Home | History | Annotate | Line # | Download | only in netinet
in_pcb.c revision 1.22
      1 /*	$NetBSD: in_pcb.c,v 1.22 1995/08/12 23:59:34 mycroft Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1991, 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  *	@(#)in_pcb.c	8.2 (Berkeley) 1/4/94
     36  */
     37 
     38 #include <sys/param.h>
     39 #include <sys/systm.h>
     40 #include <sys/malloc.h>
     41 #include <sys/mbuf.h>
     42 #include <sys/protosw.h>
     43 #include <sys/socket.h>
     44 #include <sys/socketvar.h>
     45 #include <sys/ioctl.h>
     46 #include <sys/errno.h>
     47 #include <sys/time.h>
     48 #include <sys/proc.h>
     49 
     50 #include <net/if.h>
     51 #include <net/route.h>
     52 
     53 #include <netinet/in.h>
     54 #include <netinet/in_systm.h>
     55 #include <netinet/ip.h>
     56 #include <netinet/in_pcb.h>
     57 #include <netinet/in_var.h>
     58 #include <netinet/ip_var.h>
     59 
     60 struct	in_addr zeroin_addr;
     61 
     62 void
     63 in_pcbinit(table)
     64 	struct inpcbtable *table;
     65 {
     66 
     67 	CIRCLEQ_INIT(&table->inpt_queue);
     68 	table->inpt_lastport = 0;
     69 }
     70 
     71 int
     72 in_pcballoc(so, table)
     73 	struct socket *so;
     74 	struct inpcbtable *table;
     75 {
     76 	register struct inpcb *inp;
     77 
     78 	MALLOC(inp, struct inpcb *, sizeof(*inp), M_PCB, M_WAITOK);
     79 	if (inp == NULL)
     80 		return (ENOBUFS);
     81 	bzero((caddr_t)inp, sizeof(*inp));
     82 	inp->inp_table = table;
     83 	inp->inp_socket = so;
     84 	CIRCLEQ_INSERT_HEAD(&table->inpt_queue, inp, inp_queue);
     85 	so->so_pcb = (caddr_t)inp;
     86 	return (0);
     87 }
     88 
     89 int
     90 in_pcbbind(inp, nam)
     91 	register struct inpcb *inp;
     92 	struct mbuf *nam;
     93 {
     94 	register struct socket *so = inp->inp_socket;
     95 	register struct inpcbtable *table = inp->inp_table;
     96 	register struct sockaddr_in *sin;
     97 	struct proc *p = curproc;		/* XXX */
     98 	u_int16_t lport = 0;
     99 	int wild = 0, reuseport = (so->so_options & SO_REUSEPORT);
    100 	int error;
    101 
    102 	if (in_ifaddr.tqh_first == 0)
    103 		return (EADDRNOTAVAIL);
    104 	if (inp->inp_lport || inp->inp_laddr.s_addr != INADDR_ANY)
    105 		return (EINVAL);
    106 	if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0 &&
    107 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0 ||
    108 	     (so->so_options & SO_ACCEPTCONN) == 0))
    109 		wild = INPLOOKUP_WILDCARD;
    110 	if (nam) {
    111 		sin = mtod(nam, struct sockaddr_in *);
    112 		if (nam->m_len != sizeof (*sin))
    113 			return (EINVAL);
    114 #ifdef notdef
    115 		/*
    116 		 * We should check the family, but old programs
    117 		 * incorrectly fail to initialize it.
    118 		 */
    119 		if (sin->sin_family != AF_INET)
    120 			return (EAFNOSUPPORT);
    121 #endif
    122 		lport = sin->sin_port;
    123 		if (IN_MULTICAST(sin->sin_addr.s_addr)) {
    124 			/*
    125 			 * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
    126 			 * allow complete duplication of binding if
    127 			 * SO_REUSEPORT is set, or if SO_REUSEADDR is set
    128 			 * and a multicast address is bound on both
    129 			 * new and duplicated sockets.
    130 			 */
    131 			if (so->so_options & SO_REUSEADDR)
    132 				reuseport = SO_REUSEADDR|SO_REUSEPORT;
    133 		} else if (sin->sin_addr.s_addr != INADDR_ANY) {
    134 			sin->sin_port = 0;		/* yech... */
    135 			if (ifa_ifwithaddr(sintosa(sin)) == 0)
    136 				return (EADDRNOTAVAIL);
    137 		}
    138 		if (lport) {
    139 			struct inpcb *t;
    140 
    141 			/* GROSS */
    142 			if (ntohs(lport) < IPPORT_RESERVED &&
    143 			    (error = suser(p->p_ucred, &p->p_acflag)))
    144 				return (EACCES);
    145 			t = in_pcblookup(table, zeroin_addr, 0,
    146 			    sin->sin_addr, lport, wild);
    147 			if (t && (reuseport & t->inp_socket->so_options) == 0)
    148 				return (EADDRINUSE);
    149 		}
    150 		inp->inp_laddr = sin->sin_addr;
    151 	}
    152 	if (lport == 0)
    153 		do {
    154 			if (table->inpt_lastport++ < IPPORT_RESERVED ||
    155 			    table->inpt_lastport > IPPORT_USERRESERVED)
    156 				table->inpt_lastport = IPPORT_RESERVED;
    157 			lport = htons(table->inpt_lastport);
    158 		} while (in_pcblookup(table,
    159 			    zeroin_addr, 0, inp->inp_laddr, lport, wild));
    160 	inp->inp_lport = lport;
    161 	return (0);
    162 }
    163 
    164 /*
    165  * Connect from a socket to a specified address.
    166  * Both address and port must be specified in argument sin.
    167  * If don't have a local address for this socket yet,
    168  * then pick one.
    169  */
    170 int
    171 in_pcbconnect(inp, nam)
    172 	register struct inpcb *inp;
    173 	struct mbuf *nam;
    174 {
    175 	struct in_ifaddr *ia;
    176 	struct sockaddr_in *ifaddr;
    177 	register struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
    178 
    179 	if (nam->m_len != sizeof (*sin))
    180 		return (EINVAL);
    181 	if (sin->sin_family != AF_INET)
    182 		return (EAFNOSUPPORT);
    183 	if (sin->sin_port == 0)
    184 		return (EADDRNOTAVAIL);
    185 	if (in_ifaddr.tqh_first != 0) {
    186 		/*
    187 		 * If the destination address is INADDR_ANY,
    188 		 * use the primary local address.
    189 		 * If the supplied address is INADDR_BROADCAST,
    190 		 * and the primary interface supports broadcast,
    191 		 * choose the broadcast address for that interface.
    192 		 */
    193 		if (sin->sin_addr.s_addr == INADDR_ANY)
    194 			sin->sin_addr = in_ifaddr.tqh_first->ia_addr.sin_addr;
    195 		else if (sin->sin_addr.s_addr == INADDR_BROADCAST &&
    196 		  (in_ifaddr.tqh_first->ia_ifp->if_flags & IFF_BROADCAST))
    197 			sin->sin_addr = in_ifaddr.tqh_first->ia_broadaddr.sin_addr;
    198 	}
    199 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
    200 		register struct route *ro;
    201 
    202 		ia = (struct in_ifaddr *)0;
    203 		/*
    204 		 * If route is known or can be allocated now,
    205 		 * our src addr is taken from the i/f, else punt.
    206 		 */
    207 		ro = &inp->inp_route;
    208 		if (ro->ro_rt &&
    209 		    (satosin(&ro->ro_dst)->sin_addr.s_addr !=
    210 			sin->sin_addr.s_addr ||
    211 		    inp->inp_socket->so_options & SO_DONTROUTE)) {
    212 			RTFREE(ro->ro_rt);
    213 			ro->ro_rt = (struct rtentry *)0;
    214 		}
    215 		if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0 && /*XXX*/
    216 		    (ro->ro_rt == (struct rtentry *)0 ||
    217 		    ro->ro_rt->rt_ifp == (struct ifnet *)0)) {
    218 			/* No route yet, so try to acquire one */
    219 			ro->ro_dst.sa_family = AF_INET;
    220 			ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
    221 			satosin(&ro->ro_dst)->sin_addr = sin->sin_addr;
    222 			rtalloc(ro);
    223 		}
    224 		/*
    225 		 * If we found a route, use the address
    226 		 * corresponding to the outgoing interface
    227 		 * unless it is the loopback (in case a route
    228 		 * to our address on another net goes to loopback).
    229 		 */
    230 		if (ro->ro_rt && !(ro->ro_rt->rt_ifp->if_flags & IFF_LOOPBACK))
    231 			ia = ifatoia(ro->ro_rt->rt_ifa);
    232 		if (ia == 0) {
    233 			u_int16_t fport = sin->sin_port;
    234 
    235 			sin->sin_port = 0;
    236 			ia = ifatoia(ifa_ifwithdstaddr(sintosa(sin)));
    237 			if (ia == 0)
    238 				ia = ifatoia(ifa_ifwithnet(sintosa(sin)));
    239 			sin->sin_port = fport;
    240 			if (ia == 0)
    241 				ia = in_ifaddr.tqh_first;
    242 			if (ia == 0)
    243 				return (EADDRNOTAVAIL);
    244 		}
    245 		/*
    246 		 * If the destination address is multicast and an outgoing
    247 		 * interface has been set as a multicast option, use the
    248 		 * address of that interface as our source address.
    249 		 */
    250 		if (IN_MULTICAST(sin->sin_addr.s_addr) &&
    251 		    inp->inp_moptions != NULL) {
    252 			struct ip_moptions *imo;
    253 			struct ifnet *ifp;
    254 
    255 			imo = inp->inp_moptions;
    256 			if (imo->imo_multicast_ifp != NULL) {
    257 				ifp = imo->imo_multicast_ifp;
    258 				for (ia = in_ifaddr.tqh_first; ia != 0;
    259 				    ia = ia->ia_list.tqe_next)
    260 					if (ia->ia_ifp == ifp)
    261 						break;
    262 				if (ia == 0)
    263 					return (EADDRNOTAVAIL);
    264 			}
    265 		}
    266 		ifaddr = satosin(&ia->ia_addr);
    267 	}
    268 	if (in_pcblookup(inp->inp_table,
    269 	    sin->sin_addr,
    270 	    sin->sin_port,
    271 	    inp->inp_laddr.s_addr ? inp->inp_laddr : ifaddr->sin_addr,
    272 	    inp->inp_lport,
    273 	    0))
    274 		return (EADDRINUSE);
    275 	if (inp->inp_laddr.s_addr == INADDR_ANY) {
    276 		if (inp->inp_lport == 0)
    277 			(void)in_pcbbind(inp, (struct mbuf *)0);
    278 		inp->inp_laddr = ifaddr->sin_addr;
    279 	}
    280 	inp->inp_faddr = sin->sin_addr;
    281 	inp->inp_fport = sin->sin_port;
    282 	return (0);
    283 }
    284 
    285 int
    286 in_pcbdisconnect(inp)
    287 	struct inpcb *inp;
    288 {
    289 
    290 	inp->inp_faddr.s_addr = INADDR_ANY;
    291 	inp->inp_fport = 0;
    292 	if (inp->inp_socket->so_state & SS_NOFDREF)
    293 		in_pcbdetach(inp);
    294 }
    295 
    296 int
    297 in_pcbdetach(inp)
    298 	struct inpcb *inp;
    299 {
    300 	struct socket *so = inp->inp_socket;
    301 
    302 	so->so_pcb = 0;
    303 	sofree(so);
    304 	if (inp->inp_options)
    305 		(void)m_free(inp->inp_options);
    306 	if (inp->inp_route.ro_rt)
    307 		rtfree(inp->inp_route.ro_rt);
    308 	ip_freemoptions(inp->inp_moptions);
    309 	CIRCLEQ_REMOVE(&inp->inp_table->inpt_queue, inp, inp_queue);
    310 	FREE(inp, M_PCB);
    311 }
    312 
    313 int
    314 in_setsockaddr(inp, nam)
    315 	register struct inpcb *inp;
    316 	struct mbuf *nam;
    317 {
    318 	register struct sockaddr_in *sin;
    319 
    320 	nam->m_len = sizeof (*sin);
    321 	sin = mtod(nam, struct sockaddr_in *);
    322 	bzero((caddr_t)sin, sizeof (*sin));
    323 	sin->sin_family = AF_INET;
    324 	sin->sin_len = sizeof(*sin);
    325 	sin->sin_port = inp->inp_lport;
    326 	sin->sin_addr = inp->inp_laddr;
    327 }
    328 
    329 int
    330 in_setpeeraddr(inp, nam)
    331 	struct inpcb *inp;
    332 	struct mbuf *nam;
    333 {
    334 	register struct sockaddr_in *sin;
    335 
    336 	nam->m_len = sizeof (*sin);
    337 	sin = mtod(nam, struct sockaddr_in *);
    338 	bzero((caddr_t)sin, sizeof (*sin));
    339 	sin->sin_family = AF_INET;
    340 	sin->sin_len = sizeof(*sin);
    341 	sin->sin_port = inp->inp_fport;
    342 	sin->sin_addr = inp->inp_faddr;
    343 }
    344 
    345 /*
    346  * Pass some notification to all connections of a protocol
    347  * associated with address dst.  The local address and/or port numbers
    348  * may be specified to limit the search.  The "usual action" will be
    349  * taken, depending on the ctlinput cmd.  The caller must filter any
    350  * cmds that are uninteresting (e.g., no error in the map).
    351  * Call the protocol specific routine (if any) to report
    352  * any errors for each matching socket.
    353  *
    354  * Must be called at splsoftnet.
    355  */
    356 void
    357 in_pcbnotify(table, dst, fport_arg, laddr, lport_arg, errno, notify)
    358 	struct inpcbtable *table;
    359 	struct sockaddr *dst;
    360 	u_int fport_arg, lport_arg;
    361 	struct in_addr laddr;
    362 	int errno;
    363 	void (*notify) __P((struct inpcb *, int));
    364 {
    365 	register struct inpcb *inp, *oinp;
    366 	struct in_addr faddr;
    367 	u_int16_t fport = fport_arg, lport = lport_arg;
    368 
    369 	if (dst->sa_family != AF_INET)
    370 		return;
    371 	faddr = satosin(dst)->sin_addr;
    372 	if (faddr.s_addr == INADDR_ANY)
    373 		return;
    374 
    375 	for (inp = table->inpt_queue.cqh_first;
    376 	    inp != (struct inpcb *)&table->inpt_queue;) {
    377 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
    378 		    inp->inp_socket == 0 ||
    379 		    inp->inp_fport != fport ||
    380 		    inp->inp_lport != lport ||
    381 		    inp->inp_laddr.s_addr != laddr.s_addr) {
    382 			inp = inp->inp_queue.cqe_next;
    383 			continue;
    384 		}
    385 		oinp = inp;
    386 		inp = inp->inp_queue.cqe_next;
    387 		if (notify)
    388 			(*notify)(oinp, errno);
    389 	}
    390 }
    391 
    392 void
    393 in_pcbnotifyall(table, dst, errno, notify)
    394 	struct inpcbtable *table;
    395 	struct sockaddr *dst;
    396 	int errno;
    397 	void (*notify) __P((struct inpcb *, int));
    398 {
    399 	register struct inpcb *inp, *oinp;
    400 	struct in_addr faddr;
    401 
    402 	if (dst->sa_family != AF_INET)
    403 		return;
    404 	faddr = satosin(dst)->sin_addr;
    405 	if (faddr.s_addr == INADDR_ANY)
    406 		return;
    407 
    408 	for (inp = table->inpt_queue.cqh_first;
    409 	    inp != (struct inpcb *)&table->inpt_queue;) {
    410 		if (inp->inp_faddr.s_addr != faddr.s_addr ||
    411 		    inp->inp_socket == 0) {
    412 			inp = inp->inp_queue.cqe_next;
    413 			continue;
    414 		}
    415 		oinp = inp;
    416 		inp = inp->inp_queue.cqe_next;
    417 		if (notify)
    418 			(*notify)(oinp, errno);
    419 	}
    420 }
    421 
    422 /*
    423  * Check for alternatives when higher level complains
    424  * about service problems.  For now, invalidate cached
    425  * routing information.  If the route was created dynamically
    426  * (by a redirect), time to try a default gateway again.
    427  */
    428 int
    429 in_losing(inp)
    430 	struct inpcb *inp;
    431 {
    432 	register struct rtentry *rt;
    433 	struct rt_addrinfo info;
    434 
    435 	if ((rt = inp->inp_route.ro_rt)) {
    436 		inp->inp_route.ro_rt = 0;
    437 		bzero((caddr_t)&info, sizeof(info));
    438 		info.rti_info[RTAX_DST] = &inp->inp_route.ro_dst;
    439 		info.rti_info[RTAX_GATEWAY] = rt->rt_gateway;
    440 		info.rti_info[RTAX_NETMASK] = rt_mask(rt);
    441 		rt_missmsg(RTM_LOSING, &info, rt->rt_flags, 0);
    442 		if (rt->rt_flags & RTF_DYNAMIC)
    443 			(void) rtrequest(RTM_DELETE, rt_key(rt),
    444 				rt->rt_gateway, rt_mask(rt), rt->rt_flags,
    445 				(struct rtentry **)0);
    446 		else
    447 		/*
    448 		 * A new route can be allocated
    449 		 * the next time output is attempted.
    450 		 */
    451 			rtfree(rt);
    452 	}
    453 }
    454 
    455 /*
    456  * After a routing change, flush old routing
    457  * and allocate a (hopefully) better one.
    458  */
    459 void
    460 in_rtchange(inp, errno)
    461 	register struct inpcb *inp;
    462 	int errno;
    463 {
    464 	if (inp->inp_route.ro_rt) {
    465 		rtfree(inp->inp_route.ro_rt);
    466 		inp->inp_route.ro_rt = 0;
    467 		/*
    468 		 * A new route can be allocated the next time
    469 		 * output is attempted.
    470 		 */
    471 	}
    472 }
    473 
    474 struct inpcb *
    475 in_pcblookup(table, faddr, fport_arg, laddr, lport_arg, flags)
    476 	struct inpcbtable *table;
    477 	struct in_addr faddr, laddr;
    478 	u_int fport_arg, lport_arg;
    479 	int flags;
    480 {
    481 	register struct inpcb *inp, *match = 0;
    482 	int matchwild = 3, wildcard;
    483 	u_int16_t fport = fport_arg, lport = lport_arg;
    484 
    485 	for (inp = table->inpt_queue.cqh_first;
    486 	    inp != (struct inpcb *)&table->inpt_queue;
    487 	    inp = inp->inp_queue.cqe_next) {
    488 		if (inp->inp_lport != lport)
    489 			continue;
    490 		wildcard = 0;
    491 		if (inp->inp_laddr.s_addr != INADDR_ANY) {
    492 			if (laddr.s_addr == INADDR_ANY)
    493 				wildcard++;
    494 			else if (inp->inp_laddr.s_addr != laddr.s_addr)
    495 				continue;
    496 		} else {
    497 			if (laddr.s_addr != INADDR_ANY)
    498 				wildcard++;
    499 		}
    500 		if (inp->inp_faddr.s_addr != INADDR_ANY) {
    501 			if (faddr.s_addr == INADDR_ANY)
    502 				wildcard++;
    503 			else if (inp->inp_faddr.s_addr != faddr.s_addr ||
    504 			    inp->inp_fport != fport)
    505 				continue;
    506 		} else {
    507 			if (faddr.s_addr != INADDR_ANY)
    508 				wildcard++;
    509 		}
    510 		if (wildcard && (flags & INPLOOKUP_WILDCARD) == 0)
    511 			continue;
    512 		if (wildcard < matchwild) {
    513 			match = inp;
    514 			matchwild = wildcard;
    515 			if (matchwild == 0)
    516 				break;
    517 		}
    518 	}
    519 	return (match);
    520 }
    521