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/stat.h> 43 #include <sys/systm.h> 44 #include <sys/vnode.h> 45 #include <sys/vmmeter.h> 46 47 #include <vm/vm.h> 48 #include <vm/vm_object.h> 49 #include <vm/vm_page.h> 50 #include <vm/vm_pager.h> 51 #include <vm/vm_extern.h> 52 53 #include <vfs/tmpfs/tmpfs.h> 54 #include <vfs/tmpfs/tmpfs_vnops.h> 55 56 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *); 57 58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a, 59 struct tmpfs_dirent *b); 60 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare); 61 62 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, 63 struct tmpfs_dirent *b); 64 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent, 65 rb_cookienode, tmpfs_dirtree_compare_cookie); 66 67 68 /* --------------------------------------------------------------------- */ 69 70 /* 71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 73 * using the credentials of the process 'p'. 74 * 75 * If the node type is set to 'VDIR', then the parent parameter must point 76 * to the parent directory of the node being created. It may only be NULL 77 * while allocating the root node. 78 * 79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 80 * specifies the device the node represents. 81 * 82 * If the node type is set to 'VLNK', then the parameter target specifies 83 * the file name of the target file for the symbolic link that is being 84 * created. 85 * 86 * Note that new nodes are retrieved from the available list if it has 87 * items or, if it is empty, from the node pool as long as there is enough 88 * space to create them. 89 * 90 * Returns zero on success or an appropriate error code on failure. 91 */ 92 int 93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 94 uid_t uid, gid_t gid, mode_t mode, 95 char *target, int rmajor, int rminor, 96 struct tmpfs_node **node) 97 { 98 struct tmpfs_node *nnode; 99 struct timespec ts; 100 udev_t rdev; 101 102 KKASSERT(IFF(type == VLNK, target != NULL)); 103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL)); 104 105 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 106 return (ENOSPC); 107 108 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK); 109 if (nnode == NULL) 110 return (ENOSPC); 111 112 /* Generic initialization. */ 113 nnode->tn_type = type; 114 vfs_timestamp(&ts); 115 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime 116 = ts.tv_sec; 117 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec 118 = ts.tv_nsec; 119 nnode->tn_uid = uid; 120 nnode->tn_gid = gid; 121 nnode->tn_mode = mode; 122 nnode->tn_id = tmpfs_fetch_ino(tmp); 123 nnode->tn_advlock.init_done = 0; 124 KKASSERT(nnode->tn_links == 0); 125 126 /* Type-specific initialization. */ 127 switch (nnode->tn_type) { 128 case VBLK: 129 case VCHR: 130 rdev = makeudev(rmajor, rminor); 131 if (rdev == NOUDEV) { 132 objcache_put(tmp->tm_node_pool, nnode); 133 return(EINVAL); 134 } 135 nnode->tn_rdev = rdev; 136 break; 137 138 case VDIR: 139 RB_INIT(&nnode->tn_dir.tn_dirtree); 140 RB_INIT(&nnode->tn_dir.tn_cookietree); 141 nnode->tn_size = 0; 142 break; 143 144 case VFIFO: 145 /* FALLTHROUGH */ 146 case VSOCK: 147 break; 148 149 case VLNK: 150 nnode->tn_size = strlen(target); 151 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone, 152 M_WAITOK | M_NULLOK); 153 if (nnode->tn_link == NULL) { 154 objcache_put(tmp->tm_node_pool, nnode); 155 return (ENOSPC); 156 } 157 bcopy(target, nnode->tn_link, nnode->tn_size); 158 nnode->tn_link[nnode->tn_size] = '\0'; 159 break; 160 161 case VREG: 162 nnode->tn_reg.tn_aobj = 163 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0); 164 nnode->tn_reg.tn_aobj_pages = 0; 165 nnode->tn_size = 0; 166 break; 167 168 default: 169 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 170 } 171 172 TMPFS_NODE_LOCK(nnode); 173 TMPFS_LOCK(tmp); 174 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 175 tmp->tm_nodes_inuse++; 176 TMPFS_UNLOCK(tmp); 177 TMPFS_NODE_UNLOCK(nnode); 178 179 *node = nnode; 180 return 0; 181 } 182 183 /* --------------------------------------------------------------------- */ 184 185 /* 186 * Destroys the node pointed to by node from the file system 'tmp'. 187 * If the node does not belong to the given mount point, the results are 188 * unpredicted. 189 * 190 * If the node references a directory; no entries are allowed because 191 * their removal could need a recursive algorithm, something forbidden in 192 * kernel space. Furthermore, there is not need to provide such 193 * functionality (recursive removal) because the only primitives offered 194 * to the user are the removal of empty directories and the deletion of 195 * individual files. 196 * 197 * Note that nodes are not really deleted; in fact, when a node has been 198 * allocated, it cannot be deleted during the whole life of the file 199 * system. Instead, they are moved to the available list and remain there 200 * until reused. 201 * 202 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it. 203 */ 204 void 205 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 206 { 207 vm_pindex_t pages = 0; 208 209 #ifdef INVARIANTS 210 TMPFS_ASSERT_ELOCKED(node); 211 KKASSERT(node->tn_vnode == NULL); 212 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 213 #endif 214 215 TMPFS_LOCK(tmp); 216 LIST_REMOVE(node, tn_entries); 217 tmp->tm_nodes_inuse--; 218 TMPFS_UNLOCK(tmp); 219 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */ 220 221 switch (node->tn_type) { 222 case VNON: 223 /* Do not do anything. VNON is provided to let the 224 * allocation routine clean itself easily by avoiding 225 * duplicating code in it. */ 226 /* FALLTHROUGH */ 227 case VBLK: 228 /* FALLTHROUGH */ 229 case VCHR: 230 /* FALLTHROUGH */ 231 break; 232 case VDIR: 233 /* 234 * The parent link can be NULL if this is the root 235 * node or if it is a directory node that was rmdir'd. 236 * 237 * XXX what if node is a directory which still contains 238 * directory entries (e.g. due to a forced umount) ? 239 */ 240 node->tn_size = 0; 241 KKASSERT(node->tn_dir.tn_parent == NULL); 242 243 /* 244 * If the root node is being destroyed don't leave a 245 * dangling pointer in tmpfs_mount. 246 */ 247 if (node == tmp->tm_root) 248 tmp->tm_root = NULL; 249 break; 250 case VFIFO: 251 /* FALLTHROUGH */ 252 case VSOCK: 253 break; 254 255 case VLNK: 256 kfree(node->tn_link, tmp->tm_name_zone); 257 node->tn_link = NULL; 258 node->tn_size = 0; 259 break; 260 261 case VREG: 262 if (node->tn_reg.tn_aobj != NULL) 263 vm_object_deallocate(node->tn_reg.tn_aobj); 264 node->tn_reg.tn_aobj = NULL; 265 pages = node->tn_reg.tn_aobj_pages; 266 break; 267 268 default: 269 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 270 } 271 272 /* 273 * Clean up fields for the next allocation. The objcache only ctors 274 * new allocations. 275 */ 276 tmpfs_node_ctor(node, NULL, 0); 277 objcache_put(tmp->tm_node_pool, node); 278 /* node is now invalid */ 279 280 atomic_add_long(&tmp->tm_pages_used, -(long)pages); 281 } 282 283 /* --------------------------------------------------------------------- */ 284 285 /* 286 * Allocates a new directory entry for the node node with a name of name. 287 * The new directory entry is returned in *de. 288 * 289 * The link count of node is increased by one to reflect the new object 290 * referencing it. 291 * 292 * Returns zero on success or an appropriate error code on failure. 293 */ 294 int 295 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 296 const char *name, uint16_t len, struct tmpfs_dirent **de) 297 { 298 struct tmpfs_dirent *nde; 299 300 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK); 301 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK); 302 if (nde->td_name == NULL) { 303 objcache_put(tmp->tm_dirent_pool, nde); 304 *de = NULL; 305 return (ENOSPC); 306 } 307 nde->td_namelen = len; 308 bcopy(name, nde->td_name, len); 309 nde->td_name[len] = '\0'; 310 311 nde->td_node = node; 312 313 TMPFS_NODE_LOCK(node); 314 ++node->tn_links; 315 TMPFS_NODE_UNLOCK(node); 316 317 *de = nde; 318 319 return 0; 320 } 321 322 /* --------------------------------------------------------------------- */ 323 324 /* 325 * Frees a directory entry. It is the caller's responsibility to destroy 326 * the node referenced by it if needed. 327 * 328 * The link count of node is decreased by one to reflect the removal of an 329 * object that referenced it. This only happens if 'node_exists' is true; 330 * otherwise the function will not access the node referred to by the 331 * directory entry, as it may already have been released from the outside. 332 */ 333 void 334 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 335 { 336 struct tmpfs_node *node; 337 338 node = de->td_node; 339 340 TMPFS_NODE_LOCK(node); 341 TMPFS_ASSERT_ELOCKED(node); 342 KKASSERT(node->tn_links > 0); 343 node->tn_links--; 344 TMPFS_NODE_UNLOCK(node); 345 346 kfree(de->td_name, tmp->tm_name_zone); 347 de->td_namelen = 0; 348 de->td_name = NULL; 349 de->td_node = NULL; 350 objcache_put(tmp->tm_dirent_pool, de); 351 } 352 353 /* --------------------------------------------------------------------- */ 354 355 /* 356 * Allocates a new vnode for the node node or returns a new reference to 357 * an existing one if the node had already a vnode referencing it. The 358 * resulting locked vnode is returned in *vpp. 359 * 360 * Returns zero on success or an appropriate error code on failure. 361 */ 362 int 363 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 364 struct vnode **vpp) 365 { 366 int error = 0; 367 struct vnode *vp; 368 369 loop: 370 /* 371 * Interlocked extraction from node. This can race many things. 372 * We have to get a soft reference on the vnode while we hold 373 * the node locked, then acquire it properly and check for races. 374 */ 375 TMPFS_NODE_LOCK(node); 376 if ((vp = node->tn_vnode) != NULL) { 377 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 378 vhold(vp); 379 TMPFS_NODE_UNLOCK(node); 380 381 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) { 382 vdrop(vp); 383 goto loop; 384 } 385 if (node->tn_vnode != vp) { 386 vput(vp); 387 vdrop(vp); 388 goto loop; 389 } 390 vdrop(vp); 391 goto out; 392 } 393 /* vp is NULL */ 394 395 /* 396 * This should never happen. 397 */ 398 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) { 399 TMPFS_NODE_UNLOCK(node); 400 error = ENOENT; 401 goto out; 402 } 403 404 /* 405 * Interlock against other calls to tmpfs_alloc_vp() trying to 406 * allocate and assign a vp to node. 407 */ 408 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 409 node->tn_vpstate |= TMPFS_VNODE_WANT; 410 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH, 411 "tmpfs_alloc_vp", 0); 412 TMPFS_NODE_UNLOCK(node); 413 if (error) 414 return error; 415 goto loop; 416 } 417 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 418 TMPFS_NODE_UNLOCK(node); 419 420 /* 421 * Allocate a new vnode (may block). The ALLOCATING flag should 422 * prevent a race against someone else assigning node->tn_vnode. 423 */ 424 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE); 425 if (error != 0) 426 goto unlock; 427 428 KKASSERT(node->tn_vnode == NULL); 429 KKASSERT(vp != NULL); 430 vp->v_data = node; 431 vp->v_type = node->tn_type; 432 433 /* Type-specific initialization. */ 434 switch (node->tn_type) { 435 case VBLK: 436 /* FALLTHROUGH */ 437 case VCHR: 438 /* FALLTHROUGH */ 439 case VSOCK: 440 break; 441 case VREG: 442 vinitvmio(vp, node->tn_size, TMPFS_BLKMASK, -1); 443 break; 444 case VLNK: 445 break; 446 case VFIFO: 447 vp->v_ops = &mp->mnt_vn_fifo_ops; 448 break; 449 case VDIR: 450 break; 451 452 default: 453 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 454 } 455 456 457 unlock: 458 TMPFS_NODE_LOCK(node); 459 460 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 461 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 462 node->tn_vnode = vp; 463 464 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 465 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 466 TMPFS_NODE_UNLOCK(node); 467 wakeup(&node->tn_vpstate); 468 } else { 469 TMPFS_NODE_UNLOCK(node); 470 } 471 472 out: 473 *vpp = vp; 474 475 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 476 #ifdef INVARIANTS 477 TMPFS_NODE_LOCK(node); 478 KKASSERT(*vpp == node->tn_vnode); 479 TMPFS_NODE_UNLOCK(node); 480 #endif 481 482 return error; 483 } 484 485 /* --------------------------------------------------------------------- */ 486 487 /* 488 * Destroys the association between the vnode vp and the node it 489 * references. 490 */ 491 void 492 tmpfs_free_vp(struct vnode *vp) 493 { 494 struct tmpfs_node *node; 495 496 node = VP_TO_TMPFS_NODE(vp); 497 498 TMPFS_NODE_LOCK(node); 499 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0); 500 node->tn_vnode = NULL; 501 vp->v_data = NULL; 502 TMPFS_NODE_UNLOCK(node); 503 } 504 505 /* --------------------------------------------------------------------- */ 506 507 /* 508 * Allocates a new file of type 'type' and adds it to the parent directory 509 * 'dvp'; this addition is done using the component name given in 'cnp'. 510 * The ownership of the new file is automatically assigned based on the 511 * credentials of the caller (through 'cnp'), the group is set based on 512 * the parent directory and the mode is determined from the 'vap' argument. 513 * If successful, *vpp holds a vnode to the newly created file and zero 514 * is returned. Otherwise *vpp is NULL and the function returns an 515 * appropriate error code. 516 */ 517 int 518 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 519 struct namecache *ncp, struct ucred *cred, char *target) 520 { 521 int error; 522 struct tmpfs_dirent *de; 523 struct tmpfs_mount *tmp; 524 struct tmpfs_node *dnode; 525 struct tmpfs_node *node; 526 527 tmp = VFS_TO_TMPFS(dvp->v_mount); 528 dnode = VP_TO_TMPFS_DIR(dvp); 529 *vpp = NULL; 530 531 /* 532 * If the directory was removed but a process was CD'd into it, 533 * we do not allow any more file/dir creation within it. Otherwise 534 * we will lose track of it. 535 */ 536 KKASSERT(dnode->tn_type == VDIR); 537 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) 538 return ENOENT; 539 540 /* 541 * Make sure the link count does not overflow. 542 */ 543 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) 544 return EMLINK; 545 546 /* Allocate a node that represents the new file. */ 547 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 548 dnode->tn_gid, vap->va_mode, target, 549 vap->va_rmajor, vap->va_rminor, &node); 550 if (error != 0) 551 return error; 552 TMPFS_NODE_LOCK(node); 553 554 /* Allocate a directory entry that points to the new file. */ 555 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de); 556 if (error != 0) { 557 tmpfs_free_node(tmp, node); 558 /* eats node lock */ 559 return error; 560 } 561 562 /* Allocate a vnode for the new file. */ 563 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 564 if (error != 0) { 565 tmpfs_free_dirent(tmp, de); 566 tmpfs_free_node(tmp, node); 567 /* eats node lock */ 568 return error; 569 } 570 571 /* 572 * Now that all required items are allocated, we can proceed to 573 * insert the new node into the directory, an operation that 574 * cannot fail. 575 */ 576 tmpfs_dir_attach(dnode, de); 577 TMPFS_NODE_UNLOCK(node); 578 579 return error; 580 } 581 582 /* --------------------------------------------------------------------- */ 583 584 /* 585 * Attaches the directory entry de to the directory represented by dnode. 586 * Note that this does not change the link count of the node pointed by 587 * the directory entry, as this is done by tmpfs_alloc_dirent. 588 */ 589 void 590 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 591 { 592 struct tmpfs_node *node = de->td_node; 593 594 TMPFS_NODE_LOCK(dnode); 595 if (node && node->tn_type == VDIR) { 596 TMPFS_NODE_LOCK(node); 597 ++node->tn_links; 598 node->tn_status |= TMPFS_NODE_CHANGED; 599 node->tn_dir.tn_parent = dnode; 600 ++dnode->tn_links; 601 TMPFS_NODE_UNLOCK(node); 602 } 603 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 604 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 605 dnode->tn_size += sizeof(struct tmpfs_dirent); 606 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 607 TMPFS_NODE_MODIFIED; 608 TMPFS_NODE_UNLOCK(dnode); 609 } 610 611 /* --------------------------------------------------------------------- */ 612 613 /* 614 * Detaches the directory entry de from the directory represented by dnode. 615 * Note that this does not change the link count of the node pointed by 616 * the directory entry, as this is done by tmpfs_free_dirent. 617 */ 618 void 619 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 620 { 621 struct tmpfs_node *node = de->td_node; 622 623 TMPFS_NODE_LOCK(dnode); 624 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 625 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 626 dnode->tn_size -= sizeof(struct tmpfs_dirent); 627 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 628 TMPFS_NODE_MODIFIED; 629 TMPFS_NODE_UNLOCK(dnode); 630 631 /* 632 * Clean out the tn_parent pointer immediately when removing a 633 * directory. 634 * 635 * Removal of the parent linkage also cleans out the extra tn_links 636 * count we had on both node and dnode. 637 * 638 * node can be NULL (typ during a forced umount), in which case 639 * the mount code is dealing with the linkages from a linked list 640 * scan. 641 */ 642 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) { 643 TMPFS_NODE_LOCK(dnode); 644 TMPFS_NODE_LOCK(node); 645 KKASSERT(node->tn_dir.tn_parent == dnode); 646 dnode->tn_links--; 647 node->tn_links--; 648 node->tn_dir.tn_parent = NULL; 649 TMPFS_NODE_UNLOCK(node); 650 TMPFS_NODE_UNLOCK(dnode); 651 } 652 } 653 654 /* --------------------------------------------------------------------- */ 655 656 /* 657 * Looks for a directory entry in the directory represented by node. 658 * 'ncp' describes the name of the entry to look for. Note that the . 659 * and .. components are not allowed as they do not physically exist 660 * within directories. 661 * 662 * Returns a pointer to the entry when found, otherwise NULL. 663 * 664 * Caller must hold the node locked (shared ok) 665 */ 666 struct tmpfs_dirent * 667 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 668 struct namecache *ncp) 669 { 670 struct tmpfs_dirent *de; 671 int len = ncp->nc_nlen; 672 struct tmpfs_dirent wanted; 673 674 wanted.td_namelen = len; 675 wanted.td_name = ncp->nc_name; 676 677 TMPFS_VALIDATE_DIR(node); 678 679 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted); 680 681 KKASSERT(f == NULL || f == de->td_node); 682 683 return de; 684 } 685 686 /* --------------------------------------------------------------------- */ 687 688 /* 689 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 690 * directory and returns it in the uio space. The function returns 0 691 * on success, -1 if there was not enough space in the uio structure to 692 * hold the directory entry or an appropriate error code if another 693 * error happens. 694 */ 695 int 696 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 697 { 698 int error; 699 700 TMPFS_VALIDATE_DIR(node); 701 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 702 703 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, ".")) 704 return -1; 705 if (error == 0) 706 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 707 return error; 708 } 709 710 /* --------------------------------------------------------------------- */ 711 712 /* 713 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 714 * directory and returns it in the uio space. The function returns 0 715 * on success, -1 if there was not enough space in the uio structure to 716 * hold the directory entry or an appropriate error code if another 717 * error happens. 718 */ 719 int 720 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 721 struct uio *uio) 722 { 723 int error; 724 ino_t d_ino; 725 726 TMPFS_VALIDATE_DIR(node); 727 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 728 729 if (node->tn_dir.tn_parent) { 730 TMPFS_NODE_LOCK(node); 731 if (node->tn_dir.tn_parent) 732 d_ino = node->tn_dir.tn_parent->tn_id; 733 else 734 d_ino = tmp->tm_root->tn_id; 735 TMPFS_NODE_UNLOCK(node); 736 } else { 737 d_ino = tmp->tm_root->tn_id; 738 } 739 740 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, "..")) 741 return -1; 742 if (error == 0) { 743 struct tmpfs_dirent *de; 744 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree); 745 if (de == NULL) 746 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 747 else 748 uio->uio_offset = tmpfs_dircookie(de); 749 } 750 return error; 751 } 752 753 /* --------------------------------------------------------------------- */ 754 755 /* 756 * Lookup a directory entry by its associated cookie. 757 * 758 * Must be called with the directory node locked (shared ok) 759 */ 760 struct lubycookie_info { 761 off_t cookie; 762 struct tmpfs_dirent *de; 763 }; 764 765 static int 766 lubycookie_cmp(struct tmpfs_dirent *de, void *arg) 767 { 768 struct lubycookie_info *info = arg; 769 off_t cookie = tmpfs_dircookie(de); 770 771 if (cookie < info->cookie) 772 return(-1); 773 if (cookie > info->cookie) 774 return(1); 775 return(0); 776 } 777 778 static int 779 lubycookie_callback(struct tmpfs_dirent *de, void *arg) 780 { 781 struct lubycookie_info *info = arg; 782 783 if (tmpfs_dircookie(de) == info->cookie) { 784 info->de = de; 785 return(-1); 786 } 787 return(0); 788 } 789 790 struct tmpfs_dirent * 791 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 792 { 793 struct lubycookie_info info; 794 795 info.cookie = cookie; 796 info.de = NULL; 797 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree, 798 lubycookie_cmp, lubycookie_callback, &info); 799 return (info.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 * Caller must hold the node locked (shared ok) 812 */ 813 int 814 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 815 { 816 int error; 817 off_t startcookie; 818 struct tmpfs_dirent *de; 819 820 TMPFS_VALIDATE_DIR(node); 821 822 /* 823 * Locate the first directory entry we have to return. We have cached 824 * the last readdir in the node, so use those values if appropriate. 825 * Otherwise do a linear scan to find the requested entry. 826 */ 827 startcookie = uio->uio_offset; 828 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 829 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 830 831 if (startcookie == TMPFS_DIRCOOKIE_EOF) 832 return 0; 833 834 de = tmpfs_dir_lookupbycookie(node, startcookie); 835 if (de == NULL) 836 return EINVAL; 837 838 /* 839 * Read as much entries as possible; i.e., until we reach the end of 840 * the directory or we exhaust uio space. 841 */ 842 do { 843 ino_t d_ino; 844 uint8_t d_type; 845 846 /* Create a dirent structure representing the current 847 * tmpfs_node and fill it. */ 848 d_ino = de->td_node->tn_id; 849 switch (de->td_node->tn_type) { 850 case VBLK: 851 d_type = DT_BLK; 852 break; 853 854 case VCHR: 855 d_type = DT_CHR; 856 break; 857 858 case VDIR: 859 d_type = DT_DIR; 860 break; 861 862 case VFIFO: 863 d_type = DT_FIFO; 864 break; 865 866 case VLNK: 867 d_type = DT_LNK; 868 break; 869 870 case VREG: 871 d_type = DT_REG; 872 break; 873 874 case VSOCK: 875 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 KKASSERT(de->td_namelen < 256); /* 255 + 1 */ 883 884 if (vop_write_dirent(&error, uio, d_ino, d_type, 885 de->td_namelen, de->td_name)) { 886 error = -1; 887 break; 888 } 889 890 (*cntp)++; 891 de = RB_NEXT(tmpfs_dirtree_cookie, 892 node->tn_dir.tn_cookietree, de); 893 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 894 895 /* Update the offset and cache. */ 896 if (de == NULL) { 897 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 898 } else { 899 uio->uio_offset = tmpfs_dircookie(de); 900 } 901 902 return error; 903 } 904 905 /* --------------------------------------------------------------------- */ 906 907 /* 908 * Resizes the aobj associated to the regular file pointed to by vp to 909 * the size newsize. 'vp' must point to a vnode that represents a regular 910 * file. 'newsize' must be positive. 911 * 912 * pass trivial as 1 when buf content will be overwritten, otherwise set 0 913 * to be zero filled. 914 * 915 * Returns zero on success or an appropriate error code on failure. 916 * 917 * Caller must hold the node exclusively locked. 918 */ 919 int 920 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 921 { 922 int error; 923 vm_pindex_t newpages, oldpages; 924 struct tmpfs_mount *tmp; 925 struct tmpfs_node *node; 926 off_t oldsize; 927 928 #ifdef INVARIANTS 929 KKASSERT(vp->v_type == VREG); 930 KKASSERT(newsize >= 0); 931 #endif 932 933 node = VP_TO_TMPFS_NODE(vp); 934 tmp = VFS_TO_TMPFS(vp->v_mount); 935 936 /* 937 * Convert the old and new sizes to the number of pages needed to 938 * store them. It may happen that we do not need to do anything 939 * because the last allocated page can accommodate the change on 940 * its own. 941 */ 942 oldsize = node->tn_size; 943 oldpages = round_page64(oldsize) / PAGE_SIZE; 944 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 945 newpages = round_page64(newsize) / PAGE_SIZE; 946 947 if (newpages > oldpages && 948 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 949 error = ENOSPC; 950 goto out; 951 } 952 node->tn_reg.tn_aobj_pages = newpages; 953 node->tn_size = newsize; 954 955 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages)); 956 957 /* 958 * When adjusting the vnode filesize and its VM object we must 959 * also adjust our backing VM object (aobj). The blocksize 960 * used must match the block sized we use for the buffer cache. 961 * 962 * The backing VM object contains no VM pages, only swap 963 * assignments. 964 */ 965 if (newsize < oldsize) { 966 vm_pindex_t osize; 967 vm_pindex_t nsize; 968 vm_object_t aobj; 969 970 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0); 971 aobj = node->tn_reg.tn_aobj; 972 if (aobj) { 973 osize = aobj->size; 974 nsize = vp->v_object->size; 975 if (nsize < osize) { 976 aobj->size = osize; 977 swap_pager_freespace(aobj, nsize, 978 osize - nsize); 979 } 980 } 981 } else { 982 vm_object_t aobj; 983 984 error = nvextendbuf(vp, oldsize, newsize, 985 TMPFS_BLKSIZE, TMPFS_BLKSIZE, 986 -1, -1, trivial); 987 aobj = node->tn_reg.tn_aobj; 988 if (aobj) 989 aobj->size = vp->v_object->size; 990 } 991 992 out: 993 return error; 994 } 995 996 /* --------------------------------------------------------------------- */ 997 998 /* 999 * Change flags of the given vnode. 1000 * Caller should execute tmpfs_update on vp after a successful execution. 1001 * The vnode must be locked on entry and remain locked on exit. 1002 */ 1003 int 1004 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred) 1005 { 1006 int error; 1007 struct tmpfs_node *node; 1008 int flags; 1009 1010 KKASSERT(vn_islocked(vp)); 1011 1012 node = VP_TO_TMPFS_NODE(vp); 1013 flags = node->tn_flags; 1014 1015 /* Disallow this operation if the file system is mounted read-only. */ 1016 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1017 return EROFS; 1018 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1019 1020 /* Actually change the flags on the node itself */ 1021 if (error == 0) { 1022 TMPFS_NODE_LOCK(node); 1023 node->tn_flags = flags; 1024 node->tn_status |= TMPFS_NODE_CHANGED; 1025 TMPFS_NODE_UNLOCK(node); 1026 } 1027 1028 KKASSERT(vn_islocked(vp)); 1029 1030 return error; 1031 } 1032 1033 /* --------------------------------------------------------------------- */ 1034 1035 /* 1036 * Change access mode on the given vnode. 1037 * Caller should execute tmpfs_update on vp after a successful execution. 1038 * The vnode must be locked on entry and remain locked on exit. 1039 */ 1040 int 1041 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1042 { 1043 struct tmpfs_node *node; 1044 mode_t cur_mode; 1045 int error; 1046 1047 KKASSERT(vn_islocked(vp)); 1048 1049 node = VP_TO_TMPFS_NODE(vp); 1050 1051 /* Disallow this operation if the file system is mounted read-only. */ 1052 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1053 return EROFS; 1054 1055 /* Immutable or append-only files cannot be modified, either. */ 1056 if (node->tn_flags & (IMMUTABLE | APPEND)) 1057 return EPERM; 1058 1059 cur_mode = node->tn_mode; 1060 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1061 &cur_mode); 1062 1063 if (error == 0 && 1064 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1065 TMPFS_NODE_LOCK(node); 1066 node->tn_mode &= ~ALLPERMS; 1067 node->tn_mode |= cur_mode & ALLPERMS; 1068 1069 node->tn_status |= TMPFS_NODE_CHANGED; 1070 TMPFS_NODE_UNLOCK(node); 1071 } 1072 1073 KKASSERT(vn_islocked(vp)); 1074 1075 return 0; 1076 } 1077 1078 /* --------------------------------------------------------------------- */ 1079 1080 /* 1081 * Change ownership of the given vnode. At least one of uid or gid must 1082 * be different than VNOVAL. If one is set to that value, the attribute 1083 * is unchanged. 1084 * Caller should execute tmpfs_update on vp after a successful execution. 1085 * The vnode must be locked on entry and remain locked on exit. 1086 */ 1087 int 1088 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1089 { 1090 mode_t cur_mode; 1091 uid_t cur_uid; 1092 gid_t cur_gid; 1093 struct tmpfs_node *node; 1094 int error; 1095 1096 KKASSERT(vn_islocked(vp)); 1097 node = VP_TO_TMPFS_NODE(vp); 1098 1099 /* Disallow this operation if the file system is mounted read-only. */ 1100 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1101 return EROFS; 1102 1103 /* Immutable or append-only files cannot be modified, either. */ 1104 if (node->tn_flags & (IMMUTABLE | APPEND)) 1105 return EPERM; 1106 1107 cur_uid = node->tn_uid; 1108 cur_gid = node->tn_gid; 1109 cur_mode = node->tn_mode; 1110 error = vop_helper_chown(vp, uid, gid, cred, 1111 &cur_uid, &cur_gid, &cur_mode); 1112 1113 if (error == 0) { 1114 TMPFS_NODE_LOCK(node); 1115 if (cur_uid != node->tn_uid || 1116 cur_gid != node->tn_gid || 1117 cur_mode != node->tn_mode) { 1118 node->tn_uid = cur_uid; 1119 node->tn_gid = cur_gid; 1120 node->tn_mode = cur_mode; 1121 node->tn_status |= TMPFS_NODE_CHANGED; 1122 } 1123 TMPFS_NODE_UNLOCK(node); 1124 } 1125 1126 return error; 1127 } 1128 1129 /* --------------------------------------------------------------------- */ 1130 1131 /* 1132 * Change size of the given vnode. 1133 * Caller should execute tmpfs_update on vp after a successful execution. 1134 * The vnode must be locked on entry and remain locked on exit. 1135 */ 1136 int 1137 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1138 { 1139 int error; 1140 struct tmpfs_node *node; 1141 1142 KKASSERT(vn_islocked(vp)); 1143 1144 node = VP_TO_TMPFS_NODE(vp); 1145 1146 /* Decide whether this is a valid operation based on the file type. */ 1147 error = 0; 1148 switch (vp->v_type) { 1149 case VDIR: 1150 return EISDIR; 1151 1152 case VREG: 1153 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1154 return EROFS; 1155 break; 1156 1157 case VBLK: 1158 /* FALLTHROUGH */ 1159 case VCHR: 1160 /* FALLTHROUGH */ 1161 case VFIFO: 1162 /* Allow modifications of special files even if in the file 1163 * system is mounted read-only (we are not modifying the 1164 * files themselves, but the objects they represent). */ 1165 return 0; 1166 1167 default: 1168 /* Anything else is unsupported. */ 1169 return EOPNOTSUPP; 1170 } 1171 1172 /* Immutable or append-only files cannot be modified, either. */ 1173 if (node->tn_flags & (IMMUTABLE | APPEND)) 1174 return EPERM; 1175 1176 error = tmpfs_truncate(vp, size); 1177 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1178 * for us, as will update tn_status; no need to do that here. */ 1179 1180 KKASSERT(vn_islocked(vp)); 1181 1182 return error; 1183 } 1184 1185 /* --------------------------------------------------------------------- */ 1186 1187 /* 1188 * Change access and modification times of the given vnode. 1189 * Caller should execute tmpfs_update on vp after a successful execution. 1190 * The vnode must be locked on entry and remain locked on exit. 1191 */ 1192 int 1193 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1194 int vaflags, struct ucred *cred) 1195 { 1196 struct tmpfs_node *node; 1197 1198 KKASSERT(vn_islocked(vp)); 1199 1200 node = VP_TO_TMPFS_NODE(vp); 1201 1202 /* Disallow this operation if the file system is mounted read-only. */ 1203 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1204 return EROFS; 1205 1206 /* Immutable or append-only files cannot be modified, either. */ 1207 if (node->tn_flags & (IMMUTABLE | APPEND)) 1208 return EPERM; 1209 1210 TMPFS_NODE_LOCK(node); 1211 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1212 node->tn_status |= TMPFS_NODE_ACCESSED; 1213 1214 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1215 node->tn_status |= TMPFS_NODE_MODIFIED; 1216 1217 TMPFS_NODE_UNLOCK(node); 1218 1219 tmpfs_itimes(vp, atime, mtime); 1220 1221 KKASSERT(vn_islocked(vp)); 1222 1223 return 0; 1224 } 1225 1226 /* --------------------------------------------------------------------- */ 1227 /* Sync timestamps */ 1228 void 1229 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1230 const struct timespec *mod) 1231 { 1232 struct tmpfs_node *node; 1233 struct timespec now; 1234 1235 node = VP_TO_TMPFS_NODE(vp); 1236 1237 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1238 TMPFS_NODE_CHANGED)) == 0) 1239 return; 1240 1241 vfs_timestamp(&now); 1242 1243 TMPFS_NODE_LOCK(node); 1244 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1245 if (acc == NULL) 1246 acc = &now; 1247 node->tn_atime = acc->tv_sec; 1248 node->tn_atimensec = acc->tv_nsec; 1249 } 1250 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1251 if (mod == NULL) 1252 mod = &now; 1253 node->tn_mtime = mod->tv_sec; 1254 node->tn_mtimensec = mod->tv_nsec; 1255 } 1256 if (node->tn_status & TMPFS_NODE_CHANGED) { 1257 node->tn_ctime = now.tv_sec; 1258 node->tn_ctimensec = now.tv_nsec; 1259 } 1260 node->tn_status &= 1261 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1262 TMPFS_NODE_UNLOCK(node); 1263 } 1264 1265 /* --------------------------------------------------------------------- */ 1266 1267 void 1268 tmpfs_update(struct vnode *vp) 1269 { 1270 tmpfs_itimes(vp, NULL, NULL); 1271 } 1272 1273 /* --------------------------------------------------------------------- */ 1274 1275 /* 1276 * Caller must hold an exclusive node lock. 1277 */ 1278 int 1279 tmpfs_truncate(struct vnode *vp, off_t length) 1280 { 1281 int error; 1282 struct tmpfs_node *node; 1283 1284 node = VP_TO_TMPFS_NODE(vp); 1285 1286 if (length < 0) { 1287 error = EINVAL; 1288 goto out; 1289 } 1290 1291 if (node->tn_size == length) { 1292 error = 0; 1293 goto out; 1294 } 1295 1296 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1297 return (EFBIG); 1298 1299 1300 error = tmpfs_reg_resize(vp, length, 1); 1301 1302 if (error == 0) 1303 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1304 1305 out: 1306 tmpfs_update(vp); 1307 1308 return error; 1309 } 1310 1311 /* --------------------------------------------------------------------- */ 1312 1313 static ino_t 1314 tmpfs_fetch_ino(struct tmpfs_mount *tmp) 1315 { 1316 ino_t ret; 1317 1318 TMPFS_LOCK(tmp); 1319 ret = tmp->tm_ino++; 1320 TMPFS_UNLOCK(tmp); 1321 1322 return (ret); 1323 } 1324 1325 static int 1326 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1327 { 1328 if (a->td_namelen > b->td_namelen) 1329 return 1; 1330 else if (a->td_namelen < b->td_namelen) 1331 return -1; 1332 else 1333 return strncmp(a->td_name, b->td_name, a->td_namelen); 1334 } 1335 1336 static int 1337 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1338 { 1339 if (a < b) 1340 return(-1); 1341 if (a > b) 1342 return(1); 1343 return 0; 1344 } 1345