xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision aa6ac96e01825b3efcab953441f85adbf9815e0f)
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 /* NOTE: unp_token MUST be held */
141 static __inline void
142 unp_reference(struct unpcb *unp)
143 {
144 	atomic_add_int(&unp->unp_refcnt, 1);
145 }
146 
147 /* NOTE: unp_token MUST be held */
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 = msg->base.nm_so->so_pcb;
229 	if (!UNP_ISATTACHED(unp)) {
230 		error = EINVAL;
231 	} else {
232 		struct unpcb *unp2 = unp->unp_conn;
233 
234 		/*
235 		 * Pass back name of connected socket,
236 		 * if it was bound and we are still connected
237 		 * (our peer may have closed already!).
238 		 */
239 		if (unp2 && unp2->unp_addr) {
240 			unp_reference(unp2);
241 			*msg->accept.nm_nam = dup_sockaddr(
242 				(struct sockaddr *)unp2->unp_addr);
243 			unp_free(unp2);
244 		} else {
245 			*msg->accept.nm_nam = dup_sockaddr(&sun_noname);
246 		}
247 		error = 0;
248 	}
249 	lwkt_reltoken(&unp_token);
250 	lwkt_replymsg(&msg->lmsg, error);
251 }
252 
253 static void
254 uipc_attach(netmsg_t msg)
255 {
256 	struct unpcb *unp;
257 	int error;
258 
259 	lwkt_gettoken(&unp_token);
260 	unp = msg->base.nm_so->so_pcb;
261 	KASSERT(unp == NULL, ("double unp attach"));
262 	error = unp_attach(msg->base.nm_so, msg->attach.nm_ai);
263 	lwkt_reltoken(&unp_token);
264 	lwkt_replymsg(&msg->lmsg, error);
265 }
266 
267 static void
268 uipc_bind(netmsg_t msg)
269 {
270 	struct unpcb *unp;
271 	int error;
272 
273 	lwkt_gettoken(&unp_token);
274 	unp = msg->base.nm_so->so_pcb;
275 	if (UNP_ISATTACHED(unp))
276 		error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td);
277 	else
278 		error = EINVAL;
279 	lwkt_reltoken(&unp_token);
280 	lwkt_replymsg(&msg->lmsg, error);
281 }
282 
283 static void
284 uipc_connect(netmsg_t msg)
285 {
286 	int error;
287 
288 	error = unp_connect(msg->base.nm_so, msg->connect.nm_nam,
289 	    msg->connect.nm_td);
290 	lwkt_replymsg(&msg->lmsg, error);
291 }
292 
293 static void
294 uipc_connect2(netmsg_t msg)
295 {
296 	int error;
297 
298 	error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2);
299 	lwkt_replymsg(&msg->lmsg, error);
300 }
301 
302 /* control is EOPNOTSUPP */
303 
304 static void
305 uipc_detach(netmsg_t msg)
306 {
307 	struct unpcb *unp;
308 	int error;
309 
310 	lwkt_gettoken(&unp_token);
311 	unp = unp_getsocktoken(msg->base.nm_so);
312 
313 	if (UNP_ISATTACHED(unp)) {
314 		unp_setflags(unp, UNP_DETACHED);
315 		unp_drop(unp, 0);
316 		unp_free(unp);
317 		error = 0;
318 	} else {
319 		error = EINVAL;
320 	}
321 
322 	unp_reltoken(unp);
323 	lwkt_reltoken(&unp_token);
324 
325 	lwkt_replymsg(&msg->lmsg, error);
326 }
327 
328 static void
329 uipc_disconnect(netmsg_t msg)
330 {
331 	struct unpcb *unp;
332 	int error;
333 
334 	lwkt_gettoken(&unp_token);
335 	unp = unp_getsocktoken(msg->base.nm_so);
336 
337 	if (UNP_ISATTACHED(unp)) {
338 		unp_disconnect(unp, 0);
339 		error = 0;
340 	} else {
341 		error = EINVAL;
342 	}
343 
344 	unp_reltoken(unp);
345 	lwkt_reltoken(&unp_token);
346 
347 	lwkt_replymsg(&msg->lmsg, error);
348 }
349 
350 static void
351 uipc_listen(netmsg_t msg)
352 {
353 	struct unpcb *unp;
354 	int error;
355 
356 	lwkt_gettoken(&unp_token);
357 	unp = msg->base.nm_so->so_pcb;
358 	if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL)
359 		error = EINVAL;
360 	else
361 		error = unp_listen(unp, msg->listen.nm_td);
362 	lwkt_reltoken(&unp_token);
363 	lwkt_replymsg(&msg->lmsg, error);
364 }
365 
366 static void
367 uipc_peeraddr(netmsg_t msg)
368 {
369 	struct unpcb *unp;
370 	int error;
371 
372 	lwkt_gettoken(&unp_token);
373 	unp = msg->base.nm_so->so_pcb;
374 	if (!UNP_ISATTACHED(unp)) {
375 		error = EINVAL;
376 	} else if (unp->unp_conn && unp->unp_conn->unp_addr) {
377 		struct unpcb *unp2 = unp->unp_conn;
378 
379 		unp_reference(unp2);
380 		*msg->peeraddr.nm_nam = dup_sockaddr(
381 				(struct sockaddr *)unp2->unp_addr);
382 		unp_free(unp2);
383 		error = 0;
384 	} else {
385 		/*
386 		 * XXX: It seems that this test always fails even when
387 		 * connection is established.  So, this else clause is
388 		 * added as workaround to return PF_LOCAL sockaddr.
389 		 */
390 		*msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname);
391 		error = 0;
392 	}
393 	lwkt_reltoken(&unp_token);
394 	lwkt_replymsg(&msg->lmsg, error);
395 }
396 
397 static void
398 uipc_rcvd(netmsg_t msg)
399 {
400 	struct unpcb *unp, *unp2;
401 	struct socket *so;
402 	struct socket *so2;
403 	int error;
404 
405 	/*
406 	 * so_pcb is only modified with both the global and the unp
407 	 * pool token held.
408 	 */
409 	so = msg->base.nm_so;
410 	unp = unp_getsocktoken(so);
411 
412 	if (!UNP_ISATTACHED(unp)) {
413 		error = EINVAL;
414 		goto done;
415 	}
416 
417 	switch (so->so_type) {
418 	case SOCK_DGRAM:
419 		panic("uipc_rcvd DGRAM?");
420 		/*NOTREACHED*/
421 	case SOCK_STREAM:
422 	case SOCK_SEQPACKET:
423 		if (unp->unp_conn == NULL)
424 			break;
425 		unp2 = unp->unp_conn;	/* protected by pool token */
426 
427 		/*
428 		 * Because we are transfering mbufs directly to the
429 		 * peer socket we have to use SSB_STOP on the sender
430 		 * to prevent it from building up infinite mbufs.
431 		 *
432 		 * As in several places in this module w ehave to ref unp2
433 		 * to ensure that it does not get ripped out from under us
434 		 * if we block on the so2 token or in sowwakeup().
435 		 */
436 		so2 = unp2->unp_socket;
437 		unp_reference(unp2);
438 		lwkt_gettoken(&so2->so_rcv.ssb_token);
439 		if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
440 		    so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
441 		) {
442 			atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP);
443 
444 			sowwakeup(so2);
445 		}
446 		lwkt_reltoken(&so2->so_rcv.ssb_token);
447 		unp_free(unp2);
448 		break;
449 	default:
450 		panic("uipc_rcvd unknown socktype");
451 		/*NOTREACHED*/
452 	}
453 	error = 0;
454 done:
455 	unp_reltoken(unp);
456 	lwkt_replymsg(&msg->lmsg, error);
457 }
458 
459 /* pru_rcvoob is EOPNOTSUPP */
460 
461 static void
462 uipc_send(netmsg_t msg)
463 {
464 	struct unpcb *unp, *unp2;
465 	struct socket *so;
466 	struct socket *so2;
467 	struct mbuf *control;
468 	struct mbuf *m;
469 	int error = 0;
470 
471 	so = msg->base.nm_so;
472 	control = msg->send.nm_control;
473 	m = msg->send.nm_m;
474 
475 	/*
476 	 * so_pcb is only modified with both the global and the unp
477 	 * pool token held.
478 	 */
479 	so = msg->base.nm_so;
480 	unp = unp_getsocktoken(so);
481 
482 	if (!UNP_ISATTACHED(unp)) {
483 		error = EINVAL;
484 		goto release;
485 	}
486 
487 	if (msg->send.nm_flags & PRUS_OOB) {
488 		error = EOPNOTSUPP;
489 		goto release;
490 	}
491 
492 	wakeup_start_delayed();
493 
494 	if (control && (error = unp_internalize(control, msg->send.nm_td)))
495 		goto release;
496 
497 	switch (so->so_type) {
498 	case SOCK_DGRAM:
499 	{
500 		struct sockaddr *from;
501 
502 		if (msg->send.nm_addr) {
503 			if (unp->unp_conn) {
504 				error = EISCONN;
505 				break;
506 			}
507 			error = unp_find_lockref(msg->send.nm_addr,
508 			    msg->send.nm_td, so->so_type, &unp2);
509 			if (error)
510 				break;
511 			/*
512 			 * NOTE:
513 			 * unp2 is locked and referenced.
514 			 *
515 			 * We could unlock unp2 now, since it was checked
516 			 * and referenced.
517 			 */
518 			unp_reltoken(unp2);
519 		} else {
520 			if (unp->unp_conn == NULL) {
521 				error = ENOTCONN;
522 				break;
523 			}
524 			/* XXX racy. */
525 			unp2 = unp->unp_conn;
526 			unp_reference(unp2);
527 		}
528 		/* NOTE: unp2 is referenced. */
529 		so2 = unp2->unp_socket;
530 
531 		if (unp->unp_addr)
532 			from = (struct sockaddr *)unp->unp_addr;
533 		else
534 			from = &sun_noname;
535 
536 		lwkt_gettoken(&so2->so_rcv.ssb_token);
537 		if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
538 			sorwakeup(so2);
539 			m = NULL;
540 			control = NULL;
541 		} else {
542 			error = ENOBUFS;
543 		}
544 		lwkt_reltoken(&so2->so_rcv.ssb_token);
545 
546 		unp_free(unp2);
547 		break;
548 	}
549 
550 	case SOCK_STREAM:
551 	case SOCK_SEQPACKET:
552 		/* Connect if not connected yet. */
553 		/*
554 		 * Note: A better implementation would complain
555 		 * if not equal to the peer's address.
556 		 */
557 		if (!(so->so_state & SS_ISCONNECTED)) {
558 			if (msg->send.nm_addr) {
559 				error = unp_connect(so,
560 						    msg->send.nm_addr,
561 						    msg->send.nm_td);
562 				if (error)
563 					break;	/* XXX */
564 			} else {
565 				error = ENOTCONN;
566 				break;
567 			}
568 		}
569 
570 		if (so->so_state & SS_CANTSENDMORE) {
571 			error = EPIPE;
572 			break;
573 		}
574 		if (unp->unp_conn == NULL)
575 			panic("uipc_send connected but no connection?");
576 		unp2 = unp->unp_conn;
577 		so2 = unp2->unp_socket;
578 
579 		unp_reference(unp2);
580 
581 		/*
582 		 * Send to paired receive port, and then reduce
583 		 * send buffer hiwater marks to maintain backpressure.
584 		 * Wake up readers.
585 		 */
586 		lwkt_gettoken(&so2->so_rcv.ssb_token);
587 		if (control) {
588 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
589 				control = NULL;
590 				m = NULL;
591 			}
592 		} else if (so->so_type == SOCK_SEQPACKET) {
593 			sbappendrecord(&so2->so_rcv.sb, m);
594 			m = NULL;
595 		} else {
596 			sbappend(&so2->so_rcv.sb, m);
597 			m = NULL;
598 		}
599 
600 		/*
601 		 * Because we are transfering mbufs directly to the
602 		 * peer socket we have to use SSB_STOP on the sender
603 		 * to prevent it from building up infinite mbufs.
604 		 */
605 		if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
606 		    so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
607 		) {
608 			atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP);
609 		}
610 		lwkt_reltoken(&so2->so_rcv.ssb_token);
611 		sorwakeup(so2);
612 
613 		unp_free(unp2);
614 		break;
615 
616 	default:
617 		panic("uipc_send unknown socktype");
618 	}
619 
620 	/*
621 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
622 	 */
623 	if (msg->send.nm_flags & PRUS_EOF) {
624 		socantsendmore(so);
625 		unp_shutdown(unp);
626 	}
627 
628 	if (control && error != 0)
629 		unp_dispose(control);
630 release:
631 	unp_reltoken(unp);
632 	wakeup_end_delayed();
633 
634 	if (control)
635 		m_freem(control);
636 	if (m)
637 		m_freem(m);
638 	lwkt_replymsg(&msg->lmsg, error);
639 }
640 
641 /*
642  * MPSAFE
643  */
644 static void
645 uipc_sense(netmsg_t msg)
646 {
647 	struct unpcb *unp;
648 	struct socket *so;
649 	struct stat *sb;
650 	int error;
651 
652 	so = msg->base.nm_so;
653 	sb = msg->sense.nm_stat;
654 
655 	/*
656 	 * so_pcb is only modified with both the global and the unp
657 	 * pool token held.
658 	 */
659 	unp = unp_getsocktoken(so);
660 
661 	if (!UNP_ISATTACHED(unp)) {
662 		error = EINVAL;
663 		goto done;
664 	}
665 
666 	sb->st_blksize = so->so_snd.ssb_hiwat;
667 	sb->st_dev = NOUDEV;
668 	if (unp->unp_ino == 0) {	/* make up a non-zero inode number */
669 		spin_lock(&unp_ino_spin);
670 		unp->unp_ino = unp_ino++;
671 		spin_unlock(&unp_ino_spin);
672 	}
673 	sb->st_ino = unp->unp_ino;
674 	error = 0;
675 done:
676 	unp_reltoken(unp);
677 	lwkt_replymsg(&msg->lmsg, error);
678 }
679 
680 static void
681 uipc_shutdown(netmsg_t msg)
682 {
683 	struct socket *so;
684 	struct unpcb *unp;
685 	int error;
686 
687 	/*
688 	 * so_pcb is only modified with both the global and the unp
689 	 * pool token held.
690 	 */
691 	so = msg->base.nm_so;
692 	unp = unp_getsocktoken(so);
693 
694 	if (UNP_ISATTACHED(unp)) {
695 		socantsendmore(so);
696 		unp_shutdown(unp);
697 		error = 0;
698 	} else {
699 		error = EINVAL;
700 	}
701 
702 	unp_reltoken(unp);
703 	lwkt_replymsg(&msg->lmsg, error);
704 }
705 
706 static void
707 uipc_sockaddr(netmsg_t msg)
708 {
709 	struct socket *so;
710 	struct unpcb *unp;
711 	int error;
712 
713 	/*
714 	 * so_pcb is only modified with both the global and the unp
715 	 * pool token held.
716 	 */
717 	so = msg->base.nm_so;
718 	unp = unp_getsocktoken(so);
719 
720 	if (UNP_ISATTACHED(unp)) {
721 		if (unp->unp_addr) {
722 			*msg->sockaddr.nm_nam =
723 				dup_sockaddr((struct sockaddr *)unp->unp_addr);
724 		}
725 		error = 0;
726 	} else {
727 		error = EINVAL;
728 	}
729 
730 	unp_reltoken(unp);
731 	lwkt_replymsg(&msg->lmsg, error);
732 }
733 
734 struct pr_usrreqs uipc_usrreqs = {
735 	.pru_abort = uipc_abort,
736 	.pru_accept = uipc_accept,
737 	.pru_attach = uipc_attach,
738 	.pru_bind = uipc_bind,
739 	.pru_connect = uipc_connect,
740 	.pru_connect2 = uipc_connect2,
741 	.pru_control = pr_generic_notsupp,
742 	.pru_detach = uipc_detach,
743 	.pru_disconnect = uipc_disconnect,
744 	.pru_listen = uipc_listen,
745 	.pru_peeraddr = uipc_peeraddr,
746 	.pru_rcvd = uipc_rcvd,
747 	.pru_rcvoob = pr_generic_notsupp,
748 	.pru_send = uipc_send,
749 	.pru_sense = uipc_sense,
750 	.pru_shutdown = uipc_shutdown,
751 	.pru_sockaddr = uipc_sockaddr,
752 	.pru_sosend = sosend,
753 	.pru_soreceive = soreceive
754 };
755 
756 void
757 uipc_ctloutput(netmsg_t msg)
758 {
759 	struct socket *so;
760 	struct sockopt *sopt;
761 	struct unpcb *unp;
762 	int error = 0;
763 
764 	lwkt_gettoken(&unp_token);
765 	so = msg->base.nm_so;
766 	sopt = msg->ctloutput.nm_sopt;
767 	unp = so->so_pcb;
768 
769 	switch (sopt->sopt_dir) {
770 	case SOPT_GET:
771 		switch (sopt->sopt_name) {
772 		case LOCAL_PEERCRED:
773 			if (unp->unp_flags & UNP_HAVEPC)
774 				soopt_from_kbuf(sopt, &unp->unp_peercred,
775 						sizeof(unp->unp_peercred));
776 			else {
777 				if (so->so_type == SOCK_STREAM)
778 					error = ENOTCONN;
779 				else if (so->so_type == SOCK_SEQPACKET)
780 					error = ENOTCONN;
781 				else
782 					error = EINVAL;
783 			}
784 			break;
785 		default:
786 			error = EOPNOTSUPP;
787 			break;
788 		}
789 		break;
790 	case SOPT_SET:
791 	default:
792 		error = EOPNOTSUPP;
793 		break;
794 	}
795 	lwkt_reltoken(&unp_token);
796 	lwkt_replymsg(&msg->lmsg, error);
797 }
798 
799 /*
800  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
801  * for stream sockets, although the total for sender and receiver is
802  * actually only PIPSIZ.
803  *
804  * Datagram sockets really use the sendspace as the maximum datagram size,
805  * and don't really want to reserve the sendspace.  Their recvspace should
806  * be large enough for at least one max-size datagram plus address.
807  *
808  * We want the local send/recv space to be significant larger then lo0's
809  * mtu of 16384.
810  */
811 #ifndef PIPSIZ
812 #define	PIPSIZ	57344
813 #endif
814 static u_long	unpst_sendspace = PIPSIZ;
815 static u_long	unpst_recvspace = PIPSIZ;
816 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
817 static u_long	unpdg_recvspace = 4*1024;
818 
819 static int	unp_rights;			/* file descriptors in flight */
820 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin, "unp_spin");
821 
822 SYSCTL_DECL(_net_local_seqpacket);
823 SYSCTL_DECL(_net_local_stream);
824 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
825     &unpst_sendspace, 0, "Size of stream socket send buffer");
826 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
827     &unpst_recvspace, 0, "Size of stream socket receive buffer");
828 
829 SYSCTL_DECL(_net_local_dgram);
830 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
831     &unpdg_sendspace, 0, "Max datagram socket size");
832 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
833     &unpdg_recvspace, 0, "Size of datagram socket receive buffer");
834 
835 SYSCTL_DECL(_net_local);
836 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
837    "File descriptors in flight");
838 
839 static int
840 unp_attach(struct socket *so, struct pru_attach_info *ai)
841 {
842 	struct unpcb *unp;
843 	int error;
844 
845 	lwkt_gettoken(&unp_token);
846 
847 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
848 		switch (so->so_type) {
849 		case SOCK_STREAM:
850 		case SOCK_SEQPACKET:
851 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
852 					  ai->sb_rlimit);
853 			break;
854 
855 		case SOCK_DGRAM:
856 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
857 					  ai->sb_rlimit);
858 			break;
859 
860 		default:
861 			panic("unp_attach");
862 		}
863 		if (error)
864 			goto failed;
865 	}
866 
867 	/*
868 	 * In order to support sendfile we have to set either SSB_STOPSUPP
869 	 * or SSB_PREALLOC.  Unix domain sockets use the SSB_STOP flow
870 	 * control mechanism.
871 	 */
872 	if (so->so_type == SOCK_STREAM) {
873 		atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP);
874 		atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP);
875 	}
876 
877 	unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK);
878 	if (unp == NULL) {
879 		error = ENOBUFS;
880 		goto failed;
881 	}
882 	unp->unp_refcnt = 1;
883 	unp->unp_gencnt = ++unp_gencnt;
884 	unp_count++;
885 	LIST_INIT(&unp->unp_refs);
886 	unp->unp_socket = so;
887 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
888 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
889 			 : &unp_shead, unp, unp_link);
890 	so->so_pcb = (caddr_t)unp;
891 	soreference(so);
892 	error = 0;
893 failed:
894 	lwkt_reltoken(&unp_token);
895 	return error;
896 }
897 
898 static void
899 unp_detach(struct unpcb *unp)
900 {
901 	struct socket *so;
902 
903 	lwkt_gettoken(&unp_token);
904 	lwkt_getpooltoken(unp);
905 
906 	LIST_REMOVE(unp, unp_link);	/* both tokens required */
907 	unp->unp_gencnt = ++unp_gencnt;
908 	--unp_count;
909 	if (unp->unp_vnode) {
910 		unp->unp_vnode->v_socket = NULL;
911 		vrele(unp->unp_vnode);
912 		unp->unp_vnode = NULL;
913 	}
914 	soisdisconnected(unp->unp_socket);
915 	so = unp->unp_socket;
916 	soreference(so);		/* for delayed sorflush */
917 	KKASSERT(so->so_pcb == unp);
918 	so->so_pcb = NULL;		/* both tokens required */
919 	unp->unp_socket = NULL;
920 	sofree(so);		/* remove pcb ref */
921 
922 	if (unp_rights) {
923 		/*
924 		 * Normally the receive buffer is flushed later,
925 		 * in sofree, but if our receive buffer holds references
926 		 * to descriptors that are now garbage, we will dispose
927 		 * of those descriptor references after the garbage collector
928 		 * gets them (resulting in a "panic: closef: count < 0").
929 		 */
930 		sorflush(so);
931 		unp_gc();
932 	}
933 	sofree(so);
934 	lwkt_relpooltoken(unp);
935 	lwkt_reltoken(&unp_token);
936 
937 	KASSERT(unp->unp_conn == NULL, ("unp is still connected"));
938 	KASSERT(LIST_EMPTY(&unp->unp_refs), ("unp still has references"));
939 
940 	if (unp->unp_addr)
941 		kfree(unp->unp_addr, M_SONAME);
942 	kfree(unp, M_UNPCB);
943 }
944 
945 static int
946 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
947 {
948 	struct proc *p = td->td_proc;
949 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
950 	struct vnode *vp;
951 	struct vattr vattr;
952 	int error, namelen;
953 	struct nlookupdata nd;
954 	char buf[SOCK_MAXADDRLEN];
955 
956 	lwkt_gettoken(&unp_token);
957 	if (unp->unp_vnode != NULL) {
958 		error = EINVAL;
959 		goto failed;
960 	}
961 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
962 	if (namelen <= 0) {
963 		error = EINVAL;
964 		goto failed;
965 	}
966 	strncpy(buf, soun->sun_path, namelen);
967 	buf[namelen] = 0;	/* null-terminate the string */
968 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
969 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
970 	if (error == 0)
971 		error = nlookup(&nd);
972 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
973 		error = EADDRINUSE;
974 	if (error)
975 		goto done;
976 
977 	VATTR_NULL(&vattr);
978 	vattr.va_type = VSOCK;
979 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
980 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
981 	if (error == 0) {
982 		if (unp->unp_vnode == NULL) {
983 			vp->v_socket = unp->unp_socket;
984 			unp->unp_vnode = vp;
985 			unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
986 			vn_unlock(vp);
987 		} else {
988 			vput(vp);		/* late race */
989 			error = EINVAL;
990 		}
991 	}
992 done:
993 	nlookup_done(&nd);
994 failed:
995 	lwkt_reltoken(&unp_token);
996 	return (error);
997 }
998 
999 static int
1000 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1001 {
1002 	struct unpcb *unp, *unp2;
1003 	int error, flags = 0;
1004 
1005 	lwkt_gettoken(&unp_token);
1006 
1007 	unp = unp_getsocktoken(so);
1008 	if (!UNP_ISATTACHED(unp)) {
1009 		error = EINVAL;
1010 		goto failed;
1011 	}
1012 
1013 	if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) {
1014 		error = EISCONN;
1015 		goto failed;
1016 	}
1017 
1018 	flags = UNP_CONNECTING;
1019 	unp_setflags(unp, flags);
1020 
1021 	error = unp_find_lockref(nam, td, so->so_type, &unp2);
1022 	if (error)
1023 		goto failed;
1024 	/*
1025 	 * NOTE:
1026 	 * unp2 is locked and referenced.
1027 	 */
1028 
1029 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1030 		struct socket *so2, *so3;
1031 		struct unpcb *unp3;
1032 
1033 		so2 = unp2->unp_socket;
1034 		if (!(so2->so_options & SO_ACCEPTCONN) ||
1035 		    (so3 = sonewconn_faddr(so2, 0, NULL,
1036 		     TRUE /* keep ref */)) == NULL) {
1037 			error = ECONNREFUSED;
1038 			goto done;
1039 		}
1040 		/* so3 has a socket reference. */
1041 
1042 		unp3 = unp_getsocktoken(so3);
1043 		if (!UNP_ISATTACHED(unp3)) {
1044 			unp_reltoken(unp3);
1045 			/*
1046 			 * Already aborted; we only need to drop the
1047 			 * socket reference held by sonewconn_faddr().
1048 			 */
1049 			sofree(so3);
1050 			error = ECONNREFUSED;
1051 			goto done;
1052 		}
1053 		unp_reference(unp3);
1054 		/*
1055 		 * NOTE:
1056 		 * unp3 is locked and referenced.
1057 		 */
1058 
1059 		/*
1060 		 * Release so3 socket reference held by sonewconn_faddr().
1061 		 * Since we have referenced unp3, neither unp3 nor so3 will
1062 		 * be destroyed here.
1063 		 */
1064 		sofree(so3);
1065 
1066 		if (unp2->unp_addr != NULL) {
1067 			unp3->unp_addr = (struct sockaddr_un *)
1068 			    dup_sockaddr((struct sockaddr *)unp2->unp_addr);
1069 		}
1070 
1071 		/*
1072 		 * unp_peercred management:
1073 		 *
1074 		 * The connecter's (client's) credentials are copied
1075 		 * from its process structure at the time of connect()
1076 		 * (which is now).
1077 		 */
1078 		cru2x(td->td_proc->p_ucred, &unp3->unp_peercred);
1079 		unp_setflags(unp3, UNP_HAVEPC);
1080 		/*
1081 		 * The receiver's (server's) credentials are copied
1082 		 * from the unp_peercred member of socket on which the
1083 		 * former called listen(); unp_listen() cached that
1084 		 * process's credentials at that time so we can use
1085 		 * them now.
1086 		 */
1087 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1088 		    ("unp_connect: listener without cached peercred"));
1089 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1090 		    sizeof(unp->unp_peercred));
1091 		unp_setflags(unp, UNP_HAVEPC);
1092 
1093 		error = unp_connect_pair(unp, unp3);
1094 		if (error) {
1095 			/* XXX we need a better name */
1096 			soabort_oncpu(so3);
1097 		}
1098 
1099 		/* Done with unp3 */
1100 		unp_free(unp3);
1101 		unp_reltoken(unp3);
1102 	} else {
1103 		error = unp_connect_pair(unp, unp2);
1104 	}
1105 done:
1106 	unp_free(unp2);
1107 	unp_reltoken(unp2);
1108 failed:
1109 	if (flags)
1110 		unp_clrflags(unp, flags);
1111 	unp_reltoken(unp);
1112 
1113 	lwkt_reltoken(&unp_token);
1114 	return (error);
1115 }
1116 
1117 /*
1118  * Connect two unix domain sockets together.
1119  *
1120  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1121  *	 pool token also be held.
1122  */
1123 int
1124 unp_connect2(struct socket *so, struct socket *so2)
1125 {
1126 	struct unpcb *unp, *unp2;
1127 	int error;
1128 
1129 	lwkt_gettoken(&unp_token);
1130 	if (so2->so_type != so->so_type) {
1131 		lwkt_reltoken(&unp_token);
1132 		return (EPROTOTYPE);
1133 	}
1134 	unp = unp_getsocktoken(so);
1135 	unp2 = unp_getsocktoken(so2);
1136 
1137 	if (!UNP_ISATTACHED(unp)) {
1138 		error = EINVAL;
1139 		goto done;
1140 	}
1141 	if (!UNP_ISATTACHED(unp2)) {
1142 		error = ECONNREFUSED;
1143 		goto done;
1144 	}
1145 
1146 	if (unp->unp_conn != NULL) {
1147 		error = EISCONN;
1148 		goto done;
1149 	}
1150 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1151 	    unp2->unp_conn != NULL) {
1152 		error = EISCONN;
1153 		goto done;
1154 	}
1155 
1156 	error = unp_connect_pair(unp, unp2);
1157 done:
1158 	unp_reltoken(unp2);
1159 	unp_reltoken(unp);
1160 	lwkt_reltoken(&unp_token);
1161 	return (error);
1162 }
1163 
1164 /*
1165  * Disconnect a unix domain socket pair.
1166  *
1167  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1168  *	 pool token also be held.
1169  */
1170 static void
1171 unp_disconnect(struct unpcb *unp, int error)
1172 {
1173 	struct socket *so = unp->unp_socket;
1174 	struct unpcb *unp2;
1175 
1176 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
1177 	UNP_ASSERT_TOKEN_HELD(unp);
1178 
1179 	if (error)
1180 		so->so_error = error;
1181 
1182 	while ((unp2 = unp->unp_conn) != NULL) {
1183 		lwkt_getpooltoken(unp2);
1184 		if (unp2 == unp->unp_conn)
1185 			break;
1186 		lwkt_relpooltoken(unp2);
1187 	}
1188 	if (unp2 == NULL)
1189 		return;
1190 	/* unp2 is locked. */
1191 
1192 	KASSERT((unp2->unp_flags & UNP_DROPPED) == 0, ("unp2 was dropped"));
1193 
1194 	unp->unp_conn = NULL;
1195 
1196 	switch (so->so_type) {
1197 	case SOCK_DGRAM:
1198 		LIST_REMOVE(unp, unp_reflink);
1199 		soclrstate(so, SS_ISCONNECTED);
1200 		break;
1201 
1202 	case SOCK_STREAM:
1203 	case SOCK_SEQPACKET:
1204 		/*
1205 		 * Keep a reference before clearing the unp_conn
1206 		 * to avoid racing uipc_detach()/uipc_abort() in
1207 		 * other thread.
1208 		 */
1209 		unp_reference(unp2);
1210 		KASSERT(unp2->unp_conn == unp, ("unp_conn mismatch"));
1211 		unp2->unp_conn = NULL;
1212 
1213 		soisdisconnected(so);
1214 		soisdisconnected(unp2->unp_socket);
1215 
1216 		unp_free(unp2);
1217 		break;
1218 	}
1219 
1220 	lwkt_relpooltoken(unp2);
1221 }
1222 
1223 #ifdef notdef
1224 void
1225 unp_abort(struct unpcb *unp)
1226 {
1227 	lwkt_gettoken(&unp_token);
1228 	unp_free(unp);
1229 	lwkt_reltoken(&unp_token);
1230 }
1231 #endif
1232 
1233 static int
1234 prison_unpcb(struct thread *td, struct unpcb *unp)
1235 {
1236 	struct proc *p;
1237 
1238 	if (td == NULL)
1239 		return (0);
1240 	if ((p = td->td_proc) == NULL)
1241 		return (0);
1242 	if (!p->p_ucred->cr_prison)
1243 		return (0);
1244 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
1245 		return (0);
1246 	return (1);
1247 }
1248 
1249 static int
1250 unp_pcblist(SYSCTL_HANDLER_ARGS)
1251 {
1252 	int error, i, n;
1253 	struct unpcb *unp, **unp_list;
1254 	unp_gen_t gencnt;
1255 	struct unp_head *head;
1256 
1257 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1258 
1259 	KKASSERT(curproc != NULL);
1260 
1261 	/*
1262 	 * The process of preparing the PCB list is too time-consuming and
1263 	 * resource-intensive to repeat twice on every request.
1264 	 */
1265 	if (req->oldptr == NULL) {
1266 		n = unp_count;
1267 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
1268 		return 0;
1269 	}
1270 
1271 	if (req->newptr != NULL)
1272 		return EPERM;
1273 
1274 	lwkt_gettoken(&unp_token);
1275 
1276 	/*
1277 	 * OK, now we're committed to doing something.
1278 	 */
1279 	gencnt = unp_gencnt;
1280 	n = unp_count;
1281 
1282 	unp_list = kmalloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1283 
1284 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1285 	     unp = LIST_NEXT(unp, unp_link)) {
1286 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->td, unp))
1287 			unp_list[i++] = unp;
1288 	}
1289 	n = i;			/* in case we lost some during malloc */
1290 
1291 	error = 0;
1292 	for (i = 0; i < n; i++) {
1293 		unp = unp_list[i];
1294 		if (unp->unp_gencnt <= gencnt) {
1295 			struct xunpcb xu;
1296 			xu.xu_len = sizeof xu;
1297 			xu.xu_unpp = unp;
1298 			/*
1299 			 * XXX - need more locking here to protect against
1300 			 * connect/disconnect races for SMP.
1301 			 */
1302 			if (unp->unp_addr)
1303 				bcopy(unp->unp_addr, &xu.xu_addr,
1304 				      unp->unp_addr->sun_len);
1305 			if (unp->unp_conn && unp->unp_conn->unp_addr)
1306 				bcopy(unp->unp_conn->unp_addr,
1307 				      &xu.xu_caddr,
1308 				      unp->unp_conn->unp_addr->sun_len);
1309 			bcopy(unp, &xu.xu_unp, sizeof *unp);
1310 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
1311 			error = SYSCTL_OUT(req, &xu, sizeof xu);
1312 		}
1313 	}
1314 	lwkt_reltoken(&unp_token);
1315 	kfree(unp_list, M_TEMP);
1316 
1317 	return error;
1318 }
1319 
1320 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1321 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1322 	    "List of active local datagram sockets");
1323 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1324 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1325 	    "List of active local stream sockets");
1326 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
1327 	    (caddr_t)(long)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1328 	    "List of active local seqpacket stream sockets");
1329 
1330 static void
1331 unp_shutdown(struct unpcb *unp)
1332 {
1333 	struct socket *so;
1334 
1335 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1336 	     unp->unp_socket->so_type == SOCK_SEQPACKET) &&
1337 	    unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
1338 		socantrcvmore(so);
1339 	}
1340 }
1341 
1342 #ifdef notdef
1343 void
1344 unp_drain(void)
1345 {
1346 	lwkt_gettoken(&unp_token);
1347 	lwkt_reltoken(&unp_token);
1348 }
1349 #endif
1350 
1351 int
1352 unp_externalize(struct mbuf *rights)
1353 {
1354 	struct thread *td = curthread;
1355 	struct proc *p = td->td_proc;		/* XXX */
1356 	struct lwp *lp = td->td_lwp;
1357 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
1358 	int *fdp;
1359 	int i;
1360 	struct file **rp;
1361 	struct file *fp;
1362 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
1363 		/ sizeof (struct file *);
1364 	int f;
1365 
1366 	lwkt_gettoken(&unp_token);
1367 
1368 	/*
1369 	 * if the new FD's will not fit, then we free them all
1370 	 */
1371 	if (!fdavail(p, newfds)) {
1372 		rp = (struct file **)CMSG_DATA(cm);
1373 		for (i = 0; i < newfds; i++) {
1374 			fp = *rp;
1375 			/*
1376 			 * zero the pointer before calling unp_discard,
1377 			 * since it may end up in unp_gc()..
1378 			 */
1379 			*rp++ = NULL;
1380 			unp_discard(fp, NULL);
1381 		}
1382 		lwkt_reltoken(&unp_token);
1383 		return (EMSGSIZE);
1384 	}
1385 
1386 	/*
1387 	 * now change each pointer to an fd in the global table to
1388 	 * an integer that is the index to the local fd table entry
1389 	 * that we set up to point to the global one we are transferring.
1390 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1391 	 * then do it in forward order. In that case, an integer will
1392 	 * always come in the same place or before its corresponding
1393 	 * struct file pointer.
1394 	 * If sizeof (struct file *) is smaller than sizeof int, then
1395 	 * do it in reverse order.
1396 	 */
1397 	if (sizeof (struct file *) >= sizeof (int)) {
1398 		fdp = (int *)CMSG_DATA(cm);
1399 		rp = (struct file **)CMSG_DATA(cm);
1400 		for (i = 0; i < newfds; i++) {
1401 			if (fdalloc(p, 0, &f))
1402 				panic("unp_externalize");
1403 			fp = *rp++;
1404 			unp_fp_externalize(lp, fp, f);
1405 			*fdp++ = f;
1406 		}
1407 	} else {
1408 		fdp = (int *)CMSG_DATA(cm) + newfds - 1;
1409 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
1410 		for (i = 0; i < newfds; i++) {
1411 			if (fdalloc(p, 0, &f))
1412 				panic("unp_externalize");
1413 			fp = *rp--;
1414 			unp_fp_externalize(lp, fp, f);
1415 			*fdp-- = f;
1416 		}
1417 	}
1418 
1419 	/*
1420 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1421 	 * differs.
1422 	 */
1423 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1424 	rights->m_len = cm->cmsg_len;
1425 
1426 	lwkt_reltoken(&unp_token);
1427 	return (0);
1428 }
1429 
1430 static void
1431 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd)
1432 {
1433 	struct file *fx;
1434 	int error;
1435 
1436 	lwkt_gettoken(&unp_token);
1437 
1438 	if (lp) {
1439 		KKASSERT(fd >= 0);
1440 		if (fp->f_flag & FREVOKED) {
1441 			kprintf("Warning: revoked fp exiting unix socket\n");
1442 			fx = NULL;
1443 			error = falloc(lp, &fx, NULL);
1444 			if (error == 0)
1445 				fsetfd(lp->lwp_proc->p_fd, fx, fd);
1446 			else
1447 				fsetfd(lp->lwp_proc->p_fd, NULL, fd);
1448 			fdrop(fx);
1449 		} else {
1450 			fsetfd(lp->lwp_proc->p_fd, fp, fd);
1451 		}
1452 	}
1453 	spin_lock(&unp_spin);
1454 	fp->f_msgcount--;
1455 	unp_rights--;
1456 	spin_unlock(&unp_spin);
1457 	fdrop(fp);
1458 
1459 	lwkt_reltoken(&unp_token);
1460 }
1461 
1462 
1463 void
1464 unp_init(void)
1465 {
1466 	LIST_INIT(&unp_dhead);
1467 	LIST_INIT(&unp_shead);
1468 	spin_init(&unp_spin, "unpinit");
1469 }
1470 
1471 static int
1472 unp_internalize(struct mbuf *control, struct thread *td)
1473 {
1474 	struct proc *p = td->td_proc;
1475 	struct filedesc *fdescp;
1476 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1477 	struct file **rp;
1478 	struct file *fp;
1479 	int i, fd, *fdp;
1480 	struct cmsgcred *cmcred;
1481 	int oldfds;
1482 	u_int newlen;
1483 	int error;
1484 
1485 	KKASSERT(p);
1486 	lwkt_gettoken(&unp_token);
1487 
1488 	fdescp = p->p_fd;
1489 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1490 	    cm->cmsg_level != SOL_SOCKET ||
1491 	    CMSG_ALIGN(cm->cmsg_len) != control->m_len) {
1492 		error = EINVAL;
1493 		goto done;
1494 	}
1495 
1496 	/*
1497 	 * Fill in credential information.
1498 	 */
1499 	if (cm->cmsg_type == SCM_CREDS) {
1500 		cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1501 		cmcred->cmcred_pid = p->p_pid;
1502 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1503 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1504 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1505 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1506 							CMGROUP_MAX);
1507 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1508 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1509 		error = 0;
1510 		goto done;
1511 	}
1512 
1513 	/*
1514 	 * cmsghdr may not be aligned, do not allow calculation(s) to
1515 	 * go negative.
1516 	 */
1517 	if (cm->cmsg_len < CMSG_LEN(0)) {
1518 		error = EINVAL;
1519 		goto done;
1520 	}
1521 
1522 	oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof (int);
1523 
1524 	/*
1525 	 * check that all the FDs passed in refer to legal OPEN files
1526 	 * If not, reject the entire operation.
1527 	 */
1528 	fdp = (int *)CMSG_DATA(cm);
1529 	for (i = 0; i < oldfds; i++) {
1530 		fd = *fdp++;
1531 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1532 		    fdescp->fd_files[fd].fp == NULL) {
1533 			error = EBADF;
1534 			goto done;
1535 		}
1536 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) {
1537 			error = EOPNOTSUPP;
1538 			goto done;
1539 		}
1540 	}
1541 	/*
1542 	 * Now replace the integer FDs with pointers to
1543 	 * the associated global file table entry..
1544 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1545 	 * enough, return E2BIG.
1546 	 */
1547 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1548 	if (newlen > MCLBYTES) {
1549 		error = E2BIG;
1550 		goto done;
1551 	}
1552 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1553 		if (control->m_flags & M_EXT) {
1554 			error = E2BIG;
1555 			goto done;
1556 		}
1557 		MCLGET(control, M_WAITOK);
1558 		if (!(control->m_flags & M_EXT)) {
1559 			error = ENOBUFS;
1560 			goto done;
1561 		}
1562 
1563 		/* copy the data to the cluster */
1564 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1565 		cm = mtod(control, struct cmsghdr *);
1566 	}
1567 
1568 	/*
1569 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1570 	 * differs.
1571 	 */
1572 	cm->cmsg_len = newlen;
1573 	control->m_len = CMSG_ALIGN(newlen);
1574 
1575 	/*
1576 	 * Transform the file descriptors into struct file pointers.
1577 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1578 	 * then do it in reverse order so that the int won't get until
1579 	 * we're done.
1580 	 * If sizeof (struct file *) is smaller than sizeof int, then
1581 	 * do it in forward order.
1582 	 */
1583 	if (sizeof (struct file *) >= sizeof (int)) {
1584 		fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1585 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1586 		for (i = 0; i < oldfds; i++) {
1587 			fp = fdescp->fd_files[*fdp--].fp;
1588 			*rp-- = fp;
1589 			fhold(fp);
1590 			spin_lock(&unp_spin);
1591 			fp->f_msgcount++;
1592 			unp_rights++;
1593 			spin_unlock(&unp_spin);
1594 		}
1595 	} else {
1596 		fdp = (int *)CMSG_DATA(cm);
1597 		rp = (struct file **)CMSG_DATA(cm);
1598 		for (i = 0; i < oldfds; i++) {
1599 			fp = fdescp->fd_files[*fdp++].fp;
1600 			*rp++ = fp;
1601 			fhold(fp);
1602 			spin_lock(&unp_spin);
1603 			fp->f_msgcount++;
1604 			unp_rights++;
1605 			spin_unlock(&unp_spin);
1606 		}
1607 	}
1608 	error = 0;
1609 done:
1610 	lwkt_reltoken(&unp_token);
1611 	return error;
1612 }
1613 
1614 /*
1615  * Garbage collect in-transit file descriptors that get lost due to
1616  * loops (i.e. when a socket is sent to another process over itself,
1617  * and more complex situations).
1618  *
1619  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1620  */
1621 
1622 struct unp_gc_info {
1623 	struct file **extra_ref;
1624 	struct file *locked_fp;
1625 	int defer;
1626 	int index;
1627 	int maxindex;
1628 };
1629 
1630 static void
1631 unp_gc(void)
1632 {
1633 	struct unp_gc_info info;
1634 	static boolean_t unp_gcing;
1635 	struct file **fpp;
1636 	int i;
1637 
1638 	/*
1639 	 * Only one gc can be in-progress at any given moment
1640 	 */
1641 	spin_lock(&unp_spin);
1642 	if (unp_gcing) {
1643 		spin_unlock(&unp_spin);
1644 		return;
1645 	}
1646 	unp_gcing = TRUE;
1647 	spin_unlock(&unp_spin);
1648 
1649 	lwkt_gettoken(&unp_token);
1650 
1651 	/*
1652 	 * Before going through all this, set all FDs to be NOT defered
1653 	 * and NOT externally accessible (not marked).  During the scan
1654 	 * a fd can be marked externally accessible but we may or may not
1655 	 * be able to immediately process it (controlled by FDEFER).
1656 	 *
1657 	 * If we loop sleep a bit.  The complexity of the topology can cause
1658 	 * multiple loops.  Also failure to acquire the socket's so_rcv
1659 	 * token can cause us to loop.
1660 	 */
1661 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1662 	do {
1663 		info.defer = 0;
1664 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1665 		if (info.defer)
1666 			tsleep(&info, 0, "gcagain", 1);
1667 	} while (info.defer);
1668 
1669 	/*
1670 	 * We grab an extra reference to each of the file table entries
1671 	 * that are not otherwise accessible and then free the rights
1672 	 * that are stored in messages on them.
1673 	 *
1674 	 * The bug in the orginal code is a little tricky, so I'll describe
1675 	 * what's wrong with it here.
1676 	 *
1677 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1678 	 * times -- consider the case of sockets A and B that contain
1679 	 * references to each other.  On a last close of some other socket,
1680 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1681 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1682 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1683 	 * results in the following chain.  Closef calls soo_close, which
1684 	 * calls soclose.   Soclose calls first (through the switch
1685 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1686 	 * returns because the previous instance had set unp_gcing, and
1687 	 * we return all the way back to soclose, which marks the socket
1688 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1689 	 * to free up the rights that are queued in messages on the socket A,
1690 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1691 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1692 	 * instance of unp_discard just calls closef on B.
1693 	 *
1694 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1695 	 * which results in another closef on A.  Unfortunately, A is already
1696 	 * being closed, and the descriptor has already been marked with
1697 	 * SS_NOFDREF, and soclose panics at this point.
1698 	 *
1699 	 * Here, we first take an extra reference to each inaccessible
1700 	 * descriptor.  Then, we call sorflush ourself, since we know
1701 	 * it is a Unix domain socket anyhow.  After we destroy all the
1702 	 * rights carried in messages, we do a last closef to get rid
1703 	 * of our extra reference.  This is the last close, and the
1704 	 * unp_detach etc will shut down the socket.
1705 	 *
1706 	 * 91/09/19, bsy@cs.cmu.edu
1707 	 */
1708 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1709 	info.maxindex = 256;
1710 
1711 	do {
1712 		/*
1713 		 * Look for matches
1714 		 */
1715 		info.index = 0;
1716 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1717 
1718 		/*
1719 		 * For each FD on our hit list, do the following two things
1720 		 */
1721 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1722 			struct file *tfp = *fpp;
1723 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1724 				sorflush((struct socket *)(tfp->f_data));
1725 		}
1726 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1727 			closef(*fpp, NULL);
1728 	} while (info.index == info.maxindex);
1729 
1730 	lwkt_reltoken(&unp_token);
1731 
1732 	kfree((caddr_t)info.extra_ref, M_FILE);
1733 	unp_gcing = FALSE;
1734 }
1735 
1736 /*
1737  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1738  */
1739 static int
1740 unp_gc_checkrefs(struct file *fp, void *data)
1741 {
1742 	struct unp_gc_info *info = data;
1743 
1744 	if (fp->f_count == 0)
1745 		return(0);
1746 	if (info->index == info->maxindex)
1747 		return(-1);
1748 
1749 	/*
1750 	 * If all refs are from msgs, and it's not marked accessible
1751 	 * then it must be referenced from some unreachable cycle
1752 	 * of (shut-down) FDs, so include it in our
1753 	 * list of FDs to remove
1754 	 */
1755 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1756 		info->extra_ref[info->index++] = fp;
1757 		fhold(fp);
1758 	}
1759 	return(0);
1760 }
1761 
1762 /*
1763  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1764  */
1765 static int
1766 unp_gc_clearmarks(struct file *fp, void *data __unused)
1767 {
1768 	atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1769 	return(0);
1770 }
1771 
1772 /*
1773  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1774  */
1775 static int
1776 unp_gc_checkmarks(struct file *fp, void *data)
1777 {
1778 	struct unp_gc_info *info = data;
1779 	struct socket *so;
1780 
1781 	/*
1782 	 * If the file is not open, skip it.  Make sure it isn't marked
1783 	 * defered or we could loop forever, in case we somehow race
1784 	 * something.
1785 	 */
1786 	if (fp->f_count == 0) {
1787 		if (fp->f_flag & FDEFER)
1788 			atomic_clear_int(&fp->f_flag, FDEFER);
1789 		return(0);
1790 	}
1791 	/*
1792 	 * If we already marked it as 'defer'  in a
1793 	 * previous pass, then try process it this time
1794 	 * and un-mark it
1795 	 */
1796 	if (fp->f_flag & FDEFER) {
1797 		atomic_clear_int(&fp->f_flag, FDEFER);
1798 	} else {
1799 		/*
1800 		 * if it's not defered, then check if it's
1801 		 * already marked.. if so skip it
1802 		 */
1803 		if (fp->f_flag & FMARK)
1804 			return(0);
1805 		/*
1806 		 * If all references are from messages
1807 		 * in transit, then skip it. it's not
1808 		 * externally accessible.
1809 		 */
1810 		if (fp->f_count == fp->f_msgcount)
1811 			return(0);
1812 		/*
1813 		 * If it got this far then it must be
1814 		 * externally accessible.
1815 		 */
1816 		atomic_set_int(&fp->f_flag, FMARK);
1817 	}
1818 
1819 	/*
1820 	 * either it was defered, or it is externally
1821 	 * accessible and not already marked so.
1822 	 * Now check if it is possibly one of OUR sockets.
1823 	 */
1824 	if (fp->f_type != DTYPE_SOCKET ||
1825 	    (so = (struct socket *)fp->f_data) == NULL) {
1826 		return(0);
1827 	}
1828 	if (so->so_proto->pr_domain != &localdomain ||
1829 	    !(so->so_proto->pr_flags & PR_RIGHTS)) {
1830 		return(0);
1831 	}
1832 
1833 	/*
1834 	 * So, Ok, it's one of our sockets and it IS externally accessible
1835 	 * (or was defered).  Now we look to see if we hold any file
1836 	 * descriptors in its message buffers.  Follow those links and mark
1837 	 * them as accessible too.
1838 	 *
1839 	 * We are holding multiple spinlocks here, if we cannot get the
1840 	 * token non-blocking defer until the next loop.
1841 	 */
1842 	info->locked_fp = fp;
1843 	if (lwkt_trytoken(&so->so_rcv.ssb_token)) {
1844 		unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1845 		lwkt_reltoken(&so->so_rcv.ssb_token);
1846 	} else {
1847 		atomic_set_int(&fp->f_flag, FDEFER);
1848 		++info->defer;
1849 	}
1850 	return (0);
1851 }
1852 
1853 /*
1854  * Scan all unix domain sockets and replace any revoked file pointers
1855  * found with the dummy file pointer fx.  We don't worry about races
1856  * against file pointers being read out as those are handled in the
1857  * externalize code.
1858  */
1859 
1860 #define REVOKE_GC_MAXFILES	32
1861 
1862 struct unp_revoke_gc_info {
1863 	struct file	*fx;
1864 	struct file	*fary[REVOKE_GC_MAXFILES];
1865 	int		fcount;
1866 };
1867 
1868 void
1869 unp_revoke_gc(struct file *fx)
1870 {
1871 	struct unp_revoke_gc_info info;
1872 	int i;
1873 
1874 	lwkt_gettoken(&unp_token);
1875 	info.fx = fx;
1876 	do {
1877 		info.fcount = 0;
1878 		allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1879 		for (i = 0; i < info.fcount; ++i)
1880 			unp_fp_externalize(NULL, info.fary[i], -1);
1881 	} while (info.fcount == REVOKE_GC_MAXFILES);
1882 	lwkt_reltoken(&unp_token);
1883 }
1884 
1885 /*
1886  * Check for and replace revoked descriptors.
1887  *
1888  * WARNING:  This routine is not allowed to block.
1889  */
1890 static int
1891 unp_revoke_gc_check(struct file *fps, void *vinfo)
1892 {
1893 	struct unp_revoke_gc_info *info = vinfo;
1894 	struct file *fp;
1895 	struct socket *so;
1896 	struct mbuf *m0;
1897 	struct mbuf *m;
1898 	struct file **rp;
1899 	struct cmsghdr *cm;
1900 	int i;
1901 	int qfds;
1902 
1903 	/*
1904 	 * Is this a unix domain socket with rights-passing abilities?
1905 	 */
1906 	if (fps->f_type != DTYPE_SOCKET)
1907 		return (0);
1908 	if ((so = (struct socket *)fps->f_data) == NULL)
1909 		return(0);
1910 	if (so->so_proto->pr_domain != &localdomain)
1911 		return(0);
1912 	if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
1913 		return(0);
1914 
1915 	/*
1916 	 * Scan the mbufs for control messages and replace any revoked
1917 	 * descriptors we find.
1918 	 */
1919 	lwkt_gettoken(&so->so_rcv.ssb_token);
1920 	m0 = so->so_rcv.ssb_mb;
1921 	while (m0) {
1922 		for (m = m0; m; m = m->m_next) {
1923 			if (m->m_type != MT_CONTROL)
1924 				continue;
1925 			if (m->m_len < sizeof(*cm))
1926 				continue;
1927 			cm = mtod(m, struct cmsghdr *);
1928 			if (cm->cmsg_level != SOL_SOCKET ||
1929 			    cm->cmsg_type != SCM_RIGHTS) {
1930 				continue;
1931 			}
1932 			qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
1933 			rp = (struct file **)CMSG_DATA(cm);
1934 			for (i = 0; i < qfds; i++) {
1935 				fp = rp[i];
1936 				if (fp->f_flag & FREVOKED) {
1937 					kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
1938 					fhold(info->fx);
1939 					info->fx->f_msgcount++;
1940 					unp_rights++;
1941 					rp[i] = info->fx;
1942 					info->fary[info->fcount++] = fp;
1943 				}
1944 				if (info->fcount == REVOKE_GC_MAXFILES)
1945 					break;
1946 			}
1947 			if (info->fcount == REVOKE_GC_MAXFILES)
1948 				break;
1949 		}
1950 		m0 = m0->m_nextpkt;
1951 		if (info->fcount == REVOKE_GC_MAXFILES)
1952 			break;
1953 	}
1954 	lwkt_reltoken(&so->so_rcv.ssb_token);
1955 
1956 	/*
1957 	 * Stop the scan if we filled up our array.
1958 	 */
1959 	if (info->fcount == REVOKE_GC_MAXFILES)
1960 		return(-1);
1961 	return(0);
1962 }
1963 
1964 /*
1965  * Dispose of the fp's stored in a mbuf.
1966  *
1967  * The dds loop can cause additional fps to be entered onto the
1968  * list while it is running, flattening out the operation and avoiding
1969  * a deep kernel stack recursion.
1970  */
1971 void
1972 unp_dispose(struct mbuf *m)
1973 {
1974 	unp_defdiscard_t dds;
1975 
1976 	lwkt_gettoken(&unp_token);
1977 	++unp_defdiscard_nest;
1978 	if (m) {
1979 		unp_scan(m, unp_discard, NULL);
1980 	}
1981 	if (unp_defdiscard_nest == 1) {
1982 		while ((dds = unp_defdiscard_base) != NULL) {
1983 			unp_defdiscard_base = dds->next;
1984 			closef(dds->fp, NULL);
1985 			kfree(dds, M_UNPCB);
1986 		}
1987 	}
1988 	--unp_defdiscard_nest;
1989 	lwkt_reltoken(&unp_token);
1990 }
1991 
1992 static int
1993 unp_listen(struct unpcb *unp, struct thread *td)
1994 {
1995 	struct proc *p = td->td_proc;
1996 
1997 	KKASSERT(p);
1998 	lwkt_gettoken(&unp_token);
1999 	cru2x(p->p_ucred, &unp->unp_peercred);
2000 	unp_setflags(unp, UNP_HAVEPCCACHED);
2001 	lwkt_reltoken(&unp_token);
2002 	return (0);
2003 }
2004 
2005 static void
2006 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
2007 {
2008 	struct mbuf *m;
2009 	struct file **rp;
2010 	struct cmsghdr *cm;
2011 	int i;
2012 	int qfds;
2013 
2014 	while (m0) {
2015 		for (m = m0; m; m = m->m_next) {
2016 			if (m->m_type == MT_CONTROL &&
2017 			    m->m_len >= sizeof(*cm)) {
2018 				cm = mtod(m, struct cmsghdr *);
2019 				if (cm->cmsg_level != SOL_SOCKET ||
2020 				    cm->cmsg_type != SCM_RIGHTS)
2021 					continue;
2022 				qfds = (cm->cmsg_len - CMSG_LEN(0)) /
2023 					sizeof(void *);
2024 				rp = (struct file **)CMSG_DATA(cm);
2025 				for (i = 0; i < qfds; i++)
2026 					(*op)(*rp++, data);
2027 				break;		/* XXX, but saves time */
2028 			}
2029 		}
2030 		m0 = m0->m_nextpkt;
2031 	}
2032 }
2033 
2034 /*
2035  * Mark visibility.  info->defer is recalculated on every pass.
2036  */
2037 static void
2038 unp_mark(struct file *fp, void *data)
2039 {
2040 	struct unp_gc_info *info = data;
2041 
2042 	if ((fp->f_flag & FMARK) == 0) {
2043 		++info->defer;
2044 		atomic_set_int(&fp->f_flag, FMARK | FDEFER);
2045 	} else if (fp->f_flag & FDEFER) {
2046 		++info->defer;
2047 	}
2048 }
2049 
2050 /*
2051  * Discard a fp previously held in a unix domain socket mbuf.  To
2052  * avoid blowing out the kernel stack due to contrived chain-reactions
2053  * we may have to defer the operation to a higher procedural level.
2054  *
2055  * Caller holds unp_token
2056  */
2057 static void
2058 unp_discard(struct file *fp, void *data __unused)
2059 {
2060 	unp_defdiscard_t dds;
2061 
2062 	spin_lock(&unp_spin);
2063 	fp->f_msgcount--;
2064 	unp_rights--;
2065 	spin_unlock(&unp_spin);
2066 
2067 	if (unp_defdiscard_nest) {
2068 		dds = kmalloc(sizeof(*dds), M_UNPCB, M_WAITOK|M_ZERO);
2069 		dds->fp = fp;
2070 		dds->next = unp_defdiscard_base;
2071 		unp_defdiscard_base = dds;
2072 	} else {
2073 		closef(fp, NULL);
2074 	}
2075 }
2076 
2077 static int
2078 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type,
2079     struct unpcb **unp_ret)
2080 {
2081 	struct proc *p = td->td_proc;
2082 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
2083 	struct vnode *vp = NULL;
2084 	struct socket *so;
2085 	struct unpcb *unp;
2086 	int error, len;
2087 	struct nlookupdata nd;
2088 	char buf[SOCK_MAXADDRLEN];
2089 
2090 	*unp_ret = NULL;
2091 
2092 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2093 	if (len <= 0) {
2094 		error = EINVAL;
2095 		goto failed;
2096 	}
2097 	strncpy(buf, soun->sun_path, len);
2098 	buf[len] = 0;
2099 
2100 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
2101 	if (error == 0)
2102 		error = nlookup(&nd);
2103 	if (error == 0)
2104 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2105 	nlookup_done(&nd);
2106 	if (error) {
2107 		vp = NULL;
2108 		goto failed;
2109 	}
2110 
2111 	if (vp->v_type != VSOCK) {
2112 		error = ENOTSOCK;
2113 		goto failed;
2114 	}
2115 	error = VOP_EACCESS(vp, VWRITE, p->p_ucred);
2116 	if (error)
2117 		goto failed;
2118 	so = vp->v_socket;
2119 	if (so == NULL) {
2120 		error = ECONNREFUSED;
2121 		goto failed;
2122 	}
2123 	if (so->so_type != type) {
2124 		error = EPROTOTYPE;
2125 		goto failed;
2126 	}
2127 
2128 	/* Lock this unp. */
2129 	unp = unp_getsocktoken(so);
2130 	if (!UNP_ISATTACHED(unp)) {
2131 		unp_reltoken(unp);
2132 		error = ECONNREFUSED;
2133 		goto failed;
2134 	}
2135 	/* And keep this unp referenced. */
2136 	unp_reference(unp);
2137 
2138 	/* Done! */
2139 	*unp_ret = unp;
2140 	error = 0;
2141 failed:
2142 	if (vp != NULL)
2143 		vput(vp);
2144 	return error;
2145 }
2146 
2147 static int
2148 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2)
2149 {
2150 	struct socket *so = unp->unp_socket;
2151 	struct socket *so2 = unp2->unp_socket;
2152 
2153 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2154 	UNP_ASSERT_TOKEN_HELD(unp);
2155 	UNP_ASSERT_TOKEN_HELD(unp2);
2156 
2157 	KASSERT(so->so_type == so2->so_type,
2158 	    ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type));
2159 
2160 	if (!UNP_ISATTACHED(unp))
2161 		return EINVAL;
2162 	if (!UNP_ISATTACHED(unp2))
2163 		return ECONNREFUSED;
2164 
2165 	KASSERT(unp->unp_conn == NULL, ("unp is already connected"));
2166 	unp->unp_conn = unp2;
2167 
2168 	switch (so->so_type) {
2169 	case SOCK_DGRAM:
2170 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2171 		soisconnected(so);
2172 		break;
2173 
2174 	case SOCK_STREAM:
2175 	case SOCK_SEQPACKET:
2176 		KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected"));
2177 		unp2->unp_conn = unp;
2178 		soisconnected(so);
2179 		soisconnected(so2);
2180 		break;
2181 
2182 	default:
2183 		panic("unp_connect_pair: unknown socket type %d", so->so_type);
2184 	}
2185 	return 0;
2186 }
2187 
2188 static void
2189 unp_drop(struct unpcb *unp, int error)
2190 {
2191 	struct unpcb *unp2;
2192 
2193 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2194 	UNP_ASSERT_TOKEN_HELD(unp);
2195 	KASSERT(unp->unp_flags & UNP_DETACHED, ("unp is not detached"));
2196 
2197 	unp_disconnect(unp, error);
2198 
2199 	while ((unp2 = LIST_FIRST(&unp->unp_refs)) != NULL) {
2200 		lwkt_getpooltoken(unp2);
2201 		unp_disconnect(unp2, ECONNRESET);
2202 		lwkt_relpooltoken(unp2);
2203 	}
2204 	unp_setflags(unp, UNP_DROPPED);
2205 }
2206