xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 35794)
1 /*
2  * Copyright (c) 1982, 1986, 1988 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are permitted
6  * provided that the above copyright notice and this paragraph are
7  * duplicated in all such forms and that any documentation,
8  * advertising materials, and other materials related to such
9  * distribution and use acknowledge that the software was developed
10  * by the University of California, Berkeley.  The name of the
11  * University may not be used to endorse or promote products derived
12  * from this software without specific prior written permission.
13  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
14  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
15  * WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE.
16  *
17  *	@(#)tcp_usrreq.c	7.11 (Berkeley) 10/12/88
18  */
19 
20 #include "param.h"
21 #include "systm.h"
22 #include "malloc.h"
23 #include "mbuf.h"
24 #include "socket.h"
25 #include "socketvar.h"
26 #include "protosw.h"
27 #include "errno.h"
28 #include "stat.h"
29 
30 #include "../net/if.h"
31 #include "../net/route.h"
32 
33 #include "in.h"
34 #include "in_pcb.h"
35 #include "in_systm.h"
36 #include "ip.h"
37 #include "ip_var.h"
38 #include "tcp.h"
39 #include "tcp_fsm.h"
40 #include "tcp_seq.h"
41 #include "tcp_timer.h"
42 #include "tcp_var.h"
43 #include "tcpip.h"
44 #include "tcp_debug.h"
45 
46 /*
47  * TCP protocol interface to socket abstraction.
48  */
49 extern	char *tcpstates[];
50 struct	tcpcb *tcp_newtcpcb();
51 
52 /*
53  * Process a TCP user request for TCP tb.  If this is a send request
54  * then m is the mbuf chain of send data.  If this is a timer expiration
55  * (called from the software clock routine), then timertype tells which timer.
56  */
57 /*ARGSUSED*/
58 tcp_usrreq(so, req, m, nam, rights)
59 	struct socket *so;
60 	int req;
61 	struct mbuf *m, *nam, *rights;
62 {
63 	register struct inpcb *inp;
64 	register struct tcpcb *tp;
65 	int s;
66 	int error = 0;
67 	int ostate;
68 
69 	if (req == PRU_CONTROL)
70 		return (in_control(so, (int)m, (caddr_t)nam,
71 			(struct ifnet *)rights));
72 	if (rights && rights->m_len)
73 		return (EINVAL);
74 
75 	s = splnet();
76 	inp = sotoinpcb(so);
77 	/*
78 	 * When a TCP is attached to a socket, then there will be
79 	 * a (struct inpcb) pointed at by the socket, and this
80 	 * structure will point at a subsidary (struct tcpcb).
81 	 */
82 	if (inp == 0 && req != PRU_ATTACH) {
83 		splx(s);
84 		return (EINVAL);		/* XXX */
85 	}
86 	if (inp) {
87 		tp = intotcpcb(inp);
88 		/* WHAT IF TP IS 0? */
89 #ifdef KPROF
90 		tcp_acounts[tp->t_state][req]++;
91 #endif
92 		ostate = tp->t_state;
93 	} else
94 		ostate = 0;
95 	switch (req) {
96 
97 	/*
98 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
99 	 * and an internet control block.
100 	 */
101 	case PRU_ATTACH:
102 		if (inp) {
103 			error = EISCONN;
104 			break;
105 		}
106 		error = tcp_attach(so);
107 		if (error)
108 			break;
109 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
110 			so->so_linger = TCP_LINGERTIME;
111 		tp = sototcpcb(so);
112 		break;
113 
114 	/*
115 	 * PRU_DETACH detaches the TCP protocol from the socket.
116 	 * If the protocol state is non-embryonic, then can't
117 	 * do this directly: have to initiate a PRU_DISCONNECT,
118 	 * which may finish later; embryonic TCB's can just
119 	 * be discarded here.
120 	 */
121 	case PRU_DETACH:
122 		if (tp->t_state > TCPS_LISTEN)
123 			tp = tcp_disconnect(tp);
124 		else
125 			tp = tcp_close(tp);
126 		break;
127 
128 	/*
129 	 * Give the socket an address.
130 	 */
131 	case PRU_BIND:
132 		error = in_pcbbind(inp, nam);
133 		if (error)
134 			break;
135 		break;
136 
137 	/*
138 	 * Prepare to accept connections.
139 	 */
140 	case PRU_LISTEN:
141 		if (inp->inp_lport == 0)
142 			error = in_pcbbind(inp, (struct mbuf *)0);
143 		if (error == 0)
144 			tp->t_state = TCPS_LISTEN;
145 		break;
146 
147 	/*
148 	 * Initiate connection to peer.
149 	 * Create a template for use in transmissions on this connection.
150 	 * Enter SYN_SENT state, and mark socket as connecting.
151 	 * Start keep-alive timer, and seed output sequence space.
152 	 * Send initial segment on connection.
153 	 */
154 	case PRU_CONNECT:
155 		if (inp->inp_lport == 0) {
156 			error = in_pcbbind(inp, (struct mbuf *)0);
157 			if (error)
158 				break;
159 		}
160 		error = in_pcbconnect(inp, nam);
161 		if (error)
162 			break;
163 		tp->t_template = tcp_template(tp);
164 		if (tp->t_template == 0) {
165 			in_pcbdisconnect(inp);
166 			error = ENOBUFS;
167 			break;
168 		}
169 		soisconnecting(so);
170 		tcpstat.tcps_connattempt++;
171 		tp->t_state = TCPS_SYN_SENT;
172 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
173 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
174 		tcp_sendseqinit(tp);
175 		error = tcp_output(tp);
176 		break;
177 
178 	/*
179 	 * Create a TCP connection between two sockets.
180 	 */
181 	case PRU_CONNECT2:
182 		error = EOPNOTSUPP;
183 		break;
184 
185 	/*
186 	 * Initiate disconnect from peer.
187 	 * If connection never passed embryonic stage, just drop;
188 	 * else if don't need to let data drain, then can just drop anyways,
189 	 * else have to begin TCP shutdown process: mark socket disconnecting,
190 	 * drain unread data, state switch to reflect user close, and
191 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
192 	 * when peer sends FIN and acks ours.
193 	 *
194 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
195 	 */
196 	case PRU_DISCONNECT:
197 		tp = tcp_disconnect(tp);
198 		break;
199 
200 	/*
201 	 * Accept a connection.  Essentially all the work is
202 	 * done at higher levels; just return the address
203 	 * of the peer, storing through addr.
204 	 */
205 	case PRU_ACCEPT: {
206 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
207 
208 		nam->m_len = sizeof (struct sockaddr_in);
209 		sin->sin_family = AF_INET;
210 		sin->sin_port = inp->inp_fport;
211 		sin->sin_addr = inp->inp_faddr;
212 		break;
213 		}
214 
215 	/*
216 	 * Mark the connection as being incapable of further output.
217 	 */
218 	case PRU_SHUTDOWN:
219 		socantsendmore(so);
220 		tp = tcp_usrclosed(tp);
221 		if (tp)
222 			error = tcp_output(tp);
223 		break;
224 
225 	/*
226 	 * After a receive, possibly send window update to peer.
227 	 */
228 	case PRU_RCVD:
229 		(void) tcp_output(tp);
230 		break;
231 
232 	/*
233 	 * Do a send by putting data in output queue and updating urgent
234 	 * marker if URG set.  Possibly send more data.
235 	 */
236 	case PRU_SEND:
237 		sbappend(&so->so_snd, m);
238 		error = tcp_output(tp);
239 		break;
240 
241 	/*
242 	 * Abort the TCP.
243 	 */
244 	case PRU_ABORT:
245 		tp = tcp_drop(tp, ECONNABORTED);
246 		break;
247 
248 	case PRU_SENSE:
249 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
250 		(void) splx(s);
251 		return (0);
252 
253 	case PRU_RCVOOB:
254 		if ((so->so_oobmark == 0 &&
255 		    (so->so_state & SS_RCVATMARK) == 0) ||
256 		    so->so_options & SO_OOBINLINE ||
257 		    tp->t_oobflags & TCPOOB_HADDATA) {
258 			error = EINVAL;
259 			break;
260 		}
261 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
262 			error = EWOULDBLOCK;
263 			break;
264 		}
265 		m->m_len = 1;
266 		*mtod(m, caddr_t) = tp->t_iobc;
267 		if (((int)nam & MSG_PEEK) == 0)
268 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
269 		break;
270 
271 	case PRU_SENDOOB:
272 		if (sbspace(&so->so_snd) < -512) {
273 			m_freem(m);
274 			error = ENOBUFS;
275 			break;
276 		}
277 		/*
278 		 * According to RFC961 (Assigned Protocols),
279 		 * the urgent pointer points to the last octet
280 		 * of urgent data.  We continue, however,
281 		 * to consider it to indicate the first octet
282 		 * of data past the urgent section.
283 		 * Otherwise, snd_up should be one lower.
284 		 */
285 		sbappend(&so->so_snd, m);
286 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
287 		tp->t_force = 1;
288 		error = tcp_output(tp);
289 		tp->t_force = 0;
290 		break;
291 
292 	case PRU_SOCKADDR:
293 		in_setsockaddr(inp, nam);
294 		break;
295 
296 	case PRU_PEERADDR:
297 		in_setpeeraddr(inp, nam);
298 		break;
299 
300 	/*
301 	 * TCP slow timer went off; going through this
302 	 * routine for tracing's sake.
303 	 */
304 	case PRU_SLOWTIMO:
305 		tp = tcp_timers(tp, (int)nam);
306 		req |= (int)nam << 8;		/* for debug's sake */
307 		break;
308 
309 	default:
310 		panic("tcp_usrreq");
311 	}
312 	if (tp && (so->so_options & SO_DEBUG))
313 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
314 	splx(s);
315 	return (error);
316 }
317 
318 tcp_ctloutput(op, so, level, optname, mp)
319 	int op;
320 	struct socket *so;
321 	int level, optname;
322 	struct mbuf **mp;
323 {
324 	int error = 0;
325 	struct inpcb *inp = sotoinpcb(so);
326 	register struct tcpcb *tp = intotcpcb(inp);
327 	register struct mbuf *m;
328 
329 	if (level != IPPROTO_TCP)
330 		return (ip_ctloutput(op, so, level, optname, mp));
331 
332 	switch (op) {
333 
334 	case PRCO_SETOPT:
335 		m = *mp;
336 		switch (optname) {
337 
338 		case TCP_NODELAY:
339 			if (m == NULL || m->m_len < sizeof (int))
340 				error = EINVAL;
341 			else if (*mtod(m, int *))
342 				tp->t_flags |= TF_NODELAY;
343 			else
344 				tp->t_flags &= ~TF_NODELAY;
345 			break;
346 
347 		case TCP_MAXSEG:	/* not yet */
348 		default:
349 			error = EINVAL;
350 			break;
351 		}
352 		if (m)
353 			(void) m_free(m);
354 		break;
355 
356 	case PRCO_GETOPT:
357 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
358 		m->m_len = sizeof(int);
359 
360 		switch (optname) {
361 		case TCP_NODELAY:
362 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
363 			break;
364 		case TCP_MAXSEG:
365 			*mtod(m, int *) = tp->t_maxseg;
366 			break;
367 		default:
368 			error = EINVAL;
369 			break;
370 		}
371 		break;
372 	}
373 	return (error);
374 }
375 
376 u_long	tcp_sendspace = 1024*4;
377 u_long	tcp_recvspace = 1024*4;
378 
379 /*
380  * Attach TCP protocol to socket, allocating
381  * internet protocol control block, tcp control block,
382  * bufer space, and entering LISTEN state if to accept connections.
383  */
384 tcp_attach(so)
385 	struct socket *so;
386 {
387 	register struct tcpcb *tp;
388 	struct inpcb *inp;
389 	int error;
390 
391 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
392 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
393 		if (error)
394 			return (error);
395 	}
396 	error = in_pcballoc(so, &tcb);
397 	if (error)
398 		return (error);
399 	inp = sotoinpcb(so);
400 	tp = tcp_newtcpcb(inp);
401 	if (tp == 0) {
402 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
403 
404 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
405 		in_pcbdetach(inp);
406 		so->so_state |= nofd;
407 		return (ENOBUFS);
408 	}
409 	tp->t_state = TCPS_CLOSED;
410 	return (0);
411 }
412 
413 /*
414  * Initiate (or continue) disconnect.
415  * If embryonic state, just send reset (once).
416  * If in ``let data drain'' option and linger null, just drop.
417  * Otherwise (hard), mark socket disconnecting and drop
418  * current input data; switch states based on user close, and
419  * send segment to peer (with FIN).
420  */
421 struct tcpcb *
422 tcp_disconnect(tp)
423 	register struct tcpcb *tp;
424 {
425 	struct socket *so = tp->t_inpcb->inp_socket;
426 
427 	if (tp->t_state < TCPS_ESTABLISHED)
428 		tp = tcp_close(tp);
429 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
430 		tp = tcp_drop(tp, 0);
431 	else {
432 		soisdisconnecting(so);
433 		sbflush(&so->so_rcv);
434 		tp = tcp_usrclosed(tp);
435 		if (tp)
436 			(void) tcp_output(tp);
437 	}
438 	return (tp);
439 }
440 
441 /*
442  * User issued close, and wish to trail through shutdown states:
443  * if never received SYN, just forget it.  If got a SYN from peer,
444  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
445  * If already got a FIN from peer, then almost done; go to LAST_ACK
446  * state.  In all other cases, have already sent FIN to peer (e.g.
447  * after PRU_SHUTDOWN), and just have to play tedious game waiting
448  * for peer to send FIN or not respond to keep-alives, etc.
449  * We can let the user exit from the close as soon as the FIN is acked.
450  */
451 struct tcpcb *
452 tcp_usrclosed(tp)
453 	register struct tcpcb *tp;
454 {
455 
456 	switch (tp->t_state) {
457 
458 	case TCPS_CLOSED:
459 	case TCPS_LISTEN:
460 	case TCPS_SYN_SENT:
461 		tp->t_state = TCPS_CLOSED;
462 		tp = tcp_close(tp);
463 		break;
464 
465 	case TCPS_SYN_RECEIVED:
466 	case TCPS_ESTABLISHED:
467 		tp->t_state = TCPS_FIN_WAIT_1;
468 		break;
469 
470 	case TCPS_CLOSE_WAIT:
471 		tp->t_state = TCPS_LAST_ACK;
472 		break;
473 	}
474 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
475 		soisdisconnected(tp->t_inpcb->inp_socket);
476 	return (tp);
477 }
478