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