git.oblomov.eu Git - linux-2.6/blob - 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         int fs_gen;
 255         b_blocknr_t blocknr;
 256         char *p = NULL;
 257         int chars;
 258         int ret;
 259         int result;
 260         int done = 0;
 261         unsigned long offset;
 262
 263         // prepare the key to look for the 'block'-th block of file
 264         make_cpu_key(&key, inode,
 265                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 266                      3);
 267
 268       research:
 269         result = search_for_position_by_key(inode->i_sb, &key, &path);
 270         if (result != POSITION_FOUND) {
 271                 pathrelse(&path);
 272                 if (p)
 273                         kunmap(bh_result->b_page);
 274                 if (result == IO_ERROR)
 275                         return -EIO;
 276                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 277                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 278                 if ((args & GET_BLOCK_NO_HOLE)
 279                     && !PageUptodate(bh_result->b_page)) {
 280                         return -ENOENT;
 281                 }
 282                 return 0;
 283         }
 284         //
 285         bh = get_last_bh(&path);
 286         ih = get_ih(&path);
 287         if (is_indirect_le_ih(ih)) {
 288                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 289
 290                 /* FIXME: here we could cache indirect item or part of it in
 291                    the inode to avoid search_by_key in case of subsequent
 292                    access to file */
 293                 blocknr = get_block_num(ind_item, path.pos_in_item);
 294                 ret = 0;
 295                 if (blocknr) {
 296                         map_bh(bh_result, inode->i_sb, blocknr);
 297                         if (path.pos_in_item ==
 298                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 299                                 set_buffer_boundary(bh_result);
 300                         }
 301                 } else
 302                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 303                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 304                 if ((args & GET_BLOCK_NO_HOLE)
 305                             && !PageUptodate(bh_result->b_page)) {
 306                         ret = -ENOENT;
 307                 }
 308
 309                 pathrelse(&path);
 310                 if (p)
 311                         kunmap(bh_result->b_page);
 312                 return ret;
 313         }
 314         // requested data are in direct item(s)
 315         if (!(args & GET_BLOCK_READ_DIRECT)) {
 316                 // we are called by bmap. FIXME: we can not map block of file
 317                 // when it is stored in direct item(s)
 318                 pathrelse(&path);
 319                 if (p)
 320                         kunmap(bh_result->b_page);
 321                 return -ENOENT;
 322         }
 323
 324         /* if we've got a direct item, and the buffer or page was uptodate,
 325          ** we don't want to pull data off disk again.  skip to the
 326          ** end, where we map the buffer and return
 327          */
 328         if (buffer_uptodate(bh_result)) {
 329                 goto finished;
 330         } else
 331                 /*
 332                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 333                  ** pages without any buffers.  If the page is up to date, we don't want
 334                  ** read old data off disk.  Set the up to date bit on the buffer instead
 335                  ** and jump to the end
 336                  */
 337         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 338                 set_buffer_uptodate(bh_result);
 339                 goto finished;
 340         }
 341         // read file tail into part of page
 342         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 343         fs_gen = get_generation(inode->i_sb);
 344         copy_item_head(&tmp_ih, ih);
 345
 346         /* we only want to kmap if we are reading the tail into the page.
 347          ** this is not the common case, so we don't kmap until we are
 348          ** sure we need to.  But, this means the item might move if
 349          ** kmap schedules
 350          */
 351         if (!p) {
 352                 p = (char *)kmap(bh_result->b_page);
 353                 if (fs_changed(fs_gen, inode->i_sb)
 354                     && item_moved(&tmp_ih, &path)) {
 355                         goto research;
 356                 }
 357         }
 358         p += offset;
 359         memset(p, 0, inode->i_sb->s_blocksize);
 360         do {
 361                 if (!is_direct_le_ih(ih)) {
 362                         BUG();
 363                 }
 364                 /* make sure we don't read more bytes than actually exist in
 365                  ** the file.  This can happen in odd cases where i_size isn't
 366                  ** correct, and when direct item padding results in a few
 367                  ** extra bytes at the end of the direct item
 368                  */
 369                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 370                         break;
 371                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 372                         chars =
 373                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 374                             path.pos_in_item;
 375                         done = 1;
 376                 } else {
 377                         chars = ih_item_len(ih) - path.pos_in_item;
 378                 }
 379                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 380
 381                 if (done)
 382                         break;
 383
 384                 p += chars;
 385
 386                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 387                         // we done, if read direct item is not the last item of
 388                         // node FIXME: we could try to check right delimiting key
 389                         // to see whether direct item continues in the right
 390                         // neighbor or rely on i_size
 391                         break;
 392
 393                 // update key to look for the next piece
 394                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 395                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 396                 if (result != POSITION_FOUND)
 397                         // i/o error most likely
 398                         break;
 399                 bh = get_last_bh(&path);
 400                 ih = get_ih(&path);
 401         } while (1);
 402
 403         flush_dcache_page(bh_result->b_page);
 404         kunmap(bh_result->b_page);
 405
 406       finished:
 407         pathrelse(&path);
 408
 409         if (result == IO_ERROR)
 410                 return -EIO;
 411
 412         /* this buffer has valid data, but isn't valid for io.  mapping it to
 413          * block #0 tells the rest of reiserfs it just has a tail in it
 414          */
 415         map_bh(bh_result, inode->i_sb, 0);
 416         set_buffer_uptodate(bh_result);
 417         return 0;
 418 }
 419
 420 // this is called to create file map. So, _get_block_create_0 will not
 421 // read direct item
 422 static int reiserfs_bmap(struct inode *inode, sector_t block,
 423                          struct buffer_head *bh_result, int create)
 424 {
 425         if (!file_capable(inode, block))
 426                 return -EFBIG;
 427
 428         reiserfs_write_lock(inode->i_sb);
 429         /* do not read the direct item */
 430         _get_block_create_0(inode, block, bh_result, 0);
 431         reiserfs_write_unlock(inode->i_sb);
 432         return 0;
 433 }
 434
 435 /* special version of get_block that is only used by grab_tail_page right
 436 ** now.  It is sent to block_prepare_write, and when you try to get a
 437 ** block past the end of the file (or a block from a hole) it returns
 438 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 439 ** be able to do i/o on the buffers returned, unless an error value
 440 ** is also returned.
 441 **
 442 ** So, this allows block_prepare_write to be used for reading a single block
 443 ** in a page.  Where it does not produce a valid page for holes, or past the
 444 ** end of the file.  This turns out to be exactly what we need for reading
 445 ** tails for conversion.
 446 **
 447 ** The point of the wrapper is forcing a certain value for create, even
 448 ** though the VFS layer is calling this function with create==1.  If you
 449 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 450 ** don't use this function.
 451 */
 452 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 453                                        struct buffer_head *bh_result,
 454                                        int create)
 455 {
 456         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 457 }
 458
 459 /* This is special helper for reiserfs_get_block in case we are executing
 460    direct_IO request. */
 461 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 462                                          sector_t iblock,
 463                                          struct buffer_head *bh_result,
 464                                          int create)
 465 {
 466         int ret;
 467
 468         bh_result->b_page = NULL;
 469
 470         /* We set the b_size before reiserfs_get_block call since it is
 471            referenced in convert_tail_for_hole() that may be called from
 472            reiserfs_get_block() */
 473         bh_result->b_size = (1 << inode->i_blkbits);
 474
 475         ret = reiserfs_get_block(inode, iblock, bh_result,
 476                                  create | GET_BLOCK_NO_DANGLE);
 477         if (ret)
 478                 goto out;
 479
 480         /* don't allow direct io onto tail pages */
 481         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 482                 /* make sure future calls to the direct io funcs for this offset
 483                  ** in the file fail by unmapping the buffer
 484                  */
 485                 clear_buffer_mapped(bh_result);
 486                 ret = -EINVAL;
 487         }
 488         /* Possible unpacked tail. Flush the data before pages have
 489            disappeared */
 490         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 491                 int err;
 492                 lock_kernel();
 493                 err = reiserfs_commit_for_inode(inode);
 494                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 495                 unlock_kernel();
 496                 if (err < 0)
 497                         ret = err;
 498         }
 499       out:
 500         return ret;
 501 }
 502
 503 /*
 504 ** helper function for when reiserfs_get_block is called for a hole
 505 ** but the file tail is still in a direct item
 506 ** bh_result is the buffer head for the hole
 507 ** tail_offset is the offset of the start of the tail in the file
 508 **
 509 ** This calls prepare_write, which will start a new transaction
 510 ** you should not be in a transaction, or have any paths held when you
 511 ** call this.
 512 */
 513 static int convert_tail_for_hole(struct inode *inode,
 514                                  struct buffer_head *bh_result,
 515                                  loff_t tail_offset)
 516 {
 517         unsigned long index;
 518         unsigned long tail_end;
 519         unsigned long tail_start;
 520         struct page *tail_page;
 521         struct page *hole_page = bh_result->b_page;
 522         int retval = 0;
 523
 524         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 525                 return -EIO;
 526
 527         /* always try to read until the end of the block */
 528         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 529         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 530
 531         index = tail_offset >> PAGE_CACHE_SHIFT;
 532         /* hole_page can be zero in case of direct_io, we are sure
 533            that we cannot get here if we write with O_DIRECT into
 534            tail page */
 535         if (!hole_page || index != hole_page->index) {
 536                 tail_page = grab_cache_page(inode->i_mapping, index);
 537                 retval = -ENOMEM;
 538                 if (!tail_page) {
 539                         goto out;
 540                 }
 541         } else {
 542                 tail_page = hole_page;
 543         }
 544
 545         /* we don't have to make sure the conversion did not happen while
 546          ** we were locking the page because anyone that could convert
 547          ** must first take i_mutex.
 548          **
 549          ** We must fix the tail page for writing because it might have buffers
 550          ** that are mapped, but have a block number of 0.  This indicates tail
 551          ** data that has been read directly into the page, and block_prepare_write
 552          ** won't trigger a get_block in this case.
 553          */
 554         fix_tail_page_for_writing(tail_page);
 555         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 556         if (retval)
 557                 goto unlock;
 558
 559         /* tail conversion might change the data in the page */
 560         flush_dcache_page(tail_page);
 561
 562         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 563
 564       unlock:
 565         if (tail_page != hole_page) {
 566                 unlock_page(tail_page);
 567                 page_cache_release(tail_page);
 568         }
 569       out:
 570         return retval;
 571 }
 572
 573 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 574                                   sector_t block,
 575                                   struct inode *inode,
 576                                   b_blocknr_t * allocated_block_nr,
 577                                   struct treepath *path, int flags)
 578 {
 579         BUG_ON(!th->t_trans_id);
 580
 581 #ifdef REISERFS_PREALLOCATE
 582         if (!(flags & GET_BLOCK_NO_IMUX)) {
 583                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 584                                                   path, block);
 585         }
 586 #endif
 587         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 588                                          block);
 589 }
 590
 591 int reiserfs_get_block(struct inode *inode, sector_t block,
 592                        struct buffer_head *bh_result, int create)
 593 {
 594         int repeat, retval = 0;
 595         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 596         INITIALIZE_PATH(path);
 597         int pos_in_item;
 598         struct cpu_key key;
 599         struct buffer_head *bh, *unbh = NULL;
 600         struct item_head *ih, tmp_ih;
 601         __le32 *item;
 602         int done;
 603         int fs_gen;
 604         struct reiserfs_transaction_handle *th = NULL;
 605         /* space reserved in transaction batch:
 606            . 3 balancings in direct->indirect conversion
 607            . 1 block involved into reiserfs_update_sd()
 608            XXX in practically impossible worst case direct2indirect()
 609            can incur (much) more than 3 balancings.
 610            quota update for user, group */
 611         int jbegin_count =
 612             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 613             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 614         int version;
 615         int dangle = 1;
 616         loff_t new_offset =
 617             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 618
 619         /* bad.... */
 620         reiserfs_write_lock(inode->i_sb);
 621         version = get_inode_item_key_version(inode);
 622
 623         if (!file_capable(inode, block)) {
 624                 reiserfs_write_unlock(inode->i_sb);
 625                 return -EFBIG;
 626         }
 627
 628         /* if !create, we aren't changing the FS, so we don't need to
 629          ** log anything, so we don't need to start a transaction
 630          */
 631         if (!(create & GET_BLOCK_CREATE)) {
 632                 int ret;
 633                 /* find number of block-th logical block of the file */
 634                 ret = _get_block_create_0(inode, block, bh_result,
 635                                           create | GET_BLOCK_READ_DIRECT);
 636                 reiserfs_write_unlock(inode->i_sb);
 637                 return ret;
 638         }
 639         /*
 640          * if we're already in a transaction, make sure to close
 641          * any new transactions we start in this func
 642          */
 643         if ((create & GET_BLOCK_NO_DANGLE) ||
 644             reiserfs_transaction_running(inode->i_sb))
 645                 dangle = 0;
 646
 647         /* If file is of such a size, that it might have a tail and tails are enabled
 648          ** we should mark it as possibly needing tail packing on close
 649          */
 650         if ((have_large_tails(inode->i_sb)
 651              && inode->i_size < i_block_size(inode) * 4)
 652             || (have_small_tails(inode->i_sb)
 653                 && inode->i_size < i_block_size(inode)))
 654                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 655
 656         /* set the key of the first byte in the 'block'-th block of file */
 657         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 658         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 659               start_trans:
 660                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 661                 if (!th) {
 662                         retval = -ENOMEM;
 663                         goto failure;
 664                 }
 665                 reiserfs_update_inode_transaction(inode);
 666         }
 667       research:
 668
 669         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 670         if (retval == IO_ERROR) {
 671                 retval = -EIO;
 672                 goto failure;
 673         }
 674
 675         bh = get_last_bh(&path);
 676         ih = get_ih(&path);
 677         item = get_item(&path);
 678         pos_in_item = path.pos_in_item;
 679
 680         fs_gen = get_generation(inode->i_sb);
 681         copy_item_head(&tmp_ih, ih);
 682
 683         if (allocation_needed
 684             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 685                 /* we have to allocate block for the unformatted node */
 686                 if (!th) {
 687                         pathrelse(&path);
 688                         goto start_trans;
 689                 }
 690
 691                 repeat =
 692                     _allocate_block(th, block, inode, &allocated_block_nr,
 693                                     &path, create);
 694
 695                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 696                         /* restart the transaction to give the journal a chance to free
 697                          ** some blocks.  releases the path, so we have to go back to
 698                          ** research if we succeed on the second try
 699                          */
 700                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 701                         retval = restart_transaction(th, inode, &path);
 702                         if (retval)
 703                                 goto failure;
 704                         repeat =
 705                             _allocate_block(th, block, inode,
 706                                             &allocated_block_nr, NULL, create);
 707
 708                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 709                                 goto research;
 710                         }
 711                         if (repeat == QUOTA_EXCEEDED)
 712                                 retval = -EDQUOT;
 713                         else
 714                                 retval = -ENOSPC;
 715                         goto failure;
 716                 }
 717
 718                 if (fs_changed(fs_gen, inode->i_sb)
 719                     && item_moved(&tmp_ih, &path)) {
 720                         goto research;
 721                 }
 722         }
 723
 724         if (indirect_item_found(retval, ih)) {
 725                 b_blocknr_t unfm_ptr;
 726                 /* 'block'-th block is in the file already (there is
 727                    corresponding cell in some indirect item). But it may be
 728                    zero unformatted node pointer (hole) */
 729                 unfm_ptr = get_block_num(item, pos_in_item);
 730                 if (unfm_ptr == 0) {
 731                         /* use allocated block to plug the hole */
 732                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 733                         if (fs_changed(fs_gen, inode->i_sb)
 734                             && item_moved(&tmp_ih, &path)) {
 735                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 736                                                                  bh);
 737                                 goto research;
 738                         }
 739                         set_buffer_new(bh_result);
 740                         if (buffer_dirty(bh_result)
 741                             && reiserfs_data_ordered(inode->i_sb))
 742                                 reiserfs_add_ordered_list(inode, bh_result);
 743                         put_block_num(item, pos_in_item, allocated_block_nr);
 744                         unfm_ptr = allocated_block_nr;
 745                         journal_mark_dirty(th, inode->i_sb, bh);
 746                         reiserfs_update_sd(th, inode);
 747                 }
 748                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 749                 pathrelse(&path);
 750                 retval = 0;
 751                 if (!dangle && th)
 752                         retval = reiserfs_end_persistent_transaction(th);
 753
 754                 reiserfs_write_unlock(inode->i_sb);
 755
 756                 /* the item was found, so new blocks were not added to the file
 757                  ** there is no need to make sure the inode is updated with this
 758                  ** transaction
 759                  */
 760                 return retval;
 761         }
 762
 763         if (!th) {
 764                 pathrelse(&path);
 765                 goto start_trans;
 766         }
 767
 768         /* desired position is not found or is in the direct item. We have
 769            to append file with holes up to 'block'-th block converting
 770            direct items to indirect one if necessary */
 771         done = 0;
 772         do {
 773                 if (is_statdata_le_ih(ih)) {
 774                         __le32 unp = 0;
 775                         struct cpu_key tmp_key;
 776
 777                         /* indirect item has to be inserted */
 778                         make_le_item_head(&tmp_ih, &key, version, 1,
 779                                           TYPE_INDIRECT, UNFM_P_SIZE,
 780                                           0 /* free_space */ );
 781
 782                         if (cpu_key_k_offset(&key) == 1) {
 783                                 /* we are going to add 'block'-th block to the file. Use
 784                                    allocated block for that */
 785                                 unp = cpu_to_le32(allocated_block_nr);
 786                                 set_block_dev_mapped(bh_result,
 787                                                      allocated_block_nr, inode);
 788                                 set_buffer_new(bh_result);
 789                                 done = 1;
 790                         }
 791                         tmp_key = key;  // ;)
 792                         set_cpu_key_k_offset(&tmp_key, 1);
 793                         PATH_LAST_POSITION(&path)++;
 794
 795                         retval =
 796                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 797                                                  inode, (char *)&unp);
 798                         if (retval) {
 799                                 reiserfs_free_block(th, inode,
 800                                                     allocated_block_nr, 1);
 801                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 802                         }
 803                         //mark_tail_converted (inode);
 804                 } else if (is_direct_le_ih(ih)) {
 805                         /* direct item has to be converted */
 806                         loff_t tail_offset;
 807
 808                         tail_offset =
 809                             ((le_ih_k_offset(ih) -
 810                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 811                         if (tail_offset == cpu_key_k_offset(&key)) {
 812                                 /* direct item we just found fits into block we have
 813                                    to map. Convert it into unformatted node: use
 814                                    bh_result for the conversion */
 815                                 set_block_dev_mapped(bh_result,
 816                                                      allocated_block_nr, inode);
 817                                 unbh = bh_result;
 818                                 done = 1;
 819                         } else {
 820                                 /* we have to padd file tail stored in direct item(s)
 821                                    up to block size and convert it to unformatted
 822                                    node. FIXME: this should also get into page cache */
 823
 824                                 pathrelse(&path);
 825                                 /*
 826                                  * ugly, but we can only end the transaction if
 827                                  * we aren't nested
 828                                  */
 829                                 BUG_ON(!th->t_refcount);
 830                                 if (th->t_refcount == 1) {
 831                                         retval =
 832                                             reiserfs_end_persistent_transaction
 833                                             (th);
 834                                         th = NULL;
 835                                         if (retval)
 836                                                 goto failure;
 837                                 }
 838
 839                                 retval =
 840                                     convert_tail_for_hole(inode, bh_result,
 841                                                           tail_offset);
 842                                 if (retval) {
 843                                         if (retval != -ENOSPC)
 844                                                 reiserfs_error(inode->i_sb,
 845                                                         "clm-6004",
 846                                                         "convert tail failed "
 847                                                         "inode %lu, error %d",
 848                                                         inode->i_ino,
 849                                                         retval);
 850                                         if (allocated_block_nr) {
 851                                                 /* the bitmap, the super, and the stat data == 3 */
 852                                                 if (!th)
 853                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 854                                                 if (th)
 855                                                         reiserfs_free_block(th,
 856                                                                             inode,
 857                                                                             allocated_block_nr,
 858                                                                             1);
 859                                         }
 860                                         goto failure;
 861                                 }
 862                                 goto research;
 863                         }
 864                         retval =
 865                             direct2indirect(th, inode, &path, unbh,
 866                                             tail_offset);
 867                         if (retval) {
 868                                 reiserfs_unmap_buffer(unbh);
 869                                 reiserfs_free_block(th, inode,
 870                                                     allocated_block_nr, 1);
 871                                 goto failure;
 872                         }
 873                         /* it is important the set_buffer_uptodate is done after
 874                          ** the direct2indirect.  The buffer might contain valid
 875                          ** data newer than the data on disk (read by readpage, changed,
 876                          ** and then sent here by writepage).  direct2indirect needs
 877                          ** to know if unbh was already up to date, so it can decide
 878                          ** if the data in unbh needs to be replaced with data from
 879                          ** the disk
 880                          */
 881                         set_buffer_uptodate(unbh);
 882
 883                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 884                            buffer will disappear shortly, so it should not be added to
 885                          */
 886                         if (unbh->b_page) {
 887                                 /* we've converted the tail, so we must
 888                                  ** flush unbh before the transaction commits
 889                                  */
 890                                 reiserfs_add_tail_list(inode, unbh);
 891
 892                                 /* mark it dirty now to prevent commit_write from adding
 893                                  ** this buffer to the inode's dirty buffer list
 894                                  */
 895                                 /*
 896                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 897                                  * It's still atomic, but it sets the page dirty too,
 898                                  * which makes it eligible for writeback at any time by the
 899                                  * VM (which was also the case with __mark_buffer_dirty())
 900                                  */
 901                                 mark_buffer_dirty(unbh);
 902                         }
 903                 } else {
 904                         /* append indirect item with holes if needed, when appending
 905                            pointer to 'block'-th block use block, which is already
 906                            allocated */
 907                         struct cpu_key tmp_key;
 908                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 909                         // one block which is a fastpath
 910                         unp_t *un;
 911                         __u64 max_to_insert =
 912                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 913                             UNFM_P_SIZE;
 914                         __u64 blocks_needed;
 915
 916                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 917                                "vs-804: invalid position for append");
 918                         /* indirect item has to be appended, set up key of that position */
 919                         make_cpu_key(&tmp_key, inode,
 920                                      le_key_k_offset(version,
 921                                                      &(ih->ih_key)) +
 922                                      op_bytes_number(ih,
 923                                                      inode->i_sb->s_blocksize),
 924                                      //pos_in_item * inode->i_sb->s_blocksize,
 925                                      TYPE_INDIRECT, 3); // key type is unimportant
 926
 927                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 928                                "green-805: invalid offset");
 929                         blocks_needed =
 930                             1 +
 931                             ((cpu_key_k_offset(&key) -
 932                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 933                              s_blocksize_bits);
 934
 935                         if (blocks_needed == 1) {
 936                                 un = &unf_single;
 937                         } else {
 938                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 939                                 if (!un) {
 940                                         un = &unf_single;
 941                                         blocks_needed = 1;
 942                                         max_to_insert = 0;
 943                                 }
 944                         }
 945                         if (blocks_needed <= max_to_insert) {
 946                                 /* we are going to add target block to the file. Use allocated
 947                                    block for that */
 948                                 un[blocks_needed - 1] =
 949                                     cpu_to_le32(allocated_block_nr);
 950                                 set_block_dev_mapped(bh_result,
 951                                                      allocated_block_nr, inode);
 952                                 set_buffer_new(bh_result);
 953                                 done = 1;
 954                         } else {
 955                                 /* paste hole to the indirect item */
 956                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 957                                    only have space for one block */
 958                                 blocks_needed =
 959                                     max_to_insert ? max_to_insert : 1;
 960                         }
 961                         retval =
 962                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 963                                                      (char *)un,
 964                                                      UNFM_P_SIZE *
 965                                                      blocks_needed);
 966
 967                         if (blocks_needed != 1)
 968                                 kfree(un);
 969
 970                         if (retval) {
 971                                 reiserfs_free_block(th, inode,
 972                                                     allocated_block_nr, 1);
 973                                 goto failure;
 974                         }
 975                         if (!done) {
 976                                 /* We need to mark new file size in case this function will be
 977                                    interrupted/aborted later on. And we may do this only for
 978                                    holes. */
 979                                 inode->i_size +=
 980                                     inode->i_sb->s_blocksize * blocks_needed;
 981                         }
 982                 }
 983
 984                 if (done == 1)
 985                         break;
 986
 987                 /* this loop could log more blocks than we had originally asked
 988                  ** for.  So, we have to allow the transaction to end if it is
 989                  ** too big or too full.  Update the inode so things are
 990                  ** consistent if we crash before the function returns
 991                  **
 992                  ** release the path so that anybody waiting on the path before
 993                  ** ending their transaction will be able to continue.
 994                  */
 995                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 996                         retval = restart_transaction(th, inode, &path);
 997                         if (retval)
 998                                 goto failure;
 999                 }
1000                 /* inserting indirect pointers for a hole can take a
1001                  ** long time.  reschedule if needed
1002                  */
1003                 cond_resched();
1004
1005                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1006                 if (retval == IO_ERROR) {
1007                         retval = -EIO;
1008                         goto failure;
1009                 }
1010                 if (retval == POSITION_FOUND) {
1011                         reiserfs_warning(inode->i_sb, "vs-825",
1012                                          "%K should not be found", &key);
1013                         retval = -EEXIST;
1014                         if (allocated_block_nr)
1015                                 reiserfs_free_block(th, inode,
1016                                                     allocated_block_nr, 1);
1017                         pathrelse(&path);
1018                         goto failure;
1019                 }
1020                 bh = get_last_bh(&path);
1021                 ih = get_ih(&path);
1022                 item = get_item(&path);
1023                 pos_in_item = path.pos_in_item;
1024         } while (1);
1025
1026         retval = 0;
1027
1028       failure:
1029         if (th && (!dangle || (retval && !th->t_trans_id))) {
1030                 int err;
1031                 if (th->t_trans_id)
1032                         reiserfs_update_sd(th, inode);
1033                 err = reiserfs_end_persistent_transaction(th);
1034                 if (err)
1035                         retval = err;
1036         }
1037
1038         reiserfs_write_unlock(inode->i_sb);
1039         reiserfs_check_path(&path);
1040         return retval;
1041 }
1042
1043 static int
1044 reiserfs_readpages(struct file *file, struct address_space *mapping,
1045                    struct list_head *pages, unsigned nr_pages)
1046 {
1047         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1048 }
1049
1050 /* Compute real number of used bytes by file
1051  * Following three functions can go away when we'll have enough space in stat item
1052  */
1053 static int real_space_diff(struct inode *inode, int sd_size)
1054 {
1055         int bytes;
1056         loff_t blocksize = inode->i_sb->s_blocksize;
1057
1058         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1059                 return sd_size;
1060
1061         /* End of file is also in full block with indirect reference, so round
1062          ** up to the next block.
1063          **
1064          ** there is just no way to know if the tail is actually packed
1065          ** on the file, so we have to assume it isn't.  When we pack the
1066          ** tail, we add 4 bytes to pretend there really is an unformatted
1067          ** node pointer
1068          */
1069         bytes =
1070             ((inode->i_size +
1071               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1072             sd_size;
1073         return bytes;
1074 }
1075
1076 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1077                                         int sd_size)
1078 {
1079         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1080                 return inode->i_size +
1081                     (loff_t) (real_space_diff(inode, sd_size));
1082         }
1083         return ((loff_t) real_space_diff(inode, sd_size)) +
1084             (((loff_t) blocks) << 9);
1085 }
1086
1087 /* Compute number of blocks used by file in ReiserFS counting */
1088 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1089 {
1090         loff_t bytes = inode_get_bytes(inode);
1091         loff_t real_space = real_space_diff(inode, sd_size);
1092
1093         /* keeps fsck and non-quota versions of reiserfs happy */
1094         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1095                 bytes += (loff_t) 511;
1096         }
1097
1098         /* files from before the quota patch might i_blocks such that
1099          ** bytes < real_space.  Deal with that here to prevent it from
1100          ** going negative.
1101          */
1102         if (bytes < real_space)
1103                 return 0;
1104         return (bytes - real_space) >> 9;
1105 }
1106
1107 //
1108 // BAD: new directories have stat data of new type and all other items
1109 // of old type. Version stored in the inode says about body items, so
1110 // in update_stat_data we can not rely on inode, but have to check
1111 // item version directly
1112 //
1113
1114 // called by read_locked_inode
1115 static void init_inode(struct inode *inode, struct treepath *path)
1116 {
1117         struct buffer_head *bh;
1118         struct item_head *ih;
1119         __u32 rdev;
1120         //int version = ITEM_VERSION_1;
1121
1122         bh = PATH_PLAST_BUFFER(path);
1123         ih = PATH_PITEM_HEAD(path);
1124
1125         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1126
1127         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1128         REISERFS_I(inode)->i_flags = 0;
1129         REISERFS_I(inode)->i_prealloc_block = 0;
1130         REISERFS_I(inode)->i_prealloc_count = 0;
1131         REISERFS_I(inode)->i_trans_id = 0;
1132         REISERFS_I(inode)->i_jl = NULL;
1133         mutex_init(&(REISERFS_I(inode)->i_mmap));
1134         reiserfs_init_acl_access(inode);
1135         reiserfs_init_acl_default(inode);
1136         reiserfs_init_xattr_rwsem(inode);
1137
1138         if (stat_data_v1(ih)) {
1139                 struct stat_data_v1 *sd =
1140                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1141                 unsigned long blocks;
1142
1143                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1144                 set_inode_sd_version(inode, STAT_DATA_V1);
1145                 inode->i_mode = sd_v1_mode(sd);
1146                 inode->i_nlink = sd_v1_nlink(sd);
1147                 inode->i_uid = sd_v1_uid(sd);
1148                 inode->i_gid = sd_v1_gid(sd);
1149                 inode->i_size = sd_v1_size(sd);
1150                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1151                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1152                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1153                 inode->i_atime.tv_nsec = 0;
1154                 inode->i_ctime.tv_nsec = 0;
1155                 inode->i_mtime.tv_nsec = 0;
1156
1157                 inode->i_blocks = sd_v1_blocks(sd);
1158                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1159                 blocks = (inode->i_size + 511) >> 9;
1160                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1161                 if (inode->i_blocks > blocks) {
1162                         // there was a bug in <=3.5.23 when i_blocks could take negative
1163                         // values. Starting from 3.5.17 this value could even be stored in
1164                         // stat data. For such files we set i_blocks based on file
1165                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1166                         // only updated if file's inode will ever change
1167                         inode->i_blocks = blocks;
1168                 }
1169
1170                 rdev = sd_v1_rdev(sd);
1171                 REISERFS_I(inode)->i_first_direct_byte =
1172                     sd_v1_first_direct_byte(sd);
1173                 /* an early bug in the quota code can give us an odd number for the
1174                  ** block count.  This is incorrect, fix it here.
1175                  */
1176                 if (inode->i_blocks & 1) {
1177                         inode->i_blocks++;
1178                 }
1179                 inode_set_bytes(inode,
1180                                 to_real_used_space(inode, inode->i_blocks,
1181                                                    SD_V1_SIZE));
1182                 /* nopack is initially zero for v1 objects. For v2 objects,
1183                    nopack is initialised from sd_attrs */
1184                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1185         } else {
1186                 // new stat data found, but object may have old items
1187                 // (directories and symlinks)
1188                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1189
1190                 inode->i_mode = sd_v2_mode(sd);
1191                 inode->i_nlink = sd_v2_nlink(sd);
1192                 inode->i_uid = sd_v2_uid(sd);
1193                 inode->i_size = sd_v2_size(sd);
1194                 inode->i_gid = sd_v2_gid(sd);
1195                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1196                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1197                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1198                 inode->i_ctime.tv_nsec = 0;
1199                 inode->i_mtime.tv_nsec = 0;
1200                 inode->i_atime.tv_nsec = 0;
1201                 inode->i_blocks = sd_v2_blocks(sd);
1202                 rdev = sd_v2_rdev(sd);
1203                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1204                         inode->i_generation =
1205                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1206                 else
1207                         inode->i_generation = sd_v2_generation(sd);
1208
1209                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1210                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1211                 else
1212                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1213                 REISERFS_I(inode)->i_first_direct_byte = 0;
1214                 set_inode_sd_version(inode, STAT_DATA_V2);
1215                 inode_set_bytes(inode,
1216                                 to_real_used_space(inode, inode->i_blocks,
1217                                                    SD_V2_SIZE));
1218                 /* read persistent inode attributes from sd and initalise
1219                    generic inode flags from them */
1220                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1221                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1222         }
1223
1224         pathrelse(path);
1225         if (S_ISREG(inode->i_mode)) {
1226                 inode->i_op = &reiserfs_file_inode_operations;
1227                 inode->i_fop = &reiserfs_file_operations;
1228                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1229         } else if (S_ISDIR(inode->i_mode)) {
1230                 inode->i_op = &reiserfs_dir_inode_operations;
1231                 inode->i_fop = &reiserfs_dir_operations;
1232         } else if (S_ISLNK(inode->i_mode)) {
1233                 inode->i_op = &reiserfs_symlink_inode_operations;
1234                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1235         } else {
1236                 inode->i_blocks = 0;
1237                 inode->i_op = &reiserfs_special_inode_operations;
1238                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1239         }
1240 }
1241
1242 // update new stat data with inode fields
1243 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1244 {
1245         struct stat_data *sd_v2 = (struct stat_data *)sd;
1246         __u16 flags;
1247
1248         set_sd_v2_mode(sd_v2, inode->i_mode);
1249         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1250         set_sd_v2_uid(sd_v2, inode->i_uid);
1251         set_sd_v2_size(sd_v2, size);
1252         set_sd_v2_gid(sd_v2, inode->i_gid);
1253         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1254         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1255         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1256         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1257         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1258                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1259         else
1260                 set_sd_v2_generation(sd_v2, inode->i_generation);
1261         flags = REISERFS_I(inode)->i_attrs;
1262         i_attrs_to_sd_attrs(inode, &flags);
1263         set_sd_v2_attrs(sd_v2, flags);
1264 }
1265
1266 // used to copy inode's fields to old stat data
1267 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1268 {
1269         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1270
1271         set_sd_v1_mode(sd_v1, inode->i_mode);
1272         set_sd_v1_uid(sd_v1, inode->i_uid);
1273         set_sd_v1_gid(sd_v1, inode->i_gid);
1274         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1275         set_sd_v1_size(sd_v1, size);
1276         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1277         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1278         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1279
1280         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1281                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1282         else
1283                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1284
1285         // Sigh. i_first_direct_byte is back
1286         set_sd_v1_first_direct_byte(sd_v1,
1287                                     REISERFS_I(inode)->i_first_direct_byte);
1288 }
1289
1290 /* NOTE, you must prepare the buffer head before sending it here,
1291 ** and then log it after the call
1292 */
1293 static void update_stat_data(struct treepath *path, struct inode *inode,
1294                              loff_t size)
1295 {
1296         struct buffer_head *bh;
1297         struct item_head *ih;
1298
1299         bh = PATH_PLAST_BUFFER(path);
1300         ih = PATH_PITEM_HEAD(path);
1301
1302         if (!is_statdata_le_ih(ih))
1303                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1304                                INODE_PKEY(inode), ih);
1305
1306         if (stat_data_v1(ih)) {
1307                 // path points to old stat data
1308                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1309         } else {
1310                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1311         }
1312
1313         return;
1314 }
1315
1316 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1317                              struct inode *inode, loff_t size)
1318 {
1319         struct cpu_key key;
1320         INITIALIZE_PATH(path);
1321         struct buffer_head *bh;
1322         int fs_gen;
1323         struct item_head *ih, tmp_ih;
1324         int retval;
1325
1326         BUG_ON(!th->t_trans_id);
1327
1328         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1329
1330         for (;;) {
1331                 int pos;
1332                 /* look for the object's stat data */
1333                 retval = search_item(inode->i_sb, &key, &path);
1334                 if (retval == IO_ERROR) {
1335                         reiserfs_error(inode->i_sb, "vs-13050",
1336                                        "i/o failure occurred trying to "
1337                                        "update %K stat data", &key);
1338                         return;
1339                 }
1340                 if (retval == ITEM_NOT_FOUND) {
1341                         pos = PATH_LAST_POSITION(&path);
1342                         pathrelse(&path);
1343                         if (inode->i_nlink == 0) {
1344                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1345                                 return;
1346                         }
1347                         reiserfs_warning(inode->i_sb, "vs-13060",
1348                                          "stat data of object %k (nlink == %d) "
1349                                          "not found (pos %d)",
1350                                          INODE_PKEY(inode), inode->i_nlink,
1351                                          pos);
1352                         reiserfs_check_path(&path);
1353                         return;
1354                 }
1355
1356                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1357                  ** FS might change.  We have to detect that, and loop back to the
1358                  ** search if the stat data item has moved
1359                  */
1360                 bh = get_last_bh(&path);
1361                 ih = get_ih(&path);
1362                 copy_item_head(&tmp_ih, ih);
1363                 fs_gen = get_generation(inode->i_sb);
1364                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1365                 if (fs_changed(fs_gen, inode->i_sb)
1366                     && item_moved(&tmp_ih, &path)) {
1367                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1368                         continue;       /* Stat_data item has been moved after scheduling. */
1369                 }
1370                 break;
1371         }
1372         update_stat_data(&path, inode, size);
1373         journal_mark_dirty(th, th->t_super, bh);
1374         pathrelse(&path);
1375         return;
1376 }
1377
1378 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1379 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1380 ** and clear the key in the private portion of the inode, otherwise a
1381 ** corresponding iput might try to delete whatever object the inode last
1382 ** represented.
1383 */
1384 static void reiserfs_make_bad_inode(struct inode *inode)
1385 {
1386         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1387         make_bad_inode(inode);
1388 }
1389
1390 //
1391 // initially this function was derived from minix or ext2's analog and
1392 // evolved as the prototype did
1393 //
1394
1395 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1396 {
1397         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1398         inode->i_ino = args->objectid;
1399         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1400         return 0;
1401 }
1402
1403 /* looks for stat data in the tree, and fills up the fields of in-core
1404    inode stat data fields */
1405 void reiserfs_read_locked_inode(struct inode *inode,
1406                                 struct reiserfs_iget_args *args)
1407 {
1408         INITIALIZE_PATH(path_to_sd);
1409         struct cpu_key key;
1410         unsigned long dirino;
1411         int retval;
1412
1413         dirino = args->dirid;
1414
1415         /* set version 1, version 2 could be used too, because stat data
1416            key is the same in both versions */
1417         key.version = KEY_FORMAT_3_5;
1418         key.on_disk_key.k_dir_id = dirino;
1419         key.on_disk_key.k_objectid = inode->i_ino;
1420         key.on_disk_key.k_offset = 0;
1421         key.on_disk_key.k_type = 0;
1422
1423         /* look for the object's stat data */
1424         retval = search_item(inode->i_sb, &key, &path_to_sd);
1425         if (retval == IO_ERROR) {
1426                 reiserfs_error(inode->i_sb, "vs-13070",
1427                                "i/o failure occurred trying to find "
1428                                "stat data of %K", &key);
1429                 reiserfs_make_bad_inode(inode);
1430                 return;
1431         }
1432         if (retval != ITEM_FOUND) {
1433                 /* a stale NFS handle can trigger this without it being an error */
1434                 pathrelse(&path_to_sd);
1435                 reiserfs_make_bad_inode(inode);
1436                 inode->i_nlink = 0;
1437                 return;
1438         }
1439
1440         init_inode(inode, &path_to_sd);
1441
1442         /* It is possible that knfsd is trying to access inode of a file
1443            that is being removed from the disk by some other thread. As we
1444            update sd on unlink all that is required is to check for nlink
1445            here. This bug was first found by Sizif when debugging
1446            SquidNG/Butterfly, forgotten, and found again after Philippe
1447            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1448
1449            More logical fix would require changes in fs/inode.c:iput() to
1450            remove inode from hash-table _after_ fs cleaned disk stuff up and
1451            in iget() to return NULL if I_FREEING inode is found in
1452            hash-table. */
1453         /* Currently there is one place where it's ok to meet inode with
1454            nlink==0: processing of open-unlinked and half-truncated files
1455            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1456         if ((inode->i_nlink == 0) &&
1457             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1458                 reiserfs_warning(inode->i_sb, "vs-13075",
1459                                  "dead inode read from disk %K. "
1460                                  "This is likely to be race with knfsd. Ignore",
1461                                  &key);
1462                 reiserfs_make_bad_inode(inode);
1463         }
1464
1465         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1466
1467 }
1468
1469 /**
1470  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1471  *
1472  * @inode:    inode from hash table to check
1473  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1474  *
1475  * This function is called by iget5_locked() to distinguish reiserfs inodes
1476  * having the same inode numbers. Such inodes can only exist due to some
1477  * error condition. One of them should be bad. Inodes with identical
1478  * inode numbers (objectids) are distinguished by parent directory ids.
1479  *
1480  */
1481 int reiserfs_find_actor(struct inode *inode, void *opaque)
1482 {
1483         struct reiserfs_iget_args *args;
1484
1485         args = opaque;
1486         /* args is already in CPU order */
1487         return (inode->i_ino == args->objectid) &&
1488             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1489 }
1490
1491 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1492 {
1493         struct inode *inode;
1494         struct reiserfs_iget_args args;
1495
1496         args.objectid = key->on_disk_key.k_objectid;
1497         args.dirid = key->on_disk_key.k_dir_id;
1498         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1499                              reiserfs_find_actor, reiserfs_init_locked_inode,
1500                              (void *)(&args));
1501         if (!inode)
1502                 return ERR_PTR(-ENOMEM);
1503
1504         if (inode->i_state & I_NEW) {
1505                 reiserfs_read_locked_inode(inode, &args);
1506                 unlock_new_inode(inode);
1507         }
1508
1509         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1510                 /* either due to i/o error or a stale NFS handle */
1511                 iput(inode);
1512                 inode = NULL;
1513         }
1514         return inode;
1515 }
1516
1517 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1518         u32 objectid, u32 dir_id, u32 generation)
1519
1520 {
1521         struct cpu_key key;
1522         struct inode *inode;
1523
1524         key.on_disk_key.k_objectid = objectid;
1525         key.on_disk_key.k_dir_id = dir_id;
1526         reiserfs_write_lock(sb);
1527         inode = reiserfs_iget(sb, &key);
1528         if (inode && !IS_ERR(inode) && generation != 0 &&
1529             generation != inode->i_generation) {
1530                 iput(inode);
1531                 inode = NULL;
1532         }
1533         reiserfs_write_unlock(sb);
1534
1535         return d_obtain_alias(inode);
1536 }
1537
1538 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1539                 int fh_len, int fh_type)
1540 {
1541         /* fhtype happens to reflect the number of u32s encoded.
1542          * due to a bug in earlier code, fhtype might indicate there
1543          * are more u32s then actually fitted.
1544          * so if fhtype seems to be more than len, reduce fhtype.
1545          * Valid types are:
1546          *   2 - objectid + dir_id - legacy support
1547          *   3 - objectid + dir_id + generation
1548          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1549          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1550          *   6 - as above plus generation of directory
1551          * 6 does not fit in NFSv2 handles
1552          */
1553         if (fh_type > fh_len) {
1554                 if (fh_type != 6 || fh_len != 5)
1555                         reiserfs_warning(sb, "reiserfs-13077",
1556                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1557                                 fh_type, fh_len);
1558                 fh_type = 5;
1559         }
1560
1561         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1562                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1563 }
1564
1565 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1566                 int fh_len, int fh_type)
1567 {
1568         if (fh_type < 4)
1569                 return NULL;
1570
1571         return reiserfs_get_dentry(sb,
1572                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1573                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1574                 (fh_type == 6) ? fid->raw[5] : 0);
1575 }
1576
1577 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1578                        int need_parent)
1579 {
1580         struct inode *inode = dentry->d_inode;
1581         int maxlen = *lenp;
1582
1583         if (maxlen < 3)
1584                 return 255;
1585
1586         data[0] = inode->i_ino;
1587         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1588         data[2] = inode->i_generation;
1589         *lenp = 3;
1590         /* no room for directory info? return what we've stored so far */
1591         if (maxlen < 5 || !need_parent)
1592                 return 3;
1593
1594         spin_lock(&dentry->d_lock);
1595         inode = dentry->d_parent->d_inode;
1596         data[3] = inode->i_ino;
1597         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1598         *lenp = 5;
1599         if (maxlen >= 6) {
1600                 data[5] = inode->i_generation;
1601                 *lenp = 6;
1602         }
1603         spin_unlock(&dentry->d_lock);
1604         return *lenp;
1605 }
1606
1607 /* looks for stat data, then copies fields to it, marks the buffer
1608    containing stat data as dirty */
1609 /* reiserfs inodes are never really dirty, since the dirty inode call
1610 ** always logs them.  This call allows the VFS inode marking routines
1611 ** to properly mark inodes for datasync and such, but only actually
1612 ** does something when called for a synchronous update.
1613 */
1614 int reiserfs_write_inode(struct inode *inode, int do_sync)
1615 {
1616         struct reiserfs_transaction_handle th;
1617         int jbegin_count = 1;
1618
1619         if (inode->i_sb->s_flags & MS_RDONLY)
1620                 return -EROFS;
1621         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1622          ** these cases are just when the system needs ram, not when the
1623          ** inode needs to reach disk for safety, and they can safely be
1624          ** ignored because the altered inode has already been logged.
1625          */
1626         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1627                 reiserfs_write_lock(inode->i_sb);
1628                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1629                         reiserfs_update_sd(&th, inode);
1630                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1631                 }
1632                 reiserfs_write_unlock(inode->i_sb);
1633         }
1634         return 0;
1635 }
1636
1637 /* stat data of new object is inserted already, this inserts the item
1638    containing "." and ".." entries */
1639 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1640                                   struct inode *inode,
1641                                   struct item_head *ih, struct treepath *path,
1642                                   struct inode *dir)
1643 {
1644         struct super_block *sb = th->t_super;
1645         char empty_dir[EMPTY_DIR_SIZE];
1646         char *body = empty_dir;
1647         struct cpu_key key;
1648         int retval;
1649
1650         BUG_ON(!th->t_trans_id);
1651
1652         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1653                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1654                       TYPE_DIRENTRY, 3 /*key length */ );
1655
1656         /* compose item head for new item. Directories consist of items of
1657            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1658            is done by reiserfs_new_inode */
1659         if (old_format_only(sb)) {
1660                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1661                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1662
1663                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1664                                        ih->ih_key.k_objectid,
1665                                        INODE_PKEY(dir)->k_dir_id,
1666                                        INODE_PKEY(dir)->k_objectid);
1667         } else {
1668                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1669                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1670
1671                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1672                                     ih->ih_key.k_objectid,
1673                                     INODE_PKEY(dir)->k_dir_id,
1674                                     INODE_PKEY(dir)->k_objectid);
1675         }
1676
1677         /* look for place in the tree for new item */
1678         retval = search_item(sb, &key, path);
1679         if (retval == IO_ERROR) {
1680                 reiserfs_error(sb, "vs-13080",
1681                                "i/o failure occurred creating new directory");
1682                 return -EIO;
1683         }
1684         if (retval == ITEM_FOUND) {
1685                 pathrelse(path);
1686                 reiserfs_warning(sb, "vs-13070",
1687                                  "object with this key exists (%k)",
1688                                  &(ih->ih_key));
1689                 return -EEXIST;
1690         }
1691
1692         /* insert item, that is empty directory item */
1693         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1694 }
1695
1696 /* stat data of object has been inserted, this inserts the item
1697    containing the body of symlink */
1698 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1699                                 struct item_head *ih,
1700                                 struct treepath *path, const char *symname,
1701                                 int item_len)
1702 {
1703         struct super_block *sb = th->t_super;
1704         struct cpu_key key;
1705         int retval;
1706
1707         BUG_ON(!th->t_trans_id);
1708
1709         _make_cpu_key(&key, KEY_FORMAT_3_5,
1710                       le32_to_cpu(ih->ih_key.k_dir_id),
1711                       le32_to_cpu(ih->ih_key.k_objectid),
1712                       1, TYPE_DIRECT, 3 /*key length */ );
1713
1714         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1715                           0 /*free_space */ );
1716
1717         /* look for place in the tree for new item */
1718         retval = search_item(sb, &key, path);
1719         if (retval == IO_ERROR) {
1720                 reiserfs_error(sb, "vs-13080",
1721                                "i/o failure occurred creating new symlink");
1722                 return -EIO;
1723         }
1724         if (retval == ITEM_FOUND) {
1725                 pathrelse(path);
1726                 reiserfs_warning(sb, "vs-13080",
1727                                  "object with this key exists (%k)",
1728                                  &(ih->ih_key));
1729                 return -EEXIST;
1730         }
1731
1732         /* insert item, that is body of symlink */
1733         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1734 }
1735
1736 /* inserts the stat data into the tree, and then calls
1737    reiserfs_new_directory (to insert ".", ".." item if new object is
1738    directory) or reiserfs_new_symlink (to insert symlink body if new
1739    object is symlink) or nothing (if new object is regular file)
1740
1741    NOTE! uid and gid must already be set in the inode.  If we return
1742    non-zero due to an error, we have to drop the quota previously allocated
1743    for the fresh inode.  This can only be done outside a transaction, so
1744    if we return non-zero, we also end the transaction.  */
1745 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1746                        struct inode *dir, int mode, const char *symname,
1747                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1748                           strlen (symname) for symlinks) */
1749                        loff_t i_size, struct dentry *dentry,
1750                        struct inode *inode,
1751                        struct reiserfs_security_handle *security)
1752 {
1753         struct super_block *sb;
1754         struct reiserfs_iget_args args;
1755         INITIALIZE_PATH(path_to_key);
1756         struct cpu_key key;
1757         struct item_head ih;
1758         struct stat_data sd;
1759         int retval;
1760         int err;
1761
1762         BUG_ON(!th->t_trans_id);
1763
1764         if (vfs_dq_alloc_inode(inode)) {
1765                 err = -EDQUOT;
1766                 goto out_end_trans;
1767         }
1768         if (!dir->i_nlink) {
1769                 err = -EPERM;
1770                 goto out_bad_inode;
1771         }
1772
1773         sb = dir->i_sb;
1774
1775         /* item head of new item */
1776         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1777         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1778         if (!ih.ih_key.k_objectid) {
1779                 err = -ENOMEM;
1780                 goto out_bad_inode;
1781         }
1782         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1783         if (old_format_only(sb))
1784                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1785                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1786         else
1787                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1788                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1789         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1790         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1791         if (insert_inode_locked4(inode, args.objectid,
1792                              reiserfs_find_actor, &args) < 0) {
1793                 err = -EINVAL;
1794                 goto out_bad_inode;
1795         }
1796         if (old_format_only(sb))
1797                 /* not a perfect generation count, as object ids can be reused, but
1798                  ** this is as good as reiserfs can do right now.
1799                  ** note that the private part of inode isn't filled in yet, we have
1800                  ** to use the directory.
1801                  */
1802                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1803         else
1804 #if defined( USE_INODE_GENERATION_COUNTER )
1805                 inode->i_generation =
1806                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1807 #else
1808                 inode->i_generation = ++event;
1809 #endif
1810
1811         /* fill stat data */
1812         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1813
1814         /* uid and gid must already be set by the caller for quota init */
1815
1816         /* symlink cannot be immutable or append only, right? */
1817         if (S_ISLNK(inode->i_mode))
1818                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1819
1820         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1821         inode->i_size = i_size;
1822         inode->i_blocks = 0;
1823         inode->i_bytes = 0;
1824         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1825             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1826
1827         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1828         REISERFS_I(inode)->i_flags = 0;
1829         REISERFS_I(inode)->i_prealloc_block = 0;
1830         REISERFS_I(inode)->i_prealloc_count = 0;
1831         REISERFS_I(inode)->i_trans_id = 0;
1832         REISERFS_I(inode)->i_jl = NULL;
1833         REISERFS_I(inode)->i_attrs =
1834             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1835         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1836         mutex_init(&(REISERFS_I(inode)->i_mmap));
1837         reiserfs_init_acl_access(inode);
1838         reiserfs_init_acl_default(inode);
1839         reiserfs_init_xattr_rwsem(inode);
1840
1841         /* key to search for correct place for new stat data */
1842         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1843                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1844                       TYPE_STAT_DATA, 3 /*key length */ );
1845
1846         /* find proper place for inserting of stat data */
1847         retval = search_item(sb, &key, &path_to_key);
1848         if (retval == IO_ERROR) {
1849                 err = -EIO;
1850                 goto out_bad_inode;
1851         }
1852         if (retval == ITEM_FOUND) {
1853                 pathrelse(&path_to_key);
1854                 err = -EEXIST;
1855                 goto out_bad_inode;
1856         }
1857         if (old_format_only(sb)) {
1858                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1859                         pathrelse(&path_to_key);
1860                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1861                         err = -EINVAL;
1862                         goto out_bad_inode;
1863                 }
1864                 inode2sd_v1(&sd, inode, inode->i_size);
1865         } else {
1866                 inode2sd(&sd, inode, inode->i_size);
1867         }
1868         // store in in-core inode the key of stat data and version all
1869         // object items will have (directory items will have old offset
1870         // format, other new objects will consist of new items)
1871         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1872                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1873         else
1874                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1875         if (old_format_only(sb))
1876                 set_inode_sd_version(inode, STAT_DATA_V1);
1877         else
1878                 set_inode_sd_version(inode, STAT_DATA_V2);
1879
1880         /* insert the stat data into the tree */
1881 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1882         if (REISERFS_I(dir)->new_packing_locality)
1883                 th->displace_new_blocks = 1;
1884 #endif
1885         retval =
1886             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1887                                  (char *)(&sd));
1888         if (retval) {
1889                 err = retval;
1890                 reiserfs_check_path(&path_to_key);
1891                 goto out_bad_inode;
1892         }
1893 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1894         if (!th->displace_new_blocks)
1895                 REISERFS_I(dir)->new_packing_locality = 0;
1896 #endif
1897         if (S_ISDIR(mode)) {
1898                 /* insert item with "." and ".." */
1899                 retval =
1900                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1901         }
1902
1903         if (S_ISLNK(mode)) {
1904                 /* insert body of symlink */
1905                 if (!old_format_only(sb))
1906                         i_size = ROUND_UP(i_size);
1907                 retval =
1908                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1909                                          i_size);
1910         }
1911         if (retval) {
1912                 err = retval;
1913                 reiserfs_check_path(&path_to_key);
1914                 journal_end(th, th->t_super, th->t_blocks_allocated);
1915                 goto out_inserted_sd;
1916         }
1917
1918         if (reiserfs_posixacl(inode->i_sb)) {
1919                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1920                 if (retval) {
1921                         err = retval;
1922                         reiserfs_check_path(&path_to_key);
1923                         journal_end(th, th->t_super, th->t_blocks_allocated);
1924                         goto out_inserted_sd;
1925                 }
1926         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1927                 reiserfs_warning(inode->i_sb, "jdm-13090",
1928                                  "ACLs aren't enabled in the fs, "
1929                                  "but vfs thinks they are!");
1930         } else if (IS_PRIVATE(dir))
1931                 inode->i_flags |= S_PRIVATE;
1932
1933         if (security->name) {
1934                 retval = reiserfs_security_write(th, inode, security);
1935                 if (retval) {
1936                         err = retval;
1937                         reiserfs_check_path(&path_to_key);
1938                         retval = journal_end(th, th->t_super,
1939                                              th->t_blocks_allocated);
1940                         if (retval)
1941                                 err = retval;
1942                         goto out_inserted_sd;
1943                 }
1944         }
1945
1946         reiserfs_update_sd(th, inode);
1947         reiserfs_check_path(&path_to_key);
1948
1949         return 0;
1950
1951 /* it looks like you can easily compress these two goto targets into
1952  * one.  Keeping it like this doesn't actually hurt anything, and they
1953  * are place holders for what the quota code actually needs.
1954  */
1955       out_bad_inode:
1956         /* Invalidate the object, nothing was inserted yet */
1957         INODE_PKEY(inode)->k_objectid = 0;
1958
1959         /* Quota change must be inside a transaction for journaling */
1960         vfs_dq_free_inode(inode);
1961
1962       out_end_trans:
1963         journal_end(th, th->t_super, th->t_blocks_allocated);
1964         /* Drop can be outside and it needs more credits so it's better to have it outside */
1965         vfs_dq_drop(inode);
1966         inode->i_flags |= S_NOQUOTA;
1967         make_bad_inode(inode);
1968
1969       out_inserted_sd:
1970         inode->i_nlink = 0;
1971         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1972         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1973         iput(inode);
1974         return err;
1975 }
1976
1977 /*
1978 ** finds the tail page in the page cache,
1979 ** reads the last block in.
1980 **
1981 ** On success, page_result is set to a locked, pinned page, and bh_result
1982 ** is set to an up to date buffer for the last block in the file.  returns 0.
1983 **
1984 ** tail conversion is not done, so bh_result might not be valid for writing
1985 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1986 ** trying to write the block.
1987 **
1988 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1989 */
1990 static int grab_tail_page(struct inode *inode,
1991                           struct page **page_result,
1992                           struct buffer_head **bh_result)
1993 {
1994
1995         /* we want the page with the last byte in the file,
1996          ** not the page that will hold the next byte for appending
1997          */
1998         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1999         unsigned long pos = 0;
2000         unsigned long start = 0;
2001         unsigned long blocksize = inode->i_sb->s_blocksize;
2002         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2003         struct buffer_head *bh;
2004         struct buffer_head *head;
2005         struct page *page;
2006         int error;
2007
2008         /* we know that we are only called with inode->i_size > 0.
2009          ** we also know that a file tail can never be as big as a block
2010          ** If i_size % blocksize == 0, our file is currently block aligned
2011          ** and it won't need converting or zeroing after a truncate.
2012          */
2013         if ((offset & (blocksize - 1)) == 0) {
2014                 return -ENOENT;
2015         }
2016         page = grab_cache_page(inode->i_mapping, index);
2017         error = -ENOMEM;
2018         if (!page) {
2019                 goto out;
2020         }
2021         /* start within the page of the last block in the file */
2022         start = (offset / blocksize) * blocksize;
2023
2024         error = block_prepare_write(page, start, offset,
2025                                     reiserfs_get_block_create_0);
2026         if (error)
2027                 goto unlock;
2028
2029         head = page_buffers(page);
2030         bh = head;
2031         do {
2032                 if (pos >= start) {
2033                         break;
2034                 }
2035                 bh = bh->b_this_page;
2036                 pos += blocksize;
2037         } while (bh != head);
2038
2039         if (!buffer_uptodate(bh)) {
2040                 /* note, this should never happen, prepare_write should
2041                  ** be taking care of this for us.  If the buffer isn't up to date,
2042                  ** I've screwed up the code to find the buffer, or the code to
2043                  ** call prepare_write
2044                  */
2045                 reiserfs_error(inode->i_sb, "clm-6000",
2046                                "error reading block %lu", bh->b_blocknr);
2047                 error = -EIO;
2048                 goto unlock;
2049         }
2050         *bh_result = bh;
2051         *page_result = page;
2052
2053       out:
2054         return error;
2055
2056       unlock:
2057         unlock_page(page);
2058         page_cache_release(page);
2059         return error;
2060 }
2061
2062 /*
2063 ** vfs version of truncate file.  Must NOT be called with
2064 ** a transaction already started.
2065 **
2066 ** some code taken from block_truncate_page
2067 */
2068 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2069 {
2070         struct reiserfs_transaction_handle th;
2071         /* we want the offset for the first byte after the end of the file */
2072         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2073         unsigned blocksize = inode->i_sb->s_blocksize;
2074         unsigned length;
2075         struct page *page = NULL;
2076         int error;
2077         struct buffer_head *bh = NULL;
2078         int err2;
2079
2080         reiserfs_write_lock(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(inode->i_sb);
2150         return 0;
2151       out:
2152         if (page) {
2153                 unlock_page(page);
2154                 page_cache_release(page);
2155         }
2156         reiserfs_write_unlock(inode->i_sb);
2157         return error;
2158 }
2159
2160 static int map_block_for_writepage(struct inode *inode,
2161                                    struct buffer_head *bh_result,
2162                                    unsigned long block)
2163 {
2164         struct reiserfs_transaction_handle th;
2165         int fs_gen;
2166         struct item_head tmp_ih;
2167         struct item_head *ih;
2168         struct buffer_head *bh;
2169         __le32 *item;
2170         struct cpu_key key;
2171         INITIALIZE_PATH(path);
2172         int pos_in_item;
2173         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2174         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2175         int retval;
2176         int use_get_block = 0;
2177         int bytes_copied = 0;
2178         int copy_size;
2179         int trans_running = 0;
2180
2181         /* catch places below that try to log something without starting a trans */
2182         th.t_trans_id = 0;
2183
2184         if (!buffer_uptodate(bh_result)) {
2185                 return -EIO;
2186         }
2187
2188         kmap(bh_result->b_page);
2189       start_over:
2190         reiserfs_write_lock(inode->i_sb);
2191         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2192
2193       research:
2194         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2195         if (retval != POSITION_FOUND) {
2196                 use_get_block = 1;
2197                 goto out;
2198         }
2199
2200         bh = get_last_bh(&path);
2201         ih = get_ih(&path);
2202         item = get_item(&path);
2203         pos_in_item = path.pos_in_item;
2204
2205         /* we've found an unformatted node */
2206         if (indirect_item_found(retval, ih)) {
2207                 if (bytes_copied > 0) {
2208                         reiserfs_warning(inode->i_sb, "clm-6002",
2209                                          "bytes_copied %d", bytes_copied);
2210                 }
2211                 if (!get_block_num(item, pos_in_item)) {
2212                         /* crap, we are writing to a hole */
2213                         use_get_block = 1;
2214                         goto out;
2215                 }
2216                 set_block_dev_mapped(bh_result,
2217                                      get_block_num(item, pos_in_item), inode);
2218         } else if (is_direct_le_ih(ih)) {
2219                 char *p;
2220                 p = page_address(bh_result->b_page);
2221                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2222                 copy_size = ih_item_len(ih) - pos_in_item;
2223
2224                 fs_gen = get_generation(inode->i_sb);
2225                 copy_item_head(&tmp_ih, ih);
2226
2227                 if (!trans_running) {
2228                         /* vs-3050 is gone, no need to drop the path */
2229                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2230                         if (retval)
2231                                 goto out;
2232                         reiserfs_update_inode_transaction(inode);
2233                         trans_running = 1;
2234                         if (fs_changed(fs_gen, inode->i_sb)
2235                             && item_moved(&tmp_ih, &path)) {
2236                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2237                                                                  bh);
2238                                 goto research;
2239                         }
2240                 }
2241
2242                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2243
2244                 if (fs_changed(fs_gen, inode->i_sb)
2245                     && item_moved(&tmp_ih, &path)) {
2246                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2247                         goto research;
2248                 }
2249
2250                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2251                        copy_size);
2252
2253                 journal_mark_dirty(&th, inode->i_sb, bh);
2254                 bytes_copied += copy_size;
2255                 set_block_dev_mapped(bh_result, 0, inode);
2256
2257                 /* are there still bytes left? */
2258                 if (bytes_copied < bh_result->b_size &&
2259                     (byte_offset + bytes_copied) < inode->i_size) {
2260                         set_cpu_key_k_offset(&key,
2261                                              cpu_key_k_offset(&key) +
2262                                              copy_size);
2263                         goto research;
2264                 }
2265         } else {
2266                 reiserfs_warning(inode->i_sb, "clm-6003",
2267                                  "bad item inode %lu", inode->i_ino);
2268                 retval = -EIO;
2269                 goto out;
2270         }
2271         retval = 0;
2272
2273       out:
2274         pathrelse(&path);
2275         if (trans_running) {
2276                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2277                 if (err)
2278                         retval = err;
2279                 trans_running = 0;
2280         }
2281         reiserfs_write_unlock(inode->i_sb);
2282
2283         /* this is where we fill in holes in the file. */
2284         if (use_get_block) {
2285                 retval = reiserfs_get_block(inode, block, bh_result,
2286                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2287                                             | GET_BLOCK_NO_DANGLE);
2288                 if (!retval) {
2289                         if (!buffer_mapped(bh_result)
2290                             || bh_result->b_blocknr == 0) {
2291                                 /* get_block failed to find a mapped unformatted node. */
2292                                 use_get_block = 0;
2293                                 goto start_over;
2294                         }
2295                 }
2296         }
2297         kunmap(bh_result->b_page);
2298
2299         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2300                 /* we've copied data from the page into the direct item, so the
2301                  * buffer in the page is now clean, mark it to reflect that.
2302                  */
2303                 lock_buffer(bh_result);
2304                 clear_buffer_dirty(bh_result);
2305                 unlock_buffer(bh_result);
2306         }
2307         return retval;
2308 }
2309
2310 /*
2311  * mason@suse.com: updated in 2.5.54 to follow the same general io
2312  * start/recovery path as __block_write_full_page, along with special
2313  * code to handle reiserfs tails.
2314  */
2315 static int reiserfs_write_full_page(struct page *page,
2316                                     struct writeback_control *wbc)
2317 {
2318         struct inode *inode = page->mapping->host;
2319         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2320         int error = 0;
2321         unsigned long block;
2322         sector_t last_block;
2323         struct buffer_head *head, *bh;
2324         int partial = 0;
2325         int nr = 0;
2326         int checked = PageChecked(page);
2327         struct reiserfs_transaction_handle th;
2328         struct super_block *s = inode->i_sb;
2329         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2330         th.t_trans_id = 0;
2331
2332         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2333         if (checked && (current->flags & PF_MEMALLOC)) {
2334                 redirty_page_for_writepage(wbc, page);
2335                 unlock_page(page);
2336                 return 0;
2337         }
2338
2339         /* The page dirty bit is cleared before writepage is called, which
2340          * means we have to tell create_empty_buffers to make dirty buffers
2341          * The page really should be up to date at this point, so tossing
2342          * in the BH_Uptodate is just a sanity check.
2343          */
2344         if (!page_has_buffers(page)) {
2345                 create_empty_buffers(page, s->s_blocksize,
2346                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2347         }
2348         head = page_buffers(page);
2349
2350         /* last page in the file, zero out any contents past the
2351          ** last byte in the file
2352          */
2353         if (page->index >= end_index) {
2354                 unsigned last_offset;
2355
2356                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2357                 /* no file contents in this page */
2358                 if (page->index >= end_index + 1 || !last_offset) {
2359                         unlock_page(page);
2360                         return 0;
2361                 }
2362                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2363         }
2364         bh = head;
2365         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2366         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2367         /* first map all the buffers, logging any direct items we find */
2368         do {
2369                 if (block > last_block) {
2370                         /*
2371                          * This can happen when the block size is less than
2372                          * the page size.  The corresponding bytes in the page
2373                          * were zero filled above
2374                          */
2375                         clear_buffer_dirty(bh);
2376                         set_buffer_uptodate(bh);
2377                 } else if ((checked || buffer_dirty(bh)) &&
2378                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2379                                                        && bh->b_blocknr ==
2380                                                        0))) {
2381                         /* not mapped yet, or it points to a direct item, search
2382                          * the btree for the mapping info, and log any direct
2383                          * items found
2384                          */
2385                         if ((error = map_block_for_writepage(inode, bh, block))) {
2386                                 goto fail;
2387                         }
2388                 }
2389                 bh = bh->b_this_page;
2390                 block++;
2391         } while (bh != head);
2392
2393         /*
2394          * we start the transaction after map_block_for_writepage,
2395          * because it can create holes in the file (an unbounded operation).
2396          * starting it here, we can make a reliable estimate for how many
2397          * blocks we're going to log
2398          */
2399         if (checked) {
2400                 ClearPageChecked(page);
2401                 reiserfs_write_lock(s);
2402                 error = journal_begin(&th, s, bh_per_page + 1);
2403                 if (error) {
2404                         reiserfs_write_unlock(s);
2405                         goto fail;
2406                 }
2407                 reiserfs_update_inode_transaction(inode);
2408         }
2409         /* now go through and lock any dirty buffers on the page */
2410         do {
2411                 get_bh(bh);
2412                 if (!buffer_mapped(bh))
2413                         continue;
2414                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2415                         continue;
2416
2417                 if (checked) {
2418                         reiserfs_prepare_for_journal(s, bh, 1);
2419                         journal_mark_dirty(&th, s, bh);
2420                         continue;
2421                 }
2422                 /* from this point on, we know the buffer is mapped to a
2423                  * real block and not a direct item
2424                  */
2425                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2426                         lock_buffer(bh);
2427                 } else {
2428                         if (!trylock_buffer(bh)) {
2429                                 redirty_page_for_writepage(wbc, page);
2430                                 continue;
2431                         }
2432                 }
2433                 if (test_clear_buffer_dirty(bh)) {
2434                         mark_buffer_async_write(bh);
2435                 } else {
2436                         unlock_buffer(bh);
2437                 }
2438         } while ((bh = bh->b_this_page) != head);
2439
2440         if (checked) {
2441                 error = journal_end(&th, s, bh_per_page + 1);
2442                 reiserfs_write_unlock(s);
2443                 if (error)
2444                         goto fail;
2445         }
2446         BUG_ON(PageWriteback(page));
2447         set_page_writeback(page);
2448         unlock_page(page);
2449
2450         /*
2451          * since any buffer might be the only dirty buffer on the page,
2452          * the first submit_bh can bring the page out of writeback.
2453          * be careful with the buffers.
2454          */
2455         do {
2456                 struct buffer_head *next = bh->b_this_page;
2457                 if (buffer_async_write(bh)) {
2458                         submit_bh(WRITE, bh);
2459                         nr++;
2460                 }
2461                 put_bh(bh);
2462                 bh = next;
2463         } while (bh != head);
2464
2465         error = 0;
2466       done:
2467         if (nr == 0) {
2468                 /*
2469                  * if this page only had a direct item, it is very possible for
2470                  * no io to be required without there being an error.  Or,
2471                  * someone else could have locked them and sent them down the
2472                  * pipe without locking the page
2473                  */
2474                 bh = head;
2475                 do {
2476                         if (!buffer_uptodate(bh)) {
2477                                 partial = 1;
2478                                 break;
2479                         }
2480                         bh = bh->b_this_page;
2481                 } while (bh != head);
2482                 if (!partial)
2483                         SetPageUptodate(page);
2484                 end_page_writeback(page);
2485         }
2486         return error;
2487
2488       fail:
2489         /* catches various errors, we need to make sure any valid dirty blocks
2490          * get to the media.  The page is currently locked and not marked for
2491          * writeback
2492          */
2493         ClearPageUptodate(page);
2494         bh = head;
2495         do {
2496                 get_bh(bh);
2497                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2498                         lock_buffer(bh);
2499                         mark_buffer_async_write(bh);
2500                 } else {
2501                         /*
2502                          * clear any dirty bits that might have come from getting
2503                          * attached to a dirty page
2504                          */
2505                         clear_buffer_dirty(bh);
2506                 }
2507                 bh = bh->b_this_page;
2508         } while (bh != head);
2509         SetPageError(page);
2510         BUG_ON(PageWriteback(page));
2511         set_page_writeback(page);
2512         unlock_page(page);
2513         do {
2514                 struct buffer_head *next = bh->b_this_page;
2515                 if (buffer_async_write(bh)) {
2516                         clear_buffer_dirty(bh);
2517                         submit_bh(WRITE, bh);
2518                         nr++;
2519                 }
2520                 put_bh(bh);
2521                 bh = next;
2522         } while (bh != head);
2523         goto done;
2524 }
2525
2526 static int reiserfs_readpage(struct file *f, struct page *page)
2527 {
2528         return block_read_full_page(page, reiserfs_get_block);
2529 }
2530
2531 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2532 {
2533         struct inode *inode = page->mapping->host;
2534         reiserfs_wait_on_write_block(inode->i_sb);
2535         return reiserfs_write_full_page(page, wbc);
2536 }
2537
2538 static int reiserfs_write_begin(struct file *file,
2539                                 struct address_space *mapping,
2540                                 loff_t pos, unsigned len, unsigned flags,
2541                                 struct page **pagep, void **fsdata)
2542 {
2543         struct inode *inode;
2544         struct page *page;
2545         pgoff_t index;
2546         int ret;
2547         int old_ref = 0;
2548
2549         inode = mapping->host;
2550         *fsdata = 0;
2551         if (flags & AOP_FLAG_CONT_EXPAND &&
2552             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2553                 pos ++;
2554                 *fsdata = (void *)(unsigned long)flags;
2555         }
2556
2557         index = pos >> PAGE_CACHE_SHIFT;
2558         page = grab_cache_page_write_begin(mapping, index, flags);
2559         if (!page)
2560                 return -ENOMEM;
2561         *pagep = page;
2562
2563         reiserfs_wait_on_write_block(inode->i_sb);
2564         fix_tail_page_for_writing(page);
2565         if (reiserfs_transaction_running(inode->i_sb)) {
2566                 struct reiserfs_transaction_handle *th;
2567                 th = (struct reiserfs_transaction_handle *)current->
2568                     journal_info;
2569                 BUG_ON(!th->t_refcount);
2570                 BUG_ON(!th->t_trans_id);
2571                 old_ref = th->t_refcount;
2572                 th->t_refcount++;
2573         }
2574         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2575                                 reiserfs_get_block);
2576         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2577                 struct reiserfs_transaction_handle *th = current->journal_info;
2578                 /* this gets a little ugly.  If reiserfs_get_block returned an
2579                  * error and left a transacstion running, we've got to close it,
2580                  * and we've got to free handle if it was a persistent transaction.
2581                  *
2582                  * But, if we had nested into an existing transaction, we need
2583                  * to just drop the ref count on the handle.
2584                  *
2585                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2586                  * and it was a persistent trans.  Otherwise, it was nested above.
2587                  */
2588                 if (th->t_refcount > old_ref) {
2589                         if (old_ref)
2590                                 th->t_refcount--;
2591                         else {
2592                                 int err;
2593                                 reiserfs_write_lock(inode->i_sb);
2594                                 err = reiserfs_end_persistent_transaction(th);
2595                                 reiserfs_write_unlock(inode->i_sb);
2596                                 if (err)
2597                                         ret = err;
2598                         }
2599                 }
2600         }
2601         if (ret) {
2602                 unlock_page(page);
2603                 page_cache_release(page);
2604         }
2605         return ret;
2606 }
2607
2608 int reiserfs_prepare_write(struct file *f, struct page *page,
2609                            unsigned from, unsigned to)
2610 {
2611         struct inode *inode = page->mapping->host;
2612         int ret;
2613         int old_ref = 0;
2614
2615         reiserfs_wait_on_write_block(inode->i_sb);
2616         fix_tail_page_for_writing(page);
2617         if (reiserfs_transaction_running(inode->i_sb)) {
2618                 struct reiserfs_transaction_handle *th;
2619                 th = (struct reiserfs_transaction_handle *)current->
2620                     journal_info;
2621                 BUG_ON(!th->t_refcount);
2622                 BUG_ON(!th->t_trans_id);
2623                 old_ref = th->t_refcount;
2624                 th->t_refcount++;
2625         }
2626
2627         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2628         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2629                 struct reiserfs_transaction_handle *th = current->journal_info;
2630                 /* this gets a little ugly.  If reiserfs_get_block returned an
2631                  * error and left a transacstion running, we've got to close it,
2632                  * and we've got to free handle if it was a persistent transaction.
2633                  *
2634                  * But, if we had nested into an existing transaction, we need
2635                  * to just drop the ref count on the handle.
2636                  *
2637                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2638                  * and it was a persistent trans.  Otherwise, it was nested above.
2639                  */
2640                 if (th->t_refcount > old_ref) {
2641                         if (old_ref)
2642                                 th->t_refcount--;
2643                         else {
2644                                 int err;
2645                                 reiserfs_write_lock(inode->i_sb);
2646                                 err = reiserfs_end_persistent_transaction(th);
2647                                 reiserfs_write_unlock(inode->i_sb);
2648                                 if (err)
2649                                         ret = err;
2650                         }
2651                 }
2652         }
2653         return ret;
2654
2655 }
2656
2657 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2658 {
2659         return generic_block_bmap(as, block, reiserfs_bmap);
2660 }
2661
2662 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2663                               loff_t pos, unsigned len, unsigned copied,
2664                               struct page *page, void *fsdata)
2665 {
2666         struct inode *inode = page->mapping->host;
2667         int ret = 0;
2668         int update_sd = 0;
2669         struct reiserfs_transaction_handle *th;
2670         unsigned start;
2671
2672         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2673                 pos ++;
2674
2675         reiserfs_wait_on_write_block(inode->i_sb);
2676         if (reiserfs_transaction_running(inode->i_sb))
2677                 th = current->journal_info;
2678         else
2679                 th = NULL;
2680
2681         start = pos & (PAGE_CACHE_SIZE - 1);
2682         if (unlikely(copied < len)) {
2683                 if (!PageUptodate(page))
2684                         copied = 0;
2685
2686                 page_zero_new_buffers(page, start + copied, start + len);
2687         }
2688         flush_dcache_page(page);
2689
2690         reiserfs_commit_page(inode, page, start, start + copied);
2691
2692         /* generic_commit_write does this for us, but does not update the
2693          ** transaction tracking stuff when the size changes.  So, we have
2694          ** to do the i_size updates here.
2695          */
2696         pos += copied;
2697         if (pos > inode->i_size) {
2698                 struct reiserfs_transaction_handle myth;
2699                 reiserfs_write_lock(inode->i_sb);
2700                 /* If the file have grown beyond the border where it
2701                    can have a tail, unmark it as needing a tail
2702                    packing */
2703                 if ((have_large_tails(inode->i_sb)
2704                      && inode->i_size > i_block_size(inode) * 4)
2705                     || (have_small_tails(inode->i_sb)
2706                         && inode->i_size > i_block_size(inode)))
2707                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2708
2709                 ret = journal_begin(&myth, inode->i_sb, 1);
2710                 if (ret) {
2711                         reiserfs_write_unlock(inode->i_sb);
2712                         goto journal_error;
2713                 }
2714                 reiserfs_update_inode_transaction(inode);
2715                 inode->i_size = pos;
2716                 /*
2717                  * this will just nest into our transaction.  It's important
2718                  * to use mark_inode_dirty so the inode gets pushed around on the
2719                  * dirty lists, and so that O_SYNC works as expected
2720                  */
2721                 mark_inode_dirty(inode);
2722                 reiserfs_update_sd(&myth, inode);
2723                 update_sd = 1;
2724                 ret = journal_end(&myth, inode->i_sb, 1);
2725                 reiserfs_write_unlock(inode->i_sb);
2726                 if (ret)
2727                         goto journal_error;
2728         }
2729         if (th) {
2730                 reiserfs_write_lock(inode->i_sb);
2731                 if (!update_sd)
2732                         mark_inode_dirty(inode);
2733                 ret = reiserfs_end_persistent_transaction(th);
2734                 reiserfs_write_unlock(inode->i_sb);
2735                 if (ret)
2736                         goto out;
2737         }
2738
2739       out:
2740         unlock_page(page);
2741         page_cache_release(page);
2742         return ret == 0 ? copied : ret;
2743
2744       journal_error:
2745         if (th) {
2746                 reiserfs_write_lock(inode->i_sb);
2747                 if (!update_sd)
2748                         reiserfs_update_sd(th, inode);
2749                 ret = reiserfs_end_persistent_transaction(th);
2750                 reiserfs_write_unlock(inode->i_sb);
2751         }
2752
2753         goto out;
2754 }
2755
2756 int reiserfs_commit_write(struct file *f, struct page *page,
2757                           unsigned from, unsigned to)
2758 {
2759         struct inode *inode = page->mapping->host;
2760         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2761         int ret = 0;
2762         int update_sd = 0;
2763         struct reiserfs_transaction_handle *th = NULL;
2764
2765         reiserfs_wait_on_write_block(inode->i_sb);
2766         if (reiserfs_transaction_running(inode->i_sb)) {
2767                 th = current->journal_info;
2768         }
2769         reiserfs_commit_page(inode, page, from, to);
2770
2771         /* generic_commit_write does this for us, but does not update the
2772          ** transaction tracking stuff when the size changes.  So, we have
2773          ** to do the i_size updates here.
2774          */
2775         if (pos > inode->i_size) {
2776                 struct reiserfs_transaction_handle myth;
2777                 reiserfs_write_lock(inode->i_sb);
2778                 /* If the file have grown beyond the border where it
2779                    can have a tail, unmark it as needing a tail
2780                    packing */
2781                 if ((have_large_tails(inode->i_sb)
2782                      && inode->i_size > i_block_size(inode) * 4)
2783                     || (have_small_tails(inode->i_sb)
2784                         && inode->i_size > i_block_size(inode)))
2785                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2786
2787                 ret = journal_begin(&myth, inode->i_sb, 1);
2788                 if (ret) {
2789                         reiserfs_write_unlock(inode->i_sb);
2790                         goto journal_error;
2791                 }
2792                 reiserfs_update_inode_transaction(inode);
2793                 inode->i_size = pos;
2794                 /*
2795                  * this will just nest into our transaction.  It's important
2796                  * to use mark_inode_dirty so the inode gets pushed around on the
2797                  * dirty lists, and so that O_SYNC works as expected
2798                  */
2799                 mark_inode_dirty(inode);
2800                 reiserfs_update_sd(&myth, inode);
2801                 update_sd = 1;
2802                 ret = journal_end(&myth, inode->i_sb, 1);
2803                 reiserfs_write_unlock(inode->i_sb);
2804                 if (ret)
2805                         goto journal_error;
2806         }
2807         if (th) {
2808                 reiserfs_write_lock(inode->i_sb);
2809                 if (!update_sd)
2810                         mark_inode_dirty(inode);
2811                 ret = reiserfs_end_persistent_transaction(th);
2812                 reiserfs_write_unlock(inode->i_sb);
2813                 if (ret)
2814                         goto out;
2815         }
2816
2817       out:
2818         return ret;
2819
2820       journal_error:
2821         if (th) {
2822                 reiserfs_write_lock(inode->i_sb);
2823                 if (!update_sd)
2824                         reiserfs_update_sd(th, inode);
2825                 ret = reiserfs_end_persistent_transaction(th);
2826                 reiserfs_write_unlock(inode->i_sb);
2827         }
2828
2829         return ret;
2830 }
2831
2832 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2833 {
2834         if (reiserfs_attrs(inode->i_sb)) {
2835                 if (sd_attrs & REISERFS_SYNC_FL)
2836                         inode->i_flags |= S_SYNC;
2837                 else
2838                         inode->i_flags &= ~S_SYNC;
2839                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2840                         inode->i_flags |= S_IMMUTABLE;
2841                 else
2842                         inode->i_flags &= ~S_IMMUTABLE;
2843                 if (sd_attrs & REISERFS_APPEND_FL)
2844                         inode->i_flags |= S_APPEND;
2845                 else
2846                         inode->i_flags &= ~S_APPEND;
2847                 if (sd_attrs & REISERFS_NOATIME_FL)
2848                         inode->i_flags |= S_NOATIME;
2849                 else
2850                         inode->i_flags &= ~S_NOATIME;
2851                 if (sd_attrs & REISERFS_NOTAIL_FL)
2852                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2853                 else
2854                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2855         }
2856 }
2857
2858 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2859 {
2860         if (reiserfs_attrs(inode->i_sb)) {
2861                 if (inode->i_flags & S_IMMUTABLE)
2862                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2863                 else
2864                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2865                 if (inode->i_flags & S_SYNC)
2866                         *sd_attrs |= REISERFS_SYNC_FL;
2867                 else
2868                         *sd_attrs &= ~REISERFS_SYNC_FL;
2869                 if (inode->i_flags & S_NOATIME)
2870                         *sd_attrs |= REISERFS_NOATIME_FL;
2871                 else
2872                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2873                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2874                         *sd_attrs |= REISERFS_NOTAIL_FL;
2875                 else
2876                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2877         }
2878 }
2879
2880 /* decide if this buffer needs to stay around for data logging or ordered
2881 ** write purposes
2882 */
2883 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2884 {
2885         int ret = 1;
2886         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2887
2888         lock_buffer(bh);
2889         spin_lock(&j->j_dirty_buffers_lock);
2890         if (!buffer_mapped(bh)) {
2891                 goto free_jh;
2892         }
2893         /* the page is locked, and the only places that log a data buffer
2894          * also lock the page.
2895          */
2896         if (reiserfs_file_data_log(inode)) {
2897                 /*
2898                  * very conservative, leave the buffer pinned if
2899                  * anyone might need it.
2900                  */
2901                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2902                         ret = 0;
2903                 }
2904         } else  if (buffer_dirty(bh)) {
2905                 struct reiserfs_journal_list *jl;
2906                 struct reiserfs_jh *jh = bh->b_private;
2907
2908                 /* why is this safe?
2909                  * reiserfs_setattr updates i_size in the on disk
2910                  * stat data before allowing vmtruncate to be called.
2911                  *
2912                  * If buffer was put onto the ordered list for this
2913                  * transaction, we know for sure either this transaction
2914                  * or an older one already has updated i_size on disk,
2915                  * and this ordered data won't be referenced in the file
2916                  * if we crash.
2917                  *
2918                  * if the buffer was put onto the ordered list for an older
2919                  * transaction, we need to leave it around
2920                  */
2921                 if (jh && (jl = jh->jl)
2922                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2923                         ret = 0;
2924         }
2925       free_jh:
2926         if (ret && bh->b_private) {
2927                 reiserfs_free_jh(bh);
2928         }
2929         spin_unlock(&j->j_dirty_buffers_lock);
2930         unlock_buffer(bh);
2931         return ret;
2932 }
2933
2934 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2935 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2936 {
2937         struct buffer_head *head, *bh, *next;
2938         struct inode *inode = page->mapping->host;
2939         unsigned int curr_off = 0;
2940         int ret = 1;
2941
2942         BUG_ON(!PageLocked(page));
2943
2944         if (offset == 0)
2945                 ClearPageChecked(page);
2946
2947         if (!page_has_buffers(page))
2948                 goto out;
2949
2950         head = page_buffers(page);
2951         bh = head;
2952         do {
2953                 unsigned int next_off = curr_off + bh->b_size;
2954                 next = bh->b_this_page;
2955
2956                 /*
2957                  * is this block fully invalidated?
2958                  */
2959                 if (offset <= curr_off) {
2960                         if (invalidatepage_can_drop(inode, bh))
2961                                 reiserfs_unmap_buffer(bh);
2962                         else
2963                                 ret = 0;
2964                 }
2965                 curr_off = next_off;
2966                 bh = next;
2967         } while (bh != head);
2968
2969         /*
2970          * We release buffers only if the entire page is being invalidated.
2971          * The get_block cached value has been unconditionally invalidated,
2972          * so real IO is not possible anymore.
2973          */
2974         if (!offset && ret) {
2975                 ret = try_to_release_page(page, 0);
2976                 /* maybe should BUG_ON(!ret); - neilb */
2977         }
2978       out:
2979         return;
2980 }
2981
2982 static int reiserfs_set_page_dirty(struct page *page)
2983 {
2984         struct inode *inode = page->mapping->host;
2985         if (reiserfs_file_data_log(inode)) {
2986                 SetPageChecked(page);
2987                 return __set_page_dirty_nobuffers(page);
2988         }
2989         return __set_page_dirty_buffers(page);
2990 }
2991
2992 /*
2993  * Returns 1 if the page's buffers were dropped.  The page is locked.
2994  *
2995  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2996  * in the buffers at page_buffers(page).
2997  *
2998  * even in -o notail mode, we can't be sure an old mount without -o notail
2999  * didn't create files with tails.
3000  */
3001 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3002 {
3003         struct inode *inode = page->mapping->host;
3004         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3005         struct buffer_head *head;
3006         struct buffer_head *bh;
3007         int ret = 1;
3008
3009         WARN_ON(PageChecked(page));
3010         spin_lock(&j->j_dirty_buffers_lock);
3011         head = page_buffers(page);
3012         bh = head;
3013         do {
3014                 if (bh->b_private) {
3015                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3016                                 reiserfs_free_jh(bh);
3017                         } else {
3018                                 ret = 0;
3019                                 break;
3020                         }
3021                 }
3022                 bh = bh->b_this_page;
3023         } while (bh != head);
3024         if (ret)
3025                 ret = try_to_free_buffers(page);
3026         spin_unlock(&j->j_dirty_buffers_lock);
3027         return ret;
3028 }
3029
3030 /* We thank Mingming Cao for helping us understand in great detail what
3031    to do in this section of the code. */
3032 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3033                                   const struct iovec *iov, loff_t offset,
3034                                   unsigned long nr_segs)
3035 {
3036         struct file *file = iocb->ki_filp;
3037         struct inode *inode = file->f_mapping->host;
3038
3039         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3040                                   offset, nr_segs,
3041                                   reiserfs_get_blocks_direct_io, NULL);
3042 }
3043
3044 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3045 {
3046         struct inode *inode = dentry->d_inode;
3047         int error;
3048         unsigned int ia_valid;
3049
3050         /* must be turned off for recursive notify_change calls */
3051         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3052
3053         reiserfs_write_lock(inode->i_sb);
3054         if (attr->ia_valid & ATTR_SIZE) {
3055                 /* version 2 items will be caught by the s_maxbytes check
3056                  ** done for us in vmtruncate
3057                  */
3058                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3059                     attr->ia_size > MAX_NON_LFS) {
3060                         error = -EFBIG;
3061                         goto out;
3062                 }
3063                 /* fill in hole pointers in the expanding truncate case. */
3064                 if (attr->ia_size > inode->i_size) {
3065                         error = generic_cont_expand_simple(inode, attr->ia_size);
3066                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3067                                 int err;
3068                                 struct reiserfs_transaction_handle th;
3069                                 /* we're changing at most 2 bitmaps, inode + super */
3070                                 err = journal_begin(&th, inode->i_sb, 4);
3071                                 if (!err) {
3072                                         reiserfs_discard_prealloc(&th, inode);
3073                                         err = journal_end(&th, inode->i_sb, 4);
3074                                 }
3075                                 if (err)
3076                                         error = err;
3077                         }
3078                         if (error)
3079                                 goto out;
3080                         /*
3081                          * file size is changed, ctime and mtime are
3082                          * to be updated
3083                          */
3084                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3085                 }
3086         }
3087
3088         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3089              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3090             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3091                 /* stat data of format v3.5 has 16 bit uid and gid */
3092                 error = -EINVAL;
3093                 goto out;
3094         }
3095
3096         error = inode_change_ok(inode, attr);
3097         if (!error) {
3098                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3099                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3100                         error = reiserfs_chown_xattrs(inode, attr);
3101
3102                         if (!error) {
3103                                 struct reiserfs_transaction_handle th;
3104                                 int jbegin_count =
3105                                     2 *
3106                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3107                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3108                                     2;
3109
3110                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3111                                 error =
3112                                     journal_begin(&th, inode->i_sb,
3113                                                   jbegin_count);
3114                                 if (error)
3115                                         goto out;
3116                                 error =
3117                                     vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
3118                                 if (error) {
3119                                         journal_end(&th, inode->i_sb,
3120                                                     jbegin_count);
3121                                         goto out;
3122                                 }
3123                                 /* Update corresponding info in inode so that everything is in
3124                                  * one transaction */
3125                                 if (attr->ia_valid & ATTR_UID)
3126                                         inode->i_uid = attr->ia_uid;
3127                                 if (attr->ia_valid & ATTR_GID)
3128                                         inode->i_gid = attr->ia_gid;
3129                                 mark_inode_dirty(inode);
3130                                 error =
3131                                     journal_end(&th, inode->i_sb, jbegin_count);
3132                         }
3133                 }
3134                 if (!error)
3135                         error = inode_setattr(inode, attr);
3136         }
3137
3138         if (!error && reiserfs_posixacl(inode->i_sb)) {
3139                 if (attr->ia_valid & ATTR_MODE)
3140                         error = reiserfs_acl_chmod(inode);
3141         }
3142
3143       out:
3144         reiserfs_write_unlock(inode->i_sb);
3145         return error;
3146 }
3147
3148 const struct address_space_operations reiserfs_address_space_operations = {
3149         .writepage = reiserfs_writepage,
3150         .readpage = reiserfs_readpage,
3151         .readpages = reiserfs_readpages,
3152         .releasepage = reiserfs_releasepage,
3153         .invalidatepage = reiserfs_invalidatepage,
3154         .sync_page = block_sync_page,
3155         .write_begin = reiserfs_write_begin,
3156         .write_end = reiserfs_write_end,
3157         .bmap = reiserfs_aop_bmap,
3158         .direct_IO = reiserfs_direct_IO,
3159         .set_page_dirty = reiserfs_set_page_dirty,
3160 };