xref: /netbsd-src/sys/kern/uipc_usrreq.c (revision d9158b13b5dfe46201430699a3f7a235ecf28df3)
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 1.9 1994/06/08 11:28:47 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 	struct vattr vattr;
392 	int error;
393 	struct nameidata nd;
394 
395 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
396 	    soun->sun_path, p);
397 	if (unp->unp_vnode != NULL)
398 		return (EINVAL);
399 	if (nam->m_len == MLEN) {
400 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
401 			return (EINVAL);
402 	} else
403 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
404 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
405 	if (error = namei(&nd))
406 		return (error);
407 	vp = nd.ni_vp;
408 	if (vp != NULL) {
409 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
410 		if (nd.ni_dvp == vp)
411 			vrele(nd.ni_dvp);
412 		else
413 			vput(nd.ni_dvp);
414 		vrele(vp);
415 		return (EADDRINUSE);
416 	}
417 	VATTR_NULL(&vattr);
418 	vattr.va_type = VSOCK;
419 	vattr.va_mode = ACCESSPERMS;
420 	LEASE_CHECK(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
421 	if (error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr))
422 		return (error);
423 	vp = nd.ni_vp;
424 	vp->v_socket = unp->unp_socket;
425 	unp->unp_vnode = vp;
426 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
427 	VOP_UNLOCK(vp);
428 	return (0);
429 }
430 
431 int
432 unp_connect(so, nam, p)
433 	struct socket *so;
434 	struct mbuf *nam;
435 	struct proc *p;
436 {
437 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
438 	register struct vnode *vp;
439 	register struct socket *so2, *so3;
440 	struct unpcb *unp2, *unp3;
441 	int error;
442 	struct nameidata nd;
443 
444 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
445 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
446 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
447 			return (EMSGSIZE);
448 	} else
449 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
450 	if (error = namei(&nd))
451 		return (error);
452 	vp = nd.ni_vp;
453 	if (vp->v_type != VSOCK) {
454 		error = ENOTSOCK;
455 		goto bad;
456 	}
457 	if (error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p))
458 		goto bad;
459 	so2 = vp->v_socket;
460 	if (so2 == 0) {
461 		error = ECONNREFUSED;
462 		goto bad;
463 	}
464 	if (so->so_type != so2->so_type) {
465 		error = EPROTOTYPE;
466 		goto bad;
467 	}
468 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
469 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
470 		    (so3 = sonewconn(so2, 0)) == 0) {
471 			error = ECONNREFUSED;
472 			goto bad;
473 		}
474 		unp2 = sotounpcb(so2);
475 		unp3 = sotounpcb(so3);
476 		if (unp2->unp_addr)
477 			unp3->unp_addr =
478 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
479 		so2 = so3;
480 	}
481 	error = unp_connect2(so, so2);
482 bad:
483 	vput(vp);
484 	return (error);
485 }
486 
487 int
488 unp_connect2(so, so2)
489 	register struct socket *so;
490 	register struct socket *so2;
491 {
492 	register struct unpcb *unp = sotounpcb(so);
493 	register struct unpcb *unp2;
494 
495 	if (so2->so_type != so->so_type)
496 		return (EPROTOTYPE);
497 	unp2 = sotounpcb(so2);
498 	unp->unp_conn = unp2;
499 	switch (so->so_type) {
500 
501 	case SOCK_DGRAM:
502 		unp->unp_nextref = unp2->unp_refs;
503 		unp2->unp_refs = unp;
504 		soisconnected(so);
505 		break;
506 
507 	case SOCK_STREAM:
508 		unp2->unp_conn = unp;
509 		soisconnected(so);
510 		soisconnected(so2);
511 		break;
512 
513 	default:
514 		panic("unp_connect2");
515 	}
516 	return (0);
517 }
518 
519 void
520 unp_disconnect(unp)
521 	struct unpcb *unp;
522 {
523 	register struct unpcb *unp2 = unp->unp_conn;
524 
525 	if (unp2 == 0)
526 		return;
527 	unp->unp_conn = 0;
528 	switch (unp->unp_socket->so_type) {
529 
530 	case SOCK_DGRAM:
531 		if (unp2->unp_refs == unp)
532 			unp2->unp_refs = unp->unp_nextref;
533 		else {
534 			unp2 = unp2->unp_refs;
535 			for (;;) {
536 				if (unp2 == 0)
537 					panic("unp_disconnect");
538 				if (unp2->unp_nextref == unp)
539 					break;
540 				unp2 = unp2->unp_nextref;
541 			}
542 			unp2->unp_nextref = unp->unp_nextref;
543 		}
544 		unp->unp_nextref = 0;
545 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
546 		break;
547 
548 	case SOCK_STREAM:
549 		soisdisconnected(unp->unp_socket);
550 		unp2->unp_conn = 0;
551 		soisdisconnected(unp2->unp_socket);
552 		break;
553 	}
554 }
555 
556 #ifdef notdef
557 unp_abort(unp)
558 	struct unpcb *unp;
559 {
560 
561 	unp_detach(unp);
562 }
563 #endif
564 
565 void
566 unp_shutdown(unp)
567 	struct unpcb *unp;
568 {
569 	struct socket *so;
570 
571 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
572 	    (so = unp->unp_conn->unp_socket))
573 		socantrcvmore(so);
574 }
575 
576 void
577 unp_drop(unp, errno)
578 	struct unpcb *unp;
579 	int errno;
580 {
581 	struct socket *so = unp->unp_socket;
582 
583 	so->so_error = errno;
584 	unp_disconnect(unp);
585 	if (so->so_head) {
586 		so->so_pcb = (caddr_t) 0;
587 		m_freem(unp->unp_addr);
588 		(void) m_free(dtom(unp));
589 		sofree(so);
590 	}
591 }
592 
593 #ifdef notdef
594 unp_drain()
595 {
596 
597 }
598 #endif
599 
600 int
601 unp_externalize(rights)
602 	struct mbuf *rights;
603 {
604 	struct proc *p = curproc;		/* XXX */
605 	register int i;
606 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
607 	register struct file **rp = (struct file **)(cm + 1);
608 	register struct file *fp;
609 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
610 	int f;
611 
612 	if (!fdavail(p, newfds)) {
613 		for (i = 0; i < newfds; i++) {
614 			fp = *rp;
615 			unp_discard(fp);
616 			*rp++ = 0;
617 		}
618 		return (EMSGSIZE);
619 	}
620 	for (i = 0; i < newfds; i++) {
621 		if (fdalloc(p, 0, &f))
622 			panic("unp_externalize");
623 		fp = *rp;
624 		p->p_fd->fd_ofiles[f] = fp;
625 		fp->f_msgcount--;
626 		unp_rights--;
627 		*(int *)rp++ = f;
628 	}
629 	return (0);
630 }
631 
632 int
633 unp_internalize(control, p)
634 	struct mbuf *control;
635 	struct proc *p;
636 {
637 	struct filedesc *fdp = p->p_fd;
638 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
639 	register struct file **rp;
640 	register struct file *fp;
641 	register int i, fd;
642 	int oldfds;
643 
644 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
645 	    cm->cmsg_len != control->m_len)
646 		return (EINVAL);
647 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
648 	rp = (struct file **)(cm + 1);
649 	for (i = 0; i < oldfds; i++) {
650 		fd = *(int *)rp++;
651 		if ((unsigned)fd >= fdp->fd_nfiles ||
652 		    fdp->fd_ofiles[fd] == NULL)
653 			return (EBADF);
654 	}
655 	rp = (struct file **)(cm + 1);
656 	for (i = 0; i < oldfds; i++) {
657 		fp = fdp->fd_ofiles[*(int *)rp];
658 		*rp++ = fp;
659 		fp->f_count++;
660 		fp->f_msgcount++;
661 		unp_rights++;
662 	}
663 	return (0);
664 }
665 
666 int	unp_defer, unp_gcing;
667 extern	struct domain unixdomain;
668 
669 void
670 unp_gc()
671 {
672 	register struct file *fp, *nextfp;
673 	register struct socket *so;
674 	struct file **extra_ref, **fpp;
675 	int nunref, i;
676 
677 	if (unp_gcing)
678 		return;
679 	unp_gcing = 1;
680 	unp_defer = 0;
681 	for (fp = filehead; fp; fp = fp->f_filef)
682 		fp->f_flag &= ~(FMARK|FDEFER);
683 	do {
684 		for (fp = filehead; fp; fp = fp->f_filef) {
685 			if (fp->f_count == 0)
686 				continue;
687 			if (fp->f_flag & FDEFER) {
688 				fp->f_flag &= ~FDEFER;
689 				unp_defer--;
690 			} else {
691 				if (fp->f_flag & FMARK)
692 					continue;
693 				if (fp->f_count == fp->f_msgcount)
694 					continue;
695 				fp->f_flag |= FMARK;
696 			}
697 			if (fp->f_type != DTYPE_SOCKET ||
698 			    (so = (struct socket *)fp->f_data) == 0)
699 				continue;
700 			if (so->so_proto->pr_domain != &unixdomain ||
701 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
702 				continue;
703 #ifdef notdef
704 			if (so->so_rcv.sb_flags & SB_LOCK) {
705 				/*
706 				 * This is problematical; it's not clear
707 				 * we need to wait for the sockbuf to be
708 				 * unlocked (on a uniprocessor, at least),
709 				 * and it's also not clear what to do
710 				 * if sbwait returns an error due to receipt
711 				 * of a signal.  If sbwait does return
712 				 * an error, we'll go into an infinite
713 				 * loop.  Delete all of this for now.
714 				 */
715 				(void) sbwait(&so->so_rcv);
716 				goto restart;
717 			}
718 #endif
719 			unp_scan(so->so_rcv.sb_mb, unp_mark);
720 		}
721 	} while (unp_defer);
722 	/*
723 	 * We grab an extra reference to each of the file table entries
724 	 * that are not otherwise accessible and then free the rights
725 	 * that are stored in messages on them.
726 	 *
727 	 * The bug in the orginal code is a little tricky, so I'll describe
728 	 * what's wrong with it here.
729 	 *
730 	 * It is incorrect to simply unp_discard each entry for f_msgcount
731 	 * times -- consider the case of sockets A and B that contain
732 	 * references to each other.  On a last close of some other socket,
733 	 * we trigger a gc since the number of outstanding rights (unp_rights)
734 	 * is non-zero.  If during the sweep phase the gc code un_discards,
735 	 * we end up doing a (full) closef on the descriptor.  A closef on A
736 	 * results in the following chain.  Closef calls soo_close, which
737 	 * calls soclose.   Soclose calls first (through the switch
738 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
739 	 * returns because the previous instance had set unp_gcing, and
740 	 * we return all the way back to soclose, which marks the socket
741 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
742 	 * to free up the rights that are queued in messages on the socket A,
743 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
744 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
745 	 * instance of unp_discard just calls closef on B.
746 	 *
747 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
748 	 * which results in another closef on A.  Unfortunately, A is already
749 	 * being closed, and the descriptor has already been marked with
750 	 * SS_NOFDREF, and soclose panics at this point.
751 	 *
752 	 * Here, we first take an extra reference to each inaccessible
753 	 * descriptor.  Then, we call sorflush ourself, since we know
754 	 * it is a Unix domain socket anyhow.  After we destroy all the
755 	 * rights carried in messages, we do a last closef to get rid
756 	 * of our extra reference.  This is the last close, and the
757 	 * unp_detach etc will shut down the socket.
758 	 *
759 	 * 91/09/19, bsy@cs.cmu.edu
760 	 */
761 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
762 	for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) {
763 		nextfp = fp->f_filef;
764 		if (fp->f_count == 0)
765 			continue;
766 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
767 			*fpp++ = fp;
768 			nunref++;
769 			fp->f_count++;
770 		}
771 	}
772 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
773 		sorflush((struct socket *)(*fpp)->f_data);
774 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
775 		closef(*fpp);
776 	free((caddr_t)extra_ref, M_FILE);
777 	unp_gcing = 0;
778 }
779 
780 void
781 unp_dispose(m)
782 	struct mbuf *m;
783 {
784 
785 	if (m)
786 		unp_scan(m, unp_discard);
787 }
788 
789 void
790 unp_scan(m0, op)
791 	register struct mbuf *m0;
792 	void (*op) __P((struct file *));
793 {
794 	register struct mbuf *m;
795 	register struct file **rp;
796 	register struct cmsghdr *cm;
797 	register int i;
798 	int qfds;
799 
800 	while (m0) {
801 		for (m = m0; m; m = m->m_next)
802 			if (m->m_type == MT_CONTROL &&
803 			    m->m_len >= sizeof(*cm)) {
804 				cm = mtod(m, struct cmsghdr *);
805 				if (cm->cmsg_level != SOL_SOCKET ||
806 				    cm->cmsg_type != SCM_RIGHTS)
807 					continue;
808 				qfds = (cm->cmsg_len - sizeof *cm)
809 						/ sizeof (struct file *);
810 				rp = (struct file **)(cm + 1);
811 				for (i = 0; i < qfds; i++)
812 					(*op)(*rp++);
813 				break;		/* XXX, but saves time */
814 			}
815 		m0 = m0->m_act;
816 	}
817 }
818 
819 void
820 unp_mark(fp)
821 	struct file *fp;
822 {
823 
824 	if (fp->f_flag & FMARK)
825 		return;
826 	unp_defer++;
827 	fp->f_flag |= (FMARK|FDEFER);
828 }
829 
830 void
831 unp_discard(fp)
832 	struct file *fp;
833 {
834 
835 	fp->f_msgcount--;
836 	unp_rights--;
837 	(void) closef(fp, (struct proc *)NULL);
838 }
839