sin6_addr.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1060 
1061 	if (ip6) {
1062 		/*
1063 		 * XXX: We assume that when IPV6 is non NULL,
1064 		 * M and OFF are valid.
1065 		 */
1066 		struct in6_addr s;
1067 
1068 		/* translate addresses into internal form */
1069 		memcpy(&s, &ip6->ip6_src, sizeof(s));
1070 		if (IN6_IS_ADDR_LINKLOCAL(&s) != 0 && m != NULL &&
1071 		    m->m_pkthdr.rcvif != NULL)
1072 			s.s6_addr16[1] = htons(m->m_pkthdr.rcvif->if_index);
1073 
1074 		/* check if we can safely examine src and dst ports */
1075 		if (m->m_pkthdr.len < off + sizeof(th))
1076 			return;
1077 
1078 		if (m->m_len < off + sizeof(th)) {
1079 			/*
1080 			 * this should be rare case
1081 			 * because now MINCLSIZE is "(MHLEN + 1)",
1082 			 * so we compromise on this copy...
1083 			 */
1084 			m_copydata(m, off, sizeof(th), (caddr_t)&th);
1085 			thp = &th;
1086 		} else
1087 			thp = (struct tcphdr *)(mtod(m, caddr_t) + off);
1088 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, thp->th_dport,
1089 			      &s, thp->th_sport, cmd, notify);
1090 	} else
1091 		in6_pcbnotify(&tcb, (struct sockaddr *)&sa6, 0, &zeroin6_addr,
1092 			      0, cmd, notify);
1093 }
1094 #endif /* INET6 */
1095 
1096 /*
1097  * Check if the supplied TCP sequence number is a sequence number
1098  * for a sent but unacknowledged packet on the given TCP session.
1099  */
1100 int
1101 tcp_seq_vs_sess(inp, tcp_sequence)
1102 	struct inpcb *inp;
1103 	tcp_seq tcp_sequence;
1104 {
1105 	struct tcpcb *tp = intotcpcb(inp);
1106 	/*
1107 	 * If the sequence number is less than that of the last
1108 	 * unacknowledged packet, or greater than that of the
1109 	 * last sent, the given sequence number is not that
1110 	 * of a sent but unacknowledged packet for this session.
1111 	 */
1112 	if (SEQ_LT(tcp_sequence, tp->snd_una) ||
1113 			SEQ_GT(tcp_sequence, tp->snd_max)) {
1114 		return(0);
1115 	} else {
1116 		return(1);
1117 	}
1118 }
1119 
1120 /*
1121  * When a source quench is received, close congestion window
1122  * to one segment.  We will gradually open it again as we proceed.
1123  */
1124 void
1125 tcp_quench(inp, errno)
1126 	struct inpcb *inp;
1127 	int errno;
1128 {
1129 	struct tcpcb *tp = intotcpcb(inp);
1130 
1131 	if (tp)
1132 		tp->snd_cwnd = tp->t_maxseg;
1133 }
1134 
1135 /*
1136  * When a ICMP unreachable is recieved, drop the
1137  * TCP connection, depending on the sysctl
1138  * icmp_like_rst_syn_sent_only, it only drops
1139  * the session if it's in SYN-SENT state
1140  */
1141 void
1142 tcp_drop_syn_sent(inp, errno)
1143 	struct inpcb *inp;
1144 	int errno;
1145 {
1146 	struct tcpcb *tp = intotcpcb(inp);
1147 	if((tp) && ((icmp_like_rst_syn_sent_only == 0) ||
1148 			(tp->t_state == TCPS_SYN_SENT)))
1149 		tcp_drop(tp, errno);
1150 }
1151 
1152 /*
1153  * When `need fragmentation' ICMP is received, update our idea of the MSS
1154  * based on the new value in the route.  Also nudge TCP to send something,
1155  * since we know the packet we just sent was dropped.
1156  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1157  */
1158 void
1159 tcp_mtudisc(inp, errno)
1160 	struct inpcb *inp;
1161 	int errno;
1162 {
1163 	struct tcpcb *tp = intotcpcb(inp);
1164 	struct rtentry *rt;
1165 	struct rmxp_tao *taop;
1166 	struct socket *so = inp->inp_socket;
1167 	int offered;
1168 	int mss;
1169 #ifdef INET6
1170 	int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1171 #endif /* INET6 */
1172 
1173 	if (tp) {
1174 #ifdef INET6
1175 		if (isipv6)
1176 			rt = tcp_rtlookup6(inp);
1177 		else
1178 #endif /* INET6 */
1179 		rt = tcp_rtlookup(inp);
1180 		if (!rt || !rt->rt_rmx.rmx_mtu) {
1181 			tp->t_maxopd = tp->t_maxseg =
1182 #ifdef INET6
1183 				isipv6 ? tcp_v6mssdflt :
1184 #endif /* INET6 */
1185 				tcp_mssdflt;
1186 			return;
1187 		}
1188 		taop = rmx_taop(rt->rt_rmx);
1189 		offered = taop->tao_mssopt;
1190 		mss = rt->rt_rmx.rmx_mtu -
1191 #ifdef INET6
1192 			(isipv6 ?
1193 			 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1194 #endif /* INET6 */
1195 			 sizeof(struct tcpiphdr)
1196 #ifdef INET6
1197 			 )
1198 #endif /* INET6 */
1199 			;
1200 
1201 		if (offered)
1202 			mss = min(mss, offered);
1203 		/*
1204 		 * XXX - The above conditional probably violates the TCP
1205 		 * spec.  The problem is that, since we don't know the
1206 		 * other end's MSS, we are supposed to use a conservative
1207 		 * default.  But, if we do that, then MTU discovery will
1208 		 * never actually take place, because the conservative
1209 		 * default is much less than the MTUs typically seen
1210 		 * on the Internet today.  For the moment, we'll sweep
1211 		 * this under the carpet.
1212 		 *
1213 		 * The conservative default might not actually be a problem
1214 		 * if the only case this occurs is when sending an initial
1215 		 * SYN with options and data to a host we've never talked
1216 		 * to before.  Then, they will reply with an MSS value which
1217 		 * will get recorded and the new parameters should get
1218 		 * recomputed.  For Further Study.
1219 		 */
1220 		if (tp->t_maxopd <= mss)
1221 			return;
1222 		tp->t_maxopd = mss;
1223 
1224 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1225 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1226 			mss -= TCPOLEN_TSTAMP_APPA;
1227 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1228 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1229 			mss -= TCPOLEN_CC_APPA;
1230 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1231 		if (mss > MCLBYTES)
1232 			mss &= ~(MCLBYTES-1);
1233 #else
1234 		if (mss > MCLBYTES)
1235 			mss = mss / MCLBYTES * MCLBYTES;
1236 #endif
1237 		if (so->so_snd.sb_hiwat < mss)
1238 			mss = so->so_snd.sb_hiwat;
1239 
1240 		tp->t_maxseg = mss;
1241 
1242 		tcpstat.tcps_mturesent++;
1243 		tp->t_rtttime = 0;
1244 		tp->snd_nxt = tp->snd_una;
1245 		tcp_output(tp);
1246 	}
1247 }
1248 
1249 /*
1250  * Look-up the routing entry to the peer of this inpcb.  If no route
1251  * is found and it cannot be allocated the return NULL.  This routine
1252  * is called by TCP routines that access the rmx structure and by tcp_mss
1253  * to get the interface MTU.
1254  */
1255 struct rtentry *
1256 tcp_rtlookup(inp)
1257 	struct inpcb *inp;
1258 {
1259 	struct route *ro;
1260 	struct rtentry *rt;
1261 
1262 	ro = &inp->inp_route;
1263 	rt = ro->ro_rt;
1264 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1265 		/* No route yet, so try to acquire one */
1266 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
1267 			ro->ro_dst.sa_family = AF_INET;
1268 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
1269 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1270 				inp->inp_faddr;
1271 			rtalloc(ro);
1272 			rt = ro->ro_rt;
1273 		}
1274 	}
1275 	return rt;
1276 }
1277 
1278 #ifdef INET6
1279 struct rtentry *
1280 tcp_rtlookup6(inp)
1281 	struct inpcb *inp;
1282 {
1283 	struct route_in6 *ro6;
1284 	struct rtentry *rt;
1285 
1286 	ro6 = &inp->in6p_route;
1287 	rt = ro6->ro_rt;
1288 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1289 		/* No route yet, so try to acquire one */
1290 		if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
1291 			ro6->ro_dst.sin6_family = AF_INET6;
1292 			ro6->ro_dst.sin6_len = sizeof(ro6->ro_dst);
1293 			ro6->ro_dst.sin6_addr = inp->in6p_faddr;
1294 			rtalloc((struct route *)ro6);
1295 			rt = ro6->ro_rt;
1296 		}
1297 	}
1298 	return rt;
1299 }
1300 #endif /* INET6 */
1301 
1302 #ifdef IPSEC
1303 /* compute ESP/AH header size for TCP, including outer IP header. */
1304 size_t
1305 ipsec_hdrsiz_tcp(tp)
1306 	struct tcpcb *tp;
1307 {
1308 	struct inpcb *inp;
1309 	struct mbuf *m;
1310 	size_t hdrsiz;
1311 	struct ip *ip;
1312 #ifdef INET6
1313 	struct ip6_hdr *ip6;
1314 #endif /* INET6 */
1315 	struct tcphdr *th;
1316 
1317 	if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
1318 		return 0;
1319 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1320 	if (!m)
1321 		return 0;
1322 
1323 #ifdef INET6
1324 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1325 		ip6 = mtod(m, struct ip6_hdr *);
1326 		th = (struct tcphdr *)(ip6 + 1);
1327 		m->m_pkthdr.len = m->m_len =
1328 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1329 		bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip6,
1330 		      sizeof(struct ip6_hdr));
1331 		bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1332 		      sizeof(struct tcphdr));
1333 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1334 	} else
1335 #endif /* INET6 */
1336       {
1337 	ip = mtod(m, struct ip *);
1338 	th = (struct tcphdr *)(ip + 1);
1339 	m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1340 	bcopy((caddr_t)tp->t_template->tt_ipgen, (caddr_t)ip,
1341 	      sizeof(struct ip));
1342 	bcopy((caddr_t)&tp->t_template->tt_t, (caddr_t)th,
1343 	      sizeof(struct tcphdr));
1344 	hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1345       }
1346 
1347 	m_free(m);
1348 	return hdrsiz;
1349 }
1350 #endif /*IPSEC*/
1351 
1352 /*
1353  * Return a pointer to the cached information about the remote host.
1354  * The cached information is stored in the protocol specific part of
1355  * the route metrics.
1356  */
1357 struct rmxp_tao *
1358 tcp_gettaocache(inp)
1359 	struct inpcb *inp;
1360 {
1361 	struct rtentry *rt;
1362 
1363 #ifdef INET6
1364 	if ((inp->inp_vflag & INP_IPV6) != 0)
1365 		rt = tcp_rtlookup6(inp);
1366 	else
1367 #endif /* INET6 */
1368 	rt = tcp_rtlookup(inp);
1369 
1370 	/* Make sure this is a host route and is up. */
1371 	if (rt == NULL ||
1372 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1373 		return NULL;
1374 
1375 	return rmx_taop(rt->rt_rmx);
1376 }
1377 
1378 /*
1379  * Clear all the TAO cache entries, called from tcp_init.
1380  *
1381  * XXX
1382  * This routine is just an empty one, because we assume that the routing
1383  * routing tables are initialized at the same time when TCP, so there is
1384  * nothing in the cache left over.
1385  */
1386 static void
1387 tcp_cleartaocache()
1388 {
1389 }
1390