xref: /freebsd/sys/kern/uipc_mqueue.c (revision 2be1a816b9ff69588e55be0a84cbe2a31efc0f2f)
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 <sys/param.h>
49 #include <sys/kernel.h>
50 #include <sys/systm.h>
51 #include <sys/limits.h>
52 #include <sys/buf.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/fcntl.h>
57 #include <sys/file.h>
58 #include <sys/filedesc.h>
59 #include <sys/lock.h>
60 #include <sys/malloc.h>
61 #include <sys/module.h>
62 #include <sys/mount.h>
63 #include <sys/mqueue.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/posix4.h>
67 #include <sys/poll.h>
68 #include <sys/priv.h>
69 #include <sys/proc.h>
70 #include <sys/queue.h>
71 #include <sys/sysproto.h>
72 #include <sys/stat.h>
73 #include <sys/syscall.h>
74 #include <sys/syscallsubr.h>
75 #include <sys/sysent.h>
76 #include <sys/sx.h>
77 #include <sys/sysctl.h>
78 #include <sys/taskqueue.h>
79 #include <sys/unistd.h>
80 #include <sys/vnode.h>
81 #include <machine/atomic.h>
82 
83 /*
84  * Limits and constants
85  */
86 #define	MQFS_NAMELEN		NAME_MAX
87 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
88 
89 /* node types */
90 typedef enum {
91 	mqfstype_none = 0,
92 	mqfstype_root,
93 	mqfstype_dir,
94 	mqfstype_this,
95 	mqfstype_parent,
96 	mqfstype_file,
97 	mqfstype_symlink,
98 } mqfs_type_t;
99 
100 struct mqfs_node;
101 
102 /*
103  * mqfs_info: describes a mqfs instance
104  */
105 struct mqfs_info {
106 	struct sx		mi_lock;
107 	struct mqfs_node	*mi_root;
108 	struct unrhdr		*mi_unrhdr;
109 };
110 
111 struct mqfs_vdata {
112 	LIST_ENTRY(mqfs_vdata)	mv_link;
113 	struct mqfs_node	*mv_node;
114 	struct vnode		*mv_vnode;
115 	struct task		mv_task;
116 };
117 
118 /*
119  * mqfs_node: describes a node (file or directory) within a mqfs
120  */
121 struct mqfs_node {
122 	char			mn_name[MQFS_NAMELEN+1];
123 	struct mqfs_info	*mn_info;
124 	struct mqfs_node	*mn_parent;
125 	LIST_HEAD(,mqfs_node)	mn_children;
126 	LIST_ENTRY(mqfs_node)	mn_sibling;
127 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
128 	int			mn_refcount;
129 	mqfs_type_t		mn_type;
130 	int			mn_deleted;
131 	u_int32_t		mn_fileno;
132 	void			*mn_data;
133 	struct timespec		mn_birth;
134 	struct timespec		mn_ctime;
135 	struct timespec		mn_atime;
136 	struct timespec		mn_mtime;
137 	uid_t			mn_uid;
138 	gid_t			mn_gid;
139 	int			mn_mode;
140 };
141 
142 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
143 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
144 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
145 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
146 				(fp)->f_data)->mn_data))
147 
148 TAILQ_HEAD(msgq, mqueue_msg);
149 
150 struct mqueue;
151 
152 struct mqueue_notifier {
153 	LIST_ENTRY(mqueue_notifier)	nt_link;
154 	struct sigevent			nt_sigev;
155 	ksiginfo_t			nt_ksi;
156 	struct proc			*nt_proc;
157 };
158 
159 struct mqueue {
160 	struct mtx	mq_mutex;
161 	int		mq_flags;
162 	long		mq_maxmsg;
163 	long		mq_msgsize;
164 	long		mq_curmsgs;
165 	long		mq_totalbytes;
166 	struct msgq	mq_msgq;
167 	int		mq_receivers;
168 	int		mq_senders;
169 	struct selinfo	mq_rsel;
170 	struct selinfo	mq_wsel;
171 	struct mqueue_notifier	*mq_notifier;
172 };
173 
174 #define	MQ_RSEL		0x01
175 #define	MQ_WSEL		0x02
176 
177 struct mqueue_msg {
178 	TAILQ_ENTRY(mqueue_msg)	msg_link;
179 	unsigned int	msg_prio;
180 	unsigned int	msg_size;
181 	/* following real data... */
182 };
183 
184 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
185 	"POSIX real time message queue");
186 
187 static int	default_maxmsg  = 10;
188 static int	default_msgsize = 1024;
189 
190 static int	maxmsg = 100;
191 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
192     &maxmsg, 0, "Default maximum messages in queue");
193 static int	maxmsgsize = 16384;
194 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
195     &maxmsgsize, 0, "Default maximum message size");
196 static int	maxmq = 100;
197 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
198     &maxmq, 0, "maximum message queues");
199 static int	curmq = 0;
200 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
201     &curmq, 0, "current message queue number");
202 static int	unloadable = 0;
203 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
204 
205 static eventhandler_tag exit_tag;
206 
207 /* Only one instance per-system */
208 static struct mqfs_info		mqfs_data;
209 static uma_zone_t		mqnode_zone;
210 static uma_zone_t		mqueue_zone;
211 static uma_zone_t		mvdata_zone;
212 static uma_zone_t		mqnoti_zone;
213 static struct vop_vector	mqfs_vnodeops;
214 static struct fileops		mqueueops;
215 
216 /*
217  * Directory structure construction and manipulation
218  */
219 #ifdef notyet
220 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
221 	const char *name, int namelen, struct ucred *cred, int mode);
222 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
223 	const char *name, int namelen, struct ucred *cred, int mode);
224 #endif
225 
226 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
227 	const char *name, int namelen, struct ucred *cred, int mode);
228 static int	mqfs_destroy(struct mqfs_node *mn);
229 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
230 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
231 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
232 
233 /*
234  * Message queue construction and maniplation
235  */
236 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
237 static void	mqueue_free(struct mqueue *mq);
238 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
239 			size_t msg_len, unsigned msg_prio, int waitok,
240 			const struct timespec *abs_timeout);
241 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
242 			size_t msg_len, unsigned *msg_prio, int waitok,
243 			const struct timespec *abs_timeout);
244 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
245 			int timo);
246 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
247 			int timo);
248 static void	mqueue_send_notification(struct mqueue *mq);
249 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
250 static void	mq_proc_exit(void *arg, struct proc *p);
251 
252 /*
253  * kqueue filters
254  */
255 static void	filt_mqdetach(struct knote *kn);
256 static int	filt_mqread(struct knote *kn, long hint);
257 static int	filt_mqwrite(struct knote *kn, long hint);
258 
259 struct filterops mq_rfiltops =
260 	{ 1, NULL, filt_mqdetach, filt_mqread };
261 struct filterops mq_wfiltops =
262 	{ 1, NULL, filt_mqdetach, filt_mqwrite };
263 
264 /*
265  * Initialize fileno bitmap
266  */
267 static void
268 mqfs_fileno_init(struct mqfs_info *mi)
269 {
270 	struct unrhdr *up;
271 
272 	up = new_unrhdr(1, INT_MAX, NULL);
273 	mi->mi_unrhdr = up;
274 }
275 
276 /*
277  * Tear down fileno bitmap
278  */
279 static void
280 mqfs_fileno_uninit(struct mqfs_info *mi)
281 {
282 	struct unrhdr *up;
283 
284 	up = mi->mi_unrhdr;
285 	mi->mi_unrhdr = NULL;
286 	delete_unrhdr(up);
287 }
288 
289 /*
290  * Allocate a file number
291  */
292 static void
293 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
294 {
295 	/* make sure our parent has a file number */
296 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
297 		mqfs_fileno_alloc(mi, mn->mn_parent);
298 
299 	switch (mn->mn_type) {
300 	case mqfstype_root:
301 	case mqfstype_dir:
302 	case mqfstype_file:
303 	case mqfstype_symlink:
304 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
305 		break;
306 	case mqfstype_this:
307 		KASSERT(mn->mn_parent != NULL,
308 		    ("mqfstype_this node has no parent"));
309 		mn->mn_fileno = mn->mn_parent->mn_fileno;
310 		break;
311 	case mqfstype_parent:
312 		KASSERT(mn->mn_parent != NULL,
313 		    ("mqfstype_parent node has no parent"));
314 		if (mn->mn_parent == mi->mi_root) {
315 			mn->mn_fileno = mn->mn_parent->mn_fileno;
316 			break;
317 		}
318 		KASSERT(mn->mn_parent->mn_parent != NULL,
319 		    ("mqfstype_parent node has no grandparent"));
320 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
321 		break;
322 	default:
323 		KASSERT(0,
324 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
325 			mn->mn_type));
326 		break;
327 	}
328 }
329 
330 /*
331  * Release a file number
332  */
333 static void
334 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
335 {
336 	switch (mn->mn_type) {
337 	case mqfstype_root:
338 	case mqfstype_dir:
339 	case mqfstype_file:
340 	case mqfstype_symlink:
341 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
342 		break;
343 	case mqfstype_this:
344 	case mqfstype_parent:
345 		/* ignore these, as they don't "own" their file number */
346 		break;
347 	default:
348 		KASSERT(0,
349 		    ("mqfs_fileno_free() called for unknown type node: %d",
350 			mn->mn_type));
351 		break;
352 	}
353 }
354 
355 static __inline struct mqfs_node *
356 mqnode_alloc(void)
357 {
358 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
359 }
360 
361 static __inline void
362 mqnode_free(struct mqfs_node *node)
363 {
364 	uma_zfree(mqnode_zone, node);
365 }
366 
367 static __inline void
368 mqnode_addref(struct mqfs_node *node)
369 {
370 	atomic_fetchadd_int(&node->mn_refcount, 1);
371 }
372 
373 static __inline void
374 mqnode_release(struct mqfs_node *node)
375 {
376 	int old, exp;
377 
378 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
379 	if (node->mn_type == mqfstype_dir ||
380 	    node->mn_type == mqfstype_root)
381 		exp = 3; /* include . and .. */
382 	else
383 		exp = 1;
384 	if (old == exp)
385 		mqfs_destroy(node);
386 }
387 
388 /*
389  * Add a node to a directory
390  */
391 static int
392 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
393 {
394 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
395 	KASSERT(parent->mn_info != NULL,
396 	    ("%s(): parent has no mn_info", __func__));
397 	KASSERT(parent->mn_type == mqfstype_dir ||
398 	    parent->mn_type == mqfstype_root,
399 	    ("%s(): parent is not a directory", __func__));
400 
401 	node->mn_info = parent->mn_info;
402 	node->mn_parent = parent;
403 	LIST_INIT(&node->mn_children);
404 	LIST_INIT(&node->mn_vnodes);
405 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
406 	mqnode_addref(parent);
407 	return (0);
408 }
409 
410 static struct mqfs_node *
411 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
412 	int nodetype)
413 {
414 	struct mqfs_node *node;
415 
416 	node = mqnode_alloc();
417 	strncpy(node->mn_name, name, namelen);
418 	node->mn_type = nodetype;
419 	node->mn_refcount = 1;
420 	getnanotime(&node->mn_birth);
421 	node->mn_ctime = node->mn_atime = node->mn_mtime
422 		= node->mn_birth;
423 	node->mn_uid = cred->cr_uid;
424 	node->mn_gid = cred->cr_gid;
425 	node->mn_mode = mode;
426 	return (node);
427 }
428 
429 /*
430  * Create a file
431  */
432 static struct mqfs_node *
433 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
434 	struct ucred *cred, int mode)
435 {
436 	struct mqfs_node *node;
437 
438 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
439 	if (mqfs_add_node(parent, node) != 0) {
440 		mqnode_free(node);
441 		return (NULL);
442 	}
443 	return (node);
444 }
445 
446 /*
447  * Add . and .. to a directory
448  */
449 static int
450 mqfs_fixup_dir(struct mqfs_node *parent)
451 {
452 	struct mqfs_node *dir;
453 
454 	dir = mqnode_alloc();
455 	dir->mn_name[0] = '.';
456 	dir->mn_type = mqfstype_this;
457 	dir->mn_refcount = 1;
458 	if (mqfs_add_node(parent, dir) != 0) {
459 		mqnode_free(dir);
460 		return (-1);
461 	}
462 
463 	dir = mqnode_alloc();
464 	dir->mn_name[0] = dir->mn_name[1] = '.';
465 	dir->mn_type = mqfstype_parent;
466 	dir->mn_refcount = 1;
467 
468 	if (mqfs_add_node(parent, dir) != 0) {
469 		mqnode_free(dir);
470 		return (-1);
471 	}
472 
473 	return (0);
474 }
475 
476 #ifdef notyet
477 
478 /*
479  * Create a directory
480  */
481 static struct mqfs_node *
482 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
483 	struct ucred *cred, int mode)
484 {
485 	struct mqfs_node *node;
486 
487 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
488 	if (mqfs_add_node(parent, node) != 0) {
489 		mqnode_free(node);
490 		return (NULL);
491 	}
492 
493 	if (mqfs_fixup_dir(node) != 0) {
494 		mqfs_destroy(node);
495 		return (NULL);
496 	}
497 	return (node);
498 }
499 
500 /*
501  * Create a symlink
502  */
503 static struct mqfs_node *
504 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
505 	struct ucred *cred, int mode)
506 {
507 	struct mqfs_node *node;
508 
509 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
510 	if (mqfs_add_node(parent, node) != 0) {
511 		mqnode_free(node);
512 		return (NULL);
513 	}
514 	return (node);
515 }
516 
517 #endif
518 
519 /*
520  * Destroy a node or a tree of nodes
521  */
522 static int
523 mqfs_destroy(struct mqfs_node *node)
524 {
525 	struct mqfs_node *parent;
526 
527 	KASSERT(node != NULL,
528 	    ("%s(): node is NULL", __func__));
529 	KASSERT(node->mn_info != NULL,
530 	    ("%s(): node has no mn_info", __func__));
531 
532 	/* destroy children */
533 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
534 		while (! LIST_EMPTY(&node->mn_children))
535 			mqfs_destroy(LIST_FIRST(&node->mn_children));
536 
537 	/* unlink from parent */
538 	if ((parent = node->mn_parent) != NULL) {
539 		KASSERT(parent->mn_info == node->mn_info,
540 		    ("%s(): parent has different mn_info", __func__));
541 		LIST_REMOVE(node, mn_sibling);
542 	}
543 
544 	if (node->mn_fileno != 0)
545 		mqfs_fileno_free(node->mn_info, node);
546 	if (node->mn_data != NULL)
547 		mqueue_free(node->mn_data);
548 	mqnode_free(node);
549 	return (0);
550 }
551 
552 /*
553  * Mount a mqfs instance
554  */
555 static int
556 mqfs_mount(struct mount *mp, struct thread *td)
557 {
558 	struct statfs *sbp;
559 
560 	if (mp->mnt_flag & MNT_UPDATE)
561 		return (EOPNOTSUPP);
562 
563 	mp->mnt_data = &mqfs_data;
564 	MNT_ILOCK(mp);
565 	mp->mnt_flag |= MNT_LOCAL;
566 	mp->mnt_kern_flag |= MNTK_MPSAFE;
567 	MNT_IUNLOCK(mp);
568 	vfs_getnewfsid(mp);
569 
570 	sbp = &mp->mnt_stat;
571 	vfs_mountedfrom(mp, "mqueue");
572 	sbp->f_bsize = PAGE_SIZE;
573 	sbp->f_iosize = PAGE_SIZE;
574 	sbp->f_blocks = 1;
575 	sbp->f_bfree = 0;
576 	sbp->f_bavail = 0;
577 	sbp->f_files = 1;
578 	sbp->f_ffree = 0;
579 	return (0);
580 }
581 
582 /*
583  * Unmount a mqfs instance
584  */
585 static int
586 mqfs_unmount(struct mount *mp, int mntflags, struct thread *td)
587 {
588 	int error;
589 
590 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0, td);
591 	return (error);
592 }
593 
594 /*
595  * Return a root vnode
596  */
597 static int
598 mqfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
599 {
600 	struct mqfs_info *mqfs;
601 	int ret;
602 
603 	mqfs = VFSTOMQFS(mp);
604 	sx_xlock(&mqfs->mi_lock);
605 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
606 	sx_xunlock(&mqfs->mi_lock);
607 	return (ret);
608 }
609 
610 /*
611  * Return filesystem stats
612  */
613 static int
614 mqfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
615 {
616 	/* XXX update statistics */
617 	return (0);
618 }
619 
620 /*
621  * Initialize a mqfs instance
622  */
623 static int
624 mqfs_init(struct vfsconf *vfc)
625 {
626 	struct mqfs_node *root;
627 	struct mqfs_info *mi;
628 
629 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
630 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
631 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
632 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
633 	mvdata_zone = uma_zcreate("mvdata",
634 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
635 		NULL, UMA_ALIGN_PTR, 0);
636 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
637 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
638 	mi = &mqfs_data;
639 	sx_init(&mi->mi_lock, "mqfs lock");
640 	/* set up the root diretory */
641 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
642 		mqfstype_root);
643 	root->mn_info = mi;
644 	LIST_INIT(&root->mn_children);
645 	LIST_INIT(&root->mn_vnodes);
646 	mi->mi_root = root;
647 	mqfs_fileno_init(mi);
648 	mqfs_fileno_alloc(mi, root);
649 	mqfs_fixup_dir(root);
650 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
651 	    EVENTHANDLER_PRI_ANY);
652 	mq_fdclose = mqueue_fdclose;
653 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
654 	return (0);
655 }
656 
657 /*
658  * Destroy a mqfs instance
659  */
660 static int
661 mqfs_uninit(struct vfsconf *vfc)
662 {
663 	struct mqfs_info *mi;
664 
665 	if (!unloadable)
666 		return (EOPNOTSUPP);
667 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
668 	mi = &mqfs_data;
669 	mqfs_destroy(mi->mi_root);
670 	mi->mi_root = NULL;
671 	mqfs_fileno_uninit(mi);
672 	sx_destroy(&mi->mi_lock);
673 	uma_zdestroy(mqnode_zone);
674 	uma_zdestroy(mqueue_zone);
675 	uma_zdestroy(mvdata_zone);
676 	uma_zdestroy(mqnoti_zone);
677 	return (0);
678 }
679 
680 /*
681  * task routine
682  */
683 static void
684 do_recycle(void *context, int pending __unused)
685 {
686 	struct vnode *vp = (struct vnode *)context;
687 
688 	vrecycle(vp, curthread);
689 	vdrop(vp);
690 }
691 
692 /*
693  * Allocate a vnode
694  */
695 static int
696 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
697 {
698 	struct mqfs_vdata *vd;
699 	int error;
700 
701 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
702 		if (vd->mv_vnode->v_mount == mp)
703 			break;
704 	}
705 
706 	if (vd != NULL) {
707 		*vpp = vd->mv_vnode;
708 		vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
709 		return (0);
710 	}
711 
712 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
713 	if (error)
714 		return (error);
715 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
716 	error = insmntque(*vpp, mp);
717 	if (error != 0) {
718 		*vpp = NULLVP;
719 		return (error);
720 	}
721 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
722 	(*vpp)->v_data = vd;
723 	vd->mv_vnode = *vpp;
724 	vd->mv_node = pn;
725 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
726 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
727 	mqnode_addref(pn);
728 	switch (pn->mn_type) {
729 	case mqfstype_root:
730 		(*vpp)->v_vflag = VV_ROOT;
731 		/* fall through */
732 	case mqfstype_dir:
733 	case mqfstype_this:
734 	case mqfstype_parent:
735 		(*vpp)->v_type = VDIR;
736 		break;
737 	case mqfstype_file:
738 		(*vpp)->v_type = VREG;
739 		break;
740 	case mqfstype_symlink:
741 		(*vpp)->v_type = VLNK;
742 		break;
743 	case mqfstype_none:
744 		KASSERT(0, ("mqfs_allocf called for null node\n"));
745 	default:
746 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
747 	}
748 	return (0);
749 }
750 
751 /*
752  * Search a directory entry
753  */
754 static struct mqfs_node *
755 mqfs_search(struct mqfs_node *pd, const char *name, int len)
756 {
757 	struct mqfs_node *pn;
758 
759 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
760 		if (strncmp(pn->mn_name, name, len) == 0)
761 			return (pn);
762 	}
763 	return (NULL);
764 }
765 
766 /*
767  * Look up a file or directory.
768  */
769 static int
770 mqfs_lookupx(struct vop_cachedlookup_args *ap)
771 {
772 	struct componentname *cnp;
773 	struct vnode *dvp, **vpp;
774 	struct mqfs_node *pd;
775 	struct mqfs_node *pn;
776 	int nameiop, flags, error, namelen;
777 	char *pname;
778 	struct thread *td;
779 
780 	cnp = ap->a_cnp;
781 	vpp = ap->a_vpp;
782 	dvp = ap->a_dvp;
783 	pname = cnp->cn_nameptr;
784 	namelen = cnp->cn_namelen;
785 	td = cnp->cn_thread;
786 	flags = cnp->cn_flags;
787 	nameiop = cnp->cn_nameiop;
788 	pd = VTON(dvp);
789 	pn = NULL;
790 	*vpp = NULLVP;
791 
792 	if (dvp->v_type != VDIR)
793 		return (ENOTDIR);
794 
795 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
796 	if (error)
797 		return (error);
798 
799 	/* shortcut: check if the name is too long */
800 	if (cnp->cn_namelen >= MQFS_NAMELEN)
801 		return (ENOENT);
802 
803 	/* self */
804 	if (namelen == 1 && pname[0] == '.') {
805 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
806 			return (EINVAL);
807 		pn = pd;
808 		*vpp = dvp;
809 		VREF(dvp);
810 		return (0);
811 	}
812 
813 	/* parent */
814 	if (cnp->cn_flags & ISDOTDOT) {
815 		if (dvp->v_vflag & VV_ROOT)
816 			return (EIO);
817 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
818 			return (EINVAL);
819 		VOP_UNLOCK(dvp, 0);
820 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
821 		pn = pd->mn_parent;
822 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
823 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
824 		return (error);
825 	}
826 
827 	/* named node */
828 	pn = mqfs_search(pd, pname, namelen);
829 
830 	/* found */
831 	if (pn != NULL) {
832 		/* DELETE */
833 		if (nameiop == DELETE && (flags & ISLASTCN)) {
834 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
835 			if (error)
836 				return (error);
837 			if (*vpp == dvp) {
838 				VREF(dvp);
839 				*vpp = dvp;
840 				return (0);
841 			}
842 		}
843 
844 		/* allocate vnode */
845 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
846 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
847 			cache_enter(dvp, *vpp, cnp);
848 		return (error);
849 	}
850 
851 	/* not found */
852 
853 	/* will create a new entry in the directory ? */
854 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
855 	    && (flags & ISLASTCN)) {
856 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
857 		if (error)
858 			return (error);
859 		cnp->cn_flags |= SAVENAME;
860 		return (EJUSTRETURN);
861 	}
862 	return (ENOENT);
863 }
864 
865 #if 0
866 struct vop_lookup_args {
867 	struct vop_generic_args a_gen;
868 	struct vnode *a_dvp;
869 	struct vnode **a_vpp;
870 	struct componentname *a_cnp;
871 };
872 #endif
873 
874 /*
875  * vnode lookup operation
876  */
877 static int
878 mqfs_lookup(struct vop_cachedlookup_args *ap)
879 {
880 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
881 	int rc;
882 
883 	sx_xlock(&mqfs->mi_lock);
884 	rc = mqfs_lookupx(ap);
885 	sx_xunlock(&mqfs->mi_lock);
886 	return (rc);
887 }
888 
889 #if 0
890 struct vop_create_args {
891 	struct vnode *a_dvp;
892 	struct vnode **a_vpp;
893 	struct componentname *a_cnp;
894 	struct vattr *a_vap;
895 };
896 #endif
897 
898 /*
899  * vnode creation operation
900  */
901 static int
902 mqfs_create(struct vop_create_args *ap)
903 {
904 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
905 	struct componentname *cnp = ap->a_cnp;
906 	struct mqfs_node *pd;
907 	struct mqfs_node *pn;
908 	struct mqueue *mq;
909 	int error;
910 
911 	pd = VTON(ap->a_dvp);
912 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
913 		return (ENOTDIR);
914 	mq = mqueue_alloc(NULL);
915 	if (mq == NULL)
916 		return (EAGAIN);
917 	sx_xlock(&mqfs->mi_lock);
918 #if 0
919 	/* named node */
920 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
921 	if (pn != NULL) {
922 		mqueue_free(mq);
923 		sx_xunlock(&mqfs->mi_lock);
924 		return (EEXIST);
925 	}
926 #else
927 	if ((cnp->cn_flags & HASBUF) == 0)
928 		panic("%s: no name", __func__);
929 #endif
930 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
931 		cnp->cn_cred, ap->a_vap->va_mode);
932 	if (pn == NULL)
933 		error = ENOSPC;
934 	else {
935 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
936 		if (error)
937 			mqfs_destroy(pn);
938 		else
939 			pn->mn_data = mq;
940 	}
941 	sx_xunlock(&mqfs->mi_lock);
942 	if (error)
943 		mqueue_free(mq);
944 	return (error);
945 }
946 
947 /*
948  * Remove an entry
949  */
950 static
951 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
952 {
953 	struct mqfs_node *parent;
954 	struct mqfs_vdata *vd;
955 	int error = 0;
956 
957 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
958 
959 	if (ucred->cr_uid != pn->mn_uid &&
960 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
961 		error = EACCES;
962 	else if (!pn->mn_deleted) {
963 		parent = pn->mn_parent;
964 		pn->mn_parent = NULL;
965 		pn->mn_deleted = 1;
966 		LIST_REMOVE(pn, mn_sibling);
967 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
968 			cache_purge(vd->mv_vnode);
969 			vhold(vd->mv_vnode);
970 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
971 		}
972 		mqnode_release(pn);
973 		mqnode_release(parent);
974 	} else
975 		error = ENOENT;
976 	return (error);
977 }
978 
979 #if 0
980 struct vop_remove_args {
981 	struct vnode *a_dvp;
982 	struct vnode *a_vp;
983 	struct componentname *a_cnp;
984 };
985 #endif
986 
987 /*
988  * vnode removal operation
989  */
990 static int
991 mqfs_remove(struct vop_remove_args *ap)
992 {
993 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
994 	struct mqfs_node *pn;
995 	int error;
996 
997 	if (ap->a_vp->v_type == VDIR)
998                 return (EPERM);
999 	pn = VTON(ap->a_vp);
1000 	sx_xlock(&mqfs->mi_lock);
1001 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1002 	sx_xunlock(&mqfs->mi_lock);
1003 	return (error);
1004 }
1005 
1006 #if 0
1007 struct vop_inactive_args {
1008 	struct vnode *a_vp;
1009 	struct thread *a_td;
1010 };
1011 #endif
1012 
1013 static int
1014 mqfs_inactive(struct vop_inactive_args *ap)
1015 {
1016 	struct mqfs_node *pn = VTON(ap->a_vp);
1017 
1018 	if (pn->mn_deleted)
1019 		vrecycle(ap->a_vp, ap->a_td);
1020 	return (0);
1021 }
1022 
1023 #if 0
1024 struct vop_reclaim_args {
1025 	struct vop_generic_args a_gen;
1026 	struct vnode *a_vp;
1027 	struct thread *a_td;
1028 };
1029 #endif
1030 
1031 static int
1032 mqfs_reclaim(struct vop_reclaim_args *ap)
1033 {
1034 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1035 	struct vnode *vp = ap->a_vp;
1036 	struct mqfs_node *pn;
1037 	struct mqfs_vdata *vd;
1038 
1039 	vd = vp->v_data;
1040 	pn = vd->mv_node;
1041 	sx_xlock(&mqfs->mi_lock);
1042 	vp->v_data = NULL;
1043 	LIST_REMOVE(vd, mv_link);
1044 	uma_zfree(mvdata_zone, vd);
1045 	mqnode_release(pn);
1046 	sx_xunlock(&mqfs->mi_lock);
1047 	return (0);
1048 }
1049 
1050 #if 0
1051 struct vop_open_args {
1052 	struct vop_generic_args a_gen;
1053 	struct vnode *a_vp;
1054 	int a_mode;
1055 	struct ucred *a_cred;
1056 	struct thread *a_td;
1057 	int a_fdidx;
1058 };
1059 #endif
1060 
1061 static int
1062 mqfs_open(struct vop_open_args *ap)
1063 {
1064 	return (0);
1065 }
1066 
1067 #if 0
1068 struct vop_close_args {
1069 	struct vop_generic_args a_gen;
1070 	struct vnode *a_vp;
1071 	int a_fflag;
1072 	struct ucred *a_cred;
1073 	struct thread *a_td;
1074 };
1075 #endif
1076 
1077 static int
1078 mqfs_close(struct vop_close_args *ap)
1079 {
1080 	return (0);
1081 }
1082 
1083 #if 0
1084 struct vop_access_args {
1085 	struct vop_generic_args a_gen;
1086 	struct vnode *a_vp;
1087 	int a_mode;
1088 	struct ucred *a_cred;
1089 	struct thread *a_td;
1090 };
1091 #endif
1092 
1093 /*
1094  * Verify permissions
1095  */
1096 static int
1097 mqfs_access(struct vop_access_args *ap)
1098 {
1099 	struct vnode *vp = ap->a_vp;
1100 	struct vattr vattr;
1101 	int error;
1102 
1103 	error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1104 	if (error)
1105 		return (error);
1106 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1107 	    vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1108 	return (error);
1109 }
1110 
1111 #if 0
1112 struct vop_getattr_args {
1113 	struct vop_generic_args a_gen;
1114 	struct vnode *a_vp;
1115 	struct vattr *a_vap;
1116 	struct ucred *a_cred;
1117 	struct thread *a_td;
1118 };
1119 #endif
1120 
1121 /*
1122  * Get file attributes
1123  */
1124 static int
1125 mqfs_getattr(struct vop_getattr_args *ap)
1126 {
1127 	struct vnode *vp = ap->a_vp;
1128 	struct mqfs_node *pn = VTON(vp);
1129 	struct vattr *vap = ap->a_vap;
1130 	int error = 0;
1131 
1132 	VATTR_NULL(vap);
1133 	vap->va_type = vp->v_type;
1134 	vap->va_mode = pn->mn_mode;
1135 	vap->va_nlink = 1;
1136 	vap->va_uid = pn->mn_uid;
1137 	vap->va_gid = pn->mn_gid;
1138 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1139 	vap->va_fileid = pn->mn_fileno;
1140 	vap->va_size = 0;
1141 	vap->va_blocksize = PAGE_SIZE;
1142 	vap->va_bytes = vap->va_size = 0;
1143 	vap->va_atime = pn->mn_atime;
1144 	vap->va_mtime = pn->mn_mtime;
1145 	vap->va_ctime = pn->mn_ctime;
1146 	vap->va_birthtime = pn->mn_birth;
1147 	vap->va_gen = 0;
1148 	vap->va_flags = 0;
1149 	vap->va_rdev = 0;
1150 	vap->va_bytes = 0;
1151 	vap->va_filerev = 0;
1152 	vap->va_vaflags = 0;
1153 	return (error);
1154 }
1155 
1156 #if 0
1157 struct vop_setattr_args {
1158 	struct vop_generic_args a_gen;
1159 	struct vnode *a_vp;
1160 	struct vattr *a_vap;
1161 	struct ucred *a_cred;
1162 	struct thread *a_td;
1163 };
1164 #endif
1165 /*
1166  * Set attributes
1167  */
1168 static int
1169 mqfs_setattr(struct vop_setattr_args *ap)
1170 {
1171 	struct mqfs_node *pn;
1172 	struct vattr *vap;
1173 	struct vnode *vp;
1174 	int c, error;
1175 	uid_t uid;
1176 	gid_t gid;
1177 
1178 	vap = ap->a_vap;
1179 	vp = ap->a_vp;
1180 	if ((vap->va_type != VNON) ||
1181 	    (vap->va_nlink != VNOVAL) ||
1182 	    (vap->va_fsid != VNOVAL) ||
1183 	    (vap->va_fileid != VNOVAL) ||
1184 	    (vap->va_blocksize != VNOVAL) ||
1185 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1186 	    (vap->va_rdev != VNOVAL) ||
1187 	    ((int)vap->va_bytes != VNOVAL) ||
1188 	    (vap->va_gen != VNOVAL)) {
1189 		return (EINVAL);
1190 	}
1191 
1192 	pn = VTON(vp);
1193 
1194 	error = c = 0;
1195 	if (vap->va_uid == (uid_t)VNOVAL)
1196 		uid = pn->mn_uid;
1197 	else
1198 		uid = vap->va_uid;
1199 	if (vap->va_gid == (gid_t)VNOVAL)
1200 		gid = pn->mn_gid;
1201 	else
1202 		gid = vap->va_gid;
1203 
1204 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1205 		/*
1206 		 * To modify the ownership of a file, must possess VADMIN
1207 		 * for that file.
1208 		 */
1209 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1210 			return (error);
1211 
1212 		/*
1213 		 * XXXRW: Why is there a privilege check here: shouldn't the
1214 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1215 		 * below definitely right?
1216 		 */
1217 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1218 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1219 		    (error = priv_check(ap->a_td, PRIV_MQ_ADMIN)) != 0)
1220 			return (error);
1221 		pn->mn_uid = uid;
1222 		pn->mn_gid = gid;
1223 		c = 1;
1224 	}
1225 
1226 	if (vap->va_mode != (mode_t)VNOVAL) {
1227 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1228 		    (error = priv_check(ap->a_td, PRIV_MQ_ADMIN)))
1229 			return (error);
1230 		pn->mn_mode = vap->va_mode;
1231 		c = 1;
1232 	}
1233 
1234 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1235 		/* See the comment in ufs_vnops::ufs_setattr(). */
1236 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1237 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1238 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1239 			return (error);
1240 		if (vap->va_atime.tv_sec != VNOVAL) {
1241 			pn->mn_atime = vap->va_atime;
1242 		}
1243 		if (vap->va_mtime.tv_sec != VNOVAL) {
1244 			pn->mn_mtime = vap->va_mtime;
1245 		}
1246 		c = 1;
1247 	}
1248 	if (c) {
1249 		vfs_timestamp(&pn->mn_ctime);
1250 	}
1251 	return (0);
1252 }
1253 
1254 #if 0
1255 struct vop_read_args {
1256 	struct vop_generic_args a_gen;
1257 	struct vnode *a_vp;
1258 	struct uio *a_uio;
1259 	int a_ioflag;
1260 	struct ucred *a_cred;
1261 };
1262 #endif
1263 
1264 /*
1265  * Read from a file
1266  */
1267 static int
1268 mqfs_read(struct vop_read_args *ap)
1269 {
1270 	char buf[80];
1271 	struct vnode *vp = ap->a_vp;
1272 	struct uio *uio = ap->a_uio;
1273 	struct mqfs_node *pn;
1274 	struct mqueue *mq;
1275 	int len, error;
1276 
1277 	if (vp->v_type != VREG)
1278 		return (EINVAL);
1279 
1280 	pn = VTON(vp);
1281 	mq = VTOMQ(vp);
1282 	snprintf(buf, sizeof(buf),
1283 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1284 		mq->mq_totalbytes,
1285 		mq->mq_maxmsg,
1286 		mq->mq_curmsgs,
1287 		mq->mq_msgsize);
1288 	buf[sizeof(buf)-1] = '\0';
1289 	len = strlen(buf);
1290 	error = uiomove_frombuf(buf, len, uio);
1291 	return (error);
1292 }
1293 
1294 #if 0
1295 struct vop_readdir_args {
1296 	struct vop_generic_args a_gen;
1297 	struct vnode *a_vp;
1298 	struct uio *a_uio;
1299 	struct ucred *a_cred;
1300 	int *a_eofflag;
1301 	int *a_ncookies;
1302 	u_long **a_cookies;
1303 };
1304 #endif
1305 
1306 /*
1307  * Return directory entries.
1308  */
1309 static int
1310 mqfs_readdir(struct vop_readdir_args *ap)
1311 {
1312 	struct vnode *vp;
1313 	struct mqfs_info *mi;
1314 	struct mqfs_node *pd;
1315 	struct mqfs_node *pn;
1316 	struct dirent entry;
1317 	struct uio *uio;
1318 	int *tmp_ncookies = NULL;
1319 	off_t offset;
1320 	int error, i;
1321 
1322 	vp = ap->a_vp;
1323 	mi = VFSTOMQFS(vp->v_mount);
1324 	pd = VTON(vp);
1325 	uio = ap->a_uio;
1326 
1327 	if (vp->v_type != VDIR)
1328 		return (ENOTDIR);
1329 
1330 	if (uio->uio_offset < 0)
1331 		return (EINVAL);
1332 
1333 	if (ap->a_ncookies != NULL) {
1334 		tmp_ncookies = ap->a_ncookies;
1335 		*ap->a_ncookies = 0;
1336 		ap->a_ncookies = NULL;
1337         }
1338 
1339 	error = 0;
1340 	offset = 0;
1341 
1342 	sx_xlock(&mi->mi_lock);
1343 
1344 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1345 		entry.d_reclen = sizeof(entry);
1346 		if (!pn->mn_fileno)
1347 			mqfs_fileno_alloc(mi, pn);
1348 		entry.d_fileno = pn->mn_fileno;
1349 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1350 			entry.d_name[i] = pn->mn_name[i];
1351 		entry.d_name[i] = 0;
1352 		entry.d_namlen = i;
1353 		switch (pn->mn_type) {
1354 		case mqfstype_root:
1355 		case mqfstype_dir:
1356 		case mqfstype_this:
1357 		case mqfstype_parent:
1358 			entry.d_type = DT_DIR;
1359 			break;
1360 		case mqfstype_file:
1361 			entry.d_type = DT_REG;
1362 			break;
1363 		case mqfstype_symlink:
1364 			entry.d_type = DT_LNK;
1365 			break;
1366 		default:
1367 			panic("%s has unexpected node type: %d", pn->mn_name,
1368 				pn->mn_type);
1369 		}
1370 		if (entry.d_reclen > uio->uio_resid)
1371                         break;
1372 		if (offset >= uio->uio_offset) {
1373 			error = vfs_read_dirent(ap, &entry, offset);
1374                         if (error)
1375                                 break;
1376                 }
1377                 offset += entry.d_reclen;
1378 	}
1379 	sx_xunlock(&mi->mi_lock);
1380 
1381 	uio->uio_offset = offset;
1382 
1383 	if (tmp_ncookies != NULL)
1384 		ap->a_ncookies = tmp_ncookies;
1385 
1386 	return (error);
1387 }
1388 
1389 #ifdef notyet
1390 
1391 #if 0
1392 struct vop_mkdir_args {
1393 	struct vnode *a_dvp;
1394 	struvt vnode **a_vpp;
1395 	struvt componentname *a_cnp;
1396 	struct vattr *a_vap;
1397 };
1398 #endif
1399 
1400 /*
1401  * Create a directory.
1402  */
1403 static int
1404 mqfs_mkdir(struct vop_mkdir_args *ap)
1405 {
1406 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1407 	struct componentname *cnp = ap->a_cnp;
1408 	struct mqfs_node *pd = VTON(ap->a_dvp);
1409 	struct mqfs_node *pn;
1410 	int error;
1411 
1412 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1413 		return (ENOTDIR);
1414 	sx_xlock(&mqfs->mi_lock);
1415 #if 0
1416 	/* named node */
1417 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1418 	if (pn != NULL) {
1419 		sx_xunlock(&mqfs->mi_lock);
1420 		return (EEXIST);
1421 	}
1422 #else
1423 	if ((cnp->cn_flags & HASBUF) == 0)
1424 		panic("%s: no name", __func__);
1425 #endif
1426 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1427 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1428 	if (pn == NULL)
1429 		error = ENOSPC;
1430 	else
1431 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1432 	sx_xunlock(&mqfs->mi_lock);
1433 	return (error);
1434 }
1435 
1436 #if 0
1437 struct vop_rmdir_args {
1438 	struct vnode *a_dvp;
1439 	struct vnode *a_vp;
1440 	struct componentname *a_cnp;
1441 };
1442 #endif
1443 
1444 /*
1445  * Remove a directory.
1446  */
1447 static int
1448 mqfs_rmdir(struct vop_rmdir_args *ap)
1449 {
1450 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1451 	struct mqfs_node *pn = VTON(ap->a_vp);
1452 	struct mqfs_node *pt;
1453 
1454 	if (pn->mn_type != mqfstype_dir)
1455 		return (ENOTDIR);
1456 
1457 	sx_xlock(&mqfs->mi_lock);
1458 	if (pn->mn_deleted) {
1459 		sx_xunlock(&mqfs->mi_lock);
1460 		return (ENOENT);
1461 	}
1462 
1463 	pt = LIST_FIRST(&pn->mn_children);
1464 	pt = LIST_NEXT(pt, mn_sibling);
1465 	pt = LIST_NEXT(pt, mn_sibling);
1466 	if (pt != NULL) {
1467 		sx_xunlock(&mqfs->mi_lock);
1468 		return (ENOTEMPTY);
1469 	}
1470 	pt = pn->mn_parent;
1471 	pn->mn_parent = NULL;
1472 	pn->mn_deleted = 1;
1473 	LIST_REMOVE(pn, mn_sibling);
1474 	mqnode_release(pn);
1475 	mqnode_release(pt);
1476 	sx_xunlock(&mqfs->mi_lock);
1477 	cache_purge(ap->a_vp);
1478 	return (0);
1479 }
1480 
1481 #endif /* notyet */
1482 
1483 /*
1484  * Allocate a message queue
1485  */
1486 static struct mqueue *
1487 mqueue_alloc(const struct mq_attr *attr)
1488 {
1489 	struct mqueue *mq;
1490 
1491 	if (curmq >= maxmq)
1492 		return (NULL);
1493 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1494 	TAILQ_INIT(&mq->mq_msgq);
1495 	if (attr != NULL) {
1496 		mq->mq_maxmsg = attr->mq_maxmsg;
1497 		mq->mq_msgsize = attr->mq_msgsize;
1498 	} else {
1499 		mq->mq_maxmsg = default_maxmsg;
1500 		mq->mq_msgsize = default_msgsize;
1501 	}
1502 	mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1503 	knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1504 	knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1505 	atomic_add_int(&curmq, 1);
1506 	return (mq);
1507 }
1508 
1509 /*
1510  * Destroy a message queue
1511  */
1512 static void
1513 mqueue_free(struct mqueue *mq)
1514 {
1515 	struct mqueue_msg *msg;
1516 
1517 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1518 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1519 		FREE(msg, M_MQUEUEDATA);
1520 	}
1521 
1522 	mtx_destroy(&mq->mq_mutex);
1523 	knlist_destroy(&mq->mq_rsel.si_note);
1524 	knlist_destroy(&mq->mq_wsel.si_note);
1525 	uma_zfree(mqueue_zone, mq);
1526 	atomic_add_int(&curmq, -1);
1527 }
1528 
1529 /*
1530  * Load a message from user space
1531  */
1532 static struct mqueue_msg *
1533 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1534 {
1535 	struct mqueue_msg *msg;
1536 	size_t len;
1537 	int error;
1538 
1539 	len = sizeof(struct mqueue_msg) + msg_size;
1540 	MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1541 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1542 	    msg_size);
1543 	if (error) {
1544 		FREE(msg, M_MQUEUEDATA);
1545 		msg = NULL;
1546 	} else {
1547 		msg->msg_size = msg_size;
1548 		msg->msg_prio = msg_prio;
1549 	}
1550 	return (msg);
1551 }
1552 
1553 /*
1554  * Save a message to user space
1555  */
1556 static int
1557 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1558 {
1559 	int error;
1560 
1561 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1562 		msg->msg_size);
1563 	if (error == 0 && msg_prio != NULL)
1564 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1565 	return (error);
1566 }
1567 
1568 /*
1569  * Free a message's memory
1570  */
1571 static __inline void
1572 mqueue_freemsg(struct mqueue_msg *msg)
1573 {
1574 	FREE(msg, M_MQUEUEDATA);
1575 }
1576 
1577 /*
1578  * Send a message. if waitok is false, thread will not be
1579  * blocked if there is no data in queue, otherwise, absolute
1580  * time will be checked.
1581  */
1582 int
1583 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1584 	size_t msg_len, unsigned msg_prio, int waitok,
1585 	const struct timespec *abs_timeout)
1586 {
1587 	struct mqueue_msg *msg;
1588 	struct timespec ets, ts, ts2;
1589 	struct timeval tv;
1590 	int error;
1591 
1592 	if (msg_prio >= MQ_PRIO_MAX)
1593 		return (EINVAL);
1594 	if (msg_len > mq->mq_msgsize)
1595 		return (EMSGSIZE);
1596 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1597 	if (msg == NULL)
1598 		return (EFAULT);
1599 
1600 	/* O_NONBLOCK case */
1601 	if (!waitok) {
1602 		error = _mqueue_send(mq, msg, -1);
1603 		if (error)
1604 			goto bad;
1605 		return (0);
1606 	}
1607 
1608 	/* we allow a null timeout (wait forever) */
1609 	if (abs_timeout == NULL) {
1610 		error = _mqueue_send(mq, msg, 0);
1611 		if (error)
1612 			goto bad;
1613 		return (0);
1614 	}
1615 
1616 	/* send it before checking time */
1617 	error = _mqueue_send(mq, msg, -1);
1618 	if (error == 0)
1619 		return (0);
1620 
1621 	if (error != EAGAIN)
1622 		goto bad;
1623 
1624 	error = copyin(abs_timeout, &ets, sizeof(ets));
1625 	if (error != 0)
1626 		goto bad;
1627 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1628 		error = EINVAL;
1629 		goto bad;
1630 	}
1631 	for (;;) {
1632 		ts2 = ets;
1633 		getnanotime(&ts);
1634 		timespecsub(&ts2, &ts);
1635 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1636 			error = ETIMEDOUT;
1637 			break;
1638 		}
1639 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1640 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1641 		if (error != ETIMEDOUT)
1642 			break;
1643 	}
1644 	if (error == 0)
1645 		return (0);
1646 bad:
1647 	mqueue_freemsg(msg);
1648 	return (error);
1649 }
1650 
1651 /*
1652  * Common routine to send a message
1653  */
1654 static int
1655 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1656 {
1657 	struct mqueue_msg *msg2;
1658 	int error = 0;
1659 
1660 	mtx_lock(&mq->mq_mutex);
1661 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1662 		if (timo < 0) {
1663 			mtx_unlock(&mq->mq_mutex);
1664 			return (EAGAIN);
1665 		}
1666 		mq->mq_senders++;
1667 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1668 			    PCATCH, "mqsend", timo);
1669 		mq->mq_senders--;
1670 		if (error == EAGAIN)
1671 			error = ETIMEDOUT;
1672 	}
1673 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1674 		mtx_unlock(&mq->mq_mutex);
1675 		return (error);
1676 	}
1677 	error = 0;
1678 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1679 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1680 	} else {
1681 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1682 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1683 		} else {
1684 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1685 				if (msg2->msg_prio < msg->msg_prio)
1686 					break;
1687 			}
1688 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1689 		}
1690 	}
1691 	mq->mq_curmsgs++;
1692 	mq->mq_totalbytes += msg->msg_size;
1693 	if (mq->mq_receivers)
1694 		wakeup_one(&mq->mq_receivers);
1695 	else if (mq->mq_notifier != NULL)
1696 		mqueue_send_notification(mq);
1697 	if (mq->mq_flags & MQ_RSEL) {
1698 		mq->mq_flags &= ~MQ_RSEL;
1699 		selwakeup(&mq->mq_rsel);
1700 	}
1701 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1702 	mtx_unlock(&mq->mq_mutex);
1703 	return (0);
1704 }
1705 
1706 /*
1707  * Send realtime a signal to process which registered itself
1708  * successfully by mq_notify.
1709  */
1710 static void
1711 mqueue_send_notification(struct mqueue *mq)
1712 {
1713 	struct mqueue_notifier *nt;
1714 	struct proc *p;
1715 
1716 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1717 	nt = mq->mq_notifier;
1718 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1719 		p = nt->nt_proc;
1720 		PROC_LOCK(p);
1721 		if (!KSI_ONQ(&nt->nt_ksi))
1722 			psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1723 		PROC_UNLOCK(p);
1724 	}
1725 	mq->mq_notifier = NULL;
1726 }
1727 
1728 /*
1729  * Get a message. if waitok is false, thread will not be
1730  * blocked if there is no data in queue, otherwise, absolute
1731  * time will be checked.
1732  */
1733 int
1734 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1735 	size_t msg_len, unsigned *msg_prio, int waitok,
1736 	const struct timespec *abs_timeout)
1737 {
1738 	struct mqueue_msg *msg;
1739 	struct timespec ets, ts, ts2;
1740 	struct timeval tv;
1741 	int error;
1742 
1743 	if (msg_len < mq->mq_msgsize)
1744 		return (EMSGSIZE);
1745 
1746 	/* O_NONBLOCK case */
1747 	if (!waitok) {
1748 		error = _mqueue_recv(mq, &msg, -1);
1749 		if (error)
1750 			return (error);
1751 		goto received;
1752 	}
1753 
1754 	/* we allow a null timeout (wait forever). */
1755 	if (abs_timeout == NULL) {
1756 		error = _mqueue_recv(mq, &msg, 0);
1757 		if (error)
1758 			return (error);
1759 		goto received;
1760 	}
1761 
1762 	/* try to get a message before checking time */
1763 	error = _mqueue_recv(mq, &msg, -1);
1764 	if (error == 0)
1765 		goto received;
1766 
1767 	if (error != EAGAIN)
1768 		return (error);
1769 
1770 	error = copyin(abs_timeout, &ets, sizeof(ets));
1771 	if (error != 0)
1772 		return (error);
1773 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1774 		error = EINVAL;
1775 		return (error);
1776 	}
1777 
1778 	for (;;) {
1779 		ts2 = ets;
1780 		getnanotime(&ts);
1781 		timespecsub(&ts2, &ts);
1782 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1783 			error = ETIMEDOUT;
1784 			return (error);
1785 		}
1786 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1787 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1788 		if (error == 0)
1789 			break;
1790 		if (error != ETIMEDOUT)
1791 			return (error);
1792 	}
1793 
1794 received:
1795 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1796 	if (error == 0) {
1797 		curthread->td_retval[0] = msg->msg_size;
1798 		curthread->td_retval[1] = 0;
1799 	}
1800 	mqueue_freemsg(msg);
1801 	return (error);
1802 }
1803 
1804 /*
1805  * Common routine to receive a message
1806  */
1807 static int
1808 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1809 {
1810 	int error = 0;
1811 
1812 	mtx_lock(&mq->mq_mutex);
1813 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1814 		if (timo < 0) {
1815 			mtx_unlock(&mq->mq_mutex);
1816 			return (EAGAIN);
1817 		}
1818 		mq->mq_receivers++;
1819 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1820 			    PCATCH, "mqrecv", timo);
1821 		mq->mq_receivers--;
1822 		if (error == EAGAIN)
1823 			error = ETIMEDOUT;
1824 	}
1825 	if (*msg != NULL) {
1826 		error = 0;
1827 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1828 		mq->mq_curmsgs--;
1829 		mq->mq_totalbytes -= (*msg)->msg_size;
1830 		if (mq->mq_senders)
1831 			wakeup_one(&mq->mq_senders);
1832 		if (mq->mq_flags & MQ_WSEL) {
1833 			mq->mq_flags &= ~MQ_WSEL;
1834 			selwakeup(&mq->mq_wsel);
1835 		}
1836 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1837 	}
1838 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1839 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1840 		mqueue_send_notification(mq);
1841 	}
1842 	mtx_unlock(&mq->mq_mutex);
1843 	return (error);
1844 }
1845 
1846 static __inline struct mqueue_notifier *
1847 notifier_alloc(void)
1848 {
1849 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1850 }
1851 
1852 static __inline void
1853 notifier_free(struct mqueue_notifier *p)
1854 {
1855 	uma_zfree(mqnoti_zone, p);
1856 }
1857 
1858 static struct mqueue_notifier *
1859 notifier_search(struct proc *p, int fd)
1860 {
1861 	struct mqueue_notifier *nt;
1862 
1863 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1864 		if (nt->nt_ksi.ksi_mqd == fd)
1865 			break;
1866 	}
1867 	return (nt);
1868 }
1869 
1870 static __inline void
1871 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1872 {
1873 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1874 }
1875 
1876 static __inline void
1877 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1878 {
1879 	LIST_REMOVE(nt, nt_link);
1880 	notifier_free(nt);
1881 }
1882 
1883 static void
1884 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1885 {
1886 	struct mqueue_notifier *nt;
1887 
1888 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1889 	PROC_LOCK(p);
1890 	nt = notifier_search(p, fd);
1891 	if (nt != NULL) {
1892 		if (mq->mq_notifier == nt)
1893 			mq->mq_notifier = NULL;
1894 		sigqueue_take(&nt->nt_ksi);
1895 		notifier_delete(p, nt);
1896 	}
1897 	PROC_UNLOCK(p);
1898 }
1899 
1900 /*
1901  * Syscall to open a message queue.
1902  */
1903 int
1904 kmq_open(struct thread *td, struct kmq_open_args *uap)
1905 {
1906 	char path[MQFS_NAMELEN + 1];
1907 	struct mq_attr attr, *pattr;
1908 	struct mqfs_node *pn;
1909 	struct filedesc *fdp;
1910 	struct file *fp;
1911 	struct mqueue *mq;
1912 	int fd, error, len, flags, cmode;
1913 
1914 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1915 		return (EINVAL);
1916 
1917 	fdp = td->td_proc->p_fd;
1918 	flags = FFLAGS(uap->flags);
1919 	cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1920 	mq = NULL;
1921 	if ((flags & O_CREAT) && (uap->attr != NULL)) {
1922 		error = copyin(uap->attr, &attr, sizeof(attr));
1923 		if (error)
1924 			return (error);
1925 		if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1926 			return (EINVAL);
1927 		if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1928 			return (EINVAL);
1929 		pattr = &attr;
1930 	} else
1931 		pattr = NULL;
1932 
1933 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1934         if (error)
1935 		return (error);
1936 
1937 	/*
1938 	 * The first character of name must be a slash  (/) character
1939 	 * and the remaining characters of name cannot include any slash
1940 	 * characters.
1941 	 */
1942 	len = strlen(path);
1943 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1944 		return (EINVAL);
1945 
1946 	error = falloc(td, &fp, &fd);
1947 	if (error)
1948 		return (error);
1949 
1950 	sx_xlock(&mqfs_data.mi_lock);
1951 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1952 	if (pn == NULL) {
1953 		if (!(flags & O_CREAT)) {
1954 			error = ENOENT;
1955 		} else {
1956 			mq = mqueue_alloc(pattr);
1957 			if (mq == NULL) {
1958 				error = ENFILE;
1959 			} else {
1960 				pn = mqfs_create_file(mqfs_data.mi_root,
1961 				         path + 1, len - 1, td->td_ucred,
1962 					 cmode);
1963 				if (pn == NULL) {
1964 					error = ENOSPC;
1965 					mqueue_free(mq);
1966 				}
1967 			}
1968 		}
1969 
1970 		if (error == 0) {
1971 			pn->mn_data = mq;
1972 		}
1973 	} else {
1974 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1975 			error = EEXIST;
1976 		} else {
1977 			int acc_mode = 0;
1978 
1979 			if (flags & FREAD)
1980 				acc_mode |= VREAD;
1981 			if (flags & FWRITE)
1982 				acc_mode |= VWRITE;
1983 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1984 				    pn->mn_gid, acc_mode, td->td_ucred, NULL);
1985 		}
1986 	}
1987 
1988 	if (error) {
1989 		sx_xunlock(&mqfs_data.mi_lock);
1990 		fdclose(fdp, fp, fd, td);
1991 		fdrop(fp, td);
1992 		return (error);
1993 	}
1994 
1995 	mqnode_addref(pn);
1996 	sx_xunlock(&mqfs_data.mi_lock);
1997 
1998 	finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
1999 	    &mqueueops);
2000 
2001 	FILEDESC_XLOCK(fdp);
2002 	if (fdp->fd_ofiles[fd] == fp)
2003 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2004 	FILEDESC_XUNLOCK(fdp);
2005 	td->td_retval[0] = fd;
2006 	fdrop(fp, td);
2007 	return (0);
2008 }
2009 
2010 /*
2011  * Syscall to unlink a message queue.
2012  */
2013 int
2014 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2015 {
2016 	char path[MQFS_NAMELEN+1];
2017 	struct mqfs_node *pn;
2018 	int error, len;
2019 
2020 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2021         if (error)
2022 		return (error);
2023 
2024 	len = strlen(path);
2025 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
2026 		return (EINVAL);
2027 
2028 	sx_xlock(&mqfs_data.mi_lock);
2029 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2030 	if (pn != NULL)
2031 		error = do_unlink(pn, td->td_ucred);
2032 	else
2033 		error = ENOENT;
2034 	sx_xunlock(&mqfs_data.mi_lock);
2035 	return (error);
2036 }
2037 
2038 typedef int (*_fgetf)(struct thread *, int, struct file **);
2039 
2040 /*
2041  * Get message queue by giving file slot
2042  */
2043 static int
2044 _getmq(struct thread *td, int fd, _fgetf func,
2045        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2046 {
2047 	struct mqfs_node *pn;
2048 	int error;
2049 
2050 	error = func(td, fd, fpp);
2051 	if (error)
2052 		return (error);
2053 	if (&mqueueops != (*fpp)->f_ops) {
2054 		fdrop(*fpp, td);
2055 		return (EBADF);
2056 	}
2057 	pn = (*fpp)->f_data;
2058 	if (ppn)
2059 		*ppn = pn;
2060 	if (pmq)
2061 		*pmq = pn->mn_data;
2062 	return (0);
2063 }
2064 
2065 static __inline int
2066 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2067 	struct mqueue **pmq)
2068 {
2069 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2070 }
2071 
2072 static __inline int
2073 getmq_read(struct thread *td, int fd, struct file **fpp,
2074 	 struct mqfs_node **ppn, struct mqueue **pmq)
2075 {
2076 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2077 }
2078 
2079 static __inline int
2080 getmq_write(struct thread *td, int fd, struct file **fpp,
2081 	struct mqfs_node **ppn, struct mqueue **pmq)
2082 {
2083 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2084 }
2085 
2086 int
2087 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2088 {
2089 	struct mqueue *mq;
2090 	struct file *fp;
2091 	struct mq_attr attr, oattr;
2092 	u_int oflag, flag;
2093 	int error;
2094 
2095 	if (uap->attr) {
2096 		error = copyin(uap->attr, &attr, sizeof(attr));
2097 		if (error)
2098 			return (error);
2099 		if (attr.mq_flags & ~O_NONBLOCK)
2100 			return (EINVAL);
2101 	}
2102 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2103 	if (error)
2104 		return (error);
2105 	oattr.mq_maxmsg  = mq->mq_maxmsg;
2106 	oattr.mq_msgsize = mq->mq_msgsize;
2107 	oattr.mq_curmsgs = mq->mq_curmsgs;
2108 	if (uap->attr) {
2109 		do {
2110 			oflag = flag = fp->f_flag;
2111 			flag &= ~O_NONBLOCK;
2112 			flag |= (attr.mq_flags & O_NONBLOCK);
2113 		} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
2114 	} else
2115 		oflag = fp->f_flag;
2116 	oattr.mq_flags = (O_NONBLOCK & oflag);
2117 	fdrop(fp, td);
2118 	if (uap->oattr)
2119 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2120 	return (error);
2121 }
2122 
2123 int
2124 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2125 {
2126 	struct mqueue *mq;
2127 	struct file *fp;
2128 	int error;
2129 	int waitok;
2130 
2131 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2132 	if (error)
2133 		return (error);
2134 	waitok = !(fp->f_flag & O_NONBLOCK);
2135 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2136 		uap->msg_prio, waitok, uap->abs_timeout);
2137 	fdrop(fp, td);
2138 	return (error);
2139 }
2140 
2141 int
2142 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2143 {
2144 	struct mqueue *mq;
2145 	struct file *fp;
2146 	int error, waitok;
2147 
2148 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2149 	if (error)
2150 		return (error);
2151 	waitok = !(fp->f_flag & O_NONBLOCK);
2152 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2153 		uap->msg_prio, waitok, uap->abs_timeout);
2154 	fdrop(fp, td);
2155 	return (error);
2156 }
2157 
2158 int
2159 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2160 {
2161 	struct sigevent ev;
2162 	struct filedesc *fdp;
2163 	struct proc *p;
2164 	struct mqueue *mq;
2165 	struct file *fp;
2166 	struct mqueue_notifier *nt, *newnt = NULL;
2167 	int error;
2168 
2169 	p = td->td_proc;
2170 	fdp = td->td_proc->p_fd;
2171 	if (uap->sigev) {
2172 		error = copyin(uap->sigev, &ev, sizeof(ev));
2173 		if (error)
2174 			return (error);
2175 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2176 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2177 		    ev.sigev_notify != SIGEV_NONE)
2178 			return (EINVAL);
2179 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2180 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2181 			!_SIG_VALID(ev.sigev_signo))
2182 			return (EINVAL);
2183 	}
2184 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2185 	if (error)
2186 		return (error);
2187 again:
2188 	FILEDESC_SLOCK(fdp);
2189 	if (fget_locked(fdp, uap->mqd) != fp) {
2190 		FILEDESC_SUNLOCK(fdp);
2191 		error = EBADF;
2192 		goto out;
2193 	}
2194 	mtx_lock(&mq->mq_mutex);
2195 	FILEDESC_SUNLOCK(fdp);
2196 	if (uap->sigev != NULL) {
2197 		if (mq->mq_notifier != NULL) {
2198 			error = EBUSY;
2199 		} else {
2200 			PROC_LOCK(p);
2201 			nt = notifier_search(p, uap->mqd);
2202 			if (nt == NULL) {
2203 				if (newnt == NULL) {
2204 					PROC_UNLOCK(p);
2205 					mtx_unlock(&mq->mq_mutex);
2206 					newnt = notifier_alloc();
2207 					goto again;
2208 				}
2209 			}
2210 
2211 			if (nt != NULL) {
2212 				sigqueue_take(&nt->nt_ksi);
2213 				if (newnt != NULL) {
2214 					notifier_free(newnt);
2215 					newnt = NULL;
2216 				}
2217 			} else {
2218 				nt = newnt;
2219 				newnt = NULL;
2220 				ksiginfo_init(&nt->nt_ksi);
2221 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2222 				nt->nt_ksi.ksi_code = SI_MESGQ;
2223 				nt->nt_proc = p;
2224 				nt->nt_ksi.ksi_mqd = uap->mqd;
2225 				notifier_insert(p, nt);
2226 			}
2227 			nt->nt_sigev = ev;
2228 			mq->mq_notifier = nt;
2229 			PROC_UNLOCK(p);
2230 			/*
2231 			 * if there is no receivers and message queue
2232 			 * is not empty, we should send notification
2233 			 * as soon as possible.
2234 			 */
2235 			if (mq->mq_receivers == 0 &&
2236 			    !TAILQ_EMPTY(&mq->mq_msgq))
2237 				mqueue_send_notification(mq);
2238 		}
2239 	} else {
2240 		notifier_remove(p, mq, uap->mqd);
2241 	}
2242 	mtx_unlock(&mq->mq_mutex);
2243 
2244 out:
2245 	fdrop(fp, td);
2246 	if (newnt != NULL)
2247 		notifier_free(newnt);
2248 	return (error);
2249 }
2250 
2251 static void
2252 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2253 {
2254 	struct filedesc *fdp;
2255 	struct mqueue *mq;
2256 
2257 	fdp = td->td_proc->p_fd;
2258 	FILEDESC_LOCK_ASSERT(fdp);
2259 
2260 	if (fp->f_ops == &mqueueops) {
2261 		mq = FPTOMQ(fp);
2262 		mtx_lock(&mq->mq_mutex);
2263 		notifier_remove(td->td_proc, mq, fd);
2264 
2265 		/* have to wakeup thread in same process */
2266 		if (mq->mq_flags & MQ_RSEL) {
2267 			mq->mq_flags &= ~MQ_RSEL;
2268 			selwakeup(&mq->mq_rsel);
2269 		}
2270 		if (mq->mq_flags & MQ_WSEL) {
2271 			mq->mq_flags &= ~MQ_WSEL;
2272 			selwakeup(&mq->mq_wsel);
2273 		}
2274 		mtx_unlock(&mq->mq_mutex);
2275 	}
2276 }
2277 
2278 static void
2279 mq_proc_exit(void *arg __unused, struct proc *p)
2280 {
2281 	struct filedesc *fdp;
2282 	struct file *fp;
2283 	struct mqueue *mq;
2284 	int i;
2285 
2286 	fdp = p->p_fd;
2287 	FILEDESC_SLOCK(fdp);
2288 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2289 		fp = fget_locked(fdp, i);
2290 		if (fp != NULL && fp->f_ops == &mqueueops) {
2291 			mq = FPTOMQ(fp);
2292 			mtx_lock(&mq->mq_mutex);
2293 			notifier_remove(p, FPTOMQ(fp), i);
2294 			mtx_unlock(&mq->mq_mutex);
2295 		}
2296 	}
2297 	FILEDESC_SUNLOCK(fdp);
2298 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2299 }
2300 
2301 static int
2302 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2303 	int flags, struct thread *td)
2304 {
2305 	return (EOPNOTSUPP);
2306 }
2307 
2308 static int
2309 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2310 	int flags, struct thread *td)
2311 {
2312 	return (EOPNOTSUPP);
2313 }
2314 
2315 static int
2316 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred,
2317     struct thread *td)
2318 {
2319 
2320 	return (EINVAL);
2321 }
2322 
2323 static int
2324 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2325 	struct ucred *active_cred, struct thread *td)
2326 {
2327 	return (ENOTTY);
2328 }
2329 
2330 static int
2331 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2332 	struct thread *td)
2333 {
2334 	struct mqueue *mq = FPTOMQ(fp);
2335 	int revents = 0;
2336 
2337 	mtx_lock(&mq->mq_mutex);
2338 	if (events & (POLLIN | POLLRDNORM)) {
2339 		if (mq->mq_curmsgs) {
2340 			revents |= events & (POLLIN | POLLRDNORM);
2341 		} else {
2342 			mq->mq_flags |= MQ_RSEL;
2343 			selrecord(td, &mq->mq_rsel);
2344  		}
2345 	}
2346 	if (events & POLLOUT) {
2347 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2348 			revents |= POLLOUT;
2349 		else {
2350 			mq->mq_flags |= MQ_WSEL;
2351 			selrecord(td, &mq->mq_wsel);
2352 		}
2353 	}
2354 	mtx_unlock(&mq->mq_mutex);
2355 	return (revents);
2356 }
2357 
2358 static int
2359 mqf_close(struct file *fp, struct thread *td)
2360 {
2361 	struct mqfs_node *pn;
2362 
2363 	fp->f_ops = &badfileops;
2364 	pn = fp->f_data;
2365 	fp->f_data = NULL;
2366 	sx_xlock(&mqfs_data.mi_lock);
2367 	mqnode_release(pn);
2368 	sx_xunlock(&mqfs_data.mi_lock);
2369 	return (0);
2370 }
2371 
2372 static int
2373 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2374 	struct thread *td)
2375 {
2376 	struct mqfs_node *pn = fp->f_data;
2377 
2378 	bzero(st, sizeof *st);
2379 	st->st_atimespec = pn->mn_atime;
2380 	st->st_mtimespec = pn->mn_mtime;
2381 	st->st_ctimespec = pn->mn_ctime;
2382 	st->st_birthtimespec = pn->mn_birth;
2383 	st->st_uid = pn->mn_uid;
2384 	st->st_gid = pn->mn_gid;
2385 	st->st_mode = S_IFIFO | pn->mn_mode;
2386 	return (0);
2387 }
2388 
2389 static int
2390 mqf_kqfilter(struct file *fp, struct knote *kn)
2391 {
2392 	struct mqueue *mq = FPTOMQ(fp);
2393 	int error = 0;
2394 
2395 	if (kn->kn_filter == EVFILT_READ) {
2396 		kn->kn_fop = &mq_rfiltops;
2397 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2398 	} else if (kn->kn_filter == EVFILT_WRITE) {
2399 		kn->kn_fop = &mq_wfiltops;
2400 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2401 	} else
2402 		error = EINVAL;
2403 	return (error);
2404 }
2405 
2406 static void
2407 filt_mqdetach(struct knote *kn)
2408 {
2409 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2410 
2411 	if (kn->kn_filter == EVFILT_READ)
2412 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2413 	else if (kn->kn_filter == EVFILT_WRITE)
2414 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2415 	else
2416 		panic("filt_mqdetach");
2417 }
2418 
2419 static int
2420 filt_mqread(struct knote *kn, long hint)
2421 {
2422 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2423 
2424 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2425 	return (mq->mq_curmsgs != 0);
2426 }
2427 
2428 static int
2429 filt_mqwrite(struct knote *kn, long hint)
2430 {
2431 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2432 
2433 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2434 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2435 }
2436 
2437 static struct fileops mqueueops = {
2438 	.fo_read		= mqf_read,
2439 	.fo_write		= mqf_write,
2440 	.fo_truncate		= mqf_truncate,
2441 	.fo_ioctl		= mqf_ioctl,
2442 	.fo_poll		= mqf_poll,
2443 	.fo_kqfilter		= mqf_kqfilter,
2444 	.fo_stat		= mqf_stat,
2445 	.fo_close		= mqf_close
2446 };
2447 
2448 static struct vop_vector mqfs_vnodeops = {
2449 	.vop_default 		= &default_vnodeops,
2450 	.vop_access		= mqfs_access,
2451 	.vop_cachedlookup	= mqfs_lookup,
2452 	.vop_lookup		= vfs_cache_lookup,
2453 	.vop_reclaim		= mqfs_reclaim,
2454 	.vop_create		= mqfs_create,
2455 	.vop_remove		= mqfs_remove,
2456 	.vop_inactive		= mqfs_inactive,
2457 	.vop_open		= mqfs_open,
2458 	.vop_close		= mqfs_close,
2459 	.vop_getattr		= mqfs_getattr,
2460 	.vop_setattr		= mqfs_setattr,
2461 	.vop_read		= mqfs_read,
2462 	.vop_write		= VOP_EOPNOTSUPP,
2463 	.vop_readdir		= mqfs_readdir,
2464 	.vop_mkdir		= VOP_EOPNOTSUPP,
2465 	.vop_rmdir		= VOP_EOPNOTSUPP
2466 };
2467 
2468 static struct vfsops mqfs_vfsops = {
2469 	.vfs_init 		= mqfs_init,
2470 	.vfs_uninit		= mqfs_uninit,
2471 	.vfs_mount		= mqfs_mount,
2472 	.vfs_unmount		= mqfs_unmount,
2473 	.vfs_root		= mqfs_root,
2474 	.vfs_statfs		= mqfs_statfs,
2475 };
2476 
2477 SYSCALL_MODULE_HELPER(kmq_open);
2478 SYSCALL_MODULE_HELPER(kmq_setattr);
2479 SYSCALL_MODULE_HELPER(kmq_timedsend);
2480 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2481 SYSCALL_MODULE_HELPER(kmq_notify);
2482 SYSCALL_MODULE_HELPER(kmq_unlink);
2483 
2484 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2485 MODULE_VERSION(mqueuefs, 1);
2486