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",
 846                                                          "convert tail failed "
 847                                                          "inode %lu, error %d",
 848                                                                  inode->i_ino,
 849                                                                  retval);
 850                                         if (allocated_block_nr) {
 851                                                 /* the bitmap, the super, and the stat data == 3 */
 852                                                 if (!th)
 853                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 854                                                 if (th)
 855                                                         reiserfs_free_block(th,
 856                                                                             inode,
 857                                                                             allocated_block_nr,
 858                                                                             1);
 859                                         }
 860                                         goto failure;
 861                                 }
 862                                 goto research;
 863                         }
 864                         retval =
 865                             direct2indirect(th, inode, &path, unbh,
 866                                             tail_offset);
 867                         if (retval) {
 868                                 reiserfs_unmap_buffer(unbh);
 869                                 reiserfs_free_block(th, inode,
 870                                                     allocated_block_nr, 1);
 871                                 goto failure;
 872                         }
 873                         /* it is important the set_buffer_uptodate is done after
 874                          ** the direct2indirect.  The buffer might contain valid
 875                          ** data newer than the data on disk (read by readpage, changed,
 876                          ** and then sent here by writepage).  direct2indirect needs
 877                          ** to know if unbh was already up to date, so it can decide
 878                          ** if the data in unbh needs to be replaced with data from
 879                          ** the disk
 880                          */
 881                         set_buffer_uptodate(unbh);
 882
 883                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 884                            buffer will disappear shortly, so it should not be added to
 885                          */
 886                         if (unbh->b_page) {
 887                                 /* we've converted the tail, so we must
 888                                  ** flush unbh before the transaction commits
 889                                  */
 890                                 reiserfs_add_tail_list(inode, unbh);
 891
 892                                 /* mark it dirty now to prevent commit_write from adding
 893                                  ** this buffer to the inode's dirty buffer list
 894                                  */
 895                                 /*
 896                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 897                                  * It's still atomic, but it sets the page dirty too,
 898                                  * which makes it eligible for writeback at any time by the
 899                                  * VM (which was also the case with __mark_buffer_dirty())
 900                                  */
 901                                 mark_buffer_dirty(unbh);
 902                         }
 903                 } else {
 904                         /* append indirect item with holes if needed, when appending
 905                            pointer to 'block'-th block use block, which is already
 906                            allocated */
 907                         struct cpu_key tmp_key;
 908                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 909                         // one block which is a fastpath
 910                         unp_t *un;
 911                         __u64 max_to_insert =
 912                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 913                             UNFM_P_SIZE;
 914                         __u64 blocks_needed;
 915
 916                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 917                                "vs-804: invalid position for append");
 918                         /* indirect item has to be appended, set up key of that position */
 919                         make_cpu_key(&tmp_key, inode,
 920                                      le_key_k_offset(version,
 921                                                      &(ih->ih_key)) +
 922                                      op_bytes_number(ih,
 923                                                      inode->i_sb->s_blocksize),
 924                                      //pos_in_item * inode->i_sb->s_blocksize,
 925                                      TYPE_INDIRECT, 3); // key type is unimportant
 926
 927                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 928                                "green-805: invalid offset");
 929                         blocks_needed =
 930                             1 +
 931                             ((cpu_key_k_offset(&key) -
 932                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 933                              s_blocksize_bits);
 934
 935                         if (blocks_needed == 1) {
 936                                 un = &unf_single;
 937                         } else {
 938                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 939                                 if (!un) {
 940                                         un = &unf_single;
 941                                         blocks_needed = 1;
 942                                         max_to_insert = 0;
 943                                 }
 944                         }
 945                         if (blocks_needed <= max_to_insert) {
 946                                 /* we are going to add target block to the file. Use allocated
 947                                    block for that */
 948                                 un[blocks_needed - 1] =
 949                                     cpu_to_le32(allocated_block_nr);
 950                                 set_block_dev_mapped(bh_result,
 951                                                      allocated_block_nr, inode);
 952                                 set_buffer_new(bh_result);
 953                                 done = 1;
 954                         } else {
 955                                 /* paste hole to the indirect item */
 956                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 957                                    only have space for one block */
 958                                 blocks_needed =
 959                                     max_to_insert ? max_to_insert : 1;
 960                         }
 961                         retval =
 962                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 963                                                      (char *)un,
 964                                                      UNFM_P_SIZE *
 965                                                      blocks_needed);
 966
 967                         if (blocks_needed != 1)
 968                                 kfree(un);
 969
 970                         if (retval) {
 971                                 reiserfs_free_block(th, inode,
 972                                                     allocated_block_nr, 1);
 973                                 goto failure;
 974                         }
 975                         if (!done) {
 976                                 /* We need to mark new file size in case this function will be
 977                                    interrupted/aborted later on. And we may do this only for
 978                                    holes. */
 979                                 inode->i_size +=
 980                                     inode->i_sb->s_blocksize * blocks_needed;
 981                         }
 982                 }
 983
 984                 if (done == 1)
 985                         break;
 986
 987                 /* this loop could log more blocks than we had originally asked
 988                  ** for.  So, we have to allow the transaction to end if it is
 989                  ** too big or too full.  Update the inode so things are
 990                  ** consistent if we crash before the function returns
 991                  **
 992                  ** release the path so that anybody waiting on the path before
 993                  ** ending their transaction will be able to continue.
 994                  */
 995                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 996                         retval = restart_transaction(th, inode, &path);
 997                         if (retval)
 998                                 goto failure;
 999                 }
1000                 /* inserting indirect pointers for a hole can take a
1001                  ** long time.  reschedule if needed
1002                  */
1003                 cond_resched();
1004
1005                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1006                 if (retval == IO_ERROR) {
1007                         retval = -EIO;
1008                         goto failure;
1009                 }
1010                 if (retval == POSITION_FOUND) {
1011                         reiserfs_warning(inode->i_sb, "vs-825",
1012                                          "%K should not be found", &key);
1013                         retval = -EEXIST;
1014                         if (allocated_block_nr)
1015                                 reiserfs_free_block(th, inode,
1016                                                     allocated_block_nr, 1);
1017                         pathrelse(&path);
1018                         goto failure;
1019                 }
1020                 bh = get_last_bh(&path);
1021                 ih = get_ih(&path);
1022                 item = get_item(&path);
1023                 pos_in_item = path.pos_in_item;
1024         } while (1);
1025
1026         retval = 0;
1027
1028       failure:
1029         if (th && (!dangle || (retval && !th->t_trans_id))) {
1030                 int err;
1031                 if (th->t_trans_id)
1032                         reiserfs_update_sd(th, inode);
1033                 err = reiserfs_end_persistent_transaction(th);
1034                 if (err)
1035                         retval = err;
1036         }
1037
1038         reiserfs_write_unlock(inode->i_sb);
1039         reiserfs_check_path(&path);
1040         return retval;
1041 }
1042
1043 static int
1044 reiserfs_readpages(struct file *file, struct address_space *mapping,
1045                    struct list_head *pages, unsigned nr_pages)
1046 {
1047         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1048 }
1049
1050 /* Compute real number of used bytes by file
1051  * Following three functions can go away when we'll have enough space in stat item
1052  */
1053 static int real_space_diff(struct inode *inode, int sd_size)
1054 {
1055         int bytes;
1056         loff_t blocksize = inode->i_sb->s_blocksize;
1057
1058         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1059                 return sd_size;
1060
1061         /* End of file is also in full block with indirect reference, so round
1062          ** up to the next block.
1063          **
1064          ** there is just no way to know if the tail is actually packed
1065          ** on the file, so we have to assume it isn't.  When we pack the
1066          ** tail, we add 4 bytes to pretend there really is an unformatted
1067          ** node pointer
1068          */
1069         bytes =
1070             ((inode->i_size +
1071               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1072             sd_size;
1073         return bytes;
1074 }
1075
1076 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1077                                         int sd_size)
1078 {
1079         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1080                 return inode->i_size +
1081                     (loff_t) (real_space_diff(inode, sd_size));
1082         }
1083         return ((loff_t) real_space_diff(inode, sd_size)) +
1084             (((loff_t) blocks) << 9);
1085 }
1086
1087 /* Compute number of blocks used by file in ReiserFS counting */
1088 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1089 {
1090         loff_t bytes = inode_get_bytes(inode);
1091         loff_t real_space = real_space_diff(inode, sd_size);
1092
1093         /* keeps fsck and non-quota versions of reiserfs happy */
1094         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1095                 bytes += (loff_t) 511;
1096         }
1097
1098         /* files from before the quota patch might i_blocks such that
1099          ** bytes < real_space.  Deal with that here to prevent it from
1100          ** going negative.
1101          */
1102         if (bytes < real_space)
1103                 return 0;
1104         return (bytes - real_space) >> 9;
1105 }
1106
1107 //
1108 // BAD: new directories have stat data of new type and all other items
1109 // of old type. Version stored in the inode says about body items, so
1110 // in update_stat_data we can not rely on inode, but have to check
1111 // item version directly
1112 //
1113
1114 // called by read_locked_inode
1115 static void init_inode(struct inode *inode, struct treepath *path)
1116 {
1117         struct buffer_head *bh;
1118         struct item_head *ih;
1119         __u32 rdev;
1120         //int version = ITEM_VERSION_1;
1121
1122         bh = PATH_PLAST_BUFFER(path);
1123         ih = PATH_PITEM_HEAD(path);
1124
1125         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1126
1127         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1128         REISERFS_I(inode)->i_flags = 0;
1129         REISERFS_I(inode)->i_prealloc_block = 0;
1130         REISERFS_I(inode)->i_prealloc_count = 0;
1131         REISERFS_I(inode)->i_trans_id = 0;
1132         REISERFS_I(inode)->i_jl = NULL;
1133         mutex_init(&(REISERFS_I(inode)->i_mmap));
1134         reiserfs_init_acl_access(inode);
1135         reiserfs_init_acl_default(inode);
1136         reiserfs_init_xattr_rwsem(inode);
1137
1138         if (stat_data_v1(ih)) {
1139                 struct stat_data_v1 *sd =
1140                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1141                 unsigned long blocks;
1142
1143                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1144                 set_inode_sd_version(inode, STAT_DATA_V1);
1145                 inode->i_mode = sd_v1_mode(sd);
1146                 inode->i_nlink = sd_v1_nlink(sd);
1147                 inode->i_uid = sd_v1_uid(sd);
1148                 inode->i_gid = sd_v1_gid(sd);
1149                 inode->i_size = sd_v1_size(sd);
1150                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1151                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1152                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1153                 inode->i_atime.tv_nsec = 0;
1154                 inode->i_ctime.tv_nsec = 0;
1155                 inode->i_mtime.tv_nsec = 0;
1156
1157                 inode->i_blocks = sd_v1_blocks(sd);
1158                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1159                 blocks = (inode->i_size + 511) >> 9;
1160                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1161                 if (inode->i_blocks > blocks) {
1162                         // there was a bug in <=3.5.23 when i_blocks could take negative
1163                         // values. Starting from 3.5.17 this value could even be stored in
1164                         // stat data. For such files we set i_blocks based on file
1165                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1166                         // only updated if file's inode will ever change
1167                         inode->i_blocks = blocks;
1168                 }
1169
1170                 rdev = sd_v1_rdev(sd);
1171                 REISERFS_I(inode)->i_first_direct_byte =
1172                     sd_v1_first_direct_byte(sd);
1173                 /* an early bug in the quota code can give us an odd number for the
1174                  ** block count.  This is incorrect, fix it here.
1175                  */
1176                 if (inode->i_blocks & 1) {
1177                         inode->i_blocks++;
1178                 }
1179                 inode_set_bytes(inode,
1180                                 to_real_used_space(inode, inode->i_blocks,
1181                                                    SD_V1_SIZE));
1182                 /* nopack is initially zero for v1 objects. For v2 objects,
1183                    nopack is initialised from sd_attrs */
1184                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1185         } else {
1186                 // new stat data found, but object may have old items
1187                 // (directories and symlinks)
1188                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1189
1190                 inode->i_mode = sd_v2_mode(sd);
1191                 inode->i_nlink = sd_v2_nlink(sd);
1192                 inode->i_uid = sd_v2_uid(sd);
1193                 inode->i_size = sd_v2_size(sd);
1194                 inode->i_gid = sd_v2_gid(sd);
1195                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1196                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1197                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1198                 inode->i_ctime.tv_nsec = 0;
1199                 inode->i_mtime.tv_nsec = 0;
1200                 inode->i_atime.tv_nsec = 0;
1201                 inode->i_blocks = sd_v2_blocks(sd);
1202                 rdev = sd_v2_rdev(sd);
1203                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1204                         inode->i_generation =
1205                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1206                 else
1207                         inode->i_generation = sd_v2_generation(sd);
1208
1209                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1210                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1211                 else
1212                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1213                 REISERFS_I(inode)->i_first_direct_byte = 0;
1214                 set_inode_sd_version(inode, STAT_DATA_V2);
1215                 inode_set_bytes(inode,
1216                                 to_real_used_space(inode, inode->i_blocks,
1217                                                    SD_V2_SIZE));
1218                 /* read persistent inode attributes from sd and initalise
1219                    generic inode flags from them */
1220                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1221                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1222         }
1223
1224         pathrelse(path);
1225         if (S_ISREG(inode->i_mode)) {
1226                 inode->i_op = &reiserfs_file_inode_operations;
1227                 inode->i_fop = &reiserfs_file_operations;
1228                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1229         } else if (S_ISDIR(inode->i_mode)) {
1230                 inode->i_op = &reiserfs_dir_inode_operations;
1231                 inode->i_fop = &reiserfs_dir_operations;
1232         } else if (S_ISLNK(inode->i_mode)) {
1233                 inode->i_op = &reiserfs_symlink_inode_operations;
1234                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1235         } else {
1236                 inode->i_blocks = 0;
1237                 inode->i_op = &reiserfs_special_inode_operations;
1238                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1239         }
1240 }
1241
1242 // update new stat data with inode fields
1243 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1244 {
1245         struct stat_data *sd_v2 = (struct stat_data *)sd;
1246         __u16 flags;
1247
1248         set_sd_v2_mode(sd_v2, inode->i_mode);
1249         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1250         set_sd_v2_uid(sd_v2, inode->i_uid);
1251         set_sd_v2_size(sd_v2, size);
1252         set_sd_v2_gid(sd_v2, inode->i_gid);
1253         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1254         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1255         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1256         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1257         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1258                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1259         else
1260                 set_sd_v2_generation(sd_v2, inode->i_generation);
1261         flags = REISERFS_I(inode)->i_attrs;
1262         i_attrs_to_sd_attrs(inode, &flags);
1263         set_sd_v2_attrs(sd_v2, flags);
1264 }
1265
1266 // used to copy inode's fields to old stat data
1267 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1268 {
1269         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1270
1271         set_sd_v1_mode(sd_v1, inode->i_mode);
1272         set_sd_v1_uid(sd_v1, inode->i_uid);
1273         set_sd_v1_gid(sd_v1, inode->i_gid);
1274         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1275         set_sd_v1_size(sd_v1, size);
1276         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1277         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1278         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1279
1280         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1281                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1282         else
1283                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1284
1285         // Sigh. i_first_direct_byte is back
1286         set_sd_v1_first_direct_byte(sd_v1,
1287                                     REISERFS_I(inode)->i_first_direct_byte);
1288 }
1289
1290 /* NOTE, you must prepare the buffer head before sending it here,
1291 ** and then log it after the call
1292 */
1293 static void update_stat_data(struct treepath *path, struct inode *inode,
1294                              loff_t size)
1295 {
1296         struct buffer_head *bh;
1297         struct item_head *ih;
1298
1299         bh = PATH_PLAST_BUFFER(path);
1300         ih = PATH_PITEM_HEAD(path);
1301
1302         if (!is_statdata_le_ih(ih))
1303                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1304                                INODE_PKEY(inode), ih);
1305
1306         if (stat_data_v1(ih)) {
1307                 // path points to old stat data
1308                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1309         } else {
1310                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1311         }
1312
1313         return;
1314 }
1315
1316 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1317                              struct inode *inode, loff_t size)
1318 {
1319         struct cpu_key key;
1320         INITIALIZE_PATH(path);
1321         struct buffer_head *bh;
1322         int fs_gen;
1323         struct item_head *ih, tmp_ih;
1324         int retval;
1325
1326         BUG_ON(!th->t_trans_id);
1327
1328         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1329
1330         for (;;) {
1331                 int pos;
1332                 /* look for the object's stat data */
1333                 retval = search_item(inode->i_sb, &key, &path);
1334                 if (retval == IO_ERROR) {
1335                         reiserfs_warning(inode->i_sb, "vs-13050",
1336                                          "i/o failure occurred trying to "
1337                                          "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, "vs-13060",
1349                                          "stat data of object %k (nlink == %d) "
1350                                          "not found (pos %d)",
1351                                          INODE_PKEY(inode), inode->i_nlink,
1352                                          pos);
1353                         reiserfs_check_path(&path);
1354                         return;
1355                 }
1356
1357                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1358                  ** FS might change.  We have to detect that, and loop back to the
1359                  ** search if the stat data item has moved
1360                  */
1361                 bh = get_last_bh(&path);
1362                 ih = get_ih(&path);
1363                 copy_item_head(&tmp_ih, ih);
1364                 fs_gen = get_generation(inode->i_sb);
1365                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1366                 if (fs_changed(fs_gen, inode->i_sb)
1367                     && item_moved(&tmp_ih, &path)) {
1368                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1369                         continue;       /* Stat_data item has been moved after scheduling. */
1370                 }
1371                 break;
1372         }
1373         update_stat_data(&path, inode, size);
1374         journal_mark_dirty(th, th->t_super, bh);
1375         pathrelse(&path);
1376         return;
1377 }
1378
1379 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1380 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1381 ** and clear the key in the private portion of the inode, otherwise a
1382 ** corresponding iput might try to delete whatever object the inode last
1383 ** represented.
1384 */
1385 static void reiserfs_make_bad_inode(struct inode *inode)
1386 {
1387         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1388         make_bad_inode(inode);
1389 }
1390
1391 //
1392 // initially this function was derived from minix or ext2's analog and
1393 // evolved as the prototype did
1394 //
1395
1396 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1397 {
1398         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1399         inode->i_ino = args->objectid;
1400         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1401         return 0;
1402 }
1403
1404 /* looks for stat data in the tree, and fills up the fields of in-core
1405    inode stat data fields */
1406 void reiserfs_read_locked_inode(struct inode *inode,
1407                                 struct reiserfs_iget_args *args)
1408 {
1409         INITIALIZE_PATH(path_to_sd);
1410         struct cpu_key key;
1411         unsigned long dirino;
1412         int retval;
1413
1414         dirino = args->dirid;
1415
1416         /* set version 1, version 2 could be used too, because stat data
1417            key is the same in both versions */
1418         key.version = KEY_FORMAT_3_5;
1419         key.on_disk_key.k_dir_id = dirino;
1420         key.on_disk_key.k_objectid = inode->i_ino;
1421         key.on_disk_key.k_offset = 0;
1422         key.on_disk_key.k_type = 0;
1423
1424         /* look for the object's stat data */
1425         retval = search_item(inode->i_sb, &key, &path_to_sd);
1426         if (retval == IO_ERROR) {
1427                 reiserfs_warning(inode->i_sb, "vs-13070",
1428                                  "i/o failure occurred trying to find "
1429                                  "stat data of %K", &key);
1430                 reiserfs_make_bad_inode(inode);
1431                 return;
1432         }
1433         if (retval != ITEM_FOUND) {
1434                 /* a stale NFS handle can trigger this without it being an error */
1435                 pathrelse(&path_to_sd);
1436                 reiserfs_make_bad_inode(inode);
1437                 inode->i_nlink = 0;
1438                 return;
1439         }
1440
1441         init_inode(inode, &path_to_sd);
1442
1443         /* It is possible that knfsd is trying to access inode of a file
1444            that is being removed from the disk by some other thread. As we
1445            update sd on unlink all that is required is to check for nlink
1446            here. This bug was first found by Sizif when debugging
1447            SquidNG/Butterfly, forgotten, and found again after Philippe
1448            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1449
1450            More logical fix would require changes in fs/inode.c:iput() to
1451            remove inode from hash-table _after_ fs cleaned disk stuff up and
1452            in iget() to return NULL if I_FREEING inode is found in
1453            hash-table. */
1454         /* Currently there is one place where it's ok to meet inode with
1455            nlink==0: processing of open-unlinked and half-truncated files
1456            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1457         if ((inode->i_nlink == 0) &&
1458             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1459                 reiserfs_warning(inode->i_sb, "vs-13075",
1460                                  "dead inode read from disk %K. "
1461                                  "This is likely to be race with knfsd. Ignore",
1462                                  &key);
1463                 reiserfs_make_bad_inode(inode);
1464         }
1465
1466         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1467
1468 }
1469
1470 /**
1471  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1472  *
1473  * @inode:    inode from hash table to check
1474  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1475  *
1476  * This function is called by iget5_locked() to distinguish reiserfs inodes
1477  * having the same inode numbers. Such inodes can only exist due to some
1478  * error condition. One of them should be bad. Inodes with identical
1479  * inode numbers (objectids) are distinguished by parent directory ids.
1480  *
1481  */
1482 int reiserfs_find_actor(struct inode *inode, void *opaque)
1483 {
1484         struct reiserfs_iget_args *args;
1485
1486         args = opaque;
1487         /* args is already in CPU order */
1488         return (inode->i_ino == args->objectid) &&
1489             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1490 }
1491
1492 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1493 {
1494         struct inode *inode;
1495         struct reiserfs_iget_args args;
1496
1497         args.objectid = key->on_disk_key.k_objectid;
1498         args.dirid = key->on_disk_key.k_dir_id;
1499         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1500                              reiserfs_find_actor, reiserfs_init_locked_inode,
1501                              (void *)(&args));
1502         if (!inode)
1503                 return ERR_PTR(-ENOMEM);
1504
1505         if (inode->i_state & I_NEW) {
1506                 reiserfs_read_locked_inode(inode, &args);
1507                 unlock_new_inode(inode);
1508         }
1509
1510         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1511                 /* either due to i/o error or a stale NFS handle */
1512                 iput(inode);
1513                 inode = NULL;
1514         }
1515         return inode;
1516 }
1517
1518 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1519         u32 objectid, u32 dir_id, u32 generation)
1520
1521 {
1522         struct cpu_key key;
1523         struct inode *inode;
1524
1525         key.on_disk_key.k_objectid = objectid;
1526         key.on_disk_key.k_dir_id = dir_id;
1527         reiserfs_write_lock(sb);
1528         inode = reiserfs_iget(sb, &key);
1529         if (inode && !IS_ERR(inode) && generation != 0 &&
1530             generation != inode->i_generation) {
1531                 iput(inode);
1532                 inode = NULL;
1533         }
1534         reiserfs_write_unlock(sb);
1535
1536         return d_obtain_alias(inode);
1537 }
1538
1539 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1540                 int fh_len, int fh_type)
1541 {
1542         /* fhtype happens to reflect the number of u32s encoded.
1543          * due to a bug in earlier code, fhtype might indicate there
1544          * are more u32s then actually fitted.
1545          * so if fhtype seems to be more than len, reduce fhtype.
1546          * Valid types are:
1547          *   2 - objectid + dir_id - legacy support
1548          *   3 - objectid + dir_id + generation
1549          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1550          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1551          *   6 - as above plus generation of directory
1552          * 6 does not fit in NFSv2 handles
1553          */
1554         if (fh_type > fh_len) {
1555                 if (fh_type != 6 || fh_len != 5)
1556                         reiserfs_warning(sb, "reiserfs-13077",
1557                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1558                                 fh_type, fh_len);
1559                 fh_type = 5;
1560         }
1561
1562         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1563                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1564 }
1565
1566 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1567                 int fh_len, int fh_type)
1568 {
1569         if (fh_type < 4)
1570                 return NULL;
1571
1572         return reiserfs_get_dentry(sb,
1573                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1574                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1575                 (fh_type == 6) ? fid->raw[5] : 0);
1576 }
1577
1578 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1579                        int need_parent)
1580 {
1581         struct inode *inode = dentry->d_inode;
1582         int maxlen = *lenp;
1583
1584         if (maxlen < 3)
1585                 return 255;
1586
1587         data[0] = inode->i_ino;
1588         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1589         data[2] = inode->i_generation;
1590         *lenp = 3;
1591         /* no room for directory info? return what we've stored so far */
1592         if (maxlen < 5 || !need_parent)
1593                 return 3;
1594
1595         spin_lock(&dentry->d_lock);
1596         inode = dentry->d_parent->d_inode;
1597         data[3] = inode->i_ino;
1598         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1599         *lenp = 5;
1600         if (maxlen >= 6) {
1601                 data[5] = inode->i_generation;
1602                 *lenp = 6;
1603         }
1604         spin_unlock(&dentry->d_lock);
1605         return *lenp;
1606 }
1607
1608 /* looks for stat data, then copies fields to it, marks the buffer
1609    containing stat data as dirty */
1610 /* reiserfs inodes are never really dirty, since the dirty inode call
1611 ** always logs them.  This call allows the VFS inode marking routines
1612 ** to properly mark inodes for datasync and such, but only actually
1613 ** does something when called for a synchronous update.
1614 */
1615 int reiserfs_write_inode(struct inode *inode, int do_sync)
1616 {
1617         struct reiserfs_transaction_handle th;
1618         int jbegin_count = 1;
1619
1620         if (inode->i_sb->s_flags & MS_RDONLY)
1621                 return -EROFS;
1622         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1623          ** these cases are just when the system needs ram, not when the
1624          ** inode needs to reach disk for safety, and they can safely be
1625          ** ignored because the altered inode has already been logged.
1626          */
1627         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1628                 reiserfs_write_lock(inode->i_sb);
1629                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1630                         reiserfs_update_sd(&th, inode);
1631                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1632                 }
1633                 reiserfs_write_unlock(inode->i_sb);
1634         }
1635         return 0;
1636 }
1637
1638 /* stat data of new object is inserted already, this inserts the item
1639    containing "." and ".." entries */
1640 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1641                                   struct inode *inode,
1642                                   struct item_head *ih, struct treepath *path,
1643                                   struct inode *dir)
1644 {
1645         struct super_block *sb = th->t_super;
1646         char empty_dir[EMPTY_DIR_SIZE];
1647         char *body = empty_dir;
1648         struct cpu_key key;
1649         int retval;
1650
1651         BUG_ON(!th->t_trans_id);
1652
1653         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1654                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1655                       TYPE_DIRENTRY, 3 /*key length */ );
1656
1657         /* compose item head for new item. Directories consist of items of
1658            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1659            is done by reiserfs_new_inode */
1660         if (old_format_only(sb)) {
1661                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1662                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1663
1664                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1665                                        ih->ih_key.k_objectid,
1666                                        INODE_PKEY(dir)->k_dir_id,
1667                                        INODE_PKEY(dir)->k_objectid);
1668         } else {
1669                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1670                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1671
1672                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1673                                     ih->ih_key.k_objectid,
1674                                     INODE_PKEY(dir)->k_dir_id,
1675                                     INODE_PKEY(dir)->k_objectid);
1676         }
1677
1678         /* look for place in the tree for new item */
1679         retval = search_item(sb, &key, path);
1680         if (retval == IO_ERROR) {
1681                 reiserfs_warning(sb, "vs-13080",
1682                                  "i/o failure occurred creating new directory");
1683                 return -EIO;
1684         }
1685         if (retval == ITEM_FOUND) {
1686                 pathrelse(path);
1687                 reiserfs_warning(sb, "vs-13070",
1688                                  "object with this key exists (%k)",
1689                                  &(ih->ih_key));
1690                 return -EEXIST;
1691         }
1692
1693         /* insert item, that is empty directory item */
1694         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1695 }
1696
1697 /* stat data of object has been inserted, this inserts the item
1698    containing the body of symlink */
1699 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1700                                 struct item_head *ih,
1701                                 struct treepath *path, const char *symname,
1702                                 int item_len)
1703 {
1704         struct super_block *sb = th->t_super;
1705         struct cpu_key key;
1706         int retval;
1707
1708         BUG_ON(!th->t_trans_id);
1709
1710         _make_cpu_key(&key, KEY_FORMAT_3_5,
1711                       le32_to_cpu(ih->ih_key.k_dir_id),
1712                       le32_to_cpu(ih->ih_key.k_objectid),
1713                       1, TYPE_DIRECT, 3 /*key length */ );
1714
1715         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1716                           0 /*free_space */ );
1717
1718         /* look for place in the tree for new item */
1719         retval = search_item(sb, &key, path);
1720         if (retval == IO_ERROR) {
1721                 reiserfs_warning(sb, "vs-13080",
1722                                  "i/o failure occurred creating new symlink");
1723                 return -EIO;
1724         }
1725         if (retval == ITEM_FOUND) {
1726                 pathrelse(path);
1727                 reiserfs_warning(sb, "vs-13080",
1728                                  "object with this key exists (%k)",
1729                                  &(ih->ih_key));
1730                 return -EEXIST;
1731         }
1732
1733         /* insert item, that is body of symlink */
1734         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1735 }
1736
1737 /* inserts the stat data into the tree, and then calls
1738    reiserfs_new_directory (to insert ".", ".." item if new object is
1739    directory) or reiserfs_new_symlink (to insert symlink body if new
1740    object is symlink) or nothing (if new object is regular file)
1741
1742    NOTE! uid and gid must already be set in the inode.  If we return
1743    non-zero due to an error, we have to drop the quota previously allocated
1744    for the fresh inode.  This can only be done outside a transaction, so
1745    if we return non-zero, we also end the transaction.  */
1746 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1747                        struct inode *dir, int mode, const char *symname,
1748                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1749                           strlen (symname) for symlinks) */
1750                        loff_t i_size, struct dentry *dentry,
1751                        struct inode *inode)
1752 {
1753         struct super_block *sb;
1754         struct reiserfs_iget_args args;
1755         INITIALIZE_PATH(path_to_key);
1756         struct cpu_key key;
1757         struct item_head ih;
1758         struct stat_data sd;
1759         int retval;
1760         int err;
1761
1762         BUG_ON(!th->t_trans_id);
1763
1764         if (DQUOT_ALLOC_INODE(inode)) {
1765                 err = -EDQUOT;
1766                 goto out_end_trans;
1767         }
1768         if (!dir->i_nlink) {
1769                 err = -EPERM;
1770                 goto out_bad_inode;
1771         }
1772
1773         sb = dir->i_sb;
1774
1775         /* item head of new item */
1776         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1777         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1778         if (!ih.ih_key.k_objectid) {
1779                 err = -ENOMEM;
1780                 goto out_bad_inode;
1781         }
1782         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1783         if (old_format_only(sb))
1784                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1785                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1786         else
1787                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1788                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1789         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1790         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1791         if (insert_inode_locked4(inode, args.objectid,
1792                              reiserfs_find_actor, &args) < 0) {
1793                 err = -EINVAL;
1794                 goto out_bad_inode;
1795         }
1796         if (old_format_only(sb))
1797                 /* not a perfect generation count, as object ids can be reused, but
1798                  ** this is as good as reiserfs can do right now.
1799                  ** note that the private part of inode isn't filled in yet, we have
1800                  ** to use the directory.
1801                  */
1802                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1803         else
1804 #if defined( USE_INODE_GENERATION_COUNTER )
1805                 inode->i_generation =
1806                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1807 #else
1808                 inode->i_generation = ++event;
1809 #endif
1810
1811         /* fill stat data */
1812         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1813
1814         /* uid and gid must already be set by the caller for quota init */
1815
1816         /* symlink cannot be immutable or append only, right? */
1817         if (S_ISLNK(inode->i_mode))
1818                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1819
1820         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1821         inode->i_size = i_size;
1822         inode->i_blocks = 0;
1823         inode->i_bytes = 0;
1824         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1825             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1826
1827         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1828         REISERFS_I(inode)->i_flags = 0;
1829         REISERFS_I(inode)->i_prealloc_block = 0;
1830         REISERFS_I(inode)->i_prealloc_count = 0;
1831         REISERFS_I(inode)->i_trans_id = 0;
1832         REISERFS_I(inode)->i_jl = NULL;
1833         REISERFS_I(inode)->i_attrs =
1834             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1835         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1836         mutex_init(&(REISERFS_I(inode)->i_mmap));
1837         reiserfs_init_acl_access(inode);
1838         reiserfs_init_acl_default(inode);
1839         reiserfs_init_xattr_rwsem(inode);
1840
1841         /* key to search for correct place for new stat data */
1842         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1843                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1844                       TYPE_STAT_DATA, 3 /*key length */ );
1845
1846         /* find proper place for inserting of stat data */
1847         retval = search_item(sb, &key, &path_to_key);
1848         if (retval == IO_ERROR) {
1849                 err = -EIO;
1850                 goto out_bad_inode;
1851         }
1852         if (retval == ITEM_FOUND) {
1853                 pathrelse(&path_to_key);
1854                 err = -EEXIST;
1855                 goto out_bad_inode;
1856         }
1857         if (old_format_only(sb)) {
1858                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1859                         pathrelse(&path_to_key);
1860                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1861                         err = -EINVAL;
1862                         goto out_bad_inode;
1863                 }
1864                 inode2sd_v1(&sd, inode, inode->i_size);
1865         } else {
1866                 inode2sd(&sd, inode, inode->i_size);
1867         }
1868         // store in in-core inode the key of stat data and version all
1869         // object items will have (directory items will have old offset
1870         // format, other new objects will consist of new items)
1871         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1872                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1873         else
1874                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1875         if (old_format_only(sb))
1876                 set_inode_sd_version(inode, STAT_DATA_V1);
1877         else
1878                 set_inode_sd_version(inode, STAT_DATA_V2);
1879
1880         /* insert the stat data into the tree */
1881 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1882         if (REISERFS_I(dir)->new_packing_locality)
1883                 th->displace_new_blocks = 1;
1884 #endif
1885         retval =
1886             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1887                                  (char *)(&sd));
1888         if (retval) {
1889                 err = retval;
1890                 reiserfs_check_path(&path_to_key);
1891                 goto out_bad_inode;
1892         }
1893 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1894         if (!th->displace_new_blocks)
1895                 REISERFS_I(dir)->new_packing_locality = 0;
1896 #endif
1897         if (S_ISDIR(mode)) {
1898                 /* insert item with "." and ".." */
1899                 retval =
1900                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1901         }
1902
1903         if (S_ISLNK(mode)) {
1904                 /* insert body of symlink */
1905                 if (!old_format_only(sb))
1906                         i_size = ROUND_UP(i_size);
1907                 retval =
1908                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1909                                          i_size);
1910         }
1911         if (retval) {
1912                 err = retval;
1913                 reiserfs_check_path(&path_to_key);
1914                 journal_end(th, th->t_super, th->t_blocks_allocated);
1915                 goto out_inserted_sd;
1916         }
1917
1918         /* XXX CHECK THIS */
1919         if (reiserfs_posixacl(inode->i_sb)) {
1920                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1921                 if (retval) {
1922                         err = retval;
1923                         reiserfs_check_path(&path_to_key);
1924                         journal_end(th, th->t_super, th->t_blocks_allocated);
1925                         goto out_inserted_sd;
1926                 }
1927         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1928                 reiserfs_warning(inode->i_sb, "jdm-13090",
1929                                  "ACLs aren't enabled in the fs, "
1930                                  "but vfs thinks they are!");
1931         } else if (is_reiserfs_priv_object(dir)) {
1932                 reiserfs_mark_inode_private(inode);
1933         }
1934
1935         reiserfs_update_sd(th, inode);
1936         reiserfs_check_path(&path_to_key);
1937
1938         return 0;
1939
1940 /* it looks like you can easily compress these two goto targets into
1941  * one.  Keeping it like this doesn't actually hurt anything, and they
1942  * are place holders for what the quota code actually needs.
1943  */
1944       out_bad_inode:
1945         /* Invalidate the object, nothing was inserted yet */
1946         INODE_PKEY(inode)->k_objectid = 0;
1947
1948         /* Quota change must be inside a transaction for journaling */
1949         DQUOT_FREE_INODE(inode);
1950
1951       out_end_trans:
1952         journal_end(th, th->t_super, th->t_blocks_allocated);
1953         /* Drop can be outside and it needs more credits so it's better to have it outside */
1954         DQUOT_DROP(inode);
1955         inode->i_flags |= S_NOQUOTA;
1956         make_bad_inode(inode);
1957
1958       out_inserted_sd:
1959         inode->i_nlink = 0;
1960         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1961         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1962
1963         /* If we were inheriting an ACL, we need to release the lock so that
1964          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1965          * code really needs to be reworked, but this will take care of it
1966          * for now. -jeffm */
1967 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1968         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1969                 reiserfs_write_unlock_xattrs(dir->i_sb);
1970                 iput(inode);
1971                 reiserfs_write_lock_xattrs(dir->i_sb);
1972         } else
1973 #endif
1974                 iput(inode);
1975         return err;
1976 }
1977
1978 /*
1979 ** finds the tail page in the page cache,
1980 ** reads the last block in.
1981 **
1982 ** On success, page_result is set to a locked, pinned page, and bh_result
1983 ** is set to an up to date buffer for the last block in the file.  returns 0.
1984 **
1985 ** tail conversion is not done, so bh_result might not be valid for writing
1986 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1987 ** trying to write the block.
1988 **
1989 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1990 */
1991 static int grab_tail_page(struct inode *p_s_inode,
1992                           struct page **page_result,
1993                           struct buffer_head **bh_result)
1994 {
1995
1996         /* we want the page with the last byte in the file,
1997          ** not the page that will hold the next byte for appending
1998          */
1999         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2000         unsigned long pos = 0;
2001         unsigned long start = 0;
2002         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2003         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2004         struct buffer_head *bh;
2005         struct buffer_head *head;
2006         struct page *page;
2007         int error;
2008
2009         /* we know that we are only called with inode->i_size > 0.
2010          ** we also know that a file tail can never be as big as a block
2011          ** If i_size % blocksize == 0, our file is currently block aligned
2012          ** and it won't need converting or zeroing after a truncate.
2013          */
2014         if ((offset & (blocksize - 1)) == 0) {
2015                 return -ENOENT;
2016         }
2017         page = grab_cache_page(p_s_inode->i_mapping, index);
2018         error = -ENOMEM;
2019         if (!page) {
2020                 goto out;
2021         }
2022         /* start within the page of the last block in the file */
2023         start = (offset / blocksize) * blocksize;
2024
2025         error = block_prepare_write(page, start, offset,
2026                                     reiserfs_get_block_create_0);
2027         if (error)
2028                 goto unlock;
2029
2030         head = page_buffers(page);
2031         bh = head;
2032         do {
2033                 if (pos >= start) {
2034                         break;
2035                 }
2036                 bh = bh->b_this_page;
2037                 pos += blocksize;
2038         } while (bh != head);
2039
2040         if (!buffer_uptodate(bh)) {
2041                 /* note, this should never happen, prepare_write should
2042                  ** be taking care of this for us.  If the buffer isn't up to date,
2043                  ** I've screwed up the code to find the buffer, or the code to
2044                  ** call prepare_write
2045                  */
2046                 reiserfs_warning(p_s_inode->i_sb, "clm-6000",
2047                                  "error reading block %lu on dev %s",
2048                                  bh->b_blocknr,
2049                                  reiserfs_bdevname(p_s_inode->i_sb));
2050                 error = -EIO;
2051                 goto unlock;
2052         }
2053         *bh_result = bh;
2054         *page_result = page;
2055
2056       out:
2057         return error;
2058
2059       unlock:
2060         unlock_page(page);
2061         page_cache_release(page);
2062         return error;
2063 }
2064
2065 /*
2066 ** vfs version of truncate file.  Must NOT be called with
2067 ** a transaction already started.
2068 **
2069 ** some code taken from block_truncate_page
2070 */
2071 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2072 {
2073         struct reiserfs_transaction_handle th;
2074         /* we want the offset for the first byte after the end of the file */
2075         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2076         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2077         unsigned length;
2078         struct page *page = NULL;
2079         int error;
2080         struct buffer_head *bh = NULL;
2081         int err2;
2082
2083         reiserfs_write_lock(p_s_inode->i_sb);
2084
2085         if (p_s_inode->i_size > 0) {
2086                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2087                         // -ENOENT means we truncated past the end of the file,
2088                         // and get_block_create_0 could not find a block to read in,
2089                         // which is ok.
2090                         if (error != -ENOENT)
2091                                 reiserfs_warning(p_s_inode->i_sb, "clm-6001",
2092                                                  "grab_tail_page failed %d",
2093                                                  error);
2094                         page = NULL;
2095                         bh = NULL;
2096                 }
2097         }
2098
2099         /* so, if page != NULL, we have a buffer head for the offset at
2100          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2101          ** then we have an unformatted node.  Otherwise, we have a direct item,
2102          ** and no zeroing is required on disk.  We zero after the truncate,
2103          ** because the truncate might pack the item anyway
2104          ** (it will unmap bh if it packs).
2105          */
2106         /* it is enough to reserve space in transaction for 2 balancings:
2107            one for "save" link adding and another for the first
2108            cut_from_item. 1 is for update_sd */
2109         error = journal_begin(&th, p_s_inode->i_sb,
2110                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2111         if (error)
2112                 goto out;
2113         reiserfs_update_inode_transaction(p_s_inode);
2114         if (update_timestamps)
2115                 /* we are doing real truncate: if the system crashes before the last
2116                    transaction of truncating gets committed - on reboot the file
2117                    either appears truncated properly or not truncated at all */
2118                 add_save_link(&th, p_s_inode, 1);
2119         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2120         error =
2121             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2122         if (error)
2123                 goto out;
2124
2125         /* check reiserfs_do_truncate after ending the transaction */
2126         if (err2) {
2127                 error = err2;
2128                 goto out;
2129         }
2130
2131         if (update_timestamps) {
2132                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2133                 if (error)
2134                         goto out;
2135         }
2136
2137         if (page) {
2138                 length = offset & (blocksize - 1);
2139                 /* if we are not on a block boundary */
2140                 if (length) {
2141                         length = blocksize - length;
2142                         zero_user(page, offset, length);
2143                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2144                                 mark_buffer_dirty(bh);
2145                         }
2146                 }
2147                 unlock_page(page);
2148                 page_cache_release(page);
2149         }
2150
2151         reiserfs_write_unlock(p_s_inode->i_sb);
2152         return 0;
2153       out:
2154         if (page) {
2155                 unlock_page(page);
2156                 page_cache_release(page);
2157         }
2158         reiserfs_write_unlock(p_s_inode->i_sb);
2159         return error;
2160 }
2161
2162 static int map_block_for_writepage(struct inode *inode,
2163                                    struct buffer_head *bh_result,
2164                                    unsigned long block)
2165 {
2166         struct reiserfs_transaction_handle th;
2167         int fs_gen;
2168         struct item_head tmp_ih;
2169         struct item_head *ih;
2170         struct buffer_head *bh;
2171         __le32 *item;
2172         struct cpu_key key;
2173         INITIALIZE_PATH(path);
2174         int pos_in_item;
2175         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2176         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2177         int retval;
2178         int use_get_block = 0;
2179         int bytes_copied = 0;
2180         int copy_size;
2181         int trans_running = 0;
2182
2183         /* catch places below that try to log something without starting a trans */
2184         th.t_trans_id = 0;
2185
2186         if (!buffer_uptodate(bh_result)) {
2187                 return -EIO;
2188         }
2189
2190         kmap(bh_result->b_page);
2191       start_over:
2192         reiserfs_write_lock(inode->i_sb);
2193         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2194
2195       research:
2196         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2197         if (retval != POSITION_FOUND) {
2198                 use_get_block = 1;
2199                 goto out;
2200         }
2201
2202         bh = get_last_bh(&path);
2203         ih = get_ih(&path);
2204         item = get_item(&path);
2205         pos_in_item = path.pos_in_item;
2206
2207         /* we've found an unformatted node */
2208         if (indirect_item_found(retval, ih)) {
2209                 if (bytes_copied > 0) {
2210                         reiserfs_warning(inode->i_sb, "clm-6002",
2211                                          "bytes_copied %d", bytes_copied);
2212                 }
2213                 if (!get_block_num(item, pos_in_item)) {
2214                         /* crap, we are writing to a hole */
2215                         use_get_block = 1;
2216                         goto out;
2217                 }
2218                 set_block_dev_mapped(bh_result,
2219                                      get_block_num(item, pos_in_item), inode);
2220         } else if (is_direct_le_ih(ih)) {
2221                 char *p;
2222                 p = page_address(bh_result->b_page);
2223                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2224                 copy_size = ih_item_len(ih) - pos_in_item;
2225
2226                 fs_gen = get_generation(inode->i_sb);
2227                 copy_item_head(&tmp_ih, ih);
2228
2229                 if (!trans_running) {
2230                         /* vs-3050 is gone, no need to drop the path */
2231                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2232                         if (retval)
2233                                 goto out;
2234                         reiserfs_update_inode_transaction(inode);
2235                         trans_running = 1;
2236                         if (fs_changed(fs_gen, inode->i_sb)
2237                             && item_moved(&tmp_ih, &path)) {
2238                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2239                                                                  bh);
2240                                 goto research;
2241                         }
2242                 }
2243
2244                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2245
2246                 if (fs_changed(fs_gen, inode->i_sb)
2247                     && item_moved(&tmp_ih, &path)) {
2248                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2249                         goto research;
2250                 }
2251
2252                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2253                        copy_size);
2254
2255                 journal_mark_dirty(&th, inode->i_sb, bh);
2256                 bytes_copied += copy_size;
2257                 set_block_dev_mapped(bh_result, 0, inode);
2258
2259                 /* are there still bytes left? */
2260                 if (bytes_copied < bh_result->b_size &&
2261                     (byte_offset + bytes_copied) < inode->i_size) {
2262                         set_cpu_key_k_offset(&key,
2263                                              cpu_key_k_offset(&key) +
2264                                              copy_size);
2265                         goto research;
2266                 }
2267         } else {
2268                 reiserfs_warning(inode->i_sb, "clm-6003",
2269                                  "bad item inode %lu", inode->i_ino);
2270                 retval = -EIO;
2271                 goto out;
2272         }
2273         retval = 0;
2274
2275       out:
2276         pathrelse(&path);
2277         if (trans_running) {
2278                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2279                 if (err)
2280                         retval = err;
2281                 trans_running = 0;
2282         }
2283         reiserfs_write_unlock(inode->i_sb);
2284
2285         /* this is where we fill in holes in the file. */
2286         if (use_get_block) {
2287                 retval = reiserfs_get_block(inode, block, bh_result,
2288                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2289                                             | GET_BLOCK_NO_DANGLE);
2290                 if (!retval) {
2291                         if (!buffer_mapped(bh_result)
2292                             || bh_result->b_blocknr == 0) {
2293                                 /* get_block failed to find a mapped unformatted node. */
2294                                 use_get_block = 0;
2295                                 goto start_over;
2296                         }
2297                 }
2298         }
2299         kunmap(bh_result->b_page);
2300
2301         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2302                 /* we've copied data from the page into the direct item, so the
2303                  * buffer in the page is now clean, mark it to reflect that.
2304                  */
2305                 lock_buffer(bh_result);
2306                 clear_buffer_dirty(bh_result);
2307                 unlock_buffer(bh_result);
2308         }
2309         return retval;
2310 }
2311
2312 /*
2313  * mason@suse.com: updated in 2.5.54 to follow the same general io
2314  * start/recovery path as __block_write_full_page, along with special
2315  * code to handle reiserfs tails.
2316  */
2317 static int reiserfs_write_full_page(struct page *page,
2318                                     struct writeback_control *wbc)
2319 {
2320         struct inode *inode = page->mapping->host;
2321         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2322         int error = 0;
2323         unsigned long block;
2324         sector_t last_block;
2325         struct buffer_head *head, *bh;
2326         int partial = 0;
2327         int nr = 0;
2328         int checked = PageChecked(page);
2329         struct reiserfs_transaction_handle th;
2330         struct super_block *s = inode->i_sb;
2331         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2332         th.t_trans_id = 0;
2333
2334         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2335         if (checked && (current->flags & PF_MEMALLOC)) {
2336                 redirty_page_for_writepage(wbc, page);
2337                 unlock_page(page);
2338                 return 0;
2339         }
2340
2341         /* The page dirty bit is cleared before writepage is called, which
2342          * means we have to tell create_empty_buffers to make dirty buffers
2343          * The page really should be up to date at this point, so tossing
2344          * in the BH_Uptodate is just a sanity check.
2345          */
2346         if (!page_has_buffers(page)) {
2347                 create_empty_buffers(page, s->s_blocksize,
2348                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2349         }
2350         head = page_buffers(page);
2351
2352         /* last page in the file, zero out any contents past the
2353          ** last byte in the file
2354          */
2355         if (page->index >= end_index) {
2356                 unsigned last_offset;
2357
2358                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2359                 /* no file contents in this page */
2360                 if (page->index >= end_index + 1 || !last_offset) {
2361                         unlock_page(page);
2362                         return 0;
2363                 }
2364                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2365         }
2366         bh = head;
2367         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2368         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2369         /* first map all the buffers, logging any direct items we find */
2370         do {
2371                 if (block > last_block) {
2372                         /*
2373                          * This can happen when the block size is less than
2374                          * the page size.  The corresponding bytes in the page
2375                          * were zero filled above
2376                          */
2377                         clear_buffer_dirty(bh);
2378                         set_buffer_uptodate(bh);
2379                 } else if ((checked || buffer_dirty(bh)) &&
2380                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2381                                                        && bh->b_blocknr ==
2382                                                        0))) {
2383                         /* not mapped yet, or it points to a direct item, search
2384                          * the btree for the mapping info, and log any direct
2385                          * items found
2386                          */
2387                         if ((error = map_block_for_writepage(inode, bh, block))) {
2388                                 goto fail;
2389                         }
2390                 }
2391                 bh = bh->b_this_page;
2392                 block++;
2393         } while (bh != head);
2394
2395         /*
2396          * we start the transaction after map_block_for_writepage,
2397          * because it can create holes in the file (an unbounded operation).
2398          * starting it here, we can make a reliable estimate for how many
2399          * blocks we're going to log
2400          */
2401         if (checked) {
2402                 ClearPageChecked(page);
2403                 reiserfs_write_lock(s);
2404                 error = journal_begin(&th, s, bh_per_page + 1);
2405                 if (error) {
2406                         reiserfs_write_unlock(s);
2407                         goto fail;
2408                 }
2409                 reiserfs_update_inode_transaction(inode);
2410         }
2411         /* now go through and lock any dirty buffers on the page */
2412         do {
2413                 get_bh(bh);
2414                 if (!buffer_mapped(bh))
2415                         continue;
2416                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2417                         continue;
2418
2419                 if (checked) {
2420                         reiserfs_prepare_for_journal(s, bh, 1);
2421                         journal_mark_dirty(&th, s, bh);
2422                         continue;
2423                 }
2424                 /* from this point on, we know the buffer is mapped to a
2425                  * real block and not a direct item
2426                  */
2427                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2428                         lock_buffer(bh);
2429                 } else {
2430                         if (!trylock_buffer(bh)) {
2431                                 redirty_page_for_writepage(wbc, page);
2432                                 continue;
2433                         }
2434                 }
2435                 if (test_clear_buffer_dirty(bh)) {
2436                         mark_buffer_async_write(bh);
2437                 } else {
2438                         unlock_buffer(bh);
2439                 }
2440         } while ((bh = bh->b_this_page) != head);
2441
2442         if (checked) {
2443                 error = journal_end(&th, s, bh_per_page + 1);
2444                 reiserfs_write_unlock(s);
2445                 if (error)
2446                         goto fail;
2447         }
2448         BUG_ON(PageWriteback(page));
2449         set_page_writeback(page);
2450         unlock_page(page);
2451
2452         /*
2453          * since any buffer might be the only dirty buffer on the page,
2454          * the first submit_bh can bring the page out of writeback.
2455          * be careful with the buffers.
2456          */
2457         do {
2458                 struct buffer_head *next = bh->b_this_page;
2459                 if (buffer_async_write(bh)) {
2460                         submit_bh(WRITE, bh);
2461                         nr++;
2462                 }
2463                 put_bh(bh);
2464                 bh = next;
2465         } while (bh != head);
2466
2467         error = 0;
2468       done:
2469         if (nr == 0) {
2470                 /*
2471                  * if this page only had a direct item, it is very possible for
2472                  * no io to be required without there being an error.  Or,
2473                  * someone else could have locked them and sent them down the
2474                  * pipe without locking the page
2475                  */
2476                 bh = head;
2477                 do {
2478                         if (!buffer_uptodate(bh)) {
2479                                 partial = 1;
2480                                 break;
2481                         }
2482                         bh = bh->b_this_page;
2483                 } while (bh != head);
2484                 if (!partial)
2485                         SetPageUptodate(page);
2486                 end_page_writeback(page);
2487         }
2488         return error;
2489
2490       fail:
2491         /* catches various errors, we need to make sure any valid dirty blocks
2492          * get to the media.  The page is currently locked and not marked for
2493          * writeback
2494          */
2495         ClearPageUptodate(page);
2496         bh = head;
2497         do {
2498                 get_bh(bh);
2499                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2500                         lock_buffer(bh);
2501                         mark_buffer_async_write(bh);
2502                 } else {
2503                         /*
2504                          * clear any dirty bits that might have come from getting
2505                          * attached to a dirty page
2506                          */
2507                         clear_buffer_dirty(bh);
2508                 }
2509                 bh = bh->b_this_page;
2510         } while (bh != head);
2511         SetPageError(page);
2512         BUG_ON(PageWriteback(page));
2513         set_page_writeback(page);
2514         unlock_page(page);
2515         do {
2516                 struct buffer_head *next = bh->b_this_page;
2517                 if (buffer_async_write(bh)) {
2518                         clear_buffer_dirty(bh);
2519                         submit_bh(WRITE, bh);
2520                         nr++;
2521                 }
2522                 put_bh(bh);
2523                 bh = next;
2524         } while (bh != head);
2525         goto done;
2526 }
2527
2528 static int reiserfs_readpage(struct file *f, struct page *page)
2529 {
2530         return block_read_full_page(page, reiserfs_get_block);
2531 }
2532
2533 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2534 {
2535         struct inode *inode = page->mapping->host;
2536         reiserfs_wait_on_write_block(inode->i_sb);
2537         return reiserfs_write_full_page(page, wbc);
2538 }
2539
2540 static int reiserfs_write_begin(struct file *file,
2541                                 struct address_space *mapping,
2542                                 loff_t pos, unsigned len, unsigned flags,
2543                                 struct page **pagep, void **fsdata)
2544 {
2545         struct inode *inode;
2546         struct page *page;
2547         pgoff_t index;
2548         int ret;
2549         int old_ref = 0;
2550
2551         inode = mapping->host;
2552         *fsdata = 0;
2553         if (flags & AOP_FLAG_CONT_EXPAND &&
2554             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2555                 pos ++;
2556                 *fsdata = (void *)(unsigned long)flags;
2557         }
2558
2559         index = pos >> PAGE_CACHE_SHIFT;
2560         page = grab_cache_page_write_begin(mapping, index, flags);
2561         if (!page)
2562                 return -ENOMEM;
2563         *pagep = page;
2564
2565         reiserfs_wait_on_write_block(inode->i_sb);
2566         fix_tail_page_for_writing(page);
2567         if (reiserfs_transaction_running(inode->i_sb)) {
2568                 struct reiserfs_transaction_handle *th;
2569                 th = (struct reiserfs_transaction_handle *)current->
2570                     journal_info;
2571                 BUG_ON(!th->t_refcount);
2572                 BUG_ON(!th->t_trans_id);
2573                 old_ref = th->t_refcount;
2574                 th->t_refcount++;
2575         }
2576         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2577                                 reiserfs_get_block);
2578         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2579                 struct reiserfs_transaction_handle *th = current->journal_info;
2580                 /* this gets a little ugly.  If reiserfs_get_block returned an
2581                  * error and left a transacstion running, we've got to close it,
2582                  * and we've got to free handle if it was a persistent transaction.
2583                  *
2584                  * But, if we had nested into an existing transaction, we need
2585                  * to just drop the ref count on the handle.
2586                  *
2587                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2588                  * and it was a persistent trans.  Otherwise, it was nested above.
2589                  */
2590                 if (th->t_refcount > old_ref) {
2591                         if (old_ref)
2592                                 th->t_refcount--;
2593                         else {
2594                                 int err;
2595                                 reiserfs_write_lock(inode->i_sb);
2596                                 err = reiserfs_end_persistent_transaction(th);
2597                                 reiserfs_write_unlock(inode->i_sb);
2598                                 if (err)
2599                                         ret = err;
2600                         }
2601                 }
2602         }
2603         if (ret) {
2604                 unlock_page(page);
2605                 page_cache_release(page);
2606         }
2607         return ret;
2608 }
2609
2610 int reiserfs_prepare_write(struct file *f, struct page *page,
2611                            unsigned from, unsigned to)
2612 {
2613         struct inode *inode = page->mapping->host;
2614         int ret;
2615         int old_ref = 0;
2616
2617         reiserfs_wait_on_write_block(inode->i_sb);
2618         fix_tail_page_for_writing(page);
2619         if (reiserfs_transaction_running(inode->i_sb)) {
2620                 struct reiserfs_transaction_handle *th;
2621                 th = (struct reiserfs_transaction_handle *)current->
2622                     journal_info;
2623                 BUG_ON(!th->t_refcount);
2624                 BUG_ON(!th->t_trans_id);
2625                 old_ref = th->t_refcount;
2626                 th->t_refcount++;
2627         }
2628
2629         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2630         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2631                 struct reiserfs_transaction_handle *th = current->journal_info;
2632                 /* this gets a little ugly.  If reiserfs_get_block returned an
2633                  * error and left a transacstion running, we've got to close it,
2634                  * and we've got to free handle if it was a persistent transaction.
2635                  *
2636                  * But, if we had nested into an existing transaction, we need
2637                  * to just drop the ref count on the handle.
2638                  *
2639                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2640                  * and it was a persistent trans.  Otherwise, it was nested above.
2641                  */
2642                 if (th->t_refcount > old_ref) {
2643                         if (old_ref)
2644                                 th->t_refcount--;
2645                         else {
2646                                 int err;
2647                                 reiserfs_write_lock(inode->i_sb);
2648                                 err = reiserfs_end_persistent_transaction(th);
2649                                 reiserfs_write_unlock(inode->i_sb);
2650                                 if (err)
2651                                         ret = err;
2652                         }
2653                 }
2654         }
2655         return ret;
2656
2657 }
2658
2659 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2660 {
2661         return generic_block_bmap(as, block, reiserfs_bmap);
2662 }
2663
2664 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2665                               loff_t pos, unsigned len, unsigned copied,
2666                               struct page *page, void *fsdata)
2667 {
2668         struct inode *inode = page->mapping->host;
2669         int ret = 0;
2670         int update_sd = 0;
2671         struct reiserfs_transaction_handle *th;
2672         unsigned start;
2673
2674         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2675                 pos ++;
2676
2677         reiserfs_wait_on_write_block(inode->i_sb);
2678         if (reiserfs_transaction_running(inode->i_sb))
2679                 th = current->journal_info;
2680         else
2681                 th = NULL;
2682
2683         start = pos & (PAGE_CACHE_SIZE - 1);
2684         if (unlikely(copied < len)) {
2685                 if (!PageUptodate(page))
2686                         copied = 0;
2687
2688                 page_zero_new_buffers(page, start + copied, start + len);
2689         }
2690         flush_dcache_page(page);
2691
2692         reiserfs_commit_page(inode, page, start, start + copied);
2693
2694         /* generic_commit_write does this for us, but does not update the
2695          ** transaction tracking stuff when the size changes.  So, we have
2696          ** to do the i_size updates here.
2697          */
2698         pos += copied;
2699         if (pos > inode->i_size) {
2700                 struct reiserfs_transaction_handle myth;
2701                 reiserfs_write_lock(inode->i_sb);
2702                 /* If the file have grown beyond the border where it
2703                    can have a tail, unmark it as needing a tail
2704                    packing */
2705                 if ((have_large_tails(inode->i_sb)
2706                      && inode->i_size > i_block_size(inode) * 4)
2707                     || (have_small_tails(inode->i_sb)
2708                         && inode->i_size > i_block_size(inode)))
2709                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2710
2711                 ret = journal_begin(&myth, inode->i_sb, 1);
2712                 if (ret) {
2713                         reiserfs_write_unlock(inode->i_sb);
2714                         goto journal_error;
2715                 }
2716                 reiserfs_update_inode_transaction(inode);
2717                 inode->i_size = pos;
2718                 /*
2719                  * this will just nest into our transaction.  It's important
2720                  * to use mark_inode_dirty so the inode gets pushed around on the
2721                  * dirty lists, and so that O_SYNC works as expected
2722                  */
2723                 mark_inode_dirty(inode);
2724                 reiserfs_update_sd(&myth, inode);
2725                 update_sd = 1;
2726                 ret = journal_end(&myth, inode->i_sb, 1);
2727                 reiserfs_write_unlock(inode->i_sb);
2728                 if (ret)
2729                         goto journal_error;
2730         }
2731         if (th) {
2732                 reiserfs_write_lock(inode->i_sb);
2733                 if (!update_sd)
2734                         mark_inode_dirty(inode);
2735                 ret = reiserfs_end_persistent_transaction(th);
2736                 reiserfs_write_unlock(inode->i_sb);
2737                 if (ret)
2738                         goto out;
2739         }
2740
2741       out:
2742         unlock_page(page);
2743         page_cache_release(page);
2744         return ret == 0 ? copied : ret;
2745
2746       journal_error:
2747         if (th) {
2748                 reiserfs_write_lock(inode->i_sb);
2749                 if (!update_sd)
2750                         reiserfs_update_sd(th, inode);
2751                 ret = reiserfs_end_persistent_transaction(th);
2752                 reiserfs_write_unlock(inode->i_sb);
2753         }
2754
2755         goto out;
2756 }
2757
2758 int reiserfs_commit_write(struct file *f, struct page *page,
2759                           unsigned from, unsigned to)
2760 {
2761         struct inode *inode = page->mapping->host;
2762         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2763         int ret = 0;
2764         int update_sd = 0;
2765         struct reiserfs_transaction_handle *th = NULL;
2766
2767         reiserfs_wait_on_write_block(inode->i_sb);
2768         if (reiserfs_transaction_running(inode->i_sb)) {
2769                 th = current->journal_info;
2770         }
2771         reiserfs_commit_page(inode, page, from, to);
2772
2773         /* generic_commit_write does this for us, but does not update the
2774          ** transaction tracking stuff when the size changes.  So, we have
2775          ** to do the i_size updates here.
2776          */
2777         if (pos > inode->i_size) {
2778                 struct reiserfs_transaction_handle myth;
2779                 reiserfs_write_lock(inode->i_sb);
2780                 /* If the file have grown beyond the border where it
2781                    can have a tail, unmark it as needing a tail
2782                    packing */
2783                 if ((have_large_tails(inode->i_sb)
2784                      && inode->i_size > i_block_size(inode) * 4)
2785                     || (have_small_tails(inode->i_sb)
2786                         && inode->i_size > i_block_size(inode)))
2787                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2788
2789                 ret = journal_begin(&myth, inode->i_sb, 1);
2790                 if (ret) {
2791                         reiserfs_write_unlock(inode->i_sb);
2792                         goto journal_error;
2793                 }
2794                 reiserfs_update_inode_transaction(inode);
2795                 inode->i_size = pos;
2796                 /*
2797                  * this will just nest into our transaction.  It's important
2798                  * to use mark_inode_dirty so the inode gets pushed around on the
2799                  * dirty lists, and so that O_SYNC works as expected
2800                  */
2801                 mark_inode_dirty(inode);
2802                 reiserfs_update_sd(&myth, inode);
2803                 update_sd = 1;
2804                 ret = journal_end(&myth, inode->i_sb, 1);
2805                 reiserfs_write_unlock(inode->i_sb);
2806                 if (ret)
2807                         goto journal_error;
2808         }
2809         if (th) {
2810                 reiserfs_write_lock(inode->i_sb);
2811                 if (!update_sd)
2812                         mark_inode_dirty(inode);
2813                 ret = reiserfs_end_persistent_transaction(th);
2814                 reiserfs_write_unlock(inode->i_sb);
2815                 if (ret)
2816                         goto out;
2817         }
2818
2819       out:
2820         return ret;
2821
2822       journal_error:
2823         if (th) {
2824                 reiserfs_write_lock(inode->i_sb);
2825                 if (!update_sd)
2826                         reiserfs_update_sd(th, inode);
2827                 ret = reiserfs_end_persistent_transaction(th);
2828                 reiserfs_write_unlock(inode->i_sb);
2829         }
2830
2831         return ret;
2832 }
2833
2834 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2835 {
2836         if (reiserfs_attrs(inode->i_sb)) {
2837                 if (sd_attrs & REISERFS_SYNC_FL)
2838                         inode->i_flags |= S_SYNC;
2839                 else
2840                         inode->i_flags &= ~S_SYNC;
2841                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2842                         inode->i_flags |= S_IMMUTABLE;
2843                 else
2844                         inode->i_flags &= ~S_IMMUTABLE;
2845                 if (sd_attrs & REISERFS_APPEND_FL)
2846                         inode->i_flags |= S_APPEND;
2847                 else
2848                         inode->i_flags &= ~S_APPEND;
2849                 if (sd_attrs & REISERFS_NOATIME_FL)
2850                         inode->i_flags |= S_NOATIME;
2851                 else
2852                         inode->i_flags &= ~S_NOATIME;
2853                 if (sd_attrs & REISERFS_NOTAIL_FL)
2854                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2855                 else
2856                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2857         }
2858 }
2859
2860 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2861 {
2862         if (reiserfs_attrs(inode->i_sb)) {
2863                 if (inode->i_flags & S_IMMUTABLE)
2864                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2865                 else
2866                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2867                 if (inode->i_flags & S_SYNC)
2868                         *sd_attrs |= REISERFS_SYNC_FL;
2869                 else
2870                         *sd_attrs &= ~REISERFS_SYNC_FL;
2871                 if (inode->i_flags & S_NOATIME)
2872                         *sd_attrs |= REISERFS_NOATIME_FL;
2873                 else
2874                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2875                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2876                         *sd_attrs |= REISERFS_NOTAIL_FL;
2877                 else
2878                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2879         }
2880 }
2881
2882 /* decide if this buffer needs to stay around for data logging or ordered
2883 ** write purposes
2884 */
2885 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2886 {
2887         int ret = 1;
2888         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2889
2890         lock_buffer(bh);
2891         spin_lock(&j->j_dirty_buffers_lock);
2892         if (!buffer_mapped(bh)) {
2893                 goto free_jh;
2894         }
2895         /* the page is locked, and the only places that log a data buffer
2896          * also lock the page.
2897          */
2898         if (reiserfs_file_data_log(inode)) {
2899                 /*
2900                  * very conservative, leave the buffer pinned if
2901                  * anyone might need it.
2902                  */
2903                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2904                         ret = 0;
2905                 }
2906         } else  if (buffer_dirty(bh)) {
2907                 struct reiserfs_journal_list *jl;
2908                 struct reiserfs_jh *jh = bh->b_private;
2909
2910                 /* why is this safe?
2911                  * reiserfs_setattr updates i_size in the on disk
2912                  * stat data before allowing vmtruncate to be called.
2913                  *
2914                  * If buffer was put onto the ordered list for this
2915                  * transaction, we know for sure either this transaction
2916                  * or an older one already has updated i_size on disk,
2917                  * and this ordered data won't be referenced in the file
2918                  * if we crash.
2919                  *
2920                  * if the buffer was put onto the ordered list for an older
2921                  * transaction, we need to leave it around
2922                  */
2923                 if (jh && (jl = jh->jl)
2924                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2925                         ret = 0;
2926         }
2927       free_jh:
2928         if (ret && bh->b_private) {
2929                 reiserfs_free_jh(bh);
2930         }
2931         spin_unlock(&j->j_dirty_buffers_lock);
2932         unlock_buffer(bh);
2933         return ret;
2934 }
2935
2936 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2937 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2938 {
2939         struct buffer_head *head, *bh, *next;
2940         struct inode *inode = page->mapping->host;
2941         unsigned int curr_off = 0;
2942         int ret = 1;
2943
2944         BUG_ON(!PageLocked(page));
2945
2946         if (offset == 0)
2947                 ClearPageChecked(page);
2948
2949         if (!page_has_buffers(page))
2950                 goto out;
2951
2952         head = page_buffers(page);
2953         bh = head;
2954         do {
2955                 unsigned int next_off = curr_off + bh->b_size;
2956                 next = bh->b_this_page;
2957
2958                 /*
2959                  * is this block fully invalidated?
2960                  */
2961                 if (offset <= curr_off) {
2962                         if (invalidatepage_can_drop(inode, bh))
2963                                 reiserfs_unmap_buffer(bh);
2964                         else
2965                                 ret = 0;
2966                 }
2967                 curr_off = next_off;
2968                 bh = next;
2969         } while (bh != head);
2970
2971         /*
2972          * We release buffers only if the entire page is being invalidated.
2973          * The get_block cached value has been unconditionally invalidated,
2974          * so real IO is not possible anymore.
2975          */
2976         if (!offset && ret) {
2977                 ret = try_to_release_page(page, 0);
2978                 /* maybe should BUG_ON(!ret); - neilb */
2979         }
2980       out:
2981         return;
2982 }
2983
2984 static int reiserfs_set_page_dirty(struct page *page)
2985 {
2986         struct inode *inode = page->mapping->host;
2987         if (reiserfs_file_data_log(inode)) {
2988                 SetPageChecked(page);
2989                 return __set_page_dirty_nobuffers(page);
2990         }
2991         return __set_page_dirty_buffers(page);
2992 }
2993
2994 /*
2995  * Returns 1 if the page's buffers were dropped.  The page is locked.
2996  *
2997  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2998  * in the buffers at page_buffers(page).
2999  *
3000  * even in -o notail mode, we can't be sure an old mount without -o notail
3001  * didn't create files with tails.
3002  */
3003 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3004 {
3005         struct inode *inode = page->mapping->host;
3006         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3007         struct buffer_head *head;
3008         struct buffer_head *bh;
3009         int ret = 1;
3010
3011         WARN_ON(PageChecked(page));
3012         spin_lock(&j->j_dirty_buffers_lock);
3013         head = page_buffers(page);
3014         bh = head;
3015         do {
3016                 if (bh->b_private) {
3017                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3018                                 reiserfs_free_jh(bh);
3019                         } else {
3020                                 ret = 0;
3021                                 break;
3022                         }
3023                 }
3024                 bh = bh->b_this_page;
3025         } while (bh != head);
3026         if (ret)
3027                 ret = try_to_free_buffers(page);
3028         spin_unlock(&j->j_dirty_buffers_lock);
3029         return ret;
3030 }
3031
3032 /* We thank Mingming Cao for helping us understand in great detail what
3033    to do in this section of the code. */
3034 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3035                                   const struct iovec *iov, loff_t offset,
3036                                   unsigned long nr_segs)
3037 {
3038         struct file *file = iocb->ki_filp;
3039         struct inode *inode = file->f_mapping->host;
3040
3041         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3042                                   offset, nr_segs,
3043                                   reiserfs_get_blocks_direct_io, NULL);
3044 }
3045
3046 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3047 {
3048         struct inode *inode = dentry->d_inode;
3049         int error;
3050         unsigned int ia_valid;
3051
3052         /* must be turned off for recursive notify_change calls */
3053         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3054
3055         reiserfs_write_lock(inode->i_sb);
3056         if (attr->ia_valid & ATTR_SIZE) {
3057                 /* version 2 items will be caught by the s_maxbytes check
3058                  ** done for us in vmtruncate
3059                  */
3060                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3061                     attr->ia_size > MAX_NON_LFS) {
3062                         error = -EFBIG;
3063                         goto out;
3064                 }
3065                 /* fill in hole pointers in the expanding truncate case. */
3066                 if (attr->ia_size > inode->i_size) {
3067                         error = generic_cont_expand_simple(inode, attr->ia_size);
3068                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3069                                 int err;
3070                                 struct reiserfs_transaction_handle th;
3071                                 /* we're changing at most 2 bitmaps, inode + super */
3072                                 err = journal_begin(&th, inode->i_sb, 4);
3073                                 if (!err) {
3074                                         reiserfs_discard_prealloc(&th, inode);
3075                                         err = journal_end(&th, inode->i_sb, 4);
3076                                 }
3077                                 if (err)
3078                                         error = err;
3079                         }
3080                         if (error)
3081                                 goto out;
3082                         /*
3083                          * file size is changed, ctime and mtime are
3084                          * to be updated
3085                          */
3086                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3087                 }
3088         }
3089
3090         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3091              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3092             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3093                 /* stat data of format v3.5 has 16 bit uid and gid */
3094                 error = -EINVAL;
3095                 goto out;
3096         }
3097
3098         error = inode_change_ok(inode, attr);
3099         if (!error) {
3100                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3101                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3102                         error = reiserfs_chown_xattrs(inode, attr);
3103
3104                         if (!error) {
3105                                 struct reiserfs_transaction_handle th;
3106                                 int jbegin_count =
3107                                     2 *
3108                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3109                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3110                                     2;
3111
3112                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3113                                 error =
3114                                     journal_begin(&th, inode->i_sb,
3115                                                   jbegin_count);
3116                                 if (error)
3117                                         goto out;
3118                                 error =
3119                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3120                                 if (error) {
3121                                         journal_end(&th, inode->i_sb,
3122                                                     jbegin_count);
3123                                         goto out;
3124                                 }
3125                                 /* Update corresponding info in inode so that everything is in
3126                                  * one transaction */
3127                                 if (attr->ia_valid & ATTR_UID)
3128                                         inode->i_uid = attr->ia_uid;
3129                                 if (attr->ia_valid & ATTR_GID)
3130                                         inode->i_gid = attr->ia_gid;
3131                                 mark_inode_dirty(inode);
3132                                 error =
3133                                     journal_end(&th, inode->i_sb, jbegin_count);
3134                         }
3135                 }
3136                 if (!error)
3137                         error = inode_setattr(inode, attr);
3138         }
3139
3140         if (!error && reiserfs_posixacl(inode->i_sb)) {
3141                 if (attr->ia_valid & ATTR_MODE)
3142                         error = reiserfs_acl_chmod(inode);
3143         }
3144
3145       out:
3146         reiserfs_write_unlock(inode->i_sb);
3147         return error;
3148 }
3149
3150 const struct address_space_operations reiserfs_address_space_operations = {
3151         .writepage = reiserfs_writepage,
3152         .readpage = reiserfs_readpage,
3153         .readpages = reiserfs_readpages,
3154         .releasepage = reiserfs_releasepage,
3155         .invalidatepage = reiserfs_invalidatepage,
3156         .sync_page = block_sync_page,
3157         .write_begin = reiserfs_write_begin,
3158         .write_end = reiserfs_write_end,
3159         .bmap = reiserfs_aop_bmap,
3160         .direct_IO = reiserfs_direct_IO,
3161         .set_page_dirty = reiserfs_set_page_dirty,
3162 };