1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Efficient memory file system supporting functions. 35 */ 36 37 #include <sys/kernel.h> 38 #include <sys/param.h> 39 #include <sys/namei.h> 40 #include <sys/priv.h> 41 #include <sys/proc.h> 42 #include <sys/spinlock2.h> 43 #include <sys/stat.h> 44 #include <sys/systm.h> 45 #include <sys/vnode.h> 46 #include <sys/vmmeter.h> 47 48 #include <sys/mplock2.h> 49 50 #include <vm/vm.h> 51 #include <vm/vm_object.h> 52 #include <vm/vm_page.h> 53 #include <vm/vm_pager.h> 54 #include <vm/vm_extern.h> 55 56 #include <vfs/tmpfs/tmpfs.h> 57 #include <vfs/tmpfs/tmpfs_fifoops.h> 58 #include <vfs/tmpfs/tmpfs_vnops.h> 59 60 static ino_t t_ino = 2; 61 static struct spinlock ino_lock; 62 static ino_t tmpfs_fetch_ino(void); 63 64 /* --------------------------------------------------------------------- */ 65 66 /* 67 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 68 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 69 * using the credentials of the process 'p'. 70 * 71 * If the node type is set to 'VDIR', then the parent parameter must point 72 * to the parent directory of the node being created. It may only be NULL 73 * while allocating the root node. 74 * 75 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 76 * specifies the device the node represents. 77 * 78 * If the node type is set to 'VLNK', then the parameter target specifies 79 * the file name of the target file for the symbolic link that is being 80 * created. 81 * 82 * Note that new nodes are retrieved from the available list if it has 83 * items or, if it is empty, from the node pool as long as there is enough 84 * space to create them. 85 * 86 * Returns zero on success or an appropriate error code on failure. 87 */ 88 int 89 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 90 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 91 char *target, int rmajor, int rminor, struct tmpfs_node **node) 92 { 93 struct tmpfs_node *nnode; 94 struct timespec ts; 95 udev_t rdev; 96 97 /* If the root directory of the 'tmp' file system is not yet 98 * allocated, this must be the request to do it. */ 99 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 100 101 KKASSERT(IFF(type == VLNK, target != NULL)); 102 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL)); 103 104 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 105 return (ENOSPC); 106 107 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK); 108 if (nnode == NULL) 109 return (ENOSPC); 110 111 /* Generic initialization. */ 112 nnode->tn_type = type; 113 vfs_timestamp(&ts); 114 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime 115 = ts.tv_sec; 116 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec 117 = ts.tv_nsec; 118 nnode->tn_uid = uid; 119 nnode->tn_gid = gid; 120 nnode->tn_mode = mode; 121 nnode->tn_id = tmpfs_fetch_ino(); 122 nnode->tn_advlock.init_done = 0; 123 124 /* Type-specific initialization. */ 125 switch (nnode->tn_type) { 126 case VBLK: 127 case VCHR: 128 rdev = makeudev(rmajor, rminor); 129 if (rdev == NOUDEV) { 130 objcache_put(tmp->tm_node_pool, nnode); 131 return(EINVAL); 132 } 133 nnode->tn_rdev = rdev; 134 break; 135 136 case VDIR: 137 TAILQ_INIT(&nnode->tn_dir.tn_dirhead); 138 KKASSERT(parent != nnode); 139 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 140 nnode->tn_dir.tn_parent = parent; 141 nnode->tn_dir.tn_readdir_lastn = 0; 142 nnode->tn_dir.tn_readdir_lastp = NULL; 143 nnode->tn_links++; 144 nnode->tn_size = 0; 145 if (parent) { 146 TMPFS_NODE_LOCK(parent); 147 parent->tn_links++; 148 TMPFS_NODE_UNLOCK(parent); 149 } 150 break; 151 152 case VFIFO: 153 /* FALLTHROUGH */ 154 case VSOCK: 155 break; 156 157 case VLNK: 158 nnode->tn_size = strlen(target); 159 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone, 160 M_WAITOK | M_NULLOK); 161 if (nnode->tn_link == NULL) { 162 objcache_put(tmp->tm_node_pool, nnode); 163 return (ENOSPC); 164 } 165 bcopy(target, nnode->tn_link, nnode->tn_size); 166 nnode->tn_link[nnode->tn_size] = '\0'; 167 break; 168 169 case VREG: 170 nnode->tn_reg.tn_aobj = 171 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0); 172 nnode->tn_reg.tn_aobj_pages = 0; 173 nnode->tn_size = 0; 174 break; 175 176 default: 177 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 178 } 179 180 TMPFS_NODE_LOCK(nnode); 181 TMPFS_LOCK(tmp); 182 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 183 tmp->tm_nodes_inuse++; 184 TMPFS_UNLOCK(tmp); 185 TMPFS_NODE_UNLOCK(nnode); 186 187 *node = nnode; 188 return 0; 189 } 190 191 /* --------------------------------------------------------------------- */ 192 193 /* 194 * Destroys the node pointed to by node from the file system 'tmp'. 195 * If the node does not belong to the given mount point, the results are 196 * unpredicted. 197 * 198 * If the node references a directory; no entries are allowed because 199 * their removal could need a recursive algorithm, something forbidden in 200 * kernel space. Furthermore, there is not need to provide such 201 * functionality (recursive removal) because the only primitives offered 202 * to the user are the removal of empty directories and the deletion of 203 * individual files. 204 * 205 * Note that nodes are not really deleted; in fact, when a node has been 206 * allocated, it cannot be deleted during the whole life of the file 207 * system. Instead, they are moved to the available list and remain there 208 * until reused. 209 */ 210 void 211 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 212 { 213 vm_pindex_t pages = 0; 214 215 #ifdef INVARIANTS 216 TMPFS_ASSERT_ELOCKED(node); 217 KKASSERT(node->tn_vnode == NULL); 218 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 219 #endif 220 221 TMPFS_LOCK(tmp); 222 LIST_REMOVE(node, tn_entries); 223 tmp->tm_nodes_inuse--; 224 TMPFS_UNLOCK(tmp); 225 TMPFS_NODE_UNLOCK(node); 226 227 switch (node->tn_type) { 228 case VNON: 229 /* Do not do anything. VNON is provided to let the 230 * allocation routine clean itself easily by avoiding 231 * duplicating code in it. */ 232 /* FALLTHROUGH */ 233 case VBLK: 234 /* FALLTHROUGH */ 235 case VCHR: 236 /* FALLTHROUGH */ 237 break; 238 case VDIR: 239 /* 240 * The parent link can be NULL if this is the root 241 * node. 242 */ 243 node->tn_links--; 244 node->tn_size = 0; 245 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root); 246 if (node->tn_dir.tn_parent) { 247 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 248 node->tn_dir.tn_parent->tn_links--; 249 250 /* 251 * If the parent directory has no more links and 252 * no vnode ref nothing is going to come along 253 * and clean it up unless we do it here. 254 */ 255 if (node->tn_dir.tn_parent->tn_links == 0 && 256 node->tn_dir.tn_parent->tn_vnode == NULL) { 257 tmpfs_free_node(tmp, node->tn_dir.tn_parent); 258 /* eats parent lock */ 259 } else { 260 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 261 } 262 node->tn_dir.tn_parent = NULL; 263 } 264 265 /* 266 * If the root node is being destroyed don't leave a 267 * dangling pointer in tmpfs_mount. 268 */ 269 if (node == tmp->tm_root) 270 tmp->tm_root = NULL; 271 break; 272 case VFIFO: 273 /* FALLTHROUGH */ 274 case VSOCK: 275 break; 276 277 case VLNK: 278 kfree(node->tn_link, tmp->tm_name_zone); 279 node->tn_link = NULL; 280 node->tn_size = 0; 281 break; 282 283 case VREG: 284 if (node->tn_reg.tn_aobj != NULL) 285 vm_object_deallocate(node->tn_reg.tn_aobj); 286 node->tn_reg.tn_aobj = NULL; 287 pages = node->tn_reg.tn_aobj_pages; 288 break; 289 290 default: 291 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 292 } 293 294 /* 295 * Clean up fields for the next allocation. The objcache only ctors 296 * new allocations. 297 */ 298 tmpfs_node_ctor(node, NULL, 0); 299 objcache_put(tmp->tm_node_pool, node); 300 /* node is now invalid */ 301 302 TMPFS_LOCK(tmp); 303 tmp->tm_pages_used -= pages; 304 TMPFS_UNLOCK(tmp); 305 } 306 307 /* --------------------------------------------------------------------- */ 308 309 /* 310 * Allocates a new directory entry for the node node with a name of name. 311 * The new directory entry is returned in *de. 312 * 313 * The link count of node is increased by one to reflect the new object 314 * referencing it. 315 * 316 * Returns zero on success or an appropriate error code on failure. 317 */ 318 int 319 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 320 const char *name, uint16_t len, struct tmpfs_dirent **de) 321 { 322 struct tmpfs_dirent *nde; 323 324 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK); 325 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK); 326 if (nde->td_name == NULL) { 327 objcache_put(tmp->tm_dirent_pool, nde); 328 *de = NULL; 329 return (ENOSPC); 330 } 331 nde->td_namelen = len; 332 bcopy(name, nde->td_name, len); 333 nde->td_name[len] = '\0'; 334 335 nde->td_node = node; 336 337 TMPFS_NODE_LOCK(node); 338 node->tn_links++; 339 TMPFS_NODE_UNLOCK(node); 340 341 *de = nde; 342 343 return 0; 344 } 345 346 /* --------------------------------------------------------------------- */ 347 348 /* 349 * Frees a directory entry. It is the caller's responsibility to destroy 350 * the node referenced by it if needed. 351 * 352 * The link count of node is decreased by one to reflect the removal of an 353 * object that referenced it. This only happens if 'node_exists' is true; 354 * otherwise the function will not access the node referred to by the 355 * directory entry, as it may already have been released from the outside. 356 */ 357 void 358 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 359 { 360 struct tmpfs_node *node; 361 362 node = de->td_node; 363 364 TMPFS_NODE_LOCK(node); 365 TMPFS_ASSERT_ELOCKED(node); 366 KKASSERT(node->tn_links > 0); 367 node->tn_links--; 368 TMPFS_NODE_UNLOCK(node); 369 370 kfree(de->td_name, tmp->tm_name_zone); 371 de->td_namelen = 0; 372 de->td_name = NULL; 373 de->td_node = NULL; 374 objcache_put(tmp->tm_dirent_pool, de); 375 } 376 377 /* --------------------------------------------------------------------- */ 378 379 /* 380 * Allocates a new vnode for the node node or returns a new reference to 381 * an existing one if the node had already a vnode referencing it. The 382 * resulting locked vnode is returned in *vpp. 383 * 384 * Returns zero on success or an appropriate error code on failure. 385 */ 386 int 387 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 388 struct vnode **vpp) 389 { 390 int error = 0; 391 struct vnode *vp; 392 393 loop: 394 /* 395 * Interlocked extraction from node. This can race many things. 396 * We have to get a soft reference on the vnode while we hold 397 * the node locked, then acquire it properly and check for races. 398 */ 399 TMPFS_NODE_LOCK(node); 400 if ((vp = node->tn_vnode) != NULL) { 401 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 402 vhold_interlocked(vp); 403 TMPFS_NODE_UNLOCK(node); 404 405 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) { 406 vdrop(vp); 407 goto loop; 408 } 409 if (node->tn_vnode != vp) { 410 vput(vp); 411 vdrop(vp); 412 goto loop; 413 } 414 vdrop(vp); 415 goto out; 416 } 417 /* vp is NULL */ 418 419 /* 420 * This should never happen. 421 */ 422 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) { 423 TMPFS_NODE_UNLOCK(node); 424 error = ENOENT; 425 goto out; 426 } 427 428 /* 429 * Interlock against other calls to tmpfs_alloc_vp() trying to 430 * allocate and assign a vp to node. 431 */ 432 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 433 node->tn_vpstate |= TMPFS_VNODE_WANT; 434 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH, 435 "tmpfs_alloc_vp", 0); 436 TMPFS_NODE_UNLOCK(node); 437 if (error) 438 return error; 439 goto loop; 440 } 441 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 442 TMPFS_NODE_UNLOCK(node); 443 444 /* 445 * Allocate a new vnode (may block). The ALLOCATING flag should 446 * prevent a race against someone else assigning node->tn_vnode. 447 */ 448 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE); 449 if (error != 0) 450 goto unlock; 451 452 KKASSERT(node->tn_vnode == NULL); 453 KKASSERT(vp != NULL); 454 vp->v_data = node; 455 vp->v_type = node->tn_type; 456 457 /* Type-specific initialization. */ 458 switch (node->tn_type) { 459 case VBLK: 460 /* FALLTHROUGH */ 461 case VCHR: 462 /* FALLTHROUGH */ 463 case VSOCK: 464 break; 465 case VREG: 466 vinitvmio(vp, node->tn_size, BMASK, -1); 467 break; 468 case VLNK: 469 break; 470 case VFIFO: 471 vp->v_ops = &mp->mnt_vn_fifo_ops; 472 break; 473 case VDIR: 474 break; 475 476 default: 477 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 478 } 479 480 insmntque(vp, mp); 481 482 unlock: 483 TMPFS_NODE_LOCK(node); 484 485 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 486 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 487 node->tn_vnode = vp; 488 489 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 490 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 491 TMPFS_NODE_UNLOCK(node); 492 wakeup(&node->tn_vpstate); 493 } else { 494 TMPFS_NODE_UNLOCK(node); 495 } 496 497 out: 498 *vpp = vp; 499 500 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 501 #ifdef INVARIANTS 502 TMPFS_NODE_LOCK(node); 503 KKASSERT(*vpp == node->tn_vnode); 504 TMPFS_NODE_UNLOCK(node); 505 #endif 506 507 return error; 508 } 509 510 /* --------------------------------------------------------------------- */ 511 512 /* 513 * Destroys the association between the vnode vp and the node it 514 * references. 515 */ 516 void 517 tmpfs_free_vp(struct vnode *vp) 518 { 519 struct tmpfs_node *node; 520 521 node = VP_TO_TMPFS_NODE(vp); 522 523 TMPFS_NODE_LOCK(node); 524 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0); 525 node->tn_vnode = NULL; 526 TMPFS_NODE_UNLOCK(node); 527 vp->v_data = NULL; 528 } 529 530 /* --------------------------------------------------------------------- */ 531 532 /* 533 * Allocates a new file of type 'type' and adds it to the parent directory 534 * 'dvp'; this addition is done using the component name given in 'cnp'. 535 * The ownership of the new file is automatically assigned based on the 536 * credentials of the caller (through 'cnp'), the group is set based on 537 * the parent directory and the mode is determined from the 'vap' argument. 538 * If successful, *vpp holds a vnode to the newly created file and zero 539 * is returned. Otherwise *vpp is NULL and the function returns an 540 * appropriate error code. 541 */ 542 int 543 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 544 struct namecache *ncp, struct ucred *cred, char *target) 545 { 546 int error; 547 struct tmpfs_dirent *de; 548 struct tmpfs_mount *tmp; 549 struct tmpfs_node *dnode; 550 struct tmpfs_node *node; 551 struct tmpfs_node *parent; 552 553 tmp = VFS_TO_TMPFS(dvp->v_mount); 554 dnode = VP_TO_TMPFS_DIR(dvp); 555 *vpp = NULL; 556 557 /* If the entry we are creating is a directory, we cannot overflow 558 * the number of links of its parent, because it will get a new 559 * link. */ 560 if (vap->va_type == VDIR) { 561 /* Ensure that we do not overflow the maximum number of links 562 * imposed by the system. */ 563 KKASSERT(dnode->tn_links <= LINK_MAX); 564 if (dnode->tn_links == LINK_MAX) { 565 return EMLINK; 566 } 567 568 parent = dnode; 569 KKASSERT(parent != NULL); 570 } else 571 parent = NULL; 572 573 /* Allocate a node that represents the new file. */ 574 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 575 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node); 576 if (error != 0) 577 return error; 578 TMPFS_NODE_LOCK(node); 579 580 /* Allocate a directory entry that points to the new file. */ 581 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de); 582 if (error != 0) { 583 tmpfs_free_node(tmp, node); 584 /* eats node lock */ 585 return error; 586 } 587 588 /* Allocate a vnode for the new file. */ 589 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 590 if (error != 0) { 591 tmpfs_free_dirent(tmp, de); 592 tmpfs_free_node(tmp, node); 593 /* eats node lock */ 594 return error; 595 } 596 597 /* Now that all required items are allocated, we can proceed to 598 * insert the new node into the directory, an operation that 599 * cannot fail. */ 600 tmpfs_dir_attach(dnode, de); 601 TMPFS_NODE_UNLOCK(node); 602 603 return error; 604 } 605 606 /* --------------------------------------------------------------------- */ 607 608 /* 609 * Attaches the directory entry de to the directory represented by vp. 610 * Note that this does not change the link count of the node pointed by 611 * the directory entry, as this is done by tmpfs_alloc_dirent. 612 */ 613 void 614 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 615 { 616 TMPFS_NODE_LOCK(dnode); 617 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries); 618 619 TMPFS_ASSERT_ELOCKED(dnode); 620 dnode->tn_size += sizeof(struct tmpfs_dirent); 621 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 622 TMPFS_NODE_MODIFIED; 623 TMPFS_NODE_UNLOCK(dnode); 624 } 625 626 /* --------------------------------------------------------------------- */ 627 628 /* 629 * Detaches the directory entry de from the directory represented by vp. 630 * Note that this does not change the link count of the node pointed by 631 * the directory entry, as this is done by tmpfs_free_dirent. 632 */ 633 void 634 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 635 { 636 TMPFS_NODE_LOCK(dnode); 637 if (dnode->tn_dir.tn_readdir_lastp == de) { 638 dnode->tn_dir.tn_readdir_lastn = 0; 639 dnode->tn_dir.tn_readdir_lastp = NULL; 640 } 641 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries); 642 643 TMPFS_ASSERT_ELOCKED(dnode); 644 dnode->tn_size -= sizeof(struct tmpfs_dirent); 645 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 646 TMPFS_NODE_MODIFIED; 647 TMPFS_NODE_UNLOCK(dnode); 648 } 649 650 /* --------------------------------------------------------------------- */ 651 652 /* 653 * Looks for a directory entry in the directory represented by node. 654 * 'ncp' describes the name of the entry to look for. Note that the . 655 * and .. components are not allowed as they do not physically exist 656 * within directories. 657 * 658 * Returns a pointer to the entry when found, otherwise NULL. 659 */ 660 struct tmpfs_dirent * 661 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 662 struct namecache *ncp) 663 { 664 struct tmpfs_dirent *de; 665 int len = ncp->nc_nlen; 666 667 TMPFS_VALIDATE_DIR(node); 668 669 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 670 if (f != NULL && de->td_node != f) 671 continue; 672 if (len == de->td_namelen) { 673 if (!memcmp(ncp->nc_name, de->td_name, len)) 674 break; 675 } 676 } 677 678 TMPFS_NODE_LOCK(node); 679 node->tn_status |= TMPFS_NODE_ACCESSED; 680 TMPFS_NODE_UNLOCK(node); 681 682 return de; 683 } 684 685 /* --------------------------------------------------------------------- */ 686 687 /* 688 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 689 * directory and returns it in the uio space. The function returns 0 690 * on success, -1 if there was not enough space in the uio structure to 691 * hold the directory entry or an appropriate error code if another 692 * error happens. 693 */ 694 int 695 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 696 { 697 int error; 698 struct dirent dent; 699 int dirsize; 700 701 TMPFS_VALIDATE_DIR(node); 702 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 703 704 dent.d_ino = node->tn_id; 705 dent.d_type = DT_DIR; 706 dent.d_namlen = 1; 707 dent.d_name[0] = '.'; 708 dent.d_name[1] = '\0'; 709 dirsize = _DIRENT_DIRSIZ(&dent); 710 711 if (dirsize > uio->uio_resid) 712 error = -1; 713 else { 714 error = uiomove((caddr_t)&dent, dirsize, uio); 715 if (error == 0) 716 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 717 } 718 719 TMPFS_NODE_LOCK(node); 720 node->tn_status |= TMPFS_NODE_ACCESSED; 721 TMPFS_NODE_UNLOCK(node); 722 723 return error; 724 } 725 726 /* --------------------------------------------------------------------- */ 727 728 /* 729 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 730 * directory and returns it in the uio space. The function returns 0 731 * on success, -1 if there was not enough space in the uio structure to 732 * hold the directory entry or an appropriate error code if another 733 * error happens. 734 */ 735 int 736 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 737 struct uio *uio) 738 { 739 int error; 740 struct dirent dent; 741 int dirsize; 742 743 TMPFS_VALIDATE_DIR(node); 744 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 745 746 if (node->tn_dir.tn_parent) { 747 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 748 dent.d_ino = node->tn_dir.tn_parent->tn_id; 749 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 750 } else { 751 dent.d_ino = tmp->tm_root->tn_id; 752 } 753 754 dent.d_type = DT_DIR; 755 dent.d_namlen = 2; 756 dent.d_name[0] = '.'; 757 dent.d_name[1] = '.'; 758 dent.d_name[2] = '\0'; 759 dirsize = _DIRENT_DIRSIZ(&dent); 760 761 if (dirsize > uio->uio_resid) 762 error = -1; 763 else { 764 error = uiomove((caddr_t)&dent, dirsize, uio); 765 if (error == 0) { 766 struct tmpfs_dirent *de; 767 768 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 769 if (de == NULL) 770 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 771 else 772 uio->uio_offset = tmpfs_dircookie(de); 773 } 774 } 775 776 TMPFS_NODE_LOCK(node); 777 node->tn_status |= TMPFS_NODE_ACCESSED; 778 TMPFS_NODE_UNLOCK(node); 779 780 return error; 781 } 782 783 /* --------------------------------------------------------------------- */ 784 785 /* 786 * Lookup a directory entry by its associated cookie. 787 */ 788 struct tmpfs_dirent * 789 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 790 { 791 struct tmpfs_dirent *de; 792 793 if (cookie == node->tn_dir.tn_readdir_lastn && 794 node->tn_dir.tn_readdir_lastp != NULL) { 795 return node->tn_dir.tn_readdir_lastp; 796 } 797 798 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 799 if (tmpfs_dircookie(de) == cookie) { 800 break; 801 } 802 } 803 804 return de; 805 } 806 807 /* --------------------------------------------------------------------- */ 808 809 /* 810 * Helper function for tmpfs_readdir. Returns as much directory entries 811 * as can fit in the uio space. The read starts at uio->uio_offset. 812 * The function returns 0 on success, -1 if there was not enough space 813 * in the uio structure to hold the directory entry or an appropriate 814 * error code if another error happens. 815 */ 816 int 817 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 818 { 819 int error; 820 off_t startcookie; 821 struct tmpfs_dirent *de; 822 823 TMPFS_VALIDATE_DIR(node); 824 825 /* Locate the first directory entry we have to return. We have cached 826 * the last readdir in the node, so use those values if appropriate. 827 * Otherwise do a linear scan to find the requested entry. */ 828 startcookie = uio->uio_offset; 829 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 830 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 831 if (startcookie == TMPFS_DIRCOOKIE_EOF) { 832 return 0; 833 } else { 834 de = tmpfs_dir_lookupbycookie(node, startcookie); 835 } 836 if (de == NULL) { 837 return EINVAL; 838 } 839 840 /* Read as much entries as possible; i.e., until we reach the end of 841 * the directory or we exhaust uio space. */ 842 do { 843 struct dirent d; 844 int reclen; 845 846 /* Create a dirent structure representing the current 847 * tmpfs_node and fill it. */ 848 d.d_ino = de->td_node->tn_id; 849 switch (de->td_node->tn_type) { 850 case VBLK: 851 d.d_type = DT_BLK; 852 break; 853 854 case VCHR: 855 d.d_type = DT_CHR; 856 break; 857 858 case VDIR: 859 d.d_type = DT_DIR; 860 break; 861 862 case VFIFO: 863 d.d_type = DT_FIFO; 864 break; 865 866 case VLNK: 867 d.d_type = DT_LNK; 868 break; 869 870 case VREG: 871 d.d_type = DT_REG; 872 break; 873 874 case VSOCK: 875 d.d_type = DT_SOCK; 876 break; 877 878 default: 879 panic("tmpfs_dir_getdents: type %p %d", 880 de->td_node, (int)de->td_node->tn_type); 881 } 882 d.d_namlen = de->td_namelen; 883 KKASSERT(de->td_namelen < sizeof(d.d_name)); 884 bcopy(de->td_name, d.d_name, d.d_namlen); 885 d.d_name[d.d_namlen] = '\0'; 886 reclen = _DIRENT_RECLEN(d.d_namlen); 887 888 /* Stop reading if the directory entry we are treating is 889 * bigger than the amount of data that can be returned. */ 890 if (reclen > uio->uio_resid) { 891 error = -1; 892 break; 893 } 894 895 /* Copy the new dirent structure into the output buffer and 896 * advance pointers. */ 897 error = uiomove((caddr_t)&d, reclen, uio); 898 899 (*cntp)++; 900 de = TAILQ_NEXT(de, td_entries); 901 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 902 903 /* Update the offset and cache. */ 904 if (de == NULL) { 905 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 906 node->tn_dir.tn_readdir_lastn = 0; 907 node->tn_dir.tn_readdir_lastp = NULL; 908 } else { 909 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de); 910 node->tn_dir.tn_readdir_lastp = de; 911 } 912 node->tn_status |= TMPFS_NODE_ACCESSED; 913 914 return error; 915 } 916 917 /* --------------------------------------------------------------------- */ 918 919 /* 920 * Resizes the aobj associated to the regular file pointed to by vp to 921 * the size newsize. 'vp' must point to a vnode that represents a regular 922 * file. 'newsize' must be positive. 923 * 924 * pass trivial as 1 when buf content will be overwritten, otherwise set 0 925 * to be zero filled. 926 * 927 * Returns zero on success or an appropriate error code on failure. 928 */ 929 int 930 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 931 { 932 int error; 933 vm_pindex_t newpages, oldpages; 934 struct tmpfs_mount *tmp; 935 struct tmpfs_node *node; 936 off_t oldsize; 937 938 #ifdef INVARIANTS 939 KKASSERT(vp->v_type == VREG); 940 KKASSERT(newsize >= 0); 941 #endif 942 943 node = VP_TO_TMPFS_NODE(vp); 944 tmp = VFS_TO_TMPFS(vp->v_mount); 945 946 /* Convert the old and new sizes to the number of pages needed to 947 * store them. It may happen that we do not need to do anything 948 * because the last allocated page can accommodate the change on 949 * its own. */ 950 oldsize = node->tn_size; 951 oldpages = round_page64(oldsize) / PAGE_SIZE; 952 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 953 newpages = round_page64(newsize) / PAGE_SIZE; 954 955 if (newpages > oldpages && 956 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 957 error = ENOSPC; 958 goto out; 959 } 960 961 TMPFS_LOCK(tmp); 962 tmp->tm_pages_used += (newpages - oldpages); 963 TMPFS_UNLOCK(tmp); 964 965 TMPFS_NODE_LOCK(node); 966 node->tn_reg.tn_aobj_pages = newpages; 967 node->tn_size = newsize; 968 TMPFS_NODE_UNLOCK(node); 969 970 /* 971 * When adjusting the vnode filesize and its VM object we must 972 * also adjust our backing VM object (aobj). The blocksize 973 * used must match the block sized we use for the buffer cache. 974 * 975 * The backing VM object contains no VM pages, only swap 976 * assignments. 977 */ 978 if (newsize < oldsize) { 979 vm_pindex_t osize; 980 vm_pindex_t nsize; 981 vm_object_t aobj; 982 983 error = nvtruncbuf(vp, newsize, BSIZE, -1); 984 aobj = node->tn_reg.tn_aobj; 985 if (aobj) { 986 osize = aobj->size; 987 nsize = vp->v_object->size; 988 if (nsize < osize) { 989 aobj->size = osize; 990 swap_pager_freespace(aobj, nsize, 991 osize - nsize); 992 } 993 } 994 } else { 995 vm_object_t aobj; 996 997 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE, 998 -1, -1, trivial); 999 aobj = node->tn_reg.tn_aobj; 1000 if (aobj) 1001 aobj->size = vp->v_object->size; 1002 } 1003 1004 out: 1005 return error; 1006 } 1007 1008 /* --------------------------------------------------------------------- */ 1009 1010 /* 1011 * Change flags of the given vnode. 1012 * Caller should execute tmpfs_update on vp after a successful execution. 1013 * The vnode must be locked on entry and remain locked on exit. 1014 */ 1015 int 1016 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred) 1017 { 1018 int error; 1019 struct tmpfs_node *node; 1020 int flags; 1021 1022 KKASSERT(vn_islocked(vp)); 1023 1024 node = VP_TO_TMPFS_NODE(vp); 1025 flags = node->tn_flags; 1026 1027 /* Disallow this operation if the file system is mounted read-only. */ 1028 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1029 return EROFS; 1030 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1031 1032 /* 1033 * Unprivileged processes are not permitted to unset system 1034 * flags, or modify flags if any system flags are set. 1035 * 1036 * Silently enforce SF_NOCACHE on the root tmpfs vnode so 1037 * tmpfs data is not double-cached by swapcache. 1038 */ 1039 if (error == 0) { 1040 TMPFS_NODE_LOCK(node); 1041 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1042 if (vp->v_flag & VROOT) 1043 flags |= SF_NOCACHE; 1044 node->tn_flags = flags; 1045 } else { 1046 if (node->tn_flags & (SF_NOUNLINK | SF_IMMUTABLE | 1047 SF_APPEND) || 1048 (flags & UF_SETTABLE) != flags) { 1049 error = EPERM; 1050 } else { 1051 node->tn_flags &= SF_SETTABLE; 1052 node->tn_flags |= (flags & UF_SETTABLE); 1053 } 1054 } 1055 node->tn_status |= TMPFS_NODE_CHANGED; 1056 TMPFS_NODE_UNLOCK(node); 1057 } 1058 1059 KKASSERT(vn_islocked(vp)); 1060 1061 return error; 1062 } 1063 1064 /* --------------------------------------------------------------------- */ 1065 1066 /* 1067 * Change access mode on the given vnode. 1068 * Caller should execute tmpfs_update on vp after a successful execution. 1069 * The vnode must be locked on entry and remain locked on exit. 1070 */ 1071 int 1072 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1073 { 1074 struct tmpfs_node *node; 1075 mode_t cur_mode; 1076 int error; 1077 1078 KKASSERT(vn_islocked(vp)); 1079 1080 node = VP_TO_TMPFS_NODE(vp); 1081 1082 /* Disallow this operation if the file system is mounted read-only. */ 1083 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1084 return EROFS; 1085 1086 /* Immutable or append-only files cannot be modified, either. */ 1087 if (node->tn_flags & (IMMUTABLE | APPEND)) 1088 return EPERM; 1089 1090 cur_mode = node->tn_mode; 1091 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1092 &cur_mode); 1093 1094 if (error == 0 && 1095 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1096 TMPFS_NODE_LOCK(node); 1097 node->tn_mode &= ~ALLPERMS; 1098 node->tn_mode |= cur_mode & ALLPERMS; 1099 1100 node->tn_status |= TMPFS_NODE_CHANGED; 1101 TMPFS_NODE_UNLOCK(node); 1102 } 1103 1104 KKASSERT(vn_islocked(vp)); 1105 1106 return 0; 1107 } 1108 1109 /* --------------------------------------------------------------------- */ 1110 1111 /* 1112 * Change ownership of the given vnode. At least one of uid or gid must 1113 * be different than VNOVAL. If one is set to that value, the attribute 1114 * is unchanged. 1115 * Caller should execute tmpfs_update on vp after a successful execution. 1116 * The vnode must be locked on entry and remain locked on exit. 1117 */ 1118 int 1119 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1120 { 1121 mode_t cur_mode; 1122 uid_t cur_uid; 1123 gid_t cur_gid; 1124 struct tmpfs_node *node; 1125 int error; 1126 1127 KKASSERT(vn_islocked(vp)); 1128 node = VP_TO_TMPFS_NODE(vp); 1129 1130 /* Disallow this operation if the file system is mounted read-only. */ 1131 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1132 return EROFS; 1133 1134 /* Immutable or append-only files cannot be modified, either. */ 1135 if (node->tn_flags & (IMMUTABLE | APPEND)) 1136 return EPERM; 1137 1138 cur_uid = node->tn_uid; 1139 cur_gid = node->tn_gid; 1140 cur_mode = node->tn_mode; 1141 error = vop_helper_chown(vp, uid, gid, cred, 1142 &cur_uid, &cur_gid, &cur_mode); 1143 1144 if (error == 0) { 1145 TMPFS_NODE_LOCK(node); 1146 if (cur_uid != node->tn_uid || 1147 cur_gid != node->tn_gid || 1148 cur_mode != node->tn_mode) { 1149 node->tn_uid = cur_uid; 1150 node->tn_gid = cur_gid; 1151 node->tn_mode = cur_mode; 1152 node->tn_status |= TMPFS_NODE_CHANGED; 1153 } 1154 TMPFS_NODE_UNLOCK(node); 1155 } 1156 1157 return error; 1158 } 1159 1160 /* --------------------------------------------------------------------- */ 1161 1162 /* 1163 * Change size of the given vnode. 1164 * Caller should execute tmpfs_update on vp after a successful execution. 1165 * The vnode must be locked on entry and remain locked on exit. 1166 */ 1167 int 1168 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1169 { 1170 int error; 1171 struct tmpfs_node *node; 1172 1173 KKASSERT(vn_islocked(vp)); 1174 1175 node = VP_TO_TMPFS_NODE(vp); 1176 1177 /* Decide whether this is a valid operation based on the file type. */ 1178 error = 0; 1179 switch (vp->v_type) { 1180 case VDIR: 1181 return EISDIR; 1182 1183 case VREG: 1184 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1185 return EROFS; 1186 break; 1187 1188 case VBLK: 1189 /* FALLTHROUGH */ 1190 case VCHR: 1191 /* FALLTHROUGH */ 1192 case VFIFO: 1193 /* Allow modifications of special files even if in the file 1194 * system is mounted read-only (we are not modifying the 1195 * files themselves, but the objects they represent). */ 1196 return 0; 1197 1198 default: 1199 /* Anything else is unsupported. */ 1200 return EOPNOTSUPP; 1201 } 1202 1203 /* Immutable or append-only files cannot be modified, either. */ 1204 if (node->tn_flags & (IMMUTABLE | APPEND)) 1205 return EPERM; 1206 1207 error = tmpfs_truncate(vp, size); 1208 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1209 * for us, as will update tn_status; no need to do that here. */ 1210 1211 KKASSERT(vn_islocked(vp)); 1212 1213 return error; 1214 } 1215 1216 /* --------------------------------------------------------------------- */ 1217 1218 /* 1219 * Change access and modification times of the given vnode. 1220 * Caller should execute tmpfs_update on vp after a successful execution. 1221 * The vnode must be locked on entry and remain locked on exit. 1222 */ 1223 int 1224 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1225 int vaflags, struct ucred *cred) 1226 { 1227 struct tmpfs_node *node; 1228 1229 KKASSERT(vn_islocked(vp)); 1230 1231 node = VP_TO_TMPFS_NODE(vp); 1232 1233 /* Disallow this operation if the file system is mounted read-only. */ 1234 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1235 return EROFS; 1236 1237 /* Immutable or append-only files cannot be modified, either. */ 1238 if (node->tn_flags & (IMMUTABLE | APPEND)) 1239 return EPERM; 1240 1241 TMPFS_NODE_LOCK(node); 1242 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1243 node->tn_status |= TMPFS_NODE_ACCESSED; 1244 1245 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1246 node->tn_status |= TMPFS_NODE_MODIFIED; 1247 1248 TMPFS_NODE_UNLOCK(node); 1249 1250 tmpfs_itimes(vp, atime, mtime); 1251 1252 KKASSERT(vn_islocked(vp)); 1253 1254 return 0; 1255 } 1256 1257 /* --------------------------------------------------------------------- */ 1258 /* Sync timestamps */ 1259 void 1260 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1261 const struct timespec *mod) 1262 { 1263 struct tmpfs_node *node; 1264 struct timespec now; 1265 1266 node = VP_TO_TMPFS_NODE(vp); 1267 1268 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1269 TMPFS_NODE_CHANGED)) == 0) 1270 return; 1271 1272 vfs_timestamp(&now); 1273 1274 TMPFS_NODE_LOCK(node); 1275 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1276 if (acc == NULL) 1277 acc = &now; 1278 node->tn_atime = acc->tv_sec; 1279 node->tn_atimensec = acc->tv_nsec; 1280 } 1281 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1282 if (mod == NULL) 1283 mod = &now; 1284 node->tn_mtime = mod->tv_sec; 1285 node->tn_mtimensec = mod->tv_nsec; 1286 } 1287 if (node->tn_status & TMPFS_NODE_CHANGED) { 1288 node->tn_ctime = now.tv_sec; 1289 node->tn_ctimensec = now.tv_nsec; 1290 } 1291 node->tn_status &= 1292 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1293 TMPFS_NODE_UNLOCK(node); 1294 } 1295 1296 /* --------------------------------------------------------------------- */ 1297 1298 void 1299 tmpfs_update(struct vnode *vp) 1300 { 1301 1302 tmpfs_itimes(vp, NULL, NULL); 1303 } 1304 1305 /* --------------------------------------------------------------------- */ 1306 1307 int 1308 tmpfs_truncate(struct vnode *vp, off_t length) 1309 { 1310 int error; 1311 struct tmpfs_node *node; 1312 1313 node = VP_TO_TMPFS_NODE(vp); 1314 1315 if (length < 0) { 1316 error = EINVAL; 1317 goto out; 1318 } 1319 1320 if (node->tn_size == length) { 1321 error = 0; 1322 goto out; 1323 } 1324 1325 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1326 return (EFBIG); 1327 1328 1329 error = tmpfs_reg_resize(vp, length, 1); 1330 1331 if (error == 0) { 1332 TMPFS_NODE_LOCK(node); 1333 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1334 TMPFS_NODE_UNLOCK(node); 1335 } 1336 1337 out: 1338 tmpfs_update(vp); 1339 1340 return error; 1341 } 1342 1343 /* --------------------------------------------------------------------- */ 1344 1345 static ino_t 1346 tmpfs_fetch_ino(void) 1347 { 1348 ino_t ret; 1349 1350 spin_lock(&ino_lock); 1351 ret = t_ino++; 1352 spin_unlock(&ino_lock); 1353 1354 return ret; 1355 } 1356