git.oblomov.eu Git - linux-2.6/blob - fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <asm/uaccess.h>
  14 #include <asm/unaligned.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/mpage.h>
  17 #include <linux/writeback.h>
  18 #include <linux/quotaops.h>
  19
  20 extern int reiserfs_default_io_size;    /* default io size devuned in super.c */
  21
  22 static int reiserfs_commit_write(struct file *f, struct page *page,
  23                                  unsigned from, unsigned to);
  24 static int reiserfs_prepare_write(struct file *f, struct page *page,
  25                                   unsigned from, unsigned to);
  26
  27 void reiserfs_delete_inode(struct inode *inode)
  28 {
  29         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  30         int jbegin_count =
  31             JOURNAL_PER_BALANCE_CNT * 2 +
  32             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  33         struct reiserfs_transaction_handle th;
  34         int err;
  35
  36         truncate_inode_pages(&inode->i_data, 0);
  37
  38         reiserfs_write_lock(inode->i_sb);
  39
  40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  42                 reiserfs_delete_xattrs(inode);
  43
  44                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  45                         goto out;
  46                 reiserfs_update_inode_transaction(inode);
  47
  48                 err = reiserfs_delete_object(&th, inode);
  49
  50                 /* Do quota update inside a transaction for journaled quotas. We must do that
  51                  * after delete_object so that quota updates go into the same transaction as
  52                  * stat data deletion */
  53                 if (!err)
  54                         DQUOT_FREE_INODE(inode);
  55
  56                 if (journal_end(&th, inode->i_sb, jbegin_count))
  57                         goto out;
  58
  59                 /* check return value from reiserfs_delete_object after
  60                  * ending the transaction
  61                  */
  62                 if (err)
  63                     goto out;
  64
  65                 /* all items of file are deleted, so we can remove "save" link */
  66                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  67                                                                  * about an error here */
  68         } else {
  69                 /* no object items are in the tree */
  70                 ;
  71         }
  72       out:
  73         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  74         inode->i_blocks = 0;
  75         reiserfs_write_unlock(inode->i_sb);
  76 }
  77
  78 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  79                           __u32 objectid, loff_t offset, int type, int length)
  80 {
  81         key->version = version;
  82
  83         key->on_disk_key.k_dir_id = dirid;
  84         key->on_disk_key.k_objectid = objectid;
  85         set_cpu_key_k_offset(key, offset);
  86         set_cpu_key_k_type(key, type);
  87         key->key_length = length;
  88 }
  89
  90 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  91    offset and type of key */
  92 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  93                   int type, int length)
  94 {
  95         _make_cpu_key(key, get_inode_item_key_version(inode),
  96                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
  97                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
  98                       length);
  99 }
 100
 101 //
 102 // when key is 0, do not set version and short key
 103 //
 104 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 105                               int version,
 106                               loff_t offset, int type, int length,
 107                               int entry_count /*or ih_free_space */ )
 108 {
 109         if (key) {
 110                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 111                 ih->ih_key.k_objectid =
 112                     cpu_to_le32(key->on_disk_key.k_objectid);
 113         }
 114         put_ih_version(ih, version);
 115         set_le_ih_k_offset(ih, offset);
 116         set_le_ih_k_type(ih, type);
 117         put_ih_item_len(ih, length);
 118         /*    set_ih_free_space (ih, 0); */
 119         // for directory items it is entry count, for directs and stat
 120         // datas - 0xffff, for indirects - 0
 121         put_ih_entry_count(ih, entry_count);
 122 }
 123
 124 //
 125 // FIXME: we might cache recently accessed indirect item
 126
 127 // Ugh.  Not too eager for that....
 128 //  I cut the code until such time as I see a convincing argument (benchmark).
 129 // I don't want a bloated inode struct..., and I don't like code complexity....
 130
 131 /* cutting the code is fine, since it really isn't in use yet and is easy
 132 ** to add back in.  But, Vladimir has a really good idea here.  Think
 133 ** about what happens for reading a file.  For each page,
 134 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 135 ** an indirect item.  This indirect item has X number of pointers, where
 136 ** X is a big number if we've done the block allocation right.  But,
 137 ** we only use one or two of these pointers during each call to readpage,
 138 ** needlessly researching again later on.
 139 **
 140 ** The size of the cache could be dynamic based on the size of the file.
 141 **
 142 ** I'd also like to see us cache the location the stat data item, since
 143 ** we are needlessly researching for that frequently.
 144 **
 145 ** --chris
 146 */
 147
 148 /* If this page has a file tail in it, and
 149 ** it was read in by get_block_create_0, the page data is valid,
 150 ** but tail is still sitting in a direct item, and we can't write to
 151 ** it.  So, look through this page, and check all the mapped buffers
 152 ** to make sure they have valid block numbers.  Any that don't need
 153 ** to be unmapped, so that block_prepare_write will correctly call
 154 ** reiserfs_get_block to convert the tail into an unformatted node
 155 */
 156 static inline void fix_tail_page_for_writing(struct page *page)
 157 {
 158         struct buffer_head *head, *next, *bh;
 159
 160         if (page && page_has_buffers(page)) {
 161                 head = page_buffers(page);
 162                 bh = head;
 163                 do {
 164                         next = bh->b_this_page;
 165                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 166                                 reiserfs_unmap_buffer(bh);
 167                         }
 168                         bh = next;
 169                 } while (bh != head);
 170         }
 171 }
 172
 173 /* reiserfs_get_block does not need to allocate a block only if it has been
 174    done already or non-hole position has been found in the indirect item */
 175 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 176                                     struct item_head *ih,
 177                                     __le32 * item, int pos_in_item)
 178 {
 179         if (allocated)
 180                 return 0;
 181         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 182             get_block_num(item, pos_in_item))
 183                 return 0;
 184         return 1;
 185 }
 186
 187 static inline int indirect_item_found(int retval, struct item_head *ih)
 188 {
 189         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 190 }
 191
 192 static inline void set_block_dev_mapped(struct buffer_head *bh,
 193                                         b_blocknr_t block, struct inode *inode)
 194 {
 195         map_bh(bh, inode->i_sb, block);
 196 }
 197
 198 //
 199 // files which were created in the earlier version can not be longer,
 200 // than 2 gb
 201 //
 202 static int file_capable(struct inode *inode, long block)
 203 {
 204         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 205             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 206                 return 1;
 207
 208         return 0;
 209 }
 210
 211 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
 212                                    struct inode *inode, struct path *path)
 213 {
 214         struct super_block *s = th->t_super;
 215         int len = th->t_blocks_allocated;
 216         int err;
 217
 218         BUG_ON(!th->t_trans_id);
 219         BUG_ON(!th->t_refcount);
 220
 221         /* we cannot restart while nested */
 222         if (th->t_refcount > 1) {
 223                 return 0;
 224         }
 225         pathrelse(path);
 226         reiserfs_update_sd(th, inode);
 227         err = journal_end(th, s, len);
 228         if (!err) {
 229                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 230                 if (!err)
 231                         reiserfs_update_inode_transaction(inode);
 232         }
 233         return err;
 234 }
 235
 236 // it is called by get_block when create == 0. Returns block number
 237 // for 'block'-th logical block of file. When it hits direct item it
 238 // returns 0 (being called from bmap) or read direct item into piece
 239 // of page (bh_result)
 240
 241 // Please improve the english/clarity in the comment above, as it is
 242 // hard to understand.
 243
 244 static int _get_block_create_0(struct inode *inode, long block,
 245                                struct buffer_head *bh_result, int args)
 246 {
 247         INITIALIZE_PATH(path);
 248         struct cpu_key key;
 249         struct buffer_head *bh;
 250         struct item_head *ih, tmp_ih;
 251         int fs_gen;
 252         int blocknr;
 253         char *p = NULL;
 254         int chars;
 255         int ret;
 256         int result;
 257         int done = 0;
 258         unsigned long offset;
 259
 260         // prepare the key to look for the 'block'-th block of file
 261         make_cpu_key(&key, inode,
 262                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 263                      3);
 264
 265       research:
 266         result = search_for_position_by_key(inode->i_sb, &key, &path);
 267         if (result != POSITION_FOUND) {
 268                 pathrelse(&path);
 269                 if (p)
 270                         kunmap(bh_result->b_page);
 271                 if (result == IO_ERROR)
 272                         return -EIO;
 273                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 274                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 275                 if ((args & GET_BLOCK_NO_HOLE)
 276                     && !PageUptodate(bh_result->b_page)) {
 277                         return -ENOENT;
 278                 }
 279                 return 0;
 280         }
 281         //
 282         bh = get_last_bh(&path);
 283         ih = get_ih(&path);
 284         if (is_indirect_le_ih(ih)) {
 285                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 286
 287                 /* FIXME: here we could cache indirect item or part of it in
 288                    the inode to avoid search_by_key in case of subsequent
 289                    access to file */
 290                 blocknr = get_block_num(ind_item, path.pos_in_item);
 291                 ret = 0;
 292                 if (blocknr) {
 293                         map_bh(bh_result, inode->i_sb, blocknr);
 294                         if (path.pos_in_item ==
 295                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 296                                 set_buffer_boundary(bh_result);
 297                         }
 298                 } else
 299                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 300                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 301                 if ((args & GET_BLOCK_NO_HOLE)
 302                             && !PageUptodate(bh_result->b_page)) {
 303                         ret = -ENOENT;
 304                 }
 305
 306                 pathrelse(&path);
 307                 if (p)
 308                         kunmap(bh_result->b_page);
 309                 return ret;
 310         }
 311         // requested data are in direct item(s)
 312         if (!(args & GET_BLOCK_READ_DIRECT)) {
 313                 // we are called by bmap. FIXME: we can not map block of file
 314                 // when it is stored in direct item(s)
 315                 pathrelse(&path);
 316                 if (p)
 317                         kunmap(bh_result->b_page);
 318                 return -ENOENT;
 319         }
 320
 321         /* if we've got a direct item, and the buffer or page was uptodate,
 322          ** we don't want to pull data off disk again.  skip to the
 323          ** end, where we map the buffer and return
 324          */
 325         if (buffer_uptodate(bh_result)) {
 326                 goto finished;
 327         } else
 328                 /*
 329                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 330                  ** pages without any buffers.  If the page is up to date, we don't want
 331                  ** read old data off disk.  Set the up to date bit on the buffer instead
 332                  ** and jump to the end
 333                  */
 334         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 335                 set_buffer_uptodate(bh_result);
 336                 goto finished;
 337         }
 338         // read file tail into part of page
 339         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 340         fs_gen = get_generation(inode->i_sb);
 341         copy_item_head(&tmp_ih, ih);
 342
 343         /* we only want to kmap if we are reading the tail into the page.
 344          ** this is not the common case, so we don't kmap until we are
 345          ** sure we need to.  But, this means the item might move if
 346          ** kmap schedules
 347          */
 348         if (!p) {
 349                 p = (char *)kmap(bh_result->b_page);
 350                 if (fs_changed(fs_gen, inode->i_sb)
 351                     && item_moved(&tmp_ih, &path)) {
 352                         goto research;
 353                 }
 354         }
 355         p += offset;
 356         memset(p, 0, inode->i_sb->s_blocksize);
 357         do {
 358                 if (!is_direct_le_ih(ih)) {
 359                         BUG();
 360                 }
 361                 /* make sure we don't read more bytes than actually exist in
 362                  ** the file.  This can happen in odd cases where i_size isn't
 363                  ** correct, and when direct item padding results in a few
 364                  ** extra bytes at the end of the direct item
 365                  */
 366                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 367                         break;
 368                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 369                         chars =
 370                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 371                             path.pos_in_item;
 372                         done = 1;
 373                 } else {
 374                         chars = ih_item_len(ih) - path.pos_in_item;
 375                 }
 376                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 377
 378                 if (done)
 379                         break;
 380
 381                 p += chars;
 382
 383                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 384                         // we done, if read direct item is not the last item of
 385                         // node FIXME: we could try to check right delimiting key
 386                         // to see whether direct item continues in the right
 387                         // neighbor or rely on i_size
 388                         break;
 389
 390                 // update key to look for the next piece
 391                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 392                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 393                 if (result != POSITION_FOUND)
 394                         // i/o error most likely
 395                         break;
 396                 bh = get_last_bh(&path);
 397                 ih = get_ih(&path);
 398         } while (1);
 399
 400         flush_dcache_page(bh_result->b_page);
 401         kunmap(bh_result->b_page);
 402
 403       finished:
 404         pathrelse(&path);
 405
 406         if (result == IO_ERROR)
 407                 return -EIO;
 408
 409         /* this buffer has valid data, but isn't valid for io.  mapping it to
 410          * block #0 tells the rest of reiserfs it just has a tail in it
 411          */
 412         map_bh(bh_result, inode->i_sb, 0);
 413         set_buffer_uptodate(bh_result);
 414         return 0;
 415 }
 416
 417 // this is called to create file map. So, _get_block_create_0 will not
 418 // read direct item
 419 static int reiserfs_bmap(struct inode *inode, sector_t block,
 420                          struct buffer_head *bh_result, int create)
 421 {
 422         if (!file_capable(inode, block))
 423                 return -EFBIG;
 424
 425         reiserfs_write_lock(inode->i_sb);
 426         /* do not read the direct item */
 427         _get_block_create_0(inode, block, bh_result, 0);
 428         reiserfs_write_unlock(inode->i_sb);
 429         return 0;
 430 }
 431
 432 /* special version of get_block that is only used by grab_tail_page right
 433 ** now.  It is sent to block_prepare_write, and when you try to get a
 434 ** block past the end of the file (or a block from a hole) it returns
 435 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 436 ** be able to do i/o on the buffers returned, unless an error value
 437 ** is also returned.
 438 **
 439 ** So, this allows block_prepare_write to be used for reading a single block
 440 ** in a page.  Where it does not produce a valid page for holes, or past the
 441 ** end of the file.  This turns out to be exactly what we need for reading
 442 ** tails for conversion.
 443 **
 444 ** The point of the wrapper is forcing a certain value for create, even
 445 ** though the VFS layer is calling this function with create==1.  If you
 446 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 447 ** don't use this function.
 448 */
 449 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 450                                        struct buffer_head *bh_result,
 451                                        int create)
 452 {
 453         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 454 }
 455
 456 /* This is special helper for reiserfs_get_block in case we are executing
 457    direct_IO request. */
 458 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 459                                          sector_t iblock,
 460                                          struct buffer_head *bh_result,
 461                                          int create)
 462 {
 463         int ret;
 464
 465         bh_result->b_page = NULL;
 466
 467         /* We set the b_size before reiserfs_get_block call since it is
 468            referenced in convert_tail_for_hole() that may be called from
 469            reiserfs_get_block() */
 470         bh_result->b_size = (1 << inode->i_blkbits);
 471
 472         ret = reiserfs_get_block(inode, iblock, bh_result,
 473                                  create | GET_BLOCK_NO_DANGLE);
 474         if (ret)
 475                 goto out;
 476
 477         /* don't allow direct io onto tail pages */
 478         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 479                 /* make sure future calls to the direct io funcs for this offset
 480                  ** in the file fail by unmapping the buffer
 481                  */
 482                 clear_buffer_mapped(bh_result);
 483                 ret = -EINVAL;
 484         }
 485         /* Possible unpacked tail. Flush the data before pages have
 486            disappeared */
 487         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 488                 int err;
 489                 lock_kernel();
 490                 err = reiserfs_commit_for_inode(inode);
 491                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 492                 unlock_kernel();
 493                 if (err < 0)
 494                         ret = err;
 495         }
 496       out:
 497         return ret;
 498 }
 499
 500 /*
 501 ** helper function for when reiserfs_get_block is called for a hole
 502 ** but the file tail is still in a direct item
 503 ** bh_result is the buffer head for the hole
 504 ** tail_offset is the offset of the start of the tail in the file
 505 **
 506 ** This calls prepare_write, which will start a new transaction
 507 ** you should not be in a transaction, or have any paths held when you
 508 ** call this.
 509 */
 510 static int convert_tail_for_hole(struct inode *inode,
 511                                  struct buffer_head *bh_result,
 512                                  loff_t tail_offset)
 513 {
 514         unsigned long index;
 515         unsigned long tail_end;
 516         unsigned long tail_start;
 517         struct page *tail_page;
 518         struct page *hole_page = bh_result->b_page;
 519         int retval = 0;
 520
 521         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 522                 return -EIO;
 523
 524         /* always try to read until the end of the block */
 525         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 526         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 527
 528         index = tail_offset >> PAGE_CACHE_SHIFT;
 529         /* hole_page can be zero in case of direct_io, we are sure
 530            that we cannot get here if we write with O_DIRECT into
 531            tail page */
 532         if (!hole_page || index != hole_page->index) {
 533                 tail_page = grab_cache_page(inode->i_mapping, index);
 534                 retval = -ENOMEM;
 535                 if (!tail_page) {
 536                         goto out;
 537                 }
 538         } else {
 539                 tail_page = hole_page;
 540         }
 541
 542         /* we don't have to make sure the conversion did not happen while
 543          ** we were locking the page because anyone that could convert
 544          ** must first take i_mutex.
 545          **
 546          ** We must fix the tail page for writing because it might have buffers
 547          ** that are mapped, but have a block number of 0.  This indicates tail
 548          ** data that has been read directly into the page, and block_prepare_write
 549          ** won't trigger a get_block in this case.
 550          */
 551         fix_tail_page_for_writing(tail_page);
 552         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 553         if (retval)
 554                 goto unlock;
 555
 556         /* tail conversion might change the data in the page */
 557         flush_dcache_page(tail_page);
 558
 559         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 560
 561       unlock:
 562         if (tail_page != hole_page) {
 563                 unlock_page(tail_page);
 564                 page_cache_release(tail_page);
 565         }
 566       out:
 567         return retval;
 568 }
 569
 570 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 571                                   long block,
 572                                   struct inode *inode,
 573                                   b_blocknr_t * allocated_block_nr,
 574                                   struct path *path, int flags)
 575 {
 576         BUG_ON(!th->t_trans_id);
 577
 578 #ifdef REISERFS_PREALLOCATE
 579         if (!(flags & GET_BLOCK_NO_IMUX)) {
 580                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 581                                                   path, block);
 582         }
 583 #endif
 584         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 585                                          block);
 586 }
 587
 588 int reiserfs_get_block(struct inode *inode, sector_t block,
 589                        struct buffer_head *bh_result, int create)
 590 {
 591         int repeat, retval = 0;
 592         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 593         INITIALIZE_PATH(path);
 594         int pos_in_item;
 595         struct cpu_key key;
 596         struct buffer_head *bh, *unbh = NULL;
 597         struct item_head *ih, tmp_ih;
 598         __le32 *item;
 599         int done;
 600         int fs_gen;
 601         struct reiserfs_transaction_handle *th = NULL;
 602         /* space reserved in transaction batch:
 603            . 3 balancings in direct->indirect conversion
 604            . 1 block involved into reiserfs_update_sd()
 605            XXX in practically impossible worst case direct2indirect()
 606            can incur (much) more than 3 balancings.
 607            quota update for user, group */
 608         int jbegin_count =
 609             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 610             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 611         int version;
 612         int dangle = 1;
 613         loff_t new_offset =
 614             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 615
 616         /* bad.... */
 617         reiserfs_write_lock(inode->i_sb);
 618         version = get_inode_item_key_version(inode);
 619
 620         if (!file_capable(inode, block)) {
 621                 reiserfs_write_unlock(inode->i_sb);
 622                 return -EFBIG;
 623         }
 624
 625         /* if !create, we aren't changing the FS, so we don't need to
 626          ** log anything, so we don't need to start a transaction
 627          */
 628         if (!(create & GET_BLOCK_CREATE)) {
 629                 int ret;
 630                 /* find number of block-th logical block of the file */
 631                 ret = _get_block_create_0(inode, block, bh_result,
 632                                           create | GET_BLOCK_READ_DIRECT);
 633                 reiserfs_write_unlock(inode->i_sb);
 634                 return ret;
 635         }
 636         /*
 637          * if we're already in a transaction, make sure to close
 638          * any new transactions we start in this func
 639          */
 640         if ((create & GET_BLOCK_NO_DANGLE) ||
 641             reiserfs_transaction_running(inode->i_sb))
 642                 dangle = 0;
 643
 644         /* If file is of such a size, that it might have a tail and tails are enabled
 645          ** we should mark it as possibly needing tail packing on close
 646          */
 647         if ((have_large_tails(inode->i_sb)
 648              && inode->i_size < i_block_size(inode) * 4)
 649             || (have_small_tails(inode->i_sb)
 650                 && inode->i_size < i_block_size(inode)))
 651                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 652
 653         /* set the key of the first byte in the 'block'-th block of file */
 654         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 655         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 656               start_trans:
 657                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 658                 if (!th) {
 659                         retval = -ENOMEM;
 660                         goto failure;
 661                 }
 662                 reiserfs_update_inode_transaction(inode);
 663         }
 664       research:
 665
 666         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 667         if (retval == IO_ERROR) {
 668                 retval = -EIO;
 669                 goto failure;
 670         }
 671
 672         bh = get_last_bh(&path);
 673         ih = get_ih(&path);
 674         item = get_item(&path);
 675         pos_in_item = path.pos_in_item;
 676
 677         fs_gen = get_generation(inode->i_sb);
 678         copy_item_head(&tmp_ih, ih);
 679
 680         if (allocation_needed
 681             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 682                 /* we have to allocate block for the unformatted node */
 683                 if (!th) {
 684                         pathrelse(&path);
 685                         goto start_trans;
 686                 }
 687
 688                 repeat =
 689                     _allocate_block(th, block, inode, &allocated_block_nr,
 690                                     &path, create);
 691
 692                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 693                         /* restart the transaction to give the journal a chance to free
 694                          ** some blocks.  releases the path, so we have to go back to
 695                          ** research if we succeed on the second try
 696                          */
 697                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 698                         retval = restart_transaction(th, inode, &path);
 699                         if (retval)
 700                                 goto failure;
 701                         repeat =
 702                             _allocate_block(th, block, inode,
 703                                             &allocated_block_nr, NULL, create);
 704
 705                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 706                                 goto research;
 707                         }
 708                         if (repeat == QUOTA_EXCEEDED)
 709                                 retval = -EDQUOT;
 710                         else
 711                                 retval = -ENOSPC;
 712                         goto failure;
 713                 }
 714
 715                 if (fs_changed(fs_gen, inode->i_sb)
 716                     && item_moved(&tmp_ih, &path)) {
 717                         goto research;
 718                 }
 719         }
 720
 721         if (indirect_item_found(retval, ih)) {
 722                 b_blocknr_t unfm_ptr;
 723                 /* 'block'-th block is in the file already (there is
 724                    corresponding cell in some indirect item). But it may be
 725                    zero unformatted node pointer (hole) */
 726                 unfm_ptr = get_block_num(item, pos_in_item);
 727                 if (unfm_ptr == 0) {
 728                         /* use allocated block to plug the hole */
 729                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 730                         if (fs_changed(fs_gen, inode->i_sb)
 731                             && item_moved(&tmp_ih, &path)) {
 732                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 733                                                                  bh);
 734                                 goto research;
 735                         }
 736                         set_buffer_new(bh_result);
 737                         if (buffer_dirty(bh_result)
 738                             && reiserfs_data_ordered(inode->i_sb))
 739                                 reiserfs_add_ordered_list(inode, bh_result);
 740                         put_block_num(item, pos_in_item, allocated_block_nr);
 741                         unfm_ptr = allocated_block_nr;
 742                         journal_mark_dirty(th, inode->i_sb, bh);
 743                         reiserfs_update_sd(th, inode);
 744                 }
 745                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 746                 pathrelse(&path);
 747                 retval = 0;
 748                 if (!dangle && th)
 749                         retval = reiserfs_end_persistent_transaction(th);
 750
 751                 reiserfs_write_unlock(inode->i_sb);
 752
 753                 /* the item was found, so new blocks were not added to the file
 754                  ** there is no need to make sure the inode is updated with this
 755                  ** transaction
 756                  */
 757                 return retval;
 758         }
 759
 760         if (!th) {
 761                 pathrelse(&path);
 762                 goto start_trans;
 763         }
 764
 765         /* desired position is not found or is in the direct item. We have
 766            to append file with holes up to 'block'-th block converting
 767            direct items to indirect one if necessary */
 768         done = 0;
 769         do {
 770                 if (is_statdata_le_ih(ih)) {
 771                         __le32 unp = 0;
 772                         struct cpu_key tmp_key;
 773
 774                         /* indirect item has to be inserted */
 775                         make_le_item_head(&tmp_ih, &key, version, 1,
 776                                           TYPE_INDIRECT, UNFM_P_SIZE,
 777                                           0 /* free_space */ );
 778
 779                         if (cpu_key_k_offset(&key) == 1) {
 780                                 /* we are going to add 'block'-th block to the file. Use
 781                                    allocated block for that */
 782                                 unp = cpu_to_le32(allocated_block_nr);
 783                                 set_block_dev_mapped(bh_result,
 784                                                      allocated_block_nr, inode);
 785                                 set_buffer_new(bh_result);
 786                                 done = 1;
 787                         }
 788                         tmp_key = key;  // ;)
 789                         set_cpu_key_k_offset(&tmp_key, 1);
 790                         PATH_LAST_POSITION(&path)++;
 791
 792                         retval =
 793                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 794                                                  inode, (char *)&unp);
 795                         if (retval) {
 796                                 reiserfs_free_block(th, inode,
 797                                                     allocated_block_nr, 1);
 798                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 799                         }
 800                         //mark_tail_converted (inode);
 801                 } else if (is_direct_le_ih(ih)) {
 802                         /* direct item has to be converted */
 803                         loff_t tail_offset;
 804
 805                         tail_offset =
 806                             ((le_ih_k_offset(ih) -
 807                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 808                         if (tail_offset == cpu_key_k_offset(&key)) {
 809                                 /* direct item we just found fits into block we have
 810                                    to map. Convert it into unformatted node: use
 811                                    bh_result for the conversion */
 812                                 set_block_dev_mapped(bh_result,
 813                                                      allocated_block_nr, inode);
 814                                 unbh = bh_result;
 815                                 done = 1;
 816                         } else {
 817                                 /* we have to padd file tail stored in direct item(s)
 818                                    up to block size and convert it to unformatted
 819                                    node. FIXME: this should also get into page cache */
 820
 821                                 pathrelse(&path);
 822                                 /*
 823                                  * ugly, but we can only end the transaction if
 824                                  * we aren't nested
 825                                  */
 826                                 BUG_ON(!th->t_refcount);
 827                                 if (th->t_refcount == 1) {
 828                                         retval =
 829                                             reiserfs_end_persistent_transaction
 830                                             (th);
 831                                         th = NULL;
 832                                         if (retval)
 833                                                 goto failure;
 834                                 }
 835
 836                                 retval =
 837                                     convert_tail_for_hole(inode, bh_result,
 838                                                           tail_offset);
 839                                 if (retval) {
 840                                         if (retval != -ENOSPC)
 841                                                 reiserfs_warning(inode->i_sb,
 842                                                                  "clm-6004: convert tail failed inode %lu, error %d",
 843                                                                  inode->i_ino,
 844                                                                  retval);
 845                                         if (allocated_block_nr) {
 846                                                 /* the bitmap, the super, and the stat data == 3 */
 847                                                 if (!th)
 848                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 849                                                 if (th)
 850                                                         reiserfs_free_block(th,
 851                                                                             inode,
 852                                                                             allocated_block_nr,
 853                                                                             1);
 854                                         }
 855                                         goto failure;
 856                                 }
 857                                 goto research;
 858                         }
 859                         retval =
 860                             direct2indirect(th, inode, &path, unbh,
 861                                             tail_offset);
 862                         if (retval) {
 863                                 reiserfs_unmap_buffer(unbh);
 864                                 reiserfs_free_block(th, inode,
 865                                                     allocated_block_nr, 1);
 866                                 goto failure;
 867                         }
 868                         /* it is important the set_buffer_uptodate is done after
 869                          ** the direct2indirect.  The buffer might contain valid
 870                          ** data newer than the data on disk (read by readpage, changed,
 871                          ** and then sent here by writepage).  direct2indirect needs
 872                          ** to know if unbh was already up to date, so it can decide
 873                          ** if the data in unbh needs to be replaced with data from
 874                          ** the disk
 875                          */
 876                         set_buffer_uptodate(unbh);
 877
 878                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 879                            buffer will disappear shortly, so it should not be added to
 880                          */
 881                         if (unbh->b_page) {
 882                                 /* we've converted the tail, so we must
 883                                  ** flush unbh before the transaction commits
 884                                  */
 885                                 reiserfs_add_tail_list(inode, unbh);
 886
 887                                 /* mark it dirty now to prevent commit_write from adding
 888                                  ** this buffer to the inode's dirty buffer list
 889                                  */
 890                                 /*
 891                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 892                                  * It's still atomic, but it sets the page dirty too,
 893                                  * which makes it eligible for writeback at any time by the
 894                                  * VM (which was also the case with __mark_buffer_dirty())
 895                                  */
 896                                 mark_buffer_dirty(unbh);
 897                         }
 898                 } else {
 899                         /* append indirect item with holes if needed, when appending
 900                            pointer to 'block'-th block use block, which is already
 901                            allocated */
 902                         struct cpu_key tmp_key;
 903                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 904                         // one block which is a fastpath
 905                         unp_t *un;
 906                         __u64 max_to_insert =
 907                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 908                             UNFM_P_SIZE;
 909                         __u64 blocks_needed;
 910
 911                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 912                                "vs-804: invalid position for append");
 913                         /* indirect item has to be appended, set up key of that position */
 914                         make_cpu_key(&tmp_key, inode,
 915                                      le_key_k_offset(version,
 916                                                      &(ih->ih_key)) +
 917                                      op_bytes_number(ih,
 918                                                      inode->i_sb->s_blocksize),
 919                                      //pos_in_item * inode->i_sb->s_blocksize,
 920                                      TYPE_INDIRECT, 3); // key type is unimportant
 921
 922                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 923                                "green-805: invalid offset");
 924                         blocks_needed =
 925                             1 +
 926                             ((cpu_key_k_offset(&key) -
 927                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 928                              s_blocksize_bits);
 929
 930                         if (blocks_needed == 1) {
 931                                 un = &unf_single;
 932                         } else {
 933                                 un = kmalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 934                                 if (!un) {
 935                                         un = &unf_single;
 936                                         blocks_needed = 1;
 937                                         max_to_insert = 0;
 938                                 } else
 939                                         memset(un, 0,
 940                                                UNFM_P_SIZE * min(blocks_needed,
 941                                                                  max_to_insert));
 942                         }
 943                         if (blocks_needed <= max_to_insert) {
 944                                 /* we are going to add target block to the file. Use allocated
 945                                    block for that */
 946                                 un[blocks_needed - 1] =
 947                                     cpu_to_le32(allocated_block_nr);
 948                                 set_block_dev_mapped(bh_result,
 949                                                      allocated_block_nr, inode);
 950                                 set_buffer_new(bh_result);
 951                                 done = 1;
 952                         } else {
 953                                 /* paste hole to the indirect item */
 954                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 955                                    only have space for one block */
 956                                 blocks_needed =
 957                                     max_to_insert ? max_to_insert : 1;
 958                         }
 959                         retval =
 960                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 961                                                      (char *)un,
 962                                                      UNFM_P_SIZE *
 963                                                      blocks_needed);
 964
 965                         if (blocks_needed != 1)
 966                                 kfree(un);
 967
 968                         if (retval) {
 969                                 reiserfs_free_block(th, inode,
 970                                                     allocated_block_nr, 1);
 971                                 goto failure;
 972                         }
 973                         if (!done) {
 974                                 /* We need to mark new file size in case this function will be
 975                                    interrupted/aborted later on. And we may do this only for
 976                                    holes. */
 977                                 inode->i_size +=
 978                                     inode->i_sb->s_blocksize * blocks_needed;
 979                         }
 980                 }
 981
 982                 if (done == 1)
 983                         break;
 984
 985                 /* this loop could log more blocks than we had originally asked
 986                  ** for.  So, we have to allow the transaction to end if it is
 987                  ** too big or too full.  Update the inode so things are
 988                  ** consistent if we crash before the function returns
 989                  **
 990                  ** release the path so that anybody waiting on the path before
 991                  ** ending their transaction will be able to continue.
 992                  */
 993                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 994                         retval = restart_transaction(th, inode, &path);
 995                         if (retval)
 996                                 goto failure;
 997                 }
 998                 /* inserting indirect pointers for a hole can take a
 999                  ** long time.  reschedule if needed
1000                  */
1001                 cond_resched();
1002
1003                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1004                 if (retval == IO_ERROR) {
1005                         retval = -EIO;
1006                         goto failure;
1007                 }
1008                 if (retval == POSITION_FOUND) {
1009                         reiserfs_warning(inode->i_sb,
1010                                          "vs-825: reiserfs_get_block: "
1011                                          "%K should not be found", &key);
1012                         retval = -EEXIST;
1013                         if (allocated_block_nr)
1014                                 reiserfs_free_block(th, inode,
1015                                                     allocated_block_nr, 1);
1016                         pathrelse(&path);
1017                         goto failure;
1018                 }
1019                 bh = get_last_bh(&path);
1020                 ih = get_ih(&path);
1021                 item = get_item(&path);
1022                 pos_in_item = path.pos_in_item;
1023         } while (1);
1024
1025         retval = 0;
1026
1027       failure:
1028         if (th && (!dangle || (retval && !th->t_trans_id))) {
1029                 int err;
1030                 if (th->t_trans_id)
1031                         reiserfs_update_sd(th, inode);
1032                 err = reiserfs_end_persistent_transaction(th);
1033                 if (err)
1034                         retval = err;
1035         }
1036
1037         reiserfs_write_unlock(inode->i_sb);
1038         reiserfs_check_path(&path);
1039         return retval;
1040 }
1041
1042 static int
1043 reiserfs_readpages(struct file *file, struct address_space *mapping,
1044                    struct list_head *pages, unsigned nr_pages)
1045 {
1046         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1047 }
1048
1049 /* Compute real number of used bytes by file
1050  * Following three functions can go away when we'll have enough space in stat item
1051  */
1052 static int real_space_diff(struct inode *inode, int sd_size)
1053 {
1054         int bytes;
1055         loff_t blocksize = inode->i_sb->s_blocksize;
1056
1057         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1058                 return sd_size;
1059
1060         /* End of file is also in full block with indirect reference, so round
1061          ** up to the next block.
1062          **
1063          ** there is just no way to know if the tail is actually packed
1064          ** on the file, so we have to assume it isn't.  When we pack the
1065          ** tail, we add 4 bytes to pretend there really is an unformatted
1066          ** node pointer
1067          */
1068         bytes =
1069             ((inode->i_size +
1070               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1071             sd_size;
1072         return bytes;
1073 }
1074
1075 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1076                                         int sd_size)
1077 {
1078         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1079                 return inode->i_size +
1080                     (loff_t) (real_space_diff(inode, sd_size));
1081         }
1082         return ((loff_t) real_space_diff(inode, sd_size)) +
1083             (((loff_t) blocks) << 9);
1084 }
1085
1086 /* Compute number of blocks used by file in ReiserFS counting */
1087 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1088 {
1089         loff_t bytes = inode_get_bytes(inode);
1090         loff_t real_space = real_space_diff(inode, sd_size);
1091
1092         /* keeps fsck and non-quota versions of reiserfs happy */
1093         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1094                 bytes += (loff_t) 511;
1095         }
1096
1097         /* files from before the quota patch might i_blocks such that
1098          ** bytes < real_space.  Deal with that here to prevent it from
1099          ** going negative.
1100          */
1101         if (bytes < real_space)
1102                 return 0;
1103         return (bytes - real_space) >> 9;
1104 }
1105
1106 //
1107 // BAD: new directories have stat data of new type and all other items
1108 // of old type. Version stored in the inode says about body items, so
1109 // in update_stat_data we can not rely on inode, but have to check
1110 // item version directly
1111 //
1112
1113 // called by read_locked_inode
1114 static void init_inode(struct inode *inode, struct path *path)
1115 {
1116         struct buffer_head *bh;
1117         struct item_head *ih;
1118         __u32 rdev;
1119         //int version = ITEM_VERSION_1;
1120
1121         bh = PATH_PLAST_BUFFER(path);
1122         ih = PATH_PITEM_HEAD(path);
1123
1124         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1125         inode->i_blksize = reiserfs_default_io_size;
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         REISERFS_I(inode)->i_acl_access = NULL;
1134         REISERFS_I(inode)->i_acl_default = NULL;
1135         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1136
1137         if (stat_data_v1(ih)) {
1138                 struct stat_data_v1 *sd =
1139                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1140                 unsigned long blocks;
1141
1142                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1143                 set_inode_sd_version(inode, STAT_DATA_V1);
1144                 inode->i_mode = sd_v1_mode(sd);
1145                 inode->i_nlink = sd_v1_nlink(sd);
1146                 inode->i_uid = sd_v1_uid(sd);
1147                 inode->i_gid = sd_v1_gid(sd);
1148                 inode->i_size = sd_v1_size(sd);
1149                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1150                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1151                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1152                 inode->i_atime.tv_nsec = 0;
1153                 inode->i_ctime.tv_nsec = 0;
1154                 inode->i_mtime.tv_nsec = 0;
1155
1156                 inode->i_blocks = sd_v1_blocks(sd);
1157                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1158                 blocks = (inode->i_size + 511) >> 9;
1159                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1160                 if (inode->i_blocks > blocks) {
1161                         // there was a bug in <=3.5.23 when i_blocks could take negative
1162                         // values. Starting from 3.5.17 this value could even be stored in
1163                         // stat data. For such files we set i_blocks based on file
1164                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1165                         // only updated if file's inode will ever change
1166                         inode->i_blocks = blocks;
1167                 }
1168
1169                 rdev = sd_v1_rdev(sd);
1170                 REISERFS_I(inode)->i_first_direct_byte =
1171                     sd_v1_first_direct_byte(sd);
1172                 /* an early bug in the quota code can give us an odd number for the
1173                  ** block count.  This is incorrect, fix it here.
1174                  */
1175                 if (inode->i_blocks & 1) {
1176                         inode->i_blocks++;
1177                 }
1178                 inode_set_bytes(inode,
1179                                 to_real_used_space(inode, inode->i_blocks,
1180                                                    SD_V1_SIZE));
1181                 /* nopack is initially zero for v1 objects. For v2 objects,
1182                    nopack is initialised from sd_attrs */
1183                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1184         } else {
1185                 // new stat data found, but object may have old items
1186                 // (directories and symlinks)
1187                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1188
1189                 inode->i_mode = sd_v2_mode(sd);
1190                 inode->i_nlink = sd_v2_nlink(sd);
1191                 inode->i_uid = sd_v2_uid(sd);
1192                 inode->i_size = sd_v2_size(sd);
1193                 inode->i_gid = sd_v2_gid(sd);
1194                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1195                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1196                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1197                 inode->i_ctime.tv_nsec = 0;
1198                 inode->i_mtime.tv_nsec = 0;
1199                 inode->i_atime.tv_nsec = 0;
1200                 inode->i_blocks = sd_v2_blocks(sd);
1201                 rdev = sd_v2_rdev(sd);
1202                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1203                         inode->i_generation =
1204                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1205                 else
1206                         inode->i_generation = sd_v2_generation(sd);
1207
1208                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1209                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1210                 else
1211                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1212                 REISERFS_I(inode)->i_first_direct_byte = 0;
1213                 set_inode_sd_version(inode, STAT_DATA_V2);
1214                 inode_set_bytes(inode,
1215                                 to_real_used_space(inode, inode->i_blocks,
1216                                                    SD_V2_SIZE));
1217                 /* read persistent inode attributes from sd and initalise
1218                    generic inode flags from them */
1219                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1220                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1221         }
1222
1223         pathrelse(path);
1224         if (S_ISREG(inode->i_mode)) {
1225                 inode->i_op = &reiserfs_file_inode_operations;
1226                 inode->i_fop = &reiserfs_file_operations;
1227                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1228         } else if (S_ISDIR(inode->i_mode)) {
1229                 inode->i_op = &reiserfs_dir_inode_operations;
1230                 inode->i_fop = &reiserfs_dir_operations;
1231         } else if (S_ISLNK(inode->i_mode)) {
1232                 inode->i_op = &reiserfs_symlink_inode_operations;
1233                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1234         } else {
1235                 inode->i_blocks = 0;
1236                 inode->i_op = &reiserfs_special_inode_operations;
1237                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1238         }
1239 }
1240
1241 // update new stat data with inode fields
1242 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1243 {
1244         struct stat_data *sd_v2 = (struct stat_data *)sd;
1245         __u16 flags;
1246
1247         set_sd_v2_mode(sd_v2, inode->i_mode);
1248         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1249         set_sd_v2_uid(sd_v2, inode->i_uid);
1250         set_sd_v2_size(sd_v2, size);
1251         set_sd_v2_gid(sd_v2, inode->i_gid);
1252         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1253         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1254         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1255         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1256         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1257                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1258         else
1259                 set_sd_v2_generation(sd_v2, inode->i_generation);
1260         flags = REISERFS_I(inode)->i_attrs;
1261         i_attrs_to_sd_attrs(inode, &flags);
1262         set_sd_v2_attrs(sd_v2, flags);
1263 }
1264
1265 // used to copy inode's fields to old stat data
1266 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1267 {
1268         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1269
1270         set_sd_v1_mode(sd_v1, inode->i_mode);
1271         set_sd_v1_uid(sd_v1, inode->i_uid);
1272         set_sd_v1_gid(sd_v1, inode->i_gid);
1273         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1274         set_sd_v1_size(sd_v1, size);
1275         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1276         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1277         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1278
1279         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1280                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1281         else
1282                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1283
1284         // Sigh. i_first_direct_byte is back
1285         set_sd_v1_first_direct_byte(sd_v1,
1286                                     REISERFS_I(inode)->i_first_direct_byte);
1287 }
1288
1289 /* NOTE, you must prepare the buffer head before sending it here,
1290 ** and then log it after the call
1291 */
1292 static void update_stat_data(struct path *path, struct inode *inode,
1293                              loff_t size)
1294 {
1295         struct buffer_head *bh;
1296         struct item_head *ih;
1297
1298         bh = PATH_PLAST_BUFFER(path);
1299         ih = PATH_PITEM_HEAD(path);
1300
1301         if (!is_statdata_le_ih(ih))
1302                 reiserfs_panic(inode->i_sb,
1303                                "vs-13065: update_stat_data: key %k, found item %h",
1304                                INODE_PKEY(inode), ih);
1305
1306         if (stat_data_v1(ih)) {
1307                 // path points to old stat data
1308                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1309         } else {
1310                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1311         }
1312
1313         return;
1314 }
1315
1316 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1317                              struct inode *inode, loff_t size)
1318 {
1319         struct cpu_key key;
1320         INITIALIZE_PATH(path);
1321         struct buffer_head *bh;
1322         int fs_gen;
1323         struct item_head *ih, tmp_ih;
1324         int retval;
1325
1326         BUG_ON(!th->t_trans_id);
1327
1328         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1329
1330         for (;;) {
1331                 int pos;
1332                 /* look for the object's stat data */
1333                 retval = search_item(inode->i_sb, &key, &path);
1334                 if (retval == IO_ERROR) {
1335                         reiserfs_warning(inode->i_sb,
1336                                          "vs-13050: reiserfs_update_sd: "
1337                                          "i/o failure occurred trying to update %K stat data",
1338                                          &key);
1339                         return;
1340                 }
1341                 if (retval == ITEM_NOT_FOUND) {
1342                         pos = PATH_LAST_POSITION(&path);
1343                         pathrelse(&path);
1344                         if (inode->i_nlink == 0) {
1345                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1346                                 return;
1347                         }
1348                         reiserfs_warning(inode->i_sb,
1349                                          "vs-13060: reiserfs_update_sd: "
1350                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1351                                          INODE_PKEY(inode), inode->i_nlink,
1352                                          pos);
1353                         reiserfs_check_path(&path);
1354                         return;
1355                 }
1356
1357                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1358                  ** FS might change.  We have to detect that, and loop back to the
1359                  ** search if the stat data item has moved
1360                  */
1361                 bh = get_last_bh(&path);
1362                 ih = get_ih(&path);
1363                 copy_item_head(&tmp_ih, ih);
1364                 fs_gen = get_generation(inode->i_sb);
1365                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1366                 if (fs_changed(fs_gen, inode->i_sb)
1367                     && item_moved(&tmp_ih, &path)) {
1368                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1369                         continue;       /* Stat_data item has been moved after scheduling. */
1370                 }
1371                 break;
1372         }
1373         update_stat_data(&path, inode, size);
1374         journal_mark_dirty(th, th->t_super, bh);
1375         pathrelse(&path);
1376         return;
1377 }
1378
1379 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1380 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1381 ** and clear the key in the private portion of the inode, otherwise a
1382 ** corresponding iput might try to delete whatever object the inode last
1383 ** represented.
1384 */
1385 static void reiserfs_make_bad_inode(struct inode *inode)
1386 {
1387         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1388         make_bad_inode(inode);
1389 }
1390
1391 //
1392 // initially this function was derived from minix or ext2's analog and
1393 // evolved as the prototype did
1394 //
1395
1396 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1397 {
1398         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1399         inode->i_ino = args->objectid;
1400         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1401         return 0;
1402 }
1403
1404 /* looks for stat data in the tree, and fills up the fields of in-core
1405    inode stat data fields */
1406 void reiserfs_read_locked_inode(struct inode *inode,
1407                                 struct reiserfs_iget_args *args)
1408 {
1409         INITIALIZE_PATH(path_to_sd);
1410         struct cpu_key key;
1411         unsigned long dirino;
1412         int retval;
1413
1414         dirino = args->dirid;
1415
1416         /* set version 1, version 2 could be used too, because stat data
1417            key is the same in both versions */
1418         key.version = KEY_FORMAT_3_5;
1419         key.on_disk_key.k_dir_id = dirino;
1420         key.on_disk_key.k_objectid = inode->i_ino;
1421         key.on_disk_key.k_offset = 0;
1422         key.on_disk_key.k_type = 0;
1423
1424         /* look for the object's stat data */
1425         retval = search_item(inode->i_sb, &key, &path_to_sd);
1426         if (retval == IO_ERROR) {
1427                 reiserfs_warning(inode->i_sb,
1428                                  "vs-13070: reiserfs_read_locked_inode: "
1429                                  "i/o failure occurred trying to find stat data of %K",
1430                                  &key);
1431                 reiserfs_make_bad_inode(inode);
1432                 return;
1433         }
1434         if (retval != ITEM_FOUND) {
1435                 /* a stale NFS handle can trigger this without it being an error */
1436                 pathrelse(&path_to_sd);
1437                 reiserfs_make_bad_inode(inode);
1438                 inode->i_nlink = 0;
1439                 return;
1440         }
1441
1442         init_inode(inode, &path_to_sd);
1443
1444         /* It is possible that knfsd is trying to access inode of a file
1445            that is being removed from the disk by some other thread. As we
1446            update sd on unlink all that is required is to check for nlink
1447            here. This bug was first found by Sizif when debugging
1448            SquidNG/Butterfly, forgotten, and found again after Philippe
1449            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1450
1451            More logical fix would require changes in fs/inode.c:iput() to
1452            remove inode from hash-table _after_ fs cleaned disk stuff up and
1453            in iget() to return NULL if I_FREEING inode is found in
1454            hash-table. */
1455         /* Currently there is one place where it's ok to meet inode with
1456            nlink==0: processing of open-unlinked and half-truncated files
1457            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1458         if ((inode->i_nlink == 0) &&
1459             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1460                 reiserfs_warning(inode->i_sb,
1461                                  "vs-13075: reiserfs_read_locked_inode: "
1462                                  "dead inode read from disk %K. "
1463                                  "This is likely to be race with knfsd. Ignore",
1464                                  &key);
1465                 reiserfs_make_bad_inode(inode);
1466         }
1467
1468         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1469
1470 }
1471
1472 /**
1473  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1474  *
1475  * @inode:    inode from hash table to check
1476  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1477  *
1478  * This function is called by iget5_locked() to distinguish reiserfs inodes
1479  * having the same inode numbers. Such inodes can only exist due to some
1480  * error condition. One of them should be bad. Inodes with identical
1481  * inode numbers (objectids) are distinguished by parent directory ids.
1482  *
1483  */
1484 int reiserfs_find_actor(struct inode *inode, void *opaque)
1485 {
1486         struct reiserfs_iget_args *args;
1487
1488         args = opaque;
1489         /* args is already in CPU order */
1490         return (inode->i_ino == args->objectid) &&
1491             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1492 }
1493
1494 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1495 {
1496         struct inode *inode;
1497         struct reiserfs_iget_args args;
1498
1499         args.objectid = key->on_disk_key.k_objectid;
1500         args.dirid = key->on_disk_key.k_dir_id;
1501         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1502                              reiserfs_find_actor, reiserfs_init_locked_inode,
1503                              (void *)(&args));
1504         if (!inode)
1505                 return ERR_PTR(-ENOMEM);
1506
1507         if (inode->i_state & I_NEW) {
1508                 reiserfs_read_locked_inode(inode, &args);
1509                 unlock_new_inode(inode);
1510         }
1511
1512         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1513                 /* either due to i/o error or a stale NFS handle */
1514                 iput(inode);
1515                 inode = NULL;
1516         }
1517         return inode;
1518 }
1519
1520 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1521 {
1522         __u32 *data = vobjp;
1523         struct cpu_key key;
1524         struct dentry *result;
1525         struct inode *inode;
1526
1527         key.on_disk_key.k_objectid = data[0];
1528         key.on_disk_key.k_dir_id = data[1];
1529         reiserfs_write_lock(sb);
1530         inode = reiserfs_iget(sb, &key);
1531         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1532             data[2] != inode->i_generation) {
1533                 iput(inode);
1534                 inode = NULL;
1535         }
1536         reiserfs_write_unlock(sb);
1537         if (!inode)
1538                 inode = ERR_PTR(-ESTALE);
1539         if (IS_ERR(inode))
1540                 return ERR_PTR(PTR_ERR(inode));
1541         result = d_alloc_anon(inode);
1542         if (!result) {
1543                 iput(inode);
1544                 return ERR_PTR(-ENOMEM);
1545         }
1546         return result;
1547 }
1548
1549 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1550                                   int len, int fhtype,
1551                                   int (*acceptable) (void *contect,
1552                                                      struct dentry * de),
1553                                   void *context)
1554 {
1555         __u32 obj[3], parent[3];
1556
1557         /* fhtype happens to reflect the number of u32s encoded.
1558          * due to a bug in earlier code, fhtype might indicate there
1559          * are more u32s then actually fitted.
1560          * so if fhtype seems to be more than len, reduce fhtype.
1561          * Valid types are:
1562          *   2 - objectid + dir_id - legacy support
1563          *   3 - objectid + dir_id + generation
1564          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1565          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1566          *   6 - as above plus generation of directory
1567          * 6 does not fit in NFSv2 handles
1568          */
1569         if (fhtype > len) {
1570                 if (fhtype != 6 || len != 5)
1571                         reiserfs_warning(sb,
1572                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1573                                          fhtype, len);
1574                 fhtype = 5;
1575         }
1576
1577         obj[0] = data[0];
1578         obj[1] = data[1];
1579         if (fhtype == 3 || fhtype >= 5)
1580                 obj[2] = data[2];
1581         else
1582                 obj[2] = 0;     /* generation number */
1583
1584         if (fhtype >= 4) {
1585                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1586                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1587                 if (fhtype == 6)
1588                         parent[2] = data[5];
1589                 else
1590                         parent[2] = 0;
1591         }
1592         return sb->s_export_op->find_exported_dentry(sb, obj,
1593                                                      fhtype < 4 ? NULL : parent,
1594                                                      acceptable, context);
1595 }
1596
1597 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1598                        int need_parent)
1599 {
1600         struct inode *inode = dentry->d_inode;
1601         int maxlen = *lenp;
1602
1603         if (maxlen < 3)
1604                 return 255;
1605
1606         data[0] = inode->i_ino;
1607         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1608         data[2] = inode->i_generation;
1609         *lenp = 3;
1610         /* no room for directory info? return what we've stored so far */
1611         if (maxlen < 5 || !need_parent)
1612                 return 3;
1613
1614         spin_lock(&dentry->d_lock);
1615         inode = dentry->d_parent->d_inode;
1616         data[3] = inode->i_ino;
1617         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1618         *lenp = 5;
1619         if (maxlen >= 6) {
1620                 data[5] = inode->i_generation;
1621                 *lenp = 6;
1622         }
1623         spin_unlock(&dentry->d_lock);
1624         return *lenp;
1625 }
1626
1627 /* looks for stat data, then copies fields to it, marks the buffer
1628    containing stat data as dirty */
1629 /* reiserfs inodes are never really dirty, since the dirty inode call
1630 ** always logs them.  This call allows the VFS inode marking routines
1631 ** to properly mark inodes for datasync and such, but only actually
1632 ** does something when called for a synchronous update.
1633 */
1634 int reiserfs_write_inode(struct inode *inode, int do_sync)
1635 {
1636         struct reiserfs_transaction_handle th;
1637         int jbegin_count = 1;
1638
1639         if (inode->i_sb->s_flags & MS_RDONLY)
1640                 return -EROFS;
1641         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1642          ** these cases are just when the system needs ram, not when the
1643          ** inode needs to reach disk for safety, and they can safely be
1644          ** ignored because the altered inode has already been logged.
1645          */
1646         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1647                 reiserfs_write_lock(inode->i_sb);
1648                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1649                         reiserfs_update_sd(&th, inode);
1650                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1651                 }
1652                 reiserfs_write_unlock(inode->i_sb);
1653         }
1654         return 0;
1655 }
1656
1657 /* stat data of new object is inserted already, this inserts the item
1658    containing "." and ".." entries */
1659 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1660                                   struct inode *inode,
1661                                   struct item_head *ih, struct path *path,
1662                                   struct inode *dir)
1663 {
1664         struct super_block *sb = th->t_super;
1665         char empty_dir[EMPTY_DIR_SIZE];
1666         char *body = empty_dir;
1667         struct cpu_key key;
1668         int retval;
1669
1670         BUG_ON(!th->t_trans_id);
1671
1672         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1673                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1674                       TYPE_DIRENTRY, 3 /*key length */ );
1675
1676         /* compose item head for new item. Directories consist of items of
1677            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1678            is done by reiserfs_new_inode */
1679         if (old_format_only(sb)) {
1680                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1681                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1682
1683                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1684                                        ih->ih_key.k_objectid,
1685                                        INODE_PKEY(dir)->k_dir_id,
1686                                        INODE_PKEY(dir)->k_objectid);
1687         } else {
1688                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1689                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1690
1691                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1692                                     ih->ih_key.k_objectid,
1693                                     INODE_PKEY(dir)->k_dir_id,
1694                                     INODE_PKEY(dir)->k_objectid);
1695         }
1696
1697         /* look for place in the tree for new item */
1698         retval = search_item(sb, &key, path);
1699         if (retval == IO_ERROR) {
1700                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1701                                  "i/o failure occurred creating new directory");
1702                 return -EIO;
1703         }
1704         if (retval == ITEM_FOUND) {
1705                 pathrelse(path);
1706                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1707                                  "object with this key exists (%k)",
1708                                  &(ih->ih_key));
1709                 return -EEXIST;
1710         }
1711
1712         /* insert item, that is empty directory item */
1713         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1714 }
1715
1716 /* stat data of object has been inserted, this inserts the item
1717    containing the body of symlink */
1718 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1719                                 struct item_head *ih,
1720                                 struct path *path, const char *symname,
1721                                 int item_len)
1722 {
1723         struct super_block *sb = th->t_super;
1724         struct cpu_key key;
1725         int retval;
1726
1727         BUG_ON(!th->t_trans_id);
1728
1729         _make_cpu_key(&key, KEY_FORMAT_3_5,
1730                       le32_to_cpu(ih->ih_key.k_dir_id),
1731                       le32_to_cpu(ih->ih_key.k_objectid),
1732                       1, TYPE_DIRECT, 3 /*key length */ );
1733
1734         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1735                           0 /*free_space */ );
1736
1737         /* look for place in the tree for new item */
1738         retval = search_item(sb, &key, path);
1739         if (retval == IO_ERROR) {
1740                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1741                                  "i/o failure occurred creating new symlink");
1742                 return -EIO;
1743         }
1744         if (retval == ITEM_FOUND) {
1745                 pathrelse(path);
1746                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1747                                  "object with this key exists (%k)",
1748                                  &(ih->ih_key));
1749                 return -EEXIST;
1750         }
1751
1752         /* insert item, that is body of symlink */
1753         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1754 }
1755
1756 /* inserts the stat data into the tree, and then calls
1757    reiserfs_new_directory (to insert ".", ".." item if new object is
1758    directory) or reiserfs_new_symlink (to insert symlink body if new
1759    object is symlink) or nothing (if new object is regular file)
1760
1761    NOTE! uid and gid must already be set in the inode.  If we return
1762    non-zero due to an error, we have to drop the quota previously allocated
1763    for the fresh inode.  This can only be done outside a transaction, so
1764    if we return non-zero, we also end the transaction.  */
1765 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1766                        struct inode *dir, int mode, const char *symname,
1767                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1768                           strlen (symname) for symlinks) */
1769                        loff_t i_size, struct dentry *dentry,
1770                        struct inode *inode)
1771 {
1772         struct super_block *sb;
1773         INITIALIZE_PATH(path_to_key);
1774         struct cpu_key key;
1775         struct item_head ih;
1776         struct stat_data sd;
1777         int retval;
1778         int err;
1779
1780         BUG_ON(!th->t_trans_id);
1781
1782         if (DQUOT_ALLOC_INODE(inode)) {
1783                 err = -EDQUOT;
1784                 goto out_end_trans;
1785         }
1786         if (!dir || !dir->i_nlink) {
1787                 err = -EPERM;
1788                 goto out_bad_inode;
1789         }
1790
1791         sb = dir->i_sb;
1792
1793         /* item head of new item */
1794         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1795         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1796         if (!ih.ih_key.k_objectid) {
1797                 err = -ENOMEM;
1798                 goto out_bad_inode;
1799         }
1800         if (old_format_only(sb))
1801                 /* not a perfect generation count, as object ids can be reused, but
1802                  ** this is as good as reiserfs can do right now.
1803                  ** note that the private part of inode isn't filled in yet, we have
1804                  ** to use the directory.
1805                  */
1806                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1807         else
1808 #if defined( USE_INODE_GENERATION_COUNTER )
1809                 inode->i_generation =
1810                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1811 #else
1812                 inode->i_generation = ++event;
1813 #endif
1814
1815         /* fill stat data */
1816         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1817
1818         /* uid and gid must already be set by the caller for quota init */
1819
1820         /* symlink cannot be immutable or append only, right? */
1821         if (S_ISLNK(inode->i_mode))
1822                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1823
1824         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1825         inode->i_size = i_size;
1826         inode->i_blocks = 0;
1827         inode->i_bytes = 0;
1828         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1829             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1830
1831         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1832         REISERFS_I(inode)->i_flags = 0;
1833         REISERFS_I(inode)->i_prealloc_block = 0;
1834         REISERFS_I(inode)->i_prealloc_count = 0;
1835         REISERFS_I(inode)->i_trans_id = 0;
1836         REISERFS_I(inode)->i_jl = NULL;
1837         REISERFS_I(inode)->i_attrs =
1838             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1839         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1840         REISERFS_I(inode)->i_acl_access = NULL;
1841         REISERFS_I(inode)->i_acl_default = NULL;
1842         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1843
1844         if (old_format_only(sb))
1845                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1846                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1847         else
1848                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1849                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1850
1851         /* key to search for correct place for new stat data */
1852         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1853                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1854                       TYPE_STAT_DATA, 3 /*key length */ );
1855
1856         /* find proper place for inserting of stat data */
1857         retval = search_item(sb, &key, &path_to_key);
1858         if (retval == IO_ERROR) {
1859                 err = -EIO;
1860                 goto out_bad_inode;
1861         }
1862         if (retval == ITEM_FOUND) {
1863                 pathrelse(&path_to_key);
1864                 err = -EEXIST;
1865                 goto out_bad_inode;
1866         }
1867         if (old_format_only(sb)) {
1868                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1869                         pathrelse(&path_to_key);
1870                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1871                         err = -EINVAL;
1872                         goto out_bad_inode;
1873                 }
1874                 inode2sd_v1(&sd, inode, inode->i_size);
1875         } else {
1876                 inode2sd(&sd, inode, inode->i_size);
1877         }
1878         // these do not go to on-disk stat data
1879         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1880         inode->i_blksize = reiserfs_default_io_size;
1881
1882         // store in in-core inode the key of stat data and version all
1883         // object items will have (directory items will have old offset
1884         // format, other new objects will consist of new items)
1885         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1886         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1887                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1888         else
1889                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1890         if (old_format_only(sb))
1891                 set_inode_sd_version(inode, STAT_DATA_V1);
1892         else
1893                 set_inode_sd_version(inode, STAT_DATA_V2);
1894
1895         /* insert the stat data into the tree */
1896 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1897         if (REISERFS_I(dir)->new_packing_locality)
1898                 th->displace_new_blocks = 1;
1899 #endif
1900         retval =
1901             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1902                                  (char *)(&sd));
1903         if (retval) {
1904                 err = retval;
1905                 reiserfs_check_path(&path_to_key);
1906                 goto out_bad_inode;
1907         }
1908 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1909         if (!th->displace_new_blocks)
1910                 REISERFS_I(dir)->new_packing_locality = 0;
1911 #endif
1912         if (S_ISDIR(mode)) {
1913                 /* insert item with "." and ".." */
1914                 retval =
1915                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1916         }
1917
1918         if (S_ISLNK(mode)) {
1919                 /* insert body of symlink */
1920                 if (!old_format_only(sb))
1921                         i_size = ROUND_UP(i_size);
1922                 retval =
1923                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1924                                          i_size);
1925         }
1926         if (retval) {
1927                 err = retval;
1928                 reiserfs_check_path(&path_to_key);
1929                 journal_end(th, th->t_super, th->t_blocks_allocated);
1930                 goto out_inserted_sd;
1931         }
1932
1933         /* XXX CHECK THIS */
1934         if (reiserfs_posixacl(inode->i_sb)) {
1935                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1936                 if (retval) {
1937                         err = retval;
1938                         reiserfs_check_path(&path_to_key);
1939                         journal_end(th, th->t_super, th->t_blocks_allocated);
1940                         goto out_inserted_sd;
1941                 }
1942         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1943                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1944                                  "but vfs thinks they are!");
1945         } else if (is_reiserfs_priv_object(dir)) {
1946                 reiserfs_mark_inode_private(inode);
1947         }
1948
1949         insert_inode_hash(inode);
1950         reiserfs_update_sd(th, inode);
1951         reiserfs_check_path(&path_to_key);
1952
1953         return 0;
1954
1955 /* it looks like you can easily compress these two goto targets into
1956  * one.  Keeping it like this doesn't actually hurt anything, and they
1957  * are place holders for what the quota code actually needs.
1958  */
1959       out_bad_inode:
1960         /* Invalidate the object, nothing was inserted yet */
1961         INODE_PKEY(inode)->k_objectid = 0;
1962
1963         /* Quota change must be inside a transaction for journaling */
1964         DQUOT_FREE_INODE(inode);
1965
1966       out_end_trans:
1967         journal_end(th, th->t_super, th->t_blocks_allocated);
1968         /* Drop can be outside and it needs more credits so it's better to have it outside */
1969         DQUOT_DROP(inode);
1970         inode->i_flags |= S_NOQUOTA;
1971         make_bad_inode(inode);
1972
1973       out_inserted_sd:
1974         inode->i_nlink = 0;
1975         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1976
1977         /* If we were inheriting an ACL, we need to release the lock so that
1978          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1979          * code really needs to be reworked, but this will take care of it
1980          * for now. -jeffm */
1981         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1982                 reiserfs_write_unlock_xattrs(dir->i_sb);
1983                 iput(inode);
1984                 reiserfs_write_lock_xattrs(dir->i_sb);
1985         } else
1986                 iput(inode);
1987         return err;
1988 }
1989
1990 /*
1991 ** finds the tail page in the page cache,
1992 ** reads the last block in.
1993 **
1994 ** On success, page_result is set to a locked, pinned page, and bh_result
1995 ** is set to an up to date buffer for the last block in the file.  returns 0.
1996 **
1997 ** tail conversion is not done, so bh_result might not be valid for writing
1998 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1999 ** trying to write the block.
2000 **
2001 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2002 */
2003 static int grab_tail_page(struct inode *p_s_inode,
2004                           struct page **page_result,
2005                           struct buffer_head **bh_result)
2006 {
2007
2008         /* we want the page with the last byte in the file,
2009          ** not the page that will hold the next byte for appending
2010          */
2011         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2012         unsigned long pos = 0;
2013         unsigned long start = 0;
2014         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2015         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2016         struct buffer_head *bh;
2017         struct buffer_head *head;
2018         struct page *page;
2019         int error;
2020
2021         /* we know that we are only called with inode->i_size > 0.
2022          ** we also know that a file tail can never be as big as a block
2023          ** If i_size % blocksize == 0, our file is currently block aligned
2024          ** and it won't need converting or zeroing after a truncate.
2025          */
2026         if ((offset & (blocksize - 1)) == 0) {
2027                 return -ENOENT;
2028         }
2029         page = grab_cache_page(p_s_inode->i_mapping, index);
2030         error = -ENOMEM;
2031         if (!page) {
2032                 goto out;
2033         }
2034         /* start within the page of the last block in the file */
2035         start = (offset / blocksize) * blocksize;
2036
2037         error = block_prepare_write(page, start, offset,
2038                                     reiserfs_get_block_create_0);
2039         if (error)
2040                 goto unlock;
2041
2042         head = page_buffers(page);
2043         bh = head;
2044         do {
2045                 if (pos >= start) {
2046                         break;
2047                 }
2048                 bh = bh->b_this_page;
2049                 pos += blocksize;
2050         } while (bh != head);
2051
2052         if (!buffer_uptodate(bh)) {
2053                 /* note, this should never happen, prepare_write should
2054                  ** be taking care of this for us.  If the buffer isn't up to date,
2055                  ** I've screwed up the code to find the buffer, or the code to
2056                  ** call prepare_write
2057                  */
2058                 reiserfs_warning(p_s_inode->i_sb,
2059                                  "clm-6000: error reading block %lu on dev %s",
2060                                  bh->b_blocknr,
2061                                  reiserfs_bdevname(p_s_inode->i_sb));
2062                 error = -EIO;
2063                 goto unlock;
2064         }
2065         *bh_result = bh;
2066         *page_result = page;
2067
2068       out:
2069         return error;
2070
2071       unlock:
2072         unlock_page(page);
2073         page_cache_release(page);
2074         return error;
2075 }
2076
2077 /*
2078 ** vfs version of truncate file.  Must NOT be called with
2079 ** a transaction already started.
2080 **
2081 ** some code taken from block_truncate_page
2082 */
2083 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2084 {
2085         struct reiserfs_transaction_handle th;
2086         /* we want the offset for the first byte after the end of the file */
2087         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2088         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2089         unsigned length;
2090         struct page *page = NULL;
2091         int error;
2092         struct buffer_head *bh = NULL;
2093         int err2;
2094
2095         reiserfs_write_lock(p_s_inode->i_sb);
2096
2097         if (p_s_inode->i_size > 0) {
2098                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2099                         // -ENOENT means we truncated past the end of the file,
2100                         // and get_block_create_0 could not find a block to read in,
2101                         // which is ok.
2102                         if (error != -ENOENT)
2103                                 reiserfs_warning(p_s_inode->i_sb,
2104                                                  "clm-6001: grab_tail_page failed %d",
2105                                                  error);
2106                         page = NULL;
2107                         bh = NULL;
2108                 }
2109         }
2110
2111         /* so, if page != NULL, we have a buffer head for the offset at
2112          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2113          ** then we have an unformatted node.  Otherwise, we have a direct item,
2114          ** and no zeroing is required on disk.  We zero after the truncate,
2115          ** because the truncate might pack the item anyway
2116          ** (it will unmap bh if it packs).
2117          */
2118         /* it is enough to reserve space in transaction for 2 balancings:
2119            one for "save" link adding and another for the first
2120            cut_from_item. 1 is for update_sd */
2121         error = journal_begin(&th, p_s_inode->i_sb,
2122                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2123         if (error)
2124                 goto out;
2125         reiserfs_update_inode_transaction(p_s_inode);
2126         if (update_timestamps)
2127                 /* we are doing real truncate: if the system crashes before the last
2128                    transaction of truncating gets committed - on reboot the file
2129                    either appears truncated properly or not truncated at all */
2130                 add_save_link(&th, p_s_inode, 1);
2131         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2132         error =
2133             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2134         if (error)
2135                 goto out;
2136
2137         /* check reiserfs_do_truncate after ending the transaction */
2138         if (err2) {
2139                 error = err2;
2140                 goto out;
2141         }
2142
2143         if (update_timestamps) {
2144                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2145                 if (error)
2146                         goto out;
2147         }
2148
2149         if (page) {
2150                 length = offset & (blocksize - 1);
2151                 /* if we are not on a block boundary */
2152                 if (length) {
2153                         char *kaddr;
2154
2155                         length = blocksize - length;
2156                         kaddr = kmap_atomic(page, KM_USER0);
2157                         memset(kaddr + offset, 0, length);
2158                         flush_dcache_page(page);
2159                         kunmap_atomic(kaddr, KM_USER0);
2160                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2161                                 mark_buffer_dirty(bh);
2162                         }
2163                 }
2164                 unlock_page(page);
2165                 page_cache_release(page);
2166         }
2167
2168         reiserfs_write_unlock(p_s_inode->i_sb);
2169         return 0;
2170       out:
2171         if (page) {
2172                 unlock_page(page);
2173                 page_cache_release(page);
2174         }
2175         reiserfs_write_unlock(p_s_inode->i_sb);
2176         return error;
2177 }
2178
2179 static int map_block_for_writepage(struct inode *inode,
2180                                    struct buffer_head *bh_result,
2181                                    unsigned long block)
2182 {
2183         struct reiserfs_transaction_handle th;
2184         int fs_gen;
2185         struct item_head tmp_ih;
2186         struct item_head *ih;
2187         struct buffer_head *bh;
2188         __le32 *item;
2189         struct cpu_key key;
2190         INITIALIZE_PATH(path);
2191         int pos_in_item;
2192         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2193         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2194         int retval;
2195         int use_get_block = 0;
2196         int bytes_copied = 0;
2197         int copy_size;
2198         int trans_running = 0;
2199
2200         /* catch places below that try to log something without starting a trans */
2201         th.t_trans_id = 0;
2202
2203         if (!buffer_uptodate(bh_result)) {
2204                 return -EIO;
2205         }
2206
2207         kmap(bh_result->b_page);
2208       start_over:
2209         reiserfs_write_lock(inode->i_sb);
2210         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2211
2212       research:
2213         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2214         if (retval != POSITION_FOUND) {
2215                 use_get_block = 1;
2216                 goto out;
2217         }
2218
2219         bh = get_last_bh(&path);
2220         ih = get_ih(&path);
2221         item = get_item(&path);
2222         pos_in_item = path.pos_in_item;
2223
2224         /* we've found an unformatted node */
2225         if (indirect_item_found(retval, ih)) {
2226                 if (bytes_copied > 0) {
2227                         reiserfs_warning(inode->i_sb,
2228                                          "clm-6002: bytes_copied %d",
2229                                          bytes_copied);
2230                 }
2231                 if (!get_block_num(item, pos_in_item)) {
2232                         /* crap, we are writing to a hole */
2233                         use_get_block = 1;
2234                         goto out;
2235                 }
2236                 set_block_dev_mapped(bh_result,
2237                                      get_block_num(item, pos_in_item), inode);
2238         } else if (is_direct_le_ih(ih)) {
2239                 char *p;
2240                 p = page_address(bh_result->b_page);
2241                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2242                 copy_size = ih_item_len(ih) - pos_in_item;
2243
2244                 fs_gen = get_generation(inode->i_sb);
2245                 copy_item_head(&tmp_ih, ih);
2246
2247                 if (!trans_running) {
2248                         /* vs-3050 is gone, no need to drop the path */
2249                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2250                         if (retval)
2251                                 goto out;
2252                         reiserfs_update_inode_transaction(inode);
2253                         trans_running = 1;
2254                         if (fs_changed(fs_gen, inode->i_sb)
2255                             && item_moved(&tmp_ih, &path)) {
2256                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2257                                                                  bh);
2258                                 goto research;
2259                         }
2260                 }
2261
2262                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2263
2264                 if (fs_changed(fs_gen, inode->i_sb)
2265                     && item_moved(&tmp_ih, &path)) {
2266                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2267                         goto research;
2268                 }
2269
2270                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2271                        copy_size);
2272
2273                 journal_mark_dirty(&th, inode->i_sb, bh);
2274                 bytes_copied += copy_size;
2275                 set_block_dev_mapped(bh_result, 0, inode);
2276
2277                 /* are there still bytes left? */
2278                 if (bytes_copied < bh_result->b_size &&
2279                     (byte_offset + bytes_copied) < inode->i_size) {
2280                         set_cpu_key_k_offset(&key,
2281                                              cpu_key_k_offset(&key) +
2282                                              copy_size);
2283                         goto research;
2284                 }
2285         } else {
2286                 reiserfs_warning(inode->i_sb,
2287                                  "clm-6003: bad item inode %lu, device %s",
2288                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2289                 retval = -EIO;
2290                 goto out;
2291         }
2292         retval = 0;
2293
2294       out:
2295         pathrelse(&path);
2296         if (trans_running) {
2297                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2298                 if (err)
2299                         retval = err;
2300                 trans_running = 0;
2301         }
2302         reiserfs_write_unlock(inode->i_sb);
2303
2304         /* this is where we fill in holes in the file. */
2305         if (use_get_block) {
2306                 retval = reiserfs_get_block(inode, block, bh_result,
2307                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2308                                             | GET_BLOCK_NO_DANGLE);
2309                 if (!retval) {
2310                         if (!buffer_mapped(bh_result)
2311                             || bh_result->b_blocknr == 0) {
2312                                 /* get_block failed to find a mapped unformatted node. */
2313                                 use_get_block = 0;
2314                                 goto start_over;
2315                         }
2316                 }
2317         }
2318         kunmap(bh_result->b_page);
2319
2320         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2321                 /* we've copied data from the page into the direct item, so the
2322                  * buffer in the page is now clean, mark it to reflect that.
2323                  */
2324                 lock_buffer(bh_result);
2325                 clear_buffer_dirty(bh_result);
2326                 unlock_buffer(bh_result);
2327         }
2328         return retval;
2329 }
2330
2331 /*
2332  * mason@suse.com: updated in 2.5.54 to follow the same general io
2333  * start/recovery path as __block_write_full_page, along with special
2334  * code to handle reiserfs tails.
2335  */
2336 static int reiserfs_write_full_page(struct page *page,
2337                                     struct writeback_control *wbc)
2338 {
2339         struct inode *inode = page->mapping->host;
2340         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2341         int error = 0;
2342         unsigned long block;
2343         sector_t last_block;
2344         struct buffer_head *head, *bh;
2345         int partial = 0;
2346         int nr = 0;
2347         int checked = PageChecked(page);
2348         struct reiserfs_transaction_handle th;
2349         struct super_block *s = inode->i_sb;
2350         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2351         th.t_trans_id = 0;
2352
2353         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2354         if (checked && (current->flags & PF_MEMALLOC)) {
2355                 redirty_page_for_writepage(wbc, page);
2356                 unlock_page(page);
2357                 return 0;
2358         }
2359
2360         /* The page dirty bit is cleared before writepage is called, which
2361          * means we have to tell create_empty_buffers to make dirty buffers
2362          * The page really should be up to date at this point, so tossing
2363          * in the BH_Uptodate is just a sanity check.
2364          */
2365         if (!page_has_buffers(page)) {
2366                 create_empty_buffers(page, s->s_blocksize,
2367                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2368         }
2369         head = page_buffers(page);
2370
2371         /* last page in the file, zero out any contents past the
2372          ** last byte in the file
2373          */
2374         if (page->index >= end_index) {
2375                 char *kaddr;
2376                 unsigned last_offset;
2377
2378                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2379                 /* no file contents in this page */
2380                 if (page->index >= end_index + 1 || !last_offset) {
2381                         unlock_page(page);
2382                         return 0;
2383                 }
2384                 kaddr = kmap_atomic(page, KM_USER0);
2385                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2386                 flush_dcache_page(page);
2387                 kunmap_atomic(kaddr, KM_USER0);
2388         }
2389         bh = head;
2390         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2391         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2392         /* first map all the buffers, logging any direct items we find */
2393         do {
2394                 if (block > last_block) {
2395                         /*
2396                          * This can happen when the block size is less than
2397                          * the page size.  The corresponding bytes in the page
2398                          * were zero filled above
2399                          */
2400                         clear_buffer_dirty(bh);
2401                         set_buffer_uptodate(bh);
2402                 } else if ((checked || buffer_dirty(bh)) &&
2403                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2404                                                        && bh->b_blocknr ==
2405                                                        0))) {
2406                         /* not mapped yet, or it points to a direct item, search
2407                          * the btree for the mapping info, and log any direct
2408                          * items found
2409                          */
2410                         if ((error = map_block_for_writepage(inode, bh, block))) {
2411                                 goto fail;
2412                         }
2413                 }
2414                 bh = bh->b_this_page;
2415                 block++;
2416         } while (bh != head);
2417
2418         /*
2419          * we start the transaction after map_block_for_writepage,
2420          * because it can create holes in the file (an unbounded operation).
2421          * starting it here, we can make a reliable estimate for how many
2422          * blocks we're going to log
2423          */
2424         if (checked) {
2425                 ClearPageChecked(page);
2426                 reiserfs_write_lock(s);
2427                 error = journal_begin(&th, s, bh_per_page + 1);
2428                 if (error) {
2429                         reiserfs_write_unlock(s);
2430                         goto fail;
2431                 }
2432                 reiserfs_update_inode_transaction(inode);
2433         }
2434         /* now go through and lock any dirty buffers on the page */
2435         do {
2436                 get_bh(bh);
2437                 if (!buffer_mapped(bh))
2438                         continue;
2439                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2440                         continue;
2441
2442                 if (checked) {
2443                         reiserfs_prepare_for_journal(s, bh, 1);
2444                         journal_mark_dirty(&th, s, bh);
2445                         continue;
2446                 }
2447                 /* from this point on, we know the buffer is mapped to a
2448                  * real block and not a direct item
2449                  */
2450                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2451                         lock_buffer(bh);
2452                 } else {
2453                         if (test_set_buffer_locked(bh)) {
2454                                 redirty_page_for_writepage(wbc, page);
2455                                 continue;
2456                         }
2457                 }
2458                 if (test_clear_buffer_dirty(bh)) {
2459                         mark_buffer_async_write(bh);
2460                 } else {
2461                         unlock_buffer(bh);
2462                 }
2463         } while ((bh = bh->b_this_page) != head);
2464
2465         if (checked) {
2466                 error = journal_end(&th, s, bh_per_page + 1);
2467                 reiserfs_write_unlock(s);
2468                 if (error)
2469                         goto fail;
2470         }
2471         BUG_ON(PageWriteback(page));
2472         set_page_writeback(page);
2473         unlock_page(page);
2474
2475         /*
2476          * since any buffer might be the only dirty buffer on the page,
2477          * the first submit_bh can bring the page out of writeback.
2478          * be careful with the buffers.
2479          */
2480         do {
2481                 struct buffer_head *next = bh->b_this_page;
2482                 if (buffer_async_write(bh)) {
2483                         submit_bh(WRITE, bh);
2484                         nr++;
2485                 }
2486                 put_bh(bh);
2487                 bh = next;
2488         } while (bh != head);
2489
2490         error = 0;
2491       done:
2492         if (nr == 0) {
2493                 /*
2494                  * if this page only had a direct item, it is very possible for
2495                  * no io to be required without there being an error.  Or,
2496                  * someone else could have locked them and sent them down the
2497                  * pipe without locking the page
2498                  */
2499                 bh = head;
2500                 do {
2501                         if (!buffer_uptodate(bh)) {
2502                                 partial = 1;
2503                                 break;
2504                         }
2505                         bh = bh->b_this_page;
2506                 } while (bh != head);
2507                 if (!partial)
2508                         SetPageUptodate(page);
2509                 end_page_writeback(page);
2510         }
2511         return error;
2512
2513       fail:
2514         /* catches various errors, we need to make sure any valid dirty blocks
2515          * get to the media.  The page is currently locked and not marked for
2516          * writeback
2517          */
2518         ClearPageUptodate(page);
2519         bh = head;
2520         do {
2521                 get_bh(bh);
2522                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2523                         lock_buffer(bh);
2524                         mark_buffer_async_write(bh);
2525                 } else {
2526                         /*
2527                          * clear any dirty bits that might have come from getting
2528                          * attached to a dirty page
2529                          */
2530                         clear_buffer_dirty(bh);
2531                 }
2532                 bh = bh->b_this_page;
2533         } while (bh != head);
2534         SetPageError(page);
2535         BUG_ON(PageWriteback(page));
2536         set_page_writeback(page);
2537         unlock_page(page);
2538         do {
2539                 struct buffer_head *next = bh->b_this_page;
2540                 if (buffer_async_write(bh)) {
2541                         clear_buffer_dirty(bh);
2542                         submit_bh(WRITE, bh);
2543                         nr++;
2544                 }
2545                 put_bh(bh);
2546                 bh = next;
2547         } while (bh != head);
2548         goto done;
2549 }
2550
2551 static int reiserfs_readpage(struct file *f, struct page *page)
2552 {
2553         return block_read_full_page(page, reiserfs_get_block);
2554 }
2555
2556 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2557 {
2558         struct inode *inode = page->mapping->host;
2559         reiserfs_wait_on_write_block(inode->i_sb);
2560         return reiserfs_write_full_page(page, wbc);
2561 }
2562
2563 static int reiserfs_prepare_write(struct file *f, struct page *page,
2564                                   unsigned from, unsigned to)
2565 {
2566         struct inode *inode = page->mapping->host;
2567         int ret;
2568         int old_ref = 0;
2569
2570         reiserfs_wait_on_write_block(inode->i_sb);
2571         fix_tail_page_for_writing(page);
2572         if (reiserfs_transaction_running(inode->i_sb)) {
2573                 struct reiserfs_transaction_handle *th;
2574                 th = (struct reiserfs_transaction_handle *)current->
2575                     journal_info;
2576                 BUG_ON(!th->t_refcount);
2577                 BUG_ON(!th->t_trans_id);
2578                 old_ref = th->t_refcount;
2579                 th->t_refcount++;
2580         }
2581
2582         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2583         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2584                 struct reiserfs_transaction_handle *th = current->journal_info;
2585                 /* this gets a little ugly.  If reiserfs_get_block returned an
2586                  * error and left a transacstion running, we've got to close it,
2587                  * and we've got to free handle if it was a persistent transaction.
2588                  *
2589                  * But, if we had nested into an existing transaction, we need
2590                  * to just drop the ref count on the handle.
2591                  *
2592                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2593                  * and it was a persistent trans.  Otherwise, it was nested above.
2594                  */
2595                 if (th->t_refcount > old_ref) {
2596                         if (old_ref)
2597                                 th->t_refcount--;
2598                         else {
2599                                 int err;
2600                                 reiserfs_write_lock(inode->i_sb);
2601                                 err = reiserfs_end_persistent_transaction(th);
2602                                 reiserfs_write_unlock(inode->i_sb);
2603                                 if (err)
2604                                         ret = err;
2605                         }
2606                 }
2607         }
2608         return ret;
2609
2610 }
2611
2612 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2613 {
2614         return generic_block_bmap(as, block, reiserfs_bmap);
2615 }
2616
2617 static int reiserfs_commit_write(struct file *f, struct page *page,
2618                                  unsigned from, unsigned to)
2619 {
2620         struct inode *inode = page->mapping->host;
2621         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2622         int ret = 0;
2623         int update_sd = 0;
2624         struct reiserfs_transaction_handle *th = NULL;
2625
2626         reiserfs_wait_on_write_block(inode->i_sb);
2627         if (reiserfs_transaction_running(inode->i_sb)) {
2628                 th = current->journal_info;
2629         }
2630         reiserfs_commit_page(inode, page, from, to);
2631
2632         /* generic_commit_write does this for us, but does not update the
2633          ** transaction tracking stuff when the size changes.  So, we have
2634          ** to do the i_size updates here.
2635          */
2636         if (pos > inode->i_size) {
2637                 struct reiserfs_transaction_handle myth;
2638                 reiserfs_write_lock(inode->i_sb);
2639                 /* If the file have grown beyond the border where it
2640                    can have a tail, unmark it as needing a tail
2641                    packing */
2642                 if ((have_large_tails(inode->i_sb)
2643                      && inode->i_size > i_block_size(inode) * 4)
2644                     || (have_small_tails(inode->i_sb)
2645                         && inode->i_size > i_block_size(inode)))
2646                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2647
2648                 ret = journal_begin(&myth, inode->i_sb, 1);
2649                 if (ret) {
2650                         reiserfs_write_unlock(inode->i_sb);
2651                         goto journal_error;
2652                 }
2653                 reiserfs_update_inode_transaction(inode);
2654                 inode->i_size = pos;
2655                 /*
2656                  * this will just nest into our transaction.  It's important
2657                  * to use mark_inode_dirty so the inode gets pushed around on the
2658                  * dirty lists, and so that O_SYNC works as expected
2659                  */
2660                 mark_inode_dirty(inode);
2661                 reiserfs_update_sd(&myth, inode);
2662                 update_sd = 1;
2663                 ret = journal_end(&myth, inode->i_sb, 1);
2664                 reiserfs_write_unlock(inode->i_sb);
2665                 if (ret)
2666                         goto journal_error;
2667         }
2668         if (th) {
2669                 reiserfs_write_lock(inode->i_sb);
2670                 if (!update_sd)
2671                         mark_inode_dirty(inode);
2672                 ret = reiserfs_end_persistent_transaction(th);
2673                 reiserfs_write_unlock(inode->i_sb);
2674                 if (ret)
2675                         goto out;
2676         }
2677
2678       out:
2679         return ret;
2680
2681       journal_error:
2682         if (th) {
2683                 reiserfs_write_lock(inode->i_sb);
2684                 if (!update_sd)
2685                         reiserfs_update_sd(th, inode);
2686                 ret = reiserfs_end_persistent_transaction(th);
2687                 reiserfs_write_unlock(inode->i_sb);
2688         }
2689
2690         return ret;
2691 }
2692
2693 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2694 {
2695         if (reiserfs_attrs(inode->i_sb)) {
2696                 if (sd_attrs & REISERFS_SYNC_FL)
2697                         inode->i_flags |= S_SYNC;
2698                 else
2699                         inode->i_flags &= ~S_SYNC;
2700                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2701                         inode->i_flags |= S_IMMUTABLE;
2702                 else
2703                         inode->i_flags &= ~S_IMMUTABLE;
2704                 if (sd_attrs & REISERFS_APPEND_FL)
2705                         inode->i_flags |= S_APPEND;
2706                 else
2707                         inode->i_flags &= ~S_APPEND;
2708                 if (sd_attrs & REISERFS_NOATIME_FL)
2709                         inode->i_flags |= S_NOATIME;
2710                 else
2711                         inode->i_flags &= ~S_NOATIME;
2712                 if (sd_attrs & REISERFS_NOTAIL_FL)
2713                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2714                 else
2715                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2716         }
2717 }
2718
2719 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2720 {
2721         if (reiserfs_attrs(inode->i_sb)) {
2722                 if (inode->i_flags & S_IMMUTABLE)
2723                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2724                 else
2725                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2726                 if (inode->i_flags & S_SYNC)
2727                         *sd_attrs |= REISERFS_SYNC_FL;
2728                 else
2729                         *sd_attrs &= ~REISERFS_SYNC_FL;
2730                 if (inode->i_flags & S_NOATIME)
2731                         *sd_attrs |= REISERFS_NOATIME_FL;
2732                 else
2733                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2734                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2735                         *sd_attrs |= REISERFS_NOTAIL_FL;
2736                 else
2737                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2738         }
2739 }
2740
2741 /* decide if this buffer needs to stay around for data logging or ordered
2742 ** write purposes
2743 */
2744 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2745 {
2746         int ret = 1;
2747         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2748
2749         lock_buffer(bh);
2750         spin_lock(&j->j_dirty_buffers_lock);
2751         if (!buffer_mapped(bh)) {
2752                 goto free_jh;
2753         }
2754         /* the page is locked, and the only places that log a data buffer
2755          * also lock the page.
2756          */
2757         if (reiserfs_file_data_log(inode)) {
2758                 /*
2759                  * very conservative, leave the buffer pinned if
2760                  * anyone might need it.
2761                  */
2762                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2763                         ret = 0;
2764                 }
2765         } else  if (buffer_dirty(bh)) {
2766                 struct reiserfs_journal_list *jl;
2767                 struct reiserfs_jh *jh = bh->b_private;
2768
2769                 /* why is this safe?
2770                  * reiserfs_setattr updates i_size in the on disk
2771                  * stat data before allowing vmtruncate to be called.
2772                  *
2773                  * If buffer was put onto the ordered list for this
2774                  * transaction, we know for sure either this transaction
2775                  * or an older one already has updated i_size on disk,
2776                  * and this ordered data won't be referenced in the file
2777                  * if we crash.
2778                  *
2779                  * if the buffer was put onto the ordered list for an older
2780                  * transaction, we need to leave it around
2781                  */
2782                 if (jh && (jl = jh->jl)
2783                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2784                         ret = 0;
2785         }
2786       free_jh:
2787         if (ret && bh->b_private) {
2788                 reiserfs_free_jh(bh);
2789         }
2790         spin_unlock(&j->j_dirty_buffers_lock);
2791         unlock_buffer(bh);
2792         return ret;
2793 }
2794
2795 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2796 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2797 {
2798         struct buffer_head *head, *bh, *next;
2799         struct inode *inode = page->mapping->host;
2800         unsigned int curr_off = 0;
2801         int ret = 1;
2802
2803         BUG_ON(!PageLocked(page));
2804
2805         if (offset == 0)
2806                 ClearPageChecked(page);
2807
2808         if (!page_has_buffers(page))
2809                 goto out;
2810
2811         head = page_buffers(page);
2812         bh = head;
2813         do {
2814                 unsigned int next_off = curr_off + bh->b_size;
2815                 next = bh->b_this_page;
2816
2817                 /*
2818                  * is this block fully invalidated?
2819                  */
2820                 if (offset <= curr_off) {
2821                         if (invalidatepage_can_drop(inode, bh))
2822                                 reiserfs_unmap_buffer(bh);
2823                         else
2824                                 ret = 0;
2825                 }
2826                 curr_off = next_off;
2827                 bh = next;
2828         } while (bh != head);
2829
2830         /*
2831          * We release buffers only if the entire page is being invalidated.
2832          * The get_block cached value has been unconditionally invalidated,
2833          * so real IO is not possible anymore.
2834          */
2835         if (!offset && ret) {
2836                 ret = try_to_release_page(page, 0);
2837                 /* maybe should BUG_ON(!ret); - neilb */
2838         }
2839       out:
2840         return;
2841 }
2842
2843 static int reiserfs_set_page_dirty(struct page *page)
2844 {
2845         struct inode *inode = page->mapping->host;
2846         if (reiserfs_file_data_log(inode)) {
2847                 SetPageChecked(page);
2848                 return __set_page_dirty_nobuffers(page);
2849         }
2850         return __set_page_dirty_buffers(page);
2851 }
2852
2853 /*
2854  * Returns 1 if the page's buffers were dropped.  The page is locked.
2855  *
2856  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2857  * in the buffers at page_buffers(page).
2858  *
2859  * even in -o notail mode, we can't be sure an old mount without -o notail
2860  * didn't create files with tails.
2861  */
2862 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2863 {
2864         struct inode *inode = page->mapping->host;
2865         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2866         struct buffer_head *head;
2867         struct buffer_head *bh;
2868         int ret = 1;
2869
2870         WARN_ON(PageChecked(page));
2871         spin_lock(&j->j_dirty_buffers_lock);
2872         head = page_buffers(page);
2873         bh = head;
2874         do {
2875                 if (bh->b_private) {
2876                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2877                                 reiserfs_free_jh(bh);
2878                         } else {
2879                                 ret = 0;
2880                                 break;
2881                         }
2882                 }
2883                 bh = bh->b_this_page;
2884         } while (bh != head);
2885         if (ret)
2886                 ret = try_to_free_buffers(page);
2887         spin_unlock(&j->j_dirty_buffers_lock);
2888         return ret;
2889 }
2890
2891 /* We thank Mingming Cao for helping us understand in great detail what
2892    to do in this section of the code. */
2893 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2894                                   const struct iovec *iov, loff_t offset,
2895                                   unsigned long nr_segs)
2896 {
2897         struct file *file = iocb->ki_filp;
2898         struct inode *inode = file->f_mapping->host;
2899
2900         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2901                                   offset, nr_segs,
2902                                   reiserfs_get_blocks_direct_io, NULL);
2903 }
2904
2905 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2906 {
2907         struct inode *inode = dentry->d_inode;
2908         int error;
2909         unsigned int ia_valid = attr->ia_valid;
2910         reiserfs_write_lock(inode->i_sb);
2911         if (attr->ia_valid & ATTR_SIZE) {
2912                 /* version 2 items will be caught by the s_maxbytes check
2913                  ** done for us in vmtruncate
2914                  */
2915                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2916                     attr->ia_size > MAX_NON_LFS) {
2917                         error = -EFBIG;
2918                         goto out;
2919                 }
2920                 /* fill in hole pointers in the expanding truncate case. */
2921                 if (attr->ia_size > inode->i_size) {
2922                         error = generic_cont_expand(inode, attr->ia_size);
2923                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2924                                 int err;
2925                                 struct reiserfs_transaction_handle th;
2926                                 /* we're changing at most 2 bitmaps, inode + super */
2927                                 err = journal_begin(&th, inode->i_sb, 4);
2928                                 if (!err) {
2929                                         reiserfs_discard_prealloc(&th, inode);
2930                                         err = journal_end(&th, inode->i_sb, 4);
2931                                 }
2932                                 if (err)
2933                                         error = err;
2934                         }
2935                         if (error)
2936                                 goto out;
2937                         /*
2938                          * file size is changed, ctime and mtime are
2939                          * to be updated
2940                          */
2941                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
2942                 }
2943         }
2944
2945         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2946              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2947             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2948                 /* stat data of format v3.5 has 16 bit uid and gid */
2949                 error = -EINVAL;
2950                 goto out;
2951         }
2952
2953         error = inode_change_ok(inode, attr);
2954         if (!error) {
2955                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2956                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2957                         error = reiserfs_chown_xattrs(inode, attr);
2958
2959                         if (!error) {
2960                                 struct reiserfs_transaction_handle th;
2961                                 int jbegin_count =
2962                                     2 *
2963                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2964                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2965                                     2;
2966
2967                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2968                                 error =
2969                                     journal_begin(&th, inode->i_sb,
2970                                                   jbegin_count);
2971                                 if (error)
2972                                         goto out;
2973                                 error =
2974                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2975                                 if (error) {
2976                                         journal_end(&th, inode->i_sb,
2977                                                     jbegin_count);
2978                                         goto out;
2979                                 }
2980                                 /* Update corresponding info in inode so that everything is in
2981                                  * one transaction */
2982                                 if (attr->ia_valid & ATTR_UID)
2983                                         inode->i_uid = attr->ia_uid;
2984                                 if (attr->ia_valid & ATTR_GID)
2985                                         inode->i_gid = attr->ia_gid;
2986                                 mark_inode_dirty(inode);
2987                                 error =
2988                                     journal_end(&th, inode->i_sb, jbegin_count);
2989                         }
2990                 }
2991                 if (!error)
2992                         error = inode_setattr(inode, attr);
2993         }
2994
2995         if (!error && reiserfs_posixacl(inode->i_sb)) {
2996                 if (attr->ia_valid & ATTR_MODE)
2997                         error = reiserfs_acl_chmod(inode);
2998         }
2999
3000       out:
3001         reiserfs_write_unlock(inode->i_sb);
3002         return error;
3003 }
3004
3005 const struct address_space_operations reiserfs_address_space_operations = {
3006         .writepage = reiserfs_writepage,
3007         .readpage = reiserfs_readpage,
3008         .readpages = reiserfs_readpages,
3009         .releasepage = reiserfs_releasepage,
3010         .invalidatepage = reiserfs_invalidatepage,
3011         .sync_page = block_sync_page,
3012         .prepare_write = reiserfs_prepare_write,
3013         .commit_write = reiserfs_commit_write,
3014         .bmap = reiserfs_aop_bmap,
3015         .direct_IO = reiserfs_direct_IO,
3016         .set_page_dirty = reiserfs_set_page_dirty,
3017 };