fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/exportfs.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <asm/uaccess.h>
  15 #include <asm/unaligned.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/mpage.h>
  18 #include <linux/writeback.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/swap.h>
  21
  22 int reiserfs_commit_write(struct file *f, struct page *page,
  23                           unsigned from, unsigned to);
  24 int reiserfs_prepare_write(struct file *f, struct page *page,
  25                            unsigned from, unsigned to);
  26
  27 void reiserfs_delete_inode(struct inode *inode)
  28 {
  29         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  30         int jbegin_count =
  31             JOURNAL_PER_BALANCE_CNT * 2 +
  32             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  33         struct reiserfs_transaction_handle th;
  34         int err;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37
  38         reiserfs_write_lock(inode->i_sb);
  39
  40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  42                 reiserfs_delete_xattrs(inode);
  43
  44                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  45                         goto out;
  46                 reiserfs_update_inode_transaction(inode);
  47
  48                 reiserfs_discard_prealloc(&th, inode);
  49
  50                 err = reiserfs_delete_object(&th, inode);
  51
  52                 /* Do quota update inside a transaction for journaled quotas. We must do that
  53                  * after delete_object so that quota updates go into the same transaction as
  54                  * stat data deletion */
  55                 if (!err)
  56                         DQUOT_FREE_INODE(inode);
  57
  58                 if (journal_end(&th, inode->i_sb, jbegin_count))
  59                         goto out;
  60
  61                 /* check return value from reiserfs_delete_object after
  62                  * ending the transaction
  63                  */
  64                 if (err)
  65                     goto out;
  66
  67                 /* all items of file are deleted, so we can remove "save" link */
  68                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  69                                                                  * about an error here */
  70         } else {
  71                 /* no object items are in the tree */
  72                 ;
  73         }
  74       out:
  75         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  76         inode->i_blocks = 0;
  77         reiserfs_write_unlock(inode->i_sb);
  78 }
  79
  80 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  81                           __u32 objectid, loff_t offset, int type, int length)
  82 {
  83         key->version = version;
  84
  85         key->on_disk_key.k_dir_id = dirid;
  86         key->on_disk_key.k_objectid = objectid;
  87         set_cpu_key_k_offset(key, offset);
  88         set_cpu_key_k_type(key, type);
  89         key->key_length = length;
  90 }
  91
  92 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  93    offset and type of key */
  94 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  95                   int type, int length)
  96 {
  97         _make_cpu_key(key, get_inode_item_key_version(inode),
  98                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
  99                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 100                       length);
 101 }
 102
 103 //
 104 // when key is 0, do not set version and short key
 105 //
 106 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 107                               int version,
 108                               loff_t offset, int type, int length,
 109                               int entry_count /*or ih_free_space */ )
 110 {
 111         if (key) {
 112                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 113                 ih->ih_key.k_objectid =
 114                     cpu_to_le32(key->on_disk_key.k_objectid);
 115         }
 116         put_ih_version(ih, version);
 117         set_le_ih_k_offset(ih, offset);
 118         set_le_ih_k_type(ih, type);
 119         put_ih_item_len(ih, length);
 120         /*    set_ih_free_space (ih, 0); */
 121         // for directory items it is entry count, for directs and stat
 122         // datas - 0xffff, for indirects - 0
 123         put_ih_entry_count(ih, entry_count);
 124 }
 125
 126 //
 127 // FIXME: we might cache recently accessed indirect item
 128
 129 // Ugh.  Not too eager for that....
 130 //  I cut the code until such time as I see a convincing argument (benchmark).
 131 // I don't want a bloated inode struct..., and I don't like code complexity....
 132
 133 /* cutting the code is fine, since it really isn't in use yet and is easy
 134 ** to add back in.  But, Vladimir has a really good idea here.  Think
 135 ** about what happens for reading a file.  For each page,
 136 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 137 ** an indirect item.  This indirect item has X number of pointers, where
 138 ** X is a big number if we've done the block allocation right.  But,
 139 ** we only use one or two of these pointers during each call to readpage,
 140 ** needlessly researching again later on.
 141 **
 142 ** The size of the cache could be dynamic based on the size of the file.
 143 **
 144 ** I'd also like to see us cache the location the stat data item, since
 145 ** we are needlessly researching for that frequently.
 146 **
 147 ** --chris
 148 */
 149
 150 /* If this page has a file tail in it, and
 151 ** it was read in by get_block_create_0, the page data is valid,
 152 ** but tail is still sitting in a direct item, and we can't write to
 153 ** it.  So, look through this page, and check all the mapped buffers
 154 ** to make sure they have valid block numbers.  Any that don't need
 155 ** to be unmapped, so that block_prepare_write will correctly call
 156 ** reiserfs_get_block to convert the tail into an unformatted node
 157 */
 158 static inline void fix_tail_page_for_writing(struct page *page)
 159 {
 160         struct buffer_head *head, *next, *bh;
 161
 162         if (page && page_has_buffers(page)) {
 163                 head = page_buffers(page);
 164                 bh = head;
 165                 do {
 166                         next = bh->b_this_page;
 167                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 168                                 reiserfs_unmap_buffer(bh);
 169                         }
 170                         bh = next;
 171                 } while (bh != head);
 172         }
 173 }
 174
 175 /* reiserfs_get_block does not need to allocate a block only if it has been
 176    done already or non-hole position has been found in the indirect item */
 177 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 178                                     struct item_head *ih,
 179                                     __le32 * item, int pos_in_item)
 180 {
 181         if (allocated)
 182                 return 0;
 183         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 184             get_block_num(item, pos_in_item))
 185                 return 0;
 186         return 1;
 187 }
 188
 189 static inline int indirect_item_found(int retval, struct item_head *ih)
 190 {
 191         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 192 }
 193
 194 static inline void set_block_dev_mapped(struct buffer_head *bh,
 195                                         b_blocknr_t block, struct inode *inode)
 196 {
 197         map_bh(bh, inode->i_sb, block);
 198 }
 199
 200 //
 201 // files which were created in the earlier version can not be longer,
 202 // than 2 gb
 203 //
 204 static int file_capable(struct inode *inode, sector_t block)
 205 {
 206         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 207             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 208                 return 1;
 209
 210         return 0;
 211 }
 212
 213 static int restart_transaction(struct reiserfs_transaction_handle *th,
 214                                struct inode *inode, struct treepath *path)
 215 {
 216         struct super_block *s = th->t_super;
 217         int len = th->t_blocks_allocated;
 218         int err;
 219
 220         BUG_ON(!th->t_trans_id);
 221         BUG_ON(!th->t_refcount);
 222
 223         pathrelse(path);
 224
 225         /* we cannot restart while nested */
 226         if (th->t_refcount > 1) {
 227                 return 0;
 228         }
 229         reiserfs_update_sd(th, inode);
 230         err = journal_end(th, s, len);
 231         if (!err) {
 232                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 233                 if (!err)
 234                         reiserfs_update_inode_transaction(inode);
 235         }
 236         return err;
 237 }
 238
 239 // it is called by get_block when create == 0. Returns block number
 240 // for 'block'-th logical block of file. When it hits direct item it
 241 // returns 0 (being called from bmap) or read direct item into piece
 242 // of page (bh_result)
 243
 244 // Please improve the english/clarity in the comment above, as it is
 245 // hard to understand.
 246
 247 static int _get_block_create_0(struct inode *inode, sector_t block,
 248                                struct buffer_head *bh_result, int args)
 249 {
 250         INITIALIZE_PATH(path);
 251         struct cpu_key key;
 252         struct buffer_head *bh;
 253         struct item_head *ih, tmp_ih;
 254         int fs_gen;
 255         b_blocknr_t blocknr;
 256         char *p = NULL;
 257         int chars;
 258         int ret;
 259         int result;
 260         int done = 0;
 261         unsigned long offset;
 262
 263         // prepare the key to look for the 'block'-th block of file
 264         make_cpu_key(&key, inode,
 265                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 266                      3);
 267
 268       research:
 269         result = search_for_position_by_key(inode->i_sb, &key, &path);
 270         if (result != POSITION_FOUND) {
 271                 pathrelse(&path);
 272                 if (p)
 273                         kunmap(bh_result->b_page);
 274                 if (result == IO_ERROR)
 275                         return -EIO;
 276                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 277                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 278                 if ((args & GET_BLOCK_NO_HOLE)
 279                     && !PageUptodate(bh_result->b_page)) {
 280                         return -ENOENT;
 281                 }
 282                 return 0;
 283         }
 284         //
 285         bh = get_last_bh(&path);
 286         ih = get_ih(&path);
 287         if (is_indirect_le_ih(ih)) {
 288                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 289
 290                 /* FIXME: here we could cache indirect item or part of it in
 291                    the inode to avoid search_by_key in case of subsequent
 292                    access to file */
 293                 blocknr = get_block_num(ind_item, path.pos_in_item);
 294                 ret = 0;
 295                 if (blocknr) {
 296                         map_bh(bh_result, inode->i_sb, blocknr);
 297                         if (path.pos_in_item ==
 298                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 299                                 set_buffer_boundary(bh_result);
 300                         }
 301                 } else
 302                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 303                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 304                 if ((args & GET_BLOCK_NO_HOLE)
 305                             && !PageUptodate(bh_result->b_page)) {
 306                         ret = -ENOENT;
 307                 }
 308
 309                 pathrelse(&path);
 310                 if (p)
 311                         kunmap(bh_result->b_page);
 312                 return ret;
 313         }
 314         // requested data are in direct item(s)
 315         if (!(args & GET_BLOCK_READ_DIRECT)) {
 316                 // we are called by bmap. FIXME: we can not map block of file
 317                 // when it is stored in direct item(s)
 318                 pathrelse(&path);
 319                 if (p)
 320                         kunmap(bh_result->b_page);
 321                 return -ENOENT;
 322         }
 323
 324         /* if we've got a direct item, and the buffer or page was uptodate,
 325          ** we don't want to pull data off disk again.  skip to the
 326          ** end, where we map the buffer and return
 327          */
 328         if (buffer_uptodate(bh_result)) {
 329                 goto finished;
 330         } else
 331                 /*
 332                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 333                  ** pages without any buffers.  If the page is up to date, we don't want
 334                  ** read old data off disk.  Set the up to date bit on the buffer instead
 335                  ** and jump to the end
 336                  */
 337         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 338                 set_buffer_uptodate(bh_result);
 339                 goto finished;
 340         }
 341         // read file tail into part of page
 342         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 343         fs_gen = get_generation(inode->i_sb);
 344         copy_item_head(&tmp_ih, ih);
 345
 346         /* we only want to kmap if we are reading the tail into the page.
 347          ** this is not the common case, so we don't kmap until we are
 348          ** sure we need to.  But, this means the item might move if
 349          ** kmap schedules
 350          */
 351         if (!p) {
 352                 p = (char *)kmap(bh_result->b_page);
 353                 if (fs_changed(fs_gen, inode->i_sb)
 354                     && item_moved(&tmp_ih, &path)) {
 355                         goto research;
 356                 }
 357         }
 358         p += offset;
 359         memset(p, 0, inode->i_sb->s_blocksize);
 360         do {
 361                 if (!is_direct_le_ih(ih)) {
 362                         BUG();
 363                 }
 364                 /* make sure we don't read more bytes than actually exist in
 365                  ** the file.  This can happen in odd cases where i_size isn't
 366                  ** correct, and when direct item padding results in a few
 367                  ** extra bytes at the end of the direct item
 368                  */
 369                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 370                         break;
 371                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 372                         chars =
 373                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 374                             path.pos_in_item;
 375                         done = 1;
 376                 } else {
 377                         chars = ih_item_len(ih) - path.pos_in_item;
 378                 }
 379                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 380
 381                 if (done)
 382                         break;
 383
 384                 p += chars;
 385
 386                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 387                         // we done, if read direct item is not the last item of
 388                         // node FIXME: we could try to check right delimiting key
 389                         // to see whether direct item continues in the right
 390                         // neighbor or rely on i_size
 391                         break;
 392
 393                 // update key to look for the next piece
 394                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 395                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 396                 if (result != POSITION_FOUND)
 397                         // i/o error most likely
 398                         break;
 399                 bh = get_last_bh(&path);
 400                 ih = get_ih(&path);
 401         } while (1);
 402
 403         flush_dcache_page(bh_result->b_page);
 404         kunmap(bh_result->b_page);
 405
 406       finished:
 407         pathrelse(&path);
 408
 409         if (result == IO_ERROR)
 410                 return -EIO;
 411
 412         /* this buffer has valid data, but isn't valid for io.  mapping it to
 413          * block #0 tells the rest of reiserfs it just has a tail in it
 414          */
 415         map_bh(bh_result, inode->i_sb, 0);
 416         set_buffer_uptodate(bh_result);
 417         return 0;
 418 }
 419
 420 // this is called to create file map. So, _get_block_create_0 will not
 421 // read direct item
 422 static int reiserfs_bmap(struct inode *inode, sector_t block,
 423                          struct buffer_head *bh_result, int create)
 424 {
 425         if (!file_capable(inode, block))
 426                 return -EFBIG;
 427
 428         reiserfs_write_lock(inode->i_sb);
 429         /* do not read the direct item */
 430         _get_block_create_0(inode, block, bh_result, 0);
 431         reiserfs_write_unlock(inode->i_sb);
 432         return 0;
 433 }
 434
 435 /* special version of get_block that is only used by grab_tail_page right
 436 ** now.  It is sent to block_prepare_write, and when you try to get a
 437 ** block past the end of the file (or a block from a hole) it returns
 438 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 439 ** be able to do i/o on the buffers returned, unless an error value
 440 ** is also returned.
 441 **
 442 ** So, this allows block_prepare_write to be used for reading a single block
 443 ** in a page.  Where it does not produce a valid page for holes, or past the
 444 ** end of the file.  This turns out to be exactly what we need for reading
 445 ** tails for conversion.
 446 **
 447 ** The point of the wrapper is forcing a certain value for create, even
 448 ** though the VFS layer is calling this function with create==1.  If you
 449 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 450 ** don't use this function.
 451 */
 452 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 453                                        struct buffer_head *bh_result,
 454                                        int create)
 455 {
 456         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 457 }
 458
 459 /* This is special helper for reiserfs_get_block in case we are executing
 460    direct_IO request. */
 461 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 462                                          sector_t iblock,
 463                                          struct buffer_head *bh_result,
 464                                          int create)
 465 {
 466         int ret;
 467
 468         bh_result->b_page = NULL;
 469
 470         /* We set the b_size before reiserfs_get_block call since it is
 471            referenced in convert_tail_for_hole() that may be called from
 472            reiserfs_get_block() */
 473         bh_result->b_size = (1 << inode->i_blkbits);
 474
 475         ret = reiserfs_get_block(inode, iblock, bh_result,
 476                                  create | GET_BLOCK_NO_DANGLE);
 477         if (ret)
 478                 goto out;
 479
 480         /* don't allow direct io onto tail pages */
 481         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 482                 /* make sure future calls to the direct io funcs for this offset
 483                  ** in the file fail by unmapping the buffer
 484                  */
 485                 clear_buffer_mapped(bh_result);
 486                 ret = -EINVAL;
 487         }
 488         /* Possible unpacked tail. Flush the data before pages have
 489            disappeared */
 490         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 491                 int err;
 492                 lock_kernel();
 493                 err = reiserfs_commit_for_inode(inode);
 494                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 495                 unlock_kernel();
 496                 if (err < 0)
 497                         ret = err;
 498         }
 499       out:
 500         return ret;
 501 }
 502
 503 /*
 504 ** helper function for when reiserfs_get_block is called for a hole
 505 ** but the file tail is still in a direct item
 506 ** bh_result is the buffer head for the hole
 507 ** tail_offset is the offset of the start of the tail in the file
 508 **
 509 ** This calls prepare_write, which will start a new transaction
 510 ** you should not be in a transaction, or have any paths held when you
 511 ** call this.
 512 */
 513 static int convert_tail_for_hole(struct inode *inode,
 514                                  struct buffer_head *bh_result,
 515                                  loff_t tail_offset)
 516 {
 517         unsigned long index;
 518         unsigned long tail_end;
 519         unsigned long tail_start;
 520         struct page *tail_page;
 521         struct page *hole_page = bh_result->b_page;
 522         int retval = 0;
 523
 524         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 525                 return -EIO;
 526
 527         /* always try to read until the end of the block */
 528         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 529         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 530
 531         index = tail_offset >> PAGE_CACHE_SHIFT;
 532         /* hole_page can be zero in case of direct_io, we are sure
 533            that we cannot get here if we write with O_DIRECT into
 534            tail page */
 535         if (!hole_page || index != hole_page->index) {
 536                 tail_page = grab_cache_page(inode->i_mapping, index);
 537                 retval = -ENOMEM;
 538                 if (!tail_page) {
 539                         goto out;
 540                 }
 541         } else {
 542                 tail_page = hole_page;
 543         }
 544
 545         /* we don't have to make sure the conversion did not happen while
 546          ** we were locking the page because anyone that could convert
 547          ** must first take i_mutex.
 548          **
 549          ** We must fix the tail page for writing because it might have buffers
 550          ** that are mapped, but have a block number of 0.  This indicates tail
 551          ** data that has been read directly into the page, and block_prepare_write
 552          ** won't trigger a get_block in this case.
 553          */
 554         fix_tail_page_for_writing(tail_page);
 555         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 556         if (retval)
 557                 goto unlock;
 558
 559         /* tail conversion might change the data in the page */
 560         flush_dcache_page(tail_page);
 561
 562         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 563
 564       unlock:
 565         if (tail_page != hole_page) {
 566                 unlock_page(tail_page);
 567                 page_cache_release(tail_page);
 568         }
 569       out:
 570         return retval;
 571 }
 572
 573 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 574                                   sector_t block,
 575                                   struct inode *inode,
 576                                   b_blocknr_t * allocated_block_nr,
 577                                   struct treepath *path, int flags)
 578 {
 579         BUG_ON(!th->t_trans_id);
 580
 581 #ifdef REISERFS_PREALLOCATE
 582         if (!(flags & GET_BLOCK_NO_IMUX)) {
 583                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 584                                                   path, block);
 585         }
 586 #endif
 587         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 588                                          block);
 589 }
 590
 591 int reiserfs_get_block(struct inode *inode, sector_t block,
 592                        struct buffer_head *bh_result, int create)
 593 {
 594         int repeat, retval = 0;
 595         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 596         INITIALIZE_PATH(path);
 597         int pos_in_item;
 598         struct cpu_key key;
 599         struct buffer_head *bh, *unbh = NULL;
 600         struct item_head *ih, tmp_ih;
 601         __le32 *item;
 602         int done;
 603         int fs_gen;
 604         struct reiserfs_transaction_handle *th = NULL;
 605         /* space reserved in transaction batch:
 606            . 3 balancings in direct->indirect conversion
 607            . 1 block involved into reiserfs_update_sd()
 608            XXX in practically impossible worst case direct2indirect()
 609            can incur (much) more than 3 balancings.
 610            quota update for user, group */
 611         int jbegin_count =
 612             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 613             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 614         int version;
 615         int dangle = 1;
 616         loff_t new_offset =
 617             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 618
 619         /* bad.... */
 620         reiserfs_write_lock(inode->i_sb);
 621         version = get_inode_item_key_version(inode);
 622
 623         if (!file_capable(inode, block)) {
 624                 reiserfs_write_unlock(inode->i_sb);
 625                 return -EFBIG;
 626         }
 627
 628         /* if !create, we aren't changing the FS, so we don't need to
 629          ** log anything, so we don't need to start a transaction
 630          */
 631         if (!(create & GET_BLOCK_CREATE)) {
 632                 int ret;
 633                 /* find number of block-th logical block of the file */
 634                 ret = _get_block_create_0(inode, block, bh_result,
 635                                           create | GET_BLOCK_READ_DIRECT);
 636                 reiserfs_write_unlock(inode->i_sb);
 637                 return ret;
 638         }
 639         /*
 640          * if we're already in a transaction, make sure to close
 641          * any new transactions we start in this func
 642          */
 643         if ((create & GET_BLOCK_NO_DANGLE) ||
 644             reiserfs_transaction_running(inode->i_sb))
 645                 dangle = 0;
 646
 647         /* If file is of such a size, that it might have a tail and tails are enabled
 648          ** we should mark it as possibly needing tail packing on close
 649          */
 650         if ((have_large_tails(inode->i_sb)
 651              && inode->i_size < i_block_size(inode) * 4)
 652             || (have_small_tails(inode->i_sb)
 653                 && inode->i_size < i_block_size(inode)))
 654                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 655
 656         /* set the key of the first byte in the 'block'-th block of file */
 657         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 658         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 659               start_trans:
 660                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 661                 if (!th) {
 662                         retval = -ENOMEM;
 663                         goto failure;
 664                 }
 665                 reiserfs_update_inode_transaction(inode);
 666         }
 667       research:
 668
 669         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 670         if (retval == IO_ERROR) {
 671                 retval = -EIO;
 672                 goto failure;
 673         }
 674
 675         bh = get_last_bh(&path);
 676         ih = get_ih(&path);
 677         item = get_item(&path);
 678         pos_in_item = path.pos_in_item;
 679
 680         fs_gen = get_generation(inode->i_sb);
 681         copy_item_head(&tmp_ih, ih);
 682
 683         if (allocation_needed
 684             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 685                 /* we have to allocate block for the unformatted node */
 686                 if (!th) {
 687                         pathrelse(&path);
 688                         goto start_trans;
 689                 }
 690
 691                 repeat =
 692                     _allocate_block(th, block, inode, &allocated_block_nr,
 693                                     &path, create);
 694
 695                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 696                         /* restart the transaction to give the journal a chance to free
 697                          ** some blocks.  releases the path, so we have to go back to
 698                          ** research if we succeed on the second try
 699                          */
 700                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 701                         retval = restart_transaction(th, inode, &path);
 702                         if (retval)
 703                                 goto failure;
 704                         repeat =
 705                             _allocate_block(th, block, inode,
 706                                             &allocated_block_nr, NULL, create);
 707
 708                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 709                                 goto research;
 710                         }
 711                         if (repeat == QUOTA_EXCEEDED)
 712                                 retval = -EDQUOT;
 713                         else
 714                                 retval = -ENOSPC;
 715                         goto failure;
 716                 }
 717
 718                 if (fs_changed(fs_gen, inode->i_sb)
 719                     && item_moved(&tmp_ih, &path)) {
 720                         goto research;
 721                 }
 722         }
 723
 724         if (indirect_item_found(retval, ih)) {
 725                 b_blocknr_t unfm_ptr;
 726                 /* 'block'-th block is in the file already (there is
 727                    corresponding cell in some indirect item). But it may be
 728                    zero unformatted node pointer (hole) */
 729                 unfm_ptr = get_block_num(item, pos_in_item);
 730                 if (unfm_ptr == 0) {
 731                         /* use allocated block to plug the hole */
 732                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 733                         if (fs_changed(fs_gen, inode->i_sb)
 734                             && item_moved(&tmp_ih, &path)) {
 735                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 736                                                                  bh);
 737                                 goto research;
 738                         }
 739                         set_buffer_new(bh_result);
 740                         if (buffer_dirty(bh_result)
 741                             && reiserfs_data_ordered(inode->i_sb))
 742                                 reiserfs_add_ordered_list(inode, bh_result);
 743                         put_block_num(item, pos_in_item, allocated_block_nr);
 744                         unfm_ptr = allocated_block_nr;
 745                         journal_mark_dirty(th, inode->i_sb, bh);
 746                         reiserfs_update_sd(th, inode);
 747                 }
 748                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 749                 pathrelse(&path);
 750                 retval = 0;
 751                 if (!dangle && th)
 752                         retval = reiserfs_end_persistent_transaction(th);
 753
 754                 reiserfs_write_unlock(inode->i_sb);
 755
 756                 /* the item was found, so new blocks were not added to the file
 757                  ** there is no need to make sure the inode is updated with this
 758                  ** transaction
 759                  */
 760                 return retval;
 761         }
 762
 763         if (!th) {
 764                 pathrelse(&path);
 765                 goto start_trans;
 766         }
 767
 768         /* desired position is not found or is in the direct item. We have
 769            to append file with holes up to 'block'-th block converting
 770            direct items to indirect one if necessary */
 771         done = 0;
 772         do {
 773                 if (is_statdata_le_ih(ih)) {
 774                         __le32 unp = 0;
 775                         struct cpu_key tmp_key;
 776
 777                         /* indirect item has to be inserted */
 778                         make_le_item_head(&tmp_ih, &key, version, 1,
 779                                           TYPE_INDIRECT, UNFM_P_SIZE,
 780                                           0 /* free_space */ );
 781
 782                         if (cpu_key_k_offset(&key) == 1) {
 783                                 /* we are going to add 'block'-th block to the file. Use
 784                                    allocated block for that */
 785                                 unp = cpu_to_le32(allocated_block_nr);
 786                                 set_block_dev_mapped(bh_result,
 787                                                      allocated_block_nr, inode);
 788                                 set_buffer_new(bh_result);
 789                                 done = 1;
 790                         }
 791                         tmp_key = key;  // ;)
 792                         set_cpu_key_k_offset(&tmp_key, 1);
 793                         PATH_LAST_POSITION(&path)++;
 794
 795                         retval =
 796                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 797                                                  inode, (char *)&unp);
 798                         if (retval) {
 799                                 reiserfs_free_block(th, inode,
 800                                                     allocated_block_nr, 1);
 801                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 802                         }
 803                         //mark_tail_converted (inode);
 804                 } else if (is_direct_le_ih(ih)) {
 805                         /* direct item has to be converted */
 806                         loff_t tail_offset;
 807
 808                         tail_offset =
 809                             ((le_ih_k_offset(ih) -
 810                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 811                         if (tail_offset == cpu_key_k_offset(&key)) {
 812                                 /* direct item we just found fits into block we have
 813                                    to map. Convert it into unformatted node: use
 814                                    bh_result for the conversion */
 815                                 set_block_dev_mapped(bh_result,
 816                                                      allocated_block_nr, inode);
 817                                 unbh = bh_result;
 818                                 done = 1;
 819                         } else {
 820                                 /* we have to padd file tail stored in direct item(s)
 821                                    up to block size and convert it to unformatted
 822                                    node. FIXME: this should also get into page cache */
 823
 824                                 pathrelse(&path);
 825                                 /*
 826                                  * ugly, but we can only end the transaction if
 827                                  * we aren't nested
 828                                  */
 829                                 BUG_ON(!th->t_refcount);
 830                                 if (th->t_refcount == 1) {
 831                                         retval =
 832                                             reiserfs_end_persistent_transaction
 833                                             (th);
 834                                         th = NULL;
 835                                         if (retval)
 836                                                 goto failure;
 837                                 }
 838
 839                                 retval =
 840                                     convert_tail_for_hole(inode, bh_result,
 841                                                           tail_offset);
 842                                 if (retval) {
 843                                         if (retval != -ENOSPC)
 844                                                 reiserfs_warning(inode->i_sb,
 845                                                          "clm-6004",
 846                                                          "convert tail failed "
 847                                                          "inode %lu, error %d",
 848                                                                  inode->i_ino,
 849                                                                  retval);
 850                                         if (allocated_block_nr) {
 851                                                 /* the bitmap, the super, and the stat data == 3 */
 852                                                 if (!th)
 853                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 854                                                 if (th)
 855                                                         reiserfs_free_block(th,
 856                                                                             inode,
 857                                                                             allocated_block_nr,
 858                                                                             1);
 859                                         }
 860                                         goto failure;
 861                                 }
 862                                 goto research;
 863                         }
 864                         retval =
 865                             direct2indirect(th, inode, &path, unbh,
 866                                             tail_offset);
 867                         if (retval) {
 868                                 reiserfs_unmap_buffer(unbh);
 869                                 reiserfs_free_block(th, inode,
 870                                                     allocated_block_nr, 1);
 871                                 goto failure;
 872                         }
 873                         /* it is important the set_buffer_uptodate is done after
 874                          ** the direct2indirect.  The buffer might contain valid
 875                          ** data newer than the data on disk (read by readpage, changed,
 876                          ** and then sent here by writepage).  direct2indirect needs
 877                          ** to know if unbh was already up to date, so it can decide
 878                          ** if the data in unbh needs to be replaced with data from
 879                          ** the disk
 880                          */
 881                         set_buffer_uptodate(unbh);
 882
 883                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 884                            buffer will disappear shortly, so it should not be added to
 885                          */
 886                         if (unbh->b_page) {
 887                                 /* we've converted the tail, so we must
 888                                  ** flush unbh before the transaction commits
 889                                  */
 890                                 reiserfs_add_tail_list(inode, unbh);
 891
 892                                 /* mark it dirty now to prevent commit_write from adding
 893                                  ** this buffer to the inode's dirty buffer list
 894                                  */
 895                                 /*
 896                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 897                                  * It's still atomic, but it sets the page dirty too,
 898                                  * which makes it eligible for writeback at any time by the
 899                                  * VM (which was also the case with __mark_buffer_dirty())
 900                                  */
 901                                 mark_buffer_dirty(unbh);
 902                         }
 903                 } else {
 904                         /* append indirect item with holes if needed, when appending
 905                            pointer to 'block'-th block use block, which is already
 906                            allocated */
 907                         struct cpu_key tmp_key;
 908                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 909                         // one block which is a fastpath
 910                         unp_t *un;
 911                         __u64 max_to_insert =
 912                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 913                             UNFM_P_SIZE;
 914                         __u64 blocks_needed;
 915
 916                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 917                                "vs-804: invalid position for append");
 918                         /* indirect item has to be appended, set up key of that position */
 919                         make_cpu_key(&tmp_key, inode,
 920                                      le_key_k_offset(version,
 921                                                      &(ih->ih_key)) +
 922                                      op_bytes_number(ih,
 923                                                      inode->i_sb->s_blocksize),
 924                                      //pos_in_item * inode->i_sb->s_blocksize,
 925                                      TYPE_INDIRECT, 3); // key type is unimportant
 926
 927                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 928                                "green-805: invalid offset");
 929                         blocks_needed =
 930                             1 +
 931                             ((cpu_key_k_offset(&key) -
 932                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 933                              s_blocksize_bits);
 934
 935                         if (blocks_needed == 1) {
 936                                 un = &unf_single;
 937                         } else {
 938                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 939                                 if (!un) {
 940                                         un = &unf_single;
 941                                         blocks_needed = 1;
 942                                         max_to_insert = 0;
 943                                 }
 944                         }
 945                         if (blocks_needed <= max_to_insert) {
 946                                 /* we are going to add target block to the file. Use allocated
 947                                    block for that */
 948                                 un[blocks_needed - 1] =
 949                                     cpu_to_le32(allocated_block_nr);
 950                                 set_block_dev_mapped(bh_result,
 951                                                      allocated_block_nr, inode);
 952                                 set_buffer_new(bh_result);
 953                                 done = 1;
 954                         } else {
 955                                 /* paste hole to the indirect item */
 956                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 957                                    only have space for one block */
 958                                 blocks_needed =
 959                                     max_to_insert ? max_to_insert : 1;
 960                         }
 961                         retval =
 962                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 963                                                      (char *)un,
 964                                                      UNFM_P_SIZE *
 965                                                      blocks_needed);
 966
 967                         if (blocks_needed != 1)
 968                                 kfree(un);
 969
 970                         if (retval) {
 971                                 reiserfs_free_block(th, inode,
 972                                                     allocated_block_nr, 1);
 973                                 goto failure;
 974                         }
 975                         if (!done) {
 976                                 /* We need to mark new file size in case this function will be
 977                                    interrupted/aborted later on. And we may do this only for
 978                                    holes. */
 979                                 inode->i_size +=
 980                                     inode->i_sb->s_blocksize * blocks_needed;
 981                         }
 982                 }
 983
 984                 if (done == 1)
 985                         break;
 986
 987                 /* this loop could log more blocks than we had originally asked
 988                  ** for.  So, we have to allow the transaction to end if it is
 989                  ** too big or too full.  Update the inode so things are
 990                  ** consistent if we crash before the function returns
 991                  **
 992                  ** release the path so that anybody waiting on the path before
 993                  ** ending their transaction will be able to continue.
 994                  */
 995                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 996                         retval = restart_transaction(th, inode, &path);
 997                         if (retval)
 998                                 goto failure;
 999                 }
1000                 /* inserting indirect pointers for a hole can take a
1001                  ** long time.  reschedule if needed
1002                  */
1003                 cond_resched();
1004
1005                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1006                 if (retval == IO_ERROR) {
1007                         retval = -EIO;
1008                         goto failure;
1009                 }
1010                 if (retval == POSITION_FOUND) {
1011                         reiserfs_warning(inode->i_sb, "vs-825",
1012                                          "%K should not be found", &key);
1013                         retval = -EEXIST;
1014                         if (allocated_block_nr)
1015                                 reiserfs_free_block(th, inode,
1016                                                     allocated_block_nr, 1);
1017                         pathrelse(&path);
1018                         goto failure;
1019                 }
1020                 bh = get_last_bh(&path);
1021                 ih = get_ih(&path);
1022                 item = get_item(&path);
1023                 pos_in_item = path.pos_in_item;
1024         } while (1);
1025
1026         retval = 0;
1027
1028       failure:
1029         if (th && (!dangle || (retval && !th->t_trans_id))) {
1030                 int err;
1031                 if (th->t_trans_id)
1032                         reiserfs_update_sd(th, inode);
1033                 err = reiserfs_end_persistent_transaction(th);
1034                 if (err)
1035                         retval = err;
1036         }
1037
1038         reiserfs_write_unlock(inode->i_sb);
1039         reiserfs_check_path(&path);
1040         return retval;
1041 }
1042
1043 static int
1044 reiserfs_readpages(struct file *file, struct address_space *mapping,
1045                    struct list_head *pages, unsigned nr_pages)
1046 {
1047         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1048 }
1049
1050 /* Compute real number of used bytes by file
1051  * Following three functions can go away when we'll have enough space in stat item
1052  */
1053 static int real_space_diff(struct inode *inode, int sd_size)
1054 {
1055         int bytes;
1056         loff_t blocksize = inode->i_sb->s_blocksize;
1057
1058         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1059                 return sd_size;
1060
1061         /* End of file is also in full block with indirect reference, so round
1062          ** up to the next block.
1063          **
1064          ** there is just no way to know if the tail is actually packed
1065          ** on the file, so we have to assume it isn't.  When we pack the
1066          ** tail, we add 4 bytes to pretend there really is an unformatted
1067          ** node pointer
1068          */
1069         bytes =
1070             ((inode->i_size +
1071               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1072             sd_size;
1073         return bytes;
1074 }
1075
1076 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1077                                         int sd_size)
1078 {
1079         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1080                 return inode->i_size +
1081                     (loff_t) (real_space_diff(inode, sd_size));
1082         }
1083         return ((loff_t) real_space_diff(inode, sd_size)) +
1084             (((loff_t) blocks) << 9);
1085 }
1086
1087 /* Compute number of blocks used by file in ReiserFS counting */
1088 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1089 {
1090         loff_t bytes = inode_get_bytes(inode);
1091         loff_t real_space = real_space_diff(inode, sd_size);
1092
1093         /* keeps fsck and non-quota versions of reiserfs happy */
1094         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1095                 bytes += (loff_t) 511;
1096         }
1097
1098         /* files from before the quota patch might i_blocks such that
1099          ** bytes < real_space.  Deal with that here to prevent it from
1100          ** going negative.
1101          */
1102         if (bytes < real_space)
1103                 return 0;
1104         return (bytes - real_space) >> 9;
1105 }
1106
1107 //
1108 // BAD: new directories have stat data of new type and all other items
1109 // of old type. Version stored in the inode says about body items, so
1110 // in update_stat_data we can not rely on inode, but have to check
1111 // item version directly
1112 //
1113
1114 // called by read_locked_inode
1115 static void init_inode(struct inode *inode, struct treepath *path)
1116 {
1117         struct buffer_head *bh;
1118         struct item_head *ih;
1119         __u32 rdev;
1120         //int version = ITEM_VERSION_1;
1121
1122         bh = PATH_PLAST_BUFFER(path);
1123         ih = PATH_PITEM_HEAD(path);
1124
1125         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1126
1127         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1128         REISERFS_I(inode)->i_flags = 0;
1129         REISERFS_I(inode)->i_prealloc_block = 0;
1130         REISERFS_I(inode)->i_prealloc_count = 0;
1131         REISERFS_I(inode)->i_trans_id = 0;
1132         REISERFS_I(inode)->i_jl = NULL;
1133         mutex_init(&(REISERFS_I(inode)->i_mmap));
1134         reiserfs_init_acl_access(inode);
1135         reiserfs_init_acl_default(inode);
1136         reiserfs_init_xattr_rwsem(inode);
1137
1138         if (stat_data_v1(ih)) {
1139                 struct stat_data_v1 *sd =
1140                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1141                 unsigned long blocks;
1142
1143                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1144                 set_inode_sd_version(inode, STAT_DATA_V1);
1145                 inode->i_mode = sd_v1_mode(sd);
1146                 inode->i_nlink = sd_v1_nlink(sd);
1147                 inode->i_uid = sd_v1_uid(sd);
1148                 inode->i_gid = sd_v1_gid(sd);
1149                 inode->i_size = sd_v1_size(sd);
1150                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1151                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1152                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1153                 inode->i_atime.tv_nsec = 0;
1154                 inode->i_ctime.tv_nsec = 0;
1155                 inode->i_mtime.tv_nsec = 0;
1156
1157                 inode->i_blocks = sd_v1_blocks(sd);
1158                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1159                 blocks = (inode->i_size + 511) >> 9;
1160                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1161                 if (inode->i_blocks > blocks) {
1162                         // there was a bug in <=3.5.23 when i_blocks could take negative
1163                         // values. Starting from 3.5.17 this value could even be stored in
1164                         // stat data. For such files we set i_blocks based on file
1165                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1166                         // only updated if file's inode will ever change
1167                         inode->i_blocks = blocks;
1168                 }
1169
1170                 rdev = sd_v1_rdev(sd);
1171                 REISERFS_I(inode)->i_first_direct_byte =
1172                     sd_v1_first_direct_byte(sd);
1173                 /* an early bug in the quota code can give us an odd number for the
1174                  ** block count.  This is incorrect, fix it here.
1175                  */
1176                 if (inode->i_blocks & 1) {
1177                         inode->i_blocks++;
1178                 }
1179                 inode_set_bytes(inode,
1180                                 to_real_used_space(inode, inode->i_blocks,
1181                                                    SD_V1_SIZE));
1182                 /* nopack is initially zero for v1 objects. For v2 objects,
1183                    nopack is initialised from sd_attrs */
1184                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1185         } else {
1186                 // new stat data found, but object may have old items
1187                 // (directories and symlinks)
1188                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1189
1190                 inode->i_mode = sd_v2_mode(sd);
1191                 inode->i_nlink = sd_v2_nlink(sd);
1192                 inode->i_uid = sd_v2_uid(sd);
1193                 inode->i_size = sd_v2_size(sd);
1194                 inode->i_gid = sd_v2_gid(sd);
1195                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1196                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1197                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1198                 inode->i_ctime.tv_nsec = 0;
1199                 inode->i_mtime.tv_nsec = 0;
1200                 inode->i_atime.tv_nsec = 0;
1201                 inode->i_blocks = sd_v2_blocks(sd);
1202                 rdev = sd_v2_rdev(sd);
1203                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1204                         inode->i_generation =
1205                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1206                 else
1207                         inode->i_generation = sd_v2_generation(sd);
1208
1209                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1210                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1211                 else
1212                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1213                 REISERFS_I(inode)->i_first_direct_byte = 0;
1214                 set_inode_sd_version(inode, STAT_DATA_V2);
1215                 inode_set_bytes(inode,
1216                                 to_real_used_space(inode, inode->i_blocks,
1217                                                    SD_V2_SIZE));
1218                 /* read persistent inode attributes from sd and initalise
1219                    generic inode flags from them */
1220                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1221                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1222         }
1223
1224         pathrelse(path);
1225         if (S_ISREG(inode->i_mode)) {
1226                 inode->i_op = &reiserfs_file_inode_operations;
1227                 inode->i_fop = &reiserfs_file_operations;
1228                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1229         } else if (S_ISDIR(inode->i_mode)) {
1230                 inode->i_op = &reiserfs_dir_inode_operations;
1231                 inode->i_fop = &reiserfs_dir_operations;
1232         } else if (S_ISLNK(inode->i_mode)) {
1233                 inode->i_op = &reiserfs_symlink_inode_operations;
1234                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1235         } else {
1236                 inode->i_blocks = 0;
1237                 inode->i_op = &reiserfs_special_inode_operations;
1238                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1239         }
1240 }
1241
1242 // update new stat data with inode fields
1243 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1244 {
1245         struct stat_data *sd_v2 = (struct stat_data *)sd;
1246         __u16 flags;
1247
1248         set_sd_v2_mode(sd_v2, inode->i_mode);
1249         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1250         set_sd_v2_uid(sd_v2, inode->i_uid);
1251         set_sd_v2_size(sd_v2, size);
1252         set_sd_v2_gid(sd_v2, inode->i_gid);
1253         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1254         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1255         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1256         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1257         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1258                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1259         else
1260                 set_sd_v2_generation(sd_v2, inode->i_generation);
1261         flags = REISERFS_I(inode)->i_attrs;
1262         i_attrs_to_sd_attrs(inode, &flags);
1263         set_sd_v2_attrs(sd_v2, flags);
1264 }
1265
1266 // used to copy inode's fields to old stat data
1267 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1268 {
1269         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1270
1271         set_sd_v1_mode(sd_v1, inode->i_mode);
1272         set_sd_v1_uid(sd_v1, inode->i_uid);
1273         set_sd_v1_gid(sd_v1, inode->i_gid);
1274         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1275         set_sd_v1_size(sd_v1, size);
1276         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1277         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1278         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1279
1280         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1281                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1282         else
1283                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1284
1285         // Sigh. i_first_direct_byte is back
1286         set_sd_v1_first_direct_byte(sd_v1,
1287                                     REISERFS_I(inode)->i_first_direct_byte);
1288 }
1289
1290 /* NOTE, you must prepare the buffer head before sending it here,
1291 ** and then log it after the call
1292 */
1293 static void update_stat_data(struct treepath *path, struct inode *inode,
1294                              loff_t size)
1295 {
1296         struct buffer_head *bh;
1297         struct item_head *ih;
1298
1299         bh = PATH_PLAST_BUFFER(path);
1300         ih = PATH_PITEM_HEAD(path);
1301
1302         if (!is_statdata_le_ih(ih))
1303                 reiserfs_panic(inode->i_sb,
1304                                "vs-13065: update_stat_data: key %k, found item %h",
1305                                INODE_PKEY(inode), ih);
1306
1307         if (stat_data_v1(ih)) {
1308                 // path points to old stat data
1309                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1310         } else {
1311                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1312         }
1313
1314         return;
1315 }
1316
1317 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1318                              struct inode *inode, loff_t size)
1319 {
1320         struct cpu_key key;
1321         INITIALIZE_PATH(path);
1322         struct buffer_head *bh;
1323         int fs_gen;
1324         struct item_head *ih, tmp_ih;
1325         int retval;
1326
1327         BUG_ON(!th->t_trans_id);
1328
1329         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1330
1331         for (;;) {
1332                 int pos;
1333                 /* look for the object's stat data */
1334                 retval = search_item(inode->i_sb, &key, &path);
1335                 if (retval == IO_ERROR) {
1336                         reiserfs_warning(inode->i_sb, "vs-13050",
1337                                          "i/o failure occurred trying to "
1338                                          "update %K stat data",
1339                                          &key);
1340                         return;
1341                 }
1342                 if (retval == ITEM_NOT_FOUND) {
1343                         pos = PATH_LAST_POSITION(&path);
1344                         pathrelse(&path);
1345                         if (inode->i_nlink == 0) {
1346                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1347                                 return;
1348                         }
1349                         reiserfs_warning(inode->i_sb, "vs-13060",
1350                                          "stat data of object %k (nlink == %d) "
1351                                          "not found (pos %d)",
1352                                          INODE_PKEY(inode), inode->i_nlink,
1353                                          pos);
1354                         reiserfs_check_path(&path);
1355                         return;
1356                 }
1357
1358                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1359                  ** FS might change.  We have to detect that, and loop back to the
1360                  ** search if the stat data item has moved
1361                  */
1362                 bh = get_last_bh(&path);
1363                 ih = get_ih(&path);
1364                 copy_item_head(&tmp_ih, ih);
1365                 fs_gen = get_generation(inode->i_sb);
1366                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1367                 if (fs_changed(fs_gen, inode->i_sb)
1368                     && item_moved(&tmp_ih, &path)) {
1369                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1370                         continue;       /* Stat_data item has been moved after scheduling. */
1371                 }
1372                 break;
1373         }
1374         update_stat_data(&path, inode, size);
1375         journal_mark_dirty(th, th->t_super, bh);
1376         pathrelse(&path);
1377         return;
1378 }
1379
1380 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1381 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1382 ** and clear the key in the private portion of the inode, otherwise a
1383 ** corresponding iput might try to delete whatever object the inode last
1384 ** represented.
1385 */
1386 static void reiserfs_make_bad_inode(struct inode *inode)
1387 {
1388         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1389         make_bad_inode(inode);
1390 }
1391
1392 //
1393 // initially this function was derived from minix or ext2's analog and
1394 // evolved as the prototype did
1395 //
1396
1397 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1398 {
1399         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1400         inode->i_ino = args->objectid;
1401         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1402         return 0;
1403 }
1404
1405 /* looks for stat data in the tree, and fills up the fields of in-core
1406    inode stat data fields */
1407 void reiserfs_read_locked_inode(struct inode *inode,
1408                                 struct reiserfs_iget_args *args)
1409 {
1410         INITIALIZE_PATH(path_to_sd);
1411         struct cpu_key key;
1412         unsigned long dirino;
1413         int retval;
1414
1415         dirino = args->dirid;
1416
1417         /* set version 1, version 2 could be used too, because stat data
1418            key is the same in both versions */
1419         key.version = KEY_FORMAT_3_5;
1420         key.on_disk_key.k_dir_id = dirino;
1421         key.on_disk_key.k_objectid = inode->i_ino;
1422         key.on_disk_key.k_offset = 0;
1423         key.on_disk_key.k_type = 0;
1424
1425         /* look for the object's stat data */
1426         retval = search_item(inode->i_sb, &key, &path_to_sd);
1427         if (retval == IO_ERROR) {
1428                 reiserfs_warning(inode->i_sb, "vs-13070",
1429                                  "i/o failure occurred trying to find "
1430                                  "stat data of %K", &key);
1431                 reiserfs_make_bad_inode(inode);
1432                 return;
1433         }
1434         if (retval != ITEM_FOUND) {
1435                 /* a stale NFS handle can trigger this without it being an error */
1436                 pathrelse(&path_to_sd);
1437                 reiserfs_make_bad_inode(inode);
1438                 inode->i_nlink = 0;
1439                 return;
1440         }
1441
1442         init_inode(inode, &path_to_sd);
1443
1444         /* It is possible that knfsd is trying to access inode of a file
1445            that is being removed from the disk by some other thread. As we
1446            update sd on unlink all that is required is to check for nlink
1447            here. This bug was first found by Sizif when debugging
1448            SquidNG/Butterfly, forgotten, and found again after Philippe
1449            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1450
1451            More logical fix would require changes in fs/inode.c:iput() to
1452            remove inode from hash-table _after_ fs cleaned disk stuff up and
1453            in iget() to return NULL if I_FREEING inode is found in
1454            hash-table. */
1455         /* Currently there is one place where it's ok to meet inode with
1456            nlink==0: processing of open-unlinked and half-truncated files
1457            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1458         if ((inode->i_nlink == 0) &&
1459             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1460                 reiserfs_warning(inode->i_sb, "vs-13075",
1461                                  "dead inode read from disk %K. "
1462                                  "This is likely to be race with knfsd. Ignore",
1463                                  &key);
1464                 reiserfs_make_bad_inode(inode);
1465         }
1466
1467         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1468
1469 }
1470
1471 /**
1472  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1473  *
1474  * @inode:    inode from hash table to check
1475  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1476  *
1477  * This function is called by iget5_locked() to distinguish reiserfs inodes
1478  * having the same inode numbers. Such inodes can only exist due to some
1479  * error condition. One of them should be bad. Inodes with identical
1480  * inode numbers (objectids) are distinguished by parent directory ids.
1481  *
1482  */
1483 int reiserfs_find_actor(struct inode *inode, void *opaque)
1484 {
1485         struct reiserfs_iget_args *args;
1486
1487         args = opaque;
1488         /* args is already in CPU order */
1489         return (inode->i_ino == args->objectid) &&
1490             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1491 }
1492
1493 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1494 {
1495         struct inode *inode;
1496         struct reiserfs_iget_args args;
1497
1498         args.objectid = key->on_disk_key.k_objectid;
1499         args.dirid = key->on_disk_key.k_dir_id;
1500         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1501                              reiserfs_find_actor, reiserfs_init_locked_inode,
1502                              (void *)(&args));
1503         if (!inode)
1504                 return ERR_PTR(-ENOMEM);
1505
1506         if (inode->i_state & I_NEW) {
1507                 reiserfs_read_locked_inode(inode, &args);
1508                 unlock_new_inode(inode);
1509         }
1510
1511         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1512                 /* either due to i/o error or a stale NFS handle */
1513                 iput(inode);
1514                 inode = NULL;
1515         }
1516         return inode;
1517 }
1518
1519 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1520         u32 objectid, u32 dir_id, u32 generation)
1521
1522 {
1523         struct cpu_key key;
1524         struct inode *inode;
1525
1526         key.on_disk_key.k_objectid = objectid;
1527         key.on_disk_key.k_dir_id = dir_id;
1528         reiserfs_write_lock(sb);
1529         inode = reiserfs_iget(sb, &key);
1530         if (inode && !IS_ERR(inode) && generation != 0 &&
1531             generation != inode->i_generation) {
1532                 iput(inode);
1533                 inode = NULL;
1534         }
1535         reiserfs_write_unlock(sb);
1536
1537         return d_obtain_alias(inode);
1538 }
1539
1540 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1541                 int fh_len, int fh_type)
1542 {
1543         /* fhtype happens to reflect the number of u32s encoded.
1544          * due to a bug in earlier code, fhtype might indicate there
1545          * are more u32s then actually fitted.
1546          * so if fhtype seems to be more than len, reduce fhtype.
1547          * Valid types are:
1548          *   2 - objectid + dir_id - legacy support
1549          *   3 - objectid + dir_id + generation
1550          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1551          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1552          *   6 - as above plus generation of directory
1553          * 6 does not fit in NFSv2 handles
1554          */
1555         if (fh_type > fh_len) {
1556                 if (fh_type != 6 || fh_len != 5)
1557                         reiserfs_warning(sb, "reiserfs-13077",
1558                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1559                                 fh_type, fh_len);
1560                 fh_type = 5;
1561         }
1562
1563         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1564                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1565 }
1566
1567 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1568                 int fh_len, int fh_type)
1569 {
1570         if (fh_type < 4)
1571                 return NULL;
1572
1573         return reiserfs_get_dentry(sb,
1574                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1575                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1576                 (fh_type == 6) ? fid->raw[5] : 0);
1577 }
1578
1579 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1580                        int need_parent)
1581 {
1582         struct inode *inode = dentry->d_inode;
1583         int maxlen = *lenp;
1584
1585         if (maxlen < 3)
1586                 return 255;
1587
1588         data[0] = inode->i_ino;
1589         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1590         data[2] = inode->i_generation;
1591         *lenp = 3;
1592         /* no room for directory info? return what we've stored so far */
1593         if (maxlen < 5 || !need_parent)
1594                 return 3;
1595
1596         spin_lock(&dentry->d_lock);
1597         inode = dentry->d_parent->d_inode;
1598         data[3] = inode->i_ino;
1599         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1600         *lenp = 5;
1601         if (maxlen >= 6) {
1602                 data[5] = inode->i_generation;
1603                 *lenp = 6;
1604         }
1605         spin_unlock(&dentry->d_lock);
1606         return *lenp;
1607 }
1608
1609 /* looks for stat data, then copies fields to it, marks the buffer
1610    containing stat data as dirty */
1611 /* reiserfs inodes are never really dirty, since the dirty inode call
1612 ** always logs them.  This call allows the VFS inode marking routines
1613 ** to properly mark inodes for datasync and such, but only actually
1614 ** does something when called for a synchronous update.
1615 */
1616 int reiserfs_write_inode(struct inode *inode, int do_sync)
1617 {
1618         struct reiserfs_transaction_handle th;
1619         int jbegin_count = 1;
1620
1621         if (inode->i_sb->s_flags & MS_RDONLY)
1622                 return -EROFS;
1623         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1624          ** these cases are just when the system needs ram, not when the
1625          ** inode needs to reach disk for safety, and they can safely be
1626          ** ignored because the altered inode has already been logged.
1627          */
1628         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1629                 reiserfs_write_lock(inode->i_sb);
1630                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1631                         reiserfs_update_sd(&th, inode);
1632                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1633                 }
1634                 reiserfs_write_unlock(inode->i_sb);
1635         }
1636         return 0;
1637 }
1638
1639 /* stat data of new object is inserted already, this inserts the item
1640    containing "." and ".." entries */
1641 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1642                                   struct inode *inode,
1643                                   struct item_head *ih, struct treepath *path,
1644                                   struct inode *dir)
1645 {
1646         struct super_block *sb = th->t_super;
1647         char empty_dir[EMPTY_DIR_SIZE];
1648         char *body = empty_dir;
1649         struct cpu_key key;
1650         int retval;
1651
1652         BUG_ON(!th->t_trans_id);
1653
1654         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1655                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1656                       TYPE_DIRENTRY, 3 /*key length */ );
1657
1658         /* compose item head for new item. Directories consist of items of
1659            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1660            is done by reiserfs_new_inode */
1661         if (old_format_only(sb)) {
1662                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1663                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1664
1665                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1666                                        ih->ih_key.k_objectid,
1667                                        INODE_PKEY(dir)->k_dir_id,
1668                                        INODE_PKEY(dir)->k_objectid);
1669         } else {
1670                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1671                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1672
1673                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1674                                     ih->ih_key.k_objectid,
1675                                     INODE_PKEY(dir)->k_dir_id,
1676                                     INODE_PKEY(dir)->k_objectid);
1677         }
1678
1679         /* look for place in the tree for new item */
1680         retval = search_item(sb, &key, path);
1681         if (retval == IO_ERROR) {
1682                 reiserfs_warning(sb, "vs-13080",
1683                                  "i/o failure occurred creating new directory");
1684                 return -EIO;
1685         }
1686         if (retval == ITEM_FOUND) {
1687                 pathrelse(path);
1688                 reiserfs_warning(sb, "vs-13070",
1689                                  "object with this key exists (%k)",
1690                                  &(ih->ih_key));
1691                 return -EEXIST;
1692         }
1693
1694         /* insert item, that is empty directory item */
1695         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1696 }
1697
1698 /* stat data of object has been inserted, this inserts the item
1699    containing the body of symlink */
1700 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1701                                 struct item_head *ih,
1702                                 struct treepath *path, const char *symname,
1703                                 int item_len)
1704 {
1705         struct super_block *sb = th->t_super;
1706         struct cpu_key key;
1707         int retval;
1708
1709         BUG_ON(!th->t_trans_id);
1710
1711         _make_cpu_key(&key, KEY_FORMAT_3_5,
1712                       le32_to_cpu(ih->ih_key.k_dir_id),
1713                       le32_to_cpu(ih->ih_key.k_objectid),
1714                       1, TYPE_DIRECT, 3 /*key length */ );
1715
1716         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1717                           0 /*free_space */ );
1718
1719         /* look for place in the tree for new item */
1720         retval = search_item(sb, &key, path);
1721         if (retval == IO_ERROR) {
1722                 reiserfs_warning(sb, "vs-13080",
1723                                  "i/o failure occurred creating new symlink");
1724                 return -EIO;
1725         }
1726         if (retval == ITEM_FOUND) {
1727                 pathrelse(path);
1728                 reiserfs_warning(sb, "vs-13080",
1729                                  "object with this key exists (%k)",
1730                                  &(ih->ih_key));
1731                 return -EEXIST;
1732         }
1733
1734         /* insert item, that is body of symlink */
1735         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1736 }
1737
1738 /* inserts the stat data into the tree, and then calls
1739    reiserfs_new_directory (to insert ".", ".." item if new object is
1740    directory) or reiserfs_new_symlink (to insert symlink body if new
1741    object is symlink) or nothing (if new object is regular file)
1742
1743    NOTE! uid and gid must already be set in the inode.  If we return
1744    non-zero due to an error, we have to drop the quota previously allocated
1745    for the fresh inode.  This can only be done outside a transaction, so
1746    if we return non-zero, we also end the transaction.  */
1747 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1748                        struct inode *dir, int mode, const char *symname,
1749                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1750                           strlen (symname) for symlinks) */
1751                        loff_t i_size, struct dentry *dentry,
1752                        struct inode *inode)
1753 {
1754         struct super_block *sb;
1755         struct reiserfs_iget_args args;
1756         INITIALIZE_PATH(path_to_key);
1757         struct cpu_key key;
1758         struct item_head ih;
1759         struct stat_data sd;
1760         int retval;
1761         int err;
1762
1763         BUG_ON(!th->t_trans_id);
1764
1765         if (DQUOT_ALLOC_INODE(inode)) {
1766                 err = -EDQUOT;
1767                 goto out_end_trans;
1768         }
1769         if (!dir->i_nlink) {
1770                 err = -EPERM;
1771                 goto out_bad_inode;
1772         }
1773
1774         sb = dir->i_sb;
1775
1776         /* item head of new item */
1777         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1778         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1779         if (!ih.ih_key.k_objectid) {
1780                 err = -ENOMEM;
1781                 goto out_bad_inode;
1782         }
1783         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1784         if (old_format_only(sb))
1785                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1786                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1787         else
1788                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1789                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1790         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1791         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1792         if (insert_inode_locked4(inode, args.objectid,
1793                              reiserfs_find_actor, &args) < 0) {
1794                 err = -EINVAL;
1795                 goto out_bad_inode;
1796         }
1797         if (old_format_only(sb))
1798                 /* not a perfect generation count, as object ids can be reused, but
1799                  ** this is as good as reiserfs can do right now.
1800                  ** note that the private part of inode isn't filled in yet, we have
1801                  ** to use the directory.
1802                  */
1803                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1804         else
1805 #if defined( USE_INODE_GENERATION_COUNTER )
1806                 inode->i_generation =
1807                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1808 #else
1809                 inode->i_generation = ++event;
1810 #endif
1811
1812         /* fill stat data */
1813         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1814
1815         /* uid and gid must already be set by the caller for quota init */
1816
1817         /* symlink cannot be immutable or append only, right? */
1818         if (S_ISLNK(inode->i_mode))
1819                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1820
1821         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1822         inode->i_size = i_size;
1823         inode->i_blocks = 0;
1824         inode->i_bytes = 0;
1825         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1826             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1827
1828         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1829         REISERFS_I(inode)->i_flags = 0;
1830         REISERFS_I(inode)->i_prealloc_block = 0;
1831         REISERFS_I(inode)->i_prealloc_count = 0;
1832         REISERFS_I(inode)->i_trans_id = 0;
1833         REISERFS_I(inode)->i_jl = NULL;
1834         REISERFS_I(inode)->i_attrs =
1835             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1836         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1837         mutex_init(&(REISERFS_I(inode)->i_mmap));
1838         reiserfs_init_acl_access(inode);
1839         reiserfs_init_acl_default(inode);
1840         reiserfs_init_xattr_rwsem(inode);
1841
1842         /* key to search for correct place for new stat data */
1843         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1844                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1845                       TYPE_STAT_DATA, 3 /*key length */ );
1846
1847         /* find proper place for inserting of stat data */
1848         retval = search_item(sb, &key, &path_to_key);
1849         if (retval == IO_ERROR) {
1850                 err = -EIO;
1851                 goto out_bad_inode;
1852         }
1853         if (retval == ITEM_FOUND) {
1854                 pathrelse(&path_to_key);
1855                 err = -EEXIST;
1856                 goto out_bad_inode;
1857         }
1858         if (old_format_only(sb)) {
1859                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1860                         pathrelse(&path_to_key);
1861                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1862                         err = -EINVAL;
1863                         goto out_bad_inode;
1864                 }
1865                 inode2sd_v1(&sd, inode, inode->i_size);
1866         } else {
1867                 inode2sd(&sd, inode, inode->i_size);
1868         }
1869         // store in in-core inode the key of stat data and version all
1870         // object items will have (directory items will have old offset
1871         // format, other new objects will consist of new items)
1872         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1873                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1874         else
1875                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1876         if (old_format_only(sb))
1877                 set_inode_sd_version(inode, STAT_DATA_V1);
1878         else
1879                 set_inode_sd_version(inode, STAT_DATA_V2);
1880
1881         /* insert the stat data into the tree */
1882 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1883         if (REISERFS_I(dir)->new_packing_locality)
1884                 th->displace_new_blocks = 1;
1885 #endif
1886         retval =
1887             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1888                                  (char *)(&sd));
1889         if (retval) {
1890                 err = retval;
1891                 reiserfs_check_path(&path_to_key);
1892                 goto out_bad_inode;
1893         }
1894 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1895         if (!th->displace_new_blocks)
1896                 REISERFS_I(dir)->new_packing_locality = 0;
1897 #endif
1898         if (S_ISDIR(mode)) {
1899                 /* insert item with "." and ".." */
1900                 retval =
1901                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1902         }
1903
1904         if (S_ISLNK(mode)) {
1905                 /* insert body of symlink */
1906                 if (!old_format_only(sb))
1907                         i_size = ROUND_UP(i_size);
1908                 retval =
1909                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1910                                          i_size);
1911         }
1912         if (retval) {
1913                 err = retval;
1914                 reiserfs_check_path(&path_to_key);
1915                 journal_end(th, th->t_super, th->t_blocks_allocated);
1916                 goto out_inserted_sd;
1917         }
1918
1919         /* XXX CHECK THIS */
1920         if (reiserfs_posixacl(inode->i_sb)) {
1921                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1922                 if (retval) {
1923                         err = retval;
1924                         reiserfs_check_path(&path_to_key);
1925                         journal_end(th, th->t_super, th->t_blocks_allocated);
1926                         goto out_inserted_sd;
1927                 }
1928         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1929                 reiserfs_warning(inode->i_sb, "jdm-13090",
1930                                  "ACLs aren't enabled in the fs, "
1931                                  "but vfs thinks they are!");
1932         } else if (is_reiserfs_priv_object(dir)) {
1933                 reiserfs_mark_inode_private(inode);
1934         }
1935
1936         reiserfs_update_sd(th, inode);
1937         reiserfs_check_path(&path_to_key);
1938
1939         return 0;
1940
1941 /* it looks like you can easily compress these two goto targets into
1942  * one.  Keeping it like this doesn't actually hurt anything, and they
1943  * are place holders for what the quota code actually needs.
1944  */
1945       out_bad_inode:
1946         /* Invalidate the object, nothing was inserted yet */
1947         INODE_PKEY(inode)->k_objectid = 0;
1948
1949         /* Quota change must be inside a transaction for journaling */
1950         DQUOT_FREE_INODE(inode);
1951
1952       out_end_trans:
1953         journal_end(th, th->t_super, th->t_blocks_allocated);
1954         /* Drop can be outside and it needs more credits so it's better to have it outside */
1955         DQUOT_DROP(inode);
1956         inode->i_flags |= S_NOQUOTA;
1957         make_bad_inode(inode);
1958
1959       out_inserted_sd:
1960         inode->i_nlink = 0;
1961         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1962         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1963
1964         /* If we were inheriting an ACL, we need to release the lock so that
1965          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1966          * code really needs to be reworked, but this will take care of it
1967          * for now. -jeffm */
1968 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1969         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1970                 reiserfs_write_unlock_xattrs(dir->i_sb);
1971                 iput(inode);
1972                 reiserfs_write_lock_xattrs(dir->i_sb);
1973         } else
1974 #endif
1975                 iput(inode);
1976         return err;
1977 }
1978
1979 /*
1980 ** finds the tail page in the page cache,
1981 ** reads the last block in.
1982 **
1983 ** On success, page_result is set to a locked, pinned page, and bh_result
1984 ** is set to an up to date buffer for the last block in the file.  returns 0.
1985 **
1986 ** tail conversion is not done, so bh_result might not be valid for writing
1987 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1988 ** trying to write the block.
1989 **
1990 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1991 */
1992 static int grab_tail_page(struct inode *p_s_inode,
1993                           struct page **page_result,
1994                           struct buffer_head **bh_result)
1995 {
1996
1997         /* we want the page with the last byte in the file,
1998          ** not the page that will hold the next byte for appending
1999          */
2000         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2001         unsigned long pos = 0;
2002         unsigned long start = 0;
2003         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2004         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2005         struct buffer_head *bh;
2006         struct buffer_head *head;
2007         struct page *page;
2008         int error;
2009
2010         /* we know that we are only called with inode->i_size > 0.
2011          ** we also know that a file tail can never be as big as a block
2012          ** If i_size % blocksize == 0, our file is currently block aligned
2013          ** and it won't need converting or zeroing after a truncate.
2014          */
2015         if ((offset & (blocksize - 1)) == 0) {
2016                 return -ENOENT;
2017         }
2018         page = grab_cache_page(p_s_inode->i_mapping, index);
2019         error = -ENOMEM;
2020         if (!page) {
2021                 goto out;
2022         }
2023         /* start within the page of the last block in the file */
2024         start = (offset / blocksize) * blocksize;
2025
2026         error = block_prepare_write(page, start, offset,
2027                                     reiserfs_get_block_create_0);
2028         if (error)
2029                 goto unlock;
2030
2031         head = page_buffers(page);
2032         bh = head;
2033         do {
2034                 if (pos >= start) {
2035                         break;
2036                 }
2037                 bh = bh->b_this_page;
2038                 pos += blocksize;
2039         } while (bh != head);
2040
2041         if (!buffer_uptodate(bh)) {
2042                 /* note, this should never happen, prepare_write should
2043                  ** be taking care of this for us.  If the buffer isn't up to date,
2044                  ** I've screwed up the code to find the buffer, or the code to
2045                  ** call prepare_write
2046                  */
2047                 reiserfs_warning(p_s_inode->i_sb, "clm-6000",
2048                                  "error reading block %lu on dev %s",
2049                                  bh->b_blocknr,
2050                                  reiserfs_bdevname(p_s_inode->i_sb));
2051                 error = -EIO;
2052                 goto unlock;
2053         }
2054         *bh_result = bh;
2055         *page_result = page;
2056
2057       out:
2058         return error;
2059
2060       unlock:
2061         unlock_page(page);
2062         page_cache_release(page);
2063         return error;
2064 }
2065
2066 /*
2067 ** vfs version of truncate file.  Must NOT be called with
2068 ** a transaction already started.
2069 **
2070 ** some code taken from block_truncate_page
2071 */
2072 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2073 {
2074         struct reiserfs_transaction_handle th;
2075         /* we want the offset for the first byte after the end of the file */
2076         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2077         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2078         unsigned length;
2079         struct page *page = NULL;
2080         int error;
2081         struct buffer_head *bh = NULL;
2082         int err2;
2083
2084         reiserfs_write_lock(p_s_inode->i_sb);
2085
2086         if (p_s_inode->i_size > 0) {
2087                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2088                         // -ENOENT means we truncated past the end of the file,
2089                         // and get_block_create_0 could not find a block to read in,
2090                         // which is ok.
2091                         if (error != -ENOENT)
2092                                 reiserfs_warning(p_s_inode->i_sb, "clm-6001",
2093                                                  "grab_tail_page failed %d",
2094                                                  error);
2095                         page = NULL;
2096                         bh = NULL;
2097                 }
2098         }
2099
2100         /* so, if page != NULL, we have a buffer head for the offset at
2101          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2102          ** then we have an unformatted node.  Otherwise, we have a direct item,
2103          ** and no zeroing is required on disk.  We zero after the truncate,
2104          ** because the truncate might pack the item anyway
2105          ** (it will unmap bh if it packs).
2106          */
2107         /* it is enough to reserve space in transaction for 2 balancings:
2108            one for "save" link adding and another for the first
2109            cut_from_item. 1 is for update_sd */
2110         error = journal_begin(&th, p_s_inode->i_sb,
2111                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2112         if (error)
2113                 goto out;
2114         reiserfs_update_inode_transaction(p_s_inode);
2115         if (update_timestamps)
2116                 /* we are doing real truncate: if the system crashes before the last
2117                    transaction of truncating gets committed - on reboot the file
2118                    either appears truncated properly or not truncated at all */
2119                 add_save_link(&th, p_s_inode, 1);
2120         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2121         error =
2122             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2123         if (error)
2124                 goto out;
2125
2126         /* check reiserfs_do_truncate after ending the transaction */
2127         if (err2) {
2128                 error = err2;
2129                 goto out;
2130         }
2131
2132         if (update_timestamps) {
2133                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2134                 if (error)
2135                         goto out;
2136         }
2137
2138         if (page) {
2139                 length = offset & (blocksize - 1);
2140                 /* if we are not on a block boundary */
2141                 if (length) {
2142                         length = blocksize - length;
2143                         zero_user(page, offset, length);
2144                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2145                                 mark_buffer_dirty(bh);
2146                         }
2147                 }
2148                 unlock_page(page);
2149                 page_cache_release(page);
2150         }
2151
2152         reiserfs_write_unlock(p_s_inode->i_sb);
2153         return 0;
2154       out:
2155         if (page) {
2156                 unlock_page(page);
2157                 page_cache_release(page);
2158         }
2159         reiserfs_write_unlock(p_s_inode->i_sb);
2160         return error;
2161 }
2162
2163 static int map_block_for_writepage(struct inode *inode,
2164                                    struct buffer_head *bh_result,
2165                                    unsigned long block)
2166 {
2167         struct reiserfs_transaction_handle th;
2168         int fs_gen;
2169         struct item_head tmp_ih;
2170         struct item_head *ih;
2171         struct buffer_head *bh;
2172         __le32 *item;
2173         struct cpu_key key;
2174         INITIALIZE_PATH(path);
2175         int pos_in_item;
2176         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2177         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2178         int retval;
2179         int use_get_block = 0;
2180         int bytes_copied = 0;
2181         int copy_size;
2182         int trans_running = 0;
2183
2184         /* catch places below that try to log something without starting a trans */
2185         th.t_trans_id = 0;
2186
2187         if (!buffer_uptodate(bh_result)) {
2188                 return -EIO;
2189         }
2190
2191         kmap(bh_result->b_page);
2192       start_over:
2193         reiserfs_write_lock(inode->i_sb);
2194         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2195
2196       research:
2197         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2198         if (retval != POSITION_FOUND) {
2199                 use_get_block = 1;
2200                 goto out;
2201         }
2202
2203         bh = get_last_bh(&path);
2204         ih = get_ih(&path);
2205         item = get_item(&path);
2206         pos_in_item = path.pos_in_item;
2207
2208         /* we've found an unformatted node */
2209         if (indirect_item_found(retval, ih)) {
2210                 if (bytes_copied > 0) {
2211                         reiserfs_warning(inode->i_sb, "clm-6002",
2212                                          "bytes_copied %d", bytes_copied);
2213                 }
2214                 if (!get_block_num(item, pos_in_item)) {
2215                         /* crap, we are writing to a hole */
2216                         use_get_block = 1;
2217                         goto out;
2218                 }
2219                 set_block_dev_mapped(bh_result,
2220                                      get_block_num(item, pos_in_item), inode);
2221         } else if (is_direct_le_ih(ih)) {
2222                 char *p;
2223                 p = page_address(bh_result->b_page);
2224                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2225                 copy_size = ih_item_len(ih) - pos_in_item;
2226
2227                 fs_gen = get_generation(inode->i_sb);
2228                 copy_item_head(&tmp_ih, ih);
2229
2230                 if (!trans_running) {
2231                         /* vs-3050 is gone, no need to drop the path */
2232                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2233                         if (retval)
2234                                 goto out;
2235                         reiserfs_update_inode_transaction(inode);
2236                         trans_running = 1;
2237                         if (fs_changed(fs_gen, inode->i_sb)
2238                             && item_moved(&tmp_ih, &path)) {
2239                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2240                                                                  bh);
2241                                 goto research;
2242                         }
2243                 }
2244
2245                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2246
2247                 if (fs_changed(fs_gen, inode->i_sb)
2248                     && item_moved(&tmp_ih, &path)) {
2249                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2250                         goto research;
2251                 }
2252
2253                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2254                        copy_size);
2255
2256                 journal_mark_dirty(&th, inode->i_sb, bh);
2257                 bytes_copied += copy_size;
2258                 set_block_dev_mapped(bh_result, 0, inode);
2259
2260                 /* are there still bytes left? */
2261                 if (bytes_copied < bh_result->b_size &&
2262                     (byte_offset + bytes_copied) < inode->i_size) {
2263                         set_cpu_key_k_offset(&key,
2264                                              cpu_key_k_offset(&key) +
2265                                              copy_size);
2266                         goto research;
2267                 }
2268         } else {
2269                 reiserfs_warning(inode->i_sb, "clm-6003",
2270                                  "bad item inode %lu", inode->i_ino);
2271                 retval = -EIO;
2272                 goto out;
2273         }
2274         retval = 0;
2275
2276       out:
2277         pathrelse(&path);
2278         if (trans_running) {
2279                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2280                 if (err)
2281                         retval = err;
2282                 trans_running = 0;
2283         }
2284         reiserfs_write_unlock(inode->i_sb);
2285
2286         /* this is where we fill in holes in the file. */
2287         if (use_get_block) {
2288                 retval = reiserfs_get_block(inode, block, bh_result,
2289                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2290                                             | GET_BLOCK_NO_DANGLE);
2291                 if (!retval) {
2292                         if (!buffer_mapped(bh_result)
2293                             || bh_result->b_blocknr == 0) {
2294                                 /* get_block failed to find a mapped unformatted node. */
2295                                 use_get_block = 0;
2296                                 goto start_over;
2297                         }
2298                 }
2299         }
2300         kunmap(bh_result->b_page);
2301
2302         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2303                 /* we've copied data from the page into the direct item, so the
2304                  * buffer in the page is now clean, mark it to reflect that.
2305                  */
2306                 lock_buffer(bh_result);
2307                 clear_buffer_dirty(bh_result);
2308                 unlock_buffer(bh_result);
2309         }
2310         return retval;
2311 }
2312
2313 /*
2314  * mason@suse.com: updated in 2.5.54 to follow the same general io
2315  * start/recovery path as __block_write_full_page, along with special
2316  * code to handle reiserfs tails.
2317  */
2318 static int reiserfs_write_full_page(struct page *page,
2319                                     struct writeback_control *wbc)
2320 {
2321         struct inode *inode = page->mapping->host;
2322         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2323         int error = 0;
2324         unsigned long block;
2325         sector_t last_block;
2326         struct buffer_head *head, *bh;
2327         int partial = 0;
2328         int nr = 0;
2329         int checked = PageChecked(page);
2330         struct reiserfs_transaction_handle th;
2331         struct super_block *s = inode->i_sb;
2332         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2333         th.t_trans_id = 0;
2334
2335         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2336         if (checked && (current->flags & PF_MEMALLOC)) {
2337                 redirty_page_for_writepage(wbc, page);
2338                 unlock_page(page);
2339                 return 0;
2340         }
2341
2342         /* The page dirty bit is cleared before writepage is called, which
2343          * means we have to tell create_empty_buffers to make dirty buffers
2344          * The page really should be up to date at this point, so tossing
2345          * in the BH_Uptodate is just a sanity check.
2346          */
2347         if (!page_has_buffers(page)) {
2348                 create_empty_buffers(page, s->s_blocksize,
2349                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2350         }
2351         head = page_buffers(page);
2352
2353         /* last page in the file, zero out any contents past the
2354          ** last byte in the file
2355          */
2356         if (page->index >= end_index) {
2357                 unsigned last_offset;
2358
2359                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2360                 /* no file contents in this page */
2361                 if (page->index >= end_index + 1 || !last_offset) {
2362                         unlock_page(page);
2363                         return 0;
2364                 }
2365                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2366         }
2367         bh = head;
2368         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2369         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2370         /* first map all the buffers, logging any direct items we find */
2371         do {
2372                 if (block > last_block) {
2373                         /*
2374                          * This can happen when the block size is less than
2375                          * the page size.  The corresponding bytes in the page
2376                          * were zero filled above
2377                          */
2378                         clear_buffer_dirty(bh);
2379                         set_buffer_uptodate(bh);
2380                 } else if ((checked || buffer_dirty(bh)) &&
2381                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2382                                                        && bh->b_blocknr ==
2383                                                        0))) {
2384                         /* not mapped yet, or it points to a direct item, search
2385                          * the btree for the mapping info, and log any direct
2386                          * items found
2387                          */
2388                         if ((error = map_block_for_writepage(inode, bh, block))) {
2389                                 goto fail;
2390                         }
2391                 }
2392                 bh = bh->b_this_page;
2393                 block++;
2394         } while (bh != head);
2395
2396         /*
2397          * we start the transaction after map_block_for_writepage,
2398          * because it can create holes in the file (an unbounded operation).
2399          * starting it here, we can make a reliable estimate for how many
2400          * blocks we're going to log
2401          */
2402         if (checked) {
2403                 ClearPageChecked(page);
2404                 reiserfs_write_lock(s);
2405                 error = journal_begin(&th, s, bh_per_page + 1);
2406                 if (error) {
2407                         reiserfs_write_unlock(s);
2408                         goto fail;
2409                 }
2410                 reiserfs_update_inode_transaction(inode);
2411         }
2412         /* now go through and lock any dirty buffers on the page */
2413         do {
2414                 get_bh(bh);
2415                 if (!buffer_mapped(bh))
2416                         continue;
2417                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2418                         continue;
2419
2420                 if (checked) {
2421                         reiserfs_prepare_for_journal(s, bh, 1);
2422                         journal_mark_dirty(&th, s, bh);
2423                         continue;
2424                 }
2425                 /* from this point on, we know the buffer is mapped to a
2426                  * real block and not a direct item
2427                  */
2428                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2429                         lock_buffer(bh);
2430                 } else {
2431                         if (!trylock_buffer(bh)) {
2432                                 redirty_page_for_writepage(wbc, page);
2433                                 continue;
2434                         }
2435                 }
2436                 if (test_clear_buffer_dirty(bh)) {
2437                         mark_buffer_async_write(bh);
2438                 } else {
2439                         unlock_buffer(bh);
2440                 }
2441         } while ((bh = bh->b_this_page) != head);
2442
2443         if (checked) {
2444                 error = journal_end(&th, s, bh_per_page + 1);
2445                 reiserfs_write_unlock(s);
2446                 if (error)
2447                         goto fail;
2448         }
2449         BUG_ON(PageWriteback(page));
2450         set_page_writeback(page);
2451         unlock_page(page);
2452
2453         /*
2454          * since any buffer might be the only dirty buffer on the page,
2455          * the first submit_bh can bring the page out of writeback.
2456          * be careful with the buffers.
2457          */
2458         do {
2459                 struct buffer_head *next = bh->b_this_page;
2460                 if (buffer_async_write(bh)) {
2461                         submit_bh(WRITE, bh);
2462                         nr++;
2463                 }
2464                 put_bh(bh);
2465                 bh = next;
2466         } while (bh != head);
2467
2468         error = 0;
2469       done:
2470         if (nr == 0) {
2471                 /*
2472                  * if this page only had a direct item, it is very possible for
2473                  * no io to be required without there being an error.  Or,
2474                  * someone else could have locked them and sent them down the
2475                  * pipe without locking the page
2476                  */
2477                 bh = head;
2478                 do {
2479                         if (!buffer_uptodate(bh)) {
2480                                 partial = 1;
2481                                 break;
2482                         }
2483                         bh = bh->b_this_page;
2484                 } while (bh != head);
2485                 if (!partial)
2486                         SetPageUptodate(page);
2487                 end_page_writeback(page);
2488         }
2489         return error;
2490
2491       fail:
2492         /* catches various errors, we need to make sure any valid dirty blocks
2493          * get to the media.  The page is currently locked and not marked for
2494          * writeback
2495          */
2496         ClearPageUptodate(page);
2497         bh = head;
2498         do {
2499                 get_bh(bh);
2500                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2501                         lock_buffer(bh);
2502                         mark_buffer_async_write(bh);
2503                 } else {
2504                         /*
2505                          * clear any dirty bits that might have come from getting
2506                          * attached to a dirty page
2507                          */
2508                         clear_buffer_dirty(bh);
2509                 }
2510                 bh = bh->b_this_page;
2511         } while (bh != head);
2512         SetPageError(page);
2513         BUG_ON(PageWriteback(page));
2514         set_page_writeback(page);
2515         unlock_page(page);
2516         do {
2517                 struct buffer_head *next = bh->b_this_page;
2518                 if (buffer_async_write(bh)) {
2519                         clear_buffer_dirty(bh);
2520                         submit_bh(WRITE, bh);
2521                         nr++;
2522                 }
2523                 put_bh(bh);
2524                 bh = next;
2525         } while (bh != head);
2526         goto done;
2527 }
2528
2529 static int reiserfs_readpage(struct file *f, struct page *page)
2530 {
2531         return block_read_full_page(page, reiserfs_get_block);
2532 }
2533
2534 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2535 {
2536         struct inode *inode = page->mapping->host;
2537         reiserfs_wait_on_write_block(inode->i_sb);
2538         return reiserfs_write_full_page(page, wbc);
2539 }
2540
2541 static int reiserfs_write_begin(struct file *file,
2542                                 struct address_space *mapping,
2543                                 loff_t pos, unsigned len, unsigned flags,
2544                                 struct page **pagep, void **fsdata)
2545 {
2546         struct inode *inode;
2547         struct page *page;
2548         pgoff_t index;
2549         int ret;
2550         int old_ref = 0;
2551
2552         inode = mapping->host;
2553         *fsdata = 0;
2554         if (flags & AOP_FLAG_CONT_EXPAND &&
2555             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2556                 pos ++;
2557                 *fsdata = (void *)(unsigned long)flags;
2558         }
2559
2560         index = pos >> PAGE_CACHE_SHIFT;
2561         page = grab_cache_page_write_begin(mapping, index, flags);
2562         if (!page)
2563                 return -ENOMEM;
2564         *pagep = page;
2565
2566         reiserfs_wait_on_write_block(inode->i_sb);
2567         fix_tail_page_for_writing(page);
2568         if (reiserfs_transaction_running(inode->i_sb)) {
2569                 struct reiserfs_transaction_handle *th;
2570                 th = (struct reiserfs_transaction_handle *)current->
2571                     journal_info;
2572                 BUG_ON(!th->t_refcount);
2573                 BUG_ON(!th->t_trans_id);
2574                 old_ref = th->t_refcount;
2575                 th->t_refcount++;
2576         }
2577         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2578                                 reiserfs_get_block);
2579         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2580                 struct reiserfs_transaction_handle *th = current->journal_info;
2581                 /* this gets a little ugly.  If reiserfs_get_block returned an
2582                  * error and left a transacstion running, we've got to close it,
2583                  * and we've got to free handle if it was a persistent transaction.
2584                  *
2585                  * But, if we had nested into an existing transaction, we need
2586                  * to just drop the ref count on the handle.
2587                  *
2588                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2589                  * and it was a persistent trans.  Otherwise, it was nested above.
2590                  */
2591                 if (th->t_refcount > old_ref) {
2592                         if (old_ref)
2593                                 th->t_refcount--;
2594                         else {
2595                                 int err;
2596                                 reiserfs_write_lock(inode->i_sb);
2597                                 err = reiserfs_end_persistent_transaction(th);
2598                                 reiserfs_write_unlock(inode->i_sb);
2599                                 if (err)
2600                                         ret = err;
2601                         }
2602                 }
2603         }
2604         if (ret) {
2605                 unlock_page(page);
2606                 page_cache_release(page);
2607         }
2608         return ret;
2609 }
2610
2611 int reiserfs_prepare_write(struct file *f, struct page *page,
2612                            unsigned from, unsigned to)
2613 {
2614         struct inode *inode = page->mapping->host;
2615         int ret;
2616         int old_ref = 0;
2617
2618         reiserfs_wait_on_write_block(inode->i_sb);
2619         fix_tail_page_for_writing(page);
2620         if (reiserfs_transaction_running(inode->i_sb)) {
2621                 struct reiserfs_transaction_handle *th;
2622                 th = (struct reiserfs_transaction_handle *)current->
2623                     journal_info;
2624                 BUG_ON(!th->t_refcount);
2625                 BUG_ON(!th->t_trans_id);
2626                 old_ref = th->t_refcount;
2627                 th->t_refcount++;
2628         }
2629
2630         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2631         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2632                 struct reiserfs_transaction_handle *th = current->journal_info;
2633                 /* this gets a little ugly.  If reiserfs_get_block returned an
2634                  * error and left a transacstion running, we've got to close it,
2635                  * and we've got to free handle if it was a persistent transaction.
2636                  *
2637                  * But, if we had nested into an existing transaction, we need
2638                  * to just drop the ref count on the handle.
2639                  *
2640                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2641                  * and it was a persistent trans.  Otherwise, it was nested above.
2642                  */
2643                 if (th->t_refcount > old_ref) {
2644                         if (old_ref)
2645                                 th->t_refcount--;
2646                         else {
2647                                 int err;
2648                                 reiserfs_write_lock(inode->i_sb);
2649                                 err = reiserfs_end_persistent_transaction(th);
2650                                 reiserfs_write_unlock(inode->i_sb);
2651                                 if (err)
2652                                         ret = err;
2653                         }
2654                 }
2655         }
2656         return ret;
2657
2658 }
2659
2660 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2661 {
2662         return generic_block_bmap(as, block, reiserfs_bmap);
2663 }
2664
2665 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2666                               loff_t pos, unsigned len, unsigned copied,
2667                               struct page *page, void *fsdata)
2668 {
2669         struct inode *inode = page->mapping->host;
2670         int ret = 0;
2671         int update_sd = 0;
2672         struct reiserfs_transaction_handle *th;
2673         unsigned start;
2674
2675         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2676                 pos ++;
2677
2678         reiserfs_wait_on_write_block(inode->i_sb);
2679         if (reiserfs_transaction_running(inode->i_sb))
2680                 th = current->journal_info;
2681         else
2682                 th = NULL;
2683
2684         start = pos & (PAGE_CACHE_SIZE - 1);
2685         if (unlikely(copied < len)) {
2686                 if (!PageUptodate(page))
2687                         copied = 0;
2688
2689                 page_zero_new_buffers(page, start + copied, start + len);
2690         }
2691         flush_dcache_page(page);
2692
2693         reiserfs_commit_page(inode, page, start, start + copied);
2694
2695         /* generic_commit_write does this for us, but does not update the
2696          ** transaction tracking stuff when the size changes.  So, we have
2697          ** to do the i_size updates here.
2698          */
2699         pos += copied;
2700         if (pos > inode->i_size) {
2701                 struct reiserfs_transaction_handle myth;
2702                 reiserfs_write_lock(inode->i_sb);
2703                 /* If the file have grown beyond the border where it
2704                    can have a tail, unmark it as needing a tail
2705                    packing */
2706                 if ((have_large_tails(inode->i_sb)
2707                      && inode->i_size > i_block_size(inode) * 4)
2708                     || (have_small_tails(inode->i_sb)
2709                         && inode->i_size > i_block_size(inode)))
2710                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2711
2712                 ret = journal_begin(&myth, inode->i_sb, 1);
2713                 if (ret) {
2714                         reiserfs_write_unlock(inode->i_sb);
2715                         goto journal_error;
2716                 }
2717                 reiserfs_update_inode_transaction(inode);
2718                 inode->i_size = pos;
2719                 /*
2720                  * this will just nest into our transaction.  It's important
2721                  * to use mark_inode_dirty so the inode gets pushed around on the
2722                  * dirty lists, and so that O_SYNC works as expected
2723                  */
2724                 mark_inode_dirty(inode);
2725                 reiserfs_update_sd(&myth, inode);
2726                 update_sd = 1;
2727                 ret = journal_end(&myth, inode->i_sb, 1);
2728                 reiserfs_write_unlock(inode->i_sb);
2729                 if (ret)
2730                         goto journal_error;
2731         }
2732         if (th) {
2733                 reiserfs_write_lock(inode->i_sb);
2734                 if (!update_sd)
2735                         mark_inode_dirty(inode);
2736                 ret = reiserfs_end_persistent_transaction(th);
2737                 reiserfs_write_unlock(inode->i_sb);
2738                 if (ret)
2739                         goto out;
2740         }
2741
2742       out:
2743         unlock_page(page);
2744         page_cache_release(page);
2745         return ret == 0 ? copied : ret;
2746
2747       journal_error:
2748         if (th) {
2749                 reiserfs_write_lock(inode->i_sb);
2750                 if (!update_sd)
2751                         reiserfs_update_sd(th, inode);
2752                 ret = reiserfs_end_persistent_transaction(th);
2753                 reiserfs_write_unlock(inode->i_sb);
2754         }
2755
2756         goto out;
2757 }
2758
2759 int reiserfs_commit_write(struct file *f, struct page *page,
2760                           unsigned from, unsigned to)
2761 {
2762         struct inode *inode = page->mapping->host;
2763         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2764         int ret = 0;
2765         int update_sd = 0;
2766         struct reiserfs_transaction_handle *th = NULL;
2767
2768         reiserfs_wait_on_write_block(inode->i_sb);
2769         if (reiserfs_transaction_running(inode->i_sb)) {
2770                 th = current->journal_info;
2771         }
2772         reiserfs_commit_page(inode, page, from, to);
2773
2774         /* generic_commit_write does this for us, but does not update the
2775          ** transaction tracking stuff when the size changes.  So, we have
2776          ** to do the i_size updates here.
2777          */
2778         if (pos > inode->i_size) {
2779                 struct reiserfs_transaction_handle myth;
2780                 reiserfs_write_lock(inode->i_sb);
2781                 /* If the file have grown beyond the border where it
2782                    can have a tail, unmark it as needing a tail
2783                    packing */
2784                 if ((have_large_tails(inode->i_sb)
2785                      && inode->i_size > i_block_size(inode) * 4)
2786                     || (have_small_tails(inode->i_sb)
2787                         && inode->i_size > i_block_size(inode)))
2788                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2789
2790                 ret = journal_begin(&myth, inode->i_sb, 1);
2791                 if (ret) {
2792                         reiserfs_write_unlock(inode->i_sb);
2793                         goto journal_error;
2794                 }
2795                 reiserfs_update_inode_transaction(inode);
2796                 inode->i_size = pos;
2797                 /*
2798                  * this will just nest into our transaction.  It's important
2799                  * to use mark_inode_dirty so the inode gets pushed around on the
2800                  * dirty lists, and so that O_SYNC works as expected
2801                  */
2802                 mark_inode_dirty(inode);
2803                 reiserfs_update_sd(&myth, inode);
2804                 update_sd = 1;
2805                 ret = journal_end(&myth, inode->i_sb, 1);
2806                 reiserfs_write_unlock(inode->i_sb);
2807                 if (ret)
2808                         goto journal_error;
2809         }
2810         if (th) {
2811                 reiserfs_write_lock(inode->i_sb);
2812                 if (!update_sd)
2813                         mark_inode_dirty(inode);
2814                 ret = reiserfs_end_persistent_transaction(th);
2815                 reiserfs_write_unlock(inode->i_sb);
2816                 if (ret)
2817                         goto out;
2818         }
2819
2820       out:
2821         return ret;
2822
2823       journal_error:
2824         if (th) {
2825                 reiserfs_write_lock(inode->i_sb);
2826                 if (!update_sd)
2827                         reiserfs_update_sd(th, inode);
2828                 ret = reiserfs_end_persistent_transaction(th);
2829                 reiserfs_write_unlock(inode->i_sb);
2830         }
2831
2832         return ret;
2833 }
2834
2835 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2836 {
2837         if (reiserfs_attrs(inode->i_sb)) {
2838                 if (sd_attrs & REISERFS_SYNC_FL)
2839                         inode->i_flags |= S_SYNC;
2840                 else
2841                         inode->i_flags &= ~S_SYNC;
2842                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2843                         inode->i_flags |= S_IMMUTABLE;
2844                 else
2845                         inode->i_flags &= ~S_IMMUTABLE;
2846                 if (sd_attrs & REISERFS_APPEND_FL)
2847                         inode->i_flags |= S_APPEND;
2848                 else
2849                         inode->i_flags &= ~S_APPEND;
2850                 if (sd_attrs & REISERFS_NOATIME_FL)
2851                         inode->i_flags |= S_NOATIME;
2852                 else
2853                         inode->i_flags &= ~S_NOATIME;
2854                 if (sd_attrs & REISERFS_NOTAIL_FL)
2855                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2856                 else
2857                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2858         }
2859 }
2860
2861 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2862 {
2863         if (reiserfs_attrs(inode->i_sb)) {
2864                 if (inode->i_flags & S_IMMUTABLE)
2865                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2866                 else
2867                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2868                 if (inode->i_flags & S_SYNC)
2869                         *sd_attrs |= REISERFS_SYNC_FL;
2870                 else
2871                         *sd_attrs &= ~REISERFS_SYNC_FL;
2872                 if (inode->i_flags & S_NOATIME)
2873                         *sd_attrs |= REISERFS_NOATIME_FL;
2874                 else
2875                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2876                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2877                         *sd_attrs |= REISERFS_NOTAIL_FL;
2878                 else
2879                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2880         }
2881 }
2882
2883 /* decide if this buffer needs to stay around for data logging or ordered
2884 ** write purposes
2885 */
2886 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2887 {
2888         int ret = 1;
2889         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2890
2891         lock_buffer(bh);
2892         spin_lock(&j->j_dirty_buffers_lock);
2893         if (!buffer_mapped(bh)) {
2894                 goto free_jh;
2895         }
2896         /* the page is locked, and the only places that log a data buffer
2897          * also lock the page.
2898          */
2899         if (reiserfs_file_data_log(inode)) {
2900                 /*
2901                  * very conservative, leave the buffer pinned if
2902                  * anyone might need it.
2903                  */
2904                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2905                         ret = 0;
2906                 }
2907         } else  if (buffer_dirty(bh)) {
2908                 struct reiserfs_journal_list *jl;
2909                 struct reiserfs_jh *jh = bh->b_private;
2910
2911                 /* why is this safe?
2912                  * reiserfs_setattr updates i_size in the on disk
2913                  * stat data before allowing vmtruncate to be called.
2914                  *
2915                  * If buffer was put onto the ordered list for this
2916                  * transaction, we know for sure either this transaction
2917                  * or an older one already has updated i_size on disk,
2918                  * and this ordered data won't be referenced in the file
2919                  * if we crash.
2920                  *
2921                  * if the buffer was put onto the ordered list for an older
2922                  * transaction, we need to leave it around
2923                  */
2924                 if (jh && (jl = jh->jl)
2925                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2926                         ret = 0;
2927         }
2928       free_jh:
2929         if (ret && bh->b_private) {
2930                 reiserfs_free_jh(bh);
2931         }
2932         spin_unlock(&j->j_dirty_buffers_lock);
2933         unlock_buffer(bh);
2934         return ret;
2935 }
2936
2937 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2938 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2939 {
2940         struct buffer_head *head, *bh, *next;
2941         struct inode *inode = page->mapping->host;
2942         unsigned int curr_off = 0;
2943         int ret = 1;
2944
2945         BUG_ON(!PageLocked(page));
2946
2947         if (offset == 0)
2948                 ClearPageChecked(page);
2949
2950         if (!page_has_buffers(page))
2951                 goto out;
2952
2953         head = page_buffers(page);
2954         bh = head;
2955         do {
2956                 unsigned int next_off = curr_off + bh->b_size;
2957                 next = bh->b_this_page;
2958
2959                 /*
2960                  * is this block fully invalidated?
2961                  */
2962                 if (offset <= curr_off) {
2963                         if (invalidatepage_can_drop(inode, bh))
2964                                 reiserfs_unmap_buffer(bh);
2965                         else
2966                                 ret = 0;
2967                 }
2968                 curr_off = next_off;
2969                 bh = next;
2970         } while (bh != head);
2971
2972         /*
2973          * We release buffers only if the entire page is being invalidated.
2974          * The get_block cached value has been unconditionally invalidated,
2975          * so real IO is not possible anymore.
2976          */
2977         if (!offset && ret) {
2978                 ret = try_to_release_page(page, 0);
2979                 /* maybe should BUG_ON(!ret); - neilb */
2980         }
2981       out:
2982         return;
2983 }
2984
2985 static int reiserfs_set_page_dirty(struct page *page)
2986 {
2987         struct inode *inode = page->mapping->host;
2988         if (reiserfs_file_data_log(inode)) {
2989                 SetPageChecked(page);
2990                 return __set_page_dirty_nobuffers(page);
2991         }
2992         return __set_page_dirty_buffers(page);
2993 }
2994
2995 /*
2996  * Returns 1 if the page's buffers were dropped.  The page is locked.
2997  *
2998  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2999  * in the buffers at page_buffers(page).
3000  *
3001  * even in -o notail mode, we can't be sure an old mount without -o notail
3002  * didn't create files with tails.
3003  */
3004 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3005 {
3006         struct inode *inode = page->mapping->host;
3007         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3008         struct buffer_head *head;
3009         struct buffer_head *bh;
3010         int ret = 1;
3011
3012         WARN_ON(PageChecked(page));
3013         spin_lock(&j->j_dirty_buffers_lock);
3014         head = page_buffers(page);
3015         bh = head;
3016         do {
3017                 if (bh->b_private) {
3018                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3019                                 reiserfs_free_jh(bh);
3020                         } else {
3021                                 ret = 0;
3022                                 break;
3023                         }
3024                 }
3025                 bh = bh->b_this_page;
3026         } while (bh != head);
3027         if (ret)
3028                 ret = try_to_free_buffers(page);
3029         spin_unlock(&j->j_dirty_buffers_lock);
3030         return ret;
3031 }
3032
3033 /* We thank Mingming Cao for helping us understand in great detail what
3034    to do in this section of the code. */
3035 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3036                                   const struct iovec *iov, loff_t offset,
3037                                   unsigned long nr_segs)
3038 {
3039         struct file *file = iocb->ki_filp;
3040         struct inode *inode = file->f_mapping->host;
3041
3042         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3043                                   offset, nr_segs,
3044                                   reiserfs_get_blocks_direct_io, NULL);
3045 }
3046
3047 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3048 {
3049         struct inode *inode = dentry->d_inode;
3050         int error;
3051         unsigned int ia_valid;
3052
3053         /* must be turned off for recursive notify_change calls */
3054         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3055
3056         reiserfs_write_lock(inode->i_sb);
3057         if (attr->ia_valid & ATTR_SIZE) {
3058                 /* version 2 items will be caught by the s_maxbytes check
3059                  ** done for us in vmtruncate
3060                  */
3061                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3062                     attr->ia_size > MAX_NON_LFS) {
3063                         error = -EFBIG;
3064                         goto out;
3065                 }
3066                 /* fill in hole pointers in the expanding truncate case. */
3067                 if (attr->ia_size > inode->i_size) {
3068                         error = generic_cont_expand_simple(inode, attr->ia_size);
3069                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3070                                 int err;
3071                                 struct reiserfs_transaction_handle th;
3072                                 /* we're changing at most 2 bitmaps, inode + super */
3073                                 err = journal_begin(&th, inode->i_sb, 4);
3074                                 if (!err) {
3075                                         reiserfs_discard_prealloc(&th, inode);
3076                                         err = journal_end(&th, inode->i_sb, 4);
3077                                 }
3078                                 if (err)
3079                                         error = err;
3080                         }
3081                         if (error)
3082                                 goto out;
3083                         /*
3084                          * file size is changed, ctime and mtime are
3085                          * to be updated
3086                          */
3087                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3088                 }
3089         }
3090
3091         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3092              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3093             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3094                 /* stat data of format v3.5 has 16 bit uid and gid */
3095                 error = -EINVAL;
3096                 goto out;
3097         }
3098
3099         error = inode_change_ok(inode, attr);
3100         if (!error) {
3101                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3102                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3103                         error = reiserfs_chown_xattrs(inode, attr);
3104
3105                         if (!error) {
3106                                 struct reiserfs_transaction_handle th;
3107                                 int jbegin_count =
3108                                     2 *
3109                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3110                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3111                                     2;
3112
3113                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3114                                 error =
3115                                     journal_begin(&th, inode->i_sb,
3116                                                   jbegin_count);
3117                                 if (error)
3118                                         goto out;
3119                                 error =
3120                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3121                                 if (error) {
3122                                         journal_end(&th, inode->i_sb,
3123                                                     jbegin_count);
3124                                         goto out;
3125                                 }
3126                                 /* Update corresponding info in inode so that everything is in
3127                                  * one transaction */
3128                                 if (attr->ia_valid & ATTR_UID)
3129                                         inode->i_uid = attr->ia_uid;
3130                                 if (attr->ia_valid & ATTR_GID)
3131                                         inode->i_gid = attr->ia_gid;
3132                                 mark_inode_dirty(inode);
3133                                 error =
3134                                     journal_end(&th, inode->i_sb, jbegin_count);
3135                         }
3136                 }
3137                 if (!error)
3138                         error = inode_setattr(inode, attr);
3139         }
3140
3141         if (!error && reiserfs_posixacl(inode->i_sb)) {
3142                 if (attr->ia_valid & ATTR_MODE)
3143                         error = reiserfs_acl_chmod(inode);
3144         }
3145
3146       out:
3147         reiserfs_write_unlock(inode->i_sb);
3148         return error;
3149 }
3150
3151 const struct address_space_operations reiserfs_address_space_operations = {
3152         .writepage = reiserfs_writepage,
3153         .readpage = reiserfs_readpage,
3154         .readpages = reiserfs_readpages,
3155         .releasepage = reiserfs_releasepage,
3156         .invalidatepage = reiserfs_invalidatepage,
3157         .sync_page = block_sync_page,
3158         .write_begin = reiserfs_write_begin,
3159         .write_end = reiserfs_write_end,
3160         .bmap = reiserfs_aop_bmap,
3161         .direct_IO = reiserfs_direct_IO,
3162         .set_page_dirty = reiserfs_set_page_dirty,
3163 };