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/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <asm/uaccess.h>
  15 #include <asm/unaligned.h>
  16 #include <linux/buffer_head.h>
  17 #include <linux/mpage.h>
  18 #include <linux/writeback.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/swap.h>
  21
  22 int reiserfs_commit_write(struct file *f, struct page *page,
  23                           unsigned from, unsigned to);
  24
  25 void reiserfs_evict_inode(struct inode *inode)
  26 {
  27         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
  28         int jbegin_count =
  29             JOURNAL_PER_BALANCE_CNT * 2 +
  30             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  31         struct reiserfs_transaction_handle th;
  32         int depth;
  33         int err;
  34
  35         if (!inode->i_nlink && !is_bad_inode(inode))
  36                 dquot_initialize(inode);
  37
  38         truncate_inode_pages(&inode->i_data, 0);
  39         if (inode->i_nlink)
  40                 goto no_delete;
  41
  42         depth = reiserfs_write_lock_once(inode->i_sb);
  43
  44         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
  45         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
  46                 reiserfs_delete_xattrs(inode);
  47
  48                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  49                         goto out;
  50                 reiserfs_update_inode_transaction(inode);
  51
  52                 reiserfs_discard_prealloc(&th, inode);
  53
  54                 err = reiserfs_delete_object(&th, inode);
  55
  56                 /* Do quota update inside a transaction for journaled quotas. We must do that
  57                  * after delete_object so that quota updates go into the same transaction as
  58                  * stat data deletion */
  59                 if (!err)
  60                         dquot_free_inode(inode);
  61
  62                 if (journal_end(&th, inode->i_sb, jbegin_count))
  63                         goto out;
  64
  65                 /* check return value from reiserfs_delete_object after
  66                  * ending the transaction
  67                  */
  68                 if (err)
  69                     goto out;
  70
  71                 /* all items of file are deleted, so we can remove "save" link */
  72                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
  73                                                                  * about an error here */
  74         } else {
  75                 /* no object items are in the tree */
  76                 ;
  77         }
  78       out:
  79         end_writeback(inode);   /* note this must go after the journal_end to prevent deadlock */
  80         dquot_drop(inode);
  81         inode->i_blocks = 0;
  82         reiserfs_write_unlock_once(inode->i_sb, depth);
  83         return;
  84
  85 no_delete:
  86         end_writeback(inode);
  87         dquot_drop(inode);
  88 }
  89
  90 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
  91                           __u32 objectid, loff_t offset, int type, int length)
  92 {
  93         key->version = version;
  94
  95         key->on_disk_key.k_dir_id = dirid;
  96         key->on_disk_key.k_objectid = objectid;
  97         set_cpu_key_k_offset(key, offset);
  98         set_cpu_key_k_type(key, type);
  99         key->key_length = length;
 100 }
 101
 102 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
 103    offset and type of key */
 104 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 105                   int type, int length)
 106 {
 107         _make_cpu_key(key, get_inode_item_key_version(inode),
 108                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 109                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 110                       length);
 111 }
 112
 113 //
 114 // when key is 0, do not set version and short key
 115 //
 116 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 117                               int version,
 118                               loff_t offset, int type, int length,
 119                               int entry_count /*or ih_free_space */ )
 120 {
 121         if (key) {
 122                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 123                 ih->ih_key.k_objectid =
 124                     cpu_to_le32(key->on_disk_key.k_objectid);
 125         }
 126         put_ih_version(ih, version);
 127         set_le_ih_k_offset(ih, offset);
 128         set_le_ih_k_type(ih, type);
 129         put_ih_item_len(ih, length);
 130         /*    set_ih_free_space (ih, 0); */
 131         // for directory items it is entry count, for directs and stat
 132         // datas - 0xffff, for indirects - 0
 133         put_ih_entry_count(ih, entry_count);
 134 }
 135
 136 //
 137 // FIXME: we might cache recently accessed indirect item
 138
 139 // Ugh.  Not too eager for that....
 140 //  I cut the code until such time as I see a convincing argument (benchmark).
 141 // I don't want a bloated inode struct..., and I don't like code complexity....
 142
 143 /* cutting the code is fine, since it really isn't in use yet and is easy
 144 ** to add back in.  But, Vladimir has a really good idea here.  Think
 145 ** about what happens for reading a file.  For each page,
 146 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
 147 ** an indirect item.  This indirect item has X number of pointers, where
 148 ** X is a big number if we've done the block allocation right.  But,
 149 ** we only use one or two of these pointers during each call to readpage,
 150 ** needlessly researching again later on.
 151 **
 152 ** The size of the cache could be dynamic based on the size of the file.
 153 **
 154 ** I'd also like to see us cache the location the stat data item, since
 155 ** we are needlessly researching for that frequently.
 156 **
 157 ** --chris
 158 */
 159
 160 /* If this page has a file tail in it, and
 161 ** it was read in by get_block_create_0, the page data is valid,
 162 ** but tail is still sitting in a direct item, and we can't write to
 163 ** it.  So, look through this page, and check all the mapped buffers
 164 ** to make sure they have valid block numbers.  Any that don't need
 165 ** to be unmapped, so that __block_write_begin will correctly call
 166 ** reiserfs_get_block to convert the tail into an unformatted node
 167 */
 168 static inline void fix_tail_page_for_writing(struct page *page)
 169 {
 170         struct buffer_head *head, *next, *bh;
 171
 172         if (page && page_has_buffers(page)) {
 173                 head = page_buffers(page);
 174                 bh = head;
 175                 do {
 176                         next = bh->b_this_page;
 177                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 178                                 reiserfs_unmap_buffer(bh);
 179                         }
 180                         bh = next;
 181                 } while (bh != head);
 182         }
 183 }
 184
 185 /* reiserfs_get_block does not need to allocate a block only if it has been
 186    done already or non-hole position has been found in the indirect item */
 187 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 188                                     struct item_head *ih,
 189                                     __le32 * item, int pos_in_item)
 190 {
 191         if (allocated)
 192                 return 0;
 193         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 194             get_block_num(item, pos_in_item))
 195                 return 0;
 196         return 1;
 197 }
 198
 199 static inline int indirect_item_found(int retval, struct item_head *ih)
 200 {
 201         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 202 }
 203
 204 static inline void set_block_dev_mapped(struct buffer_head *bh,
 205                                         b_blocknr_t block, struct inode *inode)
 206 {
 207         map_bh(bh, inode->i_sb, block);
 208 }
 209
 210 //
 211 // files which were created in the earlier version can not be longer,
 212 // than 2 gb
 213 //
 214 static int file_capable(struct inode *inode, sector_t block)
 215 {
 216         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
 217             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
 218                 return 1;
 219
 220         return 0;
 221 }
 222
 223 static int restart_transaction(struct reiserfs_transaction_handle *th,
 224                                struct inode *inode, struct treepath *path)
 225 {
 226         struct super_block *s = th->t_super;
 227         int len = th->t_blocks_allocated;
 228         int err;
 229
 230         BUG_ON(!th->t_trans_id);
 231         BUG_ON(!th->t_refcount);
 232
 233         pathrelse(path);
 234
 235         /* we cannot restart while nested */
 236         if (th->t_refcount > 1) {
 237                 return 0;
 238         }
 239         reiserfs_update_sd(th, inode);
 240         err = journal_end(th, s, len);
 241         if (!err) {
 242                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 243                 if (!err)
 244                         reiserfs_update_inode_transaction(inode);
 245         }
 246         return err;
 247 }
 248
 249 // it is called by get_block when create == 0. Returns block number
 250 // for 'block'-th logical block of file. When it hits direct item it
 251 // returns 0 (being called from bmap) or read direct item into piece
 252 // of page (bh_result)
 253
 254 // Please improve the english/clarity in the comment above, as it is
 255 // hard to understand.
 256
 257 static int _get_block_create_0(struct inode *inode, sector_t block,
 258                                struct buffer_head *bh_result, int args)
 259 {
 260         INITIALIZE_PATH(path);
 261         struct cpu_key key;
 262         struct buffer_head *bh;
 263         struct item_head *ih, tmp_ih;
 264         b_blocknr_t blocknr;
 265         char *p = NULL;
 266         int chars;
 267         int ret;
 268         int result;
 269         int done = 0;
 270         unsigned long offset;
 271
 272         // prepare the key to look for the 'block'-th block of file
 273         make_cpu_key(&key, inode,
 274                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 275                      3);
 276
 277         result = search_for_position_by_key(inode->i_sb, &key, &path);
 278         if (result != POSITION_FOUND) {
 279                 pathrelse(&path);
 280                 if (p)
 281                         kunmap(bh_result->b_page);
 282                 if (result == IO_ERROR)
 283                         return -EIO;
 284                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 285                 // That there is some MMAPED data associated with it that is yet to be written to disk.
 286                 if ((args & GET_BLOCK_NO_HOLE)
 287                     && !PageUptodate(bh_result->b_page)) {
 288                         return -ENOENT;
 289                 }
 290                 return 0;
 291         }
 292         //
 293         bh = get_last_bh(&path);
 294         ih = get_ih(&path);
 295         if (is_indirect_le_ih(ih)) {
 296                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
 297
 298                 /* FIXME: here we could cache indirect item or part of it in
 299                    the inode to avoid search_by_key in case of subsequent
 300                    access to file */
 301                 blocknr = get_block_num(ind_item, path.pos_in_item);
 302                 ret = 0;
 303                 if (blocknr) {
 304                         map_bh(bh_result, inode->i_sb, blocknr);
 305                         if (path.pos_in_item ==
 306                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 307                                 set_buffer_boundary(bh_result);
 308                         }
 309                 } else
 310                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
 311                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
 312                 if ((args & GET_BLOCK_NO_HOLE)
 313                             && !PageUptodate(bh_result->b_page)) {
 314                         ret = -ENOENT;
 315                 }
 316
 317                 pathrelse(&path);
 318                 if (p)
 319                         kunmap(bh_result->b_page);
 320                 return ret;
 321         }
 322         // requested data are in direct item(s)
 323         if (!(args & GET_BLOCK_READ_DIRECT)) {
 324                 // we are called by bmap. FIXME: we can not map block of file
 325                 // when it is stored in direct item(s)
 326                 pathrelse(&path);
 327                 if (p)
 328                         kunmap(bh_result->b_page);
 329                 return -ENOENT;
 330         }
 331
 332         /* if we've got a direct item, and the buffer or page was uptodate,
 333          ** we don't want to pull data off disk again.  skip to the
 334          ** end, where we map the buffer and return
 335          */
 336         if (buffer_uptodate(bh_result)) {
 337                 goto finished;
 338         } else
 339                 /*
 340                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
 341                  ** pages without any buffers.  If the page is up to date, we don't want
 342                  ** read old data off disk.  Set the up to date bit on the buffer instead
 343                  ** and jump to the end
 344                  */
 345         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 346                 set_buffer_uptodate(bh_result);
 347                 goto finished;
 348         }
 349         // read file tail into part of page
 350         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
 351         copy_item_head(&tmp_ih, ih);
 352
 353         /* we only want to kmap if we are reading the tail into the page.
 354          ** this is not the common case, so we don't kmap until we are
 355          ** sure we need to.  But, this means the item might move if
 356          ** kmap schedules
 357          */
 358         if (!p)
 359                 p = (char *)kmap(bh_result->b_page);
 360
 361         p += offset;
 362         memset(p, 0, inode->i_sb->s_blocksize);
 363         do {
 364                 if (!is_direct_le_ih(ih)) {
 365                         BUG();
 366                 }
 367                 /* make sure we don't read more bytes than actually exist in
 368                  ** the file.  This can happen in odd cases where i_size isn't
 369                  ** correct, and when direct item padding results in a few
 370                  ** extra bytes at the end of the direct item
 371                  */
 372                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 373                         break;
 374                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 375                         chars =
 376                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 377                             path.pos_in_item;
 378                         done = 1;
 379                 } else {
 380                         chars = ih_item_len(ih) - path.pos_in_item;
 381                 }
 382                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
 383
 384                 if (done)
 385                         break;
 386
 387                 p += chars;
 388
 389                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 390                         // we done, if read direct item is not the last item of
 391                         // node FIXME: we could try to check right delimiting key
 392                         // to see whether direct item continues in the right
 393                         // neighbor or rely on i_size
 394                         break;
 395
 396                 // update key to look for the next piece
 397                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 398                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 399                 if (result != POSITION_FOUND)
 400                         // i/o error most likely
 401                         break;
 402                 bh = get_last_bh(&path);
 403                 ih = get_ih(&path);
 404         } while (1);
 405
 406         flush_dcache_page(bh_result->b_page);
 407         kunmap(bh_result->b_page);
 408
 409       finished:
 410         pathrelse(&path);
 411
 412         if (result == IO_ERROR)
 413                 return -EIO;
 414
 415         /* this buffer has valid data, but isn't valid for io.  mapping it to
 416          * block #0 tells the rest of reiserfs it just has a tail in it
 417          */
 418         map_bh(bh_result, inode->i_sb, 0);
 419         set_buffer_uptodate(bh_result);
 420         return 0;
 421 }
 422
 423 // this is called to create file map. So, _get_block_create_0 will not
 424 // read direct item
 425 static int reiserfs_bmap(struct inode *inode, sector_t block,
 426                          struct buffer_head *bh_result, int create)
 427 {
 428         if (!file_capable(inode, block))
 429                 return -EFBIG;
 430
 431         reiserfs_write_lock(inode->i_sb);
 432         /* do not read the direct item */
 433         _get_block_create_0(inode, block, bh_result, 0);
 434         reiserfs_write_unlock(inode->i_sb);
 435         return 0;
 436 }
 437
 438 /* special version of get_block that is only used by grab_tail_page right
 439 ** now.  It is sent to __block_write_begin, and when you try to get a
 440 ** block past the end of the file (or a block from a hole) it returns
 441 ** -ENOENT instead of a valid buffer.  __block_write_begin expects to
 442 ** be able to do i/o on the buffers returned, unless an error value
 443 ** is also returned.
 444 **
 445 ** So, this allows __block_write_begin to be used for reading a single block
 446 ** in a page.  Where it does not produce a valid page for holes, or past the
 447 ** end of the file.  This turns out to be exactly what we need for reading
 448 ** tails for conversion.
 449 **
 450 ** The point of the wrapper is forcing a certain value for create, even
 451 ** though the VFS layer is calling this function with create==1.  If you
 452 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 453 ** don't use this function.
 454 */
 455 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 456                                        struct buffer_head *bh_result,
 457                                        int create)
 458 {
 459         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 460 }
 461
 462 /* This is special helper for reiserfs_get_block in case we are executing
 463    direct_IO request. */
 464 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 465                                          sector_t iblock,
 466                                          struct buffer_head *bh_result,
 467                                          int create)
 468 {
 469         int ret;
 470
 471         bh_result->b_page = NULL;
 472
 473         /* We set the b_size before reiserfs_get_block call since it is
 474            referenced in convert_tail_for_hole() that may be called from
 475            reiserfs_get_block() */
 476         bh_result->b_size = (1 << inode->i_blkbits);
 477
 478         ret = reiserfs_get_block(inode, iblock, bh_result,
 479                                  create | GET_BLOCK_NO_DANGLE);
 480         if (ret)
 481                 goto out;
 482
 483         /* don't allow direct io onto tail pages */
 484         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 485                 /* make sure future calls to the direct io funcs for this offset
 486                  ** in the file fail by unmapping the buffer
 487                  */
 488                 clear_buffer_mapped(bh_result);
 489                 ret = -EINVAL;
 490         }
 491         /* Possible unpacked tail. Flush the data before pages have
 492            disappeared */
 493         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 494                 int err;
 495
 496                 reiserfs_write_lock(inode->i_sb);
 497
 498                 err = reiserfs_commit_for_inode(inode);
 499                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 500
 501                 reiserfs_write_unlock(inode->i_sb);
 502
 503                 if (err < 0)
 504                         ret = err;
 505         }
 506       out:
 507         return ret;
 508 }
 509
 510 /*
 511 ** helper function for when reiserfs_get_block is called for a hole
 512 ** but the file tail is still in a direct item
 513 ** bh_result is the buffer head for the hole
 514 ** tail_offset is the offset of the start of the tail in the file
 515 **
 516 ** This calls prepare_write, which will start a new transaction
 517 ** you should not be in a transaction, or have any paths held when you
 518 ** call this.
 519 */
 520 static int convert_tail_for_hole(struct inode *inode,
 521                                  struct buffer_head *bh_result,
 522                                  loff_t tail_offset)
 523 {
 524         unsigned long index;
 525         unsigned long tail_end;
 526         unsigned long tail_start;
 527         struct page *tail_page;
 528         struct page *hole_page = bh_result->b_page;
 529         int retval = 0;
 530
 531         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 532                 return -EIO;
 533
 534         /* always try to read until the end of the block */
 535         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
 536         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 537
 538         index = tail_offset >> PAGE_CACHE_SHIFT;
 539         /* hole_page can be zero in case of direct_io, we are sure
 540            that we cannot get here if we write with O_DIRECT into
 541            tail page */
 542         if (!hole_page || index != hole_page->index) {
 543                 tail_page = grab_cache_page(inode->i_mapping, index);
 544                 retval = -ENOMEM;
 545                 if (!tail_page) {
 546                         goto out;
 547                 }
 548         } else {
 549                 tail_page = hole_page;
 550         }
 551
 552         /* we don't have to make sure the conversion did not happen while
 553          ** we were locking the page because anyone that could convert
 554          ** must first take i_mutex.
 555          **
 556          ** We must fix the tail page for writing because it might have buffers
 557          ** that are mapped, but have a block number of 0.  This indicates tail
 558          ** data that has been read directly into the page, and
 559          ** __block_write_begin won't trigger a get_block in this case.
 560          */
 561         fix_tail_page_for_writing(tail_page);
 562         retval = __reiserfs_write_begin(tail_page, tail_start,
 563                                       tail_end - tail_start);
 564         if (retval)
 565                 goto unlock;
 566
 567         /* tail conversion might change the data in the page */
 568         flush_dcache_page(tail_page);
 569
 570         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 571
 572       unlock:
 573         if (tail_page != hole_page) {
 574                 unlock_page(tail_page);
 575                 page_cache_release(tail_page);
 576         }
 577       out:
 578         return retval;
 579 }
 580
 581 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 582                                   sector_t block,
 583                                   struct inode *inode,
 584                                   b_blocknr_t * allocated_block_nr,
 585                                   struct treepath *path, int flags)
 586 {
 587         BUG_ON(!th->t_trans_id);
 588
 589 #ifdef REISERFS_PREALLOCATE
 590         if (!(flags & GET_BLOCK_NO_IMUX)) {
 591                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 592                                                   path, block);
 593         }
 594 #endif
 595         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 596                                          block);
 597 }
 598
 599 int reiserfs_get_block(struct inode *inode, sector_t block,
 600                        struct buffer_head *bh_result, int create)
 601 {
 602         int repeat, retval = 0;
 603         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
 604         INITIALIZE_PATH(path);
 605         int pos_in_item;
 606         struct cpu_key key;
 607         struct buffer_head *bh, *unbh = NULL;
 608         struct item_head *ih, tmp_ih;
 609         __le32 *item;
 610         int done;
 611         int fs_gen;
 612         int lock_depth;
 613         struct reiserfs_transaction_handle *th = NULL;
 614         /* space reserved in transaction batch:
 615            . 3 balancings in direct->indirect conversion
 616            . 1 block involved into reiserfs_update_sd()
 617            XXX in practically impossible worst case direct2indirect()
 618            can incur (much) more than 3 balancings.
 619            quota update for user, group */
 620         int jbegin_count =
 621             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 622             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 623         int version;
 624         int dangle = 1;
 625         loff_t new_offset =
 626             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 627
 628         lock_depth = reiserfs_write_lock_once(inode->i_sb);
 629         version = get_inode_item_key_version(inode);
 630
 631         if (!file_capable(inode, block)) {
 632                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 633                 return -EFBIG;
 634         }
 635
 636         /* if !create, we aren't changing the FS, so we don't need to
 637          ** log anything, so we don't need to start a transaction
 638          */
 639         if (!(create & GET_BLOCK_CREATE)) {
 640                 int ret;
 641                 /* find number of block-th logical block of the file */
 642                 ret = _get_block_create_0(inode, block, bh_result,
 643                                           create | GET_BLOCK_READ_DIRECT);
 644                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 645                 return ret;
 646         }
 647         /*
 648          * if we're already in a transaction, make sure to close
 649          * any new transactions we start in this func
 650          */
 651         if ((create & GET_BLOCK_NO_DANGLE) ||
 652             reiserfs_transaction_running(inode->i_sb))
 653                 dangle = 0;
 654
 655         /* If file is of such a size, that it might have a tail and tails are enabled
 656          ** we should mark it as possibly needing tail packing on close
 657          */
 658         if ((have_large_tails(inode->i_sb)
 659              && inode->i_size < i_block_size(inode) * 4)
 660             || (have_small_tails(inode->i_sb)
 661                 && inode->i_size < i_block_size(inode)))
 662                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 663
 664         /* set the key of the first byte in the 'block'-th block of file */
 665         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 666         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 667               start_trans:
 668                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 669                 if (!th) {
 670                         retval = -ENOMEM;
 671                         goto failure;
 672                 }
 673                 reiserfs_update_inode_transaction(inode);
 674         }
 675       research:
 676
 677         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 678         if (retval == IO_ERROR) {
 679                 retval = -EIO;
 680                 goto failure;
 681         }
 682
 683         bh = get_last_bh(&path);
 684         ih = get_ih(&path);
 685         item = get_item(&path);
 686         pos_in_item = path.pos_in_item;
 687
 688         fs_gen = get_generation(inode->i_sb);
 689         copy_item_head(&tmp_ih, ih);
 690
 691         if (allocation_needed
 692             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 693                 /* we have to allocate block for the unformatted node */
 694                 if (!th) {
 695                         pathrelse(&path);
 696                         goto start_trans;
 697                 }
 698
 699                 repeat =
 700                     _allocate_block(th, block, inode, &allocated_block_nr,
 701                                     &path, create);
 702
 703                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 704                         /* restart the transaction to give the journal a chance to free
 705                          ** some blocks.  releases the path, so we have to go back to
 706                          ** research if we succeed on the second try
 707                          */
 708                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 709                         retval = restart_transaction(th, inode, &path);
 710                         if (retval)
 711                                 goto failure;
 712                         repeat =
 713                             _allocate_block(th, block, inode,
 714                                             &allocated_block_nr, NULL, create);
 715
 716                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 717                                 goto research;
 718                         }
 719                         if (repeat == QUOTA_EXCEEDED)
 720                                 retval = -EDQUOT;
 721                         else
 722                                 retval = -ENOSPC;
 723                         goto failure;
 724                 }
 725
 726                 if (fs_changed(fs_gen, inode->i_sb)
 727                     && item_moved(&tmp_ih, &path)) {
 728                         goto research;
 729                 }
 730         }
 731
 732         if (indirect_item_found(retval, ih)) {
 733                 b_blocknr_t unfm_ptr;
 734                 /* 'block'-th block is in the file already (there is
 735                    corresponding cell in some indirect item). But it may be
 736                    zero unformatted node pointer (hole) */
 737                 unfm_ptr = get_block_num(item, pos_in_item);
 738                 if (unfm_ptr == 0) {
 739                         /* use allocated block to plug the hole */
 740                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 741                         if (fs_changed(fs_gen, inode->i_sb)
 742                             && item_moved(&tmp_ih, &path)) {
 743                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 744                                                                  bh);
 745                                 goto research;
 746                         }
 747                         set_buffer_new(bh_result);
 748                         if (buffer_dirty(bh_result)
 749                             && reiserfs_data_ordered(inode->i_sb))
 750                                 reiserfs_add_ordered_list(inode, bh_result);
 751                         put_block_num(item, pos_in_item, allocated_block_nr);
 752                         unfm_ptr = allocated_block_nr;
 753                         journal_mark_dirty(th, inode->i_sb, bh);
 754                         reiserfs_update_sd(th, inode);
 755                 }
 756                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 757                 pathrelse(&path);
 758                 retval = 0;
 759                 if (!dangle && th)
 760                         retval = reiserfs_end_persistent_transaction(th);
 761
 762                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
 763
 764                 /* the item was found, so new blocks were not added to the file
 765                  ** there is no need to make sure the inode is updated with this
 766                  ** transaction
 767                  */
 768                 return retval;
 769         }
 770
 771         if (!th) {
 772                 pathrelse(&path);
 773                 goto start_trans;
 774         }
 775
 776         /* desired position is not found or is in the direct item. We have
 777            to append file with holes up to 'block'-th block converting
 778            direct items to indirect one if necessary */
 779         done = 0;
 780         do {
 781                 if (is_statdata_le_ih(ih)) {
 782                         __le32 unp = 0;
 783                         struct cpu_key tmp_key;
 784
 785                         /* indirect item has to be inserted */
 786                         make_le_item_head(&tmp_ih, &key, version, 1,
 787                                           TYPE_INDIRECT, UNFM_P_SIZE,
 788                                           0 /* free_space */ );
 789
 790                         if (cpu_key_k_offset(&key) == 1) {
 791                                 /* we are going to add 'block'-th block to the file. Use
 792                                    allocated block for that */
 793                                 unp = cpu_to_le32(allocated_block_nr);
 794                                 set_block_dev_mapped(bh_result,
 795                                                      allocated_block_nr, inode);
 796                                 set_buffer_new(bh_result);
 797                                 done = 1;
 798                         }
 799                         tmp_key = key;  // ;)
 800                         set_cpu_key_k_offset(&tmp_key, 1);
 801                         PATH_LAST_POSITION(&path)++;
 802
 803                         retval =
 804                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 805                                                  inode, (char *)&unp);
 806                         if (retval) {
 807                                 reiserfs_free_block(th, inode,
 808                                                     allocated_block_nr, 1);
 809                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
 810                         }
 811                         //mark_tail_converted (inode);
 812                 } else if (is_direct_le_ih(ih)) {
 813                         /* direct item has to be converted */
 814                         loff_t tail_offset;
 815
 816                         tail_offset =
 817                             ((le_ih_k_offset(ih) -
 818                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 819                         if (tail_offset == cpu_key_k_offset(&key)) {
 820                                 /* direct item we just found fits into block we have
 821                                    to map. Convert it into unformatted node: use
 822                                    bh_result for the conversion */
 823                                 set_block_dev_mapped(bh_result,
 824                                                      allocated_block_nr, inode);
 825                                 unbh = bh_result;
 826                                 done = 1;
 827                         } else {
 828                                 /* we have to padd file tail stored in direct item(s)
 829                                    up to block size and convert it to unformatted
 830                                    node. FIXME: this should also get into page cache */
 831
 832                                 pathrelse(&path);
 833                                 /*
 834                                  * ugly, but we can only end the transaction if
 835                                  * we aren't nested
 836                                  */
 837                                 BUG_ON(!th->t_refcount);
 838                                 if (th->t_refcount == 1) {
 839                                         retval =
 840                                             reiserfs_end_persistent_transaction
 841                                             (th);
 842                                         th = NULL;
 843                                         if (retval)
 844                                                 goto failure;
 845                                 }
 846
 847                                 retval =
 848                                     convert_tail_for_hole(inode, bh_result,
 849                                                           tail_offset);
 850                                 if (retval) {
 851                                         if (retval != -ENOSPC)
 852                                                 reiserfs_error(inode->i_sb,
 853                                                         "clm-6004",
 854                                                         "convert tail failed "
 855                                                         "inode %lu, error %d",
 856                                                         inode->i_ino,
 857                                                         retval);
 858                                         if (allocated_block_nr) {
 859                                                 /* the bitmap, the super, and the stat data == 3 */
 860                                                 if (!th)
 861                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 862                                                 if (th)
 863                                                         reiserfs_free_block(th,
 864                                                                             inode,
 865                                                                             allocated_block_nr,
 866                                                                             1);
 867                                         }
 868                                         goto failure;
 869                                 }
 870                                 goto research;
 871                         }
 872                         retval =
 873                             direct2indirect(th, inode, &path, unbh,
 874                                             tail_offset);
 875                         if (retval) {
 876                                 reiserfs_unmap_buffer(unbh);
 877                                 reiserfs_free_block(th, inode,
 878                                                     allocated_block_nr, 1);
 879                                 goto failure;
 880                         }
 881                         /* it is important the set_buffer_uptodate is done after
 882                          ** the direct2indirect.  The buffer might contain valid
 883                          ** data newer than the data on disk (read by readpage, changed,
 884                          ** and then sent here by writepage).  direct2indirect needs
 885                          ** to know if unbh was already up to date, so it can decide
 886                          ** if the data in unbh needs to be replaced with data from
 887                          ** the disk
 888                          */
 889                         set_buffer_uptodate(unbh);
 890
 891                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
 892                            buffer will disappear shortly, so it should not be added to
 893                          */
 894                         if (unbh->b_page) {
 895                                 /* we've converted the tail, so we must
 896                                  ** flush unbh before the transaction commits
 897                                  */
 898                                 reiserfs_add_tail_list(inode, unbh);
 899
 900                                 /* mark it dirty now to prevent commit_write from adding
 901                                  ** this buffer to the inode's dirty buffer list
 902                                  */
 903                                 /*
 904                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 905                                  * It's still atomic, but it sets the page dirty too,
 906                                  * which makes it eligible for writeback at any time by the
 907                                  * VM (which was also the case with __mark_buffer_dirty())
 908                                  */
 909                                 mark_buffer_dirty(unbh);
 910                         }
 911                 } else {
 912                         /* append indirect item with holes if needed, when appending
 913                            pointer to 'block'-th block use block, which is already
 914                            allocated */
 915                         struct cpu_key tmp_key;
 916                         unp_t unf_single = 0;   // We use this in case we need to allocate only
 917                         // one block which is a fastpath
 918                         unp_t *un;
 919                         __u64 max_to_insert =
 920                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
 921                             UNFM_P_SIZE;
 922                         __u64 blocks_needed;
 923
 924                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
 925                                "vs-804: invalid position for append");
 926                         /* indirect item has to be appended, set up key of that position */
 927                         make_cpu_key(&tmp_key, inode,
 928                                      le_key_k_offset(version,
 929                                                      &(ih->ih_key)) +
 930                                      op_bytes_number(ih,
 931                                                      inode->i_sb->s_blocksize),
 932                                      //pos_in_item * inode->i_sb->s_blocksize,
 933                                      TYPE_INDIRECT, 3); // key type is unimportant
 934
 935                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
 936                                "green-805: invalid offset");
 937                         blocks_needed =
 938                             1 +
 939                             ((cpu_key_k_offset(&key) -
 940                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
 941                              s_blocksize_bits);
 942
 943                         if (blocks_needed == 1) {
 944                                 un = &unf_single;
 945                         } else {
 946                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
 947                                 if (!un) {
 948                                         un = &unf_single;
 949                                         blocks_needed = 1;
 950                                         max_to_insert = 0;
 951                                 }
 952                         }
 953                         if (blocks_needed <= max_to_insert) {
 954                                 /* we are going to add target block to the file. Use allocated
 955                                    block for that */
 956                                 un[blocks_needed - 1] =
 957                                     cpu_to_le32(allocated_block_nr);
 958                                 set_block_dev_mapped(bh_result,
 959                                                      allocated_block_nr, inode);
 960                                 set_buffer_new(bh_result);
 961                                 done = 1;
 962                         } else {
 963                                 /* paste hole to the indirect item */
 964                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
 965                                    only have space for one block */
 966                                 blocks_needed =
 967                                     max_to_insert ? max_to_insert : 1;
 968                         }
 969                         retval =
 970                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
 971                                                      (char *)un,
 972                                                      UNFM_P_SIZE *
 973                                                      blocks_needed);
 974
 975                         if (blocks_needed != 1)
 976                                 kfree(un);
 977
 978                         if (retval) {
 979                                 reiserfs_free_block(th, inode,
 980                                                     allocated_block_nr, 1);
 981                                 goto failure;
 982                         }
 983                         if (!done) {
 984                                 /* We need to mark new file size in case this function will be
 985                                    interrupted/aborted later on. And we may do this only for
 986                                    holes. */
 987                                 inode->i_size +=
 988                                     inode->i_sb->s_blocksize * blocks_needed;
 989                         }
 990                 }
 991
 992                 if (done == 1)
 993                         break;
 994
 995                 /* this loop could log more blocks than we had originally asked
 996                  ** for.  So, we have to allow the transaction to end if it is
 997                  ** too big or too full.  Update the inode so things are
 998                  ** consistent if we crash before the function returns
 999                  **
1000                  ** release the path so that anybody waiting on the path before
1001                  ** ending their transaction will be able to continue.
1002                  */
1003                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1004                         retval = restart_transaction(th, inode, &path);
1005                         if (retval)
1006                                 goto failure;
1007                 }
1008                 /*
1009                  * inserting indirect pointers for a hole can take a
1010                  * long time.  reschedule if needed and also release the write
1011                  * lock for others.
1012                  */
1013                 if (need_resched()) {
1014                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1015                         schedule();
1016                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1017                 }
1018
1019                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1020                 if (retval == IO_ERROR) {
1021                         retval = -EIO;
1022                         goto failure;
1023                 }
1024                 if (retval == POSITION_FOUND) {
1025                         reiserfs_warning(inode->i_sb, "vs-825",
1026                                          "%K should not be found", &key);
1027                         retval = -EEXIST;
1028                         if (allocated_block_nr)
1029                                 reiserfs_free_block(th, inode,
1030                                                     allocated_block_nr, 1);
1031                         pathrelse(&path);
1032                         goto failure;
1033                 }
1034                 bh = get_last_bh(&path);
1035                 ih = get_ih(&path);
1036                 item = get_item(&path);
1037                 pos_in_item = path.pos_in_item;
1038         } while (1);
1039
1040         retval = 0;
1041
1042       failure:
1043         if (th && (!dangle || (retval && !th->t_trans_id))) {
1044                 int err;
1045                 if (th->t_trans_id)
1046                         reiserfs_update_sd(th, inode);
1047                 err = reiserfs_end_persistent_transaction(th);
1048                 if (err)
1049                         retval = err;
1050         }
1051
1052         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1053         reiserfs_check_path(&path);
1054         return retval;
1055 }
1056
1057 static int
1058 reiserfs_readpages(struct file *file, struct address_space *mapping,
1059                    struct list_head *pages, unsigned nr_pages)
1060 {
1061         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1062 }
1063
1064 /* Compute real number of used bytes by file
1065  * Following three functions can go away when we'll have enough space in stat item
1066  */
1067 static int real_space_diff(struct inode *inode, int sd_size)
1068 {
1069         int bytes;
1070         loff_t blocksize = inode->i_sb->s_blocksize;
1071
1072         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1073                 return sd_size;
1074
1075         /* End of file is also in full block with indirect reference, so round
1076          ** up to the next block.
1077          **
1078          ** there is just no way to know if the tail is actually packed
1079          ** on the file, so we have to assume it isn't.  When we pack the
1080          ** tail, we add 4 bytes to pretend there really is an unformatted
1081          ** node pointer
1082          */
1083         bytes =
1084             ((inode->i_size +
1085               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1086             sd_size;
1087         return bytes;
1088 }
1089
1090 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1091                                         int sd_size)
1092 {
1093         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1094                 return inode->i_size +
1095                     (loff_t) (real_space_diff(inode, sd_size));
1096         }
1097         return ((loff_t) real_space_diff(inode, sd_size)) +
1098             (((loff_t) blocks) << 9);
1099 }
1100
1101 /* Compute number of blocks used by file in ReiserFS counting */
1102 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1103 {
1104         loff_t bytes = inode_get_bytes(inode);
1105         loff_t real_space = real_space_diff(inode, sd_size);
1106
1107         /* keeps fsck and non-quota versions of reiserfs happy */
1108         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1109                 bytes += (loff_t) 511;
1110         }
1111
1112         /* files from before the quota patch might i_blocks such that
1113          ** bytes < real_space.  Deal with that here to prevent it from
1114          ** going negative.
1115          */
1116         if (bytes < real_space)
1117                 return 0;
1118         return (bytes - real_space) >> 9;
1119 }
1120
1121 //
1122 // BAD: new directories have stat data of new type and all other items
1123 // of old type. Version stored in the inode says about body items, so
1124 // in update_stat_data we can not rely on inode, but have to check
1125 // item version directly
1126 //
1127
1128 // called by read_locked_inode
1129 static void init_inode(struct inode *inode, struct treepath *path)
1130 {
1131         struct buffer_head *bh;
1132         struct item_head *ih;
1133         __u32 rdev;
1134         //int version = ITEM_VERSION_1;
1135
1136         bh = PATH_PLAST_BUFFER(path);
1137         ih = PATH_PITEM_HEAD(path);
1138
1139         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1140
1141         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1142         REISERFS_I(inode)->i_flags = 0;
1143         REISERFS_I(inode)->i_prealloc_block = 0;
1144         REISERFS_I(inode)->i_prealloc_count = 0;
1145         REISERFS_I(inode)->i_trans_id = 0;
1146         REISERFS_I(inode)->i_jl = NULL;
1147         reiserfs_init_xattr_rwsem(inode);
1148
1149         if (stat_data_v1(ih)) {
1150                 struct stat_data_v1 *sd =
1151                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1152                 unsigned long blocks;
1153
1154                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1155                 set_inode_sd_version(inode, STAT_DATA_V1);
1156                 inode->i_mode = sd_v1_mode(sd);
1157                 set_nlink(inode, sd_v1_nlink(sd));
1158                 inode->i_uid = sd_v1_uid(sd);
1159                 inode->i_gid = sd_v1_gid(sd);
1160                 inode->i_size = sd_v1_size(sd);
1161                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1162                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1163                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1164                 inode->i_atime.tv_nsec = 0;
1165                 inode->i_ctime.tv_nsec = 0;
1166                 inode->i_mtime.tv_nsec = 0;
1167
1168                 inode->i_blocks = sd_v1_blocks(sd);
1169                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1170                 blocks = (inode->i_size + 511) >> 9;
1171                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1172                 if (inode->i_blocks > blocks) {
1173                         // there was a bug in <=3.5.23 when i_blocks could take negative
1174                         // values. Starting from 3.5.17 this value could even be stored in
1175                         // stat data. For such files we set i_blocks based on file
1176                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1177                         // only updated if file's inode will ever change
1178                         inode->i_blocks = blocks;
1179                 }
1180
1181                 rdev = sd_v1_rdev(sd);
1182                 REISERFS_I(inode)->i_first_direct_byte =
1183                     sd_v1_first_direct_byte(sd);
1184                 /* an early bug in the quota code can give us an odd number for the
1185                  ** block count.  This is incorrect, fix it here.
1186                  */
1187                 if (inode->i_blocks & 1) {
1188                         inode->i_blocks++;
1189                 }
1190                 inode_set_bytes(inode,
1191                                 to_real_used_space(inode, inode->i_blocks,
1192                                                    SD_V1_SIZE));
1193                 /* nopack is initially zero for v1 objects. For v2 objects,
1194                    nopack is initialised from sd_attrs */
1195                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1196         } else {
1197                 // new stat data found, but object may have old items
1198                 // (directories and symlinks)
1199                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1200
1201                 inode->i_mode = sd_v2_mode(sd);
1202                 set_nlink(inode, sd_v2_nlink(sd));
1203                 inode->i_uid = sd_v2_uid(sd);
1204                 inode->i_size = sd_v2_size(sd);
1205                 inode->i_gid = sd_v2_gid(sd);
1206                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1207                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1208                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1209                 inode->i_ctime.tv_nsec = 0;
1210                 inode->i_mtime.tv_nsec = 0;
1211                 inode->i_atime.tv_nsec = 0;
1212                 inode->i_blocks = sd_v2_blocks(sd);
1213                 rdev = sd_v2_rdev(sd);
1214                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1215                         inode->i_generation =
1216                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1217                 else
1218                         inode->i_generation = sd_v2_generation(sd);
1219
1220                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1221                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1222                 else
1223                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1224                 REISERFS_I(inode)->i_first_direct_byte = 0;
1225                 set_inode_sd_version(inode, STAT_DATA_V2);
1226                 inode_set_bytes(inode,
1227                                 to_real_used_space(inode, inode->i_blocks,
1228                                                    SD_V2_SIZE));
1229                 /* read persistent inode attributes from sd and initialise
1230                    generic inode flags from them */
1231                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1232                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1233         }
1234
1235         pathrelse(path);
1236         if (S_ISREG(inode->i_mode)) {
1237                 inode->i_op = &reiserfs_file_inode_operations;
1238                 inode->i_fop = &reiserfs_file_operations;
1239                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1240         } else if (S_ISDIR(inode->i_mode)) {
1241                 inode->i_op = &reiserfs_dir_inode_operations;
1242                 inode->i_fop = &reiserfs_dir_operations;
1243         } else if (S_ISLNK(inode->i_mode)) {
1244                 inode->i_op = &reiserfs_symlink_inode_operations;
1245                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1246         } else {
1247                 inode->i_blocks = 0;
1248                 inode->i_op = &reiserfs_special_inode_operations;
1249                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1250         }
1251 }
1252
1253 // update new stat data with inode fields
1254 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1255 {
1256         struct stat_data *sd_v2 = (struct stat_data *)sd;
1257         __u16 flags;
1258
1259         set_sd_v2_mode(sd_v2, inode->i_mode);
1260         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1261         set_sd_v2_uid(sd_v2, inode->i_uid);
1262         set_sd_v2_size(sd_v2, size);
1263         set_sd_v2_gid(sd_v2, inode->i_gid);
1264         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1265         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1266         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1267         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1268         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1269                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1270         else
1271                 set_sd_v2_generation(sd_v2, inode->i_generation);
1272         flags = REISERFS_I(inode)->i_attrs;
1273         i_attrs_to_sd_attrs(inode, &flags);
1274         set_sd_v2_attrs(sd_v2, flags);
1275 }
1276
1277 // used to copy inode's fields to old stat data
1278 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1279 {
1280         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1281
1282         set_sd_v1_mode(sd_v1, inode->i_mode);
1283         set_sd_v1_uid(sd_v1, inode->i_uid);
1284         set_sd_v1_gid(sd_v1, inode->i_gid);
1285         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1286         set_sd_v1_size(sd_v1, size);
1287         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1288         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1289         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1290
1291         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1292                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1293         else
1294                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1295
1296         // Sigh. i_first_direct_byte is back
1297         set_sd_v1_first_direct_byte(sd_v1,
1298                                     REISERFS_I(inode)->i_first_direct_byte);
1299 }
1300
1301 /* NOTE, you must prepare the buffer head before sending it here,
1302 ** and then log it after the call
1303 */
1304 static void update_stat_data(struct treepath *path, struct inode *inode,
1305                              loff_t size)
1306 {
1307         struct buffer_head *bh;
1308         struct item_head *ih;
1309
1310         bh = PATH_PLAST_BUFFER(path);
1311         ih = PATH_PITEM_HEAD(path);
1312
1313         if (!is_statdata_le_ih(ih))
1314                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1315                                INODE_PKEY(inode), ih);
1316
1317         if (stat_data_v1(ih)) {
1318                 // path points to old stat data
1319                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1320         } else {
1321                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1322         }
1323
1324         return;
1325 }
1326
1327 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1328                              struct inode *inode, loff_t size)
1329 {
1330         struct cpu_key key;
1331         INITIALIZE_PATH(path);
1332         struct buffer_head *bh;
1333         int fs_gen;
1334         struct item_head *ih, tmp_ih;
1335         int retval;
1336
1337         BUG_ON(!th->t_trans_id);
1338
1339         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1340
1341         for (;;) {
1342                 int pos;
1343                 /* look for the object's stat data */
1344                 retval = search_item(inode->i_sb, &key, &path);
1345                 if (retval == IO_ERROR) {
1346                         reiserfs_error(inode->i_sb, "vs-13050",
1347                                        "i/o failure occurred trying to "
1348                                        "update %K stat data", &key);
1349                         return;
1350                 }
1351                 if (retval == ITEM_NOT_FOUND) {
1352                         pos = PATH_LAST_POSITION(&path);
1353                         pathrelse(&path);
1354                         if (inode->i_nlink == 0) {
1355                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1356                                 return;
1357                         }
1358                         reiserfs_warning(inode->i_sb, "vs-13060",
1359                                          "stat data of object %k (nlink == %d) "
1360                                          "not found (pos %d)",
1361                                          INODE_PKEY(inode), inode->i_nlink,
1362                                          pos);
1363                         reiserfs_check_path(&path);
1364                         return;
1365                 }
1366
1367                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1368                  ** FS might change.  We have to detect that, and loop back to the
1369                  ** search if the stat data item has moved
1370                  */
1371                 bh = get_last_bh(&path);
1372                 ih = get_ih(&path);
1373                 copy_item_head(&tmp_ih, ih);
1374                 fs_gen = get_generation(inode->i_sb);
1375                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1376                 if (fs_changed(fs_gen, inode->i_sb)
1377                     && item_moved(&tmp_ih, &path)) {
1378                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1379                         continue;       /* Stat_data item has been moved after scheduling. */
1380                 }
1381                 break;
1382         }
1383         update_stat_data(&path, inode, size);
1384         journal_mark_dirty(th, th->t_super, bh);
1385         pathrelse(&path);
1386         return;
1387 }
1388
1389 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1390 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1391 ** and clear the key in the private portion of the inode, otherwise a
1392 ** corresponding iput might try to delete whatever object the inode last
1393 ** represented.
1394 */
1395 static void reiserfs_make_bad_inode(struct inode *inode)
1396 {
1397         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1398         make_bad_inode(inode);
1399 }
1400
1401 //
1402 // initially this function was derived from minix or ext2's analog and
1403 // evolved as the prototype did
1404 //
1405
1406 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1407 {
1408         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1409         inode->i_ino = args->objectid;
1410         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1411         return 0;
1412 }
1413
1414 /* looks for stat data in the tree, and fills up the fields of in-core
1415    inode stat data fields */
1416 void reiserfs_read_locked_inode(struct inode *inode,
1417                                 struct reiserfs_iget_args *args)
1418 {
1419         INITIALIZE_PATH(path_to_sd);
1420         struct cpu_key key;
1421         unsigned long dirino;
1422         int retval;
1423
1424         dirino = args->dirid;
1425
1426         /* set version 1, version 2 could be used too, because stat data
1427            key is the same in both versions */
1428         key.version = KEY_FORMAT_3_5;
1429         key.on_disk_key.k_dir_id = dirino;
1430         key.on_disk_key.k_objectid = inode->i_ino;
1431         key.on_disk_key.k_offset = 0;
1432         key.on_disk_key.k_type = 0;
1433
1434         /* look for the object's stat data */
1435         retval = search_item(inode->i_sb, &key, &path_to_sd);
1436         if (retval == IO_ERROR) {
1437                 reiserfs_error(inode->i_sb, "vs-13070",
1438                                "i/o failure occurred trying to find "
1439                                "stat data of %K", &key);
1440                 reiserfs_make_bad_inode(inode);
1441                 return;
1442         }
1443         if (retval != ITEM_FOUND) {
1444                 /* a stale NFS handle can trigger this without it being an error */
1445                 pathrelse(&path_to_sd);
1446                 reiserfs_make_bad_inode(inode);
1447                 clear_nlink(inode);
1448                 return;
1449         }
1450
1451         init_inode(inode, &path_to_sd);
1452
1453         /* It is possible that knfsd is trying to access inode of a file
1454            that is being removed from the disk by some other thread. As we
1455            update sd on unlink all that is required is to check for nlink
1456            here. This bug was first found by Sizif when debugging
1457            SquidNG/Butterfly, forgotten, and found again after Philippe
1458            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1459
1460            More logical fix would require changes in fs/inode.c:iput() to
1461            remove inode from hash-table _after_ fs cleaned disk stuff up and
1462            in iget() to return NULL if I_FREEING inode is found in
1463            hash-table. */
1464         /* Currently there is one place where it's ok to meet inode with
1465            nlink==0: processing of open-unlinked and half-truncated files
1466            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1467         if ((inode->i_nlink == 0) &&
1468             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1469                 reiserfs_warning(inode->i_sb, "vs-13075",
1470                                  "dead inode read from disk %K. "
1471                                  "This is likely to be race with knfsd. Ignore",
1472                                  &key);
1473                 reiserfs_make_bad_inode(inode);
1474         }
1475
1476         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1477
1478         /*
1479          * Stat data v1 doesn't support ACLs.
1480          */
1481         if (get_inode_sd_version(inode) == STAT_DATA_V1)
1482                 cache_no_acl(inode);
1483 }
1484
1485 /**
1486  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1487  *
1488  * @inode:    inode from hash table to check
1489  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1490  *
1491  * This function is called by iget5_locked() to distinguish reiserfs inodes
1492  * having the same inode numbers. Such inodes can only exist due to some
1493  * error condition. One of them should be bad. Inodes with identical
1494  * inode numbers (objectids) are distinguished by parent directory ids.
1495  *
1496  */
1497 int reiserfs_find_actor(struct inode *inode, void *opaque)
1498 {
1499         struct reiserfs_iget_args *args;
1500
1501         args = opaque;
1502         /* args is already in CPU order */
1503         return (inode->i_ino == args->objectid) &&
1504             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1505 }
1506
1507 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1508 {
1509         struct inode *inode;
1510         struct reiserfs_iget_args args;
1511
1512         args.objectid = key->on_disk_key.k_objectid;
1513         args.dirid = key->on_disk_key.k_dir_id;
1514         reiserfs_write_unlock(s);
1515         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1516                              reiserfs_find_actor, reiserfs_init_locked_inode,
1517                              (void *)(&args));
1518         reiserfs_write_lock(s);
1519         if (!inode)
1520                 return ERR_PTR(-ENOMEM);
1521
1522         if (inode->i_state & I_NEW) {
1523                 reiserfs_read_locked_inode(inode, &args);
1524                 unlock_new_inode(inode);
1525         }
1526
1527         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1528                 /* either due to i/o error or a stale NFS handle */
1529                 iput(inode);
1530                 inode = NULL;
1531         }
1532         return inode;
1533 }
1534
1535 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1536         u32 objectid, u32 dir_id, u32 generation)
1537
1538 {
1539         struct cpu_key key;
1540         struct inode *inode;
1541
1542         key.on_disk_key.k_objectid = objectid;
1543         key.on_disk_key.k_dir_id = dir_id;
1544         reiserfs_write_lock(sb);
1545         inode = reiserfs_iget(sb, &key);
1546         if (inode && !IS_ERR(inode) && generation != 0 &&
1547             generation != inode->i_generation) {
1548                 iput(inode);
1549                 inode = NULL;
1550         }
1551         reiserfs_write_unlock(sb);
1552
1553         return d_obtain_alias(inode);
1554 }
1555
1556 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1557                 int fh_len, int fh_type)
1558 {
1559         /* fhtype happens to reflect the number of u32s encoded.
1560          * due to a bug in earlier code, fhtype might indicate there
1561          * are more u32s then actually fitted.
1562          * so if fhtype seems to be more than len, reduce fhtype.
1563          * Valid types are:
1564          *   2 - objectid + dir_id - legacy support
1565          *   3 - objectid + dir_id + generation
1566          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1567          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1568          *   6 - as above plus generation of directory
1569          * 6 does not fit in NFSv2 handles
1570          */
1571         if (fh_type > fh_len) {
1572                 if (fh_type != 6 || fh_len != 5)
1573                         reiserfs_warning(sb, "reiserfs-13077",
1574                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1575                                 fh_type, fh_len);
1576                 fh_type = 5;
1577         }
1578
1579         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1580                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1581 }
1582
1583 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1584                 int fh_len, int fh_type)
1585 {
1586         if (fh_type < 4)
1587                 return NULL;
1588
1589         return reiserfs_get_dentry(sb,
1590                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1591                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1592                 (fh_type == 6) ? fid->raw[5] : 0);
1593 }
1594
1595 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1596                        int need_parent)
1597 {
1598         struct inode *inode = dentry->d_inode;
1599         int maxlen = *lenp;
1600
1601         if (need_parent && (maxlen < 5)) {
1602                 *lenp = 5;
1603                 return 255;
1604         } else if (maxlen < 3) {
1605                 *lenp = 3;
1606                 return 255;
1607         }
1608
1609         data[0] = inode->i_ino;
1610         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1611         data[2] = inode->i_generation;
1612         *lenp = 3;
1613         /* no room for directory info? return what we've stored so far */
1614         if (maxlen < 5 || !need_parent)
1615                 return 3;
1616
1617         spin_lock(&dentry->d_lock);
1618         inode = dentry->d_parent->d_inode;
1619         data[3] = inode->i_ino;
1620         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1621         *lenp = 5;
1622         if (maxlen >= 6) {
1623                 data[5] = inode->i_generation;
1624                 *lenp = 6;
1625         }
1626         spin_unlock(&dentry->d_lock);
1627         return *lenp;
1628 }
1629
1630 /* looks for stat data, then copies fields to it, marks the buffer
1631    containing stat data as dirty */
1632 /* reiserfs inodes are never really dirty, since the dirty inode call
1633 ** always logs them.  This call allows the VFS inode marking routines
1634 ** to properly mark inodes for datasync and such, but only actually
1635 ** does something when called for a synchronous update.
1636 */
1637 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1638 {
1639         struct reiserfs_transaction_handle th;
1640         int jbegin_count = 1;
1641
1642         if (inode->i_sb->s_flags & MS_RDONLY)
1643                 return -EROFS;
1644         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1645          ** these cases are just when the system needs ram, not when the
1646          ** inode needs to reach disk for safety, and they can safely be
1647          ** ignored because the altered inode has already been logged.
1648          */
1649         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1650                 reiserfs_write_lock(inode->i_sb);
1651                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1652                         reiserfs_update_sd(&th, inode);
1653                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1654                 }
1655                 reiserfs_write_unlock(inode->i_sb);
1656         }
1657         return 0;
1658 }
1659
1660 /* stat data of new object is inserted already, this inserts the item
1661    containing "." and ".." entries */
1662 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1663                                   struct inode *inode,
1664                                   struct item_head *ih, struct treepath *path,
1665                                   struct inode *dir)
1666 {
1667         struct super_block *sb = th->t_super;
1668         char empty_dir[EMPTY_DIR_SIZE];
1669         char *body = empty_dir;
1670         struct cpu_key key;
1671         int retval;
1672
1673         BUG_ON(!th->t_trans_id);
1674
1675         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1676                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1677                       TYPE_DIRENTRY, 3 /*key length */ );
1678
1679         /* compose item head for new item. Directories consist of items of
1680            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1681            is done by reiserfs_new_inode */
1682         if (old_format_only(sb)) {
1683                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1684                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1685
1686                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1687                                        ih->ih_key.k_objectid,
1688                                        INODE_PKEY(dir)->k_dir_id,
1689                                        INODE_PKEY(dir)->k_objectid);
1690         } else {
1691                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1692                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1693
1694                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1695                                     ih->ih_key.k_objectid,
1696                                     INODE_PKEY(dir)->k_dir_id,
1697                                     INODE_PKEY(dir)->k_objectid);
1698         }
1699
1700         /* look for place in the tree for new item */
1701         retval = search_item(sb, &key, path);
1702         if (retval == IO_ERROR) {
1703                 reiserfs_error(sb, "vs-13080",
1704                                "i/o failure occurred creating new directory");
1705                 return -EIO;
1706         }
1707         if (retval == ITEM_FOUND) {
1708                 pathrelse(path);
1709                 reiserfs_warning(sb, "vs-13070",
1710                                  "object with this key exists (%k)",
1711                                  &(ih->ih_key));
1712                 return -EEXIST;
1713         }
1714
1715         /* insert item, that is empty directory item */
1716         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1717 }
1718
1719 /* stat data of object has been inserted, this inserts the item
1720    containing the body of symlink */
1721 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1722                                 struct item_head *ih,
1723                                 struct treepath *path, const char *symname,
1724                                 int item_len)
1725 {
1726         struct super_block *sb = th->t_super;
1727         struct cpu_key key;
1728         int retval;
1729
1730         BUG_ON(!th->t_trans_id);
1731
1732         _make_cpu_key(&key, KEY_FORMAT_3_5,
1733                       le32_to_cpu(ih->ih_key.k_dir_id),
1734                       le32_to_cpu(ih->ih_key.k_objectid),
1735                       1, TYPE_DIRECT, 3 /*key length */ );
1736
1737         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1738                           0 /*free_space */ );
1739
1740         /* look for place in the tree for new item */
1741         retval = search_item(sb, &key, path);
1742         if (retval == IO_ERROR) {
1743                 reiserfs_error(sb, "vs-13080",
1744                                "i/o failure occurred creating new symlink");
1745                 return -EIO;
1746         }
1747         if (retval == ITEM_FOUND) {
1748                 pathrelse(path);
1749                 reiserfs_warning(sb, "vs-13080",
1750                                  "object with this key exists (%k)",
1751                                  &(ih->ih_key));
1752                 return -EEXIST;
1753         }
1754
1755         /* insert item, that is body of symlink */
1756         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1757 }
1758
1759 /* inserts the stat data into the tree, and then calls
1760    reiserfs_new_directory (to insert ".", ".." item if new object is
1761    directory) or reiserfs_new_symlink (to insert symlink body if new
1762    object is symlink) or nothing (if new object is regular file)
1763
1764    NOTE! uid and gid must already be set in the inode.  If we return
1765    non-zero due to an error, we have to drop the quota previously allocated
1766    for the fresh inode.  This can only be done outside a transaction, so
1767    if we return non-zero, we also end the transaction.  */
1768 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1769                        struct inode *dir, int mode, const char *symname,
1770                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1771                           strlen (symname) for symlinks) */
1772                        loff_t i_size, struct dentry *dentry,
1773                        struct inode *inode,
1774                        struct reiserfs_security_handle *security)
1775 {
1776         struct super_block *sb;
1777         struct reiserfs_iget_args args;
1778         INITIALIZE_PATH(path_to_key);
1779         struct cpu_key key;
1780         struct item_head ih;
1781         struct stat_data sd;
1782         int retval;
1783         int err;
1784
1785         BUG_ON(!th->t_trans_id);
1786
1787         dquot_initialize(inode);
1788         err = dquot_alloc_inode(inode);
1789         if (err)
1790                 goto out_end_trans;
1791         if (!dir->i_nlink) {
1792                 err = -EPERM;
1793                 goto out_bad_inode;
1794         }
1795
1796         sb = dir->i_sb;
1797
1798         /* item head of new item */
1799         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1800         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1801         if (!ih.ih_key.k_objectid) {
1802                 err = -ENOMEM;
1803                 goto out_bad_inode;
1804         }
1805         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1806         if (old_format_only(sb))
1807                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1808                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1809         else
1810                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1811                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1812         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1813         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1814         if (insert_inode_locked4(inode, args.objectid,
1815                              reiserfs_find_actor, &args) < 0) {
1816                 err = -EINVAL;
1817                 goto out_bad_inode;
1818         }
1819         if (old_format_only(sb))
1820                 /* not a perfect generation count, as object ids can be reused, but
1821                  ** this is as good as reiserfs can do right now.
1822                  ** note that the private part of inode isn't filled in yet, we have
1823                  ** to use the directory.
1824                  */
1825                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1826         else
1827 #if defined( USE_INODE_GENERATION_COUNTER )
1828                 inode->i_generation =
1829                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1830 #else
1831                 inode->i_generation = ++event;
1832 #endif
1833
1834         /* fill stat data */
1835         set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
1836
1837         /* uid and gid must already be set by the caller for quota init */
1838
1839         /* symlink cannot be immutable or append only, right? */
1840         if (S_ISLNK(inode->i_mode))
1841                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1842
1843         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1844         inode->i_size = i_size;
1845         inode->i_blocks = 0;
1846         inode->i_bytes = 0;
1847         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1848             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1849
1850         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1851         REISERFS_I(inode)->i_flags = 0;
1852         REISERFS_I(inode)->i_prealloc_block = 0;
1853         REISERFS_I(inode)->i_prealloc_count = 0;
1854         REISERFS_I(inode)->i_trans_id = 0;
1855         REISERFS_I(inode)->i_jl = NULL;
1856         REISERFS_I(inode)->i_attrs =
1857             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1858         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1859         reiserfs_init_xattr_rwsem(inode);
1860
1861         /* key to search for correct place for new stat data */
1862         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1863                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1864                       TYPE_STAT_DATA, 3 /*key length */ );
1865
1866         /* find proper place for inserting of stat data */
1867         retval = search_item(sb, &key, &path_to_key);
1868         if (retval == IO_ERROR) {
1869                 err = -EIO;
1870                 goto out_bad_inode;
1871         }
1872         if (retval == ITEM_FOUND) {
1873                 pathrelse(&path_to_key);
1874                 err = -EEXIST;
1875                 goto out_bad_inode;
1876         }
1877         if (old_format_only(sb)) {
1878                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1879                         pathrelse(&path_to_key);
1880                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1881                         err = -EINVAL;
1882                         goto out_bad_inode;
1883                 }
1884                 inode2sd_v1(&sd, inode, inode->i_size);
1885         } else {
1886                 inode2sd(&sd, inode, inode->i_size);
1887         }
1888         // store in in-core inode the key of stat data and version all
1889         // object items will have (directory items will have old offset
1890         // format, other new objects will consist of new items)
1891         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1892                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1893         else
1894                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1895         if (old_format_only(sb))
1896                 set_inode_sd_version(inode, STAT_DATA_V1);
1897         else
1898                 set_inode_sd_version(inode, STAT_DATA_V2);
1899
1900         /* insert the stat data into the tree */
1901 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1902         if (REISERFS_I(dir)->new_packing_locality)
1903                 th->displace_new_blocks = 1;
1904 #endif
1905         retval =
1906             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1907                                  (char *)(&sd));
1908         if (retval) {
1909                 err = retval;
1910                 reiserfs_check_path(&path_to_key);
1911                 goto out_bad_inode;
1912         }
1913 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1914         if (!th->displace_new_blocks)
1915                 REISERFS_I(dir)->new_packing_locality = 0;
1916 #endif
1917         if (S_ISDIR(mode)) {
1918                 /* insert item with "." and ".." */
1919                 retval =
1920                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1921         }
1922
1923         if (S_ISLNK(mode)) {
1924                 /* insert body of symlink */
1925                 if (!old_format_only(sb))
1926                         i_size = ROUND_UP(i_size);
1927                 retval =
1928                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1929                                          i_size);
1930         }
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
1938         if (reiserfs_posixacl(inode->i_sb)) {
1939                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1940                 if (retval) {
1941                         err = retval;
1942                         reiserfs_check_path(&path_to_key);
1943                         journal_end(th, th->t_super, th->t_blocks_allocated);
1944                         goto out_inserted_sd;
1945                 }
1946         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1947                 reiserfs_warning(inode->i_sb, "jdm-13090",
1948                                  "ACLs aren't enabled in the fs, "
1949                                  "but vfs thinks they are!");
1950         } else if (IS_PRIVATE(dir))
1951                 inode->i_flags |= S_PRIVATE;
1952
1953         if (security->name) {
1954                 retval = reiserfs_security_write(th, inode, security);
1955                 if (retval) {
1956                         err = retval;
1957                         reiserfs_check_path(&path_to_key);
1958                         retval = journal_end(th, th->t_super,
1959                                              th->t_blocks_allocated);
1960                         if (retval)
1961                                 err = retval;
1962                         goto out_inserted_sd;
1963                 }
1964         }
1965
1966         reiserfs_update_sd(th, inode);
1967         reiserfs_check_path(&path_to_key);
1968
1969         return 0;
1970
1971 /* it looks like you can easily compress these two goto targets into
1972  * one.  Keeping it like this doesn't actually hurt anything, and they
1973  * are place holders for what the quota code actually needs.
1974  */
1975       out_bad_inode:
1976         /* Invalidate the object, nothing was inserted yet */
1977         INODE_PKEY(inode)->k_objectid = 0;
1978
1979         /* Quota change must be inside a transaction for journaling */
1980         dquot_free_inode(inode);
1981
1982       out_end_trans:
1983         journal_end(th, th->t_super, th->t_blocks_allocated);
1984         /* Drop can be outside and it needs more credits so it's better to have it outside */
1985         dquot_drop(inode);
1986         inode->i_flags |= S_NOQUOTA;
1987         make_bad_inode(inode);
1988
1989       out_inserted_sd:
1990         clear_nlink(inode);
1991         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1992         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1993         iput(inode);
1994         return err;
1995 }
1996
1997 /*
1998 ** finds the tail page in the page cache,
1999 ** reads the last block in.
2000 **
2001 ** On success, page_result is set to a locked, pinned page, and bh_result
2002 ** is set to an up to date buffer for the last block in the file.  returns 0.
2003 **
2004 ** tail conversion is not done, so bh_result might not be valid for writing
2005 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2006 ** trying to write the block.
2007 **
2008 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2009 */
2010 static int grab_tail_page(struct inode *inode,
2011                           struct page **page_result,
2012                           struct buffer_head **bh_result)
2013 {
2014
2015         /* we want the page with the last byte in the file,
2016          ** not the page that will hold the next byte for appending
2017          */
2018         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2019         unsigned long pos = 0;
2020         unsigned long start = 0;
2021         unsigned long blocksize = inode->i_sb->s_blocksize;
2022         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2023         struct buffer_head *bh;
2024         struct buffer_head *head;
2025         struct page *page;
2026         int error;
2027
2028         /* we know that we are only called with inode->i_size > 0.
2029          ** we also know that a file tail can never be as big as a block
2030          ** If i_size % blocksize == 0, our file is currently block aligned
2031          ** and it won't need converting or zeroing after a truncate.
2032          */
2033         if ((offset & (blocksize - 1)) == 0) {
2034                 return -ENOENT;
2035         }
2036         page = grab_cache_page(inode->i_mapping, index);
2037         error = -ENOMEM;
2038         if (!page) {
2039                 goto out;
2040         }
2041         /* start within the page of the last block in the file */
2042         start = (offset / blocksize) * blocksize;
2043
2044         error = __block_write_begin(page, start, offset - start,
2045                                     reiserfs_get_block_create_0);
2046         if (error)
2047                 goto unlock;
2048
2049         head = page_buffers(page);
2050         bh = head;
2051         do {
2052                 if (pos >= start) {
2053                         break;
2054                 }
2055                 bh = bh->b_this_page;
2056                 pos += blocksize;
2057         } while (bh != head);
2058
2059         if (!buffer_uptodate(bh)) {
2060                 /* note, this should never happen, prepare_write should
2061                  ** be taking care of this for us.  If the buffer isn't up to date,
2062                  ** I've screwed up the code to find the buffer, or the code to
2063                  ** call prepare_write
2064                  */
2065                 reiserfs_error(inode->i_sb, "clm-6000",
2066                                "error reading block %lu", bh->b_blocknr);
2067                 error = -EIO;
2068                 goto unlock;
2069         }
2070         *bh_result = bh;
2071         *page_result = page;
2072
2073       out:
2074         return error;
2075
2076       unlock:
2077         unlock_page(page);
2078         page_cache_release(page);
2079         return error;
2080 }
2081
2082 /*
2083 ** vfs version of truncate file.  Must NOT be called with
2084 ** a transaction already started.
2085 **
2086 ** some code taken from block_truncate_page
2087 */
2088 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2089 {
2090         struct reiserfs_transaction_handle th;
2091         /* we want the offset for the first byte after the end of the file */
2092         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2093         unsigned blocksize = inode->i_sb->s_blocksize;
2094         unsigned length;
2095         struct page *page = NULL;
2096         int error;
2097         struct buffer_head *bh = NULL;
2098         int err2;
2099         int lock_depth;
2100
2101         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2102
2103         if (inode->i_size > 0) {
2104                 error = grab_tail_page(inode, &page, &bh);
2105                 if (error) {
2106                         // -ENOENT means we truncated past the end of the file,
2107                         // and get_block_create_0 could not find a block to read in,
2108                         // which is ok.
2109                         if (error != -ENOENT)
2110                                 reiserfs_error(inode->i_sb, "clm-6001",
2111                                                "grab_tail_page failed %d",
2112                                                error);
2113                         page = NULL;
2114                         bh = NULL;
2115                 }
2116         }
2117
2118         /* so, if page != NULL, we have a buffer head for the offset at
2119          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2120          ** then we have an unformatted node.  Otherwise, we have a direct item,
2121          ** and no zeroing is required on disk.  We zero after the truncate,
2122          ** because the truncate might pack the item anyway
2123          ** (it will unmap bh if it packs).
2124          */
2125         /* it is enough to reserve space in transaction for 2 balancings:
2126            one for "save" link adding and another for the first
2127            cut_from_item. 1 is for update_sd */
2128         error = journal_begin(&th, inode->i_sb,
2129                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2130         if (error)
2131                 goto out;
2132         reiserfs_update_inode_transaction(inode);
2133         if (update_timestamps)
2134                 /* we are doing real truncate: if the system crashes before the last
2135                    transaction of truncating gets committed - on reboot the file
2136                    either appears truncated properly or not truncated at all */
2137                 add_save_link(&th, inode, 1);
2138         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2139         error =
2140             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2141         if (error)
2142                 goto out;
2143
2144         /* check reiserfs_do_truncate after ending the transaction */
2145         if (err2) {
2146                 error = err2;
2147                 goto out;
2148         }
2149
2150         if (update_timestamps) {
2151                 error = remove_save_link(inode, 1 /* truncate */);
2152                 if (error)
2153                         goto out;
2154         }
2155
2156         if (page) {
2157                 length = offset & (blocksize - 1);
2158                 /* if we are not on a block boundary */
2159                 if (length) {
2160                         length = blocksize - length;
2161                         zero_user(page, offset, length);
2162                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2163                                 mark_buffer_dirty(bh);
2164                         }
2165                 }
2166                 unlock_page(page);
2167                 page_cache_release(page);
2168         }
2169
2170         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2171
2172         return 0;
2173       out:
2174         if (page) {
2175                 unlock_page(page);
2176                 page_cache_release(page);
2177         }
2178
2179         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2180
2181         return error;
2182 }
2183
2184 static int map_block_for_writepage(struct inode *inode,
2185                                    struct buffer_head *bh_result,
2186                                    unsigned long block)
2187 {
2188         struct reiserfs_transaction_handle th;
2189         int fs_gen;
2190         struct item_head tmp_ih;
2191         struct item_head *ih;
2192         struct buffer_head *bh;
2193         __le32 *item;
2194         struct cpu_key key;
2195         INITIALIZE_PATH(path);
2196         int pos_in_item;
2197         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2198         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2199         int retval;
2200         int use_get_block = 0;
2201         int bytes_copied = 0;
2202         int copy_size;
2203         int trans_running = 0;
2204
2205         /* catch places below that try to log something without starting a trans */
2206         th.t_trans_id = 0;
2207
2208         if (!buffer_uptodate(bh_result)) {
2209                 return -EIO;
2210         }
2211
2212         kmap(bh_result->b_page);
2213       start_over:
2214         reiserfs_write_lock(inode->i_sb);
2215         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2216
2217       research:
2218         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2219         if (retval != POSITION_FOUND) {
2220                 use_get_block = 1;
2221                 goto out;
2222         }
2223
2224         bh = get_last_bh(&path);
2225         ih = get_ih(&path);
2226         item = get_item(&path);
2227         pos_in_item = path.pos_in_item;
2228
2229         /* we've found an unformatted node */
2230         if (indirect_item_found(retval, ih)) {
2231                 if (bytes_copied > 0) {
2232                         reiserfs_warning(inode->i_sb, "clm-6002",
2233                                          "bytes_copied %d", bytes_copied);
2234                 }
2235                 if (!get_block_num(item, pos_in_item)) {
2236                         /* crap, we are writing to a hole */
2237                         use_get_block = 1;
2238                         goto out;
2239                 }
2240                 set_block_dev_mapped(bh_result,
2241                                      get_block_num(item, pos_in_item), inode);
2242         } else if (is_direct_le_ih(ih)) {
2243                 char *p;
2244                 p = page_address(bh_result->b_page);
2245                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2246                 copy_size = ih_item_len(ih) - pos_in_item;
2247
2248                 fs_gen = get_generation(inode->i_sb);
2249                 copy_item_head(&tmp_ih, ih);
2250
2251                 if (!trans_running) {
2252                         /* vs-3050 is gone, no need to drop the path */
2253                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2254                         if (retval)
2255                                 goto out;
2256                         reiserfs_update_inode_transaction(inode);
2257                         trans_running = 1;
2258                         if (fs_changed(fs_gen, inode->i_sb)
2259                             && item_moved(&tmp_ih, &path)) {
2260                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2261                                                                  bh);
2262                                 goto research;
2263                         }
2264                 }
2265
2266                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2267
2268                 if (fs_changed(fs_gen, inode->i_sb)
2269                     && item_moved(&tmp_ih, &path)) {
2270                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2271                         goto research;
2272                 }
2273
2274                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2275                        copy_size);
2276
2277                 journal_mark_dirty(&th, inode->i_sb, bh);
2278                 bytes_copied += copy_size;
2279                 set_block_dev_mapped(bh_result, 0, inode);
2280
2281                 /* are there still bytes left? */
2282                 if (bytes_copied < bh_result->b_size &&
2283                     (byte_offset + bytes_copied) < inode->i_size) {
2284                         set_cpu_key_k_offset(&key,
2285                                              cpu_key_k_offset(&key) +
2286                                              copy_size);
2287                         goto research;
2288                 }
2289         } else {
2290                 reiserfs_warning(inode->i_sb, "clm-6003",
2291                                  "bad item inode %lu", inode->i_ino);
2292                 retval = -EIO;
2293                 goto out;
2294         }
2295         retval = 0;
2296
2297       out:
2298         pathrelse(&path);
2299         if (trans_running) {
2300                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2301                 if (err)
2302                         retval = err;
2303                 trans_running = 0;
2304         }
2305         reiserfs_write_unlock(inode->i_sb);
2306
2307         /* this is where we fill in holes in the file. */
2308         if (use_get_block) {
2309                 retval = reiserfs_get_block(inode, block, bh_result,
2310                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2311                                             | GET_BLOCK_NO_DANGLE);
2312                 if (!retval) {
2313                         if (!buffer_mapped(bh_result)
2314                             || bh_result->b_blocknr == 0) {
2315                                 /* get_block failed to find a mapped unformatted node. */
2316                                 use_get_block = 0;
2317                                 goto start_over;
2318                         }
2319                 }
2320         }
2321         kunmap(bh_result->b_page);
2322
2323         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2324                 /* we've copied data from the page into the direct item, so the
2325                  * buffer in the page is now clean, mark it to reflect that.
2326                  */
2327                 lock_buffer(bh_result);
2328                 clear_buffer_dirty(bh_result);
2329                 unlock_buffer(bh_result);
2330         }
2331         return retval;
2332 }
2333
2334 /*
2335  * mason@suse.com: updated in 2.5.54 to follow the same general io
2336  * start/recovery path as __block_write_full_page, along with special
2337  * code to handle reiserfs tails.
2338  */
2339 static int reiserfs_write_full_page(struct page *page,
2340                                     struct writeback_control *wbc)
2341 {
2342         struct inode *inode = page->mapping->host;
2343         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2344         int error = 0;
2345         unsigned long block;
2346         sector_t last_block;
2347         struct buffer_head *head, *bh;
2348         int partial = 0;
2349         int nr = 0;
2350         int checked = PageChecked(page);
2351         struct reiserfs_transaction_handle th;
2352         struct super_block *s = inode->i_sb;
2353         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2354         th.t_trans_id = 0;
2355
2356         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2357         if (checked && (current->flags & PF_MEMALLOC)) {
2358                 redirty_page_for_writepage(wbc, page);
2359                 unlock_page(page);
2360                 return 0;
2361         }
2362
2363         /* The page dirty bit is cleared before writepage is called, which
2364          * means we have to tell create_empty_buffers to make dirty buffers
2365          * The page really should be up to date at this point, so tossing
2366          * in the BH_Uptodate is just a sanity check.
2367          */
2368         if (!page_has_buffers(page)) {
2369                 create_empty_buffers(page, s->s_blocksize,
2370                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2371         }
2372         head = page_buffers(page);
2373
2374         /* last page in the file, zero out any contents past the
2375          ** last byte in the file
2376          */
2377         if (page->index >= end_index) {
2378                 unsigned last_offset;
2379
2380                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2381                 /* no file contents in this page */
2382                 if (page->index >= end_index + 1 || !last_offset) {
2383                         unlock_page(page);
2384                         return 0;
2385                 }
2386                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2387         }
2388         bh = head;
2389         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2390         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2391         /* first map all the buffers, logging any direct items we find */
2392         do {
2393                 if (block > last_block) {
2394                         /*
2395                          * This can happen when the block size is less than
2396                          * the page size.  The corresponding bytes in the page
2397                          * were zero filled above
2398                          */
2399                         clear_buffer_dirty(bh);
2400                         set_buffer_uptodate(bh);
2401                 } else if ((checked || buffer_dirty(bh)) &&
2402                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2403                                                        && bh->b_blocknr ==
2404                                                        0))) {
2405                         /* not mapped yet, or it points to a direct item, search
2406                          * the btree for the mapping info, and log any direct
2407                          * items found
2408                          */
2409                         if ((error = map_block_for_writepage(inode, bh, block))) {
2410                                 goto fail;
2411                         }
2412                 }
2413                 bh = bh->b_this_page;
2414                 block++;
2415         } while (bh != head);
2416
2417         /*
2418          * we start the transaction after map_block_for_writepage,
2419          * because it can create holes in the file (an unbounded operation).
2420          * starting it here, we can make a reliable estimate for how many
2421          * blocks we're going to log
2422          */
2423         if (checked) {
2424                 ClearPageChecked(page);
2425                 reiserfs_write_lock(s);
2426                 error = journal_begin(&th, s, bh_per_page + 1);
2427                 if (error) {
2428                         reiserfs_write_unlock(s);
2429                         goto fail;
2430                 }
2431                 reiserfs_update_inode_transaction(inode);
2432         }
2433         /* now go through and lock any dirty buffers on the page */
2434         do {
2435                 get_bh(bh);
2436                 if (!buffer_mapped(bh))
2437                         continue;
2438                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2439                         continue;
2440
2441                 if (checked) {
2442                         reiserfs_prepare_for_journal(s, bh, 1);
2443                         journal_mark_dirty(&th, s, bh);
2444                         continue;
2445                 }
2446                 /* from this point on, we know the buffer is mapped to a
2447                  * real block and not a direct item
2448                  */
2449                 if (wbc->sync_mode != WB_SYNC_NONE) {
2450                         lock_buffer(bh);
2451                 } else {
2452                         if (!trylock_buffer(bh)) {
2453                                 redirty_page_for_writepage(wbc, page);
2454                                 continue;
2455                         }
2456                 }
2457                 if (test_clear_buffer_dirty(bh)) {
2458                         mark_buffer_async_write(bh);
2459                 } else {
2460                         unlock_buffer(bh);
2461                 }
2462         } while ((bh = bh->b_this_page) != head);
2463
2464         if (checked) {
2465                 error = journal_end(&th, s, bh_per_page + 1);
2466                 reiserfs_write_unlock(s);
2467                 if (error)
2468                         goto fail;
2469         }
2470         BUG_ON(PageWriteback(page));
2471         set_page_writeback(page);
2472         unlock_page(page);
2473
2474         /*
2475          * since any buffer might be the only dirty buffer on the page,
2476          * the first submit_bh can bring the page out of writeback.
2477          * be careful with the buffers.
2478          */
2479         do {
2480                 struct buffer_head *next = bh->b_this_page;
2481                 if (buffer_async_write(bh)) {
2482                         submit_bh(WRITE, bh);
2483                         nr++;
2484                 }
2485                 put_bh(bh);
2486                 bh = next;
2487         } while (bh != head);
2488
2489         error = 0;
2490       done:
2491         if (nr == 0) {
2492                 /*
2493                  * if this page only had a direct item, it is very possible for
2494                  * no io to be required without there being an error.  Or,
2495                  * someone else could have locked them and sent them down the
2496                  * pipe without locking the page
2497                  */
2498                 bh = head;
2499                 do {
2500                         if (!buffer_uptodate(bh)) {
2501                                 partial = 1;
2502                                 break;
2503                         }
2504                         bh = bh->b_this_page;
2505                 } while (bh != head);
2506                 if (!partial)
2507                         SetPageUptodate(page);
2508                 end_page_writeback(page);
2509         }
2510         return error;
2511
2512       fail:
2513         /* catches various errors, we need to make sure any valid dirty blocks
2514          * get to the media.  The page is currently locked and not marked for
2515          * writeback
2516          */
2517         ClearPageUptodate(page);
2518         bh = head;
2519         do {
2520                 get_bh(bh);
2521                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2522                         lock_buffer(bh);
2523                         mark_buffer_async_write(bh);
2524                 } else {
2525                         /*
2526                          * clear any dirty bits that might have come from getting
2527                          * attached to a dirty page
2528                          */
2529                         clear_buffer_dirty(bh);
2530                 }
2531                 bh = bh->b_this_page;
2532         } while (bh != head);
2533         SetPageError(page);
2534         BUG_ON(PageWriteback(page));
2535         set_page_writeback(page);
2536         unlock_page(page);
2537         do {
2538                 struct buffer_head *next = bh->b_this_page;
2539                 if (buffer_async_write(bh)) {
2540                         clear_buffer_dirty(bh);
2541                         submit_bh(WRITE, bh);
2542                         nr++;
2543                 }
2544                 put_bh(bh);
2545                 bh = next;
2546         } while (bh != head);
2547         goto done;
2548 }
2549
2550 static int reiserfs_readpage(struct file *f, struct page *page)
2551 {
2552         return block_read_full_page(page, reiserfs_get_block);
2553 }
2554
2555 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2556 {
2557         struct inode *inode = page->mapping->host;
2558         reiserfs_wait_on_write_block(inode->i_sb);
2559         return reiserfs_write_full_page(page, wbc);
2560 }
2561
2562 static void reiserfs_truncate_failed_write(struct inode *inode)
2563 {
2564         truncate_inode_pages(inode->i_mapping, inode->i_size);
2565         reiserfs_truncate_file(inode, 0);
2566 }
2567
2568 static int reiserfs_write_begin(struct file *file,
2569                                 struct address_space *mapping,
2570                                 loff_t pos, unsigned len, unsigned flags,
2571                                 struct page **pagep, void **fsdata)
2572 {
2573         struct inode *inode;
2574         struct page *page;
2575         pgoff_t index;
2576         int ret;
2577         int old_ref = 0;
2578
2579         inode = mapping->host;
2580         *fsdata = 0;
2581         if (flags & AOP_FLAG_CONT_EXPAND &&
2582             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2583                 pos ++;
2584                 *fsdata = (void *)(unsigned long)flags;
2585         }
2586
2587         index = pos >> PAGE_CACHE_SHIFT;
2588         page = grab_cache_page_write_begin(mapping, index, flags);
2589         if (!page)
2590                 return -ENOMEM;
2591         *pagep = page;
2592
2593         reiserfs_wait_on_write_block(inode->i_sb);
2594         fix_tail_page_for_writing(page);
2595         if (reiserfs_transaction_running(inode->i_sb)) {
2596                 struct reiserfs_transaction_handle *th;
2597                 th = (struct reiserfs_transaction_handle *)current->
2598                     journal_info;
2599                 BUG_ON(!th->t_refcount);
2600                 BUG_ON(!th->t_trans_id);
2601                 old_ref = th->t_refcount;
2602                 th->t_refcount++;
2603         }
2604         ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2605         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2606                 struct reiserfs_transaction_handle *th = current->journal_info;
2607                 /* this gets a little ugly.  If reiserfs_get_block returned an
2608                  * error and left a transacstion running, we've got to close it,
2609                  * and we've got to free handle if it was a persistent transaction.
2610                  *
2611                  * But, if we had nested into an existing transaction, we need
2612                  * to just drop the ref count on the handle.
2613                  *
2614                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2615                  * and it was a persistent trans.  Otherwise, it was nested above.
2616                  */
2617                 if (th->t_refcount > old_ref) {
2618                         if (old_ref)
2619                                 th->t_refcount--;
2620                         else {
2621                                 int err;
2622                                 reiserfs_write_lock(inode->i_sb);
2623                                 err = reiserfs_end_persistent_transaction(th);
2624                                 reiserfs_write_unlock(inode->i_sb);
2625                                 if (err)
2626                                         ret = err;
2627                         }
2628                 }
2629         }
2630         if (ret) {
2631                 unlock_page(page);
2632                 page_cache_release(page);
2633                 /* Truncate allocated blocks */
2634                 reiserfs_truncate_failed_write(inode);
2635         }
2636         return ret;
2637 }
2638
2639 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2640 {
2641         struct inode *inode = page->mapping->host;
2642         int ret;
2643         int old_ref = 0;
2644
2645         reiserfs_write_unlock(inode->i_sb);
2646         reiserfs_wait_on_write_block(inode->i_sb);
2647         reiserfs_write_lock(inode->i_sb);
2648
2649         fix_tail_page_for_writing(page);
2650         if (reiserfs_transaction_running(inode->i_sb)) {
2651                 struct reiserfs_transaction_handle *th;
2652                 th = (struct reiserfs_transaction_handle *)current->
2653                     journal_info;
2654                 BUG_ON(!th->t_refcount);
2655                 BUG_ON(!th->t_trans_id);
2656                 old_ref = th->t_refcount;
2657                 th->t_refcount++;
2658         }
2659
2660         ret = __block_write_begin(page, from, len, reiserfs_get_block);
2661         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2662                 struct reiserfs_transaction_handle *th = current->journal_info;
2663                 /* this gets a little ugly.  If reiserfs_get_block returned an
2664                  * error and left a transacstion running, we've got to close it,
2665                  * and we've got to free handle if it was a persistent transaction.
2666                  *
2667                  * But, if we had nested into an existing transaction, we need
2668                  * to just drop the ref count on the handle.
2669                  *
2670                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2671                  * and it was a persistent trans.  Otherwise, it was nested above.
2672                  */
2673                 if (th->t_refcount > old_ref) {
2674                         if (old_ref)
2675                                 th->t_refcount--;
2676                         else {
2677                                 int err;
2678                                 reiserfs_write_lock(inode->i_sb);
2679                                 err = reiserfs_end_persistent_transaction(th);
2680                                 reiserfs_write_unlock(inode->i_sb);
2681                                 if (err)
2682                                         ret = err;
2683                         }
2684                 }
2685         }
2686         return ret;
2687
2688 }
2689
2690 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2691 {
2692         return generic_block_bmap(as, block, reiserfs_bmap);
2693 }
2694
2695 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2696                               loff_t pos, unsigned len, unsigned copied,
2697                               struct page *page, void *fsdata)
2698 {
2699         struct inode *inode = page->mapping->host;
2700         int ret = 0;
2701         int update_sd = 0;
2702         struct reiserfs_transaction_handle *th;
2703         unsigned start;
2704         int lock_depth = 0;
2705         bool locked = false;
2706
2707         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2708                 pos ++;
2709
2710         reiserfs_wait_on_write_block(inode->i_sb);
2711         if (reiserfs_transaction_running(inode->i_sb))
2712                 th = current->journal_info;
2713         else
2714                 th = NULL;
2715
2716         start = pos & (PAGE_CACHE_SIZE - 1);
2717         if (unlikely(copied < len)) {
2718                 if (!PageUptodate(page))
2719                         copied = 0;
2720
2721                 page_zero_new_buffers(page, start + copied, start + len);
2722         }
2723         flush_dcache_page(page);
2724
2725         reiserfs_commit_page(inode, page, start, start + copied);
2726
2727         /* generic_commit_write does this for us, but does not update the
2728          ** transaction tracking stuff when the size changes.  So, we have
2729          ** to do the i_size updates here.
2730          */
2731         if (pos + copied > inode->i_size) {
2732                 struct reiserfs_transaction_handle myth;
2733                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2734                 locked = true;
2735                 /* If the file have grown beyond the border where it
2736                    can have a tail, unmark it as needing a tail
2737                    packing */
2738                 if ((have_large_tails(inode->i_sb)
2739                      && inode->i_size > i_block_size(inode) * 4)
2740                     || (have_small_tails(inode->i_sb)
2741                         && inode->i_size > i_block_size(inode)))
2742                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2743
2744                 ret = journal_begin(&myth, inode->i_sb, 1);
2745                 if (ret)
2746                         goto journal_error;
2747
2748                 reiserfs_update_inode_transaction(inode);
2749                 inode->i_size = pos + copied;
2750                 /*
2751                  * this will just nest into our transaction.  It's important
2752                  * to use mark_inode_dirty so the inode gets pushed around on the
2753                  * dirty lists, and so that O_SYNC works as expected
2754                  */
2755                 mark_inode_dirty(inode);
2756                 reiserfs_update_sd(&myth, inode);
2757                 update_sd = 1;
2758                 ret = journal_end(&myth, inode->i_sb, 1);
2759                 if (ret)
2760                         goto journal_error;
2761         }
2762         if (th) {
2763                 if (!locked) {
2764                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2765                         locked = true;
2766                 }
2767                 if (!update_sd)
2768                         mark_inode_dirty(inode);
2769                 ret = reiserfs_end_persistent_transaction(th);
2770                 if (ret)
2771                         goto out;
2772         }
2773
2774       out:
2775         if (locked)
2776                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2777         unlock_page(page);
2778         page_cache_release(page);
2779
2780         if (pos + len > inode->i_size)
2781                 reiserfs_truncate_failed_write(inode);
2782
2783         return ret == 0 ? copied : ret;
2784
2785       journal_error:
2786         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2787         locked = false;
2788         if (th) {
2789                 if (!update_sd)
2790                         reiserfs_update_sd(th, inode);
2791                 ret = reiserfs_end_persistent_transaction(th);
2792         }
2793         goto out;
2794 }
2795
2796 int reiserfs_commit_write(struct file *f, struct page *page,
2797                           unsigned from, unsigned to)
2798 {
2799         struct inode *inode = page->mapping->host;
2800         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2801         int ret = 0;
2802         int update_sd = 0;
2803         struct reiserfs_transaction_handle *th = NULL;
2804
2805         reiserfs_write_unlock(inode->i_sb);
2806         reiserfs_wait_on_write_block(inode->i_sb);
2807         reiserfs_write_lock(inode->i_sb);
2808
2809         if (reiserfs_transaction_running(inode->i_sb)) {
2810                 th = current->journal_info;
2811         }
2812         reiserfs_commit_page(inode, page, from, to);
2813
2814         /* generic_commit_write does this for us, but does not update the
2815          ** transaction tracking stuff when the size changes.  So, we have
2816          ** to do the i_size updates here.
2817          */
2818         if (pos > inode->i_size) {
2819                 struct reiserfs_transaction_handle myth;
2820                 /* If the file have grown beyond the border where it
2821                    can have a tail, unmark it as needing a tail
2822                    packing */
2823                 if ((have_large_tails(inode->i_sb)
2824                      && inode->i_size > i_block_size(inode) * 4)
2825                     || (have_small_tails(inode->i_sb)
2826                         && inode->i_size > i_block_size(inode)))
2827                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2828
2829                 ret = journal_begin(&myth, inode->i_sb, 1);
2830                 if (ret)
2831                         goto journal_error;
2832
2833                 reiserfs_update_inode_transaction(inode);
2834                 inode->i_size = pos;
2835                 /*
2836                  * this will just nest into our transaction.  It's important
2837                  * to use mark_inode_dirty so the inode gets pushed around on the
2838                  * dirty lists, and so that O_SYNC works as expected
2839                  */
2840                 mark_inode_dirty(inode);
2841                 reiserfs_update_sd(&myth, inode);
2842                 update_sd = 1;
2843                 ret = journal_end(&myth, inode->i_sb, 1);
2844                 if (ret)
2845                         goto journal_error;
2846         }
2847         if (th) {
2848                 if (!update_sd)
2849                         mark_inode_dirty(inode);
2850                 ret = reiserfs_end_persistent_transaction(th);
2851                 if (ret)
2852                         goto out;
2853         }
2854
2855       out:
2856         return ret;
2857
2858       journal_error:
2859         if (th) {
2860                 if (!update_sd)
2861                         reiserfs_update_sd(th, inode);
2862                 ret = reiserfs_end_persistent_transaction(th);
2863         }
2864
2865         return ret;
2866 }
2867
2868 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2869 {
2870         if (reiserfs_attrs(inode->i_sb)) {
2871                 if (sd_attrs & REISERFS_SYNC_FL)
2872                         inode->i_flags |= S_SYNC;
2873                 else
2874                         inode->i_flags &= ~S_SYNC;
2875                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2876                         inode->i_flags |= S_IMMUTABLE;
2877                 else
2878                         inode->i_flags &= ~S_IMMUTABLE;
2879                 if (sd_attrs & REISERFS_APPEND_FL)
2880                         inode->i_flags |= S_APPEND;
2881                 else
2882                         inode->i_flags &= ~S_APPEND;
2883                 if (sd_attrs & REISERFS_NOATIME_FL)
2884                         inode->i_flags |= S_NOATIME;
2885                 else
2886                         inode->i_flags &= ~S_NOATIME;
2887                 if (sd_attrs & REISERFS_NOTAIL_FL)
2888                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2889                 else
2890                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2891         }
2892 }
2893
2894 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2895 {
2896         if (reiserfs_attrs(inode->i_sb)) {
2897                 if (inode->i_flags & S_IMMUTABLE)
2898                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2899                 else
2900                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2901                 if (inode->i_flags & S_SYNC)
2902                         *sd_attrs |= REISERFS_SYNC_FL;
2903                 else
2904                         *sd_attrs &= ~REISERFS_SYNC_FL;
2905                 if (inode->i_flags & S_NOATIME)
2906                         *sd_attrs |= REISERFS_NOATIME_FL;
2907                 else
2908                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2909                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2910                         *sd_attrs |= REISERFS_NOTAIL_FL;
2911                 else
2912                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2913         }
2914 }
2915
2916 /* decide if this buffer needs to stay around for data logging or ordered
2917 ** write purposes
2918 */
2919 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2920 {
2921         int ret = 1;
2922         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2923
2924         lock_buffer(bh);
2925         spin_lock(&j->j_dirty_buffers_lock);
2926         if (!buffer_mapped(bh)) {
2927                 goto free_jh;
2928         }
2929         /* the page is locked, and the only places that log a data buffer
2930          * also lock the page.
2931          */
2932         if (reiserfs_file_data_log(inode)) {
2933                 /*
2934                  * very conservative, leave the buffer pinned if
2935                  * anyone might need it.
2936                  */
2937                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2938                         ret = 0;
2939                 }
2940         } else  if (buffer_dirty(bh)) {
2941                 struct reiserfs_journal_list *jl;
2942                 struct reiserfs_jh *jh = bh->b_private;
2943
2944                 /* why is this safe?
2945                  * reiserfs_setattr updates i_size in the on disk
2946                  * stat data before allowing vmtruncate to be called.
2947                  *
2948                  * If buffer was put onto the ordered list for this
2949                  * transaction, we know for sure either this transaction
2950                  * or an older one already has updated i_size on disk,
2951                  * and this ordered data won't be referenced in the file
2952                  * if we crash.
2953                  *
2954                  * if the buffer was put onto the ordered list for an older
2955                  * transaction, we need to leave it around
2956                  */
2957                 if (jh && (jl = jh->jl)
2958                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2959                         ret = 0;
2960         }
2961       free_jh:
2962         if (ret && bh->b_private) {
2963                 reiserfs_free_jh(bh);
2964         }
2965         spin_unlock(&j->j_dirty_buffers_lock);
2966         unlock_buffer(bh);
2967         return ret;
2968 }
2969
2970 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2971 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2972 {
2973         struct buffer_head *head, *bh, *next;
2974         struct inode *inode = page->mapping->host;
2975         unsigned int curr_off = 0;
2976         int ret = 1;
2977
2978         BUG_ON(!PageLocked(page));
2979
2980         if (offset == 0)
2981                 ClearPageChecked(page);
2982
2983         if (!page_has_buffers(page))
2984                 goto out;
2985
2986         head = page_buffers(page);
2987         bh = head;
2988         do {
2989                 unsigned int next_off = curr_off + bh->b_size;
2990                 next = bh->b_this_page;
2991
2992                 /*
2993                  * is this block fully invalidated?
2994                  */
2995                 if (offset <= curr_off) {
2996                         if (invalidatepage_can_drop(inode, bh))
2997                                 reiserfs_unmap_buffer(bh);
2998                         else
2999                                 ret = 0;
3000                 }
3001                 curr_off = next_off;
3002                 bh = next;
3003         } while (bh != head);
3004
3005         /*
3006          * We release buffers only if the entire page is being invalidated.
3007          * The get_block cached value has been unconditionally invalidated,
3008          * so real IO is not possible anymore.
3009          */
3010         if (!offset && ret) {
3011                 ret = try_to_release_page(page, 0);
3012                 /* maybe should BUG_ON(!ret); - neilb */
3013         }
3014       out:
3015         return;
3016 }
3017
3018 static int reiserfs_set_page_dirty(struct page *page)
3019 {
3020         struct inode *inode = page->mapping->host;
3021         if (reiserfs_file_data_log(inode)) {
3022                 SetPageChecked(page);
3023                 return __set_page_dirty_nobuffers(page);
3024         }
3025         return __set_page_dirty_buffers(page);
3026 }
3027
3028 /*
3029  * Returns 1 if the page's buffers were dropped.  The page is locked.
3030  *
3031  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3032  * in the buffers at page_buffers(page).
3033  *
3034  * even in -o notail mode, we can't be sure an old mount without -o notail
3035  * didn't create files with tails.
3036  */
3037 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3038 {
3039         struct inode *inode = page->mapping->host;
3040         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3041         struct buffer_head *head;
3042         struct buffer_head *bh;
3043         int ret = 1;
3044
3045         WARN_ON(PageChecked(page));
3046         spin_lock(&j->j_dirty_buffers_lock);
3047         head = page_buffers(page);
3048         bh = head;
3049         do {
3050                 if (bh->b_private) {
3051                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3052                                 reiserfs_free_jh(bh);
3053                         } else {
3054                                 ret = 0;
3055                                 break;
3056                         }
3057                 }
3058                 bh = bh->b_this_page;
3059         } while (bh != head);
3060         if (ret)
3061                 ret = try_to_free_buffers(page);
3062         spin_unlock(&j->j_dirty_buffers_lock);
3063         return ret;
3064 }
3065
3066 /* We thank Mingming Cao for helping us understand in great detail what
3067    to do in this section of the code. */
3068 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3069                                   const struct iovec *iov, loff_t offset,
3070                                   unsigned long nr_segs)
3071 {
3072         struct file *file = iocb->ki_filp;
3073         struct inode *inode = file->f_mapping->host;
3074         ssize_t ret;
3075
3076         ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
3077                                   reiserfs_get_blocks_direct_io);
3078
3079         /*
3080          * In case of error extending write may have instantiated a few
3081          * blocks outside i_size. Trim these off again.
3082          */
3083         if (unlikely((rw & WRITE) && ret < 0)) {
3084                 loff_t isize = i_size_read(inode);
3085                 loff_t end = offset + iov_length(iov, nr_segs);
3086
3087                 if (end > isize)
3088                         vmtruncate(inode, isize);
3089         }
3090
3091         return ret;
3092 }
3093
3094 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3095 {
3096         struct inode *inode = dentry->d_inode;
3097         unsigned int ia_valid;
3098         int depth;
3099         int error;
3100
3101         error = inode_change_ok(inode, attr);
3102         if (error)
3103                 return error;
3104
3105         /* must be turned off for recursive notify_change calls */
3106         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3107
3108         depth = reiserfs_write_lock_once(inode->i_sb);
3109         if (is_quota_modification(inode, attr))
3110                 dquot_initialize(inode);
3111
3112         if (attr->ia_valid & ATTR_SIZE) {
3113                 /* version 2 items will be caught by the s_maxbytes check
3114                  ** done for us in vmtruncate
3115                  */
3116                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3117                     attr->ia_size > MAX_NON_LFS) {
3118                         error = -EFBIG;
3119                         goto out;
3120                 }
3121
3122                 inode_dio_wait(inode);
3123
3124                 /* fill in hole pointers in the expanding truncate case. */
3125                 if (attr->ia_size > inode->i_size) {
3126                         error = generic_cont_expand_simple(inode, attr->ia_size);
3127                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3128                                 int err;
3129                                 struct reiserfs_transaction_handle th;
3130                                 /* we're changing at most 2 bitmaps, inode + super */
3131                                 err = journal_begin(&th, inode->i_sb, 4);
3132                                 if (!err) {
3133                                         reiserfs_discard_prealloc(&th, inode);
3134                                         err = journal_end(&th, inode->i_sb, 4);
3135                                 }
3136                                 if (err)
3137                                         error = err;
3138                         }
3139                         if (error)
3140                                 goto out;
3141                         /*
3142                          * file size is changed, ctime and mtime are
3143                          * to be updated
3144                          */
3145                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3146                 }
3147         }
3148
3149         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3150              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3151             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3152                 /* stat data of format v3.5 has 16 bit uid and gid */
3153                 error = -EINVAL;
3154                 goto out;
3155         }
3156
3157         if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3158             (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3159                 struct reiserfs_transaction_handle th;
3160                 int jbegin_count =
3161                     2 *
3162                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3163                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3164                     2;
3165
3166                 error = reiserfs_chown_xattrs(inode, attr);
3167
3168                 if (error)
3169                         return error;
3170
3171                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3172                 error = journal_begin(&th, inode->i_sb, jbegin_count);
3173                 if (error)
3174                         goto out;
3175                 error = dquot_transfer(inode, attr);
3176                 if (error) {
3177                         journal_end(&th, inode->i_sb, jbegin_count);
3178                         goto out;
3179                 }
3180
3181                 /* Update corresponding info in inode so that everything is in
3182                  * one transaction */
3183                 if (attr->ia_valid & ATTR_UID)
3184                         inode->i_uid = attr->ia_uid;
3185                 if (attr->ia_valid & ATTR_GID)
3186                         inode->i_gid = attr->ia_gid;
3187                 mark_inode_dirty(inode);
3188                 error = journal_end(&th, inode->i_sb, jbegin_count);
3189                 if (error)
3190                         goto out;
3191         }
3192
3193         /*
3194          * Relax the lock here, as it might truncate the
3195          * inode pages and wait for inode pages locks.
3196          * To release such page lock, the owner needs the
3197          * reiserfs lock
3198          */
3199         reiserfs_write_unlock_once(inode->i_sb, depth);
3200         if ((attr->ia_valid & ATTR_SIZE) &&
3201             attr->ia_size != i_size_read(inode))
3202                 error = vmtruncate(inode, attr->ia_size);
3203
3204         if (!error) {
3205                 setattr_copy(inode, attr);
3206                 mark_inode_dirty(inode);
3207         }
3208         depth = reiserfs_write_lock_once(inode->i_sb);
3209
3210         if (!error && reiserfs_posixacl(inode->i_sb)) {
3211                 if (attr->ia_valid & ATTR_MODE)
3212                         error = reiserfs_acl_chmod(inode);
3213         }
3214
3215       out:
3216         reiserfs_write_unlock_once(inode->i_sb, depth);
3217
3218         return error;
3219 }
3220
3221 const struct address_space_operations reiserfs_address_space_operations = {
3222         .writepage = reiserfs_writepage,
3223         .readpage = reiserfs_readpage,
3224         .readpages = reiserfs_readpages,
3225         .releasepage = reiserfs_releasepage,
3226         .invalidatepage = reiserfs_invalidatepage,
3227         .write_begin = reiserfs_write_begin,
3228         .write_end = reiserfs_write_end,
3229         .bmap = reiserfs_aop_bmap,
3230         .direct_IO = reiserfs_direct_IO,
3231         .set_page_dirty = reiserfs_set_page_dirty,
3232 };