xref: /dragonfly/sys/kern/uipc_usrreq.c (revision 6b5c5d0d)
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  * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
35  * $DragonFly: src/sys/kern/uipc_usrreq.c,v 1.37 2008/01/06 16:55:51 swildner Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/domain.h>
42 #include <sys/fcntl.h>
43 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
44 #include <sys/proc.h>
45 #include <sys/file.h>
46 #include <sys/filedesc.h>
47 #include <sys/mbuf.h>
48 #include <sys/nlookup.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/resourcevar.h>
53 #include <sys/stat.h>
54 #include <sys/mount.h>
55 #include <sys/sysctl.h>
56 #include <sys/un.h>
57 #include <sys/unpcb.h>
58 #include <sys/vnode.h>
59 #include <sys/file2.h>
60 #include <sys/spinlock2.h>
61 
62 #include <vm/vm_zone.h>
63 
64 static	struct vm_zone *unp_zone;
65 static	unp_gen_t unp_gencnt;
66 static	u_int unp_count;
67 
68 static	struct unp_head unp_shead, unp_dhead;
69 
70 /*
71  * Unix communications domain.
72  *
73  * TODO:
74  *	SEQPACKET, RDM
75  *	rethink name space problems
76  *	need a proper out-of-band
77  *	lock pushdown
78  */
79 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
80 static ino_t	unp_ino;		/* prototype for fake inode numbers */
81 
82 static int     unp_attach (struct socket *, struct pru_attach_info *);
83 static void    unp_detach (struct unpcb *);
84 static int     unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
85 static int     unp_connect (struct socket *,struct sockaddr *,
86 				struct thread *);
87 static void    unp_disconnect (struct unpcb *);
88 static void    unp_shutdown (struct unpcb *);
89 static void    unp_drop (struct unpcb *, int);
90 static void    unp_gc (void);
91 static int     unp_gc_clearmarks(struct file *, void *);
92 static int     unp_gc_checkmarks(struct file *, void *);
93 static int     unp_gc_checkrefs(struct file *, void *);
94 static void    unp_scan (struct mbuf *, void (*)(struct file *, void *),
95 				void *data);
96 static void    unp_mark (struct file *, void *data);
97 static void    unp_discard (struct file *, void *);
98 static int     unp_internalize (struct mbuf *, struct thread *);
99 static int     unp_listen (struct unpcb *, struct thread *);
100 
101 static int
102 uipc_abort(struct socket *so)
103 {
104 	struct unpcb *unp = so->so_pcb;
105 
106 	if (unp == NULL)
107 		return EINVAL;
108 	unp_drop(unp, ECONNABORTED);
109 	unp_detach(unp);
110 	sofree(so);
111 	return 0;
112 }
113 
114 static int
115 uipc_accept(struct socket *so, struct sockaddr **nam)
116 {
117 	struct unpcb *unp = so->so_pcb;
118 
119 	if (unp == NULL)
120 		return EINVAL;
121 
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 = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
129 	} else {
130 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname);
131 	}
132 	return 0;
133 }
134 
135 static int
136 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai)
137 {
138 	struct unpcb *unp = so->so_pcb;
139 
140 	if (unp != NULL)
141 		return EISCONN;
142 	return unp_attach(so, ai);
143 }
144 
145 static int
146 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
147 {
148 	struct unpcb *unp = so->so_pcb;
149 
150 	if (unp == NULL)
151 		return EINVAL;
152 	return unp_bind(unp, nam, td);
153 }
154 
155 static int
156 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
157 {
158 	struct unpcb *unp = so->so_pcb;
159 
160 	if (unp == NULL)
161 		return EINVAL;
162 	return unp_connect(so, nam, td);
163 }
164 
165 static int
166 uipc_connect2(struct socket *so1, struct socket *so2)
167 {
168 	struct unpcb *unp = so1->so_pcb;
169 
170 	if (unp == NULL)
171 		return EINVAL;
172 
173 	return unp_connect2(so1, so2);
174 }
175 
176 /* control is EOPNOTSUPP */
177 
178 static int
179 uipc_detach(struct socket *so)
180 {
181 	struct unpcb *unp = so->so_pcb;
182 
183 	if (unp == NULL)
184 		return EINVAL;
185 
186 	unp_detach(unp);
187 	return 0;
188 }
189 
190 static int
191 uipc_disconnect(struct socket *so)
192 {
193 	struct unpcb *unp = so->so_pcb;
194 
195 	if (unp == NULL)
196 		return EINVAL;
197 	unp_disconnect(unp);
198 	return 0;
199 }
200 
201 static int
202 uipc_listen(struct socket *so, struct thread *td)
203 {
204 	struct unpcb *unp = so->so_pcb;
205 
206 	if (unp == NULL || unp->unp_vnode == NULL)
207 		return EINVAL;
208 	return unp_listen(unp, td);
209 }
210 
211 static int
212 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
213 {
214 	struct unpcb *unp = so->so_pcb;
215 
216 	if (unp == NULL)
217 		return EINVAL;
218 	if (unp->unp_conn && unp->unp_conn->unp_addr)
219 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
220 	else {
221 		/*
222 		 * XXX: It seems that this test always fails even when
223 		 * connection is established.  So, this else clause is
224 		 * added as workaround to return PF_LOCAL sockaddr.
225 		 */
226 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname);
227 	}
228 	return 0;
229 }
230 
231 static int
232 uipc_rcvd(struct socket *so, int flags)
233 {
234 	struct unpcb *unp = so->so_pcb;
235 	struct socket *so2;
236 	u_long newhiwat;
237 
238 	if (unp == NULL)
239 		return EINVAL;
240 	switch (so->so_type) {
241 	case SOCK_DGRAM:
242 		panic("uipc_rcvd DGRAM?");
243 		/*NOTREACHED*/
244 
245 	case SOCK_STREAM:
246 		if (unp->unp_conn == NULL)
247 			break;
248 		so2 = unp->unp_conn->unp_socket;
249 		/*
250 		 * Adjust backpressure on sender
251 		 * and wakeup any waiting to write.
252 		 */
253 		so2->so_snd.ssb_mbmax += unp->unp_mbcnt - so->so_rcv.ssb_mbcnt;
254 		unp->unp_mbcnt = so->so_rcv.ssb_mbcnt;
255 		newhiwat =
256 		    so2->so_snd.ssb_hiwat + unp->unp_cc - so->so_rcv.ssb_cc;
257 		chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.ssb_hiwat,
258 		    newhiwat, RLIM_INFINITY);
259 		unp->unp_cc = so->so_rcv.ssb_cc;
260 		sowwakeup(so2);
261 		break;
262 
263 	default:
264 		panic("uipc_rcvd unknown socktype");
265 	}
266 	return 0;
267 }
268 
269 /* pru_rcvoob is EOPNOTSUPP */
270 
271 static int
272 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
273 	  struct mbuf *control, struct thread *td)
274 {
275 	int error = 0;
276 	struct unpcb *unp = so->so_pcb;
277 	struct socket *so2;
278 	u_long newhiwat;
279 
280 	if (unp == NULL) {
281 		error = EINVAL;
282 		goto release;
283 	}
284 	if (flags & PRUS_OOB) {
285 		error = EOPNOTSUPP;
286 		goto release;
287 	}
288 
289 	if (control && (error = unp_internalize(control, td)))
290 		goto release;
291 
292 	switch (so->so_type) {
293 	case SOCK_DGRAM:
294 	{
295 		struct sockaddr *from;
296 
297 		if (nam) {
298 			if (unp->unp_conn) {
299 				error = EISCONN;
300 				break;
301 			}
302 			error = unp_connect(so, nam, td);
303 			if (error)
304 				break;
305 		} else {
306 			if (unp->unp_conn == NULL) {
307 				error = ENOTCONN;
308 				break;
309 			}
310 		}
311 		so2 = unp->unp_conn->unp_socket;
312 		if (unp->unp_addr)
313 			from = (struct sockaddr *)unp->unp_addr;
314 		else
315 			from = &sun_noname;
316 		if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
317 			sorwakeup(so2);
318 			m = NULL;
319 			control = NULL;
320 		} else {
321 			error = ENOBUFS;
322 		}
323 		if (nam)
324 			unp_disconnect(unp);
325 		break;
326 	}
327 
328 	case SOCK_STREAM:
329 		/* Connect if not connected yet. */
330 		/*
331 		 * Note: A better implementation would complain
332 		 * if not equal to the peer's address.
333 		 */
334 		if (!(so->so_state & SS_ISCONNECTED)) {
335 			if (nam) {
336 				error = unp_connect(so, nam, td);
337 				if (error)
338 					break;	/* XXX */
339 			} else {
340 				error = ENOTCONN;
341 				break;
342 			}
343 		}
344 
345 		if (so->so_state & SS_CANTSENDMORE) {
346 			error = EPIPE;
347 			break;
348 		}
349 		if (unp->unp_conn == NULL)
350 			panic("uipc_send connected but no connection?");
351 		so2 = unp->unp_conn->unp_socket;
352 		/*
353 		 * Send to paired receive port, and then reduce
354 		 * send buffer hiwater marks to maintain backpressure.
355 		 * Wake up readers.
356 		 */
357 		if (control) {
358 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
359 				control = NULL;
360 				m = NULL;
361 			}
362 		} else {
363 			sbappend(&so2->so_rcv.sb, m);
364 			m = NULL;
365 		}
366 		so->so_snd.ssb_mbmax -=
367 			so2->so_rcv.ssb_mbcnt - unp->unp_conn->unp_mbcnt;
368 		unp->unp_conn->unp_mbcnt = so2->so_rcv.ssb_mbcnt;
369 		newhiwat = so->so_snd.ssb_hiwat -
370 		    (so2->so_rcv.ssb_cc - unp->unp_conn->unp_cc);
371 		chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.ssb_hiwat,
372 		    newhiwat, RLIM_INFINITY);
373 		unp->unp_conn->unp_cc = so2->so_rcv.ssb_cc;
374 		sorwakeup(so2);
375 		break;
376 
377 	default:
378 		panic("uipc_send unknown socktype");
379 	}
380 
381 	/*
382 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
383 	 */
384 	if (flags & PRUS_EOF) {
385 		socantsendmore(so);
386 		unp_shutdown(unp);
387 	}
388 
389 	if (control && error != 0)
390 		unp_dispose(control);
391 
392 release:
393 	if (control)
394 		m_freem(control);
395 	if (m)
396 		m_freem(m);
397 	return error;
398 }
399 
400 static int
401 uipc_sense(struct socket *so, struct stat *sb)
402 {
403 	struct unpcb *unp = so->so_pcb;
404 	struct socket *so2;
405 
406 	if (unp == NULL)
407 		return EINVAL;
408 	sb->st_blksize = so->so_snd.ssb_hiwat;
409 	if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
410 		so2 = unp->unp_conn->unp_socket;
411 		sb->st_blksize += so2->so_rcv.ssb_cc;
412 	}
413 	sb->st_dev = NOUDEV;
414 	if (unp->unp_ino == 0)		/* make up a non-zero inode number */
415 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
416 	sb->st_ino = unp->unp_ino;
417 	return (0);
418 }
419 
420 static int
421 uipc_shutdown(struct socket *so)
422 {
423 	struct unpcb *unp = so->so_pcb;
424 
425 	if (unp == NULL)
426 		return EINVAL;
427 	socantsendmore(so);
428 	unp_shutdown(unp);
429 	return 0;
430 }
431 
432 static int
433 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
434 {
435 	struct unpcb *unp = so->so_pcb;
436 
437 	if (unp == NULL)
438 		return EINVAL;
439 	if (unp->unp_addr)
440 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_addr);
441 	return 0;
442 }
443 
444 struct pr_usrreqs uipc_usrreqs = {
445 	.pru_abort = uipc_abort,
446 	.pru_accept = uipc_accept,
447 	.pru_attach = uipc_attach,
448 	.pru_bind = uipc_bind,
449 	.pru_connect = uipc_connect,
450 	.pru_connect2 = uipc_connect2,
451 	.pru_control = pru_control_notsupp,
452 	.pru_detach = uipc_detach,
453 	.pru_disconnect = uipc_disconnect,
454 	.pru_listen = uipc_listen,
455 	.pru_peeraddr = uipc_peeraddr,
456 	.pru_rcvd = uipc_rcvd,
457 	.pru_rcvoob = pru_rcvoob_notsupp,
458 	.pru_send = uipc_send,
459 	.pru_sense = uipc_sense,
460 	.pru_shutdown = uipc_shutdown,
461 	.pru_sockaddr = uipc_sockaddr,
462 	.pru_sosend = sosend,
463 	.pru_soreceive = soreceive,
464 	.pru_sopoll = sopoll
465 };
466 
467 int
468 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
469 {
470 	struct unpcb *unp = so->so_pcb;
471 	int error;
472 
473 	switch (sopt->sopt_dir) {
474 	case SOPT_GET:
475 		switch (sopt->sopt_name) {
476 		case LOCAL_PEERCRED:
477 			if (unp->unp_flags & UNP_HAVEPC)
478 				error = sooptcopyout(sopt, &unp->unp_peercred,
479 				    sizeof(unp->unp_peercred));
480 			else {
481 				if (so->so_type == SOCK_STREAM)
482 					error = ENOTCONN;
483 				else
484 					error = EINVAL;
485 			}
486 			break;
487 		default:
488 			error = EOPNOTSUPP;
489 			break;
490 		}
491 		break;
492 	case SOPT_SET:
493 	default:
494 		error = EOPNOTSUPP;
495 		break;
496 	}
497 	return (error);
498 }
499 
500 /*
501  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
502  * for stream sockets, although the total for sender and receiver is
503  * actually only PIPSIZ.
504  * Datagram sockets really use the sendspace as the maximum datagram size,
505  * and don't really want to reserve the sendspace.  Their recvspace should
506  * be large enough for at least one max-size datagram plus address.
507  */
508 #ifndef PIPSIZ
509 #define	PIPSIZ	8192
510 #endif
511 static u_long	unpst_sendspace = PIPSIZ;
512 static u_long	unpst_recvspace = PIPSIZ;
513 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
514 static u_long	unpdg_recvspace = 4*1024;
515 
516 static int	unp_rights;			/* file descriptors in flight */
517 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin);
518 
519 SYSCTL_DECL(_net_local_stream);
520 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
521 	   &unpst_sendspace, 0, "");
522 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
523 	   &unpst_recvspace, 0, "");
524 
525 SYSCTL_DECL(_net_local_dgram);
526 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
527 	   &unpdg_sendspace, 0, "");
528 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
529 	   &unpdg_recvspace, 0, "");
530 
531 SYSCTL_DECL(_net_local);
532 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
533 
534 static int
535 unp_attach(struct socket *so, struct pru_attach_info *ai)
536 {
537 	struct unpcb *unp;
538 	int error;
539 
540 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
541 		switch (so->so_type) {
542 
543 		case SOCK_STREAM:
544 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
545 					  ai->sb_rlimit);
546 			break;
547 
548 		case SOCK_DGRAM:
549 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
550 					  ai->sb_rlimit);
551 			break;
552 
553 		default:
554 			panic("unp_attach");
555 		}
556 		if (error)
557 			return (error);
558 	}
559 	unp = zalloc(unp_zone);
560 	if (unp == NULL)
561 		return (ENOBUFS);
562 	bzero(unp, sizeof *unp);
563 	unp->unp_gencnt = ++unp_gencnt;
564 	unp_count++;
565 	LIST_INIT(&unp->unp_refs);
566 	unp->unp_socket = so;
567 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
568 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
569 			 : &unp_shead, unp, unp_link);
570 	so->so_pcb = (caddr_t)unp;
571 	return (0);
572 }
573 
574 static void
575 unp_detach(struct unpcb *unp)
576 {
577 	LIST_REMOVE(unp, unp_link);
578 	unp->unp_gencnt = ++unp_gencnt;
579 	--unp_count;
580 	if (unp->unp_vnode) {
581 		unp->unp_vnode->v_socket = NULL;
582 		vrele(unp->unp_vnode);
583 		unp->unp_vnode = NULL;
584 	}
585 	if (unp->unp_conn)
586 		unp_disconnect(unp);
587 	while (!LIST_EMPTY(&unp->unp_refs))
588 		unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
589 	soisdisconnected(unp->unp_socket);
590 	unp->unp_socket->so_pcb = NULL;
591 	if (unp_rights) {
592 		/*
593 		 * Normally the receive buffer is flushed later,
594 		 * in sofree, but if our receive buffer holds references
595 		 * to descriptors that are now garbage, we will dispose
596 		 * of those descriptor references after the garbage collector
597 		 * gets them (resulting in a "panic: closef: count < 0").
598 		 */
599 		sorflush(unp->unp_socket);
600 		unp_gc();
601 	}
602 	if (unp->unp_addr)
603 		kfree(unp->unp_addr, M_SONAME);
604 	zfree(unp_zone, unp);
605 }
606 
607 static int
608 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
609 {
610 	struct proc *p = td->td_proc;
611 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
612 	struct vnode *dvp;
613 	struct vnode *vp;
614 	struct vattr vattr;
615 	int error, namelen;
616 	struct nlookupdata nd;
617 	char buf[SOCK_MAXADDRLEN];
618 
619 	if (unp->unp_vnode != NULL)
620 		return (EINVAL);
621 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
622 	if (namelen <= 0)
623 		return (EINVAL);
624 	strncpy(buf, soun->sun_path, namelen);
625 	buf[namelen] = 0;	/* null-terminate the string */
626 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_LOCKVP|NLC_CREATE);
627 	if (error == 0)
628 		error = nlookup(&nd);
629 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
630 		error = EADDRINUSE;
631 	if (error == 0 && (dvp = nd.nl_nch.ncp->nc_parent->nc_vp) == NULL)
632 		error = EPERM;
633 	if (error)
634 		goto done;
635 
636 	/* vhold(dvp); - DVP can't go away */
637 	VATTR_NULL(&vattr);
638 	vattr.va_type = VSOCK;
639 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
640 	error = VOP_NCREATE(&nd.nl_nch, dvp, &vp, nd.nl_cred, &vattr);
641 	/* vdrop(dvp); */
642 	if (error == 0) {
643 		vp->v_socket = unp->unp_socket;
644 		unp->unp_vnode = vp;
645 		unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
646 		vn_unlock(vp);
647 	}
648 done:
649 	nlookup_done(&nd);
650 	return (error);
651 }
652 
653 static int
654 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
655 {
656 	struct proc *p = td->td_proc;
657 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
658 	struct vnode *vp;
659 	struct socket *so2, *so3;
660 	struct unpcb *unp, *unp2, *unp3;
661 	int error, len;
662 	struct nlookupdata nd;
663 	char buf[SOCK_MAXADDRLEN];
664 
665 	KKASSERT(p);
666 
667 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
668 	if (len <= 0)
669 		return EINVAL;
670 	strncpy(buf, soun->sun_path, len);
671 	buf[len] = 0;
672 
673 	vp = NULL;
674 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
675 	if (error == 0)
676 		error = nlookup(&nd);
677 	if (error == 0)
678 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
679 	nlookup_done(&nd);
680 	if (error)
681 		return (error);
682 
683 	if (vp->v_type != VSOCK) {
684 		error = ENOTSOCK;
685 		goto bad;
686 	}
687 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred);
688 	if (error)
689 		goto bad;
690 	so2 = vp->v_socket;
691 	if (so2 == NULL) {
692 		error = ECONNREFUSED;
693 		goto bad;
694 	}
695 	if (so->so_type != so2->so_type) {
696 		error = EPROTOTYPE;
697 		goto bad;
698 	}
699 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
700 		if (!(so2->so_options & SO_ACCEPTCONN) ||
701 		    (so3 = sonewconn(so2, 0)) == NULL) {
702 			error = ECONNREFUSED;
703 			goto bad;
704 		}
705 		unp = so->so_pcb;
706 		unp2 = so2->so_pcb;
707 		unp3 = so3->so_pcb;
708 		if (unp2->unp_addr)
709 			unp3->unp_addr = (struct sockaddr_un *)
710 				dup_sockaddr((struct sockaddr *)unp2->unp_addr);
711 
712 		/*
713 		 * unp_peercred management:
714 		 *
715 		 * The connecter's (client's) credentials are copied
716 		 * from its process structure at the time of connect()
717 		 * (which is now).
718 		 */
719 		cru2x(p->p_ucred, &unp3->unp_peercred);
720 		unp3->unp_flags |= UNP_HAVEPC;
721 		/*
722 		 * The receiver's (server's) credentials are copied
723 		 * from the unp_peercred member of socket on which the
724 		 * former called listen(); unp_listen() cached that
725 		 * process's credentials at that time so we can use
726 		 * them now.
727 		 */
728 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
729 		    ("unp_connect: listener without cached peercred"));
730 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
731 		    sizeof(unp->unp_peercred));
732 		unp->unp_flags |= UNP_HAVEPC;
733 
734 		so2 = so3;
735 	}
736 	error = unp_connect2(so, so2);
737 bad:
738 	vput(vp);
739 	return (error);
740 }
741 
742 int
743 unp_connect2(struct socket *so, struct socket *so2)
744 {
745 	struct unpcb *unp = so->so_pcb;
746 	struct unpcb *unp2;
747 
748 	if (so2->so_type != so->so_type)
749 		return (EPROTOTYPE);
750 	unp2 = so2->so_pcb;
751 	unp->unp_conn = unp2;
752 	switch (so->so_type) {
753 
754 	case SOCK_DGRAM:
755 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
756 		soisconnected(so);
757 		break;
758 
759 	case SOCK_STREAM:
760 		unp2->unp_conn = unp;
761 		soisconnected(so);
762 		soisconnected(so2);
763 		break;
764 
765 	default:
766 		panic("unp_connect2");
767 	}
768 	return (0);
769 }
770 
771 static void
772 unp_disconnect(struct unpcb *unp)
773 {
774 	struct unpcb *unp2 = unp->unp_conn;
775 
776 	if (unp2 == NULL)
777 		return;
778 
779 	unp->unp_conn = NULL;
780 
781 	switch (unp->unp_socket->so_type) {
782 	case SOCK_DGRAM:
783 		LIST_REMOVE(unp, unp_reflink);
784 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
785 		break;
786 	case SOCK_STREAM:
787 		soisdisconnected(unp->unp_socket);
788 		unp2->unp_conn = NULL;
789 		soisdisconnected(unp2->unp_socket);
790 		break;
791 	}
792 }
793 
794 #ifdef notdef
795 void
796 unp_abort(struct unpcb *unp)
797 {
798 
799 	unp_detach(unp);
800 }
801 #endif
802 
803 static int
804 prison_unpcb(struct thread *td, struct unpcb *unp)
805 {
806 	struct proc *p;
807 
808 	if (td == NULL)
809 		return (0);
810 	if ((p = td->td_proc) == NULL)
811 		return (0);
812 	if (!p->p_ucred->cr_prison)
813 		return (0);
814 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
815 		return (0);
816 	return (1);
817 }
818 
819 static int
820 unp_pcblist(SYSCTL_HANDLER_ARGS)
821 {
822 	int error, i, n;
823 	struct unpcb *unp, **unp_list;
824 	unp_gen_t gencnt;
825 	struct unp_head *head;
826 
827 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
828 
829 	KKASSERT(curproc != NULL);
830 
831 	/*
832 	 * The process of preparing the PCB list is too time-consuming and
833 	 * resource-intensive to repeat twice on every request.
834 	 */
835 	if (req->oldptr == NULL) {
836 		n = unp_count;
837 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
838 		return 0;
839 	}
840 
841 	if (req->newptr != NULL)
842 		return EPERM;
843 
844 	/*
845 	 * OK, now we're committed to doing something.
846 	 */
847 	gencnt = unp_gencnt;
848 	n = unp_count;
849 
850 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
851 
852 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
853 	     unp = LIST_NEXT(unp, unp_link)) {
854 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
855 			unp_list[i++] = unp;
856 	}
857 	n = i;			/* in case we lost some during malloc */
858 
859 	error = 0;
860 	for (i = 0; i < n; i++) {
861 		unp = unp_list[i];
862 		if (unp->unp_gencnt <= gencnt) {
863 			struct xunpcb xu;
864 			xu.xu_len = sizeof xu;
865 			xu.xu_unpp = unp;
866 			/*
867 			 * XXX - need more locking here to protect against
868 			 * connect/disconnect races for SMP.
869 			 */
870 			if (unp->unp_addr)
871 				bcopy(unp->unp_addr, &xu.xu_addr,
872 				      unp->unp_addr->sun_len);
873 			if (unp->unp_conn && unp->unp_conn->unp_addr)
874 				bcopy(unp->unp_conn->unp_addr,
875 				      &xu.xu_caddr,
876 				      unp->unp_conn->unp_addr->sun_len);
877 			bcopy(unp, &xu.xu_unp, sizeof *unp);
878 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
879 			error = SYSCTL_OUT(req, &xu, sizeof xu);
880 		}
881 	}
882 	kfree(unp_list, M_TEMP);
883 	return error;
884 }
885 
886 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
887 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
888 	    "List of active local datagram sockets");
889 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
890 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
891 	    "List of active local stream sockets");
892 
893 static void
894 unp_shutdown(struct unpcb *unp)
895 {
896 	struct socket *so;
897 
898 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn != NULL &&
899 	    (so = unp->unp_conn->unp_socket))
900 		socantrcvmore(so);
901 }
902 
903 static void
904 unp_drop(struct unpcb *unp, int err)
905 {
906 	struct socket *so = unp->unp_socket;
907 
908 	so->so_error = err;
909 	unp_disconnect(unp);
910 }
911 
912 #ifdef notdef
913 void
914 unp_drain(void)
915 {
916 
917 }
918 #endif
919 
920 int
921 unp_externalize(struct mbuf *rights)
922 {
923 	struct proc *p = curproc;		/* XXX */
924 	int i;
925 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
926 	int *fdp;
927 	struct file **rp;
928 	struct file *fp;
929 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
930 		/ sizeof (struct file *);
931 	int f;
932 
933 	/*
934 	 * if the new FD's will not fit, then we free them all
935 	 */
936 	if (!fdavail(p, newfds)) {
937 		rp = (struct file **)CMSG_DATA(cm);
938 		for (i = 0; i < newfds; i++) {
939 			fp = *rp;
940 			/*
941 			 * zero the pointer before calling unp_discard,
942 			 * since it may end up in unp_gc()..
943 			 */
944 			*rp++ = 0;
945 			unp_discard(fp, NULL);
946 		}
947 		return (EMSGSIZE);
948 	}
949 	/*
950 	 * now change each pointer to an fd in the global table to
951 	 * an integer that is the index to the local fd table entry
952 	 * that we set up to point to the global one we are transferring.
953 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
954 	 * then do it in forward order. In that case, an integer will
955 	 * always come in the same place or before its corresponding
956 	 * struct file pointer.
957 	 * If sizeof (struct file *) is smaller than sizeof int, then
958 	 * do it in reverse order.
959 	 */
960 	if (sizeof (struct file *) >= sizeof (int)) {
961 		fdp = (int *)(cm + 1);
962 		rp = (struct file **)CMSG_DATA(cm);
963 		for (i = 0; i < newfds; i++) {
964 			if (fdalloc(p, 0, &f))
965 				panic("unp_externalize");
966 			fp = *rp++;
967 			fsetfd(p, fp, f);
968 			fdrop(fp);
969 			spin_lock_wr(&unp_spin);
970 			fp->f_msgcount--;
971 			unp_rights--;
972 			spin_unlock_wr(&unp_spin);
973 			*fdp++ = f;
974 		}
975 	} else {
976 		fdp = (int *)(cm + 1) + newfds - 1;
977 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
978 		for (i = 0; i < newfds; i++) {
979 			if (fdalloc(p, 0, &f))
980 				panic("unp_externalize");
981 			fp = *rp--;
982 			fsetfd(p, fp, f);
983 			fdrop(fp);
984 			spin_lock_wr(&unp_spin);
985 			fp->f_msgcount--;
986 			unp_rights--;
987 			spin_unlock_wr(&unp_spin);
988 			*fdp-- = f;
989 		}
990 	}
991 
992 	/*
993 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
994 	 * differs.
995 	 */
996 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
997 	rights->m_len = cm->cmsg_len;
998 	return (0);
999 }
1000 
1001 void
1002 unp_init(void)
1003 {
1004 	unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
1005 	if (unp_zone == NULL)
1006 		panic("unp_init");
1007 	LIST_INIT(&unp_dhead);
1008 	LIST_INIT(&unp_shead);
1009 	spin_init(&unp_spin);
1010 }
1011 
1012 static int
1013 unp_internalize(struct mbuf *control, struct thread *td)
1014 {
1015 	struct proc *p = td->td_proc;
1016 	struct filedesc *fdescp;
1017 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1018 	struct file **rp;
1019 	struct file *fp;
1020 	int i, fd, *fdp;
1021 	struct cmsgcred *cmcred;
1022 	int oldfds;
1023 	u_int newlen;
1024 
1025 	KKASSERT(p);
1026 	fdescp = p->p_fd;
1027 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1028 	    cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
1029 		return (EINVAL);
1030 
1031 	/*
1032 	 * Fill in credential information.
1033 	 */
1034 	if (cm->cmsg_type == SCM_CREDS) {
1035 		cmcred = (struct cmsgcred *)(cm + 1);
1036 		cmcred->cmcred_pid = p->p_pid;
1037 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1038 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1039 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1040 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1041 							CMGROUP_MAX);
1042 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1043 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1044 		return(0);
1045 	}
1046 
1047 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
1048 	/*
1049 	 * check that all the FDs passed in refer to legal OPEN files
1050 	 * If not, reject the entire operation.
1051 	 */
1052 	fdp = (int *)(cm + 1);
1053 	for (i = 0; i < oldfds; i++) {
1054 		fd = *fdp++;
1055 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1056 		    fdescp->fd_files[fd].fp == NULL)
1057 			return (EBADF);
1058 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE)
1059 			return (EOPNOTSUPP);
1060 	}
1061 	/*
1062 	 * Now replace the integer FDs with pointers to
1063 	 * the associated global file table entry..
1064 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1065 	 * enough, return E2BIG.
1066 	 */
1067 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1068 	if (newlen > MCLBYTES)
1069 		return (E2BIG);
1070 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1071 		if (control->m_flags & M_EXT)
1072 			return (E2BIG);
1073 		MCLGET(control, MB_WAIT);
1074 		if (!(control->m_flags & M_EXT))
1075 			return (ENOBUFS);
1076 
1077 		/* copy the data to the cluster */
1078 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1079 		cm = mtod(control, struct cmsghdr *);
1080 	}
1081 
1082 	/*
1083 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1084 	 * differs.
1085 	 */
1086 	control->m_len = cm->cmsg_len = newlen;
1087 
1088 	/*
1089 	 * Transform the file descriptors into struct file pointers.
1090 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1091 	 * then do it in reverse order so that the int won't get until
1092 	 * we're done.
1093 	 * If sizeof (struct file *) is smaller than sizeof int, then
1094 	 * do it in forward order.
1095 	 */
1096 	if (sizeof (struct file *) >= sizeof (int)) {
1097 		fdp = (int *)(cm + 1) + oldfds - 1;
1098 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1099 		for (i = 0; i < oldfds; i++) {
1100 			fp = fdescp->fd_files[*fdp--].fp;
1101 			*rp-- = fp;
1102 			fhold(fp);
1103 			spin_lock_wr(&unp_spin);
1104 			fp->f_msgcount++;
1105 			unp_rights++;
1106 			spin_unlock_wr(&unp_spin);
1107 		}
1108 	} else {
1109 		fdp = (int *)(cm + 1);
1110 		rp = (struct file **)CMSG_DATA(cm);
1111 		for (i = 0; i < oldfds; i++) {
1112 			fp = fdescp->fd_files[*fdp++].fp;
1113 			*rp++ = fp;
1114 			fhold(fp);
1115 			spin_lock_wr(&unp_spin);
1116 			fp->f_msgcount++;
1117 			unp_rights++;
1118 			spin_unlock_wr(&unp_spin);
1119 		}
1120 	}
1121 	return (0);
1122 }
1123 
1124 /*
1125  * Garbage collect in-transit file descriptors that get lost due to
1126  * loops (i.e. when a socket is sent to another process over itself,
1127  * and more complex situations).
1128  *
1129  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1130  */
1131 
1132 struct unp_gc_info {
1133 	struct file **extra_ref;
1134 	struct file *locked_fp;
1135 	int defer;
1136 	int index;
1137 	int maxindex;
1138 };
1139 
1140 static void
1141 unp_gc(void)
1142 {
1143 	struct unp_gc_info info;
1144 	static boolean_t unp_gcing;
1145 	struct file **fpp;
1146 	int i;
1147 
1148 	spin_lock_wr(&unp_spin);
1149 	if (unp_gcing) {
1150 		spin_unlock_wr(&unp_spin);
1151 		return;
1152 	}
1153 	unp_gcing = TRUE;
1154 	spin_unlock_wr(&unp_spin);
1155 
1156 	/*
1157 	 * before going through all this, set all FDs to
1158 	 * be NOT defered and NOT externally accessible
1159 	 */
1160 	info.defer = 0;
1161 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1162 	do {
1163 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1164 	} while (info.defer);
1165 
1166 	/*
1167 	 * We grab an extra reference to each of the file table entries
1168 	 * that are not otherwise accessible and then free the rights
1169 	 * that are stored in messages on them.
1170 	 *
1171 	 * The bug in the orginal code is a little tricky, so I'll describe
1172 	 * what's wrong with it here.
1173 	 *
1174 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1175 	 * times -- consider the case of sockets A and B that contain
1176 	 * references to each other.  On a last close of some other socket,
1177 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1178 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1179 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1180 	 * results in the following chain.  Closef calls soo_close, which
1181 	 * calls soclose.   Soclose calls first (through the switch
1182 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1183 	 * returns because the previous instance had set unp_gcing, and
1184 	 * we return all the way back to soclose, which marks the socket
1185 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1186 	 * to free up the rights that are queued in messages on the socket A,
1187 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1188 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1189 	 * instance of unp_discard just calls closef on B.
1190 	 *
1191 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1192 	 * which results in another closef on A.  Unfortunately, A is already
1193 	 * being closed, and the descriptor has already been marked with
1194 	 * SS_NOFDREF, and soclose panics at this point.
1195 	 *
1196 	 * Here, we first take an extra reference to each inaccessible
1197 	 * descriptor.  Then, we call sorflush ourself, since we know
1198 	 * it is a Unix domain socket anyhow.  After we destroy all the
1199 	 * rights carried in messages, we do a last closef to get rid
1200 	 * of our extra reference.  This is the last close, and the
1201 	 * unp_detach etc will shut down the socket.
1202 	 *
1203 	 * 91/09/19, bsy@cs.cmu.edu
1204 	 */
1205 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1206 	info.maxindex = 256;
1207 
1208 	do {
1209 		/*
1210 		 * Look for matches
1211 		 */
1212 		info.index = 0;
1213 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1214 
1215 		/*
1216 		 * For each FD on our hit list, do the following two things
1217 		 */
1218 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1219 			struct file *tfp = *fpp;
1220 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1221 				sorflush((struct socket *)(tfp->f_data));
1222 		}
1223 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1224 			closef(*fpp, NULL);
1225 	} while (info.index == info.maxindex);
1226 	kfree((caddr_t)info.extra_ref, M_FILE);
1227 	unp_gcing = FALSE;
1228 }
1229 
1230 /*
1231  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1232  */
1233 static int
1234 unp_gc_checkrefs(struct file *fp, void *data)
1235 {
1236 	struct unp_gc_info *info = data;
1237 
1238 	if (fp->f_count == 0)
1239 		return(0);
1240 	if (info->index == info->maxindex)
1241 		return(-1);
1242 
1243 	/*
1244 	 * If all refs are from msgs, and it's not marked accessible
1245 	 * then it must be referenced from some unreachable cycle
1246 	 * of (shut-down) FDs, so include it in our
1247 	 * list of FDs to remove
1248 	 */
1249 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1250 		info->extra_ref[info->index++] = fp;
1251 		fhold(fp);
1252 	}
1253 	return(0);
1254 }
1255 
1256 /*
1257  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1258  */
1259 static int
1260 unp_gc_clearmarks(struct file *fp, void *data __unused)
1261 {
1262 	fp->f_flag &= ~(FMARK|FDEFER);
1263 	return(0);
1264 }
1265 
1266 /*
1267  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1268  */
1269 static int
1270 unp_gc_checkmarks(struct file *fp, void *data)
1271 {
1272 	struct unp_gc_info *info = data;
1273 	struct socket *so;
1274 
1275 	/*
1276 	 * If the file is not open, skip it
1277 	 */
1278 	if (fp->f_count == 0)
1279 		return(0);
1280 	/*
1281 	 * If we already marked it as 'defer'  in a
1282 	 * previous pass, then try process it this time
1283 	 * and un-mark it
1284 	 */
1285 	if (fp->f_flag & FDEFER) {
1286 		fp->f_flag &= ~FDEFER;
1287 		--info->defer;
1288 	} else {
1289 		/*
1290 		 * if it's not defered, then check if it's
1291 		 * already marked.. if so skip it
1292 		 */
1293 		if (fp->f_flag & FMARK)
1294 			return(0);
1295 		/*
1296 		 * If all references are from messages
1297 		 * in transit, then skip it. it's not
1298 		 * externally accessible.
1299 		 */
1300 		if (fp->f_count == fp->f_msgcount)
1301 			return(0);
1302 		/*
1303 		 * If it got this far then it must be
1304 		 * externally accessible.
1305 		 */
1306 		fp->f_flag |= FMARK;
1307 	}
1308 	/*
1309 	 * either it was defered, or it is externally
1310 	 * accessible and not already marked so.
1311 	 * Now check if it is possibly one of OUR sockets.
1312 	 */
1313 	if (fp->f_type != DTYPE_SOCKET ||
1314 	    (so = (struct socket *)fp->f_data) == NULL)
1315 		return(0);
1316 	if (so->so_proto->pr_domain != &localdomain ||
1317 	    !(so->so_proto->pr_flags & PR_RIGHTS))
1318 		return(0);
1319 #ifdef notdef
1320 	XXX note: exclusive fp->f_spin lock held
1321 	if (so->so_rcv.sb_flags & SB_LOCK) {
1322 		/*
1323 		 * This is problematical; it's not clear
1324 		 * we need to wait for the sockbuf to be
1325 		 * unlocked (on a uniprocessor, at least),
1326 		 * and it's also not clear what to do
1327 		 * if sbwait returns an error due to receipt
1328 		 * of a signal.  If sbwait does return
1329 		 * an error, we'll go into an infinite
1330 		 * loop.  Delete all of this for now.
1331 		 */
1332 		sbwait(&so->so_rcv);
1333 		goto restart;
1334 	}
1335 #endif
1336 	/*
1337 	 * So, Ok, it's one of our sockets and it IS externally
1338 	 * accessible (or was defered). Now we look
1339 	 * to see if we hold any file descriptors in its
1340 	 * message buffers. Follow those links and mark them
1341 	 * as accessible too.
1342 	 */
1343 	info->locked_fp = fp;
1344 /*	spin_lock_wr(&so->so_rcv.sb_spin); */
1345 	unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1346 /*	spin_unlock_wr(&so->so_rcv.sb_spin);*/
1347 	return (0);
1348 }
1349 
1350 void
1351 unp_dispose(struct mbuf *m)
1352 {
1353 	if (m)
1354 		unp_scan(m, unp_discard, NULL);
1355 }
1356 
1357 static int
1358 unp_listen(struct unpcb *unp, struct thread *td)
1359 {
1360 	struct proc *p = td->td_proc;
1361 
1362 	KKASSERT(p);
1363 	cru2x(p->p_ucred, &unp->unp_peercred);
1364 	unp->unp_flags |= UNP_HAVEPCCACHED;
1365 	return (0);
1366 }
1367 
1368 static void
1369 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
1370 {
1371 	struct mbuf *m;
1372 	struct file **rp;
1373 	struct cmsghdr *cm;
1374 	int i;
1375 	int qfds;
1376 
1377 	while (m0) {
1378 		for (m = m0; m; m = m->m_next) {
1379 			if (m->m_type == MT_CONTROL &&
1380 			    m->m_len >= sizeof(*cm)) {
1381 				cm = mtod(m, struct cmsghdr *);
1382 				if (cm->cmsg_level != SOL_SOCKET ||
1383 				    cm->cmsg_type != SCM_RIGHTS)
1384 					continue;
1385 				qfds = (cm->cmsg_len -
1386 					(CMSG_DATA(cm) - (u_char *)cm))
1387 						/ sizeof (struct file *);
1388 				rp = (struct file **)CMSG_DATA(cm);
1389 				for (i = 0; i < qfds; i++)
1390 					(*op)(*rp++, data);
1391 				break;		/* XXX, but saves time */
1392 			}
1393 		}
1394 		m0 = m0->m_nextpkt;
1395 	}
1396 }
1397 
1398 static void
1399 unp_mark(struct file *fp, void *data)
1400 {
1401 	struct unp_gc_info *info = data;
1402 
1403 	if (info->locked_fp != fp)
1404 		spin_lock_wr(&fp->f_spin);
1405 	if ((fp->f_flag & FMARK) == 0) {
1406 		++info->defer;
1407 		fp->f_flag |= (FMARK|FDEFER);
1408 	}
1409 	if (info->locked_fp != fp)
1410 		spin_unlock_wr(&fp->f_spin);
1411 }
1412 
1413 static void
1414 unp_discard(struct file *fp, void *data __unused)
1415 {
1416 	spin_lock_wr(&unp_spin);
1417 	fp->f_msgcount--;
1418 	unp_rights--;
1419 	spin_unlock_wr(&unp_spin);
1420 	closef(fp, NULL);
1421 }
1422 
1423