1 /**	$MirOS: src/sys/netinet/tcp_subr.c,v 1.7 2011/11/20 18:54:51 tg Exp $	*/
2 /*	$OpenBSD: tcp_subr.c,v 1.85 2004/11/25 15:32:08 markus Exp $	*/
3 /*	$NetBSD: tcp_subr.c,v 1.22 1996/02/13 23:44:00 christos Exp $	*/
4 
5 /*
6  * Copyright (c) 1982, 1986, 1988, 1990, 1993
7  *	The Regents of the University of California.  All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)COPYRIGHT	1.1 (NRL) 17 January 1995
34  *
35  * NRL grants permission for redistribution and use in source and binary
36  * forms, with or without modification, of the software and documentation
37  * created at NRL provided that the following conditions are met:
38  *
39  * 1. Redistributions of source code must retain the above copyright
40  *    notice, this list of conditions and the following disclaimer.
41  * 2. Redistributions in binary form must reproduce the above copyright
42  *    notice, this list of conditions and the following disclaimer in the
43  *    documentation and/or other materials provided with the distribution.
44  * 3. All advertising materials mentioning features or use of this software
45  *    must display the following acknowledgements:
46  * 	This product includes software developed by the University of
47  * 	California, Berkeley and its contributors.
48  * 	This product includes software developed at the Information
49  * 	Technology Division, US Naval Research Laboratory.
50  * 4. Neither the name of the NRL nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
55  * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
56  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
57  * PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL NRL OR
58  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
59  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
60  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
61  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
62  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
63  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
64  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
65  *
66  * The views and conclusions contained in the software and documentation
67  * are those of the authors and should not be interpreted as representing
68  * official policies, either expressed or implied, of the US Naval
69  * Research Laboratory (NRL).
70  */
71 
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/proc.h>
75 #include <sys/mbuf.h>
76 #include <sys/socket.h>
77 #include <sys/socketvar.h>
78 #include <sys/protosw.h>
79 #include <sys/kernel.h>
80 
81 #include <net/route.h>
82 #include <net/if.h>
83 
84 #include <netinet/in.h>
85 #include <netinet/in_systm.h>
86 #include <netinet/ip.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/ip_var.h>
89 #include <netinet/ip_icmp.h>
90 #include <netinet/tcp.h>
91 #include <netinet/tcp_fsm.h>
92 #include <netinet/tcp_seq.h>
93 #include <netinet/tcp_timer.h>
94 #include <netinet/tcp_var.h>
95 #include <netinet/tcpip.h>
96 #include <dev/rndvar.h>
97 
98 #ifdef INET6
99 #include <netinet6/in6_var.h>
100 #include <netinet6/ip6protosw.h>
101 #endif /* INET6 */
102 
103 #ifdef TCP_SIGNATURE
104 #include <syskern/md5.h>
105 #endif /* TCP_SIGNATURE */
106 
107 /* patchable/settable parameters for tcp */
108 int	tcp_mssdflt = TCP_MSS;
109 int	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
110 
111 /*
112  * Configure kernel with options "TCP_DO_RFC1323=0" to disable RFC1323 stuff.
113  * This is a good idea over slow SLIP/PPP links, because the timestamp
114  * pretty well destroys the VJ compression (any packet with a timestamp
115  * different from the previous one can't be compressed), as well as adding
116  * more overhead.
117  * XXX And it should be a settable per route characteristic (with this just
118  * used as the default).
119  */
120 #ifndef TCP_DO_RFC1323
121 #define TCP_DO_RFC1323	1
122 #endif
123 int	tcp_do_rfc1323 = TCP_DO_RFC1323;
124 
125 #ifndef TCP_DO_SACK
126 #ifdef TCP_SACK
127 #define TCP_DO_SACK	1
128 #else
129 #define TCP_DO_SACK	0
130 #endif
131 #endif
132 int	tcp_do_sack = TCP_DO_SACK;		/* RFC 2018 selective ACKs */
133 int	tcp_ack_on_push = 0;	/* set to enable immediate ACK-on-PUSH */
134 int	tcp_do_ecn = 1;		/* RFC3168 ECN enabled/disabled? */
135 int	tcp_do_rfc3390 = 0;	/* RFC3390 Increasing TCP's Initial Window */
136 
137 u_int32_t	tcp_now;
138 
139 #ifndef TCBHASHSIZE
140 #define	TCBHASHSIZE	128
141 #endif
142 int	tcbhashsize = TCBHASHSIZE;
143 
144 /* syn hash parameters */
145 #define	TCP_SYN_HASH_SIZE	293
146 #define	TCP_SYN_BUCKET_SIZE	35
147 int	tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
148 int	tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
149 int	tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
150 struct	syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
151 
152 int tcp_reass_limit = NMBCLUSTERS / 2; /* hardlimit for tcpqe_pool */
153 #ifdef TCP_SACK
154 int tcp_sackhole_limit = 32*1024; /* hardlimit for sackhl_pool */
155 #endif
156 
157 #ifdef INET6
158 extern int ip6_defhlim;
159 #endif /* INET6 */
160 
161 struct pool tcpcb_pool;
162 struct pool tcpqe_pool;
163 #ifdef TCP_SACK
164 struct pool sackhl_pool;
165 #endif
166 
167 struct tcpstat tcpstat;		/* tcp statistics */
168 tcp_seq  tcp_iss;
169 
170 /*
171  * Tcp initialization
172  */
173 void
tcp_init()174 tcp_init()
175 {
176 #ifdef TCP_COMPAT_42
177 	tcp_iss = 1;		/* wrong */
178 #endif /* TCP_COMPAT_42 */
179 	pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
180 	    NULL);
181 	pool_init(&tcpqe_pool, sizeof(struct ipqent), 0, 0, 0, "tcpqepl",
182 	    NULL);
183 	pool_sethardlimit(&tcpqe_pool, tcp_reass_limit, NULL, 0);
184 #ifdef TCP_SACK
185 	pool_init(&sackhl_pool, sizeof(struct sackhole), 0, 0, 0, "sackhlpl",
186 	    NULL);
187 	pool_sethardlimit(&sackhl_pool, tcp_sackhole_limit, NULL, 0);
188 #endif /* TCP_SACK */
189 	in_pcbinit(&tcbtable, tcbhashsize);
190 	tcp_now = arc4random() / 2;
191 
192 #ifdef INET6
193 	/*
194 	 * Since sizeof(struct ip6_hdr) > sizeof(struct ip), we
195 	 * do max length checks/computations only on the former.
196 	 */
197 	if (max_protohdr < (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)))
198 		max_protohdr = (sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
199 	if ((max_linkhdr + sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) >
200 	    MHLEN)
201 		panic("tcp_init");
202 
203 	icmp6_mtudisc_callback_register(tcp6_mtudisc_callback);
204 #endif /* INET6 */
205 
206 	/* Initialize the compressed state engine. */
207 	syn_cache_init();
208 
209 	/* Initialize timer state. */
210 	tcp_timer_init();
211 }
212 
213 /*
214  * Create template to be used to send tcp packets on a connection.
215  * Call after host entry created, allocates an mbuf and fills
216  * in a skeletal tcp/ip header, minimizing the amount of work
217  * necessary when the connection is used.
218  *
219  * To support IPv6 in addition to IPv4 and considering that the sizes of
220  * the IPv4 and IPv6 headers are not the same, we now use a separate pointer
221  * for the TCP header.  Also, we made the former tcpiphdr header pointer
222  * into just an IP overlay pointer, with casting as appropriate for v6. rja
223  */
224 struct mbuf *
tcp_template(tp)225 tcp_template(tp)
226 	struct tcpcb *tp;
227 {
228 	struct inpcb *inp = tp->t_inpcb;
229 	struct mbuf *m;
230 	struct tcphdr *th;
231 
232 	if ((m = tp->t_template) == 0) {
233 		m = m_get(M_DONTWAIT, MT_HEADER);
234 		if (m == NULL)
235 			return (NULL);
236 
237 		switch (tp->pf) {
238 		case 0:	/*default to PF_INET*/
239 #ifdef INET
240 		case AF_INET:
241 			m->m_len = sizeof(struct ip);
242 			break;
243 #endif /* INET */
244 #ifdef INET6
245 		case AF_INET6:
246 			m->m_len = sizeof(struct ip6_hdr);
247 			break;
248 #endif /* INET6 */
249 		}
250 		m->m_len += sizeof (struct tcphdr);
251 
252 		/*
253 		 * The link header, network header, TCP header, and TCP options
254 		 * all must fit in this mbuf. For now, assume the worst case of
255 		 * TCP options size. Eventually, compute this from tp flags.
256 		 */
257 		if (m->m_len + MAX_TCPOPTLEN + max_linkhdr >= MHLEN) {
258 			MCLGET(m, M_DONTWAIT);
259 			if ((m->m_flags & M_EXT) == 0) {
260 				m_free(m);
261 				return (0);
262 			}
263 		}
264 	}
265 
266 	switch(tp->pf) {
267 #ifdef INET
268 	case AF_INET:
269 		{
270 			struct ipovly *ipovly;
271 
272 			ipovly = mtod(m, struct ipovly *);
273 
274 			bzero(ipovly->ih_x1, sizeof ipovly->ih_x1);
275 			ipovly->ih_pr = IPPROTO_TCP;
276 			ipovly->ih_len = htons(sizeof (struct tcphdr));
277 			ipovly->ih_src = inp->inp_laddr;
278 			ipovly->ih_dst = inp->inp_faddr;
279 
280 			th = (struct tcphdr *)(mtod(m, caddr_t) +
281 				sizeof(struct ip));
282 			th->th_sum = in_cksum_phdr(ipovly->ih_src.s_addr,
283 			    ipovly->ih_dst.s_addr,
284 			    htons(sizeof (struct tcphdr) + IPPROTO_TCP));
285 		}
286 		break;
287 #endif /* INET */
288 #ifdef INET6
289 	case AF_INET6:
290 		{
291 			struct ip6_hdr *ip6;
292 
293 			ip6 = mtod(m, struct ip6_hdr *);
294 
295 			ip6->ip6_src = inp->inp_laddr6;
296 			ip6->ip6_dst = inp->inp_faddr6;
297 			ip6->ip6_flow = htonl(0x60000000) |
298 			    (inp->inp_flowinfo & IPV6_FLOWLABEL_MASK);
299 
300 			ip6->ip6_nxt = IPPROTO_TCP;
301 			ip6->ip6_plen = htons(sizeof(struct tcphdr)); /*XXX*/
302 			ip6->ip6_hlim = in6_selecthlim(inp, NULL);	/*XXX*/
303 
304 			th = (struct tcphdr *)(mtod(m, caddr_t) +
305 				sizeof(struct ip6_hdr));
306 			th->th_sum = 0;
307 		}
308 		break;
309 #endif /* INET6 */
310 	default:
311 		return (NULL);
312 	}
313 
314 	th->th_sport = inp->inp_lport;
315 	th->th_dport = inp->inp_fport;
316 	th->th_seq = 0;
317 	th->th_ack = 0;
318 	th->th_x2  = 0;
319 	th->th_off = 5;
320 	th->th_flags = 0;
321 	th->th_win = 0;
322 	th->th_urp = 0;
323 	return (m);
324 }
325 
326 /*
327  * Send a single message to the TCP at address specified by
328  * the given TCP/IP header.  If m == 0, then we make a copy
329  * of the tcpiphdr at ti and send directly to the addressed host.
330  * This is used to force keep alive messages out using the TCP
331  * template for a connection tp->t_template.  If flags are given
332  * then we send a message back to the TCP which originated the
333  * segment ti, and discard the mbuf containing it and any other
334  * attached mbufs.
335  *
336  * In any case the ack and sequence number of the transmitted
337  * segment are as specified by the parameters.
338  */
339 #ifdef INET6
340 /* This function looks hairy, because it was so IPv4-dependent. */
341 #endif /* INET6 */
342 void
tcp_respond(tp,template,th0,ack,seq,flags)343 tcp_respond(tp, template, th0, ack, seq, flags)
344 	struct tcpcb *tp;
345 	caddr_t template;
346 	struct tcphdr *th0;
347 	tcp_seq ack, seq;
348 	int flags;
349 {
350 	int tlen;
351 	int win = 0;
352 	struct mbuf *m = 0;
353 	struct route *ro = 0;
354 	struct tcphdr *th;
355 	struct ip *ip;
356 	struct ipovly *ih;
357 #ifdef INET6
358 	struct ip6_hdr *ip6;
359 #endif
360 	int af;		/* af on wire */
361 
362 	if (tp) {
363 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
364 		/*
365 		 * If this is called with an unconnected
366 		 * socket/tp/pcb (tp->pf is 0), we lose.
367 		 */
368 		af = tp->pf;
369 
370 		/*
371 		 * The route/route6 distinction is meaningless
372 		 * unless you're allocating space or passing parameters.
373 		 */
374 		ro = &tp->t_inpcb->inp_route;
375 	} else
376 		af = (((struct ip *)template)->ip_v == 6) ? AF_INET6 : AF_INET;
377 
378 	m = m_gethdr(M_DONTWAIT, MT_HEADER);
379 	if (m == NULL)
380 		return;
381 	m->m_data += max_linkhdr;
382 #ifdef TCP_COMPAT_42
383 	tlen = 1;
384 #else
385 	tlen = 0;
386 #endif
387 
388 #define xchg(a,b,type) do { type t; t=a; a=b; b=t; } while (0)
389 	switch (af) {
390 #ifdef INET6
391 	case AF_INET6:
392 		ip6 = mtod(m, struct ip6_hdr *);
393 		th = (struct tcphdr *)(ip6 + 1);
394 		tlen = sizeof(*ip6) + sizeof(*th);
395 		if (th0) {
396 			bcopy(template, ip6, sizeof(*ip6));
397 			bcopy(th0, th, sizeof(*th));
398 			xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
399 		} else {
400 			bcopy(template, ip6, tlen);
401 		}
402 		break;
403 #endif /* INET6 */
404 	case AF_INET:
405 		ip = mtod(m, struct ip *);
406 		th = (struct tcphdr *)(ip + 1);
407 		tlen = sizeof(*ip) + sizeof(*th);
408 		if (th0) {
409 			bcopy(template, ip, sizeof(*ip));
410 			bcopy(th0, th, sizeof(*th));
411 			xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, u_int32_t);
412 		} else {
413 			bcopy(template, ip, tlen);
414 		}
415  		break;
416  	}
417 	if (th0)
418 		xchg(th->th_dport, th->th_sport, u_int16_t);
419 	else
420 		flags = TH_ACK;
421 #undef xchg
422 
423 	m->m_len = tlen;
424 	m->m_pkthdr.len = tlen;
425 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
426 	th->th_seq = htonl(seq);
427 	th->th_ack = htonl(ack);
428 	th->th_x2 = 0;
429 	th->th_off = sizeof (struct tcphdr) >> 2;
430 	th->th_flags = flags;
431 	if (tp)
432 		win >>= tp->rcv_scale;
433 	if (win > TCP_MAXWIN)
434 		win = TCP_MAXWIN;
435 	th->th_win = htons((u_int16_t)win);
436 	th->th_urp = 0;
437 
438 	switch (af) {
439 #ifdef INET6
440 	case AF_INET6:
441 		ip6->ip6_flow = htonl(0x60000000);
442 		ip6->ip6_nxt  = IPPROTO_TCP;
443 		ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL, NULL);	/*XXX*/
444 		ip6->ip6_plen = tlen - sizeof(struct ip6_hdr);
445 		th->th_sum = 0;
446 		th->th_sum = in6_cksum(m, IPPROTO_TCP,
447 		   sizeof(struct ip6_hdr), ip6->ip6_plen);
448 		HTONS(ip6->ip6_plen);
449 		ip6_output(m, tp ? tp->t_inpcb->inp_outputopts6 : NULL,
450 		    (struct route_in6 *)ro, 0, NULL, NULL);
451 		break;
452 #endif /* INET6 */
453 	case AF_INET:
454 		ih = (struct ipovly *)ip;
455 		bzero(ih->ih_x1, sizeof ih->ih_x1);
456 		ih->ih_len = htons((u_short)tlen - sizeof(struct ip));
457 
458 		/*
459 		 * There's no point deferring to hardware checksum processing
460 		 * here, as we only send a minimal TCP packet whose checksum
461 		 * we need to compute in any case.
462 		 */
463 		th->th_sum = 0;
464 		th->th_sum = in_cksum(m, tlen);
465 		ip->ip_len = htons(tlen);
466 		ip->ip_ttl = ip_defttl;
467 		ip_output(m, (void *)NULL, ro, ip_mtudisc ? IP_MTUDISC : 0,
468 			(void *)NULL, tp ? tp->t_inpcb : (void *)NULL);
469 	}
470 }
471 
472 /*
473  * Create a new TCP control block, making an
474  * empty reassembly queue and hooking it to the argument
475  * protocol control block.
476  */
477 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp)478 tcp_newtcpcb(struct inpcb *inp)
479 {
480 	struct tcpcb *tp;
481 	int i;
482 
483 	tp = pool_get(&tcpcb_pool, PR_NOWAIT);
484 	if (tp == NULL)
485 		return ((struct tcpcb *)0);
486 	bzero((char *) tp, sizeof(struct tcpcb));
487 	LIST_INIT(&tp->segq);
488 	tp->t_maxseg = tcp_mssdflt;
489 	tp->t_maxopd = 0;
490 
491 	TCP_INIT_DELACK(tp);
492 	for (i = 0; i < TCPT_NTIMERS; i++)
493 		TCP_TIMER_INIT(tp, i);
494 	timeout_set(&tp->t_reap_to, tcp_reaper, tp);
495 
496 #ifdef TCP_SACK
497 	tp->sack_enable = tcp_do_sack;
498 #endif
499 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
500 	tp->t_inpcb = inp;
501 	/*
502 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
503 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
504 	 * reasonable initial retransmit time.
505 	 */
506 	tp->t_srtt = TCPTV_SRTTBASE;
507 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
508 	tp->t_rttmin = TCPTV_MIN;
509 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
510 	    TCPTV_MIN, TCPTV_REXMTMAX);
511 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
512 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
513 
514 	tp->t_pmtud_mtu_sent = 0;
515 	tp->t_pmtud_mss_acked = 0;
516 
517 #ifdef INET6
518 	/* we disallow IPv4 mapped address completely. */
519 	if ((inp->inp_flags & INP_IPV6) == 0)
520 		tp->pf = PF_INET;
521 	else
522 		tp->pf = PF_INET6;
523 #else
524 	tp->pf = PF_INET;
525 #endif
526 
527 #ifdef INET6
528 	if (inp->inp_flags & INP_IPV6)
529 		inp->inp_ipv6.ip6_hlim = ip6_defhlim;
530 	else
531 #endif /* INET6 */
532 		inp->inp_ip.ip_ttl = ip_defttl;
533 
534 	inp->inp_ppcb = (caddr_t)tp;
535 	return (tp);
536 }
537 
538 /*
539  * Drop a TCP connection, reporting
540  * the specified error.  If connection is synchronized,
541  * then send a RST to peer.
542  */
543 struct tcpcb *
tcp_drop(tp,errno)544 tcp_drop(tp, errno)
545 	struct tcpcb *tp;
546 	int errno;
547 {
548 	struct socket *so = tp->t_inpcb->inp_socket;
549 
550 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
551 		tp->t_state = TCPS_CLOSED;
552 		(void) tcp_output(tp);
553 		tcpstat.tcps_drops++;
554 	} else
555 		tcpstat.tcps_conndrops++;
556 	if (errno == ETIMEDOUT && tp->t_softerror)
557 		errno = tp->t_softerror;
558 	so->so_error = errno;
559 	return (tcp_close(tp));
560 }
561 
562 /*
563  * Close a TCP control block:
564  *	discard all space held by the tcp
565  *	discard internet protocol block
566  *	wake up any sleepers
567  */
568 struct tcpcb *
tcp_close(struct tcpcb * tp)569 tcp_close(struct tcpcb *tp)
570 {
571 	struct inpcb *inp = tp->t_inpcb;
572 	struct socket *so = inp->inp_socket;
573 #ifdef TCP_SACK
574 	struct sackhole *p, *q;
575 #endif
576 
577 	/* free the reassembly queue, if any */
578 	tcp_reass_lock(tp);
579 	tcp_freeq(tp);
580 	tcp_reass_unlock(tp);
581 
582 	tcp_canceltimers(tp);
583 	TCP_CLEAR_DELACK(tp);
584 	syn_cache_cleanup(tp);
585 
586 #ifdef TCP_SACK
587 	/* Free SACK holes. */
588 	q = p = tp->snd_holes;
589 	while (p != 0) {
590 		q = p->next;
591 		pool_put(&sackhl_pool, p);
592 		p = q;
593 	}
594 #endif
595 	if (tp->t_template)
596 		(void) m_free(tp->t_template);
597 
598 	tp->t_flags |= TF_DEAD;
599 	timeout_add(&tp->t_reap_to, 0);
600 
601 	inp->inp_ppcb = 0;
602 	soisdisconnected(so);
603 	in_pcbdetach(inp);
604 	return ((struct tcpcb *)0);
605 }
606 
607 void
tcp_reaper(void * arg)608 tcp_reaper(void *arg)
609 {
610 	struct tcpcb *tp = arg;
611 	int s;
612 
613 	s = splsoftnet();
614 	pool_put(&tcpcb_pool, tp);
615 	splx(s);
616 	tcpstat.tcps_closed++;
617 }
618 
619 int
tcp_freeq(struct tcpcb * tp)620 tcp_freeq(struct tcpcb *tp)
621 {
622 	struct ipqent *qe;
623 	int rv = 0;
624 
625 	while ((qe = LIST_FIRST(&tp->segq)) != NULL) {
626 		LIST_REMOVE(qe, ipqe_q);
627 		m_freem(qe->ipqe_m);
628 		pool_put(&tcpqe_pool, qe);
629 		rv = 1;
630 	}
631 	return (rv);
632 }
633 
634 void
tcp_drain()635 tcp_drain()
636 {
637 	struct inpcb *inp;
638 
639 	/* called at splimp() */
640 	CIRCLEQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
641 		struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
642 
643 		if (tp != NULL) {
644 			if (tcp_reass_lock_try(tp) == 0)
645 				continue;
646 			if (tcp_freeq(tp))
647 				tcpstat.tcps_conndrained++;
648 			tcp_reass_unlock(tp);
649 		}
650 	}
651 }
652 
653 /*
654  * Compute proper scaling value for receiver window from buffer space
655  */
656 
657 void
tcp_rscale(struct tcpcb * tp,u_long hiwat)658 tcp_rscale(struct tcpcb *tp, u_long hiwat)
659 {
660 	tp->request_r_scale = 0;
661 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
662 	       TCP_MAXWIN << tp->request_r_scale < hiwat)
663 		tp->request_r_scale++;
664 }
665 
666 /*
667  * Notify a tcp user of an asynchronous error;
668  * store error as soft error, but wake up user
669  * (for now, won't do anything until can select for soft error).
670  */
671 void
tcp_notify(inp,error)672 tcp_notify(inp, error)
673 	struct inpcb *inp;
674 	int error;
675 {
676 	struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
677 	struct socket *so = inp->inp_socket;
678 
679 	/*
680 	 * Ignore some errors if we are hooked up.
681 	 * If connection hasn't completed, has retransmitted several times,
682 	 * and receives a second error, give up now.  This is better
683 	 * than waiting a long time to establish a connection that
684 	 * can never complete.
685 	 */
686 	if (tp->t_state == TCPS_ESTABLISHED &&
687 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
688 	      error == EHOSTDOWN)) {
689 		return;
690 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
691 	    tp->t_rxtshift > 3 && tp->t_softerror)
692 		so->so_error = error;
693 	else
694 		tp->t_softerror = error;
695 	wakeup((caddr_t) &so->so_timeo);
696 	sorwakeup(so);
697 	sowwakeup(so);
698 }
699 
700 #ifdef INET6
701 void
tcp6_ctlinput(cmd,sa,d)702 tcp6_ctlinput(cmd, sa, d)
703 	int cmd;
704 	struct sockaddr *sa;
705 	void *d;
706 {
707 	struct tcphdr th;
708 	struct tcpcb *tp;
709 	void (*notify)(struct inpcb *, int) = tcp_notify;
710 	struct ip6_hdr *ip6;
711 	const struct sockaddr_in6 *sa6_src = NULL;
712 	struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa;
713 	struct inpcb *inp;
714 	struct mbuf *m;
715 	tcp_seq seq;
716 	int off;
717 	struct {
718 		u_int16_t th_sport;
719 		u_int16_t th_dport;
720 		u_int32_t th_seq;
721 	} *thp;
722 
723 	if (sa->sa_family != AF_INET6 ||
724 	    sa->sa_len != sizeof(struct sockaddr_in6) ||
725 	    IN6_IS_ADDR_UNSPECIFIED(&sa6->sin6_addr) ||
726 	    IN6_IS_ADDR_V4MAPPED(&sa6->sin6_addr))
727 		return;
728 	if ((unsigned)cmd >= PRC_NCMDS)
729 		return;
730 	else if (cmd == PRC_QUENCH) {
731 		/* XXX there's no PRC_QUENCH in IPv6 */
732 		notify = tcp_quench;
733 	} else if (PRC_IS_REDIRECT(cmd))
734 		notify = in_rtchange, d = NULL;
735 	else if (cmd == PRC_MSGSIZE)
736 		; /* special code is present, see below */
737 	else if (cmd == PRC_HOSTDEAD)
738 		d = NULL;
739 	else if (inet6ctlerrmap[cmd] == 0)
740 		return;
741 
742 	/* if the parameter is from icmp6, decode it. */
743 	if (d != NULL) {
744 		struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
745 		m = ip6cp->ip6c_m;
746 		ip6 = ip6cp->ip6c_ip6;
747 		off = ip6cp->ip6c_off;
748 		sa6_src = ip6cp->ip6c_src;
749 	} else {
750 		m = NULL;
751 		ip6 = NULL;
752 	off = 0;
753 		sa6_src = &sa6_any;
754 	}
755 
756 	if (ip6) {
757 		/*
758 		 * XXX: We assume that when ip6 is non NULL,
759 		 * M and OFF are valid.
760 		 */
761 
762 		/* check if we can safely examine src and dst ports */
763 		if (m->m_pkthdr.len < off + sizeof(*thp))
764 			return;
765 
766 		bzero(&th, sizeof(th));
767 #ifdef DIAGNOSTIC
768 		if (sizeof(*thp) > sizeof(th))
769 			panic("assumption failed in tcp6_ctlinput");
770 #endif
771 		m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
772 
773 		/*
774 		 * Check to see if we have a valid TCP connection
775 		 * corresponding to the address in the ICMPv6 message
776 		 * payload.
777 		 */
778 		inp = in6_pcbhashlookup(&tcbtable, &sa6->sin6_addr,
779 		    th.th_dport, (struct in6_addr *)&sa6_src->sin6_addr,
780 		    th.th_sport);
781 		if (cmd == PRC_MSGSIZE) {
782 			/*
783 			 * Depending on the value of "valid" and routing table
784 			 * size (mtudisc_{hi,lo}wat), we will:
785 			 * - recalcurate the new MTU and create the
786 			 *   corresponding routing entry, or
787 			 * - ignore the MTU change notification.
788 			 */
789 			icmp6_mtudisc_update((struct ip6ctlparam *)d, inp != NULL);
790 			return;
791 		}
792 		if (inp) {
793 			seq = ntohl(th.th_seq);
794 			if (inp->inp_socket &&
795 			    (tp = intotcpcb(inp)) &&
796 			    SEQ_GEQ(seq, tp->snd_una) &&
797 			    SEQ_LT(seq, tp->snd_max))
798 				notify(inp, inet6ctlerrmap[cmd]);
799 		} else if (syn_cache_count &&
800 		    (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
801 		     inet6ctlerrmap[cmd] == ENETUNREACH ||
802 		     inet6ctlerrmap[cmd] == EHOSTDOWN))
803 			syn_cache_unreach((struct sockaddr *)sa6_src,
804 			    sa, &th);
805 	} else {
806 		(void) in6_pcbnotify(&tcbtable, sa, 0,
807 		    (struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
808 	}
809 }
810 #endif
811 
812 void *
tcp_ctlinput(cmd,sa,v)813 tcp_ctlinput(cmd, sa, v)
814 	int cmd;
815 	struct sockaddr *sa;
816 	void *v;
817 {
818 	struct ip *ip = v;
819 	struct tcphdr *th;
820 	struct tcpcb *tp;
821 	struct inpcb *inp;
822 	struct in_addr faddr;
823 	tcp_seq seq;
824 	extern int inetctlerrmap[];
825 	void (*notify)(struct inpcb *, int) = tcp_notify;
826 	u_int mtu;
827 	int errno;
828 
829 	if (sa->sa_family != AF_INET)
830 		return NULL;
831 	faddr = satosin(sa)->sin_addr;
832 	if (faddr.s_addr == INADDR_ANY)
833 		return NULL;
834 
835 	if ((unsigned)cmd >= PRC_NCMDS)
836 		return NULL;
837 	errno = inetctlerrmap[cmd];
838 	if (cmd == PRC_QUENCH)
839 		notify = tcp_quench;
840 	else if (PRC_IS_REDIRECT(cmd))
841 		notify = in_rtchange, ip = 0;
842 	else if (cmd == PRC_MSGSIZE && ip_mtudisc && ip) {
843 		/*
844 		 * Verify that the packet in the icmp payload refers
845 		 * to an existing TCP connection.
846 		 */
847 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
848 		seq = ntohl(th->th_seq);
849 		inp = in_pcbhashlookup(&tcbtable,
850 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport);
851 		if (inp && (tp = intotcpcb(inp)) &&
852 		    SEQ_GEQ(seq, tp->snd_una) &&
853 		    SEQ_LT(seq, tp->snd_max)) {
854 			struct icmp *icp;
855 			icp = (struct icmp *)((caddr_t)ip -
856 					      offsetof(struct icmp, icmp_ip));
857 
858 			/*
859 			 * If the ICMP message advertises a Next-Hop MTU
860 			 * equal or larger than the maximum packet size we have
861 			 * ever sent, drop the message.
862 			 */
863 			mtu = (u_int)ntohs(icp->icmp_nextmtu);
864 			if (mtu >= tp->t_pmtud_mtu_sent)
865 				return NULL;
866 			if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
867 				/*
868 				 * Calculate new MTU, and create corresponding
869 				 * route (traditional PMTUD).
870 				 */
871 				tp->t_flags &= ~TF_PMTUD_PEND;
872 				icmp_mtudisc(icp);
873 			} else {
874 				/*
875 				 * Record the information got in the ICMP
876 				 * message; act on it later.
877 				 * If we had already recorded an ICMP message,
878 				 * replace the old one only if the new message
879 				 * refers to an older TCP segment
880 				 */
881 				if (tp->t_flags & TF_PMTUD_PEND) {
882 					if (SEQ_LT(tp->t_pmtud_th_seq, seq))
883 						return NULL;
884 				} else
885 					tp->t_flags |= TF_PMTUD_PEND;
886 				tp->t_pmtud_th_seq = seq;
887 				tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
888 				tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
889 				tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
890 				return NULL;
891 			}
892 		}
893 		notify = tcp_mtudisc, ip = 0;
894 	} else if (cmd == PRC_MTUINC)
895 		notify = tcp_mtudisc_increase, ip = 0;
896 	else if (cmd == PRC_HOSTDEAD)
897 		ip = 0;
898 	else if (errno == 0)
899 		return NULL;
900 
901 	if (ip) {
902 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
903 		inp = in_pcbhashlookup(&tcbtable,
904 		    ip->ip_dst, th->th_dport, ip->ip_src, th->th_sport);
905 		if (inp) {
906 			seq = ntohl(th->th_seq);
907 			if (inp->inp_socket &&
908 			    (tp = intotcpcb(inp)) &&
909 			    SEQ_GEQ(seq, tp->snd_una) &&
910 			    SEQ_LT(seq, tp->snd_max))
911 				notify(inp, errno);
912 		} else if (syn_cache_count &&
913 		    (inetctlerrmap[cmd] == EHOSTUNREACH ||
914 		     inetctlerrmap[cmd] == ENETUNREACH ||
915 		     inetctlerrmap[cmd] == EHOSTDOWN)) {
916 			struct sockaddr_in sin;
917 
918 			bzero(&sin, sizeof(sin));
919 			sin.sin_len = sizeof(sin);
920 			sin.sin_family = AF_INET;
921 			sin.sin_port = th->th_sport;
922 			sin.sin_addr = ip->ip_src;
923 			syn_cache_unreach((struct sockaddr *)&sin,
924 			    sa, th);
925 		}
926 	} else
927 		in_pcbnotifyall(&tcbtable, sa, errno, notify);
928 
929 	return NULL;
930 }
931 
932 /*
933  * When a source quench is received, close congestion window
934  * to one segment.  We will gradually open it again as we proceed.
935  */
936 void
tcp_quench(inp,errno)937 tcp_quench(inp, errno)
938 	struct inpcb *inp;
939 	int errno;
940 {
941 	struct tcpcb *tp = intotcpcb(inp);
942 
943 	if (tp)
944 		tp->snd_cwnd = tp->t_maxseg;
945 }
946 
947 #ifdef INET6
948 /*
949  * Path MTU Discovery handlers.
950  */
951 void
tcp6_mtudisc_callback(faddr)952 tcp6_mtudisc_callback(faddr)
953 	struct in6_addr *faddr;
954 {
955 	struct sockaddr_in6 sin6;
956 
957 	bzero(&sin6, sizeof(sin6));
958 	sin6.sin6_family = AF_INET6;
959 	sin6.sin6_len = sizeof(struct sockaddr_in6);
960 	sin6.sin6_addr = *faddr;
961 	(void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
962 	    (struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp_mtudisc);
963 }
964 #endif /* INET6 */
965 
966 /*
967  * On receipt of path MTU corrections, flush old route and replace it
968  * with the new one.  Retransmit all unacknowledged packets, to ensure
969  * that all packets will be received.
970  */
971 void
tcp_mtudisc(inp,errno)972 tcp_mtudisc(inp, errno)
973 	struct inpcb *inp;
974 	int errno;
975 {
976 	struct tcpcb *tp = intotcpcb(inp);
977 	struct rtentry *rt = in_pcbrtentry(inp);
978 	int change = 0;
979 
980 	if (tp != 0) {
981 		int orig_maxseg = tp->t_maxseg;
982 		if (rt != 0) {
983 			/*
984 			 * If this was not a host route, remove and realloc.
985 			 */
986 			if ((rt->rt_flags & RTF_HOST) == 0) {
987 				in_rtchange(inp, errno);
988 				if ((rt = in_pcbrtentry(inp)) == 0)
989 					return;
990 			}
991 			if (orig_maxseg != tp->t_maxseg ||
992 			    (rt->rt_rmx.rmx_locks & RTV_MTU))
993 				change = 1;
994 		}
995 		tcp_mss(tp, -1);
996 
997 		/*
998 		 * Resend unacknowledged packets
999 		 */
1000 		tp->snd_nxt = tp->snd_una;
1001 		if (change || errno > 0)
1002 			tcp_output(tp);
1003 	}
1004 }
1005 
1006 void
tcp_mtudisc_increase(inp,errno)1007 tcp_mtudisc_increase(inp, errno)
1008 	struct inpcb *inp;
1009 	int errno;
1010 {
1011 	struct tcpcb *tp = intotcpcb(inp);
1012 	struct rtentry *rt = in_pcbrtentry(inp);
1013 
1014 	if (tp != 0 && rt != 0) {
1015 		/*
1016 		 * If this was a host route, remove and realloc.
1017 		 */
1018 		if (rt->rt_flags & RTF_HOST)
1019 			in_rtchange(inp, errno);
1020 
1021 		/* also takes care of congestion window */
1022 		tcp_mss(tp, -1);
1023 	}
1024 }
1025 
1026 #ifdef TCP_SIGNATURE
1027 int
tcp_signature_tdb_attach()1028 tcp_signature_tdb_attach()
1029 {
1030 	return (0);
1031 }
1032 
1033 int
tcp_signature_tdb_init(tdbp,xsp,ii)1034 tcp_signature_tdb_init(tdbp, xsp, ii)
1035 	struct tdb *tdbp;
1036 	struct xformsw *xsp;
1037 	struct ipsecinit *ii;
1038 {
1039 	if ((ii->ii_authkeylen < 1) || (ii->ii_authkeylen > 80))
1040 		return (EINVAL);
1041 
1042 	tdbp->tdb_amxkey = malloc(ii->ii_authkeylen, M_XDATA, M_DONTWAIT);
1043 	if (tdbp->tdb_amxkey == NULL)
1044 		return (ENOMEM);
1045 	bcopy(ii->ii_authkey, tdbp->tdb_amxkey, ii->ii_authkeylen);
1046 	tdbp->tdb_amxkeylen = ii->ii_authkeylen;
1047 
1048 	return (0);
1049 }
1050 
1051 int
tcp_signature_tdb_zeroize(tdbp)1052 tcp_signature_tdb_zeroize(tdbp)
1053 	struct tdb *tdbp;
1054 {
1055 	if (tdbp->tdb_amxkey) {
1056 		bzero(tdbp->tdb_amxkey, tdbp->tdb_amxkeylen);
1057 		free(tdbp->tdb_amxkey, M_XDATA);
1058 		tdbp->tdb_amxkey = NULL;
1059 	}
1060 
1061 	return (0);
1062 }
1063 
1064 int
tcp_signature_tdb_input(m,tdbp,skip,protoff)1065 tcp_signature_tdb_input(m, tdbp, skip, protoff)
1066 	struct mbuf *m;
1067 	struct tdb *tdbp;
1068 	int skip, protoff;
1069 {
1070 	return (0);
1071 }
1072 
1073 int
tcp_signature_tdb_output(m,tdbp,mp,skip,protoff)1074 tcp_signature_tdb_output(m, tdbp, mp, skip, protoff)
1075 	struct mbuf *m;
1076 	struct tdb *tdbp;
1077 	struct mbuf **mp;
1078 	int skip, protoff;
1079 {
1080 	return (EINVAL);
1081 }
1082 
1083 int
tcp_signature_apply(fstate,data,len)1084 tcp_signature_apply(fstate, data, len)
1085 	caddr_t fstate;
1086 	caddr_t data;
1087 	unsigned int len;
1088 {
1089 	MD5Update((MD5_CTX *)fstate, (char *)data, len);
1090 	return 0;
1091 }
1092 #endif /* TCP_SIGNATURE */
1093 
1094 #define TCP_RNDISS_ROUNDS	16
1095 #define TCP_RNDISS_OUT	7200
1096 #define TCP_RNDISS_MAX	30000
1097 
1098 u_int8_t tcp_rndiss_sbox[128];
1099 u_int16_t tcp_rndiss_msb;
1100 u_int16_t tcp_rndiss_cnt;
1101 time_t tcp_rndiss_reseed;
1102 
1103 u_int16_t
tcp_rndiss_encrypt(val)1104 tcp_rndiss_encrypt(val)
1105 	u_int16_t val;
1106 {
1107 	u_int16_t sum = 0, i;
1108 
1109 	for (i = 0; i < TCP_RNDISS_ROUNDS; i++) {
1110 		sum += 0x79b9;
1111 		val ^= ((u_int16_t)tcp_rndiss_sbox[(val^sum) & 0x7f]) << 7;
1112 		val = ((val & 0xff) << 7) | (val >> 8);
1113 	}
1114 
1115 	return val;
1116 }
1117 
1118 void
tcp_rndiss_init()1119 tcp_rndiss_init()
1120 {
1121 	get_random_bytes(tcp_rndiss_sbox, sizeof(tcp_rndiss_sbox));
1122 
1123 	tcp_rndiss_reseed = time.tv_sec + TCP_RNDISS_OUT;
1124 	tcp_rndiss_msb = tcp_rndiss_msb == 0x8000 ? 0 : 0x8000;
1125 	tcp_rndiss_cnt = 0;
1126 }
1127 
1128 tcp_seq
tcp_rndiss_next()1129 tcp_rndiss_next()
1130 {
1131         if (tcp_rndiss_cnt >= TCP_RNDISS_MAX ||
1132 	    time.tv_sec > tcp_rndiss_reseed)
1133                 tcp_rndiss_init();
1134 
1135 	/* (arc4random() & 0x7fff) ensures a 32768 byte gap between ISS */
1136 	return ((tcp_rndiss_encrypt(tcp_rndiss_cnt++) | tcp_rndiss_msb) <<16) |
1137 		(arc4random() & 0x7fff);
1138 }
1139