xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision 32af04f740541c3e943edbdb70f13dff978147be)
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 1.44 2008/09/06 05:44:58 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 
63 static	MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
64 static	unp_gen_t unp_gencnt;
65 static	u_int unp_count;
66 
67 static	struct unp_head unp_shead, unp_dhead;
68 
69 /*
70  * Unix communications domain.
71  *
72  * TODO:
73  *	RDM
74  *	rethink name space problems
75  *	need a proper out-of-band
76  *	lock pushdown
77  */
78 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
79 static ino_t	unp_ino = 1;		/* prototype for fake inode numbers */
80 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin);
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 int     unp_revoke_gc_check(struct file *, void *);
95 static void    unp_scan (struct mbuf *, void (*)(struct file *, void *),
96 				void *data);
97 static void    unp_mark (struct file *, void *data);
98 static void    unp_discard (struct file *, void *);
99 static int     unp_internalize (struct mbuf *, struct thread *);
100 static int     unp_listen (struct unpcb *, struct thread *);
101 static void    unp_fp_externalize(struct proc *p, struct file *fp, int fd);
102 
103 static int
104 uipc_abort(struct socket *so)
105 {
106 	struct unpcb *unp = so->so_pcb;
107 
108 	if (unp == NULL)
109 		return EINVAL;
110 	unp_drop(unp, ECONNABORTED);
111 	unp_detach(unp);
112 	sofree(so);
113 	return 0;
114 }
115 
116 static int
117 uipc_accept(struct socket *so, struct sockaddr **nam)
118 {
119 	struct unpcb *unp = so->so_pcb;
120 
121 	if (unp == NULL)
122 		return EINVAL;
123 
124 	/*
125 	 * Pass back name of connected socket,
126 	 * if it was bound and we are still connected
127 	 * (our peer may have closed already!).
128 	 */
129 	if (unp->unp_conn && unp->unp_conn->unp_addr) {
130 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
131 	} else {
132 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname);
133 	}
134 	return 0;
135 }
136 
137 static int
138 uipc_attach(struct socket *so, int proto, struct pru_attach_info *ai)
139 {
140 	struct unpcb *unp = so->so_pcb;
141 
142 	if (unp != NULL)
143 		return EISCONN;
144 	return unp_attach(so, ai);
145 }
146 
147 static int
148 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
149 {
150 	struct unpcb *unp = so->so_pcb;
151 
152 	if (unp == NULL)
153 		return EINVAL;
154 	return unp_bind(unp, nam, td);
155 }
156 
157 static int
158 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
159 {
160 	struct unpcb *unp = so->so_pcb;
161 
162 	if (unp == NULL)
163 		return EINVAL;
164 	return unp_connect(so, nam, td);
165 }
166 
167 static int
168 uipc_connect2(struct socket *so1, struct socket *so2)
169 {
170 	struct unpcb *unp = so1->so_pcb;
171 
172 	if (unp == NULL)
173 		return EINVAL;
174 
175 	return unp_connect2(so1, so2);
176 }
177 
178 /* control is EOPNOTSUPP */
179 
180 static int
181 uipc_detach(struct socket *so)
182 {
183 	struct unpcb *unp = so->so_pcb;
184 
185 	if (unp == NULL)
186 		return EINVAL;
187 
188 	unp_detach(unp);
189 	return 0;
190 }
191 
192 static int
193 uipc_disconnect(struct socket *so)
194 {
195 	struct unpcb *unp = so->so_pcb;
196 
197 	if (unp == NULL)
198 		return EINVAL;
199 	unp_disconnect(unp);
200 	return 0;
201 }
202 
203 static int
204 uipc_listen(struct socket *so, struct thread *td)
205 {
206 	struct unpcb *unp = so->so_pcb;
207 
208 	if (unp == NULL || unp->unp_vnode == NULL)
209 		return EINVAL;
210 	return unp_listen(unp, td);
211 }
212 
213 static int
214 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
215 {
216 	struct unpcb *unp = so->so_pcb;
217 
218 	if (unp == NULL)
219 		return EINVAL;
220 	if (unp->unp_conn && unp->unp_conn->unp_addr)
221 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr);
222 	else {
223 		/*
224 		 * XXX: It seems that this test always fails even when
225 		 * connection is established.  So, this else clause is
226 		 * added as workaround to return PF_LOCAL sockaddr.
227 		 */
228 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname);
229 	}
230 	return 0;
231 }
232 
233 static int
234 uipc_rcvd(struct socket *so, int flags)
235 {
236 	struct unpcb *unp = so->so_pcb;
237 	struct socket *so2;
238 
239 	if (unp == NULL)
240 		return EINVAL;
241 	switch (so->so_type) {
242 	case SOCK_DGRAM:
243 		panic("uipc_rcvd DGRAM?");
244 		/*NOTREACHED*/
245 
246 	case SOCK_STREAM:
247 	case SOCK_SEQPACKET:
248 		if (unp->unp_conn == NULL)
249 			break;
250 		/*
251 		 * Because we are transfering mbufs directly to the
252 		 * peer socket we have to use SSB_STOP on the sender
253 		 * to prevent it from building up infinite mbufs.
254 		 */
255 		so2 = unp->unp_conn->unp_socket;
256 		if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
257 		    so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
258 		) {
259 			so2->so_snd.ssb_flags &= ~SSB_STOP;
260 			sowwakeup(so2);
261 		}
262 		break;
263 
264 	default:
265 		panic("uipc_rcvd unknown socktype");
266 	}
267 	return 0;
268 }
269 
270 /* pru_rcvoob is EOPNOTSUPP */
271 
272 static int
273 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
274 	  struct mbuf *control, struct thread *td)
275 {
276 	int error = 0;
277 	struct unpcb *unp = so->so_pcb;
278 	struct socket *so2;
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 	case SOCK_SEQPACKET:
330 		/* Connect if not connected yet. */
331 		/*
332 		 * Note: A better implementation would complain
333 		 * if not equal to the peer's address.
334 		 */
335 		if (!(so->so_state & SS_ISCONNECTED)) {
336 			if (nam) {
337 				error = unp_connect(so, nam, td);
338 				if (error)
339 					break;	/* XXX */
340 			} else {
341 				error = ENOTCONN;
342 				break;
343 			}
344 		}
345 
346 		if (so->so_state & SS_CANTSENDMORE) {
347 			error = EPIPE;
348 			break;
349 		}
350 		if (unp->unp_conn == NULL)
351 			panic("uipc_send connected but no connection?");
352 		so2 = unp->unp_conn->unp_socket;
353 		/*
354 		 * Send to paired receive port, and then reduce
355 		 * send buffer hiwater marks to maintain backpressure.
356 		 * Wake up readers.
357 		 */
358 		if (control) {
359 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
360 				control = NULL;
361 				m = NULL;
362 			}
363 		} else if (so->so_type == SOCK_SEQPACKET) {
364 			sbappendrecord(&so2->so_rcv.sb, m);
365 			m = NULL;
366 		} else {
367 			sbappend(&so2->so_rcv.sb, m);
368 			m = NULL;
369 		}
370 
371 		/*
372 		 * Because we are transfering mbufs directly to the
373 		 * peer socket we have to use SSB_STOP on the sender
374 		 * to prevent it from building up infinite mbufs.
375 		 */
376 		if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
377 		    so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
378 		) {
379 			so->so_snd.ssb_flags |= SSB_STOP;
380 		}
381 		sorwakeup(so2);
382 		break;
383 
384 	default:
385 		panic("uipc_send unknown socktype");
386 	}
387 
388 	/*
389 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
390 	 */
391 	if (flags & PRUS_EOF) {
392 		socantsendmore(so);
393 		unp_shutdown(unp);
394 	}
395 
396 	if (control && error != 0)
397 		unp_dispose(control);
398 
399 release:
400 	if (control)
401 		m_freem(control);
402 	if (m)
403 		m_freem(m);
404 	return error;
405 }
406 
407 /*
408  * MPSAFE
409  */
410 static int
411 uipc_sense(struct socket *so, struct stat *sb)
412 {
413 	struct unpcb *unp = so->so_pcb;
414 
415 	if (unp == NULL)
416 		return EINVAL;
417 	sb->st_blksize = so->so_snd.ssb_hiwat;
418 	sb->st_dev = NOUDEV;
419 	if (unp->unp_ino == 0) {	/* make up a non-zero inode number */
420 		spin_lock_wr(&unp_ino_spin);
421 		unp->unp_ino = unp_ino++;
422 		spin_unlock_wr(&unp_ino_spin);
423 	}
424 	sb->st_ino = unp->unp_ino;
425 	return (0);
426 }
427 
428 static int
429 uipc_shutdown(struct socket *so)
430 {
431 	struct unpcb *unp = so->so_pcb;
432 
433 	if (unp == NULL)
434 		return EINVAL;
435 	socantsendmore(so);
436 	unp_shutdown(unp);
437 	return 0;
438 }
439 
440 static int
441 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
442 {
443 	struct unpcb *unp = so->so_pcb;
444 
445 	if (unp == NULL)
446 		return EINVAL;
447 	if (unp->unp_addr)
448 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_addr);
449 	return 0;
450 }
451 
452 struct pr_usrreqs uipc_usrreqs = {
453 	.pru_abort = uipc_abort,
454 	.pru_accept = uipc_accept,
455 	.pru_attach = uipc_attach,
456 	.pru_bind = uipc_bind,
457 	.pru_connect = uipc_connect,
458 	.pru_connect2 = uipc_connect2,
459 	.pru_control = pru_control_notsupp,
460 	.pru_detach = uipc_detach,
461 	.pru_disconnect = uipc_disconnect,
462 	.pru_listen = uipc_listen,
463 	.pru_peeraddr = uipc_peeraddr,
464 	.pru_rcvd = uipc_rcvd,
465 	.pru_rcvoob = pru_rcvoob_notsupp,
466 	.pru_send = uipc_send,
467 	.pru_sense = uipc_sense,
468 	.pru_shutdown = uipc_shutdown,
469 	.pru_sockaddr = uipc_sockaddr,
470 	.pru_sosend = sosend,
471 	.pru_soreceive = soreceive,
472 	.pru_sopoll = sopoll
473 };
474 
475 int
476 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
477 {
478 	struct unpcb *unp = so->so_pcb;
479 	int error = 0;
480 
481 	switch (sopt->sopt_dir) {
482 	case SOPT_GET:
483 		switch (sopt->sopt_name) {
484 		case LOCAL_PEERCRED:
485 			if (unp->unp_flags & UNP_HAVEPC)
486 				soopt_from_kbuf(sopt, &unp->unp_peercred,
487 						sizeof(unp->unp_peercred));
488 			else {
489 				if (so->so_type == SOCK_STREAM)
490 					error = ENOTCONN;
491 				else if (so->so_type == SOCK_SEQPACKET)
492 					error = ENOTCONN;
493 				else
494 					error = EINVAL;
495 			}
496 			break;
497 		default:
498 			error = EOPNOTSUPP;
499 			break;
500 		}
501 		break;
502 	case SOPT_SET:
503 	default:
504 		error = EOPNOTSUPP;
505 		break;
506 	}
507 	return (error);
508 }
509 
510 /*
511  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
512  * for stream sockets, although the total for sender and receiver is
513  * actually only PIPSIZ.
514  *
515  * Datagram sockets really use the sendspace as the maximum datagram size,
516  * and don't really want to reserve the sendspace.  Their recvspace should
517  * be large enough for at least one max-size datagram plus address.
518  *
519  * We want the local send/recv space to be significant larger then lo0's
520  * mtu of 16384.
521  */
522 #ifndef PIPSIZ
523 #define	PIPSIZ	57344
524 #endif
525 static u_long	unpst_sendspace = PIPSIZ;
526 static u_long	unpst_recvspace = PIPSIZ;
527 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
528 static u_long	unpdg_recvspace = 4*1024;
529 
530 static int	unp_rights;			/* file descriptors in flight */
531 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin);
532 
533 SYSCTL_DECL(_net_local_seqpacket);
534 SYSCTL_DECL(_net_local_stream);
535 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
536 	   &unpst_sendspace, 0, "");
537 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
538 	   &unpst_recvspace, 0, "");
539 
540 SYSCTL_DECL(_net_local_dgram);
541 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
542 	   &unpdg_sendspace, 0, "");
543 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
544 	   &unpdg_recvspace, 0, "");
545 
546 SYSCTL_DECL(_net_local);
547 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
548 
549 static int
550 unp_attach(struct socket *so, struct pru_attach_info *ai)
551 {
552 	struct unpcb *unp;
553 	int error;
554 
555 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
556 		switch (so->so_type) {
557 
558 		case SOCK_STREAM:
559 		case SOCK_SEQPACKET:
560 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
561 					  ai->sb_rlimit);
562 			break;
563 
564 		case SOCK_DGRAM:
565 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
566 					  ai->sb_rlimit);
567 			break;
568 
569 		default:
570 			panic("unp_attach");
571 		}
572 		if (error)
573 			return (error);
574 	}
575 	unp = kmalloc(sizeof(*unp), M_UNPCB, M_NOWAIT|M_ZERO);
576 	if (unp == NULL)
577 		return (ENOBUFS);
578 	unp->unp_gencnt = ++unp_gencnt;
579 	unp_count++;
580 	LIST_INIT(&unp->unp_refs);
581 	unp->unp_socket = so;
582 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
583 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
584 			 : &unp_shead, unp, unp_link);
585 	so->so_pcb = (caddr_t)unp;
586 	so->so_port = sync_soport(so, NULL, NULL);
587 	return (0);
588 }
589 
590 static void
591 unp_detach(struct unpcb *unp)
592 {
593 	LIST_REMOVE(unp, unp_link);
594 	unp->unp_gencnt = ++unp_gencnt;
595 	--unp_count;
596 	if (unp->unp_vnode) {
597 		unp->unp_vnode->v_socket = NULL;
598 		vrele(unp->unp_vnode);
599 		unp->unp_vnode = NULL;
600 	}
601 	if (unp->unp_conn)
602 		unp_disconnect(unp);
603 	while (!LIST_EMPTY(&unp->unp_refs))
604 		unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
605 	soisdisconnected(unp->unp_socket);
606 	unp->unp_socket->so_pcb = NULL;
607 	if (unp_rights) {
608 		/*
609 		 * Normally the receive buffer is flushed later,
610 		 * in sofree, but if our receive buffer holds references
611 		 * to descriptors that are now garbage, we will dispose
612 		 * of those descriptor references after the garbage collector
613 		 * gets them (resulting in a "panic: closef: count < 0").
614 		 */
615 		sorflush(unp->unp_socket);
616 		unp_gc();
617 	}
618 	if (unp->unp_addr)
619 		kfree(unp->unp_addr, M_SONAME);
620 	kfree(unp, M_UNPCB);
621 }
622 
623 static int
624 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
625 {
626 	struct proc *p = td->td_proc;
627 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
628 	struct vnode *vp;
629 	struct vattr vattr;
630 	int error, namelen;
631 	struct nlookupdata nd;
632 	char buf[SOCK_MAXADDRLEN];
633 
634 	if (unp->unp_vnode != NULL)
635 		return (EINVAL);
636 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
637 	if (namelen <= 0)
638 		return (EINVAL);
639 	strncpy(buf, soun->sun_path, namelen);
640 	buf[namelen] = 0;	/* null-terminate the string */
641 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
642 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
643 	if (error == 0)
644 		error = nlookup(&nd);
645 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
646 		error = EADDRINUSE;
647 	if (error)
648 		goto done;
649 
650 	VATTR_NULL(&vattr);
651 	vattr.va_type = VSOCK;
652 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
653 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
654 	if (error == 0) {
655 		vp->v_socket = unp->unp_socket;
656 		unp->unp_vnode = vp;
657 		unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
658 		vn_unlock(vp);
659 	}
660 done:
661 	nlookup_done(&nd);
662 	return (error);
663 }
664 
665 static int
666 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
667 {
668 	struct proc *p = td->td_proc;
669 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
670 	struct vnode *vp;
671 	struct socket *so2, *so3;
672 	struct unpcb *unp, *unp2, *unp3;
673 	int error, len;
674 	struct nlookupdata nd;
675 	char buf[SOCK_MAXADDRLEN];
676 
677 	KKASSERT(p);
678 
679 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
680 	if (len <= 0)
681 		return EINVAL;
682 	strncpy(buf, soun->sun_path, len);
683 	buf[len] = 0;
684 
685 	vp = NULL;
686 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
687 	if (error == 0)
688 		error = nlookup(&nd);
689 	if (error == 0)
690 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
691 	nlookup_done(&nd);
692 	if (error)
693 		return (error);
694 
695 	if (vp->v_type != VSOCK) {
696 		error = ENOTSOCK;
697 		goto bad;
698 	}
699 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred);
700 	if (error)
701 		goto bad;
702 	so2 = vp->v_socket;
703 	if (so2 == NULL) {
704 		error = ECONNREFUSED;
705 		goto bad;
706 	}
707 	if (so->so_type != so2->so_type) {
708 		error = EPROTOTYPE;
709 		goto bad;
710 	}
711 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
712 		if (!(so2->so_options & SO_ACCEPTCONN) ||
713 		    (so3 = sonewconn(so2, 0)) == NULL) {
714 			error = ECONNREFUSED;
715 			goto bad;
716 		}
717 		unp = so->so_pcb;
718 		unp2 = so2->so_pcb;
719 		unp3 = so3->so_pcb;
720 		if (unp2->unp_addr)
721 			unp3->unp_addr = (struct sockaddr_un *)
722 				dup_sockaddr((struct sockaddr *)unp2->unp_addr);
723 
724 		/*
725 		 * unp_peercred management:
726 		 *
727 		 * The connecter's (client's) credentials are copied
728 		 * from its process structure at the time of connect()
729 		 * (which is now).
730 		 */
731 		cru2x(p->p_ucred, &unp3->unp_peercred);
732 		unp3->unp_flags |= UNP_HAVEPC;
733 		/*
734 		 * The receiver's (server's) credentials are copied
735 		 * from the unp_peercred member of socket on which the
736 		 * former called listen(); unp_listen() cached that
737 		 * process's credentials at that time so we can use
738 		 * them now.
739 		 */
740 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
741 		    ("unp_connect: listener without cached peercred"));
742 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
743 		    sizeof(unp->unp_peercred));
744 		unp->unp_flags |= UNP_HAVEPC;
745 
746 		so2 = so3;
747 	}
748 	error = unp_connect2(so, so2);
749 bad:
750 	vput(vp);
751 	return (error);
752 }
753 
754 int
755 unp_connect2(struct socket *so, struct socket *so2)
756 {
757 	struct unpcb *unp = so->so_pcb;
758 	struct unpcb *unp2;
759 
760 	if (so2->so_type != so->so_type)
761 		return (EPROTOTYPE);
762 	unp2 = so2->so_pcb;
763 	unp->unp_conn = unp2;
764 	switch (so->so_type) {
765 
766 	case SOCK_DGRAM:
767 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
768 		soisconnected(so);
769 		break;
770 
771 	case SOCK_STREAM:
772 	case SOCK_SEQPACKET:
773 		unp2->unp_conn = unp;
774 		soisconnected(so);
775 		soisconnected(so2);
776 		break;
777 
778 	default:
779 		panic("unp_connect2");
780 	}
781 	return (0);
782 }
783 
784 static void
785 unp_disconnect(struct unpcb *unp)
786 {
787 	struct unpcb *unp2 = unp->unp_conn;
788 
789 	if (unp2 == NULL)
790 		return;
791 
792 	unp->unp_conn = NULL;
793 
794 	switch (unp->unp_socket->so_type) {
795 	case SOCK_DGRAM:
796 		LIST_REMOVE(unp, unp_reflink);
797 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
798 		break;
799 	case SOCK_STREAM:
800 	case SOCK_SEQPACKET:
801 		soisdisconnected(unp->unp_socket);
802 		unp2->unp_conn = NULL;
803 		soisdisconnected(unp2->unp_socket);
804 		break;
805 	}
806 }
807 
808 #ifdef notdef
809 void
810 unp_abort(struct unpcb *unp)
811 {
812 
813 	unp_detach(unp);
814 }
815 #endif
816 
817 static int
818 prison_unpcb(struct thread *td, struct unpcb *unp)
819 {
820 	struct proc *p;
821 
822 	if (td == NULL)
823 		return (0);
824 	if ((p = td->td_proc) == NULL)
825 		return (0);
826 	if (!p->p_ucred->cr_prison)
827 		return (0);
828 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
829 		return (0);
830 	return (1);
831 }
832 
833 static int
834 unp_pcblist(SYSCTL_HANDLER_ARGS)
835 {
836 	int error, i, n;
837 	struct unpcb *unp, **unp_list;
838 	unp_gen_t gencnt;
839 	struct unp_head *head;
840 
841 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
842 
843 	KKASSERT(curproc != NULL);
844 
845 	/*
846 	 * The process of preparing the PCB list is too time-consuming and
847 	 * resource-intensive to repeat twice on every request.
848 	 */
849 	if (req->oldptr == NULL) {
850 		n = unp_count;
851 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
852 		return 0;
853 	}
854 
855 	if (req->newptr != NULL)
856 		return EPERM;
857 
858 	/*
859 	 * OK, now we're committed to doing something.
860 	 */
861 	gencnt = unp_gencnt;
862 	n = unp_count;
863 
864 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
865 
866 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
867 	     unp = LIST_NEXT(unp, unp_link)) {
868 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
869 			unp_list[i++] = unp;
870 	}
871 	n = i;			/* in case we lost some during malloc */
872 
873 	error = 0;
874 	for (i = 0; i < n; i++) {
875 		unp = unp_list[i];
876 		if (unp->unp_gencnt <= gencnt) {
877 			struct xunpcb xu;
878 			xu.xu_len = sizeof xu;
879 			xu.xu_unpp = unp;
880 			/*
881 			 * XXX - need more locking here to protect against
882 			 * connect/disconnect races for SMP.
883 			 */
884 			if (unp->unp_addr)
885 				bcopy(unp->unp_addr, &xu.xu_addr,
886 				      unp->unp_addr->sun_len);
887 			if (unp->unp_conn && unp->unp_conn->unp_addr)
888 				bcopy(unp->unp_conn->unp_addr,
889 				      &xu.xu_caddr,
890 				      unp->unp_conn->unp_addr->sun_len);
891 			bcopy(unp, &xu.xu_unp, sizeof *unp);
892 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
893 			error = SYSCTL_OUT(req, &xu, sizeof xu);
894 		}
895 	}
896 	kfree(unp_list, M_TEMP);
897 	return error;
898 }
899 
900 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
901 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
902 	    "List of active local datagram sockets");
903 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
904 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
905 	    "List of active local stream sockets");
906 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
907 	    (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
908 	    "List of active local seqpacket stream sockets");
909 
910 static void
911 unp_shutdown(struct unpcb *unp)
912 {
913 	struct socket *so;
914 
915 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
916 	     unp->unp_socket->so_type == SOCK_SEQPACKET) &&
917 	    unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
918 		socantrcvmore(so);
919 	}
920 }
921 
922 static void
923 unp_drop(struct unpcb *unp, int err)
924 {
925 	struct socket *so = unp->unp_socket;
926 
927 	so->so_error = err;
928 	unp_disconnect(unp);
929 }
930 
931 #ifdef notdef
932 void
933 unp_drain(void)
934 {
935 
936 }
937 #endif
938 
939 int
940 unp_externalize(struct mbuf *rights)
941 {
942 	struct proc *p = curproc;		/* XXX */
943 	int i;
944 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
945 	int *fdp;
946 	struct file **rp;
947 	struct file *fp;
948 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
949 		/ sizeof (struct file *);
950 	int f;
951 
952 	/*
953 	 * if the new FD's will not fit, then we free them all
954 	 */
955 	if (!fdavail(p, newfds)) {
956 		rp = (struct file **)CMSG_DATA(cm);
957 		for (i = 0; i < newfds; i++) {
958 			fp = *rp;
959 			/*
960 			 * zero the pointer before calling unp_discard,
961 			 * since it may end up in unp_gc()..
962 			 */
963 			*rp++ = 0;
964 			unp_discard(fp, NULL);
965 		}
966 		return (EMSGSIZE);
967 	}
968 
969 	/*
970 	 * now change each pointer to an fd in the global table to
971 	 * an integer that is the index to the local fd table entry
972 	 * that we set up to point to the global one we are transferring.
973 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
974 	 * then do it in forward order. In that case, an integer will
975 	 * always come in the same place or before its corresponding
976 	 * struct file pointer.
977 	 * If sizeof (struct file *) is smaller than sizeof int, then
978 	 * do it in reverse order.
979 	 */
980 	if (sizeof (struct file *) >= sizeof (int)) {
981 		fdp = (int *)CMSG_DATA(cm);
982 		rp = (struct file **)CMSG_DATA(cm);
983 		for (i = 0; i < newfds; i++) {
984 			if (fdalloc(p, 0, &f))
985 				panic("unp_externalize");
986 			fp = *rp++;
987 			unp_fp_externalize(p, fp, f);
988 			*fdp++ = f;
989 		}
990 	} else {
991 		fdp = (int *)CMSG_DATA(cm) + newfds - 1;
992 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
993 		for (i = 0; i < newfds; i++) {
994 			if (fdalloc(p, 0, &f))
995 				panic("unp_externalize");
996 			fp = *rp--;
997 			unp_fp_externalize(p, fp, f);
998 			*fdp-- = f;
999 		}
1000 	}
1001 
1002 	/*
1003 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1004 	 * differs.
1005 	 */
1006 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1007 	rights->m_len = cm->cmsg_len;
1008 	return (0);
1009 }
1010 
1011 static void
1012 unp_fp_externalize(struct proc *p, struct file *fp, int fd)
1013 {
1014 	struct file *fx;
1015 	int error;
1016 
1017 	if (p) {
1018 		KKASSERT(fd >= 0);
1019 		if (fp->f_flag & FREVOKED) {
1020 			kprintf("Warning: revoked fp exiting unix socket\n");
1021 			fx = NULL;
1022 			error = falloc(p, &fx, NULL);
1023 			if (error == 0)
1024 				fsetfd(p, fx, fd);
1025 			else
1026 				fsetfd(p, NULL, fd);
1027 			fdrop(fx);
1028 		} else {
1029 			fsetfd(p, fp, fd);
1030 		}
1031 	}
1032 	spin_lock_wr(&unp_spin);
1033 	fp->f_msgcount--;
1034 	unp_rights--;
1035 	spin_unlock_wr(&unp_spin);
1036 	fdrop(fp);
1037 }
1038 
1039 
1040 void
1041 unp_init(void)
1042 {
1043 	LIST_INIT(&unp_dhead);
1044 	LIST_INIT(&unp_shead);
1045 	spin_init(&unp_spin);
1046 }
1047 
1048 static int
1049 unp_internalize(struct mbuf *control, struct thread *td)
1050 {
1051 	struct proc *p = td->td_proc;
1052 	struct filedesc *fdescp;
1053 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1054 	struct file **rp;
1055 	struct file *fp;
1056 	int i, fd, *fdp;
1057 	struct cmsgcred *cmcred;
1058 	int oldfds;
1059 	u_int newlen;
1060 
1061 	KKASSERT(p);
1062 	fdescp = p->p_fd;
1063 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1064 	    cm->cmsg_level != SOL_SOCKET ||
1065 	    CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1066 		return (EINVAL);
1067 	}
1068 
1069 	/*
1070 	 * Fill in credential information.
1071 	 */
1072 	if (cm->cmsg_type == SCM_CREDS) {
1073 		cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1074 		cmcred->cmcred_pid = p->p_pid;
1075 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1076 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1077 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1078 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1079 							CMGROUP_MAX);
1080 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1081 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1082 		return(0);
1083 	}
1084 
1085 	/*
1086 	 * cmsghdr may not be aligned, do not allow calculation(s) to
1087 	 * go negative.
1088 	 */
1089 	if (cm->cmsg_len < CMSG_LEN(0))
1090 		return(EINVAL);
1091 
1092 	oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1093 
1094 	/*
1095 	 * check that all the FDs passed in refer to legal OPEN files
1096 	 * If not, reject the entire operation.
1097 	 */
1098 	fdp = (int *)CMSG_DATA(cm);
1099 	for (i = 0; i < oldfds; i++) {
1100 		fd = *fdp++;
1101 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1102 		    fdescp->fd_files[fd].fp == NULL)
1103 			return (EBADF);
1104 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE)
1105 			return (EOPNOTSUPP);
1106 	}
1107 	/*
1108 	 * Now replace the integer FDs with pointers to
1109 	 * the associated global file table entry..
1110 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1111 	 * enough, return E2BIG.
1112 	 */
1113 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1114 	if (newlen > MCLBYTES)
1115 		return (E2BIG);
1116 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1117 		if (control->m_flags & M_EXT)
1118 			return (E2BIG);
1119 		MCLGET(control, MB_WAIT);
1120 		if (!(control->m_flags & M_EXT))
1121 			return (ENOBUFS);
1122 
1123 		/* copy the data to the cluster */
1124 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1125 		cm = mtod(control, struct cmsghdr *);
1126 	}
1127 
1128 	/*
1129 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1130 	 * differs.
1131 	 */
1132 	cm->cmsg_len = newlen;
1133 	control->m_len = CMSG_ALIGN(newlen);
1134 
1135 	/*
1136 	 * Transform the file descriptors into struct file pointers.
1137 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1138 	 * then do it in reverse order so that the int won't get until
1139 	 * we're done.
1140 	 * If sizeof (struct file *) is smaller than sizeof int, then
1141 	 * do it in forward order.
1142 	 */
1143 	if (sizeof (struct file *) >= sizeof (int)) {
1144 		fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1145 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1146 		for (i = 0; i < oldfds; i++) {
1147 			fp = fdescp->fd_files[*fdp--].fp;
1148 			*rp-- = fp;
1149 			fhold(fp);
1150 			spin_lock_wr(&unp_spin);
1151 			fp->f_msgcount++;
1152 			unp_rights++;
1153 			spin_unlock_wr(&unp_spin);
1154 		}
1155 	} else {
1156 		fdp = (int *)CMSG_DATA(cm);
1157 		rp = (struct file **)CMSG_DATA(cm);
1158 		for (i = 0; i < oldfds; i++) {
1159 			fp = fdescp->fd_files[*fdp++].fp;
1160 			*rp++ = fp;
1161 			fhold(fp);
1162 			spin_lock_wr(&unp_spin);
1163 			fp->f_msgcount++;
1164 			unp_rights++;
1165 			spin_unlock_wr(&unp_spin);
1166 		}
1167 	}
1168 	return (0);
1169 }
1170 
1171 /*
1172  * Garbage collect in-transit file descriptors that get lost due to
1173  * loops (i.e. when a socket is sent to another process over itself,
1174  * and more complex situations).
1175  *
1176  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1177  */
1178 
1179 struct unp_gc_info {
1180 	struct file **extra_ref;
1181 	struct file *locked_fp;
1182 	int defer;
1183 	int index;
1184 	int maxindex;
1185 };
1186 
1187 static void
1188 unp_gc(void)
1189 {
1190 	struct unp_gc_info info;
1191 	static boolean_t unp_gcing;
1192 	struct file **fpp;
1193 	int i;
1194 
1195 	spin_lock_wr(&unp_spin);
1196 	if (unp_gcing) {
1197 		spin_unlock_wr(&unp_spin);
1198 		return;
1199 	}
1200 	unp_gcing = TRUE;
1201 	spin_unlock_wr(&unp_spin);
1202 
1203 	/*
1204 	 * before going through all this, set all FDs to
1205 	 * be NOT defered and NOT externally accessible
1206 	 */
1207 	info.defer = 0;
1208 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1209 	do {
1210 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1211 	} while (info.defer);
1212 
1213 	/*
1214 	 * We grab an extra reference to each of the file table entries
1215 	 * that are not otherwise accessible and then free the rights
1216 	 * that are stored in messages on them.
1217 	 *
1218 	 * The bug in the orginal code is a little tricky, so I'll describe
1219 	 * what's wrong with it here.
1220 	 *
1221 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1222 	 * times -- consider the case of sockets A and B that contain
1223 	 * references to each other.  On a last close of some other socket,
1224 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1225 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1226 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1227 	 * results in the following chain.  Closef calls soo_close, which
1228 	 * calls soclose.   Soclose calls first (through the switch
1229 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1230 	 * returns because the previous instance had set unp_gcing, and
1231 	 * we return all the way back to soclose, which marks the socket
1232 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1233 	 * to free up the rights that are queued in messages on the socket A,
1234 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1235 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1236 	 * instance of unp_discard just calls closef on B.
1237 	 *
1238 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1239 	 * which results in another closef on A.  Unfortunately, A is already
1240 	 * being closed, and the descriptor has already been marked with
1241 	 * SS_NOFDREF, and soclose panics at this point.
1242 	 *
1243 	 * Here, we first take an extra reference to each inaccessible
1244 	 * descriptor.  Then, we call sorflush ourself, since we know
1245 	 * it is a Unix domain socket anyhow.  After we destroy all the
1246 	 * rights carried in messages, we do a last closef to get rid
1247 	 * of our extra reference.  This is the last close, and the
1248 	 * unp_detach etc will shut down the socket.
1249 	 *
1250 	 * 91/09/19, bsy@cs.cmu.edu
1251 	 */
1252 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1253 	info.maxindex = 256;
1254 
1255 	do {
1256 		/*
1257 		 * Look for matches
1258 		 */
1259 		info.index = 0;
1260 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1261 
1262 		/*
1263 		 * For each FD on our hit list, do the following two things
1264 		 */
1265 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1266 			struct file *tfp = *fpp;
1267 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1268 				sorflush((struct socket *)(tfp->f_data));
1269 		}
1270 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1271 			closef(*fpp, NULL);
1272 	} while (info.index == info.maxindex);
1273 	kfree((caddr_t)info.extra_ref, M_FILE);
1274 	unp_gcing = FALSE;
1275 }
1276 
1277 /*
1278  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1279  */
1280 static int
1281 unp_gc_checkrefs(struct file *fp, void *data)
1282 {
1283 	struct unp_gc_info *info = data;
1284 
1285 	if (fp->f_count == 0)
1286 		return(0);
1287 	if (info->index == info->maxindex)
1288 		return(-1);
1289 
1290 	/*
1291 	 * If all refs are from msgs, and it's not marked accessible
1292 	 * then it must be referenced from some unreachable cycle
1293 	 * of (shut-down) FDs, so include it in our
1294 	 * list of FDs to remove
1295 	 */
1296 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1297 		info->extra_ref[info->index++] = fp;
1298 		fhold(fp);
1299 	}
1300 	return(0);
1301 }
1302 
1303 /*
1304  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1305  */
1306 static int
1307 unp_gc_clearmarks(struct file *fp, void *data __unused)
1308 {
1309 	atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1310 	return(0);
1311 }
1312 
1313 /*
1314  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1315  */
1316 static int
1317 unp_gc_checkmarks(struct file *fp, void *data)
1318 {
1319 	struct unp_gc_info *info = data;
1320 	struct socket *so;
1321 
1322 	/*
1323 	 * If the file is not open, skip it
1324 	 */
1325 	if (fp->f_count == 0)
1326 		return(0);
1327 	/*
1328 	 * If we already marked it as 'defer'  in a
1329 	 * previous pass, then try process it this time
1330 	 * and un-mark it
1331 	 */
1332 	if (fp->f_flag & FDEFER) {
1333 		atomic_clear_int(&fp->f_flag, FDEFER);
1334 		--info->defer;
1335 	} else {
1336 		/*
1337 		 * if it's not defered, then check if it's
1338 		 * already marked.. if so skip it
1339 		 */
1340 		if (fp->f_flag & FMARK)
1341 			return(0);
1342 		/*
1343 		 * If all references are from messages
1344 		 * in transit, then skip it. it's not
1345 		 * externally accessible.
1346 		 */
1347 		if (fp->f_count == fp->f_msgcount)
1348 			return(0);
1349 		/*
1350 		 * If it got this far then it must be
1351 		 * externally accessible.
1352 		 */
1353 		atomic_set_int(&fp->f_flag, FMARK);
1354 	}
1355 
1356 	/*
1357 	 * either it was defered, or it is externally
1358 	 * accessible and not already marked so.
1359 	 * Now check if it is possibly one of OUR sockets.
1360 	 */
1361 	if (fp->f_type != DTYPE_SOCKET ||
1362 	    (so = (struct socket *)fp->f_data) == NULL)
1363 		return(0);
1364 	if (so->so_proto->pr_domain != &localdomain ||
1365 	    !(so->so_proto->pr_flags & PR_RIGHTS))
1366 		return(0);
1367 #ifdef notdef
1368 	if (so->so_rcv.ssb_flags & SSB_LOCK) {
1369 		/*
1370 		 * This is problematical; it's not clear
1371 		 * we need to wait for the sockbuf to be
1372 		 * unlocked (on a uniprocessor, at least),
1373 		 * and it's also not clear what to do
1374 		 * if sbwait returns an error due to receipt
1375 		 * of a signal.  If sbwait does return
1376 		 * an error, we'll go into an infinite
1377 		 * loop.  Delete all of this for now.
1378 		 */
1379 		sbwait(&so->so_rcv);
1380 		goto restart;
1381 	}
1382 #endif
1383 	/*
1384 	 * So, Ok, it's one of our sockets and it IS externally
1385 	 * accessible (or was defered). Now we look
1386 	 * to see if we hold any file descriptors in its
1387 	 * message buffers. Follow those links and mark them
1388 	 * as accessible too.
1389 	 */
1390 	info->locked_fp = fp;
1391 /*	spin_lock_wr(&so->so_rcv.sb_spin); */
1392 	unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1393 /*	spin_unlock_wr(&so->so_rcv.sb_spin);*/
1394 	return (0);
1395 }
1396 
1397 /*
1398  * Scan all unix domain sockets and replace any revoked file pointers
1399  * found with the dummy file pointer fx.  We don't worry about races
1400  * against file pointers being read out as those are handled in the
1401  * externalize code.
1402  */
1403 
1404 #define REVOKE_GC_MAXFILES	32
1405 
1406 struct unp_revoke_gc_info {
1407 	struct file	*fx;
1408 	struct file	*fary[REVOKE_GC_MAXFILES];
1409 	int		fcount;
1410 };
1411 
1412 void
1413 unp_revoke_gc(struct file *fx)
1414 {
1415 	struct unp_revoke_gc_info info;
1416 	int i;
1417 
1418 	info.fx = fx;
1419 	do {
1420 		info.fcount = 0;
1421 		allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1422 		for (i = 0; i < info.fcount; ++i)
1423 			unp_fp_externalize(NULL, info.fary[i], -1);
1424 	} while (info.fcount == REVOKE_GC_MAXFILES);
1425 }
1426 
1427 /*
1428  * Check for and replace revoked descriptors.
1429  *
1430  * WARNING:  This routine is not allowed to block.
1431  */
1432 static int
1433 unp_revoke_gc_check(struct file *fps, void *vinfo)
1434 {
1435 	struct unp_revoke_gc_info *info = vinfo;
1436 	struct file *fp;
1437 	struct socket *so;
1438 	struct mbuf *m0;
1439 	struct mbuf *m;
1440 	struct file **rp;
1441 	struct cmsghdr *cm;
1442 	int i;
1443 	int qfds;
1444 
1445 	/*
1446 	 * Is this a unix domain socket with rights-passing abilities?
1447 	 */
1448 	if (fps->f_type != DTYPE_SOCKET)
1449 		return (0);
1450 	if ((so = (struct socket *)fps->f_data) == NULL)
1451 		return(0);
1452 	if (so->so_proto->pr_domain != &localdomain)
1453 		return(0);
1454 	if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1455 		return(0);
1456 
1457 	/*
1458 	 * Scan the mbufs for control messages and replace any revoked
1459 	 * descriptors we find.
1460 	 */
1461 	m0 = so->so_rcv.ssb_mb;
1462 	while (m0) {
1463 		for (m = m0; m; m = m->m_next) {
1464 			if (m->m_type != MT_CONTROL)
1465 				continue;
1466 			if (m->m_len < sizeof(*cm))
1467 				continue;
1468 			cm = mtod(m, struct cmsghdr *);
1469 			if (cm->cmsg_level != SOL_SOCKET ||
1470 			    cm->cmsg_type != SCM_RIGHTS) {
1471 				continue;
1472 			}
1473 			qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1474 			rp = (struct file **)CMSG_DATA(cm);
1475 			for (i = 0; i < qfds; i++) {
1476 				fp = rp[i];
1477 				if (fp->f_flag & FREVOKED) {
1478 					kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1479 					fhold(info->fx);
1480 					info->fx->f_msgcount++;
1481 					unp_rights++;
1482 					rp[i] = info->fx;
1483 					info->fary[info->fcount++] = fp;
1484 				}
1485 				if (info->fcount == REVOKE_GC_MAXFILES)
1486 					break;
1487 			}
1488 			if (info->fcount == REVOKE_GC_MAXFILES)
1489 				break;
1490 		}
1491 		m0 = m0->m_nextpkt;
1492 		if (info->fcount == REVOKE_GC_MAXFILES)
1493 			break;
1494 	}
1495 
1496 	/*
1497 	 * Stop the scan if we filled up our array.
1498 	 */
1499 	if (info->fcount == REVOKE_GC_MAXFILES)
1500 		return(-1);
1501 	return(0);
1502 }
1503 
1504 void
1505 unp_dispose(struct mbuf *m)
1506 {
1507 	if (m)
1508 		unp_scan(m, unp_discard, NULL);
1509 }
1510 
1511 static int
1512 unp_listen(struct unpcb *unp, struct thread *td)
1513 {
1514 	struct proc *p = td->td_proc;
1515 
1516 	KKASSERT(p);
1517 	cru2x(p->p_ucred, &unp->unp_peercred);
1518 	unp->unp_flags |= UNP_HAVEPCCACHED;
1519 	return (0);
1520 }
1521 
1522 static void
1523 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
1524 {
1525 	struct mbuf *m;
1526 	struct file **rp;
1527 	struct cmsghdr *cm;
1528 	int i;
1529 	int qfds;
1530 
1531 	while (m0) {
1532 		for (m = m0; m; m = m->m_next) {
1533 			if (m->m_type == MT_CONTROL &&
1534 			    m->m_len >= sizeof(*cm)) {
1535 				cm = mtod(m, struct cmsghdr *);
1536 				if (cm->cmsg_level != SOL_SOCKET ||
1537 				    cm->cmsg_type != SCM_RIGHTS)
1538 					continue;
1539 				qfds = (cm->cmsg_len - CMSG_LEN(0)) /
1540 					sizeof(void *);
1541 				rp = (struct file **)CMSG_DATA(cm);
1542 				for (i = 0; i < qfds; i++)
1543 					(*op)(*rp++, data);
1544 				break;		/* XXX, but saves time */
1545 			}
1546 		}
1547 		m0 = m0->m_nextpkt;
1548 	}
1549 }
1550 
1551 static void
1552 unp_mark(struct file *fp, void *data)
1553 {
1554 	struct unp_gc_info *info = data;
1555 
1556 	if ((fp->f_flag & FMARK) == 0) {
1557 		++info->defer;
1558 		atomic_set_int(&fp->f_flag, FMARK | FDEFER);
1559 	}
1560 }
1561 
1562 static void
1563 unp_discard(struct file *fp, void *data __unused)
1564 {
1565 	spin_lock_wr(&unp_spin);
1566 	fp->f_msgcount--;
1567 	unp_rights--;
1568 	spin_unlock_wr(&unp_spin);
1569 	closef(fp, NULL);
1570 }
1571 
1572