xref: /freebsd/sys/kern/uipc_mqueue.c (revision 87569f75a91f298c52a71823c04d41cf53c88889)
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/poll.h>
68 #include <sys/proc.h>
69 #include <sys/queue.h>
70 #include <sys/sysproto.h>
71 #include <sys/stat.h>
72 #include <sys/syscall.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysent.h>
75 #include <sys/sx.h>
76 #include <sys/sysctl.h>
77 #include <sys/taskqueue.h>
78 #include <sys/unistd.h>
79 #include <sys/vnode.h>
80 #include <machine/atomic.h>
81 #include <posix4/posix4.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 	mp->mnt_flag |= MNT_LOCAL;
565 	mp->mnt_kern_flag |= MNTK_MPSAFE;
566 	vfs_getnewfsid(mp);
567 
568 	sbp = &mp->mnt_stat;
569 	vfs_mountedfrom(mp, "mqueue");
570 	sbp->f_bsize = PAGE_SIZE;
571 	sbp->f_iosize = PAGE_SIZE;
572 	sbp->f_blocks = 1;
573 	sbp->f_bfree = 0;
574 	sbp->f_bavail = 0;
575 	sbp->f_files = 1;
576 	sbp->f_ffree = 0;
577 	return (0);
578 }
579 
580 /*
581  * Unmount a mqfs instance
582  */
583 static int
584 mqfs_unmount(struct mount *mp, int mntflags, struct thread *td)
585 {
586 	int error;
587 
588 	error = vflush(mp, 0, (mntflags & MNT_FORCE) ?  FORCECLOSE : 0, td);
589 	return (error);
590 }
591 
592 /*
593  * Return a root vnode
594  */
595 static int
596 mqfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td)
597 {
598 	struct mqfs_info *mqfs;
599 	int ret;
600 
601 	mqfs = VFSTOMQFS(mp);
602 	sx_xlock(&mqfs->mi_lock);
603 	ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
604 	sx_xunlock(&mqfs->mi_lock);
605 	return (ret);
606 }
607 
608 /*
609  * Return filesystem stats
610  */
611 static int
612 mqfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
613 {
614 	/* XXX update statistics */
615 	return (0);
616 }
617 
618 /*
619  * Initialize a mqfs instance
620  */
621 static int
622 mqfs_init(struct vfsconf *vfc)
623 {
624 	struct mqfs_node *root;
625 	struct mqfs_info *mi;
626 
627 	mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
628 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
629 	mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
630 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
631 	mvdata_zone = uma_zcreate("mvdata",
632 		sizeof(struct mqfs_vdata), NULL, NULL, NULL,
633 		NULL, UMA_ALIGN_PTR, 0);
634 	mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
635 		NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
636 	mi = &mqfs_data;
637 	sx_init(&mi->mi_lock, "mqfs lock");
638 	/* set up the root diretory */
639 	root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
640 		mqfstype_root);
641 	root->mn_info = mi;
642 	LIST_INIT(&root->mn_children);
643 	LIST_INIT(&root->mn_vnodes);
644 	mi->mi_root = root;
645 	mqfs_fileno_init(mi);
646 	mqfs_fileno_alloc(mi, root);
647 	mqfs_fixup_dir(root);
648 	exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
649 	    EVENTHANDLER_PRI_ANY);
650 	mq_fdclose = mqueue_fdclose;
651 	p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
652 	return (0);
653 }
654 
655 /*
656  * Destroy a mqfs instance
657  */
658 static int
659 mqfs_uninit(struct vfsconf *vfc)
660 {
661 	struct mqfs_info *mi;
662 
663 	if (!unloadable)
664 		return (EOPNOTSUPP);
665 	EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
666 	mi = &mqfs_data;
667 	mqfs_destroy(mi->mi_root);
668 	mi->mi_root = NULL;
669 	mqfs_fileno_uninit(mi);
670 	sx_destroy(&mi->mi_lock);
671 	uma_zdestroy(mqnode_zone);
672 	uma_zdestroy(mqueue_zone);
673 	uma_zdestroy(mvdata_zone);
674 	uma_zdestroy(mqnoti_zone);
675 	return (0);
676 }
677 
678 /*
679  * task routine
680  */
681 static void
682 do_recycle(void *context, int pending __unused)
683 {
684 	struct vnode *vp = (struct vnode *)context;
685 
686 	vrecycle(vp, curthread);
687 	vdrop(vp);
688 }
689 
690 /*
691  * Allocate a vnode
692  */
693 static int
694 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
695 {
696 	struct mqfs_vdata *vd;
697 	int error;
698 
699 	LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
700 		if (vd->mv_vnode->v_mount == mp)
701 			break;
702 	}
703 
704 	if (vd != NULL) {
705 		if (vget(vd->mv_vnode, 0, curthread) == 0) {
706 			*vpp = vd->mv_vnode;
707 			vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE,
708 			    curthread);
709 			return (0);
710 		}
711 		/* XXX if this can happen, we're in trouble */
712 	}
713 
714 	error = getnewvnode("mqueue", mp, &mqfs_vnodeops, vpp);
715 	if (error)
716 		return (error);
717 	vd = uma_zalloc(mvdata_zone, M_WAITOK);
718 	(*vpp)->v_data = vd;
719 	vd->mv_vnode = *vpp;
720 	vd->mv_node = pn;
721 	TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
722 	LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
723 	mqnode_addref(pn);
724 	switch (pn->mn_type) {
725 	case mqfstype_root:
726 		(*vpp)->v_vflag = VV_ROOT;
727 		/* fall through */
728 	case mqfstype_dir:
729 	case mqfstype_this:
730 	case mqfstype_parent:
731 		(*vpp)->v_type = VDIR;
732 		break;
733 	case mqfstype_file:
734 		(*vpp)->v_type = VREG;
735 		break;
736 	case mqfstype_symlink:
737 		(*vpp)->v_type = VLNK;
738 		break;
739 	case mqfstype_none:
740 		KASSERT(0, ("mqfs_allocf called for null node\n"));
741 	default:
742 		panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
743 	}
744 	vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
745 	return (0);
746 }
747 
748 /*
749  * Search a directory entry
750  */
751 static struct mqfs_node *
752 mqfs_search(struct mqfs_node *pd, const char *name, int len)
753 {
754 	struct mqfs_node *pn;
755 
756 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
757 		if (strncmp(pn->mn_name, name, len) == 0)
758 			return (pn);
759 	}
760 	return (NULL);
761 }
762 
763 /*
764  * Look up a file or directory
765  */
766 static int
767 mqfs_lookupx(struct vop_cachedlookup_args *ap)
768 {
769 	struct componentname *cnp;
770 	struct vnode *dvp, **vpp;
771 	struct mqfs_node *pd;
772 	struct mqfs_node *pn;
773 	int nameiop, flags, error, namelen;
774 	char *pname;
775 	struct thread *td;
776 
777 	cnp = ap->a_cnp;
778 	vpp = ap->a_vpp;
779 	dvp = ap->a_dvp;
780 	pname = cnp->cn_nameptr;
781 	namelen = cnp->cn_namelen;
782 	td = cnp->cn_thread;
783 	flags = cnp->cn_flags;
784 	nameiop = cnp->cn_nameiop;
785 	pd = VTON(dvp);
786 	pn = NULL;
787 	*vpp = NULLVP;
788 
789 	if (dvp->v_type != VDIR)
790 		return (ENOTDIR);
791 
792 	error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
793 	if (error)
794 		return (error);
795 
796 	/* shortcut: check if the name is too long */
797 	if (cnp->cn_namelen >= MQFS_NAMELEN)
798 		return (ENOENT);
799 
800 	/* self */
801 	if (namelen == 1 && pname[0] == '.') {
802 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
803 			return (EINVAL);
804 		pn = pd;
805 		*vpp = dvp;
806 		VREF(dvp);
807 		return (0);
808 	}
809 
810 	/* parent */
811 	if (cnp->cn_flags & ISDOTDOT) {
812 		if (dvp->v_vflag & VV_ROOT)
813 			return (EIO);
814 		if ((flags & ISLASTCN) && nameiop != LOOKUP)
815 			return (EINVAL);
816 		VOP_UNLOCK(dvp, 0, cnp->cn_thread);
817 		KASSERT(pd->mn_parent, ("non-root directory has no parent"));
818 		pn = pd->mn_parent;
819 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
820 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
821 		return (error);
822 	}
823 
824 	/* named node */
825 	pn = mqfs_search(pd, pname, namelen);
826 
827 	/* found */
828 	if (pn != NULL) {
829 		/* DELETE */
830 		if (nameiop == DELETE && (flags & ISLASTCN)) {
831 			error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
832 			if (error)
833 				return (error);
834 			if (*vpp == dvp) {
835 				VREF(dvp);
836 				*vpp = dvp;
837 				return (0);
838 			}
839 		}
840 
841 		/* allocate vnode */
842 		error = mqfs_allocv(dvp->v_mount, vpp, pn);
843 		if (error == 0 && cnp->cn_flags & MAKEENTRY)
844 			cache_enter(dvp, *vpp, cnp);
845 		return (error);
846 	}
847 
848 	/* not found */
849 
850 	/* will create a new entry in the directory ? */
851 	if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
852 	    && (flags & ISLASTCN)) {
853 		error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
854 		if (error)
855 			return (error);
856 		cnp->cn_flags |= SAVENAME;
857 		return (EJUSTRETURN);
858 	}
859 	return (ENOENT);
860 }
861 
862 #if 0
863 struct vop_lookup_args {
864 	struct vop_generic_args a_gen;
865 	struct vnode *a_dvp;
866 	struct vnode **a_vpp;
867 	struct componentname *a_cnp;
868 };
869 #endif
870 
871 /*
872  * vnode lookup operation
873  */
874 static int
875 mqfs_lookup(struct vop_cachedlookup_args *ap)
876 {
877 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
878 	int rc;
879 
880 	sx_xlock(&mqfs->mi_lock);
881 	rc = mqfs_lookupx(ap);
882 	sx_xunlock(&mqfs->mi_lock);
883 	return (rc);
884 }
885 
886 #if 0
887 struct vop_create_args {
888 	struct vnode *a_dvp;
889 	struct vnode **a_vpp;
890 	struct componentname *a_cnp;
891 	struct vattr *a_vap;
892 };
893 #endif
894 
895 /*
896  * vnode creation operation
897  */
898 static int
899 mqfs_create(struct vop_create_args *ap)
900 {
901 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
902 	struct componentname *cnp = ap->a_cnp;
903 	struct mqfs_node *pd;
904 	struct mqfs_node *pn;
905 	struct mqueue *mq;
906 	int error;
907 
908 	pd = VTON(ap->a_dvp);
909 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
910 		return (ENOTDIR);
911 	mq = mqueue_alloc(NULL);
912 	if (mq == NULL)
913 		return (EAGAIN);
914 	sx_xlock(&mqfs->mi_lock);
915 #if 0
916 	/* named node */
917 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
918 	if (pn != NULL) {
919 		mqueue_free(mq);
920 		sx_xunlock(&mqfs->mi_lock);
921 		return (EEXIST);
922 	}
923 #else
924 	if ((cnp->cn_flags & HASBUF) == 0)
925 		panic("%s: no name", __func__);
926 #endif
927 	pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
928 		cnp->cn_cred, ap->a_vap->va_mode);
929 	if (pn == NULL)
930 		error = ENOSPC;
931 	else {
932 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
933 		if (error)
934 			mqfs_destroy(pn);
935 		else
936 			pn->mn_data = mq;
937 	}
938 	sx_xunlock(&mqfs->mi_lock);
939 	if (error)
940 		mqueue_free(mq);
941 	return (error);
942 }
943 
944 /*
945  * Remove an entry
946  */
947 static
948 int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
949 {
950 	struct mqfs_node *parent;
951 	struct mqfs_vdata *vd;
952 	int error = 0;
953 
954 	sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
955 
956 	if (ucred->cr_uid != pn->mn_uid &&
957 	    (error = suser_cred(ucred, 0)) != 0)
958 		error = EACCES;
959 	else if (!pn->mn_deleted) {
960 		parent = pn->mn_parent;
961 		pn->mn_parent = NULL;
962 		pn->mn_deleted = 1;
963 		LIST_REMOVE(pn, mn_sibling);
964 		LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
965 			cache_purge(vd->mv_vnode);
966 			vhold(vd->mv_vnode);
967 			taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
968 		}
969 		mqnode_release(pn);
970 		mqnode_release(parent);
971 	} else
972 		error = ENOENT;
973 	return (error);
974 }
975 
976 #if 0
977 struct vop_remove_args {
978 	struct vnode *a_dvp;
979 	struct vnode *a_vp;
980 	struct componentname *a_cnp;
981 };
982 #endif
983 
984 /*
985  * vnode removal operation
986  */
987 static int
988 mqfs_remove(struct vop_remove_args *ap)
989 {
990 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
991 	struct mqfs_node *pn;
992 	int error;
993 
994 	if (ap->a_vp->v_type == VDIR)
995                 return (EPERM);
996 	pn = VTON(ap->a_vp);
997 	sx_xlock(&mqfs->mi_lock);
998 	error = do_unlink(pn, ap->a_cnp->cn_cred);
999 	sx_xunlock(&mqfs->mi_lock);
1000 	return (error);
1001 }
1002 
1003 #if 0
1004 struct vop_inactive_args {
1005 	struct vnode *a_vp;
1006 	struct thread *a_td;
1007 };
1008 #endif
1009 
1010 static int
1011 mqfs_inactive(struct vop_inactive_args *ap)
1012 {
1013 	struct mqfs_node *pn = VTON(ap->a_vp);
1014 
1015 	if (pn->mn_deleted)
1016 		vrecycle(ap->a_vp, ap->a_td);
1017 	return (0);
1018 }
1019 
1020 #if 0
1021 struct vop_reclaim_args {
1022 	struct vop_generic_args a_gen;
1023 	struct vnode *a_vp;
1024 	struct thread *a_td;
1025 };
1026 #endif
1027 
1028 static int
1029 mqfs_reclaim(struct vop_reclaim_args *ap)
1030 {
1031 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
1032 	struct vnode *vp = ap->a_vp;
1033 	struct mqfs_node *pn;
1034 	struct mqfs_vdata *vd;
1035 
1036 	vd = vp->v_data;
1037 	pn = vd->mv_node;
1038 	sx_xlock(&mqfs->mi_lock);
1039 	vp->v_data = NULL;
1040 	LIST_REMOVE(vd, mv_link);
1041 	uma_zfree(mvdata_zone, vd);
1042 	mqnode_release(pn);
1043 	sx_xunlock(&mqfs->mi_lock);
1044 	return (0);
1045 }
1046 
1047 #if 0
1048 struct vop_open_args {
1049 	struct vop_generic_args a_gen;
1050 	struct vnode *a_vp;
1051 	int a_mode;
1052 	struct ucred *a_cred;
1053 	struct thread *a_td;
1054 	int a_fdidx;
1055 };
1056 #endif
1057 
1058 static int
1059 mqfs_open(struct vop_open_args *ap)
1060 {
1061 	return (0);
1062 }
1063 
1064 #if 0
1065 struct vop_close_args {
1066 	struct vop_generic_args a_gen;
1067 	struct vnode *a_vp;
1068 	int a_fflag;
1069 	struct ucred *a_cred;
1070 	struct thread *a_td;
1071 };
1072 #endif
1073 
1074 static int
1075 mqfs_close(struct vop_close_args *ap)
1076 {
1077 	return (0);
1078 }
1079 
1080 #if 0
1081 struct vop_access_args {
1082 	struct vop_generic_args a_gen;
1083 	struct vnode *a_vp;
1084 	int a_mode;
1085 	struct ucred *a_cred;
1086 	struct thread *a_td;
1087 };
1088 #endif
1089 
1090 /*
1091  * Verify permissions
1092  */
1093 static int
1094 mqfs_access(struct vop_access_args *ap)
1095 {
1096 	struct vnode *vp = ap->a_vp;
1097 	struct vattr vattr;
1098 	int error;
1099 
1100 	error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
1101 	if (error)
1102 		return (error);
1103 	error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
1104 	    vattr.va_gid, ap->a_mode, ap->a_cred, NULL);
1105 	return (error);
1106 }
1107 
1108 #if 0
1109 struct vop_getattr_args {
1110 	struct vop_generic_args a_gen;
1111 	struct vnode *a_vp;
1112 	struct vattr *a_vap;
1113 	struct ucred *a_cred;
1114 	struct thread *a_td;
1115 };
1116 #endif
1117 
1118 /*
1119  * Get file attributes
1120  */
1121 static int
1122 mqfs_getattr(struct vop_getattr_args *ap)
1123 {
1124 	struct vnode *vp = ap->a_vp;
1125 	struct mqfs_node *pn = VTON(vp);
1126 	struct vattr *vap = ap->a_vap;
1127 	int error = 0;
1128 
1129 	VATTR_NULL(vap);
1130 	vap->va_type = vp->v_type;
1131 	vap->va_mode = pn->mn_mode;
1132 	vap->va_nlink = 1;
1133 	vap->va_uid = pn->mn_uid;
1134 	vap->va_gid = pn->mn_gid;
1135 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1136 	vap->va_fileid = pn->mn_fileno;
1137 	vap->va_size = 0;
1138 	vap->va_blocksize = PAGE_SIZE;
1139 	vap->va_bytes = vap->va_size = 0;
1140 	vap->va_atime = pn->mn_atime;
1141 	vap->va_mtime = pn->mn_mtime;
1142 	vap->va_ctime = pn->mn_ctime;
1143 	vap->va_birthtime = pn->mn_birth;
1144 	vap->va_gen = 0;
1145 	vap->va_flags = 0;
1146 	vap->va_rdev = 0;
1147 	vap->va_bytes = 0;
1148 	vap->va_filerev = 0;
1149 	vap->va_vaflags = 0;
1150 	return (error);
1151 }
1152 
1153 #if 0
1154 struct vop_setattr_args {
1155 	struct vop_generic_args a_gen;
1156 	struct vnode *a_vp;
1157 	struct vattr *a_vap;
1158 	struct ucred *a_cred;
1159 	struct thread *a_td;
1160 };
1161 #endif
1162 /*
1163  * Set attributes
1164  */
1165 static int
1166 mqfs_setattr(struct vop_setattr_args *ap)
1167 {
1168 	struct mqfs_node *pn;
1169 	struct vattr *vap;
1170 	struct vnode *vp;
1171 	int c, error;
1172 	uid_t uid;
1173 	gid_t gid;
1174 
1175 	vap = ap->a_vap;
1176 	vp = ap->a_vp;
1177 	if ((vap->va_type != VNON) ||
1178 	    (vap->va_nlink != VNOVAL) ||
1179 	    (vap->va_fsid != VNOVAL) ||
1180 	    (vap->va_fileid != VNOVAL) ||
1181 	    (vap->va_blocksize != VNOVAL) ||
1182 	    (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1183 	    (vap->va_rdev != VNOVAL) ||
1184 	    ((int)vap->va_bytes != VNOVAL) ||
1185 	    (vap->va_gen != VNOVAL)) {
1186 		return (EINVAL);
1187 	}
1188 
1189 	pn = VTON(vp);
1190 
1191 	error = c = 0;
1192 	if (vap->va_uid == (uid_t)VNOVAL)
1193 		uid = pn->mn_uid;
1194 	else
1195 		uid = vap->va_uid;
1196 	if (vap->va_gid == (gid_t)VNOVAL)
1197 		gid = pn->mn_gid;
1198 	else
1199 		gid = vap->va_gid;
1200 
1201 	if (uid != pn->mn_uid || gid != pn->mn_gid) {
1202 		/*
1203 		 * To modify the ownership of a file, must possess VADMIN
1204 		 * for that file.
1205 		 */
1206 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)))
1207 			return (error);
1208 		if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
1209 		    (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
1210 		    (error = suser_cred(ap->a_td->td_ucred, SUSER_ALLOWJAIL))
1211                        != 0)
1212 			return (error);
1213 		pn->mn_uid = uid;
1214 		pn->mn_gid = gid;
1215 		c = 1;
1216 	}
1217 
1218 	if (vap->va_mode != (mode_t)VNOVAL) {
1219 		if ((ap->a_cred->cr_uid != pn->mn_uid) &&
1220 		    (error = suser_cred(ap->a_td->td_ucred, SUSER_ALLOWJAIL)))
1221 			return (error);
1222 		pn->mn_mode = vap->va_mode;
1223 		c = 1;
1224 	}
1225 
1226 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1227 		/* See the comment in ufs_vnops::ufs_setattr(). */
1228 		if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, ap->a_td)) &&
1229 		    ((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
1230 		    (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, ap->a_td))))
1231 			return (error);
1232 		if (vap->va_atime.tv_sec != VNOVAL) {
1233 			pn->mn_atime = vap->va_atime;
1234 		}
1235 		if (vap->va_mtime.tv_sec != VNOVAL) {
1236 			pn->mn_mtime = vap->va_mtime;
1237 		}
1238 		c = 1;
1239 	}
1240 	if (c) {
1241 		vfs_timestamp(&pn->mn_ctime);
1242 	}
1243 	return (0);
1244 }
1245 
1246 #if 0
1247 struct vop_read_args {
1248 	struct vop_generic_args a_gen;
1249 	struct vnode *a_vp;
1250 	struct uio *a_uio;
1251 	int a_ioflag;
1252 	struct ucred *a_cred;
1253 };
1254 #endif
1255 
1256 /*
1257  * Read from a file
1258  */
1259 static int
1260 mqfs_read(struct vop_read_args *ap)
1261 {
1262 	char buf[80];
1263 	struct vnode *vp = ap->a_vp;
1264 	struct uio *uio = ap->a_uio;
1265 	struct mqfs_node *pn;
1266 	struct mqueue *mq;
1267 	int len, error;
1268 
1269 	if (vp->v_type != VREG)
1270 		return (EINVAL);
1271 
1272 	pn = VTON(vp);
1273 	mq = VTOMQ(vp);
1274 	snprintf(buf, sizeof(buf),
1275 		"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
1276 		mq->mq_totalbytes,
1277 		mq->mq_maxmsg,
1278 		mq->mq_curmsgs,
1279 		mq->mq_msgsize);
1280 	buf[sizeof(buf)-1] = '\0';
1281 	len = strlen(buf);
1282 	error = uiomove_frombuf(buf, len, uio);
1283 	return (error);
1284 }
1285 
1286 #if 0
1287 struct vop_readdir_args {
1288 	struct vop_generic_args a_gen;
1289 	struct vnode *a_vp;
1290 	struct uio *a_uio;
1291 	struct ucred *a_cred;
1292 	int *a_eofflag;
1293 	int *a_ncookies;
1294 	u_long **a_cookies;
1295 };
1296 #endif
1297 
1298 /*
1299  * Return directory entries.
1300  */
1301 static int
1302 mqfs_readdir(struct vop_readdir_args *ap)
1303 {
1304 	struct vnode *vp;
1305 	struct mqfs_info *mi;
1306 	struct mqfs_node *pd;
1307 	struct mqfs_node *pn;
1308 	struct dirent entry;
1309 	struct uio *uio;
1310 	int *tmp_ncookies = NULL;
1311 	off_t offset;
1312 	int error, i;
1313 
1314 	vp = ap->a_vp;
1315 	mi = VFSTOMQFS(vp->v_mount);
1316 	pd = VTON(vp);
1317 	uio = ap->a_uio;
1318 
1319 	if (vp->v_type != VDIR)
1320 		return (ENOTDIR);
1321 
1322 	if (uio->uio_offset < 0)
1323 		return (EINVAL);
1324 
1325 	if (ap->a_ncookies != NULL) {
1326 		tmp_ncookies = ap->a_ncookies;
1327 		*ap->a_ncookies = 0;
1328 		ap->a_ncookies = NULL;
1329         }
1330 
1331 	error = 0;
1332 	offset = 0;
1333 
1334 	sx_xlock(&mi->mi_lock);
1335 
1336 	LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
1337 		entry.d_reclen = sizeof(entry);
1338 		if (!pn->mn_fileno)
1339 			mqfs_fileno_alloc(mi, pn);
1340 		entry.d_fileno = pn->mn_fileno;
1341 		for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
1342 			entry.d_name[i] = pn->mn_name[i];
1343 		entry.d_name[i] = 0;
1344 		entry.d_namlen = i;
1345 		switch (pn->mn_type) {
1346 		case mqfstype_root:
1347 		case mqfstype_dir:
1348 		case mqfstype_this:
1349 		case mqfstype_parent:
1350 			entry.d_type = DT_DIR;
1351 			break;
1352 		case mqfstype_file:
1353 			entry.d_type = DT_REG;
1354 			break;
1355 		case mqfstype_symlink:
1356 			entry.d_type = DT_LNK;
1357 			break;
1358 		default:
1359 			panic("%s has unexpected node type: %d", pn->mn_name,
1360 				pn->mn_type);
1361 		}
1362 		if (entry.d_reclen > uio->uio_resid)
1363                         break;
1364 		if (offset >= uio->uio_offset) {
1365 			error = vfs_read_dirent(ap, &entry, offset);
1366                         if (error)
1367                                 break;
1368                 }
1369                 offset += entry.d_reclen;
1370 	}
1371 	sx_xunlock(&mi->mi_lock);
1372 
1373 	uio->uio_offset = offset;
1374 
1375 	if (tmp_ncookies != NULL)
1376 		ap->a_ncookies = tmp_ncookies;
1377 
1378 	return (error);
1379 }
1380 
1381 #ifdef notyet
1382 
1383 #if 0
1384 struct vop_mkdir_args {
1385 	struct vnode *a_dvp;
1386 	struvt vnode **a_vpp;
1387 	struvt componentname *a_cnp;
1388 	struct vattr *a_vap;
1389 };
1390 #endif
1391 
1392 /*
1393  * Create a directory.
1394  */
1395 static int
1396 mqfs_mkdir(struct vop_mkdir_args *ap)
1397 {
1398 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1399 	struct componentname *cnp = ap->a_cnp;
1400 	struct mqfs_node *pd = VTON(ap->a_dvp);
1401 	struct mqfs_node *pn;
1402 	int error;
1403 
1404 	if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
1405 		return (ENOTDIR);
1406 	sx_xlock(&mqfs->mi_lock);
1407 #if 0
1408 	/* named node */
1409 	pn = mqfs_search(pd, cnp->cn_nameptr, cnp->cn_namelen);
1410 	if (pn != NULL) {
1411 		sx_xunlock(&mqfs->mi_lock);
1412 		return (EEXIST);
1413 	}
1414 #else
1415 	if ((cnp->cn_flags & HASBUF) == 0)
1416 		panic("%s: no name", __func__);
1417 #endif
1418 	pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
1419 		ap->a_vap->cn_cred, ap->a_vap->va_mode);
1420 	if (pn == NULL)
1421 		error = ENOSPC;
1422 	else
1423 		error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
1424 	sx_xunlock(&mqfs->mi_lock);
1425 	return (error);
1426 }
1427 
1428 #if 0
1429 struct vop_rmdir_args {
1430 	struct vnode *a_dvp;
1431 	struct vnode *a_vp;
1432 	struct componentname *a_cnp;
1433 };
1434 #endif
1435 
1436 /*
1437  * Remove a directory.
1438  */
1439 static int
1440 mqfs_rmdir(struct vop_rmdir_args *ap)
1441 {
1442 	struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
1443 	struct mqfs_node *pn = VTON(ap->a_vp);
1444 	struct mqfs_node *pt;
1445 
1446 	if (pn->mn_type != mqfstype_dir)
1447 		return (ENOTDIR);
1448 
1449 	sx_xlock(&mqfs->mi_lock);
1450 	if (pn->mn_deleted) {
1451 		sx_xunlock(&mqfs->mi_lock);
1452 		return (ENOENT);
1453 	}
1454 
1455 	pt = LIST_FIRST(&pn->mn_children);
1456 	pt = LIST_NEXT(pt, mn_sibling);
1457 	pt = LIST_NEXT(pt, mn_sibling);
1458 	if (pt != NULL) {
1459 		sx_xunlock(&mqfs->mi_lock);
1460 		return (ENOTEMPTY);
1461 	}
1462 	pt = pn->mn_parent;
1463 	pn->mn_parent = NULL;
1464 	pn->mn_deleted = 1;
1465 	LIST_REMOVE(pn, mn_sibling);
1466 	mqnode_release(pn);
1467 	mqnode_release(pt);
1468 	sx_xunlock(&mqfs->mi_lock);
1469 	cache_purge(ap->a_vp);
1470 	return (0);
1471 }
1472 
1473 #endif /* notyet */
1474 
1475 /*
1476  * Allocate a message queue
1477  */
1478 static struct mqueue *
1479 mqueue_alloc(const struct mq_attr *attr)
1480 {
1481 	struct mqueue *mq;
1482 
1483 	if (curmq >= maxmq)
1484 		return (NULL);
1485 	mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
1486 	TAILQ_INIT(&mq->mq_msgq);
1487 	if (attr != NULL) {
1488 		mq->mq_maxmsg = attr->mq_maxmsg;
1489 		mq->mq_msgsize = attr->mq_msgsize;
1490 	} else {
1491 		mq->mq_maxmsg = default_maxmsg;
1492 		mq->mq_msgsize = default_msgsize;
1493 	}
1494 	mtx_init(&mq->mq_mutex, "mqueue", NULL, MTX_DEF);
1495 	knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1496 	knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL);
1497 	atomic_add_int(&curmq, 1);
1498 	return (mq);
1499 }
1500 
1501 /*
1502  * Destroy a message queue
1503  */
1504 static void
1505 mqueue_free(struct mqueue *mq)
1506 {
1507 	struct mqueue_msg *msg;
1508 
1509 	while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
1510 		TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
1511 		FREE(msg, M_MQUEUEDATA);
1512 	}
1513 
1514 	mtx_destroy(&mq->mq_mutex);
1515 	knlist_destroy(&mq->mq_rsel.si_note);
1516 	knlist_destroy(&mq->mq_wsel.si_note);
1517 	uma_zfree(mqueue_zone, mq);
1518 	atomic_add_int(&curmq, -1);
1519 }
1520 
1521 /*
1522  * Load a message from user space
1523  */
1524 static struct mqueue_msg *
1525 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
1526 {
1527 	struct mqueue_msg *msg;
1528 	size_t len;
1529 	int error;
1530 
1531 	len = sizeof(struct mqueue_msg) + msg_size;
1532 	MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK);
1533 	error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
1534 	    msg_size);
1535 	if (error) {
1536 		FREE(msg, M_MQUEUEDATA);
1537 		msg = NULL;
1538 	} else {
1539 		msg->msg_size = msg_size;
1540 		msg->msg_prio = msg_prio;
1541 	}
1542 	return (msg);
1543 }
1544 
1545 /*
1546  * Save a message to user space
1547  */
1548 static int
1549 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
1550 {
1551 	int error;
1552 
1553 	error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
1554 		msg->msg_size);
1555 	if (error == 0 && msg_prio != NULL)
1556 		error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
1557 	return (error);
1558 }
1559 
1560 /*
1561  * Free a message's memory
1562  */
1563 static __inline void
1564 mqueue_freemsg(struct mqueue_msg *msg)
1565 {
1566 	FREE(msg, M_MQUEUEDATA);
1567 }
1568 
1569 /*
1570  * Send a message. if waitok is false, thread will not be
1571  * blocked if there is no data in queue, otherwise, absolute
1572  * time will be checked.
1573  */
1574 int
1575 mqueue_send(struct mqueue *mq, const char *msg_ptr,
1576 	size_t msg_len, unsigned msg_prio, int waitok,
1577 	const struct timespec *abs_timeout)
1578 {
1579 	struct mqueue_msg *msg;
1580 	struct timespec ets, ts, ts2;
1581 	struct timeval tv;
1582 	int error;
1583 
1584 	if (msg_prio >= MQ_PRIO_MAX)
1585 		return (EINVAL);
1586 	if (msg_len > mq->mq_msgsize)
1587 		return (EMSGSIZE);
1588 	msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
1589 	if (msg == NULL)
1590 		return (EFAULT);
1591 
1592 	/* O_NONBLOCK case */
1593 	if (!waitok) {
1594 		error = _mqueue_send(mq, msg, -1);
1595 		if (error)
1596 			goto bad;
1597 		return (0);
1598 	}
1599 
1600 	/* we allow a null timeout (wait forever) */
1601 	if (abs_timeout == NULL) {
1602 		error = _mqueue_send(mq, msg, 0);
1603 		if (error)
1604 			goto bad;
1605 		return (0);
1606 	}
1607 
1608 	/* send it before checking time */
1609 	error = _mqueue_send(mq, msg, -1);
1610 	if (error == 0)
1611 		return (0);
1612 
1613 	if (error != EAGAIN)
1614 		goto bad;
1615 
1616 	error = copyin(abs_timeout, &ets, sizeof(ets));
1617 	if (error != 0)
1618 		goto bad;
1619 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1620 		error = EINVAL;
1621 		goto bad;
1622 	}
1623 	for (;;) {
1624 		ts2 = ets;
1625 		getnanotime(&ts);
1626 		timespecsub(&ts2, &ts);
1627 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1628 			error = ETIMEDOUT;
1629 			break;
1630 		}
1631 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1632 		error = _mqueue_send(mq, msg, tvtohz(&tv));
1633 		if (error != ETIMEDOUT)
1634 			break;
1635 	}
1636 	if (error == 0)
1637 		return (0);
1638 bad:
1639 	mqueue_freemsg(msg);
1640 	return (error);
1641 }
1642 
1643 /*
1644  * Common routine to send a message
1645  */
1646 static int
1647 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
1648 {
1649 	struct mqueue_msg *msg2;
1650 	int error = 0;
1651 
1652 	mtx_lock(&mq->mq_mutex);
1653 	while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
1654 		if (timo < 0) {
1655 			mtx_unlock(&mq->mq_mutex);
1656 			return (EAGAIN);
1657 		}
1658 		mq->mq_senders++;
1659 		error = msleep(&mq->mq_senders, &mq->mq_mutex,
1660 			    curthread->td_priority | PCATCH, "mqsend", timo);
1661 		mq->mq_senders--;
1662 		if (error == EAGAIN)
1663 			error = ETIMEDOUT;
1664 	}
1665 	if (mq->mq_curmsgs >= mq->mq_maxmsg) {
1666 		mtx_unlock(&mq->mq_mutex);
1667 		return (error);
1668 	}
1669 	error = 0;
1670 	if (TAILQ_EMPTY(&mq->mq_msgq)) {
1671 		TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
1672 	} else {
1673 		if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
1674 			TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
1675 		} else {
1676 			TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
1677 				if (msg2->msg_prio < msg->msg_prio)
1678 					break;
1679 			}
1680 			TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
1681 		}
1682 	}
1683 	mq->mq_curmsgs++;
1684 	mq->mq_totalbytes += msg->msg_size;
1685 	if (mq->mq_receivers)
1686 		wakeup_one(&mq->mq_receivers);
1687 	else if (mq->mq_notifier != NULL)
1688 		mqueue_send_notification(mq);
1689 	if (mq->mq_flags & MQ_RSEL) {
1690 		mq->mq_flags &= ~MQ_RSEL;
1691 		selwakeup(&mq->mq_rsel);
1692 	}
1693 	KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
1694 	mtx_unlock(&mq->mq_mutex);
1695 	return (0);
1696 }
1697 
1698 /*
1699  * Send realtime a signal to process which registered itself
1700  * successfully by mq_notify.
1701  */
1702 static void
1703 mqueue_send_notification(struct mqueue *mq)
1704 {
1705 	struct mqueue_notifier *nt;
1706 	struct proc *p;
1707 
1708 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1709 	nt = mq->mq_notifier;
1710 	if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
1711 		p = nt->nt_proc;
1712 		PROC_LOCK(p);
1713 		if (!KSI_ONQ(&nt->nt_ksi))
1714 			psignal_event(p, &nt->nt_sigev, &nt->nt_ksi);
1715 		PROC_UNLOCK(p);
1716 	}
1717 	mq->mq_notifier = NULL;
1718 }
1719 
1720 /*
1721  * Get a message. if waitok is false, thread will not be
1722  * blocked if there is no data in queue, otherwise, absolute
1723  * time will be checked.
1724  */
1725 int
1726 mqueue_receive(struct mqueue *mq, char *msg_ptr,
1727 	size_t msg_len, unsigned *msg_prio, int waitok,
1728 	const struct timespec *abs_timeout)
1729 {
1730 	struct mqueue_msg *msg;
1731 	struct timespec ets, ts, ts2;
1732 	struct timeval tv;
1733 	int error;
1734 
1735 	if (msg_len < mq->mq_msgsize)
1736 		return (EMSGSIZE);
1737 
1738 	/* O_NONBLOCK case */
1739 	if (!waitok) {
1740 		error = _mqueue_recv(mq, &msg, -1);
1741 		if (error)
1742 			return (error);
1743 		goto received;
1744 	}
1745 
1746 	/* we allow a null timeout (wait forever). */
1747 	if (abs_timeout == NULL) {
1748 		error = _mqueue_recv(mq, &msg, 0);
1749 		if (error)
1750 			return (error);
1751 		goto received;
1752 	}
1753 
1754 	/* try to get a message before checking time */
1755 	error = _mqueue_recv(mq, &msg, -1);
1756 	if (error == 0)
1757 		goto received;
1758 
1759 	if (error != EAGAIN)
1760 		return (error);
1761 
1762 	error = copyin(abs_timeout, &ets, sizeof(ets));
1763 	if (error != 0)
1764 		return (error);
1765 	if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) {
1766 		error = EINVAL;
1767 		return (error);
1768 	}
1769 
1770 	for (;;) {
1771 		ts2 = ets;
1772 		getnanotime(&ts);
1773 		timespecsub(&ts2, &ts);
1774 		if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
1775 			error = ETIMEDOUT;
1776 			return (error);
1777 		}
1778 		TIMESPEC_TO_TIMEVAL(&tv, &ts2);
1779 		error = _mqueue_recv(mq, &msg, tvtohz(&tv));
1780 		if (error == 0)
1781 			break;
1782 		if (error != ETIMEDOUT)
1783 			return (error);
1784 	}
1785 
1786 received:
1787 	error = mqueue_savemsg(msg, msg_ptr, msg_prio);
1788 	if (error == 0) {
1789 		curthread->td_retval[0] = msg->msg_size;
1790 		curthread->td_retval[1] = 0;
1791 	}
1792 	mqueue_freemsg(msg);
1793 	return (error);
1794 }
1795 
1796 /*
1797  * Common routine to receive a message
1798  */
1799 static int
1800 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
1801 {
1802 	int error = 0;
1803 
1804 	mtx_lock(&mq->mq_mutex);
1805 	while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
1806 		if (timo < 0) {
1807 			mtx_unlock(&mq->mq_mutex);
1808 			return (EAGAIN);
1809 		}
1810 		mq->mq_receivers++;
1811 		error = msleep(&mq->mq_receivers, &mq->mq_mutex,
1812 			    curthread->td_priority | PCATCH, "mqrecv", timo);
1813 		mq->mq_receivers--;
1814 		if (error == EAGAIN)
1815 			error = ETIMEDOUT;
1816 	}
1817 	if (*msg != NULL) {
1818 		error = 0;
1819 		TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
1820 		mq->mq_curmsgs--;
1821 		mq->mq_totalbytes -= (*msg)->msg_size;
1822 		if (mq->mq_senders)
1823 			wakeup_one(&mq->mq_senders);
1824 		if (mq->mq_flags & MQ_WSEL) {
1825 			mq->mq_flags &= ~MQ_WSEL;
1826 			selwakeup(&mq->mq_wsel);
1827 		}
1828 		KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
1829 	}
1830 	if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
1831 	    !TAILQ_EMPTY(&mq->mq_msgq)) {
1832 		mqueue_send_notification(mq);
1833 	}
1834 	mtx_unlock(&mq->mq_mutex);
1835 	return (error);
1836 }
1837 
1838 static __inline struct mqueue_notifier *
1839 notifier_alloc(void)
1840 {
1841 	return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
1842 }
1843 
1844 static __inline void
1845 notifier_free(struct mqueue_notifier *p)
1846 {
1847 	uma_zfree(mqnoti_zone, p);
1848 }
1849 
1850 static struct mqueue_notifier *
1851 notifier_search(struct proc *p, int fd)
1852 {
1853 	struct mqueue_notifier *nt;
1854 
1855 	LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
1856 		if (nt->nt_ksi.ksi_mqd == fd)
1857 			break;
1858 	}
1859 	return (nt);
1860 }
1861 
1862 static __inline void
1863 notifier_insert(struct proc *p, struct mqueue_notifier *nt)
1864 {
1865 	LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
1866 }
1867 
1868 static __inline void
1869 notifier_delete(struct proc *p, struct mqueue_notifier *nt)
1870 {
1871 	LIST_REMOVE(nt, nt_link);
1872 	notifier_free(nt);
1873 }
1874 
1875 static void
1876 notifier_remove(struct proc *p, struct mqueue *mq, int fd)
1877 {
1878 	struct mqueue_notifier *nt;
1879 
1880 	mtx_assert(&mq->mq_mutex, MA_OWNED);
1881 	PROC_LOCK(p);
1882 	nt = notifier_search(p, fd);
1883 	if (nt != NULL) {
1884 		if (mq->mq_notifier == nt)
1885 			mq->mq_notifier = NULL;
1886 		sigqueue_take(&nt->nt_ksi);
1887 		notifier_delete(p, nt);
1888 	}
1889 	PROC_UNLOCK(p);
1890 }
1891 
1892 /*
1893  * Syscall to open a message queue
1894  */
1895 int
1896 kmq_open(struct thread *td, struct kmq_open_args *uap)
1897 {
1898 	char path[MQFS_NAMELEN + 1];
1899 	struct mq_attr attr, *pattr;
1900 	struct mqfs_node *pn;
1901 	struct filedesc *fdp;
1902 	struct file *fp;
1903 	struct mqueue *mq;
1904 	int fd, error, len, flags, cmode;
1905 
1906 	if ((uap->flags & O_ACCMODE) == O_ACCMODE)
1907 		return (EINVAL);
1908 
1909 	fdp = td->td_proc->p_fd;
1910 	flags = FFLAGS(uap->flags);
1911 	cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
1912 	mq = NULL;
1913 	if ((flags & O_CREAT) && (uap->attr != NULL)) {
1914 		error = copyin(uap->attr, &attr, sizeof(attr));
1915 		if (error)
1916 			return (error);
1917 		if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg)
1918 			return (EINVAL);
1919 		if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize)
1920 			return (EINVAL);
1921 		pattr = &attr;
1922 	} else
1923 		pattr = NULL;
1924 
1925 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
1926         if (error)
1927 		return (error);
1928 
1929 	/*
1930 	 * The first character of name must be a slash  (/) character
1931 	 * and the remaining characters of name cannot include any slash
1932 	 * characters.
1933 	 */
1934 	len = strlen(path);
1935 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
1936 		return (EINVAL);
1937 
1938 	error = falloc(td, &fp, &fd);
1939 	if (error)
1940 		return (error);
1941 
1942 	sx_xlock(&mqfs_data.mi_lock);
1943 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
1944 	if (pn == NULL) {
1945 		if (!(flags & O_CREAT)) {
1946 			error = ENOENT;
1947 		} else {
1948 			mq = mqueue_alloc(pattr);
1949 			if (mq == NULL) {
1950 				error = ENFILE;
1951 			} else {
1952 				pn = mqfs_create_file(mqfs_data.mi_root,
1953 				         path + 1, len - 1, td->td_ucred,
1954 					 cmode);
1955 				if (pn == NULL) {
1956 					error = ENOSPC;
1957 					mqueue_free(mq);
1958 				}
1959 			}
1960 		}
1961 
1962 		if (error == 0) {
1963 			pn->mn_data = mq;
1964 		}
1965 	} else {
1966 		if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
1967 			error = EEXIST;
1968 		} else {
1969 			int acc_mode = 0;
1970 
1971 			if (flags & FREAD)
1972 				acc_mode |= VREAD;
1973 			if (flags & FWRITE)
1974 				acc_mode |= VWRITE;
1975 			error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
1976 				    pn->mn_gid, acc_mode, td->td_ucred, NULL);
1977 		}
1978 	}
1979 
1980 	if (error) {
1981 		sx_xunlock(&mqfs_data.mi_lock);
1982 		fdclose(fdp, fp, fd, td);
1983 		fdrop(fp, td);
1984 		return (error);
1985 	}
1986 
1987 	mqnode_addref(pn);
1988 	sx_xunlock(&mqfs_data.mi_lock);
1989 
1990 	FILE_LOCK(fp);
1991 	fp->f_flag = (flags & (FREAD | FWRITE | O_NONBLOCK));
1992 	fp->f_type = DTYPE_MQUEUE;
1993 	fp->f_ops = &mqueueops;
1994 	fp->f_data = pn;
1995 	FILE_UNLOCK(fp);
1996 
1997 	FILEDESC_LOCK_FAST(fdp);
1998 	if (fdp->fd_ofiles[fd] == fp)
1999 		fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
2000 	FILEDESC_UNLOCK_FAST(fdp);
2001 	td->td_retval[0] = fd;
2002 	fdrop(fp, td);
2003 	return (0);
2004 }
2005 
2006 /*
2007  * Syscall to unlink a message queue
2008  */
2009 int
2010 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
2011 {
2012 	char path[MQFS_NAMELEN+1];
2013 	struct mqfs_node *pn;
2014 	int error, len;
2015 
2016 	error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
2017         if (error)
2018 		return (error);
2019 
2020 	len = strlen(path);
2021 	if (len < 2  || path[0] != '/' || index(path + 1, '/') != NULL)
2022 		return (EINVAL);
2023 
2024 	sx_xlock(&mqfs_data.mi_lock);
2025 	pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1);
2026 	if (pn != NULL)
2027 		error = do_unlink(pn, td->td_ucred);
2028 	else
2029 		error = ENOENT;
2030 	sx_xunlock(&mqfs_data.mi_lock);
2031 	return (error);
2032 }
2033 
2034 typedef int (*_fgetf)(struct thread *, int, struct file **);
2035 
2036 /*
2037  * Get message queue by giving file slot
2038  */
2039 static int
2040 _getmq(struct thread *td, int fd, _fgetf func,
2041        struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
2042 {
2043 	struct mqfs_node *pn;
2044 	int error;
2045 
2046 	error = func(td, fd, fpp);
2047 	if (error)
2048 		return (error);
2049 	if (&mqueueops != (*fpp)->f_ops) {
2050 		fdrop(*fpp, td);
2051 		return (EBADF);
2052 	}
2053 	pn = (*fpp)->f_data;
2054 	if (ppn)
2055 		*ppn = pn;
2056 	if (pmq)
2057 		*pmq = pn->mn_data;
2058 	return (0);
2059 }
2060 
2061 static __inline int
2062 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
2063 	struct mqueue **pmq)
2064 {
2065 	return _getmq(td, fd, fget, fpp, ppn, pmq);
2066 }
2067 
2068 static __inline int
2069 getmq_read(struct thread *td, int fd, struct file **fpp,
2070 	 struct mqfs_node **ppn, struct mqueue **pmq)
2071 {
2072 	return _getmq(td, fd, fget_read, fpp, ppn, pmq);
2073 }
2074 
2075 static __inline int
2076 getmq_write(struct thread *td, int fd, struct file **fpp,
2077 	struct mqfs_node **ppn, struct mqueue **pmq)
2078 {
2079 	return _getmq(td, fd, fget_write, fpp, ppn, pmq);
2080 }
2081 
2082 /*
2083  * Syscall
2084  */
2085 int
2086 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
2087 {
2088 	struct mqueue *mq;
2089 	struct file *fp;
2090 	struct mq_attr attr, oattr;
2091 	int error;
2092 
2093 	if (uap->attr) {
2094 		error = copyin(uap->attr, &attr, sizeof(attr));
2095 		if (error)
2096 			return (error);
2097 		if (attr.mq_flags & ~O_NONBLOCK)
2098 			return (EINVAL);
2099 	}
2100 	error = getmq(td, uap->mqd, &fp, NULL, &mq);
2101 	if (error)
2102 		return (error);
2103 	oattr.mq_maxmsg  = mq->mq_maxmsg;
2104 	oattr.mq_msgsize = mq->mq_msgsize;
2105 	oattr.mq_curmsgs = mq->mq_curmsgs;
2106 	FILE_LOCK(fp);
2107 	oattr.mq_flags = (O_NONBLOCK & fp->f_flag);
2108 	if (uap->attr) {
2109 		fp->f_flag &= ~O_NONBLOCK;
2110 		fp->f_flag |= (attr.mq_flags & O_NONBLOCK);
2111 	}
2112 	FILE_UNLOCK(fp);
2113 	fdrop(fp, td);
2114 	if (uap->oattr)
2115 		error = copyout(&oattr, uap->oattr, sizeof(oattr));
2116 	return (error);
2117 }
2118 
2119 /*
2120  * Syscall
2121  */
2122 int
2123 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
2124 {
2125 	struct mqueue *mq;
2126 	struct file *fp;
2127 	int error;
2128 	int waitok;
2129 
2130 	error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
2131 	if (error)
2132 		return (error);
2133 	waitok = !(fp->f_flag & O_NONBLOCK);
2134 	error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
2135 		uap->msg_prio, waitok, uap->abs_timeout);
2136 	fdrop(fp, td);
2137 	return (error);
2138 }
2139 
2140 /*
2141  * Syscall
2142  */
2143 int
2144 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
2145 {
2146 	struct mqueue *mq;
2147 	struct file *fp;
2148 	int error, waitok;
2149 
2150 	error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
2151 	if (error)
2152 		return (error);
2153 	waitok = !(fp->f_flag & O_NONBLOCK);
2154 	error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
2155 		uap->msg_prio, waitok, uap->abs_timeout);
2156 	fdrop(fp, td);
2157 	return (error);
2158 }
2159 
2160 /*
2161  * Syscall
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_LOCK_FAST(fdp);
2194 	if (fget_locked(fdp, uap->mqd) != fp) {
2195 		FILEDESC_UNLOCK_FAST(fdp);
2196 		error = EBADF;
2197 		goto out;
2198 	}
2199 	mtx_lock(&mq->mq_mutex);
2200 	FILEDESC_UNLOCK_FAST(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, MA_OWNED);
2264 	if (fp->f_ops == &mqueueops) {
2265 		mq = FPTOMQ(fp);
2266 		mtx_lock(&mq->mq_mutex);
2267 		notifier_remove(td->td_proc, mq, fd);
2268 
2269 		/* have to wakeup thread in same process */
2270 		if (mq->mq_flags & MQ_RSEL) {
2271 			mq->mq_flags &= ~MQ_RSEL;
2272 			selwakeup(&mq->mq_rsel);
2273 		}
2274 		if (mq->mq_flags & MQ_WSEL) {
2275 			mq->mq_flags &= ~MQ_WSEL;
2276 			selwakeup(&mq->mq_wsel);
2277 		}
2278 		mtx_unlock(&mq->mq_mutex);
2279 	}
2280 }
2281 
2282 static void
2283 mq_proc_exit(void *arg __unused, struct proc *p)
2284 {
2285 	struct filedesc *fdp;
2286 	struct file *fp;
2287 	struct mqueue *mq;
2288 	int i;
2289 
2290 	fdp = p->p_fd;
2291 	FILEDESC_LOCK_FAST(fdp);
2292 	for (i = 0; i < fdp->fd_nfiles; ++i) {
2293 		fp = fget_locked(fdp, i);
2294 		if (fp != NULL && fp->f_ops == &mqueueops) {
2295 			mq = FPTOMQ(fp);
2296 			mtx_lock(&mq->mq_mutex);
2297 			notifier_remove(p, FPTOMQ(fp), i);
2298 			mtx_unlock(&mq->mq_mutex);
2299 		}
2300 	}
2301 	FILEDESC_UNLOCK_FAST(fdp);
2302 	KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
2303 }
2304 
2305 static int
2306 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
2307 	int flags, struct thread *td)
2308 {
2309 	return (EOPNOTSUPP);
2310 }
2311 
2312 static int
2313 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
2314 	int flags, struct thread *td)
2315 {
2316 	return (EOPNOTSUPP);
2317 }
2318 
2319 static int
2320 mqf_ioctl(struct file *fp, u_long cmd, void *data,
2321 	struct ucred *active_cred, struct thread *td)
2322 {
2323 	return (ENOTTY);
2324 }
2325 
2326 static int
2327 mqf_poll(struct file *fp, int events, struct ucred *active_cred,
2328 	struct thread *td)
2329 {
2330 	struct mqueue *mq = FPTOMQ(fp);
2331 	int revents = 0;
2332 
2333 	mtx_lock(&mq->mq_mutex);
2334 	if (events & (POLLIN | POLLRDNORM)) {
2335 		if (mq->mq_curmsgs) {
2336 			revents |= events & (POLLIN | POLLRDNORM);
2337 		} else {
2338 			mq->mq_flags |= MQ_RSEL;
2339 			selrecord(td, &mq->mq_rsel);
2340  		}
2341 	}
2342 	if (events & POLLOUT) {
2343 		if (mq->mq_curmsgs < mq->mq_maxmsg)
2344 			revents |= POLLOUT;
2345 		else {
2346 			mq->mq_flags |= MQ_WSEL;
2347 			selrecord(td, &mq->mq_wsel);
2348 		}
2349 	}
2350 	mtx_unlock(&mq->mq_mutex);
2351 	return (revents);
2352 }
2353 
2354 static int
2355 mqf_close(struct file *fp, struct thread *td)
2356 {
2357 	struct mqfs_node *pn;
2358 
2359 	FILE_LOCK(fp);
2360 	fp->f_ops = &badfileops;
2361 	FILE_UNLOCK(fp);
2362 	pn = fp->f_data;
2363 	fp->f_data = NULL;
2364 	sx_xlock(&mqfs_data.mi_lock);
2365 	mqnode_release(pn);
2366 	sx_xunlock(&mqfs_data.mi_lock);
2367 	return (0);
2368 }
2369 
2370 static int
2371 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
2372 	struct thread *td)
2373 {
2374 	struct mqfs_node *pn = fp->f_data;
2375 
2376 	bzero(st, sizeof *st);
2377 	st->st_atimespec = pn->mn_atime;
2378 	st->st_mtimespec = pn->mn_mtime;
2379 	st->st_ctimespec = pn->mn_ctime;
2380 	st->st_birthtimespec = pn->mn_birth;
2381 	st->st_uid = pn->mn_uid;
2382 	st->st_gid = pn->mn_gid;
2383 	st->st_mode = S_IFIFO | pn->mn_mode;
2384 	return (0);
2385 }
2386 
2387 static int
2388 mqf_kqfilter(struct file *fp, struct knote *kn)
2389 {
2390 	struct mqueue *mq = FPTOMQ(fp);
2391 	int error = 0;
2392 
2393 	if (kn->kn_filter == EVFILT_READ) {
2394 		kn->kn_fop = &mq_rfiltops;
2395 		knlist_add(&mq->mq_rsel.si_note, kn, 0);
2396 	} else if (kn->kn_filter == EVFILT_WRITE) {
2397 		kn->kn_fop = &mq_wfiltops;
2398 		knlist_add(&mq->mq_wsel.si_note, kn, 0);
2399 	} else
2400 		error = EINVAL;
2401 	return (error);
2402 }
2403 
2404 static void
2405 filt_mqdetach(struct knote *kn)
2406 {
2407 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2408 
2409 	if (kn->kn_filter == EVFILT_READ)
2410 		knlist_remove(&mq->mq_rsel.si_note, kn, 0);
2411 	else if (kn->kn_filter == EVFILT_WRITE)
2412 		knlist_remove(&mq->mq_wsel.si_note, kn, 0);
2413 	else
2414 		panic("filt_mqdetach");
2415 }
2416 
2417 static int
2418 filt_mqread(struct knote *kn, long hint)
2419 {
2420 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2421 
2422 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2423 	return (mq->mq_curmsgs != 0);
2424 }
2425 
2426 static int
2427 filt_mqwrite(struct knote *kn, long hint)
2428 {
2429 	struct mqueue *mq = FPTOMQ(kn->kn_fp);
2430 
2431 	mtx_assert(&mq->mq_mutex, MA_OWNED);
2432 	return (mq->mq_curmsgs < mq->mq_maxmsg);
2433 }
2434 
2435 static struct fileops mqueueops = {
2436 	.fo_read		= mqf_read,
2437 	.fo_write		= mqf_write,
2438 	.fo_ioctl		= mqf_ioctl,
2439 	.fo_poll		= mqf_poll,
2440 	.fo_kqfilter		= mqf_kqfilter,
2441 	.fo_stat		= mqf_stat,
2442 	.fo_close		= mqf_close
2443 };
2444 
2445 static struct vop_vector mqfs_vnodeops = {
2446 	.vop_default 		= &default_vnodeops,
2447 	.vop_access		= mqfs_access,
2448 	.vop_cachedlookup	= mqfs_lookup,
2449 	.vop_lookup		= vfs_cache_lookup,
2450 	.vop_reclaim		= mqfs_reclaim,
2451 	.vop_create		= mqfs_create,
2452 	.vop_remove		= mqfs_remove,
2453 	.vop_inactive		= mqfs_inactive,
2454 	.vop_open		= mqfs_open,
2455 	.vop_close		= mqfs_close,
2456 	.vop_getattr		= mqfs_getattr,
2457 	.vop_setattr		= mqfs_setattr,
2458 	.vop_read		= mqfs_read,
2459 	.vop_write		= VOP_EOPNOTSUPP,
2460 	.vop_readdir		= mqfs_readdir,
2461 	.vop_mkdir		= VOP_EOPNOTSUPP,
2462 	.vop_rmdir		= VOP_EOPNOTSUPP
2463 };
2464 
2465 static struct vfsops mqfs_vfsops = {
2466 	.vfs_init 		= mqfs_init,
2467 	.vfs_uninit		= mqfs_uninit,
2468 	.vfs_mount		= mqfs_mount,
2469 	.vfs_unmount		= mqfs_unmount,
2470 	.vfs_root		= mqfs_root,
2471 	.vfs_statfs		= mqfs_statfs,
2472 };
2473 
2474 SYSCALL_MODULE_HELPER(kmq_open);
2475 SYSCALL_MODULE_HELPER(kmq_setattr);
2476 SYSCALL_MODULE_HELPER(kmq_timedsend);
2477 SYSCALL_MODULE_HELPER(kmq_timedreceive);
2478 SYSCALL_MODULE_HELPER(kmq_notify);
2479 SYSCALL_MODULE_HELPER(kmq_unlink);
2480 
2481 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC);
2482 MODULE_VERSION(mqueuefs, 1);
2483