xref: /netbsd-src/sys/kern/sys_mqueue.c (revision 09afef20633f5fe63d92dfe43ee3a9380dc06883)
1 /*	$NetBSD: sys_mqueue.c,v 1.26 2009/11/01 21:46:09 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 2007-2009 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 1.26 2009/11/01 21:46:09 rmind 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/module.h>
60 #include <sys/mutex.h>
61 #include <sys/pool.h>
62 #include <sys/poll.h>
63 #include <sys/proc.h>
64 #include <sys/queue.h>
65 #include <sys/select.h>
66 #include <sys/signal.h>
67 #include <sys/signalvar.h>
68 #include <sys/stat.h>
69 #include <sys/sysctl.h>
70 #include <sys/syscall.h>
71 #include <sys/syscallvar.h>
72 #include <sys/syscallargs.h>
73 #include <sys/systm.h>
74 #include <sys/unistd.h>
75 
76 #include <miscfs/genfs/genfs.h>
77 
78 MODULE(MODULE_CLASS_MISC, mqueue, NULL);
79 
80 /* System-wide limits. */
81 static u_int			mq_open_max = MQ_OPEN_MAX;
82 static u_int			mq_prio_max = MQ_PRIO_MAX;
83 static u_int			mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
84 static u_int			mq_def_maxmsg = 32;
85 
86 static kmutex_t			mqlist_mtx;
87 static pool_cache_t		mqmsg_cache;
88 static LIST_HEAD(, mqueue)	mqueue_head;
89 
90 static int	mqueue_sysinit(void);
91 static int	mqueue_sysfini(bool);
92 static int	mq_poll_fop(file_t *, int);
93 static int	mq_stat_fop(file_t *, struct stat *);
94 static int	mq_close_fop(file_t *);
95 
96 static const struct fileops mqops = {
97 	.fo_read = fbadop_read,
98 	.fo_write = fbadop_write,
99 	.fo_ioctl = fbadop_ioctl,
100 	.fo_fcntl = fnullop_fcntl,
101 	.fo_poll = mq_poll_fop,
102 	.fo_stat = mq_stat_fop,
103 	.fo_close = mq_close_fop,
104 	.fo_kqfilter = fnullop_kqfilter,
105 	.fo_drain = fnullop_drain,
106 };
107 
108 static const struct syscall_package mqueue_syscalls[] = {
109 	{ SYS_mq_open, 0, (sy_call_t *)sys_mq_open },
110 	{ SYS_mq_close, 0, (sy_call_t *)sys_mq_close },
111 	{ SYS_mq_unlink, 0, (sy_call_t *)sys_mq_unlink },
112 	{ SYS_mq_getattr, 0, (sy_call_t *)sys_mq_getattr },
113 	{ SYS_mq_setattr, 0, (sy_call_t *)sys_mq_setattr },
114 	{ SYS_mq_notify, 0, (sy_call_t *)sys_mq_notify },
115 	{ SYS_mq_send, 0, (sy_call_t *)sys_mq_send },
116 	{ SYS_mq_receive, 0, (sy_call_t *)sys_mq_receive },
117 	{ SYS___mq_timedsend50, 0, (sy_call_t *)sys___mq_timedsend50 },
118 	{ SYS___mq_timedreceive50, 0, (sy_call_t *)sys___mq_timedreceive50 },
119 	{ 0, 0, NULL }
120 };
121 
122 /*
123  * Initialisation and unloading of POSIX message queue subsystem.
124  */
125 
126 static int
127 mqueue_sysinit(void)
128 {
129 	int error;
130 
131 	mqmsg_cache = pool_cache_init(MQ_DEF_MSGSIZE, coherency_unit,
132 	    0, 0, "mqmsgpl", NULL, IPL_NONE, NULL, NULL, NULL);
133 	mutex_init(&mqlist_mtx, MUTEX_DEFAULT, IPL_NONE);
134 	LIST_INIT(&mqueue_head);
135 
136 	error = syscall_establish(NULL, mqueue_syscalls);
137 	if (error) {
138 		(void)mqueue_sysfini(false);
139 	}
140 	return error;
141 }
142 
143 static int
144 mqueue_sysfini(bool interface)
145 {
146 
147 	if (interface) {
148 		int error;
149 		bool inuse;
150 
151 		/* Stop syscall activity. */
152 		error = syscall_disestablish(NULL, mqueue_syscalls);
153 		if (error)
154 			return error;
155 		/*
156 		 * Check if there are any message queues in use.
157 		 * TODO: We shall support forced unload.
158 		 */
159 		mutex_enter(&mqlist_mtx);
160 		inuse = !LIST_EMPTY(&mqueue_head);
161 		mutex_exit(&mqlist_mtx);
162 		if (inuse) {
163 			error = syscall_establish(NULL, mqueue_syscalls);
164 			KASSERT(error == 0);
165 			return EBUSY;
166 		}
167 	}
168 	mutex_destroy(&mqlist_mtx);
169 	pool_cache_destroy(mqmsg_cache);
170 	return 0;
171 }
172 
173 /*
174  * Module interface.
175  */
176 static int
177 mqueue_modcmd(modcmd_t cmd, void *arg)
178 {
179 
180 	switch (cmd) {
181 	case MODULE_CMD_INIT:
182 		return mqueue_sysinit();
183 	case MODULE_CMD_FINI:
184 		return mqueue_sysfini(true);
185 	default:
186 		return ENOTTY;
187 	}
188 }
189 
190 /*
191  * Free the message.
192  */
193 static void
194 mqueue_freemsg(struct mq_msg *msg, const size_t size)
195 {
196 
197 	if (size > MQ_DEF_MSGSIZE) {
198 		kmem_free(msg, size);
199 	} else {
200 		pool_cache_put(mqmsg_cache, msg);
201 	}
202 }
203 
204 /*
205  * Destroy the message queue.
206  */
207 static void
208 mqueue_destroy(struct mqueue *mq)
209 {
210 	struct mq_msg *msg;
211 	size_t msz;
212 	u_int i;
213 
214 	/* Note MQ_PQSIZE + 1. */
215 	for (i = 0; i <= MQ_PQSIZE; i++) {
216 		while ((msg = TAILQ_FIRST(&mq->mq_head[i])) != NULL) {
217 			TAILQ_REMOVE(&mq->mq_head[i], msg, msg_queue);
218 			msz = sizeof(struct mq_msg) + msg->msg_len;
219 			mqueue_freemsg(msg, msz);
220 		}
221 	}
222 	seldestroy(&mq->mq_rsel);
223 	seldestroy(&mq->mq_wsel);
224 	cv_destroy(&mq->mq_send_cv);
225 	cv_destroy(&mq->mq_recv_cv);
226 	mutex_destroy(&mq->mq_mtx);
227 	kmem_free(mq, sizeof(struct mqueue));
228 }
229 
230 /*
231  * Lookup for file name in general list of message queues.
232  *  => locks the message queue
233  */
234 static void *
235 mqueue_lookup(char *name)
236 {
237 	struct mqueue *mq;
238 	KASSERT(mutex_owned(&mqlist_mtx));
239 
240 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
241 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
242 			mutex_enter(&mq->mq_mtx);
243 			return mq;
244 		}
245 	}
246 
247 	return NULL;
248 }
249 
250 /*
251  * mqueue_get: get the mqueue from the descriptor.
252  *  => locks the message queue, if found.
253  *  => holds a reference on the file descriptor.
254  */
255 static int
256 mqueue_get(mqd_t mqd, file_t **fpr)
257 {
258 	struct mqueue *mq;
259 	file_t *fp;
260 
261 	fp = fd_getfile((int)mqd);
262 	if (__predict_false(fp == NULL)) {
263 		return EBADF;
264 	}
265 	if (__predict_false(fp->f_type != DTYPE_MQUEUE)) {
266 		fd_putfile((int)mqd);
267 		return EBADF;
268 	}
269 	mq = fp->f_data;
270 	mutex_enter(&mq->mq_mtx);
271 
272 	*fpr = fp;
273 	return 0;
274 }
275 
276 /*
277  * mqueue_linear_insert: perform linear insert according to the message
278  * priority into the reserved queue (MQ_PQRESQ).  Reserved queue is a
279  * sorted list used only when mq_prio_max is increased via sysctl.
280  */
281 static inline void
282 mqueue_linear_insert(struct mqueue *mq, struct mq_msg *msg)
283 {
284 	struct mq_msg *mit;
285 
286 	TAILQ_FOREACH(mit, &mq->mq_head[MQ_PQRESQ], msg_queue) {
287 		if (msg->msg_prio > mit->msg_prio)
288 			break;
289 	}
290 	if (mit == NULL) {
291 		TAILQ_INSERT_TAIL(&mq->mq_head[MQ_PQRESQ], msg, msg_queue);
292 	} else {
293 		TAILQ_INSERT_BEFORE(mit, msg, msg_queue);
294 	}
295 }
296 
297 static int
298 mq_stat_fop(file_t *fp, struct stat *st)
299 {
300 	struct mqueue *mq = fp->f_data;
301 
302 	memset(st, 0, sizeof(*st));
303 
304 	mutex_enter(&mq->mq_mtx);
305 	st->st_mode = mq->mq_mode;
306 	st->st_uid = mq->mq_euid;
307 	st->st_gid = mq->mq_egid;
308 	st->st_atimespec = mq->mq_atime;
309 	st->st_mtimespec = mq->mq_mtime;
310 	st->st_ctimespec = st->st_birthtimespec = mq->mq_btime;
311 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
312 	st->st_gid = kauth_cred_getegid(fp->f_cred);
313 	mutex_exit(&mq->mq_mtx);
314 
315 	return 0;
316 }
317 
318 static int
319 mq_poll_fop(file_t *fp, int events)
320 {
321 	struct mqueue *mq = fp->f_data;
322 	struct mq_attr *mqattr;
323 	int revents = 0;
324 
325 	mutex_enter(&mq->mq_mtx);
326 	mqattr = &mq->mq_attrib;
327 	if (events & (POLLIN | POLLRDNORM)) {
328 		/* Ready for receiving, if there are messages in the queue */
329 		if (mqattr->mq_curmsgs)
330 			revents |= (POLLIN | POLLRDNORM);
331 		else
332 			selrecord(curlwp, &mq->mq_rsel);
333 	}
334 	if (events & (POLLOUT | POLLWRNORM)) {
335 		/* Ready for sending, if the message queue is not full */
336 		if (mqattr->mq_curmsgs < mqattr->mq_maxmsg)
337 			revents |= (POLLOUT | POLLWRNORM);
338 		else
339 			selrecord(curlwp, &mq->mq_wsel);
340 	}
341 	mutex_exit(&mq->mq_mtx);
342 
343 	return revents;
344 }
345 
346 static int
347 mq_close_fop(file_t *fp)
348 {
349 	struct proc *p = curproc;
350 	struct mqueue *mq = fp->f_data;
351 	bool destroy;
352 
353 	mutex_enter(&mqlist_mtx);
354 	mutex_enter(&mq->mq_mtx);
355 
356 	/* Decrease the counters */
357 	p->p_mqueue_cnt--;
358 	mq->mq_refcnt--;
359 
360 	/* Remove notification if registered for this process */
361 	if (mq->mq_notify_proc == p)
362 		mq->mq_notify_proc = NULL;
363 
364 	/*
365 	 * If this is the last reference and mqueue is marked for unlink,
366 	 * remove and later destroy the message queue.
367 	 */
368 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
369 		LIST_REMOVE(mq, mq_list);
370 		destroy = true;
371 	} else
372 		destroy = false;
373 
374 	mutex_exit(&mq->mq_mtx);
375 	mutex_exit(&mqlist_mtx);
376 
377 	if (destroy)
378 		mqueue_destroy(mq);
379 
380 	return 0;
381 }
382 
383 static int
384 mqueue_access(struct mqueue *mq, mode_t mode, kauth_cred_t cred)
385 {
386 
387 	if (genfs_can_access(VNON, mq->mq_mode, mq->mq_euid,
388 	    mq->mq_egid, mode, cred)) {
389 		return EACCES;
390 	}
391 
392 	return 0;
393 }
394 
395 /*
396  * General mqueue system calls.
397  */
398 
399 int
400 sys_mq_open(struct lwp *l, const struct sys_mq_open_args *uap,
401     register_t *retval)
402 {
403 	/* {
404 		syscallarg(const char *) name;
405 		syscallarg(int) oflag;
406 		syscallarg(mode_t) mode;
407 		syscallarg(struct mq_attr) attr;
408 	} */
409 	struct proc *p = l->l_proc;
410 	struct mqueue *mq, *mq_new = NULL;
411 	file_t *fp;
412 	char *name;
413 	int mqd, error, oflag;
414 
415 	oflag = SCARG(uap, oflag);
416 
417 	/* Get the name from the user-space */
418 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
419 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
420 	if (error) {
421 		kmem_free(name, MQ_NAMELEN);
422 		return error;
423 	}
424 
425 	if (oflag & O_CREAT) {
426 		struct cwdinfo *cwdi = p->p_cwdi;
427 		struct mq_attr attr;
428 		u_int i;
429 
430 		/* Check the limit */
431 		if (p->p_mqueue_cnt == mq_open_max) {
432 			kmem_free(name, MQ_NAMELEN);
433 			return EMFILE;
434 		}
435 
436 		/* Empty name is invalid */
437 		if (name[0] == '\0') {
438 			kmem_free(name, MQ_NAMELEN);
439 			return EINVAL;
440 		}
441 
442 		/* Check for mqueue attributes */
443 		if (SCARG(uap, attr)) {
444 			error = copyin(SCARG(uap, attr), &attr,
445 			    sizeof(struct mq_attr));
446 			if (error) {
447 				kmem_free(name, MQ_NAMELEN);
448 				return error;
449 			}
450 			if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 ||
451 			    attr.mq_msgsize > mq_max_msgsize) {
452 				kmem_free(name, MQ_NAMELEN);
453 				return EINVAL;
454 			}
455 			attr.mq_curmsgs = 0;
456 		} else {
457 			memset(&attr, 0, sizeof(struct mq_attr));
458 			attr.mq_maxmsg = mq_def_maxmsg;
459 			attr.mq_msgsize =
460 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
461 		}
462 
463 		/*
464 		 * Allocate new mqueue, initialize data structures,
465 		 * copy the name, attributes and set the flag.
466 		 */
467 		mq_new = kmem_zalloc(sizeof(struct mqueue), KM_SLEEP);
468 
469 		mutex_init(&mq_new->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
470 		cv_init(&mq_new->mq_send_cv, "mqsendcv");
471 		cv_init(&mq_new->mq_recv_cv, "mqrecvcv");
472 		for (i = 0; i < (MQ_PQSIZE + 1); i++) {
473 			TAILQ_INIT(&mq_new->mq_head[i]);
474 		}
475 		selinit(&mq_new->mq_rsel);
476 		selinit(&mq_new->mq_wsel);
477 
478 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
479 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
480 
481 		CTASSERT((O_MASK & (MQ_UNLINK | MQ_RECEIVE)) == 0);
482 		mq_new->mq_attrib.mq_flags = (O_MASK & oflag);
483 
484 		/* Store mode and effective UID with GID */
485 		mq_new->mq_mode = ((SCARG(uap, mode) &
486 		    ~cwdi->cwdi_cmask) & ALLPERMS) & ~S_ISTXT;
487 		mq_new->mq_euid = kauth_cred_geteuid(l->l_cred);
488 		mq_new->mq_egid = kauth_cred_getegid(l->l_cred);
489 	}
490 
491 	/* Allocate file structure and descriptor */
492 	error = fd_allocfile(&fp, &mqd);
493 	if (error) {
494 		if (mq_new)
495 			mqueue_destroy(mq_new);
496 		kmem_free(name, MQ_NAMELEN);
497 		return error;
498 	}
499 	fp->f_type = DTYPE_MQUEUE;
500 	fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE);
501 	fp->f_ops = &mqops;
502 
503 	/* Look up for mqueue with such name */
504 	mutex_enter(&mqlist_mtx);
505 	mq = mqueue_lookup(name);
506 	if (mq) {
507 		mode_t acc_mode;
508 
509 		KASSERT(mutex_owned(&mq->mq_mtx));
510 
511 		/* Check if mqueue is not marked as unlinking */
512 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
513 			error = EACCES;
514 			goto exit;
515 		}
516 
517 		/* Fail if O_EXCL is set, and mqueue already exists */
518 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
519 			error = EEXIST;
520 			goto exit;
521 		}
522 
523 		/*
524 		 * Check the permissions.  Note the difference between
525 		 * VREAD/VWRITE and FREAD/FWRITE.
526 		 */
527 		acc_mode = 0;
528 		if (fp->f_flag & FREAD) {
529 			acc_mode |= VREAD;
530 		}
531 		if (fp->f_flag & FWRITE) {
532 			acc_mode |= VWRITE;
533 		}
534 		if (mqueue_access(mq, acc_mode, l->l_cred) != 0) {
535 			error = EACCES;
536 			goto exit;
537 		}
538 	} else {
539 		/* Fail if mqueue neither exists, nor we create it */
540 		if ((oflag & O_CREAT) == 0) {
541 			mutex_exit(&mqlist_mtx);
542 			KASSERT(mq_new == NULL);
543 			fd_abort(p, fp, mqd);
544 			kmem_free(name, MQ_NAMELEN);
545 			return ENOENT;
546 		}
547 
548 		/* Check the limit */
549 		if (p->p_mqueue_cnt == mq_open_max) {
550 			error = EMFILE;
551 			goto exit;
552 		}
553 
554 		/* Insert the queue to the list */
555 		mq = mq_new;
556 		mutex_enter(&mq->mq_mtx);
557 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
558 		mq_new = NULL;
559 		getnanotime(&mq->mq_btime);
560 		mq->mq_atime = mq->mq_mtime = mq->mq_btime;
561 	}
562 
563 	/* Increase the counters, and make descriptor ready */
564 	p->p_mqueue_cnt++;
565 	mq->mq_refcnt++;
566 	fp->f_data = mq;
567 exit:
568 	mutex_exit(&mq->mq_mtx);
569 	mutex_exit(&mqlist_mtx);
570 
571 	if (mq_new)
572 		mqueue_destroy(mq_new);
573 	if (error) {
574 		fd_abort(p, fp, mqd);
575 	} else {
576 		fd_affix(p, fp, mqd);
577 		*retval = mqd;
578 	}
579 	kmem_free(name, MQ_NAMELEN);
580 
581 	return error;
582 }
583 
584 int
585 sys_mq_close(struct lwp *l, const struct sys_mq_close_args *uap,
586     register_t *retval)
587 {
588 
589 	return sys_close(l, (const void *)uap, retval);
590 }
591 
592 /*
593  * Primary mq_recv1() function.
594  */
595 int
596 mq_recv1(mqd_t mqdes, void *msg_ptr, size_t msg_len, u_int *msg_prio,
597     struct timespec *ts, ssize_t *mlen)
598 {
599 	file_t *fp = NULL;
600 	struct mqueue *mq;
601 	struct mq_msg *msg = NULL;
602 	struct mq_attr *mqattr;
603 	u_int idx;
604 	int error;
605 
606 	/* Get the message queue */
607 	error = mqueue_get(mqdes, &fp);
608 	if (error) {
609 		return error;
610 	}
611 	mq = fp->f_data;
612 	if ((fp->f_flag & FREAD) == 0) {
613 		error = EBADF;
614 		goto error;
615 	}
616 	getnanotime(&mq->mq_atime);
617 	mqattr = &mq->mq_attrib;
618 
619 	/* Check the message size limits */
620 	if (msg_len < mqattr->mq_msgsize) {
621 		error = EMSGSIZE;
622 		goto error;
623 	}
624 
625 	/* Check if queue is empty */
626 	while (mqattr->mq_curmsgs == 0) {
627 		int t;
628 
629 		if (mqattr->mq_flags & O_NONBLOCK) {
630 			error = EAGAIN;
631 			goto error;
632 		}
633 		error = abstimeout2timo(ts, &t);
634 		if (error) {
635 			goto error;
636 		}
637 		/*
638 		 * Block until someone sends the message.
639 		 * While doing this, notification should not be sent.
640 		 */
641 		mqattr->mq_flags |= MQ_RECEIVE;
642 		error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
643 		mqattr->mq_flags &= ~MQ_RECEIVE;
644 		if (error || (mqattr->mq_flags & MQ_UNLINK)) {
645 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
646 			goto error;
647 		}
648 	}
649 
650 	/*
651 	 * Find the highest priority message, and remove it from the queue.
652 	 * At first, reserved queue is checked, bitmap is next.
653 	 */
654 	msg = TAILQ_FIRST(&mq->mq_head[MQ_PQRESQ]);
655 	if (__predict_true(msg == NULL)) {
656 		idx = ffs(mq->mq_bitmap);
657 		msg = TAILQ_FIRST(&mq->mq_head[idx]);
658 		KASSERT(msg != NULL);
659 	} else {
660 		idx = MQ_PQRESQ;
661 	}
662 	TAILQ_REMOVE(&mq->mq_head[idx], msg, msg_queue);
663 
664 	/* Unmark the bit, if last message. */
665 	if (__predict_true(idx) && TAILQ_EMPTY(&mq->mq_head[idx])) {
666 		KASSERT((MQ_PQSIZE - idx) == msg->msg_prio);
667 		mq->mq_bitmap &= ~(1 << --idx);
668 	}
669 
670 	/* Decrement the counter and signal waiter, if any */
671 	mqattr->mq_curmsgs--;
672 	cv_signal(&mq->mq_recv_cv);
673 
674 	/* Ready for sending now */
675 	selnotify(&mq->mq_wsel, POLLOUT | POLLWRNORM, 0);
676 error:
677 	mutex_exit(&mq->mq_mtx);
678 	fd_putfile((int)mqdes);
679 	if (error)
680 		return error;
681 
682 	/*
683 	 * Copy the data to the user-space.
684 	 * Note: According to POSIX, no message should be removed from the
685 	 * queue in case of fail - this would be violated.
686 	 */
687 	*mlen = msg->msg_len;
688 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
689 	if (error == 0 && msg_prio)
690 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
691 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
692 
693 	return error;
694 }
695 
696 int
697 sys_mq_receive(struct lwp *l, const struct sys_mq_receive_args *uap,
698     register_t *retval)
699 {
700 	/* {
701 		syscallarg(mqd_t) mqdes;
702 		syscallarg(char *) msg_ptr;
703 		syscallarg(size_t) msg_len;
704 		syscallarg(unsigned *) msg_prio;
705 	} */
706 	ssize_t mlen;
707 	int error;
708 
709 	error = mq_recv1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
710 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), NULL, &mlen);
711 	if (error == 0)
712 		*retval = mlen;
713 
714 	return error;
715 }
716 
717 int
718 sys___mq_timedreceive50(struct lwp *l,
719     const struct sys___mq_timedreceive50_args *uap, register_t *retval)
720 {
721 	/* {
722 		syscallarg(mqd_t) mqdes;
723 		syscallarg(char *) msg_ptr;
724 		syscallarg(size_t) msg_len;
725 		syscallarg(unsigned *) msg_prio;
726 		syscallarg(const struct timespec *) abs_timeout;
727 	} */
728 	struct timespec ts, *tsp;
729 	ssize_t mlen;
730 	int error;
731 
732 	/* Get and convert time value */
733 	if (SCARG(uap, abs_timeout)) {
734 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
735 		if (error)
736 			return error;
737 		tsp = &ts;
738 	} else {
739 		tsp = NULL;
740 	}
741 
742 	error = mq_recv1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
743 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp, &mlen);
744 	if (error == 0)
745 		*retval = mlen;
746 
747 	return error;
748 }
749 
750 /*
751  * Primary mq_send1() function.
752  */
753 int
754 mq_send1(mqd_t mqdes, const char *msg_ptr, size_t msg_len, u_int msg_prio,
755     struct timespec *ts)
756 {
757 	file_t *fp = NULL;
758 	struct mqueue *mq;
759 	struct mq_msg *msg;
760 	struct mq_attr *mqattr;
761 	struct proc *notify = NULL;
762 	ksiginfo_t ksi;
763 	size_t size;
764 	int error;
765 
766 	/* Check the priority range */
767 	if (msg_prio >= mq_prio_max)
768 		return EINVAL;
769 
770 	/* Allocate a new message */
771 	size = sizeof(struct mq_msg) + msg_len;
772 	if (size > mq_max_msgsize)
773 		return EMSGSIZE;
774 
775 	if (size > MQ_DEF_MSGSIZE) {
776 		msg = kmem_alloc(size, KM_SLEEP);
777 	} else {
778 		msg = pool_cache_get(mqmsg_cache, PR_WAITOK);
779 	}
780 
781 	/* Get the data from user-space */
782 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
783 	if (error) {
784 		mqueue_freemsg(msg, size);
785 		return error;
786 	}
787 	msg->msg_len = msg_len;
788 	msg->msg_prio = msg_prio;
789 
790 	/* Get the mqueue */
791 	error = mqueue_get(mqdes, &fp);
792 	if (error) {
793 		mqueue_freemsg(msg, size);
794 		return error;
795 	}
796 	mq = fp->f_data;
797 	if ((fp->f_flag & FWRITE) == 0) {
798 		error = EBADF;
799 		goto error;
800 	}
801 	getnanotime(&mq->mq_mtime);
802 	mqattr = &mq->mq_attrib;
803 
804 	/* Check the message size limit */
805 	if (msg_len <= 0 || msg_len > mqattr->mq_msgsize) {
806 		error = EMSGSIZE;
807 		goto error;
808 	}
809 
810 	/* Check if queue is full */
811 	while (mqattr->mq_curmsgs >= mqattr->mq_maxmsg) {
812 		int t;
813 
814 		if (mqattr->mq_flags & O_NONBLOCK) {
815 			error = EAGAIN;
816 			goto error;
817 		}
818 		error = abstimeout2timo(ts, &t);
819 		if (error) {
820 			goto error;
821 		}
822 		/* Block until queue becomes available */
823 		error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
824 		if (error || (mqattr->mq_flags & MQ_UNLINK)) {
825 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
826 			goto error;
827 		}
828 	}
829 	KASSERT(mqattr->mq_curmsgs < mqattr->mq_maxmsg);
830 
831 	/*
832 	 * Insert message into the queue, according to the priority.
833 	 * Note the difference between index and priority.
834 	 */
835 	if (__predict_true(msg_prio < MQ_PQSIZE)) {
836 		u_int idx = MQ_PQSIZE - msg_prio;
837 
838 		KASSERT(idx != MQ_PQRESQ);
839 		TAILQ_INSERT_TAIL(&mq->mq_head[idx], msg, msg_queue);
840 		mq->mq_bitmap |= (1 << --idx);
841 	} else {
842 		mqueue_linear_insert(mq, msg);
843 	}
844 
845 	/* Check for the notify */
846 	if (mqattr->mq_curmsgs == 0 && mq->mq_notify_proc &&
847 	    (mqattr->mq_flags & MQ_RECEIVE) == 0) {
848 		/* Initialize the signal */
849 		KSI_INIT(&ksi);
850 		ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
851 		ksi.ksi_code = SI_MESGQ;
852 		ksi.ksi_value = mq->mq_sig_notify.sigev_value;
853 		/* Unregister the process */
854 		notify = mq->mq_notify_proc;
855 		mq->mq_notify_proc = NULL;
856 	}
857 
858 	/* Increment the counter and signal waiter, if any */
859 	mqattr->mq_curmsgs++;
860 	cv_signal(&mq->mq_send_cv);
861 
862 	/* Ready for receiving now */
863 	selnotify(&mq->mq_rsel, POLLIN | POLLRDNORM, 0);
864 error:
865 	mutex_exit(&mq->mq_mtx);
866 	fd_putfile((int)mqdes);
867 
868 	if (error) {
869 		mqueue_freemsg(msg, size);
870 	} else if (notify) {
871 		/* Send the notify, if needed */
872 		mutex_enter(proc_lock);
873 		kpsignal(notify, &ksi, NULL);
874 		mutex_exit(proc_lock);
875 	}
876 	return error;
877 }
878 
879 int
880 sys_mq_send(struct lwp *l, const struct sys_mq_send_args *uap,
881     register_t *retval)
882 {
883 	/* {
884 		syscallarg(mqd_t) mqdes;
885 		syscallarg(const char *) msg_ptr;
886 		syscallarg(size_t) msg_len;
887 		syscallarg(unsigned) msg_prio;
888 	} */
889 
890 	return mq_send1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
891 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), NULL);
892 }
893 
894 int
895 sys___mq_timedsend50(struct lwp *l, const struct sys___mq_timedsend50_args *uap,
896     register_t *retval)
897 {
898 	/* {
899 		syscallarg(mqd_t) mqdes;
900 		syscallarg(const char *) msg_ptr;
901 		syscallarg(size_t) msg_len;
902 		syscallarg(unsigned) msg_prio;
903 		syscallarg(const struct timespec *) abs_timeout;
904 	} */
905 	struct timespec ts, *tsp;
906 	int error;
907 
908 	/* Get and convert time value */
909 	if (SCARG(uap, abs_timeout)) {
910 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
911 		if (error)
912 			return error;
913 		tsp = &ts;
914 	} else {
915 		tsp = NULL;
916 	}
917 
918 	return mq_send1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
919 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp);
920 }
921 
922 int
923 sys_mq_notify(struct lwp *l, const struct sys_mq_notify_args *uap,
924     register_t *retval)
925 {
926 	/* {
927 		syscallarg(mqd_t) mqdes;
928 		syscallarg(const struct sigevent *) notification;
929 	} */
930 	file_t *fp = NULL;
931 	struct mqueue *mq;
932 	struct sigevent sig;
933 	int error;
934 
935 	if (SCARG(uap, notification)) {
936 		/* Get the signal from user-space */
937 		error = copyin(SCARG(uap, notification), &sig,
938 		    sizeof(struct sigevent));
939 		if (error)
940 			return error;
941 	}
942 
943 	error = mqueue_get(SCARG(uap, mqdes), &fp);
944 	if (error)
945 		return error;
946 	mq = fp->f_data;
947 
948 	if (SCARG(uap, notification)) {
949 		/* Register notification: set the signal and target process */
950 		if (mq->mq_notify_proc == NULL) {
951 			memcpy(&mq->mq_sig_notify, &sig,
952 			    sizeof(struct sigevent));
953 			mq->mq_notify_proc = l->l_proc;
954 		} else {
955 			/* Fail if someone else already registered */
956 			error = EBUSY;
957 		}
958 	} else {
959 		/* Unregister the notification */
960 		mq->mq_notify_proc = NULL;
961 	}
962 	mutex_exit(&mq->mq_mtx);
963 	fd_putfile((int)SCARG(uap, mqdes));
964 
965 	return error;
966 }
967 
968 int
969 sys_mq_getattr(struct lwp *l, const struct sys_mq_getattr_args *uap,
970     register_t *retval)
971 {
972 	/* {
973 		syscallarg(mqd_t) mqdes;
974 		syscallarg(struct mq_attr *) mqstat;
975 	} */
976 	file_t *fp = NULL;
977 	struct mqueue *mq;
978 	struct mq_attr attr;
979 	int error;
980 
981 	/* Get the message queue */
982 	error = mqueue_get(SCARG(uap, mqdes), &fp);
983 	if (error)
984 		return error;
985 	mq = fp->f_data;
986 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
987 	mutex_exit(&mq->mq_mtx);
988 	fd_putfile((int)SCARG(uap, mqdes));
989 
990 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
991 }
992 
993 int
994 sys_mq_setattr(struct lwp *l, const struct sys_mq_setattr_args *uap,
995     register_t *retval)
996 {
997 	/* {
998 		syscallarg(mqd_t) mqdes;
999 		syscallarg(const struct mq_attr *) mqstat;
1000 		syscallarg(struct mq_attr *) omqstat;
1001 	} */
1002 	file_t *fp = NULL;
1003 	struct mqueue *mq;
1004 	struct mq_attr attr;
1005 	int error, nonblock;
1006 
1007 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
1008 	if (error)
1009 		return error;
1010 	nonblock = (attr.mq_flags & O_NONBLOCK);
1011 
1012 	/* Get the message queue */
1013 	error = mqueue_get(SCARG(uap, mqdes), &fp);
1014 	if (error)
1015 		return error;
1016 	mq = fp->f_data;
1017 
1018 	/* Copy the old attributes, if needed */
1019 	if (SCARG(uap, omqstat)) {
1020 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
1021 	}
1022 
1023 	/* Ignore everything, except O_NONBLOCK */
1024 	if (nonblock)
1025 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
1026 	else
1027 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
1028 
1029 	mutex_exit(&mq->mq_mtx);
1030 	fd_putfile((int)SCARG(uap, mqdes));
1031 
1032 	/*
1033 	 * Copy the data to the user-space.
1034 	 * Note: According to POSIX, the new attributes should not be set in
1035 	 * case of fail - this would be violated.
1036 	 */
1037 	if (SCARG(uap, omqstat))
1038 		error = copyout(&attr, SCARG(uap, omqstat),
1039 		    sizeof(struct mq_attr));
1040 
1041 	return error;
1042 }
1043 
1044 int
1045 sys_mq_unlink(struct lwp *l, const struct sys_mq_unlink_args *uap,
1046     register_t *retval)
1047 {
1048 	/* {
1049 		syscallarg(const char *) name;
1050 	} */
1051 	struct mqueue *mq;
1052 	char *name;
1053 	int error, refcnt = 0;
1054 
1055 	/* Get the name from the user-space */
1056 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
1057 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
1058 	if (error) {
1059 		kmem_free(name, MQ_NAMELEN);
1060 		return error;
1061 	}
1062 
1063 	/* Lookup for this file */
1064 	mutex_enter(&mqlist_mtx);
1065 	mq = mqueue_lookup(name);
1066 	if (mq == NULL) {
1067 		error = ENOENT;
1068 		goto error;
1069 	}
1070 
1071 	/* Check the permissions */
1072 	if (kauth_cred_geteuid(l->l_cred) != mq->mq_euid &&
1073 	    kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, NULL)) {
1074 		mutex_exit(&mq->mq_mtx);
1075 		error = EACCES;
1076 		goto error;
1077 	}
1078 
1079 	/* Mark message queue as unlinking, before leaving the window */
1080 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
1081 
1082 	/* Wake up all waiters, if there are such */
1083 	cv_broadcast(&mq->mq_send_cv);
1084 	cv_broadcast(&mq->mq_recv_cv);
1085 
1086 	selnotify(&mq->mq_rsel, POLLHUP, 0);
1087 	selnotify(&mq->mq_wsel, POLLHUP, 0);
1088 
1089 	refcnt = mq->mq_refcnt;
1090 	if (refcnt == 0)
1091 		LIST_REMOVE(mq, mq_list);
1092 
1093 	mutex_exit(&mq->mq_mtx);
1094 error:
1095 	mutex_exit(&mqlist_mtx);
1096 
1097 	/*
1098 	 * If there are no references - destroy the message
1099 	 * queue, otherwise, the last mq_close() will do that.
1100 	 */
1101 	if (error == 0 && refcnt == 0)
1102 		mqueue_destroy(mq);
1103 
1104 	kmem_free(name, MQ_NAMELEN);
1105 	return error;
1106 }
1107 
1108 /*
1109  * System control nodes.
1110  */
1111 
1112 SYSCTL_SETUP(sysctl_mqueue_setup, "sysctl mqueue setup")
1113 {
1114 	const struct sysctlnode *node = NULL;
1115 
1116 	sysctl_createv(clog, 0, NULL, NULL,
1117 		CTLFLAG_PERMANENT,
1118 		CTLTYPE_NODE, "kern", NULL,
1119 		NULL, 0, NULL, 0,
1120 		CTL_KERN, CTL_EOL);
1121 	sysctl_createv(clog, 0, NULL, NULL,
1122 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
1123 		CTLTYPE_INT, "posix_msg",
1124 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
1125 			     "Message Passing option to which the "
1126 			     "system attempts to conform"),
1127 		NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
1128 		CTL_KERN, CTL_CREATE, CTL_EOL);
1129 	sysctl_createv(clog, 0, NULL, &node,
1130 		CTLFLAG_PERMANENT,
1131 		CTLTYPE_NODE, "mqueue",
1132 		SYSCTL_DESCR("Message queue options"),
1133 		NULL, 0, NULL, 0,
1134 		CTL_KERN, CTL_CREATE, CTL_EOL);
1135 
1136 	if (node == NULL)
1137 		return;
1138 
1139 	sysctl_createv(clog, 0, &node, NULL,
1140 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1141 		CTLTYPE_INT, "mq_open_max",
1142 		SYSCTL_DESCR("Maximal number of message queue descriptors "
1143 			     "that process could open"),
1144 		NULL, 0, &mq_open_max, 0,
1145 		CTL_CREATE, CTL_EOL);
1146 	sysctl_createv(clog, 0, &node, NULL,
1147 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1148 		CTLTYPE_INT, "mq_prio_max",
1149 		SYSCTL_DESCR("Maximal priority of the message"),
1150 		NULL, 0, &mq_prio_max, 0,
1151 		CTL_CREATE, CTL_EOL);
1152 	sysctl_createv(clog, 0, &node, NULL,
1153 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1154 		CTLTYPE_INT, "mq_max_msgsize",
1155 		SYSCTL_DESCR("Maximal allowed size of the message"),
1156 		NULL, 0, &mq_max_msgsize, 0,
1157 		CTL_CREATE, CTL_EOL);
1158 	sysctl_createv(clog, 0, &node, NULL,
1159 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1160 		CTLTYPE_INT, "mq_def_maxmsg",
1161 		SYSCTL_DESCR("Default maximal message count"),
1162 		NULL, 0, &mq_def_maxmsg, 0,
1163 		CTL_CREATE, CTL_EOL);
1164 }
1165 
1166 /*
1167  * Debugging.
1168  */
1169 #if defined(DDB)
1170 
1171 void
1172 mqueue_print_list(void (*pr)(const char *, ...))
1173 {
1174 	struct mqueue *mq;
1175 
1176 	(*pr)("Global list of the message queues:\n");
1177 	(*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
1178 	    "Name", "Ptr", "Mode", "Flags",  "Ref",
1179 	    "MaxMsg", "MsgSze", "CurMsg");
1180 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
1181 		(*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
1182 		    mq->mq_name, mq, mq->mq_mode,
1183 		    mq->mq_attrib.mq_flags, mq->mq_refcnt,
1184 		    mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
1185 		    mq->mq_attrib.mq_curmsgs);
1186 	}
1187 }
1188 
1189 #endif /* defined(DDB) */
1190