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 *p_v_to,
  72                            const struct item_head *p_v_from)
  73 {
  74         memcpy(p_v_to, p_v_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 *p_s_1_u32, *p_s_2_u32;
 139         int n_key_length = REISERFS_SHORT_KEY_LEN;
 140
 141         p_s_1_u32 = (__u32 *) key1;
 142         p_s_2_u32 = (__u32 *) key2;
 143         for (; n_key_length--; ++p_s_1_u32, ++p_s_2_u32) {
 144                 if (le32_to_cpu(*p_s_1_u32) < le32_to_cpu(*p_s_2_u32))
 145                         return -1;
 146                 if (le32_to_cpu(*p_s_1_u32) > le32_to_cpu(*p_s_2_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  *        *p_n_pos = number of the searched element if found, else the    *
 178  *        number of the first element that is larger than p_v_key.        *
 179  **************************************************************************/
 180 /* For those not familiar with binary search: n_lbound is the leftmost item that it
 181  could be, n_rbound the rightmost item that it could be.  We examine the item
 182  halfway between n_lbound and n_rbound, and that tells us either that we can increase
 183  n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_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 *p_v_key,       /* Key to search for.                   */
 188                              const void *p_v_base,      /* First item in the array.             */
 189                              int p_n_num,       /* Number of items in the array.        */
 190                              int p_n_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, p_n_width
 196                                                    is actually the item header size not
 197                                                    the item size.                      */
 198                              int *p_n_pos       /* Number of the searched for element. */
 199     )
 200 {
 201         int n_rbound, n_lbound, n_j;
 202
 203         for (n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0)) / 2;
 204              n_lbound <= n_rbound; n_j = (n_rbound + n_lbound) / 2)
 205                 switch (comp_keys
 206                         ((struct reiserfs_key *)((char *)p_v_base +
 207                                                  n_j * p_n_width),
 208                          (struct cpu_key *)p_v_key)) {
 209                 case -1:
 210                         n_lbound = n_j + 1;
 211                         continue;
 212                 case 1:
 213                         n_rbound = n_j - 1;
 214                         continue;
 215                 case 0:
 216                         *p_n_pos = n_j;
 217                         return ITEM_FOUND;      /* Key found in the array.  */
 218                 }
 219
 220         /* bin_search did not find given key, it returns position of key,
 221            that is minimal and greater than the given one. */
 222         *p_n_pos = n_lbound;
 223         return ITEM_NOT_FOUND;
 224 }
 225
 226 #ifdef CONFIG_REISERFS_CHECK
 227 extern struct tree_balance *cur_tb;
 228 #endif
 229
 230 /* Minimal possible key. It is never in the tree. */
 231 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
 232
 233 /* Maximal possible key. It is never in the tree. */
 234 static const struct reiserfs_key MAX_KEY = {
 235         __constant_cpu_to_le32(0xffffffff),
 236         __constant_cpu_to_le32(0xffffffff),
 237         {{__constant_cpu_to_le32(0xffffffff),
 238           __constant_cpu_to_le32(0xffffffff)},}
 239 };
 240
 241 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
 242    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
 243    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
 244    case we return a special key, either MIN_KEY or MAX_KEY. */
 245 static inline const struct reiserfs_key *get_lkey(const struct treepath
 246                                                   *p_s_chk_path,
 247                                                   const struct super_block
 248                                                   *sb)
 249 {
 250         int n_position, n_path_offset = p_s_chk_path->path_length;
 251         struct buffer_head *p_s_parent;
 252
 253         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
 254                "PAP-5010: invalid offset in the path");
 255
 256         /* While not higher in path than first element. */
 257         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 258
 259                 RFALSE(!buffer_uptodate
 260                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
 261                        "PAP-5020: parent is not uptodate");
 262
 263                 /* Parent at the path is not in the tree now. */
 264                 if (!B_IS_IN_TREE
 265                     (p_s_parent =
 266                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
 267                         return &MAX_KEY;
 268                 /* Check whether position in the parent is correct. */
 269                 if ((n_position =
 270                      PATH_OFFSET_POSITION(p_s_chk_path,
 271                                           n_path_offset)) >
 272                     B_NR_ITEMS(p_s_parent))
 273                         return &MAX_KEY;
 274                 /* Check whether parent at the path really points to the child. */
 275                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
 276                     PATH_OFFSET_PBUFFER(p_s_chk_path,
 277                                         n_path_offset + 1)->b_blocknr)
 278                         return &MAX_KEY;
 279                 /* Return delimiting key if position in the parent is not equal to zero. */
 280                 if (n_position)
 281                         return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
 282         }
 283         /* Return MIN_KEY if we are in the root of the buffer tree. */
 284         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 285             b_blocknr == SB_ROOT_BLOCK(sb))
 286                 return &MIN_KEY;
 287         return &MAX_KEY;
 288 }
 289
 290 /* Get delimiting key of the buffer at the path and its right neighbor. */
 291 inline const struct reiserfs_key *get_rkey(const struct treepath *p_s_chk_path,
 292                                            const struct super_block *sb)
 293 {
 294         int n_position, n_path_offset = p_s_chk_path->path_length;
 295         struct buffer_head *p_s_parent;
 296
 297         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
 298                "PAP-5030: invalid offset in the path");
 299
 300         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 301
 302                 RFALSE(!buffer_uptodate
 303                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
 304                        "PAP-5040: parent is not uptodate");
 305
 306                 /* Parent at the path is not in the tree now. */
 307                 if (!B_IS_IN_TREE
 308                     (p_s_parent =
 309                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
 310                         return &MIN_KEY;
 311                 /* Check whether position in the parent is correct. */
 312                 if ((n_position =
 313                      PATH_OFFSET_POSITION(p_s_chk_path,
 314                                           n_path_offset)) >
 315                     B_NR_ITEMS(p_s_parent))
 316                         return &MIN_KEY;
 317                 /* Check whether parent at the path really points to the child. */
 318                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
 319                     PATH_OFFSET_PBUFFER(p_s_chk_path,
 320                                         n_path_offset + 1)->b_blocknr)
 321                         return &MIN_KEY;
 322                 /* Return delimiting key if position in the parent is not the last one. */
 323                 if (n_position != B_NR_ITEMS(p_s_parent))
 324                         return B_N_PDELIM_KEY(p_s_parent, n_position);
 325         }
 326         /* Return MAX_KEY if we are in the root of the buffer tree. */
 327         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 328             b_blocknr == SB_ROOT_BLOCK(sb))
 329                 return &MAX_KEY;
 330         return &MIN_KEY;
 331 }
 332
 333 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
 334 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
 335    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
 336    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
 337    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
 338 static inline int key_in_buffer(struct treepath *p_s_chk_path,  /* Path which should be checked.  */
 339                                 const struct cpu_key *p_s_key,  /* Key which should be checked.   */
 340                                 struct super_block *sb  /* Super block pointer.    */
 341     )
 342 {
 343
 344         RFALSE(!p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
 345                || p_s_chk_path->path_length > MAX_HEIGHT,
 346                "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
 347                p_s_key, p_s_chk_path->path_length);
 348         RFALSE(!PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
 349                "PAP-5060: device must not be NODEV");
 350
 351         if (comp_keys(get_lkey(p_s_chk_path, sb), p_s_key) == 1)
 352                 /* left delimiting key is bigger, that the key we look for */
 353                 return 0;
 354         //  if ( comp_keys(p_s_key, get_rkey(p_s_chk_path, sb)) != -1 )
 355         if (comp_keys(get_rkey(p_s_chk_path, sb), p_s_key) != 1)
 356                 /* p_s_key must be less than right delimitiing key */
 357                 return 0;
 358         return 1;
 359 }
 360
 361 int reiserfs_check_path(struct treepath *p)
 362 {
 363         RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
 364                "path not properly relsed");
 365         return 0;
 366 }
 367
 368 /* Drop the reference to each buffer in a path and restore
 369  * dirty bits clean when preparing the buffer for the log.
 370  * This version should only be called from fix_nodes() */
 371 void pathrelse_and_restore(struct super_block *sb,
 372                            struct treepath *p_s_search_path)
 373 {
 374         int n_path_offset = p_s_search_path->path_length;
 375
 376         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 377                "clm-4000: invalid path offset");
 378
 379         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
 380                 struct buffer_head *bh;
 381                 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
 382                 reiserfs_restore_prepared_buffer(sb, bh);
 383                 brelse(bh);
 384         }
 385         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 386 }
 387
 388 /* Drop the reference to each buffer in a path */
 389 void pathrelse(struct treepath *p_s_search_path)
 390 {
 391         int n_path_offset = p_s_search_path->path_length;
 392
 393         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 394                "PAP-5090: invalid path offset");
 395
 396         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
 397                 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
 398
 399         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 400 }
 401
 402 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
 403 {
 404         struct block_head *blkh;
 405         struct item_head *ih;
 406         int used_space;
 407         int prev_location;
 408         int i;
 409         int nr;
 410
 411         blkh = (struct block_head *)buf;
 412         if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
 413                 reiserfs_warning(NULL, "reiserfs-5080",
 414                                  "this should be caught earlier");
 415                 return 0;
 416         }
 417
 418         nr = blkh_nr_item(blkh);
 419         if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
 420                 /* item number is too big or too small */
 421                 reiserfs_warning(NULL, "reiserfs-5081",
 422                                  "nr_item seems wrong: %z", bh);
 423                 return 0;
 424         }
 425         ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
 426         used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
 427         if (used_space != blocksize - blkh_free_space(blkh)) {
 428                 /* free space does not match to calculated amount of use space */
 429                 reiserfs_warning(NULL, "reiserfs-5082",
 430                                  "free space seems wrong: %z", bh);
 431                 return 0;
 432         }
 433         // FIXME: it is_leaf will hit performance too much - we may have
 434         // return 1 here
 435
 436         /* check tables of item heads */
 437         ih = (struct item_head *)(buf + BLKH_SIZE);
 438         prev_location = blocksize;
 439         for (i = 0; i < nr; i++, ih++) {
 440                 if (le_ih_k_type(ih) == TYPE_ANY) {
 441                         reiserfs_warning(NULL, "reiserfs-5083",
 442                                          "wrong item type for item %h",
 443                                          ih);
 444                         return 0;
 445                 }
 446                 if (ih_location(ih) >= blocksize
 447                     || ih_location(ih) < IH_SIZE * nr) {
 448                         reiserfs_warning(NULL, "reiserfs-5084",
 449                                          "item location seems wrong: %h",
 450                                          ih);
 451                         return 0;
 452                 }
 453                 if (ih_item_len(ih) < 1
 454                     || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
 455                         reiserfs_warning(NULL, "reiserfs-5085",
 456                                          "item length seems wrong: %h",
 457                                          ih);
 458                         return 0;
 459                 }
 460                 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
 461                         reiserfs_warning(NULL, "reiserfs-5086",
 462                                          "item location seems wrong "
 463                                          "(second one): %h", ih);
 464                         return 0;
 465                 }
 466                 prev_location = ih_location(ih);
 467         }
 468
 469         // one may imagine much more checks
 470         return 1;
 471 }
 472
 473 /* returns 1 if buf looks like an internal node, 0 otherwise */
 474 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
 475 {
 476         struct block_head *blkh;
 477         int nr;
 478         int used_space;
 479
 480         blkh = (struct block_head *)buf;
 481         nr = blkh_level(blkh);
 482         if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
 483                 /* this level is not possible for internal nodes */
 484                 reiserfs_warning(NULL, "reiserfs-5087",
 485                                  "this should be caught earlier");
 486                 return 0;
 487         }
 488
 489         nr = blkh_nr_item(blkh);
 490         if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
 491                 /* for internal which is not root we might check min number of keys */
 492                 reiserfs_warning(NULL, "reiserfs-5088",
 493                                  "number of key seems wrong: %z", bh);
 494                 return 0;
 495         }
 496
 497         used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
 498         if (used_space != blocksize - blkh_free_space(blkh)) {
 499                 reiserfs_warning(NULL, "reiserfs-5089",
 500                                  "free space seems wrong: %z", bh);
 501                 return 0;
 502         }
 503         // one may imagine much more checks
 504         return 1;
 505 }
 506
 507 // make sure that bh contains formatted node of reiserfs tree of
 508 // 'level'-th level
 509 static int is_tree_node(struct buffer_head *bh, int level)
 510 {
 511         if (B_LEVEL(bh) != level) {
 512                 reiserfs_warning(NULL, "reiserfs-5090", "node level %d does "
 513                                  "not match to the expected one %d",
 514                                  B_LEVEL(bh), level);
 515                 return 0;
 516         }
 517         if (level == DISK_LEAF_NODE_LEVEL)
 518                 return is_leaf(bh->b_data, bh->b_size, bh);
 519
 520         return is_internal(bh->b_data, bh->b_size, bh);
 521 }
 522
 523 #define SEARCH_BY_KEY_READA 16
 524
 525 /* The function is NOT SCHEDULE-SAFE! */
 526 static void search_by_key_reada(struct super_block *s,
 527                                 struct buffer_head **bh,
 528                                 b_blocknr_t *b, int num)
 529 {
 530         int i, j;
 531
 532         for (i = 0; i < num; i++) {
 533                 bh[i] = sb_getblk(s, b[i]);
 534         }
 535         for (j = 0; j < i; j++) {
 536                 /*
 537                  * note, this needs attention if we are getting rid of the BKL
 538                  * you have to make sure the prepared bit isn't set on this buffer
 539                  */
 540                 if (!buffer_uptodate(bh[j]))
 541                         ll_rw_block(READA, 1, bh + j);
 542                 brelse(bh[j]);
 543         }
 544 }
 545
 546 /**************************************************************************
 547  * Algorithm   SearchByKey                                                *
 548  *             look for item in the Disk S+Tree by its key                *
 549  * Input:  sb   -  super block                                            *
 550  *         p_s_key  - pointer to the key to search                        *
 551  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
 552  *         p_s_search_path - path from the root to the needed leaf        *
 553  **************************************************************************/
 554
 555 /* This function fills up the path from the root to the leaf as it
 556    descends the tree looking for the key.  It uses reiserfs_bread to
 557    try to find buffers in the cache given their block number.  If it
 558    does not find them in the cache it reads them from disk.  For each
 559    node search_by_key finds using reiserfs_bread it then uses
 560    bin_search to look through that node.  bin_search will find the
 561    position of the block_number of the next node if it is looking
 562    through an internal node.  If it is looking through a leaf node
 563    bin_search will find the position of the item which has key either
 564    equal to given key, or which is the maximal key less than the given
 565    key.  search_by_key returns a path that must be checked for the
 566    correctness of the top of the path but need not be checked for the
 567    correctness of the bottom of the path */
 568 /* The function is NOT SCHEDULE-SAFE! */
 569 int search_by_key(struct super_block *sb, const struct cpu_key *p_s_key,        /* Key to search. */
 570                   struct treepath *p_s_search_path,/* This structure was
 571                                                    allocated and initialized
 572                                                    by the calling
 573                                                    function. It is filled up
 574                                                    by this function.  */
 575                   int n_stop_level      /* How far down the tree to search. To
 576                                            stop at leaf level - set to
 577                                            DISK_LEAF_NODE_LEVEL */
 578     )
 579 {
 580         b_blocknr_t n_block_number;
 581         int expected_level;
 582         struct buffer_head *bh;
 583         struct path_element *p_s_last_element;
 584         int n_node_level, n_retval;
 585         int right_neighbor_of_leaf_node;
 586         int fs_gen;
 587         struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
 588         b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
 589         int reada_count = 0;
 590
 591 #ifdef CONFIG_REISERFS_CHECK
 592         int n_repeat_counter = 0;
 593 #endif
 594
 595         PROC_INFO_INC(sb, search_by_key);
 596
 597         /* As we add each node to a path we increase its count.  This means that
 598            we must be careful to release all nodes in a path before we either
 599            discard the path struct or re-use the path struct, as we do here. */
 600
 601         pathrelse(p_s_search_path);
 602
 603         right_neighbor_of_leaf_node = 0;
 604
 605         /* With each iteration of this loop we search through the items in the
 606            current node, and calculate the next current node(next path element)
 607            for the next iteration of this loop.. */
 608         n_block_number = SB_ROOT_BLOCK(sb);
 609         expected_level = -1;
 610         while (1) {
 611
 612 #ifdef CONFIG_REISERFS_CHECK
 613                 if (!(++n_repeat_counter % 50000))
 614                         reiserfs_warning(sb, "PAP-5100",
 615                                          "%s: there were %d iterations of "
 616                                          "while loop looking for key %K",
 617                                          current->comm, n_repeat_counter,
 618                                          p_s_key);
 619 #endif
 620
 621                 /* prep path to have another element added to it. */
 622                 p_s_last_element =
 623                     PATH_OFFSET_PELEMENT(p_s_search_path,
 624                                          ++p_s_search_path->path_length);
 625                 fs_gen = get_generation(sb);
 626
 627                 /* Read the next tree node, and set the last element in the path to
 628                    have a pointer to it. */
 629                 if ((bh = p_s_last_element->pe_buffer =
 630                      sb_getblk(sb, n_block_number))) {
 631                         if (!buffer_uptodate(bh) && reada_count > 1)
 632                                 search_by_key_reada(sb, reada_bh,
 633                                                     reada_blocks, reada_count);
 634                         ll_rw_block(READ, 1, &bh);
 635                         wait_on_buffer(bh);
 636                         if (!buffer_uptodate(bh))
 637                                 goto io_error;
 638                 } else {
 639                       io_error:
 640                         p_s_search_path->path_length--;
 641                         pathrelse(p_s_search_path);
 642                         return IO_ERROR;
 643                 }
 644                 reada_count = 0;
 645                 if (expected_level == -1)
 646                         expected_level = SB_TREE_HEIGHT(sb);
 647                 expected_level--;
 648
 649                 /* It is possible that schedule occurred. We must check whether the key
 650                    to search is still in the tree rooted from the current buffer. If
 651                    not then repeat search from the root. */
 652                 if (fs_changed(fs_gen, sb) &&
 653                     (!B_IS_IN_TREE(bh) ||
 654                      B_LEVEL(bh) != expected_level ||
 655                      !key_in_buffer(p_s_search_path, p_s_key, sb))) {
 656                         PROC_INFO_INC(sb, search_by_key_fs_changed);
 657                         PROC_INFO_INC(sb, search_by_key_restarted);
 658                         PROC_INFO_INC(sb,
 659                                       sbk_restarted[expected_level - 1]);
 660                         pathrelse(p_s_search_path);
 661
 662                         /* Get the root block number so that we can repeat the search
 663                            starting from the root. */
 664                         n_block_number = SB_ROOT_BLOCK(sb);
 665                         expected_level = -1;
 666                         right_neighbor_of_leaf_node = 0;
 667
 668                         /* repeat search from the root */
 669                         continue;
 670                 }
 671
 672                 /* only check that the key is in the buffer if p_s_key is not
 673                    equal to the MAX_KEY. Latter case is only possible in
 674                    "finish_unfinished()" processing during mount. */
 675                 RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
 676                        !key_in_buffer(p_s_search_path, p_s_key, sb),
 677                        "PAP-5130: key is not in the buffer");
 678 #ifdef CONFIG_REISERFS_CHECK
 679                 if (cur_tb) {
 680                         print_cur_tb("5140");
 681                         reiserfs_panic(sb, "PAP-5140",
 682                                        "schedule occurred in do_balance!");
 683                 }
 684 #endif
 685
 686                 // make sure, that the node contents look like a node of
 687                 // certain level
 688                 if (!is_tree_node(bh, expected_level)) {
 689                         reiserfs_error(sb, "vs-5150",
 690                                        "invalid format found in block %ld. "
 691                                        "Fsck?", bh->b_blocknr);
 692                         pathrelse(p_s_search_path);
 693                         return IO_ERROR;
 694                 }
 695
 696                 /* ok, we have acquired next formatted node in the tree */
 697                 n_node_level = B_LEVEL(bh);
 698
 699                 PROC_INFO_BH_STAT(sb, bh, n_node_level - 1);
 700
 701                 RFALSE(n_node_level < n_stop_level,
 702                        "vs-5152: tree level (%d) is less than stop level (%d)",
 703                        n_node_level, n_stop_level);
 704
 705                 n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(bh, 0),
 706                                       B_NR_ITEMS(bh),
 707                                       (n_node_level ==
 708                                        DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
 709                                       KEY_SIZE,
 710                                       &(p_s_last_element->pe_position));
 711                 if (n_node_level == n_stop_level) {
 712                         return n_retval;
 713                 }
 714
 715                 /* we are not in the stop level */
 716                 if (n_retval == ITEM_FOUND)
 717                         /* item has been found, so we choose the pointer which is to the right of the found one */
 718                         p_s_last_element->pe_position++;
 719
 720                 /* if item was not found we choose the position which is to
 721                    the left of the found item. This requires no code,
 722                    bin_search did it already. */
 723
 724                 /* So we have chosen a position in the current node which is
 725                    an internal node.  Now we calculate child block number by
 726                    position in the node. */
 727                 n_block_number =
 728                     B_N_CHILD_NUM(bh, p_s_last_element->pe_position);
 729
 730                 /* if we are going to read leaf nodes, try for read ahead as well */
 731                 if ((p_s_search_path->reada & PATH_READA) &&
 732                     n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
 733                         int pos = p_s_last_element->pe_position;
 734                         int limit = B_NR_ITEMS(bh);
 735                         struct reiserfs_key *le_key;
 736
 737                         if (p_s_search_path->reada & PATH_READA_BACK)
 738                                 limit = 0;
 739                         while (reada_count < SEARCH_BY_KEY_READA) {
 740                                 if (pos == limit)
 741                                         break;
 742                                 reada_blocks[reada_count++] =
 743                                     B_N_CHILD_NUM(bh, pos);
 744                                 if (p_s_search_path->reada & PATH_READA_BACK)
 745                                         pos--;
 746                                 else
 747                                         pos++;
 748
 749                                 /*
 750                                  * check to make sure we're in the same object
 751                                  */
 752                                 le_key = B_N_PDELIM_KEY(bh, pos);
 753                                 if (le32_to_cpu(le_key->k_objectid) !=
 754                                     p_s_key->on_disk_key.k_objectid) {
 755                                         break;
 756                                 }
 757                         }
 758                 }
 759         }
 760 }
 761
 762 /* Form the path to an item and position in this item which contains
 763    file byte defined by p_s_key. If there is no such item
 764    corresponding to the key, we point the path to the item with
 765    maximal key less than p_s_key, and *p_n_pos_in_item is set to one
 766    past the last entry/byte in the item.  If searching for entry in a
 767    directory item, and it is not found, *p_n_pos_in_item is set to one
 768    entry more than the entry with maximal key which is less than the
 769    sought key.
 770
 771    Note that if there is no entry in this same node which is one more,
 772    then we point to an imaginary entry.  for direct items, the
 773    position is in units of bytes, for indirect items the position is
 774    in units of blocknr entries, for directory items the position is in
 775    units of directory entries.  */
 776
 777 /* The function is NOT SCHEDULE-SAFE! */
 778 int search_for_position_by_key(struct super_block *sb,  /* Pointer to the super block.          */
 779                                const struct cpu_key *p_cpu_key, /* Key to search (cpu variable)         */
 780                                struct treepath *p_s_search_path /* Filled up by this function.          */
 781     )
 782 {
 783         struct item_head *p_le_ih;      /* pointer to on-disk structure */
 784         int n_blk_size;
 785         loff_t item_offset, offset;
 786         struct reiserfs_dir_entry de;
 787         int retval;
 788
 789         /* If searching for directory entry. */
 790         if (is_direntry_cpu_key(p_cpu_key))
 791                 return search_by_entry_key(sb, p_cpu_key, p_s_search_path,
 792                                            &de);
 793
 794         /* If not searching for directory entry. */
 795
 796         /* If item is found. */
 797         retval = search_item(sb, p_cpu_key, p_s_search_path);
 798         if (retval == IO_ERROR)
 799                 return retval;
 800         if (retval == ITEM_FOUND) {
 801
 802                 RFALSE(!ih_item_len
 803                        (B_N_PITEM_HEAD
 804                         (PATH_PLAST_BUFFER(p_s_search_path),
 805                          PATH_LAST_POSITION(p_s_search_path))),
 806                        "PAP-5165: item length equals zero");
 807
 808                 pos_in_item(p_s_search_path) = 0;
 809                 return POSITION_FOUND;
 810         }
 811
 812         RFALSE(!PATH_LAST_POSITION(p_s_search_path),
 813                "PAP-5170: position equals zero");
 814
 815         /* Item is not found. Set path to the previous item. */
 816         p_le_ih =
 817             B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
 818                            --PATH_LAST_POSITION(p_s_search_path));
 819         n_blk_size = sb->s_blocksize;
 820
 821         if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
 822                 return FILE_NOT_FOUND;
 823         }
 824         // FIXME: quite ugly this far
 825
 826         item_offset = le_ih_k_offset(p_le_ih);
 827         offset = cpu_key_k_offset(p_cpu_key);
 828
 829         /* Needed byte is contained in the item pointed to by the path. */
 830         if (item_offset <= offset &&
 831             item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
 832                 pos_in_item(p_s_search_path) = offset - item_offset;
 833                 if (is_indirect_le_ih(p_le_ih)) {
 834                         pos_in_item(p_s_search_path) /= n_blk_size;
 835                 }
 836                 return POSITION_FOUND;
 837         }
 838
 839         /* Needed byte is not contained in the item pointed to by the
 840            path. Set pos_in_item out of the item. */
 841         if (is_indirect_le_ih(p_le_ih))
 842                 pos_in_item(p_s_search_path) =
 843                     ih_item_len(p_le_ih) / UNFM_P_SIZE;
 844         else
 845                 pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
 846
 847         return POSITION_NOT_FOUND;
 848 }
 849
 850 /* Compare given item and item pointed to by the path. */
 851 int comp_items(const struct item_head *stored_ih, const struct treepath *p_s_path)
 852 {
 853         struct buffer_head *bh = PATH_PLAST_BUFFER(p_s_path);
 854         struct item_head *ih;
 855
 856         /* Last buffer at the path is not in the tree. */
 857         if (!B_IS_IN_TREE(bh))
 858                 return 1;
 859
 860         /* Last path position is invalid. */
 861         if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(bh))
 862                 return 1;
 863
 864         /* we need only to know, whether it is the same item */
 865         ih = get_ih(p_s_path);
 866         return memcmp(stored_ih, ih, IH_SIZE);
 867 }
 868
 869 /* unformatted nodes are not logged anymore, ever.  This is safe
 870 ** now
 871 */
 872 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
 873
 874 // block can not be forgotten as it is in I/O or held by someone
 875 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
 876
 877 // prepare for delete or cut of direct item
 878 static inline int prepare_for_direct_item(struct treepath *path,
 879                                           struct item_head *le_ih,
 880                                           struct inode *inode,
 881                                           loff_t new_file_length, int *cut_size)
 882 {
 883         loff_t round_len;
 884
 885         if (new_file_length == max_reiserfs_offset(inode)) {
 886                 /* item has to be deleted */
 887                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 888                 return M_DELETE;
 889         }
 890         // new file gets truncated
 891         if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
 892                 //
 893                 round_len = ROUND_UP(new_file_length);
 894                 /* this was n_new_file_length < le_ih ... */
 895                 if (round_len < le_ih_k_offset(le_ih)) {
 896                         *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 897                         return M_DELETE;        /* Delete this item. */
 898                 }
 899                 /* Calculate first position and size for cutting from item. */
 900                 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
 901                 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
 902
 903                 return M_CUT;   /* Cut from this item. */
 904         }
 905
 906         // old file: items may have any length
 907
 908         if (new_file_length < le_ih_k_offset(le_ih)) {
 909                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 910                 return M_DELETE;        /* Delete this item. */
 911         }
 912         /* Calculate first position and size for cutting from item. */
 913         *cut_size = -(ih_item_len(le_ih) -
 914                       (pos_in_item(path) =
 915                        new_file_length + 1 - le_ih_k_offset(le_ih)));
 916         return M_CUT;           /* Cut from this item. */
 917 }
 918
 919 static inline int prepare_for_direntry_item(struct treepath *path,
 920                                             struct item_head *le_ih,
 921                                             struct inode *inode,
 922                                             loff_t new_file_length,
 923                                             int *cut_size)
 924 {
 925         if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
 926             new_file_length == max_reiserfs_offset(inode)) {
 927                 RFALSE(ih_entry_count(le_ih) != 2,
 928                        "PAP-5220: incorrect empty directory item (%h)", le_ih);
 929                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 930                 return M_DELETE;        /* Delete the directory item containing "." and ".." entry. */
 931         }
 932
 933         if (ih_entry_count(le_ih) == 1) {
 934                 /* Delete the directory item such as there is one record only
 935                    in this item */
 936                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 937                 return M_DELETE;
 938         }
 939
 940         /* Cut one record from the directory item. */
 941         *cut_size =
 942             -(DEH_SIZE +
 943               entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
 944         return M_CUT;
 945 }
 946
 947 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
 948
 949 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
 950     If the path points to an indirect item, remove some number of its unformatted nodes.
 951     In case of file truncate calculate whether this item must be deleted/truncated or last
 952     unformatted node of this item will be converted to a direct item.
 953     This function returns a determination of what balance mode the calling function should employ. */
 954 static char prepare_for_delete_or_cut(struct reiserfs_transaction_handle *th, struct inode *inode, struct treepath *p_s_path, const struct cpu_key *p_s_item_key, int *p_n_removed,     /* Number of unformatted nodes which were removed
 955                                                                                                                                                                                    from end of the file. */
 956                                       int *p_n_cut_size, unsigned long long n_new_file_length   /* MAX_KEY_OFFSET in case of delete. */
 957     )
 958 {
 959         struct super_block *sb = inode->i_sb;
 960         struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
 961         struct buffer_head *bh = PATH_PLAST_BUFFER(p_s_path);
 962
 963         BUG_ON(!th->t_trans_id);
 964
 965         /* Stat_data item. */
 966         if (is_statdata_le_ih(p_le_ih)) {
 967
 968                 RFALSE(n_new_file_length != max_reiserfs_offset(inode),
 969                        "PAP-5210: mode must be M_DELETE");
 970
 971                 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
 972                 return M_DELETE;
 973         }
 974
 975         /* Directory item. */
 976         if (is_direntry_le_ih(p_le_ih))
 977                 return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
 978                                                  n_new_file_length,
 979                                                  p_n_cut_size);
 980
 981         /* Direct item. */
 982         if (is_direct_le_ih(p_le_ih))
 983                 return prepare_for_direct_item(p_s_path, p_le_ih, inode,
 984                                                n_new_file_length, p_n_cut_size);
 985
 986         /* Case of an indirect item. */
 987         {
 988             int blk_size = sb->s_blocksize;
 989             struct item_head s_ih;
 990             int need_re_search;
 991             int delete = 0;
 992             int result = M_CUT;
 993             int pos = 0;
 994
 995             if ( n_new_file_length == max_reiserfs_offset (inode) ) {
 996                 /* prepare_for_delete_or_cut() is called by
 997                  * reiserfs_delete_item() */
 998                 n_new_file_length = 0;
 999                 delete = 1;
1000             }
1001
1002             do {
1003                 need_re_search = 0;
1004                 *p_n_cut_size = 0;
1005                 bh = PATH_PLAST_BUFFER(p_s_path);
1006                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1007                 pos = I_UNFM_NUM(&s_ih);
1008
1009                 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > n_new_file_length) {
1010                     __le32 *unfm;
1011                     __u32 block;
1012
1013                     /* Each unformatted block deletion may involve one additional
1014                      * bitmap block into the transaction, thereby the initial
1015                      * journal space reservation might not be enough. */
1016                     if (!delete && (*p_n_cut_size) != 0 &&
1017                         reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1018                         break;
1019                     }
1020
1021                     unfm = (__le32 *)B_I_PITEM(bh, &s_ih) + pos - 1;
1022                     block = get_block_num(unfm, 0);
1023
1024                     if (block != 0) {
1025                         reiserfs_prepare_for_journal(sb, bh, 1);
1026                         put_block_num(unfm, 0, 0);
1027                         journal_mark_dirty(th, sb, bh);
1028                         reiserfs_free_block(th, inode, block, 1);
1029                     }
1030
1031                     cond_resched();
1032
1033                     if (item_moved (&s_ih, p_s_path))  {
1034                         need_re_search = 1;
1035                         break;
1036                     }
1037
1038                     pos --;
1039                     (*p_n_removed) ++;
1040                     (*p_n_cut_size) -= UNFM_P_SIZE;
1041
1042                     if (pos == 0) {
1043                         (*p_n_cut_size) -= IH_SIZE;
1044                         result = M_DELETE;
1045                         break;
1046                     }
1047                 }
1048                 /* a trick.  If the buffer has been logged, this will do nothing.  If
1049                 ** we've broken the loop without logging it, it will restore the
1050                 ** buffer */
1051                 reiserfs_restore_prepared_buffer(sb, bh);
1052             } while (need_re_search &&
1053                      search_for_position_by_key(sb, p_s_item_key, p_s_path) == POSITION_FOUND);
1054             pos_in_item(p_s_path) = pos * UNFM_P_SIZE;
1055
1056             if (*p_n_cut_size == 0) {
1057                 /* Nothing were cut. maybe convert last unformatted node to the
1058                  * direct item? */
1059                 result = M_CONVERT;
1060             }
1061             return result;
1062         }
1063 }
1064
1065 /* Calculate number of bytes which will be deleted or cut during balance */
1066 static int calc_deleted_bytes_number(struct tree_balance *tb, char c_mode)
1067 {
1068         int n_del_size;
1069         struct item_head *p_le_ih = PATH_PITEM_HEAD(tb->tb_path);
1070
1071         if (is_statdata_le_ih(p_le_ih))
1072                 return 0;
1073
1074         n_del_size =
1075             (c_mode ==
1076              M_DELETE) ? ih_item_len(p_le_ih) : -tb->insert_size[0];
1077         if (is_direntry_le_ih(p_le_ih)) {
1078                 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1079                 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1080                 // empty size.  ick. FIXME, is this right?
1081                 //
1082                 return n_del_size;
1083         }
1084
1085         if (is_indirect_le_ih(p_le_ih))
1086                 n_del_size = (n_del_size / UNFM_P_SIZE) *
1087                                 (PATH_PLAST_BUFFER(tb->tb_path)->b_size);
1088         return n_del_size;
1089 }
1090
1091 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1092                            struct tree_balance *tb,
1093                            struct super_block *sb,
1094                            struct treepath *p_s_path, int n_size)
1095 {
1096
1097         BUG_ON(!th->t_trans_id);
1098
1099         memset(tb, '\0', sizeof(struct tree_balance));
1100         tb->transaction_handle = th;
1101         tb->tb_sb = sb;
1102         tb->tb_path = p_s_path;
1103         PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1104         PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1105         tb->insert_size[0] = n_size;
1106 }
1107
1108 void padd_item(char *item, int total_length, int length)
1109 {
1110         int i;
1111
1112         for (i = total_length; i > length;)
1113                 item[--i] = 0;
1114 }
1115
1116 #ifdef REISERQUOTA_DEBUG
1117 char key2type(struct reiserfs_key *ih)
1118 {
1119         if (is_direntry_le_key(2, ih))
1120                 return 'd';
1121         if (is_direct_le_key(2, ih))
1122                 return 'D';
1123         if (is_indirect_le_key(2, ih))
1124                 return 'i';
1125         if (is_statdata_le_key(2, ih))
1126                 return 's';
1127         return 'u';
1128 }
1129
1130 char head2type(struct item_head *ih)
1131 {
1132         if (is_direntry_le_ih(ih))
1133                 return 'd';
1134         if (is_direct_le_ih(ih))
1135                 return 'D';
1136         if (is_indirect_le_ih(ih))
1137                 return 'i';
1138         if (is_statdata_le_ih(ih))
1139                 return 's';
1140         return 'u';
1141 }
1142 #endif
1143
1144 /* Delete object item. */
1145 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the deleted item. */
1146                          const struct cpu_key *p_s_item_key,    /* Key to search for the deleted item.  */
1147                          struct inode *inode,   /* inode is here just to update
1148                                                  * i_blocks and quotas */
1149                          struct buffer_head *p_s_un_bh)
1150 {                               /* NULL or unformatted node pointer.    */
1151         struct super_block *sb = inode->i_sb;
1152         struct tree_balance s_del_balance;
1153         struct item_head s_ih;
1154         struct item_head *q_ih;
1155         int quota_cut_bytes;
1156         int n_ret_value, n_del_size, n_removed;
1157
1158 #ifdef CONFIG_REISERFS_CHECK
1159         char c_mode;
1160         int n_iter = 0;
1161 #endif
1162
1163         BUG_ON(!th->t_trans_id);
1164
1165         init_tb_struct(th, &s_del_balance, sb, p_s_path,
1166                        0 /*size is unknown */ );
1167
1168         while (1) {
1169                 n_removed = 0;
1170
1171 #ifdef CONFIG_REISERFS_CHECK
1172                 n_iter++;
1173                 c_mode =
1174 #endif
1175                     prepare_for_delete_or_cut(th, inode, p_s_path,
1176                                               p_s_item_key, &n_removed,
1177                                               &n_del_size,
1178                                               max_reiserfs_offset(inode));
1179
1180                 RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1181
1182                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1183                 s_del_balance.insert_size[0] = n_del_size;
1184
1185                 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1186                 if (n_ret_value != REPEAT_SEARCH)
1187                         break;
1188
1189                 PROC_INFO_INC(sb, delete_item_restarted);
1190
1191                 // file system changed, repeat search
1192                 n_ret_value =
1193                     search_for_position_by_key(sb, p_s_item_key, p_s_path);
1194                 if (n_ret_value == IO_ERROR)
1195                         break;
1196                 if (n_ret_value == FILE_NOT_FOUND) {
1197                         reiserfs_warning(sb, "vs-5340",
1198                                          "no items of the file %K found",
1199                                          p_s_item_key);
1200                         break;
1201                 }
1202         }                       /* while (1) */
1203
1204         if (n_ret_value != CARRY_ON) {
1205                 unfix_nodes(&s_del_balance);
1206                 return 0;
1207         }
1208         // reiserfs_delete_item returns item length when success
1209         n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1210         q_ih = get_ih(p_s_path);
1211         quota_cut_bytes = ih_item_len(q_ih);
1212
1213         /* hack so the quota code doesn't have to guess if the file
1214          ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1215          ** We test the offset because the tail might have been
1216          ** split into multiple items, and we only want to decrement for
1217          ** the unfm node once
1218          */
1219         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(q_ih)) {
1220                 if ((le_ih_k_offset(q_ih) & (sb->s_blocksize - 1)) == 1) {
1221                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1222                 } else {
1223                         quota_cut_bytes = 0;
1224                 }
1225         }
1226
1227         if (p_s_un_bh) {
1228                 int off;
1229                 char *data;
1230
1231                 /* We are in direct2indirect conversion, so move tail contents
1232                    to the unformatted node */
1233                 /* note, we do the copy before preparing the buffer because we
1234                  ** don't care about the contents of the unformatted node yet.
1235                  ** the only thing we really care about is the direct item's data
1236                  ** is in the unformatted node.
1237                  **
1238                  ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1239                  ** the unformatted node, which might schedule, meaning we'd have to
1240                  ** loop all the way back up to the start of the while loop.
1241                  **
1242                  ** The unformatted node must be dirtied later on.  We can't be
1243                  ** sure here if the entire tail has been deleted yet.
1244                  **
1245                  ** p_s_un_bh is from the page cache (all unformatted nodes are
1246                  ** from the page cache) and might be a highmem page.  So, we
1247                  ** can't use p_s_un_bh->b_data.
1248                  ** -clm
1249                  */
1250
1251                 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1252                 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1253                 memcpy(data + off,
1254                        B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1255                        n_ret_value);
1256                 kunmap_atomic(data, KM_USER0);
1257         }
1258         /* Perform balancing after all resources have been collected at once. */
1259         do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1260
1261 #ifdef REISERQUOTA_DEBUG
1262         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1263                        "reiserquota delete_item(): freeing %u, id=%u type=%c",
1264                        quota_cut_bytes, inode->i_uid, head2type(&s_ih));
1265 #endif
1266         DQUOT_FREE_SPACE_NODIRTY(inode, quota_cut_bytes);
1267
1268         /* Return deleted body length */
1269         return n_ret_value;
1270 }
1271
1272 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1273
1274  deletion of the body of the object is performed by iput(), with the
1275  result that if multiple processes are operating on a file, the
1276  deletion of the body of the file is deferred until the last process
1277  that has an open inode performs its iput().
1278
1279  writes and truncates are protected from collisions by use of
1280  semaphores.
1281
1282  creates, linking, and mknod are protected from collisions with other
1283  processes by making the reiserfs_add_entry() the last step in the
1284  creation, and then rolling back all changes if there was a collision.
1285  - Hans
1286 */
1287
1288 /* this deletes item which never gets split */
1289 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1290                                 struct inode *inode, struct reiserfs_key *key)
1291 {
1292         struct tree_balance tb;
1293         INITIALIZE_PATH(path);
1294         int item_len = 0;
1295         int tb_init = 0;
1296         struct cpu_key cpu_key;
1297         int retval;
1298         int quota_cut_bytes = 0;
1299
1300         BUG_ON(!th->t_trans_id);
1301
1302         le_key2cpu_key(&cpu_key, key);
1303
1304         while (1) {
1305                 retval = search_item(th->t_super, &cpu_key, &path);
1306                 if (retval == IO_ERROR) {
1307                         reiserfs_error(th->t_super, "vs-5350",
1308                                        "i/o failure occurred trying "
1309                                        "to delete %K", &cpu_key);
1310                         break;
1311                 }
1312                 if (retval != ITEM_FOUND) {
1313                         pathrelse(&path);
1314                         // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1315                         if (!
1316                             ((unsigned long long)
1317                              GET_HASH_VALUE(le_key_k_offset
1318                                             (le_key_version(key), key)) == 0
1319                              && (unsigned long long)
1320                              GET_GENERATION_NUMBER(le_key_k_offset
1321                                                    (le_key_version(key),
1322                                                     key)) == 1))
1323                                 reiserfs_warning(th->t_super, "vs-5355",
1324                                                  "%k not found", key);
1325                         break;
1326                 }
1327                 if (!tb_init) {
1328                         tb_init = 1;
1329                         item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1330                         init_tb_struct(th, &tb, th->t_super, &path,
1331                                        -(IH_SIZE + item_len));
1332                 }
1333                 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1334
1335                 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1336                 if (retval == REPEAT_SEARCH) {
1337                         PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1338                         continue;
1339                 }
1340
1341                 if (retval == CARRY_ON) {
1342                         do_balance(&tb, NULL, NULL, M_DELETE);
1343                         if (inode) {    /* Should we count quota for item? (we don't count quotas for save-links) */
1344 #ifdef REISERQUOTA_DEBUG
1345                                 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1346                                                "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1347                                                quota_cut_bytes, inode->i_uid,
1348                                                key2type(key));
1349 #endif
1350                                 DQUOT_FREE_SPACE_NODIRTY(inode,
1351                                                          quota_cut_bytes);
1352                         }
1353                         break;
1354                 }
1355                 // IO_ERROR, NO_DISK_SPACE, etc
1356                 reiserfs_warning(th->t_super, "vs-5360",
1357                                  "could not delete %K due to fix_nodes failure",
1358                                  &cpu_key);
1359                 unfix_nodes(&tb);
1360                 break;
1361         }
1362
1363         reiserfs_check_path(&path);
1364 }
1365
1366 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1367                            struct inode *inode)
1368 {
1369         int err;
1370         inode->i_size = 0;
1371         BUG_ON(!th->t_trans_id);
1372
1373         /* for directory this deletes item containing "." and ".." */
1374         err =
1375             reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1376         if (err)
1377                 return err;
1378
1379 #if defined( USE_INODE_GENERATION_COUNTER )
1380         if (!old_format_only(th->t_super)) {
1381                 __le32 *inode_generation;
1382
1383                 inode_generation =
1384                     &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1385                 le32_add_cpu(inode_generation, 1);
1386         }
1387 /* USE_INODE_GENERATION_COUNTER */
1388 #endif
1389         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1390
1391         return err;
1392 }
1393
1394 static void unmap_buffers(struct page *page, loff_t pos)
1395 {
1396         struct buffer_head *bh;
1397         struct buffer_head *head;
1398         struct buffer_head *next;
1399         unsigned long tail_index;
1400         unsigned long cur_index;
1401
1402         if (page) {
1403                 if (page_has_buffers(page)) {
1404                         tail_index = pos & (PAGE_CACHE_SIZE - 1);
1405                         cur_index = 0;
1406                         head = page_buffers(page);
1407                         bh = head;
1408                         do {
1409                                 next = bh->b_this_page;
1410
1411                                 /* we want to unmap the buffers that contain the tail, and
1412                                  ** all the buffers after it (since the tail must be at the
1413                                  ** end of the file).  We don't want to unmap file data
1414                                  ** before the tail, since it might be dirty and waiting to
1415                                  ** reach disk
1416                                  */
1417                                 cur_index += bh->b_size;
1418                                 if (cur_index > tail_index) {
1419                                         reiserfs_unmap_buffer(bh);
1420                                 }
1421                                 bh = next;
1422                         } while (bh != head);
1423                 }
1424         }
1425 }
1426
1427 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1428                                     struct inode *inode,
1429                                     struct page *page,
1430                                     struct treepath *p_s_path,
1431                                     const struct cpu_key *p_s_item_key,
1432                                     loff_t n_new_file_size, char *p_c_mode)
1433 {
1434         struct super_block *sb = inode->i_sb;
1435         int n_block_size = sb->s_blocksize;
1436         int cut_bytes;
1437         BUG_ON(!th->t_trans_id);
1438         BUG_ON(n_new_file_size != inode->i_size);
1439
1440         /* the page being sent in could be NULL if there was an i/o error
1441          ** reading in the last block.  The user will hit problems trying to
1442          ** read the file, but for now we just skip the indirect2direct
1443          */
1444         if (atomic_read(&inode->i_count) > 1 ||
1445             !tail_has_to_be_packed(inode) ||
1446             !page || (REISERFS_I(inode)->i_flags & i_nopack_mask)) {
1447                 /* leave tail in an unformatted node */
1448                 *p_c_mode = M_SKIP_BALANCING;
1449                 cut_bytes =
1450                     n_block_size - (n_new_file_size & (n_block_size - 1));
1451                 pathrelse(p_s_path);
1452                 return cut_bytes;
1453         }
1454         /* Permorm the conversion to a direct_item. */
1455         /* return indirect_to_direct(inode, p_s_path, p_s_item_key,
1456                                   n_new_file_size, p_c_mode); */
1457         return indirect2direct(th, inode, page, p_s_path, p_s_item_key,
1458                                n_new_file_size, p_c_mode);
1459 }
1460
1461 /* we did indirect_to_direct conversion. And we have inserted direct
1462    item successesfully, but there were no disk space to cut unfm
1463    pointer being converted. Therefore we have to delete inserted
1464    direct item(s) */
1465 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1466                                          struct inode *inode, struct treepath *path)
1467 {
1468         struct cpu_key tail_key;
1469         int tail_len;
1470         int removed;
1471         BUG_ON(!th->t_trans_id);
1472
1473         make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);      // !!!!
1474         tail_key.key_length = 4;
1475
1476         tail_len =
1477             (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1478         while (tail_len) {
1479                 /* look for the last byte of the tail */
1480                 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1481                     POSITION_NOT_FOUND)
1482                         reiserfs_panic(inode->i_sb, "vs-5615",
1483                                        "found invalid item");
1484                 RFALSE(path->pos_in_item !=
1485                        ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1486                        "vs-5616: appended bytes found");
1487                 PATH_LAST_POSITION(path)--;
1488
1489                 removed =
1490                     reiserfs_delete_item(th, path, &tail_key, inode,
1491                                          NULL /*unbh not needed */ );
1492                 RFALSE(removed <= 0
1493                        || removed > tail_len,
1494                        "vs-5617: there was tail %d bytes, removed item length %d bytes",
1495                        tail_len, removed);
1496                 tail_len -= removed;
1497                 set_cpu_key_k_offset(&tail_key,
1498                                      cpu_key_k_offset(&tail_key) - removed);
1499         }
1500         reiserfs_warning(inode->i_sb, "reiserfs-5091", "indirect_to_direct "
1501                          "conversion has been rolled back due to "
1502                          "lack of disk space");
1503         //mark_file_without_tail (inode);
1504         mark_inode_dirty(inode);
1505 }
1506
1507 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1508 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1509                            struct treepath *p_s_path,
1510                            struct cpu_key *p_s_item_key,
1511                            struct inode *inode,
1512                            struct page *page, loff_t n_new_file_size)
1513 {
1514         struct super_block *sb = inode->i_sb;
1515         /* Every function which is going to call do_balance must first
1516            create a tree_balance structure.  Then it must fill up this
1517            structure by using the init_tb_struct and fix_nodes functions.
1518            After that we can make tree balancing. */
1519         struct tree_balance s_cut_balance;
1520         struct item_head *p_le_ih;
1521         int n_cut_size = 0,     /* Amount to be cut. */
1522             n_ret_value = CARRY_ON, n_removed = 0,      /* Number of the removed unformatted nodes. */
1523             n_is_inode_locked = 0;
1524         char c_mode;            /* Mode of the balance. */
1525         int retval2 = -1;
1526         int quota_cut_bytes;
1527         loff_t tail_pos = 0;
1528
1529         BUG_ON(!th->t_trans_id);
1530
1531         init_tb_struct(th, &s_cut_balance, inode->i_sb, p_s_path,
1532                        n_cut_size);
1533
1534         /* Repeat this loop until we either cut the item without needing
1535            to balance, or we fix_nodes without schedule occurring */
1536         while (1) {
1537                 /* Determine the balance mode, position of the first byte to
1538                    be cut, and size to be cut.  In case of the indirect item
1539                    free unformatted nodes which are pointed to by the cut
1540                    pointers. */
1541
1542                 c_mode =
1543                     prepare_for_delete_or_cut(th, inode, p_s_path,
1544                                               p_s_item_key, &n_removed,
1545                                               &n_cut_size, n_new_file_size);
1546                 if (c_mode == M_CONVERT) {
1547                         /* convert last unformatted node to direct item or leave
1548                            tail in the unformatted node */
1549                         RFALSE(n_ret_value != CARRY_ON,
1550                                "PAP-5570: can not convert twice");
1551
1552                         n_ret_value =
1553                             maybe_indirect_to_direct(th, inode, page,
1554                                                      p_s_path, p_s_item_key,
1555                                                      n_new_file_size, &c_mode);
1556                         if (c_mode == M_SKIP_BALANCING)
1557                                 /* tail has been left in the unformatted node */
1558                                 return n_ret_value;
1559
1560                         n_is_inode_locked = 1;
1561
1562                         /* removing of last unformatted node will change value we
1563                            have to return to truncate. Save it */
1564                         retval2 = n_ret_value;
1565                         /*retval2 = sb->s_blocksize - (n_new_file_size & (sb->s_blocksize - 1)); */
1566
1567                         /* So, we have performed the first part of the conversion:
1568                            inserting the new direct item.  Now we are removing the
1569                            last unformatted node pointer. Set key to search for
1570                            it. */
1571                         set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1572                         p_s_item_key->key_length = 4;
1573                         n_new_file_size -=
1574                             (n_new_file_size & (sb->s_blocksize - 1));
1575                         tail_pos = n_new_file_size;
1576                         set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1577                         if (search_for_position_by_key
1578                             (sb, p_s_item_key,
1579                              p_s_path) == POSITION_NOT_FOUND) {
1580                                 print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1581                                             PATH_LAST_POSITION(p_s_path) - 1,
1582                                             PATH_LAST_POSITION(p_s_path) + 1);
1583                                 reiserfs_panic(sb, "PAP-5580", "item to "
1584                                                "convert does not exist (%K)",
1585                                                p_s_item_key);
1586                         }
1587                         continue;
1588                 }
1589                 if (n_cut_size == 0) {
1590                         pathrelse(p_s_path);
1591                         return 0;
1592                 }
1593
1594                 s_cut_balance.insert_size[0] = n_cut_size;
1595
1596                 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1597                 if (n_ret_value != REPEAT_SEARCH)
1598                         break;
1599
1600                 PROC_INFO_INC(sb, cut_from_item_restarted);
1601
1602                 n_ret_value =
1603                     search_for_position_by_key(sb, p_s_item_key, p_s_path);
1604                 if (n_ret_value == POSITION_FOUND)
1605                         continue;
1606
1607                 reiserfs_warning(sb, "PAP-5610", "item %K not found",
1608                                  p_s_item_key);
1609                 unfix_nodes(&s_cut_balance);
1610                 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1611         }                       /* while */
1612
1613         // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1614         if (n_ret_value != CARRY_ON) {
1615                 if (n_is_inode_locked) {
1616                         // FIXME: this seems to be not needed: we are always able
1617                         // to cut item
1618                         indirect_to_direct_roll_back(th, inode, p_s_path);
1619                 }
1620                 if (n_ret_value == NO_DISK_SPACE)
1621                         reiserfs_warning(sb, "reiserfs-5092",
1622                                          "NO_DISK_SPACE");
1623                 unfix_nodes(&s_cut_balance);
1624                 return -EIO;
1625         }
1626
1627         /* go ahead and perform balancing */
1628
1629         RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1630
1631         /* Calculate number of bytes that need to be cut from the item. */
1632         quota_cut_bytes =
1633             (c_mode ==
1634              M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1635             insert_size[0];
1636         if (retval2 == -1)
1637                 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1638         else
1639                 n_ret_value = retval2;
1640
1641         /* For direct items, we only change the quota when deleting the last
1642          ** item.
1643          */
1644         p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1645         if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1646                 if (c_mode == M_DELETE &&
1647                     (le_ih_k_offset(p_le_ih) & (sb->s_blocksize - 1)) ==
1648                     1) {
1649                         // FIXME: this is to keep 3.5 happy
1650                         REISERFS_I(inode)->i_first_direct_byte = U32_MAX;
1651                         quota_cut_bytes = sb->s_blocksize + UNFM_P_SIZE;
1652                 } else {
1653                         quota_cut_bytes = 0;
1654                 }
1655         }
1656 #ifdef CONFIG_REISERFS_CHECK
1657         if (n_is_inode_locked) {
1658                 struct item_head *le_ih =
1659                     PATH_PITEM_HEAD(s_cut_balance.tb_path);
1660                 /* we are going to complete indirect2direct conversion. Make
1661                    sure, that we exactly remove last unformatted node pointer
1662                    of the item */
1663                 if (!is_indirect_le_ih(le_ih))
1664                         reiserfs_panic(sb, "vs-5652",
1665                                        "item must be indirect %h", le_ih);
1666
1667                 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1668                         reiserfs_panic(sb, "vs-5653", "completing "
1669                                        "indirect2direct conversion indirect "
1670                                        "item %h being deleted must be of "
1671                                        "4 byte long", le_ih);
1672
1673                 if (c_mode == M_CUT
1674                     && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1675                         reiserfs_panic(sb, "vs-5654", "can not complete "
1676                                        "indirect2direct conversion of %h "
1677                                        "(CUT, insert_size==%d)",
1678                                        le_ih, s_cut_balance.insert_size[0]);
1679                 }
1680                 /* it would be useful to make sure, that right neighboring
1681                    item is direct item of this file */
1682         }
1683 #endif
1684
1685         do_balance(&s_cut_balance, NULL, NULL, c_mode);
1686         if (n_is_inode_locked) {
1687                 /* we've done an indirect->direct conversion.  when the data block
1688                  ** was freed, it was removed from the list of blocks that must
1689                  ** be flushed before the transaction commits, make sure to
1690                  ** unmap and invalidate it
1691                  */
1692                 unmap_buffers(page, tail_pos);
1693                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
1694         }
1695 #ifdef REISERQUOTA_DEBUG
1696         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1697                        "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1698                        quota_cut_bytes, inode->i_uid, '?');
1699 #endif
1700         DQUOT_FREE_SPACE_NODIRTY(inode, quota_cut_bytes);
1701         return n_ret_value;
1702 }
1703
1704 static void truncate_directory(struct reiserfs_transaction_handle *th,
1705                                struct inode *inode)
1706 {
1707         BUG_ON(!th->t_trans_id);
1708         if (inode->i_nlink)
1709                 reiserfs_error(inode->i_sb, "vs-5655", "link count != 0");
1710
1711         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1712         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1713         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1714         reiserfs_update_sd(th, inode);
1715         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1716         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1717 }
1718
1719 /* Truncate file to the new size. Note, this must be called with a transaction
1720    already started */
1721 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th,
1722                           struct inode *inode,  /* ->i_size contains new size */
1723                          struct page *page,     /* up to date for last block */
1724                          int update_timestamps  /* when it is called by
1725                                                    file_release to convert
1726                                                    the tail - no timestamps
1727                                                    should be updated */
1728     )
1729 {
1730         INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1731         struct item_head *p_le_ih;      /* Pointer to an item header. */
1732         struct cpu_key s_item_key;      /* Key to search for a previous file item. */
1733         loff_t n_file_size,     /* Old file size. */
1734          n_new_file_size;       /* New file size. */
1735         int n_deleted;          /* Number of deleted or truncated bytes. */
1736         int retval;
1737         int err = 0;
1738
1739         BUG_ON(!th->t_trans_id);
1740         if (!
1741             (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1742              || S_ISLNK(inode->i_mode)))
1743                 return 0;
1744
1745         if (S_ISDIR(inode->i_mode)) {
1746                 // deletion of directory - no need to update timestamps
1747                 truncate_directory(th, inode);
1748                 return 0;
1749         }
1750
1751         /* Get new file size. */
1752         n_new_file_size = inode->i_size;
1753
1754         // FIXME: note, that key type is unimportant here
1755         make_cpu_key(&s_item_key, inode, max_reiserfs_offset(inode),
1756                      TYPE_DIRECT, 3);
1757
1758         retval =
1759             search_for_position_by_key(inode->i_sb, &s_item_key,
1760                                        &s_search_path);
1761         if (retval == IO_ERROR) {
1762                 reiserfs_error(inode->i_sb, "vs-5657",
1763                                "i/o failure occurred trying to truncate %K",
1764                                &s_item_key);
1765                 err = -EIO;
1766                 goto out;
1767         }
1768         if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1769                 reiserfs_error(inode->i_sb, "PAP-5660",
1770                                "wrong result %d of search for %K", retval,
1771                                &s_item_key);
1772
1773                 err = -EIO;
1774                 goto out;
1775         }
1776
1777         s_search_path.pos_in_item--;
1778
1779         /* Get real file size (total length of all file items) */
1780         p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1781         if (is_statdata_le_ih(p_le_ih))
1782                 n_file_size = 0;
1783         else {
1784                 loff_t offset = le_ih_k_offset(p_le_ih);
1785                 int bytes =
1786                     op_bytes_number(p_le_ih, inode->i_sb->s_blocksize);
1787
1788                 /* this may mismatch with real file size: if last direct item
1789                    had no padding zeros and last unformatted node had no free
1790                    space, this file would have this file size */
1791                 n_file_size = offset + bytes - 1;
1792         }
1793         /*
1794          * are we doing a full truncate or delete, if so
1795          * kick in the reada code
1796          */
1797         if (n_new_file_size == 0)
1798                 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1799
1800         if (n_file_size == 0 || n_file_size < n_new_file_size) {
1801                 goto update_and_out;
1802         }
1803
1804         /* Update key to search for the last file item. */
1805         set_cpu_key_k_offset(&s_item_key, n_file_size);
1806
1807         do {
1808                 /* Cut or delete file item. */
1809                 n_deleted =
1810                     reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1811                                            inode, page, n_new_file_size);
1812                 if (n_deleted < 0) {
1813                         reiserfs_warning(inode->i_sb, "vs-5665",
1814                                          "reiserfs_cut_from_item failed");
1815                         reiserfs_check_path(&s_search_path);
1816                         return 0;
1817                 }
1818
1819                 RFALSE(n_deleted > n_file_size,
1820                        "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1821                        n_deleted, n_file_size, &s_item_key);
1822
1823                 /* Change key to search the last file item. */
1824                 n_file_size -= n_deleted;
1825
1826                 set_cpu_key_k_offset(&s_item_key, n_file_size);
1827
1828                 /* While there are bytes to truncate and previous file item is presented in the tree. */
1829
1830                 /*
1831                  ** This loop could take a really long time, and could log
1832                  ** many more blocks than a transaction can hold.  So, we do a polite
1833                  ** journal end here, and if the transaction needs ending, we make
1834                  ** sure the file is consistent before ending the current trans
1835                  ** and starting a new one
1836                  */
1837                 if (journal_transaction_should_end(th, 0) ||
1838                     reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1839                         int orig_len_alloc = th->t_blocks_allocated;
1840                         pathrelse(&s_search_path);
1841
1842                         if (update_timestamps) {
1843                                 inode->i_mtime = CURRENT_TIME_SEC;
1844                                 inode->i_ctime = CURRENT_TIME_SEC;
1845                         }
1846                         reiserfs_update_sd(th, inode);
1847
1848                         err = journal_end(th, inode->i_sb, orig_len_alloc);
1849                         if (err)
1850                                 goto out;
1851                         err = journal_begin(th, inode->i_sb,
1852                                             JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1853                         if (err)
1854                                 goto out;
1855                         reiserfs_update_inode_transaction(inode);
1856                 }
1857         } while (n_file_size > ROUND_UP(n_new_file_size) &&
1858                  search_for_position_by_key(inode->i_sb, &s_item_key,
1859                                             &s_search_path) == POSITION_FOUND);
1860
1861         RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1862                "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1863                n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1864
1865       update_and_out:
1866         if (update_timestamps) {
1867                 // this is truncate, not file closing
1868                 inode->i_mtime = CURRENT_TIME_SEC;
1869                 inode->i_ctime = CURRENT_TIME_SEC;
1870         }
1871         reiserfs_update_sd(th, inode);
1872
1873       out:
1874         pathrelse(&s_search_path);
1875         return err;
1876 }
1877
1878 #ifdef CONFIG_REISERFS_CHECK
1879 // this makes sure, that we __append__, not overwrite or add holes
1880 static void check_research_for_paste(struct treepath *path,
1881                                      const struct cpu_key *p_s_key)
1882 {
1883         struct item_head *found_ih = get_ih(path);
1884
1885         if (is_direct_le_ih(found_ih)) {
1886                 if (le_ih_k_offset(found_ih) +
1887                     op_bytes_number(found_ih,
1888                                     get_last_bh(path)->b_size) !=
1889                     cpu_key_k_offset(p_s_key)
1890                     || op_bytes_number(found_ih,
1891                                        get_last_bh(path)->b_size) !=
1892                     pos_in_item(path))
1893                         reiserfs_panic(NULL, "PAP-5720", "found direct item "
1894                                        "%h or position (%d) does not match "
1895                                        "to key %K", found_ih,
1896                                        pos_in_item(path), p_s_key);
1897         }
1898         if (is_indirect_le_ih(found_ih)) {
1899                 if (le_ih_k_offset(found_ih) +
1900                     op_bytes_number(found_ih,
1901                                     get_last_bh(path)->b_size) !=
1902                     cpu_key_k_offset(p_s_key)
1903                     || I_UNFM_NUM(found_ih) != pos_in_item(path)
1904                     || get_ih_free_space(found_ih) != 0)
1905                         reiserfs_panic(NULL, "PAP-5730", "found indirect "
1906                                        "item (%h) or position (%d) does not "
1907                                        "match to key (%K)",
1908                                        found_ih, pos_in_item(path), p_s_key);
1909         }
1910 }
1911 #endif                          /* config reiserfs check */
1912
1913 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1914 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_search_path,  /* Path to the pasted item.          */
1915                              const struct cpu_key *p_s_key,     /* Key to search for the needed item. */
1916                              struct inode *inode,       /* Inode item belongs to */
1917                              const char *p_c_body,      /* Pointer to the bytes to paste.    */
1918                              int n_pasted_size)
1919 {                               /* Size of pasted bytes.             */
1920         struct tree_balance s_paste_balance;
1921         int retval;
1922         int fs_gen;
1923
1924         BUG_ON(!th->t_trans_id);
1925
1926         fs_gen = get_generation(inode->i_sb);
1927
1928 #ifdef REISERQUOTA_DEBUG
1929         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1930                        "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1931                        n_pasted_size, inode->i_uid,
1932                        key2type(&(p_s_key->on_disk_key)));
1933 #endif
1934
1935         if (DQUOT_ALLOC_SPACE_NODIRTY(inode, n_pasted_size)) {
1936                 pathrelse(p_s_search_path);
1937                 return -EDQUOT;
1938         }
1939         init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
1940                        n_pasted_size);
1941 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1942         s_paste_balance.key = p_s_key->on_disk_key;
1943 #endif
1944
1945         /* DQUOT_* can schedule, must check before the fix_nodes */
1946         if (fs_changed(fs_gen, inode->i_sb)) {
1947                 goto search_again;
1948         }
1949
1950         while ((retval =
1951                 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1952                           p_c_body)) == REPEAT_SEARCH) {
1953               search_again:
1954                 /* file system changed while we were in the fix_nodes */
1955                 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1956                 retval =
1957                     search_for_position_by_key(th->t_super, p_s_key,
1958                                                p_s_search_path);
1959                 if (retval == IO_ERROR) {
1960                         retval = -EIO;
1961                         goto error_out;
1962                 }
1963                 if (retval == POSITION_FOUND) {
1964                         reiserfs_warning(inode->i_sb, "PAP-5710",
1965                                          "entry or pasted byte (%K) exists",
1966                                          p_s_key);
1967                         retval = -EEXIST;
1968                         goto error_out;
1969                 }
1970 #ifdef CONFIG_REISERFS_CHECK
1971                 check_research_for_paste(p_s_search_path, p_s_key);
1972 #endif
1973         }
1974
1975         /* Perform balancing after all resources are collected by fix_nodes, and
1976            accessing them will not risk triggering schedule. */
1977         if (retval == CARRY_ON) {
1978                 do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
1979                 return 0;
1980         }
1981         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
1982       error_out:
1983         /* this also releases the path */
1984         unfix_nodes(&s_paste_balance);
1985 #ifdef REISERQUOTA_DEBUG
1986         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1987                        "reiserquota paste_into_item(): freeing %u id=%u type=%c",
1988                        n_pasted_size, inode->i_uid,
1989                        key2type(&(p_s_key->on_disk_key)));
1990 #endif
1991         DQUOT_FREE_SPACE_NODIRTY(inode, n_pasted_size);
1992         return retval;
1993 }
1994
1995 /* Insert new item into the buffer at the path. */
1996 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the inserteded item.         */
1997                          const struct cpu_key *key, struct item_head *p_s_ih,   /* Pointer to the item header to insert. */
1998                          struct inode *inode, const char *p_c_body)
1999 {                               /* Pointer to the bytes to insert.      */
2000         struct tree_balance s_ins_balance;
2001         int retval;
2002         int fs_gen = 0;
2003         int quota_bytes = 0;
2004
2005         BUG_ON(!th->t_trans_id);
2006
2007         if (inode) {            /* Do we count quotas for item? */
2008                 fs_gen = get_generation(inode->i_sb);
2009                 quota_bytes = ih_item_len(p_s_ih);
2010
2011                 /* hack so the quota code doesn't have to guess if the file has
2012                  ** a tail, links are always tails, so there's no guessing needed
2013                  */
2014                 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2015                         quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2016                 }
2017 #ifdef REISERQUOTA_DEBUG
2018                 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2019                                "reiserquota insert_item(): allocating %u id=%u type=%c",
2020                                quota_bytes, inode->i_uid, head2type(p_s_ih));
2021 #endif
2022                 /* We can't dirty inode here. It would be immediately written but
2023                  * appropriate stat item isn't inserted yet... */
2024                 if (DQUOT_ALLOC_SPACE_NODIRTY(inode, quota_bytes)) {
2025                         pathrelse(p_s_path);
2026                         return -EDQUOT;
2027                 }
2028         }
2029         init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2030                        IH_SIZE + ih_item_len(p_s_ih));
2031 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2032         s_ins_balance.key = key->on_disk_key;
2033 #endif
2034         /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2035         if (inode && fs_changed(fs_gen, inode->i_sb)) {
2036                 goto search_again;
2037         }
2038
2039         while ((retval =
2040                 fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2041                           p_c_body)) == REPEAT_SEARCH) {
2042               search_again:
2043                 /* file system changed while we were in the fix_nodes */
2044                 PROC_INFO_INC(th->t_super, insert_item_restarted);
2045                 retval = search_item(th->t_super, key, p_s_path);
2046                 if (retval == IO_ERROR) {
2047                         retval = -EIO;
2048                         goto error_out;
2049                 }
2050                 if (retval == ITEM_FOUND) {
2051                         reiserfs_warning(th->t_super, "PAP-5760",
2052                                          "key %K already exists in the tree",
2053                                          key);
2054                         retval = -EEXIST;
2055                         goto error_out;
2056                 }
2057         }
2058
2059         /* make balancing after all resources will be collected at a time */
2060         if (retval == CARRY_ON) {
2061                 do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2062                 return 0;
2063         }
2064
2065         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2066       error_out:
2067         /* also releases the path */
2068         unfix_nodes(&s_ins_balance);
2069 #ifdef REISERQUOTA_DEBUG
2070         reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2071                        "reiserquota insert_item(): freeing %u id=%u type=%c",
2072                        quota_bytes, inode->i_uid, head2type(p_s_ih));
2073 #endif
2074         if (inode)
2075                 DQUOT_FREE_SPACE_NODIRTY(inode, quota_bytes);
2076         return retval;
2077 }