xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision 8b927cb7b72266b97393ee565d882d7fddfa4375)
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 #define UNP_DROPPED		UNP_PRIVATE3
63 
64 #define UNP_ISATTACHED(unp)	\
65     ((unp) != NULL && ((unp)->unp_flags & UNP_DETACHED) == 0)
66 
67 #ifdef INVARIANTS
68 #define UNP_ASSERT_TOKEN_HELD(unp) \
69     ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup((unp)))
70 #else	/* !INVARIANTS */
71 #define UNP_ASSERT_TOKEN_HELD(unp)
72 #endif	/* INVARIANTS */
73 
74 typedef struct unp_defdiscard {
75 	struct unp_defdiscard *next;
76 	struct file *fp;
77 } *unp_defdiscard_t;
78 
79 static	MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
80 static	unp_gen_t unp_gencnt;
81 static	u_int unp_count;
82 
83 static	struct unp_head unp_shead, unp_dhead;
84 
85 static struct lwkt_token unp_token = LWKT_TOKEN_INITIALIZER(unp_token);
86 static int unp_defdiscard_nest;
87 static unp_defdiscard_t unp_defdiscard_base;
88 
89 /*
90  * Unix communications domain.
91  *
92  * TODO:
93  *	RDM
94  *	rethink name space problems
95  *	need a proper out-of-band
96  *	lock pushdown
97  */
98 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
99 static ino_t	unp_ino = 1;		/* prototype for fake inode numbers */
100 static struct spinlock unp_ino_spin = SPINLOCK_INITIALIZER(&unp_ino_spin, "unp_ino_spin");
101 
102 static int     unp_attach (struct socket *, struct pru_attach_info *);
103 static void    unp_detach (struct unpcb *);
104 static int     unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
105 static int     unp_connect (struct socket *,struct sockaddr *,
106 				struct thread *);
107 static void    unp_disconnect(struct unpcb *, int);
108 static void    unp_shutdown (struct unpcb *);
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 static void    unp_drop(struct unpcb *unp, int error);
125 
126 /*
127  * SMP Considerations:
128  *
129  *	Since unp_token will be automaticly released upon execution of
130  *	blocking code, we need to reference unp_conn before any possible
131  *	blocking code to prevent it from being ripped behind our back.
132  *
133  *	Any adjustment to unp->unp_conn requires both the global unp_token
134  *	AND the per-unp token (lwkt_token_pool_lookup(unp)) to be held.
135  *
136  *	Any access to so_pcb to obtain unp requires the pool token for
137  *	unp to be held.
138  */
139 
140 static __inline void
141 unp_reference(struct unpcb *unp)
142 {
143 	/* 0->1 transition will not work */
144 	KKASSERT(unp->unp_refcnt > 0);
145 	atomic_add_int(&unp->unp_refcnt, 1);
146 }
147 
148 static __inline void
149 unp_free(struct unpcb *unp)
150 {
151 	KKASSERT(unp->unp_refcnt > 0);
152 	if (atomic_fetchadd_int(&unp->unp_refcnt, -1) == 1)
153 		unp_detach(unp);
154 }
155 
156 static __inline struct unpcb *
157 unp_getsocktoken(struct socket *so)
158 {
159 	struct unpcb *unp;
160 
161 	/*
162 	 * The unp pointer is invalid until we verify that it is
163 	 * good by re-checking so_pcb AFTER obtaining the token.
164 	 */
165 	while ((unp = so->so_pcb) != NULL) {
166 		lwkt_getpooltoken(unp);
167 		if (unp == so->so_pcb)
168 			break;
169 		lwkt_relpooltoken(unp);
170 	}
171 	return unp;
172 }
173 
174 static __inline void
175 unp_reltoken(struct unpcb *unp)
176 {
177 	if (unp != NULL)
178 		lwkt_relpooltoken(unp);
179 }
180 
181 static __inline void
182 unp_setflags(struct unpcb *unp, int flags)
183 {
184 	atomic_set_int(&unp->unp_flags, flags);
185 }
186 
187 static __inline void
188 unp_clrflags(struct unpcb *unp, int flags)
189 {
190 	atomic_clear_int(&unp->unp_flags, flags);
191 }
192 
193 /*
194  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
195  *	 will sofree() it when we return.
196  */
197 static void
198 uipc_abort(netmsg_t msg)
199 {
200 	struct unpcb *unp;
201 	int error;
202 
203 	lwkt_gettoken(&unp_token);
204 	unp = unp_getsocktoken(msg->base.nm_so);
205 
206 	if (UNP_ISATTACHED(unp)) {
207 		unp_setflags(unp, UNP_DETACHED);
208 		unp_drop(unp, ECONNABORTED);
209 		unp_free(unp);
210 		error = 0;
211 	} else {
212 		error = EINVAL;
213 	}
214 
215 	unp_reltoken(unp);
216 	lwkt_reltoken(&unp_token);
217 
218 	lwkt_replymsg(&msg->lmsg, error);
219 }
220 
221 static void
222 uipc_accept(netmsg_t msg)
223 {
224 	struct unpcb *unp;
225 	int error;
226 
227 	lwkt_gettoken(&unp_token);
228 	unp = unp_getsocktoken(msg->base.nm_so);
229 
230 	if (!UNP_ISATTACHED(unp)) {
231 		error = EINVAL;
232 	} else {
233 		struct unpcb *unp2 = unp->unp_conn;
234 
235 		/*
236 		 * Pass back name of connected socket,
237 		 * if it was bound and we are still connected
238 		 * (our peer may have closed already!).
239 		 */
240 		if (unp2 && unp2->unp_addr) {
241 			unp_reference(unp2);
242 			*msg->accept.nm_nam = dup_sockaddr(
243 				(struct sockaddr *)unp2->unp_addr);
244 			unp_free(unp2);
245 		} else {
246 			*msg->accept.nm_nam = dup_sockaddr(&sun_noname);
247 		}
248 		error = 0;
249 	}
250 
251 	unp_reltoken(unp);
252 	lwkt_reltoken(&unp_token);
253 
254 	lwkt_replymsg(&msg->lmsg, error);
255 }
256 
257 static void
258 uipc_attach(netmsg_t msg)
259 {
260 	int error;
261 
262 	lwkt_gettoken(&unp_token);
263 
264 	KASSERT(msg->base.nm_so->so_pcb == NULL, ("double unp attach"));
265 	error = unp_attach(msg->base.nm_so, msg->attach.nm_ai);
266 
267 	lwkt_reltoken(&unp_token);
268 	lwkt_replymsg(&msg->lmsg, error);
269 }
270 
271 static void
272 uipc_bind(netmsg_t msg)
273 {
274 	struct unpcb *unp;
275 	int error;
276 
277 	lwkt_gettoken(&unp_token);
278 	unp = unp_getsocktoken(msg->base.nm_so);
279 
280 	if (UNP_ISATTACHED(unp))
281 		error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td);
282 	else
283 		error = EINVAL;
284 
285 	unp_reltoken(unp);
286 	lwkt_reltoken(&unp_token);
287 
288 	lwkt_replymsg(&msg->lmsg, error);
289 }
290 
291 static void
292 uipc_connect(netmsg_t msg)
293 {
294 	int error;
295 
296 	error = unp_connect(msg->base.nm_so, msg->connect.nm_nam,
297 	    msg->connect.nm_td);
298 	lwkt_replymsg(&msg->lmsg, error);
299 }
300 
301 static void
302 uipc_connect2(netmsg_t msg)
303 {
304 	int error;
305 
306 	error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2);
307 	lwkt_replymsg(&msg->lmsg, error);
308 }
309 
310 /* control is EOPNOTSUPP */
311 
312 static void
313 uipc_detach(netmsg_t msg)
314 {
315 	struct unpcb *unp;
316 	int error;
317 
318 	lwkt_gettoken(&unp_token);
319 	unp = unp_getsocktoken(msg->base.nm_so);
320 
321 	if (UNP_ISATTACHED(unp)) {
322 		unp_setflags(unp, UNP_DETACHED);
323 		unp_drop(unp, 0);
324 		unp_free(unp);
325 		error = 0;
326 	} else {
327 		error = EINVAL;
328 	}
329 
330 	unp_reltoken(unp);
331 	lwkt_reltoken(&unp_token);
332 
333 	lwkt_replymsg(&msg->lmsg, error);
334 }
335 
336 static void
337 uipc_disconnect(netmsg_t msg)
338 {
339 	struct unpcb *unp;
340 	int error;
341 
342 	lwkt_gettoken(&unp_token);
343 	unp = unp_getsocktoken(msg->base.nm_so);
344 
345 	if (UNP_ISATTACHED(unp)) {
346 		unp_disconnect(unp, 0);
347 		error = 0;
348 	} else {
349 		error = EINVAL;
350 	}
351 
352 	unp_reltoken(unp);
353 	lwkt_reltoken(&unp_token);
354 
355 	lwkt_replymsg(&msg->lmsg, error);
356 }
357 
358 static void
359 uipc_listen(netmsg_t msg)
360 {
361 	struct unpcb *unp;
362 	int error;
363 
364 	lwkt_gettoken(&unp_token);
365 	unp = unp_getsocktoken(msg->base.nm_so);
366 
367 	if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL)
368 		error = EINVAL;
369 	else
370 		error = unp_listen(unp, msg->listen.nm_td);
371 
372 	unp_reltoken(unp);
373 	lwkt_reltoken(&unp_token);
374 
375 	lwkt_replymsg(&msg->lmsg, error);
376 }
377 
378 static void
379 uipc_peeraddr(netmsg_t msg)
380 {
381 	struct unpcb *unp;
382 	int error;
383 
384 	lwkt_gettoken(&unp_token);
385 	unp = unp_getsocktoken(msg->base.nm_so);
386 
387 	if (!UNP_ISATTACHED(unp)) {
388 		error = EINVAL;
389 	} else if (unp->unp_conn && unp->unp_conn->unp_addr) {
390 		struct unpcb *unp2 = unp->unp_conn;
391 
392 		unp_reference(unp2);
393 		*msg->peeraddr.nm_nam = dup_sockaddr(
394 				(struct sockaddr *)unp2->unp_addr);
395 		unp_free(unp2);
396 		error = 0;
397 	} else {
398 		/*
399 		 * XXX: It seems that this test always fails even when
400 		 * connection is established.  So, this else clause is
401 		 * added as workaround to return PF_LOCAL sockaddr.
402 		 */
403 		*msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname);
404 		error = 0;
405 	}
406 
407 	unp_reltoken(unp);
408 	lwkt_reltoken(&unp_token);
409 
410 	lwkt_replymsg(&msg->lmsg, error);
411 }
412 
413 static void
414 uipc_rcvd(netmsg_t msg)
415 {
416 	struct unpcb *unp, *unp2;
417 	struct socket *so;
418 	struct socket *so2;
419 	int error;
420 
421 	/*
422 	 * so_pcb is only modified with both the global and the unp
423 	 * pool token held.
424 	 */
425 	so = msg->base.nm_so;
426 	unp = unp_getsocktoken(so);
427 
428 	if (!UNP_ISATTACHED(unp)) {
429 		error = EINVAL;
430 		goto done;
431 	}
432 
433 	switch (so->so_type) {
434 	case SOCK_DGRAM:
435 		panic("uipc_rcvd DGRAM?");
436 		/*NOTREACHED*/
437 	case SOCK_STREAM:
438 	case SOCK_SEQPACKET:
439 		if (unp->unp_conn == NULL)
440 			break;
441 		unp2 = unp->unp_conn;	/* protected by pool token */
442 
443 		/*
444 		 * Because we are transfering mbufs directly to the
445 		 * peer socket we have to use SSB_STOP on the sender
446 		 * to prevent it from building up infinite mbufs.
447 		 *
448 		 * As in several places in this module w ehave to ref unp2
449 		 * to ensure that it does not get ripped out from under us
450 		 * if we block on the so2 token or in sowwakeup().
451 		 */
452 		so2 = unp2->unp_socket;
453 		unp_reference(unp2);
454 		lwkt_gettoken(&so2->so_rcv.ssb_token);
455 		if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
456 		    so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
457 		) {
458 			atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP);
459 
460 			sowwakeup(so2);
461 		}
462 		lwkt_reltoken(&so2->so_rcv.ssb_token);
463 		unp_free(unp2);
464 		break;
465 	default:
466 		panic("uipc_rcvd unknown socktype");
467 		/*NOTREACHED*/
468 	}
469 	error = 0;
470 done:
471 	unp_reltoken(unp);
472 	lwkt_replymsg(&msg->lmsg, error);
473 }
474 
475 /* pru_rcvoob is EOPNOTSUPP */
476 
477 static void
478 uipc_send(netmsg_t msg)
479 {
480 	struct unpcb *unp, *unp2;
481 	struct socket *so;
482 	struct socket *so2;
483 	struct mbuf *control;
484 	struct mbuf *m;
485 	int error = 0;
486 
487 	so = msg->base.nm_so;
488 	control = msg->send.nm_control;
489 	m = msg->send.nm_m;
490 
491 	/*
492 	 * so_pcb is only modified with both the global and the unp
493 	 * pool token held.
494 	 */
495 	so = msg->base.nm_so;
496 	unp = unp_getsocktoken(so);
497 
498 	if (!UNP_ISATTACHED(unp)) {
499 		error = EINVAL;
500 		goto release;
501 	}
502 
503 	if (msg->send.nm_flags & PRUS_OOB) {
504 		error = EOPNOTSUPP;
505 		goto release;
506 	}
507 
508 	wakeup_start_delayed();
509 
510 	if (control && (error = unp_internalize(control, msg->send.nm_td)))
511 		goto release;
512 
513 	switch (so->so_type) {
514 	case SOCK_DGRAM:
515 	{
516 		struct sockaddr *from;
517 
518 		if (msg->send.nm_addr) {
519 			if (unp->unp_conn) {
520 				error = EISCONN;
521 				break;
522 			}
523 			error = unp_find_lockref(msg->send.nm_addr,
524 			    msg->send.nm_td, so->so_type, &unp2);
525 			if (error)
526 				break;
527 			/*
528 			 * NOTE:
529 			 * unp2 is locked and referenced.
530 			 *
531 			 * We could unlock unp2 now, since it was checked
532 			 * and referenced.
533 			 */
534 			unp_reltoken(unp2);
535 		} else {
536 			if (unp->unp_conn == NULL) {
537 				error = ENOTCONN;
538 				break;
539 			}
540 			unp2 = unp->unp_conn;
541 			unp_reference(unp2);
542 		}
543 		/* NOTE: unp2 is referenced. */
544 		so2 = unp2->unp_socket;
545 
546 		if (unp->unp_addr)
547 			from = (struct sockaddr *)unp->unp_addr;
548 		else
549 			from = &sun_noname;
550 
551 		lwkt_gettoken(&so2->so_rcv.ssb_token);
552 		if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
553 			sorwakeup(so2);
554 			m = NULL;
555 			control = NULL;
556 		} else {
557 			error = ENOBUFS;
558 		}
559 		lwkt_reltoken(&so2->so_rcv.ssb_token);
560 
561 		unp_free(unp2);
562 		break;
563 	}
564 
565 	case SOCK_STREAM:
566 	case SOCK_SEQPACKET:
567 		/* Connect if not connected yet. */
568 		/*
569 		 * Note: A better implementation would complain
570 		 * if not equal to the peer's address.
571 		 */
572 		if (unp->unp_conn == NULL) {
573 			if (msg->send.nm_addr) {
574 				error = unp_connect(so,
575 						    msg->send.nm_addr,
576 						    msg->send.nm_td);
577 				if (error)
578 					break;	/* XXX */
579 			}
580 			/*
581 			 * NOTE:
582 			 * unp_conn still could be NULL, even if the
583 			 * above unp_connect() succeeds; since the
584 			 * current unp's token could be released due
585 			 * to blocking operations after unp_conn is
586 			 * assigned.
587 			 */
588 			if (unp->unp_conn == NULL) {
589 				error = ENOTCONN;
590 				break;
591 			}
592 		}
593 		if (so->so_state & SS_CANTSENDMORE) {
594 			error = EPIPE;
595 			break;
596 		}
597 
598 		unp2 = unp->unp_conn;
599 		KASSERT(unp2 != NULL, ("unp is not connected"));
600 		so2 = unp2->unp_socket;
601 
602 		unp_reference(unp2);
603 
604 		/*
605 		 * Send to paired receive port, and then reduce
606 		 * send buffer hiwater marks to maintain backpressure.
607 		 * Wake up readers.
608 		 */
609 		lwkt_gettoken(&so2->so_rcv.ssb_token);
610 		if (control) {
611 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
612 				control = NULL;
613 				m = NULL;
614 			}
615 		} else if (so->so_type == SOCK_SEQPACKET) {
616 			sbappendrecord(&so2->so_rcv.sb, m);
617 			m = NULL;
618 		} else {
619 			sbappend(&so2->so_rcv.sb, m);
620 			m = NULL;
621 		}
622 
623 		/*
624 		 * Because we are transfering mbufs directly to the
625 		 * peer socket we have to use SSB_STOP on the sender
626 		 * to prevent it from building up infinite mbufs.
627 		 */
628 		if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
629 		    so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
630 		) {
631 			atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP);
632 		}
633 		lwkt_reltoken(&so2->so_rcv.ssb_token);
634 		sorwakeup(so2);
635 
636 		unp_free(unp2);
637 		break;
638 
639 	default:
640 		panic("uipc_send unknown socktype");
641 	}
642 
643 	/*
644 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
645 	 */
646 	if (msg->send.nm_flags & PRUS_EOF) {
647 		socantsendmore(so);
648 		unp_shutdown(unp);
649 	}
650 
651 	if (control && error != 0)
652 		unp_dispose(control);
653 release:
654 	unp_reltoken(unp);
655 	wakeup_end_delayed();
656 
657 	if (control)
658 		m_freem(control);
659 	if (m)
660 		m_freem(m);
661 	lwkt_replymsg(&msg->lmsg, error);
662 }
663 
664 /*
665  * MPSAFE
666  */
667 static void
668 uipc_sense(netmsg_t msg)
669 {
670 	struct unpcb *unp;
671 	struct socket *so;
672 	struct stat *sb;
673 	int error;
674 
675 	so = msg->base.nm_so;
676 	sb = msg->sense.nm_stat;
677 
678 	/*
679 	 * so_pcb is only modified with both the global and the unp
680 	 * pool token held.
681 	 */
682 	unp = unp_getsocktoken(so);
683 
684 	if (!UNP_ISATTACHED(unp)) {
685 		error = EINVAL;
686 		goto done;
687 	}
688 
689 	sb->st_blksize = so->so_snd.ssb_hiwat;
690 	sb->st_dev = NOUDEV;
691 	if (unp->unp_ino == 0) {	/* make up a non-zero inode number */
692 		spin_lock(&unp_ino_spin);
693 		unp->unp_ino = unp_ino++;
694 		spin_unlock(&unp_ino_spin);
695 	}
696 	sb->st_ino = unp->unp_ino;
697 	error = 0;
698 done:
699 	unp_reltoken(unp);
700 	lwkt_replymsg(&msg->lmsg, error);
701 }
702 
703 static void
704 uipc_shutdown(netmsg_t msg)
705 {
706 	struct socket *so;
707 	struct unpcb *unp;
708 	int error;
709 
710 	/*
711 	 * so_pcb is only modified with both the global and the unp
712 	 * pool token held.
713 	 */
714 	so = msg->base.nm_so;
715 	unp = unp_getsocktoken(so);
716 
717 	if (UNP_ISATTACHED(unp)) {
718 		socantsendmore(so);
719 		unp_shutdown(unp);
720 		error = 0;
721 	} else {
722 		error = EINVAL;
723 	}
724 
725 	unp_reltoken(unp);
726 	lwkt_replymsg(&msg->lmsg, error);
727 }
728 
729 static void
730 uipc_sockaddr(netmsg_t msg)
731 {
732 	struct unpcb *unp;
733 	int error;
734 
735 	/*
736 	 * so_pcb is only modified with both the global and the unp
737 	 * pool token held.
738 	 */
739 	unp = unp_getsocktoken(msg->base.nm_so);
740 
741 	if (UNP_ISATTACHED(unp)) {
742 		if (unp->unp_addr) {
743 			*msg->sockaddr.nm_nam =
744 				dup_sockaddr((struct sockaddr *)unp->unp_addr);
745 		}
746 		error = 0;
747 	} else {
748 		error = EINVAL;
749 	}
750 
751 	unp_reltoken(unp);
752 	lwkt_replymsg(&msg->lmsg, error);
753 }
754 
755 struct pr_usrreqs uipc_usrreqs = {
756 	.pru_abort = uipc_abort,
757 	.pru_accept = uipc_accept,
758 	.pru_attach = uipc_attach,
759 	.pru_bind = uipc_bind,
760 	.pru_connect = uipc_connect,
761 	.pru_connect2 = uipc_connect2,
762 	.pru_control = pr_generic_notsupp,
763 	.pru_detach = uipc_detach,
764 	.pru_disconnect = uipc_disconnect,
765 	.pru_listen = uipc_listen,
766 	.pru_peeraddr = uipc_peeraddr,
767 	.pru_rcvd = uipc_rcvd,
768 	.pru_rcvoob = pr_generic_notsupp,
769 	.pru_send = uipc_send,
770 	.pru_sense = uipc_sense,
771 	.pru_shutdown = uipc_shutdown,
772 	.pru_sockaddr = uipc_sockaddr,
773 	.pru_sosend = sosend,
774 	.pru_soreceive = soreceive
775 };
776 
777 void
778 uipc_ctloutput(netmsg_t msg)
779 {
780 	struct socket *so;
781 	struct sockopt *sopt;
782 	struct unpcb *unp;
783 	int error = 0;
784 
785 	so = msg->base.nm_so;
786 	sopt = msg->ctloutput.nm_sopt;
787 
788 	lwkt_gettoken(&unp_token);
789 	unp = unp_getsocktoken(so);
790 
791 	if (!UNP_ISATTACHED(unp)) {
792 		error = EINVAL;
793 		goto done;
794 	}
795 
796 	switch (sopt->sopt_dir) {
797 	case SOPT_GET:
798 		switch (sopt->sopt_name) {
799 		case LOCAL_PEERCRED:
800 			if (unp->unp_flags & UNP_HAVEPC)
801 				soopt_from_kbuf(sopt, &unp->unp_peercred,
802 						sizeof(unp->unp_peercred));
803 			else {
804 				if (so->so_type == SOCK_STREAM)
805 					error = ENOTCONN;
806 				else if (so->so_type == SOCK_SEQPACKET)
807 					error = ENOTCONN;
808 				else
809 					error = EINVAL;
810 			}
811 			break;
812 		default:
813 			error = EOPNOTSUPP;
814 			break;
815 		}
816 		break;
817 	case SOPT_SET:
818 	default:
819 		error = EOPNOTSUPP;
820 		break;
821 	}
822 
823 done:
824 	unp_reltoken(unp);
825 	lwkt_reltoken(&unp_token);
826 
827 	lwkt_replymsg(&msg->lmsg, error);
828 }
829 
830 /*
831  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
832  * for stream sockets, although the total for sender and receiver is
833  * actually only PIPSIZ.
834  *
835  * Datagram sockets really use the sendspace as the maximum datagram size,
836  * and don't really want to reserve the sendspace.  Their recvspace should
837  * be large enough for at least one max-size datagram plus address.
838  *
839  * We want the local send/recv space to be significant larger then lo0's
840  * mtu of 16384.
841  */
842 #ifndef PIPSIZ
843 #define	PIPSIZ	57344
844 #endif
845 static u_long	unpst_sendspace = PIPSIZ;
846 static u_long	unpst_recvspace = PIPSIZ;
847 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
848 static u_long	unpdg_recvspace = 4*1024;
849 
850 static int	unp_rights;			/* file descriptors in flight */
851 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin, "unp_spin");
852 
853 SYSCTL_DECL(_net_local_seqpacket);
854 SYSCTL_DECL(_net_local_stream);
855 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
856     &unpst_sendspace, 0, "Size of stream socket send buffer");
857 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
858     &unpst_recvspace, 0, "Size of stream socket receive buffer");
859 
860 SYSCTL_DECL(_net_local_dgram);
861 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
862     &unpdg_sendspace, 0, "Max datagram socket size");
863 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
864     &unpdg_recvspace, 0, "Size of datagram socket receive buffer");
865 
866 SYSCTL_DECL(_net_local);
867 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
868    "File descriptors in flight");
869 
870 static int
871 unp_attach(struct socket *so, struct pru_attach_info *ai)
872 {
873 	struct unpcb *unp;
874 	int error;
875 
876 	lwkt_gettoken(&unp_token);
877 
878 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
879 		switch (so->so_type) {
880 		case SOCK_STREAM:
881 		case SOCK_SEQPACKET:
882 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
883 					  ai->sb_rlimit);
884 			break;
885 
886 		case SOCK_DGRAM:
887 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
888 					  ai->sb_rlimit);
889 			break;
890 
891 		default:
892 			panic("unp_attach");
893 		}
894 		if (error)
895 			goto failed;
896 	}
897 
898 	/*
899 	 * In order to support sendfile we have to set either SSB_STOPSUPP
900 	 * or SSB_PREALLOC.  Unix domain sockets use the SSB_STOP flow
901 	 * control mechanism.
902 	 */
903 	if (so->so_type == SOCK_STREAM) {
904 		atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP);
905 		atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP);
906 	}
907 
908 	unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK);
909 	if (unp == NULL) {
910 		error = ENOBUFS;
911 		goto failed;
912 	}
913 	unp->unp_refcnt = 1;
914 	unp->unp_gencnt = ++unp_gencnt;
915 	unp_count++;
916 	LIST_INIT(&unp->unp_refs);
917 	unp->unp_socket = so;
918 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
919 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
920 			 : &unp_shead, unp, unp_link);
921 	so->so_pcb = (caddr_t)unp;
922 	soreference(so);
923 	error = 0;
924 failed:
925 	lwkt_reltoken(&unp_token);
926 	return error;
927 }
928 
929 static void
930 unp_detach(struct unpcb *unp)
931 {
932 	struct socket *so;
933 
934 	lwkt_gettoken(&unp_token);
935 	lwkt_getpooltoken(unp);
936 
937 	LIST_REMOVE(unp, unp_link);	/* both tokens required */
938 	unp->unp_gencnt = ++unp_gencnt;
939 	--unp_count;
940 	if (unp->unp_vnode) {
941 		unp->unp_vnode->v_socket = NULL;
942 		vrele(unp->unp_vnode);
943 		unp->unp_vnode = NULL;
944 	}
945 	soisdisconnected(unp->unp_socket);
946 	so = unp->unp_socket;
947 	soreference(so);		/* for delayed sorflush */
948 	KKASSERT(so->so_pcb == unp);
949 	so->so_pcb = NULL;		/* both tokens required */
950 	unp->unp_socket = NULL;
951 	sofree(so);		/* remove pcb ref */
952 
953 	if (unp_rights) {
954 		/*
955 		 * Normally the receive buffer is flushed later,
956 		 * in sofree, but if our receive buffer holds references
957 		 * to descriptors that are now garbage, we will dispose
958 		 * of those descriptor references after the garbage collector
959 		 * gets them (resulting in a "panic: closef: count < 0").
960 		 */
961 		sorflush(so);
962 		unp_gc();
963 	}
964 	sofree(so);
965 	lwkt_relpooltoken(unp);
966 	lwkt_reltoken(&unp_token);
967 
968 	KASSERT(unp->unp_conn == NULL, ("unp is still connected"));
969 	KASSERT(LIST_EMPTY(&unp->unp_refs), ("unp still has references"));
970 
971 	if (unp->unp_addr)
972 		kfree(unp->unp_addr, M_SONAME);
973 	kfree(unp, M_UNPCB);
974 }
975 
976 static int
977 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
978 {
979 	struct proc *p = td->td_proc;
980 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
981 	struct vnode *vp;
982 	struct vattr vattr;
983 	int error, namelen;
984 	struct nlookupdata nd;
985 	char buf[SOCK_MAXADDRLEN];
986 
987 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
988 	UNP_ASSERT_TOKEN_HELD(unp);
989 
990 	if (unp->unp_vnode != NULL)
991 		return EINVAL;
992 
993 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
994 	if (namelen <= 0)
995 		return EINVAL;
996 	strncpy(buf, soun->sun_path, namelen);
997 	buf[namelen] = 0;	/* null-terminate the string */
998 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
999 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
1000 	if (error == 0)
1001 		error = nlookup(&nd);
1002 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
1003 		error = EADDRINUSE;
1004 	if (error)
1005 		goto done;
1006 
1007 	VATTR_NULL(&vattr);
1008 	vattr.va_type = VSOCK;
1009 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
1010 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
1011 	if (error == 0) {
1012 		if (unp->unp_vnode == NULL) {
1013 			vp->v_socket = unp->unp_socket;
1014 			unp->unp_vnode = vp;
1015 			unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
1016 			vn_unlock(vp);
1017 		} else {
1018 			vput(vp);		/* late race */
1019 			error = EINVAL;
1020 		}
1021 	}
1022 done:
1023 	nlookup_done(&nd);
1024 	return (error);
1025 }
1026 
1027 static int
1028 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1029 {
1030 	struct unpcb *unp, *unp2;
1031 	int error, flags = 0;
1032 
1033 	lwkt_gettoken(&unp_token);
1034 
1035 	unp = unp_getsocktoken(so);
1036 	if (!UNP_ISATTACHED(unp)) {
1037 		error = EINVAL;
1038 		goto failed;
1039 	}
1040 
1041 	if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) {
1042 		error = EISCONN;
1043 		goto failed;
1044 	}
1045 
1046 	flags = UNP_CONNECTING;
1047 	unp_setflags(unp, flags);
1048 
1049 	error = unp_find_lockref(nam, td, so->so_type, &unp2);
1050 	if (error)
1051 		goto failed;
1052 	/*
1053 	 * NOTE:
1054 	 * unp2 is locked and referenced.
1055 	 */
1056 
1057 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1058 		struct socket *so2, *so3;
1059 		struct unpcb *unp3;
1060 
1061 		so2 = unp2->unp_socket;
1062 		if (!(so2->so_options & SO_ACCEPTCONN) ||
1063 		    (so3 = sonewconn_faddr(so2, 0, NULL,
1064 		     TRUE /* keep ref */)) == NULL) {
1065 			error = ECONNREFUSED;
1066 			goto done;
1067 		}
1068 		/* so3 has a socket reference. */
1069 
1070 		unp3 = unp_getsocktoken(so3);
1071 		if (!UNP_ISATTACHED(unp3)) {
1072 			unp_reltoken(unp3);
1073 			/*
1074 			 * Already aborted; we only need to drop the
1075 			 * socket reference held by sonewconn_faddr().
1076 			 */
1077 			sofree(so3);
1078 			error = ECONNREFUSED;
1079 			goto done;
1080 		}
1081 		unp_reference(unp3);
1082 		/*
1083 		 * NOTE:
1084 		 * unp3 is locked and referenced.
1085 		 */
1086 
1087 		/*
1088 		 * Release so3 socket reference held by sonewconn_faddr().
1089 		 * Since we have referenced unp3, neither unp3 nor so3 will
1090 		 * be destroyed here.
1091 		 */
1092 		sofree(so3);
1093 
1094 		if (unp2->unp_addr != NULL) {
1095 			unp3->unp_addr = (struct sockaddr_un *)
1096 			    dup_sockaddr((struct sockaddr *)unp2->unp_addr);
1097 		}
1098 
1099 		/*
1100 		 * unp_peercred management:
1101 		 *
1102 		 * The connecter's (client's) credentials are copied
1103 		 * from its process structure at the time of connect()
1104 		 * (which is now).
1105 		 */
1106 		cru2x(td->td_proc->p_ucred, &unp3->unp_peercred);
1107 		unp_setflags(unp3, UNP_HAVEPC);
1108 		/*
1109 		 * The receiver's (server's) credentials are copied
1110 		 * from the unp_peercred member of socket on which the
1111 		 * former called listen(); unp_listen() cached that
1112 		 * process's credentials at that time so we can use
1113 		 * them now.
1114 		 */
1115 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1116 		    ("unp_connect: listener without cached peercred"));
1117 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1118 		    sizeof(unp->unp_peercred));
1119 		unp_setflags(unp, UNP_HAVEPC);
1120 
1121 		error = unp_connect_pair(unp, unp3);
1122 		if (error) {
1123 			/* XXX we need a better name */
1124 			soabort_oncpu(so3);
1125 		}
1126 
1127 		/* Done with unp3 */
1128 		unp_free(unp3);
1129 		unp_reltoken(unp3);
1130 	} else {
1131 		error = unp_connect_pair(unp, unp2);
1132 	}
1133 done:
1134 	unp_free(unp2);
1135 	unp_reltoken(unp2);
1136 failed:
1137 	if (flags)
1138 		unp_clrflags(unp, flags);
1139 	unp_reltoken(unp);
1140 
1141 	lwkt_reltoken(&unp_token);
1142 	return (error);
1143 }
1144 
1145 /*
1146  * Connect two unix domain sockets together.
1147  *
1148  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1149  *	 pool token also be held.
1150  */
1151 int
1152 unp_connect2(struct socket *so, struct socket *so2)
1153 {
1154 	struct unpcb *unp, *unp2;
1155 	int error;
1156 
1157 	lwkt_gettoken(&unp_token);
1158 	if (so2->so_type != so->so_type) {
1159 		lwkt_reltoken(&unp_token);
1160 		return (EPROTOTYPE);
1161 	}
1162 	unp = unp_getsocktoken(so);
1163 	unp2 = unp_getsocktoken(so2);
1164 
1165 	if (!UNP_ISATTACHED(unp)) {
1166 		error = EINVAL;
1167 		goto done;
1168 	}
1169 	if (!UNP_ISATTACHED(unp2)) {
1170 		error = ECONNREFUSED;
1171 		goto done;
1172 	}
1173 
1174 	if (unp->unp_conn != NULL) {
1175 		error = EISCONN;
1176 		goto done;
1177 	}
1178 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1179 	    unp2->unp_conn != NULL) {
1180 		error = EISCONN;
1181 		goto done;
1182 	}
1183 
1184 	error = unp_connect_pair(unp, unp2);
1185 done:
1186 	unp_reltoken(unp2);
1187 	unp_reltoken(unp);
1188 	lwkt_reltoken(&unp_token);
1189 	return (error);
1190 }
1191 
1192 /*
1193  * Disconnect a unix domain socket pair.
1194  *
1195  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1196  *	 pool token also be held.
1197  */
1198 static void
1199 unp_disconnect(struct unpcb *unp, int error)
1200 {
1201 	struct socket *so = unp->unp_socket;
1202 	struct unpcb *unp2;
1203 
1204 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
1205 	UNP_ASSERT_TOKEN_HELD(unp);
1206 
1207 	if (error)
1208 		so->so_error = error;
1209 
1210 	while ((unp2 = unp->unp_conn) != NULL) {
1211 		lwkt_getpooltoken(unp2);
1212 		if (unp2 == unp->unp_conn)
1213 			break;
1214 		lwkt_relpooltoken(unp2);
1215 	}
1216 	if (unp2 == NULL)
1217 		return;
1218 	/* unp2 is locked. */
1219 
1220 	KASSERT((unp2->unp_flags & UNP_DROPPED) == 0, ("unp2 was dropped"));
1221 
1222 	unp->unp_conn = NULL;
1223 
1224 	switch (so->so_type) {
1225 	case SOCK_DGRAM:
1226 		LIST_REMOVE(unp, unp_reflink);
1227 		soclrstate(so, SS_ISCONNECTED);
1228 		break;
1229 
1230 	case SOCK_STREAM:
1231 	case SOCK_SEQPACKET:
1232 		/*
1233 		 * Keep a reference before clearing the unp_conn
1234 		 * to avoid racing uipc_detach()/uipc_abort() in
1235 		 * other thread.
1236 		 */
1237 		unp_reference(unp2);
1238 		KASSERT(unp2->unp_conn == unp, ("unp_conn mismatch"));
1239 		unp2->unp_conn = NULL;
1240 
1241 		soisdisconnected(so);
1242 		soisdisconnected(unp2->unp_socket);
1243 
1244 		unp_free(unp2);
1245 		break;
1246 	}
1247 
1248 	lwkt_relpooltoken(unp2);
1249 }
1250 
1251 #ifdef notdef
1252 void
1253 unp_abort(struct unpcb *unp)
1254 {
1255 	lwkt_gettoken(&unp_token);
1256 	unp_free(unp);
1257 	lwkt_reltoken(&unp_token);
1258 }
1259 #endif
1260 
1261 static int
1262 prison_unpcb(struct thread *td, struct unpcb *unp)
1263 {
1264 	struct proc *p;
1265 
1266 	if (td == NULL)
1267 		return (0);
1268 	if ((p = td->td_proc) == NULL)
1269 		return (0);
1270 	if (!p->p_ucred->cr_prison)
1271 		return (0);
1272 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
1273 		return (0);
1274 	return (1);
1275 }
1276 
1277 static int
1278 unp_pcblist(SYSCTL_HANDLER_ARGS)
1279 {
1280 	int error, i, n;
1281 	struct unpcb *unp, **unp_list;
1282 	unp_gen_t gencnt;
1283 	struct unp_head *head;
1284 
1285 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1286 
1287 	KKASSERT(curproc != NULL);
1288 
1289 	/*
1290 	 * The process of preparing the PCB list is too time-consuming and
1291 	 * resource-intensive to repeat twice on every request.
1292 	 */
1293 	if (req->oldptr == NULL) {
1294 		n = unp_count;
1295 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
1296 		return 0;
1297 	}
1298 
1299 	if (req->newptr != NULL)
1300 		return EPERM;
1301 
1302 	lwkt_gettoken(&unp_token);
1303 
1304 	/*
1305 	 * OK, now we're committed to doing something.
1306 	 */
1307 	gencnt = unp_gencnt;
1308 	n = unp_count;
1309 
1310 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1311 
1312 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1313 	     unp = LIST_NEXT(unp, unp_link)) {
1314 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
1315 			unp_list[i++] = unp;
1316 	}
1317 	n = i;			/* in case we lost some during malloc */
1318 
1319 	error = 0;
1320 	for (i = 0; i < n; i++) {
1321 		unp = unp_list[i];
1322 		if (unp->unp_gencnt <= gencnt) {
1323 			struct xunpcb xu;
1324 			xu.xu_len = sizeof xu;
1325 			xu.xu_unpp = unp;
1326 			/*
1327 			 * XXX - need more locking here to protect against
1328 			 * connect/disconnect races for SMP.
1329 			 */
1330 			if (unp->unp_addr)
1331 				bcopy(unp->unp_addr, &xu.xu_addr,
1332 				      unp->unp_addr->sun_len);
1333 			if (unp->unp_conn && unp->unp_conn->unp_addr)
1334 				bcopy(unp->unp_conn->unp_addr,
1335 				      &xu.xu_caddr,
1336 				      unp->unp_conn->unp_addr->sun_len);
1337 			bcopy(unp, &xu.xu_unp, sizeof *unp);
1338 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
1339 			error = SYSCTL_OUT(req, &xu, sizeof xu);
1340 		}
1341 	}
1342 	lwkt_reltoken(&unp_token);
1343 	kfree(unp_list, M_TEMP);
1344 
1345 	return error;
1346 }
1347 
1348 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1349 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1350 	    "List of active local datagram sockets");
1351 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1352 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1353 	    "List of active local stream sockets");
1354 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
1355 	    (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1356 	    "List of active local seqpacket stream sockets");
1357 
1358 static void
1359 unp_shutdown(struct unpcb *unp)
1360 {
1361 	struct socket *so;
1362 
1363 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1364 	     unp->unp_socket->so_type == SOCK_SEQPACKET) &&
1365 	    unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
1366 		socantrcvmore(so);
1367 	}
1368 }
1369 
1370 #ifdef notdef
1371 void
1372 unp_drain(void)
1373 {
1374 	lwkt_gettoken(&unp_token);
1375 	lwkt_reltoken(&unp_token);
1376 }
1377 #endif
1378 
1379 int
1380 unp_externalize(struct mbuf *rights)
1381 {
1382 	struct thread *td = curthread;
1383 	struct proc *p = td->td_proc;		/* XXX */
1384 	struct lwp *lp = td->td_lwp;
1385 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
1386 	int *fdp;
1387 	int i;
1388 	struct file **rp;
1389 	struct file *fp;
1390 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
1391 		/ sizeof (struct file *);
1392 	int f;
1393 
1394 	lwkt_gettoken(&unp_token);
1395 
1396 	/*
1397 	 * if the new FD's will not fit, then we free them all
1398 	 */
1399 	if (!fdavail(p, newfds)) {
1400 		rp = (struct file **)CMSG_DATA(cm);
1401 		for (i = 0; i < newfds; i++) {
1402 			fp = *rp;
1403 			/*
1404 			 * zero the pointer before calling unp_discard,
1405 			 * since it may end up in unp_gc()..
1406 			 */
1407 			*rp++ = NULL;
1408 			unp_discard(fp, NULL);
1409 		}
1410 		lwkt_reltoken(&unp_token);
1411 		return (EMSGSIZE);
1412 	}
1413 
1414 	/*
1415 	 * now change each pointer to an fd in the global table to
1416 	 * an integer that is the index to the local fd table entry
1417 	 * that we set up to point to the global one we are transferring.
1418 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1419 	 * then do it in forward order. In that case, an integer will
1420 	 * always come in the same place or before its corresponding
1421 	 * struct file pointer.
1422 	 * If sizeof (struct file *) is smaller than sizeof int, then
1423 	 * do it in reverse order.
1424 	 */
1425 	if (sizeof (struct file *) >= sizeof (int)) {
1426 		fdp = (int *)CMSG_DATA(cm);
1427 		rp = (struct file **)CMSG_DATA(cm);
1428 		for (i = 0; i < newfds; i++) {
1429 			if (fdalloc(p, 0, &f))
1430 				panic("unp_externalize");
1431 			fp = *rp++;
1432 			unp_fp_externalize(lp, fp, f);
1433 			*fdp++ = f;
1434 		}
1435 	} else {
1436 		fdp = (int *)CMSG_DATA(cm) + newfds - 1;
1437 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
1438 		for (i = 0; i < newfds; i++) {
1439 			if (fdalloc(p, 0, &f))
1440 				panic("unp_externalize");
1441 			fp = *rp--;
1442 			unp_fp_externalize(lp, fp, f);
1443 			*fdp-- = f;
1444 		}
1445 	}
1446 
1447 	/*
1448 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1449 	 * differs.
1450 	 */
1451 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1452 	rights->m_len = cm->cmsg_len;
1453 
1454 	lwkt_reltoken(&unp_token);
1455 	return (0);
1456 }
1457 
1458 static void
1459 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd)
1460 {
1461 	struct file *fx;
1462 	int error;
1463 
1464 	lwkt_gettoken(&unp_token);
1465 
1466 	if (lp) {
1467 		KKASSERT(fd >= 0);
1468 		if (fp->f_flag & FREVOKED) {
1469 			kprintf("Warning: revoked fp exiting unix socket\n");
1470 			fx = NULL;
1471 			error = falloc(lp, &fx, NULL);
1472 			if (error == 0)
1473 				fsetfd(lp->lwp_proc->p_fd, fx, fd);
1474 			else
1475 				fsetfd(lp->lwp_proc->p_fd, NULL, fd);
1476 			fdrop(fx);
1477 		} else {
1478 			fsetfd(lp->lwp_proc->p_fd, fp, fd);
1479 		}
1480 	}
1481 	spin_lock(&unp_spin);
1482 	fp->f_msgcount--;
1483 	unp_rights--;
1484 	spin_unlock(&unp_spin);
1485 	fdrop(fp);
1486 
1487 	lwkt_reltoken(&unp_token);
1488 }
1489 
1490 
1491 void
1492 unp_init(void)
1493 {
1494 	LIST_INIT(&unp_dhead);
1495 	LIST_INIT(&unp_shead);
1496 	spin_init(&unp_spin, "unpinit");
1497 }
1498 
1499 static int
1500 unp_internalize(struct mbuf *control, struct thread *td)
1501 {
1502 	struct proc *p = td->td_proc;
1503 	struct filedesc *fdescp;
1504 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1505 	struct file **rp;
1506 	struct file *fp;
1507 	int i, fd, *fdp;
1508 	struct cmsgcred *cmcred;
1509 	int oldfds;
1510 	u_int newlen;
1511 	int error;
1512 
1513 	KKASSERT(p);
1514 	lwkt_gettoken(&unp_token);
1515 
1516 	fdescp = p->p_fd;
1517 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1518 	    cm->cmsg_level != SOL_SOCKET ||
1519 	    CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1520 		error = EINVAL;
1521 		goto done;
1522 	}
1523 
1524 	/*
1525 	 * Fill in credential information.
1526 	 */
1527 	if (cm->cmsg_type == SCM_CREDS) {
1528 		cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1529 		cmcred->cmcred_pid = p->p_pid;
1530 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1531 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1532 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1533 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1534 							CMGROUP_MAX);
1535 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1536 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1537 		error = 0;
1538 		goto done;
1539 	}
1540 
1541 	/*
1542 	 * cmsghdr may not be aligned, do not allow calculation(s) to
1543 	 * go negative.
1544 	 */
1545 	if (cm->cmsg_len < CMSG_LEN(0)) {
1546 		error = EINVAL;
1547 		goto done;
1548 	}
1549 
1550 	oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1551 
1552 	/*
1553 	 * check that all the FDs passed in refer to legal OPEN files
1554 	 * If not, reject the entire operation.
1555 	 */
1556 	fdp = (int *)CMSG_DATA(cm);
1557 	for (i = 0; i < oldfds; i++) {
1558 		fd = *fdp++;
1559 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1560 		    fdescp->fd_files[fd].fp == NULL) {
1561 			error = EBADF;
1562 			goto done;
1563 		}
1564 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) {
1565 			error = EOPNOTSUPP;
1566 			goto done;
1567 		}
1568 	}
1569 	/*
1570 	 * Now replace the integer FDs with pointers to
1571 	 * the associated global file table entry..
1572 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1573 	 * enough, return E2BIG.
1574 	 */
1575 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1576 	if (newlen > MCLBYTES) {
1577 		error = E2BIG;
1578 		goto done;
1579 	}
1580 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1581 		if (control->m_flags & M_EXT) {
1582 			error = E2BIG;
1583 			goto done;
1584 		}
1585 		MCLGET(control, M_WAITOK);
1586 		if (!(control->m_flags & M_EXT)) {
1587 			error = ENOBUFS;
1588 			goto done;
1589 		}
1590 
1591 		/* copy the data to the cluster */
1592 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1593 		cm = mtod(control, struct cmsghdr *);
1594 	}
1595 
1596 	/*
1597 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1598 	 * differs.
1599 	 */
1600 	cm->cmsg_len = newlen;
1601 	control->m_len = CMSG_ALIGN(newlen);
1602 
1603 	/*
1604 	 * Transform the file descriptors into struct file pointers.
1605 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1606 	 * then do it in reverse order so that the int won't get until
1607 	 * we're done.
1608 	 * If sizeof (struct file *) is smaller than sizeof int, then
1609 	 * do it in forward order.
1610 	 */
1611 	if (sizeof (struct file *) >= sizeof (int)) {
1612 		fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1613 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1614 		for (i = 0; i < oldfds; i++) {
1615 			fp = fdescp->fd_files[*fdp--].fp;
1616 			*rp-- = fp;
1617 			fhold(fp);
1618 			spin_lock(&unp_spin);
1619 			fp->f_msgcount++;
1620 			unp_rights++;
1621 			spin_unlock(&unp_spin);
1622 		}
1623 	} else {
1624 		fdp = (int *)CMSG_DATA(cm);
1625 		rp = (struct file **)CMSG_DATA(cm);
1626 		for (i = 0; i < oldfds; i++) {
1627 			fp = fdescp->fd_files[*fdp++].fp;
1628 			*rp++ = fp;
1629 			fhold(fp);
1630 			spin_lock(&unp_spin);
1631 			fp->f_msgcount++;
1632 			unp_rights++;
1633 			spin_unlock(&unp_spin);
1634 		}
1635 	}
1636 	error = 0;
1637 done:
1638 	lwkt_reltoken(&unp_token);
1639 	return error;
1640 }
1641 
1642 /*
1643  * Garbage collect in-transit file descriptors that get lost due to
1644  * loops (i.e. when a socket is sent to another process over itself,
1645  * and more complex situations).
1646  *
1647  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1648  */
1649 
1650 struct unp_gc_info {
1651 	struct file **extra_ref;
1652 	struct file *locked_fp;
1653 	int defer;
1654 	int index;
1655 	int maxindex;
1656 };
1657 
1658 static void
1659 unp_gc(void)
1660 {
1661 	struct unp_gc_info info;
1662 	static boolean_t unp_gcing;
1663 	struct file **fpp;
1664 	int i;
1665 
1666 	/*
1667 	 * Only one gc can be in-progress at any given moment
1668 	 */
1669 	spin_lock(&unp_spin);
1670 	if (unp_gcing) {
1671 		spin_unlock(&unp_spin);
1672 		return;
1673 	}
1674 	unp_gcing = TRUE;
1675 	spin_unlock(&unp_spin);
1676 
1677 	lwkt_gettoken(&unp_token);
1678 
1679 	/*
1680 	 * Before going through all this, set all FDs to be NOT defered
1681 	 * and NOT externally accessible (not marked).  During the scan
1682 	 * a fd can be marked externally accessible but we may or may not
1683 	 * be able to immediately process it (controlled by FDEFER).
1684 	 *
1685 	 * If we loop sleep a bit.  The complexity of the topology can cause
1686 	 * multiple loops.  Also failure to acquire the socket's so_rcv
1687 	 * token can cause us to loop.
1688 	 */
1689 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1690 	do {
1691 		info.defer = 0;
1692 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1693 		if (info.defer)
1694 			tsleep(&info, 0, "gcagain", 1);
1695 	} while (info.defer);
1696 
1697 	/*
1698 	 * We grab an extra reference to each of the file table entries
1699 	 * that are not otherwise accessible and then free the rights
1700 	 * that are stored in messages on them.
1701 	 *
1702 	 * The bug in the orginal code is a little tricky, so I'll describe
1703 	 * what's wrong with it here.
1704 	 *
1705 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1706 	 * times -- consider the case of sockets A and B that contain
1707 	 * references to each other.  On a last close of some other socket,
1708 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1709 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1710 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1711 	 * results in the following chain.  Closef calls soo_close, which
1712 	 * calls soclose.   Soclose calls first (through the switch
1713 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1714 	 * returns because the previous instance had set unp_gcing, and
1715 	 * we return all the way back to soclose, which marks the socket
1716 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1717 	 * to free up the rights that are queued in messages on the socket A,
1718 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1719 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1720 	 * instance of unp_discard just calls closef on B.
1721 	 *
1722 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1723 	 * which results in another closef on A.  Unfortunately, A is already
1724 	 * being closed, and the descriptor has already been marked with
1725 	 * SS_NOFDREF, and soclose panics at this point.
1726 	 *
1727 	 * Here, we first take an extra reference to each inaccessible
1728 	 * descriptor.  Then, we call sorflush ourself, since we know
1729 	 * it is a Unix domain socket anyhow.  After we destroy all the
1730 	 * rights carried in messages, we do a last closef to get rid
1731 	 * of our extra reference.  This is the last close, and the
1732 	 * unp_detach etc will shut down the socket.
1733 	 *
1734 	 * 91/09/19, bsy@cs.cmu.edu
1735 	 */
1736 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1737 	info.maxindex = 256;
1738 
1739 	do {
1740 		/*
1741 		 * Look for matches
1742 		 */
1743 		info.index = 0;
1744 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1745 
1746 		/*
1747 		 * For each FD on our hit list, do the following two things
1748 		 */
1749 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1750 			struct file *tfp = *fpp;
1751 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1752 				sorflush((struct socket *)(tfp->f_data));
1753 		}
1754 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1755 			closef(*fpp, NULL);
1756 	} while (info.index == info.maxindex);
1757 
1758 	lwkt_reltoken(&unp_token);
1759 
1760 	kfree((caddr_t)info.extra_ref, M_FILE);
1761 	unp_gcing = FALSE;
1762 }
1763 
1764 /*
1765  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1766  */
1767 static int
1768 unp_gc_checkrefs(struct file *fp, void *data)
1769 {
1770 	struct unp_gc_info *info = data;
1771 
1772 	if (fp->f_count == 0)
1773 		return(0);
1774 	if (info->index == info->maxindex)
1775 		return(-1);
1776 
1777 	/*
1778 	 * If all refs are from msgs, and it's not marked accessible
1779 	 * then it must be referenced from some unreachable cycle
1780 	 * of (shut-down) FDs, so include it in our
1781 	 * list of FDs to remove
1782 	 */
1783 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1784 		info->extra_ref[info->index++] = fp;
1785 		fhold(fp);
1786 	}
1787 	return(0);
1788 }
1789 
1790 /*
1791  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1792  */
1793 static int
1794 unp_gc_clearmarks(struct file *fp, void *data __unused)
1795 {
1796 	atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1797 	return(0);
1798 }
1799 
1800 /*
1801  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1802  */
1803 static int
1804 unp_gc_checkmarks(struct file *fp, void *data)
1805 {
1806 	struct unp_gc_info *info = data;
1807 	struct socket *so;
1808 
1809 	/*
1810 	 * If the file is not open, skip it.  Make sure it isn't marked
1811 	 * defered or we could loop forever, in case we somehow race
1812 	 * something.
1813 	 */
1814 	if (fp->f_count == 0) {
1815 		if (fp->f_flag & FDEFER)
1816 			atomic_clear_int(&fp->f_flag, FDEFER);
1817 		return(0);
1818 	}
1819 	/*
1820 	 * If we already marked it as 'defer'  in a
1821 	 * previous pass, then try process it this time
1822 	 * and un-mark it
1823 	 */
1824 	if (fp->f_flag & FDEFER) {
1825 		atomic_clear_int(&fp->f_flag, FDEFER);
1826 	} else {
1827 		/*
1828 		 * if it's not defered, then check if it's
1829 		 * already marked.. if so skip it
1830 		 */
1831 		if (fp->f_flag & FMARK)
1832 			return(0);
1833 		/*
1834 		 * If all references are from messages
1835 		 * in transit, then skip it. it's not
1836 		 * externally accessible.
1837 		 */
1838 		if (fp->f_count == fp->f_msgcount)
1839 			return(0);
1840 		/*
1841 		 * If it got this far then it must be
1842 		 * externally accessible.
1843 		 */
1844 		atomic_set_int(&fp->f_flag, FMARK);
1845 	}
1846 
1847 	/*
1848 	 * either it was defered, or it is externally
1849 	 * accessible and not already marked so.
1850 	 * Now check if it is possibly one of OUR sockets.
1851 	 */
1852 	if (fp->f_type != DTYPE_SOCKET ||
1853 	    (so = (struct socket *)fp->f_data) == NULL) {
1854 		return(0);
1855 	}
1856 	if (so->so_proto->pr_domain != &localdomain ||
1857 	    !(so->so_proto->pr_flags & PR_RIGHTS)) {
1858 		return(0);
1859 	}
1860 
1861 	/*
1862 	 * So, Ok, it's one of our sockets and it IS externally accessible
1863 	 * (or was defered).  Now we look to see if we hold any file
1864 	 * descriptors in its message buffers.  Follow those links and mark
1865 	 * them as accessible too.
1866 	 *
1867 	 * We are holding multiple spinlocks here, if we cannot get the
1868 	 * token non-blocking defer until the next loop.
1869 	 */
1870 	info->locked_fp = fp;
1871 	if (lwkt_trytoken(&so->so_rcv.ssb_token)) {
1872 		unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1873 		lwkt_reltoken(&so->so_rcv.ssb_token);
1874 	} else {
1875 		atomic_set_int(&fp->f_flag, FDEFER);
1876 		++info->defer;
1877 	}
1878 	return (0);
1879 }
1880 
1881 /*
1882  * Scan all unix domain sockets and replace any revoked file pointers
1883  * found with the dummy file pointer fx.  We don't worry about races
1884  * against file pointers being read out as those are handled in the
1885  * externalize code.
1886  */
1887 
1888 #define REVOKE_GC_MAXFILES	32
1889 
1890 struct unp_revoke_gc_info {
1891 	struct file	*fx;
1892 	struct file	*fary[REVOKE_GC_MAXFILES];
1893 	int		fcount;
1894 };
1895 
1896 void
1897 unp_revoke_gc(struct file *fx)
1898 {
1899 	struct unp_revoke_gc_info info;
1900 	int i;
1901 
1902 	lwkt_gettoken(&unp_token);
1903 	info.fx = fx;
1904 	do {
1905 		info.fcount = 0;
1906 		allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1907 		for (i = 0; i < info.fcount; ++i)
1908 			unp_fp_externalize(NULL, info.fary[i], -1);
1909 	} while (info.fcount == REVOKE_GC_MAXFILES);
1910 	lwkt_reltoken(&unp_token);
1911 }
1912 
1913 /*
1914  * Check for and replace revoked descriptors.
1915  *
1916  * WARNING:  This routine is not allowed to block.
1917  */
1918 static int
1919 unp_revoke_gc_check(struct file *fps, void *vinfo)
1920 {
1921 	struct unp_revoke_gc_info *info = vinfo;
1922 	struct file *fp;
1923 	struct socket *so;
1924 	struct mbuf *m0;
1925 	struct mbuf *m;
1926 	struct file **rp;
1927 	struct cmsghdr *cm;
1928 	int i;
1929 	int qfds;
1930 
1931 	/*
1932 	 * Is this a unix domain socket with rights-passing abilities?
1933 	 */
1934 	if (fps->f_type != DTYPE_SOCKET)
1935 		return (0);
1936 	if ((so = (struct socket *)fps->f_data) == NULL)
1937 		return(0);
1938 	if (so->so_proto->pr_domain != &localdomain)
1939 		return(0);
1940 	if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1941 		return(0);
1942 
1943 	/*
1944 	 * Scan the mbufs for control messages and replace any revoked
1945 	 * descriptors we find.
1946 	 */
1947 	lwkt_gettoken(&so->so_rcv.ssb_token);
1948 	m0 = so->so_rcv.ssb_mb;
1949 	while (m0) {
1950 		for (m = m0; m; m = m->m_next) {
1951 			if (m->m_type != MT_CONTROL)
1952 				continue;
1953 			if (m->m_len < sizeof(*cm))
1954 				continue;
1955 			cm = mtod(m, struct cmsghdr *);
1956 			if (cm->cmsg_level != SOL_SOCKET ||
1957 			    cm->cmsg_type != SCM_RIGHTS) {
1958 				continue;
1959 			}
1960 			qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1961 			rp = (struct file **)CMSG_DATA(cm);
1962 			for (i = 0; i < qfds; i++) {
1963 				fp = rp[i];
1964 				if (fp->f_flag & FREVOKED) {
1965 					kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1966 					fhold(info->fx);
1967 					info->fx->f_msgcount++;
1968 					unp_rights++;
1969 					rp[i] = info->fx;
1970 					info->fary[info->fcount++] = fp;
1971 				}
1972 				if (info->fcount == REVOKE_GC_MAXFILES)
1973 					break;
1974 			}
1975 			if (info->fcount == REVOKE_GC_MAXFILES)
1976 				break;
1977 		}
1978 		m0 = m0->m_nextpkt;
1979 		if (info->fcount == REVOKE_GC_MAXFILES)
1980 			break;
1981 	}
1982 	lwkt_reltoken(&so->so_rcv.ssb_token);
1983 
1984 	/*
1985 	 * Stop the scan if we filled up our array.
1986 	 */
1987 	if (info->fcount == REVOKE_GC_MAXFILES)
1988 		return(-1);
1989 	return(0);
1990 }
1991 
1992 /*
1993  * Dispose of the fp's stored in a mbuf.
1994  *
1995  * The dds loop can cause additional fps to be entered onto the
1996  * list while it is running, flattening out the operation and avoiding
1997  * a deep kernel stack recursion.
1998  */
1999 void
2000 unp_dispose(struct mbuf *m)
2001 {
2002 	unp_defdiscard_t dds;
2003 
2004 	lwkt_gettoken(&unp_token);
2005 	++unp_defdiscard_nest;
2006 	if (m) {
2007 		unp_scan(m, unp_discard, NULL);
2008 	}
2009 	if (unp_defdiscard_nest == 1) {
2010 		while ((dds = unp_defdiscard_base) != NULL) {
2011 			unp_defdiscard_base = dds->next;
2012 			closef(dds->fp, NULL);
2013 			kfree(dds, M_UNPCB);
2014 		}
2015 	}
2016 	--unp_defdiscard_nest;
2017 	lwkt_reltoken(&unp_token);
2018 }
2019 
2020 static int
2021 unp_listen(struct unpcb *unp, struct thread *td)
2022 {
2023 	struct proc *p = td->td_proc;
2024 
2025 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2026 	UNP_ASSERT_TOKEN_HELD(unp);
2027 
2028 	KKASSERT(p);
2029 	cru2x(p->p_ucred, &unp->unp_peercred);
2030 	unp_setflags(unp, UNP_HAVEPCCACHED);
2031 	return (0);
2032 }
2033 
2034 static void
2035 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
2036 {
2037 	struct mbuf *m;
2038 	struct file **rp;
2039 	struct cmsghdr *cm;
2040 	int i;
2041 	int qfds;
2042 
2043 	while (m0) {
2044 		for (m = m0; m; m = m->m_next) {
2045 			if (m->m_type == MT_CONTROL &&
2046 			    m->m_len >= sizeof(*cm)) {
2047 				cm = mtod(m, struct cmsghdr *);
2048 				if (cm->cmsg_level != SOL_SOCKET ||
2049 				    cm->cmsg_type != SCM_RIGHTS)
2050 					continue;
2051 				qfds = (cm->cmsg_len - CMSG_LEN(0)) /
2052 					sizeof(void *);
2053 				rp = (struct file **)CMSG_DATA(cm);
2054 				for (i = 0; i < qfds; i++)
2055 					(*op)(*rp++, data);
2056 				break;		/* XXX, but saves time */
2057 			}
2058 		}
2059 		m0 = m0->m_nextpkt;
2060 	}
2061 }
2062 
2063 /*
2064  * Mark visibility.  info->defer is recalculated on every pass.
2065  */
2066 static void
2067 unp_mark(struct file *fp, void *data)
2068 {
2069 	struct unp_gc_info *info = data;
2070 
2071 	if ((fp->f_flag & FMARK) == 0) {
2072 		++info->defer;
2073 		atomic_set_int(&fp->f_flag, FMARK | FDEFER);
2074 	} else if (fp->f_flag & FDEFER) {
2075 		++info->defer;
2076 	}
2077 }
2078 
2079 /*
2080  * Discard a fp previously held in a unix domain socket mbuf.  To
2081  * avoid blowing out the kernel stack due to contrived chain-reactions
2082  * we may have to defer the operation to a higher procedural level.
2083  *
2084  * Caller holds unp_token
2085  */
2086 static void
2087 unp_discard(struct file *fp, void *data __unused)
2088 {
2089 	unp_defdiscard_t dds;
2090 
2091 	spin_lock(&unp_spin);
2092 	fp->f_msgcount--;
2093 	unp_rights--;
2094 	spin_unlock(&unp_spin);
2095 
2096 	if (unp_defdiscard_nest) {
2097 		dds = kmalloc(sizeof(*dds), M_UNPCB, M_WAITOK|M_ZERO);
2098 		dds->fp = fp;
2099 		dds->next = unp_defdiscard_base;
2100 		unp_defdiscard_base = dds;
2101 	} else {
2102 		closef(fp, NULL);
2103 	}
2104 }
2105 
2106 static int
2107 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type,
2108     struct unpcb **unp_ret)
2109 {
2110 	struct proc *p = td->td_proc;
2111 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
2112 	struct vnode *vp = NULL;
2113 	struct socket *so;
2114 	struct unpcb *unp;
2115 	int error, len;
2116 	struct nlookupdata nd;
2117 	char buf[SOCK_MAXADDRLEN];
2118 
2119 	*unp_ret = NULL;
2120 
2121 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2122 	if (len <= 0) {
2123 		error = EINVAL;
2124 		goto failed;
2125 	}
2126 	strncpy(buf, soun->sun_path, len);
2127 	buf[len] = 0;
2128 
2129 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
2130 	if (error == 0)
2131 		error = nlookup(&nd);
2132 	if (error == 0)
2133 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2134 	nlookup_done(&nd);
2135 	if (error) {
2136 		vp = NULL;
2137 		goto failed;
2138 	}
2139 
2140 	if (vp->v_type != VSOCK) {
2141 		error = ENOTSOCK;
2142 		goto failed;
2143 	}
2144 	error = VOP_EACCESS(vp, VWRITE, p->p_ucred);
2145 	if (error)
2146 		goto failed;
2147 	so = vp->v_socket;
2148 	if (so == NULL) {
2149 		error = ECONNREFUSED;
2150 		goto failed;
2151 	}
2152 	if (so->so_type != type) {
2153 		error = EPROTOTYPE;
2154 		goto failed;
2155 	}
2156 
2157 	/* Lock this unp. */
2158 	unp = unp_getsocktoken(so);
2159 	if (!UNP_ISATTACHED(unp)) {
2160 		unp_reltoken(unp);
2161 		error = ECONNREFUSED;
2162 		goto failed;
2163 	}
2164 	/* And keep this unp referenced. */
2165 	unp_reference(unp);
2166 
2167 	/* Done! */
2168 	*unp_ret = unp;
2169 	error = 0;
2170 failed:
2171 	if (vp != NULL)
2172 		vput(vp);
2173 	return error;
2174 }
2175 
2176 static int
2177 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2)
2178 {
2179 	struct socket *so = unp->unp_socket;
2180 	struct socket *so2 = unp2->unp_socket;
2181 
2182 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2183 	UNP_ASSERT_TOKEN_HELD(unp);
2184 	UNP_ASSERT_TOKEN_HELD(unp2);
2185 
2186 	KASSERT(so->so_type == so2->so_type,
2187 	    ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type));
2188 
2189 	if (!UNP_ISATTACHED(unp))
2190 		return EINVAL;
2191 	if (!UNP_ISATTACHED(unp2))
2192 		return ECONNREFUSED;
2193 
2194 	KASSERT(unp->unp_conn == NULL, ("unp is already connected"));
2195 	unp->unp_conn = unp2;
2196 
2197 	switch (so->so_type) {
2198 	case SOCK_DGRAM:
2199 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2200 		soisconnected(so);
2201 		break;
2202 
2203 	case SOCK_STREAM:
2204 	case SOCK_SEQPACKET:
2205 		KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected"));
2206 		unp2->unp_conn = unp;
2207 		soisconnected(so);
2208 		soisconnected(so2);
2209 		break;
2210 
2211 	default:
2212 		panic("unp_connect_pair: unknown socket type %d", so->so_type);
2213 	}
2214 	return 0;
2215 }
2216 
2217 static void
2218 unp_drop(struct unpcb *unp, int error)
2219 {
2220 	struct unpcb *unp2;
2221 
2222 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2223 	UNP_ASSERT_TOKEN_HELD(unp);
2224 	KASSERT(unp->unp_flags & UNP_DETACHED, ("unp is not detached"));
2225 
2226 	unp_disconnect(unp, error);
2227 
2228 	while ((unp2 = LIST_FIRST(&unp->unp_refs)) != NULL) {
2229 		lwkt_getpooltoken(unp2);
2230 		unp_disconnect(unp2, ECONNRESET);
2231 		lwkt_relpooltoken(unp2);
2232 	}
2233 	unp_setflags(unp, UNP_DROPPED);
2234 }
2235