xref: /freebsd/sys/netinet/tcp_subr.c (revision afe61c15161c324a7af299a9b8457aba5afc92db)
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  */
35 
36 #include <sys/param.h>
37 #include <sys/proc.h>
38 #include <sys/systm.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/socket.h>
42 #include <sys/socketvar.h>
43 #include <sys/protosw.h>
44 #include <sys/errno.h>
45 
46 #include <net/route.h>
47 #include <net/if.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/in_systm.h>
51 #include <netinet/ip.h>
52 #include <netinet/in_pcb.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/ip_icmp.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_fsm.h>
57 #include <netinet/tcp_seq.h>
58 #include <netinet/tcp_timer.h>
59 #include <netinet/tcp_var.h>
60 #include <netinet/tcpip.h>
61 
62 /* patchable/settable parameters for tcp */
63 int 	tcp_mssdflt = TCP_MSS;
64 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
65 int	tcp_do_rfc1323 = 1;
66 
67 extern	struct inpcb *tcp_last_inpcb;
68 
69 /*
70  * Tcp initialization
71  */
72 void
73 tcp_init()
74 {
75 
76 	tcp_iss = 1;		/* wrong */
77 	tcb.inp_next = tcb.inp_prev = &tcb;
78 	if (max_protohdr < sizeof(struct tcpiphdr))
79 		max_protohdr = sizeof(struct tcpiphdr);
80 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
81 		panic("tcp_init");
82 }
83 
84 /*
85  * Create template to be used to send tcp packets on a connection.
86  * Call after host entry created, allocates an mbuf and fills
87  * in a skeletal tcp/ip header, minimizing the amount of work
88  * necessary when the connection is used.
89  */
90 struct tcpiphdr *
91 tcp_template(tp)
92 	struct tcpcb *tp;
93 {
94 	register struct inpcb *inp = tp->t_inpcb;
95 	register struct mbuf *m;
96 	register struct tcpiphdr *n;
97 
98 	if ((n = tp->t_template) == 0) {
99 		m = m_get(M_DONTWAIT, MT_HEADER);
100 		if (m == NULL)
101 			return (0);
102 		m->m_len = sizeof (struct tcpiphdr);
103 		n = mtod(m, struct tcpiphdr *);
104 	}
105 	n->ti_next = n->ti_prev = 0;
106 	n->ti_x1 = 0;
107 	n->ti_pr = IPPROTO_TCP;
108 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
109 	n->ti_src = inp->inp_laddr;
110 	n->ti_dst = inp->inp_faddr;
111 	n->ti_sport = inp->inp_lport;
112 	n->ti_dport = inp->inp_fport;
113 	n->ti_seq = 0;
114 	n->ti_ack = 0;
115 	n->ti_x2 = 0;
116 	n->ti_off = 5;
117 	n->ti_flags = 0;
118 	n->ti_win = 0;
119 	n->ti_sum = 0;
120 	n->ti_urp = 0;
121 	return (n);
122 }
123 
124 /*
125  * Send a single message to the TCP at address specified by
126  * the given TCP/IP header.  If m == 0, then we make a copy
127  * of the tcpiphdr at ti and send directly to the addressed host.
128  * This is used to force keep alive messages out using the TCP
129  * template for a connection tp->t_template.  If flags are given
130  * then we send a message back to the TCP which originated the
131  * segment ti, and discard the mbuf containing it and any other
132  * attached mbufs.
133  *
134  * In any case the ack and sequence number of the transmitted
135  * segment are as specified by the parameters.
136  */
137 void
138 tcp_respond(tp, ti, m, ack, seq, flags)
139 	struct tcpcb *tp;
140 	register struct tcpiphdr *ti;
141 	register struct mbuf *m;
142 	tcp_seq ack, seq;
143 	int flags;
144 {
145 	register int tlen;
146 	int win = 0;
147 	struct route *ro = 0;
148 
149 	if (tp) {
150 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
151 		ro = &tp->t_inpcb->inp_route;
152 	}
153 	if (m == 0) {
154 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
155 		if (m == NULL)
156 			return;
157 #ifdef TCP_COMPAT_42
158 		tlen = 1;
159 #else
160 		tlen = 0;
161 #endif
162 		m->m_data += max_linkhdr;
163 		*mtod(m, struct tcpiphdr *) = *ti;
164 		ti = mtod(m, struct tcpiphdr *);
165 		flags = TH_ACK;
166 	} else {
167 		m_freem(m->m_next);
168 		m->m_next = 0;
169 		m->m_data = (caddr_t)ti;
170 		m->m_len = sizeof (struct tcpiphdr);
171 		tlen = 0;
172 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
173 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
174 		xchg(ti->ti_dport, ti->ti_sport, u_short);
175 #undef xchg
176 	}
177 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
178 	tlen += sizeof (struct tcpiphdr);
179 	m->m_len = tlen;
180 	m->m_pkthdr.len = tlen;
181 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
182 	ti->ti_next = ti->ti_prev = 0;
183 	ti->ti_x1 = 0;
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_short) (win >> tp->rcv_scale));
191 	else
192 		ti->ti_win = htons((u_short)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 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
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 << 2;
228 	tp->t_rttmin = TCPTV_MIN;
229 	TCPT_RANGESET(tp->t_rxtcur,
230 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
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 tcpiphdr *t;
274 	struct inpcb *inp = tp->t_inpcb;
275 	struct socket *so = inp->inp_socket;
276 	register struct mbuf *m;
277 #ifdef RTV_RTT
278 	register struct rtentry *rt;
279 
280 	/*
281 	 * If we sent enough data to get some meaningful characteristics,
282 	 * save them in the routing entry.  'Enough' is arbitrarily
283 	 * defined as the sendpipesize (default 4K) * 16.  This would
284 	 * give us 16 rtt samples assuming we only get one sample per
285 	 * window (the usual case on a long haul net).  16 samples is
286 	 * enough for the srtt filter to converge to within 5% of the correct
287 	 * value; fewer samples and we could save a very bogus rtt.
288 	 *
289 	 * Don't update the default route's characteristics and don't
290 	 * update anything that the user "locked".
291 	 */
292 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
293 	    (rt = inp->inp_route.ro_rt) &&
294 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
295 		register u_long i = 0;
296 
297 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
298 			i = tp->t_srtt *
299 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
300 			if (rt->rt_rmx.rmx_rtt && i)
301 				/*
302 				 * filter this update to half the old & half
303 				 * the new values, converting scale.
304 				 * See route.h and tcp_var.h for a
305 				 * description of the scaling constants.
306 				 */
307 				rt->rt_rmx.rmx_rtt =
308 				    (rt->rt_rmx.rmx_rtt + i) / 2;
309 			else
310 				rt->rt_rmx.rmx_rtt = i;
311 		}
312 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
313 			i = tp->t_rttvar *
314 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
315 			if (rt->rt_rmx.rmx_rttvar && i)
316 				rt->rt_rmx.rmx_rttvar =
317 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
318 			else
319 				rt->rt_rmx.rmx_rttvar = i;
320 		}
321 		/*
322 		 * update the pipelimit (ssthresh) if it has been updated
323 		 * already or if a pipesize was specified & the threshhold
324 		 * got below half the pipesize.  I.e., wait for bad news
325 		 * before we start updating, then update on both good
326 		 * and bad news.
327 		 */
328 		if ((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
329 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh ||
330 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
331 			/*
332 			 * convert the limit from user data bytes to
333 			 * packets then to packet data bytes.
334 			 */
335 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
336 			if (i < 2)
337 				i = 2;
338 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
339 			if (rt->rt_rmx.rmx_ssthresh)
340 				rt->rt_rmx.rmx_ssthresh =
341 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
342 			else
343 				rt->rt_rmx.rmx_ssthresh = i;
344 		}
345 	}
346 #endif /* RTV_RTT */
347 	/* free the reassembly queue, if any */
348 	t = tp->seg_next;
349 	while (t != (struct tcpiphdr *)tp) {
350 		t = (struct tcpiphdr *)t->ti_next;
351 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
352 		remque(t->ti_prev);
353 		m_freem(m);
354 	}
355 	if (tp->t_template)
356 		(void) m_free(dtom(tp->t_template));
357 	free(tp, M_PCB);
358 	inp->inp_ppcb = 0;
359 	soisdisconnected(so);
360 	/* clobber input pcb cache if we're closing the cached connection */
361 	if (inp == tcp_last_inpcb)
362 		tcp_last_inpcb = &tcb;
363 	in_pcbdetach(inp);
364 	tcpstat.tcps_closed++;
365 	return ((struct tcpcb *)0);
366 }
367 
368 void
369 tcp_drain()
370 {
371 
372 }
373 
374 /*
375  * Notify a tcp user of an asynchronous error;
376  * store error as soft error, but wake up user
377  * (for now, won't do anything until can select for soft error).
378  */
379 void
380 tcp_notify(inp, error)
381 	struct inpcb *inp;
382 	int error;
383 {
384 	register struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
385 	register struct socket *so = inp->inp_socket;
386 
387 	/*
388 	 * Ignore some errors if we are hooked up.
389 	 * If connection hasn't completed, has retransmitted several times,
390 	 * and receives a second error, give up now.  This is better
391 	 * than waiting a long time to establish a connection that
392 	 * can never complete.
393 	 */
394 	if (tp->t_state == TCPS_ESTABLISHED &&
395 	     (error == EHOSTUNREACH || error == ENETUNREACH ||
396 	      error == EHOSTDOWN)) {
397 		return;
398 	} else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
399 	    tp->t_softerror)
400 		so->so_error = error;
401 	else
402 		tp->t_softerror = error;
403 	wakeup((caddr_t) &so->so_timeo);
404 	sorwakeup(so);
405 	sowwakeup(so);
406 }
407 
408 void
409 tcp_ctlinput(cmd, sa, ip)
410 	int cmd;
411 	struct sockaddr *sa;
412 	register struct ip *ip;
413 {
414 	register struct tcphdr *th;
415 	extern struct in_addr zeroin_addr;
416 	extern u_char inetctlerrmap[];
417 	void (*notify) __P((struct inpcb *, int)) = tcp_notify;
418 
419 	if (cmd == PRC_QUENCH)
420 		notify = tcp_quench;
421 	else if (!PRC_IS_REDIRECT(cmd) &&
422 		 ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0))
423 		return;
424 	if (ip) {
425 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
426 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
427 			cmd, notify);
428 	} else
429 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
430 }
431 
432 /*
433  * When a source quench is received, close congestion window
434  * to one segment.  We will gradually open it again as we proceed.
435  */
436 void
437 tcp_quench(inp, errno)
438 	struct inpcb *inp;
439 	int errno;
440 {
441 	struct tcpcb *tp = intotcpcb(inp);
442 
443 	if (tp)
444 		tp->snd_cwnd = tp->t_maxseg;
445 }
446