xref: /openbsd-src/sys/netinet/tcp_subr.c (revision c0dd97bfcad3dab6c31ec12b9de1274fd2d2f993)
1 /*	$OpenBSD: tcp_subr.c,v 1.166 2017/10/22 14:11:34 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 tcp_seq  tcp_iss;
129 
130 /*
131  * Tcp initialization
132  */
133 void
134 tcp_init(void)
135 {
136 	tcp_iss = 1;		/* wrong */
137 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, IPL_SOFTNET, 0,
138 	    "tcpcb", NULL);
139 	pool_init(&tcpqe_pool, sizeof(struct tcpqent), 0, IPL_SOFTNET, 0,
140 	    "tcpqe", NULL);
141 	pool_sethardlimit(&tcpqe_pool, tcp_reass_limit, NULL, 0);
142 	pool_init(&sackhl_pool, sizeof(struct sackhole), 0, IPL_SOFTNET, 0,
143 	    "sackhl", NULL);
144 	pool_sethardlimit(&sackhl_pool, tcp_sackhole_limit, NULL, 0);
145 	in_pcbinit(&tcbtable, TCB_INITIAL_HASH_SIZE);
146 	tcpcounters = counters_alloc(tcps_ncounters);
147 
148 #ifdef INET6
149 	/*
150 	 * Since sizeof(struct ip6_hdr) > sizeof(struct ip), we
151 	 * do max length checks/computations only on the former.
152 	 */
153 	if (max_protohdr < (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)))
154 		max_protohdr = (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
155 	if ((max_linkhdr + sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) >
156 	    MHLEN)
157 		panic("tcp_init");
158 
159 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
160 #endif /* INET6 */
161 
162 	/* Initialize the compressed state engine. */
163 	syn_cache_init();
164 
165 	/* Initialize timer state. */
166 	tcp_timer_init();
167 }
168 
169 /*
170  * Create template to be used to send tcp packets on a connection.
171  * Call after host entry created, allocates an mbuf and fills
172  * in a skeletal tcp/ip header, minimizing the amount of work
173  * necessary when the connection is used.
174  *
175  * To support IPv6 in addition to IPv4 and considering that the sizes of
176  * the IPv4 and IPv6 headers are not the same, we now use a separate pointer
177  * for the TCP header.  Also, we made the former tcpiphdr header pointer
178  * into just an IP overlay pointer, with casting as appropriate for v6. rja
179  */
180 struct mbuf *
181 tcp_template(struct tcpcb *tp)
182 {
183 	struct inpcb *inp = tp->t_inpcb;
184 	struct mbuf *m;
185 	struct tcphdr *th;
186 
187 	if ((m = tp->t_template) == 0) {
188 		m = m_get(M_DONTWAIT, MT_HEADER);
189 		if (m == NULL)
190 			return (0);
191 
192 		switch (tp->pf) {
193 		case 0:	/*default to PF_INET*/
194 		case AF_INET:
195 			m->m_len = sizeof(struct ip);
196 			break;
197 #ifdef INET6
198 		case AF_INET6:
199 			m->m_len = sizeof(struct ip6_hdr);
200 			break;
201 #endif /* INET6 */
202 		}
203 		m->m_len += sizeof (struct tcphdr);
204 
205 		/*
206 		 * The link header, network header, TCP header, and TCP options
207 		 * all must fit in this mbuf. For now, assume the worst case of
208 		 * TCP options size. Eventually, compute this from tp flags.
209 		 */
210 		if (m->m_len + MAX_TCPOPTLEN + max_linkhdr >= MHLEN) {
211 			MCLGET(m, M_DONTWAIT);
212 			if ((m->m_flags & M_EXT) == 0) {
213 				m_free(m);
214 				return (0);
215 			}
216 		}
217 	}
218 
219 	switch(tp->pf) {
220 	case AF_INET:
221 		{
222 			struct ipovly *ipovly;
223 
224 			ipovly = mtod(m, struct ipovly *);
225 
226 			bzero(ipovly->ih_x1, sizeof ipovly->ih_x1);
227 			ipovly->ih_pr = IPPROTO_TCP;
228 			ipovly->ih_len = htons(sizeof (struct tcphdr));
229 			ipovly->ih_src = inp->inp_laddr;
230 			ipovly->ih_dst = inp->inp_faddr;
231 
232 			th = (struct tcphdr *)(mtod(m, caddr_t) +
233 				sizeof(struct ip));
234 		}
235 		break;
236 #ifdef INET6
237 	case AF_INET6:
238 		{
239 			struct ip6_hdr *ip6;
240 
241 			ip6 = mtod(m, struct ip6_hdr *);
242 
243 			ip6->ip6_src = inp->inp_laddr6;
244 			ip6->ip6_dst = inp->inp_faddr6;
245 			ip6->ip6_flow = htonl(0x60000000) |
246 			    (inp->inp_flowinfo & IPV6_FLOWLABEL_MASK);
247 
248 			ip6->ip6_nxt = IPPROTO_TCP;
249 			ip6->ip6_plen = htons(sizeof(struct tcphdr)); /*XXX*/
250 			ip6->ip6_hlim = in6_selecthlim(inp);	/*XXX*/
251 
252 			th = (struct tcphdr *)(mtod(m, caddr_t) +
253 				sizeof(struct ip6_hdr));
254 		}
255 		break;
256 #endif /* INET6 */
257 	}
258 
259 	th->th_sport = inp->inp_lport;
260 	th->th_dport = inp->inp_fport;
261 	th->th_seq = 0;
262 	th->th_ack = 0;
263 	th->th_x2  = 0;
264 	th->th_off = 5;
265 	th->th_flags = 0;
266 	th->th_win = 0;
267 	th->th_urp = 0;
268 	th->th_sum = 0;
269 	return (m);
270 }
271 
272 /*
273  * Send a single message to the TCP at address specified by
274  * the given TCP/IP header.  If m == 0, then we make a copy
275  * of the tcpiphdr at ti and send directly to the addressed host.
276  * This is used to force keep alive messages out using the TCP
277  * template for a connection tp->t_template.  If flags are given
278  * then we send a message back to the TCP which originated the
279  * segment ti, and discard the mbuf containing it and any other
280  * attached mbufs.
281  *
282  * In any case the ack and sequence number of the transmitted
283  * segment are as specified by the parameters.
284  */
285 void
286 tcp_respond(struct tcpcb *tp, caddr_t template, struct tcphdr *th0,
287     tcp_seq ack, tcp_seq seq, int flags, u_int rtableid)
288 {
289 	int tlen;
290 	int win = 0;
291 	struct mbuf *m = NULL;
292 	struct tcphdr *th;
293 	struct ip *ip;
294 #ifdef INET6
295 	struct ip6_hdr *ip6;
296 #endif
297 	int af;		/* af on wire */
298 
299 	if (tp) {
300 		struct socket *so = tp->t_inpcb->inp_socket;
301 		win = sbspace(so, &so->so_rcv);
302 		/*
303 		 * If this is called with an unconnected
304 		 * socket/tp/pcb (tp->pf is 0), we lose.
305 		 */
306 		af = tp->pf;
307 	} else
308 		af = (((struct ip *)template)->ip_v == 6) ? AF_INET6 : AF_INET;
309 
310 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
311 	if (m == NULL)
312 		return;
313 	m->m_data += max_linkhdr;
314 	tlen = 0;
315 
316 #define xchg(a,b,type) do { type t; t=a; a=b; b=t; } while (0)
317 	switch (af) {
318 #ifdef INET6
319 	case AF_INET6:
320 		ip6 = mtod(m, struct ip6_hdr *);
321 		th = (struct tcphdr *)(ip6 + 1);
322 		tlen = sizeof(*ip6) + sizeof(*th);
323 		if (th0) {
324 			bcopy(template, ip6, sizeof(*ip6));
325 			bcopy(th0, th, sizeof(*th));
326 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
327 		} else {
328 			bcopy(template, ip6, tlen);
329 		}
330 		break;
331 #endif /* INET6 */
332 	case AF_INET:
333 		ip = mtod(m, struct ip *);
334 		th = (struct tcphdr *)(ip + 1);
335 		tlen = sizeof(*ip) + sizeof(*th);
336 		if (th0) {
337 			bcopy(template, ip, sizeof(*ip));
338 			bcopy(th0, th, sizeof(*th));
339 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, u_int32_t);
340 		} else {
341 			bcopy(template, ip, tlen);
342 		}
343 		break;
344 	}
345 	if (th0)
346 		xchg(th->th_dport, th->th_sport, u_int16_t);
347 	else
348 		flags = TH_ACK;
349 #undef xchg
350 
351 	th->th_seq = htonl(seq);
352 	th->th_ack = htonl(ack);
353 	th->th_x2 = 0;
354 	th->th_off = sizeof (struct tcphdr) >> 2;
355 	th->th_flags = flags;
356 	if (tp)
357 		win >>= tp->rcv_scale;
358 	if (win > TCP_MAXWIN)
359 		win = TCP_MAXWIN;
360 	th->th_win = htons((u_int16_t)win);
361 	th->th_urp = 0;
362 
363 	if (tp && (tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
364 	    (flags & TH_RST) == 0 && (tp->t_flags & TF_RCVD_TSTMP)) {
365 		u_int32_t *lp = (u_int32_t *)(th + 1);
366 		/* Form timestamp option as shown in appendix A of RFC 1323. */
367 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
368 		*lp++ = htonl(tcp_now + tp->ts_modulate);
369 		*lp   = htonl(tp->ts_recent);
370 		tlen += TCPOLEN_TSTAMP_APPA;
371 		th->th_off = (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_APPA) >> 2;
372 	}
373 
374 	m->m_len = tlen;
375 	m->m_pkthdr.len = tlen;
376 	m->m_pkthdr.ph_ifidx = 0;
377 	m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT;
378 
379 	/* force routing table */
380 	if (tp)
381 		m->m_pkthdr.ph_rtableid = tp->t_inpcb->inp_rtableid;
382 	else
383 		m->m_pkthdr.ph_rtableid = rtableid;
384 
385 	switch (af) {
386 #ifdef INET6
387 	case AF_INET6:
388 		ip6->ip6_flow = htonl(0x60000000);
389 		ip6->ip6_nxt  = IPPROTO_TCP;
390 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL);	/*XXX*/
391 		ip6->ip6_plen = tlen - sizeof(struct ip6_hdr);
392 		ip6->ip6_plen = htons(ip6->ip6_plen);
393 		ip6_output(m, tp ? tp->t_inpcb->inp_outputopts6 : NULL,
394 		    tp ? &tp->t_inpcb->inp_route6 : NULL,
395 		    0, NULL,
396 		    tp ? tp->t_inpcb : NULL);
397 		break;
398 #endif /* INET6 */
399 	case AF_INET:
400 		ip->ip_len = htons(tlen);
401 		ip->ip_ttl = ip_defttl;
402 		ip->ip_tos = 0;
403 		ip_output(m, NULL,
404 		    tp ? &tp->t_inpcb->inp_route : NULL,
405 		    ip_mtudisc ? IP_MTUDISC : 0, NULL,
406 		    tp ? tp->t_inpcb : NULL, 0);
407 		break;
408 	}
409 }
410 
411 /*
412  * Create a new TCP control block, making an
413  * empty reassembly queue and hooking it to the argument
414  * protocol control block.
415  */
416 struct tcpcb *
417 tcp_newtcpcb(struct inpcb *inp)
418 {
419 	struct tcpcb *tp;
420 	int i;
421 
422 	tp = pool_get(&tcpcb_pool, PR_NOWAIT|PR_ZERO);
423 	if (tp == NULL)
424 		return (NULL);
425 	TAILQ_INIT(&tp->t_segq);
426 	tp->t_maxseg = tcp_mssdflt;
427 	tp->t_maxopd = 0;
428 
429 	TCP_INIT_DELACK(tp);
430 	for (i = 0; i < TCPT_NTIMERS; i++)
431 		TCP_TIMER_INIT(tp, i);
432 	timeout_set(&tp->t_reap_to, tcp_reaper, tp);
433 
434 	tp->sack_enable = tcp_do_sack;
435 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
436 	tp->t_inpcb = inp;
437 	/*
438 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
439 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
440 	 * reasonable initial retransmit time.
441 	 */
442 	tp->t_srtt = TCPTV_SRTTBASE;
443 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ <<
444 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
445 	tp->t_rttmin = TCPTV_MIN;
446 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
447 	    TCPTV_MIN, TCPTV_REXMTMAX);
448 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
449 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
450 
451 	tp->t_pmtud_mtu_sent = 0;
452 	tp->t_pmtud_mss_acked = 0;
453 
454 #ifdef INET6
455 	/* we disallow IPv4 mapped address completely. */
456 	if ((inp->inp_flags & INP_IPV6) == 0)
457 		tp->pf = PF_INET;
458 	else
459 		tp->pf = PF_INET6;
460 #else
461 	tp->pf = PF_INET;
462 #endif
463 
464 #ifdef INET6
465 	if (inp->inp_flags & INP_IPV6)
466 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
467 	else
468 #endif /* INET6 */
469 		inp->inp_ip.ip_ttl = ip_defttl;
470 
471 	inp->inp_ppcb = (caddr_t)tp;
472 	return (tp);
473 }
474 
475 /*
476  * Drop a TCP connection, reporting
477  * the specified error.  If connection is synchronized,
478  * then send a RST to peer.
479  */
480 struct tcpcb *
481 tcp_drop(struct tcpcb *tp, int errno)
482 {
483 	struct socket *so = tp->t_inpcb->inp_socket;
484 
485 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
486 		tp->t_state = TCPS_CLOSED;
487 		(void) tcp_output(tp);
488 		tcpstat_inc(tcps_drops);
489 	} else
490 		tcpstat_inc(tcps_conndrops);
491 	if (errno == ETIMEDOUT && tp->t_softerror)
492 		errno = tp->t_softerror;
493 	so->so_error = errno;
494 	return (tcp_close(tp));
495 }
496 
497 /*
498  * Close a TCP control block:
499  *	discard all space held by the tcp
500  *	discard internet protocol block
501  *	wake up any sleepers
502  */
503 struct tcpcb *
504 tcp_close(struct tcpcb *tp)
505 {
506 	struct inpcb *inp = tp->t_inpcb;
507 	struct socket *so = inp->inp_socket;
508 	struct sackhole *p, *q;
509 
510 	/* free the reassembly queue, if any */
511 	tcp_freeq(tp);
512 
513 	tcp_canceltimers(tp);
514 	TCP_CLEAR_DELACK(tp);
515 	syn_cache_cleanup(tp);
516 
517 	/* Free SACK holes. */
518 	q = p = tp->snd_holes;
519 	while (p != 0) {
520 		q = p->next;
521 		pool_put(&sackhl_pool, p);
522 		p = q;
523 	}
524 
525 	m_free(tp->t_template);
526 
527 	tp->t_flags |= TF_DEAD;
528 	timeout_add(&tp->t_reap_to, 0);
529 
530 	inp->inp_ppcb = 0;
531 	soisdisconnected(so);
532 	in_pcbdetach(inp);
533 	return (NULL);
534 }
535 
536 void
537 tcp_reaper(void *arg)
538 {
539 	struct tcpcb *tp = arg;
540 
541 	pool_put(&tcpcb_pool, tp);
542 	tcpstat_inc(tcps_closed);
543 }
544 
545 int
546 tcp_freeq(struct tcpcb *tp)
547 {
548 	struct tcpqent *qe;
549 	int rv = 0;
550 
551 	while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) {
552 		TAILQ_REMOVE(&tp->t_segq, qe, tcpqe_q);
553 		m_freem(qe->tcpqe_m);
554 		pool_put(&tcpqe_pool, qe);
555 		rv = 1;
556 	}
557 	return (rv);
558 }
559 
560 /*
561  * Compute proper scaling value for receiver window from buffer space
562  */
563 
564 void
565 tcp_rscale(struct tcpcb *tp, u_long hiwat)
566 {
567 	tp->request_r_scale = 0;
568 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
569 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
570 		tp->request_r_scale++;
571 }
572 
573 /*
574  * Notify a tcp user of an asynchronous error;
575  * store error as soft error, but wake up user
576  * (for now, won't do anything until can select for soft error).
577  */
578 void
579 tcp_notify(struct inpcb *inp, int error)
580 {
581 	struct tcpcb *tp = intotcpcb(inp);
582 	struct socket *so = inp->inp_socket;
583 
584 	/*
585 	 * Ignore some errors if we are hooked up.
586 	 * If connection hasn't completed, has retransmitted several times,
587 	 * and receives a second error, give up now.  This is better
588 	 * than waiting a long time to establish a connection that
589 	 * can never complete.
590 	 */
591 	if (tp->t_state == TCPS_ESTABLISHED &&
592 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
593 	      error == EHOSTDOWN)) {
594 		return;
595 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
596 	    tp->t_rxtshift > 3 && tp->t_softerror)
597 		so->so_error = error;
598 	else
599 		tp->t_softerror = error;
600 	wakeup((caddr_t) &so->so_timeo);
601 	sorwakeup(so);
602 	sowwakeup(so);
603 }
604 
605 #ifdef INET6
606 void
607 tcp6_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *d)
608 {
609 	struct tcphdr th;
610 	struct tcpcb *tp;
611 	void (*notify)(struct inpcb *, int) = tcp_notify;
612 	struct ip6_hdr *ip6;
613 	const struct sockaddr_in6 *sa6_src = NULL;
614 	struct sockaddr_in6 *sa6 = satosin6(sa);
615 	struct inpcb *inp;
616 	struct mbuf *m;
617 	tcp_seq seq;
618 	int off;
619 	struct {
620 		u_int16_t th_sport;
621 		u_int16_t th_dport;
622 		u_int32_t th_seq;
623 	} *thp;
624 
625 	CTASSERT(sizeof(*thp) <= sizeof(th));
626 	if (sa->sa_family != AF_INET6 ||
627 	    sa->sa_len != sizeof(struct sockaddr_in6) ||
628 	    IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) ||
629 	    IN6_IS_ADDR_V4MAPPED(&sa6->sin6_addr))
630 		return;
631 	if ((unsigned)cmd >= PRC_NCMDS)
632 		return;
633 	else if (cmd == PRC_QUENCH) {
634 		/*
635 		 * Don't honor ICMP Source Quench messages meant for
636 		 * TCP connections.
637 		 */
638 		/* XXX there's no PRC_QUENCH in IPv6 */
639 		return;
640 	} else if (PRC_IS_REDIRECT(cmd))
641 		notify = in_rtchange, d = NULL;
642 	else if (cmd == PRC_MSGSIZE)
643 		; /* special code is present, see below */
644 	else if (cmd == PRC_HOSTDEAD)
645 		d = NULL;
646 	else if (inet6ctlerrmap[cmd] == 0)
647 		return;
648 
649 	/* if the parameter is from icmp6, decode it. */
650 	if (d != NULL) {
651 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
652 		m = ip6cp->ip6c_m;
653 		ip6 = ip6cp->ip6c_ip6;
654 		off = ip6cp->ip6c_off;
655 		sa6_src = ip6cp->ip6c_src;
656 	} else {
657 		m = NULL;
658 		ip6 = NULL;
659 		sa6_src = &sa6_any;
660 	}
661 
662 	if (ip6) {
663 		/*
664 		 * XXX: We assume that when ip6 is non NULL,
665 		 * M and OFF are valid.
666 		 */
667 
668 		/* check if we can safely examine src and dst ports */
669 		if (m->m_pkthdr.len < off + sizeof(*thp))
670 			return;
671 
672 		bzero(&th, sizeof(th));
673 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
674 
675 		/*
676 		 * Check to see if we have a valid TCP connection
677 		 * corresponding to the address in the ICMPv6 message
678 		 * payload.
679 		 */
680 		inp = in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
681 		    th.th_dport, &sa6_src->sin6_addr, th.th_sport, rdomain);
682 		if (cmd == PRC_MSGSIZE) {
683 			/*
684 			 * Depending on the value of "valid" and routing table
685 			 * size (mtudisc_{hi,lo}wat), we will:
686 			 * - recalcurate the new MTU and create the
687 			 *   corresponding routing entry, or
688 			 * - ignore the MTU change notification.
689 			 */
690 			icmp6_mtudisc_update((struct ip6ctlparam *)d, inp != NULL);
691 			return;
692 		}
693 		if (inp) {
694 			seq = ntohl(th.th_seq);
695 			if (inp->inp_socket &&
696 			    (tp = intotcpcb(inp)) &&
697 			    SEQ_GEQ(seq, tp->snd_una) &&
698 			    SEQ_LT(seq, tp->snd_max))
699 				notify(inp, inet6ctlerrmap[cmd]);
700 		} else if (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
701 		    inet6ctlerrmap[cmd] == ENETUNREACH ||
702 		    inet6ctlerrmap[cmd] == EHOSTDOWN)
703 			syn_cache_unreach((struct sockaddr *)sa6_src,
704 			    sa, &th, rdomain);
705 	} else {
706 		(void) in6_pcbnotify(&tcbtable, sa6, 0,
707 		    sa6_src, 0, rdomain, cmd, NULL, notify);
708 	}
709 }
710 #endif
711 
712 void
713 tcp_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
714 {
715 	struct ip *ip = v;
716 	struct tcphdr *th;
717 	struct tcpcb *tp;
718 	struct inpcb *inp;
719 	struct in_addr faddr;
720 	tcp_seq seq;
721 	u_int mtu;
722 	void (*notify)(struct inpcb *, int) = tcp_notify;
723 	int errno;
724 
725 	if (sa->sa_family != AF_INET)
726 		return;
727 	faddr = satosin(sa)->sin_addr;
728 	if (faddr.s_addr == INADDR_ANY)
729 		return;
730 
731 	if ((unsigned)cmd >= PRC_NCMDS)
732 		return;
733 	errno = inetctlerrmap[cmd];
734 	if (cmd == PRC_QUENCH)
735 		/*
736 		 * Don't honor ICMP Source Quench messages meant for
737 		 * TCP connections.
738 		 */
739 		return;
740 	else if (PRC_IS_REDIRECT(cmd))
741 		notify = in_rtchange, ip = 0;
742 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip) {
743 		/*
744 		 * Verify that the packet in the icmp payload refers
745 		 * to an existing TCP connection.
746 		 */
747 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
748 		seq = ntohl(th->th_seq);
749 		inp = in_pcbhashlookup(&tcbtable,
750 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
751 		    rdomain);
752 		if (inp && (tp = intotcpcb(inp)) &&
753 		    SEQ_GEQ(seq, tp->snd_una) &&
754 		    SEQ_LT(seq, tp->snd_max)) {
755 			struct icmp *icp;
756 			icp = (struct icmp *)((caddr_t)ip -
757 					      offsetof(struct icmp, icmp_ip));
758 
759 			/*
760 			 * If the ICMP message advertises a Next-Hop MTU
761 			 * equal or larger than the maximum packet size we have
762 			 * ever sent, drop the message.
763 			 */
764 			mtu = (u_int)ntohs(icp->icmp_nextmtu);
765 			if (mtu >= tp->t_pmtud_mtu_sent)
766 				return;
767 			if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
768 				/*
769 				 * Calculate new MTU, and create corresponding
770 				 * route (traditional PMTUD).
771 				 */
772 				tp->t_flags &= ~TF_PMTUD_PEND;
773 				icmp_mtudisc(icp, inp->inp_rtableid);
774 			} else {
775 				/*
776 				 * Record the information got in the ICMP
777 				 * message; act on it later.
778 				 * If we had already recorded an ICMP message,
779 				 * replace the old one only if the new message
780 				 * refers to an older TCP segment
781 				 */
782 				if (tp->t_flags & TF_PMTUD_PEND) {
783 					if (SEQ_LT(tp->t_pmtud_th_seq, seq))
784 						return;
785 				} else
786 					tp->t_flags |= TF_PMTUD_PEND;
787 				tp->t_pmtud_th_seq = seq;
788 				tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
789 				tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
790 				tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
791 				return;
792 			}
793 		} else {
794 			/* ignore if we don't have a matching connection */
795 			return;
796 		}
797 		notify = tcp_mtudisc, ip = 0;
798 	} else if (cmd == PRC_MTUINC)
799 		notify = tcp_mtudisc_increase, ip = 0;
800 	else if (cmd == PRC_HOSTDEAD)
801 		ip = 0;
802 	else if (errno == 0)
803 		return;
804 
805 	if (ip) {
806 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
807 		inp = in_pcbhashlookup(&tcbtable,
808 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
809 		    rdomain);
810 		if (inp) {
811 			seq = ntohl(th->th_seq);
812 			if (inp->inp_socket &&
813 			    (tp = intotcpcb(inp)) &&
814 			    SEQ_GEQ(seq, tp->snd_una) &&
815 			    SEQ_LT(seq, tp->snd_max))
816 				notify(inp, errno);
817 		} else if (inetctlerrmap[cmd] == EHOSTUNREACH ||
818 		    inetctlerrmap[cmd] == ENETUNREACH ||
819 		    inetctlerrmap[cmd] == EHOSTDOWN) {
820 			struct sockaddr_in sin;
821 
822 			bzero(&sin, sizeof(sin));
823 			sin.sin_len = sizeof(sin);
824 			sin.sin_family = AF_INET;
825 			sin.sin_port = th->th_sport;
826 			sin.sin_addr = ip->ip_src;
827 			syn_cache_unreach(sintosa(&sin), sa, th, rdomain);
828 		}
829 	} else
830 		in_pcbnotifyall(&tcbtable, sa, rdomain, errno, notify);
831 }
832 
833 
834 #ifdef INET6
835 /*
836  * Path MTU Discovery handlers.
837  */
838 void
839 tcp6_mtudisc_callback(struct sockaddr_in6 *sin6, u_int rdomain)
840 {
841 	(void) in6_pcbnotify(&tcbtable, sin6, 0,
842 	    &sa6_any, 0, rdomain, PRC_MSGSIZE, NULL, tcp_mtudisc);
843 }
844 #endif /* INET6 */
845 
846 /*
847  * On receipt of path MTU corrections, flush old route and replace it
848  * with the new one.  Retransmit all unacknowledged packets, to ensure
849  * that all packets will be received.
850  */
851 void
852 tcp_mtudisc(struct inpcb *inp, int errno)
853 {
854 	struct tcpcb *tp = intotcpcb(inp);
855 	struct rtentry *rt = in_pcbrtentry(inp);
856 	int change = 0;
857 
858 	if (tp != 0) {
859 		int orig_maxseg = tp->t_maxseg;
860 		if (rt != 0) {
861 			/*
862 			 * If this was not a host route, remove and realloc.
863 			 */
864 			if ((rt->rt_flags & RTF_HOST) == 0) {
865 				in_rtchange(inp, errno);
866 				if ((rt = in_pcbrtentry(inp)) == 0)
867 					return;
868 			}
869 			if (orig_maxseg != tp->t_maxseg ||
870 			    (rt->rt_locks & RTV_MTU))
871 				change = 1;
872 		}
873 		tcp_mss(tp, -1);
874 
875 		/*
876 		 * Resend unacknowledged packets
877 		 */
878 		tp->snd_nxt = tp->snd_una;
879 		if (change || errno > 0)
880 			tcp_output(tp);
881 	}
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 int tcp_secret_init;
907 u_char tcp_secret[16];
908 SHA2_CTX tcp_secret_ctx;
909 
910 void
911 tcp_set_iss_tsm(struct tcpcb *tp)
912 {
913 	SHA2_CTX ctx;
914 	union {
915 		uint8_t bytes[SHA512_DIGEST_LENGTH];
916 		uint32_t words[2];
917 	} digest;
918 	u_int rdomain = rtable_l2(tp->t_inpcb->inp_rtableid);
919 
920 	if (tcp_secret_init == 0) {
921 		arc4random_buf(tcp_secret, sizeof(tcp_secret));
922 		SHA512Init(&tcp_secret_ctx);
923 		SHA512Update(&tcp_secret_ctx, tcp_secret, sizeof(tcp_secret));
924 		tcp_secret_init = 1;
925 	}
926 	ctx = tcp_secret_ctx;
927 	SHA512Update(&ctx, &rdomain, sizeof(rdomain));
928 	SHA512Update(&ctx, &tp->t_inpcb->inp_lport, sizeof(u_short));
929 	SHA512Update(&ctx, &tp->t_inpcb->inp_fport, sizeof(u_short));
930 	if (tp->pf == AF_INET6) {
931 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr6,
932 		    sizeof(struct in6_addr));
933 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr6,
934 		    sizeof(struct in6_addr));
935 	} else {
936 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr,
937 		    sizeof(struct in_addr));
938 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr,
939 		    sizeof(struct in_addr));
940 	}
941 	SHA512Final(digest.bytes, &ctx);
942 	tcp_iss += TCP_ISS_CONN_INC;
943 	tp->iss = digest.words[0] + tcp_iss;
944 	tp->ts_modulate = digest.words[1];
945 }
946 
947 #ifdef TCP_SIGNATURE
948 int
949 tcp_signature_tdb_attach(void)
950 {
951 	return (0);
952 }
953 
954 int
955 tcp_signature_tdb_init(struct tdb *tdbp, struct xformsw *xsp,
956     struct ipsecinit *ii)
957 {
958 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
959 		return (EINVAL);
960 
961 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_NOWAIT);
962 	if (tdbp->tdb_amxkey == NULL)
963 		return (ENOMEM);
964 	bcopy(ii->ii_authkey, tdbp->tdb_amxkey, ii->ii_authkeylen);
965 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
966 
967 	return (0);
968 }
969 
970 int
971 tcp_signature_tdb_zeroize(struct tdb *tdbp)
972 {
973 	if (tdbp->tdb_amxkey) {
974 		explicit_bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
975 		free(tdbp->tdb_amxkey, M_XDATA, 0);
976 		tdbp->tdb_amxkey = NULL;
977 	}
978 
979 	return (0);
980 }
981 
982 int
983 tcp_signature_tdb_input(struct mbuf *m, struct tdb *tdbp, int skip, int protoff)
984 {
985 	return (0);
986 }
987 
988 int
989 tcp_signature_tdb_output(struct mbuf *m, struct tdb *tdbp, struct mbuf **mp,
990     int skip, int protoff)
991 {
992 	return (EINVAL);
993 }
994 
995 int
996 tcp_signature_apply(caddr_t fstate, caddr_t data, unsigned int len)
997 {
998 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
999 	return 0;
1000 }
1001 
1002 int
1003 tcp_signature(struct tdb *tdb, int af, struct mbuf *m, struct tcphdr *th,
1004     int iphlen, int doswap, char *sig)
1005 {
1006 	MD5_CTX ctx;
1007 	int len;
1008 	struct tcphdr th0;
1009 
1010 	MD5Init(&ctx);
1011 
1012 	switch(af) {
1013 	case 0:
1014 	case AF_INET: {
1015 		struct ippseudo ippseudo;
1016 		struct ip *ip;
1017 
1018 		ip = mtod(m, struct ip *);
1019 
1020 		ippseudo.ippseudo_src = ip->ip_src;
1021 		ippseudo.ippseudo_dst = ip->ip_dst;
1022 		ippseudo.ippseudo_pad = 0;
1023 		ippseudo.ippseudo_p = IPPROTO_TCP;
1024 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - iphlen);
1025 
1026 		MD5Update(&ctx, (char *)&ippseudo,
1027 		    sizeof(struct ippseudo));
1028 		break;
1029 		}
1030 #ifdef INET6
1031 	case AF_INET6: {
1032 		struct ip6_hdr_pseudo ip6pseudo;
1033 		struct ip6_hdr *ip6;
1034 
1035 		ip6 = mtod(m, struct ip6_hdr *);
1036 		bzero(&ip6pseudo, sizeof(ip6pseudo));
1037 		ip6pseudo.ip6ph_src = ip6->ip6_src;
1038 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
1039 		in6_clearscope(&ip6pseudo.ip6ph_src);
1040 		in6_clearscope(&ip6pseudo.ip6ph_dst);
1041 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
1042 		ip6pseudo.ip6ph_len = htonl(m->m_pkthdr.len - iphlen);
1043 
1044 		MD5Update(&ctx, (char *)&ip6pseudo,
1045 		    sizeof(ip6pseudo));
1046 		break;
1047 		}
1048 #endif
1049 	}
1050 
1051 	th0 = *th;
1052 	th0.th_sum = 0;
1053 
1054 	if (doswap) {
1055 		th0.th_seq = htonl(th0.th_seq);
1056 		th0.th_ack = htonl(th0.th_ack);
1057 		th0.th_win = htons(th0.th_win);
1058 		th0.th_urp = htons(th0.th_urp);
1059 	}
1060 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
1061 
1062 	len = m->m_pkthdr.len - iphlen - th->th_off * sizeof(uint32_t);
1063 
1064 	if (len > 0 &&
1065 	    m_apply(m, iphlen + th->th_off * sizeof(uint32_t), len,
1066 	    tcp_signature_apply, (caddr_t)&ctx))
1067 		return (-1);
1068 
1069 	MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen);
1070 	MD5Final(sig, &ctx);
1071 
1072 	return (0);
1073 }
1074 #endif /* TCP_SIGNATURE */
1075