1 /* 2 * This file is part of UBIFS. 3 * 4 * Copyright (C) 2006-2008 Nokia Corporation. 5 * 6 * This program is free software; you can redistribute it and/or modify it 7 * under the terms of the GNU General Public License version 2 as published by 8 * the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it will be useful, but WITHOUT 11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 13 * more details. 14 * 15 * You should have received a copy of the GNU General Public License along with 16 * this program; if not, write to the Free Software Foundation, Inc., 51 17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 18 * 19 * Authors: Adrian Hunter 20 * Artem Bityutskiy (Битюцкий Артём) 21 */ 22 23 /* 24 * This file implements TNC (Tree Node Cache) which caches indexing nodes of 25 * the UBIFS B-tree. 26 * 27 * At the moment the locking rules of the TNC tree are quite simple and 28 * straightforward. We just have a mutex and lock it when we traverse the 29 * tree. If a znode is not in memory, we read it from flash while still having 30 * the mutex locked. 31 */ 32 33 #include <linux/crc32.h> 34 #include <linux/slab.h> 35 #include "ubifs.h" 36 37 /* 38 * Returned codes of 'matches_name()' and 'fallible_matches_name()' functions. 39 * @NAME_LESS: name corresponding to the first argument is less than second 40 * @NAME_MATCHES: names match 41 * @NAME_GREATER: name corresponding to the second argument is greater than 42 * first 43 * @NOT_ON_MEDIA: node referred by zbranch does not exist on the media 44 * 45 * These constants were introduce to improve readability. 46 */ 47 enum { 48 NAME_LESS = 0, 49 NAME_MATCHES = 1, 50 NAME_GREATER = 2, 51 NOT_ON_MEDIA = 3, 52 }; 53 54 /** 55 * insert_old_idx - record an index node obsoleted since the last commit start. 56 * @c: UBIFS file-system description object 57 * @lnum: LEB number of obsoleted index node 58 * @offs: offset of obsoleted index node 59 * 60 * Returns %0 on success, and a negative error code on failure. 61 * 62 * For recovery, there must always be a complete intact version of the index on 63 * flash at all times. That is called the "old index". It is the index as at the 64 * time of the last successful commit. Many of the index nodes in the old index 65 * may be dirty, but they must not be erased until the next successful commit 66 * (at which point that index becomes the old index). 67 * 68 * That means that the garbage collection and the in-the-gaps method of 69 * committing must be able to determine if an index node is in the old index. 70 * Most of the old index nodes can be found by looking up the TNC using the 71 * 'lookup_znode()' function. However, some of the old index nodes may have 72 * been deleted from the current index or may have been changed so much that 73 * they cannot be easily found. In those cases, an entry is added to an RB-tree. 74 * That is what this function does. The RB-tree is ordered by LEB number and 75 * offset because they uniquely identify the old index node. 76 */ 77 static int insert_old_idx(struct ubifs_info *c, int lnum, int offs) 78 { 79 struct ubifs_old_idx *old_idx, *o; 80 struct rb_node **p, *parent = NULL; 81 82 old_idx = kmalloc(sizeof(struct ubifs_old_idx), GFP_NOFS); 83 if (unlikely(!old_idx)) 84 return -ENOMEM; 85 old_idx->lnum = lnum; 86 old_idx->offs = offs; 87 88 p = &c->old_idx.rb_node; 89 while (*p) { 90 parent = *p; 91 o = rb_entry(parent, struct ubifs_old_idx, rb); 92 if (lnum < o->lnum) 93 p = &(*p)->rb_left; 94 else if (lnum > o->lnum) 95 p = &(*p)->rb_right; 96 else if (offs < o->offs) 97 p = &(*p)->rb_left; 98 else if (offs > o->offs) 99 p = &(*p)->rb_right; 100 else { 101 ubifs_err(c, "old idx added twice!"); 102 kfree(old_idx); 103 return 0; 104 } 105 } 106 rb_link_node(&old_idx->rb, parent, p); 107 rb_insert_color(&old_idx->rb, &c->old_idx); 108 return 0; 109 } 110 111 /** 112 * insert_old_idx_znode - record a znode obsoleted since last commit start. 113 * @c: UBIFS file-system description object 114 * @znode: znode of obsoleted index node 115 * 116 * Returns %0 on success, and a negative error code on failure. 117 */ 118 int insert_old_idx_znode(struct ubifs_info *c, struct ubifs_znode *znode) 119 { 120 if (znode->parent) { 121 struct ubifs_zbranch *zbr; 122 123 zbr = &znode->parent->zbranch[znode->iip]; 124 if (zbr->len) 125 return insert_old_idx(c, zbr->lnum, zbr->offs); 126 } else 127 if (c->zroot.len) 128 return insert_old_idx(c, c->zroot.lnum, 129 c->zroot.offs); 130 return 0; 131 } 132 133 /** 134 * ins_clr_old_idx_znode - record a znode obsoleted since last commit start. 135 * @c: UBIFS file-system description object 136 * @znode: znode of obsoleted index node 137 * 138 * Returns %0 on success, and a negative error code on failure. 139 */ 140 static int ins_clr_old_idx_znode(struct ubifs_info *c, 141 struct ubifs_znode *znode) 142 { 143 int err; 144 145 if (znode->parent) { 146 struct ubifs_zbranch *zbr; 147 148 zbr = &znode->parent->zbranch[znode->iip]; 149 if (zbr->len) { 150 err = insert_old_idx(c, zbr->lnum, zbr->offs); 151 if (err) 152 return err; 153 zbr->lnum = 0; 154 zbr->offs = 0; 155 zbr->len = 0; 156 } 157 } else 158 if (c->zroot.len) { 159 err = insert_old_idx(c, c->zroot.lnum, c->zroot.offs); 160 if (err) 161 return err; 162 c->zroot.lnum = 0; 163 c->zroot.offs = 0; 164 c->zroot.len = 0; 165 } 166 return 0; 167 } 168 169 /** 170 * destroy_old_idx - destroy the old_idx RB-tree. 171 * @c: UBIFS file-system description object 172 * 173 * During start commit, the old_idx RB-tree is used to avoid overwriting index 174 * nodes that were in the index last commit but have since been deleted. This 175 * is necessary for recovery i.e. the old index must be kept intact until the 176 * new index is successfully written. The old-idx RB-tree is used for the 177 * in-the-gaps method of writing index nodes and is destroyed every commit. 178 */ 179 void destroy_old_idx(struct ubifs_info *c) 180 { 181 struct ubifs_old_idx *old_idx, *n; 182 183 rbtree_postorder_for_each_entry_safe(old_idx, n, &c->old_idx, rb) 184 kfree(old_idx); 185 186 c->old_idx = RB_ROOT; 187 } 188 189 /** 190 * copy_znode - copy a dirty znode. 191 * @c: UBIFS file-system description object 192 * @znode: znode to copy 193 * 194 * A dirty znode being committed may not be changed, so it is copied. 195 */ 196 static struct ubifs_znode *copy_znode(struct ubifs_info *c, 197 struct ubifs_znode *znode) 198 { 199 struct ubifs_znode *zn; 200 201 zn = kmemdup(znode, c->max_znode_sz, GFP_NOFS); 202 if (unlikely(!zn)) 203 return ERR_PTR(-ENOMEM); 204 205 zn->cnext = NULL; 206 __set_bit(DIRTY_ZNODE, &zn->flags); 207 __clear_bit(COW_ZNODE, &zn->flags); 208 209 ubifs_assert(!ubifs_zn_obsolete(znode)); 210 __set_bit(OBSOLETE_ZNODE, &znode->flags); 211 212 if (znode->level != 0) { 213 int i; 214 const int n = zn->child_cnt; 215 216 /* The children now have new parent */ 217 for (i = 0; i < n; i++) { 218 struct ubifs_zbranch *zbr = &zn->zbranch[i]; 219 220 if (zbr->znode) 221 zbr->znode->parent = zn; 222 } 223 } 224 225 atomic_long_inc(&c->dirty_zn_cnt); 226 return zn; 227 } 228 229 /** 230 * add_idx_dirt - add dirt due to a dirty znode. 231 * @c: UBIFS file-system description object 232 * @lnum: LEB number of index node 233 * @dirt: size of index node 234 * 235 * This function updates lprops dirty space and the new size of the index. 236 */ 237 static int add_idx_dirt(struct ubifs_info *c, int lnum, int dirt) 238 { 239 c->calc_idx_sz -= ALIGN(dirt, 8); 240 return ubifs_add_dirt(c, lnum, dirt); 241 } 242 243 /** 244 * dirty_cow_znode - ensure a znode is not being committed. 245 * @c: UBIFS file-system description object 246 * @zbr: branch of znode to check 247 * 248 * Returns dirtied znode on success or negative error code on failure. 249 */ 250 static struct ubifs_znode *dirty_cow_znode(struct ubifs_info *c, 251 struct ubifs_zbranch *zbr) 252 { 253 struct ubifs_znode *znode = zbr->znode; 254 struct ubifs_znode *zn; 255 int err; 256 257 if (!ubifs_zn_cow(znode)) { 258 /* znode is not being committed */ 259 if (!test_and_set_bit(DIRTY_ZNODE, &znode->flags)) { 260 atomic_long_inc(&c->dirty_zn_cnt); 261 atomic_long_dec(&c->clean_zn_cnt); 262 atomic_long_dec(&ubifs_clean_zn_cnt); 263 err = add_idx_dirt(c, zbr->lnum, zbr->len); 264 if (unlikely(err)) 265 return ERR_PTR(err); 266 } 267 return znode; 268 } 269 270 zn = copy_znode(c, znode); 271 if (IS_ERR(zn)) 272 return zn; 273 274 if (zbr->len) { 275 err = insert_old_idx(c, zbr->lnum, zbr->offs); 276 if (unlikely(err)) 277 return ERR_PTR(err); 278 err = add_idx_dirt(c, zbr->lnum, zbr->len); 279 } else 280 err = 0; 281 282 zbr->znode = zn; 283 zbr->lnum = 0; 284 zbr->offs = 0; 285 zbr->len = 0; 286 287 if (unlikely(err)) 288 return ERR_PTR(err); 289 return zn; 290 } 291 292 /** 293 * lnc_add - add a leaf node to the leaf node cache. 294 * @c: UBIFS file-system description object 295 * @zbr: zbranch of leaf node 296 * @node: leaf node 297 * 298 * Leaf nodes are non-index nodes directory entry nodes or data nodes. The 299 * purpose of the leaf node cache is to save re-reading the same leaf node over 300 * and over again. Most things are cached by VFS, however the file system must 301 * cache directory entries for readdir and for resolving hash collisions. The 302 * present implementation of the leaf node cache is extremely simple, and 303 * allows for error returns that are not used but that may be needed if a more 304 * complex implementation is created. 305 * 306 * Note, this function does not add the @node object to LNC directly, but 307 * allocates a copy of the object and adds the copy to LNC. The reason for this 308 * is that @node has been allocated outside of the TNC subsystem and will be 309 * used with @c->tnc_mutex unlock upon return from the TNC subsystem. But LNC 310 * may be changed at any time, e.g. freed by the shrinker. 311 */ 312 static int lnc_add(struct ubifs_info *c, struct ubifs_zbranch *zbr, 313 const void *node) 314 { 315 int err; 316 void *lnc_node; 317 const struct ubifs_dent_node *dent = node; 318 319 ubifs_assert(!zbr->leaf); 320 ubifs_assert(zbr->len != 0); 321 ubifs_assert(is_hash_key(c, &zbr->key)); 322 323 err = ubifs_validate_entry(c, dent); 324 if (err) { 325 dump_stack(); 326 ubifs_dump_node(c, dent); 327 return err; 328 } 329 330 lnc_node = kmemdup(node, zbr->len, GFP_NOFS); 331 if (!lnc_node) 332 /* We don't have to have the cache, so no error */ 333 return 0; 334 335 zbr->leaf = lnc_node; 336 return 0; 337 } 338 339 /** 340 * lnc_add_directly - add a leaf node to the leaf-node-cache. 341 * @c: UBIFS file-system description object 342 * @zbr: zbranch of leaf node 343 * @node: leaf node 344 * 345 * This function is similar to 'lnc_add()', but it does not create a copy of 346 * @node but inserts @node to TNC directly. 347 */ 348 static int lnc_add_directly(struct ubifs_info *c, struct ubifs_zbranch *zbr, 349 void *node) 350 { 351 int err; 352 353 ubifs_assert(!zbr->leaf); 354 ubifs_assert(zbr->len != 0); 355 356 err = ubifs_validate_entry(c, node); 357 if (err) { 358 dump_stack(); 359 ubifs_dump_node(c, node); 360 return err; 361 } 362 363 zbr->leaf = node; 364 return 0; 365 } 366 367 /** 368 * lnc_free - remove a leaf node from the leaf node cache. 369 * @zbr: zbranch of leaf node 370 * @node: leaf node 371 */ 372 static void lnc_free(struct ubifs_zbranch *zbr) 373 { 374 if (!zbr->leaf) 375 return; 376 kfree(zbr->leaf); 377 zbr->leaf = NULL; 378 } 379 380 /** 381 * tnc_read_hashed_node - read a "hashed" leaf node. 382 * @c: UBIFS file-system description object 383 * @zbr: key and position of the node 384 * @node: node is returned here 385 * 386 * This function reads a "hashed" node defined by @zbr from the leaf node cache 387 * (in it is there) or from the hash media, in which case the node is also 388 * added to LNC. Returns zero in case of success or a negative negative error 389 * code in case of failure. 390 */ 391 static int tnc_read_hashed_node(struct ubifs_info *c, struct ubifs_zbranch *zbr, 392 void *node) 393 { 394 int err; 395 396 ubifs_assert(is_hash_key(c, &zbr->key)); 397 398 if (zbr->leaf) { 399 /* Read from the leaf node cache */ 400 ubifs_assert(zbr->len != 0); 401 memcpy(node, zbr->leaf, zbr->len); 402 return 0; 403 } 404 405 err = ubifs_tnc_read_node(c, zbr, node); 406 if (err) 407 return err; 408 409 /* Add the node to the leaf node cache */ 410 err = lnc_add(c, zbr, node); 411 return err; 412 } 413 414 /** 415 * try_read_node - read a node if it is a node. 416 * @c: UBIFS file-system description object 417 * @buf: buffer to read to 418 * @type: node type 419 * @len: node length (not aligned) 420 * @lnum: LEB number of node to read 421 * @offs: offset of node to read 422 * 423 * This function tries to read a node of known type and length, checks it and 424 * stores it in @buf. This function returns %1 if a node is present and %0 if 425 * a node is not present. A negative error code is returned for I/O errors. 426 * This function performs that same function as ubifs_read_node except that 427 * it does not require that there is actually a node present and instead 428 * the return code indicates if a node was read. 429 * 430 * Note, this function does not check CRC of data nodes if @c->no_chk_data_crc 431 * is true (it is controlled by corresponding mount option). However, if 432 * @c->mounting or @c->remounting_rw is true (we are mounting or re-mounting to 433 * R/W mode), @c->no_chk_data_crc is ignored and CRC is checked. This is 434 * because during mounting or re-mounting from R/O mode to R/W mode we may read 435 * journal nodes (when replying the journal or doing the recovery) and the 436 * journal nodes may potentially be corrupted, so checking is required. 437 */ 438 static int try_read_node(const struct ubifs_info *c, void *buf, int type, 439 int len, int lnum, int offs) 440 { 441 int err, node_len; 442 struct ubifs_ch *ch = buf; 443 uint32_t crc, node_crc; 444 445 dbg_io("LEB %d:%d, %s, length %d", lnum, offs, dbg_ntype(type), len); 446 447 err = ubifs_leb_read(c, lnum, buf, offs, len, 1); 448 if (err) { 449 ubifs_err(c, "cannot read node type %d from LEB %d:%d, error %d", 450 type, lnum, offs, err); 451 return err; 452 } 453 454 if (le32_to_cpu(ch->magic) != UBIFS_NODE_MAGIC) 455 return 0; 456 457 if (ch->node_type != type) 458 return 0; 459 460 node_len = le32_to_cpu(ch->len); 461 if (node_len != len) 462 return 0; 463 464 if (type == UBIFS_DATA_NODE && c->no_chk_data_crc && !c->mounting && 465 !c->remounting_rw) 466 return 1; 467 468 crc = crc32(UBIFS_CRC32_INIT, buf + 8, node_len - 8); 469 node_crc = le32_to_cpu(ch->crc); 470 if (crc != node_crc) 471 return 0; 472 473 return 1; 474 } 475 476 /** 477 * fallible_read_node - try to read a leaf node. 478 * @c: UBIFS file-system description object 479 * @key: key of node to read 480 * @zbr: position of node 481 * @node: node returned 482 * 483 * This function tries to read a node and returns %1 if the node is read, %0 484 * if the node is not present, and a negative error code in the case of error. 485 */ 486 static int fallible_read_node(struct ubifs_info *c, const union ubifs_key *key, 487 struct ubifs_zbranch *zbr, void *node) 488 { 489 int ret; 490 491 dbg_tnck(key, "LEB %d:%d, key ", zbr->lnum, zbr->offs); 492 493 ret = try_read_node(c, node, key_type(c, key), zbr->len, zbr->lnum, 494 zbr->offs); 495 if (ret == 1) { 496 union ubifs_key node_key; 497 struct ubifs_dent_node *dent = node; 498 499 /* All nodes have key in the same place */ 500 key_read(c, &dent->key, &node_key); 501 if (keys_cmp(c, key, &node_key) != 0) 502 ret = 0; 503 } 504 if (ret == 0 && c->replaying) 505 dbg_mntk(key, "dangling branch LEB %d:%d len %d, key ", 506 zbr->lnum, zbr->offs, zbr->len); 507 return ret; 508 } 509 510 /** 511 * matches_name - determine if a direntry or xattr entry matches a given name. 512 * @c: UBIFS file-system description object 513 * @zbr: zbranch of dent 514 * @nm: name to match 515 * 516 * This function checks if xentry/direntry referred by zbranch @zbr matches name 517 * @nm. Returns %NAME_MATCHES if it does, %NAME_LESS if the name referred by 518 * @zbr is less than @nm, and %NAME_GREATER if it is greater than @nm. In case 519 * of failure, a negative error code is returned. 520 */ 521 static int matches_name(struct ubifs_info *c, struct ubifs_zbranch *zbr, 522 const struct fscrypt_name *nm) 523 { 524 struct ubifs_dent_node *dent; 525 int nlen, err; 526 527 /* If possible, match against the dent in the leaf node cache */ 528 if (!zbr->leaf) { 529 dent = kmalloc(zbr->len, GFP_NOFS); 530 if (!dent) 531 return -ENOMEM; 532 533 err = ubifs_tnc_read_node(c, zbr, dent); 534 if (err) 535 goto out_free; 536 537 /* Add the node to the leaf node cache */ 538 err = lnc_add_directly(c, zbr, dent); 539 if (err) 540 goto out_free; 541 } else 542 dent = zbr->leaf; 543 544 nlen = le16_to_cpu(dent->nlen); 545 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm))); 546 if (err == 0) { 547 if (nlen == fname_len(nm)) 548 return NAME_MATCHES; 549 else if (nlen < fname_len(nm)) 550 return NAME_LESS; 551 else 552 return NAME_GREATER; 553 } else if (err < 0) 554 return NAME_LESS; 555 else 556 return NAME_GREATER; 557 558 out_free: 559 kfree(dent); 560 return err; 561 } 562 563 /** 564 * get_znode - get a TNC znode that may not be loaded yet. 565 * @c: UBIFS file-system description object 566 * @znode: parent znode 567 * @n: znode branch slot number 568 * 569 * This function returns the znode or a negative error code. 570 */ 571 static struct ubifs_znode *get_znode(struct ubifs_info *c, 572 struct ubifs_znode *znode, int n) 573 { 574 struct ubifs_zbranch *zbr; 575 576 zbr = &znode->zbranch[n]; 577 if (zbr->znode) 578 znode = zbr->znode; 579 else 580 znode = ubifs_load_znode(c, zbr, znode, n); 581 return znode; 582 } 583 584 /** 585 * tnc_next - find next TNC entry. 586 * @c: UBIFS file-system description object 587 * @zn: znode is passed and returned here 588 * @n: znode branch slot number is passed and returned here 589 * 590 * This function returns %0 if the next TNC entry is found, %-ENOENT if there is 591 * no next entry, or a negative error code otherwise. 592 */ 593 static int tnc_next(struct ubifs_info *c, struct ubifs_znode **zn, int *n) 594 { 595 struct ubifs_znode *znode = *zn; 596 int nn = *n; 597 598 nn += 1; 599 if (nn < znode->child_cnt) { 600 *n = nn; 601 return 0; 602 } 603 while (1) { 604 struct ubifs_znode *zp; 605 606 zp = znode->parent; 607 if (!zp) 608 return -ENOENT; 609 nn = znode->iip + 1; 610 znode = zp; 611 if (nn < znode->child_cnt) { 612 znode = get_znode(c, znode, nn); 613 if (IS_ERR(znode)) 614 return PTR_ERR(znode); 615 while (znode->level != 0) { 616 znode = get_znode(c, znode, 0); 617 if (IS_ERR(znode)) 618 return PTR_ERR(znode); 619 } 620 nn = 0; 621 break; 622 } 623 } 624 *zn = znode; 625 *n = nn; 626 return 0; 627 } 628 629 /** 630 * tnc_prev - find previous TNC entry. 631 * @c: UBIFS file-system description object 632 * @zn: znode is returned here 633 * @n: znode branch slot number is passed and returned here 634 * 635 * This function returns %0 if the previous TNC entry is found, %-ENOENT if 636 * there is no next entry, or a negative error code otherwise. 637 */ 638 static int tnc_prev(struct ubifs_info *c, struct ubifs_znode **zn, int *n) 639 { 640 struct ubifs_znode *znode = *zn; 641 int nn = *n; 642 643 if (nn > 0) { 644 *n = nn - 1; 645 return 0; 646 } 647 while (1) { 648 struct ubifs_znode *zp; 649 650 zp = znode->parent; 651 if (!zp) 652 return -ENOENT; 653 nn = znode->iip - 1; 654 znode = zp; 655 if (nn >= 0) { 656 znode = get_znode(c, znode, nn); 657 if (IS_ERR(znode)) 658 return PTR_ERR(znode); 659 while (znode->level != 0) { 660 nn = znode->child_cnt - 1; 661 znode = get_znode(c, znode, nn); 662 if (IS_ERR(znode)) 663 return PTR_ERR(znode); 664 } 665 nn = znode->child_cnt - 1; 666 break; 667 } 668 } 669 *zn = znode; 670 *n = nn; 671 return 0; 672 } 673 674 /** 675 * resolve_collision - resolve a collision. 676 * @c: UBIFS file-system description object 677 * @key: key of a directory or extended attribute entry 678 * @zn: znode is returned here 679 * @n: zbranch number is passed and returned here 680 * @nm: name of the entry 681 * 682 * This function is called for "hashed" keys to make sure that the found key 683 * really corresponds to the looked up node (directory or extended attribute 684 * entry). It returns %1 and sets @zn and @n if the collision is resolved. 685 * %0 is returned if @nm is not found and @zn and @n are set to the previous 686 * entry, i.e. to the entry after which @nm could follow if it were in TNC. 687 * This means that @n may be set to %-1 if the leftmost key in @zn is the 688 * previous one. A negative error code is returned on failures. 689 */ 690 static int resolve_collision(struct ubifs_info *c, const union ubifs_key *key, 691 struct ubifs_znode **zn, int *n, 692 const struct fscrypt_name *nm) 693 { 694 int err; 695 696 err = matches_name(c, &(*zn)->zbranch[*n], nm); 697 if (unlikely(err < 0)) 698 return err; 699 if (err == NAME_MATCHES) 700 return 1; 701 702 if (err == NAME_GREATER) { 703 /* Look left */ 704 while (1) { 705 err = tnc_prev(c, zn, n); 706 if (err == -ENOENT) { 707 ubifs_assert(*n == 0); 708 *n = -1; 709 return 0; 710 } 711 if (err < 0) 712 return err; 713 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { 714 /* 715 * We have found the branch after which we would 716 * like to insert, but inserting in this znode 717 * may still be wrong. Consider the following 3 718 * znodes, in the case where we are resolving a 719 * collision with Key2. 720 * 721 * znode zp 722 * ---------------------- 723 * level 1 | Key0 | Key1 | 724 * ----------------------- 725 * | | 726 * znode za | | znode zb 727 * ------------ ------------ 728 * level 0 | Key0 | | Key2 | 729 * ------------ ------------ 730 * 731 * The lookup finds Key2 in znode zb. Lets say 732 * there is no match and the name is greater so 733 * we look left. When we find Key0, we end up 734 * here. If we return now, we will insert into 735 * znode za at slot n = 1. But that is invalid 736 * according to the parent's keys. Key2 must 737 * be inserted into znode zb. 738 * 739 * Note, this problem is not relevant for the 740 * case when we go right, because 741 * 'tnc_insert()' would correct the parent key. 742 */ 743 if (*n == (*zn)->child_cnt - 1) { 744 err = tnc_next(c, zn, n); 745 if (err) { 746 /* Should be impossible */ 747 ubifs_assert(0); 748 if (err == -ENOENT) 749 err = -EINVAL; 750 return err; 751 } 752 ubifs_assert(*n == 0); 753 *n = -1; 754 } 755 return 0; 756 } 757 err = matches_name(c, &(*zn)->zbranch[*n], nm); 758 if (err < 0) 759 return err; 760 if (err == NAME_LESS) 761 return 0; 762 if (err == NAME_MATCHES) 763 return 1; 764 ubifs_assert(err == NAME_GREATER); 765 } 766 } else { 767 int nn = *n; 768 struct ubifs_znode *znode = *zn; 769 770 /* Look right */ 771 while (1) { 772 err = tnc_next(c, &znode, &nn); 773 if (err == -ENOENT) 774 return 0; 775 if (err < 0) 776 return err; 777 if (keys_cmp(c, &znode->zbranch[nn].key, key)) 778 return 0; 779 err = matches_name(c, &znode->zbranch[nn], nm); 780 if (err < 0) 781 return err; 782 if (err == NAME_GREATER) 783 return 0; 784 *zn = znode; 785 *n = nn; 786 if (err == NAME_MATCHES) 787 return 1; 788 ubifs_assert(err == NAME_LESS); 789 } 790 } 791 } 792 793 /** 794 * fallible_matches_name - determine if a dent matches a given name. 795 * @c: UBIFS file-system description object 796 * @zbr: zbranch of dent 797 * @nm: name to match 798 * 799 * This is a "fallible" version of 'matches_name()' function which does not 800 * panic if the direntry/xentry referred by @zbr does not exist on the media. 801 * 802 * This function checks if xentry/direntry referred by zbranch @zbr matches name 803 * @nm. Returns %NAME_MATCHES it does, %NAME_LESS if the name referred by @zbr 804 * is less than @nm, %NAME_GREATER if it is greater than @nm, and @NOT_ON_MEDIA 805 * if xentry/direntry referred by @zbr does not exist on the media. A negative 806 * error code is returned in case of failure. 807 */ 808 static int fallible_matches_name(struct ubifs_info *c, 809 struct ubifs_zbranch *zbr, 810 const struct fscrypt_name *nm) 811 { 812 struct ubifs_dent_node *dent; 813 int nlen, err; 814 815 /* If possible, match against the dent in the leaf node cache */ 816 if (!zbr->leaf) { 817 dent = kmalloc(zbr->len, GFP_NOFS); 818 if (!dent) 819 return -ENOMEM; 820 821 err = fallible_read_node(c, &zbr->key, zbr, dent); 822 if (err < 0) 823 goto out_free; 824 if (err == 0) { 825 /* The node was not present */ 826 err = NOT_ON_MEDIA; 827 goto out_free; 828 } 829 ubifs_assert(err == 1); 830 831 err = lnc_add_directly(c, zbr, dent); 832 if (err) 833 goto out_free; 834 } else 835 dent = zbr->leaf; 836 837 nlen = le16_to_cpu(dent->nlen); 838 err = memcmp(dent->name, fname_name(nm), min_t(int, nlen, fname_len(nm))); 839 if (err == 0) { 840 if (nlen == fname_len(nm)) 841 return NAME_MATCHES; 842 else if (nlen < fname_len(nm)) 843 return NAME_LESS; 844 else 845 return NAME_GREATER; 846 } else if (err < 0) 847 return NAME_LESS; 848 else 849 return NAME_GREATER; 850 851 out_free: 852 kfree(dent); 853 return err; 854 } 855 856 /** 857 * fallible_resolve_collision - resolve a collision even if nodes are missing. 858 * @c: UBIFS file-system description object 859 * @key: key 860 * @zn: znode is returned here 861 * @n: branch number is passed and returned here 862 * @nm: name of directory entry 863 * @adding: indicates caller is adding a key to the TNC 864 * 865 * This is a "fallible" version of the 'resolve_collision()' function which 866 * does not panic if one of the nodes referred to by TNC does not exist on the 867 * media. This may happen when replaying the journal if a deleted node was 868 * Garbage-collected and the commit was not done. A branch that refers to a node 869 * that is not present is called a dangling branch. The following are the return 870 * codes for this function: 871 * o if @nm was found, %1 is returned and @zn and @n are set to the found 872 * branch; 873 * o if we are @adding and @nm was not found, %0 is returned; 874 * o if we are not @adding and @nm was not found, but a dangling branch was 875 * found, then %1 is returned and @zn and @n are set to the dangling branch; 876 * o a negative error code is returned in case of failure. 877 */ 878 static int fallible_resolve_collision(struct ubifs_info *c, 879 const union ubifs_key *key, 880 struct ubifs_znode **zn, int *n, 881 const struct fscrypt_name *nm, 882 int adding) 883 { 884 struct ubifs_znode *o_znode = NULL, *znode = *zn; 885 int uninitialized_var(o_n), err, cmp, unsure = 0, nn = *n; 886 887 cmp = fallible_matches_name(c, &znode->zbranch[nn], nm); 888 if (unlikely(cmp < 0)) 889 return cmp; 890 if (cmp == NAME_MATCHES) 891 return 1; 892 if (cmp == NOT_ON_MEDIA) { 893 o_znode = znode; 894 o_n = nn; 895 /* 896 * We are unlucky and hit a dangling branch straight away. 897 * Now we do not really know where to go to find the needed 898 * branch - to the left or to the right. Well, let's try left. 899 */ 900 unsure = 1; 901 } else if (!adding) 902 unsure = 1; /* Remove a dangling branch wherever it is */ 903 904 if (cmp == NAME_GREATER || unsure) { 905 /* Look left */ 906 while (1) { 907 err = tnc_prev(c, zn, n); 908 if (err == -ENOENT) { 909 ubifs_assert(*n == 0); 910 *n = -1; 911 break; 912 } 913 if (err < 0) 914 return err; 915 if (keys_cmp(c, &(*zn)->zbranch[*n].key, key)) { 916 /* See comments in 'resolve_collision()' */ 917 if (*n == (*zn)->child_cnt - 1) { 918 err = tnc_next(c, zn, n); 919 if (err) { 920 /* Should be impossible */ 921 ubifs_assert(0); 922 if (err == -ENOENT) 923 err = -EINVAL; 924 return err; 925 } 926 ubifs_assert(*n == 0); 927 *n = -1; 928 } 929 break; 930 } 931 err = fallible_matches_name(c, &(*zn)->zbranch[*n], nm); 932 if (err < 0) 933 return err; 934 if (err == NAME_MATCHES) 935 return 1; 936 if (err == NOT_ON_MEDIA) { 937 o_znode = *zn; 938 o_n = *n; 939 continue; 940 } 941 if (!adding) 942 continue; 943 if (err == NAME_LESS) 944 break; 945 else 946 unsure = 0; 947 } 948 } 949 950 if (cmp == NAME_LESS || unsure) { 951 /* Look right */ 952 *zn = znode; 953 *n = nn; 954 while (1) { 955 err = tnc_next(c, &znode, &nn); 956 if (err == -ENOENT) 957 break; 958 if (err < 0) 959 return err; 960 if (keys_cmp(c, &znode->zbranch[nn].key, key)) 961 break; 962 err = fallible_matches_name(c, &znode->zbranch[nn], nm); 963 if (err < 0) 964 return err; 965 if (err == NAME_GREATER) 966 break; 967 *zn = znode; 968 *n = nn; 969 if (err == NAME_MATCHES) 970 return 1; 971 if (err == NOT_ON_MEDIA) { 972 o_znode = znode; 973 o_n = nn; 974 } 975 } 976 } 977 978 /* Never match a dangling branch when adding */ 979 if (adding || !o_znode) 980 return 0; 981 982 dbg_mntk(key, "dangling match LEB %d:%d len %d key ", 983 o_znode->zbranch[o_n].lnum, o_znode->zbranch[o_n].offs, 984 o_znode->zbranch[o_n].len); 985 *zn = o_znode; 986 *n = o_n; 987 return 1; 988 } 989 990 /** 991 * matches_position - determine if a zbranch matches a given position. 992 * @zbr: zbranch of dent 993 * @lnum: LEB number of dent to match 994 * @offs: offset of dent to match 995 * 996 * This function returns %1 if @lnum:@offs matches, and %0 otherwise. 997 */ 998 static int matches_position(struct ubifs_zbranch *zbr, int lnum, int offs) 999 { 1000 if (zbr->lnum == lnum && zbr->offs == offs) 1001 return 1; 1002 else 1003 return 0; 1004 } 1005 1006 /** 1007 * resolve_collision_directly - resolve a collision directly. 1008 * @c: UBIFS file-system description object 1009 * @key: key of directory entry 1010 * @zn: znode is passed and returned here 1011 * @n: zbranch number is passed and returned here 1012 * @lnum: LEB number of dent node to match 1013 * @offs: offset of dent node to match 1014 * 1015 * This function is used for "hashed" keys to make sure the found directory or 1016 * extended attribute entry node is what was looked for. It is used when the 1017 * flash address of the right node is known (@lnum:@offs) which makes it much 1018 * easier to resolve collisions (no need to read entries and match full 1019 * names). This function returns %1 and sets @zn and @n if the collision is 1020 * resolved, %0 if @lnum:@offs is not found and @zn and @n are set to the 1021 * previous directory entry. Otherwise a negative error code is returned. 1022 */ 1023 static int resolve_collision_directly(struct ubifs_info *c, 1024 const union ubifs_key *key, 1025 struct ubifs_znode **zn, int *n, 1026 int lnum, int offs) 1027 { 1028 struct ubifs_znode *znode; 1029 int nn, err; 1030 1031 znode = *zn; 1032 nn = *n; 1033 if (matches_position(&znode->zbranch[nn], lnum, offs)) 1034 return 1; 1035 1036 /* Look left */ 1037 while (1) { 1038 err = tnc_prev(c, &znode, &nn); 1039 if (err == -ENOENT) 1040 break; 1041 if (err < 0) 1042 return err; 1043 if (keys_cmp(c, &znode->zbranch[nn].key, key)) 1044 break; 1045 if (matches_position(&znode->zbranch[nn], lnum, offs)) { 1046 *zn = znode; 1047 *n = nn; 1048 return 1; 1049 } 1050 } 1051 1052 /* Look right */ 1053 znode = *zn; 1054 nn = *n; 1055 while (1) { 1056 err = tnc_next(c, &znode, &nn); 1057 if (err == -ENOENT) 1058 return 0; 1059 if (err < 0) 1060 return err; 1061 if (keys_cmp(c, &znode->zbranch[nn].key, key)) 1062 return 0; 1063 *zn = znode; 1064 *n = nn; 1065 if (matches_position(&znode->zbranch[nn], lnum, offs)) 1066 return 1; 1067 } 1068 } 1069 1070 /** 1071 * dirty_cow_bottom_up - dirty a znode and its ancestors. 1072 * @c: UBIFS file-system description object 1073 * @znode: znode to dirty 1074 * 1075 * If we do not have a unique key that resides in a znode, then we cannot 1076 * dirty that znode from the top down (i.e. by using lookup_level0_dirty) 1077 * This function records the path back to the last dirty ancestor, and then 1078 * dirties the znodes on that path. 1079 */ 1080 static struct ubifs_znode *dirty_cow_bottom_up(struct ubifs_info *c, 1081 struct ubifs_znode *znode) 1082 { 1083 struct ubifs_znode *zp; 1084 int *path = c->bottom_up_buf, p = 0; 1085 1086 ubifs_assert(c->zroot.znode); 1087 ubifs_assert(znode); 1088 if (c->zroot.znode->level > BOTTOM_UP_HEIGHT) { 1089 kfree(c->bottom_up_buf); 1090 c->bottom_up_buf = kmalloc(c->zroot.znode->level * sizeof(int), 1091 GFP_NOFS); 1092 if (!c->bottom_up_buf) 1093 return ERR_PTR(-ENOMEM); 1094 path = c->bottom_up_buf; 1095 } 1096 if (c->zroot.znode->level) { 1097 /* Go up until parent is dirty */ 1098 while (1) { 1099 int n; 1100 1101 zp = znode->parent; 1102 if (!zp) 1103 break; 1104 n = znode->iip; 1105 ubifs_assert(p < c->zroot.znode->level); 1106 path[p++] = n; 1107 if (!zp->cnext && ubifs_zn_dirty(znode)) 1108 break; 1109 znode = zp; 1110 } 1111 } 1112 1113 /* Come back down, dirtying as we go */ 1114 while (1) { 1115 struct ubifs_zbranch *zbr; 1116 1117 zp = znode->parent; 1118 if (zp) { 1119 ubifs_assert(path[p - 1] >= 0); 1120 ubifs_assert(path[p - 1] < zp->child_cnt); 1121 zbr = &zp->zbranch[path[--p]]; 1122 znode = dirty_cow_znode(c, zbr); 1123 } else { 1124 ubifs_assert(znode == c->zroot.znode); 1125 znode = dirty_cow_znode(c, &c->zroot); 1126 } 1127 if (IS_ERR(znode) || !p) 1128 break; 1129 ubifs_assert(path[p - 1] >= 0); 1130 ubifs_assert(path[p - 1] < znode->child_cnt); 1131 znode = znode->zbranch[path[p - 1]].znode; 1132 } 1133 1134 return znode; 1135 } 1136 1137 /** 1138 * ubifs_lookup_level0 - search for zero-level znode. 1139 * @c: UBIFS file-system description object 1140 * @key: key to lookup 1141 * @zn: znode is returned here 1142 * @n: znode branch slot number is returned here 1143 * 1144 * This function looks up the TNC tree and search for zero-level znode which 1145 * refers key @key. The found zero-level znode is returned in @zn. There are 3 1146 * cases: 1147 * o exact match, i.e. the found zero-level znode contains key @key, then %1 1148 * is returned and slot number of the matched branch is stored in @n; 1149 * o not exact match, which means that zero-level znode does not contain 1150 * @key, then %0 is returned and slot number of the closest branch is stored 1151 * in @n; 1152 * o @key is so small that it is even less than the lowest key of the 1153 * leftmost zero-level node, then %0 is returned and %0 is stored in @n. 1154 * 1155 * Note, when the TNC tree is traversed, some znodes may be absent, then this 1156 * function reads corresponding indexing nodes and inserts them to TNC. In 1157 * case of failure, a negative error code is returned. 1158 */ 1159 int ubifs_lookup_level0(struct ubifs_info *c, const union ubifs_key *key, 1160 struct ubifs_znode **zn, int *n) 1161 { 1162 int err, exact; 1163 struct ubifs_znode *znode; 1164 unsigned long time = get_seconds(); 1165 1166 dbg_tnck(key, "search key "); 1167 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY); 1168 1169 znode = c->zroot.znode; 1170 if (unlikely(!znode)) { 1171 znode = ubifs_load_znode(c, &c->zroot, NULL, 0); 1172 if (IS_ERR(znode)) 1173 return PTR_ERR(znode); 1174 } 1175 1176 znode->time = time; 1177 1178 while (1) { 1179 struct ubifs_zbranch *zbr; 1180 1181 exact = ubifs_search_zbranch(c, znode, key, n); 1182 1183 if (znode->level == 0) 1184 break; 1185 1186 if (*n < 0) 1187 *n = 0; 1188 zbr = &znode->zbranch[*n]; 1189 1190 if (zbr->znode) { 1191 znode->time = time; 1192 znode = zbr->znode; 1193 continue; 1194 } 1195 1196 /* znode is not in TNC cache, load it from the media */ 1197 znode = ubifs_load_znode(c, zbr, znode, *n); 1198 if (IS_ERR(znode)) 1199 return PTR_ERR(znode); 1200 } 1201 1202 *zn = znode; 1203 if (exact || !is_hash_key(c, key) || *n != -1) { 1204 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); 1205 return exact; 1206 } 1207 1208 /* 1209 * Here is a tricky place. We have not found the key and this is a 1210 * "hashed" key, which may collide. The rest of the code deals with 1211 * situations like this: 1212 * 1213 * | 3 | 5 | 1214 * / \ 1215 * | 3 | 5 | | 6 | 7 | (x) 1216 * 1217 * Or more a complex example: 1218 * 1219 * | 1 | 5 | 1220 * / \ 1221 * | 1 | 3 | | 5 | 8 | 1222 * \ / 1223 * | 5 | 5 | | 6 | 7 | (x) 1224 * 1225 * In the examples, if we are looking for key "5", we may reach nodes 1226 * marked with "(x)". In this case what we have do is to look at the 1227 * left and see if there is "5" key there. If there is, we have to 1228 * return it. 1229 * 1230 * Note, this whole situation is possible because we allow to have 1231 * elements which are equivalent to the next key in the parent in the 1232 * children of current znode. For example, this happens if we split a 1233 * znode like this: | 3 | 5 | 5 | 6 | 7 |, which results in something 1234 * like this: 1235 * | 3 | 5 | 1236 * / \ 1237 * | 3 | 5 | | 5 | 6 | 7 | 1238 * ^ 1239 * And this becomes what is at the first "picture" after key "5" marked 1240 * with "^" is removed. What could be done is we could prohibit 1241 * splitting in the middle of the colliding sequence. Also, when 1242 * removing the leftmost key, we would have to correct the key of the 1243 * parent node, which would introduce additional complications. Namely, 1244 * if we changed the leftmost key of the parent znode, the garbage 1245 * collector would be unable to find it (GC is doing this when GC'ing 1246 * indexing LEBs). Although we already have an additional RB-tree where 1247 * we save such changed znodes (see 'ins_clr_old_idx_znode()') until 1248 * after the commit. But anyway, this does not look easy to implement 1249 * so we did not try this. 1250 */ 1251 err = tnc_prev(c, &znode, n); 1252 if (err == -ENOENT) { 1253 dbg_tnc("found 0, lvl %d, n -1", znode->level); 1254 *n = -1; 1255 return 0; 1256 } 1257 if (unlikely(err < 0)) 1258 return err; 1259 if (keys_cmp(c, key, &znode->zbranch[*n].key)) { 1260 dbg_tnc("found 0, lvl %d, n -1", znode->level); 1261 *n = -1; 1262 return 0; 1263 } 1264 1265 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); 1266 *zn = znode; 1267 return 1; 1268 } 1269 1270 /** 1271 * lookup_level0_dirty - search for zero-level znode dirtying. 1272 * @c: UBIFS file-system description object 1273 * @key: key to lookup 1274 * @zn: znode is returned here 1275 * @n: znode branch slot number is returned here 1276 * 1277 * This function looks up the TNC tree and search for zero-level znode which 1278 * refers key @key. The found zero-level znode is returned in @zn. There are 3 1279 * cases: 1280 * o exact match, i.e. the found zero-level znode contains key @key, then %1 1281 * is returned and slot number of the matched branch is stored in @n; 1282 * o not exact match, which means that zero-level znode does not contain @key 1283 * then %0 is returned and slot number of the closed branch is stored in 1284 * @n; 1285 * o @key is so small that it is even less than the lowest key of the 1286 * leftmost zero-level node, then %0 is returned and %-1 is stored in @n. 1287 * 1288 * Additionally all znodes in the path from the root to the located zero-level 1289 * znode are marked as dirty. 1290 * 1291 * Note, when the TNC tree is traversed, some znodes may be absent, then this 1292 * function reads corresponding indexing nodes and inserts them to TNC. In 1293 * case of failure, a negative error code is returned. 1294 */ 1295 static int lookup_level0_dirty(struct ubifs_info *c, const union ubifs_key *key, 1296 struct ubifs_znode **zn, int *n) 1297 { 1298 int err, exact; 1299 struct ubifs_znode *znode; 1300 unsigned long time = get_seconds(); 1301 1302 dbg_tnck(key, "search and dirty key "); 1303 1304 znode = c->zroot.znode; 1305 if (unlikely(!znode)) { 1306 znode = ubifs_load_znode(c, &c->zroot, NULL, 0); 1307 if (IS_ERR(znode)) 1308 return PTR_ERR(znode); 1309 } 1310 1311 znode = dirty_cow_znode(c, &c->zroot); 1312 if (IS_ERR(znode)) 1313 return PTR_ERR(znode); 1314 1315 znode->time = time; 1316 1317 while (1) { 1318 struct ubifs_zbranch *zbr; 1319 1320 exact = ubifs_search_zbranch(c, znode, key, n); 1321 1322 if (znode->level == 0) 1323 break; 1324 1325 if (*n < 0) 1326 *n = 0; 1327 zbr = &znode->zbranch[*n]; 1328 1329 if (zbr->znode) { 1330 znode->time = time; 1331 znode = dirty_cow_znode(c, zbr); 1332 if (IS_ERR(znode)) 1333 return PTR_ERR(znode); 1334 continue; 1335 } 1336 1337 /* znode is not in TNC cache, load it from the media */ 1338 znode = ubifs_load_znode(c, zbr, znode, *n); 1339 if (IS_ERR(znode)) 1340 return PTR_ERR(znode); 1341 znode = dirty_cow_znode(c, zbr); 1342 if (IS_ERR(znode)) 1343 return PTR_ERR(znode); 1344 } 1345 1346 *zn = znode; 1347 if (exact || !is_hash_key(c, key) || *n != -1) { 1348 dbg_tnc("found %d, lvl %d, n %d", exact, znode->level, *n); 1349 return exact; 1350 } 1351 1352 /* 1353 * See huge comment at 'lookup_level0_dirty()' what is the rest of the 1354 * code. 1355 */ 1356 err = tnc_prev(c, &znode, n); 1357 if (err == -ENOENT) { 1358 *n = -1; 1359 dbg_tnc("found 0, lvl %d, n -1", znode->level); 1360 return 0; 1361 } 1362 if (unlikely(err < 0)) 1363 return err; 1364 if (keys_cmp(c, key, &znode->zbranch[*n].key)) { 1365 *n = -1; 1366 dbg_tnc("found 0, lvl %d, n -1", znode->level); 1367 return 0; 1368 } 1369 1370 if (znode->cnext || !ubifs_zn_dirty(znode)) { 1371 znode = dirty_cow_bottom_up(c, znode); 1372 if (IS_ERR(znode)) 1373 return PTR_ERR(znode); 1374 } 1375 1376 dbg_tnc("found 1, lvl %d, n %d", znode->level, *n); 1377 *zn = znode; 1378 return 1; 1379 } 1380 1381 /** 1382 * maybe_leb_gced - determine if a LEB may have been garbage collected. 1383 * @c: UBIFS file-system description object 1384 * @lnum: LEB number 1385 * @gc_seq1: garbage collection sequence number 1386 * 1387 * This function determines if @lnum may have been garbage collected since 1388 * sequence number @gc_seq1. If it may have been then %1 is returned, otherwise 1389 * %0 is returned. 1390 */ 1391 static int maybe_leb_gced(struct ubifs_info *c, int lnum, int gc_seq1) 1392 { 1393 int gc_seq2, gced_lnum; 1394 1395 gced_lnum = c->gced_lnum; 1396 smp_rmb(); 1397 gc_seq2 = c->gc_seq; 1398 /* Same seq means no GC */ 1399 if (gc_seq1 == gc_seq2) 1400 return 0; 1401 /* Different by more than 1 means we don't know */ 1402 if (gc_seq1 + 1 != gc_seq2) 1403 return 1; 1404 /* 1405 * We have seen the sequence number has increased by 1. Now we need to 1406 * be sure we read the right LEB number, so read it again. 1407 */ 1408 smp_rmb(); 1409 if (gced_lnum != c->gced_lnum) 1410 return 1; 1411 /* Finally we can check lnum */ 1412 if (gced_lnum == lnum) 1413 return 1; 1414 return 0; 1415 } 1416 1417 /** 1418 * ubifs_tnc_locate - look up a file-system node and return it and its location. 1419 * @c: UBIFS file-system description object 1420 * @key: node key to lookup 1421 * @node: the node is returned here 1422 * @lnum: LEB number is returned here 1423 * @offs: offset is returned here 1424 * 1425 * This function looks up and reads node with key @key. The caller has to make 1426 * sure the @node buffer is large enough to fit the node. Returns zero in case 1427 * of success, %-ENOENT if the node was not found, and a negative error code in 1428 * case of failure. The node location can be returned in @lnum and @offs. 1429 */ 1430 int ubifs_tnc_locate(struct ubifs_info *c, const union ubifs_key *key, 1431 void *node, int *lnum, int *offs) 1432 { 1433 int found, n, err, safely = 0, gc_seq1; 1434 struct ubifs_znode *znode; 1435 struct ubifs_zbranch zbr, *zt; 1436 1437 again: 1438 mutex_lock(&c->tnc_mutex); 1439 found = ubifs_lookup_level0(c, key, &znode, &n); 1440 if (!found) { 1441 err = -ENOENT; 1442 goto out; 1443 } else if (found < 0) { 1444 err = found; 1445 goto out; 1446 } 1447 zt = &znode->zbranch[n]; 1448 if (lnum) { 1449 *lnum = zt->lnum; 1450 *offs = zt->offs; 1451 } 1452 if (is_hash_key(c, key)) { 1453 /* 1454 * In this case the leaf node cache gets used, so we pass the 1455 * address of the zbranch and keep the mutex locked 1456 */ 1457 err = tnc_read_hashed_node(c, zt, node); 1458 goto out; 1459 } 1460 if (safely) { 1461 err = ubifs_tnc_read_node(c, zt, node); 1462 goto out; 1463 } 1464 /* Drop the TNC mutex prematurely and race with garbage collection */ 1465 zbr = znode->zbranch[n]; 1466 gc_seq1 = c->gc_seq; 1467 mutex_unlock(&c->tnc_mutex); 1468 1469 if (ubifs_get_wbuf(c, zbr.lnum)) { 1470 /* We do not GC journal heads */ 1471 err = ubifs_tnc_read_node(c, &zbr, node); 1472 return err; 1473 } 1474 1475 err = fallible_read_node(c, key, &zbr, node); 1476 if (err <= 0 || maybe_leb_gced(c, zbr.lnum, gc_seq1)) { 1477 /* 1478 * The node may have been GC'ed out from under us so try again 1479 * while keeping the TNC mutex locked. 1480 */ 1481 safely = 1; 1482 goto again; 1483 } 1484 return 0; 1485 1486 out: 1487 mutex_unlock(&c->tnc_mutex); 1488 return err; 1489 } 1490 1491 /** 1492 * ubifs_tnc_get_bu_keys - lookup keys for bulk-read. 1493 * @c: UBIFS file-system description object 1494 * @bu: bulk-read parameters and results 1495 * 1496 * Lookup consecutive data node keys for the same inode that reside 1497 * consecutively in the same LEB. This function returns zero in case of success 1498 * and a negative error code in case of failure. 1499 * 1500 * Note, if the bulk-read buffer length (@bu->buf_len) is known, this function 1501 * makes sure bulk-read nodes fit the buffer. Otherwise, this function prepares 1502 * maximum possible amount of nodes for bulk-read. 1503 */ 1504 int ubifs_tnc_get_bu_keys(struct ubifs_info *c, struct bu_info *bu) 1505 { 1506 int n, err = 0, lnum = -1, uninitialized_var(offs); 1507 int uninitialized_var(len); 1508 unsigned int block = key_block(c, &bu->key); 1509 struct ubifs_znode *znode; 1510 1511 bu->cnt = 0; 1512 bu->blk_cnt = 0; 1513 bu->eof = 0; 1514 1515 mutex_lock(&c->tnc_mutex); 1516 /* Find first key */ 1517 err = ubifs_lookup_level0(c, &bu->key, &znode, &n); 1518 if (err < 0) 1519 goto out; 1520 if (err) { 1521 /* Key found */ 1522 len = znode->zbranch[n].len; 1523 /* The buffer must be big enough for at least 1 node */ 1524 if (len > bu->buf_len) { 1525 err = -EINVAL; 1526 goto out; 1527 } 1528 /* Add this key */ 1529 bu->zbranch[bu->cnt++] = znode->zbranch[n]; 1530 bu->blk_cnt += 1; 1531 lnum = znode->zbranch[n].lnum; 1532 offs = ALIGN(znode->zbranch[n].offs + len, 8); 1533 } 1534 while (1) { 1535 struct ubifs_zbranch *zbr; 1536 union ubifs_key *key; 1537 unsigned int next_block; 1538 1539 /* Find next key */ 1540 err = tnc_next(c, &znode, &n); 1541 if (err) 1542 goto out; 1543 zbr = &znode->zbranch[n]; 1544 key = &zbr->key; 1545 /* See if there is another data key for this file */ 1546 if (key_inum(c, key) != key_inum(c, &bu->key) || 1547 key_type(c, key) != UBIFS_DATA_KEY) { 1548 err = -ENOENT; 1549 goto out; 1550 } 1551 if (lnum < 0) { 1552 /* First key found */ 1553 lnum = zbr->lnum; 1554 offs = ALIGN(zbr->offs + zbr->len, 8); 1555 len = zbr->len; 1556 if (len > bu->buf_len) { 1557 err = -EINVAL; 1558 goto out; 1559 } 1560 } else { 1561 /* 1562 * The data nodes must be in consecutive positions in 1563 * the same LEB. 1564 */ 1565 if (zbr->lnum != lnum || zbr->offs != offs) 1566 goto out; 1567 offs += ALIGN(zbr->len, 8); 1568 len = ALIGN(len, 8) + zbr->len; 1569 /* Must not exceed buffer length */ 1570 if (len > bu->buf_len) 1571 goto out; 1572 } 1573 /* Allow for holes */ 1574 next_block = key_block(c, key); 1575 bu->blk_cnt += (next_block - block - 1); 1576 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) 1577 goto out; 1578 block = next_block; 1579 /* Add this key */ 1580 bu->zbranch[bu->cnt++] = *zbr; 1581 bu->blk_cnt += 1; 1582 /* See if we have room for more */ 1583 if (bu->cnt >= UBIFS_MAX_BULK_READ) 1584 goto out; 1585 if (bu->blk_cnt >= UBIFS_MAX_BULK_READ) 1586 goto out; 1587 } 1588 out: 1589 if (err == -ENOENT) { 1590 bu->eof = 1; 1591 err = 0; 1592 } 1593 bu->gc_seq = c->gc_seq; 1594 mutex_unlock(&c->tnc_mutex); 1595 if (err) 1596 return err; 1597 /* 1598 * An enormous hole could cause bulk-read to encompass too many 1599 * page cache pages, so limit the number here. 1600 */ 1601 if (bu->blk_cnt > UBIFS_MAX_BULK_READ) 1602 bu->blk_cnt = UBIFS_MAX_BULK_READ; 1603 /* 1604 * Ensure that bulk-read covers a whole number of page cache 1605 * pages. 1606 */ 1607 if (UBIFS_BLOCKS_PER_PAGE == 1 || 1608 !(bu->blk_cnt & (UBIFS_BLOCKS_PER_PAGE - 1))) 1609 return 0; 1610 if (bu->eof) { 1611 /* At the end of file we can round up */ 1612 bu->blk_cnt += UBIFS_BLOCKS_PER_PAGE - 1; 1613 return 0; 1614 } 1615 /* Exclude data nodes that do not make up a whole page cache page */ 1616 block = key_block(c, &bu->key) + bu->blk_cnt; 1617 block &= ~(UBIFS_BLOCKS_PER_PAGE - 1); 1618 while (bu->cnt) { 1619 if (key_block(c, &bu->zbranch[bu->cnt - 1].key) < block) 1620 break; 1621 bu->cnt -= 1; 1622 } 1623 return 0; 1624 } 1625 1626 /** 1627 * read_wbuf - bulk-read from a LEB with a wbuf. 1628 * @wbuf: wbuf that may overlap the read 1629 * @buf: buffer into which to read 1630 * @len: read length 1631 * @lnum: LEB number from which to read 1632 * @offs: offset from which to read 1633 * 1634 * This functions returns %0 on success or a negative error code on failure. 1635 */ 1636 static int read_wbuf(struct ubifs_wbuf *wbuf, void *buf, int len, int lnum, 1637 int offs) 1638 { 1639 const struct ubifs_info *c = wbuf->c; 1640 int rlen, overlap; 1641 1642 dbg_io("LEB %d:%d, length %d", lnum, offs, len); 1643 ubifs_assert(wbuf && lnum >= 0 && lnum < c->leb_cnt && offs >= 0); 1644 ubifs_assert(!(offs & 7) && offs < c->leb_size); 1645 ubifs_assert(offs + len <= c->leb_size); 1646 1647 spin_lock(&wbuf->lock); 1648 overlap = (lnum == wbuf->lnum && offs + len > wbuf->offs); 1649 if (!overlap) { 1650 /* We may safely unlock the write-buffer and read the data */ 1651 spin_unlock(&wbuf->lock); 1652 return ubifs_leb_read(c, lnum, buf, offs, len, 0); 1653 } 1654 1655 /* Don't read under wbuf */ 1656 rlen = wbuf->offs - offs; 1657 if (rlen < 0) 1658 rlen = 0; 1659 1660 /* Copy the rest from the write-buffer */ 1661 memcpy(buf + rlen, wbuf->buf + offs + rlen - wbuf->offs, len - rlen); 1662 spin_unlock(&wbuf->lock); 1663 1664 if (rlen > 0) 1665 /* Read everything that goes before write-buffer */ 1666 return ubifs_leb_read(c, lnum, buf, offs, rlen, 0); 1667 1668 return 0; 1669 } 1670 1671 /** 1672 * validate_data_node - validate data nodes for bulk-read. 1673 * @c: UBIFS file-system description object 1674 * @buf: buffer containing data node to validate 1675 * @zbr: zbranch of data node to validate 1676 * 1677 * This functions returns %0 on success or a negative error code on failure. 1678 */ 1679 static int validate_data_node(struct ubifs_info *c, void *buf, 1680 struct ubifs_zbranch *zbr) 1681 { 1682 union ubifs_key key1; 1683 struct ubifs_ch *ch = buf; 1684 int err, len; 1685 1686 if (ch->node_type != UBIFS_DATA_NODE) { 1687 ubifs_err(c, "bad node type (%d but expected %d)", 1688 ch->node_type, UBIFS_DATA_NODE); 1689 goto out_err; 1690 } 1691 1692 err = ubifs_check_node(c, buf, zbr->lnum, zbr->offs, 0, 0); 1693 if (err) { 1694 ubifs_err(c, "expected node type %d", UBIFS_DATA_NODE); 1695 goto out; 1696 } 1697 1698 len = le32_to_cpu(ch->len); 1699 if (len != zbr->len) { 1700 ubifs_err(c, "bad node length %d, expected %d", len, zbr->len); 1701 goto out_err; 1702 } 1703 1704 /* Make sure the key of the read node is correct */ 1705 key_read(c, buf + UBIFS_KEY_OFFSET, &key1); 1706 if (!keys_eq(c, &zbr->key, &key1)) { 1707 ubifs_err(c, "bad key in node at LEB %d:%d", 1708 zbr->lnum, zbr->offs); 1709 dbg_tnck(&zbr->key, "looked for key "); 1710 dbg_tnck(&key1, "found node's key "); 1711 goto out_err; 1712 } 1713 1714 return 0; 1715 1716 out_err: 1717 err = -EINVAL; 1718 out: 1719 ubifs_err(c, "bad node at LEB %d:%d", zbr->lnum, zbr->offs); 1720 ubifs_dump_node(c, buf); 1721 dump_stack(); 1722 return err; 1723 } 1724 1725 /** 1726 * ubifs_tnc_bulk_read - read a number of data nodes in one go. 1727 * @c: UBIFS file-system description object 1728 * @bu: bulk-read parameters and results 1729 * 1730 * This functions reads and validates the data nodes that were identified by the 1731 * 'ubifs_tnc_get_bu_keys()' function. This functions returns %0 on success, 1732 * -EAGAIN to indicate a race with GC, or another negative error code on 1733 * failure. 1734 */ 1735 int ubifs_tnc_bulk_read(struct ubifs_info *c, struct bu_info *bu) 1736 { 1737 int lnum = bu->zbranch[0].lnum, offs = bu->zbranch[0].offs, len, err, i; 1738 struct ubifs_wbuf *wbuf; 1739 void *buf; 1740 1741 len = bu->zbranch[bu->cnt - 1].offs; 1742 len += bu->zbranch[bu->cnt - 1].len - offs; 1743 if (len > bu->buf_len) { 1744 ubifs_err(c, "buffer too small %d vs %d", bu->buf_len, len); 1745 return -EINVAL; 1746 } 1747 1748 /* Do the read */ 1749 wbuf = ubifs_get_wbuf(c, lnum); 1750 if (wbuf) 1751 err = read_wbuf(wbuf, bu->buf, len, lnum, offs); 1752 else 1753 err = ubifs_leb_read(c, lnum, bu->buf, offs, len, 0); 1754 1755 /* Check for a race with GC */ 1756 if (maybe_leb_gced(c, lnum, bu->gc_seq)) 1757 return -EAGAIN; 1758 1759 if (err && err != -EBADMSG) { 1760 ubifs_err(c, "failed to read from LEB %d:%d, error %d", 1761 lnum, offs, err); 1762 dump_stack(); 1763 dbg_tnck(&bu->key, "key "); 1764 return err; 1765 } 1766 1767 /* Validate the nodes read */ 1768 buf = bu->buf; 1769 for (i = 0; i < bu->cnt; i++) { 1770 err = validate_data_node(c, buf, &bu->zbranch[i]); 1771 if (err) 1772 return err; 1773 buf = buf + ALIGN(bu->zbranch[i].len, 8); 1774 } 1775 1776 return 0; 1777 } 1778 1779 /** 1780 * do_lookup_nm- look up a "hashed" node. 1781 * @c: UBIFS file-system description object 1782 * @key: node key to lookup 1783 * @node: the node is returned here 1784 * @nm: node name 1785 * 1786 * This function looks up and reads a node which contains name hash in the key. 1787 * Since the hash may have collisions, there may be many nodes with the same 1788 * key, so we have to sequentially look to all of them until the needed one is 1789 * found. This function returns zero in case of success, %-ENOENT if the node 1790 * was not found, and a negative error code in case of failure. 1791 */ 1792 static int do_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, 1793 void *node, const struct fscrypt_name *nm) 1794 { 1795 int found, n, err; 1796 struct ubifs_znode *znode; 1797 1798 //dbg_tnck(key, "name '%.*s' key ", nm->len, nm->name); 1799 mutex_lock(&c->tnc_mutex); 1800 found = ubifs_lookup_level0(c, key, &znode, &n); 1801 if (!found) { 1802 err = -ENOENT; 1803 goto out_unlock; 1804 } else if (found < 0) { 1805 err = found; 1806 goto out_unlock; 1807 } 1808 1809 ubifs_assert(n >= 0); 1810 1811 err = resolve_collision(c, key, &znode, &n, nm); 1812 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); 1813 if (unlikely(err < 0)) 1814 goto out_unlock; 1815 if (err == 0) { 1816 err = -ENOENT; 1817 goto out_unlock; 1818 } 1819 1820 err = tnc_read_hashed_node(c, &znode->zbranch[n], node); 1821 1822 out_unlock: 1823 mutex_unlock(&c->tnc_mutex); 1824 return err; 1825 } 1826 1827 /** 1828 * ubifs_tnc_lookup_nm - look up a "hashed" node. 1829 * @c: UBIFS file-system description object 1830 * @key: node key to lookup 1831 * @node: the node is returned here 1832 * @nm: node name 1833 * 1834 * This function looks up and reads a node which contains name hash in the key. 1835 * Since the hash may have collisions, there may be many nodes with the same 1836 * key, so we have to sequentially look to all of them until the needed one is 1837 * found. This function returns zero in case of success, %-ENOENT if the node 1838 * was not found, and a negative error code in case of failure. 1839 */ 1840 int ubifs_tnc_lookup_nm(struct ubifs_info *c, const union ubifs_key *key, 1841 void *node, const struct fscrypt_name *nm) 1842 { 1843 int err, len; 1844 const struct ubifs_dent_node *dent = node; 1845 1846 /* 1847 * We assume that in most of the cases there are no name collisions and 1848 * 'ubifs_tnc_lookup()' returns us the right direntry. 1849 */ 1850 err = ubifs_tnc_lookup(c, key, node); 1851 if (err) 1852 return err; 1853 1854 len = le16_to_cpu(dent->nlen); 1855 if (fname_len(nm) == len && !memcmp(dent->name, fname_name(nm), len)) 1856 return 0; 1857 1858 /* 1859 * Unluckily, there are hash collisions and we have to iterate over 1860 * them look at each direntry with colliding name hash sequentially. 1861 */ 1862 1863 return do_lookup_nm(c, key, node, nm); 1864 } 1865 1866 static int do_lookup_dh(struct ubifs_info *c, const union ubifs_key *key, 1867 struct ubifs_dent_node *dent, uint32_t cookie) 1868 { 1869 int n, err, type = key_type(c, key); 1870 struct ubifs_znode *znode; 1871 struct ubifs_zbranch *zbr; 1872 union ubifs_key *dkey, start_key; 1873 1874 ubifs_assert(is_hash_key(c, key)); 1875 1876 lowest_dent_key(c, &start_key, key_inum(c, key)); 1877 1878 mutex_lock(&c->tnc_mutex); 1879 err = ubifs_lookup_level0(c, &start_key, &znode, &n); 1880 if (unlikely(err < 0)) 1881 goto out_unlock; 1882 1883 for (;;) { 1884 if (!err) { 1885 err = tnc_next(c, &znode, &n); 1886 if (err) 1887 goto out_unlock; 1888 } 1889 1890 zbr = &znode->zbranch[n]; 1891 dkey = &zbr->key; 1892 1893 if (key_inum(c, dkey) != key_inum(c, key) || 1894 key_type(c, dkey) != type) { 1895 err = -ENOENT; 1896 goto out_unlock; 1897 } 1898 1899 err = tnc_read_hashed_node(c, zbr, dent); 1900 if (err) 1901 goto out_unlock; 1902 1903 if (key_hash(c, key) == key_hash(c, dkey) && 1904 le32_to_cpu(dent->cookie) == cookie) 1905 goto out_unlock; 1906 } 1907 1908 out_unlock: 1909 mutex_unlock(&c->tnc_mutex); 1910 return err; 1911 } 1912 1913 /** 1914 * ubifs_tnc_lookup_dh - look up a "double hashed" node. 1915 * @c: UBIFS file-system description object 1916 * @key: node key to lookup 1917 * @node: the node is returned here 1918 * @cookie: node cookie for collision resolution 1919 * 1920 * This function looks up and reads a node which contains name hash in the key. 1921 * Since the hash may have collisions, there may be many nodes with the same 1922 * key, so we have to sequentially look to all of them until the needed one 1923 * with the same cookie value is found. 1924 * This function returns zero in case of success, %-ENOENT if the node 1925 * was not found, and a negative error code in case of failure. 1926 */ 1927 int ubifs_tnc_lookup_dh(struct ubifs_info *c, const union ubifs_key *key, 1928 void *node, uint32_t cookie) 1929 { 1930 int err; 1931 const struct ubifs_dent_node *dent = node; 1932 1933 if (!c->double_hash) 1934 return -EOPNOTSUPP; 1935 1936 /* 1937 * We assume that in most of the cases there are no name collisions and 1938 * 'ubifs_tnc_lookup()' returns us the right direntry. 1939 */ 1940 err = ubifs_tnc_lookup(c, key, node); 1941 if (err) 1942 return err; 1943 1944 if (le32_to_cpu(dent->cookie) == cookie) 1945 return 0; 1946 1947 /* 1948 * Unluckily, there are hash collisions and we have to iterate over 1949 * them look at each direntry with colliding name hash sequentially. 1950 */ 1951 return do_lookup_dh(c, key, node, cookie); 1952 } 1953 1954 /** 1955 * correct_parent_keys - correct parent znodes' keys. 1956 * @c: UBIFS file-system description object 1957 * @znode: znode to correct parent znodes for 1958 * 1959 * This is a helper function for 'tnc_insert()'. When the key of the leftmost 1960 * zbranch changes, keys of parent znodes have to be corrected. This helper 1961 * function is called in such situations and corrects the keys if needed. 1962 */ 1963 static void correct_parent_keys(const struct ubifs_info *c, 1964 struct ubifs_znode *znode) 1965 { 1966 union ubifs_key *key, *key1; 1967 1968 ubifs_assert(znode->parent); 1969 ubifs_assert(znode->iip == 0); 1970 1971 key = &znode->zbranch[0].key; 1972 key1 = &znode->parent->zbranch[0].key; 1973 1974 while (keys_cmp(c, key, key1) < 0) { 1975 key_copy(c, key, key1); 1976 znode = znode->parent; 1977 znode->alt = 1; 1978 if (!znode->parent || znode->iip) 1979 break; 1980 key1 = &znode->parent->zbranch[0].key; 1981 } 1982 } 1983 1984 /** 1985 * insert_zbranch - insert a zbranch into a znode. 1986 * @znode: znode into which to insert 1987 * @zbr: zbranch to insert 1988 * @n: slot number to insert to 1989 * 1990 * This is a helper function for 'tnc_insert()'. UBIFS does not allow "gaps" in 1991 * znode's array of zbranches and keeps zbranches consolidated, so when a new 1992 * zbranch has to be inserted to the @znode->zbranches[]' array at the @n-th 1993 * slot, zbranches starting from @n have to be moved right. 1994 */ 1995 static void insert_zbranch(struct ubifs_znode *znode, 1996 const struct ubifs_zbranch *zbr, int n) 1997 { 1998 int i; 1999 2000 ubifs_assert(ubifs_zn_dirty(znode)); 2001 2002 if (znode->level) { 2003 for (i = znode->child_cnt; i > n; i--) { 2004 znode->zbranch[i] = znode->zbranch[i - 1]; 2005 if (znode->zbranch[i].znode) 2006 znode->zbranch[i].znode->iip = i; 2007 } 2008 if (zbr->znode) 2009 zbr->znode->iip = n; 2010 } else 2011 for (i = znode->child_cnt; i > n; i--) 2012 znode->zbranch[i] = znode->zbranch[i - 1]; 2013 2014 znode->zbranch[n] = *zbr; 2015 znode->child_cnt += 1; 2016 2017 /* 2018 * After inserting at slot zero, the lower bound of the key range of 2019 * this znode may have changed. If this znode is subsequently split 2020 * then the upper bound of the key range may change, and furthermore 2021 * it could change to be lower than the original lower bound. If that 2022 * happens, then it will no longer be possible to find this znode in the 2023 * TNC using the key from the index node on flash. That is bad because 2024 * if it is not found, we will assume it is obsolete and may overwrite 2025 * it. Then if there is an unclean unmount, we will start using the 2026 * old index which will be broken. 2027 * 2028 * So we first mark znodes that have insertions at slot zero, and then 2029 * if they are split we add their lnum/offs to the old_idx tree. 2030 */ 2031 if (n == 0) 2032 znode->alt = 1; 2033 } 2034 2035 /** 2036 * tnc_insert - insert a node into TNC. 2037 * @c: UBIFS file-system description object 2038 * @znode: znode to insert into 2039 * @zbr: branch to insert 2040 * @n: slot number to insert new zbranch to 2041 * 2042 * This function inserts a new node described by @zbr into znode @znode. If 2043 * znode does not have a free slot for new zbranch, it is split. Parent znodes 2044 * are splat as well if needed. Returns zero in case of success or a negative 2045 * error code in case of failure. 2046 */ 2047 static int tnc_insert(struct ubifs_info *c, struct ubifs_znode *znode, 2048 struct ubifs_zbranch *zbr, int n) 2049 { 2050 struct ubifs_znode *zn, *zi, *zp; 2051 int i, keep, move, appending = 0; 2052 union ubifs_key *key = &zbr->key, *key1; 2053 2054 ubifs_assert(n >= 0 && n <= c->fanout); 2055 2056 /* Implement naive insert for now */ 2057 again: 2058 zp = znode->parent; 2059 if (znode->child_cnt < c->fanout) { 2060 ubifs_assert(n != c->fanout); 2061 dbg_tnck(key, "inserted at %d level %d, key ", n, znode->level); 2062 2063 insert_zbranch(znode, zbr, n); 2064 2065 /* Ensure parent's key is correct */ 2066 if (n == 0 && zp && znode->iip == 0) 2067 correct_parent_keys(c, znode); 2068 2069 return 0; 2070 } 2071 2072 /* 2073 * Unfortunately, @znode does not have more empty slots and we have to 2074 * split it. 2075 */ 2076 dbg_tnck(key, "splitting level %d, key ", znode->level); 2077 2078 if (znode->alt) 2079 /* 2080 * We can no longer be sure of finding this znode by key, so we 2081 * record it in the old_idx tree. 2082 */ 2083 ins_clr_old_idx_znode(c, znode); 2084 2085 zn = kzalloc(c->max_znode_sz, GFP_NOFS); 2086 if (!zn) 2087 return -ENOMEM; 2088 zn->parent = zp; 2089 zn->level = znode->level; 2090 2091 /* Decide where to split */ 2092 if (znode->level == 0 && key_type(c, key) == UBIFS_DATA_KEY) { 2093 /* Try not to split consecutive data keys */ 2094 if (n == c->fanout) { 2095 key1 = &znode->zbranch[n - 1].key; 2096 if (key_inum(c, key1) == key_inum(c, key) && 2097 key_type(c, key1) == UBIFS_DATA_KEY) 2098 appending = 1; 2099 } else 2100 goto check_split; 2101 } else if (appending && n != c->fanout) { 2102 /* Try not to split consecutive data keys */ 2103 appending = 0; 2104 check_split: 2105 if (n >= (c->fanout + 1) / 2) { 2106 key1 = &znode->zbranch[0].key; 2107 if (key_inum(c, key1) == key_inum(c, key) && 2108 key_type(c, key1) == UBIFS_DATA_KEY) { 2109 key1 = &znode->zbranch[n].key; 2110 if (key_inum(c, key1) != key_inum(c, key) || 2111 key_type(c, key1) != UBIFS_DATA_KEY) { 2112 keep = n; 2113 move = c->fanout - keep; 2114 zi = znode; 2115 goto do_split; 2116 } 2117 } 2118 } 2119 } 2120 2121 if (appending) { 2122 keep = c->fanout; 2123 move = 0; 2124 } else { 2125 keep = (c->fanout + 1) / 2; 2126 move = c->fanout - keep; 2127 } 2128 2129 /* 2130 * Although we don't at present, we could look at the neighbors and see 2131 * if we can move some zbranches there. 2132 */ 2133 2134 if (n < keep) { 2135 /* Insert into existing znode */ 2136 zi = znode; 2137 move += 1; 2138 keep -= 1; 2139 } else { 2140 /* Insert into new znode */ 2141 zi = zn; 2142 n -= keep; 2143 /* Re-parent */ 2144 if (zn->level != 0) 2145 zbr->znode->parent = zn; 2146 } 2147 2148 do_split: 2149 2150 __set_bit(DIRTY_ZNODE, &zn->flags); 2151 atomic_long_inc(&c->dirty_zn_cnt); 2152 2153 zn->child_cnt = move; 2154 znode->child_cnt = keep; 2155 2156 dbg_tnc("moving %d, keeping %d", move, keep); 2157 2158 /* Move zbranch */ 2159 for (i = 0; i < move; i++) { 2160 zn->zbranch[i] = znode->zbranch[keep + i]; 2161 /* Re-parent */ 2162 if (zn->level != 0) 2163 if (zn->zbranch[i].znode) { 2164 zn->zbranch[i].znode->parent = zn; 2165 zn->zbranch[i].znode->iip = i; 2166 } 2167 } 2168 2169 /* Insert new key and branch */ 2170 dbg_tnck(key, "inserting at %d level %d, key ", n, zn->level); 2171 2172 insert_zbranch(zi, zbr, n); 2173 2174 /* Insert new znode (produced by spitting) into the parent */ 2175 if (zp) { 2176 if (n == 0 && zi == znode && znode->iip == 0) 2177 correct_parent_keys(c, znode); 2178 2179 /* Locate insertion point */ 2180 n = znode->iip + 1; 2181 2182 /* Tail recursion */ 2183 zbr->key = zn->zbranch[0].key; 2184 zbr->znode = zn; 2185 zbr->lnum = 0; 2186 zbr->offs = 0; 2187 zbr->len = 0; 2188 znode = zp; 2189 2190 goto again; 2191 } 2192 2193 /* We have to split root znode */ 2194 dbg_tnc("creating new zroot at level %d", znode->level + 1); 2195 2196 zi = kzalloc(c->max_znode_sz, GFP_NOFS); 2197 if (!zi) 2198 return -ENOMEM; 2199 2200 zi->child_cnt = 2; 2201 zi->level = znode->level + 1; 2202 2203 __set_bit(DIRTY_ZNODE, &zi->flags); 2204 atomic_long_inc(&c->dirty_zn_cnt); 2205 2206 zi->zbranch[0].key = znode->zbranch[0].key; 2207 zi->zbranch[0].znode = znode; 2208 zi->zbranch[0].lnum = c->zroot.lnum; 2209 zi->zbranch[0].offs = c->zroot.offs; 2210 zi->zbranch[0].len = c->zroot.len; 2211 zi->zbranch[1].key = zn->zbranch[0].key; 2212 zi->zbranch[1].znode = zn; 2213 2214 c->zroot.lnum = 0; 2215 c->zroot.offs = 0; 2216 c->zroot.len = 0; 2217 c->zroot.znode = zi; 2218 2219 zn->parent = zi; 2220 zn->iip = 1; 2221 znode->parent = zi; 2222 znode->iip = 0; 2223 2224 return 0; 2225 } 2226 2227 /** 2228 * ubifs_tnc_add - add a node to TNC. 2229 * @c: UBIFS file-system description object 2230 * @key: key to add 2231 * @lnum: LEB number of node 2232 * @offs: node offset 2233 * @len: node length 2234 * 2235 * This function adds a node with key @key to TNC. The node may be new or it may 2236 * obsolete some existing one. Returns %0 on success or negative error code on 2237 * failure. 2238 */ 2239 int ubifs_tnc_add(struct ubifs_info *c, const union ubifs_key *key, int lnum, 2240 int offs, int len) 2241 { 2242 int found, n, err = 0; 2243 struct ubifs_znode *znode; 2244 2245 mutex_lock(&c->tnc_mutex); 2246 dbg_tnck(key, "%d:%d, len %d, key ", lnum, offs, len); 2247 found = lookup_level0_dirty(c, key, &znode, &n); 2248 if (!found) { 2249 struct ubifs_zbranch zbr; 2250 2251 zbr.znode = NULL; 2252 zbr.lnum = lnum; 2253 zbr.offs = offs; 2254 zbr.len = len; 2255 key_copy(c, key, &zbr.key); 2256 err = tnc_insert(c, znode, &zbr, n + 1); 2257 } else if (found == 1) { 2258 struct ubifs_zbranch *zbr = &znode->zbranch[n]; 2259 2260 lnc_free(zbr); 2261 err = ubifs_add_dirt(c, zbr->lnum, zbr->len); 2262 zbr->lnum = lnum; 2263 zbr->offs = offs; 2264 zbr->len = len; 2265 } else 2266 err = found; 2267 if (!err) 2268 err = dbg_check_tnc(c, 0); 2269 mutex_unlock(&c->tnc_mutex); 2270 2271 return err; 2272 } 2273 2274 /** 2275 * ubifs_tnc_replace - replace a node in the TNC only if the old node is found. 2276 * @c: UBIFS file-system description object 2277 * @key: key to add 2278 * @old_lnum: LEB number of old node 2279 * @old_offs: old node offset 2280 * @lnum: LEB number of node 2281 * @offs: node offset 2282 * @len: node length 2283 * 2284 * This function replaces a node with key @key in the TNC only if the old node 2285 * is found. This function is called by garbage collection when node are moved. 2286 * Returns %0 on success or negative error code on failure. 2287 */ 2288 int ubifs_tnc_replace(struct ubifs_info *c, const union ubifs_key *key, 2289 int old_lnum, int old_offs, int lnum, int offs, int len) 2290 { 2291 int found, n, err = 0; 2292 struct ubifs_znode *znode; 2293 2294 mutex_lock(&c->tnc_mutex); 2295 dbg_tnck(key, "old LEB %d:%d, new LEB %d:%d, len %d, key ", old_lnum, 2296 old_offs, lnum, offs, len); 2297 found = lookup_level0_dirty(c, key, &znode, &n); 2298 if (found < 0) { 2299 err = found; 2300 goto out_unlock; 2301 } 2302 2303 if (found == 1) { 2304 struct ubifs_zbranch *zbr = &znode->zbranch[n]; 2305 2306 found = 0; 2307 if (zbr->lnum == old_lnum && zbr->offs == old_offs) { 2308 lnc_free(zbr); 2309 err = ubifs_add_dirt(c, zbr->lnum, zbr->len); 2310 if (err) 2311 goto out_unlock; 2312 zbr->lnum = lnum; 2313 zbr->offs = offs; 2314 zbr->len = len; 2315 found = 1; 2316 } else if (is_hash_key(c, key)) { 2317 found = resolve_collision_directly(c, key, &znode, &n, 2318 old_lnum, old_offs); 2319 dbg_tnc("rc returned %d, znode %p, n %d, LEB %d:%d", 2320 found, znode, n, old_lnum, old_offs); 2321 if (found < 0) { 2322 err = found; 2323 goto out_unlock; 2324 } 2325 2326 if (found) { 2327 /* Ensure the znode is dirtied */ 2328 if (znode->cnext || !ubifs_zn_dirty(znode)) { 2329 znode = dirty_cow_bottom_up(c, znode); 2330 if (IS_ERR(znode)) { 2331 err = PTR_ERR(znode); 2332 goto out_unlock; 2333 } 2334 } 2335 zbr = &znode->zbranch[n]; 2336 lnc_free(zbr); 2337 err = ubifs_add_dirt(c, zbr->lnum, 2338 zbr->len); 2339 if (err) 2340 goto out_unlock; 2341 zbr->lnum = lnum; 2342 zbr->offs = offs; 2343 zbr->len = len; 2344 } 2345 } 2346 } 2347 2348 if (!found) 2349 err = ubifs_add_dirt(c, lnum, len); 2350 2351 if (!err) 2352 err = dbg_check_tnc(c, 0); 2353 2354 out_unlock: 2355 mutex_unlock(&c->tnc_mutex); 2356 return err; 2357 } 2358 2359 /** 2360 * ubifs_tnc_add_nm - add a "hashed" node to TNC. 2361 * @c: UBIFS file-system description object 2362 * @key: key to add 2363 * @lnum: LEB number of node 2364 * @offs: node offset 2365 * @len: node length 2366 * @nm: node name 2367 * 2368 * This is the same as 'ubifs_tnc_add()' but it should be used with keys which 2369 * may have collisions, like directory entry keys. 2370 */ 2371 int ubifs_tnc_add_nm(struct ubifs_info *c, const union ubifs_key *key, 2372 int lnum, int offs, int len, 2373 const struct fscrypt_name *nm) 2374 { 2375 int found, n, err = 0; 2376 struct ubifs_znode *znode; 2377 2378 mutex_lock(&c->tnc_mutex); 2379 //dbg_tnck(key, "LEB %d:%d, name '%.*s', key ", 2380 // lnum, offs, nm->len, nm->name); 2381 found = lookup_level0_dirty(c, key, &znode, &n); 2382 if (found < 0) { 2383 err = found; 2384 goto out_unlock; 2385 } 2386 2387 if (found == 1) { 2388 if (c->replaying) 2389 found = fallible_resolve_collision(c, key, &znode, &n, 2390 nm, 1); 2391 else 2392 found = resolve_collision(c, key, &znode, &n, nm); 2393 dbg_tnc("rc returned %d, znode %p, n %d", found, znode, n); 2394 if (found < 0) { 2395 err = found; 2396 goto out_unlock; 2397 } 2398 2399 /* Ensure the znode is dirtied */ 2400 if (znode->cnext || !ubifs_zn_dirty(znode)) { 2401 znode = dirty_cow_bottom_up(c, znode); 2402 if (IS_ERR(znode)) { 2403 err = PTR_ERR(znode); 2404 goto out_unlock; 2405 } 2406 } 2407 2408 if (found == 1) { 2409 struct ubifs_zbranch *zbr = &znode->zbranch[n]; 2410 2411 lnc_free(zbr); 2412 err = ubifs_add_dirt(c, zbr->lnum, zbr->len); 2413 zbr->lnum = lnum; 2414 zbr->offs = offs; 2415 zbr->len = len; 2416 goto out_unlock; 2417 } 2418 } 2419 2420 if (!found) { 2421 struct ubifs_zbranch zbr; 2422 2423 zbr.znode = NULL; 2424 zbr.lnum = lnum; 2425 zbr.offs = offs; 2426 zbr.len = len; 2427 key_copy(c, key, &zbr.key); 2428 err = tnc_insert(c, znode, &zbr, n + 1); 2429 if (err) 2430 goto out_unlock; 2431 if (c->replaying) { 2432 /* 2433 * We did not find it in the index so there may be a 2434 * dangling branch still in the index. So we remove it 2435 * by passing 'ubifs_tnc_remove_nm()' the same key but 2436 * an unmatchable name. 2437 */ 2438 struct fscrypt_name noname = { .disk_name = { .name = "", .len = 1 } }; 2439 2440 err = dbg_check_tnc(c, 0); 2441 mutex_unlock(&c->tnc_mutex); 2442 if (err) 2443 return err; 2444 return ubifs_tnc_remove_nm(c, key, &noname); 2445 } 2446 } 2447 2448 out_unlock: 2449 if (!err) 2450 err = dbg_check_tnc(c, 0); 2451 mutex_unlock(&c->tnc_mutex); 2452 return err; 2453 } 2454 2455 /** 2456 * tnc_delete - delete a znode form TNC. 2457 * @c: UBIFS file-system description object 2458 * @znode: znode to delete from 2459 * @n: zbranch slot number to delete 2460 * 2461 * This function deletes a leaf node from @n-th slot of @znode. Returns zero in 2462 * case of success and a negative error code in case of failure. 2463 */ 2464 static int tnc_delete(struct ubifs_info *c, struct ubifs_znode *znode, int n) 2465 { 2466 struct ubifs_zbranch *zbr; 2467 struct ubifs_znode *zp; 2468 int i, err; 2469 2470 /* Delete without merge for now */ 2471 ubifs_assert(znode->level == 0); 2472 ubifs_assert(n >= 0 && n < c->fanout); 2473 dbg_tnck(&znode->zbranch[n].key, "deleting key "); 2474 2475 zbr = &znode->zbranch[n]; 2476 lnc_free(zbr); 2477 2478 err = ubifs_add_dirt(c, zbr->lnum, zbr->len); 2479 if (err) { 2480 ubifs_dump_znode(c, znode); 2481 return err; 2482 } 2483 2484 /* We do not "gap" zbranch slots */ 2485 for (i = n; i < znode->child_cnt - 1; i++) 2486 znode->zbranch[i] = znode->zbranch[i + 1]; 2487 znode->child_cnt -= 1; 2488 2489 if (znode->child_cnt > 0) 2490 return 0; 2491 2492 /* 2493 * This was the last zbranch, we have to delete this znode from the 2494 * parent. 2495 */ 2496 2497 do { 2498 ubifs_assert(!ubifs_zn_obsolete(znode)); 2499 ubifs_assert(ubifs_zn_dirty(znode)); 2500 2501 zp = znode->parent; 2502 n = znode->iip; 2503 2504 atomic_long_dec(&c->dirty_zn_cnt); 2505 2506 err = insert_old_idx_znode(c, znode); 2507 if (err) 2508 return err; 2509 2510 if (znode->cnext) { 2511 __set_bit(OBSOLETE_ZNODE, &znode->flags); 2512 atomic_long_inc(&c->clean_zn_cnt); 2513 atomic_long_inc(&ubifs_clean_zn_cnt); 2514 } else 2515 kfree(znode); 2516 znode = zp; 2517 } while (znode->child_cnt == 1); /* while removing last child */ 2518 2519 /* Remove from znode, entry n - 1 */ 2520 znode->child_cnt -= 1; 2521 ubifs_assert(znode->level != 0); 2522 for (i = n; i < znode->child_cnt; i++) { 2523 znode->zbranch[i] = znode->zbranch[i + 1]; 2524 if (znode->zbranch[i].znode) 2525 znode->zbranch[i].znode->iip = i; 2526 } 2527 2528 /* 2529 * If this is the root and it has only 1 child then 2530 * collapse the tree. 2531 */ 2532 if (!znode->parent) { 2533 while (znode->child_cnt == 1 && znode->level != 0) { 2534 zp = znode; 2535 zbr = &znode->zbranch[0]; 2536 znode = get_znode(c, znode, 0); 2537 if (IS_ERR(znode)) 2538 return PTR_ERR(znode); 2539 znode = dirty_cow_znode(c, zbr); 2540 if (IS_ERR(znode)) 2541 return PTR_ERR(znode); 2542 znode->parent = NULL; 2543 znode->iip = 0; 2544 if (c->zroot.len) { 2545 err = insert_old_idx(c, c->zroot.lnum, 2546 c->zroot.offs); 2547 if (err) 2548 return err; 2549 } 2550 c->zroot.lnum = zbr->lnum; 2551 c->zroot.offs = zbr->offs; 2552 c->zroot.len = zbr->len; 2553 c->zroot.znode = znode; 2554 ubifs_assert(!ubifs_zn_obsolete(zp)); 2555 ubifs_assert(ubifs_zn_dirty(zp)); 2556 atomic_long_dec(&c->dirty_zn_cnt); 2557 2558 if (zp->cnext) { 2559 __set_bit(OBSOLETE_ZNODE, &zp->flags); 2560 atomic_long_inc(&c->clean_zn_cnt); 2561 atomic_long_inc(&ubifs_clean_zn_cnt); 2562 } else 2563 kfree(zp); 2564 } 2565 } 2566 2567 return 0; 2568 } 2569 2570 /** 2571 * ubifs_tnc_remove - remove an index entry of a node. 2572 * @c: UBIFS file-system description object 2573 * @key: key of node 2574 * 2575 * Returns %0 on success or negative error code on failure. 2576 */ 2577 int ubifs_tnc_remove(struct ubifs_info *c, const union ubifs_key *key) 2578 { 2579 int found, n, err = 0; 2580 struct ubifs_znode *znode; 2581 2582 mutex_lock(&c->tnc_mutex); 2583 dbg_tnck(key, "key "); 2584 found = lookup_level0_dirty(c, key, &znode, &n); 2585 if (found < 0) { 2586 err = found; 2587 goto out_unlock; 2588 } 2589 if (found == 1) 2590 err = tnc_delete(c, znode, n); 2591 if (!err) 2592 err = dbg_check_tnc(c, 0); 2593 2594 out_unlock: 2595 mutex_unlock(&c->tnc_mutex); 2596 return err; 2597 } 2598 2599 /** 2600 * ubifs_tnc_remove_nm - remove an index entry for a "hashed" node. 2601 * @c: UBIFS file-system description object 2602 * @key: key of node 2603 * @nm: directory entry name 2604 * 2605 * Returns %0 on success or negative error code on failure. 2606 */ 2607 int ubifs_tnc_remove_nm(struct ubifs_info *c, const union ubifs_key *key, 2608 const struct fscrypt_name *nm) 2609 { 2610 int n, err; 2611 struct ubifs_znode *znode; 2612 2613 mutex_lock(&c->tnc_mutex); 2614 //dbg_tnck(key, "%.*s, key ", nm->len, nm->name); 2615 err = lookup_level0_dirty(c, key, &znode, &n); 2616 if (err < 0) 2617 goto out_unlock; 2618 2619 if (err) { 2620 if (c->replaying) 2621 err = fallible_resolve_collision(c, key, &znode, &n, 2622 nm, 0); 2623 else 2624 err = resolve_collision(c, key, &znode, &n, nm); 2625 dbg_tnc("rc returned %d, znode %p, n %d", err, znode, n); 2626 if (err < 0) 2627 goto out_unlock; 2628 if (err) { 2629 /* Ensure the znode is dirtied */ 2630 if (znode->cnext || !ubifs_zn_dirty(znode)) { 2631 znode = dirty_cow_bottom_up(c, znode); 2632 if (IS_ERR(znode)) { 2633 err = PTR_ERR(znode); 2634 goto out_unlock; 2635 } 2636 } 2637 err = tnc_delete(c, znode, n); 2638 } 2639 } 2640 2641 out_unlock: 2642 if (!err) 2643 err = dbg_check_tnc(c, 0); 2644 mutex_unlock(&c->tnc_mutex); 2645 return err; 2646 } 2647 2648 /** 2649 * key_in_range - determine if a key falls within a range of keys. 2650 * @c: UBIFS file-system description object 2651 * @key: key to check 2652 * @from_key: lowest key in range 2653 * @to_key: highest key in range 2654 * 2655 * This function returns %1 if the key is in range and %0 otherwise. 2656 */ 2657 static int key_in_range(struct ubifs_info *c, union ubifs_key *key, 2658 union ubifs_key *from_key, union ubifs_key *to_key) 2659 { 2660 if (keys_cmp(c, key, from_key) < 0) 2661 return 0; 2662 if (keys_cmp(c, key, to_key) > 0) 2663 return 0; 2664 return 1; 2665 } 2666 2667 /** 2668 * ubifs_tnc_remove_range - remove index entries in range. 2669 * @c: UBIFS file-system description object 2670 * @from_key: lowest key to remove 2671 * @to_key: highest key to remove 2672 * 2673 * This function removes index entries starting at @from_key and ending at 2674 * @to_key. This function returns zero in case of success and a negative error 2675 * code in case of failure. 2676 */ 2677 int ubifs_tnc_remove_range(struct ubifs_info *c, union ubifs_key *from_key, 2678 union ubifs_key *to_key) 2679 { 2680 int i, n, k, err = 0; 2681 struct ubifs_znode *znode; 2682 union ubifs_key *key; 2683 2684 mutex_lock(&c->tnc_mutex); 2685 while (1) { 2686 /* Find first level 0 znode that contains keys to remove */ 2687 err = ubifs_lookup_level0(c, from_key, &znode, &n); 2688 if (err < 0) 2689 goto out_unlock; 2690 2691 if (err) 2692 key = from_key; 2693 else { 2694 err = tnc_next(c, &znode, &n); 2695 if (err == -ENOENT) { 2696 err = 0; 2697 goto out_unlock; 2698 } 2699 if (err < 0) 2700 goto out_unlock; 2701 key = &znode->zbranch[n].key; 2702 if (!key_in_range(c, key, from_key, to_key)) { 2703 err = 0; 2704 goto out_unlock; 2705 } 2706 } 2707 2708 /* Ensure the znode is dirtied */ 2709 if (znode->cnext || !ubifs_zn_dirty(znode)) { 2710 znode = dirty_cow_bottom_up(c, znode); 2711 if (IS_ERR(znode)) { 2712 err = PTR_ERR(znode); 2713 goto out_unlock; 2714 } 2715 } 2716 2717 /* Remove all keys in range except the first */ 2718 for (i = n + 1, k = 0; i < znode->child_cnt; i++, k++) { 2719 key = &znode->zbranch[i].key; 2720 if (!key_in_range(c, key, from_key, to_key)) 2721 break; 2722 lnc_free(&znode->zbranch[i]); 2723 err = ubifs_add_dirt(c, znode->zbranch[i].lnum, 2724 znode->zbranch[i].len); 2725 if (err) { 2726 ubifs_dump_znode(c, znode); 2727 goto out_unlock; 2728 } 2729 dbg_tnck(key, "removing key "); 2730 } 2731 if (k) { 2732 for (i = n + 1 + k; i < znode->child_cnt; i++) 2733 znode->zbranch[i - k] = znode->zbranch[i]; 2734 znode->child_cnt -= k; 2735 } 2736 2737 /* Now delete the first */ 2738 err = tnc_delete(c, znode, n); 2739 if (err) 2740 goto out_unlock; 2741 } 2742 2743 out_unlock: 2744 if (!err) 2745 err = dbg_check_tnc(c, 0); 2746 mutex_unlock(&c->tnc_mutex); 2747 return err; 2748 } 2749 2750 /** 2751 * ubifs_tnc_remove_ino - remove an inode from TNC. 2752 * @c: UBIFS file-system description object 2753 * @inum: inode number to remove 2754 * 2755 * This function remove inode @inum and all the extended attributes associated 2756 * with the anode from TNC and returns zero in case of success or a negative 2757 * error code in case of failure. 2758 */ 2759 int ubifs_tnc_remove_ino(struct ubifs_info *c, ino_t inum) 2760 { 2761 union ubifs_key key1, key2; 2762 struct ubifs_dent_node *xent, *pxent = NULL; 2763 struct fscrypt_name nm = {0}; 2764 2765 dbg_tnc("ino %lu", (unsigned long)inum); 2766 2767 /* 2768 * Walk all extended attribute entries and remove them together with 2769 * corresponding extended attribute inodes. 2770 */ 2771 lowest_xent_key(c, &key1, inum); 2772 while (1) { 2773 ino_t xattr_inum; 2774 int err; 2775 2776 xent = ubifs_tnc_next_ent(c, &key1, &nm); 2777 if (IS_ERR(xent)) { 2778 err = PTR_ERR(xent); 2779 if (err == -ENOENT) 2780 break; 2781 return err; 2782 } 2783 2784 xattr_inum = le64_to_cpu(xent->inum); 2785 dbg_tnc("xent '%s', ino %lu", xent->name, 2786 (unsigned long)xattr_inum); 2787 2788 fname_name(&nm) = xent->name; 2789 fname_len(&nm) = le16_to_cpu(xent->nlen); 2790 err = ubifs_tnc_remove_nm(c, &key1, &nm); 2791 if (err) { 2792 kfree(xent); 2793 return err; 2794 } 2795 2796 lowest_ino_key(c, &key1, xattr_inum); 2797 highest_ino_key(c, &key2, xattr_inum); 2798 err = ubifs_tnc_remove_range(c, &key1, &key2); 2799 if (err) { 2800 kfree(xent); 2801 return err; 2802 } 2803 2804 kfree(pxent); 2805 pxent = xent; 2806 key_read(c, &xent->key, &key1); 2807 } 2808 2809 kfree(pxent); 2810 lowest_ino_key(c, &key1, inum); 2811 highest_ino_key(c, &key2, inum); 2812 2813 return ubifs_tnc_remove_range(c, &key1, &key2); 2814 } 2815 2816 /** 2817 * ubifs_tnc_next_ent - walk directory or extended attribute entries. 2818 * @c: UBIFS file-system description object 2819 * @key: key of last entry 2820 * @nm: name of last entry found or %NULL 2821 * 2822 * This function finds and reads the next directory or extended attribute entry 2823 * after the given key (@key) if there is one. @nm is used to resolve 2824 * collisions. 2825 * 2826 * If the name of the current entry is not known and only the key is known, 2827 * @nm->name has to be %NULL. In this case the semantics of this function is a 2828 * little bit different and it returns the entry corresponding to this key, not 2829 * the next one. If the key was not found, the closest "right" entry is 2830 * returned. 2831 * 2832 * If the fist entry has to be found, @key has to contain the lowest possible 2833 * key value for this inode and @name has to be %NULL. 2834 * 2835 * This function returns the found directory or extended attribute entry node 2836 * in case of success, %-ENOENT is returned if no entry was found, and a 2837 * negative error code is returned in case of failure. 2838 */ 2839 struct ubifs_dent_node *ubifs_tnc_next_ent(struct ubifs_info *c, 2840 union ubifs_key *key, 2841 const struct fscrypt_name *nm) 2842 { 2843 int n, err, type = key_type(c, key); 2844 struct ubifs_znode *znode; 2845 struct ubifs_dent_node *dent; 2846 struct ubifs_zbranch *zbr; 2847 union ubifs_key *dkey; 2848 2849 //dbg_tnck(key, "%s ", nm->name ? (char *)nm->name : "(lowest)"); 2850 ubifs_assert(is_hash_key(c, key)); 2851 2852 mutex_lock(&c->tnc_mutex); 2853 err = ubifs_lookup_level0(c, key, &znode, &n); 2854 if (unlikely(err < 0)) 2855 goto out_unlock; 2856 2857 if (fname_len(nm) > 0) { 2858 if (err) { 2859 /* Handle collisions */ 2860 err = resolve_collision(c, key, &znode, &n, nm); 2861 dbg_tnc("rc returned %d, znode %p, n %d", 2862 err, znode, n); 2863 if (unlikely(err < 0)) 2864 goto out_unlock; 2865 } 2866 2867 /* Now find next entry */ 2868 err = tnc_next(c, &znode, &n); 2869 if (unlikely(err)) 2870 goto out_unlock; 2871 } else { 2872 /* 2873 * The full name of the entry was not given, in which case the 2874 * behavior of this function is a little different and it 2875 * returns current entry, not the next one. 2876 */ 2877 if (!err) { 2878 /* 2879 * However, the given key does not exist in the TNC 2880 * tree and @znode/@n variables contain the closest 2881 * "preceding" element. Switch to the next one. 2882 */ 2883 err = tnc_next(c, &znode, &n); 2884 if (err) 2885 goto out_unlock; 2886 } 2887 } 2888 2889 zbr = &znode->zbranch[n]; 2890 dent = kmalloc(zbr->len, GFP_NOFS); 2891 if (unlikely(!dent)) { 2892 err = -ENOMEM; 2893 goto out_unlock; 2894 } 2895 2896 /* 2897 * The above 'tnc_next()' call could lead us to the next inode, check 2898 * this. 2899 */ 2900 dkey = &zbr->key; 2901 if (key_inum(c, dkey) != key_inum(c, key) || 2902 key_type(c, dkey) != type) { 2903 err = -ENOENT; 2904 goto out_free; 2905 } 2906 2907 err = tnc_read_hashed_node(c, zbr, dent); 2908 if (unlikely(err)) 2909 goto out_free; 2910 2911 mutex_unlock(&c->tnc_mutex); 2912 return dent; 2913 2914 out_free: 2915 kfree(dent); 2916 out_unlock: 2917 mutex_unlock(&c->tnc_mutex); 2918 return ERR_PTR(err); 2919 } 2920 2921 /** 2922 * tnc_destroy_cnext - destroy left-over obsolete znodes from a failed commit. 2923 * @c: UBIFS file-system description object 2924 * 2925 * Destroy left-over obsolete znodes from a failed commit. 2926 */ 2927 static void tnc_destroy_cnext(struct ubifs_info *c) 2928 { 2929 struct ubifs_znode *cnext; 2930 2931 if (!c->cnext) 2932 return; 2933 ubifs_assert(c->cmt_state == COMMIT_BROKEN); 2934 cnext = c->cnext; 2935 do { 2936 struct ubifs_znode *znode = cnext; 2937 2938 cnext = cnext->cnext; 2939 if (ubifs_zn_obsolete(znode)) 2940 kfree(znode); 2941 } while (cnext && cnext != c->cnext); 2942 } 2943 2944 /** 2945 * ubifs_tnc_close - close TNC subsystem and free all related resources. 2946 * @c: UBIFS file-system description object 2947 */ 2948 void ubifs_tnc_close(struct ubifs_info *c) 2949 { 2950 tnc_destroy_cnext(c); 2951 if (c->zroot.znode) { 2952 long n, freed; 2953 2954 n = atomic_long_read(&c->clean_zn_cnt); 2955 freed = ubifs_destroy_tnc_subtree(c->zroot.znode); 2956 ubifs_assert(freed == n); 2957 atomic_long_sub(n, &ubifs_clean_zn_cnt); 2958 } 2959 kfree(c->gap_lebs); 2960 kfree(c->ilebs); 2961 destroy_old_idx(c); 2962 } 2963 2964 /** 2965 * left_znode - get the znode to the left. 2966 * @c: UBIFS file-system description object 2967 * @znode: znode 2968 * 2969 * This function returns a pointer to the znode to the left of @znode or NULL if 2970 * there is not one. A negative error code is returned on failure. 2971 */ 2972 static struct ubifs_znode *left_znode(struct ubifs_info *c, 2973 struct ubifs_znode *znode) 2974 { 2975 int level = znode->level; 2976 2977 while (1) { 2978 int n = znode->iip - 1; 2979 2980 /* Go up until we can go left */ 2981 znode = znode->parent; 2982 if (!znode) 2983 return NULL; 2984 if (n >= 0) { 2985 /* Now go down the rightmost branch to 'level' */ 2986 znode = get_znode(c, znode, n); 2987 if (IS_ERR(znode)) 2988 return znode; 2989 while (znode->level != level) { 2990 n = znode->child_cnt - 1; 2991 znode = get_znode(c, znode, n); 2992 if (IS_ERR(znode)) 2993 return znode; 2994 } 2995 break; 2996 } 2997 } 2998 return znode; 2999 } 3000 3001 /** 3002 * right_znode - get the znode to the right. 3003 * @c: UBIFS file-system description object 3004 * @znode: znode 3005 * 3006 * This function returns a pointer to the znode to the right of @znode or NULL 3007 * if there is not one. A negative error code is returned on failure. 3008 */ 3009 static struct ubifs_znode *right_znode(struct ubifs_info *c, 3010 struct ubifs_znode *znode) 3011 { 3012 int level = znode->level; 3013 3014 while (1) { 3015 int n = znode->iip + 1; 3016 3017 /* Go up until we can go right */ 3018 znode = znode->parent; 3019 if (!znode) 3020 return NULL; 3021 if (n < znode->child_cnt) { 3022 /* Now go down the leftmost branch to 'level' */ 3023 znode = get_znode(c, znode, n); 3024 if (IS_ERR(znode)) 3025 return znode; 3026 while (znode->level != level) { 3027 znode = get_znode(c, znode, 0); 3028 if (IS_ERR(znode)) 3029 return znode; 3030 } 3031 break; 3032 } 3033 } 3034 return znode; 3035 } 3036 3037 /** 3038 * lookup_znode - find a particular indexing node from TNC. 3039 * @c: UBIFS file-system description object 3040 * @key: index node key to lookup 3041 * @level: index node level 3042 * @lnum: index node LEB number 3043 * @offs: index node offset 3044 * 3045 * This function searches an indexing node by its first key @key and its 3046 * address @lnum:@offs. It looks up the indexing tree by pulling all indexing 3047 * nodes it traverses to TNC. This function is called for indexing nodes which 3048 * were found on the media by scanning, for example when garbage-collecting or 3049 * when doing in-the-gaps commit. This means that the indexing node which is 3050 * looked for does not have to have exactly the same leftmost key @key, because 3051 * the leftmost key may have been changed, in which case TNC will contain a 3052 * dirty znode which still refers the same @lnum:@offs. This function is clever 3053 * enough to recognize such indexing nodes. 3054 * 3055 * Note, if a znode was deleted or changed too much, then this function will 3056 * not find it. For situations like this UBIFS has the old index RB-tree 3057 * (indexed by @lnum:@offs). 3058 * 3059 * This function returns a pointer to the znode found or %NULL if it is not 3060 * found. A negative error code is returned on failure. 3061 */ 3062 static struct ubifs_znode *lookup_znode(struct ubifs_info *c, 3063 union ubifs_key *key, int level, 3064 int lnum, int offs) 3065 { 3066 struct ubifs_znode *znode, *zn; 3067 int n, nn; 3068 3069 ubifs_assert(key_type(c, key) < UBIFS_INVALID_KEY); 3070 3071 /* 3072 * The arguments have probably been read off flash, so don't assume 3073 * they are valid. 3074 */ 3075 if (level < 0) 3076 return ERR_PTR(-EINVAL); 3077 3078 /* Get the root znode */ 3079 znode = c->zroot.znode; 3080 if (!znode) { 3081 znode = ubifs_load_znode(c, &c->zroot, NULL, 0); 3082 if (IS_ERR(znode)) 3083 return znode; 3084 } 3085 /* Check if it is the one we are looking for */ 3086 if (c->zroot.lnum == lnum && c->zroot.offs == offs) 3087 return znode; 3088 /* Descend to the parent level i.e. (level + 1) */ 3089 if (level >= znode->level) 3090 return NULL; 3091 while (1) { 3092 ubifs_search_zbranch(c, znode, key, &n); 3093 if (n < 0) { 3094 /* 3095 * We reached a znode where the leftmost key is greater 3096 * than the key we are searching for. This is the same 3097 * situation as the one described in a huge comment at 3098 * the end of the 'ubifs_lookup_level0()' function. And 3099 * for exactly the same reasons we have to try to look 3100 * left before giving up. 3101 */ 3102 znode = left_znode(c, znode); 3103 if (!znode) 3104 return NULL; 3105 if (IS_ERR(znode)) 3106 return znode; 3107 ubifs_search_zbranch(c, znode, key, &n); 3108 ubifs_assert(n >= 0); 3109 } 3110 if (znode->level == level + 1) 3111 break; 3112 znode = get_znode(c, znode, n); 3113 if (IS_ERR(znode)) 3114 return znode; 3115 } 3116 /* Check if the child is the one we are looking for */ 3117 if (znode->zbranch[n].lnum == lnum && znode->zbranch[n].offs == offs) 3118 return get_znode(c, znode, n); 3119 /* If the key is unique, there is nowhere else to look */ 3120 if (!is_hash_key(c, key)) 3121 return NULL; 3122 /* 3123 * The key is not unique and so may be also in the znodes to either 3124 * side. 3125 */ 3126 zn = znode; 3127 nn = n; 3128 /* Look left */ 3129 while (1) { 3130 /* Move one branch to the left */ 3131 if (n) 3132 n -= 1; 3133 else { 3134 znode = left_znode(c, znode); 3135 if (!znode) 3136 break; 3137 if (IS_ERR(znode)) 3138 return znode; 3139 n = znode->child_cnt - 1; 3140 } 3141 /* Check it */ 3142 if (znode->zbranch[n].lnum == lnum && 3143 znode->zbranch[n].offs == offs) 3144 return get_znode(c, znode, n); 3145 /* Stop if the key is less than the one we are looking for */ 3146 if (keys_cmp(c, &znode->zbranch[n].key, key) < 0) 3147 break; 3148 } 3149 /* Back to the middle */ 3150 znode = zn; 3151 n = nn; 3152 /* Look right */ 3153 while (1) { 3154 /* Move one branch to the right */ 3155 if (++n >= znode->child_cnt) { 3156 znode = right_znode(c, znode); 3157 if (!znode) 3158 break; 3159 if (IS_ERR(znode)) 3160 return znode; 3161 n = 0; 3162 } 3163 /* Check it */ 3164 if (znode->zbranch[n].lnum == lnum && 3165 znode->zbranch[n].offs == offs) 3166 return get_znode(c, znode, n); 3167 /* Stop if the key is greater than the one we are looking for */ 3168 if (keys_cmp(c, &znode->zbranch[n].key, key) > 0) 3169 break; 3170 } 3171 return NULL; 3172 } 3173 3174 /** 3175 * is_idx_node_in_tnc - determine if an index node is in the TNC. 3176 * @c: UBIFS file-system description object 3177 * @key: key of index node 3178 * @level: index node level 3179 * @lnum: LEB number of index node 3180 * @offs: offset of index node 3181 * 3182 * This function returns %0 if the index node is not referred to in the TNC, %1 3183 * if the index node is referred to in the TNC and the corresponding znode is 3184 * dirty, %2 if an index node is referred to in the TNC and the corresponding 3185 * znode is clean, and a negative error code in case of failure. 3186 * 3187 * Note, the @key argument has to be the key of the first child. Also note, 3188 * this function relies on the fact that 0:0 is never a valid LEB number and 3189 * offset for a main-area node. 3190 */ 3191 int is_idx_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, int level, 3192 int lnum, int offs) 3193 { 3194 struct ubifs_znode *znode; 3195 3196 znode = lookup_znode(c, key, level, lnum, offs); 3197 if (!znode) 3198 return 0; 3199 if (IS_ERR(znode)) 3200 return PTR_ERR(znode); 3201 3202 return ubifs_zn_dirty(znode) ? 1 : 2; 3203 } 3204 3205 /** 3206 * is_leaf_node_in_tnc - determine if a non-indexing not is in the TNC. 3207 * @c: UBIFS file-system description object 3208 * @key: node key 3209 * @lnum: node LEB number 3210 * @offs: node offset 3211 * 3212 * This function returns %1 if the node is referred to in the TNC, %0 if it is 3213 * not, and a negative error code in case of failure. 3214 * 3215 * Note, this function relies on the fact that 0:0 is never a valid LEB number 3216 * and offset for a main-area node. 3217 */ 3218 static int is_leaf_node_in_tnc(struct ubifs_info *c, union ubifs_key *key, 3219 int lnum, int offs) 3220 { 3221 struct ubifs_zbranch *zbr; 3222 struct ubifs_znode *znode, *zn; 3223 int n, found, err, nn; 3224 const int unique = !is_hash_key(c, key); 3225 3226 found = ubifs_lookup_level0(c, key, &znode, &n); 3227 if (found < 0) 3228 return found; /* Error code */ 3229 if (!found) 3230 return 0; 3231 zbr = &znode->zbranch[n]; 3232 if (lnum == zbr->lnum && offs == zbr->offs) 3233 return 1; /* Found it */ 3234 if (unique) 3235 return 0; 3236 /* 3237 * Because the key is not unique, we have to look left 3238 * and right as well 3239 */ 3240 zn = znode; 3241 nn = n; 3242 /* Look left */ 3243 while (1) { 3244 err = tnc_prev(c, &znode, &n); 3245 if (err == -ENOENT) 3246 break; 3247 if (err) 3248 return err; 3249 if (keys_cmp(c, key, &znode->zbranch[n].key)) 3250 break; 3251 zbr = &znode->zbranch[n]; 3252 if (lnum == zbr->lnum && offs == zbr->offs) 3253 return 1; /* Found it */ 3254 } 3255 /* Look right */ 3256 znode = zn; 3257 n = nn; 3258 while (1) { 3259 err = tnc_next(c, &znode, &n); 3260 if (err) { 3261 if (err == -ENOENT) 3262 return 0; 3263 return err; 3264 } 3265 if (keys_cmp(c, key, &znode->zbranch[n].key)) 3266 break; 3267 zbr = &znode->zbranch[n]; 3268 if (lnum == zbr->lnum && offs == zbr->offs) 3269 return 1; /* Found it */ 3270 } 3271 return 0; 3272 } 3273 3274 /** 3275 * ubifs_tnc_has_node - determine whether a node is in the TNC. 3276 * @c: UBIFS file-system description object 3277 * @key: node key 3278 * @level: index node level (if it is an index node) 3279 * @lnum: node LEB number 3280 * @offs: node offset 3281 * @is_idx: non-zero if the node is an index node 3282 * 3283 * This function returns %1 if the node is in the TNC, %0 if it is not, and a 3284 * negative error code in case of failure. For index nodes, @key has to be the 3285 * key of the first child. An index node is considered to be in the TNC only if 3286 * the corresponding znode is clean or has not been loaded. 3287 */ 3288 int ubifs_tnc_has_node(struct ubifs_info *c, union ubifs_key *key, int level, 3289 int lnum, int offs, int is_idx) 3290 { 3291 int err; 3292 3293 mutex_lock(&c->tnc_mutex); 3294 if (is_idx) { 3295 err = is_idx_node_in_tnc(c, key, level, lnum, offs); 3296 if (err < 0) 3297 goto out_unlock; 3298 if (err == 1) 3299 /* The index node was found but it was dirty */ 3300 err = 0; 3301 else if (err == 2) 3302 /* The index node was found and it was clean */ 3303 err = 1; 3304 else 3305 BUG_ON(err != 0); 3306 } else 3307 err = is_leaf_node_in_tnc(c, key, lnum, offs); 3308 3309 out_unlock: 3310 mutex_unlock(&c->tnc_mutex); 3311 return err; 3312 } 3313 3314 /** 3315 * ubifs_dirty_idx_node - dirty an index node. 3316 * @c: UBIFS file-system description object 3317 * @key: index node key 3318 * @level: index node level 3319 * @lnum: index node LEB number 3320 * @offs: index node offset 3321 * 3322 * This function loads and dirties an index node so that it can be garbage 3323 * collected. The @key argument has to be the key of the first child. This 3324 * function relies on the fact that 0:0 is never a valid LEB number and offset 3325 * for a main-area node. Returns %0 on success and a negative error code on 3326 * failure. 3327 */ 3328 int ubifs_dirty_idx_node(struct ubifs_info *c, union ubifs_key *key, int level, 3329 int lnum, int offs) 3330 { 3331 struct ubifs_znode *znode; 3332 int err = 0; 3333 3334 mutex_lock(&c->tnc_mutex); 3335 znode = lookup_znode(c, key, level, lnum, offs); 3336 if (!znode) 3337 goto out_unlock; 3338 if (IS_ERR(znode)) { 3339 err = PTR_ERR(znode); 3340 goto out_unlock; 3341 } 3342 znode = dirty_cow_bottom_up(c, znode); 3343 if (IS_ERR(znode)) { 3344 err = PTR_ERR(znode); 3345 goto out_unlock; 3346 } 3347 3348 out_unlock: 3349 mutex_unlock(&c->tnc_mutex); 3350 return err; 3351 } 3352 3353 /** 3354 * dbg_check_inode_size - check if inode size is correct. 3355 * @c: UBIFS file-system description object 3356 * @inum: inode number 3357 * @size: inode size 3358 * 3359 * This function makes sure that the inode size (@size) is correct and it does 3360 * not have any pages beyond @size. Returns zero if the inode is OK, %-EINVAL 3361 * if it has a data page beyond @size, and other negative error code in case of 3362 * other errors. 3363 */ 3364 int dbg_check_inode_size(struct ubifs_info *c, const struct inode *inode, 3365 loff_t size) 3366 { 3367 int err, n; 3368 union ubifs_key from_key, to_key, *key; 3369 struct ubifs_znode *znode; 3370 unsigned int block; 3371 3372 if (!S_ISREG(inode->i_mode)) 3373 return 0; 3374 if (!dbg_is_chk_gen(c)) 3375 return 0; 3376 3377 block = (size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT; 3378 data_key_init(c, &from_key, inode->i_ino, block); 3379 highest_data_key(c, &to_key, inode->i_ino); 3380 3381 mutex_lock(&c->tnc_mutex); 3382 err = ubifs_lookup_level0(c, &from_key, &znode, &n); 3383 if (err < 0) 3384 goto out_unlock; 3385 3386 if (err) { 3387 key = &from_key; 3388 goto out_dump; 3389 } 3390 3391 err = tnc_next(c, &znode, &n); 3392 if (err == -ENOENT) { 3393 err = 0; 3394 goto out_unlock; 3395 } 3396 if (err < 0) 3397 goto out_unlock; 3398 3399 ubifs_assert(err == 0); 3400 key = &znode->zbranch[n].key; 3401 if (!key_in_range(c, key, &from_key, &to_key)) 3402 goto out_unlock; 3403 3404 out_dump: 3405 block = key_block(c, key); 3406 ubifs_err(c, "inode %lu has size %lld, but there are data at offset %lld", 3407 (unsigned long)inode->i_ino, size, 3408 ((loff_t)block) << UBIFS_BLOCK_SHIFT); 3409 mutex_unlock(&c->tnc_mutex); 3410 ubifs_dump_inode(c, inode); 3411 dump_stack(); 3412 return -EINVAL; 3413 3414 out_unlock: 3415 mutex_unlock(&c->tnc_mutex); 3416 return err; 3417 } 3418