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                         DQUOT_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_warning(inode->i_sb,
 845                                                                  "clm-6004: convert tail failed inode %lu, error %d",
 846                                                                  inode->i_ino,
 847                                                                  retval);
 848                                         if (allocated_block_nr) {
 849                                                 /* the bitmap, the super, and the stat data == 3 */
 850                                                 if (!th)
 851                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 852                                                 if (th)
 853                                                         reiserfs_free_block(th,
 854                                                                             inode,
 855                                                                             allocated_block_nr,
 856                                                                             1);
 857                                         }
 858                                         goto failure;
 859                                 }
 860                                 goto research;
 861                         }
 862                         retval =
 863                             direct2indirect(th, inode, &path, unbh,
 864                                             tail_offset);
 865                         if (retval) {
 866                                 reiserfs_unmap_buffer(unbh);
 867                                 reiserfs_free_block(th, inode,
 868                                                     allocated_block_nr, 1);
 869                                 goto failure;
 870                         }
 871                         /* it is important the set_buffer_uptodate is done after
 872                          ** the direct2indirect.  The buffer might contain valid
 873                          ** data newer than the data on disk (read by readpage, changed,
 874                          ** and then sent here by writepage).  direct2indirect needs
 875                          ** to know if unbh was already up to date, so it can decide
 876                          ** if the data in unbh needs to be replaced with data from
 877                          ** the disk
 878                          */
 879                         set_buffer_uptodate(unbh);
 880
 881                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 882                            buffer will disappear shortly, so it should not be added to
 883                          */
 884                         if (unbh->b_page) {
 885                                 /* we've converted the tail, so we must
 886                                  ** flush unbh before the transaction commits
 887                                  */
 888                                 reiserfs_add_tail_list(inode, unbh);
 889
 890                                 /* mark it dirty now to prevent commit_write from adding
 891                                  ** this buffer to the inode's dirty buffer list
 892                                  */
 893                                 /*
 894                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 895                                  * It's still atomic, but it sets the page dirty too,
 896                                  * which makes it eligible for writeback at any time by the
 897                                  * VM (which was also the case with __mark_buffer_dirty())
 898                                  */
 899                                 mark_buffer_dirty(unbh);
 900                         }
 901                 } else {
 902                         /* append indirect item with holes if needed, when appending
 903                            pointer to 'block'-th block use block, which is already
 904                            allocated */
 905                         struct cpu_key tmp_key;
 906                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 907                         // one block which is a fastpath
 908                         unp_t *un;
 909                         __u64 max_to_insert =
 910                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 911                             UNFM_P_SIZE;
 912                         __u64 blocks_needed;
 913
 914                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 915                                "vs-804: invalid position for append");
 916                         /* indirect item has to be appended, set up key of that position */
 917                         make_cpu_key(&tmp_key, inode,
 918                                      le_key_k_offset(version,
 919                                                      &(ih->ih_key)) +
 920                                      op_bytes_number(ih,
 921                                                      inode->i_sb->s_blocksize),
 922                                      //pos_in_item * inode->i_sb->s_blocksize,
 923                                      TYPE_INDIRECT, 3); // key type is unimportant
 924
 925                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 926                                "green-805: invalid offset");
 927                         blocks_needed =
 928                             1 +
 929                             ((cpu_key_k_offset(&key) -
 930                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 931                              s_blocksize_bits);
 932
 933                         if (blocks_needed == 1) {
 934                                 un = &unf_single;
 935                         } else {
 936                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 937                                 if (!un) {
 938                                         un = &unf_single;
 939                                         blocks_needed = 1;
 940                                         max_to_insert = 0;
 941                                 }
 942                         }
 943                         if (blocks_needed <= max_to_insert) {
 944                                 /* we are going to add target block to the file. Use allocated
 945                                    block for that */
 946                                 un[blocks_needed - 1] =
 947                                     cpu_to_le32(allocated_block_nr);
 948                                 set_block_dev_mapped(bh_result,
 949                                                      allocated_block_nr, inode);
 950                                 set_buffer_new(bh_result);
 951                                 done = 1;
 952                         } else {
 953                                 /* paste hole to the indirect item */
 954                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 955                                    only have space for one block */
 956                                 blocks_needed =
 957                                     max_to_insert ? max_to_insert : 1;
 958                         }
 959                         retval =
 960                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 961                                                      (char *)un,
 962                                                      UNFM_P_SIZE *
 963                                                      blocks_needed);
 964
 965                         if (blocks_needed != 1)
 966                                 kfree(un);
 967
 968                         if (retval) {
 969                                 reiserfs_free_block(th, inode,
 970                                                     allocated_block_nr, 1);
 971                                 goto failure;
 972                         }
 973                         if (!done) {
 974                                 /* We need to mark new file size in case this function will be
 975                                    interrupted/aborted later on. And we may do this only for
 976                                    holes. */
 977                                 inode->i_size +=
 978                                     inode->i_sb->s_blocksize * blocks_needed;
 979                         }
 980                 }
 981
 982                 if (done == 1)
 983                         break;
 984
 985                 /* this loop could log more blocks than we had originally asked
 986                  ** for.  So, we have to allow the transaction to end if it is
 987                  ** too big or too full.  Update the inode so things are
 988                  ** consistent if we crash before the function returns
 989                  **
 990                  ** release the path so that anybody waiting on the path before
 991                  ** ending their transaction will be able to continue.
 992                  */
 993                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 994                         retval = restart_transaction(th, inode, &path);
 995                         if (retval)
 996                                 goto failure;
 997                 }
 998                 /* inserting indirect pointers for a hole can take a
 999                  ** long time.  reschedule if needed
1000                  */
1001                 cond_resched();
1002
1003                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1004                 if (retval == IO_ERROR) {
1005                         retval = -EIO;
1006                         goto failure;
1007                 }
1008                 if (retval == POSITION_FOUND) {
1009                         reiserfs_warning(inode->i_sb,
1010                                          "vs-825: reiserfs_get_block: "
1011                                          "%K should not be found", &key);
1012                         retval = -EEXIST;
1013                         if (allocated_block_nr)
1014                                 reiserfs_free_block(th, inode,
1015                                                     allocated_block_nr, 1);
1016                         pathrelse(&path);
1017                         goto failure;
1018                 }
1019                 bh = get_last_bh(&path);
1020                 ih = get_ih(&path);
1021                 item = get_item(&path);
1022                 pos_in_item = path.pos_in_item;
1023         } while (1);
1024
1025         retval = 0;
1026
1027       failure:
1028         if (th && (!dangle || (retval && !th->t_trans_id))) {
1029                 int err;
1030                 if (th->t_trans_id)
1031                         reiserfs_update_sd(th, inode);
1032                 err = reiserfs_end_persistent_transaction(th);
1033                 if (err)
1034                         retval = err;
1035         }
1036
1037         reiserfs_write_unlock(inode->i_sb);
1038         reiserfs_check_path(&path);
1039         return retval;
1040 }
1041
1042 static int
1043 reiserfs_readpages(struct file *file, struct address_space *mapping,
1044                    struct list_head *pages, unsigned nr_pages)
1045 {
1046         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1047 }
1048
1049 /* Compute real number of used bytes by file
1050  * Following three functions can go away when we'll have enough space in stat item
1051  */
1052 static int real_space_diff(struct inode *inode, int sd_size)
1053 {
1054         int bytes;
1055         loff_t blocksize = inode->i_sb->s_blocksize;
1056
1057         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1058                 return sd_size;
1059
1060         /* End of file is also in full block with indirect reference, so round
1061          ** up to the next block.
1062          **
1063          ** there is just no way to know if the tail is actually packed
1064          ** on the file, so we have to assume it isn't.  When we pack the
1065          ** tail, we add 4 bytes to pretend there really is an unformatted
1066          ** node pointer
1067          */
1068         bytes =
1069             ((inode->i_size +
1070               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1071             sd_size;
1072         return bytes;
1073 }
1074
1075 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1076                                         int sd_size)
1077 {
1078         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1079                 return inode->i_size +
1080                     (loff_t) (real_space_diff(inode, sd_size));
1081         }
1082         return ((loff_t) real_space_diff(inode, sd_size)) +
1083             (((loff_t) blocks) << 9);
1084 }
1085
1086 /* Compute number of blocks used by file in ReiserFS counting */
1087 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1088 {
1089         loff_t bytes = inode_get_bytes(inode);
1090         loff_t real_space = real_space_diff(inode, sd_size);
1091
1092         /* keeps fsck and non-quota versions of reiserfs happy */
1093         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1094                 bytes += (loff_t) 511;
1095         }
1096
1097         /* files from before the quota patch might i_blocks such that
1098          ** bytes < real_space.  Deal with that here to prevent it from
1099          ** going negative.
1100          */
1101         if (bytes < real_space)
1102                 return 0;
1103         return (bytes - real_space) >> 9;
1104 }
1105
1106 //
1107 // BAD: new directories have stat data of new type and all other items
1108 // of old type. Version stored in the inode says about body items, so
1109 // in update_stat_data we can not rely on inode, but have to check
1110 // item version directly
1111 //
1112
1113 // called by read_locked_inode
1114 static void init_inode(struct inode *inode, struct treepath *path)
1115 {
1116         struct buffer_head *bh;
1117         struct item_head *ih;
1118         __u32 rdev;
1119         //int version = ITEM_VERSION_1;
1120
1121         bh = PATH_PLAST_BUFFER(path);
1122         ih = PATH_PITEM_HEAD(path);
1123
1124         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1125
1126         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1127         REISERFS_I(inode)->i_flags = 0;
1128         REISERFS_I(inode)->i_prealloc_block = 0;
1129         REISERFS_I(inode)->i_prealloc_count = 0;
1130         REISERFS_I(inode)->i_trans_id = 0;
1131         REISERFS_I(inode)->i_jl = NULL;
1132         mutex_init(&(REISERFS_I(inode)->i_mmap));
1133         reiserfs_init_acl_access(inode);
1134         reiserfs_init_acl_default(inode);
1135         reiserfs_init_xattr_rwsem(inode);
1136
1137         if (stat_data_v1(ih)) {
1138                 struct stat_data_v1 *sd =
1139                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1140                 unsigned long blocks;
1141
1142                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1143                 set_inode_sd_version(inode, STAT_DATA_V1);
1144                 inode->i_mode = sd_v1_mode(sd);
1145                 inode->i_nlink = sd_v1_nlink(sd);
1146                 inode->i_uid = sd_v1_uid(sd);
1147                 inode->i_gid = sd_v1_gid(sd);
1148                 inode->i_size = sd_v1_size(sd);
1149                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1150                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1151                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1152                 inode->i_atime.tv_nsec = 0;
1153                 inode->i_ctime.tv_nsec = 0;
1154                 inode->i_mtime.tv_nsec = 0;
1155
1156                 inode->i_blocks = sd_v1_blocks(sd);
1157                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1158                 blocks = (inode->i_size + 511) >> 9;
1159                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1160                 if (inode->i_blocks > blocks) {
1161                         // there was a bug in <=3.5.23 when i_blocks could take negative
1162                         // values. Starting from 3.5.17 this value could even be stored in
1163                         // stat data. For such files we set i_blocks based on file
1164                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1165                         // only updated if file's inode will ever change
1166                         inode->i_blocks = blocks;
1167                 }
1168
1169                 rdev = sd_v1_rdev(sd);
1170                 REISERFS_I(inode)->i_first_direct_byte =
1171                     sd_v1_first_direct_byte(sd);
1172                 /* an early bug in the quota code can give us an odd number for the
1173                  ** block count.  This is incorrect, fix it here.
1174                  */
1175                 if (inode->i_blocks & 1) {
1176                         inode->i_blocks++;
1177                 }
1178                 inode_set_bytes(inode,
1179                                 to_real_used_space(inode, inode->i_blocks,
1180                                                    SD_V1_SIZE));
1181                 /* nopack is initially zero for v1 objects. For v2 objects,
1182                    nopack is initialised from sd_attrs */
1183                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1184         } else {
1185                 // new stat data found, but object may have old items
1186                 // (directories and symlinks)
1187                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1188
1189                 inode->i_mode = sd_v2_mode(sd);
1190                 inode->i_nlink = sd_v2_nlink(sd);
1191                 inode->i_uid = sd_v2_uid(sd);
1192                 inode->i_size = sd_v2_size(sd);
1193                 inode->i_gid = sd_v2_gid(sd);
1194                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1195                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1196                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1197                 inode->i_ctime.tv_nsec = 0;
1198                 inode->i_mtime.tv_nsec = 0;
1199                 inode->i_atime.tv_nsec = 0;
1200                 inode->i_blocks = sd_v2_blocks(sd);
1201                 rdev = sd_v2_rdev(sd);
1202                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1203                         inode->i_generation =
1204                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1205                 else
1206                         inode->i_generation = sd_v2_generation(sd);
1207
1208                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1209                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1210                 else
1211                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1212                 REISERFS_I(inode)->i_first_direct_byte = 0;
1213                 set_inode_sd_version(inode, STAT_DATA_V2);
1214                 inode_set_bytes(inode,
1215                                 to_real_used_space(inode, inode->i_blocks,
1216                                                    SD_V2_SIZE));
1217                 /* read persistent inode attributes from sd and initalise
1218                    generic inode flags from them */
1219                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1220                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1221         }
1222
1223         pathrelse(path);
1224         if (S_ISREG(inode->i_mode)) {
1225                 inode->i_op = &reiserfs_file_inode_operations;
1226                 inode->i_fop = &reiserfs_file_operations;
1227                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1228         } else if (S_ISDIR(inode->i_mode)) {
1229                 inode->i_op = &reiserfs_dir_inode_operations;
1230                 inode->i_fop = &reiserfs_dir_operations;
1231         } else if (S_ISLNK(inode->i_mode)) {
1232                 inode->i_op = &reiserfs_symlink_inode_operations;
1233                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1234         } else {
1235                 inode->i_blocks = 0;
1236                 inode->i_op = &reiserfs_special_inode_operations;
1237                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1238         }
1239 }
1240
1241 // update new stat data with inode fields
1242 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1243 {
1244         struct stat_data *sd_v2 = (struct stat_data *)sd;
1245         __u16 flags;
1246
1247         set_sd_v2_mode(sd_v2, inode->i_mode);
1248         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1249         set_sd_v2_uid(sd_v2, inode->i_uid);
1250         set_sd_v2_size(sd_v2, size);
1251         set_sd_v2_gid(sd_v2, inode->i_gid);
1252         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1253         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1254         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1255         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1256         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1257                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1258         else
1259                 set_sd_v2_generation(sd_v2, inode->i_generation);
1260         flags = REISERFS_I(inode)->i_attrs;
1261         i_attrs_to_sd_attrs(inode, &flags);
1262         set_sd_v2_attrs(sd_v2, flags);
1263 }
1264
1265 // used to copy inode's fields to old stat data
1266 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1267 {
1268         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1269
1270         set_sd_v1_mode(sd_v1, inode->i_mode);
1271         set_sd_v1_uid(sd_v1, inode->i_uid);
1272         set_sd_v1_gid(sd_v1, inode->i_gid);
1273         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1274         set_sd_v1_size(sd_v1, size);
1275         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1276         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1277         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1278
1279         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1280                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1281         else
1282                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1283
1284         // Sigh. i_first_direct_byte is back
1285         set_sd_v1_first_direct_byte(sd_v1,
1286                                     REISERFS_I(inode)->i_first_direct_byte);
1287 }
1288
1289 /* NOTE, you must prepare the buffer head before sending it here,
1290 ** and then log it after the call
1291 */
1292 static void update_stat_data(struct treepath *path, struct inode *inode,
1293                              loff_t size)
1294 {
1295         struct buffer_head *bh;
1296         struct item_head *ih;
1297
1298         bh = PATH_PLAST_BUFFER(path);
1299         ih = PATH_PITEM_HEAD(path);
1300
1301         if (!is_statdata_le_ih(ih))
1302                 reiserfs_panic(inode->i_sb,
1303                                "vs-13065: update_stat_data: 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_warning(inode->i_sb,
1336                                          "vs-13050: reiserfs_update_sd: "
1337                                          "i/o failure occurred trying to update %K stat data",
1338                                          &key);
1339                         return;
1340                 }
1341                 if (retval == ITEM_NOT_FOUND) {
1342                         pos = PATH_LAST_POSITION(&path);
1343                         pathrelse(&path);
1344                         if (inode->i_nlink == 0) {
1345                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1346                                 return;
1347                         }
1348                         reiserfs_warning(inode->i_sb,
1349                                          "vs-13060: reiserfs_update_sd: "
1350                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1351                                          INODE_PKEY(inode), inode->i_nlink,
1352                                          pos);
1353                         reiserfs_check_path(&path);
1354                         return;
1355                 }
1356
1357                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1358                  ** FS might change.  We have to detect that, and loop back to the
1359                  ** search if the stat data item has moved
1360                  */
1361                 bh = get_last_bh(&path);
1362                 ih = get_ih(&path);
1363                 copy_item_head(&tmp_ih, ih);
1364                 fs_gen = get_generation(inode->i_sb);
1365                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1366                 if (fs_changed(fs_gen, inode->i_sb)
1367                     && item_moved(&tmp_ih, &path)) {
1368                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1369                         continue;       /* Stat_data item has been moved after scheduling. */
1370                 }
1371                 break;
1372         }
1373         update_stat_data(&path, inode, size);
1374         journal_mark_dirty(th, th->t_super, bh);
1375         pathrelse(&path);
1376         return;
1377 }
1378
1379 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1380 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1381 ** and clear the key in the private portion of the inode, otherwise a
1382 ** corresponding iput might try to delete whatever object the inode last
1383 ** represented.
1384 */
1385 static void reiserfs_make_bad_inode(struct inode *inode)
1386 {
1387         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1388         make_bad_inode(inode);
1389 }
1390
1391 //
1392 // initially this function was derived from minix or ext2's analog and
1393 // evolved as the prototype did
1394 //
1395
1396 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1397 {
1398         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1399         inode->i_ino = args->objectid;
1400         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1401         return 0;
1402 }
1403
1404 /* looks for stat data in the tree, and fills up the fields of in-core
1405    inode stat data fields */
1406 void reiserfs_read_locked_inode(struct inode *inode,
1407                                 struct reiserfs_iget_args *args)
1408 {
1409         INITIALIZE_PATH(path_to_sd);
1410         struct cpu_key key;
1411         unsigned long dirino;
1412         int retval;
1413
1414         dirino = args->dirid;
1415
1416         /* set version 1, version 2 could be used too, because stat data
1417            key is the same in both versions */
1418         key.version = KEY_FORMAT_3_5;
1419         key.on_disk_key.k_dir_id = dirino;
1420         key.on_disk_key.k_objectid = inode->i_ino;
1421         key.on_disk_key.k_offset = 0;
1422         key.on_disk_key.k_type = 0;
1423
1424         /* look for the object's stat data */
1425         retval = search_item(inode->i_sb, &key, &path_to_sd);
1426         if (retval == IO_ERROR) {
1427                 reiserfs_warning(inode->i_sb,
1428                                  "vs-13070: reiserfs_read_locked_inode: "
1429                                  "i/o failure occurred trying to find stat data of %K",
1430                                  &key);
1431                 reiserfs_make_bad_inode(inode);
1432                 return;
1433         }
1434         if (retval != ITEM_FOUND) {
1435                 /* a stale NFS handle can trigger this without it being an error */
1436                 pathrelse(&path_to_sd);
1437                 reiserfs_make_bad_inode(inode);
1438                 inode->i_nlink = 0;
1439                 return;
1440         }
1441
1442         init_inode(inode, &path_to_sd);
1443
1444         /* It is possible that knfsd is trying to access inode of a file
1445            that is being removed from the disk by some other thread. As we
1446            update sd on unlink all that is required is to check for nlink
1447            here. This bug was first found by Sizif when debugging
1448            SquidNG/Butterfly, forgotten, and found again after Philippe
1449            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1450
1451            More logical fix would require changes in fs/inode.c:iput() to
1452            remove inode from hash-table _after_ fs cleaned disk stuff up and
1453            in iget() to return NULL if I_FREEING inode is found in
1454            hash-table. */
1455         /* Currently there is one place where it's ok to meet inode with
1456            nlink==0: processing of open-unlinked and half-truncated files
1457            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1458         if ((inode->i_nlink == 0) &&
1459             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1460                 reiserfs_warning(inode->i_sb,
1461                                  "vs-13075: reiserfs_read_locked_inode: "
1462                                  "dead inode read from disk %K. "
1463                                  "This is likely to be race with knfsd. Ignore",
1464                                  &key);
1465                 reiserfs_make_bad_inode(inode);
1466         }
1467
1468         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1469
1470 }
1471
1472 /**
1473  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1474  *
1475  * @inode:    inode from hash table to check
1476  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1477  *
1478  * This function is called by iget5_locked() to distinguish reiserfs inodes
1479  * having the same inode numbers. Such inodes can only exist due to some
1480  * error condition. One of them should be bad. Inodes with identical
1481  * inode numbers (objectids) are distinguished by parent directory ids.
1482  *
1483  */
1484 int reiserfs_find_actor(struct inode *inode, void *opaque)
1485 {
1486         struct reiserfs_iget_args *args;
1487
1488         args = opaque;
1489         /* args is already in CPU order */
1490         return (inode->i_ino == args->objectid) &&
1491             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1492 }
1493
1494 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1495 {
1496         struct inode *inode;
1497         struct reiserfs_iget_args args;
1498
1499         args.objectid = key->on_disk_key.k_objectid;
1500         args.dirid = key->on_disk_key.k_dir_id;
1501         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1502                              reiserfs_find_actor, reiserfs_init_locked_inode,
1503                              (void *)(&args));
1504         if (!inode)
1505                 return ERR_PTR(-ENOMEM);
1506
1507         if (inode->i_state & I_NEW) {
1508                 reiserfs_read_locked_inode(inode, &args);
1509                 unlock_new_inode(inode);
1510         }
1511
1512         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1513                 /* either due to i/o error or a stale NFS handle */
1514                 iput(inode);
1515                 inode = NULL;
1516         }
1517         return inode;
1518 }
1519
1520 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1521         u32 objectid, u32 dir_id, u32 generation)
1522
1523 {
1524         struct cpu_key key;
1525         struct inode *inode;
1526
1527         key.on_disk_key.k_objectid = objectid;
1528         key.on_disk_key.k_dir_id = dir_id;
1529         reiserfs_write_lock(sb);
1530         inode = reiserfs_iget(sb, &key);
1531         if (inode && !IS_ERR(inode) && generation != 0 &&
1532             generation != inode->i_generation) {
1533                 iput(inode);
1534                 inode = NULL;
1535         }
1536         reiserfs_write_unlock(sb);
1537
1538         return d_obtain_alias(inode);
1539 }
1540
1541 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1542                 int fh_len, int fh_type)
1543 {
1544         /* fhtype happens to reflect the number of u32s encoded.
1545          * due to a bug in earlier code, fhtype might indicate there
1546          * are more u32s then actually fitted.
1547          * so if fhtype seems to be more than len, reduce fhtype.
1548          * Valid types are:
1549          *   2 - objectid + dir_id - legacy support
1550          *   3 - objectid + dir_id + generation
1551          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1552          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1553          *   6 - as above plus generation of directory
1554          * 6 does not fit in NFSv2 handles
1555          */
1556         if (fh_type > fh_len) {
1557                 if (fh_type != 6 || fh_len != 5)
1558                         reiserfs_warning(sb,
1559                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1560                                 fh_type, fh_len);
1561                 fh_type = 5;
1562         }
1563
1564         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1565                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1566 }
1567
1568 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1569                 int fh_len, int fh_type)
1570 {
1571         if (fh_type < 4)
1572                 return NULL;
1573
1574         return reiserfs_get_dentry(sb,
1575                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1576                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1577                 (fh_type == 6) ? fid->raw[5] : 0);
1578 }
1579
1580 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1581                        int need_parent)
1582 {
1583         struct inode *inode = dentry->d_inode;
1584         int maxlen = *lenp;
1585
1586         if (maxlen < 3)
1587                 return 255;
1588
1589         data[0] = inode->i_ino;
1590         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1591         data[2] = inode->i_generation;
1592         *lenp = 3;
1593         /* no room for directory info? return what we've stored so far */
1594         if (maxlen < 5 || !need_parent)
1595                 return 3;
1596
1597         spin_lock(&dentry->d_lock);
1598         inode = dentry->d_parent->d_inode;
1599         data[3] = inode->i_ino;
1600         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1601         *lenp = 5;
1602         if (maxlen >= 6) {
1603                 data[5] = inode->i_generation;
1604                 *lenp = 6;
1605         }
1606         spin_unlock(&dentry->d_lock);
1607         return *lenp;
1608 }
1609
1610 /* looks for stat data, then copies fields to it, marks the buffer
1611    containing stat data as dirty */
1612 /* reiserfs inodes are never really dirty, since the dirty inode call
1613 ** always logs them.  This call allows the VFS inode marking routines
1614 ** to properly mark inodes for datasync and such, but only actually
1615 ** does something when called for a synchronous update.
1616 */
1617 int reiserfs_write_inode(struct inode *inode, int do_sync)
1618 {
1619         struct reiserfs_transaction_handle th;
1620         int jbegin_count = 1;
1621
1622         if (inode->i_sb->s_flags & MS_RDONLY)
1623                 return -EROFS;
1624         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1625          ** these cases are just when the system needs ram, not when the
1626          ** inode needs to reach disk for safety, and they can safely be
1627          ** ignored because the altered inode has already been logged.
1628          */
1629         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1630                 reiserfs_write_lock(inode->i_sb);
1631                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1632                         reiserfs_update_sd(&th, inode);
1633                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1634                 }
1635                 reiserfs_write_unlock(inode->i_sb);
1636         }
1637         return 0;
1638 }
1639
1640 /* stat data of new object is inserted already, this inserts the item
1641    containing "." and ".." entries */
1642 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1643                                   struct inode *inode,
1644                                   struct item_head *ih, struct treepath *path,
1645                                   struct inode *dir)
1646 {
1647         struct super_block *sb = th->t_super;
1648         char empty_dir[EMPTY_DIR_SIZE];
1649         char *body = empty_dir;
1650         struct cpu_key key;
1651         int retval;
1652
1653         BUG_ON(!th->t_trans_id);
1654
1655         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1656                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1657                       TYPE_DIRENTRY, 3 /*key length */ );
1658
1659         /* compose item head for new item. Directories consist of items of
1660            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1661            is done by reiserfs_new_inode */
1662         if (old_format_only(sb)) {
1663                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1664                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1665
1666                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1667                                        ih->ih_key.k_objectid,
1668                                        INODE_PKEY(dir)->k_dir_id,
1669                                        INODE_PKEY(dir)->k_objectid);
1670         } else {
1671                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1672                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1673
1674                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1675                                     ih->ih_key.k_objectid,
1676                                     INODE_PKEY(dir)->k_dir_id,
1677                                     INODE_PKEY(dir)->k_objectid);
1678         }
1679
1680         /* look for place in the tree for new item */
1681         retval = search_item(sb, &key, path);
1682         if (retval == IO_ERROR) {
1683                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1684                                  "i/o failure occurred creating new directory");
1685                 return -EIO;
1686         }
1687         if (retval == ITEM_FOUND) {
1688                 pathrelse(path);
1689                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1690                                  "object with this key exists (%k)",
1691                                  &(ih->ih_key));
1692                 return -EEXIST;
1693         }
1694
1695         /* insert item, that is empty directory item */
1696         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1697 }
1698
1699 /* stat data of object has been inserted, this inserts the item
1700    containing the body of symlink */
1701 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1702                                 struct item_head *ih,
1703                                 struct treepath *path, const char *symname,
1704                                 int item_len)
1705 {
1706         struct super_block *sb = th->t_super;
1707         struct cpu_key key;
1708         int retval;
1709
1710         BUG_ON(!th->t_trans_id);
1711
1712         _make_cpu_key(&key, KEY_FORMAT_3_5,
1713                       le32_to_cpu(ih->ih_key.k_dir_id),
1714                       le32_to_cpu(ih->ih_key.k_objectid),
1715                       1, TYPE_DIRECT, 3 /*key length */ );
1716
1717         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1718                           0 /*free_space */ );
1719
1720         /* look for place in the tree for new item */
1721         retval = search_item(sb, &key, path);
1722         if (retval == IO_ERROR) {
1723                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1724                                  "i/o failure occurred creating new symlink");
1725                 return -EIO;
1726         }
1727         if (retval == ITEM_FOUND) {
1728                 pathrelse(path);
1729                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1730                                  "object with this key exists (%k)",
1731                                  &(ih->ih_key));
1732                 return -EEXIST;
1733         }
1734
1735         /* insert item, that is body of symlink */
1736         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1737 }
1738
1739 /* inserts the stat data into the tree, and then calls
1740    reiserfs_new_directory (to insert ".", ".." item if new object is
1741    directory) or reiserfs_new_symlink (to insert symlink body if new
1742    object is symlink) or nothing (if new object is regular file)
1743
1744    NOTE! uid and gid must already be set in the inode.  If we return
1745    non-zero due to an error, we have to drop the quota previously allocated
1746    for the fresh inode.  This can only be done outside a transaction, so
1747    if we return non-zero, we also end the transaction.  */
1748 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1749                        struct inode *dir, int mode, const char *symname,
1750                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1751                           strlen (symname) for symlinks) */
1752                        loff_t i_size, struct dentry *dentry,
1753                        struct inode *inode)
1754 {
1755         struct super_block *sb;
1756         struct reiserfs_iget_args args;
1757         INITIALIZE_PATH(path_to_key);
1758         struct cpu_key key;
1759         struct item_head ih;
1760         struct stat_data sd;
1761         int retval;
1762         int err;
1763
1764         BUG_ON(!th->t_trans_id);
1765
1766         if (DQUOT_ALLOC_INODE(inode)) {
1767                 err = -EDQUOT;
1768                 goto out_end_trans;
1769         }
1770         if (!dir->i_nlink) {
1771                 err = -EPERM;
1772                 goto out_bad_inode;
1773         }
1774
1775         sb = dir->i_sb;
1776
1777         /* item head of new item */
1778         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1779         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1780         if (!ih.ih_key.k_objectid) {
1781                 err = -ENOMEM;
1782                 goto out_bad_inode;
1783         }
1784         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1785         if (old_format_only(sb))
1786                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1787                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1788         else
1789                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1790                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1791         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1792         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1793         if (insert_inode_locked4(inode, args.objectid,
1794                              reiserfs_find_actor, &args) < 0) {
1795                 err = -EINVAL;
1796                 goto out_bad_inode;
1797         }
1798         if (old_format_only(sb))
1799                 /* not a perfect generation count, as object ids can be reused, but
1800                  ** this is as good as reiserfs can do right now.
1801                  ** note that the private part of inode isn't filled in yet, we have
1802                  ** to use the directory.
1803                  */
1804                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1805         else
1806 #if defined( USE_INODE_GENERATION_COUNTER )
1807                 inode->i_generation =
1808                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1809 #else
1810                 inode->i_generation = ++event;
1811 #endif
1812
1813         /* fill stat data */
1814         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1815
1816         /* uid and gid must already be set by the caller for quota init */
1817
1818         /* symlink cannot be immutable or append only, right? */
1819         if (S_ISLNK(inode->i_mode))
1820                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1821
1822         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1823         inode->i_size = i_size;
1824         inode->i_blocks = 0;
1825         inode->i_bytes = 0;
1826         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1827             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1828
1829         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1830         REISERFS_I(inode)->i_flags = 0;
1831         REISERFS_I(inode)->i_prealloc_block = 0;
1832         REISERFS_I(inode)->i_prealloc_count = 0;
1833         REISERFS_I(inode)->i_trans_id = 0;
1834         REISERFS_I(inode)->i_jl = NULL;
1835         REISERFS_I(inode)->i_attrs =
1836             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1837         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1838         mutex_init(&(REISERFS_I(inode)->i_mmap));
1839         reiserfs_init_acl_access(inode);
1840         reiserfs_init_acl_default(inode);
1841         reiserfs_init_xattr_rwsem(inode);
1842
1843         /* key to search for correct place for new stat data */
1844         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1845                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1846                       TYPE_STAT_DATA, 3 /*key length */ );
1847
1848         /* find proper place for inserting of stat data */
1849         retval = search_item(sb, &key, &path_to_key);
1850         if (retval == IO_ERROR) {
1851                 err = -EIO;
1852                 goto out_bad_inode;
1853         }
1854         if (retval == ITEM_FOUND) {
1855                 pathrelse(&path_to_key);
1856                 err = -EEXIST;
1857                 goto out_bad_inode;
1858         }
1859         if (old_format_only(sb)) {
1860                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1861                         pathrelse(&path_to_key);
1862                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1863                         err = -EINVAL;
1864                         goto out_bad_inode;
1865                 }
1866                 inode2sd_v1(&sd, inode, inode->i_size);
1867         } else {
1868                 inode2sd(&sd, inode, inode->i_size);
1869         }
1870         // store in in-core inode the key of stat data and version all
1871         // object items will have (directory items will have old offset
1872         // format, other new objects will consist of new items)
1873         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1874                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1875         else
1876                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1877         if (old_format_only(sb))
1878                 set_inode_sd_version(inode, STAT_DATA_V1);
1879         else
1880                 set_inode_sd_version(inode, STAT_DATA_V2);
1881
1882         /* insert the stat data into the tree */
1883 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1884         if (REISERFS_I(dir)->new_packing_locality)
1885                 th->displace_new_blocks = 1;
1886 #endif
1887         retval =
1888             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1889                                  (char *)(&sd));
1890         if (retval) {
1891                 err = retval;
1892                 reiserfs_check_path(&path_to_key);
1893                 goto out_bad_inode;
1894         }
1895 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1896         if (!th->displace_new_blocks)
1897                 REISERFS_I(dir)->new_packing_locality = 0;
1898 #endif
1899         if (S_ISDIR(mode)) {
1900                 /* insert item with "." and ".." */
1901                 retval =
1902                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1903         }
1904
1905         if (S_ISLNK(mode)) {
1906                 /* insert body of symlink */
1907                 if (!old_format_only(sb))
1908                         i_size = ROUND_UP(i_size);
1909                 retval =
1910                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1911                                          i_size);
1912         }
1913         if (retval) {
1914                 err = retval;
1915                 reiserfs_check_path(&path_to_key);
1916                 journal_end(th, th->t_super, th->t_blocks_allocated);
1917                 goto out_inserted_sd;
1918         }
1919
1920         /* XXX CHECK THIS */
1921         if (reiserfs_posixacl(inode->i_sb)) {
1922                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1923                 if (retval) {
1924                         err = retval;
1925                         reiserfs_check_path(&path_to_key);
1926                         journal_end(th, th->t_super, th->t_blocks_allocated);
1927                         goto out_inserted_sd;
1928                 }
1929         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1930                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1931                                  "but vfs thinks they are!");
1932         } else if (is_reiserfs_priv_object(dir)) {
1933                 reiserfs_mark_inode_private(inode);
1934         }
1935
1936         reiserfs_update_sd(th, inode);
1937         reiserfs_check_path(&path_to_key);
1938
1939         return 0;
1940
1941 /* it looks like you can easily compress these two goto targets into
1942  * one.  Keeping it like this doesn't actually hurt anything, and they
1943  * are place holders for what the quota code actually needs.
1944  */
1945       out_bad_inode:
1946         /* Invalidate the object, nothing was inserted yet */
1947         INODE_PKEY(inode)->k_objectid = 0;
1948
1949         /* Quota change must be inside a transaction for journaling */
1950         DQUOT_FREE_INODE(inode);
1951
1952       out_end_trans:
1953         journal_end(th, th->t_super, th->t_blocks_allocated);
1954         /* Drop can be outside and it needs more credits so it's better to have it outside */
1955         DQUOT_DROP(inode);
1956         inode->i_flags |= S_NOQUOTA;
1957         make_bad_inode(inode);
1958
1959       out_inserted_sd:
1960         inode->i_nlink = 0;
1961         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1962         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1963
1964         /* If we were inheriting an ACL, we need to release the lock so that
1965          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1966          * code really needs to be reworked, but this will take care of it
1967          * for now. -jeffm */
1968 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1969         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1970                 reiserfs_write_unlock_xattrs(dir->i_sb);
1971                 iput(inode);
1972                 reiserfs_write_lock_xattrs(dir->i_sb);
1973         } else
1974 #endif
1975                 iput(inode);
1976         return err;
1977 }
1978
1979 /*
1980 ** finds the tail page in the page cache,
1981 ** reads the last block in.
1982 **
1983 ** On success, page_result is set to a locked, pinned page, and bh_result
1984 ** is set to an up to date buffer for the last block in the file.  returns 0.
1985 **
1986 ** tail conversion is not done, so bh_result might not be valid for writing
1987 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1988 ** trying to write the block.
1989 **
1990 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1991 */
1992 static int grab_tail_page(struct inode *p_s_inode,
1993                           struct page **page_result,
1994                           struct buffer_head **bh_result)
1995 {
1996
1997         /* we want the page with the last byte in the file,
1998          ** not the page that will hold the next byte for appending
1999          */
2000         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2001         unsigned long pos = 0;
2002         unsigned long start = 0;
2003         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2004         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2005         struct buffer_head *bh;
2006         struct buffer_head *head;
2007         struct page *page;
2008         int error;
2009
2010         /* we know that we are only called with inode->i_size > 0.
2011          ** we also know that a file tail can never be as big as a block
2012          ** If i_size % blocksize == 0, our file is currently block aligned
2013          ** and it won't need converting or zeroing after a truncate.
2014          */
2015         if ((offset & (blocksize - 1)) == 0) {
2016                 return -ENOENT;
2017         }
2018         page = grab_cache_page(p_s_inode->i_mapping, index);
2019         error = -ENOMEM;
2020         if (!page) {
2021                 goto out;
2022         }
2023         /* start within the page of the last block in the file */
2024         start = (offset / blocksize) * blocksize;
2025
2026         error = block_prepare_write(page, start, offset,
2027                                     reiserfs_get_block_create_0);
2028         if (error)
2029                 goto unlock;
2030
2031         head = page_buffers(page);
2032         bh = head;
2033         do {
2034                 if (pos >= start) {
2035                         break;
2036                 }
2037                 bh = bh->b_this_page;
2038                 pos += blocksize;
2039         } while (bh != head);
2040
2041         if (!buffer_uptodate(bh)) {
2042                 /* note, this should never happen, prepare_write should
2043                  ** be taking care of this for us.  If the buffer isn't up to date,
2044                  ** I've screwed up the code to find the buffer, or the code to
2045                  ** call prepare_write
2046                  */
2047                 reiserfs_warning(p_s_inode->i_sb,
2048                                  "clm-6000: error reading block %lu on dev %s",
2049                                  bh->b_blocknr,
2050                                  reiserfs_bdevname(p_s_inode->i_sb));
2051                 error = -EIO;
2052                 goto unlock;
2053         }
2054         *bh_result = bh;
2055         *page_result = page;
2056
2057       out:
2058         return error;
2059
2060       unlock:
2061         unlock_page(page);
2062         page_cache_release(page);
2063         return error;
2064 }
2065
2066 /*
2067 ** vfs version of truncate file.  Must NOT be called with
2068 ** a transaction already started.
2069 **
2070 ** some code taken from block_truncate_page
2071 */
2072 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2073 {
2074         struct reiserfs_transaction_handle th;
2075         /* we want the offset for the first byte after the end of the file */
2076         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2077         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2078         unsigned length;
2079         struct page *page = NULL;
2080         int error;
2081         struct buffer_head *bh = NULL;
2082         int err2;
2083
2084         reiserfs_write_lock(p_s_inode->i_sb);
2085
2086         if (p_s_inode->i_size > 0) {
2087                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2088                         // -ENOENT means we truncated past the end of the file,
2089                         // and get_block_create_0 could not find a block to read in,
2090                         // which is ok.
2091                         if (error != -ENOENT)
2092                                 reiserfs_warning(p_s_inode->i_sb,
2093                                                  "clm-6001: grab_tail_page failed %d",
2094                                                  error);
2095                         page = NULL;
2096                         bh = NULL;
2097                 }
2098         }
2099
2100         /* so, if page != NULL, we have a buffer head for the offset at
2101          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2102          ** then we have an unformatted node.  Otherwise, we have a direct item,
2103          ** and no zeroing is required on disk.  We zero after the truncate,
2104          ** because the truncate might pack the item anyway
2105          ** (it will unmap bh if it packs).
2106          */
2107         /* it is enough to reserve space in transaction for 2 balancings:
2108            one for "save" link adding and another for the first
2109            cut_from_item. 1 is for update_sd */
2110         error = journal_begin(&th, p_s_inode->i_sb,
2111                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2112         if (error)
2113                 goto out;
2114         reiserfs_update_inode_transaction(p_s_inode);
2115         if (update_timestamps)
2116                 /* we are doing real truncate: if the system crashes before the last
2117                    transaction of truncating gets committed - on reboot the file
2118                    either appears truncated properly or not truncated at all */
2119                 add_save_link(&th, p_s_inode, 1);
2120         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2121         error =
2122             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2123         if (error)
2124                 goto out;
2125
2126         /* check reiserfs_do_truncate after ending the transaction */
2127         if (err2) {
2128                 error = err2;
2129                 goto out;
2130         }
2131
2132         if (update_timestamps) {
2133                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2134                 if (error)
2135                         goto out;
2136         }
2137
2138         if (page) {
2139                 length = offset & (blocksize - 1);
2140                 /* if we are not on a block boundary */
2141                 if (length) {
2142                         length = blocksize - length;
2143                         zero_user(page, offset, length);
2144                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2145                                 mark_buffer_dirty(bh);
2146                         }
2147                 }
2148                 unlock_page(page);
2149                 page_cache_release(page);
2150         }
2151
2152         reiserfs_write_unlock(p_s_inode->i_sb);
2153         return 0;
2154       out:
2155         if (page) {
2156                 unlock_page(page);
2157                 page_cache_release(page);
2158         }
2159         reiserfs_write_unlock(p_s_inode->i_sb);
2160         return error;
2161 }
2162
2163 static int map_block_for_writepage(struct inode *inode,
2164                                    struct buffer_head *bh_result,
2165                                    unsigned long block)
2166 {
2167         struct reiserfs_transaction_handle th;
2168         int fs_gen;
2169         struct item_head tmp_ih;
2170         struct item_head *ih;
2171         struct buffer_head *bh;
2172         __le32 *item;
2173         struct cpu_key key;
2174         INITIALIZE_PATH(path);
2175         int pos_in_item;
2176         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2177         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2178         int retval;
2179         int use_get_block = 0;
2180         int bytes_copied = 0;
2181         int copy_size;
2182         int trans_running = 0;
2183
2184         /* catch places below that try to log something without starting a trans */
2185         th.t_trans_id = 0;
2186
2187         if (!buffer_uptodate(bh_result)) {
2188                 return -EIO;
2189         }
2190
2191         kmap(bh_result->b_page);
2192       start_over:
2193         reiserfs_write_lock(inode->i_sb);
2194         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2195
2196       research:
2197         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2198         if (retval != POSITION_FOUND) {
2199                 use_get_block = 1;
2200                 goto out;
2201         }
2202
2203         bh = get_last_bh(&path);
2204         ih = get_ih(&path);
2205         item = get_item(&path);
2206         pos_in_item = path.pos_in_item;
2207
2208         /* we've found an unformatted node */
2209         if (indirect_item_found(retval, ih)) {
2210                 if (bytes_copied > 0) {
2211                         reiserfs_warning(inode->i_sb,
2212                                          "clm-6002: bytes_copied %d",
2213                                          bytes_copied);
2214                 }
2215                 if (!get_block_num(item, pos_in_item)) {
2216                         /* crap, we are writing to a hole */
2217                         use_get_block = 1;
2218                         goto out;
2219                 }
2220                 set_block_dev_mapped(bh_result,
2221                                      get_block_num(item, pos_in_item), inode);
2222         } else if (is_direct_le_ih(ih)) {
2223                 char *p;
2224                 p = page_address(bh_result->b_page);
2225                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2226                 copy_size = ih_item_len(ih) - pos_in_item;
2227
2228                 fs_gen = get_generation(inode->i_sb);
2229                 copy_item_head(&tmp_ih, ih);
2230
2231                 if (!trans_running) {
2232                         /* vs-3050 is gone, no need to drop the path */
2233                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2234                         if (retval)
2235                                 goto out;
2236                         reiserfs_update_inode_transaction(inode);
2237                         trans_running = 1;
2238                         if (fs_changed(fs_gen, inode->i_sb)
2239                             && item_moved(&tmp_ih, &path)) {
2240                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2241                                                                  bh);
2242                                 goto research;
2243                         }
2244                 }
2245
2246                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2247
2248                 if (fs_changed(fs_gen, inode->i_sb)
2249                     && item_moved(&tmp_ih, &path)) {
2250                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2251                         goto research;
2252                 }
2253
2254                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2255                        copy_size);
2256
2257                 journal_mark_dirty(&th, inode->i_sb, bh);
2258                 bytes_copied += copy_size;
2259                 set_block_dev_mapped(bh_result, 0, inode);
2260
2261                 /* are there still bytes left? */
2262                 if (bytes_copied < bh_result->b_size &&
2263                     (byte_offset + bytes_copied) < inode->i_size) {
2264                         set_cpu_key_k_offset(&key,
2265                                              cpu_key_k_offset(&key) +
2266                                              copy_size);
2267                         goto research;
2268                 }
2269         } else {
2270                 reiserfs_warning(inode->i_sb,
2271                                  "clm-6003: bad item inode %lu, device %s",
2272                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2273                 retval = -EIO;
2274                 goto out;
2275         }
2276         retval = 0;
2277
2278       out:
2279         pathrelse(&path);
2280         if (trans_running) {
2281                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2282                 if (err)
2283                         retval = err;
2284                 trans_running = 0;
2285         }
2286         reiserfs_write_unlock(inode->i_sb);
2287
2288         /* this is where we fill in holes in the file. */
2289         if (use_get_block) {
2290                 retval = reiserfs_get_block(inode, block, bh_result,
2291                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2292                                             | GET_BLOCK_NO_DANGLE);
2293                 if (!retval) {
2294                         if (!buffer_mapped(bh_result)
2295                             || bh_result->b_blocknr == 0) {
2296                                 /* get_block failed to find a mapped unformatted node. */
2297                                 use_get_block = 0;
2298                                 goto start_over;
2299                         }
2300                 }
2301         }
2302         kunmap(bh_result->b_page);
2303
2304         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2305                 /* we've copied data from the page into the direct item, so the
2306                  * buffer in the page is now clean, mark it to reflect that.
2307                  */
2308                 lock_buffer(bh_result);
2309                 clear_buffer_dirty(bh_result);
2310                 unlock_buffer(bh_result);
2311         }
2312         return retval;
2313 }
2314
2315 /*
2316  * mason@suse.com: updated in 2.5.54 to follow the same general io
2317  * start/recovery path as __block_write_full_page, along with special
2318  * code to handle reiserfs tails.
2319  */
2320 static int reiserfs_write_full_page(struct page *page,
2321                                     struct writeback_control *wbc)
2322 {
2323         struct inode *inode = page->mapping->host;
2324         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2325         int error = 0;
2326         unsigned long block;
2327         sector_t last_block;
2328         struct buffer_head *head, *bh;
2329         int partial = 0;
2330         int nr = 0;
2331         int checked = PageChecked(page);
2332         struct reiserfs_transaction_handle th;
2333         struct super_block *s = inode->i_sb;
2334         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2335         th.t_trans_id = 0;
2336
2337         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2338         if (checked && (current->flags & PF_MEMALLOC)) {
2339                 redirty_page_for_writepage(wbc, page);
2340                 unlock_page(page);
2341                 return 0;
2342         }
2343
2344         /* The page dirty bit is cleared before writepage is called, which
2345          * means we have to tell create_empty_buffers to make dirty buffers
2346          * The page really should be up to date at this point, so tossing
2347          * in the BH_Uptodate is just a sanity check.
2348          */
2349         if (!page_has_buffers(page)) {
2350                 create_empty_buffers(page, s->s_blocksize,
2351                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2352         }
2353         head = page_buffers(page);
2354
2355         /* last page in the file, zero out any contents past the
2356          ** last byte in the file
2357          */
2358         if (page->index >= end_index) {
2359                 unsigned last_offset;
2360
2361                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2362                 /* no file contents in this page */
2363                 if (page->index >= end_index + 1 || !last_offset) {
2364                         unlock_page(page);
2365                         return 0;
2366                 }
2367                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2368         }
2369         bh = head;
2370         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2371         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2372         /* first map all the buffers, logging any direct items we find */
2373         do {
2374                 if (block > last_block) {
2375                         /*
2376                          * This can happen when the block size is less than
2377                          * the page size.  The corresponding bytes in the page
2378                          * were zero filled above
2379                          */
2380                         clear_buffer_dirty(bh);
2381                         set_buffer_uptodate(bh);
2382                 } else if ((checked || buffer_dirty(bh)) &&
2383                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2384                                                        && bh->b_blocknr ==
2385                                                        0))) {
2386                         /* not mapped yet, or it points to a direct item, search
2387                          * the btree for the mapping info, and log any direct
2388                          * items found
2389                          */
2390                         if ((error = map_block_for_writepage(inode, bh, block))) {
2391                                 goto fail;
2392                         }
2393                 }
2394                 bh = bh->b_this_page;
2395                 block++;
2396         } while (bh != head);
2397
2398         /*
2399          * we start the transaction after map_block_for_writepage,
2400          * because it can create holes in the file (an unbounded operation).
2401          * starting it here, we can make a reliable estimate for how many
2402          * blocks we're going to log
2403          */
2404         if (checked) {
2405                 ClearPageChecked(page);
2406                 reiserfs_write_lock(s);
2407                 error = journal_begin(&th, s, bh_per_page + 1);
2408                 if (error) {
2409                         reiserfs_write_unlock(s);
2410                         goto fail;
2411                 }
2412                 reiserfs_update_inode_transaction(inode);
2413         }
2414         /* now go through and lock any dirty buffers on the page */
2415         do {
2416                 get_bh(bh);
2417                 if (!buffer_mapped(bh))
2418                         continue;
2419                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2420                         continue;
2421
2422                 if (checked) {
2423                         reiserfs_prepare_for_journal(s, bh, 1);
2424                         journal_mark_dirty(&th, s, bh);
2425                         continue;
2426                 }
2427                 /* from this point on, we know the buffer is mapped to a
2428                  * real block and not a direct item
2429                  */
2430                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2431                         lock_buffer(bh);
2432                 } else {
2433                         if (!trylock_buffer(bh)) {
2434                                 redirty_page_for_writepage(wbc, page);
2435                                 continue;
2436                         }
2437                 }
2438                 if (test_clear_buffer_dirty(bh)) {
2439                         mark_buffer_async_write(bh);
2440                 } else {
2441                         unlock_buffer(bh);
2442                 }
2443         } while ((bh = bh->b_this_page) != head);
2444
2445         if (checked) {
2446                 error = journal_end(&th, s, bh_per_page + 1);
2447                 reiserfs_write_unlock(s);
2448                 if (error)
2449                         goto fail;
2450         }
2451         BUG_ON(PageWriteback(page));
2452         set_page_writeback(page);
2453         unlock_page(page);
2454
2455         /*
2456          * since any buffer might be the only dirty buffer on the page,
2457          * the first submit_bh can bring the page out of writeback.
2458          * be careful with the buffers.
2459          */
2460         do {
2461                 struct buffer_head *next = bh->b_this_page;
2462                 if (buffer_async_write(bh)) {
2463                         submit_bh(WRITE, bh);
2464                         nr++;
2465                 }
2466                 put_bh(bh);
2467                 bh = next;
2468         } while (bh != head);
2469
2470         error = 0;
2471       done:
2472         if (nr == 0) {
2473                 /*
2474                  * if this page only had a direct item, it is very possible for
2475                  * no io to be required without there being an error.  Or,
2476                  * someone else could have locked them and sent them down the
2477                  * pipe without locking the page
2478                  */
2479                 bh = head;
2480                 do {
2481                         if (!buffer_uptodate(bh)) {
2482                                 partial = 1;
2483                                 break;
2484                         }
2485                         bh = bh->b_this_page;
2486                 } while (bh != head);
2487                 if (!partial)
2488                         SetPageUptodate(page);
2489                 end_page_writeback(page);
2490         }
2491         return error;
2492
2493       fail:
2494         /* catches various errors, we need to make sure any valid dirty blocks
2495          * get to the media.  The page is currently locked and not marked for
2496          * writeback
2497          */
2498         ClearPageUptodate(page);
2499         bh = head;
2500         do {
2501                 get_bh(bh);
2502                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2503                         lock_buffer(bh);
2504                         mark_buffer_async_write(bh);
2505                 } else {
2506                         /*
2507                          * clear any dirty bits that might have come from getting
2508                          * attached to a dirty page
2509                          */
2510                         clear_buffer_dirty(bh);
2511                 }
2512                 bh = bh->b_this_page;
2513         } while (bh != head);
2514         SetPageError(page);
2515         BUG_ON(PageWriteback(page));
2516         set_page_writeback(page);
2517         unlock_page(page);
2518         do {
2519                 struct buffer_head *next = bh->b_this_page;
2520                 if (buffer_async_write(bh)) {
2521                         clear_buffer_dirty(bh);
2522                         submit_bh(WRITE, bh);
2523                         nr++;
2524                 }
2525                 put_bh(bh);
2526                 bh = next;
2527         } while (bh != head);
2528         goto done;
2529 }
2530
2531 static int reiserfs_readpage(struct file *f, struct page *page)
2532 {
2533         return block_read_full_page(page, reiserfs_get_block);
2534 }
2535
2536 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2537 {
2538         struct inode *inode = page->mapping->host;
2539         reiserfs_wait_on_write_block(inode->i_sb);
2540         return reiserfs_write_full_page(page, wbc);
2541 }
2542
2543 static int reiserfs_write_begin(struct file *file,
2544                                 struct address_space *mapping,
2545                                 loff_t pos, unsigned len, unsigned flags,
2546                                 struct page **pagep, void **fsdata)
2547 {
2548         struct inode *inode;
2549         struct page *page;
2550         pgoff_t index;
2551         int ret;
2552         int old_ref = 0;
2553
2554         inode = mapping->host;
2555         *fsdata = 0;
2556         if (flags & AOP_FLAG_CONT_EXPAND &&
2557             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2558                 pos ++;
2559                 *fsdata = (void *)(unsigned long)flags;
2560         }
2561
2562         index = pos >> PAGE_CACHE_SHIFT;
2563         page = grab_cache_page_write_begin(mapping, index, flags);
2564         if (!page)
2565                 return -ENOMEM;
2566         *pagep = page;
2567
2568         reiserfs_wait_on_write_block(inode->i_sb);
2569         fix_tail_page_for_writing(page);
2570         if (reiserfs_transaction_running(inode->i_sb)) {
2571                 struct reiserfs_transaction_handle *th;
2572                 th = (struct reiserfs_transaction_handle *)current->
2573                     journal_info;
2574                 BUG_ON(!th->t_refcount);
2575                 BUG_ON(!th->t_trans_id);
2576                 old_ref = th->t_refcount;
2577                 th->t_refcount++;
2578         }
2579         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2580                                 reiserfs_get_block);
2581         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2582                 struct reiserfs_transaction_handle *th = current->journal_info;
2583                 /* this gets a little ugly.  If reiserfs_get_block returned an
2584                  * error and left a transacstion running, we've got to close it,
2585                  * and we've got to free handle if it was a persistent transaction.
2586                  *
2587                  * But, if we had nested into an existing transaction, we need
2588                  * to just drop the ref count on the handle.
2589                  *
2590                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2591                  * and it was a persistent trans.  Otherwise, it was nested above.
2592                  */
2593                 if (th->t_refcount > old_ref) {
2594                         if (old_ref)
2595                                 th->t_refcount--;
2596                         else {
2597                                 int err;
2598                                 reiserfs_write_lock(inode->i_sb);
2599                                 err = reiserfs_end_persistent_transaction(th);
2600                                 reiserfs_write_unlock(inode->i_sb);
2601                                 if (err)
2602                                         ret = err;
2603                         }
2604                 }
2605         }
2606         if (ret) {
2607                 unlock_page(page);
2608                 page_cache_release(page);
2609         }
2610         return ret;
2611 }
2612
2613 int reiserfs_prepare_write(struct file *f, struct page *page,
2614                            unsigned from, unsigned to)
2615 {
2616         struct inode *inode = page->mapping->host;
2617         int ret;
2618         int old_ref = 0;
2619
2620         reiserfs_wait_on_write_block(inode->i_sb);
2621         fix_tail_page_for_writing(page);
2622         if (reiserfs_transaction_running(inode->i_sb)) {
2623                 struct reiserfs_transaction_handle *th;
2624                 th = (struct reiserfs_transaction_handle *)current->
2625                     journal_info;
2626                 BUG_ON(!th->t_refcount);
2627                 BUG_ON(!th->t_trans_id);
2628                 old_ref = th->t_refcount;
2629                 th->t_refcount++;
2630         }
2631
2632         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2633         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2634                 struct reiserfs_transaction_handle *th = current->journal_info;
2635                 /* this gets a little ugly.  If reiserfs_get_block returned an
2636                  * error and left a transacstion running, we've got to close it,
2637                  * and we've got to free handle if it was a persistent transaction.
2638                  *
2639                  * But, if we had nested into an existing transaction, we need
2640                  * to just drop the ref count on the handle.
2641                  *
2642                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2643                  * and it was a persistent trans.  Otherwise, it was nested above.
2644                  */
2645                 if (th->t_refcount > old_ref) {
2646                         if (old_ref)
2647                                 th->t_refcount--;
2648                         else {
2649                                 int err;
2650                                 reiserfs_write_lock(inode->i_sb);
2651                                 err = reiserfs_end_persistent_transaction(th);
2652                                 reiserfs_write_unlock(inode->i_sb);
2653                                 if (err)
2654                                         ret = err;
2655                         }
2656                 }
2657         }
2658         return ret;
2659
2660 }
2661
2662 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2663 {
2664         return generic_block_bmap(as, block, reiserfs_bmap);
2665 }
2666
2667 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2668                               loff_t pos, unsigned len, unsigned copied,
2669                               struct page *page, void *fsdata)
2670 {
2671         struct inode *inode = page->mapping->host;
2672         int ret = 0;
2673         int update_sd = 0;
2674         struct reiserfs_transaction_handle *th;
2675         unsigned start;
2676
2677         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2678                 pos ++;
2679
2680         reiserfs_wait_on_write_block(inode->i_sb);
2681         if (reiserfs_transaction_running(inode->i_sb))
2682                 th = current->journal_info;
2683         else
2684                 th = NULL;
2685
2686         start = pos & (PAGE_CACHE_SIZE - 1);
2687         if (unlikely(copied < len)) {
2688                 if (!PageUptodate(page))
2689                         copied = 0;
2690
2691                 page_zero_new_buffers(page, start + copied, start + len);
2692         }
2693         flush_dcache_page(page);
2694
2695         reiserfs_commit_page(inode, page, start, start + copied);
2696
2697         /* generic_commit_write does this for us, but does not update the
2698          ** transaction tracking stuff when the size changes.  So, we have
2699          ** to do the i_size updates here.
2700          */
2701         pos += copied;
2702         if (pos > inode->i_size) {
2703                 struct reiserfs_transaction_handle myth;
2704                 reiserfs_write_lock(inode->i_sb);
2705                 /* If the file have grown beyond the border where it
2706                    can have a tail, unmark it as needing a tail
2707                    packing */
2708                 if ((have_large_tails(inode->i_sb)
2709                      && inode->i_size > i_block_size(inode) * 4)
2710                     || (have_small_tails(inode->i_sb)
2711                         && inode->i_size > i_block_size(inode)))
2712                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2713
2714                 ret = journal_begin(&myth, inode->i_sb, 1);
2715                 if (ret) {
2716                         reiserfs_write_unlock(inode->i_sb);
2717                         goto journal_error;
2718                 }
2719                 reiserfs_update_inode_transaction(inode);
2720                 inode->i_size = pos;
2721                 /*
2722                  * this will just nest into our transaction.  It's important
2723                  * to use mark_inode_dirty so the inode gets pushed around on the
2724                  * dirty lists, and so that O_SYNC works as expected
2725                  */
2726                 mark_inode_dirty(inode);
2727                 reiserfs_update_sd(&myth, inode);
2728                 update_sd = 1;
2729                 ret = journal_end(&myth, inode->i_sb, 1);
2730                 reiserfs_write_unlock(inode->i_sb);
2731                 if (ret)
2732                         goto journal_error;
2733         }
2734         if (th) {
2735                 reiserfs_write_lock(inode->i_sb);
2736                 if (!update_sd)
2737                         mark_inode_dirty(inode);
2738                 ret = reiserfs_end_persistent_transaction(th);
2739                 reiserfs_write_unlock(inode->i_sb);
2740                 if (ret)
2741                         goto out;
2742         }
2743
2744       out:
2745         unlock_page(page);
2746         page_cache_release(page);
2747         return ret == 0 ? copied : ret;
2748
2749       journal_error:
2750         if (th) {
2751                 reiserfs_write_lock(inode->i_sb);
2752                 if (!update_sd)
2753                         reiserfs_update_sd(th, inode);
2754                 ret = reiserfs_end_persistent_transaction(th);
2755                 reiserfs_write_unlock(inode->i_sb);
2756         }
2757
2758         goto out;
2759 }
2760
2761 int reiserfs_commit_write(struct file *f, struct page *page,
2762                           unsigned from, unsigned to)
2763 {
2764         struct inode *inode = page->mapping->host;
2765         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2766         int ret = 0;
2767         int update_sd = 0;
2768         struct reiserfs_transaction_handle *th = NULL;
2769
2770         reiserfs_wait_on_write_block(inode->i_sb);
2771         if (reiserfs_transaction_running(inode->i_sb)) {
2772                 th = current->journal_info;
2773         }
2774         reiserfs_commit_page(inode, page, from, to);
2775
2776         /* generic_commit_write does this for us, but does not update the
2777          ** transaction tracking stuff when the size changes.  So, we have
2778          ** to do the i_size updates here.
2779          */
2780         if (pos > inode->i_size) {
2781                 struct reiserfs_transaction_handle myth;
2782                 reiserfs_write_lock(inode->i_sb);
2783                 /* If the file have grown beyond the border where it
2784                    can have a tail, unmark it as needing a tail
2785                    packing */
2786                 if ((have_large_tails(inode->i_sb)
2787                      && inode->i_size > i_block_size(inode) * 4)
2788                     || (have_small_tails(inode->i_sb)
2789                         && inode->i_size > i_block_size(inode)))
2790                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2791
2792                 ret = journal_begin(&myth, inode->i_sb, 1);
2793                 if (ret) {
2794                         reiserfs_write_unlock(inode->i_sb);
2795                         goto journal_error;
2796                 }
2797                 reiserfs_update_inode_transaction(inode);
2798                 inode->i_size = pos;
2799                 /*
2800                  * this will just nest into our transaction.  It's important
2801                  * to use mark_inode_dirty so the inode gets pushed around on the
2802                  * dirty lists, and so that O_SYNC works as expected
2803                  */
2804                 mark_inode_dirty(inode);
2805                 reiserfs_update_sd(&myth, inode);
2806                 update_sd = 1;
2807                 ret = journal_end(&myth, inode->i_sb, 1);
2808                 reiserfs_write_unlock(inode->i_sb);
2809                 if (ret)
2810                         goto journal_error;
2811         }
2812         if (th) {
2813                 reiserfs_write_lock(inode->i_sb);
2814                 if (!update_sd)
2815                         mark_inode_dirty(inode);
2816                 ret = reiserfs_end_persistent_transaction(th);
2817                 reiserfs_write_unlock(inode->i_sb);
2818                 if (ret)
2819                         goto out;
2820         }
2821
2822       out:
2823         return ret;
2824
2825       journal_error:
2826         if (th) {
2827                 reiserfs_write_lock(inode->i_sb);
2828                 if (!update_sd)
2829                         reiserfs_update_sd(th, inode);
2830                 ret = reiserfs_end_persistent_transaction(th);
2831                 reiserfs_write_unlock(inode->i_sb);
2832         }
2833
2834         return ret;
2835 }
2836
2837 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2838 {
2839         if (reiserfs_attrs(inode->i_sb)) {
2840                 if (sd_attrs & REISERFS_SYNC_FL)
2841                         inode->i_flags |= S_SYNC;
2842                 else
2843                         inode->i_flags &= ~S_SYNC;
2844                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2845                         inode->i_flags |= S_IMMUTABLE;
2846                 else
2847                         inode->i_flags &= ~S_IMMUTABLE;
2848                 if (sd_attrs & REISERFS_APPEND_FL)
2849                         inode->i_flags |= S_APPEND;
2850                 else
2851                         inode->i_flags &= ~S_APPEND;
2852                 if (sd_attrs & REISERFS_NOATIME_FL)
2853                         inode->i_flags |= S_NOATIME;
2854                 else
2855                         inode->i_flags &= ~S_NOATIME;
2856                 if (sd_attrs & REISERFS_NOTAIL_FL)
2857                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2858                 else
2859                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2860         }
2861 }
2862
2863 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2864 {
2865         if (reiserfs_attrs(inode->i_sb)) {
2866                 if (inode->i_flags & S_IMMUTABLE)
2867                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2868                 else
2869                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2870                 if (inode->i_flags & S_SYNC)
2871                         *sd_attrs |= REISERFS_SYNC_FL;
2872                 else
2873                         *sd_attrs &= ~REISERFS_SYNC_FL;
2874                 if (inode->i_flags & S_NOATIME)
2875                         *sd_attrs |= REISERFS_NOATIME_FL;
2876                 else
2877                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2878                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2879                         *sd_attrs |= REISERFS_NOTAIL_FL;
2880                 else
2881                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2882         }
2883 }
2884
2885 /* decide if this buffer needs to stay around for data logging or ordered
2886 ** write purposes
2887 */
2888 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2889 {
2890         int ret = 1;
2891         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2892
2893         lock_buffer(bh);
2894         spin_lock(&j->j_dirty_buffers_lock);
2895         if (!buffer_mapped(bh)) {
2896                 goto free_jh;
2897         }
2898         /* the page is locked, and the only places that log a data buffer
2899          * also lock the page.
2900          */
2901         if (reiserfs_file_data_log(inode)) {
2902                 /*
2903                  * very conservative, leave the buffer pinned if
2904                  * anyone might need it.
2905                  */
2906                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2907                         ret = 0;
2908                 }
2909         } else  if (buffer_dirty(bh)) {
2910                 struct reiserfs_journal_list *jl;
2911                 struct reiserfs_jh *jh = bh->b_private;
2912
2913                 /* why is this safe?
2914                  * reiserfs_setattr updates i_size in the on disk
2915                  * stat data before allowing vmtruncate to be called.
2916                  *
2917                  * If buffer was put onto the ordered list for this
2918                  * transaction, we know for sure either this transaction
2919                  * or an older one already has updated i_size on disk,
2920                  * and this ordered data won't be referenced in the file
2921                  * if we crash.
2922                  *
2923                  * if the buffer was put onto the ordered list for an older
2924                  * transaction, we need to leave it around
2925                  */
2926                 if (jh && (jl = jh->jl)
2927                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2928                         ret = 0;
2929         }
2930       free_jh:
2931         if (ret && bh->b_private) {
2932                 reiserfs_free_jh(bh);
2933         }
2934         spin_unlock(&j->j_dirty_buffers_lock);
2935         unlock_buffer(bh);
2936         return ret;
2937 }
2938
2939 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2940 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2941 {
2942         struct buffer_head *head, *bh, *next;
2943         struct inode *inode = page->mapping->host;
2944         unsigned int curr_off = 0;
2945         int ret = 1;
2946
2947         BUG_ON(!PageLocked(page));
2948
2949         if (offset == 0)
2950                 ClearPageChecked(page);
2951
2952         if (!page_has_buffers(page))
2953                 goto out;
2954
2955         head = page_buffers(page);
2956         bh = head;
2957         do {
2958                 unsigned int next_off = curr_off + bh->b_size;
2959                 next = bh->b_this_page;
2960
2961                 /*
2962                  * is this block fully invalidated?
2963                  */
2964                 if (offset <= curr_off) {
2965                         if (invalidatepage_can_drop(inode, bh))
2966                                 reiserfs_unmap_buffer(bh);
2967                         else
2968                                 ret = 0;
2969                 }
2970                 curr_off = next_off;
2971                 bh = next;
2972         } while (bh != head);
2973
2974         /*
2975          * We release buffers only if the entire page is being invalidated.
2976          * The get_block cached value has been unconditionally invalidated,
2977          * so real IO is not possible anymore.
2978          */
2979         if (!offset && ret) {
2980                 ret = try_to_release_page(page, 0);
2981                 /* maybe should BUG_ON(!ret); - neilb */
2982         }
2983       out:
2984         return;
2985 }
2986
2987 static int reiserfs_set_page_dirty(struct page *page)
2988 {
2989         struct inode *inode = page->mapping->host;
2990         if (reiserfs_file_data_log(inode)) {
2991                 SetPageChecked(page);
2992                 return __set_page_dirty_nobuffers(page);
2993         }
2994         return __set_page_dirty_buffers(page);
2995 }
2996
2997 /*
2998  * Returns 1 if the page's buffers were dropped.  The page is locked.
2999  *
3000  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3001  * in the buffers at page_buffers(page).
3002  *
3003  * even in -o notail mode, we can't be sure an old mount without -o notail
3004  * didn't create files with tails.
3005  */
3006 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3007 {
3008         struct inode *inode = page->mapping->host;
3009         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3010         struct buffer_head *head;
3011         struct buffer_head *bh;
3012         int ret = 1;
3013
3014         WARN_ON(PageChecked(page));
3015         spin_lock(&j->j_dirty_buffers_lock);
3016         head = page_buffers(page);
3017         bh = head;
3018         do {
3019                 if (bh->b_private) {
3020                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3021                                 reiserfs_free_jh(bh);
3022                         } else {
3023                                 ret = 0;
3024                                 break;
3025                         }
3026                 }
3027                 bh = bh->b_this_page;
3028         } while (bh != head);
3029         if (ret)
3030                 ret = try_to_free_buffers(page);
3031         spin_unlock(&j->j_dirty_buffers_lock);
3032         return ret;
3033 }
3034
3035 /* We thank Mingming Cao for helping us understand in great detail what
3036    to do in this section of the code. */
3037 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3038                                   const struct iovec *iov, loff_t offset,
3039                                   unsigned long nr_segs)
3040 {
3041         struct file *file = iocb->ki_filp;
3042         struct inode *inode = file->f_mapping->host;
3043
3044         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3045                                   offset, nr_segs,
3046                                   reiserfs_get_blocks_direct_io, NULL);
3047 }
3048
3049 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3050 {
3051         struct inode *inode = dentry->d_inode;
3052         int error;
3053         unsigned int ia_valid;
3054
3055         /* must be turned off for recursive notify_change calls */
3056         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3057
3058         reiserfs_write_lock(inode->i_sb);
3059         if (attr->ia_valid & ATTR_SIZE) {
3060                 /* version 2 items will be caught by the s_maxbytes check
3061                  ** done for us in vmtruncate
3062                  */
3063                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3064                     attr->ia_size > MAX_NON_LFS) {
3065                         error = -EFBIG;
3066                         goto out;
3067                 }
3068                 /* fill in hole pointers in the expanding truncate case. */
3069                 if (attr->ia_size > inode->i_size) {
3070                         error = generic_cont_expand_simple(inode, attr->ia_size);
3071                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3072                                 int err;
3073                                 struct reiserfs_transaction_handle th;
3074                                 /* we're changing at most 2 bitmaps, inode + super */
3075                                 err = journal_begin(&th, inode->i_sb, 4);
3076                                 if (!err) {
3077                                         reiserfs_discard_prealloc(&th, inode);
3078                                         err = journal_end(&th, inode->i_sb, 4);
3079                                 }
3080                                 if (err)
3081                                         error = err;
3082                         }
3083                         if (error)
3084                                 goto out;
3085                         /*
3086                          * file size is changed, ctime and mtime are
3087                          * to be updated
3088                          */
3089                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3090                 }
3091         }
3092
3093         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3094              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3095             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3096                 /* stat data of format v3.5 has 16 bit uid and gid */
3097                 error = -EINVAL;
3098                 goto out;
3099         }
3100
3101         error = inode_change_ok(inode, attr);
3102         if (!error) {
3103                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3104                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3105                         error = reiserfs_chown_xattrs(inode, attr);
3106
3107                         if (!error) {
3108                                 struct reiserfs_transaction_handle th;
3109                                 int jbegin_count =
3110                                     2 *
3111                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3112                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3113                                     2;
3114
3115                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3116                                 error =
3117                                     journal_begin(&th, inode->i_sb,
3118                                                   jbegin_count);
3119                                 if (error)
3120                                         goto out;
3121                                 error =
3122                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3123                                 if (error) {
3124                                         journal_end(&th, inode->i_sb,
3125                                                     jbegin_count);
3126                                         goto out;
3127                                 }
3128                                 /* Update corresponding info in inode so that everything is in
3129                                  * one transaction */
3130                                 if (attr->ia_valid & ATTR_UID)
3131                                         inode->i_uid = attr->ia_uid;
3132                                 if (attr->ia_valid & ATTR_GID)
3133                                         inode->i_gid = attr->ia_gid;
3134                                 mark_inode_dirty(inode);
3135                                 error =
3136                                     journal_end(&th, inode->i_sb, jbegin_count);
3137                         }
3138                 }
3139                 if (!error)
3140                         error = inode_setattr(inode, attr);
3141         }
3142
3143         if (!error && reiserfs_posixacl(inode->i_sb)) {
3144                 if (attr->ia_valid & ATTR_MODE)
3145                         error = reiserfs_acl_chmod(inode);
3146         }
3147
3148       out:
3149         reiserfs_write_unlock(inode->i_sb);
3150         return error;
3151 }
3152
3153 const struct address_space_operations reiserfs_address_space_operations = {
3154         .writepage = reiserfs_writepage,
3155         .readpage = reiserfs_readpage,
3156         .readpages = reiserfs_readpages,
3157         .releasepage = reiserfs_releasepage,
3158         .invalidatepage = reiserfs_invalidatepage,
3159         .sync_page = block_sync_page,
3160         .write_begin = reiserfs_write_begin,
3161         .write_end = reiserfs_write_end,
3162         .bmap = reiserfs_aop_bmap,
3163         .direct_IO = reiserfs_direct_IO,
3164         .set_page_dirty = reiserfs_set_page_dirty,
3165 };