xref: /freebsd/sys/kern/uipc_usrreq.c (revision 8e6b01171e30297084bb0b4457c4183c2746aacc)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
34  *	$Id: uipc_usrreq.c,v 1.11 1995/08/16 16:13:27 bde Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/proc.h>
40 #include <sys/filedesc.h>
41 #include <sys/domain.h>
42 #include <sys/protosw.h>
43 #include <sys/stat.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_LOCAL };
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 	case PRU_SEND_EOF:
176 		if (control && (error = unp_internalize(control, p)))
177 			break;
178 		switch (so->so_type) {
179 
180 		case SOCK_DGRAM: {
181 			struct sockaddr *from;
182 
183 			if (nam) {
184 				if (unp->unp_conn) {
185 					error = EISCONN;
186 					break;
187 				}
188 				error = unp_connect(so, nam, p);
189 				if (error)
190 					break;
191 			} else {
192 				if (unp->unp_conn == 0) {
193 					error = ENOTCONN;
194 					break;
195 				}
196 			}
197 			so2 = unp->unp_conn->unp_socket;
198 			if (unp->unp_addr)
199 				from = mtod(unp->unp_addr, struct sockaddr *);
200 			else
201 				from = &sun_noname;
202 			if (sbappendaddr(&so2->so_rcv, from, m, control)) {
203 				sorwakeup(so2);
204 				m = 0;
205 				control = 0;
206 			} else
207 				error = ENOBUFS;
208 			if (nam)
209 				unp_disconnect(unp);
210 			break;
211 		}
212 
213 		case SOCK_STREAM:
214 #define	rcv (&so2->so_rcv)
215 #define	snd (&so->so_snd)
216 			/* Connect if not connected yet. */
217 			/*
218 			 * Note: A better implementation would complain
219 			 * if not equal to the peer's address.
220 			 */
221 			if ((so->so_state & SS_ISCONNECTED) == 0) {
222 				if (nam) {
223 		    			error = unp_connect(so, nam, p);
224 					if (error)
225 						break;	/* XXX */
226 				} else {
227 					error = ENOTCONN;
228 					break;
229 				}
230 			}
231 
232 			if (so->so_state & SS_CANTSENDMORE) {
233 				error = EPIPE;
234 				break;
235 			}
236 			if (unp->unp_conn == 0)
237 				panic("uipc 3");
238 			so2 = unp->unp_conn->unp_socket;
239 			/*
240 			 * Send to paired receive port, and then reduce
241 			 * send buffer hiwater marks to maintain backpressure.
242 			 * Wake up readers.
243 			 */
244 			if (control) {
245 				if (sbappendcontrol(rcv, m, control))
246 					control = 0;
247 			} else
248 				sbappend(rcv, m);
249 			snd->sb_mbmax -=
250 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
251 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
252 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
253 			unp->unp_conn->unp_cc = rcv->sb_cc;
254 			sorwakeup(so2);
255 			m = 0;
256 #undef snd
257 #undef rcv
258 			break;
259 
260 		default:
261 			panic("uipc 4");
262 		}
263 		/*
264 		 * SEND_EOF is equivalent to a SEND followed by
265 		 * a SHUTDOWN.
266 		 */
267 		if (req == PRU_SEND_EOF) {
268 			socantsendmore(so);
269 			unp_shutdown(unp);
270 		}
271 		break;
272 
273 	case PRU_ABORT:
274 		unp_drop(unp, ECONNABORTED);
275 		break;
276 
277 	case PRU_SENSE:
278 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
279 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
280 			so2 = unp->unp_conn->unp_socket;
281 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
282 		}
283 		((struct stat *) m)->st_dev = NODEV;
284 		if (unp->unp_ino == 0)
285 			unp->unp_ino = unp_ino++;
286 		((struct stat *) m)->st_ino = unp->unp_ino;
287 		return (0);
288 
289 	case PRU_RCVOOB:
290 		return (EOPNOTSUPP);
291 
292 	case PRU_SENDOOB:
293 		error = EOPNOTSUPP;
294 		break;
295 
296 	case PRU_SOCKADDR:
297 		if (unp->unp_addr) {
298 			nam->m_len = unp->unp_addr->m_len;
299 			bcopy(mtod(unp->unp_addr, caddr_t),
300 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
301 		} else
302 			nam->m_len = 0;
303 		break;
304 
305 	case PRU_PEERADDR:
306 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
307 			nam->m_len = unp->unp_conn->unp_addr->m_len;
308 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
309 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
310 		} else
311 			nam->m_len = 0;
312 		break;
313 
314 	case PRU_SLOWTIMO:
315 		break;
316 
317 	default:
318 		panic("piusrreq");
319 	}
320 release:
321 	if (control)
322 		m_freem(control);
323 	if (m)
324 		m_freem(m);
325 	return (error);
326 }
327 
328 /*
329  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
330  * for stream sockets, although the total for sender and receiver is
331  * actually only PIPSIZ.
332  * Datagram sockets really use the sendspace as the maximum datagram size,
333  * and don't really want to reserve the sendspace.  Their recvspace should
334  * be large enough for at least one max-size datagram plus address.
335  */
336 #ifndef PIPSIZ
337 #define	PIPSIZ	8192
338 #endif
339 u_long	unpst_sendspace = PIPSIZ;
340 u_long	unpst_recvspace = PIPSIZ;
341 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
342 u_long	unpdg_recvspace = 4*1024;
343 
344 int	unp_rights;			/* file descriptors in flight */
345 
346 int
347 unp_attach(so)
348 	struct socket *so;
349 {
350 	register struct mbuf *m;
351 	register struct unpcb *unp;
352 	int error;
353 
354 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
355 		switch (so->so_type) {
356 
357 		case SOCK_STREAM:
358 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
359 			break;
360 
361 		case SOCK_DGRAM:
362 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
363 			break;
364 
365 		default:
366 			panic("unp_attach");
367 		}
368 		if (error)
369 			return (error);
370 	}
371 	m = m_getclr(M_DONTWAIT, MT_PCB);
372 	if (m == NULL)
373 		return (ENOBUFS);
374 	unp = mtod(m, struct unpcb *);
375 	so->so_pcb = (caddr_t)unp;
376 	unp->unp_socket = so;
377 	return (0);
378 }
379 
380 void
381 unp_detach(unp)
382 	register struct unpcb *unp;
383 {
384 
385 	if (unp->unp_vnode) {
386 		unp->unp_vnode->v_socket = 0;
387 		vrele(unp->unp_vnode);
388 		unp->unp_vnode = 0;
389 	}
390 	if (unp->unp_conn)
391 		unp_disconnect(unp);
392 	while (unp->unp_refs)
393 		unp_drop(unp->unp_refs, ECONNRESET);
394 	soisdisconnected(unp->unp_socket);
395 	unp->unp_socket->so_pcb = 0;
396 	if (unp_rights) {
397 		/*
398 		 * Normally the receive buffer is flushed later,
399 		 * in sofree, but if our receive buffer holds references
400 		 * to descriptors that are now garbage, we will dispose
401 		 * of those descriptor references after the garbage collector
402 		 * gets them (resulting in a "panic: closef: count < 0").
403 		 */
404 		sorflush(unp->unp_socket);
405 		unp_gc();
406 	}
407 	m_freem(unp->unp_addr);
408 	(void) m_free(dtom(unp));
409 }
410 
411 int
412 unp_bind(unp, nam, p)
413 	struct unpcb *unp;
414 	struct mbuf *nam;
415 	struct proc *p;
416 {
417 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
418 	register struct vnode *vp;
419 	struct vattr vattr;
420 	int error;
421 	struct nameidata nd;
422 
423 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
424 		soun->sun_path, p);
425 	if (unp->unp_vnode != NULL)
426 		return (EINVAL);
427 	if (nam->m_len == MLEN) {
428 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
429 			return (EINVAL);
430 	} else
431 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
432 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
433 	error = namei(&nd);
434 	if (error)
435 		return (error);
436 	vp = nd.ni_vp;
437 	if (vp != NULL) {
438 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
439 		if (nd.ni_dvp == vp)
440 			vrele(nd.ni_dvp);
441 		else
442 			vput(nd.ni_dvp);
443 		vrele(vp);
444 		return (EADDRINUSE);
445 	}
446 	VATTR_NULL(&vattr);
447 	vattr.va_type = VSOCK;
448 	vattr.va_mode = ACCESSPERMS;
449 	LEASE_CHECK(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
450 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
451 	if (error)
452 		return (error);
453 	vp = nd.ni_vp;
454 	vp->v_socket = unp->unp_socket;
455 	unp->unp_vnode = vp;
456 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
457 	VOP_UNLOCK(vp);
458 	return (0);
459 }
460 
461 int
462 unp_connect(so, nam, p)
463 	struct socket *so;
464 	struct mbuf *nam;
465 	struct proc *p;
466 {
467 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
468 	register struct vnode *vp;
469 	register struct socket *so2, *so3;
470 	struct unpcb *unp2, *unp3;
471 	int error;
472 	struct nameidata nd;
473 
474 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
475 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
476 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
477 			return (EMSGSIZE);
478 	} else
479 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
480 	error = namei(&nd);
481 	if (error)
482 		return (error);
483 	vp = nd.ni_vp;
484 	if (vp->v_type != VSOCK) {
485 		error = ENOTSOCK;
486 		goto bad;
487 	}
488 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p);
489 	if (error)
490 		goto bad;
491 	so2 = vp->v_socket;
492 	if (so2 == 0) {
493 		error = ECONNREFUSED;
494 		goto bad;
495 	}
496 	if (so->so_type != so2->so_type) {
497 		error = EPROTOTYPE;
498 		goto bad;
499 	}
500 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
501 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
502 		    (so3 = sonewconn(so2, 0)) == 0) {
503 			error = ECONNREFUSED;
504 			goto bad;
505 		}
506 		unp2 = sotounpcb(so2);
507 		unp3 = sotounpcb(so3);
508 		if (unp2->unp_addr)
509 			unp3->unp_addr =
510 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
511 		so2 = so3;
512 	}
513 	error = unp_connect2(so, so2);
514 bad:
515 	vput(vp);
516 	return (error);
517 }
518 
519 int
520 unp_connect2(so, so2)
521 	register struct socket *so;
522 	register struct socket *so2;
523 {
524 	register struct unpcb *unp = sotounpcb(so);
525 	register struct unpcb *unp2;
526 
527 	if (so2->so_type != so->so_type)
528 		return (EPROTOTYPE);
529 	unp2 = sotounpcb(so2);
530 	unp->unp_conn = unp2;
531 	switch (so->so_type) {
532 
533 	case SOCK_DGRAM:
534 		unp->unp_nextref = unp2->unp_refs;
535 		unp2->unp_refs = unp;
536 		soisconnected(so);
537 		break;
538 
539 	case SOCK_STREAM:
540 		unp2->unp_conn = unp;
541 		soisconnected(so);
542 		soisconnected(so2);
543 		break;
544 
545 	default:
546 		panic("unp_connect2");
547 	}
548 	return (0);
549 }
550 
551 void
552 unp_disconnect(unp)
553 	struct unpcb *unp;
554 {
555 	register struct unpcb *unp2 = unp->unp_conn;
556 
557 	if (unp2 == 0)
558 		return;
559 	unp->unp_conn = 0;
560 	switch (unp->unp_socket->so_type) {
561 
562 	case SOCK_DGRAM:
563 		if (unp2->unp_refs == unp)
564 			unp2->unp_refs = unp->unp_nextref;
565 		else {
566 			unp2 = unp2->unp_refs;
567 			for (;;) {
568 				if (unp2 == 0)
569 					panic("unp_disconnect");
570 				if (unp2->unp_nextref == unp)
571 					break;
572 				unp2 = unp2->unp_nextref;
573 			}
574 			unp2->unp_nextref = unp->unp_nextref;
575 		}
576 		unp->unp_nextref = 0;
577 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
578 		break;
579 
580 	case SOCK_STREAM:
581 		soisdisconnected(unp->unp_socket);
582 		unp2->unp_conn = 0;
583 		soisdisconnected(unp2->unp_socket);
584 		break;
585 	}
586 }
587 
588 #ifdef notdef
589 void
590 unp_abort(unp)
591 	struct unpcb *unp;
592 {
593 
594 	unp_detach(unp);
595 }
596 #endif
597 
598 void
599 unp_shutdown(unp)
600 	struct unpcb *unp;
601 {
602 	struct socket *so;
603 
604 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
605 	    (so = unp->unp_conn->unp_socket))
606 		socantrcvmore(so);
607 }
608 
609 void
610 unp_drop(unp, errno)
611 	struct unpcb *unp;
612 	int errno;
613 {
614 	struct socket *so = unp->unp_socket;
615 
616 	so->so_error = errno;
617 	unp_disconnect(unp);
618 	if (so->so_head) {
619 		so->so_pcb = (caddr_t) 0;
620 		m_freem(unp->unp_addr);
621 		(void) m_free(dtom(unp));
622 		sofree(so);
623 	}
624 }
625 
626 #ifdef notdef
627 void
628 unp_drain()
629 {
630 
631 }
632 #endif
633 
634 int
635 unp_externalize(rights)
636 	struct mbuf *rights;
637 {
638 	struct proc *p = curproc;		/* XXX */
639 	register int i;
640 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
641 	register struct file **rp = (struct file **)(cm + 1);
642 	register struct file *fp;
643 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
644 	int f;
645 
646 	if (!fdavail(p, newfds)) {
647 		for (i = 0; i < newfds; i++) {
648 			fp = *rp;
649 			unp_discard(fp);
650 			*rp++ = 0;
651 		}
652 		return (EMSGSIZE);
653 	}
654 	for (i = 0; i < newfds; i++) {
655 		if (fdalloc(p, 0, &f))
656 			panic("unp_externalize");
657 		fp = *rp;
658 		p->p_fd->fd_ofiles[f] = fp;
659 		fp->f_msgcount--;
660 		unp_rights--;
661 		*(int *)rp++ = f;
662 	}
663 	return (0);
664 }
665 
666 int
667 unp_internalize(control, p)
668 	struct mbuf *control;
669 	struct proc *p;
670 {
671 	struct filedesc *fdp = p->p_fd;
672 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
673 	register struct file **rp;
674 	register struct file *fp;
675 	register int i, fd;
676 	int oldfds;
677 
678 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
679 	    cm->cmsg_len != control->m_len)
680 		return (EINVAL);
681 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
682 	rp = (struct file **)(cm + 1);
683 	for (i = 0; i < oldfds; i++) {
684 		fd = *(int *)rp++;
685 		if ((unsigned)fd >= fdp->fd_nfiles ||
686 		    fdp->fd_ofiles[fd] == NULL)
687 			return (EBADF);
688 	}
689 	rp = (struct file **)(cm + 1);
690 	for (i = 0; i < oldfds; i++) {
691 		fp = fdp->fd_ofiles[*(int *)rp];
692 		*rp++ = fp;
693 		fp->f_count++;
694 		fp->f_msgcount++;
695 		unp_rights++;
696 	}
697 	return (0);
698 }
699 
700 int	unp_defer, unp_gcing;
701 
702 void
703 unp_gc()
704 {
705 	register struct file *fp, *nextfp;
706 	register struct socket *so;
707 	struct file **extra_ref, **fpp;
708 	int nunref, i;
709 
710 	if (unp_gcing)
711 		return;
712 	unp_gcing = 1;
713 	unp_defer = 0;
714 	for (fp = filehead; fp; fp = fp->f_filef)
715 		fp->f_flag &= ~(FMARK|FDEFER);
716 	do {
717 		for (fp = filehead; fp; fp = fp->f_filef) {
718 			if (fp->f_count == 0)
719 				continue;
720 			if (fp->f_flag & FDEFER) {
721 				fp->f_flag &= ~FDEFER;
722 				unp_defer--;
723 			} else {
724 				if (fp->f_flag & FMARK)
725 					continue;
726 				if (fp->f_count == fp->f_msgcount)
727 					continue;
728 				fp->f_flag |= FMARK;
729 			}
730 			if (fp->f_type != DTYPE_SOCKET ||
731 			    (so = (struct socket *)fp->f_data) == 0)
732 				continue;
733 			if (so->so_proto->pr_domain != &localdomain ||
734 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
735 				continue;
736 #ifdef notdef
737 			if (so->so_rcv.sb_flags & SB_LOCK) {
738 				/*
739 				 * This is problematical; it's not clear
740 				 * we need to wait for the sockbuf to be
741 				 * unlocked (on a uniprocessor, at least),
742 				 * and it's also not clear what to do
743 				 * if sbwait returns an error due to receipt
744 				 * of a signal.  If sbwait does return
745 				 * an error, we'll go into an infinite
746 				 * loop.  Delete all of this for now.
747 				 */
748 				(void) sbwait(&so->so_rcv);
749 				goto restart;
750 			}
751 #endif
752 			unp_scan(so->so_rcv.sb_mb, unp_mark);
753 		}
754 	} while (unp_defer);
755 	/*
756 	 * We grab an extra reference to each of the file table entries
757 	 * that are not otherwise accessible and then free the rights
758 	 * that are stored in messages on them.
759 	 *
760 	 * The bug in the orginal code is a little tricky, so I'll describe
761 	 * what's wrong with it here.
762 	 *
763 	 * It is incorrect to simply unp_discard each entry for f_msgcount
764 	 * times -- consider the case of sockets A and B that contain
765 	 * references to each other.  On a last close of some other socket,
766 	 * we trigger a gc since the number of outstanding rights (unp_rights)
767 	 * is non-zero.  If during the sweep phase the gc code un_discards,
768 	 * we end up doing a (full) closef on the descriptor.  A closef on A
769 	 * results in the following chain.  Closef calls soo_close, which
770 	 * calls soclose.   Soclose calls first (through the switch
771 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
772 	 * returns because the previous instance had set unp_gcing, and
773 	 * we return all the way back to soclose, which marks the socket
774 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
775 	 * to free up the rights that are queued in messages on the socket A,
776 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
777 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
778 	 * instance of unp_discard just calls closef on B.
779 	 *
780 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
781 	 * which results in another closef on A.  Unfortunately, A is already
782 	 * being closed, and the descriptor has already been marked with
783 	 * SS_NOFDREF, and soclose panics at this point.
784 	 *
785 	 * Here, we first take an extra reference to each inaccessible
786 	 * descriptor.  Then, we call sorflush ourself, since we know
787 	 * it is a Unix domain socket anyhow.  After we destroy all the
788 	 * rights carried in messages, we do a last closef to get rid
789 	 * of our extra reference.  This is the last close, and the
790 	 * unp_detach etc will shut down the socket.
791 	 *
792 	 * 91/09/19, bsy@cs.cmu.edu
793 	 */
794 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
795 	for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) {
796 		nextfp = fp->f_filef;
797 		if (fp->f_count == 0)
798 			continue;
799 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
800 			*fpp++ = fp;
801 			nunref++;
802 			fp->f_count++;
803 		}
804 	}
805 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
806 		sorflush((struct socket *)(*fpp)->f_data);
807 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
808 		closef(*fpp,(struct proc*) NULL);
809 	free((caddr_t)extra_ref, M_FILE);
810 	unp_gcing = 0;
811 }
812 
813 void
814 unp_dispose(m)
815 	struct mbuf *m;
816 {
817 	if (m)
818 		unp_scan(m, unp_discard);
819 }
820 
821 void
822 unp_scan(m0, op)
823 	register struct mbuf *m0;
824 	void (*op)(struct file *);
825 {
826 	register struct mbuf *m;
827 	register struct file **rp;
828 	register struct cmsghdr *cm;
829 	register int i;
830 	int qfds;
831 
832 	while (m0) {
833 		for (m = m0; m; m = m->m_next)
834 			if (m->m_type == MT_CONTROL &&
835 			    m->m_len >= sizeof(*cm)) {
836 				cm = mtod(m, struct cmsghdr *);
837 				if (cm->cmsg_level != SOL_SOCKET ||
838 				    cm->cmsg_type != SCM_RIGHTS)
839 					continue;
840 				qfds = (cm->cmsg_len - sizeof *cm)
841 						/ sizeof (struct file *);
842 				rp = (struct file **)(cm + 1);
843 				for (i = 0; i < qfds; i++)
844 					(*op)(*rp++);
845 				break;		/* XXX, but saves time */
846 			}
847 		m0 = m0->m_act;
848 	}
849 }
850 
851 void
852 unp_mark(fp)
853 	struct file *fp;
854 {
855 
856 	if (fp->f_flag & FMARK)
857 		return;
858 	unp_defer++;
859 	fp->f_flag |= (FMARK|FDEFER);
860 }
861 
862 void
863 unp_discard(fp)
864 	struct file *fp;
865 {
866 
867 	fp->f_msgcount--;
868 	unp_rights--;
869 	(void) closef(fp, (struct proc *)NULL);
870 }
871