xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 1394f01b4a9e99092957ca5d824d67219565d9b5)
1 /*	$NetBSD: tcp_subr.c,v 1.26 1997/06/24 02:26:06 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)tcp_subr.c	8.1 (Berkeley) 6/10/93
36  */
37 
38 #include <sys/param.h>
39 #include <sys/proc.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.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/ip_var.h>
56 #include <netinet/ip_icmp.h>
57 #include <netinet/tcp.h>
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 
64 /* patchable/settable parameters for tcp */
65 int 	tcp_mssdflt = TCP_MSS;
66 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
67 int	tcp_do_rfc1323 = 1;
68 
69 #ifndef TCBHASHSIZE
70 #define	TCBHASHSIZE	128
71 #endif
72 int	tcbhashsize = TCBHASHSIZE;
73 
74 /*
75  * Tcp initialization
76  */
77 void
78 tcp_init()
79 {
80 
81 	tcp_iss = 1;		/* XXX wrong */
82 	in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
83 	if (max_protohdr < sizeof(struct tcpiphdr))
84 		max_protohdr = sizeof(struct tcpiphdr);
85 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
86 		panic("tcp_init");
87 }
88 
89 /*
90  * Create template to be used to send tcp packets on a connection.
91  * Call after host entry created, allocates an mbuf and fills
92  * in a skeletal tcp/ip header, minimizing the amount of work
93  * necessary when the connection is used.
94  */
95 struct tcpiphdr *
96 tcp_template(tp)
97 	struct tcpcb *tp;
98 {
99 	register struct inpcb *inp = tp->t_inpcb;
100 	register struct tcpiphdr *n;
101 
102 	if ((n = tp->t_template) == 0) {
103 		MALLOC(n, struct tcpiphdr *, sizeof (struct tcpiphdr),
104 		    M_MBUF, M_NOWAIT);
105 		if (n == NULL)
106 			return (0);
107 	}
108 	bzero(n->ti_x1, sizeof n->ti_x1);
109 	n->ti_pr = IPPROTO_TCP;
110 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
111 	n->ti_src = inp->inp_laddr;
112 	n->ti_dst = inp->inp_faddr;
113 	n->ti_sport = inp->inp_lport;
114 	n->ti_dport = inp->inp_fport;
115 	n->ti_seq = 0;
116 	n->ti_ack = 0;
117 	n->ti_x2 = 0;
118 	n->ti_off = 5;
119 	n->ti_flags = 0;
120 	n->ti_win = 0;
121 	n->ti_sum = 0;
122 	n->ti_urp = 0;
123 	return (n);
124 }
125 
126 /*
127  * Send a single message to the TCP at address specified by
128  * the given TCP/IP header.  If m == 0, then we make a copy
129  * of the tcpiphdr at ti and send directly to the addressed host.
130  * This is used to force keep alive messages out using the TCP
131  * template for a connection tp->t_template.  If flags are given
132  * then we send a message back to the TCP which originated the
133  * segment ti, and discard the mbuf containing it and any other
134  * attached mbufs.
135  *
136  * In any case the ack and sequence number of the transmitted
137  * segment are as specified by the parameters.
138  */
139 void
140 tcp_respond(tp, ti, m, ack, seq, flags)
141 	struct tcpcb *tp;
142 	register struct tcpiphdr *ti;
143 	register struct mbuf *m;
144 	tcp_seq ack, seq;
145 	int flags;
146 {
147 	register int tlen;
148 	int win = 0;
149 	struct route *ro = 0;
150 
151 	if (tp) {
152 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
153 		ro = &tp->t_inpcb->inp_route;
154 	}
155 	if (m == 0) {
156 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
157 		if (m == NULL)
158 			return;
159 #ifdef TCP_COMPAT_42
160 		tlen = 1;
161 #else
162 		tlen = 0;
163 #endif
164 		m->m_data += max_linkhdr;
165 		*mtod(m, struct tcpiphdr *) = *ti;
166 		ti = mtod(m, struct tcpiphdr *);
167 		flags = TH_ACK;
168 	} else {
169 		m_freem(m->m_next);
170 		m->m_next = 0;
171 		m->m_data = (caddr_t)ti;
172 		m->m_len = sizeof (struct tcpiphdr);
173 		tlen = 0;
174 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
175 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
176 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
177 #undef xchg
178 	}
179 	ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) + tlen));
180 	tlen += sizeof (struct tcpiphdr);
181 	m->m_len = tlen;
182 	m->m_pkthdr.len = tlen;
183 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
184 	bzero(ti->ti_x1, sizeof ti->ti_x1);
185 	ti->ti_seq = htonl(seq);
186 	ti->ti_ack = htonl(ack);
187 	ti->ti_x2 = 0;
188 	ti->ti_off = sizeof (struct tcphdr) >> 2;
189 	ti->ti_flags = flags;
190 	if (tp)
191 		ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
192 	else
193 		ti->ti_win = htons((u_int16_t)win);
194 	ti->ti_urp = 0;
195 	ti->ti_sum = 0;
196 	ti->ti_sum = in_cksum(m, tlen);
197 	((struct ip *)ti)->ip_len = tlen;
198 	((struct ip *)ti)->ip_ttl = ip_defttl;
199 	(void) ip_output(m, NULL, ro, 0, NULL);
200 }
201 
202 /*
203  * Create a new TCP control block, making an
204  * empty reassembly queue and hooking it to the argument
205  * protocol control block.
206  */
207 struct tcpcb *
208 tcp_newtcpcb(inp)
209 	struct inpcb *inp;
210 {
211 	register struct tcpcb *tp;
212 
213 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
214 	if (tp == NULL)
215 		return ((struct tcpcb *)0);
216 	bzero((caddr_t)tp, sizeof(struct tcpcb));
217 	LIST_INIT(&tp->segq);
218 	tp->t_maxseg = tcp_mssdflt;
219 
220 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
221 	tp->t_inpcb = inp;
222 	/*
223 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
224 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
225 	 * reasonable initial retransmit time.
226 	 */
227 	tp->t_srtt = TCPTV_SRTTBASE;
228 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
229 	tp->t_rttmin = TCPTV_MIN;
230 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
231 	    TCPTV_MIN, TCPTV_REXMTMAX);
232 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
233 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
234 	inp->inp_ip.ip_ttl = ip_defttl;
235 	inp->inp_ppcb = (caddr_t)tp;
236 	return (tp);
237 }
238 
239 /*
240  * Drop a TCP connection, reporting
241  * the specified error.  If connection is synchronized,
242  * then send a RST to peer.
243  */
244 struct tcpcb *
245 tcp_drop(tp, errno)
246 	register struct tcpcb *tp;
247 	int errno;
248 {
249 	struct socket *so = tp->t_inpcb->inp_socket;
250 
251 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
252 		tp->t_state = TCPS_CLOSED;
253 		(void) tcp_output(tp);
254 		tcpstat.tcps_drops++;
255 	} else
256 		tcpstat.tcps_conndrops++;
257 	if (errno == ETIMEDOUT && tp->t_softerror)
258 		errno = tp->t_softerror;
259 	so->so_error = errno;
260 	return (tcp_close(tp));
261 }
262 
263 /*
264  * Close a TCP control block:
265  *	discard all space held by the tcp
266  *	discard internet protocol block
267  *	wake up any sleepers
268  */
269 struct tcpcb *
270 tcp_close(tp)
271 	register struct tcpcb *tp;
272 {
273 	register struct ipqent *qe;
274 	struct inpcb *inp = tp->t_inpcb;
275 	struct socket *so = inp->inp_socket;
276 #ifdef RTV_RTT
277 	register struct rtentry *rt;
278 
279 	/*
280 	 * If we sent enough data to get some meaningful characteristics,
281 	 * save them in the routing entry.  'Enough' is arbitrarily
282 	 * defined as the sendpipesize (default 4K) * 16.  This would
283 	 * give us 16 rtt samples assuming we only get one sample per
284 	 * window (the usual case on a long haul net).  16 samples is
285 	 * enough for the srtt filter to converge to within 5% of the correct
286 	 * value; fewer samples and we could save a very bogus rtt.
287 	 *
288 	 * Don't update the default route's characteristics and don't
289 	 * update anything that the user "locked".
290 	 */
291 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
292 	    (rt = inp->inp_route.ro_rt) &&
293 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
294 		register u_long i = 0;
295 
296 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
297 			i = tp->t_srtt *
298 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
299 			if (rt->rt_rmx.rmx_rtt && i)
300 				/*
301 				 * filter this update to half the old & half
302 				 * the new values, converting scale.
303 				 * See route.h and tcp_var.h for a
304 				 * description of the scaling constants.
305 				 */
306 				rt->rt_rmx.rmx_rtt =
307 				    (rt->rt_rmx.rmx_rtt + i) / 2;
308 			else
309 				rt->rt_rmx.rmx_rtt = i;
310 		}
311 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
312 			i = tp->t_rttvar *
313 			    ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
314 			if (rt->rt_rmx.rmx_rttvar && i)
315 				rt->rt_rmx.rmx_rttvar =
316 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
317 			else
318 				rt->rt_rmx.rmx_rttvar = i;
319 		}
320 		/*
321 		 * update the pipelimit (ssthresh) if it has been updated
322 		 * already or if a pipesize was specified & the threshhold
323 		 * got below half the pipesize.  I.e., wait for bad news
324 		 * before we start updating, then update on both good
325 		 * and bad news.
326 		 */
327 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
328 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
329 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
330 			/*
331 			 * convert the limit from user data bytes to
332 			 * packets then to packet data bytes.
333 			 */
334 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
335 			if (i < 2)
336 				i = 2;
337 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
338 			if (rt->rt_rmx.rmx_ssthresh)
339 				rt->rt_rmx.rmx_ssthresh =
340 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
341 			else
342 				rt->rt_rmx.rmx_ssthresh = i;
343 		}
344 	}
345 #endif /* RTV_RTT */
346 	/* free the reassembly queue, if any */
347 	while ((qe = tp->segq.lh_first) != NULL) {
348 		LIST_REMOVE(qe, ipqe_q);
349 		m_freem(qe->ipqe_m);
350 		FREE(qe, M_IPQ);
351 	}
352 	if (tp->t_template)
353 		FREE(tp->t_template, M_MBUF);
354 	free(tp, M_PCB);
355 	inp->inp_ppcb = 0;
356 	soisdisconnected(so);
357 	in_pcbdetach(inp);
358 	tcpstat.tcps_closed++;
359 	return ((struct tcpcb *)0);
360 }
361 
362 void
363 tcp_drain()
364 {
365 
366 }
367 
368 /*
369  * Notify a tcp user of an asynchronous error;
370  * store error as soft error, but wake up user
371  * (for now, won't do anything until can select for soft error).
372  */
373 void
374 tcp_notify(inp, error)
375 	struct inpcb *inp;
376 	int error;
377 {
378 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
379 	register struct socket *so = inp->inp_socket;
380 
381 	/*
382 	 * Ignore some errors if we are hooked up.
383 	 * If connection hasn't completed, has retransmitted several times,
384 	 * and receives a second error, give up now.  This is better
385 	 * than waiting a long time to establish a connection that
386 	 * can never complete.
387 	 */
388 	if (tp->t_state == TCPS_ESTABLISHED &&
389 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
390 	      error == EHOSTDOWN)) {
391 		return;
392 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
393 	    tp->t_rxtshift > 3 && tp->t_softerror)
394 		so->so_error = error;
395 	else
396 		tp->t_softerror = error;
397 	wakeup((caddr_t) &so->so_timeo);
398 	sorwakeup(so);
399 	sowwakeup(so);
400 }
401 
402 void *
403 tcp_ctlinput(cmd, sa, v)
404 	int cmd;
405 	struct sockaddr *sa;
406 	register void *v;
407 {
408 	register struct ip *ip = v;
409 	register struct tcphdr *th;
410 	extern int inetctlerrmap[];
411 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
412 	int errno;
413 
414 	if ((unsigned)cmd >= PRC_NCMDS)
415 		return NULL;
416 	errno = inetctlerrmap[cmd];
417 	if (cmd == PRC_QUENCH)
418 		notify = tcp_quench;
419 	else if (PRC_IS_REDIRECT(cmd))
420 		notify = in_rtchange, ip = 0;
421 	else if (cmd == PRC_HOSTDEAD)
422 		ip = 0;
423 	else if (errno == 0)
424 		return NULL;
425 	if (ip) {
426 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
427 		in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, th->th_dport,
428 		    ip->ip_src, th->th_sport, errno, notify);
429 	} else
430 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
431 		    notify);
432 	return NULL;
433 }
434 
435 /*
436  * When a source quench is received, close congestion window
437  * to one segment.  We will gradually open it again as we proceed.
438  */
439 void
440 tcp_quench(inp, errno)
441 	struct inpcb *inp;
442 	int errno;
443 {
444 	struct tcpcb *tp = intotcpcb(inp);
445 
446 	if (tp)
447 		tp->snd_cwnd = tp->t_maxseg;
448 }
449