xref: /netbsd-src/sys/kern/sys_mqueue.c (revision 0df165c04d0a9ca1adde9ed2b890344c937954a6)
1 /*	$NetBSD: sys_mqueue.c,v 1.4 2007/11/11 23:22:24 matt Exp $	*/
2 
3 /*
4  * Copyright (c) 2007, Mindaugas Rasiukevicius <rmind at NetBSD org>
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
17  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
18  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
19  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  * POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*
29  * Implementation of POSIX message queues.
30  * Defined in the Base Definitions volume of IEEE Std 1003.1-2001.
31  *
32  * Locking
33  *  Global list of message queues and proc::p_mqueue_cnt counter are protected
34  *  by mqlist_mtx lock.  Concrete message queue and its members are protected
35  *  by mqueue::mq_mtx.
36  *
37  * Lock order:
38  * 	mqlist_mtx
39  * 	  -> mqueue::mq_mtx
40  *
41  * TODO:
42  *  - Hashing or RB-tree for the global list.
43  *  - Support for select(), poll() and perhaps kqueue().
44  */
45 
46 #include <sys/cdefs.h>
47 __KERNEL_RCSID(0, "$NetBSD: sys_mqueue.c,v 1.4 2007/11/11 23:22:24 matt Exp $");
48 
49 #include <sys/param.h>
50 #include <sys/types.h>
51 
52 #include <sys/condvar.h>
53 #include <sys/errno.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/filedesc.h>
57 #include <sys/kauth.h>
58 #include <sys/kernel.h>
59 #include <sys/kmem.h>
60 #include <sys/lwp.h>
61 #include <sys/mqueue.h>
62 #include <sys/mutex.h>
63 #include <sys/pool.h>
64 #include <sys/proc.h>
65 #include <sys/queue.h>
66 #include <sys/signal.h>
67 #include <sys/signalvar.h>
68 #include <sys/sysctl.h>
69 #include <sys/syscallargs.h>
70 #include <sys/systm.h>
71 #include <sys/unistd.h>
72 #include <sys/vnode.h>
73 
74 /* System-wide limits. */
75 static u_int			mq_open_max = MQ_OPEN_MAX;
76 static u_int			mq_prio_max = MQ_PRIO_MAX;
77 
78 static u_int			mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
79 static u_int			mq_def_maxmsg = 32;
80 
81 static kmutex_t			mqlist_mtx;
82 static struct pool		mqmsg_poll;
83 static LIST_HEAD(, mqueue)	mqueue_head =
84 	LIST_HEAD_INITIALIZER(mqueue_head);
85 
86 static int	mq_close_fop(struct file *, struct lwp *);
87 
88 static const struct fileops mqops = {
89 	fbadop_read, fbadop_write, fbadop_ioctl, fnullop_fcntl, fnullop_poll,
90 	fbadop_stat, mq_close_fop, fnullop_kqfilter
91 };
92 
93 /*
94  * Initialize POSIX message queue subsystem.
95  */
96 void
97 mqueue_sysinit(void)
98 {
99 
100 	pool_init(&mqmsg_poll, MQ_DEF_MSGSIZE, 0, 0, 0,
101 	    "mqmsg_poll", &pool_allocator_nointr, IPL_NONE);
102 	mutex_init(&mqlist_mtx, MUTEX_DEFAULT, IPL_NONE);
103 }
104 
105 /*
106  * Free the message.
107  */
108 static void
109 mqueue_freemsg(struct mq_msg *msg, const size_t size)
110 {
111 
112 	if (size > MQ_DEF_MSGSIZE)
113 		kmem_free(msg, size);
114 	else
115 		pool_put(&mqmsg_poll, msg);
116 }
117 
118 /*
119  * Destroy the message queue.
120  */
121 static void
122 mqueue_destroy(struct mqueue *mq)
123 {
124 	struct mq_msg *msg;
125 
126 	while ((msg = TAILQ_FIRST(&mq->mq_head)) != NULL) {
127 		TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
128 		mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
129 	}
130 	cv_destroy(&mq->mq_send_cv);
131 	cv_destroy(&mq->mq_recv_cv);
132 	mutex_destroy(&mq->mq_mtx);
133 	kmem_free(mq, sizeof(struct mqueue));
134 }
135 
136 /*
137  * Lookup for file name in general list of message queues.
138  *  => locks the message queue
139  */
140 static void *
141 mqueue_lookup(char *name)
142 {
143 	struct mqueue *mq;
144 	KASSERT(mutex_owned(&mqlist_mtx));
145 
146 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
147 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
148 			mutex_enter(&mq->mq_mtx);
149 			return mq;
150 		}
151 	}
152 
153 	return NULL;
154 }
155 
156 /*
157  * Check access against message queue.
158  */
159 static inline int
160 mqueue_access(struct lwp *l, struct mqueue *mq, mode_t acc_mode)
161 {
162 
163 	KASSERT(mutex_owned(&mq->mq_mtx));
164 	return vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid,
165 	    acc_mode, l->l_cred);
166 }
167 
168 /*
169  * Get the mqueue from the descriptor.
170  *  => locks the message queue
171  *  => increments the reference on file entry
172  */
173 static int
174 mqueue_get(struct lwp *l, mqd_t mqd, mode_t acc_mode, struct file **fpr)
175 {
176 	const struct proc *p = l->l_proc;
177 	struct file *fp;
178 	struct mqueue *mq;
179 
180 	/* Get the file and descriptor */
181 	fp = fd_getfile(p->p_fd, (int)mqd);
182 	if (fp == NULL)
183 		return EBADF;
184 
185 	/* Increment the reference of file entry, and lock the mqueue */
186 	FILE_USE(fp);
187 	mq = fp->f_data;
188 	*fpr = fp;
189 	mutex_enter(&mq->mq_mtx);
190 
191 	/* Check the access mode and permission if needed */
192 	if (acc_mode == VNOVAL)
193 		return 0;
194 	if ((acc_mode & fp->f_flag) == 0 || mqueue_access(l, mq, acc_mode)) {
195 		mutex_exit(&mq->mq_mtx);
196 		FILE_UNUSE(fp, l);
197 		return EPERM;
198 	}
199 
200 	return 0;
201 }
202 
203 /*
204  * Converter from struct timespec to the ticks.
205  * Used by mq_timedreceive(), mq_timedsend().
206  */
207 static int
208 abstimeout2timo(const void *uaddr, int *timo)
209 {
210 	struct timespec ts;
211 	int error;
212 
213 	error = copyin(uaddr, &ts, sizeof(struct timespec));
214 	if (error)
215 		return error;
216 
217 	/*
218 	 * According to POSIX, validation check is needed only in case of
219 	 * blocking.  Thus, set the invalid value right now, and fail latter.
220 	 */
221 	error = itimespecfix(&ts);
222 	*timo = (error == 0) ? tstohz(&ts) : -1;
223 
224 	return 0;
225 }
226 
227 static int
228 mq_close_fop(struct file *fp, struct lwp *l)
229 {
230 	struct proc *p = l->l_proc;
231 	struct mqueue *mq = fp->f_data;
232 	bool destroy;
233 
234 	mutex_enter(&mqlist_mtx);
235 	mutex_enter(&mq->mq_mtx);
236 
237 	/* Decrease the counters */
238 	p->p_mqueue_cnt--;
239 	mq->mq_refcnt--;
240 
241 	/* Remove notification if registered for this process */
242 	if (mq->mq_notify_proc == p)
243 		mq->mq_notify_proc = NULL;
244 
245 	/*
246 	 * If this is the last reference and mqueue is marked for unlink,
247 	 * remove and later destroy the message queue.
248 	 */
249 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
250 		LIST_REMOVE(mq, mq_list);
251 		destroy = true;
252 	} else
253 		destroy = false;
254 
255 	mutex_exit(&mq->mq_mtx);
256 	mutex_exit(&mqlist_mtx);
257 
258 	if (destroy)
259 		mqueue_destroy(mq);
260 
261 	return 0;
262 }
263 
264 /*
265  * General mqueue system calls.
266  */
267 
268 int
269 sys_mq_open(struct lwp *l, void *v, register_t *retval)
270 {
271 	struct sys_mq_open_args /* {
272 		syscallarg(const char *) name;
273 		syscallarg(int) oflag;
274 		syscallarg(mode_t) mode;
275 		syscallarg(struct mq_attr) attr;
276 	} */ *uap = v;
277 	struct proc *p = l->l_proc;
278 	struct mqueue *mq, *mq_new = NULL;
279 	struct file *fp;
280 	char *name;
281 	int mqd, error, oflag;
282 
283 	/* Check access mode flags */
284 	oflag = SCARG(uap, oflag);
285 	if ((oflag & (O_RDWR | O_RDONLY | O_WRONLY)) == 0)
286 		return EINVAL;
287 
288 	/* Get the name from the user-space */
289 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
290 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
291 	if (error) {
292 		kmem_free(name, MQ_NAMELEN);
293 		return error;
294 	}
295 
296 	if (oflag & O_CREAT) {
297 		struct mq_attr attr;
298 
299 		/* Check the limit */
300 		if (p->p_mqueue_cnt == mq_open_max) {
301 			kmem_free(name, MQ_NAMELEN);
302 			return EMFILE;
303 		}
304 
305 		/* Check for mqueue attributes */
306 		if (SCARG(uap, attr)) {
307 			error = copyin(SCARG(uap, attr), &attr,
308 				sizeof(struct mq_attr));
309 			if (error) {
310 				kmem_free(name, MQ_NAMELEN);
311 				return error;
312 			}
313 			if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 ||
314 			    attr.mq_msgsize > mq_max_msgsize) {
315 				kmem_free(name, MQ_NAMELEN);
316 				return EINVAL;
317 			}
318 			attr.mq_curmsgs = 0;
319 		} else {
320 			memset(&attr, 0, sizeof(struct mq_attr));
321 			attr.mq_maxmsg = mq_def_maxmsg;
322 			attr.mq_msgsize =
323 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
324 		}
325 
326 		/*
327 		 * Allocate new mqueue, initialize data structures,
328 		 * copy the name, attributes and set the flag.
329 		 */
330 		mq_new = kmem_zalloc(sizeof(struct mqueue), KM_SLEEP);
331 
332 		mutex_init(&mq_new->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
333 		cv_init(&mq_new->mq_send_cv, "mqsendcv");
334 		cv_init(&mq_new->mq_recv_cv, "mqrecvcv");
335 		TAILQ_INIT(&mq_new->mq_head);
336 
337 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
338 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
339 		mq_new->mq_attrib.mq_flags = oflag;
340 
341 		/* Store mode and effective UID with GID */
342 		mq_new->mq_mode = SCARG(uap, mode);
343 		mq_new->mq_euid = kauth_cred_geteuid(l->l_cred);
344 		mq_new->mq_egid = kauth_cred_getegid(l->l_cred);
345 	}
346 
347 	/* Allocate file structure and descriptor */
348 	error = falloc(l, &fp, &mqd);
349 	if (error) {
350 		if (mq_new)
351 			mqueue_destroy(mq_new);
352 		kmem_free(name, MQ_NAMELEN);
353 		return error;
354 	}
355 	fp->f_type = DTYPE_MQUEUE;
356 	fp->f_flag = (oflag & O_RDWR) ? (VREAD | VWRITE) :
357 	    ((oflag & O_RDONLY) ? VREAD : VWRITE);
358 	fp->f_ops = &mqops;
359 
360 	/* Look up for mqueue with such name */
361 	mutex_enter(&mqlist_mtx);
362 	mq = mqueue_lookup(name);
363 	if (mq) {
364 		KASSERT(mutex_owned(&mq->mq_mtx));
365 		/* Check if mqueue is not marked as unlinking */
366 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
367 			error = EACCES;
368 			goto exit;
369 		}
370 		/* Fail if O_EXCL is set, and mqueue already exists */
371 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
372 			error = EEXIST;
373 			goto exit;
374 		}
375 		/* Check the permission */
376 		if (mqueue_access(l, mq, fp->f_flag)) {
377 			error = EACCES;
378 			goto exit;
379 		}
380 	} else {
381 		/* Fail if mqueue neither exists, nor we create it */
382 		if ((oflag & O_CREAT) == 0) {
383 			mutex_exit(&mqlist_mtx);
384 			KASSERT(mq_new == NULL);
385 			FILE_UNUSE(fp, l);
386 			ffree(fp);
387 			fdremove(p->p_fd, mqd);
388 			kmem_free(name, MQ_NAMELEN);
389 			return ENOENT;
390 		}
391 
392 		/* Check the limit */
393 		if (p->p_mqueue_cnt == mq_open_max) {
394 			error = EMFILE;
395 			goto exit;
396 		}
397 
398 		/* Insert the queue to the list */
399 		mq = mq_new;
400 		mutex_enter(&mq->mq_mtx);
401 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
402 		mq_new = NULL;
403 	}
404 
405 	/* Increase the counters, and make descriptor ready */
406 	p->p_mqueue_cnt++;
407 	mq->mq_refcnt++;
408 	fp->f_data = mq;
409 	FILE_SET_MATURE(fp);
410 exit:
411 	mutex_exit(&mq->mq_mtx);
412 	mutex_exit(&mqlist_mtx);
413 	FILE_UNUSE(fp, l);
414 
415 	if (mq_new)
416 		mqueue_destroy(mq_new);
417 	if (error) {
418 		ffree(fp);
419 		fdremove(p->p_fd, mqd);
420 	} else
421 		*retval = mqd;
422 	kmem_free(name, MQ_NAMELEN);
423 
424 	return error;
425 }
426 
427 int
428 sys_mq_close(struct lwp *l, void *v, register_t *retval)
429 {
430 
431 	return sys_close(l, v, retval);
432 }
433 
434 /*
435  * Primary mq_receive1() function.
436  */
437 static int
438 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len,
439     unsigned *msg_prio, int t, ssize_t *mlen)
440 {
441 	struct file *fp = NULL;
442 	struct mqueue *mq;
443 	struct mq_msg *msg = NULL;
444 	int error;
445 
446 	/* Get the message queue */
447 	error = mqueue_get(l, mqdes, VREAD, &fp);
448 	if (error)
449 		return error;
450 	mq = fp->f_data;
451 
452 	/* Check the message size limits */
453 	if (msg_len < mq->mq_attrib.mq_msgsize) {
454 		error = EMSGSIZE;
455 		goto error;
456 	}
457 
458 	/* Check if queue is empty */
459 	while (TAILQ_EMPTY(&mq->mq_head)) {
460 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
461 			error = EAGAIN;
462 			goto error;
463 		}
464 		if (t < 0) {
465 			error = EINVAL;
466 			goto error;
467 		}
468 		/*
469 		 * Block until someone sends the message.
470 		 * While doing this, notification should not be sent.
471 		 */
472 		mq->mq_attrib.mq_flags |= MQ_RECEIVE;
473 		error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
474 		mq->mq_attrib.mq_flags &= ~MQ_RECEIVE;
475 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
476 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
477 			goto error;
478 		}
479 	}
480 
481 	/* Remove the message from the queue */
482 	msg = TAILQ_FIRST(&mq->mq_head);
483 	KASSERT(msg != NULL);
484 	TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
485 
486 	/* Decrement the counter and signal waiter, if any */
487 	mq->mq_attrib.mq_curmsgs--;
488 	cv_signal(&mq->mq_recv_cv);
489 error:
490 	mutex_exit(&mq->mq_mtx);
491 	FILE_UNUSE(fp, l);
492 	if (error)
493 		return error;
494 
495 	/*
496 	 * Copy the data to the user-space.
497 	 * Note: According to POSIX, no message should be removed from the
498 	 * queue in case of fail - this would be violated.
499 	 */
500 	*mlen = msg->msg_len;
501 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
502 	if (error == 0 && msg_prio)
503 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
504 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
505 
506 	return error;
507 }
508 
509 int
510 sys_mq_receive(struct lwp *l, void *v, register_t *retval)
511 {
512 	struct sys_mq_receive_args /* {
513 		syscallarg(mqd_t) mqdes;
514 		syscallarg(char *) msg_ptr;
515 		syscallarg(size_t) msg_len;
516 		syscallarg(unsigned *) msg_prio;
517 	} */ *uap = v;
518 	int error;
519 	ssize_t mlen;
520 
521 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
522 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen);
523 	if (error == 0)
524 		*retval = mlen;
525 
526 	return error;
527 }
528 
529 int
530 sys_mq_timedreceive(struct lwp *l, void *v, register_t *retval)
531 {
532 	struct sys_mq_timedreceive_args /* {
533 		syscallarg(mqd_t) mqdes;
534 		syscallarg(char *) msg_ptr;
535 		syscallarg(size_t) msg_len;
536 		syscallarg(unsigned *) msg_prio;
537 		syscallarg(const struct timespec *) abs_timeout;
538 	} */ *uap = v;
539 	int error, t;
540 	ssize_t mlen;
541 
542 	/* Get and convert time value */
543 	if (SCARG(uap, abs_timeout)) {
544 		error = abstimeout2timo(SCARG(uap, abs_timeout), &t);
545 		if (error)
546 			return error;
547 	} else
548 		t = 0;
549 
550 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
551 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t, &mlen);
552 	if (error == 0)
553 		*retval = mlen;
554 
555 	return error;
556 }
557 
558 /*
559  * Primary mq_send1() function.
560  */
561 static int
562 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len,
563     unsigned msg_prio, int t)
564 {
565 	struct file *fp = NULL;
566 	struct mqueue *mq;
567 	struct mq_msg *msg, *pos_msg;
568 	struct proc *notify = NULL;
569 	ksiginfo_t ksi;
570 	size_t size;
571 	int error;
572 
573 	/* Check the priority range */
574 	if (msg_prio >= mq_prio_max)
575 		return EINVAL;
576 
577 	/* Allocate a new message */
578 	size = sizeof(struct mq_msg) + msg_len;
579 	if (size > mq_max_msgsize)
580 		return EMSGSIZE;
581 
582 	if (size > MQ_DEF_MSGSIZE)
583 		msg = kmem_alloc(size, KM_SLEEP);
584 	else
585 		msg = pool_get(&mqmsg_poll, PR_WAITOK);
586 
587 	/* Get the data from user-space */
588 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
589 	if (error) {
590 		mqueue_freemsg(msg, size);
591 		return error;
592 	}
593 	msg->msg_len = msg_len;
594 	msg->msg_prio = msg_prio;
595 
596 	/* Get the mqueue */
597 	error = mqueue_get(l, mqdes, VWRITE, &fp);
598 	if (error) {
599 		mqueue_freemsg(msg, size);
600 		return error;
601 	}
602 	mq = fp->f_data;
603 
604 	/* Check the message size limit */
605 	if (msg_len <= 0 || msg_len > mq->mq_attrib.mq_msgsize) {
606 		error = EMSGSIZE;
607 		goto error;
608 	}
609 
610 	/* Check if queue is full */
611 	while (mq->mq_attrib.mq_curmsgs >= mq->mq_attrib.mq_maxmsg) {
612 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
613 			error = EAGAIN;
614 			goto error;
615 		}
616 		if (t < 0) {
617 			error = EINVAL;
618 			goto error;
619 		}
620 		/* Block until queue becomes available */
621 		error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
622 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
623 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
624 			goto error;
625 		}
626 	}
627 	KASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg);
628 
629 	/* Insert message into the queue, according to the priority */
630 	TAILQ_FOREACH(pos_msg, &mq->mq_head, msg_queue)
631 		if (msg->msg_prio > pos_msg->msg_prio)
632 			break;
633 	if (pos_msg == NULL)
634 		TAILQ_INSERT_TAIL(&mq->mq_head, msg, msg_queue);
635 	else
636 		TAILQ_INSERT_BEFORE(pos_msg, msg, msg_queue);
637 
638 	/* Check for the notify */
639 	if (mq->mq_attrib.mq_curmsgs == 0 && mq->mq_notify_proc &&
640 	    (mq->mq_attrib.mq_flags & MQ_RECEIVE) == 0) {
641 		/* Initialize the signal */
642 		KSI_INIT(&ksi);
643 		ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
644 		ksi.ksi_code = SI_MESGQ;
645 		ksi.ksi_value = mq->mq_sig_notify.sigev_value;
646 		/* Unregister the process */
647 		notify = mq->mq_notify_proc;
648 		mq->mq_notify_proc = NULL;
649 	}
650 
651 	/* Increment the counter and signal waiter, if any */
652 	mq->mq_attrib.mq_curmsgs++;
653 	cv_signal(&mq->mq_send_cv);
654 error:
655 	mutex_exit(&mq->mq_mtx);
656 	FILE_UNUSE(fp, l);
657 
658 	if (error) {
659 		mqueue_freemsg(msg, size);
660 	} else if (notify) {
661 		/* Send the notify, if needed */
662 		mutex_enter(&proclist_mutex);
663 		kpsignal(notify, &ksi, NULL);
664 		mutex_exit(&proclist_mutex);
665 	}
666 
667 	return error;
668 }
669 
670 int
671 sys_mq_send(struct lwp *l, void *v, register_t *retval)
672 {
673 	struct sys_mq_send_args /* {
674 		syscallarg(mqd_t) mqdes;
675 		syscallarg(const char *) msg_ptr;
676 		syscallarg(size_t) msg_len;
677 		syscallarg(unsigned) msg_prio;
678 	} */ *uap = v;
679 
680 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
681 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0);
682 }
683 
684 int
685 sys_mq_timedsend(struct lwp *l, void *v, register_t *retval)
686 {
687 	struct sys_mq_timedsend_args /* {
688 		syscallarg(mqd_t) mqdes;
689 		syscallarg(const char *) msg_ptr;
690 		syscallarg(size_t) msg_len;
691 		syscallarg(unsigned) msg_prio;
692 		syscallarg(const struct timespec *) abs_timeout;
693 	} */ *uap = v;
694 	int t;
695 
696 	/* Get and convert time value */
697 	if (SCARG(uap, abs_timeout)) {
698 		int error = abstimeout2timo(SCARG(uap, abs_timeout), &t);
699 		if (error)
700 			return error;
701 	} else
702 		t = 0;
703 
704 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
705 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t);
706 }
707 
708 int
709 sys_mq_notify(struct lwp *l, void *v, register_t *retval)
710 {
711 	struct sys_mq_notify_args /* {
712 		syscallarg(mqd_t) mqdes;
713 		syscallarg(const struct sigevent *) notification;
714 	} */ *uap = v;
715 	struct file *fp = NULL;
716 	struct mqueue *mq;
717 	struct sigevent sig;
718 	int error;
719 
720 	if (SCARG(uap, notification)) {
721 		/* Get the signal from user-space */
722 		error = copyin(SCARG(uap, notification), &sig,
723 		    sizeof(struct sigevent));
724 		if (error)
725 			return error;
726 	}
727 
728 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
729 	if (error)
730 		return error;
731 	mq = fp->f_data;
732 
733 	if (SCARG(uap, notification)) {
734 		/* Register notification: set the signal and target process */
735 		if (mq->mq_notify_proc == NULL) {
736 			memcpy(&mq->mq_sig_notify, &sig,
737 			    sizeof(struct sigevent));
738 			mq->mq_notify_proc = l->l_proc;
739 		} else {
740 			/* Fail if someone else already registered */
741 			error = EBUSY;
742 		}
743 	} else {
744 		/* Unregister the notification */
745 		mq->mq_notify_proc = NULL;
746 	}
747 	mutex_exit(&mq->mq_mtx);
748 	FILE_UNUSE(fp, l);
749 
750 	return error;
751 }
752 
753 int
754 sys_mq_getattr(struct lwp *l, void *v, register_t *retval)
755 {
756 	struct sys_mq_getattr_args /* {
757 		syscallarg(mqd_t) mqdes;
758 		syscallarg(struct mq_attr *) mqstat;
759 	} */ *uap = v;
760 	struct file *fp = NULL;
761 	struct mqueue *mq;
762 	struct mq_attr attr;
763 	int error;
764 
765 	/* Get the message queue */
766 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
767 	if (error)
768 		return error;
769 	mq = fp->f_data;
770 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
771 	mutex_exit(&mq->mq_mtx);
772 	FILE_UNUSE(fp, l);
773 
774 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
775 }
776 
777 int
778 sys_mq_setattr(struct lwp *l, void *v, register_t *retval)
779 {
780 	struct sys_mq_setattr_args /* {
781 		syscallarg(mqd_t) mqdes;
782 		syscallarg(const struct mq_attr *) mqstat;
783 		syscallarg(struct mq_attr *) omqstat;
784 	} */ *uap = v;
785 	struct file *fp = NULL;
786 	struct mqueue *mq;
787 	struct mq_attr attr;
788 	int error, nonblock;
789 
790 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
791 	if (error)
792 		return error;
793 	nonblock = (attr.mq_flags & O_NONBLOCK);
794 
795 	/* Get the message queue */
796 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
797 	if (error)
798 		return error;
799 	mq = fp->f_data;
800 
801 	/* Copy the old attributes, if needed */
802 	if (SCARG(uap, omqstat))
803 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
804 
805 	/* Ignore everything, except O_NONBLOCK */
806 	if (nonblock)
807 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
808 	else
809 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
810 
811 	mutex_exit(&mq->mq_mtx);
812 	FILE_UNUSE(fp, l);
813 
814 	/*
815 	 * Copy the data to the user-space.
816 	 * Note: According to POSIX, the new attributes should not be set in
817 	 * case of fail - this would be violated.
818 	 */
819 	if (SCARG(uap, omqstat))
820 		error = copyout(&attr, SCARG(uap, omqstat),
821 		    sizeof(struct mq_attr));
822 
823 	return error;
824 }
825 
826 int
827 sys_mq_unlink(struct lwp *l, void *v, register_t *retval)
828 {
829 	struct sys_mq_unlink_args /* {
830 		syscallarg(const char *) name;
831 	} */ *uap = v;
832 	struct mqueue *mq;
833 	char *name;
834 	int error, refcnt = 0;
835 
836 	/* Get the name from the user-space */
837 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
838 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
839 	if (error) {
840 		kmem_free(name, MQ_NAMELEN);
841 		return error;
842 	}
843 
844 	/* Lookup for this file */
845 	mutex_enter(&mqlist_mtx);
846 	mq = mqueue_lookup(name);
847 	if (mq == NULL) {
848 		error = ENOENT;
849 		goto error;
850 	}
851 
852 	/* Check the permissions */
853 	if (mqueue_access(l, mq, VWRITE)) {
854 		mutex_exit(&mq->mq_mtx);
855 		error = EACCES;
856 		goto error;
857 	}
858 
859 	/* Mark message queue as unlinking, before leaving the window */
860 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
861 
862 	/* Wake up all waiters, if there are such */
863 	cv_broadcast(&mq->mq_send_cv);
864 	cv_broadcast(&mq->mq_recv_cv);
865 
866 	refcnt = mq->mq_refcnt;
867 	if (refcnt == 0)
868 		LIST_REMOVE(mq, mq_list);
869 
870 	mutex_exit(&mq->mq_mtx);
871 error:
872 	mutex_exit(&mqlist_mtx);
873 
874 	/*
875 	 * If there are no references - destroy the message
876 	 * queue, otherwise, the last mq_close() will do that.
877 	 */
878 	if (error == 0 && refcnt == 0)
879 		mqueue_destroy(mq);
880 
881 	kmem_free(name, MQ_NAMELEN);
882 	return error;
883 }
884 
885 /*
886  * SysCtl.
887  */
888 
889 SYSCTL_SETUP(sysctl_mqueue_setup, "sysctl mqueue setup")
890 {
891 	const struct sysctlnode *node = NULL;
892 
893 	sysctl_createv(clog, 0, NULL, NULL,
894 		CTLFLAG_PERMANENT,
895 		CTLTYPE_NODE, "kern", NULL,
896 		NULL, 0, NULL, 0,
897 		CTL_KERN, CTL_EOL);
898 	sysctl_createv(clog, 0, NULL, NULL,
899 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
900 		CTLTYPE_INT, "posix_msg",
901 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
902 			     "Message Passing option to which the "
903 			     "system attempts to conform"),
904 		NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
905 		CTL_KERN, CTL_CREATE, CTL_EOL);
906 	sysctl_createv(clog, 0, NULL, &node,
907 		CTLFLAG_PERMANENT,
908 		CTLTYPE_NODE, "mqueue",
909 		SYSCTL_DESCR("Message queue options"),
910 		NULL, 0, NULL, 0,
911 		CTL_KERN, CTL_CREATE, CTL_EOL);
912 
913 	if (node == NULL)
914 		return;
915 
916 	sysctl_createv(clog, 0, &node, NULL,
917 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
918 		CTLTYPE_INT, "mq_open_max",
919 		SYSCTL_DESCR("Maximal number of message queue descriptors "
920 			     "that process could open"),
921 		NULL, 0, &mq_open_max, 0,
922 		CTL_CREATE, CTL_EOL);
923 	sysctl_createv(clog, 0, &node, NULL,
924 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
925 		CTLTYPE_INT, "mq_prio_max",
926 		SYSCTL_DESCR("Maximal priority of the message"),
927 		NULL, 0, &mq_prio_max, 0,
928 		CTL_CREATE, CTL_EOL);
929 	sysctl_createv(clog, 0, &node, NULL,
930 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
931 		CTLTYPE_INT, "mq_max_msgsize",
932 		SYSCTL_DESCR("Maximal allowed size of the message"),
933 		NULL, 0, &mq_max_msgsize, 0,
934 		CTL_CREATE, CTL_EOL);
935 	sysctl_createv(clog, 0, &node, NULL,
936 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
937 		CTLTYPE_INT, "mq_def_maxmsg",
938 		SYSCTL_DESCR("Default maximal message count"),
939 		NULL, 0, &mq_def_maxmsg, 0,
940 		CTL_CREATE, CTL_EOL);
941 }
942 
943 /*
944  * Debugging.
945  */
946 #if defined(DDB)
947 
948 void
949 mqueue_print_list(void (*pr)(const char *, ...))
950 {
951 	struct mqueue *mq;
952 
953 	(*pr)("Global list of the message queues:\n");
954 	(*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
955 	    "Name", "Ptr", "Mode", "Flags",  "Ref",
956 	    "MaxMsg", "MsgSze", "CurMsg");
957 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
958 		(*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
959 		    mq->mq_name, mq, mq->mq_mode,
960 		    mq->mq_attrib.mq_flags, mq->mq_refcnt,
961 		    mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
962 		    mq->mq_attrib.mq_curmsgs);
963 	}
964 }
965 
966 #endif /* defined(DDB) */
967