xref: /dflybsd-src/sys/kern/uipc_usrreq.c (revision cf37dc2040cea9f384bd7d3dcaf24014f441b8a6)
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 #include <sys/kern_syscall.h>
55 #include <sys/taskqueue.h>
56 
57 #include <sys/file2.h>
58 #include <sys/spinlock2.h>
59 #include <sys/socketvar2.h>
60 #include <sys/msgport2.h>
61 
62 #define UNP_DETACHED		UNP_PRIVATE1
63 #define UNP_CONNECTING		UNP_PRIVATE2
64 #define UNP_DROPPED		UNP_PRIVATE3
65 #define UNP_MARKER		UNP_PRIVATE4
66 
67 #define UNP_ISATTACHED(unp)	\
68     ((unp) != NULL && ((unp)->unp_flags & UNP_DETACHED) == 0)
69 
70 #ifdef INVARIANTS
71 #define UNP_ASSERT_TOKEN_HELD(unp) \
72     ASSERT_LWKT_TOKEN_HELD(lwkt_token_pool_lookup((unp)))
73 #else	/* !INVARIANTS */
74 #define UNP_ASSERT_TOKEN_HELD(unp)
75 #endif	/* INVARIANTS */
76 
77 struct unp_defdiscard {
78 	SLIST_ENTRY(unp_defdiscard) next;
79 	struct file *fp;
80 };
81 SLIST_HEAD(unp_defdiscard_list, unp_defdiscard);
82 
83 TAILQ_HEAD(unpcb_qhead, unpcb);
84 struct unp_global_head {
85 	struct unpcb_qhead	list;
86 	int			count;
87 };
88 
89 static	MALLOC_DEFINE(M_UNPCB, "unpcb", "unpcb struct");
90 static	unp_gen_t unp_gencnt;
91 
92 static struct unp_global_head unp_stream_head;
93 static struct unp_global_head unp_dgram_head;
94 static struct unp_global_head unp_seqpkt_head;
95 
96 static struct lwkt_token unp_token = LWKT_TOKEN_INITIALIZER(unp_token);
97 static struct taskqueue *unp_taskqueue;
98 
99 static struct unp_defdiscard_list unp_defdiscard_head;
100 static struct spinlock unp_defdiscard_spin;
101 static struct task unp_defdiscard_task;
102 
103 /*
104  * Unix communications domain.
105  *
106  * TODO:
107  *	RDM
108  *	rethink name space problems
109  *	need a proper out-of-band
110  *	lock pushdown
111  */
112 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
113 static ino_t	unp_ino = 1;		/* prototype for fake inode numbers */
114 
115 static int     unp_attach (struct socket *, struct pru_attach_info *);
116 static void    unp_detach (struct unpcb *);
117 static int     unp_bind (struct unpcb *,struct sockaddr *, struct thread *);
118 static int     unp_connect (struct socket *,struct sockaddr *,
119 				struct thread *);
120 static void    unp_disconnect(struct unpcb *, int);
121 static void    unp_shutdown (struct unpcb *);
122 static void    unp_gc (void);
123 static int     unp_gc_clearmarks(struct file *, void *);
124 static int     unp_gc_checkmarks(struct file *, void *);
125 static int     unp_gc_checkrefs(struct file *, void *);
126 static int     unp_revoke_gc_check(struct file *, void *);
127 static void    unp_scan (struct mbuf *, void (*)(struct file *, void *),
128 				void *data);
129 static void    unp_mark (struct file *, void *data);
130 static void    unp_discard (struct file *, void *);
131 static int     unp_internalize (struct mbuf *, struct thread *);
132 static int     unp_listen (struct unpcb *, struct thread *);
133 static void    unp_fp_externalize(struct lwp *lp, struct file *fp, int fd);
134 static int     unp_find_lockref(struct sockaddr *nam, struct thread *td,
135 		   short type, struct unpcb **unp_ret);
136 static int     unp_connect_pair(struct unpcb *unp, struct unpcb *unp2);
137 static void    unp_drop(struct unpcb *unp, int error);
138 static void    unp_defdiscard_taskfunc(void *, int);
139 
140 /*
141  * SMP Considerations:
142  *
143  *	Since unp_token will be automaticly released upon execution of
144  *	blocking code, we need to reference unp_conn before any possible
145  *	blocking code to prevent it from being ripped behind our back.
146  *
147  *	Any adjustment to unp->unp_conn requires both the global unp_token
148  *	AND the per-unp token (lwkt_token_pool_lookup(unp)) to be held.
149  *
150  *	Any access to so_pcb to obtain unp requires the pool token for
151  *	unp to be held.
152  */
153 
154 static __inline void
155 unp_reference(struct unpcb *unp)
156 {
157 	/* 0->1 transition will not work */
158 	KKASSERT(unp->unp_refcnt > 0);
159 	atomic_add_int(&unp->unp_refcnt, 1);
160 }
161 
162 static __inline void
163 unp_free(struct unpcb *unp)
164 {
165 	KKASSERT(unp->unp_refcnt > 0);
166 	if (atomic_fetchadd_int(&unp->unp_refcnt, -1) == 1)
167 		unp_detach(unp);
168 }
169 
170 static __inline struct unpcb *
171 unp_getsocktoken(struct socket *so)
172 {
173 	struct unpcb *unp;
174 
175 	/*
176 	 * The unp pointer is invalid until we verify that it is
177 	 * good by re-checking so_pcb AFTER obtaining the token.
178 	 */
179 	while ((unp = so->so_pcb) != NULL) {
180 		lwkt_getpooltoken(unp);
181 		if (unp == so->so_pcb)
182 			break;
183 		lwkt_relpooltoken(unp);
184 	}
185 	return unp;
186 }
187 
188 static __inline void
189 unp_reltoken(struct unpcb *unp)
190 {
191 	if (unp != NULL)
192 		lwkt_relpooltoken(unp);
193 }
194 
195 static __inline void
196 unp_setflags(struct unpcb *unp, int flags)
197 {
198 	atomic_set_int(&unp->unp_flags, flags);
199 }
200 
201 static __inline void
202 unp_clrflags(struct unpcb *unp, int flags)
203 {
204 	atomic_clear_int(&unp->unp_flags, flags);
205 }
206 
207 static __inline struct unp_global_head *
208 unp_globalhead(short type)
209 {
210 	switch (type) {
211 	case SOCK_STREAM:
212 		return &unp_stream_head;
213 	case SOCK_DGRAM:
214 		return &unp_dgram_head;
215 	case SOCK_SEQPACKET:
216 		return &unp_seqpkt_head;
217 	default:
218 		panic("unknown socket type %d", type);
219 	}
220 }
221 
222 /*
223  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
224  *	 will sofree() it when we return.
225  */
226 static void
227 uipc_abort(netmsg_t msg)
228 {
229 	struct unpcb *unp;
230 	int error;
231 
232 	lwkt_gettoken(&unp_token);
233 	unp = unp_getsocktoken(msg->base.nm_so);
234 
235 	if (UNP_ISATTACHED(unp)) {
236 		unp_setflags(unp, UNP_DETACHED);
237 		unp_drop(unp, ECONNABORTED);
238 		unp_free(unp);
239 		error = 0;
240 	} else {
241 		error = EINVAL;
242 	}
243 
244 	unp_reltoken(unp);
245 	lwkt_reltoken(&unp_token);
246 
247 	lwkt_replymsg(&msg->lmsg, error);
248 }
249 
250 static void
251 uipc_accept(netmsg_t msg)
252 {
253 	struct unpcb *unp;
254 	int error;
255 
256 	lwkt_gettoken(&unp_token);
257 	unp = unp_getsocktoken(msg->base.nm_so);
258 
259 	if (!UNP_ISATTACHED(unp)) {
260 		error = EINVAL;
261 	} else {
262 		struct unpcb *unp2 = unp->unp_conn;
263 
264 		/*
265 		 * Pass back name of connected socket,
266 		 * if it was bound and we are still connected
267 		 * (our peer may have closed already!).
268 		 */
269 		if (unp2 && unp2->unp_addr) {
270 			unp_reference(unp2);
271 			*msg->accept.nm_nam = dup_sockaddr(
272 				(struct sockaddr *)unp2->unp_addr);
273 			unp_free(unp2);
274 		} else {
275 			*msg->accept.nm_nam = dup_sockaddr(&sun_noname);
276 		}
277 		error = 0;
278 	}
279 
280 	unp_reltoken(unp);
281 	lwkt_reltoken(&unp_token);
282 
283 	lwkt_replymsg(&msg->lmsg, error);
284 }
285 
286 static void
287 uipc_attach(netmsg_t msg)
288 {
289 	int error;
290 
291 	lwkt_gettoken(&unp_token);
292 
293 	KASSERT(msg->base.nm_so->so_pcb == NULL, ("double unp attach"));
294 	error = unp_attach(msg->base.nm_so, msg->attach.nm_ai);
295 
296 	lwkt_reltoken(&unp_token);
297 	lwkt_replymsg(&msg->lmsg, error);
298 }
299 
300 static void
301 uipc_bind(netmsg_t msg)
302 {
303 	struct unpcb *unp;
304 	int error;
305 
306 	lwkt_gettoken(&unp_token);
307 	unp = unp_getsocktoken(msg->base.nm_so);
308 
309 	if (UNP_ISATTACHED(unp))
310 		error = unp_bind(unp, msg->bind.nm_nam, msg->bind.nm_td);
311 	else
312 		error = EINVAL;
313 
314 	unp_reltoken(unp);
315 	lwkt_reltoken(&unp_token);
316 
317 	lwkt_replymsg(&msg->lmsg, error);
318 }
319 
320 static void
321 uipc_connect(netmsg_t msg)
322 {
323 	int error;
324 
325 	error = unp_connect(msg->base.nm_so, msg->connect.nm_nam,
326 	    msg->connect.nm_td);
327 	lwkt_replymsg(&msg->lmsg, error);
328 }
329 
330 static void
331 uipc_connect2(netmsg_t msg)
332 {
333 	int error;
334 
335 	error = unp_connect2(msg->connect2.nm_so1, msg->connect2.nm_so2);
336 	lwkt_replymsg(&msg->lmsg, error);
337 }
338 
339 /* control is EOPNOTSUPP */
340 
341 static void
342 uipc_detach(netmsg_t msg)
343 {
344 	struct unpcb *unp;
345 	int error;
346 
347 	lwkt_gettoken(&unp_token);
348 	unp = unp_getsocktoken(msg->base.nm_so);
349 
350 	if (UNP_ISATTACHED(unp)) {
351 		unp_setflags(unp, UNP_DETACHED);
352 		unp_drop(unp, 0);
353 		unp_free(unp);
354 		error = 0;
355 	} else {
356 		error = EINVAL;
357 	}
358 
359 	unp_reltoken(unp);
360 	lwkt_reltoken(&unp_token);
361 
362 	lwkt_replymsg(&msg->lmsg, error);
363 }
364 
365 static void
366 uipc_disconnect(netmsg_t msg)
367 {
368 	struct unpcb *unp;
369 	int error;
370 
371 	lwkt_gettoken(&unp_token);
372 	unp = unp_getsocktoken(msg->base.nm_so);
373 
374 	if (UNP_ISATTACHED(unp)) {
375 		unp_disconnect(unp, 0);
376 		error = 0;
377 	} else {
378 		error = EINVAL;
379 	}
380 
381 	unp_reltoken(unp);
382 	lwkt_reltoken(&unp_token);
383 
384 	lwkt_replymsg(&msg->lmsg, error);
385 }
386 
387 static void
388 uipc_listen(netmsg_t msg)
389 {
390 	struct unpcb *unp;
391 	int error;
392 
393 	lwkt_gettoken(&unp_token);
394 	unp = unp_getsocktoken(msg->base.nm_so);
395 
396 	if (!UNP_ISATTACHED(unp) || unp->unp_vnode == NULL)
397 		error = EINVAL;
398 	else
399 		error = unp_listen(unp, msg->listen.nm_td);
400 
401 	unp_reltoken(unp);
402 	lwkt_reltoken(&unp_token);
403 
404 	lwkt_replymsg(&msg->lmsg, error);
405 }
406 
407 static void
408 uipc_peeraddr(netmsg_t msg)
409 {
410 	struct unpcb *unp;
411 	int error;
412 
413 	lwkt_gettoken(&unp_token);
414 	unp = unp_getsocktoken(msg->base.nm_so);
415 
416 	if (!UNP_ISATTACHED(unp)) {
417 		error = EINVAL;
418 	} else if (unp->unp_conn && unp->unp_conn->unp_addr) {
419 		struct unpcb *unp2 = unp->unp_conn;
420 
421 		unp_reference(unp2);
422 		*msg->peeraddr.nm_nam = dup_sockaddr(
423 				(struct sockaddr *)unp2->unp_addr);
424 		unp_free(unp2);
425 		error = 0;
426 	} else {
427 		/*
428 		 * XXX: It seems that this test always fails even when
429 		 * connection is established.  So, this else clause is
430 		 * added as workaround to return PF_LOCAL sockaddr.
431 		 */
432 		*msg->peeraddr.nm_nam = dup_sockaddr(&sun_noname);
433 		error = 0;
434 	}
435 
436 	unp_reltoken(unp);
437 	lwkt_reltoken(&unp_token);
438 
439 	lwkt_replymsg(&msg->lmsg, error);
440 }
441 
442 static void
443 uipc_rcvd(netmsg_t msg)
444 {
445 	struct unpcb *unp, *unp2;
446 	struct socket *so;
447 	struct socket *so2;
448 	int error;
449 
450 	/*
451 	 * so_pcb is only modified with both the global and the unp
452 	 * pool token held.
453 	 */
454 	so = msg->base.nm_so;
455 	unp = unp_getsocktoken(so);
456 
457 	if (!UNP_ISATTACHED(unp)) {
458 		error = EINVAL;
459 		goto done;
460 	}
461 
462 	switch (so->so_type) {
463 	case SOCK_DGRAM:
464 		panic("uipc_rcvd DGRAM?");
465 		/*NOTREACHED*/
466 	case SOCK_STREAM:
467 	case SOCK_SEQPACKET:
468 		if (unp->unp_conn == NULL)
469 			break;
470 		unp2 = unp->unp_conn;	/* protected by pool token */
471 
472 		/*
473 		 * Because we are transfering mbufs directly to the
474 		 * peer socket we have to use SSB_STOP on the sender
475 		 * to prevent it from building up infinite mbufs.
476 		 *
477 		 * As in several places in this module w ehave to ref unp2
478 		 * to ensure that it does not get ripped out from under us
479 		 * if we block on the so2 token or in sowwakeup().
480 		 */
481 		so2 = unp2->unp_socket;
482 		unp_reference(unp2);
483 		lwkt_gettoken(&so2->so_rcv.ssb_token);
484 		if (so->so_rcv.ssb_cc < so2->so_snd.ssb_hiwat &&
485 		    so->so_rcv.ssb_mbcnt < so2->so_snd.ssb_mbmax
486 		) {
487 			atomic_clear_int(&so2->so_snd.ssb_flags, SSB_STOP);
488 
489 			sowwakeup(so2);
490 		}
491 		lwkt_reltoken(&so2->so_rcv.ssb_token);
492 		unp_free(unp2);
493 		break;
494 	default:
495 		panic("uipc_rcvd unknown socktype");
496 		/*NOTREACHED*/
497 	}
498 	error = 0;
499 done:
500 	unp_reltoken(unp);
501 	lwkt_replymsg(&msg->lmsg, error);
502 }
503 
504 /* pru_rcvoob is EOPNOTSUPP */
505 
506 static void
507 uipc_send(netmsg_t msg)
508 {
509 	struct unpcb *unp, *unp2;
510 	struct socket *so;
511 	struct socket *so2;
512 	struct mbuf *control;
513 	struct mbuf *m;
514 	int error = 0;
515 
516 	so = msg->base.nm_so;
517 	control = msg->send.nm_control;
518 	m = msg->send.nm_m;
519 
520 	/*
521 	 * so_pcb is only modified with both the global and the unp
522 	 * pool token held.
523 	 */
524 	so = msg->base.nm_so;
525 	unp = unp_getsocktoken(so);
526 
527 	if (!UNP_ISATTACHED(unp)) {
528 		error = EINVAL;
529 		goto release;
530 	}
531 
532 	if (msg->send.nm_flags & PRUS_OOB) {
533 		error = EOPNOTSUPP;
534 		goto release;
535 	}
536 
537 	wakeup_start_delayed();
538 
539 	if (control && (error = unp_internalize(control, msg->send.nm_td)))
540 		goto release;
541 
542 	switch (so->so_type) {
543 	case SOCK_DGRAM:
544 	{
545 		struct sockaddr *from;
546 
547 		if (msg->send.nm_addr) {
548 			if (unp->unp_conn) {
549 				error = EISCONN;
550 				break;
551 			}
552 			error = unp_find_lockref(msg->send.nm_addr,
553 			    msg->send.nm_td, so->so_type, &unp2);
554 			if (error)
555 				break;
556 			/*
557 			 * NOTE:
558 			 * unp2 is locked and referenced.
559 			 *
560 			 * We could unlock unp2 now, since it was checked
561 			 * and referenced.
562 			 */
563 			unp_reltoken(unp2);
564 		} else {
565 			if (unp->unp_conn == NULL) {
566 				error = ENOTCONN;
567 				break;
568 			}
569 			unp2 = unp->unp_conn;
570 			unp_reference(unp2);
571 		}
572 		/* NOTE: unp2 is referenced. */
573 		so2 = unp2->unp_socket;
574 
575 		if (unp->unp_addr)
576 			from = (struct sockaddr *)unp->unp_addr;
577 		else
578 			from = &sun_noname;
579 
580 		lwkt_gettoken(&so2->so_rcv.ssb_token);
581 		if (ssb_appendaddr(&so2->so_rcv, from, m, control)) {
582 			sorwakeup(so2);
583 			m = NULL;
584 			control = NULL;
585 		} else {
586 			error = ENOBUFS;
587 		}
588 		lwkt_reltoken(&so2->so_rcv.ssb_token);
589 
590 		unp_free(unp2);
591 		break;
592 	}
593 
594 	case SOCK_STREAM:
595 	case SOCK_SEQPACKET:
596 		/* Connect if not connected yet. */
597 		/*
598 		 * Note: A better implementation would complain
599 		 * if not equal to the peer's address.
600 		 */
601 		if (unp->unp_conn == NULL) {
602 			if (msg->send.nm_addr) {
603 				error = unp_connect(so,
604 						    msg->send.nm_addr,
605 						    msg->send.nm_td);
606 				if (error)
607 					break;	/* XXX */
608 			}
609 			/*
610 			 * NOTE:
611 			 * unp_conn still could be NULL, even if the
612 			 * above unp_connect() succeeds; since the
613 			 * current unp's token could be released due
614 			 * to blocking operations after unp_conn is
615 			 * assigned.
616 			 */
617 			if (unp->unp_conn == NULL) {
618 				error = ENOTCONN;
619 				break;
620 			}
621 		}
622 		if (so->so_state & SS_CANTSENDMORE) {
623 			error = EPIPE;
624 			break;
625 		}
626 
627 		unp2 = unp->unp_conn;
628 		KASSERT(unp2 != NULL, ("unp is not connected"));
629 		so2 = unp2->unp_socket;
630 
631 		unp_reference(unp2);
632 
633 		/*
634 		 * Send to paired receive port, and then reduce
635 		 * send buffer hiwater marks to maintain backpressure.
636 		 * Wake up readers.
637 		 */
638 		lwkt_gettoken(&so2->so_rcv.ssb_token);
639 		if (control) {
640 			if (ssb_appendcontrol(&so2->so_rcv, m, control)) {
641 				control = NULL;
642 				m = NULL;
643 			}
644 		} else if (so->so_type == SOCK_SEQPACKET) {
645 			sbappendrecord(&so2->so_rcv.sb, m);
646 			m = NULL;
647 		} else {
648 			sbappend(&so2->so_rcv.sb, m);
649 			m = NULL;
650 		}
651 
652 		/*
653 		 * Because we are transfering mbufs directly to the
654 		 * peer socket we have to use SSB_STOP on the sender
655 		 * to prevent it from building up infinite mbufs.
656 		 */
657 		if (so2->so_rcv.ssb_cc >= so->so_snd.ssb_hiwat ||
658 		    so2->so_rcv.ssb_mbcnt >= so->so_snd.ssb_mbmax
659 		) {
660 			atomic_set_int(&so->so_snd.ssb_flags, SSB_STOP);
661 		}
662 		lwkt_reltoken(&so2->so_rcv.ssb_token);
663 		sorwakeup(so2);
664 
665 		unp_free(unp2);
666 		break;
667 
668 	default:
669 		panic("uipc_send unknown socktype");
670 	}
671 
672 	/*
673 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
674 	 */
675 	if (msg->send.nm_flags & PRUS_EOF) {
676 		socantsendmore(so);
677 		unp_shutdown(unp);
678 	}
679 
680 	if (control && error != 0)
681 		unp_dispose(control);
682 release:
683 	unp_reltoken(unp);
684 	wakeup_end_delayed();
685 
686 	if (control)
687 		m_freem(control);
688 	if (m)
689 		m_freem(m);
690 	lwkt_replymsg(&msg->lmsg, error);
691 }
692 
693 /*
694  * MPSAFE
695  */
696 static void
697 uipc_sense(netmsg_t msg)
698 {
699 	struct unpcb *unp;
700 	struct socket *so;
701 	struct stat *sb;
702 	int error;
703 
704 	so = msg->base.nm_so;
705 	sb = msg->sense.nm_stat;
706 
707 	/*
708 	 * so_pcb is only modified with both the global and the unp
709 	 * pool token held.
710 	 */
711 	unp = unp_getsocktoken(so);
712 
713 	if (!UNP_ISATTACHED(unp)) {
714 		error = EINVAL;
715 		goto done;
716 	}
717 
718 	sb->st_blksize = so->so_snd.ssb_hiwat;
719 	sb->st_dev = NOUDEV;
720 	if (unp->unp_ino == 0) {	/* make up a non-zero inode number */
721 		unp->unp_ino = atomic_fetchadd_long(&unp_ino, 1);
722 		if (__predict_false(unp->unp_ino == 0))
723 			unp->unp_ino = atomic_fetchadd_long(&unp_ino, 1);
724 	}
725 	sb->st_ino = unp->unp_ino;
726 	error = 0;
727 done:
728 	unp_reltoken(unp);
729 	lwkt_replymsg(&msg->lmsg, error);
730 }
731 
732 static void
733 uipc_shutdown(netmsg_t msg)
734 {
735 	struct socket *so;
736 	struct unpcb *unp;
737 	int error;
738 
739 	/*
740 	 * so_pcb is only modified with both the global and the unp
741 	 * pool token held.
742 	 */
743 	so = msg->base.nm_so;
744 	unp = unp_getsocktoken(so);
745 
746 	if (UNP_ISATTACHED(unp)) {
747 		socantsendmore(so);
748 		unp_shutdown(unp);
749 		error = 0;
750 	} else {
751 		error = EINVAL;
752 	}
753 
754 	unp_reltoken(unp);
755 	lwkt_replymsg(&msg->lmsg, error);
756 }
757 
758 static void
759 uipc_sockaddr(netmsg_t msg)
760 {
761 	struct unpcb *unp;
762 	int error;
763 
764 	/*
765 	 * so_pcb is only modified with both the global and the unp
766 	 * pool token held.
767 	 */
768 	unp = unp_getsocktoken(msg->base.nm_so);
769 
770 	if (UNP_ISATTACHED(unp)) {
771 		if (unp->unp_addr) {
772 			*msg->sockaddr.nm_nam =
773 				dup_sockaddr((struct sockaddr *)unp->unp_addr);
774 		}
775 		error = 0;
776 	} else {
777 		error = EINVAL;
778 	}
779 
780 	unp_reltoken(unp);
781 	lwkt_replymsg(&msg->lmsg, error);
782 }
783 
784 struct pr_usrreqs uipc_usrreqs = {
785 	.pru_abort = uipc_abort,
786 	.pru_accept = uipc_accept,
787 	.pru_attach = uipc_attach,
788 	.pru_bind = uipc_bind,
789 	.pru_connect = uipc_connect,
790 	.pru_connect2 = uipc_connect2,
791 	.pru_control = pr_generic_notsupp,
792 	.pru_detach = uipc_detach,
793 	.pru_disconnect = uipc_disconnect,
794 	.pru_listen = uipc_listen,
795 	.pru_peeraddr = uipc_peeraddr,
796 	.pru_rcvd = uipc_rcvd,
797 	.pru_rcvoob = pr_generic_notsupp,
798 	.pru_send = uipc_send,
799 	.pru_sense = uipc_sense,
800 	.pru_shutdown = uipc_shutdown,
801 	.pru_sockaddr = uipc_sockaddr,
802 	.pru_sosend = sosend,
803 	.pru_soreceive = soreceive
804 };
805 
806 void
807 uipc_ctloutput(netmsg_t msg)
808 {
809 	struct socket *so;
810 	struct sockopt *sopt;
811 	struct unpcb *unp;
812 	int error = 0;
813 
814 	so = msg->base.nm_so;
815 	sopt = msg->ctloutput.nm_sopt;
816 
817 	lwkt_gettoken(&unp_token);
818 	unp = unp_getsocktoken(so);
819 
820 	if (!UNP_ISATTACHED(unp)) {
821 		error = EINVAL;
822 		goto done;
823 	}
824 
825 	switch (sopt->sopt_dir) {
826 	case SOPT_GET:
827 		switch (sopt->sopt_name) {
828 		case LOCAL_PEERCRED:
829 			if (unp->unp_flags & UNP_HAVEPC)
830 				soopt_from_kbuf(sopt, &unp->unp_peercred,
831 						sizeof(unp->unp_peercred));
832 			else {
833 				if (so->so_type == SOCK_STREAM)
834 					error = ENOTCONN;
835 				else if (so->so_type == SOCK_SEQPACKET)
836 					error = ENOTCONN;
837 				else
838 					error = EINVAL;
839 			}
840 			break;
841 		default:
842 			error = EOPNOTSUPP;
843 			break;
844 		}
845 		break;
846 	case SOPT_SET:
847 	default:
848 		error = EOPNOTSUPP;
849 		break;
850 	}
851 
852 done:
853 	unp_reltoken(unp);
854 	lwkt_reltoken(&unp_token);
855 
856 	lwkt_replymsg(&msg->lmsg, error);
857 }
858 
859 /*
860  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
861  * for stream sockets, although the total for sender and receiver is
862  * actually only PIPSIZ.
863  *
864  * Datagram sockets really use the sendspace as the maximum datagram size,
865  * and don't really want to reserve the sendspace.  Their recvspace should
866  * be large enough for at least one max-size datagram plus address.
867  *
868  * We want the local send/recv space to be significant larger then lo0's
869  * mtu of 16384.
870  */
871 #ifndef PIPSIZ
872 #define	PIPSIZ	57344
873 #endif
874 static u_long	unpst_sendspace = PIPSIZ;
875 static u_long	unpst_recvspace = PIPSIZ;
876 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
877 static u_long	unpdg_recvspace = 4*1024;
878 
879 static int	unp_rights;			/* file descriptors in flight */
880 static struct spinlock unp_spin = SPINLOCK_INITIALIZER(&unp_spin, "unp_spin");
881 
882 SYSCTL_DECL(_net_local_seqpacket);
883 SYSCTL_DECL(_net_local_stream);
884 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
885     &unpst_sendspace, 0, "Size of stream socket send buffer");
886 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
887     &unpst_recvspace, 0, "Size of stream socket receive buffer");
888 
889 SYSCTL_DECL(_net_local_dgram);
890 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
891     &unpdg_sendspace, 0, "Max datagram socket size");
892 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
893     &unpdg_recvspace, 0, "Size of datagram socket receive buffer");
894 
895 SYSCTL_DECL(_net_local);
896 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
897    "File descriptors in flight");
898 
899 static int
900 unp_attach(struct socket *so, struct pru_attach_info *ai)
901 {
902 	struct unp_global_head *head;
903 	struct unpcb *unp;
904 	int error;
905 
906 	lwkt_gettoken(&unp_token);
907 
908 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
909 		switch (so->so_type) {
910 		case SOCK_STREAM:
911 		case SOCK_SEQPACKET:
912 			error = soreserve(so, unpst_sendspace, unpst_recvspace,
913 					  ai->sb_rlimit);
914 			break;
915 
916 		case SOCK_DGRAM:
917 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace,
918 					  ai->sb_rlimit);
919 			break;
920 
921 		default:
922 			panic("unp_attach");
923 		}
924 		if (error)
925 			goto failed;
926 	}
927 
928 	/*
929 	 * In order to support sendfile we have to set either SSB_STOPSUPP
930 	 * or SSB_PREALLOC.  Unix domain sockets use the SSB_STOP flow
931 	 * control mechanism.
932 	 */
933 	if (so->so_type == SOCK_STREAM) {
934 		atomic_set_int(&so->so_rcv.ssb_flags, SSB_STOPSUPP);
935 		atomic_set_int(&so->so_snd.ssb_flags, SSB_STOPSUPP);
936 	}
937 
938 	unp = kmalloc(sizeof(*unp), M_UNPCB, M_WAITOK | M_ZERO | M_NULLOK);
939 	if (unp == NULL) {
940 		error = ENOBUFS;
941 		goto failed;
942 	}
943 	unp->unp_refcnt = 1;
944 	unp->unp_gencnt = ++unp_gencnt;
945 	LIST_INIT(&unp->unp_refs);
946 	unp->unp_socket = so;
947 	unp->unp_rvnode = ai->fd_rdir;		/* jail cruft XXX JH */
948 	so->so_pcb = (caddr_t)unp;
949 	soreference(so);
950 
951 	head = unp_globalhead(so->so_type);
952 	TAILQ_INSERT_TAIL(&head->list, unp, unp_link);
953 	head->count++;
954 	error = 0;
955 failed:
956 	lwkt_reltoken(&unp_token);
957 	return error;
958 }
959 
960 static void
961 unp_detach(struct unpcb *unp)
962 {
963 	struct unp_global_head *head;
964 	struct socket *so;
965 
966 	lwkt_gettoken(&unp_token);
967 	lwkt_getpooltoken(unp);
968 
969 	so = unp->unp_socket;
970 
971 	head = unp_globalhead(so->so_type);
972 	KASSERT(head->count > 0, ("invalid unp count"));
973 	TAILQ_REMOVE(&head->list, unp, unp_link);
974 	head->count--;
975 
976 	unp->unp_gencnt = ++unp_gencnt;
977 	if (unp->unp_vnode) {
978 		unp->unp_vnode->v_socket = NULL;
979 		vrele(unp->unp_vnode);
980 		unp->unp_vnode = NULL;
981 	}
982 	soisdisconnected(so);
983 	soreference(so);		/* for delayed sorflush */
984 	KKASSERT(so->so_pcb == unp);
985 	so->so_pcb = NULL;		/* both tokens required */
986 	unp->unp_socket = NULL;
987 	sofree(so);		/* remove pcb ref */
988 
989 	if (unp_rights) {
990 		/*
991 		 * Normally the receive buffer is flushed later,
992 		 * in sofree, but if our receive buffer holds references
993 		 * to descriptors that are now garbage, we will dispose
994 		 * of those descriptor references after the garbage collector
995 		 * gets them (resulting in a "panic: closef: count < 0").
996 		 */
997 		sorflush(so);
998 		unp_gc();
999 	}
1000 	sofree(so);
1001 	lwkt_relpooltoken(unp);
1002 	lwkt_reltoken(&unp_token);
1003 
1004 	KASSERT(unp->unp_conn == NULL, ("unp is still connected"));
1005 	KASSERT(LIST_EMPTY(&unp->unp_refs), ("unp still has references"));
1006 
1007 	if (unp->unp_addr)
1008 		kfree(unp->unp_addr, M_SONAME);
1009 	kfree(unp, M_UNPCB);
1010 }
1011 
1012 static int
1013 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
1014 {
1015 	struct proc *p = td->td_proc;
1016 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1017 	struct vnode *vp;
1018 	struct vattr vattr;
1019 	int error, namelen;
1020 	struct nlookupdata nd;
1021 	char buf[SOCK_MAXADDRLEN];
1022 
1023 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
1024 	UNP_ASSERT_TOKEN_HELD(unp);
1025 
1026 	if (unp->unp_vnode != NULL)
1027 		return EINVAL;
1028 
1029 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
1030 	if (namelen <= 0)
1031 		return EINVAL;
1032 	strncpy(buf, soun->sun_path, namelen);
1033 	buf[namelen] = 0;	/* null-terminate the string */
1034 	error = nlookup_init(&nd, buf, UIO_SYSSPACE,
1035 			     NLC_LOCKVP | NLC_CREATE | NLC_REFDVP);
1036 	if (error == 0)
1037 		error = nlookup(&nd);
1038 	if (error == 0 && nd.nl_nch.ncp->nc_vp != NULL)
1039 		error = EADDRINUSE;
1040 	if (error)
1041 		goto done;
1042 
1043 	VATTR_NULL(&vattr);
1044 	vattr.va_type = VSOCK;
1045 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
1046 	error = VOP_NCREATE(&nd.nl_nch, nd.nl_dvp, &vp, nd.nl_cred, &vattr);
1047 	if (error == 0) {
1048 		if (unp->unp_vnode == NULL) {
1049 			vp->v_socket = unp->unp_socket;
1050 			unp->unp_vnode = vp;
1051 			unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam);
1052 			vn_unlock(vp);
1053 		} else {
1054 			vput(vp);		/* late race */
1055 			error = EINVAL;
1056 		}
1057 	}
1058 done:
1059 	nlookup_done(&nd);
1060 	return (error);
1061 }
1062 
1063 static int
1064 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1065 {
1066 	struct unpcb *unp, *unp2;
1067 	int error, flags = 0;
1068 
1069 	lwkt_gettoken(&unp_token);
1070 
1071 	unp = unp_getsocktoken(so);
1072 	if (!UNP_ISATTACHED(unp)) {
1073 		error = EINVAL;
1074 		goto failed;
1075 	}
1076 
1077 	if ((unp->unp_flags & UNP_CONNECTING) || unp->unp_conn != NULL) {
1078 		error = EISCONN;
1079 		goto failed;
1080 	}
1081 
1082 	flags = UNP_CONNECTING;
1083 	unp_setflags(unp, flags);
1084 
1085 	error = unp_find_lockref(nam, td, so->so_type, &unp2);
1086 	if (error)
1087 		goto failed;
1088 	/*
1089 	 * NOTE:
1090 	 * unp2 is locked and referenced.
1091 	 */
1092 
1093 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1094 		struct socket *so2, *so3;
1095 		struct unpcb *unp3;
1096 
1097 		so2 = unp2->unp_socket;
1098 		if (!(so2->so_options & SO_ACCEPTCONN) ||
1099 		    (so3 = sonewconn_faddr(so2, 0, NULL,
1100 		     TRUE /* keep ref */)) == NULL) {
1101 			error = ECONNREFUSED;
1102 			goto done;
1103 		}
1104 		/* so3 has a socket reference. */
1105 
1106 		unp3 = unp_getsocktoken(so3);
1107 		if (!UNP_ISATTACHED(unp3)) {
1108 			unp_reltoken(unp3);
1109 			/*
1110 			 * Already aborted; we only need to drop the
1111 			 * socket reference held by sonewconn_faddr().
1112 			 */
1113 			sofree(so3);
1114 			error = ECONNREFUSED;
1115 			goto done;
1116 		}
1117 		unp_reference(unp3);
1118 		/*
1119 		 * NOTE:
1120 		 * unp3 is locked and referenced.
1121 		 */
1122 
1123 		/*
1124 		 * Release so3 socket reference held by sonewconn_faddr().
1125 		 * Since we have referenced unp3, neither unp3 nor so3 will
1126 		 * be destroyed here.
1127 		 */
1128 		sofree(so3);
1129 
1130 		if (unp2->unp_addr != NULL) {
1131 			unp3->unp_addr = (struct sockaddr_un *)
1132 			    dup_sockaddr((struct sockaddr *)unp2->unp_addr);
1133 		}
1134 
1135 		/*
1136 		 * unp_peercred management:
1137 		 *
1138 		 * The connecter's (client's) credentials are copied
1139 		 * from its process structure at the time of connect()
1140 		 * (which is now).
1141 		 */
1142 		cru2x(td->td_proc->p_ucred, &unp3->unp_peercred);
1143 		unp_setflags(unp3, UNP_HAVEPC);
1144 		/*
1145 		 * The receiver's (server's) credentials are copied
1146 		 * from the unp_peercred member of socket on which the
1147 		 * former called listen(); unp_listen() cached that
1148 		 * process's credentials at that time so we can use
1149 		 * them now.
1150 		 */
1151 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1152 		    ("unp_connect: listener without cached peercred"));
1153 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1154 		    sizeof(unp->unp_peercred));
1155 		unp_setflags(unp, UNP_HAVEPC);
1156 
1157 		error = unp_connect_pair(unp, unp3);
1158 		if (error) {
1159 			/* XXX we need a better name */
1160 			soabort_oncpu(so3);
1161 		}
1162 
1163 		/* Done with unp3 */
1164 		unp_free(unp3);
1165 		unp_reltoken(unp3);
1166 	} else {
1167 		error = unp_connect_pair(unp, unp2);
1168 	}
1169 done:
1170 	unp_free(unp2);
1171 	unp_reltoken(unp2);
1172 failed:
1173 	if (flags)
1174 		unp_clrflags(unp, flags);
1175 	unp_reltoken(unp);
1176 
1177 	lwkt_reltoken(&unp_token);
1178 	return (error);
1179 }
1180 
1181 /*
1182  * Connect two unix domain sockets together.
1183  *
1184  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1185  *	 pool token also be held.
1186  */
1187 int
1188 unp_connect2(struct socket *so, struct socket *so2)
1189 {
1190 	struct unpcb *unp, *unp2;
1191 	int error;
1192 
1193 	lwkt_gettoken(&unp_token);
1194 	if (so2->so_type != so->so_type) {
1195 		lwkt_reltoken(&unp_token);
1196 		return (EPROTOTYPE);
1197 	}
1198 	unp = unp_getsocktoken(so);
1199 	unp2 = unp_getsocktoken(so2);
1200 
1201 	if (!UNP_ISATTACHED(unp)) {
1202 		error = EINVAL;
1203 		goto done;
1204 	}
1205 	if (!UNP_ISATTACHED(unp2)) {
1206 		error = ECONNREFUSED;
1207 		goto done;
1208 	}
1209 
1210 	if (unp->unp_conn != NULL) {
1211 		error = EISCONN;
1212 		goto done;
1213 	}
1214 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1215 	    unp2->unp_conn != NULL) {
1216 		error = EISCONN;
1217 		goto done;
1218 	}
1219 
1220 	error = unp_connect_pair(unp, unp2);
1221 done:
1222 	unp_reltoken(unp2);
1223 	unp_reltoken(unp);
1224 	lwkt_reltoken(&unp_token);
1225 	return (error);
1226 }
1227 
1228 /*
1229  * Disconnect a unix domain socket pair.
1230  *
1231  * NOTE: Semantics for any change to unp_conn requires that the per-unp
1232  *	 pool token also be held.
1233  */
1234 static void
1235 unp_disconnect(struct unpcb *unp, int error)
1236 {
1237 	struct socket *so = unp->unp_socket;
1238 	struct unpcb *unp2;
1239 
1240 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
1241 	UNP_ASSERT_TOKEN_HELD(unp);
1242 
1243 	if (error)
1244 		so->so_error = error;
1245 
1246 	while ((unp2 = unp->unp_conn) != NULL) {
1247 		lwkt_getpooltoken(unp2);
1248 		if (unp2 == unp->unp_conn)
1249 			break;
1250 		lwkt_relpooltoken(unp2);
1251 	}
1252 	if (unp2 == NULL)
1253 		return;
1254 	/* unp2 is locked. */
1255 
1256 	KASSERT((unp2->unp_flags & UNP_DROPPED) == 0, ("unp2 was dropped"));
1257 
1258 	unp->unp_conn = NULL;
1259 
1260 	switch (so->so_type) {
1261 	case SOCK_DGRAM:
1262 		LIST_REMOVE(unp, unp_reflink);
1263 		soclrstate(so, SS_ISCONNECTED);
1264 		break;
1265 
1266 	case SOCK_STREAM:
1267 	case SOCK_SEQPACKET:
1268 		/*
1269 		 * Keep a reference before clearing the unp_conn
1270 		 * to avoid racing uipc_detach()/uipc_abort() in
1271 		 * other thread.
1272 		 */
1273 		unp_reference(unp2);
1274 		KASSERT(unp2->unp_conn == unp, ("unp_conn mismatch"));
1275 		unp2->unp_conn = NULL;
1276 
1277 		soisdisconnected(so);
1278 		soisdisconnected(unp2->unp_socket);
1279 
1280 		unp_free(unp2);
1281 		break;
1282 	}
1283 
1284 	lwkt_relpooltoken(unp2);
1285 }
1286 
1287 #ifdef notdef
1288 void
1289 unp_abort(struct unpcb *unp)
1290 {
1291 	lwkt_gettoken(&unp_token);
1292 	unp_free(unp);
1293 	lwkt_reltoken(&unp_token);
1294 }
1295 #endif
1296 
1297 static int
1298 prison_unpcb(struct thread *td, struct unpcb *unp)
1299 {
1300 	struct proc *p;
1301 
1302 	if (td == NULL)
1303 		return (0);
1304 	if ((p = td->td_proc) == NULL)
1305 		return (0);
1306 	if (!p->p_ucred->cr_prison)
1307 		return (0);
1308 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
1309 		return (0);
1310 	return (1);
1311 }
1312 
1313 static int
1314 unp_pcblist(SYSCTL_HANDLER_ARGS)
1315 {
1316 	struct unp_global_head *head = arg1;
1317 	int error, i, n;
1318 	struct unpcb *unp, *marker;
1319 
1320 	KKASSERT(curproc != NULL);
1321 
1322 	/*
1323 	 * The process of preparing the PCB list is too time-consuming and
1324 	 * resource-intensive to repeat twice on every request.
1325 	 */
1326 	if (req->oldptr == NULL) {
1327 		n = head->count;
1328 		req->oldidx = (n + n/8) * sizeof(struct xunpcb);
1329 		return 0;
1330 	}
1331 
1332 	if (req->newptr != NULL)
1333 		return EPERM;
1334 
1335 	marker = kmalloc(sizeof(*marker), M_UNPCB, M_WAITOK | M_ZERO);
1336 	marker->unp_flags |= UNP_MARKER;
1337 
1338 	lwkt_gettoken(&unp_token);
1339 
1340 	n = head->count;
1341 	i = 0;
1342 	error = 0;
1343 
1344 	TAILQ_INSERT_HEAD(&head->list, marker, unp_link);
1345 	while ((unp = TAILQ_NEXT(marker, unp_link)) != NULL && i < n) {
1346 		struct xunpcb xu;
1347 
1348 		TAILQ_REMOVE(&head->list, marker, unp_link);
1349 		TAILQ_INSERT_AFTER(&head->list, unp, marker, unp_link);
1350 
1351 		if (unp->unp_flags & UNP_MARKER)
1352 			continue;
1353 		if (prison_unpcb(req->td, unp))
1354 			continue;
1355 
1356 		xu.xu_len = sizeof(xu);
1357 		xu.xu_unpp = unp;
1358 
1359 		/*
1360 		 * NOTE:
1361 		 * unp->unp_addr and unp->unp_conn are protected by
1362 		 * unp_token.  So if we want to get rid of unp_token
1363 		 * or reduce the coverage of unp_token, care must be
1364 		 * taken.
1365 		 */
1366 		if (unp->unp_addr) {
1367 			bcopy(unp->unp_addr, &xu.xu_addr,
1368 			      unp->unp_addr->sun_len);
1369 		}
1370 		if (unp->unp_conn && unp->unp_conn->unp_addr) {
1371 			bcopy(unp->unp_conn->unp_addr,
1372 			      &xu.xu_caddr,
1373 			      unp->unp_conn->unp_addr->sun_len);
1374 		}
1375 		bcopy(unp, &xu.xu_unp, sizeof(*unp));
1376 		sotoxsocket(unp->unp_socket, &xu.xu_socket);
1377 
1378 		/* NOTE: This could block and temporarily release unp_token */
1379 		error = SYSCTL_OUT(req, &xu, sizeof(xu));
1380 		if (error)
1381 			break;
1382 		++i;
1383 	}
1384 	TAILQ_REMOVE(&head->list, marker, unp_link);
1385 
1386 	lwkt_reltoken(&unp_token);
1387 
1388 	kfree(marker, M_UNPCB);
1389 	return error;
1390 }
1391 
1392 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1393 	    &unp_dgram_head, 0, unp_pcblist, "S,xunpcb",
1394 	    "List of active local datagram sockets");
1395 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1396 	    &unp_stream_head, 0, unp_pcblist, "S,xunpcb",
1397 	    "List of active local stream sockets");
1398 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLFLAG_RD,
1399 	    &unp_seqpkt_head, 0, unp_pcblist, "S,xunpcb",
1400 	    "List of active local seqpacket sockets");
1401 
1402 static void
1403 unp_shutdown(struct unpcb *unp)
1404 {
1405 	struct socket *so;
1406 
1407 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1408 	     unp->unp_socket->so_type == SOCK_SEQPACKET) &&
1409 	    unp->unp_conn != NULL && (so = unp->unp_conn->unp_socket)) {
1410 		socantrcvmore(so);
1411 	}
1412 }
1413 
1414 #ifdef notdef
1415 void
1416 unp_drain(void)
1417 {
1418 	lwkt_gettoken(&unp_token);
1419 	lwkt_reltoken(&unp_token);
1420 }
1421 #endif
1422 
1423 int
1424 unp_externalize(struct mbuf *rights)
1425 {
1426 	struct thread *td = curthread;
1427 	struct proc *p = td->td_proc;		/* XXX */
1428 	struct lwp *lp = td->td_lwp;
1429 	struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
1430 	int *fdp;
1431 	int i;
1432 	struct file **rp;
1433 	struct file *fp;
1434 	int newfds = (cm->cmsg_len - (CMSG_DATA(cm) - (u_char *)cm))
1435 		/ sizeof(struct file *);
1436 	int f;
1437 
1438 	lwkt_gettoken(&unp_token);
1439 
1440 	/*
1441 	 * if the new FD's will not fit, then we free them all
1442 	 */
1443 	if (!fdavail(p, newfds)) {
1444 		rp = (struct file **)CMSG_DATA(cm);
1445 		for (i = 0; i < newfds; i++) {
1446 			fp = *rp;
1447 			/*
1448 			 * zero the pointer before calling unp_discard,
1449 			 * since it may end up in unp_gc()..
1450 			 */
1451 			*rp++ = NULL;
1452 			unp_discard(fp, NULL);
1453 		}
1454 		lwkt_reltoken(&unp_token);
1455 		return (EMSGSIZE);
1456 	}
1457 
1458 	/*
1459 	 * now change each pointer to an fd in the global table to
1460 	 * an integer that is the index to the local fd table entry
1461 	 * that we set up to point to the global one we are transferring.
1462 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1463 	 * then do it in forward order. In that case, an integer will
1464 	 * always come in the same place or before its corresponding
1465 	 * struct file pointer.
1466 	 * If sizeof (struct file *) is smaller than sizeof int, then
1467 	 * do it in reverse order.
1468 	 */
1469 	if (sizeof(struct file *) >= sizeof(int)) {
1470 		fdp = (int *)CMSG_DATA(cm);
1471 		rp = (struct file **)CMSG_DATA(cm);
1472 		for (i = 0; i < newfds; i++) {
1473 			if (fdalloc(p, 0, &f)) {
1474 				int j;
1475 
1476 				/*
1477 				 * Previous fdavail() can't garantee
1478 				 * fdalloc() success due to SMP race.
1479 				 * Just clean up and return the same
1480 				 * error value as if fdavail() failed.
1481 				 */
1482 
1483 				/* Close externalized files */
1484 				for (j = 0; j < i; j++)
1485 					kern_close(fdp[j]);
1486 				/* Discard the rest of internal files */
1487 				for (; i < newfds; i++)
1488 					unp_discard(rp[i], NULL);
1489 				/* Wipe out the control message */
1490 				for (i = 0; i < newfds; i++)
1491 					rp[i] = NULL;
1492 
1493 				lwkt_reltoken(&unp_token);
1494 				return (EMSGSIZE);
1495 			}
1496 			fp = rp[i];
1497 			unp_fp_externalize(lp, fp, f);
1498 			fdp[i] = f;
1499 		}
1500 	} else {
1501 		/*
1502 		 * XXX
1503 		 * Will this ever happen?  I don't think compiler will
1504 		 * generate code for this code segment -- sephe
1505 		 */
1506 		fdp = (int *)CMSG_DATA(cm) + newfds - 1;
1507 		rp = (struct file **)CMSG_DATA(cm) + newfds - 1;
1508 		for (i = 0; i < newfds; i++) {
1509 			if (fdalloc(p, 0, &f))
1510 				panic("unp_externalize");
1511 			fp = *rp--;
1512 			unp_fp_externalize(lp, fp, f);
1513 			*fdp-- = f;
1514 		}
1515 	}
1516 
1517 	/*
1518 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1519 	 * differs.
1520 	 */
1521 	cm->cmsg_len = CMSG_LEN(newfds * sizeof(int));
1522 	rights->m_len = cm->cmsg_len;
1523 
1524 	lwkt_reltoken(&unp_token);
1525 	return (0);
1526 }
1527 
1528 static void
1529 unp_fp_externalize(struct lwp *lp, struct file *fp, int fd)
1530 {
1531 	lwkt_gettoken(&unp_token);
1532 
1533 	if (lp) {
1534 		KKASSERT(fd >= 0);
1535 		if (fp->f_flag & FREVOKED) {
1536 			struct file *fx;
1537 			int error;
1538 
1539 			kprintf("Warning: revoked fp exiting unix socket\n");
1540 			error = falloc(lp, &fx, NULL);
1541 			if (error == 0) {
1542 				fsetfd(lp->lwp_proc->p_fd, fx, fd);
1543 				fdrop(fx);
1544 			} else {
1545 				fsetfd(lp->lwp_proc->p_fd, NULL, fd);
1546 			}
1547 		} else {
1548 			fsetfd(lp->lwp_proc->p_fd, fp, fd);
1549 		}
1550 	}
1551 	spin_lock(&unp_spin);
1552 	fp->f_msgcount--;
1553 	unp_rights--;
1554 	spin_unlock(&unp_spin);
1555 	fdrop(fp);
1556 
1557 	lwkt_reltoken(&unp_token);
1558 }
1559 
1560 void
1561 unp_init(void)
1562 {
1563 	TAILQ_INIT(&unp_stream_head.list);
1564 	TAILQ_INIT(&unp_dgram_head.list);
1565 	TAILQ_INIT(&unp_seqpkt_head.list);
1566 
1567 	spin_init(&unp_spin, "unpinit");
1568 
1569 	SLIST_INIT(&unp_defdiscard_head);
1570 	spin_init(&unp_defdiscard_spin, "unpdisc");
1571 	TASK_INIT(&unp_defdiscard_task, 0, unp_defdiscard_taskfunc, NULL);
1572 
1573 	/*
1574 	 * Create taskqueue for defered discard, and stick it to
1575 	 * the last CPU.
1576 	 */
1577 	unp_taskqueue = taskqueue_create("unp_taskq", M_WAITOK,
1578 	    taskqueue_thread_enqueue, &unp_taskqueue);
1579 	taskqueue_start_threads(&unp_taskqueue, 1, TDPRI_KERN_DAEMON,
1580 	    ncpus - 1, "unp taskq");
1581 }
1582 
1583 static int
1584 unp_internalize(struct mbuf *control, struct thread *td)
1585 {
1586 	struct proc *p = td->td_proc;
1587 	struct filedesc *fdescp;
1588 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1589 	struct file **rp;
1590 	struct file *fp;
1591 	int i, fd, *fdp;
1592 	struct cmsgcred *cmcred;
1593 	int oldfds;
1594 	u_int newlen;
1595 	int error;
1596 
1597 	KKASSERT(p);
1598 
1599 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
1600 	    cm->cmsg_level != SOL_SOCKET ||
1601 	    CMSG_ALIGN(cm->cmsg_len) != control->m_len)
1602 		return EINVAL;
1603 
1604 	/*
1605 	 * Fill in credential information.
1606 	 */
1607 	if (cm->cmsg_type == SCM_CREDS) {
1608 		cmcred = (struct cmsgcred *)CMSG_DATA(cm);
1609 		cmcred->cmcred_pid = p->p_pid;
1610 		cmcred->cmcred_uid = p->p_ucred->cr_ruid;
1611 		cmcred->cmcred_gid = p->p_ucred->cr_rgid;
1612 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
1613 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
1614 							CMGROUP_MAX);
1615 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
1616 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
1617 		return 0;
1618 	}
1619 
1620 	/*
1621 	 * cmsghdr may not be aligned, do not allow calculation(s) to
1622 	 * go negative.
1623 	 */
1624 	if (cm->cmsg_len < CMSG_LEN(0))
1625 		return EINVAL;
1626 
1627 	oldfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1628 
1629 	/*
1630 	 * Now replace the integer FDs with pointers to
1631 	 * the associated global file table entry..
1632 	 * Allocate a bigger buffer as necessary. But if an cluster is not
1633 	 * enough, return E2BIG.
1634 	 */
1635 	newlen = CMSG_LEN(oldfds * sizeof(struct file *));
1636 	if (newlen > MCLBYTES)
1637 		return E2BIG;
1638 	if (newlen - control->m_len > M_TRAILINGSPACE(control)) {
1639 		if (control->m_flags & M_EXT)
1640 			return E2BIG;
1641 		MCLGET(control, M_WAITOK);
1642 		if (!(control->m_flags & M_EXT))
1643 			return ENOBUFS;
1644 
1645 		/* copy the data to the cluster */
1646 		memcpy(mtod(control, char *), cm, cm->cmsg_len);
1647 		cm = mtod(control, struct cmsghdr *);
1648 	}
1649 
1650 	fdescp = p->p_fd;
1651 	spin_lock_shared(&fdescp->fd_spin);
1652 
1653 	/*
1654 	 * check that all the FDs passed in refer to legal OPEN files
1655 	 * If not, reject the entire operation.
1656 	 */
1657 	fdp = (int *)CMSG_DATA(cm);
1658 	for (i = 0; i < oldfds; i++) {
1659 		fd = *fdp++;
1660 		if ((unsigned)fd >= fdescp->fd_nfiles ||
1661 		    fdescp->fd_files[fd].fp == NULL) {
1662 			error = EBADF;
1663 			goto done;
1664 		}
1665 		if (fdescp->fd_files[fd].fp->f_type == DTYPE_KQUEUE) {
1666 			error = EOPNOTSUPP;
1667 			goto done;
1668 		}
1669 	}
1670 
1671 	/*
1672 	 * Adjust length, in case sizeof(struct file *) and sizeof(int)
1673 	 * differs.
1674 	 */
1675 	cm->cmsg_len = newlen;
1676 	control->m_len = CMSG_ALIGN(newlen);
1677 
1678 	/*
1679 	 * Transform the file descriptors into struct file pointers.
1680 	 * If sizeof (struct file *) is bigger than or equal to sizeof int,
1681 	 * then do it in reverse order so that the int won't get until
1682 	 * we're done.
1683 	 * If sizeof (struct file *) is smaller than sizeof int, then
1684 	 * do it in forward order.
1685 	 */
1686 	if (sizeof(struct file *) >= sizeof(int)) {
1687 		fdp = (int *)CMSG_DATA(cm) + oldfds - 1;
1688 		rp = (struct file **)CMSG_DATA(cm) + oldfds - 1;
1689 		for (i = 0; i < oldfds; i++) {
1690 			fp = fdescp->fd_files[*fdp--].fp;
1691 			*rp-- = fp;
1692 			fhold(fp);
1693 			spin_lock(&unp_spin);
1694 			fp->f_msgcount++;
1695 			unp_rights++;
1696 			spin_unlock(&unp_spin);
1697 		}
1698 	} else {
1699 		/*
1700 		 * XXX
1701 		 * Will this ever happen?  I don't think compiler will
1702 		 * generate code for this code segment -- sephe
1703 		 */
1704 		fdp = (int *)CMSG_DATA(cm);
1705 		rp = (struct file **)CMSG_DATA(cm);
1706 		for (i = 0; i < oldfds; i++) {
1707 			fp = fdescp->fd_files[*fdp++].fp;
1708 			*rp++ = fp;
1709 			fhold(fp);
1710 			spin_lock(&unp_spin);
1711 			fp->f_msgcount++;
1712 			unp_rights++;
1713 			spin_unlock(&unp_spin);
1714 		}
1715 	}
1716 	error = 0;
1717 done:
1718 	spin_unlock_shared(&fdescp->fd_spin);
1719 	return error;
1720 }
1721 
1722 /*
1723  * Garbage collect in-transit file descriptors that get lost due to
1724  * loops (i.e. when a socket is sent to another process over itself,
1725  * and more complex situations).
1726  *
1727  * NOT MPSAFE - TODO socket flush code and maybe closef.  Rest is MPSAFE.
1728  */
1729 
1730 struct unp_gc_info {
1731 	struct file **extra_ref;
1732 	struct file *locked_fp;
1733 	int defer;
1734 	int index;
1735 	int maxindex;
1736 };
1737 
1738 static void
1739 unp_gc(void)
1740 {
1741 	struct unp_gc_info info;
1742 	static boolean_t unp_gcing;
1743 	struct file **fpp;
1744 	int i;
1745 
1746 	/*
1747 	 * Only one gc can be in-progress at any given moment
1748 	 */
1749 	spin_lock(&unp_spin);
1750 	if (unp_gcing) {
1751 		spin_unlock(&unp_spin);
1752 		return;
1753 	}
1754 	unp_gcing = TRUE;
1755 	spin_unlock(&unp_spin);
1756 
1757 	lwkt_gettoken(&unp_token);
1758 
1759 	/*
1760 	 * Before going through all this, set all FDs to be NOT defered
1761 	 * and NOT externally accessible (not marked).  During the scan
1762 	 * a fd can be marked externally accessible but we may or may not
1763 	 * be able to immediately process it (controlled by FDEFER).
1764 	 *
1765 	 * If we loop sleep a bit.  The complexity of the topology can cause
1766 	 * multiple loops.  Also failure to acquire the socket's so_rcv
1767 	 * token can cause us to loop.
1768 	 */
1769 	allfiles_scan_exclusive(unp_gc_clearmarks, NULL);
1770 	do {
1771 		info.defer = 0;
1772 		allfiles_scan_exclusive(unp_gc_checkmarks, &info);
1773 		if (info.defer)
1774 			tsleep(&info, 0, "gcagain", 1);
1775 	} while (info.defer);
1776 
1777 	/*
1778 	 * We grab an extra reference to each of the file table entries
1779 	 * that are not otherwise accessible and then free the rights
1780 	 * that are stored in messages on them.
1781 	 *
1782 	 * The bug in the orginal code is a little tricky, so I'll describe
1783 	 * what's wrong with it here.
1784 	 *
1785 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1786 	 * times -- consider the case of sockets A and B that contain
1787 	 * references to each other.  On a last close of some other socket,
1788 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1789 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1790 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1791 	 * results in the following chain.  Closef calls soo_close, which
1792 	 * calls soclose.   Soclose calls first (through the switch
1793 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1794 	 * returns because the previous instance had set unp_gcing, and
1795 	 * we return all the way back to soclose, which marks the socket
1796 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1797 	 * to free up the rights that are queued in messages on the socket A,
1798 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1799 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1800 	 * instance of unp_discard just calls closef on B.
1801 	 *
1802 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1803 	 * which results in another closef on A.  Unfortunately, A is already
1804 	 * being closed, and the descriptor has already been marked with
1805 	 * SS_NOFDREF, and soclose panics at this point.
1806 	 *
1807 	 * Here, we first take an extra reference to each inaccessible
1808 	 * descriptor.  Then, we call sorflush ourself, since we know
1809 	 * it is a Unix domain socket anyhow.  After we destroy all the
1810 	 * rights carried in messages, we do a last closef to get rid
1811 	 * of our extra reference.  This is the last close, and the
1812 	 * unp_detach etc will shut down the socket.
1813 	 *
1814 	 * 91/09/19, bsy@cs.cmu.edu
1815 	 */
1816 	info.extra_ref = kmalloc(256 * sizeof(struct file *), M_FILE, M_WAITOK);
1817 	info.maxindex = 256;
1818 
1819 	do {
1820 		/*
1821 		 * Look for matches
1822 		 */
1823 		info.index = 0;
1824 		allfiles_scan_exclusive(unp_gc_checkrefs, &info);
1825 
1826 		/*
1827 		 * For each FD on our hit list, do the following two things
1828 		 */
1829 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp) {
1830 			struct file *tfp = *fpp;
1831 			if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1832 				sorflush((struct socket *)(tfp->f_data));
1833 		}
1834 		for (i = info.index, fpp = info.extra_ref; --i >= 0; ++fpp)
1835 			closef(*fpp, NULL);
1836 	} while (info.index == info.maxindex);
1837 
1838 	lwkt_reltoken(&unp_token);
1839 
1840 	kfree((caddr_t)info.extra_ref, M_FILE);
1841 	unp_gcing = FALSE;
1842 }
1843 
1844 /*
1845  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1846  */
1847 static int
1848 unp_gc_checkrefs(struct file *fp, void *data)
1849 {
1850 	struct unp_gc_info *info = data;
1851 
1852 	if (fp->f_count == 0)
1853 		return(0);
1854 	if (info->index == info->maxindex)
1855 		return(-1);
1856 
1857 	/*
1858 	 * If all refs are from msgs, and it's not marked accessible
1859 	 * then it must be referenced from some unreachable cycle
1860 	 * of (shut-down) FDs, so include it in our
1861 	 * list of FDs to remove
1862 	 */
1863 	if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1864 		info->extra_ref[info->index++] = fp;
1865 		fhold(fp);
1866 	}
1867 	return(0);
1868 }
1869 
1870 /*
1871  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1872  */
1873 static int
1874 unp_gc_clearmarks(struct file *fp, void *data __unused)
1875 {
1876 	atomic_clear_int(&fp->f_flag, FMARK | FDEFER);
1877 	return(0);
1878 }
1879 
1880 /*
1881  * MPSAFE - NOTE: filehead list and file pointer spinlocked on entry
1882  */
1883 static int
1884 unp_gc_checkmarks(struct file *fp, void *data)
1885 {
1886 	struct unp_gc_info *info = data;
1887 	struct socket *so;
1888 
1889 	/*
1890 	 * If the file is not open, skip it.  Make sure it isn't marked
1891 	 * defered or we could loop forever, in case we somehow race
1892 	 * something.
1893 	 */
1894 	if (fp->f_count == 0) {
1895 		if (fp->f_flag & FDEFER)
1896 			atomic_clear_int(&fp->f_flag, FDEFER);
1897 		return(0);
1898 	}
1899 	/*
1900 	 * If we already marked it as 'defer'  in a
1901 	 * previous pass, then try process it this time
1902 	 * and un-mark it
1903 	 */
1904 	if (fp->f_flag & FDEFER) {
1905 		atomic_clear_int(&fp->f_flag, FDEFER);
1906 	} else {
1907 		/*
1908 		 * if it's not defered, then check if it's
1909 		 * already marked.. if so skip it
1910 		 */
1911 		if (fp->f_flag & FMARK)
1912 			return(0);
1913 		/*
1914 		 * If all references are from messages
1915 		 * in transit, then skip it. it's not
1916 		 * externally accessible.
1917 		 */
1918 		if (fp->f_count == fp->f_msgcount)
1919 			return(0);
1920 		/*
1921 		 * If it got this far then it must be
1922 		 * externally accessible.
1923 		 */
1924 		atomic_set_int(&fp->f_flag, FMARK);
1925 	}
1926 
1927 	/*
1928 	 * either it was defered, or it is externally
1929 	 * accessible and not already marked so.
1930 	 * Now check if it is possibly one of OUR sockets.
1931 	 */
1932 	if (fp->f_type != DTYPE_SOCKET ||
1933 	    (so = (struct socket *)fp->f_data) == NULL) {
1934 		return(0);
1935 	}
1936 	if (so->so_proto->pr_domain != &localdomain ||
1937 	    !(so->so_proto->pr_flags & PR_RIGHTS)) {
1938 		return(0);
1939 	}
1940 
1941 	/*
1942 	 * So, Ok, it's one of our sockets and it IS externally accessible
1943 	 * (or was defered).  Now we look to see if we hold any file
1944 	 * descriptors in its message buffers.  Follow those links and mark
1945 	 * them as accessible too.
1946 	 *
1947 	 * We are holding multiple spinlocks here, if we cannot get the
1948 	 * token non-blocking defer until the next loop.
1949 	 */
1950 	info->locked_fp = fp;
1951 	if (lwkt_trytoken(&so->so_rcv.ssb_token)) {
1952 		unp_scan(so->so_rcv.ssb_mb, unp_mark, info);
1953 		lwkt_reltoken(&so->so_rcv.ssb_token);
1954 	} else {
1955 		atomic_set_int(&fp->f_flag, FDEFER);
1956 		++info->defer;
1957 	}
1958 	return (0);
1959 }
1960 
1961 /*
1962  * Scan all unix domain sockets and replace any revoked file pointers
1963  * found with the dummy file pointer fx.  We don't worry about races
1964  * against file pointers being read out as those are handled in the
1965  * externalize code.
1966  */
1967 
1968 #define REVOKE_GC_MAXFILES	32
1969 
1970 struct unp_revoke_gc_info {
1971 	struct file	*fx;
1972 	struct file	*fary[REVOKE_GC_MAXFILES];
1973 	int		fcount;
1974 };
1975 
1976 void
1977 unp_revoke_gc(struct file *fx)
1978 {
1979 	struct unp_revoke_gc_info info;
1980 	int i;
1981 
1982 	lwkt_gettoken(&unp_token);
1983 	info.fx = fx;
1984 	do {
1985 		info.fcount = 0;
1986 		allfiles_scan_exclusive(unp_revoke_gc_check, &info);
1987 		for (i = 0; i < info.fcount; ++i)
1988 			unp_fp_externalize(NULL, info.fary[i], -1);
1989 	} while (info.fcount == REVOKE_GC_MAXFILES);
1990 	lwkt_reltoken(&unp_token);
1991 }
1992 
1993 /*
1994  * Check for and replace revoked descriptors.
1995  *
1996  * WARNING:  This routine is not allowed to block.
1997  */
1998 static int
1999 unp_revoke_gc_check(struct file *fps, void *vinfo)
2000 {
2001 	struct unp_revoke_gc_info *info = vinfo;
2002 	struct file *fp;
2003 	struct socket *so;
2004 	struct mbuf *m0;
2005 	struct mbuf *m;
2006 	struct file **rp;
2007 	struct cmsghdr *cm;
2008 	int i;
2009 	int qfds;
2010 
2011 	/*
2012 	 * Is this a unix domain socket with rights-passing abilities?
2013 	 */
2014 	if (fps->f_type != DTYPE_SOCKET)
2015 		return (0);
2016 	if ((so = (struct socket *)fps->f_data) == NULL)
2017 		return(0);
2018 	if (so->so_proto->pr_domain != &localdomain)
2019 		return(0);
2020 	if ((so->so_proto->pr_flags & PR_RIGHTS) == 0)
2021 		return(0);
2022 
2023 	/*
2024 	 * Scan the mbufs for control messages and replace any revoked
2025 	 * descriptors we find.
2026 	 */
2027 	lwkt_gettoken(&so->so_rcv.ssb_token);
2028 	m0 = so->so_rcv.ssb_mb;
2029 	while (m0) {
2030 		for (m = m0; m; m = m->m_next) {
2031 			if (m->m_type != MT_CONTROL)
2032 				continue;
2033 			if (m->m_len < sizeof(*cm))
2034 				continue;
2035 			cm = mtod(m, struct cmsghdr *);
2036 			if (cm->cmsg_level != SOL_SOCKET ||
2037 			    cm->cmsg_type != SCM_RIGHTS) {
2038 				continue;
2039 			}
2040 			qfds = (cm->cmsg_len - CMSG_LEN(0)) / sizeof(void *);
2041 			rp = (struct file **)CMSG_DATA(cm);
2042 			for (i = 0; i < qfds; i++) {
2043 				fp = rp[i];
2044 				if (fp->f_flag & FREVOKED) {
2045 					kprintf("Warning: Removing revoked fp from unix domain socket queue\n");
2046 					fhold(info->fx);
2047 					info->fx->f_msgcount++;
2048 					unp_rights++;
2049 					rp[i] = info->fx;
2050 					info->fary[info->fcount++] = fp;
2051 				}
2052 				if (info->fcount == REVOKE_GC_MAXFILES)
2053 					break;
2054 			}
2055 			if (info->fcount == REVOKE_GC_MAXFILES)
2056 				break;
2057 		}
2058 		m0 = m0->m_nextpkt;
2059 		if (info->fcount == REVOKE_GC_MAXFILES)
2060 			break;
2061 	}
2062 	lwkt_reltoken(&so->so_rcv.ssb_token);
2063 
2064 	/*
2065 	 * Stop the scan if we filled up our array.
2066 	 */
2067 	if (info->fcount == REVOKE_GC_MAXFILES)
2068 		return(-1);
2069 	return(0);
2070 }
2071 
2072 /*
2073  * Dispose of the fp's stored in a mbuf.
2074  *
2075  * The dds loop can cause additional fps to be entered onto the
2076  * list while it is running, flattening out the operation and avoiding
2077  * a deep kernel stack recursion.
2078  */
2079 void
2080 unp_dispose(struct mbuf *m)
2081 {
2082 	lwkt_gettoken(&unp_token);
2083 	if (m)
2084 		unp_scan(m, unp_discard, NULL);
2085 	lwkt_reltoken(&unp_token);
2086 }
2087 
2088 static int
2089 unp_listen(struct unpcb *unp, struct thread *td)
2090 {
2091 	struct proc *p = td->td_proc;
2092 
2093 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2094 	UNP_ASSERT_TOKEN_HELD(unp);
2095 
2096 	KKASSERT(p);
2097 	cru2x(p->p_ucred, &unp->unp_peercred);
2098 	unp_setflags(unp, UNP_HAVEPCCACHED);
2099 	return (0);
2100 }
2101 
2102 static void
2103 unp_scan(struct mbuf *m0, void (*op)(struct file *, void *), void *data)
2104 {
2105 	struct mbuf *m;
2106 	struct file **rp;
2107 	struct cmsghdr *cm;
2108 	int i;
2109 	int qfds;
2110 
2111 	while (m0) {
2112 		for (m = m0; m; m = m->m_next) {
2113 			if (m->m_type == MT_CONTROL &&
2114 			    m->m_len >= sizeof(*cm)) {
2115 				cm = mtod(m, struct cmsghdr *);
2116 				if (cm->cmsg_level != SOL_SOCKET ||
2117 				    cm->cmsg_type != SCM_RIGHTS)
2118 					continue;
2119 				qfds = (cm->cmsg_len - CMSG_LEN(0)) /
2120 					sizeof(void *);
2121 				rp = (struct file **)CMSG_DATA(cm);
2122 				for (i = 0; i < qfds; i++)
2123 					(*op)(*rp++, data);
2124 				break;		/* XXX, but saves time */
2125 			}
2126 		}
2127 		m0 = m0->m_nextpkt;
2128 	}
2129 }
2130 
2131 /*
2132  * Mark visibility.  info->defer is recalculated on every pass.
2133  */
2134 static void
2135 unp_mark(struct file *fp, void *data)
2136 {
2137 	struct unp_gc_info *info = data;
2138 
2139 	if ((fp->f_flag & FMARK) == 0) {
2140 		++info->defer;
2141 		atomic_set_int(&fp->f_flag, FMARK | FDEFER);
2142 	} else if (fp->f_flag & FDEFER) {
2143 		++info->defer;
2144 	}
2145 }
2146 
2147 /*
2148  * Discard a fp previously held in a unix domain socket mbuf.  To
2149  * avoid blowing out the kernel stack due to contrived chain-reactions
2150  * we may have to defer the operation to a higher procedural level.
2151  *
2152  * Caller holds unp_token
2153  */
2154 static void
2155 unp_discard(struct file *fp, void *data __unused)
2156 {
2157 	struct unp_defdiscard *d;
2158 
2159 	spin_lock(&unp_spin);
2160 	fp->f_msgcount--;
2161 	unp_rights--;
2162 	spin_unlock(&unp_spin);
2163 
2164 	d = kmalloc(sizeof(*d), M_UNPCB, M_WAITOK);
2165 	d->fp = fp;
2166 
2167 	spin_lock(&unp_defdiscard_spin);
2168 	SLIST_INSERT_HEAD(&unp_defdiscard_head, d, next);
2169 	spin_unlock(&unp_defdiscard_spin);
2170 
2171 	taskqueue_enqueue(unp_taskqueue, &unp_defdiscard_task);
2172 }
2173 
2174 static int
2175 unp_find_lockref(struct sockaddr *nam, struct thread *td, short type,
2176     struct unpcb **unp_ret)
2177 {
2178 	struct proc *p = td->td_proc;
2179 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
2180 	struct vnode *vp = NULL;
2181 	struct socket *so;
2182 	struct unpcb *unp;
2183 	int error, len;
2184 	struct nlookupdata nd;
2185 	char buf[SOCK_MAXADDRLEN];
2186 
2187 	*unp_ret = NULL;
2188 
2189 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
2190 	if (len <= 0) {
2191 		error = EINVAL;
2192 		goto failed;
2193 	}
2194 	strncpy(buf, soun->sun_path, len);
2195 	buf[len] = 0;
2196 
2197 	error = nlookup_init(&nd, buf, UIO_SYSSPACE, NLC_FOLLOW);
2198 	if (error == 0)
2199 		error = nlookup(&nd);
2200 	if (error == 0)
2201 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2202 	nlookup_done(&nd);
2203 	if (error) {
2204 		vp = NULL;
2205 		goto failed;
2206 	}
2207 
2208 	if (vp->v_type != VSOCK) {
2209 		error = ENOTSOCK;
2210 		goto failed;
2211 	}
2212 	error = VOP_EACCESS(vp, VWRITE, p->p_ucred);
2213 	if (error)
2214 		goto failed;
2215 	so = vp->v_socket;
2216 	if (so == NULL) {
2217 		error = ECONNREFUSED;
2218 		goto failed;
2219 	}
2220 	if (so->so_type != type) {
2221 		error = EPROTOTYPE;
2222 		goto failed;
2223 	}
2224 
2225 	/* Lock this unp. */
2226 	unp = unp_getsocktoken(so);
2227 	if (!UNP_ISATTACHED(unp)) {
2228 		unp_reltoken(unp);
2229 		error = ECONNREFUSED;
2230 		goto failed;
2231 	}
2232 	/* And keep this unp referenced. */
2233 	unp_reference(unp);
2234 
2235 	/* Done! */
2236 	*unp_ret = unp;
2237 	error = 0;
2238 failed:
2239 	if (vp != NULL)
2240 		vput(vp);
2241 	return error;
2242 }
2243 
2244 static int
2245 unp_connect_pair(struct unpcb *unp, struct unpcb *unp2)
2246 {
2247 	struct socket *so = unp->unp_socket;
2248 	struct socket *so2 = unp2->unp_socket;
2249 
2250 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2251 	UNP_ASSERT_TOKEN_HELD(unp);
2252 	UNP_ASSERT_TOKEN_HELD(unp2);
2253 
2254 	KASSERT(so->so_type == so2->so_type,
2255 	    ("socket type mismatch, so %d, so2 %d", so->so_type, so2->so_type));
2256 
2257 	if (!UNP_ISATTACHED(unp))
2258 		return EINVAL;
2259 	if (!UNP_ISATTACHED(unp2))
2260 		return ECONNREFUSED;
2261 
2262 	KASSERT(unp->unp_conn == NULL, ("unp is already connected"));
2263 	unp->unp_conn = unp2;
2264 
2265 	switch (so->so_type) {
2266 	case SOCK_DGRAM:
2267 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2268 		soisconnected(so);
2269 		break;
2270 
2271 	case SOCK_STREAM:
2272 	case SOCK_SEQPACKET:
2273 		KASSERT(unp2->unp_conn == NULL, ("unp2 is already connected"));
2274 		unp2->unp_conn = unp;
2275 		soisconnected(so);
2276 		soisconnected(so2);
2277 		break;
2278 
2279 	default:
2280 		panic("unp_connect_pair: unknown socket type %d", so->so_type);
2281 	}
2282 	return 0;
2283 }
2284 
2285 static void
2286 unp_drop(struct unpcb *unp, int error)
2287 {
2288 	struct unpcb *unp2;
2289 
2290 	ASSERT_LWKT_TOKEN_HELD(&unp_token);
2291 	UNP_ASSERT_TOKEN_HELD(unp);
2292 	KASSERT(unp->unp_flags & UNP_DETACHED, ("unp is not detached"));
2293 
2294 	unp_disconnect(unp, error);
2295 
2296 	while ((unp2 = LIST_FIRST(&unp->unp_refs)) != NULL) {
2297 		lwkt_getpooltoken(unp2);
2298 		unp_disconnect(unp2, ECONNRESET);
2299 		lwkt_relpooltoken(unp2);
2300 	}
2301 	unp_setflags(unp, UNP_DROPPED);
2302 }
2303 
2304 static void
2305 unp_defdiscard_taskfunc(void *arg __unused, int pending __unused)
2306 {
2307 	struct unp_defdiscard *d;
2308 
2309 	spin_lock(&unp_defdiscard_spin);
2310 	while ((d = SLIST_FIRST(&unp_defdiscard_head)) != NULL) {
2311 		SLIST_REMOVE_HEAD(&unp_defdiscard_head, next);
2312 		spin_unlock(&unp_defdiscard_spin);
2313 
2314 		closef(d->fp, NULL);
2315 		kfree(d, M_UNPCB);
2316 
2317 		spin_lock(&unp_defdiscard_spin);
2318 	}
2319 	spin_unlock(&unp_defdiscard_spin);
2320 }
2321