SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/exportfs.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <asm/uaccess.h>
  15 #include <asm/unaligned.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/mpage.h>
  18 #include <linux/writeback.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/swap.h>
  21
  22 int reiserfs_commit_write(struct file *f, struct page *page,
  23                           unsigned from, unsigned to);
  24 int reiserfs_prepare_write(struct file *f, struct page *page,
  25                            unsigned from, unsigned to);
  26
  27 void reiserfs_delete_inode(struct inode *inode)
  28 {
  29         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  30         int jbegin_count =
  31             JOURNAL_PER_BALANCE_CNT * 2 +
  32             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  33         struct reiserfs_transaction_handle th;
  34         int depth;
  35         int err;
  36
  37         if (!is_bad_inode(inode))
  38                 vfs_dq_init(inode);
  39
  40         truncate_inode_pages(&inode->i_data, 0);
  41
  42         depth = reiserfs_write_lock_once(inode->i_sb);
  43
  44         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  45         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  46                 reiserfs_delete_xattrs(inode);
  47
  48                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  49                         goto out;
  50                 reiserfs_update_inode_transaction(inode);
  51
  52                 reiserfs_discard_prealloc(&th, inode);
  53
  54                 err = reiserfs_delete_object(&th, inode);
  55
  56                 /* Do quota update inside a transaction for journaled quotas. We must do that
  57                  * after delete_object so that quota updates go into the same transaction as
  58                  * stat data deletion */
  59                 if (!err)
  60                         dquot_free_inode(inode);
  61
  62                 if (journal_end(&th, inode->i_sb, jbegin_count))
  63                         goto out;
  64
  65                 /* check return value from reiserfs_delete_object after
  66                  * ending the transaction
  67                  */
  68                 if (err)
  69                     goto out;
  70
  71                 /* all items of file are deleted, so we can remove "save" link */
  72                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  73                                                                  * about an error here */
  74         } else {
  75                 /* no object items are in the tree */
  76                 ;
  77         }
  78       out:
  79         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  80         inode->i_blocks = 0;
  81         reiserfs_write_unlock_once(inode->i_sb, depth);
  82 }
  83
  84 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  85                           __u32 objectid, loff_t offset, int type, int length)
  86 {
  87         key->version = version;
  88
  89         key->on_disk_key.k_dir_id = dirid;
  90         key->on_disk_key.k_objectid = objectid;
  91         set_cpu_key_k_offset(key, offset);
  92         set_cpu_key_k_type(key, type);
  93         key->key_length = length;
  94 }
  95
  96 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
  97    offset and type of key */
  98 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
  99                   int type, int length)
 100 {
 101         _make_cpu_key(key, get_inode_item_key_version(inode),
 102                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 103                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 104                       length);
 105 }
 106
 107 //
 108 // when key is 0, do not set version and short key
 109 //
 110 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 111                               int version,
 112                               loff_t offset, int type, int length,
 113                               int entry_count /*or ih_free_space */ )
 114 {
 115         if (key) {
 116                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 117                 ih->ih_key.k_objectid =
 118                     cpu_to_le32(key->on_disk_key.k_objectid);
 119         }
 120         put_ih_version(ih, version);
 121         set_le_ih_k_offset(ih, offset);
 122         set_le_ih_k_type(ih, type);
 123         put_ih_item_len(ih, length);
 124         /*    set_ih_free_space (ih, 0); */
 125         // for directory items it is entry count, for directs and stat
 126         // datas - 0xffff, for indirects - 0
 127         put_ih_entry_count(ih, entry_count);
 128 }
 129
 130 //
 131 // FIXME: we might cache recently accessed indirect item
 132
 133 // Ugh.  Not too eager for that....
 134 //  I cut the code until such time as I see a convincing argument (benchmark).
 135 // I don't want a bloated inode struct..., and I don't like code complexity....
 136
 137 /* cutting the code is fine, since it really isn't in use yet and is easy
 138 ** to add back in.  But, Vladimir has a really good idea here.  Think
 139 ** about what happens for reading a file.  For each page,
 140 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 141 ** an indirect item.  This indirect item has X number of pointers, where
 142 ** X is a big number if we've done the block allocation right.  But,
 143 ** we only use one or two of these pointers during each call to readpage,
 144 ** needlessly researching again later on.
 145 **
 146 ** The size of the cache could be dynamic based on the size of the file.
 147 **
 148 ** I'd also like to see us cache the location the stat data item, since
 149 ** we are needlessly researching for that frequently.
 150 **
 151 ** --chris
 152 */
 153
 154 /* If this page has a file tail in it, and
 155 ** it was read in by get_block_create_0, the page data is valid,
 156 ** but tail is still sitting in a direct item, and we can't write to
 157 ** it.  So, look through this page, and check all the mapped buffers
 158 ** to make sure they have valid block numbers.  Any that don't need
 159 ** to be unmapped, so that block_prepare_write will correctly call
 160 ** reiserfs_get_block to convert the tail into an unformatted node
 161 */
 162 static inline void fix_tail_page_for_writing(struct page *page)
 163 {
 164         struct buffer_head *head, *next, *bh;
 165
 166         if (page && page_has_buffers(page)) {
 167                 head = page_buffers(page);
 168                 bh = head;
 169                 do {
 170                         next = bh->b_this_page;
 171                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 172                                 reiserfs_unmap_buffer(bh);
 173                         }
 174                         bh = next;
 175                 } while (bh != head);
 176         }
 177 }
 178
 179 /* reiserfs_get_block does not need to allocate a block only if it has been
 180    done already or non-hole position has been found in the indirect item */
 181 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 182                                     struct item_head *ih,
 183                                     __le32 * item, int pos_in_item)
 184 {
 185         if (allocated)
 186                 return 0;
 187         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 188             get_block_num(item, pos_in_item))
 189                 return 0;
 190         return 1;
 191 }
 192
 193 static inline int indirect_item_found(int retval, struct item_head *ih)
 194 {
 195         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 196 }
 197
 198 static inline void set_block_dev_mapped(struct buffer_head *bh,
 199                                         b_blocknr_t block, struct inode *inode)
 200 {
 201         map_bh(bh, inode->i_sb, block);
 202 }
 203
 204 //
 205 // files which were created in the earlier version can not be longer,
 206 // than 2 gb
 207 //
 208 static int file_capable(struct inode *inode, sector_t block)
 209 {
 210         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 211             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 212                 return 1;
 213
 214         return 0;
 215 }
 216
 217 static int restart_transaction(struct reiserfs_transaction_handle *th,
 218                                struct inode *inode, struct treepath *path)
 219 {
 220         struct super_block *s = th->t_super;
 221         int len = th->t_blocks_allocated;
 222         int err;
 223
 224         BUG_ON(!th->t_trans_id);
 225         BUG_ON(!th->t_refcount);
 226
 227         pathrelse(path);
 228
 229         /* we cannot restart while nested */
 230         if (th->t_refcount > 1) {
 231                 return 0;
 232         }
 233         reiserfs_update_sd(th, inode);
 234         err = journal_end(th, s, len);
 235         if (!err) {
 236                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 237                 if (!err)
 238                         reiserfs_update_inode_transaction(inode);
 239         }
 240         return err;
 241 }
 242
 243 // it is called by get_block when create == 0. Returns block number
 244 // for 'block'-th logical block of file. When it hits direct item it
 245 // returns 0 (being called from bmap) or read direct item into piece
 246 // of page (bh_result)
 247
 248 // Please improve the english/clarity in the comment above, as it is
 249 // hard to understand.
 250
 251 static int _get_block_create_0(struct inode *inode, sector_t block,
 252                                struct buffer_head *bh_result, int args)
 253 {
 254         INITIALIZE_PATH(path);
 255         struct cpu_key key;
 256         struct buffer_head *bh;
 257         struct item_head *ih, tmp_ih;
 258         b_blocknr_t blocknr;
 259         char *p = NULL;
 260         int chars;
 261         int ret;
 262         int result;
 263         int done = 0;
 264         unsigned long offset;
 265
 266         // prepare the key to look for the 'block'-th block of file
 267         make_cpu_key(&key, inode,
 268                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 269                      3);
 270
 271         result = search_for_position_by_key(inode->i_sb, &key, &path);
 272         if (result != POSITION_FOUND) {
 273                 pathrelse(&path);
 274                 if (p)
 275                         kunmap(bh_result->b_page);
 276                 if (result == IO_ERROR)
 277                         return -EIO;
 278                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 279                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 280                 if ((args & GET_BLOCK_NO_HOLE)
 281                     && !PageUptodate(bh_result->b_page)) {
 282                         return -ENOENT;
 283                 }
 284                 return 0;
 285         }
 286         //
 287         bh = get_last_bh(&path);
 288         ih = get_ih(&path);
 289         if (is_indirect_le_ih(ih)) {
 290                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 291
 292                 /* FIXME: here we could cache indirect item or part of it in
 293                    the inode to avoid search_by_key in case of subsequent
 294                    access to file */
 295                 blocknr = get_block_num(ind_item, path.pos_in_item);
 296                 ret = 0;
 297                 if (blocknr) {
 298                         map_bh(bh_result, inode->i_sb, blocknr);
 299                         if (path.pos_in_item ==
 300                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 301                                 set_buffer_boundary(bh_result);
 302                         }
 303                 } else
 304                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 305                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 306                 if ((args & GET_BLOCK_NO_HOLE)
 307                             && !PageUptodate(bh_result->b_page)) {
 308                         ret = -ENOENT;
 309                 }
 310
 311                 pathrelse(&path);
 312                 if (p)
 313                         kunmap(bh_result->b_page);
 314                 return ret;
 315         }
 316         // requested data are in direct item(s)
 317         if (!(args & GET_BLOCK_READ_DIRECT)) {
 318                 // we are called by bmap. FIXME: we can not map block of file
 319                 // when it is stored in direct item(s)
 320                 pathrelse(&path);
 321                 if (p)
 322                         kunmap(bh_result->b_page);
 323                 return -ENOENT;
 324         }
 325
 326         /* if we've got a direct item, and the buffer or page was uptodate,
 327          ** we don't want to pull data off disk again.  skip to the
 328          ** end, where we map the buffer and return
 329          */
 330         if (buffer_uptodate(bh_result)) {
 331                 goto finished;
 332         } else
 333                 /*
 334                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 335                  ** pages without any buffers.  If the page is up to date, we don't want
 336                  ** read old data off disk.  Set the up to date bit on the buffer instead
 337                  ** and jump to the end
 338                  */
 339         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 340                 set_buffer_uptodate(bh_result);
 341                 goto finished;
 342         }
 343         // read file tail into part of page
 344         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 345         copy_item_head(&tmp_ih, ih);
 346
 347         /* we only want to kmap if we are reading the tail into the page.
 348          ** this is not the common case, so we don't kmap until we are
 349          ** sure we need to.  But, this means the item might move if
 350          ** kmap schedules
 351          */
 352         if (!p)
 353                 p = (char *)kmap(bh_result->b_page);
 354
 355         p += offset;
 356         memset(p, 0, inode->i_sb->s_blocksize);
 357         do {
 358                 if (!is_direct_le_ih(ih)) {
 359                         BUG();
 360                 }
 361                 /* make sure we don't read more bytes than actually exist in
 362                  ** the file.  This can happen in odd cases where i_size isn't
 363                  ** correct, and when direct item padding results in a few
 364                  ** extra bytes at the end of the direct item
 365                  */
 366                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 367                         break;
 368                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 369                         chars =
 370                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 371                             path.pos_in_item;
 372                         done = 1;
 373                 } else {
 374                         chars = ih_item_len(ih) - path.pos_in_item;
 375                 }
 376                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 377
 378                 if (done)
 379                         break;
 380
 381                 p += chars;
 382
 383                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 384                         // we done, if read direct item is not the last item of
 385                         // node FIXME: we could try to check right delimiting key
 386                         // to see whether direct item continues in the right
 387                         // neighbor or rely on i_size
 388                         break;
 389
 390                 // update key to look for the next piece
 391                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 392                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 393                 if (result != POSITION_FOUND)
 394                         // i/o error most likely
 395                         break;
 396                 bh = get_last_bh(&path);
 397                 ih = get_ih(&path);
 398         } while (1);
 399
 400         flush_dcache_page(bh_result->b_page);
 401         kunmap(bh_result->b_page);
 402
 403       finished:
 404         pathrelse(&path);
 405
 406         if (result == IO_ERROR)
 407                 return -EIO;
 408
 409         /* this buffer has valid data, but isn't valid for io.  mapping it to
 410          * block #0 tells the rest of reiserfs it just has a tail in it
 411          */
 412         map_bh(bh_result, inode->i_sb, 0);
 413         set_buffer_uptodate(bh_result);
 414         return 0;
 415 }
 416
 417 // this is called to create file map. So, _get_block_create_0 will not
 418 // read direct item
 419 static int reiserfs_bmap(struct inode *inode, sector_t block,
 420                          struct buffer_head *bh_result, int create)
 421 {
 422         if (!file_capable(inode, block))
 423                 return -EFBIG;
 424
 425         reiserfs_write_lock(inode->i_sb);
 426         /* do not read the direct item */
 427         _get_block_create_0(inode, block, bh_result, 0);
 428         reiserfs_write_unlock(inode->i_sb);
 429         return 0;
 430 }
 431
 432 /* special version of get_block that is only used by grab_tail_page right
 433 ** now.  It is sent to block_prepare_write, and when you try to get a
 434 ** block past the end of the file (or a block from a hole) it returns
 435 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 436 ** be able to do i/o on the buffers returned, unless an error value
 437 ** is also returned.
 438 **
 439 ** So, this allows block_prepare_write to be used for reading a single block
 440 ** in a page.  Where it does not produce a valid page for holes, or past the
 441 ** end of the file.  This turns out to be exactly what we need for reading
 442 ** tails for conversion.
 443 **
 444 ** The point of the wrapper is forcing a certain value for create, even
 445 ** though the VFS layer is calling this function with create==1.  If you
 446 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 447 ** don't use this function.
 448 */
 449 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 450                                        struct buffer_head *bh_result,
 451                                        int create)
 452 {
 453         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 454 }
 455
 456 /* This is special helper for reiserfs_get_block in case we are executing
 457    direct_IO request. */
 458 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 459                                          sector_t iblock,
 460                                          struct buffer_head *bh_result,
 461                                          int create)
 462 {
 463         int ret;
 464
 465         bh_result->b_page = NULL;
 466
 467         /* We set the b_size before reiserfs_get_block call since it is
 468            referenced in convert_tail_for_hole() that may be called from
 469            reiserfs_get_block() */
 470         bh_result->b_size = (1 << inode->i_blkbits);
 471
 472         ret = reiserfs_get_block(inode, iblock, bh_result,
 473                                  create | GET_BLOCK_NO_DANGLE);
 474         if (ret)
 475                 goto out;
 476
 477         /* don't allow direct io onto tail pages */
 478         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 479                 /* make sure future calls to the direct io funcs for this offset
 480                  ** in the file fail by unmapping the buffer
 481                  */
 482                 clear_buffer_mapped(bh_result);
 483                 ret = -EINVAL;
 484         }
 485         /* Possible unpacked tail. Flush the data before pages have
 486            disappeared */
 487         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 488                 int err;
 489
 490                 reiserfs_write_lock(inode->i_sb);
 491
 492                 err = reiserfs_commit_for_inode(inode);
 493                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 494
 495                 reiserfs_write_unlock(inode->i_sb);
 496
 497                 if (err < 0)
 498                         ret = err;
 499         }
 500       out:
 501         return ret;
 502 }
 503
 504 /*
 505 ** helper function for when reiserfs_get_block is called for a hole
 506 ** but the file tail is still in a direct item
 507 ** bh_result is the buffer head for the hole
 508 ** tail_offset is the offset of the start of the tail in the file
 509 **
 510 ** This calls prepare_write, which will start a new transaction
 511 ** you should not be in a transaction, or have any paths held when you
 512 ** call this.
 513 */
 514 static int convert_tail_for_hole(struct inode *inode,
 515                                  struct buffer_head *bh_result,
 516                                  loff_t tail_offset)
 517 {
 518         unsigned long index;
 519         unsigned long tail_end;
 520         unsigned long tail_start;
 521         struct page *tail_page;
 522         struct page *hole_page = bh_result->b_page;
 523         int retval = 0;
 524
 525         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 526                 return -EIO;
 527
 528         /* always try to read until the end of the block */
 529         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 530         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 531
 532         index = tail_offset >> PAGE_CACHE_SHIFT;
 533         /* hole_page can be zero in case of direct_io, we are sure
 534            that we cannot get here if we write with O_DIRECT into
 535            tail page */
 536         if (!hole_page || index != hole_page->index) {
 537                 tail_page = grab_cache_page(inode->i_mapping, index);
 538                 retval = -ENOMEM;
 539                 if (!tail_page) {
 540                         goto out;
 541                 }
 542         } else {
 543                 tail_page = hole_page;
 544         }
 545
 546         /* we don't have to make sure the conversion did not happen while
 547          ** we were locking the page because anyone that could convert
 548          ** must first take i_mutex.
 549          **
 550          ** We must fix the tail page for writing because it might have buffers
 551          ** that are mapped, but have a block number of 0.  This indicates tail
 552          ** data that has been read directly into the page, and block_prepare_write
 553          ** won't trigger a get_block in this case.
 554          */
 555         fix_tail_page_for_writing(tail_page);
 556         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 557         if (retval)
 558                 goto unlock;
 559
 560         /* tail conversion might change the data in the page */
 561         flush_dcache_page(tail_page);
 562
 563         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 564
 565       unlock:
 566         if (tail_page != hole_page) {
 567                 unlock_page(tail_page);
 568                 page_cache_release(tail_page);
 569         }
 570       out:
 571         return retval;
 572 }
 573
 574 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 575                                   sector_t block,
 576                                   struct inode *inode,
 577                                   b_blocknr_t * allocated_block_nr,
 578                                   struct treepath *path, int flags)
 579 {
 580         BUG_ON(!th->t_trans_id);
 581
 582 #ifdef REISERFS_PREALLOCATE
 583         if (!(flags & GET_BLOCK_NO_IMUX)) {
 584                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 585                                                   path, block);
 586         }
 587 #endif
 588         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 589                                          block);
 590 }
 591
 592 int reiserfs_get_block(struct inode *inode, sector_t block,
 593                        struct buffer_head *bh_result, int create)
 594 {
 595         int repeat, retval = 0;
 596         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 597         INITIALIZE_PATH(path);
 598         int pos_in_item;
 599         struct cpu_key key;
 600         struct buffer_head *bh, *unbh = NULL;
 601         struct item_head *ih, tmp_ih;
 602         __le32 *item;
 603         int done;
 604         int fs_gen;
 605         int lock_depth;
 606         struct reiserfs_transaction_handle *th = NULL;
 607         /* space reserved in transaction batch:
 608            . 3 balancings in direct->indirect conversion
 609            . 1 block involved into reiserfs_update_sd()
 610            XXX in practically impossible worst case direct2indirect()
 611            can incur (much) more than 3 balancings.
 612            quota update for user, group */
 613         int jbegin_count =
 614             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 615             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 616         int version;
 617         int dangle = 1;
 618         loff_t new_offset =
 619             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 620
 621         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 622         version = get_inode_item_key_version(inode);
 623
 624         if (!file_capable(inode, block)) {
 625                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 626                 return -EFBIG;
 627         }
 628
 629         /* if !create, we aren't changing the FS, so we don't need to
 630          ** log anything, so we don't need to start a transaction
 631          */
 632         if (!(create & GET_BLOCK_CREATE)) {
 633                 int ret;
 634                 /* find number of block-th logical block of the file */
 635                 ret = _get_block_create_0(inode, block, bh_result,
 636                                           create | GET_BLOCK_READ_DIRECT);
 637                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 638                 return ret;
 639         }
 640         /*
 641          * if we're already in a transaction, make sure to close
 642          * any new transactions we start in this func
 643          */
 644         if ((create & GET_BLOCK_NO_DANGLE) ||
 645             reiserfs_transaction_running(inode->i_sb))
 646                 dangle = 0;
 647
 648         /* If file is of such a size, that it might have a tail and tails are enabled
 649          ** we should mark it as possibly needing tail packing on close
 650          */
 651         if ((have_large_tails(inode->i_sb)
 652              && inode->i_size < i_block_size(inode) * 4)
 653             || (have_small_tails(inode->i_sb)
 654                 && inode->i_size < i_block_size(inode)))
 655                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 656
 657         /* set the key of the first byte in the 'block'-th block of file */
 658         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 659         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 660               start_trans:
 661                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 662                 if (!th) {
 663                         retval = -ENOMEM;
 664                         goto failure;
 665                 }
 666                 reiserfs_update_inode_transaction(inode);
 667         }
 668       research:
 669
 670         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 671         if (retval == IO_ERROR) {
 672                 retval = -EIO;
 673                 goto failure;
 674         }
 675
 676         bh = get_last_bh(&path);
 677         ih = get_ih(&path);
 678         item = get_item(&path);
 679         pos_in_item = path.pos_in_item;
 680
 681         fs_gen = get_generation(inode->i_sb);
 682         copy_item_head(&tmp_ih, ih);
 683
 684         if (allocation_needed
 685             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 686                 /* we have to allocate block for the unformatted node */
 687                 if (!th) {
 688                         pathrelse(&path);
 689                         goto start_trans;
 690                 }
 691
 692                 repeat =
 693                     _allocate_block(th, block, inode, &allocated_block_nr,
 694                                     &path, create);
 695
 696                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 697                         /* restart the transaction to give the journal a chance to free
 698                          ** some blocks.  releases the path, so we have to go back to
 699                          ** research if we succeed on the second try
 700                          */
 701                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 702                         retval = restart_transaction(th, inode, &path);
 703                         if (retval)
 704                                 goto failure;
 705                         repeat =
 706                             _allocate_block(th, block, inode,
 707                                             &allocated_block_nr, NULL, create);
 708
 709                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 710                                 goto research;
 711                         }
 712                         if (repeat == QUOTA_EXCEEDED)
 713                                 retval = -EDQUOT;
 714                         else
 715                                 retval = -ENOSPC;
 716                         goto failure;
 717                 }
 718
 719                 if (fs_changed(fs_gen, inode->i_sb)
 720                     && item_moved(&tmp_ih, &path)) {
 721                         goto research;
 722                 }
 723         }
 724
 725         if (indirect_item_found(retval, ih)) {
 726                 b_blocknr_t unfm_ptr;
 727                 /* 'block'-th block is in the file already (there is
 728                    corresponding cell in some indirect item). But it may be
 729                    zero unformatted node pointer (hole) */
 730                 unfm_ptr = get_block_num(item, pos_in_item);
 731                 if (unfm_ptr == 0) {
 732                         /* use allocated block to plug the hole */
 733                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 734                         if (fs_changed(fs_gen, inode->i_sb)
 735                             && item_moved(&tmp_ih, &path)) {
 736                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 737                                                                  bh);
 738                                 goto research;
 739                         }
 740                         set_buffer_new(bh_result);
 741                         if (buffer_dirty(bh_result)
 742                             && reiserfs_data_ordered(inode->i_sb))
 743                                 reiserfs_add_ordered_list(inode, bh_result);
 744                         put_block_num(item, pos_in_item, allocated_block_nr);
 745                         unfm_ptr = allocated_block_nr;
 746                         journal_mark_dirty(th, inode->i_sb, bh);
 747                         reiserfs_update_sd(th, inode);
 748                 }
 749                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 750                 pathrelse(&path);
 751                 retval = 0;
 752                 if (!dangle && th)
 753                         retval = reiserfs_end_persistent_transaction(th);
 754
 755                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 756
 757                 /* the item was found, so new blocks were not added to the file
 758                  ** there is no need to make sure the inode is updated with this
 759                  ** transaction
 760                  */
 761                 return retval;
 762         }
 763
 764         if (!th) {
 765                 pathrelse(&path);
 766                 goto start_trans;
 767         }
 768
 769         /* desired position is not found or is in the direct item. We have
 770            to append file with holes up to 'block'-th block converting
 771            direct items to indirect one if necessary */
 772         done = 0;
 773         do {
 774                 if (is_statdata_le_ih(ih)) {
 775                         __le32 unp = 0;
 776                         struct cpu_key tmp_key;
 777
 778                         /* indirect item has to be inserted */
 779                         make_le_item_head(&tmp_ih, &key, version, 1,
 780                                           TYPE_INDIRECT, UNFM_P_SIZE,
 781                                           0 /* free_space */ );
 782
 783                         if (cpu_key_k_offset(&key) == 1) {
 784                                 /* we are going to add 'block'-th block to the file. Use
 785                                    allocated block for that */
 786                                 unp = cpu_to_le32(allocated_block_nr);
 787                                 set_block_dev_mapped(bh_result,
 788                                                      allocated_block_nr, inode);
 789                                 set_buffer_new(bh_result);
 790                                 done = 1;
 791                         }
 792                         tmp_key = key;  // ;)
 793                         set_cpu_key_k_offset(&tmp_key, 1);
 794                         PATH_LAST_POSITION(&path)++;
 795
 796                         retval =
 797                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 798                                                  inode, (char *)&unp);
 799                         if (retval) {
 800                                 reiserfs_free_block(th, inode,
 801                                                     allocated_block_nr, 1);
 802                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 803                         }
 804                         //mark_tail_converted (inode);
 805                 } else if (is_direct_le_ih(ih)) {
 806                         /* direct item has to be converted */
 807                         loff_t tail_offset;
 808
 809                         tail_offset =
 810                             ((le_ih_k_offset(ih) -
 811                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 812                         if (tail_offset == cpu_key_k_offset(&key)) {
 813                                 /* direct item we just found fits into block we have
 814                                    to map. Convert it into unformatted node: use
 815                                    bh_result for the conversion */
 816                                 set_block_dev_mapped(bh_result,
 817                                                      allocated_block_nr, inode);
 818                                 unbh = bh_result;
 819                                 done = 1;
 820                         } else {
 821                                 /* we have to padd file tail stored in direct item(s)
 822                                    up to block size and convert it to unformatted
 823                                    node. FIXME: this should also get into page cache */
 824
 825                                 pathrelse(&path);
 826                                 /*
 827                                  * ugly, but we can only end the transaction if
 828                                  * we aren't nested
 829                                  */
 830                                 BUG_ON(!th->t_refcount);
 831                                 if (th->t_refcount == 1) {
 832                                         retval =
 833                                             reiserfs_end_persistent_transaction
 834                                             (th);
 835                                         th = NULL;
 836                                         if (retval)
 837                                                 goto failure;
 838                                 }
 839
 840                                 retval =
 841                                     convert_tail_for_hole(inode, bh_result,
 842                                                           tail_offset);
 843                                 if (retval) {
 844                                         if (retval != -ENOSPC)
 845                                                 reiserfs_error(inode->i_sb,
 846                                                         "clm-6004",
 847                                                         "convert tail failed "
 848                                                         "inode %lu, error %d",
 849                                                         inode->i_ino,
 850                                                         retval);
 851                                         if (allocated_block_nr) {
 852                                                 /* the bitmap, the super, and the stat data == 3 */
 853                                                 if (!th)
 854                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 855                                                 if (th)
 856                                                         reiserfs_free_block(th,
 857                                                                             inode,
 858                                                                             allocated_block_nr,
 859                                                                             1);
 860                                         }
 861                                         goto failure;
 862                                 }
 863                                 goto research;
 864                         }
 865                         retval =
 866                             direct2indirect(th, inode, &path, unbh,
 867                                             tail_offset);
 868                         if (retval) {
 869                                 reiserfs_unmap_buffer(unbh);
 870                                 reiserfs_free_block(th, inode,
 871                                                     allocated_block_nr, 1);
 872                                 goto failure;
 873                         }
 874                         /* it is important the set_buffer_uptodate is done after
 875                          ** the direct2indirect.  The buffer might contain valid
 876                          ** data newer than the data on disk (read by readpage, changed,
 877                          ** and then sent here by writepage).  direct2indirect needs
 878                          ** to know if unbh was already up to date, so it can decide
 879                          ** if the data in unbh needs to be replaced with data from
 880                          ** the disk
 881                          */
 882                         set_buffer_uptodate(unbh);
 883
 884                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 885                            buffer will disappear shortly, so it should not be added to
 886                          */
 887                         if (unbh->b_page) {
 888                                 /* we've converted the tail, so we must
 889                                  ** flush unbh before the transaction commits
 890                                  */
 891                                 reiserfs_add_tail_list(inode, unbh);
 892
 893                                 /* mark it dirty now to prevent commit_write from adding
 894                                  ** this buffer to the inode's dirty buffer list
 895                                  */
 896                                 /*
 897                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 898                                  * It's still atomic, but it sets the page dirty too,
 899                                  * which makes it eligible for writeback at any time by the
 900                                  * VM (which was also the case with __mark_buffer_dirty())
 901                                  */
 902                                 mark_buffer_dirty(unbh);
 903                         }
 904                 } else {
 905                         /* append indirect item with holes if needed, when appending
 906                            pointer to 'block'-th block use block, which is already
 907                            allocated */
 908                         struct cpu_key tmp_key;
 909                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 910                         // one block which is a fastpath
 911                         unp_t *un;
 912                         __u64 max_to_insert =
 913                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 914                             UNFM_P_SIZE;
 915                         __u64 blocks_needed;
 916
 917                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 918                                "vs-804: invalid position for append");
 919                         /* indirect item has to be appended, set up key of that position */
 920                         make_cpu_key(&tmp_key, inode,
 921                                      le_key_k_offset(version,
 922                                                      &(ih->ih_key)) +
 923                                      op_bytes_number(ih,
 924                                                      inode->i_sb->s_blocksize),
 925                                      //pos_in_item * inode->i_sb->s_blocksize,
 926                                      TYPE_INDIRECT, 3); // key type is unimportant
 927
 928                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 929                                "green-805: invalid offset");
 930                         blocks_needed =
 931                             1 +
 932                             ((cpu_key_k_offset(&key) -
 933                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 934                              s_blocksize_bits);
 935
 936                         if (blocks_needed == 1) {
 937                                 un = &unf_single;
 938                         } else {
 939                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 940                                 if (!un) {
 941                                         un = &unf_single;
 942                                         blocks_needed = 1;
 943                                         max_to_insert = 0;
 944                                 }
 945                         }
 946                         if (blocks_needed <= max_to_insert) {
 947                                 /* we are going to add target block to the file. Use allocated
 948                                    block for that */
 949                                 un[blocks_needed - 1] =
 950                                     cpu_to_le32(allocated_block_nr);
 951                                 set_block_dev_mapped(bh_result,
 952                                                      allocated_block_nr, inode);
 953                                 set_buffer_new(bh_result);
 954                                 done = 1;
 955                         } else {
 956                                 /* paste hole to the indirect item */
 957                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 958                                    only have space for one block */
 959                                 blocks_needed =
 960                                     max_to_insert ? max_to_insert : 1;
 961                         }
 962                         retval =
 963                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 964                                                      (char *)un,
 965                                                      UNFM_P_SIZE *
 966                                                      blocks_needed);
 967
 968                         if (blocks_needed != 1)
 969                                 kfree(un);
 970
 971                         if (retval) {
 972                                 reiserfs_free_block(th, inode,
 973                                                     allocated_block_nr, 1);
 974                                 goto failure;
 975                         }
 976                         if (!done) {
 977                                 /* We need to mark new file size in case this function will be
 978                                    interrupted/aborted later on. And we may do this only for
 979                                    holes. */
 980                                 inode->i_size +=
 981                                     inode->i_sb->s_blocksize * blocks_needed;
 982                         }
 983                 }
 984
 985                 if (done == 1)
 986                         break;
 987
 988                 /* this loop could log more blocks than we had originally asked
 989                  ** for.  So, we have to allow the transaction to end if it is
 990                  ** too big or too full.  Update the inode so things are
 991                  ** consistent if we crash before the function returns
 992                  **
 993                  ** release the path so that anybody waiting on the path before
 994                  ** ending their transaction will be able to continue.
 995                  */
 996                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
 997                         retval = restart_transaction(th, inode, &path);
 998                         if (retval)
 999                                 goto failure;
1000                 }
1001                 /*
1002                  * inserting indirect pointers for a hole can take a
1003                  * long time.  reschedule if needed and also release the write
1004                  * lock for others.
1005                  */
1006                 if (need_resched()) {
1007                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1008                         schedule();
1009                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1010                 }
1011
1012                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1013                 if (retval == IO_ERROR) {
1014                         retval = -EIO;
1015                         goto failure;
1016                 }
1017                 if (retval == POSITION_FOUND) {
1018                         reiserfs_warning(inode->i_sb, "vs-825",
1019                                          "%K should not be found", &key);
1020                         retval = -EEXIST;
1021                         if (allocated_block_nr)
1022                                 reiserfs_free_block(th, inode,
1023                                                     allocated_block_nr, 1);
1024                         pathrelse(&path);
1025                         goto failure;
1026                 }
1027                 bh = get_last_bh(&path);
1028                 ih = get_ih(&path);
1029                 item = get_item(&path);
1030                 pos_in_item = path.pos_in_item;
1031         } while (1);
1032
1033         retval = 0;
1034
1035       failure:
1036         if (th && (!dangle || (retval && !th->t_trans_id))) {
1037                 int err;
1038                 if (th->t_trans_id)
1039                         reiserfs_update_sd(th, inode);
1040                 err = reiserfs_end_persistent_transaction(th);
1041                 if (err)
1042                         retval = err;
1043         }
1044
1045         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1046         reiserfs_check_path(&path);
1047         return retval;
1048 }
1049
1050 static int
1051 reiserfs_readpages(struct file *file, struct address_space *mapping,
1052                    struct list_head *pages, unsigned nr_pages)
1053 {
1054         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1055 }
1056
1057 /* Compute real number of used bytes by file
1058  * Following three functions can go away when we'll have enough space in stat item
1059  */
1060 static int real_space_diff(struct inode *inode, int sd_size)
1061 {
1062         int bytes;
1063         loff_t blocksize = inode->i_sb->s_blocksize;
1064
1065         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1066                 return sd_size;
1067
1068         /* End of file is also in full block with indirect reference, so round
1069          ** up to the next block.
1070          **
1071          ** there is just no way to know if the tail is actually packed
1072          ** on the file, so we have to assume it isn't.  When we pack the
1073          ** tail, we add 4 bytes to pretend there really is an unformatted
1074          ** node pointer
1075          */
1076         bytes =
1077             ((inode->i_size +
1078               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1079             sd_size;
1080         return bytes;
1081 }
1082
1083 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1084                                         int sd_size)
1085 {
1086         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1087                 return inode->i_size +
1088                     (loff_t) (real_space_diff(inode, sd_size));
1089         }
1090         return ((loff_t) real_space_diff(inode, sd_size)) +
1091             (((loff_t) blocks) << 9);
1092 }
1093
1094 /* Compute number of blocks used by file in ReiserFS counting */
1095 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1096 {
1097         loff_t bytes = inode_get_bytes(inode);
1098         loff_t real_space = real_space_diff(inode, sd_size);
1099
1100         /* keeps fsck and non-quota versions of reiserfs happy */
1101         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1102                 bytes += (loff_t) 511;
1103         }
1104
1105         /* files from before the quota patch might i_blocks such that
1106          ** bytes < real_space.  Deal with that here to prevent it from
1107          ** going negative.
1108          */
1109         if (bytes < real_space)
1110                 return 0;
1111         return (bytes - real_space) >> 9;
1112 }
1113
1114 //
1115 // BAD: new directories have stat data of new type and all other items
1116 // of old type. Version stored in the inode says about body items, so
1117 // in update_stat_data we can not rely on inode, but have to check
1118 // item version directly
1119 //
1120
1121 // called by read_locked_inode
1122 static void init_inode(struct inode *inode, struct treepath *path)
1123 {
1124         struct buffer_head *bh;
1125         struct item_head *ih;
1126         __u32 rdev;
1127         //int version = ITEM_VERSION_1;
1128
1129         bh = PATH_PLAST_BUFFER(path);
1130         ih = PATH_PITEM_HEAD(path);
1131
1132         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1133
1134         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1135         REISERFS_I(inode)->i_flags = 0;
1136         REISERFS_I(inode)->i_prealloc_block = 0;
1137         REISERFS_I(inode)->i_prealloc_count = 0;
1138         REISERFS_I(inode)->i_trans_id = 0;
1139         REISERFS_I(inode)->i_jl = NULL;
1140         mutex_init(&(REISERFS_I(inode)->i_mmap));
1141         reiserfs_init_xattr_rwsem(inode);
1142
1143         if (stat_data_v1(ih)) {
1144                 struct stat_data_v1 *sd =
1145                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1146                 unsigned long blocks;
1147
1148                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1149                 set_inode_sd_version(inode, STAT_DATA_V1);
1150                 inode->i_mode = sd_v1_mode(sd);
1151                 inode->i_nlink = sd_v1_nlink(sd);
1152                 inode->i_uid = sd_v1_uid(sd);
1153                 inode->i_gid = sd_v1_gid(sd);
1154                 inode->i_size = sd_v1_size(sd);
1155                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1156                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1157                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1158                 inode->i_atime.tv_nsec = 0;
1159                 inode->i_ctime.tv_nsec = 0;
1160                 inode->i_mtime.tv_nsec = 0;
1161
1162                 inode->i_blocks = sd_v1_blocks(sd);
1163                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1164                 blocks = (inode->i_size + 511) >> 9;
1165                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1166                 if (inode->i_blocks > blocks) {
1167                         // there was a bug in <=3.5.23 when i_blocks could take negative
1168                         // values. Starting from 3.5.17 this value could even be stored in
1169                         // stat data. For such files we set i_blocks based on file
1170                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1171                         // only updated if file's inode will ever change
1172                         inode->i_blocks = blocks;
1173                 }
1174
1175                 rdev = sd_v1_rdev(sd);
1176                 REISERFS_I(inode)->i_first_direct_byte =
1177                     sd_v1_first_direct_byte(sd);
1178                 /* an early bug in the quota code can give us an odd number for the
1179                  ** block count.  This is incorrect, fix it here.
1180                  */
1181                 if (inode->i_blocks & 1) {
1182                         inode->i_blocks++;
1183                 }
1184                 inode_set_bytes(inode,
1185                                 to_real_used_space(inode, inode->i_blocks,
1186                                                    SD_V1_SIZE));
1187                 /* nopack is initially zero for v1 objects. For v2 objects,
1188                    nopack is initialised from sd_attrs */
1189                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1190         } else {
1191                 // new stat data found, but object may have old items
1192                 // (directories and symlinks)
1193                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1194
1195                 inode->i_mode = sd_v2_mode(sd);
1196                 inode->i_nlink = sd_v2_nlink(sd);
1197                 inode->i_uid = sd_v2_uid(sd);
1198                 inode->i_size = sd_v2_size(sd);
1199                 inode->i_gid = sd_v2_gid(sd);
1200                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1201                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1202                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1203                 inode->i_ctime.tv_nsec = 0;
1204                 inode->i_mtime.tv_nsec = 0;
1205                 inode->i_atime.tv_nsec = 0;
1206                 inode->i_blocks = sd_v2_blocks(sd);
1207                 rdev = sd_v2_rdev(sd);
1208                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1209                         inode->i_generation =
1210                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1211                 else
1212                         inode->i_generation = sd_v2_generation(sd);
1213
1214                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1215                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1216                 else
1217                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1218                 REISERFS_I(inode)->i_first_direct_byte = 0;
1219                 set_inode_sd_version(inode, STAT_DATA_V2);
1220                 inode_set_bytes(inode,
1221                                 to_real_used_space(inode, inode->i_blocks,
1222                                                    SD_V2_SIZE));
1223                 /* read persistent inode attributes from sd and initalise
1224                    generic inode flags from them */
1225                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1226                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1227         }
1228
1229         pathrelse(path);
1230         if (S_ISREG(inode->i_mode)) {
1231                 inode->i_op = &reiserfs_file_inode_operations;
1232                 inode->i_fop = &reiserfs_file_operations;
1233                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1234         } else if (S_ISDIR(inode->i_mode)) {
1235                 inode->i_op = &reiserfs_dir_inode_operations;
1236                 inode->i_fop = &reiserfs_dir_operations;
1237         } else if (S_ISLNK(inode->i_mode)) {
1238                 inode->i_op = &reiserfs_symlink_inode_operations;
1239                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1240         } else {
1241                 inode->i_blocks = 0;
1242                 inode->i_op = &reiserfs_special_inode_operations;
1243                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1244         }
1245 }
1246
1247 // update new stat data with inode fields
1248 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1249 {
1250         struct stat_data *sd_v2 = (struct stat_data *)sd;
1251         __u16 flags;
1252
1253         set_sd_v2_mode(sd_v2, inode->i_mode);
1254         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1255         set_sd_v2_uid(sd_v2, inode->i_uid);
1256         set_sd_v2_size(sd_v2, size);
1257         set_sd_v2_gid(sd_v2, inode->i_gid);
1258         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1259         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1260         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1261         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1262         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1263                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1264         else
1265                 set_sd_v2_generation(sd_v2, inode->i_generation);
1266         flags = REISERFS_I(inode)->i_attrs;
1267         i_attrs_to_sd_attrs(inode, &flags);
1268         set_sd_v2_attrs(sd_v2, flags);
1269 }
1270
1271 // used to copy inode's fields to old stat data
1272 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1273 {
1274         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1275
1276         set_sd_v1_mode(sd_v1, inode->i_mode);
1277         set_sd_v1_uid(sd_v1, inode->i_uid);
1278         set_sd_v1_gid(sd_v1, inode->i_gid);
1279         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1280         set_sd_v1_size(sd_v1, size);
1281         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1282         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1283         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1284
1285         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1286                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1287         else
1288                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1289
1290         // Sigh. i_first_direct_byte is back
1291         set_sd_v1_first_direct_byte(sd_v1,
1292                                     REISERFS_I(inode)->i_first_direct_byte);
1293 }
1294
1295 /* NOTE, you must prepare the buffer head before sending it here,
1296 ** and then log it after the call
1297 */
1298 static void update_stat_data(struct treepath *path, struct inode *inode,
1299                              loff_t size)
1300 {
1301         struct buffer_head *bh;
1302         struct item_head *ih;
1303
1304         bh = PATH_PLAST_BUFFER(path);
1305         ih = PATH_PITEM_HEAD(path);
1306
1307         if (!is_statdata_le_ih(ih))
1308                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1309                                INODE_PKEY(inode), ih);
1310
1311         if (stat_data_v1(ih)) {
1312                 // path points to old stat data
1313                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1314         } else {
1315                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1316         }
1317
1318         return;
1319 }
1320
1321 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1322                              struct inode *inode, loff_t size)
1323 {
1324         struct cpu_key key;
1325         INITIALIZE_PATH(path);
1326         struct buffer_head *bh;
1327         int fs_gen;
1328         struct item_head *ih, tmp_ih;
1329         int retval;
1330
1331         BUG_ON(!th->t_trans_id);
1332
1333         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1334
1335         for (;;) {
1336                 int pos;
1337                 /* look for the object's stat data */
1338                 retval = search_item(inode->i_sb, &key, &path);
1339                 if (retval == IO_ERROR) {
1340                         reiserfs_error(inode->i_sb, "vs-13050",
1341                                        "i/o failure occurred trying to "
1342                                        "update %K stat data", &key);
1343                         return;
1344                 }
1345                 if (retval == ITEM_NOT_FOUND) {
1346                         pos = PATH_LAST_POSITION(&path);
1347                         pathrelse(&path);
1348                         if (inode->i_nlink == 0) {
1349                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1350                                 return;
1351                         }
1352                         reiserfs_warning(inode->i_sb, "vs-13060",
1353                                          "stat data of object %k (nlink == %d) "
1354                                          "not found (pos %d)",
1355                                          INODE_PKEY(inode), inode->i_nlink,
1356                                          pos);
1357                         reiserfs_check_path(&path);
1358                         return;
1359                 }
1360
1361                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1362                  ** FS might change.  We have to detect that, and loop back to the
1363                  ** search if the stat data item has moved
1364                  */
1365                 bh = get_last_bh(&path);
1366                 ih = get_ih(&path);
1367                 copy_item_head(&tmp_ih, ih);
1368                 fs_gen = get_generation(inode->i_sb);
1369                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1370                 if (fs_changed(fs_gen, inode->i_sb)
1371                     && item_moved(&tmp_ih, &path)) {
1372                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1373                         continue;       /* Stat_data item has been moved after scheduling. */
1374                 }
1375                 break;
1376         }
1377         update_stat_data(&path, inode, size);
1378         journal_mark_dirty(th, th->t_super, bh);
1379         pathrelse(&path);
1380         return;
1381 }
1382
1383 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1384 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1385 ** and clear the key in the private portion of the inode, otherwise a
1386 ** corresponding iput might try to delete whatever object the inode last
1387 ** represented.
1388 */
1389 static void reiserfs_make_bad_inode(struct inode *inode)
1390 {
1391         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1392         make_bad_inode(inode);
1393 }
1394
1395 //
1396 // initially this function was derived from minix or ext2's analog and
1397 // evolved as the prototype did
1398 //
1399
1400 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1401 {
1402         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1403         inode->i_ino = args->objectid;
1404         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1405         return 0;
1406 }
1407
1408 /* looks for stat data in the tree, and fills up the fields of in-core
1409    inode stat data fields */
1410 void reiserfs_read_locked_inode(struct inode *inode,
1411                                 struct reiserfs_iget_args *args)
1412 {
1413         INITIALIZE_PATH(path_to_sd);
1414         struct cpu_key key;
1415         unsigned long dirino;
1416         int retval;
1417
1418         dirino = args->dirid;
1419
1420         /* set version 1, version 2 could be used too, because stat data
1421            key is the same in both versions */
1422         key.version = KEY_FORMAT_3_5;
1423         key.on_disk_key.k_dir_id = dirino;
1424         key.on_disk_key.k_objectid = inode->i_ino;
1425         key.on_disk_key.k_offset = 0;
1426         key.on_disk_key.k_type = 0;
1427
1428         /* look for the object's stat data */
1429         retval = search_item(inode->i_sb, &key, &path_to_sd);
1430         if (retval == IO_ERROR) {
1431                 reiserfs_error(inode->i_sb, "vs-13070",
1432                                "i/o failure occurred trying to find "
1433                                "stat data of %K", &key);
1434                 reiserfs_make_bad_inode(inode);
1435                 return;
1436         }
1437         if (retval != ITEM_FOUND) {
1438                 /* a stale NFS handle can trigger this without it being an error */
1439                 pathrelse(&path_to_sd);
1440                 reiserfs_make_bad_inode(inode);
1441                 inode->i_nlink = 0;
1442                 return;
1443         }
1444
1445         init_inode(inode, &path_to_sd);
1446
1447         /* It is possible that knfsd is trying to access inode of a file
1448            that is being removed from the disk by some other thread. As we
1449            update sd on unlink all that is required is to check for nlink
1450            here. This bug was first found by Sizif when debugging
1451            SquidNG/Butterfly, forgotten, and found again after Philippe
1452            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1453
1454            More logical fix would require changes in fs/inode.c:iput() to
1455            remove inode from hash-table _after_ fs cleaned disk stuff up and
1456            in iget() to return NULL if I_FREEING inode is found in
1457            hash-table. */
1458         /* Currently there is one place where it's ok to meet inode with
1459            nlink==0: processing of open-unlinked and half-truncated files
1460            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1461         if ((inode->i_nlink == 0) &&
1462             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1463                 reiserfs_warning(inode->i_sb, "vs-13075",
1464                                  "dead inode read from disk %K. "
1465                                  "This is likely to be race with knfsd. Ignore",
1466                                  &key);
1467                 reiserfs_make_bad_inode(inode);
1468         }
1469
1470         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1471
1472 }
1473
1474 /**
1475  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1476  *
1477  * @inode:    inode from hash table to check
1478  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1479  *
1480  * This function is called by iget5_locked() to distinguish reiserfs inodes
1481  * having the same inode numbers. Such inodes can only exist due to some
1482  * error condition. One of them should be bad. Inodes with identical
1483  * inode numbers (objectids) are distinguished by parent directory ids.
1484  *
1485  */
1486 int reiserfs_find_actor(struct inode *inode, void *opaque)
1487 {
1488         struct reiserfs_iget_args *args;
1489
1490         args = opaque;
1491         /* args is already in CPU order */
1492         return (inode->i_ino == args->objectid) &&
1493             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1494 }
1495
1496 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1497 {
1498         struct inode *inode;
1499         struct reiserfs_iget_args args;
1500
1501         args.objectid = key->on_disk_key.k_objectid;
1502         args.dirid = key->on_disk_key.k_dir_id;
1503         reiserfs_write_unlock(s);
1504         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1505                              reiserfs_find_actor, reiserfs_init_locked_inode,
1506                              (void *)(&args));
1507         reiserfs_write_lock(s);
1508         if (!inode)
1509                 return ERR_PTR(-ENOMEM);
1510
1511         if (inode->i_state & I_NEW) {
1512                 reiserfs_read_locked_inode(inode, &args);
1513                 unlock_new_inode(inode);
1514         }
1515
1516         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1517                 /* either due to i/o error or a stale NFS handle */
1518                 iput(inode);
1519                 inode = NULL;
1520         }
1521         return inode;
1522 }
1523
1524 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1525         u32 objectid, u32 dir_id, u32 generation)
1526
1527 {
1528         struct cpu_key key;
1529         struct inode *inode;
1530
1531         key.on_disk_key.k_objectid = objectid;
1532         key.on_disk_key.k_dir_id = dir_id;
1533         reiserfs_write_lock(sb);
1534         inode = reiserfs_iget(sb, &key);
1535         if (inode && !IS_ERR(inode) && generation != 0 &&
1536             generation != inode->i_generation) {
1537                 iput(inode);
1538                 inode = NULL;
1539         }
1540         reiserfs_write_unlock(sb);
1541
1542         return d_obtain_alias(inode);
1543 }
1544
1545 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1546                 int fh_len, int fh_type)
1547 {
1548         /* fhtype happens to reflect the number of u32s encoded.
1549          * due to a bug in earlier code, fhtype might indicate there
1550          * are more u32s then actually fitted.
1551          * so if fhtype seems to be more than len, reduce fhtype.
1552          * Valid types are:
1553          *   2 - objectid + dir_id - legacy support
1554          *   3 - objectid + dir_id + generation
1555          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1556          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1557          *   6 - as above plus generation of directory
1558          * 6 does not fit in NFSv2 handles
1559          */
1560         if (fh_type > fh_len) {
1561                 if (fh_type != 6 || fh_len != 5)
1562                         reiserfs_warning(sb, "reiserfs-13077",
1563                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1564                                 fh_type, fh_len);
1565                 fh_type = 5;
1566         }
1567
1568         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1569                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1570 }
1571
1572 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1573                 int fh_len, int fh_type)
1574 {
1575         if (fh_type < 4)
1576                 return NULL;
1577
1578         return reiserfs_get_dentry(sb,
1579                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1580                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1581                 (fh_type == 6) ? fid->raw[5] : 0);
1582 }
1583
1584 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1585                        int need_parent)
1586 {
1587         struct inode *inode = dentry->d_inode;
1588         int maxlen = *lenp;
1589
1590         if (maxlen < 3)
1591                 return 255;
1592
1593         data[0] = inode->i_ino;
1594         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1595         data[2] = inode->i_generation;
1596         *lenp = 3;
1597         /* no room for directory info? return what we've stored so far */
1598         if (maxlen < 5 || !need_parent)
1599                 return 3;
1600
1601         spin_lock(&dentry->d_lock);
1602         inode = dentry->d_parent->d_inode;
1603         data[3] = inode->i_ino;
1604         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1605         *lenp = 5;
1606         if (maxlen >= 6) {
1607                 data[5] = inode->i_generation;
1608                 *lenp = 6;
1609         }
1610         spin_unlock(&dentry->d_lock);
1611         return *lenp;
1612 }
1613
1614 /* looks for stat data, then copies fields to it, marks the buffer
1615    containing stat data as dirty */
1616 /* reiserfs inodes are never really dirty, since the dirty inode call
1617 ** always logs them.  This call allows the VFS inode marking routines
1618 ** to properly mark inodes for datasync and such, but only actually
1619 ** does something when called for a synchronous update.
1620 */
1621 int reiserfs_write_inode(struct inode *inode, int do_sync)
1622 {
1623         struct reiserfs_transaction_handle th;
1624         int jbegin_count = 1;
1625
1626         if (inode->i_sb->s_flags & MS_RDONLY)
1627                 return -EROFS;
1628         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1629          ** these cases are just when the system needs ram, not when the
1630          ** inode needs to reach disk for safety, and they can safely be
1631          ** ignored because the altered inode has already been logged.
1632          */
1633         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1634                 reiserfs_write_lock(inode->i_sb);
1635                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1636                         reiserfs_update_sd(&th, inode);
1637                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1638                 }
1639                 reiserfs_write_unlock(inode->i_sb);
1640         }
1641         return 0;
1642 }
1643
1644 /* stat data of new object is inserted already, this inserts the item
1645    containing "." and ".." entries */
1646 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1647                                   struct inode *inode,
1648                                   struct item_head *ih, struct treepath *path,
1649                                   struct inode *dir)
1650 {
1651         struct super_block *sb = th->t_super;
1652         char empty_dir[EMPTY_DIR_SIZE];
1653         char *body = empty_dir;
1654         struct cpu_key key;
1655         int retval;
1656
1657         BUG_ON(!th->t_trans_id);
1658
1659         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1660                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1661                       TYPE_DIRENTRY, 3 /*key length */ );
1662
1663         /* compose item head for new item. Directories consist of items of
1664            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1665            is done by reiserfs_new_inode */
1666         if (old_format_only(sb)) {
1667                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1668                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1669
1670                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1671                                        ih->ih_key.k_objectid,
1672                                        INODE_PKEY(dir)->k_dir_id,
1673                                        INODE_PKEY(dir)->k_objectid);
1674         } else {
1675                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1676                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1677
1678                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1679                                     ih->ih_key.k_objectid,
1680                                     INODE_PKEY(dir)->k_dir_id,
1681                                     INODE_PKEY(dir)->k_objectid);
1682         }
1683
1684         /* look for place in the tree for new item */
1685         retval = search_item(sb, &key, path);
1686         if (retval == IO_ERROR) {
1687                 reiserfs_error(sb, "vs-13080",
1688                                "i/o failure occurred creating new directory");
1689                 return -EIO;
1690         }
1691         if (retval == ITEM_FOUND) {
1692                 pathrelse(path);
1693                 reiserfs_warning(sb, "vs-13070",
1694                                  "object with this key exists (%k)",
1695                                  &(ih->ih_key));
1696                 return -EEXIST;
1697         }
1698
1699         /* insert item, that is empty directory item */
1700         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1701 }
1702
1703 /* stat data of object has been inserted, this inserts the item
1704    containing the body of symlink */
1705 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1706                                 struct item_head *ih,
1707                                 struct treepath *path, const char *symname,
1708                                 int item_len)
1709 {
1710         struct super_block *sb = th->t_super;
1711         struct cpu_key key;
1712         int retval;
1713
1714         BUG_ON(!th->t_trans_id);
1715
1716         _make_cpu_key(&key, KEY_FORMAT_3_5,
1717                       le32_to_cpu(ih->ih_key.k_dir_id),
1718                       le32_to_cpu(ih->ih_key.k_objectid),
1719                       1, TYPE_DIRECT, 3 /*key length */ );
1720
1721         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1722                           0 /*free_space */ );
1723
1724         /* look for place in the tree for new item */
1725         retval = search_item(sb, &key, path);
1726         if (retval == IO_ERROR) {
1727                 reiserfs_error(sb, "vs-13080",
1728                                "i/o failure occurred creating new symlink");
1729                 return -EIO;
1730         }
1731         if (retval == ITEM_FOUND) {
1732                 pathrelse(path);
1733                 reiserfs_warning(sb, "vs-13080",
1734                                  "object with this key exists (%k)",
1735                                  &(ih->ih_key));
1736                 return -EEXIST;
1737         }
1738
1739         /* insert item, that is body of symlink */
1740         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1741 }
1742
1743 /* inserts the stat data into the tree, and then calls
1744    reiserfs_new_directory (to insert ".", ".." item if new object is
1745    directory) or reiserfs_new_symlink (to insert symlink body if new
1746    object is symlink) or nothing (if new object is regular file)
1747
1748    NOTE! uid and gid must already be set in the inode.  If we return
1749    non-zero due to an error, we have to drop the quota previously allocated
1750    for the fresh inode.  This can only be done outside a transaction, so
1751    if we return non-zero, we also end the transaction.  */
1752 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1753                        struct inode *dir, int mode, const char *symname,
1754                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1755                           strlen (symname) for symlinks) */
1756                        loff_t i_size, struct dentry *dentry,
1757                        struct inode *inode,
1758                        struct reiserfs_security_handle *security)
1759 {
1760         struct super_block *sb;
1761         struct reiserfs_iget_args args;
1762         INITIALIZE_PATH(path_to_key);
1763         struct cpu_key key;
1764         struct item_head ih;
1765         struct stat_data sd;
1766         int retval;
1767         int err;
1768
1769         BUG_ON(!th->t_trans_id);
1770
1771         vfs_dq_init(inode);
1772         err = dquot_alloc_inode(inode);
1773         if (err)
1774                 goto out_end_trans;
1775         if (!dir->i_nlink) {
1776                 err = -EPERM;
1777                 goto out_bad_inode;
1778         }
1779
1780         sb = dir->i_sb;
1781
1782         /* item head of new item */
1783         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1784         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1785         if (!ih.ih_key.k_objectid) {
1786                 err = -ENOMEM;
1787                 goto out_bad_inode;
1788         }
1789         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1790         if (old_format_only(sb))
1791                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1792                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1793         else
1794                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1795                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1796         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1797         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1798         if (insert_inode_locked4(inode, args.objectid,
1799                              reiserfs_find_actor, &args) < 0) {
1800                 err = -EINVAL;
1801                 goto out_bad_inode;
1802         }
1803         if (old_format_only(sb))
1804                 /* not a perfect generation count, as object ids can be reused, but
1805                  ** this is as good as reiserfs can do right now.
1806                  ** note that the private part of inode isn't filled in yet, we have
1807                  ** to use the directory.
1808                  */
1809                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1810         else
1811 #if defined( USE_INODE_GENERATION_COUNTER )
1812                 inode->i_generation =
1813                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1814 #else
1815                 inode->i_generation = ++event;
1816 #endif
1817
1818         /* fill stat data */
1819         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1820
1821         /* uid and gid must already be set by the caller for quota init */
1822
1823         /* symlink cannot be immutable or append only, right? */
1824         if (S_ISLNK(inode->i_mode))
1825                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1826
1827         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1828         inode->i_size = i_size;
1829         inode->i_blocks = 0;
1830         inode->i_bytes = 0;
1831         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1832             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1833
1834         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1835         REISERFS_I(inode)->i_flags = 0;
1836         REISERFS_I(inode)->i_prealloc_block = 0;
1837         REISERFS_I(inode)->i_prealloc_count = 0;
1838         REISERFS_I(inode)->i_trans_id = 0;
1839         REISERFS_I(inode)->i_jl = NULL;
1840         REISERFS_I(inode)->i_attrs =
1841             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1842         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1843         mutex_init(&(REISERFS_I(inode)->i_mmap));
1844         reiserfs_init_xattr_rwsem(inode);
1845
1846         /* key to search for correct place for new stat data */
1847         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1848                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1849                       TYPE_STAT_DATA, 3 /*key length */ );
1850
1851         /* find proper place for inserting of stat data */
1852         retval = search_item(sb, &key, &path_to_key);
1853         if (retval == IO_ERROR) {
1854                 err = -EIO;
1855                 goto out_bad_inode;
1856         }
1857         if (retval == ITEM_FOUND) {
1858                 pathrelse(&path_to_key);
1859                 err = -EEXIST;
1860                 goto out_bad_inode;
1861         }
1862         if (old_format_only(sb)) {
1863                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1864                         pathrelse(&path_to_key);
1865                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1866                         err = -EINVAL;
1867                         goto out_bad_inode;
1868                 }
1869                 inode2sd_v1(&sd, inode, inode->i_size);
1870         } else {
1871                 inode2sd(&sd, inode, inode->i_size);
1872         }
1873         // store in in-core inode the key of stat data and version all
1874         // object items will have (directory items will have old offset
1875         // format, other new objects will consist of new items)
1876         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1877                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1878         else
1879                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1880         if (old_format_only(sb))
1881                 set_inode_sd_version(inode, STAT_DATA_V1);
1882         else
1883                 set_inode_sd_version(inode, STAT_DATA_V2);
1884
1885         /* insert the stat data into the tree */
1886 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1887         if (REISERFS_I(dir)->new_packing_locality)
1888                 th->displace_new_blocks = 1;
1889 #endif
1890         retval =
1891             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1892                                  (char *)(&sd));
1893         if (retval) {
1894                 err = retval;
1895                 reiserfs_check_path(&path_to_key);
1896                 goto out_bad_inode;
1897         }
1898 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1899         if (!th->displace_new_blocks)
1900                 REISERFS_I(dir)->new_packing_locality = 0;
1901 #endif
1902         if (S_ISDIR(mode)) {
1903                 /* insert item with "." and ".." */
1904                 retval =
1905                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1906         }
1907
1908         if (S_ISLNK(mode)) {
1909                 /* insert body of symlink */
1910                 if (!old_format_only(sb))
1911                         i_size = ROUND_UP(i_size);
1912                 retval =
1913                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1914                                          i_size);
1915         }
1916         if (retval) {
1917                 err = retval;
1918                 reiserfs_check_path(&path_to_key);
1919                 journal_end(th, th->t_super, th->t_blocks_allocated);
1920                 goto out_inserted_sd;
1921         }
1922
1923         if (reiserfs_posixacl(inode->i_sb)) {
1924                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1925                 if (retval) {
1926                         err = retval;
1927                         reiserfs_check_path(&path_to_key);
1928                         journal_end(th, th->t_super, th->t_blocks_allocated);
1929                         goto out_inserted_sd;
1930                 }
1931         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1932                 reiserfs_warning(inode->i_sb, "jdm-13090",
1933                                  "ACLs aren't enabled in the fs, "
1934                                  "but vfs thinks they are!");
1935         } else if (IS_PRIVATE(dir))
1936                 inode->i_flags |= S_PRIVATE;
1937
1938         if (security->name) {
1939                 retval = reiserfs_security_write(th, inode, security);
1940                 if (retval) {
1941                         err = retval;
1942                         reiserfs_check_path(&path_to_key);
1943                         retval = journal_end(th, th->t_super,
1944                                              th->t_blocks_allocated);
1945                         if (retval)
1946                                 err = retval;
1947                         goto out_inserted_sd;
1948                 }
1949         }
1950
1951         reiserfs_update_sd(th, inode);
1952         reiserfs_check_path(&path_to_key);
1953
1954         return 0;
1955
1956 /* it looks like you can easily compress these two goto targets into
1957  * one.  Keeping it like this doesn't actually hurt anything, and they
1958  * are place holders for what the quota code actually needs.
1959  */
1960       out_bad_inode:
1961         /* Invalidate the object, nothing was inserted yet */
1962         INODE_PKEY(inode)->k_objectid = 0;
1963
1964         /* Quota change must be inside a transaction for journaling */
1965         dquot_free_inode(inode);
1966
1967       out_end_trans:
1968         journal_end(th, th->t_super, th->t_blocks_allocated);
1969         /* Drop can be outside and it needs more credits so it's better to have it outside */
1970         dquot_drop(inode);
1971         inode->i_flags |= S_NOQUOTA;
1972         make_bad_inode(inode);
1973
1974       out_inserted_sd:
1975         inode->i_nlink = 0;
1976         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1977         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1978         iput(inode);
1979         return err;
1980 }
1981
1982 /*
1983 ** finds the tail page in the page cache,
1984 ** reads the last block in.
1985 **
1986 ** On success, page_result is set to a locked, pinned page, and bh_result
1987 ** is set to an up to date buffer for the last block in the file.  returns 0.
1988 **
1989 ** tail conversion is not done, so bh_result might not be valid for writing
1990 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1991 ** trying to write the block.
1992 **
1993 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1994 */
1995 static int grab_tail_page(struct inode *inode,
1996                           struct page **page_result,
1997                           struct buffer_head **bh_result)
1998 {
1999
2000         /* we want the page with the last byte in the file,
2001          ** not the page that will hold the next byte for appending
2002          */
2003         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2004         unsigned long pos = 0;
2005         unsigned long start = 0;
2006         unsigned long blocksize = inode->i_sb->s_blocksize;
2007         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2008         struct buffer_head *bh;
2009         struct buffer_head *head;
2010         struct page *page;
2011         int error;
2012
2013         /* we know that we are only called with inode->i_size > 0.
2014          ** we also know that a file tail can never be as big as a block
2015          ** If i_size % blocksize == 0, our file is currently block aligned
2016          ** and it won't need converting or zeroing after a truncate.
2017          */
2018         if ((offset & (blocksize - 1)) == 0) {
2019                 return -ENOENT;
2020         }
2021         page = grab_cache_page(inode->i_mapping, index);
2022         error = -ENOMEM;
2023         if (!page) {
2024                 goto out;
2025         }
2026         /* start within the page of the last block in the file */
2027         start = (offset / blocksize) * blocksize;
2028
2029         error = block_prepare_write(page, start, offset,
2030                                     reiserfs_get_block_create_0);
2031         if (error)
2032                 goto unlock;
2033
2034         head = page_buffers(page);
2035         bh = head;
2036         do {
2037                 if (pos >= start) {
2038                         break;
2039                 }
2040                 bh = bh->b_this_page;
2041                 pos += blocksize;
2042         } while (bh != head);
2043
2044         if (!buffer_uptodate(bh)) {
2045                 /* note, this should never happen, prepare_write should
2046                  ** be taking care of this for us.  If the buffer isn't up to date,
2047                  ** I've screwed up the code to find the buffer, or the code to
2048                  ** call prepare_write
2049                  */
2050                 reiserfs_error(inode->i_sb, "clm-6000",
2051                                "error reading block %lu", bh->b_blocknr);
2052                 error = -EIO;
2053                 goto unlock;
2054         }
2055         *bh_result = bh;
2056         *page_result = page;
2057
2058       out:
2059         return error;
2060
2061       unlock:
2062         unlock_page(page);
2063         page_cache_release(page);
2064         return error;
2065 }
2066
2067 /*
2068 ** vfs version of truncate file.  Must NOT be called with
2069 ** a transaction already started.
2070 **
2071 ** some code taken from block_truncate_page
2072 */
2073 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2074 {
2075         struct reiserfs_transaction_handle th;
2076         /* we want the offset for the first byte after the end of the file */
2077         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2078         unsigned blocksize = inode->i_sb->s_blocksize;
2079         unsigned length;
2080         struct page *page = NULL;
2081         int error;
2082         struct buffer_head *bh = NULL;
2083         int err2;
2084         int lock_depth;
2085
2086         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2087
2088         if (inode->i_size > 0) {
2089                 error = grab_tail_page(inode, &page, &bh);
2090                 if (error) {
2091                         // -ENOENT means we truncated past the end of the file,
2092                         // and get_block_create_0 could not find a block to read in,
2093                         // which is ok.
2094                         if (error != -ENOENT)
2095                                 reiserfs_error(inode->i_sb, "clm-6001",
2096                                                "grab_tail_page failed %d",
2097                                                error);
2098                         page = NULL;
2099                         bh = NULL;
2100                 }
2101         }
2102
2103         /* so, if page != NULL, we have a buffer head for the offset at
2104          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2105          ** then we have an unformatted node.  Otherwise, we have a direct item,
2106          ** and no zeroing is required on disk.  We zero after the truncate,
2107          ** because the truncate might pack the item anyway
2108          ** (it will unmap bh if it packs).
2109          */
2110         /* it is enough to reserve space in transaction for 2 balancings:
2111            one for "save" link adding and another for the first
2112            cut_from_item. 1 is for update_sd */
2113         error = journal_begin(&th, inode->i_sb,
2114                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2115         if (error)
2116                 goto out;
2117         reiserfs_update_inode_transaction(inode);
2118         if (update_timestamps)
2119                 /* we are doing real truncate: if the system crashes before the last
2120                    transaction of truncating gets committed - on reboot the file
2121                    either appears truncated properly or not truncated at all */
2122                 add_save_link(&th, inode, 1);
2123         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2124         error =
2125             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2126         if (error)
2127                 goto out;
2128
2129         /* check reiserfs_do_truncate after ending the transaction */
2130         if (err2) {
2131                 error = err2;
2132                 goto out;
2133         }
2134
2135         if (update_timestamps) {
2136                 error = remove_save_link(inode, 1 /* truncate */);
2137                 if (error)
2138                         goto out;
2139         }
2140
2141         if (page) {
2142                 length = offset & (blocksize - 1);
2143                 /* if we are not on a block boundary */
2144                 if (length) {
2145                         length = blocksize - length;
2146                         zero_user(page, offset, length);
2147                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2148                                 mark_buffer_dirty(bh);
2149                         }
2150                 }
2151                 unlock_page(page);
2152                 page_cache_release(page);
2153         }
2154
2155         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2156
2157         return 0;
2158       out:
2159         if (page) {
2160                 unlock_page(page);
2161                 page_cache_release(page);
2162         }
2163
2164         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2165
2166         return error;
2167 }
2168
2169 static int map_block_for_writepage(struct inode *inode,
2170                                    struct buffer_head *bh_result,
2171                                    unsigned long block)
2172 {
2173         struct reiserfs_transaction_handle th;
2174         int fs_gen;
2175         struct item_head tmp_ih;
2176         struct item_head *ih;
2177         struct buffer_head *bh;
2178         __le32 *item;
2179         struct cpu_key key;
2180         INITIALIZE_PATH(path);
2181         int pos_in_item;
2182         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2183         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2184         int retval;
2185         int use_get_block = 0;
2186         int bytes_copied = 0;
2187         int copy_size;
2188         int trans_running = 0;
2189
2190         /* catch places below that try to log something without starting a trans */
2191         th.t_trans_id = 0;
2192
2193         if (!buffer_uptodate(bh_result)) {
2194                 return -EIO;
2195         }
2196
2197         kmap(bh_result->b_page);
2198       start_over:
2199         reiserfs_write_lock(inode->i_sb);
2200         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2201
2202       research:
2203         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2204         if (retval != POSITION_FOUND) {
2205                 use_get_block = 1;
2206                 goto out;
2207         }
2208
2209         bh = get_last_bh(&path);
2210         ih = get_ih(&path);
2211         item = get_item(&path);
2212         pos_in_item = path.pos_in_item;
2213
2214         /* we've found an unformatted node */
2215         if (indirect_item_found(retval, ih)) {
2216                 if (bytes_copied > 0) {
2217                         reiserfs_warning(inode->i_sb, "clm-6002",
2218                                          "bytes_copied %d", bytes_copied);
2219                 }
2220                 if (!get_block_num(item, pos_in_item)) {
2221                         /* crap, we are writing to a hole */
2222                         use_get_block = 1;
2223                         goto out;
2224                 }
2225                 set_block_dev_mapped(bh_result,
2226                                      get_block_num(item, pos_in_item), inode);
2227         } else if (is_direct_le_ih(ih)) {
2228                 char *p;
2229                 p = page_address(bh_result->b_page);
2230                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2231                 copy_size = ih_item_len(ih) - pos_in_item;
2232
2233                 fs_gen = get_generation(inode->i_sb);
2234                 copy_item_head(&tmp_ih, ih);
2235
2236                 if (!trans_running) {
2237                         /* vs-3050 is gone, no need to drop the path */
2238                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2239                         if (retval)
2240                                 goto out;
2241                         reiserfs_update_inode_transaction(inode);
2242                         trans_running = 1;
2243                         if (fs_changed(fs_gen, inode->i_sb)
2244                             && item_moved(&tmp_ih, &path)) {
2245                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2246                                                                  bh);
2247                                 goto research;
2248                         }
2249                 }
2250
2251                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2252
2253                 if (fs_changed(fs_gen, inode->i_sb)
2254                     && item_moved(&tmp_ih, &path)) {
2255                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2256                         goto research;
2257                 }
2258
2259                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2260                        copy_size);
2261
2262                 journal_mark_dirty(&th, inode->i_sb, bh);
2263                 bytes_copied += copy_size;
2264                 set_block_dev_mapped(bh_result, 0, inode);
2265
2266                 /* are there still bytes left? */
2267                 if (bytes_copied < bh_result->b_size &&
2268                     (byte_offset + bytes_copied) < inode->i_size) {
2269                         set_cpu_key_k_offset(&key,
2270                                              cpu_key_k_offset(&key) +
2271                                              copy_size);
2272                         goto research;
2273                 }
2274         } else {
2275                 reiserfs_warning(inode->i_sb, "clm-6003",
2276                                  "bad item inode %lu", inode->i_ino);
2277                 retval = -EIO;
2278                 goto out;
2279         }
2280         retval = 0;
2281
2282       out:
2283         pathrelse(&path);
2284         if (trans_running) {
2285                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2286                 if (err)
2287                         retval = err;
2288                 trans_running = 0;
2289         }
2290         reiserfs_write_unlock(inode->i_sb);
2291
2292         /* this is where we fill in holes in the file. */
2293         if (use_get_block) {
2294                 retval = reiserfs_get_block(inode, block, bh_result,
2295                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2296                                             | GET_BLOCK_NO_DANGLE);
2297                 if (!retval) {
2298                         if (!buffer_mapped(bh_result)
2299                             || bh_result->b_blocknr == 0) {
2300                                 /* get_block failed to find a mapped unformatted node. */
2301                                 use_get_block = 0;
2302                                 goto start_over;
2303                         }
2304                 }
2305         }
2306         kunmap(bh_result->b_page);
2307
2308         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2309                 /* we've copied data from the page into the direct item, so the
2310                  * buffer in the page is now clean, mark it to reflect that.
2311                  */
2312                 lock_buffer(bh_result);
2313                 clear_buffer_dirty(bh_result);
2314                 unlock_buffer(bh_result);
2315         }
2316         return retval;
2317 }
2318
2319 /*
2320  * mason@suse.com: updated in 2.5.54 to follow the same general io
2321  * start/recovery path as __block_write_full_page, along with special
2322  * code to handle reiserfs tails.
2323  */
2324 static int reiserfs_write_full_page(struct page *page,
2325                                     struct writeback_control *wbc)
2326 {
2327         struct inode *inode = page->mapping->host;
2328         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2329         int error = 0;
2330         unsigned long block;
2331         sector_t last_block;
2332         struct buffer_head *head, *bh;
2333         int partial = 0;
2334         int nr = 0;
2335         int checked = PageChecked(page);
2336         struct reiserfs_transaction_handle th;
2337         struct super_block *s = inode->i_sb;
2338         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2339         th.t_trans_id = 0;
2340
2341         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2342         if (checked && (current->flags & PF_MEMALLOC)) {
2343                 redirty_page_for_writepage(wbc, page);
2344                 unlock_page(page);
2345                 return 0;
2346         }
2347
2348         /* The page dirty bit is cleared before writepage is called, which
2349          * means we have to tell create_empty_buffers to make dirty buffers
2350          * The page really should be up to date at this point, so tossing
2351          * in the BH_Uptodate is just a sanity check.
2352          */
2353         if (!page_has_buffers(page)) {
2354                 create_empty_buffers(page, s->s_blocksize,
2355                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2356         }
2357         head = page_buffers(page);
2358
2359         /* last page in the file, zero out any contents past the
2360          ** last byte in the file
2361          */
2362         if (page->index >= end_index) {
2363                 unsigned last_offset;
2364
2365                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2366                 /* no file contents in this page */
2367                 if (page->index >= end_index + 1 || !last_offset) {
2368                         unlock_page(page);
2369                         return 0;
2370                 }
2371                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2372         }
2373         bh = head;
2374         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2375         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2376         /* first map all the buffers, logging any direct items we find */
2377         do {
2378                 if (block > last_block) {
2379                         /*
2380                          * This can happen when the block size is less than
2381                          * the page size.  The corresponding bytes in the page
2382                          * were zero filled above
2383                          */
2384                         clear_buffer_dirty(bh);
2385                         set_buffer_uptodate(bh);
2386                 } else if ((checked || buffer_dirty(bh)) &&
2387                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2388                                                        && bh->b_blocknr ==
2389                                                        0))) {
2390                         /* not mapped yet, or it points to a direct item, search
2391                          * the btree for the mapping info, and log any direct
2392                          * items found
2393                          */
2394                         if ((error = map_block_for_writepage(inode, bh, block))) {
2395                                 goto fail;
2396                         }
2397                 }
2398                 bh = bh->b_this_page;
2399                 block++;
2400         } while (bh != head);
2401
2402         /*
2403          * we start the transaction after map_block_for_writepage,
2404          * because it can create holes in the file (an unbounded operation).
2405          * starting it here, we can make a reliable estimate for how many
2406          * blocks we're going to log
2407          */
2408         if (checked) {
2409                 ClearPageChecked(page);
2410                 reiserfs_write_lock(s);
2411                 error = journal_begin(&th, s, bh_per_page + 1);
2412                 if (error) {
2413                         reiserfs_write_unlock(s);
2414                         goto fail;
2415                 }
2416                 reiserfs_update_inode_transaction(inode);
2417         }
2418         /* now go through and lock any dirty buffers on the page */
2419         do {
2420                 get_bh(bh);
2421                 if (!buffer_mapped(bh))
2422                         continue;
2423                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2424                         continue;
2425
2426                 if (checked) {
2427                         reiserfs_prepare_for_journal(s, bh, 1);
2428                         journal_mark_dirty(&th, s, bh);
2429                         continue;
2430                 }
2431                 /* from this point on, we know the buffer is mapped to a
2432                  * real block and not a direct item
2433                  */
2434                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2435                         lock_buffer(bh);
2436                 } else {
2437                         if (!trylock_buffer(bh)) {
2438                                 redirty_page_for_writepage(wbc, page);
2439                                 continue;
2440                         }
2441                 }
2442                 if (test_clear_buffer_dirty(bh)) {
2443                         mark_buffer_async_write(bh);
2444                 } else {
2445                         unlock_buffer(bh);
2446                 }
2447         } while ((bh = bh->b_this_page) != head);
2448
2449         if (checked) {
2450                 error = journal_end(&th, s, bh_per_page + 1);
2451                 reiserfs_write_unlock(s);
2452                 if (error)
2453                         goto fail;
2454         }
2455         BUG_ON(PageWriteback(page));
2456         set_page_writeback(page);
2457         unlock_page(page);
2458
2459         /*
2460          * since any buffer might be the only dirty buffer on the page,
2461          * the first submit_bh can bring the page out of writeback.
2462          * be careful with the buffers.
2463          */
2464         do {
2465                 struct buffer_head *next = bh->b_this_page;
2466                 if (buffer_async_write(bh)) {
2467                         submit_bh(WRITE, bh);
2468                         nr++;
2469                 }
2470                 put_bh(bh);
2471                 bh = next;
2472         } while (bh != head);
2473
2474         error = 0;
2475       done:
2476         if (nr == 0) {
2477                 /*
2478                  * if this page only had a direct item, it is very possible for
2479                  * no io to be required without there being an error.  Or,
2480                  * someone else could have locked them and sent them down the
2481                  * pipe without locking the page
2482                  */
2483                 bh = head;
2484                 do {
2485                         if (!buffer_uptodate(bh)) {
2486                                 partial = 1;
2487                                 break;
2488                         }
2489                         bh = bh->b_this_page;
2490                 } while (bh != head);
2491                 if (!partial)
2492                         SetPageUptodate(page);
2493                 end_page_writeback(page);
2494         }
2495         return error;
2496
2497       fail:
2498         /* catches various errors, we need to make sure any valid dirty blocks
2499          * get to the media.  The page is currently locked and not marked for
2500          * writeback
2501          */
2502         ClearPageUptodate(page);
2503         bh = head;
2504         do {
2505                 get_bh(bh);
2506                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2507                         lock_buffer(bh);
2508                         mark_buffer_async_write(bh);
2509                 } else {
2510                         /*
2511                          * clear any dirty bits that might have come from getting
2512                          * attached to a dirty page
2513                          */
2514                         clear_buffer_dirty(bh);
2515                 }
2516                 bh = bh->b_this_page;
2517         } while (bh != head);
2518         SetPageError(page);
2519         BUG_ON(PageWriteback(page));
2520         set_page_writeback(page);
2521         unlock_page(page);
2522         do {
2523                 struct buffer_head *next = bh->b_this_page;
2524                 if (buffer_async_write(bh)) {
2525                         clear_buffer_dirty(bh);
2526                         submit_bh(WRITE, bh);
2527                         nr++;
2528                 }
2529                 put_bh(bh);
2530                 bh = next;
2531         } while (bh != head);
2532         goto done;
2533 }
2534
2535 static int reiserfs_readpage(struct file *f, struct page *page)
2536 {
2537         return block_read_full_page(page, reiserfs_get_block);
2538 }
2539
2540 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2541 {
2542         struct inode *inode = page->mapping->host;
2543         reiserfs_wait_on_write_block(inode->i_sb);
2544         return reiserfs_write_full_page(page, wbc);
2545 }
2546
2547 static void reiserfs_truncate_failed_write(struct inode *inode)
2548 {
2549         truncate_inode_pages(inode->i_mapping, inode->i_size);
2550         reiserfs_truncate_file(inode, 0);
2551 }
2552
2553 static int reiserfs_write_begin(struct file *file,
2554                                 struct address_space *mapping,
2555                                 loff_t pos, unsigned len, unsigned flags,
2556                                 struct page **pagep, void **fsdata)
2557 {
2558         struct inode *inode;
2559         struct page *page;
2560         pgoff_t index;
2561         int ret;
2562         int old_ref = 0;
2563
2564         inode = mapping->host;
2565         *fsdata = 0;
2566         if (flags & AOP_FLAG_CONT_EXPAND &&
2567             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2568                 pos ++;
2569                 *fsdata = (void *)(unsigned long)flags;
2570         }
2571
2572         index = pos >> PAGE_CACHE_SHIFT;
2573         page = grab_cache_page_write_begin(mapping, index, flags);
2574         if (!page)
2575                 return -ENOMEM;
2576         *pagep = page;
2577
2578         reiserfs_wait_on_write_block(inode->i_sb);
2579         fix_tail_page_for_writing(page);
2580         if (reiserfs_transaction_running(inode->i_sb)) {
2581                 struct reiserfs_transaction_handle *th;
2582                 th = (struct reiserfs_transaction_handle *)current->
2583                     journal_info;
2584                 BUG_ON(!th->t_refcount);
2585                 BUG_ON(!th->t_trans_id);
2586                 old_ref = th->t_refcount;
2587                 th->t_refcount++;
2588         }
2589         ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2590                                 reiserfs_get_block);
2591         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2592                 struct reiserfs_transaction_handle *th = current->journal_info;
2593                 /* this gets a little ugly.  If reiserfs_get_block returned an
2594                  * error and left a transacstion running, we've got to close it,
2595                  * and we've got to free handle if it was a persistent transaction.
2596                  *
2597                  * But, if we had nested into an existing transaction, we need
2598                  * to just drop the ref count on the handle.
2599                  *
2600                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2601                  * and it was a persistent trans.  Otherwise, it was nested above.
2602                  */
2603                 if (th->t_refcount > old_ref) {
2604                         if (old_ref)
2605                                 th->t_refcount--;
2606                         else {
2607                                 int err;
2608                                 reiserfs_write_lock(inode->i_sb);
2609                                 err = reiserfs_end_persistent_transaction(th);
2610                                 reiserfs_write_unlock(inode->i_sb);
2611                                 if (err)
2612                                         ret = err;
2613                         }
2614                 }
2615         }
2616         if (ret) {
2617                 unlock_page(page);
2618                 page_cache_release(page);
2619                 /* Truncate allocated blocks */
2620                 reiserfs_truncate_failed_write(inode);
2621         }
2622         return ret;
2623 }
2624
2625 int reiserfs_prepare_write(struct file *f, struct page *page,
2626                            unsigned from, unsigned to)
2627 {
2628         struct inode *inode = page->mapping->host;
2629         int ret;
2630         int old_ref = 0;
2631
2632         reiserfs_write_unlock(inode->i_sb);
2633         reiserfs_wait_on_write_block(inode->i_sb);
2634         reiserfs_write_lock(inode->i_sb);
2635
2636         fix_tail_page_for_writing(page);
2637         if (reiserfs_transaction_running(inode->i_sb)) {
2638                 struct reiserfs_transaction_handle *th;
2639                 th = (struct reiserfs_transaction_handle *)current->
2640                     journal_info;
2641                 BUG_ON(!th->t_refcount);
2642                 BUG_ON(!th->t_trans_id);
2643                 old_ref = th->t_refcount;
2644                 th->t_refcount++;
2645         }
2646
2647         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2648         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2649                 struct reiserfs_transaction_handle *th = current->journal_info;
2650                 /* this gets a little ugly.  If reiserfs_get_block returned an
2651                  * error and left a transacstion running, we've got to close it,
2652                  * and we've got to free handle if it was a persistent transaction.
2653                  *
2654                  * But, if we had nested into an existing transaction, we need
2655                  * to just drop the ref count on the handle.
2656                  *
2657                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2658                  * and it was a persistent trans.  Otherwise, it was nested above.
2659                  */
2660                 if (th->t_refcount > old_ref) {
2661                         if (old_ref)
2662                                 th->t_refcount--;
2663                         else {
2664                                 int err;
2665                                 reiserfs_write_lock(inode->i_sb);
2666                                 err = reiserfs_end_persistent_transaction(th);
2667                                 reiserfs_write_unlock(inode->i_sb);
2668                                 if (err)
2669                                         ret = err;
2670                         }
2671                 }
2672         }
2673         return ret;
2674
2675 }
2676
2677 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2678 {
2679         return generic_block_bmap(as, block, reiserfs_bmap);
2680 }
2681
2682 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2683                               loff_t pos, unsigned len, unsigned copied,
2684                               struct page *page, void *fsdata)
2685 {
2686         struct inode *inode = page->mapping->host;
2687         int ret = 0;
2688         int update_sd = 0;
2689         struct reiserfs_transaction_handle *th;
2690         unsigned start;
2691         int lock_depth = 0;
2692         bool locked = false;
2693
2694         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2695                 pos ++;
2696
2697         reiserfs_wait_on_write_block(inode->i_sb);
2698         if (reiserfs_transaction_running(inode->i_sb))
2699                 th = current->journal_info;
2700         else
2701                 th = NULL;
2702
2703         start = pos & (PAGE_CACHE_SIZE - 1);
2704         if (unlikely(copied < len)) {
2705                 if (!PageUptodate(page))
2706                         copied = 0;
2707
2708                 page_zero_new_buffers(page, start + copied, start + len);
2709         }
2710         flush_dcache_page(page);
2711
2712         reiserfs_commit_page(inode, page, start, start + copied);
2713
2714         /* generic_commit_write does this for us, but does not update the
2715          ** transaction tracking stuff when the size changes.  So, we have
2716          ** to do the i_size updates here.
2717          */
2718         if (pos + copied > inode->i_size) {
2719                 struct reiserfs_transaction_handle myth;
2720                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2721                 locked = true;
2722                 /* If the file have grown beyond the border where it
2723                    can have a tail, unmark it as needing a tail
2724                    packing */
2725                 if ((have_large_tails(inode->i_sb)
2726                      && inode->i_size > i_block_size(inode) * 4)
2727                     || (have_small_tails(inode->i_sb)
2728                         && inode->i_size > i_block_size(inode)))
2729                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2730
2731                 ret = journal_begin(&myth, inode->i_sb, 1);
2732                 if (ret)
2733                         goto journal_error;
2734
2735                 reiserfs_update_inode_transaction(inode);
2736                 inode->i_size = pos + copied;
2737                 /*
2738                  * this will just nest into our transaction.  It's important
2739                  * to use mark_inode_dirty so the inode gets pushed around on the
2740                  * dirty lists, and so that O_SYNC works as expected
2741                  */
2742                 mark_inode_dirty(inode);
2743                 reiserfs_update_sd(&myth, inode);
2744                 update_sd = 1;
2745                 ret = journal_end(&myth, inode->i_sb, 1);
2746                 if (ret)
2747                         goto journal_error;
2748         }
2749         if (th) {
2750                 if (!locked) {
2751                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2752                         locked = true;
2753                 }
2754                 if (!update_sd)
2755                         mark_inode_dirty(inode);
2756                 ret = reiserfs_end_persistent_transaction(th);
2757                 if (ret)
2758                         goto out;
2759         }
2760
2761       out:
2762         if (locked)
2763                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2764         unlock_page(page);
2765         page_cache_release(page);
2766
2767         if (pos + len > inode->i_size)
2768                 reiserfs_truncate_failed_write(inode);
2769
2770         return ret == 0 ? copied : ret;
2771
2772       journal_error:
2773         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2774         locked = false;
2775         if (th) {
2776                 if (!update_sd)
2777                         reiserfs_update_sd(th, inode);
2778                 ret = reiserfs_end_persistent_transaction(th);
2779         }
2780         goto out;
2781 }
2782
2783 int reiserfs_commit_write(struct file *f, struct page *page,
2784                           unsigned from, unsigned to)
2785 {
2786         struct inode *inode = page->mapping->host;
2787         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2788         int ret = 0;
2789         int update_sd = 0;
2790         struct reiserfs_transaction_handle *th = NULL;
2791
2792         reiserfs_write_unlock(inode->i_sb);
2793         reiserfs_wait_on_write_block(inode->i_sb);
2794         reiserfs_write_lock(inode->i_sb);
2795
2796         if (reiserfs_transaction_running(inode->i_sb)) {
2797                 th = current->journal_info;
2798         }
2799         reiserfs_commit_page(inode, page, from, to);
2800
2801         /* generic_commit_write does this for us, but does not update the
2802          ** transaction tracking stuff when the size changes.  So, we have
2803          ** to do the i_size updates here.
2804          */
2805         if (pos > inode->i_size) {
2806                 struct reiserfs_transaction_handle myth;
2807                 /* If the file have grown beyond the border where it
2808                    can have a tail, unmark it as needing a tail
2809                    packing */
2810                 if ((have_large_tails(inode->i_sb)
2811                      && inode->i_size > i_block_size(inode) * 4)
2812                     || (have_small_tails(inode->i_sb)
2813                         && inode->i_size > i_block_size(inode)))
2814                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2815
2816                 ret = journal_begin(&myth, inode->i_sb, 1);
2817                 if (ret)
2818                         goto journal_error;
2819
2820                 reiserfs_update_inode_transaction(inode);
2821                 inode->i_size = pos;
2822                 /*
2823                  * this will just nest into our transaction.  It's important
2824                  * to use mark_inode_dirty so the inode gets pushed around on the
2825                  * dirty lists, and so that O_SYNC works as expected
2826                  */
2827                 mark_inode_dirty(inode);
2828                 reiserfs_update_sd(&myth, inode);
2829                 update_sd = 1;
2830                 ret = journal_end(&myth, inode->i_sb, 1);
2831                 if (ret)
2832                         goto journal_error;
2833         }
2834         if (th) {
2835                 if (!update_sd)
2836                         mark_inode_dirty(inode);
2837                 ret = reiserfs_end_persistent_transaction(th);
2838                 if (ret)
2839                         goto out;
2840         }
2841
2842       out:
2843         return ret;
2844
2845       journal_error:
2846         if (th) {
2847                 if (!update_sd)
2848                         reiserfs_update_sd(th, inode);
2849                 ret = reiserfs_end_persistent_transaction(th);
2850         }
2851
2852         return ret;
2853 }
2854
2855 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2856 {
2857         if (reiserfs_attrs(inode->i_sb)) {
2858                 if (sd_attrs & REISERFS_SYNC_FL)
2859                         inode->i_flags |= S_SYNC;
2860                 else
2861                         inode->i_flags &= ~S_SYNC;
2862                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2863                         inode->i_flags |= S_IMMUTABLE;
2864                 else
2865                         inode->i_flags &= ~S_IMMUTABLE;
2866                 if (sd_attrs & REISERFS_APPEND_FL)
2867                         inode->i_flags |= S_APPEND;
2868                 else
2869                         inode->i_flags &= ~S_APPEND;
2870                 if (sd_attrs & REISERFS_NOATIME_FL)
2871                         inode->i_flags |= S_NOATIME;
2872                 else
2873                         inode->i_flags &= ~S_NOATIME;
2874                 if (sd_attrs & REISERFS_NOTAIL_FL)
2875                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2876                 else
2877                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2878         }
2879 }
2880
2881 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2882 {
2883         if (reiserfs_attrs(inode->i_sb)) {
2884                 if (inode->i_flags & S_IMMUTABLE)
2885                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2886                 else
2887                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2888                 if (inode->i_flags & S_SYNC)
2889                         *sd_attrs |= REISERFS_SYNC_FL;
2890                 else
2891                         *sd_attrs &= ~REISERFS_SYNC_FL;
2892                 if (inode->i_flags & S_NOATIME)
2893                         *sd_attrs |= REISERFS_NOATIME_FL;
2894                 else
2895                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2896                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2897                         *sd_attrs |= REISERFS_NOTAIL_FL;
2898                 else
2899                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2900         }
2901 }
2902
2903 /* decide if this buffer needs to stay around for data logging or ordered
2904 ** write purposes
2905 */
2906 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2907 {
2908         int ret = 1;
2909         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2910
2911         lock_buffer(bh);
2912         spin_lock(&j->j_dirty_buffers_lock);
2913         if (!buffer_mapped(bh)) {
2914                 goto free_jh;
2915         }
2916         /* the page is locked, and the only places that log a data buffer
2917          * also lock the page.
2918          */
2919         if (reiserfs_file_data_log(inode)) {
2920                 /*
2921                  * very conservative, leave the buffer pinned if
2922                  * anyone might need it.
2923                  */
2924                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2925                         ret = 0;
2926                 }
2927         } else  if (buffer_dirty(bh)) {
2928                 struct reiserfs_journal_list *jl;
2929                 struct reiserfs_jh *jh = bh->b_private;
2930
2931                 /* why is this safe?
2932                  * reiserfs_setattr updates i_size in the on disk
2933                  * stat data before allowing vmtruncate to be called.
2934                  *
2935                  * If buffer was put onto the ordered list for this
2936                  * transaction, we know for sure either this transaction
2937                  * or an older one already has updated i_size on disk,
2938                  * and this ordered data won't be referenced in the file
2939                  * if we crash.
2940                  *
2941                  * if the buffer was put onto the ordered list for an older
2942                  * transaction, we need to leave it around
2943                  */
2944                 if (jh && (jl = jh->jl)
2945                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2946                         ret = 0;
2947         }
2948       free_jh:
2949         if (ret && bh->b_private) {
2950                 reiserfs_free_jh(bh);
2951         }
2952         spin_unlock(&j->j_dirty_buffers_lock);
2953         unlock_buffer(bh);
2954         return ret;
2955 }
2956
2957 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2958 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2959 {
2960         struct buffer_head *head, *bh, *next;
2961         struct inode *inode = page->mapping->host;
2962         unsigned int curr_off = 0;
2963         int ret = 1;
2964
2965         BUG_ON(!PageLocked(page));
2966
2967         if (offset == 0)
2968                 ClearPageChecked(page);
2969
2970         if (!page_has_buffers(page))
2971                 goto out;
2972
2973         head = page_buffers(page);
2974         bh = head;
2975         do {
2976                 unsigned int next_off = curr_off + bh->b_size;
2977                 next = bh->b_this_page;
2978
2979                 /*
2980                  * is this block fully invalidated?
2981                  */
2982                 if (offset <= curr_off) {
2983                         if (invalidatepage_can_drop(inode, bh))
2984                                 reiserfs_unmap_buffer(bh);
2985                         else
2986                                 ret = 0;
2987                 }
2988                 curr_off = next_off;
2989                 bh = next;
2990         } while (bh != head);
2991
2992         /*
2993          * We release buffers only if the entire page is being invalidated.
2994          * The get_block cached value has been unconditionally invalidated,
2995          * so real IO is not possible anymore.
2996          */
2997         if (!offset && ret) {
2998                 ret = try_to_release_page(page, 0);
2999                 /* maybe should BUG_ON(!ret); - neilb */
3000         }
3001       out:
3002         return;
3003 }
3004
3005 static int reiserfs_set_page_dirty(struct page *page)
3006 {
3007         struct inode *inode = page->mapping->host;
3008         if (reiserfs_file_data_log(inode)) {
3009                 SetPageChecked(page);
3010                 return __set_page_dirty_nobuffers(page);
3011         }
3012         return __set_page_dirty_buffers(page);
3013 }
3014
3015 /*
3016  * Returns 1 if the page's buffers were dropped.  The page is locked.
3017  *
3018  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3019  * in the buffers at page_buffers(page).
3020  *
3021  * even in -o notail mode, we can't be sure an old mount without -o notail
3022  * didn't create files with tails.
3023  */
3024 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3025 {
3026         struct inode *inode = page->mapping->host;
3027         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3028         struct buffer_head *head;
3029         struct buffer_head *bh;
3030         int ret = 1;
3031
3032         WARN_ON(PageChecked(page));
3033         spin_lock(&j->j_dirty_buffers_lock);
3034         head = page_buffers(page);
3035         bh = head;
3036         do {
3037                 if (bh->b_private) {
3038                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3039                                 reiserfs_free_jh(bh);
3040                         } else {
3041                                 ret = 0;
3042                                 break;
3043                         }
3044                 }
3045                 bh = bh->b_this_page;
3046         } while (bh != head);
3047         if (ret)
3048                 ret = try_to_free_buffers(page);
3049         spin_unlock(&j->j_dirty_buffers_lock);
3050         return ret;
3051 }
3052
3053 /* We thank Mingming Cao for helping us understand in great detail what
3054    to do in this section of the code. */
3055 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3056                                   const struct iovec *iov, loff_t offset,
3057                                   unsigned long nr_segs)
3058 {
3059         struct file *file = iocb->ki_filp;
3060         struct inode *inode = file->f_mapping->host;
3061
3062         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3063                                   offset, nr_segs,
3064                                   reiserfs_get_blocks_direct_io, NULL);
3065 }
3066
3067 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3068 {
3069         struct inode *inode = dentry->d_inode;
3070         unsigned int ia_valid;
3071         int depth;
3072         int error;
3073
3074         /* must be turned off for recursive notify_change calls */
3075         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3076
3077         depth = reiserfs_write_lock_once(inode->i_sb);
3078         if (attr->ia_valid & ATTR_SIZE) {
3079                 vfs_dq_init(inode);
3080
3081                 /* version 2 items will be caught by the s_maxbytes check
3082                  ** done for us in vmtruncate
3083                  */
3084                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3085                     attr->ia_size > MAX_NON_LFS) {
3086                         error = -EFBIG;
3087                         goto out;
3088                 }
3089                 /* fill in hole pointers in the expanding truncate case. */
3090                 if (attr->ia_size > inode->i_size) {
3091                         error = generic_cont_expand_simple(inode, attr->ia_size);
3092                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3093                                 int err;
3094                                 struct reiserfs_transaction_handle th;
3095                                 /* we're changing at most 2 bitmaps, inode + super */
3096                                 err = journal_begin(&th, inode->i_sb, 4);
3097                                 if (!err) {
3098                                         reiserfs_discard_prealloc(&th, inode);
3099                                         err = journal_end(&th, inode->i_sb, 4);
3100                                 }
3101                                 if (err)
3102                                         error = err;
3103                         }
3104                         if (error)
3105                                 goto out;
3106                         /*
3107                          * file size is changed, ctime and mtime are
3108                          * to be updated
3109                          */
3110                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3111                 }
3112         }
3113
3114         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3115              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3116             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3117                 /* stat data of format v3.5 has 16 bit uid and gid */
3118                 error = -EINVAL;
3119                 goto out;
3120         }
3121
3122         error = inode_change_ok(inode, attr);
3123         if (!error) {
3124                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3125                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3126                         error = reiserfs_chown_xattrs(inode, attr);
3127
3128                         if (!error) {
3129                                 struct reiserfs_transaction_handle th;
3130                                 int jbegin_count =
3131                                     2 *
3132                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3133                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3134                                     2;
3135
3136                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3137                                 error =
3138                                     journal_begin(&th, inode->i_sb,
3139                                                   jbegin_count);
3140                                 if (error)
3141                                         goto out;
3142                                 error = dquot_transfer(inode, attr);
3143                                 if (error) {
3144                                         journal_end(&th, inode->i_sb,
3145                                                     jbegin_count);
3146                                         goto out;
3147                                 }
3148                                 /* Update corresponding info in inode so that everything is in
3149                                  * one transaction */
3150                                 if (attr->ia_valid & ATTR_UID)
3151                                         inode->i_uid = attr->ia_uid;
3152                                 if (attr->ia_valid & ATTR_GID)
3153                                         inode->i_gid = attr->ia_gid;
3154                                 mark_inode_dirty(inode);
3155                                 error =
3156                                     journal_end(&th, inode->i_sb, jbegin_count);
3157                         }
3158                 }
3159                 if (!error) {
3160                         /*
3161                          * Relax the lock here, as it might truncate the
3162                          * inode pages and wait for inode pages locks.
3163                          * To release such page lock, the owner needs the
3164                          * reiserfs lock
3165                          */
3166                         reiserfs_write_unlock_once(inode->i_sb, depth);
3167                         error = inode_setattr(inode, attr);
3168                         depth = reiserfs_write_lock_once(inode->i_sb);
3169                 }
3170         }
3171
3172         if (!error && reiserfs_posixacl(inode->i_sb)) {
3173                 if (attr->ia_valid & ATTR_MODE)
3174                         error = reiserfs_acl_chmod(inode);
3175         }
3176
3177       out:
3178         reiserfs_write_unlock_once(inode->i_sb, depth);
3179
3180         return error;
3181 }
3182
3183 const struct address_space_operations reiserfs_address_space_operations = {
3184         .writepage = reiserfs_writepage,
3185         .readpage = reiserfs_readpage,
3186         .readpages = reiserfs_readpages,
3187         .releasepage = reiserfs_releasepage,
3188         .invalidatepage = reiserfs_invalidatepage,
3189         .sync_page = block_sync_page,
3190         .write_begin = reiserfs_write_begin,
3191         .write_end = reiserfs_write_end,
3192         .bmap = reiserfs_aop_bmap,
3193         .direct_IO = reiserfs_direct_IO,
3194         .set_page_dirty = reiserfs_set_page_dirty,
3195 };