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  * decrement_counters_in_path
  27  * reiserfs_check_path
  28  * pathrelse_and_restore
  29  * pathrelse
  30  * search_by_key_reada
  31  * search_by_key
  32  * search_for_position_by_key
  33  * comp_items
  34  * prepare_for_direct_item
  35  * prepare_for_direntry_item
  36  * prepare_for_delete_or_cut
  37  * calc_deleted_bytes_number
  38  * init_tb_struct
  39  * padd_item
  40  * reiserfs_delete_item
  41  * reiserfs_delete_solid_item
  42  * reiserfs_delete_object
  43  * maybe_indirect_to_direct
  44  * indirect_to_direct_roll_back
  45  * reiserfs_cut_from_item
  46  * truncate_directory
  47  * reiserfs_do_truncate
  48  * reiserfs_paste_into_item
  49  * reiserfs_insert_item
  50  */
  51
  52 #include <linux/time.h>
  53 #include <linux/string.h>
  54 #include <linux/pagemap.h>
  55 #include <linux/reiserfs_fs.h>
  56 #include <linux/buffer_head.h>
  57 #include <linux/quotaops.h>
  58
  59 /* Does the buffer contain a disk block which is in the tree. */
  60 inline int B_IS_IN_TREE(const struct buffer_head *p_s_bh)
  61 {
  62
  63         RFALSE(B_LEVEL(p_s_bh) > MAX_HEIGHT,
  64                "PAP-1010: block (%b) has too big level (%z)", p_s_bh, p_s_bh);
  65
  66         return (B_LEVEL(p_s_bh) != FREE_LEVEL);
  67 }
  68
  69 //
  70 // to gets item head in le form
  71 //
  72 inline void copy_item_head(struct item_head *p_v_to,
  73                            const struct item_head *p_v_from)
  74 {
  75         memcpy(p_v_to, p_v_from, IH_SIZE);
  76 }
  77
  78 /* k1 is pointer to on-disk structure which is stored in little-endian
  79    form. k2 is pointer to cpu variable. For key of items of the same
  80    object this returns 0.
  81    Returns: -1 if key1 < key2
  82    0 if key1 == key2
  83    1 if key1 > key2 */
  84 inline int comp_short_keys(const struct reiserfs_key *le_key,
  85                            const struct cpu_key *cpu_key)
  86 {
  87         __u32 n;
  88         n = le32_to_cpu(le_key->k_dir_id);
  89         if (n < cpu_key->on_disk_key.k_dir_id)
  90                 return -1;
  91         if (n > cpu_key->on_disk_key.k_dir_id)
  92                 return 1;
  93         n = le32_to_cpu(le_key->k_objectid);
  94         if (n < cpu_key->on_disk_key.k_objectid)
  95                 return -1;
  96         if (n > cpu_key->on_disk_key.k_objectid)
  97                 return 1;
  98         return 0;
  99 }
 100
 101 /* k1 is pointer to on-disk structure which is stored in little-endian
 102    form. k2 is pointer to cpu variable.
 103    Compare keys using all 4 key fields.
 104    Returns: -1 if key1 < key2 0
 105    if key1 = key2 1 if key1 > key2 */
 106 static inline int comp_keys(const struct reiserfs_key *le_key,
 107                             const struct cpu_key *cpu_key)
 108 {
 109         int retval;
 110
 111         retval = comp_short_keys(le_key, cpu_key);
 112         if (retval)
 113                 return retval;
 114         if (le_key_k_offset(le_key_version(le_key), le_key) <
 115             cpu_key_k_offset(cpu_key))
 116                 return -1;
 117         if (le_key_k_offset(le_key_version(le_key), le_key) >
 118             cpu_key_k_offset(cpu_key))
 119                 return 1;
 120
 121         if (cpu_key->key_length == 3)
 122                 return 0;
 123
 124         /* this part is needed only when tail conversion is in progress */
 125         if (le_key_k_type(le_key_version(le_key), le_key) <
 126             cpu_key_k_type(cpu_key))
 127                 return -1;
 128
 129         if (le_key_k_type(le_key_version(le_key), le_key) >
 130             cpu_key_k_type(cpu_key))
 131                 return 1;
 132
 133         return 0;
 134 }
 135
 136 inline int comp_short_le_keys(const struct reiserfs_key *key1,
 137                               const struct reiserfs_key *key2)
 138 {
 139         __u32 *p_s_1_u32, *p_s_2_u32;
 140         int n_key_length = REISERFS_SHORT_KEY_LEN;
 141
 142         p_s_1_u32 = (__u32 *) key1;
 143         p_s_2_u32 = (__u32 *) key2;
 144         for (; n_key_length--; ++p_s_1_u32, ++p_s_2_u32) {
 145                 if (le32_to_cpu(*p_s_1_u32) < le32_to_cpu(*p_s_2_u32))
 146                         return -1;
 147                 if (le32_to_cpu(*p_s_1_u32) > le32_to_cpu(*p_s_2_u32))
 148                         return 1;
 149         }
 150         return 0;
 151 }
 152
 153 inline void le_key2cpu_key(struct cpu_key *to, const struct reiserfs_key *from)
 154 {
 155         int version;
 156         to->on_disk_key.k_dir_id = le32_to_cpu(from->k_dir_id);
 157         to->on_disk_key.k_objectid = le32_to_cpu(from->k_objectid);
 158
 159         // find out version of the key
 160         version = le_key_version(from);
 161         to->version = version;
 162         to->on_disk_key.k_offset = le_key_k_offset(version, from);
 163         to->on_disk_key.k_type = le_key_k_type(version, from);
 164 }
 165
 166 // this does not say which one is bigger, it only returns 1 if keys
 167 // are not equal, 0 otherwise
 168 inline int comp_le_keys(const struct reiserfs_key *k1,
 169                         const struct reiserfs_key *k2)
 170 {
 171         return memcmp(k1, k2, sizeof(struct reiserfs_key));
 172 }
 173
 174 /**************************************************************************
 175  *  Binary search toolkit function                                        *
 176  *  Search for an item in the array by the item key                       *
 177  *  Returns:    1 if found,  0 if not found;                              *
 178  *        *p_n_pos = number of the searched element if found, else the    *
 179  *        number of the first element that is larger than p_v_key.        *
 180  **************************************************************************/
 181 /* For those not familiar with binary search: n_lbound is the leftmost item that it
 182  could be, n_rbound the rightmost item that it could be.  We examine the item
 183  halfway between n_lbound and n_rbound, and that tells us either that we can increase
 184  n_lbound, or decrease n_rbound, or that we have found it, or if n_lbound <= n_rbound that
 185  there are no possible items, and we have not found it. With each examination we
 186  cut the number of possible items it could be by one more than half rounded down,
 187  or we find it. */
 188 static inline int bin_search(const void *p_v_key,       /* Key to search for.                   */
 189                              const void *p_v_base,      /* First item in the array.             */
 190                              int p_n_num,       /* Number of items in the array.        */
 191                              int p_n_width,     /* Item size in the array.
 192                                                    searched. Lest the reader be
 193                                                    confused, note that this is crafted
 194                                                    as a general function, and when it
 195                                                    is applied specifically to the array
 196                                                    of item headers in a node, p_n_width
 197                                                    is actually the item header size not
 198                                                    the item size.                      */
 199                              int *p_n_pos       /* Number of the searched for element. */
 200     )
 201 {
 202         int n_rbound, n_lbound, n_j;
 203
 204         for (n_j = ((n_rbound = p_n_num - 1) + (n_lbound = 0)) / 2;
 205              n_lbound <= n_rbound; n_j = (n_rbound + n_lbound) / 2)
 206                 switch (comp_keys
 207                         ((struct reiserfs_key *)((char *)p_v_base +
 208                                                  n_j * p_n_width),
 209                          (struct cpu_key *)p_v_key)) {
 210                 case -1:
 211                         n_lbound = n_j + 1;
 212                         continue;
 213                 case 1:
 214                         n_rbound = n_j - 1;
 215                         continue;
 216                 case 0:
 217                         *p_n_pos = n_j;
 218                         return ITEM_FOUND;      /* Key found in the array.  */
 219                 }
 220
 221         /* bin_search did not find given key, it returns position of key,
 222            that is minimal and greater than the given one. */
 223         *p_n_pos = n_lbound;
 224         return ITEM_NOT_FOUND;
 225 }
 226
 227 #ifdef CONFIG_REISERFS_CHECK
 228 extern struct tree_balance *cur_tb;
 229 #endif
 230
 231 /* Minimal possible key. It is never in the tree. */
 232 const struct reiserfs_key MIN_KEY = { 0, 0, {{0, 0},} };
 233
 234 /* Maximal possible key. It is never in the tree. */
 235 static const struct reiserfs_key MAX_KEY = {
 236         __constant_cpu_to_le32(0xffffffff),
 237         __constant_cpu_to_le32(0xffffffff),
 238         {{__constant_cpu_to_le32(0xffffffff),
 239           __constant_cpu_to_le32(0xffffffff)},}
 240 };
 241
 242 /* Get delimiting key of the buffer by looking for it in the buffers in the path, starting from the bottom
 243    of the path, and going upwards.  We must check the path's validity at each step.  If the key is not in
 244    the path, there is no delimiting key in the tree (buffer is first or last buffer in tree), and in this
 245    case we return a special key, either MIN_KEY or MAX_KEY. */
 246 static inline const struct reiserfs_key *get_lkey(const struct treepath
 247                                                   *p_s_chk_path,
 248                                                   const struct super_block
 249                                                   *p_s_sb)
 250 {
 251         int n_position, n_path_offset = p_s_chk_path->path_length;
 252         struct buffer_head *p_s_parent;
 253
 254         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
 255                "PAP-5010: invalid offset in the path");
 256
 257         /* While not higher in path than first element. */
 258         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 259
 260                 RFALSE(!buffer_uptodate
 261                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
 262                        "PAP-5020: parent is not uptodate");
 263
 264                 /* Parent at the path is not in the tree now. */
 265                 if (!B_IS_IN_TREE
 266                     (p_s_parent =
 267                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
 268                         return &MAX_KEY;
 269                 /* Check whether position in the parent is correct. */
 270                 if ((n_position =
 271                      PATH_OFFSET_POSITION(p_s_chk_path,
 272                                           n_path_offset)) >
 273                     B_NR_ITEMS(p_s_parent))
 274                         return &MAX_KEY;
 275                 /* Check whether parent at the path really points to the child. */
 276                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
 277                     PATH_OFFSET_PBUFFER(p_s_chk_path,
 278                                         n_path_offset + 1)->b_blocknr)
 279                         return &MAX_KEY;
 280                 /* Return delimiting key if position in the parent is not equal to zero. */
 281                 if (n_position)
 282                         return B_N_PDELIM_KEY(p_s_parent, n_position - 1);
 283         }
 284         /* Return MIN_KEY if we are in the root of the buffer tree. */
 285         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 286             b_blocknr == SB_ROOT_BLOCK(p_s_sb))
 287                 return &MIN_KEY;
 288         return &MAX_KEY;
 289 }
 290
 291 /* Get delimiting key of the buffer at the path and its right neighbor. */
 292 inline const struct reiserfs_key *get_rkey(const struct treepath *p_s_chk_path,
 293                                            const struct super_block *p_s_sb)
 294 {
 295         int n_position, n_path_offset = p_s_chk_path->path_length;
 296         struct buffer_head *p_s_parent;
 297
 298         RFALSE(n_path_offset < FIRST_PATH_ELEMENT_OFFSET,
 299                "PAP-5030: invalid offset in the path");
 300
 301         while (n_path_offset-- > FIRST_PATH_ELEMENT_OFFSET) {
 302
 303                 RFALSE(!buffer_uptodate
 304                        (PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)),
 305                        "PAP-5040: parent is not uptodate");
 306
 307                 /* Parent at the path is not in the tree now. */
 308                 if (!B_IS_IN_TREE
 309                     (p_s_parent =
 310                      PATH_OFFSET_PBUFFER(p_s_chk_path, n_path_offset)))
 311                         return &MIN_KEY;
 312                 /* Check whether position in the parent is correct. */
 313                 if ((n_position =
 314                      PATH_OFFSET_POSITION(p_s_chk_path,
 315                                           n_path_offset)) >
 316                     B_NR_ITEMS(p_s_parent))
 317                         return &MIN_KEY;
 318                 /* Check whether parent at the path really points to the child. */
 319                 if (B_N_CHILD_NUM(p_s_parent, n_position) !=
 320                     PATH_OFFSET_PBUFFER(p_s_chk_path,
 321                                         n_path_offset + 1)->b_blocknr)
 322                         return &MIN_KEY;
 323                 /* Return delimiting key if position in the parent is not the last one. */
 324                 if (n_position != B_NR_ITEMS(p_s_parent))
 325                         return B_N_PDELIM_KEY(p_s_parent, n_position);
 326         }
 327         /* Return MAX_KEY if we are in the root of the buffer tree. */
 328         if (PATH_OFFSET_PBUFFER(p_s_chk_path, FIRST_PATH_ELEMENT_OFFSET)->
 329             b_blocknr == SB_ROOT_BLOCK(p_s_sb))
 330                 return &MAX_KEY;
 331         return &MIN_KEY;
 332 }
 333
 334 /* Check whether a key is contained in the tree rooted from a buffer at a path. */
 335 /* This works by looking at the left and right delimiting keys for the buffer in the last path_element in
 336    the path.  These delimiting keys are stored at least one level above that buffer in the tree. If the
 337    buffer is the first or last node in the tree order then one of the delimiting keys may be absent, and in
 338    this case get_lkey and get_rkey return a special key which is MIN_KEY or MAX_KEY. */
 339 static inline int key_in_buffer(struct treepath *p_s_chk_path,  /* Path which should be checked.  */
 340                                 const struct cpu_key *p_s_key,  /* Key which should be checked.   */
 341                                 struct super_block *p_s_sb      /* Super block pointer.           */
 342     )
 343 {
 344
 345         RFALSE(!p_s_key || p_s_chk_path->path_length < FIRST_PATH_ELEMENT_OFFSET
 346                || p_s_chk_path->path_length > MAX_HEIGHT,
 347                "PAP-5050: pointer to the key(%p) is NULL or invalid path length(%d)",
 348                p_s_key, p_s_chk_path->path_length);
 349         RFALSE(!PATH_PLAST_BUFFER(p_s_chk_path)->b_bdev,
 350                "PAP-5060: device must not be NODEV");
 351
 352         if (comp_keys(get_lkey(p_s_chk_path, p_s_sb), p_s_key) == 1)
 353                 /* left delimiting key is bigger, that the key we look for */
 354                 return 0;
 355         //  if ( comp_keys(p_s_key, get_rkey(p_s_chk_path, p_s_sb)) != -1 )
 356         if (comp_keys(get_rkey(p_s_chk_path, p_s_sb), p_s_key) != 1)
 357                 /* p_s_key must be less than right delimitiing key */
 358                 return 0;
 359         return 1;
 360 }
 361
 362 inline void decrement_bcount(struct buffer_head *p_s_bh)
 363 {
 364         if (p_s_bh) {
 365                 if (atomic_read(&(p_s_bh->b_count))) {
 366                         put_bh(p_s_bh);
 367                         return;
 368                 }
 369                 reiserfs_panic(NULL,
 370                                "PAP-5070: decrement_bcount: trying to free free buffer %b",
 371                                p_s_bh);
 372         }
 373 }
 374
 375 /* Decrement b_count field of the all buffers in the path. */
 376 void decrement_counters_in_path(struct treepath *p_s_search_path)
 377 {
 378         int n_path_offset = p_s_search_path->path_length;
 379
 380         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET ||
 381                n_path_offset > EXTENDED_MAX_HEIGHT - 1,
 382                "PAP-5080: invalid path offset of %d", n_path_offset);
 383
 384         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
 385                 struct buffer_head *bh;
 386
 387                 bh = PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--);
 388                 decrement_bcount(bh);
 389         }
 390         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 391 }
 392
 393 int reiserfs_check_path(struct treepath *p)
 394 {
 395         RFALSE(p->path_length != ILLEGAL_PATH_ELEMENT_OFFSET,
 396                "path not properly relsed");
 397         return 0;
 398 }
 399
 400 /* Release all buffers in the path. Restore dirty bits clean
 401 ** when preparing the buffer for the log
 402 **
 403 ** only called from fix_nodes()
 404 */
 405 void pathrelse_and_restore(struct super_block *s, struct treepath *p_s_search_path)
 406 {
 407         int n_path_offset = p_s_search_path->path_length;
 408
 409         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 410                "clm-4000: invalid path offset");
 411
 412         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET) {
 413                 reiserfs_restore_prepared_buffer(s,
 414                                                  PATH_OFFSET_PBUFFER
 415                                                  (p_s_search_path,
 416                                                   n_path_offset));
 417                 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
 418         }
 419         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 420 }
 421
 422 /* Release all buffers in the path. */
 423 void pathrelse(struct treepath *p_s_search_path)
 424 {
 425         int n_path_offset = p_s_search_path->path_length;
 426
 427         RFALSE(n_path_offset < ILLEGAL_PATH_ELEMENT_OFFSET,
 428                "PAP-5090: invalid path offset");
 429
 430         while (n_path_offset > ILLEGAL_PATH_ELEMENT_OFFSET)
 431                 brelse(PATH_OFFSET_PBUFFER(p_s_search_path, n_path_offset--));
 432
 433         p_s_search_path->path_length = ILLEGAL_PATH_ELEMENT_OFFSET;
 434 }
 435
 436 static int is_leaf(char *buf, int blocksize, struct buffer_head *bh)
 437 {
 438         struct block_head *blkh;
 439         struct item_head *ih;
 440         int used_space;
 441         int prev_location;
 442         int i;
 443         int nr;
 444
 445         blkh = (struct block_head *)buf;
 446         if (blkh_level(blkh) != DISK_LEAF_NODE_LEVEL) {
 447                 reiserfs_warning(NULL,
 448                                  "is_leaf: this should be caught earlier");
 449                 return 0;
 450         }
 451
 452         nr = blkh_nr_item(blkh);
 453         if (nr < 1 || nr > ((blocksize - BLKH_SIZE) / (IH_SIZE + MIN_ITEM_LEN))) {
 454                 /* item number is too big or too small */
 455                 reiserfs_warning(NULL, "is_leaf: nr_item seems wrong: %z", bh);
 456                 return 0;
 457         }
 458         ih = (struct item_head *)(buf + BLKH_SIZE) + nr - 1;
 459         used_space = BLKH_SIZE + IH_SIZE * nr + (blocksize - ih_location(ih));
 460         if (used_space != blocksize - blkh_free_space(blkh)) {
 461                 /* free space does not match to calculated amount of use space */
 462                 reiserfs_warning(NULL, "is_leaf: free space seems wrong: %z",
 463                                  bh);
 464                 return 0;
 465         }
 466         // FIXME: it is_leaf will hit performance too much - we may have
 467         // return 1 here
 468
 469         /* check tables of item heads */
 470         ih = (struct item_head *)(buf + BLKH_SIZE);
 471         prev_location = blocksize;
 472         for (i = 0; i < nr; i++, ih++) {
 473                 if (le_ih_k_type(ih) == TYPE_ANY) {
 474                         reiserfs_warning(NULL,
 475                                          "is_leaf: wrong item type for item %h",
 476                                          ih);
 477                         return 0;
 478                 }
 479                 if (ih_location(ih) >= blocksize
 480                     || ih_location(ih) < IH_SIZE * nr) {
 481                         reiserfs_warning(NULL,
 482                                          "is_leaf: item location seems wrong: %h",
 483                                          ih);
 484                         return 0;
 485                 }
 486                 if (ih_item_len(ih) < 1
 487                     || ih_item_len(ih) > MAX_ITEM_LEN(blocksize)) {
 488                         reiserfs_warning(NULL,
 489                                          "is_leaf: item length seems wrong: %h",
 490                                          ih);
 491                         return 0;
 492                 }
 493                 if (prev_location - ih_location(ih) != ih_item_len(ih)) {
 494                         reiserfs_warning(NULL,
 495                                          "is_leaf: item location seems wrong (second one): %h",
 496                                          ih);
 497                         return 0;
 498                 }
 499                 prev_location = ih_location(ih);
 500         }
 501
 502         // one may imagine much more checks
 503         return 1;
 504 }
 505
 506 /* returns 1 if buf looks like an internal node, 0 otherwise */
 507 static int is_internal(char *buf, int blocksize, struct buffer_head *bh)
 508 {
 509         struct block_head *blkh;
 510         int nr;
 511         int used_space;
 512
 513         blkh = (struct block_head *)buf;
 514         nr = blkh_level(blkh);
 515         if (nr <= DISK_LEAF_NODE_LEVEL || nr > MAX_HEIGHT) {
 516                 /* this level is not possible for internal nodes */
 517                 reiserfs_warning(NULL,
 518                                  "is_internal: this should be caught earlier");
 519                 return 0;
 520         }
 521
 522         nr = blkh_nr_item(blkh);
 523         if (nr > (blocksize - BLKH_SIZE - DC_SIZE) / (KEY_SIZE + DC_SIZE)) {
 524                 /* for internal which is not root we might check min number of keys */
 525                 reiserfs_warning(NULL,
 526                                  "is_internal: number of key seems wrong: %z",
 527                                  bh);
 528                 return 0;
 529         }
 530
 531         used_space = BLKH_SIZE + KEY_SIZE * nr + DC_SIZE * (nr + 1);
 532         if (used_space != blocksize - blkh_free_space(blkh)) {
 533                 reiserfs_warning(NULL,
 534                                  "is_internal: free space seems wrong: %z", bh);
 535                 return 0;
 536         }
 537         // one may imagine much more checks
 538         return 1;
 539 }
 540
 541 // make sure that bh contains formatted node of reiserfs tree of
 542 // 'level'-th level
 543 static int is_tree_node(struct buffer_head *bh, int level)
 544 {
 545         if (B_LEVEL(bh) != level) {
 546                 reiserfs_warning(NULL,
 547                                  "is_tree_node: node level %d does not match to the expected one %d",
 548                                  B_LEVEL(bh), level);
 549                 return 0;
 550         }
 551         if (level == DISK_LEAF_NODE_LEVEL)
 552                 return is_leaf(bh->b_data, bh->b_size, bh);
 553
 554         return is_internal(bh->b_data, bh->b_size, bh);
 555 }
 556
 557 #define SEARCH_BY_KEY_READA 16
 558
 559 /* The function is NOT SCHEDULE-SAFE! */
 560 static void search_by_key_reada(struct super_block *s,
 561                                 struct buffer_head **bh,
 562                                 b_blocknr_t *b, int num)
 563 {
 564         int i, j;
 565
 566         for (i = 0; i < num; i++) {
 567                 bh[i] = sb_getblk(s, b[i]);
 568         }
 569         for (j = 0; j < i; j++) {
 570                 /*
 571                  * note, this needs attention if we are getting rid of the BKL
 572                  * you have to make sure the prepared bit isn't set on this buffer
 573                  */
 574                 if (!buffer_uptodate(bh[j]))
 575                         ll_rw_block(READA, 1, bh + j);
 576                 brelse(bh[j]);
 577         }
 578 }
 579
 580 /**************************************************************************
 581  * Algorithm   SearchByKey                                                *
 582  *             look for item in the Disk S+Tree by its key                *
 583  * Input:  p_s_sb   -  super block                                        *
 584  *         p_s_key  - pointer to the key to search                        *
 585  * Output: ITEM_FOUND, ITEM_NOT_FOUND or IO_ERROR                         *
 586  *         p_s_search_path - path from the root to the needed leaf        *
 587  **************************************************************************/
 588
 589 /* This function fills up the path from the root to the leaf as it
 590    descends the tree looking for the key.  It uses reiserfs_bread to
 591    try to find buffers in the cache given their block number.  If it
 592    does not find them in the cache it reads them from disk.  For each
 593    node search_by_key finds using reiserfs_bread it then uses
 594    bin_search to look through that node.  bin_search will find the
 595    position of the block_number of the next node if it is looking
 596    through an internal node.  If it is looking through a leaf node
 597    bin_search will find the position of the item which has key either
 598    equal to given key, or which is the maximal key less than the given
 599    key.  search_by_key returns a path that must be checked for the
 600    correctness of the top of the path but need not be checked for the
 601    correctness of the bottom of the path */
 602 /* The function is NOT SCHEDULE-SAFE! */
 603 int search_by_key(struct super_block *p_s_sb, const struct cpu_key *p_s_key,    /* Key to search. */
 604                   struct treepath *p_s_search_path,/* This structure was
 605                                                    allocated and initialized
 606                                                    by the calling
 607                                                    function. It is filled up
 608                                                    by this function.  */
 609                   int n_stop_level      /* How far down the tree to search. To
 610                                            stop at leaf level - set to
 611                                            DISK_LEAF_NODE_LEVEL */
 612     )
 613 {
 614         b_blocknr_t n_block_number;
 615         int expected_level;
 616         struct buffer_head *p_s_bh;
 617         struct path_element *p_s_last_element;
 618         int n_node_level, n_retval;
 619         int right_neighbor_of_leaf_node;
 620         int fs_gen;
 621         struct buffer_head *reada_bh[SEARCH_BY_KEY_READA];
 622         b_blocknr_t reada_blocks[SEARCH_BY_KEY_READA];
 623         int reada_count = 0;
 624
 625 #ifdef CONFIG_REISERFS_CHECK
 626         int n_repeat_counter = 0;
 627 #endif
 628
 629         PROC_INFO_INC(p_s_sb, search_by_key);
 630
 631         /* As we add each node to a path we increase its count.  This means that
 632            we must be careful to release all nodes in a path before we either
 633            discard the path struct or re-use the path struct, as we do here. */
 634
 635         decrement_counters_in_path(p_s_search_path);
 636
 637         right_neighbor_of_leaf_node = 0;
 638
 639         /* With each iteration of this loop we search through the items in the
 640            current node, and calculate the next current node(next path element)
 641            for the next iteration of this loop.. */
 642         n_block_number = SB_ROOT_BLOCK(p_s_sb);
 643         expected_level = -1;
 644         while (1) {
 645
 646 #ifdef CONFIG_REISERFS_CHECK
 647                 if (!(++n_repeat_counter % 50000))
 648                         reiserfs_warning(p_s_sb, "PAP-5100: search_by_key: %s:"
 649                                          "there were %d iterations of while loop "
 650                                          "looking for key %K",
 651                                          current->comm, n_repeat_counter,
 652                                          p_s_key);
 653 #endif
 654
 655                 /* prep path to have another element added to it. */
 656                 p_s_last_element =
 657                     PATH_OFFSET_PELEMENT(p_s_search_path,
 658                                          ++p_s_search_path->path_length);
 659                 fs_gen = get_generation(p_s_sb);
 660
 661                 /* Read the next tree node, and set the last element in the path to
 662                    have a pointer to it. */
 663                 if ((p_s_bh = p_s_last_element->pe_buffer =
 664                      sb_getblk(p_s_sb, n_block_number))) {
 665                         if (!buffer_uptodate(p_s_bh) && reada_count > 1) {
 666                                 search_by_key_reada(p_s_sb, reada_bh,
 667                                                     reada_blocks, reada_count);
 668                         }
 669                         ll_rw_block(READ, 1, &p_s_bh);
 670                         wait_on_buffer(p_s_bh);
 671                         if (!buffer_uptodate(p_s_bh))
 672                                 goto io_error;
 673                 } else {
 674                       io_error:
 675                         p_s_search_path->path_length--;
 676                         pathrelse(p_s_search_path);
 677                         return IO_ERROR;
 678                 }
 679                 reada_count = 0;
 680                 if (expected_level == -1)
 681                         expected_level = SB_TREE_HEIGHT(p_s_sb);
 682                 expected_level--;
 683
 684                 /* It is possible that schedule occurred. We must check whether the key
 685                    to search is still in the tree rooted from the current buffer. If
 686                    not then repeat search from the root. */
 687                 if (fs_changed(fs_gen, p_s_sb) &&
 688                     (!B_IS_IN_TREE(p_s_bh) ||
 689                      B_LEVEL(p_s_bh) != expected_level ||
 690                      !key_in_buffer(p_s_search_path, p_s_key, p_s_sb))) {
 691                         PROC_INFO_INC(p_s_sb, search_by_key_fs_changed);
 692                         PROC_INFO_INC(p_s_sb, search_by_key_restarted);
 693                         PROC_INFO_INC(p_s_sb,
 694                                       sbk_restarted[expected_level - 1]);
 695                         decrement_counters_in_path(p_s_search_path);
 696
 697                         /* Get the root block number so that we can repeat the search
 698                            starting from the root. */
 699                         n_block_number = SB_ROOT_BLOCK(p_s_sb);
 700                         expected_level = -1;
 701                         right_neighbor_of_leaf_node = 0;
 702
 703                         /* repeat search from the root */
 704                         continue;
 705                 }
 706
 707                 /* only check that the key is in the buffer if p_s_key is not
 708                    equal to the MAX_KEY. Latter case is only possible in
 709                    "finish_unfinished()" processing during mount. */
 710                 RFALSE(comp_keys(&MAX_KEY, p_s_key) &&
 711                        !key_in_buffer(p_s_search_path, p_s_key, p_s_sb),
 712                        "PAP-5130: key is not in the buffer");
 713 #ifdef CONFIG_REISERFS_CHECK
 714                 if (cur_tb) {
 715                         print_cur_tb("5140");
 716                         reiserfs_panic(p_s_sb,
 717                                        "PAP-5140: search_by_key: schedule occurred in do_balance!");
 718                 }
 719 #endif
 720
 721                 // make sure, that the node contents look like a node of
 722                 // certain level
 723                 if (!is_tree_node(p_s_bh, expected_level)) {
 724                         reiserfs_warning(p_s_sb, "vs-5150: search_by_key: "
 725                                          "invalid format found in block %ld. Fsck?",
 726                                          p_s_bh->b_blocknr);
 727                         pathrelse(p_s_search_path);
 728                         return IO_ERROR;
 729                 }
 730
 731                 /* ok, we have acquired next formatted node in the tree */
 732                 n_node_level = B_LEVEL(p_s_bh);
 733
 734                 PROC_INFO_BH_STAT(p_s_sb, p_s_bh, n_node_level - 1);
 735
 736                 RFALSE(n_node_level < n_stop_level,
 737                        "vs-5152: tree level (%d) is less than stop level (%d)",
 738                        n_node_level, n_stop_level);
 739
 740                 n_retval = bin_search(p_s_key, B_N_PITEM_HEAD(p_s_bh, 0),
 741                                       B_NR_ITEMS(p_s_bh),
 742                                       (n_node_level ==
 743                                        DISK_LEAF_NODE_LEVEL) ? IH_SIZE :
 744                                       KEY_SIZE,
 745                                       &(p_s_last_element->pe_position));
 746                 if (n_node_level == n_stop_level) {
 747                         return n_retval;
 748                 }
 749
 750                 /* we are not in the stop level */
 751                 if (n_retval == ITEM_FOUND)
 752                         /* item has been found, so we choose the pointer which is to the right of the found one */
 753                         p_s_last_element->pe_position++;
 754
 755                 /* if item was not found we choose the position which is to
 756                    the left of the found item. This requires no code,
 757                    bin_search did it already. */
 758
 759                 /* So we have chosen a position in the current node which is
 760                    an internal node.  Now we calculate child block number by
 761                    position in the node. */
 762                 n_block_number =
 763                     B_N_CHILD_NUM(p_s_bh, p_s_last_element->pe_position);
 764
 765                 /* if we are going to read leaf nodes, try for read ahead as well */
 766                 if ((p_s_search_path->reada & PATH_READA) &&
 767                     n_node_level == DISK_LEAF_NODE_LEVEL + 1) {
 768                         int pos = p_s_last_element->pe_position;
 769                         int limit = B_NR_ITEMS(p_s_bh);
 770                         struct reiserfs_key *le_key;
 771
 772                         if (p_s_search_path->reada & PATH_READA_BACK)
 773                                 limit = 0;
 774                         while (reada_count < SEARCH_BY_KEY_READA) {
 775                                 if (pos == limit)
 776                                         break;
 777                                 reada_blocks[reada_count++] =
 778                                     B_N_CHILD_NUM(p_s_bh, pos);
 779                                 if (p_s_search_path->reada & PATH_READA_BACK)
 780                                         pos--;
 781                                 else
 782                                         pos++;
 783
 784                                 /*
 785                                  * check to make sure we're in the same object
 786                                  */
 787                                 le_key = B_N_PDELIM_KEY(p_s_bh, pos);
 788                                 if (le32_to_cpu(le_key->k_objectid) !=
 789                                     p_s_key->on_disk_key.k_objectid) {
 790                                         break;
 791                                 }
 792                         }
 793                 }
 794         }
 795 }
 796
 797 /* Form the path to an item and position in this item which contains
 798    file byte defined by p_s_key. If there is no such item
 799    corresponding to the key, we point the path to the item with
 800    maximal key less than p_s_key, and *p_n_pos_in_item is set to one
 801    past the last entry/byte in the item.  If searching for entry in a
 802    directory item, and it is not found, *p_n_pos_in_item is set to one
 803    entry more than the entry with maximal key which is less than the
 804    sought key.
 805
 806    Note that if there is no entry in this same node which is one more,
 807    then we point to an imaginary entry.  for direct items, the
 808    position is in units of bytes, for indirect items the position is
 809    in units of blocknr entries, for directory items the position is in
 810    units of directory entries.  */
 811
 812 /* The function is NOT SCHEDULE-SAFE! */
 813 int search_for_position_by_key(struct super_block *p_s_sb,      /* Pointer to the super block.          */
 814                                const struct cpu_key *p_cpu_key, /* Key to search (cpu variable)         */
 815                                struct treepath *p_s_search_path /* Filled up by this function.          */
 816     )
 817 {
 818         struct item_head *p_le_ih;      /* pointer to on-disk structure */
 819         int n_blk_size;
 820         loff_t item_offset, offset;
 821         struct reiserfs_dir_entry de;
 822         int retval;
 823
 824         /* If searching for directory entry. */
 825         if (is_direntry_cpu_key(p_cpu_key))
 826                 return search_by_entry_key(p_s_sb, p_cpu_key, p_s_search_path,
 827                                            &de);
 828
 829         /* If not searching for directory entry. */
 830
 831         /* If item is found. */
 832         retval = search_item(p_s_sb, p_cpu_key, p_s_search_path);
 833         if (retval == IO_ERROR)
 834                 return retval;
 835         if (retval == ITEM_FOUND) {
 836
 837                 RFALSE(!ih_item_len
 838                        (B_N_PITEM_HEAD
 839                         (PATH_PLAST_BUFFER(p_s_search_path),
 840                          PATH_LAST_POSITION(p_s_search_path))),
 841                        "PAP-5165: item length equals zero");
 842
 843                 pos_in_item(p_s_search_path) = 0;
 844                 return POSITION_FOUND;
 845         }
 846
 847         RFALSE(!PATH_LAST_POSITION(p_s_search_path),
 848                "PAP-5170: position equals zero");
 849
 850         /* Item is not found. Set path to the previous item. */
 851         p_le_ih =
 852             B_N_PITEM_HEAD(PATH_PLAST_BUFFER(p_s_search_path),
 853                            --PATH_LAST_POSITION(p_s_search_path));
 854         n_blk_size = p_s_sb->s_blocksize;
 855
 856         if (comp_short_keys(&(p_le_ih->ih_key), p_cpu_key)) {
 857                 return FILE_NOT_FOUND;
 858         }
 859         // FIXME: quite ugly this far
 860
 861         item_offset = le_ih_k_offset(p_le_ih);
 862         offset = cpu_key_k_offset(p_cpu_key);
 863
 864         /* Needed byte is contained in the item pointed to by the path. */
 865         if (item_offset <= offset &&
 866             item_offset + op_bytes_number(p_le_ih, n_blk_size) > offset) {
 867                 pos_in_item(p_s_search_path) = offset - item_offset;
 868                 if (is_indirect_le_ih(p_le_ih)) {
 869                         pos_in_item(p_s_search_path) /= n_blk_size;
 870                 }
 871                 return POSITION_FOUND;
 872         }
 873
 874         /* Needed byte is not contained in the item pointed to by the
 875            path. Set pos_in_item out of the item. */
 876         if (is_indirect_le_ih(p_le_ih))
 877                 pos_in_item(p_s_search_path) =
 878                     ih_item_len(p_le_ih) / UNFM_P_SIZE;
 879         else
 880                 pos_in_item(p_s_search_path) = ih_item_len(p_le_ih);
 881
 882         return POSITION_NOT_FOUND;
 883 }
 884
 885 /* Compare given item and item pointed to by the path. */
 886 int comp_items(const struct item_head *stored_ih, const struct treepath *p_s_path)
 887 {
 888         struct buffer_head *p_s_bh;
 889         struct item_head *ih;
 890
 891         /* Last buffer at the path is not in the tree. */
 892         if (!B_IS_IN_TREE(p_s_bh = PATH_PLAST_BUFFER(p_s_path)))
 893                 return 1;
 894
 895         /* Last path position is invalid. */
 896         if (PATH_LAST_POSITION(p_s_path) >= B_NR_ITEMS(p_s_bh))
 897                 return 1;
 898
 899         /* we need only to know, whether it is the same item */
 900         ih = get_ih(p_s_path);
 901         return memcmp(stored_ih, ih, IH_SIZE);
 902 }
 903
 904 /* unformatted nodes are not logged anymore, ever.  This is safe
 905 ** now
 906 */
 907 #define held_by_others(bh) (atomic_read(&(bh)->b_count) > 1)
 908
 909 // block can not be forgotten as it is in I/O or held by someone
 910 #define block_in_use(bh) (buffer_locked(bh) || (held_by_others(bh)))
 911
 912 // prepare for delete or cut of direct item
 913 static inline int prepare_for_direct_item(struct treepath *path,
 914                                           struct item_head *le_ih,
 915                                           struct inode *inode,
 916                                           loff_t new_file_length, int *cut_size)
 917 {
 918         loff_t round_len;
 919
 920         if (new_file_length == max_reiserfs_offset(inode)) {
 921                 /* item has to be deleted */
 922                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 923                 return M_DELETE;
 924         }
 925         // new file gets truncated
 926         if (get_inode_item_key_version(inode) == KEY_FORMAT_3_6) {
 927                 //
 928                 round_len = ROUND_UP(new_file_length);
 929                 /* this was n_new_file_length < le_ih ... */
 930                 if (round_len < le_ih_k_offset(le_ih)) {
 931                         *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 932                         return M_DELETE;        /* Delete this item. */
 933                 }
 934                 /* Calculate first position and size for cutting from item. */
 935                 pos_in_item(path) = round_len - (le_ih_k_offset(le_ih) - 1);
 936                 *cut_size = -(ih_item_len(le_ih) - pos_in_item(path));
 937
 938                 return M_CUT;   /* Cut from this item. */
 939         }
 940
 941         // old file: items may have any length
 942
 943         if (new_file_length < le_ih_k_offset(le_ih)) {
 944                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 945                 return M_DELETE;        /* Delete this item. */
 946         }
 947         /* Calculate first position and size for cutting from item. */
 948         *cut_size = -(ih_item_len(le_ih) -
 949                       (pos_in_item(path) =
 950                        new_file_length + 1 - le_ih_k_offset(le_ih)));
 951         return M_CUT;           /* Cut from this item. */
 952 }
 953
 954 static inline int prepare_for_direntry_item(struct treepath *path,
 955                                             struct item_head *le_ih,
 956                                             struct inode *inode,
 957                                             loff_t new_file_length,
 958                                             int *cut_size)
 959 {
 960         if (le_ih_k_offset(le_ih) == DOT_OFFSET &&
 961             new_file_length == max_reiserfs_offset(inode)) {
 962                 RFALSE(ih_entry_count(le_ih) != 2,
 963                        "PAP-5220: incorrect empty directory item (%h)", le_ih);
 964                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 965                 return M_DELETE;        /* Delete the directory item containing "." and ".." entry. */
 966         }
 967
 968         if (ih_entry_count(le_ih) == 1) {
 969                 /* Delete the directory item such as there is one record only
 970                    in this item */
 971                 *cut_size = -(IH_SIZE + ih_item_len(le_ih));
 972                 return M_DELETE;
 973         }
 974
 975         /* Cut one record from the directory item. */
 976         *cut_size =
 977             -(DEH_SIZE +
 978               entry_length(get_last_bh(path), le_ih, pos_in_item(path)));
 979         return M_CUT;
 980 }
 981
 982 #define JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD (2 * JOURNAL_PER_BALANCE_CNT + 1)
 983
 984 /*  If the path points to a directory or direct item, calculate mode and the size cut, for balance.
 985     If the path points to an indirect item, remove some number of its unformatted nodes.
 986     In case of file truncate calculate whether this item must be deleted/truncated or last
 987     unformatted node of this item will be converted to a direct item.
 988     This function returns a determination of what balance mode the calling function should employ. */
 989 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
 990                                                                                                                                                                                    from end of the file. */
 991                                       int *p_n_cut_size, unsigned long long n_new_file_length   /* MAX_KEY_OFFSET in case of delete. */
 992     )
 993 {
 994         struct super_block *p_s_sb = inode->i_sb;
 995         struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_path);
 996         struct buffer_head *p_s_bh = PATH_PLAST_BUFFER(p_s_path);
 997
 998         BUG_ON(!th->t_trans_id);
 999
1000         /* Stat_data item. */
1001         if (is_statdata_le_ih(p_le_ih)) {
1002
1003                 RFALSE(n_new_file_length != max_reiserfs_offset(inode),
1004                        "PAP-5210: mode must be M_DELETE");
1005
1006                 *p_n_cut_size = -(IH_SIZE + ih_item_len(p_le_ih));
1007                 return M_DELETE;
1008         }
1009
1010         /* Directory item. */
1011         if (is_direntry_le_ih(p_le_ih))
1012                 return prepare_for_direntry_item(p_s_path, p_le_ih, inode,
1013                                                  n_new_file_length,
1014                                                  p_n_cut_size);
1015
1016         /* Direct item. */
1017         if (is_direct_le_ih(p_le_ih))
1018                 return prepare_for_direct_item(p_s_path, p_le_ih, inode,
1019                                                n_new_file_length, p_n_cut_size);
1020
1021         /* Case of an indirect item. */
1022         {
1023             int blk_size = p_s_sb->s_blocksize;
1024             struct item_head s_ih;
1025             int need_re_search;
1026             int delete = 0;
1027             int result = M_CUT;
1028             int pos = 0;
1029
1030             if ( n_new_file_length == max_reiserfs_offset (inode) ) {
1031                 /* prepare_for_delete_or_cut() is called by
1032                  * reiserfs_delete_item() */
1033                 n_new_file_length = 0;
1034                 delete = 1;
1035             }
1036
1037             do {
1038                 need_re_search = 0;
1039                 *p_n_cut_size = 0;
1040                 p_s_bh = PATH_PLAST_BUFFER(p_s_path);
1041                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1042                 pos = I_UNFM_NUM(&s_ih);
1043
1044                 while (le_ih_k_offset (&s_ih) + (pos - 1) * blk_size > n_new_file_length) {
1045                     __le32 *unfm;
1046                     __u32 block;
1047
1048                     /* Each unformatted block deletion may involve one additional
1049                      * bitmap block into the transaction, thereby the initial
1050                      * journal space reservation might not be enough. */
1051                     if (!delete && (*p_n_cut_size) != 0 &&
1052                         reiserfs_transaction_free_space(th) < JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1053                         break;
1054                     }
1055
1056                     unfm = (__le32 *)B_I_PITEM(p_s_bh, &s_ih) + pos - 1;
1057                     block = get_block_num(unfm, 0);
1058
1059                     if (block != 0) {
1060                         reiserfs_prepare_for_journal(p_s_sb, p_s_bh, 1);
1061                         put_block_num(unfm, 0, 0);
1062                         journal_mark_dirty (th, p_s_sb, p_s_bh);
1063                         reiserfs_free_block(th, inode, block, 1);
1064                     }
1065
1066                     cond_resched();
1067
1068                     if (item_moved (&s_ih, p_s_path))  {
1069                         need_re_search = 1;
1070                         break;
1071                     }
1072
1073                     pos --;
1074                     (*p_n_removed) ++;
1075                     (*p_n_cut_size) -= UNFM_P_SIZE;
1076
1077                     if (pos == 0) {
1078                         (*p_n_cut_size) -= IH_SIZE;
1079                         result = M_DELETE;
1080                         break;
1081                     }
1082                 }
1083                 /* a trick.  If the buffer has been logged, this will do nothing.  If
1084                 ** we've broken the loop without logging it, it will restore the
1085                 ** buffer */
1086                 reiserfs_restore_prepared_buffer(p_s_sb, p_s_bh);
1087             } while (need_re_search &&
1088                      search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path) == POSITION_FOUND);
1089             pos_in_item(p_s_path) = pos * UNFM_P_SIZE;
1090
1091             if (*p_n_cut_size == 0) {
1092                 /* Nothing were cut. maybe convert last unformatted node to the
1093                  * direct item? */
1094                 result = M_CONVERT;
1095             }
1096             return result;
1097         }
1098 }
1099
1100 /* Calculate number of bytes which will be deleted or cut during balance */
1101 static int calc_deleted_bytes_number(struct tree_balance *p_s_tb, char c_mode)
1102 {
1103         int n_del_size;
1104         struct item_head *p_le_ih = PATH_PITEM_HEAD(p_s_tb->tb_path);
1105
1106         if (is_statdata_le_ih(p_le_ih))
1107                 return 0;
1108
1109         n_del_size =
1110             (c_mode ==
1111              M_DELETE) ? ih_item_len(p_le_ih) : -p_s_tb->insert_size[0];
1112         if (is_direntry_le_ih(p_le_ih)) {
1113                 // return EMPTY_DIR_SIZE; /* We delete emty directoris only. */
1114                 // we can't use EMPTY_DIR_SIZE, as old format dirs have a different
1115                 // empty size.  ick. FIXME, is this right?
1116                 //
1117                 return n_del_size;
1118         }
1119
1120         if (is_indirect_le_ih(p_le_ih))
1121                 n_del_size = (n_del_size / UNFM_P_SIZE) * (PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_size); // - get_ih_free_space (p_le_ih);
1122         return n_del_size;
1123 }
1124
1125 static void init_tb_struct(struct reiserfs_transaction_handle *th,
1126                            struct tree_balance *p_s_tb,
1127                            struct super_block *p_s_sb,
1128                            struct treepath *p_s_path, int n_size)
1129 {
1130
1131         BUG_ON(!th->t_trans_id);
1132
1133         memset(p_s_tb, '\0', sizeof(struct tree_balance));
1134         p_s_tb->transaction_handle = th;
1135         p_s_tb->tb_sb = p_s_sb;
1136         p_s_tb->tb_path = p_s_path;
1137         PATH_OFFSET_PBUFFER(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = NULL;
1138         PATH_OFFSET_POSITION(p_s_path, ILLEGAL_PATH_ELEMENT_OFFSET) = 0;
1139         p_s_tb->insert_size[0] = n_size;
1140 }
1141
1142 void padd_item(char *item, int total_length, int length)
1143 {
1144         int i;
1145
1146         for (i = total_length; i > length;)
1147                 item[--i] = 0;
1148 }
1149
1150 #ifdef REISERQUOTA_DEBUG
1151 char key2type(struct reiserfs_key *ih)
1152 {
1153         if (is_direntry_le_key(2, ih))
1154                 return 'd';
1155         if (is_direct_le_key(2, ih))
1156                 return 'D';
1157         if (is_indirect_le_key(2, ih))
1158                 return 'i';
1159         if (is_statdata_le_key(2, ih))
1160                 return 's';
1161         return 'u';
1162 }
1163
1164 char head2type(struct item_head *ih)
1165 {
1166         if (is_direntry_le_ih(ih))
1167                 return 'd';
1168         if (is_direct_le_ih(ih))
1169                 return 'D';
1170         if (is_indirect_le_ih(ih))
1171                 return 'i';
1172         if (is_statdata_le_ih(ih))
1173                 return 's';
1174         return 'u';
1175 }
1176 #endif
1177
1178 /* Delete object item. */
1179 int reiserfs_delete_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the deleted item. */
1180                          const struct cpu_key *p_s_item_key,    /* Key to search for the deleted item.  */
1181                          struct inode *p_s_inode,       /* inode is here just to update i_blocks and quotas */
1182                          struct buffer_head *p_s_un_bh)
1183 {                               /* NULL or unformatted node pointer.    */
1184         struct super_block *p_s_sb = p_s_inode->i_sb;
1185         struct tree_balance s_del_balance;
1186         struct item_head s_ih;
1187         struct item_head *q_ih;
1188         int quota_cut_bytes;
1189         int n_ret_value, n_del_size, n_removed;
1190
1191 #ifdef CONFIG_REISERFS_CHECK
1192         char c_mode;
1193         int n_iter = 0;
1194 #endif
1195
1196         BUG_ON(!th->t_trans_id);
1197
1198         init_tb_struct(th, &s_del_balance, p_s_sb, p_s_path,
1199                        0 /*size is unknown */ );
1200
1201         while (1) {
1202                 n_removed = 0;
1203
1204 #ifdef CONFIG_REISERFS_CHECK
1205                 n_iter++;
1206                 c_mode =
1207 #endif
1208                     prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1209                                               p_s_item_key, &n_removed,
1210                                               &n_del_size,
1211                                               max_reiserfs_offset(p_s_inode));
1212
1213                 RFALSE(c_mode != M_DELETE, "PAP-5320: mode must be M_DELETE");
1214
1215                 copy_item_head(&s_ih, PATH_PITEM_HEAD(p_s_path));
1216                 s_del_balance.insert_size[0] = n_del_size;
1217
1218                 n_ret_value = fix_nodes(M_DELETE, &s_del_balance, NULL, NULL);
1219                 if (n_ret_value != REPEAT_SEARCH)
1220                         break;
1221
1222                 PROC_INFO_INC(p_s_sb, delete_item_restarted);
1223
1224                 // file system changed, repeat search
1225                 n_ret_value =
1226                     search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1227                 if (n_ret_value == IO_ERROR)
1228                         break;
1229                 if (n_ret_value == FILE_NOT_FOUND) {
1230                         reiserfs_warning(p_s_sb,
1231                                          "vs-5340: reiserfs_delete_item: "
1232                                          "no items of the file %K found",
1233                                          p_s_item_key);
1234                         break;
1235                 }
1236         }                       /* while (1) */
1237
1238         if (n_ret_value != CARRY_ON) {
1239                 unfix_nodes(&s_del_balance);
1240                 return 0;
1241         }
1242         // reiserfs_delete_item returns item length when success
1243         n_ret_value = calc_deleted_bytes_number(&s_del_balance, M_DELETE);
1244         q_ih = get_ih(p_s_path);
1245         quota_cut_bytes = ih_item_len(q_ih);
1246
1247         /* hack so the quota code doesn't have to guess if the file
1248          ** has a tail.  On tail insert, we allocate quota for 1 unformatted node.
1249          ** We test the offset because the tail might have been
1250          ** split into multiple items, and we only want to decrement for
1251          ** the unfm node once
1252          */
1253         if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(q_ih)) {
1254                 if ((le_ih_k_offset(q_ih) & (p_s_sb->s_blocksize - 1)) == 1) {
1255                         quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1256                 } else {
1257                         quota_cut_bytes = 0;
1258                 }
1259         }
1260
1261         if (p_s_un_bh) {
1262                 int off;
1263                 char *data;
1264
1265                 /* We are in direct2indirect conversion, so move tail contents
1266                    to the unformatted node */
1267                 /* note, we do the copy before preparing the buffer because we
1268                  ** don't care about the contents of the unformatted node yet.
1269                  ** the only thing we really care about is the direct item's data
1270                  ** is in the unformatted node.
1271                  **
1272                  ** Otherwise, we would have to call reiserfs_prepare_for_journal on
1273                  ** the unformatted node, which might schedule, meaning we'd have to
1274                  ** loop all the way back up to the start of the while loop.
1275                  **
1276                  ** The unformatted node must be dirtied later on.  We can't be
1277                  ** sure here if the entire tail has been deleted yet.
1278                  **
1279                  ** p_s_un_bh is from the page cache (all unformatted nodes are
1280                  ** from the page cache) and might be a highmem page.  So, we
1281                  ** can't use p_s_un_bh->b_data.
1282                  ** -clm
1283                  */
1284
1285                 data = kmap_atomic(p_s_un_bh->b_page, KM_USER0);
1286                 off = ((le_ih_k_offset(&s_ih) - 1) & (PAGE_CACHE_SIZE - 1));
1287                 memcpy(data + off,
1288                        B_I_PITEM(PATH_PLAST_BUFFER(p_s_path), &s_ih),
1289                        n_ret_value);
1290                 kunmap_atomic(data, KM_USER0);
1291         }
1292         /* Perform balancing after all resources have been collected at once. */
1293         do_balance(&s_del_balance, NULL, NULL, M_DELETE);
1294
1295 #ifdef REISERQUOTA_DEBUG
1296         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
1297                        "reiserquota delete_item(): freeing %u, id=%u type=%c",
1298                        quota_cut_bytes, p_s_inode->i_uid, head2type(&s_ih));
1299 #endif
1300         vfs_dq_free_space_nodirty(p_s_inode, quota_cut_bytes);
1301
1302         /* Return deleted body length */
1303         return n_ret_value;
1304 }
1305
1306 /* Summary Of Mechanisms For Handling Collisions Between Processes:
1307
1308  deletion of the body of the object is performed by iput(), with the
1309  result that if multiple processes are operating on a file, the
1310  deletion of the body of the file is deferred until the last process
1311  that has an open inode performs its iput().
1312
1313  writes and truncates are protected from collisions by use of
1314  semaphores.
1315
1316  creates, linking, and mknod are protected from collisions with other
1317  processes by making the reiserfs_add_entry() the last step in the
1318  creation, and then rolling back all changes if there was a collision.
1319  - Hans
1320 */
1321
1322 /* this deletes item which never gets split */
1323 void reiserfs_delete_solid_item(struct reiserfs_transaction_handle *th,
1324                                 struct inode *inode, struct reiserfs_key *key)
1325 {
1326         struct tree_balance tb;
1327         INITIALIZE_PATH(path);
1328         int item_len = 0;
1329         int tb_init = 0;
1330         struct cpu_key cpu_key;
1331         int retval;
1332         int quota_cut_bytes = 0;
1333
1334         BUG_ON(!th->t_trans_id);
1335
1336         le_key2cpu_key(&cpu_key, key);
1337
1338         while (1) {
1339                 retval = search_item(th->t_super, &cpu_key, &path);
1340                 if (retval == IO_ERROR) {
1341                         reiserfs_warning(th->t_super,
1342                                          "vs-5350: reiserfs_delete_solid_item: "
1343                                          "i/o failure occurred trying to delete %K",
1344                                          &cpu_key);
1345                         break;
1346                 }
1347                 if (retval != ITEM_FOUND) {
1348                         pathrelse(&path);
1349                         // No need for a warning, if there is just no free space to insert '..' item into the newly-created subdir
1350                         if (!
1351                             ((unsigned long long)
1352                              GET_HASH_VALUE(le_key_k_offset
1353                                             (le_key_version(key), key)) == 0
1354                              && (unsigned long long)
1355                              GET_GENERATION_NUMBER(le_key_k_offset
1356                                                    (le_key_version(key),
1357                                                     key)) == 1))
1358                                 reiserfs_warning(th->t_super,
1359                                                  "vs-5355: reiserfs_delete_solid_item: %k not found",
1360                                                  key);
1361                         break;
1362                 }
1363                 if (!tb_init) {
1364                         tb_init = 1;
1365                         item_len = ih_item_len(PATH_PITEM_HEAD(&path));
1366                         init_tb_struct(th, &tb, th->t_super, &path,
1367                                        -(IH_SIZE + item_len));
1368                 }
1369                 quota_cut_bytes = ih_item_len(PATH_PITEM_HEAD(&path));
1370
1371                 retval = fix_nodes(M_DELETE, &tb, NULL, NULL);
1372                 if (retval == REPEAT_SEARCH) {
1373                         PROC_INFO_INC(th->t_super, delete_solid_item_restarted);
1374                         continue;
1375                 }
1376
1377                 if (retval == CARRY_ON) {
1378                         do_balance(&tb, NULL, NULL, M_DELETE);
1379                         if (inode) {    /* Should we count quota for item? (we don't count quotas for save-links) */
1380 #ifdef REISERQUOTA_DEBUG
1381                                 reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
1382                                                "reiserquota delete_solid_item(): freeing %u id=%u type=%c",
1383                                                quota_cut_bytes, inode->i_uid,
1384                                                key2type(key));
1385 #endif
1386                                 vfs_dq_free_space_nodirty(inode,
1387                                                          quota_cut_bytes);
1388                         }
1389                         break;
1390                 }
1391                 // IO_ERROR, NO_DISK_SPACE, etc
1392                 reiserfs_warning(th->t_super,
1393                                  "vs-5360: reiserfs_delete_solid_item: "
1394                                  "could not delete %K due to fix_nodes failure",
1395                                  &cpu_key);
1396                 unfix_nodes(&tb);
1397                 break;
1398         }
1399
1400         reiserfs_check_path(&path);
1401 }
1402
1403 int reiserfs_delete_object(struct reiserfs_transaction_handle *th,
1404                            struct inode *inode)
1405 {
1406         int err;
1407         inode->i_size = 0;
1408         BUG_ON(!th->t_trans_id);
1409
1410         /* for directory this deletes item containing "." and ".." */
1411         err =
1412             reiserfs_do_truncate(th, inode, NULL, 0 /*no timestamp updates */ );
1413         if (err)
1414                 return err;
1415
1416 #if defined( USE_INODE_GENERATION_COUNTER )
1417         if (!old_format_only(th->t_super)) {
1418                 __le32 *inode_generation;
1419
1420                 inode_generation =
1421                     &REISERFS_SB(th->t_super)->s_rs->s_inode_generation;
1422                 le32_add_cpu(inode_generation, 1);
1423         }
1424 /* USE_INODE_GENERATION_COUNTER */
1425 #endif
1426         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1427
1428         return err;
1429 }
1430
1431 static void unmap_buffers(struct page *page, loff_t pos)
1432 {
1433         struct buffer_head *bh;
1434         struct buffer_head *head;
1435         struct buffer_head *next;
1436         unsigned long tail_index;
1437         unsigned long cur_index;
1438
1439         if (page) {
1440                 if (page_has_buffers(page)) {
1441                         tail_index = pos & (PAGE_CACHE_SIZE - 1);
1442                         cur_index = 0;
1443                         head = page_buffers(page);
1444                         bh = head;
1445                         do {
1446                                 next = bh->b_this_page;
1447
1448                                 /* we want to unmap the buffers that contain the tail, and
1449                                  ** all the buffers after it (since the tail must be at the
1450                                  ** end of the file).  We don't want to unmap file data
1451                                  ** before the tail, since it might be dirty and waiting to
1452                                  ** reach disk
1453                                  */
1454                                 cur_index += bh->b_size;
1455                                 if (cur_index > tail_index) {
1456                                         reiserfs_unmap_buffer(bh);
1457                                 }
1458                                 bh = next;
1459                         } while (bh != head);
1460                 }
1461         }
1462 }
1463
1464 static int maybe_indirect_to_direct(struct reiserfs_transaction_handle *th,
1465                                     struct inode *p_s_inode,
1466                                     struct page *page,
1467                                     struct treepath *p_s_path,
1468                                     const struct cpu_key *p_s_item_key,
1469                                     loff_t n_new_file_size, char *p_c_mode)
1470 {
1471         struct super_block *p_s_sb = p_s_inode->i_sb;
1472         int n_block_size = p_s_sb->s_blocksize;
1473         int cut_bytes;
1474         BUG_ON(!th->t_trans_id);
1475         BUG_ON(n_new_file_size != p_s_inode->i_size);
1476
1477         /* the page being sent in could be NULL if there was an i/o error
1478          ** reading in the last block.  The user will hit problems trying to
1479          ** read the file, but for now we just skip the indirect2direct
1480          */
1481         if (atomic_read(&p_s_inode->i_count) > 1 ||
1482             !tail_has_to_be_packed(p_s_inode) ||
1483             !page || (REISERFS_I(p_s_inode)->i_flags & i_nopack_mask)) {
1484                 // leave tail in an unformatted node
1485                 *p_c_mode = M_SKIP_BALANCING;
1486                 cut_bytes =
1487                     n_block_size - (n_new_file_size & (n_block_size - 1));
1488                 pathrelse(p_s_path);
1489                 return cut_bytes;
1490         }
1491         /* Permorm the conversion to a direct_item. */
1492         /*return indirect_to_direct (p_s_inode, p_s_path, p_s_item_key, n_new_file_size, p_c_mode); */
1493         return indirect2direct(th, p_s_inode, page, p_s_path, p_s_item_key,
1494                                n_new_file_size, p_c_mode);
1495 }
1496
1497 /* we did indirect_to_direct conversion. And we have inserted direct
1498    item successesfully, but there were no disk space to cut unfm
1499    pointer being converted. Therefore we have to delete inserted
1500    direct item(s) */
1501 static void indirect_to_direct_roll_back(struct reiserfs_transaction_handle *th,
1502                                          struct inode *inode, struct treepath *path)
1503 {
1504         struct cpu_key tail_key;
1505         int tail_len;
1506         int removed;
1507         BUG_ON(!th->t_trans_id);
1508
1509         make_cpu_key(&tail_key, inode, inode->i_size + 1, TYPE_DIRECT, 4);      // !!!!
1510         tail_key.key_length = 4;
1511
1512         tail_len =
1513             (cpu_key_k_offset(&tail_key) & (inode->i_sb->s_blocksize - 1)) - 1;
1514         while (tail_len) {
1515                 /* look for the last byte of the tail */
1516                 if (search_for_position_by_key(inode->i_sb, &tail_key, path) ==
1517                     POSITION_NOT_FOUND)
1518                         reiserfs_panic(inode->i_sb,
1519                                        "vs-5615: indirect_to_direct_roll_back: found invalid item");
1520                 RFALSE(path->pos_in_item !=
1521                        ih_item_len(PATH_PITEM_HEAD(path)) - 1,
1522                        "vs-5616: appended bytes found");
1523                 PATH_LAST_POSITION(path)--;
1524
1525                 removed =
1526                     reiserfs_delete_item(th, path, &tail_key, inode,
1527                                          NULL /*unbh not needed */ );
1528                 RFALSE(removed <= 0
1529                        || removed > tail_len,
1530                        "vs-5617: there was tail %d bytes, removed item length %d bytes",
1531                        tail_len, removed);
1532                 tail_len -= removed;
1533                 set_cpu_key_k_offset(&tail_key,
1534                                      cpu_key_k_offset(&tail_key) - removed);
1535         }
1536         reiserfs_warning(inode->i_sb,
1537                          "indirect_to_direct_roll_back: indirect_to_direct conversion has been rolled back due to lack of disk space");
1538         //mark_file_without_tail (inode);
1539         mark_inode_dirty(inode);
1540 }
1541
1542 /* (Truncate or cut entry) or delete object item. Returns < 0 on failure */
1543 int reiserfs_cut_from_item(struct reiserfs_transaction_handle *th,
1544                            struct treepath *p_s_path,
1545                            struct cpu_key *p_s_item_key,
1546                            struct inode *p_s_inode,
1547                            struct page *page, loff_t n_new_file_size)
1548 {
1549         struct super_block *p_s_sb = p_s_inode->i_sb;
1550         /* Every function which is going to call do_balance must first
1551            create a tree_balance structure.  Then it must fill up this
1552            structure by using the init_tb_struct and fix_nodes functions.
1553            After that we can make tree balancing. */
1554         struct tree_balance s_cut_balance;
1555         struct item_head *p_le_ih;
1556         int n_cut_size = 0,     /* Amount to be cut. */
1557             n_ret_value = CARRY_ON, n_removed = 0,      /* Number of the removed unformatted nodes. */
1558             n_is_inode_locked = 0;
1559         char c_mode;            /* Mode of the balance. */
1560         int retval2 = -1;
1561         int quota_cut_bytes;
1562         loff_t tail_pos = 0;
1563
1564         BUG_ON(!th->t_trans_id);
1565
1566         init_tb_struct(th, &s_cut_balance, p_s_inode->i_sb, p_s_path,
1567                        n_cut_size);
1568
1569         /* Repeat this loop until we either cut the item without needing
1570            to balance, or we fix_nodes without schedule occurring */
1571         while (1) {
1572                 /* Determine the balance mode, position of the first byte to
1573                    be cut, and size to be cut.  In case of the indirect item
1574                    free unformatted nodes which are pointed to by the cut
1575                    pointers. */
1576
1577                 c_mode =
1578                     prepare_for_delete_or_cut(th, p_s_inode, p_s_path,
1579                                               p_s_item_key, &n_removed,
1580                                               &n_cut_size, n_new_file_size);
1581                 if (c_mode == M_CONVERT) {
1582                         /* convert last unformatted node to direct item or leave
1583                            tail in the unformatted node */
1584                         RFALSE(n_ret_value != CARRY_ON,
1585                                "PAP-5570: can not convert twice");
1586
1587                         n_ret_value =
1588                             maybe_indirect_to_direct(th, p_s_inode, page,
1589                                                      p_s_path, p_s_item_key,
1590                                                      n_new_file_size, &c_mode);
1591                         if (c_mode == M_SKIP_BALANCING)
1592                                 /* tail has been left in the unformatted node */
1593                                 return n_ret_value;
1594
1595                         n_is_inode_locked = 1;
1596
1597                         /* removing of last unformatted node will change value we
1598                            have to return to truncate. Save it */
1599                         retval2 = n_ret_value;
1600                         /*retval2 = p_s_sb->s_blocksize - (n_new_file_size & (p_s_sb->s_blocksize - 1)); */
1601
1602                         /* So, we have performed the first part of the conversion:
1603                            inserting the new direct item.  Now we are removing the
1604                            last unformatted node pointer. Set key to search for
1605                            it. */
1606                         set_cpu_key_k_type(p_s_item_key, TYPE_INDIRECT);
1607                         p_s_item_key->key_length = 4;
1608                         n_new_file_size -=
1609                             (n_new_file_size & (p_s_sb->s_blocksize - 1));
1610                         tail_pos = n_new_file_size;
1611                         set_cpu_key_k_offset(p_s_item_key, n_new_file_size + 1);
1612                         if (search_for_position_by_key
1613                             (p_s_sb, p_s_item_key,
1614                              p_s_path) == POSITION_NOT_FOUND) {
1615                                 print_block(PATH_PLAST_BUFFER(p_s_path), 3,
1616                                             PATH_LAST_POSITION(p_s_path) - 1,
1617                                             PATH_LAST_POSITION(p_s_path) + 1);
1618                                 reiserfs_panic(p_s_sb,
1619                                                "PAP-5580: reiserfs_cut_from_item: item to convert does not exist (%K)",
1620                                                p_s_item_key);
1621                         }
1622                         continue;
1623                 }
1624                 if (n_cut_size == 0) {
1625                         pathrelse(p_s_path);
1626                         return 0;
1627                 }
1628
1629                 s_cut_balance.insert_size[0] = n_cut_size;
1630
1631                 n_ret_value = fix_nodes(c_mode, &s_cut_balance, NULL, NULL);
1632                 if (n_ret_value != REPEAT_SEARCH)
1633                         break;
1634
1635                 PROC_INFO_INC(p_s_sb, cut_from_item_restarted);
1636
1637                 n_ret_value =
1638                     search_for_position_by_key(p_s_sb, p_s_item_key, p_s_path);
1639                 if (n_ret_value == POSITION_FOUND)
1640                         continue;
1641
1642                 reiserfs_warning(p_s_sb,
1643                                  "PAP-5610: reiserfs_cut_from_item: item %K not found",
1644                                  p_s_item_key);
1645                 unfix_nodes(&s_cut_balance);
1646                 return (n_ret_value == IO_ERROR) ? -EIO : -ENOENT;
1647         }                       /* while */
1648
1649         // check fix_nodes results (IO_ERROR or NO_DISK_SPACE)
1650         if (n_ret_value != CARRY_ON) {
1651                 if (n_is_inode_locked) {
1652                         // FIXME: this seems to be not needed: we are always able
1653                         // to cut item
1654                         indirect_to_direct_roll_back(th, p_s_inode, p_s_path);
1655                 }
1656                 if (n_ret_value == NO_DISK_SPACE)
1657                         reiserfs_warning(p_s_sb, "NO_DISK_SPACE");
1658                 unfix_nodes(&s_cut_balance);
1659                 return -EIO;
1660         }
1661
1662         /* go ahead and perform balancing */
1663
1664         RFALSE(c_mode == M_PASTE || c_mode == M_INSERT, "invalid mode");
1665
1666         /* Calculate number of bytes that need to be cut from the item. */
1667         quota_cut_bytes =
1668             (c_mode ==
1669              M_DELETE) ? ih_item_len(get_ih(p_s_path)) : -s_cut_balance.
1670             insert_size[0];
1671         if (retval2 == -1)
1672                 n_ret_value = calc_deleted_bytes_number(&s_cut_balance, c_mode);
1673         else
1674                 n_ret_value = retval2;
1675
1676         /* For direct items, we only change the quota when deleting the last
1677          ** item.
1678          */
1679         p_le_ih = PATH_PITEM_HEAD(s_cut_balance.tb_path);
1680         if (!S_ISLNK(p_s_inode->i_mode) && is_direct_le_ih(p_le_ih)) {
1681                 if (c_mode == M_DELETE &&
1682                     (le_ih_k_offset(p_le_ih) & (p_s_sb->s_blocksize - 1)) ==
1683                     1) {
1684                         // FIXME: this is to keep 3.5 happy
1685                         REISERFS_I(p_s_inode)->i_first_direct_byte = U32_MAX;
1686                         quota_cut_bytes = p_s_sb->s_blocksize + UNFM_P_SIZE;
1687                 } else {
1688                         quota_cut_bytes = 0;
1689                 }
1690         }
1691 #ifdef CONFIG_REISERFS_CHECK
1692         if (n_is_inode_locked) {
1693                 struct item_head *le_ih =
1694                     PATH_PITEM_HEAD(s_cut_balance.tb_path);
1695                 /* we are going to complete indirect2direct conversion. Make
1696                    sure, that we exactly remove last unformatted node pointer
1697                    of the item */
1698                 if (!is_indirect_le_ih(le_ih))
1699                         reiserfs_panic(p_s_sb,
1700                                        "vs-5652: reiserfs_cut_from_item: "
1701                                        "item must be indirect %h", le_ih);
1702
1703                 if (c_mode == M_DELETE && ih_item_len(le_ih) != UNFM_P_SIZE)
1704                         reiserfs_panic(p_s_sb,
1705                                        "vs-5653: reiserfs_cut_from_item: "
1706                                        "completing indirect2direct conversion indirect item %h "
1707                                        "being deleted must be of 4 byte long",
1708                                        le_ih);
1709
1710                 if (c_mode == M_CUT
1711                     && s_cut_balance.insert_size[0] != -UNFM_P_SIZE) {
1712                         reiserfs_panic(p_s_sb,
1713                                        "vs-5654: reiserfs_cut_from_item: "
1714                                        "can not complete indirect2direct conversion of %h (CUT, insert_size==%d)",
1715                                        le_ih, s_cut_balance.insert_size[0]);
1716                 }
1717                 /* it would be useful to make sure, that right neighboring
1718                    item is direct item of this file */
1719         }
1720 #endif
1721
1722         do_balance(&s_cut_balance, NULL, NULL, c_mode);
1723         if (n_is_inode_locked) {
1724                 /* we've done an indirect->direct conversion.  when the data block
1725                  ** was freed, it was removed from the list of blocks that must
1726                  ** be flushed before the transaction commits, make sure to
1727                  ** unmap and invalidate it
1728                  */
1729                 unmap_buffers(page, tail_pos);
1730                 REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask;
1731         }
1732 #ifdef REISERQUOTA_DEBUG
1733         reiserfs_debug(p_s_inode->i_sb, REISERFS_DEBUG_CODE,
1734                        "reiserquota cut_from_item(): freeing %u id=%u type=%c",
1735                        quota_cut_bytes, p_s_inode->i_uid, '?');
1736 #endif
1737         vfs_dq_free_space_nodirty(p_s_inode, quota_cut_bytes);
1738         return n_ret_value;
1739 }
1740
1741 static void truncate_directory(struct reiserfs_transaction_handle *th,
1742                                struct inode *inode)
1743 {
1744         BUG_ON(!th->t_trans_id);
1745         if (inode->i_nlink)
1746                 reiserfs_warning(inode->i_sb,
1747                                  "vs-5655: truncate_directory: link count != 0");
1748
1749         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), DOT_OFFSET);
1750         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_DIRENTRY);
1751         reiserfs_delete_solid_item(th, inode, INODE_PKEY(inode));
1752         reiserfs_update_sd(th, inode);
1753         set_le_key_k_offset(KEY_FORMAT_3_5, INODE_PKEY(inode), SD_OFFSET);
1754         set_le_key_k_type(KEY_FORMAT_3_5, INODE_PKEY(inode), TYPE_STAT_DATA);
1755 }
1756
1757 /* Truncate file to the new size. Note, this must be called with a transaction
1758    already started */
1759 int reiserfs_do_truncate(struct reiserfs_transaction_handle *th, struct inode *p_s_inode,       /* ->i_size contains new
1760                                                                                                    size */
1761                          struct page *page,     /* up to date for last block */
1762                          int update_timestamps  /* when it is called by
1763                                                    file_release to convert
1764                                                    the tail - no timestamps
1765                                                    should be updated */
1766     )
1767 {
1768         INITIALIZE_PATH(s_search_path); /* Path to the current object item. */
1769         struct item_head *p_le_ih;      /* Pointer to an item header. */
1770         struct cpu_key s_item_key;      /* Key to search for a previous file item. */
1771         loff_t n_file_size,     /* Old file size. */
1772          n_new_file_size;       /* New file size. */
1773         int n_deleted;          /* Number of deleted or truncated bytes. */
1774         int retval;
1775         int err = 0;
1776
1777         BUG_ON(!th->t_trans_id);
1778         if (!
1779             (S_ISREG(p_s_inode->i_mode) || S_ISDIR(p_s_inode->i_mode)
1780              || S_ISLNK(p_s_inode->i_mode)))
1781                 return 0;
1782
1783         if (S_ISDIR(p_s_inode->i_mode)) {
1784                 // deletion of directory - no need to update timestamps
1785                 truncate_directory(th, p_s_inode);
1786                 return 0;
1787         }
1788
1789         /* Get new file size. */
1790         n_new_file_size = p_s_inode->i_size;
1791
1792         // FIXME: note, that key type is unimportant here
1793         make_cpu_key(&s_item_key, p_s_inode, max_reiserfs_offset(p_s_inode),
1794                      TYPE_DIRECT, 3);
1795
1796         retval =
1797             search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1798                                        &s_search_path);
1799         if (retval == IO_ERROR) {
1800                 reiserfs_warning(p_s_inode->i_sb,
1801                                  "vs-5657: reiserfs_do_truncate: "
1802                                  "i/o failure occurred trying to truncate %K",
1803                                  &s_item_key);
1804                 err = -EIO;
1805                 goto out;
1806         }
1807         if (retval == POSITION_FOUND || retval == FILE_NOT_FOUND) {
1808                 reiserfs_warning(p_s_inode->i_sb,
1809                                  "PAP-5660: reiserfs_do_truncate: "
1810                                  "wrong result %d of search for %K", retval,
1811                                  &s_item_key);
1812
1813                 err = -EIO;
1814                 goto out;
1815         }
1816
1817         s_search_path.pos_in_item--;
1818
1819         /* Get real file size (total length of all file items) */
1820         p_le_ih = PATH_PITEM_HEAD(&s_search_path);
1821         if (is_statdata_le_ih(p_le_ih))
1822                 n_file_size = 0;
1823         else {
1824                 loff_t offset = le_ih_k_offset(p_le_ih);
1825                 int bytes =
1826                     op_bytes_number(p_le_ih, p_s_inode->i_sb->s_blocksize);
1827
1828                 /* this may mismatch with real file size: if last direct item
1829                    had no padding zeros and last unformatted node had no free
1830                    space, this file would have this file size */
1831                 n_file_size = offset + bytes - 1;
1832         }
1833         /*
1834          * are we doing a full truncate or delete, if so
1835          * kick in the reada code
1836          */
1837         if (n_new_file_size == 0)
1838                 s_search_path.reada = PATH_READA | PATH_READA_BACK;
1839
1840         if (n_file_size == 0 || n_file_size < n_new_file_size) {
1841                 goto update_and_out;
1842         }
1843
1844         /* Update key to search for the last file item. */
1845         set_cpu_key_k_offset(&s_item_key, n_file_size);
1846
1847         do {
1848                 /* Cut or delete file item. */
1849                 n_deleted =
1850                     reiserfs_cut_from_item(th, &s_search_path, &s_item_key,
1851                                            p_s_inode, page, n_new_file_size);
1852                 if (n_deleted < 0) {
1853                         reiserfs_warning(p_s_inode->i_sb,
1854                                          "vs-5665: reiserfs_do_truncate: reiserfs_cut_from_item failed");
1855                         reiserfs_check_path(&s_search_path);
1856                         return 0;
1857                 }
1858
1859                 RFALSE(n_deleted > n_file_size,
1860                        "PAP-5670: reiserfs_cut_from_item: too many bytes deleted: deleted %d, file_size %lu, item_key %K",
1861                        n_deleted, n_file_size, &s_item_key);
1862
1863                 /* Change key to search the last file item. */
1864                 n_file_size -= n_deleted;
1865
1866                 set_cpu_key_k_offset(&s_item_key, n_file_size);
1867
1868                 /* While there are bytes to truncate and previous file item is presented in the tree. */
1869
1870                 /*
1871                  ** This loop could take a really long time, and could log
1872                  ** many more blocks than a transaction can hold.  So, we do a polite
1873                  ** journal end here, and if the transaction needs ending, we make
1874                  ** sure the file is consistent before ending the current trans
1875                  ** and starting a new one
1876                  */
1877                 if (journal_transaction_should_end(th, 0) ||
1878                     reiserfs_transaction_free_space(th) <= JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD) {
1879                         int orig_len_alloc = th->t_blocks_allocated;
1880                         decrement_counters_in_path(&s_search_path);
1881
1882                         if (update_timestamps) {
1883                                 p_s_inode->i_mtime = p_s_inode->i_ctime =
1884                                     CURRENT_TIME_SEC;
1885                         }
1886                         reiserfs_update_sd(th, p_s_inode);
1887
1888                         err = journal_end(th, p_s_inode->i_sb, orig_len_alloc);
1889                         if (err)
1890                                 goto out;
1891                         err = journal_begin(th, p_s_inode->i_sb,
1892                                             JOURNAL_FOR_FREE_BLOCK_AND_UPDATE_SD + JOURNAL_PER_BALANCE_CNT * 4) ;
1893                         if (err)
1894                                 goto out;
1895                         reiserfs_update_inode_transaction(p_s_inode);
1896                 }
1897         } while (n_file_size > ROUND_UP(n_new_file_size) &&
1898                  search_for_position_by_key(p_s_inode->i_sb, &s_item_key,
1899                                             &s_search_path) == POSITION_FOUND);
1900
1901         RFALSE(n_file_size > ROUND_UP(n_new_file_size),
1902                "PAP-5680: truncate did not finish: new_file_size %Ld, current %Ld, oid %d",
1903                n_new_file_size, n_file_size, s_item_key.on_disk_key.k_objectid);
1904
1905       update_and_out:
1906         if (update_timestamps) {
1907                 // this is truncate, not file closing
1908                 p_s_inode->i_mtime = p_s_inode->i_ctime = CURRENT_TIME_SEC;
1909         }
1910         reiserfs_update_sd(th, p_s_inode);
1911
1912       out:
1913         pathrelse(&s_search_path);
1914         return err;
1915 }
1916
1917 #ifdef CONFIG_REISERFS_CHECK
1918 // this makes sure, that we __append__, not overwrite or add holes
1919 static void check_research_for_paste(struct treepath *path,
1920                                      const struct cpu_key *p_s_key)
1921 {
1922         struct item_head *found_ih = get_ih(path);
1923
1924         if (is_direct_le_ih(found_ih)) {
1925                 if (le_ih_k_offset(found_ih) +
1926                     op_bytes_number(found_ih,
1927                                     get_last_bh(path)->b_size) !=
1928                     cpu_key_k_offset(p_s_key)
1929                     || op_bytes_number(found_ih,
1930                                        get_last_bh(path)->b_size) !=
1931                     pos_in_item(path))
1932                         reiserfs_panic(NULL,
1933                                        "PAP-5720: check_research_for_paste: "
1934                                        "found direct item %h or position (%d) does not match to key %K",
1935                                        found_ih, pos_in_item(path), p_s_key);
1936         }
1937         if (is_indirect_le_ih(found_ih)) {
1938                 if (le_ih_k_offset(found_ih) +
1939                     op_bytes_number(found_ih,
1940                                     get_last_bh(path)->b_size) !=
1941                     cpu_key_k_offset(p_s_key)
1942                     || I_UNFM_NUM(found_ih) != pos_in_item(path)
1943                     || get_ih_free_space(found_ih) != 0)
1944                         reiserfs_panic(NULL,
1945                                        "PAP-5730: check_research_for_paste: "
1946                                        "found indirect item (%h) or position (%d) does not match to key (%K)",
1947                                        found_ih, pos_in_item(path), p_s_key);
1948         }
1949 }
1950 #endif                          /* config reiserfs check */
1951
1952 /* Paste bytes to the existing item. Returns bytes number pasted into the item. */
1953 int reiserfs_paste_into_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_search_path,  /* Path to the pasted item.          */
1954                              const struct cpu_key *p_s_key,     /* Key to search for the needed item. */
1955                              struct inode *inode,       /* Inode item belongs to */
1956                              const char *p_c_body,      /* Pointer to the bytes to paste.    */
1957                              int n_pasted_size)
1958 {                               /* Size of pasted bytes.             */
1959         struct tree_balance s_paste_balance;
1960         int retval;
1961         int fs_gen;
1962
1963         BUG_ON(!th->t_trans_id);
1964
1965         fs_gen = get_generation(inode->i_sb);
1966
1967 #ifdef REISERQUOTA_DEBUG
1968         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
1969                        "reiserquota paste_into_item(): allocating %u id=%u type=%c",
1970                        n_pasted_size, inode->i_uid,
1971                        key2type(&(p_s_key->on_disk_key)));
1972 #endif
1973
1974         if (vfs_dq_alloc_space_nodirty(inode, n_pasted_size)) {
1975                 pathrelse(p_s_search_path);
1976                 return -EDQUOT;
1977         }
1978         init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path,
1979                        n_pasted_size);
1980 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1981         s_paste_balance.key = p_s_key->on_disk_key;
1982 #endif
1983
1984         /* DQUOT_* can schedule, must check before the fix_nodes */
1985         if (fs_changed(fs_gen, inode->i_sb)) {
1986                 goto search_again;
1987         }
1988
1989         while ((retval =
1990                 fix_nodes(M_PASTE, &s_paste_balance, NULL,
1991                           p_c_body)) == REPEAT_SEARCH) {
1992               search_again:
1993                 /* file system changed while we were in the fix_nodes */
1994                 PROC_INFO_INC(th->t_super, paste_into_item_restarted);
1995                 retval =
1996                     search_for_position_by_key(th->t_super, p_s_key,
1997                                                p_s_search_path);
1998                 if (retval == IO_ERROR) {
1999                         retval = -EIO;
2000                         goto error_out;
2001                 }
2002                 if (retval == POSITION_FOUND) {
2003                         reiserfs_warning(inode->i_sb,
2004                                          "PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists",
2005                                          p_s_key);
2006                         retval = -EEXIST;
2007                         goto error_out;
2008                 }
2009 #ifdef CONFIG_REISERFS_CHECK
2010                 check_research_for_paste(p_s_search_path, p_s_key);
2011 #endif
2012         }
2013
2014         /* Perform balancing after all resources are collected by fix_nodes, and
2015            accessing them will not risk triggering schedule. */
2016         if (retval == CARRY_ON) {
2017                 do_balance(&s_paste_balance, NULL /*ih */ , p_c_body, M_PASTE);
2018                 return 0;
2019         }
2020         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2021       error_out:
2022         /* this also releases the path */
2023         unfix_nodes(&s_paste_balance);
2024 #ifdef REISERQUOTA_DEBUG
2025         reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2026                        "reiserquota paste_into_item(): freeing %u id=%u type=%c",
2027                        n_pasted_size, inode->i_uid,
2028                        key2type(&(p_s_key->on_disk_key)));
2029 #endif
2030         vfs_dq_free_space_nodirty(inode, n_pasted_size);
2031         return retval;
2032 }
2033
2034 /* Insert new item into the buffer at the path. */
2035 int reiserfs_insert_item(struct reiserfs_transaction_handle *th, struct treepath *p_s_path,     /* Path to the inserteded item.         */
2036                          const struct cpu_key *key, struct item_head *p_s_ih,   /* Pointer to the item header to insert. */
2037                          struct inode *inode, const char *p_c_body)
2038 {                               /* Pointer to the bytes to insert.      */
2039         struct tree_balance s_ins_balance;
2040         int retval;
2041         int fs_gen = 0;
2042         int quota_bytes = 0;
2043
2044         BUG_ON(!th->t_trans_id);
2045
2046         if (inode) {            /* Do we count quotas for item? */
2047                 fs_gen = get_generation(inode->i_sb);
2048                 quota_bytes = ih_item_len(p_s_ih);
2049
2050                 /* hack so the quota code doesn't have to guess if the file has
2051                  ** a tail, links are always tails, so there's no guessing needed
2052                  */
2053                 if (!S_ISLNK(inode->i_mode) && is_direct_le_ih(p_s_ih)) {
2054                         quota_bytes = inode->i_sb->s_blocksize + UNFM_P_SIZE;
2055                 }
2056 #ifdef REISERQUOTA_DEBUG
2057                 reiserfs_debug(inode->i_sb, REISERFS_DEBUG_CODE,
2058                                "reiserquota insert_item(): allocating %u id=%u type=%c",
2059                                quota_bytes, inode->i_uid, head2type(p_s_ih));
2060 #endif
2061                 /* We can't dirty inode here. It would be immediately written but
2062                  * appropriate stat item isn't inserted yet... */
2063                 if (vfs_dq_alloc_space_nodirty(inode, quota_bytes)) {
2064                         pathrelse(p_s_path);
2065                         return -EDQUOT;
2066                 }
2067         }
2068         init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path,
2069                        IH_SIZE + ih_item_len(p_s_ih));
2070 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2071         s_ins_balance.key = key->on_disk_key;
2072 #endif
2073         /* DQUOT_* can schedule, must check to be sure calling fix_nodes is safe */
2074         if (inode && fs_changed(fs_gen, inode->i_sb)) {
2075                 goto search_again;
2076         }
2077
2078         while ((retval =
2079                 fix_nodes(M_INSERT, &s_ins_balance, p_s_ih,
2080                           p_c_body)) == REPEAT_SEARCH) {
2081               search_again:
2082                 /* file system changed while we were in the fix_nodes */
2083                 PROC_INFO_INC(th->t_super, insert_item_restarted);
2084                 retval = search_item(th->t_super, key, p_s_path);
2085                 if (retval == IO_ERROR) {
2086                         retval = -EIO;
2087                         goto error_out;
2088                 }
2089                 if (retval == ITEM_FOUND) {
2090                         reiserfs_warning(th->t_super,
2091                                          "PAP-5760: reiserfs_insert_item: "
2092                                          "key %K already exists in the tree",
2093                                          key);
2094                         retval = -EEXIST;
2095                         goto error_out;
2096                 }
2097         }
2098
2099         /* make balancing after all resources will be collected at a time */
2100         if (retval == CARRY_ON) {
2101                 do_balance(&s_ins_balance, p_s_ih, p_c_body, M_INSERT);
2102                 return 0;
2103         }
2104
2105         retval = (retval == NO_DISK_SPACE) ? -ENOSPC : -EIO;
2106       error_out:
2107         /* also releases the path */
2108         unfix_nodes(&s_ins_balance);
2109 #ifdef REISERQUOTA_DEBUG
2110         reiserfs_debug(th->t_super, REISERFS_DEBUG_CODE,
2111                        "reiserquota insert_item(): freeing %u id=%u type=%c",
2112                        quota_bytes, inode->i_uid, head2type(p_s_ih));
2113 #endif
2114         if (inode)
2115                 vfs_dq_free_space_nodirty(inode, quota_bytes);
2116         return retval;
2117 }