Home | History | Annotate | Line # | Download | only in netinet
tcp_usrreq.c revision 1.22 
      1 /*	$NetBSD: tcp_usrreq.c,v 1.22 1996/05/23 16:13:19 mycroft Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1988, 1993
      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	8.2 (Berkeley) 1/3/94
     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 /*
     72  * TCP protocol interface to socket abstraction.
     73  */
     74 extern	char *tcpstates[];
     75 
     76 /*
     77  * Process a TCP user request for TCP tb.  If this is a send request
     78  * then m is the mbuf chain of send data.  If this is a timer expiration
     79  * (called from the software clock routine), then timertype tells which timer.
     80  */
     81 /*ARGSUSED*/
     82 int
     83 tcp_usrreq(so, req, m, nam, control, p)
     84 	struct socket *so;
     85 	int req;
     86 	struct mbuf *m, *nam, *control;
     87 	struct proc *p;
     88 {
     89 	register struct inpcb *inp;
     90 	register struct tcpcb *tp = NULL;
     91 	int s;
     92 	int error = 0;
     93 	int ostate;
     94 
     95 	if (req == PRU_CONTROL)
     96 		return (in_control(so, (long)m, (caddr_t)nam,
     97 		    (struct ifnet *)control, p));
     98 
     99 	s = splsoftnet();
    100 	inp = sotoinpcb(so);
    101 	if (control && control->m_len) {
    102 		m_freem(control);
    103 		error = EINVAL;
    104 		goto release;
    105 	}
    106 	/*
    107 	 * When a TCP is attached to a socket, then there will be
    108 	 * a (struct inpcb) pointed at by the socket, and this
    109 	 * structure will point at a subsidary (struct tcpcb).
    110 	 */
    111 	if (inp == 0 && req != PRU_ATTACH) {
    112 		error = EINVAL;
    113 		goto release;
    114 	}
    115 	if (inp) {
    116 		tp = intotcpcb(inp);
    117 		/* WHAT IF TP IS 0? */
    118 #ifdef KPROF
    119 		tcp_acounts[tp->t_state][req]++;
    120 #endif
    121 		ostate = tp->t_state;
    122 	} else
    123 		ostate = 0;
    124 
    125 	switch (req) {
    126 
    127 	/*
    128 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
    129 	 * and an internet control block.
    130 	 */
    131 	case PRU_ATTACH:
    132 		if (inp != 0) {
    133 			error = EISCONN;
    134 			break;
    135 		}
    136 		error = tcp_attach(so);
    137 		if (error)
    138 			break;
    139 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
    140 			so->so_linger = TCP_LINGERTIME * hz;
    141 		tp = sototcpcb(so);
    142 		break;
    143 
    144 	/*
    145 	 * PRU_DETACH detaches the TCP protocol from the socket.
    146 	 * If the protocol state is non-embryonic, then can't
    147 	 * do this directly: have to initiate a PRU_DISCONNECT,
    148 	 * which may finish later; embryonic TCB's can just
    149 	 * be discarded here.
    150 	 */
    151 	case PRU_DETACH:
    152 		if (tp->t_state > TCPS_LISTEN)
    153 			tp = tcp_disconnect(tp);
    154 		else
    155 			tp = tcp_close(tp);
    156 		break;
    157 
    158 	/*
    159 	 * Give the socket an address.
    160 	 */
    161 	case PRU_BIND:
    162 		error = in_pcbbind(inp, nam, p);
    163 		break;
    164 
    165 	/*
    166 	 * Prepare to accept connections.
    167 	 */
    168 	case PRU_LISTEN:
    169 		if (inp->inp_lport == 0) {
    170 			error = in_pcbbind(inp, (struct mbuf *)0,
    171 			    (struct proc *)0);
    172 			if (error)
    173 				break;
    174 		}
    175 		tp->t_state = TCPS_LISTEN;
    176 		break;
    177 
    178 	/*
    179 	 * Initiate connection to peer.
    180 	 * Create a template for use in transmissions on this connection.
    181 	 * Enter SYN_SENT state, and mark socket as connecting.
    182 	 * Start keep-alive timer, and seed output sequence space.
    183 	 * Send initial segment on connection.
    184 	 */
    185 	case PRU_CONNECT:
    186 		if (inp->inp_lport == 0) {
    187 			error = in_pcbbind(inp, (struct mbuf *)0,
    188 			    (struct proc *)0);
    189 			if (error)
    190 				break;
    191 		}
    192 		error = in_pcbconnect(inp, nam);
    193 		if (error)
    194 			break;
    195 		tp->t_template = tcp_template(tp);
    196 		if (tp->t_template == 0) {
    197 			in_pcbdisconnect(inp);
    198 			error = ENOBUFS;
    199 			break;
    200 		}
    201 		/* Compute window scaling to request.  */
    202 		while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
    203 		    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
    204 			tp->request_r_scale++;
    205 		soisconnecting(so);
    206 		tcpstat.tcps_connattempt++;
    207 		tp->t_state = TCPS_SYN_SENT;
    208 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
    209 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
    210 		tcp_sendseqinit(tp);
    211 		error = tcp_output(tp);
    212 		break;
    213 
    214 	/*
    215 	 * Create a TCP connection between two sockets.
    216 	 */
    217 	case PRU_CONNECT2:
    218 		error = EOPNOTSUPP;
    219 		break;
    220 
    221 	/*
    222 	 * Initiate disconnect from peer.
    223 	 * If connection never passed embryonic stage, just drop;
    224 	 * else if don't need to let data drain, then can just drop anyways,
    225 	 * else have to begin TCP shutdown process: mark socket disconnecting,
    226 	 * drain unread data, state switch to reflect user close, and
    227 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
    228 	 * when peer sends FIN and acks ours.
    229 	 *
    230 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
    231 	 */
    232 	case PRU_DISCONNECT:
    233 		tp = tcp_disconnect(tp);
    234 		break;
    235 
    236 	/*
    237 	 * Accept a connection.  Essentially all the work is
    238 	 * done at higher levels; just return the address
    239 	 * of the peer, storing through addr.
    240 	 */
    241 	case PRU_ACCEPT:
    242 		in_setpeeraddr(inp, nam);
    243 		break;
    244 
    245 	/*
    246 	 * Mark the connection as being incapable of further output.
    247 	 */
    248 	case PRU_SHUTDOWN:
    249 		socantsendmore(so);
    250 		tp = tcp_usrclosed(tp);
    251 		if (tp)
    252 			error = tcp_output(tp);
    253 		break;
    254 
    255 	/*
    256 	 * After a receive, possibly send window update to peer.
    257 	 */
    258 	case PRU_RCVD:
    259 		(void) tcp_output(tp);
    260 		break;
    261 
    262 	/*
    263 	 * Do a send by putting data in output queue and updating urgent
    264 	 * marker if URG set.  Possibly send more data.
    265 	 */
    266 	case PRU_SEND:
    267 		sbappend(&so->so_snd, m);
    268 		error = tcp_output(tp);
    269 		break;
    270 
    271 	/*
    272 	 * Abort the TCP.
    273 	 */
    274 	case PRU_ABORT:
    275 		tp = tcp_drop(tp, ECONNABORTED);
    276 		break;
    277 
    278 	case PRU_SENSE:
    279 		/*
    280 		 * stat: don't bother with a blocksize.
    281 		 */
    282 		splx(s);
    283 		return (0);
    284 
    285 	case PRU_RCVOOB:
    286 		if ((so->so_oobmark == 0 &&
    287 		    (so->so_state & SS_RCVATMARK) == 0) ||
    288 		    so->so_options & SO_OOBINLINE ||
    289 		    tp->t_oobflags & TCPOOB_HADDATA) {
    290 			error = EINVAL;
    291 			break;
    292 		}
    293 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
    294 			error = EWOULDBLOCK;
    295 			break;
    296 		}
    297 		m->m_len = 1;
    298 		*mtod(m, caddr_t) = tp->t_iobc;
    299 		if (((long)nam & MSG_PEEK) == 0)
    300 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
    301 		break;
    302 
    303 	case PRU_SENDOOB:
    304 		if (sbspace(&so->so_snd) < -512) {
    305 			m_freem(m);
    306 			error = ENOBUFS;
    307 			break;
    308 		}
    309 		/*
    310 		 * According to RFC961 (Assigned Protocols),
    311 		 * the urgent pointer points to the last octet
    312 		 * of urgent data.  We continue, however,
    313 		 * to consider it to indicate the first octet
    314 		 * of data past the urgent section.
    315 		 * Otherwise, snd_up should be one lower.
    316 		 */
    317 		sbappend(&so->so_snd, m);
    318 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
    319 		tp->t_force = 1;
    320 		error = tcp_output(tp);
    321 		tp->t_force = 0;
    322 		break;
    323 
    324 	case PRU_SOCKADDR:
    325 		in_setsockaddr(inp, nam);
    326 		break;
    327 
    328 	case PRU_PEERADDR:
    329 		in_setpeeraddr(inp, nam);
    330 		break;
    331 
    332 	/*
    333 	 * TCP slow timer went off; going through this
    334 	 * routine for tracing's sake.
    335 	 */
    336 	case PRU_SLOWTIMO:
    337 		tp = tcp_timers(tp, (long)nam);
    338 		req |= (long)nam << 8;		/* for debug's sake */
    339 		break;
    340 
    341 	default:
    342 		panic("tcp_usrreq");
    343 	}
    344 	if (tp && (so->so_options & SO_DEBUG))
    345 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
    346 
    347 release:
    348 	splx(s);
    349 	return (error);
    350 }
    351 
    352 int
    353 tcp_ctloutput(op, so, level, optname, mp)
    354 	int op;
    355 	struct socket *so;
    356 	int level, optname;
    357 	struct mbuf **mp;
    358 {
    359 	int error = 0, s;
    360 	struct inpcb *inp;
    361 	register struct tcpcb *tp;
    362 	register struct mbuf *m;
    363 	register int i;
    364 
    365 	s = splsoftnet();
    366 	inp = sotoinpcb(so);
    367 	if (inp == NULL) {
    368 		splx(s);
    369 		if (op == PRCO_SETOPT && *mp)
    370 			(void) m_free(*mp);
    371 		return (ECONNRESET);
    372 	}
    373 	if (level != IPPROTO_TCP) {
    374 		error = ip_ctloutput(op, so, level, optname, mp);
    375 		splx(s);
    376 		return (error);
    377 	}
    378 	tp = intotcpcb(inp);
    379 
    380 	switch (op) {
    381 
    382 	case PRCO_SETOPT:
    383 		m = *mp;
    384 		switch (optname) {
    385 
    386 		case TCP_NODELAY:
    387 			if (m == NULL || m->m_len < sizeof (int))
    388 				error = EINVAL;
    389 			else if (*mtod(m, int *))
    390 				tp->t_flags |= TF_NODELAY;
    391 			else
    392 				tp->t_flags &= ~TF_NODELAY;
    393 			break;
    394 
    395 		case TCP_MAXSEG:
    396 			if (m && (i = *mtod(m, int *)) > 0 && i <= tp->t_maxseg)
    397 				tp->t_maxseg = i;
    398 			else
    399 				error = EINVAL;
    400 			break;
    401 
    402 		default:
    403 			error = ENOPROTOOPT;
    404 			break;
    405 		}
    406 		if (m)
    407 			(void) m_free(m);
    408 		break;
    409 
    410 	case PRCO_GETOPT:
    411 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
    412 		m->m_len = sizeof(int);
    413 
    414 		switch (optname) {
    415 		case TCP_NODELAY:
    416 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
    417 			break;
    418 		case TCP_MAXSEG:
    419 			*mtod(m, int *) = tp->t_maxseg;
    420 			break;
    421 		default:
    422 			error = ENOPROTOOPT;
    423 			break;
    424 		}
    425 		break;
    426 	}
    427 	splx(s);
    428 	return (error);
    429 }
    430 
    431 #ifndef TCP_SENDSPACE
    432 #define	TCP_SENDSPACE	1024*16;
    433 #endif
    434 u_long	tcp_sendspace = TCP_SENDSPACE;
    435 #ifndef TCP_RECVSPACE
    436 #define	TCP_RECVSPACE	1024*16;
    437 #endif
    438 u_long	tcp_recvspace = TCP_RECVSPACE;
    439 
    440 /*
    441  * Attach TCP protocol to socket, allocating
    442  * internet protocol control block, tcp control block,
    443  * bufer space, and entering LISTEN state if to accept connections.
    444  */
    445 int
    446 tcp_attach(so)
    447 	struct socket *so;
    448 {
    449 	register struct tcpcb *tp;
    450 	struct inpcb *inp;
    451 	int error;
    452 
    453 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    454 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
    455 		if (error)
    456 			return (error);
    457 	}
    458 	error = in_pcballoc(so, &tcbtable);
    459 	if (error)
    460 		return (error);
    461 	inp = sotoinpcb(so);
    462 	tp = tcp_newtcpcb(inp);
    463 	if (tp == 0) {
    464 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
    465 
    466 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
    467 		in_pcbdetach(inp);
    468 		so->so_state |= nofd;
    469 		return (ENOBUFS);
    470 	}
    471 	tp->t_state = TCPS_CLOSED;
    472 	return (0);
    473 }
    474 
    475 /*
    476  * Initiate (or continue) disconnect.
    477  * If embryonic state, just send reset (once).
    478  * If in ``let data drain'' option and linger null, just drop.
    479  * Otherwise (hard), mark socket disconnecting and drop
    480  * current input data; switch states based on user close, and
    481  * send segment to peer (with FIN).
    482  */
    483 struct tcpcb *
    484 tcp_disconnect(tp)
    485 	register struct tcpcb *tp;
    486 {
    487 	struct socket *so = tp->t_inpcb->inp_socket;
    488 
    489 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    490 		tp = tcp_close(tp);
    491 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
    492 		tp = tcp_drop(tp, 0);
    493 	else {
    494 		soisdisconnecting(so);
    495 		sbflush(&so->so_rcv);
    496 		tp = tcp_usrclosed(tp);
    497 		if (tp)
    498 			(void) tcp_output(tp);
    499 	}
    500 	return (tp);
    501 }
    502 
    503 /*
    504  * User issued close, and wish to trail through shutdown states:
    505  * if never received SYN, just forget it.  If got a SYN from peer,
    506  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
    507  * If already got a FIN from peer, then almost done; go to LAST_ACK
    508  * state.  In all other cases, have already sent FIN to peer (e.g.
    509  * after PRU_SHUTDOWN), and just have to play tedious game waiting
    510  * for peer to send FIN or not respond to keep-alives, etc.
    511  * We can let the user exit from the close as soon as the FIN is acked.
    512  */
    513 struct tcpcb *
    514 tcp_usrclosed(tp)
    515 	register struct tcpcb *tp;
    516 {
    517 
    518 	switch (tp->t_state) {
    519 
    520 	case TCPS_CLOSED:
    521 	case TCPS_LISTEN:
    522 	case TCPS_SYN_SENT:
    523 		tp->t_state = TCPS_CLOSED;
    524 		tp = tcp_close(tp);
    525 		break;
    526 
    527 	case TCPS_SYN_RECEIVED:
    528 	case TCPS_ESTABLISHED:
    529 		tp->t_state = TCPS_FIN_WAIT_1;
    530 		break;
    531 
    532 	case TCPS_CLOSE_WAIT:
    533 		tp->t_state = TCPS_LAST_ACK;
    534 		break;
    535 	}
    536 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
    537 		soisdisconnected(tp->t_inpcb->inp_socket);
    538 		/*
    539 		 * If we are in FIN_WAIT_2, we arrived here because the
    540 		 * application did a shutdown of the send side.  Like the
    541 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
    542 		 * a full close, we start a timer to make sure sockets are
    543 		 * not left in FIN_WAIT_2 forever.
    544 		 */
    545 		if (tp->t_state == TCPS_FIN_WAIT_2)
    546 			tp->t_timer[TCPT_2MSL] = tcp_maxidle;
    547 	}
    548 	return (tp);
    549 }
    550 
    551 /*
    552  * Sysctl for tcp variables.
    553  */
    554 int
    555 tcp_sysctl(name, namelen, oldp, oldlenp, newp, newlen)
    556 	int *name;
    557 	u_int namelen;
    558 	void *oldp;
    559 	size_t *oldlenp;
    560 	void *newp;
    561 	size_t newlen;
    562 {
    563 
    564 	/* All sysctl names at this level are terminal. */
    565 	if (namelen != 1)
    566 		return (ENOTDIR);
    567 
    568 	switch (name[0]) {
    569 	case TCPCTL_RFC1323:
    570 		return (sysctl_int(oldp, oldlenp, newp, newlen,
    571 		    &tcp_do_rfc1323));
    572 
    573 	default:
    574 		return (ENOPROTOOPT);
    575 	}
    576 	/* NOTREACHED */
    577 }
    578