1 /* 2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README 3 */ 4 5 /* 6 * Written by Anatoly P. Pinchuk pap@namesys.botik.ru 7 * Programm System Institute 8 * Pereslavl-Zalessky Russia 9 */ 10 11 /* 12 * This file contains functions dealing with S+tree 13 * 14 * B_IS_IN_TREE 15 * copy_item_head 16 * comp_short_keys 17 * comp_keys 18 * comp_short_le_keys 19 * le_key2cpu_key 20 * comp_le_keys 21 * bin_search 22 * get_lkey 23 * get_rkey 24 * key_in_buffer 25 * decrement_bcount 26 * reiserfs_check_path 27 * pathrelse_and_restore 28 * pathrelse 29 * search_by_key_reada 30 * search_by_key 31 * search_for_position_by_key 32 * comp_items 33 * prepare_for_direct_item 34 * prepare_for_direntry_item 35 * prepare_for_delete_or_cut 36 * calc_deleted_bytes_number 37 * init_tb_struct 38 * padd_item 39 * reiserfs_delete_item 40 * reiserfs_delete_solid_item 41 * reiserfs_delete_object 42 * maybe_indirect_to_direct 43 * indirect_to_direct_roll_back 44 * reiserfs_cut_from_item 45 * truncate_directory 46 * reiserfs_do_truncate 47 * reiserfs_paste_into_item 48 * reiserfs_insert_item 49 */ 50 51 #include <linux/time.h> 52 #include <linux/string.h> 53 #include <linux/pagemap.h> 54 #include "reiserfs.h" 55 #include <linux/buffer_head.h> 56 #include <linux/quotaops.h> 57 58 /* Does the buffer contain a disk block which is in the tree. */ 59 inline int B_IS_IN_TREE(const struct buffer_head *bh) 60 { 61 62 RFALSE(B_LEVEL(bh) > MAX_HEIGHT, 63 "PAP-1010: block (%b) has too big level (%z)", bh, bh); 64 65 return (B_LEVEL(bh) != FREE_LEVEL); 66 } 67 68 // 69 // to gets item head in le form 70 // 71 inline void copy_item_head(struct item_head *to, 72 const struct item_head *from) 73 { 74 memcpy(to, from, IH_SIZE); 75 } 76 77 /* k1 is pointer to on-disk structure which is stored in little-endian 78 form. k2 is pointer to cpu variable. For key of items of the same 79 object this returns 0. 80 Returns: -1 if key1 < key2 81 0 if key1 == key2 82 1 if key1 > key2 */ 83 inline int comp_short_keys(const struct reiserfs_key *le_key, 84 const struct cpu_key *cpu_key) 85 { 86 __u32 n; 87 n = le32_to_cpu(le_key->k_dir_id); 88 if (n < cpu_key->on_disk_key.k_dir_id) 89 return -1; 90 if (n > cpu_key->on_disk_key.k_dir_id) 91 return 1; 92 n = le32_to_cpu(le_key->k_objectid); 93 if (n < cpu_key->on_disk_key.k_objectid) 94 return -1; 95 if (n > cpu_key->on_disk_key.k_objectid) 96 return 1; 97 return 0; 98 } 99 100 /* k1 is pointer to on-disk structure which is stored in little-endian 101 form. k2 is pointer to cpu variable. 102 Compare keys using all 4 key fields. 103 Returns: -1 if key1 < key2 0 104 if key1 = key2 1 if key1 > key2 */ 105 static inline int comp_keys(const struct reiserfs_key *le_key, 106 const struct cpu_key *cpu_key) 107 { 108 int retval; 109 110 retval = comp_short_keys(le_key, cpu_key); 111 if (retval) 112 return retval; 113 if (le_key_k_offset(le_key_version(le_key), le_key) < 114 cpu_key_k_offset(cpu_key)) 115 return -1; 116 if (le_key_k_offset(le_key_version(le_key), le_key) > 117 cpu_key_k_offset(cpu_key)) 118 return 1; 119 120 if (cpu_key->key_length == 3) 121 return 0; 122 123 /* this part is needed only when tail conversion is in progress */ 124 if (le_key_k_type(le_key_version(le_key), le_key) < 125 cpu_key_k_type(cpu_key)) 126 return -1; 127 128 if (le_key_k_type(le_key_version(le_key), le_key) > 129 cpu_key_k_type(cpu_key)) 130 return 1; 131 132 return 0; 133 } 134 135 inline int comp_short_le_keys(const struct reiserfs_key *key1, 136 const struct reiserfs_key *key2) 137 { 138 __u32 *k1_u32, *k2_u32; 139 int key_length = REISERFS_SHORT_KEY_LEN; 140 141 k1_u32 = (__u32 *) key1; 142 k2_u32 = (__u32 *) key2; 143 for (; key_length--; ++k1_u32, ++k2_u32) { 144 if (le32_to_cpu(*k1_u32) < le32_to_cpu(*k2_u32)) 145 return -1; 146 if (le32_to_cpu(*k1_u32) > le32_to_cpu(*k2_u32)) 147 return 1; 148 } 149 return 0; 150 } 151 152 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from) 153 { 154 int version; 155 to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id); 156 to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid); 157 158 // find out version of the key 159 version = le_key_version(from); 160 to->version = version; 161 to->on_disk_key.k_offset = le_key_k_offset(version, from); 162 to->on_disk_key.k_type = le_key_k_type(version, from); 163 } 164 165 // this does not say which one is bigger, it only returns 1 if keys 166 // are not equal, 0 otherwise 167 inline int comp_le_keys(const struct reiserfs_key *k1, 168 const struct reiserfs_key *k2) 169 { 170 return memcmp(k1, k2, sizeof(struct reiserfs_key)); 171 } 172 173 /************************************************************************** 174 * Binary search toolkit function * 175 * Search for an item in the array by the item key * 176 * Returns: 1 if found, 0 if not found; * 177 * *pos = number of the searched element if found, else the * 178 * number of the first element that is larger than key. * 179 **************************************************************************/ 180 /* For those not familiar with binary search: lbound is the leftmost item that it 181 could be, rbound the rightmost item that it could be. We examine the item 182 halfway between lbound and rbound, and that tells us either that we can increase 183 lbound, or decrease rbound, or that we have found it, or if lbound <= rbound that 184 there are no possible items, and we have not found it. With each examination we 185 cut the number of possible items it could be by one more than half rounded down, 186 or we find it. */ 187 static inline int bin_search(const void *key, /* Key to search for. */ 188 const void *base, /* First item in the array. */ 189 int num, /* Number of items in the array. */ 190 int width, /* Item size in the array. 191 searched. Lest the reader be 192 confused, note that this is crafted 193 as a general function, and when it 194 is applied specifically to the array 195 of item headers in a node, width 196 is actually the item header size not 197 the item size. */ 198 int *pos /* Number of the searched for element. */ 199 ) 200 { 201 int rbound, lbound, j; 202 203 for (j = ((rbound = num - 1) + (lbound = 0)) / 2; 204 lbound <= rbound; j = (rbound + lbound) / 2) 205 switch (comp_keys 206 ((struct reiserfs_key *)((char *)base + j * width), 207 (struct cpu_key *)key)) { 208 case -1: 209 lbound = j + 1; 210 continue; 211 case 1: 212 rbound = j - 1; 213 continue; 214 case 0: 215 *pos = j; 216 return ITEM_FOUND; /* Key found in the array. */ 217 } 218 219 /* bin_search did not find given key, it returns position of key, 220 that is minimal and greater than the given one. */ 221 *pos = lbound; 222 return ITEM_NOT_FOUND; 223 } 224 225 226 /* Minimal possible key. It is never in the tree. */ 227 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} }; 228 229 /* Maximal possible key. It is never in the tree. */ 230 static const struct reiserfs_key MAX_KEY = { 231 __constant_cpu_to_le32(0xffffffff), 232 __constant_cpu_to_le32(0xffffffff), 233 {{__constant_cpu_to_le32(0xffffffff), 234 __constant_cpu_to_le32(0xffffffff)},} 235 }; 236 237 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom 238 of the path, and going upwards. We must check the path's validity at each step. If the key is not in 239 the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this 240 case we return a special key, either MIN_KEY or MAX_KEY. */ 241 static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path, 242 const struct super_block *sb) 243 { 244 int position, path_offset = chk_path->path_length; 245 struct buffer_head *parent; 246 247 RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET, 248 "PAP-5010: invalid offset in the path"); 249 250 /* While not higher in path than first element. */ 251 while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) { 252 253 RFALSE(!buffer_uptodate 254 (PATH_OFFSET_PBUFFER(chk_path, path_offset)), 255 "PAP-5020: parent is not uptodate"); 256 257 /* Parent at the path is not in the tree now. */ 258 if (!B_IS_IN_TREE 259 (parent = 260 PATH_OFFSET_PBUFFER(chk_path, path_offset))) 261 return &MAX_KEY; 262 /* Check whether position in the parent is correct. */ 263 if ((position = 264 PATH_OFFSET_POSITION(chk_path, 265 path_offset)) > 266 B_NR_ITEMS(parent)) 267 return &MAX_KEY; 268 /* Check whether parent at the path really points to the child. */ 269 if (B_N_CHILD_NUM(parent, position) != 270 PATH_OFFSET_PBUFFER(chk_path, 271 path_offset + 1)->b_blocknr) 272 return &MAX_KEY; 273 /* Return delimiting key if position in the parent is not equal to zero. */ 274 if (position) 275 return B_N_PDELIM_KEY(parent, position - 1); 276 } 277 /* Return MIN_KEY if we are in the root of the buffer tree. */ 278 if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)-> 279 b_blocknr == SB_ROOT_BLOCK(sb)) 280 return &MIN_KEY; 281 return &MAX_KEY; 282 } 283 284 /* Get delimiting key of the buffer at the path and its right neighbor. */ 285 inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path, 286 const struct super_block *sb) 287 { 288 int position, path_offset = chk_path->path_length; 289 struct buffer_head *parent; 290 291 RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET, 292 "PAP-5030: invalid offset in the path"); 293 294 while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) { 295 296 RFALSE(!buffer_uptodate 297 (PATH_OFFSET_PBUFFER(chk_path, path_offset)), 298 "PAP-5040: parent is not uptodate"); 299 300 /* Parent at the path is not in the tree now. */ 301 if (!B_IS_IN_TREE 302 (parent = 303 PATH_OFFSET_PBUFFER(chk_path, path_offset))) 304 return &MIN_KEY; 305 /* Check whether position in the parent is correct. */ 306 if ((position = 307 PATH_OFFSET_POSITION(chk_path, 308 path_offset)) > 309 B_NR_ITEMS(parent)) 310 return &MIN_KEY; 311 /* Check whether parent at the path really points to the child. */ 312 if (B_N_CHILD_NUM(parent, position) != 313 PATH_OFFSET_PBUFFER(chk_path, 314 path_offset + 1)->b_blocknr) 315 return &MIN_KEY; 316 /* Return delimiting key if position in the parent is not the last one. */ 317 if (position != B_NR_ITEMS(parent)) 318 return B_N_PDELIM_KEY(parent, position); 319 } 320 /* Return MAX_KEY if we are in the root of the buffer tree. */ 321 if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)-> 322 b_blocknr == SB_ROOT_BLOCK(sb)) 323 return &MAX_KEY; 324 return &MIN_KEY; 325 } 326 327 /* Check whether a key is contained in the tree rooted from a buffer at a path. */ 328 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in 329 the path. These delimiting keys are stored at least one level above that buffer in the tree. If the 330 buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in 331 this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */ 332 static inline int key_in_buffer(struct treepath *chk_path, /* Path which should be checked. */ 333 const struct cpu_key *key, /* Key which should be checked. */ 334 struct super_block *sb 335 ) 336 { 337 338 RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET 339 || chk_path->path_length > MAX_HEIGHT, 340 "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)", 341 key, chk_path->path_length); 342 RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev, 343 "PAP-5060: device must not be NODEV"); 344 345 if (comp_keys(get_lkey(chk_path, sb), key) == 1) 346 /* left delimiting key is bigger, that the key we look for */ 347 return 0; 348 /* if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */ 349 if (comp_keys(get_rkey(chk_path, sb), key) != 1) 350 /* key must be less than right delimitiing key */ 351 return 0; 352 return 1; 353 } 354 355 int reiserfs_check_path(struct treepath *p) 356 { 357 RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET, 358 "path not properly relsed"); 359 return 0; 360 } 361 362 /* Drop the reference to each buffer in a path and restore 363 * dirty bits clean when preparing the buffer for the log. 364 * This version should only be called from fix_nodes() */ 365 void pathrelse_and_restore(struct super_block *sb, 366 struct treepath *search_path) 367 { 368 int path_offset = search_path->path_length; 369 370 RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET, 371 "clm-4000: invalid path offset"); 372 373 while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) { 374 struct buffer_head *bh; 375 bh = PATH_OFFSET_PBUFFER(search_path, path_offset--); 376 reiserfs_restore_prepared_buffer(sb, bh); 377 brelse(bh); 378 } 379 search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET; 380 } 381 382 /* Drop the reference to each buffer in a path */ 383 void pathrelse(struct treepath *search_path) 384 { 385 int path_offset = search_path->path_length; 386 387 RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET, 388 "PAP-5090: invalid path offset"); 389 390 while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) 391 brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--)); 392 393 search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET; 394 } 395 396 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh) 397 { 398 struct block_head *blkh; 399 struct item_head *ih; 400 int used_space; 401 int prev_location; 402 int i; 403 int nr; 404 405 blkh = (struct block_head *)buf; 406 if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) { 407 reiserfs_warning(NULL, "reiserfs-5080", 408 "this should be caught earlier"); 409 return 0; 410 } 411 412 nr = blkh_nr_item(blkh); 413 if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) { 414 /* item number is too big or too small */ 415 reiserfs_warning(NULL, "reiserfs-5081", 416 "nr_item seems wrong: %z", bh); 417 return 0; 418 } 419 ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1; 420 used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih)); 421 if (used_space != blocksize - blkh_free_space(blkh)) { 422 /* free space does not match to calculated amount of use space */ 423 reiserfs_warning(NULL, "reiserfs-5082", 424 "free space seems wrong: %z", bh); 425 return 0; 426 } 427 // FIXME: it is_leaf will hit performance too much - we may have 428 // return 1 here 429 430 /* check tables of item heads */ 431 ih = (struct item_head *)(buf + BLKH_SIZE); 432 prev_location = blocksize; 433 for (i = 0; i < nr; i++, ih++) { 434 if (le_ih_k_type(ih) == TYPE_ANY) { 435 reiserfs_warning(NULL, "reiserfs-5083", 436 "wrong item type for item %h", 437 ih); 438 return 0; 439 } 440 if (ih_location(ih) >= blocksize 441 || ih_location(ih) < IH_SIZE * nr) { 442 reiserfs_warning(NULL, "reiserfs-5084", 443 "item location seems wrong: %h", 444 ih); 445 return 0; 446 } 447 if (ih_item_len(ih) < 1 448 || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) { 449 reiserfs_warning(NULL, "reiserfs-5085", 450 "item length seems wrong: %h", 451 ih); 452 return 0; 453 } 454 if (prev_location - ih_location(ih) != ih_item_len(ih)) { 455 reiserfs_warning(NULL, "reiserfs-5086", 456 "item location seems wrong " 457 "(second one): %h", ih); 458 return 0; 459 } 460 prev_location = ih_location(ih); 461 } 462 463 // one may imagine much more checks 464 return 1; 465 } 466 467 /* returns 1 if buf looks like an internal node, 0 otherwise */ 468 static int is_internal(char *buf, int blocksize, struct buffer_head *bh) 469 { 470 struct block_head *blkh; 471 int nr; 472 int used_space; 473 474 blkh = (struct block_head *)buf; 475 nr = blkh_level(blkh); 476 if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) { 477 /* this level is not possible for internal nodes */ 478 reiserfs_warning(NULL, "reiserfs-5087", 479 "this should be caught earlier"); 480 return 0; 481 } 482 483 nr = blkh_nr_item(blkh); 484 if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) { 485 /* for internal which is not root we might check min number of keys */ 486 reiserfs_warning(NULL, "reiserfs-5088", 487 "number of key seems wrong: %z", bh); 488 return 0; 489 } 490 491 used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1); 492 if (used_space != blocksize - blkh_free_space(blkh)) { 493 reiserfs_warning(NULL, "reiserfs-5089", 494 "free space seems wrong: %z", bh); 495 return 0; 496 } 497 // one may imagine much more checks 498 return 1; 499 } 500 501 // make sure that bh contains formatted node of reiserfs tree of 502 // 'level'-th level 503 static int is_tree_node(struct buffer_head *bh, int level) 504 { 505 if (B_LEVEL(bh) != level) { 506 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does " 507 "not match to the expected one %d", 508 B_LEVEL(bh), level); 509 return 0; 510 } 511 if (level == DISK_LEAF_NODE_LEVEL) 512 return is_leaf(bh->b_data, bh->b_size, bh); 513 514 return is_internal(bh->b_data, bh->b_size, bh); 515 } 516 517 #define SEARCH_BY_KEY_READA 16 518 519 /* 520 * The function is NOT SCHEDULE-SAFE! 521 * It might unlock the write lock if we needed to wait for a block 522 * to be read. Note that in this case it won't recover the lock to avoid 523 * high contention resulting from too much lock requests, especially 524 * the caller (search_by_key) will perform other schedule-unsafe 525 * operations just after calling this function. 526 * 527 * @return depth of lock to be restored after read completes 528 */ 529 static int search_by_key_reada(struct super_block *s, 530 struct buffer_head **bh, 531 b_blocknr_t *b, int num) 532 { 533 int i, j; 534 int depth = -1; 535 536 for (i = 0; i < num; i++) { 537 bh[i] = sb_getblk(s, b[i]); 538 } 539 /* 540 * We are going to read some blocks on which we 541 * have a reference. It's safe, though we might be 542 * reading blocks concurrently changed if we release 543 * the lock. But it's still fine because we check later 544 * if the tree changed 545 */ 546 for (j = 0; j < i; j++) { 547 /* 548 * note, this needs attention if we are getting rid of the BKL 549 * you have to make sure the prepared bit isn't set on this buffer 550 */ 551 if (!buffer_uptodate(bh[j])) { 552 if (depth == -1) 553 depth = reiserfs_write_unlock_nested(s); 554 ll_rw_block(READA, 1, bh + j); 555 } 556 brelse(bh[j]); 557 } 558 return depth; 559 } 560 561 /************************************************************************** 562 * Algorithm SearchByKey * 563 * look for item in the Disk S+Tree by its key * 564 * Input: sb - super block * 565 * key - pointer to the key to search * 566 * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR * 567 * search_path - path from the root to the needed leaf * 568 **************************************************************************/ 569 570 /* This function fills up the path from the root to the leaf as it 571 descends the tree looking for the key. It uses reiserfs_bread to 572 try to find buffers in the cache given their block number. If it 573 does not find them in the cache it reads them from disk. For each 574 node search_by_key finds using reiserfs_bread it then uses 575 bin_search to look through that node. bin_search will find the 576 position of the block_number of the next node if it is looking 577 through an internal node. If it is looking through a leaf node 578 bin_search will find the position of the item which has key either 579 equal to given key, or which is the maximal key less than the given 580 key. search_by_key returns a path that must be checked for the 581 correctness of the top of the path but need not be checked for the 582 correctness of the bottom of the path */ 583 /* The function is NOT SCHEDULE-SAFE! */ 584 int search_by_key(struct super_block *sb, const struct cpu_key *key, /* Key to search. */ 585 struct treepath *search_path,/* This structure was 586 allocated and initialized 587 by the calling 588 function. It is filled up 589 by this function. */ 590 int stop_level /* How far down the tree to search. To 591 stop at leaf level - set to 592 DISK_LEAF_NODE_LEVEL */ 593 ) 594 { 595 b_blocknr_t block_number; 596 int expected_level; 597 struct buffer_head *bh; 598 struct path_element *last_element; 599 int node_level, retval; 600 int right_neighbor_of_leaf_node; 601 int fs_gen; 602 struct buffer_head *reada_bh[SEARCH_BY_KEY_READA]; 603 b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA]; 604 int reada_count = 0; 605 606 #ifdef CONFIG_REISERFS_CHECK 607 int repeat_counter = 0; 608 #endif 609 610 PROC_INFO_INC(sb, search_by_key); 611 612 /* As we add each node to a path we increase its count. This means that 613 we must be careful to release all nodes in a path before we either 614 discard the path struct or re-use the path struct, as we do here. */ 615 616 pathrelse(search_path); 617 618 right_neighbor_of_leaf_node = 0; 619 620 /* With each iteration of this loop we search through the items in the 621 current node, and calculate the next current node(next path element) 622 for the next iteration of this loop.. */ 623 block_number = SB_ROOT_BLOCK(sb); 624 expected_level = -1; 625 while (1) { 626 627 #ifdef CONFIG_REISERFS_CHECK 628 if (!(++repeat_counter % 50000)) 629 reiserfs_warning(sb, "PAP-5100", 630 "%s: there were %d iterations of " 631 "while loop looking for key %K", 632 current->comm, repeat_counter, 633 key); 634 #endif 635 636 /* prep path to have another element added to it. */ 637 last_element = 638 PATH_OFFSET_PELEMENT(search_path, 639 ++search_path->path_length); 640 fs_gen = get_generation(sb); 641 642 /* Read the next tree node, and set the last element in the path to 643 have a pointer to it. */ 644 if ((bh = last_element->pe_buffer = 645 sb_getblk(sb, block_number))) { 646 647 /* 648 * We'll need to drop the lock if we encounter any 649 * buffers that need to be read. If all of them are 650 * already up to date, we don't need to drop the lock. 651 */ 652 int depth = -1; 653 654 if (!buffer_uptodate(bh) && reada_count > 1) 655 depth = search_by_key_reada(sb, reada_bh, 656 reada_blocks, reada_count); 657 658 if (!buffer_uptodate(bh) && depth == -1) 659 depth = reiserfs_write_unlock_nested(sb); 660 661 ll_rw_block(READ, 1, &bh); 662 wait_on_buffer(bh); 663 664 if (depth != -1) 665 reiserfs_write_lock_nested(sb, depth); 666 if (!buffer_uptodate(bh)) 667 goto io_error; 668 } else { 669 io_error: 670 search_path->path_length--; 671 pathrelse(search_path); 672 return IO_ERROR; 673 } 674 reada_count = 0; 675 if (expected_level == -1) 676 expected_level = SB_TREE_HEIGHT(sb); 677 expected_level--; 678 679 /* It is possible that schedule occurred. We must check whether the key 680 to search is still in the tree rooted from the current buffer. If 681 not then repeat search from the root. */ 682 if (fs_changed(fs_gen, sb) && 683 (!B_IS_IN_TREE(bh) || 684 B_LEVEL(bh) != expected_level || 685 !key_in_buffer(search_path, key, sb))) { 686 PROC_INFO_INC(sb, search_by_key_fs_changed); 687 PROC_INFO_INC(sb, search_by_key_restarted); 688 PROC_INFO_INC(sb, 689 sbk_restarted[expected_level - 1]); 690 pathrelse(search_path); 691 692 /* Get the root block number so that we can repeat the search 693 starting from the root. */ 694 block_number = SB_ROOT_BLOCK(sb); 695 expected_level = -1; 696 right_neighbor_of_leaf_node = 0; 697 698 /* repeat search from the root */ 699 continue; 700 } 701 702 /* only check that the key is in the buffer if key is not 703 equal to the MAX_KEY. Latter case is only possible in 704 "finish_unfinished()" processing during mount. */ 705 RFALSE(comp_keys(&MAX_KEY, key) && 706 !key_in_buffer(search_path, key, sb), 707 "PAP-5130: key is not in the buffer"); 708 #ifdef CONFIG_REISERFS_CHECK 709 if (REISERFS_SB(sb)->cur_tb) { 710 print_cur_tb("5140"); 711 reiserfs_panic(sb, "PAP-5140", 712 "schedule occurred in do_balance!"); 713 } 714 #endif 715 716 // make sure, that the node contents look like a node of 717 // certain level 718 if (!is_tree_node(bh, expected_level)) { 719 reiserfs_error(sb, "vs-5150", 720 "invalid format found in block %ld. " 721 "Fsck?", bh->b_blocknr); 722 pathrelse(search_path); 723 return IO_ERROR; 724 } 725 726 /* ok, we have acquired next formatted node in the tree */ 727 node_level = B_LEVEL(bh); 728 729 PROC_INFO_BH_STAT(sb, bh, node_level - 1); 730 731 RFALSE(node_level < stop_level, 732 "vs-5152: tree level (%d) is less than stop level (%d)", 733 node_level, stop_level); 734 735 retval = bin_search(key, B_N_PITEM_HEAD(bh, 0), 736 B_NR_ITEMS(bh), 737 (node_level == 738 DISK_LEAF_NODE_LEVEL) ? IH_SIZE : 739 KEY_SIZE, 740 &(last_element->pe_position)); 741 if (node_level == stop_level) { 742 return retval; 743 } 744 745 /* we are not in the stop level */ 746 if (retval == ITEM_FOUND) 747 /* item has been found, so we choose the pointer which is to the right of the found one */ 748 last_element->pe_position++; 749 750 /* if item was not found we choose the position which is to 751 the left of the found item. This requires no code, 752 bin_search did it already. */ 753 754 /* So we have chosen a position in the current node which is 755 an internal node. Now we calculate child block number by 756 position in the node. */ 757 block_number = 758 B_N_CHILD_NUM(bh, last_element->pe_position); 759 760 /* if we are going to read leaf nodes, try for read ahead as well */ 761 if ((search_path->reada & PATH_READA) && 762 node_level == DISK_LEAF_NODE_LEVEL + 1) { 763 int pos = last_element->pe_position; 764 int limit = B_NR_ITEMS(bh); 765 struct reiserfs_key *le_key; 766 767 if (search_path->reada & PATH_READA_BACK) 768 limit = 0; 769 while (reada_count < SEARCH_BY_KEY_READA) { 770 if (pos == limit) 771 break; 772 reada_blocks[reada_count++] = 773 B_N_CHILD_NUM(bh, pos); 774 if (search_path->reada & PATH_READA_BACK) 775 pos--; 776 else 777 pos++; 778 779 /* 780 * check to make sure we're in the same object 781 */ 782 le_key = B_N_PDELIM_KEY(bh, pos); 783 if (le32_to_cpu(le_key->k_objectid) != 784 key->on_disk_key.k_objectid) { 785 break; 786 } 787 } 788 } 789 } 790 } 791 792 /* Form the path to an item and position in this item which contains 793 file byte defined by key. If there is no such item 794 corresponding to the key, we point the path to the item with 795 maximal key less than key, and *pos_in_item is set to one 796 past the last entry/byte in the item. If searching for entry in a 797 directory item, and it is not found, *pos_in_item is set to one 798 entry more than the entry with maximal key which is less than the 799 sought key. 800 801 Note that if there is no entry in this same node which is one more, 802 then we point to an imaginary entry. for direct items, the 803 position is in units of bytes, for indirect items the position is 804 in units of blocknr entries, for directory items the position is in 805 units of directory entries. */ 806 807 /* The function is NOT SCHEDULE-SAFE! */ 808 int search_for_position_by_key(struct super_block *sb, /* Pointer to the super block. */ 809 const struct cpu_key *p_cpu_key, /* Key to search (cpu variable) */ 810 struct treepath *search_path /* Filled up by this function. */ 811 ) 812 { 813 struct item_head *p_le_ih; /* pointer to on-disk structure */ 814 int blk_size; 815 loff_t item_offset, offset; 816 struct reiserfs_dir_entry de; 817 int retval; 818 819 /* If searching for directory entry. */ 820 if (is_direntry_cpu_key(p_cpu_key)) 821 return search_by_entry_key(sb, p_cpu_key, search_path, 822 &de); 823 824 /* If not searching for directory entry. */ 825 826 /* If item is found. */ 827 retval = search_item(sb, p_cpu_key, search_path); 828 if (retval == IO_ERROR) 829 return retval; 830 if (retval == ITEM_FOUND) { 831 832 RFALSE(!ih_item_len 833 (B_N_PITEM_HEAD 834 (PATH_PLAST_BUFFER(search_path), 835 PATH_LAST_POSITION(search_path))), 836 "PAP-5165: item length equals zero"); 837 838 pos_in_item(search_path) = 0; 839 return POSITION_FOUND; 840 } 841 842 RFALSE(!PATH_LAST_POSITION(search_path), 843 "PAP-5170: position equals zero"); 844 845 /* Item is not found. Set path to the previous item. */ 846 p_le_ih = 847 B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path), 848 --PATH_LAST_POSITION(search_path)); 849 blk_size = sb->s_blocksize; 850 851 if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) { 852 return FILE_NOT_FOUND; 853 } 854 // FIXME: quite ugly this far 855 856 item_offset = le_ih_k_offset(p_le_ih); 857 offset = cpu_key_k_offset(p_cpu_key); 858 859 /* Needed byte is contained in the item pointed to by the path. */ 860 if (item_offset <= offset && 861 item_offset + op_bytes_number(p_le_ih, blk_size) > offset) { 862 pos_in_item(search_path) = offset - item_offset; 863 if (is_indirect_le_ih(p_le_ih)) { 864 pos_in_item(search_path) /= blk_size; 865 } 866 return POSITION_FOUND; 867 } 868 869 /* Needed byte is not contained in the item pointed to by the 870 path. Set pos_in_item out of the item. */ 871 if (is_indirect_le_ih(p_le_ih)) 872 pos_in_item(search_path) = 873 ih_item_len(p_le_ih) / UNFM_P_SIZE; 874 else 875 pos_in_item(search_path) = ih_item_len(p_le_ih); 876 877 return POSITION_NOT_FOUND; 878 } 879 880 /* Compare given item and item pointed to by the path. */ 881 int comp_items(const struct item_head *stored_ih, const struct treepath *path) 882 { 883 struct buffer_head *bh = PATH_PLAST_BUFFER(path); 884 struct item_head *ih; 885 886 /* Last buffer at the path is not in the tree. */ 887 if (!B_IS_IN_TREE(bh)) 888 return 1; 889 890 /* Last path position is invalid. */ 891 if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh)) 892 return 1; 893 894 /* we need only to know, whether it is the same item */ 895 ih = get_ih(path); 896 return memcmp(stored_ih, ih, IH_SIZE); 897 } 898 899 /* unformatted nodes are not logged anymore, ever. This is safe 900 ** now 901 */ 902 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1) 903 904 // block can not be forgotten as it is in I/O or held by someone 905 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh))) 906 907 // prepare for delete or cut of direct item 908 static inline int prepare_for_direct_item(struct treepath *path, 909 struct item_head *le_ih, 910 struct inode *inode, 911 loff_t new_file_length, int *cut_size) 912 { 913 loff_t round_len; 914 915 if (new_file_length == max_reiserfs_offset(inode)) { 916 /* item has to be deleted */ 917 *cut_size = -(IH_SIZE + ih_item_len(le_ih)); 918 return M_DELETE; 919 } 920 // new file gets truncated 921 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) { 922 // 923 round_len = ROUND_UP(new_file_length); 924 /* this was new_file_length < le_ih ... */ 925 if (round_len < le_ih_k_offset(le_ih)) { 926 *cut_size = -(IH_SIZE + ih_item_len(le_ih)); 927 return M_DELETE; /* Delete this item. */ 928 } 929 /* Calculate first position and size for cutting from item. */ 930 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1); 931 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path)); 932 933 return M_CUT; /* Cut from this item. */ 934 } 935 936 // old file: items may have any length 937 938 if (new_file_length < le_ih_k_offset(le_ih)) { 939 *cut_size = -(IH_SIZE + ih_item_len(le_ih)); 940 return M_DELETE; /* Delete this item. */ 941 } 942 /* Calculate first position and size for cutting from item. */ 943 *cut_size = -(ih_item_len(le_ih) - 944 (pos_in_item(path) = 945 new_file_length + 1 - le_ih_k_offset(le_ih))); 946 return M_CUT; /* Cut from this item. */ 947 } 948 949 static inline int prepare_for_direntry_item(struct treepath *path, 950 struct item_head *le_ih, 951 struct inode *inode, 952 loff_t new_file_length, 953 int *cut_size) 954 { 955 if (le_ih_k_offset(le_ih) == DOT_OFFSET && 956 new_file_length == max_reiserfs_offset(inode)) { 957 RFALSE(ih_entry_count(le_ih) != 2, 958 "PAP-5220: incorrect empty directory item (%h)", le_ih); 959 *cut_size = -(IH_SIZE + ih_item_len(le_ih)); 960 return M_DELETE; /* Delete the directory item containing "." and ".." entry. */ 961 } 962 963 if (ih_entry_count(le_ih) == 1) { 964 /* Delete the directory item such as there is one record only 965 in this item */ 966 *cut_size = -(IH_SIZE + ih_item_len(le_ih)); 967 return M_DELETE; 968 } 969 970 /* Cut one record from the directory item. */ 971 *cut_size = 972 -(DEH_SIZE + 973 entry_length(get_last_bh(path), le_ih, pos_in_item(path))); 974 return M_CUT; 975 } 976 977 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1) 978 979 /* If the path points to a directory or direct item, calculate mode and the size cut, for balance. 980 If the path points to an indirect item, remove some number of its unformatted nodes. 981 In case of file truncate calculate whether this item must be deleted/truncated or last 982 unformatted node of this item will be converted to a direct item. 983 This function returns a determination of what balance mode the calling function should employ. */ 984 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *path, const struct cpu_key *item_key, int *removed, /* Number of unformatted nodes which were removed 985 from end of the file. */ 986 int *cut_size, unsigned long long new_file_length /* MAX_KEY_OFFSET in case of delete. */ 987 ) 988 { 989 struct super_block *sb = inode->i_sb; 990 struct item_head *p_le_ih = PATH_PITEM_HEAD(path); 991 struct buffer_head *bh = PATH_PLAST_BUFFER(path); 992 993 BUG_ON(!th->t_trans_id); 994 995 /* Stat_data item. */ 996 if (is_statdata_le_ih(p_le_ih)) { 997 998 RFALSE(new_file_length != max_reiserfs_offset(inode), 999 "PAP-5210: mode must be M_DELETE"); 1000 1001 *cut_size = -(IH_SIZE + ih_item_len(p_le_ih)); 1002 return M_DELETE; 1003 } 1004 1005 /* Directory item. */ 1006 if (is_direntry_le_ih(p_le_ih)) 1007 return prepare_for_direntry_item(path, p_le_ih, inode, 1008 new_file_length, 1009 cut_size); 1010 1011 /* Direct item. */ 1012 if (is_direct_le_ih(p_le_ih)) 1013 return prepare_for_direct_item(path, p_le_ih, inode, 1014 new_file_length, cut_size); 1015 1016 /* Case of an indirect item. */ 1017 { 1018 int blk_size = sb->s_blocksize; 1019 struct item_head s_ih; 1020 int need_re_search; 1021 int delete = 0; 1022 int result = M_CUT; 1023 int pos = 0; 1024 1025 if ( new_file_length == max_reiserfs_offset (inode) ) { 1026 /* prepare_for_delete_or_cut() is called by 1027 * reiserfs_delete_item() */ 1028 new_file_length = 0; 1029 delete = 1; 1030 } 1031 1032 do { 1033 need_re_search = 0; 1034 *cut_size = 0; 1035 bh = PATH_PLAST_BUFFER(path); 1036 copy_item_head(&s_ih, PATH_PITEM_HEAD(path)); 1037 pos = I_UNFM_NUM(&s_ih); 1038 1039 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) { 1040 __le32 *unfm; 1041 __u32 block; 1042 1043 /* Each unformatted block deletion may involve one additional 1044 * bitmap block into the transaction, thereby the initial 1045 * journal space reservation might not be enough. */ 1046 if (!delete && (*cut_size) != 0 && 1047 reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) 1048 break; 1049 1050 unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1; 1051 block = get_block_num(unfm, 0); 1052 1053 if (block != 0) { 1054 reiserfs_prepare_for_journal(sb, bh, 1); 1055 put_block_num(unfm, 0, 0); 1056 journal_mark_dirty(th, sb, bh); 1057 reiserfs_free_block(th, inode, block, 1); 1058 } 1059 1060 reiserfs_cond_resched(sb); 1061 1062 if (item_moved (&s_ih, path)) { 1063 need_re_search = 1; 1064 break; 1065 } 1066 1067 pos --; 1068 (*removed)++; 1069 (*cut_size) -= UNFM_P_SIZE; 1070 1071 if (pos == 0) { 1072 (*cut_size) -= IH_SIZE; 1073 result = M_DELETE; 1074 break; 1075 } 1076 } 1077 /* a trick. If the buffer has been logged, this will do nothing. If 1078 ** we've broken the loop without logging it, it will restore the 1079 ** buffer */ 1080 reiserfs_restore_prepared_buffer(sb, bh); 1081 } while (need_re_search && 1082 search_for_position_by_key(sb, item_key, path) == POSITION_FOUND); 1083 pos_in_item(path) = pos * UNFM_P_SIZE; 1084 1085 if (*cut_size == 0) { 1086 /* Nothing were cut. maybe convert last unformatted node to the 1087 * direct item? */ 1088 result = M_CONVERT; 1089 } 1090 return result; 1091 } 1092 } 1093 1094 /* Calculate number of bytes which will be deleted or cut during balance */ 1095 static int calc_deleted_bytes_number(struct tree_balance *tb, char mode) 1096 { 1097 int del_size; 1098 struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path); 1099 1100 if (is_statdata_le_ih(p_le_ih)) 1101 return 0; 1102 1103 del_size = 1104 (mode == 1105 M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0]; 1106 if (is_direntry_le_ih(p_le_ih)) { 1107 /* return EMPTY_DIR_SIZE; We delete emty directoris only. 1108 * we can't use EMPTY_DIR_SIZE, as old format dirs have a different 1109 * empty size. ick. FIXME, is this right? */ 1110 return del_size; 1111 } 1112 1113 if (is_indirect_le_ih(p_le_ih)) 1114 del_size = (del_size / UNFM_P_SIZE) * 1115 (PATH_PLAST_BUFFER(tb->tb_path)->b_size); 1116 return del_size; 1117 } 1118 1119 static void init_tb_struct(struct reiserfs_transaction_handle *th, 1120 struct tree_balance *tb, 1121 struct super_block *sb, 1122 struct treepath *path, int size) 1123 { 1124 1125 BUG_ON(!th->t_trans_id); 1126 1127 memset(tb, '\0', sizeof(struct tree_balance)); 1128 tb->transaction_handle = th; 1129 tb->tb_sb = sb; 1130 tb->tb_path = path; 1131 PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL; 1132 PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0; 1133 tb->insert_size[0] = size; 1134 } 1135 1136 void padd_item(char *item, int total_length, int length) 1137 { 1138 int i; 1139 1140 for (i = total_length; i > length;) 1141 item[--i] = 0; 1142 } 1143 1144 #ifdef REISERQUOTA_DEBUG 1145 char key2type(struct reiserfs_key *ih) 1146 { 1147 if (is_direntry_le_key(2, ih)) 1148 return 'd'; 1149 if (is_direct_le_key(2, ih)) 1150 return 'D'; 1151 if (is_indirect_le_key(2, ih)) 1152 return 'i'; 1153 if (is_statdata_le_key(2, ih)) 1154 return 's'; 1155 return 'u'; 1156 } 1157 1158 char head2type(struct item_head *ih) 1159 { 1160 if (is_direntry_le_ih(ih)) 1161 return 'd'; 1162 if (is_direct_le_ih(ih)) 1163 return 'D'; 1164 if (is_indirect_le_ih(ih)) 1165 return 'i'; 1166 if (is_statdata_le_ih(ih)) 1167 return 's'; 1168 return 'u'; 1169 } 1170 #endif 1171 1172 /* Delete object item. 1173 * th - active transaction handle 1174 * path - path to the deleted item 1175 * item_key - key to search for the deleted item 1176 * indode - used for updating i_blocks and quotas 1177 * un_bh - NULL or unformatted node pointer 1178 */ 1179 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, 1180 struct treepath *path, const struct cpu_key *item_key, 1181 struct inode *inode, struct buffer_head *un_bh) 1182 { 1183 struct super_block *sb = inode->i_sb; 1184 struct tree_balance s_del_balance; 1185 struct item_head s_ih; 1186 struct item_head *q_ih; 1187 int quota_cut_bytes; 1188 int ret_value, del_size, removed; 1189 1190 #ifdef CONFIG_REISERFS_CHECK 1191 char mode; 1192 int iter = 0; 1193 #endif 1194 1195 BUG_ON(!th->t_trans_id); 1196 1197 init_tb_struct(th, &s_del_balance, sb, path, 1198 0 /*size is unknown */ ); 1199 1200 while (1) { 1201 removed = 0; 1202 1203 #ifdef CONFIG_REISERFS_CHECK 1204 iter++; 1205 mode = 1206 #endif 1207 prepare_for_delete_or_cut(th, inode, path, 1208 item_key, &removed, 1209 &del_size, 1210 max_reiserfs_offset(inode)); 1211 1212 RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE"); 1213 1214 copy_item_head(&s_ih, PATH_PITEM_HEAD(path)); 1215 s_del_balance.insert_size[0] = del_size; 1216 1217 ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL); 1218 if (ret_value != REPEAT_SEARCH) 1219 break; 1220 1221 PROC_INFO_INC(sb, delete_item_restarted); 1222 1223 // file system changed, repeat search 1224 ret_value = 1225 search_for_position_by_key(sb, item_key, path); 1226 if (ret_value == IO_ERROR) 1227 break; 1228 if (ret_value == FILE_NOT_FOUND) { 1229 reiserfs_warning(sb, "vs-5340", 1230 "no items of the file %K found", 1231 item_key); 1232 break; 1233 } 1234 } /* while (1) */ 1235 1236 if (ret_value != CARRY_ON) { 1237 unfix_nodes(&s_del_balance); 1238 return 0; 1239 } 1240 // reiserfs_delete_item returns item length when success 1241 ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE); 1242 q_ih = get_ih(path); 1243 quota_cut_bytes = ih_item_len(q_ih); 1244 1245 /* hack so the quota code doesn't have to guess if the file 1246 ** has a tail. On tail insert, we allocate quota for 1 unformatted node. 1247 ** We test the offset because the tail might have been 1248 ** split into multiple items, and we only want to decrement for 1249 ** the unfm node once 1250 */ 1251 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) { 1252 if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) { 1253 quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE; 1254 } else { 1255 quota_cut_bytes = 0; 1256 } 1257 } 1258 1259 if (un_bh) { 1260 int off; 1261 char *data; 1262 1263 /* We are in direct2indirect conversion, so move tail contents 1264 to the unformatted node */ 1265 /* note, we do the copy before preparing the buffer because we 1266 ** don't care about the contents of the unformatted node yet. 1267 ** the only thing we really care about is the direct item's data 1268 ** is in the unformatted node. 1269 ** 1270 ** Otherwise, we would have to call reiserfs_prepare_for_journal on 1271 ** the unformatted node, which might schedule, meaning we'd have to 1272 ** loop all the way back up to the start of the while loop. 1273 ** 1274 ** The unformatted node must be dirtied later on. We can't be 1275 ** sure here if the entire tail has been deleted yet. 1276 ** 1277 ** un_bh is from the page cache (all unformatted nodes are 1278 ** from the page cache) and might be a highmem page. So, we 1279 ** can't use un_bh->b_data. 1280 ** -clm 1281 */ 1282 1283 data = kmap_atomic(un_bh->b_page); 1284 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1)); 1285 memcpy(data + off, 1286 B_I_PITEM(PATH_PLAST_BUFFER(path), &s_ih), 1287 ret_value); 1288 kunmap_atomic(data); 1289 } 1290 /* Perform balancing after all resources have been collected at once. */ 1291 do_balance(&s_del_balance, NULL, NULL, M_DELETE); 1292 1293 #ifdef REISERQUOTA_DEBUG 1294 reiserfs_debug(sb, REISERFS_DEBUG_CODE, 1295 "reiserquota delete_item(): freeing %u, id=%u type=%c", 1296 quota_cut_bytes, inode->i_uid, head2type(&s_ih)); 1297 #endif 1298 dquot_free_space_nodirty(inode, quota_cut_bytes); 1299 1300 /* Return deleted body length */ 1301 return ret_value; 1302 } 1303 1304 /* Summary Of Mechanisms For Handling Collisions Between Processes: 1305 1306 deletion of the body of the object is performed by iput(), with the 1307 result that if multiple processes are operating on a file, the 1308 deletion of the body of the file is deferred until the last process 1309 that has an open inode performs its iput(). 1310 1311 writes and truncates are protected from collisions by use of 1312 semaphores. 1313 1314 creates, linking, and mknod are protected from collisions with other 1315 processes by making the reiserfs_add_entry() the last step in the 1316 creation, and then rolling back all changes if there was a collision. 1317 - Hans 1318 */ 1319 1320 /* this deletes item which never gets split */ 1321 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th, 1322 struct inode *inode, struct reiserfs_key *key) 1323 { 1324 struct tree_balance tb; 1325 INITIALIZE_PATH(path); 1326 int item_len = 0; 1327 int tb_init = 0; 1328 struct cpu_key cpu_key; 1329 int retval; 1330 int quota_cut_bytes = 0; 1331 1332 BUG_ON(!th->t_trans_id); 1333 1334 le_key2cpu_key(&cpu_key, key); 1335 1336 while (1) { 1337 retval = search_item(th->t_super, &cpu_key, &path); 1338 if (retval == IO_ERROR) { 1339 reiserfs_error(th->t_super, "vs-5350", 1340 "i/o failure occurred trying " 1341 "to delete %K", &cpu_key); 1342 break; 1343 } 1344 if (retval != ITEM_FOUND) { 1345 pathrelse(&path); 1346 // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir 1347 if (! 1348 ((unsigned long long) 1349 GET_HASH_VALUE(le_key_k_offset 1350 (le_key_version(key), key)) == 0 1351 && (unsigned long long) 1352 GET_GENERATION_NUMBER(le_key_k_offset 1353 (le_key_version(key), 1354 key)) == 1)) 1355 reiserfs_warning(th->t_super, "vs-5355", 1356 "%k not found", key); 1357 break; 1358 } 1359 if (!tb_init) { 1360 tb_init = 1; 1361 item_len = ih_item_len(PATH_PITEM_HEAD(&path)); 1362 init_tb_struct(th, &tb, th->t_super, &path, 1363 -(IH_SIZE + item_len)); 1364 } 1365 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path)); 1366 1367 retval = fix_nodes(M_DELETE, &tb, NULL, NULL); 1368 if (retval == REPEAT_SEARCH) { 1369 PROC_INFO_INC(th->t_super, delete_solid_item_restarted); 1370 continue; 1371 } 1372 1373 if (retval == CARRY_ON) { 1374 do_balance(&tb, NULL, NULL, M_DELETE); 1375 if (inode) { /* Should we count quota for item? (we don't count quotas for save-links) */ 1376 #ifdef REISERQUOTA_DEBUG 1377 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE, 1378 "reiserquota delete_solid_item(): freeing %u id=%u type=%c", 1379 quota_cut_bytes, inode->i_uid, 1380 key2type(key)); 1381 #endif 1382 dquot_free_space_nodirty(inode, 1383 quota_cut_bytes); 1384 } 1385 break; 1386 } 1387 // IO_ERROR, NO_DISK_SPACE, etc 1388 reiserfs_warning(th->t_super, "vs-5360", 1389 "could not delete %K due to fix_nodes failure", 1390 &cpu_key); 1391 unfix_nodes(&tb); 1392 break; 1393 } 1394 1395 reiserfs_check_path(&path); 1396 } 1397 1398 int reiserfs_delete_object(struct reiserfs_transaction_handle *th, 1399 struct inode *inode) 1400 { 1401 int err; 1402 inode->i_size = 0; 1403 BUG_ON(!th->t_trans_id); 1404 1405 /* for directory this deletes item containing "." and ".." */ 1406 err = 1407 reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ ); 1408 if (err) 1409 return err; 1410 1411 #if defined( USE_INODE_GENERATION_COUNTER ) 1412 if (!old_format_only(th->t_super)) { 1413 __le32 *inode_generation; 1414 1415 inode_generation = 1416 &REISERFS_SB(th->t_super)->s_rs->s_inode_generation; 1417 le32_add_cpu(inode_generation, 1); 1418 } 1419 /* USE_INODE_GENERATION_COUNTER */ 1420 #endif 1421 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode)); 1422 1423 return err; 1424 } 1425 1426 static void unmap_buffers(struct page *page, loff_t pos) 1427 { 1428 struct buffer_head *bh; 1429 struct buffer_head *head; 1430 struct buffer_head *next; 1431 unsigned long tail_index; 1432 unsigned long cur_index; 1433 1434 if (page) { 1435 if (page_has_buffers(page)) { 1436 tail_index = pos & (PAGE_CACHE_SIZE - 1); 1437 cur_index = 0; 1438 head = page_buffers(page); 1439 bh = head; 1440 do { 1441 next = bh->b_this_page; 1442 1443 /* we want to unmap the buffers that contain the tail, and 1444 ** all the buffers after it (since the tail must be at the 1445 ** end of the file). We don't want to unmap file data 1446 ** before the tail, since it might be dirty and waiting to 1447 ** reach disk 1448 */ 1449 cur_index += bh->b_size; 1450 if (cur_index > tail_index) { 1451 reiserfs_unmap_buffer(bh); 1452 } 1453 bh = next; 1454 } while (bh != head); 1455 } 1456 } 1457 } 1458 1459 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th, 1460 struct inode *inode, 1461 struct page *page, 1462 struct treepath *path, 1463 const struct cpu_key *item_key, 1464 loff_t new_file_size, char *mode) 1465 { 1466 struct super_block *sb = inode->i_sb; 1467 int block_size = sb->s_blocksize; 1468 int cut_bytes; 1469 BUG_ON(!th->t_trans_id); 1470 BUG_ON(new_file_size != inode->i_size); 1471 1472 /* the page being sent in could be NULL if there was an i/o error 1473 ** reading in the last block. The user will hit problems trying to 1474 ** read the file, but for now we just skip the indirect2direct 1475 */ 1476 if (atomic_read(&inode->i_count) > 1 || 1477 !tail_has_to_be_packed(inode) || 1478 !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) { 1479 /* leave tail in an unformatted node */ 1480 *mode = M_SKIP_BALANCING; 1481 cut_bytes = 1482 block_size - (new_file_size & (block_size - 1)); 1483 pathrelse(path); 1484 return cut_bytes; 1485 } 1486 /* Perform the conversion to a direct_item. */ 1487 /* return indirect_to_direct(inode, path, item_key, 1488 new_file_size, mode); */ 1489 return indirect2direct(th, inode, page, path, item_key, 1490 new_file_size, mode); 1491 } 1492 1493 /* we did indirect_to_direct conversion. And we have inserted direct 1494 item successesfully, but there were no disk space to cut unfm 1495 pointer being converted. Therefore we have to delete inserted 1496 direct item(s) */ 1497 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th, 1498 struct inode *inode, struct treepath *path) 1499 { 1500 struct cpu_key tail_key; 1501 int tail_len; 1502 int removed; 1503 BUG_ON(!th->t_trans_id); 1504 1505 make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4); // !!!! 1506 tail_key.key_length = 4; 1507 1508 tail_len = 1509 (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1; 1510 while (tail_len) { 1511 /* look for the last byte of the tail */ 1512 if (search_for_position_by_key(inode->i_sb, &tail_key, path) == 1513 POSITION_NOT_FOUND) 1514 reiserfs_panic(inode->i_sb, "vs-5615", 1515 "found invalid item"); 1516 RFALSE(path->pos_in_item != 1517 ih_item_len(PATH_PITEM_HEAD(path)) - 1, 1518 "vs-5616: appended bytes found"); 1519 PATH_LAST_POSITION(path)--; 1520 1521 removed = 1522 reiserfs_delete_item(th, path, &tail_key, inode, 1523 NULL /*unbh not needed */ ); 1524 RFALSE(removed <= 0 1525 || removed > tail_len, 1526 "vs-5617: there was tail %d bytes, removed item length %d bytes", 1527 tail_len, removed); 1528 tail_len -= removed; 1529 set_cpu_key_k_offset(&tail_key, 1530 cpu_key_k_offset(&tail_key) - removed); 1531 } 1532 reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct " 1533 "conversion has been rolled back due to " 1534 "lack of disk space"); 1535 //mark_file_without_tail (inode); 1536 mark_inode_dirty(inode); 1537 } 1538 1539 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */ 1540 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th, 1541 struct treepath *path, 1542 struct cpu_key *item_key, 1543 struct inode *inode, 1544 struct page *page, loff_t new_file_size) 1545 { 1546 struct super_block *sb = inode->i_sb; 1547 /* Every function which is going to call do_balance must first 1548 create a tree_balance structure. Then it must fill up this 1549 structure by using the init_tb_struct and fix_nodes functions. 1550 After that we can make tree balancing. */ 1551 struct tree_balance s_cut_balance; 1552 struct item_head *p_le_ih; 1553 int cut_size = 0, /* Amount to be cut. */ 1554 ret_value = CARRY_ON, removed = 0, /* Number of the removed unformatted nodes. */ 1555 is_inode_locked = 0; 1556 char mode; /* Mode of the balance. */ 1557 int retval2 = -1; 1558 int quota_cut_bytes; 1559 loff_t tail_pos = 0; 1560 1561 BUG_ON(!th->t_trans_id); 1562 1563 init_tb_struct(th, &s_cut_balance, inode->i_sb, path, 1564 cut_size); 1565 1566 /* Repeat this loop until we either cut the item without needing 1567 to balance, or we fix_nodes without schedule occurring */ 1568 while (1) { 1569 /* Determine the balance mode, position of the first byte to 1570 be cut, and size to be cut. In case of the indirect item 1571 free unformatted nodes which are pointed to by the cut 1572 pointers. */ 1573 1574 mode = 1575 prepare_for_delete_or_cut(th, inode, path, 1576 item_key, &removed, 1577 &cut_size, new_file_size); 1578 if (mode == M_CONVERT) { 1579 /* convert last unformatted node to direct item or leave 1580 tail in the unformatted node */ 1581 RFALSE(ret_value != CARRY_ON, 1582 "PAP-5570: can not convert twice"); 1583 1584 ret_value = 1585 maybe_indirect_to_direct(th, inode, page, 1586 path, item_key, 1587 new_file_size, &mode); 1588 if (mode == M_SKIP_BALANCING) 1589 /* tail has been left in the unformatted node */ 1590 return ret_value; 1591 1592 is_inode_locked = 1; 1593 1594 /* removing of last unformatted node will change value we 1595 have to return to truncate. Save it */ 1596 retval2 = ret_value; 1597 /*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */ 1598 1599 /* So, we have performed the first part of the conversion: 1600 inserting the new direct item. Now we are removing the 1601 last unformatted node pointer. Set key to search for 1602 it. */ 1603 set_cpu_key_k_type(item_key, TYPE_INDIRECT); 1604 item_key->key_length = 4; 1605 new_file_size -= 1606 (new_file_size & (sb->s_blocksize - 1)); 1607 tail_pos = new_file_size; 1608 set_cpu_key_k_offset(item_key, new_file_size + 1); 1609 if (search_for_position_by_key 1610 (sb, item_key, 1611 path) == POSITION_NOT_FOUND) { 1612 print_block(PATH_PLAST_BUFFER(path), 3, 1613 PATH_LAST_POSITION(path) - 1, 1614 PATH_LAST_POSITION(path) + 1); 1615 reiserfs_panic(sb, "PAP-5580", "item to " 1616 "convert does not exist (%K)", 1617 item_key); 1618 } 1619 continue; 1620 } 1621 if (cut_size == 0) { 1622 pathrelse(path); 1623 return 0; 1624 } 1625 1626 s_cut_balance.insert_size[0] = cut_size; 1627 1628 ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL); 1629 if (ret_value != REPEAT_SEARCH) 1630 break; 1631 1632 PROC_INFO_INC(sb, cut_from_item_restarted); 1633 1634 ret_value = 1635 search_for_position_by_key(sb, item_key, path); 1636 if (ret_value == POSITION_FOUND) 1637 continue; 1638 1639 reiserfs_warning(sb, "PAP-5610", "item %K not found", 1640 item_key); 1641 unfix_nodes(&s_cut_balance); 1642 return (ret_value == IO_ERROR) ? -EIO : -ENOENT; 1643 } /* while */ 1644 1645 // check fix_nodes results (IO_ERROR or NO_DISK_SPACE) 1646 if (ret_value != CARRY_ON) { 1647 if (is_inode_locked) { 1648 // FIXME: this seems to be not needed: we are always able 1649 // to cut item 1650 indirect_to_direct_roll_back(th, inode, path); 1651 } 1652 if (ret_value == NO_DISK_SPACE) 1653 reiserfs_warning(sb, "reiserfs-5092", 1654 "NO_DISK_SPACE"); 1655 unfix_nodes(&s_cut_balance); 1656 return -EIO; 1657 } 1658 1659 /* go ahead and perform balancing */ 1660 1661 RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode"); 1662 1663 /* Calculate number of bytes that need to be cut from the item. */ 1664 quota_cut_bytes = 1665 (mode == 1666 M_DELETE) ? ih_item_len(get_ih(path)) : -s_cut_balance. 1667 insert_size[0]; 1668 if (retval2 == -1) 1669 ret_value = calc_deleted_bytes_number(&s_cut_balance, mode); 1670 else 1671 ret_value = retval2; 1672 1673 /* For direct items, we only change the quota when deleting the last 1674 ** item. 1675 */ 1676 p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path); 1677 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) { 1678 if (mode == M_DELETE && 1679 (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) == 1680 1) { 1681 // FIXME: this is to keep 3.5 happy 1682 REISERFS_I(inode)->i_first_direct_byte = U32_MAX; 1683 quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE; 1684 } else { 1685 quota_cut_bytes = 0; 1686 } 1687 } 1688 #ifdef CONFIG_REISERFS_CHECK 1689 if (is_inode_locked) { 1690 struct item_head *le_ih = 1691 PATH_PITEM_HEAD(s_cut_balance.tb_path); 1692 /* we are going to complete indirect2direct conversion. Make 1693 sure, that we exactly remove last unformatted node pointer 1694 of the item */ 1695 if (!is_indirect_le_ih(le_ih)) 1696 reiserfs_panic(sb, "vs-5652", 1697 "item must be indirect %h", le_ih); 1698 1699 if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE) 1700 reiserfs_panic(sb, "vs-5653", "completing " 1701 "indirect2direct conversion indirect " 1702 "item %h being deleted must be of " 1703 "4 byte long", le_ih); 1704 1705 if (mode == M_CUT 1706 && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) { 1707 reiserfs_panic(sb, "vs-5654", "can not complete " 1708 "indirect2direct conversion of %h " 1709 "(CUT, insert_size==%d)", 1710 le_ih, s_cut_balance.insert_size[0]); 1711 } 1712 /* it would be useful to make sure, that right neighboring 1713 item is direct item of this file */ 1714 } 1715 #endif 1716 1717 do_balance(&s_cut_balance, NULL, NULL, mode); 1718 if (is_inode_locked) { 1719 /* we've done an indirect->direct conversion. when the data block 1720 ** was freed, it was removed from the list of blocks that must 1721 ** be flushed before the transaction commits, make sure to 1722 ** unmap and invalidate it 1723 */ 1724 unmap_buffers(page, tail_pos); 1725 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask; 1726 } 1727 #ifdef REISERQUOTA_DEBUG 1728 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE, 1729 "reiserquota cut_from_item(): freeing %u id=%u type=%c", 1730 quota_cut_bytes, inode->i_uid, '?'); 1731 #endif 1732 dquot_free_space_nodirty(inode, quota_cut_bytes); 1733 return ret_value; 1734 } 1735 1736 static void truncate_directory(struct reiserfs_transaction_handle *th, 1737 struct inode *inode) 1738 { 1739 BUG_ON(!th->t_trans_id); 1740 if (inode->i_nlink) 1741 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0"); 1742 1743 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET); 1744 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY); 1745 reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode)); 1746 reiserfs_update_sd(th, inode); 1747 set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET); 1748 set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA); 1749 } 1750 1751 /* Truncate file to the new size. Note, this must be called with a transaction 1752 already started */ 1753 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, 1754 struct inode *inode, /* ->i_size contains new size */ 1755 struct page *page, /* up to date for last block */ 1756 int update_timestamps /* when it is called by 1757 file_release to convert 1758 the tail - no timestamps 1759 should be updated */ 1760 ) 1761 { 1762 INITIALIZE_PATH(s_search_path); /* Path to the current object item. */ 1763 struct item_head *p_le_ih; /* Pointer to an item header. */ 1764 struct cpu_key s_item_key; /* Key to search for a previous file item. */ 1765 loff_t file_size, /* Old file size. */ 1766 new_file_size; /* New file size. */ 1767 int deleted; /* Number of deleted or truncated bytes. */ 1768 int retval; 1769 int err = 0; 1770 1771 BUG_ON(!th->t_trans_id); 1772 if (! 1773 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 1774 || S_ISLNK(inode->i_mode))) 1775 return 0; 1776 1777 if (S_ISDIR(inode->i_mode)) { 1778 // deletion of directory - no need to update timestamps 1779 truncate_directory(th, inode); 1780 return 0; 1781 } 1782 1783 /* Get new file size. */ 1784 new_file_size = inode->i_size; 1785 1786 // FIXME: note, that key type is unimportant here 1787 make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode), 1788 TYPE_DIRECT, 3); 1789 1790 retval = 1791 search_for_position_by_key(inode->i_sb, &s_item_key, 1792 &s_search_path); 1793 if (retval == IO_ERROR) { 1794 reiserfs_error(inode->i_sb, "vs-5657", 1795 "i/o failure occurred trying to truncate %K", 1796 &s_item_key); 1797 err = -EIO; 1798 goto out; 1799 } 1800 if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) { 1801 reiserfs_error(inode->i_sb, "PAP-5660", 1802 "wrong result %d of search for %K", retval, 1803 &s_item_key); 1804 1805 err = -EIO; 1806 goto out; 1807 } 1808 1809 s_search_path.pos_in_item--; 1810 1811 /* Get real file size (total length of all file items) */ 1812 p_le_ih = PATH_PITEM_HEAD(&s_search_path); 1813 if (is_statdata_le_ih(p_le_ih)) 1814 file_size = 0; 1815 else { 1816 loff_t offset = le_ih_k_offset(p_le_ih); 1817 int bytes = 1818 op_bytes_number(p_le_ih, inode->i_sb->s_blocksize); 1819 1820 /* this may mismatch with real file size: if last direct item 1821 had no padding zeros and last unformatted node had no free 1822 space, this file would have this file size */ 1823 file_size = offset + bytes - 1; 1824 } 1825 /* 1826 * are we doing a full truncate or delete, if so 1827 * kick in the reada code 1828 */ 1829 if (new_file_size == 0) 1830 s_search_path.reada = PATH_READA | PATH_READA_BACK; 1831 1832 if (file_size == 0 || file_size < new_file_size) { 1833 goto update_and_out; 1834 } 1835 1836 /* Update key to search for the last file item. */ 1837 set_cpu_key_k_offset(&s_item_key, file_size); 1838 1839 do { 1840 /* Cut or delete file item. */ 1841 deleted = 1842 reiserfs_cut_from_item(th, &s_search_path, &s_item_key, 1843 inode, page, new_file_size); 1844 if (deleted < 0) { 1845 reiserfs_warning(inode->i_sb, "vs-5665", 1846 "reiserfs_cut_from_item failed"); 1847 reiserfs_check_path(&s_search_path); 1848 return 0; 1849 } 1850 1851 RFALSE(deleted > file_size, 1852 "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K", 1853 deleted, file_size, &s_item_key); 1854 1855 /* Change key to search the last file item. */ 1856 file_size -= deleted; 1857 1858 set_cpu_key_k_offset(&s_item_key, file_size); 1859 1860 /* While there are bytes to truncate and previous file item is presented in the tree. */ 1861 1862 /* 1863 ** This loop could take a really long time, and could log 1864 ** many more blocks than a transaction can hold. So, we do a polite 1865 ** journal end here, and if the transaction needs ending, we make 1866 ** sure the file is consistent before ending the current trans 1867 ** and starting a new one 1868 */ 1869 if (journal_transaction_should_end(th, 0) || 1870 reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) { 1871 int orig_len_alloc = th->t_blocks_allocated; 1872 pathrelse(&s_search_path); 1873 1874 if (update_timestamps) { 1875 inode->i_mtime = CURRENT_TIME_SEC; 1876 inode->i_ctime = CURRENT_TIME_SEC; 1877 } 1878 reiserfs_update_sd(th, inode); 1879 1880 err = journal_end(th, inode->i_sb, orig_len_alloc); 1881 if (err) 1882 goto out; 1883 err = journal_begin(th, inode->i_sb, 1884 JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ; 1885 if (err) 1886 goto out; 1887 reiserfs_update_inode_transaction(inode); 1888 } 1889 } while (file_size > ROUND_UP(new_file_size) && 1890 search_for_position_by_key(inode->i_sb, &s_item_key, 1891 &s_search_path) == POSITION_FOUND); 1892 1893 RFALSE(file_size > ROUND_UP(new_file_size), 1894 "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d", 1895 new_file_size, file_size, s_item_key.on_disk_key.k_objectid); 1896 1897 update_and_out: 1898 if (update_timestamps) { 1899 // this is truncate, not file closing 1900 inode->i_mtime = CURRENT_TIME_SEC; 1901 inode->i_ctime = CURRENT_TIME_SEC; 1902 } 1903 reiserfs_update_sd(th, inode); 1904 1905 out: 1906 pathrelse(&s_search_path); 1907 return err; 1908 } 1909 1910 #ifdef CONFIG_REISERFS_CHECK 1911 // this makes sure, that we __append__, not overwrite or add holes 1912 static void check_research_for_paste(struct treepath *path, 1913 const struct cpu_key *key) 1914 { 1915 struct item_head *found_ih = get_ih(path); 1916 1917 if (is_direct_le_ih(found_ih)) { 1918 if (le_ih_k_offset(found_ih) + 1919 op_bytes_number(found_ih, 1920 get_last_bh(path)->b_size) != 1921 cpu_key_k_offset(key) 1922 || op_bytes_number(found_ih, 1923 get_last_bh(path)->b_size) != 1924 pos_in_item(path)) 1925 reiserfs_panic(NULL, "PAP-5720", "found direct item " 1926 "%h or position (%d) does not match " 1927 "to key %K", found_ih, 1928 pos_in_item(path), key); 1929 } 1930 if (is_indirect_le_ih(found_ih)) { 1931 if (le_ih_k_offset(found_ih) + 1932 op_bytes_number(found_ih, 1933 get_last_bh(path)->b_size) != 1934 cpu_key_k_offset(key) 1935 || I_UNFM_NUM(found_ih) != pos_in_item(path) 1936 || get_ih_free_space(found_ih) != 0) 1937 reiserfs_panic(NULL, "PAP-5730", "found indirect " 1938 "item (%h) or position (%d) does not " 1939 "match to key (%K)", 1940 found_ih, pos_in_item(path), key); 1941 } 1942 } 1943 #endif /* config reiserfs check */ 1944 1945 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */ 1946 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *search_path, /* Path to the pasted item. */ 1947 const struct cpu_key *key, /* Key to search for the needed item. */ 1948 struct inode *inode, /* Inode item belongs to */ 1949 const char *body, /* Pointer to the bytes to paste. */ 1950 int pasted_size) 1951 { /* Size of pasted bytes. */ 1952 struct tree_balance s_paste_balance; 1953 int retval; 1954 int fs_gen; 1955 1956 BUG_ON(!th->t_trans_id); 1957 1958 fs_gen = get_generation(inode->i_sb); 1959 1960 #ifdef REISERQUOTA_DEBUG 1961 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE, 1962 "reiserquota paste_into_item(): allocating %u id=%u type=%c", 1963 pasted_size, inode->i_uid, 1964 key2type(&(key->on_disk_key))); 1965 #endif 1966 1967 reiserfs_write_unlock(inode->i_sb); 1968 retval = dquot_alloc_space_nodirty(inode, pasted_size); 1969 reiserfs_write_lock(inode->i_sb); 1970 if (retval) { 1971 pathrelse(search_path); 1972 return retval; 1973 } 1974 init_tb_struct(th, &s_paste_balance, th->t_super, search_path, 1975 pasted_size); 1976 #ifdef DISPLACE_NEW_PACKING_LOCALITIES 1977 s_paste_balance.key = key->on_disk_key; 1978 #endif 1979 1980 /* DQUOT_* can schedule, must check before the fix_nodes */ 1981 if (fs_changed(fs_gen, inode->i_sb)) { 1982 goto search_again; 1983 } 1984 1985 while ((retval = 1986 fix_nodes(M_PASTE, &s_paste_balance, NULL, 1987 body)) == REPEAT_SEARCH) { 1988 search_again: 1989 /* file system changed while we were in the fix_nodes */ 1990 PROC_INFO_INC(th->t_super, paste_into_item_restarted); 1991 retval = 1992 search_for_position_by_key(th->t_super, key, 1993 search_path); 1994 if (retval == IO_ERROR) { 1995 retval = -EIO; 1996 goto error_out; 1997 } 1998 if (retval == POSITION_FOUND) { 1999 reiserfs_warning(inode->i_sb, "PAP-5710", 2000 "entry or pasted byte (%K) exists", 2001 key); 2002 retval = -EEXIST; 2003 goto error_out; 2004 } 2005 #ifdef CONFIG_REISERFS_CHECK 2006 check_research_for_paste(search_path, key); 2007 #endif 2008 } 2009 2010 /* Perform balancing after all resources are collected by fix_nodes, and 2011 accessing them will not risk triggering schedule. */ 2012 if (retval == CARRY_ON) { 2013 do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE); 2014 return 0; 2015 } 2016 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO; 2017 error_out: 2018 /* this also releases the path */ 2019 unfix_nodes(&s_paste_balance); 2020 #ifdef REISERQUOTA_DEBUG 2021 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE, 2022 "reiserquota paste_into_item(): freeing %u id=%u type=%c", 2023 pasted_size, inode->i_uid, 2024 key2type(&(key->on_disk_key))); 2025 #endif 2026 dquot_free_space_nodirty(inode, pasted_size); 2027 return retval; 2028 } 2029 2030 /* Insert new item into the buffer at the path. 2031 * th - active transaction handle 2032 * path - path to the inserted item 2033 * ih - pointer to the item header to insert 2034 * body - pointer to the bytes to insert 2035 */ 2036 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, 2037 struct treepath *path, const struct cpu_key *key, 2038 struct item_head *ih, struct inode *inode, 2039 const char *body) 2040 { 2041 struct tree_balance s_ins_balance; 2042 int retval; 2043 int fs_gen = 0; 2044 int quota_bytes = 0; 2045 2046 BUG_ON(!th->t_trans_id); 2047 2048 if (inode) { /* Do we count quotas for item? */ 2049 fs_gen = get_generation(inode->i_sb); 2050 quota_bytes = ih_item_len(ih); 2051 2052 /* hack so the quota code doesn't have to guess if the file has 2053 ** a tail, links are always tails, so there's no guessing needed 2054 */ 2055 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih)) 2056 quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE; 2057 #ifdef REISERQUOTA_DEBUG 2058 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE, 2059 "reiserquota insert_item(): allocating %u id=%u type=%c", 2060 quota_bytes, inode->i_uid, head2type(ih)); 2061 #endif 2062 reiserfs_write_unlock(inode->i_sb); 2063 /* We can't dirty inode here. It would be immediately written but 2064 * appropriate stat item isn't inserted yet... */ 2065 retval = dquot_alloc_space_nodirty(inode, quota_bytes); 2066 reiserfs_write_lock(inode->i_sb); 2067 if (retval) { 2068 pathrelse(path); 2069 return retval; 2070 } 2071 } 2072 init_tb_struct(th, &s_ins_balance, th->t_super, path, 2073 IH_SIZE + ih_item_len(ih)); 2074 #ifdef DISPLACE_NEW_PACKING_LOCALITIES 2075 s_ins_balance.key = key->on_disk_key; 2076 #endif 2077 /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */ 2078 if (inode && fs_changed(fs_gen, inode->i_sb)) { 2079 goto search_again; 2080 } 2081 2082 while ((retval = 2083 fix_nodes(M_INSERT, &s_ins_balance, ih, 2084 body)) == REPEAT_SEARCH) { 2085 search_again: 2086 /* file system changed while we were in the fix_nodes */ 2087 PROC_INFO_INC(th->t_super, insert_item_restarted); 2088 retval = search_item(th->t_super, key, path); 2089 if (retval == IO_ERROR) { 2090 retval = -EIO; 2091 goto error_out; 2092 } 2093 if (retval == ITEM_FOUND) { 2094 reiserfs_warning(th->t_super, "PAP-5760", 2095 "key %K already exists in the tree", 2096 key); 2097 retval = -EEXIST; 2098 goto error_out; 2099 } 2100 } 2101 2102 /* make balancing after all resources will be collected at a time */ 2103 if (retval == CARRY_ON) { 2104 do_balance(&s_ins_balance, ih, body, M_INSERT); 2105 return 0; 2106 } 2107 2108 retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO; 2109 error_out: 2110 /* also releases the path */ 2111 unfix_nodes(&s_ins_balance); 2112 #ifdef REISERQUOTA_DEBUG 2113 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE, 2114 "reiserquota insert_item(): freeing %u id=%u type=%c", 2115 quota_bytes, inode->i_uid, head2type(ih)); 2116 #endif 2117 if (inode) 2118 dquot_free_space_nodirty(inode, quota_bytes); 2119 return retval; 2120 } 2121