Home | History | Annotate | Line # | Download | only in kern
uipc_usrreq.c revision 1.10
      1 /*	$NetBSD: uipc_usrreq.c,v 1.10 1994/06/29 06:33:43 cgd Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1982, 1986, 1989, 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  *	@(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
     36  */
     37 
     38 #include <sys/param.h>
     39 #include <sys/systm.h>
     40 #include <sys/proc.h>
     41 #include <sys/filedesc.h>
     42 #include <sys/domain.h>
     43 #include <sys/protosw.h>
     44 #include <sys/socket.h>
     45 #include <sys/socketvar.h>
     46 #include <sys/unpcb.h>
     47 #include <sys/un.h>
     48 #include <sys/namei.h>
     49 #include <sys/vnode.h>
     50 #include <sys/file.h>
     51 #include <sys/stat.h>
     52 #include <sys/mbuf.h>
     53 
     54 /*
     55  * Unix communications domain.
     56  *
     57  * TODO:
     58  *	SEQPACKET, RDM
     59  *	rethink name space problems
     60  *	need a proper out-of-band
     61  */
     62 struct	sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX };
     63 ino_t	unp_ino;			/* prototype for fake inode numbers */
     64 
     65 /*ARGSUSED*/
     66 int
     67 uipc_usrreq(so, req, m, nam, control)
     68 	struct socket *so;
     69 	int req;
     70 	struct mbuf *m, *nam, *control;
     71 {
     72 	struct unpcb *unp = sotounpcb(so);
     73 	register struct socket *so2;
     74 	register int error = 0;
     75 	struct proc *p = curproc;	/* XXX */
     76 
     77 	if (req == PRU_CONTROL)
     78 		return (EOPNOTSUPP);
     79 	if (req != PRU_SEND && control && control->m_len) {
     80 		error = EOPNOTSUPP;
     81 		goto release;
     82 	}
     83 	if (unp == 0 && req != PRU_ATTACH) {
     84 		error = EINVAL;
     85 		goto release;
     86 	}
     87 	switch (req) {
     88 
     89 	case PRU_ATTACH:
     90 		if (unp) {
     91 			error = EISCONN;
     92 			break;
     93 		}
     94 		error = unp_attach(so);
     95 		break;
     96 
     97 	case PRU_DETACH:
     98 		unp_detach(unp);
     99 		break;
    100 
    101 	case PRU_BIND:
    102 		error = unp_bind(unp, nam, p);
    103 		break;
    104 
    105 	case PRU_LISTEN:
    106 		if (unp->unp_vnode == 0)
    107 			error = EINVAL;
    108 		break;
    109 
    110 	case PRU_CONNECT:
    111 		error = unp_connect(so, nam, p);
    112 		break;
    113 
    114 	case PRU_CONNECT2:
    115 		error = unp_connect2(so, (struct socket *)nam);
    116 		break;
    117 
    118 	case PRU_DISCONNECT:
    119 		unp_disconnect(unp);
    120 		break;
    121 
    122 	case PRU_ACCEPT:
    123 		/*
    124 		 * Pass back name of connected socket,
    125 		 * if it was bound and we are still connected
    126 		 * (our peer may have closed already!).
    127 		 */
    128 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
    129 			nam->m_len = unp->unp_conn->unp_addr->m_len;
    130 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
    131 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
    132 		} else {
    133 			nam->m_len = sizeof(sun_noname);
    134 			*(mtod(nam, struct sockaddr *)) = sun_noname;
    135 		}
    136 		break;
    137 
    138 	case PRU_SHUTDOWN:
    139 		socantsendmore(so);
    140 		unp_shutdown(unp);
    141 		break;
    142 
    143 	case PRU_RCVD:
    144 		switch (so->so_type) {
    145 
    146 		case SOCK_DGRAM:
    147 			panic("uipc 1");
    148 			/*NOTREACHED*/
    149 
    150 		case SOCK_STREAM:
    151 #define	rcv (&so->so_rcv)
    152 #define snd (&so2->so_snd)
    153 			if (unp->unp_conn == 0)
    154 				break;
    155 			so2 = unp->unp_conn->unp_socket;
    156 			/*
    157 			 * Adjust backpressure on sender
    158 			 * and wakeup any waiting to write.
    159 			 */
    160 			snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
    161 			unp->unp_mbcnt = rcv->sb_mbcnt;
    162 			snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
    163 			unp->unp_cc = rcv->sb_cc;
    164 			sowwakeup(so2);
    165 #undef snd
    166 #undef rcv
    167 			break;
    168 
    169 		default:
    170 			panic("uipc 2");
    171 		}
    172 		break;
    173 
    174 	case PRU_SEND:
    175 		if (control && (error = unp_internalize(control, p)))
    176 			break;
    177 		switch (so->so_type) {
    178 
    179 		case SOCK_DGRAM: {
    180 			struct sockaddr *from;
    181 
    182 			if (nam) {
    183 				if (unp->unp_conn) {
    184 					error = EISCONN;
    185 					break;
    186 				}
    187 				error = unp_connect(so, nam, p);
    188 				if (error)
    189 					break;
    190 			} else {
    191 				if (unp->unp_conn == 0) {
    192 					error = ENOTCONN;
    193 					break;
    194 				}
    195 			}
    196 			so2 = unp->unp_conn->unp_socket;
    197 			if (unp->unp_addr)
    198 				from = mtod(unp->unp_addr, struct sockaddr *);
    199 			else
    200 				from = &sun_noname;
    201 			if (sbappendaddr(&so2->so_rcv, from, m, control)) {
    202 				sorwakeup(so2);
    203 				m = 0;
    204 				control = 0;
    205 			} else
    206 				error = ENOBUFS;
    207 			if (nam)
    208 				unp_disconnect(unp);
    209 			break;
    210 		}
    211 
    212 		case SOCK_STREAM:
    213 #define	rcv (&so2->so_rcv)
    214 #define	snd (&so->so_snd)
    215 			if (so->so_state & SS_CANTSENDMORE) {
    216 				error = EPIPE;
    217 				break;
    218 			}
    219 			if (unp->unp_conn == 0)
    220 				panic("uipc 3");
    221 			so2 = unp->unp_conn->unp_socket;
    222 			/*
    223 			 * Send to paired receive port, and then reduce
    224 			 * send buffer hiwater marks to maintain backpressure.
    225 			 * Wake up readers.
    226 			 */
    227 			if (control) {
    228 				if (sbappendcontrol(rcv, m, control))
    229 					control = 0;
    230 			} else
    231 				sbappend(rcv, m);
    232 			snd->sb_mbmax -=
    233 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
    234 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
    235 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
    236 			unp->unp_conn->unp_cc = rcv->sb_cc;
    237 			sorwakeup(so2);
    238 			m = 0;
    239 #undef snd
    240 #undef rcv
    241 			break;
    242 
    243 		default:
    244 			panic("uipc 4");
    245 		}
    246 		break;
    247 
    248 	case PRU_ABORT:
    249 		unp_drop(unp, ECONNABORTED);
    250 		break;
    251 
    252 	case PRU_SENSE:
    253 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
    254 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
    255 			so2 = unp->unp_conn->unp_socket;
    256 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
    257 		}
    258 		((struct stat *) m)->st_dev = NODEV;
    259 		if (unp->unp_ino == 0)
    260 			unp->unp_ino = unp_ino++;
    261 		((struct stat *) m)->st_ino = unp->unp_ino;
    262 		return (0);
    263 
    264 	case PRU_RCVOOB:
    265 		return (EOPNOTSUPP);
    266 
    267 	case PRU_SENDOOB:
    268 		error = EOPNOTSUPP;
    269 		break;
    270 
    271 	case PRU_SOCKADDR:
    272 		if (unp->unp_addr) {
    273 			nam->m_len = unp->unp_addr->m_len;
    274 			bcopy(mtod(unp->unp_addr, caddr_t),
    275 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
    276 		} else
    277 			nam->m_len = 0;
    278 		break;
    279 
    280 	case PRU_PEERADDR:
    281 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
    282 			nam->m_len = unp->unp_conn->unp_addr->m_len;
    283 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
    284 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
    285 		} else
    286 			nam->m_len = 0;
    287 		break;
    288 
    289 	case PRU_SLOWTIMO:
    290 		break;
    291 
    292 	default:
    293 		panic("piusrreq");
    294 	}
    295 release:
    296 	if (control)
    297 		m_freem(control);
    298 	if (m)
    299 		m_freem(m);
    300 	return (error);
    301 }
    302 
    303 /*
    304  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
    305  * for stream sockets, although the total for sender and receiver is
    306  * actually only PIPSIZ.
    307  * Datagram sockets really use the sendspace as the maximum datagram size,
    308  * and don't really want to reserve the sendspace.  Their recvspace should
    309  * be large enough for at least one max-size datagram plus address.
    310  */
    311 #define	PIPSIZ	4096
    312 u_long	unpst_sendspace = PIPSIZ;
    313 u_long	unpst_recvspace = PIPSIZ;
    314 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
    315 u_long	unpdg_recvspace = 4*1024;
    316 
    317 int	unp_rights;			/* file descriptors in flight */
    318 
    319 int
    320 unp_attach(so)
    321 	struct socket *so;
    322 {
    323 	register struct mbuf *m;
    324 	register struct unpcb *unp;
    325 	int error;
    326 
    327 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
    328 		switch (so->so_type) {
    329 
    330 		case SOCK_STREAM:
    331 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
    332 			break;
    333 
    334 		case SOCK_DGRAM:
    335 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
    336 			break;
    337 
    338 		default:
    339 			panic("unp_attach");
    340 		}
    341 		if (error)
    342 			return (error);
    343 	}
    344 	m = m_getclr(M_DONTWAIT, MT_PCB);
    345 	if (m == NULL)
    346 		return (ENOBUFS);
    347 	unp = mtod(m, struct unpcb *);
    348 	so->so_pcb = (caddr_t)unp;
    349 	unp->unp_socket = so;
    350 	return (0);
    351 }
    352 
    353 int
    354 unp_detach(unp)
    355 	register struct unpcb *unp;
    356 {
    357 
    358 	if (unp->unp_vnode) {
    359 		unp->unp_vnode->v_socket = 0;
    360 		vrele(unp->unp_vnode);
    361 		unp->unp_vnode = 0;
    362 	}
    363 	if (unp->unp_conn)
    364 		unp_disconnect(unp);
    365 	while (unp->unp_refs)
    366 		unp_drop(unp->unp_refs, ECONNRESET);
    367 	soisdisconnected(unp->unp_socket);
    368 	unp->unp_socket->so_pcb = 0;
    369 	m_freem(unp->unp_addr);
    370 	(void) m_free(dtom(unp));
    371 	if (unp_rights) {
    372 		/*
    373 		 * Normally the receive buffer is flushed later,
    374 		 * in sofree, but if our receive buffer holds references
    375 		 * to descriptors that are now garbage, we will dispose
    376 		 * of those descriptor references after the garbage collector
    377 		 * gets them (resulting in a "panic: closef: count < 0").
    378 		 */
    379 		sorflush(unp->unp_socket);
    380 		unp_gc();
    381 	}
    382 }
    383 
    384 int
    385 unp_bind(unp, nam, p)
    386 	struct unpcb *unp;
    387 	struct mbuf *nam;
    388 	struct proc *p;
    389 {
    390 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
    391 	register struct vnode *vp;
    392 	struct vattr vattr;
    393 	int error;
    394 	struct nameidata nd;
    395 
    396 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
    397 	    soun->sun_path, p);
    398 	if (unp->unp_vnode != NULL)
    399 		return (EINVAL);
    400 	if (nam->m_len == MLEN) {
    401 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
    402 			return (EINVAL);
    403 	} else
    404 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
    405 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
    406 	if (error = namei(&nd))
    407 		return (error);
    408 	vp = nd.ni_vp;
    409 	if (vp != NULL) {
    410 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
    411 		if (nd.ni_dvp == vp)
    412 			vrele(nd.ni_dvp);
    413 		else
    414 			vput(nd.ni_dvp);
    415 		vrele(vp);
    416 		return (EADDRINUSE);
    417 	}
    418 	VATTR_NULL(&vattr);
    419 	vattr.va_type = VSOCK;
    420 	vattr.va_mode = ACCESSPERMS;
    421 	LEASE_CHECK(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
    422 	if (error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr))
    423 		return (error);
    424 	vp = nd.ni_vp;
    425 	vp->v_socket = unp->unp_socket;
    426 	unp->unp_vnode = vp;
    427 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
    428 	VOP_UNLOCK(vp);
    429 	return (0);
    430 }
    431 
    432 int
    433 unp_connect(so, nam, p)
    434 	struct socket *so;
    435 	struct mbuf *nam;
    436 	struct proc *p;
    437 {
    438 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
    439 	register struct vnode *vp;
    440 	register struct socket *so2, *so3;
    441 	struct unpcb *unp2, *unp3;
    442 	int error;
    443 	struct nameidata nd;
    444 
    445 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
    446 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
    447 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
    448 			return (EMSGSIZE);
    449 	} else
    450 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
    451 	if (error = namei(&nd))
    452 		return (error);
    453 	vp = nd.ni_vp;
    454 	if (vp->v_type != VSOCK) {
    455 		error = ENOTSOCK;
    456 		goto bad;
    457 	}
    458 	if (error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p))
    459 		goto bad;
    460 	so2 = vp->v_socket;
    461 	if (so2 == 0) {
    462 		error = ECONNREFUSED;
    463 		goto bad;
    464 	}
    465 	if (so->so_type != so2->so_type) {
    466 		error = EPROTOTYPE;
    467 		goto bad;
    468 	}
    469 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
    470 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
    471 		    (so3 = sonewconn(so2, 0)) == 0) {
    472 			error = ECONNREFUSED;
    473 			goto bad;
    474 		}
    475 		unp2 = sotounpcb(so2);
    476 		unp3 = sotounpcb(so3);
    477 		if (unp2->unp_addr)
    478 			unp3->unp_addr =
    479 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
    480 		so2 = so3;
    481 	}
    482 	error = unp_connect2(so, so2);
    483 bad:
    484 	vput(vp);
    485 	return (error);
    486 }
    487 
    488 int
    489 unp_connect2(so, so2)
    490 	register struct socket *so;
    491 	register struct socket *so2;
    492 {
    493 	register struct unpcb *unp = sotounpcb(so);
    494 	register struct unpcb *unp2;
    495 
    496 	if (so2->so_type != so->so_type)
    497 		return (EPROTOTYPE);
    498 	unp2 = sotounpcb(so2);
    499 	unp->unp_conn = unp2;
    500 	switch (so->so_type) {
    501 
    502 	case SOCK_DGRAM:
    503 		unp->unp_nextref = unp2->unp_refs;
    504 		unp2->unp_refs = unp;
    505 		soisconnected(so);
    506 		break;
    507 
    508 	case SOCK_STREAM:
    509 		unp2->unp_conn = unp;
    510 		soisconnected(so);
    511 		soisconnected(so2);
    512 		break;
    513 
    514 	default:
    515 		panic("unp_connect2");
    516 	}
    517 	return (0);
    518 }
    519 
    520 void
    521 unp_disconnect(unp)
    522 	struct unpcb *unp;
    523 {
    524 	register struct unpcb *unp2 = unp->unp_conn;
    525 
    526 	if (unp2 == 0)
    527 		return;
    528 	unp->unp_conn = 0;
    529 	switch (unp->unp_socket->so_type) {
    530 
    531 	case SOCK_DGRAM:
    532 		if (unp2->unp_refs == unp)
    533 			unp2->unp_refs = unp->unp_nextref;
    534 		else {
    535 			unp2 = unp2->unp_refs;
    536 			for (;;) {
    537 				if (unp2 == 0)
    538 					panic("unp_disconnect");
    539 				if (unp2->unp_nextref == unp)
    540 					break;
    541 				unp2 = unp2->unp_nextref;
    542 			}
    543 			unp2->unp_nextref = unp->unp_nextref;
    544 		}
    545 		unp->unp_nextref = 0;
    546 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
    547 		break;
    548 
    549 	case SOCK_STREAM:
    550 		soisdisconnected(unp->unp_socket);
    551 		unp2->unp_conn = 0;
    552 		soisdisconnected(unp2->unp_socket);
    553 		break;
    554 	}
    555 }
    556 
    557 #ifdef notdef
    558 unp_abort(unp)
    559 	struct unpcb *unp;
    560 {
    561 
    562 	unp_detach(unp);
    563 }
    564 #endif
    565 
    566 void
    567 unp_shutdown(unp)
    568 	struct unpcb *unp;
    569 {
    570 	struct socket *so;
    571 
    572 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
    573 	    (so = unp->unp_conn->unp_socket))
    574 		socantrcvmore(so);
    575 }
    576 
    577 void
    578 unp_drop(unp, errno)
    579 	struct unpcb *unp;
    580 	int errno;
    581 {
    582 	struct socket *so = unp->unp_socket;
    583 
    584 	so->so_error = errno;
    585 	unp_disconnect(unp);
    586 	if (so->so_head) {
    587 		so->so_pcb = (caddr_t) 0;
    588 		m_freem(unp->unp_addr);
    589 		(void) m_free(dtom(unp));
    590 		sofree(so);
    591 	}
    592 }
    593 
    594 #ifdef notdef
    595 unp_drain()
    596 {
    597 
    598 }
    599 #endif
    600 
    601 int
    602 unp_externalize(rights)
    603 	struct mbuf *rights;
    604 {
    605 	struct proc *p = curproc;		/* XXX */
    606 	register int i;
    607 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
    608 	register struct file **rp = (struct file **)(cm + 1);
    609 	register struct file *fp;
    610 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
    611 	int f;
    612 
    613 	if (!fdavail(p, newfds)) {
    614 		for (i = 0; i < newfds; i++) {
    615 			fp = *rp;
    616 			unp_discard(fp);
    617 			*rp++ = 0;
    618 		}
    619 		return (EMSGSIZE);
    620 	}
    621 	for (i = 0; i < newfds; i++) {
    622 		if (fdalloc(p, 0, &f))
    623 			panic("unp_externalize");
    624 		fp = *rp;
    625 		p->p_fd->fd_ofiles[f] = fp;
    626 		fp->f_msgcount--;
    627 		unp_rights--;
    628 		*(int *)rp++ = f;
    629 	}
    630 	return (0);
    631 }
    632 
    633 int
    634 unp_internalize(control, p)
    635 	struct mbuf *control;
    636 	struct proc *p;
    637 {
    638 	struct filedesc *fdp = p->p_fd;
    639 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
    640 	register struct file **rp;
    641 	register struct file *fp;
    642 	register int i, fd;
    643 	int oldfds;
    644 
    645 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
    646 	    cm->cmsg_len != control->m_len)
    647 		return (EINVAL);
    648 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
    649 	rp = (struct file **)(cm + 1);
    650 	for (i = 0; i < oldfds; i++) {
    651 		fd = *(int *)rp++;
    652 		if ((unsigned)fd >= fdp->fd_nfiles ||
    653 		    fdp->fd_ofiles[fd] == NULL)
    654 			return (EBADF);
    655 	}
    656 	rp = (struct file **)(cm + 1);
    657 	for (i = 0; i < oldfds; i++) {
    658 		fp = fdp->fd_ofiles[*(int *)rp];
    659 		*rp++ = fp;
    660 		fp->f_count++;
    661 		fp->f_msgcount++;
    662 		unp_rights++;
    663 	}
    664 	return (0);
    665 }
    666 
    667 int	unp_defer, unp_gcing;
    668 extern	struct domain unixdomain;
    669 
    670 void
    671 unp_gc()
    672 {
    673 	register struct file *fp, *nextfp;
    674 	register struct socket *so;
    675 	struct file **extra_ref, **fpp;
    676 	int nunref, i;
    677 
    678 	if (unp_gcing)
    679 		return;
    680 	unp_gcing = 1;
    681 	unp_defer = 0;
    682 	for (fp = filehead; fp; fp = fp->f_filef)
    683 		fp->f_flag &= ~(FMARK|FDEFER);
    684 	do {
    685 		for (fp = filehead; fp; fp = fp->f_filef) {
    686 			if (fp->f_count == 0)
    687 				continue;
    688 			if (fp->f_flag & FDEFER) {
    689 				fp->f_flag &= ~FDEFER;
    690 				unp_defer--;
    691 			} else {
    692 				if (fp->f_flag & FMARK)
    693 					continue;
    694 				if (fp->f_count == fp->f_msgcount)
    695 					continue;
    696 				fp->f_flag |= FMARK;
    697 			}
    698 			if (fp->f_type != DTYPE_SOCKET ||
    699 			    (so = (struct socket *)fp->f_data) == 0)
    700 				continue;
    701 			if (so->so_proto->pr_domain != &unixdomain ||
    702 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
    703 				continue;
    704 #ifdef notdef
    705 			if (so->so_rcv.sb_flags & SB_LOCK) {
    706 				/*
    707 				 * This is problematical; it's not clear
    708 				 * we need to wait for the sockbuf to be
    709 				 * unlocked (on a uniprocessor, at least),
    710 				 * and it's also not clear what to do
    711 				 * if sbwait returns an error due to receipt
    712 				 * of a signal.  If sbwait does return
    713 				 * an error, we'll go into an infinite
    714 				 * loop.  Delete all of this for now.
    715 				 */
    716 				(void) sbwait(&so->so_rcv);
    717 				goto restart;
    718 			}
    719 #endif
    720 			unp_scan(so->so_rcv.sb_mb, unp_mark);
    721 		}
    722 	} while (unp_defer);
    723 	/*
    724 	 * We grab an extra reference to each of the file table entries
    725 	 * that are not otherwise accessible and then free the rights
    726 	 * that are stored in messages on them.
    727 	 *
    728 	 * The bug in the orginal code is a little tricky, so I'll describe
    729 	 * what's wrong with it here.
    730 	 *
    731 	 * It is incorrect to simply unp_discard each entry for f_msgcount
    732 	 * times -- consider the case of sockets A and B that contain
    733 	 * references to each other.  On a last close of some other socket,
    734 	 * we trigger a gc since the number of outstanding rights (unp_rights)
    735 	 * is non-zero.  If during the sweep phase the gc code un_discards,
    736 	 * we end up doing a (full) closef on the descriptor.  A closef on A
    737 	 * results in the following chain.  Closef calls soo_close, which
    738 	 * calls soclose.   Soclose calls first (through the switch
    739 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
    740 	 * returns because the previous instance had set unp_gcing, and
    741 	 * we return all the way back to soclose, which marks the socket
    742 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
    743 	 * to free up the rights that are queued in messages on the socket A,
    744 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
    745 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
    746 	 * instance of unp_discard just calls closef on B.
    747 	 *
    748 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
    749 	 * which results in another closef on A.  Unfortunately, A is already
    750 	 * being closed, and the descriptor has already been marked with
    751 	 * SS_NOFDREF, and soclose panics at this point.
    752 	 *
    753 	 * Here, we first take an extra reference to each inaccessible
    754 	 * descriptor.  Then, we call sorflush ourself, since we know
    755 	 * it is a Unix domain socket anyhow.  After we destroy all the
    756 	 * rights carried in messages, we do a last closef to get rid
    757 	 * of our extra reference.  This is the last close, and the
    758 	 * unp_detach etc will shut down the socket.
    759 	 *
    760 	 * 91/09/19, bsy (at) cs.cmu.edu
    761 	 */
    762 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
    763 	for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) {
    764 		nextfp = fp->f_filef;
    765 		if (fp->f_count == 0)
    766 			continue;
    767 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
    768 			*fpp++ = fp;
    769 			nunref++;
    770 			fp->f_count++;
    771 		}
    772 	}
    773 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
    774 		sorflush((struct socket *)(*fpp)->f_data);
    775 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
    776 		closef(*fpp);
    777 	free((caddr_t)extra_ref, M_FILE);
    778 	unp_gcing = 0;
    779 }
    780 
    781 void
    782 unp_dispose(m)
    783 	struct mbuf *m;
    784 {
    785 
    786 	if (m)
    787 		unp_scan(m, unp_discard);
    788 }
    789 
    790 void
    791 unp_scan(m0, op)
    792 	register struct mbuf *m0;
    793 	void (*op) __P((struct file *));
    794 {
    795 	register struct mbuf *m;
    796 	register struct file **rp;
    797 	register struct cmsghdr *cm;
    798 	register int i;
    799 	int qfds;
    800 
    801 	while (m0) {
    802 		for (m = m0; m; m = m->m_next)
    803 			if (m->m_type == MT_CONTROL &&
    804 			    m->m_len >= sizeof(*cm)) {
    805 				cm = mtod(m, struct cmsghdr *);
    806 				if (cm->cmsg_level != SOL_SOCKET ||
    807 				    cm->cmsg_type != SCM_RIGHTS)
    808 					continue;
    809 				qfds = (cm->cmsg_len - sizeof *cm)
    810 						/ sizeof (struct file *);
    811 				rp = (struct file **)(cm + 1);
    812 				for (i = 0; i < qfds; i++)
    813 					(*op)(*rp++);
    814 				break;		/* XXX, but saves time */
    815 			}
    816 		m0 = m0->m_act;
    817 	}
    818 }
    819 
    820 void
    821 unp_mark(fp)
    822 	struct file *fp;
    823 {
    824 
    825 	if (fp->f_flag & FMARK)
    826 		return;
    827 	unp_defer++;
    828 	fp->f_flag |= (FMARK|FDEFER);
    829 }
    830 
    831 void
    832 unp_discard(fp)
    833 	struct file *fp;
    834 {
    835 
    836 	fp->f_msgcount--;
    837 	unp_rights--;
    838 	(void) closef(fp, (struct proc *)NULL);
    839 }
    840