xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 32789)
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 this notice is preserved and that due credit is given
7  * to the University of California at Berkeley. The name of the University
8  * may not be used to endorse or promote products derived from this
9  * software without specific prior written permission. This software
10  * is provided ``as is'' without express or implied warranty.
11  *
12  *	@(#)tcp_usrreq.c	7.6 (Berkeley) 12/07/87
13  */
14 
15 #include "param.h"
16 #include "systm.h"
17 #include "mbuf.h"
18 #include "socket.h"
19 #include "socketvar.h"
20 #include "protosw.h"
21 #include "errno.h"
22 #include "stat.h"
23 
24 #include "../net/if.h"
25 #include "../net/route.h"
26 
27 #include "in.h"
28 #include "in_pcb.h"
29 #include "in_systm.h"
30 #include "ip.h"
31 #include "ip_var.h"
32 #include "tcp.h"
33 #include "tcp_fsm.h"
34 #include "tcp_seq.h"
35 #include "tcp_timer.h"
36 #include "tcp_var.h"
37 #include "tcpip.h"
38 #include "tcp_debug.h"
39 
40 /*
41  * TCP protocol interface to socket abstraction.
42  */
43 extern	char *tcpstates[];
44 struct	tcpcb *tcp_newtcpcb();
45 int	tcpsenderrors;
46 
47 /*
48  * Process a TCP user request for TCP tb.  If this is a send request
49  * then m is the mbuf chain of send data.  If this is a timer expiration
50  * (called from the software clock routine), then timertype tells which timer.
51  */
52 /*ARGSUSED*/
53 tcp_usrreq(so, req, m, nam, rights)
54 	struct socket *so;
55 	int req;
56 	struct mbuf *m, *nam, *rights;
57 {
58 	register struct inpcb *inp;
59 	register struct tcpcb *tp;
60 	int s;
61 	int error = 0;
62 	int ostate;
63 
64 	if (req == PRU_CONTROL)
65 		return (in_control(so, (int)m, (caddr_t)nam,
66 			(struct ifnet *)rights));
67 	if (rights && rights->m_len)
68 		return (EINVAL);
69 
70 	s = splnet();
71 	inp = sotoinpcb(so);
72 	/*
73 	 * When a TCP is attached to a socket, then there will be
74 	 * a (struct inpcb) pointed at by the socket, and this
75 	 * structure will point at a subsidary (struct tcpcb).
76 	 */
77 	if (inp == 0 && req != PRU_ATTACH) {
78 		splx(s);
79 		return (EINVAL);		/* XXX */
80 	}
81 	if (inp) {
82 		tp = intotcpcb(inp);
83 		/* WHAT IF TP IS 0? */
84 #ifdef KPROF
85 		tcp_acounts[tp->t_state][req]++;
86 #endif
87 		ostate = tp->t_state;
88 	} else
89 		ostate = 0;
90 	switch (req) {
91 
92 	/*
93 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
94 	 * and an internet control block.
95 	 */
96 	case PRU_ATTACH:
97 		if (inp) {
98 			error = EISCONN;
99 			break;
100 		}
101 		error = tcp_attach(so);
102 		if (error)
103 			break;
104 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
105 			so->so_linger = TCP_LINGERTIME;
106 		tp = sototcpcb(so);
107 		break;
108 
109 	/*
110 	 * PRU_DETACH detaches the TCP protocol from the socket.
111 	 * If the protocol state is non-embryonic, then can't
112 	 * do this directly: have to initiate a PRU_DISCONNECT,
113 	 * which may finish later; embryonic TCB's can just
114 	 * be discarded here.
115 	 */
116 	case PRU_DETACH:
117 		if (tp->t_state > TCPS_LISTEN)
118 			tp = tcp_disconnect(tp);
119 		else
120 			tp = tcp_close(tp);
121 		break;
122 
123 	/*
124 	 * Give the socket an address.
125 	 */
126 	case PRU_BIND:
127 		error = in_pcbbind(inp, nam);
128 		if (error)
129 			break;
130 		break;
131 
132 	/*
133 	 * Prepare to accept connections.
134 	 */
135 	case PRU_LISTEN:
136 		if (inp->inp_lport == 0)
137 			error = in_pcbbind(inp, (struct mbuf *)0);
138 		if (error == 0)
139 			tp->t_state = TCPS_LISTEN;
140 		break;
141 
142 	/*
143 	 * Initiate connection to peer.
144 	 * Create a template for use in transmissions on this connection.
145 	 * Enter SYN_SENT state, and mark socket as connecting.
146 	 * Start keep-alive timer, and seed output sequence space.
147 	 * Send initial segment on connection.
148 	 */
149 	case PRU_CONNECT:
150 		if (inp->inp_lport == 0) {
151 			error = in_pcbbind(inp, (struct mbuf *)0);
152 			if (error)
153 				break;
154 		}
155 		error = in_pcbconnect(inp, nam);
156 		if (error)
157 			break;
158 		tp->t_template = tcp_template(tp);
159 		if (tp->t_template == 0) {
160 			in_pcbdisconnect(inp);
161 			error = ENOBUFS;
162 			break;
163 		}
164 		soisconnecting(so);
165 		tcpstat.tcps_connattempt++;
166 		tp->t_state = TCPS_SYN_SENT;
167 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
168 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
169 		tcp_sendseqinit(tp);
170 		error = tcp_output(tp);
171 		break;
172 
173 	/*
174 	 * Create a TCP connection between two sockets.
175 	 */
176 	case PRU_CONNECT2:
177 		error = EOPNOTSUPP;
178 		break;
179 
180 	/*
181 	 * Initiate disconnect from peer.
182 	 * If connection never passed embryonic stage, just drop;
183 	 * else if don't need to let data drain, then can just drop anyways,
184 	 * else have to begin TCP shutdown process: mark socket disconnecting,
185 	 * drain unread data, state switch to reflect user close, and
186 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
187 	 * when peer sends FIN and acks ours.
188 	 *
189 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
190 	 */
191 	case PRU_DISCONNECT:
192 		tp = tcp_disconnect(tp);
193 		break;
194 
195 	/*
196 	 * Accept a connection.  Essentially all the work is
197 	 * done at higher levels; just return the address
198 	 * of the peer, storing through addr.
199 	 */
200 	case PRU_ACCEPT: {
201 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
202 
203 		nam->m_len = sizeof (struct sockaddr_in);
204 		sin->sin_family = AF_INET;
205 		sin->sin_port = inp->inp_fport;
206 		sin->sin_addr = inp->inp_faddr;
207 		break;
208 		}
209 
210 	/*
211 	 * Mark the connection as being incapable of further output.
212 	 */
213 	case PRU_SHUTDOWN:
214 		socantsendmore(so);
215 		tp = tcp_usrclosed(tp);
216 		if (tp)
217 			error = tcp_output(tp);
218 		break;
219 
220 	/*
221 	 * After a receive, possibly send window update to peer.
222 	 */
223 	case PRU_RCVD:
224 		(void) tcp_output(tp);
225 		break;
226 
227 	/*
228 	 * Do a send by putting data in output queue and updating urgent
229 	 * marker if URG set.  Possibly send more data.
230 	 */
231 	case PRU_SEND:
232 		sbappend(&so->so_snd, m);
233 		error = tcp_output(tp);
234 		if (error) {		/* XXX fix to use other path */
235 			if (error == ENOBUFS)		/* XXX */
236 				error = 0;		/* XXX */
237 			tcpsenderrors++;
238 		}
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 int	tcp_sendspace = 1024*4;
377 int	tcp_recvspace = 1024*4;
378 /*
379  * Attach TCP protocol to socket, allocating
380  * internet protocol control block, tcp control block,
381  * bufer space, and entering LISTEN state if to accept connections.
382  */
383 tcp_attach(so)
384 	struct socket *so;
385 {
386 	register struct tcpcb *tp;
387 	struct inpcb *inp;
388 	int error;
389 
390 	error = soreserve(so, tcp_sendspace, tcp_recvspace);
391 	if (error)
392 		return (error);
393 	error = in_pcballoc(so, &tcb);
394 	if (error)
395 		return (error);
396 	inp = sotoinpcb(so);
397 	tp = tcp_newtcpcb(inp);
398 	if (tp == 0) {
399 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
400 
401 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
402 		in_pcbdetach(inp);
403 		so->so_state |= nofd;
404 		return (ENOBUFS);
405 	}
406 	tp->t_state = TCPS_CLOSED;
407 	return (0);
408 }
409 
410 /*
411  * Initiate (or continue) disconnect.
412  * If embryonic state, just send reset (once).
413  * If in ``let data drain'' option and linger null, just drop.
414  * Otherwise (hard), mark socket disconnecting and drop
415  * current input data; switch states based on user close, and
416  * send segment to peer (with FIN).
417  */
418 struct tcpcb *
419 tcp_disconnect(tp)
420 	register struct tcpcb *tp;
421 {
422 	struct socket *so = tp->t_inpcb->inp_socket;
423 
424 	if (tp->t_state < TCPS_ESTABLISHED)
425 		tp = tcp_close(tp);
426 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
427 		tp = tcp_drop(tp, 0);
428 	else {
429 		soisdisconnecting(so);
430 		sbflush(&so->so_rcv);
431 		tp = tcp_usrclosed(tp);
432 		if (tp)
433 			(void) tcp_output(tp);
434 	}
435 	return (tp);
436 }
437 
438 /*
439  * User issued close, and wish to trail through shutdown states:
440  * if never received SYN, just forget it.  If got a SYN from peer,
441  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
442  * If already got a FIN from peer, then almost done; go to LAST_ACK
443  * state.  In all other cases, have already sent FIN to peer (e.g.
444  * after PRU_SHUTDOWN), and just have to play tedious game waiting
445  * for peer to send FIN or not respond to keep-alives, etc.
446  * We can let the user exit from the close as soon as the FIN is acked.
447  */
448 struct tcpcb *
449 tcp_usrclosed(tp)
450 	register struct tcpcb *tp;
451 {
452 
453 	switch (tp->t_state) {
454 
455 	case TCPS_CLOSED:
456 	case TCPS_LISTEN:
457 	case TCPS_SYN_SENT:
458 		tp->t_state = TCPS_CLOSED;
459 		tp = tcp_close(tp);
460 		break;
461 
462 	case TCPS_SYN_RECEIVED:
463 	case TCPS_ESTABLISHED:
464 		tp->t_state = TCPS_FIN_WAIT_1;
465 		break;
466 
467 	case TCPS_CLOSE_WAIT:
468 		tp->t_state = TCPS_LAST_ACK;
469 		break;
470 	}
471 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
472 		soisdisconnected(tp->t_inpcb->inp_socket);
473 	return (tp);
474 }
475