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 pn->mn_name[len] == '\0') 798 return (pn); 799 } 800 return (NULL); 801 } 802 803 /* 804 * Look up a file or directory. 805 */ 806 static int 807 mqfs_lookupx(struct vop_cachedlookup_args *ap) 808 { 809 struct componentname *cnp; 810 struct vnode *dvp, **vpp; 811 struct mqfs_node *pd; 812 struct mqfs_node *pn; 813 struct mqfs_info *mqfs; 814 int nameiop, flags, error, namelen; 815 char *pname; 816 struct thread *td; 817 818 cnp = ap->a_cnp; 819 vpp = ap->a_vpp; 820 dvp = ap->a_dvp; 821 pname = cnp->cn_nameptr; 822 namelen = cnp->cn_namelen; 823 td = cnp->cn_thread; 824 flags = cnp->cn_flags; 825 nameiop = cnp->cn_nameiop; 826 pd = VTON(dvp); 827 pn = NULL; 828 mqfs = pd->mn_info; 829 *vpp = NULLVP; 830 831 if (dvp->v_type != VDIR) 832 return (ENOTDIR); 833 834 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread); 835 if (error) 836 return (error); 837 838 /* shortcut: check if the name is too long */ 839 if (cnp->cn_namelen >= MQFS_NAMELEN) 840 return (ENOENT); 841 842 /* self */ 843 if (namelen == 1 && pname[0] == '.') { 844 if ((flags & ISLASTCN) && nameiop != LOOKUP) 845 return (EINVAL); 846 pn = pd; 847 *vpp = dvp; 848 VREF(dvp); 849 return (0); 850 } 851 852 /* parent */ 853 if (cnp->cn_flags & ISDOTDOT) { 854 if (dvp->v_vflag & VV_ROOT) 855 return (EIO); 856 if ((flags & ISLASTCN) && nameiop != LOOKUP) 857 return (EINVAL); 858 VOP_UNLOCK(dvp, 0); 859 KASSERT(pd->mn_parent, ("non-root directory has no parent")); 860 pn = pd->mn_parent; 861 error = mqfs_allocv(dvp->v_mount, vpp, pn); 862 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 863 return (error); 864 } 865 866 /* named node */ 867 sx_xlock(&mqfs->mi_lock); 868 pn = mqfs_search(pd, pname, namelen); 869 if (pn != NULL) 870 mqnode_addref(pn); 871 sx_xunlock(&mqfs->mi_lock); 872 873 /* found */ 874 if (pn != NULL) { 875 /* DELETE */ 876 if (nameiop == DELETE && (flags & ISLASTCN)) { 877 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 878 if (error) { 879 mqnode_release(pn); 880 return (error); 881 } 882 if (*vpp == dvp) { 883 VREF(dvp); 884 *vpp = dvp; 885 mqnode_release(pn); 886 return (0); 887 } 888 } 889 890 /* allocate vnode */ 891 error = mqfs_allocv(dvp->v_mount, vpp, pn); 892 mqnode_release(pn); 893 if (error == 0 && cnp->cn_flags & MAKEENTRY) 894 cache_enter(dvp, *vpp, cnp); 895 return (error); 896 } 897 898 /* not found */ 899 900 /* will create a new entry in the directory ? */ 901 if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT) 902 && (flags & ISLASTCN)) { 903 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td); 904 if (error) 905 return (error); 906 cnp->cn_flags |= SAVENAME; 907 return (EJUSTRETURN); 908 } 909 return (ENOENT); 910 } 911 912 #if 0 913 struct vop_lookup_args { 914 struct vop_generic_args a_gen; 915 struct vnode *a_dvp; 916 struct vnode **a_vpp; 917 struct componentname *a_cnp; 918 }; 919 #endif 920 921 /* 922 * vnode lookup operation 923 */ 924 static int 925 mqfs_lookup(struct vop_cachedlookup_args *ap) 926 { 927 int rc; 928 929 rc = mqfs_lookupx(ap); 930 return (rc); 931 } 932 933 #if 0 934 struct vop_create_args { 935 struct vnode *a_dvp; 936 struct vnode **a_vpp; 937 struct componentname *a_cnp; 938 struct vattr *a_vap; 939 }; 940 #endif 941 942 /* 943 * vnode creation operation 944 */ 945 static int 946 mqfs_create(struct vop_create_args *ap) 947 { 948 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 949 struct componentname *cnp = ap->a_cnp; 950 struct mqfs_node *pd; 951 struct mqfs_node *pn; 952 struct mqueue *mq; 953 int error; 954 955 pd = VTON(ap->a_dvp); 956 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) 957 return (ENOTDIR); 958 mq = mqueue_alloc(NULL); 959 if (mq == NULL) 960 return (EAGAIN); 961 sx_xlock(&mqfs->mi_lock); 962 if ((cnp->cn_flags & HASBUF) == 0) 963 panic("%s: no name", __func__); 964 pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen, 965 cnp->cn_cred, ap->a_vap->va_mode); 966 if (pn == NULL) { 967 sx_xunlock(&mqfs->mi_lock); 968 error = ENOSPC; 969 } else { 970 mqnode_addref(pn); 971 sx_xunlock(&mqfs->mi_lock); 972 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); 973 mqnode_release(pn); 974 if (error) 975 mqfs_destroy(pn); 976 else 977 pn->mn_data = mq; 978 } 979 if (error) 980 mqueue_free(mq); 981 return (error); 982 } 983 984 /* 985 * Remove an entry 986 */ 987 static 988 int do_unlink(struct mqfs_node *pn, struct ucred *ucred) 989 { 990 struct mqfs_node *parent; 991 struct mqfs_vdata *vd; 992 int error = 0; 993 994 sx_assert(&pn->mn_info->mi_lock, SX_LOCKED); 995 996 if (ucred->cr_uid != pn->mn_uid && 997 (error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0) 998 error = EACCES; 999 else if (!pn->mn_deleted) { 1000 parent = pn->mn_parent; 1001 pn->mn_parent = NULL; 1002 pn->mn_deleted = 1; 1003 LIST_REMOVE(pn, mn_sibling); 1004 LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) { 1005 cache_purge(vd->mv_vnode); 1006 vhold(vd->mv_vnode); 1007 taskqueue_enqueue(taskqueue_thread, &vd->mv_task); 1008 } 1009 mqnode_release(pn); 1010 mqnode_release(parent); 1011 } else 1012 error = ENOENT; 1013 return (error); 1014 } 1015 1016 #if 0 1017 struct vop_remove_args { 1018 struct vnode *a_dvp; 1019 struct vnode *a_vp; 1020 struct componentname *a_cnp; 1021 }; 1022 #endif 1023 1024 /* 1025 * vnode removal operation 1026 */ 1027 static int 1028 mqfs_remove(struct vop_remove_args *ap) 1029 { 1030 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1031 struct mqfs_node *pn; 1032 int error; 1033 1034 if (ap->a_vp->v_type == VDIR) 1035 return (EPERM); 1036 pn = VTON(ap->a_vp); 1037 sx_xlock(&mqfs->mi_lock); 1038 error = do_unlink(pn, ap->a_cnp->cn_cred); 1039 sx_xunlock(&mqfs->mi_lock); 1040 return (error); 1041 } 1042 1043 #if 0 1044 struct vop_inactive_args { 1045 struct vnode *a_vp; 1046 struct thread *a_td; 1047 }; 1048 #endif 1049 1050 static int 1051 mqfs_inactive(struct vop_inactive_args *ap) 1052 { 1053 struct mqfs_node *pn = VTON(ap->a_vp); 1054 1055 if (pn->mn_deleted) 1056 vrecycle(ap->a_vp, ap->a_td); 1057 return (0); 1058 } 1059 1060 #if 0 1061 struct vop_reclaim_args { 1062 struct vop_generic_args a_gen; 1063 struct vnode *a_vp; 1064 struct thread *a_td; 1065 }; 1066 #endif 1067 1068 static int 1069 mqfs_reclaim(struct vop_reclaim_args *ap) 1070 { 1071 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount); 1072 struct vnode *vp = ap->a_vp; 1073 struct mqfs_node *pn; 1074 struct mqfs_vdata *vd; 1075 1076 vd = vp->v_data; 1077 pn = vd->mv_node; 1078 sx_xlock(&mqfs->mi_lock); 1079 vp->v_data = NULL; 1080 LIST_REMOVE(vd, mv_link); 1081 uma_zfree(mvdata_zone, vd); 1082 mqnode_release(pn); 1083 sx_xunlock(&mqfs->mi_lock); 1084 return (0); 1085 } 1086 1087 #if 0 1088 struct vop_open_args { 1089 struct vop_generic_args a_gen; 1090 struct vnode *a_vp; 1091 int a_mode; 1092 struct ucred *a_cred; 1093 struct thread *a_td; 1094 int a_fdidx; 1095 }; 1096 #endif 1097 1098 static int 1099 mqfs_open(struct vop_open_args *ap) 1100 { 1101 return (0); 1102 } 1103 1104 #if 0 1105 struct vop_close_args { 1106 struct vop_generic_args a_gen; 1107 struct vnode *a_vp; 1108 int a_fflag; 1109 struct ucred *a_cred; 1110 struct thread *a_td; 1111 }; 1112 #endif 1113 1114 static int 1115 mqfs_close(struct vop_close_args *ap) 1116 { 1117 return (0); 1118 } 1119 1120 #if 0 1121 struct vop_access_args { 1122 struct vop_generic_args a_gen; 1123 struct vnode *a_vp; 1124 accmode_t a_accmode; 1125 struct ucred *a_cred; 1126 struct thread *a_td; 1127 }; 1128 #endif 1129 1130 /* 1131 * Verify permissions 1132 */ 1133 static int 1134 mqfs_access(struct vop_access_args *ap) 1135 { 1136 struct vnode *vp = ap->a_vp; 1137 struct vattr vattr; 1138 int error; 1139 1140 error = VOP_GETATTR(vp, &vattr, ap->a_cred); 1141 if (error) 1142 return (error); 1143 error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid, 1144 vattr.va_gid, ap->a_accmode, ap->a_cred, NULL); 1145 return (error); 1146 } 1147 1148 #if 0 1149 struct vop_getattr_args { 1150 struct vop_generic_args a_gen; 1151 struct vnode *a_vp; 1152 struct vattr *a_vap; 1153 struct ucred *a_cred; 1154 }; 1155 #endif 1156 1157 /* 1158 * Get file attributes 1159 */ 1160 static int 1161 mqfs_getattr(struct vop_getattr_args *ap) 1162 { 1163 struct vnode *vp = ap->a_vp; 1164 struct mqfs_node *pn = VTON(vp); 1165 struct vattr *vap = ap->a_vap; 1166 int error = 0; 1167 1168 vap->va_type = vp->v_type; 1169 vap->va_mode = pn->mn_mode; 1170 vap->va_nlink = 1; 1171 vap->va_uid = pn->mn_uid; 1172 vap->va_gid = pn->mn_gid; 1173 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 1174 vap->va_fileid = pn->mn_fileno; 1175 vap->va_size = 0; 1176 vap->va_blocksize = PAGE_SIZE; 1177 vap->va_bytes = vap->va_size = 0; 1178 vap->va_atime = pn->mn_atime; 1179 vap->va_mtime = pn->mn_mtime; 1180 vap->va_ctime = pn->mn_ctime; 1181 vap->va_birthtime = pn->mn_birth; 1182 vap->va_gen = 0; 1183 vap->va_flags = 0; 1184 vap->va_rdev = NODEV; 1185 vap->va_bytes = 0; 1186 vap->va_filerev = 0; 1187 return (error); 1188 } 1189 1190 #if 0 1191 struct vop_setattr_args { 1192 struct vop_generic_args a_gen; 1193 struct vnode *a_vp; 1194 struct vattr *a_vap; 1195 struct ucred *a_cred; 1196 }; 1197 #endif 1198 /* 1199 * Set attributes 1200 */ 1201 static int 1202 mqfs_setattr(struct vop_setattr_args *ap) 1203 { 1204 struct mqfs_node *pn; 1205 struct vattr *vap; 1206 struct vnode *vp; 1207 struct thread *td; 1208 int c, error; 1209 uid_t uid; 1210 gid_t gid; 1211 1212 td = curthread; 1213 vap = ap->a_vap; 1214 vp = ap->a_vp; 1215 if ((vap->va_type != VNON) || 1216 (vap->va_nlink != VNOVAL) || 1217 (vap->va_fsid != VNOVAL) || 1218 (vap->va_fileid != VNOVAL) || 1219 (vap->va_blocksize != VNOVAL) || 1220 (vap->va_flags != VNOVAL && vap->va_flags != 0) || 1221 (vap->va_rdev != VNOVAL) || 1222 ((int)vap->va_bytes != VNOVAL) || 1223 (vap->va_gen != VNOVAL)) { 1224 return (EINVAL); 1225 } 1226 1227 pn = VTON(vp); 1228 1229 error = c = 0; 1230 if (vap->va_uid == (uid_t)VNOVAL) 1231 uid = pn->mn_uid; 1232 else 1233 uid = vap->va_uid; 1234 if (vap->va_gid == (gid_t)VNOVAL) 1235 gid = pn->mn_gid; 1236 else 1237 gid = vap->va_gid; 1238 1239 if (uid != pn->mn_uid || gid != pn->mn_gid) { 1240 /* 1241 * To modify the ownership of a file, must possess VADMIN 1242 * for that file. 1243 */ 1244 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td))) 1245 return (error); 1246 1247 /* 1248 * XXXRW: Why is there a privilege check here: shouldn't the 1249 * check in VOP_ACCESS() be enough? Also, are the group bits 1250 * below definitely right? 1251 */ 1252 if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid || 1253 (gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) && 1254 (error = priv_check(td, PRIV_MQ_ADMIN)) != 0) 1255 return (error); 1256 pn->mn_uid = uid; 1257 pn->mn_gid = gid; 1258 c = 1; 1259 } 1260 1261 if (vap->va_mode != (mode_t)VNOVAL) { 1262 if ((ap->a_cred->cr_uid != pn->mn_uid) && 1263 (error = priv_check(td, PRIV_MQ_ADMIN))) 1264 return (error); 1265 pn->mn_mode = vap->va_mode; 1266 c = 1; 1267 } 1268 1269 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { 1270 /* See the comment in ufs_vnops::ufs_setattr(). */ 1271 if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) && 1272 ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || 1273 (error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td)))) 1274 return (error); 1275 if (vap->va_atime.tv_sec != VNOVAL) { 1276 pn->mn_atime = vap->va_atime; 1277 } 1278 if (vap->va_mtime.tv_sec != VNOVAL) { 1279 pn->mn_mtime = vap->va_mtime; 1280 } 1281 c = 1; 1282 } 1283 if (c) { 1284 vfs_timestamp(&pn->mn_ctime); 1285 } 1286 return (0); 1287 } 1288 1289 #if 0 1290 struct vop_read_args { 1291 struct vop_generic_args a_gen; 1292 struct vnode *a_vp; 1293 struct uio *a_uio; 1294 int a_ioflag; 1295 struct ucred *a_cred; 1296 }; 1297 #endif 1298 1299 /* 1300 * Read from a file 1301 */ 1302 static int 1303 mqfs_read(struct vop_read_args *ap) 1304 { 1305 char buf[80]; 1306 struct vnode *vp = ap->a_vp; 1307 struct uio *uio = ap->a_uio; 1308 struct mqfs_node *pn; 1309 struct mqueue *mq; 1310 int len, error; 1311 1312 if (vp->v_type != VREG) 1313 return (EINVAL); 1314 1315 pn = VTON(vp); 1316 mq = VTOMQ(vp); 1317 snprintf(buf, sizeof(buf), 1318 "QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n", 1319 mq->mq_totalbytes, 1320 mq->mq_maxmsg, 1321 mq->mq_curmsgs, 1322 mq->mq_msgsize); 1323 buf[sizeof(buf)-1] = '\0'; 1324 len = strlen(buf); 1325 error = uiomove_frombuf(buf, len, uio); 1326 return (error); 1327 } 1328 1329 #if 0 1330 struct vop_readdir_args { 1331 struct vop_generic_args a_gen; 1332 struct vnode *a_vp; 1333 struct uio *a_uio; 1334 struct ucred *a_cred; 1335 int *a_eofflag; 1336 int *a_ncookies; 1337 u_long **a_cookies; 1338 }; 1339 #endif 1340 1341 /* 1342 * Return directory entries. 1343 */ 1344 static int 1345 mqfs_readdir(struct vop_readdir_args *ap) 1346 { 1347 struct vnode *vp; 1348 struct mqfs_info *mi; 1349 struct mqfs_node *pd; 1350 struct mqfs_node *pn; 1351 struct dirent entry; 1352 struct uio *uio; 1353 int *tmp_ncookies = NULL; 1354 off_t offset; 1355 int error, i; 1356 1357 vp = ap->a_vp; 1358 mi = VFSTOMQFS(vp->v_mount); 1359 pd = VTON(vp); 1360 uio = ap->a_uio; 1361 1362 if (vp->v_type != VDIR) 1363 return (ENOTDIR); 1364 1365 if (uio->uio_offset < 0) 1366 return (EINVAL); 1367 1368 if (ap->a_ncookies != NULL) { 1369 tmp_ncookies = ap->a_ncookies; 1370 *ap->a_ncookies = 0; 1371 ap->a_ncookies = NULL; 1372 } 1373 1374 error = 0; 1375 offset = 0; 1376 1377 sx_xlock(&mi->mi_lock); 1378 1379 LIST_FOREACH(pn, &pd->mn_children, mn_sibling) { 1380 entry.d_reclen = sizeof(entry); 1381 if (!pn->mn_fileno) 1382 mqfs_fileno_alloc(mi, pn); 1383 entry.d_fileno = pn->mn_fileno; 1384 for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i) 1385 entry.d_name[i] = pn->mn_name[i]; 1386 entry.d_name[i] = 0; 1387 entry.d_namlen = i; 1388 switch (pn->mn_type) { 1389 case mqfstype_root: 1390 case mqfstype_dir: 1391 case mqfstype_this: 1392 case mqfstype_parent: 1393 entry.d_type = DT_DIR; 1394 break; 1395 case mqfstype_file: 1396 entry.d_type = DT_REG; 1397 break; 1398 case mqfstype_symlink: 1399 entry.d_type = DT_LNK; 1400 break; 1401 default: 1402 panic("%s has unexpected node type: %d", pn->mn_name, 1403 pn->mn_type); 1404 } 1405 if (entry.d_reclen > uio->uio_resid) 1406 break; 1407 if (offset >= uio->uio_offset) { 1408 error = vfs_read_dirent(ap, &entry, offset); 1409 if (error) 1410 break; 1411 } 1412 offset += entry.d_reclen; 1413 } 1414 sx_xunlock(&mi->mi_lock); 1415 1416 uio->uio_offset = offset; 1417 1418 if (tmp_ncookies != NULL) 1419 ap->a_ncookies = tmp_ncookies; 1420 1421 return (error); 1422 } 1423 1424 #ifdef notyet 1425 1426 #if 0 1427 struct vop_mkdir_args { 1428 struct vnode *a_dvp; 1429 struvt vnode **a_vpp; 1430 struvt componentname *a_cnp; 1431 struct vattr *a_vap; 1432 }; 1433 #endif 1434 1435 /* 1436 * Create a directory. 1437 */ 1438 static int 1439 mqfs_mkdir(struct vop_mkdir_args *ap) 1440 { 1441 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1442 struct componentname *cnp = ap->a_cnp; 1443 struct mqfs_node *pd = VTON(ap->a_dvp); 1444 struct mqfs_node *pn; 1445 int error; 1446 1447 if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir) 1448 return (ENOTDIR); 1449 sx_xlock(&mqfs->mi_lock); 1450 if ((cnp->cn_flags & HASBUF) == 0) 1451 panic("%s: no name", __func__); 1452 pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen, 1453 ap->a_vap->cn_cred, ap->a_vap->va_mode); 1454 if (pn != NULL) 1455 mqnode_addref(pn); 1456 sx_xunlock(&mqfs->mi_lock); 1457 if (pn == NULL) { 1458 error = ENOSPC; 1459 } else { 1460 error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn); 1461 mqnode_release(pn); 1462 } 1463 return (error); 1464 } 1465 1466 #if 0 1467 struct vop_rmdir_args { 1468 struct vnode *a_dvp; 1469 struct vnode *a_vp; 1470 struct componentname *a_cnp; 1471 }; 1472 #endif 1473 1474 /* 1475 * Remove a directory. 1476 */ 1477 static int 1478 mqfs_rmdir(struct vop_rmdir_args *ap) 1479 { 1480 struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount); 1481 struct mqfs_node *pn = VTON(ap->a_vp); 1482 struct mqfs_node *pt; 1483 1484 if (pn->mn_type != mqfstype_dir) 1485 return (ENOTDIR); 1486 1487 sx_xlock(&mqfs->mi_lock); 1488 if (pn->mn_deleted) { 1489 sx_xunlock(&mqfs->mi_lock); 1490 return (ENOENT); 1491 } 1492 1493 pt = LIST_FIRST(&pn->mn_children); 1494 pt = LIST_NEXT(pt, mn_sibling); 1495 pt = LIST_NEXT(pt, mn_sibling); 1496 if (pt != NULL) { 1497 sx_xunlock(&mqfs->mi_lock); 1498 return (ENOTEMPTY); 1499 } 1500 pt = pn->mn_parent; 1501 pn->mn_parent = NULL; 1502 pn->mn_deleted = 1; 1503 LIST_REMOVE(pn, mn_sibling); 1504 mqnode_release(pn); 1505 mqnode_release(pt); 1506 sx_xunlock(&mqfs->mi_lock); 1507 cache_purge(ap->a_vp); 1508 return (0); 1509 } 1510 1511 #endif /* notyet */ 1512 1513 /* 1514 * Allocate a message queue 1515 */ 1516 static struct mqueue * 1517 mqueue_alloc(const struct mq_attr *attr) 1518 { 1519 struct mqueue *mq; 1520 1521 if (curmq >= maxmq) 1522 return (NULL); 1523 mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO); 1524 TAILQ_INIT(&mq->mq_msgq); 1525 if (attr != NULL) { 1526 mq->mq_maxmsg = attr->mq_maxmsg; 1527 mq->mq_msgsize = attr->mq_msgsize; 1528 } else { 1529 mq->mq_maxmsg = default_maxmsg; 1530 mq->mq_msgsize = default_msgsize; 1531 } 1532 mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF); 1533 knlist_init(&mq->mq_rsel.si_note, &mq->mq_mutex, NULL, NULL, NULL); 1534 knlist_init(&mq->mq_wsel.si_note, &mq->mq_mutex, NULL, NULL, NULL); 1535 atomic_add_int(&curmq, 1); 1536 return (mq); 1537 } 1538 1539 /* 1540 * Destroy a message queue 1541 */ 1542 static void 1543 mqueue_free(struct mqueue *mq) 1544 { 1545 struct mqueue_msg *msg; 1546 1547 while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) { 1548 TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link); 1549 free(msg, M_MQUEUEDATA); 1550 } 1551 1552 mtx_destroy(&mq->mq_mutex); 1553 knlist_destroy(&mq->mq_rsel.si_note); 1554 knlist_destroy(&mq->mq_wsel.si_note); 1555 uma_zfree(mqueue_zone, mq); 1556 atomic_add_int(&curmq, -1); 1557 } 1558 1559 /* 1560 * Load a message from user space 1561 */ 1562 static struct mqueue_msg * 1563 mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio) 1564 { 1565 struct mqueue_msg *msg; 1566 size_t len; 1567 int error; 1568 1569 len = sizeof(struct mqueue_msg) + msg_size; 1570 msg = malloc(len, M_MQUEUEDATA, M_WAITOK); 1571 error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg), 1572 msg_size); 1573 if (error) { 1574 free(msg, M_MQUEUEDATA); 1575 msg = NULL; 1576 } else { 1577 msg->msg_size = msg_size; 1578 msg->msg_prio = msg_prio; 1579 } 1580 return (msg); 1581 } 1582 1583 /* 1584 * Save a message to user space 1585 */ 1586 static int 1587 mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio) 1588 { 1589 int error; 1590 1591 error = copyout(((char *)msg) + sizeof(*msg), msg_ptr, 1592 msg->msg_size); 1593 if (error == 0 && msg_prio != NULL) 1594 error = copyout(&msg->msg_prio, msg_prio, sizeof(int)); 1595 return (error); 1596 } 1597 1598 /* 1599 * Free a message's memory 1600 */ 1601 static __inline void 1602 mqueue_freemsg(struct mqueue_msg *msg) 1603 { 1604 free(msg, M_MQUEUEDATA); 1605 } 1606 1607 /* 1608 * Send a message. if waitok is false, thread will not be 1609 * blocked if there is no data in queue, otherwise, absolute 1610 * time will be checked. 1611 */ 1612 int 1613 mqueue_send(struct mqueue *mq, const char *msg_ptr, 1614 size_t msg_len, unsigned msg_prio, int waitok, 1615 const struct timespec *abs_timeout) 1616 { 1617 struct mqueue_msg *msg; 1618 struct timespec ets, ts, ts2; 1619 struct timeval tv; 1620 int error; 1621 1622 if (msg_prio >= MQ_PRIO_MAX) 1623 return (EINVAL); 1624 if (msg_len > mq->mq_msgsize) 1625 return (EMSGSIZE); 1626 msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio); 1627 if (msg == NULL) 1628 return (EFAULT); 1629 1630 /* O_NONBLOCK case */ 1631 if (!waitok) { 1632 error = _mqueue_send(mq, msg, -1); 1633 if (error) 1634 goto bad; 1635 return (0); 1636 } 1637 1638 /* we allow a null timeout (wait forever) */ 1639 if (abs_timeout == NULL) { 1640 error = _mqueue_send(mq, msg, 0); 1641 if (error) 1642 goto bad; 1643 return (0); 1644 } 1645 1646 /* send it before checking time */ 1647 error = _mqueue_send(mq, msg, -1); 1648 if (error == 0) 1649 return (0); 1650 1651 if (error != EAGAIN) 1652 goto bad; 1653 1654 error = copyin(abs_timeout, &ets, sizeof(ets)); 1655 if (error != 0) 1656 goto bad; 1657 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) { 1658 error = EINVAL; 1659 goto bad; 1660 } 1661 for (;;) { 1662 ts2 = ets; 1663 getnanotime(&ts); 1664 timespecsub(&ts2, &ts); 1665 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { 1666 error = ETIMEDOUT; 1667 break; 1668 } 1669 TIMESPEC_TO_TIMEVAL(&tv, &ts2); 1670 error = _mqueue_send(mq, msg, tvtohz(&tv)); 1671 if (error != ETIMEDOUT) 1672 break; 1673 } 1674 if (error == 0) 1675 return (0); 1676 bad: 1677 mqueue_freemsg(msg); 1678 return (error); 1679 } 1680 1681 /* 1682 * Common routine to send a message 1683 */ 1684 static int 1685 _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo) 1686 { 1687 struct mqueue_msg *msg2; 1688 int error = 0; 1689 1690 mtx_lock(&mq->mq_mutex); 1691 while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) { 1692 if (timo < 0) { 1693 mtx_unlock(&mq->mq_mutex); 1694 return (EAGAIN); 1695 } 1696 mq->mq_senders++; 1697 error = msleep(&mq->mq_senders, &mq->mq_mutex, 1698 PCATCH, "mqsend", timo); 1699 mq->mq_senders--; 1700 if (error == EAGAIN) 1701 error = ETIMEDOUT; 1702 } 1703 if (mq->mq_curmsgs >= mq->mq_maxmsg) { 1704 mtx_unlock(&mq->mq_mutex); 1705 return (error); 1706 } 1707 error = 0; 1708 if (TAILQ_EMPTY(&mq->mq_msgq)) { 1709 TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link); 1710 } else { 1711 if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) { 1712 TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link); 1713 } else { 1714 TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) { 1715 if (msg2->msg_prio < msg->msg_prio) 1716 break; 1717 } 1718 TAILQ_INSERT_BEFORE(msg2, msg, msg_link); 1719 } 1720 } 1721 mq->mq_curmsgs++; 1722 mq->mq_totalbytes += msg->msg_size; 1723 if (mq->mq_receivers) 1724 wakeup_one(&mq->mq_receivers); 1725 else if (mq->mq_notifier != NULL) 1726 mqueue_send_notification(mq); 1727 if (mq->mq_flags & MQ_RSEL) { 1728 mq->mq_flags &= ~MQ_RSEL; 1729 selwakeup(&mq->mq_rsel); 1730 } 1731 KNOTE_LOCKED(&mq->mq_rsel.si_note, 0); 1732 mtx_unlock(&mq->mq_mutex); 1733 return (0); 1734 } 1735 1736 /* 1737 * Send realtime a signal to process which registered itself 1738 * successfully by mq_notify. 1739 */ 1740 static void 1741 mqueue_send_notification(struct mqueue *mq) 1742 { 1743 struct mqueue_notifier *nt; 1744 struct proc *p; 1745 1746 mtx_assert(&mq->mq_mutex, MA_OWNED); 1747 nt = mq->mq_notifier; 1748 if (nt->nt_sigev.sigev_notify != SIGEV_NONE) { 1749 p = nt->nt_proc; 1750 PROC_LOCK(p); 1751 if (!KSI_ONQ(&nt->nt_ksi)) 1752 psignal_event(p, &nt->nt_sigev, &nt->nt_ksi); 1753 PROC_UNLOCK(p); 1754 } 1755 mq->mq_notifier = NULL; 1756 } 1757 1758 /* 1759 * Get a message. if waitok is false, thread will not be 1760 * blocked if there is no data in queue, otherwise, absolute 1761 * time will be checked. 1762 */ 1763 int 1764 mqueue_receive(struct mqueue *mq, char *msg_ptr, 1765 size_t msg_len, unsigned *msg_prio, int waitok, 1766 const struct timespec *abs_timeout) 1767 { 1768 struct mqueue_msg *msg; 1769 struct timespec ets, ts, ts2; 1770 struct timeval tv; 1771 int error; 1772 1773 if (msg_len < mq->mq_msgsize) 1774 return (EMSGSIZE); 1775 1776 /* O_NONBLOCK case */ 1777 if (!waitok) { 1778 error = _mqueue_recv(mq, &msg, -1); 1779 if (error) 1780 return (error); 1781 goto received; 1782 } 1783 1784 /* we allow a null timeout (wait forever). */ 1785 if (abs_timeout == NULL) { 1786 error = _mqueue_recv(mq, &msg, 0); 1787 if (error) 1788 return (error); 1789 goto received; 1790 } 1791 1792 /* try to get a message before checking time */ 1793 error = _mqueue_recv(mq, &msg, -1); 1794 if (error == 0) 1795 goto received; 1796 1797 if (error != EAGAIN) 1798 return (error); 1799 1800 error = copyin(abs_timeout, &ets, sizeof(ets)); 1801 if (error != 0) 1802 return (error); 1803 if (ets.tv_nsec >= 1000000000 || ets.tv_nsec < 0) { 1804 error = EINVAL; 1805 return (error); 1806 } 1807 1808 for (;;) { 1809 ts2 = ets; 1810 getnanotime(&ts); 1811 timespecsub(&ts2, &ts); 1812 if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) { 1813 error = ETIMEDOUT; 1814 return (error); 1815 } 1816 TIMESPEC_TO_TIMEVAL(&tv, &ts2); 1817 error = _mqueue_recv(mq, &msg, tvtohz(&tv)); 1818 if (error == 0) 1819 break; 1820 if (error != ETIMEDOUT) 1821 return (error); 1822 } 1823 1824 received: 1825 error = mqueue_savemsg(msg, msg_ptr, msg_prio); 1826 if (error == 0) { 1827 curthread->td_retval[0] = msg->msg_size; 1828 curthread->td_retval[1] = 0; 1829 } 1830 mqueue_freemsg(msg); 1831 return (error); 1832 } 1833 1834 /* 1835 * Common routine to receive a message 1836 */ 1837 static int 1838 _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo) 1839 { 1840 int error = 0; 1841 1842 mtx_lock(&mq->mq_mutex); 1843 while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) { 1844 if (timo < 0) { 1845 mtx_unlock(&mq->mq_mutex); 1846 return (EAGAIN); 1847 } 1848 mq->mq_receivers++; 1849 error = msleep(&mq->mq_receivers, &mq->mq_mutex, 1850 PCATCH, "mqrecv", timo); 1851 mq->mq_receivers--; 1852 if (error == EAGAIN) 1853 error = ETIMEDOUT; 1854 } 1855 if (*msg != NULL) { 1856 error = 0; 1857 TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link); 1858 mq->mq_curmsgs--; 1859 mq->mq_totalbytes -= (*msg)->msg_size; 1860 if (mq->mq_senders) 1861 wakeup_one(&mq->mq_senders); 1862 if (mq->mq_flags & MQ_WSEL) { 1863 mq->mq_flags &= ~MQ_WSEL; 1864 selwakeup(&mq->mq_wsel); 1865 } 1866 KNOTE_LOCKED(&mq->mq_wsel.si_note, 0); 1867 } 1868 if (mq->mq_notifier != NULL && mq->mq_receivers == 0 && 1869 !TAILQ_EMPTY(&mq->mq_msgq)) { 1870 mqueue_send_notification(mq); 1871 } 1872 mtx_unlock(&mq->mq_mutex); 1873 return (error); 1874 } 1875 1876 static __inline struct mqueue_notifier * 1877 notifier_alloc(void) 1878 { 1879 return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO)); 1880 } 1881 1882 static __inline void 1883 notifier_free(struct mqueue_notifier *p) 1884 { 1885 uma_zfree(mqnoti_zone, p); 1886 } 1887 1888 static struct mqueue_notifier * 1889 notifier_search(struct proc *p, int fd) 1890 { 1891 struct mqueue_notifier *nt; 1892 1893 LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) { 1894 if (nt->nt_ksi.ksi_mqd == fd) 1895 break; 1896 } 1897 return (nt); 1898 } 1899 1900 static __inline void 1901 notifier_insert(struct proc *p, struct mqueue_notifier *nt) 1902 { 1903 LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link); 1904 } 1905 1906 static __inline void 1907 notifier_delete(struct proc *p, struct mqueue_notifier *nt) 1908 { 1909 LIST_REMOVE(nt, nt_link); 1910 notifier_free(nt); 1911 } 1912 1913 static void 1914 notifier_remove(struct proc *p, struct mqueue *mq, int fd) 1915 { 1916 struct mqueue_notifier *nt; 1917 1918 mtx_assert(&mq->mq_mutex, MA_OWNED); 1919 PROC_LOCK(p); 1920 nt = notifier_search(p, fd); 1921 if (nt != NULL) { 1922 if (mq->mq_notifier == nt) 1923 mq->mq_notifier = NULL; 1924 sigqueue_take(&nt->nt_ksi); 1925 notifier_delete(p, nt); 1926 } 1927 PROC_UNLOCK(p); 1928 } 1929 1930 /* 1931 * Syscall to open a message queue. 1932 */ 1933 int 1934 kmq_open(struct thread *td, struct kmq_open_args *uap) 1935 { 1936 char path[MQFS_NAMELEN + 1]; 1937 struct mq_attr attr, *pattr; 1938 struct mqfs_node *pn; 1939 struct filedesc *fdp; 1940 struct file *fp; 1941 struct mqueue *mq; 1942 int fd, error, len, flags, cmode; 1943 1944 if ((uap->flags & O_ACCMODE) == O_ACCMODE) 1945 return (EINVAL); 1946 1947 fdp = td->td_proc->p_fd; 1948 flags = FFLAGS(uap->flags); 1949 cmode = (((uap->mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT); 1950 mq = NULL; 1951 if ((flags & O_CREAT) && (uap->attr != NULL)) { 1952 error = copyin(uap->attr, &attr, sizeof(attr)); 1953 if (error) 1954 return (error); 1955 if (attr.mq_maxmsg <= 0 || attr.mq_maxmsg > maxmsg) 1956 return (EINVAL); 1957 if (attr.mq_msgsize <= 0 || attr.mq_msgsize > maxmsgsize) 1958 return (EINVAL); 1959 pattr = &attr; 1960 } else 1961 pattr = NULL; 1962 1963 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL); 1964 if (error) 1965 return (error); 1966 1967 /* 1968 * The first character of name must be a slash (/) character 1969 * and the remaining characters of name cannot include any slash 1970 * characters. 1971 */ 1972 len = strlen(path); 1973 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL) 1974 return (EINVAL); 1975 1976 error = falloc(td, &fp, &fd); 1977 if (error) 1978 return (error); 1979 1980 sx_xlock(&mqfs_data.mi_lock); 1981 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1); 1982 if (pn == NULL) { 1983 if (!(flags & O_CREAT)) { 1984 error = ENOENT; 1985 } else { 1986 mq = mqueue_alloc(pattr); 1987 if (mq == NULL) { 1988 error = ENFILE; 1989 } else { 1990 pn = mqfs_create_file(mqfs_data.mi_root, 1991 path + 1, len - 1, td->td_ucred, 1992 cmode); 1993 if (pn == NULL) { 1994 error = ENOSPC; 1995 mqueue_free(mq); 1996 } 1997 } 1998 } 1999 2000 if (error == 0) { 2001 pn->mn_data = mq; 2002 } 2003 } else { 2004 if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) { 2005 error = EEXIST; 2006 } else { 2007 accmode_t accmode = 0; 2008 2009 if (flags & FREAD) 2010 accmode |= VREAD; 2011 if (flags & FWRITE) 2012 accmode |= VWRITE; 2013 error = vaccess(VREG, pn->mn_mode, pn->mn_uid, 2014 pn->mn_gid, accmode, td->td_ucred, NULL); 2015 } 2016 } 2017 2018 if (error) { 2019 sx_xunlock(&mqfs_data.mi_lock); 2020 fdclose(fdp, fp, fd, td); 2021 fdrop(fp, td); 2022 return (error); 2023 } 2024 2025 mqnode_addref(pn); 2026 sx_xunlock(&mqfs_data.mi_lock); 2027 2028 finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn, 2029 &mqueueops); 2030 2031 FILEDESC_XLOCK(fdp); 2032 if (fdp->fd_ofiles[fd] == fp) 2033 fdp->fd_ofileflags[fd] |= UF_EXCLOSE; 2034 FILEDESC_XUNLOCK(fdp); 2035 td->td_retval[0] = fd; 2036 fdrop(fp, td); 2037 return (0); 2038 } 2039 2040 /* 2041 * Syscall to unlink a message queue. 2042 */ 2043 int 2044 kmq_unlink(struct thread *td, struct kmq_unlink_args *uap) 2045 { 2046 char path[MQFS_NAMELEN+1]; 2047 struct mqfs_node *pn; 2048 int error, len; 2049 2050 error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL); 2051 if (error) 2052 return (error); 2053 2054 len = strlen(path); 2055 if (len < 2 || path[0] != '/' || index(path + 1, '/') != NULL) 2056 return (EINVAL); 2057 2058 sx_xlock(&mqfs_data.mi_lock); 2059 pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1); 2060 if (pn != NULL) 2061 error = do_unlink(pn, td->td_ucred); 2062 else 2063 error = ENOENT; 2064 sx_xunlock(&mqfs_data.mi_lock); 2065 return (error); 2066 } 2067 2068 typedef int (*_fgetf)(struct thread *, int, struct file **); 2069 2070 /* 2071 * Get message queue by giving file slot 2072 */ 2073 static int 2074 _getmq(struct thread *td, int fd, _fgetf func, 2075 struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) 2076 { 2077 struct mqfs_node *pn; 2078 int error; 2079 2080 error = func(td, fd, fpp); 2081 if (error) 2082 return (error); 2083 if (&mqueueops != (*fpp)->f_ops) { 2084 fdrop(*fpp, td); 2085 return (EBADF); 2086 } 2087 pn = (*fpp)->f_data; 2088 if (ppn) 2089 *ppn = pn; 2090 if (pmq) 2091 *pmq = pn->mn_data; 2092 return (0); 2093 } 2094 2095 static __inline int 2096 getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn, 2097 struct mqueue **pmq) 2098 { 2099 return _getmq(td, fd, fget, fpp, ppn, pmq); 2100 } 2101 2102 static __inline int 2103 getmq_read(struct thread *td, int fd, struct file **fpp, 2104 struct mqfs_node **ppn, struct mqueue **pmq) 2105 { 2106 return _getmq(td, fd, fget_read, fpp, ppn, pmq); 2107 } 2108 2109 static __inline int 2110 getmq_write(struct thread *td, int fd, struct file **fpp, 2111 struct mqfs_node **ppn, struct mqueue **pmq) 2112 { 2113 return _getmq(td, fd, fget_write, fpp, ppn, pmq); 2114 } 2115 2116 int 2117 kmq_setattr(struct thread *td, struct kmq_setattr_args *uap) 2118 { 2119 struct mqueue *mq; 2120 struct file *fp; 2121 struct mq_attr attr, oattr; 2122 u_int oflag, flag; 2123 int error; 2124 2125 if (uap->attr) { 2126 error = copyin(uap->attr, &attr, sizeof(attr)); 2127 if (error) 2128 return (error); 2129 if (attr.mq_flags & ~O_NONBLOCK) 2130 return (EINVAL); 2131 } 2132 error = getmq(td, uap->mqd, &fp, NULL, &mq); 2133 if (error) 2134 return (error); 2135 oattr.mq_maxmsg = mq->mq_maxmsg; 2136 oattr.mq_msgsize = mq->mq_msgsize; 2137 oattr.mq_curmsgs = mq->mq_curmsgs; 2138 if (uap->attr) { 2139 do { 2140 oflag = flag = fp->f_flag; 2141 flag &= ~O_NONBLOCK; 2142 flag |= (attr.mq_flags & O_NONBLOCK); 2143 } while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0); 2144 } else 2145 oflag = fp->f_flag; 2146 oattr.mq_flags = (O_NONBLOCK & oflag); 2147 fdrop(fp, td); 2148 if (uap->oattr) 2149 error = copyout(&oattr, uap->oattr, sizeof(oattr)); 2150 return (error); 2151 } 2152 2153 int 2154 kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap) 2155 { 2156 struct mqueue *mq; 2157 struct file *fp; 2158 int error; 2159 int waitok; 2160 2161 error = getmq_read(td, uap->mqd, &fp, NULL, &mq); 2162 if (error) 2163 return (error); 2164 waitok = !(fp->f_flag & O_NONBLOCK); 2165 error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len, 2166 uap->msg_prio, waitok, uap->abs_timeout); 2167 fdrop(fp, td); 2168 return (error); 2169 } 2170 2171 int 2172 kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap) 2173 { 2174 struct mqueue *mq; 2175 struct file *fp; 2176 int error, waitok; 2177 2178 error = getmq_write(td, uap->mqd, &fp, NULL, &mq); 2179 if (error) 2180 return (error); 2181 waitok = !(fp->f_flag & O_NONBLOCK); 2182 error = mqueue_send(mq, uap->msg_ptr, uap->msg_len, 2183 uap->msg_prio, waitok, uap->abs_timeout); 2184 fdrop(fp, td); 2185 return (error); 2186 } 2187 2188 int 2189 kmq_notify(struct thread *td, struct kmq_notify_args *uap) 2190 { 2191 struct sigevent ev; 2192 struct filedesc *fdp; 2193 struct proc *p; 2194 struct mqueue *mq; 2195 struct file *fp; 2196 struct mqueue_notifier *nt, *newnt = NULL; 2197 int error; 2198 2199 p = td->td_proc; 2200 fdp = td->td_proc->p_fd; 2201 if (uap->sigev) { 2202 error = copyin(uap->sigev, &ev, sizeof(ev)); 2203 if (error) 2204 return (error); 2205 if (ev.sigev_notify != SIGEV_SIGNAL && 2206 ev.sigev_notify != SIGEV_THREAD_ID && 2207 ev.sigev_notify != SIGEV_NONE) 2208 return (EINVAL); 2209 if ((ev.sigev_notify == SIGEV_SIGNAL || 2210 ev.sigev_notify == SIGEV_THREAD_ID) && 2211 !_SIG_VALID(ev.sigev_signo)) 2212 return (EINVAL); 2213 } 2214 error = getmq(td, uap->mqd, &fp, NULL, &mq); 2215 if (error) 2216 return (error); 2217 again: 2218 FILEDESC_SLOCK(fdp); 2219 if (fget_locked(fdp, uap->mqd) != fp) { 2220 FILEDESC_SUNLOCK(fdp); 2221 error = EBADF; 2222 goto out; 2223 } 2224 mtx_lock(&mq->mq_mutex); 2225 FILEDESC_SUNLOCK(fdp); 2226 if (uap->sigev != NULL) { 2227 if (mq->mq_notifier != NULL) { 2228 error = EBUSY; 2229 } else { 2230 PROC_LOCK(p); 2231 nt = notifier_search(p, uap->mqd); 2232 if (nt == NULL) { 2233 if (newnt == NULL) { 2234 PROC_UNLOCK(p); 2235 mtx_unlock(&mq->mq_mutex); 2236 newnt = notifier_alloc(); 2237 goto again; 2238 } 2239 } 2240 2241 if (nt != NULL) { 2242 sigqueue_take(&nt->nt_ksi); 2243 if (newnt != NULL) { 2244 notifier_free(newnt); 2245 newnt = NULL; 2246 } 2247 } else { 2248 nt = newnt; 2249 newnt = NULL; 2250 ksiginfo_init(&nt->nt_ksi); 2251 nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT; 2252 nt->nt_ksi.ksi_code = SI_MESGQ; 2253 nt->nt_proc = p; 2254 nt->nt_ksi.ksi_mqd = uap->mqd; 2255 notifier_insert(p, nt); 2256 } 2257 nt->nt_sigev = ev; 2258 mq->mq_notifier = nt; 2259 PROC_UNLOCK(p); 2260 /* 2261 * if there is no receivers and message queue 2262 * is not empty, we should send notification 2263 * as soon as possible. 2264 */ 2265 if (mq->mq_receivers == 0 && 2266 !TAILQ_EMPTY(&mq->mq_msgq)) 2267 mqueue_send_notification(mq); 2268 } 2269 } else { 2270 notifier_remove(p, mq, uap->mqd); 2271 } 2272 mtx_unlock(&mq->mq_mutex); 2273 2274 out: 2275 fdrop(fp, td); 2276 if (newnt != NULL) 2277 notifier_free(newnt); 2278 return (error); 2279 } 2280 2281 static void 2282 mqueue_fdclose(struct thread *td, int fd, struct file *fp) 2283 { 2284 struct filedesc *fdp; 2285 struct mqueue *mq; 2286 2287 fdp = td->td_proc->p_fd; 2288 FILEDESC_LOCK_ASSERT(fdp); 2289 2290 if (fp->f_ops == &mqueueops) { 2291 mq = FPTOMQ(fp); 2292 mtx_lock(&mq->mq_mutex); 2293 notifier_remove(td->td_proc, mq, fd); 2294 2295 /* have to wakeup thread in same process */ 2296 if (mq->mq_flags & MQ_RSEL) { 2297 mq->mq_flags &= ~MQ_RSEL; 2298 selwakeup(&mq->mq_rsel); 2299 } 2300 if (mq->mq_flags & MQ_WSEL) { 2301 mq->mq_flags &= ~MQ_WSEL; 2302 selwakeup(&mq->mq_wsel); 2303 } 2304 mtx_unlock(&mq->mq_mutex); 2305 } 2306 } 2307 2308 static void 2309 mq_proc_exit(void *arg __unused, struct proc *p) 2310 { 2311 struct filedesc *fdp; 2312 struct file *fp; 2313 struct mqueue *mq; 2314 int i; 2315 2316 fdp = p->p_fd; 2317 FILEDESC_SLOCK(fdp); 2318 for (i = 0; i < fdp->fd_nfiles; ++i) { 2319 fp = fget_locked(fdp, i); 2320 if (fp != NULL && fp->f_ops == &mqueueops) { 2321 mq = FPTOMQ(fp); 2322 mtx_lock(&mq->mq_mutex); 2323 notifier_remove(p, FPTOMQ(fp), i); 2324 mtx_unlock(&mq->mq_mutex); 2325 } 2326 } 2327 FILEDESC_SUNLOCK(fdp); 2328 KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left")); 2329 } 2330 2331 static int 2332 mqf_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 2333 int flags, struct thread *td) 2334 { 2335 return (EOPNOTSUPP); 2336 } 2337 2338 static int 2339 mqf_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 2340 int flags, struct thread *td) 2341 { 2342 return (EOPNOTSUPP); 2343 } 2344 2345 static int 2346 mqf_truncate(struct file *fp, off_t length, struct ucred *active_cred, 2347 struct thread *td) 2348 { 2349 2350 return (EINVAL); 2351 } 2352 2353 static int 2354 mqf_ioctl(struct file *fp, u_long cmd, void *data, 2355 struct ucred *active_cred, struct thread *td) 2356 { 2357 return (ENOTTY); 2358 } 2359 2360 static int 2361 mqf_poll(struct file *fp, int events, struct ucred *active_cred, 2362 struct thread *td) 2363 { 2364 struct mqueue *mq = FPTOMQ(fp); 2365 int revents = 0; 2366 2367 mtx_lock(&mq->mq_mutex); 2368 if (events & (POLLIN | POLLRDNORM)) { 2369 if (mq->mq_curmsgs) { 2370 revents |= events & (POLLIN | POLLRDNORM); 2371 } else { 2372 mq->mq_flags |= MQ_RSEL; 2373 selrecord(td, &mq->mq_rsel); 2374 } 2375 } 2376 if (events & POLLOUT) { 2377 if (mq->mq_curmsgs < mq->mq_maxmsg) 2378 revents |= POLLOUT; 2379 else { 2380 mq->mq_flags |= MQ_WSEL; 2381 selrecord(td, &mq->mq_wsel); 2382 } 2383 } 2384 mtx_unlock(&mq->mq_mutex); 2385 return (revents); 2386 } 2387 2388 static int 2389 mqf_close(struct file *fp, struct thread *td) 2390 { 2391 struct mqfs_node *pn; 2392 2393 fp->f_ops = &badfileops; 2394 pn = fp->f_data; 2395 fp->f_data = NULL; 2396 sx_xlock(&mqfs_data.mi_lock); 2397 mqnode_release(pn); 2398 sx_xunlock(&mqfs_data.mi_lock); 2399 return (0); 2400 } 2401 2402 static int 2403 mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred, 2404 struct thread *td) 2405 { 2406 struct mqfs_node *pn = fp->f_data; 2407 2408 bzero(st, sizeof *st); 2409 st->st_atimespec = pn->mn_atime; 2410 st->st_mtimespec = pn->mn_mtime; 2411 st->st_ctimespec = pn->mn_ctime; 2412 st->st_birthtimespec = pn->mn_birth; 2413 st->st_uid = pn->mn_uid; 2414 st->st_gid = pn->mn_gid; 2415 st->st_mode = S_IFIFO | pn->mn_mode; 2416 return (0); 2417 } 2418 2419 static int 2420 mqf_kqfilter(struct file *fp, struct knote *kn) 2421 { 2422 struct mqueue *mq = FPTOMQ(fp); 2423 int error = 0; 2424 2425 if (kn->kn_filter == EVFILT_READ) { 2426 kn->kn_fop = &mq_rfiltops; 2427 knlist_add(&mq->mq_rsel.si_note, kn, 0); 2428 } else if (kn->kn_filter == EVFILT_WRITE) { 2429 kn->kn_fop = &mq_wfiltops; 2430 knlist_add(&mq->mq_wsel.si_note, kn, 0); 2431 } else 2432 error = EINVAL; 2433 return (error); 2434 } 2435 2436 static void 2437 filt_mqdetach(struct knote *kn) 2438 { 2439 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2440 2441 if (kn->kn_filter == EVFILT_READ) 2442 knlist_remove(&mq->mq_rsel.si_note, kn, 0); 2443 else if (kn->kn_filter == EVFILT_WRITE) 2444 knlist_remove(&mq->mq_wsel.si_note, kn, 0); 2445 else 2446 panic("filt_mqdetach"); 2447 } 2448 2449 static int 2450 filt_mqread(struct knote *kn, long hint) 2451 { 2452 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2453 2454 mtx_assert(&mq->mq_mutex, MA_OWNED); 2455 return (mq->mq_curmsgs != 0); 2456 } 2457 2458 static int 2459 filt_mqwrite(struct knote *kn, long hint) 2460 { 2461 struct mqueue *mq = FPTOMQ(kn->kn_fp); 2462 2463 mtx_assert(&mq->mq_mutex, MA_OWNED); 2464 return (mq->mq_curmsgs < mq->mq_maxmsg); 2465 } 2466 2467 static struct fileops mqueueops = { 2468 .fo_read = mqf_read, 2469 .fo_write = mqf_write, 2470 .fo_truncate = mqf_truncate, 2471 .fo_ioctl = mqf_ioctl, 2472 .fo_poll = mqf_poll, 2473 .fo_kqfilter = mqf_kqfilter, 2474 .fo_stat = mqf_stat, 2475 .fo_close = mqf_close 2476 }; 2477 2478 static struct vop_vector mqfs_vnodeops = { 2479 .vop_default = &default_vnodeops, 2480 .vop_access = mqfs_access, 2481 .vop_cachedlookup = mqfs_lookup, 2482 .vop_lookup = vfs_cache_lookup, 2483 .vop_reclaim = mqfs_reclaim, 2484 .vop_create = mqfs_create, 2485 .vop_remove = mqfs_remove, 2486 .vop_inactive = mqfs_inactive, 2487 .vop_open = mqfs_open, 2488 .vop_close = mqfs_close, 2489 .vop_getattr = mqfs_getattr, 2490 .vop_setattr = mqfs_setattr, 2491 .vop_read = mqfs_read, 2492 .vop_write = VOP_EOPNOTSUPP, 2493 .vop_readdir = mqfs_readdir, 2494 .vop_mkdir = VOP_EOPNOTSUPP, 2495 .vop_rmdir = VOP_EOPNOTSUPP 2496 }; 2497 2498 static struct vfsops mqfs_vfsops = { 2499 .vfs_init = mqfs_init, 2500 .vfs_uninit = mqfs_uninit, 2501 .vfs_mount = mqfs_mount, 2502 .vfs_unmount = mqfs_unmount, 2503 .vfs_root = mqfs_root, 2504 .vfs_statfs = mqfs_statfs, 2505 }; 2506 2507 SYSCALL_MODULE_HELPER(kmq_open); 2508 SYSCALL_MODULE_HELPER(kmq_setattr); 2509 SYSCALL_MODULE_HELPER(kmq_timedsend); 2510 SYSCALL_MODULE_HELPER(kmq_timedreceive); 2511 SYSCALL_MODULE_HELPER(kmq_notify); 2512 SYSCALL_MODULE_HELPER(kmq_unlink); 2513 2514 VFS_SET(mqfs_vfsops, mqueuefs, VFCF_SYNTHETIC); 2515 MODULE_VERSION(mqueuefs, 1); 2516