xref: /freebsd/sys/kern/uipc_mqueue.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2005 David Xu <davidxu@freebsd.org>
5  * Copyright (c) 2016-2017 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * Portions of this software were developed by BAE Systems, the University of
9  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
10  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
11  * Computing (TC) research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 /*
37  * POSIX message queue implementation.
38  *
39  * 1) A mqueue filesystem can be mounted, each message queue appears
40  *    in mounted directory, user can change queue's permission and
41  *    ownership, or remove a queue. Manually creating a file in the
42  *    directory causes a message queue to be created in the kernel with
43  *    default message queue attributes applied and same name used, this
44  *    method is not advocated since mq_open syscall allows user to specify
45  *    different attributes. Also the file system can be mounted multiple
46  *    times at different mount points but shows same contents.
47  *
48  * 2) Standard POSIX message queue API. The syscalls do not use vfs layer,
49  *    but directly operate on internal data structure, this allows user to
50  *    use the IPC facility without having to mount mqueue file system.
51  */
52 
53 #include <sys/cdefs.h>
54 __FBSDID("$FreeBSD$");
55 
56 #include "opt_capsicum.h"
57 
58 #include <sys/param.h>
59 #include <sys/kernel.h>
60 #include <sys/systm.h>
61 #include <sys/limits.h>
62 #include <sys/malloc.h>
63 #include <sys/buf.h>
64 #include <sys/capsicum.h>
65 #include <sys/dirent.h>
66 #include <sys/event.h>
67 #include <sys/eventhandler.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/filedesc.h>
71 #include <sys/jail.h>
72 #include <sys/lock.h>
73 #include <sys/module.h>
74 #include <sys/mount.h>
75 #include <sys/mqueue.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
78 #include <sys/posix4.h>
79 #include <sys/poll.h>
80 #include <sys/priv.h>
81 #include <sys/proc.h>
82 #include <sys/queue.h>
83 #include <sys/sysproto.h>
84 #include <sys/stat.h>
85 #include <sys/syscall.h>
86 #include <sys/syscallsubr.h>
87 #include <sys/sysent.h>
88 #include <sys/sx.h>
89 #include <sys/sysctl.h>
90 #include <sys/taskqueue.h>
91 #include <sys/unistd.h>
92 #include <sys/user.h>
93 #include <sys/vnode.h>
94 #include <machine/atomic.h>
95 
96 #include <security/audit/audit.h>
97 
98 FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
99 
100 /*
101  * Limits and constants
102  */
103 #define	MQFS_NAMELEN		NAME_MAX
104 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
105 
106 /* node types */
107 typedef enum {
108 	mqfstype_none = 0,
109 	mqfstype_root,
110 	mqfstype_dir,
111 	mqfstype_this,
112 	mqfstype_parent,
113 	mqfstype_file,
114 	mqfstype_symlink,
115 } mqfs_type_t;
116 
117 struct mqfs_node;
118 
119 /*
120  * mqfs_info: describes a mqfs instance
121  */
122 struct mqfs_info {
123 	struct sx		mi_lock;
124 	struct mqfs_node	*mi_root;
125 	struct unrhdr		*mi_unrhdr;
126 };
127 
128 struct mqfs_vdata {
129 	LIST_ENTRY(mqfs_vdata)	mv_link;
130 	struct mqfs_node	*mv_node;
131 	struct vnode		*mv_vnode;
132 	struct task		mv_task;
133 };
134 
135 /*
136  * mqfs_node: describes a node (file or directory) within a mqfs
137  */
138 struct mqfs_node {
139 	char			mn_name[MQFS_NAMELEN+1];
140 	struct mqfs_info	*mn_info;
141 	struct mqfs_node	*mn_parent;
142 	LIST_HEAD(,mqfs_node)	mn_children;
143 	LIST_ENTRY(mqfs_node)	mn_sibling;
144 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
145 	const void		*mn_pr_root;
146 	int			mn_refcount;
147 	mqfs_type_t		mn_type;
148 	int			mn_deleted;
149 	uint32_t		mn_fileno;
150 	void			*mn_data;
151 	struct timespec		mn_birth;
152 	struct timespec		mn_ctime;
153 	struct timespec		mn_atime;
154 	struct timespec		mn_mtime;
155 	uid_t			mn_uid;
156 	gid_t			mn_gid;
157 	int			mn_mode;
158 };
159 
160 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
161 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
162 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
163 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
164 				(fp)->f_data)->mn_data))
165 
166 TAILQ_HEAD(msgq, mqueue_msg);
167 
168 struct mqueue;
169 
170 struct mqueue_notifier {
171 	LIST_ENTRY(mqueue_notifier)	nt_link;
172 	struct sigevent			nt_sigev;
173 	ksiginfo_t			nt_ksi;
174 	struct proc			*nt_proc;
175 };
176 
177 struct mqueue {
178 	struct mtx	mq_mutex;
179 	int		mq_flags;
180 	long		mq_maxmsg;
181 	long		mq_msgsize;
182 	long		mq_curmsgs;
183 	long		mq_totalbytes;
184 	struct msgq	mq_msgq;
185 	int		mq_receivers;
186 	int		mq_senders;
187 	struct selinfo	mq_rsel;
188 	struct selinfo	mq_wsel;
189 	struct mqueue_notifier	*mq_notifier;
190 };
191 
192 #define	MQ_RSEL		0x01
193 #define	MQ_WSEL		0x02
194 
195 struct mqueue_msg {
196 	TAILQ_ENTRY(mqueue_msg)	msg_link;
197 	unsigned int	msg_prio;
198 	unsigned int	msg_size;
199 	/* following real data... */
200 };
201 
202 static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203 	"POSIX real time message queue");
204 
205 static int	default_maxmsg  = 10;
206 static int	default_msgsize = 1024;
207 
208 static int	maxmsg = 100;
209 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
210     &maxmsg, 0, "Default maximum messages in queue");
211 static int	maxmsgsize = 16384;
212 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
213     &maxmsgsize, 0, "Default maximum message size");
214 static int	maxmq = 100;
215 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
216     &maxmq, 0, "maximum message queues");
217 static int	curmq = 0;
218 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
219     &curmq, 0, "current message queue number");
220 static int	unloadable = 0;
221 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
222 
223 static eventhandler_tag exit_tag;
224 
225 /* Only one instance per-system */
226 static struct mqfs_info		mqfs_data;
227 static uma_zone_t		mqnode_zone;
228 static uma_zone_t		mqueue_zone;
229 static uma_zone_t		mvdata_zone;
230 static uma_zone_t		mqnoti_zone;
231 static struct vop_vector	mqfs_vnodeops;
232 static struct fileops		mqueueops;
233 static unsigned			mqfs_osd_jail_slot;
234 
235 /*
236  * Directory structure construction and manipulation
237  */
238 #ifdef notyet
239 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
240 	const char *name, int namelen, struct ucred *cred, int mode);
241 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
242 	const char *name, int namelen, struct ucred *cred, int mode);
243 #endif
244 
245 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
246 	const char *name, int namelen, struct ucred *cred, int mode);
247 static int	mqfs_destroy(struct mqfs_node *mn);
248 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
249 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
250 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
251 static int	mqfs_prison_remove(void *obj, void *data);
252 
253 /*
254  * Message queue construction and maniplation
255  */
256 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
257 static void	mqueue_free(struct mqueue *mq);
258 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
259 			size_t msg_len, unsigned msg_prio, int waitok,
260 			const struct timespec *abs_timeout);
261 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
262 			size_t msg_len, unsigned *msg_prio, int waitok,
263 			const struct timespec *abs_timeout);
264 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
265 			int timo);
266 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
267 			int timo);
268 static void	mqueue_send_notification(struct mqueue *mq);
269 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
270 static void	mq_proc_exit(void *arg, struct proc *p);
271 
272 /*
273  * kqueue filters
274  */
275 static void	filt_mqdetach(struct knote *kn);
276 static int	filt_mqread(struct knote *kn, long hint);
277 static int	filt_mqwrite(struct knote *kn, long hint);
278 
279 struct filterops mq_rfiltops = {
280 	.f_isfd = 1,
281 	.f_detach = filt_mqdetach,
282 	.f_event = filt_mqread,
283 };
284 struct filterops mq_wfiltops = {
285 	.f_isfd = 1,
286 	.f_detach = filt_mqdetach,
287 	.f_event = filt_mqwrite,
288 };
289 
290 /*
291  * Initialize fileno bitmap
292  */
293 static void
294 mqfs_fileno_init(struct mqfs_info *mi)
295 {
296 	struct unrhdr *up;
297 
298 	up = new_unrhdr(1, INT_MAX, NULL);
299 	mi->mi_unrhdr = up;
300 }
301 
302 /*
303  * Tear down fileno bitmap
304  */
305 static void
306 mqfs_fileno_uninit(struct mqfs_info *mi)
307 {
308 	struct unrhdr *up;
309 
310 	up = mi->mi_unrhdr;
311 	mi->mi_unrhdr = NULL;
312 	delete_unrhdr(up);
313 }
314 
315 /*
316  * Allocate a file number
317  */
318 static void
319 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
320 {
321 	/* make sure our parent has a file number */
322 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
323 		mqfs_fileno_alloc(mi, mn->mn_parent);
324 
325 	switch (mn->mn_type) {
326 	case mqfstype_root:
327 	case mqfstype_dir:
328 	case mqfstype_file:
329 	case mqfstype_symlink:
330 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
331 		break;
332 	case mqfstype_this:
333 		KASSERT(mn->mn_parent != NULL,
334 		    ("mqfstype_this node has no parent"));
335 		mn->mn_fileno = mn->mn_parent->mn_fileno;
336 		break;
337 	case mqfstype_parent:
338 		KASSERT(mn->mn_parent != NULL,
339 		    ("mqfstype_parent node has no parent"));
340 		if (mn->mn_parent == mi->mi_root) {
341 			mn->mn_fileno = mn->mn_parent->mn_fileno;
342 			break;
343 		}
344 		KASSERT(mn->mn_parent->mn_parent != NULL,
345 		    ("mqfstype_parent node has no grandparent"));
346 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
347 		break;
348 	default:
349 		KASSERT(0,
350 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
351 			mn->mn_type));
352 		break;
353 	}
354 }
355 
356 /*
357  * Release a file number
358  */
359 static void
360 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
361 {
362 	switch (mn->mn_type) {
363 	case mqfstype_root:
364 	case mqfstype_dir:
365 	case mqfstype_file:
366 	case mqfstype_symlink:
367 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
368 		break;
369 	case mqfstype_this:
370 	case mqfstype_parent:
371 		/* ignore these, as they don't "own" their file number */
372 		break;
373 	default:
374 		KASSERT(0,
375 		    ("mqfs_fileno_free() called for unknown type node: %d",
376 			mn->mn_type));
377 		break;
378 	}
379 }
380 
381 static __inline struct mqfs_node *
382 mqnode_alloc(void)
383 {
384 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
385 }
386 
387 static __inline void
388 mqnode_free(struct mqfs_node *node)
389 {
390 	uma_zfree(mqnode_zone, node);
391 }
392 
393 static __inline void
394 mqnode_addref(struct mqfs_node *node)
395 {
396 	atomic_add_int(&node->mn_refcount, 1);
397 }
398 
399 static __inline void
400 mqnode_release(struct mqfs_node *node)
401 {
402 	struct mqfs_info *mqfs;
403 	int old, exp;
404 
405 	mqfs = node->mn_info;
406 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
407 	if (node->mn_type == mqfstype_dir ||
408 	    node->mn_type == mqfstype_root)
409 		exp = 3; /* include . and .. */
410 	else
411 		exp = 1;
412 	if (old == exp) {
413 		int locked = sx_xlocked(&mqfs->mi_lock);
414 		if (!locked)
415 			sx_xlock(&mqfs->mi_lock);
416 		mqfs_destroy(node);
417 		if (!locked)
418 			sx_xunlock(&mqfs->mi_lock);
419 	}
420 }
421 
422 /*
423  * Add a node to a directory
424  */
425 static int
426 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
427 {
428 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
429 	KASSERT(parent->mn_info != NULL,
430 	    ("%s(): parent has no mn_info", __func__));
431 	KASSERT(parent->mn_type == mqfstype_dir ||
432 	    parent->mn_type == mqfstype_root,
433 	    ("%s(): parent is not a directory", __func__));
434 
435 	node->mn_info = parent->mn_info;
436 	node->mn_parent = parent;
437 	LIST_INIT(&node->mn_children);
438 	LIST_INIT(&node->mn_vnodes);
439 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
440 	mqnode_addref(parent);
441 	return (0);
442 }
443 
444 static struct mqfs_node *
445 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
446 	int nodetype)
447 {
448 	struct mqfs_node *node;
449 
450 	node = mqnode_alloc();
451 	strncpy(node->mn_name, name, namelen);
452 	node->mn_pr_root = cred->cr_prison->pr_root;
453 	node->mn_type = nodetype;
454 	node->mn_refcount = 1;
455 	vfs_timestamp(&node->mn_birth);
456 	node->mn_ctime = node->mn_atime = node->mn_mtime
457 		= node->mn_birth;
458 	node->mn_uid = cred->cr_uid;
459 	node->mn_gid = cred->cr_gid;
460 	node->mn_mode = mode;
461 	return (node);
462 }
463 
464 /*
465  * Create a file
466  */
467 static struct mqfs_node *
468 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
469 	struct ucred *cred, int mode)
470 {
471 	struct mqfs_node *node;
472 
473 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
474 	if (mqfs_add_node(parent, node) != 0) {
475 		mqnode_free(node);
476 		return (NULL);
477 	}
478 	return (node);
479 }
480 
481 /*
482  * Add . and .. to a directory
483  */
484 static int
485 mqfs_fixup_dir(struct mqfs_node *parent)
486 {
487 	struct mqfs_node *dir;
488 
489 	dir = mqnode_alloc();
490 	dir->mn_name[0] = '.';
491 	dir->mn_type = mqfstype_this;
492 	dir->mn_refcount = 1;
493 	if (mqfs_add_node(parent, dir) != 0) {
494 		mqnode_free(dir);
495 		return (-1);
496 	}
497 
498 	dir = mqnode_alloc();
499 	dir->mn_name[0] = dir->mn_name[1] = '.';
500 	dir->mn_type = mqfstype_parent;
501 	dir->mn_refcount = 1;
502 
503 	if (mqfs_add_node(parent, dir) != 0) {
504 		mqnode_free(dir);
505 		return (-1);
506 	}
507 
508 	return (0);
509 }
510 
511 #ifdef notyet
512 
513 /*
514  * Create a directory
515  */
516 static struct mqfs_node *
517 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
518 	struct ucred *cred, int mode)
519 {
520 	struct mqfs_node *node;
521 
522 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
523 	if (mqfs_add_node(parent, node) != 0) {
524 		mqnode_free(node);
525 		return (NULL);
526 	}
527 
528 	if (mqfs_fixup_dir(node) != 0) {
529 		mqfs_destroy(node);
530 		return (NULL);
531 	}
532 	return (node);
533 }
534 
535 /*
536  * Create a symlink
537  */
538 static struct mqfs_node *
539 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
540 	struct ucred *cred, int mode)
541 {
542 	struct mqfs_node *node;
543 
544 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
545 	if (mqfs_add_node(parent, node) != 0) {
546 		mqnode_free(node);
547 		return (NULL);
548 	}
549 	return (node);
550 }
551 
552 #endif
553 
554 /*
555  * Destroy a node or a tree of nodes
556  */
557 static int
558 mqfs_destroy(struct mqfs_node *node)
559 {
560 	struct mqfs_node *parent;
561 
562 	KASSERT(node != NULL,
563 	    ("%s(): node is NULL", __func__));
564 	KASSERT(node->mn_info != NULL,
565 	    ("%s(): node has no mn_info", __func__));
566 
567 	/* destroy children */
568 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
569 		while (! LIST_EMPTY(&node->mn_children))
570 			mqfs_destroy(LIST_FIRST(&node->mn_children));
571 
572 	/* unlink from parent */
573 	if ((parent = node->mn_parent) != NULL) {
574 		KASSERT(parent->mn_info == node->mn_info,
575 		    ("%s(): parent has different mn_info", __func__));
576 		LIST_REMOVE(node, mn_sibling);
577 	}
578 
579 	if (node->mn_fileno != 0)
580 		mqfs_fileno_free(node->mn_info, node);
581 	if (node->mn_data != NULL)
582 		mqueue_free(node->mn_data);
583 	mqnode_free(node);
584 	return (0);
585 }
586 
587 /*
588  * Mount a mqfs instance
589  */
590 static int
591 mqfs_mount(struct mount *mp)
592 {
593 	struct statfs *sbp;
594 
595 	if (mp->mnt_flag & MNT_UPDATE)
596 		return (EOPNOTSUPP);
597 
598 	mp->mnt_data = &mqfs_data;
599 	MNT_ILOCK(mp);
600 	mp->mnt_flag |= MNT_LOCAL;
601 	MNT_IUNLOCK(mp);
602 	vfs_getnewfsid(mp);
603 
604 	sbp = &mp->mnt_stat;
605 	vfs_mountedfrom(mp, "mqueue");
606 	sbp->f_bsize = PAGE_SIZE;
607 	sbp->f_iosize = PAGE_SIZE;
608 	sbp->f_blocks = 1;
609 	sbp->f_bfree = 0;
610 	sbp->f_bavail = 0;
611 	sbp->f_files = 1;
612 	sbp->f_ffree = 0;
613 	return (0);
614 }
615 
616 /*
617  * Unmount a mqfs instance
618  */
619 static int
620 mqfs_unmount(struct mount *mp, int mntflags)
621 {
622 	int error;
623 
624 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0,
625 	    curthread);
626 	return (error);
627 }
628 
629 /*
630  * Return a root vnode
631  */
632 static int
633 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
634 {
635 	struct mqfs_info *mqfs;
636 	int ret;
637 
638 	mqfs = VFSTOMQFS(mp);
639 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
640 	return (ret);
641 }
642 
643 /*
644  * Return filesystem stats
645  */
646 static int
647 mqfs_statfs(struct mount *mp, struct statfs *sbp)
648 {
649 	/* XXX update statistics */
650 	return (0);
651 }
652 
653 /*
654  * Initialize a mqfs instance
655  */
656 static int
657 mqfs_init(struct vfsconf *vfc)
658 {
659 	struct mqfs_node *root;
660 	struct mqfs_info *mi;
661 	osd_method_t methods[PR_MAXMETHOD] = {
662 	    [PR_METHOD_REMOVE] = mqfs_prison_remove,
663 	};
664 
665 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
666 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
667 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
668 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
669 	mvdata_zone = uma_zcreate("mvdata",
670 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
671 		NULL, UMA_ALIGN_PTR, 0);
672 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
673 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
674 	mi = &mqfs_data;
675 	sx_init(&mi->mi_lock, "mqfs lock");
676 	/* set up the root diretory */
677 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
678 		mqfstype_root);
679 	root->mn_info = mi;
680 	LIST_INIT(&root->mn_children);
681 	LIST_INIT(&root->mn_vnodes);
682 	mi->mi_root = root;
683 	mqfs_fileno_init(mi);
684 	mqfs_fileno_alloc(mi, root);
685 	mqfs_fixup_dir(root);
686 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
687 	    EVENTHANDLER_PRI_ANY);
688 	mq_fdclose = mqueue_fdclose;
689 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
690 	mqfs_osd_jail_slot = osd_jail_register(NULL, methods);
691 	return (0);
692 }
693 
694 /*
695  * Destroy a mqfs instance
696  */
697 static int
698 mqfs_uninit(struct vfsconf *vfc)
699 {
700 	struct mqfs_info *mi;
701 
702 	if (!unloadable)
703 		return (EOPNOTSUPP);
704 	osd_jail_deregister(mqfs_osd_jail_slot);
705 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
706 	mi = &mqfs_data;
707 	mqfs_destroy(mi->mi_root);
708 	mi->mi_root = NULL;
709 	mqfs_fileno_uninit(mi);
710 	sx_destroy(&mi->mi_lock);
711 	uma_zdestroy(mqnode_zone);
712 	uma_zdestroy(mqueue_zone);
713 	uma_zdestroy(mvdata_zone);
714 	uma_zdestroy(mqnoti_zone);
715 	return (0);
716 }
717 
718 /*
719  * task routine
720  */
721 static void
722 do_recycle(void *context, int pending __unused)
723 {
724 	struct vnode *vp = (struct vnode *)context;
725 
726 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
727 	vrecycle(vp);
728 	VOP_UNLOCK(vp);
729 	vdrop(vp);
730 }
731 
732 /*
733  * Allocate a vnode
734  */
735 static int
736 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
737 {
738 	struct mqfs_vdata *vd;
739 	struct mqfs_info  *mqfs;
740 	struct vnode *newvpp;
741 	int error;
742 
743 	mqfs = pn->mn_info;
744 	*vpp = NULL;
745 	sx_xlock(&mqfs->mi_lock);
746 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
747 		if (vd->mv_vnode->v_mount == mp) {
748 			vhold(vd->mv_vnode);
749 			break;
750 		}
751 	}
752 
753 	if (vd != NULL) {
754 found:
755 		*vpp = vd->mv_vnode;
756 		sx_xunlock(&mqfs->mi_lock);
757 		error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE);
758 		vdrop(*vpp);
759 		return (error);
760 	}
761 	sx_xunlock(&mqfs->mi_lock);
762 
763 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
764 	if (error)
765 		return (error);
766 	vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
767 	error = insmntque(newvpp, mp);
768 	if (error != 0)
769 		return (error);
770 
771 	sx_xlock(&mqfs->mi_lock);
772 	/*
773 	 * Check if it has already been allocated
774 	 * while we were blocked.
775 	 */
776 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
777 		if (vd->mv_vnode->v_mount == mp) {
778 			vhold(vd->mv_vnode);
779 			sx_xunlock(&mqfs->mi_lock);
780 
781 			vgone(newvpp);
782 			vput(newvpp);
783 			goto found;
784 		}
785 	}
786 
787 	*vpp = newvpp;
788 
789 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
790 	(*vpp)->v_data = vd;
791 	vd->mv_vnode = *vpp;
792 	vd->mv_node = pn;
793 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
794 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
795 	mqnode_addref(pn);
796 	switch (pn->mn_type) {
797 	case mqfstype_root:
798 		(*vpp)->v_vflag = VV_ROOT;
799 		/* fall through */
800 	case mqfstype_dir:
801 	case mqfstype_this:
802 	case mqfstype_parent:
803 		(*vpp)->v_type = VDIR;
804 		break;
805 	case mqfstype_file:
806 		(*vpp)->v_type = VREG;
807 		break;
808 	case mqfstype_symlink:
809 		(*vpp)->v_type = VLNK;
810 		break;
811 	case mqfstype_none:
812 		KASSERT(0, ("mqfs_allocf called for null node\n"));
813 	default:
814 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
815 	}
816 	sx_xunlock(&mqfs->mi_lock);
817 	return (0);
818 }
819 
820 /*
821  * Search a directory entry
822  */
823 static struct mqfs_node *
824 mqfs_search(struct mqfs_node *pd, const char *name, int len, struct ucred *cred)
825 {
826 	struct mqfs_node *pn;
827 	const void *pr_root;
828 
829 	sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
830 	pr_root = cred->cr_prison->pr_root;
831 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
832 		/* Only match names within the same prison root directory */
833 		if ((pn->mn_pr_root == NULL || pn->mn_pr_root == pr_root) &&
834 		    strncmp(pn->mn_name, name, len) == 0 &&
835 		    pn->mn_name[len] == '\0')
836 			return (pn);
837 	}
838 	return (NULL);
839 }
840 
841 /*
842  * Look up a file or directory.
843  */
844 static int
845 mqfs_lookupx(struct vop_cachedlookup_args *ap)
846 {
847 	struct componentname *cnp;
848 	struct vnode *dvp, **vpp;
849 	struct mqfs_node *pd;
850 	struct mqfs_node *pn;
851 	struct mqfs_info *mqfs;
852 	int nameiop, flags, error, namelen;
853 	char *pname;
854 	struct thread *td;
855 
856 	td = curthread;
857 	cnp = ap->a_cnp;
858 	vpp = ap->a_vpp;
859 	dvp = ap->a_dvp;
860 	pname = cnp->cn_nameptr;
861 	namelen = cnp->cn_namelen;
862 	flags = cnp->cn_flags;
863 	nameiop = cnp->cn_nameiop;
864 	pd = VTON(dvp);
865 	pn = NULL;
866 	mqfs = pd->mn_info;
867 	*vpp = NULLVP;
868 
869 	if (dvp->v_type != VDIR)
870 		return (ENOTDIR);
871 
872 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
873 	if (error)
874 		return (error);
875 
876 	/* shortcut: check if the name is too long */
877 	if (cnp->cn_namelen >= MQFS_NAMELEN)
878 		return (ENOENT);
879 
880 	/* self */
881 	if (namelen == 1 && pname[0] == '.') {
882 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
883 			return (EINVAL);
884 		pn = pd;
885 		*vpp = dvp;
886 		VREF(dvp);
887 		return (0);
888 	}
889 
890 	/* parent */
891 	if (cnp->cn_flags & ISDOTDOT) {
892 		if (dvp->v_vflag & VV_ROOT)
893 			return (EIO);
894 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
895 			return (EINVAL);
896 		VOP_UNLOCK(dvp);
897 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
898 		pn = pd->mn_parent;
899 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
900 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
901 		return (error);
902 	}
903 
904 	/* named node */
905 	sx_xlock(&mqfs->mi_lock);
906 	pn = mqfs_search(pd, pname, namelen, cnp->cn_cred);
907 	if (pn != NULL)
908 		mqnode_addref(pn);
909 	sx_xunlock(&mqfs->mi_lock);
910 
911 	/* found */
912 	if (pn != NULL) {
913 		/* DELETE */
914 		if (nameiop == DELETE && (flags & ISLASTCN)) {
915 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
916 			if (error) {
917 				mqnode_release(pn);
918 				return (error);
919 			}
920 			if (*vpp == dvp) {
921 				VREF(dvp);
922 				*vpp = dvp;
923 				mqnode_release(pn);
924 				return (0);
925 			}
926 		}
927 
928 		/* allocate vnode */
929 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
930 		mqnode_release(pn);
931 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
932 			cache_enter(dvp, *vpp, cnp);
933 		return (error);
934 	}
935 
936 	/* not found */
937 
938 	/* will create a new entry in the directory ? */
939 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
940 	    && (flags & ISLASTCN)) {
941 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
942 		if (error)
943 			return (error);
944 		return (EJUSTRETURN);
945 	}
946 	return (ENOENT);
947 }
948 
949 #if 0
950 struct vop_lookup_args {
951 	struct vop_generic_args a_gen;
952 	struct vnode *a_dvp;
953 	struct vnode **a_vpp;
954 	struct componentname *a_cnp;
955 };
956 #endif
957 
958 /*
959  * vnode lookup operation
960  */
961 static int
962 mqfs_lookup(struct vop_cachedlookup_args *ap)
963 {
964 	int rc;
965 
966 	rc = mqfs_lookupx(ap);
967 	return (rc);
968 }
969 
970 #if 0
971 struct vop_create_args {
972 	struct vnode *a_dvp;
973 	struct vnode **a_vpp;
974 	struct componentname *a_cnp;
975 	struct vattr *a_vap;
976 };
977 #endif
978 
979 /*
980  * vnode creation operation
981  */
982 static int
983 mqfs_create(struct vop_create_args *ap)
984 {
985 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
986 	struct componentname *cnp = ap->a_cnp;
987 	struct mqfs_node *pd;
988 	struct mqfs_node *pn;
989 	struct mqueue *mq;
990 	int error;
991 
992 	pd = VTON(ap->a_dvp);
993 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
994 		return (ENOTDIR);
995 	mq = mqueue_alloc(NULL);
996 	if (mq == NULL)
997 		return (EAGAIN);
998 	sx_xlock(&mqfs->mi_lock);
999 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
1000 		cnp->cn_cred, ap->a_vap->va_mode);
1001 	if (pn == NULL) {
1002 		sx_xunlock(&mqfs->mi_lock);
1003 		error = ENOSPC;
1004 	} else {
1005 		mqnode_addref(pn);
1006 		sx_xunlock(&mqfs->mi_lock);
1007 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1008 		mqnode_release(pn);
1009 		if (error)
1010 			mqfs_destroy(pn);
1011 		else
1012 			pn->mn_data = mq;
1013 	}
1014 	if (error)
1015 		mqueue_free(mq);
1016 	return (error);
1017 }
1018 
1019 /*
1020  * Remove an entry
1021  */
1022 static
1023 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
1024 {
1025 	struct mqfs_node *parent;
1026 	struct mqfs_vdata *vd;
1027 	int error = 0;
1028 
1029 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
1030 
1031 	if (ucred->cr_uid != pn->mn_uid &&
1032 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN)) != 0)
1033 		error = EACCES;
1034 	else if (!pn->mn_deleted) {
1035 		parent = pn->mn_parent;
1036 		pn->mn_parent = NULL;
1037 		pn->mn_deleted = 1;
1038 		LIST_REMOVE(pn, mn_sibling);
1039 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1040 			cache_purge(vd->mv_vnode);
1041 			vhold(vd->mv_vnode);
1042 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1043 		}
1044 		mqnode_release(pn);
1045 		mqnode_release(parent);
1046 	} else
1047 		error = ENOENT;
1048 	return (error);
1049 }
1050 
1051 #if 0
1052 struct vop_remove_args {
1053 	struct vnode *a_dvp;
1054 	struct vnode *a_vp;
1055 	struct componentname *a_cnp;
1056 };
1057 #endif
1058 
1059 /*
1060  * vnode removal operation
1061  */
1062 static int
1063 mqfs_remove(struct vop_remove_args *ap)
1064 {
1065 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1066 	struct mqfs_node *pn;
1067 	int error;
1068 
1069 	if (ap->a_vp->v_type == VDIR)
1070                 return (EPERM);
1071 	pn = VTON(ap->a_vp);
1072 	sx_xlock(&mqfs->mi_lock);
1073 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1074 	sx_xunlock(&mqfs->mi_lock);
1075 	return (error);
1076 }
1077 
1078 #if 0
1079 struct vop_inactive_args {
1080 	struct vnode *a_vp;
1081 	struct thread *a_td;
1082 };
1083 #endif
1084 
1085 static int
1086 mqfs_inactive(struct vop_inactive_args *ap)
1087 {
1088 	struct mqfs_node *pn = VTON(ap->a_vp);
1089 
1090 	if (pn->mn_deleted)
1091 		vrecycle(ap->a_vp);
1092 	return (0);
1093 }
1094 
1095 #if 0
1096 struct vop_reclaim_args {
1097 	struct vop_generic_args a_gen;
1098 	struct vnode *a_vp;
1099 };
1100 #endif
1101 
1102 static int
1103 mqfs_reclaim(struct vop_reclaim_args *ap)
1104 {
1105 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1106 	struct vnode *vp = ap->a_vp;
1107 	struct mqfs_node *pn;
1108 	struct mqfs_vdata *vd;
1109 
1110 	vd = vp->v_data;
1111 	pn = vd->mv_node;
1112 	sx_xlock(&mqfs->mi_lock);
1113 	vp->v_data = NULL;
1114 	LIST_REMOVE(vd, mv_link);
1115 	uma_zfree(mvdata_zone, vd);
1116 	mqnode_release(pn);
1117 	sx_xunlock(&mqfs->mi_lock);
1118 	return (0);
1119 }
1120 
1121 #if 0
1122 struct vop_open_args {
1123 	struct vop_generic_args a_gen;
1124 	struct vnode *a_vp;
1125 	int a_mode;
1126 	struct ucred *a_cred;
1127 	struct thread *a_td;
1128 	struct file *a_fp;
1129 };
1130 #endif
1131 
1132 static int
1133 mqfs_open(struct vop_open_args *ap)
1134 {
1135 	return (0);
1136 }
1137 
1138 #if 0
1139 struct vop_close_args {
1140 	struct vop_generic_args a_gen;
1141 	struct vnode *a_vp;
1142 	int a_fflag;
1143 	struct ucred *a_cred;
1144 	struct thread *a_td;
1145 };
1146 #endif
1147 
1148 static int
1149 mqfs_close(struct vop_close_args *ap)
1150 {
1151 	return (0);
1152 }
1153 
1154 #if 0
1155 struct vop_access_args {
1156 	struct vop_generic_args a_gen;
1157 	struct vnode *a_vp;
1158 	accmode_t a_accmode;
1159 	struct ucred *a_cred;
1160 	struct thread *a_td;
1161 };
1162 #endif
1163 
1164 /*
1165  * Verify permissions
1166  */
1167 static int
1168 mqfs_access(struct vop_access_args *ap)
1169 {
1170 	struct vnode *vp = ap->a_vp;
1171 	struct vattr vattr;
1172 	int error;
1173 
1174 	error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1175 	if (error)
1176 		return (error);
1177 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, vattr.va_gid,
1178 	    ap->a_accmode, ap->a_cred);
1179 	return (error);
1180 }
1181 
1182 #if 0
1183 struct vop_getattr_args {
1184 	struct vop_generic_args a_gen;
1185 	struct vnode *a_vp;
1186 	struct vattr *a_vap;
1187 	struct ucred *a_cred;
1188 };
1189 #endif
1190 
1191 /*
1192  * Get file attributes
1193  */
1194 static int
1195 mqfs_getattr(struct vop_getattr_args *ap)
1196 {
1197 	struct vnode *vp = ap->a_vp;
1198 	struct mqfs_node *pn = VTON(vp);
1199 	struct vattr *vap = ap->a_vap;
1200 	int error = 0;
1201 
1202 	vap->va_type = vp->v_type;
1203 	vap->va_mode = pn->mn_mode;
1204 	vap->va_nlink = 1;
1205 	vap->va_uid = pn->mn_uid;
1206 	vap->va_gid = pn->mn_gid;
1207 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1208 	vap->va_fileid = pn->mn_fileno;
1209 	vap->va_size = 0;
1210 	vap->va_blocksize = PAGE_SIZE;
1211 	vap->va_bytes = vap->va_size = 0;
1212 	vap->va_atime = pn->mn_atime;
1213 	vap->va_mtime = pn->mn_mtime;
1214 	vap->va_ctime = pn->mn_ctime;
1215 	vap->va_birthtime = pn->mn_birth;
1216 	vap->va_gen = 0;
1217 	vap->va_flags = 0;
1218 	vap->va_rdev = NODEV;
1219 	vap->va_bytes = 0;
1220 	vap->va_filerev = 0;
1221 	return (error);
1222 }
1223 
1224 #if 0
1225 struct vop_setattr_args {
1226 	struct vop_generic_args a_gen;
1227 	struct vnode *a_vp;
1228 	struct vattr *a_vap;
1229 	struct ucred *a_cred;
1230 };
1231 #endif
1232 /*
1233  * Set attributes
1234  */
1235 static int
1236 mqfs_setattr(struct vop_setattr_args *ap)
1237 {
1238 	struct mqfs_node *pn;
1239 	struct vattr *vap;
1240 	struct vnode *vp;
1241 	struct thread *td;
1242 	int c, error;
1243 	uid_t uid;
1244 	gid_t gid;
1245 
1246 	td = curthread;
1247 	vap = ap->a_vap;
1248 	vp = ap->a_vp;
1249 	if ((vap->va_type != VNON) ||
1250 	    (vap->va_nlink != VNOVAL) ||
1251 	    (vap->va_fsid != VNOVAL) ||
1252 	    (vap->va_fileid != VNOVAL) ||
1253 	    (vap->va_blocksize != VNOVAL) ||
1254 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1255 	    (vap->va_rdev != VNOVAL) ||
1256 	    ((int)vap->va_bytes != VNOVAL) ||
1257 	    (vap->va_gen != VNOVAL)) {
1258 		return (EINVAL);
1259 	}
1260 
1261 	pn = VTON(vp);
1262 
1263 	error = c = 0;
1264 	if (vap->va_uid == (uid_t)VNOVAL)
1265 		uid = pn->mn_uid;
1266 	else
1267 		uid = vap->va_uid;
1268 	if (vap->va_gid == (gid_t)VNOVAL)
1269 		gid = pn->mn_gid;
1270 	else
1271 		gid = vap->va_gid;
1272 
1273 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1274 		/*
1275 		 * To modify the ownership of a file, must possess VADMIN
1276 		 * for that file.
1277 		 */
1278 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1279 			return (error);
1280 
1281 		/*
1282 		 * XXXRW: Why is there a privilege check here: shouldn't the
1283 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1284 		 * below definitely right?
1285 		 */
1286 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1287 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1288 		    (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1289 			return (error);
1290 		pn->mn_uid = uid;
1291 		pn->mn_gid = gid;
1292 		c = 1;
1293 	}
1294 
1295 	if (vap->va_mode != (mode_t)VNOVAL) {
1296 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1297 		    (error = priv_check(td, PRIV_MQ_ADMIN)))
1298 			return (error);
1299 		pn->mn_mode = vap->va_mode;
1300 		c = 1;
1301 	}
1302 
1303 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1304 		/* See the comment in ufs_vnops::ufs_setattr(). */
1305 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1306 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1307 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1308 			return (error);
1309 		if (vap->va_atime.tv_sec != VNOVAL) {
1310 			pn->mn_atime = vap->va_atime;
1311 		}
1312 		if (vap->va_mtime.tv_sec != VNOVAL) {
1313 			pn->mn_mtime = vap->va_mtime;
1314 		}
1315 		c = 1;
1316 	}
1317 	if (c) {
1318 		vfs_timestamp(&pn->mn_ctime);
1319 	}
1320 	return (0);
1321 }
1322 
1323 #if 0
1324 struct vop_read_args {
1325 	struct vop_generic_args a_gen;
1326 	struct vnode *a_vp;
1327 	struct uio *a_uio;
1328 	int a_ioflag;
1329 	struct ucred *a_cred;
1330 };
1331 #endif
1332 
1333 /*
1334  * Read from a file
1335  */
1336 static int
1337 mqfs_read(struct vop_read_args *ap)
1338 {
1339 	char buf[80];
1340 	struct vnode *vp = ap->a_vp;
1341 	struct uio *uio = ap->a_uio;
1342 	struct mqueue *mq;
1343 	int len, error;
1344 
1345 	if (vp->v_type != VREG)
1346 		return (EINVAL);
1347 
1348 	mq = VTOMQ(vp);
1349 	snprintf(buf, sizeof(buf),
1350 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1351 		mq->mq_totalbytes,
1352 		mq->mq_maxmsg,
1353 		mq->mq_curmsgs,
1354 		mq->mq_msgsize);
1355 	buf[sizeof(buf)-1] = '\0';
1356 	len = strlen(buf);
1357 	error = uiomove_frombuf(buf, len, uio);
1358 	return (error);
1359 }
1360 
1361 #if 0
1362 struct vop_readdir_args {
1363 	struct vop_generic_args a_gen;
1364 	struct vnode *a_vp;
1365 	struct uio *a_uio;
1366 	struct ucred *a_cred;
1367 	int *a_eofflag;
1368 	int *a_ncookies;
1369 	uint64_t **a_cookies;
1370 };
1371 #endif
1372 
1373 /*
1374  * Return directory entries.
1375  */
1376 static int
1377 mqfs_readdir(struct vop_readdir_args *ap)
1378 {
1379 	struct vnode *vp;
1380 	struct mqfs_info *mi;
1381 	struct mqfs_node *pd;
1382 	struct mqfs_node *pn;
1383 	struct dirent entry;
1384 	struct uio *uio;
1385 	const void *pr_root;
1386 	int *tmp_ncookies = NULL;
1387 	off_t offset;
1388 	int error, i;
1389 
1390 	vp = ap->a_vp;
1391 	mi = VFSTOMQFS(vp->v_mount);
1392 	pd = VTON(vp);
1393 	uio = ap->a_uio;
1394 
1395 	if (vp->v_type != VDIR)
1396 		return (ENOTDIR);
1397 
1398 	if (uio->uio_offset < 0)
1399 		return (EINVAL);
1400 
1401 	if (ap->a_ncookies != NULL) {
1402 		tmp_ncookies = ap->a_ncookies;
1403 		*ap->a_ncookies = 0;
1404 		ap->a_ncookies = NULL;
1405         }
1406 
1407 	error = 0;
1408 	offset = 0;
1409 
1410 	pr_root = ap->a_cred->cr_prison->pr_root;
1411 	sx_xlock(&mi->mi_lock);
1412 
1413 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1414 		entry.d_reclen = sizeof(entry);
1415 
1416 		/*
1417 		 * Only show names within the same prison root directory
1418 		 * (or not associated with a prison, e.g. "." and "..").
1419 		 */
1420 		if (pn->mn_pr_root != NULL && pn->mn_pr_root != pr_root)
1421 			continue;
1422 		if (!pn->mn_fileno)
1423 			mqfs_fileno_alloc(mi, pn);
1424 		entry.d_fileno = pn->mn_fileno;
1425 		entry.d_off = offset + entry.d_reclen;
1426 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1427 			entry.d_name[i] = pn->mn_name[i];
1428 		entry.d_namlen = i;
1429 		switch (pn->mn_type) {
1430 		case mqfstype_root:
1431 		case mqfstype_dir:
1432 		case mqfstype_this:
1433 		case mqfstype_parent:
1434 			entry.d_type = DT_DIR;
1435 			break;
1436 		case mqfstype_file:
1437 			entry.d_type = DT_REG;
1438 			break;
1439 		case mqfstype_symlink:
1440 			entry.d_type = DT_LNK;
1441 			break;
1442 		default:
1443 			panic("%s has unexpected node type: %d", pn->mn_name,
1444 				pn->mn_type);
1445 		}
1446 		dirent_terminate(&entry);
1447 		if (entry.d_reclen > uio->uio_resid)
1448                         break;
1449 		if (offset >= uio->uio_offset) {
1450 			error = vfs_read_dirent(ap, &entry, offset);
1451                         if (error)
1452                                 break;
1453                 }
1454                 offset += entry.d_reclen;
1455 	}
1456 	sx_xunlock(&mi->mi_lock);
1457 
1458 	uio->uio_offset = offset;
1459 
1460 	if (tmp_ncookies != NULL)
1461 		ap->a_ncookies = tmp_ncookies;
1462 
1463 	return (error);
1464 }
1465 
1466 #ifdef notyet
1467 
1468 #if 0
1469 struct vop_mkdir_args {
1470 	struct vnode *a_dvp;
1471 	struvt vnode **a_vpp;
1472 	struvt componentname *a_cnp;
1473 	struct vattr *a_vap;
1474 };
1475 #endif
1476 
1477 /*
1478  * Create a directory.
1479  */
1480 static int
1481 mqfs_mkdir(struct vop_mkdir_args *ap)
1482 {
1483 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1484 	struct componentname *cnp = ap->a_cnp;
1485 	struct mqfs_node *pd = VTON(ap->a_dvp);
1486 	struct mqfs_node *pn;
1487 	int error;
1488 
1489 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1490 		return (ENOTDIR);
1491 	sx_xlock(&mqfs->mi_lock);
1492 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1493 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1494 	if (pn != NULL)
1495 		mqnode_addref(pn);
1496 	sx_xunlock(&mqfs->mi_lock);
1497 	if (pn == NULL) {
1498 		error = ENOSPC;
1499 	} else {
1500 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1501 		mqnode_release(pn);
1502 	}
1503 	return (error);
1504 }
1505 
1506 #if 0
1507 struct vop_rmdir_args {
1508 	struct vnode *a_dvp;
1509 	struct vnode *a_vp;
1510 	struct componentname *a_cnp;
1511 };
1512 #endif
1513 
1514 /*
1515  * Remove a directory.
1516  */
1517 static int
1518 mqfs_rmdir(struct vop_rmdir_args *ap)
1519 {
1520 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1521 	struct mqfs_node *pn = VTON(ap->a_vp);
1522 	struct mqfs_node *pt;
1523 
1524 	if (pn->mn_type != mqfstype_dir)
1525 		return (ENOTDIR);
1526 
1527 	sx_xlock(&mqfs->mi_lock);
1528 	if (pn->mn_deleted) {
1529 		sx_xunlock(&mqfs->mi_lock);
1530 		return (ENOENT);
1531 	}
1532 
1533 	pt = LIST_FIRST(&pn->mn_children);
1534 	pt = LIST_NEXT(pt, mn_sibling);
1535 	pt = LIST_NEXT(pt, mn_sibling);
1536 	if (pt != NULL) {
1537 		sx_xunlock(&mqfs->mi_lock);
1538 		return (ENOTEMPTY);
1539 	}
1540 	pt = pn->mn_parent;
1541 	pn->mn_parent = NULL;
1542 	pn->mn_deleted = 1;
1543 	LIST_REMOVE(pn, mn_sibling);
1544 	mqnode_release(pn);
1545 	mqnode_release(pt);
1546 	sx_xunlock(&mqfs->mi_lock);
1547 	cache_purge(ap->a_vp);
1548 	return (0);
1549 }
1550 
1551 #endif /* notyet */
1552 
1553 /*
1554  * See if this prison root is obsolete, and clean up associated queues if it is.
1555  */
1556 static int
1557 mqfs_prison_remove(void *obj, void *data __unused)
1558 {
1559 	const struct prison *pr = obj;
1560 	struct prison *tpr;
1561 	struct mqfs_node *pn, *tpn;
1562 	struct vnode *pr_root;
1563 
1564 	pr_root = pr->pr_root;
1565 	if (pr->pr_parent->pr_root == pr_root)
1566 		return (0);
1567 	TAILQ_FOREACH(tpr, &allprison, pr_list) {
1568 		if (tpr != pr && tpr->pr_root == pr_root)
1569 			return (0);
1570 	}
1571 	/*
1572 	 * No jails are rooted in this directory anymore,
1573 	 * so no queues should be either.
1574 	 */
1575 	sx_xlock(&mqfs_data.mi_lock);
1576 	LIST_FOREACH_SAFE(pn, &mqfs_data.mi_root->mn_children,
1577 	    mn_sibling, tpn) {
1578 		if (pn->mn_pr_root == pr_root)
1579 			(void)do_unlink(pn, curthread->td_ucred);
1580 	}
1581 	sx_xunlock(&mqfs_data.mi_lock);
1582 	return (0);
1583 }
1584 
1585 /*
1586  * Allocate a message queue
1587  */
1588 static struct mqueue *
1589 mqueue_alloc(const struct mq_attr *attr)
1590 {
1591 	struct mqueue *mq;
1592 
1593 	if (curmq >= maxmq)
1594 		return (NULL);
1595 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1596 	TAILQ_INIT(&mq->mq_msgq);
1597 	if (attr != NULL) {
1598 		mq->mq_maxmsg = attr->mq_maxmsg;
1599 		mq->mq_msgsize = attr->mq_msgsize;
1600 	} else {
1601 		mq->mq_maxmsg = default_maxmsg;
1602 		mq->mq_msgsize = default_msgsize;
1603 	}
1604 	mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1605 	knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1606 	knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1607 	atomic_add_int(&curmq, 1);
1608 	return (mq);
1609 }
1610 
1611 /*
1612  * Destroy a message queue
1613  */
1614 static void
1615 mqueue_free(struct mqueue *mq)
1616 {
1617 	struct mqueue_msg *msg;
1618 
1619 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1620 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1621 		free(msg, M_MQUEUEDATA);
1622 	}
1623 
1624 	mtx_destroy(&mq->mq_mutex);
1625 	seldrain(&mq->mq_rsel);
1626 	seldrain(&mq->mq_wsel);
1627 	knlist_destroy(&mq->mq_rsel.si_note);
1628 	knlist_destroy(&mq->mq_wsel.si_note);
1629 	uma_zfree(mqueue_zone, mq);
1630 	atomic_add_int(&curmq, -1);
1631 }
1632 
1633 /*
1634  * Load a message from user space
1635  */
1636 static struct mqueue_msg *
1637 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1638 {
1639 	struct mqueue_msg *msg;
1640 	size_t len;
1641 	int error;
1642 
1643 	len = sizeof(struct mqueue_msg) + msg_size;
1644 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1645 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1646 	    msg_size);
1647 	if (error) {
1648 		free(msg, M_MQUEUEDATA);
1649 		msg = NULL;
1650 	} else {
1651 		msg->msg_size = msg_size;
1652 		msg->msg_prio = msg_prio;
1653 	}
1654 	return (msg);
1655 }
1656 
1657 /*
1658  * Save a message to user space
1659  */
1660 static int
1661 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1662 {
1663 	int error;
1664 
1665 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1666 		msg->msg_size);
1667 	if (error == 0 && msg_prio != NULL)
1668 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1669 	return (error);
1670 }
1671 
1672 /*
1673  * Free a message's memory
1674  */
1675 static __inline void
1676 mqueue_freemsg(struct mqueue_msg *msg)
1677 {
1678 	free(msg, M_MQUEUEDATA);
1679 }
1680 
1681 /*
1682  * Send a message. if waitok is false, thread will not be
1683  * blocked if there is no data in queue, otherwise, absolute
1684  * time will be checked.
1685  */
1686 int
1687 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1688 	size_t msg_len, unsigned msg_prio, int waitok,
1689 	const struct timespec *abs_timeout)
1690 {
1691 	struct mqueue_msg *msg;
1692 	struct timespec ts, ts2;
1693 	struct timeval tv;
1694 	int error;
1695 
1696 	if (msg_prio >= MQ_PRIO_MAX)
1697 		return (EINVAL);
1698 	if (msg_len > mq->mq_msgsize)
1699 		return (EMSGSIZE);
1700 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1701 	if (msg == NULL)
1702 		return (EFAULT);
1703 
1704 	/* O_NONBLOCK case */
1705 	if (!waitok) {
1706 		error = _mqueue_send(mq, msg, -1);
1707 		if (error)
1708 			goto bad;
1709 		return (0);
1710 	}
1711 
1712 	/* we allow a null timeout (wait forever) */
1713 	if (abs_timeout == NULL) {
1714 		error = _mqueue_send(mq, msg, 0);
1715 		if (error)
1716 			goto bad;
1717 		return (0);
1718 	}
1719 
1720 	/* send it before checking time */
1721 	error = _mqueue_send(mq, msg, -1);
1722 	if (error == 0)
1723 		return (0);
1724 
1725 	if (error != EAGAIN)
1726 		goto bad;
1727 
1728 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1729 		error = EINVAL;
1730 		goto bad;
1731 	}
1732 	for (;;) {
1733 		getnanotime(&ts);
1734 		timespecsub(abs_timeout, &ts, &ts2);
1735 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1736 			error = ETIMEDOUT;
1737 			break;
1738 		}
1739 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1740 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1741 		if (error != ETIMEDOUT)
1742 			break;
1743 	}
1744 	if (error == 0)
1745 		return (0);
1746 bad:
1747 	mqueue_freemsg(msg);
1748 	return (error);
1749 }
1750 
1751 /*
1752  * Common routine to send a message
1753  */
1754 static int
1755 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1756 {
1757 	struct mqueue_msg *msg2;
1758 	int error = 0;
1759 
1760 	mtx_lock(&mq->mq_mutex);
1761 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1762 		if (timo < 0) {
1763 			mtx_unlock(&mq->mq_mutex);
1764 			return (EAGAIN);
1765 		}
1766 		mq->mq_senders++;
1767 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1768 			    PCATCH, "mqsend", timo);
1769 		mq->mq_senders--;
1770 		if (error == EAGAIN)
1771 			error = ETIMEDOUT;
1772 	}
1773 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1774 		mtx_unlock(&mq->mq_mutex);
1775 		return (error);
1776 	}
1777 	error = 0;
1778 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1779 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1780 	} else {
1781 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1782 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1783 		} else {
1784 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1785 				if (msg2->msg_prio < msg->msg_prio)
1786 					break;
1787 			}
1788 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1789 		}
1790 	}
1791 	mq->mq_curmsgs++;
1792 	mq->mq_totalbytes += msg->msg_size;
1793 	if (mq->mq_receivers)
1794 		wakeup_one(&mq->mq_receivers);
1795 	else if (mq->mq_notifier != NULL)
1796 		mqueue_send_notification(mq);
1797 	if (mq->mq_flags & MQ_RSEL) {
1798 		mq->mq_flags &= ~MQ_RSEL;
1799 		selwakeup(&mq->mq_rsel);
1800 	}
1801 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1802 	mtx_unlock(&mq->mq_mutex);
1803 	return (0);
1804 }
1805 
1806 /*
1807  * Send realtime a signal to process which registered itself
1808  * successfully by mq_notify.
1809  */
1810 static void
1811 mqueue_send_notification(struct mqueue *mq)
1812 {
1813 	struct mqueue_notifier *nt;
1814 	struct thread *td;
1815 	struct proc *p;
1816 	int error;
1817 
1818 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1819 	nt = mq->mq_notifier;
1820 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1821 		p = nt->nt_proc;
1822 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1823 		if (error) {
1824 			mq->mq_notifier = NULL;
1825 			return;
1826 		}
1827 		if (!KSI_ONQ(&nt->nt_ksi)) {
1828 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1829 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1830 		}
1831 		PROC_UNLOCK(p);
1832 	}
1833 	mq->mq_notifier = NULL;
1834 }
1835 
1836 /*
1837  * Get a message. if waitok is false, thread will not be
1838  * blocked if there is no data in queue, otherwise, absolute
1839  * time will be checked.
1840  */
1841 int
1842 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1843 	size_t msg_len, unsigned *msg_prio, int waitok,
1844 	const struct timespec *abs_timeout)
1845 {
1846 	struct mqueue_msg *msg;
1847 	struct timespec ts, ts2;
1848 	struct timeval tv;
1849 	int error;
1850 
1851 	if (msg_len < mq->mq_msgsize)
1852 		return (EMSGSIZE);
1853 
1854 	/* O_NONBLOCK case */
1855 	if (!waitok) {
1856 		error = _mqueue_recv(mq, &msg, -1);
1857 		if (error)
1858 			return (error);
1859 		goto received;
1860 	}
1861 
1862 	/* we allow a null timeout (wait forever). */
1863 	if (abs_timeout == NULL) {
1864 		error = _mqueue_recv(mq, &msg, 0);
1865 		if (error)
1866 			return (error);
1867 		goto received;
1868 	}
1869 
1870 	/* try to get a message before checking time */
1871 	error = _mqueue_recv(mq, &msg, -1);
1872 	if (error == 0)
1873 		goto received;
1874 
1875 	if (error != EAGAIN)
1876 		return (error);
1877 
1878 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1879 		error = EINVAL;
1880 		return (error);
1881 	}
1882 
1883 	for (;;) {
1884 		getnanotime(&ts);
1885 		timespecsub(abs_timeout, &ts, &ts2);
1886 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1887 			error = ETIMEDOUT;
1888 			return (error);
1889 		}
1890 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1891 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1892 		if (error == 0)
1893 			break;
1894 		if (error != ETIMEDOUT)
1895 			return (error);
1896 	}
1897 
1898 received:
1899 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1900 	if (error == 0) {
1901 		curthread->td_retval[0] = msg->msg_size;
1902 		curthread->td_retval[1] = 0;
1903 	}
1904 	mqueue_freemsg(msg);
1905 	return (error);
1906 }
1907 
1908 /*
1909  * Common routine to receive a message
1910  */
1911 static int
1912 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1913 {
1914 	int error = 0;
1915 
1916 	mtx_lock(&mq->mq_mutex);
1917 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1918 		if (timo < 0) {
1919 			mtx_unlock(&mq->mq_mutex);
1920 			return (EAGAIN);
1921 		}
1922 		mq->mq_receivers++;
1923 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1924 			    PCATCH, "mqrecv", timo);
1925 		mq->mq_receivers--;
1926 		if (error == EAGAIN)
1927 			error = ETIMEDOUT;
1928 	}
1929 	if (*msg != NULL) {
1930 		error = 0;
1931 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1932 		mq->mq_curmsgs--;
1933 		mq->mq_totalbytes -= (*msg)->msg_size;
1934 		if (mq->mq_senders)
1935 			wakeup_one(&mq->mq_senders);
1936 		if (mq->mq_flags & MQ_WSEL) {
1937 			mq->mq_flags &= ~MQ_WSEL;
1938 			selwakeup(&mq->mq_wsel);
1939 		}
1940 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1941 	}
1942 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1943 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1944 		mqueue_send_notification(mq);
1945 	}
1946 	mtx_unlock(&mq->mq_mutex);
1947 	return (error);
1948 }
1949 
1950 static __inline struct mqueue_notifier *
1951 notifier_alloc(void)
1952 {
1953 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1954 }
1955 
1956 static __inline void
1957 notifier_free(struct mqueue_notifier *p)
1958 {
1959 	uma_zfree(mqnoti_zone, p);
1960 }
1961 
1962 static struct mqueue_notifier *
1963 notifier_search(struct proc *p, int fd)
1964 {
1965 	struct mqueue_notifier *nt;
1966 
1967 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1968 		if (nt->nt_ksi.ksi_mqd == fd)
1969 			break;
1970 	}
1971 	return (nt);
1972 }
1973 
1974 static __inline void
1975 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1976 {
1977 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1978 }
1979 
1980 static __inline void
1981 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1982 {
1983 	LIST_REMOVE(nt, nt_link);
1984 	notifier_free(nt);
1985 }
1986 
1987 static void
1988 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1989 {
1990 	struct mqueue_notifier *nt;
1991 
1992 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1993 	PROC_LOCK(p);
1994 	nt = notifier_search(p, fd);
1995 	if (nt != NULL) {
1996 		if (mq->mq_notifier == nt)
1997 			mq->mq_notifier = NULL;
1998 		sigqueue_take(&nt->nt_ksi);
1999 		notifier_delete(p, nt);
2000 	}
2001 	PROC_UNLOCK(p);
2002 }
2003 
2004 static int
2005 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
2006     const struct mq_attr *attr)
2007 {
2008 	char path[MQFS_NAMELEN + 1];
2009 	struct mqfs_node *pn;
2010 	struct pwddesc *pdp;
2011 	struct file *fp;
2012 	struct mqueue *mq;
2013 	int fd, error, len, cmode;
2014 
2015 	AUDIT_ARG_FFLAGS(flags);
2016 	AUDIT_ARG_MODE(mode);
2017 
2018 	pdp = td->td_proc->p_pd;
2019 	cmode = (((mode & ~pdp->pd_cmask) & ALLPERMS) & ~S_ISTXT);
2020 	mq = NULL;
2021 	if ((flags & O_CREAT) != 0 && attr != NULL) {
2022 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
2023 			return (EINVAL);
2024 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
2025 			return (EINVAL);
2026 	}
2027 
2028 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
2029         if (error)
2030 		return (error);
2031 
2032 	/*
2033 	 * The first character of name must be a slash  (/) character
2034 	 * and the remaining characters of name cannot include any slash
2035 	 * characters.
2036 	 */
2037 	len = strlen(path);
2038 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2039 		return (EINVAL);
2040 	/*
2041 	 * "." and ".." are magic directories, populated on the fly, and cannot
2042 	 * be opened as queues.
2043 	 */
2044 	if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0)
2045 		return (EINVAL);
2046 	AUDIT_ARG_UPATH1_CANON(path);
2047 
2048 	error = falloc(td, &fp, &fd, O_CLOEXEC);
2049 	if (error)
2050 		return (error);
2051 
2052 	sx_xlock(&mqfs_data.mi_lock);
2053 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
2054 	if (pn == NULL) {
2055 		if (!(flags & O_CREAT)) {
2056 			error = ENOENT;
2057 		} else {
2058 			mq = mqueue_alloc(attr);
2059 			if (mq == NULL) {
2060 				error = ENFILE;
2061 			} else {
2062 				pn = mqfs_create_file(mqfs_data.mi_root,
2063 				         path + 1, len - 1, td->td_ucred,
2064 					 cmode);
2065 				if (pn == NULL) {
2066 					error = ENOSPC;
2067 					mqueue_free(mq);
2068 				}
2069 			}
2070 		}
2071 
2072 		if (error == 0) {
2073 			pn->mn_data = mq;
2074 		}
2075 	} else {
2076 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2077 			error = EEXIST;
2078 		} else {
2079 			accmode_t accmode = 0;
2080 
2081 			if (flags & FREAD)
2082 				accmode |= VREAD;
2083 			if (flags & FWRITE)
2084 				accmode |= VWRITE;
2085 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2086 			    pn->mn_gid, accmode, td->td_ucred);
2087 		}
2088 	}
2089 
2090 	if (error) {
2091 		sx_xunlock(&mqfs_data.mi_lock);
2092 		fdclose(td, fp, fd);
2093 		fdrop(fp, td);
2094 		return (error);
2095 	}
2096 
2097 	mqnode_addref(pn);
2098 	sx_xunlock(&mqfs_data.mi_lock);
2099 
2100 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2101 	    &mqueueops);
2102 
2103 	td->td_retval[0] = fd;
2104 	fdrop(fp, td);
2105 	return (0);
2106 }
2107 
2108 /*
2109  * Syscall to open a message queue.
2110  */
2111 int
2112 sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
2113 {
2114 	struct mq_attr attr;
2115 	int flags, error;
2116 
2117 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2118 		return (EINVAL);
2119 	flags = FFLAGS(uap->flags);
2120 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2121 		error = copyin(uap->attr, &attr, sizeof(attr));
2122 		if (error)
2123 			return (error);
2124 	}
2125 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2126 	    uap->attr != NULL ? &attr : NULL));
2127 }
2128 
2129 /*
2130  * Syscall to unlink a message queue.
2131  */
2132 int
2133 sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2134 {
2135 	char path[MQFS_NAMELEN+1];
2136 	struct mqfs_node *pn;
2137 	int error, len;
2138 
2139 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2140         if (error)
2141 		return (error);
2142 
2143 	len = strlen(path);
2144 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2145 		return (EINVAL);
2146 	if (strcmp(path, "/.") == 0 || strcmp(path, "/..") == 0)
2147 		return (EINVAL);
2148 	AUDIT_ARG_UPATH1_CANON(path);
2149 
2150 	sx_xlock(&mqfs_data.mi_lock);
2151 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
2152 	if (pn != NULL)
2153 		error = do_unlink(pn, td->td_ucred);
2154 	else
2155 		error = ENOENT;
2156 	sx_xunlock(&mqfs_data.mi_lock);
2157 	return (error);
2158 }
2159 
2160 typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **);
2161 
2162 /*
2163  * Get message queue by giving file slot
2164  */
2165 static int
2166 _getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func,
2167        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2168 {
2169 	struct mqfs_node *pn;
2170 	int error;
2171 
2172 	error = func(td, fd, rightsp, fpp);
2173 	if (error)
2174 		return (error);
2175 	if (&mqueueops != (*fpp)->f_ops) {
2176 		fdrop(*fpp, td);
2177 		return (EBADF);
2178 	}
2179 	pn = (*fpp)->f_data;
2180 	if (ppn)
2181 		*ppn = pn;
2182 	if (pmq)
2183 		*pmq = pn->mn_data;
2184 	return (0);
2185 }
2186 
2187 static __inline int
2188 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2189 	struct mqueue **pmq)
2190 {
2191 
2192 	return _getmq(td, fd, &cap_event_rights, fget,
2193 	    fpp, ppn, pmq);
2194 }
2195 
2196 static __inline int
2197 getmq_read(struct thread *td, int fd, struct file **fpp,
2198 	 struct mqfs_node **ppn, struct mqueue **pmq)
2199 {
2200 
2201 	return _getmq(td, fd, &cap_read_rights, fget_read,
2202 	    fpp, ppn, pmq);
2203 }
2204 
2205 static __inline int
2206 getmq_write(struct thread *td, int fd, struct file **fpp,
2207 	struct mqfs_node **ppn, struct mqueue **pmq)
2208 {
2209 
2210 	return _getmq(td, fd, &cap_write_rights, fget_write,
2211 	    fpp, ppn, pmq);
2212 }
2213 
2214 static int
2215 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2216     struct mq_attr *oattr)
2217 {
2218 	struct mqueue *mq;
2219 	struct file *fp;
2220 	u_int oflag, flag;
2221 	int error;
2222 
2223 	AUDIT_ARG_FD(mqd);
2224 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2225 		return (EINVAL);
2226 	error = getmq(td, mqd, &fp, NULL, &mq);
2227 	if (error)
2228 		return (error);
2229 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2230 	oattr->mq_msgsize = mq->mq_msgsize;
2231 	oattr->mq_curmsgs = mq->mq_curmsgs;
2232 	if (attr != NULL) {
2233 		do {
2234 			oflag = flag = fp->f_flag;
2235 			flag &= ~O_NONBLOCK;
2236 			flag |= (attr->mq_flags & O_NONBLOCK);
2237 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2238 	} else
2239 		oflag = fp->f_flag;
2240 	oattr->mq_flags = (O_NONBLOCK & oflag);
2241 	fdrop(fp, td);
2242 	return (error);
2243 }
2244 
2245 int
2246 sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2247 {
2248 	struct mq_attr attr, oattr;
2249 	int error;
2250 
2251 	if (uap->attr != NULL) {
2252 		error = copyin(uap->attr, &attr, sizeof(attr));
2253 		if (error != 0)
2254 			return (error);
2255 	}
2256 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2257 	    &oattr);
2258 	if (error == 0 && uap->oattr != NULL) {
2259 		bzero(oattr.__reserved, sizeof(oattr.__reserved));
2260 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2261 	}
2262 	return (error);
2263 }
2264 
2265 int
2266 sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2267 {
2268 	struct mqueue *mq;
2269 	struct file *fp;
2270 	struct timespec *abs_timeout, ets;
2271 	int error;
2272 	int waitok;
2273 
2274 	AUDIT_ARG_FD(uap->mqd);
2275 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2276 	if (error)
2277 		return (error);
2278 	if (uap->abs_timeout != NULL) {
2279 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2280 		if (error != 0)
2281 			goto out;
2282 		abs_timeout = &ets;
2283 	} else
2284 		abs_timeout = NULL;
2285 	waitok = !(fp->f_flag & O_NONBLOCK);
2286 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2287 		uap->msg_prio, waitok, abs_timeout);
2288 out:
2289 	fdrop(fp, td);
2290 	return (error);
2291 }
2292 
2293 int
2294 sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2295 {
2296 	struct mqueue *mq;
2297 	struct file *fp;
2298 	struct timespec *abs_timeout, ets;
2299 	int error, waitok;
2300 
2301 	AUDIT_ARG_FD(uap->mqd);
2302 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2303 	if (error)
2304 		return (error);
2305 	if (uap->abs_timeout != NULL) {
2306 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2307 		if (error != 0)
2308 			goto out;
2309 		abs_timeout = &ets;
2310 	} else
2311 		abs_timeout = NULL;
2312 	waitok = !(fp->f_flag & O_NONBLOCK);
2313 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2314 		uap->msg_prio, waitok, abs_timeout);
2315 out:
2316 	fdrop(fp, td);
2317 	return (error);
2318 }
2319 
2320 static int
2321 kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev)
2322 {
2323 	struct filedesc *fdp;
2324 	struct proc *p;
2325 	struct mqueue *mq;
2326 	struct file *fp, *fp2;
2327 	struct mqueue_notifier *nt, *newnt = NULL;
2328 	int error;
2329 
2330 	AUDIT_ARG_FD(mqd);
2331 	if (sigev != NULL) {
2332 		if (sigev->sigev_notify != SIGEV_SIGNAL &&
2333 		    sigev->sigev_notify != SIGEV_THREAD_ID &&
2334 		    sigev->sigev_notify != SIGEV_NONE)
2335 			return (EINVAL);
2336 		if ((sigev->sigev_notify == SIGEV_SIGNAL ||
2337 		    sigev->sigev_notify == SIGEV_THREAD_ID) &&
2338 		    !_SIG_VALID(sigev->sigev_signo))
2339 			return (EINVAL);
2340 	}
2341 	p = td->td_proc;
2342 	fdp = td->td_proc->p_fd;
2343 	error = getmq(td, mqd, &fp, NULL, &mq);
2344 	if (error)
2345 		return (error);
2346 again:
2347 	FILEDESC_SLOCK(fdp);
2348 	fp2 = fget_noref(fdp, mqd);
2349 	if (fp2 == NULL) {
2350 		FILEDESC_SUNLOCK(fdp);
2351 		error = EBADF;
2352 		goto out;
2353 	}
2354 #ifdef CAPABILITIES
2355 	error = cap_check(cap_rights(fdp, mqd), &cap_event_rights);
2356 	if (error) {
2357 		FILEDESC_SUNLOCK(fdp);
2358 		goto out;
2359 	}
2360 #endif
2361 	if (fp2 != fp) {
2362 		FILEDESC_SUNLOCK(fdp);
2363 		error = EBADF;
2364 		goto out;
2365 	}
2366 	mtx_lock(&mq->mq_mutex);
2367 	FILEDESC_SUNLOCK(fdp);
2368 	if (sigev != NULL) {
2369 		if (mq->mq_notifier != NULL) {
2370 			error = EBUSY;
2371 		} else {
2372 			PROC_LOCK(p);
2373 			nt = notifier_search(p, mqd);
2374 			if (nt == NULL) {
2375 				if (newnt == NULL) {
2376 					PROC_UNLOCK(p);
2377 					mtx_unlock(&mq->mq_mutex);
2378 					newnt = notifier_alloc();
2379 					goto again;
2380 				}
2381 			}
2382 
2383 			if (nt != NULL) {
2384 				sigqueue_take(&nt->nt_ksi);
2385 				if (newnt != NULL) {
2386 					notifier_free(newnt);
2387 					newnt = NULL;
2388 				}
2389 			} else {
2390 				nt = newnt;
2391 				newnt = NULL;
2392 				ksiginfo_init(&nt->nt_ksi);
2393 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2394 				nt->nt_ksi.ksi_code = SI_MESGQ;
2395 				nt->nt_proc = p;
2396 				nt->nt_ksi.ksi_mqd = mqd;
2397 				notifier_insert(p, nt);
2398 			}
2399 			nt->nt_sigev = *sigev;
2400 			mq->mq_notifier = nt;
2401 			PROC_UNLOCK(p);
2402 			/*
2403 			 * if there is no receivers and message queue
2404 			 * is not empty, we should send notification
2405 			 * as soon as possible.
2406 			 */
2407 			if (mq->mq_receivers == 0 &&
2408 			    !TAILQ_EMPTY(&mq->mq_msgq))
2409 				mqueue_send_notification(mq);
2410 		}
2411 	} else {
2412 		notifier_remove(p, mq, mqd);
2413 	}
2414 	mtx_unlock(&mq->mq_mutex);
2415 
2416 out:
2417 	fdrop(fp, td);
2418 	if (newnt != NULL)
2419 		notifier_free(newnt);
2420 	return (error);
2421 }
2422 
2423 int
2424 sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2425 {
2426 	struct sigevent ev, *evp;
2427 	int error;
2428 
2429 	if (uap->sigev == NULL) {
2430 		evp = NULL;
2431 	} else {
2432 		error = copyin(uap->sigev, &ev, sizeof(ev));
2433 		if (error != 0)
2434 			return (error);
2435 		evp = &ev;
2436 	}
2437 	return (kern_kmq_notify(td, uap->mqd, evp));
2438 }
2439 
2440 static void
2441 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2442 {
2443 	struct mqueue *mq;
2444 #ifdef INVARIANTS
2445 	struct filedesc *fdp;
2446 
2447 	fdp = td->td_proc->p_fd;
2448 	FILEDESC_LOCK_ASSERT(fdp);
2449 #endif
2450 
2451 	if (fp->f_ops == &mqueueops) {
2452 		mq = FPTOMQ(fp);
2453 		mtx_lock(&mq->mq_mutex);
2454 		notifier_remove(td->td_proc, mq, fd);
2455 
2456 		/* have to wakeup thread in same process */
2457 		if (mq->mq_flags & MQ_RSEL) {
2458 			mq->mq_flags &= ~MQ_RSEL;
2459 			selwakeup(&mq->mq_rsel);
2460 		}
2461 		if (mq->mq_flags & MQ_WSEL) {
2462 			mq->mq_flags &= ~MQ_WSEL;
2463 			selwakeup(&mq->mq_wsel);
2464 		}
2465 		mtx_unlock(&mq->mq_mutex);
2466 	}
2467 }
2468 
2469 static void
2470 mq_proc_exit(void *arg __unused, struct proc *p)
2471 {
2472 	struct filedesc *fdp;
2473 	struct file *fp;
2474 	struct mqueue *mq;
2475 	int i;
2476 
2477 	fdp = p->p_fd;
2478 	FILEDESC_SLOCK(fdp);
2479 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2480 		fp = fget_noref(fdp, i);
2481 		if (fp != NULL && fp->f_ops == &mqueueops) {
2482 			mq = FPTOMQ(fp);
2483 			mtx_lock(&mq->mq_mutex);
2484 			notifier_remove(p, FPTOMQ(fp), i);
2485 			mtx_unlock(&mq->mq_mutex);
2486 		}
2487 	}
2488 	FILEDESC_SUNLOCK(fdp);
2489 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2490 }
2491 
2492 static int
2493 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2494 	struct thread *td)
2495 {
2496 	struct mqueue *mq = FPTOMQ(fp);
2497 	int revents = 0;
2498 
2499 	mtx_lock(&mq->mq_mutex);
2500 	if (events & (POLLIN | POLLRDNORM)) {
2501 		if (mq->mq_curmsgs) {
2502 			revents |= events & (POLLIN | POLLRDNORM);
2503 		} else {
2504 			mq->mq_flags |= MQ_RSEL;
2505 			selrecord(td, &mq->mq_rsel);
2506  		}
2507 	}
2508 	if (events & POLLOUT) {
2509 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2510 			revents |= POLLOUT;
2511 		else {
2512 			mq->mq_flags |= MQ_WSEL;
2513 			selrecord(td, &mq->mq_wsel);
2514 		}
2515 	}
2516 	mtx_unlock(&mq->mq_mutex);
2517 	return (revents);
2518 }
2519 
2520 static int
2521 mqf_close(struct file *fp, struct thread *td)
2522 {
2523 	struct mqfs_node *pn;
2524 
2525 	fp->f_ops = &badfileops;
2526 	pn = fp->f_data;
2527 	fp->f_data = NULL;
2528 	sx_xlock(&mqfs_data.mi_lock);
2529 	mqnode_release(pn);
2530 	sx_xunlock(&mqfs_data.mi_lock);
2531 	return (0);
2532 }
2533 
2534 static int
2535 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
2536 {
2537 	struct mqfs_node *pn = fp->f_data;
2538 
2539 	bzero(st, sizeof *st);
2540 	sx_xlock(&mqfs_data.mi_lock);
2541 	st->st_atim = pn->mn_atime;
2542 	st->st_mtim = pn->mn_mtime;
2543 	st->st_ctim = pn->mn_ctime;
2544 	st->st_birthtim = pn->mn_birth;
2545 	st->st_uid = pn->mn_uid;
2546 	st->st_gid = pn->mn_gid;
2547 	st->st_mode = S_IFIFO | pn->mn_mode;
2548 	sx_xunlock(&mqfs_data.mi_lock);
2549 	return (0);
2550 }
2551 
2552 static int
2553 mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2554     struct thread *td)
2555 {
2556 	struct mqfs_node *pn;
2557 	int error;
2558 
2559 	error = 0;
2560 	pn = fp->f_data;
2561 	sx_xlock(&mqfs_data.mi_lock);
2562 	error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
2563 	    active_cred);
2564 	if (error != 0)
2565 		goto out;
2566 	pn->mn_mode = mode & ACCESSPERMS;
2567 out:
2568 	sx_xunlock(&mqfs_data.mi_lock);
2569 	return (error);
2570 }
2571 
2572 static int
2573 mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2574     struct thread *td)
2575 {
2576 	struct mqfs_node *pn;
2577 	int error;
2578 
2579 	error = 0;
2580 	pn = fp->f_data;
2581 	sx_xlock(&mqfs_data.mi_lock);
2582 	if (uid == (uid_t)-1)
2583 		uid = pn->mn_uid;
2584 	if (gid == (gid_t)-1)
2585 		gid = pn->mn_gid;
2586 	if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
2587 	    (gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
2588 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN)))
2589 		goto out;
2590 	pn->mn_uid = uid;
2591 	pn->mn_gid = gid;
2592 out:
2593 	sx_xunlock(&mqfs_data.mi_lock);
2594 	return (error);
2595 }
2596 
2597 static int
2598 mqf_kqfilter(struct file *fp, struct knote *kn)
2599 {
2600 	struct mqueue *mq = FPTOMQ(fp);
2601 	int error = 0;
2602 
2603 	if (kn->kn_filter == EVFILT_READ) {
2604 		kn->kn_fop = &mq_rfiltops;
2605 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2606 	} else if (kn->kn_filter == EVFILT_WRITE) {
2607 		kn->kn_fop = &mq_wfiltops;
2608 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2609 	} else
2610 		error = EINVAL;
2611 	return (error);
2612 }
2613 
2614 static void
2615 filt_mqdetach(struct knote *kn)
2616 {
2617 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2618 
2619 	if (kn->kn_filter == EVFILT_READ)
2620 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2621 	else if (kn->kn_filter == EVFILT_WRITE)
2622 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2623 	else
2624 		panic("filt_mqdetach");
2625 }
2626 
2627 static int
2628 filt_mqread(struct knote *kn, long hint)
2629 {
2630 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2631 
2632 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2633 	return (mq->mq_curmsgs != 0);
2634 }
2635 
2636 static int
2637 filt_mqwrite(struct knote *kn, long hint)
2638 {
2639 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2640 
2641 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2642 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2643 }
2644 
2645 static int
2646 mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2647 {
2648 
2649 	kif->kf_type = KF_TYPE_MQUEUE;
2650 	return (0);
2651 }
2652 
2653 static struct fileops mqueueops = {
2654 	.fo_read		= invfo_rdwr,
2655 	.fo_write		= invfo_rdwr,
2656 	.fo_truncate		= invfo_truncate,
2657 	.fo_ioctl		= invfo_ioctl,
2658 	.fo_poll		= mqf_poll,
2659 	.fo_kqfilter		= mqf_kqfilter,
2660 	.fo_stat		= mqf_stat,
2661 	.fo_close		= mqf_close,
2662 	.fo_chmod		= mqf_chmod,
2663 	.fo_chown		= mqf_chown,
2664 	.fo_sendfile		= invfo_sendfile,
2665 	.fo_fill_kinfo		= mqf_fill_kinfo,
2666 	.fo_flags		= DFLAG_PASSABLE,
2667 };
2668 
2669 static struct vop_vector mqfs_vnodeops = {
2670 	.vop_default 		= &default_vnodeops,
2671 	.vop_access		= mqfs_access,
2672 	.vop_cachedlookup	= mqfs_lookup,
2673 	.vop_lookup		= vfs_cache_lookup,
2674 	.vop_reclaim		= mqfs_reclaim,
2675 	.vop_create		= mqfs_create,
2676 	.vop_remove		= mqfs_remove,
2677 	.vop_inactive		= mqfs_inactive,
2678 	.vop_open		= mqfs_open,
2679 	.vop_close		= mqfs_close,
2680 	.vop_getattr		= mqfs_getattr,
2681 	.vop_setattr		= mqfs_setattr,
2682 	.vop_read		= mqfs_read,
2683 	.vop_write		= VOP_EOPNOTSUPP,
2684 	.vop_readdir		= mqfs_readdir,
2685 	.vop_mkdir		= VOP_EOPNOTSUPP,
2686 	.vop_rmdir		= VOP_EOPNOTSUPP
2687 };
2688 VFS_VOP_VECTOR_REGISTER(mqfs_vnodeops);
2689 
2690 static struct vfsops mqfs_vfsops = {
2691 	.vfs_init 		= mqfs_init,
2692 	.vfs_uninit		= mqfs_uninit,
2693 	.vfs_mount		= mqfs_mount,
2694 	.vfs_unmount		= mqfs_unmount,
2695 	.vfs_root		= mqfs_root,
2696 	.vfs_statfs		= mqfs_statfs,
2697 };
2698 
2699 static struct vfsconf mqueuefs_vfsconf = {
2700 	.vfc_version = VFS_VERSION,
2701 	.vfc_name = "mqueuefs",
2702 	.vfc_vfsops = &mqfs_vfsops,
2703 	.vfc_typenum = -1,
2704 	.vfc_flags = VFCF_SYNTHETIC
2705 };
2706 
2707 static struct syscall_helper_data mq_syscalls[] = {
2708 	SYSCALL_INIT_HELPER(kmq_open),
2709 	SYSCALL_INIT_HELPER_F(kmq_setattr, SYF_CAPENABLED),
2710 	SYSCALL_INIT_HELPER_F(kmq_timedsend, SYF_CAPENABLED),
2711 	SYSCALL_INIT_HELPER_F(kmq_timedreceive, SYF_CAPENABLED),
2712 	SYSCALL_INIT_HELPER_F(kmq_notify, SYF_CAPENABLED),
2713 	SYSCALL_INIT_HELPER(kmq_unlink),
2714 	SYSCALL_INIT_LAST
2715 };
2716 
2717 #ifdef COMPAT_FREEBSD32
2718 #include <compat/freebsd32/freebsd32.h>
2719 #include <compat/freebsd32/freebsd32_proto.h>
2720 #include <compat/freebsd32/freebsd32_signal.h>
2721 #include <compat/freebsd32/freebsd32_syscall.h>
2722 #include <compat/freebsd32/freebsd32_util.h>
2723 
2724 static void
2725 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2726 {
2727 
2728 	to->mq_flags = from->mq_flags;
2729 	to->mq_maxmsg = from->mq_maxmsg;
2730 	to->mq_msgsize = from->mq_msgsize;
2731 	to->mq_curmsgs = from->mq_curmsgs;
2732 }
2733 
2734 static void
2735 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2736 {
2737 
2738 	to->mq_flags = from->mq_flags;
2739 	to->mq_maxmsg = from->mq_maxmsg;
2740 	to->mq_msgsize = from->mq_msgsize;
2741 	to->mq_curmsgs = from->mq_curmsgs;
2742 }
2743 
2744 int
2745 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2746 {
2747 	struct mq_attr attr;
2748 	struct mq_attr32 attr32;
2749 	int flags, error;
2750 
2751 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2752 		return (EINVAL);
2753 	flags = FFLAGS(uap->flags);
2754 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2755 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2756 		if (error)
2757 			return (error);
2758 		mq_attr_from32(&attr32, &attr);
2759 	}
2760 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2761 	    uap->attr != NULL ? &attr : NULL));
2762 }
2763 
2764 int
2765 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2766 {
2767 	struct mq_attr attr, oattr;
2768 	struct mq_attr32 attr32, oattr32;
2769 	int error;
2770 
2771 	if (uap->attr != NULL) {
2772 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2773 		if (error != 0)
2774 			return (error);
2775 		mq_attr_from32(&attr32, &attr);
2776 	}
2777 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2778 	    &oattr);
2779 	if (error == 0 && uap->oattr != NULL) {
2780 		mq_attr_to32(&oattr, &oattr32);
2781 		bzero(oattr32.__reserved, sizeof(oattr32.__reserved));
2782 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2783 	}
2784 	return (error);
2785 }
2786 
2787 int
2788 freebsd32_kmq_timedsend(struct thread *td,
2789     struct freebsd32_kmq_timedsend_args *uap)
2790 {
2791 	struct mqueue *mq;
2792 	struct file *fp;
2793 	struct timespec32 ets32;
2794 	struct timespec *abs_timeout, ets;
2795 	int error;
2796 	int waitok;
2797 
2798 	AUDIT_ARG_FD(uap->mqd);
2799 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2800 	if (error)
2801 		return (error);
2802 	if (uap->abs_timeout != NULL) {
2803 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2804 		if (error != 0)
2805 			goto out;
2806 		CP(ets32, ets, tv_sec);
2807 		CP(ets32, ets, tv_nsec);
2808 		abs_timeout = &ets;
2809 	} else
2810 		abs_timeout = NULL;
2811 	waitok = !(fp->f_flag & O_NONBLOCK);
2812 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2813 		uap->msg_prio, waitok, abs_timeout);
2814 out:
2815 	fdrop(fp, td);
2816 	return (error);
2817 }
2818 
2819 int
2820 freebsd32_kmq_timedreceive(struct thread *td,
2821     struct freebsd32_kmq_timedreceive_args *uap)
2822 {
2823 	struct mqueue *mq;
2824 	struct file *fp;
2825 	struct timespec32 ets32;
2826 	struct timespec *abs_timeout, ets;
2827 	int error, waitok;
2828 
2829 	AUDIT_ARG_FD(uap->mqd);
2830 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2831 	if (error)
2832 		return (error);
2833 	if (uap->abs_timeout != NULL) {
2834 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2835 		if (error != 0)
2836 			goto out;
2837 		CP(ets32, ets, tv_sec);
2838 		CP(ets32, ets, tv_nsec);
2839 		abs_timeout = &ets;
2840 	} else
2841 		abs_timeout = NULL;
2842 	waitok = !(fp->f_flag & O_NONBLOCK);
2843 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2844 		uap->msg_prio, waitok, abs_timeout);
2845 out:
2846 	fdrop(fp, td);
2847 	return (error);
2848 }
2849 
2850 int
2851 freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap)
2852 {
2853 	struct sigevent ev, *evp;
2854 	struct sigevent32 ev32;
2855 	int error;
2856 
2857 	if (uap->sigev == NULL) {
2858 		evp = NULL;
2859 	} else {
2860 		error = copyin(uap->sigev, &ev32, sizeof(ev32));
2861 		if (error != 0)
2862 			return (error);
2863 		error = convert_sigevent32(&ev32, &ev);
2864 		if (error != 0)
2865 			return (error);
2866 		evp = &ev;
2867 	}
2868 	return (kern_kmq_notify(td, uap->mqd, evp));
2869 }
2870 
2871 static struct syscall_helper_data mq32_syscalls[] = {
2872 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2873 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_setattr, SYF_CAPENABLED),
2874 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedsend, SYF_CAPENABLED),
2875 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_timedreceive, SYF_CAPENABLED),
2876 	SYSCALL32_INIT_HELPER_F(freebsd32_kmq_notify, SYF_CAPENABLED),
2877 	SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
2878 	SYSCALL_INIT_LAST
2879 };
2880 #endif
2881 
2882 static int
2883 mqinit(void)
2884 {
2885 	int error;
2886 
2887 	error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD);
2888 	if (error != 0)
2889 		return (error);
2890 #ifdef COMPAT_FREEBSD32
2891 	error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD);
2892 	if (error != 0)
2893 		return (error);
2894 #endif
2895 	return (0);
2896 }
2897 
2898 static int
2899 mqunload(void)
2900 {
2901 
2902 #ifdef COMPAT_FREEBSD32
2903 	syscall32_helper_unregister(mq32_syscalls);
2904 #endif
2905 	syscall_helper_unregister(mq_syscalls);
2906 	return (0);
2907 }
2908 
2909 static int
2910 mq_modload(struct module *module, int cmd, void *arg)
2911 {
2912 	int error = 0;
2913 
2914 	error = vfs_modevent(module, cmd, arg);
2915 	if (error != 0)
2916 		return (error);
2917 
2918 	switch (cmd) {
2919 	case MOD_LOAD:
2920 		error = mqinit();
2921 		if (error != 0)
2922 			mqunload();
2923 		break;
2924 	case MOD_UNLOAD:
2925 		error = mqunload();
2926 		break;
2927 	default:
2928 		break;
2929 	}
2930 	return (error);
2931 }
2932 
2933 static moduledata_t mqueuefs_mod = {
2934 	"mqueuefs",
2935 	mq_modload,
2936 	&mqueuefs_vfsconf
2937 };
2938 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2939 MODULE_VERSION(mqueuefs, 1);
2940