xref: /freebsd/sys/kern/uipc_mqueue.c (revision 35a04710d7286aa9538917fd7f8e417dbee95b82)
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 		if (vget(vd->mv_vnode, 0, curthread) == 0) {
708 			*vpp = vd->mv_vnode;
709 			vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE,
710 			    curthread);
711 			return (0);
712 		}
713 		/* XXX if this can happen, we're in trouble */
714 	}
715 
716 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
717 	if (error)
718 		return (error);
719 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, curthread);
720 	error = insmntque(*vpp, mp);
721 	if (error != 0) {
722 		*vpp = NULLVP;
723 		return (error);
724 	}
725 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
726 	(*vpp)->v_data = vd;
727 	vd->mv_vnode = *vpp;
728 	vd->mv_node = pn;
729 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
730 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
731 	mqnode_addref(pn);
732 	switch (pn->mn_type) {
733 	case mqfstype_root:
734 		(*vpp)->v_vflag = VV_ROOT;
735 		/* fall through */
736 	case mqfstype_dir:
737 	case mqfstype_this:
738 	case mqfstype_parent:
739 		(*vpp)->v_type = VDIR;
740 		break;
741 	case mqfstype_file:
742 		(*vpp)->v_type = VREG;
743 		break;
744 	case mqfstype_symlink:
745 		(*vpp)->v_type = VLNK;
746 		break;
747 	case mqfstype_none:
748 		KASSERT(0, ("mqfs_allocf called for null node\n"));
749 	default:
750 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
751 	}
752 	return (0);
753 }
754 
755 /*
756  * Search a directory entry
757  */
758 static struct mqfs_node *
759 mqfs_search(struct mqfs_node *pd, const char *name, int len)
760 {
761 	struct mqfs_node *pn;
762 
763 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
764 		if (strncmp(pn->mn_name, name, len) == 0)
765 			return (pn);
766 	}
767 	return (NULL);
768 }
769 
770 /*
771  * Look up a file or directory.
772  */
773 static int
774 mqfs_lookupx(struct vop_cachedlookup_args *ap)
775 {
776 	struct componentname *cnp;
777 	struct vnode *dvp, **vpp;
778 	struct mqfs_node *pd;
779 	struct mqfs_node *pn;
780 	int nameiop, flags, error, namelen;
781 	char *pname;
782 	struct thread *td;
783 
784 	cnp = ap->a_cnp;
785 	vpp = ap->a_vpp;
786 	dvp = ap->a_dvp;
787 	pname = cnp->cn_nameptr;
788 	namelen = cnp->cn_namelen;
789 	td = cnp->cn_thread;
790 	flags = cnp->cn_flags;
791 	nameiop = cnp->cn_nameiop;
792 	pd = VTON(dvp);
793 	pn = NULL;
794 	*vpp = NULLVP;
795 
796 	if (dvp->v_type != VDIR)
797 		return (ENOTDIR);
798 
799 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
800 	if (error)
801 		return (error);
802 
803 	/* shortcut: check if the name is too long */
804 	if (cnp->cn_namelen >= MQFS_NAMELEN)
805 		return (ENOENT);
806 
807 	/* self */
808 	if (namelen == 1 && pname[0] == '.') {
809 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
810 			return (EINVAL);
811 		pn = pd;
812 		*vpp = dvp;
813 		VREF(dvp);
814 		return (0);
815 	}
816 
817 	/* parent */
818 	if (cnp->cn_flags & ISDOTDOT) {
819 		if (dvp->v_vflag & VV_ROOT)
820 			return (EIO);
821 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
822 			return (EINVAL);
823 		VOP_UNLOCK(dvp, 0, cnp->cn_thread);
824 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
825 		pn = pd->mn_parent;
826 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
827 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
828 		return (error);
829 	}
830 
831 	/* named node */
832 	pn = mqfs_search(pd, pname, namelen);
833 
834 	/* found */
835 	if (pn != NULL) {
836 		/* DELETE */
837 		if (nameiop == DELETE && (flags & ISLASTCN)) {
838 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
839 			if (error)
840 				return (error);
841 			if (*vpp == dvp) {
842 				VREF(dvp);
843 				*vpp = dvp;
844 				return (0);
845 			}
846 		}
847 
848 		/* allocate vnode */
849 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
850 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
851 			cache_enter(dvp, *vpp, cnp);
852 		return (error);
853 	}
854 
855 	/* not found */
856 
857 	/* will create a new entry in the directory ? */
858 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
859 	    && (flags & ISLASTCN)) {
860 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
861 		if (error)
862 			return (error);
863 		cnp->cn_flags |= SAVENAME;
864 		return (EJUSTRETURN);
865 	}
866 	return (ENOENT);
867 }
868 
869 #if 0
870 struct vop_lookup_args {
871 	struct vop_generic_args a_gen;
872 	struct vnode *a_dvp;
873 	struct vnode **a_vpp;
874 	struct componentname *a_cnp;
875 };
876 #endif
877 
878 /*
879  * vnode lookup operation
880  */
881 static int
882 mqfs_lookup(struct vop_cachedlookup_args *ap)
883 {
884 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
885 	int rc;
886 
887 	sx_xlock(&mqfs->mi_lock);
888 	rc = mqfs_lookupx(ap);
889 	sx_xunlock(&mqfs->mi_lock);
890 	return (rc);
891 }
892 
893 #if 0
894 struct vop_create_args {
895 	struct vnode *a_dvp;
896 	struct vnode **a_vpp;
897 	struct componentname *a_cnp;
898 	struct vattr *a_vap;
899 };
900 #endif
901 
902 /*
903  * vnode creation operation
904  */
905 static int
906 mqfs_create(struct vop_create_args *ap)
907 {
908 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
909 	struct componentname *cnp = ap->a_cnp;
910 	struct mqfs_node *pd;
911 	struct mqfs_node *pn;
912 	struct mqueue *mq;
913 	int error;
914 
915 	pd = VTON(ap->a_dvp);
916 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
917 		return (ENOTDIR);
918 	mq = mqueue_alloc(NULL);
919 	if (mq == NULL)
920 		return (EAGAIN);
921 	sx_xlock(&mqfs->mi_lock);
922 #if 0
923 	/* named node */
924 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
925 	if (pn != NULL) {
926 		mqueue_free(mq);
927 		sx_xunlock(&mqfs->mi_lock);
928 		return (EEXIST);
929 	}
930 #else
931 	if ((cnp->cn_flags & HASBUF) == 0)
932 		panic("%s: no name", __func__);
933 #endif
934 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
935 		cnp->cn_cred, ap->a_vap->va_mode);
936 	if (pn == NULL)
937 		error = ENOSPC;
938 	else {
939 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
940 		if (error)
941 			mqfs_destroy(pn);
942 		else
943 			pn->mn_data = mq;
944 	}
945 	sx_xunlock(&mqfs->mi_lock);
946 	if (error)
947 		mqueue_free(mq);
948 	return (error);
949 }
950 
951 /*
952  * Remove an entry
953  */
954 static
955 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
956 {
957 	struct mqfs_node *parent;
958 	struct mqfs_vdata *vd;
959 	int error = 0;
960 
961 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
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(ap->a_td, PRIV_MQ_ADMIN)) != 0)
1224 			return (error);
1225 		pn->mn_uid = uid;
1226 		pn->mn_gid = gid;
1227 		c = 1;
1228 	}
1229 
1230 	if (vap->va_mode != (mode_t)VNOVAL) {
1231 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1232 		    (error = priv_check(ap->a_td, PRIV_MQ_ADMIN)))
1233 			return (error);
1234 		pn->mn_mode = vap->va_mode;
1235 		c = 1;
1236 	}
1237 
1238 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1239 		/* See the comment in ufs_vnops::ufs_setattr(). */
1240 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1241 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1242 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1243 			return (error);
1244 		if (vap->va_atime.tv_sec != VNOVAL) {
1245 			pn->mn_atime = vap->va_atime;
1246 		}
1247 		if (vap->va_mtime.tv_sec != VNOVAL) {
1248 			pn->mn_mtime = vap->va_mtime;
1249 		}
1250 		c = 1;
1251 	}
1252 	if (c) {
1253 		vfs_timestamp(&pn->mn_ctime);
1254 	}
1255 	return (0);
1256 }
1257 
1258 #if 0
1259 struct vop_read_args {
1260 	struct vop_generic_args a_gen;
1261 	struct vnode *a_vp;
1262 	struct uio *a_uio;
1263 	int a_ioflag;
1264 	struct ucred *a_cred;
1265 };
1266 #endif
1267 
1268 /*
1269  * Read from a file
1270  */
1271 static int
1272 mqfs_read(struct vop_read_args *ap)
1273 {
1274 	char buf[80];
1275 	struct vnode *vp = ap->a_vp;
1276 	struct uio *uio = ap->a_uio;
1277 	struct mqfs_node *pn;
1278 	struct mqueue *mq;
1279 	int len, error;
1280 
1281 	if (vp->v_type != VREG)
1282 		return (EINVAL);
1283 
1284 	pn = VTON(vp);
1285 	mq = VTOMQ(vp);
1286 	snprintf(buf, sizeof(buf),
1287 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1288 		mq->mq_totalbytes,
1289 		mq->mq_maxmsg,
1290 		mq->mq_curmsgs,
1291 		mq->mq_msgsize);
1292 	buf[sizeof(buf)-1] = '\0';
1293 	len = strlen(buf);
1294 	error = uiomove_frombuf(buf, len, uio);
1295 	return (error);
1296 }
1297 
1298 #if 0
1299 struct vop_readdir_args {
1300 	struct vop_generic_args a_gen;
1301 	struct vnode *a_vp;
1302 	struct uio *a_uio;
1303 	struct ucred *a_cred;
1304 	int *a_eofflag;
1305 	int *a_ncookies;
1306 	u_long **a_cookies;
1307 };
1308 #endif
1309 
1310 /*
1311  * Return directory entries.
1312  */
1313 static int
1314 mqfs_readdir(struct vop_readdir_args *ap)
1315 {
1316 	struct vnode *vp;
1317 	struct mqfs_info *mi;
1318 	struct mqfs_node *pd;
1319 	struct mqfs_node *pn;
1320 	struct dirent entry;
1321 	struct uio *uio;
1322 	int *tmp_ncookies = NULL;
1323 	off_t offset;
1324 	int error, i;
1325 
1326 	vp = ap->a_vp;
1327 	mi = VFSTOMQFS(vp->v_mount);
1328 	pd = VTON(vp);
1329 	uio = ap->a_uio;
1330 
1331 	if (vp->v_type != VDIR)
1332 		return (ENOTDIR);
1333 
1334 	if (uio->uio_offset < 0)
1335 		return (EINVAL);
1336 
1337 	if (ap->a_ncookies != NULL) {
1338 		tmp_ncookies = ap->a_ncookies;
1339 		*ap->a_ncookies = 0;
1340 		ap->a_ncookies = NULL;
1341         }
1342 
1343 	error = 0;
1344 	offset = 0;
1345 
1346 	sx_xlock(&mi->mi_lock);
1347 
1348 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1349 		entry.d_reclen = sizeof(entry);
1350 		if (!pn->mn_fileno)
1351 			mqfs_fileno_alloc(mi, pn);
1352 		entry.d_fileno = pn->mn_fileno;
1353 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1354 			entry.d_name[i] = pn->mn_name[i];
1355 		entry.d_name[i] = 0;
1356 		entry.d_namlen = i;
1357 		switch (pn->mn_type) {
1358 		case mqfstype_root:
1359 		case mqfstype_dir:
1360 		case mqfstype_this:
1361 		case mqfstype_parent:
1362 			entry.d_type = DT_DIR;
1363 			break;
1364 		case mqfstype_file:
1365 			entry.d_type = DT_REG;
1366 			break;
1367 		case mqfstype_symlink:
1368 			entry.d_type = DT_LNK;
1369 			break;
1370 		default:
1371 			panic("%s has unexpected node type: %d", pn->mn_name,
1372 				pn->mn_type);
1373 		}
1374 		if (entry.d_reclen > uio->uio_resid)
1375                         break;
1376 		if (offset >= uio->uio_offset) {
1377 			error = vfs_read_dirent(ap, &entry, offset);
1378                         if (error)
1379                                 break;
1380                 }
1381                 offset += entry.d_reclen;
1382 	}
1383 	sx_xunlock(&mi->mi_lock);
1384 
1385 	uio->uio_offset = offset;
1386 
1387 	if (tmp_ncookies != NULL)
1388 		ap->a_ncookies = tmp_ncookies;
1389 
1390 	return (error);
1391 }
1392 
1393 #ifdef notyet
1394 
1395 #if 0
1396 struct vop_mkdir_args {
1397 	struct vnode *a_dvp;
1398 	struvt vnode **a_vpp;
1399 	struvt componentname *a_cnp;
1400 	struct vattr *a_vap;
1401 };
1402 #endif
1403 
1404 /*
1405  * Create a directory.
1406  */
1407 static int
1408 mqfs_mkdir(struct vop_mkdir_args *ap)
1409 {
1410 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1411 	struct componentname *cnp = ap->a_cnp;
1412 	struct mqfs_node *pd = VTON(ap->a_dvp);
1413 	struct mqfs_node *pn;
1414 	int error;
1415 
1416 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1417 		return (ENOTDIR);
1418 	sx_xlock(&mqfs->mi_lock);
1419 #if 0
1420 	/* named node */
1421 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1422 	if (pn != NULL) {
1423 		sx_xunlock(&mqfs->mi_lock);
1424 		return (EEXIST);
1425 	}
1426 #else
1427 	if ((cnp->cn_flags & HASBUF) == 0)
1428 		panic("%s: no name", __func__);
1429 #endif
1430 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1431 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1432 	if (pn == NULL)
1433 		error = ENOSPC;
1434 	else
1435 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1436 	sx_xunlock(&mqfs->mi_lock);
1437 	return (error);
1438 }
1439 
1440 #if 0
1441 struct vop_rmdir_args {
1442 	struct vnode *a_dvp;
1443 	struct vnode *a_vp;
1444 	struct componentname *a_cnp;
1445 };
1446 #endif
1447 
1448 /*
1449  * Remove a directory.
1450  */
1451 static int
1452 mqfs_rmdir(struct vop_rmdir_args *ap)
1453 {
1454 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1455 	struct mqfs_node *pn = VTON(ap->a_vp);
1456 	struct mqfs_node *pt;
1457 
1458 	if (pn->mn_type != mqfstype_dir)
1459 		return (ENOTDIR);
1460 
1461 	sx_xlock(&mqfs->mi_lock);
1462 	if (pn->mn_deleted) {
1463 		sx_xunlock(&mqfs->mi_lock);
1464 		return (ENOENT);
1465 	}
1466 
1467 	pt = LIST_FIRST(&pn->mn_children);
1468 	pt = LIST_NEXT(pt, mn_sibling);
1469 	pt = LIST_NEXT(pt, mn_sibling);
1470 	if (pt != NULL) {
1471 		sx_xunlock(&mqfs->mi_lock);
1472 		return (ENOTEMPTY);
1473 	}
1474 	pt = pn->mn_parent;
1475 	pn->mn_parent = NULL;
1476 	pn->mn_deleted = 1;
1477 	LIST_REMOVE(pn, mn_sibling);
1478 	mqnode_release(pn);
1479 	mqnode_release(pt);
1480 	sx_xunlock(&mqfs->mi_lock);
1481 	cache_purge(ap->a_vp);
1482 	return (0);
1483 }
1484 
1485 #endif /* notyet */
1486 
1487 /*
1488  * Allocate a message queue
1489  */
1490 static struct mqueue *
1491 mqueue_alloc(const struct mq_attr *attr)
1492 {
1493 	struct mqueue *mq;
1494 
1495 	if (curmq >= maxmq)
1496 		return (NULL);
1497 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1498 	TAILQ_INIT(&mq->mq_msgq);
1499 	if (attr != NULL) {
1500 		mq->mq_maxmsg = attr->mq_maxmsg;
1501 		mq->mq_msgsize = attr->mq_msgsize;
1502 	} else {
1503 		mq->mq_maxmsg = default_maxmsg;
1504 		mq->mq_msgsize = default_msgsize;
1505 	}
1506 	mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1507 	knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1508 	knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1509 	atomic_add_int(&curmq, 1);
1510 	return (mq);
1511 }
1512 
1513 /*
1514  * Destroy a message queue
1515  */
1516 static void
1517 mqueue_free(struct mqueue *mq)
1518 {
1519 	struct mqueue_msg *msg;
1520 
1521 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1522 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1523 		FREE(msg, M_MQUEUEDATA);
1524 	}
1525 
1526 	mtx_destroy(&mq->mq_mutex);
1527 	knlist_destroy(&mq->mq_rsel.si_note);
1528 	knlist_destroy(&mq->mq_wsel.si_note);
1529 	uma_zfree(mqueue_zone, mq);
1530 	atomic_add_int(&curmq, -1);
1531 }
1532 
1533 /*
1534  * Load a message from user space
1535  */
1536 static struct mqueue_msg *
1537 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1538 {
1539 	struct mqueue_msg *msg;
1540 	size_t len;
1541 	int error;
1542 
1543 	len = sizeof(struct mqueue_msg) + msg_size;
1544 	MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1545 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1546 	    msg_size);
1547 	if (error) {
1548 		FREE(msg, M_MQUEUEDATA);
1549 		msg = NULL;
1550 	} else {
1551 		msg->msg_size = msg_size;
1552 		msg->msg_prio = msg_prio;
1553 	}
1554 	return (msg);
1555 }
1556 
1557 /*
1558  * Save a message to user space
1559  */
1560 static int
1561 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1562 {
1563 	int error;
1564 
1565 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1566 		msg->msg_size);
1567 	if (error == 0 && msg_prio != NULL)
1568 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1569 	return (error);
1570 }
1571 
1572 /*
1573  * Free a message's memory
1574  */
1575 static __inline void
1576 mqueue_freemsg(struct mqueue_msg *msg)
1577 {
1578 	FREE(msg, M_MQUEUEDATA);
1579 }
1580 
1581 /*
1582  * Send a message. if waitok is false, thread will not be
1583  * blocked if there is no data in queue, otherwise, absolute
1584  * time will be checked.
1585  */
1586 int
1587 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1588 	size_t msg_len, unsigned msg_prio, int waitok,
1589 	const struct timespec *abs_timeout)
1590 {
1591 	struct mqueue_msg *msg;
1592 	struct timespec ets, ts, ts2;
1593 	struct timeval tv;
1594 	int error;
1595 
1596 	if (msg_prio >= MQ_PRIO_MAX)
1597 		return (EINVAL);
1598 	if (msg_len > mq->mq_msgsize)
1599 		return (EMSGSIZE);
1600 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1601 	if (msg == NULL)
1602 		return (EFAULT);
1603 
1604 	/* O_NONBLOCK case */
1605 	if (!waitok) {
1606 		error = _mqueue_send(mq, msg, -1);
1607 		if (error)
1608 			goto bad;
1609 		return (0);
1610 	}
1611 
1612 	/* we allow a null timeout (wait forever) */
1613 	if (abs_timeout == NULL) {
1614 		error = _mqueue_send(mq, msg, 0);
1615 		if (error)
1616 			goto bad;
1617 		return (0);
1618 	}
1619 
1620 	/* send it before checking time */
1621 	error = _mqueue_send(mq, msg, -1);
1622 	if (error == 0)
1623 		return (0);
1624 
1625 	if (error != EAGAIN)
1626 		goto bad;
1627 
1628 	error = copyin(abs_timeout, &ets, sizeof(ets));
1629 	if (error != 0)
1630 		goto bad;
1631 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1632 		error = EINVAL;
1633 		goto bad;
1634 	}
1635 	for (;;) {
1636 		ts2 = ets;
1637 		getnanotime(&ts);
1638 		timespecsub(&ts2, &ts);
1639 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1640 			error = ETIMEDOUT;
1641 			break;
1642 		}
1643 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1644 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1645 		if (error != ETIMEDOUT)
1646 			break;
1647 	}
1648 	if (error == 0)
1649 		return (0);
1650 bad:
1651 	mqueue_freemsg(msg);
1652 	return (error);
1653 }
1654 
1655 /*
1656  * Common routine to send a message
1657  */
1658 static int
1659 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1660 {
1661 	struct mqueue_msg *msg2;
1662 	int error = 0;
1663 
1664 	mtx_lock(&mq->mq_mutex);
1665 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1666 		if (timo < 0) {
1667 			mtx_unlock(&mq->mq_mutex);
1668 			return (EAGAIN);
1669 		}
1670 		mq->mq_senders++;
1671 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1672 			    PCATCH, "mqsend", timo);
1673 		mq->mq_senders--;
1674 		if (error == EAGAIN)
1675 			error = ETIMEDOUT;
1676 	}
1677 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1678 		mtx_unlock(&mq->mq_mutex);
1679 		return (error);
1680 	}
1681 	error = 0;
1682 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1683 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1684 	} else {
1685 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1686 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1687 		} else {
1688 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1689 				if (msg2->msg_prio < msg->msg_prio)
1690 					break;
1691 			}
1692 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1693 		}
1694 	}
1695 	mq->mq_curmsgs++;
1696 	mq->mq_totalbytes += msg->msg_size;
1697 	if (mq->mq_receivers)
1698 		wakeup_one(&mq->mq_receivers);
1699 	else if (mq->mq_notifier != NULL)
1700 		mqueue_send_notification(mq);
1701 	if (mq->mq_flags & MQ_RSEL) {
1702 		mq->mq_flags &= ~MQ_RSEL;
1703 		selwakeup(&mq->mq_rsel);
1704 	}
1705 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1706 	mtx_unlock(&mq->mq_mutex);
1707 	return (0);
1708 }
1709 
1710 /*
1711  * Send realtime a signal to process which registered itself
1712  * successfully by mq_notify.
1713  */
1714 static void
1715 mqueue_send_notification(struct mqueue *mq)
1716 {
1717 	struct mqueue_notifier *nt;
1718 	struct proc *p;
1719 
1720 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1721 	nt = mq->mq_notifier;
1722 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1723 		p = nt->nt_proc;
1724 		PROC_LOCK(p);
1725 		if (!KSI_ONQ(&nt->nt_ksi))
1726 			psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1727 		PROC_UNLOCK(p);
1728 	}
1729 	mq->mq_notifier = NULL;
1730 }
1731 
1732 /*
1733  * Get a message. if waitok is false, thread will not be
1734  * blocked if there is no data in queue, otherwise, absolute
1735  * time will be checked.
1736  */
1737 int
1738 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1739 	size_t msg_len, unsigned *msg_prio, int waitok,
1740 	const struct timespec *abs_timeout)
1741 {
1742 	struct mqueue_msg *msg;
1743 	struct timespec ets, ts, ts2;
1744 	struct timeval tv;
1745 	int error;
1746 
1747 	if (msg_len < mq->mq_msgsize)
1748 		return (EMSGSIZE);
1749 
1750 	/* O_NONBLOCK case */
1751 	if (!waitok) {
1752 		error = _mqueue_recv(mq, &msg, -1);
1753 		if (error)
1754 			return (error);
1755 		goto received;
1756 	}
1757 
1758 	/* we allow a null timeout (wait forever). */
1759 	if (abs_timeout == NULL) {
1760 		error = _mqueue_recv(mq, &msg, 0);
1761 		if (error)
1762 			return (error);
1763 		goto received;
1764 	}
1765 
1766 	/* try to get a message before checking time */
1767 	error = _mqueue_recv(mq, &msg, -1);
1768 	if (error == 0)
1769 		goto received;
1770 
1771 	if (error != EAGAIN)
1772 		return (error);
1773 
1774 	error = copyin(abs_timeout, &ets, sizeof(ets));
1775 	if (error != 0)
1776 		return (error);
1777 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1778 		error = EINVAL;
1779 		return (error);
1780 	}
1781 
1782 	for (;;) {
1783 		ts2 = ets;
1784 		getnanotime(&ts);
1785 		timespecsub(&ts2, &ts);
1786 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1787 			error = ETIMEDOUT;
1788 			return (error);
1789 		}
1790 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1791 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1792 		if (error == 0)
1793 			break;
1794 		if (error != ETIMEDOUT)
1795 			return (error);
1796 	}
1797 
1798 received:
1799 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1800 	if (error == 0) {
1801 		curthread->td_retval[0] = msg->msg_size;
1802 		curthread->td_retval[1] = 0;
1803 	}
1804 	mqueue_freemsg(msg);
1805 	return (error);
1806 }
1807 
1808 /*
1809  * Common routine to receive a message
1810  */
1811 static int
1812 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1813 {
1814 	int error = 0;
1815 
1816 	mtx_lock(&mq->mq_mutex);
1817 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1818 		if (timo < 0) {
1819 			mtx_unlock(&mq->mq_mutex);
1820 			return (EAGAIN);
1821 		}
1822 		mq->mq_receivers++;
1823 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1824 			    PCATCH, "mqrecv", timo);
1825 		mq->mq_receivers--;
1826 		if (error == EAGAIN)
1827 			error = ETIMEDOUT;
1828 	}
1829 	if (*msg != NULL) {
1830 		error = 0;
1831 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1832 		mq->mq_curmsgs--;
1833 		mq->mq_totalbytes -= (*msg)->msg_size;
1834 		if (mq->mq_senders)
1835 			wakeup_one(&mq->mq_senders);
1836 		if (mq->mq_flags & MQ_WSEL) {
1837 			mq->mq_flags &= ~MQ_WSEL;
1838 			selwakeup(&mq->mq_wsel);
1839 		}
1840 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1841 	}
1842 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1843 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1844 		mqueue_send_notification(mq);
1845 	}
1846 	mtx_unlock(&mq->mq_mutex);
1847 	return (error);
1848 }
1849 
1850 static __inline struct mqueue_notifier *
1851 notifier_alloc(void)
1852 {
1853 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1854 }
1855 
1856 static __inline void
1857 notifier_free(struct mqueue_notifier *p)
1858 {
1859 	uma_zfree(mqnoti_zone, p);
1860 }
1861 
1862 static struct mqueue_notifier *
1863 notifier_search(struct proc *p, int fd)
1864 {
1865 	struct mqueue_notifier *nt;
1866 
1867 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1868 		if (nt->nt_ksi.ksi_mqd == fd)
1869 			break;
1870 	}
1871 	return (nt);
1872 }
1873 
1874 static __inline void
1875 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1876 {
1877 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1878 }
1879 
1880 static __inline void
1881 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1882 {
1883 	LIST_REMOVE(nt, nt_link);
1884 	notifier_free(nt);
1885 }
1886 
1887 static void
1888 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1889 {
1890 	struct mqueue_notifier *nt;
1891 
1892 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1893 	PROC_LOCK(p);
1894 	nt = notifier_search(p, fd);
1895 	if (nt != NULL) {
1896 		if (mq->mq_notifier == nt)
1897 			mq->mq_notifier = NULL;
1898 		sigqueue_take(&nt->nt_ksi);
1899 		notifier_delete(p, nt);
1900 	}
1901 	PROC_UNLOCK(p);
1902 }
1903 
1904 /*
1905  * Syscall to open a message queue.
1906  */
1907 int
1908 kmq_open(struct thread *td, struct kmq_open_args *uap)
1909 {
1910 	char path[MQFS_NAMELEN + 1];
1911 	struct mq_attr attr, *pattr;
1912 	struct mqfs_node *pn;
1913 	struct filedesc *fdp;
1914 	struct file *fp;
1915 	struct mqueue *mq;
1916 	int fd, error, len, flags, cmode;
1917 
1918 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1919 		return (EINVAL);
1920 
1921 	fdp = td->td_proc->p_fd;
1922 	flags = FFLAGS(uap->flags);
1923 	cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1924 	mq = NULL;
1925 	if ((flags & O_CREAT) && (uap->attr != NULL)) {
1926 		error = copyin(uap->attr, &attr, sizeof(attr));
1927 		if (error)
1928 			return (error);
1929 		if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1930 			return (EINVAL);
1931 		if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1932 			return (EINVAL);
1933 		pattr = &attr;
1934 	} else
1935 		pattr = NULL;
1936 
1937 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1938         if (error)
1939 		return (error);
1940 
1941 	/*
1942 	 * The first character of name must be a slash  (/) character
1943 	 * and the remaining characters of name cannot include any slash
1944 	 * characters.
1945 	 */
1946 	len = strlen(path);
1947 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1948 		return (EINVAL);
1949 
1950 	error = falloc(td, &fp, &fd);
1951 	if (error)
1952 		return (error);
1953 
1954 	sx_xlock(&mqfs_data.mi_lock);
1955 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1956 	if (pn == NULL) {
1957 		if (!(flags & O_CREAT)) {
1958 			error = ENOENT;
1959 		} else {
1960 			mq = mqueue_alloc(pattr);
1961 			if (mq == NULL) {
1962 				error = ENFILE;
1963 			} else {
1964 				pn = mqfs_create_file(mqfs_data.mi_root,
1965 				         path + 1, len - 1, td->td_ucred,
1966 					 cmode);
1967 				if (pn == NULL) {
1968 					error = ENOSPC;
1969 					mqueue_free(mq);
1970 				}
1971 			}
1972 		}
1973 
1974 		if (error == 0) {
1975 			pn->mn_data = mq;
1976 		}
1977 	} else {
1978 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1979 			error = EEXIST;
1980 		} else {
1981 			int acc_mode = 0;
1982 
1983 			if (flags & FREAD)
1984 				acc_mode |= VREAD;
1985 			if (flags & FWRITE)
1986 				acc_mode |= VWRITE;
1987 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1988 				    pn->mn_gid, acc_mode, td->td_ucred, NULL);
1989 		}
1990 	}
1991 
1992 	if (error) {
1993 		sx_xunlock(&mqfs_data.mi_lock);
1994 		fdclose(fdp, fp, fd, td);
1995 		fdrop(fp, td);
1996 		return (error);
1997 	}
1998 
1999 	mqnode_addref(pn);
2000 	sx_xunlock(&mqfs_data.mi_lock);
2001 
2002 	FILE_LOCK(fp);
2003 	fp->f_flag = (flags & (FREAD | FWRITE | O_NONBLOCK));
2004 	fp->f_type = DTYPE_MQUEUE;
2005 	fp->f_data = pn;
2006 	fp->f_ops = &mqueueops;
2007 	FILE_UNLOCK(fp);
2008 
2009 	FILEDESC_XLOCK(fdp);
2010 	if (fdp->fd_ofiles[fd] == fp)
2011 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2012 	FILEDESC_XUNLOCK(fdp);
2013 	td->td_retval[0] = fd;
2014 	fdrop(fp, td);
2015 	return (0);
2016 }
2017 
2018 /*
2019  * Syscall to unlink a message queue.
2020  */
2021 int
2022 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2023 {
2024 	char path[MQFS_NAMELEN+1];
2025 	struct mqfs_node *pn;
2026 	int error, len;
2027 
2028 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2029         if (error)
2030 		return (error);
2031 
2032 	len = strlen(path);
2033 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
2034 		return (EINVAL);
2035 
2036 	sx_xlock(&mqfs_data.mi_lock);
2037 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2038 	if (pn != NULL)
2039 		error = do_unlink(pn, td->td_ucred);
2040 	else
2041 		error = ENOENT;
2042 	sx_xunlock(&mqfs_data.mi_lock);
2043 	return (error);
2044 }
2045 
2046 typedef int (*_fgetf)(struct thread *, int, struct file **);
2047 
2048 /*
2049  * Get message queue by giving file slot
2050  */
2051 static int
2052 _getmq(struct thread *td, int fd, _fgetf func,
2053        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2054 {
2055 	struct mqfs_node *pn;
2056 	int error;
2057 
2058 	error = func(td, fd, fpp);
2059 	if (error)
2060 		return (error);
2061 	if (&mqueueops != (*fpp)->f_ops) {
2062 		fdrop(*fpp, td);
2063 		return (EBADF);
2064 	}
2065 	pn = (*fpp)->f_data;
2066 	if (ppn)
2067 		*ppn = pn;
2068 	if (pmq)
2069 		*pmq = pn->mn_data;
2070 	return (0);
2071 }
2072 
2073 static __inline int
2074 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2075 	struct mqueue **pmq)
2076 {
2077 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2078 }
2079 
2080 static __inline int
2081 getmq_read(struct thread *td, int fd, struct file **fpp,
2082 	 struct mqfs_node **ppn, struct mqueue **pmq)
2083 {
2084 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2085 }
2086 
2087 static __inline int
2088 getmq_write(struct thread *td, int fd, struct file **fpp,
2089 	struct mqfs_node **ppn, struct mqueue **pmq)
2090 {
2091 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2092 }
2093 
2094 int
2095 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2096 {
2097 	struct mqueue *mq;
2098 	struct file *fp;
2099 	struct mq_attr attr, oattr;
2100 	int error;
2101 
2102 	if (uap->attr) {
2103 		error = copyin(uap->attr, &attr, sizeof(attr));
2104 		if (error)
2105 			return (error);
2106 		if (attr.mq_flags & ~O_NONBLOCK)
2107 			return (EINVAL);
2108 	}
2109 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2110 	if (error)
2111 		return (error);
2112 	oattr.mq_maxmsg  = mq->mq_maxmsg;
2113 	oattr.mq_msgsize = mq->mq_msgsize;
2114 	oattr.mq_curmsgs = mq->mq_curmsgs;
2115 	FILE_LOCK(fp);
2116 	oattr.mq_flags = (O_NONBLOCK & fp->f_flag);
2117 	if (uap->attr) {
2118 		fp->f_flag &= ~O_NONBLOCK;
2119 		fp->f_flag |= (attr.mq_flags & O_NONBLOCK);
2120 	}
2121 	FILE_UNLOCK(fp);
2122 	fdrop(fp, td);
2123 	if (uap->oattr)
2124 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2125 	return (error);
2126 }
2127 
2128 int
2129 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2130 {
2131 	struct mqueue *mq;
2132 	struct file *fp;
2133 	int error;
2134 	int waitok;
2135 
2136 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2137 	if (error)
2138 		return (error);
2139 	waitok = !(fp->f_flag & O_NONBLOCK);
2140 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2141 		uap->msg_prio, waitok, uap->abs_timeout);
2142 	fdrop(fp, td);
2143 	return (error);
2144 }
2145 
2146 int
2147 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2148 {
2149 	struct mqueue *mq;
2150 	struct file *fp;
2151 	int error, waitok;
2152 
2153 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2154 	if (error)
2155 		return (error);
2156 	waitok = !(fp->f_flag & O_NONBLOCK);
2157 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2158 		uap->msg_prio, waitok, uap->abs_timeout);
2159 	fdrop(fp, td);
2160 	return (error);
2161 }
2162 
2163 int
2164 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2165 {
2166 	struct sigevent ev;
2167 	struct filedesc *fdp;
2168 	struct proc *p;
2169 	struct mqueue *mq;
2170 	struct file *fp;
2171 	struct mqueue_notifier *nt, *newnt = NULL;
2172 	int error;
2173 
2174 	p = td->td_proc;
2175 	fdp = td->td_proc->p_fd;
2176 	if (uap->sigev) {
2177 		error = copyin(uap->sigev, &ev, sizeof(ev));
2178 		if (error)
2179 			return (error);
2180 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2181 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2182 		    ev.sigev_notify != SIGEV_NONE)
2183 			return (EINVAL);
2184 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2185 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2186 			!_SIG_VALID(ev.sigev_signo))
2187 			return (EINVAL);
2188 	}
2189 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2190 	if (error)
2191 		return (error);
2192 again:
2193 	FILEDESC_SLOCK(fdp);
2194 	if (fget_locked(fdp, uap->mqd) != fp) {
2195 		FILEDESC_SUNLOCK(fdp);
2196 		error = EBADF;
2197 		goto out;
2198 	}
2199 	mtx_lock(&mq->mq_mutex);
2200 	FILEDESC_SUNLOCK(fdp);
2201 	if (uap->sigev != NULL) {
2202 		if (mq->mq_notifier != NULL) {
2203 			error = EBUSY;
2204 		} else {
2205 			PROC_LOCK(p);
2206 			nt = notifier_search(p, uap->mqd);
2207 			if (nt == NULL) {
2208 				if (newnt == NULL) {
2209 					PROC_UNLOCK(p);
2210 					mtx_unlock(&mq->mq_mutex);
2211 					newnt = notifier_alloc();
2212 					goto again;
2213 				}
2214 			}
2215 
2216 			if (nt != NULL) {
2217 				sigqueue_take(&nt->nt_ksi);
2218 				if (newnt != NULL) {
2219 					notifier_free(newnt);
2220 					newnt = NULL;
2221 				}
2222 			} else {
2223 				nt = newnt;
2224 				newnt = NULL;
2225 				ksiginfo_init(&nt->nt_ksi);
2226 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2227 				nt->nt_ksi.ksi_code = SI_MESGQ;
2228 				nt->nt_proc = p;
2229 				nt->nt_ksi.ksi_mqd = uap->mqd;
2230 				notifier_insert(p, nt);
2231 			}
2232 			nt->nt_sigev = ev;
2233 			mq->mq_notifier = nt;
2234 			PROC_UNLOCK(p);
2235 			/*
2236 			 * if there is no receivers and message queue
2237 			 * is not empty, we should send notification
2238 			 * as soon as possible.
2239 			 */
2240 			if (mq->mq_receivers == 0 &&
2241 			    !TAILQ_EMPTY(&mq->mq_msgq))
2242 				mqueue_send_notification(mq);
2243 		}
2244 	} else {
2245 		notifier_remove(p, mq, uap->mqd);
2246 	}
2247 	mtx_unlock(&mq->mq_mutex);
2248 
2249 out:
2250 	fdrop(fp, td);
2251 	if (newnt != NULL)
2252 		notifier_free(newnt);
2253 	return (error);
2254 }
2255 
2256 static void
2257 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2258 {
2259 	struct filedesc *fdp;
2260 	struct mqueue *mq;
2261 
2262 	fdp = td->td_proc->p_fd;
2263 	FILEDESC_LOCK_ASSERT(fdp);
2264 
2265 	if (fp->f_ops == &mqueueops) {
2266 		mq = FPTOMQ(fp);
2267 		mtx_lock(&mq->mq_mutex);
2268 		notifier_remove(td->td_proc, mq, fd);
2269 
2270 		/* have to wakeup thread in same process */
2271 		if (mq->mq_flags & MQ_RSEL) {
2272 			mq->mq_flags &= ~MQ_RSEL;
2273 			selwakeup(&mq->mq_rsel);
2274 		}
2275 		if (mq->mq_flags & MQ_WSEL) {
2276 			mq->mq_flags &= ~MQ_WSEL;
2277 			selwakeup(&mq->mq_wsel);
2278 		}
2279 		mtx_unlock(&mq->mq_mutex);
2280 	}
2281 }
2282 
2283 static void
2284 mq_proc_exit(void *arg __unused, struct proc *p)
2285 {
2286 	struct filedesc *fdp;
2287 	struct file *fp;
2288 	struct mqueue *mq;
2289 	int i;
2290 
2291 	fdp = p->p_fd;
2292 	FILEDESC_SLOCK(fdp);
2293 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2294 		fp = fget_locked(fdp, i);
2295 		if (fp != NULL && fp->f_ops == &mqueueops) {
2296 			mq = FPTOMQ(fp);
2297 			mtx_lock(&mq->mq_mutex);
2298 			notifier_remove(p, FPTOMQ(fp), i);
2299 			mtx_unlock(&mq->mq_mutex);
2300 		}
2301 	}
2302 	FILEDESC_SUNLOCK(fdp);
2303 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2304 }
2305 
2306 static int
2307 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2308 	int flags, struct thread *td)
2309 {
2310 	return (EOPNOTSUPP);
2311 }
2312 
2313 static int
2314 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2315 	int flags, struct thread *td)
2316 {
2317 	return (EOPNOTSUPP);
2318 }
2319 
2320 static int
2321 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2322 	struct ucred *active_cred, struct thread *td)
2323 {
2324 	return (ENOTTY);
2325 }
2326 
2327 static int
2328 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2329 	struct thread *td)
2330 {
2331 	struct mqueue *mq = FPTOMQ(fp);
2332 	int revents = 0;
2333 
2334 	mtx_lock(&mq->mq_mutex);
2335 	if (events & (POLLIN | POLLRDNORM)) {
2336 		if (mq->mq_curmsgs) {
2337 			revents |= events & (POLLIN | POLLRDNORM);
2338 		} else {
2339 			mq->mq_flags |= MQ_RSEL;
2340 			selrecord(td, &mq->mq_rsel);
2341  		}
2342 	}
2343 	if (events & POLLOUT) {
2344 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2345 			revents |= POLLOUT;
2346 		else {
2347 			mq->mq_flags |= MQ_WSEL;
2348 			selrecord(td, &mq->mq_wsel);
2349 		}
2350 	}
2351 	mtx_unlock(&mq->mq_mutex);
2352 	return (revents);
2353 }
2354 
2355 static int
2356 mqf_close(struct file *fp, struct thread *td)
2357 {
2358 	struct mqfs_node *pn;
2359 
2360 	fp->f_ops = &badfileops;
2361 	pn = fp->f_data;
2362 	fp->f_data = NULL;
2363 	sx_xlock(&mqfs_data.mi_lock);
2364 	mqnode_release(pn);
2365 	sx_xunlock(&mqfs_data.mi_lock);
2366 	return (0);
2367 }
2368 
2369 static int
2370 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2371 	struct thread *td)
2372 {
2373 	struct mqfs_node *pn = fp->f_data;
2374 
2375 	bzero(st, sizeof *st);
2376 	st->st_atimespec = pn->mn_atime;
2377 	st->st_mtimespec = pn->mn_mtime;
2378 	st->st_ctimespec = pn->mn_ctime;
2379 	st->st_birthtimespec = pn->mn_birth;
2380 	st->st_uid = pn->mn_uid;
2381 	st->st_gid = pn->mn_gid;
2382 	st->st_mode = S_IFIFO | pn->mn_mode;
2383 	return (0);
2384 }
2385 
2386 static int
2387 mqf_kqfilter(struct file *fp, struct knote *kn)
2388 {
2389 	struct mqueue *mq = FPTOMQ(fp);
2390 	int error = 0;
2391 
2392 	if (kn->kn_filter == EVFILT_READ) {
2393 		kn->kn_fop = &mq_rfiltops;
2394 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2395 	} else if (kn->kn_filter == EVFILT_WRITE) {
2396 		kn->kn_fop = &mq_wfiltops;
2397 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2398 	} else
2399 		error = EINVAL;
2400 	return (error);
2401 }
2402 
2403 static void
2404 filt_mqdetach(struct knote *kn)
2405 {
2406 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2407 
2408 	if (kn->kn_filter == EVFILT_READ)
2409 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2410 	else if (kn->kn_filter == EVFILT_WRITE)
2411 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2412 	else
2413 		panic("filt_mqdetach");
2414 }
2415 
2416 static int
2417 filt_mqread(struct knote *kn, long hint)
2418 {
2419 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2420 
2421 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2422 	return (mq->mq_curmsgs != 0);
2423 }
2424 
2425 static int
2426 filt_mqwrite(struct knote *kn, long hint)
2427 {
2428 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2429 
2430 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2431 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2432 }
2433 
2434 static struct fileops mqueueops = {
2435 	.fo_read		= mqf_read,
2436 	.fo_write		= mqf_write,
2437 	.fo_ioctl		= mqf_ioctl,
2438 	.fo_poll		= mqf_poll,
2439 	.fo_kqfilter		= mqf_kqfilter,
2440 	.fo_stat		= mqf_stat,
2441 	.fo_close		= mqf_close
2442 };
2443 
2444 static struct vop_vector mqfs_vnodeops = {
2445 	.vop_default 		= &default_vnodeops,
2446 	.vop_access		= mqfs_access,
2447 	.vop_cachedlookup	= mqfs_lookup,
2448 	.vop_lookup		= vfs_cache_lookup,
2449 	.vop_reclaim		= mqfs_reclaim,
2450 	.vop_create		= mqfs_create,
2451 	.vop_remove		= mqfs_remove,
2452 	.vop_inactive		= mqfs_inactive,
2453 	.vop_open		= mqfs_open,
2454 	.vop_close		= mqfs_close,
2455 	.vop_getattr		= mqfs_getattr,
2456 	.vop_setattr		= mqfs_setattr,
2457 	.vop_read		= mqfs_read,
2458 	.vop_write		= VOP_EOPNOTSUPP,
2459 	.vop_readdir		= mqfs_readdir,
2460 	.vop_mkdir		= VOP_EOPNOTSUPP,
2461 	.vop_rmdir		= VOP_EOPNOTSUPP
2462 };
2463 
2464 static struct vfsops mqfs_vfsops = {
2465 	.vfs_init 		= mqfs_init,
2466 	.vfs_uninit		= mqfs_uninit,
2467 	.vfs_mount		= mqfs_mount,
2468 	.vfs_unmount		= mqfs_unmount,
2469 	.vfs_root		= mqfs_root,
2470 	.vfs_statfs		= mqfs_statfs,
2471 };
2472 
2473 SYSCALL_MODULE_HELPER(kmq_open);
2474 SYSCALL_MODULE_HELPER(kmq_setattr);
2475 SYSCALL_MODULE_HELPER(kmq_timedsend);
2476 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2477 SYSCALL_MODULE_HELPER(kmq_notify);
2478 SYSCALL_MODULE_HELPER(kmq_unlink);
2479 
2480 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2481 MODULE_VERSION(mqueuefs, 1);
2482