xref: /freebsd/sys/kern/uipc_mqueue.c (revision 2b743a9e9ddc6736208dc8ca1ce06ce64ad20a19)
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/limits.h>
60 #include <sys/lock.h>
61 #include <sys/malloc.h>
62 #include <sys/module.h>
63 #include <sys/mount.h>
64 #include <sys/mqueue.h>
65 #include <sys/mutex.h>
66 #include <sys/namei.h>
67 #include <sys/posix4.h>
68 #include <sys/poll.h>
69 #include <sys/priv.h>
70 #include <sys/proc.h>
71 #include <sys/queue.h>
72 #include <sys/sysproto.h>
73 #include <sys/stat.h>
74 #include <sys/syscall.h>
75 #include <sys/syscallsubr.h>
76 #include <sys/sysent.h>
77 #include <sys/sx.h>
78 #include <sys/sysctl.h>
79 #include <sys/taskqueue.h>
80 #include <sys/unistd.h>
81 #include <sys/vnode.h>
82 #include <machine/atomic.h>
83 
84 /*
85  * Limits and constants
86  */
87 #define	MQFS_NAMELEN		NAME_MAX
88 #define MQFS_DELEN		(8 + MQFS_NAMELEN)
89 
90 /* node types */
91 typedef enum {
92 	mqfstype_none = 0,
93 	mqfstype_root,
94 	mqfstype_dir,
95 	mqfstype_this,
96 	mqfstype_parent,
97 	mqfstype_file,
98 	mqfstype_symlink,
99 } mqfs_type_t;
100 
101 struct mqfs_node;
102 
103 /*
104  * mqfs_info: describes a mqfs instance
105  */
106 struct mqfs_info {
107 	struct sx		mi_lock;
108 	struct mqfs_node	*mi_root;
109 	struct unrhdr		*mi_unrhdr;
110 };
111 
112 struct mqfs_vdata {
113 	LIST_ENTRY(mqfs_vdata)	mv_link;
114 	struct mqfs_node	*mv_node;
115 	struct vnode		*mv_vnode;
116 	struct task		mv_task;
117 };
118 
119 /*
120  * mqfs_node: describes a node (file or directory) within a mqfs
121  */
122 struct mqfs_node {
123 	char			mn_name[MQFS_NAMELEN+1];
124 	struct mqfs_info	*mn_info;
125 	struct mqfs_node	*mn_parent;
126 	LIST_HEAD(,mqfs_node)	mn_children;
127 	LIST_ENTRY(mqfs_node)	mn_sibling;
128 	LIST_HEAD(,mqfs_vdata)	mn_vnodes;
129 	int			mn_refcount;
130 	mqfs_type_t		mn_type;
131 	int			mn_deleted;
132 	u_int32_t		mn_fileno;
133 	void			*mn_data;
134 	struct timespec		mn_birth;
135 	struct timespec		mn_ctime;
136 	struct timespec		mn_atime;
137 	struct timespec		mn_mtime;
138 	uid_t			mn_uid;
139 	gid_t			mn_gid;
140 	int			mn_mode;
141 };
142 
143 #define	VTON(vp)	(((struct mqfs_vdata *)((vp)->v_data))->mv_node)
144 #define VTOMQ(vp) 	((struct mqueue *)(VTON(vp)->mn_data))
145 #define	VFSTOMQFS(m)	((struct mqfs_info *)((m)->mnt_data))
146 #define	FPTOMQ(fp)	((struct mqueue *)(((struct mqfs_node *) \
147 				(fp)->f_data)->mn_data))
148 
149 TAILQ_HEAD(msgq, mqueue_msg);
150 
151 struct mqueue;
152 
153 struct mqueue_notifier {
154 	LIST_ENTRY(mqueue_notifier)	nt_link;
155 	struct sigevent			nt_sigev;
156 	ksiginfo_t			nt_ksi;
157 	struct proc			*nt_proc;
158 };
159 
160 struct mqueue {
161 	struct mtx	mq_mutex;
162 	int		mq_flags;
163 	long		mq_maxmsg;
164 	long		mq_msgsize;
165 	long		mq_curmsgs;
166 	long		mq_totalbytes;
167 	struct msgq	mq_msgq;
168 	int		mq_receivers;
169 	int		mq_senders;
170 	struct selinfo	mq_rsel;
171 	struct selinfo	mq_wsel;
172 	struct mqueue_notifier	*mq_notifier;
173 };
174 
175 #define	MQ_RSEL		0x01
176 #define	MQ_WSEL		0x02
177 
178 struct mqueue_msg {
179 	TAILQ_ENTRY(mqueue_msg)	msg_link;
180 	unsigned int	msg_prio;
181 	unsigned int	msg_size;
182 	/* following real data... */
183 };
184 
185 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
186 	"POSIX real time message queue");
187 
188 static int	default_maxmsg  = 10;
189 static int	default_msgsize = 1024;
190 
191 static int	maxmsg = 100;
192 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
193     &maxmsg, 0, "Default maximum messages in queue");
194 static int	maxmsgsize = 16384;
195 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
196     &maxmsgsize, 0, "Default maximum message size");
197 static int	maxmq = 100;
198 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
199     &maxmq, 0, "maximum message queues");
200 static int	curmq = 0;
201 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
202     &curmq, 0, "current message queue number");
203 static int	unloadable = 0;
204 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
205 
206 static eventhandler_tag exit_tag;
207 
208 /* Only one instance per-system */
209 static struct mqfs_info		mqfs_data;
210 static uma_zone_t		mqnode_zone;
211 static uma_zone_t		mqueue_zone;
212 static uma_zone_t		mvdata_zone;
213 static uma_zone_t		mqnoti_zone;
214 static struct vop_vector	mqfs_vnodeops;
215 static struct fileops		mqueueops;
216 
217 /*
218  * Directory structure construction and manipulation
219  */
220 #ifdef notyet
221 static struct mqfs_node	*mqfs_create_dir(struct mqfs_node *parent,
222 	const char *name, int namelen, struct ucred *cred, int mode);
223 static struct mqfs_node	*mqfs_create_link(struct mqfs_node *parent,
224 	const char *name, int namelen, struct ucred *cred, int mode);
225 #endif
226 
227 static struct mqfs_node	*mqfs_create_file(struct mqfs_node *parent,
228 	const char *name, int namelen, struct ucred *cred, int mode);
229 static int	mqfs_destroy(struct mqfs_node *mn);
230 static void	mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
231 static void	mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
232 static int	mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
233 
234 /*
235  * Message queue construction and maniplation
236  */
237 static struct mqueue	*mqueue_alloc(const struct mq_attr *attr);
238 static void	mqueue_free(struct mqueue *mq);
239 static int	mqueue_send(struct mqueue *mq, const char *msg_ptr,
240 			size_t msg_len, unsigned msg_prio, int waitok,
241 			const struct timespec *abs_timeout);
242 static int	mqueue_receive(struct mqueue *mq, char *msg_ptr,
243 			size_t msg_len, unsigned *msg_prio, int waitok,
244 			const struct timespec *abs_timeout);
245 static int	_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
246 			int timo);
247 static int	_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
248 			int timo);
249 static void	mqueue_send_notification(struct mqueue *mq);
250 static void	mqueue_fdclose(struct thread *td, int fd, struct file *fp);
251 static void	mq_proc_exit(void *arg, struct proc *p);
252 
253 /*
254  * kqueue filters
255  */
256 static void	filt_mqdetach(struct knote *kn);
257 static int	filt_mqread(struct knote *kn, long hint);
258 static int	filt_mqwrite(struct knote *kn, long hint);
259 
260 struct filterops mq_rfiltops =
261 	{ 1, NULL, filt_mqdetach, filt_mqread };
262 struct filterops mq_wfiltops =
263 	{ 1, NULL, filt_mqdetach, filt_mqwrite };
264 
265 /*
266  * Initialize fileno bitmap
267  */
268 static void
269 mqfs_fileno_init(struct mqfs_info *mi)
270 {
271 	struct unrhdr *up;
272 
273 	up = new_unrhdr(1, INT_MAX, NULL);
274 	mi->mi_unrhdr = up;
275 }
276 
277 /*
278  * Tear down fileno bitmap
279  */
280 static void
281 mqfs_fileno_uninit(struct mqfs_info *mi)
282 {
283 	struct unrhdr *up;
284 
285 	up = mi->mi_unrhdr;
286 	mi->mi_unrhdr = NULL;
287 	delete_unrhdr(up);
288 }
289 
290 /*
291  * Allocate a file number
292  */
293 static void
294 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
295 {
296 	/* make sure our parent has a file number */
297 	if (mn->mn_parent && !mn->mn_parent->mn_fileno)
298 		mqfs_fileno_alloc(mi, mn->mn_parent);
299 
300 	switch (mn->mn_type) {
301 	case mqfstype_root:
302 	case mqfstype_dir:
303 	case mqfstype_file:
304 	case mqfstype_symlink:
305 		mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
306 		break;
307 	case mqfstype_this:
308 		KASSERT(mn->mn_parent != NULL,
309 		    ("mqfstype_this node has no parent"));
310 		mn->mn_fileno = mn->mn_parent->mn_fileno;
311 		break;
312 	case mqfstype_parent:
313 		KASSERT(mn->mn_parent != NULL,
314 		    ("mqfstype_parent node has no parent"));
315 		if (mn->mn_parent == mi->mi_root) {
316 			mn->mn_fileno = mn->mn_parent->mn_fileno;
317 			break;
318 		}
319 		KASSERT(mn->mn_parent->mn_parent != NULL,
320 		    ("mqfstype_parent node has no grandparent"));
321 		mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
322 		break;
323 	default:
324 		KASSERT(0,
325 		    ("mqfs_fileno_alloc() called for unknown type node: %d",
326 			mn->mn_type));
327 		break;
328 	}
329 }
330 
331 /*
332  * Release a file number
333  */
334 static void
335 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
336 {
337 	switch (mn->mn_type) {
338 	case mqfstype_root:
339 	case mqfstype_dir:
340 	case mqfstype_file:
341 	case mqfstype_symlink:
342 		free_unr(mi->mi_unrhdr, mn->mn_fileno);
343 		break;
344 	case mqfstype_this:
345 	case mqfstype_parent:
346 		/* ignore these, as they don't "own" their file number */
347 		break;
348 	default:
349 		KASSERT(0,
350 		    ("mqfs_fileno_free() called for unknown type node: %d",
351 			mn->mn_type));
352 		break;
353 	}
354 }
355 
356 static __inline struct mqfs_node *
357 mqnode_alloc(void)
358 {
359 	return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
360 }
361 
362 static __inline void
363 mqnode_free(struct mqfs_node *node)
364 {
365 	uma_zfree(mqnode_zone, node);
366 }
367 
368 static __inline void
369 mqnode_addref(struct mqfs_node *node)
370 {
371 	atomic_fetchadd_int(&node->mn_refcount, 1);
372 }
373 
374 static __inline void
375 mqnode_release(struct mqfs_node *node)
376 {
377 	int old, exp;
378 
379 	old = atomic_fetchadd_int(&node->mn_refcount, -1);
380 	if (node->mn_type == mqfstype_dir ||
381 	    node->mn_type == mqfstype_root)
382 		exp = 3; /* include . and .. */
383 	else
384 		exp = 1;
385 	if (old == exp)
386 		mqfs_destroy(node);
387 }
388 
389 /*
390  * Add a node to a directory
391  */
392 static int
393 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
394 {
395 	KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
396 	KASSERT(parent->mn_info != NULL,
397 	    ("%s(): parent has no mn_info", __func__));
398 	KASSERT(parent->mn_type == mqfstype_dir ||
399 	    parent->mn_type == mqfstype_root,
400 	    ("%s(): parent is not a directory", __func__));
401 
402 	node->mn_info = parent->mn_info;
403 	node->mn_parent = parent;
404 	LIST_INIT(&node->mn_children);
405 	LIST_INIT(&node->mn_vnodes);
406 	LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
407 	mqnode_addref(parent);
408 	return (0);
409 }
410 
411 static struct mqfs_node *
412 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
413 	int nodetype)
414 {
415 	struct mqfs_node *node;
416 
417 	node = mqnode_alloc();
418 	strncpy(node->mn_name, name, namelen);
419 	node->mn_type = nodetype;
420 	node->mn_refcount = 1;
421 	getnanotime(&node->mn_birth);
422 	node->mn_ctime = node->mn_atime = node->mn_mtime
423 		= node->mn_birth;
424 	node->mn_uid = cred->cr_uid;
425 	node->mn_gid = cred->cr_gid;
426 	node->mn_mode = mode;
427 	return (node);
428 }
429 
430 /*
431  * Create a file
432  */
433 static struct mqfs_node *
434 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
435 	struct ucred *cred, int mode)
436 {
437 	struct mqfs_node *node;
438 
439 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
440 	if (mqfs_add_node(parent, node) != 0) {
441 		mqnode_free(node);
442 		return (NULL);
443 	}
444 	return (node);
445 }
446 
447 /*
448  * Add . and .. to a directory
449  */
450 static int
451 mqfs_fixup_dir(struct mqfs_node *parent)
452 {
453 	struct mqfs_node *dir;
454 
455 	dir = mqnode_alloc();
456 	dir->mn_name[0] = '.';
457 	dir->mn_type = mqfstype_this;
458 	dir->mn_refcount = 1;
459 	if (mqfs_add_node(parent, dir) != 0) {
460 		mqnode_free(dir);
461 		return (-1);
462 	}
463 
464 	dir = mqnode_alloc();
465 	dir->mn_name[0] = dir->mn_name[1] = '.';
466 	dir->mn_type = mqfstype_parent;
467 	dir->mn_refcount = 1;
468 
469 	if (mqfs_add_node(parent, dir) != 0) {
470 		mqnode_free(dir);
471 		return (-1);
472 	}
473 
474 	return (0);
475 }
476 
477 #ifdef notyet
478 
479 /*
480  * Create a directory
481  */
482 static struct mqfs_node *
483 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
484 	struct ucred *cred, int mode)
485 {
486 	struct mqfs_node *node;
487 
488 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
489 	if (mqfs_add_node(parent, node) != 0) {
490 		mqnode_free(node);
491 		return (NULL);
492 	}
493 
494 	if (mqfs_fixup_dir(node) != 0) {
495 		mqfs_destroy(node);
496 		return (NULL);
497 	}
498 	return (node);
499 }
500 
501 /*
502  * Create a symlink
503  */
504 static struct mqfs_node *
505 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
506 	struct ucred *cred, int mode)
507 {
508 	struct mqfs_node *node;
509 
510 	node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
511 	if (mqfs_add_node(parent, node) != 0) {
512 		mqnode_free(node);
513 		return (NULL);
514 	}
515 	return (node);
516 }
517 
518 #endif
519 
520 /*
521  * Destroy a node or a tree of nodes
522  */
523 static int
524 mqfs_destroy(struct mqfs_node *node)
525 {
526 	struct mqfs_node *parent;
527 
528 	KASSERT(node != NULL,
529 	    ("%s(): node is NULL", __func__));
530 	KASSERT(node->mn_info != NULL,
531 	    ("%s(): node has no mn_info", __func__));
532 
533 	/* destroy children */
534 	if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
535 		while (! LIST_EMPTY(&node->mn_children))
536 			mqfs_destroy(LIST_FIRST(&node->mn_children));
537 
538 	/* unlink from parent */
539 	if ((parent = node->mn_parent) != NULL) {
540 		KASSERT(parent->mn_info == node->mn_info,
541 		    ("%s(): parent has different mn_info", __func__));
542 		LIST_REMOVE(node, mn_sibling);
543 	}
544 
545 	if (node->mn_fileno != 0)
546 		mqfs_fileno_free(node->mn_info, node);
547 	if (node->mn_data != NULL)
548 		mqueue_free(node->mn_data);
549 	mqnode_free(node);
550 	return (0);
551 }
552 
553 /*
554  * Mount a mqfs instance
555  */
556 static int
557 mqfs_mount(struct mount *mp, struct thread *td)
558 {
559 	struct statfs *sbp;
560 
561 	if (mp->mnt_flag & MNT_UPDATE)
562 		return (EOPNOTSUPP);
563 
564 	mp->mnt_data = &mqfs_data;
565 	MNT_ILOCK(mp);
566 	mp->mnt_flag |= MNT_LOCAL;
567 	mp->mnt_kern_flag |= MNTK_MPSAFE;
568 	MNT_IUNLOCK(mp);
569 	vfs_getnewfsid(mp);
570 
571 	sbp = &mp->mnt_stat;
572 	vfs_mountedfrom(mp, "mqueue");
573 	sbp->f_bsize = PAGE_SIZE;
574 	sbp->f_iosize = PAGE_SIZE;
575 	sbp->f_blocks = 1;
576 	sbp->f_bfree = 0;
577 	sbp->f_bavail = 0;
578 	sbp->f_files = 1;
579 	sbp->f_ffree = 0;
580 	return (0);
581 }
582 
583 /*
584  * Unmount a mqfs instance
585  */
586 static int
587 mqfs_unmount(struct mount *mp, int mntflags, struct thread *td)
588 {
589 	int error;
590 
591 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0, td);
592 	return (error);
593 }
594 
595 /*
596  * Return a root vnode
597  */
598 static int
599 mqfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
600 {
601 	struct mqfs_info *mqfs;
602 	int ret;
603 
604 	mqfs = VFSTOMQFS(mp);
605 	sx_xlock(&mqfs->mi_lock);
606 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
607 	sx_xunlock(&mqfs->mi_lock);
608 	return (ret);
609 }
610 
611 /*
612  * Return filesystem stats
613  */
614 static int
615 mqfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
616 {
617 	/* XXX update statistics */
618 	return (0);
619 }
620 
621 /*
622  * Initialize a mqfs instance
623  */
624 static int
625 mqfs_init(struct vfsconf *vfc)
626 {
627 	struct mqfs_node *root;
628 	struct mqfs_info *mi;
629 
630 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
631 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
632 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
633 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
634 	mvdata_zone = uma_zcreate("mvdata",
635 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
636 		NULL, UMA_ALIGN_PTR, 0);
637 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
638 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
639 	mi = &mqfs_data;
640 	sx_init(&mi->mi_lock, "mqfs lock");
641 	/* set up the root diretory */
642 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
643 		mqfstype_root);
644 	root->mn_info = mi;
645 	LIST_INIT(&root->mn_children);
646 	LIST_INIT(&root->mn_vnodes);
647 	mi->mi_root = root;
648 	mqfs_fileno_init(mi);
649 	mqfs_fileno_alloc(mi, root);
650 	mqfs_fixup_dir(root);
651 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
652 	    EVENTHANDLER_PRI_ANY);
653 	mq_fdclose = mqueue_fdclose;
654 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
655 	return (0);
656 }
657 
658 /*
659  * Destroy a mqfs instance
660  */
661 static int
662 mqfs_uninit(struct vfsconf *vfc)
663 {
664 	struct mqfs_info *mi;
665 
666 	if (!unloadable)
667 		return (EOPNOTSUPP);
668 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
669 	mi = &mqfs_data;
670 	mqfs_destroy(mi->mi_root);
671 	mi->mi_root = NULL;
672 	mqfs_fileno_uninit(mi);
673 	sx_destroy(&mi->mi_lock);
674 	uma_zdestroy(mqnode_zone);
675 	uma_zdestroy(mqueue_zone);
676 	uma_zdestroy(mvdata_zone);
677 	uma_zdestroy(mqnoti_zone);
678 	return (0);
679 }
680 
681 /*
682  * task routine
683  */
684 static void
685 do_recycle(void *context, int pending __unused)
686 {
687 	struct vnode *vp = (struct vnode *)context;
688 
689 	vrecycle(vp, curthread);
690 	vdrop(vp);
691 }
692 
693 /*
694  * Allocate a vnode
695  */
696 static int
697 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
698 {
699 	struct mqfs_vdata *vd;
700 	int error;
701 
702 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
703 		if (vd->mv_vnode->v_mount == mp)
704 			break;
705 	}
706 
707 	if (vd != NULL) {
708 		if (vget(vd->mv_vnode, 0, curthread) == 0) {
709 			*vpp = vd->mv_vnode;
710 			vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE,
711 			    curthread);
712 			return (0);
713 		}
714 		/* XXX if this can happen, we're in trouble */
715 	}
716 
717 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
718 	if (error)
719 		return (error);
720 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
721 	(*vpp)->v_data = vd;
722 	vd->mv_vnode = *vpp;
723 	vd->mv_node = pn;
724 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
725 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
726 	mqnode_addref(pn);
727 	switch (pn->mn_type) {
728 	case mqfstype_root:
729 		(*vpp)->v_vflag = VV_ROOT;
730 		/* fall through */
731 	case mqfstype_dir:
732 	case mqfstype_this:
733 	case mqfstype_parent:
734 		(*vpp)->v_type = VDIR;
735 		break;
736 	case mqfstype_file:
737 		(*vpp)->v_type = VREG;
738 		break;
739 	case mqfstype_symlink:
740 		(*vpp)->v_type = VLNK;
741 		break;
742 	case mqfstype_none:
743 		KASSERT(0, ("mqfs_allocf called for null node\n"));
744 	default:
745 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
746 	}
747 	vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
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, cnp->cn_thread);
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, td);
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 	/*
960 	 * XXXRW: Other instances of the message queue primitive are
961 	 * allowed in jail?
962 	 */
963 	if (ucred->cr_uid != pn->mn_uid &&
964 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
965 		error = EACCES;
966 	else if (!pn->mn_deleted) {
967 		parent = pn->mn_parent;
968 		pn->mn_parent = NULL;
969 		pn->mn_deleted = 1;
970 		LIST_REMOVE(pn, mn_sibling);
971 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
972 			cache_purge(vd->mv_vnode);
973 			vhold(vd->mv_vnode);
974 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
975 		}
976 		mqnode_release(pn);
977 		mqnode_release(parent);
978 	} else
979 		error = ENOENT;
980 	return (error);
981 }
982 
983 #if 0
984 struct vop_remove_args {
985 	struct vnode *a_dvp;
986 	struct vnode *a_vp;
987 	struct componentname *a_cnp;
988 };
989 #endif
990 
991 /*
992  * vnode removal operation
993  */
994 static int
995 mqfs_remove(struct vop_remove_args *ap)
996 {
997 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
998 	struct mqfs_node *pn;
999 	int error;
1000 
1001 	if (ap->a_vp->v_type == VDIR)
1002                 return (EPERM);
1003 	pn = VTON(ap->a_vp);
1004 	sx_xlock(&mqfs->mi_lock);
1005 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1006 	sx_xunlock(&mqfs->mi_lock);
1007 	return (error);
1008 }
1009 
1010 #if 0
1011 struct vop_inactive_args {
1012 	struct vnode *a_vp;
1013 	struct thread *a_td;
1014 };
1015 #endif
1016 
1017 static int
1018 mqfs_inactive(struct vop_inactive_args *ap)
1019 {
1020 	struct mqfs_node *pn = VTON(ap->a_vp);
1021 
1022 	if (pn->mn_deleted)
1023 		vrecycle(ap->a_vp, ap->a_td);
1024 	return (0);
1025 }
1026 
1027 #if 0
1028 struct vop_reclaim_args {
1029 	struct vop_generic_args a_gen;
1030 	struct vnode *a_vp;
1031 	struct thread *a_td;
1032 };
1033 #endif
1034 
1035 static int
1036 mqfs_reclaim(struct vop_reclaim_args *ap)
1037 {
1038 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1039 	struct vnode *vp = ap->a_vp;
1040 	struct mqfs_node *pn;
1041 	struct mqfs_vdata *vd;
1042 
1043 	vd = vp->v_data;
1044 	pn = vd->mv_node;
1045 	sx_xlock(&mqfs->mi_lock);
1046 	vp->v_data = NULL;
1047 	LIST_REMOVE(vd, mv_link);
1048 	uma_zfree(mvdata_zone, vd);
1049 	mqnode_release(pn);
1050 	sx_xunlock(&mqfs->mi_lock);
1051 	return (0);
1052 }
1053 
1054 #if 0
1055 struct vop_open_args {
1056 	struct vop_generic_args a_gen;
1057 	struct vnode *a_vp;
1058 	int a_mode;
1059 	struct ucred *a_cred;
1060 	struct thread *a_td;
1061 	int a_fdidx;
1062 };
1063 #endif
1064 
1065 static int
1066 mqfs_open(struct vop_open_args *ap)
1067 {
1068 	return (0);
1069 }
1070 
1071 #if 0
1072 struct vop_close_args {
1073 	struct vop_generic_args a_gen;
1074 	struct vnode *a_vp;
1075 	int a_fflag;
1076 	struct ucred *a_cred;
1077 	struct thread *a_td;
1078 };
1079 #endif
1080 
1081 static int
1082 mqfs_close(struct vop_close_args *ap)
1083 {
1084 	return (0);
1085 }
1086 
1087 #if 0
1088 struct vop_access_args {
1089 	struct vop_generic_args a_gen;
1090 	struct vnode *a_vp;
1091 	int a_mode;
1092 	struct ucred *a_cred;
1093 	struct thread *a_td;
1094 };
1095 #endif
1096 
1097 /*
1098  * Verify permissions
1099  */
1100 static int
1101 mqfs_access(struct vop_access_args *ap)
1102 {
1103 	struct vnode *vp = ap->a_vp;
1104 	struct vattr vattr;
1105 	int error;
1106 
1107 	error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1108 	if (error)
1109 		return (error);
1110 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1111 	    vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1112 	return (error);
1113 }
1114 
1115 #if 0
1116 struct vop_getattr_args {
1117 	struct vop_generic_args a_gen;
1118 	struct vnode *a_vp;
1119 	struct vattr *a_vap;
1120 	struct ucred *a_cred;
1121 	struct thread *a_td;
1122 };
1123 #endif
1124 
1125 /*
1126  * Get file attributes
1127  */
1128 static int
1129 mqfs_getattr(struct vop_getattr_args *ap)
1130 {
1131 	struct vnode *vp = ap->a_vp;
1132 	struct mqfs_node *pn = VTON(vp);
1133 	struct vattr *vap = ap->a_vap;
1134 	int error = 0;
1135 
1136 	VATTR_NULL(vap);
1137 	vap->va_type = vp->v_type;
1138 	vap->va_mode = pn->mn_mode;
1139 	vap->va_nlink = 1;
1140 	vap->va_uid = pn->mn_uid;
1141 	vap->va_gid = pn->mn_gid;
1142 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1143 	vap->va_fileid = pn->mn_fileno;
1144 	vap->va_size = 0;
1145 	vap->va_blocksize = PAGE_SIZE;
1146 	vap->va_bytes = vap->va_size = 0;
1147 	vap->va_atime = pn->mn_atime;
1148 	vap->va_mtime = pn->mn_mtime;
1149 	vap->va_ctime = pn->mn_ctime;
1150 	vap->va_birthtime = pn->mn_birth;
1151 	vap->va_gen = 0;
1152 	vap->va_flags = 0;
1153 	vap->va_rdev = 0;
1154 	vap->va_bytes = 0;
1155 	vap->va_filerev = 0;
1156 	vap->va_vaflags = 0;
1157 	return (error);
1158 }
1159 
1160 #if 0
1161 struct vop_setattr_args {
1162 	struct vop_generic_args a_gen;
1163 	struct vnode *a_vp;
1164 	struct vattr *a_vap;
1165 	struct ucred *a_cred;
1166 	struct thread *a_td;
1167 };
1168 #endif
1169 /*
1170  * Set attributes
1171  */
1172 static int
1173 mqfs_setattr(struct vop_setattr_args *ap)
1174 {
1175 	struct mqfs_node *pn;
1176 	struct vattr *vap;
1177 	struct vnode *vp;
1178 	int c, error;
1179 	uid_t uid;
1180 	gid_t gid;
1181 
1182 	vap = ap->a_vap;
1183 	vp = ap->a_vp;
1184 	if ((vap->va_type != VNON) ||
1185 	    (vap->va_nlink != VNOVAL) ||
1186 	    (vap->va_fsid != VNOVAL) ||
1187 	    (vap->va_fileid != VNOVAL) ||
1188 	    (vap->va_blocksize != VNOVAL) ||
1189 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1190 	    (vap->va_rdev != VNOVAL) ||
1191 	    ((int)vap->va_bytes != VNOVAL) ||
1192 	    (vap->va_gen != VNOVAL)) {
1193 		return (EINVAL);
1194 	}
1195 
1196 	pn = VTON(vp);
1197 
1198 	error = c = 0;
1199 	if (vap->va_uid == (uid_t)VNOVAL)
1200 		uid = pn->mn_uid;
1201 	else
1202 		uid = vap->va_uid;
1203 	if (vap->va_gid == (gid_t)VNOVAL)
1204 		gid = pn->mn_gid;
1205 	else
1206 		gid = vap->va_gid;
1207 
1208 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1209 		/*
1210 		 * To modify the ownership of a file, must possess VADMIN
1211 		 * for that file.
1212 		 */
1213 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1214 			return (error);
1215 
1216 		/*
1217 		 * XXXRW: Why is there a privilege check here: shouldn't the
1218 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1219 		 * below definitely right?
1220 		 */
1221 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1222 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1223 		    (error = priv_check_cred(ap->a_td->td_ucred,
1224 		    PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)) != 0)
1225 			return (error);
1226 		pn->mn_uid = uid;
1227 		pn->mn_gid = gid;
1228 		c = 1;
1229 	}
1230 
1231 	if (vap->va_mode != (mode_t)VNOVAL) {
1232 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1233 		    (error = priv_check_cred(ap->a_td->td_ucred,
1234 		    PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)))
1235 			return (error);
1236 		pn->mn_mode = vap->va_mode;
1237 		c = 1;
1238 	}
1239 
1240 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1241 		/* See the comment in ufs_vnops::ufs_setattr(). */
1242 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1243 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1244 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1245 			return (error);
1246 		if (vap->va_atime.tv_sec != VNOVAL) {
1247 			pn->mn_atime = vap->va_atime;
1248 		}
1249 		if (vap->va_mtime.tv_sec != VNOVAL) {
1250 			pn->mn_mtime = vap->va_mtime;
1251 		}
1252 		c = 1;
1253 	}
1254 	if (c) {
1255 		vfs_timestamp(&pn->mn_ctime);
1256 	}
1257 	return (0);
1258 }
1259 
1260 #if 0
1261 struct vop_read_args {
1262 	struct vop_generic_args a_gen;
1263 	struct vnode *a_vp;
1264 	struct uio *a_uio;
1265 	int a_ioflag;
1266 	struct ucred *a_cred;
1267 };
1268 #endif
1269 
1270 /*
1271  * Read from a file
1272  */
1273 static int
1274 mqfs_read(struct vop_read_args *ap)
1275 {
1276 	char buf[80];
1277 	struct vnode *vp = ap->a_vp;
1278 	struct uio *uio = ap->a_uio;
1279 	struct mqfs_node *pn;
1280 	struct mqueue *mq;
1281 	int len, error;
1282 
1283 	if (vp->v_type != VREG)
1284 		return (EINVAL);
1285 
1286 	pn = VTON(vp);
1287 	mq = VTOMQ(vp);
1288 	snprintf(buf, sizeof(buf),
1289 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1290 		mq->mq_totalbytes,
1291 		mq->mq_maxmsg,
1292 		mq->mq_curmsgs,
1293 		mq->mq_msgsize);
1294 	buf[sizeof(buf)-1] = '\0';
1295 	len = strlen(buf);
1296 	error = uiomove_frombuf(buf, len, uio);
1297 	return (error);
1298 }
1299 
1300 #if 0
1301 struct vop_readdir_args {
1302 	struct vop_generic_args a_gen;
1303 	struct vnode *a_vp;
1304 	struct uio *a_uio;
1305 	struct ucred *a_cred;
1306 	int *a_eofflag;
1307 	int *a_ncookies;
1308 	u_long **a_cookies;
1309 };
1310 #endif
1311 
1312 /*
1313  * Return directory entries.
1314  */
1315 static int
1316 mqfs_readdir(struct vop_readdir_args *ap)
1317 {
1318 	struct vnode *vp;
1319 	struct mqfs_info *mi;
1320 	struct mqfs_node *pd;
1321 	struct mqfs_node *pn;
1322 	struct dirent entry;
1323 	struct uio *uio;
1324 	int *tmp_ncookies = NULL;
1325 	off_t offset;
1326 	int error, i;
1327 
1328 	vp = ap->a_vp;
1329 	mi = VFSTOMQFS(vp->v_mount);
1330 	pd = VTON(vp);
1331 	uio = ap->a_uio;
1332 
1333 	if (vp->v_type != VDIR)
1334 		return (ENOTDIR);
1335 
1336 	if (uio->uio_offset < 0)
1337 		return (EINVAL);
1338 
1339 	if (ap->a_ncookies != NULL) {
1340 		tmp_ncookies = ap->a_ncookies;
1341 		*ap->a_ncookies = 0;
1342 		ap->a_ncookies = NULL;
1343         }
1344 
1345 	error = 0;
1346 	offset = 0;
1347 
1348 	sx_xlock(&mi->mi_lock);
1349 
1350 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1351 		entry.d_reclen = sizeof(entry);
1352 		if (!pn->mn_fileno)
1353 			mqfs_fileno_alloc(mi, pn);
1354 		entry.d_fileno = pn->mn_fileno;
1355 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1356 			entry.d_name[i] = pn->mn_name[i];
1357 		entry.d_name[i] = 0;
1358 		entry.d_namlen = i;
1359 		switch (pn->mn_type) {
1360 		case mqfstype_root:
1361 		case mqfstype_dir:
1362 		case mqfstype_this:
1363 		case mqfstype_parent:
1364 			entry.d_type = DT_DIR;
1365 			break;
1366 		case mqfstype_file:
1367 			entry.d_type = DT_REG;
1368 			break;
1369 		case mqfstype_symlink:
1370 			entry.d_type = DT_LNK;
1371 			break;
1372 		default:
1373 			panic("%s has unexpected node type: %d", pn->mn_name,
1374 				pn->mn_type);
1375 		}
1376 		if (entry.d_reclen > uio->uio_resid)
1377                         break;
1378 		if (offset >= uio->uio_offset) {
1379 			error = vfs_read_dirent(ap, &entry, offset);
1380                         if (error)
1381                                 break;
1382                 }
1383                 offset += entry.d_reclen;
1384 	}
1385 	sx_xunlock(&mi->mi_lock);
1386 
1387 	uio->uio_offset = offset;
1388 
1389 	if (tmp_ncookies != NULL)
1390 		ap->a_ncookies = tmp_ncookies;
1391 
1392 	return (error);
1393 }
1394 
1395 #ifdef notyet
1396 
1397 #if 0
1398 struct vop_mkdir_args {
1399 	struct vnode *a_dvp;
1400 	struvt vnode **a_vpp;
1401 	struvt componentname *a_cnp;
1402 	struct vattr *a_vap;
1403 };
1404 #endif
1405 
1406 /*
1407  * Create a directory.
1408  */
1409 static int
1410 mqfs_mkdir(struct vop_mkdir_args *ap)
1411 {
1412 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1413 	struct componentname *cnp = ap->a_cnp;
1414 	struct mqfs_node *pd = VTON(ap->a_dvp);
1415 	struct mqfs_node *pn;
1416 	int error;
1417 
1418 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1419 		return (ENOTDIR);
1420 	sx_xlock(&mqfs->mi_lock);
1421 #if 0
1422 	/* named node */
1423 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1424 	if (pn != NULL) {
1425 		sx_xunlock(&mqfs->mi_lock);
1426 		return (EEXIST);
1427 	}
1428 #else
1429 	if ((cnp->cn_flags & HASBUF) == 0)
1430 		panic("%s: no name", __func__);
1431 #endif
1432 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1433 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1434 	if (pn == NULL)
1435 		error = ENOSPC;
1436 	else
1437 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1438 	sx_xunlock(&mqfs->mi_lock);
1439 	return (error);
1440 }
1441 
1442 #if 0
1443 struct vop_rmdir_args {
1444 	struct vnode *a_dvp;
1445 	struct vnode *a_vp;
1446 	struct componentname *a_cnp;
1447 };
1448 #endif
1449 
1450 /*
1451  * Remove a directory.
1452  */
1453 static int
1454 mqfs_rmdir(struct vop_rmdir_args *ap)
1455 {
1456 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1457 	struct mqfs_node *pn = VTON(ap->a_vp);
1458 	struct mqfs_node *pt;
1459 
1460 	if (pn->mn_type != mqfstype_dir)
1461 		return (ENOTDIR);
1462 
1463 	sx_xlock(&mqfs->mi_lock);
1464 	if (pn->mn_deleted) {
1465 		sx_xunlock(&mqfs->mi_lock);
1466 		return (ENOENT);
1467 	}
1468 
1469 	pt = LIST_FIRST(&pn->mn_children);
1470 	pt = LIST_NEXT(pt, mn_sibling);
1471 	pt = LIST_NEXT(pt, mn_sibling);
1472 	if (pt != NULL) {
1473 		sx_xunlock(&mqfs->mi_lock);
1474 		return (ENOTEMPTY);
1475 	}
1476 	pt = pn->mn_parent;
1477 	pn->mn_parent = NULL;
1478 	pn->mn_deleted = 1;
1479 	LIST_REMOVE(pn, mn_sibling);
1480 	mqnode_release(pn);
1481 	mqnode_release(pt);
1482 	sx_xunlock(&mqfs->mi_lock);
1483 	cache_purge(ap->a_vp);
1484 	return (0);
1485 }
1486 
1487 #endif /* notyet */
1488 
1489 /*
1490  * Allocate a message queue
1491  */
1492 static struct mqueue *
1493 mqueue_alloc(const struct mq_attr *attr)
1494 {
1495 	struct mqueue *mq;
1496 
1497 	if (curmq >= maxmq)
1498 		return (NULL);
1499 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1500 	TAILQ_INIT(&mq->mq_msgq);
1501 	if (attr != NULL) {
1502 		mq->mq_maxmsg = attr->mq_maxmsg;
1503 		mq->mq_msgsize = attr->mq_msgsize;
1504 	} else {
1505 		mq->mq_maxmsg = default_maxmsg;
1506 		mq->mq_msgsize = default_msgsize;
1507 	}
1508 	mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1509 	knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1510 	knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1511 	atomic_add_int(&curmq, 1);
1512 	return (mq);
1513 }
1514 
1515 /*
1516  * Destroy a message queue
1517  */
1518 static void
1519 mqueue_free(struct mqueue *mq)
1520 {
1521 	struct mqueue_msg *msg;
1522 
1523 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1524 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1525 		FREE(msg, M_MQUEUEDATA);
1526 	}
1527 
1528 	mtx_destroy(&mq->mq_mutex);
1529 	knlist_destroy(&mq->mq_rsel.si_note);
1530 	knlist_destroy(&mq->mq_wsel.si_note);
1531 	uma_zfree(mqueue_zone, mq);
1532 	atomic_add_int(&curmq, -1);
1533 }
1534 
1535 /*
1536  * Load a message from user space
1537  */
1538 static struct mqueue_msg *
1539 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1540 {
1541 	struct mqueue_msg *msg;
1542 	size_t len;
1543 	int error;
1544 
1545 	len = sizeof(struct mqueue_msg) + msg_size;
1546 	MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1547 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1548 	    msg_size);
1549 	if (error) {
1550 		FREE(msg, M_MQUEUEDATA);
1551 		msg = NULL;
1552 	} else {
1553 		msg->msg_size = msg_size;
1554 		msg->msg_prio = msg_prio;
1555 	}
1556 	return (msg);
1557 }
1558 
1559 /*
1560  * Save a message to user space
1561  */
1562 static int
1563 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1564 {
1565 	int error;
1566 
1567 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1568 		msg->msg_size);
1569 	if (error == 0 && msg_prio != NULL)
1570 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1571 	return (error);
1572 }
1573 
1574 /*
1575  * Free a message's memory
1576  */
1577 static __inline void
1578 mqueue_freemsg(struct mqueue_msg *msg)
1579 {
1580 	FREE(msg, M_MQUEUEDATA);
1581 }
1582 
1583 /*
1584  * Send a message. if waitok is false, thread will not be
1585  * blocked if there is no data in queue, otherwise, absolute
1586  * time will be checked.
1587  */
1588 int
1589 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1590 	size_t msg_len, unsigned msg_prio, int waitok,
1591 	const struct timespec *abs_timeout)
1592 {
1593 	struct mqueue_msg *msg;
1594 	struct timespec ets, ts, ts2;
1595 	struct timeval tv;
1596 	int error;
1597 
1598 	if (msg_prio >= MQ_PRIO_MAX)
1599 		return (EINVAL);
1600 	if (msg_len > mq->mq_msgsize)
1601 		return (EMSGSIZE);
1602 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1603 	if (msg == NULL)
1604 		return (EFAULT);
1605 
1606 	/* O_NONBLOCK case */
1607 	if (!waitok) {
1608 		error = _mqueue_send(mq, msg, -1);
1609 		if (error)
1610 			goto bad;
1611 		return (0);
1612 	}
1613 
1614 	/* we allow a null timeout (wait forever) */
1615 	if (abs_timeout == NULL) {
1616 		error = _mqueue_send(mq, msg, 0);
1617 		if (error)
1618 			goto bad;
1619 		return (0);
1620 	}
1621 
1622 	/* send it before checking time */
1623 	error = _mqueue_send(mq, msg, -1);
1624 	if (error == 0)
1625 		return (0);
1626 
1627 	if (error != EAGAIN)
1628 		goto bad;
1629 
1630 	error = copyin(abs_timeout, &ets, sizeof(ets));
1631 	if (error != 0)
1632 		goto bad;
1633 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1634 		error = EINVAL;
1635 		goto bad;
1636 	}
1637 	for (;;) {
1638 		ts2 = ets;
1639 		getnanotime(&ts);
1640 		timespecsub(&ts2, &ts);
1641 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1642 			error = ETIMEDOUT;
1643 			break;
1644 		}
1645 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1646 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1647 		if (error != ETIMEDOUT)
1648 			break;
1649 	}
1650 	if (error == 0)
1651 		return (0);
1652 bad:
1653 	mqueue_freemsg(msg);
1654 	return (error);
1655 }
1656 
1657 /*
1658  * Common routine to send a message
1659  */
1660 static int
1661 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1662 {
1663 	struct mqueue_msg *msg2;
1664 	int error = 0;
1665 
1666 	mtx_lock(&mq->mq_mutex);
1667 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1668 		if (timo < 0) {
1669 			mtx_unlock(&mq->mq_mutex);
1670 			return (EAGAIN);
1671 		}
1672 		mq->mq_senders++;
1673 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1674 			    PCATCH, "mqsend", timo);
1675 		mq->mq_senders--;
1676 		if (error == EAGAIN)
1677 			error = ETIMEDOUT;
1678 	}
1679 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1680 		mtx_unlock(&mq->mq_mutex);
1681 		return (error);
1682 	}
1683 	error = 0;
1684 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1685 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1686 	} else {
1687 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1688 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1689 		} else {
1690 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1691 				if (msg2->msg_prio < msg->msg_prio)
1692 					break;
1693 			}
1694 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1695 		}
1696 	}
1697 	mq->mq_curmsgs++;
1698 	mq->mq_totalbytes += msg->msg_size;
1699 	if (mq->mq_receivers)
1700 		wakeup_one(&mq->mq_receivers);
1701 	else if (mq->mq_notifier != NULL)
1702 		mqueue_send_notification(mq);
1703 	if (mq->mq_flags & MQ_RSEL) {
1704 		mq->mq_flags &= ~MQ_RSEL;
1705 		selwakeup(&mq->mq_rsel);
1706 	}
1707 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1708 	mtx_unlock(&mq->mq_mutex);
1709 	return (0);
1710 }
1711 
1712 /*
1713  * Send realtime a signal to process which registered itself
1714  * successfully by mq_notify.
1715  */
1716 static void
1717 mqueue_send_notification(struct mqueue *mq)
1718 {
1719 	struct mqueue_notifier *nt;
1720 	struct proc *p;
1721 
1722 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1723 	nt = mq->mq_notifier;
1724 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1725 		p = nt->nt_proc;
1726 		PROC_LOCK(p);
1727 		if (!KSI_ONQ(&nt->nt_ksi))
1728 			psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1729 		PROC_UNLOCK(p);
1730 	}
1731 	mq->mq_notifier = NULL;
1732 }
1733 
1734 /*
1735  * Get a message. if waitok is false, thread will not be
1736  * blocked if there is no data in queue, otherwise, absolute
1737  * time will be checked.
1738  */
1739 int
1740 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1741 	size_t msg_len, unsigned *msg_prio, int waitok,
1742 	const struct timespec *abs_timeout)
1743 {
1744 	struct mqueue_msg *msg;
1745 	struct timespec ets, ts, ts2;
1746 	struct timeval tv;
1747 	int error;
1748 
1749 	if (msg_len < mq->mq_msgsize)
1750 		return (EMSGSIZE);
1751 
1752 	/* O_NONBLOCK case */
1753 	if (!waitok) {
1754 		error = _mqueue_recv(mq, &msg, -1);
1755 		if (error)
1756 			return (error);
1757 		goto received;
1758 	}
1759 
1760 	/* we allow a null timeout (wait forever). */
1761 	if (abs_timeout == NULL) {
1762 		error = _mqueue_recv(mq, &msg, 0);
1763 		if (error)
1764 			return (error);
1765 		goto received;
1766 	}
1767 
1768 	/* try to get a message before checking time */
1769 	error = _mqueue_recv(mq, &msg, -1);
1770 	if (error == 0)
1771 		goto received;
1772 
1773 	if (error != EAGAIN)
1774 		return (error);
1775 
1776 	error = copyin(abs_timeout, &ets, sizeof(ets));
1777 	if (error != 0)
1778 		return (error);
1779 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1780 		error = EINVAL;
1781 		return (error);
1782 	}
1783 
1784 	for (;;) {
1785 		ts2 = ets;
1786 		getnanotime(&ts);
1787 		timespecsub(&ts2, &ts);
1788 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1789 			error = ETIMEDOUT;
1790 			return (error);
1791 		}
1792 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1793 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1794 		if (error == 0)
1795 			break;
1796 		if (error != ETIMEDOUT)
1797 			return (error);
1798 	}
1799 
1800 received:
1801 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1802 	if (error == 0) {
1803 		curthread->td_retval[0] = msg->msg_size;
1804 		curthread->td_retval[1] = 0;
1805 	}
1806 	mqueue_freemsg(msg);
1807 	return (error);
1808 }
1809 
1810 /*
1811  * Common routine to receive a message
1812  */
1813 static int
1814 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1815 {
1816 	int error = 0;
1817 
1818 	mtx_lock(&mq->mq_mutex);
1819 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1820 		if (timo < 0) {
1821 			mtx_unlock(&mq->mq_mutex);
1822 			return (EAGAIN);
1823 		}
1824 		mq->mq_receivers++;
1825 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1826 			    PCATCH, "mqrecv", timo);
1827 		mq->mq_receivers--;
1828 		if (error == EAGAIN)
1829 			error = ETIMEDOUT;
1830 	}
1831 	if (*msg != NULL) {
1832 		error = 0;
1833 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1834 		mq->mq_curmsgs--;
1835 		mq->mq_totalbytes -= (*msg)->msg_size;
1836 		if (mq->mq_senders)
1837 			wakeup_one(&mq->mq_senders);
1838 		if (mq->mq_flags & MQ_WSEL) {
1839 			mq->mq_flags &= ~MQ_WSEL;
1840 			selwakeup(&mq->mq_wsel);
1841 		}
1842 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1843 	}
1844 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1845 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1846 		mqueue_send_notification(mq);
1847 	}
1848 	mtx_unlock(&mq->mq_mutex);
1849 	return (error);
1850 }
1851 
1852 static __inline struct mqueue_notifier *
1853 notifier_alloc(void)
1854 {
1855 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1856 }
1857 
1858 static __inline void
1859 notifier_free(struct mqueue_notifier *p)
1860 {
1861 	uma_zfree(mqnoti_zone, p);
1862 }
1863 
1864 static struct mqueue_notifier *
1865 notifier_search(struct proc *p, int fd)
1866 {
1867 	struct mqueue_notifier *nt;
1868 
1869 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1870 		if (nt->nt_ksi.ksi_mqd == fd)
1871 			break;
1872 	}
1873 	return (nt);
1874 }
1875 
1876 static __inline void
1877 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1878 {
1879 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1880 }
1881 
1882 static __inline void
1883 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1884 {
1885 	LIST_REMOVE(nt, nt_link);
1886 	notifier_free(nt);
1887 }
1888 
1889 static void
1890 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1891 {
1892 	struct mqueue_notifier *nt;
1893 
1894 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1895 	PROC_LOCK(p);
1896 	nt = notifier_search(p, fd);
1897 	if (nt != NULL) {
1898 		if (mq->mq_notifier == nt)
1899 			mq->mq_notifier = NULL;
1900 		sigqueue_take(&nt->nt_ksi);
1901 		notifier_delete(p, nt);
1902 	}
1903 	PROC_UNLOCK(p);
1904 }
1905 
1906 /*
1907  * Syscall to open a message queue
1908  */
1909 int
1910 kmq_open(struct thread *td, struct kmq_open_args *uap)
1911 {
1912 	char path[MQFS_NAMELEN + 1];
1913 	struct mq_attr attr, *pattr;
1914 	struct mqfs_node *pn;
1915 	struct filedesc *fdp;
1916 	struct file *fp;
1917 	struct mqueue *mq;
1918 	int fd, error, len, flags, cmode;
1919 
1920 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1921 		return (EINVAL);
1922 
1923 	fdp = td->td_proc->p_fd;
1924 	flags = FFLAGS(uap->flags);
1925 	cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1926 	mq = NULL;
1927 	if ((flags & O_CREAT) && (uap->attr != NULL)) {
1928 		error = copyin(uap->attr, &attr, sizeof(attr));
1929 		if (error)
1930 			return (error);
1931 		if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1932 			return (EINVAL);
1933 		if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1934 			return (EINVAL);
1935 		pattr = &attr;
1936 	} else
1937 		pattr = NULL;
1938 
1939 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1940         if (error)
1941 		return (error);
1942 
1943 	/*
1944 	 * The first character of name must be a slash  (/) character
1945 	 * and the remaining characters of name cannot include any slash
1946 	 * characters.
1947 	 */
1948 	len = strlen(path);
1949 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1950 		return (EINVAL);
1951 
1952 	error = falloc(td, &fp, &fd);
1953 	if (error)
1954 		return (error);
1955 
1956 	sx_xlock(&mqfs_data.mi_lock);
1957 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1958 	if (pn == NULL) {
1959 		if (!(flags & O_CREAT)) {
1960 			error = ENOENT;
1961 		} else {
1962 			mq = mqueue_alloc(pattr);
1963 			if (mq == NULL) {
1964 				error = ENFILE;
1965 			} else {
1966 				pn = mqfs_create_file(mqfs_data.mi_root,
1967 				         path + 1, len - 1, td->td_ucred,
1968 					 cmode);
1969 				if (pn == NULL) {
1970 					error = ENOSPC;
1971 					mqueue_free(mq);
1972 				}
1973 			}
1974 		}
1975 
1976 		if (error == 0) {
1977 			pn->mn_data = mq;
1978 		}
1979 	} else {
1980 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1981 			error = EEXIST;
1982 		} else {
1983 			int acc_mode = 0;
1984 
1985 			if (flags & FREAD)
1986 				acc_mode |= VREAD;
1987 			if (flags & FWRITE)
1988 				acc_mode |= VWRITE;
1989 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1990 				    pn->mn_gid, acc_mode, td->td_ucred, NULL);
1991 		}
1992 	}
1993 
1994 	if (error) {
1995 		sx_xunlock(&mqfs_data.mi_lock);
1996 		fdclose(fdp, fp, fd, td);
1997 		fdrop(fp, td);
1998 		return (error);
1999 	}
2000 
2001 	mqnode_addref(pn);
2002 	sx_xunlock(&mqfs_data.mi_lock);
2003 
2004 	FILE_LOCK(fp);
2005 	fp->f_flag = (flags & (FREAD | FWRITE | O_NONBLOCK));
2006 	fp->f_type = DTYPE_MQUEUE;
2007 	fp->f_ops = &mqueueops;
2008 	fp->f_data = pn;
2009 	FILE_UNLOCK(fp);
2010 
2011 	FILEDESC_LOCK_FAST(fdp);
2012 	if (fdp->fd_ofiles[fd] == fp)
2013 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2014 	FILEDESC_UNLOCK_FAST(fdp);
2015 	td->td_retval[0] = fd;
2016 	fdrop(fp, td);
2017 	return (0);
2018 }
2019 
2020 /*
2021  * Syscall to unlink a message queue
2022  */
2023 int
2024 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2025 {
2026 	char path[MQFS_NAMELEN+1];
2027 	struct mqfs_node *pn;
2028 	int error, len;
2029 
2030 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2031         if (error)
2032 		return (error);
2033 
2034 	len = strlen(path);
2035 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
2036 		return (EINVAL);
2037 
2038 	sx_xlock(&mqfs_data.mi_lock);
2039 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2040 	if (pn != NULL)
2041 		error = do_unlink(pn, td->td_ucred);
2042 	else
2043 		error = ENOENT;
2044 	sx_xunlock(&mqfs_data.mi_lock);
2045 	return (error);
2046 }
2047 
2048 typedef int (*_fgetf)(struct thread *, int, struct file **);
2049 
2050 /*
2051  * Get message queue by giving file slot
2052  */
2053 static int
2054 _getmq(struct thread *td, int fd, _fgetf func,
2055        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2056 {
2057 	struct mqfs_node *pn;
2058 	int error;
2059 
2060 	error = func(td, fd, fpp);
2061 	if (error)
2062 		return (error);
2063 	if (&mqueueops != (*fpp)->f_ops) {
2064 		fdrop(*fpp, td);
2065 		return (EBADF);
2066 	}
2067 	pn = (*fpp)->f_data;
2068 	if (ppn)
2069 		*ppn = pn;
2070 	if (pmq)
2071 		*pmq = pn->mn_data;
2072 	return (0);
2073 }
2074 
2075 static __inline int
2076 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2077 	struct mqueue **pmq)
2078 {
2079 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2080 }
2081 
2082 static __inline int
2083 getmq_read(struct thread *td, int fd, struct file **fpp,
2084 	 struct mqfs_node **ppn, struct mqueue **pmq)
2085 {
2086 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2087 }
2088 
2089 static __inline int
2090 getmq_write(struct thread *td, int fd, struct file **fpp,
2091 	struct mqfs_node **ppn, struct mqueue **pmq)
2092 {
2093 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2094 }
2095 
2096 /*
2097  * Syscall
2098  */
2099 int
2100 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2101 {
2102 	struct mqueue *mq;
2103 	struct file *fp;
2104 	struct mq_attr attr, oattr;
2105 	int error;
2106 
2107 	if (uap->attr) {
2108 		error = copyin(uap->attr, &attr, sizeof(attr));
2109 		if (error)
2110 			return (error);
2111 		if (attr.mq_flags & ~O_NONBLOCK)
2112 			return (EINVAL);
2113 	}
2114 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2115 	if (error)
2116 		return (error);
2117 	oattr.mq_maxmsg  = mq->mq_maxmsg;
2118 	oattr.mq_msgsize = mq->mq_msgsize;
2119 	oattr.mq_curmsgs = mq->mq_curmsgs;
2120 	FILE_LOCK(fp);
2121 	oattr.mq_flags = (O_NONBLOCK & fp->f_flag);
2122 	if (uap->attr) {
2123 		fp->f_flag &= ~O_NONBLOCK;
2124 		fp->f_flag |= (attr.mq_flags & O_NONBLOCK);
2125 	}
2126 	FILE_UNLOCK(fp);
2127 	fdrop(fp, td);
2128 	if (uap->oattr)
2129 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2130 	return (error);
2131 }
2132 
2133 /*
2134  * Syscall
2135  */
2136 int
2137 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2138 {
2139 	struct mqueue *mq;
2140 	struct file *fp;
2141 	int error;
2142 	int waitok;
2143 
2144 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2145 	if (error)
2146 		return (error);
2147 	waitok = !(fp->f_flag & O_NONBLOCK);
2148 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2149 		uap->msg_prio, waitok, uap->abs_timeout);
2150 	fdrop(fp, td);
2151 	return (error);
2152 }
2153 
2154 /*
2155  * Syscall
2156  */
2157 int
2158 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2159 {
2160 	struct mqueue *mq;
2161 	struct file *fp;
2162 	int error, waitok;
2163 
2164 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2165 	if (error)
2166 		return (error);
2167 	waitok = !(fp->f_flag & O_NONBLOCK);
2168 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2169 		uap->msg_prio, waitok, uap->abs_timeout);
2170 	fdrop(fp, td);
2171 	return (error);
2172 }
2173 
2174 /*
2175  * Syscall
2176  */
2177 int
2178 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2179 {
2180 	struct sigevent ev;
2181 	struct filedesc *fdp;
2182 	struct proc *p;
2183 	struct mqueue *mq;
2184 	struct file *fp;
2185 	struct mqueue_notifier *nt, *newnt = NULL;
2186 	int error;
2187 
2188 	p = td->td_proc;
2189 	fdp = td->td_proc->p_fd;
2190 	if (uap->sigev) {
2191 		error = copyin(uap->sigev, &ev, sizeof(ev));
2192 		if (error)
2193 			return (error);
2194 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2195 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2196 		    ev.sigev_notify != SIGEV_NONE)
2197 			return (EINVAL);
2198 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2199 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2200 			!_SIG_VALID(ev.sigev_signo))
2201 			return (EINVAL);
2202 	}
2203 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2204 	if (error)
2205 		return (error);
2206 again:
2207 	FILEDESC_LOCK_FAST(fdp);
2208 	if (fget_locked(fdp, uap->mqd) != fp) {
2209 		FILEDESC_UNLOCK_FAST(fdp);
2210 		error = EBADF;
2211 		goto out;
2212 	}
2213 	mtx_lock(&mq->mq_mutex);
2214 	FILEDESC_UNLOCK_FAST(fdp);
2215 	if (uap->sigev != NULL) {
2216 		if (mq->mq_notifier != NULL) {
2217 			error = EBUSY;
2218 		} else {
2219 			PROC_LOCK(p);
2220 			nt = notifier_search(p, uap->mqd);
2221 			if (nt == NULL) {
2222 				if (newnt == NULL) {
2223 					PROC_UNLOCK(p);
2224 					mtx_unlock(&mq->mq_mutex);
2225 					newnt = notifier_alloc();
2226 					goto again;
2227 				}
2228 			}
2229 
2230 			if (nt != NULL) {
2231 				sigqueue_take(&nt->nt_ksi);
2232 				if (newnt != NULL) {
2233 					notifier_free(newnt);
2234 					newnt = NULL;
2235 				}
2236 			} else {
2237 				nt = newnt;
2238 				newnt = NULL;
2239 				ksiginfo_init(&nt->nt_ksi);
2240 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2241 				nt->nt_ksi.ksi_code = SI_MESGQ;
2242 				nt->nt_proc = p;
2243 				nt->nt_ksi.ksi_mqd = uap->mqd;
2244 				notifier_insert(p, nt);
2245 			}
2246 			nt->nt_sigev = ev;
2247 			mq->mq_notifier = nt;
2248 			PROC_UNLOCK(p);
2249 			/*
2250 			 * if there is no receivers and message queue
2251 			 * is not empty, we should send notification
2252 			 * as soon as possible.
2253 			 */
2254 			if (mq->mq_receivers == 0 &&
2255 			    !TAILQ_EMPTY(&mq->mq_msgq))
2256 				mqueue_send_notification(mq);
2257 		}
2258 	} else {
2259 		notifier_remove(p, mq, uap->mqd);
2260 	}
2261 	mtx_unlock(&mq->mq_mutex);
2262 
2263 out:
2264 	fdrop(fp, td);
2265 	if (newnt != NULL)
2266 		notifier_free(newnt);
2267 	return (error);
2268 }
2269 
2270 static void
2271 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2272 {
2273 	struct filedesc *fdp;
2274 	struct mqueue *mq;
2275 
2276 	fdp = td->td_proc->p_fd;
2277 	FILEDESC_LOCK_ASSERT(fdp, MA_OWNED);
2278 	if (fp->f_ops == &mqueueops) {
2279 		mq = FPTOMQ(fp);
2280 		mtx_lock(&mq->mq_mutex);
2281 		notifier_remove(td->td_proc, mq, fd);
2282 
2283 		/* have to wakeup thread in same process */
2284 		if (mq->mq_flags & MQ_RSEL) {
2285 			mq->mq_flags &= ~MQ_RSEL;
2286 			selwakeup(&mq->mq_rsel);
2287 		}
2288 		if (mq->mq_flags & MQ_WSEL) {
2289 			mq->mq_flags &= ~MQ_WSEL;
2290 			selwakeup(&mq->mq_wsel);
2291 		}
2292 		mtx_unlock(&mq->mq_mutex);
2293 	}
2294 }
2295 
2296 static void
2297 mq_proc_exit(void *arg __unused, struct proc *p)
2298 {
2299 	struct filedesc *fdp;
2300 	struct file *fp;
2301 	struct mqueue *mq;
2302 	int i;
2303 
2304 	fdp = p->p_fd;
2305 	FILEDESC_LOCK_FAST(fdp);
2306 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2307 		fp = fget_locked(fdp, i);
2308 		if (fp != NULL && fp->f_ops == &mqueueops) {
2309 			mq = FPTOMQ(fp);
2310 			mtx_lock(&mq->mq_mutex);
2311 			notifier_remove(p, FPTOMQ(fp), i);
2312 			mtx_unlock(&mq->mq_mutex);
2313 		}
2314 	}
2315 	FILEDESC_UNLOCK_FAST(fdp);
2316 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2317 }
2318 
2319 static int
2320 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2321 	int flags, struct thread *td)
2322 {
2323 	return (EOPNOTSUPP);
2324 }
2325 
2326 static int
2327 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2328 	int flags, struct thread *td)
2329 {
2330 	return (EOPNOTSUPP);
2331 }
2332 
2333 static int
2334 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2335 	struct ucred *active_cred, struct thread *td)
2336 {
2337 	return (ENOTTY);
2338 }
2339 
2340 static int
2341 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2342 	struct thread *td)
2343 {
2344 	struct mqueue *mq = FPTOMQ(fp);
2345 	int revents = 0;
2346 
2347 	mtx_lock(&mq->mq_mutex);
2348 	if (events & (POLLIN | POLLRDNORM)) {
2349 		if (mq->mq_curmsgs) {
2350 			revents |= events & (POLLIN | POLLRDNORM);
2351 		} else {
2352 			mq->mq_flags |= MQ_RSEL;
2353 			selrecord(td, &mq->mq_rsel);
2354  		}
2355 	}
2356 	if (events & POLLOUT) {
2357 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2358 			revents |= POLLOUT;
2359 		else {
2360 			mq->mq_flags |= MQ_WSEL;
2361 			selrecord(td, &mq->mq_wsel);
2362 		}
2363 	}
2364 	mtx_unlock(&mq->mq_mutex);
2365 	return (revents);
2366 }
2367 
2368 static int
2369 mqf_close(struct file *fp, struct thread *td)
2370 {
2371 	struct mqfs_node *pn;
2372 
2373 	FILE_LOCK(fp);
2374 	fp->f_ops = &badfileops;
2375 	FILE_UNLOCK(fp);
2376 	pn = fp->f_data;
2377 	fp->f_data = NULL;
2378 	sx_xlock(&mqfs_data.mi_lock);
2379 	mqnode_release(pn);
2380 	sx_xunlock(&mqfs_data.mi_lock);
2381 	return (0);
2382 }
2383 
2384 static int
2385 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2386 	struct thread *td)
2387 {
2388 	struct mqfs_node *pn = fp->f_data;
2389 
2390 	bzero(st, sizeof *st);
2391 	st->st_atimespec = pn->mn_atime;
2392 	st->st_mtimespec = pn->mn_mtime;
2393 	st->st_ctimespec = pn->mn_ctime;
2394 	st->st_birthtimespec = pn->mn_birth;
2395 	st->st_uid = pn->mn_uid;
2396 	st->st_gid = pn->mn_gid;
2397 	st->st_mode = S_IFIFO | pn->mn_mode;
2398 	return (0);
2399 }
2400 
2401 static int
2402 mqf_kqfilter(struct file *fp, struct knote *kn)
2403 {
2404 	struct mqueue *mq = FPTOMQ(fp);
2405 	int error = 0;
2406 
2407 	if (kn->kn_filter == EVFILT_READ) {
2408 		kn->kn_fop = &mq_rfiltops;
2409 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2410 	} else if (kn->kn_filter == EVFILT_WRITE) {
2411 		kn->kn_fop = &mq_wfiltops;
2412 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2413 	} else
2414 		error = EINVAL;
2415 	return (error);
2416 }
2417 
2418 static void
2419 filt_mqdetach(struct knote *kn)
2420 {
2421 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2422 
2423 	if (kn->kn_filter == EVFILT_READ)
2424 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2425 	else if (kn->kn_filter == EVFILT_WRITE)
2426 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2427 	else
2428 		panic("filt_mqdetach");
2429 }
2430 
2431 static int
2432 filt_mqread(struct knote *kn, long hint)
2433 {
2434 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2435 
2436 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2437 	return (mq->mq_curmsgs != 0);
2438 }
2439 
2440 static int
2441 filt_mqwrite(struct knote *kn, long hint)
2442 {
2443 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2444 
2445 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2446 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2447 }
2448 
2449 static struct fileops mqueueops = {
2450 	.fo_read		= mqf_read,
2451 	.fo_write		= mqf_write,
2452 	.fo_ioctl		= mqf_ioctl,
2453 	.fo_poll		= mqf_poll,
2454 	.fo_kqfilter		= mqf_kqfilter,
2455 	.fo_stat		= mqf_stat,
2456 	.fo_close		= mqf_close
2457 };
2458 
2459 static struct vop_vector mqfs_vnodeops = {
2460 	.vop_default 		= &default_vnodeops,
2461 	.vop_access		= mqfs_access,
2462 	.vop_cachedlookup	= mqfs_lookup,
2463 	.vop_lookup		= vfs_cache_lookup,
2464 	.vop_reclaim		= mqfs_reclaim,
2465 	.vop_create		= mqfs_create,
2466 	.vop_remove		= mqfs_remove,
2467 	.vop_inactive		= mqfs_inactive,
2468 	.vop_open		= mqfs_open,
2469 	.vop_close		= mqfs_close,
2470 	.vop_getattr		= mqfs_getattr,
2471 	.vop_setattr		= mqfs_setattr,
2472 	.vop_read		= mqfs_read,
2473 	.vop_write		= VOP_EOPNOTSUPP,
2474 	.vop_readdir		= mqfs_readdir,
2475 	.vop_mkdir		= VOP_EOPNOTSUPP,
2476 	.vop_rmdir		= VOP_EOPNOTSUPP
2477 };
2478 
2479 static struct vfsops mqfs_vfsops = {
2480 	.vfs_init 		= mqfs_init,
2481 	.vfs_uninit		= mqfs_uninit,
2482 	.vfs_mount		= mqfs_mount,
2483 	.vfs_unmount		= mqfs_unmount,
2484 	.vfs_root		= mqfs_root,
2485 	.vfs_statfs		= mqfs_statfs,
2486 };
2487 
2488 SYSCALL_MODULE_HELPER(kmq_open);
2489 SYSCALL_MODULE_HELPER(kmq_setattr);
2490 SYSCALL_MODULE_HELPER(kmq_timedsend);
2491 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2492 SYSCALL_MODULE_HELPER(kmq_notify);
2493 SYSCALL_MODULE_HELPER(kmq_unlink);
2494 
2495 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2496 MODULE_VERSION(mqueuefs, 1);
2497