root/fs/reiserfs/inode.c

DEFINITIONS

1. reiserfs_evict_inode
2. _make_cpu_key
3. make_cpu_key
4. make_le_item_head
5. fix_tail_page_for_writing
6. allocation_needed
7. indirect_item_found
8. set_block_dev_mapped
9. file_capable
10. restart_transaction
11. _get_block_create_0
12. reiserfs_bmap
13. reiserfs_get_block_create_0
14. reiserfs_get_blocks_direct_io
15. convert_tail_for_hole
16. _allocate_block
17. reiserfs_get_block
18. reiserfs_readpages
19. real_space_diff
20. to_real_used_space
21. to_fake_used_blocks
22. init_inode
23. inode2sd
24. inode2sd_v1
25. update_stat_data
26. reiserfs_update_sd_size
27. reiserfs_make_bad_inode
28. reiserfs_init_locked_inode
29. reiserfs_read_locked_inode
30. reiserfs_find_actor
31. reiserfs_iget
32. reiserfs_get_dentry
33. reiserfs_fh_to_dentry
34. reiserfs_fh_to_parent
35. reiserfs_encode_fh
36. reiserfs_write_inode
37. reiserfs_new_directory
38. reiserfs_new_symlink
39. reiserfs_new_inode
40. grab_tail_page
41. reiserfs_truncate_file
42. map_block_for_writepage
43. reiserfs_write_full_page
44. reiserfs_readpage
45. reiserfs_writepage
46. reiserfs_truncate_failed_write
47. reiserfs_write_begin
48. __reiserfs_write_begin
49. reiserfs_aop_bmap
50. reiserfs_write_end
51. reiserfs_commit_write
52. sd_attrs_to_i_attrs
53. invalidatepage_can_drop
54. reiserfs_invalidatepage
55. reiserfs_set_page_dirty
56. reiserfs_releasepage
57. reiserfs_direct_IO
58. reiserfs_setattr

   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 "reiserfs.h"
   8 #include "acl.h"
   9 #include "xattr.h"
  10 #include <linux/exportfs.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/highmem.h>
  13 #include <linux/slab.h>
  14 #include <linux/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 #include <linux/uio.h>
  22 #include <linux/bio.h>
  23 
  24 int reiserfs_commit_write(struct file *f, struct page *page,
  25                           unsigned from, unsigned to);
  26 
  27 void reiserfs_evict_inode(struct inode *inode)
  28 {
  29         /*
  30          * We need blocks for transaction + (user+group) quota
  31          * update (possibly delete)
  32          */
  33         int jbegin_count =
  34             JOURNAL_PER_BALANCE_CNT * 2 +
  35             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
  36         struct reiserfs_transaction_handle th;
  37         int err;
  38 
  39         if (!inode->i_nlink && !is_bad_inode(inode))
  40                 dquot_initialize(inode);
  41 
  42         truncate_inode_pages_final(&inode->i_data);
  43         if (inode->i_nlink)
  44                 goto no_delete;
  45 
  46         /*
  47          * The = 0 happens when we abort creating a new inode
  48          * for some reason like lack of space..
  49          * also handles bad_inode case
  50          */
  51         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {
  52 
  53                 reiserfs_delete_xattrs(inode);
  54 
  55                 reiserfs_write_lock(inode->i_sb);
  56 
  57                 if (journal_begin(&th, inode->i_sb, jbegin_count))
  58                         goto out;
  59                 reiserfs_update_inode_transaction(inode);
  60 
  61                 reiserfs_discard_prealloc(&th, inode);
  62 
  63                 err = reiserfs_delete_object(&th, inode);
  64 
  65                 /*
  66                  * Do quota update inside a transaction for journaled quotas.
  67                  * We must do that after delete_object so that quota updates
  68                  * go into the same transaction as stat data deletion
  69                  */
  70                 if (!err) {
  71                         int depth = reiserfs_write_unlock_nested(inode->i_sb);
  72                         dquot_free_inode(inode);
  73                         reiserfs_write_lock_nested(inode->i_sb, depth);
  74                 }
  75 
  76                 if (journal_end(&th))
  77                         goto out;
  78 
  79                 /*
  80                  * check return value from reiserfs_delete_object after
  81                  * ending the transaction
  82                  */
  83                 if (err)
  84                     goto out;
  85 
  86                 /*
  87                  * all items of file are deleted, so we can remove
  88                  * "save" link
  89                  * we can't do anything about an error here
  90                  */
  91                 remove_save_link(inode, 0 /* not truncate */);
  92 out:
  93                 reiserfs_write_unlock(inode->i_sb);
  94         } else {
  95                 /* no object items are in the tree */
  96                 ;
  97         }
  98 
  99         /* note this must go after the journal_end to prevent deadlock */
 100         clear_inode(inode);
 101 
 102         dquot_drop(inode);
 103         inode->i_blocks = 0;
 104         return;
 105 
 106 no_delete:
 107         clear_inode(inode);
 108         dquot_drop(inode);
 109 }
 110 
 111 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
 112                           __u32 objectid, loff_t offset, int type, int length)
 113 {
 114         key->version = version;
 115 
 116         key->on_disk_key.k_dir_id = dirid;
 117         key->on_disk_key.k_objectid = objectid;
 118         set_cpu_key_k_offset(key, offset);
 119         set_cpu_key_k_type(key, type);
 120         key->key_length = length;
 121 }
 122 
 123 /*
 124  * take base of inode_key (it comes from inode always) (dirid, objectid)
 125  * and version from an inode, set offset and type of key
 126  */
 127 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
 128                   int type, int length)
 129 {
 130         _make_cpu_key(key, get_inode_item_key_version(inode),
 131                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
 132                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
 133                       length);
 134 }
 135 
 136 /* when key is 0, do not set version and short key */
 137 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
 138                               int version,
 139                               loff_t offset, int type, int length,
 140                               int entry_count /*or ih_free_space */ )
 141 {
 142         if (key) {
 143                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
 144                 ih->ih_key.k_objectid =
 145                     cpu_to_le32(key->on_disk_key.k_objectid);
 146         }
 147         put_ih_version(ih, version);
 148         set_le_ih_k_offset(ih, offset);
 149         set_le_ih_k_type(ih, type);
 150         put_ih_item_len(ih, length);
 151         /*    set_ih_free_space (ih, 0); */
 152         /*
 153          * for directory items it is entry count, for directs and stat
 154          * datas - 0xffff, for indirects - 0
 155          */
 156         put_ih_entry_count(ih, entry_count);
 157 }
 158 
 159 /*
 160  * FIXME: we might cache recently accessed indirect item
 161  * Ugh.  Not too eager for that....
 162  * I cut the code until such time as I see a convincing argument (benchmark).
 163  * I don't want a bloated inode struct..., and I don't like code complexity....
 164  */
 165 
 166 /*
 167  * cutting the code is fine, since it really isn't in use yet and is easy
 168  * to add back in.  But, Vladimir has a really good idea here.  Think
 169  * about what happens for reading a file.  For each page,
 170  * The VFS layer calls reiserfs_readpage, who searches the tree to find
 171  * an indirect item.  This indirect item has X number of pointers, where
 172  * X is a big number if we've done the block allocation right.  But,
 173  * we only use one or two of these pointers during each call to readpage,
 174  * needlessly researching again later on.
 175  *
 176  * The size of the cache could be dynamic based on the size of the file.
 177  *
 178  * I'd also like to see us cache the location the stat data item, since
 179  * we are needlessly researching for that frequently.
 180  *
 181  * --chris
 182  */
 183 
 184 /*
 185  * If this page has a file tail in it, and
 186  * it was read in by get_block_create_0, the page data is valid,
 187  * but tail is still sitting in a direct item, and we can't write to
 188  * it.  So, look through this page, and check all the mapped buffers
 189  * to make sure they have valid block numbers.  Any that don't need
 190  * to be unmapped, so that __block_write_begin will correctly call
 191  * reiserfs_get_block to convert the tail into an unformatted node
 192  */
 193 static inline void fix_tail_page_for_writing(struct page *page)
 194 {
 195         struct buffer_head *head, *next, *bh;
 196 
 197         if (page && page_has_buffers(page)) {
 198                 head = page_buffers(page);
 199                 bh = head;
 200                 do {
 201                         next = bh->b_this_page;
 202                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
 203                                 reiserfs_unmap_buffer(bh);
 204                         }
 205                         bh = next;
 206                 } while (bh != head);
 207         }
 208 }
 209 
 210 /*
 211  * reiserfs_get_block does not need to allocate a block only if it has been
 212  * done already or non-hole position has been found in the indirect item
 213  */
 214 static inline int allocation_needed(int retval, b_blocknr_t allocated,
 215                                     struct item_head *ih,
 216                                     __le32 * item, int pos_in_item)
 217 {
 218         if (allocated)
 219                 return 0;
 220         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
 221             get_block_num(item, pos_in_item))
 222                 return 0;
 223         return 1;
 224 }
 225 
 226 static inline int indirect_item_found(int retval, struct item_head *ih)
 227 {
 228         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
 229 }
 230 
 231 static inline void set_block_dev_mapped(struct buffer_head *bh,
 232                                         b_blocknr_t block, struct inode *inode)
 233 {
 234         map_bh(bh, inode->i_sb, block);
 235 }
 236 
 237 /*
 238  * files which were created in the earlier version can not be longer,
 239  * than 2 gb
 240  */
 241 static int file_capable(struct inode *inode, sector_t block)
 242 {
 243         /* it is new file. */
 244         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||
 245             /* old file, but 'block' is inside of 2gb */
 246             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))
 247                 return 1;
 248 
 249         return 0;
 250 }
 251 
 252 static int restart_transaction(struct reiserfs_transaction_handle *th,
 253                                struct inode *inode, struct treepath *path)
 254 {
 255         struct super_block *s = th->t_super;
 256         int err;
 257 
 258         BUG_ON(!th->t_trans_id);
 259         BUG_ON(!th->t_refcount);
 260 
 261         pathrelse(path);
 262 
 263         /* we cannot restart while nested */
 264         if (th->t_refcount > 1) {
 265                 return 0;
 266         }
 267         reiserfs_update_sd(th, inode);
 268         err = journal_end(th);
 269         if (!err) {
 270                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
 271                 if (!err)
 272                         reiserfs_update_inode_transaction(inode);
 273         }
 274         return err;
 275 }
 276 
 277 /*
 278  * it is called by get_block when create == 0. Returns block number
 279  * for 'block'-th logical block of file. When it hits direct item it
 280  * returns 0 (being called from bmap) or read direct item into piece
 281  * of page (bh_result)
 282  * Please improve the english/clarity in the comment above, as it is
 283  * hard to understand.
 284  */
 285 static int _get_block_create_0(struct inode *inode, sector_t block,
 286                                struct buffer_head *bh_result, int args)
 287 {
 288         INITIALIZE_PATH(path);
 289         struct cpu_key key;
 290         struct buffer_head *bh;
 291         struct item_head *ih, tmp_ih;
 292         b_blocknr_t blocknr;
 293         char *p = NULL;
 294         int chars;
 295         int ret;
 296         int result;
 297         int done = 0;
 298         unsigned long offset;
 299 
 300         /* prepare the key to look for the 'block'-th block of file */
 301         make_cpu_key(&key, inode,
 302                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
 303                      3);
 304 
 305         result = search_for_position_by_key(inode->i_sb, &key, &path);
 306         if (result != POSITION_FOUND) {
 307                 pathrelse(&path);
 308                 if (p)
 309                         kunmap(bh_result->b_page);
 310                 if (result == IO_ERROR)
 311                         return -EIO;
 312                 /*
 313                  * We do not return -ENOENT if there is a hole but page is
 314                  * uptodate, because it means that there is some MMAPED data
 315                  * associated with it that is yet to be written to disk.
 316                  */
 317                 if ((args & GET_BLOCK_NO_HOLE)
 318                     && !PageUptodate(bh_result->b_page)) {
 319                         return -ENOENT;
 320                 }
 321                 return 0;
 322         }
 323 
 324         bh = get_last_bh(&path);
 325         ih = tp_item_head(&path);
 326         if (is_indirect_le_ih(ih)) {
 327                 __le32 *ind_item = (__le32 *) ih_item_body(bh, ih);
 328 
 329                 /*
 330                  * FIXME: here we could cache indirect item or part of it in
 331                  * the inode to avoid search_by_key in case of subsequent
 332                  * access to file
 333                  */
 334                 blocknr = get_block_num(ind_item, path.pos_in_item);
 335                 ret = 0;
 336                 if (blocknr) {
 337                         map_bh(bh_result, inode->i_sb, blocknr);
 338                         if (path.pos_in_item ==
 339                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
 340                                 set_buffer_boundary(bh_result);
 341                         }
 342                 } else
 343                         /*
 344                          * We do not return -ENOENT if there is a hole but
 345                          * page is uptodate, because it means that there is
 346                          * some MMAPED data associated with it that is
 347                          * yet to be written to disk.
 348                          */
 349                 if ((args & GET_BLOCK_NO_HOLE)
 350                             && !PageUptodate(bh_result->b_page)) {
 351                         ret = -ENOENT;
 352                 }
 353 
 354                 pathrelse(&path);
 355                 if (p)
 356                         kunmap(bh_result->b_page);
 357                 return ret;
 358         }
 359         /* requested data are in direct item(s) */
 360         if (!(args & GET_BLOCK_READ_DIRECT)) {
 361                 /*
 362                  * we are called by bmap. FIXME: we can not map block of file
 363                  * when it is stored in direct item(s)
 364                  */
 365                 pathrelse(&path);
 366                 if (p)
 367                         kunmap(bh_result->b_page);
 368                 return -ENOENT;
 369         }
 370 
 371         /*
 372          * if we've got a direct item, and the buffer or page was uptodate,
 373          * we don't want to pull data off disk again.  skip to the
 374          * end, where we map the buffer and return
 375          */
 376         if (buffer_uptodate(bh_result)) {
 377                 goto finished;
 378         } else
 379                 /*
 380                  * grab_tail_page can trigger calls to reiserfs_get_block on
 381                  * up to date pages without any buffers.  If the page is up
 382                  * to date, we don't want read old data off disk.  Set the up
 383                  * to date bit on the buffer instead and jump to the end
 384                  */
 385         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
 386                 set_buffer_uptodate(bh_result);
 387                 goto finished;
 388         }
 389         /* read file tail into part of page */
 390         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_SIZE - 1);
 391         copy_item_head(&tmp_ih, ih);
 392 
 393         /*
 394          * we only want to kmap if we are reading the tail into the page.
 395          * this is not the common case, so we don't kmap until we are
 396          * sure we need to.  But, this means the item might move if
 397          * kmap schedules
 398          */
 399         if (!p)
 400                 p = (char *)kmap(bh_result->b_page);
 401 
 402         p += offset;
 403         memset(p, 0, inode->i_sb->s_blocksize);
 404         do {
 405                 if (!is_direct_le_ih(ih)) {
 406                         BUG();
 407                 }
 408                 /*
 409                  * make sure we don't read more bytes than actually exist in
 410                  * the file.  This can happen in odd cases where i_size isn't
 411                  * correct, and when direct item padding results in a few
 412                  * extra bytes at the end of the direct item
 413                  */
 414                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
 415                         break;
 416                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
 417                         chars =
 418                             inode->i_size - (le_ih_k_offset(ih) - 1) -
 419                             path.pos_in_item;
 420                         done = 1;
 421                 } else {
 422                         chars = ih_item_len(ih) - path.pos_in_item;
 423                 }
 424                 memcpy(p, ih_item_body(bh, ih) + path.pos_in_item, chars);
 425 
 426                 if (done)
 427                         break;
 428 
 429                 p += chars;
 430 
 431                 /*
 432                  * we done, if read direct item is not the last item of
 433                  * node FIXME: we could try to check right delimiting key
 434                  * to see whether direct item continues in the right
 435                  * neighbor or rely on i_size
 436                  */
 437                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
 438                         break;
 439 
 440                 /* update key to look for the next piece */
 441                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
 442                 result = search_for_position_by_key(inode->i_sb, &key, &path);
 443                 if (result != POSITION_FOUND)
 444                         /* i/o error most likely */
 445                         break;
 446                 bh = get_last_bh(&path);
 447                 ih = tp_item_head(&path);
 448         } while (1);
 449 
 450         flush_dcache_page(bh_result->b_page);
 451         kunmap(bh_result->b_page);
 452 
 453 finished:
 454         pathrelse(&path);
 455 
 456         if (result == IO_ERROR)
 457                 return -EIO;
 458 
 459         /*
 460          * this buffer has valid data, but isn't valid for io.  mapping it to
 461          * block #0 tells the rest of reiserfs it just has a tail in it
 462          */
 463         map_bh(bh_result, inode->i_sb, 0);
 464         set_buffer_uptodate(bh_result);
 465         return 0;
 466 }
 467 
 468 /*
 469  * this is called to create file map. So, _get_block_create_0 will not
 470  * read direct item
 471  */
 472 static int reiserfs_bmap(struct inode *inode, sector_t block,
 473                          struct buffer_head *bh_result, int create)
 474 {
 475         if (!file_capable(inode, block))
 476                 return -EFBIG;
 477 
 478         reiserfs_write_lock(inode->i_sb);
 479         /* do not read the direct item */
 480         _get_block_create_0(inode, block, bh_result, 0);
 481         reiserfs_write_unlock(inode->i_sb);
 482         return 0;
 483 }
 484 
 485 /*
 486  * special version of get_block that is only used by grab_tail_page right
 487  * now.  It is sent to __block_write_begin, and when you try to get a
 488  * block past the end of the file (or a block from a hole) it returns
 489  * -ENOENT instead of a valid buffer.  __block_write_begin expects to
 490  * be able to do i/o on the buffers returned, unless an error value
 491  * is also returned.
 492  *
 493  * So, this allows __block_write_begin to be used for reading a single block
 494  * in a page.  Where it does not produce a valid page for holes, or past the
 495  * end of the file.  This turns out to be exactly what we need for reading
 496  * tails for conversion.
 497  *
 498  * The point of the wrapper is forcing a certain value for create, even
 499  * though the VFS layer is calling this function with create==1.  If you
 500  * don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
 501  * don't use this function.
 502 */
 503 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
 504                                        struct buffer_head *bh_result,
 505                                        int create)
 506 {
 507         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
 508 }
 509 
 510 /*
 511  * This is special helper for reiserfs_get_block in case we are executing
 512  * direct_IO request.
 513  */
 514 static int reiserfs_get_blocks_direct_io(struct inode *inode,
 515                                          sector_t iblock,
 516                                          struct buffer_head *bh_result,
 517                                          int create)
 518 {
 519         int ret;
 520 
 521         bh_result->b_page = NULL;
 522 
 523         /*
 524          * We set the b_size before reiserfs_get_block call since it is
 525          * referenced in convert_tail_for_hole() that may be called from
 526          * reiserfs_get_block()
 527          */
 528         bh_result->b_size = i_blocksize(inode);
 529 
 530         ret = reiserfs_get_block(inode, iblock, bh_result,
 531                                  create | GET_BLOCK_NO_DANGLE);
 532         if (ret)
 533                 goto out;
 534 
 535         /* don't allow direct io onto tail pages */
 536         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
 537                 /*
 538                  * make sure future calls to the direct io funcs for this
 539                  * offset in the file fail by unmapping the buffer
 540                  */
 541                 clear_buffer_mapped(bh_result);
 542                 ret = -EINVAL;
 543         }
 544 
 545         /*
 546          * Possible unpacked tail. Flush the data before pages have
 547          * disappeared
 548          */
 549         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
 550                 int err;
 551 
 552                 reiserfs_write_lock(inode->i_sb);
 553 
 554                 err = reiserfs_commit_for_inode(inode);
 555                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 556 
 557                 reiserfs_write_unlock(inode->i_sb);
 558 
 559                 if (err < 0)
 560                         ret = err;
 561         }
 562 out:
 563         return ret;
 564 }
 565 
 566 /*
 567  * helper function for when reiserfs_get_block is called for a hole
 568  * but the file tail is still in a direct item
 569  * bh_result is the buffer head for the hole
 570  * tail_offset is the offset of the start of the tail in the file
 571  *
 572  * This calls prepare_write, which will start a new transaction
 573  * you should not be in a transaction, or have any paths held when you
 574  * call this.
 575  */
 576 static int convert_tail_for_hole(struct inode *inode,
 577                                  struct buffer_head *bh_result,
 578                                  loff_t tail_offset)
 579 {
 580         unsigned long index;
 581         unsigned long tail_end;
 582         unsigned long tail_start;
 583         struct page *tail_page;
 584         struct page *hole_page = bh_result->b_page;
 585         int retval = 0;
 586 
 587         if ((tail_offset & (bh_result->b_size - 1)) != 1)
 588                 return -EIO;
 589 
 590         /* always try to read until the end of the block */
 591         tail_start = tail_offset & (PAGE_SIZE - 1);
 592         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
 593 
 594         index = tail_offset >> PAGE_SHIFT;
 595         /*
 596          * hole_page can be zero in case of direct_io, we are sure
 597          * that we cannot get here if we write with O_DIRECT into tail page
 598          */
 599         if (!hole_page || index != hole_page->index) {
 600                 tail_page = grab_cache_page(inode->i_mapping, index);
 601                 retval = -ENOMEM;
 602                 if (!tail_page) {
 603                         goto out;
 604                 }
 605         } else {
 606                 tail_page = hole_page;
 607         }
 608 
 609         /*
 610          * we don't have to make sure the conversion did not happen while
 611          * we were locking the page because anyone that could convert
 612          * must first take i_mutex.
 613          *
 614          * We must fix the tail page for writing because it might have buffers
 615          * that are mapped, but have a block number of 0.  This indicates tail
 616          * data that has been read directly into the page, and
 617          * __block_write_begin won't trigger a get_block in this case.
 618          */
 619         fix_tail_page_for_writing(tail_page);
 620         retval = __reiserfs_write_begin(tail_page, tail_start,
 621                                       tail_end - tail_start);
 622         if (retval)
 623                 goto unlock;
 624 
 625         /* tail conversion might change the data in the page */
 626         flush_dcache_page(tail_page);
 627 
 628         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
 629 
 630 unlock:
 631         if (tail_page != hole_page) {
 632                 unlock_page(tail_page);
 633                 put_page(tail_page);
 634         }
 635 out:
 636         return retval;
 637 }
 638 
 639 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
 640                                   sector_t block,
 641                                   struct inode *inode,
 642                                   b_blocknr_t * allocated_block_nr,
 643                                   struct treepath *path, int flags)
 644 {
 645         BUG_ON(!th->t_trans_id);
 646 
 647 #ifdef REISERFS_PREALLOCATE
 648         if (!(flags & GET_BLOCK_NO_IMUX)) {
 649                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
 650                                                   path, block);
 651         }
 652 #endif
 653         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
 654                                          block);
 655 }
 656 
 657 int reiserfs_get_block(struct inode *inode, sector_t block,
 658                        struct buffer_head *bh_result, int create)
 659 {
 660         int repeat, retval = 0;
 661         /* b_blocknr_t is (unsigned) 32 bit int*/
 662         b_blocknr_t allocated_block_nr = 0;
 663         INITIALIZE_PATH(path);
 664         int pos_in_item;
 665         struct cpu_key key;
 666         struct buffer_head *bh, *unbh = NULL;
 667         struct item_head *ih, tmp_ih;
 668         __le32 *item;
 669         int done;
 670         int fs_gen;
 671         struct reiserfs_transaction_handle *th = NULL;
 672         /*
 673          * space reserved in transaction batch:
 674          * . 3 balancings in direct->indirect conversion
 675          * . 1 block involved into reiserfs_update_sd()
 676          * XXX in practically impossible worst case direct2indirect()
 677          * can incur (much) more than 3 balancings.
 678          * quota update for user, group
 679          */
 680         int jbegin_count =
 681             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 682             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
 683         int version;
 684         int dangle = 1;
 685         loff_t new_offset =
 686             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
 687 
 688         reiserfs_write_lock(inode->i_sb);
 689         version = get_inode_item_key_version(inode);
 690 
 691         if (!file_capable(inode, block)) {
 692                 reiserfs_write_unlock(inode->i_sb);
 693                 return -EFBIG;
 694         }
 695 
 696         /*
 697          * if !create, we aren't changing the FS, so we don't need to
 698          * log anything, so we don't need to start a transaction
 699          */
 700         if (!(create & GET_BLOCK_CREATE)) {
 701                 int ret;
 702                 /* find number of block-th logical block of the file */
 703                 ret = _get_block_create_0(inode, block, bh_result,
 704                                           create | GET_BLOCK_READ_DIRECT);
 705                 reiserfs_write_unlock(inode->i_sb);
 706                 return ret;
 707         }
 708 
 709         /*
 710          * if we're already in a transaction, make sure to close
 711          * any new transactions we start in this func
 712          */
 713         if ((create & GET_BLOCK_NO_DANGLE) ||
 714             reiserfs_transaction_running(inode->i_sb))
 715                 dangle = 0;
 716 
 717         /*
 718          * If file is of such a size, that it might have a tail and
 719          * tails are enabled  we should mark it as possibly needing
 720          * tail packing on close
 721          */
 722         if ((have_large_tails(inode->i_sb)
 723              && inode->i_size < i_block_size(inode) * 4)
 724             || (have_small_tails(inode->i_sb)
 725                 && inode->i_size < i_block_size(inode)))
 726                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 727 
 728         /* set the key of the first byte in the 'block'-th block of file */
 729         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
 730         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
 731 start_trans:
 732                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
 733                 if (!th) {
 734                         retval = -ENOMEM;
 735                         goto failure;
 736                 }
 737                 reiserfs_update_inode_transaction(inode);
 738         }
 739 research:
 740 
 741         retval = search_for_position_by_key(inode->i_sb, &key, &path);
 742         if (retval == IO_ERROR) {
 743                 retval = -EIO;
 744                 goto failure;
 745         }
 746 
 747         bh = get_last_bh(&path);
 748         ih = tp_item_head(&path);
 749         item = tp_item_body(&path);
 750         pos_in_item = path.pos_in_item;
 751 
 752         fs_gen = get_generation(inode->i_sb);
 753         copy_item_head(&tmp_ih, ih);
 754 
 755         if (allocation_needed
 756             (retval, allocated_block_nr, ih, item, pos_in_item)) {
 757                 /* we have to allocate block for the unformatted node */
 758                 if (!th) {
 759                         pathrelse(&path);
 760                         goto start_trans;
 761                 }
 762 
 763                 repeat =
 764                     _allocate_block(th, block, inode, &allocated_block_nr,
 765                                     &path, create);
 766 
 767                 /*
 768                  * restart the transaction to give the journal a chance to free
 769                  * some blocks.  releases the path, so we have to go back to
 770                  * research if we succeed on the second try
 771                  */
 772                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
 773                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
 774                         retval = restart_transaction(th, inode, &path);
 775                         if (retval)
 776                                 goto failure;
 777                         repeat =
 778                             _allocate_block(th, block, inode,
 779                                             &allocated_block_nr, NULL, create);
 780 
 781                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
 782                                 goto research;
 783                         }
 784                         if (repeat == QUOTA_EXCEEDED)
 785                                 retval = -EDQUOT;
 786                         else
 787                                 retval = -ENOSPC;
 788                         goto failure;
 789                 }
 790 
 791                 if (fs_changed(fs_gen, inode->i_sb)
 792                     && item_moved(&tmp_ih, &path)) {
 793                         goto research;
 794                 }
 795         }
 796 
 797         if (indirect_item_found(retval, ih)) {
 798                 b_blocknr_t unfm_ptr;
 799                 /*
 800                  * 'block'-th block is in the file already (there is
 801                  * corresponding cell in some indirect item). But it may be
 802                  * zero unformatted node pointer (hole)
 803                  */
 804                 unfm_ptr = get_block_num(item, pos_in_item);
 805                 if (unfm_ptr == 0) {
 806                         /* use allocated block to plug the hole */
 807                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
 808                         if (fs_changed(fs_gen, inode->i_sb)
 809                             && item_moved(&tmp_ih, &path)) {
 810                                 reiserfs_restore_prepared_buffer(inode->i_sb,
 811                                                                  bh);
 812                                 goto research;
 813                         }
 814                         set_buffer_new(bh_result);
 815                         if (buffer_dirty(bh_result)
 816                             && reiserfs_data_ordered(inode->i_sb))
 817                                 reiserfs_add_ordered_list(inode, bh_result);
 818                         put_block_num(item, pos_in_item, allocated_block_nr);
 819                         unfm_ptr = allocated_block_nr;
 820                         journal_mark_dirty(th, bh);
 821                         reiserfs_update_sd(th, inode);
 822                 }
 823                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
 824                 pathrelse(&path);
 825                 retval = 0;
 826                 if (!dangle && th)
 827                         retval = reiserfs_end_persistent_transaction(th);
 828 
 829                 reiserfs_write_unlock(inode->i_sb);
 830 
 831                 /*
 832                  * the item was found, so new blocks were not added to the file
 833                  * there is no need to make sure the inode is updated with this
 834                  * transaction
 835                  */
 836                 return retval;
 837         }
 838 
 839         if (!th) {
 840                 pathrelse(&path);
 841                 goto start_trans;
 842         }
 843 
 844         /*
 845          * desired position is not found or is in the direct item. We have
 846          * to append file with holes up to 'block'-th block converting
 847          * direct items to indirect one if necessary
 848          */
 849         done = 0;
 850         do {
 851                 if (is_statdata_le_ih(ih)) {
 852                         __le32 unp = 0;
 853                         struct cpu_key tmp_key;
 854 
 855                         /* indirect item has to be inserted */
 856                         make_le_item_head(&tmp_ih, &key, version, 1,
 857                                           TYPE_INDIRECT, UNFM_P_SIZE,
 858                                           0 /* free_space */ );
 859 
 860                         /*
 861                          * we are going to add 'block'-th block to the file.
 862                          * Use allocated block for that
 863                          */
 864                         if (cpu_key_k_offset(&key) == 1) {
 865                                 unp = cpu_to_le32(allocated_block_nr);
 866                                 set_block_dev_mapped(bh_result,
 867                                                      allocated_block_nr, inode);
 868                                 set_buffer_new(bh_result);
 869                                 done = 1;
 870                         }
 871                         tmp_key = key;  /* ;) */
 872                         set_cpu_key_k_offset(&tmp_key, 1);
 873                         PATH_LAST_POSITION(&path)++;
 874 
 875                         retval =
 876                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
 877                                                  inode, (char *)&unp);
 878                         if (retval) {
 879                                 reiserfs_free_block(th, inode,
 880                                                     allocated_block_nr, 1);
 881                                 /*
 882                                  * retval == -ENOSPC, -EDQUOT or -EIO
 883                                  * or -EEXIST
 884                                  */
 885                                 goto failure;
 886                         }
 887                 } else if (is_direct_le_ih(ih)) {
 888                         /* direct item has to be converted */
 889                         loff_t tail_offset;
 890 
 891                         tail_offset =
 892                             ((le_ih_k_offset(ih) -
 893                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
 894 
 895                         /*
 896                          * direct item we just found fits into block we have
 897                          * to map. Convert it into unformatted node: use
 898                          * bh_result for the conversion
 899                          */
 900                         if (tail_offset == cpu_key_k_offset(&key)) {
 901                                 set_block_dev_mapped(bh_result,
 902                                                      allocated_block_nr, inode);
 903                                 unbh = bh_result;
 904                                 done = 1;
 905                         } else {
 906                                 /*
 907                                  * we have to pad file tail stored in direct
 908                                  * item(s) up to block size and convert it
 909                                  * to unformatted node. FIXME: this should
 910                                  * also get into page cache
 911                                  */
 912 
 913                                 pathrelse(&path);
 914                                 /*
 915                                  * ugly, but we can only end the transaction if
 916                                  * we aren't nested
 917                                  */
 918                                 BUG_ON(!th->t_refcount);
 919                                 if (th->t_refcount == 1) {
 920                                         retval =
 921                                             reiserfs_end_persistent_transaction
 922                                             (th);
 923                                         th = NULL;
 924                                         if (retval)
 925                                                 goto failure;
 926                                 }
 927 
 928                                 retval =
 929                                     convert_tail_for_hole(inode, bh_result,
 930                                                           tail_offset);
 931                                 if (retval) {
 932                                         if (retval != -ENOSPC)
 933                                                 reiserfs_error(inode->i_sb,
 934                                                         "clm-6004",
 935                                                         "convert tail failed "
 936                                                         "inode %lu, error %d",
 937                                                         inode->i_ino,
 938                                                         retval);
 939                                         if (allocated_block_nr) {
 940                                                 /*
 941                                                  * the bitmap, the super,
 942                                                  * and the stat data == 3
 943                                                  */
 944                                                 if (!th)
 945                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
 946                                                 if (th)
 947                                                         reiserfs_free_block(th,
 948                                                                             inode,
 949                                                                             allocated_block_nr,
 950                                                                             1);
 951                                         }
 952                                         goto failure;
 953                                 }
 954                                 goto research;
 955                         }
 956                         retval =
 957                             direct2indirect(th, inode, &path, unbh,
 958                                             tail_offset);
 959                         if (retval) {
 960                                 reiserfs_unmap_buffer(unbh);
 961                                 reiserfs_free_block(th, inode,
 962                                                     allocated_block_nr, 1);
 963                                 goto failure;
 964                         }
 965                         /*
 966                          * it is important the set_buffer_uptodate is done
 967                          * after the direct2indirect.  The buffer might
 968                          * contain valid data newer than the data on disk
 969                          * (read by readpage, changed, and then sent here by
 970                          * writepage).  direct2indirect needs to know if unbh
 971                          * was already up to date, so it can decide if the
 972                          * data in unbh needs to be replaced with data from
 973                          * the disk
 974                          */
 975                         set_buffer_uptodate(unbh);
 976 
 977                         /*
 978                          * unbh->b_page == NULL in case of DIRECT_IO request,
 979                          * this means buffer will disappear shortly, so it
 980                          * should not be added to
 981                          */
 982                         if (unbh->b_page) {
 983                                 /*
 984                                  * we've converted the tail, so we must
 985                                  * flush unbh before the transaction commits
 986                                  */
 987                                 reiserfs_add_tail_list(inode, unbh);
 988 
 989                                 /*
 990                                  * mark it dirty now to prevent commit_write
 991                                  * from adding this buffer to the inode's
 992                                  * dirty buffer list
 993                                  */
 994                                 /*
 995                                  * AKPM: changed __mark_buffer_dirty to
 996                                  * mark_buffer_dirty().  It's still atomic,
 997                                  * but it sets the page dirty too, which makes
 998                                  * it eligible for writeback at any time by the
 999                                  * VM (which was also the case with
1000                                  * __mark_buffer_dirty())
1001                                  */
1002                                 mark_buffer_dirty(unbh);
1003                         }
1004                 } else {
1005                         /*
1006                          * append indirect item with holes if needed, when
1007                          * appending pointer to 'block'-th block use block,
1008                          * which is already allocated
1009                          */
1010                         struct cpu_key tmp_key;
1011                         /*
1012                          * We use this in case we need to allocate
1013                          * only one block which is a fastpath
1014                          */
1015                         unp_t unf_single = 0;
1016                         unp_t *un;
1017                         __u64 max_to_insert =
1018                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
1019                             UNFM_P_SIZE;
1020                         __u64 blocks_needed;
1021 
1022                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
1023                                "vs-804: invalid position for append");
1024                         /*
1025                          * indirect item has to be appended,
1026                          * set up key of that position
1027                          * (key type is unimportant)
1028                          */
1029                         make_cpu_key(&tmp_key, inode,
1030                                      le_key_k_offset(version,
1031                                                      &ih->ih_key) +
1032                                      op_bytes_number(ih,
1033                                                      inode->i_sb->s_blocksize),
1034                                      TYPE_INDIRECT, 3);
1035 
1036                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
1037                                "green-805: invalid offset");
1038                         blocks_needed =
1039                             1 +
1040                             ((cpu_key_k_offset(&key) -
1041                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
1042                              s_blocksize_bits);
1043 
1044                         if (blocks_needed == 1) {
1045                                 un = &unf_single;
1046                         } else {
1047                                 un = kcalloc(min(blocks_needed, max_to_insert),
1048                                              UNFM_P_SIZE, GFP_NOFS);
1049                                 if (!un) {
1050                                         un = &unf_single;
1051                                         blocks_needed = 1;
1052                                         max_to_insert = 0;
1053                                 }
1054                         }
1055                         if (blocks_needed <= max_to_insert) {
1056                                 /*
1057                                  * we are going to add target block to
1058                                  * the file. Use allocated block for that
1059                                  */
1060                                 un[blocks_needed - 1] =
1061                                     cpu_to_le32(allocated_block_nr);
1062                                 set_block_dev_mapped(bh_result,
1063                                                      allocated_block_nr, inode);
1064                                 set_buffer_new(bh_result);
1065                                 done = 1;
1066                         } else {
1067                                 /* paste hole to the indirect item */
1068                                 /*
1069                                  * If kmalloc failed, max_to_insert becomes
1070                                  * zero and it means we only have space for
1071                                  * one block
1072                                  */
1073                                 blocks_needed =
1074                                     max_to_insert ? max_to_insert : 1;
1075                         }
1076                         retval =
1077                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
1078                                                      (char *)un,
1079                                                      UNFM_P_SIZE *
1080                                                      blocks_needed);
1081 
1082                         if (blocks_needed != 1)
1083                                 kfree(un);
1084 
1085                         if (retval) {
1086                                 reiserfs_free_block(th, inode,
1087                                                     allocated_block_nr, 1);
1088                                 goto failure;
1089                         }
1090                         if (!done) {
1091                                 /*
1092                                  * We need to mark new file size in case
1093                                  * this function will be interrupted/aborted
1094                                  * later on. And we may do this only for
1095                                  * holes.
1096                                  */
1097                                 inode->i_size +=
1098                                     inode->i_sb->s_blocksize * blocks_needed;
1099                         }
1100                 }
1101 
1102                 if (done == 1)
1103                         break;
1104 
1105                 /*
1106                  * this loop could log more blocks than we had originally
1107                  * asked for.  So, we have to allow the transaction to end
1108                  * if it is too big or too full.  Update the inode so things
1109                  * are consistent if we crash before the function returns
1110                  * release the path so that anybody waiting on the path before
1111                  * ending their transaction will be able to continue.
1112                  */
1113                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1114                         retval = restart_transaction(th, inode, &path);
1115                         if (retval)
1116                                 goto failure;
1117                 }
1118                 /*
1119                  * inserting indirect pointers for a hole can take a
1120                  * long time.  reschedule if needed and also release the write
1121                  * lock for others.
1122                  */
1123                 reiserfs_cond_resched(inode->i_sb);
1124 
1125                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1126                 if (retval == IO_ERROR) {
1127                         retval = -EIO;
1128                         goto failure;
1129                 }
1130                 if (retval == POSITION_FOUND) {
1131                         reiserfs_warning(inode->i_sb, "vs-825",
1132                                          "%K should not be found", &key);
1133                         retval = -EEXIST;
1134                         if (allocated_block_nr)
1135                                 reiserfs_free_block(th, inode,
1136                                                     allocated_block_nr, 1);
1137                         pathrelse(&path);
1138                         goto failure;
1139                 }
1140                 bh = get_last_bh(&path);
1141                 ih = tp_item_head(&path);
1142                 item = tp_item_body(&path);
1143                 pos_in_item = path.pos_in_item;
1144         } while (1);
1145 
1146         retval = 0;
1147 
1148 failure:
1149         if (th && (!dangle || (retval && !th->t_trans_id))) {
1150                 int err;
1151                 if (th->t_trans_id)
1152                         reiserfs_update_sd(th, inode);
1153                 err = reiserfs_end_persistent_transaction(th);
1154                 if (err)
1155                         retval = err;
1156         }
1157 
1158         reiserfs_write_unlock(inode->i_sb);
1159         reiserfs_check_path(&path);
1160         return retval;
1161 }
1162 
1163 static int
1164 reiserfs_readpages(struct file *file, struct address_space *mapping,
1165                    struct list_head *pages, unsigned nr_pages)
1166 {
1167         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1168 }
1169 
1170 /*
1171  * Compute real number of used bytes by file
1172  * Following three functions can go away when we'll have enough space in
1173  * stat item
1174  */
1175 static int real_space_diff(struct inode *inode, int sd_size)
1176 {
1177         int bytes;
1178         loff_t blocksize = inode->i_sb->s_blocksize;
1179 
1180         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1181                 return sd_size;
1182 
1183         /*
1184          * End of file is also in full block with indirect reference, so round
1185          * up to the next block.
1186          *
1187          * there is just no way to know if the tail is actually packed
1188          * on the file, so we have to assume it isn't.  When we pack the
1189          * tail, we add 4 bytes to pretend there really is an unformatted
1190          * node pointer
1191          */
1192         bytes =
1193             ((inode->i_size +
1194               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1195             sd_size;
1196         return bytes;
1197 }
1198 
1199 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1200                                         int sd_size)
1201 {
1202         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1203                 return inode->i_size +
1204                     (loff_t) (real_space_diff(inode, sd_size));
1205         }
1206         return ((loff_t) real_space_diff(inode, sd_size)) +
1207             (((loff_t) blocks) << 9);
1208 }
1209 
1210 /* Compute number of blocks used by file in ReiserFS counting */
1211 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1212 {
1213         loff_t bytes = inode_get_bytes(inode);
1214         loff_t real_space = real_space_diff(inode, sd_size);
1215 
1216         /* keeps fsck and non-quota versions of reiserfs happy */
1217         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1218                 bytes += (loff_t) 511;
1219         }
1220 
1221         /*
1222          * files from before the quota patch might i_blocks such that
1223          * bytes < real_space.  Deal with that here to prevent it from
1224          * going negative.
1225          */
1226         if (bytes < real_space)
1227                 return 0;
1228         return (bytes - real_space) >> 9;
1229 }
1230 
1231 /*
1232  * BAD: new directories have stat data of new type and all other items
1233  * of old type. Version stored in the inode says about body items, so
1234  * in update_stat_data we can not rely on inode, but have to check
1235  * item version directly
1236  */
1237 
1238 /* called by read_locked_inode */
1239 static void init_inode(struct inode *inode, struct treepath *path)
1240 {
1241         struct buffer_head *bh;
1242         struct item_head *ih;
1243         __u32 rdev;
1244 
1245         bh = PATH_PLAST_BUFFER(path);
1246         ih = tp_item_head(path);
1247 
1248         copy_key(INODE_PKEY(inode), &ih->ih_key);
1249 
1250         INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
1251         REISERFS_I(inode)->i_flags = 0;
1252         REISERFS_I(inode)->i_prealloc_block = 0;
1253         REISERFS_I(inode)->i_prealloc_count = 0;
1254         REISERFS_I(inode)->i_trans_id = 0;
1255         REISERFS_I(inode)->i_jl = NULL;
1256         reiserfs_init_xattr_rwsem(inode);
1257 
1258         if (stat_data_v1(ih)) {
1259                 struct stat_data_v1 *sd =
1260                     (struct stat_data_v1 *)ih_item_body(bh, ih);
1261                 unsigned long blocks;
1262 
1263                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1264                 set_inode_sd_version(inode, STAT_DATA_V1);
1265                 inode->i_mode = sd_v1_mode(sd);
1266                 set_nlink(inode, sd_v1_nlink(sd));
1267                 i_uid_write(inode, sd_v1_uid(sd));
1268                 i_gid_write(inode, sd_v1_gid(sd));
1269                 inode->i_size = sd_v1_size(sd);
1270                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1271                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1272                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1273                 inode->i_atime.tv_nsec = 0;
1274                 inode->i_ctime.tv_nsec = 0;
1275                 inode->i_mtime.tv_nsec = 0;
1276 
1277                 inode->i_blocks = sd_v1_blocks(sd);
1278                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1279                 blocks = (inode->i_size + 511) >> 9;
1280                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1281 
1282                 /*
1283                  * there was a bug in <=3.5.23 when i_blocks could take
1284                  * negative values. Starting from 3.5.17 this value could
1285                  * even be stored in stat data. For such files we set
1286                  * i_blocks based on file size. Just 2 notes: this can be
1287                  * wrong for sparse files. On-disk value will be only
1288                  * updated if file's inode will ever change
1289                  */
1290                 if (inode->i_blocks > blocks) {
1291                         inode->i_blocks = blocks;
1292                 }
1293 
1294                 rdev = sd_v1_rdev(sd);
1295                 REISERFS_I(inode)->i_first_direct_byte =
1296                     sd_v1_first_direct_byte(sd);
1297 
1298                 /*
1299                  * an early bug in the quota code can give us an odd
1300                  * number for the block count.  This is incorrect, fix it here.
1301                  */
1302                 if (inode->i_blocks & 1) {
1303                         inode->i_blocks++;
1304                 }
1305                 inode_set_bytes(inode,
1306                                 to_real_used_space(inode, inode->i_blocks,
1307                                                    SD_V1_SIZE));
1308                 /*
1309                  * nopack is initially zero for v1 objects. For v2 objects,
1310                  * nopack is initialised from sd_attrs
1311                  */
1312                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1313         } else {
1314                 /*
1315                  * new stat data found, but object may have old items
1316                  * (directories and symlinks)
1317                  */
1318                 struct stat_data *sd = (struct stat_data *)ih_item_body(bh, ih);
1319 
1320                 inode->i_mode = sd_v2_mode(sd);
1321                 set_nlink(inode, sd_v2_nlink(sd));
1322                 i_uid_write(inode, sd_v2_uid(sd));
1323                 inode->i_size = sd_v2_size(sd);
1324                 i_gid_write(inode, sd_v2_gid(sd));
1325                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1326                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1327                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1328                 inode->i_ctime.tv_nsec = 0;
1329                 inode->i_mtime.tv_nsec = 0;
1330                 inode->i_atime.tv_nsec = 0;
1331                 inode->i_blocks = sd_v2_blocks(sd);
1332                 rdev = sd_v2_rdev(sd);
1333                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1334                         inode->i_generation =
1335                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1336                 else
1337                         inode->i_generation = sd_v2_generation(sd);
1338 
1339                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1340                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1341                 else
1342                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1343                 REISERFS_I(inode)->i_first_direct_byte = 0;
1344                 set_inode_sd_version(inode, STAT_DATA_V2);
1345                 inode_set_bytes(inode,
1346                                 to_real_used_space(inode, inode->i_blocks,
1347                                                    SD_V2_SIZE));
1348                 /*
1349                  * read persistent inode attributes from sd and initialise
1350                  * generic inode flags from them
1351                  */
1352                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1353                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1354         }
1355 
1356         pathrelse(path);
1357         if (S_ISREG(inode->i_mode)) {
1358                 inode->i_op = &reiserfs_file_inode_operations;
1359                 inode->i_fop = &reiserfs_file_operations;
1360                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1361         } else if (S_ISDIR(inode->i_mode)) {
1362                 inode->i_op = &reiserfs_dir_inode_operations;
1363                 inode->i_fop = &reiserfs_dir_operations;
1364         } else if (S_ISLNK(inode->i_mode)) {
1365                 inode->i_op = &reiserfs_symlink_inode_operations;
1366                 inode_nohighmem(inode);
1367                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1368         } else {
1369                 inode->i_blocks = 0;
1370                 inode->i_op = &reiserfs_special_inode_operations;
1371                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1372         }
1373 }
1374 
1375 /* update new stat data with inode fields */
1376 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1377 {
1378         struct stat_data *sd_v2 = (struct stat_data *)sd;
1379 
1380         set_sd_v2_mode(sd_v2, inode->i_mode);
1381         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1382         set_sd_v2_uid(sd_v2, i_uid_read(inode));
1383         set_sd_v2_size(sd_v2, size);
1384         set_sd_v2_gid(sd_v2, i_gid_read(inode));
1385         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1386         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1387         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1388         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1389         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1390                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1391         else
1392                 set_sd_v2_generation(sd_v2, inode->i_generation);
1393         set_sd_v2_attrs(sd_v2, REISERFS_I(inode)->i_attrs);
1394 }
1395 
1396 /* used to copy inode's fields to old stat data */
1397 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1398 {
1399         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1400 
1401         set_sd_v1_mode(sd_v1, inode->i_mode);
1402         set_sd_v1_uid(sd_v1, i_uid_read(inode));
1403         set_sd_v1_gid(sd_v1, i_gid_read(inode));
1404         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1405         set_sd_v1_size(sd_v1, size);
1406         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1407         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1408         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1409 
1410         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1411                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1412         else
1413                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1414 
1415         /* Sigh. i_first_direct_byte is back */
1416         set_sd_v1_first_direct_byte(sd_v1,
1417                                     REISERFS_I(inode)->i_first_direct_byte);
1418 }
1419 
1420 /*
1421  * NOTE, you must prepare the buffer head before sending it here,
1422  * and then log it after the call
1423  */
1424 static void update_stat_data(struct treepath *path, struct inode *inode,
1425                              loff_t size)
1426 {
1427         struct buffer_head *bh;
1428         struct item_head *ih;
1429 
1430         bh = PATH_PLAST_BUFFER(path);
1431         ih = tp_item_head(path);
1432 
1433         if (!is_statdata_le_ih(ih))
1434                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1435                                INODE_PKEY(inode), ih);
1436 
1437         /* path points to old stat data */
1438         if (stat_data_v1(ih)) {
1439                 inode2sd_v1(ih_item_body(bh, ih), inode, size);
1440         } else {
1441                 inode2sd(ih_item_body(bh, ih), inode, size);
1442         }
1443 
1444         return;
1445 }
1446 
1447 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1448                              struct inode *inode, loff_t size)
1449 {
1450         struct cpu_key key;
1451         INITIALIZE_PATH(path);
1452         struct buffer_head *bh;
1453         int fs_gen;
1454         struct item_head *ih, tmp_ih;
1455         int retval;
1456 
1457         BUG_ON(!th->t_trans_id);
1458 
1459         /* key type is unimportant */
1460         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1461 
1462         for (;;) {
1463                 int pos;
1464                 /* look for the object's stat data */
1465                 retval = search_item(inode->i_sb, &key, &path);
1466                 if (retval == IO_ERROR) {
1467                         reiserfs_error(inode->i_sb, "vs-13050",
1468                                        "i/o failure occurred trying to "
1469                                        "update %K stat data", &key);
1470                         return;
1471                 }
1472                 if (retval == ITEM_NOT_FOUND) {
1473                         pos = PATH_LAST_POSITION(&path);
1474                         pathrelse(&path);
1475                         if (inode->i_nlink == 0) {
1476                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1477                                 return;
1478                         }
1479                         reiserfs_warning(inode->i_sb, "vs-13060",
1480                                          "stat data of object %k (nlink == %d) "
1481                                          "not found (pos %d)",
1482                                          INODE_PKEY(inode), inode->i_nlink,
1483                                          pos);
1484                         reiserfs_check_path(&path);
1485                         return;
1486                 }
1487 
1488                 /*
1489                  * sigh, prepare_for_journal might schedule.  When it
1490                  * schedules the FS might change.  We have to detect that,
1491                  * and loop back to the search if the stat data item has moved
1492                  */
1493                 bh = get_last_bh(&path);
1494                 ih = tp_item_head(&path);
1495                 copy_item_head(&tmp_ih, ih);
1496                 fs_gen = get_generation(inode->i_sb);
1497                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1498 
1499                 /* Stat_data item has been moved after scheduling. */
1500                 if (fs_changed(fs_gen, inode->i_sb)
1501                     && item_moved(&tmp_ih, &path)) {
1502                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1503                         continue;
1504                 }
1505                 break;
1506         }
1507         update_stat_data(&path, inode, size);
1508         journal_mark_dirty(th, bh);
1509         pathrelse(&path);
1510         return;
1511 }
1512 
1513 /*
1514  * reiserfs_read_locked_inode is called to read the inode off disk, and it
1515  * does a make_bad_inode when things go wrong.  But, we need to make sure
1516  * and clear the key in the private portion of the inode, otherwise a
1517  * corresponding iput might try to delete whatever object the inode last
1518  * represented.
1519  */
1520 static void reiserfs_make_bad_inode(struct inode *inode)
1521 {
1522         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1523         make_bad_inode(inode);
1524 }
1525 
1526 /*
1527  * initially this function was derived from minix or ext2's analog and
1528  * evolved as the prototype did
1529  */
1530 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1531 {
1532         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1533         inode->i_ino = args->objectid;
1534         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1535         return 0;
1536 }
1537 
1538 /*
1539  * looks for stat data in the tree, and fills up the fields of in-core
1540  * inode stat data fields
1541  */
1542 void reiserfs_read_locked_inode(struct inode *inode,
1543                                 struct reiserfs_iget_args *args)
1544 {
1545         INITIALIZE_PATH(path_to_sd);
1546         struct cpu_key key;
1547         unsigned long dirino;
1548         int retval;
1549 
1550         dirino = args->dirid;
1551 
1552         /*
1553          * set version 1, version 2 could be used too, because stat data
1554          * key is the same in both versions
1555          */
1556         key.version = KEY_FORMAT_3_5;
1557         key.on_disk_key.k_dir_id = dirino;
1558         key.on_disk_key.k_objectid = inode->i_ino;
1559         key.on_disk_key.k_offset = 0;
1560         key.on_disk_key.k_type = 0;
1561 
1562         /* look for the object's stat data */
1563         retval = search_item(inode->i_sb, &key, &path_to_sd);
1564         if (retval == IO_ERROR) {
1565                 reiserfs_error(inode->i_sb, "vs-13070",
1566                                "i/o failure occurred trying to find "
1567                                "stat data of %K", &key);
1568                 reiserfs_make_bad_inode(inode);
1569                 return;
1570         }
1571 
1572         /* a stale NFS handle can trigger this without it being an error */
1573         if (retval != ITEM_FOUND) {
1574                 pathrelse(&path_to_sd);
1575                 reiserfs_make_bad_inode(inode);
1576                 clear_nlink(inode);
1577                 return;
1578         }
1579 
1580         init_inode(inode, &path_to_sd);
1581 
1582         /*
1583          * It is possible that knfsd is trying to access inode of a file
1584          * that is being removed from the disk by some other thread. As we
1585          * update sd on unlink all that is required is to check for nlink
1586          * here. This bug was first found by Sizif when debugging
1587          * SquidNG/Butterfly, forgotten, and found again after Philippe
1588          * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1589 
1590          * More logical fix would require changes in fs/inode.c:iput() to
1591          * remove inode from hash-table _after_ fs cleaned disk stuff up and
1592          * in iget() to return NULL if I_FREEING inode is found in
1593          * hash-table.
1594          */
1595 
1596         /*
1597          * Currently there is one place where it's ok to meet inode with
1598          * nlink==0: processing of open-unlinked and half-truncated files
1599          * during mount (fs/reiserfs/super.c:finish_unfinished()).
1600          */
1601         if ((inode->i_nlink == 0) &&
1602             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1603                 reiserfs_warning(inode->i_sb, "vs-13075",
1604                                  "dead inode read from disk %K. "
1605                                  "This is likely to be race with knfsd. Ignore",
1606                                  &key);
1607                 reiserfs_make_bad_inode(inode);
1608         }
1609 
1610         /* init inode should be relsing */
1611         reiserfs_check_path(&path_to_sd);
1612 
1613         /*
1614          * Stat data v1 doesn't support ACLs.
1615          */
1616         if (get_inode_sd_version(inode) == STAT_DATA_V1)
1617                 cache_no_acl(inode);
1618 }
1619 
1620 /*
1621  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1622  *
1623  * @inode:    inode from hash table to check
1624  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1625  *
1626  * This function is called by iget5_locked() to distinguish reiserfs inodes
1627  * having the same inode numbers. Such inodes can only exist due to some
1628  * error condition. One of them should be bad. Inodes with identical
1629  * inode numbers (objectids) are distinguished by parent directory ids.
1630  *
1631  */
1632 int reiserfs_find_actor(struct inode *inode, void *opaque)
1633 {
1634         struct reiserfs_iget_args *args;
1635 
1636         args = opaque;
1637         /* args is already in CPU order */
1638         return (inode->i_ino == args->objectid) &&
1639             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1640 }
1641 
1642 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1643 {
1644         struct inode *inode;
1645         struct reiserfs_iget_args args;
1646         int depth;
1647 
1648         args.objectid = key->on_disk_key.k_objectid;
1649         args.dirid = key->on_disk_key.k_dir_id;
1650         depth = reiserfs_write_unlock_nested(s);
1651         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1652                              reiserfs_find_actor, reiserfs_init_locked_inode,
1653                              (void *)(&args));
1654         reiserfs_write_lock_nested(s, depth);
1655         if (!inode)
1656                 return ERR_PTR(-ENOMEM);
1657 
1658         if (inode->i_state & I_NEW) {
1659                 reiserfs_read_locked_inode(inode, &args);
1660                 unlock_new_inode(inode);
1661         }
1662 
1663         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1664                 /* either due to i/o error or a stale NFS handle */
1665                 iput(inode);
1666                 inode = NULL;
1667         }
1668         return inode;
1669 }
1670 
1671 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1672         u32 objectid, u32 dir_id, u32 generation)
1673 
1674 {
1675         struct cpu_key key;
1676         struct inode *inode;
1677 
1678         key.on_disk_key.k_objectid = objectid;
1679         key.on_disk_key.k_dir_id = dir_id;
1680         reiserfs_write_lock(sb);
1681         inode = reiserfs_iget(sb, &key);
1682         if (inode && !IS_ERR(inode) && generation != 0 &&
1683             generation != inode->i_generation) {
1684                 iput(inode);
1685                 inode = NULL;
1686         }
1687         reiserfs_write_unlock(sb);
1688 
1689         return d_obtain_alias(inode);
1690 }
1691 
1692 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1693                 int fh_len, int fh_type)
1694 {
1695         /*
1696          * fhtype happens to reflect the number of u32s encoded.
1697          * due to a bug in earlier code, fhtype might indicate there
1698          * are more u32s then actually fitted.
1699          * so if fhtype seems to be more than len, reduce fhtype.
1700          * Valid types are:
1701          *   2 - objectid + dir_id - legacy support
1702          *   3 - objectid + dir_id + generation
1703          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1704          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1705          *   6 - as above plus generation of directory
1706          * 6 does not fit in NFSv2 handles
1707          */
1708         if (fh_type > fh_len) {
1709                 if (fh_type != 6 || fh_len != 5)
1710                         reiserfs_warning(sb, "reiserfs-13077",
1711                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1712                                 fh_type, fh_len);
1713                 fh_type = fh_len;
1714         }
1715         if (fh_len < 2)
1716                 return NULL;
1717 
1718         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1719                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1720 }
1721 
1722 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1723                 int fh_len, int fh_type)
1724 {
1725         if (fh_type > fh_len)
1726                 fh_type = fh_len;
1727         if (fh_type < 4)
1728                 return NULL;
1729 
1730         return reiserfs_get_dentry(sb,
1731                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1732                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1733                 (fh_type == 6) ? fid->raw[5] : 0);
1734 }
1735 
1736 int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1737                        struct inode *parent)
1738 {
1739         int maxlen = *lenp;
1740 
1741         if (parent && (maxlen < 5)) {
1742                 *lenp = 5;
1743                 return FILEID_INVALID;
1744         } else if (maxlen < 3) {
1745                 *lenp = 3;
1746                 return FILEID_INVALID;
1747         }
1748 
1749         data[0] = inode->i_ino;
1750         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1751         data[2] = inode->i_generation;
1752         *lenp = 3;
1753         if (parent) {
1754                 data[3] = parent->i_ino;
1755                 data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1756                 *lenp = 5;
1757                 if (maxlen >= 6) {
1758                         data[5] = parent->i_generation;
1759                         *lenp = 6;
1760                 }
1761         }
1762         return *lenp;
1763 }
1764 
1765 /*
1766  * looks for stat data, then copies fields to it, marks the buffer
1767  * containing stat data as dirty
1768  */
1769 /*
1770  * reiserfs inodes are never really dirty, since the dirty inode call
1771  * always logs them.  This call allows the VFS inode marking routines
1772  * to properly mark inodes for datasync and such, but only actually
1773  * does something when called for a synchronous update.
1774  */
1775 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1776 {
1777         struct reiserfs_transaction_handle th;
1778         int jbegin_count = 1;
1779 
1780         if (sb_rdonly(inode->i_sb))
1781                 return -EROFS;
1782         /*
1783          * memory pressure can sometimes initiate write_inode calls with
1784          * sync == 1,
1785          * these cases are just when the system needs ram, not when the
1786          * inode needs to reach disk for safety, and they can safely be
1787          * ignored because the altered inode has already been logged.
1788          */
1789         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1790                 reiserfs_write_lock(inode->i_sb);
1791                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1792                         reiserfs_update_sd(&th, inode);
1793                         journal_end_sync(&th);
1794                 }
1795                 reiserfs_write_unlock(inode->i_sb);
1796         }
1797         return 0;
1798 }
1799 
1800 /*
1801  * stat data of new object is inserted already, this inserts the item
1802  * containing "." and ".." entries
1803  */
1804 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1805                                   struct inode *inode,
1806                                   struct item_head *ih, struct treepath *path,
1807                                   struct inode *dir)
1808 {
1809         struct super_block *sb = th->t_super;
1810         char empty_dir[EMPTY_DIR_SIZE];
1811         char *body = empty_dir;
1812         struct cpu_key key;
1813         int retval;
1814 
1815         BUG_ON(!th->t_trans_id);
1816 
1817         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1818                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1819                       TYPE_DIRENTRY, 3 /*key length */ );
1820 
1821         /*
1822          * compose item head for new item. Directories consist of items of
1823          * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1824          * is done by reiserfs_new_inode
1825          */
1826         if (old_format_only(sb)) {
1827                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1828                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1829 
1830                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1831                                        ih->ih_key.k_objectid,
1832                                        INODE_PKEY(dir)->k_dir_id,
1833                                        INODE_PKEY(dir)->k_objectid);
1834         } else {
1835                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1836                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1837 
1838                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1839                                     ih->ih_key.k_objectid,
1840                                     INODE_PKEY(dir)->k_dir_id,
1841                                     INODE_PKEY(dir)->k_objectid);
1842         }
1843 
1844         /* look for place in the tree for new item */
1845         retval = search_item(sb, &key, path);
1846         if (retval == IO_ERROR) {
1847                 reiserfs_error(sb, "vs-13080",
1848                                "i/o failure occurred creating new directory");
1849                 return -EIO;
1850         }
1851         if (retval == ITEM_FOUND) {
1852                 pathrelse(path);
1853                 reiserfs_warning(sb, "vs-13070",
1854                                  "object with this key exists (%k)",
1855                                  &(ih->ih_key));
1856                 return -EEXIST;
1857         }
1858 
1859         /* insert item, that is empty directory item */
1860         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1861 }
1862 
1863 /*
1864  * stat data of object has been inserted, this inserts the item
1865  * containing the body of symlink
1866  */
1867 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1868                                 struct inode *inode,
1869                                 struct item_head *ih,
1870                                 struct treepath *path, const char *symname,
1871                                 int item_len)
1872 {
1873         struct super_block *sb = th->t_super;
1874         struct cpu_key key;
1875         int retval;
1876 
1877         BUG_ON(!th->t_trans_id);
1878 
1879         _make_cpu_key(&key, KEY_FORMAT_3_5,
1880                       le32_to_cpu(ih->ih_key.k_dir_id),
1881                       le32_to_cpu(ih->ih_key.k_objectid),
1882                       1, TYPE_DIRECT, 3 /*key length */ );
1883 
1884         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1885                           0 /*free_space */ );
1886 
1887         /* look for place in the tree for new item */
1888         retval = search_item(sb, &key, path);
1889         if (retval == IO_ERROR) {
1890                 reiserfs_error(sb, "vs-13080",
1891                                "i/o failure occurred creating new symlink");
1892                 return -EIO;
1893         }
1894         if (retval == ITEM_FOUND) {
1895                 pathrelse(path);
1896                 reiserfs_warning(sb, "vs-13080",
1897                                  "object with this key exists (%k)",
1898                                  &(ih->ih_key));
1899                 return -EEXIST;
1900         }
1901 
1902         /* insert item, that is body of symlink */
1903         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1904 }
1905 
1906 /*
1907  * inserts the stat data into the tree, and then calls
1908  * reiserfs_new_directory (to insert ".", ".." item if new object is
1909  * directory) or reiserfs_new_symlink (to insert symlink body if new
1910  * object is symlink) or nothing (if new object is regular file)
1911 
1912  * NOTE! uid and gid must already be set in the inode.  If we return
1913  * non-zero due to an error, we have to drop the quota previously allocated
1914  * for the fresh inode.  This can only be done outside a transaction, so
1915  * if we return non-zero, we also end the transaction.
1916  *
1917  * @th: active transaction handle
1918  * @dir: parent directory for new inode
1919  * @mode: mode of new inode
1920  * @symname: symlink contents if inode is symlink
1921  * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1922  *         symlinks
1923  * @inode: inode to be filled
1924  * @security: optional security context to associate with this inode
1925  */
1926 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1927                        struct inode *dir, umode_t mode, const char *symname,
1928                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1929                           strlen (symname) for symlinks) */
1930                        loff_t i_size, struct dentry *dentry,
1931                        struct inode *inode,
1932                        struct reiserfs_security_handle *security)
1933 {
1934         struct super_block *sb = dir->i_sb;
1935         struct reiserfs_iget_args args;
1936         INITIALIZE_PATH(path_to_key);
1937         struct cpu_key key;
1938         struct item_head ih;
1939         struct stat_data sd;
1940         int retval;
1941         int err;
1942         int depth;
1943 
1944         BUG_ON(!th->t_trans_id);
1945 
1946         depth = reiserfs_write_unlock_nested(sb);
1947         err = dquot_alloc_inode(inode);
1948         reiserfs_write_lock_nested(sb, depth);
1949         if (err)
1950                 goto out_end_trans;
1951         if (!dir->i_nlink) {
1952                 err = -EPERM;
1953                 goto out_bad_inode;
1954         }
1955 
1956         /* item head of new item */
1957         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1958         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1959         if (!ih.ih_key.k_objectid) {
1960                 err = -ENOMEM;
1961                 goto out_bad_inode;
1962         }
1963         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1964         if (old_format_only(sb))
1965                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1966                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1967         else
1968                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1969                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1970         memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1971         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1972 
1973         depth = reiserfs_write_unlock_nested(inode->i_sb);
1974         err = insert_inode_locked4(inode, args.objectid,
1975                              reiserfs_find_actor, &args);
1976         reiserfs_write_lock_nested(inode->i_sb, depth);
1977         if (err) {
1978                 err = -EINVAL;
1979                 goto out_bad_inode;
1980         }
1981 
1982         if (old_format_only(sb))
1983                 /*
1984                  * not a perfect generation count, as object ids can be reused,
1985                  * but this is as good as reiserfs can do right now.
1986                  * note that the private part of inode isn't filled in yet,
1987                  * we have to use the directory.
1988                  */
1989                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1990         else
1991 #if defined( USE_INODE_GENERATION_COUNTER )
1992                 inode->i_generation =
1993                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1994 #else
1995                 inode->i_generation = ++event;
1996 #endif
1997 
1998         /* fill stat data */
1999         set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
2000 
2001         /* uid and gid must already be set by the caller for quota init */
2002 
2003         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
2004         inode->i_size = i_size;
2005         inode->i_blocks = 0;
2006         inode->i_bytes = 0;
2007         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
2008             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
2009 
2010         INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
2011         REISERFS_I(inode)->i_flags = 0;
2012         REISERFS_I(inode)->i_prealloc_block = 0;
2013         REISERFS_I(inode)->i_prealloc_count = 0;
2014         REISERFS_I(inode)->i_trans_id = 0;
2015         REISERFS_I(inode)->i_jl = NULL;
2016         REISERFS_I(inode)->i_attrs =
2017             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2018         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2019         reiserfs_init_xattr_rwsem(inode);
2020 
2021         /* key to search for correct place for new stat data */
2022         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2023                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2024                       TYPE_STAT_DATA, 3 /*key length */ );
2025 
2026         /* find proper place for inserting of stat data */
2027         retval = search_item(sb, &key, &path_to_key);
2028         if (retval == IO_ERROR) {
2029                 err = -EIO;
2030                 goto out_bad_inode;
2031         }
2032         if (retval == ITEM_FOUND) {
2033                 pathrelse(&path_to_key);
2034                 err = -EEXIST;
2035                 goto out_bad_inode;
2036         }
2037         if (old_format_only(sb)) {
2038                 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
2039                 if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2040                         pathrelse(&path_to_key);
2041                         err = -EINVAL;
2042                         goto out_bad_inode;
2043                 }
2044                 inode2sd_v1(&sd, inode, inode->i_size);
2045         } else {
2046                 inode2sd(&sd, inode, inode->i_size);
2047         }
2048         /*
2049          * store in in-core inode the key of stat data and version all
2050          * object items will have (directory items will have old offset
2051          * format, other new objects will consist of new items)
2052          */
2053         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2054                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2055         else
2056                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2057         if (old_format_only(sb))
2058                 set_inode_sd_version(inode, STAT_DATA_V1);
2059         else
2060                 set_inode_sd_version(inode, STAT_DATA_V2);
2061 
2062         /* insert the stat data into the tree */
2063 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2064         if (REISERFS_I(dir)->new_packing_locality)
2065                 th->displace_new_blocks = 1;
2066 #endif
2067         retval =
2068             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2069                                  (char *)(&sd));
2070         if (retval) {
2071                 err = retval;
2072                 reiserfs_check_path(&path_to_key);
2073                 goto out_bad_inode;
2074         }
2075 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2076         if (!th->displace_new_blocks)
2077                 REISERFS_I(dir)->new_packing_locality = 0;
2078 #endif
2079         if (S_ISDIR(mode)) {
2080                 /* insert item with "." and ".." */
2081                 retval =
2082                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2083         }
2084 
2085         if (S_ISLNK(mode)) {
2086                 /* insert body of symlink */
2087                 if (!old_format_only(sb))
2088                         i_size = ROUND_UP(i_size);
2089                 retval =
2090                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2091                                          i_size);
2092         }
2093         if (retval) {
2094                 err = retval;
2095                 reiserfs_check_path(&path_to_key);
2096                 journal_end(th);
2097                 goto out_inserted_sd;
2098         }
2099 
2100         /*
2101          * Mark it private if we're creating the privroot
2102          * or something under it.
2103          */
2104         if (IS_PRIVATE(dir) || dentry == REISERFS_SB(sb)->priv_root) {
2105                 inode->i_flags |= S_PRIVATE;
2106                 inode->i_opflags &= ~IOP_XATTR;
2107         }
2108 
2109         if (reiserfs_posixacl(inode->i_sb)) {
2110                 reiserfs_write_unlock(inode->i_sb);
2111                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2112                 reiserfs_write_lock(inode->i_sb);
2113                 if (retval) {
2114                         err = retval;
2115                         reiserfs_check_path(&path_to_key);
2116                         journal_end(th);
2117                         goto out_inserted_sd;
2118                 }
2119         } else if (inode->i_sb->s_flags & SB_POSIXACL) {
2120                 reiserfs_warning(inode->i_sb, "jdm-13090",
2121                                  "ACLs aren't enabled in the fs, "
2122                                  "but vfs thinks they are!");
2123         }
2124 
2125         if (security->name) {
2126                 reiserfs_write_unlock(inode->i_sb);
2127                 retval = reiserfs_security_write(th, inode, security);
2128                 reiserfs_write_lock(inode->i_sb);
2129                 if (retval) {
2130                         err = retval;
2131                         reiserfs_check_path(&path_to_key);
2132                         retval = journal_end(th);
2133                         if (retval)
2134                                 err = retval;
2135                         goto out_inserted_sd;
2136                 }
2137         }
2138 
2139         reiserfs_update_sd(th, inode);
2140         reiserfs_check_path(&path_to_key);
2141 
2142         return 0;
2143 
2144 out_bad_inode:
2145         /* Invalidate the object, nothing was inserted yet */
2146         INODE_PKEY(inode)->k_objectid = 0;
2147 
2148         /* Quota change must be inside a transaction for journaling */
2149         depth = reiserfs_write_unlock_nested(inode->i_sb);
2150         dquot_free_inode(inode);
2151         reiserfs_write_lock_nested(inode->i_sb, depth);
2152 
2153 out_end_trans:
2154         journal_end(th);
2155         /*
2156          * Drop can be outside and it needs more credits so it's better
2157          * to have it outside
2158          */
2159         depth = reiserfs_write_unlock_nested(inode->i_sb);
2160         dquot_drop(inode);
2161         reiserfs_write_lock_nested(inode->i_sb, depth);
2162         inode->i_flags |= S_NOQUOTA;
2163         make_bad_inode(inode);
2164 
2165 out_inserted_sd:
2166         clear_nlink(inode);
2167         th->t_trans_id = 0;     /* so the caller can't use this handle later */
2168         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
2169         iput(inode);
2170         return err;
2171 }
2172 
2173 /*
2174  * finds the tail page in the page cache,
2175  * reads the last block in.
2176  *
2177  * On success, page_result is set to a locked, pinned page, and bh_result
2178  * is set to an up to date buffer for the last block in the file.  returns 0.
2179  *
2180  * tail conversion is not done, so bh_result might not be valid for writing
2181  * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2182  * trying to write the block.
2183  *
2184  * on failure, nonzero is returned, page_result and bh_result are untouched.
2185  */
2186 static int grab_tail_page(struct inode *inode,
2187                           struct page **page_result,
2188                           struct buffer_head **bh_result)
2189 {
2190 
2191         /*
2192          * we want the page with the last byte in the file,
2193          * not the page that will hold the next byte for appending
2194          */
2195         unsigned long index = (inode->i_size - 1) >> PAGE_SHIFT;
2196         unsigned long pos = 0;
2197         unsigned long start = 0;
2198         unsigned long blocksize = inode->i_sb->s_blocksize;
2199         unsigned long offset = (inode->i_size) & (PAGE_SIZE - 1);
2200         struct buffer_head *bh;
2201         struct buffer_head *head;
2202         struct page *page;
2203         int error;
2204 
2205         /*
2206          * we know that we are only called with inode->i_size > 0.
2207          * we also know that a file tail can never be as big as a block
2208          * If i_size % blocksize == 0, our file is currently block aligned
2209          * and it won't need converting or zeroing after a truncate.
2210          */
2211         if ((offset & (blocksize - 1)) == 0) {
2212                 return -ENOENT;
2213         }
2214         page = grab_cache_page(inode->i_mapping, index);
2215         error = -ENOMEM;
2216         if (!page) {
2217                 goto out;
2218         }
2219         /* start within the page of the last block in the file */
2220         start = (offset / blocksize) * blocksize;
2221 
2222         error = __block_write_begin(page, start, offset - start,
2223                                     reiserfs_get_block_create_0);
2224         if (error)
2225                 goto unlock;
2226 
2227         head = page_buffers(page);
2228         bh = head;
2229         do {
2230                 if (pos >= start) {
2231                         break;
2232                 }
2233                 bh = bh->b_this_page;
2234                 pos += blocksize;
2235         } while (bh != head);
2236 
2237         if (!buffer_uptodate(bh)) {
2238                 /*
2239                  * note, this should never happen, prepare_write should be
2240                  * taking care of this for us.  If the buffer isn't up to
2241                  * date, I've screwed up the code to find the buffer, or the
2242                  * code to call prepare_write
2243                  */
2244                 reiserfs_error(inode->i_sb, "clm-6000",
2245                                "error reading block %lu", bh->b_blocknr);
2246                 error = -EIO;
2247                 goto unlock;
2248         }
2249         *bh_result = bh;
2250         *page_result = page;
2251 
2252 out:
2253         return error;
2254 
2255 unlock:
2256         unlock_page(page);
2257         put_page(page);
2258         return error;
2259 }
2260 
2261 /*
2262  * vfs version of truncate file.  Must NOT be called with
2263  * a transaction already started.
2264  *
2265  * some code taken from block_truncate_page
2266  */
2267 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2268 {
2269         struct reiserfs_transaction_handle th;
2270         /* we want the offset for the first byte after the end of the file */
2271         unsigned long offset = inode->i_size & (PAGE_SIZE - 1);
2272         unsigned blocksize = inode->i_sb->s_blocksize;
2273         unsigned length;
2274         struct page *page = NULL;
2275         int error;
2276         struct buffer_head *bh = NULL;
2277         int err2;
2278 
2279         reiserfs_write_lock(inode->i_sb);
2280 
2281         if (inode->i_size > 0) {
2282                 error = grab_tail_page(inode, &page, &bh);
2283                 if (error) {
2284                         /*
2285                          * -ENOENT means we truncated past the end of the
2286                          * file, and get_block_create_0 could not find a
2287                          * block to read in, which is ok.
2288                          */
2289                         if (error != -ENOENT)
2290                                 reiserfs_error(inode->i_sb, "clm-6001",
2291                                                "grab_tail_page failed %d",
2292                                                error);
2293                         page = NULL;
2294                         bh = NULL;
2295                 }
2296         }
2297 
2298         /*
2299          * so, if page != NULL, we have a buffer head for the offset at
2300          * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2301          * then we have an unformatted node.  Otherwise, we have a direct item,
2302          * and no zeroing is required on disk.  We zero after the truncate,
2303          * because the truncate might pack the item anyway
2304          * (it will unmap bh if it packs).
2305          *
2306          * it is enough to reserve space in transaction for 2 balancings:
2307          * one for "save" link adding and another for the first
2308          * cut_from_item. 1 is for update_sd
2309          */
2310         error = journal_begin(&th, inode->i_sb,
2311                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2312         if (error)
2313                 goto out;
2314         reiserfs_update_inode_transaction(inode);
2315         if (update_timestamps)
2316                 /*
2317                  * we are doing real truncate: if the system crashes
2318                  * before the last transaction of truncating gets committed
2319                  * - on reboot the file either appears truncated properly
2320                  * or not truncated at all
2321                  */
2322                 add_save_link(&th, inode, 1);
2323         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2324         error = journal_end(&th);
2325         if (error)
2326                 goto out;
2327 
2328         /* check reiserfs_do_truncate after ending the transaction */
2329         if (err2) {
2330                 error = err2;
2331                 goto out;
2332         }
2333         
2334         if (update_timestamps) {
2335                 error = remove_save_link(inode, 1 /* truncate */);
2336                 if (error)
2337                         goto out;
2338         }
2339 
2340         if (page) {
2341                 length = offset & (blocksize - 1);
2342                 /* if we are not on a block boundary */
2343                 if (length) {
2344                         length = blocksize - length;
2345                         zero_user(page, offset, length);
2346                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2347                                 mark_buffer_dirty(bh);
2348                         }
2349                 }
2350                 unlock_page(page);
2351                 put_page(page);
2352         }
2353 
2354         reiserfs_write_unlock(inode->i_sb);
2355 
2356         return 0;
2357 out:
2358         if (page) {
2359                 unlock_page(page);
2360                 put_page(page);
2361         }
2362 
2363         reiserfs_write_unlock(inode->i_sb);
2364 
2365         return error;
2366 }
2367 
2368 static int map_block_for_writepage(struct inode *inode,
2369                                    struct buffer_head *bh_result,
2370                                    unsigned long block)
2371 {
2372         struct reiserfs_transaction_handle th;
2373         int fs_gen;
2374         struct item_head tmp_ih;
2375         struct item_head *ih;
2376         struct buffer_head *bh;
2377         __le32 *item;
2378         struct cpu_key key;
2379         INITIALIZE_PATH(path);
2380         int pos_in_item;
2381         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2382         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2383         int retval;
2384         int use_get_block = 0;
2385         int bytes_copied = 0;
2386         int copy_size;
2387         int trans_running = 0;
2388 
2389         /*
2390          * catch places below that try to log something without
2391          * starting a trans
2392          */
2393         th.t_trans_id = 0;
2394 
2395         if (!buffer_uptodate(bh_result)) {
2396                 return -EIO;
2397         }
2398 
2399         kmap(bh_result->b_page);
2400 start_over:
2401         reiserfs_write_lock(inode->i_sb);
2402         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2403 
2404 research:
2405         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2406         if (retval != POSITION_FOUND) {
2407                 use_get_block = 1;
2408                 goto out;
2409         }
2410 
2411         bh = get_last_bh(&path);
2412         ih = tp_item_head(&path);
2413         item = tp_item_body(&path);
2414         pos_in_item = path.pos_in_item;
2415 
2416         /* we've found an unformatted node */
2417         if (indirect_item_found(retval, ih)) {
2418                 if (bytes_copied > 0) {
2419                         reiserfs_warning(inode->i_sb, "clm-6002",
2420                                          "bytes_copied %d", bytes_copied);
2421                 }
2422                 if (!get_block_num(item, pos_in_item)) {
2423                         /* crap, we are writing to a hole */
2424                         use_get_block = 1;
2425                         goto out;
2426                 }
2427                 set_block_dev_mapped(bh_result,
2428                                      get_block_num(item, pos_in_item), inode);
2429         } else if (is_direct_le_ih(ih)) {
2430                 char *p;
2431                 p = page_address(bh_result->b_page);
2432                 p += (byte_offset - 1) & (PAGE_SIZE - 1);
2433                 copy_size = ih_item_len(ih) - pos_in_item;
2434 
2435                 fs_gen = get_generation(inode->i_sb);
2436                 copy_item_head(&tmp_ih, ih);
2437 
2438                 if (!trans_running) {
2439                         /* vs-3050 is gone, no need to drop the path */
2440                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2441                         if (retval)
2442                                 goto out;
2443                         reiserfs_update_inode_transaction(inode);
2444                         trans_running = 1;
2445                         if (fs_changed(fs_gen, inode->i_sb)
2446                             && item_moved(&tmp_ih, &path)) {
2447                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2448                                                                  bh);
2449                                 goto research;
2450                         }
2451                 }
2452 
2453                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2454 
2455                 if (fs_changed(fs_gen, inode->i_sb)
2456                     && item_moved(&tmp_ih, &path)) {
2457                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2458                         goto research;
2459                 }
2460 
2461                 memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2462                        copy_size);
2463 
2464                 journal_mark_dirty(&th, bh);
2465                 bytes_copied += copy_size;
2466                 set_block_dev_mapped(bh_result, 0, inode);
2467 
2468                 /* are there still bytes left? */
2469                 if (bytes_copied < bh_result->b_size &&
2470                     (byte_offset + bytes_copied) < inode->i_size) {
2471                         set_cpu_key_k_offset(&key,
2472                                              cpu_key_k_offset(&key) +
2473                                              copy_size);
2474                         goto research;
2475                 }
2476         } else {
2477                 reiserfs_warning(inode->i_sb, "clm-6003",
2478                                  "bad item inode %lu", inode->i_ino);
2479                 retval = -EIO;
2480                 goto out;
2481         }
2482         retval = 0;
2483 
2484 out:
2485         pathrelse(&path);
2486         if (trans_running) {
2487                 int err = journal_end(&th);
2488                 if (err)
2489                         retval = err;
2490                 trans_running = 0;
2491         }
2492         reiserfs_write_unlock(inode->i_sb);
2493 
2494         /* this is where we fill in holes in the file. */
2495         if (use_get_block) {
2496                 retval = reiserfs_get_block(inode, block, bh_result,
2497                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2498                                             | GET_BLOCK_NO_DANGLE);
2499                 if (!retval) {
2500                         if (!buffer_mapped(bh_result)
2501                             || bh_result->b_blocknr == 0) {
2502                                 /* get_block failed to find a mapped unformatted node. */
2503                                 use_get_block = 0;
2504                                 goto start_over;
2505                         }
2506                 }
2507         }
2508         kunmap(bh_result->b_page);
2509 
2510         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2511                 /*
2512                  * we've copied data from the page into the direct item, so the
2513                  * buffer in the page is now clean, mark it to reflect that.
2514                  */
2515                 lock_buffer(bh_result);
2516                 clear_buffer_dirty(bh_result);
2517                 unlock_buffer(bh_result);
2518         }
2519         return retval;
2520 }
2521 
2522 /*
2523  * mason@suse.com: updated in 2.5.54 to follow the same general io
2524  * start/recovery path as __block_write_full_page, along with special
2525  * code to handle reiserfs tails.
2526  */
2527 static int reiserfs_write_full_page(struct page *page,
2528                                     struct writeback_control *wbc)
2529 {
2530         struct inode *inode = page->mapping->host;
2531         unsigned long end_index = inode->i_size >> PAGE_SHIFT;
2532         int error = 0;
2533         unsigned long block;
2534         sector_t last_block;
2535         struct buffer_head *head, *bh;
2536         int partial = 0;
2537         int nr = 0;
2538         int checked = PageChecked(page);
2539         struct reiserfs_transaction_handle th;
2540         struct super_block *s = inode->i_sb;
2541         int bh_per_page = PAGE_SIZE / s->s_blocksize;
2542         th.t_trans_id = 0;
2543 
2544         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2545         if (checked && (current->flags & PF_MEMALLOC)) {
2546                 redirty_page_for_writepage(wbc, page);
2547                 unlock_page(page);
2548                 return 0;
2549         }
2550 
2551         /*
2552          * The page dirty bit is cleared before writepage is called, which
2553          * means we have to tell create_empty_buffers to make dirty buffers
2554          * The page really should be up to date at this point, so tossing
2555          * in the BH_Uptodate is just a sanity check.
2556          */
2557         if (!page_has_buffers(page)) {
2558                 create_empty_buffers(page, s->s_blocksize,
2559                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2560         }
2561         head = page_buffers(page);
2562 
2563         /*
2564          * last page in the file, zero out any contents past the
2565          * last byte in the file
2566          */
2567         if (page->index >= end_index) {
2568                 unsigned last_offset;
2569 
2570                 last_offset = inode->i_size & (PAGE_SIZE - 1);
2571                 /* no file contents in this page */
2572                 if (page->index >= end_index + 1 || !last_offset) {
2573                         unlock_page(page);
2574                         return 0;
2575                 }
2576                 zero_user_segment(page, last_offset, PAGE_SIZE);
2577         }
2578         bh = head;
2579         block = page->index << (PAGE_SHIFT - s->s_blocksize_bits);
2580         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2581         /* first map all the buffers, logging any direct items we find */
2582         do {
2583                 if (block > last_block) {
2584                         /*
2585                          * This can happen when the block size is less than
2586                          * the page size.  The corresponding bytes in the page
2587                          * were zero filled above
2588                          */
2589                         clear_buffer_dirty(bh);
2590                         set_buffer_uptodate(bh);
2591                 } else if ((checked || buffer_dirty(bh)) &&
2592                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2593                                                        && bh->b_blocknr ==
2594                                                        0))) {
2595                         /*
2596                          * not mapped yet, or it points to a direct item, search
2597                          * the btree for the mapping info, and log any direct
2598                          * items found
2599                          */
2600                         if ((error = map_block_for_writepage(inode, bh, block))) {
2601                                 goto fail;
2602                         }
2603                 }
2604                 bh = bh->b_this_page;
2605                 block++;
2606         } while (bh != head);
2607 
2608         /*
2609          * we start the transaction after map_block_for_writepage,
2610          * because it can create holes in the file (an unbounded operation).
2611          * starting it here, we can make a reliable estimate for how many
2612          * blocks we're going to log
2613          */
2614         if (checked) {
2615                 ClearPageChecked(page);
2616                 reiserfs_write_lock(s);
2617                 error = journal_begin(&th, s, bh_per_page + 1);
2618                 if (error) {
2619                         reiserfs_write_unlock(s);
2620                         goto fail;
2621                 }
2622                 reiserfs_update_inode_transaction(inode);
2623         }
2624         /* now go through and lock any dirty buffers on the page */
2625         do {
2626                 get_bh(bh);
2627                 if (!buffer_mapped(bh))
2628                         continue;
2629                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2630                         continue;
2631 
2632                 if (checked) {
2633                         reiserfs_prepare_for_journal(s, bh, 1);
2634                         journal_mark_dirty(&th, bh);
2635                         continue;
2636                 }
2637                 /*
2638                  * from this point on, we know the buffer is mapped to a
2639                  * real block and not a direct item
2640                  */
2641                 if (wbc->sync_mode != WB_SYNC_NONE) {
2642                         lock_buffer(bh);
2643                 } else {
2644                         if (!trylock_buffer(bh)) {
2645                                 redirty_page_for_writepage(wbc, page);
2646                                 continue;
2647                         }
2648                 }
2649                 if (test_clear_buffer_dirty(bh)) {
2650                         mark_buffer_async_write(bh);
2651                 } else {
2652                         unlock_buffer(bh);
2653                 }
2654         } while ((bh = bh->b_this_page) != head);
2655 
2656         if (checked) {
2657                 error = journal_end(&th);
2658                 reiserfs_write_unlock(s);
2659                 if (error)
2660                         goto fail;
2661         }
2662         BUG_ON(PageWriteback(page));
2663         set_page_writeback(page);
2664         unlock_page(page);
2665 
2666         /*
2667          * since any buffer might be the only dirty buffer on the page,
2668          * the first submit_bh can bring the page out of writeback.
2669          * be careful with the buffers.
2670          */
2671         do {
2672                 struct buffer_head *next = bh->b_this_page;
2673                 if (buffer_async_write(bh)) {
2674                         submit_bh(REQ_OP_WRITE, 0, bh);
2675                         nr++;
2676                 }
2677                 put_bh(bh);
2678                 bh = next;
2679         } while (bh != head);
2680 
2681         error = 0;
2682 done:
2683         if (nr == 0) {
2684                 /*
2685                  * if this page only had a direct item, it is very possible for
2686                  * no io to be required without there being an error.  Or,
2687                  * someone else could have locked them and sent them down the
2688                  * pipe without locking the page
2689                  */
2690                 bh = head;
2691                 do {
2692                         if (!buffer_uptodate(bh)) {
2693                                 partial = 1;
2694                                 break;
2695                         }
2696                         bh = bh->b_this_page;
2697                 } while (bh != head);
2698                 if (!partial)
2699                         SetPageUptodate(page);
2700                 end_page_writeback(page);
2701         }
2702         return error;
2703 
2704 fail:
2705         /*
2706          * catches various errors, we need to make sure any valid dirty blocks
2707          * get to the media.  The page is currently locked and not marked for
2708          * writeback
2709          */
2710         ClearPageUptodate(page);
2711         bh = head;
2712         do {
2713                 get_bh(bh);
2714                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2715                         lock_buffer(bh);
2716                         mark_buffer_async_write(bh);
2717                 } else {
2718                         /*
2719                          * clear any dirty bits that might have come from
2720                          * getting attached to a dirty page
2721                          */
2722                         clear_buffer_dirty(bh);
2723                 }
2724                 bh = bh->b_this_page;
2725         } while (bh != head);
2726         SetPageError(page);
2727         BUG_ON(PageWriteback(page));
2728         set_page_writeback(page);
2729         unlock_page(page);
2730         do {
2731                 struct buffer_head *next = bh->b_this_page;
2732                 if (buffer_async_write(bh)) {
2733                         clear_buffer_dirty(bh);
2734                         submit_bh(REQ_OP_WRITE, 0, bh);
2735                         nr++;
2736                 }
2737                 put_bh(bh);
2738                 bh = next;
2739         } while (bh != head);
2740         goto done;
2741 }
2742 
2743 static int reiserfs_readpage(struct file *f, struct page *page)
2744 {
2745         return block_read_full_page(page, reiserfs_get_block);
2746 }
2747 
2748 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2749 {
2750         struct inode *inode = page->mapping->host;
2751         reiserfs_wait_on_write_block(inode->i_sb);
2752         return reiserfs_write_full_page(page, wbc);
2753 }
2754 
2755 static void reiserfs_truncate_failed_write(struct inode *inode)
2756 {
2757         truncate_inode_pages(inode->i_mapping, inode->i_size);
2758         reiserfs_truncate_file(inode, 0);
2759 }
2760 
2761 static int reiserfs_write_begin(struct file *file,
2762                                 struct address_space *mapping,
2763                                 loff_t pos, unsigned len, unsigned flags,
2764                                 struct page **pagep, void **fsdata)
2765 {
2766         struct inode *inode;
2767         struct page *page;
2768         pgoff_t index;
2769         int ret;
2770         int old_ref = 0;
2771 
2772         inode = mapping->host;
2773         *fsdata = NULL;
2774         if (flags & AOP_FLAG_CONT_EXPAND &&
2775             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2776                 pos ++;
2777                 *fsdata = (void *)(unsigned long)flags;
2778         }
2779 
2780         index = pos >> PAGE_SHIFT;
2781         page = grab_cache_page_write_begin(mapping, index, flags);
2782         if (!page)
2783                 return -ENOMEM;
2784         *pagep = page;
2785 
2786         reiserfs_wait_on_write_block(inode->i_sb);
2787         fix_tail_page_for_writing(page);
2788         if (reiserfs_transaction_running(inode->i_sb)) {
2789                 struct reiserfs_transaction_handle *th;
2790                 th = (struct reiserfs_transaction_handle *)current->
2791                     journal_info;
2792                 BUG_ON(!th->t_refcount);
2793                 BUG_ON(!th->t_trans_id);
2794                 old_ref = th->t_refcount;
2795                 th->t_refcount++;
2796         }
2797         ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2798         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2799                 struct reiserfs_transaction_handle *th = current->journal_info;
2800                 /*
2801                  * this gets a little ugly.  If reiserfs_get_block returned an
2802                  * error and left a transacstion running, we've got to close
2803                  * it, and we've got to free handle if it was a persistent
2804                  * transaction.
2805                  *
2806                  * But, if we had nested into an existing transaction, we need
2807                  * to just drop the ref count on the handle.
2808                  *
2809                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2810                  * and it was a persistent trans.  Otherwise, it was nested
2811                  * above.
2812                  */
2813                 if (th->t_refcount > old_ref) {
2814                         if (old_ref)
2815                                 th->t_refcount--;
2816                         else {
2817                                 int err;
2818                                 reiserfs_write_lock(inode->i_sb);
2819                                 err = reiserfs_end_persistent_transaction(th);
2820                                 reiserfs_write_unlock(inode->i_sb);
2821                                 if (err)
2822                                         ret = err;
2823                         }
2824                 }
2825         }
2826         if (ret) {
2827                 unlock_page(page);
2828                 put_page(page);
2829                 /* Truncate allocated blocks */
2830                 reiserfs_truncate_failed_write(inode);
2831         }
2832         return ret;
2833 }
2834 
2835 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2836 {
2837         struct inode *inode = page->mapping->host;
2838         int ret;
2839         int old_ref = 0;
2840         int depth;
2841 
2842         depth = reiserfs_write_unlock_nested(inode->i_sb);
2843         reiserfs_wait_on_write_block(inode->i_sb);
2844         reiserfs_write_lock_nested(inode->i_sb, depth);
2845 
2846         fix_tail_page_for_writing(page);
2847         if (reiserfs_transaction_running(inode->i_sb)) {
2848                 struct reiserfs_transaction_handle *th;
2849                 th = (struct reiserfs_transaction_handle *)current->
2850                     journal_info;
2851                 BUG_ON(!th->t_refcount);
2852                 BUG_ON(!th->t_trans_id);
2853                 old_ref = th->t_refcount;
2854                 th->t_refcount++;
2855         }
2856 
2857         ret = __block_write_begin(page, from, len, reiserfs_get_block);
2858         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2859                 struct reiserfs_transaction_handle *th = current->journal_info;
2860                 /*
2861                  * this gets a little ugly.  If reiserfs_get_block returned an
2862                  * error and left a transacstion running, we've got to close
2863                  * it, and we've got to free handle if it was a persistent
2864                  * transaction.
2865                  *
2866                  * But, if we had nested into an existing transaction, we need
2867                  * to just drop the ref count on the handle.
2868                  *
2869                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2870                  * and it was a persistent trans.  Otherwise, it was nested
2871                  * above.
2872                  */
2873                 if (th->t_refcount > old_ref) {
2874                         if (old_ref)
2875                                 th->t_refcount--;
2876                         else {
2877                                 int err;
2878                                 reiserfs_write_lock(inode->i_sb);
2879                                 err = reiserfs_end_persistent_transaction(th);
2880                                 reiserfs_write_unlock(inode->i_sb);
2881                                 if (err)
2882                                         ret = err;
2883                         }
2884                 }
2885         }
2886         return ret;
2887 
2888 }
2889 
2890 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2891 {
2892         return generic_block_bmap(as, block, reiserfs_bmap);
2893 }
2894 
2895 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2896                               loff_t pos, unsigned len, unsigned copied,
2897                               struct page *page, void *fsdata)
2898 {
2899         struct inode *inode = page->mapping->host;
2900         int ret = 0;
2901         int update_sd = 0;
2902         struct reiserfs_transaction_handle *th;
2903         unsigned start;
2904         bool locked = false;
2905 
2906         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2907                 pos ++;
2908 
2909         reiserfs_wait_on_write_block(inode->i_sb);
2910         if (reiserfs_transaction_running(inode->i_sb))
2911                 th = current->journal_info;
2912         else
2913                 th = NULL;
2914 
2915         start = pos & (PAGE_SIZE - 1);
2916         if (unlikely(copied < len)) {
2917                 if (!PageUptodate(page))
2918                         copied = 0;
2919 
2920                 page_zero_new_buffers(page, start + copied, start + len);
2921         }
2922         flush_dcache_page(page);
2923 
2924         reiserfs_commit_page(inode, page, start, start + copied);
2925 
2926         /*
2927          * generic_commit_write does this for us, but does not update the
2928          * transaction tracking stuff when the size changes.  So, we have
2929          * to do the i_size updates here.
2930          */
2931         if (pos + copied > inode->i_size) {
2932                 struct reiserfs_transaction_handle myth;
2933                 reiserfs_write_lock(inode->i_sb);
2934                 locked = true;
2935                 /*
2936                  * If the file have grown beyond the border where it
2937                  * can have a tail, unmark it as needing a tail
2938                  * packing
2939                  */
2940                 if ((have_large_tails(inode->i_sb)
2941                      && inode->i_size > i_block_size(inode) * 4)
2942                     || (have_small_tails(inode->i_sb)
2943                         && inode->i_size > i_block_size(inode)))
2944                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2945 
2946                 ret = journal_begin(&myth, inode->i_sb, 1);
2947                 if (ret)
2948                         goto journal_error;
2949 
2950                 reiserfs_update_inode_transaction(inode);
2951                 inode->i_size = pos + copied;
2952                 /*
2953                  * this will just nest into our transaction.  It's important
2954                  * to use mark_inode_dirty so the inode gets pushed around on
2955                  * the dirty lists, and so that O_SYNC works as expected
2956                  */
2957                 mark_inode_dirty(inode);
2958                 reiserfs_update_sd(&myth, inode);
2959                 update_sd = 1;
2960                 ret = journal_end(&myth);
2961                 if (ret)
2962                         goto journal_error;
2963         }
2964         if (th) {
2965                 if (!locked) {
2966                         reiserfs_write_lock(inode->i_sb);
2967                         locked = true;
2968                 }
2969                 if (!update_sd)
2970                         mark_inode_dirty(inode);
2971                 ret = reiserfs_end_persistent_transaction(th);
2972                 if (ret)
2973                         goto out;
2974         }
2975 
2976 out:
2977         if (locked)
2978                 reiserfs_write_unlock(inode->i_sb);
2979         unlock_page(page);
2980         put_page(page);
2981 
2982         if (pos + len > inode->i_size)
2983                 reiserfs_truncate_failed_write(inode);
2984 
2985         return ret == 0 ? copied : ret;
2986 
2987 journal_error:
2988         reiserfs_write_unlock(inode->i_sb);
2989         locked = false;
2990         if (th) {
2991                 if (!update_sd)
2992                         reiserfs_update_sd(th, inode);
2993                 ret = reiserfs_end_persistent_transaction(th);
2994         }
2995         goto out;
2996 }
2997 
2998 int reiserfs_commit_write(struct file *f, struct page *page,
2999                           unsigned from, unsigned to)
3000 {
3001         struct inode *inode = page->mapping->host;
3002         loff_t pos = ((loff_t) page->index << PAGE_SHIFT) + to;
3003         int ret = 0;
3004         int update_sd = 0;
3005         struct reiserfs_transaction_handle *th = NULL;
3006         int depth;
3007 
3008         depth = reiserfs_write_unlock_nested(inode->i_sb);
3009         reiserfs_wait_on_write_block(inode->i_sb);
3010         reiserfs_write_lock_nested(inode->i_sb, depth);
3011 
3012         if (reiserfs_transaction_running(inode->i_sb)) {
3013                 th = current->journal_info;
3014         }
3015         reiserfs_commit_page(inode, page, from, to);
3016 
3017         /*
3018          * generic_commit_write does this for us, but does not update the
3019          * transaction tracking stuff when the size changes.  So, we have
3020          * to do the i_size updates here.
3021          */
3022         if (pos > inode->i_size) {
3023                 struct reiserfs_transaction_handle myth;
3024                 /*
3025                  * If the file have grown beyond the border where it
3026                  * can have a tail, unmark it as needing a tail
3027                  * packing
3028                  */
3029                 if ((have_large_tails(inode->i_sb)
3030                      && inode->i_size > i_block_size(inode) * 4)
3031                     || (have_small_tails(inode->i_sb)
3032                         && inode->i_size > i_block_size(inode)))
3033                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3034 
3035                 ret = journal_begin(&myth, inode->i_sb, 1);
3036                 if (ret)
3037                         goto journal_error;
3038 
3039                 reiserfs_update_inode_transaction(inode);
3040                 inode->i_size = pos;
3041                 /*
3042                  * this will just nest into our transaction.  It's important
3043                  * to use mark_inode_dirty so the inode gets pushed around
3044                  * on the dirty lists, and so that O_SYNC works as expected
3045                  */
3046                 mark_inode_dirty(inode);
3047                 reiserfs_update_sd(&myth, inode);
3048                 update_sd = 1;
3049                 ret = journal_end(&myth);
3050                 if (ret)
3051                         goto journal_error;
3052         }
3053         if (th) {
3054                 if (!update_sd)
3055                         mark_inode_dirty(inode);
3056                 ret = reiserfs_end_persistent_transaction(th);
3057                 if (ret)
3058                         goto out;
3059         }
3060 
3061 out:
3062         return ret;
3063 
3064 journal_error:
3065         if (th) {
3066                 if (!update_sd)
3067                         reiserfs_update_sd(th, inode);
3068                 ret = reiserfs_end_persistent_transaction(th);
3069         }
3070 
3071         return ret;
3072 }
3073 
3074 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3075 {
3076         if (reiserfs_attrs(inode->i_sb)) {
3077                 if (sd_attrs & REISERFS_SYNC_FL)
3078                         inode->i_flags |= S_SYNC;
3079                 else
3080                         inode->i_flags &= ~S_SYNC;
3081                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
3082                         inode->i_flags |= S_IMMUTABLE;
3083                 else
3084                         inode->i_flags &= ~S_IMMUTABLE;
3085                 if (sd_attrs & REISERFS_APPEND_FL)
3086                         inode->i_flags |= S_APPEND;
3087                 else
3088                         inode->i_flags &= ~S_APPEND;
3089                 if (sd_attrs & REISERFS_NOATIME_FL)
3090                         inode->i_flags |= S_NOATIME;
3091                 else
3092                         inode->i_flags &= ~S_NOATIME;
3093                 if (sd_attrs & REISERFS_NOTAIL_FL)
3094                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
3095                 else
3096                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3097         }
3098 }
3099 
3100 /*
3101  * decide if this buffer needs to stay around for data logging or ordered
3102  * write purposes
3103  */
3104 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
3105 {
3106         int ret = 1;
3107         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3108 
3109         lock_buffer(bh);
3110         spin_lock(&j->j_dirty_buffers_lock);
3111         if (!buffer_mapped(bh)) {
3112                 goto free_jh;
3113         }
3114         /*
3115          * the page is locked, and the only places that log a data buffer
3116          * also lock the page.
3117          */
3118         if (reiserfs_file_data_log(inode)) {
3119                 /*
3120                  * very conservative, leave the buffer pinned if
3121                  * anyone might need it.
3122                  */
3123                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3124                         ret = 0;
3125                 }
3126         } else  if (buffer_dirty(bh)) {
3127                 struct reiserfs_journal_list *jl;
3128                 struct reiserfs_jh *jh = bh->b_private;
3129 
3130                 /*
3131                  * why is this safe?
3132                  * reiserfs_setattr updates i_size in the on disk
3133                  * stat data before allowing vmtruncate to be called.
3134                  *
3135                  * If buffer was put onto the ordered list for this
3136                  * transaction, we know for sure either this transaction
3137                  * or an older one already has updated i_size on disk,
3138                  * and this ordered data won't be referenced in the file
3139                  * if we crash.
3140                  *
3141                  * if the buffer was put onto the ordered list for an older
3142                  * transaction, we need to leave it around
3143                  */
3144                 if (jh && (jl = jh->jl)
3145                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3146                         ret = 0;
3147         }
3148 free_jh:
3149         if (ret && bh->b_private) {
3150                 reiserfs_free_jh(bh);
3151         }
3152         spin_unlock(&j->j_dirty_buffers_lock);
3153         unlock_buffer(bh);
3154         return ret;
3155 }
3156 
3157 /* clm -- taken from fs/buffer.c:block_invalidate_page */
3158 static void reiserfs_invalidatepage(struct page *page, unsigned int offset,
3159                                     unsigned int length)
3160 {
3161         struct buffer_head *head, *bh, *next;
3162         struct inode *inode = page->mapping->host;
3163         unsigned int curr_off = 0;
3164         unsigned int stop = offset + length;
3165         int partial_page = (offset || length < PAGE_SIZE);
3166         int ret = 1;
3167 
3168         BUG_ON(!PageLocked(page));
3169 
3170         if (!partial_page)
3171                 ClearPageChecked(page);
3172 
3173         if (!page_has_buffers(page))
3174                 goto out;
3175 
3176         head = page_buffers(page);
3177         bh = head;
3178         do {
3179                 unsigned int next_off = curr_off + bh->b_size;
3180                 next = bh->b_this_page;
3181 
3182                 if (next_off > stop)
3183                         goto out;
3184 
3185                 /*
3186                  * is this block fully invalidated?
3187                  */
3188                 if (offset <= curr_off) {
3189                         if (invalidatepage_can_drop(inode, bh))
3190                                 reiserfs_unmap_buffer(bh);
3191                         else
3192                                 ret = 0;
3193                 }
3194                 curr_off = next_off;
3195                 bh = next;
3196         } while (bh != head);
3197 
3198         /*
3199          * We release buffers only if the entire page is being invalidated.
3200          * The get_block cached value has been unconditionally invalidated,
3201          * so real IO is not possible anymore.
3202          */
3203         if (!partial_page && ret) {
3204                 ret = try_to_release_page(page, 0);
3205                 /* maybe should BUG_ON(!ret); - neilb */
3206         }
3207 out:
3208         return;
3209 }
3210 
3211 static int reiserfs_set_page_dirty(struct page *page)
3212 {
3213         struct inode *inode = page->mapping->host;
3214         if (reiserfs_file_data_log(inode)) {
3215                 SetPageChecked(page);
3216                 return __set_page_dirty_nobuffers(page);
3217         }
3218         return __set_page_dirty_buffers(page);
3219 }
3220 
3221 /*
3222  * Returns 1 if the page's buffers were dropped.  The page is locked.
3223  *
3224  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3225  * in the buffers at page_buffers(page).
3226  *
3227  * even in -o notail mode, we can't be sure an old mount without -o notail
3228  * didn't create files with tails.
3229  */
3230 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3231 {
3232         struct inode *inode = page->mapping->host;
3233         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3234         struct buffer_head *head;
3235         struct buffer_head *bh;
3236         int ret = 1;
3237 
3238         WARN_ON(PageChecked(page));
3239         spin_lock(&j->j_dirty_buffers_lock);
3240         head = page_buffers(page);
3241         bh = head;
3242         do {
3243                 if (bh->b_private) {
3244                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3245                                 reiserfs_free_jh(bh);
3246                         } else {
3247                                 ret = 0;
3248                                 break;
3249                         }
3250                 }
3251                 bh = bh->b_this_page;
3252         } while (bh != head);
3253         if (ret)
3254                 ret = try_to_free_buffers(page);
3255         spin_unlock(&j->j_dirty_buffers_lock);
3256         return ret;
3257 }
3258 
3259 /*
3260  * We thank Mingming Cao for helping us understand in great detail what
3261  * to do in this section of the code.
3262  */
3263 static ssize_t reiserfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3264 {
3265         struct file *file = iocb->ki_filp;
3266         struct inode *inode = file->f_mapping->host;
3267         size_t count = iov_iter_count(iter);
3268         ssize_t ret;
3269 
3270         ret = blockdev_direct_IO(iocb, inode, iter,
3271                                  reiserfs_get_blocks_direct_io);
3272 
3273         /*
3274          * In case of error extending write may have instantiated a few
3275          * blocks outside i_size. Trim these off again.
3276          */
3277         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
3278                 loff_t isize = i_size_read(inode);
3279                 loff_t end = iocb->ki_pos + count;
3280 
3281                 if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3282                         truncate_setsize(inode, isize);
3283                         reiserfs_vfs_truncate_file(inode);
3284                 }
3285         }
3286 
3287         return ret;
3288 }
3289 
3290 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3291 {
3292         struct inode *inode = d_inode(dentry);
3293         unsigned int ia_valid;
3294         int error;
3295 
3296         error = setattr_prepare(dentry, attr);
3297         if (error)
3298                 return error;
3299 
3300         /* must be turned off for recursive notify_change calls */
3301         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3302 
3303         if (is_quota_modification(inode, attr)) {
3304                 error = dquot_initialize(inode);
3305                 if (error)
3306                         return error;
3307         }
3308         reiserfs_write_lock(inode->i_sb);
3309         if (attr->ia_valid & ATTR_SIZE) {
3310                 /*
3311                  * version 2 items will be caught by the s_maxbytes check
3312                  * done for us in vmtruncate
3313                  */
3314                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3315                     attr->ia_size > MAX_NON_LFS) {
3316                         reiserfs_write_unlock(inode->i_sb);
3317                         error = -EFBIG;
3318                         goto out;
3319                 }
3320 
3321                 inode_dio_wait(inode);
3322 
3323                 /* fill in hole pointers in the expanding truncate case. */
3324                 if (attr->ia_size > inode->i_size) {
3325                         error = generic_cont_expand_simple(inode, attr->ia_size);
3326                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3327                                 int err;
3328                                 struct reiserfs_transaction_handle th;
3329                                 /* we're changing at most 2 bitmaps, inode + super */
3330                                 err = journal_begin(&th, inode->i_sb, 4);
3331                                 if (!err) {
3332                                         reiserfs_discard_prealloc(&th, inode);
3333                                         err = journal_end(&th);
3334                                 }
3335                                 if (err)
3336                                         error = err;
3337                         }
3338                         if (error) {
3339                                 reiserfs_write_unlock(inode->i_sb);
3340                                 goto out;
3341                         }
3342                         /*
3343                          * file size is changed, ctime and mtime are
3344                          * to be updated
3345                          */
3346                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3347                 }
3348         }
3349         reiserfs_write_unlock(inode->i_sb);
3350 
3351         if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3352              ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3353             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3354                 /* stat data of format v3.5 has 16 bit uid and gid */
3355                 error = -EINVAL;
3356                 goto out;
3357         }
3358 
3359         if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3360             (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3361                 struct reiserfs_transaction_handle th;
3362                 int jbegin_count =
3363                     2 *
3364                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3365                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3366                     2;
3367 
3368                 error = reiserfs_chown_xattrs(inode, attr);
3369 
3370                 if (error)
3371                         return error;
3372 
3373                 /*
3374                  * (user+group)*(old+new) structure - we count quota
3375                  * info and , inode write (sb, inode)
3376                  */
3377                 reiserfs_write_lock(inode->i_sb);
3378                 error = journal_begin(&th, inode->i_sb, jbegin_count);
3379                 reiserfs_write_unlock(inode->i_sb);
3380                 if (error)
3381                         goto out;
3382                 error = dquot_transfer(inode, attr);
3383                 reiserfs_write_lock(inode->i_sb);
3384                 if (error) {
3385                         journal_end(&th);
3386                         reiserfs_write_unlock(inode->i_sb);
3387                         goto out;
3388                 }
3389 
3390                 /*
3391                  * Update corresponding info in inode so that everything
3392                  * is in one transaction
3393                  */
3394                 if (attr->ia_valid & ATTR_UID)
3395                         inode->i_uid = attr->ia_uid;
3396                 if (attr->ia_valid & ATTR_GID)
3397                         inode->i_gid = attr->ia_gid;
3398                 mark_inode_dirty(inode);
3399                 error = journal_end(&th);
3400                 reiserfs_write_unlock(inode->i_sb);
3401                 if (error)
3402                         goto out;
3403         }
3404 
3405         if ((attr->ia_valid & ATTR_SIZE) &&
3406             attr->ia_size != i_size_read(inode)) {
3407                 error = inode_newsize_ok(inode, attr->ia_size);
3408                 if (!error) {
3409                         /*
3410                          * Could race against reiserfs_file_release
3411                          * if called from NFS, so take tailpack mutex.
3412                          */
3413                         mutex_lock(&REISERFS_I(inode)->tailpack);
3414                         truncate_setsize(inode, attr->ia_size);
3415                         reiserfs_truncate_file(inode, 1);
3416                         mutex_unlock(&REISERFS_I(inode)->tailpack);
3417                 }
3418         }
3419 
3420         if (!error) {
3421                 setattr_copy(inode, attr);
3422                 mark_inode_dirty(inode);
3423         }
3424 
3425         if (!error && reiserfs_posixacl(inode->i_sb)) {
3426                 if (attr->ia_valid & ATTR_MODE)
3427                         error = reiserfs_acl_chmod(inode);
3428         }
3429 
3430 out:
3431         return error;
3432 }
3433 
3434 const struct address_space_operations reiserfs_address_space_operations = {
3435         .writepage = reiserfs_writepage,
3436         .readpage = reiserfs_readpage,
3437         .readpages = reiserfs_readpages,
3438         .releasepage = reiserfs_releasepage,
3439         .invalidatepage = reiserfs_invalidatepage,
3440         .write_begin = reiserfs_write_begin,
3441         .write_end = reiserfs_write_end,
3442         .bmap = reiserfs_aop_bmap,
3443         .direct_IO = reiserfs_direct_IO,
3444         .set_page_dirty = reiserfs_set_page_dirty,
3445 };