xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision 6d503aa4fca08913853b2b35d54d3aff30a6a78e)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
30  * $FreeBSD: src/sys/kern/uipc_usrreq.c,v 1.54.2.10 2003/03/04 17:28:09 nectar Exp $
31  */
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/domain.h>
37 #include <sys/fcntl.h>
38 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
39 #include <sys/proc.h>
40 #include <sys/file.h>
41 #include <sys/filedesc.h>
42 #include <sys/mbuf.h>
43 #include <sys/nlookup.h>
44 #include <sys/protosw.h>
45 #include <sys/socket.h>
46 #include <sys/socketvar.h>
47 #include <sys/resourcevar.h>
48 #include <sys/stat.h>
49 #include <sys/mount.h>
50 #include <sys/sysctl.h>
51 #include <sys/un.h>
52 #include <sys/unpcb.h>
53 #include <sys/vnode.h>
54 
55 #include <sys/file2.h>
56 #include <sys/spinlock2.h>
57 #include <sys/socketvar2.h>
58 #include <sys/msgport2.h>
59 
60 #define UNP_DETACHED		UNP_PRIVATE1
61 #define UNP_CONNECTING		UNP_PRIVATE2
62 
63 #define UNP_ISATTACHED(unp)	\
64     ((unp) != NULL && ((unp)->unp_flags & UNP_DETACHED) == 0)
65 
66 #ifdef INVARIANTS
67 #define UNP_ASSERT_TOKEN_HELD(unp) \
68     ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup((unp)))
69 #else	/* !INVARIANTS */
70 #define UNP_ASSERT_TOKEN_HELD(unp)
71 #endif	/* INVARIANTS */
72 
73 typedef struct unp_defdiscard {
74 	struct unp_defdiscard *next;
75 	struct file *fp;
76 } *unp_defdiscard_t;
77 
78 static	MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
79 static	unp_gen_t unp_gencnt;
80 static	u_int unp_count;
81 
82 static	struct unp_head unp_shead, unp_dhead;
83 
84 static struct lwkt_token unp_token = LWKT_TOKEN_INITIALIZER(unp_token);
85 static int unp_defdiscard_nest;
86 static unp_defdiscard_t unp_defdiscard_base;
87 
88 /*
89  * Unix communications domain.
90  *
91  * TODO:
92  *	RDM
93  *	rethink name space problems
94  *	need a proper out-of-band
95  *	lock pushdown
96  */
97 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
98 static ino_t	unp_ino = 1;		/* prototype for fake inode numbers */
99 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin, "unp_ino_spin");
100 
101 static int     unp_attach (struct socket *, struct pru_attach_info *);
102 static void    unp_detach (struct unpcb *);
103 static int     unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
104 static int     unp_connect (struct socket *,struct sockaddr *,
105 				struct thread *);
106 static void    unp_disconnect (struct unpcb *);
107 static void    unp_shutdown (struct unpcb *);
108 static void    unp_drop (struct unpcb *, int);
109 static void    unp_gc (void);
110 static int     unp_gc_clearmarks(struct file *, void *);
111 static int     unp_gc_checkmarks(struct file *, void *);
112 static int     unp_gc_checkrefs(struct file *, void *);
113 static int     unp_revoke_gc_check(struct file *, void *);
114 static void    unp_scan (struct mbuf *, void (*)(struct file *, void *),
115 				void *data);
116 static void    unp_mark (struct file *, void *data);
117 static void    unp_discard (struct file *, void *);
118 static int     unp_internalize (struct mbuf *, struct thread *);
119 static int     unp_listen (struct unpcb *, struct thread *);
120 static void    unp_fp_externalize(struct lwp *lp, struct file *fp, int fd);
121 static int     unp_find_lockref(struct sockaddr *nam, struct thread *td,
122 		   short type, struct unpcb **unp_ret);
123 static int     unp_connect_pair(struct unpcb *unp, struct unpcb *unp2);
124 
125 /*
126  * SMP Considerations:
127  *
128  *	Since unp_token will be automaticly released upon execution of
129  *	blocking code, we need to reference unp_conn before any possible
130  *	blocking code to prevent it from being ripped behind our back.
131  *
132  *	Any adjustment to unp->unp_conn requires both the global unp_token
133  *	AND the per-unp token (lwkt_token_pool_lookup(unp)) to be held.
134  *
135  *	Any access to so_pcb to obtain unp requires the pool token for
136  *	unp to be held.
137  */
138 
139 /* NOTE: unp_token MUST be held */
140 static __inline void
141 unp_reference(struct unpcb *unp)
142 {
143 	atomic_add_int(&unp->unp_refcnt, 1);
144 }
145 
146 /* NOTE: unp_token MUST be held */
147 static __inline void
148 unp_free(struct unpcb *unp)
149 {
150 	KKASSERT(unp->unp_refcnt > 0);
151 	if (atomic_fetchadd_int(&unp->unp_refcnt, -1) == 1)
152 		unp_detach(unp);
153 }
154 
155 static __inline struct unpcb *
156 unp_getsocktoken(struct socket *so)
157 {
158 	struct unpcb *unp;
159 
160 	/*
161 	 * The unp pointer is invalid until we verify that it is
162 	 * good by re-checking so_pcb AFTER obtaining the token.
163 	 */
164 	while ((unp = so->so_pcb) != NULL) {
165 		lwkt_getpooltoken(unp);
166 		if (unp == so->so_pcb)
167 			break;
168 		lwkt_relpooltoken(unp);
169 	}
170 	return unp;
171 }
172 
173 static __inline void
174 unp_reltoken(struct unpcb *unp)
175 {
176 	if (unp != NULL)
177 		lwkt_relpooltoken(unp);
178 }
179 
180 static __inline void
181 unp_setflags(struct unpcb *unp, int flags)
182 {
183 	atomic_set_int(&unp->unp_flags, flags);
184 }
185 
186 static __inline void
187 unp_clrflags(struct unpcb *unp, int flags)
188 {
189 	atomic_clear_int(&unp->unp_flags, flags);
190 }
191 
192 /*
193  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
194  *	 will sofree() it when we return.
195  */
196 static void
197 uipc_abort(netmsg_t msg)
198 {
199 	struct unpcb *unp;
200 	int error;
201 
202 	lwkt_gettoken(&unp_token);
203 	unp = unp_getsocktoken(msg->base.nm_so);
204 
205 	if (UNP_ISATTACHED(unp)) {
206 		unp_setflags(unp, UNP_DETACHED);
207 		unp_drop(unp, ECONNABORTED);
208 		unp_free(unp);
209 		error = 0;
210 	} else {
211 		error = EINVAL;
212 	}
213 
214 	unp_reltoken(unp);
215 	lwkt_reltoken(&unp_token);
216 
217 	lwkt_replymsg(&msg->lmsg, error);
218 }
219 
220 static void
221 uipc_accept(netmsg_t msg)
222 {
223 	struct unpcb *unp;
224 	int error;
225 
226 	lwkt_gettoken(&unp_token);
227 	unp = msg->base.nm_so->so_pcb;
228 	if (!UNP_ISATTACHED(unp)) {
229 		error = EINVAL;
230 	} else {
231 		struct unpcb *unp2 = unp->unp_conn;
232 
233 		/*
234 		 * Pass back name of connected socket,
235 		 * if it was bound and we are still connected
236 		 * (our peer may have closed already!).
237 		 */
238 		if (unp2 && unp2->unp_addr) {
239 			unp_reference(unp2);
240 			*msg->accept.nm_nam = dup_sockaddr(
241 				(struct sockaddr *)unp2->unp_addr);
242 			unp_free(unp2);
243 		} else {
244 			*msg->accept.nm_nam = dup_sockaddr(&sun_noname);
245 		}
246 		error = 0;
247 	}
248 	lwkt_reltoken(&unp_token);
249 	lwkt_replymsg(&msg->lmsg, error);
250 }
251 
252 static void
253 uipc_attach(netmsg_t msg)
254 {
255 	struct unpcb *unp;
256 	int error;
257 
258 	lwkt_gettoken(&unp_token);
259 	unp = msg->base.nm_so->so_pcb;
260 	KASSERT(unp == NULL, ("double unp attach"));
261 	error = unp_attach(msg->base.nm_so, msg->attach.nm_ai);
262 	lwkt_reltoken(&unp_token);
263 	lwkt_replymsg(&msg->lmsg, error);
264 }
265 
266 static void
267 uipc_bind(netmsg_t msg)
268 {
269 	struct unpcb *unp;
270 	int error;
271 
272 	lwkt_gettoken(&unp_token);
273 	unp = msg->base.nm_so->so_pcb;
274 	if (UNP_ISATTACHED(unp))
275 		error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td);
276 	else
277 		error = EINVAL;
278 	lwkt_reltoken(&unp_token);
279 	lwkt_replymsg(&msg->lmsg, error);
280 }
281 
282 static void
283 uipc_connect(netmsg_t msg)
284 {
285 	int error;
286 
287 	error = unp_connect(msg->base.nm_so, msg->connect.nm_nam,
288 	    msg->connect.nm_td);
289 	lwkt_replymsg(&msg->lmsg, error);
290 }
291 
292 static void
293 uipc_connect2(netmsg_t msg)
294 {
295 	int error;
296 
297 	error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2);
298 	lwkt_replymsg(&msg->lmsg, error);
299 }
300 
301 /* control is EOPNOTSUPP */
302 
303 static void
304 uipc_detach(netmsg_t msg)
305 {
306 	struct unpcb *unp;
307 	int error;
308 
309 	lwkt_gettoken(&unp_token);
310 	unp = unp_getsocktoken(msg->base.nm_so);
311 
312 	if (UNP_ISATTACHED(unp)) {
313 		unp_setflags(unp, UNP_DETACHED);
314 		unp_free(unp);
315 		error = 0;
316 	} else {
317 		error = EINVAL;
318 	}
319 
320 	unp_reltoken(unp);
321 	lwkt_reltoken(&unp_token);
322 
323 	lwkt_replymsg(&msg->lmsg, error);
324 }
325 
326 static void
327 uipc_disconnect(netmsg_t msg)
328 {
329 	struct unpcb *unp;
330 	int error;
331 
332 	lwkt_gettoken(&unp_token);
333 	unp = msg->base.nm_so->so_pcb;
334 	if (UNP_ISATTACHED(unp)) {
335 		unp_disconnect(unp);
336 		error = 0;
337 	} else {
338 		error = EINVAL;
339 	}
340 	lwkt_reltoken(&unp_token);
341 	lwkt_replymsg(&msg->lmsg, error);
342 }
343 
344 static void
345 uipc_listen(netmsg_t msg)
346 {
347 	struct unpcb *unp;
348 	int error;
349 
350 	lwkt_gettoken(&unp_token);
351 	unp = msg->base.nm_so->so_pcb;
352 	if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL)
353 		error = EINVAL;
354 	else
355 		error = unp_listen(unp, msg->listen.nm_td);
356 	lwkt_reltoken(&unp_token);
357 	lwkt_replymsg(&msg->lmsg, error);
358 }
359 
360 static void
361 uipc_peeraddr(netmsg_t msg)
362 {
363 	struct unpcb *unp;
364 	int error;
365 
366 	lwkt_gettoken(&unp_token);
367 	unp = msg->base.nm_so->so_pcb;
368 	if (!UNP_ISATTACHED(unp)) {
369 		error = EINVAL;
370 	} else if (unp->unp_conn && unp->unp_conn->unp_addr) {
371 		struct unpcb *unp2 = unp->unp_conn;
372 
373 		unp_reference(unp2);
374 		*msg->peeraddr.nm_nam = dup_sockaddr(
375 				(struct sockaddr *)unp2->unp_addr);
376 		unp_free(unp2);
377 		error = 0;
378 	} else {
379 		/*
380 		 * XXX: It seems that this test always fails even when
381 		 * connection is established.  So, this else clause is
382 		 * added as workaround to return PF_LOCAL sockaddr.
383 		 */
384 		*msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname);
385 		error = 0;
386 	}
387 	lwkt_reltoken(&unp_token);
388 	lwkt_replymsg(&msg->lmsg, error);
389 }
390 
391 static void
392 uipc_rcvd(netmsg_t msg)
393 {
394 	struct unpcb *unp, *unp2;
395 	struct socket *so;
396 	struct socket *so2;
397 	int error;
398 
399 	/*
400 	 * so_pcb is only modified with both the global and the unp
401 	 * pool token held.
402 	 */
403 	so = msg->base.nm_so;
404 	unp = unp_getsocktoken(so);
405 
406 	if (!UNP_ISATTACHED(unp)) {
407 		error = EINVAL;
408 		goto done;
409 	}
410 
411 	switch (so->so_type) {
412 	case SOCK_DGRAM:
413 		panic("uipc_rcvd DGRAM?");
414 		/*NOTREACHED*/
415 	case SOCK_STREAM:
416 	case SOCK_SEQPACKET:
417 		if (unp->unp_conn == NULL)
418 			break;
419 		unp2 = unp->unp_conn;	/* protected by pool token */
420 
421 		/*
422 		 * Because we are transfering mbufs directly to the
423 		 * peer socket we have to use SSB_STOP on the sender
424 		 * to prevent it from building up infinite mbufs.
425 		 *
426 		 * As in several places in this module w ehave to ref unp2
427 		 * to ensure that it does not get ripped out from under us
428 		 * if we block on the so2 token or in sowwakeup().
429 		 */
430 		so2 = unp2->unp_socket;
431 		unp_reference(unp2);
432 		lwkt_gettoken(&so2->so_rcv.ssb_token);
433 		if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
434 		    so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
435 		) {
436 			atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP);
437 
438 			sowwakeup(so2);
439 		}
440 		lwkt_reltoken(&so2->so_rcv.ssb_token);
441 		unp_free(unp2);
442 		break;
443 	default:
444 		panic("uipc_rcvd unknown socktype");
445 		/*NOTREACHED*/
446 	}
447 	error = 0;
448 done:
449 	unp_reltoken(unp);
450 	lwkt_replymsg(&msg->lmsg, error);
451 }
452 
453 /* pru_rcvoob is EOPNOTSUPP */
454 
455 static void
456 uipc_send(netmsg_t msg)
457 {
458 	struct unpcb *unp, *unp2;
459 	struct socket *so;
460 	struct socket *so2;
461 	struct mbuf *control;
462 	struct mbuf *m;
463 	int error = 0;
464 
465 	so = msg->base.nm_so;
466 	control = msg->send.nm_control;
467 	m = msg->send.nm_m;
468 
469 	/*
470 	 * so_pcb is only modified with both the global and the unp
471 	 * pool token held.
472 	 */
473 	so = msg->base.nm_so;
474 	unp = unp_getsocktoken(so);
475 
476 	if (!UNP_ISATTACHED(unp)) {
477 		error = EINVAL;
478 		goto release;
479 	}
480 
481 	if (msg->send.nm_flags & PRUS_OOB) {
482 		error = EOPNOTSUPP;
483 		goto release;
484 	}
485 
486 	wakeup_start_delayed();
487 
488 	if (control && (error = unp_internalize(control, msg->send.nm_td)))
489 		goto release;
490 
491 	switch (so->so_type) {
492 	case SOCK_DGRAM:
493 	{
494 		struct sockaddr *from;
495 
496 		if (msg->send.nm_addr) {
497 			if (unp->unp_conn) {
498 				error = EISCONN;
499 				break;
500 			}
501 			error = unp_find_lockref(msg->send.nm_addr,
502 			    msg->send.nm_td, so->so_type, &unp2);
503 			if (error)
504 				break;
505 			/*
506 			 * NOTE:
507 			 * unp2 is locked and referenced.
508 			 *
509 			 * We could unlock unp2 now, since it was checked
510 			 * and referenced.
511 			 */
512 			unp_reltoken(unp2);
513 		} else {
514 			if (unp->unp_conn == NULL) {
515 				error = ENOTCONN;
516 				break;
517 			}
518 			/* XXX racy. */
519 			unp2 = unp->unp_conn;
520 			unp_reference(unp2);
521 		}
522 		/* NOTE: unp2 is referenced. */
523 		so2 = unp2->unp_socket;
524 
525 		if (unp->unp_addr)
526 			from = (struct sockaddr *)unp->unp_addr;
527 		else
528 			from = &sun_noname;
529 
530 		lwkt_gettoken(&so2->so_rcv.ssb_token);
531 		if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
532 			sorwakeup(so2);
533 			m = NULL;
534 			control = NULL;
535 		} else {
536 			error = ENOBUFS;
537 		}
538 		lwkt_reltoken(&so2->so_rcv.ssb_token);
539 
540 		unp_free(unp2);
541 		break;
542 	}
543 
544 	case SOCK_STREAM:
545 	case SOCK_SEQPACKET:
546 		/* Connect if not connected yet. */
547 		/*
548 		 * Note: A better implementation would complain
549 		 * if not equal to the peer's address.
550 		 */
551 		if (!(so->so_state & SS_ISCONNECTED)) {
552 			if (msg->send.nm_addr) {
553 				error = unp_connect(so,
554 						    msg->send.nm_addr,
555 						    msg->send.nm_td);
556 				if (error)
557 					break;	/* XXX */
558 			} else {
559 				error = ENOTCONN;
560 				break;
561 			}
562 		}
563 
564 		if (so->so_state & SS_CANTSENDMORE) {
565 			error = EPIPE;
566 			break;
567 		}
568 		if (unp->unp_conn == NULL)
569 			panic("uipc_send connected but no connection?");
570 		unp2 = unp->unp_conn;
571 		so2 = unp2->unp_socket;
572 
573 		unp_reference(unp2);
574 
575 		/*
576 		 * Send to paired receive port, and then reduce
577 		 * send buffer hiwater marks to maintain backpressure.
578 		 * Wake up readers.
579 		 */
580 		lwkt_gettoken(&so2->so_rcv.ssb_token);
581 		if (control) {
582 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
583 				control = NULL;
584 				m = NULL;
585 			}
586 		} else if (so->so_type == SOCK_SEQPACKET) {
587 			sbappendrecord(&so2->so_rcv.sb, m);
588 			m = NULL;
589 		} else {
590 			sbappend(&so2->so_rcv.sb, m);
591 			m = NULL;
592 		}
593 
594 		/*
595 		 * Because we are transfering mbufs directly to the
596 		 * peer socket we have to use SSB_STOP on the sender
597 		 * to prevent it from building up infinite mbufs.
598 		 */
599 		if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
600 		    so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
601 		) {
602 			atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP);
603 		}
604 		lwkt_reltoken(&so2->so_rcv.ssb_token);
605 		sorwakeup(so2);
606 
607 		unp_free(unp2);
608 		break;
609 
610 	default:
611 		panic("uipc_send unknown socktype");
612 	}
613 
614 	/*
615 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
616 	 */
617 	if (msg->send.nm_flags & PRUS_EOF) {
618 		socantsendmore(so);
619 		unp_shutdown(unp);
620 	}
621 
622 	if (control && error != 0)
623 		unp_dispose(control);
624 release:
625 	unp_reltoken(unp);
626 	wakeup_end_delayed();
627 
628 	if (control)
629 		m_freem(control);
630 	if (m)
631 		m_freem(m);
632 	lwkt_replymsg(&msg->lmsg, error);
633 }
634 
635 /*
636  * MPSAFE
637  */
638 static void
639 uipc_sense(netmsg_t msg)
640 {
641 	struct unpcb *unp;
642 	struct socket *so;
643 	struct stat *sb;
644 	int error;
645 
646 	so = msg->base.nm_so;
647 	sb = msg->sense.nm_stat;
648 
649 	/*
650 	 * so_pcb is only modified with both the global and the unp
651 	 * pool token held.
652 	 */
653 	unp = unp_getsocktoken(so);
654 
655 	if (!UNP_ISATTACHED(unp)) {
656 		error = EINVAL;
657 		goto done;
658 	}
659 
660 	sb->st_blksize = so->so_snd.ssb_hiwat;
661 	sb->st_dev = NOUDEV;
662 	if (unp->unp_ino == 0) {	/* make up a non-zero inode number */
663 		spin_lock(&unp_ino_spin);
664 		unp->unp_ino = unp_ino++;
665 		spin_unlock(&unp_ino_spin);
666 	}
667 	sb->st_ino = unp->unp_ino;
668 	error = 0;
669 done:
670 	unp_reltoken(unp);
671 	lwkt_replymsg(&msg->lmsg, error);
672 }
673 
674 static void
675 uipc_shutdown(netmsg_t msg)
676 {
677 	struct socket *so;
678 	struct unpcb *unp;
679 	int error;
680 
681 	/*
682 	 * so_pcb is only modified with both the global and the unp
683 	 * pool token held.
684 	 */
685 	so = msg->base.nm_so;
686 	unp = unp_getsocktoken(so);
687 
688 	if (UNP_ISATTACHED(unp)) {
689 		socantsendmore(so);
690 		unp_shutdown(unp);
691 		error = 0;
692 	} else {
693 		error = EINVAL;
694 	}
695 
696 	unp_reltoken(unp);
697 	lwkt_replymsg(&msg->lmsg, error);
698 }
699 
700 static void
701 uipc_sockaddr(netmsg_t msg)
702 {
703 	struct socket *so;
704 	struct unpcb *unp;
705 	int error;
706 
707 	/*
708 	 * so_pcb is only modified with both the global and the unp
709 	 * pool token held.
710 	 */
711 	so = msg->base.nm_so;
712 	unp = unp_getsocktoken(so);
713 
714 	if (UNP_ISATTACHED(unp)) {
715 		if (unp->unp_addr) {
716 			*msg->sockaddr.nm_nam =
717 				dup_sockaddr((struct sockaddr *)unp->unp_addr);
718 		}
719 		error = 0;
720 	} else {
721 		error = EINVAL;
722 	}
723 
724 	unp_reltoken(unp);
725 	lwkt_replymsg(&msg->lmsg, error);
726 }
727 
728 struct pr_usrreqs uipc_usrreqs = {
729 	.pru_abort = uipc_abort,
730 	.pru_accept = uipc_accept,
731 	.pru_attach = uipc_attach,
732 	.pru_bind = uipc_bind,
733 	.pru_connect = uipc_connect,
734 	.pru_connect2 = uipc_connect2,
735 	.pru_control = pr_generic_notsupp,
736 	.pru_detach = uipc_detach,
737 	.pru_disconnect = uipc_disconnect,
738 	.pru_listen = uipc_listen,
739 	.pru_peeraddr = uipc_peeraddr,
740 	.pru_rcvd = uipc_rcvd,
741 	.pru_rcvoob = pr_generic_notsupp,
742 	.pru_send = uipc_send,
743 	.pru_sense = uipc_sense,
744 	.pru_shutdown = uipc_shutdown,
745 	.pru_sockaddr = uipc_sockaddr,
746 	.pru_sosend = sosend,
747 	.pru_soreceive = soreceive
748 };
749 
750 void
751 uipc_ctloutput(netmsg_t msg)
752 {
753 	struct socket *so;
754 	struct sockopt *sopt;
755 	struct unpcb *unp;
756 	int error = 0;
757 
758 	lwkt_gettoken(&unp_token);
759 	so = msg->base.nm_so;
760 	sopt = msg->ctloutput.nm_sopt;
761 	unp = so->so_pcb;
762 
763 	switch (sopt->sopt_dir) {
764 	case SOPT_GET:
765 		switch (sopt->sopt_name) {
766 		case LOCAL_PEERCRED:
767 			if (unp->unp_flags & UNP_HAVEPC)
768 				soopt_from_kbuf(sopt, &unp->unp_peercred,
769 						sizeof(unp->unp_peercred));
770 			else {
771 				if (so->so_type == SOCK_STREAM)
772 					error = ENOTCONN;
773 				else if (so->so_type == SOCK_SEQPACKET)
774 					error = ENOTCONN;
775 				else
776 					error = EINVAL;
777 			}
778 			break;
779 		default:
780 			error = EOPNOTSUPP;
781 			break;
782 		}
783 		break;
784 	case SOPT_SET:
785 	default:
786 		error = EOPNOTSUPP;
787 		break;
788 	}
789 	lwkt_reltoken(&unp_token);
790 	lwkt_replymsg(&msg->lmsg, error);
791 }
792 
793 /*
794  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
795  * for stream sockets, although the total for sender and receiver is
796  * actually only PIPSIZ.
797  *
798  * Datagram sockets really use the sendspace as the maximum datagram size,
799  * and don't really want to reserve the sendspace.  Their recvspace should
800  * be large enough for at least one max-size datagram plus address.
801  *
802  * We want the local send/recv space to be significant larger then lo0's
803  * mtu of 16384.
804  */
805 #ifndef PIPSIZ
806 #define	PIPSIZ	57344
807 #endif
808 static u_long	unpst_sendspace = PIPSIZ;
809 static u_long	unpst_recvspace = PIPSIZ;
810 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
811 static u_long	unpdg_recvspace = 4*1024;
812 
813 static int	unp_rights;			/* file descriptors in flight */
814 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin, "unp_spin");
815 
816 SYSCTL_DECL(_net_local_seqpacket);
817 SYSCTL_DECL(_net_local_stream);
818 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
819     &unpst_sendspace, 0, "Size of stream socket send buffer");
820 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
821     &unpst_recvspace, 0, "Size of stream socket receive buffer");
822 
823 SYSCTL_DECL(_net_local_dgram);
824 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
825     &unpdg_sendspace, 0, "Max datagram socket size");
826 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
827     &unpdg_recvspace, 0, "Size of datagram socket receive buffer");
828 
829 SYSCTL_DECL(_net_local);
830 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
831    "File descriptors in flight");
832 
833 static int
834 unp_attach(struct socket *so, struct pru_attach_info *ai)
835 {
836 	struct unpcb *unp;
837 	int error;
838 
839 	lwkt_gettoken(&unp_token);
840 
841 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
842 		switch (so->so_type) {
843 		case SOCK_STREAM:
844 		case SOCK_SEQPACKET:
845 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
846 					  ai->sb_rlimit);
847 			break;
848 
849 		case SOCK_DGRAM:
850 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
851 					  ai->sb_rlimit);
852 			break;
853 
854 		default:
855 			panic("unp_attach");
856 		}
857 		if (error)
858 			goto failed;
859 	}
860 
861 	/*
862 	 * In order to support sendfile we have to set either SSB_STOPSUPP
863 	 * or SSB_PREALLOC.  Unix domain sockets use the SSB_STOP flow
864 	 * control mechanism.
865 	 */
866 	if (so->so_type == SOCK_STREAM) {
867 		atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP);
868 		atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP);
869 	}
870 
871 	unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK);
872 	if (unp == NULL) {
873 		error = ENOBUFS;
874 		goto failed;
875 	}
876 	unp->unp_refcnt = 1;
877 	unp->unp_gencnt = ++unp_gencnt;
878 	unp_count++;
879 	LIST_INIT(&unp->unp_refs);
880 	unp->unp_socket = so;
881 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
882 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
883 			 : &unp_shead, unp, unp_link);
884 	so->so_pcb = (caddr_t)unp;
885 	soreference(so);
886 	error = 0;
887 failed:
888 	lwkt_reltoken(&unp_token);
889 	return error;
890 }
891 
892 static void
893 unp_detach(struct unpcb *unp)
894 {
895 	struct socket *so;
896 
897 	lwkt_gettoken(&unp_token);
898 	lwkt_getpooltoken(unp);
899 
900 	LIST_REMOVE(unp, unp_link);	/* both tokens required */
901 	unp->unp_gencnt = ++unp_gencnt;
902 	--unp_count;
903 	if (unp->unp_vnode) {
904 		unp->unp_vnode->v_socket = NULL;
905 		vrele(unp->unp_vnode);
906 		unp->unp_vnode = NULL;
907 	}
908 	if (unp->unp_conn)
909 		unp_disconnect(unp);
910 	while (!LIST_EMPTY(&unp->unp_refs))
911 		unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET);
912 	soisdisconnected(unp->unp_socket);
913 	so = unp->unp_socket;
914 	soreference(so);		/* for delayed sorflush */
915 	KKASSERT(so->so_pcb == unp);
916 	so->so_pcb = NULL;		/* both tokens required */
917 	unp->unp_socket = NULL;
918 	sofree(so);		/* remove pcb ref */
919 
920 	if (unp_rights) {
921 		/*
922 		 * Normally the receive buffer is flushed later,
923 		 * in sofree, but if our receive buffer holds references
924 		 * to descriptors that are now garbage, we will dispose
925 		 * of those descriptor references after the garbage collector
926 		 * gets them (resulting in a "panic: closef: count < 0").
927 		 */
928 		sorflush(so);
929 		unp_gc();
930 	}
931 	sofree(so);
932 	lwkt_relpooltoken(unp);
933 	lwkt_reltoken(&unp_token);
934 
935 	if (unp->unp_addr)
936 		kfree(unp->unp_addr, M_SONAME);
937 	kfree(unp, M_UNPCB);
938 }
939 
940 static int
941 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
942 {
943 	struct proc *p = td->td_proc;
944 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
945 	struct vnode *vp;
946 	struct vattr vattr;
947 	int error, namelen;
948 	struct nlookupdata nd;
949 	char buf[SOCK_MAXADDRLEN];
950 
951 	lwkt_gettoken(&unp_token);
952 	if (unp->unp_vnode != NULL) {
953 		error = EINVAL;
954 		goto failed;
955 	}
956 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
957 	if (namelen <= 0) {
958 		error = EINVAL;
959 		goto failed;
960 	}
961 	strncpy(buf, soun->sun_path, namelen);
962 	buf[namelen] = 0;	/* null-terminate the string */
963 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
964 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
965 	if (error == 0)
966 		error = nlookup(&nd);
967 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
968 		error = EADDRINUSE;
969 	if (error)
970 		goto done;
971 
972 	VATTR_NULL(&vattr);
973 	vattr.va_type = VSOCK;
974 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
975 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
976 	if (error == 0) {
977 		if (unp->unp_vnode == NULL) {
978 			vp->v_socket = unp->unp_socket;
979 			unp->unp_vnode = vp;
980 			unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
981 			vn_unlock(vp);
982 		} else {
983 			vput(vp);		/* late race */
984 			error = EINVAL;
985 		}
986 	}
987 done:
988 	nlookup_done(&nd);
989 failed:
990 	lwkt_reltoken(&unp_token);
991 	return (error);
992 }
993 
994 static int
995 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
996 {
997 	struct unpcb *unp, *unp2;
998 	int error, flags = 0;
999 
1000 	lwkt_gettoken(&unp_token);
1001 
1002 	unp = unp_getsocktoken(so);
1003 	if (!UNP_ISATTACHED(unp)) {
1004 		error = EINVAL;
1005 		goto failed;
1006 	}
1007 
1008 	if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) {
1009 		error = EISCONN;
1010 		goto failed;
1011 	}
1012 
1013 	flags = UNP_CONNECTING;
1014 	unp_setflags(unp, flags);
1015 
1016 	error = unp_find_lockref(nam, td, so->so_type, &unp2);
1017 	if (error)
1018 		goto failed;
1019 	/*
1020 	 * NOTE:
1021 	 * unp2 is locked and referenced.
1022 	 */
1023 
1024 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1025 		struct socket *so2, *so3;
1026 		struct unpcb *unp3;
1027 
1028 		so2 = unp2->unp_socket;
1029 		if (!(so2->so_options & SO_ACCEPTCONN) ||
1030 		    (so3 = sonewconn_faddr(so2, 0, NULL,
1031 		     TRUE /* keep ref */)) == NULL) {
1032 			error = ECONNREFUSED;
1033 			goto done;
1034 		}
1035 		/* so3 has a socket reference. */
1036 
1037 		unp3 = unp_getsocktoken(so3);
1038 		if (!UNP_ISATTACHED(unp3)) {
1039 			unp_reltoken(unp3);
1040 			/*
1041 			 * Already aborted; we only need to drop the
1042 			 * socket reference held by sonewconn_faddr().
1043 			 */
1044 			sofree(so3);
1045 			error = ECONNREFUSED;
1046 			goto done;
1047 		}
1048 		unp_reference(unp3);
1049 		/*
1050 		 * NOTE:
1051 		 * unp3 is locked and referenced.
1052 		 */
1053 
1054 		/*
1055 		 * Release so3 socket reference held by sonewconn_faddr().
1056 		 * Since we have referenced unp3, neither unp3 nor so3 will
1057 		 * be destroyed here.
1058 		 */
1059 		sofree(so3);
1060 
1061 		if (unp2->unp_addr != NULL) {
1062 			unp3->unp_addr = (struct sockaddr_un *)
1063 			    dup_sockaddr((struct sockaddr *)unp2->unp_addr);
1064 		}
1065 
1066 		/*
1067 		 * unp_peercred management:
1068 		 *
1069 		 * The connecter's (client's) credentials are copied
1070 		 * from its process structure at the time of connect()
1071 		 * (which is now).
1072 		 */
1073 		cru2x(td->td_proc->p_ucred, &unp3->unp_peercred);
1074 		unp_setflags(unp3, UNP_HAVEPC);
1075 		/*
1076 		 * The receiver's (server's) credentials are copied
1077 		 * from the unp_peercred member of socket on which the
1078 		 * former called listen(); unp_listen() cached that
1079 		 * process's credentials at that time so we can use
1080 		 * them now.
1081 		 */
1082 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1083 		    ("unp_connect: listener without cached peercred"));
1084 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1085 		    sizeof(unp->unp_peercred));
1086 		unp_setflags(unp, UNP_HAVEPC);
1087 
1088 		error = unp_connect_pair(unp, unp3);
1089 		if (error) {
1090 			/* XXX we need a better name */
1091 			soabort_oncpu(so3);
1092 		}
1093 
1094 		/* Done with unp3 */
1095 		unp_free(unp3);
1096 		unp_reltoken(unp3);
1097 	} else {
1098 		error = unp_connect_pair(unp, unp2);
1099 	}
1100 done:
1101 	unp_free(unp2);
1102 	unp_reltoken(unp2);
1103 failed:
1104 	if (flags)
1105 		unp_clrflags(unp, flags);
1106 	unp_reltoken(unp);
1107 
1108 	lwkt_reltoken(&unp_token);
1109 	return (error);
1110 }
1111 
1112 /*
1113  * Connect two unix domain sockets together.
1114  *
1115  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1116  *	 pool token also be held.
1117  */
1118 int
1119 unp_connect2(struct socket *so, struct socket *so2)
1120 {
1121 	struct unpcb *unp, *unp2;
1122 	int error;
1123 
1124 	lwkt_gettoken(&unp_token);
1125 	if (so2->so_type != so->so_type) {
1126 		lwkt_reltoken(&unp_token);
1127 		return (EPROTOTYPE);
1128 	}
1129 	unp = unp_getsocktoken(so);
1130 	unp2 = unp_getsocktoken(so2);
1131 
1132 	if (!UNP_ISATTACHED(unp)) {
1133 		error = EINVAL;
1134 		goto done;
1135 	}
1136 	if (!UNP_ISATTACHED(unp2)) {
1137 		error = ECONNREFUSED;
1138 		goto done;
1139 	}
1140 
1141 	if (unp->unp_conn != NULL) {
1142 		error = EISCONN;
1143 		goto done;
1144 	}
1145 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1146 	    unp2->unp_conn != NULL) {
1147 		error = EISCONN;
1148 		goto done;
1149 	}
1150 
1151 	error = unp_connect_pair(unp, unp2);
1152 done:
1153 	unp_reltoken(unp2);
1154 	unp_reltoken(unp);
1155 	lwkt_reltoken(&unp_token);
1156 	return (error);
1157 }
1158 
1159 /*
1160  * Disconnect a unix domain socket pair.
1161  *
1162  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1163  *	 pool token also be held.
1164  */
1165 static void
1166 unp_disconnect(struct unpcb *unp)
1167 {
1168 	struct unpcb *unp2;
1169 
1170 	lwkt_gettoken(&unp_token);
1171 	lwkt_getpooltoken(unp);
1172 
1173 	while ((unp2 = unp->unp_conn) != NULL) {
1174 		lwkt_getpooltoken(unp2);
1175 		if (unp2 == unp->unp_conn)
1176 			break;
1177 		lwkt_relpooltoken(unp2);
1178 	}
1179 	if (unp2 == NULL)
1180 		goto done;
1181 
1182 	unp->unp_conn = NULL;
1183 
1184 	switch (unp->unp_socket->so_type) {
1185 	case SOCK_DGRAM:
1186 		LIST_REMOVE(unp, unp_reflink);
1187 		soclrstate(unp->unp_socket, SS_ISCONNECTED);
1188 		break;
1189 
1190 	case SOCK_STREAM:
1191 	case SOCK_SEQPACKET:
1192 		unp_reference(unp2);
1193 		unp2->unp_conn = NULL;
1194 
1195 		soisdisconnected(unp->unp_socket);
1196 		soisdisconnected(unp2->unp_socket);
1197 
1198 		unp_free(unp2);
1199 		break;
1200 	}
1201 	lwkt_relpooltoken(unp2);
1202 done:
1203 	lwkt_relpooltoken(unp);
1204 	lwkt_reltoken(&unp_token);
1205 }
1206 
1207 #ifdef notdef
1208 void
1209 unp_abort(struct unpcb *unp)
1210 {
1211 	lwkt_gettoken(&unp_token);
1212 	unp_free(unp);
1213 	lwkt_reltoken(&unp_token);
1214 }
1215 #endif
1216 
1217 static int
1218 prison_unpcb(struct thread *td, struct unpcb *unp)
1219 {
1220 	struct proc *p;
1221 
1222 	if (td == NULL)
1223 		return (0);
1224 	if ((p = td->td_proc) == NULL)
1225 		return (0);
1226 	if (!p->p_ucred->cr_prison)
1227 		return (0);
1228 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
1229 		return (0);
1230 	return (1);
1231 }
1232 
1233 static int
1234 unp_pcblist(SYSCTL_HANDLER_ARGS)
1235 {
1236 	int error, i, n;
1237 	struct unpcb *unp, **unp_list;
1238 	unp_gen_t gencnt;
1239 	struct unp_head *head;
1240 
1241 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1242 
1243 	KKASSERT(curproc != NULL);
1244 
1245 	/*
1246 	 * The process of preparing the PCB list is too time-consuming and
1247 	 * resource-intensive to repeat twice on every request.
1248 	 */
1249 	if (req->oldptr == NULL) {
1250 		n = unp_count;
1251 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
1252 		return 0;
1253 	}
1254 
1255 	if (req->newptr != NULL)
1256 		return EPERM;
1257 
1258 	lwkt_gettoken(&unp_token);
1259 
1260 	/*
1261 	 * OK, now we're committed to doing something.
1262 	 */
1263 	gencnt = unp_gencnt;
1264 	n = unp_count;
1265 
1266 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1267 
1268 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1269 	     unp = LIST_NEXT(unp, unp_link)) {
1270 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
1271 			unp_list[i++] = unp;
1272 	}
1273 	n = i;			/* in case we lost some during malloc */
1274 
1275 	error = 0;
1276 	for (i = 0; i < n; i++) {
1277 		unp = unp_list[i];
1278 		if (unp->unp_gencnt <= gencnt) {
1279 			struct xunpcb xu;
1280 			xu.xu_len = sizeof xu;
1281 			xu.xu_unpp = unp;
1282 			/*
1283 			 * XXX - need more locking here to protect against
1284 			 * connect/disconnect races for SMP.
1285 			 */
1286 			if (unp->unp_addr)
1287 				bcopy(unp->unp_addr, &xu.xu_addr,
1288 				      unp->unp_addr->sun_len);
1289 			if (unp->unp_conn && unp->unp_conn->unp_addr)
1290 				bcopy(unp->unp_conn->unp_addr,
1291 				      &xu.xu_caddr,
1292 				      unp->unp_conn->unp_addr->sun_len);
1293 			bcopy(unp, &xu.xu_unp, sizeof *unp);
1294 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
1295 			error = SYSCTL_OUT(req, &xu, sizeof xu);
1296 		}
1297 	}
1298 	lwkt_reltoken(&unp_token);
1299 	kfree(unp_list, M_TEMP);
1300 
1301 	return error;
1302 }
1303 
1304 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1305 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1306 	    "List of active local datagram sockets");
1307 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1308 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1309 	    "List of active local stream sockets");
1310 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
1311 	    (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1312 	    "List of active local seqpacket stream sockets");
1313 
1314 static void
1315 unp_shutdown(struct unpcb *unp)
1316 {
1317 	struct socket *so;
1318 
1319 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1320 	     unp->unp_socket->so_type == SOCK_SEQPACKET) &&
1321 	    unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
1322 		socantrcvmore(so);
1323 	}
1324 }
1325 
1326 static void
1327 unp_drop(struct unpcb *unp, int err)
1328 {
1329 	struct socket *so = unp->unp_socket;
1330 
1331 	so->so_error = err;
1332 	unp_disconnect(unp);
1333 }
1334 
1335 #ifdef notdef
1336 void
1337 unp_drain(void)
1338 {
1339 	lwkt_gettoken(&unp_token);
1340 	lwkt_reltoken(&unp_token);
1341 }
1342 #endif
1343 
1344 int
1345 unp_externalize(struct mbuf *rights)
1346 {
1347 	struct thread *td = curthread;
1348 	struct proc *p = td->td_proc;		/* XXX */
1349 	struct lwp *lp = td->td_lwp;
1350 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
1351 	int *fdp;
1352 	int i;
1353 	struct file **rp;
1354 	struct file *fp;
1355 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
1356 		/ sizeof (struct file *);
1357 	int f;
1358 
1359 	lwkt_gettoken(&unp_token);
1360 
1361 	/*
1362 	 * if the new FD's will not fit, then we free them all
1363 	 */
1364 	if (!fdavail(p, newfds)) {
1365 		rp = (struct file **)CMSG_DATA(cm);
1366 		for (i = 0; i < newfds; i++) {
1367 			fp = *rp;
1368 			/*
1369 			 * zero the pointer before calling unp_discard,
1370 			 * since it may end up in unp_gc()..
1371 			 */
1372 			*rp++ = NULL;
1373 			unp_discard(fp, NULL);
1374 		}
1375 		lwkt_reltoken(&unp_token);
1376 		return (EMSGSIZE);
1377 	}
1378 
1379 	/*
1380 	 * now change each pointer to an fd in the global table to
1381 	 * an integer that is the index to the local fd table entry
1382 	 * that we set up to point to the global one we are transferring.
1383 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1384 	 * then do it in forward order. In that case, an integer will
1385 	 * always come in the same place or before its corresponding
1386 	 * struct file pointer.
1387 	 * If sizeof (struct file *) is smaller than sizeof int, then
1388 	 * do it in reverse order.
1389 	 */
1390 	if (sizeof (struct file *) >= sizeof (int)) {
1391 		fdp = (int *)CMSG_DATA(cm);
1392 		rp = (struct file **)CMSG_DATA(cm);
1393 		for (i = 0; i < newfds; i++) {
1394 			if (fdalloc(p, 0, &f))
1395 				panic("unp_externalize");
1396 			fp = *rp++;
1397 			unp_fp_externalize(lp, fp, f);
1398 			*fdp++ = f;
1399 		}
1400 	} else {
1401 		fdp = (int *)CMSG_DATA(cm) + newfds - 1;
1402 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
1403 		for (i = 0; i < newfds; i++) {
1404 			if (fdalloc(p, 0, &f))
1405 				panic("unp_externalize");
1406 			fp = *rp--;
1407 			unp_fp_externalize(lp, fp, f);
1408 			*fdp-- = f;
1409 		}
1410 	}
1411 
1412 	/*
1413 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1414 	 * differs.
1415 	 */
1416 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1417 	rights->m_len = cm->cmsg_len;
1418 
1419 	lwkt_reltoken(&unp_token);
1420 	return (0);
1421 }
1422 
1423 static void
1424 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd)
1425 {
1426 	struct file *fx;
1427 	int error;
1428 
1429 	lwkt_gettoken(&unp_token);
1430 
1431 	if (lp) {
1432 		KKASSERT(fd >= 0);
1433 		if (fp->f_flag & FREVOKED) {
1434 			kprintf("Warning: revoked fp exiting unix socket\n");
1435 			fx = NULL;
1436 			error = falloc(lp, &fx, NULL);
1437 			if (error == 0)
1438 				fsetfd(lp->lwp_proc->p_fd, fx, fd);
1439 			else
1440 				fsetfd(lp->lwp_proc->p_fd, NULL, fd);
1441 			fdrop(fx);
1442 		} else {
1443 			fsetfd(lp->lwp_proc->p_fd, fp, fd);
1444 		}
1445 	}
1446 	spin_lock(&unp_spin);
1447 	fp->f_msgcount--;
1448 	unp_rights--;
1449 	spin_unlock(&unp_spin);
1450 	fdrop(fp);
1451 
1452 	lwkt_reltoken(&unp_token);
1453 }
1454 
1455 
1456 void
1457 unp_init(void)
1458 {
1459 	LIST_INIT(&unp_dhead);
1460 	LIST_INIT(&unp_shead);
1461 	spin_init(&unp_spin, "unpinit");
1462 }
1463 
1464 static int
1465 unp_internalize(struct mbuf *control, struct thread *td)
1466 {
1467 	struct proc *p = td->td_proc;
1468 	struct filedesc *fdescp;
1469 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1470 	struct file **rp;
1471 	struct file *fp;
1472 	int i, fd, *fdp;
1473 	struct cmsgcred *cmcred;
1474 	int oldfds;
1475 	u_int newlen;
1476 	int error;
1477 
1478 	KKASSERT(p);
1479 	lwkt_gettoken(&unp_token);
1480 
1481 	fdescp = p->p_fd;
1482 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1483 	    cm->cmsg_level != SOL_SOCKET ||
1484 	    CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1485 		error = EINVAL;
1486 		goto done;
1487 	}
1488 
1489 	/*
1490 	 * Fill in credential information.
1491 	 */
1492 	if (cm->cmsg_type == SCM_CREDS) {
1493 		cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1494 		cmcred->cmcred_pid = p->p_pid;
1495 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1496 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1497 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1498 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1499 							CMGROUP_MAX);
1500 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1501 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1502 		error = 0;
1503 		goto done;
1504 	}
1505 
1506 	/*
1507 	 * cmsghdr may not be aligned, do not allow calculation(s) to
1508 	 * go negative.
1509 	 */
1510 	if (cm->cmsg_len < CMSG_LEN(0)) {
1511 		error = EINVAL;
1512 		goto done;
1513 	}
1514 
1515 	oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1516 
1517 	/*
1518 	 * check that all the FDs passed in refer to legal OPEN files
1519 	 * If not, reject the entire operation.
1520 	 */
1521 	fdp = (int *)CMSG_DATA(cm);
1522 	for (i = 0; i < oldfds; i++) {
1523 		fd = *fdp++;
1524 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1525 		    fdescp->fd_files[fd].fp == NULL) {
1526 			error = EBADF;
1527 			goto done;
1528 		}
1529 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) {
1530 			error = EOPNOTSUPP;
1531 			goto done;
1532 		}
1533 	}
1534 	/*
1535 	 * Now replace the integer FDs with pointers to
1536 	 * the associated global file table entry..
1537 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1538 	 * enough, return E2BIG.
1539 	 */
1540 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1541 	if (newlen > MCLBYTES) {
1542 		error = E2BIG;
1543 		goto done;
1544 	}
1545 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1546 		if (control->m_flags & M_EXT) {
1547 			error = E2BIG;
1548 			goto done;
1549 		}
1550 		MCLGET(control, M_WAITOK);
1551 		if (!(control->m_flags & M_EXT)) {
1552 			error = ENOBUFS;
1553 			goto done;
1554 		}
1555 
1556 		/* copy the data to the cluster */
1557 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1558 		cm = mtod(control, struct cmsghdr *);
1559 	}
1560 
1561 	/*
1562 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1563 	 * differs.
1564 	 */
1565 	cm->cmsg_len = newlen;
1566 	control->m_len = CMSG_ALIGN(newlen);
1567 
1568 	/*
1569 	 * Transform the file descriptors into struct file pointers.
1570 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1571 	 * then do it in reverse order so that the int won't get until
1572 	 * we're done.
1573 	 * If sizeof (struct file *) is smaller than sizeof int, then
1574 	 * do it in forward order.
1575 	 */
1576 	if (sizeof (struct file *) >= sizeof (int)) {
1577 		fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1578 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1579 		for (i = 0; i < oldfds; i++) {
1580 			fp = fdescp->fd_files[*fdp--].fp;
1581 			*rp-- = fp;
1582 			fhold(fp);
1583 			spin_lock(&unp_spin);
1584 			fp->f_msgcount++;
1585 			unp_rights++;
1586 			spin_unlock(&unp_spin);
1587 		}
1588 	} else {
1589 		fdp = (int *)CMSG_DATA(cm);
1590 		rp = (struct file **)CMSG_DATA(cm);
1591 		for (i = 0; i < oldfds; i++) {
1592 			fp = fdescp->fd_files[*fdp++].fp;
1593 			*rp++ = fp;
1594 			fhold(fp);
1595 			spin_lock(&unp_spin);
1596 			fp->f_msgcount++;
1597 			unp_rights++;
1598 			spin_unlock(&unp_spin);
1599 		}
1600 	}
1601 	error = 0;
1602 done:
1603 	lwkt_reltoken(&unp_token);
1604 	return error;
1605 }
1606 
1607 /*
1608  * Garbage collect in-transit file descriptors that get lost due to
1609  * loops (i.e. when a socket is sent to another process over itself,
1610  * and more complex situations).
1611  *
1612  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1613  */
1614 
1615 struct unp_gc_info {
1616 	struct file **extra_ref;
1617 	struct file *locked_fp;
1618 	int defer;
1619 	int index;
1620 	int maxindex;
1621 };
1622 
1623 static void
1624 unp_gc(void)
1625 {
1626 	struct unp_gc_info info;
1627 	static boolean_t unp_gcing;
1628 	struct file **fpp;
1629 	int i;
1630 
1631 	/*
1632 	 * Only one gc can be in-progress at any given moment
1633 	 */
1634 	spin_lock(&unp_spin);
1635 	if (unp_gcing) {
1636 		spin_unlock(&unp_spin);
1637 		return;
1638 	}
1639 	unp_gcing = TRUE;
1640 	spin_unlock(&unp_spin);
1641 
1642 	lwkt_gettoken(&unp_token);
1643 
1644 	/*
1645 	 * Before going through all this, set all FDs to be NOT defered
1646 	 * and NOT externally accessible (not marked).  During the scan
1647 	 * a fd can be marked externally accessible but we may or may not
1648 	 * be able to immediately process it (controlled by FDEFER).
1649 	 *
1650 	 * If we loop sleep a bit.  The complexity of the topology can cause
1651 	 * multiple loops.  Also failure to acquire the socket's so_rcv
1652 	 * token can cause us to loop.
1653 	 */
1654 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1655 	do {
1656 		info.defer = 0;
1657 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1658 		if (info.defer)
1659 			tsleep(&info, 0, "gcagain", 1);
1660 	} while (info.defer);
1661 
1662 	/*
1663 	 * We grab an extra reference to each of the file table entries
1664 	 * that are not otherwise accessible and then free the rights
1665 	 * that are stored in messages on them.
1666 	 *
1667 	 * The bug in the orginal code is a little tricky, so I'll describe
1668 	 * what's wrong with it here.
1669 	 *
1670 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1671 	 * times -- consider the case of sockets A and B that contain
1672 	 * references to each other.  On a last close of some other socket,
1673 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1674 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1675 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1676 	 * results in the following chain.  Closef calls soo_close, which
1677 	 * calls soclose.   Soclose calls first (through the switch
1678 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1679 	 * returns because the previous instance had set unp_gcing, and
1680 	 * we return all the way back to soclose, which marks the socket
1681 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1682 	 * to free up the rights that are queued in messages on the socket A,
1683 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1684 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1685 	 * instance of unp_discard just calls closef on B.
1686 	 *
1687 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1688 	 * which results in another closef on A.  Unfortunately, A is already
1689 	 * being closed, and the descriptor has already been marked with
1690 	 * SS_NOFDREF, and soclose panics at this point.
1691 	 *
1692 	 * Here, we first take an extra reference to each inaccessible
1693 	 * descriptor.  Then, we call sorflush ourself, since we know
1694 	 * it is a Unix domain socket anyhow.  After we destroy all the
1695 	 * rights carried in messages, we do a last closef to get rid
1696 	 * of our extra reference.  This is the last close, and the
1697 	 * unp_detach etc will shut down the socket.
1698 	 *
1699 	 * 91/09/19, bsy@cs.cmu.edu
1700 	 */
1701 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1702 	info.maxindex = 256;
1703 
1704 	do {
1705 		/*
1706 		 * Look for matches
1707 		 */
1708 		info.index = 0;
1709 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1710 
1711 		/*
1712 		 * For each FD on our hit list, do the following two things
1713 		 */
1714 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1715 			struct file *tfp = *fpp;
1716 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1717 				sorflush((struct socket *)(tfp->f_data));
1718 		}
1719 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1720 			closef(*fpp, NULL);
1721 	} while (info.index == info.maxindex);
1722 
1723 	lwkt_reltoken(&unp_token);
1724 
1725 	kfree((caddr_t)info.extra_ref, M_FILE);
1726 	unp_gcing = FALSE;
1727 }
1728 
1729 /*
1730  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1731  */
1732 static int
1733 unp_gc_checkrefs(struct file *fp, void *data)
1734 {
1735 	struct unp_gc_info *info = data;
1736 
1737 	if (fp->f_count == 0)
1738 		return(0);
1739 	if (info->index == info->maxindex)
1740 		return(-1);
1741 
1742 	/*
1743 	 * If all refs are from msgs, and it's not marked accessible
1744 	 * then it must be referenced from some unreachable cycle
1745 	 * of (shut-down) FDs, so include it in our
1746 	 * list of FDs to remove
1747 	 */
1748 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1749 		info->extra_ref[info->index++] = fp;
1750 		fhold(fp);
1751 	}
1752 	return(0);
1753 }
1754 
1755 /*
1756  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1757  */
1758 static int
1759 unp_gc_clearmarks(struct file *fp, void *data __unused)
1760 {
1761 	atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1762 	return(0);
1763 }
1764 
1765 /*
1766  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1767  */
1768 static int
1769 unp_gc_checkmarks(struct file *fp, void *data)
1770 {
1771 	struct unp_gc_info *info = data;
1772 	struct socket *so;
1773 
1774 	/*
1775 	 * If the file is not open, skip it.  Make sure it isn't marked
1776 	 * defered or we could loop forever, in case we somehow race
1777 	 * something.
1778 	 */
1779 	if (fp->f_count == 0) {
1780 		if (fp->f_flag & FDEFER)
1781 			atomic_clear_int(&fp->f_flag, FDEFER);
1782 		return(0);
1783 	}
1784 	/*
1785 	 * If we already marked it as 'defer'  in a
1786 	 * previous pass, then try process it this time
1787 	 * and un-mark it
1788 	 */
1789 	if (fp->f_flag & FDEFER) {
1790 		atomic_clear_int(&fp->f_flag, FDEFER);
1791 	} else {
1792 		/*
1793 		 * if it's not defered, then check if it's
1794 		 * already marked.. if so skip it
1795 		 */
1796 		if (fp->f_flag & FMARK)
1797 			return(0);
1798 		/*
1799 		 * If all references are from messages
1800 		 * in transit, then skip it. it's not
1801 		 * externally accessible.
1802 		 */
1803 		if (fp->f_count == fp->f_msgcount)
1804 			return(0);
1805 		/*
1806 		 * If it got this far then it must be
1807 		 * externally accessible.
1808 		 */
1809 		atomic_set_int(&fp->f_flag, FMARK);
1810 	}
1811 
1812 	/*
1813 	 * either it was defered, or it is externally
1814 	 * accessible and not already marked so.
1815 	 * Now check if it is possibly one of OUR sockets.
1816 	 */
1817 	if (fp->f_type != DTYPE_SOCKET ||
1818 	    (so = (struct socket *)fp->f_data) == NULL) {
1819 		return(0);
1820 	}
1821 	if (so->so_proto->pr_domain != &localdomain ||
1822 	    !(so->so_proto->pr_flags & PR_RIGHTS)) {
1823 		return(0);
1824 	}
1825 
1826 	/*
1827 	 * So, Ok, it's one of our sockets and it IS externally accessible
1828 	 * (or was defered).  Now we look to see if we hold any file
1829 	 * descriptors in its message buffers.  Follow those links and mark
1830 	 * them as accessible too.
1831 	 *
1832 	 * We are holding multiple spinlocks here, if we cannot get the
1833 	 * token non-blocking defer until the next loop.
1834 	 */
1835 	info->locked_fp = fp;
1836 	if (lwkt_trytoken(&so->so_rcv.ssb_token)) {
1837 		unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1838 		lwkt_reltoken(&so->so_rcv.ssb_token);
1839 	} else {
1840 		atomic_set_int(&fp->f_flag, FDEFER);
1841 		++info->defer;
1842 	}
1843 	return (0);
1844 }
1845 
1846 /*
1847  * Scan all unix domain sockets and replace any revoked file pointers
1848  * found with the dummy file pointer fx.  We don't worry about races
1849  * against file pointers being read out as those are handled in the
1850  * externalize code.
1851  */
1852 
1853 #define REVOKE_GC_MAXFILES	32
1854 
1855 struct unp_revoke_gc_info {
1856 	struct file	*fx;
1857 	struct file	*fary[REVOKE_GC_MAXFILES];
1858 	int		fcount;
1859 };
1860 
1861 void
1862 unp_revoke_gc(struct file *fx)
1863 {
1864 	struct unp_revoke_gc_info info;
1865 	int i;
1866 
1867 	lwkt_gettoken(&unp_token);
1868 	info.fx = fx;
1869 	do {
1870 		info.fcount = 0;
1871 		allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1872 		for (i = 0; i < info.fcount; ++i)
1873 			unp_fp_externalize(NULL, info.fary[i], -1);
1874 	} while (info.fcount == REVOKE_GC_MAXFILES);
1875 	lwkt_reltoken(&unp_token);
1876 }
1877 
1878 /*
1879  * Check for and replace revoked descriptors.
1880  *
1881  * WARNING:  This routine is not allowed to block.
1882  */
1883 static int
1884 unp_revoke_gc_check(struct file *fps, void *vinfo)
1885 {
1886 	struct unp_revoke_gc_info *info = vinfo;
1887 	struct file *fp;
1888 	struct socket *so;
1889 	struct mbuf *m0;
1890 	struct mbuf *m;
1891 	struct file **rp;
1892 	struct cmsghdr *cm;
1893 	int i;
1894 	int qfds;
1895 
1896 	/*
1897 	 * Is this a unix domain socket with rights-passing abilities?
1898 	 */
1899 	if (fps->f_type != DTYPE_SOCKET)
1900 		return (0);
1901 	if ((so = (struct socket *)fps->f_data) == NULL)
1902 		return(0);
1903 	if (so->so_proto->pr_domain != &localdomain)
1904 		return(0);
1905 	if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1906 		return(0);
1907 
1908 	/*
1909 	 * Scan the mbufs for control messages and replace any revoked
1910 	 * descriptors we find.
1911 	 */
1912 	lwkt_gettoken(&so->so_rcv.ssb_token);
1913 	m0 = so->so_rcv.ssb_mb;
1914 	while (m0) {
1915 		for (m = m0; m; m = m->m_next) {
1916 			if (m->m_type != MT_CONTROL)
1917 				continue;
1918 			if (m->m_len < sizeof(*cm))
1919 				continue;
1920 			cm = mtod(m, struct cmsghdr *);
1921 			if (cm->cmsg_level != SOL_SOCKET ||
1922 			    cm->cmsg_type != SCM_RIGHTS) {
1923 				continue;
1924 			}
1925 			qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1926 			rp = (struct file **)CMSG_DATA(cm);
1927 			for (i = 0; i < qfds; i++) {
1928 				fp = rp[i];
1929 				if (fp->f_flag & FREVOKED) {
1930 					kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1931 					fhold(info->fx);
1932 					info->fx->f_msgcount++;
1933 					unp_rights++;
1934 					rp[i] = info->fx;
1935 					info->fary[info->fcount++] = fp;
1936 				}
1937 				if (info->fcount == REVOKE_GC_MAXFILES)
1938 					break;
1939 			}
1940 			if (info->fcount == REVOKE_GC_MAXFILES)
1941 				break;
1942 		}
1943 		m0 = m0->m_nextpkt;
1944 		if (info->fcount == REVOKE_GC_MAXFILES)
1945 			break;
1946 	}
1947 	lwkt_reltoken(&so->so_rcv.ssb_token);
1948 
1949 	/*
1950 	 * Stop the scan if we filled up our array.
1951 	 */
1952 	if (info->fcount == REVOKE_GC_MAXFILES)
1953 		return(-1);
1954 	return(0);
1955 }
1956 
1957 /*
1958  * Dispose of the fp's stored in a mbuf.
1959  *
1960  * The dds loop can cause additional fps to be entered onto the
1961  * list while it is running, flattening out the operation and avoiding
1962  * a deep kernel stack recursion.
1963  */
1964 void
1965 unp_dispose(struct mbuf *m)
1966 {
1967 	unp_defdiscard_t dds;
1968 
1969 	lwkt_gettoken(&unp_token);
1970 	++unp_defdiscard_nest;
1971 	if (m) {
1972 		unp_scan(m, unp_discard, NULL);
1973 	}
1974 	if (unp_defdiscard_nest == 1) {
1975 		while ((dds = unp_defdiscard_base) != NULL) {
1976 			unp_defdiscard_base = dds->next;
1977 			closef(dds->fp, NULL);
1978 			kfree(dds, M_UNPCB);
1979 		}
1980 	}
1981 	--unp_defdiscard_nest;
1982 	lwkt_reltoken(&unp_token);
1983 }
1984 
1985 static int
1986 unp_listen(struct unpcb *unp, struct thread *td)
1987 {
1988 	struct proc *p = td->td_proc;
1989 
1990 	KKASSERT(p);
1991 	lwkt_gettoken(&unp_token);
1992 	cru2x(p->p_ucred, &unp->unp_peercred);
1993 	unp_setflags(unp, UNP_HAVEPCCACHED);
1994 	lwkt_reltoken(&unp_token);
1995 	return (0);
1996 }
1997 
1998 static void
1999 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
2000 {
2001 	struct mbuf *m;
2002 	struct file **rp;
2003 	struct cmsghdr *cm;
2004 	int i;
2005 	int qfds;
2006 
2007 	while (m0) {
2008 		for (m = m0; m; m = m->m_next) {
2009 			if (m->m_type == MT_CONTROL &&
2010 			    m->m_len >= sizeof(*cm)) {
2011 				cm = mtod(m, struct cmsghdr *);
2012 				if (cm->cmsg_level != SOL_SOCKET ||
2013 				    cm->cmsg_type != SCM_RIGHTS)
2014 					continue;
2015 				qfds = (cm->cmsg_len - CMSG_LEN(0)) /
2016 					sizeof(void *);
2017 				rp = (struct file **)CMSG_DATA(cm);
2018 				for (i = 0; i < qfds; i++)
2019 					(*op)(*rp++, data);
2020 				break;		/* XXX, but saves time */
2021 			}
2022 		}
2023 		m0 = m0->m_nextpkt;
2024 	}
2025 }
2026 
2027 /*
2028  * Mark visibility.  info->defer is recalculated on every pass.
2029  */
2030 static void
2031 unp_mark(struct file *fp, void *data)
2032 {
2033 	struct unp_gc_info *info = data;
2034 
2035 	if ((fp->f_flag & FMARK) == 0) {
2036 		++info->defer;
2037 		atomic_set_int(&fp->f_flag, FMARK | FDEFER);
2038 	} else if (fp->f_flag & FDEFER) {
2039 		++info->defer;
2040 	}
2041 }
2042 
2043 /*
2044  * Discard a fp previously held in a unix domain socket mbuf.  To
2045  * avoid blowing out the kernel stack due to contrived chain-reactions
2046  * we may have to defer the operation to a higher procedural level.
2047  *
2048  * Caller holds unp_token
2049  */
2050 static void
2051 unp_discard(struct file *fp, void *data __unused)
2052 {
2053 	unp_defdiscard_t dds;
2054 
2055 	spin_lock(&unp_spin);
2056 	fp->f_msgcount--;
2057 	unp_rights--;
2058 	spin_unlock(&unp_spin);
2059 
2060 	if (unp_defdiscard_nest) {
2061 		dds = kmalloc(sizeof(*dds), M_UNPCB, M_WAITOK|M_ZERO);
2062 		dds->fp = fp;
2063 		dds->next = unp_defdiscard_base;
2064 		unp_defdiscard_base = dds;
2065 	} else {
2066 		closef(fp, NULL);
2067 	}
2068 }
2069 
2070 static int
2071 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type,
2072     struct unpcb **unp_ret)
2073 {
2074 	struct proc *p = td->td_proc;
2075 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
2076 	struct vnode *vp = NULL;
2077 	struct socket *so;
2078 	struct unpcb *unp;
2079 	int error, len;
2080 	struct nlookupdata nd;
2081 	char buf[SOCK_MAXADDRLEN];
2082 
2083 	*unp_ret = NULL;
2084 
2085 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2086 	if (len <= 0) {
2087 		error = EINVAL;
2088 		goto failed;
2089 	}
2090 	strncpy(buf, soun->sun_path, len);
2091 	buf[len] = 0;
2092 
2093 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
2094 	if (error == 0)
2095 		error = nlookup(&nd);
2096 	if (error == 0)
2097 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2098 	nlookup_done(&nd);
2099 	if (error) {
2100 		vp = NULL;
2101 		goto failed;
2102 	}
2103 
2104 	if (vp->v_type != VSOCK) {
2105 		error = ENOTSOCK;
2106 		goto failed;
2107 	}
2108 	error = VOP_EACCESS(vp, VWRITE, p->p_ucred);
2109 	if (error)
2110 		goto failed;
2111 	so = vp->v_socket;
2112 	if (so == NULL) {
2113 		error = ECONNREFUSED;
2114 		goto failed;
2115 	}
2116 	if (so->so_type != type) {
2117 		error = EPROTOTYPE;
2118 		goto failed;
2119 	}
2120 
2121 	/* Lock this unp. */
2122 	unp = unp_getsocktoken(so);
2123 	if (!UNP_ISATTACHED(unp)) {
2124 		unp_reltoken(unp);
2125 		error = ECONNREFUSED;
2126 		goto failed;
2127 	}
2128 	/* And keep this unp referenced. */
2129 	unp_reference(unp);
2130 
2131 	/* Done! */
2132 	*unp_ret = unp;
2133 	error = 0;
2134 failed:
2135 	if (vp != NULL)
2136 		vput(vp);
2137 	return error;
2138 }
2139 
2140 static int
2141 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2)
2142 {
2143 	struct socket *so = unp->unp_socket;
2144 	struct socket *so2 = unp2->unp_socket;
2145 
2146 	UNP_ASSERT_TOKEN_HELD(unp);
2147 	UNP_ASSERT_TOKEN_HELD(unp2);
2148 
2149 	KASSERT(so->so_type == so2->so_type,
2150 	    ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type));
2151 
2152 	if (!UNP_ISATTACHED(unp))
2153 		return EINVAL;
2154 	if (!UNP_ISATTACHED(unp2))
2155 		return ECONNREFUSED;
2156 
2157 	KASSERT(unp->unp_conn == NULL, ("unp is already connected"));
2158 	unp->unp_conn = unp2;
2159 
2160 	switch (so->so_type) {
2161 	case SOCK_DGRAM:
2162 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2163 		soisconnected(so);
2164 		break;
2165 
2166 	case SOCK_STREAM:
2167 	case SOCK_SEQPACKET:
2168 		KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected"));
2169 		unp2->unp_conn = unp;
2170 		soisconnected(so);
2171 		soisconnected(so2);
2172 		break;
2173 
2174 	default:
2175 		panic("unp_connect_pair: unknown socket type %d", so->so_type);
2176 	}
2177 	return 0;
2178 }
2179