xref: /netbsd-src/sys/kern/sys_mqueue.c (revision c8a35b6227034951e874c2def577388e79ede4a5)
1 /*	$NetBSD: sys_mqueue.c,v 1.13 2009/01/11 02:45:52 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 2007, 2008 Mindaugas Rasiukevicius <rmind at NetBSD org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * Implementation of POSIX message queues.
31  * Defined in the Base Definitions volume of IEEE Std 1003.1-2001.
32  *
33  * Locking
34  *
35  * Global list of message queues (mqueue_head) and proc_t::p_mqueue_cnt
36  * counter are protected by mqlist_mtx lock.  The very message queue and
37  * its members are protected by mqueue::mq_mtx.
38  *
39  * Lock order:
40  * 	mqlist_mtx
41  * 	  -> mqueue::mq_mtx
42  */
43 
44 #include <sys/cdefs.h>
45 __KERNEL_RCSID(0, "$NetBSD: sys_mqueue.c,v 1.13 2009/01/11 02:45:52 christos Exp $");
46 
47 #include <sys/param.h>
48 #include <sys/types.h>
49 #include <sys/condvar.h>
50 #include <sys/errno.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/kauth.h>
55 #include <sys/kernel.h>
56 #include <sys/kmem.h>
57 #include <sys/lwp.h>
58 #include <sys/mqueue.h>
59 #include <sys/mutex.h>
60 #include <sys/pool.h>
61 #include <sys/poll.h>
62 #include <sys/proc.h>
63 #include <sys/queue.h>
64 #include <sys/select.h>
65 #include <sys/signal.h>
66 #include <sys/signalvar.h>
67 #include <sys/stat.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 pool_cache_t		mqmsg_cache;
83 static LIST_HEAD(, mqueue)	mqueue_head =
84 	LIST_HEAD_INITIALIZER(mqueue_head);
85 
86 static int	mq_poll_fop(file_t *, int);
87 static int	mq_close_fop(file_t *);
88 
89 #define	FNOVAL	-1
90 
91 static const struct fileops mqops = {
92 	fbadop_read, fbadop_write, fbadop_ioctl, fnullop_fcntl, mq_poll_fop,
93 	fbadop_stat, mq_close_fop, fnullop_kqfilter
94 };
95 
96 /*
97  * Initialize POSIX message queue subsystem.
98  */
99 void
100 mqueue_sysinit(void)
101 {
102 
103 	mqmsg_cache = pool_cache_init(MQ_DEF_MSGSIZE, coherency_unit,
104 	    0, 0, "mqmsgpl", NULL, IPL_NONE, NULL, NULL, NULL);
105 	mutex_init(&mqlist_mtx, MUTEX_DEFAULT, IPL_NONE);
106 }
107 
108 /*
109  * Free the message.
110  */
111 static void
112 mqueue_freemsg(struct mq_msg *msg, const size_t size)
113 {
114 
115 	if (size > MQ_DEF_MSGSIZE)
116 		kmem_free(msg, size);
117 	else
118 		pool_cache_put(mqmsg_cache, msg);
119 }
120 
121 /*
122  * Destroy the message queue.
123  */
124 static void
125 mqueue_destroy(struct mqueue *mq)
126 {
127 	struct mq_msg *msg;
128 
129 	while ((msg = TAILQ_FIRST(&mq->mq_head)) != NULL) {
130 		TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
131 		mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
132 	}
133 	seldestroy(&mq->mq_rsel);
134 	seldestroy(&mq->mq_wsel);
135 	cv_destroy(&mq->mq_send_cv);
136 	cv_destroy(&mq->mq_recv_cv);
137 	mutex_destroy(&mq->mq_mtx);
138 	kmem_free(mq, sizeof(struct mqueue));
139 }
140 
141 /*
142  * Lookup for file name in general list of message queues.
143  *  => locks the message queue
144  */
145 static void *
146 mqueue_lookup(char *name)
147 {
148 	struct mqueue *mq;
149 	KASSERT(mutex_owned(&mqlist_mtx));
150 
151 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
152 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
153 			mutex_enter(&mq->mq_mtx);
154 			return mq;
155 		}
156 	}
157 
158 	return NULL;
159 }
160 
161 /*
162  * Check access against message queue.
163  */
164 static inline int
165 mqueue_access(struct lwp *l, struct mqueue *mq, int access)
166 {
167 	mode_t acc_mode = 0;
168 
169 	KASSERT(mutex_owned(&mq->mq_mtx));
170 	KASSERT(access != FNOVAL);
171 
172 	/* Note the difference between VREAD/VWRITE and FREAD/FWRITE */
173 	if (access & FREAD)
174 		acc_mode |= VREAD;
175 	if (access & FWRITE)
176 		acc_mode |= VWRITE;
177 
178 	return vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid,
179 	    acc_mode, l->l_cred);
180 }
181 
182 /*
183  * Get the mqueue from the descriptor.
184  *  => locks the message queue, if found
185  *  => increments the reference on file entry
186  */
187 static int
188 mqueue_get(struct lwp *l, mqd_t mqd, int access, file_t **fpr)
189 {
190 	file_t *fp;
191 	struct mqueue *mq;
192 
193 	/* Get the file and descriptor */
194 	fp = fd_getfile((int)mqd);
195 	if (fp == NULL)
196 		return EBADF;
197 
198 	/* Increment the reference of file entry, and lock the mqueue */
199 	mq = fp->f_data;
200 	*fpr = fp;
201 	mutex_enter(&mq->mq_mtx);
202 	if (access == FNOVAL) {
203 		KASSERT(mutex_owned(&mq->mq_mtx));
204 		return 0;
205 	}
206 
207 	/* Check the access mode and permission */
208 	if ((fp->f_flag & access) != access || mqueue_access(l, mq, access)) {
209 		mutex_exit(&mq->mq_mtx);
210 		fd_putfile((int)mqd);
211 		return EPERM;
212 	}
213 	return 0;
214 }
215 
216 /*
217  * Converter from struct timespec to the ticks.
218  * Used by mq_timedreceive(), mq_timedsend().
219  */
220 int
221 abstimeout2timo(struct timespec *ts, int *timo)
222 {
223 	int error;
224 
225 	/*
226 	 * According to POSIX, validation check is needed only in case of
227 	 * blocking.  Thus, set the invalid value right now, and fail latter.
228 	 */
229 	error = itimespecfix(ts);
230 	*timo = (error == 0) ? tstohz(ts) : -1;
231 
232 	return 0;
233 }
234 
235 static int
236 mq_poll_fop(file_t *fp, int events)
237 {
238 	struct mqueue *mq = fp->f_data;
239 	int revents = 0;
240 
241 	mutex_enter(&mq->mq_mtx);
242 	if (events & (POLLIN | POLLRDNORM)) {
243 		/* Ready for receiving, if there are messages in the queue */
244 		if (mq->mq_attrib.mq_curmsgs)
245 			revents |= (POLLIN | POLLRDNORM);
246 		else
247 			selrecord(curlwp, &mq->mq_rsel);
248 	}
249 	if (events & (POLLOUT | POLLWRNORM)) {
250 		/* Ready for sending, if the message queue is not full */
251 		if (mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg)
252 			revents |= (POLLOUT | POLLWRNORM);
253 		else
254 			selrecord(curlwp, &mq->mq_wsel);
255 	}
256 	mutex_exit(&mq->mq_mtx);
257 
258 	return revents;
259 }
260 
261 static int
262 mq_close_fop(file_t *fp)
263 {
264 	struct proc *p = curproc;
265 	struct mqueue *mq = fp->f_data;
266 	bool destroy;
267 
268 	mutex_enter(&mqlist_mtx);
269 	mutex_enter(&mq->mq_mtx);
270 
271 	/* Decrease the counters */
272 	p->p_mqueue_cnt--;
273 	mq->mq_refcnt--;
274 
275 	/* Remove notification if registered for this process */
276 	if (mq->mq_notify_proc == p)
277 		mq->mq_notify_proc = NULL;
278 
279 	/*
280 	 * If this is the last reference and mqueue is marked for unlink,
281 	 * remove and later destroy the message queue.
282 	 */
283 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
284 		LIST_REMOVE(mq, mq_list);
285 		destroy = true;
286 	} else
287 		destroy = false;
288 
289 	mutex_exit(&mq->mq_mtx);
290 	mutex_exit(&mqlist_mtx);
291 
292 	if (destroy)
293 		mqueue_destroy(mq);
294 
295 	return 0;
296 }
297 
298 /*
299  * General mqueue system calls.
300  */
301 
302 int
303 sys_mq_open(struct lwp *l, const struct sys_mq_open_args *uap,
304     register_t *retval)
305 {
306 	/* {
307 		syscallarg(const char *) name;
308 		syscallarg(int) oflag;
309 		syscallarg(mode_t) mode;
310 		syscallarg(struct mq_attr) attr;
311 	} */
312 	struct proc *p = l->l_proc;
313 	struct mqueue *mq, *mq_new = NULL;
314 	file_t *fp;
315 	char *name;
316 	int mqd, error, oflag;
317 
318 	/* Check access mode flags */
319 	oflag = SCARG(uap, oflag);
320 	if ((oflag & O_ACCMODE) == 0)
321 		return EINVAL;
322 
323 	/* Get the name from the user-space */
324 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
325 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
326 	if (error) {
327 		kmem_free(name, MQ_NAMELEN);
328 		return error;
329 	}
330 
331 	if (oflag & O_CREAT) {
332 		struct cwdinfo *cwdi = p->p_cwdi;
333 		struct mq_attr attr;
334 
335 		/* Check the limit */
336 		if (p->p_mqueue_cnt == mq_open_max) {
337 			kmem_free(name, MQ_NAMELEN);
338 			return EMFILE;
339 		}
340 
341 		/* Check for mqueue attributes */
342 		if (SCARG(uap, attr)) {
343 			error = copyin(SCARG(uap, attr), &attr,
344 				sizeof(struct mq_attr));
345 			if (error) {
346 				kmem_free(name, MQ_NAMELEN);
347 				return error;
348 			}
349 			if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 ||
350 			    attr.mq_msgsize > mq_max_msgsize) {
351 				kmem_free(name, MQ_NAMELEN);
352 				return EINVAL;
353 			}
354 			attr.mq_curmsgs = 0;
355 		} else {
356 			memset(&attr, 0, sizeof(struct mq_attr));
357 			attr.mq_maxmsg = mq_def_maxmsg;
358 			attr.mq_msgsize =
359 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
360 		}
361 
362 		/*
363 		 * Allocate new mqueue, initialize data structures,
364 		 * copy the name, attributes and set the flag.
365 		 */
366 		mq_new = kmem_zalloc(sizeof(struct mqueue), KM_SLEEP);
367 
368 		mutex_init(&mq_new->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
369 		cv_init(&mq_new->mq_send_cv, "mqsendcv");
370 		cv_init(&mq_new->mq_recv_cv, "mqrecvcv");
371 		TAILQ_INIT(&mq_new->mq_head);
372 		selinit(&mq_new->mq_rsel);
373 		selinit(&mq_new->mq_wsel);
374 
375 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
376 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
377 		mq_new->mq_attrib.mq_flags = oflag;
378 
379 		/* Store mode and effective UID with GID */
380 		mq_new->mq_mode = ((SCARG(uap, mode) &
381 		    ~cwdi->cwdi_cmask) & ALLPERMS) & ~S_ISTXT;
382 		mq_new->mq_euid = kauth_cred_geteuid(l->l_cred);
383 		mq_new->mq_egid = kauth_cred_getegid(l->l_cred);
384 	}
385 
386 	/* Allocate file structure and descriptor */
387 	error = fd_allocfile(&fp, &mqd);
388 	if (error) {
389 		if (mq_new)
390 			mqueue_destroy(mq_new);
391 		kmem_free(name, MQ_NAMELEN);
392 		return error;
393 	}
394 	fp->f_type = DTYPE_MQUEUE;
395 	fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE);
396 	fp->f_ops = &mqops;
397 
398 	/* Look up for mqueue with such name */
399 	mutex_enter(&mqlist_mtx);
400 	mq = mqueue_lookup(name);
401 	if (mq) {
402 		KASSERT(mutex_owned(&mq->mq_mtx));
403 
404 		/* Check if mqueue is not marked as unlinking */
405 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
406 			error = EACCES;
407 			goto exit;
408 		}
409 		/* Fail if O_EXCL is set, and mqueue already exists */
410 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
411 			error = EEXIST;
412 			goto exit;
413 		}
414 		/* Check the permission */
415 		if (mqueue_access(l, mq, fp->f_flag)) {
416 			error = EACCES;
417 			goto exit;
418 		}
419 	} else {
420 		/* Fail if mqueue neither exists, nor we create it */
421 		if ((oflag & O_CREAT) == 0) {
422 			mutex_exit(&mqlist_mtx);
423 			KASSERT(mq_new == NULL);
424 			fd_abort(p, fp, mqd);
425 			kmem_free(name, MQ_NAMELEN);
426 			return ENOENT;
427 		}
428 
429 		/* Check the limit */
430 		if (p->p_mqueue_cnt == mq_open_max) {
431 			error = EMFILE;
432 			goto exit;
433 		}
434 
435 		/* Insert the queue to the list */
436 		mq = mq_new;
437 		mutex_enter(&mq->mq_mtx);
438 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
439 		mq_new = NULL;
440 	}
441 
442 	/* Increase the counters, and make descriptor ready */
443 	p->p_mqueue_cnt++;
444 	mq->mq_refcnt++;
445 	fp->f_data = mq;
446 exit:
447 	mutex_exit(&mq->mq_mtx);
448 	mutex_exit(&mqlist_mtx);
449 
450 	if (mq_new)
451 		mqueue_destroy(mq_new);
452 	if (error) {
453 		fd_abort(p, fp, mqd);
454 	} else {
455 		fd_affix(p, fp, mqd);
456 		*retval = mqd;
457 	}
458 	kmem_free(name, MQ_NAMELEN);
459 
460 	return error;
461 }
462 
463 int
464 sys_mq_close(struct lwp *l, const struct sys_mq_close_args *uap,
465     register_t *retval)
466 {
467 
468 	return sys_close(l, (const void *)uap, retval);
469 }
470 
471 /*
472  * Primary mq_receive1() function.
473  */
474 int
475 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len,
476     unsigned *msg_prio, int t, ssize_t *mlen)
477 {
478 	file_t *fp = NULL;
479 	struct mqueue *mq;
480 	struct mq_msg *msg = NULL;
481 	int error;
482 
483 	/* Get the message queue */
484 	error = mqueue_get(l, mqdes, FREAD, &fp);
485 	if (error)
486 		return error;
487 	mq = fp->f_data;
488 
489 	/* Check the message size limits */
490 	if (msg_len < mq->mq_attrib.mq_msgsize) {
491 		error = EMSGSIZE;
492 		goto error;
493 	}
494 
495 	/* Check if queue is empty */
496 	while (TAILQ_EMPTY(&mq->mq_head)) {
497 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
498 			error = EAGAIN;
499 			goto error;
500 		}
501 		if (t < 0) {
502 			error = EINVAL;
503 			goto error;
504 		}
505 		/*
506 		 * Block until someone sends the message.
507 		 * While doing this, notification should not be sent.
508 		 */
509 		mq->mq_attrib.mq_flags |= MQ_RECEIVE;
510 		error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
511 		mq->mq_attrib.mq_flags &= ~MQ_RECEIVE;
512 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
513 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
514 			goto error;
515 		}
516 	}
517 
518 	/* Remove the message from the queue */
519 	msg = TAILQ_FIRST(&mq->mq_head);
520 	KASSERT(msg != NULL);
521 	TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
522 
523 	/* Decrement the counter and signal waiter, if any */
524 	mq->mq_attrib.mq_curmsgs--;
525 	cv_signal(&mq->mq_recv_cv);
526 
527 	/* Ready for sending now */
528 	selnotify(&mq->mq_wsel, POLLOUT | POLLWRNORM, 0);
529 error:
530 	mutex_exit(&mq->mq_mtx);
531 	fd_putfile((int)mqdes);
532 	if (error)
533 		return error;
534 
535 	/*
536 	 * Copy the data to the user-space.
537 	 * Note: According to POSIX, no message should be removed from the
538 	 * queue in case of fail - this would be violated.
539 	 */
540 	*mlen = msg->msg_len;
541 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
542 	if (error == 0 && msg_prio)
543 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
544 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
545 
546 	return error;
547 }
548 
549 int
550 sys_mq_receive(struct lwp *l, const struct sys_mq_receive_args *uap,
551     register_t *retval)
552 {
553 	/* {
554 		syscallarg(mqd_t) mqdes;
555 		syscallarg(char *) msg_ptr;
556 		syscallarg(size_t) msg_len;
557 		syscallarg(unsigned *) msg_prio;
558 	} */
559 	int error;
560 	ssize_t mlen;
561 
562 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
563 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen);
564 	if (error == 0)
565 		*retval = mlen;
566 
567 	return error;
568 }
569 
570 int
571 sys___mq_timedreceive50(struct lwp *l,
572     const struct sys___mq_timedreceive50_args *uap, register_t *retval)
573 {
574 	/* {
575 		syscallarg(mqd_t) mqdes;
576 		syscallarg(char *) msg_ptr;
577 		syscallarg(size_t) msg_len;
578 		syscallarg(unsigned *) msg_prio;
579 		syscallarg(const struct timespec *) abs_timeout;
580 	} */
581 	int error, t;
582 	ssize_t mlen;
583 	struct timespec ts;
584 
585 	/* Get and convert time value */
586 	if (SCARG(uap, abs_timeout)) {
587 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
588 		if (error)
589 			return error;
590 
591 		error = abstimeout2timo(&ts, &t);
592 		if (error)
593 			return error;
594 	} else
595 		t = 0;
596 
597 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
598 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t, &mlen);
599 	if (error == 0)
600 		*retval = mlen;
601 
602 	return error;
603 }
604 
605 /*
606  * Primary mq_send1() function.
607  */
608 int
609 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len,
610     unsigned msg_prio, int t)
611 {
612 	file_t *fp = NULL;
613 	struct mqueue *mq;
614 	struct mq_msg *msg, *pos_msg;
615 	struct proc *notify = NULL;
616 	ksiginfo_t ksi;
617 	size_t size;
618 	int error;
619 
620 	/* Check the priority range */
621 	if (msg_prio >= mq_prio_max)
622 		return EINVAL;
623 
624 	/* Allocate a new message */
625 	size = sizeof(struct mq_msg) + msg_len;
626 	if (size > mq_max_msgsize)
627 		return EMSGSIZE;
628 
629 	if (size > MQ_DEF_MSGSIZE)
630 		msg = kmem_alloc(size, KM_SLEEP);
631 	else
632 		msg = pool_cache_get(mqmsg_cache, PR_WAITOK);
633 
634 	/* Get the data from user-space */
635 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
636 	if (error) {
637 		mqueue_freemsg(msg, size);
638 		return error;
639 	}
640 	msg->msg_len = msg_len;
641 	msg->msg_prio = msg_prio;
642 
643 	/* Get the mqueue */
644 	error = mqueue_get(l, mqdes, FWRITE, &fp);
645 	if (error) {
646 		mqueue_freemsg(msg, size);
647 		return error;
648 	}
649 	mq = fp->f_data;
650 
651 	/* Check the message size limit */
652 	if (msg_len <= 0 || msg_len > mq->mq_attrib.mq_msgsize) {
653 		error = EMSGSIZE;
654 		goto error;
655 	}
656 
657 	/* Check if queue is full */
658 	while (mq->mq_attrib.mq_curmsgs >= mq->mq_attrib.mq_maxmsg) {
659 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
660 			error = EAGAIN;
661 			goto error;
662 		}
663 		if (t < 0) {
664 			error = EINVAL;
665 			goto error;
666 		}
667 		/* Block until queue becomes available */
668 		error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
669 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
670 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
671 			goto error;
672 		}
673 	}
674 	KASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg);
675 
676 	/* Insert message into the queue, according to the priority */
677 	TAILQ_FOREACH(pos_msg, &mq->mq_head, msg_queue)
678 		if (msg->msg_prio > pos_msg->msg_prio)
679 			break;
680 	if (pos_msg == NULL)
681 		TAILQ_INSERT_TAIL(&mq->mq_head, msg, msg_queue);
682 	else
683 		TAILQ_INSERT_BEFORE(pos_msg, msg, msg_queue);
684 
685 	/* Check for the notify */
686 	if (mq->mq_attrib.mq_curmsgs == 0 && mq->mq_notify_proc &&
687 	    (mq->mq_attrib.mq_flags & MQ_RECEIVE) == 0) {
688 		/* Initialize the signal */
689 		KSI_INIT(&ksi);
690 		ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
691 		ksi.ksi_code = SI_MESGQ;
692 		ksi.ksi_value = mq->mq_sig_notify.sigev_value;
693 		/* Unregister the process */
694 		notify = mq->mq_notify_proc;
695 		mq->mq_notify_proc = NULL;
696 	}
697 
698 	/* Increment the counter and signal waiter, if any */
699 	mq->mq_attrib.mq_curmsgs++;
700 	cv_signal(&mq->mq_send_cv);
701 
702 	/* Ready for receiving now */
703 	selnotify(&mq->mq_rsel, POLLIN | POLLRDNORM, 0);
704 error:
705 	mutex_exit(&mq->mq_mtx);
706 	fd_putfile((int)mqdes);
707 
708 	if (error) {
709 		mqueue_freemsg(msg, size);
710 	} else if (notify) {
711 		/* Send the notify, if needed */
712 		mutex_enter(proc_lock);
713 		kpsignal(notify, &ksi, NULL);
714 		mutex_exit(proc_lock);
715 	}
716 
717 	return error;
718 }
719 
720 int
721 sys_mq_send(struct lwp *l, const struct sys_mq_send_args *uap,
722     register_t *retval)
723 {
724 	/* {
725 		syscallarg(mqd_t) mqdes;
726 		syscallarg(const char *) msg_ptr;
727 		syscallarg(size_t) msg_len;
728 		syscallarg(unsigned) msg_prio;
729 	} */
730 
731 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
732 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0);
733 }
734 
735 int
736 sys___mq_timedsend50(struct lwp *l, const struct sys___mq_timedsend50_args *uap,
737     register_t *retval)
738 {
739 	/* {
740 		syscallarg(mqd_t) mqdes;
741 		syscallarg(const char *) msg_ptr;
742 		syscallarg(size_t) msg_len;
743 		syscallarg(unsigned) msg_prio;
744 		syscallarg(const struct timespec *) abs_timeout;
745 	} */
746 	int t;
747 	struct timespec ts;
748 	int error;
749 
750 	/* Get and convert time value */
751 	if (SCARG(uap, abs_timeout)) {
752 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
753 		if (error)
754 			return error;
755 		error = abstimeout2timo(&ts, &t);
756 		if (error)
757 			return error;
758 	} else
759 		t = 0;
760 
761 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
762 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t);
763 }
764 
765 int
766 sys_mq_notify(struct lwp *l, const struct sys_mq_notify_args *uap,
767     register_t *retval)
768 {
769 	/* {
770 		syscallarg(mqd_t) mqdes;
771 		syscallarg(const struct sigevent *) notification;
772 	} */
773 	file_t *fp = NULL;
774 	struct mqueue *mq;
775 	struct sigevent sig;
776 	int error;
777 
778 	if (SCARG(uap, notification)) {
779 		/* Get the signal from user-space */
780 		error = copyin(SCARG(uap, notification), &sig,
781 		    sizeof(struct sigevent));
782 		if (error)
783 			return error;
784 	}
785 
786 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
787 	if (error)
788 		return error;
789 	mq = fp->f_data;
790 
791 	if (SCARG(uap, notification)) {
792 		/* Register notification: set the signal and target process */
793 		if (mq->mq_notify_proc == NULL) {
794 			memcpy(&mq->mq_sig_notify, &sig,
795 			    sizeof(struct sigevent));
796 			mq->mq_notify_proc = l->l_proc;
797 		} else {
798 			/* Fail if someone else already registered */
799 			error = EBUSY;
800 		}
801 	} else {
802 		/* Unregister the notification */
803 		mq->mq_notify_proc = NULL;
804 	}
805 	mutex_exit(&mq->mq_mtx);
806 	fd_putfile((int)SCARG(uap, mqdes));
807 
808 	return error;
809 }
810 
811 int
812 sys_mq_getattr(struct lwp *l, const struct sys_mq_getattr_args *uap,
813     register_t *retval)
814 {
815 	/* {
816 		syscallarg(mqd_t) mqdes;
817 		syscallarg(struct mq_attr *) mqstat;
818 	} */
819 	file_t *fp = NULL;
820 	struct mqueue *mq;
821 	struct mq_attr attr;
822 	int error;
823 
824 	/* Get the message queue */
825 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
826 	if (error)
827 		return error;
828 	mq = fp->f_data;
829 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
830 	mutex_exit(&mq->mq_mtx);
831 	fd_putfile((int)SCARG(uap, mqdes));
832 
833 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
834 }
835 
836 int
837 sys_mq_setattr(struct lwp *l, const struct sys_mq_setattr_args *uap,
838     register_t *retval)
839 {
840 	/* {
841 		syscallarg(mqd_t) mqdes;
842 		syscallarg(const struct mq_attr *) mqstat;
843 		syscallarg(struct mq_attr *) omqstat;
844 	} */
845 	file_t *fp = NULL;
846 	struct mqueue *mq;
847 	struct mq_attr attr;
848 	int error, nonblock;
849 
850 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
851 	if (error)
852 		return error;
853 	nonblock = (attr.mq_flags & O_NONBLOCK);
854 
855 	/* Get the message queue */
856 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
857 	if (error)
858 		return error;
859 	mq = fp->f_data;
860 
861 	/* Copy the old attributes, if needed */
862 	if (SCARG(uap, omqstat))
863 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
864 
865 	/* Ignore everything, except O_NONBLOCK */
866 	if (nonblock)
867 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
868 	else
869 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
870 
871 	mutex_exit(&mq->mq_mtx);
872 	fd_putfile((int)SCARG(uap, mqdes));
873 
874 	/*
875 	 * Copy the data to the user-space.
876 	 * Note: According to POSIX, the new attributes should not be set in
877 	 * case of fail - this would be violated.
878 	 */
879 	if (SCARG(uap, omqstat))
880 		error = copyout(&attr, SCARG(uap, omqstat),
881 		    sizeof(struct mq_attr));
882 
883 	return error;
884 }
885 
886 int
887 sys_mq_unlink(struct lwp *l, const struct sys_mq_unlink_args *uap,
888     register_t *retval)
889 {
890 	/* {
891 		syscallarg(const char *) name;
892 	} */
893 	struct mqueue *mq;
894 	char *name;
895 	int error, refcnt = 0;
896 
897 	/* Get the name from the user-space */
898 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
899 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
900 	if (error) {
901 		kmem_free(name, MQ_NAMELEN);
902 		return error;
903 	}
904 
905 	/* Lookup for this file */
906 	mutex_enter(&mqlist_mtx);
907 	mq = mqueue_lookup(name);
908 	if (mq == NULL) {
909 		error = ENOENT;
910 		goto error;
911 	}
912 
913 	/* Check the permissions */
914 	if (mqueue_access(l, mq, FWRITE)) {
915 		mutex_exit(&mq->mq_mtx);
916 		error = EACCES;
917 		goto error;
918 	}
919 
920 	/* Mark message queue as unlinking, before leaving the window */
921 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
922 
923 	/* Wake up all waiters, if there are such */
924 	cv_broadcast(&mq->mq_send_cv);
925 	cv_broadcast(&mq->mq_recv_cv);
926 
927 	selnotify(&mq->mq_rsel, POLLHUP, 0);
928 	selnotify(&mq->mq_wsel, POLLHUP, 0);
929 
930 	refcnt = mq->mq_refcnt;
931 	if (refcnt == 0)
932 		LIST_REMOVE(mq, mq_list);
933 
934 	mutex_exit(&mq->mq_mtx);
935 error:
936 	mutex_exit(&mqlist_mtx);
937 
938 	/*
939 	 * If there are no references - destroy the message
940 	 * queue, otherwise, the last mq_close() will do that.
941 	 */
942 	if (error == 0 && refcnt == 0)
943 		mqueue_destroy(mq);
944 
945 	kmem_free(name, MQ_NAMELEN);
946 	return error;
947 }
948 
949 /*
950  * SysCtl.
951  */
952 
953 SYSCTL_SETUP(sysctl_mqueue_setup, "sysctl mqueue setup")
954 {
955 	const struct sysctlnode *node = NULL;
956 
957 	sysctl_createv(clog, 0, NULL, NULL,
958 		CTLFLAG_PERMANENT,
959 		CTLTYPE_NODE, "kern", NULL,
960 		NULL, 0, NULL, 0,
961 		CTL_KERN, CTL_EOL);
962 	sysctl_createv(clog, 0, NULL, NULL,
963 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
964 		CTLTYPE_INT, "posix_msg",
965 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
966 			     "Message Passing option to which the "
967 			     "system attempts to conform"),
968 		NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
969 		CTL_KERN, CTL_CREATE, CTL_EOL);
970 	sysctl_createv(clog, 0, NULL, &node,
971 		CTLFLAG_PERMANENT,
972 		CTLTYPE_NODE, "mqueue",
973 		SYSCTL_DESCR("Message queue options"),
974 		NULL, 0, NULL, 0,
975 		CTL_KERN, CTL_CREATE, CTL_EOL);
976 
977 	if (node == NULL)
978 		return;
979 
980 	sysctl_createv(clog, 0, &node, NULL,
981 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
982 		CTLTYPE_INT, "mq_open_max",
983 		SYSCTL_DESCR("Maximal number of message queue descriptors "
984 			     "that process could open"),
985 		NULL, 0, &mq_open_max, 0,
986 		CTL_CREATE, CTL_EOL);
987 	sysctl_createv(clog, 0, &node, NULL,
988 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
989 		CTLTYPE_INT, "mq_prio_max",
990 		SYSCTL_DESCR("Maximal priority of the message"),
991 		NULL, 0, &mq_prio_max, 0,
992 		CTL_CREATE, CTL_EOL);
993 	sysctl_createv(clog, 0, &node, NULL,
994 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
995 		CTLTYPE_INT, "mq_max_msgsize",
996 		SYSCTL_DESCR("Maximal allowed size of the message"),
997 		NULL, 0, &mq_max_msgsize, 0,
998 		CTL_CREATE, CTL_EOL);
999 	sysctl_createv(clog, 0, &node, NULL,
1000 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1001 		CTLTYPE_INT, "mq_def_maxmsg",
1002 		SYSCTL_DESCR("Default maximal message count"),
1003 		NULL, 0, &mq_def_maxmsg, 0,
1004 		CTL_CREATE, CTL_EOL);
1005 }
1006 
1007 /*
1008  * Debugging.
1009  */
1010 #if defined(DDB)
1011 
1012 void
1013 mqueue_print_list(void (*pr)(const char *, ...))
1014 {
1015 	struct mqueue *mq;
1016 
1017 	(*pr)("Global list of the message queues:\n");
1018 	(*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
1019 	    "Name", "Ptr", "Mode", "Flags",  "Ref",
1020 	    "MaxMsg", "MsgSze", "CurMsg");
1021 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
1022 		(*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
1023 		    mq->mq_name, mq, mq->mq_mode,
1024 		    mq->mq_attrib.mq_flags, mq->mq_refcnt,
1025 		    mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
1026 		    mq->mq_attrib.mq_curmsgs);
1027 	}
1028 }
1029 
1030 #endif /* defined(DDB) */
1031