xref: /freebsd/sys/kern/uipc_usrreq.c (revision 42c159fe388a3765f69860c84183700af37aca8a)
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  * $FreeBSD$
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/domain.h>
42 #include <sys/filedesc.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
45 #include <sys/file.h>
46 #include <sys/mutex.h>
47 #include <sys/mbuf.h>
48 #include <sys/namei.h>
49 #include <sys/proc.h>
50 #include <sys/protosw.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/resourcevar.h>
54 #include <sys/stat.h>
55 #include <sys/sysctl.h>
56 #include <sys/un.h>
57 #include <sys/unpcb.h>
58 #include <sys/vnode.h>
59 #include <sys/jail.h>
60 #include <sys/sx.h>
61 
62 #include <vm/uma.h>
63 
64 static uma_zone_t 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 *);
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 *, struct thread *);
86 static void    unp_disconnect(struct unpcb *);
87 static void    unp_shutdown(struct unpcb *);
88 static void    unp_drop(struct unpcb *, int);
89 static void    unp_gc(void);
90 static void    unp_scan(struct mbuf *, void (*)(struct file *));
91 static void    unp_mark(struct file *);
92 static void    unp_discard(struct file *);
93 static void    unp_freerights(struct file **, int);
94 static int     unp_internalize(struct mbuf **, struct thread *);
95 static int     unp_listen(struct unpcb *, struct thread *);
96 
97 static int
98 uipc_abort(struct socket *so)
99 {
100 	struct unpcb *unp = sotounpcb(so);
101 
102 	if (unp == 0)
103 		return EINVAL;
104 	unp_drop(unp, ECONNABORTED);
105 	unp_detach(unp);
106 	sotryfree(so);
107 	return 0;
108 }
109 
110 static int
111 uipc_accept(struct socket *so, struct sockaddr **nam)
112 {
113 	struct unpcb *unp = sotounpcb(so);
114 
115 	if (unp == 0)
116 		return EINVAL;
117 
118 	/*
119 	 * Pass back name of connected socket,
120 	 * if it was bound and we are still connected
121 	 * (our peer may have closed already!).
122 	 */
123 	if (unp->unp_conn && unp->unp_conn->unp_addr) {
124 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
125 				    1);
126 	} else {
127 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
128 	}
129 	return 0;
130 }
131 
132 static int
133 uipc_attach(struct socket *so, int proto, struct thread *td)
134 {
135 	struct unpcb *unp = sotounpcb(so);
136 
137 	if (unp != 0)
138 		return EISCONN;
139 	return unp_attach(so);
140 }
141 
142 static int
143 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
144 {
145 	struct unpcb *unp = sotounpcb(so);
146 
147 	if (unp == 0)
148 		return EINVAL;
149 
150 	return unp_bind(unp, nam, td);
151 }
152 
153 static int
154 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
155 {
156 	struct unpcb *unp = sotounpcb(so);
157 
158 	if (unp == 0)
159 		return EINVAL;
160 	return unp_connect(so, nam, curthread);
161 }
162 
163 static int
164 uipc_connect2(struct socket *so1, struct socket *so2)
165 {
166 	struct unpcb *unp = sotounpcb(so1);
167 
168 	if (unp == 0)
169 		return EINVAL;
170 
171 	return unp_connect2(so1, so2);
172 }
173 
174 /* control is EOPNOTSUPP */
175 
176 static int
177 uipc_detach(struct socket *so)
178 {
179 	struct unpcb *unp = sotounpcb(so);
180 
181 	if (unp == 0)
182 		return EINVAL;
183 
184 	unp_detach(unp);
185 	return 0;
186 }
187 
188 static int
189 uipc_disconnect(struct socket *so)
190 {
191 	struct unpcb *unp = sotounpcb(so);
192 
193 	if (unp == 0)
194 		return EINVAL;
195 	unp_disconnect(unp);
196 	return 0;
197 }
198 
199 static int
200 uipc_listen(struct socket *so, struct thread *td)
201 {
202 	struct unpcb *unp = sotounpcb(so);
203 
204 	if (unp == 0 || unp->unp_vnode == 0)
205 		return EINVAL;
206 	return unp_listen(unp, td);
207 }
208 
209 static int
210 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
211 {
212 	struct unpcb *unp = sotounpcb(so);
213 
214 	if (unp == 0)
215 		return EINVAL;
216 	if (unp->unp_conn && unp->unp_conn->unp_addr)
217 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
218 				    1);
219 	return 0;
220 }
221 
222 static int
223 uipc_rcvd(struct socket *so, int flags)
224 {
225 	struct unpcb *unp = sotounpcb(so);
226 	struct socket *so2;
227 	u_long newhiwat;
228 
229 	if (unp == 0)
230 		return EINVAL;
231 	switch (so->so_type) {
232 	case SOCK_DGRAM:
233 		panic("uipc_rcvd DGRAM?");
234 		/*NOTREACHED*/
235 
236 	case SOCK_STREAM:
237 		if (unp->unp_conn == 0)
238 			break;
239 		so2 = unp->unp_conn->unp_socket;
240 		/*
241 		 * Adjust backpressure on sender
242 		 * and wakeup any waiting to write.
243 		 */
244 		so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
245 		unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
246 		newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
247 		    so->so_rcv.sb_cc;
248 		(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
249 		    newhiwat, RLIM_INFINITY);
250 		unp->unp_cc = so->so_rcv.sb_cc;
251 		sowwakeup(so2);
252 		break;
253 
254 	default:
255 		panic("uipc_rcvd unknown socktype");
256 	}
257 	return 0;
258 }
259 
260 /* pru_rcvoob is EOPNOTSUPP */
261 
262 static int
263 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
264 	  struct mbuf *control, struct thread *td)
265 {
266 	int error = 0;
267 	struct unpcb *unp = sotounpcb(so);
268 	struct socket *so2;
269 	u_long newhiwat;
270 
271 	if (unp == 0) {
272 		error = EINVAL;
273 		goto release;
274 	}
275 	if (flags & PRUS_OOB) {
276 		error = EOPNOTSUPP;
277 		goto release;
278 	}
279 
280 	if (control && (error = unp_internalize(&control, td)))
281 		goto release;
282 
283 	switch (so->so_type) {
284 	case SOCK_DGRAM:
285 	{
286 		struct sockaddr *from;
287 
288 		if (nam) {
289 			if (unp->unp_conn) {
290 				error = EISCONN;
291 				break;
292 			}
293 			error = unp_connect(so, nam, td);
294 			if (error)
295 				break;
296 		} else {
297 			if (unp->unp_conn == 0) {
298 				error = ENOTCONN;
299 				break;
300 			}
301 		}
302 		so2 = unp->unp_conn->unp_socket;
303 		if (unp->unp_addr)
304 			from = (struct sockaddr *)unp->unp_addr;
305 		else
306 			from = &sun_noname;
307 		if (sbappendaddr(&so2->so_rcv, from, m, control)) {
308 			sorwakeup(so2);
309 			m = 0;
310 			control = 0;
311 		} else
312 			error = ENOBUFS;
313 		if (nam)
314 			unp_disconnect(unp);
315 		break;
316 	}
317 
318 	case SOCK_STREAM:
319 		/* Connect if not connected yet. */
320 		/*
321 		 * Note: A better implementation would complain
322 		 * if not equal to the peer's address.
323 		 */
324 		if ((so->so_state & SS_ISCONNECTED) == 0) {
325 			if (nam) {
326 				error = unp_connect(so, nam, td);
327 				if (error)
328 					break;	/* XXX */
329 			} else {
330 				error = ENOTCONN;
331 				break;
332 			}
333 		}
334 
335 		if (so->so_state & SS_CANTSENDMORE) {
336 			error = EPIPE;
337 			break;
338 		}
339 		if (unp->unp_conn == 0)
340 			panic("uipc_send connected but no connection?");
341 		so2 = unp->unp_conn->unp_socket;
342 		/*
343 		 * Send to paired receive port, and then reduce
344 		 * send buffer hiwater marks to maintain backpressure.
345 		 * Wake up readers.
346 		 */
347 		if (control) {
348 			if (sbappendcontrol(&so2->so_rcv, m, control))
349 				control = 0;
350 		} else
351 			sbappend(&so2->so_rcv, m);
352 		so->so_snd.sb_mbmax -=
353 			so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
354 		unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
355 		newhiwat = so->so_snd.sb_hiwat -
356 		    (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
357 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
358 		    newhiwat, RLIM_INFINITY);
359 		unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
360 		sorwakeup(so2);
361 		m = 0;
362 		break;
363 
364 	default:
365 		panic("uipc_send unknown socktype");
366 	}
367 
368 	/*
369 	 * SEND_EOF is equivalent to a SEND followed by
370 	 * a SHUTDOWN.
371 	 */
372 	if (flags & PRUS_EOF) {
373 		socantsendmore(so);
374 		unp_shutdown(unp);
375 	}
376 
377 	if (control && error != 0)
378 		unp_dispose(control);
379 
380 release:
381 	if (control)
382 		m_freem(control);
383 	if (m)
384 		m_freem(m);
385 	return error;
386 }
387 
388 static int
389 uipc_sense(struct socket *so, struct stat *sb)
390 {
391 	struct unpcb *unp = sotounpcb(so);
392 	struct socket *so2;
393 
394 	if (unp == 0)
395 		return EINVAL;
396 	sb->st_blksize = so->so_snd.sb_hiwat;
397 	if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
398 		so2 = unp->unp_conn->unp_socket;
399 		sb->st_blksize += so2->so_rcv.sb_cc;
400 	}
401 	sb->st_dev = NOUDEV;
402 	if (unp->unp_ino == 0)
403 		unp->unp_ino = unp_ino++;
404 	sb->st_ino = unp->unp_ino;
405 	return (0);
406 }
407 
408 static int
409 uipc_shutdown(struct socket *so)
410 {
411 	struct unpcb *unp = sotounpcb(so);
412 
413 	if (unp == 0)
414 		return EINVAL;
415 	socantsendmore(so);
416 	unp_shutdown(unp);
417 	return 0;
418 }
419 
420 static int
421 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
422 {
423 	struct unpcb *unp = sotounpcb(so);
424 
425 	if (unp == 0)
426 		return EINVAL;
427 	if (unp->unp_addr)
428 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
429 	else
430 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
431 	return 0;
432 }
433 
434 struct pr_usrreqs uipc_usrreqs = {
435 	uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
436 	uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
437 	uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
438 	uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
439 	sosend, soreceive, sopoll
440 };
441 
442 int
443 uipc_ctloutput(so, sopt)
444 	struct socket *so;
445 	struct sockopt *sopt;
446 {
447 	struct unpcb *unp = sotounpcb(so);
448 	int error;
449 
450 	switch (sopt->sopt_dir) {
451 	case SOPT_GET:
452 		switch (sopt->sopt_name) {
453 		case LOCAL_PEERCRED:
454 			if (unp->unp_flags & UNP_HAVEPC)
455 				error = sooptcopyout(sopt, &unp->unp_peercred,
456 				    sizeof(unp->unp_peercred));
457 			else {
458 				if (so->so_type == SOCK_STREAM)
459 					error = ENOTCONN;
460 				else
461 					error = EINVAL;
462 			}
463 			break;
464 		default:
465 			error = EOPNOTSUPP;
466 			break;
467 		}
468 		break;
469 	case SOPT_SET:
470 	default:
471 		error = EOPNOTSUPP;
472 		break;
473 	}
474 	return (error);
475 }
476 
477 /*
478  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
479  * for stream sockets, although the total for sender and receiver is
480  * actually only PIPSIZ.
481  * Datagram sockets really use the sendspace as the maximum datagram size,
482  * and don't really want to reserve the sendspace.  Their recvspace should
483  * be large enough for at least one max-size datagram plus address.
484  */
485 #ifndef PIPSIZ
486 #define	PIPSIZ	8192
487 #endif
488 static u_long	unpst_sendspace = PIPSIZ;
489 static u_long	unpst_recvspace = PIPSIZ;
490 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
491 static u_long	unpdg_recvspace = 4*1024;
492 
493 static int	unp_rights;			/* file descriptors in flight */
494 
495 SYSCTL_DECL(_net_local_stream);
496 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
497 	   &unpst_sendspace, 0, "");
498 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
499 	   &unpst_recvspace, 0, "");
500 SYSCTL_DECL(_net_local_dgram);
501 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
502 	   &unpdg_sendspace, 0, "");
503 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
504 	   &unpdg_recvspace, 0, "");
505 SYSCTL_DECL(_net_local);
506 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
507 
508 static int
509 unp_attach(so)
510 	struct socket *so;
511 {
512 	register struct unpcb *unp;
513 	int error;
514 
515 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
516 		switch (so->so_type) {
517 
518 		case SOCK_STREAM:
519 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
520 			break;
521 
522 		case SOCK_DGRAM:
523 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
524 			break;
525 
526 		default:
527 			panic("unp_attach");
528 		}
529 		if (error)
530 			return (error);
531 	}
532 	unp = uma_zalloc(unp_zone, M_WAITOK);
533 	if (unp == NULL)
534 		return (ENOBUFS);
535 	bzero(unp, sizeof *unp);
536 	unp->unp_gencnt = ++unp_gencnt;
537 	unp_count++;
538 	LIST_INIT(&unp->unp_refs);
539 	unp->unp_socket = so;
540 	FILEDESC_LOCK(curproc->p_fd);
541 	unp->unp_rvnode = curthread->td_proc->p_fd->fd_rdir;
542 	FILEDESC_UNLOCK(curproc->p_fd);
543 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
544 			 : &unp_shead, unp, unp_link);
545 	so->so_pcb = (caddr_t)unp;
546 	return (0);
547 }
548 
549 static void
550 unp_detach(unp)
551 	register struct unpcb *unp;
552 {
553 	LIST_REMOVE(unp, unp_link);
554 	unp->unp_gencnt = ++unp_gencnt;
555 	--unp_count;
556 	if (unp->unp_vnode) {
557 		unp->unp_vnode->v_socket = 0;
558 		vrele(unp->unp_vnode);
559 		unp->unp_vnode = 0;
560 	}
561 	if (unp->unp_conn)
562 		unp_disconnect(unp);
563 	while (!LIST_EMPTY(&unp->unp_refs))
564 		unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
565 	soisdisconnected(unp->unp_socket);
566 	unp->unp_socket->so_pcb = 0;
567 	if (unp_rights) {
568 		/*
569 		 * Normally the receive buffer is flushed later,
570 		 * in sofree, but if our receive buffer holds references
571 		 * to descriptors that are now garbage, we will dispose
572 		 * of those descriptor references after the garbage collector
573 		 * gets them (resulting in a "panic: closef: count < 0").
574 		 */
575 		sorflush(unp->unp_socket);
576 		unp_gc();
577 	}
578 	if (unp->unp_addr)
579 		FREE(unp->unp_addr, M_SONAME);
580 	uma_zfree(unp_zone, unp);
581 }
582 
583 static int
584 unp_bind(unp, nam, td)
585 	struct unpcb *unp;
586 	struct sockaddr *nam;
587 	struct thread *td;
588 {
589 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
590 	struct vnode *vp;
591 	struct mount *mp;
592 	struct vattr vattr;
593 	int error, namelen;
594 	struct nameidata nd;
595 	char *buf;
596 
597 	if (unp->unp_vnode != NULL)
598 		return (EINVAL);
599 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
600 	if (namelen <= 0)
601 		return EINVAL;
602 	buf = malloc(SOCK_MAXADDRLEN, M_TEMP, M_WAITOK);
603 	strncpy(buf, soun->sun_path, namelen);
604 	buf[namelen] = 0;	/* null-terminate the string */
605 restart:
606 	NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT, UIO_SYSSPACE,
607 	    buf, td);
608 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
609 	error = namei(&nd);
610 	if (error) {
611 		free(buf, M_TEMP);
612 		return (error);
613 	}
614 	vp = nd.ni_vp;
615 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
616 		NDFREE(&nd, NDF_ONLY_PNBUF);
617 		if (nd.ni_dvp == vp)
618 			vrele(nd.ni_dvp);
619 		else
620 			vput(nd.ni_dvp);
621 		if (vp != NULL) {
622 			vrele(vp);
623 			free(buf, M_TEMP);
624 			return (EADDRINUSE);
625 		}
626 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
627 		if (error) {
628 			free(buf, M_TEMP);
629 			return (error);
630 		}
631 		goto restart;
632 	}
633 	VATTR_NULL(&vattr);
634 	vattr.va_type = VSOCK;
635 	FILEDESC_LOCK(td->td_proc->p_fd);
636 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
637 	FILEDESC_UNLOCK(td->td_proc->p_fd);
638 	VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
639 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
640 	NDFREE(&nd, NDF_ONLY_PNBUF);
641 	vput(nd.ni_dvp);
642 	if (error) {
643 		free(buf, M_TEMP);
644 		return (error);
645 	}
646 	vp = nd.ni_vp;
647 	vp->v_socket = unp->unp_socket;
648 	unp->unp_vnode = vp;
649 	unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
650 	VOP_UNLOCK(vp, 0, td);
651 	vn_finished_write(mp);
652 	free(buf, M_TEMP);
653 	return (0);
654 }
655 
656 static int
657 unp_connect(so, nam, td)
658 	struct socket *so;
659 	struct sockaddr *nam;
660 	struct thread *td;
661 {
662 	register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
663 	register struct vnode *vp;
664 	register struct socket *so2, *so3;
665 	struct unpcb *unp, *unp2, *unp3;
666 	int error, len;
667 	struct nameidata nd;
668 	char buf[SOCK_MAXADDRLEN];
669 
670 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
671 	if (len <= 0)
672 		return EINVAL;
673 	strncpy(buf, soun->sun_path, len);
674 	buf[len] = 0;
675 
676 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
677 	error = namei(&nd);
678 	if (error)
679 		return (error);
680 	vp = nd.ni_vp;
681 	NDFREE(&nd, NDF_ONLY_PNBUF);
682 	if (vp->v_type != VSOCK) {
683 		error = ENOTSOCK;
684 		goto bad;
685 	}
686 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
687 	if (error)
688 		goto bad;
689 	so2 = vp->v_socket;
690 	if (so2 == 0) {
691 		error = ECONNREFUSED;
692 		goto bad;
693 	}
694 	if (so->so_type != so2->so_type) {
695 		error = EPROTOTYPE;
696 		goto bad;
697 	}
698 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
699 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
700 		    (so3 = sonewconn(so2, 0)) == 0) {
701 			error = ECONNREFUSED;
702 			goto bad;
703 		}
704 		unp = sotounpcb(so);
705 		unp2 = sotounpcb(so2);
706 		unp3 = sotounpcb(so3);
707 		if (unp2->unp_addr)
708 			unp3->unp_addr = (struct sockaddr_un *)
709 				dup_sockaddr((struct sockaddr *)
710 					     unp2->unp_addr, 1);
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(td->td_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(so, so2)
744 	register struct socket *so;
745 	register struct socket *so2;
746 {
747 	register struct unpcb *unp = sotounpcb(so);
748 	register struct unpcb *unp2;
749 
750 	if (so2->so_type != so->so_type)
751 		return (EPROTOTYPE);
752 	unp2 = sotounpcb(so2);
753 	unp->unp_conn = unp2;
754 	switch (so->so_type) {
755 
756 	case SOCK_DGRAM:
757 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
758 		soisconnected(so);
759 		break;
760 
761 	case SOCK_STREAM:
762 		unp2->unp_conn = unp;
763 		soisconnected(so);
764 		soisconnected(so2);
765 		break;
766 
767 	default:
768 		panic("unp_connect2");
769 	}
770 	return (0);
771 }
772 
773 static void
774 unp_disconnect(unp)
775 	struct unpcb *unp;
776 {
777 	register struct unpcb *unp2 = unp->unp_conn;
778 
779 	if (unp2 == 0)
780 		return;
781 	unp->unp_conn = 0;
782 	switch (unp->unp_socket->so_type) {
783 
784 	case SOCK_DGRAM:
785 		LIST_REMOVE(unp, unp_reflink);
786 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
787 		break;
788 
789 	case SOCK_STREAM:
790 		soisdisconnected(unp->unp_socket);
791 		unp2->unp_conn = 0;
792 		soisdisconnected(unp2->unp_socket);
793 		break;
794 	}
795 }
796 
797 #ifdef notdef
798 void
799 unp_abort(unp)
800 	struct unpcb *unp;
801 {
802 
803 	unp_detach(unp);
804 }
805 #endif
806 
807 static int
808 unp_pcblist(SYSCTL_HANDLER_ARGS)
809 {
810 	int error, i, n;
811 	struct unpcb *unp, **unp_list;
812 	unp_gen_t gencnt;
813 	struct xunpgen *xug;
814 	struct unp_head *head;
815 	struct xunpcb *xu;
816 
817 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
818 
819 	/*
820 	 * The process of preparing the PCB list is too time-consuming and
821 	 * resource-intensive to repeat twice on every request.
822 	 */
823 	if (req->oldptr == 0) {
824 		n = unp_count;
825 		req->oldidx = 2 * (sizeof *xug)
826 			+ (n + n/8) * sizeof(struct xunpcb);
827 		return 0;
828 	}
829 
830 	if (req->newptr != 0)
831 		return EPERM;
832 
833 	/*
834 	 * OK, now we're committed to doing something.
835 	 */
836 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
837 	gencnt = unp_gencnt;
838 	n = unp_count;
839 
840 	xug->xug_len = sizeof *xug;
841 	xug->xug_count = n;
842 	xug->xug_gen = gencnt;
843 	xug->xug_sogen = so_gencnt;
844 	error = SYSCTL_OUT(req, xug, sizeof *xug);
845 	if (error) {
846 		free(xug, M_TEMP);
847 		return error;
848 	}
849 
850 	unp_list = malloc(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) {
855 			if (cr_cansee(req->td->td_ucred,
856 			    unp->unp_socket->so_cred))
857 				continue;
858 			unp_list[i++] = unp;
859 		}
860 	}
861 	n = i;			/* in case we lost some during malloc */
862 
863 	error = 0;
864 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK);
865 	for (i = 0; i < n; i++) {
866 		unp = unp_list[i];
867 		if (unp->unp_gencnt <= gencnt) {
868 			xu->xu_len = sizeof *xu;
869 			xu->xu_unpp = unp;
870 			/*
871 			 * XXX - need more locking here to protect against
872 			 * connect/disconnect races for SMP.
873 			 */
874 			if (unp->unp_addr)
875 				bcopy(unp->unp_addr, &xu->xu_addr,
876 				      unp->unp_addr->sun_len);
877 			if (unp->unp_conn && unp->unp_conn->unp_addr)
878 				bcopy(unp->unp_conn->unp_addr,
879 				      &xu->xu_caddr,
880 				      unp->unp_conn->unp_addr->sun_len);
881 			bcopy(unp, &xu->xu_unp, sizeof *unp);
882 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
883 			error = SYSCTL_OUT(req, xu, sizeof *xu);
884 		}
885 	}
886 	free(xu, M_TEMP);
887 	if (!error) {
888 		/*
889 		 * Give the user an updated idea of our state.
890 		 * If the generation differs from what we told
891 		 * her before, she knows that something happened
892 		 * while we were processing this request, and it
893 		 * might be necessary to retry.
894 		 */
895 		xug->xug_gen = unp_gencnt;
896 		xug->xug_sogen = so_gencnt;
897 		xug->xug_count = unp_count;
898 		error = SYSCTL_OUT(req, xug, sizeof *xug);
899 	}
900 	free(unp_list, M_TEMP);
901 	free(xug, M_TEMP);
902 	return error;
903 }
904 
905 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
906 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
907 	    "List of active local datagram sockets");
908 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
909 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
910 	    "List of active local stream sockets");
911 
912 static void
913 unp_shutdown(unp)
914 	struct unpcb *unp;
915 {
916 	struct socket *so;
917 
918 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
919 	    (so = unp->unp_conn->unp_socket))
920 		socantrcvmore(so);
921 }
922 
923 static void
924 unp_drop(unp, errno)
925 	struct unpcb *unp;
926 	int errno;
927 {
928 	struct socket *so = unp->unp_socket;
929 
930 	so->so_error = errno;
931 	unp_disconnect(unp);
932 }
933 
934 #ifdef notdef
935 void
936 unp_drain()
937 {
938 
939 }
940 #endif
941 
942 static void
943 unp_freerights(rp, fdcount)
944 	struct file **rp;
945 	int fdcount;
946 {
947 	int i;
948 	struct file *fp;
949 
950 	for (i = 0; i < fdcount; i++) {
951 		fp = *rp;
952 		/*
953 		 * zero the pointer before calling
954 		 * unp_discard since it may end up
955 		 * in unp_gc()..
956 		 */
957 		*rp++ = 0;
958 		unp_discard(fp);
959 	}
960 }
961 
962 int
963 unp_externalize(control, controlp)
964 	struct mbuf *control, **controlp;
965 {
966 	struct thread *td = curthread;		/* XXX */
967 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
968 	int i;
969 	int *fdp;
970 	struct file **rp;
971 	struct file *fp;
972 	void *data;
973 	socklen_t clen = control->m_len, datalen;
974 	int error, newfds;
975 	int f;
976 	u_int newlen;
977 
978 	error = 0;
979 	if (controlp != NULL) /* controlp == NULL => free control messages */
980 		*controlp = NULL;
981 
982 	while (cm != NULL) {
983 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
984 			error = EINVAL;
985 			break;
986 		}
987 
988 		data = CMSG_DATA(cm);
989 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
990 
991 		if (cm->cmsg_level == SOL_SOCKET
992 		    && cm->cmsg_type == SCM_RIGHTS) {
993 			newfds = datalen / sizeof(struct file *);
994 			rp = data;
995 
996 			/* If we're not outputting the discriptors free them. */
997 			if (error || controlp == NULL) {
998 				unp_freerights(rp, newfds);
999 				goto next;
1000 			}
1001 			FILEDESC_LOCK(td->td_proc->p_fd);
1002 			/* if the new FD's will not fit free them.  */
1003 			if (!fdavail(td, newfds)) {
1004 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1005 				error = EMSGSIZE;
1006 				unp_freerights(rp, newfds);
1007 				goto next;
1008 			}
1009 			/*
1010 			 * now change each pointer to an fd in the global
1011 			 * table to an integer that is the index to the
1012 			 * local fd table entry that we set up to point
1013 			 * to the global one we are transferring.
1014 			 */
1015 			newlen = newfds * sizeof(int);
1016 			*controlp = sbcreatecontrol(NULL, newlen,
1017 			    SCM_RIGHTS, SOL_SOCKET);
1018 			if (*controlp == NULL) {
1019 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1020 				error = E2BIG;
1021 				unp_freerights(rp, newfds);
1022 				goto next;
1023 			}
1024 
1025 			fdp = (int *)
1026 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1027 			for (i = 0; i < newfds; i++) {
1028 				if (fdalloc(td, 0, &f))
1029 					panic("unp_externalize fdalloc failed");
1030 				fp = *rp++;
1031 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1032 				FILE_LOCK(fp);
1033 				fp->f_msgcount--;
1034 				FILE_UNLOCK(fp);
1035 				unp_rights--;
1036 				*fdp++ = f;
1037 			}
1038 			FILEDESC_UNLOCK(td->td_proc->p_fd);
1039 		} else { /* We can just copy anything else across */
1040 			if (error || controlp == NULL)
1041 				goto next;
1042 			*controlp = sbcreatecontrol(NULL, datalen,
1043 			    cm->cmsg_type, cm->cmsg_level);
1044 			if (*controlp == NULL) {
1045 				error = ENOBUFS;
1046 				goto next;
1047 			}
1048 			bcopy(data,
1049 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1050 			    datalen);
1051 		}
1052 
1053 		controlp = &(*controlp)->m_next;
1054 
1055 next:
1056 		if (CMSG_SPACE(datalen) < clen) {
1057 			clen -= CMSG_SPACE(datalen);
1058 			cm = (struct cmsghdr *)
1059 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1060 		} else {
1061 			clen = 0;
1062 			cm = NULL;
1063 		}
1064 	}
1065 
1066 	m_freem(control);
1067 
1068 	return (error);
1069 }
1070 
1071 void
1072 unp_init(void)
1073 {
1074 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1075 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1076 	uma_zone_set_max(unp_zone, nmbclusters);
1077 	if (unp_zone == 0)
1078 		panic("unp_init");
1079 	LIST_INIT(&unp_dhead);
1080 	LIST_INIT(&unp_shead);
1081 }
1082 
1083 #ifndef MIN
1084 #define	MIN(a,b) (((a)<(b))?(a):(b))
1085 #endif
1086 
1087 static int
1088 unp_internalize(controlp, td)
1089 	struct mbuf **controlp;
1090 	struct thread *td;
1091 {
1092 	struct mbuf *control = *controlp;
1093 	struct proc *p = td->td_proc;
1094 	struct filedesc *fdescp = p->p_fd;
1095 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1096 	struct cmsgcred *cmcred;
1097 	struct file **rp;
1098 	struct file *fp;
1099 	struct timeval *tv;
1100 	int i, fd, *fdp;
1101 	void *data;
1102 	socklen_t clen = control->m_len, datalen;
1103 	int error, oldfds;
1104 	u_int newlen;
1105 
1106 	error = 0;
1107 	*controlp = NULL;
1108 
1109 	while (cm != NULL) {
1110 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1111 		    || cm->cmsg_len > clen) {
1112 			error = EINVAL;
1113 			goto out;
1114 		}
1115 
1116 		data = CMSG_DATA(cm);
1117 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1118 
1119 		switch (cm->cmsg_type) {
1120 		/*
1121 		 * Fill in credential information.
1122 		 */
1123 		case SCM_CREDS:
1124 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1125 			    SCM_CREDS, SOL_SOCKET);
1126 			if (*controlp == NULL) {
1127 				error = ENOBUFS;
1128 				goto out;
1129 			}
1130 
1131 			cmcred = (struct cmsgcred *)
1132 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1133 			cmcred->cmcred_pid = p->p_pid;
1134 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1135 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1136 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1137 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1138 							CMGROUP_MAX);
1139 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1140 				cmcred->cmcred_groups[i] =
1141 				    td->td_ucred->cr_groups[i];
1142 			break;
1143 
1144 		case SCM_RIGHTS:
1145 			oldfds = datalen / sizeof (int);
1146 			/*
1147 			 * check that all the FDs passed in refer to legal files
1148 			 * If not, reject the entire operation.
1149 			 */
1150 			fdp = data;
1151 			FILEDESC_LOCK(fdescp);
1152 			for (i = 0; i < oldfds; i++) {
1153 				fd = *fdp++;
1154 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1155 				    fdescp->fd_ofiles[fd] == NULL) {
1156 					FILEDESC_UNLOCK(fdescp);
1157 					error = EBADF;
1158 					goto out;
1159 				}
1160 			}
1161 			/*
1162 			 * Now replace the integer FDs with pointers to
1163 			 * the associated global file table entry..
1164 			 */
1165 			newlen = oldfds * sizeof(struct file *);
1166 			*controlp = sbcreatecontrol(NULL, newlen,
1167 			    SCM_RIGHTS, SOL_SOCKET);
1168 			if (*controlp == NULL) {
1169 				FILEDESC_UNLOCK(fdescp);
1170 				error = E2BIG;
1171 				goto out;
1172 			}
1173 
1174 			fdp = data;
1175 			rp = (struct file **)
1176 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1177 			for (i = 0; i < oldfds; i++) {
1178 				fp = fdescp->fd_ofiles[*fdp++];
1179 				*rp++ = fp;
1180 				FILE_LOCK(fp);
1181 				fp->f_count++;
1182 				fp->f_msgcount++;
1183 				FILE_UNLOCK(fp);
1184 				unp_rights++;
1185 			}
1186 			FILEDESC_UNLOCK(fdescp);
1187 			break;
1188 
1189 		case SCM_TIMESTAMP:
1190 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1191 			    SCM_TIMESTAMP, SOL_SOCKET);
1192 			if (*controlp == NULL) {
1193 				error = ENOBUFS;
1194 				goto out;
1195 			}
1196 			tv = (struct timeval *)
1197 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1198 			microtime(tv);
1199 			break;
1200 
1201 		default:
1202 			error = EINVAL;
1203 			goto out;
1204 		}
1205 
1206 		controlp = &(*controlp)->m_next;
1207 
1208 		if (CMSG_SPACE(datalen) < clen) {
1209 			clen -= CMSG_SPACE(datalen);
1210 			cm = (struct cmsghdr *)
1211 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1212 		} else {
1213 			clen = 0;
1214 			cm = NULL;
1215 		}
1216 	}
1217 
1218 out:
1219 	m_freem(control);
1220 
1221 	return (error);
1222 }
1223 
1224 static int	unp_defer, unp_gcing;
1225 
1226 static void
1227 unp_gc()
1228 {
1229 	register struct file *fp, *nextfp;
1230 	register struct socket *so;
1231 	struct file **extra_ref, **fpp;
1232 	int nunref, i;
1233 
1234 	if (unp_gcing)
1235 		return;
1236 	unp_gcing = 1;
1237 	unp_defer = 0;
1238 	/*
1239 	 * before going through all this, set all FDs to
1240 	 * be NOT defered and NOT externally accessible
1241 	 */
1242 	sx_slock(&filelist_lock);
1243 	LIST_FOREACH(fp, &filehead, f_list)
1244 		fp->f_gcflag &= ~(FMARK|FDEFER);
1245 	do {
1246 		LIST_FOREACH(fp, &filehead, f_list) {
1247 			FILE_LOCK(fp);
1248 			/*
1249 			 * If the file is not open, skip it
1250 			 */
1251 			if (fp->f_count == 0) {
1252 				FILE_UNLOCK(fp);
1253 				continue;
1254 			}
1255 			/*
1256 			 * If we already marked it as 'defer'  in a
1257 			 * previous pass, then try process it this time
1258 			 * and un-mark it
1259 			 */
1260 			if (fp->f_gcflag & FDEFER) {
1261 				fp->f_gcflag &= ~FDEFER;
1262 				unp_defer--;
1263 			} else {
1264 				/*
1265 				 * if it's not defered, then check if it's
1266 				 * already marked.. if so skip it
1267 				 */
1268 				if (fp->f_gcflag & FMARK) {
1269 					FILE_UNLOCK(fp);
1270 					continue;
1271 				}
1272 				/*
1273 				 * If all references are from messages
1274 				 * in transit, then skip it. it's not
1275 				 * externally accessible.
1276 				 */
1277 				if (fp->f_count == fp->f_msgcount) {
1278 					FILE_UNLOCK(fp);
1279 					continue;
1280 				}
1281 				/*
1282 				 * If it got this far then it must be
1283 				 * externally accessible.
1284 				 */
1285 				fp->f_gcflag |= FMARK;
1286 			}
1287 			/*
1288 			 * either it was defered, or it is externally
1289 			 * accessible and not already marked so.
1290 			 * Now check if it is possibly one of OUR sockets.
1291 			 */
1292 			if (fp->f_type != DTYPE_SOCKET ||
1293 			    (so = (struct socket *)fp->f_data) == 0) {
1294 				FILE_UNLOCK(fp);
1295 				continue;
1296 			}
1297 			FILE_UNLOCK(fp);
1298 			if (so->so_proto->pr_domain != &localdomain ||
1299 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1300 				continue;
1301 #ifdef notdef
1302 			if (so->so_rcv.sb_flags & SB_LOCK) {
1303 				/*
1304 				 * This is problematical; it's not clear
1305 				 * we need to wait for the sockbuf to be
1306 				 * unlocked (on a uniprocessor, at least),
1307 				 * and it's also not clear what to do
1308 				 * if sbwait returns an error due to receipt
1309 				 * of a signal.  If sbwait does return
1310 				 * an error, we'll go into an infinite
1311 				 * loop.  Delete all of this for now.
1312 				 */
1313 				(void) sbwait(&so->so_rcv);
1314 				goto restart;
1315 			}
1316 #endif
1317 			/*
1318 			 * So, Ok, it's one of our sockets and it IS externally
1319 			 * accessible (or was defered). Now we look
1320 			 * to see if we hold any file descriptors in its
1321 			 * message buffers. Follow those links and mark them
1322 			 * as accessible too.
1323 			 */
1324 			unp_scan(so->so_rcv.sb_mb, unp_mark);
1325 		}
1326 	} while (unp_defer);
1327 	sx_sunlock(&filelist_lock);
1328 	/*
1329 	 * We grab an extra reference to each of the file table entries
1330 	 * that are not otherwise accessible and then free the rights
1331 	 * that are stored in messages on them.
1332 	 *
1333 	 * The bug in the orginal code is a little tricky, so I'll describe
1334 	 * what's wrong with it here.
1335 	 *
1336 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1337 	 * times -- consider the case of sockets A and B that contain
1338 	 * references to each other.  On a last close of some other socket,
1339 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1340 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1341 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1342 	 * results in the following chain.  Closef calls soo_close, which
1343 	 * calls soclose.   Soclose calls first (through the switch
1344 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1345 	 * returns because the previous instance had set unp_gcing, and
1346 	 * we return all the way back to soclose, which marks the socket
1347 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1348 	 * to free up the rights that are queued in messages on the socket A,
1349 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1350 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1351 	 * instance of unp_discard just calls closef on B.
1352 	 *
1353 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1354 	 * which results in another closef on A.  Unfortunately, A is already
1355 	 * being closed, and the descriptor has already been marked with
1356 	 * SS_NOFDREF, and soclose panics at this point.
1357 	 *
1358 	 * Here, we first take an extra reference to each inaccessible
1359 	 * descriptor.  Then, we call sorflush ourself, since we know
1360 	 * it is a Unix domain socket anyhow.  After we destroy all the
1361 	 * rights carried in messages, we do a last closef to get rid
1362 	 * of our extra reference.  This is the last close, and the
1363 	 * unp_detach etc will shut down the socket.
1364 	 *
1365 	 * 91/09/19, bsy@cs.cmu.edu
1366 	 */
1367 	extra_ref = malloc(nfiles * sizeof(struct file *), M_TEMP, M_WAITOK);
1368 	sx_slock(&filelist_lock);
1369 	for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref; fp != 0;
1370 	    fp = nextfp) {
1371 		nextfp = LIST_NEXT(fp, f_list);
1372 		FILE_LOCK(fp);
1373 		/*
1374 		 * If it's not open, skip it
1375 		 */
1376 		if (fp->f_count == 0) {
1377 			FILE_UNLOCK(fp);
1378 			continue;
1379 		}
1380 		/*
1381 		 * If all refs are from msgs, and it's not marked accessible
1382 		 * then it must be referenced from some unreachable cycle
1383 		 * of (shut-down) FDs, so include it in our
1384 		 * list of FDs to remove
1385 		 */
1386 		if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1387 			*fpp++ = fp;
1388 			nunref++;
1389 			fp->f_count++;
1390 		}
1391 		FILE_UNLOCK(fp);
1392 	}
1393 	sx_sunlock(&filelist_lock);
1394 	/*
1395 	 * for each FD on our hit list, do the following two things
1396 	 */
1397 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1398 		struct file *tfp = *fpp;
1399 		FILE_LOCK(tfp);
1400 		if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL) {
1401 			FILE_UNLOCK(tfp);
1402 			sorflush((struct socket *)(tfp->f_data));
1403 		} else
1404 			FILE_UNLOCK(tfp);
1405 	}
1406 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1407 		closef(*fpp, (struct thread *) NULL);
1408 	free((caddr_t)extra_ref, M_TEMP);
1409 	unp_gcing = 0;
1410 }
1411 
1412 void
1413 unp_dispose(m)
1414 	struct mbuf *m;
1415 {
1416 
1417 	if (m)
1418 		unp_scan(m, unp_discard);
1419 }
1420 
1421 static int
1422 unp_listen(unp, td)
1423 	struct unpcb *unp;
1424 	struct thread *td;
1425 {
1426 
1427 	cru2x(td->td_ucred, &unp->unp_peercred);
1428 	unp->unp_flags |= UNP_HAVEPCCACHED;
1429 	return (0);
1430 }
1431 
1432 static void
1433 unp_scan(m0, op)
1434 	register struct mbuf *m0;
1435 	void (*op)(struct file *);
1436 {
1437 	struct mbuf *m;
1438 	struct file **rp;
1439 	struct cmsghdr *cm;
1440 	void *data;
1441 	int i;
1442 	socklen_t clen, datalen;
1443 	int qfds;
1444 
1445 	while (m0) {
1446 		for (m = m0; m; m = m->m_next) {
1447 			if (m->m_type != MT_CONTROL)
1448 				continue;
1449 
1450 			cm = mtod(m, struct cmsghdr *);
1451 			clen = m->m_len;
1452 
1453 			while (cm != NULL) {
1454 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1455 					break;
1456 
1457 				data = CMSG_DATA(cm);
1458 				datalen = (caddr_t)cm + cm->cmsg_len
1459 				    - (caddr_t)data;
1460 
1461 				if (cm->cmsg_level == SOL_SOCKET &&
1462 				    cm->cmsg_type == SCM_RIGHTS) {
1463 					qfds = datalen / sizeof (struct file *);
1464 					rp = data;
1465 					for (i = 0; i < qfds; i++)
1466 						(*op)(*rp++);
1467 				}
1468 
1469 				if (CMSG_SPACE(datalen) < clen) {
1470 					clen -= CMSG_SPACE(datalen);
1471 					cm = (struct cmsghdr *)
1472 					    ((caddr_t)cm + CMSG_SPACE(datalen));
1473 				} else {
1474 					clen = 0;
1475 					cm = NULL;
1476 				}
1477 			}
1478 		}
1479 		m0 = m0->m_act;
1480 	}
1481 }
1482 
1483 static void
1484 unp_mark(fp)
1485 	struct file *fp;
1486 {
1487 	if (fp->f_gcflag & FMARK)
1488 		return;
1489 	unp_defer++;
1490 	fp->f_gcflag |= (FMARK|FDEFER);
1491 }
1492 
1493 static void
1494 unp_discard(fp)
1495 	struct file *fp;
1496 {
1497 	FILE_LOCK(fp);
1498 	fp->f_msgcount--;
1499 	unp_rights--;
1500 	FILE_UNLOCK(fp);
1501 	(void) closef(fp, (struct thread *)NULL);
1502 }
1503