fs/reiserfs/stree.c

   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 <linux/reiserfs_fs.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 #ifdef CONFIG_REISERFS_CHECK
 226 extern struct tree_balance *cur_tb;
 227 #endif
 228
 229 /* Minimal possible key. It is never in the tree. */
 230 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
 231
 232 /* Maximal possible key. It is never in the tree. */
 233 static const struct reiserfs_key MAX_KEY = {
 234         __constant_cpu_to_le32(0xffffffff),
 235         __constant_cpu_to_le32(0xffffffff),
 236         {{__constant_cpu_to_le32(0xffffffff),
 237           __constant_cpu_to_le32(0xffffffff)},}
 238 };
 239
 240 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
 241    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
 242    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
 243    case we return a special key, either MIN_KEY or MAX_KEY. */
 244 static inline const struct reiserfs_key *get_lkey(const struct treepath *chk_path,
 245                                                   const struct super_block *sb)
 246 {
 247         int position, path_offset = chk_path->path_length;
 248         struct buffer_head *parent;
 249
 250         RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
 251                "PAP-5010: invalid offset in the path");
 252
 253         /* While not higher in path than first element. */
 254         while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 255
 256                 RFALSE(!buffer_uptodate
 257                        (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
 258                        "PAP-5020: parent is not uptodate");
 259
 260                 /* Parent at the path is not in the tree now. */
 261                 if (!B_IS_IN_TREE
 262                     (parent =
 263                      PATH_OFFSET_PBUFFER(chk_path, path_offset)))
 264                         return &MAX_KEY;
 265                 /* Check whether position in the parent is correct. */
 266                 if ((position =
 267                      PATH_OFFSET_POSITION(chk_path,
 268                                           path_offset)) >
 269                     B_NR_ITEMS(parent))
 270                         return &MAX_KEY;
 271                 /* Check whether parent at the path really points to the child. */
 272                 if (B_N_CHILD_NUM(parent, position) !=
 273                     PATH_OFFSET_PBUFFER(chk_path,
 274                                         path_offset + 1)->b_blocknr)
 275                         return &MAX_KEY;
 276                 /* Return delimiting key if position in the parent is not equal to zero. */
 277                 if (position)
 278                         return B_N_PDELIM_KEY(parent, position - 1);
 279         }
 280         /* Return MIN_KEY if we are in the root of the buffer tree. */
 281         if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 282             b_blocknr == SB_ROOT_BLOCK(sb))
 283                 return &MIN_KEY;
 284         return &MAX_KEY;
 285 }
 286
 287 /* Get delimiting key of the buffer at the path and its right neighbor. */
 288 inline const struct reiserfs_key *get_rkey(const struct treepath *chk_path,
 289                                            const struct super_block *sb)
 290 {
 291         int position, path_offset = chk_path->path_length;
 292         struct buffer_head *parent;
 293
 294         RFALSE(path_offset < FIRST_PATH_ELEMENT_OFFSET,
 295                "PAP-5030: invalid offset in the path");
 296
 297         while (path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 298
 299                 RFALSE(!buffer_uptodate
 300                        (PATH_OFFSET_PBUFFER(chk_path, path_offset)),
 301                        "PAP-5040: parent is not uptodate");
 302
 303                 /* Parent at the path is not in the tree now. */
 304                 if (!B_IS_IN_TREE
 305                     (parent =
 306                      PATH_OFFSET_PBUFFER(chk_path, path_offset)))
 307                         return &MIN_KEY;
 308                 /* Check whether position in the parent is correct. */
 309                 if ((position =
 310                      PATH_OFFSET_POSITION(chk_path,
 311                                           path_offset)) >
 312                     B_NR_ITEMS(parent))
 313                         return &MIN_KEY;
 314                 /* Check whether parent at the path really points to the child. */
 315                 if (B_N_CHILD_NUM(parent, position) !=
 316                     PATH_OFFSET_PBUFFER(chk_path,
 317                                         path_offset + 1)->b_blocknr)
 318                         return &MIN_KEY;
 319                 /* Return delimiting key if position in the parent is not the last one. */
 320                 if (position != B_NR_ITEMS(parent))
 321                         return B_N_PDELIM_KEY(parent, position);
 322         }
 323         /* Return MAX_KEY if we are in the root of the buffer tree. */
 324         if (PATH_OFFSET_PBUFFER(chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 325             b_blocknr == SB_ROOT_BLOCK(sb))
 326                 return &MAX_KEY;
 327         return &MIN_KEY;
 328 }
 329
 330 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
 331 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
 332    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
 333    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
 334    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
 335 static inline int key_in_buffer(struct treepath *chk_path,      /* Path which should be checked.  */
 336                                 const struct cpu_key *key,      /* Key which should be checked.   */
 337                                 struct super_block *sb
 338     )
 339 {
 340
 341         RFALSE(!key || chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
 342                || chk_path->path_length > MAX_HEIGHT,
 343                "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
 344                key, chk_path->path_length);
 345         RFALSE(!PATH_PLAST_BUFFER(chk_path)->b_bdev,
 346                "PAP-5060: device must not be NODEV");
 347
 348         if (comp_keys(get_lkey(chk_path, sb), key) == 1)
 349                 /* left delimiting key is bigger, that the key we look for */
 350                 return 0;
 351         /*  if ( comp_keys(key, get_rkey(chk_path, sb)) != -1 ) */
 352         if (comp_keys(get_rkey(chk_path, sb), key) != 1)
 353                 /* key must be less than right delimitiing key */
 354                 return 0;
 355         return 1;
 356 }
 357
 358 int reiserfs_check_path(struct treepath *p)
 359 {
 360         RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
 361                "path not properly relsed");
 362         return 0;
 363 }
 364
 365 /* Drop the reference to each buffer in a path and restore
 366  * dirty bits clean when preparing the buffer for the log.
 367  * This version should only be called from fix_nodes() */
 368 void pathrelse_and_restore(struct super_block *sb,
 369                            struct treepath *search_path)
 370 {
 371         int path_offset = search_path->path_length;
 372
 373         RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 374                "clm-4000: invalid path offset");
 375
 376         while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
 377                 struct buffer_head *bh;
 378                 bh = PATH_OFFSET_PBUFFER(search_path, path_offset--);
 379                 reiserfs_restore_prepared_buffer(sb, bh);
 380                 brelse(bh);
 381         }
 382         search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 383 }
 384
 385 /* Drop the reference to each buffer in a path */
 386 void pathrelse(struct treepath *search_path)
 387 {
 388         int path_offset = search_path->path_length;
 389
 390         RFALSE(path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 391                "PAP-5090: invalid path offset");
 392
 393         while (path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
 394                 brelse(PATH_OFFSET_PBUFFER(search_path, path_offset--));
 395
 396         search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 397 }
 398
 399 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
 400 {
 401         struct block_head *blkh;
 402         struct item_head *ih;
 403         int used_space;
 404         int prev_location;
 405         int i;
 406         int nr;
 407
 408         blkh = (struct block_head *)buf;
 409         if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
 410                 reiserfs_warning(NULL, "reiserfs-5080",
 411                                  "this should be caught earlier");
 412                 return 0;
 413         }
 414
 415         nr = blkh_nr_item(blkh);
 416         if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
 417                 /* item number is too big or too small */
 418                 reiserfs_warning(NULL, "reiserfs-5081",
 419                                  "nr_item seems wrong: %z", bh);
 420                 return 0;
 421         }
 422         ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
 423         used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
 424         if (used_space != blocksize - blkh_free_space(blkh)) {
 425                 /* free space does not match to calculated amount of use space */
 426                 reiserfs_warning(NULL, "reiserfs-5082",
 427                                  "free space seems wrong: %z", bh);
 428                 return 0;
 429         }
 430         // FIXME: it is_leaf will hit performance too much - we may have
 431         // return 1 here
 432
 433         /* check tables of item heads */
 434         ih = (struct item_head *)(buf + BLKH_SIZE);
 435         prev_location = blocksize;
 436         for (i = 0; i < nr; i++, ih++) {
 437                 if (le_ih_k_type(ih) == TYPE_ANY) {
 438                         reiserfs_warning(NULL, "reiserfs-5083",
 439                                          "wrong item type for item %h",
 440                                          ih);
 441                         return 0;
 442                 }
 443                 if (ih_location(ih) >= blocksize
 444                     || ih_location(ih) < IH_SIZE * nr) {
 445                         reiserfs_warning(NULL, "reiserfs-5084",
 446                                          "item location seems wrong: %h",
 447                                          ih);
 448                         return 0;
 449                 }
 450                 if (ih_item_len(ih) < 1
 451                     || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
 452                         reiserfs_warning(NULL, "reiserfs-5085",
 453                                          "item length seems wrong: %h",
 454                                          ih);
 455                         return 0;
 456                 }
 457                 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
 458                         reiserfs_warning(NULL, "reiserfs-5086",
 459                                          "item location seems wrong "
 460                                          "(second one): %h", ih);
 461                         return 0;
 462                 }
 463                 prev_location = ih_location(ih);
 464         }
 465
 466         // one may imagine much more checks
 467         return 1;
 468 }
 469
 470 /* returns 1 if buf looks like an internal node, 0 otherwise */
 471 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
 472 {
 473         struct block_head *blkh;
 474         int nr;
 475         int used_space;
 476
 477         blkh = (struct block_head *)buf;
 478         nr = blkh_level(blkh);
 479         if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
 480                 /* this level is not possible for internal nodes */
 481                 reiserfs_warning(NULL, "reiserfs-5087",
 482                                  "this should be caught earlier");
 483                 return 0;
 484         }
 485
 486         nr = blkh_nr_item(blkh);
 487         if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
 488                 /* for internal which is not root we might check min number of keys */
 489                 reiserfs_warning(NULL, "reiserfs-5088",
 490                                  "number of key seems wrong: %z", bh);
 491                 return 0;
 492         }
 493
 494         used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
 495         if (used_space != blocksize - blkh_free_space(blkh)) {
 496                 reiserfs_warning(NULL, "reiserfs-5089",
 497                                  "free space seems wrong: %z", bh);
 498                 return 0;
 499         }
 500         // one may imagine much more checks
 501         return 1;
 502 }
 503
 504 // make sure that bh contains formatted node of reiserfs tree of
 505 // 'level'-th level
 506 static int is_tree_node(struct buffer_head *bh, int level)
 507 {
 508         if (B_LEVEL(bh) != level) {
 509                 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
 510                                  "not match to the expected one %d",
 511                                  B_LEVEL(bh), level);
 512                 return 0;
 513         }
 514         if (level == DISK_LEAF_NODE_LEVEL)
 515                 return is_leaf(bh->b_data, bh->b_size, bh);
 516
 517         return is_internal(bh->b_data, bh->b_size, bh);
 518 }
 519
 520 #define SEARCH_BY_KEY_READA 16
 521
 522 /* The function is NOT SCHEDULE-SAFE! */
 523 static void search_by_key_reada(struct super_block *s,
 524                                 struct buffer_head **bh,
 525                                 b_blocknr_t *b, int num)
 526 {
 527         int i, j;
 528
 529         for (i = 0; i < num; i++) {
 530                 bh[i] = sb_getblk(s, b[i]);
 531         }
 532         /*
 533          * We are going to read some blocks on which we
 534          * have a reference. It's safe, though we might be
 535          * reading blocks concurrently changed if we release
 536          * the lock. But it's still fine because we check later
 537          * if the tree changed
 538          */
 539         reiserfs_write_unlock(s);
 540         for (j = 0; j < i; j++) {
 541                 /*
 542                  * note, this needs attention if we are getting rid of the BKL
 543                  * you have to make sure the prepared bit isn't set on this buffer
 544                  */
 545                 if (!buffer_uptodate(bh[j]))
 546                         ll_rw_block(READA, 1, bh + j);
 547                 brelse(bh[j]);
 548         }
 549 }
 550
 551 /**************************************************************************
 552  * Algorithm   SearchByKey                                                *
 553  *             look for item in the Disk S+Tree by its key                *
 554  * Input:  sb   -  super block                                            *
 555  *         key  - pointer to the key to search                            *
 556  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
 557  *         search_path - path from the root to the needed leaf            *
 558  **************************************************************************/
 559
 560 /* This function fills up the path from the root to the leaf as it
 561    descends the tree looking for the key.  It uses reiserfs_bread to
 562    try to find buffers in the cache given their block number.  If it
 563    does not find them in the cache it reads them from disk.  For each
 564    node search_by_key finds using reiserfs_bread it then uses
 565    bin_search to look through that node.  bin_search will find the
 566    position of the block_number of the next node if it is looking
 567    through an internal node.  If it is looking through a leaf node
 568    bin_search will find the position of the item which has key either
 569    equal to given key, or which is the maximal key less than the given
 570    key.  search_by_key returns a path that must be checked for the
 571    correctness of the top of the path but need not be checked for the
 572    correctness of the bottom of the path */
 573 /* The function is NOT SCHEDULE-SAFE! */
 574 int search_by_key(struct super_block *sb, const struct cpu_key *key,    /* Key to search. */
 575                   struct treepath *search_path,/* This structure was
 576                                                    allocated and initialized
 577                                                    by the calling
 578                                                    function. It is filled up
 579                                                    by this function.  */
 580                   int stop_level        /* How far down the tree to search. To
 581                                            stop at leaf level - set to
 582                                            DISK_LEAF_NODE_LEVEL */
 583     )
 584 {
 585         b_blocknr_t block_number;
 586         int expected_level;
 587         struct buffer_head *bh;
 588         struct path_element *last_element;
 589         int node_level, retval;
 590         int right_neighbor_of_leaf_node;
 591         int fs_gen;
 592         struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
 593         b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
 594         int reada_count = 0;
 595
 596 #ifdef CONFIG_REISERFS_CHECK
 597         int repeat_counter = 0;
 598 #endif
 599
 600         PROC_INFO_INC(sb, search_by_key);
 601
 602         /* As we add each node to a path we increase its count.  This means that
 603            we must be careful to release all nodes in a path before we either
 604            discard the path struct or re-use the path struct, as we do here. */
 605
 606         pathrelse(search_path);
 607
 608         right_neighbor_of_leaf_node = 0;
 609
 610         /* With each iteration of this loop we search through the items in the
 611            current node, and calculate the next current node(next path element)
 612            for the next iteration of this loop.. */
 613         block_number = SB_ROOT_BLOCK(sb);
 614         expected_level = -1;
 615         while (1) {
 616
 617 #ifdef CONFIG_REISERFS_CHECK
 618                 if (!(++repeat_counter % 50000))
 619                         reiserfs_warning(sb, "PAP-5100",
 620                                          "%s: there were %d iterations of "
 621                                          "while loop looking for key %K",
 622                                          current->comm, repeat_counter,
 623                                          key);
 624 #endif
 625
 626                 /* prep path to have another element added to it. */
 627                 last_element =
 628                     PATH_OFFSET_PELEMENT(search_path,
 629                                          ++search_path->path_length);
 630                 fs_gen = get_generation(sb);
 631
 632                 /* Read the next tree node, and set the last element in the path to
 633                    have a pointer to it. */
 634                 if ((bh = last_element->pe_buffer =
 635                      sb_getblk(sb, block_number))) {
 636                         if (!buffer_uptodate(bh) && reada_count > 1)
 637                                 /* will unlock the write lock */
 638                                 search_by_key_reada(sb, reada_bh,
 639                                                     reada_blocks, reada_count);
 640                         else
 641                                 reiserfs_write_unlock(sb);
 642                         ll_rw_block(READ, 1, &bh);
 643                         wait_on_buffer(bh);
 644                         reiserfs_write_lock(sb);
 645                         if (!buffer_uptodate(bh))
 646                                 goto io_error;
 647                 } else {
 648                       io_error:
 649                         search_path->path_length--;
 650                         pathrelse(search_path);
 651                         return IO_ERROR;
 652                 }
 653                 reada_count = 0;
 654                 if (expected_level == -1)
 655                         expected_level = SB_TREE_HEIGHT(sb);
 656                 expected_level--;
 657
 658                 /* It is possible that schedule occurred. We must check whether the key
 659                    to search is still in the tree rooted from the current buffer. If
 660                    not then repeat search from the root. */
 661                 if (fs_changed(fs_gen, sb) &&
 662                     (!B_IS_IN_TREE(bh) ||
 663                      B_LEVEL(bh) != expected_level ||
 664                      !key_in_buffer(search_path, key, sb))) {
 665                         PROC_INFO_INC(sb, search_by_key_fs_changed);
 666                         PROC_INFO_INC(sb, search_by_key_restarted);
 667                         PROC_INFO_INC(sb,
 668                                       sbk_restarted[expected_level - 1]);
 669                         pathrelse(search_path);
 670
 671                         /* Get the root block number so that we can repeat the search
 672                            starting from the root. */
 673                         block_number = SB_ROOT_BLOCK(sb);
 674                         expected_level = -1;
 675                         right_neighbor_of_leaf_node = 0;
 676
 677                         /* repeat search from the root */
 678                         continue;
 679                 }
 680
 681                 /* only check that the key is in the buffer if key is not
 682                    equal to the MAX_KEY. Latter case is only possible in
 683                    "finish_unfinished()" processing during mount. */
 684                 RFALSE(comp_keys(&MAX_KEY, key) &&
 685                        !key_in_buffer(search_path, key, sb),
 686                        "PAP-5130: key is not in the buffer");
 687 #ifdef CONFIG_REISERFS_CHECK
 688                 if (cur_tb) {
 689                         print_cur_tb("5140");
 690                         reiserfs_panic(sb, "PAP-5140",
 691                                        "schedule occurred in do_balance!");
 692                 }
 693 #endif
 694
 695                 // make sure, that the node contents look like a node of
 696                 // certain level
 697                 if (!is_tree_node(bh, expected_level)) {
 698                         reiserfs_error(sb, "vs-5150",
 699                                        "invalid format found in block %ld. "
 700                                        "Fsck?", bh->b_blocknr);
 701                         pathrelse(search_path);
 702                         return IO_ERROR;
 703                 }
 704
 705                 /* ok, we have acquired next formatted node in the tree */
 706                 node_level = B_LEVEL(bh);
 707
 708                 PROC_INFO_BH_STAT(sb, bh, node_level - 1);
 709
 710                 RFALSE(node_level < stop_level,
 711                        "vs-5152: tree level (%d) is less than stop level (%d)",
 712                        node_level, stop_level);
 713
 714                 retval = bin_search(key, B_N_PITEM_HEAD(bh, 0),
 715                                       B_NR_ITEMS(bh),
 716                                       (node_level ==
 717                                        DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
 718                                       KEY_SIZE,
 719                                       &(last_element->pe_position));
 720                 if (node_level == stop_level) {
 721                         return retval;
 722                 }
 723
 724                 /* we are not in the stop level */
 725                 if (retval == ITEM_FOUND)
 726                         /* item has been found, so we choose the pointer which is to the right of the found one */
 727                         last_element->pe_position++;
 728
 729                 /* if item was not found we choose the position which is to
 730                    the left of the found item. This requires no code,
 731                    bin_search did it already. */
 732
 733                 /* So we have chosen a position in the current node which is
 734                    an internal node.  Now we calculate child block number by
 735                    position in the node. */
 736                 block_number =
 737                     B_N_CHILD_NUM(bh, last_element->pe_position);
 738
 739                 /* if we are going to read leaf nodes, try for read ahead as well */
 740                 if ((search_path->reada & PATH_READA) &&
 741                     node_level == DISK_LEAF_NODE_LEVEL + 1) {
 742                         int pos = last_element->pe_position;
 743                         int limit = B_NR_ITEMS(bh);
 744                         struct reiserfs_key *le_key;
 745
 746                         if (search_path->reada & PATH_READA_BACK)
 747                                 limit = 0;
 748                         while (reada_count < SEARCH_BY_KEY_READA) {
 749                                 if (pos == limit)
 750                                         break;
 751                                 reada_blocks[reada_count++] =
 752                                     B_N_CHILD_NUM(bh, pos);
 753                                 if (search_path->reada & PATH_READA_BACK)
 754                                         pos--;
 755                                 else
 756                                         pos++;
 757
 758                                 /*
 759                                  * check to make sure we're in the same object
 760                                  */
 761                                 le_key = B_N_PDELIM_KEY(bh, pos);
 762                                 if (le32_to_cpu(le_key->k_objectid) !=
 763                                     key->on_disk_key.k_objectid) {
 764                                         break;
 765                                 }
 766                         }
 767                 }
 768         }
 769 }
 770
 771 /* Form the path to an item and position in this item which contains
 772    file byte defined by key. If there is no such item
 773    corresponding to the key, we point the path to the item with
 774    maximal key less than key, and *pos_in_item is set to one
 775    past the last entry/byte in the item.  If searching for entry in a
 776    directory item, and it is not found, *pos_in_item is set to one
 777    entry more than the entry with maximal key which is less than the
 778    sought key.
 779
 780    Note that if there is no entry in this same node which is one more,
 781    then we point to an imaginary entry.  for direct items, the
 782    position is in units of bytes, for indirect items the position is
 783    in units of blocknr entries, for directory items the position is in
 784    units of directory entries.  */
 785
 786 /* The function is NOT SCHEDULE-SAFE! */
 787 int search_for_position_by_key(struct super_block *sb,  /* Pointer to the super block.          */
 788                                const struct cpu_key *p_cpu_key, /* Key to search (cpu variable)         */
 789                                struct treepath *search_path     /* Filled up by this function.          */
 790     )
 791 {
 792         struct item_head *p_le_ih;      /* pointer to on-disk structure */
 793         int blk_size;
 794         loff_t item_offset, offset;
 795         struct reiserfs_dir_entry de;
 796         int retval;
 797
 798         /* If searching for directory entry. */
 799         if (is_direntry_cpu_key(p_cpu_key))
 800                 return search_by_entry_key(sb, p_cpu_key, search_path,
 801                                            &de);
 802
 803         /* If not searching for directory entry. */
 804
 805         /* If item is found. */
 806         retval = search_item(sb, p_cpu_key, search_path);
 807         if (retval == IO_ERROR)
 808                 return retval;
 809         if (retval == ITEM_FOUND) {
 810
 811                 RFALSE(!ih_item_len
 812                        (B_N_PITEM_HEAD
 813                         (PATH_PLAST_BUFFER(search_path),
 814                          PATH_LAST_POSITION(search_path))),
 815                        "PAP-5165: item length equals zero");
 816
 817                 pos_in_item(search_path) = 0;
 818                 return POSITION_FOUND;
 819         }
 820
 821         RFALSE(!PATH_LAST_POSITION(search_path),
 822                "PAP-5170: position equals zero");
 823
 824         /* Item is not found. Set path to the previous item. */
 825         p_le_ih =
 826             B_N_PITEM_HEAD(PATH_PLAST_BUFFER(search_path),
 827                            --PATH_LAST_POSITION(search_path));
 828         blk_size = sb->s_blocksize;
 829
 830         if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
 831                 return FILE_NOT_FOUND;
 832         }
 833         // FIXME: quite ugly this far
 834
 835         item_offset = le_ih_k_offset(p_le_ih);
 836         offset = cpu_key_k_offset(p_cpu_key);
 837
 838         /* Needed byte is contained in the item pointed to by the path. */
 839         if (item_offset <= offset &&
 840             item_offset + op_bytes_number(p_le_ih, blk_size) > offset) {
 841                 pos_in_item(search_path) = offset - item_offset;
 842                 if (is_indirect_le_ih(p_le_ih)) {
 843                         pos_in_item(search_path) /= blk_size;
 844                 }
 845                 return POSITION_FOUND;
 846         }
 847
 848         /* Needed byte is not contained in the item pointed to by the
 849            path. Set pos_in_item out of the item. */
 850         if (is_indirect_le_ih(p_le_ih))
 851                 pos_in_item(search_path) =
 852                     ih_item_len(p_le_ih) / UNFM_P_SIZE;
 853         else
 854                 pos_in_item(search_path) = ih_item_len(p_le_ih);
 855
 856         return POSITION_NOT_FOUND;
 857 }
 858
 859 /* Compare given item and item pointed to by the path. */
 860 int comp_items(const struct item_head *stored_ih, const struct treepath *path)
 861 {
 862         struct buffer_head *bh = PATH_PLAST_BUFFER(path);
 863         struct item_head *ih;
 864
 865         /* Last buffer at the path is not in the tree. */
 866         if (!B_IS_IN_TREE(bh))
 867                 return 1;
 868
 869         /* Last path position is invalid. */
 870         if (PATH_LAST_POSITION(path) >= B_NR_ITEMS(bh))
 871                 return 1;
 872
 873         /* we need only to know, whether it is the same item */
 874         ih = get_ih(path);
 875         return memcmp(stored_ih, ih, IH_SIZE);
 876 }
 877
 878 /* unformatted nodes are not logged anymore, ever.  This is safe
 879 ** now
 880 */
 881 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
 882
 883 // block can not be forgotten as it is in I/O or held by someone
 884 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
 885
 886 // prepare for delete or cut of direct item
 887 static inline int prepare_for_direct_item(struct treepath *path,
 888                                           struct item_head *le_ih,
 889                                           struct inode *inode,
 890                                           loff_t new_file_length, int *cut_size)
 891 {
 892         loff_t round_len;
 893
 894         if (new_file_length == max_reiserfs_offset(inode)) {
 895                 /* item has to be deleted */
 896                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 897                 return M_DELETE;
 898         }
 899         // new file gets truncated
 900         if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
 901                 //
 902                 round_len = ROUND_UP(new_file_length);
 903                 /* this was new_file_length < le_ih ... */
 904                 if (round_len < le_ih_k_offset(le_ih)) {
 905                         *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 906                         return M_DELETE;        /* Delete this item. */
 907                 }
 908                 /* Calculate first position and size for cutting from item. */
 909                 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
 910                 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
 911
 912                 return M_CUT;   /* Cut from this item. */
 913         }
 914
 915         // old file: items may have any length
 916
 917         if (new_file_length < le_ih_k_offset(le_ih)) {
 918                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 919                 return M_DELETE;        /* Delete this item. */
 920         }
 921         /* Calculate first position and size for cutting from item. */
 922         *cut_size = -(ih_item_len(le_ih) -
 923                       (pos_in_item(path) =
 924                        new_file_length + 1 - le_ih_k_offset(le_ih)));
 925         return M_CUT;           /* Cut from this item. */
 926 }
 927
 928 static inline int prepare_for_direntry_item(struct treepath *path,
 929                                             struct item_head *le_ih,
 930                                             struct inode *inode,
 931                                             loff_t new_file_length,
 932                                             int *cut_size)
 933 {
 934         if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
 935             new_file_length == max_reiserfs_offset(inode)) {
 936                 RFALSE(ih_entry_count(le_ih) != 2,
 937                        "PAP-5220: incorrect empty directory item (%h)", le_ih);
 938                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 939                 return M_DELETE;        /* Delete the directory item containing "." and ".." entry. */
 940         }
 941
 942         if (ih_entry_count(le_ih) == 1) {
 943                 /* Delete the directory item such as there is one record only
 944                    in this item */
 945                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 946                 return M_DELETE;
 947         }
 948
 949         /* Cut one record from the directory item. */
 950         *cut_size =
 951             -(DEH_SIZE +
 952               entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
 953         return M_CUT;
 954 }
 955
 956 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
 957
 958 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
 959     If the path points to an indirect item, remove some number of its unformatted nodes.
 960     In case of file truncate calculate whether this item must be deleted/truncated or last
 961     unformatted node of this item will be converted to a direct item.
 962     This function returns a determination of what balance mode the calling function should employ. */
 963 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
 964                                                                                                                                                                                    from end of the file. */
 965                                       int *cut_size, unsigned long long new_file_length /* MAX_KEY_OFFSET in case of delete. */
 966     )
 967 {
 968         struct super_block *sb = inode->i_sb;
 969         struct item_head *p_le_ih = PATH_PITEM_HEAD(path);
 970         struct buffer_head *bh = PATH_PLAST_BUFFER(path);
 971
 972         BUG_ON(!th->t_trans_id);
 973
 974         /* Stat_data item. */
 975         if (is_statdata_le_ih(p_le_ih)) {
 976
 977                 RFALSE(new_file_length != max_reiserfs_offset(inode),
 978                        "PAP-5210: mode must be M_DELETE");
 979
 980                 *cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
 981                 return M_DELETE;
 982         }
 983
 984         /* Directory item. */
 985         if (is_direntry_le_ih(p_le_ih))
 986                 return prepare_for_direntry_item(path, p_le_ih, inode,
 987                                                  new_file_length,
 988                                                  cut_size);
 989
 990         /* Direct item. */
 991         if (is_direct_le_ih(p_le_ih))
 992                 return prepare_for_direct_item(path, p_le_ih, inode,
 993                                                new_file_length, cut_size);
 994
 995         /* Case of an indirect item. */
 996         {
 997             int blk_size = sb->s_blocksize;
 998             struct item_head s_ih;
 999             int need_re_search;
1000             int delete = 0;
1001             int result = M_CUT;
1002             int pos = 0;
1003
1004             if ( new_file_length == max_reiserfs_offset (inode) ) {
1005                 /* prepare_for_delete_or_cut() is called by
1006                  * reiserfs_delete_item() */
1007                 new_file_length = 0;
1008                 delete = 1;
1009             }
1010
1011             do {
1012                 need_re_search = 0;
1013                 *cut_size = 0;
1014                 bh = PATH_PLAST_BUFFER(path);
1015                 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1016                 pos = I_UNFM_NUM(&s_ih);
1017
1018                 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > new_file_length) {
1019                     __le32 *unfm;
1020                     __u32 block;
1021
1022                     /* Each unformatted block deletion may involve one additional
1023                      * bitmap block into the transaction, thereby the initial
1024                      * journal space reservation might not be enough. */
1025                     if (!delete && (*cut_size) != 0 &&
1026                         reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD)
1027                         break;
1028
1029                     unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1;
1030                     block = get_block_num(unfm, 0);
1031
1032                     if (block != 0) {
1033                         reiserfs_prepare_for_journal(sb, bh, 1);
1034                         put_block_num(unfm, 0, 0);
1035                         journal_mark_dirty(th, sb, bh);
1036                         reiserfs_free_block(th, inode, block, 1);
1037                     }
1038
1039                     reiserfs_write_unlock(sb);
1040                     cond_resched();
1041                     reiserfs_write_lock(sb);
1042
1043                     if (item_moved (&s_ih, path))  {
1044                         need_re_search = 1;
1045                         break;
1046                     }
1047
1048                     pos --;
1049                     (*removed)++;
1050                     (*cut_size) -= UNFM_P_SIZE;
1051
1052                     if (pos == 0) {
1053                         (*cut_size) -= IH_SIZE;
1054                         result = M_DELETE;
1055                         break;
1056                     }
1057                 }
1058                 /* a trick.  If the buffer has been logged, this will do nothing.  If
1059                 ** we've broken the loop without logging it, it will restore the
1060                 ** buffer */
1061                 reiserfs_restore_prepared_buffer(sb, bh);
1062             } while (need_re_search &&
1063                      search_for_position_by_key(sb, item_key, path) == POSITION_FOUND);
1064             pos_in_item(path) = pos * UNFM_P_SIZE;
1065
1066             if (*cut_size == 0) {
1067                 /* Nothing were cut. maybe convert last unformatted node to the
1068                  * direct item? */
1069                 result = M_CONVERT;
1070             }
1071             return result;
1072         }
1073 }
1074
1075 /* Calculate number of bytes which will be deleted or cut during balance */
1076 static int calc_deleted_bytes_number(struct tree_balance *tb, char mode)
1077 {
1078         int del_size;
1079         struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path);
1080
1081         if (is_statdata_le_ih(p_le_ih))
1082                 return 0;
1083
1084         del_size =
1085             (mode ==
1086              M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
1087         if (is_direntry_le_ih(p_le_ih)) {
1088                 /* return EMPTY_DIR_SIZE; We delete emty directoris only.
1089                  * we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1090                  * empty size.  ick. FIXME, is this right? */
1091                 return del_size;
1092         }
1093
1094         if (is_indirect_le_ih(p_le_ih))
1095                 del_size = (del_size / UNFM_P_SIZE) *
1096                                 (PATH_PLAST_BUFFER(tb->tb_path)->b_size);
1097         return del_size;
1098 }
1099
1100 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1101                            struct tree_balance *tb,
1102                            struct super_block *sb,
1103                            struct treepath *path, int size)
1104 {
1105
1106         BUG_ON(!th->t_trans_id);
1107
1108         memset(tb, '\0', sizeof(struct tree_balance));
1109         tb->transaction_handle = th;
1110         tb->tb_sb = sb;
1111         tb->tb_path = path;
1112         PATH_OFFSET_PBUFFER(path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1113         PATH_OFFSET_POSITION(path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1114         tb->insert_size[0] = size;
1115 }
1116
1117 void padd_item(char *item, int total_length, int length)
1118 {
1119         int i;
1120
1121         for (i = total_length; i > length;)
1122                 item[--i] = 0;
1123 }
1124
1125 #ifdef REISERQUOTA_DEBUG
1126 char key2type(struct reiserfs_key *ih)
1127 {
1128         if (is_direntry_le_key(2, ih))
1129                 return 'd';
1130         if (is_direct_le_key(2, ih))
1131                 return 'D';
1132         if (is_indirect_le_key(2, ih))
1133                 return 'i';
1134         if (is_statdata_le_key(2, ih))
1135                 return 's';
1136         return 'u';
1137 }
1138
1139 char head2type(struct item_head *ih)
1140 {
1141         if (is_direntry_le_ih(ih))
1142                 return 'd';
1143         if (is_direct_le_ih(ih))
1144                 return 'D';
1145         if (is_indirect_le_ih(ih))
1146                 return 'i';
1147         if (is_statdata_le_ih(ih))
1148                 return 's';
1149         return 'u';
1150 }
1151 #endif
1152
1153 /* Delete object item.
1154  * th       - active transaction handle
1155  * path     - path to the deleted item
1156  * item_key - key to search for the deleted item
1157  * indode   - used for updating i_blocks and quotas
1158  * un_bh    - NULL or unformatted node pointer
1159  */
1160 int reiserfs_delete_item(struct reiserfs_transaction_handle *th,
1161                          struct treepath *path, const struct cpu_key *item_key,
1162                          struct inode *inode, struct buffer_head *un_bh)
1163 {
1164         struct super_block *sb = inode->i_sb;
1165         struct tree_balance s_del_balance;
1166         struct item_head s_ih;
1167         struct item_head *q_ih;
1168         int quota_cut_bytes;
1169         int ret_value, del_size, removed;
1170
1171 #ifdef CONFIG_REISERFS_CHECK
1172         char mode;
1173         int iter = 0;
1174 #endif
1175
1176         BUG_ON(!th->t_trans_id);
1177
1178         init_tb_struct(th, &s_del_balance, sb, path,
1179                        0 /*size is unknown */ );
1180
1181         while (1) {
1182                 removed = 0;
1183
1184 #ifdef CONFIG_REISERFS_CHECK
1185                 iter++;
1186                 mode =
1187 #endif
1188                     prepare_for_delete_or_cut(th, inode, path,
1189                                               item_key, &removed,
1190                                               &del_size,
1191                                               max_reiserfs_offset(inode));
1192
1193                 RFALSE(mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1194
1195                 copy_item_head(&s_ih, PATH_PITEM_HEAD(path));
1196                 s_del_balance.insert_size[0] = del_size;
1197
1198                 ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1199                 if (ret_value != REPEAT_SEARCH)
1200                         break;
1201
1202                 PROC_INFO_INC(sb, delete_item_restarted);
1203
1204                 // file system changed, repeat search
1205                 ret_value =
1206                     search_for_position_by_key(sb, item_key, path);
1207                 if (ret_value == IO_ERROR)
1208                         break;
1209                 if (ret_value == FILE_NOT_FOUND) {
1210                         reiserfs_warning(sb, "vs-5340",
1211                                          "no items of the file %K found",
1212                                          item_key);
1213                         break;
1214                 }
1215         }                       /* while (1) */
1216
1217         if (ret_value != CARRY_ON) {
1218                 unfix_nodes(&s_del_balance);
1219                 return 0;
1220         }
1221         // reiserfs_delete_item returns item length when success
1222         ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1223         q_ih = get_ih(path);
1224         quota_cut_bytes = ih_item_len(q_ih);
1225
1226         /* hack so the quota code doesn't have to guess if the file
1227          ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1228          ** We test the offset because the tail might have been
1229          ** split into multiple items, and we only want to decrement for
1230          ** the unfm node once
1231          */
1232         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
1233                 if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1234                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1235                 } else {
1236                         quota_cut_bytes = 0;
1237                 }
1238         }
1239
1240         if (un_bh) {
1241                 int off;
1242                 char *data;
1243
1244                 /* We are in direct2indirect conversion, so move tail contents
1245                    to the unformatted node */
1246                 /* note, we do the copy before preparing the buffer because we
1247                  ** don't care about the contents of the unformatted node yet.
1248                  ** the only thing we really care about is the direct item's data
1249                  ** is in the unformatted node.
1250                  **
1251                  ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1252                  ** the unformatted node, which might schedule, meaning we'd have to
1253                  ** loop all the way back up to the start of the while loop.
1254                  **
1255                  ** The unformatted node must be dirtied later on.  We can't be
1256                  ** sure here if the entire tail has been deleted yet.
1257                  **
1258                  ** un_bh is from the page cache (all unformatted nodes are
1259                  ** from the page cache) and might be a highmem page.  So, we
1260                  ** can't use un_bh->b_data.
1261                  ** -clm
1262                  */
1263
1264                 data = kmap_atomic(un_bh->b_page, KM_USER0);
1265                 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1266                 memcpy(data + off,
1267                        B_I_PITEM(PATH_PLAST_BUFFER(path), &s_ih),
1268                        ret_value);
1269                 kunmap_atomic(data, KM_USER0);
1270         }
1271         /* Perform balancing after all resources have been collected at once. */
1272         do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1273
1274 #ifdef REISERQUOTA_DEBUG
1275         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1276                        "reiserquota delete_item(): freeing %u, id=%u type=%c",
1277                        quota_cut_bytes, inode->i_uid, head2type(&s_ih));
1278 #endif
1279         vfs_dq_free_space_nodirty(inode, quota_cut_bytes);
1280
1281         /* Return deleted body length */
1282         return ret_value;
1283 }
1284
1285 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1286
1287  deletion of the body of the object is performed by iput(), with the
1288  result that if multiple processes are operating on a file, the
1289  deletion of the body of the file is deferred until the last process
1290  that has an open inode performs its iput().
1291
1292  writes and truncates are protected from collisions by use of
1293  semaphores.
1294
1295  creates, linking, and mknod are protected from collisions with other
1296  processes by making the reiserfs_add_entry() the last step in the
1297  creation, and then rolling back all changes if there was a collision.
1298  - Hans
1299 */
1300
1301 /* this deletes item which never gets split */
1302 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1303                                 struct inode *inode, struct reiserfs_key *key)
1304 {
1305         struct tree_balance tb;
1306         INITIALIZE_PATH(path);
1307         int item_len = 0;
1308         int tb_init = 0;
1309         struct cpu_key cpu_key;
1310         int retval;
1311         int quota_cut_bytes = 0;
1312
1313         BUG_ON(!th->t_trans_id);
1314
1315         le_key2cpu_key(&cpu_key, key);
1316
1317         while (1) {
1318                 retval = search_item(th->t_super, &cpu_key, &path);
1319                 if (retval == IO_ERROR) {
1320                         reiserfs_error(th->t_super, "vs-5350",
1321                                        "i/o failure occurred trying "
1322                                        "to delete %K", &cpu_key);
1323                         break;
1324                 }
1325                 if (retval != ITEM_FOUND) {
1326                         pathrelse(&path);
1327                         // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1328                         if (!
1329                             ((unsigned long long)
1330                              GET_HASH_VALUE(le_key_k_offset
1331                                             (le_key_version(key), key)) == 0
1332                              && (unsigned long long)
1333                              GET_GENERATION_NUMBER(le_key_k_offset
1334                                                    (le_key_version(key),
1335                                                     key)) == 1))
1336                                 reiserfs_warning(th->t_super, "vs-5355",
1337                                                  "%k not found", key);
1338                         break;
1339                 }
1340                 if (!tb_init) {
1341                         tb_init = 1;
1342                         item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1343                         init_tb_struct(th, &tb, th->t_super, &path,
1344                                        -(IH_SIZE + item_len));
1345                 }
1346                 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1347
1348                 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1349                 if (retval == REPEAT_SEARCH) {
1350                         PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1351                         continue;
1352                 }
1353
1354                 if (retval == CARRY_ON) {
1355                         do_balance(&tb, NULL, NULL, M_DELETE);
1356                         if (inode) {    /* Should we count quota for item? (we don't count quotas for save-links) */
1357 #ifdef REISERQUOTA_DEBUG
1358                                 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1359                                                "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1360                                                quota_cut_bytes, inode->i_uid,
1361                                                key2type(key));
1362 #endif
1363                                 vfs_dq_free_space_nodirty(inode,
1364                                                          quota_cut_bytes);
1365                         }
1366                         break;
1367                 }
1368                 // IO_ERROR, NO_DISK_SPACE, etc
1369                 reiserfs_warning(th->t_super, "vs-5360",
1370                                  "could not delete %K due to fix_nodes failure",
1371                                  &cpu_key);
1372                 unfix_nodes(&tb);
1373                 break;
1374         }
1375
1376         reiserfs_check_path(&path);
1377 }
1378
1379 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1380                            struct inode *inode)
1381 {
1382         int err;
1383         inode->i_size = 0;
1384         BUG_ON(!th->t_trans_id);
1385
1386         /* for directory this deletes item containing "." and ".." */
1387         err =
1388             reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1389         if (err)
1390                 return err;
1391
1392 #if defined( USE_INODE_GENERATION_COUNTER )
1393         if (!old_format_only(th->t_super)) {
1394                 __le32 *inode_generation;
1395
1396                 inode_generation =
1397                     &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1398                 le32_add_cpu(inode_generation, 1);
1399         }
1400 /* USE_INODE_GENERATION_COUNTER */
1401 #endif
1402         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1403
1404         return err;
1405 }
1406
1407 static void unmap_buffers(struct page *page, loff_t pos)
1408 {
1409         struct buffer_head *bh;
1410         struct buffer_head *head;
1411         struct buffer_head *next;
1412         unsigned long tail_index;
1413         unsigned long cur_index;
1414
1415         if (page) {
1416                 if (page_has_buffers(page)) {
1417                         tail_index = pos & (PAGE_CACHE_SIZE - 1);
1418                         cur_index = 0;
1419                         head = page_buffers(page);
1420                         bh = head;
1421                         do {
1422                                 next = bh->b_this_page;
1423
1424                                 /* we want to unmap the buffers that contain the tail, and
1425                                  ** all the buffers after it (since the tail must be at the
1426                                  ** end of the file).  We don't want to unmap file data
1427                                  ** before the tail, since it might be dirty and waiting to
1428                                  ** reach disk
1429                                  */
1430                                 cur_index += bh->b_size;
1431                                 if (cur_index > tail_index) {
1432                                         reiserfs_unmap_buffer(bh);
1433                                 }
1434                                 bh = next;
1435                         } while (bh != head);
1436                 }
1437         }
1438 }
1439
1440 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1441                                     struct inode *inode,
1442                                     struct page *page,
1443                                     struct treepath *path,
1444                                     const struct cpu_key *item_key,
1445                                     loff_t new_file_size, char *mode)
1446 {
1447         struct super_block *sb = inode->i_sb;
1448         int block_size = sb->s_blocksize;
1449         int cut_bytes;
1450         BUG_ON(!th->t_trans_id);
1451         BUG_ON(new_file_size != inode->i_size);
1452
1453         /* the page being sent in could be NULL if there was an i/o error
1454          ** reading in the last block.  The user will hit problems trying to
1455          ** read the file, but for now we just skip the indirect2direct
1456          */
1457         if (atomic_read(&inode->i_count) > 1 ||
1458             !tail_has_to_be_packed(inode) ||
1459             !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
1460                 /* leave tail in an unformatted node */
1461                 *mode = M_SKIP_BALANCING;
1462                 cut_bytes =
1463                     block_size - (new_file_size & (block_size - 1));
1464                 pathrelse(path);
1465                 return cut_bytes;
1466         }
1467         /* Perform the conversion to a direct_item. */
1468         /* return indirect_to_direct(inode, path, item_key,
1469                                   new_file_size, mode); */
1470         return indirect2direct(th, inode, page, path, item_key,
1471                                new_file_size, mode);
1472 }
1473
1474 /* we did indirect_to_direct conversion. And we have inserted direct
1475    item successesfully, but there were no disk space to cut unfm
1476    pointer being converted. Therefore we have to delete inserted
1477    direct item(s) */
1478 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1479                                          struct inode *inode, struct treepath *path)
1480 {
1481         struct cpu_key tail_key;
1482         int tail_len;
1483         int removed;
1484         BUG_ON(!th->t_trans_id);
1485
1486         make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);      // !!!!
1487         tail_key.key_length = 4;
1488
1489         tail_len =
1490             (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1491         while (tail_len) {
1492                 /* look for the last byte of the tail */
1493                 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1494                     POSITION_NOT_FOUND)
1495                         reiserfs_panic(inode->i_sb, "vs-5615",
1496                                        "found invalid item");
1497                 RFALSE(path->pos_in_item !=
1498                        ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1499                        "vs-5616: appended bytes found");
1500                 PATH_LAST_POSITION(path)--;
1501
1502                 removed =
1503                     reiserfs_delete_item(th, path, &tail_key, inode,
1504                                          NULL /*unbh not needed */ );
1505                 RFALSE(removed <= 0
1506                        || removed > tail_len,
1507                        "vs-5617: there was tail %d bytes, removed item length %d bytes",
1508                        tail_len, removed);
1509                 tail_len -= removed;
1510                 set_cpu_key_k_offset(&tail_key,
1511                                      cpu_key_k_offset(&tail_key) - removed);
1512         }
1513         reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1514                          "conversion has been rolled back due to "
1515                          "lack of disk space");
1516         //mark_file_without_tail (inode);
1517         mark_inode_dirty(inode);
1518 }
1519
1520 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1521 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1522                            struct treepath *path,
1523                            struct cpu_key *item_key,
1524                            struct inode *inode,
1525                            struct page *page, loff_t new_file_size)
1526 {
1527         struct super_block *sb = inode->i_sb;
1528         /* Every function which is going to call do_balance must first
1529            create a tree_balance structure.  Then it must fill up this
1530            structure by using the init_tb_struct and fix_nodes functions.
1531            After that we can make tree balancing. */
1532         struct tree_balance s_cut_balance;
1533         struct item_head *p_le_ih;
1534         int cut_size = 0,       /* Amount to be cut. */
1535             ret_value = CARRY_ON, removed = 0,  /* Number of the removed unformatted nodes. */
1536             is_inode_locked = 0;
1537         char mode;              /* Mode of the balance. */
1538         int retval2 = -1;
1539         int quota_cut_bytes;
1540         loff_t tail_pos = 0;
1541
1542         BUG_ON(!th->t_trans_id);
1543
1544         init_tb_struct(th, &s_cut_balance, inode->i_sb, path,
1545                        cut_size);
1546
1547         /* Repeat this loop until we either cut the item without needing
1548            to balance, or we fix_nodes without schedule occurring */
1549         while (1) {
1550                 /* Determine the balance mode, position of the first byte to
1551                    be cut, and size to be cut.  In case of the indirect item
1552                    free unformatted nodes which are pointed to by the cut
1553                    pointers. */
1554
1555                 mode =
1556                     prepare_for_delete_or_cut(th, inode, path,
1557                                               item_key, &removed,
1558                                               &cut_size, new_file_size);
1559                 if (mode == M_CONVERT) {
1560                         /* convert last unformatted node to direct item or leave
1561                            tail in the unformatted node */
1562                         RFALSE(ret_value != CARRY_ON,
1563                                "PAP-5570: can not convert twice");
1564
1565                         ret_value =
1566                             maybe_indirect_to_direct(th, inode, page,
1567                                                      path, item_key,
1568                                                      new_file_size, &mode);
1569                         if (mode == M_SKIP_BALANCING)
1570                                 /* tail has been left in the unformatted node */
1571                                 return ret_value;
1572
1573                         is_inode_locked = 1;
1574
1575                         /* removing of last unformatted node will change value we
1576                            have to return to truncate. Save it */
1577                         retval2 = ret_value;
1578                         /*retval2 = sb->s_blocksize - (new_file_size & (sb->s_blocksize - 1)); */
1579
1580                         /* So, we have performed the first part of the conversion:
1581                            inserting the new direct item.  Now we are removing the
1582                            last unformatted node pointer. Set key to search for
1583                            it. */
1584                         set_cpu_key_k_type(item_key, TYPE_INDIRECT);
1585                         item_key->key_length = 4;
1586                         new_file_size -=
1587                             (new_file_size & (sb->s_blocksize - 1));
1588                         tail_pos = new_file_size;
1589                         set_cpu_key_k_offset(item_key, new_file_size + 1);
1590                         if (search_for_position_by_key
1591                             (sb, item_key,
1592                              path) == POSITION_NOT_FOUND) {
1593                                 print_block(PATH_PLAST_BUFFER(path), 3,
1594                                             PATH_LAST_POSITION(path) - 1,
1595                                             PATH_LAST_POSITION(path) + 1);
1596                                 reiserfs_panic(sb, "PAP-5580", "item to "
1597                                                "convert does not exist (%K)",
1598                                                item_key);
1599                         }
1600                         continue;
1601                 }
1602                 if (cut_size == 0) {
1603                         pathrelse(path);
1604                         return 0;
1605                 }
1606
1607                 s_cut_balance.insert_size[0] = cut_size;
1608
1609                 ret_value = fix_nodes(mode, &s_cut_balance, NULL, NULL);
1610                 if (ret_value != REPEAT_SEARCH)
1611                         break;
1612
1613                 PROC_INFO_INC(sb, cut_from_item_restarted);
1614
1615                 ret_value =
1616                     search_for_position_by_key(sb, item_key, path);
1617                 if (ret_value == POSITION_FOUND)
1618                         continue;
1619
1620                 reiserfs_warning(sb, "PAP-5610", "item %K not found",
1621                                  item_key);
1622                 unfix_nodes(&s_cut_balance);
1623                 return (ret_value == IO_ERROR) ? -EIO : -ENOENT;
1624         }                       /* while */
1625
1626         // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1627         if (ret_value != CARRY_ON) {
1628                 if (is_inode_locked) {
1629                         // FIXME: this seems to be not needed: we are always able
1630                         // to cut item
1631                         indirect_to_direct_roll_back(th, inode, path);
1632                 }
1633                 if (ret_value == NO_DISK_SPACE)
1634                         reiserfs_warning(sb, "reiserfs-5092",
1635                                          "NO_DISK_SPACE");
1636                 unfix_nodes(&s_cut_balance);
1637                 return -EIO;
1638         }
1639
1640         /* go ahead and perform balancing */
1641
1642         RFALSE(mode == M_PASTE || mode == M_INSERT, "invalid mode");
1643
1644         /* Calculate number of bytes that need to be cut from the item. */
1645         quota_cut_bytes =
1646             (mode ==
1647              M_DELETE) ? ih_item_len(get_ih(path)) : -s_cut_balance.
1648             insert_size[0];
1649         if (retval2 == -1)
1650                 ret_value = calc_deleted_bytes_number(&s_cut_balance, mode);
1651         else
1652                 ret_value = retval2;
1653
1654         /* For direct items, we only change the quota when deleting the last
1655          ** item.
1656          */
1657         p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1658         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1659                 if (mode == M_DELETE &&
1660                     (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
1661                     1) {
1662                         // FIXME: this is to keep 3.5 happy
1663                         REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
1664                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1665                 } else {
1666                         quota_cut_bytes = 0;
1667                 }
1668         }
1669 #ifdef CONFIG_REISERFS_CHECK
1670         if (is_inode_locked) {
1671                 struct item_head *le_ih =
1672                     PATH_PITEM_HEAD(s_cut_balance.tb_path);
1673                 /* we are going to complete indirect2direct conversion. Make
1674                    sure, that we exactly remove last unformatted node pointer
1675                    of the item */
1676                 if (!is_indirect_le_ih(le_ih))
1677                         reiserfs_panic(sb, "vs-5652",
1678                                        "item must be indirect %h", le_ih);
1679
1680                 if (mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1681                         reiserfs_panic(sb, "vs-5653", "completing "
1682                                        "indirect2direct conversion indirect "
1683                                        "item %h being deleted must be of "
1684                                        "4 byte long", le_ih);
1685
1686                 if (mode == M_CUT
1687                     && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1688                         reiserfs_panic(sb, "vs-5654", "can not complete "
1689                                        "indirect2direct conversion of %h "
1690                                        "(CUT, insert_size==%d)",
1691                                        le_ih, s_cut_balance.insert_size[0]);
1692                 }
1693                 /* it would be useful to make sure, that right neighboring
1694                    item is direct item of this file */
1695         }
1696 #endif
1697
1698         do_balance(&s_cut_balance, NULL, NULL, mode);
1699         if (is_inode_locked) {
1700                 /* we've done an indirect->direct conversion.  when the data block
1701                  ** was freed, it was removed from the list of blocks that must
1702                  ** be flushed before the transaction commits, make sure to
1703                  ** unmap and invalidate it
1704                  */
1705                 unmap_buffers(page, tail_pos);
1706                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
1707         }
1708 #ifdef REISERQUOTA_DEBUG
1709         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1710                        "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1711                        quota_cut_bytes, inode->i_uid, '?');
1712 #endif
1713         vfs_dq_free_space_nodirty(inode, quota_cut_bytes);
1714         return ret_value;
1715 }
1716
1717 static void truncate_directory(struct reiserfs_transaction_handle *th,
1718                                struct inode *inode)
1719 {
1720         BUG_ON(!th->t_trans_id);
1721         if (inode->i_nlink)
1722                 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1723
1724         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1725         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1726         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1727         reiserfs_update_sd(th, inode);
1728         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1729         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1730 }
1731
1732 /* Truncate file to the new size. Note, this must be called with a transaction
1733    already started */
1734 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1735                           struct inode *inode,  /* ->i_size contains new size */
1736                          struct page *page,     /* up to date for last block */
1737                          int update_timestamps  /* when it is called by
1738                                                    file_release to convert
1739                                                    the tail - no timestamps
1740                                                    should be updated */
1741     )
1742 {
1743         INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1744         struct item_head *p_le_ih;      /* Pointer to an item header. */
1745         struct cpu_key s_item_key;      /* Key to search for a previous file item. */
1746         loff_t file_size,       /* Old file size. */
1747          new_file_size; /* New file size. */
1748         int deleted;            /* Number of deleted or truncated bytes. */
1749         int retval;
1750         int err = 0;
1751
1752         BUG_ON(!th->t_trans_id);
1753         if (!
1754             (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1755              || S_ISLNK(inode->i_mode)))
1756                 return 0;
1757
1758         if (S_ISDIR(inode->i_mode)) {
1759                 // deletion of directory - no need to update timestamps
1760                 truncate_directory(th, inode);
1761                 return 0;
1762         }
1763
1764         /* Get new file size. */
1765         new_file_size = inode->i_size;
1766
1767         // FIXME: note, that key type is unimportant here
1768         make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
1769                      TYPE_DIRECT, 3);
1770
1771         retval =
1772             search_for_position_by_key(inode->i_sb, &s_item_key,
1773                                        &s_search_path);
1774         if (retval == IO_ERROR) {
1775                 reiserfs_error(inode->i_sb, "vs-5657",
1776                                "i/o failure occurred trying to truncate %K",
1777                                &s_item_key);
1778                 err = -EIO;
1779                 goto out;
1780         }
1781         if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1782                 reiserfs_error(inode->i_sb, "PAP-5660",
1783                                "wrong result %d of search for %K", retval,
1784                                &s_item_key);
1785
1786                 err = -EIO;
1787                 goto out;
1788         }
1789
1790         s_search_path.pos_in_item--;
1791
1792         /* Get real file size (total length of all file items) */
1793         p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1794         if (is_statdata_le_ih(p_le_ih))
1795                 file_size = 0;
1796         else {
1797                 loff_t offset = le_ih_k_offset(p_le_ih);
1798                 int bytes =
1799                     op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
1800
1801                 /* this may mismatch with real file size: if last direct item
1802                    had no padding zeros and last unformatted node had no free
1803                    space, this file would have this file size */
1804                 file_size = offset + bytes - 1;
1805         }
1806         /*
1807          * are we doing a full truncate or delete, if so
1808          * kick in the reada code
1809          */
1810         if (new_file_size == 0)
1811                 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1812
1813         if (file_size == 0 || file_size < new_file_size) {
1814                 goto update_and_out;
1815         }
1816
1817         /* Update key to search for the last file item. */
1818         set_cpu_key_k_offset(&s_item_key, file_size);
1819
1820         do {
1821                 /* Cut or delete file item. */
1822                 deleted =
1823                     reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1824                                            inode, page, new_file_size);
1825                 if (deleted < 0) {
1826                         reiserfs_warning(inode->i_sb, "vs-5665",
1827                                          "reiserfs_cut_from_item failed");
1828                         reiserfs_check_path(&s_search_path);
1829                         return 0;
1830                 }
1831
1832                 RFALSE(deleted > file_size,
1833                        "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1834                        deleted, file_size, &s_item_key);
1835
1836                 /* Change key to search the last file item. */
1837                 file_size -= deleted;
1838
1839                 set_cpu_key_k_offset(&s_item_key, file_size);
1840
1841                 /* While there are bytes to truncate and previous file item is presented in the tree. */
1842
1843                 /*
1844                  ** This loop could take a really long time, and could log
1845                  ** many more blocks than a transaction can hold.  So, we do a polite
1846                  ** journal end here, and if the transaction needs ending, we make
1847                  ** sure the file is consistent before ending the current trans
1848                  ** and starting a new one
1849                  */
1850                 if (journal_transaction_should_end(th, 0) ||
1851                     reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1852                         int orig_len_alloc = th->t_blocks_allocated;
1853                         pathrelse(&s_search_path);
1854
1855                         if (update_timestamps) {
1856                                 inode->i_mtime = CURRENT_TIME_SEC;
1857                                 inode->i_ctime = CURRENT_TIME_SEC;
1858                         }
1859                         reiserfs_update_sd(th, inode);
1860
1861                         err = journal_end(th, inode->i_sb, orig_len_alloc);
1862                         if (err)
1863                                 goto out;
1864                         err = journal_begin(th, inode->i_sb,
1865                                             JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1866                         if (err)
1867                                 goto out;
1868                         reiserfs_update_inode_transaction(inode);
1869                 }
1870         } while (file_size > ROUND_UP(new_file_size) &&
1871                  search_for_position_by_key(inode->i_sb, &s_item_key,
1872                                             &s_search_path) == POSITION_FOUND);
1873
1874         RFALSE(file_size > ROUND_UP(new_file_size),
1875                "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1876                new_file_size, file_size, s_item_key.on_disk_key.k_objectid);
1877
1878       update_and_out:
1879         if (update_timestamps) {
1880                 // this is truncate, not file closing
1881                 inode->i_mtime = CURRENT_TIME_SEC;
1882                 inode->i_ctime = CURRENT_TIME_SEC;
1883         }
1884         reiserfs_update_sd(th, inode);
1885
1886       out:
1887         pathrelse(&s_search_path);
1888         return err;
1889 }
1890
1891 #ifdef CONFIG_REISERFS_CHECK
1892 // this makes sure, that we __append__, not overwrite or add holes
1893 static void check_research_for_paste(struct treepath *path,
1894                                      const struct cpu_key *key)
1895 {
1896         struct item_head *found_ih = get_ih(path);
1897
1898         if (is_direct_le_ih(found_ih)) {
1899                 if (le_ih_k_offset(found_ih) +
1900                     op_bytes_number(found_ih,
1901                                     get_last_bh(path)->b_size) !=
1902                     cpu_key_k_offset(key)
1903                     || op_bytes_number(found_ih,
1904                                        get_last_bh(path)->b_size) !=
1905                     pos_in_item(path))
1906                         reiserfs_panic(NULL, "PAP-5720", "found direct item "
1907                                        "%h or position (%d) does not match "
1908                                        "to key %K", found_ih,
1909                                        pos_in_item(path), key);
1910         }
1911         if (is_indirect_le_ih(found_ih)) {
1912                 if (le_ih_k_offset(found_ih) +
1913                     op_bytes_number(found_ih,
1914                                     get_last_bh(path)->b_size) !=
1915                     cpu_key_k_offset(key)
1916                     || I_UNFM_NUM(found_ih) != pos_in_item(path)
1917                     || get_ih_free_space(found_ih) != 0)
1918                         reiserfs_panic(NULL, "PAP-5730", "found indirect "
1919                                        "item (%h) or position (%d) does not "
1920                                        "match to key (%K)",
1921                                        found_ih, pos_in_item(path), key);
1922         }
1923 }
1924 #endif                          /* config reiserfs check */
1925
1926 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1927 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *search_path,      /* Path to the pasted item.       */
1928                              const struct cpu_key *key, /* Key to search for the needed item. */
1929                              struct inode *inode,       /* Inode item belongs to */
1930                              const char *body,  /* Pointer to the bytes to paste.    */
1931                              int pasted_size)
1932 {                               /* Size of pasted bytes.             */
1933         struct tree_balance s_paste_balance;
1934         int retval;
1935         int fs_gen;
1936
1937         BUG_ON(!th->t_trans_id);
1938
1939         fs_gen = get_generation(inode->i_sb);
1940
1941 #ifdef REISERQUOTA_DEBUG
1942         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1943                        "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1944                        pasted_size, inode->i_uid,
1945                        key2type(&(key->on_disk_key)));
1946 #endif
1947
1948         if (vfs_dq_alloc_space_nodirty(inode, pasted_size)) {
1949                 pathrelse(search_path);
1950                 return -EDQUOT;
1951         }
1952         init_tb_struct(th, &s_paste_balance, th->t_super, search_path,
1953                        pasted_size);
1954 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1955         s_paste_balance.key = key->on_disk_key;
1956 #endif
1957
1958         /* DQUOT_* can schedule, must check before the fix_nodes */
1959         if (fs_changed(fs_gen, inode->i_sb)) {
1960                 goto search_again;
1961         }
1962
1963         while ((retval =
1964                 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1965                           body)) == REPEAT_SEARCH) {
1966               search_again:
1967                 /* file system changed while we were in the fix_nodes */
1968                 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1969                 retval =
1970                     search_for_position_by_key(th->t_super, key,
1971                                                search_path);
1972                 if (retval == IO_ERROR) {
1973                         retval = -EIO;
1974                         goto error_out;
1975                 }
1976                 if (retval == POSITION_FOUND) {
1977                         reiserfs_warning(inode->i_sb, "PAP-5710",
1978                                          "entry or pasted byte (%K) exists",
1979                                          key);
1980                         retval = -EEXIST;
1981                         goto error_out;
1982                 }
1983 #ifdef CONFIG_REISERFS_CHECK
1984                 check_research_for_paste(search_path, key);
1985 #endif
1986         }
1987
1988         /* Perform balancing after all resources are collected by fix_nodes, and
1989            accessing them will not risk triggering schedule. */
1990         if (retval == CARRY_ON) {
1991                 do_balance(&s_paste_balance, NULL /*ih */ , body, M_PASTE);
1992                 return 0;
1993         }
1994         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1995       error_out:
1996         /* this also releases the path */
1997         unfix_nodes(&s_paste_balance);
1998 #ifdef REISERQUOTA_DEBUG
1999         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2000                        "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2001                        pasted_size, inode->i_uid,
2002                        key2type(&(key->on_disk_key)));
2003 #endif
2004         vfs_dq_free_space_nodirty(inode, pasted_size);
2005         return retval;
2006 }
2007
2008 /* Insert new item into the buffer at the path.
2009  * th   - active transaction handle
2010  * path - path to the inserted item
2011  * ih   - pointer to the item header to insert
2012  * body - pointer to the bytes to insert
2013  */
2014 int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
2015                          struct treepath *path, const struct cpu_key *key,
2016                          struct item_head *ih, struct inode *inode,
2017                          const char *body)
2018 {
2019         struct tree_balance s_ins_balance;
2020         int retval;
2021         int fs_gen = 0;
2022         int quota_bytes = 0;
2023
2024         BUG_ON(!th->t_trans_id);
2025
2026         if (inode) {            /* Do we count quotas for item? */
2027                 fs_gen = get_generation(inode->i_sb);
2028                 quota_bytes = ih_item_len(ih);
2029
2030                 /* hack so the quota code doesn't have to guess if the file has
2031                  ** a tail, links are always tails, so there's no guessing needed
2032                  */
2033                 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(ih))
2034                         quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2035 #ifdef REISERQUOTA_DEBUG
2036                 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2037                                "reiserquota insert_item(): allocating %u id=%u type=%c",
2038                                quota_bytes, inode->i_uid, head2type(ih));
2039 #endif
2040                 /* We can't dirty inode here. It would be immediately written but
2041                  * appropriate stat item isn't inserted yet... */
2042                 if (vfs_dq_alloc_space_nodirty(inode, quota_bytes)) {
2043                         pathrelse(path);
2044                         return -EDQUOT;
2045                 }
2046         }
2047         init_tb_struct(th, &s_ins_balance, th->t_super, path,
2048                        IH_SIZE + ih_item_len(ih));
2049 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2050         s_ins_balance.key = key->on_disk_key;
2051 #endif
2052         /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2053         if (inode && fs_changed(fs_gen, inode->i_sb)) {
2054                 goto search_again;
2055         }
2056
2057         while ((retval =
2058                 fix_nodes(M_INSERT, &s_ins_balance, ih,
2059                           body)) == REPEAT_SEARCH) {
2060               search_again:
2061                 /* file system changed while we were in the fix_nodes */
2062                 PROC_INFO_INC(th->t_super, insert_item_restarted);
2063                 retval = search_item(th->t_super, key, path);
2064                 if (retval == IO_ERROR) {
2065                         retval = -EIO;
2066                         goto error_out;
2067                 }
2068                 if (retval == ITEM_FOUND) {
2069                         reiserfs_warning(th->t_super, "PAP-5760",
2070                                          "key %K already exists in the tree",
2071                                          key);
2072                         retval = -EEXIST;
2073                         goto error_out;
2074                 }
2075         }
2076
2077         /* make balancing after all resources will be collected at a time */
2078         if (retval == CARRY_ON) {
2079                 do_balance(&s_ins_balance, ih, body, M_INSERT);
2080                 return 0;
2081         }
2082
2083         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2084       error_out:
2085         /* also releases the path */
2086         unfix_nodes(&s_ins_balance);
2087 #ifdef REISERQUOTA_DEBUG
2088         reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2089                        "reiserquota insert_item(): freeing %u id=%u type=%c",
2090                        quota_bytes, inode->i_uid, head2type(ih));
2091 #endif
2092         if (inode)
2093                 vfs_dq_free_space_nodirty(inode, quota_bytes);
2094         return retval;
2095 }