xref: /openbsd-src/sys/netinet/tcp_subr.c (revision c7e8ea31cd41a963f06f0a8ba93948b06aa6b4a4)
1 /*	$OpenBSD: tcp_subr.c,v 1.165 2017/06/26 09:32:32 mpi 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 #ifdef TCP_SACK
108 int	tcp_do_sack = 1;	/* RFC 2018 selective ACKs */
109 #endif
110 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
111 #ifdef TCP_ECN
112 int	tcp_do_ecn = 0;		/* RFC3168 ECN enabled/disabled? */
113 #endif
114 int	tcp_do_rfc3390 = 2;	/* Increase TCP's Initial Window to 10*mss */
115 
116 u_int32_t	tcp_now = 1;
117 
118 #ifndef TCB_INITIAL_HASH_SIZE
119 #define	TCB_INITIAL_HASH_SIZE	128
120 #endif
121 
122 int tcp_reass_limit = NMBCLUSTERS / 8; /* hardlimit for tcpqe_pool */
123 #ifdef TCP_SACK
124 int tcp_sackhole_limit = 32*1024; /* hardlimit for sackhl_pool */
125 #endif
126 
127 struct pool tcpcb_pool;
128 struct pool tcpqe_pool;
129 #ifdef TCP_SACK
130 struct pool sackhl_pool;
131 #endif
132 
133 struct cpumem *tcpcounters;		/* tcp statistics */
134 tcp_seq  tcp_iss;
135 
136 /*
137  * Tcp initialization
138  */
139 void
140 tcp_init(void)
141 {
142 	tcp_iss = 1;		/* wrong */
143 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, IPL_SOFTNET, 0,
144 	    "tcpcb", NULL);
145 	pool_init(&tcpqe_pool, sizeof(struct tcpqent), 0, IPL_SOFTNET, 0,
146 	    "tcpqe", NULL);
147 	pool_sethardlimit(&tcpqe_pool, tcp_reass_limit, NULL, 0);
148 #ifdef TCP_SACK
149 	pool_init(&sackhl_pool, sizeof(struct sackhole), 0, IPL_SOFTNET, 0,
150 	    "sackhl", NULL);
151 	pool_sethardlimit(&sackhl_pool, tcp_sackhole_limit, NULL, 0);
152 #endif /* TCP_SACK */
153 	in_pcbinit(&tcbtable, TCB_INITIAL_HASH_SIZE);
154 	tcpcounters = counters_alloc(tcps_ncounters);
155 
156 #ifdef INET6
157 	/*
158 	 * Since sizeof(struct ip6_hdr) > sizeof(struct ip), we
159 	 * do max length checks/computations only on the former.
160 	 */
161 	if (max_protohdr < (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)))
162 		max_protohdr = (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
163 	if ((max_linkhdr + sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) >
164 	    MHLEN)
165 		panic("tcp_init");
166 
167 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
168 #endif /* INET6 */
169 
170 	/* Initialize the compressed state engine. */
171 	syn_cache_init();
172 
173 	/* Initialize timer state. */
174 	tcp_timer_init();
175 }
176 
177 /*
178  * Create template to be used to send tcp packets on a connection.
179  * Call after host entry created, allocates an mbuf and fills
180  * in a skeletal tcp/ip header, minimizing the amount of work
181  * necessary when the connection is used.
182  *
183  * To support IPv6 in addition to IPv4 and considering that the sizes of
184  * the IPv4 and IPv6 headers are not the same, we now use a separate pointer
185  * for the TCP header.  Also, we made the former tcpiphdr header pointer
186  * into just an IP overlay pointer, with casting as appropriate for v6. rja
187  */
188 struct mbuf *
189 tcp_template(struct tcpcb *tp)
190 {
191 	struct inpcb *inp = tp->t_inpcb;
192 	struct mbuf *m;
193 	struct tcphdr *th;
194 
195 	if ((m = tp->t_template) == 0) {
196 		m = m_get(M_DONTWAIT, MT_HEADER);
197 		if (m == NULL)
198 			return (0);
199 
200 		switch (tp->pf) {
201 		case 0:	/*default to PF_INET*/
202 		case AF_INET:
203 			m->m_len = sizeof(struct ip);
204 			break;
205 #ifdef INET6
206 		case AF_INET6:
207 			m->m_len = sizeof(struct ip6_hdr);
208 			break;
209 #endif /* INET6 */
210 		}
211 		m->m_len += sizeof (struct tcphdr);
212 
213 		/*
214 		 * The link header, network header, TCP header, and TCP options
215 		 * all must fit in this mbuf. For now, assume the worst case of
216 		 * TCP options size. Eventually, compute this from tp flags.
217 		 */
218 		if (m->m_len + MAX_TCPOPTLEN + max_linkhdr >= MHLEN) {
219 			MCLGET(m, M_DONTWAIT);
220 			if ((m->m_flags & M_EXT) == 0) {
221 				m_free(m);
222 				return (0);
223 			}
224 		}
225 	}
226 
227 	switch(tp->pf) {
228 	case AF_INET:
229 		{
230 			struct ipovly *ipovly;
231 
232 			ipovly = mtod(m, struct ipovly *);
233 
234 			bzero(ipovly->ih_x1, sizeof ipovly->ih_x1);
235 			ipovly->ih_pr = IPPROTO_TCP;
236 			ipovly->ih_len = htons(sizeof (struct tcphdr));
237 			ipovly->ih_src = inp->inp_laddr;
238 			ipovly->ih_dst = inp->inp_faddr;
239 
240 			th = (struct tcphdr *)(mtod(m, caddr_t) +
241 				sizeof(struct ip));
242 		}
243 		break;
244 #ifdef INET6
245 	case AF_INET6:
246 		{
247 			struct ip6_hdr *ip6;
248 
249 			ip6 = mtod(m, struct ip6_hdr *);
250 
251 			ip6->ip6_src = inp->inp_laddr6;
252 			ip6->ip6_dst = inp->inp_faddr6;
253 			ip6->ip6_flow = htonl(0x60000000) |
254 			    (inp->inp_flowinfo & IPV6_FLOWLABEL_MASK);
255 
256 			ip6->ip6_nxt = IPPROTO_TCP;
257 			ip6->ip6_plen = htons(sizeof(struct tcphdr)); /*XXX*/
258 			ip6->ip6_hlim = in6_selecthlim(inp);	/*XXX*/
259 
260 			th = (struct tcphdr *)(mtod(m, caddr_t) +
261 				sizeof(struct ip6_hdr));
262 		}
263 		break;
264 #endif /* INET6 */
265 	}
266 
267 	th->th_sport = inp->inp_lport;
268 	th->th_dport = inp->inp_fport;
269 	th->th_seq = 0;
270 	th->th_ack = 0;
271 	th->th_x2  = 0;
272 	th->th_off = 5;
273 	th->th_flags = 0;
274 	th->th_win = 0;
275 	th->th_urp = 0;
276 	th->th_sum = 0;
277 	return (m);
278 }
279 
280 /*
281  * Send a single message to the TCP at address specified by
282  * the given TCP/IP header.  If m == 0, then we make a copy
283  * of the tcpiphdr at ti and send directly to the addressed host.
284  * This is used to force keep alive messages out using the TCP
285  * template for a connection tp->t_template.  If flags are given
286  * then we send a message back to the TCP which originated the
287  * segment ti, and discard the mbuf containing it and any other
288  * attached mbufs.
289  *
290  * In any case the ack and sequence number of the transmitted
291  * segment are as specified by the parameters.
292  */
293 void
294 tcp_respond(struct tcpcb *tp, caddr_t template, struct tcphdr *th0,
295     tcp_seq ack, tcp_seq seq, int flags, u_int rtableid)
296 {
297 	int tlen;
298 	int win = 0;
299 	struct mbuf *m = NULL;
300 	struct tcphdr *th;
301 	struct ip *ip;
302 #ifdef INET6
303 	struct ip6_hdr *ip6;
304 #endif
305 	int af;		/* af on wire */
306 
307 	if (tp) {
308 		struct socket *so = tp->t_inpcb->inp_socket;
309 		win = sbspace(so, &so->so_rcv);
310 		/*
311 		 * If this is called with an unconnected
312 		 * socket/tp/pcb (tp->pf is 0), we lose.
313 		 */
314 		af = tp->pf;
315 	} else
316 		af = (((struct ip *)template)->ip_v == 6) ? AF_INET6 : AF_INET;
317 
318 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
319 	if (m == NULL)
320 		return;
321 	m->m_data += max_linkhdr;
322 	tlen = 0;
323 
324 #define xchg(a,b,type) do { type t; t=a; a=b; b=t; } while (0)
325 	switch (af) {
326 #ifdef INET6
327 	case AF_INET6:
328 		ip6 = mtod(m, struct ip6_hdr *);
329 		th = (struct tcphdr *)(ip6 + 1);
330 		tlen = sizeof(*ip6) + sizeof(*th);
331 		if (th0) {
332 			bcopy(template, ip6, sizeof(*ip6));
333 			bcopy(th0, th, sizeof(*th));
334 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
335 		} else {
336 			bcopy(template, ip6, tlen);
337 		}
338 		break;
339 #endif /* INET6 */
340 	case AF_INET:
341 		ip = mtod(m, struct ip *);
342 		th = (struct tcphdr *)(ip + 1);
343 		tlen = sizeof(*ip) + sizeof(*th);
344 		if (th0) {
345 			bcopy(template, ip, sizeof(*ip));
346 			bcopy(th0, th, sizeof(*th));
347 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, u_int32_t);
348 		} else {
349 			bcopy(template, ip, tlen);
350 		}
351 		break;
352 	}
353 	if (th0)
354 		xchg(th->th_dport, th->th_sport, u_int16_t);
355 	else
356 		flags = TH_ACK;
357 #undef xchg
358 
359 	th->th_seq = htonl(seq);
360 	th->th_ack = htonl(ack);
361 	th->th_x2 = 0;
362 	th->th_off = sizeof (struct tcphdr) >> 2;
363 	th->th_flags = flags;
364 	if (tp)
365 		win >>= tp->rcv_scale;
366 	if (win > TCP_MAXWIN)
367 		win = TCP_MAXWIN;
368 	th->th_win = htons((u_int16_t)win);
369 	th->th_urp = 0;
370 
371 	if (tp && (tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
372 	    (flags & TH_RST) == 0 && (tp->t_flags & TF_RCVD_TSTMP)) {
373 		u_int32_t *lp = (u_int32_t *)(th + 1);
374 		/* Form timestamp option as shown in appendix A of RFC 1323. */
375 		*lp++ = htonl(TCPOPT_TSTAMP_HDR);
376 		*lp++ = htonl(tcp_now + tp->ts_modulate);
377 		*lp   = htonl(tp->ts_recent);
378 		tlen += TCPOLEN_TSTAMP_APPA;
379 		th->th_off = (sizeof(struct tcphdr) + TCPOLEN_TSTAMP_APPA) >> 2;
380 	}
381 
382 	m->m_len = tlen;
383 	m->m_pkthdr.len = tlen;
384 	m->m_pkthdr.ph_ifidx = 0;
385 	m->m_pkthdr.csum_flags |= M_TCP_CSUM_OUT;
386 
387 	/* force routing table */
388 	if (tp)
389 		m->m_pkthdr.ph_rtableid = tp->t_inpcb->inp_rtableid;
390 	else
391 		m->m_pkthdr.ph_rtableid = rtableid;
392 
393 	switch (af) {
394 #ifdef INET6
395 	case AF_INET6:
396 		ip6->ip6_flow = htonl(0x60000000);
397 		ip6->ip6_nxt  = IPPROTO_TCP;
398 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL);	/*XXX*/
399 		ip6->ip6_plen = tlen - sizeof(struct ip6_hdr);
400 		ip6->ip6_plen = htons(ip6->ip6_plen);
401 		ip6_output(m, tp ? tp->t_inpcb->inp_outputopts6 : NULL,
402 		    tp ? &tp->t_inpcb->inp_route6 : NULL,
403 		    0, NULL,
404 		    tp ? tp->t_inpcb : NULL);
405 		break;
406 #endif /* INET6 */
407 	case AF_INET:
408 		ip->ip_len = htons(tlen);
409 		ip->ip_ttl = ip_defttl;
410 		ip->ip_tos = 0;
411 		ip_output(m, NULL,
412 		    tp ? &tp->t_inpcb->inp_route : NULL,
413 		    ip_mtudisc ? IP_MTUDISC : 0, NULL,
414 		    tp ? tp->t_inpcb : NULL, 0);
415 		break;
416 	}
417 }
418 
419 /*
420  * Create a new TCP control block, making an
421  * empty reassembly queue and hooking it to the argument
422  * protocol control block.
423  */
424 struct tcpcb *
425 tcp_newtcpcb(struct inpcb *inp)
426 {
427 	struct tcpcb *tp;
428 	int i;
429 
430 	tp = pool_get(&tcpcb_pool, PR_NOWAIT|PR_ZERO);
431 	if (tp == NULL)
432 		return (NULL);
433 	TAILQ_INIT(&tp->t_segq);
434 	tp->t_maxseg = tcp_mssdflt;
435 	tp->t_maxopd = 0;
436 
437 	TCP_INIT_DELACK(tp);
438 	for (i = 0; i < TCPT_NTIMERS; i++)
439 		TCP_TIMER_INIT(tp, i);
440 	timeout_set(&tp->t_reap_to, tcp_reaper, tp);
441 
442 #ifdef TCP_SACK
443 	tp->sack_enable = tcp_do_sack;
444 #endif
445 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
446 	tp->t_inpcb = inp;
447 	/*
448 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
449 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
450 	 * reasonable initial retransmit time.
451 	 */
452 	tp->t_srtt = TCPTV_SRTTBASE;
453 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ <<
454 	    (TCP_RTTVAR_SHIFT + TCP_RTT_BASE_SHIFT - 1);
455 	tp->t_rttmin = TCPTV_MIN;
456 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
457 	    TCPTV_MIN, TCPTV_REXMTMAX);
458 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
459 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
460 
461 	tp->t_pmtud_mtu_sent = 0;
462 	tp->t_pmtud_mss_acked = 0;
463 
464 #ifdef INET6
465 	/* we disallow IPv4 mapped address completely. */
466 	if ((inp->inp_flags & INP_IPV6) == 0)
467 		tp->pf = PF_INET;
468 	else
469 		tp->pf = PF_INET6;
470 #else
471 	tp->pf = PF_INET;
472 #endif
473 
474 #ifdef INET6
475 	if (inp->inp_flags & INP_IPV6)
476 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
477 	else
478 #endif /* INET6 */
479 		inp->inp_ip.ip_ttl = ip_defttl;
480 
481 	inp->inp_ppcb = (caddr_t)tp;
482 	return (tp);
483 }
484 
485 /*
486  * Drop a TCP connection, reporting
487  * the specified error.  If connection is synchronized,
488  * then send a RST to peer.
489  */
490 struct tcpcb *
491 tcp_drop(struct tcpcb *tp, int errno)
492 {
493 	struct socket *so = tp->t_inpcb->inp_socket;
494 
495 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
496 		tp->t_state = TCPS_CLOSED;
497 		(void) tcp_output(tp);
498 		tcpstat_inc(tcps_drops);
499 	} else
500 		tcpstat_inc(tcps_conndrops);
501 	if (errno == ETIMEDOUT && tp->t_softerror)
502 		errno = tp->t_softerror;
503 	so->so_error = errno;
504 	return (tcp_close(tp));
505 }
506 
507 /*
508  * Close a TCP control block:
509  *	discard all space held by the tcp
510  *	discard internet protocol block
511  *	wake up any sleepers
512  */
513 struct tcpcb *
514 tcp_close(struct tcpcb *tp)
515 {
516 	struct inpcb *inp = tp->t_inpcb;
517 	struct socket *so = inp->inp_socket;
518 #ifdef TCP_SACK
519 	struct sackhole *p, *q;
520 #endif
521 
522 	/* free the reassembly queue, if any */
523 	tcp_freeq(tp);
524 
525 	tcp_canceltimers(tp);
526 	TCP_CLEAR_DELACK(tp);
527 	syn_cache_cleanup(tp);
528 
529 #ifdef TCP_SACK
530 	/* Free SACK holes. */
531 	q = p = tp->snd_holes;
532 	while (p != 0) {
533 		q = p->next;
534 		pool_put(&sackhl_pool, p);
535 		p = q;
536 	}
537 #endif
538 	m_free(tp->t_template);
539 
540 	tp->t_flags |= TF_DEAD;
541 	timeout_add(&tp->t_reap_to, 0);
542 
543 	inp->inp_ppcb = 0;
544 	soisdisconnected(so);
545 	in_pcbdetach(inp);
546 	return (NULL);
547 }
548 
549 void
550 tcp_reaper(void *arg)
551 {
552 	struct tcpcb *tp = arg;
553 
554 	pool_put(&tcpcb_pool, tp);
555 	tcpstat_inc(tcps_closed);
556 }
557 
558 int
559 tcp_freeq(struct tcpcb *tp)
560 {
561 	struct tcpqent *qe;
562 	int rv = 0;
563 
564 	while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) {
565 		TAILQ_REMOVE(&tp->t_segq, qe, tcpqe_q);
566 		m_freem(qe->tcpqe_m);
567 		pool_put(&tcpqe_pool, qe);
568 		rv = 1;
569 	}
570 	return (rv);
571 }
572 
573 /*
574  * Compute proper scaling value for receiver window from buffer space
575  */
576 
577 void
578 tcp_rscale(struct tcpcb *tp, u_long hiwat)
579 {
580 	tp->request_r_scale = 0;
581 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
582 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
583 		tp->request_r_scale++;
584 }
585 
586 /*
587  * Notify a tcp user of an asynchronous error;
588  * store error as soft error, but wake up user
589  * (for now, won't do anything until can select for soft error).
590  */
591 void
592 tcp_notify(struct inpcb *inp, int error)
593 {
594 	struct tcpcb *tp = intotcpcb(inp);
595 	struct socket *so = inp->inp_socket;
596 
597 	/*
598 	 * Ignore some errors if we are hooked up.
599 	 * If connection hasn't completed, has retransmitted several times,
600 	 * and receives a second error, give up now.  This is better
601 	 * than waiting a long time to establish a connection that
602 	 * can never complete.
603 	 */
604 	if (tp->t_state == TCPS_ESTABLISHED &&
605 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
606 	      error == EHOSTDOWN)) {
607 		return;
608 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
609 	    tp->t_rxtshift > 3 && tp->t_softerror)
610 		so->so_error = error;
611 	else
612 		tp->t_softerror = error;
613 	wakeup((caddr_t) &so->so_timeo);
614 	sorwakeup(so);
615 	sowwakeup(so);
616 }
617 
618 #ifdef INET6
619 void
620 tcp6_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *d)
621 {
622 	struct tcphdr th;
623 	struct tcpcb *tp;
624 	void (*notify)(struct inpcb *, int) = tcp_notify;
625 	struct ip6_hdr *ip6;
626 	const struct sockaddr_in6 *sa6_src = NULL;
627 	struct sockaddr_in6 *sa6 = satosin6(sa);
628 	struct inpcb *inp;
629 	struct mbuf *m;
630 	tcp_seq seq;
631 	int off;
632 	struct {
633 		u_int16_t th_sport;
634 		u_int16_t th_dport;
635 		u_int32_t th_seq;
636 	} *thp;
637 
638 	CTASSERT(sizeof(*thp) <= sizeof(th));
639 	if (sa->sa_family != AF_INET6 ||
640 	    sa->sa_len != sizeof(struct sockaddr_in6) ||
641 	    IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) ||
642 	    IN6_IS_ADDR_V4MAPPED(&sa6->sin6_addr))
643 		return;
644 	if ((unsigned)cmd >= PRC_NCMDS)
645 		return;
646 	else if (cmd == PRC_QUENCH) {
647 		/*
648 		 * Don't honor ICMP Source Quench messages meant for
649 		 * TCP connections.
650 		 */
651 		/* XXX there's no PRC_QUENCH in IPv6 */
652 		return;
653 	} else if (PRC_IS_REDIRECT(cmd))
654 		notify = in_rtchange, d = NULL;
655 	else if (cmd == PRC_MSGSIZE)
656 		; /* special code is present, see below */
657 	else if (cmd == PRC_HOSTDEAD)
658 		d = NULL;
659 	else if (inet6ctlerrmap[cmd] == 0)
660 		return;
661 
662 	/* if the parameter is from icmp6, decode it. */
663 	if (d != NULL) {
664 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
665 		m = ip6cp->ip6c_m;
666 		ip6 = ip6cp->ip6c_ip6;
667 		off = ip6cp->ip6c_off;
668 		sa6_src = ip6cp->ip6c_src;
669 	} else {
670 		m = NULL;
671 		ip6 = NULL;
672 		sa6_src = &sa6_any;
673 	}
674 
675 	if (ip6) {
676 		/*
677 		 * XXX: We assume that when ip6 is non NULL,
678 		 * M and OFF are valid.
679 		 */
680 
681 		/* check if we can safely examine src and dst ports */
682 		if (m->m_pkthdr.len < off + sizeof(*thp))
683 			return;
684 
685 		bzero(&th, sizeof(th));
686 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
687 
688 		/*
689 		 * Check to see if we have a valid TCP connection
690 		 * corresponding to the address in the ICMPv6 message
691 		 * payload.
692 		 */
693 		inp = in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
694 		    th.th_dport, &sa6_src->sin6_addr, th.th_sport, rdomain);
695 		if (cmd == PRC_MSGSIZE) {
696 			/*
697 			 * Depending on the value of "valid" and routing table
698 			 * size (mtudisc_{hi,lo}wat), we will:
699 			 * - recalcurate the new MTU and create the
700 			 *   corresponding routing entry, or
701 			 * - ignore the MTU change notification.
702 			 */
703 			icmp6_mtudisc_update((struct ip6ctlparam *)d, inp != NULL);
704 			return;
705 		}
706 		if (inp) {
707 			seq = ntohl(th.th_seq);
708 			if (inp->inp_socket &&
709 			    (tp = intotcpcb(inp)) &&
710 			    SEQ_GEQ(seq, tp->snd_una) &&
711 			    SEQ_LT(seq, tp->snd_max))
712 				notify(inp, inet6ctlerrmap[cmd]);
713 		} else if (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
714 		    inet6ctlerrmap[cmd] == ENETUNREACH ||
715 		    inet6ctlerrmap[cmd] == EHOSTDOWN)
716 			syn_cache_unreach((struct sockaddr *)sa6_src,
717 			    sa, &th, rdomain);
718 	} else {
719 		(void) in6_pcbnotify(&tcbtable, sa6, 0,
720 		    sa6_src, 0, rdomain, cmd, NULL, notify);
721 	}
722 }
723 #endif
724 
725 void
726 tcp_ctlinput(int cmd, struct sockaddr *sa, u_int rdomain, void *v)
727 {
728 	struct ip *ip = v;
729 	struct tcphdr *th;
730 	struct tcpcb *tp;
731 	struct inpcb *inp;
732 	struct in_addr faddr;
733 	tcp_seq seq;
734 	u_int mtu;
735 	void (*notify)(struct inpcb *, int) = tcp_notify;
736 	int errno;
737 
738 	if (sa->sa_family != AF_INET)
739 		return;
740 	faddr = satosin(sa)->sin_addr;
741 	if (faddr.s_addr == INADDR_ANY)
742 		return;
743 
744 	if ((unsigned)cmd >= PRC_NCMDS)
745 		return;
746 	errno = inetctlerrmap[cmd];
747 	if (cmd == PRC_QUENCH)
748 		/*
749 		 * Don't honor ICMP Source Quench messages meant for
750 		 * TCP connections.
751 		 */
752 		return;
753 	else if (PRC_IS_REDIRECT(cmd))
754 		notify = in_rtchange, ip = 0;
755 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip) {
756 		/*
757 		 * Verify that the packet in the icmp payload refers
758 		 * to an existing TCP connection.
759 		 */
760 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
761 		seq = ntohl(th->th_seq);
762 		inp = in_pcbhashlookup(&tcbtable,
763 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
764 		    rdomain);
765 		if (inp && (tp = intotcpcb(inp)) &&
766 		    SEQ_GEQ(seq, tp->snd_una) &&
767 		    SEQ_LT(seq, tp->snd_max)) {
768 			struct icmp *icp;
769 			icp = (struct icmp *)((caddr_t)ip -
770 					      offsetof(struct icmp, icmp_ip));
771 
772 			/*
773 			 * If the ICMP message advertises a Next-Hop MTU
774 			 * equal or larger than the maximum packet size we have
775 			 * ever sent, drop the message.
776 			 */
777 			mtu = (u_int)ntohs(icp->icmp_nextmtu);
778 			if (mtu >= tp->t_pmtud_mtu_sent)
779 				return;
780 			if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
781 				/*
782 				 * Calculate new MTU, and create corresponding
783 				 * route (traditional PMTUD).
784 				 */
785 				tp->t_flags &= ~TF_PMTUD_PEND;
786 				icmp_mtudisc(icp, inp->inp_rtableid);
787 			} else {
788 				/*
789 				 * Record the information got in the ICMP
790 				 * message; act on it later.
791 				 * If we had already recorded an ICMP message,
792 				 * replace the old one only if the new message
793 				 * refers to an older TCP segment
794 				 */
795 				if (tp->t_flags & TF_PMTUD_PEND) {
796 					if (SEQ_LT(tp->t_pmtud_th_seq, seq))
797 						return;
798 				} else
799 					tp->t_flags |= TF_PMTUD_PEND;
800 				tp->t_pmtud_th_seq = seq;
801 				tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
802 				tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
803 				tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
804 				return;
805 			}
806 		} else {
807 			/* ignore if we don't have a matching connection */
808 			return;
809 		}
810 		notify = tcp_mtudisc, ip = 0;
811 	} else if (cmd == PRC_MTUINC)
812 		notify = tcp_mtudisc_increase, ip = 0;
813 	else if (cmd == PRC_HOSTDEAD)
814 		ip = 0;
815 	else if (errno == 0)
816 		return;
817 
818 	if (ip) {
819 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
820 		inp = in_pcbhashlookup(&tcbtable,
821 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport,
822 		    rdomain);
823 		if (inp) {
824 			seq = ntohl(th->th_seq);
825 			if (inp->inp_socket &&
826 			    (tp = intotcpcb(inp)) &&
827 			    SEQ_GEQ(seq, tp->snd_una) &&
828 			    SEQ_LT(seq, tp->snd_max))
829 				notify(inp, errno);
830 		} else if (inetctlerrmap[cmd] == EHOSTUNREACH ||
831 		    inetctlerrmap[cmd] == ENETUNREACH ||
832 		    inetctlerrmap[cmd] == EHOSTDOWN) {
833 			struct sockaddr_in sin;
834 
835 			bzero(&sin, sizeof(sin));
836 			sin.sin_len = sizeof(sin);
837 			sin.sin_family = AF_INET;
838 			sin.sin_port = th->th_sport;
839 			sin.sin_addr = ip->ip_src;
840 			syn_cache_unreach(sintosa(&sin), sa, th, rdomain);
841 		}
842 	} else
843 		in_pcbnotifyall(&tcbtable, sa, rdomain, errno, notify);
844 }
845 
846 
847 #ifdef INET6
848 /*
849  * Path MTU Discovery handlers.
850  */
851 void
852 tcp6_mtudisc_callback(struct sockaddr_in6 *sin6, u_int rdomain)
853 {
854 	(void) in6_pcbnotify(&tcbtable, sin6, 0,
855 	    &sa6_any, 0, rdomain, PRC_MSGSIZE, NULL, tcp_mtudisc);
856 }
857 #endif /* INET6 */
858 
859 /*
860  * On receipt of path MTU corrections, flush old route and replace it
861  * with the new one.  Retransmit all unacknowledged packets, to ensure
862  * that all packets will be received.
863  */
864 void
865 tcp_mtudisc(struct inpcb *inp, int errno)
866 {
867 	struct tcpcb *tp = intotcpcb(inp);
868 	struct rtentry *rt = in_pcbrtentry(inp);
869 	int change = 0;
870 
871 	if (tp != 0) {
872 		int orig_maxseg = tp->t_maxseg;
873 		if (rt != 0) {
874 			/*
875 			 * If this was not a host route, remove and realloc.
876 			 */
877 			if ((rt->rt_flags & RTF_HOST) == 0) {
878 				in_rtchange(inp, errno);
879 				if ((rt = in_pcbrtentry(inp)) == 0)
880 					return;
881 			}
882 			if (orig_maxseg != tp->t_maxseg ||
883 			    (rt->rt_locks & RTV_MTU))
884 				change = 1;
885 		}
886 		tcp_mss(tp, -1);
887 
888 		/*
889 		 * Resend unacknowledged packets
890 		 */
891 		tp->snd_nxt = tp->snd_una;
892 		if (change || errno > 0)
893 			tcp_output(tp);
894 	}
895 }
896 
897 void
898 tcp_mtudisc_increase(struct inpcb *inp, int errno)
899 {
900 	struct tcpcb *tp = intotcpcb(inp);
901 	struct rtentry *rt = in_pcbrtentry(inp);
902 
903 	if (tp != 0 && rt != 0) {
904 		/*
905 		 * If this was a host route, remove and realloc.
906 		 */
907 		if (rt->rt_flags & RTF_HOST)
908 			in_rtchange(inp, errno);
909 
910 		/* also takes care of congestion window */
911 		tcp_mss(tp, -1);
912 	}
913 }
914 
915 /*
916  * Generate new ISNs with a method based on RFC1948
917  */
918 #define TCP_ISS_CONN_INC 4096
919 int tcp_secret_init;
920 u_char tcp_secret[16];
921 SHA2_CTX tcp_secret_ctx;
922 
923 void
924 tcp_set_iss_tsm(struct tcpcb *tp)
925 {
926 	SHA2_CTX ctx;
927 	union {
928 		uint8_t bytes[SHA512_DIGEST_LENGTH];
929 		uint32_t words[2];
930 	} digest;
931 	u_int rdomain = rtable_l2(tp->t_inpcb->inp_rtableid);
932 
933 	if (tcp_secret_init == 0) {
934 		arc4random_buf(tcp_secret, sizeof(tcp_secret));
935 		SHA512Init(&tcp_secret_ctx);
936 		SHA512Update(&tcp_secret_ctx, tcp_secret, sizeof(tcp_secret));
937 		tcp_secret_init = 1;
938 	}
939 	ctx = tcp_secret_ctx;
940 	SHA512Update(&ctx, &rdomain, sizeof(rdomain));
941 	SHA512Update(&ctx, &tp->t_inpcb->inp_lport, sizeof(u_short));
942 	SHA512Update(&ctx, &tp->t_inpcb->inp_fport, sizeof(u_short));
943 	if (tp->pf == AF_INET6) {
944 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr6,
945 		    sizeof(struct in6_addr));
946 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr6,
947 		    sizeof(struct in6_addr));
948 	} else {
949 		SHA512Update(&ctx, &tp->t_inpcb->inp_laddr,
950 		    sizeof(struct in_addr));
951 		SHA512Update(&ctx, &tp->t_inpcb->inp_faddr,
952 		    sizeof(struct in_addr));
953 	}
954 	SHA512Final(digest.bytes, &ctx);
955 	tcp_iss += TCP_ISS_CONN_INC;
956 	tp->iss = digest.words[0] + tcp_iss;
957 	tp->ts_modulate = digest.words[1];
958 }
959 
960 #ifdef TCP_SIGNATURE
961 int
962 tcp_signature_tdb_attach(void)
963 {
964 	return (0);
965 }
966 
967 int
968 tcp_signature_tdb_init(struct tdb *tdbp, struct xformsw *xsp,
969     struct ipsecinit *ii)
970 {
971 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
972 		return (EINVAL);
973 
974 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_NOWAIT);
975 	if (tdbp->tdb_amxkey == NULL)
976 		return (ENOMEM);
977 	bcopy(ii->ii_authkey, tdbp->tdb_amxkey, ii->ii_authkeylen);
978 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
979 
980 	return (0);
981 }
982 
983 int
984 tcp_signature_tdb_zeroize(struct tdb *tdbp)
985 {
986 	if (tdbp->tdb_amxkey) {
987 		explicit_bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
988 		free(tdbp->tdb_amxkey, M_XDATA, 0);
989 		tdbp->tdb_amxkey = NULL;
990 	}
991 
992 	return (0);
993 }
994 
995 int
996 tcp_signature_tdb_input(struct mbuf *m, struct tdb *tdbp, int skip, int protoff)
997 {
998 	return (0);
999 }
1000 
1001 int
1002 tcp_signature_tdb_output(struct mbuf *m, struct tdb *tdbp, struct mbuf **mp,
1003     int skip, int protoff)
1004 {
1005 	return (EINVAL);
1006 }
1007 
1008 int
1009 tcp_signature_apply(caddr_t fstate, caddr_t data, unsigned int len)
1010 {
1011 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
1012 	return 0;
1013 }
1014 
1015 int
1016 tcp_signature(struct tdb *tdb, int af, struct mbuf *m, struct tcphdr *th,
1017     int iphlen, int doswap, char *sig)
1018 {
1019 	MD5_CTX ctx;
1020 	int len;
1021 	struct tcphdr th0;
1022 
1023 	MD5Init(&ctx);
1024 
1025 	switch(af) {
1026 	case 0:
1027 	case AF_INET: {
1028 		struct ippseudo ippseudo;
1029 		struct ip *ip;
1030 
1031 		ip = mtod(m, struct ip *);
1032 
1033 		ippseudo.ippseudo_src = ip->ip_src;
1034 		ippseudo.ippseudo_dst = ip->ip_dst;
1035 		ippseudo.ippseudo_pad = 0;
1036 		ippseudo.ippseudo_p = IPPROTO_TCP;
1037 		ippseudo.ippseudo_len = htons(m->m_pkthdr.len - iphlen);
1038 
1039 		MD5Update(&ctx, (char *)&ippseudo,
1040 		    sizeof(struct ippseudo));
1041 		break;
1042 		}
1043 #ifdef INET6
1044 	case AF_INET6: {
1045 		struct ip6_hdr_pseudo ip6pseudo;
1046 		struct ip6_hdr *ip6;
1047 
1048 		ip6 = mtod(m, struct ip6_hdr *);
1049 		bzero(&ip6pseudo, sizeof(ip6pseudo));
1050 		ip6pseudo.ip6ph_src = ip6->ip6_src;
1051 		ip6pseudo.ip6ph_dst = ip6->ip6_dst;
1052 		in6_clearscope(&ip6pseudo.ip6ph_src);
1053 		in6_clearscope(&ip6pseudo.ip6ph_dst);
1054 		ip6pseudo.ip6ph_nxt = IPPROTO_TCP;
1055 		ip6pseudo.ip6ph_len = htonl(m->m_pkthdr.len - iphlen);
1056 
1057 		MD5Update(&ctx, (char *)&ip6pseudo,
1058 		    sizeof(ip6pseudo));
1059 		break;
1060 		}
1061 #endif
1062 	}
1063 
1064 	th0 = *th;
1065 	th0.th_sum = 0;
1066 
1067 	if (doswap) {
1068 		th0.th_seq = htonl(th0.th_seq);
1069 		th0.th_ack = htonl(th0.th_ack);
1070 		th0.th_win = htons(th0.th_win);
1071 		th0.th_urp = htons(th0.th_urp);
1072 	}
1073 	MD5Update(&ctx, (char *)&th0, sizeof(th0));
1074 
1075 	len = m->m_pkthdr.len - iphlen - th->th_off * sizeof(uint32_t);
1076 
1077 	if (len > 0 &&
1078 	    m_apply(m, iphlen + th->th_off * sizeof(uint32_t), len,
1079 	    tcp_signature_apply, (caddr_t)&ctx))
1080 		return (-1);
1081 
1082 	MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen);
1083 	MD5Final(sig, &ctx);
1084 
1085 	return (0);
1086 }
1087 #endif /* TCP_SIGNATURE */
1088