xref: /netbsd-src/sys/netinet/tcp_subr.c (revision 61f282557f0bc41c0b762c629a2f4c14be8b7591)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990 Regents of the University of California.
3  * 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	7.20 (Berkeley) 12/1/90
34  */
35 
36 #include "param.h"
37 #include "systm.h"
38 #include "malloc.h"
39 #include "mbuf.h"
40 #include "socket.h"
41 #include "socketvar.h"
42 #include "protosw.h"
43 #include "errno.h"
44 
45 #include "../net/route.h"
46 #include "../net/if.h"
47 
48 #include "in.h"
49 #include "in_systm.h"
50 #include "ip.h"
51 #include "in_pcb.h"
52 #include "ip_var.h"
53 #include "ip_icmp.h"
54 #include "tcp.h"
55 #include "tcp_fsm.h"
56 #include "tcp_seq.h"
57 #include "tcp_timer.h"
58 #include "tcp_var.h"
59 #include "tcpip.h"
60 
61 /* patchable/settable parameters for tcp */
62 int	tcp_ttl = TCP_TTL;
63 int 	tcp_mssdflt = TCP_MSS;
64 int 	tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
65 
66 extern	struct inpcb *tcp_last_inpcb;
67 
68 /*
69  * Tcp initialization
70  */
71 tcp_init()
72 {
73 
74 	tcp_iss = 1;		/* wrong */
75 	tcb.inp_next = tcb.inp_prev = &tcb;
76 	if (max_protohdr < sizeof(struct tcpiphdr))
77 		max_protohdr = sizeof(struct tcpiphdr);
78 	if (max_linkhdr + sizeof(struct tcpiphdr) > MHLEN)
79 		panic("tcp_init");
80 }
81 
82 /*
83  * Create template to be used to send tcp packets on a connection.
84  * Call after host entry created, allocates an mbuf and fills
85  * in a skeletal tcp/ip header, minimizing the amount of work
86  * necessary when the connection is used.
87  */
88 struct tcpiphdr *
89 tcp_template(tp)
90 	struct tcpcb *tp;
91 {
92 	register struct inpcb *inp = tp->t_inpcb;
93 	register struct mbuf *m;
94 	register struct tcpiphdr *n;
95 
96 	if ((n = tp->t_template) == 0) {
97 		m = m_get(M_DONTWAIT, MT_HEADER);
98 		if (m == NULL)
99 			return (0);
100 		m->m_len = sizeof (struct tcpiphdr);
101 		n = mtod(m, struct tcpiphdr *);
102 	}
103 	n->ti_next = n->ti_prev = 0;
104 	n->ti_x1 = 0;
105 	n->ti_pr = IPPROTO_TCP;
106 	n->ti_len = htons(sizeof (struct tcpiphdr) - sizeof (struct ip));
107 	n->ti_src = inp->inp_laddr;
108 	n->ti_dst = inp->inp_faddr;
109 	n->ti_sport = inp->inp_lport;
110 	n->ti_dport = inp->inp_fport;
111 	n->ti_seq = 0;
112 	n->ti_ack = 0;
113 	n->ti_x2 = 0;
114 	n->ti_off = 5;
115 	n->ti_flags = 0;
116 	n->ti_win = 0;
117 	n->ti_sum = 0;
118 	n->ti_urp = 0;
119 	return (n);
120 }
121 
122 /*
123  * Send a single message to the TCP at address specified by
124  * the given TCP/IP header.  If m == 0, then we make a copy
125  * of the tcpiphdr at ti and send directly to the addressed host.
126  * This is used to force keep alive messages out using the TCP
127  * template for a connection tp->t_template.  If flags are given
128  * then we send a message back to the TCP which originated the
129  * segment ti, and discard the mbuf containing it and any other
130  * attached mbufs.
131  *
132  * In any case the ack and sequence number of the transmitted
133  * segment are as specified by the parameters.
134  */
135 tcp_respond(tp, ti, m, ack, seq, flags)
136 	struct tcpcb *tp;
137 	register struct tcpiphdr *ti;
138 	register struct mbuf *m;
139 	tcp_seq ack, seq;
140 	int flags;
141 {
142 	register int tlen;
143 	int win = 0;
144 	struct route *ro = 0;
145 
146 	if (tp) {
147 		win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
148 		ro = &tp->t_inpcb->inp_route;
149 	}
150 	if (m == 0) {
151 		m = m_gethdr(M_DONTWAIT, MT_HEADER);
152 		if (m == NULL)
153 			return;
154 #ifdef TCP_COMPAT_42
155 		tlen = 1;
156 #else
157 		tlen = 0;
158 #endif
159 		m->m_data += max_linkhdr;
160 		*mtod(m, struct tcpiphdr *) = *ti;
161 		ti = mtod(m, struct tcpiphdr *);
162 		flags = TH_ACK;
163 	} else {
164 		m_freem(m->m_next);
165 		m->m_next = 0;
166 		m->m_data = (caddr_t)ti;
167 		m->m_len = sizeof (struct tcpiphdr);
168 		tlen = 0;
169 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
170 		xchg(ti->ti_dst.s_addr, ti->ti_src.s_addr, u_long);
171 		xchg(ti->ti_dport, ti->ti_sport, u_short);
172 #undef xchg
173 	}
174 	ti->ti_len = htons((u_short)(sizeof (struct tcphdr) + tlen));
175 	tlen += sizeof (struct tcpiphdr);
176 	m->m_len = tlen;
177 	m->m_pkthdr.len = tlen;
178 	m->m_pkthdr.rcvif = (struct ifnet *) 0;
179 	ti->ti_next = ti->ti_prev = 0;
180 	ti->ti_x1 = 0;
181 	ti->ti_seq = htonl(seq);
182 	ti->ti_ack = htonl(ack);
183 	ti->ti_x2 = 0;
184 	ti->ti_off = sizeof (struct tcphdr) >> 2;
185 	ti->ti_flags = flags;
186 	ti->ti_win = htons((u_short)win);
187 	ti->ti_urp = 0;
188 	ti->ti_sum = in_cksum(m, tlen);
189 	((struct ip *)ti)->ip_len = tlen;
190 	((struct ip *)ti)->ip_ttl = tcp_ttl;
191 	(void) ip_output(m, (struct mbuf *)0, ro, 0);
192 }
193 
194 /*
195  * Create a new TCP control block, making an
196  * empty reassembly queue and hooking it to the argument
197  * protocol control block.
198  */
199 struct tcpcb *
200 tcp_newtcpcb(inp)
201 	struct inpcb *inp;
202 {
203 	struct mbuf *m = m_getclr(M_DONTWAIT, MT_PCB);
204 	register struct tcpcb *tp;
205 
206 	if (m == NULL)
207 		return ((struct tcpcb *)0);
208 	tp = mtod(m, struct tcpcb *);
209 	tp->seg_next = tp->seg_prev = (struct tcpiphdr *)tp;
210 	tp->t_maxseg = tcp_mssdflt;
211 
212 	tp->t_flags = 0;		/* sends options! */
213 	tp->t_inpcb = inp;
214 	/*
215 	 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
216 	 * rtt estimate.  Set rttvar so that srtt + 2 * rttvar gives
217 	 * reasonable initial retransmit time.
218 	 */
219 	tp->t_srtt = TCPTV_SRTTBASE;
220 	tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << 2;
221 	tp->t_rttmin = TCPTV_MIN;
222 	TCPT_RANGESET(tp->t_rxtcur,
223 	    ((TCPTV_SRTTBASE >> 2) + (TCPTV_SRTTDFLT << 2)) >> 1,
224 	    TCPTV_MIN, TCPTV_REXMTMAX);
225 	tp->snd_cwnd = TCP_MAXWIN;
226 	tp->snd_ssthresh = TCP_MAXWIN;
227 	inp->inp_ip.ip_ttl = tcp_ttl;
228 	inp->inp_ppcb = (caddr_t)tp;
229 	return (tp);
230 }
231 
232 /*
233  * Drop a TCP connection, reporting
234  * the specified error.  If connection is synchronized,
235  * then send a RST to peer.
236  */
237 struct tcpcb *
238 tcp_drop(tp, errno)
239 	register struct tcpcb *tp;
240 	int errno;
241 {
242 	struct socket *so = tp->t_inpcb->inp_socket;
243 
244 	if (TCPS_HAVERCVDSYN(tp->t_state)) {
245 		tp->t_state = TCPS_CLOSED;
246 		(void) tcp_output(tp);
247 		tcpstat.tcps_drops++;
248 	} else
249 		tcpstat.tcps_conndrops++;
250 	if (errno == ETIMEDOUT && tp->t_softerror)
251 		errno = tp->t_softerror;
252 	so->so_error = errno;
253 	return (tcp_close(tp));
254 }
255 
256 /*
257  * Close a TCP control block:
258  *	discard all space held by the tcp
259  *	discard internet protocol block
260  *	wake up any sleepers
261  */
262 struct tcpcb *
263 tcp_close(tp)
264 	register struct tcpcb *tp;
265 {
266 	register struct tcpiphdr *t;
267 	struct inpcb *inp = tp->t_inpcb;
268 	struct socket *so = inp->inp_socket;
269 	register struct mbuf *m;
270 #ifdef RTV_RTT
271 	register struct rtentry *rt;
272 
273 	/*
274 	 * If we sent enough data to get some meaningful characteristics,
275 	 * save them in the routing entry.  'Enough' is arbitrarily
276 	 * defined as the sendpipesize (default 4K) * 16.  This would
277 	 * give us 16 rtt samples assuming we only get one sample per
278 	 * window (the usual case on a long haul net).  16 samples is
279 	 * enough for the srtt filter to converge to within 5% of the correct
280 	 * value; fewer samples and we could save a very bogus rtt.
281 	 *
282 	 * Don't update the default route's characteristics and don't
283 	 * update anything that the user "locked".
284 	 */
285 	if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
286 	    (rt = inp->inp_route.ro_rt) &&
287 	    ((struct sockaddr_in *)rt_key(rt))->sin_addr.s_addr != INADDR_ANY) {
288 		register u_long i;
289 
290 		if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
291 			i = tp->t_srtt *
292 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
293 			if (rt->rt_rmx.rmx_rtt && i)
294 				/*
295 				 * filter this update to half the old & half
296 				 * the new values, converting scale.
297 				 * See route.h and tcp_var.h for a
298 				 * description of the scaling constants.
299 				 */
300 				rt->rt_rmx.rmx_rtt =
301 				    (rt->rt_rmx.rmx_rtt + i) / 2;
302 			else
303 				rt->rt_rmx.rmx_rtt = i;
304 		}
305 		if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
306 			i = tp->t_rttvar *
307 			    (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
308 			if (rt->rt_rmx.rmx_rttvar && i)
309 				rt->rt_rmx.rmx_rttvar =
310 				    (rt->rt_rmx.rmx_rttvar + i) / 2;
311 			else
312 				rt->rt_rmx.rmx_rttvar = i;
313 		}
314 		/*
315 		 * update the pipelimit (ssthresh) if it has been updated
316 		 * already or if a pipesize was specified & the threshhold
317 		 * got below half the pipesize.  I.e., wait for bad news
318 		 * before we start updating, then update on both good
319 		 * and bad news.
320 		 */
321 		if ((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
322 		    (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh ||
323 		    i < (rt->rt_rmx.rmx_sendpipe / 2)) {
324 			/*
325 			 * convert the limit from user data bytes to
326 			 * packets then to packet data bytes.
327 			 */
328 			i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
329 			if (i < 2)
330 				i = 2;
331 			i *= (u_long)(tp->t_maxseg + sizeof (struct tcpiphdr));
332 			if (rt->rt_rmx.rmx_ssthresh)
333 				rt->rt_rmx.rmx_ssthresh =
334 				    (rt->rt_rmx.rmx_ssthresh + i) / 2;
335 			else
336 				rt->rt_rmx.rmx_ssthresh = i;
337 		}
338 	}
339 #endif RTV_RTT
340 	/* free the reassembly queue, if any */
341 	t = tp->seg_next;
342 	while (t != (struct tcpiphdr *)tp) {
343 		t = (struct tcpiphdr *)t->ti_next;
344 		m = REASS_MBUF((struct tcpiphdr *)t->ti_prev);
345 		remque(t->ti_prev);
346 		m_freem(m);
347 	}
348 	if (tp->t_template)
349 		(void) m_free(dtom(tp->t_template));
350 	(void) m_free(dtom(tp));
351 	inp->inp_ppcb = 0;
352 	soisdisconnected(so);
353 	/* clobber input pcb cache if we're closing the cached connection */
354 	if (inp == tcp_last_inpcb)
355 		tcp_last_inpcb = &tcb;
356 	in_pcbdetach(inp);
357 	tcpstat.tcps_closed++;
358 	return ((struct tcpcb *)0);
359 }
360 
361 tcp_drain()
362 {
363 
364 }
365 
366 /*
367  * Notify a tcp user of an asynchronous error;
368  * store error as soft error, but wake up user
369  * (for now, won't do anything until can select for soft error).
370  */
371 tcp_notify(inp, error)
372 	register struct inpcb *inp;
373 	int error;
374 {
375 
376 	((struct tcpcb *)inp->inp_ppcb)->t_softerror = error;
377 	wakeup((caddr_t) &inp->inp_socket->so_timeo);
378 	sorwakeup(inp->inp_socket);
379 	sowwakeup(inp->inp_socket);
380 }
381 
382 tcp_ctlinput(cmd, sa, ip)
383 	int cmd;
384 	struct sockaddr *sa;
385 	register struct ip *ip;
386 {
387 	register struct tcphdr *th;
388 	extern struct in_addr zeroin_addr;
389 	extern u_char inetctlerrmap[];
390 	int (*notify)() = tcp_notify, tcp_quench();
391 
392 	if (cmd == PRC_QUENCH)
393 		notify = tcp_quench;
394 	else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
395 		return;
396 	if (ip) {
397 		th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
398 		in_pcbnotify(&tcb, sa, th->th_dport, ip->ip_src, th->th_sport,
399 			cmd, notify);
400 	} else
401 		in_pcbnotify(&tcb, sa, 0, zeroin_addr, 0, cmd, notify);
402 }
403 
404 /*
405  * When a source quench is received, close congestion window
406  * to one segment.  We will gradually open it again as we proceed.
407  */
408 tcp_quench(inp)
409 	struct inpcb *inp;
410 {
411 	struct tcpcb *tp = intotcpcb(inp);
412 
413 	if (tp)
414 		tp->snd_cwnd = tp->t_maxseg;
415 }
416