xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 34855)
1 /*
2  * Copyright (c) 1982, 1986 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	7.10 (Berkeley) 06/29/88
18  */
19 
20 #include "param.h"
21 #include "systm.h"
22 #include "mbuf.h"
23 #include "socket.h"
24 #include "socketvar.h"
25 #include "protosw.h"
26 #include "errno.h"
27 #include "stat.h"
28 
29 #include "../net/if.h"
30 #include "../net/route.h"
31 
32 #include "in.h"
33 #include "in_pcb.h"
34 #include "in_systm.h"
35 #include "ip.h"
36 #include "ip_var.h"
37 #include "tcp.h"
38 #include "tcp_fsm.h"
39 #include "tcp_seq.h"
40 #include "tcp_timer.h"
41 #include "tcp_var.h"
42 #include "tcpip.h"
43 #include "tcp_debug.h"
44 
45 /*
46  * TCP protocol interface to socket abstraction.
47  */
48 extern	char *tcpstates[];
49 struct	tcpcb *tcp_newtcpcb();
50 
51 /*
52  * Process a TCP user request for TCP tb.  If this is a send request
53  * then m is the mbuf chain of send data.  If this is a timer expiration
54  * (called from the software clock routine), then timertype tells which timer.
55  */
56 /*ARGSUSED*/
57 tcp_usrreq(so, req, m, nam, rights)
58 	struct socket *so;
59 	int req;
60 	struct mbuf *m, *nam, *rights;
61 {
62 	register struct inpcb *inp;
63 	register struct tcpcb *tp;
64 	int s;
65 	int error = 0;
66 	int ostate;
67 
68 	if (req == PRU_CONTROL)
69 		return (in_control(so, (int)m, (caddr_t)nam,
70 			(struct ifnet *)rights));
71 	if (rights && rights->m_len)
72 		return (EINVAL);
73 
74 	s = splnet();
75 	inp = sotoinpcb(so);
76 	/*
77 	 * When a TCP is attached to a socket, then there will be
78 	 * a (struct inpcb) pointed at by the socket, and this
79 	 * structure will point at a subsidary (struct tcpcb).
80 	 */
81 	if (inp == 0 && req != PRU_ATTACH) {
82 		splx(s);
83 		return (EINVAL);		/* XXX */
84 	}
85 	if (inp) {
86 		tp = intotcpcb(inp);
87 		/* WHAT IF TP IS 0? */
88 #ifdef KPROF
89 		tcp_acounts[tp->t_state][req]++;
90 #endif
91 		ostate = tp->t_state;
92 	} else
93 		ostate = 0;
94 	switch (req) {
95 
96 	/*
97 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
98 	 * and an internet control block.
99 	 */
100 	case PRU_ATTACH:
101 		if (inp) {
102 			error = EISCONN;
103 			break;
104 		}
105 		error = tcp_attach(so);
106 		if (error)
107 			break;
108 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
109 			so->so_linger = TCP_LINGERTIME;
110 		tp = sototcpcb(so);
111 		break;
112 
113 	/*
114 	 * PRU_DETACH detaches the TCP protocol from the socket.
115 	 * If the protocol state is non-embryonic, then can't
116 	 * do this directly: have to initiate a PRU_DISCONNECT,
117 	 * which may finish later; embryonic TCB's can just
118 	 * be discarded here.
119 	 */
120 	case PRU_DETACH:
121 		if (tp->t_state > TCPS_LISTEN)
122 			tp = tcp_disconnect(tp);
123 		else
124 			tp = tcp_close(tp);
125 		break;
126 
127 	/*
128 	 * Give the socket an address.
129 	 */
130 	case PRU_BIND:
131 		error = in_pcbbind(inp, nam);
132 		if (error)
133 			break;
134 		break;
135 
136 	/*
137 	 * Prepare to accept connections.
138 	 */
139 	case PRU_LISTEN:
140 		if (inp->inp_lport == 0)
141 			error = in_pcbbind(inp, (struct mbuf *)0);
142 		if (error == 0)
143 			tp->t_state = TCPS_LISTEN;
144 		break;
145 
146 	/*
147 	 * Initiate connection to peer.
148 	 * Create a template for use in transmissions on this connection.
149 	 * Enter SYN_SENT state, and mark socket as connecting.
150 	 * Start keep-alive timer, and seed output sequence space.
151 	 * Send initial segment on connection.
152 	 */
153 	case PRU_CONNECT:
154 		if (inp->inp_lport == 0) {
155 			error = in_pcbbind(inp, (struct mbuf *)0);
156 			if (error)
157 				break;
158 		}
159 		error = in_pcbconnect(inp, nam);
160 		if (error)
161 			break;
162 		tp->t_template = tcp_template(tp);
163 		if (tp->t_template == 0) {
164 			in_pcbdisconnect(inp);
165 			error = ENOBUFS;
166 			break;
167 		}
168 		soisconnecting(so);
169 		tcpstat.tcps_connattempt++;
170 		tp->t_state = TCPS_SYN_SENT;
171 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
172 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
173 		tcp_sendseqinit(tp);
174 		error = tcp_output(tp);
175 		break;
176 
177 	/*
178 	 * Create a TCP connection between two sockets.
179 	 */
180 	case PRU_CONNECT2:
181 		error = EOPNOTSUPP;
182 		break;
183 
184 	/*
185 	 * Initiate disconnect from peer.
186 	 * If connection never passed embryonic stage, just drop;
187 	 * else if don't need to let data drain, then can just drop anyways,
188 	 * else have to begin TCP shutdown process: mark socket disconnecting,
189 	 * drain unread data, state switch to reflect user close, and
190 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
191 	 * when peer sends FIN and acks ours.
192 	 *
193 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
194 	 */
195 	case PRU_DISCONNECT:
196 		tp = tcp_disconnect(tp);
197 		break;
198 
199 	/*
200 	 * Accept a connection.  Essentially all the work is
201 	 * done at higher levels; just return the address
202 	 * of the peer, storing through addr.
203 	 */
204 	case PRU_ACCEPT: {
205 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
206 
207 		nam->m_len = sizeof (struct sockaddr_in);
208 		sin->sin_family = AF_INET;
209 		sin->sin_port = inp->inp_fport;
210 		sin->sin_addr = inp->inp_faddr;
211 		break;
212 		}
213 
214 	/*
215 	 * Mark the connection as being incapable of further output.
216 	 */
217 	case PRU_SHUTDOWN:
218 		socantsendmore(so);
219 		tp = tcp_usrclosed(tp);
220 		if (tp)
221 			error = tcp_output(tp);
222 		break;
223 
224 	/*
225 	 * After a receive, possibly send window update to peer.
226 	 */
227 	case PRU_RCVD:
228 		(void) tcp_output(tp);
229 		break;
230 
231 	/*
232 	 * Do a send by putting data in output queue and updating urgent
233 	 * marker if URG set.  Possibly send more data.
234 	 */
235 	case PRU_SEND:
236 		sbappend(&so->so_snd, m);
237 		error = tcp_output(tp);
238 		break;
239 
240 	/*
241 	 * Abort the TCP.
242 	 */
243 	case PRU_ABORT:
244 		tp = tcp_drop(tp, ECONNABORTED);
245 		break;
246 
247 	case PRU_SENSE:
248 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
249 		(void) splx(s);
250 		return (0);
251 
252 	case PRU_RCVOOB:
253 		if ((so->so_oobmark == 0 &&
254 		    (so->so_state & SS_RCVATMARK) == 0) ||
255 		    so->so_options & SO_OOBINLINE ||
256 		    tp->t_oobflags & TCPOOB_HADDATA) {
257 			error = EINVAL;
258 			break;
259 		}
260 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
261 			error = EWOULDBLOCK;
262 			break;
263 		}
264 		m->m_len = 1;
265 		*mtod(m, caddr_t) = tp->t_iobc;
266 		if (((int)nam & MSG_PEEK) == 0)
267 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
268 		break;
269 
270 	case PRU_SENDOOB:
271 		if (sbspace(&so->so_snd) < -512) {
272 			m_freem(m);
273 			error = ENOBUFS;
274 			break;
275 		}
276 		/*
277 		 * According to RFC961 (Assigned Protocols),
278 		 * the urgent pointer points to the last octet
279 		 * of urgent data.  We continue, however,
280 		 * to consider it to indicate the first octet
281 		 * of data past the urgent section.
282 		 * Otherwise, snd_up should be one lower.
283 		 */
284 		sbappend(&so->so_snd, m);
285 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
286 		tp->t_force = 1;
287 		error = tcp_output(tp);
288 		tp->t_force = 0;
289 		break;
290 
291 	case PRU_SOCKADDR:
292 		in_setsockaddr(inp, nam);
293 		break;
294 
295 	case PRU_PEERADDR:
296 		in_setpeeraddr(inp, nam);
297 		break;
298 
299 	/*
300 	 * TCP slow timer went off; going through this
301 	 * routine for tracing's sake.
302 	 */
303 	case PRU_SLOWTIMO:
304 		tp = tcp_timers(tp, (int)nam);
305 		req |= (int)nam << 8;		/* for debug's sake */
306 		break;
307 
308 	default:
309 		panic("tcp_usrreq");
310 	}
311 	if (tp && (so->so_options & SO_DEBUG))
312 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
313 	splx(s);
314 	return (error);
315 }
316 
317 tcp_ctloutput(op, so, level, optname, mp)
318 	int op;
319 	struct socket *so;
320 	int level, optname;
321 	struct mbuf **mp;
322 {
323 	int error = 0;
324 	struct inpcb *inp = sotoinpcb(so);
325 	register struct tcpcb *tp = intotcpcb(inp);
326 	register struct mbuf *m;
327 
328 	if (level != IPPROTO_TCP)
329 		return (ip_ctloutput(op, so, level, optname, mp));
330 
331 	switch (op) {
332 
333 	case PRCO_SETOPT:
334 		m = *mp;
335 		switch (optname) {
336 
337 		case TCP_NODELAY:
338 			if (m == NULL || m->m_len < sizeof (int))
339 				error = EINVAL;
340 			else if (*mtod(m, int *))
341 				tp->t_flags |= TF_NODELAY;
342 			else
343 				tp->t_flags &= ~TF_NODELAY;
344 			break;
345 
346 		case TCP_MAXSEG:	/* not yet */
347 		default:
348 			error = EINVAL;
349 			break;
350 		}
351 		if (m)
352 			(void) m_free(m);
353 		break;
354 
355 	case PRCO_GETOPT:
356 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
357 		m->m_len = sizeof(int);
358 
359 		switch (optname) {
360 		case TCP_NODELAY:
361 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
362 			break;
363 		case TCP_MAXSEG:
364 			*mtod(m, int *) = tp->t_maxseg;
365 			break;
366 		default:
367 			error = EINVAL;
368 			break;
369 		}
370 		break;
371 	}
372 	return (error);
373 }
374 
375 u_long	tcp_sendspace = 1024*4;
376 u_long	tcp_recvspace = 1024*4;
377 /*
378  * Attach TCP protocol to socket, allocating
379  * internet protocol control block, tcp control block,
380  * bufer space, and entering LISTEN state if to accept connections.
381  */
382 tcp_attach(so)
383 	struct socket *so;
384 {
385 	register struct tcpcb *tp;
386 	struct inpcb *inp;
387 	int error;
388 
389 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
390 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
391 		if (error)
392 			return (error);
393 	}
394 	error = in_pcballoc(so, &tcb);
395 	if (error)
396 		return (error);
397 	inp = sotoinpcb(so);
398 	tp = tcp_newtcpcb(inp);
399 	if (tp == 0) {
400 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
401 
402 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
403 		in_pcbdetach(inp);
404 		so->so_state |= nofd;
405 		return (ENOBUFS);
406 	}
407 	tp->t_state = TCPS_CLOSED;
408 	return (0);
409 }
410 
411 /*
412  * Initiate (or continue) disconnect.
413  * If embryonic state, just send reset (once).
414  * If in ``let data drain'' option and linger null, just drop.
415  * Otherwise (hard), mark socket disconnecting and drop
416  * current input data; switch states based on user close, and
417  * send segment to peer (with FIN).
418  */
419 struct tcpcb *
420 tcp_disconnect(tp)
421 	register struct tcpcb *tp;
422 {
423 	struct socket *so = tp->t_inpcb->inp_socket;
424 
425 	if (tp->t_state < TCPS_ESTABLISHED)
426 		tp = tcp_close(tp);
427 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
428 		tp = tcp_drop(tp, 0);
429 	else {
430 		soisdisconnecting(so);
431 		sbflush(&so->so_rcv);
432 		tp = tcp_usrclosed(tp);
433 		if (tp)
434 			(void) tcp_output(tp);
435 	}
436 	return (tp);
437 }
438 
439 /*
440  * User issued close, and wish to trail through shutdown states:
441  * if never received SYN, just forget it.  If got a SYN from peer,
442  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
443  * If already got a FIN from peer, then almost done; go to LAST_ACK
444  * state.  In all other cases, have already sent FIN to peer (e.g.
445  * after PRU_SHUTDOWN), and just have to play tedious game waiting
446  * for peer to send FIN or not respond to keep-alives, etc.
447  * We can let the user exit from the close as soon as the FIN is acked.
448  */
449 struct tcpcb *
450 tcp_usrclosed(tp)
451 	register struct tcpcb *tp;
452 {
453 
454 	switch (tp->t_state) {
455 
456 	case TCPS_CLOSED:
457 	case TCPS_LISTEN:
458 	case TCPS_SYN_SENT:
459 		tp->t_state = TCPS_CLOSED;
460 		tp = tcp_close(tp);
461 		break;
462 
463 	case TCPS_SYN_RECEIVED:
464 	case TCPS_ESTABLISHED:
465 		tp->t_state = TCPS_FIN_WAIT_1;
466 		break;
467 
468 	case TCPS_CLOSE_WAIT:
469 		tp->t_state = TCPS_LAST_ACK;
470 		break;
471 	}
472 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
473 		soisdisconnected(tp->t_inpcb->inp_socket);
474 	return (tp);
475 }
476