xref: /netbsd-src/sys/kern/uipc_usrreq.c (revision 6cf6fe02a981b55727c49c3d37b0d8191a98c0ee)
1 /*	$NetBSD: uipc_usrreq.c,v 1.171 2014/09/05 09:20:59 matt Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000, 2004, 2008, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)uipc_usrreq.c	8.9 (Berkeley) 5/14/95
62  */
63 
64 /*
65  * Copyright (c) 1997 Christopher G. Demetriou.  All rights reserved.
66  *
67  * Redistribution and use in source and binary forms, with or without
68  * modification, are permitted provided that the following conditions
69  * are met:
70  * 1. Redistributions of source code must retain the above copyright
71  *    notice, this list of conditions and the following disclaimer.
72  * 2. Redistributions in binary form must reproduce the above copyright
73  *    notice, this list of conditions and the following disclaimer in the
74  *    documentation and/or other materials provided with the distribution.
75  * 3. All advertising materials mentioning features or use of this software
76  *    must display the following acknowledgement:
77  *	This product includes software developed by the University of
78  *	California, Berkeley and its contributors.
79  * 4. Neither the name of the University nor the names of its contributors
80  *    may be used to endorse or promote products derived from this software
81  *    without specific prior written permission.
82  *
83  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
84  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
85  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
86  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
87  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
88  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
89  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
90  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
91  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
92  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
93  * SUCH DAMAGE.
94  *
95  *	@(#)uipc_usrreq.c	8.9 (Berkeley) 5/14/95
96  */
97 
98 #include <sys/cdefs.h>
99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.171 2014/09/05 09:20:59 matt Exp $");
100 
101 #include <sys/param.h>
102 #include <sys/systm.h>
103 #include <sys/proc.h>
104 #include <sys/filedesc.h>
105 #include <sys/domain.h>
106 #include <sys/protosw.h>
107 #include <sys/socket.h>
108 #include <sys/socketvar.h>
109 #include <sys/unpcb.h>
110 #include <sys/un.h>
111 #include <sys/namei.h>
112 #include <sys/vnode.h>
113 #include <sys/file.h>
114 #include <sys/stat.h>
115 #include <sys/mbuf.h>
116 #include <sys/kauth.h>
117 #include <sys/kmem.h>
118 #include <sys/atomic.h>
119 #include <sys/uidinfo.h>
120 #include <sys/kernel.h>
121 #include <sys/kthread.h>
122 
123 /*
124  * Unix communications domain.
125  *
126  * TODO:
127  *	RDM
128  *	rethink name space problems
129  *	need a proper out-of-band
130  *
131  * Notes on locking:
132  *
133  * The generic rules noted in uipc_socket2.c apply.  In addition:
134  *
135  * o We have a global lock, uipc_lock.
136  *
137  * o All datagram sockets are locked by uipc_lock.
138  *
139  * o For stream socketpairs, the two endpoints are created sharing the same
140  *   independent lock.  Sockets presented to PRU_CONNECT2 must already have
141  *   matching locks.
142  *
143  * o Stream sockets created via socket() start life with their own
144  *   independent lock.
145  *
146  * o Stream connections to a named endpoint are slightly more complicated.
147  *   Sockets that have called listen() have their lock pointer mutated to
148  *   the global uipc_lock.  When establishing a connection, the connecting
149  *   socket also has its lock mutated to uipc_lock, which matches the head
150  *   (listening socket).  We create a new socket for accept() to return, and
151  *   that also shares the head's lock.  Until the connection is completely
152  *   done on both ends, all three sockets are locked by uipc_lock.  Once the
153  *   connection is complete, the association with the head's lock is broken.
154  *   The connecting socket and the socket returned from accept() have their
155  *   lock pointers mutated away from uipc_lock, and back to the connecting
156  *   socket's original, independent lock.  The head continues to be locked
157  *   by uipc_lock.
158  *
159  * o If uipc_lock is determined to be a significant source of contention,
160  *   it could easily be hashed out.  It is difficult to simply make it an
161  *   independent lock because of visibility / garbage collection issues:
162  *   if a socket has been associated with a lock at any point, that lock
163  *   must remain valid until the socket is no longer visible in the system.
164  *   The lock must not be freed or otherwise destroyed until any sockets
165  *   that had referenced it have also been destroyed.
166  */
167 const struct sockaddr_un sun_noname = {
168 	.sun_len = offsetof(struct sockaddr_un, sun_path),
169 	.sun_family = AF_LOCAL,
170 };
171 ino_t	unp_ino;			/* prototype for fake inode numbers */
172 
173 static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *);
174 static void   unp_discard_later(file_t *);
175 static void   unp_discard_now(file_t *);
176 static void   unp_disconnect1(struct unpcb *);
177 static bool   unp_drop(struct unpcb *, int);
178 static int    unp_internalize(struct mbuf **);
179 static void   unp_mark(file_t *);
180 static void   unp_scan(struct mbuf *, void (*)(file_t *), int);
181 static void   unp_shutdown1(struct unpcb *);
182 static void   unp_thread(void *);
183 static void   unp_thread_kick(void);
184 
185 static kmutex_t *uipc_lock;
186 
187 static kcondvar_t unp_thread_cv;
188 static lwp_t *unp_thread_lwp;
189 static SLIST_HEAD(,file) unp_thread_discard;
190 static int unp_defer;
191 
192 /*
193  * Initialize Unix protocols.
194  */
195 void
196 uipc_init(void)
197 {
198 	int error;
199 
200 	uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
201 	cv_init(&unp_thread_cv, "unpgc");
202 
203 	error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread,
204 	    NULL, &unp_thread_lwp, "unpgc");
205 	if (error != 0)
206 		panic("uipc_init %d", error);
207 }
208 
209 /*
210  * A connection succeeded: disassociate both endpoints from the head's
211  * lock, and make them share their own lock.  There is a race here: for
212  * a very brief time one endpoint will be locked by a different lock
213  * than the other end.  However, since the current thread holds the old
214  * lock (the listening socket's lock, the head) access can still only be
215  * made to one side of the connection.
216  */
217 static void
218 unp_setpeerlocks(struct socket *so, struct socket *so2)
219 {
220 	struct unpcb *unp;
221 	kmutex_t *lock;
222 
223 	KASSERT(solocked2(so, so2));
224 
225 	/*
226 	 * Bail out if either end of the socket is not yet fully
227 	 * connected or accepted.  We only break the lock association
228 	 * with the head when the pair of sockets stand completely
229 	 * on their own.
230 	 */
231 	KASSERT(so->so_head == NULL);
232 	if (so2->so_head != NULL)
233 		return;
234 
235 	/*
236 	 * Drop references to old lock.  A third reference (from the
237 	 * queue head) must be held as we still hold its lock.  Bonus:
238 	 * we don't need to worry about garbage collecting the lock.
239 	 */
240 	lock = so->so_lock;
241 	KASSERT(lock == uipc_lock);
242 	mutex_obj_free(lock);
243 	mutex_obj_free(lock);
244 
245 	/*
246 	 * Grab stream lock from the initiator and share between the two
247 	 * endpoints.  Issue memory barrier to ensure all modifications
248 	 * become globally visible before the lock change.  so2 is
249 	 * assumed not to have a stream lock, because it was created
250 	 * purely for the server side to accept this connection and
251 	 * started out life using the domain-wide lock.
252 	 */
253 	unp = sotounpcb(so);
254 	KASSERT(unp->unp_streamlock != NULL);
255 	KASSERT(sotounpcb(so2)->unp_streamlock == NULL);
256 	lock = unp->unp_streamlock;
257 	unp->unp_streamlock = NULL;
258 	mutex_obj_hold(lock);
259 	membar_exit();
260 	/*
261 	 * possible race if lock is not held - see comment in
262 	 * uipc_usrreq(PRU_ACCEPT).
263 	 */
264 	KASSERT(mutex_owned(lock));
265 	solockreset(so, lock);
266 	solockreset(so2, lock);
267 }
268 
269 /*
270  * Reset a socket's lock back to the domain-wide lock.
271  */
272 static void
273 unp_resetlock(struct socket *so)
274 {
275 	kmutex_t *olock, *nlock;
276 	struct unpcb *unp;
277 
278 	KASSERT(solocked(so));
279 
280 	olock = so->so_lock;
281 	nlock = uipc_lock;
282 	if (olock == nlock)
283 		return;
284 	unp = sotounpcb(so);
285 	KASSERT(unp->unp_streamlock == NULL);
286 	unp->unp_streamlock = olock;
287 	mutex_obj_hold(nlock);
288 	mutex_enter(nlock);
289 	solockreset(so, nlock);
290 	mutex_exit(olock);
291 }
292 
293 static void
294 unp_free(struct unpcb *unp)
295 {
296 	if (unp->unp_addr)
297 		free(unp->unp_addr, M_SONAME);
298 	if (unp->unp_streamlock != NULL)
299 		mutex_obj_free(unp->unp_streamlock);
300 	kmem_free(unp, sizeof(*unp));
301 }
302 
303 static int
304 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp)
305 {
306 	struct socket *so2;
307 	const struct sockaddr_un *sun;
308 
309 	/* XXX: server side closed the socket */
310 	if (unp->unp_conn == NULL)
311 		return ECONNREFUSED;
312 	so2 = unp->unp_conn->unp_socket;
313 
314 	KASSERT(solocked(so2));
315 
316 	if (unp->unp_addr)
317 		sun = unp->unp_addr;
318 	else
319 		sun = &sun_noname;
320 	if (unp->unp_conn->unp_flags & UNP_WANTCRED)
321 		control = unp_addsockcred(curlwp, control);
322 	if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m,
323 	    control) == 0) {
324 		so2->so_rcv.sb_overflowed++;
325 		unp_dispose(control);
326 		m_freem(control);
327 		m_freem(m);
328 		return (ENOBUFS);
329 	} else {
330 		sorwakeup(so2);
331 		return (0);
332 	}
333 }
334 
335 static void
336 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr)
337 {
338 	const struct sockaddr_un *sun;
339 	struct unpcb *unp;
340 	bool ext;
341 
342 	KASSERT(solocked(so));
343 	unp = sotounpcb(so);
344 	ext = false;
345 
346 	for (;;) {
347 		sun = NULL;
348 		if (peeraddr) {
349 			if (unp->unp_conn && unp->unp_conn->unp_addr)
350 				sun = unp->unp_conn->unp_addr;
351 		} else {
352 			if (unp->unp_addr)
353 				sun = unp->unp_addr;
354 		}
355 		if (sun == NULL)
356 			sun = &sun_noname;
357 		nam->m_len = sun->sun_len;
358 		if (nam->m_len > MLEN && !ext) {
359 			sounlock(so);
360 			MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK);
361 			solock(so);
362 			ext = true;
363 		} else {
364 			KASSERT(nam->m_len <= MAXPATHLEN * 2);
365 			memcpy(mtod(nam, void *), sun, (size_t)nam->m_len);
366 			break;
367 		}
368 	}
369 }
370 
371 static int
372 unp_rcvd(struct socket *so, int flags, struct lwp *l)
373 {
374 	struct unpcb *unp = sotounpcb(so);
375 	struct socket *so2;
376 	u_int newhiwat;
377 
378 	KASSERT(solocked(so));
379 	KASSERT(unp != NULL);
380 
381 	switch (so->so_type) {
382 
383 	case SOCK_DGRAM:
384 		panic("uipc 1");
385 		/*NOTREACHED*/
386 
387 	case SOCK_SEQPACKET: /* FALLTHROUGH */
388 	case SOCK_STREAM:
389 #define	rcv (&so->so_rcv)
390 #define snd (&so2->so_snd)
391 		if (unp->unp_conn == 0)
392 			break;
393 		so2 = unp->unp_conn->unp_socket;
394 		KASSERT(solocked2(so, so2));
395 		/*
396 		 * Adjust backpressure on sender
397 		 * and wakeup any waiting to write.
398 		 */
399 		snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
400 		unp->unp_mbcnt = rcv->sb_mbcnt;
401 		newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc;
402 		(void)chgsbsize(so2->so_uidinfo,
403 		    &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
404 		unp->unp_cc = rcv->sb_cc;
405 		sowwakeup(so2);
406 #undef snd
407 #undef rcv
408 		break;
409 
410 	default:
411 		panic("uipc 2");
412 	}
413 
414 	return 0;
415 }
416 
417 static int
418 unp_recvoob(struct socket *so, struct mbuf *m, int flags)
419 {
420 	KASSERT(solocked(so));
421 
422 	return EOPNOTSUPP;
423 }
424 
425 static int
426 unp_send(struct socket *so, struct mbuf *m, struct mbuf *nam,
427     struct mbuf *control, struct lwp *l)
428 {
429 	struct unpcb *unp = sotounpcb(so);
430 	int error = 0;
431 	u_int newhiwat;
432 	struct socket *so2;
433 
434 	KASSERT(solocked(so));
435 	KASSERT(unp != NULL);
436 	KASSERT(m != NULL);
437 
438 	/*
439 	 * Note: unp_internalize() rejects any control message
440 	 * other than SCM_RIGHTS, and only allows one.  This
441 	 * has the side-effect of preventing a caller from
442 	 * forging SCM_CREDS.
443 	 */
444 	if (control) {
445 		sounlock(so);
446 		error = unp_internalize(&control);
447 		solock(so);
448 		if (error != 0) {
449 			m_freem(control);
450 			m_freem(m);
451 			return error;
452 		}
453 	}
454 
455 	switch (so->so_type) {
456 
457 	case SOCK_DGRAM: {
458 		KASSERT(so->so_lock == uipc_lock);
459 		if (nam) {
460 			if ((so->so_state & SS_ISCONNECTED) != 0)
461 				error = EISCONN;
462 			else {
463 				/*
464 				 * Note: once connected, the
465 				 * socket's lock must not be
466 				 * dropped until we have sent
467 				 * the message and disconnected.
468 				 * This is necessary to prevent
469 				 * intervening control ops, like
470 				 * another connection.
471 				 */
472 				error = unp_connect(so, nam, l);
473 			}
474 		} else {
475 			if ((so->so_state & SS_ISCONNECTED) == 0)
476 				error = ENOTCONN;
477 		}
478 		if (error) {
479 			unp_dispose(control);
480 			m_freem(control);
481 			m_freem(m);
482 			return error;
483 		}
484 		error = unp_output(m, control, unp);
485 		if (nam)
486 			unp_disconnect1(unp);
487 		break;
488 	}
489 
490 	case SOCK_SEQPACKET: /* FALLTHROUGH */
491 	case SOCK_STREAM:
492 #define	rcv (&so2->so_rcv)
493 #define	snd (&so->so_snd)
494 		if (unp->unp_conn == NULL) {
495 			error = ENOTCONN;
496 			break;
497 		}
498 		so2 = unp->unp_conn->unp_socket;
499 		KASSERT(solocked2(so, so2));
500 		if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
501 			/*
502 			 * Credentials are passed only once on
503 			 * SOCK_STREAM and SOCK_SEQPACKET.
504 			 */
505 			unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
506 			control = unp_addsockcred(l, control);
507 		}
508 		/*
509 		 * Send to paired receive port, and then reduce
510 		 * send buffer hiwater marks to maintain backpressure.
511 		 * Wake up readers.
512 		 */
513 		if (control) {
514 			if (sbappendcontrol(rcv, m, control) != 0)
515 				control = NULL;
516 		} else {
517 			switch(so->so_type) {
518 			case SOCK_SEQPACKET:
519 				sbappendrecord(rcv, m);
520 				break;
521 			case SOCK_STREAM:
522 				sbappend(rcv, m);
523 				break;
524 			default:
525 				panic("uipc_usrreq");
526 				break;
527 			}
528 		}
529 		snd->sb_mbmax -=
530 		    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
531 		unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
532 		newhiwat = snd->sb_hiwat -
533 		    (rcv->sb_cc - unp->unp_conn->unp_cc);
534 		(void)chgsbsize(so->so_uidinfo,
535 		    &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
536 		unp->unp_conn->unp_cc = rcv->sb_cc;
537 		sorwakeup(so2);
538 #undef snd
539 #undef rcv
540 		if (control != NULL) {
541 			unp_dispose(control);
542 			m_freem(control);
543 		}
544 		break;
545 
546 	default:
547 		panic("uipc 4");
548 	}
549 
550 	return error;
551 }
552 
553 static int
554 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control)
555 {
556 	KASSERT(solocked(so));
557 
558 	m_freem(m);
559 	m_freem(control);
560 
561 	return EOPNOTSUPP;
562 }
563 
564 static int
565 unp_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam,
566     struct mbuf *control, struct lwp *l)
567 {
568 
569 	KASSERT(req != PRU_ATTACH);
570 	KASSERT(req != PRU_DETACH);
571 	KASSERT(req != PRU_ACCEPT);
572 	KASSERT(req != PRU_BIND);
573 	KASSERT(req != PRU_LISTEN);
574 	KASSERT(req != PRU_CONNECT);
575 	KASSERT(req != PRU_CONNECT2);
576 	KASSERT(req != PRU_DISCONNECT);
577 	KASSERT(req != PRU_SHUTDOWN);
578 	KASSERT(req != PRU_ABORT);
579 	KASSERT(req != PRU_CONTROL);
580 	KASSERT(req != PRU_SENSE);
581 	KASSERT(req != PRU_PEERADDR);
582 	KASSERT(req != PRU_SOCKADDR);
583 	KASSERT(req != PRU_RCVD);
584 	KASSERT(req != PRU_RCVOOB);
585 	KASSERT(req != PRU_SEND);
586 	KASSERT(req != PRU_SENDOOB);
587 	KASSERT(req != PRU_PURGEIF);
588 
589 	KASSERT(solocked(so));
590 
591 	if (sotounpcb(so) == NULL)
592 		return EINVAL;
593 
594 	panic("piusrreq");
595 
596 	return 0;
597 }
598 
599 /*
600  * Unix domain socket option processing.
601  */
602 int
603 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt)
604 {
605 	struct unpcb *unp = sotounpcb(so);
606 	int optval = 0, error = 0;
607 
608 	KASSERT(solocked(so));
609 
610 	if (sopt->sopt_level != 0) {
611 		error = ENOPROTOOPT;
612 	} else switch (op) {
613 
614 	case PRCO_SETOPT:
615 		switch (sopt->sopt_name) {
616 		case LOCAL_CREDS:
617 		case LOCAL_CONNWAIT:
618 			error = sockopt_getint(sopt, &optval);
619 			if (error)
620 				break;
621 			switch (sopt->sopt_name) {
622 #define	OPTSET(bit) \
623 	if (optval) \
624 		unp->unp_flags |= (bit); \
625 	else \
626 		unp->unp_flags &= ~(bit);
627 
628 			case LOCAL_CREDS:
629 				OPTSET(UNP_WANTCRED);
630 				break;
631 			case LOCAL_CONNWAIT:
632 				OPTSET(UNP_CONNWAIT);
633 				break;
634 			}
635 			break;
636 #undef OPTSET
637 
638 		default:
639 			error = ENOPROTOOPT;
640 			break;
641 		}
642 		break;
643 
644 	case PRCO_GETOPT:
645 		sounlock(so);
646 		switch (sopt->sopt_name) {
647 		case LOCAL_PEEREID:
648 			if (unp->unp_flags & UNP_EIDSVALID) {
649 				error = sockopt_set(sopt,
650 				    &unp->unp_connid, sizeof(unp->unp_connid));
651 			} else {
652 				error = EINVAL;
653 			}
654 			break;
655 		case LOCAL_CREDS:
656 #define	OPTBIT(bit)	(unp->unp_flags & (bit) ? 1 : 0)
657 
658 			optval = OPTBIT(UNP_WANTCRED);
659 			error = sockopt_setint(sopt, optval);
660 			break;
661 #undef OPTBIT
662 
663 		default:
664 			error = ENOPROTOOPT;
665 			break;
666 		}
667 		solock(so);
668 		break;
669 	}
670 	return (error);
671 }
672 
673 /*
674  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
675  * for stream sockets, although the total for sender and receiver is
676  * actually only PIPSIZ.
677  * Datagram sockets really use the sendspace as the maximum datagram size,
678  * and don't really want to reserve the sendspace.  Their recvspace should
679  * be large enough for at least one max-size datagram plus address.
680  */
681 #define	PIPSIZ	4096
682 u_long	unpst_sendspace = PIPSIZ;
683 u_long	unpst_recvspace = PIPSIZ;
684 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
685 u_long	unpdg_recvspace = 4*1024;
686 
687 u_int	unp_rights;			/* files in flight */
688 u_int	unp_rights_ratio = 2;		/* limit, fraction of maxfiles */
689 
690 static int
691 unp_attach(struct socket *so, int proto)
692 {
693 	struct unpcb *unp = sotounpcb(so);
694 	u_long sndspc, rcvspc;
695 	int error;
696 
697 	KASSERT(unp == NULL);
698 
699 	switch (so->so_type) {
700 	case SOCK_SEQPACKET:
701 		/* FALLTHROUGH */
702 	case SOCK_STREAM:
703 		if (so->so_lock == NULL) {
704 			so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
705 			solock(so);
706 		}
707 		sndspc = unpst_sendspace;
708 		rcvspc = unpst_recvspace;
709 		break;
710 
711 	case SOCK_DGRAM:
712 		if (so->so_lock == NULL) {
713 			mutex_obj_hold(uipc_lock);
714 			so->so_lock = uipc_lock;
715 			solock(so);
716 		}
717 		sndspc = unpdg_sendspace;
718 		rcvspc = unpdg_recvspace;
719 		break;
720 
721 	default:
722 		panic("unp_attach");
723 	}
724 
725 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
726 		error = soreserve(so, sndspc, rcvspc);
727 		if (error) {
728 			return error;
729 		}
730 	}
731 
732 	unp = kmem_zalloc(sizeof(*unp), KM_SLEEP);
733 	nanotime(&unp->unp_ctime);
734 	unp->unp_socket = so;
735 	so->so_pcb = unp;
736 
737 	KASSERT(solocked(so));
738 	return 0;
739 }
740 
741 static void
742 unp_detach(struct socket *so)
743 {
744 	struct unpcb *unp;
745 	vnode_t *vp;
746 
747 	unp = sotounpcb(so);
748 	KASSERT(unp != NULL);
749 	KASSERT(solocked(so));
750  retry:
751 	if ((vp = unp->unp_vnode) != NULL) {
752 		sounlock(so);
753 		/* Acquire v_interlock to protect against unp_connect(). */
754 		/* XXXAD racy */
755 		mutex_enter(vp->v_interlock);
756 		vp->v_socket = NULL;
757 		mutex_exit(vp->v_interlock);
758 		vrele(vp);
759 		solock(so);
760 		unp->unp_vnode = NULL;
761 	}
762 	if (unp->unp_conn)
763 		unp_disconnect1(unp);
764 	while (unp->unp_refs) {
765 		KASSERT(solocked2(so, unp->unp_refs->unp_socket));
766 		if (unp_drop(unp->unp_refs, ECONNRESET)) {
767 			solock(so);
768 			goto retry;
769 		}
770 	}
771 	soisdisconnected(so);
772 	so->so_pcb = NULL;
773 	if (unp_rights) {
774 		/*
775 		 * Normally the receive buffer is flushed later, in sofree,
776 		 * but if our receive buffer holds references to files that
777 		 * are now garbage, we will enqueue those file references to
778 		 * the garbage collector and kick it into action.
779 		 */
780 		sorflush(so);
781 		unp_free(unp);
782 		unp_thread_kick();
783 	} else
784 		unp_free(unp);
785 }
786 
787 static int
788 unp_accept(struct socket *so, struct mbuf *nam)
789 {
790 	struct unpcb *unp = sotounpcb(so);
791 	struct socket *so2;
792 
793 	KASSERT(solocked(so));
794 	KASSERT(nam != NULL);
795 
796 	/* XXX code review required to determine if unp can ever be NULL */
797 	if (unp == NULL)
798 		return EINVAL;
799 
800 	KASSERT(so->so_lock == uipc_lock);
801 	/*
802 	 * Mark the initiating STREAM socket as connected *ONLY*
803 	 * after it's been accepted.  This prevents a client from
804 	 * overrunning a server and receiving ECONNREFUSED.
805 	 */
806 	if (unp->unp_conn == NULL) {
807 		/*
808 		 * This will use the empty socket and will not
809 		 * allocate.
810 		 */
811 		unp_setaddr(so, nam, true);
812 		return 0;
813 	}
814 	so2 = unp->unp_conn->unp_socket;
815 	if (so2->so_state & SS_ISCONNECTING) {
816 		KASSERT(solocked2(so, so->so_head));
817 		KASSERT(solocked2(so2, so->so_head));
818 		soisconnected(so2);
819 	}
820 	/*
821 	 * If the connection is fully established, break the
822 	 * association with uipc_lock and give the connected
823 	 * pair a separate lock to share.
824 	 * There is a race here: sotounpcb(so2)->unp_streamlock
825 	 * is not locked, so when changing so2->so_lock
826 	 * another thread can grab it while so->so_lock is still
827 	 * pointing to the (locked) uipc_lock.
828 	 * this should be harmless, except that this makes
829 	 * solocked2() and solocked() unreliable.
830 	 * Another problem is that unp_setaddr() expects the
831 	 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock
832 	 * fixes both issues.
833 	 */
834 	mutex_enter(sotounpcb(so2)->unp_streamlock);
835 	unp_setpeerlocks(so2, so);
836 	/*
837 	 * Only now return peer's address, as we may need to
838 	 * block in order to allocate memory.
839 	 *
840 	 * XXX Minor race: connection can be broken while
841 	 * lock is dropped in unp_setaddr().  We will return
842 	 * error == 0 and sun_noname as the peer address.
843 	 */
844 	unp_setaddr(so, nam, true);
845 	/* so_lock now points to unp_streamlock */
846 	mutex_exit(so2->so_lock);
847 	return 0;
848 }
849 
850 static int
851 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp)
852 {
853 	return EOPNOTSUPP;
854 }
855 
856 static int
857 unp_stat(struct socket *so, struct stat *ub)
858 {
859 	struct unpcb *unp;
860 	struct socket *so2;
861 
862 	KASSERT(solocked(so));
863 
864 	unp = sotounpcb(so);
865 	if (unp == NULL)
866 		return EINVAL;
867 
868 	ub->st_blksize = so->so_snd.sb_hiwat;
869 	switch (so->so_type) {
870 	case SOCK_SEQPACKET: /* FALLTHROUGH */
871 	case SOCK_STREAM:
872 		if (unp->unp_conn == 0)
873 			break;
874 
875 		so2 = unp->unp_conn->unp_socket;
876 		KASSERT(solocked2(so, so2));
877 		ub->st_blksize += so2->so_rcv.sb_cc;
878 		break;
879 	default:
880 		break;
881 	}
882 	ub->st_dev = NODEV;
883 	if (unp->unp_ino == 0)
884 		unp->unp_ino = unp_ino++;
885 	ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime;
886 	ub->st_ino = unp->unp_ino;
887 	return (0);
888 }
889 
890 static int
891 unp_peeraddr(struct socket *so, struct mbuf *nam)
892 {
893 	KASSERT(solocked(so));
894 	KASSERT(sotounpcb(so) != NULL);
895 	KASSERT(nam != NULL);
896 
897 	unp_setaddr(so, nam, true);
898 	return 0;
899 }
900 
901 static int
902 unp_sockaddr(struct socket *so, struct mbuf *nam)
903 {
904 	KASSERT(solocked(so));
905 	KASSERT(sotounpcb(so) != NULL);
906 	KASSERT(nam != NULL);
907 
908 	unp_setaddr(so, nam, false);
909 	return 0;
910 }
911 
912 /*
913  * Allocate the new sockaddr.  We have to allocate one
914  * extra byte so that we can ensure that the pathname
915  * is nul-terminated. Note that unlike linux, we don't
916  * include in the address length the NUL in the path
917  * component, because doing so, would exceed sizeof(sockaddr_un)
918  * for fully occupied pathnames. Linux is also inconsistent,
919  * because it does not include the NUL in the length of
920  * what it calls "abstract" unix sockets.
921  */
922 static struct sockaddr_un *
923 makeun(struct mbuf *nam, size_t *addrlen) {
924 	struct sockaddr_un *sun;
925 
926 	*addrlen = nam->m_len + 1;
927 	sun = malloc(*addrlen, M_SONAME, M_WAITOK);
928 	m_copydata(nam, 0, nam->m_len, (void *)sun);
929 	*(((char *)sun) + nam->m_len) = '\0';
930 	sun->sun_len = strlen(sun->sun_path) +
931 	    offsetof(struct sockaddr_un, sun_path);
932 	return sun;
933 }
934 
935 static int
936 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l)
937 {
938 	struct sockaddr_un *sun;
939 	struct unpcb *unp;
940 	vnode_t *vp;
941 	struct vattr vattr;
942 	size_t addrlen;
943 	int error;
944 	struct pathbuf *pb;
945 	struct nameidata nd;
946 	proc_t *p;
947 
948 	unp = sotounpcb(so);
949 
950 	KASSERT(solocked(so));
951 	KASSERT(unp != NULL);
952 	KASSERT(nam != NULL);
953 
954 	if (unp->unp_vnode != NULL)
955 		return (EINVAL);
956 	if ((unp->unp_flags & UNP_BUSY) != 0) {
957 		/*
958 		 * EALREADY may not be strictly accurate, but since this
959 		 * is a major application error it's hardly a big deal.
960 		 */
961 		return (EALREADY);
962 	}
963 	unp->unp_flags |= UNP_BUSY;
964 	sounlock(so);
965 
966 	p = l->l_proc;
967 	sun = makeun(nam, &addrlen);
968 
969 	pb = pathbuf_create(sun->sun_path);
970 	if (pb == NULL) {
971 		error = ENOMEM;
972 		goto bad;
973 	}
974 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb);
975 
976 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
977 	if ((error = namei(&nd)) != 0) {
978 		pathbuf_destroy(pb);
979 		goto bad;
980 	}
981 	vp = nd.ni_vp;
982 	if (vp != NULL) {
983 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
984 		if (nd.ni_dvp == vp)
985 			vrele(nd.ni_dvp);
986 		else
987 			vput(nd.ni_dvp);
988 		vrele(vp);
989 		pathbuf_destroy(pb);
990 		error = EADDRINUSE;
991 		goto bad;
992 	}
993 	vattr_null(&vattr);
994 	vattr.va_type = VSOCK;
995 	vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask);
996 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
997 	if (error) {
998 		vput(nd.ni_dvp);
999 		pathbuf_destroy(pb);
1000 		goto bad;
1001 	}
1002 	vp = nd.ni_vp;
1003 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1004 	solock(so);
1005 	vp->v_socket = unp->unp_socket;
1006 	unp->unp_vnode = vp;
1007 	unp->unp_addrlen = addrlen;
1008 	unp->unp_addr = sun;
1009 	unp->unp_connid.unp_pid = p->p_pid;
1010 	unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
1011 	unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
1012 	unp->unp_flags |= UNP_EIDSBIND;
1013 	VOP_UNLOCK(vp);
1014 	vput(nd.ni_dvp);
1015 	unp->unp_flags &= ~UNP_BUSY;
1016 	pathbuf_destroy(pb);
1017 	return (0);
1018 
1019  bad:
1020 	free(sun, M_SONAME);
1021 	solock(so);
1022 	unp->unp_flags &= ~UNP_BUSY;
1023 	return (error);
1024 }
1025 
1026 static int
1027 unp_listen(struct socket *so, struct lwp *l)
1028 {
1029 	struct unpcb *unp = sotounpcb(so);
1030 
1031 	KASSERT(solocked(so));
1032 	KASSERT(unp != NULL);
1033 
1034 	/*
1035 	 * If the socket can accept a connection, it must be
1036 	 * locked by uipc_lock.
1037 	 */
1038 	unp_resetlock(so);
1039 	if (unp->unp_vnode == NULL)
1040 		return EINVAL;
1041 
1042 	return 0;
1043 }
1044 
1045 static int
1046 unp_disconnect(struct socket *so)
1047 {
1048 	KASSERT(solocked(so));
1049 	KASSERT(sotounpcb(so) != NULL);
1050 
1051 	unp_disconnect1(sotounpcb(so));
1052 	return 0;
1053 }
1054 
1055 static int
1056 unp_shutdown(struct socket *so)
1057 {
1058 	KASSERT(solocked(so));
1059 	KASSERT(sotounpcb(so) != NULL);
1060 
1061 	socantsendmore(so);
1062 	unp_shutdown1(sotounpcb(so));
1063 	return 0;
1064 }
1065 
1066 static int
1067 unp_abort(struct socket *so)
1068 {
1069 	KASSERT(solocked(so));
1070 	KASSERT(sotounpcb(so) != NULL);
1071 
1072 	(void)unp_drop(sotounpcb(so), ECONNABORTED);
1073 	KASSERT(so->so_head == NULL);
1074 	KASSERT(so->so_pcb != NULL);
1075 	unp_detach(so);
1076 	return 0;
1077 }
1078 
1079 static int
1080 unp_connect1(struct socket *so, struct socket *so2)
1081 {
1082 	struct unpcb *unp = sotounpcb(so);
1083 	struct unpcb *unp2;
1084 
1085 	if (so2->so_type != so->so_type)
1086 		return EPROTOTYPE;
1087 
1088 	/*
1089 	 * All three sockets involved must be locked by same lock:
1090 	 *
1091 	 * local endpoint (so)
1092 	 * remote endpoint (so2)
1093 	 * queue head (so2->so_head, only if PR_CONNREQUIRED)
1094 	 */
1095 	KASSERT(solocked2(so, so2));
1096 	KASSERT(so->so_head == NULL);
1097 	if (so2->so_head != NULL) {
1098 		KASSERT(so2->so_lock == uipc_lock);
1099 		KASSERT(solocked2(so2, so2->so_head));
1100 	}
1101 
1102 	unp2 = sotounpcb(so2);
1103 	unp->unp_conn = unp2;
1104 	switch (so->so_type) {
1105 
1106 	case SOCK_DGRAM:
1107 		unp->unp_nextref = unp2->unp_refs;
1108 		unp2->unp_refs = unp;
1109 		soisconnected(so);
1110 		break;
1111 
1112 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1113 	case SOCK_STREAM:
1114 
1115 		/*
1116 		 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers
1117 		 * which are unp_connect() or unp_connect2().
1118 		 */
1119 
1120 		break;
1121 
1122 	default:
1123 		panic("unp_connect1");
1124 	}
1125 
1126 	return 0;
1127 }
1128 
1129 int
1130 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l)
1131 {
1132 	struct sockaddr_un *sun;
1133 	vnode_t *vp;
1134 	struct socket *so2, *so3;
1135 	struct unpcb *unp, *unp2, *unp3;
1136 	size_t addrlen;
1137 	int error;
1138 	struct pathbuf *pb;
1139 	struct nameidata nd;
1140 
1141 	unp = sotounpcb(so);
1142 	if ((unp->unp_flags & UNP_BUSY) != 0) {
1143 		/*
1144 		 * EALREADY may not be strictly accurate, but since this
1145 		 * is a major application error it's hardly a big deal.
1146 		 */
1147 		return (EALREADY);
1148 	}
1149 	unp->unp_flags |= UNP_BUSY;
1150 	sounlock(so);
1151 
1152 	sun = makeun(nam, &addrlen);
1153 	pb = pathbuf_create(sun->sun_path);
1154 	if (pb == NULL) {
1155 		error = ENOMEM;
1156 		goto bad2;
1157 	}
1158 
1159 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
1160 
1161 	if ((error = namei(&nd)) != 0) {
1162 		pathbuf_destroy(pb);
1163 		goto bad2;
1164 	}
1165 	vp = nd.ni_vp;
1166 	if (vp->v_type != VSOCK) {
1167 		error = ENOTSOCK;
1168 		goto bad;
1169 	}
1170 	pathbuf_destroy(pb);
1171 	if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0)
1172 		goto bad;
1173 	/* Acquire v_interlock to protect against unp_detach(). */
1174 	mutex_enter(vp->v_interlock);
1175 	so2 = vp->v_socket;
1176 	if (so2 == NULL) {
1177 		mutex_exit(vp->v_interlock);
1178 		error = ECONNREFUSED;
1179 		goto bad;
1180 	}
1181 	if (so->so_type != so2->so_type) {
1182 		mutex_exit(vp->v_interlock);
1183 		error = EPROTOTYPE;
1184 		goto bad;
1185 	}
1186 	solock(so);
1187 	unp_resetlock(so);
1188 	mutex_exit(vp->v_interlock);
1189 	if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1190 		/*
1191 		 * This may seem somewhat fragile but is OK: if we can
1192 		 * see SO_ACCEPTCONN set on the endpoint, then it must
1193 		 * be locked by the domain-wide uipc_lock.
1194 		 */
1195 		KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 ||
1196 		    so2->so_lock == uipc_lock);
1197 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
1198 		    (so3 = sonewconn(so2, false)) == NULL) {
1199 			error = ECONNREFUSED;
1200 			sounlock(so);
1201 			goto bad;
1202 		}
1203 		unp2 = sotounpcb(so2);
1204 		unp3 = sotounpcb(so3);
1205 		if (unp2->unp_addr) {
1206 			unp3->unp_addr = malloc(unp2->unp_addrlen,
1207 			    M_SONAME, M_WAITOK);
1208 			memcpy(unp3->unp_addr, unp2->unp_addr,
1209 			    unp2->unp_addrlen);
1210 			unp3->unp_addrlen = unp2->unp_addrlen;
1211 		}
1212 		unp3->unp_flags = unp2->unp_flags;
1213 		unp3->unp_connid.unp_pid = l->l_proc->p_pid;
1214 		unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
1215 		unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
1216 		unp3->unp_flags |= UNP_EIDSVALID;
1217 		if (unp2->unp_flags & UNP_EIDSBIND) {
1218 			unp->unp_connid = unp2->unp_connid;
1219 			unp->unp_flags |= UNP_EIDSVALID;
1220 		}
1221 		so2 = so3;
1222 	}
1223 	error = unp_connect1(so, so2);
1224 	if (error) {
1225 		sounlock(so);
1226 		goto bad;
1227 	}
1228 	unp2 = sotounpcb(so2);
1229 	switch (so->so_type) {
1230 
1231 	/*
1232 	 * SOCK_DGRAM and default cases are handled in prior call to
1233 	 * unp_connect1(), do not add a default case without fixing
1234 	 * unp_connect1().
1235 	 */
1236 
1237 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1238 	case SOCK_STREAM:
1239 		unp2->unp_conn = unp;
1240 		if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)
1241 			soisconnecting(so);
1242 		else
1243 			soisconnected(so);
1244 		soisconnected(so2);
1245 		/*
1246 		 * If the connection is fully established, break the
1247 		 * association with uipc_lock and give the connected
1248 		 * pair a seperate lock to share.
1249 		 */
1250 		KASSERT(so2->so_head != NULL);
1251 		unp_setpeerlocks(so, so2);
1252 		break;
1253 
1254 	}
1255 	sounlock(so);
1256  bad:
1257 	vput(vp);
1258  bad2:
1259 	free(sun, M_SONAME);
1260 	solock(so);
1261 	unp->unp_flags &= ~UNP_BUSY;
1262 	return (error);
1263 }
1264 
1265 int
1266 unp_connect2(struct socket *so, struct socket *so2)
1267 {
1268 	struct unpcb *unp = sotounpcb(so);
1269 	struct unpcb *unp2;
1270 	int error = 0;
1271 
1272 	KASSERT(solocked2(so, so2));
1273 
1274 	error = unp_connect1(so, so2);
1275 	if (error)
1276 		return error;
1277 
1278 	unp2 = sotounpcb(so2);
1279 	switch (so->so_type) {
1280 
1281 	/*
1282 	 * SOCK_DGRAM and default cases are handled in prior call to
1283 	 * unp_connect1(), do not add a default case without fixing
1284 	 * unp_connect1().
1285 	 */
1286 
1287 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1288 	case SOCK_STREAM:
1289 		unp2->unp_conn = unp;
1290 		soisconnected(so);
1291 		soisconnected(so2);
1292 		break;
1293 
1294 	}
1295 	return error;
1296 }
1297 
1298 static void
1299 unp_disconnect1(struct unpcb *unp)
1300 {
1301 	struct unpcb *unp2 = unp->unp_conn;
1302 	struct socket *so;
1303 
1304 	if (unp2 == 0)
1305 		return;
1306 	unp->unp_conn = 0;
1307 	so = unp->unp_socket;
1308 	switch (so->so_type) {
1309 	case SOCK_DGRAM:
1310 		if (unp2->unp_refs == unp)
1311 			unp2->unp_refs = unp->unp_nextref;
1312 		else {
1313 			unp2 = unp2->unp_refs;
1314 			for (;;) {
1315 				KASSERT(solocked2(so, unp2->unp_socket));
1316 				if (unp2 == 0)
1317 					panic("unp_disconnect1");
1318 				if (unp2->unp_nextref == unp)
1319 					break;
1320 				unp2 = unp2->unp_nextref;
1321 			}
1322 			unp2->unp_nextref = unp->unp_nextref;
1323 		}
1324 		unp->unp_nextref = 0;
1325 		so->so_state &= ~SS_ISCONNECTED;
1326 		break;
1327 
1328 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1329 	case SOCK_STREAM:
1330 		KASSERT(solocked2(so, unp2->unp_socket));
1331 		soisdisconnected(so);
1332 		unp2->unp_conn = 0;
1333 		soisdisconnected(unp2->unp_socket);
1334 		break;
1335 	}
1336 }
1337 
1338 static void
1339 unp_shutdown1(struct unpcb *unp)
1340 {
1341 	struct socket *so;
1342 
1343 	switch(unp->unp_socket->so_type) {
1344 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1345 	case SOCK_STREAM:
1346 		if (unp->unp_conn && (so = unp->unp_conn->unp_socket))
1347 			socantrcvmore(so);
1348 		break;
1349 	default:
1350 		break;
1351 	}
1352 }
1353 
1354 static bool
1355 unp_drop(struct unpcb *unp, int errno)
1356 {
1357 	struct socket *so = unp->unp_socket;
1358 
1359 	KASSERT(solocked(so));
1360 
1361 	so->so_error = errno;
1362 	unp_disconnect1(unp);
1363 	if (so->so_head) {
1364 		so->so_pcb = NULL;
1365 		/* sofree() drops the socket lock */
1366 		sofree(so);
1367 		unp_free(unp);
1368 		return true;
1369 	}
1370 	return false;
1371 }
1372 
1373 #ifdef notdef
1374 unp_drain(void)
1375 {
1376 
1377 }
1378 #endif
1379 
1380 int
1381 unp_externalize(struct mbuf *rights, struct lwp *l, int flags)
1382 {
1383 	struct cmsghdr * const cm = mtod(rights, struct cmsghdr *);
1384 	struct proc * const p = l->l_proc;
1385 	file_t **rp;
1386 	int error = 0;
1387 
1388 	const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
1389 	    sizeof(file_t *);
1390 	if (nfds == 0)
1391 		goto noop;
1392 
1393 	int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP);
1394 	rw_enter(&p->p_cwdi->cwdi_lock, RW_READER);
1395 
1396 	/* Make sure the recipient should be able to see the files.. */
1397 	rp = (file_t **)CMSG_DATA(cm);
1398 	for (size_t i = 0; i < nfds; i++) {
1399 		file_t * const fp = *rp++;
1400 		if (fp == NULL) {
1401 			error = EINVAL;
1402 			goto out;
1403 		}
1404 		/*
1405 		 * If we are in a chroot'ed directory, and
1406 		 * someone wants to pass us a directory, make
1407 		 * sure it's inside the subtree we're allowed
1408 		 * to access.
1409 		 */
1410 		if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) {
1411 			vnode_t *vp = fp->f_vnode;
1412 			if ((vp->v_type == VDIR) &&
1413 			    !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) {
1414 				error = EPERM;
1415 				goto out;
1416 			}
1417 		}
1418 	}
1419 
1420  restart:
1421 	/*
1422 	 * First loop -- allocate file descriptor table slots for the
1423 	 * new files.
1424 	 */
1425 	for (size_t i = 0; i < nfds; i++) {
1426 		if ((error = fd_alloc(p, 0, &fdp[i])) != 0) {
1427 			/*
1428 			 * Back out what we've done so far.
1429 			 */
1430 			while (i-- > 0) {
1431 				fd_abort(p, NULL, fdp[i]);
1432 			}
1433 			if (error == ENOSPC) {
1434 				fd_tryexpand(p);
1435 				error = 0;
1436 				goto restart;
1437 			}
1438 			/*
1439 			 * This is the error that has historically
1440 			 * been returned, and some callers may
1441 			 * expect it.
1442 			 */
1443 			error = EMSGSIZE;
1444 			goto out;
1445 		}
1446 	}
1447 
1448 	/*
1449 	 * Now that adding them has succeeded, update all of the
1450 	 * file passing state and affix the descriptors.
1451 	 */
1452 	rp = (file_t **)CMSG_DATA(cm);
1453 	int *ofdp = (int *)CMSG_DATA(cm);
1454 	for (size_t i = 0; i < nfds; i++) {
1455 		file_t * const fp = *rp++;
1456 		const int fd = fdp[i];
1457 		atomic_dec_uint(&unp_rights);
1458 		fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
1459 		fd_affix(p, fp, fd);
1460 		/*
1461 		 * Done with this file pointer, replace it with a fd;
1462 		 */
1463 		*ofdp++ = fd;
1464 		mutex_enter(&fp->f_lock);
1465 		fp->f_msgcount--;
1466 		mutex_exit(&fp->f_lock);
1467 		/*
1468 		 * Note that fd_affix() adds a reference to the file.
1469 		 * The file may already have been closed by another
1470 		 * LWP in the process, so we must drop the reference
1471 		 * added by unp_internalize() with closef().
1472 		 */
1473 		closef(fp);
1474 	}
1475 
1476 	/*
1477 	 * Adjust length, in case of transition from large file_t
1478 	 * pointers to ints.
1479 	 */
1480 	if (sizeof(file_t *) != sizeof(int)) {
1481 		cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
1482 		rights->m_len = CMSG_SPACE(nfds * sizeof(int));
1483 	}
1484  out:
1485 	if (__predict_false(error != 0)) {
1486 		file_t **const fpp = (file_t **)CMSG_DATA(cm);
1487 		for (size_t i = 0; i < nfds; i++)
1488 			unp_discard_now(fpp[i]);
1489 		/*
1490 		 * Truncate the array so that nobody will try to interpret
1491 		 * what is now garbage in it.
1492 		 */
1493 		cm->cmsg_len = CMSG_LEN(0);
1494 		rights->m_len = CMSG_SPACE(0);
1495 	}
1496 	rw_exit(&p->p_cwdi->cwdi_lock);
1497 	kmem_free(fdp, nfds * sizeof(int));
1498 
1499  noop:
1500 	/*
1501 	 * Don't disclose kernel memory in the alignment space.
1502 	 */
1503 	KASSERT(cm->cmsg_len <= rights->m_len);
1504 	memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len -
1505 	    cm->cmsg_len);
1506 	return error;
1507 }
1508 
1509 static int
1510 unp_internalize(struct mbuf **controlp)
1511 {
1512 	filedesc_t *fdescp = curlwp->l_fd;
1513 	struct mbuf *control = *controlp;
1514 	struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
1515 	file_t **rp, **files;
1516 	file_t *fp;
1517 	int i, fd, *fdp;
1518 	int nfds, error;
1519 	u_int maxmsg;
1520 
1521 	error = 0;
1522 	newcm = NULL;
1523 
1524 	/* Sanity check the control message header. */
1525 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
1526 	    cm->cmsg_len > control->m_len ||
1527 	    cm->cmsg_len < CMSG_ALIGN(sizeof(*cm)))
1528 		return (EINVAL);
1529 
1530 	/*
1531 	 * Verify that the file descriptors are valid, and acquire
1532 	 * a reference to each.
1533 	 */
1534 	nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
1535 	fdp = (int *)CMSG_DATA(cm);
1536 	maxmsg = maxfiles / unp_rights_ratio;
1537 	for (i = 0; i < nfds; i++) {
1538 		fd = *fdp++;
1539 		if (atomic_inc_uint_nv(&unp_rights) > maxmsg) {
1540 			atomic_dec_uint(&unp_rights);
1541 			nfds = i;
1542 			error = EAGAIN;
1543 			goto out;
1544 		}
1545 		if ((fp = fd_getfile(fd)) == NULL
1546 		    || fp->f_type == DTYPE_KQUEUE) {
1547 		    	if (fp)
1548 		    		fd_putfile(fd);
1549 			atomic_dec_uint(&unp_rights);
1550 			nfds = i;
1551 			error = EBADF;
1552 			goto out;
1553 		}
1554 	}
1555 
1556 	/* Allocate new space and copy header into it. */
1557 	newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK);
1558 	if (newcm == NULL) {
1559 		error = E2BIG;
1560 		goto out;
1561 	}
1562 	memcpy(newcm, cm, sizeof(struct cmsghdr));
1563 	files = (file_t **)CMSG_DATA(newcm);
1564 
1565 	/*
1566 	 * Transform the file descriptors into file_t pointers, in
1567 	 * reverse order so that if pointers are bigger than ints, the
1568 	 * int won't get until we're done.  No need to lock, as we have
1569 	 * already validated the descriptors with fd_getfile().
1570 	 */
1571 	fdp = (int *)CMSG_DATA(cm) + nfds;
1572 	rp = files + nfds;
1573 	for (i = 0; i < nfds; i++) {
1574 		fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file;
1575 		KASSERT(fp != NULL);
1576 		mutex_enter(&fp->f_lock);
1577 		*--rp = fp;
1578 		fp->f_count++;
1579 		fp->f_msgcount++;
1580 		mutex_exit(&fp->f_lock);
1581 	}
1582 
1583  out:
1584  	/* Release descriptor references. */
1585 	fdp = (int *)CMSG_DATA(cm);
1586 	for (i = 0; i < nfds; i++) {
1587 		fd_putfile(*fdp++);
1588 		if (error != 0) {
1589 			atomic_dec_uint(&unp_rights);
1590 		}
1591 	}
1592 
1593 	if (error == 0) {
1594 		if (control->m_flags & M_EXT) {
1595 			m_freem(control);
1596 			*controlp = control = m_get(M_WAIT, MT_CONTROL);
1597 		}
1598 		MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)),
1599 		    M_MBUF, NULL, NULL);
1600 		cm = newcm;
1601 		/*
1602 		 * Adjust message & mbuf to note amount of space
1603 		 * actually used.
1604 		 */
1605 		cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *));
1606 		control->m_len = CMSG_SPACE(nfds * sizeof(file_t *));
1607 	}
1608 
1609 	return error;
1610 }
1611 
1612 struct mbuf *
1613 unp_addsockcred(struct lwp *l, struct mbuf *control)
1614 {
1615 	struct sockcred *sc;
1616 	struct mbuf *m;
1617 	void *p;
1618 
1619 	m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)),
1620 		SCM_CREDS, SOL_SOCKET, M_WAITOK);
1621 	if (m == NULL)
1622 		return control;
1623 
1624 	sc = p;
1625 	sc->sc_uid = kauth_cred_getuid(l->l_cred);
1626 	sc->sc_euid = kauth_cred_geteuid(l->l_cred);
1627 	sc->sc_gid = kauth_cred_getgid(l->l_cred);
1628 	sc->sc_egid = kauth_cred_getegid(l->l_cred);
1629 	sc->sc_ngroups = kauth_cred_ngroups(l->l_cred);
1630 
1631 	for (int i = 0; i < sc->sc_ngroups; i++)
1632 		sc->sc_groups[i] = kauth_cred_group(l->l_cred, i);
1633 
1634 	return m_add(control, m);
1635 }
1636 
1637 /*
1638  * Do a mark-sweep GC of files in the system, to free up any which are
1639  * caught in flight to an about-to-be-closed socket.  Additionally,
1640  * process deferred file closures.
1641  */
1642 static void
1643 unp_gc(file_t *dp)
1644 {
1645 	extern	struct domain unixdomain;
1646 	file_t *fp, *np;
1647 	struct socket *so, *so1;
1648 	u_int i, oflags, rflags;
1649 	bool didwork;
1650 
1651 	KASSERT(curlwp == unp_thread_lwp);
1652 	KASSERT(mutex_owned(&filelist_lock));
1653 
1654 	/*
1655 	 * First, process deferred file closures.
1656 	 */
1657 	while (!SLIST_EMPTY(&unp_thread_discard)) {
1658 		fp = SLIST_FIRST(&unp_thread_discard);
1659 		KASSERT(fp->f_unpcount > 0);
1660 		KASSERT(fp->f_count > 0);
1661 		KASSERT(fp->f_msgcount > 0);
1662 		KASSERT(fp->f_count >= fp->f_unpcount);
1663 		KASSERT(fp->f_count >= fp->f_msgcount);
1664 		KASSERT(fp->f_msgcount >= fp->f_unpcount);
1665 		SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist);
1666 		i = fp->f_unpcount;
1667 		fp->f_unpcount = 0;
1668 		mutex_exit(&filelist_lock);
1669 		for (; i != 0; i--) {
1670 			unp_discard_now(fp);
1671 		}
1672 		mutex_enter(&filelist_lock);
1673 	}
1674 
1675 	/*
1676 	 * Clear mark bits.  Ensure that we don't consider new files
1677 	 * entering the file table during this loop (they will not have
1678 	 * FSCAN set).
1679 	 */
1680 	unp_defer = 0;
1681 	LIST_FOREACH(fp, &filehead, f_list) {
1682 		for (oflags = fp->f_flag;; oflags = rflags) {
1683 			rflags = atomic_cas_uint(&fp->f_flag, oflags,
1684 			    (oflags | FSCAN) & ~(FMARK|FDEFER));
1685 			if (__predict_true(oflags == rflags)) {
1686 				break;
1687 			}
1688 		}
1689 	}
1690 
1691 	/*
1692 	 * Iterate over the set of sockets, marking ones believed (based on
1693 	 * refcount) to be referenced from a process, and marking for rescan
1694 	 * sockets which are queued on a socket.  Recan continues descending
1695 	 * and searching for sockets referenced by sockets (FDEFER), until
1696 	 * there are no more socket->socket references to be discovered.
1697 	 */
1698 	do {
1699 		didwork = false;
1700 		for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1701 			KASSERT(mutex_owned(&filelist_lock));
1702 			np = LIST_NEXT(fp, f_list);
1703 			mutex_enter(&fp->f_lock);
1704 			if ((fp->f_flag & FDEFER) != 0) {
1705 				atomic_and_uint(&fp->f_flag, ~FDEFER);
1706 				unp_defer--;
1707 				KASSERT(fp->f_count != 0);
1708 			} else {
1709 				if (fp->f_count == 0 ||
1710 				    (fp->f_flag & FMARK) != 0 ||
1711 				    fp->f_count == fp->f_msgcount ||
1712 				    fp->f_unpcount != 0) {
1713 					mutex_exit(&fp->f_lock);
1714 					continue;
1715 				}
1716 			}
1717 			atomic_or_uint(&fp->f_flag, FMARK);
1718 
1719 			if (fp->f_type != DTYPE_SOCKET ||
1720 			    (so = fp->f_socket) == NULL ||
1721 			    so->so_proto->pr_domain != &unixdomain ||
1722 			    (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1723 				mutex_exit(&fp->f_lock);
1724 				continue;
1725 			}
1726 
1727 			/* Gain file ref, mark our position, and unlock. */
1728 			didwork = true;
1729 			LIST_INSERT_AFTER(fp, dp, f_list);
1730 			fp->f_count++;
1731 			mutex_exit(&fp->f_lock);
1732 			mutex_exit(&filelist_lock);
1733 
1734 			/*
1735 			 * Mark files referenced from sockets queued on the
1736 			 * accept queue as well.
1737 			 */
1738 			solock(so);
1739 			unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1740 			if ((so->so_options & SO_ACCEPTCONN) != 0) {
1741 				TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1742 					unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1743 				}
1744 				TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1745 					unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1746 				}
1747 			}
1748 			sounlock(so);
1749 
1750 			/* Re-lock and restart from where we left off. */
1751 			closef(fp);
1752 			mutex_enter(&filelist_lock);
1753 			np = LIST_NEXT(dp, f_list);
1754 			LIST_REMOVE(dp, f_list);
1755 		}
1756 		/*
1757 		 * Bail early if we did nothing in the loop above.  Could
1758 		 * happen because of concurrent activity causing unp_defer
1759 		 * to get out of sync.
1760 		 */
1761 	} while (unp_defer != 0 && didwork);
1762 
1763 	/*
1764 	 * Sweep pass.
1765 	 *
1766 	 * We grab an extra reference to each of the files that are
1767 	 * not otherwise accessible and then free the rights that are
1768 	 * stored in messages on them.
1769 	 */
1770 	for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1771 		KASSERT(mutex_owned(&filelist_lock));
1772 		np = LIST_NEXT(fp, f_list);
1773 		mutex_enter(&fp->f_lock);
1774 
1775 		/*
1776 		 * Ignore non-sockets.
1777 		 * Ignore dead sockets, or sockets with pending close.
1778 		 * Ignore sockets obviously referenced elsewhere.
1779 		 * Ignore sockets marked as referenced by our scan.
1780 		 * Ignore new sockets that did not exist during the scan.
1781 		 */
1782 		if (fp->f_type != DTYPE_SOCKET ||
1783 		    fp->f_count == 0 || fp->f_unpcount != 0 ||
1784 		    fp->f_count != fp->f_msgcount ||
1785 		    (fp->f_flag & (FMARK | FSCAN)) != FSCAN) {
1786 			mutex_exit(&fp->f_lock);
1787 			continue;
1788 		}
1789 
1790 		/* Gain file ref, mark our position, and unlock. */
1791 		LIST_INSERT_AFTER(fp, dp, f_list);
1792 		fp->f_count++;
1793 		mutex_exit(&fp->f_lock);
1794 		mutex_exit(&filelist_lock);
1795 
1796 		/*
1797 		 * Flush all data from the socket's receive buffer.
1798 		 * This will cause files referenced only by the
1799 		 * socket to be queued for close.
1800 		 */
1801 		so = fp->f_socket;
1802 		solock(so);
1803 		sorflush(so);
1804 		sounlock(so);
1805 
1806 		/* Re-lock and restart from where we left off. */
1807 		closef(fp);
1808 		mutex_enter(&filelist_lock);
1809 		np = LIST_NEXT(dp, f_list);
1810 		LIST_REMOVE(dp, f_list);
1811 	}
1812 }
1813 
1814 /*
1815  * Garbage collector thread.  While SCM_RIGHTS messages are in transit,
1816  * wake once per second to garbage collect.  Run continually while we
1817  * have deferred closes to process.
1818  */
1819 static void
1820 unp_thread(void *cookie)
1821 {
1822 	file_t *dp;
1823 
1824 	/* Allocate a dummy file for our scans. */
1825 	if ((dp = fgetdummy()) == NULL) {
1826 		panic("unp_thread");
1827 	}
1828 
1829 	mutex_enter(&filelist_lock);
1830 	for (;;) {
1831 		KASSERT(mutex_owned(&filelist_lock));
1832 		if (SLIST_EMPTY(&unp_thread_discard)) {
1833 			if (unp_rights != 0) {
1834 				(void)cv_timedwait(&unp_thread_cv,
1835 				    &filelist_lock, hz);
1836 			} else {
1837 				cv_wait(&unp_thread_cv, &filelist_lock);
1838 			}
1839 		}
1840 		unp_gc(dp);
1841 	}
1842 	/* NOTREACHED */
1843 }
1844 
1845 /*
1846  * Kick the garbage collector into action if there is something for
1847  * it to process.
1848  */
1849 static void
1850 unp_thread_kick(void)
1851 {
1852 
1853 	if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) {
1854 		mutex_enter(&filelist_lock);
1855 		cv_signal(&unp_thread_cv);
1856 		mutex_exit(&filelist_lock);
1857 	}
1858 }
1859 
1860 void
1861 unp_dispose(struct mbuf *m)
1862 {
1863 
1864 	if (m)
1865 		unp_scan(m, unp_discard_later, 1);
1866 }
1867 
1868 void
1869 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard)
1870 {
1871 	struct mbuf *m;
1872 	file_t **rp, *fp;
1873 	struct cmsghdr *cm;
1874 	int i, qfds;
1875 
1876 	while (m0) {
1877 		for (m = m0; m; m = m->m_next) {
1878 			if (m->m_type != MT_CONTROL ||
1879 			    m->m_len < sizeof(*cm)) {
1880 			    	continue;
1881 			}
1882 			cm = mtod(m, struct cmsghdr *);
1883 			if (cm->cmsg_level != SOL_SOCKET ||
1884 			    cm->cmsg_type != SCM_RIGHTS)
1885 				continue;
1886 			qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1887 			    / sizeof(file_t *);
1888 			rp = (file_t **)CMSG_DATA(cm);
1889 			for (i = 0; i < qfds; i++) {
1890 				fp = *rp;
1891 				if (discard) {
1892 					*rp = 0;
1893 				}
1894 				(*op)(fp);
1895 				rp++;
1896 			}
1897 		}
1898 		m0 = m0->m_nextpkt;
1899 	}
1900 }
1901 
1902 void
1903 unp_mark(file_t *fp)
1904 {
1905 
1906 	if (fp == NULL)
1907 		return;
1908 
1909 	/* If we're already deferred, don't screw up the defer count */
1910 	mutex_enter(&fp->f_lock);
1911 	if (fp->f_flag & (FMARK | FDEFER)) {
1912 		mutex_exit(&fp->f_lock);
1913 		return;
1914 	}
1915 
1916 	/*
1917 	 * Minimize the number of deferrals...  Sockets are the only type of
1918 	 * file which can hold references to another file, so just mark
1919 	 * other files, and defer unmarked sockets for the next pass.
1920 	 */
1921 	if (fp->f_type == DTYPE_SOCKET) {
1922 		unp_defer++;
1923 		KASSERT(fp->f_count != 0);
1924 		atomic_or_uint(&fp->f_flag, FDEFER);
1925 	} else {
1926 		atomic_or_uint(&fp->f_flag, FMARK);
1927 	}
1928 	mutex_exit(&fp->f_lock);
1929 }
1930 
1931 static void
1932 unp_discard_now(file_t *fp)
1933 {
1934 
1935 	if (fp == NULL)
1936 		return;
1937 
1938 	KASSERT(fp->f_count > 0);
1939 	KASSERT(fp->f_msgcount > 0);
1940 
1941 	mutex_enter(&fp->f_lock);
1942 	fp->f_msgcount--;
1943 	mutex_exit(&fp->f_lock);
1944 	atomic_dec_uint(&unp_rights);
1945 	(void)closef(fp);
1946 }
1947 
1948 static void
1949 unp_discard_later(file_t *fp)
1950 {
1951 
1952 	if (fp == NULL)
1953 		return;
1954 
1955 	KASSERT(fp->f_count > 0);
1956 	KASSERT(fp->f_msgcount > 0);
1957 
1958 	mutex_enter(&filelist_lock);
1959 	if (fp->f_unpcount++ == 0) {
1960 		SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist);
1961 	}
1962 	mutex_exit(&filelist_lock);
1963 }
1964 
1965 const struct pr_usrreqs unp_usrreqs = {
1966 	.pr_attach	= unp_attach,
1967 	.pr_detach	= unp_detach,
1968 	.pr_accept	= unp_accept,
1969 	.pr_bind	= unp_bind,
1970 	.pr_listen	= unp_listen,
1971 	.pr_connect	= unp_connect,
1972 	.pr_connect2	= unp_connect2,
1973 	.pr_disconnect	= unp_disconnect,
1974 	.pr_shutdown	= unp_shutdown,
1975 	.pr_abort	= unp_abort,
1976 	.pr_ioctl	= unp_ioctl,
1977 	.pr_stat	= unp_stat,
1978 	.pr_peeraddr	= unp_peeraddr,
1979 	.pr_sockaddr	= unp_sockaddr,
1980 	.pr_rcvd	= unp_rcvd,
1981 	.pr_recvoob	= unp_recvoob,
1982 	.pr_send	= unp_send,
1983 	.pr_sendoob	= unp_sendoob,
1984 	.pr_generic	= unp_usrreq,
1985 };
1986