fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/exportfs.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <linux/slab.h>
  15 #include <asm/uaccess.h>
  16 #include <asm/unaligned.h>
  17 #include <linux/buffer_head.h>
  18 #include <linux/mpage.h>
  19 #include <linux/writeback.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/swap.h>
  22
  23 int reiserfs_commit_write(struct file *f, struct page *page,
  24                           unsigned from, unsigned to);
  25 int reiserfs_prepare_write(struct file *f, struct page *page,
  26                            unsigned from, unsigned to);
  27
  28 void reiserfs_delete_inode(struct inode *inode)
  29 {
  30         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  31         int jbegin_count =
  32             JOURNAL_PER_BALANCE_CNT * 2 +
  33             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  34         struct reiserfs_transaction_handle th;
  35         int depth;
  36         int err;
  37
  38         if (!is_bad_inode(inode))
  39                 dquot_initialize(inode);
  40
  41         truncate_inode_pages(&inode->i_data, 0);
  42
  43         depth = reiserfs_write_lock_once(inode->i_sb);
  44
  45         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  46         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  47                 reiserfs_delete_xattrs(inode);
  48
  49                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  50                         goto out;
  51                 reiserfs_update_inode_transaction(inode);
  52
  53                 reiserfs_discard_prealloc(&th, inode);
  54
  55                 err = reiserfs_delete_object(&th, inode);
  56
  57                 /* Do quota update inside a transaction for journaled quotas. We must do that
  58                  * after delete_object so that quota updates go into the same transaction as
  59                  * stat data deletion */
  60                 if (!err)
  61                         dquot_free_inode(inode);
  62
  63                 if (journal_end(&th, inode->i_sb, jbegin_count))
  64                         goto out;
  65
  66                 /* check return value from reiserfs_delete_object after
  67                  * ending the transaction
  68                  */
  69                 if (err)
  70                     goto out;
  71
  72                 /* all items of file are deleted, so we can remove "save" link */
  73                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  74                                                                  * about an error here */
  75         } else {
  76                 /* no object items are in the tree */
  77                 ;
  78         }
  79       out:
  80         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  81         inode->i_blocks = 0;
  82         reiserfs_write_unlock_once(inode->i_sb, depth);
  83 }
  84
  85 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  86                           __u32 objectid, loff_t offset, int type, int length)
  87 {
  88         key->version = version;
  89
  90         key->on_disk_key.k_dir_id = dirid;
  91         key->on_disk_key.k_objectid = objectid;
  92         set_cpu_key_k_offset(key, offset);
  93         set_cpu_key_k_type(key, type);
  94         key->key_length = length;
  95 }
  96
  97 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  98    offset and type of key */
  99 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 100                   int type, int length)
 101 {
 102         _make_cpu_key(key, get_inode_item_key_version(inode),
 103                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 104                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 105                       length);
 106 }
 107
 108 //
 109 // when key is 0, do not set version and short key
 110 //
 111 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 112                               int version,
 113                               loff_t offset, int type, int length,
 114                               int entry_count /*or ih_free_space */ )
 115 {
 116         if (key) {
 117                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 118                 ih->ih_key.k_objectid =
 119                     cpu_to_le32(key->on_disk_key.k_objectid);
 120         }
 121         put_ih_version(ih, version);
 122         set_le_ih_k_offset(ih, offset);
 123         set_le_ih_k_type(ih, type);
 124         put_ih_item_len(ih, length);
 125         /*    set_ih_free_space (ih, 0); */
 126         // for directory items it is entry count, for directs and stat
 127         // datas - 0xffff, for indirects - 0
 128         put_ih_entry_count(ih, entry_count);
 129 }
 130
 131 //
 132 // FIXME: we might cache recently accessed indirect item
 133
 134 // Ugh.  Not too eager for that....
 135 //  I cut the code until such time as I see a convincing argument (benchmark).
 136 // I don't want a bloated inode struct..., and I don't like code complexity....
 137
 138 /* cutting the code is fine, since it really isn't in use yet and is easy
 139 ** to add back in.  But, Vladimir has a really good idea here.  Think
 140 ** about what happens for reading a file.  For each page,
 141 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 142 ** an indirect item.  This indirect item has X number of pointers, where
 143 ** X is a big number if we've done the block allocation right.  But,
 144 ** we only use one or two of these pointers during each call to readpage,
 145 ** needlessly researching again later on.
 146 **
 147 ** The size of the cache could be dynamic based on the size of the file.
 148 **
 149 ** I'd also like to see us cache the location the stat data item, since
 150 ** we are needlessly researching for that frequently.
 151 **
 152 ** --chris
 153 */
 154
 155 /* If this page has a file tail in it, and
 156 ** it was read in by get_block_create_0, the page data is valid,
 157 ** but tail is still sitting in a direct item, and we can't write to
 158 ** it.  So, look through this page, and check all the mapped buffers
 159 ** to make sure they have valid block numbers.  Any that don't need
 160 ** to be unmapped, so that block_prepare_write will correctly call
 161 ** reiserfs_get_block to convert the tail into an unformatted node
 162 */
 163 static inline void fix_tail_page_for_writing(struct page *page)
 164 {
 165         struct buffer_head *head, *next, *bh;
 166
 167         if (page && page_has_buffers(page)) {
 168                 head = page_buffers(page);
 169                 bh = head;
 170                 do {
 171                         next = bh->b_this_page;
 172                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 173                                 reiserfs_unmap_buffer(bh);
 174                         }
 175                         bh = next;
 176                 } while (bh != head);
 177         }
 178 }
 179
 180 /* reiserfs_get_block does not need to allocate a block only if it has been
 181    done already or non-hole position has been found in the indirect item */
 182 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 183                                     struct item_head *ih,
 184                                     __le32 * item, int pos_in_item)
 185 {
 186         if (allocated)
 187                 return 0;
 188         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 189             get_block_num(item, pos_in_item))
 190                 return 0;
 191         return 1;
 192 }
 193
 194 static inline int indirect_item_found(int retval, struct item_head *ih)
 195 {
 196         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 197 }
 198
 199 static inline void set_block_dev_mapped(struct buffer_head *bh,
 200                                         b_blocknr_t block, struct inode *inode)
 201 {
 202         map_bh(bh, inode->i_sb, block);
 203 }
 204
 205 //
 206 // files which were created in the earlier version can not be longer,
 207 // than 2 gb
 208 //
 209 static int file_capable(struct inode *inode, sector_t block)
 210 {
 211         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 212             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 213                 return 1;
 214
 215         return 0;
 216 }
 217
 218 static int restart_transaction(struct reiserfs_transaction_handle *th,
 219                                struct inode *inode, struct treepath *path)
 220 {
 221         struct super_block *s = th->t_super;
 222         int len = th->t_blocks_allocated;
 223         int err;
 224
 225         BUG_ON(!th->t_trans_id);
 226         BUG_ON(!th->t_refcount);
 227
 228         pathrelse(path);
 229
 230         /* we cannot restart while nested */
 231         if (th->t_refcount > 1) {
 232                 return 0;
 233         }
 234         reiserfs_update_sd(th, inode);
 235         err = journal_end(th, s, len);
 236         if (!err) {
 237                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 238                 if (!err)
 239                         reiserfs_update_inode_transaction(inode);
 240         }
 241         return err;
 242 }
 243
 244 // it is called by get_block when create == 0. Returns block number
 245 // for 'block'-th logical block of file. When it hits direct item it
 246 // returns 0 (being called from bmap) or read direct item into piece
 247 // of page (bh_result)
 248
 249 // Please improve the english/clarity in the comment above, as it is
 250 // hard to understand.
 251
 252 static int _get_block_create_0(struct inode *inode, sector_t block,
 253                                struct buffer_head *bh_result, int args)
 254 {
 255         INITIALIZE_PATH(path);
 256         struct cpu_key key;
 257         struct buffer_head *bh;
 258         struct item_head *ih, tmp_ih;
 259         b_blocknr_t blocknr;
 260         char *p = NULL;
 261         int chars;
 262         int ret;
 263         int result;
 264         int done = 0;
 265         unsigned long offset;
 266
 267         // prepare the key to look for the 'block'-th block of file
 268         make_cpu_key(&key, inode,
 269                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 270                      3);
 271
 272         result = search_for_position_by_key(inode->i_sb, &key, &path);
 273         if (result != POSITION_FOUND) {
 274                 pathrelse(&path);
 275                 if (p)
 276                         kunmap(bh_result->b_page);
 277                 if (result == IO_ERROR)
 278                         return -EIO;
 279                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 280                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 281                 if ((args & GET_BLOCK_NO_HOLE)
 282                     && !PageUptodate(bh_result->b_page)) {
 283                         return -ENOENT;
 284                 }
 285                 return 0;
 286         }
 287         //
 288         bh = get_last_bh(&path);
 289         ih = get_ih(&path);
 290         if (is_indirect_le_ih(ih)) {
 291                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 292
 293                 /* FIXME: here we could cache indirect item or part of it in
 294                    the inode to avoid search_by_key in case of subsequent
 295                    access to file */
 296                 blocknr = get_block_num(ind_item, path.pos_in_item);
 297                 ret = 0;
 298                 if (blocknr) {
 299                         map_bh(bh_result, inode->i_sb, blocknr);
 300                         if (path.pos_in_item ==
 301                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 302                                 set_buffer_boundary(bh_result);
 303                         }
 304                 } else
 305                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 306                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 307                 if ((args & GET_BLOCK_NO_HOLE)
 308                             && !PageUptodate(bh_result->b_page)) {
 309                         ret = -ENOENT;
 310                 }
 311
 312                 pathrelse(&path);
 313                 if (p)
 314                         kunmap(bh_result->b_page);
 315                 return ret;
 316         }
 317         // requested data are in direct item(s)
 318         if (!(args & GET_BLOCK_READ_DIRECT)) {
 319                 // we are called by bmap. FIXME: we can not map block of file
 320                 // when it is stored in direct item(s)
 321                 pathrelse(&path);
 322                 if (p)
 323                         kunmap(bh_result->b_page);
 324                 return -ENOENT;
 325         }
 326
 327         /* if we've got a direct item, and the buffer or page was uptodate,
 328          ** we don't want to pull data off disk again.  skip to the
 329          ** end, where we map the buffer and return
 330          */
 331         if (buffer_uptodate(bh_result)) {
 332                 goto finished;
 333         } else
 334                 /*
 335                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 336                  ** pages without any buffers.  If the page is up to date, we don't want
 337                  ** read old data off disk.  Set the up to date bit on the buffer instead
 338                  ** and jump to the end
 339                  */
 340         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 341                 set_buffer_uptodate(bh_result);
 342                 goto finished;
 343         }
 344         // read file tail into part of page
 345         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 346         copy_item_head(&tmp_ih, ih);
 347
 348         /* we only want to kmap if we are reading the tail into the page.
 349          ** this is not the common case, so we don't kmap until we are
 350          ** sure we need to.  But, this means the item might move if
 351          ** kmap schedules
 352          */
 353         if (!p)
 354                 p = (char *)kmap(bh_result->b_page);
 355
 356         p += offset;
 357         memset(p, 0, inode->i_sb->s_blocksize);
 358         do {
 359                 if (!is_direct_le_ih(ih)) {
 360                         BUG();
 361                 }
 362                 /* make sure we don't read more bytes than actually exist in
 363                  ** the file.  This can happen in odd cases where i_size isn't
 364                  ** correct, and when direct item padding results in a few
 365                  ** extra bytes at the end of the direct item
 366                  */
 367                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 368                         break;
 369                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 370                         chars =
 371                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 372                             path.pos_in_item;
 373                         done = 1;
 374                 } else {
 375                         chars = ih_item_len(ih) - path.pos_in_item;
 376                 }
 377                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 378
 379                 if (done)
 380                         break;
 381
 382                 p += chars;
 383
 384                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 385                         // we done, if read direct item is not the last item of
 386                         // node FIXME: we could try to check right delimiting key
 387                         // to see whether direct item continues in the right
 388                         // neighbor or rely on i_size
 389                         break;
 390
 391                 // update key to look for the next piece
 392                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 393                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 394                 if (result != POSITION_FOUND)
 395                         // i/o error most likely
 396                         break;
 397                 bh = get_last_bh(&path);
 398                 ih = get_ih(&path);
 399         } while (1);
 400
 401         flush_dcache_page(bh_result->b_page);
 402         kunmap(bh_result->b_page);
 403
 404       finished:
 405         pathrelse(&path);
 406
 407         if (result == IO_ERROR)
 408                 return -EIO;
 409
 410         /* this buffer has valid data, but isn't valid for io.  mapping it to
 411          * block #0 tells the rest of reiserfs it just has a tail in it
 412          */
 413         map_bh(bh_result, inode->i_sb, 0);
 414         set_buffer_uptodate(bh_result);
 415         return 0;
 416 }
 417
 418 // this is called to create file map. So, _get_block_create_0 will not
 419 // read direct item
 420 static int reiserfs_bmap(struct inode *inode, sector_t block,
 421                          struct buffer_head *bh_result, int create)
 422 {
 423         if (!file_capable(inode, block))
 424                 return -EFBIG;
 425
 426         reiserfs_write_lock(inode->i_sb);
 427         /* do not read the direct item */
 428         _get_block_create_0(inode, block, bh_result, 0);
 429         reiserfs_write_unlock(inode->i_sb);
 430         return 0;
 431 }
 432
 433 /* special version of get_block that is only used by grab_tail_page right
 434 ** now.  It is sent to block_prepare_write, and when you try to get a
 435 ** block past the end of the file (or a block from a hole) it returns
 436 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 437 ** be able to do i/o on the buffers returned, unless an error value
 438 ** is also returned.
 439 **
 440 ** So, this allows block_prepare_write to be used for reading a single block
 441 ** in a page.  Where it does not produce a valid page for holes, or past the
 442 ** end of the file.  This turns out to be exactly what we need for reading
 443 ** tails for conversion.
 444 **
 445 ** The point of the wrapper is forcing a certain value for create, even
 446 ** though the VFS layer is calling this function with create==1.  If you
 447 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 448 ** don't use this function.
 449 */
 450 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 451                                        struct buffer_head *bh_result,
 452                                        int create)
 453 {
 454         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 455 }
 456
 457 /* This is special helper for reiserfs_get_block in case we are executing
 458    direct_IO request. */
 459 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 460                                          sector_t iblock,
 461                                          struct buffer_head *bh_result,
 462                                          int create)
 463 {
 464         int ret;
 465
 466         bh_result->b_page = NULL;
 467
 468         /* We set the b_size before reiserfs_get_block call since it is
 469            referenced in convert_tail_for_hole() that may be called from
 470            reiserfs_get_block() */
 471         bh_result->b_size = (1 << inode->i_blkbits);
 472
 473         ret = reiserfs_get_block(inode, iblock, bh_result,
 474                                  create | GET_BLOCK_NO_DANGLE);
 475         if (ret)
 476                 goto out;
 477
 478         /* don't allow direct io onto tail pages */
 479         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 480                 /* make sure future calls to the direct io funcs for this offset
 481                  ** in the file fail by unmapping the buffer
 482                  */
 483                 clear_buffer_mapped(bh_result);
 484                 ret = -EINVAL;
 485         }
 486         /* Possible unpacked tail. Flush the data before pages have
 487            disappeared */
 488         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 489                 int err;
 490
 491                 reiserfs_write_lock(inode->i_sb);
 492
 493                 err = reiserfs_commit_for_inode(inode);
 494                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 495
 496                 reiserfs_write_unlock(inode->i_sb);
 497
 498                 if (err < 0)
 499                         ret = err;
 500         }
 501       out:
 502         return ret;
 503 }
 504
 505 /*
 506 ** helper function for when reiserfs_get_block is called for a hole
 507 ** but the file tail is still in a direct item
 508 ** bh_result is the buffer head for the hole
 509 ** tail_offset is the offset of the start of the tail in the file
 510 **
 511 ** This calls prepare_write, which will start a new transaction
 512 ** you should not be in a transaction, or have any paths held when you
 513 ** call this.
 514 */
 515 static int convert_tail_for_hole(struct inode *inode,
 516                                  struct buffer_head *bh_result,
 517                                  loff_t tail_offset)
 518 {
 519         unsigned long index;
 520         unsigned long tail_end;
 521         unsigned long tail_start;
 522         struct page *tail_page;
 523         struct page *hole_page = bh_result->b_page;
 524         int retval = 0;
 525
 526         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 527                 return -EIO;
 528
 529         /* always try to read until the end of the block */
 530         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 531         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 532
 533         index = tail_offset >> PAGE_CACHE_SHIFT;
 534         /* hole_page can be zero in case of direct_io, we are sure
 535            that we cannot get here if we write with O_DIRECT into
 536            tail page */
 537         if (!hole_page || index != hole_page->index) {
 538                 tail_page = grab_cache_page(inode->i_mapping, index);
 539                 retval = -ENOMEM;
 540                 if (!tail_page) {
 541                         goto out;
 542                 }
 543         } else {
 544                 tail_page = hole_page;
 545         }
 546
 547         /* we don't have to make sure the conversion did not happen while
 548          ** we were locking the page because anyone that could convert
 549          ** must first take i_mutex.
 550          **
 551          ** We must fix the tail page for writing because it might have buffers
 552          ** that are mapped, but have a block number of 0.  This indicates tail
 553          ** data that has been read directly into the page, and block_prepare_write
 554          ** won't trigger a get_block in this case.
 555          */
 556         fix_tail_page_for_writing(tail_page);
 557         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 558         if (retval)
 559                 goto unlock;
 560
 561         /* tail conversion might change the data in the page */
 562         flush_dcache_page(tail_page);
 563
 564         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 565
 566       unlock:
 567         if (tail_page != hole_page) {
 568                 unlock_page(tail_page);
 569                 page_cache_release(tail_page);
 570         }
 571       out:
 572         return retval;
 573 }
 574
 575 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 576                                   sector_t block,
 577                                   struct inode *inode,
 578                                   b_blocknr_t * allocated_block_nr,
 579                                   struct treepath *path, int flags)
 580 {
 581         BUG_ON(!th->t_trans_id);
 582
 583 #ifdef REISERFS_PREALLOCATE
 584         if (!(flags & GET_BLOCK_NO_IMUX)) {
 585                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 586                                                   path, block);
 587         }
 588 #endif
 589         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 590                                          block);
 591 }
 592
 593 int reiserfs_get_block(struct inode *inode, sector_t block,
 594                        struct buffer_head *bh_result, int create)
 595 {
 596         int repeat, retval = 0;
 597         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 598         INITIALIZE_PATH(path);
 599         int pos_in_item;
 600         struct cpu_key key;
 601         struct buffer_head *bh, *unbh = NULL;
 602         struct item_head *ih, tmp_ih;
 603         __le32 *item;
 604         int done;
 605         int fs_gen;
 606         int lock_depth;
 607         struct reiserfs_transaction_handle *th = NULL;
 608         /* space reserved in transaction batch:
 609            . 3 balancings in direct->indirect conversion
 610            . 1 block involved into reiserfs_update_sd()
 611            XXX in practically impossible worst case direct2indirect()
 612            can incur (much) more than 3 balancings.
 613            quota update for user, group */
 614         int jbegin_count =
 615             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 616             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 617         int version;
 618         int dangle = 1;
 619         loff_t new_offset =
 620             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 621
 622         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 623         version = get_inode_item_key_version(inode);
 624
 625         if (!file_capable(inode, block)) {
 626                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 627                 return -EFBIG;
 628         }
 629
 630         /* if !create, we aren't changing the FS, so we don't need to
 631          ** log anything, so we don't need to start a transaction
 632          */
 633         if (!(create & GET_BLOCK_CREATE)) {
 634                 int ret;
 635                 /* find number of block-th logical block of the file */
 636                 ret = _get_block_create_0(inode, block, bh_result,
 637                                           create | GET_BLOCK_READ_DIRECT);
 638                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 639                 return ret;
 640         }
 641         /*
 642          * if we're already in a transaction, make sure to close
 643          * any new transactions we start in this func
 644          */
 645         if ((create & GET_BLOCK_NO_DANGLE) ||
 646             reiserfs_transaction_running(inode->i_sb))
 647                 dangle = 0;
 648
 649         /* If file is of such a size, that it might have a tail and tails are enabled
 650          ** we should mark it as possibly needing tail packing on close
 651          */
 652         if ((have_large_tails(inode->i_sb)
 653              && inode->i_size < i_block_size(inode) * 4)
 654             || (have_small_tails(inode->i_sb)
 655                 && inode->i_size < i_block_size(inode)))
 656                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 657
 658         /* set the key of the first byte in the 'block'-th block of file */
 659         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 660         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 661               start_trans:
 662                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 663                 if (!th) {
 664                         retval = -ENOMEM;
 665                         goto failure;
 666                 }
 667                 reiserfs_update_inode_transaction(inode);
 668         }
 669       research:
 670
 671         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 672         if (retval == IO_ERROR) {
 673                 retval = -EIO;
 674                 goto failure;
 675         }
 676
 677         bh = get_last_bh(&path);
 678         ih = get_ih(&path);
 679         item = get_item(&path);
 680         pos_in_item = path.pos_in_item;
 681
 682         fs_gen = get_generation(inode->i_sb);
 683         copy_item_head(&tmp_ih, ih);
 684
 685         if (allocation_needed
 686             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 687                 /* we have to allocate block for the unformatted node */
 688                 if (!th) {
 689                         pathrelse(&path);
 690                         goto start_trans;
 691                 }
 692
 693                 repeat =
 694                     _allocate_block(th, block, inode, &allocated_block_nr,
 695                                     &path, create);
 696
 697                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 698                         /* restart the transaction to give the journal a chance to free
 699                          ** some blocks.  releases the path, so we have to go back to
 700                          ** research if we succeed on the second try
 701                          */
 702                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 703                         retval = restart_transaction(th, inode, &path);
 704                         if (retval)
 705                                 goto failure;
 706                         repeat =
 707                             _allocate_block(th, block, inode,
 708                                             &allocated_block_nr, NULL, create);
 709
 710                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 711                                 goto research;
 712                         }
 713                         if (repeat == QUOTA_EXCEEDED)
 714                                 retval = -EDQUOT;
 715                         else
 716                                 retval = -ENOSPC;
 717                         goto failure;
 718                 }
 719
 720                 if (fs_changed(fs_gen, inode->i_sb)
 721                     && item_moved(&tmp_ih, &path)) {
 722                         goto research;
 723                 }
 724         }
 725
 726         if (indirect_item_found(retval, ih)) {
 727                 b_blocknr_t unfm_ptr;
 728                 /* 'block'-th block is in the file already (there is
 729                    corresponding cell in some indirect item). But it may be
 730                    zero unformatted node pointer (hole) */
 731                 unfm_ptr = get_block_num(item, pos_in_item);
 732                 if (unfm_ptr == 0) {
 733                         /* use allocated block to plug the hole */
 734                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 735                         if (fs_changed(fs_gen, inode->i_sb)
 736                             && item_moved(&tmp_ih, &path)) {
 737                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 738                                                                  bh);
 739                                 goto research;
 740                         }
 741                         set_buffer_new(bh_result);
 742                         if (buffer_dirty(bh_result)
 743                             && reiserfs_data_ordered(inode->i_sb))
 744                                 reiserfs_add_ordered_list(inode, bh_result);
 745                         put_block_num(item, pos_in_item, allocated_block_nr);
 746                         unfm_ptr = allocated_block_nr;
 747                         journal_mark_dirty(th, inode->i_sb, bh);
 748                         reiserfs_update_sd(th, inode);
 749                 }
 750                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 751                 pathrelse(&path);
 752                 retval = 0;
 753                 if (!dangle && th)
 754                         retval = reiserfs_end_persistent_transaction(th);
 755
 756                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 757
 758                 /* the item was found, so new blocks were not added to the file
 759                  ** there is no need to make sure the inode is updated with this
 760                  ** transaction
 761                  */
 762                 return retval;
 763         }
 764
 765         if (!th) {
 766                 pathrelse(&path);
 767                 goto start_trans;
 768         }
 769
 770         /* desired position is not found or is in the direct item. We have
 771            to append file with holes up to 'block'-th block converting
 772            direct items to indirect one if necessary */
 773         done = 0;
 774         do {
 775                 if (is_statdata_le_ih(ih)) {
 776                         __le32 unp = 0;
 777                         struct cpu_key tmp_key;
 778
 779                         /* indirect item has to be inserted */
 780                         make_le_item_head(&tmp_ih, &key, version, 1,
 781                                           TYPE_INDIRECT, UNFM_P_SIZE,
 782                                           0 /* free_space */ );
 783
 784                         if (cpu_key_k_offset(&key) == 1) {
 785                                 /* we are going to add 'block'-th block to the file. Use
 786                                    allocated block for that */
 787                                 unp = cpu_to_le32(allocated_block_nr);
 788                                 set_block_dev_mapped(bh_result,
 789                                                      allocated_block_nr, inode);
 790                                 set_buffer_new(bh_result);
 791                                 done = 1;
 792                         }
 793                         tmp_key = key;  // ;)
 794                         set_cpu_key_k_offset(&tmp_key, 1);
 795                         PATH_LAST_POSITION(&path)++;
 796
 797                         retval =
 798                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 799                                                  inode, (char *)&unp);
 800                         if (retval) {
 801                                 reiserfs_free_block(th, inode,
 802                                                     allocated_block_nr, 1);
 803                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 804                         }
 805                         //mark_tail_converted (inode);
 806                 } else if (is_direct_le_ih(ih)) {
 807                         /* direct item has to be converted */
 808                         loff_t tail_offset;
 809
 810                         tail_offset =
 811                             ((le_ih_k_offset(ih) -
 812                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 813                         if (tail_offset == cpu_key_k_offset(&key)) {
 814                                 /* direct item we just found fits into block we have
 815                                    to map. Convert it into unformatted node: use
 816                                    bh_result for the conversion */
 817                                 set_block_dev_mapped(bh_result,
 818                                                      allocated_block_nr, inode);
 819                                 unbh = bh_result;
 820                                 done = 1;
 821                         } else {
 822                                 /* we have to padd file tail stored in direct item(s)
 823                                    up to block size and convert it to unformatted
 824                                    node. FIXME: this should also get into page cache */
 825
 826                                 pathrelse(&path);
 827                                 /*
 828                                  * ugly, but we can only end the transaction if
 829                                  * we aren't nested
 830                                  */
 831                                 BUG_ON(!th->t_refcount);
 832                                 if (th->t_refcount == 1) {
 833                                         retval =
 834                                             reiserfs_end_persistent_transaction
 835                                             (th);
 836                                         th = NULL;
 837                                         if (retval)
 838                                                 goto failure;
 839                                 }
 840
 841                                 retval =
 842                                     convert_tail_for_hole(inode, bh_result,
 843                                                           tail_offset);
 844                                 if (retval) {
 845                                         if (retval != -ENOSPC)
 846                                                 reiserfs_error(inode->i_sb,
 847                                                         "clm-6004",
 848                                                         "convert tail failed "
 849                                                         "inode %lu, error %d",
 850                                                         inode->i_ino,
 851                                                         retval);
 852                                         if (allocated_block_nr) {
 853                                                 /* the bitmap, the super, and the stat data == 3 */
 854                                                 if (!th)
 855                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 856                                                 if (th)
 857                                                         reiserfs_free_block(th,
 858                                                                             inode,
 859                                                                             allocated_block_nr,
 860                                                                             1);
 861                                         }
 862                                         goto failure;
 863                                 }
 864                                 goto research;
 865                         }
 866                         retval =
 867                             direct2indirect(th, inode, &path, unbh,
 868                                             tail_offset);
 869                         if (retval) {
 870                                 reiserfs_unmap_buffer(unbh);
 871                                 reiserfs_free_block(th, inode,
 872                                                     allocated_block_nr, 1);
 873                                 goto failure;
 874                         }
 875                         /* it is important the set_buffer_uptodate is done after
 876                          ** the direct2indirect.  The buffer might contain valid
 877                          ** data newer than the data on disk (read by readpage, changed,
 878                          ** and then sent here by writepage).  direct2indirect needs
 879                          ** to know if unbh was already up to date, so it can decide
 880                          ** if the data in unbh needs to be replaced with data from
 881                          ** the disk
 882                          */
 883                         set_buffer_uptodate(unbh);
 884
 885                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 886                            buffer will disappear shortly, so it should not be added to
 887                          */
 888                         if (unbh->b_page) {
 889                                 /* we've converted the tail, so we must
 890                                  ** flush unbh before the transaction commits
 891                                  */
 892                                 reiserfs_add_tail_list(inode, unbh);
 893
 894                                 /* mark it dirty now to prevent commit_write from adding
 895                                  ** this buffer to the inode's dirty buffer list
 896                                  */
 897                                 /*
 898                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 899                                  * It's still atomic, but it sets the page dirty too,
 900                                  * which makes it eligible for writeback at any time by the
 901                                  * VM (which was also the case with __mark_buffer_dirty())
 902                                  */
 903                                 mark_buffer_dirty(unbh);
 904                         }
 905                 } else {
 906                         /* append indirect item with holes if needed, when appending
 907                            pointer to 'block'-th block use block, which is already
 908                            allocated */
 909                         struct cpu_key tmp_key;
 910                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 911                         // one block which is a fastpath
 912                         unp_t *un;
 913                         __u64 max_to_insert =
 914                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 915                             UNFM_P_SIZE;
 916                         __u64 blocks_needed;
 917
 918                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 919                                "vs-804: invalid position for append");
 920                         /* indirect item has to be appended, set up key of that position */
 921                         make_cpu_key(&tmp_key, inode,
 922                                      le_key_k_offset(version,
 923                                                      &(ih->ih_key)) +
 924                                      op_bytes_number(ih,
 925                                                      inode->i_sb->s_blocksize),
 926                                      //pos_in_item * inode->i_sb->s_blocksize,
 927                                      TYPE_INDIRECT, 3); // key type is unimportant
 928
 929                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 930                                "green-805: invalid offset");
 931                         blocks_needed =
 932                             1 +
 933                             ((cpu_key_k_offset(&key) -
 934                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 935                              s_blocksize_bits);
 936
 937                         if (blocks_needed == 1) {
 938                                 un = &unf_single;
 939                         } else {
 940                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 941                                 if (!un) {
 942                                         un = &unf_single;
 943                                         blocks_needed = 1;
 944                                         max_to_insert = 0;
 945                                 }
 946                         }
 947                         if (blocks_needed <= max_to_insert) {
 948                                 /* we are going to add target block to the file. Use allocated
 949                                    block for that */
 950                                 un[blocks_needed - 1] =
 951                                     cpu_to_le32(allocated_block_nr);
 952                                 set_block_dev_mapped(bh_result,
 953                                                      allocated_block_nr, inode);
 954                                 set_buffer_new(bh_result);
 955                                 done = 1;
 956                         } else {
 957                                 /* paste hole to the indirect item */
 958                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 959                                    only have space for one block */
 960                                 blocks_needed =
 961                                     max_to_insert ? max_to_insert : 1;
 962                         }
 963                         retval =
 964                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 965                                                      (char *)un,
 966                                                      UNFM_P_SIZE *
 967                                                      blocks_needed);
 968
 969                         if (blocks_needed != 1)
 970                                 kfree(un);
 971
 972                         if (retval) {
 973                                 reiserfs_free_block(th, inode,
 974                                                     allocated_block_nr, 1);
 975                                 goto failure;
 976                         }
 977                         if (!done) {
 978                                 /* We need to mark new file size in case this function will be
 979                                    interrupted/aborted later on. And we may do this only for
 980                                    holes. */
 981                                 inode->i_size +=
 982                                     inode->i_sb->s_blocksize * blocks_needed;
 983                         }
 984                 }
 985
 986                 if (done == 1)
 987                         break;
 988
 989                 /* this loop could log more blocks than we had originally asked
 990                  ** for.  So, we have to allow the transaction to end if it is
 991                  ** too big or too full.  Update the inode so things are
 992                  ** consistent if we crash before the function returns
 993                  **
 994                  ** release the path so that anybody waiting on the path before
 995                  ** ending their transaction will be able to continue.
 996                  */
 997                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 998                         retval = restart_transaction(th, inode, &path);
 999                         if (retval)
1000                                 goto failure;
1001                 }
1002                 /*
1003                  * inserting indirect pointers for a hole can take a
1004                  * long time.  reschedule if needed and also release the write
1005                  * lock for others.
1006                  */
1007                 if (need_resched()) {
1008                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1009                         schedule();
1010                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1011                 }
1012
1013                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1014                 if (retval == IO_ERROR) {
1015                         retval = -EIO;
1016                         goto failure;
1017                 }
1018                 if (retval == POSITION_FOUND) {
1019                         reiserfs_warning(inode->i_sb, "vs-825",
1020                                          "%K should not be found", &key);
1021                         retval = -EEXIST;
1022                         if (allocated_block_nr)
1023                                 reiserfs_free_block(th, inode,
1024                                                     allocated_block_nr, 1);
1025                         pathrelse(&path);
1026                         goto failure;
1027                 }
1028                 bh = get_last_bh(&path);
1029                 ih = get_ih(&path);
1030                 item = get_item(&path);
1031                 pos_in_item = path.pos_in_item;
1032         } while (1);
1033
1034         retval = 0;
1035
1036       failure:
1037         if (th && (!dangle || (retval && !th->t_trans_id))) {
1038                 int err;
1039                 if (th->t_trans_id)
1040                         reiserfs_update_sd(th, inode);
1041                 err = reiserfs_end_persistent_transaction(th);
1042                 if (err)
1043                         retval = err;
1044         }
1045
1046         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1047         reiserfs_check_path(&path);
1048         return retval;
1049 }
1050
1051 static int
1052 reiserfs_readpages(struct file *file, struct address_space *mapping,
1053                    struct list_head *pages, unsigned nr_pages)
1054 {
1055         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1056 }
1057
1058 /* Compute real number of used bytes by file
1059  * Following three functions can go away when we'll have enough space in stat item
1060  */
1061 static int real_space_diff(struct inode *inode, int sd_size)
1062 {
1063         int bytes;
1064         loff_t blocksize = inode->i_sb->s_blocksize;
1065
1066         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1067                 return sd_size;
1068
1069         /* End of file is also in full block with indirect reference, so round
1070          ** up to the next block.
1071          **
1072          ** there is just no way to know if the tail is actually packed
1073          ** on the file, so we have to assume it isn't.  When we pack the
1074          ** tail, we add 4 bytes to pretend there really is an unformatted
1075          ** node pointer
1076          */
1077         bytes =
1078             ((inode->i_size +
1079               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1080             sd_size;
1081         return bytes;
1082 }
1083
1084 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1085                                         int sd_size)
1086 {
1087         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1088                 return inode->i_size +
1089                     (loff_t) (real_space_diff(inode, sd_size));
1090         }
1091         return ((loff_t) real_space_diff(inode, sd_size)) +
1092             (((loff_t) blocks) << 9);
1093 }
1094
1095 /* Compute number of blocks used by file in ReiserFS counting */
1096 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1097 {
1098         loff_t bytes = inode_get_bytes(inode);
1099         loff_t real_space = real_space_diff(inode, sd_size);
1100
1101         /* keeps fsck and non-quota versions of reiserfs happy */
1102         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1103                 bytes += (loff_t) 511;
1104         }
1105
1106         /* files from before the quota patch might i_blocks such that
1107          ** bytes < real_space.  Deal with that here to prevent it from
1108          ** going negative.
1109          */
1110         if (bytes < real_space)
1111                 return 0;
1112         return (bytes - real_space) >> 9;
1113 }
1114
1115 //
1116 // BAD: new directories have stat data of new type and all other items
1117 // of old type. Version stored in the inode says about body items, so
1118 // in update_stat_data we can not rely on inode, but have to check
1119 // item version directly
1120 //
1121
1122 // called by read_locked_inode
1123 static void init_inode(struct inode *inode, struct treepath *path)
1124 {
1125         struct buffer_head *bh;
1126         struct item_head *ih;
1127         __u32 rdev;
1128         //int version = ITEM_VERSION_1;
1129
1130         bh = PATH_PLAST_BUFFER(path);
1131         ih = PATH_PITEM_HEAD(path);
1132
1133         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1134
1135         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1136         REISERFS_I(inode)->i_flags = 0;
1137         REISERFS_I(inode)->i_prealloc_block = 0;
1138         REISERFS_I(inode)->i_prealloc_count = 0;
1139         REISERFS_I(inode)->i_trans_id = 0;
1140         REISERFS_I(inode)->i_jl = NULL;
1141         mutex_init(&(REISERFS_I(inode)->i_mmap));
1142         reiserfs_init_xattr_rwsem(inode);
1143
1144         if (stat_data_v1(ih)) {
1145                 struct stat_data_v1 *sd =
1146                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1147                 unsigned long blocks;
1148
1149                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1150                 set_inode_sd_version(inode, STAT_DATA_V1);
1151                 inode->i_mode = sd_v1_mode(sd);
1152                 inode->i_nlink = sd_v1_nlink(sd);
1153                 inode->i_uid = sd_v1_uid(sd);
1154                 inode->i_gid = sd_v1_gid(sd);
1155                 inode->i_size = sd_v1_size(sd);
1156                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1157                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1158                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1159                 inode->i_atime.tv_nsec = 0;
1160                 inode->i_ctime.tv_nsec = 0;
1161                 inode->i_mtime.tv_nsec = 0;
1162
1163                 inode->i_blocks = sd_v1_blocks(sd);
1164                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1165                 blocks = (inode->i_size + 511) >> 9;
1166                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1167                 if (inode->i_blocks > blocks) {
1168                         // there was a bug in <=3.5.23 when i_blocks could take negative
1169                         // values. Starting from 3.5.17 this value could even be stored in
1170                         // stat data. For such files we set i_blocks based on file
1171                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1172                         // only updated if file's inode will ever change
1173                         inode->i_blocks = blocks;
1174                 }
1175
1176                 rdev = sd_v1_rdev(sd);
1177                 REISERFS_I(inode)->i_first_direct_byte =
1178                     sd_v1_first_direct_byte(sd);
1179                 /* an early bug in the quota code can give us an odd number for the
1180                  ** block count.  This is incorrect, fix it here.
1181                  */
1182                 if (inode->i_blocks & 1) {
1183                         inode->i_blocks++;
1184                 }
1185                 inode_set_bytes(inode,
1186                                 to_real_used_space(inode, inode->i_blocks,
1187                                                    SD_V1_SIZE));
1188                 /* nopack is initially zero for v1 objects. For v2 objects,
1189                    nopack is initialised from sd_attrs */
1190                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1191         } else {
1192                 // new stat data found, but object may have old items
1193                 // (directories and symlinks)
1194                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1195
1196                 inode->i_mode = sd_v2_mode(sd);
1197                 inode->i_nlink = sd_v2_nlink(sd);
1198                 inode->i_uid = sd_v2_uid(sd);
1199                 inode->i_size = sd_v2_size(sd);
1200                 inode->i_gid = sd_v2_gid(sd);
1201                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1202                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1203                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1204                 inode->i_ctime.tv_nsec = 0;
1205                 inode->i_mtime.tv_nsec = 0;
1206                 inode->i_atime.tv_nsec = 0;
1207                 inode->i_blocks = sd_v2_blocks(sd);
1208                 rdev = sd_v2_rdev(sd);
1209                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1210                         inode->i_generation =
1211                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1212                 else
1213                         inode->i_generation = sd_v2_generation(sd);
1214
1215                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1216                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1217                 else
1218                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1219                 REISERFS_I(inode)->i_first_direct_byte = 0;
1220                 set_inode_sd_version(inode, STAT_DATA_V2);
1221                 inode_set_bytes(inode,
1222                                 to_real_used_space(inode, inode->i_blocks,
1223                                                    SD_V2_SIZE));
1224                 /* read persistent inode attributes from sd and initialise
1225                    generic inode flags from them */
1226                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1227                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1228         }
1229
1230         pathrelse(path);
1231         if (S_ISREG(inode->i_mode)) {
1232                 inode->i_op = &reiserfs_file_inode_operations;
1233                 inode->i_fop = &reiserfs_file_operations;
1234                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1235         } else if (S_ISDIR(inode->i_mode)) {
1236                 inode->i_op = &reiserfs_dir_inode_operations;
1237                 inode->i_fop = &reiserfs_dir_operations;
1238         } else if (S_ISLNK(inode->i_mode)) {
1239                 inode->i_op = &reiserfs_symlink_inode_operations;
1240                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1241         } else {
1242                 inode->i_blocks = 0;
1243                 inode->i_op = &reiserfs_special_inode_operations;
1244                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1245         }
1246 }
1247
1248 // update new stat data with inode fields
1249 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1250 {
1251         struct stat_data *sd_v2 = (struct stat_data *)sd;
1252         __u16 flags;
1253
1254         set_sd_v2_mode(sd_v2, inode->i_mode);
1255         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1256         set_sd_v2_uid(sd_v2, inode->i_uid);
1257         set_sd_v2_size(sd_v2, size);
1258         set_sd_v2_gid(sd_v2, inode->i_gid);
1259         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1260         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1261         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1262         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1263         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1264                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1265         else
1266                 set_sd_v2_generation(sd_v2, inode->i_generation);
1267         flags = REISERFS_I(inode)->i_attrs;
1268         i_attrs_to_sd_attrs(inode, &flags);
1269         set_sd_v2_attrs(sd_v2, flags);
1270 }
1271
1272 // used to copy inode's fields to old stat data
1273 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1274 {
1275         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1276
1277         set_sd_v1_mode(sd_v1, inode->i_mode);
1278         set_sd_v1_uid(sd_v1, inode->i_uid);
1279         set_sd_v1_gid(sd_v1, inode->i_gid);
1280         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1281         set_sd_v1_size(sd_v1, size);
1282         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1283         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1284         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1285
1286         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1287                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1288         else
1289                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1290
1291         // Sigh. i_first_direct_byte is back
1292         set_sd_v1_first_direct_byte(sd_v1,
1293                                     REISERFS_I(inode)->i_first_direct_byte);
1294 }
1295
1296 /* NOTE, you must prepare the buffer head before sending it here,
1297 ** and then log it after the call
1298 */
1299 static void update_stat_data(struct treepath *path, struct inode *inode,
1300                              loff_t size)
1301 {
1302         struct buffer_head *bh;
1303         struct item_head *ih;
1304
1305         bh = PATH_PLAST_BUFFER(path);
1306         ih = PATH_PITEM_HEAD(path);
1307
1308         if (!is_statdata_le_ih(ih))
1309                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1310                                INODE_PKEY(inode), ih);
1311
1312         if (stat_data_v1(ih)) {
1313                 // path points to old stat data
1314                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1315         } else {
1316                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1317         }
1318
1319         return;
1320 }
1321
1322 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1323                              struct inode *inode, loff_t size)
1324 {
1325         struct cpu_key key;
1326         INITIALIZE_PATH(path);
1327         struct buffer_head *bh;
1328         int fs_gen;
1329         struct item_head *ih, tmp_ih;
1330         int retval;
1331
1332         BUG_ON(!th->t_trans_id);
1333
1334         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1335
1336         for (;;) {
1337                 int pos;
1338                 /* look for the object's stat data */
1339                 retval = search_item(inode->i_sb, &key, &path);
1340                 if (retval == IO_ERROR) {
1341                         reiserfs_error(inode->i_sb, "vs-13050",
1342                                        "i/o failure occurred trying to "
1343                                        "update %K stat data", &key);
1344                         return;
1345                 }
1346                 if (retval == ITEM_NOT_FOUND) {
1347                         pos = PATH_LAST_POSITION(&path);
1348                         pathrelse(&path);
1349                         if (inode->i_nlink == 0) {
1350                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1351                                 return;
1352                         }
1353                         reiserfs_warning(inode->i_sb, "vs-13060",
1354                                          "stat data of object %k (nlink == %d) "
1355                                          "not found (pos %d)",
1356                                          INODE_PKEY(inode), inode->i_nlink,
1357                                          pos);
1358                         reiserfs_check_path(&path);
1359                         return;
1360                 }
1361
1362                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1363                  ** FS might change.  We have to detect that, and loop back to the
1364                  ** search if the stat data item has moved
1365                  */
1366                 bh = get_last_bh(&path);
1367                 ih = get_ih(&path);
1368                 copy_item_head(&tmp_ih, ih);
1369                 fs_gen = get_generation(inode->i_sb);
1370                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1371                 if (fs_changed(fs_gen, inode->i_sb)
1372                     && item_moved(&tmp_ih, &path)) {
1373                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1374                         continue;       /* Stat_data item has been moved after scheduling. */
1375                 }
1376                 break;
1377         }
1378         update_stat_data(&path, inode, size);
1379         journal_mark_dirty(th, th->t_super, bh);
1380         pathrelse(&path);
1381         return;
1382 }
1383
1384 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1385 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1386 ** and clear the key in the private portion of the inode, otherwise a
1387 ** corresponding iput might try to delete whatever object the inode last
1388 ** represented.
1389 */
1390 static void reiserfs_make_bad_inode(struct inode *inode)
1391 {
1392         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1393         make_bad_inode(inode);
1394 }
1395
1396 //
1397 // initially this function was derived from minix or ext2's analog and
1398 // evolved as the prototype did
1399 //
1400
1401 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1402 {
1403         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1404         inode->i_ino = args->objectid;
1405         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1406         return 0;
1407 }
1408
1409 /* looks for stat data in the tree, and fills up the fields of in-core
1410    inode stat data fields */
1411 void reiserfs_read_locked_inode(struct inode *inode,
1412                                 struct reiserfs_iget_args *args)
1413 {
1414         INITIALIZE_PATH(path_to_sd);
1415         struct cpu_key key;
1416         unsigned long dirino;
1417         int retval;
1418
1419         dirino = args->dirid;
1420
1421         /* set version 1, version 2 could be used too, because stat data
1422            key is the same in both versions */
1423         key.version = KEY_FORMAT_3_5;
1424         key.on_disk_key.k_dir_id = dirino;
1425         key.on_disk_key.k_objectid = inode->i_ino;
1426         key.on_disk_key.k_offset = 0;
1427         key.on_disk_key.k_type = 0;
1428
1429         /* look for the object's stat data */
1430         retval = search_item(inode->i_sb, &key, &path_to_sd);
1431         if (retval == IO_ERROR) {
1432                 reiserfs_error(inode->i_sb, "vs-13070",
1433                                "i/o failure occurred trying to find "
1434                                "stat data of %K", &key);
1435                 reiserfs_make_bad_inode(inode);
1436                 return;
1437         }
1438         if (retval != ITEM_FOUND) {
1439                 /* a stale NFS handle can trigger this without it being an error */
1440                 pathrelse(&path_to_sd);
1441                 reiserfs_make_bad_inode(inode);
1442                 inode->i_nlink = 0;
1443                 return;
1444         }
1445
1446         init_inode(inode, &path_to_sd);
1447
1448         /* It is possible that knfsd is trying to access inode of a file
1449            that is being removed from the disk by some other thread. As we
1450            update sd on unlink all that is required is to check for nlink
1451            here. This bug was first found by Sizif when debugging
1452            SquidNG/Butterfly, forgotten, and found again after Philippe
1453            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1454
1455            More logical fix would require changes in fs/inode.c:iput() to
1456            remove inode from hash-table _after_ fs cleaned disk stuff up and
1457            in iget() to return NULL if I_FREEING inode is found in
1458            hash-table. */
1459         /* Currently there is one place where it's ok to meet inode with
1460            nlink==0: processing of open-unlinked and half-truncated files
1461            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1462         if ((inode->i_nlink == 0) &&
1463             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1464                 reiserfs_warning(inode->i_sb, "vs-13075",
1465                                  "dead inode read from disk %K. "
1466                                  "This is likely to be race with knfsd. Ignore",
1467                                  &key);
1468                 reiserfs_make_bad_inode(inode);
1469         }
1470
1471         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1472
1473 }
1474
1475 /**
1476  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1477  *
1478  * @inode:    inode from hash table to check
1479  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1480  *
1481  * This function is called by iget5_locked() to distinguish reiserfs inodes
1482  * having the same inode numbers. Such inodes can only exist due to some
1483  * error condition. One of them should be bad. Inodes with identical
1484  * inode numbers (objectids) are distinguished by parent directory ids.
1485  *
1486  */
1487 int reiserfs_find_actor(struct inode *inode, void *opaque)
1488 {
1489         struct reiserfs_iget_args *args;
1490
1491         args = opaque;
1492         /* args is already in CPU order */
1493         return (inode->i_ino == args->objectid) &&
1494             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1495 }
1496
1497 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1498 {
1499         struct inode *inode;
1500         struct reiserfs_iget_args args;
1501
1502         args.objectid = key->on_disk_key.k_objectid;
1503         args.dirid = key->on_disk_key.k_dir_id;
1504         reiserfs_write_unlock(s);
1505         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1506                              reiserfs_find_actor, reiserfs_init_locked_inode,
1507                              (void *)(&args));
1508         reiserfs_write_lock(s);
1509         if (!inode)
1510                 return ERR_PTR(-ENOMEM);
1511
1512         if (inode->i_state & I_NEW) {
1513                 reiserfs_read_locked_inode(inode, &args);
1514                 unlock_new_inode(inode);
1515         }
1516
1517         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1518                 /* either due to i/o error or a stale NFS handle */
1519                 iput(inode);
1520                 inode = NULL;
1521         }
1522         return inode;
1523 }
1524
1525 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1526         u32 objectid, u32 dir_id, u32 generation)
1527
1528 {
1529         struct cpu_key key;
1530         struct inode *inode;
1531
1532         key.on_disk_key.k_objectid = objectid;
1533         key.on_disk_key.k_dir_id = dir_id;
1534         reiserfs_write_lock(sb);
1535         inode = reiserfs_iget(sb, &key);
1536         if (inode && !IS_ERR(inode) && generation != 0 &&
1537             generation != inode->i_generation) {
1538                 iput(inode);
1539                 inode = NULL;
1540         }
1541         reiserfs_write_unlock(sb);
1542
1543         return d_obtain_alias(inode);
1544 }
1545
1546 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1547                 int fh_len, int fh_type)
1548 {
1549         /* fhtype happens to reflect the number of u32s encoded.
1550          * due to a bug in earlier code, fhtype might indicate there
1551          * are more u32s then actually fitted.
1552          * so if fhtype seems to be more than len, reduce fhtype.
1553          * Valid types are:
1554          *   2 - objectid + dir_id - legacy support
1555          *   3 - objectid + dir_id + generation
1556          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1557          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1558          *   6 - as above plus generation of directory
1559          * 6 does not fit in NFSv2 handles
1560          */
1561         if (fh_type > fh_len) {
1562                 if (fh_type != 6 || fh_len != 5)
1563                         reiserfs_warning(sb, "reiserfs-13077",
1564                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1565                                 fh_type, fh_len);
1566                 fh_type = 5;
1567         }
1568
1569         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1570                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1571 }
1572
1573 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1574                 int fh_len, int fh_type)
1575 {
1576         if (fh_type < 4)
1577                 return NULL;
1578
1579         return reiserfs_get_dentry(sb,
1580                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1581                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1582                 (fh_type == 6) ? fid->raw[5] : 0);
1583 }
1584
1585 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1586                        int need_parent)
1587 {
1588         struct inode *inode = dentry->d_inode;
1589         int maxlen = *lenp;
1590
1591         if (maxlen < 3)
1592                 return 255;
1593
1594         data[0] = inode->i_ino;
1595         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1596         data[2] = inode->i_generation;
1597         *lenp = 3;
1598         /* no room for directory info? return what we've stored so far */
1599         if (maxlen < 5 || !need_parent)
1600                 return 3;
1601
1602         spin_lock(&dentry->d_lock);
1603         inode = dentry->d_parent->d_inode;
1604         data[3] = inode->i_ino;
1605         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1606         *lenp = 5;
1607         if (maxlen >= 6) {
1608                 data[5] = inode->i_generation;
1609                 *lenp = 6;
1610         }
1611         spin_unlock(&dentry->d_lock);
1612         return *lenp;
1613 }
1614
1615 /* looks for stat data, then copies fields to it, marks the buffer
1616    containing stat data as dirty */
1617 /* reiserfs inodes are never really dirty, since the dirty inode call
1618 ** always logs them.  This call allows the VFS inode marking routines
1619 ** to properly mark inodes for datasync and such, but only actually
1620 ** does something when called for a synchronous update.
1621 */
1622 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1623 {
1624         struct reiserfs_transaction_handle th;
1625         int jbegin_count = 1;
1626
1627         if (inode->i_sb->s_flags & MS_RDONLY)
1628                 return -EROFS;
1629         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1630          ** these cases are just when the system needs ram, not when the
1631          ** inode needs to reach disk for safety, and they can safely be
1632          ** ignored because the altered inode has already been logged.
1633          */
1634         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1635                 reiserfs_write_lock(inode->i_sb);
1636                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1637                         reiserfs_update_sd(&th, inode);
1638                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1639                 }
1640                 reiserfs_write_unlock(inode->i_sb);
1641         }
1642         return 0;
1643 }
1644
1645 /* stat data of new object is inserted already, this inserts the item
1646    containing "." and ".." entries */
1647 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1648                                   struct inode *inode,
1649                                   struct item_head *ih, struct treepath *path,
1650                                   struct inode *dir)
1651 {
1652         struct super_block *sb = th->t_super;
1653         char empty_dir[EMPTY_DIR_SIZE];
1654         char *body = empty_dir;
1655         struct cpu_key key;
1656         int retval;
1657
1658         BUG_ON(!th->t_trans_id);
1659
1660         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1661                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1662                       TYPE_DIRENTRY, 3 /*key length */ );
1663
1664         /* compose item head for new item. Directories consist of items of
1665            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1666            is done by reiserfs_new_inode */
1667         if (old_format_only(sb)) {
1668                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1669                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1670
1671                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1672                                        ih->ih_key.k_objectid,
1673                                        INODE_PKEY(dir)->k_dir_id,
1674                                        INODE_PKEY(dir)->k_objectid);
1675         } else {
1676                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1677                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1678
1679                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1680                                     ih->ih_key.k_objectid,
1681                                     INODE_PKEY(dir)->k_dir_id,
1682                                     INODE_PKEY(dir)->k_objectid);
1683         }
1684
1685         /* look for place in the tree for new item */
1686         retval = search_item(sb, &key, path);
1687         if (retval == IO_ERROR) {
1688                 reiserfs_error(sb, "vs-13080",
1689                                "i/o failure occurred creating new directory");
1690                 return -EIO;
1691         }
1692         if (retval == ITEM_FOUND) {
1693                 pathrelse(path);
1694                 reiserfs_warning(sb, "vs-13070",
1695                                  "object with this key exists (%k)",
1696                                  &(ih->ih_key));
1697                 return -EEXIST;
1698         }
1699
1700         /* insert item, that is empty directory item */
1701         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1702 }
1703
1704 /* stat data of object has been inserted, this inserts the item
1705    containing the body of symlink */
1706 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1707                                 struct item_head *ih,
1708                                 struct treepath *path, const char *symname,
1709                                 int item_len)
1710 {
1711         struct super_block *sb = th->t_super;
1712         struct cpu_key key;
1713         int retval;
1714
1715         BUG_ON(!th->t_trans_id);
1716
1717         _make_cpu_key(&key, KEY_FORMAT_3_5,
1718                       le32_to_cpu(ih->ih_key.k_dir_id),
1719                       le32_to_cpu(ih->ih_key.k_objectid),
1720                       1, TYPE_DIRECT, 3 /*key length */ );
1721
1722         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1723                           0 /*free_space */ );
1724
1725         /* look for place in the tree for new item */
1726         retval = search_item(sb, &key, path);
1727         if (retval == IO_ERROR) {
1728                 reiserfs_error(sb, "vs-13080",
1729                                "i/o failure occurred creating new symlink");
1730                 return -EIO;
1731         }
1732         if (retval == ITEM_FOUND) {
1733                 pathrelse(path);
1734                 reiserfs_warning(sb, "vs-13080",
1735                                  "object with this key exists (%k)",
1736                                  &(ih->ih_key));
1737                 return -EEXIST;
1738         }
1739
1740         /* insert item, that is body of symlink */
1741         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1742 }
1743
1744 /* inserts the stat data into the tree, and then calls
1745    reiserfs_new_directory (to insert ".", ".." item if new object is
1746    directory) or reiserfs_new_symlink (to insert symlink body if new
1747    object is symlink) or nothing (if new object is regular file)
1748
1749    NOTE! uid and gid must already be set in the inode.  If we return
1750    non-zero due to an error, we have to drop the quota previously allocated
1751    for the fresh inode.  This can only be done outside a transaction, so
1752    if we return non-zero, we also end the transaction.  */
1753 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1754                        struct inode *dir, int mode, const char *symname,
1755                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1756                           strlen (symname) for symlinks) */
1757                        loff_t i_size, struct dentry *dentry,
1758                        struct inode *inode,
1759                        struct reiserfs_security_handle *security)
1760 {
1761         struct super_block *sb;
1762         struct reiserfs_iget_args args;
1763         INITIALIZE_PATH(path_to_key);
1764         struct cpu_key key;
1765         struct item_head ih;
1766         struct stat_data sd;
1767         int retval;
1768         int err;
1769
1770         BUG_ON(!th->t_trans_id);
1771
1772         dquot_initialize(inode);
1773         err = dquot_alloc_inode(inode);
1774         if (err)
1775                 goto out_end_trans;
1776         if (!dir->i_nlink) {
1777                 err = -EPERM;
1778                 goto out_bad_inode;
1779         }
1780
1781         sb = dir->i_sb;
1782
1783         /* item head of new item */
1784         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1785         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1786         if (!ih.ih_key.k_objectid) {
1787                 err = -ENOMEM;
1788                 goto out_bad_inode;
1789         }
1790         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1791         if (old_format_only(sb))
1792                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1793                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1794         else
1795                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1796                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1797         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1798         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1799         if (insert_inode_locked4(inode, args.objectid,
1800                              reiserfs_find_actor, &args) < 0) {
1801                 err = -EINVAL;
1802                 goto out_bad_inode;
1803         }
1804         if (old_format_only(sb))
1805                 /* not a perfect generation count, as object ids can be reused, but
1806                  ** this is as good as reiserfs can do right now.
1807                  ** note that the private part of inode isn't filled in yet, we have
1808                  ** to use the directory.
1809                  */
1810                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1811         else
1812 #if defined( USE_INODE_GENERATION_COUNTER )
1813                 inode->i_generation =
1814                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1815 #else
1816                 inode->i_generation = ++event;
1817 #endif
1818
1819         /* fill stat data */
1820         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1821
1822         /* uid and gid must already be set by the caller for quota init */
1823
1824         /* symlink cannot be immutable or append only, right? */
1825         if (S_ISLNK(inode->i_mode))
1826                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1827
1828         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1829         inode->i_size = i_size;
1830         inode->i_blocks = 0;
1831         inode->i_bytes = 0;
1832         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1833             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1834
1835         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1836         REISERFS_I(inode)->i_flags = 0;
1837         REISERFS_I(inode)->i_prealloc_block = 0;
1838         REISERFS_I(inode)->i_prealloc_count = 0;
1839         REISERFS_I(inode)->i_trans_id = 0;
1840         REISERFS_I(inode)->i_jl = NULL;
1841         REISERFS_I(inode)->i_attrs =
1842             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1843         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1844         mutex_init(&(REISERFS_I(inode)->i_mmap));
1845         reiserfs_init_xattr_rwsem(inode);
1846
1847         /* key to search for correct place for new stat data */
1848         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1849                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1850                       TYPE_STAT_DATA, 3 /*key length */ );
1851
1852         /* find proper place for inserting of stat data */
1853         retval = search_item(sb, &key, &path_to_key);
1854         if (retval == IO_ERROR) {
1855                 err = -EIO;
1856                 goto out_bad_inode;
1857         }
1858         if (retval == ITEM_FOUND) {
1859                 pathrelse(&path_to_key);
1860                 err = -EEXIST;
1861                 goto out_bad_inode;
1862         }
1863         if (old_format_only(sb)) {
1864                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1865                         pathrelse(&path_to_key);
1866                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1867                         err = -EINVAL;
1868                         goto out_bad_inode;
1869                 }
1870                 inode2sd_v1(&sd, inode, inode->i_size);
1871         } else {
1872                 inode2sd(&sd, inode, inode->i_size);
1873         }
1874         // store in in-core inode the key of stat data and version all
1875         // object items will have (directory items will have old offset
1876         // format, other new objects will consist of new items)
1877         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1878                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1879         else
1880                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1881         if (old_format_only(sb))
1882                 set_inode_sd_version(inode, STAT_DATA_V1);
1883         else
1884                 set_inode_sd_version(inode, STAT_DATA_V2);
1885
1886         /* insert the stat data into the tree */
1887 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1888         if (REISERFS_I(dir)->new_packing_locality)
1889                 th->displace_new_blocks = 1;
1890 #endif
1891         retval =
1892             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1893                                  (char *)(&sd));
1894         if (retval) {
1895                 err = retval;
1896                 reiserfs_check_path(&path_to_key);
1897                 goto out_bad_inode;
1898         }
1899 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1900         if (!th->displace_new_blocks)
1901                 REISERFS_I(dir)->new_packing_locality = 0;
1902 #endif
1903         if (S_ISDIR(mode)) {
1904                 /* insert item with "." and ".." */
1905                 retval =
1906                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1907         }
1908
1909         if (S_ISLNK(mode)) {
1910                 /* insert body of symlink */
1911                 if (!old_format_only(sb))
1912                         i_size = ROUND_UP(i_size);
1913                 retval =
1914                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1915                                          i_size);
1916         }
1917         if (retval) {
1918                 err = retval;
1919                 reiserfs_check_path(&path_to_key);
1920                 journal_end(th, th->t_super, th->t_blocks_allocated);
1921                 goto out_inserted_sd;
1922         }
1923
1924         if (reiserfs_posixacl(inode->i_sb)) {
1925                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
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         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1933                 reiserfs_warning(inode->i_sb, "jdm-13090",
1934                                  "ACLs aren't enabled in the fs, "
1935                                  "but vfs thinks they are!");
1936         } else if (IS_PRIVATE(dir))
1937                 inode->i_flags |= S_PRIVATE;
1938
1939         if (security->name) {
1940                 retval = reiserfs_security_write(th, inode, security);
1941                 if (retval) {
1942                         err = retval;
1943                         reiserfs_check_path(&path_to_key);
1944                         retval = journal_end(th, th->t_super,
1945                                              th->t_blocks_allocated);
1946                         if (retval)
1947                                 err = retval;
1948                         goto out_inserted_sd;
1949                 }
1950         }
1951
1952         reiserfs_update_sd(th, inode);
1953         reiserfs_check_path(&path_to_key);
1954
1955         return 0;
1956
1957 /* it looks like you can easily compress these two goto targets into
1958  * one.  Keeping it like this doesn't actually hurt anything, and they
1959  * are place holders for what the quota code actually needs.
1960  */
1961       out_bad_inode:
1962         /* Invalidate the object, nothing was inserted yet */
1963         INODE_PKEY(inode)->k_objectid = 0;
1964
1965         /* Quota change must be inside a transaction for journaling */
1966         dquot_free_inode(inode);
1967
1968       out_end_trans:
1969         journal_end(th, th->t_super, th->t_blocks_allocated);
1970         /* Drop can be outside and it needs more credits so it's better to have it outside */
1971         dquot_drop(inode);
1972         inode->i_flags |= S_NOQUOTA;
1973         make_bad_inode(inode);
1974
1975       out_inserted_sd:
1976         inode->i_nlink = 0;
1977         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1978         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1979         iput(inode);
1980         return err;
1981 }
1982
1983 /*
1984 ** finds the tail page in the page cache,
1985 ** reads the last block in.
1986 **
1987 ** On success, page_result is set to a locked, pinned page, and bh_result
1988 ** is set to an up to date buffer for the last block in the file.  returns 0.
1989 **
1990 ** tail conversion is not done, so bh_result might not be valid for writing
1991 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1992 ** trying to write the block.
1993 **
1994 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1995 */
1996 static int grab_tail_page(struct inode *inode,
1997                           struct page **page_result,
1998                           struct buffer_head **bh_result)
1999 {
2000
2001         /* we want the page with the last byte in the file,
2002          ** not the page that will hold the next byte for appending
2003          */
2004         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2005         unsigned long pos = 0;
2006         unsigned long start = 0;
2007         unsigned long blocksize = inode->i_sb->s_blocksize;
2008         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2009         struct buffer_head *bh;
2010         struct buffer_head *head;
2011         struct page *page;
2012         int error;
2013
2014         /* we know that we are only called with inode->i_size > 0.
2015          ** we also know that a file tail can never be as big as a block
2016          ** If i_size % blocksize == 0, our file is currently block aligned
2017          ** and it won't need converting or zeroing after a truncate.
2018          */
2019         if ((offset & (blocksize - 1)) == 0) {
2020                 return -ENOENT;
2021         }
2022         page = grab_cache_page(inode->i_mapping, index);
2023         error = -ENOMEM;
2024         if (!page) {
2025                 goto out;
2026         }
2027         /* start within the page of the last block in the file */
2028         start = (offset / blocksize) * blocksize;
2029
2030         error = block_prepare_write(page, start, offset,
2031                                     reiserfs_get_block_create_0);
2032         if (error)
2033                 goto unlock;
2034
2035         head = page_buffers(page);
2036         bh = head;
2037         do {
2038                 if (pos >= start) {
2039                         break;
2040                 }
2041                 bh = bh->b_this_page;
2042                 pos += blocksize;
2043         } while (bh != head);
2044
2045         if (!buffer_uptodate(bh)) {
2046                 /* note, this should never happen, prepare_write should
2047                  ** be taking care of this for us.  If the buffer isn't up to date,
2048                  ** I've screwed up the code to find the buffer, or the code to
2049                  ** call prepare_write
2050                  */
2051                 reiserfs_error(inode->i_sb, "clm-6000",
2052                                "error reading block %lu", bh->b_blocknr);
2053                 error = -EIO;
2054                 goto unlock;
2055         }
2056         *bh_result = bh;
2057         *page_result = page;
2058
2059       out:
2060         return error;
2061
2062       unlock:
2063         unlock_page(page);
2064         page_cache_release(page);
2065         return error;
2066 }
2067
2068 /*
2069 ** vfs version of truncate file.  Must NOT be called with
2070 ** a transaction already started.
2071 **
2072 ** some code taken from block_truncate_page
2073 */
2074 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2075 {
2076         struct reiserfs_transaction_handle th;
2077         /* we want the offset for the first byte after the end of the file */
2078         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2079         unsigned blocksize = inode->i_sb->s_blocksize;
2080         unsigned length;
2081         struct page *page = NULL;
2082         int error;
2083         struct buffer_head *bh = NULL;
2084         int err2;
2085         int lock_depth;
2086
2087         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2088
2089         if (inode->i_size > 0) {
2090                 error = grab_tail_page(inode, &page, &bh);
2091                 if (error) {
2092                         // -ENOENT means we truncated past the end of the file,
2093                         // and get_block_create_0 could not find a block to read in,
2094                         // which is ok.
2095                         if (error != -ENOENT)
2096                                 reiserfs_error(inode->i_sb, "clm-6001",
2097                                                "grab_tail_page failed %d",
2098                                                error);
2099                         page = NULL;
2100                         bh = NULL;
2101                 }
2102         }
2103
2104         /* so, if page != NULL, we have a buffer head for the offset at
2105          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2106          ** then we have an unformatted node.  Otherwise, we have a direct item,
2107          ** and no zeroing is required on disk.  We zero after the truncate,
2108          ** because the truncate might pack the item anyway
2109          ** (it will unmap bh if it packs).
2110          */
2111         /* it is enough to reserve space in transaction for 2 balancings:
2112            one for "save" link adding and another for the first
2113            cut_from_item. 1 is for update_sd */
2114         error = journal_begin(&th, inode->i_sb,
2115                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2116         if (error)
2117                 goto out;
2118         reiserfs_update_inode_transaction(inode);
2119         if (update_timestamps)
2120                 /* we are doing real truncate: if the system crashes before the last
2121                    transaction of truncating gets committed - on reboot the file
2122                    either appears truncated properly or not truncated at all */
2123                 add_save_link(&th, inode, 1);
2124         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2125         error =
2126             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2127         if (error)
2128                 goto out;
2129
2130         /* check reiserfs_do_truncate after ending the transaction */
2131         if (err2) {
2132                 error = err2;
2133                 goto out;
2134         }
2135
2136         if (update_timestamps) {
2137                 error = remove_save_link(inode, 1 /* truncate */);
2138                 if (error)
2139                         goto out;
2140         }
2141
2142         if (page) {
2143                 length = offset & (blocksize - 1);
2144                 /* if we are not on a block boundary */
2145                 if (length) {
2146                         length = blocksize - length;
2147                         zero_user(page, offset, length);
2148                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2149                                 mark_buffer_dirty(bh);
2150                         }
2151                 }
2152                 unlock_page(page);
2153                 page_cache_release(page);
2154         }
2155
2156         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2157
2158         return 0;
2159       out:
2160         if (page) {
2161                 unlock_page(page);
2162                 page_cache_release(page);
2163         }
2164
2165         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2166
2167         return error;
2168 }
2169
2170 static int map_block_for_writepage(struct inode *inode,
2171                                    struct buffer_head *bh_result,
2172                                    unsigned long block)
2173 {
2174         struct reiserfs_transaction_handle th;
2175         int fs_gen;
2176         struct item_head tmp_ih;
2177         struct item_head *ih;
2178         struct buffer_head *bh;
2179         __le32 *item;
2180         struct cpu_key key;
2181         INITIALIZE_PATH(path);
2182         int pos_in_item;
2183         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2184         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2185         int retval;
2186         int use_get_block = 0;
2187         int bytes_copied = 0;
2188         int copy_size;
2189         int trans_running = 0;
2190
2191         /* catch places below that try to log something without starting a trans */
2192         th.t_trans_id = 0;
2193
2194         if (!buffer_uptodate(bh_result)) {
2195                 return -EIO;
2196         }
2197
2198         kmap(bh_result->b_page);
2199       start_over:
2200         reiserfs_write_lock(inode->i_sb);
2201         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2202
2203       research:
2204         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2205         if (retval != POSITION_FOUND) {
2206                 use_get_block = 1;
2207                 goto out;
2208         }
2209
2210         bh = get_last_bh(&path);
2211         ih = get_ih(&path);
2212         item = get_item(&path);
2213         pos_in_item = path.pos_in_item;
2214
2215         /* we've found an unformatted node */
2216         if (indirect_item_found(retval, ih)) {
2217                 if (bytes_copied > 0) {
2218                         reiserfs_warning(inode->i_sb, "clm-6002",
2219                                          "bytes_copied %d", bytes_copied);
2220                 }
2221                 if (!get_block_num(item, pos_in_item)) {
2222                         /* crap, we are writing to a hole */
2223                         use_get_block = 1;
2224                         goto out;
2225                 }
2226                 set_block_dev_mapped(bh_result,
2227                                      get_block_num(item, pos_in_item), inode);
2228         } else if (is_direct_le_ih(ih)) {
2229                 char *p;
2230                 p = page_address(bh_result->b_page);
2231                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2232                 copy_size = ih_item_len(ih) - pos_in_item;
2233
2234                 fs_gen = get_generation(inode->i_sb);
2235                 copy_item_head(&tmp_ih, ih);
2236
2237                 if (!trans_running) {
2238                         /* vs-3050 is gone, no need to drop the path */
2239                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2240                         if (retval)
2241                                 goto out;
2242                         reiserfs_update_inode_transaction(inode);
2243                         trans_running = 1;
2244                         if (fs_changed(fs_gen, inode->i_sb)
2245                             && item_moved(&tmp_ih, &path)) {
2246                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2247                                                                  bh);
2248                                 goto research;
2249                         }
2250                 }
2251
2252                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2253
2254                 if (fs_changed(fs_gen, inode->i_sb)
2255                     && item_moved(&tmp_ih, &path)) {
2256                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2257                         goto research;
2258                 }
2259
2260                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2261                        copy_size);
2262
2263                 journal_mark_dirty(&th, inode->i_sb, bh);
2264                 bytes_copied += copy_size;
2265                 set_block_dev_mapped(bh_result, 0, inode);
2266
2267                 /* are there still bytes left? */
2268                 if (bytes_copied < bh_result->b_size &&
2269                     (byte_offset + bytes_copied) < inode->i_size) {
2270                         set_cpu_key_k_offset(&key,
2271                                              cpu_key_k_offset(&key) +
2272                                              copy_size);
2273                         goto research;
2274                 }
2275         } else {
2276                 reiserfs_warning(inode->i_sb, "clm-6003",
2277                                  "bad item inode %lu", inode->i_ino);
2278                 retval = -EIO;
2279                 goto out;
2280         }
2281         retval = 0;
2282
2283       out:
2284         pathrelse(&path);
2285         if (trans_running) {
2286                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2287                 if (err)
2288                         retval = err;
2289                 trans_running = 0;
2290         }
2291         reiserfs_write_unlock(inode->i_sb);
2292
2293         /* this is where we fill in holes in the file. */
2294         if (use_get_block) {
2295                 retval = reiserfs_get_block(inode, block, bh_result,
2296                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2297                                             | GET_BLOCK_NO_DANGLE);
2298                 if (!retval) {
2299                         if (!buffer_mapped(bh_result)
2300                             || bh_result->b_blocknr == 0) {
2301                                 /* get_block failed to find a mapped unformatted node. */
2302                                 use_get_block = 0;
2303                                 goto start_over;
2304                         }
2305                 }
2306         }
2307         kunmap(bh_result->b_page);
2308
2309         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2310                 /* we've copied data from the page into the direct item, so the
2311                  * buffer in the page is now clean, mark it to reflect that.
2312                  */
2313                 lock_buffer(bh_result);
2314                 clear_buffer_dirty(bh_result);
2315                 unlock_buffer(bh_result);
2316         }
2317         return retval;
2318 }
2319
2320 /*
2321  * mason@suse.com: updated in 2.5.54 to follow the same general io
2322  * start/recovery path as __block_write_full_page, along with special
2323  * code to handle reiserfs tails.
2324  */
2325 static int reiserfs_write_full_page(struct page *page,
2326                                     struct writeback_control *wbc)
2327 {
2328         struct inode *inode = page->mapping->host;
2329         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2330         int error = 0;
2331         unsigned long block;
2332         sector_t last_block;
2333         struct buffer_head *head, *bh;
2334         int partial = 0;
2335         int nr = 0;
2336         int checked = PageChecked(page);
2337         struct reiserfs_transaction_handle th;
2338         struct super_block *s = inode->i_sb;
2339         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2340         th.t_trans_id = 0;
2341
2342         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2343         if (checked && (current->flags & PF_MEMALLOC)) {
2344                 redirty_page_for_writepage(wbc, page);
2345                 unlock_page(page);
2346                 return 0;
2347         }
2348
2349         /* The page dirty bit is cleared before writepage is called, which
2350          * means we have to tell create_empty_buffers to make dirty buffers
2351          * The page really should be up to date at this point, so tossing
2352          * in the BH_Uptodate is just a sanity check.
2353          */
2354         if (!page_has_buffers(page)) {
2355                 create_empty_buffers(page, s->s_blocksize,
2356                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2357         }
2358         head = page_buffers(page);
2359
2360         /* last page in the file, zero out any contents past the
2361          ** last byte in the file
2362          */
2363         if (page->index >= end_index) {
2364                 unsigned last_offset;
2365
2366                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2367                 /* no file contents in this page */
2368                 if (page->index >= end_index + 1 || !last_offset) {
2369                         unlock_page(page);
2370                         return 0;
2371                 }
2372                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2373         }
2374         bh = head;
2375         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2376         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2377         /* first map all the buffers, logging any direct items we find */
2378         do {
2379                 if (block > last_block) {
2380                         /*
2381                          * This can happen when the block size is less than
2382                          * the page size.  The corresponding bytes in the page
2383                          * were zero filled above
2384                          */
2385                         clear_buffer_dirty(bh);
2386                         set_buffer_uptodate(bh);
2387                 } else if ((checked || buffer_dirty(bh)) &&
2388                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2389                                                        && bh->b_blocknr ==
2390                                                        0))) {
2391                         /* not mapped yet, or it points to a direct item, search
2392                          * the btree for the mapping info, and log any direct
2393                          * items found
2394                          */
2395                         if ((error = map_block_for_writepage(inode, bh, block))) {
2396                                 goto fail;
2397                         }
2398                 }
2399                 bh = bh->b_this_page;
2400                 block++;
2401         } while (bh != head);
2402
2403         /*
2404          * we start the transaction after map_block_for_writepage,
2405          * because it can create holes in the file (an unbounded operation).
2406          * starting it here, we can make a reliable estimate for how many
2407          * blocks we're going to log
2408          */
2409         if (checked) {
2410                 ClearPageChecked(page);
2411                 reiserfs_write_lock(s);
2412                 error = journal_begin(&th, s, bh_per_page + 1);
2413                 if (error) {
2414                         reiserfs_write_unlock(s);
2415                         goto fail;
2416                 }
2417                 reiserfs_update_inode_transaction(inode);
2418         }
2419         /* now go through and lock any dirty buffers on the page */
2420         do {
2421                 get_bh(bh);
2422                 if (!buffer_mapped(bh))
2423                         continue;
2424                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2425                         continue;
2426
2427                 if (checked) {
2428                         reiserfs_prepare_for_journal(s, bh, 1);
2429                         journal_mark_dirty(&th, s, bh);
2430                         continue;
2431                 }
2432                 /* from this point on, we know the buffer is mapped to a
2433                  * real block and not a direct item
2434                  */
2435                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2436                         lock_buffer(bh);
2437                 } else {
2438                         if (!trylock_buffer(bh)) {
2439                                 redirty_page_for_writepage(wbc, page);
2440                                 continue;
2441                         }
2442                 }
2443                 if (test_clear_buffer_dirty(bh)) {
2444                         mark_buffer_async_write(bh);
2445                 } else {
2446                         unlock_buffer(bh);
2447                 }
2448         } while ((bh = bh->b_this_page) != head);
2449
2450         if (checked) {
2451                 error = journal_end(&th, s, bh_per_page + 1);
2452                 reiserfs_write_unlock(s);
2453                 if (error)
2454                         goto fail;
2455         }
2456         BUG_ON(PageWriteback(page));
2457         set_page_writeback(page);
2458         unlock_page(page);
2459
2460         /*
2461          * since any buffer might be the only dirty buffer on the page,
2462          * the first submit_bh can bring the page out of writeback.
2463          * be careful with the buffers.
2464          */
2465         do {
2466                 struct buffer_head *next = bh->b_this_page;
2467                 if (buffer_async_write(bh)) {
2468                         submit_bh(WRITE, bh);
2469                         nr++;
2470                 }
2471                 put_bh(bh);
2472                 bh = next;
2473         } while (bh != head);
2474
2475         error = 0;
2476       done:
2477         if (nr == 0) {
2478                 /*
2479                  * if this page only had a direct item, it is very possible for
2480                  * no io to be required without there being an error.  Or,
2481                  * someone else could have locked them and sent them down the
2482                  * pipe without locking the page
2483                  */
2484                 bh = head;
2485                 do {
2486                         if (!buffer_uptodate(bh)) {
2487                                 partial = 1;
2488                                 break;
2489                         }
2490                         bh = bh->b_this_page;
2491                 } while (bh != head);
2492                 if (!partial)
2493                         SetPageUptodate(page);
2494                 end_page_writeback(page);
2495         }
2496         return error;
2497
2498       fail:
2499         /* catches various errors, we need to make sure any valid dirty blocks
2500          * get to the media.  The page is currently locked and not marked for
2501          * writeback
2502          */
2503         ClearPageUptodate(page);
2504         bh = head;
2505         do {
2506                 get_bh(bh);
2507                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2508                         lock_buffer(bh);
2509                         mark_buffer_async_write(bh);
2510                 } else {
2511                         /*
2512                          * clear any dirty bits that might have come from getting
2513                          * attached to a dirty page
2514                          */
2515                         clear_buffer_dirty(bh);
2516                 }
2517                 bh = bh->b_this_page;
2518         } while (bh != head);
2519         SetPageError(page);
2520         BUG_ON(PageWriteback(page));
2521         set_page_writeback(page);
2522         unlock_page(page);
2523         do {
2524                 struct buffer_head *next = bh->b_this_page;
2525                 if (buffer_async_write(bh)) {
2526                         clear_buffer_dirty(bh);
2527                         submit_bh(WRITE, bh);
2528                         nr++;
2529                 }
2530                 put_bh(bh);
2531                 bh = next;
2532         } while (bh != head);
2533         goto done;
2534 }
2535
2536 static int reiserfs_readpage(struct file *f, struct page *page)
2537 {
2538         return block_read_full_page(page, reiserfs_get_block);
2539 }
2540
2541 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2542 {
2543         struct inode *inode = page->mapping->host;
2544         reiserfs_wait_on_write_block(inode->i_sb);
2545         return reiserfs_write_full_page(page, wbc);
2546 }
2547
2548 static void reiserfs_truncate_failed_write(struct inode *inode)
2549 {
2550         truncate_inode_pages(inode->i_mapping, inode->i_size);
2551         reiserfs_truncate_file(inode, 0);
2552 }
2553
2554 static int reiserfs_write_begin(struct file *file,
2555                                 struct address_space *mapping,
2556                                 loff_t pos, unsigned len, unsigned flags,
2557                                 struct page **pagep, void **fsdata)
2558 {
2559         struct inode *inode;
2560         struct page *page;
2561         pgoff_t index;
2562         int ret;
2563         int old_ref = 0;
2564
2565         inode = mapping->host;
2566         *fsdata = 0;
2567         if (flags & AOP_FLAG_CONT_EXPAND &&
2568             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2569                 pos ++;
2570                 *fsdata = (void *)(unsigned long)flags;
2571         }
2572
2573         index = pos >> PAGE_CACHE_SHIFT;
2574         page = grab_cache_page_write_begin(mapping, index, flags);
2575         if (!page)
2576                 return -ENOMEM;
2577         *pagep = page;
2578
2579         reiserfs_wait_on_write_block(inode->i_sb);
2580         fix_tail_page_for_writing(page);
2581         if (reiserfs_transaction_running(inode->i_sb)) {
2582                 struct reiserfs_transaction_handle *th;
2583                 th = (struct reiserfs_transaction_handle *)current->
2584                     journal_info;
2585                 BUG_ON(!th->t_refcount);
2586                 BUG_ON(!th->t_trans_id);
2587                 old_ref = th->t_refcount;
2588                 th->t_refcount++;
2589         }
2590         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2591                                 reiserfs_get_block);
2592         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2593                 struct reiserfs_transaction_handle *th = current->journal_info;
2594                 /* this gets a little ugly.  If reiserfs_get_block returned an
2595                  * error and left a transacstion running, we've got to close it,
2596                  * and we've got to free handle if it was a persistent transaction.
2597                  *
2598                  * But, if we had nested into an existing transaction, we need
2599                  * to just drop the ref count on the handle.
2600                  *
2601                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2602                  * and it was a persistent trans.  Otherwise, it was nested above.
2603                  */
2604                 if (th->t_refcount > old_ref) {
2605                         if (old_ref)
2606                                 th->t_refcount--;
2607                         else {
2608                                 int err;
2609                                 reiserfs_write_lock(inode->i_sb);
2610                                 err = reiserfs_end_persistent_transaction(th);
2611                                 reiserfs_write_unlock(inode->i_sb);
2612                                 if (err)
2613                                         ret = err;
2614                         }
2615                 }
2616         }
2617         if (ret) {
2618                 unlock_page(page);
2619                 page_cache_release(page);
2620                 /* Truncate allocated blocks */
2621                 reiserfs_truncate_failed_write(inode);
2622         }
2623         return ret;
2624 }
2625
2626 int reiserfs_prepare_write(struct file *f, struct page *page,
2627                            unsigned from, unsigned to)
2628 {
2629         struct inode *inode = page->mapping->host;
2630         int ret;
2631         int old_ref = 0;
2632
2633         reiserfs_write_unlock(inode->i_sb);
2634         reiserfs_wait_on_write_block(inode->i_sb);
2635         reiserfs_write_lock(inode->i_sb);
2636
2637         fix_tail_page_for_writing(page);
2638         if (reiserfs_transaction_running(inode->i_sb)) {
2639                 struct reiserfs_transaction_handle *th;
2640                 th = (struct reiserfs_transaction_handle *)current->
2641                     journal_info;
2642                 BUG_ON(!th->t_refcount);
2643                 BUG_ON(!th->t_trans_id);
2644                 old_ref = th->t_refcount;
2645                 th->t_refcount++;
2646         }
2647
2648         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2649         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2650                 struct reiserfs_transaction_handle *th = current->journal_info;
2651                 /* this gets a little ugly.  If reiserfs_get_block returned an
2652                  * error and left a transacstion running, we've got to close it,
2653                  * and we've got to free handle if it was a persistent transaction.
2654                  *
2655                  * But, if we had nested into an existing transaction, we need
2656                  * to just drop the ref count on the handle.
2657                  *
2658                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2659                  * and it was a persistent trans.  Otherwise, it was nested above.
2660                  */
2661                 if (th->t_refcount > old_ref) {
2662                         if (old_ref)
2663                                 th->t_refcount--;
2664                         else {
2665                                 int err;
2666                                 reiserfs_write_lock(inode->i_sb);
2667                                 err = reiserfs_end_persistent_transaction(th);
2668                                 reiserfs_write_unlock(inode->i_sb);
2669                                 if (err)
2670                                         ret = err;
2671                         }
2672                 }
2673         }
2674         return ret;
2675
2676 }
2677
2678 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2679 {
2680         return generic_block_bmap(as, block, reiserfs_bmap);
2681 }
2682
2683 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2684                               loff_t pos, unsigned len, unsigned copied,
2685                               struct page *page, void *fsdata)
2686 {
2687         struct inode *inode = page->mapping->host;
2688         int ret = 0;
2689         int update_sd = 0;
2690         struct reiserfs_transaction_handle *th;
2691         unsigned start;
2692         int lock_depth = 0;
2693         bool locked = false;
2694
2695         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2696                 pos ++;
2697
2698         reiserfs_wait_on_write_block(inode->i_sb);
2699         if (reiserfs_transaction_running(inode->i_sb))
2700                 th = current->journal_info;
2701         else
2702                 th = NULL;
2703
2704         start = pos & (PAGE_CACHE_SIZE - 1);
2705         if (unlikely(copied < len)) {
2706                 if (!PageUptodate(page))
2707                         copied = 0;
2708
2709                 page_zero_new_buffers(page, start + copied, start + len);
2710         }
2711         flush_dcache_page(page);
2712
2713         reiserfs_commit_page(inode, page, start, start + copied);
2714
2715         /* generic_commit_write does this for us, but does not update the
2716          ** transaction tracking stuff when the size changes.  So, we have
2717          ** to do the i_size updates here.
2718          */
2719         if (pos + copied > inode->i_size) {
2720                 struct reiserfs_transaction_handle myth;
2721                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2722                 locked = true;
2723                 /* If the file have grown beyond the border where it
2724                    can have a tail, unmark it as needing a tail
2725                    packing */
2726                 if ((have_large_tails(inode->i_sb)
2727                      && inode->i_size > i_block_size(inode) * 4)
2728                     || (have_small_tails(inode->i_sb)
2729                         && inode->i_size > i_block_size(inode)))
2730                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2731
2732                 ret = journal_begin(&myth, inode->i_sb, 1);
2733                 if (ret)
2734                         goto journal_error;
2735
2736                 reiserfs_update_inode_transaction(inode);
2737                 inode->i_size = pos + copied;
2738                 /*
2739                  * this will just nest into our transaction.  It's important
2740                  * to use mark_inode_dirty so the inode gets pushed around on the
2741                  * dirty lists, and so that O_SYNC works as expected
2742                  */
2743                 mark_inode_dirty(inode);
2744                 reiserfs_update_sd(&myth, inode);
2745                 update_sd = 1;
2746                 ret = journal_end(&myth, inode->i_sb, 1);
2747                 if (ret)
2748                         goto journal_error;
2749         }
2750         if (th) {
2751                 if (!locked) {
2752                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2753                         locked = true;
2754                 }
2755                 if (!update_sd)
2756                         mark_inode_dirty(inode);
2757                 ret = reiserfs_end_persistent_transaction(th);
2758                 if (ret)
2759                         goto out;
2760         }
2761
2762       out:
2763         if (locked)
2764                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2765         unlock_page(page);
2766         page_cache_release(page);
2767
2768         if (pos + len > inode->i_size)
2769                 reiserfs_truncate_failed_write(inode);
2770
2771         return ret == 0 ? copied : ret;
2772
2773       journal_error:
2774         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2775         locked = false;
2776         if (th) {
2777                 if (!update_sd)
2778                         reiserfs_update_sd(th, inode);
2779                 ret = reiserfs_end_persistent_transaction(th);
2780         }
2781         goto out;
2782 }
2783
2784 int reiserfs_commit_write(struct file *f, struct page *page,
2785                           unsigned from, unsigned to)
2786 {
2787         struct inode *inode = page->mapping->host;
2788         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2789         int ret = 0;
2790         int update_sd = 0;
2791         struct reiserfs_transaction_handle *th = NULL;
2792
2793         reiserfs_write_unlock(inode->i_sb);
2794         reiserfs_wait_on_write_block(inode->i_sb);
2795         reiserfs_write_lock(inode->i_sb);
2796
2797         if (reiserfs_transaction_running(inode->i_sb)) {
2798                 th = current->journal_info;
2799         }
2800         reiserfs_commit_page(inode, page, from, to);
2801
2802         /* generic_commit_write does this for us, but does not update the
2803          ** transaction tracking stuff when the size changes.  So, we have
2804          ** to do the i_size updates here.
2805          */
2806         if (pos > inode->i_size) {
2807                 struct reiserfs_transaction_handle myth;
2808                 /* If the file have grown beyond the border where it
2809                    can have a tail, unmark it as needing a tail
2810                    packing */
2811                 if ((have_large_tails(inode->i_sb)
2812                      && inode->i_size > i_block_size(inode) * 4)
2813                     || (have_small_tails(inode->i_sb)
2814                         && inode->i_size > i_block_size(inode)))
2815                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2816
2817                 ret = journal_begin(&myth, inode->i_sb, 1);
2818                 if (ret)
2819                         goto journal_error;
2820
2821                 reiserfs_update_inode_transaction(inode);
2822                 inode->i_size = pos;
2823                 /*
2824                  * this will just nest into our transaction.  It's important
2825                  * to use mark_inode_dirty so the inode gets pushed around on the
2826                  * dirty lists, and so that O_SYNC works as expected
2827                  */
2828                 mark_inode_dirty(inode);
2829                 reiserfs_update_sd(&myth, inode);
2830                 update_sd = 1;
2831                 ret = journal_end(&myth, inode->i_sb, 1);
2832                 if (ret)
2833                         goto journal_error;
2834         }
2835         if (th) {
2836                 if (!update_sd)
2837                         mark_inode_dirty(inode);
2838                 ret = reiserfs_end_persistent_transaction(th);
2839                 if (ret)
2840                         goto out;
2841         }
2842
2843       out:
2844         return ret;
2845
2846       journal_error:
2847         if (th) {
2848                 if (!update_sd)
2849                         reiserfs_update_sd(th, inode);
2850                 ret = reiserfs_end_persistent_transaction(th);
2851         }
2852
2853         return ret;
2854 }
2855
2856 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2857 {
2858         if (reiserfs_attrs(inode->i_sb)) {
2859                 if (sd_attrs & REISERFS_SYNC_FL)
2860                         inode->i_flags |= S_SYNC;
2861                 else
2862                         inode->i_flags &= ~S_SYNC;
2863                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2864                         inode->i_flags |= S_IMMUTABLE;
2865                 else
2866                         inode->i_flags &= ~S_IMMUTABLE;
2867                 if (sd_attrs & REISERFS_APPEND_FL)
2868                         inode->i_flags |= S_APPEND;
2869                 else
2870                         inode->i_flags &= ~S_APPEND;
2871                 if (sd_attrs & REISERFS_NOATIME_FL)
2872                         inode->i_flags |= S_NOATIME;
2873                 else
2874                         inode->i_flags &= ~S_NOATIME;
2875                 if (sd_attrs & REISERFS_NOTAIL_FL)
2876                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2877                 else
2878                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2879         }
2880 }
2881
2882 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2883 {
2884         if (reiserfs_attrs(inode->i_sb)) {
2885                 if (inode->i_flags & S_IMMUTABLE)
2886                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2887                 else
2888                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2889                 if (inode->i_flags & S_SYNC)
2890                         *sd_attrs |= REISERFS_SYNC_FL;
2891                 else
2892                         *sd_attrs &= ~REISERFS_SYNC_FL;
2893                 if (inode->i_flags & S_NOATIME)
2894                         *sd_attrs |= REISERFS_NOATIME_FL;
2895                 else
2896                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2897                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2898                         *sd_attrs |= REISERFS_NOTAIL_FL;
2899                 else
2900                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2901         }
2902 }
2903
2904 /* decide if this buffer needs to stay around for data logging or ordered
2905 ** write purposes
2906 */
2907 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2908 {
2909         int ret = 1;
2910         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2911
2912         lock_buffer(bh);
2913         spin_lock(&j->j_dirty_buffers_lock);
2914         if (!buffer_mapped(bh)) {
2915                 goto free_jh;
2916         }
2917         /* the page is locked, and the only places that log a data buffer
2918          * also lock the page.
2919          */
2920         if (reiserfs_file_data_log(inode)) {
2921                 /*
2922                  * very conservative, leave the buffer pinned if
2923                  * anyone might need it.
2924                  */
2925                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2926                         ret = 0;
2927                 }
2928         } else  if (buffer_dirty(bh)) {
2929                 struct reiserfs_journal_list *jl;
2930                 struct reiserfs_jh *jh = bh->b_private;
2931
2932                 /* why is this safe?
2933                  * reiserfs_setattr updates i_size in the on disk
2934                  * stat data before allowing vmtruncate to be called.
2935                  *
2936                  * If buffer was put onto the ordered list for this
2937                  * transaction, we know for sure either this transaction
2938                  * or an older one already has updated i_size on disk,
2939                  * and this ordered data won't be referenced in the file
2940                  * if we crash.
2941                  *
2942                  * if the buffer was put onto the ordered list for an older
2943                  * transaction, we need to leave it around
2944                  */
2945                 if (jh && (jl = jh->jl)
2946                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2947                         ret = 0;
2948         }
2949       free_jh:
2950         if (ret && bh->b_private) {
2951                 reiserfs_free_jh(bh);
2952         }
2953         spin_unlock(&j->j_dirty_buffers_lock);
2954         unlock_buffer(bh);
2955         return ret;
2956 }
2957
2958 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2959 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2960 {
2961         struct buffer_head *head, *bh, *next;
2962         struct inode *inode = page->mapping->host;
2963         unsigned int curr_off = 0;
2964         int ret = 1;
2965
2966         BUG_ON(!PageLocked(page));
2967
2968         if (offset == 0)
2969                 ClearPageChecked(page);
2970
2971         if (!page_has_buffers(page))
2972                 goto out;
2973
2974         head = page_buffers(page);
2975         bh = head;
2976         do {
2977                 unsigned int next_off = curr_off + bh->b_size;
2978                 next = bh->b_this_page;
2979
2980                 /*
2981                  * is this block fully invalidated?
2982                  */
2983                 if (offset <= curr_off) {
2984                         if (invalidatepage_can_drop(inode, bh))
2985                                 reiserfs_unmap_buffer(bh);
2986                         else
2987                                 ret = 0;
2988                 }
2989                 curr_off = next_off;
2990                 bh = next;
2991         } while (bh != head);
2992
2993         /*
2994          * We release buffers only if the entire page is being invalidated.
2995          * The get_block cached value has been unconditionally invalidated,
2996          * so real IO is not possible anymore.
2997          */
2998         if (!offset && ret) {
2999                 ret = try_to_release_page(page, 0);
3000                 /* maybe should BUG_ON(!ret); - neilb */
3001         }
3002       out:
3003         return;
3004 }
3005
3006 static int reiserfs_set_page_dirty(struct page *page)
3007 {
3008         struct inode *inode = page->mapping->host;
3009         if (reiserfs_file_data_log(inode)) {
3010                 SetPageChecked(page);
3011                 return __set_page_dirty_nobuffers(page);
3012         }
3013         return __set_page_dirty_buffers(page);
3014 }
3015
3016 /*
3017  * Returns 1 if the page's buffers were dropped.  The page is locked.
3018  *
3019  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3020  * in the buffers at page_buffers(page).
3021  *
3022  * even in -o notail mode, we can't be sure an old mount without -o notail
3023  * didn't create files with tails.
3024  */
3025 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3026 {
3027         struct inode *inode = page->mapping->host;
3028         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3029         struct buffer_head *head;
3030         struct buffer_head *bh;
3031         int ret = 1;
3032
3033         WARN_ON(PageChecked(page));
3034         spin_lock(&j->j_dirty_buffers_lock);
3035         head = page_buffers(page);
3036         bh = head;
3037         do {
3038                 if (bh->b_private) {
3039                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3040                                 reiserfs_free_jh(bh);
3041                         } else {
3042                                 ret = 0;
3043                                 break;
3044                         }
3045                 }
3046                 bh = bh->b_this_page;
3047         } while (bh != head);
3048         if (ret)
3049                 ret = try_to_free_buffers(page);
3050         spin_unlock(&j->j_dirty_buffers_lock);
3051         return ret;
3052 }
3053
3054 /* We thank Mingming Cao for helping us understand in great detail what
3055    to do in this section of the code. */
3056 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3057                                   const struct iovec *iov, loff_t offset,
3058                                   unsigned long nr_segs)
3059 {
3060         struct file *file = iocb->ki_filp;
3061         struct inode *inode = file->f_mapping->host;
3062
3063         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3064                                   offset, nr_segs,
3065                                   reiserfs_get_blocks_direct_io, NULL);
3066 }
3067
3068 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3069 {
3070         struct inode *inode = dentry->d_inode;
3071         unsigned int ia_valid;
3072         int depth;
3073         int error;
3074
3075         /* must be turned off for recursive notify_change calls */
3076         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3077
3078         depth = reiserfs_write_lock_once(inode->i_sb);
3079         if (is_quota_modification(inode, attr))
3080                 dquot_initialize(inode);
3081
3082         if (attr->ia_valid & ATTR_SIZE) {
3083                 /* version 2 items will be caught by the s_maxbytes check
3084                  ** done for us in vmtruncate
3085                  */
3086                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3087                     attr->ia_size > MAX_NON_LFS) {
3088                         error = -EFBIG;
3089                         goto out;
3090                 }
3091                 /* fill in hole pointers in the expanding truncate case. */
3092                 if (attr->ia_size > inode->i_size) {
3093                         error = generic_cont_expand_simple(inode, attr->ia_size);
3094                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3095                                 int err;
3096                                 struct reiserfs_transaction_handle th;
3097                                 /* we're changing at most 2 bitmaps, inode + super */
3098                                 err = journal_begin(&th, inode->i_sb, 4);
3099                                 if (!err) {
3100                                         reiserfs_discard_prealloc(&th, inode);
3101                                         err = journal_end(&th, inode->i_sb, 4);
3102                                 }
3103                                 if (err)
3104                                         error = err;
3105                         }
3106                         if (error)
3107                                 goto out;
3108                         /*
3109                          * file size is changed, ctime and mtime are
3110                          * to be updated
3111                          */
3112                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3113                 }
3114         }
3115
3116         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3117              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3118             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3119                 /* stat data of format v3.5 has 16 bit uid and gid */
3120                 error = -EINVAL;
3121                 goto out;
3122         }
3123
3124         error = inode_change_ok(inode, attr);
3125         if (!error) {
3126                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3127                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3128                         error = reiserfs_chown_xattrs(inode, attr);
3129
3130                         if (!error) {
3131                                 struct reiserfs_transaction_handle th;
3132                                 int jbegin_count =
3133                                     2 *
3134                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3135                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3136                                     2;
3137
3138                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3139                                 error =
3140                                     journal_begin(&th, inode->i_sb,
3141                                                   jbegin_count);
3142                                 if (error)
3143                                         goto out;
3144                                 error = dquot_transfer(inode, attr);
3145                                 if (error) {
3146                                         journal_end(&th, inode->i_sb,
3147                                                     jbegin_count);
3148                                         goto out;
3149                                 }
3150                                 /* Update corresponding info in inode so that everything is in
3151                                  * one transaction */
3152                                 if (attr->ia_valid & ATTR_UID)
3153                                         inode->i_uid = attr->ia_uid;
3154                                 if (attr->ia_valid & ATTR_GID)
3155                                         inode->i_gid = attr->ia_gid;
3156                                 mark_inode_dirty(inode);
3157                                 error =
3158                                     journal_end(&th, inode->i_sb, jbegin_count);
3159                         }
3160                 }
3161                 if (!error) {
3162                         /*
3163                          * Relax the lock here, as it might truncate the
3164                          * inode pages and wait for inode pages locks.
3165                          * To release such page lock, the owner needs the
3166                          * reiserfs lock
3167                          */
3168                         reiserfs_write_unlock_once(inode->i_sb, depth);
3169                         error = inode_setattr(inode, attr);
3170                         depth = reiserfs_write_lock_once(inode->i_sb);
3171                 }
3172         }
3173
3174         if (!error && reiserfs_posixacl(inode->i_sb)) {
3175                 if (attr->ia_valid & ATTR_MODE)
3176                         error = reiserfs_acl_chmod(inode);
3177         }
3178
3179       out:
3180         reiserfs_write_unlock_once(inode->i_sb, depth);
3181
3182         return error;
3183 }
3184
3185 const struct address_space_operations reiserfs_address_space_operations = {
3186         .writepage = reiserfs_writepage,
3187         .readpage = reiserfs_readpage,
3188         .readpages = reiserfs_readpages,
3189         .releasepage = reiserfs_releasepage,
3190         .invalidatepage = reiserfs_invalidatepage,
3191         .sync_page = block_sync_page,
3192         .write_begin = reiserfs_write_begin,
3193         .write_end = reiserfs_write_end,
3194         .bmap = reiserfs_aop_bmap,
3195         .direct_IO = reiserfs_direct_IO,
3196         .set_page_dirty = reiserfs_set_page_dirty,
3197 };