xref: /openbsd-src/sys/netinet/tcp_subr.c (revision ae3cb403620ab940fbaabb3055fac045a63d56b7)
1 /*	$OpenBSD: tcp_subr.c,v 1.167 2017/12/07 16:52:21 mikeb 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/mbuf.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/timeout.h>
77 #include <sys/protosw.h>
78 #include <sys/kernel.h>
79 #include <sys/pool.h>
80 
81 #include <net/route.h>
82 
83 #include <netinet/in.h>
84 #include <netinet/ip.h>
85 #include <netinet/in_pcb.h>
86 #include <netinet/ip_var.h>
87 #include <netinet/ip_icmp.h>
88 #include <netinet/tcp.h>
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 
94 #ifdef INET6
95 #include <netinet6/ip6protosw.h>
96 #endif /* INET6 */
97 
98 #include <crypto/md5.h>
99 #include <crypto/sha2.h>
100 
101 /* patchable/settable parameters for tcp */
102 int	tcp_mssdflt = TCP_MSS;
103 int	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
104 
105 /* values controllable via sysctl */
106 int	tcp_do_rfc1323 = 1;
107 int	tcp_do_sack = 1;	/* RFC 2018 selective ACKs */
108 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
109 #ifdef TCP_ECN
110 int	tcp_do_ecn = 0;		/* RFC3168 ECN enabled/disabled? */
111 #endif
112 int	tcp_do_rfc3390 = 2;	/* Increase TCP's Initial Window to 10*mss */
113 
114 u_int32_t	tcp_now = 1;
115 
116 #ifndef TCB_INITIAL_HASH_SIZE
117 #define	TCB_INITIAL_HASH_SIZE	128
118 #endif
119 
120 int tcp_reass_limit = NMBCLUSTERS / 8; /* hardlimit for tcpqe_pool */
121 int tcp_sackhole_limit = 32*1024; /* hardlimit for sackhl_pool */
122 
123 struct pool tcpcb_pool;
124 struct pool tcpqe_pool;
125 struct pool sackhl_pool;
126 
127 struct cpumem *tcpcounters;		/* tcp statistics */
128 
129 u_char   tcp_secret[16];
130 SHA2_CTX tcp_secret_ctx;
131 tcp_seq  tcp_iss;
132 
133 /*
134  * Tcp initialization
135  */
136 void
137 tcp_init(void)
138 {
139 	tcp_iss = 1;		/* wrong */
140 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, IPL_SOFTNET, 0,
141 	    "tcpcb", NULL);
142 	pool_init(&tcpqe_pool, sizeof(struct tcpqent), 0, IPL_SOFTNET, 0,
143 	    "tcpqe", NULL);
144 	pool_sethardlimit(&tcpqe_pool, tcp_reass_limit, NULL, 0);
145 	pool_init(&sackhl_pool, sizeof(struct sackhole), 0, IPL_SOFTNET, 0,
146 	    "sackhl", NULL);
147 	pool_sethardlimit(&sackhl_pool, tcp_sackhole_limit, NULL, 0);
148 	in_pcbinit(&tcbtable, TCB_INITIAL_HASH_SIZE);
149 	tcpcounters = counters_alloc(tcps_ncounters);
150 
151 	arc4random_buf(tcp_secret, sizeof(tcp_secret));
152 	SHA512Init(&tcp_secret_ctx);
153 	SHA512Update(&tcp_secret_ctx, tcp_secret, sizeof(tcp_secret));
154 
155 #ifdef INET6
156 	/*
157 	 * Since sizeof(struct ip6_hdr) > sizeof(struct ip), we
158 	 * do max length checks/computations only on the former.
159 	 */
160 	if (max_protohdr < (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)))
161 		max_protohdr = (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
162 	if ((max_linkhdr + sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) >
163 	    MHLEN)
164 		panic("tcp_init");
165 
166 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
167 #endif /* INET6 */
168 
169 	/* Initialize the compressed state engine. */
170 	syn_cache_init();
171 
172 	/* Initialize timer state. */
173 	tcp_timer_init();
174 }
175 
176 /*
177  * Create template to be used to send tcp packets on a connection.
178  * Call after host entry created, allocates an mbuf and fills
179  * in a skeletal tcp/ip header, minimizing the amount of work
180  * necessary when the connection is used.
181  *
182  * To support IPv6 in addition to IPv4 and considering that the sizes of
183  * the IPv4 and IPv6 headers are not the same, we now use a separate pointer
184  * for the TCP header.  Also, we made the former tcpiphdr header pointer
185  * into just an IP overlay pointer, with casting as appropriate for v6. rja
186  */
187 struct mbuf *
188 tcp_template(struct tcpcb *tp)
189 {
190 	struct inpcb *inp = tp->t_inpcb;
191 	struct mbuf *m;
192 	struct tcphdr *th;
193 
194 	if ((m = tp->t_template) == 0) {
195 		m = m_get(M_DONTWAIT, MT_HEADER);
196 		if (m == NULL)
197 			return (0);
198 
199 		switch (tp->pf) {
200 		case 0:	/*default to PF_INET*/
201 		case AF_INET:
202 			m->m_len = sizeof(struct ip);
203 			break;
204 #ifdef INET6
205 		case AF_INET6:
206 			m->m_len = sizeof(struct ip6_hdr);
207 			break;
208 #endif /* INET6 */
209 		}
210 		m->m_len += sizeof (struct tcphdr);
211 
212 		/*
213 		 * The link header, network header, TCP header, and TCP options
214 		 * all must fit in this mbuf. For now, assume the worst case of
215 		 * TCP options size. Eventually, compute this from tp flags.
216 		 */
217 		if (m->m_len + MAX_TCPOPTLEN + max_linkhdr >= MHLEN) {
218 			MCLGET(m, M_DONTWAIT);
219 			if ((m->m_flags & M_EXT) == 0) {
220 				m_free(m);
221 				return (0);
222 			}
223 		}
224 	}
225 
226 	switch(tp->pf) {
227 	case AF_INET:
228 		{
229 			struct ipovly *ipovly;
230 
231 			ipovly = mtod(m, struct ipovly *);
232 
233 			bzero(ipovly->ih_x1, sizeof ipovly->ih_x1);
234 			ipovly->ih_pr = IPPROTO_TCP;
235 			ipovly->ih_len = htons(sizeof (struct tcphdr));
236 			ipovly->ih_src = inp->inp_laddr;
237 			ipovly->ih_dst = inp->inp_faddr;
238 
239 			th = (struct tcphdr *)(mtod(m, caddr_t) +
240 				sizeof(struct ip));
241 		}
242 		break;
243 #ifdef INET6
244 	case AF_INET6:
245 		{
246 			struct ip6_hdr *ip6;
247 
248 			ip6 = mtod(m, struct ip6_hdr *);
249 
250 			ip6->ip6_src = inp->inp_laddr6;
251 			ip6->ip6_dst = inp->inp_faddr6;
252 			ip6->ip6_flow = htonl(0x60000000) |
253 			    (inp->inp_flowinfo & IPV6_FLOWLABEL_MASK);
254 
255 			ip6->ip6_nxt = IPPROTO_TCP;
256 			ip6->ip6_plen = htons(sizeof(struct tcphdr)); /*XXX*/
257 			ip6->ip6_hlim = in6_selecthlim(inp);	/*XXX*/
258 
259 			th = (struct tcphdr *)(mtod(m, caddr_t) +
260 				sizeof(struct ip6_hdr));
261 		}
262 		break;
263 #endif /* INET6 */
264 	}
265 
266 	th->th_sport = inp->inp_lport;
267 	th->th_dport = inp->inp_fport;
268 	th->th_seq = 0;
269 	th->th_ack = 0;
270 	th->th_x2  = 0;
271 	th->th_off = 5;
272 	th->th_flags = 0;
273 	th->th_win = 0;
274 	th->th_urp = 0;
275 	th->th_sum = 0;
276 	return (m);
277 }
278 
279 /*
280  * Send a single message to the TCP at address specified by
281  * the given TCP/IP header.  If m == 0, then we make a copy
282  * of the tcpiphdr at ti and send directly to the addressed host.
283  * This is used to force keep alive messages out using the TCP
284  * template for a connection tp->t_template.  If flags are given
285  * then we send a message back to the TCP which originated the
286  * segment ti, and discard the mbuf containing it and any other
287  * attached mbufs.
288  *
289  * In any case the ack and sequence number of the transmitted
290  * segment are as specified by the parameters.
291  */
292 void
293 tcp_respond(struct tcpcb *tp, caddr_t template, struct tcphdr *th0,
294     tcp_seq ack, tcp_seq seq, int flags, u_int rtableid)
295 {
296 	int tlen;
297 	int win = 0;
298 	struct mbuf *m = NULL;
299 	struct tcphdr *th;
300 	struct ip *ip;
301 #ifdef INET6
302 	struct ip6_hdr *ip6;
303 #endif
304 	int af;		/* af on wire */
305 
306 	if (tp) {
307 		struct socket *so = tp->t_inpcb->inp_socket;
308 		win = sbspace(so, &so->so_rcv);
309 		/*
310 		 * If this is called with an unconnected
311 		 * socket/tp/pcb (tp->pf is 0), we lose.
312 		 */
313 		af = tp->pf;
314 	} else
315 		af = (((struct ip *)template)->ip_v == 6) ? AF_INET6 : AF_INET;
316 
317 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
318 	if (m == NULL)
319 		return;
320 	m->m_data += max_linkhdr;
321 	tlen = 0;
322 
323 #define xchg(a,b,type) do { type t; t=a; a=b; b=t; } while (0)
324 	switch (af) {
325 #ifdef INET6
326 	case AF_INET6:
327 		ip6 = mtod(m, struct ip6_hdr *);
328 		th = (struct tcphdr *)(ip6 + 1);
329 		tlen = sizeof(*ip6) + sizeof(*th);
330 		if (th0) {
331 			bcopy(template, ip6, sizeof(*ip6));
332 			bcopy(th0, th, sizeof(*th));
333 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
334 		} else {
335 			bcopy(template, ip6, tlen);
336 		}
337 		break;
338 #endif /* INET6 */
339 	case AF_INET:
340 		ip = mtod(m, struct ip *);
341 		th = (struct tcphdr *)(ip + 1);
342 		tlen = sizeof(*ip) + sizeof(*th);
343 		if (th0) {
344 			bcopy(template, ip, sizeof(*ip));
345 			bcopy(th0, th, sizeof(*th));
346 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, u_int32_t);
347 		} else {
348 			bcopy(template, ip, tlen);
349 		}
350 		break;
351 	}
352 	if (th0)
353 		xchg(th->th_dport, th->th_sport, u_int16_t);
354 	else
355 		flags = TH_ACK;
356 #undef xchg
357 
358 	th->th_seq = htonl(seq);
359 	th->th_ack = htonl(ack);
360 	th->th_x2 = 0;
361 	th->th_off = sizeof (struct tcphdr) >> 2;
362 	th->th_flags = flags;
363 	if (tp)
364 		win >>= tp->rcv_scale;
365 	if (win > TCP_MAXWIN)
366 		win = TCP_MAXWIN;
367 	th->th_win = htons((u_int16_t)win);
368 	th->th_urp = 0;
369 
370 	if (tp && (tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
371 	    (flags & TH_RST) == 0 && (tp->t_flags & TF_RCVD_TSTMP)) {
372 		u_int32_t *lp = (u_int32_t *)(th + 1);
373 		/* Form timestamp option as shown in appendix A of RFC 1323. */
374 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
375 		*lp++ = htonl(tcp_now + tp->ts_modulate);
376 		*lp   = htonl(tp->ts_recent);
377 		tlen += TCPOLEN_TSTAMP_APPA;
378 		th->th_off = (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_APPA) >> 2;
379 	}
380 
381 	m->m_len = tlen;
382 	m->m_pkthdr.len = tlen;
383 	m->m_pkthdr.ph_ifidx = 0;
384 	m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT;
385 
386 	/* force routing table */
387 	if (tp)
388 		m->m_pkthdr.ph_rtableid = tp->t_inpcb->inp_rtableid;
389 	else
390 		m->m_pkthdr.ph_rtableid = rtableid;
391 
392 	switch (af) {
393 #ifdef INET6
394 	case AF_INET6:
395 		ip6->ip6_flow = htonl(0x60000000);
396 		ip6->ip6_nxt  = IPPROTO_TCP;
397 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL);	/*XXX*/
398 		ip6->ip6_plen = tlen - sizeof(struct ip6_hdr);
399 		ip6->ip6_plen = htons(ip6->ip6_plen);
400 		ip6_output(m, tp ? tp->t_inpcb->inp_outputopts6 : NULL,
401 		    tp ? &tp->t_inpcb->inp_route6 : NULL,
402 		    0, NULL,
403 		    tp ? tp->t_inpcb : NULL);
404 		break;
405 #endif /* INET6 */
406 	case AF_INET:
407 		ip->ip_len = htons(tlen);
408 		ip->ip_ttl = ip_defttl;
409 		ip->ip_tos = 0;
410 		ip_output(m, NULL,
411 		    tp ? &tp->t_inpcb->inp_route : NULL,
412 		    ip_mtudisc ? IP_MTUDISC : 0, NULL,
413 		    tp ? tp->t_inpcb : NULL, 0);
414 		break;
415 	}
416 }
417 
418 /*
419  * Create a new TCP control block, making an
420  * empty reassembly queue and hooking it to the argument
421  * protocol control block.
422  */
423 struct tcpcb *
424 tcp_newtcpcb(struct inpcb *inp)
425 {
426 	struct tcpcb *tp;
427 	int i;
428 
429 	tp = pool_get(&tcpcb_pool, PR_NOWAIT|PR_ZERO);
430 	if (tp == NULL)
431 		return (NULL);
432 	TAILQ_INIT(&tp->t_segq);
433 	tp->t_maxseg = tcp_mssdflt;
434 	tp->t_maxopd = 0;
435 
436 	TCP_INIT_DELACK(tp);
437 	for (i = 0; i < TCPT_NTIMERS; i++)
438 		TCP_TIMER_INIT(tp, i);
439 	timeout_set(&tp->t_reap_to, tcp_reaper, tp);
440 
441 	tp->sack_enable = tcp_do_sack;
442 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
443 	tp->t_inpcb = inp;
444 	/*
445 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
446 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
447 	 * reasonable initial retransmit time.
448 	 */
449 	tp->t_srtt = TCPTV_SRTTBASE;
450 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ <<
451 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
452 	tp->t_rttmin = TCPTV_MIN;
453 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
454 	    TCPTV_MIN, TCPTV_REXMTMAX);
455 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
456 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
457 
458 	tp->t_pmtud_mtu_sent = 0;
459 	tp->t_pmtud_mss_acked = 0;
460 
461 #ifdef INET6
462 	/* we disallow IPv4 mapped address completely. */
463 	if ((inp->inp_flags & INP_IPV6) == 0)
464 		tp->pf = PF_INET;
465 	else
466 		tp->pf = PF_INET6;
467 #else
468 	tp->pf = PF_INET;
469 #endif
470 
471 #ifdef INET6
472 	if (inp->inp_flags & INP_IPV6)
473 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
474 	else
475 #endif /* INET6 */
476 		inp->inp_ip.ip_ttl = ip_defttl;
477 
478 	inp->inp_ppcb = (caddr_t)tp;
479 	return (tp);
480 }
481 
482 /*
483  * Drop a TCP connection, reporting
484  * the specified error.  If connection is synchronized,
485  * then send a RST to peer.
486  */
487 struct tcpcb *
488 tcp_drop(struct tcpcb *tp, int errno)
489 {
490 	struct socket *so = tp->t_inpcb->inp_socket;
491 
492 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
493 		tp->t_state = TCPS_CLOSED;
494 		(void) tcp_output(tp);
495 		tcpstat_inc(tcps_drops);
496 	} else
497 		tcpstat_inc(tcps_conndrops);
498 	if (errno == ETIMEDOUT && tp->t_softerror)
499 		errno = tp->t_softerror;
500 	so->so_error = errno;
501 	return (tcp_close(tp));
502 }
503 
504 /*
505  * Close a TCP control block:
506  *	discard all space held by the tcp
507  *	discard internet protocol block
508  *	wake up any sleepers
509  */
510 struct tcpcb *
511 tcp_close(struct tcpcb *tp)
512 {
513 	struct inpcb *inp = tp->t_inpcb;
514 	struct socket *so = inp->inp_socket;
515 	struct sackhole *p, *q;
516 
517 	/* free the reassembly queue, if any */
518 	tcp_freeq(tp);
519 
520 	tcp_canceltimers(tp);
521 	TCP_CLEAR_DELACK(tp);
522 	syn_cache_cleanup(tp);
523 
524 	/* Free SACK holes. */
525 	q = p = tp->snd_holes;
526 	while (p != 0) {
527 		q = p->next;
528 		pool_put(&sackhl_pool, p);
529 		p = q;
530 	}
531 
532 	m_free(tp->t_template);
533 
534 	tp->t_flags |= TF_DEAD;
535 	timeout_add(&tp->t_reap_to, 0);
536 
537 	inp->inp_ppcb = 0;
538 	soisdisconnected(so);
539 	in_pcbdetach(inp);
540 	return (NULL);
541 }
542 
543 void
544 tcp_reaper(void *arg)
545 {
546 	struct tcpcb *tp = arg;
547 
548 	pool_put(&tcpcb_pool, tp);
549 	tcpstat_inc(tcps_closed);
550 }
551 
552 int
553 tcp_freeq(struct tcpcb *tp)
554 {
555 	struct tcpqent *qe;
556 	int rv = 0;
557 
558 	while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) {
559 		TAILQ_REMOVE(&tp->t_segq, qe, tcpqe_q);
560 		m_freem(qe->tcpqe_m);
561 		pool_put(&tcpqe_pool, qe);
562 		rv = 1;
563 	}
564 	return (rv);
565 }
566 
567 /*
568  * Compute proper scaling value for receiver window from buffer space
569  */
570 
571 void
572 tcp_rscale(struct tcpcb *tp, u_long hiwat)
573 {
574 	tp->request_r_scale = 0;
575 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
576 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
577 		tp->request_r_scale++;
578 }
579 
580 /*
581  * Notify a tcp user of an asynchronous error;
582  * store error as soft error, but wake up user
583  * (for now, won't do anything until can select for soft error).
584  */
585 void
586 tcp_notify(struct inpcb *inp, int error)
587 {
588 	struct tcpcb *tp = intotcpcb(inp);
589 	struct socket *so = inp->inp_socket;
590 
591 	/*
592 	 * Ignore some errors if we are hooked up.
593 	 * If connection hasn't completed, has retransmitted several times,
594 	 * and receives a second error, give up now.  This is better
595 	 * than waiting a long time to establish a connection that
596 	 * can never complete.
597 	 */
598 	if (tp->t_state == TCPS_ESTABLISHED &&
599 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
600 	      error == EHOSTDOWN)) {
601 		return;
602 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
603 	    tp->t_rxtshift > 3 && tp->t_softerror)
604 		so->so_error = error;
605 	else
606 		tp->t_softerror = error;
607 	wakeup((caddr_t) &so->so_timeo);
608 	sorwakeup(so);
609 	sowwakeup(so);
610 }
611 
612 #ifdef INET6
613 void
614 tcp6_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *d)
615 {
616 	struct tcphdr th;
617 	struct tcpcb *tp;
618 	void (*notify)(struct inpcb *, int) = tcp_notify;
619 	struct ip6_hdr *ip6;
620 	const struct sockaddr_in6 *sa6_src = NULL;
621 	struct sockaddr_in6 *sa6 = satosin6(sa);
622 	struct inpcb *inp;
623 	struct mbuf *m;
624 	tcp_seq seq;
625 	int off;
626 	struct {
627 		u_int16_t th_sport;
628 		u_int16_t th_dport;
629 		u_int32_t th_seq;
630 	} *thp;
631 
632 	CTASSERT(sizeof(*thp) <= sizeof(th));
633 	if (sa->sa_family != AF_INET6 ||
634 	    sa->sa_len != sizeof(struct sockaddr_in6) ||
635 	    IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) ||
636 	    IN6_IS_ADDR_V4MAPPED(&sa6->sin6_addr))
637 		return;
638 	if ((unsigned)cmd >= PRC_NCMDS)
639 		return;
640 	else if (cmd == PRC_QUENCH) {
641 		/*
642 		 * Don't honor ICMP Source Quench messages meant for
643 		 * TCP connections.
644 		 */
645 		/* XXX there's no PRC_QUENCH in IPv6 */
646 		return;
647 	} else if (PRC_IS_REDIRECT(cmd))
648 		notify = in_rtchange, d = NULL;
649 	else if (cmd == PRC_MSGSIZE)
650 		; /* special code is present, see below */
651 	else if (cmd == PRC_HOSTDEAD)
652 		d = NULL;
653 	else if (inet6ctlerrmap[cmd] == 0)
654 		return;
655 
656 	/* if the parameter is from icmp6, decode it. */
657 	if (d != NULL) {
658 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
659 		m = ip6cp->ip6c_m;
660 		ip6 = ip6cp->ip6c_ip6;
661 		off = ip6cp->ip6c_off;
662 		sa6_src = ip6cp->ip6c_src;
663 	} else {
664 		m = NULL;
665 		ip6 = NULL;
666 		sa6_src = &sa6_any;
667 	}
668 
669 	if (ip6) {
670 		/*
671 		 * XXX: We assume that when ip6 is non NULL,
672 		 * M and OFF are valid.
673 		 */
674 
675 		/* check if we can safely examine src and dst ports */
676 		if (m->m_pkthdr.len < off + sizeof(*thp))
677 			return;
678 
679 		bzero(&th, sizeof(th));
680 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
681 
682 		/*
683 		 * Check to see if we have a valid TCP connection
684 		 * corresponding to the address in the ICMPv6 message
685 		 * payload.
686 		 */
687 		inp = in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
688 		    th.th_dport, &sa6_src->sin6_addr, th.th_sport, rdomain);
689 		if (cmd == PRC_MSGSIZE) {
690 			/*
691 			 * Depending on the value of "valid" and routing table
692 			 * size (mtudisc_{hi,lo}wat), we will:
693 			 * - recalcurate the new MTU and create the
694 			 *   corresponding routing entry, or
695 			 * - ignore the MTU change notification.
696 			 */
697 			icmp6_mtudisc_update((struct ip6ctlparam *)d, inp != NULL);
698 			return;
699 		}
700 		if (inp) {
701 			seq = ntohl(th.th_seq);
702 			if (inp->inp_socket &&
703 			    (tp = intotcpcb(inp)) &&
704 			    SEQ_GEQ(seq, tp->snd_una) &&
705 			    SEQ_LT(seq, tp->snd_max))
706 				notify(inp, inet6ctlerrmap[cmd]);
707 		} else if (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
708 		    inet6ctlerrmap[cmd] == ENETUNREACH ||
709 		    inet6ctlerrmap[cmd] == EHOSTDOWN)
710 			syn_cache_unreach((struct sockaddr *)sa6_src,
711 			    sa, &th, rdomain);
712 	} else {
713 		(void) in6_pcbnotify(&tcbtable, sa6, 0,
714 		    sa6_src, 0, rdomain, cmd, NULL, notify);
715 	}
716 }
717 #endif
718 
719 void
720 tcp_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
721 {
722 	struct ip *ip = v;
723 	struct tcphdr *th;
724 	struct tcpcb *tp;
725 	struct inpcb *inp;
726 	struct in_addr faddr;
727 	tcp_seq seq;
728 	u_int mtu;
729 	void (*notify)(struct inpcb *, int) = tcp_notify;
730 	int errno;
731 
732 	if (sa->sa_family != AF_INET)
733 		return;
734 	faddr = satosin(sa)->sin_addr;
735 	if (faddr.s_addr == INADDR_ANY)
736 		return;
737 
738 	if ((unsigned)cmd >= PRC_NCMDS)
739 		return;
740 	errno = inetctlerrmap[cmd];
741 	if (cmd == PRC_QUENCH)
742 		/*
743 		 * Don't honor ICMP Source Quench messages meant for
744 		 * TCP connections.
745 		 */
746 		return;
747 	else if (PRC_IS_REDIRECT(cmd))
748 		notify = in_rtchange, ip = 0;
749 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip) {
750 		/*
751 		 * Verify that the packet in the icmp payload refers
752 		 * to an existing TCP connection.
753 		 */
754 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
755 		seq = ntohl(th->th_seq);
756 		inp = in_pcbhashlookup(&tcbtable,
757 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
758 		    rdomain);
759 		if (inp && (tp = intotcpcb(inp)) &&
760 		    SEQ_GEQ(seq, tp->snd_una) &&
761 		    SEQ_LT(seq, tp->snd_max)) {
762 			struct icmp *icp;
763 			icp = (struct icmp *)((caddr_t)ip -
764 					      offsetof(struct icmp, icmp_ip));
765 
766 			/*
767 			 * If the ICMP message advertises a Next-Hop MTU
768 			 * equal or larger than the maximum packet size we have
769 			 * ever sent, drop the message.
770 			 */
771 			mtu = (u_int)ntohs(icp->icmp_nextmtu);
772 			if (mtu >= tp->t_pmtud_mtu_sent)
773 				return;
774 			if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
775 				/*
776 				 * Calculate new MTU, and create corresponding
777 				 * route (traditional PMTUD).
778 				 */
779 				tp->t_flags &= ~TF_PMTUD_PEND;
780 				icmp_mtudisc(icp, inp->inp_rtableid);
781 			} else {
782 				/*
783 				 * Record the information got in the ICMP
784 				 * message; act on it later.
785 				 * If we had already recorded an ICMP message,
786 				 * replace the old one only if the new message
787 				 * refers to an older TCP segment
788 				 */
789 				if (tp->t_flags & TF_PMTUD_PEND) {
790 					if (SEQ_LT(tp->t_pmtud_th_seq, seq))
791 						return;
792 				} else
793 					tp->t_flags |= TF_PMTUD_PEND;
794 				tp->t_pmtud_th_seq = seq;
795 				tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
796 				tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
797 				tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
798 				return;
799 			}
800 		} else {
801 			/* ignore if we don't have a matching connection */
802 			return;
803 		}
804 		notify = tcp_mtudisc, ip = 0;
805 	} else if (cmd == PRC_MTUINC)
806 		notify = tcp_mtudisc_increase, ip = 0;
807 	else if (cmd == PRC_HOSTDEAD)
808 		ip = 0;
809 	else if (errno == 0)
810 		return;
811 
812 	if (ip) {
813 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
814 		inp = in_pcbhashlookup(&tcbtable,
815 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
816 		    rdomain);
817 		if (inp) {
818 			seq = ntohl(th->th_seq);
819 			if (inp->inp_socket &&
820 			    (tp = intotcpcb(inp)) &&
821 			    SEQ_GEQ(seq, tp->snd_una) &&
822 			    SEQ_LT(seq, tp->snd_max))
823 				notify(inp, errno);
824 		} else if (inetctlerrmap[cmd] == EHOSTUNREACH ||
825 		    inetctlerrmap[cmd] == ENETUNREACH ||
826 		    inetctlerrmap[cmd] == EHOSTDOWN) {
827 			struct sockaddr_in sin;
828 
829 			bzero(&sin, sizeof(sin));
830 			sin.sin_len = sizeof(sin);
831 			sin.sin_family = AF_INET;
832 			sin.sin_port = th->th_sport;
833 			sin.sin_addr = ip->ip_src;
834 			syn_cache_unreach(sintosa(&sin), sa, th, rdomain);
835 		}
836 	} else
837 		in_pcbnotifyall(&tcbtable, sa, rdomain, errno, notify);
838 }
839 
840 
841 #ifdef INET6
842 /*
843  * Path MTU Discovery handlers.
844  */
845 void
846 tcp6_mtudisc_callback(struct sockaddr_in6 *sin6, u_int rdomain)
847 {
848 	(void) in6_pcbnotify(&tcbtable, sin6, 0,
849 	    &sa6_any, 0, rdomain, PRC_MSGSIZE, NULL, tcp_mtudisc);
850 }
851 #endif /* INET6 */
852 
853 /*
854  * On receipt of path MTU corrections, flush old route and replace it
855  * with the new one.  Retransmit all unacknowledged packets, to ensure
856  * that all packets will be received.
857  */
858 void
859 tcp_mtudisc(struct inpcb *inp, int errno)
860 {
861 	struct tcpcb *tp = intotcpcb(inp);
862 	struct rtentry *rt = in_pcbrtentry(inp);
863 	int change = 0;
864 
865 	if (tp != 0) {
866 		int orig_maxseg = tp->t_maxseg;
867 		if (rt != 0) {
868 			/*
869 			 * If this was not a host route, remove and realloc.
870 			 */
871 			if ((rt->rt_flags & RTF_HOST) == 0) {
872 				in_rtchange(inp, errno);
873 				if ((rt = in_pcbrtentry(inp)) == 0)
874 					return;
875 			}
876 			if (orig_maxseg != tp->t_maxseg ||
877 			    (rt->rt_locks & RTV_MTU))
878 				change = 1;
879 		}
880 		tcp_mss(tp, -1);
881 
882 		/*
883 		 * Resend unacknowledged packets
884 		 */
885 		tp->snd_nxt = tp->snd_una;
886 		if (change || errno > 0)
887 			tcp_output(tp);
888 	}
889 }
890 
891 void
892 tcp_mtudisc_increase(struct inpcb *inp, int errno)
893 {
894 	struct tcpcb *tp = intotcpcb(inp);
895 	struct rtentry *rt = in_pcbrtentry(inp);
896 
897 	if (tp != 0 && rt != 0) {
898 		/*
899 		 * If this was a host route, remove and realloc.
900 		 */
901 		if (rt->rt_flags & RTF_HOST)
902 			in_rtchange(inp, errno);
903 
904 		/* also takes care of congestion window */
905 		tcp_mss(tp, -1);
906 	}
907 }
908 
909 /*
910  * Generate new ISNs with a method based on RFC1948
911  */
912 #define TCP_ISS_CONN_INC 4096
913 
914 void
915 tcp_set_iss_tsm(struct tcpcb *tp)
916 {
917 	SHA2_CTX ctx;
918 	union {
919 		uint8_t bytes[SHA512_DIGEST_LENGTH];
920 		uint32_t words[2];
921 	} digest;
922 	u_int rdomain = rtable_l2(tp->t_inpcb->inp_rtableid);
923 
924 	ctx = tcp_secret_ctx;
925 	SHA512Update(&ctx, &rdomain, sizeof(rdomain));
926 	SHA512Update(&ctx, &tp->t_inpcb->inp_lport, sizeof(u_short));
927 	SHA512Update(&ctx, &tp->t_inpcb->inp_fport, sizeof(u_short));
928 	if (tp->pf == AF_INET6) {
929 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr6,
930 		    sizeof(struct in6_addr));
931 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr6,
932 		    sizeof(struct in6_addr));
933 	} else {
934 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr,
935 		    sizeof(struct in_addr));
936 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr,
937 		    sizeof(struct in_addr));
938 	}
939 	SHA512Final(digest.bytes, &ctx);
940 	tcp_iss += TCP_ISS_CONN_INC;
941 	tp->iss = digest.words[0] + tcp_iss;
942 	tp->ts_modulate = digest.words[1];
943 }
944 
945 #ifdef TCP_SIGNATURE
946 int
947 tcp_signature_tdb_attach(void)
948 {
949 	return (0);
950 }
951 
952 int
953 tcp_signature_tdb_init(struct tdb *tdbp, struct xformsw *xsp,
954     struct ipsecinit *ii)
955 {
956 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
957 		return (EINVAL);
958 
959 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_NOWAIT);
960 	if (tdbp->tdb_amxkey == NULL)
961 		return (ENOMEM);
962 	bcopy(ii->ii_authkey, tdbp->tdb_amxkey, ii->ii_authkeylen);
963 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
964 
965 	return (0);
966 }
967 
968 int
969 tcp_signature_tdb_zeroize(struct tdb *tdbp)
970 {
971 	if (tdbp->tdb_amxkey) {
972 		explicit_bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
973 		free(tdbp->tdb_amxkey, M_XDATA, 0);
974 		tdbp->tdb_amxkey = NULL;
975 	}
976 
977 	return (0);
978 }
979 
980 int
981 tcp_signature_tdb_input(struct mbuf *m, struct tdb *tdbp, int skip, int protoff)
982 {
983 	return (0);
984 }
985 
986 int
987 tcp_signature_tdb_output(struct mbuf *m, struct tdb *tdbp, struct mbuf **mp,
988     int skip, int protoff)
989 {
990 	return (EINVAL);
991 }
992 
993 int
994 tcp_signature_apply(caddr_t fstate, caddr_t data, unsigned int len)
995 {
996 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
997 	return 0;
998 }
999 
1000 int
1001 tcp_signature(struct tdb *tdb, int af, struct mbuf *m, struct tcphdr *th,
1002     int iphlen, int doswap, char *sig)
1003 {
1004 	MD5_CTX ctx;
1005 	int len;
1006 	struct tcphdr th0;
1007 
1008 	MD5Init(&ctx);
1009 
1010 	switch(af) {
1011 	case 0:
1012 	case AF_INET: {
1013 		struct ippseudo ippseudo;
1014 		struct ip *ip;
1015 
1016 		ip = mtod(m, struct ip *);
1017 
1018 		ippseudo.ippseudo_src = ip->ip_src;
1019 		ippseudo.ippseudo_dst = ip->ip_dst;
1020 		ippseudo.ippseudo_pad = 0;
1021 		ippseudo.ippseudo_p = IPPROTO_TCP;
1022 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - iphlen);
1023 
1024 		MD5Update(&ctx, (char *)&ippseudo,
1025 		    sizeof(struct ippseudo));
1026 		break;
1027 		}
1028 #ifdef INET6
1029 	case AF_INET6: {
1030 		struct ip6_hdr_pseudo ip6pseudo;
1031 		struct ip6_hdr *ip6;
1032 
1033 		ip6 = mtod(m, struct ip6_hdr *);
1034 		bzero(&ip6pseudo, sizeof(ip6pseudo));
1035 		ip6pseudo.ip6ph_src = ip6->ip6_src;
1036 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
1037 		in6_clearscope(&ip6pseudo.ip6ph_src);
1038 		in6_clearscope(&ip6pseudo.ip6ph_dst);
1039 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
1040 		ip6pseudo.ip6ph_len = htonl(m->m_pkthdr.len - iphlen);
1041 
1042 		MD5Update(&ctx, (char *)&ip6pseudo,
1043 		    sizeof(ip6pseudo));
1044 		break;
1045 		}
1046 #endif
1047 	}
1048 
1049 	th0 = *th;
1050 	th0.th_sum = 0;
1051 
1052 	if (doswap) {
1053 		th0.th_seq = htonl(th0.th_seq);
1054 		th0.th_ack = htonl(th0.th_ack);
1055 		th0.th_win = htons(th0.th_win);
1056 		th0.th_urp = htons(th0.th_urp);
1057 	}
1058 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
1059 
1060 	len = m->m_pkthdr.len - iphlen - th->th_off * sizeof(uint32_t);
1061 
1062 	if (len > 0 &&
1063 	    m_apply(m, iphlen + th->th_off * sizeof(uint32_t), len,
1064 	    tcp_signature_apply, (caddr_t)&ctx))
1065 		return (-1);
1066 
1067 	MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen);
1068 	MD5Final(sig, &ctx);
1069 
1070 	return (0);
1071 }
1072 #endif /* TCP_SIGNATURE */
1073