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/smp_lock.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <asm/uaccess.h>
  14 #include <asm/unaligned.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/mpage.h>
  17 #include <linux/writeback.h>
  18 #include <linux/quotaops.h>
  19 #include <linux/vs_dlimit.h>
  20 #include <linux/vs_tag.h>
  21
  22 static int reiserfs_commit_write(struct file *f, struct page *page,
  23                                  unsigned from, unsigned to);
  24 static int reiserfs_prepare_write(struct file *f, struct page *page,
  25                                   unsigned from, unsigned to);
  26
  27 void reiserfs_delete_inode(struct inode *inode)
  28 {
  29         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  30         int jbegin_count =
  31             JOURNAL_PER_BALANCE_CNT * 2 +
  32             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  33         struct reiserfs_transaction_handle th;
  34         int err;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37
  38         reiserfs_write_lock(inode->i_sb);
  39
  40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  42                 reiserfs_delete_xattrs(inode);
  43
  44                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  45                         goto out;
  46                 reiserfs_update_inode_transaction(inode);
  47
  48                 err = reiserfs_delete_object(&th, inode);
  49
  50                 /* Do quota update inside a transaction for journaled quotas. We must do that
  51                  * after delete_object so that quota updates go into the same transaction as
  52                  * stat data deletion */
  53                 if (!err)
  54                         DQUOT_FREE_INODE(inode);
  55                 DLIMIT_FREE_INODE(inode);
  56
  57                 if (journal_end(&th, inode->i_sb, jbegin_count))
  58                         goto out;
  59
  60                 /* check return value from reiserfs_delete_object after
  61                  * ending the transaction
  62                  */
  63                 if (err)
  64                     goto out;
  65
  66                 /* all items of file are deleted, so we can remove "save" link */
  67                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  68                                                                  * about an error here */
  69         } else {
  70                 /* no object items are in the tree */
  71                 ;
  72         }
  73       out:
  74         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  75         inode->i_blocks = 0;
  76         reiserfs_write_unlock(inode->i_sb);
  77 }
  78
  79 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  80                           __u32 objectid, loff_t offset, int type, int length)
  81 {
  82         key->version = version;
  83
  84         key->on_disk_key.k_dir_id = dirid;
  85         key->on_disk_key.k_objectid = objectid;
  86         set_cpu_key_k_offset(key, offset);
  87         set_cpu_key_k_type(key, type);
  88         key->key_length = length;
  89 }
  90
  91 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  92    offset and type of key */
  93 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  94                   int type, int length)
  95 {
  96         _make_cpu_key(key, get_inode_item_key_version(inode),
  97                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
  98                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
  99                       length);
 100 }
 101
 102 //
 103 // when key is 0, do not set version and short key
 104 //
 105 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 106                               int version,
 107                               loff_t offset, int type, int length,
 108                               int entry_count /*or ih_free_space */ )
 109 {
 110         if (key) {
 111                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 112                 ih->ih_key.k_objectid =
 113                     cpu_to_le32(key->on_disk_key.k_objectid);
 114         }
 115         put_ih_version(ih, version);
 116         set_le_ih_k_offset(ih, offset);
 117         set_le_ih_k_type(ih, type);
 118         put_ih_item_len(ih, length);
 119         /*    set_ih_free_space (ih, 0); */
 120         // for directory items it is entry count, for directs and stat
 121         // datas - 0xffff, for indirects - 0
 122         put_ih_entry_count(ih, entry_count);
 123 }
 124
 125 //
 126 // FIXME: we might cache recently accessed indirect item
 127
 128 // Ugh.  Not too eager for that....
 129 //  I cut the code until such time as I see a convincing argument (benchmark).
 130 // I don't want a bloated inode struct..., and I don't like code complexity....
 131
 132 /* cutting the code is fine, since it really isn't in use yet and is easy
 133 ** to add back in.  But, Vladimir has a really good idea here.  Think
 134 ** about what happens for reading a file.  For each page,
 135 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 136 ** an indirect item.  This indirect item has X number of pointers, where
 137 ** X is a big number if we've done the block allocation right.  But,
 138 ** we only use one or two of these pointers during each call to readpage,
 139 ** needlessly researching again later on.
 140 **
 141 ** The size of the cache could be dynamic based on the size of the file.
 142 **
 143 ** I'd also like to see us cache the location the stat data item, since
 144 ** we are needlessly researching for that frequently.
 145 **
 146 ** --chris
 147 */
 148
 149 /* If this page has a file tail in it, and
 150 ** it was read in by get_block_create_0, the page data is valid,
 151 ** but tail is still sitting in a direct item, and we can't write to
 152 ** it.  So, look through this page, and check all the mapped buffers
 153 ** to make sure they have valid block numbers.  Any that don't need
 154 ** to be unmapped, so that block_prepare_write will correctly call
 155 ** reiserfs_get_block to convert the tail into an unformatted node
 156 */
 157 static inline void fix_tail_page_for_writing(struct page *page)
 158 {
 159         struct buffer_head *head, *next, *bh;
 160
 161         if (page && page_has_buffers(page)) {
 162                 head = page_buffers(page);
 163                 bh = head;
 164                 do {
 165                         next = bh->b_this_page;
 166                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 167                                 reiserfs_unmap_buffer(bh);
 168                         }
 169                         bh = next;
 170                 } while (bh != head);
 171         }
 172 }
 173
 174 /* reiserfs_get_block does not need to allocate a block only if it has been
 175    done already or non-hole position has been found in the indirect item */
 176 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 177                                     struct item_head *ih,
 178                                     __le32 * item, int pos_in_item)
 179 {
 180         if (allocated)
 181                 return 0;
 182         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 183             get_block_num(item, pos_in_item))
 184                 return 0;
 185         return 1;
 186 }
 187
 188 static inline int indirect_item_found(int retval, struct item_head *ih)
 189 {
 190         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 191 }
 192
 193 static inline void set_block_dev_mapped(struct buffer_head *bh,
 194                                         b_blocknr_t block, struct inode *inode)
 195 {
 196         map_bh(bh, inode->i_sb, block);
 197 }
 198
 199 //
 200 // files which were created in the earlier version can not be longer,
 201 // than 2 gb
 202 //
 203 static int file_capable(struct inode *inode, long block)
 204 {
 205         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 206             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 207                 return 1;
 208
 209         return 0;
 210 }
 211
 212 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
 213                                    struct inode *inode, struct treepath *path)
 214 {
 215         struct super_block *s = th->t_super;
 216         int len = th->t_blocks_allocated;
 217         int err;
 218
 219         BUG_ON(!th->t_trans_id);
 220         BUG_ON(!th->t_refcount);
 221
 222         pathrelse(path);
 223
 224         /* we cannot restart while nested */
 225         if (th->t_refcount > 1) {
 226                 return 0;
 227         }
 228         reiserfs_update_sd(th, inode);
 229         err = journal_end(th, s, len);
 230         if (!err) {
 231                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 232                 if (!err)
 233                         reiserfs_update_inode_transaction(inode);
 234         }
 235         return err;
 236 }
 237
 238 // it is called by get_block when create == 0. Returns block number
 239 // for 'block'-th logical block of file. When it hits direct item it
 240 // returns 0 (being called from bmap) or read direct item into piece
 241 // of page (bh_result)
 242
 243 // Please improve the english/clarity in the comment above, as it is
 244 // hard to understand.
 245
 246 static int _get_block_create_0(struct inode *inode, long block,
 247                                struct buffer_head *bh_result, int args)
 248 {
 249         INITIALIZE_PATH(path);
 250         struct cpu_key key;
 251         struct buffer_head *bh;
 252         struct item_head *ih, tmp_ih;
 253         int fs_gen;
 254         int blocknr;
 255         char *p = NULL;
 256         int chars;
 257         int ret;
 258         int result;
 259         int done = 0;
 260         unsigned long offset;
 261
 262         // prepare the key to look for the 'block'-th block of file
 263         make_cpu_key(&key, inode,
 264                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 265                      3);
 266
 267       research:
 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         fs_gen = get_generation(inode->i_sb);
 343         copy_item_head(&tmp_ih, ih);
 344
 345         /* we only want to kmap if we are reading the tail into the page.
 346          ** this is not the common case, so we don't kmap until we are
 347          ** sure we need to.  But, this means the item might move if
 348          ** kmap schedules
 349          */
 350         if (!p) {
 351                 p = (char *)kmap(bh_result->b_page);
 352                 if (fs_changed(fs_gen, inode->i_sb)
 353                     && item_moved(&tmp_ih, &path)) {
 354                         goto research;
 355                 }
 356         }
 357         p += offset;
 358         memset(p, 0, inode->i_sb->s_blocksize);
 359         do {
 360                 if (!is_direct_le_ih(ih)) {
 361                         BUG();
 362                 }
 363                 /* make sure we don't read more bytes than actually exist in
 364                  ** the file.  This can happen in odd cases where i_size isn't
 365                  ** correct, and when direct item padding results in a few
 366                  ** extra bytes at the end of the direct item
 367                  */
 368                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 369                         break;
 370                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 371                         chars =
 372                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 373                             path.pos_in_item;
 374                         done = 1;
 375                 } else {
 376                         chars = ih_item_len(ih) - path.pos_in_item;
 377                 }
 378                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 379
 380                 if (done)
 381                         break;
 382
 383                 p += chars;
 384
 385                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 386                         // we done, if read direct item is not the last item of
 387                         // node FIXME: we could try to check right delimiting key
 388                         // to see whether direct item continues in the right
 389                         // neighbor or rely on i_size
 390                         break;
 391
 392                 // update key to look for the next piece
 393                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 394                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 395                 if (result != POSITION_FOUND)
 396                         // i/o error most likely
 397                         break;
 398                 bh = get_last_bh(&path);
 399                 ih = get_ih(&path);
 400         } while (1);
 401
 402         flush_dcache_page(bh_result->b_page);
 403         kunmap(bh_result->b_page);
 404
 405       finished:
 406         pathrelse(&path);
 407
 408         if (result == IO_ERROR)
 409                 return -EIO;
 410
 411         /* this buffer has valid data, but isn't valid for io.  mapping it to
 412          * block #0 tells the rest of reiserfs it just has a tail in it
 413          */
 414         map_bh(bh_result, inode->i_sb, 0);
 415         set_buffer_uptodate(bh_result);
 416         return 0;
 417 }
 418
 419 // this is called to create file map. So, _get_block_create_0 will not
 420 // read direct item
 421 static int reiserfs_bmap(struct inode *inode, sector_t block,
 422                          struct buffer_head *bh_result, int create)
 423 {
 424         if (!file_capable(inode, block))
 425                 return -EFBIG;
 426
 427         reiserfs_write_lock(inode->i_sb);
 428         /* do not read the direct item */
 429         _get_block_create_0(inode, block, bh_result, 0);
 430         reiserfs_write_unlock(inode->i_sb);
 431         return 0;
 432 }
 433
 434 /* special version of get_block that is only used by grab_tail_page right
 435 ** now.  It is sent to block_prepare_write, and when you try to get a
 436 ** block past the end of the file (or a block from a hole) it returns
 437 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 438 ** be able to do i/o on the buffers returned, unless an error value
 439 ** is also returned.
 440 **
 441 ** So, this allows block_prepare_write to be used for reading a single block
 442 ** in a page.  Where it does not produce a valid page for holes, or past the
 443 ** end of the file.  This turns out to be exactly what we need for reading
 444 ** tails for conversion.
 445 **
 446 ** The point of the wrapper is forcing a certain value for create, even
 447 ** though the VFS layer is calling this function with create==1.  If you
 448 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 449 ** don't use this function.
 450 */
 451 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 452                                        struct buffer_head *bh_result,
 453                                        int create)
 454 {
 455         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 456 }
 457
 458 /* This is special helper for reiserfs_get_block in case we are executing
 459    direct_IO request. */
 460 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 461                                          sector_t iblock,
 462                                          struct buffer_head *bh_result,
 463                                          int create)
 464 {
 465         int ret;
 466
 467         bh_result->b_page = NULL;
 468
 469         /* We set the b_size before reiserfs_get_block call since it is
 470            referenced in convert_tail_for_hole() that may be called from
 471            reiserfs_get_block() */
 472         bh_result->b_size = (1 << inode->i_blkbits);
 473
 474         ret = reiserfs_get_block(inode, iblock, bh_result,
 475                                  create | GET_BLOCK_NO_DANGLE);
 476         if (ret)
 477                 goto out;
 478
 479         /* don't allow direct io onto tail pages */
 480         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 481                 /* make sure future calls to the direct io funcs for this offset
 482                  ** in the file fail by unmapping the buffer
 483                  */
 484                 clear_buffer_mapped(bh_result);
 485                 ret = -EINVAL;
 486         }
 487         /* Possible unpacked tail. Flush the data before pages have
 488            disappeared */
 489         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 490                 int err;
 491                 lock_kernel();
 492                 err = reiserfs_commit_for_inode(inode);
 493                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 494                 unlock_kernel();
 495                 if (err < 0)
 496                         ret = err;
 497         }
 498       out:
 499         return ret;
 500 }
 501
 502 /*
 503 ** helper function for when reiserfs_get_block is called for a hole
 504 ** but the file tail is still in a direct item
 505 ** bh_result is the buffer head for the hole
 506 ** tail_offset is the offset of the start of the tail in the file
 507 **
 508 ** This calls prepare_write, which will start a new transaction
 509 ** you should not be in a transaction, or have any paths held when you
 510 ** call this.
 511 */
 512 static int convert_tail_for_hole(struct inode *inode,
 513                                  struct buffer_head *bh_result,
 514                                  loff_t tail_offset)
 515 {
 516         unsigned long index;
 517         unsigned long tail_end;
 518         unsigned long tail_start;
 519         struct page *tail_page;
 520         struct page *hole_page = bh_result->b_page;
 521         int retval = 0;
 522
 523         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 524                 return -EIO;
 525
 526         /* always try to read until the end of the block */
 527         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 528         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 529
 530         index = tail_offset >> PAGE_CACHE_SHIFT;
 531         /* hole_page can be zero in case of direct_io, we are sure
 532            that we cannot get here if we write with O_DIRECT into
 533            tail page */
 534         if (!hole_page || index != hole_page->index) {
 535                 tail_page = grab_cache_page(inode->i_mapping, index);
 536                 retval = -ENOMEM;
 537                 if (!tail_page) {
 538                         goto out;
 539                 }
 540         } else {
 541                 tail_page = hole_page;
 542         }
 543
 544         /* we don't have to make sure the conversion did not happen while
 545          ** we were locking the page because anyone that could convert
 546          ** must first take i_mutex.
 547          **
 548          ** We must fix the tail page for writing because it might have buffers
 549          ** that are mapped, but have a block number of 0.  This indicates tail
 550          ** data that has been read directly into the page, and block_prepare_write
 551          ** won't trigger a get_block in this case.
 552          */
 553         fix_tail_page_for_writing(tail_page);
 554         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 555         if (retval)
 556                 goto unlock;
 557
 558         /* tail conversion might change the data in the page */
 559         flush_dcache_page(tail_page);
 560
 561         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 562
 563       unlock:
 564         if (tail_page != hole_page) {
 565                 unlock_page(tail_page);
 566                 page_cache_release(tail_page);
 567         }
 568       out:
 569         return retval;
 570 }
 571
 572 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 573                                   long block,
 574                                   struct inode *inode,
 575                                   b_blocknr_t * allocated_block_nr,
 576                                   struct treepath *path, int flags)
 577 {
 578         BUG_ON(!th->t_trans_id);
 579
 580 #ifdef REISERFS_PREALLOCATE
 581         if (!(flags & GET_BLOCK_NO_IMUX)) {
 582                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 583                                                   path, block);
 584         }
 585 #endif
 586         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 587                                          block);
 588 }
 589
 590 int reiserfs_get_block(struct inode *inode, sector_t block,
 591                        struct buffer_head *bh_result, int create)
 592 {
 593         int repeat, retval = 0;
 594         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 595         INITIALIZE_PATH(path);
 596         int pos_in_item;
 597         struct cpu_key key;
 598         struct buffer_head *bh, *unbh = NULL;
 599         struct item_head *ih, tmp_ih;
 600         __le32 *item;
 601         int done;
 602         int fs_gen;
 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         /* bad.... */
 619         reiserfs_write_lock(inode->i_sb);
 620         version = get_inode_item_key_version(inode);
 621
 622         if (!file_capable(inode, block)) {
 623                 reiserfs_write_unlock(inode->i_sb);
 624                 return -EFBIG;
 625         }
 626
 627         /* if !create, we aren't changing the FS, so we don't need to
 628          ** log anything, so we don't need to start a transaction
 629          */
 630         if (!(create & GET_BLOCK_CREATE)) {
 631                 int ret;
 632                 /* find number of block-th logical block of the file */
 633                 ret = _get_block_create_0(inode, block, bh_result,
 634                                           create | GET_BLOCK_READ_DIRECT);
 635                 reiserfs_write_unlock(inode->i_sb);
 636                 return ret;
 637         }
 638         /*
 639          * if we're already in a transaction, make sure to close
 640          * any new transactions we start in this func
 641          */
 642         if ((create & GET_BLOCK_NO_DANGLE) ||
 643             reiserfs_transaction_running(inode->i_sb))
 644                 dangle = 0;
 645
 646         /* If file is of such a size, that it might have a tail and tails are enabled
 647          ** we should mark it as possibly needing tail packing on close
 648          */
 649         if ((have_large_tails(inode->i_sb)
 650              && inode->i_size < i_block_size(inode) * 4)
 651             || (have_small_tails(inode->i_sb)
 652                 && inode->i_size < i_block_size(inode)))
 653                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 654
 655         /* set the key of the first byte in the 'block'-th block of file */
 656         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 657         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 658               start_trans:
 659                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 660                 if (!th) {
 661                         retval = -ENOMEM;
 662                         goto failure;
 663                 }
 664                 reiserfs_update_inode_transaction(inode);
 665         }
 666       research:
 667
 668         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 669         if (retval == IO_ERROR) {
 670                 retval = -EIO;
 671                 goto failure;
 672         }
 673
 674         bh = get_last_bh(&path);
 675         ih = get_ih(&path);
 676         item = get_item(&path);
 677         pos_in_item = path.pos_in_item;
 678
 679         fs_gen = get_generation(inode->i_sb);
 680         copy_item_head(&tmp_ih, ih);
 681
 682         if (allocation_needed
 683             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 684                 /* we have to allocate block for the unformatted node */
 685                 if (!th) {
 686                         pathrelse(&path);
 687                         goto start_trans;
 688                 }
 689
 690                 repeat =
 691                     _allocate_block(th, block, inode, &allocated_block_nr,
 692                                     &path, create);
 693
 694                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 695                         /* restart the transaction to give the journal a chance to free
 696                          ** some blocks.  releases the path, so we have to go back to
 697                          ** research if we succeed on the second try
 698                          */
 699                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 700                         retval = restart_transaction(th, inode, &path);
 701                         if (retval)
 702                                 goto failure;
 703                         repeat =
 704                             _allocate_block(th, block, inode,
 705                                             &allocated_block_nr, NULL, create);
 706
 707                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 708                                 goto research;
 709                         }
 710                         if (repeat == QUOTA_EXCEEDED)
 711                                 retval = -EDQUOT;
 712                         else
 713                                 retval = -ENOSPC;
 714                         goto failure;
 715                 }
 716
 717                 if (fs_changed(fs_gen, inode->i_sb)
 718                     && item_moved(&tmp_ih, &path)) {
 719                         goto research;
 720                 }
 721         }
 722
 723         if (indirect_item_found(retval, ih)) {
 724                 b_blocknr_t unfm_ptr;
 725                 /* 'block'-th block is in the file already (there is
 726                    corresponding cell in some indirect item). But it may be
 727                    zero unformatted node pointer (hole) */
 728                 unfm_ptr = get_block_num(item, pos_in_item);
 729                 if (unfm_ptr == 0) {
 730                         /* use allocated block to plug the hole */
 731                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 732                         if (fs_changed(fs_gen, inode->i_sb)
 733                             && item_moved(&tmp_ih, &path)) {
 734                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 735                                                                  bh);
 736                                 goto research;
 737                         }
 738                         set_buffer_new(bh_result);
 739                         if (buffer_dirty(bh_result)
 740                             && reiserfs_data_ordered(inode->i_sb))
 741                                 reiserfs_add_ordered_list(inode, bh_result);
 742                         put_block_num(item, pos_in_item, allocated_block_nr);
 743                         unfm_ptr = allocated_block_nr;
 744                         journal_mark_dirty(th, inode->i_sb, bh);
 745                         reiserfs_update_sd(th, inode);
 746                 }
 747                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 748                 pathrelse(&path);
 749                 retval = 0;
 750                 if (!dangle && th)
 751                         retval = reiserfs_end_persistent_transaction(th);
 752
 753                 reiserfs_write_unlock(inode->i_sb);
 754
 755                 /* the item was found, so new blocks were not added to the file
 756                  ** there is no need to make sure the inode is updated with this
 757                  ** transaction
 758                  */
 759                 return retval;
 760         }
 761
 762         if (!th) {
 763                 pathrelse(&path);
 764                 goto start_trans;
 765         }
 766
 767         /* desired position is not found or is in the direct item. We have
 768            to append file with holes up to 'block'-th block converting
 769            direct items to indirect one if necessary */
 770         done = 0;
 771         do {
 772                 if (is_statdata_le_ih(ih)) {
 773                         __le32 unp = 0;
 774                         struct cpu_key tmp_key;
 775
 776                         /* indirect item has to be inserted */
 777                         make_le_item_head(&tmp_ih, &key, version, 1,
 778                                           TYPE_INDIRECT, UNFM_P_SIZE,
 779                                           0 /* free_space */ );
 780
 781                         if (cpu_key_k_offset(&key) == 1) {
 782                                 /* we are going to add 'block'-th block to the file. Use
 783                                    allocated block for that */
 784                                 unp = cpu_to_le32(allocated_block_nr);
 785                                 set_block_dev_mapped(bh_result,
 786                                                      allocated_block_nr, inode);
 787                                 set_buffer_new(bh_result);
 788                                 done = 1;
 789                         }
 790                         tmp_key = key;  // ;)
 791                         set_cpu_key_k_offset(&tmp_key, 1);
 792                         PATH_LAST_POSITION(&path)++;
 793
 794                         retval =
 795                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 796                                                  inode, (char *)&unp);
 797                         if (retval) {
 798                                 reiserfs_free_block(th, inode,
 799                                                     allocated_block_nr, 1);
 800                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 801                         }
 802                         //mark_tail_converted (inode);
 803                 } else if (is_direct_le_ih(ih)) {
 804                         /* direct item has to be converted */
 805                         loff_t tail_offset;
 806
 807                         tail_offset =
 808                             ((le_ih_k_offset(ih) -
 809                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 810                         if (tail_offset == cpu_key_k_offset(&key)) {
 811                                 /* direct item we just found fits into block we have
 812                                    to map. Convert it into unformatted node: use
 813                                    bh_result for the conversion */
 814                                 set_block_dev_mapped(bh_result,
 815                                                      allocated_block_nr, inode);
 816                                 unbh = bh_result;
 817                                 done = 1;
 818                         } else {
 819                                 /* we have to padd file tail stored in direct item(s)
 820                                    up to block size and convert it to unformatted
 821                                    node. FIXME: this should also get into page cache */
 822
 823                                 pathrelse(&path);
 824                                 /*
 825                                  * ugly, but we can only end the transaction if
 826                                  * we aren't nested
 827                                  */
 828                                 BUG_ON(!th->t_refcount);
 829                                 if (th->t_refcount == 1) {
 830                                         retval =
 831                                             reiserfs_end_persistent_transaction
 832                                             (th);
 833                                         th = NULL;
 834                                         if (retval)
 835                                                 goto failure;
 836                                 }
 837
 838                                 retval =
 839                                     convert_tail_for_hole(inode, bh_result,
 840                                                           tail_offset);
 841                                 if (retval) {
 842                                         if (retval != -ENOSPC)
 843                                                 reiserfs_warning(inode->i_sb,
 844                                                                  "clm-6004: convert tail failed inode %lu, error %d",
 845                                                                  inode->i_ino,
 846                                                                  retval);
 847                                         if (allocated_block_nr) {
 848                                                 /* the bitmap, the super, and the stat data == 3 */
 849                                                 if (!th)
 850                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 851                                                 if (th)
 852                                                         reiserfs_free_block(th,
 853                                                                             inode,
 854                                                                             allocated_block_nr,
 855                                                                             1);
 856                                         }
 857                                         goto failure;
 858                                 }
 859                                 goto research;
 860                         }
 861                         retval =
 862                             direct2indirect(th, inode, &path, unbh,
 863                                             tail_offset);
 864                         if (retval) {
 865                                 reiserfs_unmap_buffer(unbh);
 866                                 reiserfs_free_block(th, inode,
 867                                                     allocated_block_nr, 1);
 868                                 goto failure;
 869                         }
 870                         /* it is important the set_buffer_uptodate is done after
 871                          ** the direct2indirect.  The buffer might contain valid
 872                          ** data newer than the data on disk (read by readpage, changed,
 873                          ** and then sent here by writepage).  direct2indirect needs
 874                          ** to know if unbh was already up to date, so it can decide
 875                          ** if the data in unbh needs to be replaced with data from
 876                          ** the disk
 877                          */
 878                         set_buffer_uptodate(unbh);
 879
 880                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 881                            buffer will disappear shortly, so it should not be added to
 882                          */
 883                         if (unbh->b_page) {
 884                                 /* we've converted the tail, so we must
 885                                  ** flush unbh before the transaction commits
 886                                  */
 887                                 reiserfs_add_tail_list(inode, unbh);
 888
 889                                 /* mark it dirty now to prevent commit_write from adding
 890                                  ** this buffer to the inode's dirty buffer list
 891                                  */
 892                                 /*
 893                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 894                                  * It's still atomic, but it sets the page dirty too,
 895                                  * which makes it eligible for writeback at any time by the
 896                                  * VM (which was also the case with __mark_buffer_dirty())
 897                                  */
 898                                 mark_buffer_dirty(unbh);
 899                         }
 900                 } else {
 901                         /* append indirect item with holes if needed, when appending
 902                            pointer to 'block'-th block use block, which is already
 903                            allocated */
 904                         struct cpu_key tmp_key;
 905                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 906                         // one block which is a fastpath
 907                         unp_t *un;
 908                         __u64 max_to_insert =
 909                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 910                             UNFM_P_SIZE;
 911                         __u64 blocks_needed;
 912
 913                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 914                                "vs-804: invalid position for append");
 915                         /* indirect item has to be appended, set up key of that position */
 916                         make_cpu_key(&tmp_key, inode,
 917                                      le_key_k_offset(version,
 918                                                      &(ih->ih_key)) +
 919                                      op_bytes_number(ih,
 920                                                      inode->i_sb->s_blocksize),
 921                                      //pos_in_item * inode->i_sb->s_blocksize,
 922                                      TYPE_INDIRECT, 3); // key type is unimportant
 923
 924                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 925                                "green-805: invalid offset");
 926                         blocks_needed =
 927                             1 +
 928                             ((cpu_key_k_offset(&key) -
 929                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 930                              s_blocksize_bits);
 931
 932                         if (blocks_needed == 1) {
 933                                 un = &unf_single;
 934                         } else {
 935                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 936                                 if (!un) {
 937                                         un = &unf_single;
 938                                         blocks_needed = 1;
 939                                         max_to_insert = 0;
 940                                 }
 941                         }
 942                         if (blocks_needed <= max_to_insert) {
 943                                 /* we are going to add target block to the file. Use allocated
 944                                    block for that */
 945                                 un[blocks_needed - 1] =
 946                                     cpu_to_le32(allocated_block_nr);
 947                                 set_block_dev_mapped(bh_result,
 948                                                      allocated_block_nr, inode);
 949                                 set_buffer_new(bh_result);
 950                                 done = 1;
 951                         } else {
 952                                 /* paste hole to the indirect item */
 953                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 954                                    only have space for one block */
 955                                 blocks_needed =
 956                                     max_to_insert ? max_to_insert : 1;
 957                         }
 958                         retval =
 959                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 960                                                      (char *)un,
 961                                                      UNFM_P_SIZE *
 962                                                      blocks_needed);
 963
 964                         if (blocks_needed != 1)
 965                                 kfree(un);
 966
 967                         if (retval) {
 968                                 reiserfs_free_block(th, inode,
 969                                                     allocated_block_nr, 1);
 970                                 goto failure;
 971                         }
 972                         if (!done) {
 973                                 /* We need to mark new file size in case this function will be
 974                                    interrupted/aborted later on. And we may do this only for
 975                                    holes. */
 976                                 inode->i_size +=
 977                                     inode->i_sb->s_blocksize * blocks_needed;
 978                         }
 979                 }
 980
 981                 if (done == 1)
 982                         break;
 983
 984                 /* this loop could log more blocks than we had originally asked
 985                  ** for.  So, we have to allow the transaction to end if it is
 986                  ** too big or too full.  Update the inode so things are
 987                  ** consistent if we crash before the function returns
 988                  **
 989                  ** release the path so that anybody waiting on the path before
 990                  ** ending their transaction will be able to continue.
 991                  */
 992                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 993                         retval = restart_transaction(th, inode, &path);
 994                         if (retval)
 995                                 goto failure;
 996                 }
 997                 /* inserting indirect pointers for a hole can take a
 998                  ** long time.  reschedule if needed
 999                  */
1000                 cond_resched();
1001
1002                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1003                 if (retval == IO_ERROR) {
1004                         retval = -EIO;
1005                         goto failure;
1006                 }
1007                 if (retval == POSITION_FOUND) {
1008                         reiserfs_warning(inode->i_sb,
1009                                          "vs-825: reiserfs_get_block: "
1010                                          "%K should not be found", &key);
1011                         retval = -EEXIST;
1012                         if (allocated_block_nr)
1013                                 reiserfs_free_block(th, inode,
1014                                                     allocated_block_nr, 1);
1015                         pathrelse(&path);
1016                         goto failure;
1017                 }
1018                 bh = get_last_bh(&path);
1019                 ih = get_ih(&path);
1020                 item = get_item(&path);
1021                 pos_in_item = path.pos_in_item;
1022         } while (1);
1023
1024         retval = 0;
1025
1026       failure:
1027         if (th && (!dangle || (retval && !th->t_trans_id))) {
1028                 int err;
1029                 if (th->t_trans_id)
1030                         reiserfs_update_sd(th, inode);
1031                 err = reiserfs_end_persistent_transaction(th);
1032                 if (err)
1033                         retval = err;
1034         }
1035
1036         reiserfs_write_unlock(inode->i_sb);
1037         reiserfs_check_path(&path);
1038         return retval;
1039 }
1040
1041 static int
1042 reiserfs_readpages(struct file *file, struct address_space *mapping,
1043                    struct list_head *pages, unsigned nr_pages)
1044 {
1045         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1046 }
1047
1048 /* Compute real number of used bytes by file
1049  * Following three functions can go away when we'll have enough space in stat item
1050  */
1051 static int real_space_diff(struct inode *inode, int sd_size)
1052 {
1053         int bytes;
1054         loff_t blocksize = inode->i_sb->s_blocksize;
1055
1056         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1057                 return sd_size;
1058
1059         /* End of file is also in full block with indirect reference, so round
1060          ** up to the next block.
1061          **
1062          ** there is just no way to know if the tail is actually packed
1063          ** on the file, so we have to assume it isn't.  When we pack the
1064          ** tail, we add 4 bytes to pretend there really is an unformatted
1065          ** node pointer
1066          */
1067         bytes =
1068             ((inode->i_size +
1069               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1070             sd_size;
1071         return bytes;
1072 }
1073
1074 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1075                                         int sd_size)
1076 {
1077         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1078                 return inode->i_size +
1079                     (loff_t) (real_space_diff(inode, sd_size));
1080         }
1081         return ((loff_t) real_space_diff(inode, sd_size)) +
1082             (((loff_t) blocks) << 9);
1083 }
1084
1085 /* Compute number of blocks used by file in ReiserFS counting */
1086 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1087 {
1088         loff_t bytes = inode_get_bytes(inode);
1089         loff_t real_space = real_space_diff(inode, sd_size);
1090
1091         /* keeps fsck and non-quota versions of reiserfs happy */
1092         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1093                 bytes += (loff_t) 511;
1094         }
1095
1096         /* files from before the quota patch might i_blocks such that
1097          ** bytes < real_space.  Deal with that here to prevent it from
1098          ** going negative.
1099          */
1100         if (bytes < real_space)
1101                 return 0;
1102         return (bytes - real_space) >> 9;
1103 }
1104
1105 //
1106 // BAD: new directories have stat data of new type and all other items
1107 // of old type. Version stored in the inode says about body items, so
1108 // in update_stat_data we can not rely on inode, but have to check
1109 // item version directly
1110 //
1111
1112 // called by read_locked_inode
1113 static void init_inode(struct inode *inode, struct treepath *path)
1114 {
1115         struct buffer_head *bh;
1116         struct item_head *ih;
1117         __u32 rdev;
1118         uid_t uid;
1119         gid_t gid;
1120         //int version = ITEM_VERSION_1;
1121
1122         bh = PATH_PLAST_BUFFER(path);
1123         ih = PATH_PITEM_HEAD(path);
1124
1125         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1126
1127         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1128         REISERFS_I(inode)->i_flags = 0;
1129         REISERFS_I(inode)->i_prealloc_block = 0;
1130         REISERFS_I(inode)->i_prealloc_count = 0;
1131         REISERFS_I(inode)->i_trans_id = 0;
1132         REISERFS_I(inode)->i_jl = NULL;
1133         mutex_init(&(REISERFS_I(inode)->i_mmap));
1134         reiserfs_init_acl_access(inode);
1135         reiserfs_init_acl_default(inode);
1136         reiserfs_init_xattr_rwsem(inode);
1137
1138         if (stat_data_v1(ih)) {
1139                 struct stat_data_v1 *sd =
1140                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1141                 unsigned long blocks;
1142
1143                 uid = sd_v1_uid(sd);
1144                 gid = sd_v1_gid(sd);
1145
1146                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1147                 set_inode_sd_version(inode, STAT_DATA_V1);
1148                 inode->i_mode = sd_v1_mode(sd);
1149                 inode->i_nlink = sd_v1_nlink(sd);
1150                 inode->i_size = sd_v1_size(sd);
1151                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1152                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1153                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1154                 inode->i_atime.tv_nsec = 0;
1155                 inode->i_ctime.tv_nsec = 0;
1156                 inode->i_mtime.tv_nsec = 0;
1157
1158                 inode->i_blocks = sd_v1_blocks(sd);
1159                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1160                 blocks = (inode->i_size + 511) >> 9;
1161                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1162                 if (inode->i_blocks > blocks) {
1163                         // there was a bug in <=3.5.23 when i_blocks could take negative
1164                         // values. Starting from 3.5.17 this value could even be stored in
1165                         // stat data. For such files we set i_blocks based on file
1166                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1167                         // only updated if file's inode will ever change
1168                         inode->i_blocks = blocks;
1169                 }
1170
1171                 rdev = sd_v1_rdev(sd);
1172                 REISERFS_I(inode)->i_first_direct_byte =
1173                     sd_v1_first_direct_byte(sd);
1174                 /* an early bug in the quota code can give us an odd number for the
1175                  ** block count.  This is incorrect, fix it here.
1176                  */
1177                 if (inode->i_blocks & 1) {
1178                         inode->i_blocks++;
1179                 }
1180                 inode_set_bytes(inode,
1181                                 to_real_used_space(inode, inode->i_blocks,
1182                                                    SD_V1_SIZE));
1183                 /* nopack is initially zero for v1 objects. For v2 objects,
1184                    nopack is initialised from sd_attrs */
1185                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1186         } else {
1187                 // new stat data found, but object may have old items
1188                 // (directories and symlinks)
1189                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1190
1191                 uid    = sd_v2_uid(sd);
1192                 gid    = sd_v2_gid(sd);
1193
1194                 inode->i_mode = sd_v2_mode(sd);
1195                 inode->i_nlink = sd_v2_nlink(sd);
1196                 inode->i_size = sd_v2_size(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         inode->i_uid = INOTAG_UID(DX_TAG(inode), uid, gid);
1227         inode->i_gid = INOTAG_GID(DX_TAG(inode), uid, gid);
1228         inode->i_tag = INOTAG_TAG(DX_TAG(inode), uid, gid, 0);
1229
1230         pathrelse(path);
1231         if (S_ISREG(inode->i_mode)) {
1232                 inode->i_op = &reiserfs_file_inode_operations;
1233                 inode->i_fop = &reiserfs_file_operations;
1234                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1235         } else if (S_ISDIR(inode->i_mode)) {
1236                 inode->i_op = &reiserfs_dir_inode_operations;
1237                 inode->i_fop = &reiserfs_dir_operations;
1238         } else if (S_ISLNK(inode->i_mode)) {
1239                 inode->i_op = &reiserfs_symlink_inode_operations;
1240                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1241         } else {
1242                 inode->i_blocks = 0;
1243                 inode->i_op = &reiserfs_special_inode_operations;
1244                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1245         }
1246 }
1247
1248 // update new stat data with inode fields
1249 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1250 {
1251         struct stat_data *sd_v2 = (struct stat_data *)sd;
1252         uid_t uid = TAGINO_UID(DX_TAG(inode), inode->i_uid, inode->i_tag);
1253         gid_t gid = TAGINO_GID(DX_TAG(inode), inode->i_gid, inode->i_tag);
1254         __u16 flags;
1255
1256         set_sd_v2_uid(sd_v2, uid);
1257         set_sd_v2_gid(sd_v2, gid);
1258         set_sd_v2_mode(sd_v2, inode->i_mode);
1259         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1260         set_sd_v2_size(sd_v2, size);
1261         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1262         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1263         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1264         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1265         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1266                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1267         else
1268                 set_sd_v2_generation(sd_v2, inode->i_generation);
1269         flags = REISERFS_I(inode)->i_attrs;
1270         i_attrs_to_sd_attrs(inode, &flags);
1271         set_sd_v2_attrs(sd_v2, flags);
1272 }
1273
1274 // used to copy inode's fields to old stat data
1275 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1276 {
1277         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1278
1279         set_sd_v1_mode(sd_v1, inode->i_mode);
1280         set_sd_v1_uid(sd_v1, inode->i_uid);
1281         set_sd_v1_gid(sd_v1, inode->i_gid);
1282         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1283         set_sd_v1_size(sd_v1, size);
1284         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1285         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1286         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1287
1288         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1289                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1290         else
1291                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1292
1293         // Sigh. i_first_direct_byte is back
1294         set_sd_v1_first_direct_byte(sd_v1,
1295                                     REISERFS_I(inode)->i_first_direct_byte);
1296 }
1297
1298 /* NOTE, you must prepare the buffer head before sending it here,
1299 ** and then log it after the call
1300 */
1301 static void update_stat_data(struct treepath *path, struct inode *inode,
1302                              loff_t size)
1303 {
1304         struct buffer_head *bh;
1305         struct item_head *ih;
1306
1307         bh = PATH_PLAST_BUFFER(path);
1308         ih = PATH_PITEM_HEAD(path);
1309
1310         if (!is_statdata_le_ih(ih))
1311                 reiserfs_panic(inode->i_sb,
1312                                "vs-13065: update_stat_data: key %k, found item %h",
1313                                INODE_PKEY(inode), ih);
1314
1315         if (stat_data_v1(ih)) {
1316                 // path points to old stat data
1317                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1318         } else {
1319                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1320         }
1321
1322         return;
1323 }
1324
1325 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1326                              struct inode *inode, loff_t size)
1327 {
1328         struct cpu_key key;
1329         INITIALIZE_PATH(path);
1330         struct buffer_head *bh;
1331         int fs_gen;
1332         struct item_head *ih, tmp_ih;
1333         int retval;
1334
1335         BUG_ON(!th->t_trans_id);
1336
1337         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1338
1339         for (;;) {
1340                 int pos;
1341                 /* look for the object's stat data */
1342                 retval = search_item(inode->i_sb, &key, &path);
1343                 if (retval == IO_ERROR) {
1344                         reiserfs_warning(inode->i_sb,
1345                                          "vs-13050: reiserfs_update_sd: "
1346                                          "i/o failure occurred trying to update %K stat data",
1347                                          &key);
1348                         return;
1349                 }
1350                 if (retval == ITEM_NOT_FOUND) {
1351                         pos = PATH_LAST_POSITION(&path);
1352                         pathrelse(&path);
1353                         if (inode->i_nlink == 0) {
1354                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1355                                 return;
1356                         }
1357                         reiserfs_warning(inode->i_sb,
1358                                          "vs-13060: reiserfs_update_sd: "
1359                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1360                                          INODE_PKEY(inode), inode->i_nlink,
1361                                          pos);
1362                         reiserfs_check_path(&path);
1363                         return;
1364                 }
1365
1366                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1367                  ** FS might change.  We have to detect that, and loop back to the
1368                  ** search if the stat data item has moved
1369                  */
1370                 bh = get_last_bh(&path);
1371                 ih = get_ih(&path);
1372                 copy_item_head(&tmp_ih, ih);
1373                 fs_gen = get_generation(inode->i_sb);
1374                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1375                 if (fs_changed(fs_gen, inode->i_sb)
1376                     && item_moved(&tmp_ih, &path)) {
1377                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1378                         continue;       /* Stat_data item has been moved after scheduling. */
1379                 }
1380                 break;
1381         }
1382         update_stat_data(&path, inode, size);
1383         journal_mark_dirty(th, th->t_super, bh);
1384         pathrelse(&path);
1385         return;
1386 }
1387
1388 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1389 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1390 ** and clear the key in the private portion of the inode, otherwise a
1391 ** corresponding iput might try to delete whatever object the inode last
1392 ** represented.
1393 */
1394 static void reiserfs_make_bad_inode(struct inode *inode)
1395 {
1396         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1397         make_bad_inode(inode);
1398 }
1399
1400 //
1401 // initially this function was derived from minix or ext2's analog and
1402 // evolved as the prototype did
1403 //
1404
1405 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1406 {
1407         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1408         inode->i_ino = args->objectid;
1409         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1410         return 0;
1411 }
1412
1413 /* looks for stat data in the tree, and fills up the fields of in-core
1414    inode stat data fields */
1415 void reiserfs_read_locked_inode(struct inode *inode,
1416                                 struct reiserfs_iget_args *args)
1417 {
1418         INITIALIZE_PATH(path_to_sd);
1419         struct cpu_key key;
1420         unsigned long dirino;
1421         int retval;
1422
1423         dirino = args->dirid;
1424
1425         /* set version 1, version 2 could be used too, because stat data
1426            key is the same in both versions */
1427         key.version = KEY_FORMAT_3_5;
1428         key.on_disk_key.k_dir_id = dirino;
1429         key.on_disk_key.k_objectid = inode->i_ino;
1430         key.on_disk_key.k_offset = 0;
1431         key.on_disk_key.k_type = 0;
1432
1433         /* look for the object's stat data */
1434         retval = search_item(inode->i_sb, &key, &path_to_sd);
1435         if (retval == IO_ERROR) {
1436                 reiserfs_warning(inode->i_sb,
1437                                  "vs-13070: reiserfs_read_locked_inode: "
1438                                  "i/o failure occurred trying to find stat data of %K",
1439                                  &key);
1440                 reiserfs_make_bad_inode(inode);
1441                 return;
1442         }
1443         if (retval != ITEM_FOUND) {
1444                 /* a stale NFS handle can trigger this without it being an error */
1445                 pathrelse(&path_to_sd);
1446                 reiserfs_make_bad_inode(inode);
1447                 inode->i_nlink = 0;
1448                 return;
1449         }
1450
1451         init_inode(inode, &path_to_sd);
1452
1453         /* It is possible that knfsd is trying to access inode of a file
1454            that is being removed from the disk by some other thread. As we
1455            update sd on unlink all that is required is to check for nlink
1456            here. This bug was first found by Sizif when debugging
1457            SquidNG/Butterfly, forgotten, and found again after Philippe
1458            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1459
1460            More logical fix would require changes in fs/inode.c:iput() to
1461            remove inode from hash-table _after_ fs cleaned disk stuff up and
1462            in iget() to return NULL if I_FREEING inode is found in
1463            hash-table. */
1464         /* Currently there is one place where it's ok to meet inode with
1465            nlink==0: processing of open-unlinked and half-truncated files
1466            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1467         if ((inode->i_nlink == 0) &&
1468             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1469                 reiserfs_warning(inode->i_sb,
1470                                  "vs-13075: reiserfs_read_locked_inode: "
1471                                  "dead inode read from disk %K. "
1472                                  "This is likely to be race with knfsd. Ignore",
1473                                  &key);
1474                 reiserfs_make_bad_inode(inode);
1475         }
1476
1477         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1478
1479 }
1480
1481 /**
1482  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1483  *
1484  * @inode:    inode from hash table to check
1485  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1486  *
1487  * This function is called by iget5_locked() to distinguish reiserfs inodes
1488  * having the same inode numbers. Such inodes can only exist due to some
1489  * error condition. One of them should be bad. Inodes with identical
1490  * inode numbers (objectids) are distinguished by parent directory ids.
1491  *
1492  */
1493 int reiserfs_find_actor(struct inode *inode, void *opaque)
1494 {
1495         struct reiserfs_iget_args *args;
1496
1497         args = opaque;
1498         /* args is already in CPU order */
1499         return (inode->i_ino == args->objectid) &&
1500             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1501 }
1502
1503 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1504 {
1505         struct inode *inode;
1506         struct reiserfs_iget_args args;
1507
1508         args.objectid = key->on_disk_key.k_objectid;
1509         args.dirid = key->on_disk_key.k_dir_id;
1510         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1511                              reiserfs_find_actor, reiserfs_init_locked_inode,
1512                              (void *)(&args));
1513         if (!inode)
1514                 return ERR_PTR(-ENOMEM);
1515
1516         if (inode->i_state & I_NEW) {
1517                 reiserfs_read_locked_inode(inode, &args);
1518                 unlock_new_inode(inode);
1519         }
1520
1521         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1522                 /* either due to i/o error or a stale NFS handle */
1523                 iput(inode);
1524                 inode = NULL;
1525         }
1526         return inode;
1527 }
1528
1529 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1530 {
1531         __u32 *data = vobjp;
1532         struct cpu_key key;
1533         struct dentry *result;
1534         struct inode *inode;
1535
1536         key.on_disk_key.k_objectid = data[0];
1537         key.on_disk_key.k_dir_id = data[1];
1538         reiserfs_write_lock(sb);
1539         inode = reiserfs_iget(sb, &key);
1540         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1541             data[2] != inode->i_generation) {
1542                 iput(inode);
1543                 inode = NULL;
1544         }
1545         reiserfs_write_unlock(sb);
1546         if (!inode)
1547                 inode = ERR_PTR(-ESTALE);
1548         if (IS_ERR(inode))
1549                 return ERR_PTR(PTR_ERR(inode));
1550         result = d_alloc_anon(inode);
1551         if (!result) {
1552                 iput(inode);
1553                 return ERR_PTR(-ENOMEM);
1554         }
1555         return result;
1556 }
1557
1558 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1559                                   int len, int fhtype,
1560                                   int (*acceptable) (void *contect,
1561                                                      struct dentry * de),
1562                                   void *context)
1563 {
1564         __u32 obj[3], parent[3];
1565
1566         /* fhtype happens to reflect the number of u32s encoded.
1567          * due to a bug in earlier code, fhtype might indicate there
1568          * are more u32s then actually fitted.
1569          * so if fhtype seems to be more than len, reduce fhtype.
1570          * Valid types are:
1571          *   2 - objectid + dir_id - legacy support
1572          *   3 - objectid + dir_id + generation
1573          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1574          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1575          *   6 - as above plus generation of directory
1576          * 6 does not fit in NFSv2 handles
1577          */
1578         if (fhtype > len) {
1579                 if (fhtype != 6 || len != 5)
1580                         reiserfs_warning(sb,
1581                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1582                                          fhtype, len);
1583                 fhtype = 5;
1584         }
1585
1586         obj[0] = data[0];
1587         obj[1] = data[1];
1588         if (fhtype == 3 || fhtype >= 5)
1589                 obj[2] = data[2];
1590         else
1591                 obj[2] = 0;     /* generation number */
1592
1593         if (fhtype >= 4) {
1594                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1595                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1596                 if (fhtype == 6)
1597                         parent[2] = data[5];
1598                 else
1599                         parent[2] = 0;
1600         }
1601         return sb->s_export_op->find_exported_dentry(sb, obj,
1602                                                      fhtype < 4 ? NULL : parent,
1603                                                      acceptable, context);
1604 }
1605
1606 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1607                        int need_parent)
1608 {
1609         struct inode *inode = dentry->d_inode;
1610         int maxlen = *lenp;
1611
1612         if (maxlen < 3)
1613                 return 255;
1614
1615         data[0] = inode->i_ino;
1616         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1617         data[2] = inode->i_generation;
1618         *lenp = 3;
1619         /* no room for directory info? return what we've stored so far */
1620         if (maxlen < 5 || !need_parent)
1621                 return 3;
1622
1623         spin_lock(&dentry->d_lock);
1624         inode = dentry->d_parent->d_inode;
1625         data[3] = inode->i_ino;
1626         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1627         *lenp = 5;
1628         if (maxlen >= 6) {
1629                 data[5] = inode->i_generation;
1630                 *lenp = 6;
1631         }
1632         spin_unlock(&dentry->d_lock);
1633         return *lenp;
1634 }
1635
1636 /* looks for stat data, then copies fields to it, marks the buffer
1637    containing stat data as dirty */
1638 /* reiserfs inodes are never really dirty, since the dirty inode call
1639 ** always logs them.  This call allows the VFS inode marking routines
1640 ** to properly mark inodes for datasync and such, but only actually
1641 ** does something when called for a synchronous update.
1642 */
1643 int reiserfs_write_inode(struct inode *inode, int do_sync)
1644 {
1645         struct reiserfs_transaction_handle th;
1646         int jbegin_count = 1;
1647
1648         if (inode->i_sb->s_flags & MS_RDONLY)
1649                 return -EROFS;
1650         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1651          ** these cases are just when the system needs ram, not when the
1652          ** inode needs to reach disk for safety, and they can safely be
1653          ** ignored because the altered inode has already been logged.
1654          */
1655         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1656                 reiserfs_write_lock(inode->i_sb);
1657                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1658                         reiserfs_update_sd(&th, inode);
1659                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1660                 }
1661                 reiserfs_write_unlock(inode->i_sb);
1662         }
1663         return 0;
1664 }
1665
1666 /* stat data of new object is inserted already, this inserts the item
1667    containing "." and ".." entries */
1668 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1669                                   struct inode *inode,
1670                                   struct item_head *ih, struct treepath *path,
1671                                   struct inode *dir)
1672 {
1673         struct super_block *sb = th->t_super;
1674         char empty_dir[EMPTY_DIR_SIZE];
1675         char *body = empty_dir;
1676         struct cpu_key key;
1677         int retval;
1678
1679         BUG_ON(!th->t_trans_id);
1680
1681         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1682                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1683                       TYPE_DIRENTRY, 3 /*key length */ );
1684
1685         /* compose item head for new item. Directories consist of items of
1686            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1687            is done by reiserfs_new_inode */
1688         if (old_format_only(sb)) {
1689                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1690                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1691
1692                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1693                                        ih->ih_key.k_objectid,
1694                                        INODE_PKEY(dir)->k_dir_id,
1695                                        INODE_PKEY(dir)->k_objectid);
1696         } else {
1697                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1698                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1699
1700                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1701                                     ih->ih_key.k_objectid,
1702                                     INODE_PKEY(dir)->k_dir_id,
1703                                     INODE_PKEY(dir)->k_objectid);
1704         }
1705
1706         /* look for place in the tree for new item */
1707         retval = search_item(sb, &key, path);
1708         if (retval == IO_ERROR) {
1709                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1710                                  "i/o failure occurred creating new directory");
1711                 return -EIO;
1712         }
1713         if (retval == ITEM_FOUND) {
1714                 pathrelse(path);
1715                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1716                                  "object with this key exists (%k)",
1717                                  &(ih->ih_key));
1718                 return -EEXIST;
1719         }
1720
1721         /* insert item, that is empty directory item */
1722         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1723 }
1724
1725 /* stat data of object has been inserted, this inserts the item
1726    containing the body of symlink */
1727 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1728                                 struct item_head *ih,
1729                                 struct treepath *path, const char *symname,
1730                                 int item_len)
1731 {
1732         struct super_block *sb = th->t_super;
1733         struct cpu_key key;
1734         int retval;
1735
1736         BUG_ON(!th->t_trans_id);
1737
1738         _make_cpu_key(&key, KEY_FORMAT_3_5,
1739                       le32_to_cpu(ih->ih_key.k_dir_id),
1740                       le32_to_cpu(ih->ih_key.k_objectid),
1741                       1, TYPE_DIRECT, 3 /*key length */ );
1742
1743         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1744                           0 /*free_space */ );
1745
1746         /* look for place in the tree for new item */
1747         retval = search_item(sb, &key, path);
1748         if (retval == IO_ERROR) {
1749                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1750                                  "i/o failure occurred creating new symlink");
1751                 return -EIO;
1752         }
1753         if (retval == ITEM_FOUND) {
1754                 pathrelse(path);
1755                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1756                                  "object with this key exists (%k)",
1757                                  &(ih->ih_key));
1758                 return -EEXIST;
1759         }
1760
1761         /* insert item, that is body of symlink */
1762         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1763 }
1764
1765 /* inserts the stat data into the tree, and then calls
1766    reiserfs_new_directory (to insert ".", ".." item if new object is
1767    directory) or reiserfs_new_symlink (to insert symlink body if new
1768    object is symlink) or nothing (if new object is regular file)
1769
1770    NOTE! uid and gid must already be set in the inode.  If we return
1771    non-zero due to an error, we have to drop the quota previously allocated
1772    for the fresh inode.  This can only be done outside a transaction, so
1773    if we return non-zero, we also end the transaction.  */
1774 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1775                        struct inode *dir, int mode, const char *symname,
1776                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1777                           strlen (symname) for symlinks) */
1778                        loff_t i_size, struct dentry *dentry,
1779                        struct inode *inode)
1780 {
1781         struct super_block *sb;
1782         INITIALIZE_PATH(path_to_key);
1783         struct cpu_key key;
1784         struct item_head ih;
1785         struct stat_data sd;
1786         int retval;
1787         int err;
1788
1789         BUG_ON(!th->t_trans_id);
1790
1791         if (DLIMIT_ALLOC_INODE(inode)) {
1792                 err = -ENOSPC;
1793                 goto out_bad_dlimit;
1794         }
1795         if (DQUOT_ALLOC_INODE(inode)) {
1796                 err = -EDQUOT;
1797                 goto out_end_trans;
1798         }
1799         if (!dir->i_nlink) {
1800                 err = -EPERM;
1801                 goto out_bad_inode;
1802         }
1803
1804         sb = dir->i_sb;
1805
1806         /* item head of new item */
1807         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1808         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1809         if (!ih.ih_key.k_objectid) {
1810                 err = -ENOMEM;
1811                 goto out_bad_inode;
1812         }
1813         if (old_format_only(sb))
1814                 /* not a perfect generation count, as object ids can be reused, but
1815                  ** this is as good as reiserfs can do right now.
1816                  ** note that the private part of inode isn't filled in yet, we have
1817                  ** to use the directory.
1818                  */
1819                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1820         else
1821 #if defined( USE_INODE_GENERATION_COUNTER )
1822                 inode->i_generation =
1823                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1824 #else
1825                 inode->i_generation = ++event;
1826 #endif
1827
1828         /* fill stat data */
1829         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1830
1831         /* uid and gid must already be set by the caller for quota init */
1832
1833         /* symlink cannot be immutable or append only, right? */
1834         if (S_ISLNK(inode->i_mode))
1835                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1836
1837         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1838         inode->i_size = i_size;
1839         inode->i_blocks = 0;
1840         inode->i_bytes = 0;
1841         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1842             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1843
1844         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1845         REISERFS_I(inode)->i_flags = 0;
1846         REISERFS_I(inode)->i_prealloc_block = 0;
1847         REISERFS_I(inode)->i_prealloc_count = 0;
1848         REISERFS_I(inode)->i_trans_id = 0;
1849         REISERFS_I(inode)->i_jl = NULL;
1850         REISERFS_I(inode)->i_attrs =
1851             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1852         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1853         mutex_init(&(REISERFS_I(inode)->i_mmap));
1854         reiserfs_init_acl_access(inode);
1855         reiserfs_init_acl_default(inode);
1856         reiserfs_init_xattr_rwsem(inode);
1857
1858         if (old_format_only(sb))
1859                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1860                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1861         else
1862                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1863                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1864
1865         /* key to search for correct place for new stat data */
1866         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1867                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1868                       TYPE_STAT_DATA, 3 /*key length */ );
1869
1870         /* find proper place for inserting of stat data */
1871         retval = search_item(sb, &key, &path_to_key);
1872         if (retval == IO_ERROR) {
1873                 err = -EIO;
1874                 goto out_bad_inode;
1875         }
1876         if (retval == ITEM_FOUND) {
1877                 pathrelse(&path_to_key);
1878                 err = -EEXIST;
1879                 goto out_bad_inode;
1880         }
1881         if (old_format_only(sb)) {
1882                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1883                         pathrelse(&path_to_key);
1884                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1885                         err = -EINVAL;
1886                         goto out_bad_inode;
1887                 }
1888                 inode2sd_v1(&sd, inode, inode->i_size);
1889         } else {
1890                 inode2sd(&sd, inode, inode->i_size);
1891         }
1892         // these do not go to on-disk stat data
1893         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1894
1895         // store in in-core inode the key of stat data and version all
1896         // object items will have (directory items will have old offset
1897         // format, other new objects will consist of new items)
1898         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1899         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1900                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1901         else
1902                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1903         if (old_format_only(sb))
1904                 set_inode_sd_version(inode, STAT_DATA_V1);
1905         else
1906                 set_inode_sd_version(inode, STAT_DATA_V2);
1907
1908         /* insert the stat data into the tree */
1909 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1910         if (REISERFS_I(dir)->new_packing_locality)
1911                 th->displace_new_blocks = 1;
1912 #endif
1913         retval =
1914             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1915                                  (char *)(&sd));
1916         if (retval) {
1917                 err = retval;
1918                 reiserfs_check_path(&path_to_key);
1919                 goto out_bad_inode;
1920         }
1921 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1922         if (!th->displace_new_blocks)
1923                 REISERFS_I(dir)->new_packing_locality = 0;
1924 #endif
1925         if (S_ISDIR(mode)) {
1926                 /* insert item with "." and ".." */
1927                 retval =
1928                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1929         }
1930
1931         if (S_ISLNK(mode)) {
1932                 /* insert body of symlink */
1933                 if (!old_format_only(sb))
1934                         i_size = ROUND_UP(i_size);
1935                 retval =
1936                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1937                                          i_size);
1938         }
1939         if (retval) {
1940                 err = retval;
1941                 reiserfs_check_path(&path_to_key);
1942                 journal_end(th, th->t_super, th->t_blocks_allocated);
1943                 goto out_inserted_sd;
1944         }
1945
1946         /* XXX CHECK THIS */
1947         if (reiserfs_posixacl(inode->i_sb)) {
1948                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1949                 if (retval) {
1950                         err = retval;
1951                         reiserfs_check_path(&path_to_key);
1952                         journal_end(th, th->t_super, th->t_blocks_allocated);
1953                         goto out_inserted_sd;
1954                 }
1955         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1956                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1957                                  "but vfs thinks they are!");
1958         } else if (is_reiserfs_priv_object(dir)) {
1959                 reiserfs_mark_inode_private(inode);
1960         }
1961
1962         insert_inode_hash(inode);
1963         reiserfs_update_sd(th, inode);
1964         reiserfs_check_path(&path_to_key);
1965
1966         return 0;
1967
1968 /* it looks like you can easily compress these two goto targets into
1969  * one.  Keeping it like this doesn't actually hurt anything, and they
1970  * are place holders for what the quota code actually needs.
1971  */
1972       out_bad_inode:
1973         /* Invalidate the object, nothing was inserted yet */
1974         INODE_PKEY(inode)->k_objectid = 0;
1975
1976         /* Quota change must be inside a transaction for journaling */
1977         DQUOT_FREE_INODE(inode);
1978
1979       out_end_trans:
1980         DLIMIT_FREE_INODE(inode);
1981
1982       out_bad_dlimit:
1983         journal_end(th, th->t_super, th->t_blocks_allocated);
1984         /* Drop can be outside and it needs more credits so it's better to have it outside */
1985         DQUOT_DROP(inode);
1986         inode->i_flags |= S_NOQUOTA;
1987         make_bad_inode(inode);
1988
1989       out_inserted_sd:
1990         inode->i_nlink = 0;
1991         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1992
1993         /* If we were inheriting an ACL, we need to release the lock so that
1994          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1995          * code really needs to be reworked, but this will take care of it
1996          * for now. -jeffm */
1997 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1998         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1999                 reiserfs_write_unlock_xattrs(dir->i_sb);
2000                 iput(inode);
2001                 reiserfs_write_lock_xattrs(dir->i_sb);
2002         } else
2003 #endif
2004                 iput(inode);
2005         return err;
2006 }
2007
2008 /*
2009 ** finds the tail page in the page cache,
2010 ** reads the last block in.
2011 **
2012 ** On success, page_result is set to a locked, pinned page, and bh_result
2013 ** is set to an up to date buffer for the last block in the file.  returns 0.
2014 **
2015 ** tail conversion is not done, so bh_result might not be valid for writing
2016 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2017 ** trying to write the block.
2018 **
2019 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2020 */
2021 static int grab_tail_page(struct inode *p_s_inode,
2022                           struct page **page_result,
2023                           struct buffer_head **bh_result)
2024 {
2025
2026         /* we want the page with the last byte in the file,
2027          ** not the page that will hold the next byte for appending
2028          */
2029         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2030         unsigned long pos = 0;
2031         unsigned long start = 0;
2032         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2033         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2034         struct buffer_head *bh;
2035         struct buffer_head *head;
2036         struct page *page;
2037         int error;
2038
2039         /* we know that we are only called with inode->i_size > 0.
2040          ** we also know that a file tail can never be as big as a block
2041          ** If i_size % blocksize == 0, our file is currently block aligned
2042          ** and it won't need converting or zeroing after a truncate.
2043          */
2044         if ((offset & (blocksize - 1)) == 0) {
2045                 return -ENOENT;
2046         }
2047         page = grab_cache_page(p_s_inode->i_mapping, index);
2048         error = -ENOMEM;
2049         if (!page) {
2050                 goto out;
2051         }
2052         /* start within the page of the last block in the file */
2053         start = (offset / blocksize) * blocksize;
2054
2055         error = block_prepare_write(page, start, offset,
2056                                     reiserfs_get_block_create_0);
2057         if (error)
2058                 goto unlock;
2059
2060         head = page_buffers(page);
2061         bh = head;
2062         do {
2063                 if (pos >= start) {
2064                         break;
2065                 }
2066                 bh = bh->b_this_page;
2067                 pos += blocksize;
2068         } while (bh != head);
2069
2070         if (!buffer_uptodate(bh)) {
2071                 /* note, this should never happen, prepare_write should
2072                  ** be taking care of this for us.  If the buffer isn't up to date,
2073                  ** I've screwed up the code to find the buffer, or the code to
2074                  ** call prepare_write
2075                  */
2076                 reiserfs_warning(p_s_inode->i_sb,
2077                                  "clm-6000: error reading block %lu on dev %s",
2078                                  bh->b_blocknr,
2079                                  reiserfs_bdevname(p_s_inode->i_sb));
2080                 error = -EIO;
2081                 goto unlock;
2082         }
2083         *bh_result = bh;
2084         *page_result = page;
2085
2086       out:
2087         return error;
2088
2089       unlock:
2090         unlock_page(page);
2091         page_cache_release(page);
2092         return error;
2093 }
2094
2095 /*
2096 ** vfs version of truncate file.  Must NOT be called with
2097 ** a transaction already started.
2098 **
2099 ** some code taken from block_truncate_page
2100 */
2101 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2102 {
2103         struct reiserfs_transaction_handle th;
2104         /* we want the offset for the first byte after the end of the file */
2105         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2106         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2107         unsigned length;
2108         struct page *page = NULL;
2109         int error;
2110         struct buffer_head *bh = NULL;
2111         int err2;
2112
2113         reiserfs_write_lock(p_s_inode->i_sb);
2114
2115         if (p_s_inode->i_size > 0) {
2116                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2117                         // -ENOENT means we truncated past the end of the file,
2118                         // and get_block_create_0 could not find a block to read in,
2119                         // which is ok.
2120                         if (error != -ENOENT)
2121                                 reiserfs_warning(p_s_inode->i_sb,
2122                                                  "clm-6001: grab_tail_page failed %d",
2123                                                  error);
2124                         page = NULL;
2125                         bh = NULL;
2126                 }
2127         }
2128
2129         /* so, if page != NULL, we have a buffer head for the offset at
2130          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2131          ** then we have an unformatted node.  Otherwise, we have a direct item,
2132          ** and no zeroing is required on disk.  We zero after the truncate,
2133          ** because the truncate might pack the item anyway
2134          ** (it will unmap bh if it packs).
2135          */
2136         /* it is enough to reserve space in transaction for 2 balancings:
2137            one for "save" link adding and another for the first
2138            cut_from_item. 1 is for update_sd */
2139         error = journal_begin(&th, p_s_inode->i_sb,
2140                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2141         if (error)
2142                 goto out;
2143         reiserfs_update_inode_transaction(p_s_inode);
2144         if (update_timestamps)
2145                 /* we are doing real truncate: if the system crashes before the last
2146                    transaction of truncating gets committed - on reboot the file
2147                    either appears truncated properly or not truncated at all */
2148                 add_save_link(&th, p_s_inode, 1);
2149         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2150         error =
2151             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2152         if (error)
2153                 goto out;
2154
2155         /* check reiserfs_do_truncate after ending the transaction */
2156         if (err2) {
2157                 error = err2;
2158                 goto out;
2159         }
2160
2161         if (update_timestamps) {
2162                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2163                 if (error)
2164                         goto out;
2165         }
2166
2167         if (page) {
2168                 length = offset & (blocksize - 1);
2169                 /* if we are not on a block boundary */
2170                 if (length) {
2171                         char *kaddr;
2172
2173                         length = blocksize - length;
2174                         kaddr = kmap_atomic(page, KM_USER0);
2175                         memset(kaddr + offset, 0, length);
2176                         flush_dcache_page(page);
2177                         kunmap_atomic(kaddr, KM_USER0);
2178                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2179                                 mark_buffer_dirty(bh);
2180                         }
2181                 }
2182                 unlock_page(page);
2183                 page_cache_release(page);
2184         }
2185
2186         reiserfs_write_unlock(p_s_inode->i_sb);
2187         return 0;
2188       out:
2189         if (page) {
2190                 unlock_page(page);
2191                 page_cache_release(page);
2192         }
2193         reiserfs_write_unlock(p_s_inode->i_sb);
2194         return error;
2195 }
2196
2197 static int map_block_for_writepage(struct inode *inode,
2198                                    struct buffer_head *bh_result,
2199                                    unsigned long block)
2200 {
2201         struct reiserfs_transaction_handle th;
2202         int fs_gen;
2203         struct item_head tmp_ih;
2204         struct item_head *ih;
2205         struct buffer_head *bh;
2206         __le32 *item;
2207         struct cpu_key key;
2208         INITIALIZE_PATH(path);
2209         int pos_in_item;
2210         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2211         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2212         int retval;
2213         int use_get_block = 0;
2214         int bytes_copied = 0;
2215         int copy_size;
2216         int trans_running = 0;
2217
2218         /* catch places below that try to log something without starting a trans */
2219         th.t_trans_id = 0;
2220
2221         if (!buffer_uptodate(bh_result)) {
2222                 return -EIO;
2223         }
2224
2225         kmap(bh_result->b_page);
2226       start_over:
2227         reiserfs_write_lock(inode->i_sb);
2228         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2229
2230       research:
2231         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2232         if (retval != POSITION_FOUND) {
2233                 use_get_block = 1;
2234                 goto out;
2235         }
2236
2237         bh = get_last_bh(&path);
2238         ih = get_ih(&path);
2239         item = get_item(&path);
2240         pos_in_item = path.pos_in_item;
2241
2242         /* we've found an unformatted node */
2243         if (indirect_item_found(retval, ih)) {
2244                 if (bytes_copied > 0) {
2245                         reiserfs_warning(inode->i_sb,
2246                                          "clm-6002: bytes_copied %d",
2247                                          bytes_copied);
2248                 }
2249                 if (!get_block_num(item, pos_in_item)) {
2250                         /* crap, we are writing to a hole */
2251                         use_get_block = 1;
2252                         goto out;
2253                 }
2254                 set_block_dev_mapped(bh_result,
2255                                      get_block_num(item, pos_in_item), inode);
2256         } else if (is_direct_le_ih(ih)) {
2257                 char *p;
2258                 p = page_address(bh_result->b_page);
2259                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2260                 copy_size = ih_item_len(ih) - pos_in_item;
2261
2262                 fs_gen = get_generation(inode->i_sb);
2263                 copy_item_head(&tmp_ih, ih);
2264
2265                 if (!trans_running) {
2266                         /* vs-3050 is gone, no need to drop the path */
2267                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2268                         if (retval)
2269                                 goto out;
2270                         reiserfs_update_inode_transaction(inode);
2271                         trans_running = 1;
2272                         if (fs_changed(fs_gen, inode->i_sb)
2273                             && item_moved(&tmp_ih, &path)) {
2274                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2275                                                                  bh);
2276                                 goto research;
2277                         }
2278                 }
2279
2280                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2281
2282                 if (fs_changed(fs_gen, inode->i_sb)
2283                     && item_moved(&tmp_ih, &path)) {
2284                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2285                         goto research;
2286                 }
2287
2288                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2289                        copy_size);
2290
2291                 journal_mark_dirty(&th, inode->i_sb, bh);
2292                 bytes_copied += copy_size;
2293                 set_block_dev_mapped(bh_result, 0, inode);
2294
2295                 /* are there still bytes left? */
2296                 if (bytes_copied < bh_result->b_size &&
2297                     (byte_offset + bytes_copied) < inode->i_size) {
2298                         set_cpu_key_k_offset(&key,
2299                                              cpu_key_k_offset(&key) +
2300                                              copy_size);
2301                         goto research;
2302                 }
2303         } else {
2304                 reiserfs_warning(inode->i_sb,
2305                                  "clm-6003: bad item inode %lu, device %s",
2306                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2307                 retval = -EIO;
2308                 goto out;
2309         }
2310         retval = 0;
2311
2312       out:
2313         pathrelse(&path);
2314         if (trans_running) {
2315                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2316                 if (err)
2317                         retval = err;
2318                 trans_running = 0;
2319         }
2320         reiserfs_write_unlock(inode->i_sb);
2321
2322         /* this is where we fill in holes in the file. */
2323         if (use_get_block) {
2324                 retval = reiserfs_get_block(inode, block, bh_result,
2325                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2326                                             | GET_BLOCK_NO_DANGLE);
2327                 if (!retval) {
2328                         if (!buffer_mapped(bh_result)
2329                             || bh_result->b_blocknr == 0) {
2330                                 /* get_block failed to find a mapped unformatted node. */
2331                                 use_get_block = 0;
2332                                 goto start_over;
2333                         }
2334                 }
2335         }
2336         kunmap(bh_result->b_page);
2337
2338         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2339                 /* we've copied data from the page into the direct item, so the
2340                  * buffer in the page is now clean, mark it to reflect that.
2341                  */
2342                 lock_buffer(bh_result);
2343                 clear_buffer_dirty(bh_result);
2344                 unlock_buffer(bh_result);
2345         }
2346         return retval;
2347 }
2348
2349 /*
2350  * mason@suse.com: updated in 2.5.54 to follow the same general io
2351  * start/recovery path as __block_write_full_page, along with special
2352  * code to handle reiserfs tails.
2353  */
2354 static int reiserfs_write_full_page(struct page *page,
2355                                     struct writeback_control *wbc)
2356 {
2357         struct inode *inode = page->mapping->host;
2358         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2359         int error = 0;
2360         unsigned long block;
2361         sector_t last_block;
2362         struct buffer_head *head, *bh;
2363         int partial = 0;
2364         int nr = 0;
2365         int checked = PageChecked(page);
2366         struct reiserfs_transaction_handle th;
2367         struct super_block *s = inode->i_sb;
2368         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2369         th.t_trans_id = 0;
2370
2371         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2372         if (checked && (current->flags & PF_MEMALLOC)) {
2373                 redirty_page_for_writepage(wbc, page);
2374                 unlock_page(page);
2375                 return 0;
2376         }
2377
2378         /* The page dirty bit is cleared before writepage is called, which
2379          * means we have to tell create_empty_buffers to make dirty buffers
2380          * The page really should be up to date at this point, so tossing
2381          * in the BH_Uptodate is just a sanity check.
2382          */
2383         if (!page_has_buffers(page)) {
2384                 create_empty_buffers(page, s->s_blocksize,
2385                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2386         }
2387         head = page_buffers(page);
2388
2389         /* last page in the file, zero out any contents past the
2390          ** last byte in the file
2391          */
2392         if (page->index >= end_index) {
2393                 char *kaddr;
2394                 unsigned last_offset;
2395
2396                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2397                 /* no file contents in this page */
2398                 if (page->index >= end_index + 1 || !last_offset) {
2399                         unlock_page(page);
2400                         return 0;
2401                 }
2402                 kaddr = kmap_atomic(page, KM_USER0);
2403                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2404                 flush_dcache_page(page);
2405                 kunmap_atomic(kaddr, KM_USER0);
2406         }
2407         bh = head;
2408         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2409         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2410         /* first map all the buffers, logging any direct items we find */
2411         do {
2412                 if (block > last_block) {
2413                         /*
2414                          * This can happen when the block size is less than
2415                          * the page size.  The corresponding bytes in the page
2416                          * were zero filled above
2417                          */
2418                         clear_buffer_dirty(bh);
2419                         set_buffer_uptodate(bh);
2420                 } else if ((checked || buffer_dirty(bh)) &&
2421                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2422                                                        && bh->b_blocknr ==
2423                                                        0))) {
2424                         /* not mapped yet, or it points to a direct item, search
2425                          * the btree for the mapping info, and log any direct
2426                          * items found
2427                          */
2428                         if ((error = map_block_for_writepage(inode, bh, block))) {
2429                                 goto fail;
2430                         }
2431                 }
2432                 bh = bh->b_this_page;
2433                 block++;
2434         } while (bh != head);
2435
2436         /*
2437          * we start the transaction after map_block_for_writepage,
2438          * because it can create holes in the file (an unbounded operation).
2439          * starting it here, we can make a reliable estimate for how many
2440          * blocks we're going to log
2441          */
2442         if (checked) {
2443                 ClearPageChecked(page);
2444                 reiserfs_write_lock(s);
2445                 error = journal_begin(&th, s, bh_per_page + 1);
2446                 if (error) {
2447                         reiserfs_write_unlock(s);
2448                         goto fail;
2449                 }
2450                 reiserfs_update_inode_transaction(inode);
2451         }
2452         /* now go through and lock any dirty buffers on the page */
2453         do {
2454                 get_bh(bh);
2455                 if (!buffer_mapped(bh))
2456                         continue;
2457                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2458                         continue;
2459
2460                 if (checked) {
2461                         reiserfs_prepare_for_journal(s, bh, 1);
2462                         journal_mark_dirty(&th, s, bh);
2463                         continue;
2464                 }
2465                 /* from this point on, we know the buffer is mapped to a
2466                  * real block and not a direct item
2467                  */
2468                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2469                         lock_buffer(bh);
2470                 } else {
2471                         if (test_set_buffer_locked(bh)) {
2472                                 redirty_page_for_writepage(wbc, page);
2473                                 continue;
2474                         }
2475                 }
2476                 if (test_clear_buffer_dirty(bh)) {
2477                         mark_buffer_async_write(bh);
2478                 } else {
2479                         unlock_buffer(bh);
2480                 }
2481         } while ((bh = bh->b_this_page) != head);
2482
2483         if (checked) {
2484                 error = journal_end(&th, s, bh_per_page + 1);
2485                 reiserfs_write_unlock(s);
2486                 if (error)
2487                         goto fail;
2488         }
2489         BUG_ON(PageWriteback(page));
2490         set_page_writeback(page);
2491         unlock_page(page);
2492
2493         /*
2494          * since any buffer might be the only dirty buffer on the page,
2495          * the first submit_bh can bring the page out of writeback.
2496          * be careful with the buffers.
2497          */
2498         do {
2499                 struct buffer_head *next = bh->b_this_page;
2500                 if (buffer_async_write(bh)) {
2501                         submit_bh(WRITE, bh);
2502                         nr++;
2503                 }
2504                 put_bh(bh);
2505                 bh = next;
2506         } while (bh != head);
2507
2508         error = 0;
2509       done:
2510         if (nr == 0) {
2511                 /*
2512                  * if this page only had a direct item, it is very possible for
2513                  * no io to be required without there being an error.  Or,
2514                  * someone else could have locked them and sent them down the
2515                  * pipe without locking the page
2516                  */
2517                 bh = head;
2518                 do {
2519                         if (!buffer_uptodate(bh)) {
2520                                 partial = 1;
2521                                 break;
2522                         }
2523                         bh = bh->b_this_page;
2524                 } while (bh != head);
2525                 if (!partial)
2526                         SetPageUptodate(page);
2527                 end_page_writeback(page);
2528         }
2529         return error;
2530
2531       fail:
2532         /* catches various errors, we need to make sure any valid dirty blocks
2533          * get to the media.  The page is currently locked and not marked for
2534          * writeback
2535          */
2536         ClearPageUptodate(page);
2537         bh = head;
2538         do {
2539                 get_bh(bh);
2540                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2541                         lock_buffer(bh);
2542                         mark_buffer_async_write(bh);
2543                 } else {
2544                         /*
2545                          * clear any dirty bits that might have come from getting
2546                          * attached to a dirty page
2547                          */
2548                         clear_buffer_dirty(bh);
2549                 }
2550                 bh = bh->b_this_page;
2551         } while (bh != head);
2552         SetPageError(page);
2553         BUG_ON(PageWriteback(page));
2554         set_page_writeback(page);
2555         unlock_page(page);
2556         do {
2557                 struct buffer_head *next = bh->b_this_page;
2558                 if (buffer_async_write(bh)) {
2559                         clear_buffer_dirty(bh);
2560                         submit_bh(WRITE, bh);
2561                         nr++;
2562                 }
2563                 put_bh(bh);
2564                 bh = next;
2565         } while (bh != head);
2566         goto done;
2567 }
2568
2569 static int reiserfs_readpage(struct file *f, struct page *page)
2570 {
2571         return block_read_full_page(page, reiserfs_get_block);
2572 }
2573
2574 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2575 {
2576         struct inode *inode = page->mapping->host;
2577         reiserfs_wait_on_write_block(inode->i_sb);
2578         return reiserfs_write_full_page(page, wbc);
2579 }
2580
2581 static int reiserfs_prepare_write(struct file *f, struct page *page,
2582                                   unsigned from, unsigned to)
2583 {
2584         struct inode *inode = page->mapping->host;
2585         int ret;
2586         int old_ref = 0;
2587
2588         reiserfs_wait_on_write_block(inode->i_sb);
2589         fix_tail_page_for_writing(page);
2590         if (reiserfs_transaction_running(inode->i_sb)) {
2591                 struct reiserfs_transaction_handle *th;
2592                 th = (struct reiserfs_transaction_handle *)current->
2593                     journal_info;
2594                 BUG_ON(!th->t_refcount);
2595                 BUG_ON(!th->t_trans_id);
2596                 old_ref = th->t_refcount;
2597                 th->t_refcount++;
2598         }
2599
2600         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2601         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2602                 struct reiserfs_transaction_handle *th = current->journal_info;
2603                 /* this gets a little ugly.  If reiserfs_get_block returned an
2604                  * error and left a transacstion running, we've got to close it,
2605                  * and we've got to free handle if it was a persistent transaction.
2606                  *
2607                  * But, if we had nested into an existing transaction, we need
2608                  * to just drop the ref count on the handle.
2609                  *
2610                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2611                  * and it was a persistent trans.  Otherwise, it was nested above.
2612                  */
2613                 if (th->t_refcount > old_ref) {
2614                         if (old_ref)
2615                                 th->t_refcount--;
2616                         else {
2617                                 int err;
2618                                 reiserfs_write_lock(inode->i_sb);
2619                                 err = reiserfs_end_persistent_transaction(th);
2620                                 reiserfs_write_unlock(inode->i_sb);
2621                                 if (err)
2622                                         ret = err;
2623                         }
2624                 }
2625         }
2626         return ret;
2627
2628 }
2629
2630 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2631 {
2632         return generic_block_bmap(as, block, reiserfs_bmap);
2633 }
2634
2635 static int reiserfs_commit_write(struct file *f, struct page *page,
2636                                  unsigned from, unsigned to)
2637 {
2638         struct inode *inode = page->mapping->host;
2639         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2640         int ret = 0;
2641         int update_sd = 0;
2642         struct reiserfs_transaction_handle *th = NULL;
2643
2644         reiserfs_wait_on_write_block(inode->i_sb);
2645         if (reiserfs_transaction_running(inode->i_sb)) {
2646                 th = current->journal_info;
2647         }
2648         reiserfs_commit_page(inode, page, from, to);
2649
2650         /* generic_commit_write does this for us, but does not update the
2651          ** transaction tracking stuff when the size changes.  So, we have
2652          ** to do the i_size updates here.
2653          */
2654         if (pos > inode->i_size) {
2655                 struct reiserfs_transaction_handle myth;
2656                 reiserfs_write_lock(inode->i_sb);
2657                 /* If the file have grown beyond the border where it
2658                    can have a tail, unmark it as needing a tail
2659                    packing */
2660                 if ((have_large_tails(inode->i_sb)
2661                      && inode->i_size > i_block_size(inode) * 4)
2662                     || (have_small_tails(inode->i_sb)
2663                         && inode->i_size > i_block_size(inode)))
2664                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2665
2666                 ret = journal_begin(&myth, inode->i_sb, 1);
2667                 if (ret) {
2668                         reiserfs_write_unlock(inode->i_sb);
2669                         goto journal_error;
2670                 }
2671                 reiserfs_update_inode_transaction(inode);
2672                 inode->i_size = pos;
2673                 /*
2674                  * this will just nest into our transaction.  It's important
2675                  * to use mark_inode_dirty so the inode gets pushed around on the
2676                  * dirty lists, and so that O_SYNC works as expected
2677                  */
2678                 mark_inode_dirty(inode);
2679                 reiserfs_update_sd(&myth, inode);
2680                 update_sd = 1;
2681                 ret = journal_end(&myth, inode->i_sb, 1);
2682                 reiserfs_write_unlock(inode->i_sb);
2683                 if (ret)
2684                         goto journal_error;
2685         }
2686         if (th) {
2687                 reiserfs_write_lock(inode->i_sb);
2688                 if (!update_sd)
2689                         mark_inode_dirty(inode);
2690                 ret = reiserfs_end_persistent_transaction(th);
2691                 reiserfs_write_unlock(inode->i_sb);
2692                 if (ret)
2693                         goto out;
2694         }
2695
2696       out:
2697         return ret;
2698
2699       journal_error:
2700         if (th) {
2701                 reiserfs_write_lock(inode->i_sb);
2702                 if (!update_sd)
2703                         reiserfs_update_sd(th, inode);
2704                 ret = reiserfs_end_persistent_transaction(th);
2705                 reiserfs_write_unlock(inode->i_sb);
2706         }
2707
2708         return ret;
2709 }
2710
2711 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2712 {
2713         if (reiserfs_attrs(inode->i_sb)) {
2714                 if (sd_attrs & REISERFS_SYNC_FL)
2715                         inode->i_flags |= S_SYNC;
2716                 else
2717                         inode->i_flags &= ~S_SYNC;
2718                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2719                         inode->i_flags |= S_IMMUTABLE;
2720                 else
2721                         inode->i_flags &= ~S_IMMUTABLE;
2722                 if (sd_attrs & REISERFS_IUNLINK_FL)
2723                         inode->i_flags |= S_IUNLINK;
2724                 else
2725                         inode->i_flags &= ~S_IUNLINK;
2726                 if (sd_attrs & REISERFS_BARRIER_FL)
2727                         inode->i_flags |= S_BARRIER;
2728                 else
2729                         inode->i_flags &= ~S_BARRIER;
2730                 if (sd_attrs & REISERFS_APPEND_FL)
2731                         inode->i_flags |= S_APPEND;
2732                 else
2733                         inode->i_flags &= ~S_APPEND;
2734                 if (sd_attrs & REISERFS_NOATIME_FL)
2735                         inode->i_flags |= S_NOATIME;
2736                 else
2737                         inode->i_flags &= ~S_NOATIME;
2738                 if (sd_attrs & REISERFS_NOTAIL_FL)
2739                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2740                 else
2741                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2742         }
2743 }
2744
2745 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2746 {
2747         if (reiserfs_attrs(inode->i_sb)) {
2748                 if (inode->i_flags & S_IMMUTABLE)
2749                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2750                 else
2751                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2752                 if (inode->i_flags & S_IUNLINK)
2753                         *sd_attrs |= REISERFS_IUNLINK_FL;
2754                 else
2755                         *sd_attrs &= ~REISERFS_IUNLINK_FL;
2756                 if (inode->i_flags & S_BARRIER)
2757                         *sd_attrs |= REISERFS_BARRIER_FL;
2758                 else
2759                         *sd_attrs &= ~REISERFS_BARRIER_FL;
2760                 if (inode->i_flags & S_SYNC)
2761                         *sd_attrs |= REISERFS_SYNC_FL;
2762                 else
2763                         *sd_attrs &= ~REISERFS_SYNC_FL;
2764                 if (inode->i_flags & S_NOATIME)
2765                         *sd_attrs |= REISERFS_NOATIME_FL;
2766                 else
2767                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2768                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2769                         *sd_attrs |= REISERFS_NOTAIL_FL;
2770                 else
2771                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2772         }
2773 }
2774
2775 /* decide if this buffer needs to stay around for data logging or ordered
2776 ** write purposes
2777 */
2778 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2779 {
2780         int ret = 1;
2781         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2782
2783         lock_buffer(bh);
2784         spin_lock(&j->j_dirty_buffers_lock);
2785         if (!buffer_mapped(bh)) {
2786                 goto free_jh;
2787         }
2788         /* the page is locked, and the only places that log a data buffer
2789          * also lock the page.
2790          */
2791         if (reiserfs_file_data_log(inode)) {
2792                 /*
2793                  * very conservative, leave the buffer pinned if
2794                  * anyone might need it.
2795                  */
2796                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2797                         ret = 0;
2798                 }
2799         } else  if (buffer_dirty(bh)) {
2800                 struct reiserfs_journal_list *jl;
2801                 struct reiserfs_jh *jh = bh->b_private;
2802
2803                 /* why is this safe?
2804                  * reiserfs_setattr updates i_size in the on disk
2805                  * stat data before allowing vmtruncate to be called.
2806                  *
2807                  * If buffer was put onto the ordered list for this
2808                  * transaction, we know for sure either this transaction
2809                  * or an older one already has updated i_size on disk,
2810                  * and this ordered data won't be referenced in the file
2811                  * if we crash.
2812                  *
2813                  * if the buffer was put onto the ordered list for an older
2814                  * transaction, we need to leave it around
2815                  */
2816                 if (jh && (jl = jh->jl)
2817                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2818                         ret = 0;
2819         }
2820       free_jh:
2821         if (ret && bh->b_private) {
2822                 reiserfs_free_jh(bh);
2823         }
2824         spin_unlock(&j->j_dirty_buffers_lock);
2825         unlock_buffer(bh);
2826         return ret;
2827 }
2828
2829 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2830 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2831 {
2832         struct buffer_head *head, *bh, *next;
2833         struct inode *inode = page->mapping->host;
2834         unsigned int curr_off = 0;
2835         int ret = 1;
2836
2837         BUG_ON(!PageLocked(page));
2838
2839         if (offset == 0)
2840                 ClearPageChecked(page);
2841
2842         if (!page_has_buffers(page))
2843                 goto out;
2844
2845         head = page_buffers(page);
2846         bh = head;
2847         do {
2848                 unsigned int next_off = curr_off + bh->b_size;
2849                 next = bh->b_this_page;
2850
2851                 /*
2852                  * is this block fully invalidated?
2853                  */
2854                 if (offset <= curr_off) {
2855                         if (invalidatepage_can_drop(inode, bh))
2856                                 reiserfs_unmap_buffer(bh);
2857                         else
2858                                 ret = 0;
2859                 }
2860                 curr_off = next_off;
2861                 bh = next;
2862         } while (bh != head);
2863
2864         /*
2865          * We release buffers only if the entire page is being invalidated.
2866          * The get_block cached value has been unconditionally invalidated,
2867          * so real IO is not possible anymore.
2868          */
2869         if (!offset && ret) {
2870                 ret = try_to_release_page(page, 0);
2871                 /* maybe should BUG_ON(!ret); - neilb */
2872         }
2873       out:
2874         return;
2875 }
2876
2877 static int reiserfs_set_page_dirty(struct page *page)
2878 {
2879         struct inode *inode = page->mapping->host;
2880         if (reiserfs_file_data_log(inode)) {
2881                 SetPageChecked(page);
2882                 return __set_page_dirty_nobuffers(page);
2883         }
2884         return __set_page_dirty_buffers(page);
2885 }
2886
2887 /*
2888  * Returns 1 if the page's buffers were dropped.  The page is locked.
2889  *
2890  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2891  * in the buffers at page_buffers(page).
2892  *
2893  * even in -o notail mode, we can't be sure an old mount without -o notail
2894  * didn't create files with tails.
2895  */
2896 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2897 {
2898         struct inode *inode = page->mapping->host;
2899         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2900         struct buffer_head *head;
2901         struct buffer_head *bh;
2902         int ret = 1;
2903
2904         WARN_ON(PageChecked(page));
2905         spin_lock(&j->j_dirty_buffers_lock);
2906         head = page_buffers(page);
2907         bh = head;
2908         do {
2909                 if (bh->b_private) {
2910                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2911                                 reiserfs_free_jh(bh);
2912                         } else {
2913                                 ret = 0;
2914                                 break;
2915                         }
2916                 }
2917                 bh = bh->b_this_page;
2918         } while (bh != head);
2919         if (ret)
2920                 ret = try_to_free_buffers(page);
2921         spin_unlock(&j->j_dirty_buffers_lock);
2922         return ret;
2923 }
2924
2925 /* We thank Mingming Cao for helping us understand in great detail what
2926    to do in this section of the code. */
2927 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2928                                   const struct iovec *iov, loff_t offset,
2929                                   unsigned long nr_segs)
2930 {
2931         struct file *file = iocb->ki_filp;
2932         struct inode *inode = file->f_mapping->host;
2933
2934         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2935                                   offset, nr_segs,
2936                                   reiserfs_get_blocks_direct_io, NULL);
2937 }
2938
2939 int reiserfs_sync_flags(struct inode *inode)
2940 {
2941         u16 oldflags, newflags;
2942
2943         oldflags = REISERFS_I(inode)->i_attrs;
2944         newflags = oldflags;
2945         i_attrs_to_sd_attrs(inode, &newflags);
2946
2947         if (oldflags ^ newflags) {
2948                 REISERFS_I(inode)->i_attrs = newflags;
2949                 inode->i_ctime = CURRENT_TIME_SEC;
2950                 mark_inode_dirty(inode);
2951         }
2952         return 0;
2953 }
2954
2955 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2956 {
2957         struct inode *inode = dentry->d_inode;
2958         int error;
2959         unsigned int ia_valid = attr->ia_valid;
2960         reiserfs_write_lock(inode->i_sb);
2961         if (attr->ia_valid & ATTR_SIZE) {
2962                 /* version 2 items will be caught by the s_maxbytes check
2963                  ** done for us in vmtruncate
2964                  */
2965                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2966                     attr->ia_size > MAX_NON_LFS) {
2967                         error = -EFBIG;
2968                         goto out;
2969                 }
2970                 /* fill in hole pointers in the expanding truncate case. */
2971                 if (attr->ia_size > inode->i_size) {
2972                         error = generic_cont_expand(inode, attr->ia_size);
2973                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2974                                 int err;
2975                                 struct reiserfs_transaction_handle th;
2976                                 /* we're changing at most 2 bitmaps, inode + super */
2977                                 err = journal_begin(&th, inode->i_sb, 4);
2978                                 if (!err) {
2979                                         reiserfs_discard_prealloc(&th, inode);
2980                                         err = journal_end(&th, inode->i_sb, 4);
2981                                 }
2982                                 if (err)
2983                                         error = err;
2984                         }
2985                         if (error)
2986                                 goto out;
2987                         /*
2988                          * file size is changed, ctime and mtime are
2989                          * to be updated
2990                          */
2991                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
2992                 }
2993         }
2994
2995         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2996              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2997             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2998                 /* stat data of format v3.5 has 16 bit uid and gid */
2999                 error = -EINVAL;
3000                 goto out;
3001         }
3002
3003         error = inode_change_ok(inode, attr);
3004
3005         if (!error) {
3006                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3007                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
3008                     (ia_valid & ATTR_TAG && attr->ia_tag != inode->i_tag)) {
3009                         error = reiserfs_chown_xattrs(inode, attr);
3010
3011                         if (!error) {
3012                                 struct reiserfs_transaction_handle th;
3013                                 int jbegin_count =
3014                                     2 *
3015                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3016                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3017                                     2;
3018
3019                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3020                                 error =
3021                                     journal_begin(&th, inode->i_sb,
3022                                                   jbegin_count);
3023                                 if (error)
3024                                         goto out;
3025                                 error =
3026                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3027                                 if (error) {
3028                                         journal_end(&th, inode->i_sb,
3029                                                     jbegin_count);
3030                                         goto out;
3031                                 }
3032                                 /* Update corresponding info in inode so that everything is in
3033                                  * one transaction */
3034                                 if (attr->ia_valid & ATTR_UID)
3035                                         inode->i_uid = attr->ia_uid;
3036                                 if (attr->ia_valid & ATTR_GID)
3037                                         inode->i_gid = attr->ia_gid;
3038                                 if ((attr->ia_valid & ATTR_TAG) &&
3039                                         IS_TAGGED(inode))
3040                                         inode->i_tag = attr->ia_tag;
3041                                 mark_inode_dirty(inode);
3042                                 error =
3043                                     journal_end(&th, inode->i_sb, jbegin_count);
3044                         }
3045                 }
3046                 if (!error)
3047                         error = inode_setattr(inode, attr);
3048         }
3049
3050         if (!error && reiserfs_posixacl(inode->i_sb)) {
3051                 if (attr->ia_valid & ATTR_MODE)
3052                         error = reiserfs_acl_chmod(inode);
3053         }
3054
3055       out:
3056         reiserfs_write_unlock(inode->i_sb);
3057         return error;
3058 }
3059
3060 const struct address_space_operations reiserfs_address_space_operations = {
3061         .writepage = reiserfs_writepage,
3062         .readpage = reiserfs_readpage,
3063         .readpages = reiserfs_readpages,
3064         .releasepage = reiserfs_releasepage,
3065         .invalidatepage = reiserfs_invalidatepage,
3066         .sync_page = block_sync_page,
3067         .prepare_write = reiserfs_prepare_write,
3068         .commit_write = reiserfs_commit_write,
3069         .bmap = reiserfs_aop_bmap,
3070         .direct_IO = reiserfs_direct_IO,
3071         .set_page_dirty = reiserfs_set_page_dirty,
3072 };