fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include <linux/reiserfs_fs.h>
   8 #include <linux/reiserfs_acl.h>
   9 #include <linux/reiserfs_xattr.h>
  10 #include <linux/exportfs.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/highmem.h>
  14 #include <linux/slab.h>
  15 #include <asm/uaccess.h>
  16 #include <asm/unaligned.h>
  17 #include <linux/buffer_head.h>
  18 #include <linux/mpage.h>
  19 #include <linux/writeback.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/swap.h>
  22
  23 int reiserfs_commit_write(struct file *f, struct page *page,
  24                           unsigned from, unsigned to);
  25
  26 void reiserfs_evict_inode(struct inode *inode)
  27 {
  28         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  29         int jbegin_count =
  30             JOURNAL_PER_BALANCE_CNT * 2 +
  31             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  32         struct reiserfs_transaction_handle th;
  33         int depth;
  34         int err;
  35
  36         if (!inode->i_nlink && !is_bad_inode(inode))
  37                 dquot_initialize(inode);
  38
  39         truncate_inode_pages(&inode->i_data, 0);
  40         if (inode->i_nlink)
  41                 goto no_delete;
  42
  43         depth = reiserfs_write_lock_once(inode->i_sb);
  44
  45         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  46         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  47                 reiserfs_delete_xattrs(inode);
  48
  49                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  50                         goto out;
  51                 reiserfs_update_inode_transaction(inode);
  52
  53                 reiserfs_discard_prealloc(&th, inode);
  54
  55                 err = reiserfs_delete_object(&th, inode);
  56
  57                 /* Do quota update inside a transaction for journaled quotas. We must do that
  58                  * after delete_object so that quota updates go into the same transaction as
  59                  * stat data deletion */
  60                 if (!err)
  61                         dquot_free_inode(inode);
  62
  63                 if (journal_end(&th, inode->i_sb, jbegin_count))
  64                         goto out;
  65
  66                 /* check return value from reiserfs_delete_object after
  67                  * ending the transaction
  68                  */
  69                 if (err)
  70                     goto out;
  71
  72                 /* all items of file are deleted, so we can remove "save" link */
  73                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  74                                                                  * about an error here */
  75         } else {
  76                 /* no object items are in the tree */
  77                 ;
  78         }
  79       out:
  80         end_writeback(inode);   /* note this must go after the journal_end to prevent deadlock */
  81         dquot_drop(inode);
  82         inode->i_blocks = 0;
  83         reiserfs_write_unlock_once(inode->i_sb, depth);
  84         return;
  85
  86 no_delete:
  87         end_writeback(inode);
  88         dquot_drop(inode);
  89 }
  90
  91 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  92                           __u32 objectid, loff_t offset, int type, int length)
  93 {
  94         key->version = version;
  95
  96         key->on_disk_key.k_dir_id = dirid;
  97         key->on_disk_key.k_objectid = objectid;
  98         set_cpu_key_k_offset(key, offset);
  99         set_cpu_key_k_type(key, type);
 100         key->key_length = length;
 101 }
 102
 103 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
 104    offset and type of key */
 105 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 106                   int type, int length)
 107 {
 108         _make_cpu_key(key, get_inode_item_key_version(inode),
 109                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 110                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 111                       length);
 112 }
 113
 114 //
 115 // when key is 0, do not set version and short key
 116 //
 117 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 118                               int version,
 119                               loff_t offset, int type, int length,
 120                               int entry_count /*or ih_free_space */ )
 121 {
 122         if (key) {
 123                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 124                 ih->ih_key.k_objectid =
 125                     cpu_to_le32(key->on_disk_key.k_objectid);
 126         }
 127         put_ih_version(ih, version);
 128         set_le_ih_k_offset(ih, offset);
 129         set_le_ih_k_type(ih, type);
 130         put_ih_item_len(ih, length);
 131         /*    set_ih_free_space (ih, 0); */
 132         // for directory items it is entry count, for directs and stat
 133         // datas - 0xffff, for indirects - 0
 134         put_ih_entry_count(ih, entry_count);
 135 }
 136
 137 //
 138 // FIXME: we might cache recently accessed indirect item
 139
 140 // Ugh.  Not too eager for that....
 141 //  I cut the code until such time as I see a convincing argument (benchmark).
 142 // I don't want a bloated inode struct..., and I don't like code complexity....
 143
 144 /* cutting the code is fine, since it really isn't in use yet and is easy
 145 ** to add back in.  But, Vladimir has a really good idea here.  Think
 146 ** about what happens for reading a file.  For each page,
 147 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 148 ** an indirect item.  This indirect item has X number of pointers, where
 149 ** X is a big number if we've done the block allocation right.  But,
 150 ** we only use one or two of these pointers during each call to readpage,
 151 ** needlessly researching again later on.
 152 **
 153 ** The size of the cache could be dynamic based on the size of the file.
 154 **
 155 ** I'd also like to see us cache the location the stat data item, since
 156 ** we are needlessly researching for that frequently.
 157 **
 158 ** --chris
 159 */
 160
 161 /* If this page has a file tail in it, and
 162 ** it was read in by get_block_create_0, the page data is valid,
 163 ** but tail is still sitting in a direct item, and we can't write to
 164 ** it.  So, look through this page, and check all the mapped buffers
 165 ** to make sure they have valid block numbers.  Any that don't need
 166 ** to be unmapped, so that __block_write_begin will correctly call
 167 ** reiserfs_get_block to convert the tail into an unformatted node
 168 */
 169 static inline void fix_tail_page_for_writing(struct page *page)
 170 {
 171         struct buffer_head *head, *next, *bh;
 172
 173         if (page && page_has_buffers(page)) {
 174                 head = page_buffers(page);
 175                 bh = head;
 176                 do {
 177                         next = bh->b_this_page;
 178                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 179                                 reiserfs_unmap_buffer(bh);
 180                         }
 181                         bh = next;
 182                 } while (bh != head);
 183         }
 184 }
 185
 186 /* reiserfs_get_block does not need to allocate a block only if it has been
 187    done already or non-hole position has been found in the indirect item */
 188 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 189                                     struct item_head *ih,
 190                                     __le32 * item, int pos_in_item)
 191 {
 192         if (allocated)
 193                 return 0;
 194         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 195             get_block_num(item, pos_in_item))
 196                 return 0;
 197         return 1;
 198 }
 199
 200 static inline int indirect_item_found(int retval, struct item_head *ih)
 201 {
 202         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 203 }
 204
 205 static inline void set_block_dev_mapped(struct buffer_head *bh,
 206                                         b_blocknr_t block, struct inode *inode)
 207 {
 208         map_bh(bh, inode->i_sb, block);
 209 }
 210
 211 //
 212 // files which were created in the earlier version can not be longer,
 213 // than 2 gb
 214 //
 215 static int file_capable(struct inode *inode, sector_t block)
 216 {
 217         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 218             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 219                 return 1;
 220
 221         return 0;
 222 }
 223
 224 static int restart_transaction(struct reiserfs_transaction_handle *th,
 225                                struct inode *inode, struct treepath *path)
 226 {
 227         struct super_block *s = th->t_super;
 228         int len = th->t_blocks_allocated;
 229         int err;
 230
 231         BUG_ON(!th->t_trans_id);
 232         BUG_ON(!th->t_refcount);
 233
 234         pathrelse(path);
 235
 236         /* we cannot restart while nested */
 237         if (th->t_refcount > 1) {
 238                 return 0;
 239         }
 240         reiserfs_update_sd(th, inode);
 241         err = journal_end(th, s, len);
 242         if (!err) {
 243                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 244                 if (!err)
 245                         reiserfs_update_inode_transaction(inode);
 246         }
 247         return err;
 248 }
 249
 250 // it is called by get_block when create == 0. Returns block number
 251 // for 'block'-th logical block of file. When it hits direct item it
 252 // returns 0 (being called from bmap) or read direct item into piece
 253 // of page (bh_result)
 254
 255 // Please improve the english/clarity in the comment above, as it is
 256 // hard to understand.
 257
 258 static int _get_block_create_0(struct inode *inode, sector_t block,
 259                                struct buffer_head *bh_result, int args)
 260 {
 261         INITIALIZE_PATH(path);
 262         struct cpu_key key;
 263         struct buffer_head *bh;
 264         struct item_head *ih, tmp_ih;
 265         b_blocknr_t blocknr;
 266         char *p = NULL;
 267         int chars;
 268         int ret;
 269         int result;
 270         int done = 0;
 271         unsigned long offset;
 272
 273         // prepare the key to look for the 'block'-th block of file
 274         make_cpu_key(&key, inode,
 275                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 276                      3);
 277
 278         result = search_for_position_by_key(inode->i_sb, &key, &path);
 279         if (result != POSITION_FOUND) {
 280                 pathrelse(&path);
 281                 if (p)
 282                         kunmap(bh_result->b_page);
 283                 if (result == IO_ERROR)
 284                         return -EIO;
 285                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 286                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 287                 if ((args & GET_BLOCK_NO_HOLE)
 288                     && !PageUptodate(bh_result->b_page)) {
 289                         return -ENOENT;
 290                 }
 291                 return 0;
 292         }
 293         //
 294         bh = get_last_bh(&path);
 295         ih = get_ih(&path);
 296         if (is_indirect_le_ih(ih)) {
 297                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 298
 299                 /* FIXME: here we could cache indirect item or part of it in
 300                    the inode to avoid search_by_key in case of subsequent
 301                    access to file */
 302                 blocknr = get_block_num(ind_item, path.pos_in_item);
 303                 ret = 0;
 304                 if (blocknr) {
 305                         map_bh(bh_result, inode->i_sb, blocknr);
 306                         if (path.pos_in_item ==
 307                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 308                                 set_buffer_boundary(bh_result);
 309                         }
 310                 } else
 311                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 312                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 313                 if ((args & GET_BLOCK_NO_HOLE)
 314                             && !PageUptodate(bh_result->b_page)) {
 315                         ret = -ENOENT;
 316                 }
 317
 318                 pathrelse(&path);
 319                 if (p)
 320                         kunmap(bh_result->b_page);
 321                 return ret;
 322         }
 323         // requested data are in direct item(s)
 324         if (!(args & GET_BLOCK_READ_DIRECT)) {
 325                 // we are called by bmap. FIXME: we can not map block of file
 326                 // when it is stored in direct item(s)
 327                 pathrelse(&path);
 328                 if (p)
 329                         kunmap(bh_result->b_page);
 330                 return -ENOENT;
 331         }
 332
 333         /* if we've got a direct item, and the buffer or page was uptodate,
 334          ** we don't want to pull data off disk again.  skip to the
 335          ** end, where we map the buffer and return
 336          */
 337         if (buffer_uptodate(bh_result)) {
 338                 goto finished;
 339         } else
 340                 /*
 341                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 342                  ** pages without any buffers.  If the page is up to date, we don't want
 343                  ** read old data off disk.  Set the up to date bit on the buffer instead
 344                  ** and jump to the end
 345                  */
 346         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 347                 set_buffer_uptodate(bh_result);
 348                 goto finished;
 349         }
 350         // read file tail into part of page
 351         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 352         copy_item_head(&tmp_ih, ih);
 353
 354         /* we only want to kmap if we are reading the tail into the page.
 355          ** this is not the common case, so we don't kmap until we are
 356          ** sure we need to.  But, this means the item might move if
 357          ** kmap schedules
 358          */
 359         if (!p)
 360                 p = (char *)kmap(bh_result->b_page);
 361
 362         p += offset;
 363         memset(p, 0, inode->i_sb->s_blocksize);
 364         do {
 365                 if (!is_direct_le_ih(ih)) {
 366                         BUG();
 367                 }
 368                 /* make sure we don't read more bytes than actually exist in
 369                  ** the file.  This can happen in odd cases where i_size isn't
 370                  ** correct, and when direct item padding results in a few
 371                  ** extra bytes at the end of the direct item
 372                  */
 373                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 374                         break;
 375                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 376                         chars =
 377                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 378                             path.pos_in_item;
 379                         done = 1;
 380                 } else {
 381                         chars = ih_item_len(ih) - path.pos_in_item;
 382                 }
 383                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 384
 385                 if (done)
 386                         break;
 387
 388                 p += chars;
 389
 390                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 391                         // we done, if read direct item is not the last item of
 392                         // node FIXME: we could try to check right delimiting key
 393                         // to see whether direct item continues in the right
 394                         // neighbor or rely on i_size
 395                         break;
 396
 397                 // update key to look for the next piece
 398                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 399                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 400                 if (result != POSITION_FOUND)
 401                         // i/o error most likely
 402                         break;
 403                 bh = get_last_bh(&path);
 404                 ih = get_ih(&path);
 405         } while (1);
 406
 407         flush_dcache_page(bh_result->b_page);
 408         kunmap(bh_result->b_page);
 409
 410       finished:
 411         pathrelse(&path);
 412
 413         if (result == IO_ERROR)
 414                 return -EIO;
 415
 416         /* this buffer has valid data, but isn't valid for io.  mapping it to
 417          * block #0 tells the rest of reiserfs it just has a tail in it
 418          */
 419         map_bh(bh_result, inode->i_sb, 0);
 420         set_buffer_uptodate(bh_result);
 421         return 0;
 422 }
 423
 424 // this is called to create file map. So, _get_block_create_0 will not
 425 // read direct item
 426 static int reiserfs_bmap(struct inode *inode, sector_t block,
 427                          struct buffer_head *bh_result, int create)
 428 {
 429         if (!file_capable(inode, block))
 430                 return -EFBIG;
 431
 432         reiserfs_write_lock(inode->i_sb);
 433         /* do not read the direct item */
 434         _get_block_create_0(inode, block, bh_result, 0);
 435         reiserfs_write_unlock(inode->i_sb);
 436         return 0;
 437 }
 438
 439 /* special version of get_block that is only used by grab_tail_page right
 440 ** now.  It is sent to __block_write_begin, and when you try to get a
 441 ** block past the end of the file (or a block from a hole) it returns
 442 ** -ENOENT instead of a valid buffer.  __block_write_begin expects to
 443 ** be able to do i/o on the buffers returned, unless an error value
 444 ** is also returned.
 445 **
 446 ** So, this allows __block_write_begin to be used for reading a single block
 447 ** in a page.  Where it does not produce a valid page for holes, or past the
 448 ** end of the file.  This turns out to be exactly what we need for reading
 449 ** tails for conversion.
 450 **
 451 ** The point of the wrapper is forcing a certain value for create, even
 452 ** though the VFS layer is calling this function with create==1.  If you
 453 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 454 ** don't use this function.
 455 */
 456 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 457                                        struct buffer_head *bh_result,
 458                                        int create)
 459 {
 460         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 461 }
 462
 463 /* This is special helper for reiserfs_get_block in case we are executing
 464    direct_IO request. */
 465 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 466                                          sector_t iblock,
 467                                          struct buffer_head *bh_result,
 468                                          int create)
 469 {
 470         int ret;
 471
 472         bh_result->b_page = NULL;
 473
 474         /* We set the b_size before reiserfs_get_block call since it is
 475            referenced in convert_tail_for_hole() that may be called from
 476            reiserfs_get_block() */
 477         bh_result->b_size = (1 << inode->i_blkbits);
 478
 479         ret = reiserfs_get_block(inode, iblock, bh_result,
 480                                  create | GET_BLOCK_NO_DANGLE);
 481         if (ret)
 482                 goto out;
 483
 484         /* don't allow direct io onto tail pages */
 485         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 486                 /* make sure future calls to the direct io funcs for this offset
 487                  ** in the file fail by unmapping the buffer
 488                  */
 489                 clear_buffer_mapped(bh_result);
 490                 ret = -EINVAL;
 491         }
 492         /* Possible unpacked tail. Flush the data before pages have
 493            disappeared */
 494         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 495                 int err;
 496
 497                 reiserfs_write_lock(inode->i_sb);
 498
 499                 err = reiserfs_commit_for_inode(inode);
 500                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 501
 502                 reiserfs_write_unlock(inode->i_sb);
 503
 504                 if (err < 0)
 505                         ret = err;
 506         }
 507       out:
 508         return ret;
 509 }
 510
 511 /*
 512 ** helper function for when reiserfs_get_block is called for a hole
 513 ** but the file tail is still in a direct item
 514 ** bh_result is the buffer head for the hole
 515 ** tail_offset is the offset of the start of the tail in the file
 516 **
 517 ** This calls prepare_write, which will start a new transaction
 518 ** you should not be in a transaction, or have any paths held when you
 519 ** call this.
 520 */
 521 static int convert_tail_for_hole(struct inode *inode,
 522                                  struct buffer_head *bh_result,
 523                                  loff_t tail_offset)
 524 {
 525         unsigned long index;
 526         unsigned long tail_end;
 527         unsigned long tail_start;
 528         struct page *tail_page;
 529         struct page *hole_page = bh_result->b_page;
 530         int retval = 0;
 531
 532         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 533                 return -EIO;
 534
 535         /* always try to read until the end of the block */
 536         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 537         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 538
 539         index = tail_offset >> PAGE_CACHE_SHIFT;
 540         /* hole_page can be zero in case of direct_io, we are sure
 541            that we cannot get here if we write with O_DIRECT into
 542            tail page */
 543         if (!hole_page || index != hole_page->index) {
 544                 tail_page = grab_cache_page(inode->i_mapping, index);
 545                 retval = -ENOMEM;
 546                 if (!tail_page) {
 547                         goto out;
 548                 }
 549         } else {
 550                 tail_page = hole_page;
 551         }
 552
 553         /* we don't have to make sure the conversion did not happen while
 554          ** we were locking the page because anyone that could convert
 555          ** must first take i_mutex.
 556          **
 557          ** We must fix the tail page for writing because it might have buffers
 558          ** that are mapped, but have a block number of 0.  This indicates tail
 559          ** data that has been read directly into the page, and
 560          ** __block_write_begin won't trigger a get_block in this case.
 561          */
 562         fix_tail_page_for_writing(tail_page);
 563         retval = __reiserfs_write_begin(tail_page, tail_start,
 564                                       tail_end - tail_start);
 565         if (retval)
 566                 goto unlock;
 567
 568         /* tail conversion might change the data in the page */
 569         flush_dcache_page(tail_page);
 570
 571         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 572
 573       unlock:
 574         if (tail_page != hole_page) {
 575                 unlock_page(tail_page);
 576                 page_cache_release(tail_page);
 577         }
 578       out:
 579         return retval;
 580 }
 581
 582 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 583                                   sector_t block,
 584                                   struct inode *inode,
 585                                   b_blocknr_t * allocated_block_nr,
 586                                   struct treepath *path, int flags)
 587 {
 588         BUG_ON(!th->t_trans_id);
 589
 590 #ifdef REISERFS_PREALLOCATE
 591         if (!(flags & GET_BLOCK_NO_IMUX)) {
 592                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 593                                                   path, block);
 594         }
 595 #endif
 596         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 597                                          block);
 598 }
 599
 600 int reiserfs_get_block(struct inode *inode, sector_t block,
 601                        struct buffer_head *bh_result, int create)
 602 {
 603         int repeat, retval = 0;
 604         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 605         INITIALIZE_PATH(path);
 606         int pos_in_item;
 607         struct cpu_key key;
 608         struct buffer_head *bh, *unbh = NULL;
 609         struct item_head *ih, tmp_ih;
 610         __le32 *item;
 611         int done;
 612         int fs_gen;
 613         int lock_depth;
 614         struct reiserfs_transaction_handle *th = NULL;
 615         /* space reserved in transaction batch:
 616            . 3 balancings in direct->indirect conversion
 617            . 1 block involved into reiserfs_update_sd()
 618            XXX in practically impossible worst case direct2indirect()
 619            can incur (much) more than 3 balancings.
 620            quota update for user, group */
 621         int jbegin_count =
 622             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 623             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 624         int version;
 625         int dangle = 1;
 626         loff_t new_offset =
 627             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 628
 629         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 630         version = get_inode_item_key_version(inode);
 631
 632         if (!file_capable(inode, block)) {
 633                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 634                 return -EFBIG;
 635         }
 636
 637         /* if !create, we aren't changing the FS, so we don't need to
 638          ** log anything, so we don't need to start a transaction
 639          */
 640         if (!(create & GET_BLOCK_CREATE)) {
 641                 int ret;
 642                 /* find number of block-th logical block of the file */
 643                 ret = _get_block_create_0(inode, block, bh_result,
 644                                           create | GET_BLOCK_READ_DIRECT);
 645                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 646                 return ret;
 647         }
 648         /*
 649          * if we're already in a transaction, make sure to close
 650          * any new transactions we start in this func
 651          */
 652         if ((create & GET_BLOCK_NO_DANGLE) ||
 653             reiserfs_transaction_running(inode->i_sb))
 654                 dangle = 0;
 655
 656         /* If file is of such a size, that it might have a tail and tails are enabled
 657          ** we should mark it as possibly needing tail packing on close
 658          */
 659         if ((have_large_tails(inode->i_sb)
 660              && inode->i_size < i_block_size(inode) * 4)
 661             || (have_small_tails(inode->i_sb)
 662                 && inode->i_size < i_block_size(inode)))
 663                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 664
 665         /* set the key of the first byte in the 'block'-th block of file */
 666         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 667         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 668               start_trans:
 669                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 670                 if (!th) {
 671                         retval = -ENOMEM;
 672                         goto failure;
 673                 }
 674                 reiserfs_update_inode_transaction(inode);
 675         }
 676       research:
 677
 678         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 679         if (retval == IO_ERROR) {
 680                 retval = -EIO;
 681                 goto failure;
 682         }
 683
 684         bh = get_last_bh(&path);
 685         ih = get_ih(&path);
 686         item = get_item(&path);
 687         pos_in_item = path.pos_in_item;
 688
 689         fs_gen = get_generation(inode->i_sb);
 690         copy_item_head(&tmp_ih, ih);
 691
 692         if (allocation_needed
 693             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 694                 /* we have to allocate block for the unformatted node */
 695                 if (!th) {
 696                         pathrelse(&path);
 697                         goto start_trans;
 698                 }
 699
 700                 repeat =
 701                     _allocate_block(th, block, inode, &allocated_block_nr,
 702                                     &path, create);
 703
 704                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 705                         /* restart the transaction to give the journal a chance to free
 706                          ** some blocks.  releases the path, so we have to go back to
 707                          ** research if we succeed on the second try
 708                          */
 709                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 710                         retval = restart_transaction(th, inode, &path);
 711                         if (retval)
 712                                 goto failure;
 713                         repeat =
 714                             _allocate_block(th, block, inode,
 715                                             &allocated_block_nr, NULL, create);
 716
 717                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 718                                 goto research;
 719                         }
 720                         if (repeat == QUOTA_EXCEEDED)
 721                                 retval = -EDQUOT;
 722                         else
 723                                 retval = -ENOSPC;
 724                         goto failure;
 725                 }
 726
 727                 if (fs_changed(fs_gen, inode->i_sb)
 728                     && item_moved(&tmp_ih, &path)) {
 729                         goto research;
 730                 }
 731         }
 732
 733         if (indirect_item_found(retval, ih)) {
 734                 b_blocknr_t unfm_ptr;
 735                 /* 'block'-th block is in the file already (there is
 736                    corresponding cell in some indirect item). But it may be
 737                    zero unformatted node pointer (hole) */
 738                 unfm_ptr = get_block_num(item, pos_in_item);
 739                 if (unfm_ptr == 0) {
 740                         /* use allocated block to plug the hole */
 741                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 742                         if (fs_changed(fs_gen, inode->i_sb)
 743                             && item_moved(&tmp_ih, &path)) {
 744                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 745                                                                  bh);
 746                                 goto research;
 747                         }
 748                         set_buffer_new(bh_result);
 749                         if (buffer_dirty(bh_result)
 750                             && reiserfs_data_ordered(inode->i_sb))
 751                                 reiserfs_add_ordered_list(inode, bh_result);
 752                         put_block_num(item, pos_in_item, allocated_block_nr);
 753                         unfm_ptr = allocated_block_nr;
 754                         journal_mark_dirty(th, inode->i_sb, bh);
 755                         reiserfs_update_sd(th, inode);
 756                 }
 757                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 758                 pathrelse(&path);
 759                 retval = 0;
 760                 if (!dangle && th)
 761                         retval = reiserfs_end_persistent_transaction(th);
 762
 763                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 764
 765                 /* the item was found, so new blocks were not added to the file
 766                  ** there is no need to make sure the inode is updated with this
 767                  ** transaction
 768                  */
 769                 return retval;
 770         }
 771
 772         if (!th) {
 773                 pathrelse(&path);
 774                 goto start_trans;
 775         }
 776
 777         /* desired position is not found or is in the direct item. We have
 778            to append file with holes up to 'block'-th block converting
 779            direct items to indirect one if necessary */
 780         done = 0;
 781         do {
 782                 if (is_statdata_le_ih(ih)) {
 783                         __le32 unp = 0;
 784                         struct cpu_key tmp_key;
 785
 786                         /* indirect item has to be inserted */
 787                         make_le_item_head(&tmp_ih, &key, version, 1,
 788                                           TYPE_INDIRECT, UNFM_P_SIZE,
 789                                           0 /* free_space */ );
 790
 791                         if (cpu_key_k_offset(&key) == 1) {
 792                                 /* we are going to add 'block'-th block to the file. Use
 793                                    allocated block for that */
 794                                 unp = cpu_to_le32(allocated_block_nr);
 795                                 set_block_dev_mapped(bh_result,
 796                                                      allocated_block_nr, inode);
 797                                 set_buffer_new(bh_result);
 798                                 done = 1;
 799                         }
 800                         tmp_key = key;  // ;)
 801                         set_cpu_key_k_offset(&tmp_key, 1);
 802                         PATH_LAST_POSITION(&path)++;
 803
 804                         retval =
 805                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 806                                                  inode, (char *)&unp);
 807                         if (retval) {
 808                                 reiserfs_free_block(th, inode,
 809                                                     allocated_block_nr, 1);
 810                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 811                         }
 812                         //mark_tail_converted (inode);
 813                 } else if (is_direct_le_ih(ih)) {
 814                         /* direct item has to be converted */
 815                         loff_t tail_offset;
 816
 817                         tail_offset =
 818                             ((le_ih_k_offset(ih) -
 819                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 820                         if (tail_offset == cpu_key_k_offset(&key)) {
 821                                 /* direct item we just found fits into block we have
 822                                    to map. Convert it into unformatted node: use
 823                                    bh_result for the conversion */
 824                                 set_block_dev_mapped(bh_result,
 825                                                      allocated_block_nr, inode);
 826                                 unbh = bh_result;
 827                                 done = 1;
 828                         } else {
 829                                 /* we have to padd file tail stored in direct item(s)
 830                                    up to block size and convert it to unformatted
 831                                    node. FIXME: this should also get into page cache */
 832
 833                                 pathrelse(&path);
 834                                 /*
 835                                  * ugly, but we can only end the transaction if
 836                                  * we aren't nested
 837                                  */
 838                                 BUG_ON(!th->t_refcount);
 839                                 if (th->t_refcount == 1) {
 840                                         retval =
 841                                             reiserfs_end_persistent_transaction
 842                                             (th);
 843                                         th = NULL;
 844                                         if (retval)
 845                                                 goto failure;
 846                                 }
 847
 848                                 retval =
 849                                     convert_tail_for_hole(inode, bh_result,
 850                                                           tail_offset);
 851                                 if (retval) {
 852                                         if (retval != -ENOSPC)
 853                                                 reiserfs_error(inode->i_sb,
 854                                                         "clm-6004",
 855                                                         "convert tail failed "
 856                                                         "inode %lu, error %d",
 857                                                         inode->i_ino,
 858                                                         retval);
 859                                         if (allocated_block_nr) {
 860                                                 /* the bitmap, the super, and the stat data == 3 */
 861                                                 if (!th)
 862                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 863                                                 if (th)
 864                                                         reiserfs_free_block(th,
 865                                                                             inode,
 866                                                                             allocated_block_nr,
 867                                                                             1);
 868                                         }
 869                                         goto failure;
 870                                 }
 871                                 goto research;
 872                         }
 873                         retval =
 874                             direct2indirect(th, inode, &path, unbh,
 875                                             tail_offset);
 876                         if (retval) {
 877                                 reiserfs_unmap_buffer(unbh);
 878                                 reiserfs_free_block(th, inode,
 879                                                     allocated_block_nr, 1);
 880                                 goto failure;
 881                         }
 882                         /* it is important the set_buffer_uptodate is done after
 883                          ** the direct2indirect.  The buffer might contain valid
 884                          ** data newer than the data on disk (read by readpage, changed,
 885                          ** and then sent here by writepage).  direct2indirect needs
 886                          ** to know if unbh was already up to date, so it can decide
 887                          ** if the data in unbh needs to be replaced with data from
 888                          ** the disk
 889                          */
 890                         set_buffer_uptodate(unbh);
 891
 892                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 893                            buffer will disappear shortly, so it should not be added to
 894                          */
 895                         if (unbh->b_page) {
 896                                 /* we've converted the tail, so we must
 897                                  ** flush unbh before the transaction commits
 898                                  */
 899                                 reiserfs_add_tail_list(inode, unbh);
 900
 901                                 /* mark it dirty now to prevent commit_write from adding
 902                                  ** this buffer to the inode's dirty buffer list
 903                                  */
 904                                 /*
 905                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 906                                  * It's still atomic, but it sets the page dirty too,
 907                                  * which makes it eligible for writeback at any time by the
 908                                  * VM (which was also the case with __mark_buffer_dirty())
 909                                  */
 910                                 mark_buffer_dirty(unbh);
 911                         }
 912                 } else {
 913                         /* append indirect item with holes if needed, when appending
 914                            pointer to 'block'-th block use block, which is already
 915                            allocated */
 916                         struct cpu_key tmp_key;
 917                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 918                         // one block which is a fastpath
 919                         unp_t *un;
 920                         __u64 max_to_insert =
 921                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 922                             UNFM_P_SIZE;
 923                         __u64 blocks_needed;
 924
 925                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 926                                "vs-804: invalid position for append");
 927                         /* indirect item has to be appended, set up key of that position */
 928                         make_cpu_key(&tmp_key, inode,
 929                                      le_key_k_offset(version,
 930                                                      &(ih->ih_key)) +
 931                                      op_bytes_number(ih,
 932                                                      inode->i_sb->s_blocksize),
 933                                      //pos_in_item * inode->i_sb->s_blocksize,
 934                                      TYPE_INDIRECT, 3); // key type is unimportant
 935
 936                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 937                                "green-805: invalid offset");
 938                         blocks_needed =
 939                             1 +
 940                             ((cpu_key_k_offset(&key) -
 941                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 942                              s_blocksize_bits);
 943
 944                         if (blocks_needed == 1) {
 945                                 un = &unf_single;
 946                         } else {
 947                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 948                                 if (!un) {
 949                                         un = &unf_single;
 950                                         blocks_needed = 1;
 951                                         max_to_insert = 0;
 952                                 }
 953                         }
 954                         if (blocks_needed <= max_to_insert) {
 955                                 /* we are going to add target block to the file. Use allocated
 956                                    block for that */
 957                                 un[blocks_needed - 1] =
 958                                     cpu_to_le32(allocated_block_nr);
 959                                 set_block_dev_mapped(bh_result,
 960                                                      allocated_block_nr, inode);
 961                                 set_buffer_new(bh_result);
 962                                 done = 1;
 963                         } else {
 964                                 /* paste hole to the indirect item */
 965                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 966                                    only have space for one block */
 967                                 blocks_needed =
 968                                     max_to_insert ? max_to_insert : 1;
 969                         }
 970                         retval =
 971                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 972                                                      (char *)un,
 973                                                      UNFM_P_SIZE *
 974                                                      blocks_needed);
 975
 976                         if (blocks_needed != 1)
 977                                 kfree(un);
 978
 979                         if (retval) {
 980                                 reiserfs_free_block(th, inode,
 981                                                     allocated_block_nr, 1);
 982                                 goto failure;
 983                         }
 984                         if (!done) {
 985                                 /* We need to mark new file size in case this function will be
 986                                    interrupted/aborted later on. And we may do this only for
 987                                    holes. */
 988                                 inode->i_size +=
 989                                     inode->i_sb->s_blocksize * blocks_needed;
 990                         }
 991                 }
 992
 993                 if (done == 1)
 994                         break;
 995
 996                 /* this loop could log more blocks than we had originally asked
 997                  ** for.  So, we have to allow the transaction to end if it is
 998                  ** too big or too full.  Update the inode so things are
 999                  ** consistent if we crash before the function returns
1000                  **
1001                  ** release the path so that anybody waiting on the path before
1002                  ** ending their transaction will be able to continue.
1003                  */
1004                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1005                         retval = restart_transaction(th, inode, &path);
1006                         if (retval)
1007                                 goto failure;
1008                 }
1009                 /*
1010                  * inserting indirect pointers for a hole can take a
1011                  * long time.  reschedule if needed and also release the write
1012                  * lock for others.
1013                  */
1014                 if (need_resched()) {
1015                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1016                         schedule();
1017                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1018                 }
1019
1020                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1021                 if (retval == IO_ERROR) {
1022                         retval = -EIO;
1023                         goto failure;
1024                 }
1025                 if (retval == POSITION_FOUND) {
1026                         reiserfs_warning(inode->i_sb, "vs-825",
1027                                          "%K should not be found", &key);
1028                         retval = -EEXIST;
1029                         if (allocated_block_nr)
1030                                 reiserfs_free_block(th, inode,
1031                                                     allocated_block_nr, 1);
1032                         pathrelse(&path);
1033                         goto failure;
1034                 }
1035                 bh = get_last_bh(&path);
1036                 ih = get_ih(&path);
1037                 item = get_item(&path);
1038                 pos_in_item = path.pos_in_item;
1039         } while (1);
1040
1041         retval = 0;
1042
1043       failure:
1044         if (th && (!dangle || (retval && !th->t_trans_id))) {
1045                 int err;
1046                 if (th->t_trans_id)
1047                         reiserfs_update_sd(th, inode);
1048                 err = reiserfs_end_persistent_transaction(th);
1049                 if (err)
1050                         retval = err;
1051         }
1052
1053         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1054         reiserfs_check_path(&path);
1055         return retval;
1056 }
1057
1058 static int
1059 reiserfs_readpages(struct file *file, struct address_space *mapping,
1060                    struct list_head *pages, unsigned nr_pages)
1061 {
1062         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1063 }
1064
1065 /* Compute real number of used bytes by file
1066  * Following three functions can go away when we'll have enough space in stat item
1067  */
1068 static int real_space_diff(struct inode *inode, int sd_size)
1069 {
1070         int bytes;
1071         loff_t blocksize = inode->i_sb->s_blocksize;
1072
1073         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1074                 return sd_size;
1075
1076         /* End of file is also in full block with indirect reference, so round
1077          ** up to the next block.
1078          **
1079          ** there is just no way to know if the tail is actually packed
1080          ** on the file, so we have to assume it isn't.  When we pack the
1081          ** tail, we add 4 bytes to pretend there really is an unformatted
1082          ** node pointer
1083          */
1084         bytes =
1085             ((inode->i_size +
1086               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1087             sd_size;
1088         return bytes;
1089 }
1090
1091 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1092                                         int sd_size)
1093 {
1094         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1095                 return inode->i_size +
1096                     (loff_t) (real_space_diff(inode, sd_size));
1097         }
1098         return ((loff_t) real_space_diff(inode, sd_size)) +
1099             (((loff_t) blocks) << 9);
1100 }
1101
1102 /* Compute number of blocks used by file in ReiserFS counting */
1103 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1104 {
1105         loff_t bytes = inode_get_bytes(inode);
1106         loff_t real_space = real_space_diff(inode, sd_size);
1107
1108         /* keeps fsck and non-quota versions of reiserfs happy */
1109         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1110                 bytes += (loff_t) 511;
1111         }
1112
1113         /* files from before the quota patch might i_blocks such that
1114          ** bytes < real_space.  Deal with that here to prevent it from
1115          ** going negative.
1116          */
1117         if (bytes < real_space)
1118                 return 0;
1119         return (bytes - real_space) >> 9;
1120 }
1121
1122 //
1123 // BAD: new directories have stat data of new type and all other items
1124 // of old type. Version stored in the inode says about body items, so
1125 // in update_stat_data we can not rely on inode, but have to check
1126 // item version directly
1127 //
1128
1129 // called by read_locked_inode
1130 static void init_inode(struct inode *inode, struct treepath *path)
1131 {
1132         struct buffer_head *bh;
1133         struct item_head *ih;
1134         __u32 rdev;
1135         //int version = ITEM_VERSION_1;
1136
1137         bh = PATH_PLAST_BUFFER(path);
1138         ih = PATH_PITEM_HEAD(path);
1139
1140         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1141
1142         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1143         REISERFS_I(inode)->i_flags = 0;
1144         REISERFS_I(inode)->i_prealloc_block = 0;
1145         REISERFS_I(inode)->i_prealloc_count = 0;
1146         REISERFS_I(inode)->i_trans_id = 0;
1147         REISERFS_I(inode)->i_jl = NULL;
1148         reiserfs_init_xattr_rwsem(inode);
1149
1150         if (stat_data_v1(ih)) {
1151                 struct stat_data_v1 *sd =
1152                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1153                 unsigned long blocks;
1154
1155                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1156                 set_inode_sd_version(inode, STAT_DATA_V1);
1157                 inode->i_mode = sd_v1_mode(sd);
1158                 inode->i_nlink = sd_v1_nlink(sd);
1159                 inode->i_uid = sd_v1_uid(sd);
1160                 inode->i_gid = sd_v1_gid(sd);
1161                 inode->i_size = sd_v1_size(sd);
1162                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1163                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1164                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1165                 inode->i_atime.tv_nsec = 0;
1166                 inode->i_ctime.tv_nsec = 0;
1167                 inode->i_mtime.tv_nsec = 0;
1168
1169                 inode->i_blocks = sd_v1_blocks(sd);
1170                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1171                 blocks = (inode->i_size + 511) >> 9;
1172                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1173                 if (inode->i_blocks > blocks) {
1174                         // there was a bug in <=3.5.23 when i_blocks could take negative
1175                         // values. Starting from 3.5.17 this value could even be stored in
1176                         // stat data. For such files we set i_blocks based on file
1177                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1178                         // only updated if file's inode will ever change
1179                         inode->i_blocks = blocks;
1180                 }
1181
1182                 rdev = sd_v1_rdev(sd);
1183                 REISERFS_I(inode)->i_first_direct_byte =
1184                     sd_v1_first_direct_byte(sd);
1185                 /* an early bug in the quota code can give us an odd number for the
1186                  ** block count.  This is incorrect, fix it here.
1187                  */
1188                 if (inode->i_blocks & 1) {
1189                         inode->i_blocks++;
1190                 }
1191                 inode_set_bytes(inode,
1192                                 to_real_used_space(inode, inode->i_blocks,
1193                                                    SD_V1_SIZE));
1194                 /* nopack is initially zero for v1 objects. For v2 objects,
1195                    nopack is initialised from sd_attrs */
1196                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1197         } else {
1198                 // new stat data found, but object may have old items
1199                 // (directories and symlinks)
1200                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1201
1202                 inode->i_mode = sd_v2_mode(sd);
1203                 inode->i_nlink = sd_v2_nlink(sd);
1204                 inode->i_uid = sd_v2_uid(sd);
1205                 inode->i_size = sd_v2_size(sd);
1206                 inode->i_gid = sd_v2_gid(sd);
1207                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1208                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1209                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1210                 inode->i_ctime.tv_nsec = 0;
1211                 inode->i_mtime.tv_nsec = 0;
1212                 inode->i_atime.tv_nsec = 0;
1213                 inode->i_blocks = sd_v2_blocks(sd);
1214                 rdev = sd_v2_rdev(sd);
1215                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1216                         inode->i_generation =
1217                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1218                 else
1219                         inode->i_generation = sd_v2_generation(sd);
1220
1221                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1222                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1223                 else
1224                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1225                 REISERFS_I(inode)->i_first_direct_byte = 0;
1226                 set_inode_sd_version(inode, STAT_DATA_V2);
1227                 inode_set_bytes(inode,
1228                                 to_real_used_space(inode, inode->i_blocks,
1229                                                    SD_V2_SIZE));
1230                 /* read persistent inode attributes from sd and initialise
1231                    generic inode flags from them */
1232                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1233                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1234         }
1235
1236         pathrelse(path);
1237         if (S_ISREG(inode->i_mode)) {
1238                 inode->i_op = &reiserfs_file_inode_operations;
1239                 inode->i_fop = &reiserfs_file_operations;
1240                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1241         } else if (S_ISDIR(inode->i_mode)) {
1242                 inode->i_op = &reiserfs_dir_inode_operations;
1243                 inode->i_fop = &reiserfs_dir_operations;
1244         } else if (S_ISLNK(inode->i_mode)) {
1245                 inode->i_op = &reiserfs_symlink_inode_operations;
1246                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1247         } else {
1248                 inode->i_blocks = 0;
1249                 inode->i_op = &reiserfs_special_inode_operations;
1250                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1251         }
1252 }
1253
1254 // update new stat data with inode fields
1255 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1256 {
1257         struct stat_data *sd_v2 = (struct stat_data *)sd;
1258         __u16 flags;
1259
1260         set_sd_v2_mode(sd_v2, inode->i_mode);
1261         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1262         set_sd_v2_uid(sd_v2, inode->i_uid);
1263         set_sd_v2_size(sd_v2, size);
1264         set_sd_v2_gid(sd_v2, inode->i_gid);
1265         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1266         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1267         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1268         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1269         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1270                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1271         else
1272                 set_sd_v2_generation(sd_v2, inode->i_generation);
1273         flags = REISERFS_I(inode)->i_attrs;
1274         i_attrs_to_sd_attrs(inode, &flags);
1275         set_sd_v2_attrs(sd_v2, flags);
1276 }
1277
1278 // used to copy inode's fields to old stat data
1279 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1280 {
1281         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1282
1283         set_sd_v1_mode(sd_v1, inode->i_mode);
1284         set_sd_v1_uid(sd_v1, inode->i_uid);
1285         set_sd_v1_gid(sd_v1, inode->i_gid);
1286         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1287         set_sd_v1_size(sd_v1, size);
1288         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1289         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1290         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1291
1292         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1293                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1294         else
1295                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1296
1297         // Sigh. i_first_direct_byte is back
1298         set_sd_v1_first_direct_byte(sd_v1,
1299                                     REISERFS_I(inode)->i_first_direct_byte);
1300 }
1301
1302 /* NOTE, you must prepare the buffer head before sending it here,
1303 ** and then log it after the call
1304 */
1305 static void update_stat_data(struct treepath *path, struct inode *inode,
1306                              loff_t size)
1307 {
1308         struct buffer_head *bh;
1309         struct item_head *ih;
1310
1311         bh = PATH_PLAST_BUFFER(path);
1312         ih = PATH_PITEM_HEAD(path);
1313
1314         if (!is_statdata_le_ih(ih))
1315                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1316                                INODE_PKEY(inode), ih);
1317
1318         if (stat_data_v1(ih)) {
1319                 // path points to old stat data
1320                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1321         } else {
1322                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1323         }
1324
1325         return;
1326 }
1327
1328 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1329                              struct inode *inode, loff_t size)
1330 {
1331         struct cpu_key key;
1332         INITIALIZE_PATH(path);
1333         struct buffer_head *bh;
1334         int fs_gen;
1335         struct item_head *ih, tmp_ih;
1336         int retval;
1337
1338         BUG_ON(!th->t_trans_id);
1339
1340         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1341
1342         for (;;) {
1343                 int pos;
1344                 /* look for the object's stat data */
1345                 retval = search_item(inode->i_sb, &key, &path);
1346                 if (retval == IO_ERROR) {
1347                         reiserfs_error(inode->i_sb, "vs-13050",
1348                                        "i/o failure occurred trying to "
1349                                        "update %K stat data", &key);
1350                         return;
1351                 }
1352                 if (retval == ITEM_NOT_FOUND) {
1353                         pos = PATH_LAST_POSITION(&path);
1354                         pathrelse(&path);
1355                         if (inode->i_nlink == 0) {
1356                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1357                                 return;
1358                         }
1359                         reiserfs_warning(inode->i_sb, "vs-13060",
1360                                          "stat data of object %k (nlink == %d) "
1361                                          "not found (pos %d)",
1362                                          INODE_PKEY(inode), inode->i_nlink,
1363                                          pos);
1364                         reiserfs_check_path(&path);
1365                         return;
1366                 }
1367
1368                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1369                  ** FS might change.  We have to detect that, and loop back to the
1370                  ** search if the stat data item has moved
1371                  */
1372                 bh = get_last_bh(&path);
1373                 ih = get_ih(&path);
1374                 copy_item_head(&tmp_ih, ih);
1375                 fs_gen = get_generation(inode->i_sb);
1376                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1377                 if (fs_changed(fs_gen, inode->i_sb)
1378                     && item_moved(&tmp_ih, &path)) {
1379                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1380                         continue;       /* Stat_data item has been moved after scheduling. */
1381                 }
1382                 break;
1383         }
1384         update_stat_data(&path, inode, size);
1385         journal_mark_dirty(th, th->t_super, bh);
1386         pathrelse(&path);
1387         return;
1388 }
1389
1390 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1391 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1392 ** and clear the key in the private portion of the inode, otherwise a
1393 ** corresponding iput might try to delete whatever object the inode last
1394 ** represented.
1395 */
1396 static void reiserfs_make_bad_inode(struct inode *inode)
1397 {
1398         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1399         make_bad_inode(inode);
1400 }
1401
1402 //
1403 // initially this function was derived from minix or ext2's analog and
1404 // evolved as the prototype did
1405 //
1406
1407 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1408 {
1409         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1410         inode->i_ino = args->objectid;
1411         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1412         return 0;
1413 }
1414
1415 /* looks for stat data in the tree, and fills up the fields of in-core
1416    inode stat data fields */
1417 void reiserfs_read_locked_inode(struct inode *inode,
1418                                 struct reiserfs_iget_args *args)
1419 {
1420         INITIALIZE_PATH(path_to_sd);
1421         struct cpu_key key;
1422         unsigned long dirino;
1423         int retval;
1424
1425         dirino = args->dirid;
1426
1427         /* set version 1, version 2 could be used too, because stat data
1428            key is the same in both versions */
1429         key.version = KEY_FORMAT_3_5;
1430         key.on_disk_key.k_dir_id = dirino;
1431         key.on_disk_key.k_objectid = inode->i_ino;
1432         key.on_disk_key.k_offset = 0;
1433         key.on_disk_key.k_type = 0;
1434
1435         /* look for the object's stat data */
1436         retval = search_item(inode->i_sb, &key, &path_to_sd);
1437         if (retval == IO_ERROR) {
1438                 reiserfs_error(inode->i_sb, "vs-13070",
1439                                "i/o failure occurred trying to find "
1440                                "stat data of %K", &key);
1441                 reiserfs_make_bad_inode(inode);
1442                 return;
1443         }
1444         if (retval != ITEM_FOUND) {
1445                 /* a stale NFS handle can trigger this without it being an error */
1446                 pathrelse(&path_to_sd);
1447                 reiserfs_make_bad_inode(inode);
1448                 inode->i_nlink = 0;
1449                 return;
1450         }
1451
1452         init_inode(inode, &path_to_sd);
1453
1454         /* It is possible that knfsd is trying to access inode of a file
1455            that is being removed from the disk by some other thread. As we
1456            update sd on unlink all that is required is to check for nlink
1457            here. This bug was first found by Sizif when debugging
1458            SquidNG/Butterfly, forgotten, and found again after Philippe
1459            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1460
1461            More logical fix would require changes in fs/inode.c:iput() to
1462            remove inode from hash-table _after_ fs cleaned disk stuff up and
1463            in iget() to return NULL if I_FREEING inode is found in
1464            hash-table. */
1465         /* Currently there is one place where it's ok to meet inode with
1466            nlink==0: processing of open-unlinked and half-truncated files
1467            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1468         if ((inode->i_nlink == 0) &&
1469             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1470                 reiserfs_warning(inode->i_sb, "vs-13075",
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         reiserfs_write_unlock(s);
1511         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1512                              reiserfs_find_actor, reiserfs_init_locked_inode,
1513                              (void *)(&args));
1514         reiserfs_write_lock(s);
1515         if (!inode)
1516                 return ERR_PTR(-ENOMEM);
1517
1518         if (inode->i_state & I_NEW) {
1519                 reiserfs_read_locked_inode(inode, &args);
1520                 unlock_new_inode(inode);
1521         }
1522
1523         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1524                 /* either due to i/o error or a stale NFS handle */
1525                 iput(inode);
1526                 inode = NULL;
1527         }
1528         return inode;
1529 }
1530
1531 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1532         u32 objectid, u32 dir_id, u32 generation)
1533
1534 {
1535         struct cpu_key key;
1536         struct inode *inode;
1537
1538         key.on_disk_key.k_objectid = objectid;
1539         key.on_disk_key.k_dir_id = dir_id;
1540         reiserfs_write_lock(sb);
1541         inode = reiserfs_iget(sb, &key);
1542         if (inode && !IS_ERR(inode) && generation != 0 &&
1543             generation != inode->i_generation) {
1544                 iput(inode);
1545                 inode = NULL;
1546         }
1547         reiserfs_write_unlock(sb);
1548
1549         return d_obtain_alias(inode);
1550 }
1551
1552 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1553                 int fh_len, int fh_type)
1554 {
1555         /* fhtype happens to reflect the number of u32s encoded.
1556          * due to a bug in earlier code, fhtype might indicate there
1557          * are more u32s then actually fitted.
1558          * so if fhtype seems to be more than len, reduce fhtype.
1559          * Valid types are:
1560          *   2 - objectid + dir_id - legacy support
1561          *   3 - objectid + dir_id + generation
1562          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1563          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1564          *   6 - as above plus generation of directory
1565          * 6 does not fit in NFSv2 handles
1566          */
1567         if (fh_type > fh_len) {
1568                 if (fh_type != 6 || fh_len != 5)
1569                         reiserfs_warning(sb, "reiserfs-13077",
1570                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1571                                 fh_type, fh_len);
1572                 fh_type = 5;
1573         }
1574
1575         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1576                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1577 }
1578
1579 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1580                 int fh_len, int fh_type)
1581 {
1582         if (fh_type < 4)
1583                 return NULL;
1584
1585         return reiserfs_get_dentry(sb,
1586                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1587                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1588                 (fh_type == 6) ? fid->raw[5] : 0);
1589 }
1590
1591 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1592                        int need_parent)
1593 {
1594         struct inode *inode = dentry->d_inode;
1595         int maxlen = *lenp;
1596
1597         if (maxlen < 3)
1598                 return 255;
1599
1600         data[0] = inode->i_ino;
1601         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1602         data[2] = inode->i_generation;
1603         *lenp = 3;
1604         /* no room for directory info? return what we've stored so far */
1605         if (maxlen < 5 || !need_parent)
1606                 return 3;
1607
1608         spin_lock(&dentry->d_lock);
1609         inode = dentry->d_parent->d_inode;
1610         data[3] = inode->i_ino;
1611         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1612         *lenp = 5;
1613         if (maxlen >= 6) {
1614                 data[5] = inode->i_generation;
1615                 *lenp = 6;
1616         }
1617         spin_unlock(&dentry->d_lock);
1618         return *lenp;
1619 }
1620
1621 /* looks for stat data, then copies fields to it, marks the buffer
1622    containing stat data as dirty */
1623 /* reiserfs inodes are never really dirty, since the dirty inode call
1624 ** always logs them.  This call allows the VFS inode marking routines
1625 ** to properly mark inodes for datasync and such, but only actually
1626 ** does something when called for a synchronous update.
1627 */
1628 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1629 {
1630         struct reiserfs_transaction_handle th;
1631         int jbegin_count = 1;
1632
1633         if (inode->i_sb->s_flags & MS_RDONLY)
1634                 return -EROFS;
1635         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1636          ** these cases are just when the system needs ram, not when the
1637          ** inode needs to reach disk for safety, and they can safely be
1638          ** ignored because the altered inode has already been logged.
1639          */
1640         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1641                 reiserfs_write_lock(inode->i_sb);
1642                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1643                         reiserfs_update_sd(&th, inode);
1644                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1645                 }
1646                 reiserfs_write_unlock(inode->i_sb);
1647         }
1648         return 0;
1649 }
1650
1651 /* stat data of new object is inserted already, this inserts the item
1652    containing "." and ".." entries */
1653 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1654                                   struct inode *inode,
1655                                   struct item_head *ih, struct treepath *path,
1656                                   struct inode *dir)
1657 {
1658         struct super_block *sb = th->t_super;
1659         char empty_dir[EMPTY_DIR_SIZE];
1660         char *body = empty_dir;
1661         struct cpu_key key;
1662         int retval;
1663
1664         BUG_ON(!th->t_trans_id);
1665
1666         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1667                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1668                       TYPE_DIRENTRY, 3 /*key length */ );
1669
1670         /* compose item head for new item. Directories consist of items of
1671            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1672            is done by reiserfs_new_inode */
1673         if (old_format_only(sb)) {
1674                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1675                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1676
1677                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1678                                        ih->ih_key.k_objectid,
1679                                        INODE_PKEY(dir)->k_dir_id,
1680                                        INODE_PKEY(dir)->k_objectid);
1681         } else {
1682                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1683                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1684
1685                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1686                                     ih->ih_key.k_objectid,
1687                                     INODE_PKEY(dir)->k_dir_id,
1688                                     INODE_PKEY(dir)->k_objectid);
1689         }
1690
1691         /* look for place in the tree for new item */
1692         retval = search_item(sb, &key, path);
1693         if (retval == IO_ERROR) {
1694                 reiserfs_error(sb, "vs-13080",
1695                                "i/o failure occurred creating new directory");
1696                 return -EIO;
1697         }
1698         if (retval == ITEM_FOUND) {
1699                 pathrelse(path);
1700                 reiserfs_warning(sb, "vs-13070",
1701                                  "object with this key exists (%k)",
1702                                  &(ih->ih_key));
1703                 return -EEXIST;
1704         }
1705
1706         /* insert item, that is empty directory item */
1707         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1708 }
1709
1710 /* stat data of object has been inserted, this inserts the item
1711    containing the body of symlink */
1712 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1713                                 struct item_head *ih,
1714                                 struct treepath *path, const char *symname,
1715                                 int item_len)
1716 {
1717         struct super_block *sb = th->t_super;
1718         struct cpu_key key;
1719         int retval;
1720
1721         BUG_ON(!th->t_trans_id);
1722
1723         _make_cpu_key(&key, KEY_FORMAT_3_5,
1724                       le32_to_cpu(ih->ih_key.k_dir_id),
1725                       le32_to_cpu(ih->ih_key.k_objectid),
1726                       1, TYPE_DIRECT, 3 /*key length */ );
1727
1728         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1729                           0 /*free_space */ );
1730
1731         /* look for place in the tree for new item */
1732         retval = search_item(sb, &key, path);
1733         if (retval == IO_ERROR) {
1734                 reiserfs_error(sb, "vs-13080",
1735                                "i/o failure occurred creating new symlink");
1736                 return -EIO;
1737         }
1738         if (retval == ITEM_FOUND) {
1739                 pathrelse(path);
1740                 reiserfs_warning(sb, "vs-13080",
1741                                  "object with this key exists (%k)",
1742                                  &(ih->ih_key));
1743                 return -EEXIST;
1744         }
1745
1746         /* insert item, that is body of symlink */
1747         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1748 }
1749
1750 /* inserts the stat data into the tree, and then calls
1751    reiserfs_new_directory (to insert ".", ".." item if new object is
1752    directory) or reiserfs_new_symlink (to insert symlink body if new
1753    object is symlink) or nothing (if new object is regular file)
1754
1755    NOTE! uid and gid must already be set in the inode.  If we return
1756    non-zero due to an error, we have to drop the quota previously allocated
1757    for the fresh inode.  This can only be done outside a transaction, so
1758    if we return non-zero, we also end the transaction.  */
1759 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1760                        struct inode *dir, int mode, const char *symname,
1761                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1762                           strlen (symname) for symlinks) */
1763                        loff_t i_size, struct dentry *dentry,
1764                        struct inode *inode,
1765                        struct reiserfs_security_handle *security)
1766 {
1767         struct super_block *sb;
1768         struct reiserfs_iget_args args;
1769         INITIALIZE_PATH(path_to_key);
1770         struct cpu_key key;
1771         struct item_head ih;
1772         struct stat_data sd;
1773         int retval;
1774         int err;
1775
1776         BUG_ON(!th->t_trans_id);
1777
1778         dquot_initialize(inode);
1779         err = dquot_alloc_inode(inode);
1780         if (err)
1781                 goto out_end_trans;
1782         if (!dir->i_nlink) {
1783                 err = -EPERM;
1784                 goto out_bad_inode;
1785         }
1786
1787         sb = dir->i_sb;
1788
1789         /* item head of new item */
1790         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1791         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1792         if (!ih.ih_key.k_objectid) {
1793                 err = -ENOMEM;
1794                 goto out_bad_inode;
1795         }
1796         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1797         if (old_format_only(sb))
1798                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1799                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1800         else
1801                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1802                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1803         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1804         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1805         if (insert_inode_locked4(inode, args.objectid,
1806                              reiserfs_find_actor, &args) < 0) {
1807                 err = -EINVAL;
1808                 goto out_bad_inode;
1809         }
1810         if (old_format_only(sb))
1811                 /* not a perfect generation count, as object ids can be reused, but
1812                  ** this is as good as reiserfs can do right now.
1813                  ** note that the private part of inode isn't filled in yet, we have
1814                  ** to use the directory.
1815                  */
1816                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1817         else
1818 #if defined( USE_INODE_GENERATION_COUNTER )
1819                 inode->i_generation =
1820                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1821 #else
1822                 inode->i_generation = ++event;
1823 #endif
1824
1825         /* fill stat data */
1826         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1827
1828         /* uid and gid must already be set by the caller for quota init */
1829
1830         /* symlink cannot be immutable or append only, right? */
1831         if (S_ISLNK(inode->i_mode))
1832                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1833
1834         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1835         inode->i_size = i_size;
1836         inode->i_blocks = 0;
1837         inode->i_bytes = 0;
1838         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1839             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1840
1841         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1842         REISERFS_I(inode)->i_flags = 0;
1843         REISERFS_I(inode)->i_prealloc_block = 0;
1844         REISERFS_I(inode)->i_prealloc_count = 0;
1845         REISERFS_I(inode)->i_trans_id = 0;
1846         REISERFS_I(inode)->i_jl = NULL;
1847         REISERFS_I(inode)->i_attrs =
1848             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1849         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1850         reiserfs_init_xattr_rwsem(inode);
1851
1852         /* key to search for correct place for new stat data */
1853         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1854                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1855                       TYPE_STAT_DATA, 3 /*key length */ );
1856
1857         /* find proper place for inserting of stat data */
1858         retval = search_item(sb, &key, &path_to_key);
1859         if (retval == IO_ERROR) {
1860                 err = -EIO;
1861                 goto out_bad_inode;
1862         }
1863         if (retval == ITEM_FOUND) {
1864                 pathrelse(&path_to_key);
1865                 err = -EEXIST;
1866                 goto out_bad_inode;
1867         }
1868         if (old_format_only(sb)) {
1869                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1870                         pathrelse(&path_to_key);
1871                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1872                         err = -EINVAL;
1873                         goto out_bad_inode;
1874                 }
1875                 inode2sd_v1(&sd, inode, inode->i_size);
1876         } else {
1877                 inode2sd(&sd, inode, inode->i_size);
1878         }
1879         // store in in-core inode the key of stat data and version all
1880         // object items will have (directory items will have old offset
1881         // format, other new objects will consist of new items)
1882         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1883                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1884         else
1885                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1886         if (old_format_only(sb))
1887                 set_inode_sd_version(inode, STAT_DATA_V1);
1888         else
1889                 set_inode_sd_version(inode, STAT_DATA_V2);
1890
1891         /* insert the stat data into the tree */
1892 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1893         if (REISERFS_I(dir)->new_packing_locality)
1894                 th->displace_new_blocks = 1;
1895 #endif
1896         retval =
1897             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1898                                  (char *)(&sd));
1899         if (retval) {
1900                 err = retval;
1901                 reiserfs_check_path(&path_to_key);
1902                 goto out_bad_inode;
1903         }
1904 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1905         if (!th->displace_new_blocks)
1906                 REISERFS_I(dir)->new_packing_locality = 0;
1907 #endif
1908         if (S_ISDIR(mode)) {
1909                 /* insert item with "." and ".." */
1910                 retval =
1911                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1912         }
1913
1914         if (S_ISLNK(mode)) {
1915                 /* insert body of symlink */
1916                 if (!old_format_only(sb))
1917                         i_size = ROUND_UP(i_size);
1918                 retval =
1919                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1920                                          i_size);
1921         }
1922         if (retval) {
1923                 err = retval;
1924                 reiserfs_check_path(&path_to_key);
1925                 journal_end(th, th->t_super, th->t_blocks_allocated);
1926                 goto out_inserted_sd;
1927         }
1928
1929         if (reiserfs_posixacl(inode->i_sb)) {
1930                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1931                 if (retval) {
1932                         err = retval;
1933                         reiserfs_check_path(&path_to_key);
1934                         journal_end(th, th->t_super, th->t_blocks_allocated);
1935                         goto out_inserted_sd;
1936                 }
1937         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1938                 reiserfs_warning(inode->i_sb, "jdm-13090",
1939                                  "ACLs aren't enabled in the fs, "
1940                                  "but vfs thinks they are!");
1941         } else if (IS_PRIVATE(dir))
1942                 inode->i_flags |= S_PRIVATE;
1943
1944         if (security->name) {
1945                 retval = reiserfs_security_write(th, inode, security);
1946                 if (retval) {
1947                         err = retval;
1948                         reiserfs_check_path(&path_to_key);
1949                         retval = journal_end(th, th->t_super,
1950                                              th->t_blocks_allocated);
1951                         if (retval)
1952                                 err = retval;
1953                         goto out_inserted_sd;
1954                 }
1955         }
1956
1957         reiserfs_update_sd(th, inode);
1958         reiserfs_check_path(&path_to_key);
1959
1960         return 0;
1961
1962 /* it looks like you can easily compress these two goto targets into
1963  * one.  Keeping it like this doesn't actually hurt anything, and they
1964  * are place holders for what the quota code actually needs.
1965  */
1966       out_bad_inode:
1967         /* Invalidate the object, nothing was inserted yet */
1968         INODE_PKEY(inode)->k_objectid = 0;
1969
1970         /* Quota change must be inside a transaction for journaling */
1971         dquot_free_inode(inode);
1972
1973       out_end_trans:
1974         journal_end(th, th->t_super, th->t_blocks_allocated);
1975         /* Drop can be outside and it needs more credits so it's better to have it outside */
1976         dquot_drop(inode);
1977         inode->i_flags |= S_NOQUOTA;
1978         make_bad_inode(inode);
1979
1980       out_inserted_sd:
1981         inode->i_nlink = 0;
1982         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1983         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1984         iput(inode);
1985         return err;
1986 }
1987
1988 /*
1989 ** finds the tail page in the page cache,
1990 ** reads the last block in.
1991 **
1992 ** On success, page_result is set to a locked, pinned page, and bh_result
1993 ** is set to an up to date buffer for the last block in the file.  returns 0.
1994 **
1995 ** tail conversion is not done, so bh_result might not be valid for writing
1996 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1997 ** trying to write the block.
1998 **
1999 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2000 */
2001 static int grab_tail_page(struct inode *inode,
2002                           struct page **page_result,
2003                           struct buffer_head **bh_result)
2004 {
2005
2006         /* we want the page with the last byte in the file,
2007          ** not the page that will hold the next byte for appending
2008          */
2009         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2010         unsigned long pos = 0;
2011         unsigned long start = 0;
2012         unsigned long blocksize = inode->i_sb->s_blocksize;
2013         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2014         struct buffer_head *bh;
2015         struct buffer_head *head;
2016         struct page *page;
2017         int error;
2018
2019         /* we know that we are only called with inode->i_size > 0.
2020          ** we also know that a file tail can never be as big as a block
2021          ** If i_size % blocksize == 0, our file is currently block aligned
2022          ** and it won't need converting or zeroing after a truncate.
2023          */
2024         if ((offset & (blocksize - 1)) == 0) {
2025                 return -ENOENT;
2026         }
2027         page = grab_cache_page(inode->i_mapping, index);
2028         error = -ENOMEM;
2029         if (!page) {
2030                 goto out;
2031         }
2032         /* start within the page of the last block in the file */
2033         start = (offset / blocksize) * blocksize;
2034
2035         error = __block_write_begin(page, start, offset - start,
2036                                     reiserfs_get_block_create_0);
2037         if (error)
2038                 goto unlock;
2039
2040         head = page_buffers(page);
2041         bh = head;
2042         do {
2043                 if (pos >= start) {
2044                         break;
2045                 }
2046                 bh = bh->b_this_page;
2047                 pos += blocksize;
2048         } while (bh != head);
2049
2050         if (!buffer_uptodate(bh)) {
2051                 /* note, this should never happen, prepare_write should
2052                  ** be taking care of this for us.  If the buffer isn't up to date,
2053                  ** I've screwed up the code to find the buffer, or the code to
2054                  ** call prepare_write
2055                  */
2056                 reiserfs_error(inode->i_sb, "clm-6000",
2057                                "error reading block %lu", bh->b_blocknr);
2058                 error = -EIO;
2059                 goto unlock;
2060         }
2061         *bh_result = bh;
2062         *page_result = page;
2063
2064       out:
2065         return error;
2066
2067       unlock:
2068         unlock_page(page);
2069         page_cache_release(page);
2070         return error;
2071 }
2072
2073 /*
2074 ** vfs version of truncate file.  Must NOT be called with
2075 ** a transaction already started.
2076 **
2077 ** some code taken from block_truncate_page
2078 */
2079 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2080 {
2081         struct reiserfs_transaction_handle th;
2082         /* we want the offset for the first byte after the end of the file */
2083         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2084         unsigned blocksize = inode->i_sb->s_blocksize;
2085         unsigned length;
2086         struct page *page = NULL;
2087         int error;
2088         struct buffer_head *bh = NULL;
2089         int err2;
2090         int lock_depth;
2091
2092         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2093
2094         if (inode->i_size > 0) {
2095                 error = grab_tail_page(inode, &page, &bh);
2096                 if (error) {
2097                         // -ENOENT means we truncated past the end of the file,
2098                         // and get_block_create_0 could not find a block to read in,
2099                         // which is ok.
2100                         if (error != -ENOENT)
2101                                 reiserfs_error(inode->i_sb, "clm-6001",
2102                                                "grab_tail_page failed %d",
2103                                                error);
2104                         page = NULL;
2105                         bh = NULL;
2106                 }
2107         }
2108
2109         /* so, if page != NULL, we have a buffer head for the offset at
2110          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2111          ** then we have an unformatted node.  Otherwise, we have a direct item,
2112          ** and no zeroing is required on disk.  We zero after the truncate,
2113          ** because the truncate might pack the item anyway
2114          ** (it will unmap bh if it packs).
2115          */
2116         /* it is enough to reserve space in transaction for 2 balancings:
2117            one for "save" link adding and another for the first
2118            cut_from_item. 1 is for update_sd */
2119         error = journal_begin(&th, inode->i_sb,
2120                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2121         if (error)
2122                 goto out;
2123         reiserfs_update_inode_transaction(inode);
2124         if (update_timestamps)
2125                 /* we are doing real truncate: if the system crashes before the last
2126                    transaction of truncating gets committed - on reboot the file
2127                    either appears truncated properly or not truncated at all */
2128                 add_save_link(&th, inode, 1);
2129         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2130         error =
2131             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2132         if (error)
2133                 goto out;
2134
2135         /* check reiserfs_do_truncate after ending the transaction */
2136         if (err2) {
2137                 error = err2;
2138                 goto out;
2139         }
2140
2141         if (update_timestamps) {
2142                 error = remove_save_link(inode, 1 /* truncate */);
2143                 if (error)
2144                         goto out;
2145         }
2146
2147         if (page) {
2148                 length = offset & (blocksize - 1);
2149                 /* if we are not on a block boundary */
2150                 if (length) {
2151                         length = blocksize - length;
2152                         zero_user(page, offset, length);
2153                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2154                                 mark_buffer_dirty(bh);
2155                         }
2156                 }
2157                 unlock_page(page);
2158                 page_cache_release(page);
2159         }
2160
2161         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2162
2163         return 0;
2164       out:
2165         if (page) {
2166                 unlock_page(page);
2167                 page_cache_release(page);
2168         }
2169
2170         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2171
2172         return error;
2173 }
2174
2175 static int map_block_for_writepage(struct inode *inode,
2176                                    struct buffer_head *bh_result,
2177                                    unsigned long block)
2178 {
2179         struct reiserfs_transaction_handle th;
2180         int fs_gen;
2181         struct item_head tmp_ih;
2182         struct item_head *ih;
2183         struct buffer_head *bh;
2184         __le32 *item;
2185         struct cpu_key key;
2186         INITIALIZE_PATH(path);
2187         int pos_in_item;
2188         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2189         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2190         int retval;
2191         int use_get_block = 0;
2192         int bytes_copied = 0;
2193         int copy_size;
2194         int trans_running = 0;
2195
2196         /* catch places below that try to log something without starting a trans */
2197         th.t_trans_id = 0;
2198
2199         if (!buffer_uptodate(bh_result)) {
2200                 return -EIO;
2201         }
2202
2203         kmap(bh_result->b_page);
2204       start_over:
2205         reiserfs_write_lock(inode->i_sb);
2206         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2207
2208       research:
2209         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2210         if (retval != POSITION_FOUND) {
2211                 use_get_block = 1;
2212                 goto out;
2213         }
2214
2215         bh = get_last_bh(&path);
2216         ih = get_ih(&path);
2217         item = get_item(&path);
2218         pos_in_item = path.pos_in_item;
2219
2220         /* we've found an unformatted node */
2221         if (indirect_item_found(retval, ih)) {
2222                 if (bytes_copied > 0) {
2223                         reiserfs_warning(inode->i_sb, "clm-6002",
2224                                          "bytes_copied %d", bytes_copied);
2225                 }
2226                 if (!get_block_num(item, pos_in_item)) {
2227                         /* crap, we are writing to a hole */
2228                         use_get_block = 1;
2229                         goto out;
2230                 }
2231                 set_block_dev_mapped(bh_result,
2232                                      get_block_num(item, pos_in_item), inode);
2233         } else if (is_direct_le_ih(ih)) {
2234                 char *p;
2235                 p = page_address(bh_result->b_page);
2236                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2237                 copy_size = ih_item_len(ih) - pos_in_item;
2238
2239                 fs_gen = get_generation(inode->i_sb);
2240                 copy_item_head(&tmp_ih, ih);
2241
2242                 if (!trans_running) {
2243                         /* vs-3050 is gone, no need to drop the path */
2244                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2245                         if (retval)
2246                                 goto out;
2247                         reiserfs_update_inode_transaction(inode);
2248                         trans_running = 1;
2249                         if (fs_changed(fs_gen, inode->i_sb)
2250                             && item_moved(&tmp_ih, &path)) {
2251                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2252                                                                  bh);
2253                                 goto research;
2254                         }
2255                 }
2256
2257                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2258
2259                 if (fs_changed(fs_gen, inode->i_sb)
2260                     && item_moved(&tmp_ih, &path)) {
2261                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2262                         goto research;
2263                 }
2264
2265                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2266                        copy_size);
2267
2268                 journal_mark_dirty(&th, inode->i_sb, bh);
2269                 bytes_copied += copy_size;
2270                 set_block_dev_mapped(bh_result, 0, inode);
2271
2272                 /* are there still bytes left? */
2273                 if (bytes_copied < bh_result->b_size &&
2274                     (byte_offset + bytes_copied) < inode->i_size) {
2275                         set_cpu_key_k_offset(&key,
2276                                              cpu_key_k_offset(&key) +
2277                                              copy_size);
2278                         goto research;
2279                 }
2280         } else {
2281                 reiserfs_warning(inode->i_sb, "clm-6003",
2282                                  "bad item inode %lu", inode->i_ino);
2283                 retval = -EIO;
2284                 goto out;
2285         }
2286         retval = 0;
2287
2288       out:
2289         pathrelse(&path);
2290         if (trans_running) {
2291                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2292                 if (err)
2293                         retval = err;
2294                 trans_running = 0;
2295         }
2296         reiserfs_write_unlock(inode->i_sb);
2297
2298         /* this is where we fill in holes in the file. */
2299         if (use_get_block) {
2300                 retval = reiserfs_get_block(inode, block, bh_result,
2301                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2302                                             | GET_BLOCK_NO_DANGLE);
2303                 if (!retval) {
2304                         if (!buffer_mapped(bh_result)
2305                             || bh_result->b_blocknr == 0) {
2306                                 /* get_block failed to find a mapped unformatted node. */
2307                                 use_get_block = 0;
2308                                 goto start_over;
2309                         }
2310                 }
2311         }
2312         kunmap(bh_result->b_page);
2313
2314         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2315                 /* we've copied data from the page into the direct item, so the
2316                  * buffer in the page is now clean, mark it to reflect that.
2317                  */
2318                 lock_buffer(bh_result);
2319                 clear_buffer_dirty(bh_result);
2320                 unlock_buffer(bh_result);
2321         }
2322         return retval;
2323 }
2324
2325 /*
2326  * mason@suse.com: updated in 2.5.54 to follow the same general io
2327  * start/recovery path as __block_write_full_page, along with special
2328  * code to handle reiserfs tails.
2329  */
2330 static int reiserfs_write_full_page(struct page *page,
2331                                     struct writeback_control *wbc)
2332 {
2333         struct inode *inode = page->mapping->host;
2334         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2335         int error = 0;
2336         unsigned long block;
2337         sector_t last_block;
2338         struct buffer_head *head, *bh;
2339         int partial = 0;
2340         int nr = 0;
2341         int checked = PageChecked(page);
2342         struct reiserfs_transaction_handle th;
2343         struct super_block *s = inode->i_sb;
2344         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2345         th.t_trans_id = 0;
2346
2347         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2348         if (checked && (current->flags & PF_MEMALLOC)) {
2349                 redirty_page_for_writepage(wbc, page);
2350                 unlock_page(page);
2351                 return 0;
2352         }
2353
2354         /* The page dirty bit is cleared before writepage is called, which
2355          * means we have to tell create_empty_buffers to make dirty buffers
2356          * The page really should be up to date at this point, so tossing
2357          * in the BH_Uptodate is just a sanity check.
2358          */
2359         if (!page_has_buffers(page)) {
2360                 create_empty_buffers(page, s->s_blocksize,
2361                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2362         }
2363         head = page_buffers(page);
2364
2365         /* last page in the file, zero out any contents past the
2366          ** last byte in the file
2367          */
2368         if (page->index >= end_index) {
2369                 unsigned last_offset;
2370
2371                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2372                 /* no file contents in this page */
2373                 if (page->index >= end_index + 1 || !last_offset) {
2374                         unlock_page(page);
2375                         return 0;
2376                 }
2377                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2378         }
2379         bh = head;
2380         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2381         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2382         /* first map all the buffers, logging any direct items we find */
2383         do {
2384                 if (block > last_block) {
2385                         /*
2386                          * This can happen when the block size is less than
2387                          * the page size.  The corresponding bytes in the page
2388                          * were zero filled above
2389                          */
2390                         clear_buffer_dirty(bh);
2391                         set_buffer_uptodate(bh);
2392                 } else if ((checked || buffer_dirty(bh)) &&
2393                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2394                                                        && bh->b_blocknr ==
2395                                                        0))) {
2396                         /* not mapped yet, or it points to a direct item, search
2397                          * the btree for the mapping info, and log any direct
2398                          * items found
2399                          */
2400                         if ((error = map_block_for_writepage(inode, bh, block))) {
2401                                 goto fail;
2402                         }
2403                 }
2404                 bh = bh->b_this_page;
2405                 block++;
2406         } while (bh != head);
2407
2408         /*
2409          * we start the transaction after map_block_for_writepage,
2410          * because it can create holes in the file (an unbounded operation).
2411          * starting it here, we can make a reliable estimate for how many
2412          * blocks we're going to log
2413          */
2414         if (checked) {
2415                 ClearPageChecked(page);
2416                 reiserfs_write_lock(s);
2417                 error = journal_begin(&th, s, bh_per_page + 1);
2418                 if (error) {
2419                         reiserfs_write_unlock(s);
2420                         goto fail;
2421                 }
2422                 reiserfs_update_inode_transaction(inode);
2423         }
2424         /* now go through and lock any dirty buffers on the page */
2425         do {
2426                 get_bh(bh);
2427                 if (!buffer_mapped(bh))
2428                         continue;
2429                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2430                         continue;
2431
2432                 if (checked) {
2433                         reiserfs_prepare_for_journal(s, bh, 1);
2434                         journal_mark_dirty(&th, s, bh);
2435                         continue;
2436                 }
2437                 /* from this point on, we know the buffer is mapped to a
2438                  * real block and not a direct item
2439                  */
2440                 if (wbc->sync_mode != WB_SYNC_NONE) {
2441                         lock_buffer(bh);
2442                 } else {
2443                         if (!trylock_buffer(bh)) {
2444                                 redirty_page_for_writepage(wbc, page);
2445                                 continue;
2446                         }
2447                 }
2448                 if (test_clear_buffer_dirty(bh)) {
2449                         mark_buffer_async_write(bh);
2450                 } else {
2451                         unlock_buffer(bh);
2452                 }
2453         } while ((bh = bh->b_this_page) != head);
2454
2455         if (checked) {
2456                 error = journal_end(&th, s, bh_per_page + 1);
2457                 reiserfs_write_unlock(s);
2458                 if (error)
2459                         goto fail;
2460         }
2461         BUG_ON(PageWriteback(page));
2462         set_page_writeback(page);
2463         unlock_page(page);
2464
2465         /*
2466          * since any buffer might be the only dirty buffer on the page,
2467          * the first submit_bh can bring the page out of writeback.
2468          * be careful with the buffers.
2469          */
2470         do {
2471                 struct buffer_head *next = bh->b_this_page;
2472                 if (buffer_async_write(bh)) {
2473                         submit_bh(WRITE, bh);
2474                         nr++;
2475                 }
2476                 put_bh(bh);
2477                 bh = next;
2478         } while (bh != head);
2479
2480         error = 0;
2481       done:
2482         if (nr == 0) {
2483                 /*
2484                  * if this page only had a direct item, it is very possible for
2485                  * no io to be required without there being an error.  Or,
2486                  * someone else could have locked them and sent them down the
2487                  * pipe without locking the page
2488                  */
2489                 bh = head;
2490                 do {
2491                         if (!buffer_uptodate(bh)) {
2492                                 partial = 1;
2493                                 break;
2494                         }
2495                         bh = bh->b_this_page;
2496                 } while (bh != head);
2497                 if (!partial)
2498                         SetPageUptodate(page);
2499                 end_page_writeback(page);
2500         }
2501         return error;
2502
2503       fail:
2504         /* catches various errors, we need to make sure any valid dirty blocks
2505          * get to the media.  The page is currently locked and not marked for
2506          * writeback
2507          */
2508         ClearPageUptodate(page);
2509         bh = head;
2510         do {
2511                 get_bh(bh);
2512                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2513                         lock_buffer(bh);
2514                         mark_buffer_async_write(bh);
2515                 } else {
2516                         /*
2517                          * clear any dirty bits that might have come from getting
2518                          * attached to a dirty page
2519                          */
2520                         clear_buffer_dirty(bh);
2521                 }
2522                 bh = bh->b_this_page;
2523         } while (bh != head);
2524         SetPageError(page);
2525         BUG_ON(PageWriteback(page));
2526         set_page_writeback(page);
2527         unlock_page(page);
2528         do {
2529                 struct buffer_head *next = bh->b_this_page;
2530                 if (buffer_async_write(bh)) {
2531                         clear_buffer_dirty(bh);
2532                         submit_bh(WRITE, bh);
2533                         nr++;
2534                 }
2535                 put_bh(bh);
2536                 bh = next;
2537         } while (bh != head);
2538         goto done;
2539 }
2540
2541 static int reiserfs_readpage(struct file *f, struct page *page)
2542 {
2543         return block_read_full_page(page, reiserfs_get_block);
2544 }
2545
2546 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2547 {
2548         struct inode *inode = page->mapping->host;
2549         reiserfs_wait_on_write_block(inode->i_sb);
2550         return reiserfs_write_full_page(page, wbc);
2551 }
2552
2553 static void reiserfs_truncate_failed_write(struct inode *inode)
2554 {
2555         truncate_inode_pages(inode->i_mapping, inode->i_size);
2556         reiserfs_truncate_file(inode, 0);
2557 }
2558
2559 static int reiserfs_write_begin(struct file *file,
2560                                 struct address_space *mapping,
2561                                 loff_t pos, unsigned len, unsigned flags,
2562                                 struct page **pagep, void **fsdata)
2563 {
2564         struct inode *inode;
2565         struct page *page;
2566         pgoff_t index;
2567         int ret;
2568         int old_ref = 0;
2569
2570         inode = mapping->host;
2571         *fsdata = 0;
2572         if (flags & AOP_FLAG_CONT_EXPAND &&
2573             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2574                 pos ++;
2575                 *fsdata = (void *)(unsigned long)flags;
2576         }
2577
2578         index = pos >> PAGE_CACHE_SHIFT;
2579         page = grab_cache_page_write_begin(mapping, index, flags);
2580         if (!page)
2581                 return -ENOMEM;
2582         *pagep = page;
2583
2584         reiserfs_wait_on_write_block(inode->i_sb);
2585         fix_tail_page_for_writing(page);
2586         if (reiserfs_transaction_running(inode->i_sb)) {
2587                 struct reiserfs_transaction_handle *th;
2588                 th = (struct reiserfs_transaction_handle *)current->
2589                     journal_info;
2590                 BUG_ON(!th->t_refcount);
2591                 BUG_ON(!th->t_trans_id);
2592                 old_ref = th->t_refcount;
2593                 th->t_refcount++;
2594         }
2595         ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2596         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2597                 struct reiserfs_transaction_handle *th = current->journal_info;
2598                 /* this gets a little ugly.  If reiserfs_get_block returned an
2599                  * error and left a transacstion running, we've got to close it,
2600                  * and we've got to free handle if it was a persistent transaction.
2601                  *
2602                  * But, if we had nested into an existing transaction, we need
2603                  * to just drop the ref count on the handle.
2604                  *
2605                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2606                  * and it was a persistent trans.  Otherwise, it was nested above.
2607                  */
2608                 if (th->t_refcount > old_ref) {
2609                         if (old_ref)
2610                                 th->t_refcount--;
2611                         else {
2612                                 int err;
2613                                 reiserfs_write_lock(inode->i_sb);
2614                                 err = reiserfs_end_persistent_transaction(th);
2615                                 reiserfs_write_unlock(inode->i_sb);
2616                                 if (err)
2617                                         ret = err;
2618                         }
2619                 }
2620         }
2621         if (ret) {
2622                 unlock_page(page);
2623                 page_cache_release(page);
2624                 /* Truncate allocated blocks */
2625                 reiserfs_truncate_failed_write(inode);
2626         }
2627         return ret;
2628 }
2629
2630 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2631 {
2632         struct inode *inode = page->mapping->host;
2633         int ret;
2634         int old_ref = 0;
2635
2636         reiserfs_write_unlock(inode->i_sb);
2637         reiserfs_wait_on_write_block(inode->i_sb);
2638         reiserfs_write_lock(inode->i_sb);
2639
2640         fix_tail_page_for_writing(page);
2641         if (reiserfs_transaction_running(inode->i_sb)) {
2642                 struct reiserfs_transaction_handle *th;
2643                 th = (struct reiserfs_transaction_handle *)current->
2644                     journal_info;
2645                 BUG_ON(!th->t_refcount);
2646                 BUG_ON(!th->t_trans_id);
2647                 old_ref = th->t_refcount;
2648                 th->t_refcount++;
2649         }
2650
2651         ret = __block_write_begin(page, from, len, reiserfs_get_block);
2652         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2653                 struct reiserfs_transaction_handle *th = current->journal_info;
2654                 /* this gets a little ugly.  If reiserfs_get_block returned an
2655                  * error and left a transacstion running, we've got to close it,
2656                  * and we've got to free handle if it was a persistent transaction.
2657                  *
2658                  * But, if we had nested into an existing transaction, we need
2659                  * to just drop the ref count on the handle.
2660                  *
2661                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2662                  * and it was a persistent trans.  Otherwise, it was nested above.
2663                  */
2664                 if (th->t_refcount > old_ref) {
2665                         if (old_ref)
2666                                 th->t_refcount--;
2667                         else {
2668                                 int err;
2669                                 reiserfs_write_lock(inode->i_sb);
2670                                 err = reiserfs_end_persistent_transaction(th);
2671                                 reiserfs_write_unlock(inode->i_sb);
2672                                 if (err)
2673                                         ret = err;
2674                         }
2675                 }
2676         }
2677         return ret;
2678
2679 }
2680
2681 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2682 {
2683         return generic_block_bmap(as, block, reiserfs_bmap);
2684 }
2685
2686 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2687                               loff_t pos, unsigned len, unsigned copied,
2688                               struct page *page, void *fsdata)
2689 {
2690         struct inode *inode = page->mapping->host;
2691         int ret = 0;
2692         int update_sd = 0;
2693         struct reiserfs_transaction_handle *th;
2694         unsigned start;
2695         int lock_depth = 0;
2696         bool locked = false;
2697
2698         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2699                 pos ++;
2700
2701         reiserfs_wait_on_write_block(inode->i_sb);
2702         if (reiserfs_transaction_running(inode->i_sb))
2703                 th = current->journal_info;
2704         else
2705                 th = NULL;
2706
2707         start = pos & (PAGE_CACHE_SIZE - 1);
2708         if (unlikely(copied < len)) {
2709                 if (!PageUptodate(page))
2710                         copied = 0;
2711
2712                 page_zero_new_buffers(page, start + copied, start + len);
2713         }
2714         flush_dcache_page(page);
2715
2716         reiserfs_commit_page(inode, page, start, start + copied);
2717
2718         /* generic_commit_write does this for us, but does not update the
2719          ** transaction tracking stuff when the size changes.  So, we have
2720          ** to do the i_size updates here.
2721          */
2722         if (pos + copied > inode->i_size) {
2723                 struct reiserfs_transaction_handle myth;
2724                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2725                 locked = true;
2726                 /* If the file have grown beyond the border where it
2727                    can have a tail, unmark it as needing a tail
2728                    packing */
2729                 if ((have_large_tails(inode->i_sb)
2730                      && inode->i_size > i_block_size(inode) * 4)
2731                     || (have_small_tails(inode->i_sb)
2732                         && inode->i_size > i_block_size(inode)))
2733                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2734
2735                 ret = journal_begin(&myth, inode->i_sb, 1);
2736                 if (ret)
2737                         goto journal_error;
2738
2739                 reiserfs_update_inode_transaction(inode);
2740                 inode->i_size = pos + copied;
2741                 /*
2742                  * this will just nest into our transaction.  It's important
2743                  * to use mark_inode_dirty so the inode gets pushed around on the
2744                  * dirty lists, and so that O_SYNC works as expected
2745                  */
2746                 mark_inode_dirty(inode);
2747                 reiserfs_update_sd(&myth, inode);
2748                 update_sd = 1;
2749                 ret = journal_end(&myth, inode->i_sb, 1);
2750                 if (ret)
2751                         goto journal_error;
2752         }
2753         if (th) {
2754                 if (!locked) {
2755                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2756                         locked = true;
2757                 }
2758                 if (!update_sd)
2759                         mark_inode_dirty(inode);
2760                 ret = reiserfs_end_persistent_transaction(th);
2761                 if (ret)
2762                         goto out;
2763         }
2764
2765       out:
2766         if (locked)
2767                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2768         unlock_page(page);
2769         page_cache_release(page);
2770
2771         if (pos + len > inode->i_size)
2772                 reiserfs_truncate_failed_write(inode);
2773
2774         return ret == 0 ? copied : ret;
2775
2776       journal_error:
2777         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2778         locked = false;
2779         if (th) {
2780                 if (!update_sd)
2781                         reiserfs_update_sd(th, inode);
2782                 ret = reiserfs_end_persistent_transaction(th);
2783         }
2784         goto out;
2785 }
2786
2787 int reiserfs_commit_write(struct file *f, struct page *page,
2788                           unsigned from, unsigned to)
2789 {
2790         struct inode *inode = page->mapping->host;
2791         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2792         int ret = 0;
2793         int update_sd = 0;
2794         struct reiserfs_transaction_handle *th = NULL;
2795
2796         reiserfs_write_unlock(inode->i_sb);
2797         reiserfs_wait_on_write_block(inode->i_sb);
2798         reiserfs_write_lock(inode->i_sb);
2799
2800         if (reiserfs_transaction_running(inode->i_sb)) {
2801                 th = current->journal_info;
2802         }
2803         reiserfs_commit_page(inode, page, from, to);
2804
2805         /* generic_commit_write does this for us, but does not update the
2806          ** transaction tracking stuff when the size changes.  So, we have
2807          ** to do the i_size updates here.
2808          */
2809         if (pos > inode->i_size) {
2810                 struct reiserfs_transaction_handle myth;
2811                 /* If the file have grown beyond the border where it
2812                    can have a tail, unmark it as needing a tail
2813                    packing */
2814                 if ((have_large_tails(inode->i_sb)
2815                      && inode->i_size > i_block_size(inode) * 4)
2816                     || (have_small_tails(inode->i_sb)
2817                         && inode->i_size > i_block_size(inode)))
2818                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2819
2820                 ret = journal_begin(&myth, inode->i_sb, 1);
2821                 if (ret)
2822                         goto journal_error;
2823
2824                 reiserfs_update_inode_transaction(inode);
2825                 inode->i_size = pos;
2826                 /*
2827                  * this will just nest into our transaction.  It's important
2828                  * to use mark_inode_dirty so the inode gets pushed around on the
2829                  * dirty lists, and so that O_SYNC works as expected
2830                  */
2831                 mark_inode_dirty(inode);
2832                 reiserfs_update_sd(&myth, inode);
2833                 update_sd = 1;
2834                 ret = journal_end(&myth, inode->i_sb, 1);
2835                 if (ret)
2836                         goto journal_error;
2837         }
2838         if (th) {
2839                 if (!update_sd)
2840                         mark_inode_dirty(inode);
2841                 ret = reiserfs_end_persistent_transaction(th);
2842                 if (ret)
2843                         goto out;
2844         }
2845
2846       out:
2847         return ret;
2848
2849       journal_error:
2850         if (th) {
2851                 if (!update_sd)
2852                         reiserfs_update_sd(th, inode);
2853                 ret = reiserfs_end_persistent_transaction(th);
2854         }
2855
2856         return ret;
2857 }
2858
2859 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2860 {
2861         if (reiserfs_attrs(inode->i_sb)) {
2862                 if (sd_attrs & REISERFS_SYNC_FL)
2863                         inode->i_flags |= S_SYNC;
2864                 else
2865                         inode->i_flags &= ~S_SYNC;
2866                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2867                         inode->i_flags |= S_IMMUTABLE;
2868                 else
2869                         inode->i_flags &= ~S_IMMUTABLE;
2870                 if (sd_attrs & REISERFS_APPEND_FL)
2871                         inode->i_flags |= S_APPEND;
2872                 else
2873                         inode->i_flags &= ~S_APPEND;
2874                 if (sd_attrs & REISERFS_NOATIME_FL)
2875                         inode->i_flags |= S_NOATIME;
2876                 else
2877                         inode->i_flags &= ~S_NOATIME;
2878                 if (sd_attrs & REISERFS_NOTAIL_FL)
2879                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2880                 else
2881                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2882         }
2883 }
2884
2885 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2886 {
2887         if (reiserfs_attrs(inode->i_sb)) {
2888                 if (inode->i_flags & S_IMMUTABLE)
2889                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2890                 else
2891                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2892                 if (inode->i_flags & S_SYNC)
2893                         *sd_attrs |= REISERFS_SYNC_FL;
2894                 else
2895                         *sd_attrs &= ~REISERFS_SYNC_FL;
2896                 if (inode->i_flags & S_NOATIME)
2897                         *sd_attrs |= REISERFS_NOATIME_FL;
2898                 else
2899                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2900                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2901                         *sd_attrs |= REISERFS_NOTAIL_FL;
2902                 else
2903                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2904         }
2905 }
2906
2907 /* decide if this buffer needs to stay around for data logging or ordered
2908 ** write purposes
2909 */
2910 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2911 {
2912         int ret = 1;
2913         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2914
2915         lock_buffer(bh);
2916         spin_lock(&j->j_dirty_buffers_lock);
2917         if (!buffer_mapped(bh)) {
2918                 goto free_jh;
2919         }
2920         /* the page is locked, and the only places that log a data buffer
2921          * also lock the page.
2922          */
2923         if (reiserfs_file_data_log(inode)) {
2924                 /*
2925                  * very conservative, leave the buffer pinned if
2926                  * anyone might need it.
2927                  */
2928                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2929                         ret = 0;
2930                 }
2931         } else  if (buffer_dirty(bh)) {
2932                 struct reiserfs_journal_list *jl;
2933                 struct reiserfs_jh *jh = bh->b_private;
2934
2935                 /* why is this safe?
2936                  * reiserfs_setattr updates i_size in the on disk
2937                  * stat data before allowing vmtruncate to be called.
2938                  *
2939                  * If buffer was put onto the ordered list for this
2940                  * transaction, we know for sure either this transaction
2941                  * or an older one already has updated i_size on disk,
2942                  * and this ordered data won't be referenced in the file
2943                  * if we crash.
2944                  *
2945                  * if the buffer was put onto the ordered list for an older
2946                  * transaction, we need to leave it around
2947                  */
2948                 if (jh && (jl = jh->jl)
2949                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2950                         ret = 0;
2951         }
2952       free_jh:
2953         if (ret && bh->b_private) {
2954                 reiserfs_free_jh(bh);
2955         }
2956         spin_unlock(&j->j_dirty_buffers_lock);
2957         unlock_buffer(bh);
2958         return ret;
2959 }
2960
2961 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2962 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2963 {
2964         struct buffer_head *head, *bh, *next;
2965         struct inode *inode = page->mapping->host;
2966         unsigned int curr_off = 0;
2967         int ret = 1;
2968
2969         BUG_ON(!PageLocked(page));
2970
2971         if (offset == 0)
2972                 ClearPageChecked(page);
2973
2974         if (!page_has_buffers(page))
2975                 goto out;
2976
2977         head = page_buffers(page);
2978         bh = head;
2979         do {
2980                 unsigned int next_off = curr_off + bh->b_size;
2981                 next = bh->b_this_page;
2982
2983                 /*
2984                  * is this block fully invalidated?
2985                  */
2986                 if (offset <= curr_off) {
2987                         if (invalidatepage_can_drop(inode, bh))
2988                                 reiserfs_unmap_buffer(bh);
2989                         else
2990                                 ret = 0;
2991                 }
2992                 curr_off = next_off;
2993                 bh = next;
2994         } while (bh != head);
2995
2996         /*
2997          * We release buffers only if the entire page is being invalidated.
2998          * The get_block cached value has been unconditionally invalidated,
2999          * so real IO is not possible anymore.
3000          */
3001         if (!offset && ret) {
3002                 ret = try_to_release_page(page, 0);
3003                 /* maybe should BUG_ON(!ret); - neilb */
3004         }
3005       out:
3006         return;
3007 }
3008
3009 static int reiserfs_set_page_dirty(struct page *page)
3010 {
3011         struct inode *inode = page->mapping->host;
3012         if (reiserfs_file_data_log(inode)) {
3013                 SetPageChecked(page);
3014                 return __set_page_dirty_nobuffers(page);
3015         }
3016         return __set_page_dirty_buffers(page);
3017 }
3018
3019 /*
3020  * Returns 1 if the page's buffers were dropped.  The page is locked.
3021  *
3022  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3023  * in the buffers at page_buffers(page).
3024  *
3025  * even in -o notail mode, we can't be sure an old mount without -o notail
3026  * didn't create files with tails.
3027  */
3028 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3029 {
3030         struct inode *inode = page->mapping->host;
3031         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3032         struct buffer_head *head;
3033         struct buffer_head *bh;
3034         int ret = 1;
3035
3036         WARN_ON(PageChecked(page));
3037         spin_lock(&j->j_dirty_buffers_lock);
3038         head = page_buffers(page);
3039         bh = head;
3040         do {
3041                 if (bh->b_private) {
3042                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3043                                 reiserfs_free_jh(bh);
3044                         } else {
3045                                 ret = 0;
3046                                 break;
3047                         }
3048                 }
3049                 bh = bh->b_this_page;
3050         } while (bh != head);
3051         if (ret)
3052                 ret = try_to_free_buffers(page);
3053         spin_unlock(&j->j_dirty_buffers_lock);
3054         return ret;
3055 }
3056
3057 /* We thank Mingming Cao for helping us understand in great detail what
3058    to do in this section of the code. */
3059 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3060                                   const struct iovec *iov, loff_t offset,
3061                                   unsigned long nr_segs)
3062 {
3063         struct file *file = iocb->ki_filp;
3064         struct inode *inode = file->f_mapping->host;
3065         ssize_t ret;
3066
3067         ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3068                                   offset, nr_segs,
3069                                   reiserfs_get_blocks_direct_io, NULL);
3070
3071         /*
3072          * In case of error extending write may have instantiated a few
3073          * blocks outside i_size. Trim these off again.
3074          */
3075         if (unlikely((rw & WRITE) && ret < 0)) {
3076                 loff_t isize = i_size_read(inode);
3077                 loff_t end = offset + iov_length(iov, nr_segs);
3078
3079                 if (end > isize)
3080                         vmtruncate(inode, isize);
3081         }
3082
3083         return ret;
3084 }
3085
3086 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3087 {
3088         struct inode *inode = dentry->d_inode;
3089         unsigned int ia_valid;
3090         int depth;
3091         int error;
3092
3093         error = inode_change_ok(inode, attr);
3094         if (error)
3095                 return error;
3096
3097         /* must be turned off for recursive notify_change calls */
3098         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3099
3100         depth = reiserfs_write_lock_once(inode->i_sb);
3101         if (is_quota_modification(inode, attr))
3102                 dquot_initialize(inode);
3103
3104         if (attr->ia_valid & ATTR_SIZE) {
3105                 /* version 2 items will be caught by the s_maxbytes check
3106                  ** done for us in vmtruncate
3107                  */
3108                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3109                     attr->ia_size > MAX_NON_LFS) {
3110                         error = -EFBIG;
3111                         goto out;
3112                 }
3113                 /* fill in hole pointers in the expanding truncate case. */
3114                 if (attr->ia_size > inode->i_size) {
3115                         error = generic_cont_expand_simple(inode, attr->ia_size);
3116                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3117                                 int err;
3118                                 struct reiserfs_transaction_handle th;
3119                                 /* we're changing at most 2 bitmaps, inode + super */
3120                                 err = journal_begin(&th, inode->i_sb, 4);
3121                                 if (!err) {
3122                                         reiserfs_discard_prealloc(&th, inode);
3123                                         err = journal_end(&th, inode->i_sb, 4);
3124                                 }
3125                                 if (err)
3126                                         error = err;
3127                         }
3128                         if (error)
3129                                 goto out;
3130                         /*
3131                          * file size is changed, ctime and mtime are
3132                          * to be updated
3133                          */
3134                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3135                 }
3136         }
3137
3138         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3139              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3140             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3141                 /* stat data of format v3.5 has 16 bit uid and gid */
3142                 error = -EINVAL;
3143                 goto out;
3144         }
3145
3146         if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3147             (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3148                 struct reiserfs_transaction_handle th;
3149                 int jbegin_count =
3150                     2 *
3151                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3152                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3153                     2;
3154
3155                 error = reiserfs_chown_xattrs(inode, attr);
3156
3157                 if (error)
3158                         return error;
3159
3160                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3161                 error = journal_begin(&th, inode->i_sb, jbegin_count);
3162                 if (error)
3163                         goto out;
3164                 error = dquot_transfer(inode, attr);
3165                 if (error) {
3166                         journal_end(&th, inode->i_sb, jbegin_count);
3167                         goto out;
3168                 }
3169
3170                 /* Update corresponding info in inode so that everything is in
3171                  * one transaction */
3172                 if (attr->ia_valid & ATTR_UID)
3173                         inode->i_uid = attr->ia_uid;
3174                 if (attr->ia_valid & ATTR_GID)
3175                         inode->i_gid = attr->ia_gid;
3176                 mark_inode_dirty(inode);
3177                 error = journal_end(&th, inode->i_sb, jbegin_count);
3178                 if (error)
3179                         goto out;
3180         }
3181
3182         /*
3183          * Relax the lock here, as it might truncate the
3184          * inode pages and wait for inode pages locks.
3185          * To release such page lock, the owner needs the
3186          * reiserfs lock
3187          */
3188         reiserfs_write_unlock_once(inode->i_sb, depth);
3189         if ((attr->ia_valid & ATTR_SIZE) &&
3190             attr->ia_size != i_size_read(inode))
3191                 error = vmtruncate(inode, attr->ia_size);
3192
3193         if (!error) {
3194                 setattr_copy(inode, attr);
3195                 mark_inode_dirty(inode);
3196         }
3197         depth = reiserfs_write_lock_once(inode->i_sb);
3198
3199         if (!error && reiserfs_posixacl(inode->i_sb)) {
3200                 if (attr->ia_valid & ATTR_MODE)
3201                         error = reiserfs_acl_chmod(inode);
3202         }
3203
3204       out:
3205         reiserfs_write_unlock_once(inode->i_sb, depth);
3206
3207         return error;
3208 }
3209
3210 const struct address_space_operations reiserfs_address_space_operations = {
3211         .writepage = reiserfs_writepage,
3212         .readpage = reiserfs_readpage,
3213         .readpages = reiserfs_readpages,
3214         .releasepage = reiserfs_releasepage,
3215         .invalidatepage = reiserfs_invalidatepage,
3216         .sync_page = block_sync_page,
3217         .write_begin = reiserfs_write_begin,
3218         .write_end = reiserfs_write_end,
3219         .bmap = reiserfs_aop_bmap,
3220         .direct_IO = reiserfs_direct_IO,
3221         .set_page_dirty = reiserfs_set_page_dirty,
3222 };