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 dentry *result;
1526         struct inode *inode;
1527
1528         key.on_disk_key.k_objectid = objectid;
1529         key.on_disk_key.k_dir_id = dir_id;
1530         reiserfs_write_lock(sb);
1531         inode = reiserfs_iget(sb, &key);
1532         if (inode && !IS_ERR(inode) && generation != 0 &&
1533             generation != inode->i_generation) {
1534                 iput(inode);
1535                 inode = NULL;
1536         }
1537         reiserfs_write_unlock(sb);
1538         if (!inode)
1539                 inode = ERR_PTR(-ESTALE);
1540         if (IS_ERR(inode))
1541                 return ERR_CAST(inode);
1542         result = d_alloc_anon(inode);
1543         if (!result) {
1544                 iput(inode);
1545                 return ERR_PTR(-ENOMEM);
1546         }
1547         return result;
1548 }
1549
1550 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1551                 int fh_len, int fh_type)
1552 {
1553         /* fhtype happens to reflect the number of u32s encoded.
1554          * due to a bug in earlier code, fhtype might indicate there
1555          * are more u32s then actually fitted.
1556          * so if fhtype seems to be more than len, reduce fhtype.
1557          * Valid types are:
1558          *   2 - objectid + dir_id - legacy support
1559          *   3 - objectid + dir_id + generation
1560          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1561          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1562          *   6 - as above plus generation of directory
1563          * 6 does not fit in NFSv2 handles
1564          */
1565         if (fh_type > fh_len) {
1566                 if (fh_type != 6 || fh_len != 5)
1567                         reiserfs_warning(sb,
1568                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1569                                 fh_type, fh_len);
1570                 fh_type = 5;
1571         }
1572
1573         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1574                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1575 }
1576
1577 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1578                 int fh_len, int fh_type)
1579 {
1580         if (fh_type < 4)
1581                 return NULL;
1582
1583         return reiserfs_get_dentry(sb,
1584                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1585                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1586                 (fh_type == 6) ? fid->raw[5] : 0);
1587 }
1588
1589 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1590                        int need_parent)
1591 {
1592         struct inode *inode = dentry->d_inode;
1593         int maxlen = *lenp;
1594
1595         if (maxlen < 3)
1596                 return 255;
1597
1598         data[0] = inode->i_ino;
1599         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1600         data[2] = inode->i_generation;
1601         *lenp = 3;
1602         /* no room for directory info? return what we've stored so far */
1603         if (maxlen < 5 || !need_parent)
1604                 return 3;
1605
1606         spin_lock(&dentry->d_lock);
1607         inode = dentry->d_parent->d_inode;
1608         data[3] = inode->i_ino;
1609         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1610         *lenp = 5;
1611         if (maxlen >= 6) {
1612                 data[5] = inode->i_generation;
1613                 *lenp = 6;
1614         }
1615         spin_unlock(&dentry->d_lock);
1616         return *lenp;
1617 }
1618
1619 /* looks for stat data, then copies fields to it, marks the buffer
1620    containing stat data as dirty */
1621 /* reiserfs inodes are never really dirty, since the dirty inode call
1622 ** always logs them.  This call allows the VFS inode marking routines
1623 ** to properly mark inodes for datasync and such, but only actually
1624 ** does something when called for a synchronous update.
1625 */
1626 int reiserfs_write_inode(struct inode *inode, int do_sync)
1627 {
1628         struct reiserfs_transaction_handle th;
1629         int jbegin_count = 1;
1630
1631         if (inode->i_sb->s_flags & MS_RDONLY)
1632                 return -EROFS;
1633         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1634          ** these cases are just when the system needs ram, not when the
1635          ** inode needs to reach disk for safety, and they can safely be
1636          ** ignored because the altered inode has already been logged.
1637          */
1638         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1639                 reiserfs_write_lock(inode->i_sb);
1640                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1641                         reiserfs_update_sd(&th, inode);
1642                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1643                 }
1644                 reiserfs_write_unlock(inode->i_sb);
1645         }
1646         return 0;
1647 }
1648
1649 /* stat data of new object is inserted already, this inserts the item
1650    containing "." and ".." entries */
1651 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1652                                   struct inode *inode,
1653                                   struct item_head *ih, struct treepath *path,
1654                                   struct inode *dir)
1655 {
1656         struct super_block *sb = th->t_super;
1657         char empty_dir[EMPTY_DIR_SIZE];
1658         char *body = empty_dir;
1659         struct cpu_key key;
1660         int retval;
1661
1662         BUG_ON(!th->t_trans_id);
1663
1664         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1665                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1666                       TYPE_DIRENTRY, 3 /*key length */ );
1667
1668         /* compose item head for new item. Directories consist of items of
1669            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1670            is done by reiserfs_new_inode */
1671         if (old_format_only(sb)) {
1672                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1673                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1674
1675                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1676                                        ih->ih_key.k_objectid,
1677                                        INODE_PKEY(dir)->k_dir_id,
1678                                        INODE_PKEY(dir)->k_objectid);
1679         } else {
1680                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1681                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1682
1683                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1684                                     ih->ih_key.k_objectid,
1685                                     INODE_PKEY(dir)->k_dir_id,
1686                                     INODE_PKEY(dir)->k_objectid);
1687         }
1688
1689         /* look for place in the tree for new item */
1690         retval = search_item(sb, &key, path);
1691         if (retval == IO_ERROR) {
1692                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1693                                  "i/o failure occurred creating new directory");
1694                 return -EIO;
1695         }
1696         if (retval == ITEM_FOUND) {
1697                 pathrelse(path);
1698                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1699                                  "object with this key exists (%k)",
1700                                  &(ih->ih_key));
1701                 return -EEXIST;
1702         }
1703
1704         /* insert item, that is empty directory item */
1705         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1706 }
1707
1708 /* stat data of object has been inserted, this inserts the item
1709    containing the body of symlink */
1710 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1711                                 struct item_head *ih,
1712                                 struct treepath *path, const char *symname,
1713                                 int item_len)
1714 {
1715         struct super_block *sb = th->t_super;
1716         struct cpu_key key;
1717         int retval;
1718
1719         BUG_ON(!th->t_trans_id);
1720
1721         _make_cpu_key(&key, KEY_FORMAT_3_5,
1722                       le32_to_cpu(ih->ih_key.k_dir_id),
1723                       le32_to_cpu(ih->ih_key.k_objectid),
1724                       1, TYPE_DIRECT, 3 /*key length */ );
1725
1726         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1727                           0 /*free_space */ );
1728
1729         /* look for place in the tree for new item */
1730         retval = search_item(sb, &key, path);
1731         if (retval == IO_ERROR) {
1732                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1733                                  "i/o failure occurred creating new symlink");
1734                 return -EIO;
1735         }
1736         if (retval == ITEM_FOUND) {
1737                 pathrelse(path);
1738                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1739                                  "object with this key exists (%k)",
1740                                  &(ih->ih_key));
1741                 return -EEXIST;
1742         }
1743
1744         /* insert item, that is body of symlink */
1745         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1746 }
1747
1748 /* inserts the stat data into the tree, and then calls
1749    reiserfs_new_directory (to insert ".", ".." item if new object is
1750    directory) or reiserfs_new_symlink (to insert symlink body if new
1751    object is symlink) or nothing (if new object is regular file)
1752
1753    NOTE! uid and gid must already be set in the inode.  If we return
1754    non-zero due to an error, we have to drop the quota previously allocated
1755    for the fresh inode.  This can only be done outside a transaction, so
1756    if we return non-zero, we also end the transaction.  */
1757 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1758                        struct inode *dir, int mode, const char *symname,
1759                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1760                           strlen (symname) for symlinks) */
1761                        loff_t i_size, struct dentry *dentry,
1762                        struct inode *inode)
1763 {
1764         struct super_block *sb;
1765         INITIALIZE_PATH(path_to_key);
1766         struct cpu_key key;
1767         struct item_head ih;
1768         struct stat_data sd;
1769         int retval;
1770         int err;
1771
1772         BUG_ON(!th->t_trans_id);
1773
1774         if (DQUOT_ALLOC_INODE(inode)) {
1775                 err = -EDQUOT;
1776                 goto out_end_trans;
1777         }
1778         if (!dir->i_nlink) {
1779                 err = -EPERM;
1780                 goto out_bad_inode;
1781         }
1782
1783         sb = dir->i_sb;
1784
1785         /* item head of new item */
1786         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1787         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1788         if (!ih.ih_key.k_objectid) {
1789                 err = -ENOMEM;
1790                 goto out_bad_inode;
1791         }
1792         if (old_format_only(sb))
1793                 /* not a perfect generation count, as object ids can be reused, but
1794                  ** this is as good as reiserfs can do right now.
1795                  ** note that the private part of inode isn't filled in yet, we have
1796                  ** to use the directory.
1797                  */
1798                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1799         else
1800 #if defined( USE_INODE_GENERATION_COUNTER )
1801                 inode->i_generation =
1802                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1803 #else
1804                 inode->i_generation = ++event;
1805 #endif
1806
1807         /* fill stat data */
1808         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1809
1810         /* uid and gid must already be set by the caller for quota init */
1811
1812         /* symlink cannot be immutable or append only, right? */
1813         if (S_ISLNK(inode->i_mode))
1814                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1815
1816         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1817         inode->i_size = i_size;
1818         inode->i_blocks = 0;
1819         inode->i_bytes = 0;
1820         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1821             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1822
1823         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1824         REISERFS_I(inode)->i_flags = 0;
1825         REISERFS_I(inode)->i_prealloc_block = 0;
1826         REISERFS_I(inode)->i_prealloc_count = 0;
1827         REISERFS_I(inode)->i_trans_id = 0;
1828         REISERFS_I(inode)->i_jl = NULL;
1829         REISERFS_I(inode)->i_attrs =
1830             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1831         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1832         mutex_init(&(REISERFS_I(inode)->i_mmap));
1833         reiserfs_init_acl_access(inode);
1834         reiserfs_init_acl_default(inode);
1835         reiserfs_init_xattr_rwsem(inode);
1836
1837         if (old_format_only(sb))
1838                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1839                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1840         else
1841                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1842                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1843
1844         /* key to search for correct place for new stat data */
1845         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1846                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1847                       TYPE_STAT_DATA, 3 /*key length */ );
1848
1849         /* find proper place for inserting of stat data */
1850         retval = search_item(sb, &key, &path_to_key);
1851         if (retval == IO_ERROR) {
1852                 err = -EIO;
1853                 goto out_bad_inode;
1854         }
1855         if (retval == ITEM_FOUND) {
1856                 pathrelse(&path_to_key);
1857                 err = -EEXIST;
1858                 goto out_bad_inode;
1859         }
1860         if (old_format_only(sb)) {
1861                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1862                         pathrelse(&path_to_key);
1863                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1864                         err = -EINVAL;
1865                         goto out_bad_inode;
1866                 }
1867                 inode2sd_v1(&sd, inode, inode->i_size);
1868         } else {
1869                 inode2sd(&sd, inode, inode->i_size);
1870         }
1871         // these do not go to on-disk stat data
1872         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1873
1874         // store in in-core inode the key of stat data and version all
1875         // object items will have (directory items will have old offset
1876         // format, other new objects will consist of new items)
1877         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1878         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1879                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1880         else
1881                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1882         if (old_format_only(sb))
1883                 set_inode_sd_version(inode, STAT_DATA_V1);
1884         else
1885                 set_inode_sd_version(inode, STAT_DATA_V2);
1886
1887         /* insert the stat data into the tree */
1888 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1889         if (REISERFS_I(dir)->new_packing_locality)
1890                 th->displace_new_blocks = 1;
1891 #endif
1892         retval =
1893             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1894                                  (char *)(&sd));
1895         if (retval) {
1896                 err = retval;
1897                 reiserfs_check_path(&path_to_key);
1898                 goto out_bad_inode;
1899         }
1900 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1901         if (!th->displace_new_blocks)
1902                 REISERFS_I(dir)->new_packing_locality = 0;
1903 #endif
1904         if (S_ISDIR(mode)) {
1905                 /* insert item with "." and ".." */
1906                 retval =
1907                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1908         }
1909
1910         if (S_ISLNK(mode)) {
1911                 /* insert body of symlink */
1912                 if (!old_format_only(sb))
1913                         i_size = ROUND_UP(i_size);
1914                 retval =
1915                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1916                                          i_size);
1917         }
1918         if (retval) {
1919                 err = retval;
1920                 reiserfs_check_path(&path_to_key);
1921                 journal_end(th, th->t_super, th->t_blocks_allocated);
1922                 goto out_inserted_sd;
1923         }
1924
1925         /* XXX CHECK THIS */
1926         if (reiserfs_posixacl(inode->i_sb)) {
1927                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1928                 if (retval) {
1929                         err = retval;
1930                         reiserfs_check_path(&path_to_key);
1931                         journal_end(th, th->t_super, th->t_blocks_allocated);
1932                         goto out_inserted_sd;
1933                 }
1934         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1935                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1936                                  "but vfs thinks they are!");
1937         } else if (is_reiserfs_priv_object(dir)) {
1938                 reiserfs_mark_inode_private(inode);
1939         }
1940
1941         insert_inode_hash(inode);
1942         reiserfs_update_sd(th, inode);
1943         reiserfs_check_path(&path_to_key);
1944
1945         return 0;
1946
1947 /* it looks like you can easily compress these two goto targets into
1948  * one.  Keeping it like this doesn't actually hurt anything, and they
1949  * are place holders for what the quota code actually needs.
1950  */
1951       out_bad_inode:
1952         /* Invalidate the object, nothing was inserted yet */
1953         INODE_PKEY(inode)->k_objectid = 0;
1954
1955         /* Quota change must be inside a transaction for journaling */
1956         DQUOT_FREE_INODE(inode);
1957
1958       out_end_trans:
1959         journal_end(th, th->t_super, th->t_blocks_allocated);
1960         /* Drop can be outside and it needs more credits so it's better to have it outside */
1961         DQUOT_DROP(inode);
1962         inode->i_flags |= S_NOQUOTA;
1963         make_bad_inode(inode);
1964
1965       out_inserted_sd:
1966         inode->i_nlink = 0;
1967         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1968
1969         /* If we were inheriting an ACL, we need to release the lock so that
1970          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1971          * code really needs to be reworked, but this will take care of it
1972          * for now. -jeffm */
1973 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1974         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1975                 reiserfs_write_unlock_xattrs(dir->i_sb);
1976                 iput(inode);
1977                 reiserfs_write_lock_xattrs(dir->i_sb);
1978         } else
1979 #endif
1980                 iput(inode);
1981         return err;
1982 }
1983
1984 /*
1985 ** finds the tail page in the page cache,
1986 ** reads the last block in.
1987 **
1988 ** On success, page_result is set to a locked, pinned page, and bh_result
1989 ** is set to an up to date buffer for the last block in the file.  returns 0.
1990 **
1991 ** tail conversion is not done, so bh_result might not be valid for writing
1992 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1993 ** trying to write the block.
1994 **
1995 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1996 */
1997 static int grab_tail_page(struct inode *p_s_inode,
1998                           struct page **page_result,
1999                           struct buffer_head **bh_result)
2000 {
2001
2002         /* we want the page with the last byte in the file,
2003          ** not the page that will hold the next byte for appending
2004          */
2005         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2006         unsigned long pos = 0;
2007         unsigned long start = 0;
2008         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2009         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2010         struct buffer_head *bh;
2011         struct buffer_head *head;
2012         struct page *page;
2013         int error;
2014
2015         /* we know that we are only called with inode->i_size > 0.
2016          ** we also know that a file tail can never be as big as a block
2017          ** If i_size % blocksize == 0, our file is currently block aligned
2018          ** and it won't need converting or zeroing after a truncate.
2019          */
2020         if ((offset & (blocksize - 1)) == 0) {
2021                 return -ENOENT;
2022         }
2023         page = grab_cache_page(p_s_inode->i_mapping, index);
2024         error = -ENOMEM;
2025         if (!page) {
2026                 goto out;
2027         }
2028         /* start within the page of the last block in the file */
2029         start = (offset / blocksize) * blocksize;
2030
2031         error = block_prepare_write(page, start, offset,
2032                                     reiserfs_get_block_create_0);
2033         if (error)
2034                 goto unlock;
2035
2036         head = page_buffers(page);
2037         bh = head;
2038         do {
2039                 if (pos >= start) {
2040                         break;
2041                 }
2042                 bh = bh->b_this_page;
2043                 pos += blocksize;
2044         } while (bh != head);
2045
2046         if (!buffer_uptodate(bh)) {
2047                 /* note, this should never happen, prepare_write should
2048                  ** be taking care of this for us.  If the buffer isn't up to date,
2049                  ** I've screwed up the code to find the buffer, or the code to
2050                  ** call prepare_write
2051                  */
2052                 reiserfs_warning(p_s_inode->i_sb,
2053                                  "clm-6000: error reading block %lu on dev %s",
2054                                  bh->b_blocknr,
2055                                  reiserfs_bdevname(p_s_inode->i_sb));
2056                 error = -EIO;
2057                 goto unlock;
2058         }
2059         *bh_result = bh;
2060         *page_result = page;
2061
2062       out:
2063         return error;
2064
2065       unlock:
2066         unlock_page(page);
2067         page_cache_release(page);
2068         return error;
2069 }
2070
2071 /*
2072 ** vfs version of truncate file.  Must NOT be called with
2073 ** a transaction already started.
2074 **
2075 ** some code taken from block_truncate_page
2076 */
2077 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2078 {
2079         struct reiserfs_transaction_handle th;
2080         /* we want the offset for the first byte after the end of the file */
2081         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2082         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2083         unsigned length;
2084         struct page *page = NULL;
2085         int error;
2086         struct buffer_head *bh = NULL;
2087         int err2;
2088
2089         reiserfs_write_lock(p_s_inode->i_sb);
2090
2091         if (p_s_inode->i_size > 0) {
2092                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2093                         // -ENOENT means we truncated past the end of the file,
2094                         // and get_block_create_0 could not find a block to read in,
2095                         // which is ok.
2096                         if (error != -ENOENT)
2097                                 reiserfs_warning(p_s_inode->i_sb,
2098                                                  "clm-6001: grab_tail_page failed %d",
2099                                                  error);
2100                         page = NULL;
2101                         bh = NULL;
2102                 }
2103         }
2104
2105         /* so, if page != NULL, we have a buffer head for the offset at
2106          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2107          ** then we have an unformatted node.  Otherwise, we have a direct item,
2108          ** and no zeroing is required on disk.  We zero after the truncate,
2109          ** because the truncate might pack the item anyway
2110          ** (it will unmap bh if it packs).
2111          */
2112         /* it is enough to reserve space in transaction for 2 balancings:
2113            one for "save" link adding and another for the first
2114            cut_from_item. 1 is for update_sd */
2115         error = journal_begin(&th, p_s_inode->i_sb,
2116                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2117         if (error)
2118                 goto out;
2119         reiserfs_update_inode_transaction(p_s_inode);
2120         if (update_timestamps)
2121                 /* we are doing real truncate: if the system crashes before the last
2122                    transaction of truncating gets committed - on reboot the file
2123                    either appears truncated properly or not truncated at all */
2124                 add_save_link(&th, p_s_inode, 1);
2125         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2126         error =
2127             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2128         if (error)
2129                 goto out;
2130
2131         /* check reiserfs_do_truncate after ending the transaction */
2132         if (err2) {
2133                 error = err2;
2134                 goto out;
2135         }
2136
2137         if (update_timestamps) {
2138                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2139                 if (error)
2140                         goto out;
2141         }
2142
2143         if (page) {
2144                 length = offset & (blocksize - 1);
2145                 /* if we are not on a block boundary */
2146                 if (length) {
2147                         length = blocksize - length;
2148                         zero_user(page, offset, length);
2149                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2150                                 mark_buffer_dirty(bh);
2151                         }
2152                 }
2153                 unlock_page(page);
2154                 page_cache_release(page);
2155         }
2156
2157         reiserfs_write_unlock(p_s_inode->i_sb);
2158         return 0;
2159       out:
2160         if (page) {
2161                 unlock_page(page);
2162                 page_cache_release(page);
2163         }
2164         reiserfs_write_unlock(p_s_inode->i_sb);
2165         return error;
2166 }
2167
2168 static int map_block_for_writepage(struct inode *inode,
2169                                    struct buffer_head *bh_result,
2170                                    unsigned long block)
2171 {
2172         struct reiserfs_transaction_handle th;
2173         int fs_gen;
2174         struct item_head tmp_ih;
2175         struct item_head *ih;
2176         struct buffer_head *bh;
2177         __le32 *item;
2178         struct cpu_key key;
2179         INITIALIZE_PATH(path);
2180         int pos_in_item;
2181         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2182         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2183         int retval;
2184         int use_get_block = 0;
2185         int bytes_copied = 0;
2186         int copy_size;
2187         int trans_running = 0;
2188
2189         /* catch places below that try to log something without starting a trans */
2190         th.t_trans_id = 0;
2191
2192         if (!buffer_uptodate(bh_result)) {
2193                 return -EIO;
2194         }
2195
2196         kmap(bh_result->b_page);
2197       start_over:
2198         reiserfs_write_lock(inode->i_sb);
2199         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2200
2201       research:
2202         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2203         if (retval != POSITION_FOUND) {
2204                 use_get_block = 1;
2205                 goto out;
2206         }
2207
2208         bh = get_last_bh(&path);
2209         ih = get_ih(&path);
2210         item = get_item(&path);
2211         pos_in_item = path.pos_in_item;
2212
2213         /* we've found an unformatted node */
2214         if (indirect_item_found(retval, ih)) {
2215                 if (bytes_copied > 0) {
2216                         reiserfs_warning(inode->i_sb,
2217                                          "clm-6002: bytes_copied %d",
2218                                          bytes_copied);
2219                 }
2220                 if (!get_block_num(item, pos_in_item)) {
2221                         /* crap, we are writing to a hole */
2222                         use_get_block = 1;
2223                         goto out;
2224                 }
2225                 set_block_dev_mapped(bh_result,
2226                                      get_block_num(item, pos_in_item), inode);
2227         } else if (is_direct_le_ih(ih)) {
2228                 char *p;
2229                 p = page_address(bh_result->b_page);
2230                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2231                 copy_size = ih_item_len(ih) - pos_in_item;
2232
2233                 fs_gen = get_generation(inode->i_sb);
2234                 copy_item_head(&tmp_ih, ih);
2235
2236                 if (!trans_running) {
2237                         /* vs-3050 is gone, no need to drop the path */
2238                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2239                         if (retval)
2240                                 goto out;
2241                         reiserfs_update_inode_transaction(inode);
2242                         trans_running = 1;
2243                         if (fs_changed(fs_gen, inode->i_sb)
2244                             && item_moved(&tmp_ih, &path)) {
2245                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2246                                                                  bh);
2247                                 goto research;
2248                         }
2249                 }
2250
2251                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2252
2253                 if (fs_changed(fs_gen, inode->i_sb)
2254                     && item_moved(&tmp_ih, &path)) {
2255                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2256                         goto research;
2257                 }
2258
2259                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2260                        copy_size);
2261
2262                 journal_mark_dirty(&th, inode->i_sb, bh);
2263                 bytes_copied += copy_size;
2264                 set_block_dev_mapped(bh_result, 0, inode);
2265
2266                 /* are there still bytes left? */
2267                 if (bytes_copied < bh_result->b_size &&
2268                     (byte_offset + bytes_copied) < inode->i_size) {
2269                         set_cpu_key_k_offset(&key,
2270                                              cpu_key_k_offset(&key) +
2271                                              copy_size);
2272                         goto research;
2273                 }
2274         } else {
2275                 reiserfs_warning(inode->i_sb,
2276                                  "clm-6003: bad item inode %lu, device %s",
2277                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2278                 retval = -EIO;
2279                 goto out;
2280         }
2281         retval = 0;
2282
2283       out:
2284         pathrelse(&path);
2285         if (trans_running) {
2286                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2287                 if (err)
2288                         retval = err;
2289                 trans_running = 0;
2290         }
2291         reiserfs_write_unlock(inode->i_sb);
2292
2293         /* this is where we fill in holes in the file. */
2294         if (use_get_block) {
2295                 retval = reiserfs_get_block(inode, block, bh_result,
2296                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2297                                             | GET_BLOCK_NO_DANGLE);
2298                 if (!retval) {
2299                         if (!buffer_mapped(bh_result)
2300                             || bh_result->b_blocknr == 0) {
2301                                 /* get_block failed to find a mapped unformatted node. */
2302                                 use_get_block = 0;
2303                                 goto start_over;
2304                         }
2305                 }
2306         }
2307         kunmap(bh_result->b_page);
2308
2309         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2310                 /* we've copied data from the page into the direct item, so the
2311                  * buffer in the page is now clean, mark it to reflect that.
2312                  */
2313                 lock_buffer(bh_result);
2314                 clear_buffer_dirty(bh_result);
2315                 unlock_buffer(bh_result);
2316         }
2317         return retval;
2318 }
2319
2320 /*
2321  * mason@suse.com: updated in 2.5.54 to follow the same general io
2322  * start/recovery path as __block_write_full_page, along with special
2323  * code to handle reiserfs tails.
2324  */
2325 static int reiserfs_write_full_page(struct page *page,
2326                                     struct writeback_control *wbc)
2327 {
2328         struct inode *inode = page->mapping->host;
2329         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2330         int error = 0;
2331         unsigned long block;
2332         sector_t last_block;
2333         struct buffer_head *head, *bh;
2334         int partial = 0;
2335         int nr = 0;
2336         int checked = PageChecked(page);
2337         struct reiserfs_transaction_handle th;
2338         struct super_block *s = inode->i_sb;
2339         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2340         th.t_trans_id = 0;
2341
2342         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2343         if (checked && (current->flags & PF_MEMALLOC)) {
2344                 redirty_page_for_writepage(wbc, page);
2345                 unlock_page(page);
2346                 return 0;
2347         }
2348
2349         /* The page dirty bit is cleared before writepage is called, which
2350          * means we have to tell create_empty_buffers to make dirty buffers
2351          * The page really should be up to date at this point, so tossing
2352          * in the BH_Uptodate is just a sanity check.
2353          */
2354         if (!page_has_buffers(page)) {
2355                 create_empty_buffers(page, s->s_blocksize,
2356                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2357         }
2358         head = page_buffers(page);
2359
2360         /* last page in the file, zero out any contents past the
2361          ** last byte in the file
2362          */
2363         if (page->index >= end_index) {
2364                 unsigned last_offset;
2365
2366                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2367                 /* no file contents in this page */
2368                 if (page->index >= end_index + 1 || !last_offset) {
2369                         unlock_page(page);
2370                         return 0;
2371                 }
2372                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2373         }
2374         bh = head;
2375         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2376         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2377         /* first map all the buffers, logging any direct items we find */
2378         do {
2379                 if (block > last_block) {
2380                         /*
2381                          * This can happen when the block size is less than
2382                          * the page size.  The corresponding bytes in the page
2383                          * were zero filled above
2384                          */
2385                         clear_buffer_dirty(bh);
2386                         set_buffer_uptodate(bh);
2387                 } else if ((checked || buffer_dirty(bh)) &&
2388                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2389                                                        && bh->b_blocknr ==
2390                                                        0))) {
2391                         /* not mapped yet, or it points to a direct item, search
2392                          * the btree for the mapping info, and log any direct
2393                          * items found
2394                          */
2395                         if ((error = map_block_for_writepage(inode, bh, block))) {
2396                                 goto fail;
2397                         }
2398                 }
2399                 bh = bh->b_this_page;
2400                 block++;
2401         } while (bh != head);
2402
2403         /*
2404          * we start the transaction after map_block_for_writepage,
2405          * because it can create holes in the file (an unbounded operation).
2406          * starting it here, we can make a reliable estimate for how many
2407          * blocks we're going to log
2408          */
2409         if (checked) {
2410                 ClearPageChecked(page);
2411                 reiserfs_write_lock(s);
2412                 error = journal_begin(&th, s, bh_per_page + 1);
2413                 if (error) {
2414                         reiserfs_write_unlock(s);
2415                         goto fail;
2416                 }
2417                 reiserfs_update_inode_transaction(inode);
2418         }
2419         /* now go through and lock any dirty buffers on the page */
2420         do {
2421                 get_bh(bh);
2422                 if (!buffer_mapped(bh))
2423                         continue;
2424                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2425                         continue;
2426
2427                 if (checked) {
2428                         reiserfs_prepare_for_journal(s, bh, 1);
2429                         journal_mark_dirty(&th, s, bh);
2430                         continue;
2431                 }
2432                 /* from this point on, we know the buffer is mapped to a
2433                  * real block and not a direct item
2434                  */
2435                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2436                         lock_buffer(bh);
2437                 } else {
2438                         if (!trylock_buffer(bh)) {
2439                                 redirty_page_for_writepage(wbc, page);
2440                                 continue;
2441                         }
2442                 }
2443                 if (test_clear_buffer_dirty(bh)) {
2444                         mark_buffer_async_write(bh);
2445                 } else {
2446                         unlock_buffer(bh);
2447                 }
2448         } while ((bh = bh->b_this_page) != head);
2449
2450         if (checked) {
2451                 error = journal_end(&th, s, bh_per_page + 1);
2452                 reiserfs_write_unlock(s);
2453                 if (error)
2454                         goto fail;
2455         }
2456         BUG_ON(PageWriteback(page));
2457         set_page_writeback(page);
2458         unlock_page(page);
2459
2460         /*
2461          * since any buffer might be the only dirty buffer on the page,
2462          * the first submit_bh can bring the page out of writeback.
2463          * be careful with the buffers.
2464          */
2465         do {
2466                 struct buffer_head *next = bh->b_this_page;
2467                 if (buffer_async_write(bh)) {
2468                         submit_bh(WRITE, bh);
2469                         nr++;
2470                 }
2471                 put_bh(bh);
2472                 bh = next;
2473         } while (bh != head);
2474
2475         error = 0;
2476       done:
2477         if (nr == 0) {
2478                 /*
2479                  * if this page only had a direct item, it is very possible for
2480                  * no io to be required without there being an error.  Or,
2481                  * someone else could have locked them and sent them down the
2482                  * pipe without locking the page
2483                  */
2484                 bh = head;
2485                 do {
2486                         if (!buffer_uptodate(bh)) {
2487                                 partial = 1;
2488                                 break;
2489                         }
2490                         bh = bh->b_this_page;
2491                 } while (bh != head);
2492                 if (!partial)
2493                         SetPageUptodate(page);
2494                 end_page_writeback(page);
2495         }
2496         return error;
2497
2498       fail:
2499         /* catches various errors, we need to make sure any valid dirty blocks
2500          * get to the media.  The page is currently locked and not marked for
2501          * writeback
2502          */
2503         ClearPageUptodate(page);
2504         bh = head;
2505         do {
2506                 get_bh(bh);
2507                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2508                         lock_buffer(bh);
2509                         mark_buffer_async_write(bh);
2510                 } else {
2511                         /*
2512                          * clear any dirty bits that might have come from getting
2513                          * attached to a dirty page
2514                          */
2515                         clear_buffer_dirty(bh);
2516                 }
2517                 bh = bh->b_this_page;
2518         } while (bh != head);
2519         SetPageError(page);
2520         BUG_ON(PageWriteback(page));
2521         set_page_writeback(page);
2522         unlock_page(page);
2523         do {
2524                 struct buffer_head *next = bh->b_this_page;
2525                 if (buffer_async_write(bh)) {
2526                         clear_buffer_dirty(bh);
2527                         submit_bh(WRITE, bh);
2528                         nr++;
2529                 }
2530                 put_bh(bh);
2531                 bh = next;
2532         } while (bh != head);
2533         goto done;
2534 }
2535
2536 static int reiserfs_readpage(struct file *f, struct page *page)
2537 {
2538         return block_read_full_page(page, reiserfs_get_block);
2539 }
2540
2541 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2542 {
2543         struct inode *inode = page->mapping->host;
2544         reiserfs_wait_on_write_block(inode->i_sb);
2545         return reiserfs_write_full_page(page, wbc);
2546 }
2547
2548 static int reiserfs_write_begin(struct file *file,
2549                                 struct address_space *mapping,
2550                                 loff_t pos, unsigned len, unsigned flags,
2551                                 struct page **pagep, void **fsdata)
2552 {
2553         struct inode *inode;
2554         struct page *page;
2555         pgoff_t index;
2556         int ret;
2557         int old_ref = 0;
2558
2559         inode = mapping->host;
2560         *fsdata = 0;
2561         if (flags & AOP_FLAG_CONT_EXPAND &&
2562             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2563                 pos ++;
2564                 *fsdata = (void *)(unsigned long)flags;
2565         }
2566
2567         index = pos >> PAGE_CACHE_SHIFT;
2568         page = __grab_cache_page(mapping, index);
2569         if (!page)
2570                 return -ENOMEM;
2571         *pagep = page;
2572
2573         reiserfs_wait_on_write_block(inode->i_sb);
2574         fix_tail_page_for_writing(page);
2575         if (reiserfs_transaction_running(inode->i_sb)) {
2576                 struct reiserfs_transaction_handle *th;
2577                 th = (struct reiserfs_transaction_handle *)current->
2578                     journal_info;
2579                 BUG_ON(!th->t_refcount);
2580                 BUG_ON(!th->t_trans_id);
2581                 old_ref = th->t_refcount;
2582                 th->t_refcount++;
2583         }
2584         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2585                                 reiserfs_get_block);
2586         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2587                 struct reiserfs_transaction_handle *th = current->journal_info;
2588                 /* this gets a little ugly.  If reiserfs_get_block returned an
2589                  * error and left a transacstion running, we've got to close it,
2590                  * and we've got to free handle if it was a persistent transaction.
2591                  *
2592                  * But, if we had nested into an existing transaction, we need
2593                  * to just drop the ref count on the handle.
2594                  *
2595                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2596                  * and it was a persistent trans.  Otherwise, it was nested above.
2597                  */
2598                 if (th->t_refcount > old_ref) {
2599                         if (old_ref)
2600                                 th->t_refcount--;
2601                         else {
2602                                 int err;
2603                                 reiserfs_write_lock(inode->i_sb);
2604                                 err = reiserfs_end_persistent_transaction(th);
2605                                 reiserfs_write_unlock(inode->i_sb);
2606                                 if (err)
2607                                         ret = err;
2608                         }
2609                 }
2610         }
2611         if (ret) {
2612                 unlock_page(page);
2613                 page_cache_release(page);
2614         }
2615         return ret;
2616 }
2617
2618 int reiserfs_prepare_write(struct file *f, struct page *page,
2619                            unsigned from, unsigned to)
2620 {
2621         struct inode *inode = page->mapping->host;
2622         int ret;
2623         int old_ref = 0;
2624
2625         reiserfs_wait_on_write_block(inode->i_sb);
2626         fix_tail_page_for_writing(page);
2627         if (reiserfs_transaction_running(inode->i_sb)) {
2628                 struct reiserfs_transaction_handle *th;
2629                 th = (struct reiserfs_transaction_handle *)current->
2630                     journal_info;
2631                 BUG_ON(!th->t_refcount);
2632                 BUG_ON(!th->t_trans_id);
2633                 old_ref = th->t_refcount;
2634                 th->t_refcount++;
2635         }
2636
2637         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2638         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2639                 struct reiserfs_transaction_handle *th = current->journal_info;
2640                 /* this gets a little ugly.  If reiserfs_get_block returned an
2641                  * error and left a transacstion running, we've got to close it,
2642                  * and we've got to free handle if it was a persistent transaction.
2643                  *
2644                  * But, if we had nested into an existing transaction, we need
2645                  * to just drop the ref count on the handle.
2646                  *
2647                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2648                  * and it was a persistent trans.  Otherwise, it was nested above.
2649                  */
2650                 if (th->t_refcount > old_ref) {
2651                         if (old_ref)
2652                                 th->t_refcount--;
2653                         else {
2654                                 int err;
2655                                 reiserfs_write_lock(inode->i_sb);
2656                                 err = reiserfs_end_persistent_transaction(th);
2657                                 reiserfs_write_unlock(inode->i_sb);
2658                                 if (err)
2659                                         ret = err;
2660                         }
2661                 }
2662         }
2663         return ret;
2664
2665 }
2666
2667 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2668 {
2669         return generic_block_bmap(as, block, reiserfs_bmap);
2670 }
2671
2672 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2673                               loff_t pos, unsigned len, unsigned copied,
2674                               struct page *page, void *fsdata)
2675 {
2676         struct inode *inode = page->mapping->host;
2677         int ret = 0;
2678         int update_sd = 0;
2679         struct reiserfs_transaction_handle *th;
2680         unsigned start;
2681
2682         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2683                 pos ++;
2684
2685         reiserfs_wait_on_write_block(inode->i_sb);
2686         if (reiserfs_transaction_running(inode->i_sb))
2687                 th = current->journal_info;
2688         else
2689                 th = NULL;
2690
2691         start = pos & (PAGE_CACHE_SIZE - 1);
2692         if (unlikely(copied < len)) {
2693                 if (!PageUptodate(page))
2694                         copied = 0;
2695
2696                 page_zero_new_buffers(page, start + copied, start + len);
2697         }
2698         flush_dcache_page(page);
2699
2700         reiserfs_commit_page(inode, page, start, start + copied);
2701
2702         /* generic_commit_write does this for us, but does not update the
2703          ** transaction tracking stuff when the size changes.  So, we have
2704          ** to do the i_size updates here.
2705          */
2706         pos += copied;
2707         if (pos > inode->i_size) {
2708                 struct reiserfs_transaction_handle myth;
2709                 reiserfs_write_lock(inode->i_sb);
2710                 /* If the file have grown beyond the border where it
2711                    can have a tail, unmark it as needing a tail
2712                    packing */
2713                 if ((have_large_tails(inode->i_sb)
2714                      && inode->i_size > i_block_size(inode) * 4)
2715                     || (have_small_tails(inode->i_sb)
2716                         && inode->i_size > i_block_size(inode)))
2717                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2718
2719                 ret = journal_begin(&myth, inode->i_sb, 1);
2720                 if (ret) {
2721                         reiserfs_write_unlock(inode->i_sb);
2722                         goto journal_error;
2723                 }
2724                 reiserfs_update_inode_transaction(inode);
2725                 inode->i_size = pos;
2726                 /*
2727                  * this will just nest into our transaction.  It's important
2728                  * to use mark_inode_dirty so the inode gets pushed around on the
2729                  * dirty lists, and so that O_SYNC works as expected
2730                  */
2731                 mark_inode_dirty(inode);
2732                 reiserfs_update_sd(&myth, inode);
2733                 update_sd = 1;
2734                 ret = journal_end(&myth, inode->i_sb, 1);
2735                 reiserfs_write_unlock(inode->i_sb);
2736                 if (ret)
2737                         goto journal_error;
2738         }
2739         if (th) {
2740                 reiserfs_write_lock(inode->i_sb);
2741                 if (!update_sd)
2742                         mark_inode_dirty(inode);
2743                 ret = reiserfs_end_persistent_transaction(th);
2744                 reiserfs_write_unlock(inode->i_sb);
2745                 if (ret)
2746                         goto out;
2747         }
2748
2749       out:
2750         unlock_page(page);
2751         page_cache_release(page);
2752         return ret == 0 ? copied : ret;
2753
2754       journal_error:
2755         if (th) {
2756                 reiserfs_write_lock(inode->i_sb);
2757                 if (!update_sd)
2758                         reiserfs_update_sd(th, inode);
2759                 ret = reiserfs_end_persistent_transaction(th);
2760                 reiserfs_write_unlock(inode->i_sb);
2761         }
2762
2763         goto out;
2764 }
2765
2766 int reiserfs_commit_write(struct file *f, struct page *page,
2767                           unsigned from, unsigned to)
2768 {
2769         struct inode *inode = page->mapping->host;
2770         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2771         int ret = 0;
2772         int update_sd = 0;
2773         struct reiserfs_transaction_handle *th = NULL;
2774
2775         reiserfs_wait_on_write_block(inode->i_sb);
2776         if (reiserfs_transaction_running(inode->i_sb)) {
2777                 th = current->journal_info;
2778         }
2779         reiserfs_commit_page(inode, page, from, to);
2780
2781         /* generic_commit_write does this for us, but does not update the
2782          ** transaction tracking stuff when the size changes.  So, we have
2783          ** to do the i_size updates here.
2784          */
2785         if (pos > inode->i_size) {
2786                 struct reiserfs_transaction_handle myth;
2787                 reiserfs_write_lock(inode->i_sb);
2788                 /* If the file have grown beyond the border where it
2789                    can have a tail, unmark it as needing a tail
2790                    packing */
2791                 if ((have_large_tails(inode->i_sb)
2792                      && inode->i_size > i_block_size(inode) * 4)
2793                     || (have_small_tails(inode->i_sb)
2794                         && inode->i_size > i_block_size(inode)))
2795                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2796
2797                 ret = journal_begin(&myth, inode->i_sb, 1);
2798                 if (ret) {
2799                         reiserfs_write_unlock(inode->i_sb);
2800                         goto journal_error;
2801                 }
2802                 reiserfs_update_inode_transaction(inode);
2803                 inode->i_size = pos;
2804                 /*
2805                  * this will just nest into our transaction.  It's important
2806                  * to use mark_inode_dirty so the inode gets pushed around on the
2807                  * dirty lists, and so that O_SYNC works as expected
2808                  */
2809                 mark_inode_dirty(inode);
2810                 reiserfs_update_sd(&myth, inode);
2811                 update_sd = 1;
2812                 ret = journal_end(&myth, inode->i_sb, 1);
2813                 reiserfs_write_unlock(inode->i_sb);
2814                 if (ret)
2815                         goto journal_error;
2816         }
2817         if (th) {
2818                 reiserfs_write_lock(inode->i_sb);
2819                 if (!update_sd)
2820                         mark_inode_dirty(inode);
2821                 ret = reiserfs_end_persistent_transaction(th);
2822                 reiserfs_write_unlock(inode->i_sb);
2823                 if (ret)
2824                         goto out;
2825         }
2826
2827       out:
2828         return ret;
2829
2830       journal_error:
2831         if (th) {
2832                 reiserfs_write_lock(inode->i_sb);
2833                 if (!update_sd)
2834                         reiserfs_update_sd(th, inode);
2835                 ret = reiserfs_end_persistent_transaction(th);
2836                 reiserfs_write_unlock(inode->i_sb);
2837         }
2838
2839         return ret;
2840 }
2841
2842 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2843 {
2844         if (reiserfs_attrs(inode->i_sb)) {
2845                 if (sd_attrs & REISERFS_SYNC_FL)
2846                         inode->i_flags |= S_SYNC;
2847                 else
2848                         inode->i_flags &= ~S_SYNC;
2849                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2850                         inode->i_flags |= S_IMMUTABLE;
2851                 else
2852                         inode->i_flags &= ~S_IMMUTABLE;
2853                 if (sd_attrs & REISERFS_APPEND_FL)
2854                         inode->i_flags |= S_APPEND;
2855                 else
2856                         inode->i_flags &= ~S_APPEND;
2857                 if (sd_attrs & REISERFS_NOATIME_FL)
2858                         inode->i_flags |= S_NOATIME;
2859                 else
2860                         inode->i_flags &= ~S_NOATIME;
2861                 if (sd_attrs & REISERFS_NOTAIL_FL)
2862                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2863                 else
2864                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2865         }
2866 }
2867
2868 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2869 {
2870         if (reiserfs_attrs(inode->i_sb)) {
2871                 if (inode->i_flags & S_IMMUTABLE)
2872                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2873                 else
2874                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2875                 if (inode->i_flags & S_SYNC)
2876                         *sd_attrs |= REISERFS_SYNC_FL;
2877                 else
2878                         *sd_attrs &= ~REISERFS_SYNC_FL;
2879                 if (inode->i_flags & S_NOATIME)
2880                         *sd_attrs |= REISERFS_NOATIME_FL;
2881                 else
2882                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2883                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2884                         *sd_attrs |= REISERFS_NOTAIL_FL;
2885                 else
2886                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2887         }
2888 }
2889
2890 /* decide if this buffer needs to stay around for data logging or ordered
2891 ** write purposes
2892 */
2893 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2894 {
2895         int ret = 1;
2896         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2897
2898         lock_buffer(bh);
2899         spin_lock(&j->j_dirty_buffers_lock);
2900         if (!buffer_mapped(bh)) {
2901                 goto free_jh;
2902         }
2903         /* the page is locked, and the only places that log a data buffer
2904          * also lock the page.
2905          */
2906         if (reiserfs_file_data_log(inode)) {
2907                 /*
2908                  * very conservative, leave the buffer pinned if
2909                  * anyone might need it.
2910                  */
2911                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2912                         ret = 0;
2913                 }
2914         } else  if (buffer_dirty(bh)) {
2915                 struct reiserfs_journal_list *jl;
2916                 struct reiserfs_jh *jh = bh->b_private;
2917
2918                 /* why is this safe?
2919                  * reiserfs_setattr updates i_size in the on disk
2920                  * stat data before allowing vmtruncate to be called.
2921                  *
2922                  * If buffer was put onto the ordered list for this
2923                  * transaction, we know for sure either this transaction
2924                  * or an older one already has updated i_size on disk,
2925                  * and this ordered data won't be referenced in the file
2926                  * if we crash.
2927                  *
2928                  * if the buffer was put onto the ordered list for an older
2929                  * transaction, we need to leave it around
2930                  */
2931                 if (jh && (jl = jh->jl)
2932                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2933                         ret = 0;
2934         }
2935       free_jh:
2936         if (ret && bh->b_private) {
2937                 reiserfs_free_jh(bh);
2938         }
2939         spin_unlock(&j->j_dirty_buffers_lock);
2940         unlock_buffer(bh);
2941         return ret;
2942 }
2943
2944 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2945 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2946 {
2947         struct buffer_head *head, *bh, *next;
2948         struct inode *inode = page->mapping->host;
2949         unsigned int curr_off = 0;
2950         int ret = 1;
2951
2952         BUG_ON(!PageLocked(page));
2953
2954         if (offset == 0)
2955                 ClearPageChecked(page);
2956
2957         if (!page_has_buffers(page))
2958                 goto out;
2959
2960         head = page_buffers(page);
2961         bh = head;
2962         do {
2963                 unsigned int next_off = curr_off + bh->b_size;
2964                 next = bh->b_this_page;
2965
2966                 /*
2967                  * is this block fully invalidated?
2968                  */
2969                 if (offset <= curr_off) {
2970                         if (invalidatepage_can_drop(inode, bh))
2971                                 reiserfs_unmap_buffer(bh);
2972                         else
2973                                 ret = 0;
2974                 }
2975                 curr_off = next_off;
2976                 bh = next;
2977         } while (bh != head);
2978
2979         /*
2980          * We release buffers only if the entire page is being invalidated.
2981          * The get_block cached value has been unconditionally invalidated,
2982          * so real IO is not possible anymore.
2983          */
2984         if (!offset && ret) {
2985                 ret = try_to_release_page(page, 0);
2986                 /* maybe should BUG_ON(!ret); - neilb */
2987         }
2988       out:
2989         return;
2990 }
2991
2992 static int reiserfs_set_page_dirty(struct page *page)
2993 {
2994         struct inode *inode = page->mapping->host;
2995         if (reiserfs_file_data_log(inode)) {
2996                 SetPageChecked(page);
2997                 return __set_page_dirty_nobuffers(page);
2998         }
2999         return __set_page_dirty_buffers(page);
3000 }
3001
3002 /*
3003  * Returns 1 if the page's buffers were dropped.  The page is locked.
3004  *
3005  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3006  * in the buffers at page_buffers(page).
3007  *
3008  * even in -o notail mode, we can't be sure an old mount without -o notail
3009  * didn't create files with tails.
3010  */
3011 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3012 {
3013         struct inode *inode = page->mapping->host;
3014         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3015         struct buffer_head *head;
3016         struct buffer_head *bh;
3017         int ret = 1;
3018
3019         WARN_ON(PageChecked(page));
3020         spin_lock(&j->j_dirty_buffers_lock);
3021         head = page_buffers(page);
3022         bh = head;
3023         do {
3024                 if (bh->b_private) {
3025                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3026                                 reiserfs_free_jh(bh);
3027                         } else {
3028                                 ret = 0;
3029                                 break;
3030                         }
3031                 }
3032                 bh = bh->b_this_page;
3033         } while (bh != head);
3034         if (ret)
3035                 ret = try_to_free_buffers(page);
3036         spin_unlock(&j->j_dirty_buffers_lock);
3037         return ret;
3038 }
3039
3040 /* We thank Mingming Cao for helping us understand in great detail what
3041    to do in this section of the code. */
3042 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3043                                   const struct iovec *iov, loff_t offset,
3044                                   unsigned long nr_segs)
3045 {
3046         struct file *file = iocb->ki_filp;
3047         struct inode *inode = file->f_mapping->host;
3048
3049         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3050                                   offset, nr_segs,
3051                                   reiserfs_get_blocks_direct_io, NULL);
3052 }
3053
3054 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3055 {
3056         struct inode *inode = dentry->d_inode;
3057         int error;
3058         unsigned int ia_valid;
3059
3060         /* must be turned off for recursive notify_change calls */
3061         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3062
3063         reiserfs_write_lock(inode->i_sb);
3064         if (attr->ia_valid & ATTR_SIZE) {
3065                 /* version 2 items will be caught by the s_maxbytes check
3066                  ** done for us in vmtruncate
3067                  */
3068                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3069                     attr->ia_size > MAX_NON_LFS) {
3070                         error = -EFBIG;
3071                         goto out;
3072                 }
3073                 /* fill in hole pointers in the expanding truncate case. */
3074                 if (attr->ia_size > inode->i_size) {
3075                         error = generic_cont_expand_simple(inode, attr->ia_size);
3076                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3077                                 int err;
3078                                 struct reiserfs_transaction_handle th;
3079                                 /* we're changing at most 2 bitmaps, inode + super */
3080                                 err = journal_begin(&th, inode->i_sb, 4);
3081                                 if (!err) {
3082                                         reiserfs_discard_prealloc(&th, inode);
3083                                         err = journal_end(&th, inode->i_sb, 4);
3084                                 }
3085                                 if (err)
3086                                         error = err;
3087                         }
3088                         if (error)
3089                                 goto out;
3090                         /*
3091                          * file size is changed, ctime and mtime are
3092                          * to be updated
3093                          */
3094                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3095                 }
3096         }
3097
3098         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3099              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3100             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3101                 /* stat data of format v3.5 has 16 bit uid and gid */
3102                 error = -EINVAL;
3103                 goto out;
3104         }
3105
3106         error = inode_change_ok(inode, attr);
3107         if (!error) {
3108                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3109                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3110                         error = reiserfs_chown_xattrs(inode, attr);
3111
3112                         if (!error) {
3113                                 struct reiserfs_transaction_handle th;
3114                                 int jbegin_count =
3115                                     2 *
3116                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3117                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3118                                     2;
3119
3120                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3121                                 error =
3122                                     journal_begin(&th, inode->i_sb,
3123                                                   jbegin_count);
3124                                 if (error)
3125                                         goto out;
3126                                 error =
3127                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3128                                 if (error) {
3129                                         journal_end(&th, inode->i_sb,
3130                                                     jbegin_count);
3131                                         goto out;
3132                                 }
3133                                 /* Update corresponding info in inode so that everything is in
3134                                  * one transaction */
3135                                 if (attr->ia_valid & ATTR_UID)
3136                                         inode->i_uid = attr->ia_uid;
3137                                 if (attr->ia_valid & ATTR_GID)
3138                                         inode->i_gid = attr->ia_gid;
3139                                 mark_inode_dirty(inode);
3140                                 error =
3141                                     journal_end(&th, inode->i_sb, jbegin_count);
3142                         }
3143                 }
3144                 if (!error)
3145                         error = inode_setattr(inode, attr);
3146         }
3147
3148         if (!error && reiserfs_posixacl(inode->i_sb)) {
3149                 if (attr->ia_valid & ATTR_MODE)
3150                         error = reiserfs_acl_chmod(inode);
3151         }
3152
3153       out:
3154         reiserfs_write_unlock(inode->i_sb);
3155         return error;
3156 }
3157
3158 const struct address_space_operations reiserfs_address_space_operations = {
3159         .writepage = reiserfs_writepage,
3160         .readpage = reiserfs_readpage,
3161         .readpages = reiserfs_readpages,
3162         .releasepage = reiserfs_releasepage,
3163         .invalidatepage = reiserfs_invalidatepage,
3164         .sync_page = block_sync_page,
3165         .write_begin = reiserfs_write_begin,
3166         .write_end = reiserfs_write_end,
3167         .bmap = reiserfs_aop_bmap,
3168         .direct_IO = reiserfs_direct_IO,
3169         .set_page_dirty = reiserfs_set_page_dirty,
3170 };