xref: /freebsd/sys/kern/uipc_mqueue.c (revision 94942af266ac119ede0ca836f9aa5a5ac0582938)
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 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, curthread);
721 	error = insmntque(*vpp, mp);
722 	if (error != 0) {
723 		*vpp = NULLVP;
724 		return (error);
725 	}
726 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
727 	(*vpp)->v_data = vd;
728 	vd->mv_vnode = *vpp;
729 	vd->mv_node = pn;
730 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
731 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
732 	mqnode_addref(pn);
733 	switch (pn->mn_type) {
734 	case mqfstype_root:
735 		(*vpp)->v_vflag = VV_ROOT;
736 		/* fall through */
737 	case mqfstype_dir:
738 	case mqfstype_this:
739 	case mqfstype_parent:
740 		(*vpp)->v_type = VDIR;
741 		break;
742 	case mqfstype_file:
743 		(*vpp)->v_type = VREG;
744 		break;
745 	case mqfstype_symlink:
746 		(*vpp)->v_type = VLNK;
747 		break;
748 	case mqfstype_none:
749 		KASSERT(0, ("mqfs_allocf called for null node\n"));
750 	default:
751 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
752 	}
753 	return (0);
754 }
755 
756 /*
757  * Search a directory entry
758  */
759 static struct mqfs_node *
760 mqfs_search(struct mqfs_node *pd, const char *name, int len)
761 {
762 	struct mqfs_node *pn;
763 
764 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
765 		if (strncmp(pn->mn_name, name, len) == 0)
766 			return (pn);
767 	}
768 	return (NULL);
769 }
770 
771 /*
772  * Look up a file or directory.
773  */
774 static int
775 mqfs_lookupx(struct vop_cachedlookup_args *ap)
776 {
777 	struct componentname *cnp;
778 	struct vnode *dvp, **vpp;
779 	struct mqfs_node *pd;
780 	struct mqfs_node *pn;
781 	int nameiop, flags, error, namelen;
782 	char *pname;
783 	struct thread *td;
784 
785 	cnp = ap->a_cnp;
786 	vpp = ap->a_vpp;
787 	dvp = ap->a_dvp;
788 	pname = cnp->cn_nameptr;
789 	namelen = cnp->cn_namelen;
790 	td = cnp->cn_thread;
791 	flags = cnp->cn_flags;
792 	nameiop = cnp->cn_nameiop;
793 	pd = VTON(dvp);
794 	pn = NULL;
795 	*vpp = NULLVP;
796 
797 	if (dvp->v_type != VDIR)
798 		return (ENOTDIR);
799 
800 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
801 	if (error)
802 		return (error);
803 
804 	/* shortcut: check if the name is too long */
805 	if (cnp->cn_namelen >= MQFS_NAMELEN)
806 		return (ENOENT);
807 
808 	/* self */
809 	if (namelen == 1 && pname[0] == '.') {
810 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
811 			return (EINVAL);
812 		pn = pd;
813 		*vpp = dvp;
814 		VREF(dvp);
815 		return (0);
816 	}
817 
818 	/* parent */
819 	if (cnp->cn_flags & ISDOTDOT) {
820 		if (dvp->v_vflag & VV_ROOT)
821 			return (EIO);
822 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
823 			return (EINVAL);
824 		VOP_UNLOCK(dvp, 0, cnp->cn_thread);
825 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
826 		pn = pd->mn_parent;
827 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
828 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
829 		return (error);
830 	}
831 
832 	/* named node */
833 	pn = mqfs_search(pd, pname, namelen);
834 
835 	/* found */
836 	if (pn != NULL) {
837 		/* DELETE */
838 		if (nameiop == DELETE && (flags & ISLASTCN)) {
839 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
840 			if (error)
841 				return (error);
842 			if (*vpp == dvp) {
843 				VREF(dvp);
844 				*vpp = dvp;
845 				return (0);
846 			}
847 		}
848 
849 		/* allocate vnode */
850 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
851 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
852 			cache_enter(dvp, *vpp, cnp);
853 		return (error);
854 	}
855 
856 	/* not found */
857 
858 	/* will create a new entry in the directory ? */
859 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
860 	    && (flags & ISLASTCN)) {
861 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
862 		if (error)
863 			return (error);
864 		cnp->cn_flags |= SAVENAME;
865 		return (EJUSTRETURN);
866 	}
867 	return (ENOENT);
868 }
869 
870 #if 0
871 struct vop_lookup_args {
872 	struct vop_generic_args a_gen;
873 	struct vnode *a_dvp;
874 	struct vnode **a_vpp;
875 	struct componentname *a_cnp;
876 };
877 #endif
878 
879 /*
880  * vnode lookup operation
881  */
882 static int
883 mqfs_lookup(struct vop_cachedlookup_args *ap)
884 {
885 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
886 	int rc;
887 
888 	sx_xlock(&mqfs->mi_lock);
889 	rc = mqfs_lookupx(ap);
890 	sx_xunlock(&mqfs->mi_lock);
891 	return (rc);
892 }
893 
894 #if 0
895 struct vop_create_args {
896 	struct vnode *a_dvp;
897 	struct vnode **a_vpp;
898 	struct componentname *a_cnp;
899 	struct vattr *a_vap;
900 };
901 #endif
902 
903 /*
904  * vnode creation operation
905  */
906 static int
907 mqfs_create(struct vop_create_args *ap)
908 {
909 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
910 	struct componentname *cnp = ap->a_cnp;
911 	struct mqfs_node *pd;
912 	struct mqfs_node *pn;
913 	struct mqueue *mq;
914 	int error;
915 
916 	pd = VTON(ap->a_dvp);
917 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
918 		return (ENOTDIR);
919 	mq = mqueue_alloc(NULL);
920 	if (mq == NULL)
921 		return (EAGAIN);
922 	sx_xlock(&mqfs->mi_lock);
923 #if 0
924 	/* named node */
925 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
926 	if (pn != NULL) {
927 		mqueue_free(mq);
928 		sx_xunlock(&mqfs->mi_lock);
929 		return (EEXIST);
930 	}
931 #else
932 	if ((cnp->cn_flags & HASBUF) == 0)
933 		panic("%s: no name", __func__);
934 #endif
935 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
936 		cnp->cn_cred, ap->a_vap->va_mode);
937 	if (pn == NULL)
938 		error = ENOSPC;
939 	else {
940 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
941 		if (error)
942 			mqfs_destroy(pn);
943 		else
944 			pn->mn_data = mq;
945 	}
946 	sx_xunlock(&mqfs->mi_lock);
947 	if (error)
948 		mqueue_free(mq);
949 	return (error);
950 }
951 
952 /*
953  * Remove an entry
954  */
955 static
956 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
957 {
958 	struct mqfs_node *parent;
959 	struct mqfs_vdata *vd;
960 	int error = 0;
961 
962 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
963 
964 	if (ucred->cr_uid != pn->mn_uid &&
965 	    (error = priv_check_cred(ucred, PRIV_MQ_ADMIN,
966 	    SUSER_ALLOWJAIL)) != 0)
967 		error = EACCES;
968 	else if (!pn->mn_deleted) {
969 		parent = pn->mn_parent;
970 		pn->mn_parent = NULL;
971 		pn->mn_deleted = 1;
972 		LIST_REMOVE(pn, mn_sibling);
973 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
974 			cache_purge(vd->mv_vnode);
975 			vhold(vd->mv_vnode);
976 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
977 		}
978 		mqnode_release(pn);
979 		mqnode_release(parent);
980 	} else
981 		error = ENOENT;
982 	return (error);
983 }
984 
985 #if 0
986 struct vop_remove_args {
987 	struct vnode *a_dvp;
988 	struct vnode *a_vp;
989 	struct componentname *a_cnp;
990 };
991 #endif
992 
993 /*
994  * vnode removal operation
995  */
996 static int
997 mqfs_remove(struct vop_remove_args *ap)
998 {
999 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1000 	struct mqfs_node *pn;
1001 	int error;
1002 
1003 	if (ap->a_vp->v_type == VDIR)
1004                 return (EPERM);
1005 	pn = VTON(ap->a_vp);
1006 	sx_xlock(&mqfs->mi_lock);
1007 	error = do_unlink(pn, ap->a_cnp->cn_cred);
1008 	sx_xunlock(&mqfs->mi_lock);
1009 	return (error);
1010 }
1011 
1012 #if 0
1013 struct vop_inactive_args {
1014 	struct vnode *a_vp;
1015 	struct thread *a_td;
1016 };
1017 #endif
1018 
1019 static int
1020 mqfs_inactive(struct vop_inactive_args *ap)
1021 {
1022 	struct mqfs_node *pn = VTON(ap->a_vp);
1023 
1024 	if (pn->mn_deleted)
1025 		vrecycle(ap->a_vp, ap->a_td);
1026 	return (0);
1027 }
1028 
1029 #if 0
1030 struct vop_reclaim_args {
1031 	struct vop_generic_args a_gen;
1032 	struct vnode *a_vp;
1033 	struct thread *a_td;
1034 };
1035 #endif
1036 
1037 static int
1038 mqfs_reclaim(struct vop_reclaim_args *ap)
1039 {
1040 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1041 	struct vnode *vp = ap->a_vp;
1042 	struct mqfs_node *pn;
1043 	struct mqfs_vdata *vd;
1044 
1045 	vd = vp->v_data;
1046 	pn = vd->mv_node;
1047 	sx_xlock(&mqfs->mi_lock);
1048 	vp->v_data = NULL;
1049 	LIST_REMOVE(vd, mv_link);
1050 	uma_zfree(mvdata_zone, vd);
1051 	mqnode_release(pn);
1052 	sx_xunlock(&mqfs->mi_lock);
1053 	return (0);
1054 }
1055 
1056 #if 0
1057 struct vop_open_args {
1058 	struct vop_generic_args a_gen;
1059 	struct vnode *a_vp;
1060 	int a_mode;
1061 	struct ucred *a_cred;
1062 	struct thread *a_td;
1063 	int a_fdidx;
1064 };
1065 #endif
1066 
1067 static int
1068 mqfs_open(struct vop_open_args *ap)
1069 {
1070 	return (0);
1071 }
1072 
1073 #if 0
1074 struct vop_close_args {
1075 	struct vop_generic_args a_gen;
1076 	struct vnode *a_vp;
1077 	int a_fflag;
1078 	struct ucred *a_cred;
1079 	struct thread *a_td;
1080 };
1081 #endif
1082 
1083 static int
1084 mqfs_close(struct vop_close_args *ap)
1085 {
1086 	return (0);
1087 }
1088 
1089 #if 0
1090 struct vop_access_args {
1091 	struct vop_generic_args a_gen;
1092 	struct vnode *a_vp;
1093 	int a_mode;
1094 	struct ucred *a_cred;
1095 	struct thread *a_td;
1096 };
1097 #endif
1098 
1099 /*
1100  * Verify permissions
1101  */
1102 static int
1103 mqfs_access(struct vop_access_args *ap)
1104 {
1105 	struct vnode *vp = ap->a_vp;
1106 	struct vattr vattr;
1107 	int error;
1108 
1109 	error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1110 	if (error)
1111 		return (error);
1112 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1113 	    vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1114 	return (error);
1115 }
1116 
1117 #if 0
1118 struct vop_getattr_args {
1119 	struct vop_generic_args a_gen;
1120 	struct vnode *a_vp;
1121 	struct vattr *a_vap;
1122 	struct ucred *a_cred;
1123 	struct thread *a_td;
1124 };
1125 #endif
1126 
1127 /*
1128  * Get file attributes
1129  */
1130 static int
1131 mqfs_getattr(struct vop_getattr_args *ap)
1132 {
1133 	struct vnode *vp = ap->a_vp;
1134 	struct mqfs_node *pn = VTON(vp);
1135 	struct vattr *vap = ap->a_vap;
1136 	int error = 0;
1137 
1138 	VATTR_NULL(vap);
1139 	vap->va_type = vp->v_type;
1140 	vap->va_mode = pn->mn_mode;
1141 	vap->va_nlink = 1;
1142 	vap->va_uid = pn->mn_uid;
1143 	vap->va_gid = pn->mn_gid;
1144 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1145 	vap->va_fileid = pn->mn_fileno;
1146 	vap->va_size = 0;
1147 	vap->va_blocksize = PAGE_SIZE;
1148 	vap->va_bytes = vap->va_size = 0;
1149 	vap->va_atime = pn->mn_atime;
1150 	vap->va_mtime = pn->mn_mtime;
1151 	vap->va_ctime = pn->mn_ctime;
1152 	vap->va_birthtime = pn->mn_birth;
1153 	vap->va_gen = 0;
1154 	vap->va_flags = 0;
1155 	vap->va_rdev = 0;
1156 	vap->va_bytes = 0;
1157 	vap->va_filerev = 0;
1158 	vap->va_vaflags = 0;
1159 	return (error);
1160 }
1161 
1162 #if 0
1163 struct vop_setattr_args {
1164 	struct vop_generic_args a_gen;
1165 	struct vnode *a_vp;
1166 	struct vattr *a_vap;
1167 	struct ucred *a_cred;
1168 	struct thread *a_td;
1169 };
1170 #endif
1171 /*
1172  * Set attributes
1173  */
1174 static int
1175 mqfs_setattr(struct vop_setattr_args *ap)
1176 {
1177 	struct mqfs_node *pn;
1178 	struct vattr *vap;
1179 	struct vnode *vp;
1180 	int c, error;
1181 	uid_t uid;
1182 	gid_t gid;
1183 
1184 	vap = ap->a_vap;
1185 	vp = ap->a_vp;
1186 	if ((vap->va_type != VNON) ||
1187 	    (vap->va_nlink != VNOVAL) ||
1188 	    (vap->va_fsid != VNOVAL) ||
1189 	    (vap->va_fileid != VNOVAL) ||
1190 	    (vap->va_blocksize != VNOVAL) ||
1191 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1192 	    (vap->va_rdev != VNOVAL) ||
1193 	    ((int)vap->va_bytes != VNOVAL) ||
1194 	    (vap->va_gen != VNOVAL)) {
1195 		return (EINVAL);
1196 	}
1197 
1198 	pn = VTON(vp);
1199 
1200 	error = c = 0;
1201 	if (vap->va_uid == (uid_t)VNOVAL)
1202 		uid = pn->mn_uid;
1203 	else
1204 		uid = vap->va_uid;
1205 	if (vap->va_gid == (gid_t)VNOVAL)
1206 		gid = pn->mn_gid;
1207 	else
1208 		gid = vap->va_gid;
1209 
1210 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1211 		/*
1212 		 * To modify the ownership of a file, must possess VADMIN
1213 		 * for that file.
1214 		 */
1215 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1216 			return (error);
1217 
1218 		/*
1219 		 * XXXRW: Why is there a privilege check here: shouldn't the
1220 		 * check in VOP_ACCESS() be enough?  Also, are the group bits
1221 		 * below definitely right?
1222 		 */
1223 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1224 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1225 		    (error = priv_check_cred(ap->a_td->td_ucred,
1226 		    PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)) != 0)
1227 			return (error);
1228 		pn->mn_uid = uid;
1229 		pn->mn_gid = gid;
1230 		c = 1;
1231 	}
1232 
1233 	if (vap->va_mode != (mode_t)VNOVAL) {
1234 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1235 		    (error = priv_check_cred(ap->a_td->td_ucred,
1236 		    PRIV_MQ_ADMIN, SUSER_ALLOWJAIL)))
1237 			return (error);
1238 		pn->mn_mode = vap->va_mode;
1239 		c = 1;
1240 	}
1241 
1242 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1243 		/* See the comment in ufs_vnops::ufs_setattr(). */
1244 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1245 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1246 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1247 			return (error);
1248 		if (vap->va_atime.tv_sec != VNOVAL) {
1249 			pn->mn_atime = vap->va_atime;
1250 		}
1251 		if (vap->va_mtime.tv_sec != VNOVAL) {
1252 			pn->mn_mtime = vap->va_mtime;
1253 		}
1254 		c = 1;
1255 	}
1256 	if (c) {
1257 		vfs_timestamp(&pn->mn_ctime);
1258 	}
1259 	return (0);
1260 }
1261 
1262 #if 0
1263 struct vop_read_args {
1264 	struct vop_generic_args a_gen;
1265 	struct vnode *a_vp;
1266 	struct uio *a_uio;
1267 	int a_ioflag;
1268 	struct ucred *a_cred;
1269 };
1270 #endif
1271 
1272 /*
1273  * Read from a file
1274  */
1275 static int
1276 mqfs_read(struct vop_read_args *ap)
1277 {
1278 	char buf[80];
1279 	struct vnode *vp = ap->a_vp;
1280 	struct uio *uio = ap->a_uio;
1281 	struct mqfs_node *pn;
1282 	struct mqueue *mq;
1283 	int len, error;
1284 
1285 	if (vp->v_type != VREG)
1286 		return (EINVAL);
1287 
1288 	pn = VTON(vp);
1289 	mq = VTOMQ(vp);
1290 	snprintf(buf, sizeof(buf),
1291 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1292 		mq->mq_totalbytes,
1293 		mq->mq_maxmsg,
1294 		mq->mq_curmsgs,
1295 		mq->mq_msgsize);
1296 	buf[sizeof(buf)-1] = '\0';
1297 	len = strlen(buf);
1298 	error = uiomove_frombuf(buf, len, uio);
1299 	return (error);
1300 }
1301 
1302 #if 0
1303 struct vop_readdir_args {
1304 	struct vop_generic_args a_gen;
1305 	struct vnode *a_vp;
1306 	struct uio *a_uio;
1307 	struct ucred *a_cred;
1308 	int *a_eofflag;
1309 	int *a_ncookies;
1310 	u_long **a_cookies;
1311 };
1312 #endif
1313 
1314 /*
1315  * Return directory entries.
1316  */
1317 static int
1318 mqfs_readdir(struct vop_readdir_args *ap)
1319 {
1320 	struct vnode *vp;
1321 	struct mqfs_info *mi;
1322 	struct mqfs_node *pd;
1323 	struct mqfs_node *pn;
1324 	struct dirent entry;
1325 	struct uio *uio;
1326 	int *tmp_ncookies = NULL;
1327 	off_t offset;
1328 	int error, i;
1329 
1330 	vp = ap->a_vp;
1331 	mi = VFSTOMQFS(vp->v_mount);
1332 	pd = VTON(vp);
1333 	uio = ap->a_uio;
1334 
1335 	if (vp->v_type != VDIR)
1336 		return (ENOTDIR);
1337 
1338 	if (uio->uio_offset < 0)
1339 		return (EINVAL);
1340 
1341 	if (ap->a_ncookies != NULL) {
1342 		tmp_ncookies = ap->a_ncookies;
1343 		*ap->a_ncookies = 0;
1344 		ap->a_ncookies = NULL;
1345         }
1346 
1347 	error = 0;
1348 	offset = 0;
1349 
1350 	sx_xlock(&mi->mi_lock);
1351 
1352 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1353 		entry.d_reclen = sizeof(entry);
1354 		if (!pn->mn_fileno)
1355 			mqfs_fileno_alloc(mi, pn);
1356 		entry.d_fileno = pn->mn_fileno;
1357 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1358 			entry.d_name[i] = pn->mn_name[i];
1359 		entry.d_name[i] = 0;
1360 		entry.d_namlen = i;
1361 		switch (pn->mn_type) {
1362 		case mqfstype_root:
1363 		case mqfstype_dir:
1364 		case mqfstype_this:
1365 		case mqfstype_parent:
1366 			entry.d_type = DT_DIR;
1367 			break;
1368 		case mqfstype_file:
1369 			entry.d_type = DT_REG;
1370 			break;
1371 		case mqfstype_symlink:
1372 			entry.d_type = DT_LNK;
1373 			break;
1374 		default:
1375 			panic("%s has unexpected node type: %d", pn->mn_name,
1376 				pn->mn_type);
1377 		}
1378 		if (entry.d_reclen > uio->uio_resid)
1379                         break;
1380 		if (offset >= uio->uio_offset) {
1381 			error = vfs_read_dirent(ap, &entry, offset);
1382                         if (error)
1383                                 break;
1384                 }
1385                 offset += entry.d_reclen;
1386 	}
1387 	sx_xunlock(&mi->mi_lock);
1388 
1389 	uio->uio_offset = offset;
1390 
1391 	if (tmp_ncookies != NULL)
1392 		ap->a_ncookies = tmp_ncookies;
1393 
1394 	return (error);
1395 }
1396 
1397 #ifdef notyet
1398 
1399 #if 0
1400 struct vop_mkdir_args {
1401 	struct vnode *a_dvp;
1402 	struvt vnode **a_vpp;
1403 	struvt componentname *a_cnp;
1404 	struct vattr *a_vap;
1405 };
1406 #endif
1407 
1408 /*
1409  * Create a directory.
1410  */
1411 static int
1412 mqfs_mkdir(struct vop_mkdir_args *ap)
1413 {
1414 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1415 	struct componentname *cnp = ap->a_cnp;
1416 	struct mqfs_node *pd = VTON(ap->a_dvp);
1417 	struct mqfs_node *pn;
1418 	int error;
1419 
1420 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1421 		return (ENOTDIR);
1422 	sx_xlock(&mqfs->mi_lock);
1423 #if 0
1424 	/* named node */
1425 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1426 	if (pn != NULL) {
1427 		sx_xunlock(&mqfs->mi_lock);
1428 		return (EEXIST);
1429 	}
1430 #else
1431 	if ((cnp->cn_flags & HASBUF) == 0)
1432 		panic("%s: no name", __func__);
1433 #endif
1434 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1435 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1436 	if (pn == NULL)
1437 		error = ENOSPC;
1438 	else
1439 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1440 	sx_xunlock(&mqfs->mi_lock);
1441 	return (error);
1442 }
1443 
1444 #if 0
1445 struct vop_rmdir_args {
1446 	struct vnode *a_dvp;
1447 	struct vnode *a_vp;
1448 	struct componentname *a_cnp;
1449 };
1450 #endif
1451 
1452 /*
1453  * Remove a directory.
1454  */
1455 static int
1456 mqfs_rmdir(struct vop_rmdir_args *ap)
1457 {
1458 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1459 	struct mqfs_node *pn = VTON(ap->a_vp);
1460 	struct mqfs_node *pt;
1461 
1462 	if (pn->mn_type != mqfstype_dir)
1463 		return (ENOTDIR);
1464 
1465 	sx_xlock(&mqfs->mi_lock);
1466 	if (pn->mn_deleted) {
1467 		sx_xunlock(&mqfs->mi_lock);
1468 		return (ENOENT);
1469 	}
1470 
1471 	pt = LIST_FIRST(&pn->mn_children);
1472 	pt = LIST_NEXT(pt, mn_sibling);
1473 	pt = LIST_NEXT(pt, mn_sibling);
1474 	if (pt != NULL) {
1475 		sx_xunlock(&mqfs->mi_lock);
1476 		return (ENOTEMPTY);
1477 	}
1478 	pt = pn->mn_parent;
1479 	pn->mn_parent = NULL;
1480 	pn->mn_deleted = 1;
1481 	LIST_REMOVE(pn, mn_sibling);
1482 	mqnode_release(pn);
1483 	mqnode_release(pt);
1484 	sx_xunlock(&mqfs->mi_lock);
1485 	cache_purge(ap->a_vp);
1486 	return (0);
1487 }
1488 
1489 #endif /* notyet */
1490 
1491 /*
1492  * Allocate a message queue
1493  */
1494 static struct mqueue *
1495 mqueue_alloc(const struct mq_attr *attr)
1496 {
1497 	struct mqueue *mq;
1498 
1499 	if (curmq >= maxmq)
1500 		return (NULL);
1501 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1502 	TAILQ_INIT(&mq->mq_msgq);
1503 	if (attr != NULL) {
1504 		mq->mq_maxmsg = attr->mq_maxmsg;
1505 		mq->mq_msgsize = attr->mq_msgsize;
1506 	} else {
1507 		mq->mq_maxmsg = default_maxmsg;
1508 		mq->mq_msgsize = default_msgsize;
1509 	}
1510 	mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1511 	knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1512 	knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1513 	atomic_add_int(&curmq, 1);
1514 	return (mq);
1515 }
1516 
1517 /*
1518  * Destroy a message queue
1519  */
1520 static void
1521 mqueue_free(struct mqueue *mq)
1522 {
1523 	struct mqueue_msg *msg;
1524 
1525 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1526 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1527 		FREE(msg, M_MQUEUEDATA);
1528 	}
1529 
1530 	mtx_destroy(&mq->mq_mutex);
1531 	knlist_destroy(&mq->mq_rsel.si_note);
1532 	knlist_destroy(&mq->mq_wsel.si_note);
1533 	uma_zfree(mqueue_zone, mq);
1534 	atomic_add_int(&curmq, -1);
1535 }
1536 
1537 /*
1538  * Load a message from user space
1539  */
1540 static struct mqueue_msg *
1541 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1542 {
1543 	struct mqueue_msg *msg;
1544 	size_t len;
1545 	int error;
1546 
1547 	len = sizeof(struct mqueue_msg) + msg_size;
1548 	MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1549 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1550 	    msg_size);
1551 	if (error) {
1552 		FREE(msg, M_MQUEUEDATA);
1553 		msg = NULL;
1554 	} else {
1555 		msg->msg_size = msg_size;
1556 		msg->msg_prio = msg_prio;
1557 	}
1558 	return (msg);
1559 }
1560 
1561 /*
1562  * Save a message to user space
1563  */
1564 static int
1565 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1566 {
1567 	int error;
1568 
1569 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1570 		msg->msg_size);
1571 	if (error == 0 && msg_prio != NULL)
1572 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1573 	return (error);
1574 }
1575 
1576 /*
1577  * Free a message's memory
1578  */
1579 static __inline void
1580 mqueue_freemsg(struct mqueue_msg *msg)
1581 {
1582 	FREE(msg, M_MQUEUEDATA);
1583 }
1584 
1585 /*
1586  * Send a message. if waitok is false, thread will not be
1587  * blocked if there is no data in queue, otherwise, absolute
1588  * time will be checked.
1589  */
1590 int
1591 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1592 	size_t msg_len, unsigned msg_prio, int waitok,
1593 	const struct timespec *abs_timeout)
1594 {
1595 	struct mqueue_msg *msg;
1596 	struct timespec ets, ts, ts2;
1597 	struct timeval tv;
1598 	int error;
1599 
1600 	if (msg_prio >= MQ_PRIO_MAX)
1601 		return (EINVAL);
1602 	if (msg_len > mq->mq_msgsize)
1603 		return (EMSGSIZE);
1604 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1605 	if (msg == NULL)
1606 		return (EFAULT);
1607 
1608 	/* O_NONBLOCK case */
1609 	if (!waitok) {
1610 		error = _mqueue_send(mq, msg, -1);
1611 		if (error)
1612 			goto bad;
1613 		return (0);
1614 	}
1615 
1616 	/* we allow a null timeout (wait forever) */
1617 	if (abs_timeout == NULL) {
1618 		error = _mqueue_send(mq, msg, 0);
1619 		if (error)
1620 			goto bad;
1621 		return (0);
1622 	}
1623 
1624 	/* send it before checking time */
1625 	error = _mqueue_send(mq, msg, -1);
1626 	if (error == 0)
1627 		return (0);
1628 
1629 	if (error != EAGAIN)
1630 		goto bad;
1631 
1632 	error = copyin(abs_timeout, &ets, sizeof(ets));
1633 	if (error != 0)
1634 		goto bad;
1635 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1636 		error = EINVAL;
1637 		goto bad;
1638 	}
1639 	for (;;) {
1640 		ts2 = ets;
1641 		getnanotime(&ts);
1642 		timespecsub(&ts2, &ts);
1643 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1644 			error = ETIMEDOUT;
1645 			break;
1646 		}
1647 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1648 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1649 		if (error != ETIMEDOUT)
1650 			break;
1651 	}
1652 	if (error == 0)
1653 		return (0);
1654 bad:
1655 	mqueue_freemsg(msg);
1656 	return (error);
1657 }
1658 
1659 /*
1660  * Common routine to send a message
1661  */
1662 static int
1663 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1664 {
1665 	struct mqueue_msg *msg2;
1666 	int error = 0;
1667 
1668 	mtx_lock(&mq->mq_mutex);
1669 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1670 		if (timo < 0) {
1671 			mtx_unlock(&mq->mq_mutex);
1672 			return (EAGAIN);
1673 		}
1674 		mq->mq_senders++;
1675 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1676 			    PCATCH, "mqsend", timo);
1677 		mq->mq_senders--;
1678 		if (error == EAGAIN)
1679 			error = ETIMEDOUT;
1680 	}
1681 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1682 		mtx_unlock(&mq->mq_mutex);
1683 		return (error);
1684 	}
1685 	error = 0;
1686 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1687 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1688 	} else {
1689 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1690 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1691 		} else {
1692 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1693 				if (msg2->msg_prio < msg->msg_prio)
1694 					break;
1695 			}
1696 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1697 		}
1698 	}
1699 	mq->mq_curmsgs++;
1700 	mq->mq_totalbytes += msg->msg_size;
1701 	if (mq->mq_receivers)
1702 		wakeup_one(&mq->mq_receivers);
1703 	else if (mq->mq_notifier != NULL)
1704 		mqueue_send_notification(mq);
1705 	if (mq->mq_flags & MQ_RSEL) {
1706 		mq->mq_flags &= ~MQ_RSEL;
1707 		selwakeup(&mq->mq_rsel);
1708 	}
1709 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1710 	mtx_unlock(&mq->mq_mutex);
1711 	return (0);
1712 }
1713 
1714 /*
1715  * Send realtime a signal to process which registered itself
1716  * successfully by mq_notify.
1717  */
1718 static void
1719 mqueue_send_notification(struct mqueue *mq)
1720 {
1721 	struct mqueue_notifier *nt;
1722 	struct proc *p;
1723 
1724 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1725 	nt = mq->mq_notifier;
1726 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1727 		p = nt->nt_proc;
1728 		PROC_LOCK(p);
1729 		if (!KSI_ONQ(&nt->nt_ksi))
1730 			psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1731 		PROC_UNLOCK(p);
1732 	}
1733 	mq->mq_notifier = NULL;
1734 }
1735 
1736 /*
1737  * Get a message. if waitok is false, thread will not be
1738  * blocked if there is no data in queue, otherwise, absolute
1739  * time will be checked.
1740  */
1741 int
1742 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1743 	size_t msg_len, unsigned *msg_prio, int waitok,
1744 	const struct timespec *abs_timeout)
1745 {
1746 	struct mqueue_msg *msg;
1747 	struct timespec ets, ts, ts2;
1748 	struct timeval tv;
1749 	int error;
1750 
1751 	if (msg_len < mq->mq_msgsize)
1752 		return (EMSGSIZE);
1753 
1754 	/* O_NONBLOCK case */
1755 	if (!waitok) {
1756 		error = _mqueue_recv(mq, &msg, -1);
1757 		if (error)
1758 			return (error);
1759 		goto received;
1760 	}
1761 
1762 	/* we allow a null timeout (wait forever). */
1763 	if (abs_timeout == NULL) {
1764 		error = _mqueue_recv(mq, &msg, 0);
1765 		if (error)
1766 			return (error);
1767 		goto received;
1768 	}
1769 
1770 	/* try to get a message before checking time */
1771 	error = _mqueue_recv(mq, &msg, -1);
1772 	if (error == 0)
1773 		goto received;
1774 
1775 	if (error != EAGAIN)
1776 		return (error);
1777 
1778 	error = copyin(abs_timeout, &ets, sizeof(ets));
1779 	if (error != 0)
1780 		return (error);
1781 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1782 		error = EINVAL;
1783 		return (error);
1784 	}
1785 
1786 	for (;;) {
1787 		ts2 = ets;
1788 		getnanotime(&ts);
1789 		timespecsub(&ts2, &ts);
1790 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1791 			error = ETIMEDOUT;
1792 			return (error);
1793 		}
1794 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1795 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1796 		if (error == 0)
1797 			break;
1798 		if (error != ETIMEDOUT)
1799 			return (error);
1800 	}
1801 
1802 received:
1803 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1804 	if (error == 0) {
1805 		curthread->td_retval[0] = msg->msg_size;
1806 		curthread->td_retval[1] = 0;
1807 	}
1808 	mqueue_freemsg(msg);
1809 	return (error);
1810 }
1811 
1812 /*
1813  * Common routine to receive a message
1814  */
1815 static int
1816 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1817 {
1818 	int error = 0;
1819 
1820 	mtx_lock(&mq->mq_mutex);
1821 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1822 		if (timo < 0) {
1823 			mtx_unlock(&mq->mq_mutex);
1824 			return (EAGAIN);
1825 		}
1826 		mq->mq_receivers++;
1827 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1828 			    PCATCH, "mqrecv", timo);
1829 		mq->mq_receivers--;
1830 		if (error == EAGAIN)
1831 			error = ETIMEDOUT;
1832 	}
1833 	if (*msg != NULL) {
1834 		error = 0;
1835 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1836 		mq->mq_curmsgs--;
1837 		mq->mq_totalbytes -= (*msg)->msg_size;
1838 		if (mq->mq_senders)
1839 			wakeup_one(&mq->mq_senders);
1840 		if (mq->mq_flags & MQ_WSEL) {
1841 			mq->mq_flags &= ~MQ_WSEL;
1842 			selwakeup(&mq->mq_wsel);
1843 		}
1844 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1845 	}
1846 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1847 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1848 		mqueue_send_notification(mq);
1849 	}
1850 	mtx_unlock(&mq->mq_mutex);
1851 	return (error);
1852 }
1853 
1854 static __inline struct mqueue_notifier *
1855 notifier_alloc(void)
1856 {
1857 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1858 }
1859 
1860 static __inline void
1861 notifier_free(struct mqueue_notifier *p)
1862 {
1863 	uma_zfree(mqnoti_zone, p);
1864 }
1865 
1866 static struct mqueue_notifier *
1867 notifier_search(struct proc *p, int fd)
1868 {
1869 	struct mqueue_notifier *nt;
1870 
1871 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1872 		if (nt->nt_ksi.ksi_mqd == fd)
1873 			break;
1874 	}
1875 	return (nt);
1876 }
1877 
1878 static __inline void
1879 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1880 {
1881 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1882 }
1883 
1884 static __inline void
1885 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1886 {
1887 	LIST_REMOVE(nt, nt_link);
1888 	notifier_free(nt);
1889 }
1890 
1891 static void
1892 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1893 {
1894 	struct mqueue_notifier *nt;
1895 
1896 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1897 	PROC_LOCK(p);
1898 	nt = notifier_search(p, fd);
1899 	if (nt != NULL) {
1900 		if (mq->mq_notifier == nt)
1901 			mq->mq_notifier = NULL;
1902 		sigqueue_take(&nt->nt_ksi);
1903 		notifier_delete(p, nt);
1904 	}
1905 	PROC_UNLOCK(p);
1906 }
1907 
1908 /*
1909  * Syscall to open a message queue.
1910  */
1911 int
1912 kmq_open(struct thread *td, struct kmq_open_args *uap)
1913 {
1914 	char path[MQFS_NAMELEN + 1];
1915 	struct mq_attr attr, *pattr;
1916 	struct mqfs_node *pn;
1917 	struct filedesc *fdp;
1918 	struct file *fp;
1919 	struct mqueue *mq;
1920 	int fd, error, len, flags, cmode;
1921 
1922 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1923 		return (EINVAL);
1924 
1925 	fdp = td->td_proc->p_fd;
1926 	flags = FFLAGS(uap->flags);
1927 	cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1928 	mq = NULL;
1929 	if ((flags & O_CREAT) && (uap->attr != NULL)) {
1930 		error = copyin(uap->attr, &attr, sizeof(attr));
1931 		if (error)
1932 			return (error);
1933 		if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1934 			return (EINVAL);
1935 		if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1936 			return (EINVAL);
1937 		pattr = &attr;
1938 	} else
1939 		pattr = NULL;
1940 
1941 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1942         if (error)
1943 		return (error);
1944 
1945 	/*
1946 	 * The first character of name must be a slash  (/) character
1947 	 * and the remaining characters of name cannot include any slash
1948 	 * characters.
1949 	 */
1950 	len = strlen(path);
1951 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1952 		return (EINVAL);
1953 
1954 	error = falloc(td, &fp, &fd);
1955 	if (error)
1956 		return (error);
1957 
1958 	sx_xlock(&mqfs_data.mi_lock);
1959 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1960 	if (pn == NULL) {
1961 		if (!(flags & O_CREAT)) {
1962 			error = ENOENT;
1963 		} else {
1964 			mq = mqueue_alloc(pattr);
1965 			if (mq == NULL) {
1966 				error = ENFILE;
1967 			} else {
1968 				pn = mqfs_create_file(mqfs_data.mi_root,
1969 				         path + 1, len - 1, td->td_ucred,
1970 					 cmode);
1971 				if (pn == NULL) {
1972 					error = ENOSPC;
1973 					mqueue_free(mq);
1974 				}
1975 			}
1976 		}
1977 
1978 		if (error == 0) {
1979 			pn->mn_data = mq;
1980 		}
1981 	} else {
1982 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1983 			error = EEXIST;
1984 		} else {
1985 			int acc_mode = 0;
1986 
1987 			if (flags & FREAD)
1988 				acc_mode |= VREAD;
1989 			if (flags & FWRITE)
1990 				acc_mode |= VWRITE;
1991 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1992 				    pn->mn_gid, acc_mode, td->td_ucred, NULL);
1993 		}
1994 	}
1995 
1996 	if (error) {
1997 		sx_xunlock(&mqfs_data.mi_lock);
1998 		fdclose(fdp, fp, fd, td);
1999 		fdrop(fp, td);
2000 		return (error);
2001 	}
2002 
2003 	mqnode_addref(pn);
2004 	sx_xunlock(&mqfs_data.mi_lock);
2005 
2006 	FILE_LOCK(fp);
2007 	fp->f_flag = (flags & (FREAD | FWRITE | O_NONBLOCK));
2008 	fp->f_type = DTYPE_MQUEUE;
2009 	fp->f_data = pn;
2010 	fp->f_ops = &mqueueops;
2011 	FILE_UNLOCK(fp);
2012 
2013 	FILEDESC_XLOCK(fdp);
2014 	if (fdp->fd_ofiles[fd] == fp)
2015 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2016 	FILEDESC_XUNLOCK(fdp);
2017 	td->td_retval[0] = fd;
2018 	fdrop(fp, td);
2019 	return (0);
2020 }
2021 
2022 /*
2023  * Syscall to unlink a message queue.
2024  */
2025 int
2026 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2027 {
2028 	char path[MQFS_NAMELEN+1];
2029 	struct mqfs_node *pn;
2030 	int error, len;
2031 
2032 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2033         if (error)
2034 		return (error);
2035 
2036 	len = strlen(path);
2037 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
2038 		return (EINVAL);
2039 
2040 	sx_xlock(&mqfs_data.mi_lock);
2041 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2042 	if (pn != NULL)
2043 		error = do_unlink(pn, td->td_ucred);
2044 	else
2045 		error = ENOENT;
2046 	sx_xunlock(&mqfs_data.mi_lock);
2047 	return (error);
2048 }
2049 
2050 typedef int (*_fgetf)(struct thread *, int, struct file **);
2051 
2052 /*
2053  * Get message queue by giving file slot
2054  */
2055 static int
2056 _getmq(struct thread *td, int fd, _fgetf func,
2057        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2058 {
2059 	struct mqfs_node *pn;
2060 	int error;
2061 
2062 	error = func(td, fd, fpp);
2063 	if (error)
2064 		return (error);
2065 	if (&mqueueops != (*fpp)->f_ops) {
2066 		fdrop(*fpp, td);
2067 		return (EBADF);
2068 	}
2069 	pn = (*fpp)->f_data;
2070 	if (ppn)
2071 		*ppn = pn;
2072 	if (pmq)
2073 		*pmq = pn->mn_data;
2074 	return (0);
2075 }
2076 
2077 static __inline int
2078 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2079 	struct mqueue **pmq)
2080 {
2081 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2082 }
2083 
2084 static __inline int
2085 getmq_read(struct thread *td, int fd, struct file **fpp,
2086 	 struct mqfs_node **ppn, struct mqueue **pmq)
2087 {
2088 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2089 }
2090 
2091 static __inline int
2092 getmq_write(struct thread *td, int fd, struct file **fpp,
2093 	struct mqfs_node **ppn, struct mqueue **pmq)
2094 {
2095 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2096 }
2097 
2098 int
2099 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2100 {
2101 	struct mqueue *mq;
2102 	struct file *fp;
2103 	struct mq_attr attr, oattr;
2104 	int error;
2105 
2106 	if (uap->attr) {
2107 		error = copyin(uap->attr, &attr, sizeof(attr));
2108 		if (error)
2109 			return (error);
2110 		if (attr.mq_flags & ~O_NONBLOCK)
2111 			return (EINVAL);
2112 	}
2113 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2114 	if (error)
2115 		return (error);
2116 	oattr.mq_maxmsg  = mq->mq_maxmsg;
2117 	oattr.mq_msgsize = mq->mq_msgsize;
2118 	oattr.mq_curmsgs = mq->mq_curmsgs;
2119 	FILE_LOCK(fp);
2120 	oattr.mq_flags = (O_NONBLOCK & fp->f_flag);
2121 	if (uap->attr) {
2122 		fp->f_flag &= ~O_NONBLOCK;
2123 		fp->f_flag |= (attr.mq_flags & O_NONBLOCK);
2124 	}
2125 	FILE_UNLOCK(fp);
2126 	fdrop(fp, td);
2127 	if (uap->oattr)
2128 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2129 	return (error);
2130 }
2131 
2132 int
2133 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2134 {
2135 	struct mqueue *mq;
2136 	struct file *fp;
2137 	int error;
2138 	int waitok;
2139 
2140 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2141 	if (error)
2142 		return (error);
2143 	waitok = !(fp->f_flag & O_NONBLOCK);
2144 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2145 		uap->msg_prio, waitok, uap->abs_timeout);
2146 	fdrop(fp, td);
2147 	return (error);
2148 }
2149 
2150 int
2151 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2152 {
2153 	struct mqueue *mq;
2154 	struct file *fp;
2155 	int error, waitok;
2156 
2157 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2158 	if (error)
2159 		return (error);
2160 	waitok = !(fp->f_flag & O_NONBLOCK);
2161 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2162 		uap->msg_prio, waitok, uap->abs_timeout);
2163 	fdrop(fp, td);
2164 	return (error);
2165 }
2166 
2167 int
2168 kmq_notify(struct thread *td, struct kmq_notify_args *uap)
2169 {
2170 	struct sigevent ev;
2171 	struct filedesc *fdp;
2172 	struct proc *p;
2173 	struct mqueue *mq;
2174 	struct file *fp;
2175 	struct mqueue_notifier *nt, *newnt = NULL;
2176 	int error;
2177 
2178 	p = td->td_proc;
2179 	fdp = td->td_proc->p_fd;
2180 	if (uap->sigev) {
2181 		error = copyin(uap->sigev, &ev, sizeof(ev));
2182 		if (error)
2183 			return (error);
2184 		if (ev.sigev_notify != SIGEV_SIGNAL &&
2185 		    ev.sigev_notify != SIGEV_THREAD_ID &&
2186 		    ev.sigev_notify != SIGEV_NONE)
2187 			return (EINVAL);
2188 		if ((ev.sigev_notify == SIGEV_SIGNAL ||
2189 		     ev.sigev_notify == SIGEV_THREAD_ID) &&
2190 			!_SIG_VALID(ev.sigev_signo))
2191 			return (EINVAL);
2192 	}
2193 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2194 	if (error)
2195 		return (error);
2196 again:
2197 	FILEDESC_SLOCK(fdp);
2198 	if (fget_locked(fdp, uap->mqd) != fp) {
2199 		FILEDESC_SUNLOCK(fdp);
2200 		error = EBADF;
2201 		goto out;
2202 	}
2203 	mtx_lock(&mq->mq_mutex);
2204 	FILEDESC_SUNLOCK(fdp);
2205 	if (uap->sigev != NULL) {
2206 		if (mq->mq_notifier != NULL) {
2207 			error = EBUSY;
2208 		} else {
2209 			PROC_LOCK(p);
2210 			nt = notifier_search(p, uap->mqd);
2211 			if (nt == NULL) {
2212 				if (newnt == NULL) {
2213 					PROC_UNLOCK(p);
2214 					mtx_unlock(&mq->mq_mutex);
2215 					newnt = notifier_alloc();
2216 					goto again;
2217 				}
2218 			}
2219 
2220 			if (nt != NULL) {
2221 				sigqueue_take(&nt->nt_ksi);
2222 				if (newnt != NULL) {
2223 					notifier_free(newnt);
2224 					newnt = NULL;
2225 				}
2226 			} else {
2227 				nt = newnt;
2228 				newnt = NULL;
2229 				ksiginfo_init(&nt->nt_ksi);
2230 				nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
2231 				nt->nt_ksi.ksi_code = SI_MESGQ;
2232 				nt->nt_proc = p;
2233 				nt->nt_ksi.ksi_mqd = uap->mqd;
2234 				notifier_insert(p, nt);
2235 			}
2236 			nt->nt_sigev = ev;
2237 			mq->mq_notifier = nt;
2238 			PROC_UNLOCK(p);
2239 			/*
2240 			 * if there is no receivers and message queue
2241 			 * is not empty, we should send notification
2242 			 * as soon as possible.
2243 			 */
2244 			if (mq->mq_receivers == 0 &&
2245 			    !TAILQ_EMPTY(&mq->mq_msgq))
2246 				mqueue_send_notification(mq);
2247 		}
2248 	} else {
2249 		notifier_remove(p, mq, uap->mqd);
2250 	}
2251 	mtx_unlock(&mq->mq_mutex);
2252 
2253 out:
2254 	fdrop(fp, td);
2255 	if (newnt != NULL)
2256 		notifier_free(newnt);
2257 	return (error);
2258 }
2259 
2260 static void
2261 mqueue_fdclose(struct thread *td, int fd, struct file *fp)
2262 {
2263 	struct filedesc *fdp;
2264 	struct mqueue *mq;
2265 
2266 	fdp = td->td_proc->p_fd;
2267 	FILEDESC_LOCK_ASSERT(fdp);
2268 
2269 	if (fp->f_ops == &mqueueops) {
2270 		mq = FPTOMQ(fp);
2271 		mtx_lock(&mq->mq_mutex);
2272 		notifier_remove(td->td_proc, mq, fd);
2273 
2274 		/* have to wakeup thread in same process */
2275 		if (mq->mq_flags & MQ_RSEL) {
2276 			mq->mq_flags &= ~MQ_RSEL;
2277 			selwakeup(&mq->mq_rsel);
2278 		}
2279 		if (mq->mq_flags & MQ_WSEL) {
2280 			mq->mq_flags &= ~MQ_WSEL;
2281 			selwakeup(&mq->mq_wsel);
2282 		}
2283 		mtx_unlock(&mq->mq_mutex);
2284 	}
2285 }
2286 
2287 static void
2288 mq_proc_exit(void *arg __unused, struct proc *p)
2289 {
2290 	struct filedesc *fdp;
2291 	struct file *fp;
2292 	struct mqueue *mq;
2293 	int i;
2294 
2295 	fdp = p->p_fd;
2296 	FILEDESC_SLOCK(fdp);
2297 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2298 		fp = fget_locked(fdp, i);
2299 		if (fp != NULL && fp->f_ops == &mqueueops) {
2300 			mq = FPTOMQ(fp);
2301 			mtx_lock(&mq->mq_mutex);
2302 			notifier_remove(p, FPTOMQ(fp), i);
2303 			mtx_unlock(&mq->mq_mutex);
2304 		}
2305 	}
2306 	FILEDESC_SUNLOCK(fdp);
2307 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2308 }
2309 
2310 static int
2311 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2312 	int flags, struct thread *td)
2313 {
2314 	return (EOPNOTSUPP);
2315 }
2316 
2317 static int
2318 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2319 	int flags, struct thread *td)
2320 {
2321 	return (EOPNOTSUPP);
2322 }
2323 
2324 static int
2325 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2326 	struct ucred *active_cred, struct thread *td)
2327 {
2328 	return (ENOTTY);
2329 }
2330 
2331 static int
2332 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2333 	struct thread *td)
2334 {
2335 	struct mqueue *mq = FPTOMQ(fp);
2336 	int revents = 0;
2337 
2338 	mtx_lock(&mq->mq_mutex);
2339 	if (events & (POLLIN | POLLRDNORM)) {
2340 		if (mq->mq_curmsgs) {
2341 			revents |= events & (POLLIN | POLLRDNORM);
2342 		} else {
2343 			mq->mq_flags |= MQ_RSEL;
2344 			selrecord(td, &mq->mq_rsel);
2345  		}
2346 	}
2347 	if (events & POLLOUT) {
2348 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2349 			revents |= POLLOUT;
2350 		else {
2351 			mq->mq_flags |= MQ_WSEL;
2352 			selrecord(td, &mq->mq_wsel);
2353 		}
2354 	}
2355 	mtx_unlock(&mq->mq_mutex);
2356 	return (revents);
2357 }
2358 
2359 static int
2360 mqf_close(struct file *fp, struct thread *td)
2361 {
2362 	struct mqfs_node *pn;
2363 
2364 	fp->f_ops = &badfileops;
2365 	pn = fp->f_data;
2366 	fp->f_data = NULL;
2367 	sx_xlock(&mqfs_data.mi_lock);
2368 	mqnode_release(pn);
2369 	sx_xunlock(&mqfs_data.mi_lock);
2370 	return (0);
2371 }
2372 
2373 static int
2374 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2375 	struct thread *td)
2376 {
2377 	struct mqfs_node *pn = fp->f_data;
2378 
2379 	bzero(st, sizeof *st);
2380 	st->st_atimespec = pn->mn_atime;
2381 	st->st_mtimespec = pn->mn_mtime;
2382 	st->st_ctimespec = pn->mn_ctime;
2383 	st->st_birthtimespec = pn->mn_birth;
2384 	st->st_uid = pn->mn_uid;
2385 	st->st_gid = pn->mn_gid;
2386 	st->st_mode = S_IFIFO | pn->mn_mode;
2387 	return (0);
2388 }
2389 
2390 static int
2391 mqf_kqfilter(struct file *fp, struct knote *kn)
2392 {
2393 	struct mqueue *mq = FPTOMQ(fp);
2394 	int error = 0;
2395 
2396 	if (kn->kn_filter == EVFILT_READ) {
2397 		kn->kn_fop = &mq_rfiltops;
2398 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2399 	} else if (kn->kn_filter == EVFILT_WRITE) {
2400 		kn->kn_fop = &mq_wfiltops;
2401 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2402 	} else
2403 		error = EINVAL;
2404 	return (error);
2405 }
2406 
2407 static void
2408 filt_mqdetach(struct knote *kn)
2409 {
2410 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2411 
2412 	if (kn->kn_filter == EVFILT_READ)
2413 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2414 	else if (kn->kn_filter == EVFILT_WRITE)
2415 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2416 	else
2417 		panic("filt_mqdetach");
2418 }
2419 
2420 static int
2421 filt_mqread(struct knote *kn, long hint)
2422 {
2423 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2424 
2425 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2426 	return (mq->mq_curmsgs != 0);
2427 }
2428 
2429 static int
2430 filt_mqwrite(struct knote *kn, long hint)
2431 {
2432 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2433 
2434 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2435 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2436 }
2437 
2438 static struct fileops mqueueops = {
2439 	.fo_read		= mqf_read,
2440 	.fo_write		= mqf_write,
2441 	.fo_ioctl		= mqf_ioctl,
2442 	.fo_poll		= mqf_poll,
2443 	.fo_kqfilter		= mqf_kqfilter,
2444 	.fo_stat		= mqf_stat,
2445 	.fo_close		= mqf_close
2446 };
2447 
2448 static struct vop_vector mqfs_vnodeops = {
2449 	.vop_default 		= &default_vnodeops,
2450 	.vop_access		= mqfs_access,
2451 	.vop_cachedlookup	= mqfs_lookup,
2452 	.vop_lookup		= vfs_cache_lookup,
2453 	.vop_reclaim		= mqfs_reclaim,
2454 	.vop_create		= mqfs_create,
2455 	.vop_remove		= mqfs_remove,
2456 	.vop_inactive		= mqfs_inactive,
2457 	.vop_open		= mqfs_open,
2458 	.vop_close		= mqfs_close,
2459 	.vop_getattr		= mqfs_getattr,
2460 	.vop_setattr		= mqfs_setattr,
2461 	.vop_read		= mqfs_read,
2462 	.vop_write		= VOP_EOPNOTSUPP,
2463 	.vop_readdir		= mqfs_readdir,
2464 	.vop_mkdir		= VOP_EOPNOTSUPP,
2465 	.vop_rmdir		= VOP_EOPNOTSUPP
2466 };
2467 
2468 static struct vfsops mqfs_vfsops = {
2469 	.vfs_init 		= mqfs_init,
2470 	.vfs_uninit		= mqfs_uninit,
2471 	.vfs_mount		= mqfs_mount,
2472 	.vfs_unmount		= mqfs_unmount,
2473 	.vfs_root		= mqfs_root,
2474 	.vfs_statfs		= mqfs_statfs,
2475 };
2476 
2477 SYSCALL_MODULE_HELPER(kmq_open);
2478 SYSCALL_MODULE_HELPER(kmq_setattr);
2479 SYSCALL_MODULE_HELPER(kmq_timedsend);
2480 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2481 SYSCALL_MODULE_HELPER(kmq_notify);
2482 SYSCALL_MODULE_HELPER(kmq_unlink);
2483 
2484 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2485 MODULE_VERSION(mqueuefs, 1);
2486