xref: /netbsd-src/sys/kern/uipc_usrreq.c (revision d0fed6c87ddc40a8bffa6f99e7433ddfc864dd83)
1 /*	$NetBSD: uipc_usrreq.c,v 1.24 1997/04/10 01:51:21 cgd Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Christopher G. Demetriou.  All rights reserved.
5  * Copyright (c) 1982, 1986, 1989, 1991, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
37  */
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/proc.h>
42 #include <sys/filedesc.h>
43 #include <sys/domain.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/unpcb.h>
48 #include <sys/un.h>
49 #include <sys/namei.h>
50 #include <sys/vnode.h>
51 #include <sys/file.h>
52 #include <sys/stat.h>
53 #include <sys/mbuf.h>
54 
55 /*
56  * Unix communications domain.
57  *
58  * TODO:
59  *	SEQPACKET, RDM
60  *	rethink name space problems
61  *	need a proper out-of-band
62  */
63 struct	sockaddr_un sun_noname = { sizeof(sun_noname), AF_UNIX };
64 ino_t	unp_ino;			/* prototype for fake inode numbers */
65 
66 int
67 unp_output(m, control, unp)
68 	struct mbuf *m, *control;
69 	struct unpcb *unp;
70 {
71 	struct socket *so2;
72 	struct sockaddr_un *sun;
73 
74 	so2 = unp->unp_conn->unp_socket;
75 	if (unp->unp_addr)
76 		sun = unp->unp_addr;
77 	else
78 		sun = &sun_noname;
79 	if (sbappendaddr(&so2->so_rcv, (struct sockaddr *)sun, m,
80 	    control) == 0) {
81 		m_freem(control);
82 		m_freem(m);
83 		return (EINVAL);
84 	} else {
85 		sorwakeup(so2);
86 		return (0);
87 	}
88 }
89 
90 void
91 unp_setsockaddr(unp, nam)
92 	register struct unpcb *unp;
93 	struct mbuf *nam;
94 {
95 	struct sockaddr_un *sun;
96 
97 	if (unp->unp_addr)
98 		sun = unp->unp_addr;
99 	else
100 		sun = &sun_noname;
101 	nam->m_len = sun->sun_len;
102 	bcopy(sun, mtod(nam, caddr_t), (size_t)nam->m_len);
103 }
104 
105 void
106 unp_setpeeraddr(unp, nam)
107 	register struct unpcb *unp;
108 	struct mbuf *nam;
109 {
110 	struct sockaddr_un *sun;
111 
112 	if (unp->unp_conn && unp->unp_conn->unp_addr)
113 		sun = unp->unp_conn->unp_addr;
114 	else
115 		sun = &sun_noname;
116 	nam->m_len = sun->sun_len;
117 	bcopy(sun, mtod(nam, caddr_t), (size_t)nam->m_len);
118 }
119 
120 /*ARGSUSED*/
121 int
122 uipc_usrreq(so, req, m, nam, control, p)
123 	struct socket *so;
124 	int req;
125 	struct mbuf *m, *nam, *control;
126 	struct proc *p;
127 {
128 	struct unpcb *unp = sotounpcb(so);
129 	register struct socket *so2;
130 	register int error = 0;
131 
132 	if (req == PRU_CONTROL)
133 		return (EOPNOTSUPP);
134 
135 #ifdef DIAGNOSTIC
136 	if (req != PRU_SEND && req != PRU_SENDOOB && control)
137 		panic("uipc_usrreq: unexpected control mbuf");
138 #endif
139 	if (unp == 0 && req != PRU_ATTACH) {
140 		error = EINVAL;
141 		goto release;
142 	}
143 
144 	switch (req) {
145 
146 	case PRU_ATTACH:
147 		if (unp != 0) {
148 			error = EISCONN;
149 			break;
150 		}
151 		error = unp_attach(so);
152 		break;
153 
154 	case PRU_DETACH:
155 		unp_detach(unp);
156 		break;
157 
158 	case PRU_BIND:
159 		error = unp_bind(unp, nam, p);
160 		break;
161 
162 	case PRU_LISTEN:
163 		if (unp->unp_vnode == 0)
164 			error = EINVAL;
165 		break;
166 
167 	case PRU_CONNECT:
168 		error = unp_connect(so, nam, p);
169 		break;
170 
171 	case PRU_CONNECT2:
172 		error = unp_connect2(so, (struct socket *)nam);
173 		break;
174 
175 	case PRU_DISCONNECT:
176 		unp_disconnect(unp);
177 		break;
178 
179 	case PRU_ACCEPT:
180 		unp_setpeeraddr(unp, nam);
181 		break;
182 
183 	case PRU_SHUTDOWN:
184 		socantsendmore(so);
185 		unp_shutdown(unp);
186 		break;
187 
188 	case PRU_RCVD:
189 		switch (so->so_type) {
190 
191 		case SOCK_DGRAM:
192 			panic("uipc 1");
193 			/*NOTREACHED*/
194 
195 		case SOCK_STREAM:
196 #define	rcv (&so->so_rcv)
197 #define snd (&so2->so_snd)
198 			if (unp->unp_conn == 0)
199 				break;
200 			so2 = unp->unp_conn->unp_socket;
201 			/*
202 			 * Adjust backpressure on sender
203 			 * and wakeup any waiting to write.
204 			 */
205 			snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
206 			unp->unp_mbcnt = rcv->sb_mbcnt;
207 			snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
208 			unp->unp_cc = rcv->sb_cc;
209 			sowwakeup(so2);
210 #undef snd
211 #undef rcv
212 			break;
213 
214 		default:
215 			panic("uipc 2");
216 		}
217 		break;
218 
219 	case PRU_SEND:
220 		if (control && (error = unp_internalize(control, p)))
221 			break;
222 		switch (so->so_type) {
223 
224 		case SOCK_DGRAM: {
225 			if (nam) {
226 				if ((so->so_state & SS_ISCONNECTED) != 0) {
227 					error = EISCONN;
228 					goto die;
229 				}
230 				error = unp_connect(so, nam, p);
231 				if (error) {
232 				die:
233 					m_freem(control);
234 					m_freem(m);
235 					break;
236 				}
237 			} else {
238 				if ((so->so_state & SS_ISCONNECTED) == 0) {
239 					error = ENOTCONN;
240 					goto die;
241 				}
242 			}
243 			error = unp_output(m, control, unp);
244 			if (nam)
245 				unp_disconnect(unp);
246 			break;
247 		}
248 
249 		case SOCK_STREAM:
250 #define	rcv (&so2->so_rcv)
251 #define	snd (&so->so_snd)
252 			if (unp->unp_conn == 0)
253 				panic("uipc 3");
254 			so2 = unp->unp_conn->unp_socket;
255 			/*
256 			 * Send to paired receive port, and then reduce
257 			 * send buffer hiwater marks to maintain backpressure.
258 			 * Wake up readers.
259 			 */
260 			if (control) {
261 				if (sbappendcontrol(rcv, m, control) == 0)
262 					m_freem(control);
263 			} else
264 				sbappend(rcv, m);
265 			snd->sb_mbmax -=
266 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
267 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
268 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
269 			unp->unp_conn->unp_cc = rcv->sb_cc;
270 			sorwakeup(so2);
271 #undef snd
272 #undef rcv
273 			break;
274 
275 		default:
276 			panic("uipc 4");
277 		}
278 		break;
279 
280 	case PRU_ABORT:
281 		unp_drop(unp, ECONNABORTED);
282 		break;
283 
284 	case PRU_SENSE:
285 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
286 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
287 			so2 = unp->unp_conn->unp_socket;
288 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
289 		}
290 		((struct stat *) m)->st_dev = NODEV;
291 		if (unp->unp_ino == 0)
292 			unp->unp_ino = unp_ino++;
293 		((struct stat *) m)->st_ino = unp->unp_ino;
294 		return (0);
295 
296 	case PRU_RCVOOB:
297 		error = EOPNOTSUPP;
298 		break;
299 
300 	case PRU_SENDOOB:
301 		m_freem(control);
302 		m_freem(m);
303 		error = EOPNOTSUPP;
304 		break;
305 
306 	case PRU_SOCKADDR:
307 		unp_setsockaddr(unp, nam);
308 		break;
309 
310 	case PRU_PEERADDR:
311 		unp_setpeeraddr(unp, nam);
312 		break;
313 
314 	default:
315 		panic("piusrreq");
316 	}
317 
318 release:
319 	return (error);
320 }
321 
322 /*
323  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
324  * for stream sockets, although the total for sender and receiver is
325  * actually only PIPSIZ.
326  * Datagram sockets really use the sendspace as the maximum datagram size,
327  * and don't really want to reserve the sendspace.  Their recvspace should
328  * be large enough for at least one max-size datagram plus address.
329  */
330 #define	PIPSIZ	4096
331 u_long	unpst_sendspace = PIPSIZ;
332 u_long	unpst_recvspace = PIPSIZ;
333 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
334 u_long	unpdg_recvspace = 4*1024;
335 
336 int	unp_rights;			/* file descriptors in flight */
337 
338 int
339 unp_attach(so)
340 	struct socket *so;
341 {
342 	register struct unpcb *unp;
343 	int error;
344 
345 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
346 		switch (so->so_type) {
347 
348 		case SOCK_STREAM:
349 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
350 			break;
351 
352 		case SOCK_DGRAM:
353 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
354 			break;
355 
356 		default:
357 			panic("unp_attach");
358 		}
359 		if (error)
360 			return (error);
361 	}
362 	unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT);
363 	if (unp == NULL)
364 		return (ENOBUFS);
365 	bzero((caddr_t)unp, sizeof(*unp));
366 	unp->unp_socket = so;
367 	so->so_pcb = unp;
368 	return (0);
369 }
370 
371 void
372 unp_detach(unp)
373 	register struct unpcb *unp;
374 {
375 
376 	if (unp->unp_vnode) {
377 		unp->unp_vnode->v_socket = 0;
378 		vrele(unp->unp_vnode);
379 		unp->unp_vnode = 0;
380 	}
381 	if (unp->unp_conn)
382 		unp_disconnect(unp);
383 	while (unp->unp_refs)
384 		unp_drop(unp->unp_refs, ECONNRESET);
385 	soisdisconnected(unp->unp_socket);
386 	unp->unp_socket->so_pcb = 0;
387 	if (unp->unp_addr)
388 		m_freem(dtom(unp->unp_addr));
389 	if (unp_rights) {
390 		/*
391 		 * Normally the receive buffer is flushed later,
392 		 * in sofree, but if our receive buffer holds references
393 		 * to descriptors that are now garbage, we will dispose
394 		 * of those descriptor references after the garbage collector
395 		 * gets them (resulting in a "panic: closef: count < 0").
396 		 */
397 		sorflush(unp->unp_socket);
398 		free(unp, M_PCB);
399 		unp_gc();
400 	} else
401 		free(unp, M_PCB);
402 }
403 
404 int
405 unp_bind(unp, nam, p)
406 	struct unpcb *unp;
407 	struct mbuf *nam;
408 	struct proc *p;
409 {
410 	struct sockaddr_un *sun = mtod(nam, struct sockaddr_un *);
411 	register struct vnode *vp;
412 	struct vattr vattr;
413 	int error;
414 	struct nameidata nd;
415 
416 	if (unp->unp_vnode != 0)
417 		return (EINVAL);
418 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
419 	    sun->sun_path, p);
420 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
421 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
422 			return (EINVAL);
423 	} else
424 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
425 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
426 	if ((error = namei(&nd)) != 0)
427 		return (error);
428 	vp = nd.ni_vp;
429 	if (vp != NULL) {
430 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
431 		if (nd.ni_dvp == vp)
432 			vrele(nd.ni_dvp);
433 		else
434 			vput(nd.ni_dvp);
435 		vrele(vp);
436 		return (EADDRINUSE);
437 	}
438 	VATTR_NULL(&vattr);
439 	vattr.va_type = VSOCK;
440 	vattr.va_mode = ACCESSPERMS;
441 	VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
442 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
443 	if (error)
444 		return (error);
445 	vp = nd.ni_vp;
446 	vp->v_socket = unp->unp_socket;
447 	unp->unp_vnode = vp;
448 	unp->unp_addr =
449 	    mtod(m_copy(nam, 0, (int)M_COPYALL), struct sockaddr_un *);
450 	VOP_UNLOCK(vp);
451 	return (0);
452 }
453 
454 int
455 unp_connect(so, nam, p)
456 	struct socket *so;
457 	struct mbuf *nam;
458 	struct proc *p;
459 {
460 	register struct sockaddr_un *sun = mtod(nam, struct sockaddr_un *);
461 	register struct vnode *vp;
462 	register struct socket *so2, *so3;
463 	struct unpcb *unp2, *unp3;
464 	int error;
465 	struct nameidata nd;
466 
467 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, sun->sun_path, p);
468 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
469 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
470 			return (EINVAL);
471 	} else
472 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
473 	if ((error = namei(&nd)) != 0)
474 		return (error);
475 	vp = nd.ni_vp;
476 	if (vp->v_type != VSOCK) {
477 		error = ENOTSOCK;
478 		goto bad;
479 	}
480 	if ((error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p)) != 0)
481 		goto bad;
482 	so2 = vp->v_socket;
483 	if (so2 == 0) {
484 		error = ECONNREFUSED;
485 		goto bad;
486 	}
487 	if (so->so_type != so2->so_type) {
488 		error = EPROTOTYPE;
489 		goto bad;
490 	}
491 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
492 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
493 		    (so3 = sonewconn(so2, 0)) == 0) {
494 			error = ECONNREFUSED;
495 			goto bad;
496 		}
497 		unp2 = sotounpcb(so2);
498 		unp3 = sotounpcb(so3);
499 		if (unp2->unp_addr)
500 			unp3->unp_addr = mtod(m_copy(dtom(unp2->unp_addr), 0,
501 			    (int)M_COPYALL), struct sockaddr_un *);
502 		so2 = so3;
503 	}
504 	error = unp_connect2(so, so2);
505 bad:
506 	vput(vp);
507 	return (error);
508 }
509 
510 int
511 unp_connect2(so, so2)
512 	register struct socket *so;
513 	register struct socket *so2;
514 {
515 	register struct unpcb *unp = sotounpcb(so);
516 	register struct unpcb *unp2;
517 
518 	if (so2->so_type != so->so_type)
519 		return (EPROTOTYPE);
520 	unp2 = sotounpcb(so2);
521 	unp->unp_conn = unp2;
522 	switch (so->so_type) {
523 
524 	case SOCK_DGRAM:
525 		unp->unp_nextref = unp2->unp_refs;
526 		unp2->unp_refs = unp;
527 		soisconnected(so);
528 		break;
529 
530 	case SOCK_STREAM:
531 		unp2->unp_conn = unp;
532 		soisconnected(so);
533 		soisconnected(so2);
534 		break;
535 
536 	default:
537 		panic("unp_connect2");
538 	}
539 	return (0);
540 }
541 
542 void
543 unp_disconnect(unp)
544 	struct unpcb *unp;
545 {
546 	register struct unpcb *unp2 = unp->unp_conn;
547 
548 	if (unp2 == 0)
549 		return;
550 	unp->unp_conn = 0;
551 	switch (unp->unp_socket->so_type) {
552 
553 	case SOCK_DGRAM:
554 		if (unp2->unp_refs == unp)
555 			unp2->unp_refs = unp->unp_nextref;
556 		else {
557 			unp2 = unp2->unp_refs;
558 			for (;;) {
559 				if (unp2 == 0)
560 					panic("unp_disconnect");
561 				if (unp2->unp_nextref == unp)
562 					break;
563 				unp2 = unp2->unp_nextref;
564 			}
565 			unp2->unp_nextref = unp->unp_nextref;
566 		}
567 		unp->unp_nextref = 0;
568 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
569 		break;
570 
571 	case SOCK_STREAM:
572 		soisdisconnected(unp->unp_socket);
573 		unp2->unp_conn = 0;
574 		soisdisconnected(unp2->unp_socket);
575 		break;
576 	}
577 }
578 
579 #ifdef notdef
580 unp_abort(unp)
581 	struct unpcb *unp;
582 {
583 
584 	unp_detach(unp);
585 }
586 #endif
587 
588 void
589 unp_shutdown(unp)
590 	struct unpcb *unp;
591 {
592 	struct socket *so;
593 
594 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
595 	    (so = unp->unp_conn->unp_socket))
596 		socantrcvmore(so);
597 }
598 
599 void
600 unp_drop(unp, errno)
601 	struct unpcb *unp;
602 	int errno;
603 {
604 	struct socket *so = unp->unp_socket;
605 
606 	so->so_error = errno;
607 	unp_disconnect(unp);
608 	if (so->so_head) {
609 		so->so_pcb = 0;
610 		sofree(so);
611 		if (unp->unp_addr)
612 			m_freem(dtom(unp->unp_addr));
613 		free(unp, M_PCB);
614 	}
615 }
616 
617 #ifdef notdef
618 unp_drain()
619 {
620 
621 }
622 #endif
623 
624 int
625 unp_externalize(rights)
626 	struct mbuf *rights;
627 {
628 	struct proc *p = curproc;		/* XXX */
629 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
630 	register int i, *fdp = (int *)(cm + 1);
631 	register struct file **rp = (struct file **)ALIGN(cm + 1);
632 	register struct file *fp;
633 	int nfds = (cm->cmsg_len - ALIGN(sizeof(*cm))) / sizeof (struct file *);
634 	int f;
635 
636 	/* Make sure that the recipient has space */
637 	if (!fdavail(p, nfds)) {
638 		for (i = 0; i < nfds; i++) {
639 			fp = *rp;
640 			unp_discard(fp);
641 			*rp++ = 0;
642 		}
643 		return (EMSGSIZE);
644 	}
645 
646 	/*
647 	 * Add file to the recipient's open file table, converting them
648 	 * to integer file descriptors as we go.  Done in forward order
649 	 * because an integer will always come in the same place or before
650 	 * its corresponding struct file pointer.
651 	 */
652 	for (i = 0; i < nfds; i++) {
653 		if (fdalloc(p, 0, &f))
654 			panic("unp_externalize");
655 		fp = *rp;
656 		p->p_fd->fd_ofiles[f] = fp;
657 		fp->f_msgcount--;
658 		unp_rights--;
659 		*fdp++ = f;
660 	}
661 
662 	/*
663 	 * Adjust length, in case of transition from large struct file
664 	 * pointers to ints.
665 	 */
666 	cm->cmsg_len = sizeof(*cm) + (nfds * sizeof(int));
667 	rights->m_len = cm->cmsg_len;
668 	return (0);
669 }
670 
671 int
672 unp_internalize(control, p)
673 	struct mbuf *control;
674 	struct proc *p;
675 {
676 	struct filedesc *fdescp = p->p_fd;
677 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
678 	register struct file **rp;
679 	register struct file *fp;
680 	register int i, fd, *fdp;
681 	int nfds;
682 	u_int neededspace;
683 
684 	/* Sanity check the control message header */
685 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
686 	    cm->cmsg_len != control->m_len)
687 		return (EINVAL);
688 
689 	/* Verify that the file descriptors are valid */
690 	nfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
691 	fdp = (int *)(cm + 1);
692 	for (i = 0; i < nfds; i++) {
693 		fd = *fdp++;
694 		if ((unsigned)fd >= fdescp->fd_nfiles ||
695 		    fdescp->fd_ofiles[fd] == NULL)
696 			return (EBADF);
697 	}
698 
699 	/* Make sure we have room for the struct file pointers */
700 morespace:
701 	neededspace = (ALIGN(sizeof (*cm)) + nfds * sizeof (struct file *)) -
702 		control->m_len;
703 	if (neededspace > M_TRAILINGSPACE(control)) {
704 
705 		/* if we already have a cluster, the message is just too big */
706 		if (control->m_flags & M_EXT)
707 			return (E2BIG);
708 
709 		/* allocate a cluster and try again */
710 		MCLGET(control, M_WAIT);
711 		if ((control->m_flags & M_EXT) == 0)
712 			return (ENOBUFS);	/* allocation failed */
713 
714 		/* copy the data to the cluster */
715 		bcopy(cm, mtod(control, char *), cm->cmsg_len);
716 		cm = mtod(control, struct cmsghdr *);
717 		goto morespace;
718 	}
719 
720 	/* adjust message & mbuf to note amount of space actually used. */
721 	cm->cmsg_len += neededspace;
722 	control->m_len = cm->cmsg_len;
723 
724 	/*
725 	 * Transform the file descriptors into struct file pointers, in
726 	 * reverse order so that if pointers are bigger than ints, the
727 	 * int won't get until we're done.
728 	 */
729 	fdp = ((int *)(cm + 1)) + nfds - 1;
730 	rp = ((struct file **)ALIGN(cm + 1)) + nfds - 1;
731 	for (i = 0; i < nfds; i++) {
732 		fp = fdescp->fd_ofiles[*fdp];
733 		*rp-- = fp;
734 		fp->f_count++;
735 		fp->f_msgcount++;
736 		unp_rights++;
737 	}
738 	return (0);
739 }
740 
741 int	unp_defer, unp_gcing;
742 extern	struct domain unixdomain;
743 
744 void
745 unp_gc()
746 {
747 	register struct file *fp, *nextfp;
748 	register struct socket *so;
749 	struct file **extra_ref, **fpp;
750 	int nunref, i;
751 
752 	if (unp_gcing)
753 		return;
754 	unp_gcing = 1;
755 	unp_defer = 0;
756 	for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next)
757 		fp->f_flag &= ~(FMARK|FDEFER);
758 	do {
759 		for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) {
760 			if (fp->f_count == 0)
761 				continue;
762 			if (fp->f_flag & FDEFER) {
763 				fp->f_flag &= ~FDEFER;
764 				unp_defer--;
765 			} else {
766 				if (fp->f_flag & FMARK)
767 					continue;
768 				if (fp->f_count == fp->f_msgcount)
769 					continue;
770 				fp->f_flag |= FMARK;
771 			}
772 			if (fp->f_type != DTYPE_SOCKET ||
773 			    (so = (struct socket *)fp->f_data) == 0)
774 				continue;
775 			if (so->so_proto->pr_domain != &unixdomain ||
776 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
777 				continue;
778 #ifdef notdef
779 			if (so->so_rcv.sb_flags & SB_LOCK) {
780 				/*
781 				 * This is problematical; it's not clear
782 				 * we need to wait for the sockbuf to be
783 				 * unlocked (on a uniprocessor, at least),
784 				 * and it's also not clear what to do
785 				 * if sbwait returns an error due to receipt
786 				 * of a signal.  If sbwait does return
787 				 * an error, we'll go into an infinite
788 				 * loop.  Delete all of this for now.
789 				 */
790 				(void) sbwait(&so->so_rcv);
791 				goto restart;
792 			}
793 #endif
794 			unp_scan(so->so_rcv.sb_mb, unp_mark);
795 		}
796 	} while (unp_defer);
797 	/*
798 	 * We grab an extra reference to each of the file table entries
799 	 * that are not otherwise accessible and then free the rights
800 	 * that are stored in messages on them.
801 	 *
802 	 * The bug in the orginal code is a little tricky, so I'll describe
803 	 * what's wrong with it here.
804 	 *
805 	 * It is incorrect to simply unp_discard each entry for f_msgcount
806 	 * times -- consider the case of sockets A and B that contain
807 	 * references to each other.  On a last close of some other socket,
808 	 * we trigger a gc since the number of outstanding rights (unp_rights)
809 	 * is non-zero.  If during the sweep phase the gc code un_discards,
810 	 * we end up doing a (full) closef on the descriptor.  A closef on A
811 	 * results in the following chain.  Closef calls soo_close, which
812 	 * calls soclose.   Soclose calls first (through the switch
813 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
814 	 * returns because the previous instance had set unp_gcing, and
815 	 * we return all the way back to soclose, which marks the socket
816 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
817 	 * to free up the rights that are queued in messages on the socket A,
818 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
819 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
820 	 * instance of unp_discard just calls closef on B.
821 	 *
822 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
823 	 * which results in another closef on A.  Unfortunately, A is already
824 	 * being closed, and the descriptor has already been marked with
825 	 * SS_NOFDREF, and soclose panics at this point.
826 	 *
827 	 * Here, we first take an extra reference to each inaccessible
828 	 * descriptor.  Then, we call sorflush ourself, since we know
829 	 * it is a Unix domain socket anyhow.  After we destroy all the
830 	 * rights carried in messages, we do a last closef to get rid
831 	 * of our extra reference.  This is the last close, and the
832 	 * unp_detach etc will shut down the socket.
833 	 *
834 	 * 91/09/19, bsy@cs.cmu.edu
835 	 */
836 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
837 	for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0;
838 	    fp = nextfp) {
839 		nextfp = fp->f_list.le_next;
840 		if (fp->f_count == 0)
841 			continue;
842 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
843 			*fpp++ = fp;
844 			nunref++;
845 			fp->f_count++;
846 		}
847 	}
848 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
849 		sorflush((struct socket *)(*fpp)->f_data);
850 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
851 		(void) closef(*fpp, (struct proc *)0);
852 	free((caddr_t)extra_ref, M_FILE);
853 	unp_gcing = 0;
854 }
855 
856 void
857 unp_dispose(m)
858 	struct mbuf *m;
859 {
860 
861 	if (m)
862 		unp_scan(m, unp_discard);
863 }
864 
865 void
866 unp_scan(m0, op)
867 	register struct mbuf *m0;
868 	void (*op) __P((struct file *));
869 {
870 	register struct mbuf *m;
871 	register struct file **rp;
872 	register struct cmsghdr *cm;
873 	register int i;
874 	int qfds;
875 
876 	while (m0) {
877 		for (m = m0; m; m = m->m_next)
878 			if (m->m_type == MT_CONTROL &&
879 			    m->m_len >= sizeof(*cm)) {
880 				cm = mtod(m, struct cmsghdr *);
881 				if (cm->cmsg_level != SOL_SOCKET ||
882 				    cm->cmsg_type != SCM_RIGHTS)
883 					continue;
884 				qfds = (cm->cmsg_len - sizeof *cm)
885 						/ sizeof (struct file *);
886 				rp = (struct file **)(cm + 1);
887 				for (i = 0; i < qfds; i++)
888 					(*op)(*rp++);
889 				break;		/* XXX, but saves time */
890 			}
891 		m0 = m0->m_act;
892 	}
893 }
894 
895 void
896 unp_mark(fp)
897 	struct file *fp;
898 {
899 
900 	if (fp->f_flag & FMARK)
901 		return;
902 	unp_defer++;
903 	fp->f_flag |= (FMARK|FDEFER);
904 }
905 
906 void
907 unp_discard(fp)
908 	struct file *fp;
909 {
910 
911 	fp->f_msgcount--;
912 	unp_rights--;
913 	(void) closef(fp, (struct proc *)0);
914 }
915