xref: /freebsd/sys/netinet/tcp_subr.c (revision a316b26e50bbed7cf655fbba726ab87d8ab7599d)
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 1.4 1994/10/02 17:48:44 phk 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 #include <netinet/tcp_fsm.h>
58 #include <netinet/tcp_seq.h>
59 #include <netinet/tcp_timer.h>
60 #include <netinet/tcp_var.h>
61 #include <netinet/tcpip.h>
62 
63 /* patchable/settable parameters for tcp */
64 int 	tcp_mssdflt = TCP_MSS;
65 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
66 int	tcp_do_rfc1323 = 1;
67 
68 extern	struct inpcb *tcp_last_inpcb;
69 
70 /*
71  * Tcp initialization
72  */
73 void
74 tcp_init()
75 {
76 
77 	tcp_iss = 1;		/* wrong */
78 	tcb.inp_next = tcb.inp_prev = &tcb;
79 	if (max_protohdr < sizeof(struct tcpiphdr))
80 		max_protohdr = sizeof(struct tcpiphdr);
81 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
82 		panic("tcp_init");
83 }
84 
85 /*
86  * Create template to be used to send tcp packets on a connection.
87  * Call after host entry created, allocates an mbuf and fills
88  * in a skeletal tcp/ip header, minimizing the amount of work
89  * necessary when the connection is used.
90  */
91 struct tcpiphdr *
92 tcp_template(tp)
93 	struct tcpcb *tp;
94 {
95 	register struct inpcb *inp = tp->t_inpcb;
96 	register struct mbuf *m;
97 	register struct tcpiphdr *n;
98 
99 	if ((n = tp->t_template) == 0) {
100 		m = m_get(M_DONTWAIT, MT_HEADER);
101 		if (m == NULL)
102 			return (0);
103 		m->m_len = sizeof (struct tcpiphdr);
104 		n = mtod(m, struct tcpiphdr *);
105 	}
106 	n->ti_next = n->ti_prev = 0;
107 	n->ti_x1 = 0;
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_long);
175 		xchg(ti->ti_dport, ti->ti_sport, u_short);
176 #undef xchg
177 	}
178 	ti->ti_len = htons((u_short)(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 	ti->ti_next = ti->ti_prev = 0;
184 	ti->ti_x1 = 0;
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_short) (win >> tp->rcv_scale));
192 	else
193 		ti->ti_win = htons((u_short)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((char *) tp, sizeof(struct tcpcb));
217 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
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 << 2;
229 	tp->t_rttmin = TCPTV_MIN;
230 	TCPT_RANGESET(tp->t_rxtcur,
231 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
232 	    TCPTV_MIN, TCPTV_REXMTMAX);
233 	tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
234 	tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
235 	inp->inp_ip.ip_ttl = ip_defttl;
236 	inp->inp_ppcb = (caddr_t)tp;
237 	return (tp);
238 }
239 
240 /*
241  * Drop a TCP connection, reporting
242  * the specified error.  If connection is synchronized,
243  * then send a RST to peer.
244  */
245 struct tcpcb *
246 tcp_drop(tp, errno)
247 	register struct tcpcb *tp;
248 	int errno;
249 {
250 	struct socket *so = tp->t_inpcb->inp_socket;
251 
252 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
253 		tp->t_state = TCPS_CLOSED;
254 		(void) tcp_output(tp);
255 		tcpstat.tcps_drops++;
256 	} else
257 		tcpstat.tcps_conndrops++;
258 	if (errno == ETIMEDOUT && tp->t_softerror)
259 		errno = tp->t_softerror;
260 	so->so_error = errno;
261 	return (tcp_close(tp));
262 }
263 
264 /*
265  * Close a TCP control block:
266  *	discard all space held by the tcp
267  *	discard internet protocol block
268  *	wake up any sleepers
269  */
270 struct tcpcb *
271 tcp_close(tp)
272 	register struct tcpcb *tp;
273 {
274 	register struct tcpiphdr *t;
275 	struct inpcb *inp = tp->t_inpcb;
276 	struct socket *so = inp->inp_socket;
277 	register struct mbuf *m;
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 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
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) != 0) && 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 	t = tp->seg_next;
350 	while (t != (struct tcpiphdr *)tp) {
351 		t = (struct tcpiphdr *)t->ti_next;
352 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
353 		remque(t->ti_prev);
354 		m_freem(m);
355 	}
356 	if (tp->t_template)
357 		(void) m_free(dtom(tp->t_template));
358 	free(tp, M_PCB);
359 	inp->inp_ppcb = 0;
360 	soisdisconnected(so);
361 	/* clobber input pcb cache if we're closing the cached connection */
362 	if (inp == tcp_last_inpcb)
363 		tcp_last_inpcb = &tcb;
364 	in_pcbdetach(inp);
365 	tcpstat.tcps_closed++;
366 	return ((struct tcpcb *)0);
367 }
368 
369 void
370 tcp_drain()
371 {
372 
373 }
374 
375 /*
376  * Notify a tcp user of an asynchronous error;
377  * store error as soft error, but wake up user
378  * (for now, won't do anything until can select for soft error).
379  */
380 void
381 tcp_notify(inp, error)
382 	struct inpcb *inp;
383 	int error;
384 {
385 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
386 	register struct socket *so = inp->inp_socket;
387 
388 	/*
389 	 * Ignore some errors if we are hooked up.
390 	 * If connection hasn't completed, has retransmitted several times,
391 	 * and receives a second error, give up now.  This is better
392 	 * than waiting a long time to establish a connection that
393 	 * can never complete.
394 	 */
395 	if (tp->t_state == TCPS_ESTABLISHED &&
396 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
397 	      error == EHOSTDOWN)) {
398 		return;
399 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
400 	    tp->t_softerror)
401 		so->so_error = error;
402 	else
403 		tp->t_softerror = error;
404 	wakeup((caddr_t) &so->so_timeo);
405 	sorwakeup(so);
406 	sowwakeup(so);
407 }
408 
409 void
410 tcp_ctlinput(cmd, sa, ip)
411 	int cmd;
412 	struct sockaddr *sa;
413 	register struct ip *ip;
414 {
415 	register struct tcphdr *th;
416 	extern struct in_addr zeroin_addr;
417 	extern u_char inetctlerrmap[];
418 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
419 
420 	if (cmd == PRC_QUENCH)
421 		notify = tcp_quench;
422 	else if (!PRC_IS_REDIRECT(cmd) &&
423 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
424 		return;
425 	if (ip) {
426 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
427 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
428 			cmd, notify);
429 	} else
430 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
431 }
432 
433 /*
434  * When a source quench is received, close congestion window
435  * to one segment.  We will gradually open it again as we proceed.
436  */
437 void
438 tcp_quench(inp, errno)
439 	struct inpcb *inp;
440 	int errno;
441 {
442 	struct tcpcb *tp = intotcpcb(inp);
443 
444 	if (tp)
445 		tp->snd_cwnd = tp->t_maxseg;
446 }
447