xref: /freebsd/sys/netinet/tcp_subr.c (revision c17d43407fe04133a94055b0dbc7ea8965654a9f)
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 #include <sys/random.h>
57 
58 #include <vm/uma.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 #ifdef INET6
96 #include <netinet6/ipsec6.h>
97 #endif
98 #endif /*IPSEC*/
99 
100 #include <machine/in_cksum.h>
101 #include <sys/md5.h>
102 
103 int 	tcp_mssdflt = TCP_MSS;
104 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
105     &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
106 
107 #ifdef INET6
108 int	tcp_v6mssdflt = TCP6_MSS;
109 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
110 	CTLFLAG_RW, &tcp_v6mssdflt , 0,
111 	"Default TCP Maximum Segment Size for IPv6");
112 #endif
113 
114 #if 0
115 static int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
116 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
117     &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
118 #endif
119 
120 int	tcp_do_rfc1323 = 1;
121 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
122     &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
123 
124 int	tcp_do_rfc1644 = 0;
125 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
126     &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
127 
128 static int	tcp_tcbhashsize = 0;
129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
130      &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
131 
132 static int	do_tcpdrain = 1;
133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
134      "Enable tcp_drain routine for extra help when low on mbufs");
135 
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
137     &tcbinfo.ipi_count, 0, "Number of active PCBs");
138 
139 static int	icmp_may_rst = 1;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
141     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
142 
143 static int	tcp_strict_rfc1948 = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW,
145     &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly");
146 
147 static int	tcp_isn_reseed_interval = 0;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
149     &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
150 
151 static void	tcp_cleartaocache(void);
152 static void	tcp_notify(struct inpcb *, int);
153 
154 /*
155  * Target size of TCP PCB hash tables. Must be a power of two.
156  *
157  * Note that this can be overridden by the kernel environment
158  * variable net.inet.tcp.tcbhashsize
159  */
160 #ifndef TCBHASHSIZE
161 #define TCBHASHSIZE	512
162 #endif
163 
164 /*
165  * This is the actual shape of what we allocate using the zone
166  * allocator.  Doing it this way allows us to protect both structures
167  * using the same generation count, and also eliminates the overhead
168  * of allocating tcpcbs separately.  By hiding the structure here,
169  * we avoid changing most of the rest of the code (although it needs
170  * to be changed, eventually, for greater efficiency).
171  */
172 #define	ALIGNMENT	32
173 #define	ALIGNM1		(ALIGNMENT - 1)
174 struct	inp_tp {
175 	union {
176 		struct	inpcb inp;
177 		char	align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
178 	} inp_tp_u;
179 	struct	tcpcb tcb;
180 	struct	callout inp_tp_rexmt, inp_tp_persist, inp_tp_keep, inp_tp_2msl;
181 	struct	callout inp_tp_delack;
182 };
183 #undef ALIGNMENT
184 #undef ALIGNM1
185 
186 /*
187  * Tcp initialization
188  */
189 void
190 tcp_init()
191 {
192 	int hashsize = TCBHASHSIZE;
193 
194 	tcp_ccgen = 1;
195 	tcp_cleartaocache();
196 
197 	tcp_delacktime = TCPTV_DELACK;
198 	tcp_keepinit = TCPTV_KEEP_INIT;
199 	tcp_keepidle = TCPTV_KEEP_IDLE;
200 	tcp_keepintvl = TCPTV_KEEPINTVL;
201 	tcp_maxpersistidle = TCPTV_KEEP_IDLE;
202 	tcp_msl = TCPTV_MSL;
203 
204 	LIST_INIT(&tcb);
205 	tcbinfo.listhead = &tcb;
206 	TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
207 	if (!powerof2(hashsize)) {
208 		printf("WARNING: TCB hash size not a power of 2\n");
209 		hashsize = 512; /* safe default */
210 	}
211 	tcp_tcbhashsize = hashsize;
212 	tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
213 	tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
214 					&tcbinfo.porthashmask);
215 	tcbinfo.ipi_zone = uma_zcreate("tcpcb", sizeof(struct inp_tp),
216 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
217 	uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
218 #ifdef INET6
219 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
220 #else /* INET6 */
221 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
222 #endif /* INET6 */
223 	if (max_protohdr < TCP_MINPROTOHDR)
224 		max_protohdr = TCP_MINPROTOHDR;
225 	if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
226 		panic("tcp_init");
227 #undef TCP_MINPROTOHDR
228 
229 	syncache_init();
230 }
231 
232 /*
233  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
234  * tcp_template used to store this data in mbufs, but we now recopy it out
235  * of the tcpcb each time to conserve mbufs.
236  */
237 void
238 tcp_fillheaders(tp, ip_ptr, tcp_ptr)
239 	struct tcpcb *tp;
240 	void *ip_ptr;
241 	void *tcp_ptr;
242 {
243 	struct inpcb *inp = tp->t_inpcb;
244 	struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
245 
246 #ifdef INET6
247 	if ((inp->inp_vflag & INP_IPV6) != 0) {
248 		struct ip6_hdr *ip6;
249 
250 		ip6 = (struct ip6_hdr *)ip_ptr;
251 		ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
252 			(inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
253 		ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
254 			(IPV6_VERSION & IPV6_VERSION_MASK);
255 		ip6->ip6_nxt = IPPROTO_TCP;
256 		ip6->ip6_plen = sizeof(struct tcphdr);
257 		ip6->ip6_src = inp->in6p_laddr;
258 		ip6->ip6_dst = inp->in6p_faddr;
259 		tcp_hdr->th_sum = 0;
260 	} else
261 #endif
262 	{
263 	struct ip *ip = (struct ip *) ip_ptr;
264 
265 	ip->ip_vhl = IP_VHL_BORING;
266 	ip->ip_tos = 0;
267 	ip->ip_len = 0;
268 	ip->ip_id = 0;
269 	ip->ip_off = 0;
270 	ip->ip_ttl = 0;
271 	ip->ip_sum = 0;
272 	ip->ip_p = IPPROTO_TCP;
273 	ip->ip_src = inp->inp_laddr;
274 	ip->ip_dst = inp->inp_faddr;
275 	tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
276 		htons(sizeof(struct tcphdr) + IPPROTO_TCP));
277 	}
278 
279 	tcp_hdr->th_sport = inp->inp_lport;
280 	tcp_hdr->th_dport = inp->inp_fport;
281 	tcp_hdr->th_seq = 0;
282 	tcp_hdr->th_ack = 0;
283 	tcp_hdr->th_x2 = 0;
284 	tcp_hdr->th_off = 5;
285 	tcp_hdr->th_flags = 0;
286 	tcp_hdr->th_win = 0;
287 	tcp_hdr->th_urp = 0;
288 }
289 
290 /*
291  * Create template to be used to send tcp packets on a connection.
292  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
293  * use for this function is in keepalives, which use tcp_respond.
294  */
295 struct tcptemp *
296 tcp_maketemplate(tp)
297 	struct tcpcb *tp;
298 {
299 	struct mbuf *m;
300 	struct tcptemp *n;
301 
302 	m = m_get(M_DONTWAIT, MT_HEADER);
303 	if (m == NULL)
304 		return (0);
305 	m->m_len = sizeof(struct tcptemp);
306 	n = mtod(m, struct tcptemp *);
307 
308 	tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
309 	return (n);
310 }
311 
312 /*
313  * Send a single message to the TCP at address specified by
314  * the given TCP/IP header.  If m == 0, then we make a copy
315  * of the tcpiphdr at ti and send directly to the addressed host.
316  * This is used to force keep alive messages out using the TCP
317  * template for a connection.  If flags are given then we send
318  * a message back to the TCP which originated the * segment ti,
319  * and discard the mbuf containing it and any other attached mbufs.
320  *
321  * In any case the ack and sequence number of the transmitted
322  * segment are as specified by the parameters.
323  *
324  * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
325  */
326 void
327 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
328 	struct tcpcb *tp;
329 	void *ipgen;
330 	register struct tcphdr *th;
331 	register struct mbuf *m;
332 	tcp_seq ack, seq;
333 	int flags;
334 {
335 	register int tlen;
336 	int win = 0;
337 	struct route *ro = 0;
338 	struct route sro;
339 	struct ip *ip;
340 	struct tcphdr *nth;
341 #ifdef INET6
342 	struct route_in6 *ro6 = 0;
343 	struct route_in6 sro6;
344 	struct ip6_hdr *ip6;
345 	int isipv6;
346 #endif /* INET6 */
347 	int ipflags = 0;
348 
349 #ifdef INET6
350 	isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
351 	ip6 = ipgen;
352 #endif /* INET6 */
353 	ip = ipgen;
354 
355 	if (tp) {
356 		if (!(flags & TH_RST)) {
357 			win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
358 			if (win > (long)TCP_MAXWIN << tp->rcv_scale)
359 				win = (long)TCP_MAXWIN << tp->rcv_scale;
360 		}
361 #ifdef INET6
362 		if (isipv6)
363 			ro6 = &tp->t_inpcb->in6p_route;
364 		else
365 #endif /* INET6 */
366 		ro = &tp->t_inpcb->inp_route;
367 	} else {
368 #ifdef INET6
369 		if (isipv6) {
370 			ro6 = &sro6;
371 			bzero(ro6, sizeof *ro6);
372 		} else
373 #endif /* INET6 */
374 	      {
375 		ro = &sro;
376 		bzero(ro, sizeof *ro);
377 	      }
378 	}
379 	if (m == 0) {
380 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
381 		if (m == NULL)
382 			return;
383 		tlen = 0;
384 		m->m_data += max_linkhdr;
385 #ifdef INET6
386 		if (isipv6) {
387 			bcopy((caddr_t)ip6, mtod(m, caddr_t),
388 			      sizeof(struct ip6_hdr));
389 			ip6 = mtod(m, struct ip6_hdr *);
390 			nth = (struct tcphdr *)(ip6 + 1);
391 		} else
392 #endif /* INET6 */
393 	      {
394 		bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
395 		ip = mtod(m, struct ip *);
396 		nth = (struct tcphdr *)(ip + 1);
397 	      }
398 		bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
399 		flags = TH_ACK;
400 	} else {
401 		m_freem(m->m_next);
402 		m->m_next = 0;
403 		m->m_data = (caddr_t)ipgen;
404 		/* m_len is set later */
405 		tlen = 0;
406 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
407 #ifdef INET6
408 		if (isipv6) {
409 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
410 			nth = (struct tcphdr *)(ip6 + 1);
411 		} else
412 #endif /* INET6 */
413 	      {
414 		xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
415 		nth = (struct tcphdr *)(ip + 1);
416 	      }
417 		if (th != nth) {
418 			/*
419 			 * this is usually a case when an extension header
420 			 * exists between the IPv6 header and the
421 			 * TCP header.
422 			 */
423 			nth->th_sport = th->th_sport;
424 			nth->th_dport = th->th_dport;
425 		}
426 		xchg(nth->th_dport, nth->th_sport, n_short);
427 #undef xchg
428 	}
429 #ifdef INET6
430 	if (isipv6) {
431 		ip6->ip6_flow = 0;
432 		ip6->ip6_vfc = IPV6_VERSION;
433 		ip6->ip6_nxt = IPPROTO_TCP;
434 		ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
435 						tlen));
436 		tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
437 	} else
438 #endif
439       {
440 	tlen += sizeof (struct tcpiphdr);
441 	ip->ip_len = tlen;
442 	ip->ip_ttl = ip_defttl;
443       }
444 	m->m_len = tlen;
445 	m->m_pkthdr.len = tlen;
446 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
447 	nth->th_seq = htonl(seq);
448 	nth->th_ack = htonl(ack);
449 	nth->th_x2 = 0;
450 	nth->th_off = sizeof (struct tcphdr) >> 2;
451 	nth->th_flags = flags;
452 	if (tp)
453 		nth->th_win = htons((u_short) (win >> tp->rcv_scale));
454 	else
455 		nth->th_win = htons((u_short)win);
456 	nth->th_urp = 0;
457 #ifdef INET6
458 	if (isipv6) {
459 		nth->th_sum = 0;
460 		nth->th_sum = in6_cksum(m, IPPROTO_TCP,
461 					sizeof(struct ip6_hdr),
462 					tlen - sizeof(struct ip6_hdr));
463 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
464 					       ro6 && ro6->ro_rt ?
465 					       ro6->ro_rt->rt_ifp :
466 					       NULL);
467 	} else
468 #endif /* INET6 */
469       {
470         nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
471 	    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
472         m->m_pkthdr.csum_flags = CSUM_TCP;
473         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
474       }
475 #ifdef TCPDEBUG
476 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
477 		tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
478 #endif
479 #ifdef IPSEC
480 	if (ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
481 		m_freem(m);
482 		return;
483 	}
484 #endif
485 #ifdef INET6
486 	if (isipv6) {
487 		(void)ip6_output(m, NULL, ro6, ipflags, NULL, NULL);
488 		if (ro6 == &sro6 && ro6->ro_rt) {
489 			RTFREE(ro6->ro_rt);
490 			ro6->ro_rt = NULL;
491 		}
492 	} else
493 #endif /* INET6 */
494       {
495 	(void) ip_output(m, NULL, ro, ipflags, NULL);
496 	if (ro == &sro && ro->ro_rt) {
497 		RTFREE(ro->ro_rt);
498 		ro->ro_rt = NULL;
499 	}
500       }
501 }
502 
503 /*
504  * Create a new TCP control block, making an
505  * empty reassembly queue and hooking it to the argument
506  * protocol control block.  The `inp' parameter must have
507  * come from the zone allocator set up in tcp_init().
508  */
509 struct tcpcb *
510 tcp_newtcpcb(inp)
511 	struct inpcb *inp;
512 {
513 	struct inp_tp *it;
514 	register struct tcpcb *tp;
515 #ifdef INET6
516 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
517 #endif /* INET6 */
518 
519 	it = (struct inp_tp *)inp;
520 	tp = &it->tcb;
521 	bzero((char *) tp, sizeof(struct tcpcb));
522 	LIST_INIT(&tp->t_segq);
523 	tp->t_maxseg = tp->t_maxopd =
524 #ifdef INET6
525 		isipv6 ? tcp_v6mssdflt :
526 #endif /* INET6 */
527 		tcp_mssdflt;
528 
529 	/* Set up our timeouts. */
530 	callout_init(tp->tt_rexmt = &it->inp_tp_rexmt, 0);
531 	callout_init(tp->tt_persist = &it->inp_tp_persist, 0);
532 	callout_init(tp->tt_keep = &it->inp_tp_keep, 0);
533 	callout_init(tp->tt_2msl = &it->inp_tp_2msl, 0);
534 	callout_init(tp->tt_delack = &it->inp_tp_delack, 0);
535 
536 	if (tcp_do_rfc1323)
537 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
538 	if (tcp_do_rfc1644)
539 		tp->t_flags |= TF_REQ_CC;
540 	tp->t_inpcb = inp;	/* XXX */
541 	/*
542 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
543 	 * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
544 	 * reasonable initial retransmit time.
545 	 */
546 	tp->t_srtt = TCPTV_SRTTBASE;
547 	tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
548 	tp->t_rttmin = TCPTV_MIN;
549 	tp->t_rxtcur = TCPTV_RTOBASE;
550 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
551 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
552 	tp->t_rcvtime = ticks;
553         /*
554 	 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
555 	 * because the socket may be bound to an IPv6 wildcard address,
556 	 * which may match an IPv4-mapped IPv6 address.
557 	 */
558 	inp->inp_ip_ttl = ip_defttl;
559 	inp->inp_ppcb = (caddr_t)tp;
560 	return (tp);		/* XXX */
561 }
562 
563 /*
564  * Drop a TCP connection, reporting
565  * the specified error.  If connection is synchronized,
566  * then send a RST to peer.
567  */
568 struct tcpcb *
569 tcp_drop(tp, errno)
570 	register struct tcpcb *tp;
571 	int errno;
572 {
573 	struct socket *so = tp->t_inpcb->inp_socket;
574 
575 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
576 		tp->t_state = TCPS_CLOSED;
577 		(void) tcp_output(tp);
578 		tcpstat.tcps_drops++;
579 	} else
580 		tcpstat.tcps_conndrops++;
581 	if (errno == ETIMEDOUT && tp->t_softerror)
582 		errno = tp->t_softerror;
583 	so->so_error = errno;
584 	return (tcp_close(tp));
585 }
586 
587 /*
588  * Close a TCP control block:
589  *	discard all space held by the tcp
590  *	discard internet protocol block
591  *	wake up any sleepers
592  */
593 struct tcpcb *
594 tcp_close(tp)
595 	register struct tcpcb *tp;
596 {
597 	register struct tseg_qent *q;
598 	struct inpcb *inp = tp->t_inpcb;
599 	struct socket *so = inp->inp_socket;
600 #ifdef INET6
601 	int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
602 #endif /* INET6 */
603 	register struct rtentry *rt;
604 	int dosavessthresh;
605 
606 	/*
607 	 * Make sure that all of our timers are stopped before we
608 	 * delete the PCB.
609 	 */
610 	callout_stop(tp->tt_rexmt);
611 	callout_stop(tp->tt_persist);
612 	callout_stop(tp->tt_keep);
613 	callout_stop(tp->tt_2msl);
614 	callout_stop(tp->tt_delack);
615 
616 	/*
617 	 * If we got enough samples through the srtt filter,
618 	 * save the rtt and rttvar in the routing entry.
619 	 * 'Enough' is arbitrarily defined as the 16 samples.
620 	 * 16 samples is enough for the srtt filter to converge
621 	 * to within 5% of the correct value; fewer samples and
622 	 * we could save a very bogus rtt.
623 	 *
624 	 * Don't update the default route's characteristics and don't
625 	 * update anything that the user "locked".
626 	 */
627 	if (tp->t_rttupdated >= 16) {
628 		register u_long i = 0;
629 #ifdef INET6
630 		if (isipv6) {
631 			struct sockaddr_in6 *sin6;
632 
633 			if ((rt = inp->in6p_route.ro_rt) == NULL)
634 				goto no_valid_rt;
635 			sin6 = (struct sockaddr_in6 *)rt_key(rt);
636 			if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
637 				goto no_valid_rt;
638 		}
639 		else
640 #endif /* INET6 */
641 		if ((rt = inp->inp_route.ro_rt) == NULL ||
642 		    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr
643 		    == INADDR_ANY)
644 			goto no_valid_rt;
645 
646 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
647 			i = tp->t_srtt *
648 			    (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
649 			if (rt->rt_rmx.rmx_rtt && i)
650 				/*
651 				 * filter this update to half the old & half
652 				 * the new values, converting scale.
653 				 * See route.h and tcp_var.h for a
654 				 * description of the scaling constants.
655 				 */
656 				rt->rt_rmx.rmx_rtt =
657 				    (rt->rt_rmx.rmx_rtt + i) / 2;
658 			else
659 				rt->rt_rmx.rmx_rtt = i;
660 			tcpstat.tcps_cachedrtt++;
661 		}
662 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
663 			i = tp->t_rttvar *
664 			    (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
665 			if (rt->rt_rmx.rmx_rttvar && i)
666 				rt->rt_rmx.rmx_rttvar =
667 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
668 			else
669 				rt->rt_rmx.rmx_rttvar = i;
670 			tcpstat.tcps_cachedrttvar++;
671 		}
672 		/*
673 		 * The old comment here said:
674 		 * update the pipelimit (ssthresh) if it has been updated
675 		 * already or if a pipesize was specified & the threshhold
676 		 * got below half the pipesize.  I.e., wait for bad news
677 		 * before we start updating, then update on both good
678 		 * and bad news.
679 		 *
680 		 * But we want to save the ssthresh even if no pipesize is
681 		 * specified explicitly in the route, because such
682 		 * connections still have an implicit pipesize specified
683 		 * by the global tcp_sendspace.  In the absence of a reliable
684 		 * way to calculate the pipesize, it will have to do.
685 		 */
686 		i = tp->snd_ssthresh;
687 		if (rt->rt_rmx.rmx_sendpipe != 0)
688 			dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
689 		else
690 			dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
691 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
692 		     i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
693 		    || dosavessthresh) {
694 			/*
695 			 * convert the limit from user data bytes to
696 			 * packets then to packet data bytes.
697 			 */
698 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
699 			if (i < 2)
700 				i = 2;
701 			i *= (u_long)(tp->t_maxseg +
702 #ifdef INET6
703 				      (isipv6 ? sizeof (struct ip6_hdr) +
704 					       sizeof (struct tcphdr) :
705 #endif
706 				       sizeof (struct tcpiphdr)
707 #ifdef INET6
708 				       )
709 #endif
710 				      );
711 			if (rt->rt_rmx.rmx_ssthresh)
712 				rt->rt_rmx.rmx_ssthresh =
713 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
714 			else
715 				rt->rt_rmx.rmx_ssthresh = i;
716 			tcpstat.tcps_cachedssthresh++;
717 		}
718 	}
719     no_valid_rt:
720 	/* free the reassembly queue, if any */
721 	while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
722 		LIST_REMOVE(q, tqe_q);
723 		m_freem(q->tqe_m);
724 		FREE(q, M_TSEGQ);
725 	}
726 	inp->inp_ppcb = NULL;
727 	soisdisconnected(so);
728 #ifdef INET6
729 	if (INP_CHECK_SOCKAF(so, AF_INET6))
730 		in6_pcbdetach(inp);
731 	else
732 #endif /* INET6 */
733 	in_pcbdetach(inp);
734 	tcpstat.tcps_closed++;
735 	return ((struct tcpcb *)0);
736 }
737 
738 void
739 tcp_drain()
740 {
741 	if (do_tcpdrain)
742 	{
743 		struct inpcb *inpb;
744 		struct tcpcb *tcpb;
745 		struct tseg_qent *te;
746 
747 	/*
748 	 * Walk the tcpbs, if existing, and flush the reassembly queue,
749 	 * if there is one...
750 	 * XXX: The "Net/3" implementation doesn't imply that the TCP
751 	 *      reassembly queue should be flushed, but in a situation
752 	 * 	where we're really low on mbufs, this is potentially
753 	 *  	usefull.
754 	 */
755 		LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
756 			if ((tcpb = intotcpcb(inpb))) {
757 				while ((te = LIST_FIRST(&tcpb->t_segq))
758 			            != NULL) {
759 					LIST_REMOVE(te, tqe_q);
760 					m_freem(te->tqe_m);
761 					FREE(te, M_TSEGQ);
762 				}
763 			}
764 		}
765 	}
766 }
767 
768 /*
769  * Notify a tcp user of an asynchronous error;
770  * store error as soft error, but wake up user
771  * (for now, won't do anything until can select for soft error).
772  *
773  * Do not wake up user since there currently is no mechanism for
774  * reporting soft errors (yet - a kqueue filter may be added).
775  */
776 static void
777 tcp_notify(inp, error)
778 	struct inpcb *inp;
779 	int error;
780 {
781 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
782 
783 	/*
784 	 * Ignore some errors if we are hooked up.
785 	 * If connection hasn't completed, has retransmitted several times,
786 	 * and receives a second error, give up now.  This is better
787 	 * than waiting a long time to establish a connection that
788 	 * can never complete.
789 	 */
790 	if (tp->t_state == TCPS_ESTABLISHED &&
791 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
792 	      error == EHOSTDOWN)) {
793 		return;
794 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
795 	    tp->t_softerror)
796 		tcp_drop(tp, error);
797 	else
798 		tp->t_softerror = error;
799 #if 0
800 	wakeup((caddr_t) &so->so_timeo);
801 	sorwakeup(so);
802 	sowwakeup(so);
803 #endif
804 }
805 
806 static int
807 tcp_pcblist(SYSCTL_HANDLER_ARGS)
808 {
809 	int error, i, n, s;
810 	struct inpcb *inp, **inp_list;
811 	inp_gen_t gencnt;
812 	struct xinpgen xig;
813 
814 	/*
815 	 * The process of preparing the TCB list is too time-consuming and
816 	 * resource-intensive to repeat twice on every request.
817 	 */
818 	if (req->oldptr == 0) {
819 		n = tcbinfo.ipi_count;
820 		req->oldidx = 2 * (sizeof xig)
821 			+ (n + n/8) * sizeof(struct xtcpcb);
822 		return 0;
823 	}
824 
825 	if (req->newptr != 0)
826 		return EPERM;
827 
828 	/*
829 	 * OK, now we're committed to doing something.
830 	 */
831 	s = splnet();
832 	gencnt = tcbinfo.ipi_gencnt;
833 	n = tcbinfo.ipi_count;
834 	splx(s);
835 
836 	xig.xig_len = sizeof xig;
837 	xig.xig_count = n;
838 	xig.xig_gen = gencnt;
839 	xig.xig_sogen = so_gencnt;
840 	error = SYSCTL_OUT(req, &xig, sizeof xig);
841 	if (error)
842 		return error;
843 
844 	inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
845 	if (inp_list == 0)
846 		return ENOMEM;
847 
848 	s = splnet();
849 	for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
850 	     inp = LIST_NEXT(inp, inp_list)) {
851 		if (inp->inp_gencnt <= gencnt) {
852 			if (cr_canseesocket(req->td->td_ucred,
853 			    inp->inp_socket))
854 				continue;
855 			inp_list[i++] = inp;
856 		}
857 	}
858 	splx(s);
859 	n = i;
860 
861 	error = 0;
862 	for (i = 0; i < n; i++) {
863 		inp = inp_list[i];
864 		if (inp->inp_gencnt <= gencnt) {
865 			struct xtcpcb xt;
866 			caddr_t inp_ppcb;
867 			xt.xt_len = sizeof xt;
868 			/* XXX should avoid extra copy */
869 			bcopy(inp, &xt.xt_inp, sizeof *inp);
870 			inp_ppcb = inp->inp_ppcb;
871 			if (inp_ppcb != NULL)
872 				bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
873 			else
874 				bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
875 			if (inp->inp_socket)
876 				sotoxsocket(inp->inp_socket, &xt.xt_socket);
877 			error = SYSCTL_OUT(req, &xt, sizeof xt);
878 		}
879 	}
880 	if (!error) {
881 		/*
882 		 * Give the user an updated idea of our state.
883 		 * If the generation differs from what we told
884 		 * her before, she knows that something happened
885 		 * while we were processing this request, and it
886 		 * might be necessary to retry.
887 		 */
888 		s = splnet();
889 		xig.xig_gen = tcbinfo.ipi_gencnt;
890 		xig.xig_sogen = so_gencnt;
891 		xig.xig_count = tcbinfo.ipi_count;
892 		splx(s);
893 		error = SYSCTL_OUT(req, &xig, sizeof xig);
894 	}
895 	free(inp_list, M_TEMP);
896 	return error;
897 }
898 
899 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
900 	    tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
901 
902 static int
903 tcp_getcred(SYSCTL_HANDLER_ARGS)
904 {
905 	struct xucred xuc;
906 	struct sockaddr_in addrs[2];
907 	struct inpcb *inp;
908 	int error, s;
909 
910 	error = suser_xxx(0, req->td->td_proc, PRISON_ROOT);
911 	if (error)
912 		return (error);
913 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
914 	if (error)
915 		return (error);
916 	s = splnet();
917 	inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
918 	    addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
919 	if (inp == NULL || inp->inp_socket == NULL) {
920 		error = ENOENT;
921 		goto out;
922 	}
923 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
924 	if (error)
925 		goto out;
926 	cru2x(inp->inp_socket->so_cred, &xuc);
927 	error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
928 out:
929 	splx(s);
930 	return (error);
931 }
932 
933 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
934     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
935     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
936 
937 #ifdef INET6
938 static int
939 tcp6_getcred(SYSCTL_HANDLER_ARGS)
940 {
941 	struct xucred xuc;
942 	struct sockaddr_in6 addrs[2];
943 	struct inpcb *inp;
944 	int error, s, mapped = 0;
945 
946 	error = suser_xxx(0, req->td->td_proc, PRISON_ROOT);
947 	if (error)
948 		return (error);
949 	error = SYSCTL_IN(req, addrs, sizeof(addrs));
950 	if (error)
951 		return (error);
952 	if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
953 		if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
954 			mapped = 1;
955 		else
956 			return (EINVAL);
957 	}
958 	s = splnet();
959 	if (mapped == 1)
960 		inp = in_pcblookup_hash(&tcbinfo,
961 			*(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
962 			addrs[1].sin6_port,
963 			*(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
964 			addrs[0].sin6_port,
965 			0, NULL);
966 	else
967 		inp = in6_pcblookup_hash(&tcbinfo, &addrs[1].sin6_addr,
968 				 addrs[1].sin6_port,
969 				 &addrs[0].sin6_addr, addrs[0].sin6_port,
970 				 0, NULL);
971 	if (inp == NULL || inp->inp_socket == NULL) {
972 		error = ENOENT;
973 		goto out;
974 	}
975 	error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
976 	if (error)
977 		goto out;
978 	cru2x(inp->inp_socket->so_cred, &xuc);
979 	error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
980 out:
981 	splx(s);
982 	return (error);
983 }
984 
985 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
986     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
987     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
988 #endif
989 
990 
991 void
992 tcp_ctlinput(cmd, sa, vip)
993 	int cmd;
994 	struct sockaddr *sa;
995 	void *vip;
996 {
997 	struct ip *ip = vip;
998 	struct tcphdr *th;
999 	struct in_addr faddr;
1000 	struct inpcb *inp;
1001 	struct tcpcb *tp;
1002 	void (*notify)(struct inpcb *, int) = tcp_notify;
1003 	tcp_seq icmp_seq;
1004 	int s;
1005 
1006 	faddr = ((struct sockaddr_in *)sa)->sin_addr;
1007 	if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1008 		return;
1009 
1010 	if (cmd == PRC_QUENCH)
1011 		notify = tcp_quench;
1012 	else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1013 		cmd == PRC_UNREACH_PORT) && ip)
1014 		notify = tcp_drop_syn_sent;
1015 	else if (cmd == PRC_MSGSIZE)
1016 		notify = tcp_mtudisc;
1017 	else if (PRC_IS_REDIRECT(cmd)) {
1018 		ip = 0;
1019 		notify = in_rtchange;
1020 	} else if (cmd == PRC_HOSTDEAD)
1021 		ip = 0;
1022 	else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
1023 		return;
1024 	if (ip) {
1025 		s = splnet();
1026 		th = (struct tcphdr *)((caddr_t)ip
1027 				       + (IP_VHL_HL(ip->ip_vhl) << 2));
1028 		inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1029 		    ip->ip_src, th->th_sport, 0, NULL);
1030 		if (inp != NULL && inp->inp_socket != NULL) {
1031 			icmp_seq = htonl(th->th_seq);
1032 			tp = intotcpcb(inp);
1033 			if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1034 			    SEQ_LT(icmp_seq, tp->snd_max))
1035 				(*notify)(inp, inetctlerrmap[cmd]);
1036 		} else {
1037 			struct in_conninfo inc;
1038 
1039 			inc.inc_fport = th->th_dport;
1040 			inc.inc_lport = th->th_sport;
1041 			inc.inc_faddr = faddr;
1042 			inc.inc_laddr = ip->ip_src;
1043 #ifdef INET6
1044 			inc.inc_isipv6 = 0;
1045 #endif
1046 			syncache_unreach(&inc, th);
1047 		}
1048 		splx(s);
1049 	} else
1050 		in_pcbnotifyall(&tcb, faddr, inetctlerrmap[cmd], notify);
1051 }
1052 
1053 #ifdef INET6
1054 void
1055 tcp6_ctlinput(cmd, sa, d)
1056 	int cmd;
1057 	struct sockaddr *sa;
1058 	void *d;
1059 {
1060 	struct tcphdr th;
1061 	void (*notify)(struct inpcb *, int) = tcp_notify;
1062 	struct ip6_hdr *ip6;
1063 	struct mbuf *m;
1064 	struct ip6ctlparam *ip6cp = NULL;
1065 	const struct sockaddr_in6 *sa6_src = NULL;
1066 	int off;
1067 	struct tcp_portonly {
1068 		u_int16_t th_sport;
1069 		u_int16_t th_dport;
1070 	} *thp;
1071 
1072 	if (sa->sa_family != AF_INET6 ||
1073 	    sa->sa_len != sizeof(struct sockaddr_in6))
1074 		return;
1075 
1076 	if (cmd == PRC_QUENCH)
1077 		notify = tcp_quench;
1078 	else if (cmd == PRC_MSGSIZE)
1079 		notify = tcp_mtudisc;
1080 	else if (!PRC_IS_REDIRECT(cmd) &&
1081 		 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1082 		return;
1083 
1084 	/* if the parameter is from icmp6, decode it. */
1085 	if (d != NULL) {
1086 		ip6cp = (struct ip6ctlparam *)d;
1087 		m = ip6cp->ip6c_m;
1088 		ip6 = ip6cp->ip6c_ip6;
1089 		off = ip6cp->ip6c_off;
1090 		sa6_src = ip6cp->ip6c_src;
1091 	} else {
1092 		m = NULL;
1093 		ip6 = NULL;
1094 		off = 0;	/* fool gcc */
1095 		sa6_src = &sa6_any;
1096 	}
1097 
1098 	if (ip6) {
1099 		struct in_conninfo inc;
1100 		/*
1101 		 * XXX: We assume that when IPV6 is non NULL,
1102 		 * M and OFF are valid.
1103 		 */
1104 
1105 		/* check if we can safely examine src and dst ports */
1106 		if (m->m_pkthdr.len < off + sizeof(*thp))
1107 			return;
1108 
1109 		bzero(&th, sizeof(th));
1110 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1111 
1112 		in6_pcbnotify(&tcb, sa, th.th_dport,
1113 		    (struct sockaddr *)ip6cp->ip6c_src,
1114 		    th.th_sport, cmd, notify);
1115 
1116 		inc.inc_fport = th.th_dport;
1117 		inc.inc_lport = th.th_sport;
1118 		inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1119 		inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1120 		inc.inc_isipv6 = 1;
1121 		syncache_unreach(&inc, &th);
1122 	} else
1123 		in6_pcbnotify(&tcb, sa, 0, (const struct sockaddr *)sa6_src,
1124 			      0, cmd, notify);
1125 }
1126 #endif /* INET6 */
1127 
1128 
1129 /*
1130  * Following is where TCP initial sequence number generation occurs.
1131  *
1132  * There are two places where we must use initial sequence numbers:
1133  * 1.  In SYN-ACK packets.
1134  * 2.  In SYN packets.
1135  *
1136  * The ISNs in SYN-ACK packets have no monotonicity requirement,
1137  * and should be as unpredictable as possible to avoid the possibility
1138  * of spoofing and/or connection hijacking.  To satisfy this
1139  * requirement, SYN-ACK ISNs are generated via the arc4random()
1140  * function.  If exact RFC 1948 compliance is requested via sysctl,
1141  * these ISNs will be generated just like those in SYN packets.
1142  *
1143  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1144  * depends on this property.  In addition, these ISNs should be
1145  * unguessable so as to prevent connection hijacking.  To satisfy
1146  * the requirements of this situation, the algorithm outlined in
1147  * RFC 1948 is used to generate sequence numbers.
1148  *
1149  * For more information on the theory of operation, please see
1150  * RFC 1948.
1151  *
1152  * Implementation details:
1153  *
1154  * Time is based off the system timer, and is corrected so that it
1155  * increases by one megabyte per second.  This allows for proper
1156  * recycling on high speed LANs while still leaving over an hour
1157  * before rollover.
1158  *
1159  * Two sysctls control the generation of ISNs:
1160  *
1161  * net.inet.tcp.isn_reseed_interval controls the number of seconds
1162  * between seeding of isn_secret.  This is normally set to zero,
1163  * as reseeding should not be necessary.
1164  *
1165  * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
1166  * strictly.  When strict compliance is requested, reseeding is
1167  * disabled and SYN-ACKs will be generated in the same manner as
1168  * SYNs.  Strict mode is disabled by default.
1169  *
1170  */
1171 
1172 #define ISN_BYTES_PER_SECOND 1048576
1173 
1174 u_char isn_secret[32];
1175 int isn_last_reseed;
1176 MD5_CTX isn_ctx;
1177 
1178 tcp_seq
1179 tcp_new_isn(tp)
1180 	struct tcpcb *tp;
1181 {
1182 	u_int32_t md5_buffer[4];
1183 	tcp_seq new_isn;
1184 
1185 	/* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
1186 	if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT))
1187 	   && tcp_strict_rfc1948 == 0)
1188 		return arc4random();
1189 
1190 	/* Seed if this is the first use, reseed if requested. */
1191 	if ((isn_last_reseed == 0) ||
1192 	    ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
1193 	     (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1194 		< (u_int)ticks))) {
1195 		read_random(&isn_secret, sizeof(isn_secret));
1196 		isn_last_reseed = ticks;
1197 	}
1198 
1199 	/* Compute the md5 hash and return the ISN. */
1200 	MD5Init(&isn_ctx);
1201 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1202 	MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1203 #ifdef INET6
1204 	if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1205 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1206 			  sizeof(struct in6_addr));
1207 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1208 			  sizeof(struct in6_addr));
1209 	} else
1210 #endif
1211 	{
1212 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1213 			  sizeof(struct in_addr));
1214 		MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1215 			  sizeof(struct in_addr));
1216 	}
1217 	MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1218 	MD5Final((u_char *) &md5_buffer, &isn_ctx);
1219 	new_isn = (tcp_seq) md5_buffer[0];
1220 	new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
1221 	return new_isn;
1222 }
1223 
1224 /*
1225  * When a source quench is received, close congestion window
1226  * to one segment.  We will gradually open it again as we proceed.
1227  */
1228 void
1229 tcp_quench(inp, errno)
1230 	struct inpcb *inp;
1231 	int errno;
1232 {
1233 	struct tcpcb *tp = intotcpcb(inp);
1234 
1235 	if (tp)
1236 		tp->snd_cwnd = tp->t_maxseg;
1237 }
1238 
1239 /*
1240  * When a specific ICMP unreachable message is received and the
1241  * connection state is SYN-SENT, drop the connection.  This behavior
1242  * is controlled by the icmp_may_rst sysctl.
1243  */
1244 void
1245 tcp_drop_syn_sent(inp, errno)
1246 	struct inpcb *inp;
1247 	int errno;
1248 {
1249 	struct tcpcb *tp = intotcpcb(inp);
1250 
1251 	if (tp && tp->t_state == TCPS_SYN_SENT)
1252 		tcp_drop(tp, errno);
1253 }
1254 
1255 /*
1256  * When `need fragmentation' ICMP is received, update our idea of the MSS
1257  * based on the new value in the route.  Also nudge TCP to send something,
1258  * since we know the packet we just sent was dropped.
1259  * This duplicates some code in the tcp_mss() function in tcp_input.c.
1260  */
1261 void
1262 tcp_mtudisc(inp, errno)
1263 	struct inpcb *inp;
1264 	int errno;
1265 {
1266 	struct tcpcb *tp = intotcpcb(inp);
1267 	struct rtentry *rt;
1268 	struct rmxp_tao *taop;
1269 	struct socket *so = inp->inp_socket;
1270 	int offered;
1271 	int mss;
1272 #ifdef INET6
1273 	int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1274 #endif /* INET6 */
1275 
1276 	if (tp) {
1277 #ifdef INET6
1278 		if (isipv6)
1279 			rt = tcp_rtlookup6(&inp->inp_inc);
1280 		else
1281 #endif /* INET6 */
1282 		rt = tcp_rtlookup(&inp->inp_inc);
1283 		if (!rt || !rt->rt_rmx.rmx_mtu) {
1284 			tp->t_maxopd = tp->t_maxseg =
1285 #ifdef INET6
1286 				isipv6 ? tcp_v6mssdflt :
1287 #endif /* INET6 */
1288 				tcp_mssdflt;
1289 			return;
1290 		}
1291 		taop = rmx_taop(rt->rt_rmx);
1292 		offered = taop->tao_mssopt;
1293 		mss = rt->rt_rmx.rmx_mtu -
1294 #ifdef INET6
1295 			(isipv6 ?
1296 			 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1297 #endif /* INET6 */
1298 			 sizeof(struct tcpiphdr)
1299 #ifdef INET6
1300 			 )
1301 #endif /* INET6 */
1302 			;
1303 
1304 		if (offered)
1305 			mss = min(mss, offered);
1306 		/*
1307 		 * XXX - The above conditional probably violates the TCP
1308 		 * spec.  The problem is that, since we don't know the
1309 		 * other end's MSS, we are supposed to use a conservative
1310 		 * default.  But, if we do that, then MTU discovery will
1311 		 * never actually take place, because the conservative
1312 		 * default is much less than the MTUs typically seen
1313 		 * on the Internet today.  For the moment, we'll sweep
1314 		 * this under the carpet.
1315 		 *
1316 		 * The conservative default might not actually be a problem
1317 		 * if the only case this occurs is when sending an initial
1318 		 * SYN with options and data to a host we've never talked
1319 		 * to before.  Then, they will reply with an MSS value which
1320 		 * will get recorded and the new parameters should get
1321 		 * recomputed.  For Further Study.
1322 		 */
1323 		if (tp->t_maxopd <= mss)
1324 			return;
1325 		tp->t_maxopd = mss;
1326 
1327 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1328 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1329 			mss -= TCPOLEN_TSTAMP_APPA;
1330 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1331 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1332 			mss -= TCPOLEN_CC_APPA;
1333 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
1334 		if (mss > MCLBYTES)
1335 			mss &= ~(MCLBYTES-1);
1336 #else
1337 		if (mss > MCLBYTES)
1338 			mss = mss / MCLBYTES * MCLBYTES;
1339 #endif
1340 		if (so->so_snd.sb_hiwat < mss)
1341 			mss = so->so_snd.sb_hiwat;
1342 
1343 		tp->t_maxseg = mss;
1344 
1345 		tcpstat.tcps_mturesent++;
1346 		tp->t_rtttime = 0;
1347 		tp->snd_nxt = tp->snd_una;
1348 		tcp_output(tp);
1349 	}
1350 }
1351 
1352 /*
1353  * Look-up the routing entry to the peer of this inpcb.  If no route
1354  * is found and it cannot be allocated the return NULL.  This routine
1355  * is called by TCP routines that access the rmx structure and by tcp_mss
1356  * to get the interface MTU.
1357  */
1358 struct rtentry *
1359 tcp_rtlookup(inc)
1360 	struct in_conninfo *inc;
1361 {
1362 	struct route *ro;
1363 	struct rtentry *rt;
1364 
1365 	ro = &inc->inc_route;
1366 	rt = ro->ro_rt;
1367 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1368 		/* No route yet, so try to acquire one */
1369 		if (inc->inc_faddr.s_addr != INADDR_ANY) {
1370 			ro->ro_dst.sa_family = AF_INET;
1371 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1372 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1373 			    inc->inc_faddr;
1374 			rtalloc(ro);
1375 			rt = ro->ro_rt;
1376 		}
1377 	}
1378 	return rt;
1379 }
1380 
1381 #ifdef INET6
1382 struct rtentry *
1383 tcp_rtlookup6(inc)
1384 	struct in_conninfo *inc;
1385 {
1386 	struct route_in6 *ro6;
1387 	struct rtentry *rt;
1388 
1389 	ro6 = &inc->inc6_route;
1390 	rt = ro6->ro_rt;
1391 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
1392 		/* No route yet, so try to acquire one */
1393 		if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1394 			ro6->ro_dst.sin6_family = AF_INET6;
1395 			ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1396 			ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1397 			rtalloc((struct route *)ro6);
1398 			rt = ro6->ro_rt;
1399 		}
1400 	}
1401 	return rt;
1402 }
1403 #endif /* INET6 */
1404 
1405 #ifdef IPSEC
1406 /* compute ESP/AH header size for TCP, including outer IP header. */
1407 size_t
1408 ipsec_hdrsiz_tcp(tp)
1409 	struct tcpcb *tp;
1410 {
1411 	struct inpcb *inp;
1412 	struct mbuf *m;
1413 	size_t hdrsiz;
1414 	struct ip *ip;
1415 #ifdef INET6
1416 	struct ip6_hdr *ip6;
1417 #endif /* INET6 */
1418 	struct tcphdr *th;
1419 
1420 	if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1421 		return 0;
1422 	MGETHDR(m, M_DONTWAIT, MT_DATA);
1423 	if (!m)
1424 		return 0;
1425 
1426 #ifdef INET6
1427 	if ((inp->inp_vflag & INP_IPV6) != 0) {
1428 		ip6 = mtod(m, struct ip6_hdr *);
1429 		th = (struct tcphdr *)(ip6 + 1);
1430 		m->m_pkthdr.len = m->m_len =
1431 			sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1432 		tcp_fillheaders(tp, ip6, th);
1433 		hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1434 	} else
1435 #endif /* INET6 */
1436       {
1437 	ip = mtod(m, struct ip *);
1438 	th = (struct tcphdr *)(ip + 1);
1439 	m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1440 	tcp_fillheaders(tp, ip, th);
1441 	hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1442       }
1443 
1444 	m_free(m);
1445 	return hdrsiz;
1446 }
1447 #endif /*IPSEC*/
1448 
1449 /*
1450  * Return a pointer to the cached information about the remote host.
1451  * The cached information is stored in the protocol specific part of
1452  * the route metrics.
1453  */
1454 struct rmxp_tao *
1455 tcp_gettaocache(inc)
1456 	struct in_conninfo *inc;
1457 {
1458 	struct rtentry *rt;
1459 
1460 #ifdef INET6
1461 	if (inc->inc_isipv6)
1462 		rt = tcp_rtlookup6(inc);
1463 	else
1464 #endif /* INET6 */
1465 	rt = tcp_rtlookup(inc);
1466 
1467 	/* Make sure this is a host route and is up. */
1468 	if (rt == NULL ||
1469 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
1470 		return NULL;
1471 
1472 	return rmx_taop(rt->rt_rmx);
1473 }
1474 
1475 /*
1476  * Clear all the TAO cache entries, called from tcp_init.
1477  *
1478  * XXX
1479  * This routine is just an empty one, because we assume that the routing
1480  * routing tables are initialized at the same time when TCP, so there is
1481  * nothing in the cache left over.
1482  */
1483 static void
1484 tcp_cleartaocache()
1485 {
1486 }
1487