xref: /netbsd-src/sys/kern/sys_mqueue.c (revision b5677b36047b601b9addaaa494a58ceae82c2a6c)
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/cdefs.h>
45 __KERNEL_RCSID(0, "$NetBSD: sys_mqueue.c,v 1.16 2009/04/11 23:05:26 christos Exp $");
46 
47 #include <sys/param.h>
48 #include <sys/types.h>
49 #include <sys/condvar.h>
50 #include <sys/errno.h>
51 #include <sys/fcntl.h>
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/kauth.h>
55 #include <sys/kernel.h>
56 #include <sys/kmem.h>
57 #include <sys/lwp.h>
58 #include <sys/mqueue.h>
59 #include <sys/mutex.h>
60 #include <sys/pool.h>
61 #include <sys/poll.h>
62 #include <sys/proc.h>
63 #include <sys/queue.h>
64 #include <sys/select.h>
65 #include <sys/signal.h>
66 #include <sys/signalvar.h>
67 #include <sys/stat.h>
68 #include <sys/sysctl.h>
69 #include <sys/syscallargs.h>
70 #include <sys/systm.h>
71 #include <sys/unistd.h>
72 #include <sys/vnode.h>
73 
74 /* System-wide limits. */
75 static u_int			mq_open_max = MQ_OPEN_MAX;
76 static u_int			mq_prio_max = MQ_PRIO_MAX;
77 
78 static u_int			mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
79 static u_int			mq_def_maxmsg = 32;
80 
81 static kmutex_t			mqlist_mtx;
82 static pool_cache_t		mqmsg_cache;
83 static LIST_HEAD(, mqueue)	mqueue_head =
84 	LIST_HEAD_INITIALIZER(mqueue_head);
85 
86 static int	mq_poll_fop(file_t *, int);
87 static int	mq_stat_fop(file_t *, struct stat *);
88 static int	mq_close_fop(file_t *);
89 
90 #define	FNOVAL	-1
91 
92 static const struct fileops mqops = {
93 	.fo_read = fbadop_read,
94 	.fo_write = fbadop_write,
95 	.fo_ioctl = fbadop_ioctl,
96 	.fo_fcntl = fnullop_fcntl,
97 	.fo_poll = mq_poll_fop,
98 	.fo_stat = mq_stat_fop,
99 	.fo_close = mq_close_fop,
100 	.fo_kqfilter = fnullop_kqfilter,
101 	.fo_drain = fnullop_drain,
102 };
103 
104 /*
105  * Initialize POSIX message queue subsystem.
106  */
107 void
108 mqueue_sysinit(void)
109 {
110 
111 	mqmsg_cache = pool_cache_init(MQ_DEF_MSGSIZE, coherency_unit,
112 	    0, 0, "mqmsgpl", NULL, IPL_NONE, NULL, NULL, NULL);
113 	mutex_init(&mqlist_mtx, MUTEX_DEFAULT, IPL_NONE);
114 }
115 
116 /*
117  * Free the message.
118  */
119 static void
120 mqueue_freemsg(struct mq_msg *msg, const size_t size)
121 {
122 
123 	if (size > MQ_DEF_MSGSIZE)
124 		kmem_free(msg, size);
125 	else
126 		pool_cache_put(mqmsg_cache, msg);
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 
137 	while ((msg = TAILQ_FIRST(&mq->mq_head)) != NULL) {
138 		TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
139 		mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
140 	}
141 	seldestroy(&mq->mq_rsel);
142 	seldestroy(&mq->mq_wsel);
143 	cv_destroy(&mq->mq_send_cv);
144 	cv_destroy(&mq->mq_recv_cv);
145 	mutex_destroy(&mq->mq_mtx);
146 	kmem_free(mq, sizeof(struct mqueue));
147 }
148 
149 /*
150  * Lookup for file name in general list of message queues.
151  *  => locks the message queue
152  */
153 static void *
154 mqueue_lookup(char *name)
155 {
156 	struct mqueue *mq;
157 	KASSERT(mutex_owned(&mqlist_mtx));
158 
159 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
160 		if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
161 			mutex_enter(&mq->mq_mtx);
162 			return mq;
163 		}
164 	}
165 
166 	return NULL;
167 }
168 
169 /*
170  * Check access against message queue.
171  */
172 static inline int
173 mqueue_access(struct lwp *l, struct mqueue *mq, int access)
174 {
175 	mode_t acc_mode = 0;
176 
177 	KASSERT(mutex_owned(&mq->mq_mtx));
178 	KASSERT(access != FNOVAL);
179 
180 	/* Note the difference between VREAD/VWRITE and FREAD/FWRITE */
181 	if (access & FREAD)
182 		acc_mode |= VREAD;
183 	if (access & FWRITE)
184 		acc_mode |= VWRITE;
185 
186 	return vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid,
187 	    acc_mode, l->l_cred);
188 }
189 
190 /*
191  * Get the mqueue from the descriptor.
192  *  => locks the message queue, if found
193  *  => increments the reference on file entry
194  */
195 static int
196 mqueue_get(struct lwp *l, mqd_t mqd, int access, file_t **fpr)
197 {
198 	file_t *fp;
199 	struct mqueue *mq;
200 
201 	/* Get the file and descriptor */
202 	fp = fd_getfile((int)mqd);
203 	if (fp == NULL)
204 		return EBADF;
205 
206 	/* Increment the reference of file entry, and lock the mqueue */
207 	mq = fp->f_data;
208 	*fpr = fp;
209 	mutex_enter(&mq->mq_mtx);
210 	if (access == FNOVAL) {
211 		KASSERT(mutex_owned(&mq->mq_mtx));
212 		return 0;
213 	}
214 
215 	/* Check the access mode and permission */
216 	if ((fp->f_flag & access) != access || mqueue_access(l, mq, access)) {
217 		mutex_exit(&mq->mq_mtx);
218 		fd_putfile((int)mqd);
219 		return EPERM;
220 	}
221 	return 0;
222 }
223 
224 /*
225  * Converter from struct timespec to the ticks.
226  * Used by mq_timedreceive(), mq_timedsend().
227  */
228 int
229 abstimeout2timo(struct timespec *ts, int *timo)
230 {
231 	int error;
232 
233 	/*
234 	 * According to POSIX, validation check is needed only in case of
235 	 * blocking.  Thus, set the invalid value right now, and fail latter.
236 	 */
237 	error = itimespecfix(ts);
238 	*timo = (error == 0) ? tstohz(ts) : -1;
239 
240 	return 0;
241 }
242 
243 static int
244 mq_stat_fop(file_t *fp, struct stat *st)
245 {
246 	struct mqueue *mq = fp->f_data;
247 
248 	(void)memset(st, 0, sizeof(*st));
249 
250 	mutex_enter(&mq->mq_mtx);
251 	st->st_mode = mq->mq_mode;
252 	st->st_uid = mq->mq_euid;
253 	st->st_gid = mq->mq_egid;
254 	st->st_atimespec = mq->mq_atime;
255 	st->st_mtimespec = mq->mq_mtime;
256 	st->st_ctimespec = st->st_birthtimespec = mq->mq_btime;
257 	st->st_uid = kauth_cred_geteuid(fp->f_cred);
258 	st->st_gid = kauth_cred_getegid(fp->f_cred);
259 	mutex_exit(&mq->mq_mtx);
260 
261 	return 0;
262 }
263 
264 static int
265 mq_poll_fop(file_t *fp, int events)
266 {
267 	struct mqueue *mq = fp->f_data;
268 	int revents = 0;
269 
270 	mutex_enter(&mq->mq_mtx);
271 	if (events & (POLLIN | POLLRDNORM)) {
272 		/* Ready for receiving, if there are messages in the queue */
273 		if (mq->mq_attrib.mq_curmsgs)
274 			revents |= (POLLIN | POLLRDNORM);
275 		else
276 			selrecord(curlwp, &mq->mq_rsel);
277 	}
278 	if (events & (POLLOUT | POLLWRNORM)) {
279 		/* Ready for sending, if the message queue is not full */
280 		if (mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg)
281 			revents |= (POLLOUT | POLLWRNORM);
282 		else
283 			selrecord(curlwp, &mq->mq_wsel);
284 	}
285 	mutex_exit(&mq->mq_mtx);
286 
287 	return revents;
288 }
289 
290 static int
291 mq_close_fop(file_t *fp)
292 {
293 	struct proc *p = curproc;
294 	struct mqueue *mq = fp->f_data;
295 	bool destroy;
296 
297 	mutex_enter(&mqlist_mtx);
298 	mutex_enter(&mq->mq_mtx);
299 
300 	/* Decrease the counters */
301 	p->p_mqueue_cnt--;
302 	mq->mq_refcnt--;
303 
304 	/* Remove notification if registered for this process */
305 	if (mq->mq_notify_proc == p)
306 		mq->mq_notify_proc = NULL;
307 
308 	/*
309 	 * If this is the last reference and mqueue is marked for unlink,
310 	 * remove and later destroy the message queue.
311 	 */
312 	if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
313 		LIST_REMOVE(mq, mq_list);
314 		destroy = true;
315 	} else
316 		destroy = false;
317 
318 	mutex_exit(&mq->mq_mtx);
319 	mutex_exit(&mqlist_mtx);
320 
321 	if (destroy)
322 		mqueue_destroy(mq);
323 
324 	return 0;
325 }
326 
327 /*
328  * General mqueue system calls.
329  */
330 
331 int
332 sys_mq_open(struct lwp *l, const struct sys_mq_open_args *uap,
333     register_t *retval)
334 {
335 	/* {
336 		syscallarg(const char *) name;
337 		syscallarg(int) oflag;
338 		syscallarg(mode_t) mode;
339 		syscallarg(struct mq_attr) attr;
340 	} */
341 	struct proc *p = l->l_proc;
342 	struct mqueue *mq, *mq_new = NULL;
343 	file_t *fp;
344 	char *name;
345 	int mqd, error, oflag;
346 
347 	/* Check access mode flags */
348 	oflag = SCARG(uap, oflag);
349 	if ((oflag & O_ACCMODE) == 0)
350 		return EINVAL;
351 
352 	/* Get the name from the user-space */
353 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
354 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
355 	if (error) {
356 		kmem_free(name, MQ_NAMELEN);
357 		return error;
358 	}
359 
360 	if (oflag & O_CREAT) {
361 		struct cwdinfo *cwdi = p->p_cwdi;
362 		struct mq_attr attr;
363 
364 		/* Check the limit */
365 		if (p->p_mqueue_cnt == mq_open_max) {
366 			kmem_free(name, MQ_NAMELEN);
367 			return EMFILE;
368 		}
369 
370 		/* Check for mqueue attributes */
371 		if (SCARG(uap, attr)) {
372 			error = copyin(SCARG(uap, attr), &attr,
373 				sizeof(struct mq_attr));
374 			if (error) {
375 				kmem_free(name, MQ_NAMELEN);
376 				return error;
377 			}
378 			if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 ||
379 			    attr.mq_msgsize > mq_max_msgsize) {
380 				kmem_free(name, MQ_NAMELEN);
381 				return EINVAL;
382 			}
383 			attr.mq_curmsgs = 0;
384 		} else {
385 			memset(&attr, 0, sizeof(struct mq_attr));
386 			attr.mq_maxmsg = mq_def_maxmsg;
387 			attr.mq_msgsize =
388 			    MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
389 		}
390 
391 		/*
392 		 * Allocate new mqueue, initialize data structures,
393 		 * copy the name, attributes and set the flag.
394 		 */
395 		mq_new = kmem_zalloc(sizeof(struct mqueue), KM_SLEEP);
396 
397 		mutex_init(&mq_new->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
398 		cv_init(&mq_new->mq_send_cv, "mqsendcv");
399 		cv_init(&mq_new->mq_recv_cv, "mqrecvcv");
400 		TAILQ_INIT(&mq_new->mq_head);
401 		selinit(&mq_new->mq_rsel);
402 		selinit(&mq_new->mq_wsel);
403 
404 		strlcpy(mq_new->mq_name, name, MQ_NAMELEN);
405 		memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr));
406 		mq_new->mq_attrib.mq_flags = oflag;
407 
408 		/* Store mode and effective UID with GID */
409 		mq_new->mq_mode = ((SCARG(uap, mode) &
410 		    ~cwdi->cwdi_cmask) & ALLPERMS) & ~S_ISTXT;
411 		mq_new->mq_euid = kauth_cred_geteuid(l->l_cred);
412 		mq_new->mq_egid = kauth_cred_getegid(l->l_cred);
413 	}
414 
415 	/* Allocate file structure and descriptor */
416 	error = fd_allocfile(&fp, &mqd);
417 	if (error) {
418 		if (mq_new)
419 			mqueue_destroy(mq_new);
420 		kmem_free(name, MQ_NAMELEN);
421 		return error;
422 	}
423 	fp->f_type = DTYPE_MQUEUE;
424 	fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE);
425 	fp->f_ops = &mqops;
426 
427 	/* Look up for mqueue with such name */
428 	mutex_enter(&mqlist_mtx);
429 	mq = mqueue_lookup(name);
430 	if (mq) {
431 		KASSERT(mutex_owned(&mq->mq_mtx));
432 
433 		/* Check if mqueue is not marked as unlinking */
434 		if (mq->mq_attrib.mq_flags & MQ_UNLINK) {
435 			error = EACCES;
436 			goto exit;
437 		}
438 		/* Fail if O_EXCL is set, and mqueue already exists */
439 		if ((oflag & O_CREAT) && (oflag & O_EXCL)) {
440 			error = EEXIST;
441 			goto exit;
442 		}
443 		/* Check the permission */
444 		if (mqueue_access(l, mq, fp->f_flag)) {
445 			error = EACCES;
446 			goto exit;
447 		}
448 	} else {
449 		/* Fail if mqueue neither exists, nor we create it */
450 		if ((oflag & O_CREAT) == 0) {
451 			mutex_exit(&mqlist_mtx);
452 			KASSERT(mq_new == NULL);
453 			fd_abort(p, fp, mqd);
454 			kmem_free(name, MQ_NAMELEN);
455 			return ENOENT;
456 		}
457 
458 		/* Check the limit */
459 		if (p->p_mqueue_cnt == mq_open_max) {
460 			error = EMFILE;
461 			goto exit;
462 		}
463 
464 		/* Insert the queue to the list */
465 		mq = mq_new;
466 		mutex_enter(&mq->mq_mtx);
467 		LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
468 		mq_new = NULL;
469 		getnanotime(&mq->mq_btime);
470 		mq->mq_atime = mq->mq_mtime = mq->mq_btime;
471 	}
472 
473 	/* Increase the counters, and make descriptor ready */
474 	p->p_mqueue_cnt++;
475 	mq->mq_refcnt++;
476 	fp->f_data = mq;
477 exit:
478 	mutex_exit(&mq->mq_mtx);
479 	mutex_exit(&mqlist_mtx);
480 
481 	if (mq_new)
482 		mqueue_destroy(mq_new);
483 	if (error) {
484 		fd_abort(p, fp, mqd);
485 	} else {
486 		fd_affix(p, fp, mqd);
487 		*retval = mqd;
488 	}
489 	kmem_free(name, MQ_NAMELEN);
490 
491 	return error;
492 }
493 
494 int
495 sys_mq_close(struct lwp *l, const struct sys_mq_close_args *uap,
496     register_t *retval)
497 {
498 
499 	return sys_close(l, (const void *)uap, retval);
500 }
501 
502 /*
503  * Primary mq_receive1() function.
504  */
505 int
506 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len,
507     unsigned *msg_prio, int t, ssize_t *mlen)
508 {
509 	file_t *fp = NULL;
510 	struct mqueue *mq;
511 	struct mq_msg *msg = NULL;
512 	int error;
513 
514 	/* Get the message queue */
515 	error = mqueue_get(l, mqdes, FREAD, &fp);
516 	if (error)
517 		return error;
518 	mq = fp->f_data;
519 
520 	getnanotime(&mq->mq_atime);
521 	/* Check the message size limits */
522 	if (msg_len < mq->mq_attrib.mq_msgsize) {
523 		error = EMSGSIZE;
524 		goto error;
525 	}
526 
527 	/* Check if queue is empty */
528 	while (TAILQ_EMPTY(&mq->mq_head)) {
529 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
530 			error = EAGAIN;
531 			goto error;
532 		}
533 		if (t < 0) {
534 			error = EINVAL;
535 			goto error;
536 		}
537 		/*
538 		 * Block until someone sends the message.
539 		 * While doing this, notification should not be sent.
540 		 */
541 		mq->mq_attrib.mq_flags |= MQ_RECEIVE;
542 		error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
543 		mq->mq_attrib.mq_flags &= ~MQ_RECEIVE;
544 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
545 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
546 			goto error;
547 		}
548 	}
549 
550 	/* Remove the message from the queue */
551 	msg = TAILQ_FIRST(&mq->mq_head);
552 	KASSERT(msg != NULL);
553 	TAILQ_REMOVE(&mq->mq_head, msg, msg_queue);
554 
555 	/* Decrement the counter and signal waiter, if any */
556 	mq->mq_attrib.mq_curmsgs--;
557 	cv_signal(&mq->mq_recv_cv);
558 
559 	/* Ready for sending now */
560 	selnotify(&mq->mq_wsel, POLLOUT | POLLWRNORM, 0);
561 error:
562 	mutex_exit(&mq->mq_mtx);
563 	fd_putfile((int)mqdes);
564 	if (error)
565 		return error;
566 
567 	/*
568 	 * Copy the data to the user-space.
569 	 * Note: According to POSIX, no message should be removed from the
570 	 * queue in case of fail - this would be violated.
571 	 */
572 	*mlen = msg->msg_len;
573 	error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len);
574 	if (error == 0 && msg_prio)
575 		error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned));
576 	mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len);
577 
578 	return error;
579 }
580 
581 int
582 sys_mq_receive(struct lwp *l, const struct sys_mq_receive_args *uap,
583     register_t *retval)
584 {
585 	/* {
586 		syscallarg(mqd_t) mqdes;
587 		syscallarg(char *) msg_ptr;
588 		syscallarg(size_t) msg_len;
589 		syscallarg(unsigned *) msg_prio;
590 	} */
591 	int error;
592 	ssize_t mlen;
593 
594 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
595 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen);
596 	if (error == 0)
597 		*retval = mlen;
598 
599 	return error;
600 }
601 
602 int
603 sys___mq_timedreceive50(struct lwp *l,
604     const struct sys___mq_timedreceive50_args *uap, register_t *retval)
605 {
606 	/* {
607 		syscallarg(mqd_t) mqdes;
608 		syscallarg(char *) msg_ptr;
609 		syscallarg(size_t) msg_len;
610 		syscallarg(unsigned *) msg_prio;
611 		syscallarg(const struct timespec *) abs_timeout;
612 	} */
613 	int error, t;
614 	ssize_t mlen;
615 	struct timespec ts;
616 
617 	/* Get and convert time value */
618 	if (SCARG(uap, abs_timeout)) {
619 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
620 		if (error)
621 			return error;
622 
623 		error = abstimeout2timo(&ts, &t);
624 		if (error)
625 			return error;
626 	} else
627 		t = 0;
628 
629 	error = mq_receive1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
630 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t, &mlen);
631 	if (error == 0)
632 		*retval = mlen;
633 
634 	return error;
635 }
636 
637 /*
638  * Primary mq_send1() function.
639  */
640 int
641 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len,
642     unsigned msg_prio, int t)
643 {
644 	file_t *fp = NULL;
645 	struct mqueue *mq;
646 	struct mq_msg *msg, *pos_msg;
647 	struct proc *notify = NULL;
648 	ksiginfo_t ksi;
649 	size_t size;
650 	int error;
651 
652 	/* Check the priority range */
653 	if (msg_prio >= mq_prio_max)
654 		return EINVAL;
655 
656 	/* Allocate a new message */
657 	size = sizeof(struct mq_msg) + msg_len;
658 	if (size > mq_max_msgsize)
659 		return EMSGSIZE;
660 
661 	if (size > MQ_DEF_MSGSIZE)
662 		msg = kmem_alloc(size, KM_SLEEP);
663 	else
664 		msg = pool_cache_get(mqmsg_cache, PR_WAITOK);
665 
666 	/* Get the data from user-space */
667 	error = copyin(msg_ptr, msg->msg_ptr, msg_len);
668 	if (error) {
669 		mqueue_freemsg(msg, size);
670 		return error;
671 	}
672 	msg->msg_len = msg_len;
673 	msg->msg_prio = msg_prio;
674 
675 	/* Get the mqueue */
676 	error = mqueue_get(l, mqdes, FWRITE, &fp);
677 	if (error) {
678 		mqueue_freemsg(msg, size);
679 		return error;
680 	}
681 	mq = fp->f_data;
682 
683 	getnanotime(&mq->mq_mtime);
684 
685 	/* Check the message size limit */
686 	if (msg_len <= 0 || msg_len > mq->mq_attrib.mq_msgsize) {
687 		error = EMSGSIZE;
688 		goto error;
689 	}
690 
691 	/* Check if queue is full */
692 	while (mq->mq_attrib.mq_curmsgs >= mq->mq_attrib.mq_maxmsg) {
693 		if (mq->mq_attrib.mq_flags & O_NONBLOCK) {
694 			error = EAGAIN;
695 			goto error;
696 		}
697 		if (t < 0) {
698 			error = EINVAL;
699 			goto error;
700 		}
701 		/* Block until queue becomes available */
702 		error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
703 		if (error || (mq->mq_attrib.mq_flags & MQ_UNLINK)) {
704 			error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
705 			goto error;
706 		}
707 	}
708 	KASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg);
709 
710 	/* Insert message into the queue, according to the priority */
711 	TAILQ_FOREACH(pos_msg, &mq->mq_head, msg_queue)
712 		if (msg->msg_prio > pos_msg->msg_prio)
713 			break;
714 	if (pos_msg == NULL)
715 		TAILQ_INSERT_TAIL(&mq->mq_head, msg, msg_queue);
716 	else
717 		TAILQ_INSERT_BEFORE(pos_msg, msg, msg_queue);
718 
719 	/* Check for the notify */
720 	if (mq->mq_attrib.mq_curmsgs == 0 && mq->mq_notify_proc &&
721 	    (mq->mq_attrib.mq_flags & MQ_RECEIVE) == 0) {
722 		/* Initialize the signal */
723 		KSI_INIT(&ksi);
724 		ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
725 		ksi.ksi_code = SI_MESGQ;
726 		ksi.ksi_value = mq->mq_sig_notify.sigev_value;
727 		/* Unregister the process */
728 		notify = mq->mq_notify_proc;
729 		mq->mq_notify_proc = NULL;
730 	}
731 
732 	/* Increment the counter and signal waiter, if any */
733 	mq->mq_attrib.mq_curmsgs++;
734 	cv_signal(&mq->mq_send_cv);
735 
736 	/* Ready for receiving now */
737 	selnotify(&mq->mq_rsel, POLLIN | POLLRDNORM, 0);
738 error:
739 	mutex_exit(&mq->mq_mtx);
740 	fd_putfile((int)mqdes);
741 
742 	if (error) {
743 		mqueue_freemsg(msg, size);
744 	} else if (notify) {
745 		/* Send the notify, if needed */
746 		mutex_enter(proc_lock);
747 		kpsignal(notify, &ksi, NULL);
748 		mutex_exit(proc_lock);
749 	}
750 
751 	return error;
752 }
753 
754 int
755 sys_mq_send(struct lwp *l, const struct sys_mq_send_args *uap,
756     register_t *retval)
757 {
758 	/* {
759 		syscallarg(mqd_t) mqdes;
760 		syscallarg(const char *) msg_ptr;
761 		syscallarg(size_t) msg_len;
762 		syscallarg(unsigned) msg_prio;
763 	} */
764 
765 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
766 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), 0);
767 }
768 
769 int
770 sys___mq_timedsend50(struct lwp *l, const struct sys___mq_timedsend50_args *uap,
771     register_t *retval)
772 {
773 	/* {
774 		syscallarg(mqd_t) mqdes;
775 		syscallarg(const char *) msg_ptr;
776 		syscallarg(size_t) msg_len;
777 		syscallarg(unsigned) msg_prio;
778 		syscallarg(const struct timespec *) abs_timeout;
779 	} */
780 	int t;
781 	struct timespec ts;
782 	int error;
783 
784 	/* Get and convert time value */
785 	if (SCARG(uap, abs_timeout)) {
786 		error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
787 		if (error)
788 			return error;
789 		error = abstimeout2timo(&ts, &t);
790 		if (error)
791 			return error;
792 	} else
793 		t = 0;
794 
795 	return mq_send1(l, SCARG(uap, mqdes), SCARG(uap, msg_ptr),
796 	    SCARG(uap, msg_len), SCARG(uap, msg_prio), t);
797 }
798 
799 int
800 sys_mq_notify(struct lwp *l, const struct sys_mq_notify_args *uap,
801     register_t *retval)
802 {
803 	/* {
804 		syscallarg(mqd_t) mqdes;
805 		syscallarg(const struct sigevent *) notification;
806 	} */
807 	file_t *fp = NULL;
808 	struct mqueue *mq;
809 	struct sigevent sig;
810 	int error;
811 
812 	if (SCARG(uap, notification)) {
813 		/* Get the signal from user-space */
814 		error = copyin(SCARG(uap, notification), &sig,
815 		    sizeof(struct sigevent));
816 		if (error)
817 			return error;
818 	}
819 
820 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
821 	if (error)
822 		return error;
823 	mq = fp->f_data;
824 
825 	if (SCARG(uap, notification)) {
826 		/* Register notification: set the signal and target process */
827 		if (mq->mq_notify_proc == NULL) {
828 			memcpy(&mq->mq_sig_notify, &sig,
829 			    sizeof(struct sigevent));
830 			mq->mq_notify_proc = l->l_proc;
831 		} else {
832 			/* Fail if someone else already registered */
833 			error = EBUSY;
834 		}
835 	} else {
836 		/* Unregister the notification */
837 		mq->mq_notify_proc = NULL;
838 	}
839 	mutex_exit(&mq->mq_mtx);
840 	fd_putfile((int)SCARG(uap, mqdes));
841 
842 	return error;
843 }
844 
845 int
846 sys_mq_getattr(struct lwp *l, const struct sys_mq_getattr_args *uap,
847     register_t *retval)
848 {
849 	/* {
850 		syscallarg(mqd_t) mqdes;
851 		syscallarg(struct mq_attr *) mqstat;
852 	} */
853 	file_t *fp = NULL;
854 	struct mqueue *mq;
855 	struct mq_attr attr;
856 	int error;
857 
858 	/* Get the message queue */
859 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
860 	if (error)
861 		return error;
862 	mq = fp->f_data;
863 	memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
864 	mutex_exit(&mq->mq_mtx);
865 	fd_putfile((int)SCARG(uap, mqdes));
866 
867 	return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
868 }
869 
870 int
871 sys_mq_setattr(struct lwp *l, const struct sys_mq_setattr_args *uap,
872     register_t *retval)
873 {
874 	/* {
875 		syscallarg(mqd_t) mqdes;
876 		syscallarg(const struct mq_attr *) mqstat;
877 		syscallarg(struct mq_attr *) omqstat;
878 	} */
879 	file_t *fp = NULL;
880 	struct mqueue *mq;
881 	struct mq_attr attr;
882 	int error, nonblock;
883 
884 	error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
885 	if (error)
886 		return error;
887 	nonblock = (attr.mq_flags & O_NONBLOCK);
888 
889 	/* Get the message queue */
890 	error = mqueue_get(l, SCARG(uap, mqdes), FNOVAL, &fp);
891 	if (error)
892 		return error;
893 	mq = fp->f_data;
894 
895 	/* Copy the old attributes, if needed */
896 	if (SCARG(uap, omqstat))
897 		memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
898 
899 	/* Ignore everything, except O_NONBLOCK */
900 	if (nonblock)
901 		mq->mq_attrib.mq_flags |= O_NONBLOCK;
902 	else
903 		mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
904 
905 	mutex_exit(&mq->mq_mtx);
906 	fd_putfile((int)SCARG(uap, mqdes));
907 
908 	/*
909 	 * Copy the data to the user-space.
910 	 * Note: According to POSIX, the new attributes should not be set in
911 	 * case of fail - this would be violated.
912 	 */
913 	if (SCARG(uap, omqstat))
914 		error = copyout(&attr, SCARG(uap, omqstat),
915 		    sizeof(struct mq_attr));
916 
917 	return error;
918 }
919 
920 int
921 sys_mq_unlink(struct lwp *l, const struct sys_mq_unlink_args *uap,
922     register_t *retval)
923 {
924 	/* {
925 		syscallarg(const char *) name;
926 	} */
927 	struct mqueue *mq;
928 	char *name;
929 	int error, refcnt = 0;
930 
931 	/* Get the name from the user-space */
932 	name = kmem_zalloc(MQ_NAMELEN, KM_SLEEP);
933 	error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
934 	if (error) {
935 		kmem_free(name, MQ_NAMELEN);
936 		return error;
937 	}
938 
939 	/* Lookup for this file */
940 	mutex_enter(&mqlist_mtx);
941 	mq = mqueue_lookup(name);
942 	if (mq == NULL) {
943 		error = ENOENT;
944 		goto error;
945 	}
946 
947 	/* Check the permissions */
948 	if (mqueue_access(l, mq, FWRITE)) {
949 		mutex_exit(&mq->mq_mtx);
950 		error = EACCES;
951 		goto error;
952 	}
953 
954 	/* Mark message queue as unlinking, before leaving the window */
955 	mq->mq_attrib.mq_flags |= MQ_UNLINK;
956 
957 	/* Wake up all waiters, if there are such */
958 	cv_broadcast(&mq->mq_send_cv);
959 	cv_broadcast(&mq->mq_recv_cv);
960 
961 	selnotify(&mq->mq_rsel, POLLHUP, 0);
962 	selnotify(&mq->mq_wsel, POLLHUP, 0);
963 
964 	refcnt = mq->mq_refcnt;
965 	if (refcnt == 0)
966 		LIST_REMOVE(mq, mq_list);
967 
968 	mutex_exit(&mq->mq_mtx);
969 error:
970 	mutex_exit(&mqlist_mtx);
971 
972 	/*
973 	 * If there are no references - destroy the message
974 	 * queue, otherwise, the last mq_close() will do that.
975 	 */
976 	if (error == 0 && refcnt == 0)
977 		mqueue_destroy(mq);
978 
979 	kmem_free(name, MQ_NAMELEN);
980 	return error;
981 }
982 
983 /*
984  * SysCtl.
985  */
986 
987 SYSCTL_SETUP(sysctl_mqueue_setup, "sysctl mqueue setup")
988 {
989 	const struct sysctlnode *node = NULL;
990 
991 	sysctl_createv(clog, 0, NULL, NULL,
992 		CTLFLAG_PERMANENT,
993 		CTLTYPE_NODE, "kern", NULL,
994 		NULL, 0, NULL, 0,
995 		CTL_KERN, CTL_EOL);
996 	sysctl_createv(clog, 0, NULL, NULL,
997 		CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
998 		CTLTYPE_INT, "posix_msg",
999 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
1000 			     "Message Passing option to which the "
1001 			     "system attempts to conform"),
1002 		NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
1003 		CTL_KERN, CTL_CREATE, CTL_EOL);
1004 	sysctl_createv(clog, 0, NULL, &node,
1005 		CTLFLAG_PERMANENT,
1006 		CTLTYPE_NODE, "mqueue",
1007 		SYSCTL_DESCR("Message queue options"),
1008 		NULL, 0, NULL, 0,
1009 		CTL_KERN, CTL_CREATE, CTL_EOL);
1010 
1011 	if (node == NULL)
1012 		return;
1013 
1014 	sysctl_createv(clog, 0, &node, NULL,
1015 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1016 		CTLTYPE_INT, "mq_open_max",
1017 		SYSCTL_DESCR("Maximal number of message queue descriptors "
1018 			     "that process could open"),
1019 		NULL, 0, &mq_open_max, 0,
1020 		CTL_CREATE, CTL_EOL);
1021 	sysctl_createv(clog, 0, &node, NULL,
1022 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1023 		CTLTYPE_INT, "mq_prio_max",
1024 		SYSCTL_DESCR("Maximal priority of the message"),
1025 		NULL, 0, &mq_prio_max, 0,
1026 		CTL_CREATE, CTL_EOL);
1027 	sysctl_createv(clog, 0, &node, NULL,
1028 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1029 		CTLTYPE_INT, "mq_max_msgsize",
1030 		SYSCTL_DESCR("Maximal allowed size of the message"),
1031 		NULL, 0, &mq_max_msgsize, 0,
1032 		CTL_CREATE, CTL_EOL);
1033 	sysctl_createv(clog, 0, &node, NULL,
1034 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1035 		CTLTYPE_INT, "mq_def_maxmsg",
1036 		SYSCTL_DESCR("Default maximal message count"),
1037 		NULL, 0, &mq_def_maxmsg, 0,
1038 		CTL_CREATE, CTL_EOL);
1039 }
1040 
1041 /*
1042  * Debugging.
1043  */
1044 #if defined(DDB)
1045 
1046 void
1047 mqueue_print_list(void (*pr)(const char *, ...))
1048 {
1049 	struct mqueue *mq;
1050 
1051 	(*pr)("Global list of the message queues:\n");
1052 	(*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
1053 	    "Name", "Ptr", "Mode", "Flags",  "Ref",
1054 	    "MaxMsg", "MsgSze", "CurMsg");
1055 	LIST_FOREACH(mq, &mqueue_head, mq_list) {
1056 		(*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
1057 		    mq->mq_name, mq, mq->mq_mode,
1058 		    mq->mq_attrib.mq_flags, mq->mq_refcnt,
1059 		    mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
1060 		    mq->mq_attrib.mq_curmsgs);
1061 	}
1062 }
1063 
1064 #endif /* defined(DDB) */
1065