xref: /dragonfly/sys/kern/sys_mqueue.c (revision 2b7dbe20)
1 /*	$NetBSD: sys_mqueue.c,v 1.16 2009/04/11 23:05:26 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/param.h>
45 #include <sys/types.h>
46 #include <sys/errno.h>
47 #include <sys/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filedesc.h>
50 #include <sys/ucred.h>
51 #include <sys/priv.h>
52 #include <sys/kernel.h>
53 #include <sys/malloc.h>
54 #include <sys/mqueue.h>
55 #include <sys/proc.h>
56 #include <sys/queue.h>
57 #include <sys/event.h>
58 #include <sys/serialize.h>
59 #include <sys/signal.h>
60 #include <sys/signalvar.h>
61 #include <sys/spinlock.h>
62 #include <sys/spinlock2.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysmsg.h>
66 #include <sys/systm.h>
67 #include <sys/lock.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 
71 /* System-wide limits. */
72 static u_int			mq_open_max = MQ_OPEN_MAX;
73 static u_int			mq_prio_max = MQ_PRIO_MAX;
74 static u_int			mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
75 static u_int			mq_def_maxmsg = 32;
76 static u_int			mq_max_maxmsg = 16 * 32;
77 
78 struct lock			mqlist_mtx;
79 static LIST_HEAD(, mqueue)	mqueue_head =
80 	LIST_HEAD_INITIALIZER(mqueue_head);
81 
82 typedef struct	file file_t;	/* XXX: Should we put this in sys/types.h ? */
83 
84 /* Function prototypes */
85 static int	mq_stat_fop(file_t *, struct stat *, struct ucred *cred);
86 static int	mq_close_fop(file_t *);
87 static int	mq_kqfilter_fop(struct file *fp, struct knote *kn);
88 static void	mqfilter_read_detach(struct knote *kn);
89 static void	mqfilter_write_detach(struct knote *kn);
90 static int	mqfilter_read(struct knote *kn, long hint);
91 static int	mqfilter_write(struct knote *kn, long hint);
92 
93 /* Some time-related utility functions */
94 static int	tstohz(const struct timespec *ts);
95 
96 /* File operations vector */
97 static struct fileops mqops = {
98 	.fo_read = badfo_readwrite,
99 	.fo_write = badfo_readwrite,
100 	.fo_ioctl = badfo_ioctl,
101 	.fo_stat = mq_stat_fop,
102 	.fo_close = mq_close_fop,
103 	.fo_kqfilter = mq_kqfilter_fop,
104 	.fo_shutdown = badfo_shutdown
105 };
106 
107 /* Define a new malloc type for message queues */
108 MALLOC_DECLARE(M_MQBUF);
109 MALLOC_DEFINE(M_MQBUF, "mqueues", "Buffers to message queues");
110 
111 /*
112  * Initialize POSIX message queue subsystem.
113  */
114 void
115 mqueue_sysinit(void)
116 {
117 	lockinit(&mqlist_mtx, "mqlist_mtx", 0, LK_CANRECURSE);
118 }
119 
120 /*
121  * Free the message.
122  */
123 static void
124 mqueue_freemsg(struct mq_msg *msg, const size_t size)
125 {
126 	kfree(msg, M_MQBUF);
127 }
128 
129 /*
130  * Destroy the message queue.
131  */
132 static void
133 mqueue_destroy(struct mqueue *mq)
134 {
135 	struct mq_msg *msg;
136 	size_t msz;
137 	u_int i;
138 
139 	/* Note MQ_PQSIZE + 1. */
140 	for (i = 0; i < MQ_PQSIZE + 1; i++) {
141 		while ((msg = TAILQ_FIRST(&mq->mq_head[i])) != NULL) {
142 			TAILQ_REMOVE(&mq->mq_head[i], msg, msg_queue);
143 			msz = sizeof(struct mq_msg) + msg->msg_len;
144 			mqueue_freemsg(msg, msz);
145 		}
146 	}
147 	lockuninit(&mq->mq_mtx);
148 	kfree(mq, M_MQBUF);
149 }
150 
151 /*
152  * Lookup for file name in general list of message queues.
153  *  => locks the message queue
154  */
155 static void *
156 mqueue_lookup(char *name)
157 {
158 	struct mqueue *mq;
159 
160 	KKASSERT(lockstatus(&mqlist_mtx, curthread));
161 
162 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
163 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
164 			lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
165 			return mq;
166 		}
167 	}
168 
169 	return NULL;
170 }
171 
172 /*
173  * mqueue_get: get the mqueue from the descriptor.
174  *  => locks the message queue, if found.
175  *  => holds a reference on the file descriptor.
176  */
177 static int
178 mqueue_get(struct lwp *l, mqd_t mqd, file_t **fpr)
179 {
180 	struct mqueue *mq;
181 	file_t *fp;
182 
183 	fp = holdfp(curthread, (int)mqd, -1);	/* XXX: Why -1 ? */
184 	if (__predict_false(fp == NULL))
185 		return EBADF;
186 
187 	if (__predict_false(fp->f_type != DTYPE_MQUEUE)) {
188 		fdrop(fp);
189 		return EBADF;
190 	}
191 	mq = fp->f_data;
192 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
193 
194 	*fpr = fp;
195 	return 0;
196 }
197 
198 /*
199  * mqueue_linear_insert: perform linear insert according to the message
200  * priority into the reserved queue (MQ_PQRESQ).  Reserved queue is a
201  * sorted list used only when mq_prio_max is increased via sysctl.
202  */
203 static inline void
204 mqueue_linear_insert(struct mqueue *mq, struct mq_msg *msg)
205 {
206 	struct mq_msg *mit;
207 
208 	TAILQ_FOREACH(mit, &mq->mq_head[MQ_PQRESQ], msg_queue) {
209 		if (msg->msg_prio > mit->msg_prio)
210 			break;
211 	}
212 	if (mit == NULL) {
213 		TAILQ_INSERT_TAIL(&mq->mq_head[MQ_PQRESQ], msg, msg_queue);
214 	} else {
215 		TAILQ_INSERT_BEFORE(mit, msg, msg_queue);
216 	}
217 }
218 
219 /*
220  * Compute number of ticks in the specified amount of time.
221  */
222 static int
223 tstohz(const struct timespec *ts)
224 {
225 	struct timeval tv;
226 
227 	/*
228 	 * usec has great enough resolution for hz, so convert to a
229 	 * timeval and use tvtohz() above.
230 	 */
231 	TIMESPEC_TO_TIMEVAL(&tv, ts);
232 	return tvtohz_high(&tv);	/* XXX Why _high() and not _low() ? */
233 }
234 
235 /*
236  * Converter from struct timespec to the ticks.
237  * Used by mq_timedreceive(), mq_timedsend().
238  */
239 int
240 abstimeout2timo(struct timespec *ts, int *timo)
241 {
242 	struct timespec tsd;
243 	int error;
244 
245 	error = itimespecfix(ts);
246 	if (error) {
247 		return error;
248 	}
249 	getnanotime(&tsd);
250 	timespecsub(ts, &tsd, ts);
251 	if (ts->tv_sec < 0 || (ts->tv_sec == 0 && ts->tv_nsec <= 0)) {
252 		return ETIMEDOUT;
253 	}
254 	*timo = tstohz(ts);
255 	KKASSERT(*timo != 0);
256 
257 	return 0;
258 }
259 
260 static int
261 mq_stat_fop(file_t *fp, struct stat *st, struct ucred *cred)
262 {
263 	struct mqueue *mq = fp->f_data;
264 
265 	(void)memset(st, 0, sizeof(*st));
266 
267 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
268 	st->st_mode = mq->mq_mode;
269 	st->st_uid = mq->mq_euid;
270 	st->st_gid = mq->mq_egid;
271 	st->st_atimespec = mq->mq_atime;
272 	st->st_mtimespec = mq->mq_mtime;
273 	/*st->st_ctimespec = st->st_birthtimespec = mq->mq_btime;*/
274 	st->st_uid = fp->f_cred->cr_uid;
275 	st->st_gid = fp->f_cred->cr_svgid;
276 	lockmgr(&mq->mq_mtx, LK_RELEASE);
277 
278 	return 0;
279 }
280 
281 static struct filterops mqfiltops_read =
282 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, mqfilter_read_detach, mqfilter_read };
283 static struct filterops mqfiltops_write =
284 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, mqfilter_write_detach, mqfilter_write };
285 
286 static int
287 mq_kqfilter_fop(struct file *fp, struct knote *kn)
288 {
289 	struct mqueue *mq = fp->f_data;
290 	struct klist *klist;
291 
292 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
293 
294 	switch (kn->kn_filter) {
295 	case EVFILT_READ:
296 		kn->kn_fop = &mqfiltops_read;
297 		kn->kn_hook = (caddr_t)mq;
298 		klist = &mq->mq_rkq.ki_note;
299 		break;
300 	case EVFILT_WRITE:
301 		kn->kn_fop = &mqfiltops_write;
302 		kn->kn_hook = (caddr_t)mq;
303 		klist = &mq->mq_wkq.ki_note;
304 		break;
305 	default:
306 		lockmgr(&mq->mq_mtx, LK_RELEASE);
307 		return (EOPNOTSUPP);
308 	}
309 
310 	knote_insert(klist, kn);
311 	lockmgr(&mq->mq_mtx, LK_RELEASE);
312 
313 	return (0);
314 }
315 
316 static void
317 mqfilter_read_detach(struct knote *kn)
318 {
319 	struct mqueue *mq = (struct mqueue *)kn->kn_hook;
320 
321 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
322 	struct klist *klist = &mq->mq_rkq.ki_note;
323 	knote_remove(klist, kn);
324 	lockmgr(&mq->mq_mtx, LK_RELEASE);
325 }
326 
327 static void
328 mqfilter_write_detach(struct knote *kn)
329 {
330 	struct mqueue *mq = (struct mqueue *)kn->kn_hook;
331 
332 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
333 	struct klist *klist = &mq->mq_wkq.ki_note;
334 	knote_remove(klist, kn);
335 	lockmgr(&mq->mq_mtx, LK_RELEASE);
336 }
337 
338 static int
339 mqfilter_read(struct knote *kn, long hint)
340 {
341 	struct mqueue *mq = (struct mqueue *)kn->kn_hook;
342 	struct mq_attr *mqattr;
343 	int ready = 0;
344 
345 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
346 	mqattr = &mq->mq_attrib;
347 	/* Ready for receiving, if there are messages in the queue */
348 	if (mqattr->mq_curmsgs)
349 		ready = 1;
350 	lockmgr(&mq->mq_mtx, LK_RELEASE);
351 
352 	return (ready);
353 }
354 
355 static int
356 mqfilter_write(struct knote *kn, long hint)
357 {
358 	struct mqueue *mq = (struct mqueue *)kn->kn_hook;
359 	struct mq_attr *mqattr;
360 	int ready = 0;
361 
362 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
363 	mqattr = &mq->mq_attrib;
364 	/* Ready for sending, if the message queue is not full */
365 	if (mqattr->mq_curmsgs < mqattr->mq_maxmsg)
366 		ready = 1;
367 	lockmgr(&mq->mq_mtx, LK_RELEASE);
368 
369 	return (ready);
370 }
371 
372 static int
373 mq_close_fop(file_t *fp)
374 {
375 	struct proc *p = curproc;
376 	struct mqueue *mq = fp->f_data;
377 	bool destroy;
378 
379 	lockmgr(&mqlist_mtx, LK_EXCLUSIVE);
380 	lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
381 
382 	/* Decrease the counters */
383 	p->p_mqueue_cnt--;
384 	mq->mq_refcnt--;
385 
386 	/* Remove notification if registered for this process */
387 	if (mq->mq_notify_proc == p)
388 		mq->mq_notify_proc = NULL;
389 
390 	/*
391 	 * If this is the last reference and mqueue is marked for unlink,
392 	 * remove and later destroy the message queue.
393 	 */
394 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
395 		LIST_REMOVE(mq, mq_list);
396 		destroy = true;
397 	} else
398 		destroy = false;
399 
400 	lockmgr(&mq->mq_mtx, LK_RELEASE);
401 	lockmgr(&mqlist_mtx, LK_RELEASE);
402 
403 	if (destroy)
404 		mqueue_destroy(mq);
405 
406 	return 0;
407 }
408 
409 /*
410  * General mqueue system calls.
411  */
412 
413 int
414 sys_mq_open(struct sysmsg *sysmsg, const struct mq_open_args *uap)
415 {
416 	/* {
417 		syscallarg(const char *) name;
418 		syscallarg(int) oflag;
419 		syscallarg(mode_t) mode;
420 		syscallarg(struct mq_attr) attr;
421 	} */
422 	struct thread *td = curthread;
423 	struct proc *p = td->td_proc;
424 	struct filedesc *fdp = p->p_fd;
425 	struct mqueue *mq, *mq_new = NULL;
426 	file_t *fp;
427 	char *name;
428 	int mqd, error, oflag;
429 
430 	/* Check access mode flags */
431 	oflag = SCARG(uap, oflag);
432 	if ((oflag & O_ACCMODE) == (O_WRONLY | O_RDWR)) {
433 		return EINVAL;
434 	}
435 
436 	/* Get the name from the user-space */
437 	name = kmalloc(MQ_NAMELEN, M_MQBUF, M_WAITOK | M_ZERO | M_NULLOK);
438 	if (name == NULL)
439 		return (ENOMEM);
440 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
441 	if (error) {
442 		kfree(name, M_MQBUF);
443 		return error;
444 	}
445 
446 	if (oflag & O_CREAT) {
447 		struct mq_attr attr;
448 		u_int i;
449 
450 		/* Check the limit */
451 		if (p->p_mqueue_cnt == mq_open_max) {
452 			kfree(name, M_MQBUF);
453 			return EMFILE;
454 		}
455 
456 		/* Empty name is invalid */
457 		if (name[0] == '\0') {
458 			kfree(name, M_MQBUF);
459 			return EINVAL;
460 		}
461 
462 		/* Check for mqueue attributes */
463 		if (SCARG(uap, attr)) {
464 			error = copyin(SCARG(uap, attr), &attr,
465 				sizeof(struct mq_attr));
466 			if (error) {
467 				kfree(name, M_MQBUF);
468 				return error;
469 			}
470 			if (attr.mq_maxmsg <= 0 ||
471 			    attr.mq_maxmsg > mq_max_maxmsg ||
472 			    attr.mq_msgsize <= 0 ||
473 			    attr.mq_msgsize > mq_max_msgsize) {
474 				kfree(name, M_MQBUF);
475 				return EINVAL;
476 			}
477 			attr.mq_curmsgs = 0;
478 		} else {
479 			memset(&attr, 0, sizeof(struct mq_attr));
480 			attr.mq_maxmsg = mq_def_maxmsg;
481 			attr.mq_msgsize =
482 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
483 		}
484 
485 		/*
486 		 * Allocate new mqueue, initialize data structures,
487 		 * copy the name, attributes and set the flag.
488 		 */
489 		mq_new = kmalloc(sizeof(struct mqueue), M_MQBUF,
490 					M_WAITOK | M_ZERO | M_NULLOK);
491 		if (mq_new == NULL) {
492 			kfree(name, M_MQBUF);
493 			return (ENOMEM);
494 		}
495 
496 		lockinit(&mq_new->mq_mtx, "mq_new->mq_mtx", 0, LK_CANRECURSE);
497 		for (i = 0; i < (MQ_PQSIZE + 1); i++) {
498 			TAILQ_INIT(&mq_new->mq_head[i]);
499 		}
500 
501 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
502 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
503 
504 		/*CTASSERT((O_MASK & (MQ_UNLINK | MQ_RECEIVE)) == 0);*/
505 		/* mq_new->mq_attrib.mq_flags = (O_MASK & oflag); */
506 		mq_new->mq_attrib.mq_flags = oflag;
507 
508 		/* Store mode and effective UID with GID */
509 		mq_new->mq_mode = ((SCARG(uap, mode) &
510 		    ~p->p_fd->fd_cmask) & ALLPERMS) & ~S_ISTXT;
511 		mq_new->mq_euid = td->td_ucred->cr_uid;
512 		mq_new->mq_egid = td->td_ucred->cr_svgid;
513 	}
514 
515 	/* Allocate file structure and descriptor */
516 	error = falloc(td->td_lwp, &fp, &mqd);
517 	if (error) {
518 		if (mq_new)
519 			mqueue_destroy(mq_new);
520 		kfree(name, M_MQBUF);
521 		return error;
522 	}
523 	fp->f_type = DTYPE_MQUEUE;
524 	fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE);
525 	fp->f_ops = &mqops;
526 
527 	/* Look up for mqueue with such name */
528 	lockmgr(&mqlist_mtx, LK_EXCLUSIVE);
529 	mq = mqueue_lookup(name);
530 	if (mq) {
531 		int acc_mode;
532 
533 		KKASSERT(lockstatus(&mq->mq_mtx, curthread));
534 
535 		/* Check if mqueue is not marked as unlinking */
536 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
537 			error = EACCES;
538 			goto exit;
539 		}
540 		/* Fail if O_EXCL is set, and mqueue already exists */
541 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
542 			error = EEXIST;
543 			goto exit;
544 		}
545 
546 		/*
547 		 * Check the permissions. Note the difference between
548 		 * VREAD/VWRITE and FREAD/FWRITE.
549 		 */
550 		acc_mode = 0;
551 		if (fp->f_flag & FREAD) {
552 			acc_mode |= VREAD;
553 		}
554 		if (fp->f_flag & FWRITE) {
555 			acc_mode |= VWRITE;
556 		}
557 		if (vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid,
558 			acc_mode, td->td_ucred)) {
559 
560 			error = EACCES;
561 			goto exit;
562 		}
563 	} else {
564 		/* Fail if mqueue neither exists, nor we create it */
565 		if ((oflag & O_CREAT) == 0) {
566 			lockmgr(&mqlist_mtx, LK_RELEASE);
567 			KKASSERT(mq_new == NULL);
568 			fsetfd(fdp, NULL, mqd);
569 			fp->f_ops = &badfileops;
570 			fdrop(fp);
571 			kfree(name, M_MQBUF);
572 			return ENOENT;
573 		}
574 
575 		/* Check the limit */
576 		if (p->p_mqueue_cnt == mq_open_max) {
577 			error = EMFILE;
578 			goto exit;
579 		}
580 
581 		/* Insert the queue to the list */
582 		mq = mq_new;
583 		lockmgr(&mq->mq_mtx, LK_EXCLUSIVE);
584 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
585 		mq_new = NULL;
586 		getnanotime(&mq->mq_btime);
587 		mq->mq_atime = mq->mq_mtime = mq->mq_btime;
588 	}
589 
590 	/* Increase the counters, and make descriptor ready */
591 	p->p_mqueue_cnt++;
592 	mq->mq_refcnt++;
593 	fp->f_data = mq;
594 exit:
595 	lockmgr(&mq->mq_mtx, LK_RELEASE);
596 	lockmgr(&mqlist_mtx, LK_RELEASE);
597 
598 	if (mq_new)
599 		mqueue_destroy(mq_new);
600 	if (error) {
601 		fsetfd(fdp, NULL, mqd);
602 		fp->f_ops = &badfileops;
603 	} else {
604 		fsetfd(fdp, fp, mqd);
605 		sysmsg->sysmsg_result = mqd;
606 	}
607 	fdrop(fp);
608 	kfree(name, M_MQBUF);
609 
610 	return error;
611 }
612 
613 int
614 sys_mq_close(struct sysmsg *sysmsg, const struct mq_close_args *uap)
615 {
616 	return sys_close(sysmsg, (const void *)uap);
617 }
618 
619 /*
620  * Primary mq_receive1() function.
621  */
622 int
623 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len,
624     unsigned *msg_prio, struct timespec *ts, ssize_t *mlen)
625 {
626 	file_t *fp = NULL;
627 	struct mqueue *mq;
628 	struct mq_msg *msg = NULL;
629 	struct mq_attr *mqattr;
630 	u_int idx;
631 	int error;
632 
633 	/* Get the message queue */
634 	error = mqueue_get(l, mqdes, &fp);
635 	if (error) {
636 		return error;
637 	}
638 	mq = fp->f_data;
639 	if ((fp->f_flag & FREAD) == 0) {
640 		error = EBADF;
641 		goto error;
642 	}
643 	getnanotime(&mq->mq_atime);
644 	mqattr = &mq->mq_attrib;
645 
646 	/* Check the message size limits */
647 	if (msg_len < mqattr->mq_msgsize) {
648 		error = EMSGSIZE;
649 		goto error;
650 	}
651 
652 	/* Check if queue is empty */
653 	while (mqattr->mq_curmsgs == 0) {
654 		int t;
655 
656 		if (mqattr->mq_flags & O_NONBLOCK) {
657 			error = EAGAIN;
658 			goto error;
659 		}
660 		if (ts) {
661 			error = abstimeout2timo(ts, &t);
662 			if (error)
663 				goto error;
664 		} else
665 			t = 0;
666 		/*
667 		 * Block until someone sends the message.
668 		 * While doing this, notification should not be sent.
669 		 */
670 		mqattr->mq_flags |= MQ_RECEIVE;
671 		error = lksleep(&mq->mq_send_cv, &mq->mq_mtx, PCATCH, "mqsend", t);
672 		mqattr->mq_flags &= ~MQ_RECEIVE;
673 		if (error || (mqattr->mq_flags & MQ_UNLINK)) {
674 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
675 			goto error;
676 		}
677 	}
678 
679 
680 	/*
681 	 * Find the highest priority message, and remove it from the queue.
682 	 * At first, reserved queue is checked, bitmap is next.
683 	 */
684 	msg = TAILQ_FIRST(&mq->mq_head[MQ_PQRESQ]);
685 	if (__predict_true(msg == NULL)) {
686 		idx = ffs(mq->mq_bitmap);
687 		msg = TAILQ_FIRST(&mq->mq_head[idx]);
688 		KKASSERT(msg != NULL);
689 	} else {
690 		idx = MQ_PQRESQ;
691 	}
692 	TAILQ_REMOVE(&mq->mq_head[idx], msg, msg_queue);
693 
694 	/* Unmark the bit, if last message. */
695 	if (__predict_true(idx) && TAILQ_EMPTY(&mq->mq_head[idx])) {
696 		KKASSERT((MQ_PQSIZE - idx) == msg->msg_prio);
697 		mq->mq_bitmap &= ~(1 << --idx);
698 	}
699 
700 	/* Decrement the counter and signal waiter, if any */
701 	mqattr->mq_curmsgs--;
702 	wakeup_one(&mq->mq_recv_cv);
703 
704 	/* Ready for sending now */
705 	KNOTE(&mq->mq_wkq.ki_note, 0);
706 error:
707 	lockmgr(&mq->mq_mtx, LK_RELEASE);
708 	fdrop(fp);
709 	if (error)
710 		return error;
711 
712 	/*
713 	 * Copy the data to the user-space.
714 	 * Note: According to POSIX, no message should be removed from the
715 	 * queue in case of fail - this would be violated.
716 	 */
717 	*mlen = msg->msg_len;
718 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
719 	if (error == 0 && msg_prio)
720 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
721 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
722 
723 	return error;
724 }
725 
726 int
727 sys_mq_receive(struct sysmsg *sysmsg, const struct mq_receive_args *uap)
728 {
729 	/* {
730 		syscallarg(mqd_t) mqdes;
731 		syscallarg(char *) msg_ptr;
732 		syscallarg(size_t) msg_len;
733 		syscallarg(unsigned *) msg_prio;
734 	} */
735 	ssize_t mlen;
736 	int error;
737 
738 	error = mq_receive1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
739 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen);
740 	if (error == 0)
741 		sysmsg->sysmsg_result = mlen;
742 
743 	return error;
744 }
745 
746 int
747 sys_mq_timedreceive(struct sysmsg *sysmsg, const struct mq_timedreceive_args *uap)
748 {
749 	/* {
750 		syscallarg(mqd_t) mqdes;
751 		syscallarg(char *) msg_ptr;
752 		syscallarg(size_t) msg_len;
753 		syscallarg(unsigned *) msg_prio;
754 		syscallarg(const struct timespec *) abs_timeout;
755 	} */
756 	int error;
757 	ssize_t mlen;
758 	struct timespec ts, *tsp;
759 
760 	/* Get and convert time value */
761 	if (SCARG(uap, abs_timeout)) {
762 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
763 		if (error)
764 			return error;
765 		tsp = &ts;
766 	} else {
767 		tsp = NULL;
768 	}
769 
770 	error = mq_receive1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
771 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp, &mlen);
772 	if (error == 0)
773 		sysmsg->sysmsg_result = mlen;
774 
775 	return error;
776 }
777 
778 /*
779  * Primary mq_send1() function.
780  */
781 int
782 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len,
783     unsigned msg_prio, struct timespec *ts)
784 {
785 	file_t *fp = NULL;
786 	struct mqueue *mq;
787 	struct mq_msg *msg;
788 	struct mq_attr *mqattr;
789 	struct proc *notify = NULL;
790 	/*ksiginfo_t ksi;*/
791 	size_t size;
792 	int error;
793 
794 	/* Check the priority range */
795 	if (msg_prio >= mq_prio_max)
796 		return EINVAL;
797 
798 	/* Allocate a new message */
799 	size = sizeof(struct mq_msg) + msg_len;
800 	if (size > mq_max_msgsize)
801 		return EMSGSIZE;
802 
803 	msg = kmalloc(size, M_MQBUF, M_WAITOK | M_NULLOK);
804 	if (msg == NULL)
805 		return (ENOMEM);
806 
807 
808 	/* Get the data from user-space */
809 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
810 	if (error) {
811 		mqueue_freemsg(msg, size);
812 		return error;
813 	}
814 	msg->msg_len = msg_len;
815 	msg->msg_prio = msg_prio;
816 
817 	/* Get the mqueue */
818 	error = mqueue_get(l, mqdes, &fp);
819 	if (error) {
820 		mqueue_freemsg(msg, size);
821 		return error;
822 	}
823 	mq = fp->f_data;
824 	if ((fp->f_flag & FWRITE) == 0) {
825 		error = EBADF;
826 		goto error;
827 	}
828 	getnanotime(&mq->mq_mtime);
829 	mqattr = &mq->mq_attrib;
830 
831 	/* Check the message size limit */
832 	if (msg_len <= 0 || msg_len > mqattr->mq_msgsize) {
833 		error = EMSGSIZE;
834 		goto error;
835 	}
836 
837 	/* Check if queue is full */
838 	while (mqattr->mq_curmsgs >= mqattr->mq_maxmsg) {
839 		int t;
840 
841 		if (mqattr->mq_flags & O_NONBLOCK) {
842 			error = EAGAIN;
843 			goto error;
844 		}
845 		if (ts) {
846 			error = abstimeout2timo(ts, &t);
847 			if (error)
848 				goto error;
849 		} else
850 			t = 0;
851 		/* Block until queue becomes available */
852 		error = lksleep(&mq->mq_recv_cv, &mq->mq_mtx, PCATCH, "mqrecv", t);
853 		if (error || (mqattr->mq_flags & MQ_UNLINK)) {
854 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
855 			goto error;
856 		}
857 	}
858 	KKASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg);
859 
860 	/*
861 	 * Insert message into the queue, according to the priority.
862 	 * Note the difference between index and priority.
863 	 */
864 	if (__predict_true(msg_prio < MQ_PQSIZE)) {
865 		u_int idx = MQ_PQSIZE - msg_prio;
866 
867 		KKASSERT(idx != MQ_PQRESQ);
868 		TAILQ_INSERT_TAIL(&mq->mq_head[idx], msg, msg_queue);
869 		mq->mq_bitmap |= (1 << --idx);
870 	} else {
871 		mqueue_linear_insert(mq, msg);
872 	}
873 
874 	/* Check for the notify */
875 	if (mqattr->mq_curmsgs == 0 && mq->mq_notify_proc &&
876 	    (mqattr->mq_flags & MQ_RECEIVE) == 0 &&
877 	    mq->mq_sig_notify.sigev_notify == SIGEV_SIGNAL) {
878 		/* Initialize the signal */
879 		/*KSI_INIT(&ksi);*/
880 		/*ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;*/
881 		/*ksi.ksi_code = SI_MESGQ;*/
882 		/*ksi.ksi_value = mq->mq_sig_notify.sigev_value;*/
883 		/* Unregister the process */
884 		notify = mq->mq_notify_proc;
885 		mq->mq_notify_proc = NULL;
886 	}
887 
888 	/* Increment the counter and signal waiter, if any */
889 	mqattr->mq_curmsgs++;
890 	wakeup_one(&mq->mq_send_cv);
891 
892 	/* Ready for receiving now */
893 	KNOTE(&mq->mq_rkq.ki_note, 0);
894 error:
895 	if (error) {
896 		lockmgr(&mq->mq_mtx, LK_RELEASE);
897 		fdrop(fp);
898 		mqueue_freemsg(msg, size);
899 	} else if (notify) {
900 		PHOLD(notify);
901 		lockmgr(&mq->mq_mtx, LK_RELEASE);
902 		fdrop(fp);
903 		/* Send the notify, if needed */
904 		/*kpsignal(notify, &ksi, NULL);*/
905 		ksignal(notify, mq->mq_sig_notify.sigev_signo);
906 		PRELE(notify);
907 	} else {
908 		lockmgr(&mq->mq_mtx, LK_RELEASE);
909 		fdrop(fp);
910 	}
911 	return error;
912 }
913 
914 int
915 sys_mq_send(struct sysmsg *sysmsg, const struct mq_send_args *uap)
916 {
917 	/* {
918 		syscallarg(mqd_t) mqdes;
919 		syscallarg(const char *) msg_ptr;
920 		syscallarg(size_t) msg_len;
921 		syscallarg(unsigned) msg_prio;
922 	} */
923 
924 	return mq_send1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
925 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0);
926 }
927 
928 int
929 sys_mq_timedsend(struct sysmsg *sysmsg, const struct mq_timedsend_args *uap)
930 {
931 	/* {
932 		syscallarg(mqd_t) mqdes;
933 		syscallarg(const char *) msg_ptr;
934 		syscallarg(size_t) msg_len;
935 		syscallarg(unsigned) msg_prio;
936 		syscallarg(const struct timespec *) abs_timeout;
937 	} */
938 	struct timespec ts, *tsp;
939 	int error;
940 
941 	/* Get and convert time value */
942 	if (SCARG(uap, abs_timeout)) {
943 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
944 		if (error)
945 			return error;
946 		tsp = &ts;
947 	} else {
948 		tsp = NULL;
949 	}
950 
951 	return mq_send1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
952 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp);
953 }
954 
955 int
956 sys_mq_notify(struct sysmsg *sysmsg, const struct mq_notify_args *uap)
957 {
958 	/* {
959 		syscallarg(mqd_t) mqdes;
960 		syscallarg(const struct sigevent *) notification;
961 	} */
962 	file_t *fp = NULL;
963 	struct mqueue *mq;
964 	struct sigevent sig;
965 	int error;
966 
967 	if (SCARG(uap, notification)) {
968 		/* Get the signal from user-space */
969 		error = copyin(SCARG(uap, notification), &sig,
970 		    sizeof(struct sigevent));
971 		if (error)
972 			return error;
973 		if (sig.sigev_notify == SIGEV_SIGNAL &&
974 		    (sig.sigev_signo <= 0 || sig.sigev_signo >= NSIG))
975 			return EINVAL;
976 	}
977 
978 	error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp);
979 	if (error)
980 		return error;
981 	mq = fp->f_data;
982 
983 	if (SCARG(uap, notification)) {
984 		/* Register notification: set the signal and target process */
985 		if (mq->mq_notify_proc == NULL) {
986 			memcpy(&mq->mq_sig_notify, &sig,
987 			    sizeof(struct sigevent));
988 			mq->mq_notify_proc = curproc;
989 		} else {
990 			/* Fail if someone else already registered */
991 			error = EBUSY;
992 		}
993 	} else {
994 		/* Unregister the notification */
995 		mq->mq_notify_proc = NULL;
996 	}
997 	lockmgr(&mq->mq_mtx, LK_RELEASE);
998 	fdrop(fp);
999 
1000 	return error;
1001 }
1002 
1003 int
1004 sys_mq_getattr(struct sysmsg *sysmsg, const struct mq_getattr_args *uap)
1005 {
1006 	/* {
1007 		syscallarg(mqd_t) mqdes;
1008 		syscallarg(struct mq_attr *) mqstat;
1009 	} */
1010 	file_t *fp = NULL;
1011 	struct mqueue *mq;
1012 	struct mq_attr attr;
1013 	int error;
1014 
1015 	/* Get the message queue */
1016 	error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp);
1017 	if (error)
1018 		return error;
1019 	mq = fp->f_data;
1020 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
1021 	lockmgr(&mq->mq_mtx, LK_RELEASE);
1022 	fdrop(fp);
1023 
1024 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
1025 }
1026 
1027 int
1028 sys_mq_setattr(struct sysmsg *sysmsg, const struct mq_setattr_args *uap)
1029 {
1030 	/* {
1031 		syscallarg(mqd_t) mqdes;
1032 		syscallarg(const struct mq_attr *) mqstat;
1033 		syscallarg(struct mq_attr *) omqstat;
1034 	} */
1035 	file_t *fp = NULL;
1036 	struct mqueue *mq;
1037 	struct mq_attr attr;
1038 	int error, nonblock;
1039 
1040 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
1041 	if (error)
1042 		return error;
1043 	nonblock = (attr.mq_flags & O_NONBLOCK);
1044 
1045 	/* Get the message queue */
1046 	error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp);
1047 	if (error)
1048 		return error;
1049 	mq = fp->f_data;
1050 
1051 	/* Copy the old attributes, if needed */
1052 	if (SCARG(uap, omqstat)) {
1053 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
1054 	}
1055 
1056 	/* Ignore everything, except O_NONBLOCK */
1057 	if (nonblock)
1058 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
1059 	else
1060 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
1061 
1062 	lockmgr(&mq->mq_mtx, LK_RELEASE);
1063 	fdrop(fp);
1064 
1065 	/*
1066 	 * Copy the data to the user-space.
1067 	 * Note: According to POSIX, the new attributes should not be set in
1068 	 * case of fail - this would be violated.
1069 	 */
1070 	if (SCARG(uap, omqstat))
1071 		error = copyout(&attr, SCARG(uap, omqstat),
1072 		    sizeof(struct mq_attr));
1073 
1074 	return error;
1075 }
1076 
1077 int
1078 sys_mq_unlink(struct sysmsg *sysmsg, const struct mq_unlink_args *uap)
1079 {
1080 	/* {
1081 		syscallarg(const char *) name;
1082 	} */
1083 	struct thread *td = curthread;
1084 	struct mqueue *mq;
1085 	char *name;
1086 	int error, refcnt = 0;
1087 
1088 	/* Get the name from the user-space */
1089 	name = kmalloc(MQ_NAMELEN, M_MQBUF, M_WAITOK | M_ZERO | M_NULLOK);
1090 	if (name == NULL)
1091 		return (ENOMEM);
1092 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
1093 	if (error) {
1094 		kfree(name, M_MQBUF);
1095 		return error;
1096 	}
1097 
1098 	/* Lookup for this file */
1099 	lockmgr(&mqlist_mtx, LK_EXCLUSIVE);
1100 	mq = mqueue_lookup(name);
1101 	if (mq == NULL) {
1102 		error = ENOENT;
1103 		goto error;
1104 	}
1105 
1106 	/* Check the permissions */
1107 	if (td->td_ucred->cr_uid != mq->mq_euid &&
1108 	    priv_check(td, PRIV_ROOT) != 0) {
1109 		lockmgr(&mq->mq_mtx, LK_RELEASE);
1110 		error = EACCES;
1111 		goto error;
1112 	}
1113 
1114 	/* Mark message queue as unlinking, before leaving the window */
1115 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
1116 
1117 	/* Wake up all waiters, if there are such */
1118 	wakeup(&mq->mq_send_cv);
1119 	wakeup(&mq->mq_recv_cv);
1120 
1121 	KNOTE(&mq->mq_rkq.ki_note, 0);
1122 	KNOTE(&mq->mq_wkq.ki_note, 0);
1123 
1124 	refcnt = mq->mq_refcnt;
1125 	if (refcnt == 0)
1126 		LIST_REMOVE(mq, mq_list);
1127 
1128 	lockmgr(&mq->mq_mtx, LK_RELEASE);
1129 error:
1130 	lockmgr(&mqlist_mtx, LK_RELEASE);
1131 
1132 	/*
1133 	 * If there are no references - destroy the message
1134 	 * queue, otherwise, the last mq_close() will do that.
1135 	 */
1136 	if (error == 0 && refcnt == 0)
1137 		mqueue_destroy(mq);
1138 
1139 	kfree(name, M_MQBUF);
1140 	return error;
1141 }
1142 
1143 /*
1144  * SysCtl.
1145  */
1146 SYSCTL_NODE(_kern, OID_AUTO, mqueue,
1147     CTLFLAG_RW, 0, "Message queue options");
1148 
1149 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_open_max,
1150     CTLFLAG_RW, &mq_open_max, 0,
1151     "Maximal number of message queue descriptors per process");
1152 
1153 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_prio_max,
1154     CTLFLAG_RW, &mq_prio_max, 0,
1155     "Maximal priority of the message");
1156 
1157 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_max_msgsize,
1158     CTLFLAG_RW, &mq_max_msgsize, 0,
1159     "Maximal allowed size of the message");
1160 
1161 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_def_maxmsg,
1162     CTLFLAG_RW, &mq_def_maxmsg, 0,
1163     "Default maximal message count");
1164 
1165 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_max_maxmsg,
1166     CTLFLAG_RW, &mq_max_maxmsg, 0,
1167     "Maximal allowed message count");
1168 
1169 SYSINIT(sys_mqueue_init, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, mqueue_sysinit, NULL);
1170