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 <stdbool.h> 45 #include <sys/param.h> 46 #include <sys/types.h> 47 #include <sys/errno.h> 48 #include <sys/fcntl.h> 49 #include <sys/file.h> 50 #include <sys/filedesc.h> 51 #include <sys/ucred.h> 52 #include <sys/priv.h> 53 #include <sys/kernel.h> 54 #include <sys/malloc.h> 55 #include <sys/mqueue.h> 56 #include <sys/objcache.h> 57 #include <sys/poll.h> 58 #include <sys/proc.h> 59 #include <sys/queue.h> 60 #include <sys/select.h> 61 #include <sys/serialize.h> 62 #include <sys/signal.h> 63 #include <sys/signalvar.h> 64 #include <sys/spinlock.h> 65 #include <sys/spinlock2.h> 66 #include <sys/stat.h> 67 #include <sys/sysctl.h> 68 #include <sys/sysproto.h> 69 #include <sys/systm.h> 70 #include <sys/lock.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 static u_int mq_max_msgsize = 16 * MQ_DEF_MSGSIZE; 78 static u_int mq_def_maxmsg = 32; 79 80 struct lock mqlist_mtx; 81 static struct objcache * mqmsg_cache; 82 static LIST_HEAD(, mqueue) mqueue_head = 83 LIST_HEAD_INITIALIZER(mqueue_head); 84 85 typedef struct file file_t; /* XXX: Should we put this in sys/types.h ? */ 86 87 /* Function prototypes */ 88 static int mq_poll_fop(file_t *, int, struct ucred *cred); 89 static int mq_stat_fop(file_t *, struct stat *, struct ucred *cred); 90 static int mq_close_fop(file_t *); 91 92 /* Some time-related utility functions */ 93 static int itimespecfix(struct timespec *ts); 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_poll = mq_poll_fop, 102 .fo_stat = mq_stat_fop, 103 .fo_close = mq_close_fop, 104 .fo_kqfilter = badfo_kqfilter, 105 .fo_shutdown = badfo_shutdown 106 }; 107 108 /* Define a new malloc type for message queues */ 109 MALLOC_DECLARE(M_MQBUF); 110 MALLOC_DEFINE(M_MQBUF, "mqueues", "Buffers to message queues"); 111 112 /* Malloc arguments for object cache */ 113 struct objcache_malloc_args mqueue_malloc_args = { 114 sizeof(struct mqueue), M_MQBUF }; 115 116 /* Spinlock around the process list */ 117 extern struct spinlock allproc_spin; 118 119 /* 120 * Initialize POSIX message queue subsystem. 121 */ 122 void 123 mqueue_sysinit(void) 124 { 125 mqmsg_cache = objcache_create("mqmsg_cache", 126 0, /* infinite depot's capacity */ 127 0, /* default magazine's capacity */ 128 NULL, /* constructor */ 129 NULL, /* deconstructor */ 130 NULL, 131 objcache_malloc_alloc, 132 objcache_malloc_free, 133 &mqueue_malloc_args); 134 135 lockinit(&mqlist_mtx, "mqlist_mtx", 0, LK_CANRECURSE); 136 } 137 138 /* 139 * Free the message. 140 */ 141 static void 142 mqueue_freemsg(struct mq_msg *msg, const size_t size) 143 { 144 145 if (size > MQ_DEF_MSGSIZE) 146 kfree(msg, M_MQBUF); 147 else 148 objcache_put(mqmsg_cache, msg); 149 } 150 151 /* 152 * Destroy the message queue. 153 */ 154 static void 155 mqueue_destroy(struct mqueue *mq) 156 { 157 struct mq_msg *msg; 158 size_t msz; 159 u_int i; 160 161 /* Note MQ_PQSIZE + 1. */ 162 for (i = 0; i < MQ_PQSIZE + 1; i++) { 163 while ((msg = TAILQ_FIRST(&mq->mq_head[i])) != NULL) { 164 TAILQ_REMOVE(&mq->mq_head[i], msg, msg_queue); 165 msz = sizeof(struct mq_msg) + msg->msg_len; 166 mqueue_freemsg(msg, msz); 167 } 168 } 169 lockuninit(&mq->mq_mtx); 170 kfree(mq, M_MQBUF); 171 } 172 173 /* 174 * Lookup for file name in general list of message queues. 175 * => locks the message queue 176 */ 177 static void * 178 mqueue_lookup(char *name) 179 { 180 struct mqueue *mq; 181 182 KKASSERT(lockstatus(&mqlist_mtx, curthread)); 183 184 LIST_FOREACH(mq, &mqueue_head, mq_list) { 185 if (strncmp(mq->mq_name, name, MQ_NAMELEN) == 0) { 186 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 187 return mq; 188 } 189 } 190 191 return NULL; 192 } 193 194 /* 195 * mqueue_get: get the mqueue from the descriptor. 196 * => locks the message queue, if found. 197 * => hold a reference on the file descriptor. 198 */ 199 static int 200 mqueue_get(struct lwp *l, mqd_t mqd, file_t **fpr) 201 { 202 struct mqueue *mq; 203 file_t *fp; 204 205 fp = holdfp(curproc->p_fd, (int)mqd, -1); /* XXX: Why -1 ? */ 206 if (__predict_false(fp == NULL)) 207 return EBADF; 208 209 if (__predict_false(fp->f_type != DTYPE_MQUEUE)) { 210 fdrop(fp); 211 return EBADF; 212 } 213 mq = fp->f_data; 214 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 215 216 *fpr = fp; 217 return 0; 218 } 219 220 /* 221 * mqueue_linear_insert: perform linear insert according to the message 222 * priority into the reserved queue (MQ_PQRESQ). Reserved queue is a 223 * sorted list used only when mq_prio_max is increased via sysctl. 224 */ 225 static inline void 226 mqueue_linear_insert(struct mqueue *mq, struct mq_msg *msg) 227 { 228 struct mq_msg *mit; 229 230 TAILQ_FOREACH(mit, &mq->mq_head[MQ_PQRESQ], msg_queue) { 231 if (msg->msg_prio > mit->msg_prio) 232 break; 233 } 234 if (mit == NULL) { 235 TAILQ_INSERT_TAIL(&mq->mq_head[MQ_PQRESQ], msg, msg_queue); 236 } else { 237 TAILQ_INSERT_BEFORE(mit, msg, msg_queue); 238 } 239 } 240 241 /* 242 * Validate input. 243 */ 244 int 245 itimespecfix(struct timespec *ts) 246 { 247 if (ts->tv_sec < 0 || ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000) 248 return (EINVAL); 249 if (ts->tv_sec == 0 && ts->tv_nsec != 0 && ts->tv_nsec < tick * 1000) 250 ts->tv_nsec = tick * 1000; 251 return (0); 252 } 253 254 /* 255 * Compute number of ticks in the specified amount of time. 256 */ 257 int 258 tstohz(const struct timespec *ts) 259 { 260 struct timeval tv; 261 262 /* 263 * usec has great enough resolution for hz, so convert to a 264 * timeval and use tvtohz() above. 265 */ 266 TIMESPEC_TO_TIMEVAL(&tv, ts); 267 return tvtohz_high(&tv); /* XXX Why _high() and not _low() ? */ 268 } 269 270 /* 271 * Converter from struct timespec to the ticks. 272 * Used by mq_timedreceive(), mq_timedsend(). 273 */ 274 int 275 abstimeout2timo(struct timespec *ts, int *timo) 276 { 277 struct timespec tsd; 278 int error; 279 280 getnanotime(&tsd); 281 timespecsub(ts, &tsd); 282 if (ts->tv_sec < 0 || (ts->tv_sec == 0 && ts->tv_nsec <= 0)) { 283 return ETIMEDOUT; 284 } 285 error = itimespecfix(ts); 286 if (error) { 287 return error; 288 } 289 *timo = tstohz(ts); 290 KKASSERT(*timo != 0); 291 292 return 0; 293 } 294 295 static int 296 mq_stat_fop(file_t *fp, struct stat *st, struct ucred *cred) 297 { 298 struct mqueue *mq = fp->f_data; 299 300 (void)memset(st, 0, sizeof(*st)); 301 302 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 303 st->st_mode = mq->mq_mode; 304 st->st_uid = mq->mq_euid; 305 st->st_gid = mq->mq_egid; 306 st->st_atimespec = mq->mq_atime; 307 st->st_mtimespec = mq->mq_mtime; 308 /*st->st_ctimespec = st->st_birthtimespec = mq->mq_btime;*/ 309 st->st_uid = fp->f_cred->cr_uid; 310 st->st_gid = fp->f_cred->cr_svgid; 311 lockmgr(&mq->mq_mtx, LK_RELEASE); 312 313 return 0; 314 } 315 316 static int 317 mq_poll_fop(file_t *fp, int events, struct ucred *cred) 318 { 319 struct mqueue *mq = fp->f_data; 320 int revents = 0; 321 322 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 323 if (events & (POLLIN | POLLRDNORM)) { 324 /* Ready for receiving, if there are messages in the queue */ 325 if (mq->mq_attrib.mq_curmsgs) 326 revents |= (POLLIN | POLLRDNORM); 327 else 328 selrecord(curthread, &mq->mq_rsel); 329 } 330 if (events & (POLLOUT | POLLWRNORM)) { 331 /* Ready for sending, if the message queue is not full */ 332 if (mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg) 333 revents |= (POLLOUT | POLLWRNORM); 334 else 335 selrecord(curthread, &mq->mq_wsel); 336 } 337 lockmgr(&mq->mq_mtx, LK_RELEASE); 338 339 return revents; 340 } 341 342 static int 343 mq_close_fop(file_t *fp) 344 { 345 struct proc *p = curproc; 346 struct mqueue *mq = fp->f_data; 347 bool destroy; 348 349 lockmgr(&mqlist_mtx, LK_EXCLUSIVE); 350 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 351 352 /* Decrease the counters */ 353 p->p_mqueue_cnt--; 354 mq->mq_refcnt--; 355 356 /* Remove notification if registered for this process */ 357 if (mq->mq_notify_proc == p) 358 mq->mq_notify_proc = NULL; 359 360 /* 361 * If this is the last reference and mqueue is marked for unlink, 362 * remove and later destroy the message queue. 363 */ 364 if (mq->mq_refcnt == 0 && (mq->mq_attrib.mq_flags & MQ_UNLINK)) { 365 LIST_REMOVE(mq, mq_list); 366 destroy = true; 367 } else 368 destroy = false; 369 370 lockmgr(&mq->mq_mtx, LK_RELEASE); 371 lockmgr(&mqlist_mtx, LK_RELEASE); 372 373 if (destroy) 374 mqueue_destroy(mq); 375 376 return 0; 377 } 378 379 /* 380 * General mqueue system calls. 381 */ 382 383 int 384 sys_mq_open(struct mq_open_args *uap) 385 { 386 /* { 387 syscallarg(const char *) name; 388 syscallarg(int) oflag; 389 syscallarg(mode_t) mode; 390 syscallarg(struct mq_attr) attr; 391 } */ 392 struct proc *p = curproc; 393 struct mqueue *mq, *mq_new = NULL; 394 file_t *fp; 395 char *name; 396 int mqd, error, oflag; 397 398 /* Check access mode flags */ 399 oflag = SCARG(uap, oflag); 400 if ((oflag & O_ACCMODE) == (O_WRONLY | O_RDWR)) { 401 return EINVAL; 402 } 403 404 /* Get the name from the user-space */ 405 name = kmalloc(MQ_NAMELEN, M_MQBUF, M_WAITOK | M_ZERO); 406 error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL); 407 if (error) { 408 kfree(name, M_MQBUF); 409 return error; 410 } 411 412 if (oflag & O_CREAT) { 413 struct mq_attr attr; 414 u_int i; 415 416 /* Check the limit */ 417 if (p->p_mqueue_cnt == mq_open_max) { 418 kfree(name, M_MQBUF); 419 return EMFILE; 420 } 421 422 /* Empty name is invalid */ 423 if (name[0] == '\0') { 424 kfree(name, M_MQBUF); 425 return EINVAL; 426 } 427 428 /* Check for mqueue attributes */ 429 if (SCARG(uap, attr)) { 430 error = copyin(SCARG(uap, attr), &attr, 431 sizeof(struct mq_attr)); 432 if (error) { 433 kfree(name, M_MQBUF); 434 return error; 435 } 436 if (attr.mq_maxmsg <= 0 || attr.mq_msgsize <= 0 || 437 attr.mq_msgsize > mq_max_msgsize) { 438 kfree(name, M_MQBUF); 439 return EINVAL; 440 } 441 attr.mq_curmsgs = 0; 442 } else { 443 memset(&attr, 0, sizeof(struct mq_attr)); 444 attr.mq_maxmsg = mq_def_maxmsg; 445 attr.mq_msgsize = 446 MQ_DEF_MSGSIZE - sizeof(struct mq_msg); 447 } 448 449 /* 450 * Allocate new mqueue, initialize data structures, 451 * copy the name, attributes and set the flag. 452 */ 453 mq_new = kmalloc(sizeof(struct mqueue), M_MQBUF, M_WAITOK | M_ZERO); 454 455 lockinit(&mq_new->mq_mtx, "mq_new->mq_mtx", 0, LK_CANRECURSE); 456 for (i = 0; i < (MQ_PQSIZE + 1); i++) { 457 TAILQ_INIT(&mq_new->mq_head[i]); 458 } 459 460 strlcpy(mq_new->mq_name, name, MQ_NAMELEN); 461 memcpy(&mq_new->mq_attrib, &attr, sizeof(struct mq_attr)); 462 463 /*CTASSERT((O_MASK & (MQ_UNLINK | MQ_RECEIVE)) == 0);*/ 464 /* mq_new->mq_attrib.mq_flags = (O_MASK & oflag); */ 465 mq_new->mq_attrib.mq_flags = oflag; 466 467 /* Store mode and effective UID with GID */ 468 mq_new->mq_mode = ((SCARG(uap, mode) & 469 ~p->p_fd->fd_cmask) & ALLPERMS) & ~S_ISTXT; 470 mq_new->mq_euid = curproc->p_ucred->cr_uid; 471 mq_new->mq_egid = curproc->p_ucred->cr_svgid; 472 } 473 474 /* Allocate file structure and descriptor */ 475 error = falloc(curproc, &fp, &mqd); 476 if (error) { 477 if (mq_new) 478 mqueue_destroy(mq_new); 479 kfree(name, M_MQBUF); 480 return error; 481 } 482 fp->f_type = DTYPE_MQUEUE; 483 fp->f_flag = FFLAGS(oflag) & (FREAD | FWRITE); 484 fp->f_ops = &mqops; 485 486 /* Look up for mqueue with such name */ 487 lockmgr(&mqlist_mtx, LK_EXCLUSIVE); 488 mq = mqueue_lookup(name); 489 if (mq) { 490 int acc_mode; 491 492 KKASSERT(lockstatus(&mq->mq_mtx, curthread)); 493 494 /* Check if mqueue is not marked as unlinking */ 495 if (mq->mq_attrib.mq_flags & MQ_UNLINK) { 496 error = EACCES; 497 goto exit; 498 } 499 /* Fail if O_EXCL is set, and mqueue already exists */ 500 if ((oflag & O_CREAT) && (oflag & O_EXCL)) { 501 error = EEXIST; 502 goto exit; 503 } 504 505 /* 506 * Check the permissions. Note the difference between 507 * VREAD/VWRITE and FREAD/FWRITE. 508 */ 509 acc_mode = 0; 510 if (fp->f_flag & FREAD) { 511 acc_mode |= VREAD; 512 } 513 if (fp->f_flag & FWRITE) { 514 acc_mode |= VWRITE; 515 } 516 if (vaccess(VNON, mq->mq_mode, mq->mq_euid, mq->mq_egid, 517 acc_mode, curproc->p_ucred)) { 518 519 error = EACCES; 520 goto exit; 521 } 522 } else { 523 /* Fail if mqueue neither exists, nor we create it */ 524 if ((oflag & O_CREAT) == 0) { 525 lockmgr(&mqlist_mtx, LK_RELEASE); 526 KKASSERT(mq_new == NULL); 527 fsetfd(curproc, NULL, mqd); 528 fp->f_ops = &badfileops; 529 fdrop(fp); 530 kfree(name, M_MQBUF); 531 return ENOENT; 532 } 533 534 /* Check the limit */ 535 if (p->p_mqueue_cnt == mq_open_max) { 536 error = EMFILE; 537 goto exit; 538 } 539 540 /* Insert the queue to the list */ 541 mq = mq_new; 542 lockmgr(&mq->mq_mtx, LK_EXCLUSIVE); 543 LIST_INSERT_HEAD(&mqueue_head, mq, mq_list); 544 mq_new = NULL; 545 getnanotime(&mq->mq_btime); 546 mq->mq_atime = mq->mq_mtime = mq->mq_btime; 547 } 548 549 /* Increase the counters, and make descriptor ready */ 550 p->p_mqueue_cnt++; 551 mq->mq_refcnt++; 552 fp->f_data = mq; 553 exit: 554 lockmgr(&mq->mq_mtx, LK_RELEASE); 555 lockmgr(&mqlist_mtx, LK_RELEASE); 556 557 if (mq_new) 558 mqueue_destroy(mq_new); 559 if (error) { 560 fsetfd(curproc, NULL, mqd); 561 fp->f_ops = &badfileops; 562 } else { 563 fsetfd(p, fp, mqd); 564 uap->sysmsg_result = mqd; 565 } 566 fdrop(fp); 567 kfree(name, M_MQBUF); 568 569 return error; 570 } 571 572 int 573 sys_mq_close(struct mq_close_args *uap) 574 { 575 return sys_close((void *)uap); 576 } 577 578 /* 579 * Primary mq_receive1() function. 580 */ 581 int 582 mq_receive1(struct lwp *l, mqd_t mqdes, void *msg_ptr, size_t msg_len, 583 unsigned *msg_prio, struct timespec *ts, ssize_t *mlen) 584 { 585 file_t *fp = NULL; 586 struct mqueue *mq; 587 struct mq_msg *msg = NULL; 588 struct mq_attr *mqattr; 589 u_int idx; 590 int error; 591 592 /* Get the message queue */ 593 error = mqueue_get(l, mqdes, &fp); 594 if (error) { 595 return error; 596 } 597 mq = fp->f_data; 598 if ((fp->f_flag & FREAD) == 0) { 599 error = EBADF; 600 goto error; 601 } 602 getnanotime(&mq->mq_atime); 603 mqattr = &mq->mq_attrib; 604 605 /* Check the message size limits */ 606 if (msg_len < mqattr->mq_msgsize) { 607 error = EMSGSIZE; 608 goto error; 609 } 610 611 /* Check if queue is empty */ 612 while (mqattr->mq_curmsgs == 0) { 613 int t; 614 615 if (mqattr->mq_flags & O_NONBLOCK) { 616 error = EAGAIN; 617 goto error; 618 } 619 error = abstimeout2timo(ts, &t); 620 if (error) { 621 goto error; 622 } 623 /* 624 * Block until someone sends the message. 625 * While doing this, notification should not be sent. 626 */ 627 mqattr->mq_flags |= MQ_RECEIVE; 628 error = tsleep(&mq->mq_send_cv, PCATCH, "mqsend", t); 629 mqattr->mq_flags &= ~MQ_RECEIVE; 630 if (error || (mqattr->mq_flags & MQ_UNLINK)) { 631 error = (error == EWOULDBLOCK) ? ETIMEDOUT : EINTR; 632 goto error; 633 } 634 } 635 636 637 /* 638 * Find the highest priority message, and remove it from the queue. 639 * At first, reserved queue is checked, bitmap is next. 640 */ 641 msg = TAILQ_FIRST(&mq->mq_head[MQ_PQRESQ]); 642 if (__predict_true(msg == NULL)) { 643 idx = ffs(mq->mq_bitmap); 644 msg = TAILQ_FIRST(&mq->mq_head[idx]); 645 KKASSERT(msg != NULL); 646 } else { 647 idx = MQ_PQRESQ; 648 } 649 TAILQ_REMOVE(&mq->mq_head[idx], msg, msg_queue); 650 651 /* Unmark the bit, if last message. */ 652 if (__predict_true(idx) && TAILQ_EMPTY(&mq->mq_head[idx])) { 653 KKASSERT((MQ_PQSIZE - idx) == msg->msg_prio); 654 mq->mq_bitmap &= ~(1 << --idx); 655 } 656 657 /* Decrement the counter and signal waiter, if any */ 658 mqattr->mq_curmsgs--; 659 wakeup_one(&mq->mq_recv_cv); 660 661 /* Ready for sending now */ 662 selwakeup(&mq->mq_wsel); 663 error: 664 lockmgr(&mq->mq_mtx, LK_RELEASE); 665 fdrop(fp); 666 if (error) 667 return error; 668 669 /* 670 * Copy the data to the user-space. 671 * Note: According to POSIX, no message should be removed from the 672 * queue in case of fail - this would be violated. 673 */ 674 *mlen = msg->msg_len; 675 error = copyout(msg->msg_ptr, msg_ptr, msg->msg_len); 676 if (error == 0 && msg_prio) 677 error = copyout(&msg->msg_prio, msg_prio, sizeof(unsigned)); 678 mqueue_freemsg(msg, sizeof(struct mq_msg) + msg->msg_len); 679 680 return error; 681 } 682 683 int 684 sys_mq_receive(struct mq_receive_args *uap) 685 { 686 /* { 687 syscallarg(mqd_t) mqdes; 688 syscallarg(char *) msg_ptr; 689 syscallarg(size_t) msg_len; 690 syscallarg(unsigned *) msg_prio; 691 } */ 692 ssize_t mlen; 693 int error; 694 695 error = mq_receive1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr), 696 SCARG(uap, msg_len), SCARG(uap, msg_prio), 0, &mlen); 697 if (error == 0) 698 uap->sysmsg_result = mlen; 699 700 return error; 701 } 702 703 int 704 sys_mq_timedreceive(struct mq_timedreceive_args *uap) 705 { 706 /* { 707 syscallarg(mqd_t) mqdes; 708 syscallarg(char *) msg_ptr; 709 syscallarg(size_t) msg_len; 710 syscallarg(unsigned *) msg_prio; 711 syscallarg(const struct timespec *) abs_timeout; 712 } */ 713 int error; 714 ssize_t mlen; 715 struct timespec ts, *tsp; 716 717 /* Get and convert time value */ 718 if (SCARG(uap, abs_timeout)) { 719 error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts)); 720 if (error) 721 return error; 722 tsp = &ts; 723 } else { 724 tsp = NULL; 725 } 726 727 error = mq_receive1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr), 728 SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp, &mlen); 729 if (error == 0) 730 uap->sysmsg_result = mlen; 731 732 return error; 733 } 734 735 /* 736 * Primary mq_send1() function. 737 */ 738 int 739 mq_send1(struct lwp *l, mqd_t mqdes, const char *msg_ptr, size_t msg_len, 740 unsigned msg_prio, struct timespec *ts) 741 { 742 file_t *fp = NULL; 743 struct mqueue *mq; 744 struct mq_msg *msg; 745 struct mq_attr *mqattr; 746 struct proc *notify = NULL; 747 /*ksiginfo_t ksi;*/ 748 size_t size; 749 int error; 750 751 /* Check the priority range */ 752 if (msg_prio >= mq_prio_max) 753 return EINVAL; 754 755 /* Allocate a new message */ 756 size = sizeof(struct mq_msg) + msg_len; 757 if (size > mq_max_msgsize) 758 return EMSGSIZE; 759 760 if (size > MQ_DEF_MSGSIZE) { 761 msg = kmalloc(size, M_MQBUF, M_WAITOK); 762 } else { 763 msg = objcache_get(mqmsg_cache, M_WAITOK); 764 } 765 766 /* Get the data from user-space */ 767 error = copyin(msg_ptr, msg->msg_ptr, msg_len); 768 if (error) { 769 mqueue_freemsg(msg, size); 770 return error; 771 } 772 msg->msg_len = msg_len; 773 msg->msg_prio = msg_prio; 774 775 /* Get the mqueue */ 776 error = mqueue_get(l, mqdes, &fp); 777 if (error) { 778 mqueue_freemsg(msg, size); 779 return error; 780 } 781 mq = fp->f_data; 782 if ((fp->f_flag & FWRITE) == 0) { 783 error = EBADF; 784 goto error; 785 } 786 getnanotime(&mq->mq_mtime); 787 mqattr = &mq->mq_attrib; 788 789 /* Check the message size limit */ 790 if (msg_len <= 0 || msg_len > mqattr->mq_msgsize) { 791 error = EMSGSIZE; 792 goto error; 793 } 794 795 /* Check if queue is full */ 796 while (mqattr->mq_curmsgs >= mqattr->mq_maxmsg) { 797 int t; 798 799 if (mqattr->mq_flags & O_NONBLOCK) { 800 error = EAGAIN; 801 goto error; 802 } 803 error = abstimeout2timo(ts, &t); 804 if (error) { 805 goto error; 806 } 807 /* Block until queue becomes available */ 808 error = tsleep(&mq->mq_recv_cv, PCATCH, "mqrecv", t); 809 if (error || (mqattr->mq_flags & MQ_UNLINK)) { 810 error = (error == EWOULDBLOCK) ? ETIMEDOUT : error; 811 goto error; 812 } 813 } 814 KKASSERT(mq->mq_attrib.mq_curmsgs < mq->mq_attrib.mq_maxmsg); 815 816 /* 817 * Insert message into the queue, according to the priority. 818 * Note the difference between index and priority. 819 */ 820 if (__predict_true(msg_prio < MQ_PQSIZE)) { 821 u_int idx = MQ_PQSIZE - msg_prio; 822 823 KKASSERT(idx != MQ_PQRESQ); 824 TAILQ_INSERT_TAIL(&mq->mq_head[idx], msg, msg_queue); 825 mq->mq_bitmap |= (1 << --idx); 826 } else { 827 mqueue_linear_insert(mq, msg); 828 } 829 830 /* Check for the notify */ 831 if (mqattr->mq_curmsgs == 0 && mq->mq_notify_proc && 832 (mqattr->mq_flags & MQ_RECEIVE) == 0) { 833 /* Initialize the signal */ 834 /*KSI_INIT(&ksi);*/ 835 /*ksi.ksi_signo = mq->mq_sig_notify.sigev_signo;*/ 836 /*ksi.ksi_code = SI_MESGQ;*/ 837 /*ksi.ksi_value = mq->mq_sig_notify.sigev_value;*/ 838 /* Unregister the process */ 839 notify = mq->mq_notify_proc; 840 mq->mq_notify_proc = NULL; 841 } 842 843 /* Increment the counter and signal waiter, if any */ 844 mqattr->mq_curmsgs++; 845 wakeup_one(&mq->mq_send_cv); 846 847 /* Ready for receiving now */ 848 selwakeup(&mq->mq_rsel); 849 error: 850 lockmgr(&mq->mq_mtx, LK_RELEASE); 851 fdrop(fp); 852 853 if (error) { 854 mqueue_freemsg(msg, size); 855 } else if (notify) { 856 /* Send the notify, if needed */ 857 spin_lock_wr(&allproc_spin); 858 /*kpsignal(notify, &ksi, NULL);*/ 859 ksignal(notify, mq->mq_sig_notify.sigev_signo); 860 spin_unlock_wr(&allproc_spin); 861 } 862 863 return error; 864 } 865 866 int 867 sys_mq_send(struct mq_send_args *uap) 868 { 869 /* { 870 syscallarg(mqd_t) mqdes; 871 syscallarg(const char *) msg_ptr; 872 syscallarg(size_t) msg_len; 873 syscallarg(unsigned) msg_prio; 874 } */ 875 876 return mq_send1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr), 877 SCARG(uap, msg_len), SCARG(uap, msg_prio), 0); 878 } 879 880 int 881 sys_mq_timedsend(struct mq_timedsend_args *uap) 882 { 883 /* { 884 syscallarg(mqd_t) mqdes; 885 syscallarg(const char *) msg_ptr; 886 syscallarg(size_t) msg_len; 887 syscallarg(unsigned) msg_prio; 888 syscallarg(const struct timespec *) abs_timeout; 889 } */ 890 struct timespec ts, *tsp; 891 int error; 892 893 /* Get and convert time value */ 894 if (SCARG(uap, abs_timeout)) { 895 error = copyin(SCARG(uap, abs_timeout), &ts, sizeof(ts)); 896 if (error) 897 return error; 898 tsp = &ts; 899 } else { 900 tsp = NULL; 901 } 902 903 return mq_send1(curthread->td_lwp, SCARG(uap, mqdes), SCARG(uap, msg_ptr), 904 SCARG(uap, msg_len), SCARG(uap, msg_prio), tsp); 905 } 906 907 int 908 sys_mq_notify(struct mq_notify_args *uap) 909 { 910 /* { 911 syscallarg(mqd_t) mqdes; 912 syscallarg(const struct sigevent *) notification; 913 } */ 914 file_t *fp = NULL; 915 struct mqueue *mq; 916 struct sigevent sig; 917 int error; 918 919 if (SCARG(uap, notification)) { 920 /* Get the signal from user-space */ 921 error = copyin(SCARG(uap, notification), &sig, 922 sizeof(struct sigevent)); 923 if (error) 924 return error; 925 } 926 927 error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp); 928 if (error) 929 return error; 930 mq = fp->f_data; 931 932 if (SCARG(uap, notification)) { 933 /* Register notification: set the signal and target process */ 934 if (mq->mq_notify_proc == NULL) { 935 memcpy(&mq->mq_sig_notify, &sig, 936 sizeof(struct sigevent)); 937 mq->mq_notify_proc = curproc; 938 } else { 939 /* Fail if someone else already registered */ 940 error = EBUSY; 941 } 942 } else { 943 /* Unregister the notification */ 944 mq->mq_notify_proc = NULL; 945 } 946 lockmgr(&mq->mq_mtx, LK_RELEASE); 947 fdrop(fp); 948 949 return error; 950 } 951 952 int 953 sys_mq_getattr(struct mq_getattr_args *uap) 954 { 955 /* { 956 syscallarg(mqd_t) mqdes; 957 syscallarg(struct mq_attr *) mqstat; 958 } */ 959 file_t *fp = NULL; 960 struct mqueue *mq; 961 struct mq_attr attr; 962 int error; 963 964 /* Get the message queue */ 965 error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp); 966 if (error) 967 return error; 968 mq = fp->f_data; 969 memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr)); 970 lockmgr(&mq->mq_mtx, LK_RELEASE); 971 fdrop(fp); 972 973 return copyout(&attr, SCARG(uap, mqstat), sizeof(struct mq_attr)); 974 } 975 976 int 977 sys_mq_setattr(struct mq_setattr_args *uap) 978 { 979 /* { 980 syscallarg(mqd_t) mqdes; 981 syscallarg(const struct mq_attr *) mqstat; 982 syscallarg(struct mq_attr *) omqstat; 983 } */ 984 file_t *fp = NULL; 985 struct mqueue *mq; 986 struct mq_attr attr; 987 int error, nonblock; 988 989 error = copyin(SCARG(uap, mqstat), &attr, sizeof(struct mq_attr)); 990 if (error) 991 return error; 992 nonblock = (attr.mq_flags & O_NONBLOCK); 993 994 /* Get the message queue */ 995 error = mqueue_get(curthread->td_lwp, SCARG(uap, mqdes), &fp); 996 if (error) 997 return error; 998 mq = fp->f_data; 999 1000 /* Copy the old attributes, if needed */ 1001 if (SCARG(uap, omqstat)) 1002 memcpy(&attr, &mq->mq_attrib, sizeof(struct mq_attr)); 1003 1004 /* Ignore everything, except O_NONBLOCK */ 1005 if (nonblock) 1006 mq->mq_attrib.mq_flags |= O_NONBLOCK; 1007 else 1008 mq->mq_attrib.mq_flags &= ~O_NONBLOCK; 1009 1010 lockmgr(&mq->mq_mtx, LK_RELEASE); 1011 fdrop(fp); 1012 1013 /* 1014 * Copy the data to the user-space. 1015 * Note: According to POSIX, the new attributes should not be set in 1016 * case of fail - this would be violated. 1017 */ 1018 if (SCARG(uap, omqstat)) 1019 error = copyout(&attr, SCARG(uap, omqstat), 1020 sizeof(struct mq_attr)); 1021 1022 return error; 1023 } 1024 1025 int 1026 sys_mq_unlink(struct mq_unlink_args *uap) 1027 { 1028 /* { 1029 syscallarg(const char *) name; 1030 } */ 1031 struct mqueue *mq; 1032 char *name; 1033 int error, refcnt = 0; 1034 1035 /* Get the name from the user-space */ 1036 name = kmalloc(MQ_NAMELEN, M_MQBUF, M_WAITOK | M_ZERO); 1037 error = copyinstr(SCARG(uap, name), name, MQ_NAMELEN - 1, NULL); 1038 if (error) { 1039 kfree(name, M_MQBUF); 1040 return error; 1041 } 1042 1043 /* Lookup for this file */ 1044 lockmgr(&mqlist_mtx, LK_EXCLUSIVE); 1045 mq = mqueue_lookup(name); 1046 if (mq == NULL) { 1047 error = ENOENT; 1048 goto error; 1049 } 1050 1051 /* Check the permissions */ 1052 if (curproc->p_ucred->cr_uid != mq->mq_euid && 1053 priv_check(curthread, PRIV_ROOT) != 0) { 1054 lockmgr(&mq->mq_mtx, LK_RELEASE); 1055 error = EACCES; 1056 goto error; 1057 } 1058 1059 /* Mark message queue as unlinking, before leaving the window */ 1060 mq->mq_attrib.mq_flags |= MQ_UNLINK; 1061 1062 /* Wake up all waiters, if there are such */ 1063 wakeup(&mq->mq_send_cv); 1064 wakeup(&mq->mq_recv_cv); 1065 1066 selwakeup(&mq->mq_rsel); 1067 selwakeup(&mq->mq_wsel); 1068 1069 refcnt = mq->mq_refcnt; 1070 if (refcnt == 0) 1071 LIST_REMOVE(mq, mq_list); 1072 1073 lockmgr(&mq->mq_mtx, LK_RELEASE); 1074 error: 1075 lockmgr(&mqlist_mtx, LK_RELEASE); 1076 1077 /* 1078 * If there are no references - destroy the message 1079 * queue, otherwise, the last mq_close() will do that. 1080 */ 1081 if (error == 0 && refcnt == 0) 1082 mqueue_destroy(mq); 1083 1084 kfree(name, M_MQBUF); 1085 return error; 1086 } 1087 1088 /* 1089 * SysCtl. 1090 */ 1091 SYSCTL_NODE(_kern, OID_AUTO, mqueue, 1092 CTLFLAG_RW, 0, "Message queue options"); 1093 1094 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_open_max, 1095 CTLFLAG_RW, &mq_open_max, 0, 1096 "Maximal number of message queue descriptors per process"); 1097 1098 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_prio_max, 1099 CTLFLAG_RW, &mq_prio_max, 0, 1100 "Maximal priority of the message"); 1101 1102 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_max_msgsize, 1103 CTLFLAG_RW, &mq_max_msgsize, 0, 1104 "Maximal allowed size of the message"); 1105 1106 SYSCTL_INT(_kern_mqueue, OID_AUTO, mq_def_maxmsg, 1107 CTLFLAG_RW, &mq_def_maxmsg, 0, 1108 "Default maximal message count"); 1109 1110 SYSINIT(sys_mqueue_init, SI_SUB_PRE_DRIVERS, SI_ORDER_ANY, mqueue_sysinit, NULL); 1111