Home | History | Annotate | Line # | Download | only in netinet
tcp_usrreq.c revision 1.24 
      1 /*	$NetBSD: tcp_usrreq.c,v 1.24 1997/06/12 18:41:14 kleink 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 #ifdef DIAGNOSTIC
    102 	if (req != PRU_SEND && req != PRU_SENDOOB && control)
    103 		panic("tcp_usrreq: unexpected control mbuf");
    104 #endif
    105 	/*
    106 	 * When a TCP is attached to a socket, then there will be
    107 	 * a (struct inpcb) pointed at by the socket, and this
    108 	 * structure will point at a subsidary (struct tcpcb).
    109 	 */
    110 	if (inp == 0 && req != PRU_ATTACH) {
    111 		error = EINVAL;
    112 		goto release;
    113 	}
    114 	if (inp) {
    115 		tp = intotcpcb(inp);
    116 		/* WHAT IF TP IS 0? */
    117 #ifdef KPROF
    118 		tcp_acounts[tp->t_state][req]++;
    119 #endif
    120 		ostate = tp->t_state;
    121 	} else
    122 		ostate = 0;
    123 
    124 	switch (req) {
    125 
    126 	/*
    127 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
    128 	 * and an internet control block.
    129 	 */
    130 	case PRU_ATTACH:
    131 		if (inp != 0) {
    132 			error = EISCONN;
    133 			break;
    134 		}
    135 		error = tcp_attach(so);
    136 		if (error)
    137 			break;
    138 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
    139 			so->so_linger = TCP_LINGERTIME * hz;
    140 		tp = sototcpcb(so);
    141 		break;
    142 
    143 	/*
    144 	 * PRU_DETACH detaches the TCP protocol from the socket.
    145 	 */
    146 	case PRU_DETACH:
    147 		tp = tcp_disconnect(tp);
    148 		break;
    149 
    150 	/*
    151 	 * Give the socket an address.
    152 	 */
    153 	case PRU_BIND:
    154 		error = in_pcbbind(inp, nam, p);
    155 		break;
    156 
    157 	/*
    158 	 * Prepare to accept connections.
    159 	 */
    160 	case PRU_LISTEN:
    161 		if (inp->inp_lport == 0) {
    162 			error = in_pcbbind(inp, (struct mbuf *)0,
    163 			    (struct proc *)0);
    164 			if (error)
    165 				break;
    166 		}
    167 		tp->t_state = TCPS_LISTEN;
    168 		break;
    169 
    170 	/*
    171 	 * Initiate connection to peer.
    172 	 * Create a template for use in transmissions on this connection.
    173 	 * Enter SYN_SENT state, and mark socket as connecting.
    174 	 * Start keep-alive timer, and seed output sequence space.
    175 	 * Send initial segment on connection.
    176 	 */
    177 	case PRU_CONNECT:
    178 		if (inp->inp_lport == 0) {
    179 			error = in_pcbbind(inp, (struct mbuf *)0,
    180 			    (struct proc *)0);
    181 			if (error)
    182 				break;
    183 		}
    184 		error = in_pcbconnect(inp, nam);
    185 		if (error)
    186 			break;
    187 		tp->t_template = tcp_template(tp);
    188 		if (tp->t_template == 0) {
    189 			in_pcbdisconnect(inp);
    190 			error = ENOBUFS;
    191 			break;
    192 		}
    193 		/* Compute window scaling to request.  */
    194 		while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
    195 		    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
    196 			tp->request_r_scale++;
    197 		soisconnecting(so);
    198 		tcpstat.tcps_connattempt++;
    199 		tp->t_state = TCPS_SYN_SENT;
    200 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
    201 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
    202 		tcp_sendseqinit(tp);
    203 		error = tcp_output(tp);
    204 		break;
    205 
    206 	/*
    207 	 * Create a TCP connection between two sockets.
    208 	 */
    209 	case PRU_CONNECT2:
    210 		error = EOPNOTSUPP;
    211 		break;
    212 
    213 	/*
    214 	 * Initiate disconnect from peer.
    215 	 * If connection never passed embryonic stage, just drop;
    216 	 * else if don't need to let data drain, then can just drop anyways,
    217 	 * else have to begin TCP shutdown process: mark socket disconnecting,
    218 	 * drain unread data, state switch to reflect user close, and
    219 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
    220 	 * when peer sends FIN and acks ours.
    221 	 *
    222 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
    223 	 */
    224 	case PRU_DISCONNECT:
    225 		tp = tcp_disconnect(tp);
    226 		break;
    227 
    228 	/*
    229 	 * Accept a connection.  Essentially all the work is
    230 	 * done at higher levels; just return the address
    231 	 * of the peer, storing through addr.
    232 	 */
    233 	case PRU_ACCEPT:
    234 		in_setpeeraddr(inp, nam);
    235 		break;
    236 
    237 	/*
    238 	 * Mark the connection as being incapable of further output.
    239 	 */
    240 	case PRU_SHUTDOWN:
    241 		socantsendmore(so);
    242 		tp = tcp_usrclosed(tp);
    243 		if (tp)
    244 			error = tcp_output(tp);
    245 		break;
    246 
    247 	/*
    248 	 * After a receive, possibly send window update to peer.
    249 	 */
    250 	case PRU_RCVD:
    251 		(void) tcp_output(tp);
    252 		break;
    253 
    254 	/*
    255 	 * Do a send by putting data in output queue and updating urgent
    256 	 * marker if URG set.  Possibly send more data.
    257 	 */
    258 	case PRU_SEND:
    259 		if (control && control->m_len) {
    260 			m_freem(control);
    261 			m_freem(m);
    262 			error = EINVAL;
    263 			break;
    264 		}
    265 		sbappend(&so->so_snd, m);
    266 		error = tcp_output(tp);
    267 		break;
    268 
    269 	/*
    270 	 * Abort the TCP.
    271 	 */
    272 	case PRU_ABORT:
    273 		tp = tcp_drop(tp, ECONNABORTED);
    274 		break;
    275 
    276 	case PRU_SENSE:
    277 		/*
    278 		 * stat: don't bother with a blocksize.
    279 		 */
    280 		splx(s);
    281 		return (0);
    282 
    283 	case PRU_RCVOOB:
    284 		if (control && control->m_len) {
    285 			m_freem(control);
    286 			m_freem(m);
    287 			error = EINVAL;
    288 			break;
    289 		}
    290 		if ((so->so_oobmark == 0 &&
    291 		    (so->so_state & SS_RCVATMARK) == 0) ||
    292 		    so->so_options & SO_OOBINLINE ||
    293 		    tp->t_oobflags & TCPOOB_HADDATA) {
    294 			error = EINVAL;
    295 			break;
    296 		}
    297 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
    298 			error = EWOULDBLOCK;
    299 			break;
    300 		}
    301 		m->m_len = 1;
    302 		*mtod(m, caddr_t) = tp->t_iobc;
    303 		if (((long)nam & MSG_PEEK) == 0)
    304 			tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
    305 		break;
    306 
    307 	case PRU_SENDOOB:
    308 		if (sbspace(&so->so_snd) < -512) {
    309 			m_freem(m);
    310 			error = ENOBUFS;
    311 			break;
    312 		}
    313 		/*
    314 		 * According to RFC961 (Assigned Protocols),
    315 		 * the urgent pointer points to the last octet
    316 		 * of urgent data.  We continue, however,
    317 		 * to consider it to indicate the first octet
    318 		 * of data past the urgent section.
    319 		 * Otherwise, snd_up should be one lower.
    320 		 */
    321 		sbappend(&so->so_snd, m);
    322 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
    323 		tp->t_force = 1;
    324 		error = tcp_output(tp);
    325 		tp->t_force = 0;
    326 		break;
    327 
    328 	case PRU_SOCKADDR:
    329 		in_setsockaddr(inp, nam);
    330 		break;
    331 
    332 	case PRU_PEERADDR:
    333 		in_setpeeraddr(inp, nam);
    334 		break;
    335 
    336 	/*
    337 	 * TCP slow timer went off; going through this
    338 	 * routine for tracing's sake.
    339 	 */
    340 	case PRU_SLOWTIMO:
    341 		tp = tcp_timers(tp, (long)nam);
    342 		req |= (long)nam << 8;		/* for debug's sake */
    343 		break;
    344 
    345 	default:
    346 		panic("tcp_usrreq");
    347 	}
    348 	if (tp && (so->so_options & SO_DEBUG))
    349 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
    350 
    351 release:
    352 	splx(s);
    353 	return (error);
    354 }
    355 
    356 int
    357 tcp_ctloutput(op, so, level, optname, mp)
    358 	int op;
    359 	struct socket *so;
    360 	int level, optname;
    361 	struct mbuf **mp;
    362 {
    363 	int error = 0, s;
    364 	struct inpcb *inp;
    365 	register struct tcpcb *tp;
    366 	register struct mbuf *m;
    367 	register int i;
    368 
    369 	s = splsoftnet();
    370 	inp = sotoinpcb(so);
    371 	if (inp == NULL) {
    372 		splx(s);
    373 		if (op == PRCO_SETOPT && *mp)
    374 			(void) m_free(*mp);
    375 		return (ECONNRESET);
    376 	}
    377 	if (level != IPPROTO_TCP) {
    378 		error = ip_ctloutput(op, so, level, optname, mp);
    379 		splx(s);
    380 		return (error);
    381 	}
    382 	tp = intotcpcb(inp);
    383 
    384 	switch (op) {
    385 
    386 	case PRCO_SETOPT:
    387 		m = *mp;
    388 		switch (optname) {
    389 
    390 		case TCP_NODELAY:
    391 			if (m == NULL || m->m_len < sizeof (int))
    392 				error = EINVAL;
    393 			else if (*mtod(m, int *))
    394 				tp->t_flags |= TF_NODELAY;
    395 			else
    396 				tp->t_flags &= ~TF_NODELAY;
    397 			break;
    398 
    399 		case TCP_MAXSEG:
    400 			if (m && (i = *mtod(m, int *)) > 0 && i <= tp->t_maxseg)
    401 				tp->t_maxseg = i;
    402 			else
    403 				error = EINVAL;
    404 			break;
    405 
    406 		default:
    407 			error = ENOPROTOOPT;
    408 			break;
    409 		}
    410 		if (m)
    411 			(void) m_free(m);
    412 		break;
    413 
    414 	case PRCO_GETOPT:
    415 		*mp = m = m_get(M_WAIT, MT_SOOPTS);
    416 		m->m_len = sizeof(int);
    417 
    418 		switch (optname) {
    419 		case TCP_NODELAY:
    420 			*mtod(m, int *) = tp->t_flags & TF_NODELAY;
    421 			break;
    422 		case TCP_MAXSEG:
    423 			*mtod(m, int *) = tp->t_maxseg;
    424 			break;
    425 		default:
    426 			error = ENOPROTOOPT;
    427 			break;
    428 		}
    429 		break;
    430 	}
    431 	splx(s);
    432 	return (error);
    433 }
    434 
    435 #ifndef TCP_SENDSPACE
    436 #define	TCP_SENDSPACE	1024*16;
    437 #endif
    438 u_long	tcp_sendspace = TCP_SENDSPACE;
    439 #ifndef TCP_RECVSPACE
    440 #define	TCP_RECVSPACE	1024*16;
    441 #endif
    442 u_long	tcp_recvspace = TCP_RECVSPACE;
    443 
    444 /*
    445  * Attach TCP protocol to socket, allocating
    446  * internet protocol control block, tcp control block,
    447  * bufer space, and entering LISTEN state if to accept connections.
    448  */
    449 int
    450 tcp_attach(so)
    451 	struct socket *so;
    452 {
    453 	register struct tcpcb *tp;
    454 	struct inpcb *inp;
    455 	int error;
    456 
    457 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    458 		error = soreserve(so, tcp_sendspace, tcp_recvspace);
    459 		if (error)
    460 			return (error);
    461 	}
    462 	error = in_pcballoc(so, &tcbtable);
    463 	if (error)
    464 		return (error);
    465 	inp = sotoinpcb(so);
    466 	tp = tcp_newtcpcb(inp);
    467 	if (tp == 0) {
    468 		int nofd = so->so_state & SS_NOFDREF;	/* XXX */
    469 
    470 		so->so_state &= ~SS_NOFDREF;	/* don't free the socket yet */
    471 		in_pcbdetach(inp);
    472 		so->so_state |= nofd;
    473 		return (ENOBUFS);
    474 	}
    475 	tp->t_state = TCPS_CLOSED;
    476 	return (0);
    477 }
    478 
    479 /*
    480  * Initiate (or continue) disconnect.
    481  * If embryonic state, just send reset (once).
    482  * If in ``let data drain'' option and linger null, just drop.
    483  * Otherwise (hard), mark socket disconnecting and drop
    484  * current input data; switch states based on user close, and
    485  * send segment to peer (with FIN).
    486  */
    487 struct tcpcb *
    488 tcp_disconnect(tp)
    489 	register struct tcpcb *tp;
    490 {
    491 	struct socket *so = tp->t_inpcb->inp_socket;
    492 
    493 	if (TCPS_HAVEESTABLISHED(tp->t_state) == 0)
    494 		tp = tcp_close(tp);
    495 	else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
    496 		tp = tcp_drop(tp, 0);
    497 	else {
    498 		soisdisconnecting(so);
    499 		sbflush(&so->so_rcv);
    500 		tp = tcp_usrclosed(tp);
    501 		if (tp)
    502 			(void) tcp_output(tp);
    503 	}
    504 	return (tp);
    505 }
    506 
    507 /*
    508  * User issued close, and wish to trail through shutdown states:
    509  * if never received SYN, just forget it.  If got a SYN from peer,
    510  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
    511  * If already got a FIN from peer, then almost done; go to LAST_ACK
    512  * state.  In all other cases, have already sent FIN to peer (e.g.
    513  * after PRU_SHUTDOWN), and just have to play tedious game waiting
    514  * for peer to send FIN or not respond to keep-alives, etc.
    515  * We can let the user exit from the close as soon as the FIN is acked.
    516  */
    517 struct tcpcb *
    518 tcp_usrclosed(tp)
    519 	register struct tcpcb *tp;
    520 {
    521 
    522 	switch (tp->t_state) {
    523 
    524 	case TCPS_CLOSED:
    525 	case TCPS_LISTEN:
    526 	case TCPS_SYN_SENT:
    527 		tp->t_state = TCPS_CLOSED;
    528 		tp = tcp_close(tp);
    529 		break;
    530 
    531 	case TCPS_SYN_RECEIVED:
    532 	case TCPS_ESTABLISHED:
    533 		tp->t_state = TCPS_FIN_WAIT_1;
    534 		break;
    535 
    536 	case TCPS_CLOSE_WAIT:
    537 		tp->t_state = TCPS_LAST_ACK;
    538 		break;
    539 	}
    540 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
    541 		soisdisconnected(tp->t_inpcb->inp_socket);
    542 		/*
    543 		 * If we are in FIN_WAIT_2, we arrived here because the
    544 		 * application did a shutdown of the send side.  Like the
    545 		 * case of a transition from FIN_WAIT_1 to FIN_WAIT_2 after
    546 		 * a full close, we start a timer to make sure sockets are
    547 		 * not left in FIN_WAIT_2 forever.
    548 		 */
    549 		if (tp->t_state == TCPS_FIN_WAIT_2)
    550 			tp->t_timer[TCPT_2MSL] = tcp_maxidle;
    551 	}
    552 	return (tp);
    553 }
    554 
    555 /*
    556  * Sysctl for tcp variables.
    557  */
    558 int
    559 tcp_sysctl(name, namelen, oldp, oldlenp, newp, newlen)
    560 	int *name;
    561 	u_int namelen;
    562 	void *oldp;
    563 	size_t *oldlenp;
    564 	void *newp;
    565 	size_t newlen;
    566 {
    567 
    568 	/* All sysctl names at this level are terminal. */
    569 	if (namelen != 1)
    570 		return (ENOTDIR);
    571 
    572 	switch (name[0]) {
    573 	case TCPCTL_RFC1323:
    574 		return (sysctl_int(oldp, oldlenp, newp, newlen,
    575 		    &tcp_do_rfc1323));
    576 
    577 	default:
    578 		return (ENOPROTOOPT);
    579 	}
    580 	/* NOTREACHED */
    581 }
    582