xref: /openbsd-src/sys/netinet/tcp_subr.c (revision db3296cf5c1dd9058ceecc3a29fe4aaa0bd26000)
1 /*	$OpenBSD: tcp_subr.c,v 1.68 2003/07/09 22:03:16 itojun Exp $	*/
2 /*	$NetBSD: tcp_subr.c,v 1.22 1996/02/13 23:44:00 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1982, 1986, 1988, 1990, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)COPYRIGHT	1.1 (NRL) 17 January 1995
33  *
34  * NRL grants permission for redistribution and use in source and binary
35  * forms, with or without modification, of the software and documentation
36  * created at NRL provided that the following conditions are met:
37  *
38  * 1. Redistributions of source code must retain the above copyright
39  *    notice, this list of conditions and the following disclaimer.
40  * 2. Redistributions in binary form must reproduce the above copyright
41  *    notice, this list of conditions and the following disclaimer in the
42  *    documentation and/or other materials provided with the distribution.
43  * 3. All advertising materials mentioning features or use of this software
44  *    must display the following acknowledgements:
45  * 	This product includes software developed by the University of
46  * 	California, Berkeley and its contributors.
47  * 	This product includes software developed at the Information
48  * 	Technology Division, US Naval Research Laboratory.
49  * 4. Neither the name of the NRL nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
54  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
55  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
56  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
57  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
58  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
59  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
60  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
61  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
62  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
63  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
64  *
65  * The views and conclusions contained in the software and documentation
66  * are those of the authors and should not be interpreted as representing
67  * official policies, either expressed or implied, of the US Naval
68  * Research Laboratory (NRL).
69  */
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/proc.h>
74 #include <sys/mbuf.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/protosw.h>
78 #include <sys/kernel.h>
79 
80 #include <net/route.h>
81 #include <net/if.h>
82 
83 #include <netinet/in.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/ip.h>
86 #include <netinet/in_pcb.h>
87 #include <netinet/ip_var.h>
88 #include <netinet/ip_icmp.h>
89 #include <netinet/tcp.h>
90 #include <netinet/tcp_fsm.h>
91 #include <netinet/tcp_seq.h>
92 #include <netinet/tcp_timer.h>
93 #include <netinet/tcp_var.h>
94 #include <netinet/tcpip.h>
95 #include <dev/rndvar.h>
96 
97 #ifdef INET6
98 #include <netinet6/in6_var.h>
99 #include <netinet6/ip6protosw.h>
100 #endif /* INET6 */
101 
102 #ifdef TCP_SIGNATURE
103 #include <sys/md5k.h>
104 #endif /* TCP_SIGNATURE */
105 
106 /* patchable/settable parameters for tcp */
107 int	tcp_mssdflt = TCP_MSS;
108 int	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
109 
110 /*
111  * Configure kernel with options "TCP_DO_RFC1323=0" to disable RFC1323 stuff.
112  * This is a good idea over slow SLIP/PPP links, because the timestamp
113  * pretty well destroys the VJ compression (any packet with a timestamp
114  * different from the previous one can't be compressed), as well as adding
115  * more overhead.
116  * XXX And it should be a settable per route characteristic (with this just
117  * used as the default).
118  */
119 #ifndef TCP_DO_RFC1323
120 #define TCP_DO_RFC1323	1
121 #endif
122 int	tcp_do_rfc1323 = TCP_DO_RFC1323;
123 
124 #ifndef TCP_DO_SACK
125 #ifdef TCP_SACK
126 #define TCP_DO_SACK	1
127 #else
128 #define TCP_DO_SACK	0
129 #endif
130 #endif
131 int	tcp_do_sack = TCP_DO_SACK;		/* RFC 2018 selective ACKs */
132 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
133 int	tcp_do_ecn = 0;		/* RFC3168 ECN enabled/disabled? */
134 
135 u_int32_t	tcp_now;
136 
137 #ifndef TCBHASHSIZE
138 #define	TCBHASHSIZE	128
139 #endif
140 int	tcbhashsize = TCBHASHSIZE;
141 
142 #ifdef INET6
143 extern int ip6_defhlim;
144 #endif /* INET6 */
145 
146 struct pool tcpcb_pool;
147 #ifdef TCP_SACK
148 struct pool sackhl_pool;
149 #endif
150 
151 int	tcp_freeq(struct tcpcb *);
152 
153 struct tcpstat tcpstat;		/* tcp statistics */
154 tcp_seq  tcp_iss;
155 
156 /*
157  * Tcp initialization
158  */
159 void
160 tcp_init()
161 {
162 #ifdef TCP_COMPAT_42
163 	tcp_iss = 1;		/* wrong */
164 #endif /* TCP_COMPAT_42 */
165 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
166 	    NULL);
167 #ifdef TCP_SACK
168 	pool_init(&sackhl_pool, sizeof(struct sackhole), 0, 0, 0, "sackhlpl",
169 	    NULL);
170 #endif /* TCP_SACK */
171 	in_pcbinit(&tcbtable, tcbhashsize);
172 	tcp_now = arc4random() / 2;
173 
174 #ifdef INET6
175 	/*
176 	 * Since sizeof(struct ip6_hdr) > sizeof(struct ip), we
177 	 * do max length checks/computations only on the former.
178 	 */
179 	if (max_protohdr < (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)))
180 		max_protohdr = (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
181 	if ((max_linkhdr + sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) >
182 	    MHLEN)
183 		panic("tcp_init");
184 
185 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
186 #endif /* INET6 */
187 
188 	/* Initialize timer state. */
189 	tcp_timer_init();
190 }
191 
192 /*
193  * Create template to be used to send tcp packets on a connection.
194  * Call after host entry created, allocates an mbuf and fills
195  * in a skeletal tcp/ip header, minimizing the amount of work
196  * necessary when the connection is used.
197  *
198  * To support IPv6 in addition to IPv4 and considering that the sizes of
199  * the IPv4 and IPv6 headers are not the same, we now use a separate pointer
200  * for the TCP header.  Also, we made the former tcpiphdr header pointer
201  * into just an IP overlay pointer, with casting as appropriate for v6. rja
202  */
203 struct mbuf *
204 tcp_template(tp)
205 	struct tcpcb *tp;
206 {
207 	register struct inpcb *inp = tp->t_inpcb;
208 	register struct mbuf *m;
209 	register struct tcphdr *th;
210 
211 	if ((m = tp->t_template) == 0) {
212 		m = m_get(M_DONTWAIT, MT_HEADER);
213 		if (m == NULL)
214 			return (0);
215 
216 		switch (tp->pf) {
217 		case 0:	/*default to PF_INET*/
218 #ifdef INET
219 		case AF_INET:
220 			m->m_len = sizeof(struct ip);
221 			break;
222 #endif /* INET */
223 #ifdef INET6
224 		case AF_INET6:
225 			m->m_len = sizeof(struct ip6_hdr);
226 			break;
227 #endif /* INET6 */
228 		}
229 		m->m_len += sizeof (struct tcphdr);
230 
231 		/*
232 		 * The link header, network header, TCP header, and TCP options
233 		 * all must fit in this mbuf. For now, assume the worst case of
234 		 * TCP options size. Eventually, compute this from tp flags.
235 		 */
236 		if (m->m_len + MAX_TCPOPTLEN + max_linkhdr >= MHLEN) {
237 			MCLGET(m, M_DONTWAIT);
238 			if ((m->m_flags & M_EXT) == 0) {
239 				m_free(m);
240 				return (0);
241 			}
242 		}
243 	}
244 
245 	switch(tp->pf) {
246 #ifdef INET
247 	case AF_INET:
248 		{
249 			struct ipovly *ipovly;
250 
251 			ipovly = mtod(m, struct ipovly *);
252 
253 			bzero(ipovly->ih_x1, sizeof ipovly->ih_x1);
254 			ipovly->ih_pr = IPPROTO_TCP;
255 			ipovly->ih_len = htons(sizeof (struct tcphdr));
256 			ipovly->ih_src = inp->inp_laddr;
257 			ipovly->ih_dst = inp->inp_faddr;
258 
259 			th = (struct tcphdr *)(mtod(m, caddr_t) +
260 				sizeof(struct ip));
261 			th->th_sum = in_cksum_phdr(ipovly->ih_src.s_addr,
262 			    ipovly->ih_dst.s_addr,
263 			    htons(sizeof (struct tcphdr) + IPPROTO_TCP));
264 		}
265 		break;
266 #endif /* INET */
267 #ifdef INET6
268 	case AF_INET6:
269 		{
270 			struct ip6_hdr *ip6;
271 
272 			ip6 = mtod(m, struct ip6_hdr *);
273 
274 			ip6->ip6_src = inp->inp_laddr6;
275 			ip6->ip6_dst = inp->inp_faddr6;
276 			ip6->ip6_flow = htonl(0x60000000) |
277 			    (inp->inp_ipv6.ip6_flow & htonl(0x0fffffff));
278 
279 			ip6->ip6_nxt = IPPROTO_TCP;
280 			ip6->ip6_plen = htons(sizeof(struct tcphdr)); /*XXX*/
281 			ip6->ip6_hlim = in6_selecthlim(inp, NULL);	/*XXX*/
282 
283 			th = (struct tcphdr *)(mtod(m, caddr_t) +
284 				sizeof(struct ip6_hdr));
285 			th->th_sum = 0;
286 		}
287 		break;
288 #endif /* INET6 */
289 	}
290 
291 	th->th_sport = inp->inp_lport;
292 	th->th_dport = inp->inp_fport;
293 	th->th_seq = 0;
294 	th->th_ack = 0;
295 	th->th_x2  = 0;
296 	th->th_off = 5;
297 	th->th_flags = 0;
298 	th->th_win = 0;
299 	th->th_urp = 0;
300 	return (m);
301 }
302 
303 /*
304  * Send a single message to the TCP at address specified by
305  * the given TCP/IP header.  If m == 0, then we make a copy
306  * of the tcpiphdr at ti and send directly to the addressed host.
307  * This is used to force keep alive messages out using the TCP
308  * template for a connection tp->t_template.  If flags are given
309  * then we send a message back to the TCP which originated the
310  * segment ti, and discard the mbuf containing it and any other
311  * attached mbufs.
312  *
313  * In any case the ack and sequence number of the transmitted
314  * segment are as specified by the parameters.
315  */
316 #ifdef INET6
317 /* This function looks hairy, because it was so IPv4-dependent. */
318 #endif /* INET6 */
319 void
320 tcp_respond(tp, template, m, ack, seq, flags)
321 	struct tcpcb *tp;
322 	caddr_t template;
323 	register struct mbuf *m;
324 	tcp_seq ack, seq;
325 	int flags;
326 {
327 	register int tlen;
328 	int win = 0;
329 	struct route *ro = 0;
330 	register struct tcphdr *th;
331 	register struct tcpiphdr *ti = (struct tcpiphdr *)template;
332 	int af;		/* af on wire */
333 
334 	if (tp) {
335 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
336 		/*
337 		 * If this is called with an unconnected
338 		 * socket/tp/pcb (tp->pf is 0), we lose.
339 		 */
340 		af = tp->pf;
341 
342 		/*
343 		 * The route/route6 distinction is meaningless
344 		 * unless you're allocating space or passing parameters.
345 		 */
346 		ro = &tp->t_inpcb->inp_route;
347 	} else
348 		af = (((struct ip *)ti)->ip_v == 6) ? AF_INET6 : AF_INET;
349 	if (m == 0) {
350 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
351 		if (m == NULL)
352 			return;
353 #ifdef TCP_COMPAT_42
354 		tlen = 1;
355 #else
356 		tlen = 0;
357 #endif
358 		m->m_data += max_linkhdr;
359 		switch (af) {
360 #ifdef INET6
361 		case AF_INET6:
362 			bcopy(ti, mtod(m, caddr_t), sizeof(struct tcphdr) +
363 			    sizeof(struct ip6_hdr));
364 			break;
365 #endif /* INET6 */
366 		case AF_INET:
367 			bcopy(ti, mtod(m, caddr_t), sizeof(struct tcphdr) +
368 			    sizeof(struct ip));
369 			break;
370 		}
371 
372 		ti = mtod(m, struct tcpiphdr *);
373 		flags = TH_ACK;
374 	} else {
375 		m_freem(m->m_next);
376 		m->m_next = 0;
377 		m->m_data = (caddr_t)ti;
378 		tlen = 0;
379 #define xchg(a,b,type) do { type t; t=a; a=b; b=t; } while (0)
380 		switch (af) {
381 #ifdef INET6
382 		case AF_INET6:
383 			m->m_len = sizeof(struct tcphdr) + sizeof(struct ip6_hdr);
384 			xchg(((struct ip6_hdr *)ti)->ip6_dst,
385 			    ((struct ip6_hdr *)ti)->ip6_src, struct in6_addr);
386 			th = (void *)((caddr_t)ti + sizeof(struct ip6_hdr));
387 			break;
388 #endif /* INET6 */
389 		case AF_INET:
390 			m->m_len = sizeof (struct tcpiphdr);
391 			xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
392 			th = (void *)((caddr_t)ti + sizeof(struct ip));
393 			break;
394 		}
395 		xchg(th->th_dport, th->th_sport, u_int16_t);
396 #undef xchg
397 	}
398 	switch (af) {
399 #ifdef INET6
400 	case AF_INET6:
401 		tlen += sizeof(struct tcphdr) + sizeof(struct ip6_hdr);
402 		th = (struct tcphdr *)((caddr_t)ti + sizeof(struct ip6_hdr));
403 		break;
404 #endif /* INET6 */
405 	case AF_INET:
406 		ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) + tlen));
407 		tlen += sizeof (struct tcpiphdr);
408 		th = (struct tcphdr *)((caddr_t)ti + sizeof(struct ip));
409 		break;
410 	}
411 
412 	m->m_len = tlen;
413 	m->m_pkthdr.len = tlen;
414 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
415 	th->th_seq = htonl(seq);
416 	th->th_ack = htonl(ack);
417 	th->th_x2 = 0;
418 	th->th_off = sizeof (struct tcphdr) >> 2;
419 	th->th_flags = flags;
420 	if (tp)
421 		win >>= tp->rcv_scale;
422 	if (win > TCP_MAXWIN)
423 		win = TCP_MAXWIN;
424 	th->th_win = htons((u_int16_t)win);
425 	th->th_urp = 0;
426 
427 	switch (af) {
428 #ifdef INET6
429 	case AF_INET6:
430 		((struct ip6_hdr *)ti)->ip6_flow   = htonl(0x60000000);
431 		((struct ip6_hdr *)ti)->ip6_nxt  = IPPROTO_TCP;
432 		((struct ip6_hdr *)ti)->ip6_hlim =
433 			in6_selecthlim(tp ? tp->t_inpcb : NULL, NULL);	/*XXX*/
434 		((struct ip6_hdr *)ti)->ip6_plen = tlen - sizeof(struct ip6_hdr);
435 		th->th_sum = 0;
436 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
437 		   sizeof(struct ip6_hdr), ((struct ip6_hdr *)ti)->ip6_plen);
438 		HTONS(((struct ip6_hdr *)ti)->ip6_plen);
439 		ip6_output(m, tp ? tp->t_inpcb->inp_outputopts6 : NULL,
440 		    (struct route_in6 *)ro, 0, NULL, NULL);
441 		break;
442 #endif /* INET6 */
443 	case AF_INET:
444 		bzero(ti->ti_x1, sizeof ti->ti_x1);
445 		ti->ti_len = htons((u_short)tlen - sizeof(struct ip));
446 
447 		/*
448 		 * There's no point deferring to hardware checksum processing
449 		 * here, as we only send a minimal TCP packet whose checksum
450 		 * we need to compute in any case.
451 		 */
452 		th->th_sum = 0;
453 		th->th_sum = in_cksum(m, tlen);
454 		((struct ip *)ti)->ip_len = htons(tlen);
455 		((struct ip *)ti)->ip_ttl = ip_defttl;
456 		ip_output(m, (void *)NULL, ro, ip_mtudisc ? IP_MTUDISC : 0,
457 			(void *)NULL, tp ? tp->t_inpcb : (void *)NULL);
458 	}
459 }
460 
461 /*
462  * Create a new TCP control block, making an
463  * empty reassembly queue and hooking it to the argument
464  * protocol control block.
465  */
466 struct tcpcb *
467 tcp_newtcpcb(struct inpcb *inp)
468 {
469 	struct tcpcb *tp;
470 	int i;
471 
472 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
473 	if (tp == NULL)
474 		return ((struct tcpcb *)0);
475 	bzero((char *) tp, sizeof(struct tcpcb));
476 	LIST_INIT(&tp->segq);
477 	tp->t_maxseg = tcp_mssdflt;
478 	tp->t_maxopd = 0;
479 
480 	TCP_INIT_DELACK(tp);
481 	for (i = 0; i < TCPT_NTIMERS; i++)
482 		TCP_TIMER_INIT(tp, i);
483 
484 #ifdef TCP_SACK
485 	tp->sack_disable = tcp_do_sack ? 0 : 1;
486 #endif
487 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
488 	tp->t_inpcb = inp;
489 	/*
490 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
491 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
492 	 * reasonable initial retransmit time.
493 	 */
494 	tp->t_srtt = TCPTV_SRTTBASE;
495 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
496 	tp->t_rttmin = TCPTV_MIN;
497 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
498 	    TCPTV_MIN, TCPTV_REXMTMAX);
499 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
500 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
501 #ifdef INET6
502 	/* we disallow IPv4 mapped address completely. */
503 	if ((inp->inp_flags & INP_IPV6) == 0)
504 		tp->pf = PF_INET;
505 	else
506 		tp->pf = PF_INET6;
507 #else
508 	tp->pf = PF_INET;
509 #endif
510 
511 #ifdef INET6
512 	if (inp->inp_flags & INP_IPV6)
513 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
514 	else
515 #endif /* INET6 */
516 		inp->inp_ip.ip_ttl = ip_defttl;
517 
518 	inp->inp_ppcb = (caddr_t)tp;
519 	return (tp);
520 }
521 
522 /*
523  * Drop a TCP connection, reporting
524  * the specified error.  If connection is synchronized,
525  * then send a RST to peer.
526  */
527 struct tcpcb *
528 tcp_drop(tp, errno)
529 	register struct tcpcb *tp;
530 	int errno;
531 {
532 	struct socket *so = tp->t_inpcb->inp_socket;
533 
534 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
535 		tp->t_state = TCPS_CLOSED;
536 		(void) tcp_output(tp);
537 		tcpstat.tcps_drops++;
538 	} else
539 		tcpstat.tcps_conndrops++;
540 	if (errno == ETIMEDOUT && tp->t_softerror)
541 		errno = tp->t_softerror;
542 	so->so_error = errno;
543 	return (tcp_close(tp));
544 }
545 
546 /*
547  * Close a TCP control block:
548  *	discard all space held by the tcp
549  *	discard internet protocol block
550  *	wake up any sleepers
551  */
552 struct tcpcb *
553 tcp_close(struct tcpcb *tp)
554 {
555 	struct inpcb *inp = tp->t_inpcb;
556 	struct socket *so = inp->inp_socket;
557 #ifdef TCP_SACK
558 	struct sackhole *p, *q;
559 #endif
560 #ifdef RTV_RTT
561 	register struct rtentry *rt;
562 #ifdef INET6
563 	register int bound_to_specific = 0;  /* I.e. non-default */
564 
565 	/*
566 	 * This code checks the nature of the route for this connection.
567 	 * Normally this is done by two simple checks in the next
568 	 * INET/INET6 ifdef block, but because of two possible lower layers,
569 	 * that check is done here.
570 	 *
571 	 * Perhaps should be doing this only for a RTF_HOST route.
572 	 */
573 	rt = inp->inp_route.ro_rt;  /* Same for route or route6. */
574 	if (tp->pf == PF_INET6) {
575 		if (rt)
576 			bound_to_specific =
577 			    !(IN6_IS_ADDR_UNSPECIFIED(&
578 			    ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr));
579 	} else {
580 		if (rt)
581 			bound_to_specific =
582 			    (((struct sockaddr_in *)rt_key(rt))->
583 			    sin_addr.s_addr != INADDR_ANY);
584 	}
585 #endif /* INET6 */
586 
587 	/*
588 	 * If we sent enough data to get some meaningful characteristics,
589 	 * save them in the routing entry.  'Enough' is arbitrarily
590 	 * defined as the sendpipesize (default 4K) * 16.  This would
591 	 * give us 16 rtt samples assuming we only get one sample per
592 	 * window (the usual case on a long haul net).  16 samples is
593 	 * enough for the srtt filter to converge to within 5% of the correct
594 	 * value; fewer samples and we could save a very bogus rtt.
595 	 *
596 	 * Don't update the default route's characteristics and don't
597 	 * update anything that the user "locked".
598 	 */
599 #ifdef INET6
600 	/*
601 	 * Note that rt and bound_to_specific are set above.
602 	 */
603 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
604 	    rt && bound_to_specific) {
605 #else /* INET6 */
606 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
607 	    (rt = inp->inp_route.ro_rt) &&
608 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
609 #endif /* INET6 */
610 		register u_long i = 0;
611 
612 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
613 			i = tp->t_srtt *
614 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
615 			if (rt->rt_rmx.rmx_rtt && i)
616 				/*
617 				 * filter this update to half the old & half
618 				 * the new values, converting scale.
619 				 * See route.h and tcp_var.h for a
620 				 * description of the scaling constants.
621 				 */
622 				rt->rt_rmx.rmx_rtt =
623 				    (rt->rt_rmx.rmx_rtt + i) / 2;
624 			else
625 				rt->rt_rmx.rmx_rtt = i;
626 		}
627 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
628 			i = tp->t_rttvar *
629 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
630 			if (rt->rt_rmx.rmx_rttvar && i)
631 				rt->rt_rmx.rmx_rttvar =
632 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
633 			else
634 				rt->rt_rmx.rmx_rttvar = i;
635 		}
636 		/*
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 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
644 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
645 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
646 			/*
647 			 * convert the limit from user data bytes to
648 			 * packets then to packet data bytes.
649 			 */
650 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
651 			if (i < 2)
652 				i = 2;
653 #ifdef INET6
654 			if (tp->pf == PF_INET6)
655 				i *= (u_long)(tp->t_maxseg + sizeof (struct tcphdr)
656 				    + sizeof(struct ip6_hdr));
657 			else
658 #endif /* INET6 */
659 				i *= (u_long)(tp->t_maxseg +
660 				    sizeof (struct tcpiphdr));
661 
662 			if (rt->rt_rmx.rmx_ssthresh)
663 				rt->rt_rmx.rmx_ssthresh =
664 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
665 			else
666 				rt->rt_rmx.rmx_ssthresh = i;
667 		}
668 	}
669 #endif /* RTV_RTT */
670 
671 	/* free the reassembly queue, if any */
672 	tcp_freeq(tp);
673 
674 	tcp_canceltimers(tp);
675 	TCP_CLEAR_DELACK(tp);
676 
677 #ifdef TCP_SACK
678 	/* Free SACK holes. */
679 	q = p = tp->snd_holes;
680 	while (p != 0) {
681 		q = p->next;
682 		pool_put(&sackhl_pool, p);
683 		p = q;
684 	}
685 #endif
686 	if (tp->t_template)
687 		(void) m_free(tp->t_template);
688 	pool_put(&tcpcb_pool, tp);
689 	inp->inp_ppcb = 0;
690 	soisdisconnected(so);
691 	in_pcbdetach(inp);
692 	tcpstat.tcps_closed++;
693 	return ((struct tcpcb *)0);
694 }
695 
696 int
697 tcp_freeq(struct tcpcb *tp)
698 {
699 	struct ipqent *qe;
700 	int rv = 0;
701 
702 	while ((qe = LIST_FIRST(&tp->segq)) != NULL) {
703 		LIST_REMOVE(qe, ipqe_q);
704 		m_freem(qe->ipqe_m);
705 		pool_put(&ipqent_pool, qe);
706 		rv = 1;
707 	}
708 	return (rv);
709 }
710 
711 void
712 tcp_drain()
713 {
714 
715 }
716 
717 /*
718  * Compute proper scaling value for receiver window from buffer space
719  */
720 
721 void
722 tcp_rscale(struct tcpcb *tp, u_long hiwat)
723 {
724 	tp->request_r_scale = 0;
725 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
726 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
727 		tp->request_r_scale++;
728 }
729 
730 /*
731  * Notify a tcp user of an asynchronous error;
732  * store error as soft error, but wake up user
733  * (for now, won't do anything until can select for soft error).
734  */
735 void
736 tcp_notify(inp, error)
737 	struct inpcb *inp;
738 	int error;
739 {
740 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
741 	register struct socket *so = inp->inp_socket;
742 
743 	/*
744 	 * Ignore some errors if we are hooked up.
745 	 * If connection hasn't completed, has retransmitted several times,
746 	 * and receives a second error, give up now.  This is better
747 	 * than waiting a long time to establish a connection that
748 	 * can never complete.
749 	 */
750 	if (tp->t_state == TCPS_ESTABLISHED &&
751 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
752 	      error == EHOSTDOWN)) {
753 		return;
754 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
755 	    tp->t_rxtshift > 3 && tp->t_softerror)
756 		so->so_error = error;
757 	else
758 		tp->t_softerror = error;
759 	wakeup((caddr_t) &so->so_timeo);
760 	sorwakeup(so);
761 	sowwakeup(so);
762 }
763 
764 #ifdef INET6
765 void
766 tcp6_ctlinput(cmd, sa, d)
767 	int cmd;
768 	struct sockaddr *sa;
769 	void *d;
770 {
771 	struct tcphdr th;
772 	void (*notify)(struct inpcb *, int) = tcp_notify;
773 	struct ip6_hdr *ip6;
774 	const struct sockaddr_in6 *sa6_src = NULL;
775 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
776 	struct mbuf *m;
777 	int off;
778 	struct {
779 		u_int16_t th_sport;
780 		u_int16_t th_dport;
781 	} *thp;
782 
783 	if (sa->sa_family != AF_INET6 ||
784 	    sa->sa_len != sizeof(struct sockaddr_in6))
785 		return;
786 	if ((unsigned)cmd >= PRC_NCMDS)
787 		return;
788 	else if (cmd == PRC_QUENCH) {
789 		/* XXX there's no PRC_QUENCH in IPv6 */
790 		notify = tcp_quench;
791 	} else if (PRC_IS_REDIRECT(cmd))
792 		notify = in_rtchange, d = NULL;
793 	else if (cmd == PRC_MSGSIZE)
794 		; /* special code is present, see below */
795 	else if (cmd == PRC_HOSTDEAD)
796 		d = NULL;
797 	else if (inet6ctlerrmap[cmd] == 0)
798 		return;
799 
800 	/* if the parameter is from icmp6, decode it. */
801 	if (d != NULL) {
802 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
803 		m = ip6cp->ip6c_m;
804 		ip6 = ip6cp->ip6c_ip6;
805 		off = ip6cp->ip6c_off;
806 		sa6_src = ip6cp->ip6c_src;
807 	} else {
808 		m = NULL;
809 		ip6 = NULL;
810 		sa6_src = &sa6_any;
811 	}
812 
813 	if (ip6) {
814 		/*
815 		 * XXX: We assume that when ip6 is non NULL,
816 		 * M and OFF are valid.
817 		 */
818 
819 		/* check if we can safely examine src and dst ports */
820 		if (m->m_pkthdr.len < off + sizeof(*thp))
821 			return;
822 
823 		bzero(&th, sizeof(th));
824 #ifdef DIAGNOSTIC
825 		if (sizeof(*thp) > sizeof(th))
826 			panic("assumption failed in tcp6_ctlinput");
827 #endif
828 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
829 
830 		if (cmd == PRC_MSGSIZE) {
831 			int valid = 0;
832 
833 			/*
834 			 * Check to see if we have a valid TCP connection
835 			 * corresponding to the address in the ICMPv6 message
836 			 * payload.
837 			 */
838 			if (in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
839 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
840 			    th.th_sport))
841 				valid++;
842 			else if (in_pcblookup(&tcbtable, &sa6->sin6_addr,
843 			    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
844 			    th.th_sport, INPLOOKUP_IPV6))
845 				valid++;
846 
847 			/*
848 			 * Depending on the value of "valid" and routing table
849 			 * size (mtudisc_{hi,lo}wat), we will:
850 			 * - recalcurate the new MTU and create the
851 			 *   corresponding routing entry, or
852 			 * - ignore the MTU change notification.
853 			 */
854 			icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
855 
856 			return;
857 		}
858 
859 		(void) in6_pcbnotify(&tcbtable, sa, th.th_dport,
860 		    (struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
861 	} else {
862 		(void) in6_pcbnotify(&tcbtable, sa, 0,
863 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
864 	}
865 }
866 #endif
867 
868 void *
869 tcp_ctlinput(cmd, sa, v)
870 	int cmd;
871 	struct sockaddr *sa;
872 	register void *v;
873 {
874 	register struct ip *ip = v;
875 	register struct tcphdr *th;
876 	extern int inetctlerrmap[];
877 	void (*notify)(struct inpcb *, int) = tcp_notify;
878 	int errno;
879 
880 	if (sa->sa_family != AF_INET)
881 		return NULL;
882 
883 	if ((unsigned)cmd >= PRC_NCMDS)
884 		return NULL;
885 	errno = inetctlerrmap[cmd];
886 	if (cmd == PRC_QUENCH)
887 		notify = tcp_quench;
888 	else if (PRC_IS_REDIRECT(cmd))
889 		notify = in_rtchange, ip = 0;
890 	else if (cmd == PRC_MSGSIZE && ip_mtudisc) {
891 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
892 		/*
893 		 * Verify that the packet in the icmp payload refers
894 		 * to an existing TCP connection.
895 		 */
896 		if (in_pcblookup(&tcbtable,
897 				 &ip->ip_dst, th->th_dport,
898 				 &ip->ip_src, th->th_sport,
899 				 INPLOOKUP_WILDCARD)) {
900 			struct icmp *icp;
901 			icp = (struct icmp *)((caddr_t)ip -
902 					      offsetof(struct icmp, icmp_ip));
903 
904 			/* Calculate new mtu and create corresponding route */
905 			icmp_mtudisc(icp);
906 		}
907 		notify = tcp_mtudisc, ip = 0;
908 	} else if (cmd == PRC_MTUINC)
909 		notify = tcp_mtudisc_increase, ip = 0;
910 	else if (cmd == PRC_HOSTDEAD)
911 		ip = 0;
912 	else if (errno == 0)
913 		return NULL;
914 
915 	if (ip) {
916 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
917 		in_pcbnotify(&tcbtable, sa, th->th_dport, ip->ip_src,
918 			     th->th_sport, errno, notify);
919 	} else
920 		in_pcbnotifyall(&tcbtable, sa, errno, notify);
921 
922 	return NULL;
923 }
924 
925 /*
926  * When a source quench is received, close congestion window
927  * to one segment.  We will gradually open it again as we proceed.
928  */
929 void
930 tcp_quench(inp, errno)
931 	struct inpcb *inp;
932 	int errno;
933 {
934 	struct tcpcb *tp = intotcpcb(inp);
935 
936 	if (tp)
937 		tp->snd_cwnd = tp->t_maxseg;
938 }
939 
940 #ifdef INET6
941 /*
942  * Path MTU Discovery handlers.
943  */
944 void
945 tcp6_mtudisc_callback(faddr)
946 	struct in6_addr *faddr;
947 {
948 	struct sockaddr_in6 sin6;
949 
950 	bzero(&sin6, sizeof(sin6));
951 	sin6.sin6_family = AF_INET6;
952 	sin6.sin6_len = sizeof(struct sockaddr_in6);
953 	sin6.sin6_addr = *faddr;
954 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
955 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp_mtudisc);
956 }
957 #endif /* INET6 */
958 
959 /*
960  * On receipt of path MTU corrections, flush old route and replace it
961  * with the new one.  Retransmit all unacknowledged packets, to ensure
962  * that all packets will be received.
963  */
964 void
965 tcp_mtudisc(inp, errno)
966 	struct inpcb *inp;
967 	int errno;
968 {
969 	struct tcpcb *tp = intotcpcb(inp);
970 	struct rtentry *rt = in_pcbrtentry(inp);
971 
972 	if (tp != 0) {
973 		if (rt != 0) {
974 			/*
975 			 * If this was not a host route, remove and realloc.
976 			 */
977 			if ((rt->rt_flags & RTF_HOST) == 0) {
978 				in_rtchange(inp, errno);
979 				if ((rt = in_pcbrtentry(inp)) == 0)
980 					return;
981 			}
982 
983 			if (rt->rt_rmx.rmx_mtu != 0) {
984 				/* also takes care of congestion window */
985 				tcp_mss(tp, -1);
986 			}
987 		}
988 
989 		/*
990 		 * Resend unacknowledged packets.
991 		 */
992 		tp->snd_nxt = tp->snd_una;
993 		tcp_output(tp);
994 	}
995 }
996 
997 void
998 tcp_mtudisc_increase(inp, errno)
999 	struct inpcb *inp;
1000 	int errno;
1001 {
1002 	struct tcpcb *tp = intotcpcb(inp);
1003 	struct rtentry *rt = in_pcbrtentry(inp);
1004 
1005 	if (tp != 0 && rt != 0) {
1006 		/*
1007 		 * If this was a host route, remove and realloc.
1008 		 */
1009 		if (rt->rt_flags & RTF_HOST)
1010 			in_rtchange(inp, errno);
1011 
1012 		/* also takes care of congestion window */
1013 		tcp_mss(tp, -1);
1014 	}
1015 }
1016 
1017 #ifdef TCP_SIGNATURE
1018 int
1019 tcp_signature_tdb_attach()
1020 {
1021 	return (0);
1022 }
1023 
1024 int
1025 tcp_signature_tdb_init(tdbp, xsp, ii)
1026 	struct tdb *tdbp;
1027 	struct xformsw *xsp;
1028 	struct ipsecinit *ii;
1029 {
1030 	char *c;
1031 #define isdigit(c)	  (((c) >= '0') && ((c) <= '9'))
1032 #define isalpha(c)	( (((c) >= 'A') && ((c) <= 'Z')) || \
1033 			  (((c) >= 'a') && ((c) <= 'z')) )
1034 
1035 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
1036 		return (EINVAL);
1037 
1038 	c = (char *)ii->ii_authkey;
1039 
1040 	while (c < (char *)ii->ii_authkey + ii->ii_authkeylen - 1) {
1041 		if (isdigit(*c)) {
1042 			if (*(c + 1) == ' ')
1043 				return (EINVAL);
1044 		} else {
1045 			if (!isalpha(*c))
1046 				return (EINVAL);
1047 		}
1048 
1049 		c++;
1050 	}
1051 
1052 	if (!isdigit(*c) && !isalpha(*c))
1053 		return (EINVAL);
1054 
1055 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_DONTWAIT);
1056 	if (tdbp->tdb_amxkey == NULL)
1057 		return (ENOMEM);
1058 	bcopy(ii->ii_authkey, tdbp->tdb_amxkey, ii->ii_authkeylen);
1059 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
1060 
1061 	return (0);
1062 }
1063 
1064 int
1065 tcp_signature_tdb_zeroize(tdbp)
1066 	struct tdb *tdbp;
1067 {
1068 	if (tdbp->tdb_amxkey) {
1069 		bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
1070 		free(tdbp->tdb_amxkey, M_XDATA);
1071 		tdbp->tdb_amxkey = NULL;
1072 	}
1073 
1074 	return (0);
1075 }
1076 
1077 int
1078 tcp_signature_tdb_input(m, tdbp, skip, protoff)
1079 	struct mbuf *m;
1080 	struct tdb *tdbp;
1081 	int skip, protoff;
1082 {
1083 	return (0);
1084 }
1085 
1086 int
1087 tcp_signature_tdb_output(m, tdbp, mp, skip, protoff)
1088 	struct mbuf *m;
1089 	struct tdb *tdbp;
1090 	struct mbuf **mp;
1091 	int skip, protoff;
1092 {
1093 	return (EINVAL);
1094 }
1095 
1096 int
1097 tcp_signature_apply(fstate, data, len)
1098 	caddr_t fstate;
1099 	caddr_t data;
1100 	unsigned int len;
1101 {
1102 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
1103 	return 0;
1104 }
1105 #endif /* TCP_SIGNATURE */
1106 
1107 #define TCP_RNDISS_ROUNDS	16
1108 #define TCP_RNDISS_OUT	7200
1109 #define TCP_RNDISS_MAX	30000
1110 
1111 u_int8_t tcp_rndiss_sbox[128];
1112 u_int16_t tcp_rndiss_msb;
1113 u_int16_t tcp_rndiss_cnt;
1114 long tcp_rndiss_reseed;
1115 
1116 u_int16_t
1117 tcp_rndiss_encrypt(val)
1118 	u_int16_t val;
1119 {
1120 	u_int16_t sum = 0, i;
1121 
1122 	for (i = 0; i < TCP_RNDISS_ROUNDS; i++) {
1123 		sum += 0x79b9;
1124 		val ^= ((u_int16_t)tcp_rndiss_sbox[(val^sum) & 0x7f]) << 7;
1125 		val = ((val & 0xff) << 7) | (val >> 8);
1126 	}
1127 
1128 	return val;
1129 }
1130 
1131 void
1132 tcp_rndiss_init()
1133 {
1134 	get_random_bytes(tcp_rndiss_sbox, sizeof(tcp_rndiss_sbox));
1135 
1136 	tcp_rndiss_reseed = time.tv_sec + TCP_RNDISS_OUT;
1137 	tcp_rndiss_msb = tcp_rndiss_msb == 0x8000 ? 0 : 0x8000;
1138 	tcp_rndiss_cnt = 0;
1139 }
1140 
1141 tcp_seq
1142 tcp_rndiss_next()
1143 {
1144         if (tcp_rndiss_cnt >= TCP_RNDISS_MAX ||
1145 	    time.tv_sec > tcp_rndiss_reseed)
1146                 tcp_rndiss_init();
1147 
1148 	/* (arc4random() & 0x7fff) ensures a 32768 byte gap between ISS */
1149 	return ((tcp_rndiss_encrypt(tcp_rndiss_cnt++) | tcp_rndiss_msb) <<16) |
1150 		(arc4random() & 0x7fff);
1151 }
1152 
1153