xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 9573504567626934c7ee01c7dce0c4bb1dfe7403)
1 /*	$NetBSD: tcp_subr.c,v 1.20 1995/11/21 01:07:41 cgd 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 extern	struct inpcb *tcp_last_inpcb;
70 
71 /*
72  * Tcp initialization
73  */
74 void
75 tcp_init()
76 {
77 
78 	tcp_iss = 1;		/* wrong */
79 	in_pcbinit(&tcbtable);
80 	if (max_protohdr < sizeof(struct tcpiphdr))
81 		max_protohdr = sizeof(struct tcpiphdr);
82 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
83 		panic("tcp_init");
84 }
85 
86 /*
87  * Create template to be used to send tcp packets on a connection.
88  * Call after host entry created, allocates an mbuf and fills
89  * in a skeletal tcp/ip header, minimizing the amount of work
90  * necessary when the connection is used.
91  */
92 struct tcpiphdr *
93 tcp_template(tp)
94 	struct tcpcb *tp;
95 {
96 	register struct inpcb *inp = tp->t_inpcb;
97 	register struct mbuf *m;
98 	register struct tcpiphdr *n;
99 
100 	if ((n = tp->t_template) == 0) {
101 		m = m_get(M_DONTWAIT, MT_HEADER);
102 		if (m == NULL)
103 			return (0);
104 		m->m_len = sizeof (struct tcpiphdr);
105 		n = mtod(m, struct tcpiphdr *);
106 	}
107 	bzero(n->ti_x1, sizeof n->ti_x1);
108 	n->ti_pr = IPPROTO_TCP;
109 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
110 	n->ti_src = inp->inp_laddr;
111 	n->ti_dst = inp->inp_faddr;
112 	n->ti_sport = inp->inp_lport;
113 	n->ti_dport = inp->inp_fport;
114 	n->ti_seq = 0;
115 	n->ti_ack = 0;
116 	n->ti_x2 = 0;
117 	n->ti_off = 5;
118 	n->ti_flags = 0;
119 	n->ti_win = 0;
120 	n->ti_sum = 0;
121 	n->ti_urp = 0;
122 	return (n);
123 }
124 
125 /*
126  * Send a single message to the TCP at address specified by
127  * the given TCP/IP header.  If m == 0, then we make a copy
128  * of the tcpiphdr at ti and send directly to the addressed host.
129  * This is used to force keep alive messages out using the TCP
130  * template for a connection tp->t_template.  If flags are given
131  * then we send a message back to the TCP which originated the
132  * segment ti, and discard the mbuf containing it and any other
133  * attached mbufs.
134  *
135  * In any case the ack and sequence number of the transmitted
136  * segment are as specified by the parameters.
137  */
138 void
139 tcp_respond(tp, ti, m, ack, seq, flags)
140 	struct tcpcb *tp;
141 	register struct tcpiphdr *ti;
142 	register struct mbuf *m;
143 	tcp_seq ack, seq;
144 	int flags;
145 {
146 	register int tlen;
147 	int win = 0;
148 	struct route *ro = 0;
149 
150 	if (tp) {
151 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
152 		ro = &tp->t_inpcb->inp_route;
153 	}
154 	if (m == 0) {
155 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
156 		if (m == NULL)
157 			return;
158 #ifdef TCP_COMPAT_42
159 		tlen = 1;
160 #else
161 		tlen = 0;
162 #endif
163 		m->m_data += max_linkhdr;
164 		*mtod(m, struct tcpiphdr *) = *ti;
165 		ti = mtod(m, struct tcpiphdr *);
166 		flags = TH_ACK;
167 	} else {
168 		m_freem(m->m_next);
169 		m->m_next = 0;
170 		m->m_data = (caddr_t)ti;
171 		m->m_len = sizeof (struct tcpiphdr);
172 		tlen = 0;
173 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
174 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_int32_t);
175 		xchg(ti->ti_dport, ti->ti_sport, u_int16_t);
176 #undef xchg
177 	}
178 	ti->ti_len = htons((u_int16_t)(sizeof (struct tcphdr) + tlen));
179 	tlen += sizeof (struct tcpiphdr);
180 	m->m_len = tlen;
181 	m->m_pkthdr.len = tlen;
182 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
183 	bzero(ti->ti_x1, sizeof ti->ti_x1);
184 	ti->ti_seq = htonl(seq);
185 	ti->ti_ack = htonl(ack);
186 	ti->ti_x2 = 0;
187 	ti->ti_off = sizeof (struct tcphdr) >> 2;
188 	ti->ti_flags = flags;
189 	if (tp)
190 		ti->ti_win = htons((u_int16_t) (win >> tp->rcv_scale));
191 	else
192 		ti->ti_win = htons((u_int16_t)win);
193 	ti->ti_urp = 0;
194 	ti->ti_sum = 0;
195 	ti->ti_sum = in_cksum(m, tlen);
196 	((struct ip *)ti)->ip_len = tlen;
197 	((struct ip *)ti)->ip_ttl = ip_defttl;
198 	(void) ip_output(m, NULL, ro, 0, NULL);
199 }
200 
201 /*
202  * Create a new TCP control block, making an
203  * empty reassembly queue and hooking it to the argument
204  * protocol control block.
205  */
206 struct tcpcb *
207 tcp_newtcpcb(inp)
208 	struct inpcb *inp;
209 {
210 	register struct tcpcb *tp;
211 
212 	tp = malloc(sizeof(*tp), M_PCB, M_NOWAIT);
213 	if (tp == NULL)
214 		return ((struct tcpcb *)0);
215 	bzero((char *) tp, sizeof(struct tcpcb));
216 	LIST_INIT(&tp->segq);
217 	tp->t_maxseg = tcp_mssdflt;
218 
219 	tp->t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0;
220 	tp->t_inpcb = inp;
221 	/*
222 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
223 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
224 	 * reasonable initial retransmit time.
225 	 */
226 	tp->t_srtt = TCPTV_SRTTBASE;
227 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
228 	tp->t_rttmin = TCPTV_MIN;
229 	TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
230 	    TCPTV_MIN, TCPTV_REXMTMAX);
231 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
232 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
233 	inp->inp_ip.ip_ttl = ip_defttl;
234 	inp->inp_ppcb = (caddr_t)tp;
235 	return (tp);
236 }
237 
238 /*
239  * Drop a TCP connection, reporting
240  * the specified error.  If connection is synchronized,
241  * then send a RST to peer.
242  */
243 struct tcpcb *
244 tcp_drop(tp, errno)
245 	register struct tcpcb *tp;
246 	int errno;
247 {
248 	struct socket *so = tp->t_inpcb->inp_socket;
249 
250 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
251 		tp->t_state = TCPS_CLOSED;
252 		(void) tcp_output(tp);
253 		tcpstat.tcps_drops++;
254 	} else
255 		tcpstat.tcps_conndrops++;
256 	if (errno == ETIMEDOUT && tp->t_softerror)
257 		errno = tp->t_softerror;
258 	so->so_error = errno;
259 	return (tcp_close(tp));
260 }
261 
262 /*
263  * Close a TCP control block:
264  *	discard all space held by the tcp
265  *	discard internet protocol block
266  *	wake up any sleepers
267  */
268 struct tcpcb *
269 tcp_close(tp)
270 	register struct tcpcb *tp;
271 {
272 	register struct ipqent *qe;
273 	struct inpcb *inp = tp->t_inpcb;
274 	struct socket *so = inp->inp_socket;
275 	register struct mbuf *m;
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 	    satosin(rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
294 		register u_long i;
295 
296 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
297 			i = tp->t_srtt *
298 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
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_SCALE));
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 		(void) m_free(dtom(tp->t_template));
354 	free(tp, M_PCB);
355 	inp->inp_ppcb = 0;
356 	soisdisconnected(so);
357 	/* clobber input pcb cache if we're closing the cached connection */
358 	if (inp == tcp_last_inpcb)
359 		tcp_last_inpcb = 0;
360 	in_pcbdetach(inp);
361 	tcpstat.tcps_closed++;
362 	return ((struct tcpcb *)0);
363 }
364 
365 void
366 tcp_drain()
367 {
368 
369 }
370 
371 /*
372  * Notify a tcp user of an asynchronous error;
373  * store error as soft error, but wake up user
374  * (for now, won't do anything until can select for soft error).
375  */
376 void
377 tcp_notify(inp, error)
378 	struct inpcb *inp;
379 	int error;
380 {
381 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
382 	register struct socket *so = inp->inp_socket;
383 
384 	/*
385 	 * Ignore some errors if we are hooked up.
386 	 * If connection hasn't completed, has retransmitted several times,
387 	 * and receives a second error, give up now.  This is better
388 	 * than waiting a long time to establish a connection that
389 	 * can never complete.
390 	 */
391 	if (tp->t_state == TCPS_ESTABLISHED &&
392 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
393 	      error == EHOSTDOWN)) {
394 		return;
395 	} else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
396 	    tp->t_rxtshift > 3 && tp->t_softerror)
397 		so->so_error = error;
398 	else
399 		tp->t_softerror = error;
400 	wakeup((caddr_t) &so->so_timeo);
401 	sorwakeup(so);
402 	sowwakeup(so);
403 }
404 
405 void
406 tcp_ctlinput(cmd, sa, ip)
407 	int cmd;
408 	struct sockaddr *sa;
409 	register struct ip *ip;
410 {
411 	register struct tcphdr *th;
412 	extern struct in_addr zeroin_addr;
413 	extern int inetctlerrmap[];
414 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
415 	int errno;
416 
417 	if ((unsigned)cmd >= PRC_NCMDS)
418 		return;
419 	errno = inetctlerrmap[cmd];
420 	if (cmd == PRC_QUENCH)
421 		notify = tcp_quench;
422 	else if (PRC_IS_REDIRECT(cmd))
423 		notify = in_rtchange, ip = 0;
424 	else if (cmd == PRC_HOSTDEAD)
425 		ip = 0;
426 	else if (errno == 0)
427 		return;
428 	if (ip) {
429 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
430 		in_pcbnotify(&tcbtable, sa, th->th_dport, ip->ip_src,
431 		    th->th_sport, errno, notify);
432 	} else
433 		in_pcbnotifyall(&tcbtable, sa, errno, notify);
434 }
435 
436 /*
437  * When a source quench is received, close congestion window
438  * to one segment.  We will gradually open it again as we proceed.
439  */
440 void
441 tcp_quench(inp, errno)
442 	struct inpcb *inp;
443 	int errno;
444 {
445 	struct tcpcb *tp = intotcpcb(inp);
446 
447 	if (tp)
448 		tp->snd_cwnd = tp->t_maxseg;
449 }
450