xref: /netbsd-src/sys/netinet/tcp_usrreq.c (revision 2a399c6883d870daece976daec6ffa7bb7f934ce)
1 /*	$NetBSD: tcp_usrreq.c,v 1.33 1998/01/05 10:32:12 thorpej Exp $	*/
2 
3 /*
4  * Copyright (c) 1982, 1986, 1988, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. All advertising materials mentioning features or use of this software
16  *    must display the following acknowledgement:
17  *	This product includes software developed by the University of
18  *	California, Berkeley and its contributors.
19  * 4. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  *
35  *	@(#)tcp_usrreq.c	8.5 (Berkeley) 6/21/95
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/protosw.h>
46 #include <sys/errno.h>
47 #include <sys/stat.h>
48 #include <sys/proc.h>
49 #include <sys/ucred.h>
50 
51 #include <vm/vm.h>
52 #include <sys/sysctl.h>
53 
54 #include <net/if.h>
55 #include <net/route.h>
56 
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip.h>
61 #include <netinet/in_pcb.h>
62 #include <netinet/ip_var.h>
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_fsm.h>
65 #include <netinet/tcp_seq.h>
66 #include <netinet/tcp_timer.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet/tcpip.h>
69 #include <netinet/tcp_debug.h>
70 
71 #include "opt_tcp_recvspace.h"
72 #include "opt_tcp_sendspace.h"
73 
74 /*
75  * TCP protocol interface to socket abstraction.
76  */
77 extern	char *tcpstates[];
78 
79 /*
80  * Process a TCP user request for TCP tb.  If this is a send request
81  * then m is the mbuf chain of send data.  If this is a timer expiration
82  * (called from the software clock routine), then timertype tells which timer.
83  */
84 /*ARGSUSED*/
85 int
86 tcp_usrreq(so, req, m, nam, control, p)
87 	struct socket *so;
88 	int req;
89 	struct mbuf *m, *nam, *control;
90 	struct proc *p;
91 {
92 	register struct inpcb *inp;
93 	register struct tcpcb *tp = NULL;
94 	int s;
95 	int error = 0;
96 	int ostate;
97 
98 	if (req == PRU_CONTROL)
99 		return (in_control(so, (long)m, (caddr_t)nam,
100 		    (struct ifnet *)control, p));
101 
102 	s = splsoftnet();
103 	inp = sotoinpcb(so);
104 #ifdef DIAGNOSTIC
105 	if (req != PRU_SEND && req != PRU_SENDOOB && control)
106 		panic("tcp_usrreq: unexpected control mbuf");
107 #endif
108 	/*
109 	 * When a TCP is attached to a socket, then there will be
110 	 * a (struct inpcb) pointed at by the socket, and this
111 	 * structure will point at a subsidary (struct tcpcb).
112 	 */
113 	if (inp == 0 && req != PRU_ATTACH) {
114 		error = EINVAL;
115 		goto release;
116 	}
117 	if (inp) {
118 		tp = intotcpcb(inp);
119 		/* WHAT IF TP IS 0? */
120 #ifdef KPROF
121 		tcp_acounts[tp->t_state][req]++;
122 #endif
123 		ostate = tp->t_state;
124 	} else
125 		ostate = 0;
126 
127 	switch (req) {
128 
129 	/*
130 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
131 	 * and an internet control block.
132 	 */
133 	case PRU_ATTACH:
134 		if (inp != 0) {
135 			error = EISCONN;
136 			break;
137 		}
138 		error = tcp_attach(so);
139 		if (error)
140 			break;
141 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
142 			so->so_linger = TCP_LINGERTIME;
143 		tp = sototcpcb(so);
144 		break;
145 
146 	/*
147 	 * PRU_DETACH detaches the TCP protocol from the socket.
148 	 */
149 	case PRU_DETACH:
150 		tp = tcp_disconnect(tp);
151 		break;
152 
153 	/*
154 	 * Give the socket an address.
155 	 */
156 	case PRU_BIND:
157 		error = in_pcbbind(inp, nam, p);
158 		break;
159 
160 	/*
161 	 * Prepare to accept connections.
162 	 */
163 	case PRU_LISTEN:
164 		if (inp->inp_lport == 0) {
165 			error = in_pcbbind(inp, (struct mbuf *)0,
166 			    (struct proc *)0);
167 			if (error)
168 				break;
169 		}
170 		tp->t_state = TCPS_LISTEN;
171 		break;
172 
173 	/*
174 	 * Initiate connection to peer.
175 	 * Create a template for use in transmissions on this connection.
176 	 * Enter SYN_SENT state, and mark socket as connecting.
177 	 * Start keep-alive timer, and seed output sequence space.
178 	 * Send initial segment on connection.
179 	 */
180 	case PRU_CONNECT:
181 		if (inp->inp_lport == 0) {
182 			error = in_pcbbind(inp, (struct mbuf *)0,
183 			    (struct proc *)0);
184 			if (error)
185 				break;
186 		}
187 		error = in_pcbconnect(inp, nam);
188 		if (error)
189 			break;
190 		tp->t_template = tcp_template(tp);
191 		if (tp->t_template == 0) {
192 			in_pcbdisconnect(inp);
193 			error = ENOBUFS;
194 			break;
195 		}
196 		/* Compute window scaling to request.  */
197 		while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
198 		    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
199 			tp->request_r_scale++;
200 		soisconnecting(so);
201 		tcpstat.tcps_connattempt++;
202 		tp->t_state = TCPS_SYN_SENT;
203 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
204 		tp->iss = tcp_new_iss(tp, sizeof(struct tcpcb), 0);
205 		tcp_sendseqinit(tp);
206 		error = tcp_output(tp);
207 		break;
208 
209 	/*
210 	 * Create a TCP connection between two sockets.
211 	 */
212 	case PRU_CONNECT2:
213 		error = EOPNOTSUPP;
214 		break;
215 
216 	/*
217 	 * Initiate disconnect from peer.
218 	 * If connection never passed embryonic stage, just drop;
219 	 * else if don't need to let data drain, then can just drop anyways,
220 	 * else have to begin TCP shutdown process: mark socket disconnecting,
221 	 * drain unread data, state switch to reflect user close, and
222 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
223 	 * when peer sends FIN and acks ours.
224 	 *
225 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
226 	 */
227 	case PRU_DISCONNECT:
228 		tp = tcp_disconnect(tp);
229 		break;
230 
231 	/*
232 	 * Accept a connection.  Essentially all the work is
233 	 * done at higher levels; just return the address
234 	 * of the peer, storing through addr.
235 	 */
236 	case PRU_ACCEPT:
237 		in_setpeeraddr(inp, nam);
238 		break;
239 
240 	/*
241 	 * Mark the connection as being incapable of further output.
242 	 */
243 	case PRU_SHUTDOWN:
244 		socantsendmore(so);
245 		tp = tcp_usrclosed(tp);
246 		if (tp)
247 			error = tcp_output(tp);
248 		break;
249 
250 	/*
251 	 * After a receive, possibly send window update to peer.
252 	 */
253 	case PRU_RCVD:
254 		(void) tcp_output(tp);
255 		break;
256 
257 	/*
258 	 * Do a send by putting data in output queue and updating urgent
259 	 * marker if URG set.  Possibly send more data.
260 	 */
261 	case PRU_SEND:
262 		if (control && control->m_len) {
263 			m_freem(control);
264 			m_freem(m);
265 			error = EINVAL;
266 			break;
267 		}
268 		sbappend(&so->so_snd, m);
269 		error = tcp_output(tp);
270 		break;
271 
272 	/*
273 	 * Abort the TCP.
274 	 */
275 	case PRU_ABORT:
276 		tp = tcp_drop(tp, ECONNABORTED);
277 		break;
278 
279 	case PRU_SENSE:
280 		/*
281 		 * stat: don't bother with a blocksize.
282 		 */
283 		splx(s);
284 		return (0);
285 
286 	case PRU_RCVOOB:
287 		if (control && control->m_len) {
288 			m_freem(control);
289 			m_freem(m);
290 			error = EINVAL;
291 			break;
292 		}
293 		if ((so->so_oobmark == 0 &&
294 		    (so->so_state & SS_RCVATMARK) == 0) ||
295 		    so->so_options & SO_OOBINLINE ||
296 		    tp->t_oobflags & TCPOOB_HADDATA) {
297 			error = EINVAL;
298 			break;
299 		}
300 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
301 			error = EWOULDBLOCK;
302 			break;
303 		}
304 		m->m_len = 1;
305 		*mtod(m, caddr_t) = tp->t_iobc;
306 		if (((long)nam & MSG_PEEK) == 0)
307 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
308 		break;
309 
310 	case PRU_SENDOOB:
311 		if (sbspace(&so->so_snd) < -512) {
312 			m_freem(m);
313 			error = ENOBUFS;
314 			break;
315 		}
316 		/*
317 		 * According to RFC961 (Assigned Protocols),
318 		 * the urgent pointer points to the last octet
319 		 * of urgent data.  We continue, however,
320 		 * to consider it to indicate the first octet
321 		 * of data past the urgent section.
322 		 * Otherwise, snd_up should be one lower.
323 		 */
324 		sbappend(&so->so_snd, m);
325 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
326 		tp->t_force = 1;
327 		error = tcp_output(tp);
328 		tp->t_force = 0;
329 		break;
330 
331 	case PRU_SOCKADDR:
332 		in_setsockaddr(inp, nam);
333 		break;
334 
335 	case PRU_PEERADDR:
336 		in_setpeeraddr(inp, nam);
337 		break;
338 
339 	/*
340 	 * TCP slow timer went off; going through this
341 	 * routine for tracing's sake.
342 	 */
343 	case PRU_SLOWTIMO:
344 		tp = tcp_timers(tp, (long)nam);
345 		req |= (long)nam << 8;		/* for debug's sake */
346 		break;
347 
348 	default:
349 		panic("tcp_usrreq");
350 	}
351 	if (tp && (so->so_options & SO_DEBUG))
352 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
353 
354 release:
355 	splx(s);
356 	return (error);
357 }
358 
359 int
360 tcp_ctloutput(op, so, level, optname, mp)
361 	int op;
362 	struct socket *so;
363 	int level, optname;
364 	struct mbuf **mp;
365 {
366 	int error = 0, s;
367 	struct inpcb *inp;
368 	register struct tcpcb *tp;
369 	register struct mbuf *m;
370 	register int i;
371 
372 	s = splsoftnet();
373 	inp = sotoinpcb(so);
374 	if (inp == NULL) {
375 		splx(s);
376 		if (op == PRCO_SETOPT && *mp)
377 			(void) m_free(*mp);
378 		return (ECONNRESET);
379 	}
380 	if (level != IPPROTO_TCP) {
381 		error = ip_ctloutput(op, so, level, optname, mp);
382 		splx(s);
383 		return (error);
384 	}
385 	tp = intotcpcb(inp);
386 
387 	switch (op) {
388 
389 	case PRCO_SETOPT:
390 		m = *mp;
391 		switch (optname) {
392 
393 		case TCP_NODELAY:
394 			if (m == NULL || m->m_len < sizeof (int))
395 				error = EINVAL;
396 			else if (*mtod(m, int *))
397 				tp->t_flags |= TF_NODELAY;
398 			else
399 				tp->t_flags &= ~TF_NODELAY;
400 			break;
401 
402 		case TCP_MAXSEG:
403 			if (m && (i = *mtod(m, int *)) > 0 &&
404 			    i <= tp->t_peermss)
405 				tp->t_peermss = i;  /* limit on send size */
406 			else
407 				error = EINVAL;
408 			break;
409 
410 		default:
411 			error = ENOPROTOOPT;
412 			break;
413 		}
414 		if (m)
415 			(void) m_free(m);
416 		break;
417 
418 	case PRCO_GETOPT:
419 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
420 		m->m_len = sizeof(int);
421 
422 		switch (optname) {
423 		case TCP_NODELAY:
424 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
425 			break;
426 		case TCP_MAXSEG:
427 			*mtod(m, int *) = tp->t_peermss;
428 			break;
429 		default:
430 			error = ENOPROTOOPT;
431 			break;
432 		}
433 		break;
434 	}
435 	splx(s);
436 	return (error);
437 }
438 
439 #ifndef TCP_SENDSPACE
440 #define	TCP_SENDSPACE	1024*16;
441 #endif
442 int	tcp_sendspace = TCP_SENDSPACE;
443 #ifndef TCP_RECVSPACE
444 #define	TCP_RECVSPACE	1024*16;
445 #endif
446 int	tcp_recvspace = TCP_RECVSPACE;
447 
448 /*
449  * Attach TCP protocol to socket, allocating
450  * internet protocol control block, tcp control block,
451  * bufer space, and entering LISTEN state if to accept connections.
452  */
453 int
454 tcp_attach(so)
455 	struct socket *so;
456 {
457 	register struct tcpcb *tp;
458 	struct inpcb *inp;
459 	int error;
460 
461 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
462 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
463 		if (error)
464 			return (error);
465 	}
466 	error = in_pcballoc(so, &tcbtable);
467 	if (error)
468 		return (error);
469 	inp = sotoinpcb(so);
470 	tp = tcp_newtcpcb(inp);
471 	if (tp == 0) {
472 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
473 
474 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
475 		in_pcbdetach(inp);
476 		so->so_state |= nofd;
477 		return (ENOBUFS);
478 	}
479 	tp->t_state = TCPS_CLOSED;
480 	return (0);
481 }
482 
483 /*
484  * Initiate (or continue) disconnect.
485  * If embryonic state, just send reset (once).
486  * If in ``let data drain'' option and linger null, just drop.
487  * Otherwise (hard), mark socket disconnecting and drop
488  * current input data; switch states based on user close, and
489  * send segment to peer (with FIN).
490  */
491 struct tcpcb *
492 tcp_disconnect(tp)
493 	register struct tcpcb *tp;
494 {
495 	struct socket *so = tp->t_inpcb->inp_socket;
496 
497 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
498 		tp = tcp_close(tp);
499 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
500 		tp = tcp_drop(tp, 0);
501 	else {
502 		soisdisconnecting(so);
503 		sbflush(&so->so_rcv);
504 		tp = tcp_usrclosed(tp);
505 		if (tp)
506 			(void) tcp_output(tp);
507 	}
508 	return (tp);
509 }
510 
511 /*
512  * User issued close, and wish to trail through shutdown states:
513  * if never received SYN, just forget it.  If got a SYN from peer,
514  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
515  * If already got a FIN from peer, then almost done; go to LAST_ACK
516  * state.  In all other cases, have already sent FIN to peer (e.g.
517  * after PRU_SHUTDOWN), and just have to play tedious game waiting
518  * for peer to send FIN or not respond to keep-alives, etc.
519  * We can let the user exit from the close as soon as the FIN is acked.
520  */
521 struct tcpcb *
522 tcp_usrclosed(tp)
523 	register struct tcpcb *tp;
524 {
525 
526 	switch (tp->t_state) {
527 
528 	case TCPS_CLOSED:
529 	case TCPS_LISTEN:
530 	case TCPS_SYN_SENT:
531 		tp->t_state = TCPS_CLOSED;
532 		tp = tcp_close(tp);
533 		break;
534 
535 	case TCPS_SYN_RECEIVED:
536 	case TCPS_ESTABLISHED:
537 		tp->t_state = TCPS_FIN_WAIT_1;
538 		break;
539 
540 	case TCPS_CLOSE_WAIT:
541 		tp->t_state = TCPS_LAST_ACK;
542 		break;
543 	}
544 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
545 		soisdisconnected(tp->t_inpcb->inp_socket);
546 		/*
547 		 * If we are in FIN_WAIT_2, we arrived here because the
548 		 * application did a shutdown of the send side.  Like the
549 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
550 		 * a full close, we start a timer to make sure sockets are
551 		 * not left in FIN_WAIT_2 forever.
552 		 */
553 		if (tp->t_state == TCPS_FIN_WAIT_2)
554 			tp->t_timer[TCPT_2MSL] = tcp_maxidle;
555 	}
556 	return (tp);
557 }
558 
559 /*
560  * Sysctl for tcp variables.
561  */
562 int
563 tcp_sysctl(name, namelen, oldp, oldlenp, newp, newlen)
564 	int *name;
565 	u_int namelen;
566 	void *oldp;
567 	size_t *oldlenp;
568 	void *newp;
569 	size_t newlen;
570 {
571 
572 	/* All sysctl names at this level are terminal. */
573 	if (namelen != 1)
574 		return (ENOTDIR);
575 
576 	switch (name[0]) {
577 	case TCPCTL_RFC1323:
578 		return (sysctl_int(oldp, oldlenp, newp, newlen,
579 		    &tcp_do_rfc1323));
580 	case TCPCTL_SENDSPACE:
581 		return (sysctl_int(oldp, oldlenp, newp, newlen,
582 		    &tcp_sendspace));
583 	case TCPCTL_RECVSPACE:
584 		return (sysctl_int(oldp, oldlenp, newp, newlen,
585 		    &tcp_recvspace));
586 	case TCPCTL_MSSDFLT:
587 		return (sysctl_int(oldp, oldlenp, newp, newlen,
588 		    &tcp_mssdflt));
589 	case TCPCTL_SYN_CACHE_LIMIT:
590 		return (sysctl_int(oldp, oldlenp, newp, newlen,
591 		    &tcp_syn_cache_limit));
592 	case TCPCTL_SYN_BUCKET_LIMIT:
593 		return (sysctl_int(oldp, oldlenp, newp, newlen,
594 		    &tcp_syn_bucket_limit));
595 	case TCPCTL_SYN_CACHE_INTER:
596 		return (sysctl_int(oldp, oldlenp, newp, newlen,
597 		    &tcp_syn_cache_interval));
598 	case TCPCTL_INIT_WIN:
599 		return (sysctl_int(oldp, oldlenp, newp, newlen,
600 		    &tcp_init_win));
601 	default:
602 		return (ENOPROTOOPT);
603 	}
604 	/* NOTREACHED */
605 }
606