fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/exportfs.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <asm/uaccess.h>
  15 #include <asm/unaligned.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/mpage.h>
  18 #include <linux/writeback.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/swap.h>
  21
  22 int reiserfs_commit_write(struct file *f, struct page *page,
  23                           unsigned from, unsigned to);
  24 int reiserfs_prepare_write(struct file *f, struct page *page,
  25                            unsigned from, unsigned to);
  26
  27 void reiserfs_delete_inode(struct inode *inode)
  28 {
  29         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  30         int jbegin_count =
  31             JOURNAL_PER_BALANCE_CNT * 2 +
  32             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  33         struct reiserfs_transaction_handle th;
  34         int err;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37
  38         reiserfs_write_lock(inode->i_sb);
  39
  40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  42                 reiserfs_delete_xattrs(inode);
  43
  44                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  45                         goto out;
  46                 reiserfs_update_inode_transaction(inode);
  47
  48                 reiserfs_discard_prealloc(&th, inode);
  49
  50                 err = reiserfs_delete_object(&th, inode);
  51
  52                 /* Do quota update inside a transaction for journaled quotas. We must do that
  53                  * after delete_object so that quota updates go into the same transaction as
  54                  * stat data deletion */
  55                 if (!err)
  56                         vfs_dq_free_inode(inode);
  57
  58                 if (journal_end(&th, inode->i_sb, jbegin_count))
  59                         goto out;
  60
  61                 /* check return value from reiserfs_delete_object after
  62                  * ending the transaction
  63                  */
  64                 if (err)
  65                     goto out;
  66
  67                 /* all items of file are deleted, so we can remove "save" link */
  68                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  69                                                                  * about an error here */
  70         } else {
  71                 /* no object items are in the tree */
  72                 ;
  73         }
  74       out:
  75         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  76         inode->i_blocks = 0;
  77         reiserfs_write_unlock(inode->i_sb);
  78 }
  79
  80 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  81                           __u32 objectid, loff_t offset, int type, int length)
  82 {
  83         key->version = version;
  84
  85         key->on_disk_key.k_dir_id = dirid;
  86         key->on_disk_key.k_objectid = objectid;
  87         set_cpu_key_k_offset(key, offset);
  88         set_cpu_key_k_type(key, type);
  89         key->key_length = length;
  90 }
  91
  92 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  93    offset and type of key */
  94 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  95                   int type, int length)
  96 {
  97         _make_cpu_key(key, get_inode_item_key_version(inode),
  98                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
  99                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 100                       length);
 101 }
 102
 103 //
 104 // when key is 0, do not set version and short key
 105 //
 106 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 107                               int version,
 108                               loff_t offset, int type, int length,
 109                               int entry_count /*or ih_free_space */ )
 110 {
 111         if (key) {
 112                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 113                 ih->ih_key.k_objectid =
 114                     cpu_to_le32(key->on_disk_key.k_objectid);
 115         }
 116         put_ih_version(ih, version);
 117         set_le_ih_k_offset(ih, offset);
 118         set_le_ih_k_type(ih, type);
 119         put_ih_item_len(ih, length);
 120         /*    set_ih_free_space (ih, 0); */
 121         // for directory items it is entry count, for directs and stat
 122         // datas - 0xffff, for indirects - 0
 123         put_ih_entry_count(ih, entry_count);
 124 }
 125
 126 //
 127 // FIXME: we might cache recently accessed indirect item
 128
 129 // Ugh.  Not too eager for that....
 130 //  I cut the code until such time as I see a convincing argument (benchmark).
 131 // I don't want a bloated inode struct..., and I don't like code complexity....
 132
 133 /* cutting the code is fine, since it really isn't in use yet and is easy
 134 ** to add back in.  But, Vladimir has a really good idea here.  Think
 135 ** about what happens for reading a file.  For each page,
 136 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 137 ** an indirect item.  This indirect item has X number of pointers, where
 138 ** X is a big number if we've done the block allocation right.  But,
 139 ** we only use one or two of these pointers during each call to readpage,
 140 ** needlessly researching again later on.
 141 **
 142 ** The size of the cache could be dynamic based on the size of the file.
 143 **
 144 ** I'd also like to see us cache the location the stat data item, since
 145 ** we are needlessly researching for that frequently.
 146 **
 147 ** --chris
 148 */
 149
 150 /* If this page has a file tail in it, and
 151 ** it was read in by get_block_create_0, the page data is valid,
 152 ** but tail is still sitting in a direct item, and we can't write to
 153 ** it.  So, look through this page, and check all the mapped buffers
 154 ** to make sure they have valid block numbers.  Any that don't need
 155 ** to be unmapped, so that block_prepare_write will correctly call
 156 ** reiserfs_get_block to convert the tail into an unformatted node
 157 */
 158 static inline void fix_tail_page_for_writing(struct page *page)
 159 {
 160         struct buffer_head *head, *next, *bh;
 161
 162         if (page && page_has_buffers(page)) {
 163                 head = page_buffers(page);
 164                 bh = head;
 165                 do {
 166                         next = bh->b_this_page;
 167                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 168                                 reiserfs_unmap_buffer(bh);
 169                         }
 170                         bh = next;
 171                 } while (bh != head);
 172         }
 173 }
 174
 175 /* reiserfs_get_block does not need to allocate a block only if it has been
 176    done already or non-hole position has been found in the indirect item */
 177 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 178                                     struct item_head *ih,
 179                                     __le32 * item, int pos_in_item)
 180 {
 181         if (allocated)
 182                 return 0;
 183         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 184             get_block_num(item, pos_in_item))
 185                 return 0;
 186         return 1;
 187 }
 188
 189 static inline int indirect_item_found(int retval, struct item_head *ih)
 190 {
 191         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 192 }
 193
 194 static inline void set_block_dev_mapped(struct buffer_head *bh,
 195                                         b_blocknr_t block, struct inode *inode)
 196 {
 197         map_bh(bh, inode->i_sb, block);
 198 }
 199
 200 //
 201 // files which were created in the earlier version can not be longer,
 202 // than 2 gb
 203 //
 204 static int file_capable(struct inode *inode, sector_t block)
 205 {
 206         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 207             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 208                 return 1;
 209
 210         return 0;
 211 }
 212
 213 static int restart_transaction(struct reiserfs_transaction_handle *th,
 214                                struct inode *inode, struct treepath *path)
 215 {
 216         struct super_block *s = th->t_super;
 217         int len = th->t_blocks_allocated;
 218         int err;
 219
 220         BUG_ON(!th->t_trans_id);
 221         BUG_ON(!th->t_refcount);
 222
 223         pathrelse(path);
 224
 225         /* we cannot restart while nested */
 226         if (th->t_refcount > 1) {
 227                 return 0;
 228         }
 229         reiserfs_update_sd(th, inode);
 230         err = journal_end(th, s, len);
 231         if (!err) {
 232                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 233                 if (!err)
 234                         reiserfs_update_inode_transaction(inode);
 235         }
 236         return err;
 237 }
 238
 239 // it is called by get_block when create == 0. Returns block number
 240 // for 'block'-th logical block of file. When it hits direct item it
 241 // returns 0 (being called from bmap) or read direct item into piece
 242 // of page (bh_result)
 243
 244 // Please improve the english/clarity in the comment above, as it is
 245 // hard to understand.
 246
 247 static int _get_block_create_0(struct inode *inode, sector_t block,
 248                                struct buffer_head *bh_result, int args)
 249 {
 250         INITIALIZE_PATH(path);
 251         struct cpu_key key;
 252         struct buffer_head *bh;
 253         struct item_head *ih, tmp_ih;
 254         b_blocknr_t blocknr;
 255         char *p = NULL;
 256         int chars;
 257         int ret;
 258         int result;
 259         int done = 0;
 260         unsigned long offset;
 261
 262         // prepare the key to look for the 'block'-th block of file
 263         make_cpu_key(&key, inode,
 264                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 265                      3);
 266
 267         result = search_for_position_by_key(inode->i_sb, &key, &path);
 268         if (result != POSITION_FOUND) {
 269                 pathrelse(&path);
 270                 if (p)
 271                         kunmap(bh_result->b_page);
 272                 if (result == IO_ERROR)
 273                         return -EIO;
 274                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 275                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 276                 if ((args & GET_BLOCK_NO_HOLE)
 277                     && !PageUptodate(bh_result->b_page)) {
 278                         return -ENOENT;
 279                 }
 280                 return 0;
 281         }
 282         //
 283         bh = get_last_bh(&path);
 284         ih = get_ih(&path);
 285         if (is_indirect_le_ih(ih)) {
 286                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 287
 288                 /* FIXME: here we could cache indirect item or part of it in
 289                    the inode to avoid search_by_key in case of subsequent
 290                    access to file */
 291                 blocknr = get_block_num(ind_item, path.pos_in_item);
 292                 ret = 0;
 293                 if (blocknr) {
 294                         map_bh(bh_result, inode->i_sb, blocknr);
 295                         if (path.pos_in_item ==
 296                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 297                                 set_buffer_boundary(bh_result);
 298                         }
 299                 } else
 300                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 301                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 302                 if ((args & GET_BLOCK_NO_HOLE)
 303                             && !PageUptodate(bh_result->b_page)) {
 304                         ret = -ENOENT;
 305                 }
 306
 307                 pathrelse(&path);
 308                 if (p)
 309                         kunmap(bh_result->b_page);
 310                 return ret;
 311         }
 312         // requested data are in direct item(s)
 313         if (!(args & GET_BLOCK_READ_DIRECT)) {
 314                 // we are called by bmap. FIXME: we can not map block of file
 315                 // when it is stored in direct item(s)
 316                 pathrelse(&path);
 317                 if (p)
 318                         kunmap(bh_result->b_page);
 319                 return -ENOENT;
 320         }
 321
 322         /* if we've got a direct item, and the buffer or page was uptodate,
 323          ** we don't want to pull data off disk again.  skip to the
 324          ** end, where we map the buffer and return
 325          */
 326         if (buffer_uptodate(bh_result)) {
 327                 goto finished;
 328         } else
 329                 /*
 330                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 331                  ** pages without any buffers.  If the page is up to date, we don't want
 332                  ** read old data off disk.  Set the up to date bit on the buffer instead
 333                  ** and jump to the end
 334                  */
 335         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 336                 set_buffer_uptodate(bh_result);
 337                 goto finished;
 338         }
 339         // read file tail into part of page
 340         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 341         copy_item_head(&tmp_ih, ih);
 342
 343         /* we only want to kmap if we are reading the tail into the page.
 344          ** this is not the common case, so we don't kmap until we are
 345          ** sure we need to.  But, this means the item might move if
 346          ** kmap schedules
 347          */
 348         if (!p)
 349                 p = (char *)kmap(bh_result->b_page);
 350
 351         p += offset;
 352         memset(p, 0, inode->i_sb->s_blocksize);
 353         do {
 354                 if (!is_direct_le_ih(ih)) {
 355                         BUG();
 356                 }
 357                 /* make sure we don't read more bytes than actually exist in
 358                  ** the file.  This can happen in odd cases where i_size isn't
 359                  ** correct, and when direct item padding results in a few
 360                  ** extra bytes at the end of the direct item
 361                  */
 362                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 363                         break;
 364                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 365                         chars =
 366                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 367                             path.pos_in_item;
 368                         done = 1;
 369                 } else {
 370                         chars = ih_item_len(ih) - path.pos_in_item;
 371                 }
 372                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 373
 374                 if (done)
 375                         break;
 376
 377                 p += chars;
 378
 379                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 380                         // we done, if read direct item is not the last item of
 381                         // node FIXME: we could try to check right delimiting key
 382                         // to see whether direct item continues in the right
 383                         // neighbor or rely on i_size
 384                         break;
 385
 386                 // update key to look for the next piece
 387                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 388                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 389                 if (result != POSITION_FOUND)
 390                         // i/o error most likely
 391                         break;
 392                 bh = get_last_bh(&path);
 393                 ih = get_ih(&path);
 394         } while (1);
 395
 396         flush_dcache_page(bh_result->b_page);
 397         kunmap(bh_result->b_page);
 398
 399       finished:
 400         pathrelse(&path);
 401
 402         if (result == IO_ERROR)
 403                 return -EIO;
 404
 405         /* this buffer has valid data, but isn't valid for io.  mapping it to
 406          * block #0 tells the rest of reiserfs it just has a tail in it
 407          */
 408         map_bh(bh_result, inode->i_sb, 0);
 409         set_buffer_uptodate(bh_result);
 410         return 0;
 411 }
 412
 413 // this is called to create file map. So, _get_block_create_0 will not
 414 // read direct item
 415 static int reiserfs_bmap(struct inode *inode, sector_t block,
 416                          struct buffer_head *bh_result, int create)
 417 {
 418         if (!file_capable(inode, block))
 419                 return -EFBIG;
 420
 421         reiserfs_write_lock(inode->i_sb);
 422         /* do not read the direct item */
 423         _get_block_create_0(inode, block, bh_result, 0);
 424         reiserfs_write_unlock(inode->i_sb);
 425         return 0;
 426 }
 427
 428 /* special version of get_block that is only used by grab_tail_page right
 429 ** now.  It is sent to block_prepare_write, and when you try to get a
 430 ** block past the end of the file (or a block from a hole) it returns
 431 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 432 ** be able to do i/o on the buffers returned, unless an error value
 433 ** is also returned.
 434 **
 435 ** So, this allows block_prepare_write to be used for reading a single block
 436 ** in a page.  Where it does not produce a valid page for holes, or past the
 437 ** end of the file.  This turns out to be exactly what we need for reading
 438 ** tails for conversion.
 439 **
 440 ** The point of the wrapper is forcing a certain value for create, even
 441 ** though the VFS layer is calling this function with create==1.  If you
 442 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 443 ** don't use this function.
 444 */
 445 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 446                                        struct buffer_head *bh_result,
 447                                        int create)
 448 {
 449         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 450 }
 451
 452 /* This is special helper for reiserfs_get_block in case we are executing
 453    direct_IO request. */
 454 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 455                                          sector_t iblock,
 456                                          struct buffer_head *bh_result,
 457                                          int create)
 458 {
 459         int ret;
 460
 461         bh_result->b_page = NULL;
 462
 463         /* We set the b_size before reiserfs_get_block call since it is
 464            referenced in convert_tail_for_hole() that may be called from
 465            reiserfs_get_block() */
 466         bh_result->b_size = (1 << inode->i_blkbits);
 467
 468         ret = reiserfs_get_block(inode, iblock, bh_result,
 469                                  create | GET_BLOCK_NO_DANGLE);
 470         if (ret)
 471                 goto out;
 472
 473         /* don't allow direct io onto tail pages */
 474         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 475                 /* make sure future calls to the direct io funcs for this offset
 476                  ** in the file fail by unmapping the buffer
 477                  */
 478                 clear_buffer_mapped(bh_result);
 479                 ret = -EINVAL;
 480         }
 481         /* Possible unpacked tail. Flush the data before pages have
 482            disappeared */
 483         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 484                 int err;
 485
 486                 reiserfs_write_lock(inode->i_sb);
 487
 488                 err = reiserfs_commit_for_inode(inode);
 489                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 490
 491                 reiserfs_write_unlock(inode->i_sb);
 492
 493                 if (err < 0)
 494                         ret = err;
 495         }
 496       out:
 497         return ret;
 498 }
 499
 500 /*
 501 ** helper function for when reiserfs_get_block is called for a hole
 502 ** but the file tail is still in a direct item
 503 ** bh_result is the buffer head for the hole
 504 ** tail_offset is the offset of the start of the tail in the file
 505 **
 506 ** This calls prepare_write, which will start a new transaction
 507 ** you should not be in a transaction, or have any paths held when you
 508 ** call this.
 509 */
 510 static int convert_tail_for_hole(struct inode *inode,
 511                                  struct buffer_head *bh_result,
 512                                  loff_t tail_offset)
 513 {
 514         unsigned long index;
 515         unsigned long tail_end;
 516         unsigned long tail_start;
 517         struct page *tail_page;
 518         struct page *hole_page = bh_result->b_page;
 519         int retval = 0;
 520
 521         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 522                 return -EIO;
 523
 524         /* always try to read until the end of the block */
 525         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 526         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 527
 528         index = tail_offset >> PAGE_CACHE_SHIFT;
 529         /* hole_page can be zero in case of direct_io, we are sure
 530            that we cannot get here if we write with O_DIRECT into
 531            tail page */
 532         if (!hole_page || index != hole_page->index) {
 533                 tail_page = grab_cache_page(inode->i_mapping, index);
 534                 retval = -ENOMEM;
 535                 if (!tail_page) {
 536                         goto out;
 537                 }
 538         } else {
 539                 tail_page = hole_page;
 540         }
 541
 542         /* we don't have to make sure the conversion did not happen while
 543          ** we were locking the page because anyone that could convert
 544          ** must first take i_mutex.
 545          **
 546          ** We must fix the tail page for writing because it might have buffers
 547          ** that are mapped, but have a block number of 0.  This indicates tail
 548          ** data that has been read directly into the page, and block_prepare_write
 549          ** won't trigger a get_block in this case.
 550          */
 551         fix_tail_page_for_writing(tail_page);
 552         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 553         if (retval)
 554                 goto unlock;
 555
 556         /* tail conversion might change the data in the page */
 557         flush_dcache_page(tail_page);
 558
 559         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 560
 561       unlock:
 562         if (tail_page != hole_page) {
 563                 unlock_page(tail_page);
 564                 page_cache_release(tail_page);
 565         }
 566       out:
 567         return retval;
 568 }
 569
 570 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 571                                   sector_t block,
 572                                   struct inode *inode,
 573                                   b_blocknr_t * allocated_block_nr,
 574                                   struct treepath *path, int flags)
 575 {
 576         BUG_ON(!th->t_trans_id);
 577
 578 #ifdef REISERFS_PREALLOCATE
 579         if (!(flags & GET_BLOCK_NO_IMUX)) {
 580                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 581                                                   path, block);
 582         }
 583 #endif
 584         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 585                                          block);
 586 }
 587
 588 int reiserfs_get_block(struct inode *inode, sector_t block,
 589                        struct buffer_head *bh_result, int create)
 590 {
 591         int repeat, retval = 0;
 592         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 593         INITIALIZE_PATH(path);
 594         int pos_in_item;
 595         struct cpu_key key;
 596         struct buffer_head *bh, *unbh = NULL;
 597         struct item_head *ih, tmp_ih;
 598         __le32 *item;
 599         int done;
 600         int fs_gen;
 601         int lock_depth;
 602         struct reiserfs_transaction_handle *th = NULL;
 603         /* space reserved in transaction batch:
 604            . 3 balancings in direct->indirect conversion
 605            . 1 block involved into reiserfs_update_sd()
 606            XXX in practically impossible worst case direct2indirect()
 607            can incur (much) more than 3 balancings.
 608            quota update for user, group */
 609         int jbegin_count =
 610             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 611             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 612         int version;
 613         int dangle = 1;
 614         loff_t new_offset =
 615             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 616
 617         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 618         version = get_inode_item_key_version(inode);
 619
 620         if (!file_capable(inode, block)) {
 621                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 622                 return -EFBIG;
 623         }
 624
 625         /* if !create, we aren't changing the FS, so we don't need to
 626          ** log anything, so we don't need to start a transaction
 627          */
 628         if (!(create & GET_BLOCK_CREATE)) {
 629                 int ret;
 630                 /* find number of block-th logical block of the file */
 631                 ret = _get_block_create_0(inode, block, bh_result,
 632                                           create | GET_BLOCK_READ_DIRECT);
 633                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 634                 return ret;
 635         }
 636         /*
 637          * if we're already in a transaction, make sure to close
 638          * any new transactions we start in this func
 639          */
 640         if ((create & GET_BLOCK_NO_DANGLE) ||
 641             reiserfs_transaction_running(inode->i_sb))
 642                 dangle = 0;
 643
 644         /* If file is of such a size, that it might have a tail and tails are enabled
 645          ** we should mark it as possibly needing tail packing on close
 646          */
 647         if ((have_large_tails(inode->i_sb)
 648              && inode->i_size < i_block_size(inode) * 4)
 649             || (have_small_tails(inode->i_sb)
 650                 && inode->i_size < i_block_size(inode)))
 651                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 652
 653         /* set the key of the first byte in the 'block'-th block of file */
 654         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 655         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 656               start_trans:
 657                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 658                 if (!th) {
 659                         retval = -ENOMEM;
 660                         goto failure;
 661                 }
 662                 reiserfs_update_inode_transaction(inode);
 663         }
 664       research:
 665
 666         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 667         if (retval == IO_ERROR) {
 668                 retval = -EIO;
 669                 goto failure;
 670         }
 671
 672         bh = get_last_bh(&path);
 673         ih = get_ih(&path);
 674         item = get_item(&path);
 675         pos_in_item = path.pos_in_item;
 676
 677         fs_gen = get_generation(inode->i_sb);
 678         copy_item_head(&tmp_ih, ih);
 679
 680         if (allocation_needed
 681             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 682                 /* we have to allocate block for the unformatted node */
 683                 if (!th) {
 684                         pathrelse(&path);
 685                         goto start_trans;
 686                 }
 687
 688                 repeat =
 689                     _allocate_block(th, block, inode, &allocated_block_nr,
 690                                     &path, create);
 691
 692                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 693                         /* restart the transaction to give the journal a chance to free
 694                          ** some blocks.  releases the path, so we have to go back to
 695                          ** research if we succeed on the second try
 696                          */
 697                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 698                         retval = restart_transaction(th, inode, &path);
 699                         if (retval)
 700                                 goto failure;
 701                         repeat =
 702                             _allocate_block(th, block, inode,
 703                                             &allocated_block_nr, NULL, create);
 704
 705                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 706                                 goto research;
 707                         }
 708                         if (repeat == QUOTA_EXCEEDED)
 709                                 retval = -EDQUOT;
 710                         else
 711                                 retval = -ENOSPC;
 712                         goto failure;
 713                 }
 714
 715                 if (fs_changed(fs_gen, inode->i_sb)
 716                     && item_moved(&tmp_ih, &path)) {
 717                         goto research;
 718                 }
 719         }
 720
 721         if (indirect_item_found(retval, ih)) {
 722                 b_blocknr_t unfm_ptr;
 723                 /* 'block'-th block is in the file already (there is
 724                    corresponding cell in some indirect item). But it may be
 725                    zero unformatted node pointer (hole) */
 726                 unfm_ptr = get_block_num(item, pos_in_item);
 727                 if (unfm_ptr == 0) {
 728                         /* use allocated block to plug the hole */
 729                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 730                         if (fs_changed(fs_gen, inode->i_sb)
 731                             && item_moved(&tmp_ih, &path)) {
 732                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 733                                                                  bh);
 734                                 goto research;
 735                         }
 736                         set_buffer_new(bh_result);
 737                         if (buffer_dirty(bh_result)
 738                             && reiserfs_data_ordered(inode->i_sb))
 739                                 reiserfs_add_ordered_list(inode, bh_result);
 740                         put_block_num(item, pos_in_item, allocated_block_nr);
 741                         unfm_ptr = allocated_block_nr;
 742                         journal_mark_dirty(th, inode->i_sb, bh);
 743                         reiserfs_update_sd(th, inode);
 744                 }
 745                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 746                 pathrelse(&path);
 747                 retval = 0;
 748                 if (!dangle && th)
 749                         retval = reiserfs_end_persistent_transaction(th);
 750
 751                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 752
 753                 /* the item was found, so new blocks were not added to the file
 754                  ** there is no need to make sure the inode is updated with this
 755                  ** transaction
 756                  */
 757                 return retval;
 758         }
 759
 760         if (!th) {
 761                 pathrelse(&path);
 762                 goto start_trans;
 763         }
 764
 765         /* desired position is not found or is in the direct item. We have
 766            to append file with holes up to 'block'-th block converting
 767            direct items to indirect one if necessary */
 768         done = 0;
 769         do {
 770                 if (is_statdata_le_ih(ih)) {
 771                         __le32 unp = 0;
 772                         struct cpu_key tmp_key;
 773
 774                         /* indirect item has to be inserted */
 775                         make_le_item_head(&tmp_ih, &key, version, 1,
 776                                           TYPE_INDIRECT, UNFM_P_SIZE,
 777                                           0 /* free_space */ );
 778
 779                         if (cpu_key_k_offset(&key) == 1) {
 780                                 /* we are going to add 'block'-th block to the file. Use
 781                                    allocated block for that */
 782                                 unp = cpu_to_le32(allocated_block_nr);
 783                                 set_block_dev_mapped(bh_result,
 784                                                      allocated_block_nr, inode);
 785                                 set_buffer_new(bh_result);
 786                                 done = 1;
 787                         }
 788                         tmp_key = key;  // ;)
 789                         set_cpu_key_k_offset(&tmp_key, 1);
 790                         PATH_LAST_POSITION(&path)++;
 791
 792                         retval =
 793                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 794                                                  inode, (char *)&unp);
 795                         if (retval) {
 796                                 reiserfs_free_block(th, inode,
 797                                                     allocated_block_nr, 1);
 798                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 799                         }
 800                         //mark_tail_converted (inode);
 801                 } else if (is_direct_le_ih(ih)) {
 802                         /* direct item has to be converted */
 803                         loff_t tail_offset;
 804
 805                         tail_offset =
 806                             ((le_ih_k_offset(ih) -
 807                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 808                         if (tail_offset == cpu_key_k_offset(&key)) {
 809                                 /* direct item we just found fits into block we have
 810                                    to map. Convert it into unformatted node: use
 811                                    bh_result for the conversion */
 812                                 set_block_dev_mapped(bh_result,
 813                                                      allocated_block_nr, inode);
 814                                 unbh = bh_result;
 815                                 done = 1;
 816                         } else {
 817                                 /* we have to padd file tail stored in direct item(s)
 818                                    up to block size and convert it to unformatted
 819                                    node. FIXME: this should also get into page cache */
 820
 821                                 pathrelse(&path);
 822                                 /*
 823                                  * ugly, but we can only end the transaction if
 824                                  * we aren't nested
 825                                  */
 826                                 BUG_ON(!th->t_refcount);
 827                                 if (th->t_refcount == 1) {
 828                                         retval =
 829                                             reiserfs_end_persistent_transaction
 830                                             (th);
 831                                         th = NULL;
 832                                         if (retval)
 833                                                 goto failure;
 834                                 }
 835
 836                                 retval =
 837                                     convert_tail_for_hole(inode, bh_result,
 838                                                           tail_offset);
 839                                 if (retval) {
 840                                         if (retval != -ENOSPC)
 841                                                 reiserfs_error(inode->i_sb,
 842                                                         "clm-6004",
 843                                                         "convert tail failed "
 844                                                         "inode %lu, error %d",
 845                                                         inode->i_ino,
 846                                                         retval);
 847                                         if (allocated_block_nr) {
 848                                                 /* the bitmap, the super, and the stat data == 3 */
 849                                                 if (!th)
 850                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 851                                                 if (th)
 852                                                         reiserfs_free_block(th,
 853                                                                             inode,
 854                                                                             allocated_block_nr,
 855                                                                             1);
 856                                         }
 857                                         goto failure;
 858                                 }
 859                                 goto research;
 860                         }
 861                         retval =
 862                             direct2indirect(th, inode, &path, unbh,
 863                                             tail_offset);
 864                         if (retval) {
 865                                 reiserfs_unmap_buffer(unbh);
 866                                 reiserfs_free_block(th, inode,
 867                                                     allocated_block_nr, 1);
 868                                 goto failure;
 869                         }
 870                         /* it is important the set_buffer_uptodate is done after
 871                          ** the direct2indirect.  The buffer might contain valid
 872                          ** data newer than the data on disk (read by readpage, changed,
 873                          ** and then sent here by writepage).  direct2indirect needs
 874                          ** to know if unbh was already up to date, so it can decide
 875                          ** if the data in unbh needs to be replaced with data from
 876                          ** the disk
 877                          */
 878                         set_buffer_uptodate(unbh);
 879
 880                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 881                            buffer will disappear shortly, so it should not be added to
 882                          */
 883                         if (unbh->b_page) {
 884                                 /* we've converted the tail, so we must
 885                                  ** flush unbh before the transaction commits
 886                                  */
 887                                 reiserfs_add_tail_list(inode, unbh);
 888
 889                                 /* mark it dirty now to prevent commit_write from adding
 890                                  ** this buffer to the inode's dirty buffer list
 891                                  */
 892                                 /*
 893                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 894                                  * It's still atomic, but it sets the page dirty too,
 895                                  * which makes it eligible for writeback at any time by the
 896                                  * VM (which was also the case with __mark_buffer_dirty())
 897                                  */
 898                                 mark_buffer_dirty(unbh);
 899                         }
 900                 } else {
 901                         /* append indirect item with holes if needed, when appending
 902                            pointer to 'block'-th block use block, which is already
 903                            allocated */
 904                         struct cpu_key tmp_key;
 905                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 906                         // one block which is a fastpath
 907                         unp_t *un;
 908                         __u64 max_to_insert =
 909                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 910                             UNFM_P_SIZE;
 911                         __u64 blocks_needed;
 912
 913                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 914                                "vs-804: invalid position for append");
 915                         /* indirect item has to be appended, set up key of that position */
 916                         make_cpu_key(&tmp_key, inode,
 917                                      le_key_k_offset(version,
 918                                                      &(ih->ih_key)) +
 919                                      op_bytes_number(ih,
 920                                                      inode->i_sb->s_blocksize),
 921                                      //pos_in_item * inode->i_sb->s_blocksize,
 922                                      TYPE_INDIRECT, 3); // key type is unimportant
 923
 924                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 925                                "green-805: invalid offset");
 926                         blocks_needed =
 927                             1 +
 928                             ((cpu_key_k_offset(&key) -
 929                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 930                              s_blocksize_bits);
 931
 932                         if (blocks_needed == 1) {
 933                                 un = &unf_single;
 934                         } else {
 935                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 936                                 if (!un) {
 937                                         un = &unf_single;
 938                                         blocks_needed = 1;
 939                                         max_to_insert = 0;
 940                                 }
 941                         }
 942                         if (blocks_needed <= max_to_insert) {
 943                                 /* we are going to add target block to the file. Use allocated
 944                                    block for that */
 945                                 un[blocks_needed - 1] =
 946                                     cpu_to_le32(allocated_block_nr);
 947                                 set_block_dev_mapped(bh_result,
 948                                                      allocated_block_nr, inode);
 949                                 set_buffer_new(bh_result);
 950                                 done = 1;
 951                         } else {
 952                                 /* paste hole to the indirect item */
 953                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 954                                    only have space for one block */
 955                                 blocks_needed =
 956                                     max_to_insert ? max_to_insert : 1;
 957                         }
 958                         retval =
 959                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 960                                                      (char *)un,
 961                                                      UNFM_P_SIZE *
 962                                                      blocks_needed);
 963
 964                         if (blocks_needed != 1)
 965                                 kfree(un);
 966
 967                         if (retval) {
 968                                 reiserfs_free_block(th, inode,
 969                                                     allocated_block_nr, 1);
 970                                 goto failure;
 971                         }
 972                         if (!done) {
 973                                 /* We need to mark new file size in case this function will be
 974                                    interrupted/aborted later on. And we may do this only for
 975                                    holes. */
 976                                 inode->i_size +=
 977                                     inode->i_sb->s_blocksize * blocks_needed;
 978                         }
 979                 }
 980
 981                 if (done == 1)
 982                         break;
 983
 984                 /* this loop could log more blocks than we had originally asked
 985                  ** for.  So, we have to allow the transaction to end if it is
 986                  ** too big or too full.  Update the inode so things are
 987                  ** consistent if we crash before the function returns
 988                  **
 989                  ** release the path so that anybody waiting on the path before
 990                  ** ending their transaction will be able to continue.
 991                  */
 992                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 993                         retval = restart_transaction(th, inode, &path);
 994                         if (retval)
 995                                 goto failure;
 996                 }
 997                 /*
 998                  * inserting indirect pointers for a hole can take a
 999                  * long time.  reschedule if needed and also release the write
1000                  * lock for others.
1001                  */
1002                 if (need_resched()) {
1003                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1004                         schedule();
1005                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1006                 }
1007
1008                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1009                 if (retval == IO_ERROR) {
1010                         retval = -EIO;
1011                         goto failure;
1012                 }
1013                 if (retval == POSITION_FOUND) {
1014                         reiserfs_warning(inode->i_sb, "vs-825",
1015                                          "%K should not be found", &key);
1016                         retval = -EEXIST;
1017                         if (allocated_block_nr)
1018                                 reiserfs_free_block(th, inode,
1019                                                     allocated_block_nr, 1);
1020                         pathrelse(&path);
1021                         goto failure;
1022                 }
1023                 bh = get_last_bh(&path);
1024                 ih = get_ih(&path);
1025                 item = get_item(&path);
1026                 pos_in_item = path.pos_in_item;
1027         } while (1);
1028
1029         retval = 0;
1030
1031       failure:
1032         if (th && (!dangle || (retval && !th->t_trans_id))) {
1033                 int err;
1034                 if (th->t_trans_id)
1035                         reiserfs_update_sd(th, inode);
1036                 err = reiserfs_end_persistent_transaction(th);
1037                 if (err)
1038                         retval = err;
1039         }
1040
1041         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1042         reiserfs_check_path(&path);
1043         return retval;
1044 }
1045
1046 static int
1047 reiserfs_readpages(struct file *file, struct address_space *mapping,
1048                    struct list_head *pages, unsigned nr_pages)
1049 {
1050         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1051 }
1052
1053 /* Compute real number of used bytes by file
1054  * Following three functions can go away when we'll have enough space in stat item
1055  */
1056 static int real_space_diff(struct inode *inode, int sd_size)
1057 {
1058         int bytes;
1059         loff_t blocksize = inode->i_sb->s_blocksize;
1060
1061         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1062                 return sd_size;
1063
1064         /* End of file is also in full block with indirect reference, so round
1065          ** up to the next block.
1066          **
1067          ** there is just no way to know if the tail is actually packed
1068          ** on the file, so we have to assume it isn't.  When we pack the
1069          ** tail, we add 4 bytes to pretend there really is an unformatted
1070          ** node pointer
1071          */
1072         bytes =
1073             ((inode->i_size +
1074               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1075             sd_size;
1076         return bytes;
1077 }
1078
1079 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1080                                         int sd_size)
1081 {
1082         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1083                 return inode->i_size +
1084                     (loff_t) (real_space_diff(inode, sd_size));
1085         }
1086         return ((loff_t) real_space_diff(inode, sd_size)) +
1087             (((loff_t) blocks) << 9);
1088 }
1089
1090 /* Compute number of blocks used by file in ReiserFS counting */
1091 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1092 {
1093         loff_t bytes = inode_get_bytes(inode);
1094         loff_t real_space = real_space_diff(inode, sd_size);
1095
1096         /* keeps fsck and non-quota versions of reiserfs happy */
1097         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1098                 bytes += (loff_t) 511;
1099         }
1100
1101         /* files from before the quota patch might i_blocks such that
1102          ** bytes < real_space.  Deal with that here to prevent it from
1103          ** going negative.
1104          */
1105         if (bytes < real_space)
1106                 return 0;
1107         return (bytes - real_space) >> 9;
1108 }
1109
1110 //
1111 // BAD: new directories have stat data of new type and all other items
1112 // of old type. Version stored in the inode says about body items, so
1113 // in update_stat_data we can not rely on inode, but have to check
1114 // item version directly
1115 //
1116
1117 // called by read_locked_inode
1118 static void init_inode(struct inode *inode, struct treepath *path)
1119 {
1120         struct buffer_head *bh;
1121         struct item_head *ih;
1122         __u32 rdev;
1123         //int version = ITEM_VERSION_1;
1124
1125         bh = PATH_PLAST_BUFFER(path);
1126         ih = PATH_PITEM_HEAD(path);
1127
1128         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1129
1130         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1131         REISERFS_I(inode)->i_flags = 0;
1132         REISERFS_I(inode)->i_prealloc_block = 0;
1133         REISERFS_I(inode)->i_prealloc_count = 0;
1134         REISERFS_I(inode)->i_trans_id = 0;
1135         REISERFS_I(inode)->i_jl = NULL;
1136         mutex_init(&(REISERFS_I(inode)->i_mmap));
1137         reiserfs_init_xattr_rwsem(inode);
1138
1139         if (stat_data_v1(ih)) {
1140                 struct stat_data_v1 *sd =
1141                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1142                 unsigned long blocks;
1143
1144                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1145                 set_inode_sd_version(inode, STAT_DATA_V1);
1146                 inode->i_mode = sd_v1_mode(sd);
1147                 inode->i_nlink = sd_v1_nlink(sd);
1148                 inode->i_uid = sd_v1_uid(sd);
1149                 inode->i_gid = sd_v1_gid(sd);
1150                 inode->i_size = sd_v1_size(sd);
1151                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1152                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1153                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1154                 inode->i_atime.tv_nsec = 0;
1155                 inode->i_ctime.tv_nsec = 0;
1156                 inode->i_mtime.tv_nsec = 0;
1157
1158                 inode->i_blocks = sd_v1_blocks(sd);
1159                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1160                 blocks = (inode->i_size + 511) >> 9;
1161                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1162                 if (inode->i_blocks > blocks) {
1163                         // there was a bug in <=3.5.23 when i_blocks could take negative
1164                         // values. Starting from 3.5.17 this value could even be stored in
1165                         // stat data. For such files we set i_blocks based on file
1166                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1167                         // only updated if file's inode will ever change
1168                         inode->i_blocks = blocks;
1169                 }
1170
1171                 rdev = sd_v1_rdev(sd);
1172                 REISERFS_I(inode)->i_first_direct_byte =
1173                     sd_v1_first_direct_byte(sd);
1174                 /* an early bug in the quota code can give us an odd number for the
1175                  ** block count.  This is incorrect, fix it here.
1176                  */
1177                 if (inode->i_blocks & 1) {
1178                         inode->i_blocks++;
1179                 }
1180                 inode_set_bytes(inode,
1181                                 to_real_used_space(inode, inode->i_blocks,
1182                                                    SD_V1_SIZE));
1183                 /* nopack is initially zero for v1 objects. For v2 objects,
1184                    nopack is initialised from sd_attrs */
1185                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1186         } else {
1187                 // new stat data found, but object may have old items
1188                 // (directories and symlinks)
1189                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1190
1191                 inode->i_mode = sd_v2_mode(sd);
1192                 inode->i_nlink = sd_v2_nlink(sd);
1193                 inode->i_uid = sd_v2_uid(sd);
1194                 inode->i_size = sd_v2_size(sd);
1195                 inode->i_gid = sd_v2_gid(sd);
1196                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1197                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1198                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1199                 inode->i_ctime.tv_nsec = 0;
1200                 inode->i_mtime.tv_nsec = 0;
1201                 inode->i_atime.tv_nsec = 0;
1202                 inode->i_blocks = sd_v2_blocks(sd);
1203                 rdev = sd_v2_rdev(sd);
1204                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1205                         inode->i_generation =
1206                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1207                 else
1208                         inode->i_generation = sd_v2_generation(sd);
1209
1210                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1211                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1212                 else
1213                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1214                 REISERFS_I(inode)->i_first_direct_byte = 0;
1215                 set_inode_sd_version(inode, STAT_DATA_V2);
1216                 inode_set_bytes(inode,
1217                                 to_real_used_space(inode, inode->i_blocks,
1218                                                    SD_V2_SIZE));
1219                 /* read persistent inode attributes from sd and initalise
1220                    generic inode flags from them */
1221                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1222                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1223         }
1224
1225         pathrelse(path);
1226         if (S_ISREG(inode->i_mode)) {
1227                 inode->i_op = &reiserfs_file_inode_operations;
1228                 inode->i_fop = &reiserfs_file_operations;
1229                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1230         } else if (S_ISDIR(inode->i_mode)) {
1231                 inode->i_op = &reiserfs_dir_inode_operations;
1232                 inode->i_fop = &reiserfs_dir_operations;
1233         } else if (S_ISLNK(inode->i_mode)) {
1234                 inode->i_op = &reiserfs_symlink_inode_operations;
1235                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1236         } else {
1237                 inode->i_blocks = 0;
1238                 inode->i_op = &reiserfs_special_inode_operations;
1239                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1240         }
1241 }
1242
1243 // update new stat data with inode fields
1244 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1245 {
1246         struct stat_data *sd_v2 = (struct stat_data *)sd;
1247         __u16 flags;
1248
1249         set_sd_v2_mode(sd_v2, inode->i_mode);
1250         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1251         set_sd_v2_uid(sd_v2, inode->i_uid);
1252         set_sd_v2_size(sd_v2, size);
1253         set_sd_v2_gid(sd_v2, inode->i_gid);
1254         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1255         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1256         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1257         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1258         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1259                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1260         else
1261                 set_sd_v2_generation(sd_v2, inode->i_generation);
1262         flags = REISERFS_I(inode)->i_attrs;
1263         i_attrs_to_sd_attrs(inode, &flags);
1264         set_sd_v2_attrs(sd_v2, flags);
1265 }
1266
1267 // used to copy inode's fields to old stat data
1268 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1269 {
1270         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1271
1272         set_sd_v1_mode(sd_v1, inode->i_mode);
1273         set_sd_v1_uid(sd_v1, inode->i_uid);
1274         set_sd_v1_gid(sd_v1, inode->i_gid);
1275         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1276         set_sd_v1_size(sd_v1, size);
1277         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1278         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1279         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1280
1281         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1282                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1283         else
1284                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1285
1286         // Sigh. i_first_direct_byte is back
1287         set_sd_v1_first_direct_byte(sd_v1,
1288                                     REISERFS_I(inode)->i_first_direct_byte);
1289 }
1290
1291 /* NOTE, you must prepare the buffer head before sending it here,
1292 ** and then log it after the call
1293 */
1294 static void update_stat_data(struct treepath *path, struct inode *inode,
1295                              loff_t size)
1296 {
1297         struct buffer_head *bh;
1298         struct item_head *ih;
1299
1300         bh = PATH_PLAST_BUFFER(path);
1301         ih = PATH_PITEM_HEAD(path);
1302
1303         if (!is_statdata_le_ih(ih))
1304                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1305                                INODE_PKEY(inode), ih);
1306
1307         if (stat_data_v1(ih)) {
1308                 // path points to old stat data
1309                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1310         } else {
1311                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1312         }
1313
1314         return;
1315 }
1316
1317 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1318                              struct inode *inode, loff_t size)
1319 {
1320         struct cpu_key key;
1321         INITIALIZE_PATH(path);
1322         struct buffer_head *bh;
1323         int fs_gen;
1324         struct item_head *ih, tmp_ih;
1325         int retval;
1326
1327         BUG_ON(!th->t_trans_id);
1328
1329         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1330
1331         for (;;) {
1332                 int pos;
1333                 /* look for the object's stat data */
1334                 retval = search_item(inode->i_sb, &key, &path);
1335                 if (retval == IO_ERROR) {
1336                         reiserfs_error(inode->i_sb, "vs-13050",
1337                                        "i/o failure occurred trying to "
1338                                        "update %K stat data", &key);
1339                         return;
1340                 }
1341                 if (retval == ITEM_NOT_FOUND) {
1342                         pos = PATH_LAST_POSITION(&path);
1343                         pathrelse(&path);
1344                         if (inode->i_nlink == 0) {
1345                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1346                                 return;
1347                         }
1348                         reiserfs_warning(inode->i_sb, "vs-13060",
1349                                          "stat data of object %k (nlink == %d) "
1350                                          "not found (pos %d)",
1351                                          INODE_PKEY(inode), inode->i_nlink,
1352                                          pos);
1353                         reiserfs_check_path(&path);
1354                         return;
1355                 }
1356
1357                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1358                  ** FS might change.  We have to detect that, and loop back to the
1359                  ** search if the stat data item has moved
1360                  */
1361                 bh = get_last_bh(&path);
1362                 ih = get_ih(&path);
1363                 copy_item_head(&tmp_ih, ih);
1364                 fs_gen = get_generation(inode->i_sb);
1365                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1366                 if (fs_changed(fs_gen, inode->i_sb)
1367                     && item_moved(&tmp_ih, &path)) {
1368                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1369                         continue;       /* Stat_data item has been moved after scheduling. */
1370                 }
1371                 break;
1372         }
1373         update_stat_data(&path, inode, size);
1374         journal_mark_dirty(th, th->t_super, bh);
1375         pathrelse(&path);
1376         return;
1377 }
1378
1379 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1380 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1381 ** and clear the key in the private portion of the inode, otherwise a
1382 ** corresponding iput might try to delete whatever object the inode last
1383 ** represented.
1384 */
1385 static void reiserfs_make_bad_inode(struct inode *inode)
1386 {
1387         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1388         make_bad_inode(inode);
1389 }
1390
1391 //
1392 // initially this function was derived from minix or ext2's analog and
1393 // evolved as the prototype did
1394 //
1395
1396 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1397 {
1398         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1399         inode->i_ino = args->objectid;
1400         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1401         return 0;
1402 }
1403
1404 /* looks for stat data in the tree, and fills up the fields of in-core
1405    inode stat data fields */
1406 void reiserfs_read_locked_inode(struct inode *inode,
1407                                 struct reiserfs_iget_args *args)
1408 {
1409         INITIALIZE_PATH(path_to_sd);
1410         struct cpu_key key;
1411         unsigned long dirino;
1412         int retval;
1413
1414         dirino = args->dirid;
1415
1416         /* set version 1, version 2 could be used too, because stat data
1417            key is the same in both versions */
1418         key.version = KEY_FORMAT_3_5;
1419         key.on_disk_key.k_dir_id = dirino;
1420         key.on_disk_key.k_objectid = inode->i_ino;
1421         key.on_disk_key.k_offset = 0;
1422         key.on_disk_key.k_type = 0;
1423
1424         /* look for the object's stat data */
1425         retval = search_item(inode->i_sb, &key, &path_to_sd);
1426         if (retval == IO_ERROR) {
1427                 reiserfs_error(inode->i_sb, "vs-13070",
1428                                "i/o failure occurred trying to find "
1429                                "stat data of %K", &key);
1430                 reiserfs_make_bad_inode(inode);
1431                 return;
1432         }
1433         if (retval != ITEM_FOUND) {
1434                 /* a stale NFS handle can trigger this without it being an error */
1435                 pathrelse(&path_to_sd);
1436                 reiserfs_make_bad_inode(inode);
1437                 inode->i_nlink = 0;
1438                 return;
1439         }
1440
1441         init_inode(inode, &path_to_sd);
1442
1443         /* It is possible that knfsd is trying to access inode of a file
1444            that is being removed from the disk by some other thread. As we
1445            update sd on unlink all that is required is to check for nlink
1446            here. This bug was first found by Sizif when debugging
1447            SquidNG/Butterfly, forgotten, and found again after Philippe
1448            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1449
1450            More logical fix would require changes in fs/inode.c:iput() to
1451            remove inode from hash-table _after_ fs cleaned disk stuff up and
1452            in iget() to return NULL if I_FREEING inode is found in
1453            hash-table. */
1454         /* Currently there is one place where it's ok to meet inode with
1455            nlink==0: processing of open-unlinked and half-truncated files
1456            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1457         if ((inode->i_nlink == 0) &&
1458             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1459                 reiserfs_warning(inode->i_sb, "vs-13075",
1460                                  "dead inode read from disk %K. "
1461                                  "This is likely to be race with knfsd. Ignore",
1462                                  &key);
1463                 reiserfs_make_bad_inode(inode);
1464         }
1465
1466         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1467
1468 }
1469
1470 /**
1471  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1472  *
1473  * @inode:    inode from hash table to check
1474  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1475  *
1476  * This function is called by iget5_locked() to distinguish reiserfs inodes
1477  * having the same inode numbers. Such inodes can only exist due to some
1478  * error condition. One of them should be bad. Inodes with identical
1479  * inode numbers (objectids) are distinguished by parent directory ids.
1480  *
1481  */
1482 int reiserfs_find_actor(struct inode *inode, void *opaque)
1483 {
1484         struct reiserfs_iget_args *args;
1485
1486         args = opaque;
1487         /* args is already in CPU order */
1488         return (inode->i_ino == args->objectid) &&
1489             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1490 }
1491
1492 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1493 {
1494         struct inode *inode;
1495         struct reiserfs_iget_args args;
1496
1497         args.objectid = key->on_disk_key.k_objectid;
1498         args.dirid = key->on_disk_key.k_dir_id;
1499         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1500                              reiserfs_find_actor, reiserfs_init_locked_inode,
1501                              (void *)(&args));
1502         if (!inode)
1503                 return ERR_PTR(-ENOMEM);
1504
1505         if (inode->i_state & I_NEW) {
1506                 reiserfs_read_locked_inode(inode, &args);
1507                 unlock_new_inode(inode);
1508         }
1509
1510         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1511                 /* either due to i/o error or a stale NFS handle */
1512                 iput(inode);
1513                 inode = NULL;
1514         }
1515         return inode;
1516 }
1517
1518 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1519         u32 objectid, u32 dir_id, u32 generation)
1520
1521 {
1522         struct cpu_key key;
1523         struct inode *inode;
1524
1525         key.on_disk_key.k_objectid = objectid;
1526         key.on_disk_key.k_dir_id = dir_id;
1527         reiserfs_write_lock(sb);
1528         inode = reiserfs_iget(sb, &key);
1529         if (inode && !IS_ERR(inode) && generation != 0 &&
1530             generation != inode->i_generation) {
1531                 iput(inode);
1532                 inode = NULL;
1533         }
1534         reiserfs_write_unlock(sb);
1535
1536         return d_obtain_alias(inode);
1537 }
1538
1539 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1540                 int fh_len, int fh_type)
1541 {
1542         /* fhtype happens to reflect the number of u32s encoded.
1543          * due to a bug in earlier code, fhtype might indicate there
1544          * are more u32s then actually fitted.
1545          * so if fhtype seems to be more than len, reduce fhtype.
1546          * Valid types are:
1547          *   2 - objectid + dir_id - legacy support
1548          *   3 - objectid + dir_id + generation
1549          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1550          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1551          *   6 - as above plus generation of directory
1552          * 6 does not fit in NFSv2 handles
1553          */
1554         if (fh_type > fh_len) {
1555                 if (fh_type != 6 || fh_len != 5)
1556                         reiserfs_warning(sb, "reiserfs-13077",
1557                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1558                                 fh_type, fh_len);
1559                 fh_type = 5;
1560         }
1561
1562         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1563                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1564 }
1565
1566 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1567                 int fh_len, int fh_type)
1568 {
1569         if (fh_type < 4)
1570                 return NULL;
1571
1572         return reiserfs_get_dentry(sb,
1573                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1574                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1575                 (fh_type == 6) ? fid->raw[5] : 0);
1576 }
1577
1578 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1579                        int need_parent)
1580 {
1581         struct inode *inode = dentry->d_inode;
1582         int maxlen = *lenp;
1583
1584         if (maxlen < 3)
1585                 return 255;
1586
1587         data[0] = inode->i_ino;
1588         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1589         data[2] = inode->i_generation;
1590         *lenp = 3;
1591         /* no room for directory info? return what we've stored so far */
1592         if (maxlen < 5 || !need_parent)
1593                 return 3;
1594
1595         spin_lock(&dentry->d_lock);
1596         inode = dentry->d_parent->d_inode;
1597         data[3] = inode->i_ino;
1598         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1599         *lenp = 5;
1600         if (maxlen >= 6) {
1601                 data[5] = inode->i_generation;
1602                 *lenp = 6;
1603         }
1604         spin_unlock(&dentry->d_lock);
1605         return *lenp;
1606 }
1607
1608 /* looks for stat data, then copies fields to it, marks the buffer
1609    containing stat data as dirty */
1610 /* reiserfs inodes are never really dirty, since the dirty inode call
1611 ** always logs them.  This call allows the VFS inode marking routines
1612 ** to properly mark inodes for datasync and such, but only actually
1613 ** does something when called for a synchronous update.
1614 */
1615 int reiserfs_write_inode(struct inode *inode, int do_sync)
1616 {
1617         struct reiserfs_transaction_handle th;
1618         int jbegin_count = 1;
1619
1620         if (inode->i_sb->s_flags & MS_RDONLY)
1621                 return -EROFS;
1622         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1623          ** these cases are just when the system needs ram, not when the
1624          ** inode needs to reach disk for safety, and they can safely be
1625          ** ignored because the altered inode has already been logged.
1626          */
1627         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1628                 reiserfs_write_lock(inode->i_sb);
1629                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1630                         reiserfs_update_sd(&th, inode);
1631                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1632                 }
1633                 reiserfs_write_unlock(inode->i_sb);
1634         }
1635         return 0;
1636 }
1637
1638 /* stat data of new object is inserted already, this inserts the item
1639    containing "." and ".." entries */
1640 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1641                                   struct inode *inode,
1642                                   struct item_head *ih, struct treepath *path,
1643                                   struct inode *dir)
1644 {
1645         struct super_block *sb = th->t_super;
1646         char empty_dir[EMPTY_DIR_SIZE];
1647         char *body = empty_dir;
1648         struct cpu_key key;
1649         int retval;
1650
1651         BUG_ON(!th->t_trans_id);
1652
1653         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1654                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1655                       TYPE_DIRENTRY, 3 /*key length */ );
1656
1657         /* compose item head for new item. Directories consist of items of
1658            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1659            is done by reiserfs_new_inode */
1660         if (old_format_only(sb)) {
1661                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1662                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1663
1664                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1665                                        ih->ih_key.k_objectid,
1666                                        INODE_PKEY(dir)->k_dir_id,
1667                                        INODE_PKEY(dir)->k_objectid);
1668         } else {
1669                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1670                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1671
1672                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1673                                     ih->ih_key.k_objectid,
1674                                     INODE_PKEY(dir)->k_dir_id,
1675                                     INODE_PKEY(dir)->k_objectid);
1676         }
1677
1678         /* look for place in the tree for new item */
1679         retval = search_item(sb, &key, path);
1680         if (retval == IO_ERROR) {
1681                 reiserfs_error(sb, "vs-13080",
1682                                "i/o failure occurred creating new directory");
1683                 return -EIO;
1684         }
1685         if (retval == ITEM_FOUND) {
1686                 pathrelse(path);
1687                 reiserfs_warning(sb, "vs-13070",
1688                                  "object with this key exists (%k)",
1689                                  &(ih->ih_key));
1690                 return -EEXIST;
1691         }
1692
1693         /* insert item, that is empty directory item */
1694         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1695 }
1696
1697 /* stat data of object has been inserted, this inserts the item
1698    containing the body of symlink */
1699 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1700                                 struct item_head *ih,
1701                                 struct treepath *path, const char *symname,
1702                                 int item_len)
1703 {
1704         struct super_block *sb = th->t_super;
1705         struct cpu_key key;
1706         int retval;
1707
1708         BUG_ON(!th->t_trans_id);
1709
1710         _make_cpu_key(&key, KEY_FORMAT_3_5,
1711                       le32_to_cpu(ih->ih_key.k_dir_id),
1712                       le32_to_cpu(ih->ih_key.k_objectid),
1713                       1, TYPE_DIRECT, 3 /*key length */ );
1714
1715         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1716                           0 /*free_space */ );
1717
1718         /* look for place in the tree for new item */
1719         retval = search_item(sb, &key, path);
1720         if (retval == IO_ERROR) {
1721                 reiserfs_error(sb, "vs-13080",
1722                                "i/o failure occurred creating new symlink");
1723                 return -EIO;
1724         }
1725         if (retval == ITEM_FOUND) {
1726                 pathrelse(path);
1727                 reiserfs_warning(sb, "vs-13080",
1728                                  "object with this key exists (%k)",
1729                                  &(ih->ih_key));
1730                 return -EEXIST;
1731         }
1732
1733         /* insert item, that is body of symlink */
1734         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1735 }
1736
1737 /* inserts the stat data into the tree, and then calls
1738    reiserfs_new_directory (to insert ".", ".." item if new object is
1739    directory) or reiserfs_new_symlink (to insert symlink body if new
1740    object is symlink) or nothing (if new object is regular file)
1741
1742    NOTE! uid and gid must already be set in the inode.  If we return
1743    non-zero due to an error, we have to drop the quota previously allocated
1744    for the fresh inode.  This can only be done outside a transaction, so
1745    if we return non-zero, we also end the transaction.  */
1746 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1747                        struct inode *dir, int mode, const char *symname,
1748                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1749                           strlen (symname) for symlinks) */
1750                        loff_t i_size, struct dentry *dentry,
1751                        struct inode *inode,
1752                        struct reiserfs_security_handle *security)
1753 {
1754         struct super_block *sb;
1755         struct reiserfs_iget_args args;
1756         INITIALIZE_PATH(path_to_key);
1757         struct cpu_key key;
1758         struct item_head ih;
1759         struct stat_data sd;
1760         int retval;
1761         int err;
1762
1763         BUG_ON(!th->t_trans_id);
1764
1765         if (vfs_dq_alloc_inode(inode)) {
1766                 err = -EDQUOT;
1767                 goto out_end_trans;
1768         }
1769         if (!dir->i_nlink) {
1770                 err = -EPERM;
1771                 goto out_bad_inode;
1772         }
1773
1774         sb = dir->i_sb;
1775
1776         /* item head of new item */
1777         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1778         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1779         if (!ih.ih_key.k_objectid) {
1780                 err = -ENOMEM;
1781                 goto out_bad_inode;
1782         }
1783         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1784         if (old_format_only(sb))
1785                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1786                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1787         else
1788                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1789                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1790         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1791         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1792         if (insert_inode_locked4(inode, args.objectid,
1793                              reiserfs_find_actor, &args) < 0) {
1794                 err = -EINVAL;
1795                 goto out_bad_inode;
1796         }
1797         if (old_format_only(sb))
1798                 /* not a perfect generation count, as object ids can be reused, but
1799                  ** this is as good as reiserfs can do right now.
1800                  ** note that the private part of inode isn't filled in yet, we have
1801                  ** to use the directory.
1802                  */
1803                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1804         else
1805 #if defined( USE_INODE_GENERATION_COUNTER )
1806                 inode->i_generation =
1807                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1808 #else
1809                 inode->i_generation = ++event;
1810 #endif
1811
1812         /* fill stat data */
1813         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1814
1815         /* uid and gid must already be set by the caller for quota init */
1816
1817         /* symlink cannot be immutable or append only, right? */
1818         if (S_ISLNK(inode->i_mode))
1819                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1820
1821         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1822         inode->i_size = i_size;
1823         inode->i_blocks = 0;
1824         inode->i_bytes = 0;
1825         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1826             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1827
1828         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1829         REISERFS_I(inode)->i_flags = 0;
1830         REISERFS_I(inode)->i_prealloc_block = 0;
1831         REISERFS_I(inode)->i_prealloc_count = 0;
1832         REISERFS_I(inode)->i_trans_id = 0;
1833         REISERFS_I(inode)->i_jl = NULL;
1834         REISERFS_I(inode)->i_attrs =
1835             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1836         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1837         mutex_init(&(REISERFS_I(inode)->i_mmap));
1838         reiserfs_init_xattr_rwsem(inode);
1839
1840         /* key to search for correct place for new stat data */
1841         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1842                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1843                       TYPE_STAT_DATA, 3 /*key length */ );
1844
1845         /* find proper place for inserting of stat data */
1846         retval = search_item(sb, &key, &path_to_key);
1847         if (retval == IO_ERROR) {
1848                 err = -EIO;
1849                 goto out_bad_inode;
1850         }
1851         if (retval == ITEM_FOUND) {
1852                 pathrelse(&path_to_key);
1853                 err = -EEXIST;
1854                 goto out_bad_inode;
1855         }
1856         if (old_format_only(sb)) {
1857                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1858                         pathrelse(&path_to_key);
1859                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1860                         err = -EINVAL;
1861                         goto out_bad_inode;
1862                 }
1863                 inode2sd_v1(&sd, inode, inode->i_size);
1864         } else {
1865                 inode2sd(&sd, inode, inode->i_size);
1866         }
1867         // store in in-core inode the key of stat data and version all
1868         // object items will have (directory items will have old offset
1869         // format, other new objects will consist of new items)
1870         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1871                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1872         else
1873                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1874         if (old_format_only(sb))
1875                 set_inode_sd_version(inode, STAT_DATA_V1);
1876         else
1877                 set_inode_sd_version(inode, STAT_DATA_V2);
1878
1879         /* insert the stat data into the tree */
1880 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1881         if (REISERFS_I(dir)->new_packing_locality)
1882                 th->displace_new_blocks = 1;
1883 #endif
1884         retval =
1885             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1886                                  (char *)(&sd));
1887         if (retval) {
1888                 err = retval;
1889                 reiserfs_check_path(&path_to_key);
1890                 goto out_bad_inode;
1891         }
1892 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1893         if (!th->displace_new_blocks)
1894                 REISERFS_I(dir)->new_packing_locality = 0;
1895 #endif
1896         if (S_ISDIR(mode)) {
1897                 /* insert item with "." and ".." */
1898                 retval =
1899                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1900         }
1901
1902         if (S_ISLNK(mode)) {
1903                 /* insert body of symlink */
1904                 if (!old_format_only(sb))
1905                         i_size = ROUND_UP(i_size);
1906                 retval =
1907                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1908                                          i_size);
1909         }
1910         if (retval) {
1911                 err = retval;
1912                 reiserfs_check_path(&path_to_key);
1913                 journal_end(th, th->t_super, th->t_blocks_allocated);
1914                 goto out_inserted_sd;
1915         }
1916
1917         if (reiserfs_posixacl(inode->i_sb)) {
1918                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1919                 if (retval) {
1920                         err = retval;
1921                         reiserfs_check_path(&path_to_key);
1922                         journal_end(th, th->t_super, th->t_blocks_allocated);
1923                         goto out_inserted_sd;
1924                 }
1925         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1926                 reiserfs_warning(inode->i_sb, "jdm-13090",
1927                                  "ACLs aren't enabled in the fs, "
1928                                  "but vfs thinks they are!");
1929         } else if (IS_PRIVATE(dir))
1930                 inode->i_flags |= S_PRIVATE;
1931
1932         if (security->name) {
1933                 retval = reiserfs_security_write(th, inode, security);
1934                 if (retval) {
1935                         err = retval;
1936                         reiserfs_check_path(&path_to_key);
1937                         retval = journal_end(th, th->t_super,
1938                                              th->t_blocks_allocated);
1939                         if (retval)
1940                                 err = retval;
1941                         goto out_inserted_sd;
1942                 }
1943         }
1944
1945         reiserfs_update_sd(th, inode);
1946         reiserfs_check_path(&path_to_key);
1947
1948         return 0;
1949
1950 /* it looks like you can easily compress these two goto targets into
1951  * one.  Keeping it like this doesn't actually hurt anything, and they
1952  * are place holders for what the quota code actually needs.
1953  */
1954       out_bad_inode:
1955         /* Invalidate the object, nothing was inserted yet */
1956         INODE_PKEY(inode)->k_objectid = 0;
1957
1958         /* Quota change must be inside a transaction for journaling */
1959         vfs_dq_free_inode(inode);
1960
1961       out_end_trans:
1962         journal_end(th, th->t_super, th->t_blocks_allocated);
1963         /* Drop can be outside and it needs more credits so it's better to have it outside */
1964         vfs_dq_drop(inode);
1965         inode->i_flags |= S_NOQUOTA;
1966         make_bad_inode(inode);
1967
1968       out_inserted_sd:
1969         inode->i_nlink = 0;
1970         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1971         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1972         iput(inode);
1973         return err;
1974 }
1975
1976 /*
1977 ** finds the tail page in the page cache,
1978 ** reads the last block in.
1979 **
1980 ** On success, page_result is set to a locked, pinned page, and bh_result
1981 ** is set to an up to date buffer for the last block in the file.  returns 0.
1982 **
1983 ** tail conversion is not done, so bh_result might not be valid for writing
1984 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1985 ** trying to write the block.
1986 **
1987 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1988 */
1989 static int grab_tail_page(struct inode *inode,
1990                           struct page **page_result,
1991                           struct buffer_head **bh_result)
1992 {
1993
1994         /* we want the page with the last byte in the file,
1995          ** not the page that will hold the next byte for appending
1996          */
1997         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1998         unsigned long pos = 0;
1999         unsigned long start = 0;
2000         unsigned long blocksize = inode->i_sb->s_blocksize;
2001         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2002         struct buffer_head *bh;
2003         struct buffer_head *head;
2004         struct page *page;
2005         int error;
2006
2007         /* we know that we are only called with inode->i_size > 0.
2008          ** we also know that a file tail can never be as big as a block
2009          ** If i_size % blocksize == 0, our file is currently block aligned
2010          ** and it won't need converting or zeroing after a truncate.
2011          */
2012         if ((offset & (blocksize - 1)) == 0) {
2013                 return -ENOENT;
2014         }
2015         page = grab_cache_page(inode->i_mapping, index);
2016         error = -ENOMEM;
2017         if (!page) {
2018                 goto out;
2019         }
2020         /* start within the page of the last block in the file */
2021         start = (offset / blocksize) * blocksize;
2022
2023         error = block_prepare_write(page, start, offset,
2024                                     reiserfs_get_block_create_0);
2025         if (error)
2026                 goto unlock;
2027
2028         head = page_buffers(page);
2029         bh = head;
2030         do {
2031                 if (pos >= start) {
2032                         break;
2033                 }
2034                 bh = bh->b_this_page;
2035                 pos += blocksize;
2036         } while (bh != head);
2037
2038         if (!buffer_uptodate(bh)) {
2039                 /* note, this should never happen, prepare_write should
2040                  ** be taking care of this for us.  If the buffer isn't up to date,
2041                  ** I've screwed up the code to find the buffer, or the code to
2042                  ** call prepare_write
2043                  */
2044                 reiserfs_error(inode->i_sb, "clm-6000",
2045                                "error reading block %lu", bh->b_blocknr);
2046                 error = -EIO;
2047                 goto unlock;
2048         }
2049         *bh_result = bh;
2050         *page_result = page;
2051
2052       out:
2053         return error;
2054
2055       unlock:
2056         unlock_page(page);
2057         page_cache_release(page);
2058         return error;
2059 }
2060
2061 /*
2062 ** vfs version of truncate file.  Must NOT be called with
2063 ** a transaction already started.
2064 **
2065 ** some code taken from block_truncate_page
2066 */
2067 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2068 {
2069         struct reiserfs_transaction_handle th;
2070         /* we want the offset for the first byte after the end of the file */
2071         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2072         unsigned blocksize = inode->i_sb->s_blocksize;
2073         unsigned length;
2074         struct page *page = NULL;
2075         int error;
2076         struct buffer_head *bh = NULL;
2077         int err2;
2078         int lock_depth;
2079
2080         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2081
2082         if (inode->i_size > 0) {
2083                 error = grab_tail_page(inode, &page, &bh);
2084                 if (error) {
2085                         // -ENOENT means we truncated past the end of the file,
2086                         // and get_block_create_0 could not find a block to read in,
2087                         // which is ok.
2088                         if (error != -ENOENT)
2089                                 reiserfs_error(inode->i_sb, "clm-6001",
2090                                                "grab_tail_page failed %d",
2091                                                error);
2092                         page = NULL;
2093                         bh = NULL;
2094                 }
2095         }
2096
2097         /* so, if page != NULL, we have a buffer head for the offset at
2098          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2099          ** then we have an unformatted node.  Otherwise, we have a direct item,
2100          ** and no zeroing is required on disk.  We zero after the truncate,
2101          ** because the truncate might pack the item anyway
2102          ** (it will unmap bh if it packs).
2103          */
2104         /* it is enough to reserve space in transaction for 2 balancings:
2105            one for "save" link adding and another for the first
2106            cut_from_item. 1 is for update_sd */
2107         error = journal_begin(&th, inode->i_sb,
2108                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2109         if (error)
2110                 goto out;
2111         reiserfs_update_inode_transaction(inode);
2112         if (update_timestamps)
2113                 /* we are doing real truncate: if the system crashes before the last
2114                    transaction of truncating gets committed - on reboot the file
2115                    either appears truncated properly or not truncated at all */
2116                 add_save_link(&th, inode, 1);
2117         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2118         error =
2119             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2120         if (error)
2121                 goto out;
2122
2123         /* check reiserfs_do_truncate after ending the transaction */
2124         if (err2) {
2125                 error = err2;
2126                 goto out;
2127         }
2128
2129         if (update_timestamps) {
2130                 error = remove_save_link(inode, 1 /* truncate */);
2131                 if (error)
2132                         goto out;
2133         }
2134
2135         if (page) {
2136                 length = offset & (blocksize - 1);
2137                 /* if we are not on a block boundary */
2138                 if (length) {
2139                         length = blocksize - length;
2140                         zero_user(page, offset, length);
2141                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2142                                 mark_buffer_dirty(bh);
2143                         }
2144                 }
2145                 unlock_page(page);
2146                 page_cache_release(page);
2147         }
2148
2149         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2150
2151         return 0;
2152       out:
2153         if (page) {
2154                 unlock_page(page);
2155                 page_cache_release(page);
2156         }
2157
2158         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2159
2160         return error;
2161 }
2162
2163 static int map_block_for_writepage(struct inode *inode,
2164                                    struct buffer_head *bh_result,
2165                                    unsigned long block)
2166 {
2167         struct reiserfs_transaction_handle th;
2168         int fs_gen;
2169         struct item_head tmp_ih;
2170         struct item_head *ih;
2171         struct buffer_head *bh;
2172         __le32 *item;
2173         struct cpu_key key;
2174         INITIALIZE_PATH(path);
2175         int pos_in_item;
2176         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2177         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2178         int retval;
2179         int use_get_block = 0;
2180         int bytes_copied = 0;
2181         int copy_size;
2182         int trans_running = 0;
2183
2184         /* catch places below that try to log something without starting a trans */
2185         th.t_trans_id = 0;
2186
2187         if (!buffer_uptodate(bh_result)) {
2188                 return -EIO;
2189         }
2190
2191         kmap(bh_result->b_page);
2192       start_over:
2193         reiserfs_write_lock(inode->i_sb);
2194         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2195
2196       research:
2197         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2198         if (retval != POSITION_FOUND) {
2199                 use_get_block = 1;
2200                 goto out;
2201         }
2202
2203         bh = get_last_bh(&path);
2204         ih = get_ih(&path);
2205         item = get_item(&path);
2206         pos_in_item = path.pos_in_item;
2207
2208         /* we've found an unformatted node */
2209         if (indirect_item_found(retval, ih)) {
2210                 if (bytes_copied > 0) {
2211                         reiserfs_warning(inode->i_sb, "clm-6002",
2212                                          "bytes_copied %d", bytes_copied);
2213                 }
2214                 if (!get_block_num(item, pos_in_item)) {
2215                         /* crap, we are writing to a hole */
2216                         use_get_block = 1;
2217                         goto out;
2218                 }
2219                 set_block_dev_mapped(bh_result,
2220                                      get_block_num(item, pos_in_item), inode);
2221         } else if (is_direct_le_ih(ih)) {
2222                 char *p;
2223                 p = page_address(bh_result->b_page);
2224                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2225                 copy_size = ih_item_len(ih) - pos_in_item;
2226
2227                 fs_gen = get_generation(inode->i_sb);
2228                 copy_item_head(&tmp_ih, ih);
2229
2230                 if (!trans_running) {
2231                         /* vs-3050 is gone, no need to drop the path */
2232                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2233                         if (retval)
2234                                 goto out;
2235                         reiserfs_update_inode_transaction(inode);
2236                         trans_running = 1;
2237                         if (fs_changed(fs_gen, inode->i_sb)
2238                             && item_moved(&tmp_ih, &path)) {
2239                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2240                                                                  bh);
2241                                 goto research;
2242                         }
2243                 }
2244
2245                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2246
2247                 if (fs_changed(fs_gen, inode->i_sb)
2248                     && item_moved(&tmp_ih, &path)) {
2249                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2250                         goto research;
2251                 }
2252
2253                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2254                        copy_size);
2255
2256                 journal_mark_dirty(&th, inode->i_sb, bh);
2257                 bytes_copied += copy_size;
2258                 set_block_dev_mapped(bh_result, 0, inode);
2259
2260                 /* are there still bytes left? */
2261                 if (bytes_copied < bh_result->b_size &&
2262                     (byte_offset + bytes_copied) < inode->i_size) {
2263                         set_cpu_key_k_offset(&key,
2264                                              cpu_key_k_offset(&key) +
2265                                              copy_size);
2266                         goto research;
2267                 }
2268         } else {
2269                 reiserfs_warning(inode->i_sb, "clm-6003",
2270                                  "bad item inode %lu", inode->i_ino);
2271                 retval = -EIO;
2272                 goto out;
2273         }
2274         retval = 0;
2275
2276       out:
2277         pathrelse(&path);
2278         if (trans_running) {
2279                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2280                 if (err)
2281                         retval = err;
2282                 trans_running = 0;
2283         }
2284         reiserfs_write_unlock(inode->i_sb);
2285
2286         /* this is where we fill in holes in the file. */
2287         if (use_get_block) {
2288                 retval = reiserfs_get_block(inode, block, bh_result,
2289                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2290                                             | GET_BLOCK_NO_DANGLE);
2291                 if (!retval) {
2292                         if (!buffer_mapped(bh_result)
2293                             || bh_result->b_blocknr == 0) {
2294                                 /* get_block failed to find a mapped unformatted node. */
2295                                 use_get_block = 0;
2296                                 goto start_over;
2297                         }
2298                 }
2299         }
2300         kunmap(bh_result->b_page);
2301
2302         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2303                 /* we've copied data from the page into the direct item, so the
2304                  * buffer in the page is now clean, mark it to reflect that.
2305                  */
2306                 lock_buffer(bh_result);
2307                 clear_buffer_dirty(bh_result);
2308                 unlock_buffer(bh_result);
2309         }
2310         return retval;
2311 }
2312
2313 /*
2314  * mason@suse.com: updated in 2.5.54 to follow the same general io
2315  * start/recovery path as __block_write_full_page, along with special
2316  * code to handle reiserfs tails.
2317  */
2318 static int reiserfs_write_full_page(struct page *page,
2319                                     struct writeback_control *wbc)
2320 {
2321         struct inode *inode = page->mapping->host;
2322         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2323         int error = 0;
2324         unsigned long block;
2325         sector_t last_block;
2326         struct buffer_head *head, *bh;
2327         int partial = 0;
2328         int nr = 0;
2329         int checked = PageChecked(page);
2330         struct reiserfs_transaction_handle th;
2331         struct super_block *s = inode->i_sb;
2332         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2333         th.t_trans_id = 0;
2334
2335         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2336         if (checked && (current->flags & PF_MEMALLOC)) {
2337                 redirty_page_for_writepage(wbc, page);
2338                 unlock_page(page);
2339                 return 0;
2340         }
2341
2342         /* The page dirty bit is cleared before writepage is called, which
2343          * means we have to tell create_empty_buffers to make dirty buffers
2344          * The page really should be up to date at this point, so tossing
2345          * in the BH_Uptodate is just a sanity check.
2346          */
2347         if (!page_has_buffers(page)) {
2348                 create_empty_buffers(page, s->s_blocksize,
2349                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2350         }
2351         head = page_buffers(page);
2352
2353         /* last page in the file, zero out any contents past the
2354          ** last byte in the file
2355          */
2356         if (page->index >= end_index) {
2357                 unsigned last_offset;
2358
2359                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2360                 /* no file contents in this page */
2361                 if (page->index >= end_index + 1 || !last_offset) {
2362                         unlock_page(page);
2363                         return 0;
2364                 }
2365                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2366         }
2367         bh = head;
2368         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2369         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2370         /* first map all the buffers, logging any direct items we find */
2371         do {
2372                 if (block > last_block) {
2373                         /*
2374                          * This can happen when the block size is less than
2375                          * the page size.  The corresponding bytes in the page
2376                          * were zero filled above
2377                          */
2378                         clear_buffer_dirty(bh);
2379                         set_buffer_uptodate(bh);
2380                 } else if ((checked || buffer_dirty(bh)) &&
2381                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2382                                                        && bh->b_blocknr ==
2383                                                        0))) {
2384                         /* not mapped yet, or it points to a direct item, search
2385                          * the btree for the mapping info, and log any direct
2386                          * items found
2387                          */
2388                         if ((error = map_block_for_writepage(inode, bh, block))) {
2389                                 goto fail;
2390                         }
2391                 }
2392                 bh = bh->b_this_page;
2393                 block++;
2394         } while (bh != head);
2395
2396         /*
2397          * we start the transaction after map_block_for_writepage,
2398          * because it can create holes in the file (an unbounded operation).
2399          * starting it here, we can make a reliable estimate for how many
2400          * blocks we're going to log
2401          */
2402         if (checked) {
2403                 ClearPageChecked(page);
2404                 reiserfs_write_lock(s);
2405                 error = journal_begin(&th, s, bh_per_page + 1);
2406                 if (error) {
2407                         reiserfs_write_unlock(s);
2408                         goto fail;
2409                 }
2410                 reiserfs_update_inode_transaction(inode);
2411         }
2412         /* now go through and lock any dirty buffers on the page */
2413         do {
2414                 get_bh(bh);
2415                 if (!buffer_mapped(bh))
2416                         continue;
2417                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2418                         continue;
2419
2420                 if (checked) {
2421                         reiserfs_prepare_for_journal(s, bh, 1);
2422                         journal_mark_dirty(&th, s, bh);
2423                         continue;
2424                 }
2425                 /* from this point on, we know the buffer is mapped to a
2426                  * real block and not a direct item
2427                  */
2428                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2429                         lock_buffer(bh);
2430                 } else {
2431                         if (!trylock_buffer(bh)) {
2432                                 redirty_page_for_writepage(wbc, page);
2433                                 continue;
2434                         }
2435                 }
2436                 if (test_clear_buffer_dirty(bh)) {
2437                         mark_buffer_async_write(bh);
2438                 } else {
2439                         unlock_buffer(bh);
2440                 }
2441         } while ((bh = bh->b_this_page) != head);
2442
2443         if (checked) {
2444                 error = journal_end(&th, s, bh_per_page + 1);
2445                 reiserfs_write_unlock(s);
2446                 if (error)
2447                         goto fail;
2448         }
2449         BUG_ON(PageWriteback(page));
2450         set_page_writeback(page);
2451         unlock_page(page);
2452
2453         /*
2454          * since any buffer might be the only dirty buffer on the page,
2455          * the first submit_bh can bring the page out of writeback.
2456          * be careful with the buffers.
2457          */
2458         do {
2459                 struct buffer_head *next = bh->b_this_page;
2460                 if (buffer_async_write(bh)) {
2461                         submit_bh(WRITE, bh);
2462                         nr++;
2463                 }
2464                 put_bh(bh);
2465                 bh = next;
2466         } while (bh != head);
2467
2468         error = 0;
2469       done:
2470         if (nr == 0) {
2471                 /*
2472                  * if this page only had a direct item, it is very possible for
2473                  * no io to be required without there being an error.  Or,
2474                  * someone else could have locked them and sent them down the
2475                  * pipe without locking the page
2476                  */
2477                 bh = head;
2478                 do {
2479                         if (!buffer_uptodate(bh)) {
2480                                 partial = 1;
2481                                 break;
2482                         }
2483                         bh = bh->b_this_page;
2484                 } while (bh != head);
2485                 if (!partial)
2486                         SetPageUptodate(page);
2487                 end_page_writeback(page);
2488         }
2489         return error;
2490
2491       fail:
2492         /* catches various errors, we need to make sure any valid dirty blocks
2493          * get to the media.  The page is currently locked and not marked for
2494          * writeback
2495          */
2496         ClearPageUptodate(page);
2497         bh = head;
2498         do {
2499                 get_bh(bh);
2500                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2501                         lock_buffer(bh);
2502                         mark_buffer_async_write(bh);
2503                 } else {
2504                         /*
2505                          * clear any dirty bits that might have come from getting
2506                          * attached to a dirty page
2507                          */
2508                         clear_buffer_dirty(bh);
2509                 }
2510                 bh = bh->b_this_page;
2511         } while (bh != head);
2512         SetPageError(page);
2513         BUG_ON(PageWriteback(page));
2514         set_page_writeback(page);
2515         unlock_page(page);
2516         do {
2517                 struct buffer_head *next = bh->b_this_page;
2518                 if (buffer_async_write(bh)) {
2519                         clear_buffer_dirty(bh);
2520                         submit_bh(WRITE, bh);
2521                         nr++;
2522                 }
2523                 put_bh(bh);
2524                 bh = next;
2525         } while (bh != head);
2526         goto done;
2527 }
2528
2529 static int reiserfs_readpage(struct file *f, struct page *page)
2530 {
2531         return block_read_full_page(page, reiserfs_get_block);
2532 }
2533
2534 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2535 {
2536         struct inode *inode = page->mapping->host;
2537         reiserfs_wait_on_write_block(inode->i_sb);
2538         return reiserfs_write_full_page(page, wbc);
2539 }
2540
2541 static int reiserfs_write_begin(struct file *file,
2542                                 struct address_space *mapping,
2543                                 loff_t pos, unsigned len, unsigned flags,
2544                                 struct page **pagep, void **fsdata)
2545 {
2546         struct inode *inode;
2547         struct page *page;
2548         pgoff_t index;
2549         int ret;
2550         int old_ref = 0;
2551
2552         inode = mapping->host;
2553         *fsdata = 0;
2554         if (flags & AOP_FLAG_CONT_EXPAND &&
2555             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2556                 pos ++;
2557                 *fsdata = (void *)(unsigned long)flags;
2558         }
2559
2560         index = pos >> PAGE_CACHE_SHIFT;
2561         page = grab_cache_page_write_begin(mapping, index, flags);
2562         if (!page)
2563                 return -ENOMEM;
2564         *pagep = page;
2565
2566         reiserfs_wait_on_write_block(inode->i_sb);
2567         fix_tail_page_for_writing(page);
2568         if (reiserfs_transaction_running(inode->i_sb)) {
2569                 struct reiserfs_transaction_handle *th;
2570                 th = (struct reiserfs_transaction_handle *)current->
2571                     journal_info;
2572                 BUG_ON(!th->t_refcount);
2573                 BUG_ON(!th->t_trans_id);
2574                 old_ref = th->t_refcount;
2575                 th->t_refcount++;
2576         }
2577         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2578                                 reiserfs_get_block);
2579         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2580                 struct reiserfs_transaction_handle *th = current->journal_info;
2581                 /* this gets a little ugly.  If reiserfs_get_block returned an
2582                  * error and left a transacstion running, we've got to close it,
2583                  * and we've got to free handle if it was a persistent transaction.
2584                  *
2585                  * But, if we had nested into an existing transaction, we need
2586                  * to just drop the ref count on the handle.
2587                  *
2588                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2589                  * and it was a persistent trans.  Otherwise, it was nested above.
2590                  */
2591                 if (th->t_refcount > old_ref) {
2592                         if (old_ref)
2593                                 th->t_refcount--;
2594                         else {
2595                                 int err;
2596                                 reiserfs_write_lock(inode->i_sb);
2597                                 err = reiserfs_end_persistent_transaction(th);
2598                                 reiserfs_write_unlock(inode->i_sb);
2599                                 if (err)
2600                                         ret = err;
2601                         }
2602                 }
2603         }
2604         if (ret) {
2605                 unlock_page(page);
2606                 page_cache_release(page);
2607         }
2608         return ret;
2609 }
2610
2611 int reiserfs_prepare_write(struct file *f, struct page *page,
2612                            unsigned from, unsigned to)
2613 {
2614         struct inode *inode = page->mapping->host;
2615         int ret;
2616         int old_ref = 0;
2617
2618         reiserfs_write_unlock(inode->i_sb);
2619         reiserfs_wait_on_write_block(inode->i_sb);
2620         reiserfs_write_lock(inode->i_sb);
2621
2622         fix_tail_page_for_writing(page);
2623         if (reiserfs_transaction_running(inode->i_sb)) {
2624                 struct reiserfs_transaction_handle *th;
2625                 th = (struct reiserfs_transaction_handle *)current->
2626                     journal_info;
2627                 BUG_ON(!th->t_refcount);
2628                 BUG_ON(!th->t_trans_id);
2629                 old_ref = th->t_refcount;
2630                 th->t_refcount++;
2631         }
2632
2633         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2634         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2635                 struct reiserfs_transaction_handle *th = current->journal_info;
2636                 /* this gets a little ugly.  If reiserfs_get_block returned an
2637                  * error and left a transacstion running, we've got to close it,
2638                  * and we've got to free handle if it was a persistent transaction.
2639                  *
2640                  * But, if we had nested into an existing transaction, we need
2641                  * to just drop the ref count on the handle.
2642                  *
2643                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2644                  * and it was a persistent trans.  Otherwise, it was nested above.
2645                  */
2646                 if (th->t_refcount > old_ref) {
2647                         if (old_ref)
2648                                 th->t_refcount--;
2649                         else {
2650                                 int err;
2651                                 reiserfs_write_lock(inode->i_sb);
2652                                 err = reiserfs_end_persistent_transaction(th);
2653                                 reiserfs_write_unlock(inode->i_sb);
2654                                 if (err)
2655                                         ret = err;
2656                         }
2657                 }
2658         }
2659         return ret;
2660
2661 }
2662
2663 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2664 {
2665         return generic_block_bmap(as, block, reiserfs_bmap);
2666 }
2667
2668 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2669                               loff_t pos, unsigned len, unsigned copied,
2670                               struct page *page, void *fsdata)
2671 {
2672         struct inode *inode = page->mapping->host;
2673         int ret = 0;
2674         int update_sd = 0;
2675         struct reiserfs_transaction_handle *th;
2676         unsigned start;
2677         int lock_depth = 0;
2678         bool locked = false;
2679
2680         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2681                 pos ++;
2682
2683         reiserfs_wait_on_write_block(inode->i_sb);
2684         if (reiserfs_transaction_running(inode->i_sb))
2685                 th = current->journal_info;
2686         else
2687                 th = NULL;
2688
2689         start = pos & (PAGE_CACHE_SIZE - 1);
2690         if (unlikely(copied < len)) {
2691                 if (!PageUptodate(page))
2692                         copied = 0;
2693
2694                 page_zero_new_buffers(page, start + copied, start + len);
2695         }
2696         flush_dcache_page(page);
2697
2698         reiserfs_commit_page(inode, page, start, start + copied);
2699
2700         /* generic_commit_write does this for us, but does not update the
2701          ** transaction tracking stuff when the size changes.  So, we have
2702          ** to do the i_size updates here.
2703          */
2704         pos += copied;
2705
2706         if (pos > inode->i_size) {
2707                 struct reiserfs_transaction_handle myth;
2708                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2709                 locked = true;
2710                 /* If the file have grown beyond the border where it
2711                    can have a tail, unmark it as needing a tail
2712                    packing */
2713                 if ((have_large_tails(inode->i_sb)
2714                      && inode->i_size > i_block_size(inode) * 4)
2715                     || (have_small_tails(inode->i_sb)
2716                         && inode->i_size > i_block_size(inode)))
2717                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2718
2719                 ret = journal_begin(&myth, inode->i_sb, 1);
2720                 if (ret)
2721                         goto journal_error;
2722
2723                 reiserfs_update_inode_transaction(inode);
2724                 inode->i_size = pos;
2725                 /*
2726                  * this will just nest into our transaction.  It's important
2727                  * to use mark_inode_dirty so the inode gets pushed around on the
2728                  * dirty lists, and so that O_SYNC works as expected
2729                  */
2730                 mark_inode_dirty(inode);
2731                 reiserfs_update_sd(&myth, inode);
2732                 update_sd = 1;
2733                 ret = journal_end(&myth, inode->i_sb, 1);
2734                 if (ret)
2735                         goto journal_error;
2736         }
2737         if (th) {
2738                 if (!locked) {
2739                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2740                         locked = true;
2741                 }
2742                 if (!update_sd)
2743                         mark_inode_dirty(inode);
2744                 ret = reiserfs_end_persistent_transaction(th);
2745                 if (ret)
2746                         goto out;
2747         }
2748
2749       out:
2750         if (locked)
2751                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2752         unlock_page(page);
2753         page_cache_release(page);
2754         return ret == 0 ? copied : ret;
2755
2756       journal_error:
2757         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2758         locked = false;
2759         if (th) {
2760                 if (!update_sd)
2761                         reiserfs_update_sd(th, inode);
2762                 ret = reiserfs_end_persistent_transaction(th);
2763         }
2764         goto out;
2765 }
2766
2767 int reiserfs_commit_write(struct file *f, struct page *page,
2768                           unsigned from, unsigned to)
2769 {
2770         struct inode *inode = page->mapping->host;
2771         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2772         int ret = 0;
2773         int update_sd = 0;
2774         struct reiserfs_transaction_handle *th = NULL;
2775
2776         reiserfs_write_unlock(inode->i_sb);
2777         reiserfs_wait_on_write_block(inode->i_sb);
2778         reiserfs_write_lock(inode->i_sb);
2779
2780         if (reiserfs_transaction_running(inode->i_sb)) {
2781                 th = current->journal_info;
2782         }
2783         reiserfs_commit_page(inode, page, from, to);
2784
2785         /* generic_commit_write does this for us, but does not update the
2786          ** transaction tracking stuff when the size changes.  So, we have
2787          ** to do the i_size updates here.
2788          */
2789         if (pos > inode->i_size) {
2790                 struct reiserfs_transaction_handle myth;
2791                 /* If the file have grown beyond the border where it
2792                    can have a tail, unmark it as needing a tail
2793                    packing */
2794                 if ((have_large_tails(inode->i_sb)
2795                      && inode->i_size > i_block_size(inode) * 4)
2796                     || (have_small_tails(inode->i_sb)
2797                         && inode->i_size > i_block_size(inode)))
2798                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2799
2800                 ret = journal_begin(&myth, inode->i_sb, 1);
2801                 if (ret)
2802                         goto journal_error;
2803
2804                 reiserfs_update_inode_transaction(inode);
2805                 inode->i_size = pos;
2806                 /*
2807                  * this will just nest into our transaction.  It's important
2808                  * to use mark_inode_dirty so the inode gets pushed around on the
2809                  * dirty lists, and so that O_SYNC works as expected
2810                  */
2811                 mark_inode_dirty(inode);
2812                 reiserfs_update_sd(&myth, inode);
2813                 update_sd = 1;
2814                 ret = journal_end(&myth, inode->i_sb, 1);
2815                 if (ret)
2816                         goto journal_error;
2817         }
2818         if (th) {
2819                 if (!update_sd)
2820                         mark_inode_dirty(inode);
2821                 ret = reiserfs_end_persistent_transaction(th);
2822                 if (ret)
2823                         goto out;
2824         }
2825
2826       out:
2827         return ret;
2828
2829       journal_error:
2830         if (th) {
2831                 if (!update_sd)
2832                         reiserfs_update_sd(th, inode);
2833                 ret = reiserfs_end_persistent_transaction(th);
2834         }
2835
2836         return ret;
2837 }
2838
2839 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2840 {
2841         if (reiserfs_attrs(inode->i_sb)) {
2842                 if (sd_attrs & REISERFS_SYNC_FL)
2843                         inode->i_flags |= S_SYNC;
2844                 else
2845                         inode->i_flags &= ~S_SYNC;
2846                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2847                         inode->i_flags |= S_IMMUTABLE;
2848                 else
2849                         inode->i_flags &= ~S_IMMUTABLE;
2850                 if (sd_attrs & REISERFS_APPEND_FL)
2851                         inode->i_flags |= S_APPEND;
2852                 else
2853                         inode->i_flags &= ~S_APPEND;
2854                 if (sd_attrs & REISERFS_NOATIME_FL)
2855                         inode->i_flags |= S_NOATIME;
2856                 else
2857                         inode->i_flags &= ~S_NOATIME;
2858                 if (sd_attrs & REISERFS_NOTAIL_FL)
2859                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2860                 else
2861                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2862         }
2863 }
2864
2865 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2866 {
2867         if (reiserfs_attrs(inode->i_sb)) {
2868                 if (inode->i_flags & S_IMMUTABLE)
2869                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2870                 else
2871                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2872                 if (inode->i_flags & S_SYNC)
2873                         *sd_attrs |= REISERFS_SYNC_FL;
2874                 else
2875                         *sd_attrs &= ~REISERFS_SYNC_FL;
2876                 if (inode->i_flags & S_NOATIME)
2877                         *sd_attrs |= REISERFS_NOATIME_FL;
2878                 else
2879                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2880                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2881                         *sd_attrs |= REISERFS_NOTAIL_FL;
2882                 else
2883                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2884         }
2885 }
2886
2887 /* decide if this buffer needs to stay around for data logging or ordered
2888 ** write purposes
2889 */
2890 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2891 {
2892         int ret = 1;
2893         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2894
2895         lock_buffer(bh);
2896         spin_lock(&j->j_dirty_buffers_lock);
2897         if (!buffer_mapped(bh)) {
2898                 goto free_jh;
2899         }
2900         /* the page is locked, and the only places that log a data buffer
2901          * also lock the page.
2902          */
2903         if (reiserfs_file_data_log(inode)) {
2904                 /*
2905                  * very conservative, leave the buffer pinned if
2906                  * anyone might need it.
2907                  */
2908                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2909                         ret = 0;
2910                 }
2911         } else  if (buffer_dirty(bh)) {
2912                 struct reiserfs_journal_list *jl;
2913                 struct reiserfs_jh *jh = bh->b_private;
2914
2915                 /* why is this safe?
2916                  * reiserfs_setattr updates i_size in the on disk
2917                  * stat data before allowing vmtruncate to be called.
2918                  *
2919                  * If buffer was put onto the ordered list for this
2920                  * transaction, we know for sure either this transaction
2921                  * or an older one already has updated i_size on disk,
2922                  * and this ordered data won't be referenced in the file
2923                  * if we crash.
2924                  *
2925                  * if the buffer was put onto the ordered list for an older
2926                  * transaction, we need to leave it around
2927                  */
2928                 if (jh && (jl = jh->jl)
2929                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2930                         ret = 0;
2931         }
2932       free_jh:
2933         if (ret && bh->b_private) {
2934                 reiserfs_free_jh(bh);
2935         }
2936         spin_unlock(&j->j_dirty_buffers_lock);
2937         unlock_buffer(bh);
2938         return ret;
2939 }
2940
2941 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2942 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2943 {
2944         struct buffer_head *head, *bh, *next;
2945         struct inode *inode = page->mapping->host;
2946         unsigned int curr_off = 0;
2947         int ret = 1;
2948
2949         BUG_ON(!PageLocked(page));
2950
2951         if (offset == 0)
2952                 ClearPageChecked(page);
2953
2954         if (!page_has_buffers(page))
2955                 goto out;
2956
2957         head = page_buffers(page);
2958         bh = head;
2959         do {
2960                 unsigned int next_off = curr_off + bh->b_size;
2961                 next = bh->b_this_page;
2962
2963                 /*
2964                  * is this block fully invalidated?
2965                  */
2966                 if (offset <= curr_off) {
2967                         if (invalidatepage_can_drop(inode, bh))
2968                                 reiserfs_unmap_buffer(bh);
2969                         else
2970                                 ret = 0;
2971                 }
2972                 curr_off = next_off;
2973                 bh = next;
2974         } while (bh != head);
2975
2976         /*
2977          * We release buffers only if the entire page is being invalidated.
2978          * The get_block cached value has been unconditionally invalidated,
2979          * so real IO is not possible anymore.
2980          */
2981         if (!offset && ret) {
2982                 ret = try_to_release_page(page, 0);
2983                 /* maybe should BUG_ON(!ret); - neilb */
2984         }
2985       out:
2986         return;
2987 }
2988
2989 static int reiserfs_set_page_dirty(struct page *page)
2990 {
2991         struct inode *inode = page->mapping->host;
2992         if (reiserfs_file_data_log(inode)) {
2993                 SetPageChecked(page);
2994                 return __set_page_dirty_nobuffers(page);
2995         }
2996         return __set_page_dirty_buffers(page);
2997 }
2998
2999 /*
3000  * Returns 1 if the page's buffers were dropped.  The page is locked.
3001  *
3002  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3003  * in the buffers at page_buffers(page).
3004  *
3005  * even in -o notail mode, we can't be sure an old mount without -o notail
3006  * didn't create files with tails.
3007  */
3008 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3009 {
3010         struct inode *inode = page->mapping->host;
3011         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3012         struct buffer_head *head;
3013         struct buffer_head *bh;
3014         int ret = 1;
3015
3016         WARN_ON(PageChecked(page));
3017         spin_lock(&j->j_dirty_buffers_lock);
3018         head = page_buffers(page);
3019         bh = head;
3020         do {
3021                 if (bh->b_private) {
3022                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3023                                 reiserfs_free_jh(bh);
3024                         } else {
3025                                 ret = 0;
3026                                 break;
3027                         }
3028                 }
3029                 bh = bh->b_this_page;
3030         } while (bh != head);
3031         if (ret)
3032                 ret = try_to_free_buffers(page);
3033         spin_unlock(&j->j_dirty_buffers_lock);
3034         return ret;
3035 }
3036
3037 /* We thank Mingming Cao for helping us understand in great detail what
3038    to do in this section of the code. */
3039 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3040                                   const struct iovec *iov, loff_t offset,
3041                                   unsigned long nr_segs)
3042 {
3043         struct file *file = iocb->ki_filp;
3044         struct inode *inode = file->f_mapping->host;
3045
3046         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3047                                   offset, nr_segs,
3048                                   reiserfs_get_blocks_direct_io, NULL);
3049 }
3050
3051 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3052 {
3053         struct inode *inode = dentry->d_inode;
3054         int error;
3055         unsigned int ia_valid;
3056
3057         /* must be turned off for recursive notify_change calls */
3058         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3059
3060         reiserfs_write_lock(inode->i_sb);
3061         if (attr->ia_valid & ATTR_SIZE) {
3062                 /* version 2 items will be caught by the s_maxbytes check
3063                  ** done for us in vmtruncate
3064                  */
3065                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3066                     attr->ia_size > MAX_NON_LFS) {
3067                         error = -EFBIG;
3068                         goto out;
3069                 }
3070                 /* fill in hole pointers in the expanding truncate case. */
3071                 if (attr->ia_size > inode->i_size) {
3072                         error = generic_cont_expand_simple(inode, attr->ia_size);
3073                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3074                                 int err;
3075                                 struct reiserfs_transaction_handle th;
3076                                 /* we're changing at most 2 bitmaps, inode + super */
3077                                 err = journal_begin(&th, inode->i_sb, 4);
3078                                 if (!err) {
3079                                         reiserfs_discard_prealloc(&th, inode);
3080                                         err = journal_end(&th, inode->i_sb, 4);
3081                                 }
3082                                 if (err)
3083                                         error = err;
3084                         }
3085                         if (error)
3086                                 goto out;
3087                         /*
3088                          * file size is changed, ctime and mtime are
3089                          * to be updated
3090                          */
3091                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3092                 }
3093         }
3094
3095         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3096              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3097             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3098                 /* stat data of format v3.5 has 16 bit uid and gid */
3099                 error = -EINVAL;
3100                 goto out;
3101         }
3102
3103         error = inode_change_ok(inode, attr);
3104         if (!error) {
3105                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3106                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3107                         error = reiserfs_chown_xattrs(inode, attr);
3108
3109                         if (!error) {
3110                                 struct reiserfs_transaction_handle th;
3111                                 int jbegin_count =
3112                                     2 *
3113                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3114                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3115                                     2;
3116
3117                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3118                                 error =
3119                                     journal_begin(&th, inode->i_sb,
3120                                                   jbegin_count);
3121                                 if (error)
3122                                         goto out;
3123                                 error =
3124                                     vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
3125                                 if (error) {
3126                                         journal_end(&th, inode->i_sb,
3127                                                     jbegin_count);
3128                                         goto out;
3129                                 }
3130                                 /* Update corresponding info in inode so that everything is in
3131                                  * one transaction */
3132                                 if (attr->ia_valid & ATTR_UID)
3133                                         inode->i_uid = attr->ia_uid;
3134                                 if (attr->ia_valid & ATTR_GID)
3135                                         inode->i_gid = attr->ia_gid;
3136                                 mark_inode_dirty(inode);
3137                                 error =
3138                                     journal_end(&th, inode->i_sb, jbegin_count);
3139                         }
3140                 }
3141                 if (!error)
3142                         error = inode_setattr(inode, attr);
3143         }
3144
3145         if (!error && reiserfs_posixacl(inode->i_sb)) {
3146                 if (attr->ia_valid & ATTR_MODE)
3147                         error = reiserfs_acl_chmod(inode);
3148         }
3149
3150       out:
3151         reiserfs_write_unlock(inode->i_sb);
3152         return error;
3153 }
3154
3155 const struct address_space_operations reiserfs_address_space_operations = {
3156         .writepage = reiserfs_writepage,
3157         .readpage = reiserfs_readpage,
3158         .readpages = reiserfs_readpages,
3159         .releasepage = reiserfs_releasepage,
3160         .invalidatepage = reiserfs_invalidatepage,
3161         .sync_page = block_sync_page,
3162         .write_begin = reiserfs_write_begin,
3163         .write_end = reiserfs_write_end,
3164         .bmap = reiserfs_aop_bmap,
3165         .direct_IO = reiserfs_direct_IO,
3166         .set_page_dirty = reiserfs_set_page_dirty,
3167 };