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