1 /*- 2 * Copyright (c) 2005 David Xu <davidxu@freebsd.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 */ 27 28 /* 29 * POSIX message queue implementation. 30 * 31 * 1) A mqueue filesystem can be mounted, each message queue appears 32 * in mounted directory, user can change queue's permission and 33 * ownership, or remove a queue. Manually creating a file in the 34 * directory causes a message queue to be created in the kernel with 35 * default message queue attributes applied and same name used, this 36 * method is not advocated since mq_open syscall allows user to specify 37 * different attributes. Also the file system can be mounted multiple 38 * times at different mount points but shows same contents. 39 * 40 * 2) Standard POSIX message queue API. The syscalls do not use vfs layer, 41 * but directly operate on internal data structure, this allows user to 42 * use the IPC facility without having to mount mqueue file system. 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include <sys/param.h> 49 #include <sys/kernel.h> 50 #include <sys/systm.h> 51 #include <sys/limits.h> 52 #include <sys/buf.h> 53 #include <sys/dirent.h> 54 #include <sys/event.h> 55 #include <sys/eventhandler.h> 56 #include <sys/fcntl.h> 57 #include <sys/file.h> 58 #include <sys/filedesc.h> 59 #include <sys/lock.h> 60 #include <sys/malloc.h> 61 #include <sys/module.h> 62 #include <sys/mount.h> 63 #include <sys/mqueue.h> 64 #include <sys/mutex.h> 65 #include <sys/namei.h> 66 #include <sys/posix4.h> 67 #include <sys/poll.h> 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/queue.h> 71 #include <sys/sysproto.h> 72 #include <sys/stat.h> 73 #include <sys/syscall.h> 74 #include <sys/syscallsubr.h> 75 #include <sys/sysent.h> 76 #include <sys/sx.h> 77 #include <sys/sysctl.h> 78 #include <sys/taskqueue.h> 79 #include <sys/unistd.h> 80 #include <sys/vnode.h> 81 #include <machine/atomic.h> 82 83 /* 84 * Limits and constants 85 */ 86 #define MQFS_NAMELEN NAME_MAX 87 #define MQFS_DELEN (8 + MQFS_NAMELEN) 88 89 /* node types */ 90 typedef enum { 91 mqfstype_none = 0, 92 mqfstype_root, 93 mqfstype_dir, 94 mqfstype_this, 95 mqfstype_parent, 96 mqfstype_file, 97 mqfstype_symlink, 98 } mqfs_type_t; 99 100 struct mqfs_node; 101 102 /* 103 * mqfs_info: describes a mqfs instance 104 */ 105 struct mqfs_info { 106 struct sx mi_lock; 107 struct mqfs_node *mi_root; 108 struct unrhdr *mi_unrhdr; 109 }; 110 111 struct mqfs_vdata { 112 LIST_ENTRY(mqfs_vdata) mv_link; 113 struct mqfs_node *mv_node; 114 struct vnode *mv_vnode; 115 struct task mv_task; 116 }; 117 118 /* 119 * mqfs_node: describes a node (file or directory) within a mqfs 120 */ 121 struct mqfs_node { 122 char mn_name[MQFS_NAMELEN+1]; 123 struct mqfs_info *mn_info; 124 struct mqfs_node *mn_parent; 125 LIST_HEAD(,mqfs_node) mn_children; 126 LIST_ENTRY(mqfs_node) mn_sibling; 127 LIST_HEAD(,mqfs_vdata) mn_vnodes; 128 int mn_refcount; 129 mqfs_type_t mn_type; 130 int mn_deleted; 131 u_int32_t mn_fileno; 132 void *mn_data; 133 struct timespec mn_birth; 134 struct timespec mn_ctime; 135 struct timespec mn_atime; 136 struct timespec mn_mtime; 137 uid_t mn_uid; 138 gid_t mn_gid; 139 int mn_mode; 140 }; 141 142 #define VTON(vp) (((struct mqfs_vdata *)((vp)->v_data))->mv_node) 143 #define VTOMQ(vp) ((struct mqueue *)(VTON(vp)->mn_data)) 144 #define VFSTOMQFS(m) ((struct mqfs_info *)((m)->mnt_data)) 145 #define FPTOMQ(fp) ((struct mqueue *)(((struct mqfs_node *) \ 146 (fp)->f_data)->mn_data)) 147 148 TAILQ_HEAD(msgq, mqueue_msg); 149 150 struct mqueue; 151 152 struct mqueue_notifier { 153 LIST_ENTRY(mqueue_notifier) nt_link; 154 struct sigevent nt_sigev; 155 ksiginfo_t nt_ksi; 156 struct proc *nt_proc; 157 }; 158 159 struct mqueue { 160 struct mtx mq_mutex; 161 int mq_flags; 162 long mq_maxmsg; 163 long mq_msgsize; 164 long mq_curmsgs; 165 long mq_totalbytes; 166 struct msgq mq_msgq; 167 int mq_receivers; 168 int mq_senders; 169 struct selinfo mq_rsel; 170 struct selinfo mq_wsel; 171 struct mqueue_notifier *mq_notifier; 172 }; 173 174 #define MQ_RSEL 0x01 175 #define MQ_WSEL 0x02 176 177 struct mqueue_msg { 178 TAILQ_ENTRY(mqueue_msg) msg_link; 179 unsigned int msg_prio; 180 unsigned int msg_size; 181 /* following real data... */ 182 }; 183 184 SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0, 185 "POSIX real time message queue"); 186 187 static int default_maxmsg = 10; 188 static int default_msgsize = 1024; 189 190 static int maxmsg = 100; 191 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW, 192 &maxmsg, 0, "Default maximum messages in queue"); 193 static int maxmsgsize = 16384; 194 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW, 195 &maxmsgsize, 0, "Default maximum message size"); 196 static int maxmq = 100; 197 SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW, 198 &maxmq, 0, "maximum message queues"); 199 static int curmq = 0; 200 SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW, 201 &curmq, 0, "current message queue number"); 202 static int unloadable = 0; 203 static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data"); 204 205 static eventhandler_tag exit_tag; 206 207 /* Only one instance per-system */ 208 static struct mqfs_info mqfs_data; 209 static uma_zone_t mqnode_zone; 210 static uma_zone_t mqueue_zone; 211 static uma_zone_t mvdata_zone; 212 static uma_zone_t mqnoti_zone; 213 static struct vop_vector mqfs_vnodeops; 214 static struct fileops mqueueops; 215 216 /* 217 * Directory structure construction and manipulation 218 */ 219 #ifdef notyet 220 static struct mqfs_node *mqfs_create_dir(struct mqfs_node *parent, 221 const char *name, int namelen, struct ucred *cred, int mode); 222 static struct mqfs_node *mqfs_create_link(struct mqfs_node *parent, 223 const char *name, int namelen, struct ucred *cred, int mode); 224 #endif 225 226 static struct mqfs_node *mqfs_create_file(struct mqfs_node *parent, 227 const char *name, int namelen, struct ucred *cred, int mode); 228 static int mqfs_destroy(struct mqfs_node *mn); 229 static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn); 230 static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn); 231 static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn); 232 233 /* 234 * Message queue construction and maniplation 235 */ 236 static struct mqueue *mqueue_alloc(const struct mq_attr *attr); 237 static void mqueue_free(struct mqueue *mq); 238 static int mqueue_send(struct mqueue *mq, const char *msg_ptr, 239 size_t msg_len, unsigned msg_prio, int waitok, 240 const struct timespec *abs_timeout); 241 static int mqueue_receive(struct mqueue *mq, char *msg_ptr, 242 size_t msg_len, unsigned *msg_prio, int waitok, 243 const struct timespec *abs_timeout); 244 static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, 245 int timo); 246 static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, 247 int timo); 248 static void mqueue_send_notification(struct mqueue *mq); 249 static void mqueue_fdclose(struct thread *td, int fd, struct file *fp); 250 static void mq_proc_exit(void *arg, struct proc *p); 251 252 /* 253 * kqueue filters 254 */ 255 static void filt_mqdetach(struct knote *kn); 256 static int filt_mqread(struct knote *kn, long hint); 257 static int filt_mqwrite(struct knote *kn, long hint); 258 259 struct filterops mq_rfiltops = 260 { 1, NULL, filt_mqdetach, filt_mqread }; 261 struct filterops mq_wfiltops = 262 { 1, NULL, filt_mqdetach, filt_mqwrite }; 263 264 /* 265 * Initialize fileno bitmap 266 */ 267 static void 268 mqfs_fileno_init(struct mqfs_info *mi) 269 { 270 struct unrhdr *up; 271 272 up = new_unrhdr(1, INT_MAX, NULL); 273 mi->mi_unrhdr = up; 274 } 275 276 /* 277 * Tear down fileno bitmap 278 */ 279 static void 280 mqfs_fileno_uninit(struct mqfs_info *mi) 281 { 282 struct unrhdr *up; 283 284 up = mi->mi_unrhdr; 285 mi->mi_unrhdr = NULL; 286 delete_unrhdr(up); 287 } 288 289 /* 290 * Allocate a file number 291 */ 292 static void 293 mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn) 294 { 295 /* make sure our parent has a file number */ 296 if (mn->mn_parent && !mn->mn_parent->mn_fileno) 297 mqfs_fileno_alloc(mi, mn->mn_parent); 298 299 switch (mn->mn_type) { 300 case mqfstype_root: 301 case mqfstype_dir: 302 case mqfstype_file: 303 case mqfstype_symlink: 304 mn->mn_fileno = alloc_unr(mi->mi_unrhdr); 305 break; 306 case mqfstype_this: 307 KASSERT(mn->mn_parent != NULL, 308 ("mqfstype_this node has no parent")); 309 mn->mn_fileno = mn->mn_parent->mn_fileno; 310 break; 311 case mqfstype_parent: 312 KASSERT(mn->mn_parent != NULL, 313 ("mqfstype_parent node has no parent")); 314 if (mn->mn_parent == mi->mi_root) { 315 mn->mn_fileno = mn->mn_parent->mn_fileno; 316 break; 317 } 318 KASSERT(mn->mn_parent->mn_parent != NULL, 319 ("mqfstype_parent node has no grandparent")); 320 mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno; 321 break; 322 default: 323 KASSERT(0, 324 ("mqfs_fileno_alloc() called for unknown type node: %d", 325 mn->mn_type)); 326 break; 327 } 328 } 329 330 /* 331 * Release a file number 332 */ 333 static void 334 mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn) 335 { 336 switch (mn->mn_type) { 337 case mqfstype_root: 338 case mqfstype_dir: 339 case mqfstype_file: 340 case mqfstype_symlink: 341 free_unr(mi->mi_unrhdr, mn->mn_fileno); 342 break; 343 case mqfstype_this: 344 case mqfstype_parent: 345 /* ignore these, as they don't "own" their file number */ 346 break; 347 default: 348 KASSERT(0, 349 ("mqfs_fileno_free() called for unknown type node: %d", 350 mn->mn_type)); 351 break; 352 } 353 } 354 355 static __inline struct mqfs_node * 356 mqnode_alloc(void) 357 { 358 return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO); 359 } 360 361 static __inline void 362 mqnode_free(struct mqfs_node *node) 363 { 364 uma_zfree(mqnode_zone, node); 365 } 366 367 static __inline void 368 mqnode_addref(struct mqfs_node *node) 369 { 370 atomic_fetchadd_int(&node->mn_refcount, 1); 371 } 372 373 static __inline void 374 mqnode_release(struct mqfs_node *node) 375 { 376 struct mqfs_info *mqfs; 377 int old, exp; 378 379 mqfs = node->mn_info; 380 old = atomic_fetchadd_int(&node->mn_refcount, -1); 381 if (node->mn_type == mqfstype_dir || 382 node->mn_type == mqfstype_root) 383 exp = 3; /* include . and .. */ 384 else 385 exp = 1; 386 if (old == exp) { 387 int locked = sx_xlocked(&mqfs->mi_lock); 388 if (!locked) 389 sx_xlock(&mqfs->mi_lock); 390 mqfs_destroy(node); 391 if (!locked) 392 sx_xunlock(&mqfs->mi_lock); 393 } 394 } 395 396 /* 397 * Add a node to a directory 398 */ 399 static int 400 mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node) 401 { 402 KASSERT(parent != NULL, ("%s(): parent is NULL", __func__)); 403 KASSERT(parent->mn_info != NULL, 404 ("%s(): parent has no mn_info", __func__)); 405 KASSERT(parent->mn_type == mqfstype_dir || 406 parent->mn_type == mqfstype_root, 407 ("%s(): parent is not a directory", __func__)); 408 409 node->mn_info = parent->mn_info; 410 node->mn_parent = parent; 411 LIST_INIT(&node->mn_children); 412 LIST_INIT(&node->mn_vnodes); 413 LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling); 414 mqnode_addref(parent); 415 return (0); 416 } 417 418 static struct mqfs_node * 419 mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode, 420 int nodetype) 421 { 422 struct mqfs_node *node; 423 424 node = mqnode_alloc(); 425 strncpy(node->mn_name, name, namelen); 426 node->mn_type = nodetype; 427 node->mn_refcount = 1; 428 vfs_timestamp(&node->mn_birth); 429 node->mn_ctime = node->mn_atime = node->mn_mtime 430 = node->mn_birth; 431 node->mn_uid = cred->cr_uid; 432 node->mn_gid = cred->cr_gid; 433 node->mn_mode = mode; 434 return (node); 435 } 436 437 /* 438 * Create a file 439 */ 440 static struct mqfs_node * 441 mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen, 442 struct ucred *cred, int mode) 443 { 444 struct mqfs_node *node; 445 446 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file); 447 if (mqfs_add_node(parent, node) != 0) { 448 mqnode_free(node); 449 return (NULL); 450 } 451 return (node); 452 } 453 454 /* 455 * Add . and .. to a directory 456 */ 457 static int 458 mqfs_fixup_dir(struct mqfs_node *parent) 459 { 460 struct mqfs_node *dir; 461 462 dir = mqnode_alloc(); 463 dir->mn_name[0] = '.'; 464 dir->mn_type = mqfstype_this; 465 dir->mn_refcount = 1; 466 if (mqfs_add_node(parent, dir) != 0) { 467 mqnode_free(dir); 468 return (-1); 469 } 470 471 dir = mqnode_alloc(); 472 dir->mn_name[0] = dir->mn_name[1] = '.'; 473 dir->mn_type = mqfstype_parent; 474 dir->mn_refcount = 1; 475 476 if (mqfs_add_node(parent, dir) != 0) { 477 mqnode_free(dir); 478 return (-1); 479 } 480 481 return (0); 482 } 483 484 #ifdef notyet 485 486 /* 487 * Create a directory 488 */ 489 static struct mqfs_node * 490 mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen, 491 struct ucred *cred, int mode) 492 { 493 struct mqfs_node *node; 494 495 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir); 496 if (mqfs_add_node(parent, node) != 0) { 497 mqnode_free(node); 498 return (NULL); 499 } 500 501 if (mqfs_fixup_dir(node) != 0) { 502 mqfs_destroy(node); 503 return (NULL); 504 } 505 return (node); 506 } 507 508 /* 509 * Create a symlink 510 */ 511 static struct mqfs_node * 512 mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen, 513 struct ucred *cred, int mode) 514 { 515 struct mqfs_node *node; 516 517 node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink); 518 if (mqfs_add_node(parent, node) != 0) { 519 mqnode_free(node); 520 return (NULL); 521 } 522 return (node); 523 } 524 525 #endif 526 527 /* 528 * Destroy a node or a tree of nodes 529 */ 530 static int 531 mqfs_destroy(struct mqfs_node *node) 532 { 533 struct mqfs_node *parent; 534 535 KASSERT(node != NULL, 536 ("%s(): node is NULL", __func__)); 537 KASSERT(node->mn_info != NULL, 538 ("%s(): node has no mn_info", __func__)); 539 540 /* destroy children */ 541 if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root) 542 while (! LIST_EMPTY(&node->mn_children)) 543 mqfs_destroy(LIST_FIRST(&node->mn_children)); 544 545 /* unlink from parent */ 546 if ((parent = node->mn_parent) != NULL) { 547 KASSERT(parent->mn_info == node->mn_info, 548 ("%s(): parent has different mn_info", __func__)); 549 LIST_REMOVE(node, mn_sibling); 550 } 551 552 if (node->mn_fileno != 0) 553 mqfs_fileno_free(node->mn_info, node); 554 if (node->mn_data != NULL) 555 mqueue_free(node->mn_data); 556 mqnode_free(node); 557 return (0); 558 } 559 560 /* 561 * Mount a mqfs instance 562 */ 563 static int 564 mqfs_mount(struct mount *mp, struct thread *td) 565 { 566 struct statfs *sbp; 567 568 if (mp->mnt_flag & MNT_UPDATE) 569 return (EOPNOTSUPP); 570 571 mp->mnt_data = &mqfs_data; 572 MNT_ILOCK(mp); 573 mp->mnt_flag |= MNT_LOCAL; 574 mp->mnt_kern_flag |= MNTK_MPSAFE; 575 MNT_IUNLOCK(mp); 576 vfs_getnewfsid(mp); 577 578 sbp = &mp->mnt_stat; 579 vfs_mountedfrom(mp, "mqueue"); 580 sbp->f_bsize = PAGE_SIZE; 581 sbp->f_iosize = PAGE_SIZE; 582 sbp->f_blocks = 1; 583 sbp->f_bfree = 0; 584 sbp->f_bavail = 0; 585 sbp->f_files = 1; 586 sbp->f_ffree = 0; 587 return (0); 588 } 589 590 /* 591 * Unmount a mqfs instance 592 */ 593 static int 594 mqfs_unmount(struct mount *mp, int mntflags, struct thread *td) 595 { 596 int error; 597 598 error = vflush(mp, 0, (mntflags & MNT_FORCE) ? FORCECLOSE : 0, td); 599 return (error); 600 } 601 602 /* 603 * Return a root vnode 604 */ 605 static int 606 mqfs_root(struct mount *mp, int flags, struct vnode **vpp, struct thread *td) 607 { 608 struct mqfs_info *mqfs; 609 int ret; 610 611 mqfs = VFSTOMQFS(mp); 612 ret = mqfs_allocv(mp, vpp, mqfs->mi_root); 613 return (ret); 614 } 615 616 /* 617 * Return filesystem stats 618 */ 619 static int 620 mqfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 621 { 622 /* XXX update statistics */ 623 return (0); 624 } 625 626 /* 627 * Initialize a mqfs instance 628 */ 629 static int 630 mqfs_init(struct vfsconf *vfc) 631 { 632 struct mqfs_node *root; 633 struct mqfs_info *mi; 634 635 mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node), 636 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 637 mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue), 638 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 639 mvdata_zone = uma_zcreate("mvdata", 640 sizeof(struct mqfs_vdata), NULL, NULL, NULL, 641 NULL, UMA_ALIGN_PTR, 0); 642 mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier), 643 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 644 mi = &mqfs_data; 645 sx_init(&mi->mi_lock, "mqfs lock"); 646 /* set up the root diretory */ 647 root = mqfs_create_node("/", 1, curthread->td_ucred, 01777, 648 mqfstype_root); 649 root->mn_info = mi; 650 LIST_INIT(&root->mn_children); 651 LIST_INIT(&root->mn_vnodes); 652 mi->mi_root = root; 653 mqfs_fileno_init(mi); 654 mqfs_fileno_alloc(mi, root); 655 mqfs_fixup_dir(root); 656 exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL, 657 EVENTHANDLER_PRI_ANY); 658 mq_fdclose = mqueue_fdclose; 659 p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING); 660 return (0); 661 } 662 663 /* 664 * Destroy a mqfs instance 665 */ 666 static int 667 mqfs_uninit(struct vfsconf *vfc) 668 { 669 struct mqfs_info *mi; 670 671 if (!unloadable) 672 return (EOPNOTSUPP); 673 EVENTHANDLER_DEREGISTER(process_exit, exit_tag); 674 mi = &mqfs_data; 675 mqfs_destroy(mi->mi_root); 676 mi->mi_root = NULL; 677 mqfs_fileno_uninit(mi); 678 sx_destroy(&mi->mi_lock); 679 uma_zdestroy(mqnode_zone); 680 uma_zdestroy(mqueue_zone); 681 uma_zdestroy(mvdata_zone); 682 uma_zdestroy(mqnoti_zone); 683 return (0); 684 } 685 686 /* 687 * task routine 688 */ 689 static void 690 do_recycle(void *context, int pending __unused) 691 { 692 struct vnode *vp = (struct vnode *)context; 693 694 vrecycle(vp, curthread); 695 vdrop(vp); 696 } 697 698 /* 699 * Allocate a vnode 700 */ 701 static int 702 mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn) 703 { 704 struct mqfs_vdata *vd; 705 struct mqfs_info *mqfs; 706 struct vnode *newvpp; 707 int error; 708 709 mqfs = pn->mn_info; 710 *vpp = NULL; 711 sx_xlock(&mqfs->mi_lock); 712 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { 713 if (vd->mv_vnode->v_mount == mp) { 714 vhold(vd->mv_vnode); 715 break; 716 } 717 } 718 719 if (vd != NULL) { 720 found: 721 *vpp = vd->mv_vnode; 722 sx_xunlock(&mqfs->mi_lock); 723 error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread); 724 vdrop(*vpp); 725 return (error); 726 } 727 sx_xunlock(&mqfs->mi_lock); 728 729 error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp); 730 if (error) 731 return (error); 732 vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY); 733 error = insmntque(newvpp, mp); 734 if (error != 0) 735 return (error); 736 737 sx_xlock(&mqfs->mi_lock); 738 /* 739 * Check if it has already been allocated 740 * while we were blocked. 741 */ 742 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { 743 if (vd->mv_vnode->v_mount == mp) { 744 vhold(vd->mv_vnode); 745 sx_xunlock(&mqfs->mi_lock); 746 747 vgone(newvpp); 748 vput(newvpp); 749 goto found; 750 } 751 } 752 753 *vpp = newvpp; 754 755 vd = uma_zalloc(mvdata_zone, M_WAITOK); 756 (*vpp)->v_data = vd; 757 vd->mv_vnode = *vpp; 758 vd->mv_node = pn; 759 TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp); 760 LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link); 761 mqnode_addref(pn); 762 switch (pn->mn_type) { 763 case mqfstype_root: 764 (*vpp)->v_vflag = VV_ROOT; 765 /* fall through */ 766 case mqfstype_dir: 767 case mqfstype_this: 768 case mqfstype_parent: 769 (*vpp)->v_type = VDIR; 770 break; 771 case mqfstype_file: 772 (*vpp)->v_type = VREG; 773 break; 774 case mqfstype_symlink: 775 (*vpp)->v_type = VLNK; 776 break; 777 case mqfstype_none: 778 KASSERT(0, ("mqfs_allocf called for null node\n")); 779 default: 780 panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type); 781 } 782 sx_xunlock(&mqfs->mi_lock); 783 return (0); 784 } 785 786 /* 787 * Search a directory entry 788 */ 789 static struct mqfs_node * 790 mqfs_search(struct mqfs_node *pd, const char *name, int len) 791 { 792 struct mqfs_node *pn; 793 794 sx_assert(&pd->mn_info->mi_lock, SX_LOCKED); 795 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) { 796 if (strncmp(pn->mn_name, name, len) == 0) 797 return (pn); 798 } 799 return (NULL); 800 } 801 802 /* 803 * Look up a file or directory. 804 */ 805 static int 806 mqfs_lookupx(struct vop_cachedlookup_args *ap) 807 { 808 struct componentname *cnp; 809 struct vnode *dvp, **vpp; 810 struct mqfs_node *pd; 811 struct mqfs_node *pn; 812 struct mqfs_info *mqfs; 813 int nameiop, flags, error, namelen; 814 char *pname; 815 struct thread *td; 816 817 cnp = ap->a_cnp; 818 vpp = ap->a_vpp; 819 dvp = ap->a_dvp; 820 pname = cnp->cn_nameptr; 821 namelen = cnp->cn_namelen; 822 td = cnp->cn_thread; 823 flags = cnp->cn_flags; 824 nameiop = cnp->cn_nameiop; 825 pd = VTON(dvp); 826 pn = NULL; 827 mqfs = pd->mn_info; 828 *vpp = NULLVP; 829 830 if (dvp->v_type != VDIR) 831 return (ENOTDIR); 832 833 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread); 834 if (error) 835 return (error); 836 837 /* shortcut: check if the name is too long */ 838 if (cnp->cn_namelen >= MQFS_NAMELEN) 839 return (ENOENT); 840 841 /* self */ 842 if (namelen == 1 && pname[0] == '.') { 843 if ((flags & ISLASTCN) && nameiop != LOOKUP) 844 return (EINVAL); 845 pn = pd; 846 *vpp = dvp; 847 VREF(dvp); 848 return (0); 849 } 850 851 /* parent */ 852 if (cnp->cn_flags & ISDOTDOT) { 853 if (dvp->v_vflag & VV_ROOT) 854 return (EIO); 855 if ((flags & ISLASTCN) && nameiop != LOOKUP) 856 return (EINVAL); 857 VOP_UNLOCK(dvp, 0); 858 KASSERT(pd->mn_parent, ("non-root directory has no parent")); 859 pn = pd->mn_parent; 860 error = mqfs_allocv(dvp->v_mount, vpp, pn); 861 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 862 return (error); 863 } 864 865 /* named node */ 866 sx_xlock(&mqfs->mi_lock); 867 pn = mqfs_search(pd, pname, namelen); 868 if (pn != NULL) 869 mqnode_addref(pn); 870 sx_xunlock(&mqfs->mi_lock); 871 872 /* found */ 873 if (pn != NULL) { 874 /* DELETE */ 875 if (nameiop == DELETE && (flags & ISLASTCN)) { 876 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 877 if (error) { 878 mqnode_release(pn); 879 return (error); 880 } 881 if (*vpp == dvp) { 882 VREF(dvp); 883 *vpp = dvp; 884 mqnode_release(pn); 885 return (0); 886 } 887 } 888 889 /* allocate vnode */ 890 error = mqfs_allocv(dvp->v_mount, vpp, pn); 891 mqnode_release(pn); 892 if (error == 0 && cnp->cn_flags & MAKEENTRY) 893 cache_enter(dvp, *vpp, cnp); 894 return (error); 895 } 896 897 /* not found */ 898 899 /* will create a new entry in the directory ? */ 900 if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT) 901 && (flags & ISLASTCN)) { 902 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 903 if (error) 904 return (error); 905 cnp->cn_flags |= SAVENAME; 906 return (EJUSTRETURN); 907 } 908 return (ENOENT); 909 } 910 911 #if 0 912 struct vop_lookup_args { 913 struct vop_generic_args a_gen; 914 struct vnode *a_dvp; 915 struct vnode **a_vpp; 916 struct componentname *a_cnp; 917 }; 918 #endif 919 920 /* 921 * vnode lookup operation 922 */ 923 static int 924 mqfs_lookup(struct vop_cachedlookup_args *ap) 925 { 926 int rc; 927 928 rc = mqfs_lookupx(ap); 929 return (rc); 930 } 931 932 #if 0 933 struct vop_create_args { 934 struct vnode *a_dvp; 935 struct vnode **a_vpp; 936 struct componentname *a_cnp; 937 struct vattr *a_vap; 938 }; 939 #endif 940 941 /* 942 * vnode creation operation 943 */ 944 static int 945 mqfs_create(struct vop_create_args *ap) 946 { 947 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 948 struct componentname *cnp = ap->a_cnp; 949 struct mqfs_node *pd; 950 struct mqfs_node *pn; 951 struct mqueue *mq; 952 int error; 953 954 pd = VTON(ap->a_dvp); 955 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) 956 return (ENOTDIR); 957 mq = mqueue_alloc(NULL); 958 if (mq == NULL) 959 return (EAGAIN); 960 sx_xlock(&mqfs->mi_lock); 961 if ((cnp->cn_flags & HASBUF) == 0) 962 panic("%s: no name", __func__); 963 pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen, 964 cnp->cn_cred, ap->a_vap->va_mode); 965 if (pn == NULL) { 966 sx_xunlock(&mqfs->mi_lock); 967 error = ENOSPC; 968 } else { 969 mqnode_addref(pn); 970 sx_xunlock(&mqfs->mi_lock); 971 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); 972 mqnode_release(pn); 973 if (error) 974 mqfs_destroy(pn); 975 else 976 pn->mn_data = mq; 977 } 978 if (error) 979 mqueue_free(mq); 980 return (error); 981 } 982 983 /* 984 * Remove an entry 985 */ 986 static 987 int do_unlink(struct mqfs_node *pn, struct ucred *ucred) 988 { 989 struct mqfs_node *parent; 990 struct mqfs_vdata *vd; 991 int error = 0; 992 993 sx_assert(&pn->mn_info->mi_lock, SX_LOCKED); 994 995 if (ucred->cr_uid != pn->mn_uid && 996 (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0) 997 error = EACCES; 998 else if (!pn->mn_deleted) { 999 parent = pn->mn_parent; 1000 pn->mn_parent = NULL; 1001 pn->mn_deleted = 1; 1002 LIST_REMOVE(pn, mn_sibling); 1003 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { 1004 cache_purge(vd->mv_vnode); 1005 vhold(vd->mv_vnode); 1006 taskqueue_enqueue(taskqueue_thread, &vd->mv_task); 1007 } 1008 mqnode_release(pn); 1009 mqnode_release(parent); 1010 } else 1011 error = ENOENT; 1012 return (error); 1013 } 1014 1015 #if 0 1016 struct vop_remove_args { 1017 struct vnode *a_dvp; 1018 struct vnode *a_vp; 1019 struct componentname *a_cnp; 1020 }; 1021 #endif 1022 1023 /* 1024 * vnode removal operation 1025 */ 1026 static int 1027 mqfs_remove(struct vop_remove_args *ap) 1028 { 1029 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1030 struct mqfs_node *pn; 1031 int error; 1032 1033 if (ap->a_vp->v_type == VDIR) 1034 return (EPERM); 1035 pn = VTON(ap->a_vp); 1036 sx_xlock(&mqfs->mi_lock); 1037 error = do_unlink(pn, ap->a_cnp->cn_cred); 1038 sx_xunlock(&mqfs->mi_lock); 1039 return (error); 1040 } 1041 1042 #if 0 1043 struct vop_inactive_args { 1044 struct vnode *a_vp; 1045 struct thread *a_td; 1046 }; 1047 #endif 1048 1049 static int 1050 mqfs_inactive(struct vop_inactive_args *ap) 1051 { 1052 struct mqfs_node *pn = VTON(ap->a_vp); 1053 1054 if (pn->mn_deleted) 1055 vrecycle(ap->a_vp, ap->a_td); 1056 return (0); 1057 } 1058 1059 #if 0 1060 struct vop_reclaim_args { 1061 struct vop_generic_args a_gen; 1062 struct vnode *a_vp; 1063 struct thread *a_td; 1064 }; 1065 #endif 1066 1067 static int 1068 mqfs_reclaim(struct vop_reclaim_args *ap) 1069 { 1070 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount); 1071 struct vnode *vp = ap->a_vp; 1072 struct mqfs_node *pn; 1073 struct mqfs_vdata *vd; 1074 1075 vd = vp->v_data; 1076 pn = vd->mv_node; 1077 sx_xlock(&mqfs->mi_lock); 1078 vp->v_data = NULL; 1079 LIST_REMOVE(vd, mv_link); 1080 uma_zfree(mvdata_zone, vd); 1081 mqnode_release(pn); 1082 sx_xunlock(&mqfs->mi_lock); 1083 return (0); 1084 } 1085 1086 #if 0 1087 struct vop_open_args { 1088 struct vop_generic_args a_gen; 1089 struct vnode *a_vp; 1090 int a_mode; 1091 struct ucred *a_cred; 1092 struct thread *a_td; 1093 int a_fdidx; 1094 }; 1095 #endif 1096 1097 static int 1098 mqfs_open(struct vop_open_args *ap) 1099 { 1100 return (0); 1101 } 1102 1103 #if 0 1104 struct vop_close_args { 1105 struct vop_generic_args a_gen; 1106 struct vnode *a_vp; 1107 int a_fflag; 1108 struct ucred *a_cred; 1109 struct thread *a_td; 1110 }; 1111 #endif 1112 1113 static int 1114 mqfs_close(struct vop_close_args *ap) 1115 { 1116 return (0); 1117 } 1118 1119 #if 0 1120 struct vop_access_args { 1121 struct vop_generic_args a_gen; 1122 struct vnode *a_vp; 1123 int a_mode; 1124 struct ucred *a_cred; 1125 struct thread *a_td; 1126 }; 1127 #endif 1128 1129 /* 1130 * Verify permissions 1131 */ 1132 static int 1133 mqfs_access(struct vop_access_args *ap) 1134 { 1135 struct vnode *vp = ap->a_vp; 1136 struct vattr vattr; 1137 int error; 1138 1139 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 1140 if (error) 1141 return (error); 1142 error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, 1143 vattr.va_gid, ap->a_mode, ap->a_cred, NULL); 1144 return (error); 1145 } 1146 1147 #if 0 1148 struct vop_getattr_args { 1149 struct vop_generic_args a_gen; 1150 struct vnode *a_vp; 1151 struct vattr *a_vap; 1152 struct ucred *a_cred; 1153 }; 1154 #endif 1155 1156 /* 1157 * Get file attributes 1158 */ 1159 static int 1160 mqfs_getattr(struct vop_getattr_args *ap) 1161 { 1162 struct vnode *vp = ap->a_vp; 1163 struct mqfs_node *pn = VTON(vp); 1164 struct vattr *vap = ap->a_vap; 1165 int error = 0; 1166 1167 vap->va_type = vp->v_type; 1168 vap->va_mode = pn->mn_mode; 1169 vap->va_nlink = 1; 1170 vap->va_uid = pn->mn_uid; 1171 vap->va_gid = pn->mn_gid; 1172 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 1173 vap->va_fileid = pn->mn_fileno; 1174 vap->va_size = 0; 1175 vap->va_blocksize = PAGE_SIZE; 1176 vap->va_bytes = vap->va_size = 0; 1177 vap->va_atime = pn->mn_atime; 1178 vap->va_mtime = pn->mn_mtime; 1179 vap->va_ctime = pn->mn_ctime; 1180 vap->va_birthtime = pn->mn_birth; 1181 vap->va_gen = 0; 1182 vap->va_flags = 0; 1183 vap->va_rdev = NODEV; 1184 vap->va_bytes = 0; 1185 vap->va_filerev = 0; 1186 return (error); 1187 } 1188 1189 #if 0 1190 struct vop_setattr_args { 1191 struct vop_generic_args a_gen; 1192 struct vnode *a_vp; 1193 struct vattr *a_vap; 1194 struct ucred *a_cred; 1195 }; 1196 #endif 1197 /* 1198 * Set attributes 1199 */ 1200 static int 1201 mqfs_setattr(struct vop_setattr_args *ap) 1202 { 1203 struct mqfs_node *pn; 1204 struct vattr *vap; 1205 struct vnode *vp; 1206 struct thread *td; 1207 int c, error; 1208 uid_t uid; 1209 gid_t gid; 1210 1211 td = curthread; 1212 vap = ap->a_vap; 1213 vp = ap->a_vp; 1214 if ((vap->va_type != VNON) || 1215 (vap->va_nlink != VNOVAL) || 1216 (vap->va_fsid != VNOVAL) || 1217 (vap->va_fileid != VNOVAL) || 1218 (vap->va_blocksize != VNOVAL) || 1219 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1220 (vap->va_rdev != VNOVAL) || 1221 ((int)vap->va_bytes != VNOVAL) || 1222 (vap->va_gen != VNOVAL)) { 1223 return (EINVAL); 1224 } 1225 1226 pn = VTON(vp); 1227 1228 error = c = 0; 1229 if (vap->va_uid == (uid_t)VNOVAL) 1230 uid = pn->mn_uid; 1231 else 1232 uid = vap->va_uid; 1233 if (vap->va_gid == (gid_t)VNOVAL) 1234 gid = pn->mn_gid; 1235 else 1236 gid = vap->va_gid; 1237 1238 if (uid != pn->mn_uid || gid != pn->mn_gid) { 1239 /* 1240 * To modify the ownership of a file, must possess VADMIN 1241 * for that file. 1242 */ 1243 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td))) 1244 return (error); 1245 1246 /* 1247 * XXXRW: Why is there a privilege check here: shouldn't the 1248 * check in VOP_ACCESS() be enough? Also, are the group bits 1249 * below definitely right? 1250 */ 1251 if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid || 1252 (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) && 1253 (error = priv_check(td, PRIV_MQ_ADMIN)) != 0) 1254 return (error); 1255 pn->mn_uid = uid; 1256 pn->mn_gid = gid; 1257 c = 1; 1258 } 1259 1260 if (vap->va_mode != (mode_t)VNOVAL) { 1261 if ((ap->a_cred->cr_uid != pn->mn_uid) && 1262 (error = priv_check(td, PRIV_MQ_ADMIN))) 1263 return (error); 1264 pn->mn_mode = vap->va_mode; 1265 c = 1; 1266 } 1267 1268 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1269 /* See the comment in ufs_vnops::ufs_setattr(). */ 1270 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) && 1271 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 1272 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td)))) 1273 return (error); 1274 if (vap->va_atime.tv_sec != VNOVAL) { 1275 pn->mn_atime = vap->va_atime; 1276 } 1277 if (vap->va_mtime.tv_sec != VNOVAL) { 1278 pn->mn_mtime = vap->va_mtime; 1279 } 1280 c = 1; 1281 } 1282 if (c) { 1283 vfs_timestamp(&pn->mn_ctime); 1284 } 1285 return (0); 1286 } 1287 1288 #if 0 1289 struct vop_read_args { 1290 struct vop_generic_args a_gen; 1291 struct vnode *a_vp; 1292 struct uio *a_uio; 1293 int a_ioflag; 1294 struct ucred *a_cred; 1295 }; 1296 #endif 1297 1298 /* 1299 * Read from a file 1300 */ 1301 static int 1302 mqfs_read(struct vop_read_args *ap) 1303 { 1304 char buf[80]; 1305 struct vnode *vp = ap->a_vp; 1306 struct uio *uio = ap->a_uio; 1307 struct mqfs_node *pn; 1308 struct mqueue *mq; 1309 int len, error; 1310 1311 if (vp->v_type != VREG) 1312 return (EINVAL); 1313 1314 pn = VTON(vp); 1315 mq = VTOMQ(vp); 1316 snprintf(buf, sizeof(buf), 1317 "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n", 1318 mq->mq_totalbytes, 1319 mq->mq_maxmsg, 1320 mq->mq_curmsgs, 1321 mq->mq_msgsize); 1322 buf[sizeof(buf)-1] = '\0'; 1323 len = strlen(buf); 1324 error = uiomove_frombuf(buf, len, uio); 1325 return (error); 1326 } 1327 1328 #if 0 1329 struct vop_readdir_args { 1330 struct vop_generic_args a_gen; 1331 struct vnode *a_vp; 1332 struct uio *a_uio; 1333 struct ucred *a_cred; 1334 int *a_eofflag; 1335 int *a_ncookies; 1336 u_long **a_cookies; 1337 }; 1338 #endif 1339 1340 /* 1341 * Return directory entries. 1342 */ 1343 static int 1344 mqfs_readdir(struct vop_readdir_args *ap) 1345 { 1346 struct vnode *vp; 1347 struct mqfs_info *mi; 1348 struct mqfs_node *pd; 1349 struct mqfs_node *pn; 1350 struct dirent entry; 1351 struct uio *uio; 1352 int *tmp_ncookies = NULL; 1353 off_t offset; 1354 int error, i; 1355 1356 vp = ap->a_vp; 1357 mi = VFSTOMQFS(vp->v_mount); 1358 pd = VTON(vp); 1359 uio = ap->a_uio; 1360 1361 if (vp->v_type != VDIR) 1362 return (ENOTDIR); 1363 1364 if (uio->uio_offset < 0) 1365 return (EINVAL); 1366 1367 if (ap->a_ncookies != NULL) { 1368 tmp_ncookies = ap->a_ncookies; 1369 *ap->a_ncookies = 0; 1370 ap->a_ncookies = NULL; 1371 } 1372 1373 error = 0; 1374 offset = 0; 1375 1376 sx_xlock(&mi->mi_lock); 1377 1378 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) { 1379 entry.d_reclen = sizeof(entry); 1380 if (!pn->mn_fileno) 1381 mqfs_fileno_alloc(mi, pn); 1382 entry.d_fileno = pn->mn_fileno; 1383 for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i) 1384 entry.d_name[i] = pn->mn_name[i]; 1385 entry.d_name[i] = 0; 1386 entry.d_namlen = i; 1387 switch (pn->mn_type) { 1388 case mqfstype_root: 1389 case mqfstype_dir: 1390 case mqfstype_this: 1391 case mqfstype_parent: 1392 entry.d_type = DT_DIR; 1393 break; 1394 case mqfstype_file: 1395 entry.d_type = DT_REG; 1396 break; 1397 case mqfstype_symlink: 1398 entry.d_type = DT_LNK; 1399 break; 1400 default: 1401 panic("%s has unexpected node type: %d", pn->mn_name, 1402 pn->mn_type); 1403 } 1404 if (entry.d_reclen > uio->uio_resid) 1405 break; 1406 if (offset >= uio->uio_offset) { 1407 error = vfs_read_dirent(ap, &entry, offset); 1408 if (error) 1409 break; 1410 } 1411 offset += entry.d_reclen; 1412 } 1413 sx_xunlock(&mi->mi_lock); 1414 1415 uio->uio_offset = offset; 1416 1417 if (tmp_ncookies != NULL) 1418 ap->a_ncookies = tmp_ncookies; 1419 1420 return (error); 1421 } 1422 1423 #ifdef notyet 1424 1425 #if 0 1426 struct vop_mkdir_args { 1427 struct vnode *a_dvp; 1428 struvt vnode **a_vpp; 1429 struvt componentname *a_cnp; 1430 struct vattr *a_vap; 1431 }; 1432 #endif 1433 1434 /* 1435 * Create a directory. 1436 */ 1437 static int 1438 mqfs_mkdir(struct vop_mkdir_args *ap) 1439 { 1440 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1441 struct componentname *cnp = ap->a_cnp; 1442 struct mqfs_node *pd = VTON(ap->a_dvp); 1443 struct mqfs_node *pn; 1444 int error; 1445 1446 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) 1447 return (ENOTDIR); 1448 sx_xlock(&mqfs->mi_lock); 1449 if ((cnp->cn_flags & HASBUF) == 0) 1450 panic("%s: no name", __func__); 1451 pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen, 1452 ap->a_vap->cn_cred, ap->a_vap->va_mode); 1453 if (pn != NULL) 1454 mqnode_addref(pn); 1455 sx_xunlock(&mqfs->mi_lock); 1456 if (pn == NULL) { 1457 error = ENOSPC; 1458 } else { 1459 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); 1460 mqnode_release(pn); 1461 } 1462 return (error); 1463 } 1464 1465 #if 0 1466 struct vop_rmdir_args { 1467 struct vnode *a_dvp; 1468 struct vnode *a_vp; 1469 struct componentname *a_cnp; 1470 }; 1471 #endif 1472 1473 /* 1474 * Remove a directory. 1475 */ 1476 static int 1477 mqfs_rmdir(struct vop_rmdir_args *ap) 1478 { 1479 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1480 struct mqfs_node *pn = VTON(ap->a_vp); 1481 struct mqfs_node *pt; 1482 1483 if (pn->mn_type != mqfstype_dir) 1484 return (ENOTDIR); 1485 1486 sx_xlock(&mqfs->mi_lock); 1487 if (pn->mn_deleted) { 1488 sx_xunlock(&mqfs->mi_lock); 1489 return (ENOENT); 1490 } 1491 1492 pt = LIST_FIRST(&pn->mn_children); 1493 pt = LIST_NEXT(pt, mn_sibling); 1494 pt = LIST_NEXT(pt, mn_sibling); 1495 if (pt != NULL) { 1496 sx_xunlock(&mqfs->mi_lock); 1497 return (ENOTEMPTY); 1498 } 1499 pt = pn->mn_parent; 1500 pn->mn_parent = NULL; 1501 pn->mn_deleted = 1; 1502 LIST_REMOVE(pn, mn_sibling); 1503 mqnode_release(pn); 1504 mqnode_release(pt); 1505 sx_xunlock(&mqfs->mi_lock); 1506 cache_purge(ap->a_vp); 1507 return (0); 1508 } 1509 1510 #endif /* notyet */ 1511 1512 /* 1513 * Allocate a message queue 1514 */ 1515 static struct mqueue * 1516 mqueue_alloc(const struct mq_attr *attr) 1517 { 1518 struct mqueue *mq; 1519 1520 if (curmq >= maxmq) 1521 return (NULL); 1522 mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO); 1523 TAILQ_INIT(&mq->mq_msgq); 1524 if (attr != NULL) { 1525 mq->mq_maxmsg = attr->mq_maxmsg; 1526 mq->mq_msgsize = attr->mq_msgsize; 1527 } else { 1528 mq->mq_maxmsg = default_maxmsg; 1529 mq->mq_msgsize = default_msgsize; 1530 } 1531 mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF); 1532 knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL); 1533 knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL); 1534 atomic_add_int(&curmq, 1); 1535 return (mq); 1536 } 1537 1538 /* 1539 * Destroy a message queue 1540 */ 1541 static void 1542 mqueue_free(struct mqueue *mq) 1543 { 1544 struct mqueue_msg *msg; 1545 1546 while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) { 1547 TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link); 1548 FREE(msg, M_MQUEUEDATA); 1549 } 1550 1551 mtx_destroy(&mq->mq_mutex); 1552 knlist_destroy(&mq->mq_rsel.si_note); 1553 knlist_destroy(&mq->mq_wsel.si_note); 1554 uma_zfree(mqueue_zone, mq); 1555 atomic_add_int(&curmq, -1); 1556 } 1557 1558 /* 1559 * Load a message from user space 1560 */ 1561 static struct mqueue_msg * 1562 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio) 1563 { 1564 struct mqueue_msg *msg; 1565 size_t len; 1566 int error; 1567 1568 len = sizeof(struct mqueue_msg) + msg_size; 1569 MALLOC(msg, struct mqueue_msg *, len, M_MQUEUEDATA, M_WAITOK); 1570 error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg), 1571 msg_size); 1572 if (error) { 1573 FREE(msg, M_MQUEUEDATA); 1574 msg = NULL; 1575 } else { 1576 msg->msg_size = msg_size; 1577 msg->msg_prio = msg_prio; 1578 } 1579 return (msg); 1580 } 1581 1582 /* 1583 * Save a message to user space 1584 */ 1585 static int 1586 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio) 1587 { 1588 int error; 1589 1590 error = copyout(((char *)msg) + sizeof(*msg), msg_ptr, 1591 msg->msg_size); 1592 if (error == 0 && msg_prio != NULL) 1593 error = copyout(&msg->msg_prio, msg_prio, sizeof(int)); 1594 return (error); 1595 } 1596 1597 /* 1598 * Free a message's memory 1599 */ 1600 static __inline void 1601 mqueue_freemsg(struct mqueue_msg *msg) 1602 { 1603 FREE(msg, M_MQUEUEDATA); 1604 } 1605 1606 /* 1607 * Send a message. if waitok is false, thread will not be 1608 * blocked if there is no data in queue, otherwise, absolute 1609 * time will be checked. 1610 */ 1611 int 1612 mqueue_send(struct mqueue *mq, const char *msg_ptr, 1613 size_t msg_len, unsigned msg_prio, int waitok, 1614 const struct timespec *abs_timeout) 1615 { 1616 struct mqueue_msg *msg; 1617 struct timespec ets, ts, ts2; 1618 struct timeval tv; 1619 int error; 1620 1621 if (msg_prio >= MQ_PRIO_MAX) 1622 return (EINVAL); 1623 if (msg_len > mq->mq_msgsize) 1624 return (EMSGSIZE); 1625 msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio); 1626 if (msg == NULL) 1627 return (EFAULT); 1628 1629 /* O_NONBLOCK case */ 1630 if (!waitok) { 1631 error = _mqueue_send(mq, msg, -1); 1632 if (error) 1633 goto bad; 1634 return (0); 1635 } 1636 1637 /* we allow a null timeout (wait forever) */ 1638 if (abs_timeout == NULL) { 1639 error = _mqueue_send(mq, msg, 0); 1640 if (error) 1641 goto bad; 1642 return (0); 1643 } 1644 1645 /* send it before checking time */ 1646 error = _mqueue_send(mq, msg, -1); 1647 if (error == 0) 1648 return (0); 1649 1650 if (error != EAGAIN) 1651 goto bad; 1652 1653 error = copyin(abs_timeout, &ets, sizeof(ets)); 1654 if (error != 0) 1655 goto bad; 1656 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) { 1657 error = EINVAL; 1658 goto bad; 1659 } 1660 for (;;) { 1661 ts2 = ets; 1662 getnanotime(&ts); 1663 timespecsub(&ts2, &ts); 1664 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { 1665 error = ETIMEDOUT; 1666 break; 1667 } 1668 TIMESPEC_TO_TIMEVAL(&tv, &ts2); 1669 error = _mqueue_send(mq, msg, tvtohz(&tv)); 1670 if (error != ETIMEDOUT) 1671 break; 1672 } 1673 if (error == 0) 1674 return (0); 1675 bad: 1676 mqueue_freemsg(msg); 1677 return (error); 1678 } 1679 1680 /* 1681 * Common routine to send a message 1682 */ 1683 static int 1684 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo) 1685 { 1686 struct mqueue_msg *msg2; 1687 int error = 0; 1688 1689 mtx_lock(&mq->mq_mutex); 1690 while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) { 1691 if (timo < 0) { 1692 mtx_unlock(&mq->mq_mutex); 1693 return (EAGAIN); 1694 } 1695 mq->mq_senders++; 1696 error = msleep(&mq->mq_senders, &mq->mq_mutex, 1697 PCATCH, "mqsend", timo); 1698 mq->mq_senders--; 1699 if (error == EAGAIN) 1700 error = ETIMEDOUT; 1701 } 1702 if (mq->mq_curmsgs >= mq->mq_maxmsg) { 1703 mtx_unlock(&mq->mq_mutex); 1704 return (error); 1705 } 1706 error = 0; 1707 if (TAILQ_EMPTY(&mq->mq_msgq)) { 1708 TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link); 1709 } else { 1710 if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) { 1711 TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link); 1712 } else { 1713 TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) { 1714 if (msg2->msg_prio < msg->msg_prio) 1715 break; 1716 } 1717 TAILQ_INSERT_BEFORE(msg2, msg, msg_link); 1718 } 1719 } 1720 mq->mq_curmsgs++; 1721 mq->mq_totalbytes += msg->msg_size; 1722 if (mq->mq_receivers) 1723 wakeup_one(&mq->mq_receivers); 1724 else if (mq->mq_notifier != NULL) 1725 mqueue_send_notification(mq); 1726 if (mq->mq_flags & MQ_RSEL) { 1727 mq->mq_flags &= ~MQ_RSEL; 1728 selwakeup(&mq->mq_rsel); 1729 } 1730 KNOTE_LOCKED(&mq->mq_rsel.si_note, 0); 1731 mtx_unlock(&mq->mq_mutex); 1732 return (0); 1733 } 1734 1735 /* 1736 * Send realtime a signal to process which registered itself 1737 * successfully by mq_notify. 1738 */ 1739 static void 1740 mqueue_send_notification(struct mqueue *mq) 1741 { 1742 struct mqueue_notifier *nt; 1743 struct proc *p; 1744 1745 mtx_assert(&mq->mq_mutex, MA_OWNED); 1746 nt = mq->mq_notifier; 1747 if (nt->nt_sigev.sigev_notify != SIGEV_NONE) { 1748 p = nt->nt_proc; 1749 PROC_LOCK(p); 1750 if (!KSI_ONQ(&nt->nt_ksi)) 1751 psignal_event(p, &nt->nt_sigev, &nt->nt_ksi); 1752 PROC_UNLOCK(p); 1753 } 1754 mq->mq_notifier = NULL; 1755 } 1756 1757 /* 1758 * Get a message. if waitok is false, thread will not be 1759 * blocked if there is no data in queue, otherwise, absolute 1760 * time will be checked. 1761 */ 1762 int 1763 mqueue_receive(struct mqueue *mq, char *msg_ptr, 1764 size_t msg_len, unsigned *msg_prio, int waitok, 1765 const struct timespec *abs_timeout) 1766 { 1767 struct mqueue_msg *msg; 1768 struct timespec ets, ts, ts2; 1769 struct timeval tv; 1770 int error; 1771 1772 if (msg_len < mq->mq_msgsize) 1773 return (EMSGSIZE); 1774 1775 /* O_NONBLOCK case */ 1776 if (!waitok) { 1777 error = _mqueue_recv(mq, &msg, -1); 1778 if (error) 1779 return (error); 1780 goto received; 1781 } 1782 1783 /* we allow a null timeout (wait forever). */ 1784 if (abs_timeout == NULL) { 1785 error = _mqueue_recv(mq, &msg, 0); 1786 if (error) 1787 return (error); 1788 goto received; 1789 } 1790 1791 /* try to get a message before checking time */ 1792 error = _mqueue_recv(mq, &msg, -1); 1793 if (error == 0) 1794 goto received; 1795 1796 if (error != EAGAIN) 1797 return (error); 1798 1799 error = copyin(abs_timeout, &ets, sizeof(ets)); 1800 if (error != 0) 1801 return (error); 1802 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) { 1803 error = EINVAL; 1804 return (error); 1805 } 1806 1807 for (;;) { 1808 ts2 = ets; 1809 getnanotime(&ts); 1810 timespecsub(&ts2, &ts); 1811 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { 1812 error = ETIMEDOUT; 1813 return (error); 1814 } 1815 TIMESPEC_TO_TIMEVAL(&tv, &ts2); 1816 error = _mqueue_recv(mq, &msg, tvtohz(&tv)); 1817 if (error == 0) 1818 break; 1819 if (error != ETIMEDOUT) 1820 return (error); 1821 } 1822 1823 received: 1824 error = mqueue_savemsg(msg, msg_ptr, msg_prio); 1825 if (error == 0) { 1826 curthread->td_retval[0] = msg->msg_size; 1827 curthread->td_retval[1] = 0; 1828 } 1829 mqueue_freemsg(msg); 1830 return (error); 1831 } 1832 1833 /* 1834 * Common routine to receive a message 1835 */ 1836 static int 1837 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo) 1838 { 1839 int error = 0; 1840 1841 mtx_lock(&mq->mq_mutex); 1842 while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) { 1843 if (timo < 0) { 1844 mtx_unlock(&mq->mq_mutex); 1845 return (EAGAIN); 1846 } 1847 mq->mq_receivers++; 1848 error = msleep(&mq->mq_receivers, &mq->mq_mutex, 1849 PCATCH, "mqrecv", timo); 1850 mq->mq_receivers--; 1851 if (error == EAGAIN) 1852 error = ETIMEDOUT; 1853 } 1854 if (*msg != NULL) { 1855 error = 0; 1856 TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link); 1857 mq->mq_curmsgs--; 1858 mq->mq_totalbytes -= (*msg)->msg_size; 1859 if (mq->mq_senders) 1860 wakeup_one(&mq->mq_senders); 1861 if (mq->mq_flags & MQ_WSEL) { 1862 mq->mq_flags &= ~MQ_WSEL; 1863 selwakeup(&mq->mq_wsel); 1864 } 1865 KNOTE_LOCKED(&mq->mq_wsel.si_note, 0); 1866 } 1867 if (mq->mq_notifier != NULL && mq->mq_receivers == 0 && 1868 !TAILQ_EMPTY(&mq->mq_msgq)) { 1869 mqueue_send_notification(mq); 1870 } 1871 mtx_unlock(&mq->mq_mutex); 1872 return (error); 1873 } 1874 1875 static __inline struct mqueue_notifier * 1876 notifier_alloc(void) 1877 { 1878 return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO)); 1879 } 1880 1881 static __inline void 1882 notifier_free(struct mqueue_notifier *p) 1883 { 1884 uma_zfree(mqnoti_zone, p); 1885 } 1886 1887 static struct mqueue_notifier * 1888 notifier_search(struct proc *p, int fd) 1889 { 1890 struct mqueue_notifier *nt; 1891 1892 LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) { 1893 if (nt->nt_ksi.ksi_mqd == fd) 1894 break; 1895 } 1896 return (nt); 1897 } 1898 1899 static __inline void 1900 notifier_insert(struct proc *p, struct mqueue_notifier *nt) 1901 { 1902 LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link); 1903 } 1904 1905 static __inline void 1906 notifier_delete(struct proc *p, struct mqueue_notifier *nt) 1907 { 1908 LIST_REMOVE(nt, nt_link); 1909 notifier_free(nt); 1910 } 1911 1912 static void 1913 notifier_remove(struct proc *p, struct mqueue *mq, int fd) 1914 { 1915 struct mqueue_notifier *nt; 1916 1917 mtx_assert(&mq->mq_mutex, MA_OWNED); 1918 PROC_LOCK(p); 1919 nt = notifier_search(p, fd); 1920 if (nt != NULL) { 1921 if (mq->mq_notifier == nt) 1922 mq->mq_notifier = NULL; 1923 sigqueue_take(&nt->nt_ksi); 1924 notifier_delete(p, nt); 1925 } 1926 PROC_UNLOCK(p); 1927 } 1928 1929 /* 1930 * Syscall to open a message queue. 1931 */ 1932 int 1933 kmq_open(struct thread *td, struct kmq_open_args *uap) 1934 { 1935 char path[MQFS_NAMELEN + 1]; 1936 struct mq_attr attr, *pattr; 1937 struct mqfs_node *pn; 1938 struct filedesc *fdp; 1939 struct file *fp; 1940 struct mqueue *mq; 1941 int fd, error, len, flags, cmode; 1942 1943 if ((uap->flags & O_ACCMODE) == O_ACCMODE) 1944 return (EINVAL); 1945 1946 fdp = td->td_proc->p_fd; 1947 flags = FFLAGS(uap->flags); 1948 cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT); 1949 mq = NULL; 1950 if ((flags & O_CREAT) && (uap->attr != NULL)) { 1951 error = copyin(uap->attr, &attr, sizeof(attr)); 1952 if (error) 1953 return (error); 1954 if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg) 1955 return (EINVAL); 1956 if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize) 1957 return (EINVAL); 1958 pattr = &attr; 1959 } else 1960 pattr = NULL; 1961 1962 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL); 1963 if (error) 1964 return (error); 1965 1966 /* 1967 * The first character of name must be a slash (/) character 1968 * and the remaining characters of name cannot include any slash 1969 * characters. 1970 */ 1971 len = strlen(path); 1972 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL) 1973 return (EINVAL); 1974 1975 error = falloc(td, &fp, &fd); 1976 if (error) 1977 return (error); 1978 1979 sx_xlock(&mqfs_data.mi_lock); 1980 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1); 1981 if (pn == NULL) { 1982 if (!(flags & O_CREAT)) { 1983 error = ENOENT; 1984 } else { 1985 mq = mqueue_alloc(pattr); 1986 if (mq == NULL) { 1987 error = ENFILE; 1988 } else { 1989 pn = mqfs_create_file(mqfs_data.mi_root, 1990 path + 1, len - 1, td->td_ucred, 1991 cmode); 1992 if (pn == NULL) { 1993 error = ENOSPC; 1994 mqueue_free(mq); 1995 } 1996 } 1997 } 1998 1999 if (error == 0) { 2000 pn->mn_data = mq; 2001 } 2002 } else { 2003 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) { 2004 error = EEXIST; 2005 } else { 2006 int acc_mode = 0; 2007 2008 if (flags & FREAD) 2009 acc_mode |= VREAD; 2010 if (flags & FWRITE) 2011 acc_mode |= VWRITE; 2012 error = vaccess(VREG, pn->mn_mode, pn->mn_uid, 2013 pn->mn_gid, acc_mode, td->td_ucred, NULL); 2014 } 2015 } 2016 2017 if (error) { 2018 sx_xunlock(&mqfs_data.mi_lock); 2019 fdclose(fdp, fp, fd, td); 2020 fdrop(fp, td); 2021 return (error); 2022 } 2023 2024 mqnode_addref(pn); 2025 sx_xunlock(&mqfs_data.mi_lock); 2026 2027 finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn, 2028 &mqueueops); 2029 2030 FILEDESC_XLOCK(fdp); 2031 if (fdp->fd_ofiles[fd] == fp) 2032 fdp->fd_ofileflags[fd] |= UF_EXCLOSE; 2033 FILEDESC_XUNLOCK(fdp); 2034 td->td_retval[0] = fd; 2035 fdrop(fp, td); 2036 return (0); 2037 } 2038 2039 /* 2040 * Syscall to unlink a message queue. 2041 */ 2042 int 2043 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap) 2044 { 2045 char path[MQFS_NAMELEN+1]; 2046 struct mqfs_node *pn; 2047 int error, len; 2048 2049 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL); 2050 if (error) 2051 return (error); 2052 2053 len = strlen(path); 2054 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL) 2055 return (EINVAL); 2056 2057 sx_xlock(&mqfs_data.mi_lock); 2058 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1); 2059 if (pn != NULL) 2060 error = do_unlink(pn, td->td_ucred); 2061 else 2062 error = ENOENT; 2063 sx_xunlock(&mqfs_data.mi_lock); 2064 return (error); 2065 } 2066 2067 typedef int (*_fgetf)(struct thread *, int, struct file **); 2068 2069 /* 2070 * Get message queue by giving file slot 2071 */ 2072 static int 2073 _getmq(struct thread *td, int fd, _fgetf func, 2074 struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) 2075 { 2076 struct mqfs_node *pn; 2077 int error; 2078 2079 error = func(td, fd, fpp); 2080 if (error) 2081 return (error); 2082 if (&mqueueops != (*fpp)->f_ops) { 2083 fdrop(*fpp, td); 2084 return (EBADF); 2085 } 2086 pn = (*fpp)->f_data; 2087 if (ppn) 2088 *ppn = pn; 2089 if (pmq) 2090 *pmq = pn->mn_data; 2091 return (0); 2092 } 2093 2094 static __inline int 2095 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn, 2096 struct mqueue **pmq) 2097 { 2098 return _getmq(td, fd, fget, fpp, ppn, pmq); 2099 } 2100 2101 static __inline int 2102 getmq_read(struct thread *td, int fd, struct file **fpp, 2103 struct mqfs_node **ppn, struct mqueue **pmq) 2104 { 2105 return _getmq(td, fd, fget_read, fpp, ppn, pmq); 2106 } 2107 2108 static __inline int 2109 getmq_write(struct thread *td, int fd, struct file **fpp, 2110 struct mqfs_node **ppn, struct mqueue **pmq) 2111 { 2112 return _getmq(td, fd, fget_write, fpp, ppn, pmq); 2113 } 2114 2115 int 2116 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap) 2117 { 2118 struct mqueue *mq; 2119 struct file *fp; 2120 struct mq_attr attr, oattr; 2121 u_int oflag, flag; 2122 int error; 2123 2124 if (uap->attr) { 2125 error = copyin(uap->attr, &attr, sizeof(attr)); 2126 if (error) 2127 return (error); 2128 if (attr.mq_flags & ~O_NONBLOCK) 2129 return (EINVAL); 2130 } 2131 error = getmq(td, uap->mqd, &fp, NULL, &mq); 2132 if (error) 2133 return (error); 2134 oattr.mq_maxmsg = mq->mq_maxmsg; 2135 oattr.mq_msgsize = mq->mq_msgsize; 2136 oattr.mq_curmsgs = mq->mq_curmsgs; 2137 if (uap->attr) { 2138 do { 2139 oflag = flag = fp->f_flag; 2140 flag &= ~O_NONBLOCK; 2141 flag |= (attr.mq_flags & O_NONBLOCK); 2142 } while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0); 2143 } else 2144 oflag = fp->f_flag; 2145 oattr.mq_flags = (O_NONBLOCK & oflag); 2146 fdrop(fp, td); 2147 if (uap->oattr) 2148 error = copyout(&oattr, uap->oattr, sizeof(oattr)); 2149 return (error); 2150 } 2151 2152 int 2153 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap) 2154 { 2155 struct mqueue *mq; 2156 struct file *fp; 2157 int error; 2158 int waitok; 2159 2160 error = getmq_read(td, uap->mqd, &fp, NULL, &mq); 2161 if (error) 2162 return (error); 2163 waitok = !(fp->f_flag & O_NONBLOCK); 2164 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len, 2165 uap->msg_prio, waitok, uap->abs_timeout); 2166 fdrop(fp, td); 2167 return (error); 2168 } 2169 2170 int 2171 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap) 2172 { 2173 struct mqueue *mq; 2174 struct file *fp; 2175 int error, waitok; 2176 2177 error = getmq_write(td, uap->mqd, &fp, NULL, &mq); 2178 if (error) 2179 return (error); 2180 waitok = !(fp->f_flag & O_NONBLOCK); 2181 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len, 2182 uap->msg_prio, waitok, uap->abs_timeout); 2183 fdrop(fp, td); 2184 return (error); 2185 } 2186 2187 int 2188 kmq_notify(struct thread *td, struct kmq_notify_args *uap) 2189 { 2190 struct sigevent ev; 2191 struct filedesc *fdp; 2192 struct proc *p; 2193 struct mqueue *mq; 2194 struct file *fp; 2195 struct mqueue_notifier *nt, *newnt = NULL; 2196 int error; 2197 2198 p = td->td_proc; 2199 fdp = td->td_proc->p_fd; 2200 if (uap->sigev) { 2201 error = copyin(uap->sigev, &ev, sizeof(ev)); 2202 if (error) 2203 return (error); 2204 if (ev.sigev_notify != SIGEV_SIGNAL && 2205 ev.sigev_notify != SIGEV_THREAD_ID && 2206 ev.sigev_notify != SIGEV_NONE) 2207 return (EINVAL); 2208 if ((ev.sigev_notify == SIGEV_SIGNAL || 2209 ev.sigev_notify == SIGEV_THREAD_ID) && 2210 !_SIG_VALID(ev.sigev_signo)) 2211 return (EINVAL); 2212 } 2213 error = getmq(td, uap->mqd, &fp, NULL, &mq); 2214 if (error) 2215 return (error); 2216 again: 2217 FILEDESC_SLOCK(fdp); 2218 if (fget_locked(fdp, uap->mqd) != fp) { 2219 FILEDESC_SUNLOCK(fdp); 2220 error = EBADF; 2221 goto out; 2222 } 2223 mtx_lock(&mq->mq_mutex); 2224 FILEDESC_SUNLOCK(fdp); 2225 if (uap->sigev != NULL) { 2226 if (mq->mq_notifier != NULL) { 2227 error = EBUSY; 2228 } else { 2229 PROC_LOCK(p); 2230 nt = notifier_search(p, uap->mqd); 2231 if (nt == NULL) { 2232 if (newnt == NULL) { 2233 PROC_UNLOCK(p); 2234 mtx_unlock(&mq->mq_mutex); 2235 newnt = notifier_alloc(); 2236 goto again; 2237 } 2238 } 2239 2240 if (nt != NULL) { 2241 sigqueue_take(&nt->nt_ksi); 2242 if (newnt != NULL) { 2243 notifier_free(newnt); 2244 newnt = NULL; 2245 } 2246 } else { 2247 nt = newnt; 2248 newnt = NULL; 2249 ksiginfo_init(&nt->nt_ksi); 2250 nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT; 2251 nt->nt_ksi.ksi_code = SI_MESGQ; 2252 nt->nt_proc = p; 2253 nt->nt_ksi.ksi_mqd = uap->mqd; 2254 notifier_insert(p, nt); 2255 } 2256 nt->nt_sigev = ev; 2257 mq->mq_notifier = nt; 2258 PROC_UNLOCK(p); 2259 /* 2260 * if there is no receivers and message queue 2261 * is not empty, we should send notification 2262 * as soon as possible. 2263 */ 2264 if (mq->mq_receivers == 0 && 2265 !TAILQ_EMPTY(&mq->mq_msgq)) 2266 mqueue_send_notification(mq); 2267 } 2268 } else { 2269 notifier_remove(p, mq, uap->mqd); 2270 } 2271 mtx_unlock(&mq->mq_mutex); 2272 2273 out: 2274 fdrop(fp, td); 2275 if (newnt != NULL) 2276 notifier_free(newnt); 2277 return (error); 2278 } 2279 2280 static void 2281 mqueue_fdclose(struct thread *td, int fd, struct file *fp) 2282 { 2283 struct filedesc *fdp; 2284 struct mqueue *mq; 2285 2286 fdp = td->td_proc->p_fd; 2287 FILEDESC_LOCK_ASSERT(fdp); 2288 2289 if (fp->f_ops == &mqueueops) { 2290 mq = FPTOMQ(fp); 2291 mtx_lock(&mq->mq_mutex); 2292 notifier_remove(td->td_proc, mq, fd); 2293 2294 /* have to wakeup thread in same process */ 2295 if (mq->mq_flags & MQ_RSEL) { 2296 mq->mq_flags &= ~MQ_RSEL; 2297 selwakeup(&mq->mq_rsel); 2298 } 2299 if (mq->mq_flags & MQ_WSEL) { 2300 mq->mq_flags &= ~MQ_WSEL; 2301 selwakeup(&mq->mq_wsel); 2302 } 2303 mtx_unlock(&mq->mq_mutex); 2304 } 2305 } 2306 2307 static void 2308 mq_proc_exit(void *arg __unused, struct proc *p) 2309 { 2310 struct filedesc *fdp; 2311 struct file *fp; 2312 struct mqueue *mq; 2313 int i; 2314 2315 fdp = p->p_fd; 2316 FILEDESC_SLOCK(fdp); 2317 for (i = 0; i < fdp->fd_nfiles; ++i) { 2318 fp = fget_locked(fdp, i); 2319 if (fp != NULL && fp->f_ops == &mqueueops) { 2320 mq = FPTOMQ(fp); 2321 mtx_lock(&mq->mq_mutex); 2322 notifier_remove(p, FPTOMQ(fp), i); 2323 mtx_unlock(&mq->mq_mutex); 2324 } 2325 } 2326 FILEDESC_SUNLOCK(fdp); 2327 KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left")); 2328 } 2329 2330 static int 2331 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 2332 int flags, struct thread *td) 2333 { 2334 return (EOPNOTSUPP); 2335 } 2336 2337 static int 2338 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 2339 int flags, struct thread *td) 2340 { 2341 return (EOPNOTSUPP); 2342 } 2343 2344 static int 2345 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred, 2346 struct thread *td) 2347 { 2348 2349 return (EINVAL); 2350 } 2351 2352 static int 2353 mqf_ioctl(struct file *fp, u_long cmd, void *data, 2354 struct ucred *active_cred, struct thread *td) 2355 { 2356 return (ENOTTY); 2357 } 2358 2359 static int 2360 mqf_poll(struct file *fp, int events, struct ucred *active_cred, 2361 struct thread *td) 2362 { 2363 struct mqueue *mq = FPTOMQ(fp); 2364 int revents = 0; 2365 2366 mtx_lock(&mq->mq_mutex); 2367 if (events & (POLLIN | POLLRDNORM)) { 2368 if (mq->mq_curmsgs) { 2369 revents |= events & (POLLIN | POLLRDNORM); 2370 } else { 2371 mq->mq_flags |= MQ_RSEL; 2372 selrecord(td, &mq->mq_rsel); 2373 } 2374 } 2375 if (events & POLLOUT) { 2376 if (mq->mq_curmsgs < mq->mq_maxmsg) 2377 revents |= POLLOUT; 2378 else { 2379 mq->mq_flags |= MQ_WSEL; 2380 selrecord(td, &mq->mq_wsel); 2381 } 2382 } 2383 mtx_unlock(&mq->mq_mutex); 2384 return (revents); 2385 } 2386 2387 static int 2388 mqf_close(struct file *fp, struct thread *td) 2389 { 2390 struct mqfs_node *pn; 2391 2392 fp->f_ops = &badfileops; 2393 pn = fp->f_data; 2394 fp->f_data = NULL; 2395 sx_xlock(&mqfs_data.mi_lock); 2396 mqnode_release(pn); 2397 sx_xunlock(&mqfs_data.mi_lock); 2398 return (0); 2399 } 2400 2401 static int 2402 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 2403 struct thread *td) 2404 { 2405 struct mqfs_node *pn = fp->f_data; 2406 2407 bzero(st, sizeof *st); 2408 st->st_atimespec = pn->mn_atime; 2409 st->st_mtimespec = pn->mn_mtime; 2410 st->st_ctimespec = pn->mn_ctime; 2411 st->st_birthtimespec = pn->mn_birth; 2412 st->st_uid = pn->mn_uid; 2413 st->st_gid = pn->mn_gid; 2414 st->st_mode = S_IFIFO | pn->mn_mode; 2415 return (0); 2416 } 2417 2418 static int 2419 mqf_kqfilter(struct file *fp, struct knote *kn) 2420 { 2421 struct mqueue *mq = FPTOMQ(fp); 2422 int error = 0; 2423 2424 if (kn->kn_filter == EVFILT_READ) { 2425 kn->kn_fop = &mq_rfiltops; 2426 knlist_add(&mq->mq_rsel.si_note, kn, 0); 2427 } else if (kn->kn_filter == EVFILT_WRITE) { 2428 kn->kn_fop = &mq_wfiltops; 2429 knlist_add(&mq->mq_wsel.si_note, kn, 0); 2430 } else 2431 error = EINVAL; 2432 return (error); 2433 } 2434 2435 static void 2436 filt_mqdetach(struct knote *kn) 2437 { 2438 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2439 2440 if (kn->kn_filter == EVFILT_READ) 2441 knlist_remove(&mq->mq_rsel.si_note, kn, 0); 2442 else if (kn->kn_filter == EVFILT_WRITE) 2443 knlist_remove(&mq->mq_wsel.si_note, kn, 0); 2444 else 2445 panic("filt_mqdetach"); 2446 } 2447 2448 static int 2449 filt_mqread(struct knote *kn, long hint) 2450 { 2451 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2452 2453 mtx_assert(&mq->mq_mutex, MA_OWNED); 2454 return (mq->mq_curmsgs != 0); 2455 } 2456 2457 static int 2458 filt_mqwrite(struct knote *kn, long hint) 2459 { 2460 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2461 2462 mtx_assert(&mq->mq_mutex, MA_OWNED); 2463 return (mq->mq_curmsgs < mq->mq_maxmsg); 2464 } 2465 2466 static struct fileops mqueueops = { 2467 .fo_read = mqf_read, 2468 .fo_write = mqf_write, 2469 .fo_truncate = mqf_truncate, 2470 .fo_ioctl = mqf_ioctl, 2471 .fo_poll = mqf_poll, 2472 .fo_kqfilter = mqf_kqfilter, 2473 .fo_stat = mqf_stat, 2474 .fo_close = mqf_close 2475 }; 2476 2477 static struct vop_vector mqfs_vnodeops = { 2478 .vop_default = &default_vnodeops, 2479 .vop_access = mqfs_access, 2480 .vop_cachedlookup = mqfs_lookup, 2481 .vop_lookup = vfs_cache_lookup, 2482 .vop_reclaim = mqfs_reclaim, 2483 .vop_create = mqfs_create, 2484 .vop_remove = mqfs_remove, 2485 .vop_inactive = mqfs_inactive, 2486 .vop_open = mqfs_open, 2487 .vop_close = mqfs_close, 2488 .vop_getattr = mqfs_getattr, 2489 .vop_setattr = mqfs_setattr, 2490 .vop_read = mqfs_read, 2491 .vop_write = VOP_EOPNOTSUPP, 2492 .vop_readdir = mqfs_readdir, 2493 .vop_mkdir = VOP_EOPNOTSUPP, 2494 .vop_rmdir = VOP_EOPNOTSUPP 2495 }; 2496 2497 static struct vfsops mqfs_vfsops = { 2498 .vfs_init = mqfs_init, 2499 .vfs_uninit = mqfs_uninit, 2500 .vfs_mount = mqfs_mount, 2501 .vfs_unmount = mqfs_unmount, 2502 .vfs_root = mqfs_root, 2503 .vfs_statfs = mqfs_statfs, 2504 }; 2505 2506 SYSCALL_MODULE_HELPER(kmq_open); 2507 SYSCALL_MODULE_HELPER(kmq_setattr); 2508 SYSCALL_MODULE_HELPER(kmq_timedsend); 2509 SYSCALL_MODULE_HELPER(kmq_timedreceive); 2510 SYSCALL_MODULE_HELPER(kmq_notify); 2511 SYSCALL_MODULE_HELPER(kmq_unlink); 2512 2513 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC); 2514 MODULE_VERSION(mqueuefs, 1); 2515