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 depth;
  35         int err;
  36
  37         truncate_inode_pages(&inode->i_data, 0);
  38
  39         depth = reiserfs_write_lock_once(inode->i_sb);
  40
  41         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  42         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  43                 reiserfs_delete_xattrs(inode);
  44
  45                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  46                         goto out;
  47                 reiserfs_update_inode_transaction(inode);
  48
  49                 reiserfs_discard_prealloc(&th, inode);
  50
  51                 err = reiserfs_delete_object(&th, inode);
  52
  53                 /* Do quota update inside a transaction for journaled quotas. We must do that
  54                  * after delete_object so that quota updates go into the same transaction as
  55                  * stat data deletion */
  56                 if (!err)
  57                         vfs_dq_free_inode(inode);
  58
  59                 if (journal_end(&th, inode->i_sb, jbegin_count))
  60                         goto out;
  61
  62                 /* check return value from reiserfs_delete_object after
  63                  * ending the transaction
  64                  */
  65                 if (err)
  66                     goto out;
  67
  68                 /* all items of file are deleted, so we can remove "save" link */
  69                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  70                                                                  * about an error here */
  71         } else {
  72                 /* no object items are in the tree */
  73                 ;
  74         }
  75       out:
  76         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  77         inode->i_blocks = 0;
  78         reiserfs_write_unlock_once(inode->i_sb, depth);
  79 }
  80
  81 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  82                           __u32 objectid, loff_t offset, int type, int length)
  83 {
  84         key->version = version;
  85
  86         key->on_disk_key.k_dir_id = dirid;
  87         key->on_disk_key.k_objectid = objectid;
  88         set_cpu_key_k_offset(key, offset);
  89         set_cpu_key_k_type(key, type);
  90         key->key_length = length;
  91 }
  92
  93 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  94    offset and type of key */
  95 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  96                   int type, int length)
  97 {
  98         _make_cpu_key(key, get_inode_item_key_version(inode),
  99                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 100                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 101                       length);
 102 }
 103
 104 //
 105 // when key is 0, do not set version and short key
 106 //
 107 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 108                               int version,
 109                               loff_t offset, int type, int length,
 110                               int entry_count /*or ih_free_space */ )
 111 {
 112         if (key) {
 113                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 114                 ih->ih_key.k_objectid =
 115                     cpu_to_le32(key->on_disk_key.k_objectid);
 116         }
 117         put_ih_version(ih, version);
 118         set_le_ih_k_offset(ih, offset);
 119         set_le_ih_k_type(ih, type);
 120         put_ih_item_len(ih, length);
 121         /*    set_ih_free_space (ih, 0); */
 122         // for directory items it is entry count, for directs and stat
 123         // datas - 0xffff, for indirects - 0
 124         put_ih_entry_count(ih, entry_count);
 125 }
 126
 127 //
 128 // FIXME: we might cache recently accessed indirect item
 129
 130 // Ugh.  Not too eager for that....
 131 //  I cut the code until such time as I see a convincing argument (benchmark).
 132 // I don't want a bloated inode struct..., and I don't like code complexity....
 133
 134 /* cutting the code is fine, since it really isn't in use yet and is easy
 135 ** to add back in.  But, Vladimir has a really good idea here.  Think
 136 ** about what happens for reading a file.  For each page,
 137 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 138 ** an indirect item.  This indirect item has X number of pointers, where
 139 ** X is a big number if we've done the block allocation right.  But,
 140 ** we only use one or two of these pointers during each call to readpage,
 141 ** needlessly researching again later on.
 142 **
 143 ** The size of the cache could be dynamic based on the size of the file.
 144 **
 145 ** I'd also like to see us cache the location the stat data item, since
 146 ** we are needlessly researching for that frequently.
 147 **
 148 ** --chris
 149 */
 150
 151 /* If this page has a file tail in it, and
 152 ** it was read in by get_block_create_0, the page data is valid,
 153 ** but tail is still sitting in a direct item, and we can't write to
 154 ** it.  So, look through this page, and check all the mapped buffers
 155 ** to make sure they have valid block numbers.  Any that don't need
 156 ** to be unmapped, so that block_prepare_write will correctly call
 157 ** reiserfs_get_block to convert the tail into an unformatted node
 158 */
 159 static inline void fix_tail_page_for_writing(struct page *page)
 160 {
 161         struct buffer_head *head, *next, *bh;
 162
 163         if (page && page_has_buffers(page)) {
 164                 head = page_buffers(page);
 165                 bh = head;
 166                 do {
 167                         next = bh->b_this_page;
 168                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 169                                 reiserfs_unmap_buffer(bh);
 170                         }
 171                         bh = next;
 172                 } while (bh != head);
 173         }
 174 }
 175
 176 /* reiserfs_get_block does not need to allocate a block only if it has been
 177    done already or non-hole position has been found in the indirect item */
 178 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 179                                     struct item_head *ih,
 180                                     __le32 * item, int pos_in_item)
 181 {
 182         if (allocated)
 183                 return 0;
 184         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 185             get_block_num(item, pos_in_item))
 186                 return 0;
 187         return 1;
 188 }
 189
 190 static inline int indirect_item_found(int retval, struct item_head *ih)
 191 {
 192         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 193 }
 194
 195 static inline void set_block_dev_mapped(struct buffer_head *bh,
 196                                         b_blocknr_t block, struct inode *inode)
 197 {
 198         map_bh(bh, inode->i_sb, block);
 199 }
 200
 201 //
 202 // files which were created in the earlier version can not be longer,
 203 // than 2 gb
 204 //
 205 static int file_capable(struct inode *inode, sector_t block)
 206 {
 207         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 208             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 209                 return 1;
 210
 211         return 0;
 212 }
 213
 214 static int restart_transaction(struct reiserfs_transaction_handle *th,
 215                                struct inode *inode, struct treepath *path)
 216 {
 217         struct super_block *s = th->t_super;
 218         int len = th->t_blocks_allocated;
 219         int err;
 220
 221         BUG_ON(!th->t_trans_id);
 222         BUG_ON(!th->t_refcount);
 223
 224         pathrelse(path);
 225
 226         /* we cannot restart while nested */
 227         if (th->t_refcount > 1) {
 228                 return 0;
 229         }
 230         reiserfs_update_sd(th, inode);
 231         err = journal_end(th, s, len);
 232         if (!err) {
 233                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 234                 if (!err)
 235                         reiserfs_update_inode_transaction(inode);
 236         }
 237         return err;
 238 }
 239
 240 // it is called by get_block when create == 0. Returns block number
 241 // for 'block'-th logical block of file. When it hits direct item it
 242 // returns 0 (being called from bmap) or read direct item into piece
 243 // of page (bh_result)
 244
 245 // Please improve the english/clarity in the comment above, as it is
 246 // hard to understand.
 247
 248 static int _get_block_create_0(struct inode *inode, sector_t block,
 249                                struct buffer_head *bh_result, int args)
 250 {
 251         INITIALIZE_PATH(path);
 252         struct cpu_key key;
 253         struct buffer_head *bh;
 254         struct item_head *ih, tmp_ih;
 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         result = search_for_position_by_key(inode->i_sb, &key, &path);
 269         if (result != POSITION_FOUND) {
 270                 pathrelse(&path);
 271                 if (p)
 272                         kunmap(bh_result->b_page);
 273                 if (result == IO_ERROR)
 274                         return -EIO;
 275                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 276                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 277                 if ((args & GET_BLOCK_NO_HOLE)
 278                     && !PageUptodate(bh_result->b_page)) {
 279                         return -ENOENT;
 280                 }
 281                 return 0;
 282         }
 283         //
 284         bh = get_last_bh(&path);
 285         ih = get_ih(&path);
 286         if (is_indirect_le_ih(ih)) {
 287                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 288
 289                 /* FIXME: here we could cache indirect item or part of it in
 290                    the inode to avoid search_by_key in case of subsequent
 291                    access to file */
 292                 blocknr = get_block_num(ind_item, path.pos_in_item);
 293                 ret = 0;
 294                 if (blocknr) {
 295                         map_bh(bh_result, inode->i_sb, blocknr);
 296                         if (path.pos_in_item ==
 297                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 298                                 set_buffer_boundary(bh_result);
 299                         }
 300                 } else
 301                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 302                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 303                 if ((args & GET_BLOCK_NO_HOLE)
 304                             && !PageUptodate(bh_result->b_page)) {
 305                         ret = -ENOENT;
 306                 }
 307
 308                 pathrelse(&path);
 309                 if (p)
 310                         kunmap(bh_result->b_page);
 311                 return ret;
 312         }
 313         // requested data are in direct item(s)
 314         if (!(args & GET_BLOCK_READ_DIRECT)) {
 315                 // we are called by bmap. FIXME: we can not map block of file
 316                 // when it is stored in direct item(s)
 317                 pathrelse(&path);
 318                 if (p)
 319                         kunmap(bh_result->b_page);
 320                 return -ENOENT;
 321         }
 322
 323         /* if we've got a direct item, and the buffer or page was uptodate,
 324          ** we don't want to pull data off disk again.  skip to the
 325          ** end, where we map the buffer and return
 326          */
 327         if (buffer_uptodate(bh_result)) {
 328                 goto finished;
 329         } else
 330                 /*
 331                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 332                  ** pages without any buffers.  If the page is up to date, we don't want
 333                  ** read old data off disk.  Set the up to date bit on the buffer instead
 334                  ** and jump to the end
 335                  */
 336         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 337                 set_buffer_uptodate(bh_result);
 338                 goto finished;
 339         }
 340         // read file tail into part of page
 341         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 342         copy_item_head(&tmp_ih, ih);
 343
 344         /* we only want to kmap if we are reading the tail into the page.
 345          ** this is not the common case, so we don't kmap until we are
 346          ** sure we need to.  But, this means the item might move if
 347          ** kmap schedules
 348          */
 349         if (!p)
 350                 p = (char *)kmap(bh_result->b_page);
 351
 352         p += offset;
 353         memset(p, 0, inode->i_sb->s_blocksize);
 354         do {
 355                 if (!is_direct_le_ih(ih)) {
 356                         BUG();
 357                 }
 358                 /* make sure we don't read more bytes than actually exist in
 359                  ** the file.  This can happen in odd cases where i_size isn't
 360                  ** correct, and when direct item padding results in a few
 361                  ** extra bytes at the end of the direct item
 362                  */
 363                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 364                         break;
 365                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 366                         chars =
 367                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 368                             path.pos_in_item;
 369                         done = 1;
 370                 } else {
 371                         chars = ih_item_len(ih) - path.pos_in_item;
 372                 }
 373                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 374
 375                 if (done)
 376                         break;
 377
 378                 p += chars;
 379
 380                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 381                         // we done, if read direct item is not the last item of
 382                         // node FIXME: we could try to check right delimiting key
 383                         // to see whether direct item continues in the right
 384                         // neighbor or rely on i_size
 385                         break;
 386
 387                 // update key to look for the next piece
 388                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 389                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 390                 if (result != POSITION_FOUND)
 391                         // i/o error most likely
 392                         break;
 393                 bh = get_last_bh(&path);
 394                 ih = get_ih(&path);
 395         } while (1);
 396
 397         flush_dcache_page(bh_result->b_page);
 398         kunmap(bh_result->b_page);
 399
 400       finished:
 401         pathrelse(&path);
 402
 403         if (result == IO_ERROR)
 404                 return -EIO;
 405
 406         /* this buffer has valid data, but isn't valid for io.  mapping it to
 407          * block #0 tells the rest of reiserfs it just has a tail in it
 408          */
 409         map_bh(bh_result, inode->i_sb, 0);
 410         set_buffer_uptodate(bh_result);
 411         return 0;
 412 }
 413
 414 // this is called to create file map. So, _get_block_create_0 will not
 415 // read direct item
 416 static int reiserfs_bmap(struct inode *inode, sector_t block,
 417                          struct buffer_head *bh_result, int create)
 418 {
 419         if (!file_capable(inode, block))
 420                 return -EFBIG;
 421
 422         reiserfs_write_lock(inode->i_sb);
 423         /* do not read the direct item */
 424         _get_block_create_0(inode, block, bh_result, 0);
 425         reiserfs_write_unlock(inode->i_sb);
 426         return 0;
 427 }
 428
 429 /* special version of get_block that is only used by grab_tail_page right
 430 ** now.  It is sent to block_prepare_write, and when you try to get a
 431 ** block past the end of the file (or a block from a hole) it returns
 432 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 433 ** be able to do i/o on the buffers returned, unless an error value
 434 ** is also returned.
 435 **
 436 ** So, this allows block_prepare_write to be used for reading a single block
 437 ** in a page.  Where it does not produce a valid page for holes, or past the
 438 ** end of the file.  This turns out to be exactly what we need for reading
 439 ** tails for conversion.
 440 **
 441 ** The point of the wrapper is forcing a certain value for create, even
 442 ** though the VFS layer is calling this function with create==1.  If you
 443 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 444 ** don't use this function.
 445 */
 446 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 447                                        struct buffer_head *bh_result,
 448                                        int create)
 449 {
 450         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 451 }
 452
 453 /* This is special helper for reiserfs_get_block in case we are executing
 454    direct_IO request. */
 455 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 456                                          sector_t iblock,
 457                                          struct buffer_head *bh_result,
 458                                          int create)
 459 {
 460         int ret;
 461
 462         bh_result->b_page = NULL;
 463
 464         /* We set the b_size before reiserfs_get_block call since it is
 465            referenced in convert_tail_for_hole() that may be called from
 466            reiserfs_get_block() */
 467         bh_result->b_size = (1 << inode->i_blkbits);
 468
 469         ret = reiserfs_get_block(inode, iblock, bh_result,
 470                                  create | GET_BLOCK_NO_DANGLE);
 471         if (ret)
 472                 goto out;
 473
 474         /* don't allow direct io onto tail pages */
 475         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 476                 /* make sure future calls to the direct io funcs for this offset
 477                  ** in the file fail by unmapping the buffer
 478                  */
 479                 clear_buffer_mapped(bh_result);
 480                 ret = -EINVAL;
 481         }
 482         /* Possible unpacked tail. Flush the data before pages have
 483            disappeared */
 484         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 485                 int err;
 486
 487                 reiserfs_write_lock(inode->i_sb);
 488
 489                 err = reiserfs_commit_for_inode(inode);
 490                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 491
 492                 reiserfs_write_unlock(inode->i_sb);
 493
 494                 if (err < 0)
 495                         ret = err;
 496         }
 497       out:
 498         return ret;
 499 }
 500
 501 /*
 502 ** helper function for when reiserfs_get_block is called for a hole
 503 ** but the file tail is still in a direct item
 504 ** bh_result is the buffer head for the hole
 505 ** tail_offset is the offset of the start of the tail in the file
 506 **
 507 ** This calls prepare_write, which will start a new transaction
 508 ** you should not be in a transaction, or have any paths held when you
 509 ** call this.
 510 */
 511 static int convert_tail_for_hole(struct inode *inode,
 512                                  struct buffer_head *bh_result,
 513                                  loff_t tail_offset)
 514 {
 515         unsigned long index;
 516         unsigned long tail_end;
 517         unsigned long tail_start;
 518         struct page *tail_page;
 519         struct page *hole_page = bh_result->b_page;
 520         int retval = 0;
 521
 522         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 523                 return -EIO;
 524
 525         /* always try to read until the end of the block */
 526         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 527         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 528
 529         index = tail_offset >> PAGE_CACHE_SHIFT;
 530         /* hole_page can be zero in case of direct_io, we are sure
 531            that we cannot get here if we write with O_DIRECT into
 532            tail page */
 533         if (!hole_page || index != hole_page->index) {
 534                 tail_page = grab_cache_page(inode->i_mapping, index);
 535                 retval = -ENOMEM;
 536                 if (!tail_page) {
 537                         goto out;
 538                 }
 539         } else {
 540                 tail_page = hole_page;
 541         }
 542
 543         /* we don't have to make sure the conversion did not happen while
 544          ** we were locking the page because anyone that could convert
 545          ** must first take i_mutex.
 546          **
 547          ** We must fix the tail page for writing because it might have buffers
 548          ** that are mapped, but have a block number of 0.  This indicates tail
 549          ** data that has been read directly into the page, and block_prepare_write
 550          ** won't trigger a get_block in this case.
 551          */
 552         fix_tail_page_for_writing(tail_page);
 553         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 554         if (retval)
 555                 goto unlock;
 556
 557         /* tail conversion might change the data in the page */
 558         flush_dcache_page(tail_page);
 559
 560         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 561
 562       unlock:
 563         if (tail_page != hole_page) {
 564                 unlock_page(tail_page);
 565                 page_cache_release(tail_page);
 566         }
 567       out:
 568         return retval;
 569 }
 570
 571 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 572                                   sector_t block,
 573                                   struct inode *inode,
 574                                   b_blocknr_t * allocated_block_nr,
 575                                   struct treepath *path, int flags)
 576 {
 577         BUG_ON(!th->t_trans_id);
 578
 579 #ifdef REISERFS_PREALLOCATE
 580         if (!(flags & GET_BLOCK_NO_IMUX)) {
 581                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 582                                                   path, block);
 583         }
 584 #endif
 585         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 586                                          block);
 587 }
 588
 589 int reiserfs_get_block(struct inode *inode, sector_t block,
 590                        struct buffer_head *bh_result, int create)
 591 {
 592         int repeat, retval = 0;
 593         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 594         INITIALIZE_PATH(path);
 595         int pos_in_item;
 596         struct cpu_key key;
 597         struct buffer_head *bh, *unbh = NULL;
 598         struct item_head *ih, tmp_ih;
 599         __le32 *item;
 600         int done;
 601         int fs_gen;
 602         int lock_depth;
 603         struct reiserfs_transaction_handle *th = NULL;
 604         /* space reserved in transaction batch:
 605            . 3 balancings in direct->indirect conversion
 606            . 1 block involved into reiserfs_update_sd()
 607            XXX in practically impossible worst case direct2indirect()
 608            can incur (much) more than 3 balancings.
 609            quota update for user, group */
 610         int jbegin_count =
 611             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 612             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 613         int version;
 614         int dangle = 1;
 615         loff_t new_offset =
 616             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 617
 618         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 619         version = get_inode_item_key_version(inode);
 620
 621         if (!file_capable(inode, block)) {
 622                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 623                 return -EFBIG;
 624         }
 625
 626         /* if !create, we aren't changing the FS, so we don't need to
 627          ** log anything, so we don't need to start a transaction
 628          */
 629         if (!(create & GET_BLOCK_CREATE)) {
 630                 int ret;
 631                 /* find number of block-th logical block of the file */
 632                 ret = _get_block_create_0(inode, block, bh_result,
 633                                           create | GET_BLOCK_READ_DIRECT);
 634                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 635                 return ret;
 636         }
 637         /*
 638          * if we're already in a transaction, make sure to close
 639          * any new transactions we start in this func
 640          */
 641         if ((create & GET_BLOCK_NO_DANGLE) ||
 642             reiserfs_transaction_running(inode->i_sb))
 643                 dangle = 0;
 644
 645         /* If file is of such a size, that it might have a tail and tails are enabled
 646          ** we should mark it as possibly needing tail packing on close
 647          */
 648         if ((have_large_tails(inode->i_sb)
 649              && inode->i_size < i_block_size(inode) * 4)
 650             || (have_small_tails(inode->i_sb)
 651                 && inode->i_size < i_block_size(inode)))
 652                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 653
 654         /* set the key of the first byte in the 'block'-th block of file */
 655         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 656         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 657               start_trans:
 658                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 659                 if (!th) {
 660                         retval = -ENOMEM;
 661                         goto failure;
 662                 }
 663                 reiserfs_update_inode_transaction(inode);
 664         }
 665       research:
 666
 667         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 668         if (retval == IO_ERROR) {
 669                 retval = -EIO;
 670                 goto failure;
 671         }
 672
 673         bh = get_last_bh(&path);
 674         ih = get_ih(&path);
 675         item = get_item(&path);
 676         pos_in_item = path.pos_in_item;
 677
 678         fs_gen = get_generation(inode->i_sb);
 679         copy_item_head(&tmp_ih, ih);
 680
 681         if (allocation_needed
 682             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 683                 /* we have to allocate block for the unformatted node */
 684                 if (!th) {
 685                         pathrelse(&path);
 686                         goto start_trans;
 687                 }
 688
 689                 repeat =
 690                     _allocate_block(th, block, inode, &allocated_block_nr,
 691                                     &path, create);
 692
 693                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 694                         /* restart the transaction to give the journal a chance to free
 695                          ** some blocks.  releases the path, so we have to go back to
 696                          ** research if we succeed on the second try
 697                          */
 698                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 699                         retval = restart_transaction(th, inode, &path);
 700                         if (retval)
 701                                 goto failure;
 702                         repeat =
 703                             _allocate_block(th, block, inode,
 704                                             &allocated_block_nr, NULL, create);
 705
 706                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 707                                 goto research;
 708                         }
 709                         if (repeat == QUOTA_EXCEEDED)
 710                                 retval = -EDQUOT;
 711                         else
 712                                 retval = -ENOSPC;
 713                         goto failure;
 714                 }
 715
 716                 if (fs_changed(fs_gen, inode->i_sb)
 717                     && item_moved(&tmp_ih, &path)) {
 718                         goto research;
 719                 }
 720         }
 721
 722         if (indirect_item_found(retval, ih)) {
 723                 b_blocknr_t unfm_ptr;
 724                 /* 'block'-th block is in the file already (there is
 725                    corresponding cell in some indirect item). But it may be
 726                    zero unformatted node pointer (hole) */
 727                 unfm_ptr = get_block_num(item, pos_in_item);
 728                 if (unfm_ptr == 0) {
 729                         /* use allocated block to plug the hole */
 730                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 731                         if (fs_changed(fs_gen, inode->i_sb)
 732                             && item_moved(&tmp_ih, &path)) {
 733                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 734                                                                  bh);
 735                                 goto research;
 736                         }
 737                         set_buffer_new(bh_result);
 738                         if (buffer_dirty(bh_result)
 739                             && reiserfs_data_ordered(inode->i_sb))
 740                                 reiserfs_add_ordered_list(inode, bh_result);
 741                         put_block_num(item, pos_in_item, allocated_block_nr);
 742                         unfm_ptr = allocated_block_nr;
 743                         journal_mark_dirty(th, inode->i_sb, bh);
 744                         reiserfs_update_sd(th, inode);
 745                 }
 746                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 747                 pathrelse(&path);
 748                 retval = 0;
 749                 if (!dangle && th)
 750                         retval = reiserfs_end_persistent_transaction(th);
 751
 752                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 753
 754                 /* the item was found, so new blocks were not added to the file
 755                  ** there is no need to make sure the inode is updated with this
 756                  ** transaction
 757                  */
 758                 return retval;
 759         }
 760
 761         if (!th) {
 762                 pathrelse(&path);
 763                 goto start_trans;
 764         }
 765
 766         /* desired position is not found or is in the direct item. We have
 767            to append file with holes up to 'block'-th block converting
 768            direct items to indirect one if necessary */
 769         done = 0;
 770         do {
 771                 if (is_statdata_le_ih(ih)) {
 772                         __le32 unp = 0;
 773                         struct cpu_key tmp_key;
 774
 775                         /* indirect item has to be inserted */
 776                         make_le_item_head(&tmp_ih, &key, version, 1,
 777                                           TYPE_INDIRECT, UNFM_P_SIZE,
 778                                           0 /* free_space */ );
 779
 780                         if (cpu_key_k_offset(&key) == 1) {
 781                                 /* we are going to add 'block'-th block to the file. Use
 782                                    allocated block for that */
 783                                 unp = cpu_to_le32(allocated_block_nr);
 784                                 set_block_dev_mapped(bh_result,
 785                                                      allocated_block_nr, inode);
 786                                 set_buffer_new(bh_result);
 787                                 done = 1;
 788                         }
 789                         tmp_key = key;  // ;)
 790                         set_cpu_key_k_offset(&tmp_key, 1);
 791                         PATH_LAST_POSITION(&path)++;
 792
 793                         retval =
 794                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 795                                                  inode, (char *)&unp);
 796                         if (retval) {
 797                                 reiserfs_free_block(th, inode,
 798                                                     allocated_block_nr, 1);
 799                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 800                         }
 801                         //mark_tail_converted (inode);
 802                 } else if (is_direct_le_ih(ih)) {
 803                         /* direct item has to be converted */
 804                         loff_t tail_offset;
 805
 806                         tail_offset =
 807                             ((le_ih_k_offset(ih) -
 808                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 809                         if (tail_offset == cpu_key_k_offset(&key)) {
 810                                 /* direct item we just found fits into block we have
 811                                    to map. Convert it into unformatted node: use
 812                                    bh_result for the conversion */
 813                                 set_block_dev_mapped(bh_result,
 814                                                      allocated_block_nr, inode);
 815                                 unbh = bh_result;
 816                                 done = 1;
 817                         } else {
 818                                 /* we have to padd file tail stored in direct item(s)
 819                                    up to block size and convert it to unformatted
 820                                    node. FIXME: this should also get into page cache */
 821
 822                                 pathrelse(&path);
 823                                 /*
 824                                  * ugly, but we can only end the transaction if
 825                                  * we aren't nested
 826                                  */
 827                                 BUG_ON(!th->t_refcount);
 828                                 if (th->t_refcount == 1) {
 829                                         retval =
 830                                             reiserfs_end_persistent_transaction
 831                                             (th);
 832                                         th = NULL;
 833                                         if (retval)
 834                                                 goto failure;
 835                                 }
 836
 837                                 retval =
 838                                     convert_tail_for_hole(inode, bh_result,
 839                                                           tail_offset);
 840                                 if (retval) {
 841                                         if (retval != -ENOSPC)
 842                                                 reiserfs_error(inode->i_sb,
 843                                                         "clm-6004",
 844                                                         "convert tail failed "
 845                                                         "inode %lu, error %d",
 846                                                         inode->i_ino,
 847                                                         retval);
 848                                         if (allocated_block_nr) {
 849                                                 /* the bitmap, the super, and the stat data == 3 */
 850                                                 if (!th)
 851                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 852                                                 if (th)
 853                                                         reiserfs_free_block(th,
 854                                                                             inode,
 855                                                                             allocated_block_nr,
 856                                                                             1);
 857                                         }
 858                                         goto failure;
 859                                 }
 860                                 goto research;
 861                         }
 862                         retval =
 863                             direct2indirect(th, inode, &path, unbh,
 864                                             tail_offset);
 865                         if (retval) {
 866                                 reiserfs_unmap_buffer(unbh);
 867                                 reiserfs_free_block(th, inode,
 868                                                     allocated_block_nr, 1);
 869                                 goto failure;
 870                         }
 871                         /* it is important the set_buffer_uptodate is done after
 872                          ** the direct2indirect.  The buffer might contain valid
 873                          ** data newer than the data on disk (read by readpage, changed,
 874                          ** and then sent here by writepage).  direct2indirect needs
 875                          ** to know if unbh was already up to date, so it can decide
 876                          ** if the data in unbh needs to be replaced with data from
 877                          ** the disk
 878                          */
 879                         set_buffer_uptodate(unbh);
 880
 881                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 882                            buffer will disappear shortly, so it should not be added to
 883                          */
 884                         if (unbh->b_page) {
 885                                 /* we've converted the tail, so we must
 886                                  ** flush unbh before the transaction commits
 887                                  */
 888                                 reiserfs_add_tail_list(inode, unbh);
 889
 890                                 /* mark it dirty now to prevent commit_write from adding
 891                                  ** this buffer to the inode's dirty buffer list
 892                                  */
 893                                 /*
 894                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 895                                  * It's still atomic, but it sets the page dirty too,
 896                                  * which makes it eligible for writeback at any time by the
 897                                  * VM (which was also the case with __mark_buffer_dirty())
 898                                  */
 899                                 mark_buffer_dirty(unbh);
 900                         }
 901                 } else {
 902                         /* append indirect item with holes if needed, when appending
 903                            pointer to 'block'-th block use block, which is already
 904                            allocated */
 905                         struct cpu_key tmp_key;
 906                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 907                         // one block which is a fastpath
 908                         unp_t *un;
 909                         __u64 max_to_insert =
 910                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 911                             UNFM_P_SIZE;
 912                         __u64 blocks_needed;
 913
 914                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 915                                "vs-804: invalid position for append");
 916                         /* indirect item has to be appended, set up key of that position */
 917                         make_cpu_key(&tmp_key, inode,
 918                                      le_key_k_offset(version,
 919                                                      &(ih->ih_key)) +
 920                                      op_bytes_number(ih,
 921                                                      inode->i_sb->s_blocksize),
 922                                      //pos_in_item * inode->i_sb->s_blocksize,
 923                                      TYPE_INDIRECT, 3); // key type is unimportant
 924
 925                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 926                                "green-805: invalid offset");
 927                         blocks_needed =
 928                             1 +
 929                             ((cpu_key_k_offset(&key) -
 930                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 931                              s_blocksize_bits);
 932
 933                         if (blocks_needed == 1) {
 934                                 un = &unf_single;
 935                         } else {
 936                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 937                                 if (!un) {
 938                                         un = &unf_single;
 939                                         blocks_needed = 1;
 940                                         max_to_insert = 0;
 941                                 }
 942                         }
 943                         if (blocks_needed <= max_to_insert) {
 944                                 /* we are going to add target block to the file. Use allocated
 945                                    block for that */
 946                                 un[blocks_needed - 1] =
 947                                     cpu_to_le32(allocated_block_nr);
 948                                 set_block_dev_mapped(bh_result,
 949                                                      allocated_block_nr, inode);
 950                                 set_buffer_new(bh_result);
 951                                 done = 1;
 952                         } else {
 953                                 /* paste hole to the indirect item */
 954                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 955                                    only have space for one block */
 956                                 blocks_needed =
 957                                     max_to_insert ? max_to_insert : 1;
 958                         }
 959                         retval =
 960                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 961                                                      (char *)un,
 962                                                      UNFM_P_SIZE *
 963                                                      blocks_needed);
 964
 965                         if (blocks_needed != 1)
 966                                 kfree(un);
 967
 968                         if (retval) {
 969                                 reiserfs_free_block(th, inode,
 970                                                     allocated_block_nr, 1);
 971                                 goto failure;
 972                         }
 973                         if (!done) {
 974                                 /* We need to mark new file size in case this function will be
 975                                    interrupted/aborted later on. And we may do this only for
 976                                    holes. */
 977                                 inode->i_size +=
 978                                     inode->i_sb->s_blocksize * blocks_needed;
 979                         }
 980                 }
 981
 982                 if (done == 1)
 983                         break;
 984
 985                 /* this loop could log more blocks than we had originally asked
 986                  ** for.  So, we have to allow the transaction to end if it is
 987                  ** too big or too full.  Update the inode so things are
 988                  ** consistent if we crash before the function returns
 989                  **
 990                  ** release the path so that anybody waiting on the path before
 991                  ** ending their transaction will be able to continue.
 992                  */
 993                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 994                         retval = restart_transaction(th, inode, &path);
 995                         if (retval)
 996                                 goto failure;
 997                 }
 998                 /*
 999                  * inserting indirect pointers for a hole can take a
1000                  * long time.  reschedule if needed and also release the write
1001                  * lock for others.
1002                  */
1003                 if (need_resched()) {
1004                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1005                         schedule();
1006                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1007                 }
1008
1009                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1010                 if (retval == IO_ERROR) {
1011                         retval = -EIO;
1012                         goto failure;
1013                 }
1014                 if (retval == POSITION_FOUND) {
1015                         reiserfs_warning(inode->i_sb, "vs-825",
1016                                          "%K should not be found", &key);
1017                         retval = -EEXIST;
1018                         if (allocated_block_nr)
1019                                 reiserfs_free_block(th, inode,
1020                                                     allocated_block_nr, 1);
1021                         pathrelse(&path);
1022                         goto failure;
1023                 }
1024                 bh = get_last_bh(&path);
1025                 ih = get_ih(&path);
1026                 item = get_item(&path);
1027                 pos_in_item = path.pos_in_item;
1028         } while (1);
1029
1030         retval = 0;
1031
1032       failure:
1033         if (th && (!dangle || (retval && !th->t_trans_id))) {
1034                 int err;
1035                 if (th->t_trans_id)
1036                         reiserfs_update_sd(th, inode);
1037                 err = reiserfs_end_persistent_transaction(th);
1038                 if (err)
1039                         retval = err;
1040         }
1041
1042         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1043         reiserfs_check_path(&path);
1044         return retval;
1045 }
1046
1047 static int
1048 reiserfs_readpages(struct file *file, struct address_space *mapping,
1049                    struct list_head *pages, unsigned nr_pages)
1050 {
1051         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1052 }
1053
1054 /* Compute real number of used bytes by file
1055  * Following three functions can go away when we'll have enough space in stat item
1056  */
1057 static int real_space_diff(struct inode *inode, int sd_size)
1058 {
1059         int bytes;
1060         loff_t blocksize = inode->i_sb->s_blocksize;
1061
1062         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1063                 return sd_size;
1064
1065         /* End of file is also in full block with indirect reference, so round
1066          ** up to the next block.
1067          **
1068          ** there is just no way to know if the tail is actually packed
1069          ** on the file, so we have to assume it isn't.  When we pack the
1070          ** tail, we add 4 bytes to pretend there really is an unformatted
1071          ** node pointer
1072          */
1073         bytes =
1074             ((inode->i_size +
1075               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1076             sd_size;
1077         return bytes;
1078 }
1079
1080 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1081                                         int sd_size)
1082 {
1083         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1084                 return inode->i_size +
1085                     (loff_t) (real_space_diff(inode, sd_size));
1086         }
1087         return ((loff_t) real_space_diff(inode, sd_size)) +
1088             (((loff_t) blocks) << 9);
1089 }
1090
1091 /* Compute number of blocks used by file in ReiserFS counting */
1092 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1093 {
1094         loff_t bytes = inode_get_bytes(inode);
1095         loff_t real_space = real_space_diff(inode, sd_size);
1096
1097         /* keeps fsck and non-quota versions of reiserfs happy */
1098         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1099                 bytes += (loff_t) 511;
1100         }
1101
1102         /* files from before the quota patch might i_blocks such that
1103          ** bytes < real_space.  Deal with that here to prevent it from
1104          ** going negative.
1105          */
1106         if (bytes < real_space)
1107                 return 0;
1108         return (bytes - real_space) >> 9;
1109 }
1110
1111 //
1112 // BAD: new directories have stat data of new type and all other items
1113 // of old type. Version stored in the inode says about body items, so
1114 // in update_stat_data we can not rely on inode, but have to check
1115 // item version directly
1116 //
1117
1118 // called by read_locked_inode
1119 static void init_inode(struct inode *inode, struct treepath *path)
1120 {
1121         struct buffer_head *bh;
1122         struct item_head *ih;
1123         __u32 rdev;
1124         //int version = ITEM_VERSION_1;
1125
1126         bh = PATH_PLAST_BUFFER(path);
1127         ih = PATH_PITEM_HEAD(path);
1128
1129         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1130
1131         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1132         REISERFS_I(inode)->i_flags = 0;
1133         REISERFS_I(inode)->i_prealloc_block = 0;
1134         REISERFS_I(inode)->i_prealloc_count = 0;
1135         REISERFS_I(inode)->i_trans_id = 0;
1136         REISERFS_I(inode)->i_jl = NULL;
1137         mutex_init(&(REISERFS_I(inode)->i_mmap));
1138         reiserfs_init_xattr_rwsem(inode);
1139
1140         if (stat_data_v1(ih)) {
1141                 struct stat_data_v1 *sd =
1142                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1143                 unsigned long blocks;
1144
1145                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1146                 set_inode_sd_version(inode, STAT_DATA_V1);
1147                 inode->i_mode = sd_v1_mode(sd);
1148                 inode->i_nlink = sd_v1_nlink(sd);
1149                 inode->i_uid = sd_v1_uid(sd);
1150                 inode->i_gid = sd_v1_gid(sd);
1151                 inode->i_size = sd_v1_size(sd);
1152                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1153                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1154                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1155                 inode->i_atime.tv_nsec = 0;
1156                 inode->i_ctime.tv_nsec = 0;
1157                 inode->i_mtime.tv_nsec = 0;
1158
1159                 inode->i_blocks = sd_v1_blocks(sd);
1160                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1161                 blocks = (inode->i_size + 511) >> 9;
1162                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1163                 if (inode->i_blocks > blocks) {
1164                         // there was a bug in <=3.5.23 when i_blocks could take negative
1165                         // values. Starting from 3.5.17 this value could even be stored in
1166                         // stat data. For such files we set i_blocks based on file
1167                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1168                         // only updated if file's inode will ever change
1169                         inode->i_blocks = blocks;
1170                 }
1171
1172                 rdev = sd_v1_rdev(sd);
1173                 REISERFS_I(inode)->i_first_direct_byte =
1174                     sd_v1_first_direct_byte(sd);
1175                 /* an early bug in the quota code can give us an odd number for the
1176                  ** block count.  This is incorrect, fix it here.
1177                  */
1178                 if (inode->i_blocks & 1) {
1179                         inode->i_blocks++;
1180                 }
1181                 inode_set_bytes(inode,
1182                                 to_real_used_space(inode, inode->i_blocks,
1183                                                    SD_V1_SIZE));
1184                 /* nopack is initially zero for v1 objects. For v2 objects,
1185                    nopack is initialised from sd_attrs */
1186                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1187         } else {
1188                 // new stat data found, but object may have old items
1189                 // (directories and symlinks)
1190                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1191
1192                 inode->i_mode = sd_v2_mode(sd);
1193                 inode->i_nlink = sd_v2_nlink(sd);
1194                 inode->i_uid = sd_v2_uid(sd);
1195                 inode->i_size = sd_v2_size(sd);
1196                 inode->i_gid = sd_v2_gid(sd);
1197                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1198                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1199                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1200                 inode->i_ctime.tv_nsec = 0;
1201                 inode->i_mtime.tv_nsec = 0;
1202                 inode->i_atime.tv_nsec = 0;
1203                 inode->i_blocks = sd_v2_blocks(sd);
1204                 rdev = sd_v2_rdev(sd);
1205                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1206                         inode->i_generation =
1207                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1208                 else
1209                         inode->i_generation = sd_v2_generation(sd);
1210
1211                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1212                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1213                 else
1214                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1215                 REISERFS_I(inode)->i_first_direct_byte = 0;
1216                 set_inode_sd_version(inode, STAT_DATA_V2);
1217                 inode_set_bytes(inode,
1218                                 to_real_used_space(inode, inode->i_blocks,
1219                                                    SD_V2_SIZE));
1220                 /* read persistent inode attributes from sd and initalise
1221                    generic inode flags from them */
1222                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1223                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1224         }
1225
1226         pathrelse(path);
1227         if (S_ISREG(inode->i_mode)) {
1228                 inode->i_op = &reiserfs_file_inode_operations;
1229                 inode->i_fop = &reiserfs_file_operations;
1230                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1231         } else if (S_ISDIR(inode->i_mode)) {
1232                 inode->i_op = &reiserfs_dir_inode_operations;
1233                 inode->i_fop = &reiserfs_dir_operations;
1234         } else if (S_ISLNK(inode->i_mode)) {
1235                 inode->i_op = &reiserfs_symlink_inode_operations;
1236                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1237         } else {
1238                 inode->i_blocks = 0;
1239                 inode->i_op = &reiserfs_special_inode_operations;
1240                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1241         }
1242 }
1243
1244 // update new stat data with inode fields
1245 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1246 {
1247         struct stat_data *sd_v2 = (struct stat_data *)sd;
1248         __u16 flags;
1249
1250         set_sd_v2_mode(sd_v2, inode->i_mode);
1251         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1252         set_sd_v2_uid(sd_v2, inode->i_uid);
1253         set_sd_v2_size(sd_v2, size);
1254         set_sd_v2_gid(sd_v2, inode->i_gid);
1255         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1256         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1257         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1258         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1259         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1260                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1261         else
1262                 set_sd_v2_generation(sd_v2, inode->i_generation);
1263         flags = REISERFS_I(inode)->i_attrs;
1264         i_attrs_to_sd_attrs(inode, &flags);
1265         set_sd_v2_attrs(sd_v2, flags);
1266 }
1267
1268 // used to copy inode's fields to old stat data
1269 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1270 {
1271         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1272
1273         set_sd_v1_mode(sd_v1, inode->i_mode);
1274         set_sd_v1_uid(sd_v1, inode->i_uid);
1275         set_sd_v1_gid(sd_v1, inode->i_gid);
1276         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1277         set_sd_v1_size(sd_v1, size);
1278         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1279         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1280         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1281
1282         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1283                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1284         else
1285                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1286
1287         // Sigh. i_first_direct_byte is back
1288         set_sd_v1_first_direct_byte(sd_v1,
1289                                     REISERFS_I(inode)->i_first_direct_byte);
1290 }
1291
1292 /* NOTE, you must prepare the buffer head before sending it here,
1293 ** and then log it after the call
1294 */
1295 static void update_stat_data(struct treepath *path, struct inode *inode,
1296                              loff_t size)
1297 {
1298         struct buffer_head *bh;
1299         struct item_head *ih;
1300
1301         bh = PATH_PLAST_BUFFER(path);
1302         ih = PATH_PITEM_HEAD(path);
1303
1304         if (!is_statdata_le_ih(ih))
1305                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1306                                INODE_PKEY(inode), ih);
1307
1308         if (stat_data_v1(ih)) {
1309                 // path points to old stat data
1310                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1311         } else {
1312                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1313         }
1314
1315         return;
1316 }
1317
1318 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1319                              struct inode *inode, loff_t size)
1320 {
1321         struct cpu_key key;
1322         INITIALIZE_PATH(path);
1323         struct buffer_head *bh;
1324         int fs_gen;
1325         struct item_head *ih, tmp_ih;
1326         int retval;
1327
1328         BUG_ON(!th->t_trans_id);
1329
1330         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1331
1332         for (;;) {
1333                 int pos;
1334                 /* look for the object's stat data */
1335                 retval = search_item(inode->i_sb, &key, &path);
1336                 if (retval == IO_ERROR) {
1337                         reiserfs_error(inode->i_sb, "vs-13050",
1338                                        "i/o failure occurred trying to "
1339                                        "update %K stat data", &key);
1340                         return;
1341                 }
1342                 if (retval == ITEM_NOT_FOUND) {
1343                         pos = PATH_LAST_POSITION(&path);
1344                         pathrelse(&path);
1345                         if (inode->i_nlink == 0) {
1346                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1347                                 return;
1348                         }
1349                         reiserfs_warning(inode->i_sb, "vs-13060",
1350                                          "stat data of object %k (nlink == %d) "
1351                                          "not found (pos %d)",
1352                                          INODE_PKEY(inode), inode->i_nlink,
1353                                          pos);
1354                         reiserfs_check_path(&path);
1355                         return;
1356                 }
1357
1358                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1359                  ** FS might change.  We have to detect that, and loop back to the
1360                  ** search if the stat data item has moved
1361                  */
1362                 bh = get_last_bh(&path);
1363                 ih = get_ih(&path);
1364                 copy_item_head(&tmp_ih, ih);
1365                 fs_gen = get_generation(inode->i_sb);
1366                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1367                 if (fs_changed(fs_gen, inode->i_sb)
1368                     && item_moved(&tmp_ih, &path)) {
1369                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1370                         continue;       /* Stat_data item has been moved after scheduling. */
1371                 }
1372                 break;
1373         }
1374         update_stat_data(&path, inode, size);
1375         journal_mark_dirty(th, th->t_super, bh);
1376         pathrelse(&path);
1377         return;
1378 }
1379
1380 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1381 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1382 ** and clear the key in the private portion of the inode, otherwise a
1383 ** corresponding iput might try to delete whatever object the inode last
1384 ** represented.
1385 */
1386 static void reiserfs_make_bad_inode(struct inode *inode)
1387 {
1388         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1389         make_bad_inode(inode);
1390 }
1391
1392 //
1393 // initially this function was derived from minix or ext2's analog and
1394 // evolved as the prototype did
1395 //
1396
1397 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1398 {
1399         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1400         inode->i_ino = args->objectid;
1401         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1402         return 0;
1403 }
1404
1405 /* looks for stat data in the tree, and fills up the fields of in-core
1406    inode stat data fields */
1407 void reiserfs_read_locked_inode(struct inode *inode,
1408                                 struct reiserfs_iget_args *args)
1409 {
1410         INITIALIZE_PATH(path_to_sd);
1411         struct cpu_key key;
1412         unsigned long dirino;
1413         int retval;
1414
1415         dirino = args->dirid;
1416
1417         /* set version 1, version 2 could be used too, because stat data
1418            key is the same in both versions */
1419         key.version = KEY_FORMAT_3_5;
1420         key.on_disk_key.k_dir_id = dirino;
1421         key.on_disk_key.k_objectid = inode->i_ino;
1422         key.on_disk_key.k_offset = 0;
1423         key.on_disk_key.k_type = 0;
1424
1425         /* look for the object's stat data */
1426         retval = search_item(inode->i_sb, &key, &path_to_sd);
1427         if (retval == IO_ERROR) {
1428                 reiserfs_error(inode->i_sb, "vs-13070",
1429                                "i/o failure occurred trying to find "
1430                                "stat data of %K", &key);
1431                 reiserfs_make_bad_inode(inode);
1432                 return;
1433         }
1434         if (retval != ITEM_FOUND) {
1435                 /* a stale NFS handle can trigger this without it being an error */
1436                 pathrelse(&path_to_sd);
1437                 reiserfs_make_bad_inode(inode);
1438                 inode->i_nlink = 0;
1439                 return;
1440         }
1441
1442         init_inode(inode, &path_to_sd);
1443
1444         /* It is possible that knfsd is trying to access inode of a file
1445            that is being removed from the disk by some other thread. As we
1446            update sd on unlink all that is required is to check for nlink
1447            here. This bug was first found by Sizif when debugging
1448            SquidNG/Butterfly, forgotten, and found again after Philippe
1449            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1450
1451            More logical fix would require changes in fs/inode.c:iput() to
1452            remove inode from hash-table _after_ fs cleaned disk stuff up and
1453            in iget() to return NULL if I_FREEING inode is found in
1454            hash-table. */
1455         /* Currently there is one place where it's ok to meet inode with
1456            nlink==0: processing of open-unlinked and half-truncated files
1457            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1458         if ((inode->i_nlink == 0) &&
1459             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1460                 reiserfs_warning(inode->i_sb, "vs-13075",
1461                                  "dead inode read from disk %K. "
1462                                  "This is likely to be race with knfsd. Ignore",
1463                                  &key);
1464                 reiserfs_make_bad_inode(inode);
1465         }
1466
1467         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1468
1469 }
1470
1471 /**
1472  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1473  *
1474  * @inode:    inode from hash table to check
1475  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1476  *
1477  * This function is called by iget5_locked() to distinguish reiserfs inodes
1478  * having the same inode numbers. Such inodes can only exist due to some
1479  * error condition. One of them should be bad. Inodes with identical
1480  * inode numbers (objectids) are distinguished by parent directory ids.
1481  *
1482  */
1483 int reiserfs_find_actor(struct inode *inode, void *opaque)
1484 {
1485         struct reiserfs_iget_args *args;
1486
1487         args = opaque;
1488         /* args is already in CPU order */
1489         return (inode->i_ino == args->objectid) &&
1490             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1491 }
1492
1493 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1494 {
1495         struct inode *inode;
1496         struct reiserfs_iget_args args;
1497
1498         args.objectid = key->on_disk_key.k_objectid;
1499         args.dirid = key->on_disk_key.k_dir_id;
1500         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1501                              reiserfs_find_actor, reiserfs_init_locked_inode,
1502                              (void *)(&args));
1503         if (!inode)
1504                 return ERR_PTR(-ENOMEM);
1505
1506         if (inode->i_state & I_NEW) {
1507                 reiserfs_read_locked_inode(inode, &args);
1508                 unlock_new_inode(inode);
1509         }
1510
1511         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1512                 /* either due to i/o error or a stale NFS handle */
1513                 iput(inode);
1514                 inode = NULL;
1515         }
1516         return inode;
1517 }
1518
1519 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1520         u32 objectid, u32 dir_id, u32 generation)
1521
1522 {
1523         struct cpu_key key;
1524         struct inode *inode;
1525
1526         key.on_disk_key.k_objectid = objectid;
1527         key.on_disk_key.k_dir_id = dir_id;
1528         reiserfs_write_lock(sb);
1529         inode = reiserfs_iget(sb, &key);
1530         if (inode && !IS_ERR(inode) && generation != 0 &&
1531             generation != inode->i_generation) {
1532                 iput(inode);
1533                 inode = NULL;
1534         }
1535         reiserfs_write_unlock(sb);
1536
1537         return d_obtain_alias(inode);
1538 }
1539
1540 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1541                 int fh_len, int fh_type)
1542 {
1543         /* fhtype happens to reflect the number of u32s encoded.
1544          * due to a bug in earlier code, fhtype might indicate there
1545          * are more u32s then actually fitted.
1546          * so if fhtype seems to be more than len, reduce fhtype.
1547          * Valid types are:
1548          *   2 - objectid + dir_id - legacy support
1549          *   3 - objectid + dir_id + generation
1550          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1551          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1552          *   6 - as above plus generation of directory
1553          * 6 does not fit in NFSv2 handles
1554          */
1555         if (fh_type > fh_len) {
1556                 if (fh_type != 6 || fh_len != 5)
1557                         reiserfs_warning(sb, "reiserfs-13077",
1558                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1559                                 fh_type, fh_len);
1560                 fh_type = 5;
1561         }
1562
1563         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1564                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1565 }
1566
1567 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1568                 int fh_len, int fh_type)
1569 {
1570         if (fh_type < 4)
1571                 return NULL;
1572
1573         return reiserfs_get_dentry(sb,
1574                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1575                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1576                 (fh_type == 6) ? fid->raw[5] : 0);
1577 }
1578
1579 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1580                        int need_parent)
1581 {
1582         struct inode *inode = dentry->d_inode;
1583         int maxlen = *lenp;
1584
1585         if (maxlen < 3)
1586                 return 255;
1587
1588         data[0] = inode->i_ino;
1589         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1590         data[2] = inode->i_generation;
1591         *lenp = 3;
1592         /* no room for directory info? return what we've stored so far */
1593         if (maxlen < 5 || !need_parent)
1594                 return 3;
1595
1596         spin_lock(&dentry->d_lock);
1597         inode = dentry->d_parent->d_inode;
1598         data[3] = inode->i_ino;
1599         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1600         *lenp = 5;
1601         if (maxlen >= 6) {
1602                 data[5] = inode->i_generation;
1603                 *lenp = 6;
1604         }
1605         spin_unlock(&dentry->d_lock);
1606         return *lenp;
1607 }
1608
1609 /* looks for stat data, then copies fields to it, marks the buffer
1610    containing stat data as dirty */
1611 /* reiserfs inodes are never really dirty, since the dirty inode call
1612 ** always logs them.  This call allows the VFS inode marking routines
1613 ** to properly mark inodes for datasync and such, but only actually
1614 ** does something when called for a synchronous update.
1615 */
1616 int reiserfs_write_inode(struct inode *inode, int do_sync)
1617 {
1618         struct reiserfs_transaction_handle th;
1619         int jbegin_count = 1;
1620
1621         if (inode->i_sb->s_flags & MS_RDONLY)
1622                 return -EROFS;
1623         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1624          ** these cases are just when the system needs ram, not when the
1625          ** inode needs to reach disk for safety, and they can safely be
1626          ** ignored because the altered inode has already been logged.
1627          */
1628         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1629                 reiserfs_write_lock(inode->i_sb);
1630                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1631                         reiserfs_update_sd(&th, inode);
1632                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1633                 }
1634                 reiserfs_write_unlock(inode->i_sb);
1635         }
1636         return 0;
1637 }
1638
1639 /* stat data of new object is inserted already, this inserts the item
1640    containing "." and ".." entries */
1641 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1642                                   struct inode *inode,
1643                                   struct item_head *ih, struct treepath *path,
1644                                   struct inode *dir)
1645 {
1646         struct super_block *sb = th->t_super;
1647         char empty_dir[EMPTY_DIR_SIZE];
1648         char *body = empty_dir;
1649         struct cpu_key key;
1650         int retval;
1651
1652         BUG_ON(!th->t_trans_id);
1653
1654         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1655                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1656                       TYPE_DIRENTRY, 3 /*key length */ );
1657
1658         /* compose item head for new item. Directories consist of items of
1659            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1660            is done by reiserfs_new_inode */
1661         if (old_format_only(sb)) {
1662                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1663                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1664
1665                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1666                                        ih->ih_key.k_objectid,
1667                                        INODE_PKEY(dir)->k_dir_id,
1668                                        INODE_PKEY(dir)->k_objectid);
1669         } else {
1670                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1671                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1672
1673                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1674                                     ih->ih_key.k_objectid,
1675                                     INODE_PKEY(dir)->k_dir_id,
1676                                     INODE_PKEY(dir)->k_objectid);
1677         }
1678
1679         /* look for place in the tree for new item */
1680         retval = search_item(sb, &key, path);
1681         if (retval == IO_ERROR) {
1682                 reiserfs_error(sb, "vs-13080",
1683                                "i/o failure occurred creating new directory");
1684                 return -EIO;
1685         }
1686         if (retval == ITEM_FOUND) {
1687                 pathrelse(path);
1688                 reiserfs_warning(sb, "vs-13070",
1689                                  "object with this key exists (%k)",
1690                                  &(ih->ih_key));
1691                 return -EEXIST;
1692         }
1693
1694         /* insert item, that is empty directory item */
1695         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1696 }
1697
1698 /* stat data of object has been inserted, this inserts the item
1699    containing the body of symlink */
1700 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1701                                 struct item_head *ih,
1702                                 struct treepath *path, const char *symname,
1703                                 int item_len)
1704 {
1705         struct super_block *sb = th->t_super;
1706         struct cpu_key key;
1707         int retval;
1708
1709         BUG_ON(!th->t_trans_id);
1710
1711         _make_cpu_key(&key, KEY_FORMAT_3_5,
1712                       le32_to_cpu(ih->ih_key.k_dir_id),
1713                       le32_to_cpu(ih->ih_key.k_objectid),
1714                       1, TYPE_DIRECT, 3 /*key length */ );
1715
1716         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1717                           0 /*free_space */ );
1718
1719         /* look for place in the tree for new item */
1720         retval = search_item(sb, &key, path);
1721         if (retval == IO_ERROR) {
1722                 reiserfs_error(sb, "vs-13080",
1723                                "i/o failure occurred creating new symlink");
1724                 return -EIO;
1725         }
1726         if (retval == ITEM_FOUND) {
1727                 pathrelse(path);
1728                 reiserfs_warning(sb, "vs-13080",
1729                                  "object with this key exists (%k)",
1730                                  &(ih->ih_key));
1731                 return -EEXIST;
1732         }
1733
1734         /* insert item, that is body of symlink */
1735         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1736 }
1737
1738 /* inserts the stat data into the tree, and then calls
1739    reiserfs_new_directory (to insert ".", ".." item if new object is
1740    directory) or reiserfs_new_symlink (to insert symlink body if new
1741    object is symlink) or nothing (if new object is regular file)
1742
1743    NOTE! uid and gid must already be set in the inode.  If we return
1744    non-zero due to an error, we have to drop the quota previously allocated
1745    for the fresh inode.  This can only be done outside a transaction, so
1746    if we return non-zero, we also end the transaction.  */
1747 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1748                        struct inode *dir, int mode, const char *symname,
1749                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1750                           strlen (symname) for symlinks) */
1751                        loff_t i_size, struct dentry *dentry,
1752                        struct inode *inode,
1753                        struct reiserfs_security_handle *security)
1754 {
1755         struct super_block *sb;
1756         struct reiserfs_iget_args args;
1757         INITIALIZE_PATH(path_to_key);
1758         struct cpu_key key;
1759         struct item_head ih;
1760         struct stat_data sd;
1761         int retval;
1762         int err;
1763
1764         BUG_ON(!th->t_trans_id);
1765
1766         if (vfs_dq_alloc_inode(inode)) {
1767                 err = -EDQUOT;
1768                 goto out_end_trans;
1769         }
1770         if (!dir->i_nlink) {
1771                 err = -EPERM;
1772                 goto out_bad_inode;
1773         }
1774
1775         sb = dir->i_sb;
1776
1777         /* item head of new item */
1778         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1779         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1780         if (!ih.ih_key.k_objectid) {
1781                 err = -ENOMEM;
1782                 goto out_bad_inode;
1783         }
1784         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1785         if (old_format_only(sb))
1786                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1787                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1788         else
1789                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1790                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1791         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1792         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1793         if (insert_inode_locked4(inode, args.objectid,
1794                              reiserfs_find_actor, &args) < 0) {
1795                 err = -EINVAL;
1796                 goto out_bad_inode;
1797         }
1798         if (old_format_only(sb))
1799                 /* not a perfect generation count, as object ids can be reused, but
1800                  ** this is as good as reiserfs can do right now.
1801                  ** note that the private part of inode isn't filled in yet, we have
1802                  ** to use the directory.
1803                  */
1804                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1805         else
1806 #if defined( USE_INODE_GENERATION_COUNTER )
1807                 inode->i_generation =
1808                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1809 #else
1810                 inode->i_generation = ++event;
1811 #endif
1812
1813         /* fill stat data */
1814         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1815
1816         /* uid and gid must already be set by the caller for quota init */
1817
1818         /* symlink cannot be immutable or append only, right? */
1819         if (S_ISLNK(inode->i_mode))
1820                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1821
1822         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1823         inode->i_size = i_size;
1824         inode->i_blocks = 0;
1825         inode->i_bytes = 0;
1826         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1827             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1828
1829         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1830         REISERFS_I(inode)->i_flags = 0;
1831         REISERFS_I(inode)->i_prealloc_block = 0;
1832         REISERFS_I(inode)->i_prealloc_count = 0;
1833         REISERFS_I(inode)->i_trans_id = 0;
1834         REISERFS_I(inode)->i_jl = NULL;
1835         REISERFS_I(inode)->i_attrs =
1836             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1837         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1838         mutex_init(&(REISERFS_I(inode)->i_mmap));
1839         reiserfs_init_xattr_rwsem(inode);
1840
1841         /* key to search for correct place for new stat data */
1842         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1843                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1844                       TYPE_STAT_DATA, 3 /*key length */ );
1845
1846         /* find proper place for inserting of stat data */
1847         retval = search_item(sb, &key, &path_to_key);
1848         if (retval == IO_ERROR) {
1849                 err = -EIO;
1850                 goto out_bad_inode;
1851         }
1852         if (retval == ITEM_FOUND) {
1853                 pathrelse(&path_to_key);
1854                 err = -EEXIST;
1855                 goto out_bad_inode;
1856         }
1857         if (old_format_only(sb)) {
1858                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1859                         pathrelse(&path_to_key);
1860                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1861                         err = -EINVAL;
1862                         goto out_bad_inode;
1863                 }
1864                 inode2sd_v1(&sd, inode, inode->i_size);
1865         } else {
1866                 inode2sd(&sd, inode, inode->i_size);
1867         }
1868         // store in in-core inode the key of stat data and version all
1869         // object items will have (directory items will have old offset
1870         // format, other new objects will consist of new items)
1871         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1872                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1873         else
1874                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1875         if (old_format_only(sb))
1876                 set_inode_sd_version(inode, STAT_DATA_V1);
1877         else
1878                 set_inode_sd_version(inode, STAT_DATA_V2);
1879
1880         /* insert the stat data into the tree */
1881 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1882         if (REISERFS_I(dir)->new_packing_locality)
1883                 th->displace_new_blocks = 1;
1884 #endif
1885         retval =
1886             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1887                                  (char *)(&sd));
1888         if (retval) {
1889                 err = retval;
1890                 reiserfs_check_path(&path_to_key);
1891                 goto out_bad_inode;
1892         }
1893 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1894         if (!th->displace_new_blocks)
1895                 REISERFS_I(dir)->new_packing_locality = 0;
1896 #endif
1897         if (S_ISDIR(mode)) {
1898                 /* insert item with "." and ".." */
1899                 retval =
1900                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1901         }
1902
1903         if (S_ISLNK(mode)) {
1904                 /* insert body of symlink */
1905                 if (!old_format_only(sb))
1906                         i_size = ROUND_UP(i_size);
1907                 retval =
1908                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1909                                          i_size);
1910         }
1911         if (retval) {
1912                 err = retval;
1913                 reiserfs_check_path(&path_to_key);
1914                 journal_end(th, th->t_super, th->t_blocks_allocated);
1915                 goto out_inserted_sd;
1916         }
1917
1918         if (reiserfs_posixacl(inode->i_sb)) {
1919                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1920                 if (retval) {
1921                         err = retval;
1922                         reiserfs_check_path(&path_to_key);
1923                         journal_end(th, th->t_super, th->t_blocks_allocated);
1924                         goto out_inserted_sd;
1925                 }
1926         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1927                 reiserfs_warning(inode->i_sb, "jdm-13090",
1928                                  "ACLs aren't enabled in the fs, "
1929                                  "but vfs thinks they are!");
1930         } else if (IS_PRIVATE(dir))
1931                 inode->i_flags |= S_PRIVATE;
1932
1933         if (security->name) {
1934                 retval = reiserfs_security_write(th, inode, security);
1935                 if (retval) {
1936                         err = retval;
1937                         reiserfs_check_path(&path_to_key);
1938                         retval = journal_end(th, th->t_super,
1939                                              th->t_blocks_allocated);
1940                         if (retval)
1941                                 err = retval;
1942                         goto out_inserted_sd;
1943                 }
1944         }
1945
1946         reiserfs_update_sd(th, inode);
1947         reiserfs_check_path(&path_to_key);
1948
1949         return 0;
1950
1951 /* it looks like you can easily compress these two goto targets into
1952  * one.  Keeping it like this doesn't actually hurt anything, and they
1953  * are place holders for what the quota code actually needs.
1954  */
1955       out_bad_inode:
1956         /* Invalidate the object, nothing was inserted yet */
1957         INODE_PKEY(inode)->k_objectid = 0;
1958
1959         /* Quota change must be inside a transaction for journaling */
1960         vfs_dq_free_inode(inode);
1961
1962       out_end_trans:
1963         journal_end(th, th->t_super, th->t_blocks_allocated);
1964         /* Drop can be outside and it needs more credits so it's better to have it outside */
1965         vfs_dq_drop(inode);
1966         inode->i_flags |= S_NOQUOTA;
1967         make_bad_inode(inode);
1968
1969       out_inserted_sd:
1970         inode->i_nlink = 0;
1971         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1972         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1973         iput(inode);
1974         return err;
1975 }
1976
1977 /*
1978 ** finds the tail page in the page cache,
1979 ** reads the last block in.
1980 **
1981 ** On success, page_result is set to a locked, pinned page, and bh_result
1982 ** is set to an up to date buffer for the last block in the file.  returns 0.
1983 **
1984 ** tail conversion is not done, so bh_result might not be valid for writing
1985 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1986 ** trying to write the block.
1987 **
1988 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1989 */
1990 static int grab_tail_page(struct inode *inode,
1991                           struct page **page_result,
1992                           struct buffer_head **bh_result)
1993 {
1994
1995         /* we want the page with the last byte in the file,
1996          ** not the page that will hold the next byte for appending
1997          */
1998         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1999         unsigned long pos = 0;
2000         unsigned long start = 0;
2001         unsigned long blocksize = inode->i_sb->s_blocksize;
2002         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2003         struct buffer_head *bh;
2004         struct buffer_head *head;
2005         struct page *page;
2006         int error;
2007
2008         /* we know that we are only called with inode->i_size > 0.
2009          ** we also know that a file tail can never be as big as a block
2010          ** If i_size % blocksize == 0, our file is currently block aligned
2011          ** and it won't need converting or zeroing after a truncate.
2012          */
2013         if ((offset & (blocksize - 1)) == 0) {
2014                 return -ENOENT;
2015         }
2016         page = grab_cache_page(inode->i_mapping, index);
2017         error = -ENOMEM;
2018         if (!page) {
2019                 goto out;
2020         }
2021         /* start within the page of the last block in the file */
2022         start = (offset / blocksize) * blocksize;
2023
2024         error = block_prepare_write(page, start, offset,
2025                                     reiserfs_get_block_create_0);
2026         if (error)
2027                 goto unlock;
2028
2029         head = page_buffers(page);
2030         bh = head;
2031         do {
2032                 if (pos >= start) {
2033                         break;
2034                 }
2035                 bh = bh->b_this_page;
2036                 pos += blocksize;
2037         } while (bh != head);
2038
2039         if (!buffer_uptodate(bh)) {
2040                 /* note, this should never happen, prepare_write should
2041                  ** be taking care of this for us.  If the buffer isn't up to date,
2042                  ** I've screwed up the code to find the buffer, or the code to
2043                  ** call prepare_write
2044                  */
2045                 reiserfs_error(inode->i_sb, "clm-6000",
2046                                "error reading block %lu", bh->b_blocknr);
2047                 error = -EIO;
2048                 goto unlock;
2049         }
2050         *bh_result = bh;
2051         *page_result = page;
2052
2053       out:
2054         return error;
2055
2056       unlock:
2057         unlock_page(page);
2058         page_cache_release(page);
2059         return error;
2060 }
2061
2062 /*
2063 ** vfs version of truncate file.  Must NOT be called with
2064 ** a transaction already started.
2065 **
2066 ** some code taken from block_truncate_page
2067 */
2068 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2069 {
2070         struct reiserfs_transaction_handle th;
2071         /* we want the offset for the first byte after the end of the file */
2072         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2073         unsigned blocksize = inode->i_sb->s_blocksize;
2074         unsigned length;
2075         struct page *page = NULL;
2076         int error;
2077         struct buffer_head *bh = NULL;
2078         int err2;
2079         int lock_depth;
2080
2081         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2082
2083         if (inode->i_size > 0) {
2084                 error = grab_tail_page(inode, &page, &bh);
2085                 if (error) {
2086                         // -ENOENT means we truncated past the end of the file,
2087                         // and get_block_create_0 could not find a block to read in,
2088                         // which is ok.
2089                         if (error != -ENOENT)
2090                                 reiserfs_error(inode->i_sb, "clm-6001",
2091                                                "grab_tail_page failed %d",
2092                                                error);
2093                         page = NULL;
2094                         bh = NULL;
2095                 }
2096         }
2097
2098         /* so, if page != NULL, we have a buffer head for the offset at
2099          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2100          ** then we have an unformatted node.  Otherwise, we have a direct item,
2101          ** and no zeroing is required on disk.  We zero after the truncate,
2102          ** because the truncate might pack the item anyway
2103          ** (it will unmap bh if it packs).
2104          */
2105         /* it is enough to reserve space in transaction for 2 balancings:
2106            one for "save" link adding and another for the first
2107            cut_from_item. 1 is for update_sd */
2108         error = journal_begin(&th, inode->i_sb,
2109                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2110         if (error)
2111                 goto out;
2112         reiserfs_update_inode_transaction(inode);
2113         if (update_timestamps)
2114                 /* we are doing real truncate: if the system crashes before the last
2115                    transaction of truncating gets committed - on reboot the file
2116                    either appears truncated properly or not truncated at all */
2117                 add_save_link(&th, inode, 1);
2118         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2119         error =
2120             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2121         if (error)
2122                 goto out;
2123
2124         /* check reiserfs_do_truncate after ending the transaction */
2125         if (err2) {
2126                 error = err2;
2127                 goto out;
2128         }
2129
2130         if (update_timestamps) {
2131                 error = remove_save_link(inode, 1 /* truncate */);
2132                 if (error)
2133                         goto out;
2134         }
2135
2136         if (page) {
2137                 length = offset & (blocksize - 1);
2138                 /* if we are not on a block boundary */
2139                 if (length) {
2140                         length = blocksize - length;
2141                         zero_user(page, offset, length);
2142                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2143                                 mark_buffer_dirty(bh);
2144                         }
2145                 }
2146                 unlock_page(page);
2147                 page_cache_release(page);
2148         }
2149
2150         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2151
2152         return 0;
2153       out:
2154         if (page) {
2155                 unlock_page(page);
2156                 page_cache_release(page);
2157         }
2158
2159         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2160
2161         return error;
2162 }
2163
2164 static int map_block_for_writepage(struct inode *inode,
2165                                    struct buffer_head *bh_result,
2166                                    unsigned long block)
2167 {
2168         struct reiserfs_transaction_handle th;
2169         int fs_gen;
2170         struct item_head tmp_ih;
2171         struct item_head *ih;
2172         struct buffer_head *bh;
2173         __le32 *item;
2174         struct cpu_key key;
2175         INITIALIZE_PATH(path);
2176         int pos_in_item;
2177         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2178         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2179         int retval;
2180         int use_get_block = 0;
2181         int bytes_copied = 0;
2182         int copy_size;
2183         int trans_running = 0;
2184
2185         /* catch places below that try to log something without starting a trans */
2186         th.t_trans_id = 0;
2187
2188         if (!buffer_uptodate(bh_result)) {
2189                 return -EIO;
2190         }
2191
2192         kmap(bh_result->b_page);
2193       start_over:
2194         reiserfs_write_lock(inode->i_sb);
2195         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2196
2197       research:
2198         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2199         if (retval != POSITION_FOUND) {
2200                 use_get_block = 1;
2201                 goto out;
2202         }
2203
2204         bh = get_last_bh(&path);
2205         ih = get_ih(&path);
2206         item = get_item(&path);
2207         pos_in_item = path.pos_in_item;
2208
2209         /* we've found an unformatted node */
2210         if (indirect_item_found(retval, ih)) {
2211                 if (bytes_copied > 0) {
2212                         reiserfs_warning(inode->i_sb, "clm-6002",
2213                                          "bytes_copied %d", bytes_copied);
2214                 }
2215                 if (!get_block_num(item, pos_in_item)) {
2216                         /* crap, we are writing to a hole */
2217                         use_get_block = 1;
2218                         goto out;
2219                 }
2220                 set_block_dev_mapped(bh_result,
2221                                      get_block_num(item, pos_in_item), inode);
2222         } else if (is_direct_le_ih(ih)) {
2223                 char *p;
2224                 p = page_address(bh_result->b_page);
2225                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2226                 copy_size = ih_item_len(ih) - pos_in_item;
2227
2228                 fs_gen = get_generation(inode->i_sb);
2229                 copy_item_head(&tmp_ih, ih);
2230
2231                 if (!trans_running) {
2232                         /* vs-3050 is gone, no need to drop the path */
2233                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2234                         if (retval)
2235                                 goto out;
2236                         reiserfs_update_inode_transaction(inode);
2237                         trans_running = 1;
2238                         if (fs_changed(fs_gen, inode->i_sb)
2239                             && item_moved(&tmp_ih, &path)) {
2240                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2241                                                                  bh);
2242                                 goto research;
2243                         }
2244                 }
2245
2246                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2247
2248                 if (fs_changed(fs_gen, inode->i_sb)
2249                     && item_moved(&tmp_ih, &path)) {
2250                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2251                         goto research;
2252                 }
2253
2254                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2255                        copy_size);
2256
2257                 journal_mark_dirty(&th, inode->i_sb, bh);
2258                 bytes_copied += copy_size;
2259                 set_block_dev_mapped(bh_result, 0, inode);
2260
2261                 /* are there still bytes left? */
2262                 if (bytes_copied < bh_result->b_size &&
2263                     (byte_offset + bytes_copied) < inode->i_size) {
2264                         set_cpu_key_k_offset(&key,
2265                                              cpu_key_k_offset(&key) +
2266                                              copy_size);
2267                         goto research;
2268                 }
2269         } else {
2270                 reiserfs_warning(inode->i_sb, "clm-6003",
2271                                  "bad item inode %lu", inode->i_ino);
2272                 retval = -EIO;
2273                 goto out;
2274         }
2275         retval = 0;
2276
2277       out:
2278         pathrelse(&path);
2279         if (trans_running) {
2280                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2281                 if (err)
2282                         retval = err;
2283                 trans_running = 0;
2284         }
2285         reiserfs_write_unlock(inode->i_sb);
2286
2287         /* this is where we fill in holes in the file. */
2288         if (use_get_block) {
2289                 retval = reiserfs_get_block(inode, block, bh_result,
2290                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2291                                             | GET_BLOCK_NO_DANGLE);
2292                 if (!retval) {
2293                         if (!buffer_mapped(bh_result)
2294                             || bh_result->b_blocknr == 0) {
2295                                 /* get_block failed to find a mapped unformatted node. */
2296                                 use_get_block = 0;
2297                                 goto start_over;
2298                         }
2299                 }
2300         }
2301         kunmap(bh_result->b_page);
2302
2303         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2304                 /* we've copied data from the page into the direct item, so the
2305                  * buffer in the page is now clean, mark it to reflect that.
2306                  */
2307                 lock_buffer(bh_result);
2308                 clear_buffer_dirty(bh_result);
2309                 unlock_buffer(bh_result);
2310         }
2311         return retval;
2312 }
2313
2314 /*
2315  * mason@suse.com: updated in 2.5.54 to follow the same general io
2316  * start/recovery path as __block_write_full_page, along with special
2317  * code to handle reiserfs tails.
2318  */
2319 static int reiserfs_write_full_page(struct page *page,
2320                                     struct writeback_control *wbc)
2321 {
2322         struct inode *inode = page->mapping->host;
2323         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2324         int error = 0;
2325         unsigned long block;
2326         sector_t last_block;
2327         struct buffer_head *head, *bh;
2328         int partial = 0;
2329         int nr = 0;
2330         int checked = PageChecked(page);
2331         struct reiserfs_transaction_handle th;
2332         struct super_block *s = inode->i_sb;
2333         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2334         th.t_trans_id = 0;
2335
2336         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2337         if (checked && (current->flags & PF_MEMALLOC)) {
2338                 redirty_page_for_writepage(wbc, page);
2339                 unlock_page(page);
2340                 return 0;
2341         }
2342
2343         /* The page dirty bit is cleared before writepage is called, which
2344          * means we have to tell create_empty_buffers to make dirty buffers
2345          * The page really should be up to date at this point, so tossing
2346          * in the BH_Uptodate is just a sanity check.
2347          */
2348         if (!page_has_buffers(page)) {
2349                 create_empty_buffers(page, s->s_blocksize,
2350                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2351         }
2352         head = page_buffers(page);
2353
2354         /* last page in the file, zero out any contents past the
2355          ** last byte in the file
2356          */
2357         if (page->index >= end_index) {
2358                 unsigned last_offset;
2359
2360                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2361                 /* no file contents in this page */
2362                 if (page->index >= end_index + 1 || !last_offset) {
2363                         unlock_page(page);
2364                         return 0;
2365                 }
2366                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2367         }
2368         bh = head;
2369         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2370         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2371         /* first map all the buffers, logging any direct items we find */
2372         do {
2373                 if (block > last_block) {
2374                         /*
2375                          * This can happen when the block size is less than
2376                          * the page size.  The corresponding bytes in the page
2377                          * were zero filled above
2378                          */
2379                         clear_buffer_dirty(bh);
2380                         set_buffer_uptodate(bh);
2381                 } else if ((checked || buffer_dirty(bh)) &&
2382                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2383                                                        && bh->b_blocknr ==
2384                                                        0))) {
2385                         /* not mapped yet, or it points to a direct item, search
2386                          * the btree for the mapping info, and log any direct
2387                          * items found
2388                          */
2389                         if ((error = map_block_for_writepage(inode, bh, block))) {
2390                                 goto fail;
2391                         }
2392                 }
2393                 bh = bh->b_this_page;
2394                 block++;
2395         } while (bh != head);
2396
2397         /*
2398          * we start the transaction after map_block_for_writepage,
2399          * because it can create holes in the file (an unbounded operation).
2400          * starting it here, we can make a reliable estimate for how many
2401          * blocks we're going to log
2402          */
2403         if (checked) {
2404                 ClearPageChecked(page);
2405                 reiserfs_write_lock(s);
2406                 error = journal_begin(&th, s, bh_per_page + 1);
2407                 if (error) {
2408                         reiserfs_write_unlock(s);
2409                         goto fail;
2410                 }
2411                 reiserfs_update_inode_transaction(inode);
2412         }
2413         /* now go through and lock any dirty buffers on the page */
2414         do {
2415                 get_bh(bh);
2416                 if (!buffer_mapped(bh))
2417                         continue;
2418                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2419                         continue;
2420
2421                 if (checked) {
2422                         reiserfs_prepare_for_journal(s, bh, 1);
2423                         journal_mark_dirty(&th, s, bh);
2424                         continue;
2425                 }
2426                 /* from this point on, we know the buffer is mapped to a
2427                  * real block and not a direct item
2428                  */
2429                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2430                         lock_buffer(bh);
2431                 } else {
2432                         if (!trylock_buffer(bh)) {
2433                                 redirty_page_for_writepage(wbc, page);
2434                                 continue;
2435                         }
2436                 }
2437                 if (test_clear_buffer_dirty(bh)) {
2438                         mark_buffer_async_write(bh);
2439                 } else {
2440                         unlock_buffer(bh);
2441                 }
2442         } while ((bh = bh->b_this_page) != head);
2443
2444         if (checked) {
2445                 error = journal_end(&th, s, bh_per_page + 1);
2446                 reiserfs_write_unlock(s);
2447                 if (error)
2448                         goto fail;
2449         }
2450         BUG_ON(PageWriteback(page));
2451         set_page_writeback(page);
2452         unlock_page(page);
2453
2454         /*
2455          * since any buffer might be the only dirty buffer on the page,
2456          * the first submit_bh can bring the page out of writeback.
2457          * be careful with the buffers.
2458          */
2459         do {
2460                 struct buffer_head *next = bh->b_this_page;
2461                 if (buffer_async_write(bh)) {
2462                         submit_bh(WRITE, bh);
2463                         nr++;
2464                 }
2465                 put_bh(bh);
2466                 bh = next;
2467         } while (bh != head);
2468
2469         error = 0;
2470       done:
2471         if (nr == 0) {
2472                 /*
2473                  * if this page only had a direct item, it is very possible for
2474                  * no io to be required without there being an error.  Or,
2475                  * someone else could have locked them and sent them down the
2476                  * pipe without locking the page
2477                  */
2478                 bh = head;
2479                 do {
2480                         if (!buffer_uptodate(bh)) {
2481                                 partial = 1;
2482                                 break;
2483                         }
2484                         bh = bh->b_this_page;
2485                 } while (bh != head);
2486                 if (!partial)
2487                         SetPageUptodate(page);
2488                 end_page_writeback(page);
2489         }
2490         return error;
2491
2492       fail:
2493         /* catches various errors, we need to make sure any valid dirty blocks
2494          * get to the media.  The page is currently locked and not marked for
2495          * writeback
2496          */
2497         ClearPageUptodate(page);
2498         bh = head;
2499         do {
2500                 get_bh(bh);
2501                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2502                         lock_buffer(bh);
2503                         mark_buffer_async_write(bh);
2504                 } else {
2505                         /*
2506                          * clear any dirty bits that might have come from getting
2507                          * attached to a dirty page
2508                          */
2509                         clear_buffer_dirty(bh);
2510                 }
2511                 bh = bh->b_this_page;
2512         } while (bh != head);
2513         SetPageError(page);
2514         BUG_ON(PageWriteback(page));
2515         set_page_writeback(page);
2516         unlock_page(page);
2517         do {
2518                 struct buffer_head *next = bh->b_this_page;
2519                 if (buffer_async_write(bh)) {
2520                         clear_buffer_dirty(bh);
2521                         submit_bh(WRITE, bh);
2522                         nr++;
2523                 }
2524                 put_bh(bh);
2525                 bh = next;
2526         } while (bh != head);
2527         goto done;
2528 }
2529
2530 static int reiserfs_readpage(struct file *f, struct page *page)
2531 {
2532         return block_read_full_page(page, reiserfs_get_block);
2533 }
2534
2535 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2536 {
2537         struct inode *inode = page->mapping->host;
2538         reiserfs_wait_on_write_block(inode->i_sb);
2539         return reiserfs_write_full_page(page, wbc);
2540 }
2541
2542 static void reiserfs_truncate_failed_write(struct inode *inode)
2543 {
2544         truncate_inode_pages(inode->i_mapping, inode->i_size);
2545         reiserfs_truncate_file(inode, 0);
2546 }
2547
2548 static int reiserfs_write_begin(struct file *file,
2549                                 struct address_space *mapping,
2550                                 loff_t pos, unsigned len, unsigned flags,
2551                                 struct page **pagep, void **fsdata)
2552 {
2553         struct inode *inode;
2554         struct page *page;
2555         pgoff_t index;
2556         int ret;
2557         int old_ref = 0;
2558
2559         inode = mapping->host;
2560         *fsdata = 0;
2561         if (flags & AOP_FLAG_CONT_EXPAND &&
2562             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2563                 pos ++;
2564                 *fsdata = (void *)(unsigned long)flags;
2565         }
2566
2567         index = pos >> PAGE_CACHE_SHIFT;
2568         page = grab_cache_page_write_begin(mapping, index, flags);
2569         if (!page)
2570                 return -ENOMEM;
2571         *pagep = page;
2572
2573         reiserfs_wait_on_write_block(inode->i_sb);
2574         fix_tail_page_for_writing(page);
2575         if (reiserfs_transaction_running(inode->i_sb)) {
2576                 struct reiserfs_transaction_handle *th;
2577                 th = (struct reiserfs_transaction_handle *)current->
2578                     journal_info;
2579                 BUG_ON(!th->t_refcount);
2580                 BUG_ON(!th->t_trans_id);
2581                 old_ref = th->t_refcount;
2582                 th->t_refcount++;
2583         }
2584         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2585                                 reiserfs_get_block);
2586         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2587                 struct reiserfs_transaction_handle *th = current->journal_info;
2588                 /* this gets a little ugly.  If reiserfs_get_block returned an
2589                  * error and left a transacstion running, we've got to close it,
2590                  * and we've got to free handle if it was a persistent transaction.
2591                  *
2592                  * But, if we had nested into an existing transaction, we need
2593                  * to just drop the ref count on the handle.
2594                  *
2595                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2596                  * and it was a persistent trans.  Otherwise, it was nested above.
2597                  */
2598                 if (th->t_refcount > old_ref) {
2599                         if (old_ref)
2600                                 th->t_refcount--;
2601                         else {
2602                                 int err;
2603                                 reiserfs_write_lock(inode->i_sb);
2604                                 err = reiserfs_end_persistent_transaction(th);
2605                                 reiserfs_write_unlock(inode->i_sb);
2606                                 if (err)
2607                                         ret = err;
2608                         }
2609                 }
2610         }
2611         if (ret) {
2612                 unlock_page(page);
2613                 page_cache_release(page);
2614                 /* Truncate allocated blocks */
2615                 reiserfs_truncate_failed_write(inode);
2616         }
2617         return ret;
2618 }
2619
2620 int reiserfs_prepare_write(struct file *f, struct page *page,
2621                            unsigned from, unsigned to)
2622 {
2623         struct inode *inode = page->mapping->host;
2624         int ret;
2625         int old_ref = 0;
2626
2627         reiserfs_write_unlock(inode->i_sb);
2628         reiserfs_wait_on_write_block(inode->i_sb);
2629         reiserfs_write_lock(inode->i_sb);
2630
2631         fix_tail_page_for_writing(page);
2632         if (reiserfs_transaction_running(inode->i_sb)) {
2633                 struct reiserfs_transaction_handle *th;
2634                 th = (struct reiserfs_transaction_handle *)current->
2635                     journal_info;
2636                 BUG_ON(!th->t_refcount);
2637                 BUG_ON(!th->t_trans_id);
2638                 old_ref = th->t_refcount;
2639                 th->t_refcount++;
2640         }
2641
2642         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2643         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2644                 struct reiserfs_transaction_handle *th = current->journal_info;
2645                 /* this gets a little ugly.  If reiserfs_get_block returned an
2646                  * error and left a transacstion running, we've got to close it,
2647                  * and we've got to free handle if it was a persistent transaction.
2648                  *
2649                  * But, if we had nested into an existing transaction, we need
2650                  * to just drop the ref count on the handle.
2651                  *
2652                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2653                  * and it was a persistent trans.  Otherwise, it was nested above.
2654                  */
2655                 if (th->t_refcount > old_ref) {
2656                         if (old_ref)
2657                                 th->t_refcount--;
2658                         else {
2659                                 int err;
2660                                 reiserfs_write_lock(inode->i_sb);
2661                                 err = reiserfs_end_persistent_transaction(th);
2662                                 reiserfs_write_unlock(inode->i_sb);
2663                                 if (err)
2664                                         ret = err;
2665                         }
2666                 }
2667         }
2668         return ret;
2669
2670 }
2671
2672 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2673 {
2674         return generic_block_bmap(as, block, reiserfs_bmap);
2675 }
2676
2677 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2678                               loff_t pos, unsigned len, unsigned copied,
2679                               struct page *page, void *fsdata)
2680 {
2681         struct inode *inode = page->mapping->host;
2682         int ret = 0;
2683         int update_sd = 0;
2684         struct reiserfs_transaction_handle *th;
2685         unsigned start;
2686         int lock_depth = 0;
2687         bool locked = false;
2688
2689         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2690                 pos ++;
2691
2692         reiserfs_wait_on_write_block(inode->i_sb);
2693         if (reiserfs_transaction_running(inode->i_sb))
2694                 th = current->journal_info;
2695         else
2696                 th = NULL;
2697
2698         start = pos & (PAGE_CACHE_SIZE - 1);
2699         if (unlikely(copied < len)) {
2700                 if (!PageUptodate(page))
2701                         copied = 0;
2702
2703                 page_zero_new_buffers(page, start + copied, start + len);
2704         }
2705         flush_dcache_page(page);
2706
2707         reiserfs_commit_page(inode, page, start, start + copied);
2708
2709         /* generic_commit_write does this for us, but does not update the
2710          ** transaction tracking stuff when the size changes.  So, we have
2711          ** to do the i_size updates here.
2712          */
2713         if (pos + copied > inode->i_size) {
2714                 struct reiserfs_transaction_handle myth;
2715                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2716                 locked = true;
2717                 /* If the file have grown beyond the border where it
2718                    can have a tail, unmark it as needing a tail
2719                    packing */
2720                 if ((have_large_tails(inode->i_sb)
2721                      && inode->i_size > i_block_size(inode) * 4)
2722                     || (have_small_tails(inode->i_sb)
2723                         && inode->i_size > i_block_size(inode)))
2724                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2725
2726                 ret = journal_begin(&myth, inode->i_sb, 1);
2727                 if (ret)
2728                         goto journal_error;
2729
2730                 reiserfs_update_inode_transaction(inode);
2731                 inode->i_size = pos + copied;
2732                 /*
2733                  * this will just nest into our transaction.  It's important
2734                  * to use mark_inode_dirty so the inode gets pushed around on the
2735                  * dirty lists, and so that O_SYNC works as expected
2736                  */
2737                 mark_inode_dirty(inode);
2738                 reiserfs_update_sd(&myth, inode);
2739                 update_sd = 1;
2740                 ret = journal_end(&myth, inode->i_sb, 1);
2741                 if (ret)
2742                         goto journal_error;
2743         }
2744         if (th) {
2745                 if (!locked) {
2746                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2747                         locked = true;
2748                 }
2749                 if (!update_sd)
2750                         mark_inode_dirty(inode);
2751                 ret = reiserfs_end_persistent_transaction(th);
2752                 if (ret)
2753                         goto out;
2754         }
2755
2756       out:
2757         if (locked)
2758                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2759         unlock_page(page);
2760         page_cache_release(page);
2761
2762         if (pos + len > inode->i_size)
2763                 reiserfs_truncate_failed_write(inode);
2764
2765         return ret == 0 ? copied : ret;
2766
2767       journal_error:
2768         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2769         locked = false;
2770         if (th) {
2771                 if (!update_sd)
2772                         reiserfs_update_sd(th, inode);
2773                 ret = reiserfs_end_persistent_transaction(th);
2774         }
2775         goto out;
2776 }
2777
2778 int reiserfs_commit_write(struct file *f, struct page *page,
2779                           unsigned from, unsigned to)
2780 {
2781         struct inode *inode = page->mapping->host;
2782         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2783         int ret = 0;
2784         int update_sd = 0;
2785         struct reiserfs_transaction_handle *th = NULL;
2786
2787         reiserfs_write_unlock(inode->i_sb);
2788         reiserfs_wait_on_write_block(inode->i_sb);
2789         reiserfs_write_lock(inode->i_sb);
2790
2791         if (reiserfs_transaction_running(inode->i_sb)) {
2792                 th = current->journal_info;
2793         }
2794         reiserfs_commit_page(inode, page, from, to);
2795
2796         /* generic_commit_write does this for us, but does not update the
2797          ** transaction tracking stuff when the size changes.  So, we have
2798          ** to do the i_size updates here.
2799          */
2800         if (pos > inode->i_size) {
2801                 struct reiserfs_transaction_handle myth;
2802                 /* If the file have grown beyond the border where it
2803                    can have a tail, unmark it as needing a tail
2804                    packing */
2805                 if ((have_large_tails(inode->i_sb)
2806                      && inode->i_size > i_block_size(inode) * 4)
2807                     || (have_small_tails(inode->i_sb)
2808                         && inode->i_size > i_block_size(inode)))
2809                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2810
2811                 ret = journal_begin(&myth, inode->i_sb, 1);
2812                 if (ret)
2813                         goto journal_error;
2814
2815                 reiserfs_update_inode_transaction(inode);
2816                 inode->i_size = pos;
2817                 /*
2818                  * this will just nest into our transaction.  It's important
2819                  * to use mark_inode_dirty so the inode gets pushed around on the
2820                  * dirty lists, and so that O_SYNC works as expected
2821                  */
2822                 mark_inode_dirty(inode);
2823                 reiserfs_update_sd(&myth, inode);
2824                 update_sd = 1;
2825                 ret = journal_end(&myth, inode->i_sb, 1);
2826                 if (ret)
2827                         goto journal_error;
2828         }
2829         if (th) {
2830                 if (!update_sd)
2831                         mark_inode_dirty(inode);
2832                 ret = reiserfs_end_persistent_transaction(th);
2833                 if (ret)
2834                         goto out;
2835         }
2836
2837       out:
2838         return ret;
2839
2840       journal_error:
2841         if (th) {
2842                 if (!update_sd)
2843                         reiserfs_update_sd(th, inode);
2844                 ret = reiserfs_end_persistent_transaction(th);
2845         }
2846
2847         return ret;
2848 }
2849
2850 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2851 {
2852         if (reiserfs_attrs(inode->i_sb)) {
2853                 if (sd_attrs & REISERFS_SYNC_FL)
2854                         inode->i_flags |= S_SYNC;
2855                 else
2856                         inode->i_flags &= ~S_SYNC;
2857                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2858                         inode->i_flags |= S_IMMUTABLE;
2859                 else
2860                         inode->i_flags &= ~S_IMMUTABLE;
2861                 if (sd_attrs & REISERFS_APPEND_FL)
2862                         inode->i_flags |= S_APPEND;
2863                 else
2864                         inode->i_flags &= ~S_APPEND;
2865                 if (sd_attrs & REISERFS_NOATIME_FL)
2866                         inode->i_flags |= S_NOATIME;
2867                 else
2868                         inode->i_flags &= ~S_NOATIME;
2869                 if (sd_attrs & REISERFS_NOTAIL_FL)
2870                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2871                 else
2872                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2873         }
2874 }
2875
2876 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2877 {
2878         if (reiserfs_attrs(inode->i_sb)) {
2879                 if (inode->i_flags & S_IMMUTABLE)
2880                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2881                 else
2882                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2883                 if (inode->i_flags & S_SYNC)
2884                         *sd_attrs |= REISERFS_SYNC_FL;
2885                 else
2886                         *sd_attrs &= ~REISERFS_SYNC_FL;
2887                 if (inode->i_flags & S_NOATIME)
2888                         *sd_attrs |= REISERFS_NOATIME_FL;
2889                 else
2890                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2891                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2892                         *sd_attrs |= REISERFS_NOTAIL_FL;
2893                 else
2894                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2895         }
2896 }
2897
2898 /* decide if this buffer needs to stay around for data logging or ordered
2899 ** write purposes
2900 */
2901 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2902 {
2903         int ret = 1;
2904         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2905
2906         lock_buffer(bh);
2907         spin_lock(&j->j_dirty_buffers_lock);
2908         if (!buffer_mapped(bh)) {
2909                 goto free_jh;
2910         }
2911         /* the page is locked, and the only places that log a data buffer
2912          * also lock the page.
2913          */
2914         if (reiserfs_file_data_log(inode)) {
2915                 /*
2916                  * very conservative, leave the buffer pinned if
2917                  * anyone might need it.
2918                  */
2919                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2920                         ret = 0;
2921                 }
2922         } else  if (buffer_dirty(bh)) {
2923                 struct reiserfs_journal_list *jl;
2924                 struct reiserfs_jh *jh = bh->b_private;
2925
2926                 /* why is this safe?
2927                  * reiserfs_setattr updates i_size in the on disk
2928                  * stat data before allowing vmtruncate to be called.
2929                  *
2930                  * If buffer was put onto the ordered list for this
2931                  * transaction, we know for sure either this transaction
2932                  * or an older one already has updated i_size on disk,
2933                  * and this ordered data won't be referenced in the file
2934                  * if we crash.
2935                  *
2936                  * if the buffer was put onto the ordered list for an older
2937                  * transaction, we need to leave it around
2938                  */
2939                 if (jh && (jl = jh->jl)
2940                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2941                         ret = 0;
2942         }
2943       free_jh:
2944         if (ret && bh->b_private) {
2945                 reiserfs_free_jh(bh);
2946         }
2947         spin_unlock(&j->j_dirty_buffers_lock);
2948         unlock_buffer(bh);
2949         return ret;
2950 }
2951
2952 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2953 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2954 {
2955         struct buffer_head *head, *bh, *next;
2956         struct inode *inode = page->mapping->host;
2957         unsigned int curr_off = 0;
2958         int ret = 1;
2959
2960         BUG_ON(!PageLocked(page));
2961
2962         if (offset == 0)
2963                 ClearPageChecked(page);
2964
2965         if (!page_has_buffers(page))
2966                 goto out;
2967
2968         head = page_buffers(page);
2969         bh = head;
2970         do {
2971                 unsigned int next_off = curr_off + bh->b_size;
2972                 next = bh->b_this_page;
2973
2974                 /*
2975                  * is this block fully invalidated?
2976                  */
2977                 if (offset <= curr_off) {
2978                         if (invalidatepage_can_drop(inode, bh))
2979                                 reiserfs_unmap_buffer(bh);
2980                         else
2981                                 ret = 0;
2982                 }
2983                 curr_off = next_off;
2984                 bh = next;
2985         } while (bh != head);
2986
2987         /*
2988          * We release buffers only if the entire page is being invalidated.
2989          * The get_block cached value has been unconditionally invalidated,
2990          * so real IO is not possible anymore.
2991          */
2992         if (!offset && ret) {
2993                 ret = try_to_release_page(page, 0);
2994                 /* maybe should BUG_ON(!ret); - neilb */
2995         }
2996       out:
2997         return;
2998 }
2999
3000 static int reiserfs_set_page_dirty(struct page *page)
3001 {
3002         struct inode *inode = page->mapping->host;
3003         if (reiserfs_file_data_log(inode)) {
3004                 SetPageChecked(page);
3005                 return __set_page_dirty_nobuffers(page);
3006         }
3007         return __set_page_dirty_buffers(page);
3008 }
3009
3010 /*
3011  * Returns 1 if the page's buffers were dropped.  The page is locked.
3012  *
3013  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3014  * in the buffers at page_buffers(page).
3015  *
3016  * even in -o notail mode, we can't be sure an old mount without -o notail
3017  * didn't create files with tails.
3018  */
3019 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3020 {
3021         struct inode *inode = page->mapping->host;
3022         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3023         struct buffer_head *head;
3024         struct buffer_head *bh;
3025         int ret = 1;
3026
3027         WARN_ON(PageChecked(page));
3028         spin_lock(&j->j_dirty_buffers_lock);
3029         head = page_buffers(page);
3030         bh = head;
3031         do {
3032                 if (bh->b_private) {
3033                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3034                                 reiserfs_free_jh(bh);
3035                         } else {
3036                                 ret = 0;
3037                                 break;
3038                         }
3039                 }
3040                 bh = bh->b_this_page;
3041         } while (bh != head);
3042         if (ret)
3043                 ret = try_to_free_buffers(page);
3044         spin_unlock(&j->j_dirty_buffers_lock);
3045         return ret;
3046 }
3047
3048 /* We thank Mingming Cao for helping us understand in great detail what
3049    to do in this section of the code. */
3050 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3051                                   const struct iovec *iov, loff_t offset,
3052                                   unsigned long nr_segs)
3053 {
3054         struct file *file = iocb->ki_filp;
3055         struct inode *inode = file->f_mapping->host;
3056
3057         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3058                                   offset, nr_segs,
3059                                   reiserfs_get_blocks_direct_io, NULL);
3060 }
3061
3062 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3063 {
3064         struct inode *inode = dentry->d_inode;
3065         int error;
3066         unsigned int ia_valid;
3067
3068         /* must be turned off for recursive notify_change calls */
3069         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3070
3071         reiserfs_write_lock(inode->i_sb);
3072         if (attr->ia_valid & ATTR_SIZE) {
3073                 /* version 2 items will be caught by the s_maxbytes check
3074                  ** done for us in vmtruncate
3075                  */
3076                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3077                     attr->ia_size > MAX_NON_LFS) {
3078                         error = -EFBIG;
3079                         goto out;
3080                 }
3081                 /* fill in hole pointers in the expanding truncate case. */
3082                 if (attr->ia_size > inode->i_size) {
3083                         error = generic_cont_expand_simple(inode, attr->ia_size);
3084                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3085                                 int err;
3086                                 struct reiserfs_transaction_handle th;
3087                                 /* we're changing at most 2 bitmaps, inode + super */
3088                                 err = journal_begin(&th, inode->i_sb, 4);
3089                                 if (!err) {
3090                                         reiserfs_discard_prealloc(&th, inode);
3091                                         err = journal_end(&th, inode->i_sb, 4);
3092                                 }
3093                                 if (err)
3094                                         error = err;
3095                         }
3096                         if (error)
3097                                 goto out;
3098                         /*
3099                          * file size is changed, ctime and mtime are
3100                          * to be updated
3101                          */
3102                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3103                 }
3104         }
3105
3106         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3107              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3108             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3109                 /* stat data of format v3.5 has 16 bit uid and gid */
3110                 error = -EINVAL;
3111                 goto out;
3112         }
3113
3114         error = inode_change_ok(inode, attr);
3115         if (!error) {
3116                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3117                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3118                         error = reiserfs_chown_xattrs(inode, attr);
3119
3120                         if (!error) {
3121                                 struct reiserfs_transaction_handle th;
3122                                 int jbegin_count =
3123                                     2 *
3124                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3125                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3126                                     2;
3127
3128                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3129                                 error =
3130                                     journal_begin(&th, inode->i_sb,
3131                                                   jbegin_count);
3132                                 if (error)
3133                                         goto out;
3134                                 error =
3135                                     vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
3136                                 if (error) {
3137                                         journal_end(&th, inode->i_sb,
3138                                                     jbegin_count);
3139                                         goto out;
3140                                 }
3141                                 /* Update corresponding info in inode so that everything is in
3142                                  * one transaction */
3143                                 if (attr->ia_valid & ATTR_UID)
3144                                         inode->i_uid = attr->ia_uid;
3145                                 if (attr->ia_valid & ATTR_GID)
3146                                         inode->i_gid = attr->ia_gid;
3147                                 mark_inode_dirty(inode);
3148                                 error =
3149                                     journal_end(&th, inode->i_sb, jbegin_count);
3150                         }
3151                 }
3152                 if (!error)
3153                         error = inode_setattr(inode, attr);
3154         }
3155
3156         if (!error && reiserfs_posixacl(inode->i_sb)) {
3157                 if (attr->ia_valid & ATTR_MODE)
3158                         error = reiserfs_acl_chmod(inode);
3159         }
3160
3161       out:
3162         reiserfs_write_unlock(inode->i_sb);
3163         return error;
3164 }
3165
3166 const struct address_space_operations reiserfs_address_space_operations = {
3167         .writepage = reiserfs_writepage,
3168         .readpage = reiserfs_readpage,
3169         .readpages = reiserfs_readpages,
3170         .releasepage = reiserfs_releasepage,
3171         .invalidatepage = reiserfs_invalidatepage,
3172         .sync_page = block_sync_page,
3173         .write_begin = reiserfs_write_begin,
3174         .write_end = reiserfs_write_end,
3175         .bmap = reiserfs_aop_bmap,
3176         .direct_IO = reiserfs_direct_IO,
3177         .set_page_dirty = reiserfs_set_page_dirty,
3178 };