xref: /netbsd-src/sys/kern/sys_mqueue.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: sys_mqueue.c,v 1.5 2007/11/28 19:30:56 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 2007, 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 COPYRIGHT HOLDERS AND CONTRIBUTORS
17  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF 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  *  Global list of message queues and proc::p_mqueue_cnt counter are protected
35  *  by mqlist_mtx lock.  Concrete message queue and its members are protected
36  *  by mqueue::mq_mtx.
37  *
38  * Lock order:
39  * 	mqlist_mtx
40  * 	  -> mqueue::mq_mtx
41  *
42  * TODO:
43  *  - Hashing or RB-tree for the global list.
44  *  - Support for select(), poll() and perhaps kqueue().
45  */
46 
47 #include <sys/cdefs.h>
48 __KERNEL_RCSID(0, "$NetBSD: sys_mqueue.c,v 1.5 2007/11/28 19:30:56 rmind Exp $");
49 
50 #include <sys/param.h>
51 #include <sys/types.h>
52 
53 #include <sys/condvar.h>
54 #include <sys/errno.h>
55 #include <sys/fcntl.h>
56 #include <sys/file.h>
57 #include <sys/filedesc.h>
58 #include <sys/kauth.h>
59 #include <sys/kernel.h>
60 #include <sys/kmem.h>
61 #include <sys/lwp.h>
62 #include <sys/mqueue.h>
63 #include <sys/mutex.h>
64 #include <sys/pool.h>
65 #include <sys/proc.h>
66 #include <sys/queue.h>
67 #include <sys/signal.h>
68 #include <sys/signalvar.h>
69 #include <sys/sysctl.h>
70 #include <sys/syscallargs.h>
71 #include <sys/systm.h>
72 #include <sys/unistd.h>
73 #include <sys/vnode.h>
74 
75 /* System-wide limits. */
76 static u_int			mq_open_max = MQ_OPEN_MAX;
77 static u_int			mq_prio_max = MQ_PRIO_MAX;
78 
79 static u_int			mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
80 static u_int			mq_def_maxmsg = 32;
81 
82 static kmutex_t			mqlist_mtx;
83 static struct pool		mqmsg_poll;
84 static LIST_HEAD(, mqueue)	mqueue_head =
85 	LIST_HEAD_INITIALIZER(mqueue_head);
86 
87 static int	mq_close_fop(struct file *, struct lwp *);
88 
89 static const struct fileops mqops = {
90 	fbadop_read, fbadop_write, fbadop_ioctl, fnullop_fcntl, fnullop_poll,
91 	fbadop_stat, mq_close_fop, fnullop_kqfilter
92 };
93 
94 /*
95  * Initialize POSIX message queue subsystem.
96  */
97 void
98 mqueue_sysinit(void)
99 {
100 
101 	pool_init(&mqmsg_poll, MQ_DEF_MSGSIZE, 0, 0, 0,
102 	    "mqmsg_poll", &pool_allocator_nointr, IPL_NONE);
103 	mutex_init(&mqlist_mtx, MUTEX_DEFAULT, IPL_NONE);
104 }
105 
106 /*
107  * Free the message.
108  */
109 static void
110 mqueue_freemsg(struct mq_msg *msg, const size_t size)
111 {
112 
113 	if (size > MQ_DEF_MSGSIZE)
114 		kmem_free(msg, size);
115 	else
116 		pool_put(&mqmsg_poll, msg);
117 }
118 
119 /*
120  * Destroy the message queue.
121  */
122 static void
123 mqueue_destroy(struct mqueue *mq)
124 {
125 	struct mq_msg *msg;
126 
127 	while ((msg = TAILQ_FIRST(&mq->mq_head)) != NULL) {
128 		TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
129 		mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
130 	}
131 	cv_destroy(&mq->mq_send_cv);
132 	cv_destroy(&mq->mq_recv_cv);
133 	mutex_destroy(&mq->mq_mtx);
134 	kmem_free(mq, sizeof(struct mqueue));
135 }
136 
137 /*
138  * Lookup for file name in general list of message queues.
139  *  => locks the message queue
140  */
141 static void *
142 mqueue_lookup(char *name)
143 {
144 	struct mqueue *mq;
145 	KASSERT(mutex_owned(&mqlist_mtx));
146 
147 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
148 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
149 			mutex_enter(&mq->mq_mtx);
150 			return mq;
151 		}
152 	}
153 
154 	return NULL;
155 }
156 
157 /*
158  * Check access against message queue.
159  */
160 static inline int
161 mqueue_access(struct lwp *l, struct mqueue *mq, mode_t acc_mode)
162 {
163 
164 	KASSERT(mutex_owned(&mq->mq_mtx));
165 	return vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid,
166 	    acc_mode, l->l_cred);
167 }
168 
169 /*
170  * Get the mqueue from the descriptor.
171  *  => locks the message queue
172  *  => increments the reference on file entry
173  */
174 static int
175 mqueue_get(struct lwp *l, mqd_t mqd, mode_t acc_mode, struct file **fpr)
176 {
177 	const struct proc *p = l->l_proc;
178 	struct file *fp;
179 	struct mqueue *mq;
180 
181 	/* Get the file and descriptor */
182 	fp = fd_getfile(p->p_fd, (int)mqd);
183 	if (fp == NULL)
184 		return EBADF;
185 
186 	/* Increment the reference of file entry, and lock the mqueue */
187 	FILE_USE(fp);
188 	mq = fp->f_data;
189 	*fpr = fp;
190 	mutex_enter(&mq->mq_mtx);
191 
192 	/* Check the access mode and permission if needed */
193 	if (acc_mode == VNOVAL)
194 		return 0;
195 	if ((acc_mode & fp->f_flag) == 0 || mqueue_access(l, mq, acc_mode)) {
196 		mutex_exit(&mq->mq_mtx);
197 		FILE_UNUSE(fp, l);
198 		return EPERM;
199 	}
200 
201 	return 0;
202 }
203 
204 /*
205  * Converter from struct timespec to the ticks.
206  * Used by mq_timedreceive(), mq_timedsend().
207  */
208 static int
209 abstimeout2timo(const void *uaddr, int *timo)
210 {
211 	struct timespec ts;
212 	int error;
213 
214 	error = copyin(uaddr, &ts, sizeof(struct timespec));
215 	if (error)
216 		return error;
217 
218 	/*
219 	 * According to POSIX, validation check is needed only in case of
220 	 * blocking.  Thus, set the invalid value right now, and fail latter.
221 	 */
222 	error = itimespecfix(&ts);
223 	*timo = (error == 0) ? tstohz(&ts) : -1;
224 
225 	return 0;
226 }
227 
228 static int
229 mq_close_fop(struct file *fp, struct lwp *l)
230 {
231 	struct proc *p = l->l_proc;
232 	struct mqueue *mq = fp->f_data;
233 	bool destroy;
234 
235 	mutex_enter(&mqlist_mtx);
236 	mutex_enter(&mq->mq_mtx);
237 
238 	/* Decrease the counters */
239 	p->p_mqueue_cnt--;
240 	mq->mq_refcnt--;
241 
242 	/* Remove notification if registered for this process */
243 	if (mq->mq_notify_proc == p)
244 		mq->mq_notify_proc = NULL;
245 
246 	/*
247 	 * If this is the last reference and mqueue is marked for unlink,
248 	 * remove and later destroy the message queue.
249 	 */
250 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
251 		LIST_REMOVE(mq, mq_list);
252 		destroy = true;
253 	} else
254 		destroy = false;
255 
256 	mutex_exit(&mq->mq_mtx);
257 	mutex_exit(&mqlist_mtx);
258 
259 	if (destroy)
260 		mqueue_destroy(mq);
261 
262 	return 0;
263 }
264 
265 /*
266  * General mqueue system calls.
267  */
268 
269 int
270 sys_mq_open(struct lwp *l, void *v, register_t *retval)
271 {
272 	struct sys_mq_open_args /* {
273 		syscallarg(const char *) name;
274 		syscallarg(int) oflag;
275 		syscallarg(mode_t) mode;
276 		syscallarg(struct mq_attr) attr;
277 	} */ *uap = v;
278 	struct proc *p = l->l_proc;
279 	struct mqueue *mq, *mq_new = NULL;
280 	struct file *fp;
281 	char *name;
282 	int mqd, error, oflag;
283 
284 	/* Check access mode flags */
285 	oflag = SCARG(uap, oflag);
286 	if ((oflag & (O_RDWR | O_RDONLY | O_WRONLY)) == 0)
287 		return EINVAL;
288 
289 	/* Get the name from the user-space */
290 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
291 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
292 	if (error) {
293 		kmem_free(name, MQ_NAMELEN);
294 		return error;
295 	}
296 
297 	if (oflag & O_CREAT) {
298 		struct mq_attr attr;
299 
300 		/* Check the limit */
301 		if (p->p_mqueue_cnt == mq_open_max) {
302 			kmem_free(name, MQ_NAMELEN);
303 			return EMFILE;
304 		}
305 
306 		/* Check for mqueue attributes */
307 		if (SCARG(uap, attr)) {
308 			error = copyin(SCARG(uap, attr), &attr,
309 				sizeof(struct mq_attr));
310 			if (error) {
311 				kmem_free(name, MQ_NAMELEN);
312 				return error;
313 			}
314 			if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 ||
315 			    attr.mq_msgsize > mq_max_msgsize) {
316 				kmem_free(name, MQ_NAMELEN);
317 				return EINVAL;
318 			}
319 			attr.mq_curmsgs = 0;
320 		} else {
321 			memset(&attr, 0, sizeof(struct mq_attr));
322 			attr.mq_maxmsg = mq_def_maxmsg;
323 			attr.mq_msgsize =
324 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
325 		}
326 
327 		/*
328 		 * Allocate new mqueue, initialize data structures,
329 		 * copy the name, attributes and set the flag.
330 		 */
331 		mq_new = kmem_zalloc(sizeof(struct mqueue), KM_SLEEP);
332 
333 		mutex_init(&mq_new->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
334 		cv_init(&mq_new->mq_send_cv, "mqsendcv");
335 		cv_init(&mq_new->mq_recv_cv, "mqrecvcv");
336 		TAILQ_INIT(&mq_new->mq_head);
337 
338 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
339 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
340 		mq_new->mq_attrib.mq_flags = oflag;
341 
342 		/* Store mode and effective UID with GID */
343 		mq_new->mq_mode = SCARG(uap, mode);
344 		mq_new->mq_euid = kauth_cred_geteuid(l->l_cred);
345 		mq_new->mq_egid = kauth_cred_getegid(l->l_cred);
346 	}
347 
348 	/* Allocate file structure and descriptor */
349 	error = falloc(l, &fp, &mqd);
350 	if (error) {
351 		if (mq_new)
352 			mqueue_destroy(mq_new);
353 		kmem_free(name, MQ_NAMELEN);
354 		return error;
355 	}
356 	fp->f_type = DTYPE_MQUEUE;
357 	fp->f_flag = (oflag & O_RDWR) ? (VREAD | VWRITE) :
358 	    ((oflag & O_RDONLY) ? VREAD : VWRITE);
359 	fp->f_ops = &mqops;
360 
361 	/* Look up for mqueue with such name */
362 	mutex_enter(&mqlist_mtx);
363 	mq = mqueue_lookup(name);
364 	if (mq) {
365 		KASSERT(mutex_owned(&mq->mq_mtx));
366 		/* Check if mqueue is not marked as unlinking */
367 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
368 			error = EACCES;
369 			goto exit;
370 		}
371 		/* Fail if O_EXCL is set, and mqueue already exists */
372 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
373 			error = EEXIST;
374 			goto exit;
375 		}
376 		/* Check the permission */
377 		if (mqueue_access(l, mq, fp->f_flag)) {
378 			error = EACCES;
379 			goto exit;
380 		}
381 	} else {
382 		/* Fail if mqueue neither exists, nor we create it */
383 		if ((oflag & O_CREAT) == 0) {
384 			mutex_exit(&mqlist_mtx);
385 			KASSERT(mq_new == NULL);
386 			FILE_UNUSE(fp, l);
387 			ffree(fp);
388 			fdremove(p->p_fd, mqd);
389 			kmem_free(name, MQ_NAMELEN);
390 			return ENOENT;
391 		}
392 
393 		/* Check the limit */
394 		if (p->p_mqueue_cnt == mq_open_max) {
395 			error = EMFILE;
396 			goto exit;
397 		}
398 
399 		/* Insert the queue to the list */
400 		mq = mq_new;
401 		mutex_enter(&mq->mq_mtx);
402 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
403 		mq_new = NULL;
404 	}
405 
406 	/* Increase the counters, and make descriptor ready */
407 	p->p_mqueue_cnt++;
408 	mq->mq_refcnt++;
409 	fp->f_data = mq;
410 	FILE_SET_MATURE(fp);
411 exit:
412 	mutex_exit(&mq->mq_mtx);
413 	mutex_exit(&mqlist_mtx);
414 	FILE_UNUSE(fp, l);
415 
416 	if (mq_new)
417 		mqueue_destroy(mq_new);
418 	if (error) {
419 		ffree(fp);
420 		fdremove(p->p_fd, mqd);
421 	} else
422 		*retval = mqd;
423 	kmem_free(name, MQ_NAMELEN);
424 
425 	return error;
426 }
427 
428 int
429 sys_mq_close(struct lwp *l, void *v, register_t *retval)
430 {
431 
432 	return sys_close(l, v, retval);
433 }
434 
435 /*
436  * Primary mq_receive1() function.
437  */
438 static int
439 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len,
440     unsigned *msg_prio, int t, ssize_t *mlen)
441 {
442 	struct file *fp = NULL;
443 	struct mqueue *mq;
444 	struct mq_msg *msg = NULL;
445 	int error;
446 
447 	/* Get the message queue */
448 	error = mqueue_get(l, mqdes, VREAD, &fp);
449 	if (error)
450 		return error;
451 	mq = fp->f_data;
452 
453 	/* Check the message size limits */
454 	if (msg_len < mq->mq_attrib.mq_msgsize) {
455 		error = EMSGSIZE;
456 		goto error;
457 	}
458 
459 	/* Check if queue is empty */
460 	while (TAILQ_EMPTY(&mq->mq_head)) {
461 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
462 			error = EAGAIN;
463 			goto error;
464 		}
465 		if (t < 0) {
466 			error = EINVAL;
467 			goto error;
468 		}
469 		/*
470 		 * Block until someone sends the message.
471 		 * While doing this, notification should not be sent.
472 		 */
473 		mq->mq_attrib.mq_flags |= MQ_RECEIVE;
474 		error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
475 		mq->mq_attrib.mq_flags &= ~MQ_RECEIVE;
476 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
477 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
478 			goto error;
479 		}
480 	}
481 
482 	/* Remove the message from the queue */
483 	msg = TAILQ_FIRST(&mq->mq_head);
484 	KASSERT(msg != NULL);
485 	TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
486 
487 	/* Decrement the counter and signal waiter, if any */
488 	mq->mq_attrib.mq_curmsgs--;
489 	cv_signal(&mq->mq_recv_cv);
490 error:
491 	mutex_exit(&mq->mq_mtx);
492 	FILE_UNUSE(fp, l);
493 	if (error)
494 		return error;
495 
496 	/*
497 	 * Copy the data to the user-space.
498 	 * Note: According to POSIX, no message should be removed from the
499 	 * queue in case of fail - this would be violated.
500 	 */
501 	*mlen = msg->msg_len;
502 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
503 	if (error == 0 && msg_prio)
504 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
505 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
506 
507 	return error;
508 }
509 
510 int
511 sys_mq_receive(struct lwp *l, void *v, register_t *retval)
512 {
513 	struct sys_mq_receive_args /* {
514 		syscallarg(mqd_t) mqdes;
515 		syscallarg(char *) msg_ptr;
516 		syscallarg(size_t) msg_len;
517 		syscallarg(unsigned *) msg_prio;
518 	} */ *uap = v;
519 	int error;
520 	ssize_t mlen;
521 
522 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
523 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen);
524 	if (error == 0)
525 		*retval = mlen;
526 
527 	return error;
528 }
529 
530 int
531 sys_mq_timedreceive(struct lwp *l, void *v, register_t *retval)
532 {
533 	struct sys_mq_timedreceive_args /* {
534 		syscallarg(mqd_t) mqdes;
535 		syscallarg(char *) msg_ptr;
536 		syscallarg(size_t) msg_len;
537 		syscallarg(unsigned *) msg_prio;
538 		syscallarg(const struct timespec *) abs_timeout;
539 	} */ *uap = v;
540 	int error, t;
541 	ssize_t mlen;
542 
543 	/* Get and convert time value */
544 	if (SCARG(uap, abs_timeout)) {
545 		error = abstimeout2timo(SCARG(uap, abs_timeout), &t);
546 		if (error)
547 			return error;
548 	} else
549 		t = 0;
550 
551 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
552 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t, &mlen);
553 	if (error == 0)
554 		*retval = mlen;
555 
556 	return error;
557 }
558 
559 /*
560  * Primary mq_send1() function.
561  */
562 static int
563 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len,
564     unsigned msg_prio, int t)
565 {
566 	struct file *fp = NULL;
567 	struct mqueue *mq;
568 	struct mq_msg *msg, *pos_msg;
569 	struct proc *notify = NULL;
570 	ksiginfo_t ksi;
571 	size_t size;
572 	int error;
573 
574 	/* Check the priority range */
575 	if (msg_prio >= mq_prio_max)
576 		return EINVAL;
577 
578 	/* Allocate a new message */
579 	size = sizeof(struct mq_msg) + msg_len;
580 	if (size > mq_max_msgsize)
581 		return EMSGSIZE;
582 
583 	if (size > MQ_DEF_MSGSIZE)
584 		msg = kmem_alloc(size, KM_SLEEP);
585 	else
586 		msg = pool_get(&mqmsg_poll, PR_WAITOK);
587 
588 	/* Get the data from user-space */
589 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
590 	if (error) {
591 		mqueue_freemsg(msg, size);
592 		return error;
593 	}
594 	msg->msg_len = msg_len;
595 	msg->msg_prio = msg_prio;
596 
597 	/* Get the mqueue */
598 	error = mqueue_get(l, mqdes, VWRITE, &fp);
599 	if (error) {
600 		mqueue_freemsg(msg, size);
601 		return error;
602 	}
603 	mq = fp->f_data;
604 
605 	/* Check the message size limit */
606 	if (msg_len <= 0 || msg_len > mq->mq_attrib.mq_msgsize) {
607 		error = EMSGSIZE;
608 		goto error;
609 	}
610 
611 	/* Check if queue is full */
612 	while (mq->mq_attrib.mq_curmsgs >= mq->mq_attrib.mq_maxmsg) {
613 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
614 			error = EAGAIN;
615 			goto error;
616 		}
617 		if (t < 0) {
618 			error = EINVAL;
619 			goto error;
620 		}
621 		/* Block until queue becomes available */
622 		error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
623 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
624 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
625 			goto error;
626 		}
627 	}
628 	KASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg);
629 
630 	/* Insert message into the queue, according to the priority */
631 	TAILQ_FOREACH(pos_msg, &mq->mq_head, msg_queue)
632 		if (msg->msg_prio > pos_msg->msg_prio)
633 			break;
634 	if (pos_msg == NULL)
635 		TAILQ_INSERT_TAIL(&mq->mq_head, msg, msg_queue);
636 	else
637 		TAILQ_INSERT_BEFORE(pos_msg, msg, msg_queue);
638 
639 	/* Check for the notify */
640 	if (mq->mq_attrib.mq_curmsgs == 0 && mq->mq_notify_proc &&
641 	    (mq->mq_attrib.mq_flags & MQ_RECEIVE) == 0) {
642 		/* Initialize the signal */
643 		KSI_INIT(&ksi);
644 		ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
645 		ksi.ksi_code = SI_MESGQ;
646 		ksi.ksi_value = mq->mq_sig_notify.sigev_value;
647 		/* Unregister the process */
648 		notify = mq->mq_notify_proc;
649 		mq->mq_notify_proc = NULL;
650 	}
651 
652 	/* Increment the counter and signal waiter, if any */
653 	mq->mq_attrib.mq_curmsgs++;
654 	cv_signal(&mq->mq_send_cv);
655 error:
656 	mutex_exit(&mq->mq_mtx);
657 	FILE_UNUSE(fp, l);
658 
659 	if (error) {
660 		mqueue_freemsg(msg, size);
661 	} else if (notify) {
662 		/* Send the notify, if needed */
663 		mutex_enter(&proclist_mutex);
664 		kpsignal(notify, &ksi, NULL);
665 		mutex_exit(&proclist_mutex);
666 	}
667 
668 	return error;
669 }
670 
671 int
672 sys_mq_send(struct lwp *l, void *v, register_t *retval)
673 {
674 	struct sys_mq_send_args /* {
675 		syscallarg(mqd_t) mqdes;
676 		syscallarg(const char *) msg_ptr;
677 		syscallarg(size_t) msg_len;
678 		syscallarg(unsigned) msg_prio;
679 	} */ *uap = v;
680 
681 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
682 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0);
683 }
684 
685 int
686 sys_mq_timedsend(struct lwp *l, void *v, register_t *retval)
687 {
688 	struct sys_mq_timedsend_args /* {
689 		syscallarg(mqd_t) mqdes;
690 		syscallarg(const char *) msg_ptr;
691 		syscallarg(size_t) msg_len;
692 		syscallarg(unsigned) msg_prio;
693 		syscallarg(const struct timespec *) abs_timeout;
694 	} */ *uap = v;
695 	int t;
696 
697 	/* Get and convert time value */
698 	if (SCARG(uap, abs_timeout)) {
699 		int error = abstimeout2timo(SCARG(uap, abs_timeout), &t);
700 		if (error)
701 			return error;
702 	} else
703 		t = 0;
704 
705 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
706 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t);
707 }
708 
709 int
710 sys_mq_notify(struct lwp *l, void *v, register_t *retval)
711 {
712 	struct sys_mq_notify_args /* {
713 		syscallarg(mqd_t) mqdes;
714 		syscallarg(const struct sigevent *) notification;
715 	} */ *uap = v;
716 	struct file *fp = NULL;
717 	struct mqueue *mq;
718 	struct sigevent sig;
719 	int error;
720 
721 	if (SCARG(uap, notification)) {
722 		/* Get the signal from user-space */
723 		error = copyin(SCARG(uap, notification), &sig,
724 		    sizeof(struct sigevent));
725 		if (error)
726 			return error;
727 	}
728 
729 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
730 	if (error)
731 		return error;
732 	mq = fp->f_data;
733 
734 	if (SCARG(uap, notification)) {
735 		/* Register notification: set the signal and target process */
736 		if (mq->mq_notify_proc == NULL) {
737 			memcpy(&mq->mq_sig_notify, &sig,
738 			    sizeof(struct sigevent));
739 			mq->mq_notify_proc = l->l_proc;
740 		} else {
741 			/* Fail if someone else already registered */
742 			error = EBUSY;
743 		}
744 	} else {
745 		/* Unregister the notification */
746 		mq->mq_notify_proc = NULL;
747 	}
748 	mutex_exit(&mq->mq_mtx);
749 	FILE_UNUSE(fp, l);
750 
751 	return error;
752 }
753 
754 int
755 sys_mq_getattr(struct lwp *l, void *v, register_t *retval)
756 {
757 	struct sys_mq_getattr_args /* {
758 		syscallarg(mqd_t) mqdes;
759 		syscallarg(struct mq_attr *) mqstat;
760 	} */ *uap = v;
761 	struct file *fp = NULL;
762 	struct mqueue *mq;
763 	struct mq_attr attr;
764 	int error;
765 
766 	/* Get the message queue */
767 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
768 	if (error)
769 		return error;
770 	mq = fp->f_data;
771 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
772 	mutex_exit(&mq->mq_mtx);
773 	FILE_UNUSE(fp, l);
774 
775 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
776 }
777 
778 int
779 sys_mq_setattr(struct lwp *l, void *v, register_t *retval)
780 {
781 	struct sys_mq_setattr_args /* {
782 		syscallarg(mqd_t) mqdes;
783 		syscallarg(const struct mq_attr *) mqstat;
784 		syscallarg(struct mq_attr *) omqstat;
785 	} */ *uap = v;
786 	struct file *fp = NULL;
787 	struct mqueue *mq;
788 	struct mq_attr attr;
789 	int error, nonblock;
790 
791 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
792 	if (error)
793 		return error;
794 	nonblock = (attr.mq_flags & O_NONBLOCK);
795 
796 	/* Get the message queue */
797 	error = mqueue_get(l, SCARG(uap, mqdes), VNOVAL, &fp);
798 	if (error)
799 		return error;
800 	mq = fp->f_data;
801 
802 	/* Copy the old attributes, if needed */
803 	if (SCARG(uap, omqstat))
804 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
805 
806 	/* Ignore everything, except O_NONBLOCK */
807 	if (nonblock)
808 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
809 	else
810 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
811 
812 	mutex_exit(&mq->mq_mtx);
813 	FILE_UNUSE(fp, l);
814 
815 	/*
816 	 * Copy the data to the user-space.
817 	 * Note: According to POSIX, the new attributes should not be set in
818 	 * case of fail - this would be violated.
819 	 */
820 	if (SCARG(uap, omqstat))
821 		error = copyout(&attr, SCARG(uap, omqstat),
822 		    sizeof(struct mq_attr));
823 
824 	return error;
825 }
826 
827 int
828 sys_mq_unlink(struct lwp *l, void *v, register_t *retval)
829 {
830 	struct sys_mq_unlink_args /* {
831 		syscallarg(const char *) name;
832 	} */ *uap = v;
833 	struct mqueue *mq;
834 	char *name;
835 	int error, refcnt = 0;
836 
837 	/* Get the name from the user-space */
838 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
839 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
840 	if (error) {
841 		kmem_free(name, MQ_NAMELEN);
842 		return error;
843 	}
844 
845 	/* Lookup for this file */
846 	mutex_enter(&mqlist_mtx);
847 	mq = mqueue_lookup(name);
848 	if (mq == NULL) {
849 		error = ENOENT;
850 		goto error;
851 	}
852 
853 	/* Check the permissions */
854 	if (mqueue_access(l, mq, VWRITE)) {
855 		mutex_exit(&mq->mq_mtx);
856 		error = EACCES;
857 		goto error;
858 	}
859 
860 	/* Mark message queue as unlinking, before leaving the window */
861 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
862 
863 	/* Wake up all waiters, if there are such */
864 	cv_broadcast(&mq->mq_send_cv);
865 	cv_broadcast(&mq->mq_recv_cv);
866 
867 	refcnt = mq->mq_refcnt;
868 	if (refcnt == 0)
869 		LIST_REMOVE(mq, mq_list);
870 
871 	mutex_exit(&mq->mq_mtx);
872 error:
873 	mutex_exit(&mqlist_mtx);
874 
875 	/*
876 	 * If there are no references - destroy the message
877 	 * queue, otherwise, the last mq_close() will do that.
878 	 */
879 	if (error == 0 && refcnt == 0)
880 		mqueue_destroy(mq);
881 
882 	kmem_free(name, MQ_NAMELEN);
883 	return error;
884 }
885 
886 /*
887  * SysCtl.
888  */
889 
890 SYSCTL_SETUP(sysctl_mqueue_setup, "sysctl mqueue setup")
891 {
892 	const struct sysctlnode *node = NULL;
893 
894 	sysctl_createv(clog, 0, NULL, NULL,
895 		CTLFLAG_PERMANENT,
896 		CTLTYPE_NODE, "kern", NULL,
897 		NULL, 0, NULL, 0,
898 		CTL_KERN, CTL_EOL);
899 	sysctl_createv(clog, 0, NULL, NULL,
900 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
901 		CTLTYPE_INT, "posix_msg",
902 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
903 			     "Message Passing option to which the "
904 			     "system attempts to conform"),
905 		NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
906 		CTL_KERN, CTL_CREATE, CTL_EOL);
907 	sysctl_createv(clog, 0, NULL, &node,
908 		CTLFLAG_PERMANENT,
909 		CTLTYPE_NODE, "mqueue",
910 		SYSCTL_DESCR("Message queue options"),
911 		NULL, 0, NULL, 0,
912 		CTL_KERN, CTL_CREATE, CTL_EOL);
913 
914 	if (node == NULL)
915 		return;
916 
917 	sysctl_createv(clog, 0, &node, NULL,
918 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
919 		CTLTYPE_INT, "mq_open_max",
920 		SYSCTL_DESCR("Maximal number of message queue descriptors "
921 			     "that process could open"),
922 		NULL, 0, &mq_open_max, 0,
923 		CTL_CREATE, CTL_EOL);
924 	sysctl_createv(clog, 0, &node, NULL,
925 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
926 		CTLTYPE_INT, "mq_prio_max",
927 		SYSCTL_DESCR("Maximal priority of the message"),
928 		NULL, 0, &mq_prio_max, 0,
929 		CTL_CREATE, CTL_EOL);
930 	sysctl_createv(clog, 0, &node, NULL,
931 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
932 		CTLTYPE_INT, "mq_max_msgsize",
933 		SYSCTL_DESCR("Maximal allowed size of the message"),
934 		NULL, 0, &mq_max_msgsize, 0,
935 		CTL_CREATE, CTL_EOL);
936 	sysctl_createv(clog, 0, &node, NULL,
937 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
938 		CTLTYPE_INT, "mq_def_maxmsg",
939 		SYSCTL_DESCR("Default maximal message count"),
940 		NULL, 0, &mq_def_maxmsg, 0,
941 		CTL_CREATE, CTL_EOL);
942 }
943 
944 /*
945  * Debugging.
946  */
947 #if defined(DDB)
948 
949 void
950 mqueue_print_list(void (*pr)(const char *, ...))
951 {
952 	struct mqueue *mq;
953 
954 	(*pr)("Global list of the message queues:\n");
955 	(*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
956 	    "Name", "Ptr", "Mode", "Flags",  "Ref",
957 	    "MaxMsg", "MsgSze", "CurMsg");
958 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
959 		(*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
960 		    mq->mq_name, mq, mq->mq_mode,
961 		    mq->mq_attrib.mq_flags, mq->mq_refcnt,
962 		    mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
963 		    mq->mq_attrib.mq_curmsgs);
964 	}
965 }
966 
967 #endif /* defined(DDB) */
968