xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 29155)
1 /*
2  * Copyright (c) 1982, 1986 Regents of the University of California.
3  * All rights reserved.  The Berkeley software License Agreement
4  * specifies the terms and conditions for redistribution.
5  *
6  *	@(#)tcp_usrreq.c	7.1 (Berkeley) 06/05/86
7  */
8 
9 #include "param.h"
10 #include "systm.h"
11 #include "mbuf.h"
12 #include "socket.h"
13 #include "socketvar.h"
14 #include "protosw.h"
15 #include "errno.h"
16 #include "stat.h"
17 
18 #include "../net/if.h"
19 #include "../net/route.h"
20 
21 #include "in.h"
22 #include "in_pcb.h"
23 #include "in_systm.h"
24 #include "ip.h"
25 #include "ip_var.h"
26 #include "tcp.h"
27 #include "tcp_fsm.h"
28 #include "tcp_seq.h"
29 #include "tcp_timer.h"
30 #include "tcp_var.h"
31 #include "tcpip.h"
32 #include "tcp_debug.h"
33 
34 /*
35  * TCP protocol interface to socket abstraction.
36  */
37 extern	char *tcpstates[];
38 struct	tcpcb *tcp_newtcpcb();
39 int	tcpsenderrors;
40 
41 /*
42  * Process a TCP user request for TCP tb.  If this is a send request
43  * then m is the mbuf chain of send data.  If this is a timer expiration
44  * (called from the software clock routine), then timertype tells which timer.
45  */
46 /*ARGSUSED*/
47 tcp_usrreq(so, req, m, nam, rights)
48 	struct socket *so;
49 	int req;
50 	struct mbuf *m, *nam, *rights;
51 {
52 	register struct inpcb *inp = sotoinpcb(so);
53 	register struct tcpcb *tp;
54 	int s = splnet();
55 	int error = 0;
56 	int ostate;
57 
58 	if (req == PRU_CONTROL)
59 		return (in_control(so, (int)m, (caddr_t)nam,
60 			(struct ifnet *)rights));
61 	if (rights && rights->m_len) {
62 		splx(s);
63 		return (EINVAL);
64 	}
65 	/*
66 	 * When a TCP is attached to a socket, then there will be
67 	 * a (struct inpcb) pointed at by the socket, and this
68 	 * structure will point at a subsidary (struct tcpcb).
69 	 */
70 	if (inp == 0 && req != PRU_ATTACH) {
71 		splx(s);
72 		return (EINVAL);		/* XXX */
73 	}
74 	if (inp) {
75 		tp = intotcpcb(inp);
76 		/* WHAT IF TP IS 0? */
77 #ifdef KPROF
78 		tcp_acounts[tp->t_state][req]++;
79 #endif
80 		ostate = tp->t_state;
81 	} else
82 		ostate = 0;
83 	switch (req) {
84 
85 	/*
86 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
87 	 * and an internet control block.
88 	 */
89 	case PRU_ATTACH:
90 		if (inp) {
91 			error = EISCONN;
92 			break;
93 		}
94 		error = tcp_attach(so);
95 		if (error)
96 			break;
97 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
98 			so->so_linger = TCP_LINGERTIME;
99 		tp = sototcpcb(so);
100 		break;
101 
102 	/*
103 	 * PRU_DETACH detaches the TCP protocol from the socket.
104 	 * If the protocol state is non-embryonic, then can't
105 	 * do this directly: have to initiate a PRU_DISCONNECT,
106 	 * which may finish later; embryonic TCB's can just
107 	 * be discarded here.
108 	 */
109 	case PRU_DETACH:
110 		if (tp->t_state > TCPS_LISTEN)
111 			tp = tcp_disconnect(tp);
112 		else
113 			tp = tcp_close(tp);
114 		break;
115 
116 	/*
117 	 * Give the socket an address.
118 	 */
119 	case PRU_BIND:
120 		error = in_pcbbind(inp, nam);
121 		if (error)
122 			break;
123 		break;
124 
125 	/*
126 	 * Prepare to accept connections.
127 	 */
128 	case PRU_LISTEN:
129 		if (inp->inp_lport == 0)
130 			error = in_pcbbind(inp, (struct mbuf *)0);
131 		if (error == 0)
132 			tp->t_state = TCPS_LISTEN;
133 		break;
134 
135 	/*
136 	 * Initiate connection to peer.
137 	 * Create a template for use in transmissions on this connection.
138 	 * Enter SYN_SENT state, and mark socket as connecting.
139 	 * Start keep-alive timer, and seed output sequence space.
140 	 * Send initial segment on connection.
141 	 */
142 	case PRU_CONNECT:
143 		if (inp->inp_lport == 0) {
144 			error = in_pcbbind(inp, (struct mbuf *)0);
145 			if (error)
146 				break;
147 		}
148 		error = in_pcbconnect(inp, nam);
149 		if (error)
150 			break;
151 		tp->t_template = tcp_template(tp);
152 		if (tp->t_template == 0) {
153 			in_pcbdisconnect(inp);
154 			error = ENOBUFS;
155 			break;
156 		}
157 		soisconnecting(so);
158 		tp->t_state = TCPS_SYN_SENT;
159 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
160 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
161 		tcp_sendseqinit(tp);
162 		error = tcp_output(tp);
163 		break;
164 
165 	/*
166 	 * Create a TCP connection between two sockets.
167 	 */
168 	case PRU_CONNECT2:
169 		error = EOPNOTSUPP;
170 		break;
171 
172 	/*
173 	 * Initiate disconnect from peer.
174 	 * If connection never passed embryonic stage, just drop;
175 	 * else if don't need to let data drain, then can just drop anyways,
176 	 * else have to begin TCP shutdown process: mark socket disconnecting,
177 	 * drain unread data, state switch to reflect user close, and
178 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
179 	 * when peer sends FIN and acks ours.
180 	 *
181 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
182 	 */
183 	case PRU_DISCONNECT:
184 		tp = tcp_disconnect(tp);
185 		break;
186 
187 	/*
188 	 * Accept a connection.  Essentially all the work is
189 	 * done at higher levels; just return the address
190 	 * of the peer, storing through addr.
191 	 */
192 	case PRU_ACCEPT: {
193 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
194 
195 		nam->m_len = sizeof (struct sockaddr_in);
196 		sin->sin_family = AF_INET;
197 		sin->sin_port = inp->inp_fport;
198 		sin->sin_addr = inp->inp_faddr;
199 		break;
200 		}
201 
202 	/*
203 	 * Mark the connection as being incapable of further output.
204 	 */
205 	case PRU_SHUTDOWN:
206 		socantsendmore(so);
207 		tp = tcp_usrclosed(tp);
208 		if (tp)
209 			error = tcp_output(tp);
210 		break;
211 
212 	/*
213 	 * After a receive, possibly send window update to peer.
214 	 */
215 	case PRU_RCVD:
216 		(void) tcp_output(tp);
217 		break;
218 
219 	/*
220 	 * Do a send by putting data in output queue and updating urgent
221 	 * marker if URG set.  Possibly send more data.
222 	 */
223 	case PRU_SEND:
224 		sbappend(&so->so_snd, m);
225 #ifdef notdef
226 		if (tp->t_flags & TF_PUSH)
227 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
228 #endif
229 		error = tcp_output(tp);
230 		if (error) {		/* XXX fix to use other path */
231 			if (error == ENOBUFS)		/* XXX */
232 				error = 0;		/* XXX */
233 			tcpsenderrors++;
234 		}
235 		break;
236 
237 	/*
238 	 * Abort the TCP.
239 	 */
240 	case PRU_ABORT:
241 		tp = tcp_drop(tp, ECONNABORTED);
242 		break;
243 
244 	case PRU_SENSE:
245 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
246 		return (0);
247 
248 	case PRU_RCVOOB:
249 		if ((so->so_oobmark == 0 &&
250 		    (so->so_state & SS_RCVATMARK) == 0) ||
251 		    so->so_options & SO_OOBINLINE ||
252 		    tp->t_oobflags & TCPOOB_HADDATA) {
253 			error = EINVAL;
254 			break;
255 		}
256 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
257 			error = EWOULDBLOCK;
258 			break;
259 		}
260 		m->m_len = 1;
261 		*mtod(m, caddr_t) = tp->t_iobc;
262 		if (((int)nam & MSG_PEEK) == 0)
263 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
264 		break;
265 
266 	case PRU_SENDOOB:
267 		if (sbspace(&so->so_snd) < -512) {
268 			m_freem(m);
269 			error = ENOBUFS;
270 			break;
271 		}
272 		/*
273 		 * According to RFC961 (Assigned Protocols),
274 		 * the urgent pointer points to the last octet
275 		 * of urgent data.  We continue, however,
276 		 * to consider it to indicate the first octet
277 		 * of data past the urgent section.
278 		 * Otherwise, snd_up should be one lower.
279 		 */
280 		sbappend(&so->so_snd, m);
281 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
282 		tp->t_force = 1;
283 		error = tcp_output(tp);
284 		tp->t_force = 0;
285 		break;
286 
287 	case PRU_SOCKADDR:
288 		in_setsockaddr(inp, nam);
289 		break;
290 
291 	case PRU_PEERADDR:
292 		in_setpeeraddr(inp, nam);
293 		break;
294 
295 	/*
296 	 * TCP slow timer went off; going through this
297 	 * routine for tracing's sake.
298 	 */
299 	case PRU_SLOWTIMO:
300 		tp = tcp_timers(tp, (int)nam);
301 		req |= (int)nam << 8;		/* for debug's sake */
302 		break;
303 
304 	default:
305 		panic("tcp_usrreq");
306 	}
307 	if (tp && (so->so_options & SO_DEBUG))
308 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
309 	splx(s);
310 	return (error);
311 }
312 
313 tcp_ctloutput(op, so, level, optname, mp)
314 	int op;
315 	struct socket *so;
316 	int level, optname;
317 	struct mbuf **mp;
318 {
319 	int error = 0;
320 	struct inpcb *inp = sotoinpcb(so);
321 	register struct tcpcb *tp = intotcpcb(inp);
322 	register struct mbuf *m;
323 
324 	if (level != IPPROTO_TCP)
325 		return (ip_ctloutput(op, so, level, optname, mp));
326 
327 	switch (op) {
328 
329 	case PRCO_SETOPT:
330 		m = *mp;
331 		switch (optname) {
332 
333 		case TCP_NODELAY:
334 			if (m == NULL || m->m_len < sizeof (int))
335 				error = EINVAL;
336 			else if (*mtod(m, int *))
337 				tp->t_flags |= TF_NODELAY;
338 			else
339 				tp->t_flags &= ~TF_NODELAY;
340 			break;
341 
342 		case TCP_MAXSEG:	/* not yet */
343 		default:
344 			error = EINVAL;
345 			break;
346 		}
347 		(void)m_free(m);
348 		break;
349 
350 	case PRCO_GETOPT:
351 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
352 		m->m_len = sizeof(int);
353 
354 		switch (optname) {
355 		case TCP_NODELAY:
356 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
357 			break;
358 		case TCP_MAXSEG:
359 			*mtod(m, int *) = tp->t_maxseg;
360 			break;
361 		default:
362 			error = EINVAL;
363 			break;
364 		}
365 		break;
366 	}
367 	return (error);
368 }
369 
370 int	tcp_sendspace = 1024*4;
371 int	tcp_recvspace = 1024*4;
372 /*
373  * Attach TCP protocol to socket, allocating
374  * internet protocol control block, tcp control block,
375  * bufer space, and entering LISTEN state if to accept connections.
376  */
377 tcp_attach(so)
378 	struct socket *so;
379 {
380 	register struct tcpcb *tp;
381 	struct inpcb *inp;
382 	int error;
383 
384 	error = soreserve(so, tcp_sendspace, tcp_recvspace);
385 	if (error)
386 		return (error);
387 	error = in_pcballoc(so, &tcb);
388 	if (error)
389 		return (error);
390 	inp = sotoinpcb(so);
391 	tp = tcp_newtcpcb(inp);
392 	if (tp == 0) {
393 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
394 
395 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
396 		in_pcbdetach(inp);
397 		so->so_state |= nofd;
398 		return (ENOBUFS);
399 	}
400 	tp->t_state = TCPS_CLOSED;
401 	return (0);
402 }
403 
404 /*
405  * Initiate (or continue) disconnect.
406  * If embryonic state, just send reset (once).
407  * If in ``let data drain'' option and linger null, just drop.
408  * Otherwise (hard), mark socket disconnecting and drop
409  * current input data; switch states based on user close, and
410  * send segment to peer (with FIN).
411  */
412 struct tcpcb *
413 tcp_disconnect(tp)
414 	register struct tcpcb *tp;
415 {
416 	struct socket *so = tp->t_inpcb->inp_socket;
417 
418 	if (tp->t_state < TCPS_ESTABLISHED)
419 		tp = tcp_close(tp);
420 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
421 		tp = tcp_drop(tp, 0);
422 	else {
423 		soisdisconnecting(so);
424 		sbflush(&so->so_rcv);
425 		tp = tcp_usrclosed(tp);
426 		if (tp)
427 			(void) tcp_output(tp);
428 	}
429 	return (tp);
430 }
431 
432 /*
433  * User issued close, and wish to trail through shutdown states:
434  * if never received SYN, just forget it.  If got a SYN from peer,
435  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
436  * If already got a FIN from peer, then almost done; go to LAST_ACK
437  * state.  In all other cases, have already sent FIN to peer (e.g.
438  * after PRU_SHUTDOWN), and just have to play tedious game waiting
439  * for peer to send FIN or not respond to keep-alives, etc.
440  * We can let the user exit from the close as soon as the FIN is acked.
441  */
442 struct tcpcb *
443 tcp_usrclosed(tp)
444 	register struct tcpcb *tp;
445 {
446 
447 	switch (tp->t_state) {
448 
449 	case TCPS_CLOSED:
450 	case TCPS_LISTEN:
451 	case TCPS_SYN_SENT:
452 		tp->t_state = TCPS_CLOSED;
453 		tp = tcp_close(tp);
454 		break;
455 
456 	case TCPS_SYN_RECEIVED:
457 	case TCPS_ESTABLISHED:
458 		tp->t_state = TCPS_FIN_WAIT_1;
459 		break;
460 
461 	case TCPS_CLOSE_WAIT:
462 		tp->t_state = TCPS_LAST_ACK;
463 		break;
464 	}
465 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
466 		soisdisconnected(tp->t_inpcb->inp_socket);
467 	return (tp);
468 }
469