xref: /freebsd/sys/kern/uipc_usrreq.c (revision a316b26e50bbed7cf655fbba726ab87d8ab7599d)
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  *	@(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
34  * $Id: uipc_usrreq.c,v 1.4 1994/09/28 19:55:10 phk 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_UNIX };
63 ino_t	unp_ino;			/* prototype for fake inode numbers */
64 
65 /*ARGSUSED*/
66 int
67 uipc_usrreq(so, req, m, nam, control)
68 	struct socket *so;
69 	int req;
70 	struct mbuf *m, *nam, *control;
71 {
72 	struct unpcb *unp = sotounpcb(so);
73 	register struct socket *so2;
74 	register int error = 0;
75 	struct proc *p = curproc;	/* XXX */
76 
77 	if (req == PRU_CONTROL)
78 		return (EOPNOTSUPP);
79 	if (req != PRU_SEND && control && control->m_len) {
80 		error = EOPNOTSUPP;
81 		goto release;
82 	}
83 	if (unp == 0 && req != PRU_ATTACH) {
84 		error = EINVAL;
85 		goto release;
86 	}
87 	switch (req) {
88 
89 	case PRU_ATTACH:
90 		if (unp) {
91 			error = EISCONN;
92 			break;
93 		}
94 		error = unp_attach(so);
95 		break;
96 
97 	case PRU_DETACH:
98 		unp_detach(unp);
99 		break;
100 
101 	case PRU_BIND:
102 		error = unp_bind(unp, nam, p);
103 		break;
104 
105 	case PRU_LISTEN:
106 		if (unp->unp_vnode == 0)
107 			error = EINVAL;
108 		break;
109 
110 	case PRU_CONNECT:
111 		error = unp_connect(so, nam, p);
112 		break;
113 
114 	case PRU_CONNECT2:
115 		error = unp_connect2(so, (struct socket *)nam);
116 		break;
117 
118 	case PRU_DISCONNECT:
119 		unp_disconnect(unp);
120 		break;
121 
122 	case PRU_ACCEPT:
123 		/*
124 		 * Pass back name of connected socket,
125 		 * if it was bound and we are still connected
126 		 * (our peer may have closed already!).
127 		 */
128 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
129 			nam->m_len = unp->unp_conn->unp_addr->m_len;
130 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
131 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
132 		} else {
133 			nam->m_len = sizeof(sun_noname);
134 			*(mtod(nam, struct sockaddr *)) = sun_noname;
135 		}
136 		break;
137 
138 	case PRU_SHUTDOWN:
139 		socantsendmore(so);
140 		unp_shutdown(unp);
141 		break;
142 
143 	case PRU_RCVD:
144 		switch (so->so_type) {
145 
146 		case SOCK_DGRAM:
147 			panic("uipc 1");
148 			/*NOTREACHED*/
149 
150 		case SOCK_STREAM:
151 #define	rcv (&so->so_rcv)
152 #define snd (&so2->so_snd)
153 			if (unp->unp_conn == 0)
154 				break;
155 			so2 = unp->unp_conn->unp_socket;
156 			/*
157 			 * Adjust backpressure on sender
158 			 * and wakeup any waiting to write.
159 			 */
160 			snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
161 			unp->unp_mbcnt = rcv->sb_mbcnt;
162 			snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
163 			unp->unp_cc = rcv->sb_cc;
164 			sowwakeup(so2);
165 #undef snd
166 #undef rcv
167 			break;
168 
169 		default:
170 			panic("uipc 2");
171 		}
172 		break;
173 
174 	case PRU_SEND:
175 		if (control && (error = unp_internalize(control, p)))
176 			break;
177 		switch (so->so_type) {
178 
179 		case SOCK_DGRAM: {
180 			struct sockaddr *from;
181 
182 			if (nam) {
183 				if (unp->unp_conn) {
184 					error = EISCONN;
185 					break;
186 				}
187 				error = unp_connect(so, nam, p);
188 				if (error)
189 					break;
190 			} else {
191 				if (unp->unp_conn == 0) {
192 					error = ENOTCONN;
193 					break;
194 				}
195 			}
196 			so2 = unp->unp_conn->unp_socket;
197 			if (unp->unp_addr)
198 				from = mtod(unp->unp_addr, struct sockaddr *);
199 			else
200 				from = &sun_noname;
201 			if (sbappendaddr(&so2->so_rcv, from, m, control)) {
202 				sorwakeup(so2);
203 				m = 0;
204 				control = 0;
205 			} else
206 				error = ENOBUFS;
207 			if (nam)
208 				unp_disconnect(unp);
209 			break;
210 		}
211 
212 		case SOCK_STREAM:
213 #define	rcv (&so2->so_rcv)
214 #define	snd (&so->so_snd)
215 			if (so->so_state & SS_CANTSENDMORE) {
216 				error = EPIPE;
217 				break;
218 			}
219 			if (unp->unp_conn == 0)
220 				panic("uipc 3");
221 			so2 = unp->unp_conn->unp_socket;
222 			/*
223 			 * Send to paired receive port, and then reduce
224 			 * send buffer hiwater marks to maintain backpressure.
225 			 * Wake up readers.
226 			 */
227 			if (control) {
228 				if (sbappendcontrol(rcv, m, control))
229 					control = 0;
230 			} else
231 				sbappend(rcv, m);
232 			snd->sb_mbmax -=
233 			    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
234 			unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
235 			snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
236 			unp->unp_conn->unp_cc = rcv->sb_cc;
237 			sorwakeup(so2);
238 			m = 0;
239 #undef snd
240 #undef rcv
241 			break;
242 
243 		default:
244 			panic("uipc 4");
245 		}
246 		break;
247 
248 	case PRU_ABORT:
249 		unp_drop(unp, ECONNABORTED);
250 		break;
251 
252 	case PRU_SENSE:
253 		((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
254 		if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
255 			so2 = unp->unp_conn->unp_socket;
256 			((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
257 		}
258 		((struct stat *) m)->st_dev = NODEV;
259 		if (unp->unp_ino == 0)
260 			unp->unp_ino = unp_ino++;
261 		((struct stat *) m)->st_ino = unp->unp_ino;
262 		return (0);
263 
264 	case PRU_RCVOOB:
265 		return (EOPNOTSUPP);
266 
267 	case PRU_SENDOOB:
268 		error = EOPNOTSUPP;
269 		break;
270 
271 	case PRU_SOCKADDR:
272 		if (unp->unp_addr) {
273 			nam->m_len = unp->unp_addr->m_len;
274 			bcopy(mtod(unp->unp_addr, caddr_t),
275 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
276 		} else
277 			nam->m_len = 0;
278 		break;
279 
280 	case PRU_PEERADDR:
281 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
282 			nam->m_len = unp->unp_conn->unp_addr->m_len;
283 			bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
284 			    mtod(nam, caddr_t), (unsigned)nam->m_len);
285 		} else
286 			nam->m_len = 0;
287 		break;
288 
289 	case PRU_SLOWTIMO:
290 		break;
291 
292 	default:
293 		panic("piusrreq");
294 	}
295 release:
296 	if (control)
297 		m_freem(control);
298 	if (m)
299 		m_freem(m);
300 	return (error);
301 }
302 
303 /*
304  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
305  * for stream sockets, although the total for sender and receiver is
306  * actually only PIPSIZ.
307  * Datagram sockets really use the sendspace as the maximum datagram size,
308  * and don't really want to reserve the sendspace.  Their recvspace should
309  * be large enough for at least one max-size datagram plus address.
310  */
311 #define	PIPSIZ	4096
312 u_long	unpst_sendspace = PIPSIZ;
313 u_long	unpst_recvspace = PIPSIZ;
314 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
315 u_long	unpdg_recvspace = 4*1024;
316 
317 int	unp_rights;			/* file descriptors in flight */
318 
319 int
320 unp_attach(so)
321 	struct socket *so;
322 {
323 	register struct mbuf *m;
324 	register struct unpcb *unp;
325 	int error;
326 
327 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
328 		switch (so->so_type) {
329 
330 		case SOCK_STREAM:
331 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
332 			break;
333 
334 		case SOCK_DGRAM:
335 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
336 			break;
337 
338 		default:
339 			panic("unp_attach");
340 		}
341 		if (error)
342 			return (error);
343 	}
344 	m = m_getclr(M_DONTWAIT, MT_PCB);
345 	if (m == NULL)
346 		return (ENOBUFS);
347 	unp = mtod(m, struct unpcb *);
348 	so->so_pcb = (caddr_t)unp;
349 	unp->unp_socket = so;
350 	return (0);
351 }
352 
353 void
354 unp_detach(unp)
355 	register struct unpcb *unp;
356 {
357 
358 	if (unp->unp_vnode) {
359 		unp->unp_vnode->v_socket = 0;
360 		vrele(unp->unp_vnode);
361 		unp->unp_vnode = 0;
362 	}
363 	if (unp->unp_conn)
364 		unp_disconnect(unp);
365 	while (unp->unp_refs)
366 		unp_drop(unp->unp_refs, ECONNRESET);
367 	soisdisconnected(unp->unp_socket);
368 	unp->unp_socket->so_pcb = 0;
369 	m_freem(unp->unp_addr);
370 	(void) m_free(dtom(unp));
371 	if (unp_rights) {
372 		/*
373 		 * Normally the receive buffer is flushed later,
374 		 * in sofree, but if our receive buffer holds references
375 		 * to descriptors that are now garbage, we will dispose
376 		 * of those descriptor references after the garbage collector
377 		 * gets them (resulting in a "panic: closef: count < 0").
378 		 */
379 		sorflush(unp->unp_socket);
380 		unp_gc();
381 	}
382 }
383 
384 int
385 unp_bind(unp, nam, p)
386 	struct unpcb *unp;
387 	struct mbuf *nam;
388 	struct proc *p;
389 {
390 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
391 	register struct vnode *vp;
392 	struct vattr vattr;
393 	int error;
394 	struct nameidata nd;
395 
396 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
397 		soun->sun_path, p);
398 	if (unp->unp_vnode != NULL)
399 		return (EINVAL);
400 	if (nam->m_len == MLEN) {
401 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
402 			return (EINVAL);
403 	} else
404 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
405 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
406 	error = namei(&nd);
407 	if (error)
408 		return (error);
409 	vp = nd.ni_vp;
410 	if (vp != NULL) {
411 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
412 		if (nd.ni_dvp == vp)
413 			vrele(nd.ni_dvp);
414 		else
415 			vput(nd.ni_dvp);
416 		vrele(vp);
417 		return (EADDRINUSE);
418 	}
419 	VATTR_NULL(&vattr);
420 	vattr.va_type = VSOCK;
421 	vattr.va_mode = ACCESSPERMS;
422 	LEASE_CHECK(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
423 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
424 	if (error)
425 		return (error);
426 	vp = nd.ni_vp;
427 	vp->v_socket = unp->unp_socket;
428 	unp->unp_vnode = vp;
429 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
430 	VOP_UNLOCK(vp);
431 	return (0);
432 }
433 
434 int
435 unp_connect(so, nam, p)
436 	struct socket *so;
437 	struct mbuf *nam;
438 	struct proc *p;
439 {
440 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
441 	register struct vnode *vp;
442 	register struct socket *so2, *so3;
443 	struct unpcb *unp2, *unp3;
444 	int error;
445 	struct nameidata nd;
446 
447 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
448 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
449 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
450 			return (EMSGSIZE);
451 	} else
452 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
453 	error = namei(&nd);
454 	if (error)
455 		return (error);
456 	vp = nd.ni_vp;
457 	if (vp->v_type != VSOCK) {
458 		error = ENOTSOCK;
459 		goto bad;
460 	}
461 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p);
462 	if (error)
463 		goto bad;
464 	so2 = vp->v_socket;
465 	if (so2 == 0) {
466 		error = ECONNREFUSED;
467 		goto bad;
468 	}
469 	if (so->so_type != so2->so_type) {
470 		error = EPROTOTYPE;
471 		goto bad;
472 	}
473 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
474 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
475 		    (so3 = sonewconn(so2, 0)) == 0) {
476 			error = ECONNREFUSED;
477 			goto bad;
478 		}
479 		unp2 = sotounpcb(so2);
480 		unp3 = sotounpcb(so3);
481 		if (unp2->unp_addr)
482 			unp3->unp_addr =
483 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
484 		so2 = so3;
485 	}
486 	error = unp_connect2(so, so2);
487 bad:
488 	vput(vp);
489 	return (error);
490 }
491 
492 int
493 unp_connect2(so, so2)
494 	register struct socket *so;
495 	register struct socket *so2;
496 {
497 	register struct unpcb *unp = sotounpcb(so);
498 	register struct unpcb *unp2;
499 
500 	if (so2->so_type != so->so_type)
501 		return (EPROTOTYPE);
502 	unp2 = sotounpcb(so2);
503 	unp->unp_conn = unp2;
504 	switch (so->so_type) {
505 
506 	case SOCK_DGRAM:
507 		unp->unp_nextref = unp2->unp_refs;
508 		unp2->unp_refs = unp;
509 		soisconnected(so);
510 		break;
511 
512 	case SOCK_STREAM:
513 		unp2->unp_conn = unp;
514 		soisconnected(so);
515 		soisconnected(so2);
516 		break;
517 
518 	default:
519 		panic("unp_connect2");
520 	}
521 	return (0);
522 }
523 
524 void
525 unp_disconnect(unp)
526 	struct unpcb *unp;
527 {
528 	register struct unpcb *unp2 = unp->unp_conn;
529 
530 	if (unp2 == 0)
531 		return;
532 	unp->unp_conn = 0;
533 	switch (unp->unp_socket->so_type) {
534 
535 	case SOCK_DGRAM:
536 		if (unp2->unp_refs == unp)
537 			unp2->unp_refs = unp->unp_nextref;
538 		else {
539 			unp2 = unp2->unp_refs;
540 			for (;;) {
541 				if (unp2 == 0)
542 					panic("unp_disconnect");
543 				if (unp2->unp_nextref == unp)
544 					break;
545 				unp2 = unp2->unp_nextref;
546 			}
547 			unp2->unp_nextref = unp->unp_nextref;
548 		}
549 		unp->unp_nextref = 0;
550 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
551 		break;
552 
553 	case SOCK_STREAM:
554 		soisdisconnected(unp->unp_socket);
555 		unp2->unp_conn = 0;
556 		soisdisconnected(unp2->unp_socket);
557 		break;
558 	}
559 }
560 
561 #ifdef notdef
562 void
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 void
601 unp_drain()
602 {
603 
604 }
605 #endif
606 
607 int
608 unp_externalize(rights)
609 	struct mbuf *rights;
610 {
611 	struct proc *p = curproc;		/* XXX */
612 	register int i;
613 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
614 	register struct file **rp = (struct file **)(cm + 1);
615 	register struct file *fp;
616 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
617 	int f;
618 
619 	if (!fdavail(p, newfds)) {
620 		for (i = 0; i < newfds; i++) {
621 			fp = *rp;
622 			unp_discard(fp);
623 			*rp++ = 0;
624 		}
625 		return (EMSGSIZE);
626 	}
627 	for (i = 0; i < newfds; i++) {
628 		if (fdalloc(p, 0, &f))
629 			panic("unp_externalize");
630 		fp = *rp;
631 		p->p_fd->fd_ofiles[f] = fp;
632 		fp->f_msgcount--;
633 		unp_rights--;
634 		*(int *)rp++ = f;
635 	}
636 	return (0);
637 }
638 
639 int
640 unp_internalize(control, p)
641 	struct mbuf *control;
642 	struct proc *p;
643 {
644 	struct filedesc *fdp = p->p_fd;
645 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
646 	register struct file **rp;
647 	register struct file *fp;
648 	register int i, fd;
649 	int oldfds;
650 
651 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
652 	    cm->cmsg_len != control->m_len)
653 		return (EINVAL);
654 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
655 	rp = (struct file **)(cm + 1);
656 	for (i = 0; i < oldfds; i++) {
657 		fd = *(int *)rp++;
658 		if ((unsigned)fd >= fdp->fd_nfiles ||
659 		    fdp->fd_ofiles[fd] == NULL)
660 			return (EBADF);
661 	}
662 	rp = (struct file **)(cm + 1);
663 	for (i = 0; i < oldfds; i++) {
664 		fp = fdp->fd_ofiles[*(int *)rp];
665 		*rp++ = fp;
666 		fp->f_count++;
667 		fp->f_msgcount++;
668 		unp_rights++;
669 	}
670 	return (0);
671 }
672 
673 int	unp_defer, unp_gcing;
674 extern	struct domain unixdomain;
675 
676 void
677 unp_gc()
678 {
679 	register struct file *fp, *nextfp;
680 	register struct socket *so;
681 	struct file **extra_ref, **fpp;
682 	int nunref, i;
683 
684 	if (unp_gcing)
685 		return;
686 	unp_gcing = 1;
687 	unp_defer = 0;
688 	for (fp = filehead; fp; fp = fp->f_filef)
689 		fp->f_flag &= ~(FMARK|FDEFER);
690 	do {
691 		for (fp = filehead; fp; fp = fp->f_filef) {
692 			if (fp->f_count == 0)
693 				continue;
694 			if (fp->f_flag & FDEFER) {
695 				fp->f_flag &= ~FDEFER;
696 				unp_defer--;
697 			} else {
698 				if (fp->f_flag & FMARK)
699 					continue;
700 				if (fp->f_count == fp->f_msgcount)
701 					continue;
702 				fp->f_flag |= FMARK;
703 			}
704 			if (fp->f_type != DTYPE_SOCKET ||
705 			    (so = (struct socket *)fp->f_data) == 0)
706 				continue;
707 			if (so->so_proto->pr_domain != &unixdomain ||
708 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
709 				continue;
710 #ifdef notdef
711 			if (so->so_rcv.sb_flags & SB_LOCK) {
712 				/*
713 				 * This is problematical; it's not clear
714 				 * we need to wait for the sockbuf to be
715 				 * unlocked (on a uniprocessor, at least),
716 				 * and it's also not clear what to do
717 				 * if sbwait returns an error due to receipt
718 				 * of a signal.  If sbwait does return
719 				 * an error, we'll go into an infinite
720 				 * loop.  Delete all of this for now.
721 				 */
722 				(void) sbwait(&so->so_rcv);
723 				goto restart;
724 			}
725 #endif
726 			unp_scan(so->so_rcv.sb_mb, unp_mark);
727 		}
728 	} while (unp_defer);
729 	/*
730 	 * We grab an extra reference to each of the file table entries
731 	 * that are not otherwise accessible and then free the rights
732 	 * that are stored in messages on them.
733 	 *
734 	 * The bug in the orginal code is a little tricky, so I'll describe
735 	 * what's wrong with it here.
736 	 *
737 	 * It is incorrect to simply unp_discard each entry for f_msgcount
738 	 * times -- consider the case of sockets A and B that contain
739 	 * references to each other.  On a last close of some other socket,
740 	 * we trigger a gc since the number of outstanding rights (unp_rights)
741 	 * is non-zero.  If during the sweep phase the gc code un_discards,
742 	 * we end up doing a (full) closef on the descriptor.  A closef on A
743 	 * results in the following chain.  Closef calls soo_close, which
744 	 * calls soclose.   Soclose calls first (through the switch
745 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
746 	 * returns because the previous instance had set unp_gcing, and
747 	 * we return all the way back to soclose, which marks the socket
748 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
749 	 * to free up the rights that are queued in messages on the socket A,
750 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
751 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
752 	 * instance of unp_discard just calls closef on B.
753 	 *
754 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
755 	 * which results in another closef on A.  Unfortunately, A is already
756 	 * being closed, and the descriptor has already been marked with
757 	 * SS_NOFDREF, and soclose panics at this point.
758 	 *
759 	 * Here, we first take an extra reference to each inaccessible
760 	 * descriptor.  Then, we call sorflush ourself, since we know
761 	 * it is a Unix domain socket anyhow.  After we destroy all the
762 	 * rights carried in messages, we do a last closef to get rid
763 	 * of our extra reference.  This is the last close, and the
764 	 * unp_detach etc will shut down the socket.
765 	 *
766 	 * 91/09/19, bsy@cs.cmu.edu
767 	 */
768 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
769 	for (nunref = 0, fp = filehead, fpp = extra_ref; fp; fp = nextfp) {
770 		nextfp = fp->f_filef;
771 		if (fp->f_count == 0)
772 			continue;
773 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
774 			*fpp++ = fp;
775 			nunref++;
776 			fp->f_count++;
777 		}
778 	}
779 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
780 		sorflush((struct socket *)(*fpp)->f_data);
781 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
782 		closef(*fpp,(struct proc*) NULL);
783 	free((caddr_t)extra_ref, M_FILE);
784 	unp_gcing = 0;
785 }
786 
787 void
788 unp_dispose(m)
789 	struct mbuf *m;
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)(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