1 /* $NetBSD: sys_mqueue.c,v 1.49 2024/07/01 01:35:53 christos Exp $ */
2
3 /*
4 * Copyright (c) 2007-2011 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) is protected by mqlist_lock.
36 * Each message queue and its members are protected by mqueue::mq_mtx.
37 * Note that proc_t::p_mqueue_cnt is updated atomically.
38 *
39 * Lock order:
40 *
41 * mqlist_lock ->
42 * mqueue::mq_mtx
43 */
44
45 #include <sys/cdefs.h>
46 __KERNEL_RCSID(0, "$NetBSD: sys_mqueue.c,v 1.49 2024/07/01 01:35:53 christos Exp $");
47
48 #include <sys/param.h>
49 #include <sys/types.h>
50 #include <sys/atomic.h>
51
52 #include <sys/file.h>
53 #include <sys/filedesc.h>
54 #include <sys/kauth.h>
55 #include <sys/lwp.h>
56 #include <sys/mqueue.h>
57 #include <sys/module.h>
58 #include <sys/poll.h>
59 #include <sys/select.h>
60 #include <sys/signal.h>
61 #include <sys/signalvar.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/syscall.h>
65 #include <sys/syscallvar.h>
66 #include <sys/syscallargs.h>
67
68 #include <miscfs/genfs/genfs.h>
69
70 MODULE(MODULE_CLASS_MISC, mqueue, NULL);
71
72 /* System-wide limits. */
73 u_int mq_open_max = MQ_OPEN_MAX;
74 static u_int mq_prio_max = MQ_PRIO_MAX;
75 u_int mq_max_msgsize = 16 * MQ_DEF_MSGSIZE;
76 u_int mq_def_maxmsg = 32;
77 u_int mq_max_maxmsg = 16 * 32;
78
79 static pool_cache_t mqmsg_cache __read_mostly;
80 static kmutex_t mqlist_lock __cacheline_aligned;
81 static LIST_HEAD(, mqueue) mqueue_head __cacheline_aligned;
82
83 static kauth_listener_t mq_listener;
84
85 static int mqueue_sysinit(void);
86 static int mqueue_sysfini(bool);
87 static int mq_poll_fop(file_t *, int);
88 static int mq_stat_fop(file_t *, struct stat *);
89 static int mq_close_fop(file_t *);
90
91 static const struct fileops mqops = {
92 .fo_name = "mq",
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_restart = fnullop_restart,
102 };
103
104 static const struct syscall_package mqueue_syscalls[] = {
105 { SYS_mq_open, 0, (sy_call_t *)sys_mq_open },
106 { SYS_mq_close, 0, (sy_call_t *)sys_mq_close },
107 { SYS_mq_unlink, 0, (sy_call_t *)sys_mq_unlink },
108 { SYS_mq_getattr, 0, (sy_call_t *)sys_mq_getattr },
109 { SYS_mq_setattr, 0, (sy_call_t *)sys_mq_setattr },
110 { SYS_mq_notify, 0, (sy_call_t *)sys_mq_notify },
111 { SYS_mq_send, 0, (sy_call_t *)sys_mq_send },
112 { SYS_mq_receive, 0, (sy_call_t *)sys_mq_receive },
113 { SYS___mq_timedsend50, 0, (sy_call_t *)sys___mq_timedsend50 },
114 { SYS___mq_timedreceive50, 0, (sy_call_t *)sys___mq_timedreceive50 },
115 { 0, 0, NULL }
116 };
117
118 static int
mq_listener_cb(kauth_cred_t cred,kauth_action_t action,void * cookie,void * arg0,void * arg1,void * arg2,void * arg3)119 mq_listener_cb(kauth_cred_t cred, kauth_action_t action, void *cookie,
120 void *arg0, void *arg1, void *arg2, void *arg3)
121 {
122 mqueue_t *mq;
123 int result;
124
125 if (action != KAUTH_SYSTEM_MQUEUE)
126 return KAUTH_RESULT_DEFER;
127
128 result = KAUTH_RESULT_DEFER;
129
130 mq = arg1;
131
132 if (kauth_cred_geteuid(cred) == mq->mq_euid)
133 result = KAUTH_RESULT_ALLOW;
134
135 return result;
136 }
137
138 /*
139 * Initialisation and unloading of POSIX message queue subsystem.
140 */
141
142 static int
mqueue_sysinit(void)143 mqueue_sysinit(void)
144 {
145 int error;
146
147 mqmsg_cache = pool_cache_init(MQ_DEF_MSGSIZE, coherency_unit,
148 0, 0, "mqmsgpl", NULL, IPL_NONE, NULL, NULL, NULL);
149 mutex_init(&mqlist_lock, MUTEX_DEFAULT, IPL_NONE);
150 LIST_INIT(&mqueue_head);
151
152 error = syscall_establish(NULL, mqueue_syscalls);
153 mq_listener = kauth_listen_scope(KAUTH_SCOPE_SYSTEM,
154 mq_listener_cb, NULL);
155 return error;
156 }
157
158 static int
mqueue_sysfini(bool interface)159 mqueue_sysfini(bool interface)
160 {
161
162 if (interface) {
163 int error;
164 bool inuse;
165
166 /* Stop syscall activity. */
167 error = syscall_disestablish(NULL, mqueue_syscalls);
168 if (error)
169 return error;
170 /* Check if there are any message queues in use. */
171 mutex_enter(&mqlist_lock);
172 inuse = !LIST_EMPTY(&mqueue_head);
173 mutex_exit(&mqlist_lock);
174 if (inuse) {
175 error = syscall_establish(NULL, mqueue_syscalls);
176 KASSERT(error == 0);
177 return EBUSY;
178 }
179 }
180
181 kauth_unlisten_scope(mq_listener);
182
183 mutex_destroy(&mqlist_lock);
184 pool_cache_destroy(mqmsg_cache);
185 return 0;
186 }
187
188 /*
189 * Module interface.
190 */
191 static int
mqueue_modcmd(modcmd_t cmd,void * arg)192 mqueue_modcmd(modcmd_t cmd, void *arg)
193 {
194
195 switch (cmd) {
196 case MODULE_CMD_INIT:
197 return mqueue_sysinit();
198 case MODULE_CMD_FINI:
199 return mqueue_sysfini(true);
200 default:
201 return ENOTTY;
202 }
203 }
204
205 /*
206 * Free the message.
207 */
208 static void
mqueue_freemsg(struct mq_msg * msg,const size_t size)209 mqueue_freemsg(struct mq_msg *msg, const size_t size)
210 {
211
212 if (size > MQ_DEF_MSGSIZE) {
213 kmem_free(msg, size);
214 } else {
215 pool_cache_put(mqmsg_cache, msg);
216 }
217 }
218
219 /*
220 * Destroy the message queue.
221 */
222 static void
mqueue_destroy(struct mqueue * mq)223 mqueue_destroy(struct mqueue *mq)
224 {
225 struct mq_msg *msg;
226 size_t msz;
227 u_int i;
228
229 /* Note MQ_PQSIZE + 1. */
230 for (i = 0; i <= MQ_PQSIZE; i++) {
231 while ((msg = TAILQ_FIRST(&mq->mq_head[i])) != NULL) {
232 TAILQ_REMOVE(&mq->mq_head[i], msg, msg_queue);
233 msz = sizeof(struct mq_msg) + msg->msg_len;
234 mqueue_freemsg(msg, msz);
235 }
236 }
237 if (mq->mq_name) {
238 kmem_free(mq->mq_name, MQ_NAMELEN);
239 }
240 seldestroy(&mq->mq_rsel);
241 seldestroy(&mq->mq_wsel);
242 cv_destroy(&mq->mq_send_cv);
243 cv_destroy(&mq->mq_recv_cv);
244 mutex_destroy(&mq->mq_mtx);
245 kmem_free(mq, sizeof(struct mqueue));
246 }
247
248 /*
249 * mqueue_lookup: lookup for file name in general list of message queues.
250 *
251 * => locks the message queue on success
252 */
253 static mqueue_t *
mqueue_lookup(const char * name)254 mqueue_lookup(const char *name)
255 {
256 mqueue_t *mq;
257
258 KASSERT(mutex_owned(&mqlist_lock));
259
260 LIST_FOREACH(mq, &mqueue_head, mq_list) {
261 if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) {
262 mutex_enter(&mq->mq_mtx);
263 return mq;
264 }
265 }
266 return NULL;
267 }
268
269 /*
270 * mqueue_get: get the mqueue from the descriptor.
271 *
272 * => locks the message queue, if found.
273 * => holds a reference on the file descriptor.
274 */
275 int
mqueue_get(mqd_t mqd,int fflag,mqueue_t ** mqret)276 mqueue_get(mqd_t mqd, int fflag, mqueue_t **mqret)
277 {
278 const int fd = (int)mqd;
279 mqueue_t *mq;
280 file_t *fp;
281
282 fp = fd_getfile(fd);
283 if (__predict_false(fp == NULL)) {
284 return EBADF;
285 }
286 if (__predict_false(fp->f_type != DTYPE_MQUEUE)) {
287 fd_putfile(fd);
288 return EBADF;
289 }
290 if (fflag && (fp->f_flag & fflag) == 0) {
291 fd_putfile(fd);
292 return EBADF;
293 }
294 mq = fp->f_mqueue;
295 mutex_enter(&mq->mq_mtx);
296
297 *mqret = mq;
298 return 0;
299 }
300
301 /*
302 * mqueue_linear_insert: perform linear insert according to the message
303 * priority into the reserved queue (MQ_PQRESQ). Reserved queue is a
304 * sorted list used only when mq_prio_max is increased via sysctl.
305 */
306 static inline void
mqueue_linear_insert(struct mqueue * mq,struct mq_msg * msg)307 mqueue_linear_insert(struct mqueue *mq, struct mq_msg *msg)
308 {
309 struct mq_msg *mit;
310
311 TAILQ_FOREACH(mit, &mq->mq_head[MQ_PQRESQ], msg_queue) {
312 if (msg->msg_prio > mit->msg_prio)
313 break;
314 }
315 if (mit == NULL) {
316 TAILQ_INSERT_TAIL(&mq->mq_head[MQ_PQRESQ], msg, msg_queue);
317 } else {
318 TAILQ_INSERT_BEFORE(mit, msg, msg_queue);
319 }
320 }
321
322 static int
mq_stat_fop(file_t * fp,struct stat * st)323 mq_stat_fop(file_t *fp, struct stat *st)
324 {
325 struct mqueue *mq = fp->f_mqueue;
326
327 memset(st, 0, sizeof(*st));
328
329 mutex_enter(&mq->mq_mtx);
330 st->st_mode = mq->mq_mode;
331 st->st_uid = mq->mq_euid;
332 st->st_gid = mq->mq_egid;
333 st->st_atimespec = mq->mq_atime;
334 st->st_mtimespec = mq->mq_mtime;
335 st->st_ctimespec = st->st_birthtimespec = mq->mq_btime;
336 st->st_uid = kauth_cred_geteuid(fp->f_cred);
337 st->st_gid = kauth_cred_getegid(fp->f_cred);
338 mutex_exit(&mq->mq_mtx);
339
340 return 0;
341 }
342
343 static int
mq_poll_fop(file_t * fp,int events)344 mq_poll_fop(file_t *fp, int events)
345 {
346 struct mqueue *mq = fp->f_mqueue;
347 struct mq_attr *mqattr;
348 int revents = 0;
349
350 mutex_enter(&mq->mq_mtx);
351 mqattr = &mq->mq_attrib;
352 if (events & (POLLIN | POLLRDNORM)) {
353 /* Ready for receiving, if there are messages in the queue. */
354 if (mqattr->mq_curmsgs)
355 revents |= events & (POLLIN | POLLRDNORM);
356 else
357 selrecord(curlwp, &mq->mq_rsel);
358 }
359 if (events & (POLLOUT | POLLWRNORM)) {
360 /* Ready for sending, if the message queue is not full. */
361 if (mqattr->mq_curmsgs < mqattr->mq_maxmsg)
362 revents |= events & (POLLOUT | POLLWRNORM);
363 else
364 selrecord(curlwp, &mq->mq_wsel);
365 }
366 mutex_exit(&mq->mq_mtx);
367
368 return revents;
369 }
370
371 static int
mq_close_fop(file_t * fp)372 mq_close_fop(file_t *fp)
373 {
374 proc_t *p = curproc;
375 mqueue_t *mq = fp->f_mqueue;
376 bool destroy = false;
377
378 mutex_enter(&mq->mq_mtx);
379 KASSERT(mq->mq_refcnt > 0);
380 if (--mq->mq_refcnt == 0) {
381 /* Destroy if the last reference and unlinked. */
382 destroy = (mq->mq_attrib.mq_flags & MQ_UNLINKED) != 0;
383 }
384 mutex_exit(&mq->mq_mtx);
385
386 if (destroy) {
387 mqueue_destroy(mq);
388 }
389 atomic_dec_uint(&p->p_mqueue_cnt);
390 return 0;
391 }
392
393 static int
mqueue_access(mqueue_t * mq,int access,kauth_cred_t cred)394 mqueue_access(mqueue_t *mq, int access, kauth_cred_t cred)
395 {
396 accmode_t accmode = 0;
397
398 /* Note the difference between VREAD/VWRITE and FREAD/FWRITE. */
399 if (access & FREAD) {
400 accmode |= VREAD;
401 }
402 if (access & FWRITE) {
403 accmode |= VWRITE;
404 }
405 if (genfs_can_access(NULL, cred, mq->mq_euid, mq->mq_egid,
406 mq->mq_mode, NULL, accmode)) {
407 return EACCES;
408 }
409 return 0;
410 }
411
412 static int
mqueue_create(lwp_t * l,char * name,struct mq_attr * attr,mode_t mode,int oflag,mqueue_t ** mqret)413 mqueue_create(lwp_t *l, char *name, struct mq_attr *attr, mode_t mode,
414 int oflag, mqueue_t **mqret)
415 {
416 proc_t *p = l->l_proc;
417 struct cwdinfo *cwdi = p->p_cwdi;
418 mqueue_t *mq;
419 u_int i;
420
421 /* Empty name is invalid. */
422 if (name[0] == '\0') {
423 return EINVAL;
424 }
425
426 /* Check for mqueue attributes. */
427 if (attr) {
428 if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > mq_max_maxmsg ||
429 attr->mq_msgsize <= 0 ||
430 attr->mq_msgsize > mq_max_msgsize) {
431 return EINVAL;
432 }
433 attr->mq_curmsgs = 0;
434 }
435
436 /*
437 * Allocate new message queue, initialize data structures, copy the
438 * name attributes. Note that the initial reference is set here.
439 */
440 mq = kmem_zalloc(sizeof(mqueue_t), KM_SLEEP);
441
442 mutex_init(&mq->mq_mtx, MUTEX_DEFAULT, IPL_NONE);
443 cv_init(&mq->mq_send_cv, "mqsendcv");
444 cv_init(&mq->mq_recv_cv, "mqrecvcv");
445 for (i = 0; i < (MQ_PQSIZE + 1); i++) {
446 TAILQ_INIT(&mq->mq_head[i]);
447 }
448 selinit(&mq->mq_rsel);
449 selinit(&mq->mq_wsel);
450 mq->mq_name = name;
451 mq->mq_refcnt = 1;
452
453 if (attr != NULL) {
454 memcpy(&mq->mq_attrib, attr, sizeof(struct mq_attr));
455 } else {
456 memset(&mq->mq_attrib, 0, sizeof(struct mq_attr));
457 mq->mq_attrib.mq_maxmsg = mq_def_maxmsg;
458 mq->mq_attrib.mq_msgsize = MQ_DEF_MSGSIZE - sizeof(struct mq_msg);
459 }
460
461 CTASSERT((O_MASK & (MQ_UNLINKED | MQ_RECEIVE)) == 0);
462 mq->mq_attrib.mq_flags = (O_MASK & oflag);
463
464 /* Store mode and effective UID with GID. */
465 mq->mq_mode = ((mode & ~cwdi->cwdi_cmask) & ALLPERMS) & ~S_ISTXT;
466 mq->mq_euid = kauth_cred_geteuid(l->l_cred);
467 mq->mq_egid = kauth_cred_getegid(l->l_cred);
468
469 *mqret = mq;
470 return 0;
471 }
472
473 /*
474 * Helper function for mq_open() - note that "u_name" is a userland pointer,
475 * while "attr" is a kernel pointer!
476 */
477 int
mq_handle_open(struct lwp * l,const char * u_name,int oflag,mode_t mode,struct mq_attr * attr,register_t * retval)478 mq_handle_open(struct lwp *l, const char *u_name, int oflag, mode_t mode,
479 struct mq_attr *attr, register_t *retval)
480 {
481 struct proc *p = l->l_proc;
482 struct mqueue *mq, *mq_new = NULL;
483 int mqd, error;
484 file_t *fp;
485 char *name;
486
487 /* Get the name from the user-space. */
488 name = kmem_alloc(MQ_NAMELEN, KM_SLEEP);
489 error = copyinstr(u_name, name, MQ_NAMELEN - 1, NULL);
490 if (error) {
491 kmem_free(name, MQ_NAMELEN);
492 return error;
493 }
494
495 /* Allocate file structure and descriptor. */
496 error = fd_allocfile(&fp, &mqd);
497 if (error) {
498 kmem_free(name, MQ_NAMELEN);
499 return error;
500 }
501
502 /* Account and check for the limit. */
503 if (atomic_inc_uint_nv(&p->p_mqueue_cnt) > mq_open_max) {
504 atomic_dec_uint(&p->p_mqueue_cnt);
505 error = EMFILE;
506 goto err;
507 }
508
509 fp->f_type = DTYPE_MQUEUE;
510 fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE);
511 fp->f_ops = &mqops;
512
513 if (oflag & O_CREAT) {
514 /* Create a new message queue. */
515 error = mqueue_create(l, name, attr, mode, oflag, &mq_new);
516 if (error) {
517 goto err;
518 }
519 KASSERT(mq_new != NULL);
520 }
521
522 /* Lookup for a message queue with such name. */
523 mutex_enter(&mqlist_lock);
524 mq = mqueue_lookup(name);
525 if (mq) {
526 KASSERT(mutex_owned(&mq->mq_mtx));
527 mutex_exit(&mqlist_lock);
528
529 /* Check for exclusive create. */
530 if (oflag & O_EXCL) {
531 mutex_exit(&mq->mq_mtx);
532 error = EEXIST;
533 goto err;
534 }
535
536 /* Verify permissions. */
537 if (mqueue_access(mq, fp->f_flag, l->l_cred) != 0) {
538 mutex_exit(&mq->mq_mtx);
539 error = EACCES;
540 goto err;
541 }
542
543 /* If we have the access, add a new reference. */
544 mq->mq_refcnt++;
545 mutex_exit(&mq->mq_mtx);
546 } else {
547 /* Fail if not found and not creating. */
548 if ((oflag & O_CREAT) == 0) {
549 mutex_exit(&mqlist_lock);
550 KASSERT(mq_new == NULL);
551 error = ENOENT;
552 goto err;
553 }
554
555 /* Initial timestamps. */
556 mq = mq_new;
557 getnanotime(&mq->mq_btime);
558 mq->mq_atime = mq->mq_mtime = mq->mq_btime;
559
560 /*
561 * Finally, insert message queue into the list.
562 * Note: it already has the initial reference.
563 */
564 LIST_INSERT_HEAD(&mqueue_head, mq, mq_list);
565 mutex_exit(&mqlist_lock);
566
567 mq_new = NULL;
568 name = NULL;
569 }
570 KASSERT(mq != NULL);
571 fp->f_mqueue = mq;
572 fd_affix(p, fp, mqd);
573 *retval = mqd;
574 err:
575 if (error) {
576 fd_abort(p, fp, mqd);
577 }
578 if (mq_new) {
579 /* Note: will free the 'name'. */
580 mqueue_destroy(mq_new);
581 } else if (name) {
582 kmem_free(name, MQ_NAMELEN);
583 }
584 return error;
585 }
586
587 /*
588 * General mqueue system calls.
589 */
590
591 int
sys_mq_open(struct lwp * l,const struct sys_mq_open_args * uap,register_t * retval)592 sys_mq_open(struct lwp *l, const struct sys_mq_open_args *uap,
593 register_t *retval)
594 {
595 /* {
596 syscallarg(const char *) name;
597 syscallarg(int) oflag;
598 syscallarg(mode_t) mode;
599 syscallarg(struct mq_attr) attr;
600 } */
601 struct mq_attr *attr = NULL, a;
602 int error;
603
604 if ((SCARG(uap, oflag) & O_EXEC) != 0)
605 return EINVAL;
606
607 if ((SCARG(uap, oflag) & O_CREAT) != 0 && SCARG(uap, attr) != NULL) {
608 error = copyin(SCARG(uap, attr), &a, sizeof(a));
609 if (error)
610 return error;
611 attr = &a;
612 }
613
614 return mq_handle_open(l, SCARG(uap, name), SCARG(uap, oflag),
615 SCARG(uap, mode), attr, retval);
616 }
617
618 int
sys_mq_close(struct lwp * l,const struct sys_mq_close_args * uap,register_t * retval)619 sys_mq_close(struct lwp *l, const struct sys_mq_close_args *uap,
620 register_t *retval)
621 {
622
623 return sys_close(l, (const void *)uap, retval);
624 }
625
626 /*
627 * Primary mq_recv1() function.
628 */
629 int
mq_recv1(mqd_t mqdes,void * msg_ptr,size_t msg_len,u_int * msg_prio,struct timespec * ts,ssize_t * mlen)630 mq_recv1(mqd_t mqdes, void *msg_ptr, size_t msg_len, u_int *msg_prio,
631 struct timespec *ts, ssize_t *mlen)
632 {
633 struct mqueue *mq;
634 struct mq_msg *msg = NULL;
635 struct mq_attr *mqattr;
636 u_int idx;
637 int error;
638
639 error = mqueue_get(mqdes, FREAD, &mq);
640 if (error) {
641 return 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 = ts2timo(CLOCK_REALTIME, TIMER_ABSTIME, ts, &t,
662 NULL);
663 if (error)
664 goto error;
665 } else
666 t = 0;
667 /*
668 * Block until someone sends the message.
669 * While doing this, notification should not be sent.
670 */
671 mqattr->mq_flags |= MQ_RECEIVE;
672 error = cv_timedwait_sig(&mq->mq_send_cv, &mq->mq_mtx, t);
673 mqattr->mq_flags &= ~MQ_RECEIVE;
674 if (error || (mqattr->mq_flags & MQ_UNLINKED)) {
675 error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR;
676 goto error;
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 KASSERT(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 KASSERT((MQ_PQSIZE - idx) == msg->msg_prio);
697 mq->mq_bitmap &= ~(1U << --idx);
698 }
699
700 /* Decrement the counter and signal waiter, if any */
701 mqattr->mq_curmsgs--;
702 cv_signal(&mq->mq_recv_cv);
703
704 /* Ready for sending now */
705 selnotify(&mq->mq_wsel, POLLOUT | POLLWRNORM, 0);
706 error:
707 mutex_exit(&mq->mq_mtx);
708 fd_putfile((int)mqdes);
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
sys_mq_receive(struct lwp * l,const struct sys_mq_receive_args * uap,register_t * retval)727 sys_mq_receive(struct lwp *l, const struct sys_mq_receive_args *uap,
728 register_t *retval)
729 {
730 /* {
731 syscallarg(mqd_t) mqdes;
732 syscallarg(char *) msg_ptr;
733 syscallarg(size_t) msg_len;
734 syscallarg(unsigned *) msg_prio;
735 } */
736 ssize_t mlen;
737 int error;
738
739 error = mq_recv1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
740 SCARG(uap, msg_len), SCARG(uap, msg_prio), NULL, &mlen);
741 if (error == 0)
742 *retval = mlen;
743
744 return error;
745 }
746
747 int
sys___mq_timedreceive50(struct lwp * l,const struct sys___mq_timedreceive50_args * uap,register_t * retval)748 sys___mq_timedreceive50(struct lwp *l,
749 const struct sys___mq_timedreceive50_args *uap, register_t *retval)
750 {
751 /* {
752 syscallarg(mqd_t) mqdes;
753 syscallarg(char *) msg_ptr;
754 syscallarg(size_t) msg_len;
755 syscallarg(unsigned *) msg_prio;
756 syscallarg(const struct timespec *) abs_timeout;
757 } */
758 struct timespec ts, *tsp;
759 ssize_t mlen;
760 int error;
761
762 /* Get and convert time value */
763 if (SCARG(uap, abs_timeout)) {
764 error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
765 if (error)
766 return error;
767 tsp = &ts;
768 } else {
769 tsp = NULL;
770 }
771
772 error = mq_recv1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
773 SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp, &mlen);
774 if (error == 0)
775 *retval = mlen;
776
777 return error;
778 }
779
780 /*
781 * Primary mq_send1() function.
782 */
783 int
mq_send1(mqd_t mqdes,const char * msg_ptr,size_t msg_len,u_int msg_prio,struct timespec * ts)784 mq_send1(mqd_t mqdes, const char *msg_ptr, size_t msg_len, u_int msg_prio,
785 struct timespec *ts)
786 {
787 struct mqueue *mq;
788 struct mq_msg *msg;
789 struct mq_attr *mqattr;
790 struct proc *notify = NULL;
791 ksiginfo_t ksi;
792 size_t size;
793 int error;
794
795 /* Check the priority range */
796 if (msg_prio >= mq_prio_max)
797 return EINVAL;
798
799 /* Allocate a new message */
800 if (msg_len > mq_max_msgsize)
801 return EMSGSIZE;
802 size = sizeof(struct mq_msg) + msg_len;
803 if (size > mq_max_msgsize)
804 return EMSGSIZE;
805
806 if (size > MQ_DEF_MSGSIZE) {
807 msg = kmem_alloc(size, KM_SLEEP);
808 } else {
809 msg = pool_cache_get(mqmsg_cache, PR_WAITOK);
810 }
811
812 /* Get the data from user-space */
813 error = copyin(msg_ptr, msg->msg_ptr, msg_len);
814 if (error) {
815 mqueue_freemsg(msg, size);
816 return error;
817 }
818 msg->msg_len = msg_len;
819 msg->msg_prio = msg_prio;
820
821 error = mqueue_get(mqdes, FWRITE, &mq);
822 if (error) {
823 mqueue_freemsg(msg, size);
824 return error;
825 }
826 getnanotime(&mq->mq_mtime);
827 mqattr = &mq->mq_attrib;
828
829 /* Check the message size limit */
830 if (msg_len <= 0 || msg_len > mqattr->mq_msgsize) {
831 error = EMSGSIZE;
832 goto error;
833 }
834
835 /* Check if queue is full */
836 while (mqattr->mq_curmsgs >= mqattr->mq_maxmsg) {
837 int t;
838
839 if (mqattr->mq_flags & O_NONBLOCK) {
840 error = EAGAIN;
841 goto error;
842 }
843 if (ts) {
844 error = ts2timo(CLOCK_REALTIME, TIMER_ABSTIME, ts, &t,
845 NULL);
846 if (error)
847 goto error;
848 } else
849 t = 0;
850 /* Block until queue becomes available */
851 error = cv_timedwait_sig(&mq->mq_recv_cv, &mq->mq_mtx, t);
852 if (error || (mqattr->mq_flags & MQ_UNLINKED)) {
853 error = (error == EWOULDBLOCK) ? ETIMEDOUT : error;
854 goto error;
855 }
856 }
857 KASSERT(mqattr->mq_curmsgs < mqattr->mq_maxmsg);
858
859 /*
860 * Insert message into the queue, according to the priority.
861 * Note the difference between index and priority.
862 */
863 if (__predict_true(msg_prio < MQ_PQSIZE)) {
864 u_int idx = MQ_PQSIZE - msg_prio;
865
866 KASSERT(idx != MQ_PQRESQ);
867 TAILQ_INSERT_TAIL(&mq->mq_head[idx], msg, msg_queue);
868 mq->mq_bitmap |= (1U << --idx);
869 } else {
870 mqueue_linear_insert(mq, msg);
871 }
872
873 /* Check for the notify */
874 if (mqattr->mq_curmsgs == 0 && mq->mq_notify_proc &&
875 (mqattr->mq_flags & MQ_RECEIVE) == 0 &&
876 mq->mq_sig_notify.sigev_notify == SIGEV_SIGNAL) {
877 /* Initialize the signal */
878 KSI_INIT(&ksi);
879 ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;
880 ksi.ksi_code = SI_MESGQ;
881 ksi.ksi_value = mq->mq_sig_notify.sigev_value;
882 /* Unregister the process */
883 notify = mq->mq_notify_proc;
884 mq->mq_notify_proc = NULL;
885 }
886
887 /* Increment the counter and signal waiter, if any */
888 mqattr->mq_curmsgs++;
889 cv_signal(&mq->mq_send_cv);
890
891 /* Ready for receiving now */
892 selnotify(&mq->mq_rsel, POLLIN | POLLRDNORM, 0);
893 error:
894 mutex_exit(&mq->mq_mtx);
895 fd_putfile((int)mqdes);
896
897 if (error) {
898 mqueue_freemsg(msg, size);
899 } else if (notify) {
900 /* Send the notify, if needed */
901 mutex_enter(&proc_lock);
902 kpsignal(notify, &ksi, NULL);
903 mutex_exit(&proc_lock);
904 }
905 return error;
906 }
907
908 int
sys_mq_send(struct lwp * l,const struct sys_mq_send_args * uap,register_t * retval)909 sys_mq_send(struct lwp *l, const struct sys_mq_send_args *uap,
910 register_t *retval)
911 {
912 /* {
913 syscallarg(mqd_t) mqdes;
914 syscallarg(const char *) msg_ptr;
915 syscallarg(size_t) msg_len;
916 syscallarg(unsigned) msg_prio;
917 } */
918
919 return mq_send1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
920 SCARG(uap, msg_len), SCARG(uap, msg_prio), NULL);
921 }
922
923 int
sys___mq_timedsend50(struct lwp * l,const struct sys___mq_timedsend50_args * uap,register_t * retval)924 sys___mq_timedsend50(struct lwp *l, const struct sys___mq_timedsend50_args *uap,
925 register_t *retval)
926 {
927 /* {
928 syscallarg(mqd_t) mqdes;
929 syscallarg(const char *) msg_ptr;
930 syscallarg(size_t) msg_len;
931 syscallarg(unsigned) msg_prio;
932 syscallarg(const struct timespec *) abs_timeout;
933 } */
934 struct timespec ts, *tsp;
935 int error;
936
937 /* Get and convert time value */
938 if (SCARG(uap, abs_timeout)) {
939 error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts));
940 if (error)
941 return error;
942 tsp = &ts;
943 } else {
944 tsp = NULL;
945 }
946
947 return mq_send1(SCARG(uap, mqdes), SCARG(uap, msg_ptr),
948 SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp);
949 }
950
951 int
sys_mq_notify(struct lwp * l,const struct sys_mq_notify_args * uap,register_t * retval)952 sys_mq_notify(struct lwp *l, const struct sys_mq_notify_args *uap,
953 register_t *retval)
954 {
955 /* {
956 syscallarg(mqd_t) mqdes;
957 syscallarg(const struct sigevent *) notification;
958 } */
959 struct mqueue *mq;
960 struct sigevent sig;
961 int error;
962
963 if (SCARG(uap, notification)) {
964 /* Get the signal from user-space */
965 error = copyin(SCARG(uap, notification), &sig,
966 sizeof(struct sigevent));
967 if (error)
968 return error;
969 if (sig.sigev_notify == SIGEV_SIGNAL &&
970 (sig.sigev_signo <=0 || sig.sigev_signo >= NSIG))
971 return EINVAL;
972 }
973
974 error = mqueue_get(SCARG(uap, mqdes), 0, &mq);
975 if (error) {
976 return error;
977 }
978 if (SCARG(uap, notification)) {
979 /* Register notification: set the signal and target process */
980 if (mq->mq_notify_proc == NULL) {
981 memcpy(&mq->mq_sig_notify, &sig,
982 sizeof(struct sigevent));
983 mq->mq_notify_proc = l->l_proc;
984 } else {
985 /* Fail if someone else already registered */
986 error = EBUSY;
987 }
988 } else {
989 /* Unregister the notification */
990 mq->mq_notify_proc = NULL;
991 }
992 mutex_exit(&mq->mq_mtx);
993 fd_putfile((int)SCARG(uap, mqdes));
994
995 return error;
996 }
997
998 int
sys_mq_getattr(struct lwp * l,const struct sys_mq_getattr_args * uap,register_t * retval)999 sys_mq_getattr(struct lwp *l, const struct sys_mq_getattr_args *uap,
1000 register_t *retval)
1001 {
1002 /* {
1003 syscallarg(mqd_t) mqdes;
1004 syscallarg(struct mq_attr *) mqstat;
1005 } */
1006 struct mqueue *mq;
1007 struct mq_attr attr;
1008 int error;
1009
1010 error = mqueue_get(SCARG(uap, mqdes), 0, &mq);
1011 if (error) {
1012 return error;
1013 }
1014 memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
1015 mutex_exit(&mq->mq_mtx);
1016 fd_putfile((int)SCARG(uap, mqdes));
1017
1018 return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr));
1019 }
1020
1021 int
sys_mq_setattr(struct lwp * l,const struct sys_mq_setattr_args * uap,register_t * retval)1022 sys_mq_setattr(struct lwp *l, const struct sys_mq_setattr_args *uap,
1023 register_t *retval)
1024 {
1025 /* {
1026 syscallarg(mqd_t) mqdes;
1027 syscallarg(const struct mq_attr *) mqstat;
1028 syscallarg(struct mq_attr *) omqstat;
1029 } */
1030 struct mqueue *mq;
1031 struct mq_attr attr;
1032 int error, nonblock;
1033
1034 error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr));
1035 if (error)
1036 return error;
1037 nonblock = (attr.mq_flags & O_NONBLOCK);
1038
1039 error = mqueue_get(SCARG(uap, mqdes), 0, &mq);
1040 if (error) {
1041 return error;
1042 }
1043
1044 /* Copy the old attributes, if needed */
1045 if (SCARG(uap, omqstat)) {
1046 memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr));
1047 }
1048
1049 /* Ignore everything, except O_NONBLOCK */
1050 if (nonblock)
1051 mq->mq_attrib.mq_flags |= O_NONBLOCK;
1052 else
1053 mq->mq_attrib.mq_flags &= ~O_NONBLOCK;
1054
1055 mutex_exit(&mq->mq_mtx);
1056 fd_putfile((int)SCARG(uap, mqdes));
1057
1058 /*
1059 * Copy the data to the user-space.
1060 * Note: According to POSIX, the new attributes should not be set in
1061 * case of fail - this would be violated.
1062 */
1063 if (SCARG(uap, omqstat))
1064 error = copyout(&attr, SCARG(uap, omqstat),
1065 sizeof(struct mq_attr));
1066
1067 return error;
1068 }
1069
1070 int
sys_mq_unlink(struct lwp * l,const struct sys_mq_unlink_args * uap,register_t * retval)1071 sys_mq_unlink(struct lwp *l, const struct sys_mq_unlink_args *uap,
1072 register_t *retval)
1073 {
1074 /* {
1075 syscallarg(const char *) name;
1076 } */
1077 mqueue_t *mq;
1078 char *name;
1079 int error, refcnt = 0;
1080
1081 /* Get the name from the user-space */
1082 name = kmem_alloc(MQ_NAMELEN, KM_SLEEP);
1083 error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL);
1084 if (error) {
1085 kmem_free(name, MQ_NAMELEN);
1086 return error;
1087 }
1088
1089 mutex_enter(&mqlist_lock);
1090 mq = mqueue_lookup(name);
1091 if (mq == NULL) {
1092 error = ENOENT;
1093 goto err;
1094 }
1095 KASSERT(mutex_owned(&mq->mq_mtx));
1096
1097 /* Verify permissions. */
1098 if (kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MQUEUE, 0, mq,
1099 NULL, NULL)) {
1100 mutex_exit(&mq->mq_mtx);
1101 error = EACCES;
1102 goto err;
1103 }
1104
1105 /* Remove and destroy if no references. */
1106 LIST_REMOVE(mq, mq_list);
1107 refcnt = mq->mq_refcnt;
1108 if (refcnt) {
1109 /* Mark as unlinked, if there are references. */
1110 mq->mq_attrib.mq_flags |= MQ_UNLINKED;
1111 }
1112
1113 /* Wake up waiters, if there are any. */
1114 cv_broadcast(&mq->mq_send_cv);
1115 cv_broadcast(&mq->mq_recv_cv);
1116
1117 selnotify(&mq->mq_rsel, POLLHUP, 0);
1118 selnotify(&mq->mq_wsel, POLLHUP, 0);
1119
1120 mutex_exit(&mq->mq_mtx);
1121 err:
1122 mutex_exit(&mqlist_lock);
1123 /*
1124 * If last reference - destroy the message queue. Otherwise,
1125 * the last mq_close() call will do that.
1126 */
1127 if (!error && refcnt == 0) {
1128 mqueue_destroy(mq);
1129 }
1130 kmem_free(name, MQ_NAMELEN);
1131
1132 return error;
1133 }
1134
1135 /*
1136 * System control nodes.
1137 */
1138 SYSCTL_SETUP(mqueue_sysctl_init, "mqueue systl")
1139 {
1140 const struct sysctlnode *node = NULL;
1141
1142 sysctl_createv(clog, 0, NULL, NULL,
1143 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE,
1144 CTLTYPE_INT, "posix_msg",
1145 SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
1146 "Message Passing option to which the "
1147 "system attempts to conform"),
1148 NULL, _POSIX_MESSAGE_PASSING, NULL, 0,
1149 CTL_KERN, CTL_CREATE, CTL_EOL);
1150 sysctl_createv(clog, 0, NULL, &node,
1151 CTLFLAG_PERMANENT,
1152 CTLTYPE_NODE, "mqueue",
1153 SYSCTL_DESCR("Message queue options"),
1154 NULL, 0, NULL, 0,
1155 CTL_KERN, CTL_CREATE, CTL_EOL);
1156
1157 if (node == NULL)
1158 return;
1159
1160 sysctl_createv(clog, 0, &node, NULL,
1161 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1162 CTLTYPE_INT, "mq_open_max",
1163 SYSCTL_DESCR("Maximal number of message queue descriptors "
1164 "that process could open"),
1165 NULL, 0, &mq_open_max, 0,
1166 CTL_CREATE, CTL_EOL);
1167 sysctl_createv(clog, 0, &node, NULL,
1168 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1169 CTLTYPE_INT, "mq_prio_max",
1170 SYSCTL_DESCR("Maximal priority of the message"),
1171 NULL, 0, &mq_prio_max, 0,
1172 CTL_CREATE, CTL_EOL);
1173 sysctl_createv(clog, 0, &node, NULL,
1174 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1175 CTLTYPE_INT, "mq_max_msgsize",
1176 SYSCTL_DESCR("Maximal allowed size of the message"),
1177 NULL, 0, &mq_max_msgsize, 0,
1178 CTL_CREATE, CTL_EOL);
1179 sysctl_createv(clog, 0, &node, NULL,
1180 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1181 CTLTYPE_INT, "mq_def_maxmsg",
1182 SYSCTL_DESCR("Default maximal message count"),
1183 NULL, 0, &mq_def_maxmsg, 0,
1184 CTL_CREATE, CTL_EOL);
1185 sysctl_createv(clog, 0, &node, NULL,
1186 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1187 CTLTYPE_INT, "mq_max_maxmsg",
1188 SYSCTL_DESCR("Maximal allowed message count"),
1189 NULL, 0, &mq_max_maxmsg, 0,
1190 CTL_CREATE, CTL_EOL);
1191
1192 return;
1193 }
1194
1195 /*
1196 * Debugging.
1197 */
1198 #if defined(DDB)
1199
1200 void
mqueue_print_list(void (* pr)(const char *,...))1201 mqueue_print_list(void (*pr)(const char *, ...))
1202 {
1203 struct mqueue *mq;
1204
1205 (*pr)("Global list of the message queues:\n");
1206 (*pr)("%20s %10s %8s %8s %3s %4s %4s %4s\n",
1207 "Name", "Ptr", "Mode", "Flags", "Ref",
1208 "MaxMsg", "MsgSze", "CurMsg");
1209 LIST_FOREACH(mq, &mqueue_head, mq_list) {
1210 (*pr)("%20s %10p %8x %8x %3u %6lu %6lu %6lu\n",
1211 mq->mq_name, mq, mq->mq_mode,
1212 mq->mq_attrib.mq_flags, mq->mq_refcnt,
1213 mq->mq_attrib.mq_maxmsg, mq->mq_attrib.mq_msgsize,
1214 mq->mq_attrib.mq_curmsgs);
1215 }
1216 }
1217
1218 #endif /* defined(DDB) */
1219