fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/config.h>
   6 #include <linux/time.h>
   7 #include <linux/fs.h>
   8 #include <linux/reiserfs_fs.h>
   9 #include <linux/reiserfs_acl.h>
  10 #include <linux/reiserfs_xattr.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
  21 extern int reiserfs_default_io_size;    /* default io size devuned in super.c */
  22
  23 static int reiserfs_commit_write(struct file *f, struct page *page,
  24                                  unsigned from, unsigned to);
  25 static int reiserfs_prepare_write(struct file *f, struct page *page,
  26                                   unsigned from, unsigned to);
  27
  28 void reiserfs_delete_inode(struct inode *inode)
  29 {
  30         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  31         int jbegin_count =
  32             JOURNAL_PER_BALANCE_CNT * 2 +
  33             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  34         struct reiserfs_transaction_handle th;
  35         int err;
  36
  37         truncate_inode_pages(&inode->i_data, 0);
  38
  39
  40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  42                 mutex_lock(&inode->i_mutex);
  43                 reiserfs_write_lock(inode->i_sb);
  44
  45                 reiserfs_delete_xattrs(inode);
  46
  47                 if (journal_begin(&th, inode->i_sb, jbegin_count)) {
  48                         mutex_unlock(&inode->i_mutex);
  49                         goto out;
  50                 }
  51                 reiserfs_update_inode_transaction(inode);
  52
  53                 err = reiserfs_delete_object(&th, inode);
  54
  55                 /* Do quota update inside a transaction for journaled quotas. We must do that
  56                  * after delete_object so that quota updates go into the same transaction as
  57                  * stat data deletion */
  58                 if (!err)
  59                         DQUOT_FREE_INODE(inode);
  60
  61                 if (journal_end(&th, inode->i_sb, jbegin_count)) {
  62                         mutex_unlock(&inode->i_mutex);
  63                         goto out;
  64                 }
  65
  66                 mutex_unlock(&inode->i_mutex);
  67
  68                 /* check return value from reiserfs_delete_object after
  69                  * ending the transaction
  70                  */
  71                 if (err)
  72                     goto out;
  73
  74                 /* all items of file are deleted, so we can remove "save" link */
  75                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  76                                                                  * about an error here */
  77         } else {
  78                 /* no object items are in the tree */
  79                 reiserfs_write_lock(inode->i_sb);
  80         }
  81       out:
  82         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
  83         inode->i_blocks = 0;
  84         reiserfs_write_unlock(inode->i_sb);
  85 }
  86
  87 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  88                           __u32 objectid, loff_t offset, int type, int length)
  89 {
  90         key->version = version;
  91
  92         key->on_disk_key.k_dir_id = dirid;
  93         key->on_disk_key.k_objectid = objectid;
  94         set_cpu_key_k_offset(key, offset);
  95         set_cpu_key_k_type(key, type);
  96         key->key_length = length;
  97 }
  98
  99 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
 100    offset and type of key */
 101 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 102                   int type, int length)
 103 {
 104         _make_cpu_key(key, get_inode_item_key_version(inode),
 105                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 106                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 107                       length);
 108 }
 109
 110 //
 111 // when key is 0, do not set version and short key
 112 //
 113 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 114                               int version,
 115                               loff_t offset, int type, int length,
 116                               int entry_count /*or ih_free_space */ )
 117 {
 118         if (key) {
 119                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 120                 ih->ih_key.k_objectid =
 121                     cpu_to_le32(key->on_disk_key.k_objectid);
 122         }
 123         put_ih_version(ih, version);
 124         set_le_ih_k_offset(ih, offset);
 125         set_le_ih_k_type(ih, type);
 126         put_ih_item_len(ih, length);
 127         /*    set_ih_free_space (ih, 0); */
 128         // for directory items it is entry count, for directs and stat
 129         // datas - 0xffff, for indirects - 0
 130         put_ih_entry_count(ih, entry_count);
 131 }
 132
 133 //
 134 // FIXME: we might cache recently accessed indirect item
 135
 136 // Ugh.  Not too eager for that....
 137 //  I cut the code until such time as I see a convincing argument (benchmark).
 138 // I don't want a bloated inode struct..., and I don't like code complexity....
 139
 140 /* cutting the code is fine, since it really isn't in use yet and is easy
 141 ** to add back in.  But, Vladimir has a really good idea here.  Think
 142 ** about what happens for reading a file.  For each page,
 143 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 144 ** an indirect item.  This indirect item has X number of pointers, where
 145 ** X is a big number if we've done the block allocation right.  But,
 146 ** we only use one or two of these pointers during each call to readpage,
 147 ** needlessly researching again later on.
 148 **
 149 ** The size of the cache could be dynamic based on the size of the file.
 150 **
 151 ** I'd also like to see us cache the location the stat data item, since
 152 ** we are needlessly researching for that frequently.
 153 **
 154 ** --chris
 155 */
 156
 157 /* If this page has a file tail in it, and
 158 ** it was read in by get_block_create_0, the page data is valid,
 159 ** but tail is still sitting in a direct item, and we can't write to
 160 ** it.  So, look through this page, and check all the mapped buffers
 161 ** to make sure they have valid block numbers.  Any that don't need
 162 ** to be unmapped, so that block_prepare_write will correctly call
 163 ** reiserfs_get_block to convert the tail into an unformatted node
 164 */
 165 static inline void fix_tail_page_for_writing(struct page *page)
 166 {
 167         struct buffer_head *head, *next, *bh;
 168
 169         if (page && page_has_buffers(page)) {
 170                 head = page_buffers(page);
 171                 bh = head;
 172                 do {
 173                         next = bh->b_this_page;
 174                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 175                                 reiserfs_unmap_buffer(bh);
 176                         }
 177                         bh = next;
 178                 } while (bh != head);
 179         }
 180 }
 181
 182 /* reiserfs_get_block does not need to allocate a block only if it has been
 183    done already or non-hole position has been found in the indirect item */
 184 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 185                                     struct item_head *ih,
 186                                     __le32 * item, int pos_in_item)
 187 {
 188         if (allocated)
 189                 return 0;
 190         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 191             get_block_num(item, pos_in_item))
 192                 return 0;
 193         return 1;
 194 }
 195
 196 static inline int indirect_item_found(int retval, struct item_head *ih)
 197 {
 198         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 199 }
 200
 201 static inline void set_block_dev_mapped(struct buffer_head *bh,
 202                                         b_blocknr_t block, struct inode *inode)
 203 {
 204         map_bh(bh, inode->i_sb, block);
 205 }
 206
 207 //
 208 // files which were created in the earlier version can not be longer,
 209 // than 2 gb
 210 //
 211 static int file_capable(struct inode *inode, long block)
 212 {
 213         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 214             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 215                 return 1;
 216
 217         return 0;
 218 }
 219
 220 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
 221                                    struct inode *inode, struct path *path)
 222 {
 223         struct super_block *s = th->t_super;
 224         int len = th->t_blocks_allocated;
 225         int err;
 226
 227         BUG_ON(!th->t_trans_id);
 228         BUG_ON(!th->t_refcount);
 229
 230         /* we cannot restart while nested */
 231         if (th->t_refcount > 1) {
 232                 return 0;
 233         }
 234         pathrelse(path);
 235         reiserfs_update_sd(th, inode);
 236         err = journal_end(th, s, len);
 237         if (!err) {
 238                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 239                 if (!err)
 240                         reiserfs_update_inode_transaction(inode);
 241         }
 242         return err;
 243 }
 244
 245 // it is called by get_block when create == 0. Returns block number
 246 // for 'block'-th logical block of file. When it hits direct item it
 247 // returns 0 (being called from bmap) or read direct item into piece
 248 // of page (bh_result)
 249
 250 // Please improve the english/clarity in the comment above, as it is
 251 // hard to understand.
 252
 253 static int _get_block_create_0(struct inode *inode, long block,
 254                                struct buffer_head *bh_result, int args)
 255 {
 256         INITIALIZE_PATH(path);
 257         struct cpu_key key;
 258         struct buffer_head *bh;
 259         struct item_head *ih, tmp_ih;
 260         int fs_gen;
 261         int blocknr;
 262         char *p = NULL;
 263         int chars;
 264         int ret;
 265         int result;
 266         int done = 0;
 267         unsigned long offset;
 268
 269         // prepare the key to look for the 'block'-th block of file
 270         make_cpu_key(&key, inode,
 271                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 272                      3);
 273
 274       research:
 275         result = search_for_position_by_key(inode->i_sb, &key, &path);
 276         if (result != POSITION_FOUND) {
 277                 pathrelse(&path);
 278                 if (p)
 279                         kunmap(bh_result->b_page);
 280                 if (result == IO_ERROR)
 281                         return -EIO;
 282                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 283                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 284                 if ((args & GET_BLOCK_NO_HOLE)
 285                     && !PageUptodate(bh_result->b_page)) {
 286                         return -ENOENT;
 287                 }
 288                 return 0;
 289         }
 290         //
 291         bh = get_last_bh(&path);
 292         ih = get_ih(&path);
 293         if (is_indirect_le_ih(ih)) {
 294                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 295
 296                 /* FIXME: here we could cache indirect item or part of it in
 297                    the inode to avoid search_by_key in case of subsequent
 298                    access to file */
 299                 blocknr = get_block_num(ind_item, path.pos_in_item);
 300                 ret = 0;
 301                 if (blocknr) {
 302                         map_bh(bh_result, inode->i_sb, blocknr);
 303                         if (path.pos_in_item ==
 304                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 305                                 set_buffer_boundary(bh_result);
 306                         }
 307                 } else
 308                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 309                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 310                 if ((args & GET_BLOCK_NO_HOLE)
 311                             && !PageUptodate(bh_result->b_page)) {
 312                         ret = -ENOENT;
 313                 }
 314
 315                 pathrelse(&path);
 316                 if (p)
 317                         kunmap(bh_result->b_page);
 318                 return ret;
 319         }
 320         // requested data are in direct item(s)
 321         if (!(args & GET_BLOCK_READ_DIRECT)) {
 322                 // we are called by bmap. FIXME: we can not map block of file
 323                 // when it is stored in direct item(s)
 324                 pathrelse(&path);
 325                 if (p)
 326                         kunmap(bh_result->b_page);
 327                 return -ENOENT;
 328         }
 329
 330         /* if we've got a direct item, and the buffer or page was uptodate,
 331          ** we don't want to pull data off disk again.  skip to the
 332          ** end, where we map the buffer and return
 333          */
 334         if (buffer_uptodate(bh_result)) {
 335                 goto finished;
 336         } else
 337                 /*
 338                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 339                  ** pages without any buffers.  If the page is up to date, we don't want
 340                  ** read old data off disk.  Set the up to date bit on the buffer instead
 341                  ** and jump to the end
 342                  */
 343         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 344                 set_buffer_uptodate(bh_result);
 345                 goto finished;
 346         }
 347         // read file tail into part of page
 348         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 349         fs_gen = get_generation(inode->i_sb);
 350         copy_item_head(&tmp_ih, ih);
 351
 352         /* we only want to kmap if we are reading the tail into the page.
 353          ** this is not the common case, so we don't kmap until we are
 354          ** sure we need to.  But, this means the item might move if
 355          ** kmap schedules
 356          */
 357         if (!p) {
 358                 p = (char *)kmap(bh_result->b_page);
 359                 if (fs_changed(fs_gen, inode->i_sb)
 360                     && item_moved(&tmp_ih, &path)) {
 361                         goto research;
 362                 }
 363         }
 364         p += offset;
 365         memset(p, 0, inode->i_sb->s_blocksize);
 366         do {
 367                 if (!is_direct_le_ih(ih)) {
 368                         BUG();
 369                 }
 370                 /* make sure we don't read more bytes than actually exist in
 371                  ** the file.  This can happen in odd cases where i_size isn't
 372                  ** correct, and when direct item padding results in a few
 373                  ** extra bytes at the end of the direct item
 374                  */
 375                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 376                         break;
 377                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 378                         chars =
 379                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 380                             path.pos_in_item;
 381                         done = 1;
 382                 } else {
 383                         chars = ih_item_len(ih) - path.pos_in_item;
 384                 }
 385                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 386
 387                 if (done)
 388                         break;
 389
 390                 p += chars;
 391
 392                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 393                         // we done, if read direct item is not the last item of
 394                         // node FIXME: we could try to check right delimiting key
 395                         // to see whether direct item continues in the right
 396                         // neighbor or rely on i_size
 397                         break;
 398
 399                 // update key to look for the next piece
 400                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 401                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 402                 if (result != POSITION_FOUND)
 403                         // i/o error most likely
 404                         break;
 405                 bh = get_last_bh(&path);
 406                 ih = get_ih(&path);
 407         } while (1);
 408
 409         flush_dcache_page(bh_result->b_page);
 410         kunmap(bh_result->b_page);
 411
 412       finished:
 413         pathrelse(&path);
 414
 415         if (result == IO_ERROR)
 416                 return -EIO;
 417
 418         /* this buffer has valid data, but isn't valid for io.  mapping it to
 419          * block #0 tells the rest of reiserfs it just has a tail in it
 420          */
 421         map_bh(bh_result, inode->i_sb, 0);
 422         set_buffer_uptodate(bh_result);
 423         return 0;
 424 }
 425
 426 // this is called to create file map. So, _get_block_create_0 will not
 427 // read direct item
 428 static int reiserfs_bmap(struct inode *inode, sector_t block,
 429                          struct buffer_head *bh_result, int create)
 430 {
 431         if (!file_capable(inode, block))
 432                 return -EFBIG;
 433
 434         reiserfs_write_lock(inode->i_sb);
 435         /* do not read the direct item */
 436         _get_block_create_0(inode, block, bh_result, 0);
 437         reiserfs_write_unlock(inode->i_sb);
 438         return 0;
 439 }
 440
 441 /* special version of get_block that is only used by grab_tail_page right
 442 ** now.  It is sent to block_prepare_write, and when you try to get a
 443 ** block past the end of the file (or a block from a hole) it returns
 444 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
 445 ** be able to do i/o on the buffers returned, unless an error value
 446 ** is also returned.
 447 **
 448 ** So, this allows block_prepare_write to be used for reading a single block
 449 ** in a page.  Where it does not produce a valid page for holes, or past the
 450 ** end of the file.  This turns out to be exactly what we need for reading
 451 ** tails for conversion.
 452 **
 453 ** The point of the wrapper is forcing a certain value for create, even
 454 ** though the VFS layer is calling this function with create==1.  If you
 455 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 456 ** don't use this function.
 457 */
 458 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 459                                        struct buffer_head *bh_result,
 460                                        int create)
 461 {
 462         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 463 }
 464
 465 /* This is special helper for reiserfs_get_block in case we are executing
 466    direct_IO request. */
 467 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 468                                          sector_t iblock,
 469                                          struct buffer_head *bh_result,
 470                                          int create)
 471 {
 472         int ret;
 473
 474         bh_result->b_page = NULL;
 475
 476         /* We set the b_size before reiserfs_get_block call since it is
 477            referenced in convert_tail_for_hole() that may be called from
 478            reiserfs_get_block() */
 479         bh_result->b_size = (1 << inode->i_blkbits);
 480
 481         ret = reiserfs_get_block(inode, iblock, bh_result,
 482                                  create | GET_BLOCK_NO_DANGLE);
 483         if (ret)
 484                 goto out;
 485
 486         /* don't allow direct io onto tail pages */
 487         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 488                 /* make sure future calls to the direct io funcs for this offset
 489                  ** in the file fail by unmapping the buffer
 490                  */
 491                 clear_buffer_mapped(bh_result);
 492                 ret = -EINVAL;
 493         }
 494         /* Possible unpacked tail. Flush the data before pages have
 495            disappeared */
 496         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 497                 int err;
 498                 lock_kernel();
 499                 err = reiserfs_commit_for_inode(inode);
 500                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 501                 unlock_kernel();
 502                 if (err < 0)
 503                         ret = err;
 504         }
 505       out:
 506         return ret;
 507 }
 508
 509 /*
 510 ** helper function for when reiserfs_get_block is called for a hole
 511 ** but the file tail is still in a direct item
 512 ** bh_result is the buffer head for the hole
 513 ** tail_offset is the offset of the start of the tail in the file
 514 **
 515 ** This calls prepare_write, which will start a new transaction
 516 ** you should not be in a transaction, or have any paths held when you
 517 ** call this.
 518 */
 519 static int convert_tail_for_hole(struct inode *inode,
 520                                  struct buffer_head *bh_result,
 521                                  loff_t tail_offset)
 522 {
 523         unsigned long index;
 524         unsigned long tail_end;
 525         unsigned long tail_start;
 526         struct page *tail_page;
 527         struct page *hole_page = bh_result->b_page;
 528         int retval = 0;
 529
 530         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 531                 return -EIO;
 532
 533         /* always try to read until the end of the block */
 534         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 535         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 536
 537         index = tail_offset >> PAGE_CACHE_SHIFT;
 538         /* hole_page can be zero in case of direct_io, we are sure
 539            that we cannot get here if we write with O_DIRECT into
 540            tail page */
 541         if (!hole_page || index != hole_page->index) {
 542                 tail_page = grab_cache_page(inode->i_mapping, index);
 543                 retval = -ENOMEM;
 544                 if (!tail_page) {
 545                         goto out;
 546                 }
 547         } else {
 548                 tail_page = hole_page;
 549         }
 550
 551         /* we don't have to make sure the conversion did not happen while
 552          ** we were locking the page because anyone that could convert
 553          ** must first take i_mutex.
 554          **
 555          ** We must fix the tail page for writing because it might have buffers
 556          ** that are mapped, but have a block number of 0.  This indicates tail
 557          ** data that has been read directly into the page, and block_prepare_write
 558          ** won't trigger a get_block in this case.
 559          */
 560         fix_tail_page_for_writing(tail_page);
 561         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
 562         if (retval)
 563                 goto unlock;
 564
 565         /* tail conversion might change the data in the page */
 566         flush_dcache_page(tail_page);
 567
 568         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 569
 570       unlock:
 571         if (tail_page != hole_page) {
 572                 unlock_page(tail_page);
 573                 page_cache_release(tail_page);
 574         }
 575       out:
 576         return retval;
 577 }
 578
 579 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 580                                   long block,
 581                                   struct inode *inode,
 582                                   b_blocknr_t * allocated_block_nr,
 583                                   struct path *path, int flags)
 584 {
 585         BUG_ON(!th->t_trans_id);
 586
 587 #ifdef REISERFS_PREALLOCATE
 588         if (!(flags & GET_BLOCK_NO_IMUX)) {
 589                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 590                                                   path, block);
 591         }
 592 #endif
 593         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 594                                          block);
 595 }
 596
 597 int reiserfs_get_block(struct inode *inode, sector_t block,
 598                        struct buffer_head *bh_result, int create)
 599 {
 600         int repeat, retval = 0;
 601         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 602         INITIALIZE_PATH(path);
 603         int pos_in_item;
 604         struct cpu_key key;
 605         struct buffer_head *bh, *unbh = NULL;
 606         struct item_head *ih, tmp_ih;
 607         __le32 *item;
 608         int done;
 609         int fs_gen;
 610         struct reiserfs_transaction_handle *th = NULL;
 611         /* space reserved in transaction batch:
 612            . 3 balancings in direct->indirect conversion
 613            . 1 block involved into reiserfs_update_sd()
 614            XXX in practically impossible worst case direct2indirect()
 615            can incur (much) more than 3 balancings.
 616            quota update for user, group */
 617         int jbegin_count =
 618             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 619             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 620         int version;
 621         int dangle = 1;
 622         loff_t new_offset =
 623             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 624
 625         /* bad.... */
 626         reiserfs_write_lock(inode->i_sb);
 627         version = get_inode_item_key_version(inode);
 628
 629         if (!file_capable(inode, block)) {
 630                 reiserfs_write_unlock(inode->i_sb);
 631                 return -EFBIG;
 632         }
 633
 634         /* if !create, we aren't changing the FS, so we don't need to
 635          ** log anything, so we don't need to start a transaction
 636          */
 637         if (!(create & GET_BLOCK_CREATE)) {
 638                 int ret;
 639                 /* find number of block-th logical block of the file */
 640                 ret = _get_block_create_0(inode, block, bh_result,
 641                                           create | GET_BLOCK_READ_DIRECT);
 642                 reiserfs_write_unlock(inode->i_sb);
 643                 return ret;
 644         }
 645         /*
 646          * if we're already in a transaction, make sure to close
 647          * any new transactions we start in this func
 648          */
 649         if ((create & GET_BLOCK_NO_DANGLE) ||
 650             reiserfs_transaction_running(inode->i_sb))
 651                 dangle = 0;
 652
 653         /* If file is of such a size, that it might have a tail and tails are enabled
 654          ** we should mark it as possibly needing tail packing on close
 655          */
 656         if ((have_large_tails(inode->i_sb)
 657              && inode->i_size < i_block_size(inode) * 4)
 658             || (have_small_tails(inode->i_sb)
 659                 && inode->i_size < i_block_size(inode)))
 660                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 661
 662         /* set the key of the first byte in the 'block'-th block of file */
 663         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 664         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 665               start_trans:
 666                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 667                 if (!th) {
 668                         retval = -ENOMEM;
 669                         goto failure;
 670                 }
 671                 reiserfs_update_inode_transaction(inode);
 672         }
 673       research:
 674
 675         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 676         if (retval == IO_ERROR) {
 677                 retval = -EIO;
 678                 goto failure;
 679         }
 680
 681         bh = get_last_bh(&path);
 682         ih = get_ih(&path);
 683         item = get_item(&path);
 684         pos_in_item = path.pos_in_item;
 685
 686         fs_gen = get_generation(inode->i_sb);
 687         copy_item_head(&tmp_ih, ih);
 688
 689         if (allocation_needed
 690             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 691                 /* we have to allocate block for the unformatted node */
 692                 if (!th) {
 693                         pathrelse(&path);
 694                         goto start_trans;
 695                 }
 696
 697                 repeat =
 698                     _allocate_block(th, block, inode, &allocated_block_nr,
 699                                     &path, create);
 700
 701                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 702                         /* restart the transaction to give the journal a chance to free
 703                          ** some blocks.  releases the path, so we have to go back to
 704                          ** research if we succeed on the second try
 705                          */
 706                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 707                         retval = restart_transaction(th, inode, &path);
 708                         if (retval)
 709                                 goto failure;
 710                         repeat =
 711                             _allocate_block(th, block, inode,
 712                                             &allocated_block_nr, NULL, create);
 713
 714                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 715                                 goto research;
 716                         }
 717                         if (repeat == QUOTA_EXCEEDED)
 718                                 retval = -EDQUOT;
 719                         else
 720                                 retval = -ENOSPC;
 721                         goto failure;
 722                 }
 723
 724                 if (fs_changed(fs_gen, inode->i_sb)
 725                     && item_moved(&tmp_ih, &path)) {
 726                         goto research;
 727                 }
 728         }
 729
 730         if (indirect_item_found(retval, ih)) {
 731                 b_blocknr_t unfm_ptr;
 732                 /* 'block'-th block is in the file already (there is
 733                    corresponding cell in some indirect item). But it may be
 734                    zero unformatted node pointer (hole) */
 735                 unfm_ptr = get_block_num(item, pos_in_item);
 736                 if (unfm_ptr == 0) {
 737                         /* use allocated block to plug the hole */
 738                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 739                         if (fs_changed(fs_gen, inode->i_sb)
 740                             && item_moved(&tmp_ih, &path)) {
 741                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 742                                                                  bh);
 743                                 goto research;
 744                         }
 745                         set_buffer_new(bh_result);
 746                         if (buffer_dirty(bh_result)
 747                             && reiserfs_data_ordered(inode->i_sb))
 748                                 reiserfs_add_ordered_list(inode, bh_result);
 749                         put_block_num(item, pos_in_item, allocated_block_nr);
 750                         unfm_ptr = allocated_block_nr;
 751                         journal_mark_dirty(th, inode->i_sb, bh);
 752                         reiserfs_update_sd(th, inode);
 753                 }
 754                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 755                 pathrelse(&path);
 756                 retval = 0;
 757                 if (!dangle && th)
 758                         retval = reiserfs_end_persistent_transaction(th);
 759
 760                 reiserfs_write_unlock(inode->i_sb);
 761
 762                 /* the item was found, so new blocks were not added to the file
 763                  ** there is no need to make sure the inode is updated with this
 764                  ** transaction
 765                  */
 766                 return retval;
 767         }
 768
 769         if (!th) {
 770                 pathrelse(&path);
 771                 goto start_trans;
 772         }
 773
 774         /* desired position is not found or is in the direct item. We have
 775            to append file with holes up to 'block'-th block converting
 776            direct items to indirect one if necessary */
 777         done = 0;
 778         do {
 779                 if (is_statdata_le_ih(ih)) {
 780                         __le32 unp = 0;
 781                         struct cpu_key tmp_key;
 782
 783                         /* indirect item has to be inserted */
 784                         make_le_item_head(&tmp_ih, &key, version, 1,
 785                                           TYPE_INDIRECT, UNFM_P_SIZE,
 786                                           0 /* free_space */ );
 787
 788                         if (cpu_key_k_offset(&key) == 1) {
 789                                 /* we are going to add 'block'-th block to the file. Use
 790                                    allocated block for that */
 791                                 unp = cpu_to_le32(allocated_block_nr);
 792                                 set_block_dev_mapped(bh_result,
 793                                                      allocated_block_nr, inode);
 794                                 set_buffer_new(bh_result);
 795                                 done = 1;
 796                         }
 797                         tmp_key = key;  // ;)
 798                         set_cpu_key_k_offset(&tmp_key, 1);
 799                         PATH_LAST_POSITION(&path)++;
 800
 801                         retval =
 802                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 803                                                  inode, (char *)&unp);
 804                         if (retval) {
 805                                 reiserfs_free_block(th, inode,
 806                                                     allocated_block_nr, 1);
 807                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 808                         }
 809                         //mark_tail_converted (inode);
 810                 } else if (is_direct_le_ih(ih)) {
 811                         /* direct item has to be converted */
 812                         loff_t tail_offset;
 813
 814                         tail_offset =
 815                             ((le_ih_k_offset(ih) -
 816                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 817                         if (tail_offset == cpu_key_k_offset(&key)) {
 818                                 /* direct item we just found fits into block we have
 819                                    to map. Convert it into unformatted node: use
 820                                    bh_result for the conversion */
 821                                 set_block_dev_mapped(bh_result,
 822                                                      allocated_block_nr, inode);
 823                                 unbh = bh_result;
 824                                 done = 1;
 825                         } else {
 826                                 /* we have to padd file tail stored in direct item(s)
 827                                    up to block size and convert it to unformatted
 828                                    node. FIXME: this should also get into page cache */
 829
 830                                 pathrelse(&path);
 831                                 /*
 832                                  * ugly, but we can only end the transaction if
 833                                  * we aren't nested
 834                                  */
 835                                 BUG_ON(!th->t_refcount);
 836                                 if (th->t_refcount == 1) {
 837                                         retval =
 838                                             reiserfs_end_persistent_transaction
 839                                             (th);
 840                                         th = NULL;
 841                                         if (retval)
 842                                                 goto failure;
 843                                 }
 844
 845                                 retval =
 846                                     convert_tail_for_hole(inode, bh_result,
 847                                                           tail_offset);
 848                                 if (retval) {
 849                                         if (retval != -ENOSPC)
 850                                                 reiserfs_warning(inode->i_sb,
 851                                                                  "clm-6004: convert tail failed inode %lu, error %d",
 852                                                                  inode->i_ino,
 853                                                                  retval);
 854                                         if (allocated_block_nr) {
 855                                                 /* the bitmap, the super, and the stat data == 3 */
 856                                                 if (!th)
 857                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 858                                                 if (th)
 859                                                         reiserfs_free_block(th,
 860                                                                             inode,
 861                                                                             allocated_block_nr,
 862                                                                             1);
 863                                         }
 864                                         goto failure;
 865                                 }
 866                                 goto research;
 867                         }
 868                         retval =
 869                             direct2indirect(th, inode, &path, unbh,
 870                                             tail_offset);
 871                         if (retval) {
 872                                 reiserfs_unmap_buffer(unbh);
 873                                 reiserfs_free_block(th, inode,
 874                                                     allocated_block_nr, 1);
 875                                 goto failure;
 876                         }
 877                         /* it is important the set_buffer_uptodate is done after
 878                          ** the direct2indirect.  The buffer might contain valid
 879                          ** data newer than the data on disk (read by readpage, changed,
 880                          ** and then sent here by writepage).  direct2indirect needs
 881                          ** to know if unbh was already up to date, so it can decide
 882                          ** if the data in unbh needs to be replaced with data from
 883                          ** the disk
 884                          */
 885                         set_buffer_uptodate(unbh);
 886
 887                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 888                            buffer will disappear shortly, so it should not be added to
 889                          */
 890                         if (unbh->b_page) {
 891                                 /* we've converted the tail, so we must
 892                                  ** flush unbh before the transaction commits
 893                                  */
 894                                 reiserfs_add_tail_list(inode, unbh);
 895
 896                                 /* mark it dirty now to prevent commit_write from adding
 897                                  ** this buffer to the inode's dirty buffer list
 898                                  */
 899                                 /*
 900                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 901                                  * It's still atomic, but it sets the page dirty too,
 902                                  * which makes it eligible for writeback at any time by the
 903                                  * VM (which was also the case with __mark_buffer_dirty())
 904                                  */
 905                                 mark_buffer_dirty(unbh);
 906                         }
 907                 } else {
 908                         /* append indirect item with holes if needed, when appending
 909                            pointer to 'block'-th block use block, which is already
 910                            allocated */
 911                         struct cpu_key tmp_key;
 912                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 913                         // one block which is a fastpath
 914                         unp_t *un;
 915                         __u64 max_to_insert =
 916                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 917                             UNFM_P_SIZE;
 918                         __u64 blocks_needed;
 919
 920                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 921                                "vs-804: invalid position for append");
 922                         /* indirect item has to be appended, set up key of that position */
 923                         make_cpu_key(&tmp_key, inode,
 924                                      le_key_k_offset(version,
 925                                                      &(ih->ih_key)) +
 926                                      op_bytes_number(ih,
 927                                                      inode->i_sb->s_blocksize),
 928                                      //pos_in_item * inode->i_sb->s_blocksize,
 929                                      TYPE_INDIRECT, 3); // key type is unimportant
 930
 931                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 932                                "green-805: invalid offset");
 933                         blocks_needed =
 934                             1 +
 935                             ((cpu_key_k_offset(&key) -
 936                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 937                              s_blocksize_bits);
 938
 939                         if (blocks_needed == 1) {
 940                                 un = &unf_single;
 941                         } else {
 942                                 un = kmalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
 943                                 if (!un) {
 944                                         un = &unf_single;
 945                                         blocks_needed = 1;
 946                                         max_to_insert = 0;
 947                                 } else
 948                                         memset(un, 0,
 949                                                UNFM_P_SIZE * min(blocks_needed,
 950                                                                  max_to_insert));
 951                         }
 952                         if (blocks_needed <= max_to_insert) {
 953                                 /* we are going to add target block to the file. Use allocated
 954                                    block for that */
 955                                 un[blocks_needed - 1] =
 956                                     cpu_to_le32(allocated_block_nr);
 957                                 set_block_dev_mapped(bh_result,
 958                                                      allocated_block_nr, inode);
 959                                 set_buffer_new(bh_result);
 960                                 done = 1;
 961                         } else {
 962                                 /* paste hole to the indirect item */
 963                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 964                                    only have space for one block */
 965                                 blocks_needed =
 966                                     max_to_insert ? max_to_insert : 1;
 967                         }
 968                         retval =
 969                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 970                                                      (char *)un,
 971                                                      UNFM_P_SIZE *
 972                                                      blocks_needed);
 973
 974                         if (blocks_needed != 1)
 975                                 kfree(un);
 976
 977                         if (retval) {
 978                                 reiserfs_free_block(th, inode,
 979                                                     allocated_block_nr, 1);
 980                                 goto failure;
 981                         }
 982                         if (!done) {
 983                                 /* We need to mark new file size in case this function will be
 984                                    interrupted/aborted later on. And we may do this only for
 985                                    holes. */
 986                                 inode->i_size +=
 987                                     inode->i_sb->s_blocksize * blocks_needed;
 988                         }
 989                 }
 990
 991                 if (done == 1)
 992                         break;
 993
 994                 /* this loop could log more blocks than we had originally asked
 995                  ** for.  So, we have to allow the transaction to end if it is
 996                  ** too big or too full.  Update the inode so things are
 997                  ** consistent if we crash before the function returns
 998                  **
 999                  ** release the path so that anybody waiting on the path before
1000                  ** ending their transaction will be able to continue.
1001                  */
1002                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1003                         retval = restart_transaction(th, inode, &path);
1004                         if (retval)
1005                                 goto failure;
1006                 }
1007                 /* inserting indirect pointers for a hole can take a
1008                  ** long time.  reschedule if needed
1009                  */
1010                 cond_resched();
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,
1019                                          "vs-825: reiserfs_get_block: "
1020                                          "%K should not be found", &key);
1021                         retval = -EEXIST;
1022                         if (allocated_block_nr)
1023                                 reiserfs_free_block(th, inode,
1024                                                     allocated_block_nr, 1);
1025                         pathrelse(&path);
1026                         goto failure;
1027                 }
1028                 bh = get_last_bh(&path);
1029                 ih = get_ih(&path);
1030                 item = get_item(&path);
1031                 pos_in_item = path.pos_in_item;
1032         } while (1);
1033
1034         retval = 0;
1035
1036       failure:
1037         if (th && (!dangle || (retval && !th->t_trans_id))) {
1038                 int err;
1039                 if (th->t_trans_id)
1040                         reiserfs_update_sd(th, inode);
1041                 err = reiserfs_end_persistent_transaction(th);
1042                 if (err)
1043                         retval = err;
1044         }
1045
1046         reiserfs_write_unlock(inode->i_sb);
1047         reiserfs_check_path(&path);
1048         return retval;
1049 }
1050
1051 static int
1052 reiserfs_readpages(struct file *file, struct address_space *mapping,
1053                    struct list_head *pages, unsigned nr_pages)
1054 {
1055         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1056 }
1057
1058 /* Compute real number of used bytes by file
1059  * Following three functions can go away when we'll have enough space in stat item
1060  */
1061 static int real_space_diff(struct inode *inode, int sd_size)
1062 {
1063         int bytes;
1064         loff_t blocksize = inode->i_sb->s_blocksize;
1065
1066         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1067                 return sd_size;
1068
1069         /* End of file is also in full block with indirect reference, so round
1070          ** up to the next block.
1071          **
1072          ** there is just no way to know if the tail is actually packed
1073          ** on the file, so we have to assume it isn't.  When we pack the
1074          ** tail, we add 4 bytes to pretend there really is an unformatted
1075          ** node pointer
1076          */
1077         bytes =
1078             ((inode->i_size +
1079               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1080             sd_size;
1081         return bytes;
1082 }
1083
1084 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1085                                         int sd_size)
1086 {
1087         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1088                 return inode->i_size +
1089                     (loff_t) (real_space_diff(inode, sd_size));
1090         }
1091         return ((loff_t) real_space_diff(inode, sd_size)) +
1092             (((loff_t) blocks) << 9);
1093 }
1094
1095 /* Compute number of blocks used by file in ReiserFS counting */
1096 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1097 {
1098         loff_t bytes = inode_get_bytes(inode);
1099         loff_t real_space = real_space_diff(inode, sd_size);
1100
1101         /* keeps fsck and non-quota versions of reiserfs happy */
1102         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1103                 bytes += (loff_t) 511;
1104         }
1105
1106         /* files from before the quota patch might i_blocks such that
1107          ** bytes < real_space.  Deal with that here to prevent it from
1108          ** going negative.
1109          */
1110         if (bytes < real_space)
1111                 return 0;
1112         return (bytes - real_space) >> 9;
1113 }
1114
1115 //
1116 // BAD: new directories have stat data of new type and all other items
1117 // of old type. Version stored in the inode says about body items, so
1118 // in update_stat_data we can not rely on inode, but have to check
1119 // item version directly
1120 //
1121
1122 // called by read_locked_inode
1123 static void init_inode(struct inode *inode, struct path *path)
1124 {
1125         struct buffer_head *bh;
1126         struct item_head *ih;
1127         __u32 rdev;
1128         //int version = ITEM_VERSION_1;
1129
1130         bh = PATH_PLAST_BUFFER(path);
1131         ih = PATH_PITEM_HEAD(path);
1132
1133         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1134         inode->i_blksize = reiserfs_default_io_size;
1135
1136         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1137         REISERFS_I(inode)->i_flags = 0;
1138         REISERFS_I(inode)->i_prealloc_block = 0;
1139         REISERFS_I(inode)->i_prealloc_count = 0;
1140         REISERFS_I(inode)->i_trans_id = 0;
1141         REISERFS_I(inode)->i_jl = NULL;
1142         REISERFS_I(inode)->i_acl_access = NULL;
1143         REISERFS_I(inode)->i_acl_default = NULL;
1144         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1145
1146         if (stat_data_v1(ih)) {
1147                 struct stat_data_v1 *sd =
1148                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1149                 unsigned long blocks;
1150
1151                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1152                 set_inode_sd_version(inode, STAT_DATA_V1);
1153                 inode->i_mode = sd_v1_mode(sd);
1154                 inode->i_nlink = sd_v1_nlink(sd);
1155                 inode->i_uid = sd_v1_uid(sd);
1156                 inode->i_gid = sd_v1_gid(sd);
1157                 inode->i_size = sd_v1_size(sd);
1158                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1159                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1160                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1161                 inode->i_atime.tv_nsec = 0;
1162                 inode->i_ctime.tv_nsec = 0;
1163                 inode->i_mtime.tv_nsec = 0;
1164
1165                 inode->i_blocks = sd_v1_blocks(sd);
1166                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1167                 blocks = (inode->i_size + 511) >> 9;
1168                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1169                 if (inode->i_blocks > blocks) {
1170                         // there was a bug in <=3.5.23 when i_blocks could take negative
1171                         // values. Starting from 3.5.17 this value could even be stored in
1172                         // stat data. For such files we set i_blocks based on file
1173                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1174                         // only updated if file's inode will ever change
1175                         inode->i_blocks = blocks;
1176                 }
1177
1178                 rdev = sd_v1_rdev(sd);
1179                 REISERFS_I(inode)->i_first_direct_byte =
1180                     sd_v1_first_direct_byte(sd);
1181                 /* an early bug in the quota code can give us an odd number for the
1182                  ** block count.  This is incorrect, fix it here.
1183                  */
1184                 if (inode->i_blocks & 1) {
1185                         inode->i_blocks++;
1186                 }
1187                 inode_set_bytes(inode,
1188                                 to_real_used_space(inode, inode->i_blocks,
1189                                                    SD_V1_SIZE));
1190                 /* nopack is initially zero for v1 objects. For v2 objects,
1191                    nopack is initialised from sd_attrs */
1192                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1193         } else {
1194                 // new stat data found, but object may have old items
1195                 // (directories and symlinks)
1196                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1197
1198                 inode->i_mode = sd_v2_mode(sd);
1199                 inode->i_nlink = sd_v2_nlink(sd);
1200                 inode->i_uid = sd_v2_uid(sd);
1201                 inode->i_size = sd_v2_size(sd);
1202                 inode->i_gid = sd_v2_gid(sd);
1203                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1204                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1205                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1206                 inode->i_ctime.tv_nsec = 0;
1207                 inode->i_mtime.tv_nsec = 0;
1208                 inode->i_atime.tv_nsec = 0;
1209                 inode->i_blocks = sd_v2_blocks(sd);
1210                 rdev = sd_v2_rdev(sd);
1211                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1212                         inode->i_generation =
1213                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1214                 else
1215                         inode->i_generation = sd_v2_generation(sd);
1216
1217                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1218                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1219                 else
1220                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1221                 REISERFS_I(inode)->i_first_direct_byte = 0;
1222                 set_inode_sd_version(inode, STAT_DATA_V2);
1223                 inode_set_bytes(inode,
1224                                 to_real_used_space(inode, inode->i_blocks,
1225                                                    SD_V2_SIZE));
1226                 /* read persistent inode attributes from sd and initalise
1227                    generic inode flags from them */
1228                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1229                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1230         }
1231
1232         pathrelse(path);
1233         if (S_ISREG(inode->i_mode)) {
1234                 inode->i_op = &reiserfs_file_inode_operations;
1235                 inode->i_fop = &reiserfs_file_operations;
1236                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1237         } else if (S_ISDIR(inode->i_mode)) {
1238                 inode->i_op = &reiserfs_dir_inode_operations;
1239                 inode->i_fop = &reiserfs_dir_operations;
1240         } else if (S_ISLNK(inode->i_mode)) {
1241                 inode->i_op = &reiserfs_symlink_inode_operations;
1242                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1243         } else {
1244                 inode->i_blocks = 0;
1245                 inode->i_op = &reiserfs_special_inode_operations;
1246                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1247         }
1248 }
1249
1250 // update new stat data with inode fields
1251 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1252 {
1253         struct stat_data *sd_v2 = (struct stat_data *)sd;
1254         __u16 flags;
1255
1256         set_sd_v2_mode(sd_v2, inode->i_mode);
1257         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1258         set_sd_v2_uid(sd_v2, inode->i_uid);
1259         set_sd_v2_size(sd_v2, size);
1260         set_sd_v2_gid(sd_v2, inode->i_gid);
1261         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1262         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1263         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1264         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1265         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1266                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1267         else
1268                 set_sd_v2_generation(sd_v2, inode->i_generation);
1269         flags = REISERFS_I(inode)->i_attrs;
1270         i_attrs_to_sd_attrs(inode, &flags);
1271         set_sd_v2_attrs(sd_v2, flags);
1272 }
1273
1274 // used to copy inode's fields to old stat data
1275 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1276 {
1277         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1278
1279         set_sd_v1_mode(sd_v1, inode->i_mode);
1280         set_sd_v1_uid(sd_v1, inode->i_uid);
1281         set_sd_v1_gid(sd_v1, inode->i_gid);
1282         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1283         set_sd_v1_size(sd_v1, size);
1284         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1285         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1286         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1287
1288         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1289                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1290         else
1291                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1292
1293         // Sigh. i_first_direct_byte is back
1294         set_sd_v1_first_direct_byte(sd_v1,
1295                                     REISERFS_I(inode)->i_first_direct_byte);
1296 }
1297
1298 /* NOTE, you must prepare the buffer head before sending it here,
1299 ** and then log it after the call
1300 */
1301 static void update_stat_data(struct path *path, struct inode *inode,
1302                              loff_t size)
1303 {
1304         struct buffer_head *bh;
1305         struct item_head *ih;
1306
1307         bh = PATH_PLAST_BUFFER(path);
1308         ih = PATH_PITEM_HEAD(path);
1309
1310         if (!is_statdata_le_ih(ih))
1311                 reiserfs_panic(inode->i_sb,
1312                                "vs-13065: update_stat_data: key %k, found item %h",
1313                                INODE_PKEY(inode), ih);
1314
1315         if (stat_data_v1(ih)) {
1316                 // path points to old stat data
1317                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1318         } else {
1319                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1320         }
1321
1322         return;
1323 }
1324
1325 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1326                              struct inode *inode, loff_t size)
1327 {
1328         struct cpu_key key;
1329         INITIALIZE_PATH(path);
1330         struct buffer_head *bh;
1331         int fs_gen;
1332         struct item_head *ih, tmp_ih;
1333         int retval;
1334
1335         BUG_ON(!th->t_trans_id);
1336
1337         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1338
1339         for (;;) {
1340                 int pos;
1341                 /* look for the object's stat data */
1342                 retval = search_item(inode->i_sb, &key, &path);
1343                 if (retval == IO_ERROR) {
1344                         reiserfs_warning(inode->i_sb,
1345                                          "vs-13050: reiserfs_update_sd: "
1346                                          "i/o failure occurred trying to update %K stat data",
1347                                          &key);
1348                         return;
1349                 }
1350                 if (retval == ITEM_NOT_FOUND) {
1351                         pos = PATH_LAST_POSITION(&path);
1352                         pathrelse(&path);
1353                         if (inode->i_nlink == 0) {
1354                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1355                                 return;
1356                         }
1357                         reiserfs_warning(inode->i_sb,
1358                                          "vs-13060: reiserfs_update_sd: "
1359                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1360                                          INODE_PKEY(inode), inode->i_nlink,
1361                                          pos);
1362                         reiserfs_check_path(&path);
1363                         return;
1364                 }
1365
1366                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1367                  ** FS might change.  We have to detect that, and loop back to the
1368                  ** search if the stat data item has moved
1369                  */
1370                 bh = get_last_bh(&path);
1371                 ih = get_ih(&path);
1372                 copy_item_head(&tmp_ih, ih);
1373                 fs_gen = get_generation(inode->i_sb);
1374                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1375                 if (fs_changed(fs_gen, inode->i_sb)
1376                     && item_moved(&tmp_ih, &path)) {
1377                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1378                         continue;       /* Stat_data item has been moved after scheduling. */
1379                 }
1380                 break;
1381         }
1382         update_stat_data(&path, inode, size);
1383         journal_mark_dirty(th, th->t_super, bh);
1384         pathrelse(&path);
1385         return;
1386 }
1387
1388 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1389 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1390 ** and clear the key in the private portion of the inode, otherwise a
1391 ** corresponding iput might try to delete whatever object the inode last
1392 ** represented.
1393 */
1394 static void reiserfs_make_bad_inode(struct inode *inode)
1395 {
1396         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1397         make_bad_inode(inode);
1398 }
1399
1400 //
1401 // initially this function was derived from minix or ext2's analog and
1402 // evolved as the prototype did
1403 //
1404
1405 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1406 {
1407         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1408         inode->i_ino = args->objectid;
1409         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1410         return 0;
1411 }
1412
1413 /* looks for stat data in the tree, and fills up the fields of in-core
1414    inode stat data fields */
1415 void reiserfs_read_locked_inode(struct inode *inode,
1416                                 struct reiserfs_iget_args *args)
1417 {
1418         INITIALIZE_PATH(path_to_sd);
1419         struct cpu_key key;
1420         unsigned long dirino;
1421         int retval;
1422
1423         dirino = args->dirid;
1424
1425         /* set version 1, version 2 could be used too, because stat data
1426            key is the same in both versions */
1427         key.version = KEY_FORMAT_3_5;
1428         key.on_disk_key.k_dir_id = dirino;
1429         key.on_disk_key.k_objectid = inode->i_ino;
1430         key.on_disk_key.k_offset = 0;
1431         key.on_disk_key.k_type = 0;
1432
1433         /* look for the object's stat data */
1434         retval = search_item(inode->i_sb, &key, &path_to_sd);
1435         if (retval == IO_ERROR) {
1436                 reiserfs_warning(inode->i_sb,
1437                                  "vs-13070: reiserfs_read_locked_inode: "
1438                                  "i/o failure occurred trying to find stat data of %K",
1439                                  &key);
1440                 reiserfs_make_bad_inode(inode);
1441                 return;
1442         }
1443         if (retval != ITEM_FOUND) {
1444                 /* a stale NFS handle can trigger this without it being an error */
1445                 pathrelse(&path_to_sd);
1446                 reiserfs_make_bad_inode(inode);
1447                 inode->i_nlink = 0;
1448                 return;
1449         }
1450
1451         init_inode(inode, &path_to_sd);
1452
1453         /* It is possible that knfsd is trying to access inode of a file
1454            that is being removed from the disk by some other thread. As we
1455            update sd on unlink all that is required is to check for nlink
1456            here. This bug was first found by Sizif when debugging
1457            SquidNG/Butterfly, forgotten, and found again after Philippe
1458            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1459
1460            More logical fix would require changes in fs/inode.c:iput() to
1461            remove inode from hash-table _after_ fs cleaned disk stuff up and
1462            in iget() to return NULL if I_FREEING inode is found in
1463            hash-table. */
1464         /* Currently there is one place where it's ok to meet inode with
1465            nlink==0: processing of open-unlinked and half-truncated files
1466            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1467         if ((inode->i_nlink == 0) &&
1468             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1469                 reiserfs_warning(inode->i_sb,
1470                                  "vs-13075: reiserfs_read_locked_inode: "
1471                                  "dead inode read from disk %K. "
1472                                  "This is likely to be race with knfsd. Ignore",
1473                                  &key);
1474                 reiserfs_make_bad_inode(inode);
1475         }
1476
1477         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1478
1479 }
1480
1481 /**
1482  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1483  *
1484  * @inode:    inode from hash table to check
1485  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1486  *
1487  * This function is called by iget5_locked() to distinguish reiserfs inodes
1488  * having the same inode numbers. Such inodes can only exist due to some
1489  * error condition. One of them should be bad. Inodes with identical
1490  * inode numbers (objectids) are distinguished by parent directory ids.
1491  *
1492  */
1493 int reiserfs_find_actor(struct inode *inode, void *opaque)
1494 {
1495         struct reiserfs_iget_args *args;
1496
1497         args = opaque;
1498         /* args is already in CPU order */
1499         return (inode->i_ino == args->objectid) &&
1500             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1501 }
1502
1503 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1504 {
1505         struct inode *inode;
1506         struct reiserfs_iget_args args;
1507
1508         args.objectid = key->on_disk_key.k_objectid;
1509         args.dirid = key->on_disk_key.k_dir_id;
1510         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1511                              reiserfs_find_actor, reiserfs_init_locked_inode,
1512                              (void *)(&args));
1513         if (!inode)
1514                 return ERR_PTR(-ENOMEM);
1515
1516         if (inode->i_state & I_NEW) {
1517                 reiserfs_read_locked_inode(inode, &args);
1518                 unlock_new_inode(inode);
1519         }
1520
1521         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1522                 /* either due to i/o error or a stale NFS handle */
1523                 iput(inode);
1524                 inode = NULL;
1525         }
1526         return inode;
1527 }
1528
1529 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1530 {
1531         __u32 *data = vobjp;
1532         struct cpu_key key;
1533         struct dentry *result;
1534         struct inode *inode;
1535
1536         key.on_disk_key.k_objectid = data[0];
1537         key.on_disk_key.k_dir_id = data[1];
1538         reiserfs_write_lock(sb);
1539         inode = reiserfs_iget(sb, &key);
1540         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1541             data[2] != inode->i_generation) {
1542                 iput(inode);
1543                 inode = NULL;
1544         }
1545         reiserfs_write_unlock(sb);
1546         if (!inode)
1547                 inode = ERR_PTR(-ESTALE);
1548         if (IS_ERR(inode))
1549                 return ERR_PTR(PTR_ERR(inode));
1550         result = d_alloc_anon(inode);
1551         if (!result) {
1552                 iput(inode);
1553                 return ERR_PTR(-ENOMEM);
1554         }
1555         return result;
1556 }
1557
1558 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1559                                   int len, int fhtype,
1560                                   int (*acceptable) (void *contect,
1561                                                      struct dentry * de),
1562                                   void *context)
1563 {
1564         __u32 obj[3], parent[3];
1565
1566         /* fhtype happens to reflect the number of u32s encoded.
1567          * due to a bug in earlier code, fhtype might indicate there
1568          * are more u32s then actually fitted.
1569          * so if fhtype seems to be more than len, reduce fhtype.
1570          * Valid types are:
1571          *   2 - objectid + dir_id - legacy support
1572          *   3 - objectid + dir_id + generation
1573          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1574          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1575          *   6 - as above plus generation of directory
1576          * 6 does not fit in NFSv2 handles
1577          */
1578         if (fhtype > len) {
1579                 if (fhtype != 6 || len != 5)
1580                         reiserfs_warning(sb,
1581                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1582                                          fhtype, len);
1583                 fhtype = 5;
1584         }
1585
1586         obj[0] = data[0];
1587         obj[1] = data[1];
1588         if (fhtype == 3 || fhtype >= 5)
1589                 obj[2] = data[2];
1590         else
1591                 obj[2] = 0;     /* generation number */
1592
1593         if (fhtype >= 4) {
1594                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1595                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1596                 if (fhtype == 6)
1597                         parent[2] = data[5];
1598                 else
1599                         parent[2] = 0;
1600         }
1601         return sb->s_export_op->find_exported_dentry(sb, obj,
1602                                                      fhtype < 4 ? NULL : parent,
1603                                                      acceptable, context);
1604 }
1605
1606 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1607                        int need_parent)
1608 {
1609         struct inode *inode = dentry->d_inode;
1610         int maxlen = *lenp;
1611
1612         if (maxlen < 3)
1613                 return 255;
1614
1615         data[0] = inode->i_ino;
1616         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1617         data[2] = inode->i_generation;
1618         *lenp = 3;
1619         /* no room for directory info? return what we've stored so far */
1620         if (maxlen < 5 || !need_parent)
1621                 return 3;
1622
1623         spin_lock(&dentry->d_lock);
1624         inode = dentry->d_parent->d_inode;
1625         data[3] = inode->i_ino;
1626         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1627         *lenp = 5;
1628         if (maxlen >= 6) {
1629                 data[5] = inode->i_generation;
1630                 *lenp = 6;
1631         }
1632         spin_unlock(&dentry->d_lock);
1633         return *lenp;
1634 }
1635
1636 /* looks for stat data, then copies fields to it, marks the buffer
1637    containing stat data as dirty */
1638 /* reiserfs inodes are never really dirty, since the dirty inode call
1639 ** always logs them.  This call allows the VFS inode marking routines
1640 ** to properly mark inodes for datasync and such, but only actually
1641 ** does something when called for a synchronous update.
1642 */
1643 int reiserfs_write_inode(struct inode *inode, int do_sync)
1644 {
1645         struct reiserfs_transaction_handle th;
1646         int jbegin_count = 1;
1647
1648         if (inode->i_sb->s_flags & MS_RDONLY)
1649                 return -EROFS;
1650         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1651          ** these cases are just when the system needs ram, not when the
1652          ** inode needs to reach disk for safety, and they can safely be
1653          ** ignored because the altered inode has already been logged.
1654          */
1655         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1656                 reiserfs_write_lock(inode->i_sb);
1657                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1658                         reiserfs_update_sd(&th, inode);
1659                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1660                 }
1661                 reiserfs_write_unlock(inode->i_sb);
1662         }
1663         return 0;
1664 }
1665
1666 /* stat data of new object is inserted already, this inserts the item
1667    containing "." and ".." entries */
1668 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1669                                   struct inode *inode,
1670                                   struct item_head *ih, struct path *path,
1671                                   struct inode *dir)
1672 {
1673         struct super_block *sb = th->t_super;
1674         char empty_dir[EMPTY_DIR_SIZE];
1675         char *body = empty_dir;
1676         struct cpu_key key;
1677         int retval;
1678
1679         BUG_ON(!th->t_trans_id);
1680
1681         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1682                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1683                       TYPE_DIRENTRY, 3 /*key length */ );
1684
1685         /* compose item head for new item. Directories consist of items of
1686            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1687            is done by reiserfs_new_inode */
1688         if (old_format_only(sb)) {
1689                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1690                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1691
1692                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1693                                        ih->ih_key.k_objectid,
1694                                        INODE_PKEY(dir)->k_dir_id,
1695                                        INODE_PKEY(dir)->k_objectid);
1696         } else {
1697                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1698                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1699
1700                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1701                                     ih->ih_key.k_objectid,
1702                                     INODE_PKEY(dir)->k_dir_id,
1703                                     INODE_PKEY(dir)->k_objectid);
1704         }
1705
1706         /* look for place in the tree for new item */
1707         retval = search_item(sb, &key, path);
1708         if (retval == IO_ERROR) {
1709                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1710                                  "i/o failure occurred creating new directory");
1711                 return -EIO;
1712         }
1713         if (retval == ITEM_FOUND) {
1714                 pathrelse(path);
1715                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1716                                  "object with this key exists (%k)",
1717                                  &(ih->ih_key));
1718                 return -EEXIST;
1719         }
1720
1721         /* insert item, that is empty directory item */
1722         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1723 }
1724
1725 /* stat data of object has been inserted, this inserts the item
1726    containing the body of symlink */
1727 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1728                                 struct item_head *ih,
1729                                 struct path *path, const char *symname,
1730                                 int item_len)
1731 {
1732         struct super_block *sb = th->t_super;
1733         struct cpu_key key;
1734         int retval;
1735
1736         BUG_ON(!th->t_trans_id);
1737
1738         _make_cpu_key(&key, KEY_FORMAT_3_5,
1739                       le32_to_cpu(ih->ih_key.k_dir_id),
1740                       le32_to_cpu(ih->ih_key.k_objectid),
1741                       1, TYPE_DIRECT, 3 /*key length */ );
1742
1743         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1744                           0 /*free_space */ );
1745
1746         /* look for place in the tree for new item */
1747         retval = search_item(sb, &key, path);
1748         if (retval == IO_ERROR) {
1749                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1750                                  "i/o failure occurred creating new symlink");
1751                 return -EIO;
1752         }
1753         if (retval == ITEM_FOUND) {
1754                 pathrelse(path);
1755                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1756                                  "object with this key exists (%k)",
1757                                  &(ih->ih_key));
1758                 return -EEXIST;
1759         }
1760
1761         /* insert item, that is body of symlink */
1762         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1763 }
1764
1765 /* inserts the stat data into the tree, and then calls
1766    reiserfs_new_directory (to insert ".", ".." item if new object is
1767    directory) or reiserfs_new_symlink (to insert symlink body if new
1768    object is symlink) or nothing (if new object is regular file)
1769
1770    NOTE! uid and gid must already be set in the inode.  If we return
1771    non-zero due to an error, we have to drop the quota previously allocated
1772    for the fresh inode.  This can only be done outside a transaction, so
1773    if we return non-zero, we also end the transaction.  */
1774 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1775                        struct inode *dir, int mode, const char *symname,
1776                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1777                           strlen (symname) for symlinks) */
1778                        loff_t i_size, struct dentry *dentry,
1779                        struct inode *inode)
1780 {
1781         struct super_block *sb;
1782         INITIALIZE_PATH(path_to_key);
1783         struct cpu_key key;
1784         struct item_head ih;
1785         struct stat_data sd;
1786         int retval;
1787         int err;
1788
1789         BUG_ON(!th->t_trans_id);
1790
1791         if (DQUOT_ALLOC_INODE(inode)) {
1792                 err = -EDQUOT;
1793                 goto out_end_trans;
1794         }
1795         if (!dir || !dir->i_nlink) {
1796                 err = -EPERM;
1797                 goto out_bad_inode;
1798         }
1799
1800         sb = dir->i_sb;
1801
1802         /* item head of new item */
1803         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1804         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1805         if (!ih.ih_key.k_objectid) {
1806                 err = -ENOMEM;
1807                 goto out_bad_inode;
1808         }
1809         if (old_format_only(sb))
1810                 /* not a perfect generation count, as object ids can be reused, but
1811                  ** this is as good as reiserfs can do right now.
1812                  ** note that the private part of inode isn't filled in yet, we have
1813                  ** to use the directory.
1814                  */
1815                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1816         else
1817 #if defined( USE_INODE_GENERATION_COUNTER )
1818                 inode->i_generation =
1819                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1820 #else
1821                 inode->i_generation = ++event;
1822 #endif
1823
1824         /* fill stat data */
1825         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1826
1827         /* uid and gid must already be set by the caller for quota init */
1828
1829         /* symlink cannot be immutable or append only, right? */
1830         if (S_ISLNK(inode->i_mode))
1831                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1832
1833         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1834         inode->i_size = i_size;
1835         inode->i_blocks = 0;
1836         inode->i_bytes = 0;
1837         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1838             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1839
1840         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1841         REISERFS_I(inode)->i_flags = 0;
1842         REISERFS_I(inode)->i_prealloc_block = 0;
1843         REISERFS_I(inode)->i_prealloc_count = 0;
1844         REISERFS_I(inode)->i_trans_id = 0;
1845         REISERFS_I(inode)->i_jl = NULL;
1846         REISERFS_I(inode)->i_attrs =
1847             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1848         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1849         REISERFS_I(inode)->i_acl_access = NULL;
1850         REISERFS_I(inode)->i_acl_default = NULL;
1851         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1852
1853         if (old_format_only(sb))
1854                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1855                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1856         else
1857                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1858                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1859
1860         /* key to search for correct place for new stat data */
1861         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1862                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1863                       TYPE_STAT_DATA, 3 /*key length */ );
1864
1865         /* find proper place for inserting of stat data */
1866         retval = search_item(sb, &key, &path_to_key);
1867         if (retval == IO_ERROR) {
1868                 err = -EIO;
1869                 goto out_bad_inode;
1870         }
1871         if (retval == ITEM_FOUND) {
1872                 pathrelse(&path_to_key);
1873                 err = -EEXIST;
1874                 goto out_bad_inode;
1875         }
1876         if (old_format_only(sb)) {
1877                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1878                         pathrelse(&path_to_key);
1879                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1880                         err = -EINVAL;
1881                         goto out_bad_inode;
1882                 }
1883                 inode2sd_v1(&sd, inode, inode->i_size);
1884         } else {
1885                 inode2sd(&sd, inode, inode->i_size);
1886         }
1887         // these do not go to on-disk stat data
1888         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1889         inode->i_blksize = reiserfs_default_io_size;
1890
1891         // store in in-core inode the key of stat data and version all
1892         // object items will have (directory items will have old offset
1893         // format, other new objects will consist of new items)
1894         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1895         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1896                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1897         else
1898                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1899         if (old_format_only(sb))
1900                 set_inode_sd_version(inode, STAT_DATA_V1);
1901         else
1902                 set_inode_sd_version(inode, STAT_DATA_V2);
1903
1904         /* insert the stat data into the tree */
1905 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1906         if (REISERFS_I(dir)->new_packing_locality)
1907                 th->displace_new_blocks = 1;
1908 #endif
1909         retval =
1910             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1911                                  (char *)(&sd));
1912         if (retval) {
1913                 err = retval;
1914                 reiserfs_check_path(&path_to_key);
1915                 goto out_bad_inode;
1916         }
1917 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1918         if (!th->displace_new_blocks)
1919                 REISERFS_I(dir)->new_packing_locality = 0;
1920 #endif
1921         if (S_ISDIR(mode)) {
1922                 /* insert item with "." and ".." */
1923                 retval =
1924                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1925         }
1926
1927         if (S_ISLNK(mode)) {
1928                 /* insert body of symlink */
1929                 if (!old_format_only(sb))
1930                         i_size = ROUND_UP(i_size);
1931                 retval =
1932                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1933                                          i_size);
1934         }
1935         if (retval) {
1936                 err = retval;
1937                 reiserfs_check_path(&path_to_key);
1938                 journal_end(th, th->t_super, th->t_blocks_allocated);
1939                 goto out_inserted_sd;
1940         }
1941
1942         /* XXX CHECK THIS */
1943         if (reiserfs_posixacl(inode->i_sb)) {
1944                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1945                 if (retval) {
1946                         err = retval;
1947                         reiserfs_check_path(&path_to_key);
1948                         journal_end(th, th->t_super, th->t_blocks_allocated);
1949                         goto out_inserted_sd;
1950                 }
1951         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1952                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1953                                  "but vfs thinks they are!");
1954         } else if (is_reiserfs_priv_object(dir)) {
1955                 reiserfs_mark_inode_private(inode);
1956         }
1957
1958         insert_inode_hash(inode);
1959         reiserfs_update_sd(th, inode);
1960         reiserfs_check_path(&path_to_key);
1961
1962         return 0;
1963
1964 /* it looks like you can easily compress these two goto targets into
1965  * one.  Keeping it like this doesn't actually hurt anything, and they
1966  * are place holders for what the quota code actually needs.
1967  */
1968       out_bad_inode:
1969         /* Invalidate the object, nothing was inserted yet */
1970         INODE_PKEY(inode)->k_objectid = 0;
1971
1972         /* Quota change must be inside a transaction for journaling */
1973         DQUOT_FREE_INODE(inode);
1974
1975       out_end_trans:
1976         journal_end(th, th->t_super, th->t_blocks_allocated);
1977         /* Drop can be outside and it needs more credits so it's better to have it outside */
1978         DQUOT_DROP(inode);
1979         inode->i_flags |= S_NOQUOTA;
1980         make_bad_inode(inode);
1981
1982       out_inserted_sd:
1983         inode->i_nlink = 0;
1984         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1985
1986         /* If we were inheriting an ACL, we need to release the lock so that
1987          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1988          * code really needs to be reworked, but this will take care of it
1989          * for now. -jeffm */
1990         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1991                 reiserfs_write_unlock_xattrs(dir->i_sb);
1992                 iput(inode);
1993                 reiserfs_write_lock_xattrs(dir->i_sb);
1994         } else
1995                 iput(inode);
1996         return err;
1997 }
1998
1999 /*
2000 ** finds the tail page in the page cache,
2001 ** reads the last block in.
2002 **
2003 ** On success, page_result is set to a locked, pinned page, and bh_result
2004 ** is set to an up to date buffer for the last block in the file.  returns 0.
2005 **
2006 ** tail conversion is not done, so bh_result might not be valid for writing
2007 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2008 ** trying to write the block.
2009 **
2010 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2011 */
2012 static int grab_tail_page(struct inode *p_s_inode,
2013                           struct page **page_result,
2014                           struct buffer_head **bh_result)
2015 {
2016
2017         /* we want the page with the last byte in the file,
2018          ** not the page that will hold the next byte for appending
2019          */
2020         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2021         unsigned long pos = 0;
2022         unsigned long start = 0;
2023         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2024         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2025         struct buffer_head *bh;
2026         struct buffer_head *head;
2027         struct page *page;
2028         int error;
2029
2030         /* we know that we are only called with inode->i_size > 0.
2031          ** we also know that a file tail can never be as big as a block
2032          ** If i_size % blocksize == 0, our file is currently block aligned
2033          ** and it won't need converting or zeroing after a truncate.
2034          */
2035         if ((offset & (blocksize - 1)) == 0) {
2036                 return -ENOENT;
2037         }
2038         page = grab_cache_page(p_s_inode->i_mapping, index);
2039         error = -ENOMEM;
2040         if (!page) {
2041                 goto out;
2042         }
2043         /* start within the page of the last block in the file */
2044         start = (offset / blocksize) * blocksize;
2045
2046         error = block_prepare_write(page, start, offset,
2047                                     reiserfs_get_block_create_0);
2048         if (error)
2049                 goto unlock;
2050
2051         head = page_buffers(page);
2052         bh = head;
2053         do {
2054                 if (pos >= start) {
2055                         break;
2056                 }
2057                 bh = bh->b_this_page;
2058                 pos += blocksize;
2059         } while (bh != head);
2060
2061         if (!buffer_uptodate(bh)) {
2062                 /* note, this should never happen, prepare_write should
2063                  ** be taking care of this for us.  If the buffer isn't up to date,
2064                  ** I've screwed up the code to find the buffer, or the code to
2065                  ** call prepare_write
2066                  */
2067                 reiserfs_warning(p_s_inode->i_sb,
2068                                  "clm-6000: error reading block %lu on dev %s",
2069                                  bh->b_blocknr,
2070                                  reiserfs_bdevname(p_s_inode->i_sb));
2071                 error = -EIO;
2072                 goto unlock;
2073         }
2074         *bh_result = bh;
2075         *page_result = page;
2076
2077       out:
2078         return error;
2079
2080       unlock:
2081         unlock_page(page);
2082         page_cache_release(page);
2083         return error;
2084 }
2085
2086 /*
2087 ** vfs version of truncate file.  Must NOT be called with
2088 ** a transaction already started.
2089 **
2090 ** some code taken from block_truncate_page
2091 */
2092 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2093 {
2094         struct reiserfs_transaction_handle th;
2095         /* we want the offset for the first byte after the end of the file */
2096         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2097         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2098         unsigned length;
2099         struct page *page = NULL;
2100         int error;
2101         struct buffer_head *bh = NULL;
2102         int err2;
2103
2104         reiserfs_write_lock(p_s_inode->i_sb);
2105
2106         if (p_s_inode->i_size > 0) {
2107                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2108                         // -ENOENT means we truncated past the end of the file,
2109                         // and get_block_create_0 could not find a block to read in,
2110                         // which is ok.
2111                         if (error != -ENOENT)
2112                                 reiserfs_warning(p_s_inode->i_sb,
2113                                                  "clm-6001: grab_tail_page failed %d",
2114                                                  error);
2115                         page = NULL;
2116                         bh = NULL;
2117                 }
2118         }
2119
2120         /* so, if page != NULL, we have a buffer head for the offset at
2121          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2122          ** then we have an unformatted node.  Otherwise, we have a direct item,
2123          ** and no zeroing is required on disk.  We zero after the truncate,
2124          ** because the truncate might pack the item anyway
2125          ** (it will unmap bh if it packs).
2126          */
2127         /* it is enough to reserve space in transaction for 2 balancings:
2128            one for "save" link adding and another for the first
2129            cut_from_item. 1 is for update_sd */
2130         error = journal_begin(&th, p_s_inode->i_sb,
2131                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2132         if (error)
2133                 goto out;
2134         reiserfs_update_inode_transaction(p_s_inode);
2135         if (update_timestamps)
2136                 /* we are doing real truncate: if the system crashes before the last
2137                    transaction of truncating gets committed - on reboot the file
2138                    either appears truncated properly or not truncated at all */
2139                 add_save_link(&th, p_s_inode, 1);
2140         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2141         error =
2142             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2143         if (error)
2144                 goto out;
2145
2146         /* check reiserfs_do_truncate after ending the transaction */
2147         if (err2) {
2148                 error = err2;
2149                 goto out;
2150         }
2151
2152         if (update_timestamps) {
2153                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2154                 if (error)
2155                         goto out;
2156         }
2157
2158         if (page) {
2159                 length = offset & (blocksize - 1);
2160                 /* if we are not on a block boundary */
2161                 if (length) {
2162                         char *kaddr;
2163
2164                         length = blocksize - length;
2165                         kaddr = kmap_atomic(page, KM_USER0);
2166                         memset(kaddr + offset, 0, length);
2167                         flush_dcache_page(page);
2168                         kunmap_atomic(kaddr, KM_USER0);
2169                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2170                                 mark_buffer_dirty(bh);
2171                         }
2172                 }
2173                 unlock_page(page);
2174                 page_cache_release(page);
2175         }
2176
2177         reiserfs_write_unlock(p_s_inode->i_sb);
2178         return 0;
2179       out:
2180         if (page) {
2181                 unlock_page(page);
2182                 page_cache_release(page);
2183         }
2184         reiserfs_write_unlock(p_s_inode->i_sb);
2185         return error;
2186 }
2187
2188 static int map_block_for_writepage(struct inode *inode,
2189                                    struct buffer_head *bh_result,
2190                                    unsigned long block)
2191 {
2192         struct reiserfs_transaction_handle th;
2193         int fs_gen;
2194         struct item_head tmp_ih;
2195         struct item_head *ih;
2196         struct buffer_head *bh;
2197         __le32 *item;
2198         struct cpu_key key;
2199         INITIALIZE_PATH(path);
2200         int pos_in_item;
2201         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2202         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2203         int retval;
2204         int use_get_block = 0;
2205         int bytes_copied = 0;
2206         int copy_size;
2207         int trans_running = 0;
2208
2209         /* catch places below that try to log something without starting a trans */
2210         th.t_trans_id = 0;
2211
2212         if (!buffer_uptodate(bh_result)) {
2213                 return -EIO;
2214         }
2215
2216         kmap(bh_result->b_page);
2217       start_over:
2218         reiserfs_write_lock(inode->i_sb);
2219         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2220
2221       research:
2222         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2223         if (retval != POSITION_FOUND) {
2224                 use_get_block = 1;
2225                 goto out;
2226         }
2227
2228         bh = get_last_bh(&path);
2229         ih = get_ih(&path);
2230         item = get_item(&path);
2231         pos_in_item = path.pos_in_item;
2232
2233         /* we've found an unformatted node */
2234         if (indirect_item_found(retval, ih)) {
2235                 if (bytes_copied > 0) {
2236                         reiserfs_warning(inode->i_sb,
2237                                          "clm-6002: bytes_copied %d",
2238                                          bytes_copied);
2239                 }
2240                 if (!get_block_num(item, pos_in_item)) {
2241                         /* crap, we are writing to a hole */
2242                         use_get_block = 1;
2243                         goto out;
2244                 }
2245                 set_block_dev_mapped(bh_result,
2246                                      get_block_num(item, pos_in_item), inode);
2247         } else if (is_direct_le_ih(ih)) {
2248                 char *p;
2249                 p = page_address(bh_result->b_page);
2250                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2251                 copy_size = ih_item_len(ih) - pos_in_item;
2252
2253                 fs_gen = get_generation(inode->i_sb);
2254                 copy_item_head(&tmp_ih, ih);
2255
2256                 if (!trans_running) {
2257                         /* vs-3050 is gone, no need to drop the path */
2258                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2259                         if (retval)
2260                                 goto out;
2261                         reiserfs_update_inode_transaction(inode);
2262                         trans_running = 1;
2263                         if (fs_changed(fs_gen, inode->i_sb)
2264                             && item_moved(&tmp_ih, &path)) {
2265                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2266                                                                  bh);
2267                                 goto research;
2268                         }
2269                 }
2270
2271                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2272
2273                 if (fs_changed(fs_gen, inode->i_sb)
2274                     && item_moved(&tmp_ih, &path)) {
2275                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2276                         goto research;
2277                 }
2278
2279                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2280                        copy_size);
2281
2282                 journal_mark_dirty(&th, inode->i_sb, bh);
2283                 bytes_copied += copy_size;
2284                 set_block_dev_mapped(bh_result, 0, inode);
2285
2286                 /* are there still bytes left? */
2287                 if (bytes_copied < bh_result->b_size &&
2288                     (byte_offset + bytes_copied) < inode->i_size) {
2289                         set_cpu_key_k_offset(&key,
2290                                              cpu_key_k_offset(&key) +
2291                                              copy_size);
2292                         goto research;
2293                 }
2294         } else {
2295                 reiserfs_warning(inode->i_sb,
2296                                  "clm-6003: bad item inode %lu, device %s",
2297                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2298                 retval = -EIO;
2299                 goto out;
2300         }
2301         retval = 0;
2302
2303       out:
2304         pathrelse(&path);
2305         if (trans_running) {
2306                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2307                 if (err)
2308                         retval = err;
2309                 trans_running = 0;
2310         }
2311         reiserfs_write_unlock(inode->i_sb);
2312
2313         /* this is where we fill in holes in the file. */
2314         if (use_get_block) {
2315                 retval = reiserfs_get_block(inode, block, bh_result,
2316                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2317                                             | GET_BLOCK_NO_DANGLE);
2318                 if (!retval) {
2319                         if (!buffer_mapped(bh_result)
2320                             || bh_result->b_blocknr == 0) {
2321                                 /* get_block failed to find a mapped unformatted node. */
2322                                 use_get_block = 0;
2323                                 goto start_over;
2324                         }
2325                 }
2326         }
2327         kunmap(bh_result->b_page);
2328
2329         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2330                 /* we've copied data from the page into the direct item, so the
2331                  * buffer in the page is now clean, mark it to reflect that.
2332                  */
2333                 lock_buffer(bh_result);
2334                 clear_buffer_dirty(bh_result);
2335                 unlock_buffer(bh_result);
2336         }
2337         return retval;
2338 }
2339
2340 /*
2341  * mason@suse.com: updated in 2.5.54 to follow the same general io
2342  * start/recovery path as __block_write_full_page, along with special
2343  * code to handle reiserfs tails.
2344  */
2345 static int reiserfs_write_full_page(struct page *page,
2346                                     struct writeback_control *wbc)
2347 {
2348         struct inode *inode = page->mapping->host;
2349         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2350         int error = 0;
2351         unsigned long block;
2352         sector_t last_block;
2353         struct buffer_head *head, *bh;
2354         int partial = 0;
2355         int nr = 0;
2356         int checked = PageChecked(page);
2357         struct reiserfs_transaction_handle th;
2358         struct super_block *s = inode->i_sb;
2359         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2360         th.t_trans_id = 0;
2361
2362         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2363         if (checked && (current->flags & PF_MEMALLOC)) {
2364                 redirty_page_for_writepage(wbc, page);
2365                 unlock_page(page);
2366                 return 0;
2367         }
2368
2369         /* The page dirty bit is cleared before writepage is called, which
2370          * means we have to tell create_empty_buffers to make dirty buffers
2371          * The page really should be up to date at this point, so tossing
2372          * in the BH_Uptodate is just a sanity check.
2373          */
2374         if (!page_has_buffers(page)) {
2375                 create_empty_buffers(page, s->s_blocksize,
2376                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2377         }
2378         head = page_buffers(page);
2379
2380         /* last page in the file, zero out any contents past the
2381          ** last byte in the file
2382          */
2383         if (page->index >= end_index) {
2384                 char *kaddr;
2385                 unsigned last_offset;
2386
2387                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2388                 /* no file contents in this page */
2389                 if (page->index >= end_index + 1 || !last_offset) {
2390                         unlock_page(page);
2391                         return 0;
2392                 }
2393                 kaddr = kmap_atomic(page, KM_USER0);
2394                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2395                 flush_dcache_page(page);
2396                 kunmap_atomic(kaddr, KM_USER0);
2397         }
2398         bh = head;
2399         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2400         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2401         /* first map all the buffers, logging any direct items we find */
2402         do {
2403                 if (block > last_block) {
2404                         /*
2405                          * This can happen when the block size is less than
2406                          * the page size.  The corresponding bytes in the page
2407                          * were zero filled above
2408                          */
2409                         clear_buffer_dirty(bh);
2410                         set_buffer_uptodate(bh);
2411                 } else if ((checked || buffer_dirty(bh)) &&
2412                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2413                                                        && bh->b_blocknr ==
2414                                                        0))) {
2415                         /* not mapped yet, or it points to a direct item, search
2416                          * the btree for the mapping info, and log any direct
2417                          * items found
2418                          */
2419                         if ((error = map_block_for_writepage(inode, bh, block))) {
2420                                 goto fail;
2421                         }
2422                 }
2423                 bh = bh->b_this_page;
2424                 block++;
2425         } while (bh != head);
2426
2427         /*
2428          * we start the transaction after map_block_for_writepage,
2429          * because it can create holes in the file (an unbounded operation).
2430          * starting it here, we can make a reliable estimate for how many
2431          * blocks we're going to log
2432          */
2433         if (checked) {
2434                 ClearPageChecked(page);
2435                 reiserfs_write_lock(s);
2436                 error = journal_begin(&th, s, bh_per_page + 1);
2437                 if (error) {
2438                         reiserfs_write_unlock(s);
2439                         goto fail;
2440                 }
2441                 reiserfs_update_inode_transaction(inode);
2442         }
2443         /* now go through and lock any dirty buffers on the page */
2444         do {
2445                 get_bh(bh);
2446                 if (!buffer_mapped(bh))
2447                         continue;
2448                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2449                         continue;
2450
2451                 if (checked) {
2452                         reiserfs_prepare_for_journal(s, bh, 1);
2453                         journal_mark_dirty(&th, s, bh);
2454                         continue;
2455                 }
2456                 /* from this point on, we know the buffer is mapped to a
2457                  * real block and not a direct item
2458                  */
2459                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2460                         lock_buffer(bh);
2461                 } else {
2462                         if (test_set_buffer_locked(bh)) {
2463                                 redirty_page_for_writepage(wbc, page);
2464                                 continue;
2465                         }
2466                 }
2467                 if (test_clear_buffer_dirty(bh)) {
2468                         mark_buffer_async_write(bh);
2469                 } else {
2470                         unlock_buffer(bh);
2471                 }
2472         } while ((bh = bh->b_this_page) != head);
2473
2474         if (checked) {
2475                 error = journal_end(&th, s, bh_per_page + 1);
2476                 reiserfs_write_unlock(s);
2477                 if (error)
2478                         goto fail;
2479         }
2480         BUG_ON(PageWriteback(page));
2481         set_page_writeback(page);
2482         unlock_page(page);
2483
2484         /*
2485          * since any buffer might be the only dirty buffer on the page,
2486          * the first submit_bh can bring the page out of writeback.
2487          * be careful with the buffers.
2488          */
2489         do {
2490                 struct buffer_head *next = bh->b_this_page;
2491                 if (buffer_async_write(bh)) {
2492                         submit_bh(WRITE, bh);
2493                         nr++;
2494                 }
2495                 put_bh(bh);
2496                 bh = next;
2497         } while (bh != head);
2498
2499         error = 0;
2500       done:
2501         if (nr == 0) {
2502                 /*
2503                  * if this page only had a direct item, it is very possible for
2504                  * no io to be required without there being an error.  Or,
2505                  * someone else could have locked them and sent them down the
2506                  * pipe without locking the page
2507                  */
2508                 bh = head;
2509                 do {
2510                         if (!buffer_uptodate(bh)) {
2511                                 partial = 1;
2512                                 break;
2513                         }
2514                         bh = bh->b_this_page;
2515                 } while (bh != head);
2516                 if (!partial)
2517                         SetPageUptodate(page);
2518                 end_page_writeback(page);
2519         }
2520         return error;
2521
2522       fail:
2523         /* catches various errors, we need to make sure any valid dirty blocks
2524          * get to the media.  The page is currently locked and not marked for
2525          * writeback
2526          */
2527         ClearPageUptodate(page);
2528         bh = head;
2529         do {
2530                 get_bh(bh);
2531                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2532                         lock_buffer(bh);
2533                         mark_buffer_async_write(bh);
2534                 } else {
2535                         /*
2536                          * clear any dirty bits that might have come from getting
2537                          * attached to a dirty page
2538                          */
2539                         clear_buffer_dirty(bh);
2540                 }
2541                 bh = bh->b_this_page;
2542         } while (bh != head);
2543         SetPageError(page);
2544         BUG_ON(PageWriteback(page));
2545         set_page_writeback(page);
2546         unlock_page(page);
2547         do {
2548                 struct buffer_head *next = bh->b_this_page;
2549                 if (buffer_async_write(bh)) {
2550                         clear_buffer_dirty(bh);
2551                         submit_bh(WRITE, bh);
2552                         nr++;
2553                 }
2554                 put_bh(bh);
2555                 bh = next;
2556         } while (bh != head);
2557         goto done;
2558 }
2559
2560 static int reiserfs_readpage(struct file *f, struct page *page)
2561 {
2562         return block_read_full_page(page, reiserfs_get_block);
2563 }
2564
2565 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2566 {
2567         struct inode *inode = page->mapping->host;
2568         reiserfs_wait_on_write_block(inode->i_sb);
2569         return reiserfs_write_full_page(page, wbc);
2570 }
2571
2572 static int reiserfs_prepare_write(struct file *f, struct page *page,
2573                                   unsigned from, unsigned to)
2574 {
2575         struct inode *inode = page->mapping->host;
2576         int ret;
2577         int old_ref = 0;
2578
2579         reiserfs_wait_on_write_block(inode->i_sb);
2580         fix_tail_page_for_writing(page);
2581         if (reiserfs_transaction_running(inode->i_sb)) {
2582                 struct reiserfs_transaction_handle *th;
2583                 th = (struct reiserfs_transaction_handle *)current->
2584                     journal_info;
2585                 BUG_ON(!th->t_refcount);
2586                 BUG_ON(!th->t_trans_id);
2587                 old_ref = th->t_refcount;
2588                 th->t_refcount++;
2589         }
2590
2591         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2592         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2593                 struct reiserfs_transaction_handle *th = current->journal_info;
2594                 /* this gets a little ugly.  If reiserfs_get_block returned an
2595                  * error and left a transacstion running, we've got to close it,
2596                  * and we've got to free handle if it was a persistent transaction.
2597                  *
2598                  * But, if we had nested into an existing transaction, we need
2599                  * to just drop the ref count on the handle.
2600                  *
2601                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2602                  * and it was a persistent trans.  Otherwise, it was nested above.
2603                  */
2604                 if (th->t_refcount > old_ref) {
2605                         if (old_ref)
2606                                 th->t_refcount--;
2607                         else {
2608                                 int err;
2609                                 reiserfs_write_lock(inode->i_sb);
2610                                 err = reiserfs_end_persistent_transaction(th);
2611                                 reiserfs_write_unlock(inode->i_sb);
2612                                 if (err)
2613                                         ret = err;
2614                         }
2615                 }
2616         }
2617         return ret;
2618
2619 }
2620
2621 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2622 {
2623         return generic_block_bmap(as, block, reiserfs_bmap);
2624 }
2625
2626 static int reiserfs_commit_write(struct file *f, struct page *page,
2627                                  unsigned from, unsigned to)
2628 {
2629         struct inode *inode = page->mapping->host;
2630         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2631         int ret = 0;
2632         int update_sd = 0;
2633         struct reiserfs_transaction_handle *th = NULL;
2634
2635         reiserfs_wait_on_write_block(inode->i_sb);
2636         if (reiserfs_transaction_running(inode->i_sb)) {
2637                 th = current->journal_info;
2638         }
2639         reiserfs_commit_page(inode, page, from, to);
2640
2641         /* generic_commit_write does this for us, but does not update the
2642          ** transaction tracking stuff when the size changes.  So, we have
2643          ** to do the i_size updates here.
2644          */
2645         if (pos > inode->i_size) {
2646                 struct reiserfs_transaction_handle myth;
2647                 reiserfs_write_lock(inode->i_sb);
2648                 /* If the file have grown beyond the border where it
2649                    can have a tail, unmark it as needing a tail
2650                    packing */
2651                 if ((have_large_tails(inode->i_sb)
2652                      && inode->i_size > i_block_size(inode) * 4)
2653                     || (have_small_tails(inode->i_sb)
2654                         && inode->i_size > i_block_size(inode)))
2655                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2656
2657                 ret = journal_begin(&myth, inode->i_sb, 1);
2658                 if (ret) {
2659                         reiserfs_write_unlock(inode->i_sb);
2660                         goto journal_error;
2661                 }
2662                 reiserfs_update_inode_transaction(inode);
2663                 inode->i_size = pos;
2664                 /*
2665                  * this will just nest into our transaction.  It's important
2666                  * to use mark_inode_dirty so the inode gets pushed around on the
2667                  * dirty lists, and so that O_SYNC works as expected
2668                  */
2669                 mark_inode_dirty(inode);
2670                 reiserfs_update_sd(&myth, inode);
2671                 update_sd = 1;
2672                 ret = journal_end(&myth, inode->i_sb, 1);
2673                 reiserfs_write_unlock(inode->i_sb);
2674                 if (ret)
2675                         goto journal_error;
2676         }
2677         if (th) {
2678                 reiserfs_write_lock(inode->i_sb);
2679                 if (!update_sd)
2680                         mark_inode_dirty(inode);
2681                 ret = reiserfs_end_persistent_transaction(th);
2682                 reiserfs_write_unlock(inode->i_sb);
2683                 if (ret)
2684                         goto out;
2685         }
2686
2687       out:
2688         return ret;
2689
2690       journal_error:
2691         if (th) {
2692                 reiserfs_write_lock(inode->i_sb);
2693                 if (!update_sd)
2694                         reiserfs_update_sd(th, inode);
2695                 ret = reiserfs_end_persistent_transaction(th);
2696                 reiserfs_write_unlock(inode->i_sb);
2697         }
2698
2699         return ret;
2700 }
2701
2702 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2703 {
2704         if (reiserfs_attrs(inode->i_sb)) {
2705                 if (sd_attrs & REISERFS_SYNC_FL)
2706                         inode->i_flags |= S_SYNC;
2707                 else
2708                         inode->i_flags &= ~S_SYNC;
2709                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2710                         inode->i_flags |= S_IMMUTABLE;
2711                 else
2712                         inode->i_flags &= ~S_IMMUTABLE;
2713                 if (sd_attrs & REISERFS_APPEND_FL)
2714                         inode->i_flags |= S_APPEND;
2715                 else
2716                         inode->i_flags &= ~S_APPEND;
2717                 if (sd_attrs & REISERFS_NOATIME_FL)
2718                         inode->i_flags |= S_NOATIME;
2719                 else
2720                         inode->i_flags &= ~S_NOATIME;
2721                 if (sd_attrs & REISERFS_NOTAIL_FL)
2722                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2723                 else
2724                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2725         }
2726 }
2727
2728 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2729 {
2730         if (reiserfs_attrs(inode->i_sb)) {
2731                 if (inode->i_flags & S_IMMUTABLE)
2732                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2733                 else
2734                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2735                 if (inode->i_flags & S_SYNC)
2736                         *sd_attrs |= REISERFS_SYNC_FL;
2737                 else
2738                         *sd_attrs &= ~REISERFS_SYNC_FL;
2739                 if (inode->i_flags & S_NOATIME)
2740                         *sd_attrs |= REISERFS_NOATIME_FL;
2741                 else
2742                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2743                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2744                         *sd_attrs |= REISERFS_NOTAIL_FL;
2745                 else
2746                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2747         }
2748 }
2749
2750 /* decide if this buffer needs to stay around for data logging or ordered
2751 ** write purposes
2752 */
2753 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2754 {
2755         int ret = 1;
2756         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2757
2758         lock_buffer(bh);
2759         spin_lock(&j->j_dirty_buffers_lock);
2760         if (!buffer_mapped(bh)) {
2761                 goto free_jh;
2762         }
2763         /* the page is locked, and the only places that log a data buffer
2764          * also lock the page.
2765          */
2766         if (reiserfs_file_data_log(inode)) {
2767                 /*
2768                  * very conservative, leave the buffer pinned if
2769                  * anyone might need it.
2770                  */
2771                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2772                         ret = 0;
2773                 }
2774         } else  if (buffer_dirty(bh)) {
2775                 struct reiserfs_journal_list *jl;
2776                 struct reiserfs_jh *jh = bh->b_private;
2777
2778                 /* why is this safe?
2779                  * reiserfs_setattr updates i_size in the on disk
2780                  * stat data before allowing vmtruncate to be called.
2781                  *
2782                  * If buffer was put onto the ordered list for this
2783                  * transaction, we know for sure either this transaction
2784                  * or an older one already has updated i_size on disk,
2785                  * and this ordered data won't be referenced in the file
2786                  * if we crash.
2787                  *
2788                  * if the buffer was put onto the ordered list for an older
2789                  * transaction, we need to leave it around
2790                  */
2791                 if (jh && (jl = jh->jl)
2792                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2793                         ret = 0;
2794         }
2795       free_jh:
2796         if (ret && bh->b_private) {
2797                 reiserfs_free_jh(bh);
2798         }
2799         spin_unlock(&j->j_dirty_buffers_lock);
2800         unlock_buffer(bh);
2801         return ret;
2802 }
2803
2804 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2805 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2806 {
2807         struct buffer_head *head, *bh, *next;
2808         struct inode *inode = page->mapping->host;
2809         unsigned int curr_off = 0;
2810         int ret = 1;
2811
2812         BUG_ON(!PageLocked(page));
2813
2814         if (offset == 0)
2815                 ClearPageChecked(page);
2816
2817         if (!page_has_buffers(page))
2818                 goto out;
2819
2820         head = page_buffers(page);
2821         bh = head;
2822         do {
2823                 unsigned int next_off = curr_off + bh->b_size;
2824                 next = bh->b_this_page;
2825
2826                 /*
2827                  * is this block fully invalidated?
2828                  */
2829                 if (offset <= curr_off) {
2830                         if (invalidatepage_can_drop(inode, bh))
2831                                 reiserfs_unmap_buffer(bh);
2832                         else
2833                                 ret = 0;
2834                 }
2835                 curr_off = next_off;
2836                 bh = next;
2837         } while (bh != head);
2838
2839         /*
2840          * We release buffers only if the entire page is being invalidated.
2841          * The get_block cached value has been unconditionally invalidated,
2842          * so real IO is not possible anymore.
2843          */
2844         if (!offset && ret) {
2845                 ret = try_to_release_page(page, 0);
2846                 /* maybe should BUG_ON(!ret); - neilb */
2847         }
2848       out:
2849         return;
2850 }
2851
2852 static int reiserfs_set_page_dirty(struct page *page)
2853 {
2854         struct inode *inode = page->mapping->host;
2855         if (reiserfs_file_data_log(inode)) {
2856                 SetPageChecked(page);
2857                 return __set_page_dirty_nobuffers(page);
2858         }
2859         return __set_page_dirty_buffers(page);
2860 }
2861
2862 /*
2863  * Returns 1 if the page's buffers were dropped.  The page is locked.
2864  *
2865  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2866  * in the buffers at page_buffers(page).
2867  *
2868  * even in -o notail mode, we can't be sure an old mount without -o notail
2869  * didn't create files with tails.
2870  */
2871 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2872 {
2873         struct inode *inode = page->mapping->host;
2874         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2875         struct buffer_head *head;
2876         struct buffer_head *bh;
2877         int ret = 1;
2878
2879         WARN_ON(PageChecked(page));
2880         spin_lock(&j->j_dirty_buffers_lock);
2881         head = page_buffers(page);
2882         bh = head;
2883         do {
2884                 if (bh->b_private) {
2885                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2886                                 reiserfs_free_jh(bh);
2887                         } else {
2888                                 ret = 0;
2889                                 break;
2890                         }
2891                 }
2892                 bh = bh->b_this_page;
2893         } while (bh != head);
2894         if (ret)
2895                 ret = try_to_free_buffers(page);
2896         spin_unlock(&j->j_dirty_buffers_lock);
2897         return ret;
2898 }
2899
2900 /* We thank Mingming Cao for helping us understand in great detail what
2901    to do in this section of the code. */
2902 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2903                                   const struct iovec *iov, loff_t offset,
2904                                   unsigned long nr_segs)
2905 {
2906         struct file *file = iocb->ki_filp;
2907         struct inode *inode = file->f_mapping->host;
2908
2909         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2910                                   offset, nr_segs,
2911                                   reiserfs_get_blocks_direct_io, NULL);
2912 }
2913
2914 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2915 {
2916         struct inode *inode = dentry->d_inode;
2917         int error;
2918         unsigned int ia_valid = attr->ia_valid;
2919         reiserfs_write_lock(inode->i_sb);
2920         if (attr->ia_valid & ATTR_SIZE) {
2921                 /* version 2 items will be caught by the s_maxbytes check
2922                  ** done for us in vmtruncate
2923                  */
2924                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2925                     attr->ia_size > MAX_NON_LFS) {
2926                         error = -EFBIG;
2927                         goto out;
2928                 }
2929                 /* fill in hole pointers in the expanding truncate case. */
2930                 if (attr->ia_size > inode->i_size) {
2931                         error = generic_cont_expand(inode, attr->ia_size);
2932                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2933                                 int err;
2934                                 struct reiserfs_transaction_handle th;
2935                                 /* we're changing at most 2 bitmaps, inode + super */
2936                                 err = journal_begin(&th, inode->i_sb, 4);
2937                                 if (!err) {
2938                                         reiserfs_discard_prealloc(&th, inode);
2939                                         err = journal_end(&th, inode->i_sb, 4);
2940                                 }
2941                                 if (err)
2942                                         error = err;
2943                         }
2944                         if (error)
2945                                 goto out;
2946                 }
2947         }
2948
2949         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2950              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2951             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2952                 /* stat data of format v3.5 has 16 bit uid and gid */
2953                 error = -EINVAL;
2954                 goto out;
2955         }
2956
2957         error = inode_change_ok(inode, attr);
2958         if (!error) {
2959                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2960                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2961                         error = reiserfs_chown_xattrs(inode, attr);
2962
2963                         if (!error) {
2964                                 struct reiserfs_transaction_handle th;
2965                                 int jbegin_count =
2966                                     2 *
2967                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2968                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2969                                     2;
2970
2971                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2972                                 error =
2973                                     journal_begin(&th, inode->i_sb,
2974                                                   jbegin_count);
2975                                 if (error)
2976                                         goto out;
2977                                 error =
2978                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2979                                 if (error) {
2980                                         journal_end(&th, inode->i_sb,
2981                                                     jbegin_count);
2982                                         goto out;
2983                                 }
2984                                 /* Update corresponding info in inode so that everything is in
2985                                  * one transaction */
2986                                 if (attr->ia_valid & ATTR_UID)
2987                                         inode->i_uid = attr->ia_uid;
2988                                 if (attr->ia_valid & ATTR_GID)
2989                                         inode->i_gid = attr->ia_gid;
2990                                 mark_inode_dirty(inode);
2991                                 error =
2992                                     journal_end(&th, inode->i_sb, jbegin_count);
2993                         }
2994                 }
2995                 if (!error)
2996                         error = inode_setattr(inode, attr);
2997         }
2998
2999         if (!error && reiserfs_posixacl(inode->i_sb)) {
3000                 if (attr->ia_valid & ATTR_MODE)
3001                         error = reiserfs_acl_chmod(inode);
3002         }
3003
3004       out:
3005         reiserfs_write_unlock(inode->i_sb);
3006         return error;
3007 }
3008
3009 struct address_space_operations reiserfs_address_space_operations = {
3010         .writepage = reiserfs_writepage,
3011         .readpage = reiserfs_readpage,
3012         .readpages = reiserfs_readpages,
3013         .releasepage = reiserfs_releasepage,
3014         .invalidatepage = reiserfs_invalidatepage,
3015         .sync_page = block_sync_page,
3016         .prepare_write = reiserfs_prepare_write,
3017         .commit_write = reiserfs_commit_write,
3018         .bmap = reiserfs_aop_bmap,
3019         .direct_IO = reiserfs_direct_IO,
3020         .set_page_dirty = reiserfs_set_page_dirty,
3021 };