xref: /freebsd/sys/kern/uipc_mqueue.c (revision c6ec7d31830ab1c80edae95ad5e4b9dba10c47ac)
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/capability.h>
56 #include <sys/dirent.h>
57 #include <sys/event.h>
58 #include <sys/eventhandler.h>
59 #include <sys/fcntl.h>
60 #include <sys/file.h>
61 #include <sys/filedesc.h>
62 #include <sys/lock.h>
63 #include <sys/malloc.h>
64 #include <sys/module.h>
65 #include <sys/mount.h>
66 #include <sys/mqueue.h>
67 #include <sys/mutex.h>
68 #include <sys/namei.h>
69 #include <sys/posix4.h>
70 #include <sys/poll.h>
71 #include <sys/priv.h>
72 #include <sys/proc.h>
73 #include <sys/queue.h>
74 #include <sys/sysproto.h>
75 #include <sys/stat.h>
76 #include <sys/syscall.h>
77 #include <sys/syscallsubr.h>
78 #include <sys/sysent.h>
79 #include <sys/sx.h>
80 #include <sys/sysctl.h>
81 #include <sys/taskqueue.h>
82 #include <sys/unistd.h>
83 #include <sys/vnode.h>
84 #include <machine/atomic.h>
85 
86 FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
87 
88 /*
89  * Limits and constants
90  */
91 #define	MQFS_NAMELEN		NAME_MAX
92 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
93 
94 /* node types */
95 typedef enum {
96 	mqfstype_none = 0,
97 	mqfstype_root,
98 	mqfstype_dir,
99 	mqfstype_this,
100 	mqfstype_parent,
101 	mqfstype_file,
102 	mqfstype_symlink,
103 } mqfs_type_t;
104 
105 struct mqfs_node;
106 
107 /*
108  * mqfs_info: describes a mqfs instance
109  */
110 struct mqfs_info {
111 	struct sx		mi_lock;
112 	struct mqfs_node	*mi_root;
113 	struct unrhdr		*mi_unrhdr;
114 };
115 
116 struct mqfs_vdata {
117 	LIST_ENTRY(mqfs_vdata)	mv_link;
118 	struct mqfs_node	*mv_node;
119 	struct vnode		*mv_vnode;
120 	struct task		mv_task;
121 };
122 
123 /*
124  * mqfs_node: describes a node (file or directory) within a mqfs
125  */
126 struct mqfs_node {
127 	char			mn_name[MQFS_NAMELEN+1];
128 	struct mqfs_info	*mn_info;
129 	struct mqfs_node	*mn_parent;
130 	LIST_HEAD(,mqfs_node)	mn_children;
131 	LIST_ENTRY(mqfs_node)	mn_sibling;
132 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
133 	int			mn_refcount;
134 	mqfs_type_t		mn_type;
135 	int			mn_deleted;
136 	uint32_t		mn_fileno;
137 	void			*mn_data;
138 	struct timespec		mn_birth;
139 	struct timespec		mn_ctime;
140 	struct timespec		mn_atime;
141 	struct timespec		mn_mtime;
142 	uid_t			mn_uid;
143 	gid_t			mn_gid;
144 	int			mn_mode;
145 };
146 
147 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
148 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
149 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
150 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
151 				(fp)->f_data)->mn_data))
152 
153 TAILQ_HEAD(msgq, mqueue_msg);
154 
155 struct mqueue;
156 
157 struct mqueue_notifier {
158 	LIST_ENTRY(mqueue_notifier)	nt_link;
159 	struct sigevent			nt_sigev;
160 	ksiginfo_t			nt_ksi;
161 	struct proc			*nt_proc;
162 };
163 
164 struct mqueue {
165 	struct mtx	mq_mutex;
166 	int		mq_flags;
167 	long		mq_maxmsg;
168 	long		mq_msgsize;
169 	long		mq_curmsgs;
170 	long		mq_totalbytes;
171 	struct msgq	mq_msgq;
172 	int		mq_receivers;
173 	int		mq_senders;
174 	struct selinfo	mq_rsel;
175 	struct selinfo	mq_wsel;
176 	struct mqueue_notifier	*mq_notifier;
177 };
178 
179 #define	MQ_RSEL		0x01
180 #define	MQ_WSEL		0x02
181 
182 struct mqueue_msg {
183 	TAILQ_ENTRY(mqueue_msg)	msg_link;
184 	unsigned int	msg_prio;
185 	unsigned int	msg_size;
186 	/* following real data... */
187 };
188 
189 static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
190 	"POSIX real time message queue");
191 
192 static int	default_maxmsg  = 10;
193 static int	default_msgsize = 1024;
194 
195 static int	maxmsg = 100;
196 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
197     &maxmsg, 0, "Default maximum messages in queue");
198 static int	maxmsgsize = 16384;
199 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
200     &maxmsgsize, 0, "Default maximum message size");
201 static int	maxmq = 100;
202 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
203     &maxmq, 0, "maximum message queues");
204 static int	curmq = 0;
205 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
206     &curmq, 0, "current message queue number");
207 static int	unloadable = 0;
208 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
209 
210 static eventhandler_tag exit_tag;
211 
212 /* Only one instance per-system */
213 static struct mqfs_info		mqfs_data;
214 static uma_zone_t		mqnode_zone;
215 static uma_zone_t		mqueue_zone;
216 static uma_zone_t		mvdata_zone;
217 static uma_zone_t		mqnoti_zone;
218 static struct vop_vector	mqfs_vnodeops;
219 static struct fileops		mqueueops;
220 
221 /*
222  * Directory structure construction and manipulation
223  */
224 #ifdef notyet
225 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
226 	const char *name, int namelen, struct ucred *cred, int mode);
227 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
228 	const char *name, int namelen, struct ucred *cred, int mode);
229 #endif
230 
231 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
232 	const char *name, int namelen, struct ucred *cred, int mode);
233 static int	mqfs_destroy(struct mqfs_node *mn);
234 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
235 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
236 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
237 
238 /*
239  * Message queue construction and maniplation
240  */
241 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
242 static void	mqueue_free(struct mqueue *mq);
243 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
244 			size_t msg_len, unsigned msg_prio, int waitok,
245 			const struct timespec *abs_timeout);
246 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
247 			size_t msg_len, unsigned *msg_prio, int waitok,
248 			const struct timespec *abs_timeout);
249 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
250 			int timo);
251 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
252 			int timo);
253 static void	mqueue_send_notification(struct mqueue *mq);
254 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
255 static void	mq_proc_exit(void *arg, struct proc *p);
256 
257 /*
258  * kqueue filters
259  */
260 static void	filt_mqdetach(struct knote *kn);
261 static int	filt_mqread(struct knote *kn, long hint);
262 static int	filt_mqwrite(struct knote *kn, long hint);
263 
264 struct filterops mq_rfiltops = {
265 	.f_isfd = 1,
266 	.f_detach = filt_mqdetach,
267 	.f_event = filt_mqread,
268 };
269 struct filterops mq_wfiltops = {
270 	.f_isfd = 1,
271 	.f_detach = filt_mqdetach,
272 	.f_event = filt_mqwrite,
273 };
274 
275 /*
276  * Initialize fileno bitmap
277  */
278 static void
279 mqfs_fileno_init(struct mqfs_info *mi)
280 {
281 	struct unrhdr *up;
282 
283 	up = new_unrhdr(1, INT_MAX, NULL);
284 	mi->mi_unrhdr = up;
285 }
286 
287 /*
288  * Tear down fileno bitmap
289  */
290 static void
291 mqfs_fileno_uninit(struct mqfs_info *mi)
292 {
293 	struct unrhdr *up;
294 
295 	up = mi->mi_unrhdr;
296 	mi->mi_unrhdr = NULL;
297 	delete_unrhdr(up);
298 }
299 
300 /*
301  * Allocate a file number
302  */
303 static void
304 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
305 {
306 	/* make sure our parent has a file number */
307 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
308 		mqfs_fileno_alloc(mi, mn->mn_parent);
309 
310 	switch (mn->mn_type) {
311 	case mqfstype_root:
312 	case mqfstype_dir:
313 	case mqfstype_file:
314 	case mqfstype_symlink:
315 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
316 		break;
317 	case mqfstype_this:
318 		KASSERT(mn->mn_parent != NULL,
319 		    ("mqfstype_this node has no parent"));
320 		mn->mn_fileno = mn->mn_parent->mn_fileno;
321 		break;
322 	case mqfstype_parent:
323 		KASSERT(mn->mn_parent != NULL,
324 		    ("mqfstype_parent node has no parent"));
325 		if (mn->mn_parent == mi->mi_root) {
326 			mn->mn_fileno = mn->mn_parent->mn_fileno;
327 			break;
328 		}
329 		KASSERT(mn->mn_parent->mn_parent != NULL,
330 		    ("mqfstype_parent node has no grandparent"));
331 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
332 		break;
333 	default:
334 		KASSERT(0,
335 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
336 			mn->mn_type));
337 		break;
338 	}
339 }
340 
341 /*
342  * Release a file number
343  */
344 static void
345 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
346 {
347 	switch (mn->mn_type) {
348 	case mqfstype_root:
349 	case mqfstype_dir:
350 	case mqfstype_file:
351 	case mqfstype_symlink:
352 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
353 		break;
354 	case mqfstype_this:
355 	case mqfstype_parent:
356 		/* ignore these, as they don't "own" their file number */
357 		break;
358 	default:
359 		KASSERT(0,
360 		    ("mqfs_fileno_free() called for unknown type node: %d",
361 			mn->mn_type));
362 		break;
363 	}
364 }
365 
366 static __inline struct mqfs_node *
367 mqnode_alloc(void)
368 {
369 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
370 }
371 
372 static __inline void
373 mqnode_free(struct mqfs_node *node)
374 {
375 	uma_zfree(mqnode_zone, node);
376 }
377 
378 static __inline void
379 mqnode_addref(struct mqfs_node *node)
380 {
381 	atomic_fetchadd_int(&node->mn_refcount, 1);
382 }
383 
384 static __inline void
385 mqnode_release(struct mqfs_node *node)
386 {
387 	struct mqfs_info *mqfs;
388 	int old, exp;
389 
390 	mqfs = node->mn_info;
391 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
392 	if (node->mn_type == mqfstype_dir ||
393 	    node->mn_type == mqfstype_root)
394 		exp = 3; /* include . and .. */
395 	else
396 		exp = 1;
397 	if (old == exp) {
398 		int locked = sx_xlocked(&mqfs->mi_lock);
399 		if (!locked)
400 			sx_xlock(&mqfs->mi_lock);
401 		mqfs_destroy(node);
402 		if (!locked)
403 			sx_xunlock(&mqfs->mi_lock);
404 	}
405 }
406 
407 /*
408  * Add a node to a directory
409  */
410 static int
411 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
412 {
413 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
414 	KASSERT(parent->mn_info != NULL,
415 	    ("%s(): parent has no mn_info", __func__));
416 	KASSERT(parent->mn_type == mqfstype_dir ||
417 	    parent->mn_type == mqfstype_root,
418 	    ("%s(): parent is not a directory", __func__));
419 
420 	node->mn_info = parent->mn_info;
421 	node->mn_parent = parent;
422 	LIST_INIT(&node->mn_children);
423 	LIST_INIT(&node->mn_vnodes);
424 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
425 	mqnode_addref(parent);
426 	return (0);
427 }
428 
429 static struct mqfs_node *
430 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
431 	int nodetype)
432 {
433 	struct mqfs_node *node;
434 
435 	node = mqnode_alloc();
436 	strncpy(node->mn_name, name, namelen);
437 	node->mn_type = nodetype;
438 	node->mn_refcount = 1;
439 	vfs_timestamp(&node->mn_birth);
440 	node->mn_ctime = node->mn_atime = node->mn_mtime
441 		= node->mn_birth;
442 	node->mn_uid = cred->cr_uid;
443 	node->mn_gid = cred->cr_gid;
444 	node->mn_mode = mode;
445 	return (node);
446 }
447 
448 /*
449  * Create a file
450  */
451 static struct mqfs_node *
452 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
453 	struct ucred *cred, int mode)
454 {
455 	struct mqfs_node *node;
456 
457 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
458 	if (mqfs_add_node(parent, node) != 0) {
459 		mqnode_free(node);
460 		return (NULL);
461 	}
462 	return (node);
463 }
464 
465 /*
466  * Add . and .. to a directory
467  */
468 static int
469 mqfs_fixup_dir(struct mqfs_node *parent)
470 {
471 	struct mqfs_node *dir;
472 
473 	dir = mqnode_alloc();
474 	dir->mn_name[0] = '.';
475 	dir->mn_type = mqfstype_this;
476 	dir->mn_refcount = 1;
477 	if (mqfs_add_node(parent, dir) != 0) {
478 		mqnode_free(dir);
479 		return (-1);
480 	}
481 
482 	dir = mqnode_alloc();
483 	dir->mn_name[0] = dir->mn_name[1] = '.';
484 	dir->mn_type = mqfstype_parent;
485 	dir->mn_refcount = 1;
486 
487 	if (mqfs_add_node(parent, dir) != 0) {
488 		mqnode_free(dir);
489 		return (-1);
490 	}
491 
492 	return (0);
493 }
494 
495 #ifdef notyet
496 
497 /*
498  * Create a directory
499  */
500 static struct mqfs_node *
501 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
502 	struct ucred *cred, int mode)
503 {
504 	struct mqfs_node *node;
505 
506 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
507 	if (mqfs_add_node(parent, node) != 0) {
508 		mqnode_free(node);
509 		return (NULL);
510 	}
511 
512 	if (mqfs_fixup_dir(node) != 0) {
513 		mqfs_destroy(node);
514 		return (NULL);
515 	}
516 	return (node);
517 }
518 
519 /*
520  * Create a symlink
521  */
522 static struct mqfs_node *
523 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
524 	struct ucred *cred, int mode)
525 {
526 	struct mqfs_node *node;
527 
528 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
529 	if (mqfs_add_node(parent, node) != 0) {
530 		mqnode_free(node);
531 		return (NULL);
532 	}
533 	return (node);
534 }
535 
536 #endif
537 
538 /*
539  * Destroy a node or a tree of nodes
540  */
541 static int
542 mqfs_destroy(struct mqfs_node *node)
543 {
544 	struct mqfs_node *parent;
545 
546 	KASSERT(node != NULL,
547 	    ("%s(): node is NULL", __func__));
548 	KASSERT(node->mn_info != NULL,
549 	    ("%s(): node has no mn_info", __func__));
550 
551 	/* destroy children */
552 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
553 		while (! LIST_EMPTY(&node->mn_children))
554 			mqfs_destroy(LIST_FIRST(&node->mn_children));
555 
556 	/* unlink from parent */
557 	if ((parent = node->mn_parent) != NULL) {
558 		KASSERT(parent->mn_info == node->mn_info,
559 		    ("%s(): parent has different mn_info", __func__));
560 		LIST_REMOVE(node, mn_sibling);
561 	}
562 
563 	if (node->mn_fileno != 0)
564 		mqfs_fileno_free(node->mn_info, node);
565 	if (node->mn_data != NULL)
566 		mqueue_free(node->mn_data);
567 	mqnode_free(node);
568 	return (0);
569 }
570 
571 /*
572  * Mount a mqfs instance
573  */
574 static int
575 mqfs_mount(struct mount *mp)
576 {
577 	struct statfs *sbp;
578 
579 	if (mp->mnt_flag & MNT_UPDATE)
580 		return (EOPNOTSUPP);
581 
582 	mp->mnt_data = &mqfs_data;
583 	MNT_ILOCK(mp);
584 	mp->mnt_flag |= MNT_LOCAL;
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);
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);
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 	seldrain(&mq->mq_rsel);
1565 	seldrain(&mq->mq_wsel);
1566 	knlist_destroy(&mq->mq_rsel.si_note);
1567 	knlist_destroy(&mq->mq_wsel.si_note);
1568 	uma_zfree(mqueue_zone, mq);
1569 	atomic_add_int(&curmq, -1);
1570 }
1571 
1572 /*
1573  * Load a message from user space
1574  */
1575 static struct mqueue_msg *
1576 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1577 {
1578 	struct mqueue_msg *msg;
1579 	size_t len;
1580 	int error;
1581 
1582 	len = sizeof(struct mqueue_msg) + msg_size;
1583 	msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
1584 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1585 	    msg_size);
1586 	if (error) {
1587 		free(msg, M_MQUEUEDATA);
1588 		msg = NULL;
1589 	} else {
1590 		msg->msg_size = msg_size;
1591 		msg->msg_prio = msg_prio;
1592 	}
1593 	return (msg);
1594 }
1595 
1596 /*
1597  * Save a message to user space
1598  */
1599 static int
1600 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1601 {
1602 	int error;
1603 
1604 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1605 		msg->msg_size);
1606 	if (error == 0 && msg_prio != NULL)
1607 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1608 	return (error);
1609 }
1610 
1611 /*
1612  * Free a message's memory
1613  */
1614 static __inline void
1615 mqueue_freemsg(struct mqueue_msg *msg)
1616 {
1617 	free(msg, M_MQUEUEDATA);
1618 }
1619 
1620 /*
1621  * Send a message. if waitok is false, thread will not be
1622  * blocked if there is no data in queue, otherwise, absolute
1623  * time will be checked.
1624  */
1625 int
1626 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1627 	size_t msg_len, unsigned msg_prio, int waitok,
1628 	const struct timespec *abs_timeout)
1629 {
1630 	struct mqueue_msg *msg;
1631 	struct timespec ts, ts2;
1632 	struct timeval tv;
1633 	int error;
1634 
1635 	if (msg_prio >= MQ_PRIO_MAX)
1636 		return (EINVAL);
1637 	if (msg_len > mq->mq_msgsize)
1638 		return (EMSGSIZE);
1639 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1640 	if (msg == NULL)
1641 		return (EFAULT);
1642 
1643 	/* O_NONBLOCK case */
1644 	if (!waitok) {
1645 		error = _mqueue_send(mq, msg, -1);
1646 		if (error)
1647 			goto bad;
1648 		return (0);
1649 	}
1650 
1651 	/* we allow a null timeout (wait forever) */
1652 	if (abs_timeout == NULL) {
1653 		error = _mqueue_send(mq, msg, 0);
1654 		if (error)
1655 			goto bad;
1656 		return (0);
1657 	}
1658 
1659 	/* send it before checking time */
1660 	error = _mqueue_send(mq, msg, -1);
1661 	if (error == 0)
1662 		return (0);
1663 
1664 	if (error != EAGAIN)
1665 		goto bad;
1666 
1667 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1668 		error = EINVAL;
1669 		goto bad;
1670 	}
1671 	for (;;) {
1672 		ts2 = *abs_timeout;
1673 		getnanotime(&ts);
1674 		timespecsub(&ts2, &ts);
1675 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1676 			error = ETIMEDOUT;
1677 			break;
1678 		}
1679 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1680 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1681 		if (error != ETIMEDOUT)
1682 			break;
1683 	}
1684 	if (error == 0)
1685 		return (0);
1686 bad:
1687 	mqueue_freemsg(msg);
1688 	return (error);
1689 }
1690 
1691 /*
1692  * Common routine to send a message
1693  */
1694 static int
1695 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1696 {
1697 	struct mqueue_msg *msg2;
1698 	int error = 0;
1699 
1700 	mtx_lock(&mq->mq_mutex);
1701 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1702 		if (timo < 0) {
1703 			mtx_unlock(&mq->mq_mutex);
1704 			return (EAGAIN);
1705 		}
1706 		mq->mq_senders++;
1707 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1708 			    PCATCH, "mqsend", timo);
1709 		mq->mq_senders--;
1710 		if (error == EAGAIN)
1711 			error = ETIMEDOUT;
1712 	}
1713 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1714 		mtx_unlock(&mq->mq_mutex);
1715 		return (error);
1716 	}
1717 	error = 0;
1718 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1719 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1720 	} else {
1721 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1722 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1723 		} else {
1724 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1725 				if (msg2->msg_prio < msg->msg_prio)
1726 					break;
1727 			}
1728 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1729 		}
1730 	}
1731 	mq->mq_curmsgs++;
1732 	mq->mq_totalbytes += msg->msg_size;
1733 	if (mq->mq_receivers)
1734 		wakeup_one(&mq->mq_receivers);
1735 	else if (mq->mq_notifier != NULL)
1736 		mqueue_send_notification(mq);
1737 	if (mq->mq_flags & MQ_RSEL) {
1738 		mq->mq_flags &= ~MQ_RSEL;
1739 		selwakeup(&mq->mq_rsel);
1740 	}
1741 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1742 	mtx_unlock(&mq->mq_mutex);
1743 	return (0);
1744 }
1745 
1746 /*
1747  * Send realtime a signal to process which registered itself
1748  * successfully by mq_notify.
1749  */
1750 static void
1751 mqueue_send_notification(struct mqueue *mq)
1752 {
1753 	struct mqueue_notifier *nt;
1754 	struct thread *td;
1755 	struct proc *p;
1756 	int error;
1757 
1758 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1759 	nt = mq->mq_notifier;
1760 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1761 		p = nt->nt_proc;
1762 		error = sigev_findtd(p, &nt->nt_sigev, &td);
1763 		if (error) {
1764 			mq->mq_notifier = NULL;
1765 			return;
1766 		}
1767 		if (!KSI_ONQ(&nt->nt_ksi)) {
1768 			ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
1769 			tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
1770 		}
1771 		PROC_UNLOCK(p);
1772 	}
1773 	mq->mq_notifier = NULL;
1774 }
1775 
1776 /*
1777  * Get a message. if waitok is false, thread will not be
1778  * blocked if there is no data in queue, otherwise, absolute
1779  * time will be checked.
1780  */
1781 int
1782 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1783 	size_t msg_len, unsigned *msg_prio, int waitok,
1784 	const struct timespec *abs_timeout)
1785 {
1786 	struct mqueue_msg *msg;
1787 	struct timespec ts, ts2;
1788 	struct timeval tv;
1789 	int error;
1790 
1791 	if (msg_len < mq->mq_msgsize)
1792 		return (EMSGSIZE);
1793 
1794 	/* O_NONBLOCK case */
1795 	if (!waitok) {
1796 		error = _mqueue_recv(mq, &msg, -1);
1797 		if (error)
1798 			return (error);
1799 		goto received;
1800 	}
1801 
1802 	/* we allow a null timeout (wait forever). */
1803 	if (abs_timeout == NULL) {
1804 		error = _mqueue_recv(mq, &msg, 0);
1805 		if (error)
1806 			return (error);
1807 		goto received;
1808 	}
1809 
1810 	/* try to get a message before checking time */
1811 	error = _mqueue_recv(mq, &msg, -1);
1812 	if (error == 0)
1813 		goto received;
1814 
1815 	if (error != EAGAIN)
1816 		return (error);
1817 
1818 	if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
1819 		error = EINVAL;
1820 		return (error);
1821 	}
1822 
1823 	for (;;) {
1824 		ts2 = *abs_timeout;
1825 		getnanotime(&ts);
1826 		timespecsub(&ts2, &ts);
1827 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1828 			error = ETIMEDOUT;
1829 			return (error);
1830 		}
1831 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1832 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1833 		if (error == 0)
1834 			break;
1835 		if (error != ETIMEDOUT)
1836 			return (error);
1837 	}
1838 
1839 received:
1840 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1841 	if (error == 0) {
1842 		curthread->td_retval[0] = msg->msg_size;
1843 		curthread->td_retval[1] = 0;
1844 	}
1845 	mqueue_freemsg(msg);
1846 	return (error);
1847 }
1848 
1849 /*
1850  * Common routine to receive a message
1851  */
1852 static int
1853 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1854 {
1855 	int error = 0;
1856 
1857 	mtx_lock(&mq->mq_mutex);
1858 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1859 		if (timo < 0) {
1860 			mtx_unlock(&mq->mq_mutex);
1861 			return (EAGAIN);
1862 		}
1863 		mq->mq_receivers++;
1864 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1865 			    PCATCH, "mqrecv", timo);
1866 		mq->mq_receivers--;
1867 		if (error == EAGAIN)
1868 			error = ETIMEDOUT;
1869 	}
1870 	if (*msg != NULL) {
1871 		error = 0;
1872 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1873 		mq->mq_curmsgs--;
1874 		mq->mq_totalbytes -= (*msg)->msg_size;
1875 		if (mq->mq_senders)
1876 			wakeup_one(&mq->mq_senders);
1877 		if (mq->mq_flags & MQ_WSEL) {
1878 			mq->mq_flags &= ~MQ_WSEL;
1879 			selwakeup(&mq->mq_wsel);
1880 		}
1881 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1882 	}
1883 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1884 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1885 		mqueue_send_notification(mq);
1886 	}
1887 	mtx_unlock(&mq->mq_mutex);
1888 	return (error);
1889 }
1890 
1891 static __inline struct mqueue_notifier *
1892 notifier_alloc(void)
1893 {
1894 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1895 }
1896 
1897 static __inline void
1898 notifier_free(struct mqueue_notifier *p)
1899 {
1900 	uma_zfree(mqnoti_zone, p);
1901 }
1902 
1903 static struct mqueue_notifier *
1904 notifier_search(struct proc *p, int fd)
1905 {
1906 	struct mqueue_notifier *nt;
1907 
1908 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1909 		if (nt->nt_ksi.ksi_mqd == fd)
1910 			break;
1911 	}
1912 	return (nt);
1913 }
1914 
1915 static __inline void
1916 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1917 {
1918 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1919 }
1920 
1921 static __inline void
1922 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1923 {
1924 	LIST_REMOVE(nt, nt_link);
1925 	notifier_free(nt);
1926 }
1927 
1928 static void
1929 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1930 {
1931 	struct mqueue_notifier *nt;
1932 
1933 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1934 	PROC_LOCK(p);
1935 	nt = notifier_search(p, fd);
1936 	if (nt != NULL) {
1937 		if (mq->mq_notifier == nt)
1938 			mq->mq_notifier = NULL;
1939 		sigqueue_take(&nt->nt_ksi);
1940 		notifier_delete(p, nt);
1941 	}
1942 	PROC_UNLOCK(p);
1943 }
1944 
1945 static int
1946 kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
1947     const struct mq_attr *attr)
1948 {
1949 	char path[MQFS_NAMELEN + 1];
1950 	struct mqfs_node *pn;
1951 	struct filedesc *fdp;
1952 	struct file *fp;
1953 	struct mqueue *mq;
1954 	int fd, error, len, cmode;
1955 
1956 	fdp = td->td_proc->p_fd;
1957 	cmode = (((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1958 	mq = NULL;
1959 	if ((flags & O_CREAT) != 0 && attr != NULL) {
1960 		if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
1961 			return (EINVAL);
1962 		if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
1963 			return (EINVAL);
1964 	}
1965 
1966 	error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
1967         if (error)
1968 		return (error);
1969 
1970 	/*
1971 	 * The first character of name must be a slash  (/) character
1972 	 * and the remaining characters of name cannot include any slash
1973 	 * characters.
1974 	 */
1975 	len = strlen(path);
1976 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
1977 		return (EINVAL);
1978 
1979 	error = falloc(td, &fp, &fd, 0);
1980 	if (error)
1981 		return (error);
1982 
1983 	sx_xlock(&mqfs_data.mi_lock);
1984 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1985 	if (pn == NULL) {
1986 		if (!(flags & O_CREAT)) {
1987 			error = ENOENT;
1988 		} else {
1989 			mq = mqueue_alloc(attr);
1990 			if (mq == NULL) {
1991 				error = ENFILE;
1992 			} else {
1993 				pn = mqfs_create_file(mqfs_data.mi_root,
1994 				         path + 1, len - 1, td->td_ucred,
1995 					 cmode);
1996 				if (pn == NULL) {
1997 					error = ENOSPC;
1998 					mqueue_free(mq);
1999 				}
2000 			}
2001 		}
2002 
2003 		if (error == 0) {
2004 			pn->mn_data = mq;
2005 		}
2006 	} else {
2007 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
2008 			error = EEXIST;
2009 		} else {
2010 			accmode_t accmode = 0;
2011 
2012 			if (flags & FREAD)
2013 				accmode |= VREAD;
2014 			if (flags & FWRITE)
2015 				accmode |= VWRITE;
2016 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
2017 				    pn->mn_gid, accmode, td->td_ucred, NULL);
2018 		}
2019 	}
2020 
2021 	if (error) {
2022 		sx_xunlock(&mqfs_data.mi_lock);
2023 		fdclose(fdp, fp, fd, td);
2024 		fdrop(fp, td);
2025 		return (error);
2026 	}
2027 
2028 	mqnode_addref(pn);
2029 	sx_xunlock(&mqfs_data.mi_lock);
2030 
2031 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
2032 	    &mqueueops);
2033 
2034 	FILEDESC_XLOCK(fdp);
2035 	if (fdp->fd_ofiles[fd] == fp)
2036 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2037 	FILEDESC_XUNLOCK(fdp);
2038 	td->td_retval[0] = fd;
2039 	fdrop(fp, td);
2040 	return (0);
2041 }
2042 
2043 /*
2044  * Syscall to open a message queue.
2045  */
2046 int
2047 sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
2048 {
2049 	struct mq_attr attr;
2050 	int flags, error;
2051 
2052 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
2053 		return (EINVAL);
2054 	flags = FFLAGS(uap->flags);
2055 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2056 		error = copyin(uap->attr, &attr, sizeof(attr));
2057 		if (error)
2058 			return (error);
2059 	}
2060 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2061 	    uap->attr != NULL ? &attr : NULL));
2062 }
2063 
2064 /*
2065  * Syscall to unlink a message queue.
2066  */
2067 int
2068 sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2069 {
2070 	char path[MQFS_NAMELEN+1];
2071 	struct mqfs_node *pn;
2072 	int error, len;
2073 
2074 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2075         if (error)
2076 		return (error);
2077 
2078 	len = strlen(path);
2079 	if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
2080 		return (EINVAL);
2081 
2082 	sx_xlock(&mqfs_data.mi_lock);
2083 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2084 	if (pn != NULL)
2085 		error = do_unlink(pn, td->td_ucred);
2086 	else
2087 		error = ENOENT;
2088 	sx_xunlock(&mqfs_data.mi_lock);
2089 	return (error);
2090 }
2091 
2092 typedef int (*_fgetf)(struct thread *, int, cap_rights_t, struct file **);
2093 
2094 /*
2095  * Get message queue by giving file slot
2096  */
2097 static int
2098 _getmq(struct thread *td, int fd, cap_rights_t rights, _fgetf func,
2099        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2100 {
2101 	struct mqfs_node *pn;
2102 	int error;
2103 
2104 	error = func(td, fd, rights, fpp);
2105 	if (error)
2106 		return (error);
2107 	if (&mqueueops != (*fpp)->f_ops) {
2108 		fdrop(*fpp, td);
2109 		return (EBADF);
2110 	}
2111 	pn = (*fpp)->f_data;
2112 	if (ppn)
2113 		*ppn = pn;
2114 	if (pmq)
2115 		*pmq = pn->mn_data;
2116 	return (0);
2117 }
2118 
2119 static __inline int
2120 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2121 	struct mqueue **pmq)
2122 {
2123 	return _getmq(td, fd, CAP_POLL_EVENT, fget, fpp, ppn, pmq);
2124 }
2125 
2126 static __inline int
2127 getmq_read(struct thread *td, int fd, struct file **fpp,
2128 	 struct mqfs_node **ppn, struct mqueue **pmq)
2129 {
2130 	return _getmq(td, fd, CAP_READ, fget_read, fpp, ppn, pmq);
2131 }
2132 
2133 static __inline int
2134 getmq_write(struct thread *td, int fd, struct file **fpp,
2135 	struct mqfs_node **ppn, struct mqueue **pmq)
2136 {
2137 	return _getmq(td, fd, CAP_WRITE, fget_write, fpp, ppn, pmq);
2138 }
2139 
2140 static int
2141 kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
2142     struct mq_attr *oattr)
2143 {
2144 	struct mqueue *mq;
2145 	struct file *fp;
2146 	u_int oflag, flag;
2147 	int error;
2148 
2149 	if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
2150 		return (EINVAL);
2151 	error = getmq(td, mqd, &fp, NULL, &mq);
2152 	if (error)
2153 		return (error);
2154 	oattr->mq_maxmsg  = mq->mq_maxmsg;
2155 	oattr->mq_msgsize = mq->mq_msgsize;
2156 	oattr->mq_curmsgs = mq->mq_curmsgs;
2157 	if (attr != NULL) {
2158 		do {
2159 			oflag = flag = fp->f_flag;
2160 			flag &= ~O_NONBLOCK;
2161 			flag |= (attr->mq_flags & O_NONBLOCK);
2162 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2163 	} else
2164 		oflag = fp->f_flag;
2165 	oattr->mq_flags = (O_NONBLOCK & oflag);
2166 	fdrop(fp, td);
2167 	return (error);
2168 }
2169 
2170 int
2171 sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2172 {
2173 	struct mq_attr attr, oattr;
2174 	int error;
2175 
2176 	if (uap->attr != NULL) {
2177 		error = copyin(uap->attr, &attr, sizeof(attr));
2178 		if (error != 0)
2179 			return (error);
2180 	}
2181 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2182 	    &oattr);
2183 	if (error != 0)
2184 		return (error);
2185 	if (uap->oattr != NULL)
2186 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2187 	return (error);
2188 }
2189 
2190 int
2191 sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2192 {
2193 	struct mqueue *mq;
2194 	struct file *fp;
2195 	struct timespec *abs_timeout, ets;
2196 	int error;
2197 	int waitok;
2198 
2199 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2200 	if (error)
2201 		return (error);
2202 	if (uap->abs_timeout != NULL) {
2203 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2204 		if (error != 0)
2205 			return (error);
2206 		abs_timeout = &ets;
2207 	} else
2208 		abs_timeout = NULL;
2209 	waitok = !(fp->f_flag & O_NONBLOCK);
2210 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2211 		uap->msg_prio, waitok, abs_timeout);
2212 	fdrop(fp, td);
2213 	return (error);
2214 }
2215 
2216 int
2217 sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2218 {
2219 	struct mqueue *mq;
2220 	struct file *fp;
2221 	struct timespec *abs_timeout, ets;
2222 	int error, waitok;
2223 
2224 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2225 	if (error)
2226 		return (error);
2227 	if (uap->abs_timeout != NULL) {
2228 		error = copyin(uap->abs_timeout, &ets, sizeof(ets));
2229 		if (error != 0)
2230 			return (error);
2231 		abs_timeout = &ets;
2232 	} else
2233 		abs_timeout = NULL;
2234 	waitok = !(fp->f_flag & O_NONBLOCK);
2235 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2236 		uap->msg_prio, waitok, abs_timeout);
2237 	fdrop(fp, td);
2238 	return (error);
2239 }
2240 
2241 int
2242 sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2243 {
2244 	struct sigevent ev;
2245 	struct filedesc *fdp;
2246 	struct proc *p;
2247 	struct mqueue *mq;
2248 	struct file *fp, *fp2;
2249 	struct mqueue_notifier *nt, *newnt = NULL;
2250 	int error;
2251 
2252 	p = td->td_proc;
2253 	fdp = td->td_proc->p_fd;
2254 	if (uap->sigev) {
2255 		error = copyin(uap->sigev, &ev, sizeof(ev));
2256 		if (error)
2257 			return (error);
2258 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2259 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2260 		    ev.sigev_notify != SIGEV_NONE)
2261 			return (EINVAL);
2262 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2263 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2264 			!_SIG_VALID(ev.sigev_signo))
2265 			return (EINVAL);
2266 	}
2267 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2268 	if (error)
2269 		return (error);
2270 again:
2271 	FILEDESC_SLOCK(fdp);
2272 	fp2 = fget_locked(fdp, uap->mqd);
2273 	if (fp2 == NULL) {
2274 		FILEDESC_SUNLOCK(fdp);
2275 		error = EBADF;
2276 		goto out;
2277 	}
2278 	error = cap_funwrap(fp2, CAP_POLL_EVENT, &fp2);
2279 	if (error) {
2280 		FILEDESC_SUNLOCK(fdp);
2281 		goto out;
2282 	}
2283 	if (fp2 != fp) {
2284 		FILEDESC_SUNLOCK(fdp);
2285 		error = EBADF;
2286 		goto out;
2287 	}
2288 	mtx_lock(&mq->mq_mutex);
2289 	FILEDESC_SUNLOCK(fdp);
2290 	if (uap->sigev != NULL) {
2291 		if (mq->mq_notifier != NULL) {
2292 			error = EBUSY;
2293 		} else {
2294 			PROC_LOCK(p);
2295 			nt = notifier_search(p, uap->mqd);
2296 			if (nt == NULL) {
2297 				if (newnt == NULL) {
2298 					PROC_UNLOCK(p);
2299 					mtx_unlock(&mq->mq_mutex);
2300 					newnt = notifier_alloc();
2301 					goto again;
2302 				}
2303 			}
2304 
2305 			if (nt != NULL) {
2306 				sigqueue_take(&nt->nt_ksi);
2307 				if (newnt != NULL) {
2308 					notifier_free(newnt);
2309 					newnt = NULL;
2310 				}
2311 			} else {
2312 				nt = newnt;
2313 				newnt = NULL;
2314 				ksiginfo_init(&nt->nt_ksi);
2315 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2316 				nt->nt_ksi.ksi_code = SI_MESGQ;
2317 				nt->nt_proc = p;
2318 				nt->nt_ksi.ksi_mqd = uap->mqd;
2319 				notifier_insert(p, nt);
2320 			}
2321 			nt->nt_sigev = ev;
2322 			mq->mq_notifier = nt;
2323 			PROC_UNLOCK(p);
2324 			/*
2325 			 * if there is no receivers and message queue
2326 			 * is not empty, we should send notification
2327 			 * as soon as possible.
2328 			 */
2329 			if (mq->mq_receivers == 0 &&
2330 			    !TAILQ_EMPTY(&mq->mq_msgq))
2331 				mqueue_send_notification(mq);
2332 		}
2333 	} else {
2334 		notifier_remove(p, mq, uap->mqd);
2335 	}
2336 	mtx_unlock(&mq->mq_mutex);
2337 
2338 out:
2339 	fdrop(fp, td);
2340 	if (newnt != NULL)
2341 		notifier_free(newnt);
2342 	return (error);
2343 }
2344 
2345 static void
2346 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2347 {
2348 	struct filedesc *fdp;
2349 	struct mqueue *mq;
2350 
2351 	fdp = td->td_proc->p_fd;
2352 	FILEDESC_LOCK_ASSERT(fdp);
2353 
2354 	if (fp->f_ops == &mqueueops) {
2355 		mq = FPTOMQ(fp);
2356 		mtx_lock(&mq->mq_mutex);
2357 		notifier_remove(td->td_proc, mq, fd);
2358 
2359 		/* have to wakeup thread in same process */
2360 		if (mq->mq_flags & MQ_RSEL) {
2361 			mq->mq_flags &= ~MQ_RSEL;
2362 			selwakeup(&mq->mq_rsel);
2363 		}
2364 		if (mq->mq_flags & MQ_WSEL) {
2365 			mq->mq_flags &= ~MQ_WSEL;
2366 			selwakeup(&mq->mq_wsel);
2367 		}
2368 		mtx_unlock(&mq->mq_mutex);
2369 	}
2370 }
2371 
2372 static void
2373 mq_proc_exit(void *arg __unused, struct proc *p)
2374 {
2375 	struct filedesc *fdp;
2376 	struct file *fp;
2377 	struct mqueue *mq;
2378 	int i;
2379 
2380 	fdp = p->p_fd;
2381 	FILEDESC_SLOCK(fdp);
2382 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2383 		fp = fget_locked(fdp, i);
2384 		if (fp != NULL && fp->f_ops == &mqueueops) {
2385 			mq = FPTOMQ(fp);
2386 			mtx_lock(&mq->mq_mutex);
2387 			notifier_remove(p, FPTOMQ(fp), i);
2388 			mtx_unlock(&mq->mq_mutex);
2389 		}
2390 	}
2391 	FILEDESC_SUNLOCK(fdp);
2392 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2393 }
2394 
2395 static int
2396 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2397 	int flags, struct thread *td)
2398 {
2399 	return (EOPNOTSUPP);
2400 }
2401 
2402 static int
2403 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2404 	int flags, struct thread *td)
2405 {
2406 	return (EOPNOTSUPP);
2407 }
2408 
2409 static int
2410 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred,
2411     struct thread *td)
2412 {
2413 
2414 	return (EINVAL);
2415 }
2416 
2417 static int
2418 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2419 	struct ucred *active_cred, struct thread *td)
2420 {
2421 	return (ENOTTY);
2422 }
2423 
2424 static int
2425 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2426 	struct thread *td)
2427 {
2428 	struct mqueue *mq = FPTOMQ(fp);
2429 	int revents = 0;
2430 
2431 	mtx_lock(&mq->mq_mutex);
2432 	if (events & (POLLIN | POLLRDNORM)) {
2433 		if (mq->mq_curmsgs) {
2434 			revents |= events & (POLLIN | POLLRDNORM);
2435 		} else {
2436 			mq->mq_flags |= MQ_RSEL;
2437 			selrecord(td, &mq->mq_rsel);
2438  		}
2439 	}
2440 	if (events & POLLOUT) {
2441 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2442 			revents |= POLLOUT;
2443 		else {
2444 			mq->mq_flags |= MQ_WSEL;
2445 			selrecord(td, &mq->mq_wsel);
2446 		}
2447 	}
2448 	mtx_unlock(&mq->mq_mutex);
2449 	return (revents);
2450 }
2451 
2452 static int
2453 mqf_close(struct file *fp, struct thread *td)
2454 {
2455 	struct mqfs_node *pn;
2456 
2457 	fp->f_ops = &badfileops;
2458 	pn = fp->f_data;
2459 	fp->f_data = NULL;
2460 	sx_xlock(&mqfs_data.mi_lock);
2461 	mqnode_release(pn);
2462 	sx_xunlock(&mqfs_data.mi_lock);
2463 	return (0);
2464 }
2465 
2466 static int
2467 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2468 	struct thread *td)
2469 {
2470 	struct mqfs_node *pn = fp->f_data;
2471 
2472 	bzero(st, sizeof *st);
2473 	sx_xlock(&mqfs_data.mi_lock);
2474 	st->st_atim = pn->mn_atime;
2475 	st->st_mtim = pn->mn_mtime;
2476 	st->st_ctim = pn->mn_ctime;
2477 	st->st_birthtim = pn->mn_birth;
2478 	st->st_uid = pn->mn_uid;
2479 	st->st_gid = pn->mn_gid;
2480 	st->st_mode = S_IFIFO | pn->mn_mode;
2481 	sx_xunlock(&mqfs_data.mi_lock);
2482 	return (0);
2483 }
2484 
2485 static int
2486 mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2487     struct thread *td)
2488 {
2489 	struct mqfs_node *pn;
2490 	int error;
2491 
2492 	error = 0;
2493 	pn = fp->f_data;
2494 	sx_xlock(&mqfs_data.mi_lock);
2495 	error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
2496 	    active_cred, NULL);
2497 	if (error != 0)
2498 		goto out;
2499 	pn->mn_mode = mode & ACCESSPERMS;
2500 out:
2501 	sx_xunlock(&mqfs_data.mi_lock);
2502 	return (error);
2503 }
2504 
2505 static int
2506 mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2507     struct thread *td)
2508 {
2509 	struct mqfs_node *pn;
2510 	int error;
2511 
2512 	error = 0;
2513 	pn = fp->f_data;
2514 	sx_xlock(&mqfs_data.mi_lock);
2515 	if (uid == (uid_t)-1)
2516 		uid = pn->mn_uid;
2517 	if (gid == (gid_t)-1)
2518 		gid = pn->mn_gid;
2519 	if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
2520 	    (gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
2521 	    (error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
2522 		goto out;
2523 	pn->mn_uid = uid;
2524 	pn->mn_gid = gid;
2525 out:
2526 	sx_xunlock(&mqfs_data.mi_lock);
2527 	return (error);
2528 }
2529 
2530 static int
2531 mqf_kqfilter(struct file *fp, struct knote *kn)
2532 {
2533 	struct mqueue *mq = FPTOMQ(fp);
2534 	int error = 0;
2535 
2536 	if (kn->kn_filter == EVFILT_READ) {
2537 		kn->kn_fop = &mq_rfiltops;
2538 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2539 	} else if (kn->kn_filter == EVFILT_WRITE) {
2540 		kn->kn_fop = &mq_wfiltops;
2541 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2542 	} else
2543 		error = EINVAL;
2544 	return (error);
2545 }
2546 
2547 static void
2548 filt_mqdetach(struct knote *kn)
2549 {
2550 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2551 
2552 	if (kn->kn_filter == EVFILT_READ)
2553 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2554 	else if (kn->kn_filter == EVFILT_WRITE)
2555 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2556 	else
2557 		panic("filt_mqdetach");
2558 }
2559 
2560 static int
2561 filt_mqread(struct knote *kn, long hint)
2562 {
2563 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2564 
2565 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2566 	return (mq->mq_curmsgs != 0);
2567 }
2568 
2569 static int
2570 filt_mqwrite(struct knote *kn, long hint)
2571 {
2572 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2573 
2574 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2575 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2576 }
2577 
2578 static struct fileops mqueueops = {
2579 	.fo_read		= mqf_read,
2580 	.fo_write		= mqf_write,
2581 	.fo_truncate		= mqf_truncate,
2582 	.fo_ioctl		= mqf_ioctl,
2583 	.fo_poll		= mqf_poll,
2584 	.fo_kqfilter		= mqf_kqfilter,
2585 	.fo_stat		= mqf_stat,
2586 	.fo_chmod		= mqf_chmod,
2587 	.fo_chown		= mqf_chown,
2588 	.fo_close		= mqf_close
2589 };
2590 
2591 static struct vop_vector mqfs_vnodeops = {
2592 	.vop_default 		= &default_vnodeops,
2593 	.vop_access		= mqfs_access,
2594 	.vop_cachedlookup	= mqfs_lookup,
2595 	.vop_lookup		= vfs_cache_lookup,
2596 	.vop_reclaim		= mqfs_reclaim,
2597 	.vop_create		= mqfs_create,
2598 	.vop_remove		= mqfs_remove,
2599 	.vop_inactive		= mqfs_inactive,
2600 	.vop_open		= mqfs_open,
2601 	.vop_close		= mqfs_close,
2602 	.vop_getattr		= mqfs_getattr,
2603 	.vop_setattr		= mqfs_setattr,
2604 	.vop_read		= mqfs_read,
2605 	.vop_write		= VOP_EOPNOTSUPP,
2606 	.vop_readdir		= mqfs_readdir,
2607 	.vop_mkdir		= VOP_EOPNOTSUPP,
2608 	.vop_rmdir		= VOP_EOPNOTSUPP
2609 };
2610 
2611 static struct vfsops mqfs_vfsops = {
2612 	.vfs_init 		= mqfs_init,
2613 	.vfs_uninit		= mqfs_uninit,
2614 	.vfs_mount		= mqfs_mount,
2615 	.vfs_unmount		= mqfs_unmount,
2616 	.vfs_root		= mqfs_root,
2617 	.vfs_statfs		= mqfs_statfs,
2618 };
2619 
2620 static struct vfsconf mqueuefs_vfsconf = {
2621 	.vfc_version = VFS_VERSION,
2622 	.vfc_name = "mqueuefs",
2623 	.vfc_vfsops = &mqfs_vfsops,
2624 	.vfc_typenum = -1,
2625 	.vfc_flags = VFCF_SYNTHETIC
2626 };
2627 
2628 static struct syscall_helper_data mq_syscalls[] = {
2629 	SYSCALL_INIT_HELPER(kmq_open),
2630 	SYSCALL_INIT_HELPER(kmq_setattr),
2631 	SYSCALL_INIT_HELPER(kmq_timedsend),
2632 	SYSCALL_INIT_HELPER(kmq_timedreceive),
2633 	SYSCALL_INIT_HELPER(kmq_notify),
2634 	SYSCALL_INIT_HELPER(kmq_unlink),
2635 	SYSCALL_INIT_LAST
2636 };
2637 
2638 #ifdef COMPAT_FREEBSD32
2639 #include <compat/freebsd32/freebsd32.h>
2640 #include <compat/freebsd32/freebsd32_proto.h>
2641 #include <compat/freebsd32/freebsd32_syscall.h>
2642 #include <compat/freebsd32/freebsd32_util.h>
2643 
2644 static void
2645 mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
2646 {
2647 
2648 	to->mq_flags = from->mq_flags;
2649 	to->mq_maxmsg = from->mq_maxmsg;
2650 	to->mq_msgsize = from->mq_msgsize;
2651 	to->mq_curmsgs = from->mq_curmsgs;
2652 }
2653 
2654 static void
2655 mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
2656 {
2657 
2658 	to->mq_flags = from->mq_flags;
2659 	to->mq_maxmsg = from->mq_maxmsg;
2660 	to->mq_msgsize = from->mq_msgsize;
2661 	to->mq_curmsgs = from->mq_curmsgs;
2662 }
2663 
2664 int
2665 freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
2666 {
2667 	struct mq_attr attr;
2668 	struct mq_attr32 attr32;
2669 	int flags, error;
2670 
2671 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
2672 		return (EINVAL);
2673 	flags = FFLAGS(uap->flags);
2674 	if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
2675 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2676 		if (error)
2677 			return (error);
2678 		mq_attr_from32(&attr32, &attr);
2679 	}
2680 	return (kern_kmq_open(td, uap->path, flags, uap->mode,
2681 	    uap->attr != NULL ? &attr : NULL));
2682 }
2683 
2684 int
2685 freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
2686 {
2687 	struct mq_attr attr, oattr;
2688 	struct mq_attr32 attr32, oattr32;
2689 	int error;
2690 
2691 	if (uap->attr != NULL) {
2692 		error = copyin(uap->attr, &attr32, sizeof(attr32));
2693 		if (error != 0)
2694 			return (error);
2695 		mq_attr_from32(&attr32, &attr);
2696 	}
2697 	error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
2698 	    &oattr);
2699 	if (error != 0)
2700 		return (error);
2701 	if (uap->oattr != NULL) {
2702 		mq_attr_to32(&oattr, &oattr32);
2703 		error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
2704 	}
2705 	return (error);
2706 }
2707 
2708 int
2709 freebsd32_kmq_timedsend(struct thread *td,
2710     struct freebsd32_kmq_timedsend_args *uap)
2711 {
2712 	struct mqueue *mq;
2713 	struct file *fp;
2714 	struct timespec32 ets32;
2715 	struct timespec *abs_timeout, ets;
2716 	int error;
2717 	int waitok;
2718 
2719 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2720 	if (error)
2721 		return (error);
2722 	if (uap->abs_timeout != NULL) {
2723 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2724 		if (error != 0)
2725 			return (error);
2726 		CP(ets32, ets, tv_sec);
2727 		CP(ets32, ets, tv_nsec);
2728 		abs_timeout = &ets;
2729 	} else
2730 		abs_timeout = NULL;
2731 	waitok = !(fp->f_flag & O_NONBLOCK);
2732 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2733 		uap->msg_prio, waitok, abs_timeout);
2734 	fdrop(fp, td);
2735 	return (error);
2736 }
2737 
2738 int
2739 freebsd32_kmq_timedreceive(struct thread *td,
2740     struct freebsd32_kmq_timedreceive_args *uap)
2741 {
2742 	struct mqueue *mq;
2743 	struct file *fp;
2744 	struct timespec32 ets32;
2745 	struct timespec *abs_timeout, ets;
2746 	int error, waitok;
2747 
2748 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2749 	if (error)
2750 		return (error);
2751 	if (uap->abs_timeout != NULL) {
2752 		error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
2753 		if (error != 0)
2754 			return (error);
2755 		CP(ets32, ets, tv_sec);
2756 		CP(ets32, ets, tv_nsec);
2757 		abs_timeout = &ets;
2758 	} else
2759 		abs_timeout = NULL;
2760 	waitok = !(fp->f_flag & O_NONBLOCK);
2761 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2762 		uap->msg_prio, waitok, abs_timeout);
2763 	fdrop(fp, td);
2764 	return (error);
2765 }
2766 
2767 static struct syscall_helper_data mq32_syscalls[] = {
2768 	SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
2769 	SYSCALL32_INIT_HELPER(freebsd32_kmq_setattr),
2770 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedsend),
2771 	SYSCALL32_INIT_HELPER(freebsd32_kmq_timedreceive),
2772 	SYSCALL32_INIT_HELPER_COMPAT(kmq_notify),
2773 	SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
2774 	SYSCALL_INIT_LAST
2775 };
2776 #endif
2777 
2778 static int
2779 mqinit(void)
2780 {
2781 	int error;
2782 
2783 	error = syscall_helper_register(mq_syscalls);
2784 	if (error != 0)
2785 		return (error);
2786 #ifdef COMPAT_FREEBSD32
2787 	error = syscall32_helper_register(mq32_syscalls);
2788 	if (error != 0)
2789 		return (error);
2790 #endif
2791 	return (0);
2792 }
2793 
2794 static int
2795 mqunload(void)
2796 {
2797 
2798 #ifdef COMPAT_FREEBSD32
2799 	syscall32_helper_unregister(mq32_syscalls);
2800 #endif
2801 	syscall_helper_unregister(mq_syscalls);
2802 	return (0);
2803 }
2804 
2805 static int
2806 mq_modload(struct module *module, int cmd, void *arg)
2807 {
2808 	int error = 0;
2809 
2810 	error = vfs_modevent(module, cmd, arg);
2811 	if (error != 0)
2812 		return (error);
2813 
2814 	switch (cmd) {
2815 	case MOD_LOAD:
2816 		error = mqinit();
2817 		if (error != 0)
2818 			mqunload();
2819 		break;
2820 	case MOD_UNLOAD:
2821 		error = mqunload();
2822 		break;
2823 	default:
2824 		break;
2825 	}
2826 	return (error);
2827 }
2828 
2829 static moduledata_t mqueuefs_mod = {
2830 	"mqueuefs",
2831 	mq_modload,
2832 	&mqueuefs_vfsconf
2833 };
2834 DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
2835 MODULE_VERSION(mqueuefs, 1);
2836