xref: /netbsd-src/sys/kern/uipc_usrreq.c (revision ae1bfcddc410612bc8c58b807e1830becb69a24c)
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 1.8 1994/05/04 09:50:11 mycroft 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/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/unpcb.h>
46 #include <sys/un.h>
47 #include <sys/namei.h>
48 #include <sys/vnode.h>
49 #include <sys/file.h>
50 #include <sys/stat.h>
51 #include <sys/mbuf.h>
52 
53 /*
54  * Unix communications domain.
55  *
56  * TODO:
57  *	SEQPACKET, RDM
58  *	rethink name space problems
59  *	need a proper out-of-band
60  */
61 struct	sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX };
62 ino_t	unp_ino;			/* prototype for fake inode numbers */
63 
64 /*ARGSUSED*/
65 int
66 uipc_usrreq(so, req, m, nam, control)
67 	struct socket *so;
68 	int req;
69 	struct mbuf *m, *nam, *control;
70 {
71 	struct unpcb *unp = sotounpcb(so);
72 	register struct socket *so2;
73 	register int error = 0;
74 	struct proc *p = curproc;	/* XXX */
75 
76 	if (req == PRU_CONTROL)
77 		return (EOPNOTSUPP);
78 	if (req != PRU_SEND && control && control->m_len) {
79 		error = EOPNOTSUPP;
80 		goto release;
81 	}
82 	if (unp == 0 && req != PRU_ATTACH) {
83 		error = EINVAL;
84 		goto release;
85 	}
86 	switch (req) {
87 
88 	case PRU_ATTACH:
89 		if (unp) {
90 			error = EISCONN;
91 			break;
92 		}
93 		error = unp_attach(so);
94 		break;
95 
96 	case PRU_DETACH:
97 		unp_detach(unp);
98 		break;
99 
100 	case PRU_BIND:
101 		error = unp_bind(unp, nam, p);
102 		break;
103 
104 	case PRU_LISTEN:
105 		if (unp->unp_vnode == 0)
106 			error = EINVAL;
107 		break;
108 
109 	case PRU_CONNECT:
110 		error = unp_connect(so, nam, p);
111 		break;
112 
113 	case PRU_CONNECT2:
114 		error = unp_connect2(so, (struct socket *)nam);
115 		break;
116 
117 	case PRU_DISCONNECT:
118 		unp_disconnect(unp);
119 		break;
120 
121 	case PRU_ACCEPT:
122 		/*
123 		 * Pass back name of connected socket,
124 		 * if it was bound and we are still connected
125 		 * (our peer may have closed already!).
126 		 */
127 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
128 			nam->m_len = unp->unp_conn->unp_addr->m_len;
129 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
130 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
131 		} else {
132 			nam->m_len = sizeof(sun_noname);
133 			*(mtod(nam, struct sockaddr *)) = sun_noname;
134 		}
135 		break;
136 
137 	case PRU_SHUTDOWN:
138 		socantsendmore(so);
139 		unp_shutdown(unp);
140 		break;
141 
142 	case PRU_RCVD:
143 		switch (so->so_type) {
144 
145 		case SOCK_DGRAM:
146 			panic("uipc 1");
147 			/*NOTREACHED*/
148 
149 		case SOCK_STREAM:
150 #define	rcv (&so->so_rcv)
151 #define snd (&so2->so_snd)
152 			if (unp->unp_conn == 0)
153 				break;
154 			so2 = unp->unp_conn->unp_socket;
155 			/*
156 			 * Adjust backpressure on sender
157 			 * and wakeup any waiting to write.
158 			 */
159 			snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
160 			unp->unp_mbcnt = rcv->sb_mbcnt;
161 			snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
162 			unp->unp_cc = rcv->sb_cc;
163 			sowwakeup(so2);
164 #undef snd
165 #undef rcv
166 			break;
167 
168 		default:
169 			panic("uipc 2");
170 		}
171 		break;
172 
173 	case PRU_SEND:
174 		if (control && (error = unp_internalize(control, p)))
175 			break;
176 		switch (so->so_type) {
177 
178 		case SOCK_DGRAM: {
179 			struct sockaddr *from;
180 
181 			if (nam) {
182 				if (unp->unp_conn) {
183 					error = EISCONN;
184 					break;
185 				}
186 				error = unp_connect(so, nam, p);
187 				if (error)
188 					break;
189 			} else {
190 				if (unp->unp_conn == 0) {
191 					error = ENOTCONN;
192 					break;
193 				}
194 			}
195 			so2 = unp->unp_conn->unp_socket;
196 			if (unp->unp_addr)
197 				from = mtod(unp->unp_addr, struct sockaddr *);
198 			else
199 				from = &sun_noname;
200 			if (sbappendaddr(&so2->so_rcv, from, m, control)) {
201 				sorwakeup(so2);
202 				m = 0;
203 				control = 0;
204 			} else
205 				error = ENOBUFS;
206 			if (nam)
207 				unp_disconnect(unp);
208 			break;
209 		}
210 
211 		case SOCK_STREAM:
212 #define	rcv (&so2->so_rcv)
213 #define	snd (&so->so_snd)
214 			if (so->so_state & SS_CANTSENDMORE) {
215 				error = EPIPE;
216 				break;
217 			}
218 			if (unp->unp_conn == 0)
219 				panic("uipc 3");
220 			so2 = unp->unp_conn->unp_socket;
221 			/*
222 			 * Send to paired receive port, and then reduce
223 			 * send buffer hiwater marks to maintain backpressure.
224 			 * Wake up readers.
225 			 */
226 			if (control) {
227 				if (sbappendcontrol(rcv, m, control))
228 					control = 0;
229 			} else
230 				sbappend(rcv, m);
231 			snd->sb_mbmax -=
232 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
233 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
234 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
235 			unp->unp_conn->unp_cc = rcv->sb_cc;
236 			sorwakeup(so2);
237 			m = 0;
238 #undef snd
239 #undef rcv
240 			break;
241 
242 		default:
243 			panic("uipc 4");
244 		}
245 		break;
246 
247 	case PRU_ABORT:
248 		unp_drop(unp, ECONNABORTED);
249 		break;
250 
251 	case PRU_SENSE:
252 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
253 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
254 			so2 = unp->unp_conn->unp_socket;
255 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
256 		}
257 		((struct stat *) m)->st_dev = NODEV;
258 		if (unp->unp_ino == 0)
259 			unp->unp_ino = unp_ino++;
260 		((struct stat *) m)->st_ino = unp->unp_ino;
261 		return (0);
262 
263 	case PRU_RCVOOB:
264 		return (EOPNOTSUPP);
265 
266 	case PRU_SENDOOB:
267 		error = EOPNOTSUPP;
268 		break;
269 
270 	case PRU_SOCKADDR:
271 		if (unp->unp_addr) {
272 			nam->m_len = unp->unp_addr->m_len;
273 			bcopy(mtod(unp->unp_addr, caddr_t),
274 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
275 		} else
276 			nam->m_len = 0;
277 		break;
278 
279 	case PRU_PEERADDR:
280 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
281 			nam->m_len = unp->unp_conn->unp_addr->m_len;
282 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
283 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
284 		} else
285 			nam->m_len = 0;
286 		break;
287 
288 	case PRU_SLOWTIMO:
289 		break;
290 
291 	default:
292 		panic("piusrreq");
293 	}
294 release:
295 	if (control)
296 		m_freem(control);
297 	if (m)
298 		m_freem(m);
299 	return (error);
300 }
301 
302 /*
303  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
304  * for stream sockets, although the total for sender and receiver is
305  * actually only PIPSIZ.
306  * Datagram sockets really use the sendspace as the maximum datagram size,
307  * and don't really want to reserve the sendspace.  Their recvspace should
308  * be large enough for at least one max-size datagram plus address.
309  */
310 #define	PIPSIZ	4096
311 u_long	unpst_sendspace = PIPSIZ;
312 u_long	unpst_recvspace = PIPSIZ;
313 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
314 u_long	unpdg_recvspace = 4*1024;
315 
316 int	unp_rights;			/* file descriptors in flight */
317 
318 int
319 unp_attach(so)
320 	struct socket *so;
321 {
322 	register struct mbuf *m;
323 	register struct unpcb *unp;
324 	int error;
325 
326 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
327 		switch (so->so_type) {
328 
329 		case SOCK_STREAM:
330 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
331 			break;
332 
333 		case SOCK_DGRAM:
334 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
335 			break;
336 
337 		default:
338 			panic("unp_attach");
339 		}
340 		if (error)
341 			return (error);
342 	}
343 	m = m_getclr(M_DONTWAIT, MT_PCB);
344 	if (m == NULL)
345 		return (ENOBUFS);
346 	unp = mtod(m, struct unpcb *);
347 	so->so_pcb = (caddr_t)unp;
348 	unp->unp_socket = so;
349 	return (0);
350 }
351 
352 int
353 unp_detach(unp)
354 	register struct unpcb *unp;
355 {
356 
357 	if (unp->unp_vnode) {
358 		unp->unp_vnode->v_socket = 0;
359 		vrele(unp->unp_vnode);
360 		unp->unp_vnode = 0;
361 	}
362 	if (unp->unp_conn)
363 		unp_disconnect(unp);
364 	while (unp->unp_refs)
365 		unp_drop(unp->unp_refs, ECONNRESET);
366 	soisdisconnected(unp->unp_socket);
367 	unp->unp_socket->so_pcb = 0;
368 	m_freem(unp->unp_addr);
369 	(void) m_free(dtom(unp));
370 	if (unp_rights) {
371 		/*
372 		 * Normally the receive buffer is flushed later,
373 		 * in sofree, but if our receive buffer holds references
374 		 * to descriptors that are now garbage, we will dispose
375 		 * of those descriptor references after the garbage collector
376 		 * gets them (resulting in a "panic: closef: count < 0").
377 		 */
378 		sorflush(unp->unp_socket);
379 		unp_gc();
380 	}
381 }
382 
383 int
384 unp_bind(unp, nam, p)
385 	struct unpcb *unp;
386 	struct mbuf *nam;
387 	struct proc *p;
388 {
389 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
390 	register struct vnode *vp;
391 	register struct nameidata *ndp;
392 	struct vattr vattr;
393 	int error;
394 	struct nameidata nd;
395 
396 	ndp = &nd;
397 	ndp->ni_dirp = soun->sun_path;
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 	ndp->ni_nameiop = CREATE | FOLLOW | LOCKPARENT;
407 	ndp->ni_segflg = UIO_SYSSPACE;
408 	if (error = namei(ndp, p))
409 		return (error);
410 	vp = ndp->ni_vp;
411 	if (vp != NULL) {
412 		VOP_ABORTOP(ndp);
413 		if (ndp->ni_dvp == vp)
414 			vrele(ndp->ni_dvp);
415 		else
416 			vput(ndp->ni_dvp);
417 		vrele(vp);
418 		return (EADDRINUSE);
419 	}
420 	VATTR_NULL(&vattr);
421 	vattr.va_type = VSOCK;
422 	vattr.va_mode = 0777;
423 	if (error = VOP_CREATE(ndp, &vattr, p))
424 		return (error);
425 	vp = ndp->ni_vp;
426 	vp->v_socket = unp->unp_socket;
427 	unp->unp_vnode = vp;
428 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
429 	VOP_UNLOCK(vp);
430 	return (0);
431 }
432 
433 int
434 unp_connect(so, nam, p)
435 	struct socket *so;
436 	struct mbuf *nam;
437 	struct proc *p;
438 {
439 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
440 	register struct vnode *vp;
441 	register struct socket *so2, *so3;
442 	register struct nameidata *ndp;
443 	struct unpcb *unp2, *unp3;
444 	int error;
445 	struct nameidata nd;
446 
447 	ndp = &nd;
448 	ndp->ni_dirp = soun->sun_path;
449 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
450 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
451 			return (EMSGSIZE);
452 	} else
453 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
454 	ndp->ni_nameiop = LOOKUP | FOLLOW | LOCKLEAF;
455 	ndp->ni_segflg = UIO_SYSSPACE;
456 	if (error = namei(ndp, p))
457 		return (error);
458 	vp = ndp->ni_vp;
459 	if (vp->v_type != VSOCK) {
460 		error = ENOTSOCK;
461 		goto bad;
462 	}
463 	if (error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p))
464 		goto bad;
465 	so2 = vp->v_socket;
466 	if (so2 == 0) {
467 		error = ECONNREFUSED;
468 		goto bad;
469 	}
470 	if (so->so_type != so2->so_type) {
471 		error = EPROTOTYPE;
472 		goto bad;
473 	}
474 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
475 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
476 		    (so3 = sonewconn(so2, 0)) == 0) {
477 			error = ECONNREFUSED;
478 			goto bad;
479 		}
480 		unp2 = sotounpcb(so2);
481 		unp3 = sotounpcb(so3);
482 		if (unp2->unp_addr)
483 			unp3->unp_addr =
484 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
485 		so2 = so3;
486 	}
487 	error = unp_connect2(so, so2);
488 bad:
489 	vput(vp);
490 	return (error);
491 }
492 
493 int
494 unp_connect2(so, so2)
495 	register struct socket *so;
496 	register struct socket *so2;
497 {
498 	register struct unpcb *unp = sotounpcb(so);
499 	register struct unpcb *unp2;
500 
501 	if (so2->so_type != so->so_type)
502 		return (EPROTOTYPE);
503 	unp2 = sotounpcb(so2);
504 	unp->unp_conn = unp2;
505 	switch (so->so_type) {
506 
507 	case SOCK_DGRAM:
508 		unp->unp_nextref = unp2->unp_refs;
509 		unp2->unp_refs = unp;
510 		soisconnected(so);
511 		break;
512 
513 	case SOCK_STREAM:
514 		unp2->unp_conn = unp;
515 		soisconnected(so);
516 		soisconnected(so2);
517 		break;
518 
519 	default:
520 		panic("unp_connect2");
521 	}
522 	return (0);
523 }
524 
525 void
526 unp_disconnect(unp)
527 	struct unpcb *unp;
528 {
529 	register struct unpcb *unp2 = unp->unp_conn;
530 
531 	if (unp2 == 0)
532 		return;
533 	unp->unp_conn = 0;
534 	switch (unp->unp_socket->so_type) {
535 
536 	case SOCK_DGRAM:
537 		if (unp2->unp_refs == unp)
538 			unp2->unp_refs = unp->unp_nextref;
539 		else {
540 			unp2 = unp2->unp_refs;
541 			for (;;) {
542 				if (unp2 == 0)
543 					panic("unp_disconnect");
544 				if (unp2->unp_nextref == unp)
545 					break;
546 				unp2 = unp2->unp_nextref;
547 			}
548 			unp2->unp_nextref = unp->unp_nextref;
549 		}
550 		unp->unp_nextref = 0;
551 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
552 		break;
553 
554 	case SOCK_STREAM:
555 		soisdisconnected(unp->unp_socket);
556 		unp2->unp_conn = 0;
557 		soisdisconnected(unp2->unp_socket);
558 		break;
559 	}
560 }
561 
562 #ifdef notdef
563 unp_abort(unp)
564 	struct unpcb *unp;
565 {
566 
567 	unp_detach(unp);
568 }
569 #endif
570 
571 void
572 unp_shutdown(unp)
573 	struct unpcb *unp;
574 {
575 	struct socket *so;
576 
577 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
578 	    (so = unp->unp_conn->unp_socket))
579 		socantrcvmore(so);
580 }
581 
582 void
583 unp_drop(unp, errno)
584 	struct unpcb *unp;
585 	int errno;
586 {
587 	struct socket *so = unp->unp_socket;
588 
589 	so->so_error = errno;
590 	unp_disconnect(unp);
591 	if (so->so_head) {
592 		so->so_pcb = (caddr_t) 0;
593 		m_freem(unp->unp_addr);
594 		(void) m_free(dtom(unp));
595 		sofree(so);
596 	}
597 }
598 
599 #ifdef notdef
600 unp_drain()
601 {
602 
603 }
604 #endif
605 
606 int
607 unp_externalize(rights)
608 	struct mbuf *rights;
609 {
610 	struct proc *p = curproc;		/* XXX */
611 	register int i;
612 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
613 	register struct file **rp = (struct file **)(cm + 1);
614 	register struct file *fp;
615 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
616 	int f;
617 
618 	if (!fdavail(p, newfds)) {
619 		for (i = 0; i < newfds; i++) {
620 			fp = *rp;
621 			unp_discard(fp);
622 			*rp++ = 0;
623 		}
624 		return (EMSGSIZE);
625 	}
626 	for (i = 0; i < newfds; i++) {
627 		if (fdalloc(p, 0, &f))
628 			panic("unp_externalize");
629 		fp = *rp;
630 		p->p_fd->fd_ofiles[f] = fp;
631 		fp->f_msgcount--;
632 		unp_rights--;
633 		*(int *)rp++ = f;
634 	}
635 	return (0);
636 }
637 
638 int
639 unp_internalize(control, p)
640 	struct mbuf *control;
641 	struct proc *p;
642 {
643 	struct filedesc *fdp = p->p_fd;
644 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
645 	register struct file **rp;
646 	register struct file *fp;
647 	register int i, fd;
648 	int oldfds;
649 
650 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
651 	    cm->cmsg_len != control->m_len)
652 		return (EINVAL);
653 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
654 	rp = (struct file **)(cm + 1);
655 	for (i = 0; i < oldfds; i++) {
656 		fd = *(int *)rp++;
657 		if ((unsigned)fd >= fdp->fd_nfiles ||
658 		    fdp->fd_ofiles[fd] == NULL)
659 			return (EBADF);
660 	}
661 	rp = (struct file **)(cm + 1);
662 	for (i = 0; i < oldfds; i++) {
663 		fp = fdp->fd_ofiles[*(int *)rp];
664 		*rp++ = fp;
665 		fp->f_count++;
666 		fp->f_msgcount++;
667 		unp_rights++;
668 	}
669 	return (0);
670 }
671 
672 int	unp_defer, unp_gcing;
673 extern	struct domain unixdomain;
674 
675 void
676 unp_gc()
677 {
678 	register struct file *fp, *nextfp;
679 	register struct socket *so;
680 	struct file **extra_ref, **fpp;
681 	int nunref, i;
682 
683 	if (unp_gcing)
684 		return;
685 	unp_gcing = 1;
686 	unp_defer = 0;
687 	for (fp = filehead; fp; fp = fp->f_filef)
688 		fp->f_flag &= ~(FMARK|FDEFER);
689 	do {
690 		for (fp = filehead; fp; fp = fp->f_filef) {
691 			if (fp->f_count == 0)
692 				continue;
693 			if (fp->f_flag & FDEFER) {
694 				fp->f_flag &= ~FDEFER;
695 				unp_defer--;
696 			} else {
697 				if (fp->f_flag & FMARK)
698 					continue;
699 				if (fp->f_count == fp->f_msgcount)
700 					continue;
701 				fp->f_flag |= FMARK;
702 			}
703 			if (fp->f_type != DTYPE_SOCKET ||
704 			    (so = (struct socket *)fp->f_data) == 0)
705 				continue;
706 			if (so->so_proto->pr_domain != &unixdomain ||
707 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
708 				continue;
709 #ifdef notdef
710 			if (so->so_rcv.sb_flags & SB_LOCK) {
711 				/*
712 				 * This is problematical; it's not clear
713 				 * we need to wait for the sockbuf to be
714 				 * unlocked (on a uniprocessor, at least),
715 				 * and it's also not clear what to do
716 				 * if sbwait returns an error due to receipt
717 				 * of a signal.  If sbwait does return
718 				 * an error, we'll go into an infinite
719 				 * loop.  Delete all of this for now.
720 				 */
721 				(void) sbwait(&so->so_rcv);
722 				goto restart;
723 			}
724 #endif
725 			unp_scan(so->so_rcv.sb_mb, unp_mark);
726 		}
727 	} while (unp_defer);
728 	/*
729 	 * We grab an extra reference to each of the file table entries
730 	 * that are not otherwise accessible and then free the rights
731 	 * that are stored in messages on them.
732 	 *
733 	 * The bug in the orginal code is a little tricky, so I'll describe
734 	 * what's wrong with it here.
735 	 *
736 	 * It is incorrect to simply unp_discard each entry for f_msgcount
737 	 * times -- consider the case of sockets A and B that contain
738 	 * references to each other.  On a last close of some other socket,
739 	 * we trigger a gc since the number of outstanding rights (unp_rights)
740 	 * is non-zero.  If during the sweep phase the gc code un_discards,
741 	 * we end up doing a (full) closef on the descriptor.  A closef on A
742 	 * results in the following chain.  Closef calls soo_close, which
743 	 * calls soclose.   Soclose calls first (through the switch
744 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
745 	 * returns because the previous instance had set unp_gcing, and
746 	 * we return all the way back to soclose, which marks the socket
747 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
748 	 * to free up the rights that are queued in messages on the socket A,
749 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
750 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
751 	 * instance of unp_discard just calls closef on B.
752 	 *
753 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
754 	 * which results in another closef on A.  Unfortunately, A is already
755 	 * being closed, and the descriptor has already been marked with
756 	 * SS_NOFDREF, and soclose panics at this point.
757 	 *
758 	 * Here, we first take an extra reference to each inaccessible
759 	 * descriptor.  Then, we call sorflush ourself, since we know
760 	 * it is a Unix domain socket anyhow.  After we destroy all the
761 	 * rights carried in messages, we do a last closef to get rid
762 	 * of our extra reference.  This is the last close, and the
763 	 * unp_detach etc will shut down the socket.
764 	 *
765 	 * 91/09/19, bsy@cs.cmu.edu
766 	 */
767 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
768 	for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) {
769 		nextfp = fp->f_filef;
770 		if (fp->f_count == 0)
771 			continue;
772 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
773 			*fpp++ = fp;
774 			nunref++;
775 			fp->f_count++;
776 		}
777 	}
778 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
779 		sorflush((struct socket *)(*fpp)->f_data);
780 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
781 		closef(*fpp);
782 	free((caddr_t)extra_ref, M_FILE);
783 	unp_gcing = 0;
784 }
785 
786 void
787 unp_dispose(m)
788 	struct mbuf *m;
789 {
790 
791 	if (m)
792 		unp_scan(m, unp_discard);
793 }
794 
795 void
796 unp_scan(m0, op)
797 	register struct mbuf *m0;
798 	void (*op) __P((struct file *));
799 {
800 	register struct mbuf *m;
801 	register struct file **rp;
802 	register struct cmsghdr *cm;
803 	register int i;
804 	int qfds;
805 
806 	while (m0) {
807 		for (m = m0; m; m = m->m_next)
808 			if (m->m_type == MT_CONTROL &&
809 			    m->m_len >= sizeof(*cm)) {
810 				cm = mtod(m, struct cmsghdr *);
811 				if (cm->cmsg_level != SOL_SOCKET ||
812 				    cm->cmsg_type != SCM_RIGHTS)
813 					continue;
814 				qfds = (cm->cmsg_len - sizeof *cm)
815 						/ sizeof (struct file *);
816 				rp = (struct file **)(cm + 1);
817 				for (i = 0; i < qfds; i++)
818 					(*op)(*rp++);
819 				break;		/* XXX, but saves time */
820 			}
821 		m0 = m0->m_act;
822 	}
823 }
824 
825 void
826 unp_mark(fp)
827 	struct file *fp;
828 {
829 
830 	if (fp->f_flag & FMARK)
831 		return;
832 	unp_defer++;
833 	fp->f_flag |= (FMARK|FDEFER);
834 }
835 
836 void
837 unp_discard(fp)
838 	struct file *fp;
839 {
840 
841 	fp->f_msgcount--;
842 	unp_rights--;
843 	(void) closef(fp, (struct proc *)NULL);
844 }
845