xref: /freebsd/sys/kern/uipc_mqueue.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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_compat.h"
49 
50 #include <sys/param.h>
51 #include <sys/kernel.h>
52 #include <sys/systm.h>
53 #include <sys/limits.h>
54 #include <sys/buf.h>
55 #include <sys/dirent.h>
56 #include <sys/event.h>
57 #include <sys/eventhandler.h>
58 #include <sys/fcntl.h>
59 #include <sys/file.h>
60 #include <sys/filedesc.h>
61 #include <sys/lock.h>
62 #include <sys/malloc.h>
63 #include <sys/module.h>
64 #include <sys/mount.h>
65 #include <sys/mqueue.h>
66 #include <sys/mutex.h>
67 #include <sys/namei.h>
68 #include <sys/posix4.h>
69 #include <sys/poll.h>
70 #include <sys/priv.h>
71 #include <sys/proc.h>
72 #include <sys/queue.h>
73 #include <sys/sysproto.h>
74 #include <sys/stat.h>
75 #include <sys/syscall.h>
76 #include <sys/syscallsubr.h>
77 #include <sys/sysent.h>
78 #include <sys/sx.h>
79 #include <sys/sysctl.h>
80 #include <sys/taskqueue.h>
81 #include <sys/unistd.h>
82 #include <sys/vnode.h>
83 #include <machine/atomic.h>
84 
85 FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
86 
87 /*
88  * Limits and constants
89  */
90 #define	MQFS_NAMELEN		NAME_MAX
91 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
92 
93 /* node types */
94 typedef enum {
95 	mqfstype_none = 0,
96 	mqfstype_root,
97 	mqfstype_dir,
98 	mqfstype_this,
99 	mqfstype_parent,
100 	mqfstype_file,
101 	mqfstype_symlink,
102 } mqfs_type_t;
103 
104 struct mqfs_node;
105 
106 /*
107  * mqfs_info: describes a mqfs instance
108  */
109 struct mqfs_info {
110 	struct sx		mi_lock;
111 	struct mqfs_node	*mi_root;
112 	struct unrhdr		*mi_unrhdr;
113 };
114 
115 struct mqfs_vdata {
116 	LIST_ENTRY(mqfs_vdata)	mv_link;
117 	struct mqfs_node	*mv_node;
118 	struct vnode		*mv_vnode;
119 	struct task		mv_task;
120 };
121 
122 /*
123  * mqfs_node: describes a node (file or directory) within a mqfs
124  */
125 struct mqfs_node {
126 	char			mn_name[MQFS_NAMELEN+1];
127 	struct mqfs_info	*mn_info;
128 	struct mqfs_node	*mn_parent;
129 	LIST_HEAD(,mqfs_node)	mn_children;
130 	LIST_ENTRY(mqfs_node)	mn_sibling;
131 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
132 	int			mn_refcount;
133 	mqfs_type_t		mn_type;
134 	int			mn_deleted;
135 	uint32_t		mn_fileno;
136 	void			*mn_data;
137 	struct timespec		mn_birth;
138 	struct timespec		mn_ctime;
139 	struct timespec		mn_atime;
140 	struct timespec		mn_mtime;
141 	uid_t			mn_uid;
142 	gid_t			mn_gid;
143 	int			mn_mode;
144 };
145 
146 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
147 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
148 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
149 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
150 				(fp)->f_data)->mn_data))
151 
152 TAILQ_HEAD(msgq, mqueue_msg);
153 
154 struct mqueue;
155 
156 struct mqueue_notifier {
157 	LIST_ENTRY(mqueue_notifier)	nt_link;
158 	struct sigevent			nt_sigev;
159 	ksiginfo_t			nt_ksi;
160 	struct proc			*nt_proc;
161 };
162 
163 struct mqueue {
164 	struct mtx	mq_mutex;
165 	int		mq_flags;
166 	long		mq_maxmsg;
167 	long		mq_msgsize;
168 	long		mq_curmsgs;
169 	long		mq_totalbytes;
170 	struct msgq	mq_msgq;
171 	int		mq_receivers;
172 	int		mq_senders;
173 	struct selinfo	mq_rsel;
174 	struct selinfo	mq_wsel;
175 	struct mqueue_notifier	*mq_notifier;
176 };
177 
178 #define	MQ_RSEL		0x01
179 #define	MQ_WSEL		0x02
180 
181 struct mqueue_msg {
182 	TAILQ_ENTRY(mqueue_msg)	msg_link;
183 	unsigned int	msg_prio;
184 	unsigned int	msg_size;
185 	/* following real data... */
186 };
187 
188 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
189 	"POSIX real time message queue");
190 
191 static int	default_maxmsg  = 10;
192 static int	default_msgsize = 1024;
193 
194 static int	maxmsg = 100;
195 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
196     &maxmsg, 0, "Default maximum messages in queue");
197 static int	maxmsgsize = 16384;
198 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
199     &maxmsgsize, 0, "Default maximum message size");
200 static int	maxmq = 100;
201 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
202     &maxmq, 0, "maximum message queues");
203 static int	curmq = 0;
204 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
205     &curmq, 0, "current message queue number");
206 static int	unloadable = 0;
207 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
208 
209 static eventhandler_tag exit_tag;
210 
211 /* Only one instance per-system */
212 static struct mqfs_info		mqfs_data;
213 static uma_zone_t		mqnode_zone;
214 static uma_zone_t		mqueue_zone;
215 static uma_zone_t		mvdata_zone;
216 static uma_zone_t		mqnoti_zone;
217 static struct vop_vector	mqfs_vnodeops;
218 static struct fileops		mqueueops;
219 
220 /*
221  * Directory structure construction and manipulation
222  */
223 #ifdef notyet
224 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
225 	const char *name, int namelen, struct ucred *cred, int mode);
226 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
227 	const char *name, int namelen, struct ucred *cred, int mode);
228 #endif
229 
230 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
231 	const char *name, int namelen, struct ucred *cred, int mode);
232 static int	mqfs_destroy(struct mqfs_node *mn);
233 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
234 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
235 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
236 
237 /*
238  * Message queue construction and maniplation
239  */
240 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
241 static void	mqueue_free(struct mqueue *mq);
242 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
243 			size_t msg_len, unsigned msg_prio, int waitok,
244 			const struct timespec *abs_timeout);
245 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
246 			size_t msg_len, unsigned *msg_prio, int waitok,
247 			const struct timespec *abs_timeout);
248 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
249 			int timo);
250 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
251 			int timo);
252 static void	mqueue_send_notification(struct mqueue *mq);
253 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
254 static void	mq_proc_exit(void *arg, struct proc *p);
255 
256 /*
257  * kqueue filters
258  */
259 static void	filt_mqdetach(struct knote *kn);
260 static int	filt_mqread(struct knote *kn, long hint);
261 static int	filt_mqwrite(struct knote *kn, long hint);
262 
263 struct filterops mq_rfiltops = {
264 	.f_isfd = 1,
265 	.f_detach = filt_mqdetach,
266 	.f_event = filt_mqread,
267 };
268 struct filterops mq_wfiltops = {
269 	.f_isfd = 1,
270 	.f_detach = filt_mqdetach,
271 	.f_event = filt_mqwrite,
272 };
273 
274 /*
275  * Initialize fileno bitmap
276  */
277 static void
278 mqfs_fileno_init(struct mqfs_info *mi)
279 {
280 	struct unrhdr *up;
281 
282 	up = new_unrhdr(1, INT_MAX, NULL);
283 	mi->mi_unrhdr = up;
284 }
285 
286 /*
287  * Tear down fileno bitmap
288  */
289 static void
290 mqfs_fileno_uninit(struct mqfs_info *mi)
291 {
292 	struct unrhdr *up;
293 
294 	up = mi->mi_unrhdr;
295 	mi->mi_unrhdr = NULL;
296 	delete_unrhdr(up);
297 }
298 
299 /*
300  * Allocate a file number
301  */
302 static void
303 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
304 {
305 	/* make sure our parent has a file number */
306 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
307 		mqfs_fileno_alloc(mi, mn->mn_parent);
308 
309 	switch (mn->mn_type) {
310 	case mqfstype_root:
311 	case mqfstype_dir:
312 	case mqfstype_file:
313 	case mqfstype_symlink:
314 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
315 		break;
316 	case mqfstype_this:
317 		KASSERT(mn->mn_parent != NULL,
318 		    ("mqfstype_this node has no parent"));
319 		mn->mn_fileno = mn->mn_parent->mn_fileno;
320 		break;
321 	case mqfstype_parent:
322 		KASSERT(mn->mn_parent != NULL,
323 		    ("mqfstype_parent node has no parent"));
324 		if (mn->mn_parent == mi->mi_root) {
325 			mn->mn_fileno = mn->mn_parent->mn_fileno;
326 			break;
327 		}
328 		KASSERT(mn->mn_parent->mn_parent != NULL,
329 		    ("mqfstype_parent node has no grandparent"));
330 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
331 		break;
332 	default:
333 		KASSERT(0,
334 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
335 			mn->mn_type));
336 		break;
337 	}
338 }
339 
340 /*
341  * Release a file number
342  */
343 static void
344 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
345 {
346 	switch (mn->mn_type) {
347 	case mqfstype_root:
348 	case mqfstype_dir:
349 	case mqfstype_file:
350 	case mqfstype_symlink:
351 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
352 		break;
353 	case mqfstype_this:
354 	case mqfstype_parent:
355 		/* ignore these, as they don't "own" their file number */
356 		break;
357 	default:
358 		KASSERT(0,
359 		    ("mqfs_fileno_free() called for unknown type node: %d",
360 			mn->mn_type));
361 		break;
362 	}
363 }
364 
365 static __inline struct mqfs_node *
366 mqnode_alloc(void)
367 {
368 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
369 }
370 
371 static __inline void
372 mqnode_free(struct mqfs_node *node)
373 {
374 	uma_zfree(mqnode_zone, node);
375 }
376 
377 static __inline void
378 mqnode_addref(struct mqfs_node *node)
379 {
380 	atomic_fetchadd_int(&node->mn_refcount, 1);
381 }
382 
383 static __inline void
384 mqnode_release(struct mqfs_node *node)
385 {
386 	struct mqfs_info *mqfs;
387 	int old, exp;
388 
389 	mqfs = node->mn_info;
390 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
391 	if (node->mn_type == mqfstype_dir ||
392 	    node->mn_type == mqfstype_root)
393 		exp = 3; /* include . and .. */
394 	else
395 		exp = 1;
396 	if (old == exp) {
397 		int locked = sx_xlocked(&mqfs->mi_lock);
398 		if (!locked)
399 			sx_xlock(&mqfs->mi_lock);
400 		mqfs_destroy(node);
401 		if (!locked)
402 			sx_xunlock(&mqfs->mi_lock);
403 	}
404 }
405 
406 /*
407  * Add a node to a directory
408  */
409 static int
410 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
411 {
412 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
413 	KASSERT(parent->mn_info != NULL,
414 	    ("%s(): parent has no mn_info", __func__));
415 	KASSERT(parent->mn_type == mqfstype_dir ||
416 	    parent->mn_type == mqfstype_root,
417 	    ("%s(): parent is not a directory", __func__));
418 
419 	node->mn_info = parent->mn_info;
420 	node->mn_parent = parent;
421 	LIST_INIT(&node->mn_children);
422 	LIST_INIT(&node->mn_vnodes);
423 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
424 	mqnode_addref(parent);
425 	return (0);
426 }
427 
428 static struct mqfs_node *
429 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
430 	int nodetype)
431 {
432 	struct mqfs_node *node;
433 
434 	node = mqnode_alloc();
435 	strncpy(node->mn_name, name, namelen);
436 	node->mn_type = nodetype;
437 	node->mn_refcount = 1;
438 	vfs_timestamp(&node->mn_birth);
439 	node->mn_ctime = node->mn_atime = node->mn_mtime
440 		= node->mn_birth;
441 	node->mn_uid = cred->cr_uid;
442 	node->mn_gid = cred->cr_gid;
443 	node->mn_mode = mode;
444 	return (node);
445 }
446 
447 /*
448  * Create a file
449  */
450 static struct mqfs_node *
451 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
452 	struct ucred *cred, int mode)
453 {
454 	struct mqfs_node *node;
455 
456 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
457 	if (mqfs_add_node(parent, node) != 0) {
458 		mqnode_free(node);
459 		return (NULL);
460 	}
461 	return (node);
462 }
463 
464 /*
465  * Add . and .. to a directory
466  */
467 static int
468 mqfs_fixup_dir(struct mqfs_node *parent)
469 {
470 	struct mqfs_node *dir;
471 
472 	dir = mqnode_alloc();
473 	dir->mn_name[0] = '.';
474 	dir->mn_type = mqfstype_this;
475 	dir->mn_refcount = 1;
476 	if (mqfs_add_node(parent, dir) != 0) {
477 		mqnode_free(dir);
478 		return (-1);
479 	}
480 
481 	dir = mqnode_alloc();
482 	dir->mn_name[0] = dir->mn_name[1] = '.';
483 	dir->mn_type = mqfstype_parent;
484 	dir->mn_refcount = 1;
485 
486 	if (mqfs_add_node(parent, dir) != 0) {
487 		mqnode_free(dir);
488 		return (-1);
489 	}
490 
491 	return (0);
492 }
493 
494 #ifdef notyet
495 
496 /*
497  * Create a directory
498  */
499 static struct mqfs_node *
500 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
501 	struct ucred *cred, int mode)
502 {
503 	struct mqfs_node *node;
504 
505 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
506 	if (mqfs_add_node(parent, node) != 0) {
507 		mqnode_free(node);
508 		return (NULL);
509 	}
510 
511 	if (mqfs_fixup_dir(node) != 0) {
512 		mqfs_destroy(node);
513 		return (NULL);
514 	}
515 	return (node);
516 }
517 
518 /*
519  * Create a symlink
520  */
521 static struct mqfs_node *
522 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
523 	struct ucred *cred, int mode)
524 {
525 	struct mqfs_node *node;
526 
527 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
528 	if (mqfs_add_node(parent, node) != 0) {
529 		mqnode_free(node);
530 		return (NULL);
531 	}
532 	return (node);
533 }
534 
535 #endif
536 
537 /*
538  * Destroy a node or a tree of nodes
539  */
540 static int
541 mqfs_destroy(struct mqfs_node *node)
542 {
543 	struct mqfs_node *parent;
544 
545 	KASSERT(node != NULL,
546 	    ("%s(): node is NULL", __func__));
547 	KASSERT(node->mn_info != NULL,
548 	    ("%s(): node has no mn_info", __func__));
549 
550 	/* destroy children */
551 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
552 		while (! LIST_EMPTY(&node->mn_children))
553 			mqfs_destroy(LIST_FIRST(&node->mn_children));
554 
555 	/* unlink from parent */
556 	if ((parent = node->mn_parent) != NULL) {
557 		KASSERT(parent->mn_info == node->mn_info,
558 		    ("%s(): parent has different mn_info", __func__));
559 		LIST_REMOVE(node, mn_sibling);
560 	}
561 
562 	if (node->mn_fileno != 0)
563 		mqfs_fileno_free(node->mn_info, node);
564 	if (node->mn_data != NULL)
565 		mqueue_free(node->mn_data);
566 	mqnode_free(node);
567 	return (0);
568 }
569 
570 /*
571  * Mount a mqfs instance
572  */
573 static int
574 mqfs_mount(struct mount *mp)
575 {
576 	struct statfs *sbp;
577 
578 	if (mp->mnt_flag & MNT_UPDATE)
579 		return (EOPNOTSUPP);
580 
581 	mp->mnt_data = &mqfs_data;
582 	MNT_ILOCK(mp);
583 	mp->mnt_flag |= MNT_LOCAL;
584 	mp->mnt_kern_flag |= MNTK_MPSAFE;
585 	MNT_IUNLOCK(mp);
586 	vfs_getnewfsid(mp);
587 
588 	sbp = &mp->mnt_stat;
589 	vfs_mountedfrom(mp, "mqueue");
590 	sbp->f_bsize = PAGE_SIZE;
591 	sbp->f_iosize = PAGE_SIZE;
592 	sbp->f_blocks = 1;
593 	sbp->f_bfree = 0;
594 	sbp->f_bavail = 0;
595 	sbp->f_files = 1;
596 	sbp->f_ffree = 0;
597 	return (0);
598 }
599 
600 /*
601  * Unmount a mqfs instance
602  */
603 static int
604 mqfs_unmount(struct mount *mp, int mntflags)
605 {
606 	int error;
607 
608 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0,
609 	    curthread);
610 	return (error);
611 }
612 
613 /*
614  * Return a root vnode
615  */
616 static int
617 mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
618 {
619 	struct mqfs_info *mqfs;
620 	int ret;
621 
622 	mqfs = VFSTOMQFS(mp);
623 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
624 	return (ret);
625 }
626 
627 /*
628  * Return filesystem stats
629  */
630 static int
631 mqfs_statfs(struct mount *mp, struct statfs *sbp)
632 {
633 	/* XXX update statistics */
634 	return (0);
635 }
636 
637 /*
638  * Initialize a mqfs instance
639  */
640 static int
641 mqfs_init(struct vfsconf *vfc)
642 {
643 	struct mqfs_node *root;
644 	struct mqfs_info *mi;
645 
646 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
647 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
648 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
649 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
650 	mvdata_zone = uma_zcreate("mvdata",
651 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
652 		NULL, UMA_ALIGN_PTR, 0);
653 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
654 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
655 	mi = &mqfs_data;
656 	sx_init(&mi->mi_lock, "mqfs lock");
657 	/* set up the root diretory */
658 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
659 		mqfstype_root);
660 	root->mn_info = mi;
661 	LIST_INIT(&root->mn_children);
662 	LIST_INIT(&root->mn_vnodes);
663 	mi->mi_root = root;
664 	mqfs_fileno_init(mi);
665 	mqfs_fileno_alloc(mi, root);
666 	mqfs_fixup_dir(root);
667 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
668 	    EVENTHANDLER_PRI_ANY);
669 	mq_fdclose = mqueue_fdclose;
670 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
671 	return (0);
672 }
673 
674 /*
675  * Destroy a mqfs instance
676  */
677 static int
678 mqfs_uninit(struct vfsconf *vfc)
679 {
680 	struct mqfs_info *mi;
681 
682 	if (!unloadable)
683 		return (EOPNOTSUPP);
684 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
685 	mi = &mqfs_data;
686 	mqfs_destroy(mi->mi_root);
687 	mi->mi_root = NULL;
688 	mqfs_fileno_uninit(mi);
689 	sx_destroy(&mi->mi_lock);
690 	uma_zdestroy(mqnode_zone);
691 	uma_zdestroy(mqueue_zone);
692 	uma_zdestroy(mvdata_zone);
693 	uma_zdestroy(mqnoti_zone);
694 	return (0);
695 }
696 
697 /*
698  * task routine
699  */
700 static void
701 do_recycle(void *context, int pending __unused)
702 {
703 	struct vnode *vp = (struct vnode *)context;
704 
705 	vrecycle(vp, curthread);
706 	vdrop(vp);
707 }
708 
709 /*
710  * Allocate a vnode
711  */
712 static int
713 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
714 {
715 	struct mqfs_vdata *vd;
716 	struct mqfs_info  *mqfs;
717 	struct vnode *newvpp;
718 	int error;
719 
720 	mqfs = pn->mn_info;
721 	*vpp = NULL;
722 	sx_xlock(&mqfs->mi_lock);
723 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
724 		if (vd->mv_vnode->v_mount == mp) {
725 			vhold(vd->mv_vnode);
726 			break;
727 		}
728 	}
729 
730 	if (vd != NULL) {
731 found:
732 		*vpp = vd->mv_vnode;
733 		sx_xunlock(&mqfs->mi_lock);
734 		error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
735 		vdrop(*vpp);
736 		return (error);
737 	}
738 	sx_xunlock(&mqfs->mi_lock);
739 
740 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
741 	if (error)
742 		return (error);
743 	vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
744 	error = insmntque(newvpp, mp);
745 	if (error != 0)
746 		return (error);
747 
748 	sx_xlock(&mqfs->mi_lock);
749 	/*
750 	 * Check if it has already been allocated
751 	 * while we were blocked.
752 	 */
753 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
754 		if (vd->mv_vnode->v_mount == mp) {
755 			vhold(vd->mv_vnode);
756 			sx_xunlock(&mqfs->mi_lock);
757 
758 			vgone(newvpp);
759 			vput(newvpp);
760 			goto found;
761 		}
762 	}
763 
764 	*vpp = newvpp;
765 
766 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
767 	(*vpp)->v_data = vd;
768 	vd->mv_vnode = *vpp;
769 	vd->mv_node = pn;
770 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
771 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
772 	mqnode_addref(pn);
773 	switch (pn->mn_type) {
774 	case mqfstype_root:
775 		(*vpp)->v_vflag = VV_ROOT;
776 		/* fall through */
777 	case mqfstype_dir:
778 	case mqfstype_this:
779 	case mqfstype_parent:
780 		(*vpp)->v_type = VDIR;
781 		break;
782 	case mqfstype_file:
783 		(*vpp)->v_type = VREG;
784 		break;
785 	case mqfstype_symlink:
786 		(*vpp)->v_type = VLNK;
787 		break;
788 	case mqfstype_none:
789 		KASSERT(0, ("mqfs_allocf called for null node\n"));
790 	default:
791 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
792 	}
793 	sx_xunlock(&mqfs->mi_lock);
794 	return (0);
795 }
796 
797 /*
798  * Search a directory entry
799  */
800 static struct mqfs_node *
801 mqfs_search(struct mqfs_node *pd, const char *name, int len)
802 {
803 	struct mqfs_node *pn;
804 
805 	sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
806 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
807 		if (strncmp(pn->mn_name, name, len) == 0 &&
808 		    pn->mn_name[len] == '\0')
809 			return (pn);
810 	}
811 	return (NULL);
812 }
813 
814 /*
815  * Look up a file or directory.
816  */
817 static int
818 mqfs_lookupx(struct vop_cachedlookup_args *ap)
819 {
820 	struct componentname *cnp;
821 	struct vnode *dvp, **vpp;
822 	struct mqfs_node *pd;
823 	struct mqfs_node *pn;
824 	struct mqfs_info *mqfs;
825 	int nameiop, flags, error, namelen;
826 	char *pname;
827 	struct thread *td;
828 
829 	cnp = ap->a_cnp;
830 	vpp = ap->a_vpp;
831 	dvp = ap->a_dvp;
832 	pname = cnp->cn_nameptr;
833 	namelen = cnp->cn_namelen;
834 	td = cnp->cn_thread;
835 	flags = cnp->cn_flags;
836 	nameiop = cnp->cn_nameiop;
837 	pd = VTON(dvp);
838 	pn = NULL;
839 	mqfs = pd->mn_info;
840 	*vpp = NULLVP;
841 
842 	if (dvp->v_type != VDIR)
843 		return (ENOTDIR);
844 
845 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
846 	if (error)
847 		return (error);
848 
849 	/* shortcut: check if the name is too long */
850 	if (cnp->cn_namelen >= MQFS_NAMELEN)
851 		return (ENOENT);
852 
853 	/* self */
854 	if (namelen == 1 && pname[0] == '.') {
855 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
856 			return (EINVAL);
857 		pn = pd;
858 		*vpp = dvp;
859 		VREF(dvp);
860 		return (0);
861 	}
862 
863 	/* parent */
864 	if (cnp->cn_flags & ISDOTDOT) {
865 		if (dvp->v_vflag & VV_ROOT)
866 			return (EIO);
867 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
868 			return (EINVAL);
869 		VOP_UNLOCK(dvp, 0);
870 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
871 		pn = pd->mn_parent;
872 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
873 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
874 		return (error);
875 	}
876 
877 	/* named node */
878 	sx_xlock(&mqfs->mi_lock);
879 	pn = mqfs_search(pd, pname, namelen);
880 	if (pn != NULL)
881 		mqnode_addref(pn);
882 	sx_xunlock(&mqfs->mi_lock);
883 
884 	/* found */
885 	if (pn != NULL) {
886 		/* DELETE */
887 		if (nameiop == DELETE && (flags & ISLASTCN)) {
888 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
889 			if (error) {
890 				mqnode_release(pn);
891 				return (error);
892 			}
893 			if (*vpp == dvp) {
894 				VREF(dvp);
895 				*vpp = dvp;
896 				mqnode_release(pn);
897 				return (0);
898 			}
899 		}
900 
901 		/* allocate vnode */
902 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
903 		mqnode_release(pn);
904 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
905 			cache_enter(dvp, *vpp, cnp);
906 		return (error);
907 	}
908 
909 	/* not found */
910 
911 	/* will create a new entry in the directory ? */
912 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
913 	    && (flags & ISLASTCN)) {
914 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
915 		if (error)
916 			return (error);
917 		cnp->cn_flags |= SAVENAME;
918 		return (EJUSTRETURN);
919 	}
920 	return (ENOENT);
921 }
922 
923 #if 0
924 struct vop_lookup_args {
925 	struct vop_generic_args a_gen;
926 	struct vnode *a_dvp;
927 	struct vnode **a_vpp;
928 	struct componentname *a_cnp;
929 };
930 #endif
931 
932 /*
933  * vnode lookup operation
934  */
935 static int
936 mqfs_lookup(struct vop_cachedlookup_args *ap)
937 {
938 	int rc;
939 
940 	rc = mqfs_lookupx(ap);
941 	return (rc);
942 }
943 
944 #if 0
945 struct vop_create_args {
946 	struct vnode *a_dvp;
947 	struct vnode **a_vpp;
948 	struct componentname *a_cnp;
949 	struct vattr *a_vap;
950 };
951 #endif
952 
953 /*
954  * vnode creation operation
955  */
956 static int
957 mqfs_create(struct vop_create_args *ap)
958 {
959 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
960 	struct componentname *cnp = ap->a_cnp;
961 	struct mqfs_node *pd;
962 	struct mqfs_node *pn;
963 	struct mqueue *mq;
964 	int error;
965 
966 	pd = VTON(ap->a_dvp);
967 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
968 		return (ENOTDIR);
969 	mq = mqueue_alloc(NULL);
970 	if (mq == NULL)
971 		return (EAGAIN);
972 	sx_xlock(&mqfs->mi_lock);
973 	if ((cnp->cn_flags & HASBUF) == 0)
974 		panic("%s: no name", __func__);
975 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
976 		cnp->cn_cred, ap->a_vap->va_mode);
977 	if (pn == NULL) {
978 		sx_xunlock(&mqfs->mi_lock);
979 		error = ENOSPC;
980 	} else {
981 		mqnode_addref(pn);
982 		sx_xunlock(&mqfs->mi_lock);
983 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
984 		mqnode_release(pn);
985 		if (error)
986 			mqfs_destroy(pn);
987 		else
988 			pn->mn_data = mq;
989 	}
990 	if (error)
991 		mqueue_free(mq);
992 	return (error);
993 }
994 
995 /*
996  * Remove an entry
997  */
998 static
999 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
1000 {
1001 	struct mqfs_node *parent;
1002 	struct mqfs_vdata *vd;
1003 	int error = 0;
1004 
1005 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
1006 
1007 	if (ucred->cr_uid != pn->mn_uid &&
1008 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
1009 		error = EACCES;
1010 	else if (!pn->mn_deleted) {
1011 		parent = pn->mn_parent;
1012 		pn->mn_parent = NULL;
1013 		pn->mn_deleted = 1;
1014 		LIST_REMOVE(pn, mn_sibling);
1015 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
1016 			cache_purge(vd->mv_vnode);
1017 			vhold(vd->mv_vnode);
1018 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
1019 		}
1020 		mqnode_release(pn);
1021 		mqnode_release(parent);
1022 	} else
1023 		error = ENOENT;
1024 	return (error);
1025 }
1026 
1027 #if 0
1028 struct vop_remove_args {
1029 	struct vnode *a_dvp;
1030 	struct vnode *a_vp;
1031 	struct componentname *a_cnp;
1032 };
1033 #endif
1034 
1035 /*
1036  * vnode removal operation
1037  */
1038 static int
1039 mqfs_remove(struct vop_remove_args *ap)
1040 {
1041 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1042 	struct mqfs_node *pn;
1043 	int error;
1044 
1045 	if (ap->a_vp->v_type == VDIR)
1046                 return (EPERM);
1047 	pn = VTON(ap->a_vp);
1048 	sx_xlock(&mqfs->mi_lock);
1049 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1050 	sx_xunlock(&mqfs->mi_lock);
1051 	return (error);
1052 }
1053 
1054 #if 0
1055 struct vop_inactive_args {
1056 	struct vnode *a_vp;
1057 	struct thread *a_td;
1058 };
1059 #endif
1060 
1061 static int
1062 mqfs_inactive(struct vop_inactive_args *ap)
1063 {
1064 	struct mqfs_node *pn = VTON(ap->a_vp);
1065 
1066 	if (pn->mn_deleted)
1067 		vrecycle(ap->a_vp, ap->a_td);
1068 	return (0);
1069 }
1070 
1071 #if 0
1072 struct vop_reclaim_args {
1073 	struct vop_generic_args a_gen;
1074 	struct vnode *a_vp;
1075 	struct thread *a_td;
1076 };
1077 #endif
1078 
1079 static int
1080 mqfs_reclaim(struct vop_reclaim_args *ap)
1081 {
1082 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1083 	struct vnode *vp = ap->a_vp;
1084 	struct mqfs_node *pn;
1085 	struct mqfs_vdata *vd;
1086 
1087 	vd = vp->v_data;
1088 	pn = vd->mv_node;
1089 	sx_xlock(&mqfs->mi_lock);
1090 	vp->v_data = NULL;
1091 	LIST_REMOVE(vd, mv_link);
1092 	uma_zfree(mvdata_zone, vd);
1093 	mqnode_release(pn);
1094 	sx_xunlock(&mqfs->mi_lock);
1095 	return (0);
1096 }
1097 
1098 #if 0
1099 struct vop_open_args {
1100 	struct vop_generic_args a_gen;
1101 	struct vnode *a_vp;
1102 	int a_mode;
1103 	struct ucred *a_cred;
1104 	struct thread *a_td;
1105 	struct file *a_fp;
1106 };
1107 #endif
1108 
1109 static int
1110 mqfs_open(struct vop_open_args *ap)
1111 {
1112 	return (0);
1113 }
1114 
1115 #if 0
1116 struct vop_close_args {
1117 	struct vop_generic_args a_gen;
1118 	struct vnode *a_vp;
1119 	int a_fflag;
1120 	struct ucred *a_cred;
1121 	struct thread *a_td;
1122 };
1123 #endif
1124 
1125 static int
1126 mqfs_close(struct vop_close_args *ap)
1127 {
1128 	return (0);
1129 }
1130 
1131 #if 0
1132 struct vop_access_args {
1133 	struct vop_generic_args a_gen;
1134 	struct vnode *a_vp;
1135 	accmode_t a_accmode;
1136 	struct ucred *a_cred;
1137 	struct thread *a_td;
1138 };
1139 #endif
1140 
1141 /*
1142  * Verify permissions
1143  */
1144 static int
1145 mqfs_access(struct vop_access_args *ap)
1146 {
1147 	struct vnode *vp = ap->a_vp;
1148 	struct vattr vattr;
1149 	int error;
1150 
1151 	error = VOP_GETATTR(vp, &vattr, ap->a_cred);
1152 	if (error)
1153 		return (error);
1154 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1155 	    vattr.va_gid, ap->a_accmode, ap->a_cred, NULL);
1156 	return (error);
1157 }
1158 
1159 #if 0
1160 struct vop_getattr_args {
1161 	struct vop_generic_args a_gen;
1162 	struct vnode *a_vp;
1163 	struct vattr *a_vap;
1164 	struct ucred *a_cred;
1165 };
1166 #endif
1167 
1168 /*
1169  * Get file attributes
1170  */
1171 static int
1172 mqfs_getattr(struct vop_getattr_args *ap)
1173 {
1174 	struct vnode *vp = ap->a_vp;
1175 	struct mqfs_node *pn = VTON(vp);
1176 	struct vattr *vap = ap->a_vap;
1177 	int error = 0;
1178 
1179 	vap->va_type = vp->v_type;
1180 	vap->va_mode = pn->mn_mode;
1181 	vap->va_nlink = 1;
1182 	vap->va_uid = pn->mn_uid;
1183 	vap->va_gid = pn->mn_gid;
1184 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1185 	vap->va_fileid = pn->mn_fileno;
1186 	vap->va_size = 0;
1187 	vap->va_blocksize = PAGE_SIZE;
1188 	vap->va_bytes = vap->va_size = 0;
1189 	vap->va_atime = pn->mn_atime;
1190 	vap->va_mtime = pn->mn_mtime;
1191 	vap->va_ctime = pn->mn_ctime;
1192 	vap->va_birthtime = pn->mn_birth;
1193 	vap->va_gen = 0;
1194 	vap->va_flags = 0;
1195 	vap->va_rdev = NODEV;
1196 	vap->va_bytes = 0;
1197 	vap->va_filerev = 0;
1198 	return (error);
1199 }
1200 
1201 #if 0
1202 struct vop_setattr_args {
1203 	struct vop_generic_args a_gen;
1204 	struct vnode *a_vp;
1205 	struct vattr *a_vap;
1206 	struct ucred *a_cred;
1207 };
1208 #endif
1209 /*
1210  * Set attributes
1211  */
1212 static int
1213 mqfs_setattr(struct vop_setattr_args *ap)
1214 {
1215 	struct mqfs_node *pn;
1216 	struct vattr *vap;
1217 	struct vnode *vp;
1218 	struct thread *td;
1219 	int c, error;
1220 	uid_t uid;
1221 	gid_t gid;
1222 
1223 	td = curthread;
1224 	vap = ap->a_vap;
1225 	vp = ap->a_vp;
1226 	if ((vap->va_type != VNON) ||
1227 	    (vap->va_nlink != VNOVAL) ||
1228 	    (vap->va_fsid != VNOVAL) ||
1229 	    (vap->va_fileid != VNOVAL) ||
1230 	    (vap->va_blocksize != VNOVAL) ||
1231 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1232 	    (vap->va_rdev != VNOVAL) ||
1233 	    ((int)vap->va_bytes != VNOVAL) ||
1234 	    (vap->va_gen != VNOVAL)) {
1235 		return (EINVAL);
1236 	}
1237 
1238 	pn = VTON(vp);
1239 
1240 	error = c = 0;
1241 	if (vap->va_uid == (uid_t)VNOVAL)
1242 		uid = pn->mn_uid;
1243 	else
1244 		uid = vap->va_uid;
1245 	if (vap->va_gid == (gid_t)VNOVAL)
1246 		gid = pn->mn_gid;
1247 	else
1248 		gid = vap->va_gid;
1249 
1250 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1251 		/*
1252 		 * To modify the ownership of a file, must possess VADMIN
1253 		 * for that file.
1254 		 */
1255 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
1256 			return (error);
1257 
1258 		/*
1259 		 * XXXRW: Why is there a privilege check here: shouldn't the
1260 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1261 		 * below definitely right?
1262 		 */
1263 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1264 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1265 		    (error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
1266 			return (error);
1267 		pn->mn_uid = uid;
1268 		pn->mn_gid = gid;
1269 		c = 1;
1270 	}
1271 
1272 	if (vap->va_mode != (mode_t)VNOVAL) {
1273 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1274 		    (error = priv_check(td, PRIV_MQ_ADMIN)))
1275 			return (error);
1276 		pn->mn_mode = vap->va_mode;
1277 		c = 1;
1278 	}
1279 
1280 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1281 		/* See the comment in ufs_vnops::ufs_setattr(). */
1282 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
1283 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1284 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
1285 			return (error);
1286 		if (vap->va_atime.tv_sec != VNOVAL) {
1287 			pn->mn_atime = vap->va_atime;
1288 		}
1289 		if (vap->va_mtime.tv_sec != VNOVAL) {
1290 			pn->mn_mtime = vap->va_mtime;
1291 		}
1292 		c = 1;
1293 	}
1294 	if (c) {
1295 		vfs_timestamp(&pn->mn_ctime);
1296 	}
1297 	return (0);
1298 }
1299 
1300 #if 0
1301 struct vop_read_args {
1302 	struct vop_generic_args a_gen;
1303 	struct vnode *a_vp;
1304 	struct uio *a_uio;
1305 	int a_ioflag;
1306 	struct ucred *a_cred;
1307 };
1308 #endif
1309 
1310 /*
1311  * Read from a file
1312  */
1313 static int
1314 mqfs_read(struct vop_read_args *ap)
1315 {
1316 	char buf[80];
1317 	struct vnode *vp = ap->a_vp;
1318 	struct uio *uio = ap->a_uio;
1319 	struct mqfs_node *pn;
1320 	struct mqueue *mq;
1321 	int len, error;
1322 
1323 	if (vp->v_type != VREG)
1324 		return (EINVAL);
1325 
1326 	pn = VTON(vp);
1327 	mq = VTOMQ(vp);
1328 	snprintf(buf, sizeof(buf),
1329 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1330 		mq->mq_totalbytes,
1331 		mq->mq_maxmsg,
1332 		mq->mq_curmsgs,
1333 		mq->mq_msgsize);
1334 	buf[sizeof(buf)-1] = '\0';
1335 	len = strlen(buf);
1336 	error = uiomove_frombuf(buf, len, uio);
1337 	return (error);
1338 }
1339 
1340 #if 0
1341 struct vop_readdir_args {
1342 	struct vop_generic_args a_gen;
1343 	struct vnode *a_vp;
1344 	struct uio *a_uio;
1345 	struct ucred *a_cred;
1346 	int *a_eofflag;
1347 	int *a_ncookies;
1348 	u_long **a_cookies;
1349 };
1350 #endif
1351 
1352 /*
1353  * Return directory entries.
1354  */
1355 static int
1356 mqfs_readdir(struct vop_readdir_args *ap)
1357 {
1358 	struct vnode *vp;
1359 	struct mqfs_info *mi;
1360 	struct mqfs_node *pd;
1361 	struct mqfs_node *pn;
1362 	struct dirent entry;
1363 	struct uio *uio;
1364 	int *tmp_ncookies = NULL;
1365 	off_t offset;
1366 	int error, i;
1367 
1368 	vp = ap->a_vp;
1369 	mi = VFSTOMQFS(vp->v_mount);
1370 	pd = VTON(vp);
1371 	uio = ap->a_uio;
1372 
1373 	if (vp->v_type != VDIR)
1374 		return (ENOTDIR);
1375 
1376 	if (uio->uio_offset < 0)
1377 		return (EINVAL);
1378 
1379 	if (ap->a_ncookies != NULL) {
1380 		tmp_ncookies = ap->a_ncookies;
1381 		*ap->a_ncookies = 0;
1382 		ap->a_ncookies = NULL;
1383         }
1384 
1385 	error = 0;
1386 	offset = 0;
1387 
1388 	sx_xlock(&mi->mi_lock);
1389 
1390 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1391 		entry.d_reclen = sizeof(entry);
1392 		if (!pn->mn_fileno)
1393 			mqfs_fileno_alloc(mi, pn);
1394 		entry.d_fileno = pn->mn_fileno;
1395 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1396 			entry.d_name[i] = pn->mn_name[i];
1397 		entry.d_name[i] = 0;
1398 		entry.d_namlen = i;
1399 		switch (pn->mn_type) {
1400 		case mqfstype_root:
1401 		case mqfstype_dir:
1402 		case mqfstype_this:
1403 		case mqfstype_parent:
1404 			entry.d_type = DT_DIR;
1405 			break;
1406 		case mqfstype_file:
1407 			entry.d_type = DT_REG;
1408 			break;
1409 		case mqfstype_symlink:
1410 			entry.d_type = DT_LNK;
1411 			break;
1412 		default:
1413 			panic("%s has unexpected node type: %d", pn->mn_name,
1414 				pn->mn_type);
1415 		}
1416 		if (entry.d_reclen > uio->uio_resid)
1417                         break;
1418 		if (offset >= uio->uio_offset) {
1419 			error = vfs_read_dirent(ap, &entry, offset);
1420                         if (error)
1421                                 break;
1422                 }
1423                 offset += entry.d_reclen;
1424 	}
1425 	sx_xunlock(&mi->mi_lock);
1426 
1427 	uio->uio_offset = offset;
1428 
1429 	if (tmp_ncookies != NULL)
1430 		ap->a_ncookies = tmp_ncookies;
1431 
1432 	return (error);
1433 }
1434 
1435 #ifdef notyet
1436 
1437 #if 0
1438 struct vop_mkdir_args {
1439 	struct vnode *a_dvp;
1440 	struvt vnode **a_vpp;
1441 	struvt componentname *a_cnp;
1442 	struct vattr *a_vap;
1443 };
1444 #endif
1445 
1446 /*
1447  * Create a directory.
1448  */
1449 static int
1450 mqfs_mkdir(struct vop_mkdir_args *ap)
1451 {
1452 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1453 	struct componentname *cnp = ap->a_cnp;
1454 	struct mqfs_node *pd = VTON(ap->a_dvp);
1455 	struct mqfs_node *pn;
1456 	int error;
1457 
1458 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1459 		return (ENOTDIR);
1460 	sx_xlock(&mqfs->mi_lock);
1461 	if ((cnp->cn_flags & HASBUF) == 0)
1462 		panic("%s: no name", __func__);
1463 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1464 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1465 	if (pn != NULL)
1466 		mqnode_addref(pn);
1467 	sx_xunlock(&mqfs->mi_lock);
1468 	if (pn == NULL) {
1469 		error = ENOSPC;
1470 	} else {
1471 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1472 		mqnode_release(pn);
1473 	}
1474 	return (error);
1475 }
1476 
1477 #if 0
1478 struct vop_rmdir_args {
1479 	struct vnode *a_dvp;
1480 	struct vnode *a_vp;
1481 	struct componentname *a_cnp;
1482 };
1483 #endif
1484 
1485 /*
1486  * Remove a directory.
1487  */
1488 static int
1489 mqfs_rmdir(struct vop_rmdir_args *ap)
1490 {
1491 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1492 	struct mqfs_node *pn = VTON(ap->a_vp);
1493 	struct mqfs_node *pt;
1494 
1495 	if (pn->mn_type != mqfstype_dir)
1496 		return (ENOTDIR);
1497 
1498 	sx_xlock(&mqfs->mi_lock);
1499 	if (pn->mn_deleted) {
1500 		sx_xunlock(&mqfs->mi_lock);
1501 		return (ENOENT);
1502 	}
1503 
1504 	pt = LIST_FIRST(&pn->mn_children);
1505 	pt = LIST_NEXT(pt, mn_sibling);
1506 	pt = LIST_NEXT(pt, mn_sibling);
1507 	if (pt != NULL) {
1508 		sx_xunlock(&mqfs->mi_lock);
1509 		return (ENOTEMPTY);
1510 	}
1511 	pt = pn->mn_parent;
1512 	pn->mn_parent = NULL;
1513 	pn->mn_deleted = 1;
1514 	LIST_REMOVE(pn, mn_sibling);
1515 	mqnode_release(pn);
1516 	mqnode_release(pt);
1517 	sx_xunlock(&mqfs->mi_lock);
1518 	cache_purge(ap->a_vp);
1519 	return (0);
1520 }
1521 
1522 #endif /* notyet */
1523 
1524 /*
1525  * Allocate a message queue
1526  */
1527 static struct mqueue *
1528 mqueue_alloc(const struct mq_attr *attr)
1529 {
1530 	struct mqueue *mq;
1531 
1532 	if (curmq >= maxmq)
1533 		return (NULL);
1534 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1535 	TAILQ_INIT(&mq->mq_msgq);
1536 	if (attr != NULL) {
1537 		mq->mq_maxmsg = attr->mq_maxmsg;
1538 		mq->mq_msgsize = attr->mq_msgsize;
1539 	} else {
1540 		mq->mq_maxmsg = default_maxmsg;
1541 		mq->mq_msgsize = default_msgsize;
1542 	}
1543 	mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
1544 	knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
1545 	knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
1546 	atomic_add_int(&curmq, 1);
1547 	return (mq);
1548 }
1549 
1550 /*
1551  * Destroy a message queue
1552  */
1553 static void
1554 mqueue_free(struct mqueue *mq)
1555 {
1556 	struct mqueue_msg *msg;
1557 
1558 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1559 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1560 		free(msg, M_MQUEUEDATA);
1561 	}
1562 
1563 	mtx_destroy(&mq->mq_mutex);
1564 	knlist_destroy(&mq->mq_rsel.si_note);
1565 	knlist_destroy(&mq->mq_wsel.si_note);
1566 	uma_zfree(mqueue_zone, mq);
1567 	atomic_add_int(&curmq, -1);
1568 }
1569 
1570 /*
1571  * Load a message from user space
1572  */
1573 static struct mqueue_msg *
1574 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1575 {
1576 	struct mqueue_msg *msg;
1577 	size_t len;
1578 	int error;
1579 
1580 	len = sizeof(struct mqueue_msg) + msg_size;
1581 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1582 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1583 	    msg_size);
1584 	if (error) {
1585 		free(msg, M_MQUEUEDATA);
1586 		msg = NULL;
1587 	} else {
1588 		msg->msg_size = msg_size;
1589 		msg->msg_prio = msg_prio;
1590 	}
1591 	return (msg);
1592 }
1593 
1594 /*
1595  * Save a message to user space
1596  */
1597 static int
1598 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1599 {
1600 	int error;
1601 
1602 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1603 		msg->msg_size);
1604 	if (error == 0 && msg_prio != NULL)
1605 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1606 	return (error);
1607 }
1608 
1609 /*
1610  * Free a message's memory
1611  */
1612 static __inline void
1613 mqueue_freemsg(struct mqueue_msg *msg)
1614 {
1615 	free(msg, M_MQUEUEDATA);
1616 }
1617 
1618 /*
1619  * Send a message. if waitok is false, thread will not be
1620  * blocked if there is no data in queue, otherwise, absolute
1621  * time will be checked.
1622  */
1623 int
1624 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1625 	size_t msg_len, unsigned msg_prio, int waitok,
1626 	const struct timespec *abs_timeout)
1627 {
1628 	struct mqueue_msg *msg;
1629 	struct timespec ts, ts2;
1630 	struct timeval tv;
1631 	int error;
1632 
1633 	if (msg_prio >= MQ_PRIO_MAX)
1634 		return (EINVAL);
1635 	if (msg_len > mq->mq_msgsize)
1636 		return (EMSGSIZE);
1637 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1638 	if (msg == NULL)
1639 		return (EFAULT);
1640 
1641 	/* O_NONBLOCK case */
1642 	if (!waitok) {
1643 		error = _mqueue_send(mq, msg, -1);
1644 		if (error)
1645 			goto bad;
1646 		return (0);
1647 	}
1648 
1649 	/* we allow a null timeout (wait forever) */
1650 	if (abs_timeout == NULL) {
1651 		error = _mqueue_send(mq, msg, 0);
1652 		if (error)
1653 			goto bad;
1654 		return (0);
1655 	}
1656 
1657 	/* send it before checking time */
1658 	error = _mqueue_send(mq, msg, -1);
1659 	if (error == 0)
1660 		return (0);
1661 
1662 	if (error != EAGAIN)
1663 		goto bad;
1664 
1665 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1666 		error = EINVAL;
1667 		goto bad;
1668 	}
1669 	for (;;) {
1670 		ts2 = *abs_timeout;
1671 		getnanotime(&ts);
1672 		timespecsub(&ts2, &ts);
1673 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1674 			error = ETIMEDOUT;
1675 			break;
1676 		}
1677 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1678 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1679 		if (error != ETIMEDOUT)
1680 			break;
1681 	}
1682 	if (error == 0)
1683 		return (0);
1684 bad:
1685 	mqueue_freemsg(msg);
1686 	return (error);
1687 }
1688 
1689 /*
1690  * Common routine to send a message
1691  */
1692 static int
1693 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1694 {
1695 	struct mqueue_msg *msg2;
1696 	int error = 0;
1697 
1698 	mtx_lock(&mq->mq_mutex);
1699 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1700 		if (timo < 0) {
1701 			mtx_unlock(&mq->mq_mutex);
1702 			return (EAGAIN);
1703 		}
1704 		mq->mq_senders++;
1705 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1706 			    PCATCH, "mqsend", timo);
1707 		mq->mq_senders--;
1708 		if (error == EAGAIN)
1709 			error = ETIMEDOUT;
1710 	}
1711 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1712 		mtx_unlock(&mq->mq_mutex);
1713 		return (error);
1714 	}
1715 	error = 0;
1716 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1717 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1718 	} else {
1719 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1720 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1721 		} else {
1722 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1723 				if (msg2->msg_prio < msg->msg_prio)
1724 					break;
1725 			}
1726 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1727 		}
1728 	}
1729 	mq->mq_curmsgs++;
1730 	mq->mq_totalbytes += msg->msg_size;
1731 	if (mq->mq_receivers)
1732 		wakeup_one(&mq->mq_receivers);
1733 	else if (mq->mq_notifier != NULL)
1734 		mqueue_send_notification(mq);
1735 	if (mq->mq_flags & MQ_RSEL) {
1736 		mq->mq_flags &= ~MQ_RSEL;
1737 		selwakeup(&mq->mq_rsel);
1738 	}
1739 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1740 	mtx_unlock(&mq->mq_mutex);
1741 	return (0);
1742 }
1743 
1744 /*
1745  * Send realtime a signal to process which registered itself
1746  * successfully by mq_notify.
1747  */
1748 static void
1749 mqueue_send_notification(struct mqueue *mq)
1750 {
1751 	struct mqueue_notifier *nt;
1752 	struct thread *td;
1753 	struct proc *p;
1754 	int error;
1755 
1756 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1757 	nt = mq->mq_notifier;
1758 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1759 		p = nt->nt_proc;
1760 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1761 		if (error) {
1762 			mq->mq_notifier = NULL;
1763 			return;
1764 		}
1765 		if (!KSI_ONQ(&nt->nt_ksi)) {
1766 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1767 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1768 		}
1769 		PROC_UNLOCK(p);
1770 	}
1771 	mq->mq_notifier = NULL;
1772 }
1773 
1774 /*
1775  * Get a message. if waitok is false, thread will not be
1776  * blocked if there is no data in queue, otherwise, absolute
1777  * time will be checked.
1778  */
1779 int
1780 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1781 	size_t msg_len, unsigned *msg_prio, int waitok,
1782 	const struct timespec *abs_timeout)
1783 {
1784 	struct mqueue_msg *msg;
1785 	struct timespec ts, ts2;
1786 	struct timeval tv;
1787 	int error;
1788 
1789 	if (msg_len < mq->mq_msgsize)
1790 		return (EMSGSIZE);
1791 
1792 	/* O_NONBLOCK case */
1793 	if (!waitok) {
1794 		error = _mqueue_recv(mq, &msg, -1);
1795 		if (error)
1796 			return (error);
1797 		goto received;
1798 	}
1799 
1800 	/* we allow a null timeout (wait forever). */
1801 	if (abs_timeout == NULL) {
1802 		error = _mqueue_recv(mq, &msg, 0);
1803 		if (error)
1804 			return (error);
1805 		goto received;
1806 	}
1807 
1808 	/* try to get a message before checking time */
1809 	error = _mqueue_recv(mq, &msg, -1);
1810 	if (error == 0)
1811 		goto received;
1812 
1813 	if (error != EAGAIN)
1814 		return (error);
1815 
1816 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1817 		error = EINVAL;
1818 		return (error);
1819 	}
1820 
1821 	for (;;) {
1822 		ts2 = *abs_timeout;
1823 		getnanotime(&ts);
1824 		timespecsub(&ts2, &ts);
1825 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1826 			error = ETIMEDOUT;
1827 			return (error);
1828 		}
1829 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1830 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1831 		if (error == 0)
1832 			break;
1833 		if (error != ETIMEDOUT)
1834 			return (error);
1835 	}
1836 
1837 received:
1838 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1839 	if (error == 0) {
1840 		curthread->td_retval[0] = msg->msg_size;
1841 		curthread->td_retval[1] = 0;
1842 	}
1843 	mqueue_freemsg(msg);
1844 	return (error);
1845 }
1846 
1847 /*
1848  * Common routine to receive a message
1849  */
1850 static int
1851 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1852 {
1853 	int error = 0;
1854 
1855 	mtx_lock(&mq->mq_mutex);
1856 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1857 		if (timo < 0) {
1858 			mtx_unlock(&mq->mq_mutex);
1859 			return (EAGAIN);
1860 		}
1861 		mq->mq_receivers++;
1862 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1863 			    PCATCH, "mqrecv", timo);
1864 		mq->mq_receivers--;
1865 		if (error == EAGAIN)
1866 			error = ETIMEDOUT;
1867 	}
1868 	if (*msg != NULL) {
1869 		error = 0;
1870 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1871 		mq->mq_curmsgs--;
1872 		mq->mq_totalbytes -= (*msg)->msg_size;
1873 		if (mq->mq_senders)
1874 			wakeup_one(&mq->mq_senders);
1875 		if (mq->mq_flags & MQ_WSEL) {
1876 			mq->mq_flags &= ~MQ_WSEL;
1877 			selwakeup(&mq->mq_wsel);
1878 		}
1879 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1880 	}
1881 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1882 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1883 		mqueue_send_notification(mq);
1884 	}
1885 	mtx_unlock(&mq->mq_mutex);
1886 	return (error);
1887 }
1888 
1889 static __inline struct mqueue_notifier *
1890 notifier_alloc(void)
1891 {
1892 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1893 }
1894 
1895 static __inline void
1896 notifier_free(struct mqueue_notifier *p)
1897 {
1898 	uma_zfree(mqnoti_zone, p);
1899 }
1900 
1901 static struct mqueue_notifier *
1902 notifier_search(struct proc *p, int fd)
1903 {
1904 	struct mqueue_notifier *nt;
1905 
1906 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1907 		if (nt->nt_ksi.ksi_mqd == fd)
1908 			break;
1909 	}
1910 	return (nt);
1911 }
1912 
1913 static __inline void
1914 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1915 {
1916 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1917 }
1918 
1919 static __inline void
1920 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1921 {
1922 	LIST_REMOVE(nt, nt_link);
1923 	notifier_free(nt);
1924 }
1925 
1926 static void
1927 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1928 {
1929 	struct mqueue_notifier *nt;
1930 
1931 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1932 	PROC_LOCK(p);
1933 	nt = notifier_search(p, fd);
1934 	if (nt != NULL) {
1935 		if (mq->mq_notifier == nt)
1936 			mq->mq_notifier = NULL;
1937 		sigqueue_take(&nt->nt_ksi);
1938 		notifier_delete(p, nt);
1939 	}
1940 	PROC_UNLOCK(p);
1941 }
1942 
1943 static int
1944 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
1945     const struct mq_attr *attr)
1946 {
1947 	char path[MQFS_NAMELEN + 1];
1948 	struct mqfs_node *pn;
1949 	struct filedesc *fdp;
1950 	struct file *fp;
1951 	struct mqueue *mq;
1952 	int fd, error, len, cmode;
1953 
1954 	fdp = td->td_proc->p_fd;
1955 	cmode = (((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1956 	mq = NULL;
1957 	if ((flags & O_CREAT) != 0 && attr != NULL) {
1958 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
1959 			return (EINVAL);
1960 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
1961 			return (EINVAL);
1962 	}
1963 
1964 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
1965         if (error)
1966 		return (error);
1967 
1968 	/*
1969 	 * The first character of name must be a slash  (/) character
1970 	 * and the remaining characters of name cannot include any slash
1971 	 * characters.
1972 	 */
1973 	len = strlen(path);
1974 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1975 		return (EINVAL);
1976 
1977 	error = falloc(td, &fp, &fd, 0);
1978 	if (error)
1979 		return (error);
1980 
1981 	sx_xlock(&mqfs_data.mi_lock);
1982 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1983 	if (pn == NULL) {
1984 		if (!(flags & O_CREAT)) {
1985 			error = ENOENT;
1986 		} else {
1987 			mq = mqueue_alloc(attr);
1988 			if (mq == NULL) {
1989 				error = ENFILE;
1990 			} else {
1991 				pn = mqfs_create_file(mqfs_data.mi_root,
1992 				         path + 1, len - 1, td->td_ucred,
1993 					 cmode);
1994 				if (pn == NULL) {
1995 					error = ENOSPC;
1996 					mqueue_free(mq);
1997 				}
1998 			}
1999 		}
2000 
2001 		if (error == 0) {
2002 			pn->mn_data = mq;
2003 		}
2004 	} else {
2005 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2006 			error = EEXIST;
2007 		} else {
2008 			accmode_t accmode = 0;
2009 
2010 			if (flags & FREAD)
2011 				accmode |= VREAD;
2012 			if (flags & FWRITE)
2013 				accmode |= VWRITE;
2014 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2015 				    pn->mn_gid, accmode, td->td_ucred, NULL);
2016 		}
2017 	}
2018 
2019 	if (error) {
2020 		sx_xunlock(&mqfs_data.mi_lock);
2021 		fdclose(fdp, fp, fd, td);
2022 		fdrop(fp, td);
2023 		return (error);
2024 	}
2025 
2026 	mqnode_addref(pn);
2027 	sx_xunlock(&mqfs_data.mi_lock);
2028 
2029 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2030 	    &mqueueops);
2031 
2032 	FILEDESC_XLOCK(fdp);
2033 	if (fdp->fd_ofiles[fd] == fp)
2034 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2035 	FILEDESC_XUNLOCK(fdp);
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 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)
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 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] != '/' || index(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, struct file **);
2091 
2092 /*
2093  * Get message queue by giving file slot
2094  */
2095 static int
2096 _getmq(struct thread *td, int fd, _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, 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 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2122 }
2123 
2124 static __inline int
2125 getmq_read(struct thread *td, int fd, struct file **fpp,
2126 	 struct mqfs_node **ppn, struct mqueue **pmq)
2127 {
2128 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2129 }
2130 
2131 static __inline int
2132 getmq_write(struct thread *td, int fd, struct file **fpp,
2133 	struct mqfs_node **ppn, struct mqueue **pmq)
2134 {
2135 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2136 }
2137 
2138 static int
2139 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2140     struct mq_attr *oattr)
2141 {
2142 	struct mqueue *mq;
2143 	struct file *fp;
2144 	u_int oflag, flag;
2145 	int error;
2146 
2147 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2148 		return (EINVAL);
2149 	error = getmq(td, mqd, &fp, NULL, &mq);
2150 	if (error)
2151 		return (error);
2152 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2153 	oattr->mq_msgsize = mq->mq_msgsize;
2154 	oattr->mq_curmsgs = mq->mq_curmsgs;
2155 	if (attr != NULL) {
2156 		do {
2157 			oflag = flag = fp->f_flag;
2158 			flag &= ~O_NONBLOCK;
2159 			flag |= (attr->mq_flags & O_NONBLOCK);
2160 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2161 	} else
2162 		oflag = fp->f_flag;
2163 	oattr->mq_flags = (O_NONBLOCK & oflag);
2164 	fdrop(fp, td);
2165 	return (error);
2166 }
2167 
2168 int
2169 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2170 {
2171 	struct mq_attr attr, oattr;
2172 	int error;
2173 
2174 	if (uap->attr != NULL) {
2175 		error = copyin(uap->attr, &attr, sizeof(attr));
2176 		if (error != 0)
2177 			return (error);
2178 	}
2179 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2180 	    &oattr);
2181 	if (error != 0)
2182 		return (error);
2183 	if (uap->oattr != NULL)
2184 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2185 	return (error);
2186 }
2187 
2188 int
2189 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2190 {
2191 	struct mqueue *mq;
2192 	struct file *fp;
2193 	struct timespec *abs_timeout, ets;
2194 	int error;
2195 	int waitok;
2196 
2197 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2198 	if (error)
2199 		return (error);
2200 	if (uap->abs_timeout != NULL) {
2201 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2202 		if (error != 0)
2203 			return (error);
2204 		abs_timeout = &ets;
2205 	} else
2206 		abs_timeout = NULL;
2207 	waitok = !(fp->f_flag & O_NONBLOCK);
2208 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2209 		uap->msg_prio, waitok, abs_timeout);
2210 	fdrop(fp, td);
2211 	return (error);
2212 }
2213 
2214 int
2215 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2216 {
2217 	struct mqueue *mq;
2218 	struct file *fp;
2219 	struct timespec *abs_timeout, ets;
2220 	int error, waitok;
2221 
2222 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2223 	if (error)
2224 		return (error);
2225 	if (uap->abs_timeout != NULL) {
2226 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2227 		if (error != 0)
2228 			return (error);
2229 		abs_timeout = &ets;
2230 	} else
2231 		abs_timeout = NULL;
2232 	waitok = !(fp->f_flag & O_NONBLOCK);
2233 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2234 		uap->msg_prio, waitok, abs_timeout);
2235 	fdrop(fp, td);
2236 	return (error);
2237 }
2238 
2239 int
2240 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2241 {
2242 	struct sigevent ev;
2243 	struct filedesc *fdp;
2244 	struct proc *p;
2245 	struct mqueue *mq;
2246 	struct file *fp;
2247 	struct mqueue_notifier *nt, *newnt = NULL;
2248 	int error;
2249 
2250 	p = td->td_proc;
2251 	fdp = td->td_proc->p_fd;
2252 	if (uap->sigev) {
2253 		error = copyin(uap->sigev, &ev, sizeof(ev));
2254 		if (error)
2255 			return (error);
2256 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2257 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2258 		    ev.sigev_notify != SIGEV_NONE)
2259 			return (EINVAL);
2260 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2261 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2262 			!_SIG_VALID(ev.sigev_signo))
2263 			return (EINVAL);
2264 	}
2265 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2266 	if (error)
2267 		return (error);
2268 again:
2269 	FILEDESC_SLOCK(fdp);
2270 	if (fget_locked(fdp, uap->mqd) != fp) {
2271 		FILEDESC_SUNLOCK(fdp);
2272 		error = EBADF;
2273 		goto out;
2274 	}
2275 	mtx_lock(&mq->mq_mutex);
2276 	FILEDESC_SUNLOCK(fdp);
2277 	if (uap->sigev != NULL) {
2278 		if (mq->mq_notifier != NULL) {
2279 			error = EBUSY;
2280 		} else {
2281 			PROC_LOCK(p);
2282 			nt = notifier_search(p, uap->mqd);
2283 			if (nt == NULL) {
2284 				if (newnt == NULL) {
2285 					PROC_UNLOCK(p);
2286 					mtx_unlock(&mq->mq_mutex);
2287 					newnt = notifier_alloc();
2288 					goto again;
2289 				}
2290 			}
2291 
2292 			if (nt != NULL) {
2293 				sigqueue_take(&nt->nt_ksi);
2294 				if (newnt != NULL) {
2295 					notifier_free(newnt);
2296 					newnt = NULL;
2297 				}
2298 			} else {
2299 				nt = newnt;
2300 				newnt = NULL;
2301 				ksiginfo_init(&nt->nt_ksi);
2302 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2303 				nt->nt_ksi.ksi_code = SI_MESGQ;
2304 				nt->nt_proc = p;
2305 				nt->nt_ksi.ksi_mqd = uap->mqd;
2306 				notifier_insert(p, nt);
2307 			}
2308 			nt->nt_sigev = ev;
2309 			mq->mq_notifier = nt;
2310 			PROC_UNLOCK(p);
2311 			/*
2312 			 * if there is no receivers and message queue
2313 			 * is not empty, we should send notification
2314 			 * as soon as possible.
2315 			 */
2316 			if (mq->mq_receivers == 0 &&
2317 			    !TAILQ_EMPTY(&mq->mq_msgq))
2318 				mqueue_send_notification(mq);
2319 		}
2320 	} else {
2321 		notifier_remove(p, mq, uap->mqd);
2322 	}
2323 	mtx_unlock(&mq->mq_mutex);
2324 
2325 out:
2326 	fdrop(fp, td);
2327 	if (newnt != NULL)
2328 		notifier_free(newnt);
2329 	return (error);
2330 }
2331 
2332 static void
2333 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2334 {
2335 	struct filedesc *fdp;
2336 	struct mqueue *mq;
2337 
2338 	fdp = td->td_proc->p_fd;
2339 	FILEDESC_LOCK_ASSERT(fdp);
2340 
2341 	if (fp->f_ops == &mqueueops) {
2342 		mq = FPTOMQ(fp);
2343 		mtx_lock(&mq->mq_mutex);
2344 		notifier_remove(td->td_proc, mq, fd);
2345 
2346 		/* have to wakeup thread in same process */
2347 		if (mq->mq_flags & MQ_RSEL) {
2348 			mq->mq_flags &= ~MQ_RSEL;
2349 			selwakeup(&mq->mq_rsel);
2350 		}
2351 		if (mq->mq_flags & MQ_WSEL) {
2352 			mq->mq_flags &= ~MQ_WSEL;
2353 			selwakeup(&mq->mq_wsel);
2354 		}
2355 		mtx_unlock(&mq->mq_mutex);
2356 	}
2357 }
2358 
2359 static void
2360 mq_proc_exit(void *arg __unused, struct proc *p)
2361 {
2362 	struct filedesc *fdp;
2363 	struct file *fp;
2364 	struct mqueue *mq;
2365 	int i;
2366 
2367 	fdp = p->p_fd;
2368 	FILEDESC_SLOCK(fdp);
2369 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2370 		fp = fget_locked(fdp, i);
2371 		if (fp != NULL && fp->f_ops == &mqueueops) {
2372 			mq = FPTOMQ(fp);
2373 			mtx_lock(&mq->mq_mutex);
2374 			notifier_remove(p, FPTOMQ(fp), i);
2375 			mtx_unlock(&mq->mq_mutex);
2376 		}
2377 	}
2378 	FILEDESC_SUNLOCK(fdp);
2379 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2380 }
2381 
2382 static int
2383 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2384 	int flags, struct thread *td)
2385 {
2386 	return (EOPNOTSUPP);
2387 }
2388 
2389 static int
2390 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2391 	int flags, struct thread *td)
2392 {
2393 	return (EOPNOTSUPP);
2394 }
2395 
2396 static int
2397 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred,
2398     struct thread *td)
2399 {
2400 
2401 	return (EINVAL);
2402 }
2403 
2404 static int
2405 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2406 	struct ucred *active_cred, struct thread *td)
2407 {
2408 	return (ENOTTY);
2409 }
2410 
2411 static int
2412 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2413 	struct thread *td)
2414 {
2415 	struct mqueue *mq = FPTOMQ(fp);
2416 	int revents = 0;
2417 
2418 	mtx_lock(&mq->mq_mutex);
2419 	if (events & (POLLIN | POLLRDNORM)) {
2420 		if (mq->mq_curmsgs) {
2421 			revents |= events & (POLLIN | POLLRDNORM);
2422 		} else {
2423 			mq->mq_flags |= MQ_RSEL;
2424 			selrecord(td, &mq->mq_rsel);
2425  		}
2426 	}
2427 	if (events & POLLOUT) {
2428 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2429 			revents |= POLLOUT;
2430 		else {
2431 			mq->mq_flags |= MQ_WSEL;
2432 			selrecord(td, &mq->mq_wsel);
2433 		}
2434 	}
2435 	mtx_unlock(&mq->mq_mutex);
2436 	return (revents);
2437 }
2438 
2439 static int
2440 mqf_close(struct file *fp, struct thread *td)
2441 {
2442 	struct mqfs_node *pn;
2443 
2444 	fp->f_ops = &badfileops;
2445 	pn = fp->f_data;
2446 	fp->f_data = NULL;
2447 	sx_xlock(&mqfs_data.mi_lock);
2448 	mqnode_release(pn);
2449 	sx_xunlock(&mqfs_data.mi_lock);
2450 	return (0);
2451 }
2452 
2453 static int
2454 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2455 	struct thread *td)
2456 {
2457 	struct mqfs_node *pn = fp->f_data;
2458 
2459 	bzero(st, sizeof *st);
2460 	st->st_atim = pn->mn_atime;
2461 	st->st_mtim = pn->mn_mtime;
2462 	st->st_ctim = pn->mn_ctime;
2463 	st->st_birthtim = pn->mn_birth;
2464 	st->st_uid = pn->mn_uid;
2465 	st->st_gid = pn->mn_gid;
2466 	st->st_mode = S_IFIFO | pn->mn_mode;
2467 	return (0);
2468 }
2469 
2470 static int
2471 mqf_kqfilter(struct file *fp, struct knote *kn)
2472 {
2473 	struct mqueue *mq = FPTOMQ(fp);
2474 	int error = 0;
2475 
2476 	if (kn->kn_filter == EVFILT_READ) {
2477 		kn->kn_fop = &mq_rfiltops;
2478 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2479 	} else if (kn->kn_filter == EVFILT_WRITE) {
2480 		kn->kn_fop = &mq_wfiltops;
2481 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2482 	} else
2483 		error = EINVAL;
2484 	return (error);
2485 }
2486 
2487 static void
2488 filt_mqdetach(struct knote *kn)
2489 {
2490 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2491 
2492 	if (kn->kn_filter == EVFILT_READ)
2493 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2494 	else if (kn->kn_filter == EVFILT_WRITE)
2495 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2496 	else
2497 		panic("filt_mqdetach");
2498 }
2499 
2500 static int
2501 filt_mqread(struct knote *kn, long hint)
2502 {
2503 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2504 
2505 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2506 	return (mq->mq_curmsgs != 0);
2507 }
2508 
2509 static int
2510 filt_mqwrite(struct knote *kn, long hint)
2511 {
2512 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2513 
2514 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2515 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2516 }
2517 
2518 static struct fileops mqueueops = {
2519 	.fo_read		= mqf_read,
2520 	.fo_write		= mqf_write,
2521 	.fo_truncate		= mqf_truncate,
2522 	.fo_ioctl		= mqf_ioctl,
2523 	.fo_poll		= mqf_poll,
2524 	.fo_kqfilter		= mqf_kqfilter,
2525 	.fo_stat		= mqf_stat,
2526 	.fo_close		= mqf_close
2527 };
2528 
2529 static struct vop_vector mqfs_vnodeops = {
2530 	.vop_default 		= &default_vnodeops,
2531 	.vop_access		= mqfs_access,
2532 	.vop_cachedlookup	= mqfs_lookup,
2533 	.vop_lookup		= vfs_cache_lookup,
2534 	.vop_reclaim		= mqfs_reclaim,
2535 	.vop_create		= mqfs_create,
2536 	.vop_remove		= mqfs_remove,
2537 	.vop_inactive		= mqfs_inactive,
2538 	.vop_open		= mqfs_open,
2539 	.vop_close		= mqfs_close,
2540 	.vop_getattr		= mqfs_getattr,
2541 	.vop_setattr		= mqfs_setattr,
2542 	.vop_read		= mqfs_read,
2543 	.vop_write		= VOP_EOPNOTSUPP,
2544 	.vop_readdir		= mqfs_readdir,
2545 	.vop_mkdir		= VOP_EOPNOTSUPP,
2546 	.vop_rmdir		= VOP_EOPNOTSUPP
2547 };
2548 
2549 static struct vfsops mqfs_vfsops = {
2550 	.vfs_init 		= mqfs_init,
2551 	.vfs_uninit		= mqfs_uninit,
2552 	.vfs_mount		= mqfs_mount,
2553 	.vfs_unmount		= mqfs_unmount,
2554 	.vfs_root		= mqfs_root,
2555 	.vfs_statfs		= mqfs_statfs,
2556 };
2557 
2558 static struct vfsconf mqueuefs_vfsconf = {
2559 	.vfc_version = VFS_VERSION,
2560 	.vfc_name = "mqueuefs",
2561 	.vfc_vfsops = &mqfs_vfsops,
2562 	.vfc_typenum = -1,
2563 	.vfc_flags = VFCF_SYNTHETIC
2564 };
2565 
2566 static struct syscall_helper_data mq_syscalls[] = {
2567 	SYSCALL_INIT_HELPER(kmq_open),
2568 	SYSCALL_INIT_HELPER(kmq_setattr),
2569 	SYSCALL_INIT_HELPER(kmq_timedsend),
2570 	SYSCALL_INIT_HELPER(kmq_timedreceive),
2571 	SYSCALL_INIT_HELPER(kmq_notify),
2572 	SYSCALL_INIT_HELPER(kmq_unlink),
2573 	SYSCALL_INIT_LAST
2574 };
2575 
2576 #ifdef COMPAT_FREEBSD32
2577 #include <compat/freebsd32/freebsd32.h>
2578 #include <compat/freebsd32/freebsd32_proto.h>
2579 #include <compat/freebsd32/freebsd32_syscall.h>
2580 #include <compat/freebsd32/freebsd32_util.h>
2581 
2582 static void
2583 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2584 {
2585 
2586 	to->mq_flags = from->mq_flags;
2587 	to->mq_maxmsg = from->mq_maxmsg;
2588 	to->mq_msgsize = from->mq_msgsize;
2589 	to->mq_curmsgs = from->mq_curmsgs;
2590 }
2591 
2592 static void
2593 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2594 {
2595 
2596 	to->mq_flags = from->mq_flags;
2597 	to->mq_maxmsg = from->mq_maxmsg;
2598 	to->mq_msgsize = from->mq_msgsize;
2599 	to->mq_curmsgs = from->mq_curmsgs;
2600 }
2601 
2602 int
2603 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2604 {
2605 	struct mq_attr attr;
2606 	struct mq_attr32 attr32;
2607 	int flags, error;
2608 
2609 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
2610 		return (EINVAL);
2611 	flags = FFLAGS(uap->flags);
2612 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2613 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2614 		if (error)
2615 			return (error);
2616 		mq_attr_from32(&attr32, &attr);
2617 	}
2618 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2619 	    uap->attr != NULL ? &attr : NULL));
2620 }
2621 
2622 int
2623 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2624 {
2625 	struct mq_attr attr, oattr;
2626 	struct mq_attr32 attr32, oattr32;
2627 	int error;
2628 
2629 	if (uap->attr != NULL) {
2630 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2631 		if (error != 0)
2632 			return (error);
2633 		mq_attr_from32(&attr32, &attr);
2634 	}
2635 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2636 	    &oattr);
2637 	if (error != 0)
2638 		return (error);
2639 	if (uap->oattr != NULL) {
2640 		mq_attr_to32(&oattr, &oattr32);
2641 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2642 	}
2643 	return (error);
2644 }
2645 
2646 int
2647 freebsd32_kmq_timedsend(struct thread *td,
2648     struct freebsd32_kmq_timedsend_args *uap)
2649 {
2650 	struct mqueue *mq;
2651 	struct file *fp;
2652 	struct timespec32 ets32;
2653 	struct timespec *abs_timeout, ets;
2654 	int error;
2655 	int waitok;
2656 
2657 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2658 	if (error)
2659 		return (error);
2660 	if (uap->abs_timeout != NULL) {
2661 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2662 		if (error != 0)
2663 			return (error);
2664 		CP(ets32, ets, tv_sec);
2665 		CP(ets32, ets, tv_nsec);
2666 		abs_timeout = &ets;
2667 	} else
2668 		abs_timeout = NULL;
2669 	waitok = !(fp->f_flag & O_NONBLOCK);
2670 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2671 		uap->msg_prio, waitok, abs_timeout);
2672 	fdrop(fp, td);
2673 	return (error);
2674 }
2675 
2676 int
2677 freebsd32_kmq_timedreceive(struct thread *td,
2678     struct freebsd32_kmq_timedreceive_args *uap)
2679 {
2680 	struct mqueue *mq;
2681 	struct file *fp;
2682 	struct timespec32 ets32;
2683 	struct timespec *abs_timeout, ets;
2684 	int error, waitok;
2685 
2686 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2687 	if (error)
2688 		return (error);
2689 	if (uap->abs_timeout != NULL) {
2690 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2691 		if (error != 0)
2692 			return (error);
2693 		CP(ets32, ets, tv_sec);
2694 		CP(ets32, ets, tv_nsec);
2695 		abs_timeout = &ets;
2696 	} else
2697 		abs_timeout = NULL;
2698 	waitok = !(fp->f_flag & O_NONBLOCK);
2699 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2700 		uap->msg_prio, waitok, abs_timeout);
2701 	fdrop(fp, td);
2702 	return (error);
2703 }
2704 
2705 static struct syscall_helper_data mq32_syscalls[] = {
2706 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2707 	SYSCALL32_INIT_HELPER(freebsd32_kmq_setattr),
2708 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedsend),
2709 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedreceive),
2710 	SYSCALL32_INIT_HELPER(kmq_notify),
2711 	SYSCALL32_INIT_HELPER(kmq_unlink),
2712 	SYSCALL_INIT_LAST
2713 };
2714 #endif
2715 
2716 static int
2717 mqinit(void)
2718 {
2719 	int error;
2720 
2721 	error = syscall_helper_register(mq_syscalls);
2722 	if (error != 0)
2723 		return (error);
2724 #ifdef COMPAT_FREEBSD32
2725 	error = syscall32_helper_register(mq32_syscalls);
2726 	if (error != 0)
2727 		return (error);
2728 #endif
2729 	return (0);
2730 }
2731 
2732 static int
2733 mqunload(void)
2734 {
2735 
2736 #ifdef COMPAT_FREEBSD32
2737 	syscall32_helper_unregister(mq32_syscalls);
2738 #endif
2739 	syscall_helper_unregister(mq_syscalls);
2740 	return (0);
2741 }
2742 
2743 static int
2744 mq_modload(struct module *module, int cmd, void *arg)
2745 {
2746 	int error = 0;
2747 
2748 	error = vfs_modevent(module, cmd, arg);
2749 	if (error != 0)
2750 		return (error);
2751 
2752 	switch (cmd) {
2753 	case MOD_LOAD:
2754 		error = mqinit();
2755 		if (error != 0)
2756 			mqunload();
2757 		break;
2758 	case MOD_UNLOAD:
2759 		error = mqunload();
2760 		break;
2761 	default:
2762 		break;
2763 	}
2764 	return (error);
2765 }
2766 
2767 static moduledata_t mqueuefs_mod = {
2768 	"mqueuefs",
2769 	mq_modload,
2770 	&mqueuefs_vfsconf
2771 };
2772 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2773 MODULE_VERSION(mqueuefs, 1);
2774