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