xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision 2381583524bd7679b9e475f7ce722874d1eea618)
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 1.39 2008/05/09 17:52:17 dillon 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 *vp;
613 	struct vattr vattr;
614 	int error, namelen;
615 	struct nlookupdata nd;
616 	char buf[SOCK_MAXADDRLEN];
617 
618 	if (unp->unp_vnode != NULL)
619 		return (EINVAL);
620 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
621 	if (namelen <= 0)
622 		return (EINVAL);
623 	strncpy(buf, soun->sun_path, namelen);
624 	buf[namelen] = 0;	/* null-terminate the string */
625 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
626 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
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)
632 		goto done;
633 
634 	VATTR_NULL(&vattr);
635 	vattr.va_type = VSOCK;
636 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
637 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
638 	if (error == 0) {
639 		vp->v_socket = unp->unp_socket;
640 		unp->unp_vnode = vp;
641 		unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
642 		vn_unlock(vp);
643 	}
644 done:
645 	nlookup_done(&nd);
646 	return (error);
647 }
648 
649 static int
650 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
651 {
652 	struct proc *p = td->td_proc;
653 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
654 	struct vnode *vp;
655 	struct socket *so2, *so3;
656 	struct unpcb *unp, *unp2, *unp3;
657 	int error, len;
658 	struct nlookupdata nd;
659 	char buf[SOCK_MAXADDRLEN];
660 
661 	KKASSERT(p);
662 
663 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
664 	if (len <= 0)
665 		return EINVAL;
666 	strncpy(buf, soun->sun_path, len);
667 	buf[len] = 0;
668 
669 	vp = NULL;
670 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
671 	if (error == 0)
672 		error = nlookup(&nd);
673 	if (error == 0)
674 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
675 	nlookup_done(&nd);
676 	if (error)
677 		return (error);
678 
679 	if (vp->v_type != VSOCK) {
680 		error = ENOTSOCK;
681 		goto bad;
682 	}
683 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred);
684 	if (error)
685 		goto bad;
686 	so2 = vp->v_socket;
687 	if (so2 == NULL) {
688 		error = ECONNREFUSED;
689 		goto bad;
690 	}
691 	if (so->so_type != so2->so_type) {
692 		error = EPROTOTYPE;
693 		goto bad;
694 	}
695 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
696 		if (!(so2->so_options & SO_ACCEPTCONN) ||
697 		    (so3 = sonewconn(so2, 0)) == NULL) {
698 			error = ECONNREFUSED;
699 			goto bad;
700 		}
701 		unp = so->so_pcb;
702 		unp2 = so2->so_pcb;
703 		unp3 = so3->so_pcb;
704 		if (unp2->unp_addr)
705 			unp3->unp_addr = (struct sockaddr_un *)
706 				dup_sockaddr((struct sockaddr *)unp2->unp_addr);
707 
708 		/*
709 		 * unp_peercred management:
710 		 *
711 		 * The connecter's (client's) credentials are copied
712 		 * from its process structure at the time of connect()
713 		 * (which is now).
714 		 */
715 		cru2x(p->p_ucred, &unp3->unp_peercred);
716 		unp3->unp_flags |= UNP_HAVEPC;
717 		/*
718 		 * The receiver's (server's) credentials are copied
719 		 * from the unp_peercred member of socket on which the
720 		 * former called listen(); unp_listen() cached that
721 		 * process's credentials at that time so we can use
722 		 * them now.
723 		 */
724 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
725 		    ("unp_connect: listener without cached peercred"));
726 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
727 		    sizeof(unp->unp_peercred));
728 		unp->unp_flags |= UNP_HAVEPC;
729 
730 		so2 = so3;
731 	}
732 	error = unp_connect2(so, so2);
733 bad:
734 	vput(vp);
735 	return (error);
736 }
737 
738 int
739 unp_connect2(struct socket *so, struct socket *so2)
740 {
741 	struct unpcb *unp = so->so_pcb;
742 	struct unpcb *unp2;
743 
744 	if (so2->so_type != so->so_type)
745 		return (EPROTOTYPE);
746 	unp2 = so2->so_pcb;
747 	unp->unp_conn = unp2;
748 	switch (so->so_type) {
749 
750 	case SOCK_DGRAM:
751 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
752 		soisconnected(so);
753 		break;
754 
755 	case SOCK_STREAM:
756 		unp2->unp_conn = unp;
757 		soisconnected(so);
758 		soisconnected(so2);
759 		break;
760 
761 	default:
762 		panic("unp_connect2");
763 	}
764 	return (0);
765 }
766 
767 static void
768 unp_disconnect(struct unpcb *unp)
769 {
770 	struct unpcb *unp2 = unp->unp_conn;
771 
772 	if (unp2 == NULL)
773 		return;
774 
775 	unp->unp_conn = NULL;
776 
777 	switch (unp->unp_socket->so_type) {
778 	case SOCK_DGRAM:
779 		LIST_REMOVE(unp, unp_reflink);
780 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
781 		break;
782 	case SOCK_STREAM:
783 		soisdisconnected(unp->unp_socket);
784 		unp2->unp_conn = NULL;
785 		soisdisconnected(unp2->unp_socket);
786 		break;
787 	}
788 }
789 
790 #ifdef notdef
791 void
792 unp_abort(struct unpcb *unp)
793 {
794 
795 	unp_detach(unp);
796 }
797 #endif
798 
799 static int
800 prison_unpcb(struct thread *td, struct unpcb *unp)
801 {
802 	struct proc *p;
803 
804 	if (td == NULL)
805 		return (0);
806 	if ((p = td->td_proc) == NULL)
807 		return (0);
808 	if (!p->p_ucred->cr_prison)
809 		return (0);
810 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
811 		return (0);
812 	return (1);
813 }
814 
815 static int
816 unp_pcblist(SYSCTL_HANDLER_ARGS)
817 {
818 	int error, i, n;
819 	struct unpcb *unp, **unp_list;
820 	unp_gen_t gencnt;
821 	struct unp_head *head;
822 
823 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
824 
825 	KKASSERT(curproc != NULL);
826 
827 	/*
828 	 * The process of preparing the PCB list is too time-consuming and
829 	 * resource-intensive to repeat twice on every request.
830 	 */
831 	if (req->oldptr == NULL) {
832 		n = unp_count;
833 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
834 		return 0;
835 	}
836 
837 	if (req->newptr != NULL)
838 		return EPERM;
839 
840 	/*
841 	 * OK, now we're committed to doing something.
842 	 */
843 	gencnt = unp_gencnt;
844 	n = unp_count;
845 
846 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
847 
848 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
849 	     unp = LIST_NEXT(unp, unp_link)) {
850 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
851 			unp_list[i++] = unp;
852 	}
853 	n = i;			/* in case we lost some during malloc */
854 
855 	error = 0;
856 	for (i = 0; i < n; i++) {
857 		unp = unp_list[i];
858 		if (unp->unp_gencnt <= gencnt) {
859 			struct xunpcb xu;
860 			xu.xu_len = sizeof xu;
861 			xu.xu_unpp = unp;
862 			/*
863 			 * XXX - need more locking here to protect against
864 			 * connect/disconnect races for SMP.
865 			 */
866 			if (unp->unp_addr)
867 				bcopy(unp->unp_addr, &xu.xu_addr,
868 				      unp->unp_addr->sun_len);
869 			if (unp->unp_conn && unp->unp_conn->unp_addr)
870 				bcopy(unp->unp_conn->unp_addr,
871 				      &xu.xu_caddr,
872 				      unp->unp_conn->unp_addr->sun_len);
873 			bcopy(unp, &xu.xu_unp, sizeof *unp);
874 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
875 			error = SYSCTL_OUT(req, &xu, sizeof xu);
876 		}
877 	}
878 	kfree(unp_list, M_TEMP);
879 	return error;
880 }
881 
882 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
883 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
884 	    "List of active local datagram sockets");
885 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
886 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
887 	    "List of active local stream sockets");
888 
889 static void
890 unp_shutdown(struct unpcb *unp)
891 {
892 	struct socket *so;
893 
894 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn != NULL &&
895 	    (so = unp->unp_conn->unp_socket))
896 		socantrcvmore(so);
897 }
898 
899 static void
900 unp_drop(struct unpcb *unp, int err)
901 {
902 	struct socket *so = unp->unp_socket;
903 
904 	so->so_error = err;
905 	unp_disconnect(unp);
906 }
907 
908 #ifdef notdef
909 void
910 unp_drain(void)
911 {
912 
913 }
914 #endif
915 
916 int
917 unp_externalize(struct mbuf *rights)
918 {
919 	struct proc *p = curproc;		/* XXX */
920 	int i;
921 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
922 	int *fdp;
923 	struct file **rp;
924 	struct file *fp;
925 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
926 		/ sizeof (struct file *);
927 	int f;
928 
929 	/*
930 	 * if the new FD's will not fit, then we free them all
931 	 */
932 	if (!fdavail(p, newfds)) {
933 		rp = (struct file **)CMSG_DATA(cm);
934 		for (i = 0; i < newfds; i++) {
935 			fp = *rp;
936 			/*
937 			 * zero the pointer before calling unp_discard,
938 			 * since it may end up in unp_gc()..
939 			 */
940 			*rp++ = 0;
941 			unp_discard(fp, NULL);
942 		}
943 		return (EMSGSIZE);
944 	}
945 	/*
946 	 * now change each pointer to an fd in the global table to
947 	 * an integer that is the index to the local fd table entry
948 	 * that we set up to point to the global one we are transferring.
949 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
950 	 * then do it in forward order. In that case, an integer will
951 	 * always come in the same place or before its corresponding
952 	 * struct file pointer.
953 	 * If sizeof (struct file *) is smaller than sizeof int, then
954 	 * do it in reverse order.
955 	 */
956 	if (sizeof (struct file *) >= sizeof (int)) {
957 		fdp = (int *)(cm + 1);
958 		rp = (struct file **)CMSG_DATA(cm);
959 		for (i = 0; i < newfds; i++) {
960 			if (fdalloc(p, 0, &f))
961 				panic("unp_externalize");
962 			fp = *rp++;
963 			fsetfd(p, fp, f);
964 			fdrop(fp);
965 			spin_lock_wr(&unp_spin);
966 			fp->f_msgcount--;
967 			unp_rights--;
968 			spin_unlock_wr(&unp_spin);
969 			*fdp++ = f;
970 		}
971 	} else {
972 		fdp = (int *)(cm + 1) + newfds - 1;
973 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
974 		for (i = 0; i < newfds; i++) {
975 			if (fdalloc(p, 0, &f))
976 				panic("unp_externalize");
977 			fp = *rp--;
978 			fsetfd(p, fp, f);
979 			fdrop(fp);
980 			spin_lock_wr(&unp_spin);
981 			fp->f_msgcount--;
982 			unp_rights--;
983 			spin_unlock_wr(&unp_spin);
984 			*fdp-- = f;
985 		}
986 	}
987 
988 	/*
989 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
990 	 * differs.
991 	 */
992 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
993 	rights->m_len = cm->cmsg_len;
994 	return (0);
995 }
996 
997 void
998 unp_init(void)
999 {
1000 	unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
1001 	if (unp_zone == NULL)
1002 		panic("unp_init");
1003 	LIST_INIT(&unp_dhead);
1004 	LIST_INIT(&unp_shead);
1005 	spin_init(&unp_spin);
1006 }
1007 
1008 static int
1009 unp_internalize(struct mbuf *control, struct thread *td)
1010 {
1011 	struct proc *p = td->td_proc;
1012 	struct filedesc *fdescp;
1013 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1014 	struct file **rp;
1015 	struct file *fp;
1016 	int i, fd, *fdp;
1017 	struct cmsgcred *cmcred;
1018 	int oldfds;
1019 	u_int newlen;
1020 
1021 	KKASSERT(p);
1022 	fdescp = p->p_fd;
1023 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1024 	    cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
1025 		return (EINVAL);
1026 
1027 	/*
1028 	 * Fill in credential information.
1029 	 */
1030 	if (cm->cmsg_type == SCM_CREDS) {
1031 		cmcred = (struct cmsgcred *)(cm + 1);
1032 		cmcred->cmcred_pid = p->p_pid;
1033 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1034 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1035 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1036 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1037 							CMGROUP_MAX);
1038 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1039 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1040 		return(0);
1041 	}
1042 
1043 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
1044 	/*
1045 	 * check that all the FDs passed in refer to legal OPEN files
1046 	 * If not, reject the entire operation.
1047 	 */
1048 	fdp = (int *)(cm + 1);
1049 	for (i = 0; i < oldfds; i++) {
1050 		fd = *fdp++;
1051 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1052 		    fdescp->fd_files[fd].fp == NULL)
1053 			return (EBADF);
1054 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE)
1055 			return (EOPNOTSUPP);
1056 	}
1057 	/*
1058 	 * Now replace the integer FDs with pointers to
1059 	 * the associated global file table entry..
1060 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1061 	 * enough, return E2BIG.
1062 	 */
1063 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1064 	if (newlen > MCLBYTES)
1065 		return (E2BIG);
1066 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1067 		if (control->m_flags & M_EXT)
1068 			return (E2BIG);
1069 		MCLGET(control, MB_WAIT);
1070 		if (!(control->m_flags & M_EXT))
1071 			return (ENOBUFS);
1072 
1073 		/* copy the data to the cluster */
1074 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1075 		cm = mtod(control, struct cmsghdr *);
1076 	}
1077 
1078 	/*
1079 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1080 	 * differs.
1081 	 */
1082 	control->m_len = cm->cmsg_len = newlen;
1083 
1084 	/*
1085 	 * Transform the file descriptors into struct file pointers.
1086 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1087 	 * then do it in reverse order so that the int won't get until
1088 	 * we're done.
1089 	 * If sizeof (struct file *) is smaller than sizeof int, then
1090 	 * do it in forward order.
1091 	 */
1092 	if (sizeof (struct file *) >= sizeof (int)) {
1093 		fdp = (int *)(cm + 1) + oldfds - 1;
1094 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1095 		for (i = 0; i < oldfds; i++) {
1096 			fp = fdescp->fd_files[*fdp--].fp;
1097 			*rp-- = fp;
1098 			fhold(fp);
1099 			spin_lock_wr(&unp_spin);
1100 			fp->f_msgcount++;
1101 			unp_rights++;
1102 			spin_unlock_wr(&unp_spin);
1103 		}
1104 	} else {
1105 		fdp = (int *)(cm + 1);
1106 		rp = (struct file **)CMSG_DATA(cm);
1107 		for (i = 0; i < oldfds; i++) {
1108 			fp = fdescp->fd_files[*fdp++].fp;
1109 			*rp++ = fp;
1110 			fhold(fp);
1111 			spin_lock_wr(&unp_spin);
1112 			fp->f_msgcount++;
1113 			unp_rights++;
1114 			spin_unlock_wr(&unp_spin);
1115 		}
1116 	}
1117 	return (0);
1118 }
1119 
1120 /*
1121  * Garbage collect in-transit file descriptors that get lost due to
1122  * loops (i.e. when a socket is sent to another process over itself,
1123  * and more complex situations).
1124  *
1125  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1126  */
1127 
1128 struct unp_gc_info {
1129 	struct file **extra_ref;
1130 	struct file *locked_fp;
1131 	int defer;
1132 	int index;
1133 	int maxindex;
1134 };
1135 
1136 static void
1137 unp_gc(void)
1138 {
1139 	struct unp_gc_info info;
1140 	static boolean_t unp_gcing;
1141 	struct file **fpp;
1142 	int i;
1143 
1144 	spin_lock_wr(&unp_spin);
1145 	if (unp_gcing) {
1146 		spin_unlock_wr(&unp_spin);
1147 		return;
1148 	}
1149 	unp_gcing = TRUE;
1150 	spin_unlock_wr(&unp_spin);
1151 
1152 	/*
1153 	 * before going through all this, set all FDs to
1154 	 * be NOT defered and NOT externally accessible
1155 	 */
1156 	info.defer = 0;
1157 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1158 	do {
1159 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1160 	} while (info.defer);
1161 
1162 	/*
1163 	 * We grab an extra reference to each of the file table entries
1164 	 * that are not otherwise accessible and then free the rights
1165 	 * that are stored in messages on them.
1166 	 *
1167 	 * The bug in the orginal code is a little tricky, so I'll describe
1168 	 * what's wrong with it here.
1169 	 *
1170 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1171 	 * times -- consider the case of sockets A and B that contain
1172 	 * references to each other.  On a last close of some other socket,
1173 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1174 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1175 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1176 	 * results in the following chain.  Closef calls soo_close, which
1177 	 * calls soclose.   Soclose calls first (through the switch
1178 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1179 	 * returns because the previous instance had set unp_gcing, and
1180 	 * we return all the way back to soclose, which marks the socket
1181 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1182 	 * to free up the rights that are queued in messages on the socket A,
1183 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1184 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1185 	 * instance of unp_discard just calls closef on B.
1186 	 *
1187 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1188 	 * which results in another closef on A.  Unfortunately, A is already
1189 	 * being closed, and the descriptor has already been marked with
1190 	 * SS_NOFDREF, and soclose panics at this point.
1191 	 *
1192 	 * Here, we first take an extra reference to each inaccessible
1193 	 * descriptor.  Then, we call sorflush ourself, since we know
1194 	 * it is a Unix domain socket anyhow.  After we destroy all the
1195 	 * rights carried in messages, we do a last closef to get rid
1196 	 * of our extra reference.  This is the last close, and the
1197 	 * unp_detach etc will shut down the socket.
1198 	 *
1199 	 * 91/09/19, bsy@cs.cmu.edu
1200 	 */
1201 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1202 	info.maxindex = 256;
1203 
1204 	do {
1205 		/*
1206 		 * Look for matches
1207 		 */
1208 		info.index = 0;
1209 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1210 
1211 		/*
1212 		 * For each FD on our hit list, do the following two things
1213 		 */
1214 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1215 			struct file *tfp = *fpp;
1216 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1217 				sorflush((struct socket *)(tfp->f_data));
1218 		}
1219 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1220 			closef(*fpp, NULL);
1221 	} while (info.index == info.maxindex);
1222 	kfree((caddr_t)info.extra_ref, M_FILE);
1223 	unp_gcing = FALSE;
1224 }
1225 
1226 /*
1227  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1228  */
1229 static int
1230 unp_gc_checkrefs(struct file *fp, void *data)
1231 {
1232 	struct unp_gc_info *info = data;
1233 
1234 	if (fp->f_count == 0)
1235 		return(0);
1236 	if (info->index == info->maxindex)
1237 		return(-1);
1238 
1239 	/*
1240 	 * If all refs are from msgs, and it's not marked accessible
1241 	 * then it must be referenced from some unreachable cycle
1242 	 * of (shut-down) FDs, so include it in our
1243 	 * list of FDs to remove
1244 	 */
1245 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1246 		info->extra_ref[info->index++] = fp;
1247 		fhold(fp);
1248 	}
1249 	return(0);
1250 }
1251 
1252 /*
1253  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1254  */
1255 static int
1256 unp_gc_clearmarks(struct file *fp, void *data __unused)
1257 {
1258 	fp->f_flag &= ~(FMARK|FDEFER);
1259 	return(0);
1260 }
1261 
1262 /*
1263  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1264  */
1265 static int
1266 unp_gc_checkmarks(struct file *fp, void *data)
1267 {
1268 	struct unp_gc_info *info = data;
1269 	struct socket *so;
1270 
1271 	/*
1272 	 * If the file is not open, skip it
1273 	 */
1274 	if (fp->f_count == 0)
1275 		return(0);
1276 	/*
1277 	 * If we already marked it as 'defer'  in a
1278 	 * previous pass, then try process it this time
1279 	 * and un-mark it
1280 	 */
1281 	if (fp->f_flag & FDEFER) {
1282 		fp->f_flag &= ~FDEFER;
1283 		--info->defer;
1284 	} else {
1285 		/*
1286 		 * if it's not defered, then check if it's
1287 		 * already marked.. if so skip it
1288 		 */
1289 		if (fp->f_flag & FMARK)
1290 			return(0);
1291 		/*
1292 		 * If all references are from messages
1293 		 * in transit, then skip it. it's not
1294 		 * externally accessible.
1295 		 */
1296 		if (fp->f_count == fp->f_msgcount)
1297 			return(0);
1298 		/*
1299 		 * If it got this far then it must be
1300 		 * externally accessible.
1301 		 */
1302 		fp->f_flag |= FMARK;
1303 	}
1304 	/*
1305 	 * either it was defered, or it is externally
1306 	 * accessible and not already marked so.
1307 	 * Now check if it is possibly one of OUR sockets.
1308 	 */
1309 	if (fp->f_type != DTYPE_SOCKET ||
1310 	    (so = (struct socket *)fp->f_data) == NULL)
1311 		return(0);
1312 	if (so->so_proto->pr_domain != &localdomain ||
1313 	    !(so->so_proto->pr_flags & PR_RIGHTS))
1314 		return(0);
1315 #ifdef notdef
1316 	XXX note: exclusive fp->f_spin lock held
1317 	if (so->so_rcv.sb_flags & SB_LOCK) {
1318 		/*
1319 		 * This is problematical; it's not clear
1320 		 * we need to wait for the sockbuf to be
1321 		 * unlocked (on a uniprocessor, at least),
1322 		 * and it's also not clear what to do
1323 		 * if sbwait returns an error due to receipt
1324 		 * of a signal.  If sbwait does return
1325 		 * an error, we'll go into an infinite
1326 		 * loop.  Delete all of this for now.
1327 		 */
1328 		sbwait(&so->so_rcv);
1329 		goto restart;
1330 	}
1331 #endif
1332 	/*
1333 	 * So, Ok, it's one of our sockets and it IS externally
1334 	 * accessible (or was defered). Now we look
1335 	 * to see if we hold any file descriptors in its
1336 	 * message buffers. Follow those links and mark them
1337 	 * as accessible too.
1338 	 */
1339 	info->locked_fp = fp;
1340 /*	spin_lock_wr(&so->so_rcv.sb_spin); */
1341 	unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1342 /*	spin_unlock_wr(&so->so_rcv.sb_spin);*/
1343 	return (0);
1344 }
1345 
1346 void
1347 unp_dispose(struct mbuf *m)
1348 {
1349 	if (m)
1350 		unp_scan(m, unp_discard, NULL);
1351 }
1352 
1353 static int
1354 unp_listen(struct unpcb *unp, struct thread *td)
1355 {
1356 	struct proc *p = td->td_proc;
1357 
1358 	KKASSERT(p);
1359 	cru2x(p->p_ucred, &unp->unp_peercred);
1360 	unp->unp_flags |= UNP_HAVEPCCACHED;
1361 	return (0);
1362 }
1363 
1364 static void
1365 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
1366 {
1367 	struct mbuf *m;
1368 	struct file **rp;
1369 	struct cmsghdr *cm;
1370 	int i;
1371 	int qfds;
1372 
1373 	while (m0) {
1374 		for (m = m0; m; m = m->m_next) {
1375 			if (m->m_type == MT_CONTROL &&
1376 			    m->m_len >= sizeof(*cm)) {
1377 				cm = mtod(m, struct cmsghdr *);
1378 				if (cm->cmsg_level != SOL_SOCKET ||
1379 				    cm->cmsg_type != SCM_RIGHTS)
1380 					continue;
1381 				qfds = (cm->cmsg_len -
1382 					(CMSG_DATA(cm) - (u_char *)cm))
1383 						/ sizeof (struct file *);
1384 				rp = (struct file **)CMSG_DATA(cm);
1385 				for (i = 0; i < qfds; i++)
1386 					(*op)(*rp++, data);
1387 				break;		/* XXX, but saves time */
1388 			}
1389 		}
1390 		m0 = m0->m_nextpkt;
1391 	}
1392 }
1393 
1394 static void
1395 unp_mark(struct file *fp, void *data)
1396 {
1397 	struct unp_gc_info *info = data;
1398 
1399 	if (info->locked_fp != fp)
1400 		spin_lock_wr(&fp->f_spin);
1401 	if ((fp->f_flag & FMARK) == 0) {
1402 		++info->defer;
1403 		fp->f_flag |= (FMARK|FDEFER);
1404 	}
1405 	if (info->locked_fp != fp)
1406 		spin_unlock_wr(&fp->f_spin);
1407 }
1408 
1409 static void
1410 unp_discard(struct file *fp, void *data __unused)
1411 {
1412 	spin_lock_wr(&unp_spin);
1413 	fp->f_msgcount--;
1414 	unp_rights--;
1415 	spin_unlock_wr(&unp_spin);
1416 	closef(fp, NULL);
1417 }
1418 
1419