xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 76dfffe33547c37f8bdd446e3e4ab0f3c16cea4b)
1 /*	$NetBSD: tcp_subr.c,v 1.24 1996/09/15 18:11:10 mycroft 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 mbuf *m;
101 	register struct tcpiphdr *n;
102 
103 	if ((n = tp->t_template) == 0) {
104 		m = m_get(M_DONTWAIT, MT_HEADER);
105 		if (m == NULL)
106 			return (0);
107 		m->m_len = sizeof (struct tcpiphdr);
108 		n = mtod(m, struct tcpiphdr *);
109 	}
110 	bzero(n->ti_x1, sizeof n->ti_x1);
111 	n->ti_pr = IPPROTO_TCP;
112 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
113 	n->ti_src = inp->inp_laddr;
114 	n->ti_dst = inp->inp_faddr;
115 	n->ti_sport = inp->inp_lport;
116 	n->ti_dport = inp->inp_fport;
117 	n->ti_seq = 0;
118 	n->ti_ack = 0;
119 	n->ti_x2 = 0;
120 	n->ti_off = 5;
121 	n->ti_flags = 0;
122 	n->ti_win = 0;
123 	n->ti_sum = 0;
124 	n->ti_urp = 0;
125 	return (n);
126 }
127 
128 /*
129  * Send a single message to the TCP at address specified by
130  * the given TCP/IP header.  If m == 0, then we make a copy
131  * of the tcpiphdr at ti and send directly to the addressed host.
132  * This is used to force keep alive messages out using the TCP
133  * template for a connection tp->t_template.  If flags are given
134  * then we send a message back to the TCP which originated the
135  * segment ti, and discard the mbuf containing it and any other
136  * attached mbufs.
137  *
138  * In any case the ack and sequence number of the transmitted
139  * segment are as specified by the parameters.
140  */
141 void
142 tcp_respond(tp, ti, m, ack, seq, flags)
143 	struct tcpcb *tp;
144 	register struct tcpiphdr *ti;
145 	register struct mbuf *m;
146 	tcp_seq ack, seq;
147 	int flags;
148 {
149 	register int tlen;
150 	int win = 0;
151 	struct route *ro = 0;
152 
153 	if (tp) {
154 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
155 		ro = &tp->t_inpcb->inp_route;
156 	}
157 	if (m == 0) {
158 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
159 		if (m == NULL)
160 			return;
161 #ifdef TCP_COMPAT_42
162 		tlen = 1;
163 #else
164 		tlen = 0;
165 #endif
166 		m->m_data += max_linkhdr;
167 		*mtod(m, struct tcpiphdr *) = *ti;
168 		ti = mtod(m, struct tcpiphdr *);
169 		flags = TH_ACK;
170 	} else {
171 		m_freem(m->m_next);
172 		m->m_next = 0;
173 		m->m_data = (caddr_t)ti;
174 		m->m_len = sizeof (struct tcpiphdr);
175 		tlen = 0;
176 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
177 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
178 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
179 #undef xchg
180 	}
181 	ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) + tlen));
182 	tlen += sizeof (struct tcpiphdr);
183 	m->m_len = tlen;
184 	m->m_pkthdr.len = tlen;
185 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
186 	bzero(ti->ti_x1, sizeof ti->ti_x1);
187 	ti->ti_seq = htonl(seq);
188 	ti->ti_ack = htonl(ack);
189 	ti->ti_x2 = 0;
190 	ti->ti_off = sizeof (struct tcphdr) >> 2;
191 	ti->ti_flags = flags;
192 	if (tp)
193 		ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
194 	else
195 		ti->ti_win = htons((u_int16_t)win);
196 	ti->ti_urp = 0;
197 	ti->ti_sum = 0;
198 	ti->ti_sum = in_cksum(m, tlen);
199 	((struct ip *)ti)->ip_len = tlen;
200 	((struct ip *)ti)->ip_ttl = ip_defttl;
201 	(void) ip_output(m, NULL, ro, 0, NULL);
202 }
203 
204 /*
205  * Create a new TCP control block, making an
206  * empty reassembly queue and hooking it to the argument
207  * protocol control block.
208  */
209 struct tcpcb *
210 tcp_newtcpcb(inp)
211 	struct inpcb *inp;
212 {
213 	register struct tcpcb *tp;
214 
215 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
216 	if (tp == NULL)
217 		return ((struct tcpcb *)0);
218 	bzero((caddr_t)tp, sizeof(struct tcpcb));
219 	LIST_INIT(&tp->segq);
220 	tp->t_maxseg = tcp_mssdflt;
221 
222 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
223 	tp->t_inpcb = inp;
224 	/*
225 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
226 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
227 	 * reasonable initial retransmit time.
228 	 */
229 	tp->t_srtt = TCPTV_SRTTBASE;
230 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
231 	tp->t_rttmin = TCPTV_MIN;
232 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
233 	    TCPTV_MIN, TCPTV_REXMTMAX);
234 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
235 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
236 	inp->inp_ip.ip_ttl = ip_defttl;
237 	inp->inp_ppcb = (caddr_t)tp;
238 	return (tp);
239 }
240 
241 /*
242  * Drop a TCP connection, reporting
243  * the specified error.  If connection is synchronized,
244  * then send a RST to peer.
245  */
246 struct tcpcb *
247 tcp_drop(tp, errno)
248 	register struct tcpcb *tp;
249 	int errno;
250 {
251 	struct socket *so = tp->t_inpcb->inp_socket;
252 
253 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
254 		tp->t_state = TCPS_CLOSED;
255 		(void) tcp_output(tp);
256 		tcpstat.tcps_drops++;
257 	} else
258 		tcpstat.tcps_conndrops++;
259 	if (errno == ETIMEDOUT && tp->t_softerror)
260 		errno = tp->t_softerror;
261 	so->so_error = errno;
262 	return (tcp_close(tp));
263 }
264 
265 /*
266  * Close a TCP control block:
267  *	discard all space held by the tcp
268  *	discard internet protocol block
269  *	wake up any sleepers
270  */
271 struct tcpcb *
272 tcp_close(tp)
273 	register struct tcpcb *tp;
274 {
275 	register struct ipqent *qe;
276 	struct inpcb *inp = tp->t_inpcb;
277 	struct socket *so = inp->inp_socket;
278 #ifdef RTV_RTT
279 	register struct rtentry *rt;
280 
281 	/*
282 	 * If we sent enough data to get some meaningful characteristics,
283 	 * save them in the routing entry.  'Enough' is arbitrarily
284 	 * defined as the sendpipesize (default 4K) * 16.  This would
285 	 * give us 16 rtt samples assuming we only get one sample per
286 	 * window (the usual case on a long haul net).  16 samples is
287 	 * enough for the srtt filter to converge to within 5% of the correct
288 	 * value; fewer samples and we could save a very bogus rtt.
289 	 *
290 	 * Don't update the default route's characteristics and don't
291 	 * update anything that the user "locked".
292 	 */
293 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
294 	    (rt = inp->inp_route.ro_rt) &&
295 	    !in_nullhost(satosin(rt_key(rt))->sin_addr)) {
296 		register u_long i = 0;
297 
298 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
299 			i = tp->t_srtt *
300 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
301 			if (rt->rt_rmx.rmx_rtt && i)
302 				/*
303 				 * filter this update to half the old & half
304 				 * the new values, converting scale.
305 				 * See route.h and tcp_var.h for a
306 				 * description of the scaling constants.
307 				 */
308 				rt->rt_rmx.rmx_rtt =
309 				    (rt->rt_rmx.rmx_rtt + i) / 2;
310 			else
311 				rt->rt_rmx.rmx_rtt = i;
312 		}
313 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
314 			i = tp->t_rttvar *
315 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
316 			if (rt->rt_rmx.rmx_rttvar && i)
317 				rt->rt_rmx.rmx_rttvar =
318 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
319 			else
320 				rt->rt_rmx.rmx_rttvar = i;
321 		}
322 		/*
323 		 * update the pipelimit (ssthresh) if it has been updated
324 		 * already or if a pipesize was specified & the threshhold
325 		 * got below half the pipesize.  I.e., wait for bad news
326 		 * before we start updating, then update on both good
327 		 * and bad news.
328 		 */
329 		if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
330 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
331 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
332 			/*
333 			 * convert the limit from user data bytes to
334 			 * packets then to packet data bytes.
335 			 */
336 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
337 			if (i < 2)
338 				i = 2;
339 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
340 			if (rt->rt_rmx.rmx_ssthresh)
341 				rt->rt_rmx.rmx_ssthresh =
342 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
343 			else
344 				rt->rt_rmx.rmx_ssthresh = i;
345 		}
346 	}
347 #endif /* RTV_RTT */
348 	/* free the reassembly queue, if any */
349 	while ((qe = tp->segq.lh_first) != NULL) {
350 		LIST_REMOVE(qe, ipqe_q);
351 		m_freem(qe->ipqe_m);
352 		FREE(qe, M_IPQ);
353 	}
354 	if (tp->t_template)
355 		(void) m_free(dtom(tp->t_template));
356 	free(tp, M_PCB);
357 	inp->inp_ppcb = 0;
358 	soisdisconnected(so);
359 	in_pcbdetach(inp);
360 	tcpstat.tcps_closed++;
361 	return ((struct tcpcb *)0);
362 }
363 
364 void
365 tcp_drain()
366 {
367 
368 }
369 
370 /*
371  * Notify a tcp user of an asynchronous error;
372  * store error as soft error, but wake up user
373  * (for now, won't do anything until can select for soft error).
374  */
375 void
376 tcp_notify(inp, error)
377 	struct inpcb *inp;
378 	int error;
379 {
380 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
381 	register struct socket *so = inp->inp_socket;
382 
383 	/*
384 	 * Ignore some errors if we are hooked up.
385 	 * If connection hasn't completed, has retransmitted several times,
386 	 * and receives a second error, give up now.  This is better
387 	 * than waiting a long time to establish a connection that
388 	 * can never complete.
389 	 */
390 	if (tp->t_state == TCPS_ESTABLISHED &&
391 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
392 	      error == EHOSTDOWN)) {
393 		return;
394 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
395 	    tp->t_rxtshift > 3 && tp->t_softerror)
396 		so->so_error = error;
397 	else
398 		tp->t_softerror = error;
399 	wakeup((caddr_t) &so->so_timeo);
400 	sorwakeup(so);
401 	sowwakeup(so);
402 }
403 
404 void *
405 tcp_ctlinput(cmd, sa, v)
406 	int cmd;
407 	struct sockaddr *sa;
408 	register void *v;
409 {
410 	register struct ip *ip = v;
411 	register struct tcphdr *th;
412 	extern int inetctlerrmap[];
413 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
414 	int errno;
415 
416 	if ((unsigned)cmd >= PRC_NCMDS)
417 		return NULL;
418 	errno = inetctlerrmap[cmd];
419 	if (cmd == PRC_QUENCH)
420 		notify = tcp_quench;
421 	else if (PRC_IS_REDIRECT(cmd))
422 		notify = in_rtchange, ip = 0;
423 	else if (cmd == PRC_HOSTDEAD)
424 		ip = 0;
425 	else if (errno == 0)
426 		return NULL;
427 	if (ip) {
428 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
429 		in_pcbnotify(&tcbtable, satosin(sa)->sin_addr, th->th_dport,
430 		    ip->ip_src, th->th_sport, errno, notify);
431 	} else
432 		in_pcbnotifyall(&tcbtable, satosin(sa)->sin_addr, errno,
433 		    notify);
434 	return NULL;
435 }
436 
437 /*
438  * When a source quench is received, close congestion window
439  * to one segment.  We will gradually open it again as we proceed.
440  */
441 void
442 tcp_quench(inp, errno)
443 	struct inpcb *inp;
444 	int errno;
445 {
446 	struct tcpcb *tp = intotcpcb(inp);
447 
448 	if (tp)
449 		tp->snd_cwnd = tp->t_maxseg;
450 }
451