xref: /freebsd/sys/kern/uipc_mqueue.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
1 /*-
2  * Copyright (c) 2005 David Xu <davidxu@freebsd.org>
3  * 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  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 /*
29  * POSIX message queue implementation.
30  *
31  * 1) A mqueue filesystem can be mounted, each message queue appears
32  *    in mounted directory, user can change queue's permission and
33  *    ownership, or remove a queue. Manually creating a file in the
34  *    directory causes a message queue to be created in the kernel with
35  *    default message queue attributes applied and same name used, this
36  *    method is not advocated since mq_open syscall allows user to specify
37  *    different attributes. Also the file system can be mounted multiple
38  *    times at different mount points but shows same contents.
39  *
40  * 2) Standard POSIX message queue API. The syscalls do not use vfs layer,
41  *    but directly operate on internal data structure, this allows user to
42  *    use the IPC facility without having to mount mqueue file system.
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_capsicum.h"
49 #include "opt_compat.h"
50 
51 #include <sys/param.h>
52 #include <sys/kernel.h>
53 #include <sys/systm.h>
54 #include <sys/limits.h>
55 #include <sys/buf.h>
56 #include <sys/capsicum.h>
57 #include <sys/dirent.h>
58 #include <sys/event.h>
59 #include <sys/eventhandler.h>
60 #include <sys/fcntl.h>
61 #include <sys/file.h>
62 #include <sys/filedesc.h>
63 #include <sys/lock.h>
64 #include <sys/malloc.h>
65 #include <sys/module.h>
66 #include <sys/mount.h>
67 #include <sys/mqueue.h>
68 #include <sys/mutex.h>
69 #include <sys/namei.h>
70 #include <sys/posix4.h>
71 #include <sys/poll.h>
72 #include <sys/priv.h>
73 #include <sys/proc.h>
74 #include <sys/queue.h>
75 #include <sys/sysproto.h>
76 #include <sys/stat.h>
77 #include <sys/syscall.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysent.h>
80 #include <sys/sx.h>
81 #include <sys/sysctl.h>
82 #include <sys/taskqueue.h>
83 #include <sys/unistd.h>
84 #include <sys/user.h>
85 #include <sys/vnode.h>
86 #include <machine/atomic.h>
87 
88 FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
89 
90 /*
91  * Limits and constants
92  */
93 #define	MQFS_NAMELEN		NAME_MAX
94 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
95 
96 /* node types */
97 typedef enum {
98 	mqfstype_none = 0,
99 	mqfstype_root,
100 	mqfstype_dir,
101 	mqfstype_this,
102 	mqfstype_parent,
103 	mqfstype_file,
104 	mqfstype_symlink,
105 } mqfs_type_t;
106 
107 struct mqfs_node;
108 
109 /*
110  * mqfs_info: describes a mqfs instance
111  */
112 struct mqfs_info {
113 	struct sx		mi_lock;
114 	struct mqfs_node	*mi_root;
115 	struct unrhdr		*mi_unrhdr;
116 };
117 
118 struct mqfs_vdata {
119 	LIST_ENTRY(mqfs_vdata)	mv_link;
120 	struct mqfs_node	*mv_node;
121 	struct vnode		*mv_vnode;
122 	struct task		mv_task;
123 };
124 
125 /*
126  * mqfs_node: describes a node (file or directory) within a mqfs
127  */
128 struct mqfs_node {
129 	char			mn_name[MQFS_NAMELEN+1];
130 	struct mqfs_info	*mn_info;
131 	struct mqfs_node	*mn_parent;
132 	LIST_HEAD(,mqfs_node)	mn_children;
133 	LIST_ENTRY(mqfs_node)	mn_sibling;
134 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
135 	int			mn_refcount;
136 	mqfs_type_t		mn_type;
137 	int			mn_deleted;
138 	uint32_t		mn_fileno;
139 	void			*mn_data;
140 	struct timespec		mn_birth;
141 	struct timespec		mn_ctime;
142 	struct timespec		mn_atime;
143 	struct timespec		mn_mtime;
144 	uid_t			mn_uid;
145 	gid_t			mn_gid;
146 	int			mn_mode;
147 };
148 
149 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
150 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
151 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
152 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
153 				(fp)->f_data)->mn_data))
154 
155 TAILQ_HEAD(msgq, mqueue_msg);
156 
157 struct mqueue;
158 
159 struct mqueue_notifier {
160 	LIST_ENTRY(mqueue_notifier)	nt_link;
161 	struct sigevent			nt_sigev;
162 	ksiginfo_t			nt_ksi;
163 	struct proc			*nt_proc;
164 };
165 
166 struct mqueue {
167 	struct mtx	mq_mutex;
168 	int		mq_flags;
169 	long		mq_maxmsg;
170 	long		mq_msgsize;
171 	long		mq_curmsgs;
172 	long		mq_totalbytes;
173 	struct msgq	mq_msgq;
174 	int		mq_receivers;
175 	int		mq_senders;
176 	struct selinfo	mq_rsel;
177 	struct selinfo	mq_wsel;
178 	struct mqueue_notifier	*mq_notifier;
179 };
180 
181 #define	MQ_RSEL		0x01
182 #define	MQ_WSEL		0x02
183 
184 struct mqueue_msg {
185 	TAILQ_ENTRY(mqueue_msg)	msg_link;
186 	unsigned int	msg_prio;
187 	unsigned int	msg_size;
188 	/* following real data... */
189 };
190 
191 static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
192 	"POSIX real time message queue");
193 
194 static int	default_maxmsg  = 10;
195 static int	default_msgsize = 1024;
196 
197 static int	maxmsg = 100;
198 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
199     &maxmsg, 0, "Default maximum messages in queue");
200 static int	maxmsgsize = 16384;
201 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
202     &maxmsgsize, 0, "Default maximum message size");
203 static int	maxmq = 100;
204 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
205     &maxmq, 0, "maximum message queues");
206 static int	curmq = 0;
207 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
208     &curmq, 0, "current message queue number");
209 static int	unloadable = 0;
210 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
211 
212 static eventhandler_tag exit_tag;
213 
214 /* Only one instance per-system */
215 static struct mqfs_info		mqfs_data;
216 static uma_zone_t		mqnode_zone;
217 static uma_zone_t		mqueue_zone;
218 static uma_zone_t		mvdata_zone;
219 static uma_zone_t		mqnoti_zone;
220 static struct vop_vector	mqfs_vnodeops;
221 static struct fileops		mqueueops;
222 
223 /*
224  * Directory structure construction and manipulation
225  */
226 #ifdef notyet
227 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
228 	const char *name, int namelen, struct ucred *cred, int mode);
229 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
230 	const char *name, int namelen, struct ucred *cred, int mode);
231 #endif
232 
233 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
234 	const char *name, int namelen, struct ucred *cred, int mode);
235 static int	mqfs_destroy(struct mqfs_node *mn);
236 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
237 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
238 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
239 
240 /*
241  * Message queue construction and maniplation
242  */
243 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
244 static void	mqueue_free(struct mqueue *mq);
245 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
246 			size_t msg_len, unsigned msg_prio, int waitok,
247 			const struct timespec *abs_timeout);
248 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
249 			size_t msg_len, unsigned *msg_prio, int waitok,
250 			const struct timespec *abs_timeout);
251 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
252 			int timo);
253 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
254 			int timo);
255 static void	mqueue_send_notification(struct mqueue *mq);
256 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
257 static void	mq_proc_exit(void *arg, struct proc *p);
258 
259 /*
260  * kqueue filters
261  */
262 static void	filt_mqdetach(struct knote *kn);
263 static int	filt_mqread(struct knote *kn, long hint);
264 static int	filt_mqwrite(struct knote *kn, long hint);
265 
266 struct filterops mq_rfiltops = {
267 	.f_isfd = 1,
268 	.f_detach = filt_mqdetach,
269 	.f_event = filt_mqread,
270 };
271 struct filterops mq_wfiltops = {
272 	.f_isfd = 1,
273 	.f_detach = filt_mqdetach,
274 	.f_event = filt_mqwrite,
275 };
276 
277 /*
278  * Initialize fileno bitmap
279  */
280 static void
281 mqfs_fileno_init(struct mqfs_info *mi)
282 {
283 	struct unrhdr *up;
284 
285 	up = new_unrhdr(1, INT_MAX, NULL);
286 	mi->mi_unrhdr = up;
287 }
288 
289 /*
290  * Tear down fileno bitmap
291  */
292 static void
293 mqfs_fileno_uninit(struct mqfs_info *mi)
294 {
295 	struct unrhdr *up;
296 
297 	up = mi->mi_unrhdr;
298 	mi->mi_unrhdr = NULL;
299 	delete_unrhdr(up);
300 }
301 
302 /*
303  * Allocate a file number
304  */
305 static void
306 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
307 {
308 	/* make sure our parent has a file number */
309 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
310 		mqfs_fileno_alloc(mi, mn->mn_parent);
311 
312 	switch (mn->mn_type) {
313 	case mqfstype_root:
314 	case mqfstype_dir:
315 	case mqfstype_file:
316 	case mqfstype_symlink:
317 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
318 		break;
319 	case mqfstype_this:
320 		KASSERT(mn->mn_parent != NULL,
321 		    ("mqfstype_this node has no parent"));
322 		mn->mn_fileno = mn->mn_parent->mn_fileno;
323 		break;
324 	case mqfstype_parent:
325 		KASSERT(mn->mn_parent != NULL,
326 		    ("mqfstype_parent node has no parent"));
327 		if (mn->mn_parent == mi->mi_root) {
328 			mn->mn_fileno = mn->mn_parent->mn_fileno;
329 			break;
330 		}
331 		KASSERT(mn->mn_parent->mn_parent != NULL,
332 		    ("mqfstype_parent node has no grandparent"));
333 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
334 		break;
335 	default:
336 		KASSERT(0,
337 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
338 			mn->mn_type));
339 		break;
340 	}
341 }
342 
343 /*
344  * Release a file number
345  */
346 static void
347 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
348 {
349 	switch (mn->mn_type) {
350 	case mqfstype_root:
351 	case mqfstype_dir:
352 	case mqfstype_file:
353 	case mqfstype_symlink:
354 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
355 		break;
356 	case mqfstype_this:
357 	case mqfstype_parent:
358 		/* ignore these, as they don't "own" their file number */
359 		break;
360 	default:
361 		KASSERT(0,
362 		    ("mqfs_fileno_free() called for unknown type node: %d",
363 			mn->mn_type));
364 		break;
365 	}
366 }
367 
368 static __inline struct mqfs_node *
369 mqnode_alloc(void)
370 {
371 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
372 }
373 
374 static __inline void
375 mqnode_free(struct mqfs_node *node)
376 {
377 	uma_zfree(mqnode_zone, node);
378 }
379 
380 static __inline void
381 mqnode_addref(struct mqfs_node *node)
382 {
383 	atomic_fetchadd_int(&node->mn_refcount, 1);
384 }
385 
386 static __inline void
387 mqnode_release(struct mqfs_node *node)
388 {
389 	struct mqfs_info *mqfs;
390 	int old, exp;
391 
392 	mqfs = node->mn_info;
393 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
394 	if (node->mn_type == mqfstype_dir ||
395 	    node->mn_type == mqfstype_root)
396 		exp = 3; /* include . and .. */
397 	else
398 		exp = 1;
399 	if (old == exp) {
400 		int locked = sx_xlocked(&mqfs->mi_lock);
401 		if (!locked)
402 			sx_xlock(&mqfs->mi_lock);
403 		mqfs_destroy(node);
404 		if (!locked)
405 			sx_xunlock(&mqfs->mi_lock);
406 	}
407 }
408 
409 /*
410  * Add a node to a directory
411  */
412 static int
413 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
414 {
415 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
416 	KASSERT(parent->mn_info != NULL,
417 	    ("%s(): parent has no mn_info", __func__));
418 	KASSERT(parent->mn_type == mqfstype_dir ||
419 	    parent->mn_type == mqfstype_root,
420 	    ("%s(): parent is not a directory", __func__));
421 
422 	node->mn_info = parent->mn_info;
423 	node->mn_parent = parent;
424 	LIST_INIT(&node->mn_children);
425 	LIST_INIT(&node->mn_vnodes);
426 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
427 	mqnode_addref(parent);
428 	return (0);
429 }
430 
431 static struct mqfs_node *
432 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
433 	int nodetype)
434 {
435 	struct mqfs_node *node;
436 
437 	node = mqnode_alloc();
438 	strncpy(node->mn_name, name, namelen);
439 	node->mn_type = nodetype;
440 	node->mn_refcount = 1;
441 	vfs_timestamp(&node->mn_birth);
442 	node->mn_ctime = node->mn_atime = node->mn_mtime
443 		= node->mn_birth;
444 	node->mn_uid = cred->cr_uid;
445 	node->mn_gid = cred->cr_gid;
446 	node->mn_mode = mode;
447 	return (node);
448 }
449 
450 /*
451  * Create a file
452  */
453 static struct mqfs_node *
454 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
455 	struct ucred *cred, int mode)
456 {
457 	struct mqfs_node *node;
458 
459 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
460 	if (mqfs_add_node(parent, node) != 0) {
461 		mqnode_free(node);
462 		return (NULL);
463 	}
464 	return (node);
465 }
466 
467 /*
468  * Add . and .. to a directory
469  */
470 static int
471 mqfs_fixup_dir(struct mqfs_node *parent)
472 {
473 	struct mqfs_node *dir;
474 
475 	dir = mqnode_alloc();
476 	dir->mn_name[0] = '.';
477 	dir->mn_type = mqfstype_this;
478 	dir->mn_refcount = 1;
479 	if (mqfs_add_node(parent, dir) != 0) {
480 		mqnode_free(dir);
481 		return (-1);
482 	}
483 
484 	dir = mqnode_alloc();
485 	dir->mn_name[0] = dir->mn_name[1] = '.';
486 	dir->mn_type = mqfstype_parent;
487 	dir->mn_refcount = 1;
488 
489 	if (mqfs_add_node(parent, dir) != 0) {
490 		mqnode_free(dir);
491 		return (-1);
492 	}
493 
494 	return (0);
495 }
496 
497 #ifdef notyet
498 
499 /*
500  * Create a directory
501  */
502 static struct mqfs_node *
503 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
504 	struct ucred *cred, int mode)
505 {
506 	struct mqfs_node *node;
507 
508 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
509 	if (mqfs_add_node(parent, node) != 0) {
510 		mqnode_free(node);
511 		return (NULL);
512 	}
513 
514 	if (mqfs_fixup_dir(node) != 0) {
515 		mqfs_destroy(node);
516 		return (NULL);
517 	}
518 	return (node);
519 }
520 
521 /*
522  * Create a symlink
523  */
524 static struct mqfs_node *
525 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
526 	struct ucred *cred, int mode)
527 {
528 	struct mqfs_node *node;
529 
530 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
531 	if (mqfs_add_node(parent, node) != 0) {
532 		mqnode_free(node);
533 		return (NULL);
534 	}
535 	return (node);
536 }
537 
538 #endif
539 
540 /*
541  * Destroy a node or a tree of nodes
542  */
543 static int
544 mqfs_destroy(struct mqfs_node *node)
545 {
546 	struct mqfs_node *parent;
547 
548 	KASSERT(node != NULL,
549 	    ("%s(): node is NULL", __func__));
550 	KASSERT(node->mn_info != NULL,
551 	    ("%s(): node has no mn_info", __func__));
552 
553 	/* destroy children */
554 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
555 		while (! LIST_EMPTY(&node->mn_children))
556 			mqfs_destroy(LIST_FIRST(&node->mn_children));
557 
558 	/* unlink from parent */
559 	if ((parent = node->mn_parent) != NULL) {
560 		KASSERT(parent->mn_info == node->mn_info,
561 		    ("%s(): parent has different mn_info", __func__));
562 		LIST_REMOVE(node, mn_sibling);
563 	}
564 
565 	if (node->mn_fileno != 0)
566 		mqfs_fileno_free(node->mn_info, node);
567 	if (node->mn_data != NULL)
568 		mqueue_free(node->mn_data);
569 	mqnode_free(node);
570 	return (0);
571 }
572 
573 /*
574  * Mount a mqfs instance
575  */
576 static int
577 mqfs_mount(struct mount *mp)
578 {
579 	struct statfs *sbp;
580 
581 	if (mp->mnt_flag & MNT_UPDATE)
582 		return (EOPNOTSUPP);
583 
584 	mp->mnt_data = &mqfs_data;
585 	MNT_ILOCK(mp);
586 	mp->mnt_flag |= MNT_LOCAL;
587 	MNT_IUNLOCK(mp);
588 	vfs_getnewfsid(mp);
589 
590 	sbp = &mp->mnt_stat;
591 	vfs_mountedfrom(mp, "mqueue");
592 	sbp->f_bsize = PAGE_SIZE;
593 	sbp->f_iosize = PAGE_SIZE;
594 	sbp->f_blocks = 1;
595 	sbp->f_bfree = 0;
596 	sbp->f_bavail = 0;
597 	sbp->f_files = 1;
598 	sbp->f_ffree = 0;
599 	return (0);
600 }
601 
602 /*
603  * Unmount a mqfs instance
604  */
605 static int
606 mqfs_unmount(struct mount *mp, int mntflags)
607 {
608 	int error;
609 
610 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0,
611 	    curthread);
612 	return (error);
613 }
614 
615 /*
616  * Return a root vnode
617  */
618 static int
619 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
620 {
621 	struct mqfs_info *mqfs;
622 	int ret;
623 
624 	mqfs = VFSTOMQFS(mp);
625 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
626 	return (ret);
627 }
628 
629 /*
630  * Return filesystem stats
631  */
632 static int
633 mqfs_statfs(struct mount *mp, struct statfs *sbp)
634 {
635 	/* XXX update statistics */
636 	return (0);
637 }
638 
639 /*
640  * Initialize a mqfs instance
641  */
642 static int
643 mqfs_init(struct vfsconf *vfc)
644 {
645 	struct mqfs_node *root;
646 	struct mqfs_info *mi;
647 
648 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
649 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
650 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
651 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
652 	mvdata_zone = uma_zcreate("mvdata",
653 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
654 		NULL, UMA_ALIGN_PTR, 0);
655 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
656 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
657 	mi = &mqfs_data;
658 	sx_init(&mi->mi_lock, "mqfs lock");
659 	/* set up the root diretory */
660 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
661 		mqfstype_root);
662 	root->mn_info = mi;
663 	LIST_INIT(&root->mn_children);
664 	LIST_INIT(&root->mn_vnodes);
665 	mi->mi_root = root;
666 	mqfs_fileno_init(mi);
667 	mqfs_fileno_alloc(mi, root);
668 	mqfs_fixup_dir(root);
669 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
670 	    EVENTHANDLER_PRI_ANY);
671 	mq_fdclose = mqueue_fdclose;
672 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
673 	return (0);
674 }
675 
676 /*
677  * Destroy a mqfs instance
678  */
679 static int
680 mqfs_uninit(struct vfsconf *vfc)
681 {
682 	struct mqfs_info *mi;
683 
684 	if (!unloadable)
685 		return (EOPNOTSUPP);
686 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
687 	mi = &mqfs_data;
688 	mqfs_destroy(mi->mi_root);
689 	mi->mi_root = NULL;
690 	mqfs_fileno_uninit(mi);
691 	sx_destroy(&mi->mi_lock);
692 	uma_zdestroy(mqnode_zone);
693 	uma_zdestroy(mqueue_zone);
694 	uma_zdestroy(mvdata_zone);
695 	uma_zdestroy(mqnoti_zone);
696 	return (0);
697 }
698 
699 /*
700  * task routine
701  */
702 static void
703 do_recycle(void *context, int pending __unused)
704 {
705 	struct vnode *vp = (struct vnode *)context;
706 
707 	vrecycle(vp);
708 	vdrop(vp);
709 }
710 
711 /*
712  * Allocate a vnode
713  */
714 static int
715 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
716 {
717 	struct mqfs_vdata *vd;
718 	struct mqfs_info  *mqfs;
719 	struct vnode *newvpp;
720 	int error;
721 
722 	mqfs = pn->mn_info;
723 	*vpp = NULL;
724 	sx_xlock(&mqfs->mi_lock);
725 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
726 		if (vd->mv_vnode->v_mount == mp) {
727 			vhold(vd->mv_vnode);
728 			break;
729 		}
730 	}
731 
732 	if (vd != NULL) {
733 found:
734 		*vpp = vd->mv_vnode;
735 		sx_xunlock(&mqfs->mi_lock);
736 		error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
737 		vdrop(*vpp);
738 		return (error);
739 	}
740 	sx_xunlock(&mqfs->mi_lock);
741 
742 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
743 	if (error)
744 		return (error);
745 	vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
746 	error = insmntque(newvpp, mp);
747 	if (error != 0)
748 		return (error);
749 
750 	sx_xlock(&mqfs->mi_lock);
751 	/*
752 	 * Check if it has already been allocated
753 	 * while we were blocked.
754 	 */
755 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
756 		if (vd->mv_vnode->v_mount == mp) {
757 			vhold(vd->mv_vnode);
758 			sx_xunlock(&mqfs->mi_lock);
759 
760 			vgone(newvpp);
761 			vput(newvpp);
762 			goto found;
763 		}
764 	}
765 
766 	*vpp = newvpp;
767 
768 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
769 	(*vpp)->v_data = vd;
770 	vd->mv_vnode = *vpp;
771 	vd->mv_node = pn;
772 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
773 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
774 	mqnode_addref(pn);
775 	switch (pn->mn_type) {
776 	case mqfstype_root:
777 		(*vpp)->v_vflag = VV_ROOT;
778 		/* fall through */
779 	case mqfstype_dir:
780 	case mqfstype_this:
781 	case mqfstype_parent:
782 		(*vpp)->v_type = VDIR;
783 		break;
784 	case mqfstype_file:
785 		(*vpp)->v_type = VREG;
786 		break;
787 	case mqfstype_symlink:
788 		(*vpp)->v_type = VLNK;
789 		break;
790 	case mqfstype_none:
791 		KASSERT(0, ("mqfs_allocf called for null node\n"));
792 	default:
793 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
794 	}
795 	sx_xunlock(&mqfs->mi_lock);
796 	return (0);
797 }
798 
799 /*
800  * Search a directory entry
801  */
802 static struct mqfs_node *
803 mqfs_search(struct mqfs_node *pd, const char *name, int len)
804 {
805 	struct mqfs_node *pn;
806 
807 	sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
808 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
809 		if (strncmp(pn->mn_name, name, len) == 0 &&
810 		    pn->mn_name[len] == '\0')
811 			return (pn);
812 	}
813 	return (NULL);
814 }
815 
816 /*
817  * Look up a file or directory.
818  */
819 static int
820 mqfs_lookupx(struct vop_cachedlookup_args *ap)
821 {
822 	struct componentname *cnp;
823 	struct vnode *dvp, **vpp;
824 	struct mqfs_node *pd;
825 	struct mqfs_node *pn;
826 	struct mqfs_info *mqfs;
827 	int nameiop, flags, error, namelen;
828 	char *pname;
829 	struct thread *td;
830 
831 	cnp = ap->a_cnp;
832 	vpp = ap->a_vpp;
833 	dvp = ap->a_dvp;
834 	pname = cnp->cn_nameptr;
835 	namelen = cnp->cn_namelen;
836 	td = cnp->cn_thread;
837 	flags = cnp->cn_flags;
838 	nameiop = cnp->cn_nameiop;
839 	pd = VTON(dvp);
840 	pn = NULL;
841 	mqfs = pd->mn_info;
842 	*vpp = NULLVP;
843 
844 	if (dvp->v_type != VDIR)
845 		return (ENOTDIR);
846 
847 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
848 	if (error)
849 		return (error);
850 
851 	/* shortcut: check if the name is too long */
852 	if (cnp->cn_namelen >= MQFS_NAMELEN)
853 		return (ENOENT);
854 
855 	/* self */
856 	if (namelen == 1 && pname[0] == '.') {
857 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
858 			return (EINVAL);
859 		pn = pd;
860 		*vpp = dvp;
861 		VREF(dvp);
862 		return (0);
863 	}
864 
865 	/* parent */
866 	if (cnp->cn_flags & ISDOTDOT) {
867 		if (dvp->v_vflag & VV_ROOT)
868 			return (EIO);
869 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
870 			return (EINVAL);
871 		VOP_UNLOCK(dvp, 0);
872 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
873 		pn = pd->mn_parent;
874 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
875 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
876 		return (error);
877 	}
878 
879 	/* named node */
880 	sx_xlock(&mqfs->mi_lock);
881 	pn = mqfs_search(pd, pname, namelen);
882 	if (pn != NULL)
883 		mqnode_addref(pn);
884 	sx_xunlock(&mqfs->mi_lock);
885 
886 	/* found */
887 	if (pn != NULL) {
888 		/* DELETE */
889 		if (nameiop == DELETE && (flags & ISLASTCN)) {
890 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
891 			if (error) {
892 				mqnode_release(pn);
893 				return (error);
894 			}
895 			if (*vpp == dvp) {
896 				VREF(dvp);
897 				*vpp = dvp;
898 				mqnode_release(pn);
899 				return (0);
900 			}
901 		}
902 
903 		/* allocate vnode */
904 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
905 		mqnode_release(pn);
906 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
907 			cache_enter(dvp, *vpp, cnp);
908 		return (error);
909 	}
910 
911 	/* not found */
912 
913 	/* will create a new entry in the directory ? */
914 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
915 	    && (flags & ISLASTCN)) {
916 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
917 		if (error)
918 			return (error);
919 		cnp->cn_flags |= SAVENAME;
920 		return (EJUSTRETURN);
921 	}
922 	return (ENOENT);
923 }
924 
925 #if 0
926 struct vop_lookup_args {
927 	struct vop_generic_args a_gen;
928 	struct vnode *a_dvp;
929 	struct vnode **a_vpp;
930 	struct componentname *a_cnp;
931 };
932 #endif
933 
934 /*
935  * vnode lookup operation
936  */
937 static int
938 mqfs_lookup(struct vop_cachedlookup_args *ap)
939 {
940 	int rc;
941 
942 	rc = mqfs_lookupx(ap);
943 	return (rc);
944 }
945 
946 #if 0
947 struct vop_create_args {
948 	struct vnode *a_dvp;
949 	struct vnode **a_vpp;
950 	struct componentname *a_cnp;
951 	struct vattr *a_vap;
952 };
953 #endif
954 
955 /*
956  * vnode creation operation
957  */
958 static int
959 mqfs_create(struct vop_create_args *ap)
960 {
961 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
962 	struct componentname *cnp = ap->a_cnp;
963 	struct mqfs_node *pd;
964 	struct mqfs_node *pn;
965 	struct mqueue *mq;
966 	int error;
967 
968 	pd = VTON(ap->a_dvp);
969 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
970 		return (ENOTDIR);
971 	mq = mqueue_alloc(NULL);
972 	if (mq == NULL)
973 		return (EAGAIN);
974 	sx_xlock(&mqfs->mi_lock);
975 	if ((cnp->cn_flags & HASBUF) == 0)
976 		panic("%s: no name", __func__);
977 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
978 		cnp->cn_cred, ap->a_vap->va_mode);
979 	if (pn == NULL) {
980 		sx_xunlock(&mqfs->mi_lock);
981 		error = ENOSPC;
982 	} else {
983 		mqnode_addref(pn);
984 		sx_xunlock(&mqfs->mi_lock);
985 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
986 		mqnode_release(pn);
987 		if (error)
988 			mqfs_destroy(pn);
989 		else
990 			pn->mn_data = mq;
991 	}
992 	if (error)
993 		mqueue_free(mq);
994 	return (error);
995 }
996 
997 /*
998  * Remove an entry
999  */
1000 static
1001 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
1002 {
1003 	struct mqfs_node *parent;
1004 	struct mqfs_vdata *vd;
1005 	int error = 0;
1006 
1007 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
1008 
1009 	if (ucred->cr_uid != pn->mn_uid &&
1010 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
1011 		error = EACCES;
1012 	else if (!pn->mn_deleted) {
1013 		parent = pn->mn_parent;
1014 		pn->mn_parent = NULL;
1015 		pn->mn_deleted = 1;
1016 		LIST_REMOVE(pn, mn_sibling);
1017 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1018 			cache_purge(vd->mv_vnode);
1019 			vhold(vd->mv_vnode);
1020 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1021 		}
1022 		mqnode_release(pn);
1023 		mqnode_release(parent);
1024 	} else
1025 		error = ENOENT;
1026 	return (error);
1027 }
1028 
1029 #if 0
1030 struct vop_remove_args {
1031 	struct vnode *a_dvp;
1032 	struct vnode *a_vp;
1033 	struct componentname *a_cnp;
1034 };
1035 #endif
1036 
1037 /*
1038  * vnode removal operation
1039  */
1040 static int
1041 mqfs_remove(struct vop_remove_args *ap)
1042 {
1043 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1044 	struct mqfs_node *pn;
1045 	int error;
1046 
1047 	if (ap->a_vp->v_type == VDIR)
1048                 return (EPERM);
1049 	pn = VTON(ap->a_vp);
1050 	sx_xlock(&mqfs->mi_lock);
1051 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1052 	sx_xunlock(&mqfs->mi_lock);
1053 	return (error);
1054 }
1055 
1056 #if 0
1057 struct vop_inactive_args {
1058 	struct vnode *a_vp;
1059 	struct thread *a_td;
1060 };
1061 #endif
1062 
1063 static int
1064 mqfs_inactive(struct vop_inactive_args *ap)
1065 {
1066 	struct mqfs_node *pn = VTON(ap->a_vp);
1067 
1068 	if (pn->mn_deleted)
1069 		vrecycle(ap->a_vp);
1070 	return (0);
1071 }
1072 
1073 #if 0
1074 struct vop_reclaim_args {
1075 	struct vop_generic_args a_gen;
1076 	struct vnode *a_vp;
1077 	struct thread *a_td;
1078 };
1079 #endif
1080 
1081 static int
1082 mqfs_reclaim(struct vop_reclaim_args *ap)
1083 {
1084 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1085 	struct vnode *vp = ap->a_vp;
1086 	struct mqfs_node *pn;
1087 	struct mqfs_vdata *vd;
1088 
1089 	vd = vp->v_data;
1090 	pn = vd->mv_node;
1091 	sx_xlock(&mqfs->mi_lock);
1092 	vp->v_data = NULL;
1093 	LIST_REMOVE(vd, mv_link);
1094 	uma_zfree(mvdata_zone, vd);
1095 	mqnode_release(pn);
1096 	sx_xunlock(&mqfs->mi_lock);
1097 	return (0);
1098 }
1099 
1100 #if 0
1101 struct vop_open_args {
1102 	struct vop_generic_args a_gen;
1103 	struct vnode *a_vp;
1104 	int a_mode;
1105 	struct ucred *a_cred;
1106 	struct thread *a_td;
1107 	struct file *a_fp;
1108 };
1109 #endif
1110 
1111 static int
1112 mqfs_open(struct vop_open_args *ap)
1113 {
1114 	return (0);
1115 }
1116 
1117 #if 0
1118 struct vop_close_args {
1119 	struct vop_generic_args a_gen;
1120 	struct vnode *a_vp;
1121 	int a_fflag;
1122 	struct ucred *a_cred;
1123 	struct thread *a_td;
1124 };
1125 #endif
1126 
1127 static int
1128 mqfs_close(struct vop_close_args *ap)
1129 {
1130 	return (0);
1131 }
1132 
1133 #if 0
1134 struct vop_access_args {
1135 	struct vop_generic_args a_gen;
1136 	struct vnode *a_vp;
1137 	accmode_t a_accmode;
1138 	struct ucred *a_cred;
1139 	struct thread *a_td;
1140 };
1141 #endif
1142 
1143 /*
1144  * Verify permissions
1145  */
1146 static int
1147 mqfs_access(struct vop_access_args *ap)
1148 {
1149 	struct vnode *vp = ap->a_vp;
1150 	struct vattr vattr;
1151 	int error;
1152 
1153 	error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1154 	if (error)
1155 		return (error);
1156 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1157 	    vattr.va_gid, ap->a_accmode, ap->a_cred, NULL);
1158 	return (error);
1159 }
1160 
1161 #if 0
1162 struct vop_getattr_args {
1163 	struct vop_generic_args a_gen;
1164 	struct vnode *a_vp;
1165 	struct vattr *a_vap;
1166 	struct ucred *a_cred;
1167 };
1168 #endif
1169 
1170 /*
1171  * Get file attributes
1172  */
1173 static int
1174 mqfs_getattr(struct vop_getattr_args *ap)
1175 {
1176 	struct vnode *vp = ap->a_vp;
1177 	struct mqfs_node *pn = VTON(vp);
1178 	struct vattr *vap = ap->a_vap;
1179 	int error = 0;
1180 
1181 	vap->va_type = vp->v_type;
1182 	vap->va_mode = pn->mn_mode;
1183 	vap->va_nlink = 1;
1184 	vap->va_uid = pn->mn_uid;
1185 	vap->va_gid = pn->mn_gid;
1186 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1187 	vap->va_fileid = pn->mn_fileno;
1188 	vap->va_size = 0;
1189 	vap->va_blocksize = PAGE_SIZE;
1190 	vap->va_bytes = vap->va_size = 0;
1191 	vap->va_atime = pn->mn_atime;
1192 	vap->va_mtime = pn->mn_mtime;
1193 	vap->va_ctime = pn->mn_ctime;
1194 	vap->va_birthtime = pn->mn_birth;
1195 	vap->va_gen = 0;
1196 	vap->va_flags = 0;
1197 	vap->va_rdev = NODEV;
1198 	vap->va_bytes = 0;
1199 	vap->va_filerev = 0;
1200 	return (error);
1201 }
1202 
1203 #if 0
1204 struct vop_setattr_args {
1205 	struct vop_generic_args a_gen;
1206 	struct vnode *a_vp;
1207 	struct vattr *a_vap;
1208 	struct ucred *a_cred;
1209 };
1210 #endif
1211 /*
1212  * Set attributes
1213  */
1214 static int
1215 mqfs_setattr(struct vop_setattr_args *ap)
1216 {
1217 	struct mqfs_node *pn;
1218 	struct vattr *vap;
1219 	struct vnode *vp;
1220 	struct thread *td;
1221 	int c, error;
1222 	uid_t uid;
1223 	gid_t gid;
1224 
1225 	td = curthread;
1226 	vap = ap->a_vap;
1227 	vp = ap->a_vp;
1228 	if ((vap->va_type != VNON) ||
1229 	    (vap->va_nlink != VNOVAL) ||
1230 	    (vap->va_fsid != VNOVAL) ||
1231 	    (vap->va_fileid != VNOVAL) ||
1232 	    (vap->va_blocksize != VNOVAL) ||
1233 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1234 	    (vap->va_rdev != VNOVAL) ||
1235 	    ((int)vap->va_bytes != VNOVAL) ||
1236 	    (vap->va_gen != VNOVAL)) {
1237 		return (EINVAL);
1238 	}
1239 
1240 	pn = VTON(vp);
1241 
1242 	error = c = 0;
1243 	if (vap->va_uid == (uid_t)VNOVAL)
1244 		uid = pn->mn_uid;
1245 	else
1246 		uid = vap->va_uid;
1247 	if (vap->va_gid == (gid_t)VNOVAL)
1248 		gid = pn->mn_gid;
1249 	else
1250 		gid = vap->va_gid;
1251 
1252 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1253 		/*
1254 		 * To modify the ownership of a file, must possess VADMIN
1255 		 * for that file.
1256 		 */
1257 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1258 			return (error);
1259 
1260 		/*
1261 		 * XXXRW: Why is there a privilege check here: shouldn't the
1262 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1263 		 * below definitely right?
1264 		 */
1265 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1266 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1267 		    (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1268 			return (error);
1269 		pn->mn_uid = uid;
1270 		pn->mn_gid = gid;
1271 		c = 1;
1272 	}
1273 
1274 	if (vap->va_mode != (mode_t)VNOVAL) {
1275 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1276 		    (error = priv_check(td, PRIV_MQ_ADMIN)))
1277 			return (error);
1278 		pn->mn_mode = vap->va_mode;
1279 		c = 1;
1280 	}
1281 
1282 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1283 		/* See the comment in ufs_vnops::ufs_setattr(). */
1284 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1285 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1286 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1287 			return (error);
1288 		if (vap->va_atime.tv_sec != VNOVAL) {
1289 			pn->mn_atime = vap->va_atime;
1290 		}
1291 		if (vap->va_mtime.tv_sec != VNOVAL) {
1292 			pn->mn_mtime = vap->va_mtime;
1293 		}
1294 		c = 1;
1295 	}
1296 	if (c) {
1297 		vfs_timestamp(&pn->mn_ctime);
1298 	}
1299 	return (0);
1300 }
1301 
1302 #if 0
1303 struct vop_read_args {
1304 	struct vop_generic_args a_gen;
1305 	struct vnode *a_vp;
1306 	struct uio *a_uio;
1307 	int a_ioflag;
1308 	struct ucred *a_cred;
1309 };
1310 #endif
1311 
1312 /*
1313  * Read from a file
1314  */
1315 static int
1316 mqfs_read(struct vop_read_args *ap)
1317 {
1318 	char buf[80];
1319 	struct vnode *vp = ap->a_vp;
1320 	struct uio *uio = ap->a_uio;
1321 	struct mqfs_node *pn;
1322 	struct mqueue *mq;
1323 	int len, error;
1324 
1325 	if (vp->v_type != VREG)
1326 		return (EINVAL);
1327 
1328 	pn = VTON(vp);
1329 	mq = VTOMQ(vp);
1330 	snprintf(buf, sizeof(buf),
1331 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1332 		mq->mq_totalbytes,
1333 		mq->mq_maxmsg,
1334 		mq->mq_curmsgs,
1335 		mq->mq_msgsize);
1336 	buf[sizeof(buf)-1] = '\0';
1337 	len = strlen(buf);
1338 	error = uiomove_frombuf(buf, len, uio);
1339 	return (error);
1340 }
1341 
1342 #if 0
1343 struct vop_readdir_args {
1344 	struct vop_generic_args a_gen;
1345 	struct vnode *a_vp;
1346 	struct uio *a_uio;
1347 	struct ucred *a_cred;
1348 	int *a_eofflag;
1349 	int *a_ncookies;
1350 	u_long **a_cookies;
1351 };
1352 #endif
1353 
1354 /*
1355  * Return directory entries.
1356  */
1357 static int
1358 mqfs_readdir(struct vop_readdir_args *ap)
1359 {
1360 	struct vnode *vp;
1361 	struct mqfs_info *mi;
1362 	struct mqfs_node *pd;
1363 	struct mqfs_node *pn;
1364 	struct dirent entry;
1365 	struct uio *uio;
1366 	int *tmp_ncookies = NULL;
1367 	off_t offset;
1368 	int error, i;
1369 
1370 	vp = ap->a_vp;
1371 	mi = VFSTOMQFS(vp->v_mount);
1372 	pd = VTON(vp);
1373 	uio = ap->a_uio;
1374 
1375 	if (vp->v_type != VDIR)
1376 		return (ENOTDIR);
1377 
1378 	if (uio->uio_offset < 0)
1379 		return (EINVAL);
1380 
1381 	if (ap->a_ncookies != NULL) {
1382 		tmp_ncookies = ap->a_ncookies;
1383 		*ap->a_ncookies = 0;
1384 		ap->a_ncookies = NULL;
1385         }
1386 
1387 	error = 0;
1388 	offset = 0;
1389 
1390 	sx_xlock(&mi->mi_lock);
1391 
1392 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1393 		entry.d_reclen = sizeof(entry);
1394 		if (!pn->mn_fileno)
1395 			mqfs_fileno_alloc(mi, pn);
1396 		entry.d_fileno = pn->mn_fileno;
1397 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1398 			entry.d_name[i] = pn->mn_name[i];
1399 		entry.d_name[i] = 0;
1400 		entry.d_namlen = i;
1401 		switch (pn->mn_type) {
1402 		case mqfstype_root:
1403 		case mqfstype_dir:
1404 		case mqfstype_this:
1405 		case mqfstype_parent:
1406 			entry.d_type = DT_DIR;
1407 			break;
1408 		case mqfstype_file:
1409 			entry.d_type = DT_REG;
1410 			break;
1411 		case mqfstype_symlink:
1412 			entry.d_type = DT_LNK;
1413 			break;
1414 		default:
1415 			panic("%s has unexpected node type: %d", pn->mn_name,
1416 				pn->mn_type);
1417 		}
1418 		if (entry.d_reclen > uio->uio_resid)
1419                         break;
1420 		if (offset >= uio->uio_offset) {
1421 			error = vfs_read_dirent(ap, &entry, offset);
1422                         if (error)
1423                                 break;
1424                 }
1425                 offset += entry.d_reclen;
1426 	}
1427 	sx_xunlock(&mi->mi_lock);
1428 
1429 	uio->uio_offset = offset;
1430 
1431 	if (tmp_ncookies != NULL)
1432 		ap->a_ncookies = tmp_ncookies;
1433 
1434 	return (error);
1435 }
1436 
1437 #ifdef notyet
1438 
1439 #if 0
1440 struct vop_mkdir_args {
1441 	struct vnode *a_dvp;
1442 	struvt vnode **a_vpp;
1443 	struvt componentname *a_cnp;
1444 	struct vattr *a_vap;
1445 };
1446 #endif
1447 
1448 /*
1449  * Create a directory.
1450  */
1451 static int
1452 mqfs_mkdir(struct vop_mkdir_args *ap)
1453 {
1454 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1455 	struct componentname *cnp = ap->a_cnp;
1456 	struct mqfs_node *pd = VTON(ap->a_dvp);
1457 	struct mqfs_node *pn;
1458 	int error;
1459 
1460 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1461 		return (ENOTDIR);
1462 	sx_xlock(&mqfs->mi_lock);
1463 	if ((cnp->cn_flags & HASBUF) == 0)
1464 		panic("%s: no name", __func__);
1465 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1466 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1467 	if (pn != NULL)
1468 		mqnode_addref(pn);
1469 	sx_xunlock(&mqfs->mi_lock);
1470 	if (pn == NULL) {
1471 		error = ENOSPC;
1472 	} else {
1473 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1474 		mqnode_release(pn);
1475 	}
1476 	return (error);
1477 }
1478 
1479 #if 0
1480 struct vop_rmdir_args {
1481 	struct vnode *a_dvp;
1482 	struct vnode *a_vp;
1483 	struct componentname *a_cnp;
1484 };
1485 #endif
1486 
1487 /*
1488  * Remove a directory.
1489  */
1490 static int
1491 mqfs_rmdir(struct vop_rmdir_args *ap)
1492 {
1493 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1494 	struct mqfs_node *pn = VTON(ap->a_vp);
1495 	struct mqfs_node *pt;
1496 
1497 	if (pn->mn_type != mqfstype_dir)
1498 		return (ENOTDIR);
1499 
1500 	sx_xlock(&mqfs->mi_lock);
1501 	if (pn->mn_deleted) {
1502 		sx_xunlock(&mqfs->mi_lock);
1503 		return (ENOENT);
1504 	}
1505 
1506 	pt = LIST_FIRST(&pn->mn_children);
1507 	pt = LIST_NEXT(pt, mn_sibling);
1508 	pt = LIST_NEXT(pt, mn_sibling);
1509 	if (pt != NULL) {
1510 		sx_xunlock(&mqfs->mi_lock);
1511 		return (ENOTEMPTY);
1512 	}
1513 	pt = pn->mn_parent;
1514 	pn->mn_parent = NULL;
1515 	pn->mn_deleted = 1;
1516 	LIST_REMOVE(pn, mn_sibling);
1517 	mqnode_release(pn);
1518 	mqnode_release(pt);
1519 	sx_xunlock(&mqfs->mi_lock);
1520 	cache_purge(ap->a_vp);
1521 	return (0);
1522 }
1523 
1524 #endif /* notyet */
1525 
1526 /*
1527  * Allocate a message queue
1528  */
1529 static struct mqueue *
1530 mqueue_alloc(const struct mq_attr *attr)
1531 {
1532 	struct mqueue *mq;
1533 
1534 	if (curmq >= maxmq)
1535 		return (NULL);
1536 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1537 	TAILQ_INIT(&mq->mq_msgq);
1538 	if (attr != NULL) {
1539 		mq->mq_maxmsg = attr->mq_maxmsg;
1540 		mq->mq_msgsize = attr->mq_msgsize;
1541 	} else {
1542 		mq->mq_maxmsg = default_maxmsg;
1543 		mq->mq_msgsize = default_msgsize;
1544 	}
1545 	mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1546 	knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1547 	knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1548 	atomic_add_int(&curmq, 1);
1549 	return (mq);
1550 }
1551 
1552 /*
1553  * Destroy a message queue
1554  */
1555 static void
1556 mqueue_free(struct mqueue *mq)
1557 {
1558 	struct mqueue_msg *msg;
1559 
1560 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1561 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1562 		free(msg, M_MQUEUEDATA);
1563 	}
1564 
1565 	mtx_destroy(&mq->mq_mutex);
1566 	seldrain(&mq->mq_rsel);
1567 	seldrain(&mq->mq_wsel);
1568 	knlist_destroy(&mq->mq_rsel.si_note);
1569 	knlist_destroy(&mq->mq_wsel.si_note);
1570 	uma_zfree(mqueue_zone, mq);
1571 	atomic_add_int(&curmq, -1);
1572 }
1573 
1574 /*
1575  * Load a message from user space
1576  */
1577 static struct mqueue_msg *
1578 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1579 {
1580 	struct mqueue_msg *msg;
1581 	size_t len;
1582 	int error;
1583 
1584 	len = sizeof(struct mqueue_msg) + msg_size;
1585 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1586 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1587 	    msg_size);
1588 	if (error) {
1589 		free(msg, M_MQUEUEDATA);
1590 		msg = NULL;
1591 	} else {
1592 		msg->msg_size = msg_size;
1593 		msg->msg_prio = msg_prio;
1594 	}
1595 	return (msg);
1596 }
1597 
1598 /*
1599  * Save a message to user space
1600  */
1601 static int
1602 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1603 {
1604 	int error;
1605 
1606 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1607 		msg->msg_size);
1608 	if (error == 0 && msg_prio != NULL)
1609 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1610 	return (error);
1611 }
1612 
1613 /*
1614  * Free a message's memory
1615  */
1616 static __inline void
1617 mqueue_freemsg(struct mqueue_msg *msg)
1618 {
1619 	free(msg, M_MQUEUEDATA);
1620 }
1621 
1622 /*
1623  * Send a message. if waitok is false, thread will not be
1624  * blocked if there is no data in queue, otherwise, absolute
1625  * time will be checked.
1626  */
1627 int
1628 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1629 	size_t msg_len, unsigned msg_prio, int waitok,
1630 	const struct timespec *abs_timeout)
1631 {
1632 	struct mqueue_msg *msg;
1633 	struct timespec ts, ts2;
1634 	struct timeval tv;
1635 	int error;
1636 
1637 	if (msg_prio >= MQ_PRIO_MAX)
1638 		return (EINVAL);
1639 	if (msg_len > mq->mq_msgsize)
1640 		return (EMSGSIZE);
1641 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1642 	if (msg == NULL)
1643 		return (EFAULT);
1644 
1645 	/* O_NONBLOCK case */
1646 	if (!waitok) {
1647 		error = _mqueue_send(mq, msg, -1);
1648 		if (error)
1649 			goto bad;
1650 		return (0);
1651 	}
1652 
1653 	/* we allow a null timeout (wait forever) */
1654 	if (abs_timeout == NULL) {
1655 		error = _mqueue_send(mq, msg, 0);
1656 		if (error)
1657 			goto bad;
1658 		return (0);
1659 	}
1660 
1661 	/* send it before checking time */
1662 	error = _mqueue_send(mq, msg, -1);
1663 	if (error == 0)
1664 		return (0);
1665 
1666 	if (error != EAGAIN)
1667 		goto bad;
1668 
1669 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1670 		error = EINVAL;
1671 		goto bad;
1672 	}
1673 	for (;;) {
1674 		ts2 = *abs_timeout;
1675 		getnanotime(&ts);
1676 		timespecsub(&ts2, &ts);
1677 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1678 			error = ETIMEDOUT;
1679 			break;
1680 		}
1681 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1682 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1683 		if (error != ETIMEDOUT)
1684 			break;
1685 	}
1686 	if (error == 0)
1687 		return (0);
1688 bad:
1689 	mqueue_freemsg(msg);
1690 	return (error);
1691 }
1692 
1693 /*
1694  * Common routine to send a message
1695  */
1696 static int
1697 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1698 {
1699 	struct mqueue_msg *msg2;
1700 	int error = 0;
1701 
1702 	mtx_lock(&mq->mq_mutex);
1703 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1704 		if (timo < 0) {
1705 			mtx_unlock(&mq->mq_mutex);
1706 			return (EAGAIN);
1707 		}
1708 		mq->mq_senders++;
1709 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1710 			    PCATCH, "mqsend", timo);
1711 		mq->mq_senders--;
1712 		if (error == EAGAIN)
1713 			error = ETIMEDOUT;
1714 	}
1715 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1716 		mtx_unlock(&mq->mq_mutex);
1717 		return (error);
1718 	}
1719 	error = 0;
1720 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1721 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1722 	} else {
1723 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1724 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1725 		} else {
1726 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1727 				if (msg2->msg_prio < msg->msg_prio)
1728 					break;
1729 			}
1730 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1731 		}
1732 	}
1733 	mq->mq_curmsgs++;
1734 	mq->mq_totalbytes += msg->msg_size;
1735 	if (mq->mq_receivers)
1736 		wakeup_one(&mq->mq_receivers);
1737 	else if (mq->mq_notifier != NULL)
1738 		mqueue_send_notification(mq);
1739 	if (mq->mq_flags & MQ_RSEL) {
1740 		mq->mq_flags &= ~MQ_RSEL;
1741 		selwakeup(&mq->mq_rsel);
1742 	}
1743 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1744 	mtx_unlock(&mq->mq_mutex);
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Send realtime a signal to process which registered itself
1750  * successfully by mq_notify.
1751  */
1752 static void
1753 mqueue_send_notification(struct mqueue *mq)
1754 {
1755 	struct mqueue_notifier *nt;
1756 	struct thread *td;
1757 	struct proc *p;
1758 	int error;
1759 
1760 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1761 	nt = mq->mq_notifier;
1762 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1763 		p = nt->nt_proc;
1764 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1765 		if (error) {
1766 			mq->mq_notifier = NULL;
1767 			return;
1768 		}
1769 		if (!KSI_ONQ(&nt->nt_ksi)) {
1770 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1771 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1772 		}
1773 		PROC_UNLOCK(p);
1774 	}
1775 	mq->mq_notifier = NULL;
1776 }
1777 
1778 /*
1779  * Get a message. if waitok is false, thread will not be
1780  * blocked if there is no data in queue, otherwise, absolute
1781  * time will be checked.
1782  */
1783 int
1784 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1785 	size_t msg_len, unsigned *msg_prio, int waitok,
1786 	const struct timespec *abs_timeout)
1787 {
1788 	struct mqueue_msg *msg;
1789 	struct timespec ts, ts2;
1790 	struct timeval tv;
1791 	int error;
1792 
1793 	if (msg_len < mq->mq_msgsize)
1794 		return (EMSGSIZE);
1795 
1796 	/* O_NONBLOCK case */
1797 	if (!waitok) {
1798 		error = _mqueue_recv(mq, &msg, -1);
1799 		if (error)
1800 			return (error);
1801 		goto received;
1802 	}
1803 
1804 	/* we allow a null timeout (wait forever). */
1805 	if (abs_timeout == NULL) {
1806 		error = _mqueue_recv(mq, &msg, 0);
1807 		if (error)
1808 			return (error);
1809 		goto received;
1810 	}
1811 
1812 	/* try to get a message before checking time */
1813 	error = _mqueue_recv(mq, &msg, -1);
1814 	if (error == 0)
1815 		goto received;
1816 
1817 	if (error != EAGAIN)
1818 		return (error);
1819 
1820 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1821 		error = EINVAL;
1822 		return (error);
1823 	}
1824 
1825 	for (;;) {
1826 		ts2 = *abs_timeout;
1827 		getnanotime(&ts);
1828 		timespecsub(&ts2, &ts);
1829 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1830 			error = ETIMEDOUT;
1831 			return (error);
1832 		}
1833 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1834 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1835 		if (error == 0)
1836 			break;
1837 		if (error != ETIMEDOUT)
1838 			return (error);
1839 	}
1840 
1841 received:
1842 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1843 	if (error == 0) {
1844 		curthread->td_retval[0] = msg->msg_size;
1845 		curthread->td_retval[1] = 0;
1846 	}
1847 	mqueue_freemsg(msg);
1848 	return (error);
1849 }
1850 
1851 /*
1852  * Common routine to receive a message
1853  */
1854 static int
1855 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1856 {
1857 	int error = 0;
1858 
1859 	mtx_lock(&mq->mq_mutex);
1860 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1861 		if (timo < 0) {
1862 			mtx_unlock(&mq->mq_mutex);
1863 			return (EAGAIN);
1864 		}
1865 		mq->mq_receivers++;
1866 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1867 			    PCATCH, "mqrecv", timo);
1868 		mq->mq_receivers--;
1869 		if (error == EAGAIN)
1870 			error = ETIMEDOUT;
1871 	}
1872 	if (*msg != NULL) {
1873 		error = 0;
1874 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1875 		mq->mq_curmsgs--;
1876 		mq->mq_totalbytes -= (*msg)->msg_size;
1877 		if (mq->mq_senders)
1878 			wakeup_one(&mq->mq_senders);
1879 		if (mq->mq_flags & MQ_WSEL) {
1880 			mq->mq_flags &= ~MQ_WSEL;
1881 			selwakeup(&mq->mq_wsel);
1882 		}
1883 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1884 	}
1885 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1886 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1887 		mqueue_send_notification(mq);
1888 	}
1889 	mtx_unlock(&mq->mq_mutex);
1890 	return (error);
1891 }
1892 
1893 static __inline struct mqueue_notifier *
1894 notifier_alloc(void)
1895 {
1896 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1897 }
1898 
1899 static __inline void
1900 notifier_free(struct mqueue_notifier *p)
1901 {
1902 	uma_zfree(mqnoti_zone, p);
1903 }
1904 
1905 static struct mqueue_notifier *
1906 notifier_search(struct proc *p, int fd)
1907 {
1908 	struct mqueue_notifier *nt;
1909 
1910 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1911 		if (nt->nt_ksi.ksi_mqd == fd)
1912 			break;
1913 	}
1914 	return (nt);
1915 }
1916 
1917 static __inline void
1918 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1919 {
1920 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1921 }
1922 
1923 static __inline void
1924 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1925 {
1926 	LIST_REMOVE(nt, nt_link);
1927 	notifier_free(nt);
1928 }
1929 
1930 static void
1931 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1932 {
1933 	struct mqueue_notifier *nt;
1934 
1935 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1936 	PROC_LOCK(p);
1937 	nt = notifier_search(p, fd);
1938 	if (nt != NULL) {
1939 		if (mq->mq_notifier == nt)
1940 			mq->mq_notifier = NULL;
1941 		sigqueue_take(&nt->nt_ksi);
1942 		notifier_delete(p, nt);
1943 	}
1944 	PROC_UNLOCK(p);
1945 }
1946 
1947 static int
1948 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
1949     const struct mq_attr *attr)
1950 {
1951 	char path[MQFS_NAMELEN + 1];
1952 	struct mqfs_node *pn;
1953 	struct filedesc *fdp;
1954 	struct file *fp;
1955 	struct mqueue *mq;
1956 	int fd, error, len, cmode;
1957 
1958 	fdp = td->td_proc->p_fd;
1959 	cmode = (((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1960 	mq = NULL;
1961 	if ((flags & O_CREAT) != 0 && attr != NULL) {
1962 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
1963 			return (EINVAL);
1964 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
1965 			return (EINVAL);
1966 	}
1967 
1968 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
1969         if (error)
1970 		return (error);
1971 
1972 	/*
1973 	 * The first character of name must be a slash  (/) character
1974 	 * and the remaining characters of name cannot include any slash
1975 	 * characters.
1976 	 */
1977 	len = strlen(path);
1978 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
1979 		return (EINVAL);
1980 
1981 	error = falloc(td, &fp, &fd, O_CLOEXEC);
1982 	if (error)
1983 		return (error);
1984 
1985 	sx_xlock(&mqfs_data.mi_lock);
1986 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1987 	if (pn == NULL) {
1988 		if (!(flags & O_CREAT)) {
1989 			error = ENOENT;
1990 		} else {
1991 			mq = mqueue_alloc(attr);
1992 			if (mq == NULL) {
1993 				error = ENFILE;
1994 			} else {
1995 				pn = mqfs_create_file(mqfs_data.mi_root,
1996 				         path + 1, len - 1, td->td_ucred,
1997 					 cmode);
1998 				if (pn == NULL) {
1999 					error = ENOSPC;
2000 					mqueue_free(mq);
2001 				}
2002 			}
2003 		}
2004 
2005 		if (error == 0) {
2006 			pn->mn_data = mq;
2007 		}
2008 	} else {
2009 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2010 			error = EEXIST;
2011 		} else {
2012 			accmode_t accmode = 0;
2013 
2014 			if (flags & FREAD)
2015 				accmode |= VREAD;
2016 			if (flags & FWRITE)
2017 				accmode |= VWRITE;
2018 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2019 				    pn->mn_gid, accmode, td->td_ucred, NULL);
2020 		}
2021 	}
2022 
2023 	if (error) {
2024 		sx_xunlock(&mqfs_data.mi_lock);
2025 		fdclose(fdp, fp, fd, td);
2026 		fdrop(fp, td);
2027 		return (error);
2028 	}
2029 
2030 	mqnode_addref(pn);
2031 	sx_xunlock(&mqfs_data.mi_lock);
2032 
2033 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2034 	    &mqueueops);
2035 
2036 	td->td_retval[0] = fd;
2037 	fdrop(fp, td);
2038 	return (0);
2039 }
2040 
2041 /*
2042  * Syscall to open a message queue.
2043  */
2044 int
2045 sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
2046 {
2047 	struct mq_attr attr;
2048 	int flags, error;
2049 
2050 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2051 		return (EINVAL);
2052 	flags = FFLAGS(uap->flags);
2053 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2054 		error = copyin(uap->attr, &attr, sizeof(attr));
2055 		if (error)
2056 			return (error);
2057 	}
2058 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2059 	    uap->attr != NULL ? &attr : NULL));
2060 }
2061 
2062 /*
2063  * Syscall to unlink a message queue.
2064  */
2065 int
2066 sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2067 {
2068 	char path[MQFS_NAMELEN+1];
2069 	struct mqfs_node *pn;
2070 	int error, len;
2071 
2072 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2073         if (error)
2074 		return (error);
2075 
2076 	len = strlen(path);
2077 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2078 		return (EINVAL);
2079 
2080 	sx_xlock(&mqfs_data.mi_lock);
2081 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2082 	if (pn != NULL)
2083 		error = do_unlink(pn, td->td_ucred);
2084 	else
2085 		error = ENOENT;
2086 	sx_xunlock(&mqfs_data.mi_lock);
2087 	return (error);
2088 }
2089 
2090 typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **);
2091 
2092 /*
2093  * Get message queue by giving file slot
2094  */
2095 static int
2096 _getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func,
2097        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2098 {
2099 	struct mqfs_node *pn;
2100 	int error;
2101 
2102 	error = func(td, fd, rightsp, fpp);
2103 	if (error)
2104 		return (error);
2105 	if (&mqueueops != (*fpp)->f_ops) {
2106 		fdrop(*fpp, td);
2107 		return (EBADF);
2108 	}
2109 	pn = (*fpp)->f_data;
2110 	if (ppn)
2111 		*ppn = pn;
2112 	if (pmq)
2113 		*pmq = pn->mn_data;
2114 	return (0);
2115 }
2116 
2117 static __inline int
2118 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2119 	struct mqueue **pmq)
2120 {
2121 	cap_rights_t rights;
2122 
2123 	return _getmq(td, fd, cap_rights_init(&rights, CAP_EVENT), fget,
2124 	    fpp, ppn, pmq);
2125 }
2126 
2127 static __inline int
2128 getmq_read(struct thread *td, int fd, struct file **fpp,
2129 	 struct mqfs_node **ppn, struct mqueue **pmq)
2130 {
2131 	cap_rights_t rights;
2132 
2133 	return _getmq(td, fd, cap_rights_init(&rights, CAP_READ), fget_read,
2134 	    fpp, ppn, pmq);
2135 }
2136 
2137 static __inline int
2138 getmq_write(struct thread *td, int fd, struct file **fpp,
2139 	struct mqfs_node **ppn, struct mqueue **pmq)
2140 {
2141 	cap_rights_t rights;
2142 
2143 	return _getmq(td, fd, cap_rights_init(&rights, CAP_WRITE), fget_write,
2144 	    fpp, ppn, pmq);
2145 }
2146 
2147 static int
2148 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2149     struct mq_attr *oattr)
2150 {
2151 	struct mqueue *mq;
2152 	struct file *fp;
2153 	u_int oflag, flag;
2154 	int error;
2155 
2156 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2157 		return (EINVAL);
2158 	error = getmq(td, mqd, &fp, NULL, &mq);
2159 	if (error)
2160 		return (error);
2161 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2162 	oattr->mq_msgsize = mq->mq_msgsize;
2163 	oattr->mq_curmsgs = mq->mq_curmsgs;
2164 	if (attr != NULL) {
2165 		do {
2166 			oflag = flag = fp->f_flag;
2167 			flag &= ~O_NONBLOCK;
2168 			flag |= (attr->mq_flags & O_NONBLOCK);
2169 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2170 	} else
2171 		oflag = fp->f_flag;
2172 	oattr->mq_flags = (O_NONBLOCK & oflag);
2173 	fdrop(fp, td);
2174 	return (error);
2175 }
2176 
2177 int
2178 sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2179 {
2180 	struct mq_attr attr, oattr;
2181 	int error;
2182 
2183 	if (uap->attr != NULL) {
2184 		error = copyin(uap->attr, &attr, sizeof(attr));
2185 		if (error != 0)
2186 			return (error);
2187 	}
2188 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2189 	    &oattr);
2190 	if (error != 0)
2191 		return (error);
2192 	if (uap->oattr != NULL)
2193 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2194 	return (error);
2195 }
2196 
2197 int
2198 sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2199 {
2200 	struct mqueue *mq;
2201 	struct file *fp;
2202 	struct timespec *abs_timeout, ets;
2203 	int error;
2204 	int waitok;
2205 
2206 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2207 	if (error)
2208 		return (error);
2209 	if (uap->abs_timeout != NULL) {
2210 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2211 		if (error != 0)
2212 			return (error);
2213 		abs_timeout = &ets;
2214 	} else
2215 		abs_timeout = NULL;
2216 	waitok = !(fp->f_flag & O_NONBLOCK);
2217 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2218 		uap->msg_prio, waitok, abs_timeout);
2219 	fdrop(fp, td);
2220 	return (error);
2221 }
2222 
2223 int
2224 sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2225 {
2226 	struct mqueue *mq;
2227 	struct file *fp;
2228 	struct timespec *abs_timeout, ets;
2229 	int error, waitok;
2230 
2231 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2232 	if (error)
2233 		return (error);
2234 	if (uap->abs_timeout != NULL) {
2235 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2236 		if (error != 0)
2237 			return (error);
2238 		abs_timeout = &ets;
2239 	} else
2240 		abs_timeout = NULL;
2241 	waitok = !(fp->f_flag & O_NONBLOCK);
2242 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2243 		uap->msg_prio, waitok, abs_timeout);
2244 	fdrop(fp, td);
2245 	return (error);
2246 }
2247 
2248 static int
2249 kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev)
2250 {
2251 #ifdef CAPABILITIES
2252 	cap_rights_t rights;
2253 #endif
2254 	struct filedesc *fdp;
2255 	struct proc *p;
2256 	struct mqueue *mq;
2257 	struct file *fp, *fp2;
2258 	struct mqueue_notifier *nt, *newnt = NULL;
2259 	int error;
2260 
2261 	if (sigev != NULL) {
2262 		if (sigev->sigev_notify != SIGEV_SIGNAL &&
2263 		    sigev->sigev_notify != SIGEV_THREAD_ID &&
2264 		    sigev->sigev_notify != SIGEV_NONE)
2265 			return (EINVAL);
2266 		if ((sigev->sigev_notify == SIGEV_SIGNAL ||
2267 		    sigev->sigev_notify == SIGEV_THREAD_ID) &&
2268 		    !_SIG_VALID(sigev->sigev_signo))
2269 			return (EINVAL);
2270 	}
2271 	p = td->td_proc;
2272 	fdp = td->td_proc->p_fd;
2273 	error = getmq(td, mqd, &fp, NULL, &mq);
2274 	if (error)
2275 		return (error);
2276 again:
2277 	FILEDESC_SLOCK(fdp);
2278 	fp2 = fget_locked(fdp, mqd);
2279 	if (fp2 == NULL) {
2280 		FILEDESC_SUNLOCK(fdp);
2281 		error = EBADF;
2282 		goto out;
2283 	}
2284 #ifdef CAPABILITIES
2285 	error = cap_check(cap_rights(fdp, mqd),
2286 	    cap_rights_init(&rights, CAP_EVENT));
2287 	if (error) {
2288 		FILEDESC_SUNLOCK(fdp);
2289 		goto out;
2290 	}
2291 #endif
2292 	if (fp2 != fp) {
2293 		FILEDESC_SUNLOCK(fdp);
2294 		error = EBADF;
2295 		goto out;
2296 	}
2297 	mtx_lock(&mq->mq_mutex);
2298 	FILEDESC_SUNLOCK(fdp);
2299 	if (sigev != NULL) {
2300 		if (mq->mq_notifier != NULL) {
2301 			error = EBUSY;
2302 		} else {
2303 			PROC_LOCK(p);
2304 			nt = notifier_search(p, mqd);
2305 			if (nt == NULL) {
2306 				if (newnt == NULL) {
2307 					PROC_UNLOCK(p);
2308 					mtx_unlock(&mq->mq_mutex);
2309 					newnt = notifier_alloc();
2310 					goto again;
2311 				}
2312 			}
2313 
2314 			if (nt != NULL) {
2315 				sigqueue_take(&nt->nt_ksi);
2316 				if (newnt != NULL) {
2317 					notifier_free(newnt);
2318 					newnt = NULL;
2319 				}
2320 			} else {
2321 				nt = newnt;
2322 				newnt = NULL;
2323 				ksiginfo_init(&nt->nt_ksi);
2324 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2325 				nt->nt_ksi.ksi_code = SI_MESGQ;
2326 				nt->nt_proc = p;
2327 				nt->nt_ksi.ksi_mqd = mqd;
2328 				notifier_insert(p, nt);
2329 			}
2330 			nt->nt_sigev = *sigev;
2331 			mq->mq_notifier = nt;
2332 			PROC_UNLOCK(p);
2333 			/*
2334 			 * if there is no receivers and message queue
2335 			 * is not empty, we should send notification
2336 			 * as soon as possible.
2337 			 */
2338 			if (mq->mq_receivers == 0 &&
2339 			    !TAILQ_EMPTY(&mq->mq_msgq))
2340 				mqueue_send_notification(mq);
2341 		}
2342 	} else {
2343 		notifier_remove(p, mq, mqd);
2344 	}
2345 	mtx_unlock(&mq->mq_mutex);
2346 
2347 out:
2348 	fdrop(fp, td);
2349 	if (newnt != NULL)
2350 		notifier_free(newnt);
2351 	return (error);
2352 }
2353 
2354 int
2355 sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2356 {
2357 	struct sigevent ev, *evp;
2358 	int error;
2359 
2360 	if (uap->sigev == NULL) {
2361 		evp = NULL;
2362 	} else {
2363 		error = copyin(uap->sigev, &ev, sizeof(ev));
2364 		if (error != 0)
2365 			return (error);
2366 		evp = &ev;
2367 	}
2368 	return (kern_kmq_notify(td, uap->mqd, evp));
2369 }
2370 
2371 static void
2372 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2373 {
2374 	struct filedesc *fdp;
2375 	struct mqueue *mq;
2376 
2377 	fdp = td->td_proc->p_fd;
2378 	FILEDESC_LOCK_ASSERT(fdp);
2379 
2380 	if (fp->f_ops == &mqueueops) {
2381 		mq = FPTOMQ(fp);
2382 		mtx_lock(&mq->mq_mutex);
2383 		notifier_remove(td->td_proc, mq, fd);
2384 
2385 		/* have to wakeup thread in same process */
2386 		if (mq->mq_flags & MQ_RSEL) {
2387 			mq->mq_flags &= ~MQ_RSEL;
2388 			selwakeup(&mq->mq_rsel);
2389 		}
2390 		if (mq->mq_flags & MQ_WSEL) {
2391 			mq->mq_flags &= ~MQ_WSEL;
2392 			selwakeup(&mq->mq_wsel);
2393 		}
2394 		mtx_unlock(&mq->mq_mutex);
2395 	}
2396 }
2397 
2398 static void
2399 mq_proc_exit(void *arg __unused, struct proc *p)
2400 {
2401 	struct filedesc *fdp;
2402 	struct file *fp;
2403 	struct mqueue *mq;
2404 	int i;
2405 
2406 	fdp = p->p_fd;
2407 	FILEDESC_SLOCK(fdp);
2408 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2409 		fp = fget_locked(fdp, i);
2410 		if (fp != NULL && fp->f_ops == &mqueueops) {
2411 			mq = FPTOMQ(fp);
2412 			mtx_lock(&mq->mq_mutex);
2413 			notifier_remove(p, FPTOMQ(fp), i);
2414 			mtx_unlock(&mq->mq_mutex);
2415 		}
2416 	}
2417 	FILEDESC_SUNLOCK(fdp);
2418 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2419 }
2420 
2421 static int
2422 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2423 	struct thread *td)
2424 {
2425 	struct mqueue *mq = FPTOMQ(fp);
2426 	int revents = 0;
2427 
2428 	mtx_lock(&mq->mq_mutex);
2429 	if (events & (POLLIN | POLLRDNORM)) {
2430 		if (mq->mq_curmsgs) {
2431 			revents |= events & (POLLIN | POLLRDNORM);
2432 		} else {
2433 			mq->mq_flags |= MQ_RSEL;
2434 			selrecord(td, &mq->mq_rsel);
2435  		}
2436 	}
2437 	if (events & POLLOUT) {
2438 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2439 			revents |= POLLOUT;
2440 		else {
2441 			mq->mq_flags |= MQ_WSEL;
2442 			selrecord(td, &mq->mq_wsel);
2443 		}
2444 	}
2445 	mtx_unlock(&mq->mq_mutex);
2446 	return (revents);
2447 }
2448 
2449 static int
2450 mqf_close(struct file *fp, struct thread *td)
2451 {
2452 	struct mqfs_node *pn;
2453 
2454 	fp->f_ops = &badfileops;
2455 	pn = fp->f_data;
2456 	fp->f_data = NULL;
2457 	sx_xlock(&mqfs_data.mi_lock);
2458 	mqnode_release(pn);
2459 	sx_xunlock(&mqfs_data.mi_lock);
2460 	return (0);
2461 }
2462 
2463 static int
2464 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2465 	struct thread *td)
2466 {
2467 	struct mqfs_node *pn = fp->f_data;
2468 
2469 	bzero(st, sizeof *st);
2470 	sx_xlock(&mqfs_data.mi_lock);
2471 	st->st_atim = pn->mn_atime;
2472 	st->st_mtim = pn->mn_mtime;
2473 	st->st_ctim = pn->mn_ctime;
2474 	st->st_birthtim = pn->mn_birth;
2475 	st->st_uid = pn->mn_uid;
2476 	st->st_gid = pn->mn_gid;
2477 	st->st_mode = S_IFIFO | pn->mn_mode;
2478 	sx_xunlock(&mqfs_data.mi_lock);
2479 	return (0);
2480 }
2481 
2482 static int
2483 mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2484     struct thread *td)
2485 {
2486 	struct mqfs_node *pn;
2487 	int error;
2488 
2489 	error = 0;
2490 	pn = fp->f_data;
2491 	sx_xlock(&mqfs_data.mi_lock);
2492 	error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
2493 	    active_cred, NULL);
2494 	if (error != 0)
2495 		goto out;
2496 	pn->mn_mode = mode & ACCESSPERMS;
2497 out:
2498 	sx_xunlock(&mqfs_data.mi_lock);
2499 	return (error);
2500 }
2501 
2502 static int
2503 mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2504     struct thread *td)
2505 {
2506 	struct mqfs_node *pn;
2507 	int error;
2508 
2509 	error = 0;
2510 	pn = fp->f_data;
2511 	sx_xlock(&mqfs_data.mi_lock);
2512 	if (uid == (uid_t)-1)
2513 		uid = pn->mn_uid;
2514 	if (gid == (gid_t)-1)
2515 		gid = pn->mn_gid;
2516 	if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
2517 	    (gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
2518 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
2519 		goto out;
2520 	pn->mn_uid = uid;
2521 	pn->mn_gid = gid;
2522 out:
2523 	sx_xunlock(&mqfs_data.mi_lock);
2524 	return (error);
2525 }
2526 
2527 static int
2528 mqf_kqfilter(struct file *fp, struct knote *kn)
2529 {
2530 	struct mqueue *mq = FPTOMQ(fp);
2531 	int error = 0;
2532 
2533 	if (kn->kn_filter == EVFILT_READ) {
2534 		kn->kn_fop = &mq_rfiltops;
2535 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2536 	} else if (kn->kn_filter == EVFILT_WRITE) {
2537 		kn->kn_fop = &mq_wfiltops;
2538 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2539 	} else
2540 		error = EINVAL;
2541 	return (error);
2542 }
2543 
2544 static void
2545 filt_mqdetach(struct knote *kn)
2546 {
2547 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2548 
2549 	if (kn->kn_filter == EVFILT_READ)
2550 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2551 	else if (kn->kn_filter == EVFILT_WRITE)
2552 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2553 	else
2554 		panic("filt_mqdetach");
2555 }
2556 
2557 static int
2558 filt_mqread(struct knote *kn, long hint)
2559 {
2560 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2561 
2562 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2563 	return (mq->mq_curmsgs != 0);
2564 }
2565 
2566 static int
2567 filt_mqwrite(struct knote *kn, long hint)
2568 {
2569 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2570 
2571 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2572 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2573 }
2574 
2575 static int
2576 mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2577 {
2578 
2579 	kif->kf_type = KF_TYPE_MQUEUE;
2580 	return (0);
2581 }
2582 
2583 static struct fileops mqueueops = {
2584 	.fo_read		= invfo_rdwr,
2585 	.fo_write		= invfo_rdwr,
2586 	.fo_truncate		= invfo_truncate,
2587 	.fo_ioctl		= invfo_ioctl,
2588 	.fo_poll		= mqf_poll,
2589 	.fo_kqfilter		= mqf_kqfilter,
2590 	.fo_stat		= mqf_stat,
2591 	.fo_close		= mqf_close,
2592 	.fo_chmod		= mqf_chmod,
2593 	.fo_chown		= mqf_chown,
2594 	.fo_sendfile		= invfo_sendfile,
2595 	.fo_fill_kinfo		= mqf_fill_kinfo,
2596 };
2597 
2598 static struct vop_vector mqfs_vnodeops = {
2599 	.vop_default 		= &default_vnodeops,
2600 	.vop_access		= mqfs_access,
2601 	.vop_cachedlookup	= mqfs_lookup,
2602 	.vop_lookup		= vfs_cache_lookup,
2603 	.vop_reclaim		= mqfs_reclaim,
2604 	.vop_create		= mqfs_create,
2605 	.vop_remove		= mqfs_remove,
2606 	.vop_inactive		= mqfs_inactive,
2607 	.vop_open		= mqfs_open,
2608 	.vop_close		= mqfs_close,
2609 	.vop_getattr		= mqfs_getattr,
2610 	.vop_setattr		= mqfs_setattr,
2611 	.vop_read		= mqfs_read,
2612 	.vop_write		= VOP_EOPNOTSUPP,
2613 	.vop_readdir		= mqfs_readdir,
2614 	.vop_mkdir		= VOP_EOPNOTSUPP,
2615 	.vop_rmdir		= VOP_EOPNOTSUPP
2616 };
2617 
2618 static struct vfsops mqfs_vfsops = {
2619 	.vfs_init 		= mqfs_init,
2620 	.vfs_uninit		= mqfs_uninit,
2621 	.vfs_mount		= mqfs_mount,
2622 	.vfs_unmount		= mqfs_unmount,
2623 	.vfs_root		= mqfs_root,
2624 	.vfs_statfs		= mqfs_statfs,
2625 };
2626 
2627 static struct vfsconf mqueuefs_vfsconf = {
2628 	.vfc_version = VFS_VERSION,
2629 	.vfc_name = "mqueuefs",
2630 	.vfc_vfsops = &mqfs_vfsops,
2631 	.vfc_typenum = -1,
2632 	.vfc_flags = VFCF_SYNTHETIC
2633 };
2634 
2635 static struct syscall_helper_data mq_syscalls[] = {
2636 	SYSCALL_INIT_HELPER(kmq_open),
2637 	SYSCALL_INIT_HELPER(kmq_setattr),
2638 	SYSCALL_INIT_HELPER(kmq_timedsend),
2639 	SYSCALL_INIT_HELPER(kmq_timedreceive),
2640 	SYSCALL_INIT_HELPER(kmq_notify),
2641 	SYSCALL_INIT_HELPER(kmq_unlink),
2642 	SYSCALL_INIT_LAST
2643 };
2644 
2645 #ifdef COMPAT_FREEBSD32
2646 #include <compat/freebsd32/freebsd32.h>
2647 #include <compat/freebsd32/freebsd32_proto.h>
2648 #include <compat/freebsd32/freebsd32_signal.h>
2649 #include <compat/freebsd32/freebsd32_syscall.h>
2650 #include <compat/freebsd32/freebsd32_util.h>
2651 
2652 static void
2653 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2654 {
2655 
2656 	to->mq_flags = from->mq_flags;
2657 	to->mq_maxmsg = from->mq_maxmsg;
2658 	to->mq_msgsize = from->mq_msgsize;
2659 	to->mq_curmsgs = from->mq_curmsgs;
2660 }
2661 
2662 static void
2663 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2664 {
2665 
2666 	to->mq_flags = from->mq_flags;
2667 	to->mq_maxmsg = from->mq_maxmsg;
2668 	to->mq_msgsize = from->mq_msgsize;
2669 	to->mq_curmsgs = from->mq_curmsgs;
2670 }
2671 
2672 int
2673 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2674 {
2675 	struct mq_attr attr;
2676 	struct mq_attr32 attr32;
2677 	int flags, error;
2678 
2679 	if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
2680 		return (EINVAL);
2681 	flags = FFLAGS(uap->flags);
2682 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2683 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2684 		if (error)
2685 			return (error);
2686 		mq_attr_from32(&attr32, &attr);
2687 	}
2688 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2689 	    uap->attr != NULL ? &attr : NULL));
2690 }
2691 
2692 int
2693 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2694 {
2695 	struct mq_attr attr, oattr;
2696 	struct mq_attr32 attr32, oattr32;
2697 	int error;
2698 
2699 	if (uap->attr != NULL) {
2700 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2701 		if (error != 0)
2702 			return (error);
2703 		mq_attr_from32(&attr32, &attr);
2704 	}
2705 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2706 	    &oattr);
2707 	if (error != 0)
2708 		return (error);
2709 	if (uap->oattr != NULL) {
2710 		mq_attr_to32(&oattr, &oattr32);
2711 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2712 	}
2713 	return (error);
2714 }
2715 
2716 int
2717 freebsd32_kmq_timedsend(struct thread *td,
2718     struct freebsd32_kmq_timedsend_args *uap)
2719 {
2720 	struct mqueue *mq;
2721 	struct file *fp;
2722 	struct timespec32 ets32;
2723 	struct timespec *abs_timeout, ets;
2724 	int error;
2725 	int waitok;
2726 
2727 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2728 	if (error)
2729 		return (error);
2730 	if (uap->abs_timeout != NULL) {
2731 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2732 		if (error != 0)
2733 			return (error);
2734 		CP(ets32, ets, tv_sec);
2735 		CP(ets32, ets, tv_nsec);
2736 		abs_timeout = &ets;
2737 	} else
2738 		abs_timeout = NULL;
2739 	waitok = !(fp->f_flag & O_NONBLOCK);
2740 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2741 		uap->msg_prio, waitok, abs_timeout);
2742 	fdrop(fp, td);
2743 	return (error);
2744 }
2745 
2746 int
2747 freebsd32_kmq_timedreceive(struct thread *td,
2748     struct freebsd32_kmq_timedreceive_args *uap)
2749 {
2750 	struct mqueue *mq;
2751 	struct file *fp;
2752 	struct timespec32 ets32;
2753 	struct timespec *abs_timeout, ets;
2754 	int error, waitok;
2755 
2756 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2757 	if (error)
2758 		return (error);
2759 	if (uap->abs_timeout != NULL) {
2760 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2761 		if (error != 0)
2762 			return (error);
2763 		CP(ets32, ets, tv_sec);
2764 		CP(ets32, ets, tv_nsec);
2765 		abs_timeout = &ets;
2766 	} else
2767 		abs_timeout = NULL;
2768 	waitok = !(fp->f_flag & O_NONBLOCK);
2769 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2770 		uap->msg_prio, waitok, abs_timeout);
2771 	fdrop(fp, td);
2772 	return (error);
2773 }
2774 
2775 int
2776 freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap)
2777 {
2778 	struct sigevent ev, *evp;
2779 	struct sigevent32 ev32;
2780 	int error;
2781 
2782 	if (uap->sigev == NULL) {
2783 		evp = NULL;
2784 	} else {
2785 		error = copyin(uap->sigev, &ev32, sizeof(ev32));
2786 		if (error != 0)
2787 			return (error);
2788 		error = convert_sigevent32(&ev32, &ev);
2789 		if (error != 0)
2790 			return (error);
2791 		evp = &ev;
2792 	}
2793 	return (kern_kmq_notify(td, uap->mqd, evp));
2794 }
2795 
2796 static struct syscall_helper_data mq32_syscalls[] = {
2797 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2798 	SYSCALL32_INIT_HELPER(freebsd32_kmq_setattr),
2799 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedsend),
2800 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedreceive),
2801 	SYSCALL32_INIT_HELPER(freebsd32_kmq_notify),
2802 	SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
2803 	SYSCALL_INIT_LAST
2804 };
2805 #endif
2806 
2807 static int
2808 mqinit(void)
2809 {
2810 	int error;
2811 
2812 	error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD);
2813 	if (error != 0)
2814 		return (error);
2815 #ifdef COMPAT_FREEBSD32
2816 	error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD);
2817 	if (error != 0)
2818 		return (error);
2819 #endif
2820 	return (0);
2821 }
2822 
2823 static int
2824 mqunload(void)
2825 {
2826 
2827 #ifdef COMPAT_FREEBSD32
2828 	syscall32_helper_unregister(mq32_syscalls);
2829 #endif
2830 	syscall_helper_unregister(mq_syscalls);
2831 	return (0);
2832 }
2833 
2834 static int
2835 mq_modload(struct module *module, int cmd, void *arg)
2836 {
2837 	int error = 0;
2838 
2839 	error = vfs_modevent(module, cmd, arg);
2840 	if (error != 0)
2841 		return (error);
2842 
2843 	switch (cmd) {
2844 	case MOD_LOAD:
2845 		error = mqinit();
2846 		if (error != 0)
2847 			mqunload();
2848 		break;
2849 	case MOD_UNLOAD:
2850 		error = mqunload();
2851 		break;
2852 	default:
2853 		break;
2854 	}
2855 	return (error);
2856 }
2857 
2858 static moduledata_t mqueuefs_mod = {
2859 	"mqueuefs",
2860 	mq_modload,
2861 	&mqueuefs_vfsconf
2862 };
2863 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2864 MODULE_VERSION(mqueuefs, 1);
2865