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