xref: /freebsd/sys/netinet/tcp_timewait.c (revision 17ee9d00bc1ae1e598c38f25826f861e4bc6c3ce)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
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	8.1 (Berkeley) 6/10/93
34  * $Id: tcp_subr.c,v 1.6 1995/02/09 23:13:25 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 
47 #include <net/route.h>
48 #include <net/if.h>
49 
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/ip.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip_var.h>
55 #include <netinet/ip_icmp.h>
56 #include <netinet/tcp.h>
57 #define	TCPOUTFLAGS
58 #include <netinet/tcp_fsm.h>
59 #include <netinet/tcp_seq.h>
60 #include <netinet/tcp_timer.h>
61 #include <netinet/tcp_var.h>
62 #include <netinet/tcpip.h>
63 #ifdef TCPDEBUG
64 #include <netinet/tcp_debug.h>
65 #endif
66 
67 /* patchable/settable parameters for tcp */
68 int 	tcp_mssdflt = TCP_MSS;
69 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
70 int	tcp_do_rfc1323 = 1;
71 int	tcp_do_rfc1644 = 1;
72 static	void tcp_cleartaocache(void);
73 
74 extern	struct inpcb *tcp_last_inpcb;
75 
76 /*
77  * Tcp initialization
78  */
79 void
80 tcp_init()
81 {
82 
83 	tcp_iss = 1;		/* wrong */
84 	tcp_ccgen = 1;
85 	tcp_cleartaocache();
86 	tcb.inp_next = tcb.inp_prev = &tcb;
87 	if (max_protohdr < sizeof(struct tcpiphdr))
88 		max_protohdr = sizeof(struct tcpiphdr);
89 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
90 		panic("tcp_init");
91 }
92 
93 /*
94  * Create template to be used to send tcp packets on a connection.
95  * Call after host entry created, allocates an mbuf and fills
96  * in a skeletal tcp/ip header, minimizing the amount of work
97  * necessary when the connection is used.
98  */
99 struct tcpiphdr *
100 tcp_template(tp)
101 	struct tcpcb *tp;
102 {
103 	register struct inpcb *inp = tp->t_inpcb;
104 	register struct mbuf *m;
105 	register struct tcpiphdr *n;
106 
107 	if ((n = tp->t_template) == 0) {
108 		m = m_get(M_DONTWAIT, MT_HEADER);
109 		if (m == NULL)
110 			return (0);
111 		m->m_len = sizeof (struct tcpiphdr);
112 		n = mtod(m, struct tcpiphdr *);
113 	}
114 	n->ti_next = n->ti_prev = 0;
115 	n->ti_x1 = 0;
116 	n->ti_pr = IPPROTO_TCP;
117 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
118 	n->ti_src = inp->inp_laddr;
119 	n->ti_dst = inp->inp_faddr;
120 	n->ti_sport = inp->inp_lport;
121 	n->ti_dport = inp->inp_fport;
122 	n->ti_seq = 0;
123 	n->ti_ack = 0;
124 	n->ti_x2 = 0;
125 	n->ti_off = 5;
126 	n->ti_flags = 0;
127 	n->ti_win = 0;
128 	n->ti_sum = 0;
129 	n->ti_urp = 0;
130 	return (n);
131 }
132 
133 /*
134  * Send a single message to the TCP at address specified by
135  * the given TCP/IP header.  If m == 0, then we make a copy
136  * of the tcpiphdr at ti and send directly to the addressed host.
137  * This is used to force keep alive messages out using the TCP
138  * template for a connection tp->t_template.  If flags are given
139  * then we send a message back to the TCP which originated the
140  * segment ti, and discard the mbuf containing it and any other
141  * attached mbufs.
142  *
143  * In any case the ack and sequence number of the transmitted
144  * segment are as specified by the parameters.
145  */
146 void
147 tcp_respond(tp, ti, m, ack, seq, flags)
148 	struct tcpcb *tp;
149 	register struct tcpiphdr *ti;
150 	register struct mbuf *m;
151 	tcp_seq ack, seq;
152 	int flags;
153 {
154 	register int tlen;
155 	int win = 0;
156 	struct route *ro = 0;
157 
158 	if (tp) {
159 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
160 		ro = &tp->t_inpcb->inp_route;
161 	}
162 	if (m == 0) {
163 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
164 		if (m == NULL)
165 			return;
166 #ifdef TCP_COMPAT_42
167 		tlen = 1;
168 #else
169 		tlen = 0;
170 #endif
171 		m->m_data += max_linkhdr;
172 		*mtod(m, struct tcpiphdr *) = *ti;
173 		ti = mtod(m, struct tcpiphdr *);
174 		flags = TH_ACK;
175 	} else {
176 		m_freem(m->m_next);
177 		m->m_next = 0;
178 		m->m_data = (caddr_t)ti;
179 		m->m_len = sizeof (struct tcpiphdr);
180 		tlen = 0;
181 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
182 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
183 		xchg(ti->ti_dport, ti->ti_sport, u_short);
184 #undef xchg
185 	}
186 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
187 	tlen += sizeof (struct tcpiphdr);
188 	m->m_len = tlen;
189 	m->m_pkthdr.len = tlen;
190 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
191 	ti->ti_next = ti->ti_prev = 0;
192 	ti->ti_x1 = 0;
193 	ti->ti_seq = htonl(seq);
194 	ti->ti_ack = htonl(ack);
195 	ti->ti_x2 = 0;
196 	ti->ti_off = sizeof (struct tcphdr) >> 2;
197 	ti->ti_flags = flags;
198 	if (tp)
199 		ti->ti_win = htons((u_short) (win >> tp->rcv_scale));
200 	else
201 		ti->ti_win = htons((u_short)win);
202 	ti->ti_urp = 0;
203 	ti->ti_sum = 0;
204 	ti->ti_sum = in_cksum(m, tlen);
205 	((struct ip *)ti)->ip_len = tlen;
206 	((struct ip *)ti)->ip_ttl = ip_defttl;
207 #ifdef TCPDEBUG
208 	if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
209 		tcp_trace(TA_OUTPUT, 0, tp, ti, 0);
210 #endif
211 	(void) ip_output(m, NULL, ro, 0, NULL);
212 }
213 
214 /*
215  * Create a new TCP control block, making an
216  * empty reassembly queue and hooking it to the argument
217  * protocol control block.
218  */
219 struct tcpcb *
220 tcp_newtcpcb(inp)
221 	struct inpcb *inp;
222 {
223 	register struct tcpcb *tp;
224 
225 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
226 	if (tp == NULL)
227 		return ((struct tcpcb *)0);
228 	bzero((char *) tp, sizeof(struct tcpcb));
229 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
230 	tp->t_maxseg = tp->t_maxopd = tcp_mssdflt;
231 
232 	if (tcp_do_rfc1323)
233 		tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
234 	if (tcp_do_rfc1644)
235 		tp->t_flags |= TF_REQ_CC;
236 	tp->t_inpcb = inp;
237 	/*
238 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
239 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
240 	 * reasonable initial retransmit time.
241 	 */
242 	tp->t_srtt = TCPTV_SRTTBASE;
243 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
244 	tp->t_rttmin = TCPTV_MIN;
245 	TCPT_RANGESET(tp->t_rxtcur,
246 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
247 	    TCPTV_MIN, TCPTV_REXMTMAX);
248 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
249 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
250 	inp->inp_ip.ip_ttl = ip_defttl;
251 	inp->inp_ppcb = (caddr_t)tp;
252 	return (tp);
253 }
254 
255 /*
256  * Drop a TCP connection, reporting
257  * the specified error.  If connection is synchronized,
258  * then send a RST to peer.
259  */
260 struct tcpcb *
261 tcp_drop(tp, errno)
262 	register struct tcpcb *tp;
263 	int errno;
264 {
265 	struct socket *so = tp->t_inpcb->inp_socket;
266 
267 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
268 		tp->t_state = TCPS_CLOSED;
269 		(void) tcp_output(tp);
270 		tcpstat.tcps_drops++;
271 	} else
272 		tcpstat.tcps_conndrops++;
273 	if (errno == ETIMEDOUT && tp->t_softerror)
274 		errno = tp->t_softerror;
275 	so->so_error = errno;
276 	return (tcp_close(tp));
277 }
278 
279 /*
280  * Close a TCP control block:
281  *	discard all space held by the tcp
282  *	discard internet protocol block
283  *	wake up any sleepers
284  */
285 struct tcpcb *
286 tcp_close(tp)
287 	register struct tcpcb *tp;
288 {
289 	register struct tcpiphdr *t;
290 	struct inpcb *inp = tp->t_inpcb;
291 	struct socket *so = inp->inp_socket;
292 	register struct mbuf *m;
293 #ifdef RTV_RTT
294 	register struct rtentry *rt;
295 
296 	/*
297 	 * If we sent enough data to get some meaningful characteristics,
298 	 * save them in the routing entry.  'Enough' is arbitrarily
299 	 * defined as the sendpipesize (default 4K) * 16.  This would
300 	 * give us 16 rtt samples assuming we only get one sample per
301 	 * window (the usual case on a long haul net).  16 samples is
302 	 * enough for the srtt filter to converge to within 5% of the correct
303 	 * value; fewer samples and we could save a very bogus rtt.
304 	 *
305 	 * Don't update the default route's characteristics and don't
306 	 * update anything that the user "locked".
307 	 */
308 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
309 	    (rt = inp->inp_route.ro_rt) &&
310 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
311 		register u_long i = 0;
312 
313 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
314 			i = tp->t_srtt *
315 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
316 			if (rt->rt_rmx.rmx_rtt && i)
317 				/*
318 				 * filter this update to half the old & half
319 				 * the new values, converting scale.
320 				 * See route.h and tcp_var.h for a
321 				 * description of the scaling constants.
322 				 */
323 				rt->rt_rmx.rmx_rtt =
324 				    (rt->rt_rmx.rmx_rtt + i) / 2;
325 			else
326 				rt->rt_rmx.rmx_rtt = i;
327 		}
328 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
329 			i = tp->t_rttvar *
330 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
331 			if (rt->rt_rmx.rmx_rttvar && i)
332 				rt->rt_rmx.rmx_rttvar =
333 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
334 			else
335 				rt->rt_rmx.rmx_rttvar = i;
336 		}
337 		/*
338 		 * update the pipelimit (ssthresh) if it has been updated
339 		 * already or if a pipesize was specified & the threshhold
340 		 * got below half the pipesize.  I.e., wait for bad news
341 		 * before we start updating, then update on both good
342 		 * and bad news.
343 		 */
344 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
345 		    ((i = tp->snd_ssthresh) != 0) && rt->rt_rmx.rmx_ssthresh) ||
346 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
347 			/*
348 			 * convert the limit from user data bytes to
349 			 * packets then to packet data bytes.
350 			 */
351 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
352 			if (i < 2)
353 				i = 2;
354 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
355 			if (rt->rt_rmx.rmx_ssthresh)
356 				rt->rt_rmx.rmx_ssthresh =
357 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
358 			else
359 				rt->rt_rmx.rmx_ssthresh = i;
360 		}
361 	}
362 #endif /* RTV_RTT */
363 	/* free the reassembly queue, if any */
364 	t = tp->seg_next;
365 	while (t != (struct tcpiphdr *)tp) {
366 		t = (struct tcpiphdr *)t->ti_next;
367 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
368 		remque(t->ti_prev);
369 		m_freem(m);
370 	}
371 	if (tp->t_template)
372 		(void) m_free(dtom(tp->t_template));
373 	free(tp, M_PCB);
374 	inp->inp_ppcb = 0;
375 	soisdisconnected(so);
376 	/* clobber input pcb cache if we're closing the cached connection */
377 	if (inp == tcp_last_inpcb)
378 		tcp_last_inpcb = &tcb;
379 	in_pcbdetach(inp);
380 	tcpstat.tcps_closed++;
381 	return ((struct tcpcb *)0);
382 }
383 
384 void
385 tcp_drain()
386 {
387 
388 }
389 
390 /*
391  * Notify a tcp user of an asynchronous error;
392  * store error as soft error, but wake up user
393  * (for now, won't do anything until can select for soft error).
394  */
395 void
396 tcp_notify(inp, error)
397 	struct inpcb *inp;
398 	int error;
399 {
400 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
401 	register struct socket *so = inp->inp_socket;
402 
403 	/*
404 	 * Ignore some errors if we are hooked up.
405 	 * If connection hasn't completed, has retransmitted several times,
406 	 * and receives a second error, give up now.  This is better
407 	 * than waiting a long time to establish a connection that
408 	 * can never complete.
409 	 */
410 	if (tp->t_state == TCPS_ESTABLISHED &&
411 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
412 	      error == EHOSTDOWN)) {
413 		return;
414 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
415 	    tp->t_softerror)
416 		so->so_error = error;
417 	else
418 		tp->t_softerror = error;
419 	wakeup((caddr_t) &so->so_timeo);
420 	sorwakeup(so);
421 	sowwakeup(so);
422 }
423 
424 void
425 tcp_ctlinput(cmd, sa, ip)
426 	int cmd;
427 	struct sockaddr *sa;
428 	register struct ip *ip;
429 {
430 	register struct tcphdr *th;
431 	extern struct in_addr zeroin_addr;
432 	extern u_char inetctlerrmap[];
433 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
434 
435 	if (cmd == PRC_QUENCH)
436 		notify = tcp_quench;
437 	else if (!PRC_IS_REDIRECT(cmd) &&
438 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
439 		return;
440 	if (ip) {
441 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
442 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
443 			cmd, notify);
444 	} else
445 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
446 }
447 
448 /*
449  * When a source quench is received, close congestion window
450  * to one segment.  We will gradually open it again as we proceed.
451  */
452 void
453 tcp_quench(inp, errno)
454 	struct inpcb *inp;
455 	int errno;
456 {
457 	struct tcpcb *tp = intotcpcb(inp);
458 
459 	if (tp)
460 		tp->snd_cwnd = tp->t_maxseg;
461 }
462 
463 /*
464  * Look-up the routing entry to the peer of this inpcb.  If no route
465  * is found and it cannot be allocated the return NULL.  This routine
466  * is called by TCP routines that access the rmx structure and by tcp_mss
467  * to get the interface MTU.
468  */
469 struct rtentry *
470 tcp_rtlookup(inp)
471 	struct inpcb *inp;
472 {
473 	struct route *ro;
474 	struct rtentry *rt;
475 
476 	ro = &inp->inp_route;
477 	rt = ro->ro_rt;
478 	if (rt == NULL || !(rt->rt_flags & RTF_UP)) {
479 		/* No route yet, so try to acquire one */
480 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
481 			ro->ro_dst.sa_family = AF_INET;
482 			ro->ro_dst.sa_len = sizeof(ro->ro_dst);
483 			((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
484 				inp->inp_faddr;
485 			rtalloc(ro);
486 			rt = ro->ro_rt;
487 		}
488 	}
489 	return rt;
490 }
491 
492 /*
493  * Return a pointer to the cached information about the remote host.
494  * The cached information is stored in the protocol specific part of
495  * the route metrics.
496  */
497 struct rmxp_tao *
498 tcp_gettaocache(inp)
499 	struct inpcb *inp;
500 {
501 	struct rtentry *rt = tcp_rtlookup(inp);
502 
503 	/* Make sure this is a host route and is up. */
504 	if (rt == NULL ||
505 	    (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST))
506 		return NULL;
507 
508 	return rmx_taop(rt->rt_rmx);
509 }
510 
511 /*
512  * Clear all the TAO cache entries, called from tcp_init.
513  *
514  * XXX
515  * This routine is just an empty one, because we assume that the routing
516  * routing tables are initialized at the same time when TCP, so there is
517  * nothing in the cache left over.
518  */
519 static void
520 tcp_cleartaocache(void)
521 { }
522