xref: /freebsd/sys/netinet/tcp_subr.c (revision 8e6b01171e30297084bb0b4457c4183c2746aacc)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. 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  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
34  *	$Id: tcp_subr.c,v 1.19 1995/10/12 17:37:24 wollman Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/proc.h>
39 #include <sys/systm.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/protosw.h>
45 #include <sys/errno.h>
46 #include <sys/queue.h>
47 
48 #include <net/route.h>
49 #include <net/if.h>
50 
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_pcb.h>
55 #include <netinet/in_var.h>
56 #include <netinet/ip_var.h>
57 #include <netinet/ip_icmp.h>
58 #include <netinet/tcp.h>
59 #include <netinet/tcp_fsm.h>
60 #include <netinet/tcp_seq.h>
61 #include <netinet/tcp_timer.h>
62 #include <netinet/tcp_var.h>
63 #include <netinet/tcpip.h>
64 #ifdef TCPDEBUG
65 #include <netinet/tcp_debug.h>
66 #endif
67 
68 /* patchable/settable parameters for tcp */
69 int 	tcp_mssdflt = TCP_MSS;
70 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
71 int	tcp_do_rfc1323 = 1;
72 int	tcp_do_rfc1644 = 1;
73 static	void tcp_cleartaocache(void);
74 
75 /*
76  * Target size of TCP PCB hash table. Will be rounded down to a prime
77  * number.
78  */
79 #ifndef TCBHASHSIZE
80 #define TCBHASHSIZE	128
81 #endif
82 
83 /*
84  * Tcp initialization
85  */
86 void
87 tcp_init()
88 {
89 
90 	tcp_iss = random();	/* wrong, but better than a constant */
91 	tcp_ccgen = 1;
92 	tcp_cleartaocache();
93 	LIST_INIT(&tcb);
94 	tcbinfo.listhead = &tcb;
95 	tcbinfo.hashbase = phashinit(TCBHASHSIZE, M_PCB, &tcbinfo.hashsize);
96 	if (max_protohdr < sizeof(struct tcpiphdr))
97 		max_protohdr = sizeof(struct tcpiphdr);
98 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
99 		panic("tcp_init");
100 }
101 
102 /*
103  * Create template to be used to send tcp packets on a connection.
104  * Call after host entry created, allocates an mbuf and fills
105  * in a skeletal tcp/ip header, minimizing the amount of work
106  * necessary when the connection is used.
107  */
108 struct tcpiphdr *
109 tcp_template(tp)
110 	struct tcpcb *tp;
111 {
112 	register struct inpcb *inp = tp->t_inpcb;
113 	register struct mbuf *m;
114 	register struct tcpiphdr *n;
115 
116 	if ((n = tp->t_template) == 0) {
117 		m = m_get(M_DONTWAIT, MT_HEADER);
118 		if (m == NULL)
119 			return (0);
120 		m->m_len = sizeof (struct tcpiphdr);
121 		n = mtod(m, struct tcpiphdr *);
122 	}
123 	n->ti_next = n->ti_prev = 0;
124 	n->ti_x1 = 0;
125 	n->ti_pr = IPPROTO_TCP;
126 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
127 	n->ti_src = inp->inp_laddr;
128 	n->ti_dst = inp->inp_faddr;
129 	n->ti_sport = inp->inp_lport;
130 	n->ti_dport = inp->inp_fport;
131 	n->ti_seq = 0;
132 	n->ti_ack = 0;
133 	n->ti_x2 = 0;
134 	n->ti_off = 5;
135 	n->ti_flags = 0;
136 	n->ti_win = 0;
137 	n->ti_sum = 0;
138 	n->ti_urp = 0;
139 	return (n);
140 }
141 
142 /*
143  * Send a single message to the TCP at address specified by
144  * the given TCP/IP header.  If m == 0, then we make a copy
145  * of the tcpiphdr at ti and send directly to the addressed host.
146  * This is used to force keep alive messages out using the TCP
147  * template for a connection tp->t_template.  If flags are given
148  * then we send a message back to the TCP which originated the
149  * segment ti, and discard the mbuf containing it and any other
150  * attached mbufs.
151  *
152  * In any case the ack and sequence number of the transmitted
153  * segment are as specified by the parameters.
154  */
155 void
156 tcp_respond(tp, ti, m, ack, seq, flags)
157 	struct tcpcb *tp;
158 	register struct tcpiphdr *ti;
159 	register struct mbuf *m;
160 	tcp_seq ack, seq;
161 	int flags;
162 {
163 	register int tlen;
164 	int win = 0;
165 	struct route *ro = 0;
166 
167 	if (tp) {
168 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
169 		ro = &tp->t_inpcb->inp_route;
170 	}
171 	if (m == 0) {
172 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
173 		if (m == NULL)
174 			return;
175 #ifdef TCP_COMPAT_42
176 		tlen = 1;
177 #else
178 		tlen = 0;
179 #endif
180 		m->m_data += max_linkhdr;
181 		*mtod(m, struct tcpiphdr *) = *ti;
182 		ti = mtod(m, struct tcpiphdr *);
183 		flags = TH_ACK;
184 	} else {
185 		m_freem(m->m_next);
186 		m->m_next = 0;
187 		m->m_data = (caddr_t)ti;
188 		m->m_len = sizeof (struct tcpiphdr);
189 		tlen = 0;
190 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
191 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
192 		xchg(ti->ti_dport, ti->ti_sport, u_short);
193 #undef xchg
194 	}
195 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
196 	tlen += sizeof (struct tcpiphdr);
197 	m->m_len = tlen;
198 	m->m_pkthdr.len = tlen;
199 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
200 	ti->ti_next = ti->ti_prev = 0;
201 	ti->ti_x1 = 0;
202 	ti->ti_seq = htonl(seq);
203 	ti->ti_ack = htonl(ack);
204 	ti->ti_x2 = 0;
205 	ti->ti_off = sizeof (struct tcphdr) >> 2;
206 	ti->ti_flags = flags;
207 	if (tp)
208 		ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
209 	else
210 		ti->ti_win = htons((u_short)win);
211 	ti->ti_urp = 0;
212 	ti->ti_sum = 0;
213 	ti->ti_sum = in_cksum(m, tlen);
214 	((struct ip *)ti)->ip_len = tlen;
215 	((struct ip *)ti)->ip_ttl = ip_defttl;
216 #ifdef TCPDEBUG
217 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
218 		tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
219 #endif
220 	(void) ip_output(m, NULL, ro, 0, NULL);
221 }
222 
223 /*
224  * Create a new TCP control block, making an
225  * empty reassembly queue and hooking it to the argument
226  * protocol control block.
227  */
228 struct tcpcb *
229 tcp_newtcpcb(inp)
230 	struct inpcb *inp;
231 {
232 	register struct tcpcb *tp;
233 
234 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
235 	if (tp == NULL)
236 		return ((struct tcpcb *)0);
237 	bzero((char *) tp, sizeof(struct tcpcb));
238 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
239 	tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
240 
241 	if (tcp_do_rfc1323)
242 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
243 	if (tcp_do_rfc1644)
244 		tp->t_flags |= TF_REQ_CC;
245 	tp->t_inpcb = inp;
246 	/*
247 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
248 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
249 	 * reasonable initial retransmit time.
250 	 */
251 	tp->t_srtt = TCPTV_SRTTBASE;
252 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
253 	tp->t_rttmin = TCPTV_MIN;
254 	TCPT_RANGESET(tp->t_rxtcur,
255 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
256 	    TCPTV_MIN, TCPTV_REXMTMAX);
257 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
258 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
259 	inp->inp_ip.ip_ttl = ip_defttl;
260 	inp->inp_ppcb = (caddr_t)tp;
261 	return (tp);
262 }
263 
264 /*
265  * Drop a TCP connection, reporting
266  * the specified error.  If connection is synchronized,
267  * then send a RST to peer.
268  */
269 struct tcpcb *
270 tcp_drop(tp, errno)
271 	register struct tcpcb *tp;
272 	int errno;
273 {
274 	struct socket *so = tp->t_inpcb->inp_socket;
275 
276 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
277 		tp->t_state = TCPS_CLOSED;
278 		(void) tcp_output(tp);
279 		tcpstat.tcps_drops++;
280 	} else
281 		tcpstat.tcps_conndrops++;
282 	if (errno == ETIMEDOUT && tp->t_softerror)
283 		errno = tp->t_softerror;
284 	so->so_error = errno;
285 	return (tcp_close(tp));
286 }
287 
288 /*
289  * Close a TCP control block:
290  *	discard all space held by the tcp
291  *	discard internet protocol block
292  *	wake up any sleepers
293  */
294 struct tcpcb *
295 tcp_close(tp)
296 	register struct tcpcb *tp;
297 {
298 	register struct tcpiphdr *t;
299 	struct inpcb *inp = tp->t_inpcb;
300 	struct socket *so = inp->inp_socket;
301 	register struct mbuf *m;
302 #ifdef RTV_RTT
303 	register struct rtentry *rt;
304 
305 	/*
306 	 * If we got enough samples through the srtt filter,
307 	 * save the rtt and rttvar in the routing entry.
308 	 * 'Enough' is arbitrarily defined as the 16 samples.
309 	 * 16 samples is enough for the srtt filter to converge
310 	 * to within 5% of the correct value; fewer samples and
311 	 * we could save a very bogus rtt.
312 	 *
313 	 * Don't update the default route's characteristics and don't
314 	 * update anything that the user "locked".
315 	 */
316 	if (tp->t_rttupdated >= 16 &&
317 	    (rt = inp->inp_route.ro_rt) &&
318 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
319 		register u_long i = 0;
320 
321 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
322 			i = tp->t_srtt *
323 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
324 			if (rt->rt_rmx.rmx_rtt && i)
325 				/*
326 				 * filter this update to half the old & half
327 				 * the new values, converting scale.
328 				 * See route.h and tcp_var.h for a
329 				 * description of the scaling constants.
330 				 */
331 				rt->rt_rmx.rmx_rtt =
332 				    (rt->rt_rmx.rmx_rtt + i) / 2;
333 			else
334 				rt->rt_rmx.rmx_rtt = i;
335 			tcpstat.tcps_cachedrtt++;
336 		}
337 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
338 			i = tp->t_rttvar *
339 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
340 			if (rt->rt_rmx.rmx_rttvar && i)
341 				rt->rt_rmx.rmx_rttvar =
342 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
343 			else
344 				rt->rt_rmx.rmx_rttvar = i;
345 			tcpstat.tcps_cachedrttvar++;
346 		}
347 		/*
348 		 * update the pipelimit (ssthresh) if it has been updated
349 		 * already or if a pipesize was specified & the threshhold
350 		 * got below half the pipesize.  I.e., wait for bad news
351 		 * before we start updating, then update on both good
352 		 * and bad news.
353 		 */
354 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
355 		    ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) ||
356 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
357 			/*
358 			 * convert the limit from user data bytes to
359 			 * packets then to packet data bytes.
360 			 */
361 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
362 			if (i < 2)
363 				i = 2;
364 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
365 			if (rt->rt_rmx.rmx_ssthresh)
366 				rt->rt_rmx.rmx_ssthresh =
367 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
368 			else
369 				rt->rt_rmx.rmx_ssthresh = i;
370 			tcpstat.tcps_cachedssthresh++;
371 		}
372 	}
373 #endif /* RTV_RTT */
374 	/* free the reassembly queue, if any */
375 	t = tp->seg_next;
376 	while (t != (struct tcpiphdr *)tp) {
377 		t = (struct tcpiphdr *)t->ti_next;
378 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
379 		remque(t->ti_prev);
380 		m_freem(m);
381 	}
382 	if (tp->t_template)
383 		(void) m_free(dtom(tp->t_template));
384 	free(tp, M_PCB);
385 	inp->inp_ppcb = 0;
386 	soisdisconnected(so);
387 	in_pcbdetach(inp);
388 	tcpstat.tcps_closed++;
389 	return ((struct tcpcb *)0);
390 }
391 
392 void
393 tcp_drain()
394 {
395 
396 }
397 
398 /*
399  * Notify a tcp user of an asynchronous error;
400  * store error as soft error, but wake up user
401  * (for now, won't do anything until can select for soft error).
402  */
403 void
404 tcp_notify(inp, error)
405 	struct inpcb *inp;
406 	int error;
407 {
408 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
409 	register struct socket *so = inp->inp_socket;
410 
411 	/*
412 	 * Ignore some errors if we are hooked up.
413 	 * If connection hasn't completed, has retransmitted several times,
414 	 * and receives a second error, give up now.  This is better
415 	 * than waiting a long time to establish a connection that
416 	 * can never complete.
417 	 */
418 	if (tp->t_state == TCPS_ESTABLISHED &&
419 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
420 	      error == EHOSTDOWN)) {
421 		return;
422 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
423 	    tp->t_softerror)
424 		so->so_error = error;
425 	else
426 		tp->t_softerror = error;
427 	wakeup((caddr_t) &so->so_timeo);
428 	sorwakeup(so);
429 	sowwakeup(so);
430 }
431 
432 void
433 tcp_ctlinput(cmd, sa, ip)
434 	int cmd;
435 	struct sockaddr *sa;
436 	register struct ip *ip;
437 {
438 	register struct tcphdr *th;
439 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
440 
441 	if (cmd == PRC_QUENCH)
442 		notify = tcp_quench;
443 #ifdef MTUDISC
444 	else if (cmd == PRC_MSGSIZE)
445 		notify = tcp_mtudisc;
446 #endif /* MTUDISC */
447 	else if (!PRC_IS_REDIRECT(cmd) &&
448 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
449 		return;
450 	if (ip) {
451 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
452 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
453 			cmd, notify);
454 	} else
455 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
456 }
457 
458 /*
459  * When a source quench is received, close congestion window
460  * to one segment.  We will gradually open it again as we proceed.
461  */
462 void
463 tcp_quench(inp, errno)
464 	struct inpcb *inp;
465 	int errno;
466 {
467 	struct tcpcb *tp = intotcpcb(inp);
468 
469 	if (tp)
470 		tp->snd_cwnd = tp->t_maxseg;
471 }
472 
473 #ifdef MTUDISC
474 /*
475  * When `need fragmentation' ICMP is received, update our idea of the MSS
476  * based on the new value in the route.  Also nudge TCP to send something,
477  * since we know the packet we just sent was dropped.
478  * This duplicates some code in the tcp_mss() function in tcp_input.c.
479  */
480 void
481 tcp_mtudisc(inp, errno)
482 	struct inpcb *inp;
483 	int errno;
484 {
485 	struct tcpcb *tp = intotcpcb(inp);
486 	struct rtentry *rt;
487 	struct rmxp_tao *taop;
488 	struct socket *so = inp->inp_socket;
489 	int offered;
490 	int mss;
491 
492 	if (tp) {
493 		rt = tcp_rtlookup(inp);
494 		if (!rt || !rt->rt_rmx.rmx_mtu) {
495 			tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
496 			return;
497 		}
498 		taop = rmx_taop(rt->rt_rmx);
499 		offered = taop->tao_mssopt;
500 		mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
501 		mss = min(mss, offered);
502 		if (tp->t_maxopd <= mss)
503 			return;
504 		tp->t_maxopd = mss;
505 
506 		if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
507 		    (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
508 			mss -= TCPOLEN_TSTAMP_APPA;
509 		if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
510 		    (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
511 			mss -= TCPOLEN_CC_APPA;
512 #if	(MCLBYTES & (MCLBYTES - 1)) == 0
513 		if (mss > MCLBYTES)
514 			mss &= ~(MCLBYTES-1);
515 #else
516 		if (mss > MCLBYTES)
517 			mss = mss / MCLBYTES * MCLBYTES;
518 #endif
519 		if (so->so_snd.sb_hiwat < mss)
520 			mss = so->so_snd.sb_hiwat;
521 
522 		tp->t_maxseg = mss;
523 
524 		tcpstat.tcps_mturesent++;
525 		tp->t_rtt = 0;
526 		tp->snd_nxt = tp->snd_una;
527 		tcp_output(tp);
528 	}
529 }
530 #endif /* MTUDISC */
531 
532 /*
533  * Look-up the routing entry to the peer of this inpcb.  If no route
534  * is found and it cannot be allocated the return NULL.  This routine
535  * is called by TCP routines that access the rmx structure and by tcp_mss
536  * to get the interface MTU.
537  */
538 struct rtentry *
539 tcp_rtlookup(inp)
540 	struct inpcb *inp;
541 {
542 	struct route *ro;
543 	struct rtentry *rt;
544 
545 	ro = &inp->inp_route;
546 	rt = ro->ro_rt;
547 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
548 		/* No route yet, so try to acquire one */
549 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
550 			ro->ro_dst.sa_family = AF_INET;
551 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
552 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
553 				inp->inp_faddr;
554 			rtalloc(ro);
555 			rt = ro->ro_rt;
556 		}
557 	}
558 	return rt;
559 }
560 
561 /*
562  * Return a pointer to the cached information about the remote host.
563  * The cached information is stored in the protocol specific part of
564  * the route metrics.
565  */
566 struct rmxp_tao *
567 tcp_gettaocache(inp)
568 	struct inpcb *inp;
569 {
570 	struct rtentry *rt = tcp_rtlookup(inp);
571 
572 	/* Make sure this is a host route and is up. */
573 	if (rt == NULL ||
574 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
575 		return NULL;
576 
577 	return rmx_taop(rt->rt_rmx);
578 }
579 
580 /*
581  * Clear all the TAO cache entries, called from tcp_init.
582  *
583  * XXX
584  * This routine is just an empty one, because we assume that the routing
585  * routing tables are initialized at the same time when TCP, so there is
586  * nothing in the cache left over.
587  */
588 static void
589 tcp_cleartaocache(void)
590 { }
591