xref: /openbsd-src/sys/netinet/tcp_subr.c (revision 897fc685943471cf985a0fe38ba076ea6fe74fa5)
1 /*	$OpenBSD: tcp_subr.c,v 1.170 2018/04/02 14:19:17 dhill 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 
440 	tp->sack_enable = tcp_do_sack;
441 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
442 	tp->t_inpcb = inp;
443 	/*
444 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
445 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
446 	 * reasonable initial retransmit time.
447 	 */
448 	tp->t_srtt = TCPTV_SRTTBASE;
449 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ <<
450 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
451 	tp->t_rttmin = TCPTV_MIN;
452 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
453 	    TCPTV_MIN, TCPTV_REXMTMAX);
454 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
455 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
456 
457 	tp->t_pmtud_mtu_sent = 0;
458 	tp->t_pmtud_mss_acked = 0;
459 
460 #ifdef INET6
461 	/* we disallow IPv4 mapped address completely. */
462 	if ((inp->inp_flags & INP_IPV6) == 0)
463 		tp->pf = PF_INET;
464 	else
465 		tp->pf = PF_INET6;
466 #else
467 	tp->pf = PF_INET;
468 #endif
469 
470 #ifdef INET6
471 	if (inp->inp_flags & INP_IPV6)
472 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
473 	else
474 #endif /* INET6 */
475 		inp->inp_ip.ip_ttl = ip_defttl;
476 
477 	inp->inp_ppcb = (caddr_t)tp;
478 	return (tp);
479 }
480 
481 /*
482  * Drop a TCP connection, reporting
483  * the specified error.  If connection is synchronized,
484  * then send a RST to peer.
485  */
486 struct tcpcb *
487 tcp_drop(struct tcpcb *tp, int errno)
488 {
489 	struct socket *so = tp->t_inpcb->inp_socket;
490 
491 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
492 		tp->t_state = TCPS_CLOSED;
493 		(void) tcp_output(tp);
494 		tcpstat_inc(tcps_drops);
495 	} else
496 		tcpstat_inc(tcps_conndrops);
497 	if (errno == ETIMEDOUT && tp->t_softerror)
498 		errno = tp->t_softerror;
499 	so->so_error = errno;
500 	return (tcp_close(tp));
501 }
502 
503 /*
504  * Close a TCP control block:
505  *	discard all space held by the tcp
506  *	discard internet protocol block
507  *	wake up any sleepers
508  */
509 struct tcpcb *
510 tcp_close(struct tcpcb *tp)
511 {
512 	struct inpcb *inp = tp->t_inpcb;
513 	struct socket *so = inp->inp_socket;
514 	struct sackhole *p, *q;
515 
516 	/* free the reassembly queue, if any */
517 	tcp_freeq(tp);
518 
519 	tcp_canceltimers(tp);
520 	TCP_CLEAR_DELACK(tp);
521 	syn_cache_cleanup(tp);
522 
523 	/* Free SACK holes. */
524 	q = p = tp->snd_holes;
525 	while (p != 0) {
526 		q = p->next;
527 		pool_put(&sackhl_pool, p);
528 		p = q;
529 	}
530 
531 	m_free(tp->t_template);
532 
533 	tp->t_flags |= TF_DEAD;
534 	TCP_TIMER_ARM(tp, TCPT_REAPER, 0);
535 
536 	inp->inp_ppcb = NULL;
537 	soisdisconnected(so);
538 	in_pcbdetach(inp);
539 	return (NULL);
540 }
541 
542 int
543 tcp_freeq(struct tcpcb *tp)
544 {
545 	struct tcpqent *qe;
546 	int rv = 0;
547 
548 	while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) {
549 		TAILQ_REMOVE(&tp->t_segq, qe, tcpqe_q);
550 		m_freem(qe->tcpqe_m);
551 		pool_put(&tcpqe_pool, qe);
552 		rv = 1;
553 	}
554 	return (rv);
555 }
556 
557 /*
558  * Compute proper scaling value for receiver window from buffer space
559  */
560 
561 void
562 tcp_rscale(struct tcpcb *tp, u_long hiwat)
563 {
564 	tp->request_r_scale = 0;
565 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
566 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
567 		tp->request_r_scale++;
568 }
569 
570 /*
571  * Notify a tcp user of an asynchronous error;
572  * store error as soft error, but wake up user
573  * (for now, won't do anything until can select for soft error).
574  */
575 void
576 tcp_notify(struct inpcb *inp, int error)
577 {
578 	struct tcpcb *tp = intotcpcb(inp);
579 	struct socket *so = inp->inp_socket;
580 
581 	/*
582 	 * Ignore some errors if we are hooked up.
583 	 * If connection hasn't completed, has retransmitted several times,
584 	 * and receives a second error, give up now.  This is better
585 	 * than waiting a long time to establish a connection that
586 	 * can never complete.
587 	 */
588 	if (tp->t_state == TCPS_ESTABLISHED &&
589 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
590 	      error == EHOSTDOWN)) {
591 		return;
592 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
593 	    tp->t_rxtshift > 3 && tp->t_softerror)
594 		so->so_error = error;
595 	else
596 		tp->t_softerror = error;
597 	wakeup((caddr_t) &so->so_timeo);
598 	sorwakeup(so);
599 	sowwakeup(so);
600 }
601 
602 #ifdef INET6
603 void
604 tcp6_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *d)
605 {
606 	struct tcphdr th;
607 	struct tcpcb *tp;
608 	void (*notify)(struct inpcb *, int) = tcp_notify;
609 	struct ip6_hdr *ip6;
610 	const struct sockaddr_in6 *sa6_src = NULL;
611 	struct sockaddr_in6 *sa6 = satosin6(sa);
612 	struct inpcb *inp;
613 	struct mbuf *m;
614 	tcp_seq seq;
615 	int off;
616 	struct {
617 		u_int16_t th_sport;
618 		u_int16_t th_dport;
619 		u_int32_t th_seq;
620 	} *thp;
621 
622 	CTASSERT(sizeof(*thp) <= sizeof(th));
623 	if (sa->sa_family != AF_INET6 ||
624 	    sa->sa_len != sizeof(struct sockaddr_in6) ||
625 	    IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) ||
626 	    IN6_IS_ADDR_V4MAPPED(&sa6->sin6_addr))
627 		return;
628 	if ((unsigned)cmd >= PRC_NCMDS)
629 		return;
630 	else if (cmd == PRC_QUENCH) {
631 		/*
632 		 * Don't honor ICMP Source Quench messages meant for
633 		 * TCP connections.
634 		 */
635 		/* XXX there's no PRC_QUENCH in IPv6 */
636 		return;
637 	} else if (PRC_IS_REDIRECT(cmd))
638 		notify = in_rtchange, d = NULL;
639 	else if (cmd == PRC_MSGSIZE)
640 		; /* special code is present, see below */
641 	else if (cmd == PRC_HOSTDEAD)
642 		d = NULL;
643 	else if (inet6ctlerrmap[cmd] == 0)
644 		return;
645 
646 	/* if the parameter is from icmp6, decode it. */
647 	if (d != NULL) {
648 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
649 		m = ip6cp->ip6c_m;
650 		ip6 = ip6cp->ip6c_ip6;
651 		off = ip6cp->ip6c_off;
652 		sa6_src = ip6cp->ip6c_src;
653 	} else {
654 		m = NULL;
655 		ip6 = NULL;
656 		sa6_src = &sa6_any;
657 	}
658 
659 	if (ip6) {
660 		/*
661 		 * XXX: We assume that when ip6 is non NULL,
662 		 * M and OFF are valid.
663 		 */
664 
665 		/* check if we can safely examine src and dst ports */
666 		if (m->m_pkthdr.len < off + sizeof(*thp))
667 			return;
668 
669 		bzero(&th, sizeof(th));
670 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
671 
672 		/*
673 		 * Check to see if we have a valid TCP connection
674 		 * corresponding to the address in the ICMPv6 message
675 		 * payload.
676 		 */
677 		inp = in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
678 		    th.th_dport, &sa6_src->sin6_addr, th.th_sport, rdomain);
679 		if (cmd == PRC_MSGSIZE) {
680 			/*
681 			 * Depending on the value of "valid" and routing table
682 			 * size (mtudisc_{hi,lo}wat), we will:
683 			 * - recalcurate the new MTU and create the
684 			 *   corresponding routing entry, or
685 			 * - ignore the MTU change notification.
686 			 */
687 			icmp6_mtudisc_update((struct ip6ctlparam *)d, inp != NULL);
688 			return;
689 		}
690 		if (inp) {
691 			seq = ntohl(th.th_seq);
692 			if (inp->inp_socket &&
693 			    (tp = intotcpcb(inp)) &&
694 			    SEQ_GEQ(seq, tp->snd_una) &&
695 			    SEQ_LT(seq, tp->snd_max))
696 				notify(inp, inet6ctlerrmap[cmd]);
697 		} else if (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
698 		    inet6ctlerrmap[cmd] == ENETUNREACH ||
699 		    inet6ctlerrmap[cmd] == EHOSTDOWN)
700 			syn_cache_unreach((struct sockaddr *)sa6_src,
701 			    sa, &th, rdomain);
702 	} else {
703 		(void) in6_pcbnotify(&tcbtable, sa6, 0,
704 		    sa6_src, 0, rdomain, cmd, NULL, notify);
705 	}
706 }
707 #endif
708 
709 void
710 tcp_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
711 {
712 	struct ip *ip = v;
713 	struct tcphdr *th;
714 	struct tcpcb *tp;
715 	struct inpcb *inp;
716 	struct in_addr faddr;
717 	tcp_seq seq;
718 	u_int mtu;
719 	void (*notify)(struct inpcb *, int) = tcp_notify;
720 	int errno;
721 
722 	if (sa->sa_family != AF_INET)
723 		return;
724 	faddr = satosin(sa)->sin_addr;
725 	if (faddr.s_addr == INADDR_ANY)
726 		return;
727 
728 	if ((unsigned)cmd >= PRC_NCMDS)
729 		return;
730 	errno = inetctlerrmap[cmd];
731 	if (cmd == PRC_QUENCH)
732 		/*
733 		 * Don't honor ICMP Source Quench messages meant for
734 		 * TCP connections.
735 		 */
736 		return;
737 	else if (PRC_IS_REDIRECT(cmd))
738 		notify = in_rtchange, ip = 0;
739 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip) {
740 		/*
741 		 * Verify that the packet in the icmp payload refers
742 		 * to an existing TCP connection.
743 		 */
744 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
745 		seq = ntohl(th->th_seq);
746 		inp = in_pcbhashlookup(&tcbtable,
747 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
748 		    rdomain);
749 		if (inp && (tp = intotcpcb(inp)) &&
750 		    SEQ_GEQ(seq, tp->snd_una) &&
751 		    SEQ_LT(seq, tp->snd_max)) {
752 			struct icmp *icp;
753 			icp = (struct icmp *)((caddr_t)ip -
754 					      offsetof(struct icmp, icmp_ip));
755 
756 			/*
757 			 * If the ICMP message advertises a Next-Hop MTU
758 			 * equal or larger than the maximum packet size we have
759 			 * ever sent, drop the message.
760 			 */
761 			mtu = (u_int)ntohs(icp->icmp_nextmtu);
762 			if (mtu >= tp->t_pmtud_mtu_sent)
763 				return;
764 			if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
765 				/*
766 				 * Calculate new MTU, and create corresponding
767 				 * route (traditional PMTUD).
768 				 */
769 				tp->t_flags &= ~TF_PMTUD_PEND;
770 				icmp_mtudisc(icp, inp->inp_rtableid);
771 			} else {
772 				/*
773 				 * Record the information got in the ICMP
774 				 * message; act on it later.
775 				 * If we had already recorded an ICMP message,
776 				 * replace the old one only if the new message
777 				 * refers to an older TCP segment
778 				 */
779 				if (tp->t_flags & TF_PMTUD_PEND) {
780 					if (SEQ_LT(tp->t_pmtud_th_seq, seq))
781 						return;
782 				} else
783 					tp->t_flags |= TF_PMTUD_PEND;
784 				tp->t_pmtud_th_seq = seq;
785 				tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
786 				tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
787 				tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
788 				return;
789 			}
790 		} else {
791 			/* ignore if we don't have a matching connection */
792 			return;
793 		}
794 		notify = tcp_mtudisc, ip = 0;
795 	} else if (cmd == PRC_MTUINC)
796 		notify = tcp_mtudisc_increase, ip = 0;
797 	else if (cmd == PRC_HOSTDEAD)
798 		ip = 0;
799 	else if (errno == 0)
800 		return;
801 
802 	if (ip) {
803 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
804 		inp = in_pcbhashlookup(&tcbtable,
805 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
806 		    rdomain);
807 		if (inp) {
808 			seq = ntohl(th->th_seq);
809 			if (inp->inp_socket &&
810 			    (tp = intotcpcb(inp)) &&
811 			    SEQ_GEQ(seq, tp->snd_una) &&
812 			    SEQ_LT(seq, tp->snd_max))
813 				notify(inp, errno);
814 		} else if (inetctlerrmap[cmd] == EHOSTUNREACH ||
815 		    inetctlerrmap[cmd] == ENETUNREACH ||
816 		    inetctlerrmap[cmd] == EHOSTDOWN) {
817 			struct sockaddr_in sin;
818 
819 			bzero(&sin, sizeof(sin));
820 			sin.sin_len = sizeof(sin);
821 			sin.sin_family = AF_INET;
822 			sin.sin_port = th->th_sport;
823 			sin.sin_addr = ip->ip_src;
824 			syn_cache_unreach(sintosa(&sin), sa, th, rdomain);
825 		}
826 	} else
827 		in_pcbnotifyall(&tcbtable, sa, rdomain, errno, notify);
828 }
829 
830 
831 #ifdef INET6
832 /*
833  * Path MTU Discovery handlers.
834  */
835 void
836 tcp6_mtudisc_callback(struct sockaddr_in6 *sin6, u_int rdomain)
837 {
838 	(void) in6_pcbnotify(&tcbtable, sin6, 0,
839 	    &sa6_any, 0, rdomain, PRC_MSGSIZE, NULL, tcp_mtudisc);
840 }
841 #endif /* INET6 */
842 
843 /*
844  * On receipt of path MTU corrections, flush old route and replace it
845  * with the new one.  Retransmit all unacknowledged packets, to ensure
846  * that all packets will be received.
847  */
848 void
849 tcp_mtudisc(struct inpcb *inp, int errno)
850 {
851 	struct tcpcb *tp = intotcpcb(inp);
852 	struct rtentry *rt;
853 	int change = 0;
854 
855 	if (tp == NULL)
856 		return;
857 
858 	rt = in_pcbrtentry(inp);
859 	if (rt != NULL) {
860 		int orig_maxseg = tp->t_maxseg;
861 
862 		/*
863 		 * If this was not a host route, remove and realloc.
864 		 */
865 		if ((rt->rt_flags & RTF_HOST) == 0) {
866 			in_rtchange(inp, errno);
867 			if ((rt = in_pcbrtentry(inp)) == NULL)
868 				return;
869 		}
870 		if (orig_maxseg != tp->t_maxseg ||
871 		    (rt->rt_locks & RTV_MTU))
872 			change = 1;
873 	}
874 	tcp_mss(tp, -1);
875 
876 	/*
877 	 * Resend unacknowledged packets
878 	 */
879 	tp->snd_nxt = tp->snd_una;
880 	if (change || errno > 0)
881 		tcp_output(tp);
882 }
883 
884 void
885 tcp_mtudisc_increase(struct inpcb *inp, int errno)
886 {
887 	struct tcpcb *tp = intotcpcb(inp);
888 	struct rtentry *rt = in_pcbrtentry(inp);
889 
890 	if (tp != 0 && rt != 0) {
891 		/*
892 		 * If this was a host route, remove and realloc.
893 		 */
894 		if (rt->rt_flags & RTF_HOST)
895 			in_rtchange(inp, errno);
896 
897 		/* also takes care of congestion window */
898 		tcp_mss(tp, -1);
899 	}
900 }
901 
902 /*
903  * Generate new ISNs with a method based on RFC1948
904  */
905 #define TCP_ISS_CONN_INC 4096
906 
907 void
908 tcp_set_iss_tsm(struct tcpcb *tp)
909 {
910 	SHA2_CTX ctx;
911 	union {
912 		uint8_t bytes[SHA512_DIGEST_LENGTH];
913 		uint32_t words[2];
914 	} digest;
915 	u_int rdomain = rtable_l2(tp->t_inpcb->inp_rtableid);
916 
917 	ctx = tcp_secret_ctx;
918 	SHA512Update(&ctx, &rdomain, sizeof(rdomain));
919 	SHA512Update(&ctx, &tp->t_inpcb->inp_lport, sizeof(u_short));
920 	SHA512Update(&ctx, &tp->t_inpcb->inp_fport, sizeof(u_short));
921 	if (tp->pf == AF_INET6) {
922 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr6,
923 		    sizeof(struct in6_addr));
924 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr6,
925 		    sizeof(struct in6_addr));
926 	} else {
927 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr,
928 		    sizeof(struct in_addr));
929 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr,
930 		    sizeof(struct in_addr));
931 	}
932 	SHA512Final(digest.bytes, &ctx);
933 	tcp_iss += TCP_ISS_CONN_INC;
934 	tp->iss = digest.words[0] + tcp_iss;
935 	tp->ts_modulate = digest.words[1];
936 }
937 
938 #ifdef TCP_SIGNATURE
939 int
940 tcp_signature_tdb_attach(void)
941 {
942 	return (0);
943 }
944 
945 int
946 tcp_signature_tdb_init(struct tdb *tdbp, struct xformsw *xsp,
947     struct ipsecinit *ii)
948 {
949 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
950 		return (EINVAL);
951 
952 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_NOWAIT);
953 	if (tdbp->tdb_amxkey == NULL)
954 		return (ENOMEM);
955 	memcpy(tdbp->tdb_amxkey, ii->ii_authkey, ii->ii_authkeylen);
956 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
957 
958 	return (0);
959 }
960 
961 int
962 tcp_signature_tdb_zeroize(struct tdb *tdbp)
963 {
964 	if (tdbp->tdb_amxkey) {
965 		explicit_bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
966 		free(tdbp->tdb_amxkey, M_XDATA, tdbp->tdb_amxkeylen);
967 		tdbp->tdb_amxkey = NULL;
968 	}
969 
970 	return (0);
971 }
972 
973 int
974 tcp_signature_tdb_input(struct mbuf *m, struct tdb *tdbp, int skip, int protoff)
975 {
976 	return (0);
977 }
978 
979 int
980 tcp_signature_tdb_output(struct mbuf *m, struct tdb *tdbp, struct mbuf **mp,
981     int skip, int protoff)
982 {
983 	return (EINVAL);
984 }
985 
986 int
987 tcp_signature_apply(caddr_t fstate, caddr_t data, unsigned int len)
988 {
989 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
990 	return 0;
991 }
992 
993 int
994 tcp_signature(struct tdb *tdb, int af, struct mbuf *m, struct tcphdr *th,
995     int iphlen, int doswap, char *sig)
996 {
997 	MD5_CTX ctx;
998 	int len;
999 	struct tcphdr th0;
1000 
1001 	MD5Init(&ctx);
1002 
1003 	switch(af) {
1004 	case 0:
1005 	case AF_INET: {
1006 		struct ippseudo ippseudo;
1007 		struct ip *ip;
1008 
1009 		ip = mtod(m, struct ip *);
1010 
1011 		ippseudo.ippseudo_src = ip->ip_src;
1012 		ippseudo.ippseudo_dst = ip->ip_dst;
1013 		ippseudo.ippseudo_pad = 0;
1014 		ippseudo.ippseudo_p = IPPROTO_TCP;
1015 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - iphlen);
1016 
1017 		MD5Update(&ctx, (char *)&ippseudo,
1018 		    sizeof(struct ippseudo));
1019 		break;
1020 		}
1021 #ifdef INET6
1022 	case AF_INET6: {
1023 		struct ip6_hdr_pseudo ip6pseudo;
1024 		struct ip6_hdr *ip6;
1025 
1026 		ip6 = mtod(m, struct ip6_hdr *);
1027 		bzero(&ip6pseudo, sizeof(ip6pseudo));
1028 		ip6pseudo.ip6ph_src = ip6->ip6_src;
1029 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
1030 		in6_clearscope(&ip6pseudo.ip6ph_src);
1031 		in6_clearscope(&ip6pseudo.ip6ph_dst);
1032 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
1033 		ip6pseudo.ip6ph_len = htonl(m->m_pkthdr.len - iphlen);
1034 
1035 		MD5Update(&ctx, (char *)&ip6pseudo,
1036 		    sizeof(ip6pseudo));
1037 		break;
1038 		}
1039 #endif
1040 	}
1041 
1042 	th0 = *th;
1043 	th0.th_sum = 0;
1044 
1045 	if (doswap) {
1046 		th0.th_seq = htonl(th0.th_seq);
1047 		th0.th_ack = htonl(th0.th_ack);
1048 		th0.th_win = htons(th0.th_win);
1049 		th0.th_urp = htons(th0.th_urp);
1050 	}
1051 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
1052 
1053 	len = m->m_pkthdr.len - iphlen - th->th_off * sizeof(uint32_t);
1054 
1055 	if (len > 0 &&
1056 	    m_apply(m, iphlen + th->th_off * sizeof(uint32_t), len,
1057 	    tcp_signature_apply, (caddr_t)&ctx))
1058 		return (-1);
1059 
1060 	MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen);
1061 	MD5Final(sig, &ctx);
1062 
1063 	return (0);
1064 }
1065 #endif /* TCP_SIGNATURE */
1066