fs/reiserfs/inode.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/fs.h>
   7 #include "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 kcalloc 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 void reiserfs_readahead(struct readahead_control *rac)
1164 {
1165         mpage_readahead(rac, reiserfs_get_block);
1166 }
1167
1168 /*
1169  * Compute real number of used bytes by file
1170  * Following three functions can go away when we'll have enough space in
1171  * stat item
1172  */
1173 static int real_space_diff(struct inode *inode, int sd_size)
1174 {
1175         int bytes;
1176         loff_t blocksize = inode->i_sb->s_blocksize;
1177
1178         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1179                 return sd_size;
1180
1181         /*
1182          * End of file is also in full block with indirect reference, so round
1183          * up to the next block.
1184          *
1185          * there is just no way to know if the tail is actually packed
1186          * on the file, so we have to assume it isn't.  When we pack the
1187          * tail, we add 4 bytes to pretend there really is an unformatted
1188          * node pointer
1189          */
1190         bytes =
1191             ((inode->i_size +
1192               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1193             sd_size;
1194         return bytes;
1195 }
1196
1197 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1198                                         int sd_size)
1199 {
1200         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1201                 return inode->i_size +
1202                     (loff_t) (real_space_diff(inode, sd_size));
1203         }
1204         return ((loff_t) real_space_diff(inode, sd_size)) +
1205             (((loff_t) blocks) << 9);
1206 }
1207
1208 /* Compute number of blocks used by file in ReiserFS counting */
1209 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1210 {
1211         loff_t bytes = inode_get_bytes(inode);
1212         loff_t real_space = real_space_diff(inode, sd_size);
1213
1214         /* keeps fsck and non-quota versions of reiserfs happy */
1215         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1216                 bytes += (loff_t) 511;
1217         }
1218
1219         /*
1220          * files from before the quota patch might i_blocks such that
1221          * bytes < real_space.  Deal with that here to prevent it from
1222          * going negative.
1223          */
1224         if (bytes < real_space)
1225                 return 0;
1226         return (bytes - real_space) >> 9;
1227 }
1228
1229 /*
1230  * BAD: new directories have stat data of new type and all other items
1231  * of old type. Version stored in the inode says about body items, so
1232  * in update_stat_data we can not rely on inode, but have to check
1233  * item version directly
1234  */
1235
1236 /* called by read_locked_inode */
1237 static void init_inode(struct inode *inode, struct treepath *path)
1238 {
1239         struct buffer_head *bh;
1240         struct item_head *ih;
1241         __u32 rdev;
1242
1243         bh = PATH_PLAST_BUFFER(path);
1244         ih = tp_item_head(path);
1245
1246         copy_key(INODE_PKEY(inode), &ih->ih_key);
1247
1248         INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
1249         REISERFS_I(inode)->i_flags = 0;
1250         REISERFS_I(inode)->i_prealloc_block = 0;
1251         REISERFS_I(inode)->i_prealloc_count = 0;
1252         REISERFS_I(inode)->i_trans_id = 0;
1253         REISERFS_I(inode)->i_jl = NULL;
1254         reiserfs_init_xattr_rwsem(inode);
1255
1256         if (stat_data_v1(ih)) {
1257                 struct stat_data_v1 *sd =
1258                     (struct stat_data_v1 *)ih_item_body(bh, ih);
1259                 unsigned long blocks;
1260
1261                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1262                 set_inode_sd_version(inode, STAT_DATA_V1);
1263                 inode->i_mode = sd_v1_mode(sd);
1264                 set_nlink(inode, sd_v1_nlink(sd));
1265                 i_uid_write(inode, sd_v1_uid(sd));
1266                 i_gid_write(inode, sd_v1_gid(sd));
1267                 inode->i_size = sd_v1_size(sd);
1268                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1269                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1270                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1271                 inode->i_atime.tv_nsec = 0;
1272                 inode->i_ctime.tv_nsec = 0;
1273                 inode->i_mtime.tv_nsec = 0;
1274
1275                 inode->i_blocks = sd_v1_blocks(sd);
1276                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1277                 blocks = (inode->i_size + 511) >> 9;
1278                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1279
1280                 /*
1281                  * there was a bug in <=3.5.23 when i_blocks could take
1282                  * negative values. Starting from 3.5.17 this value could
1283                  * even be stored in stat data. For such files we set
1284                  * i_blocks based on file size. Just 2 notes: this can be
1285                  * wrong for sparse files. On-disk value will be only
1286                  * updated if file's inode will ever change
1287                  */
1288                 if (inode->i_blocks > blocks) {
1289                         inode->i_blocks = blocks;
1290                 }
1291
1292                 rdev = sd_v1_rdev(sd);
1293                 REISERFS_I(inode)->i_first_direct_byte =
1294                     sd_v1_first_direct_byte(sd);
1295
1296                 /*
1297                  * an early bug in the quota code can give us an odd
1298                  * number for the block count.  This is incorrect, fix it here.
1299                  */
1300                 if (inode->i_blocks & 1) {
1301                         inode->i_blocks++;
1302                 }
1303                 inode_set_bytes(inode,
1304                                 to_real_used_space(inode, inode->i_blocks,
1305                                                    SD_V1_SIZE));
1306                 /*
1307                  * nopack is initially zero for v1 objects. For v2 objects,
1308                  * nopack is initialised from sd_attrs
1309                  */
1310                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1311         } else {
1312                 /*
1313                  * new stat data found, but object may have old items
1314                  * (directories and symlinks)
1315                  */
1316                 struct stat_data *sd = (struct stat_data *)ih_item_body(bh, ih);
1317
1318                 inode->i_mode = sd_v2_mode(sd);
1319                 set_nlink(inode, sd_v2_nlink(sd));
1320                 i_uid_write(inode, sd_v2_uid(sd));
1321                 inode->i_size = sd_v2_size(sd);
1322                 i_gid_write(inode, sd_v2_gid(sd));
1323                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1324                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1325                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1326                 inode->i_ctime.tv_nsec = 0;
1327                 inode->i_mtime.tv_nsec = 0;
1328                 inode->i_atime.tv_nsec = 0;
1329                 inode->i_blocks = sd_v2_blocks(sd);
1330                 rdev = sd_v2_rdev(sd);
1331                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1332                         inode->i_generation =
1333                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1334                 else
1335                         inode->i_generation = sd_v2_generation(sd);
1336
1337                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1338                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1339                 else
1340                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1341                 REISERFS_I(inode)->i_first_direct_byte = 0;
1342                 set_inode_sd_version(inode, STAT_DATA_V2);
1343                 inode_set_bytes(inode,
1344                                 to_real_used_space(inode, inode->i_blocks,
1345                                                    SD_V2_SIZE));
1346                 /*
1347                  * read persistent inode attributes from sd and initialise
1348                  * generic inode flags from them
1349                  */
1350                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1351                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1352         }
1353
1354         pathrelse(path);
1355         if (S_ISREG(inode->i_mode)) {
1356                 inode->i_op = &reiserfs_file_inode_operations;
1357                 inode->i_fop = &reiserfs_file_operations;
1358                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1359         } else if (S_ISDIR(inode->i_mode)) {
1360                 inode->i_op = &reiserfs_dir_inode_operations;
1361                 inode->i_fop = &reiserfs_dir_operations;
1362         } else if (S_ISLNK(inode->i_mode)) {
1363                 inode->i_op = &reiserfs_symlink_inode_operations;
1364                 inode_nohighmem(inode);
1365                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1366         } else {
1367                 inode->i_blocks = 0;
1368                 inode->i_op = &reiserfs_special_inode_operations;
1369                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1370         }
1371 }
1372
1373 /* update new stat data with inode fields */
1374 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1375 {
1376         struct stat_data *sd_v2 = (struct stat_data *)sd;
1377
1378         set_sd_v2_mode(sd_v2, inode->i_mode);
1379         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1380         set_sd_v2_uid(sd_v2, i_uid_read(inode));
1381         set_sd_v2_size(sd_v2, size);
1382         set_sd_v2_gid(sd_v2, i_gid_read(inode));
1383         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1384         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1385         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1386         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1387         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1388                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1389         else
1390                 set_sd_v2_generation(sd_v2, inode->i_generation);
1391         set_sd_v2_attrs(sd_v2, REISERFS_I(inode)->i_attrs);
1392 }
1393
1394 /* used to copy inode's fields to old stat data */
1395 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1396 {
1397         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1398
1399         set_sd_v1_mode(sd_v1, inode->i_mode);
1400         set_sd_v1_uid(sd_v1, i_uid_read(inode));
1401         set_sd_v1_gid(sd_v1, i_gid_read(inode));
1402         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1403         set_sd_v1_size(sd_v1, size);
1404         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1405         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1406         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1407
1408         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1409                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1410         else
1411                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1412
1413         /* Sigh. i_first_direct_byte is back */
1414         set_sd_v1_first_direct_byte(sd_v1,
1415                                     REISERFS_I(inode)->i_first_direct_byte);
1416 }
1417
1418 /*
1419  * NOTE, you must prepare the buffer head before sending it here,
1420  * and then log it after the call
1421  */
1422 static void update_stat_data(struct treepath *path, struct inode *inode,
1423                              loff_t size)
1424 {
1425         struct buffer_head *bh;
1426         struct item_head *ih;
1427
1428         bh = PATH_PLAST_BUFFER(path);
1429         ih = tp_item_head(path);
1430
1431         if (!is_statdata_le_ih(ih))
1432                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1433                                INODE_PKEY(inode), ih);
1434
1435         /* path points to old stat data */
1436         if (stat_data_v1(ih)) {
1437                 inode2sd_v1(ih_item_body(bh, ih), inode, size);
1438         } else {
1439                 inode2sd(ih_item_body(bh, ih), inode, size);
1440         }
1441
1442         return;
1443 }
1444
1445 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1446                              struct inode *inode, loff_t size)
1447 {
1448         struct cpu_key key;
1449         INITIALIZE_PATH(path);
1450         struct buffer_head *bh;
1451         int fs_gen;
1452         struct item_head *ih, tmp_ih;
1453         int retval;
1454
1455         BUG_ON(!th->t_trans_id);
1456
1457         /* key type is unimportant */
1458         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1459
1460         for (;;) {
1461                 int pos;
1462                 /* look for the object's stat data */
1463                 retval = search_item(inode->i_sb, &key, &path);
1464                 if (retval == IO_ERROR) {
1465                         reiserfs_error(inode->i_sb, "vs-13050",
1466                                        "i/o failure occurred trying to "
1467                                        "update %K stat data", &key);
1468                         return;
1469                 }
1470                 if (retval == ITEM_NOT_FOUND) {
1471                         pos = PATH_LAST_POSITION(&path);
1472                         pathrelse(&path);
1473                         if (inode->i_nlink == 0) {
1474                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1475                                 return;
1476                         }
1477                         reiserfs_warning(inode->i_sb, "vs-13060",
1478                                          "stat data of object %k (nlink == %d) "
1479                                          "not found (pos %d)",
1480                                          INODE_PKEY(inode), inode->i_nlink,
1481                                          pos);
1482                         reiserfs_check_path(&path);
1483                         return;
1484                 }
1485
1486                 /*
1487                  * sigh, prepare_for_journal might schedule.  When it
1488                  * schedules the FS might change.  We have to detect that,
1489                  * and loop back to the search if the stat data item has moved
1490                  */
1491                 bh = get_last_bh(&path);
1492                 ih = tp_item_head(&path);
1493                 copy_item_head(&tmp_ih, ih);
1494                 fs_gen = get_generation(inode->i_sb);
1495                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1496
1497                 /* Stat_data item has been moved after scheduling. */
1498                 if (fs_changed(fs_gen, inode->i_sb)
1499                     && item_moved(&tmp_ih, &path)) {
1500                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1501                         continue;
1502                 }
1503                 break;
1504         }
1505         update_stat_data(&path, inode, size);
1506         journal_mark_dirty(th, bh);
1507         pathrelse(&path);
1508         return;
1509 }
1510
1511 /*
1512  * reiserfs_read_locked_inode is called to read the inode off disk, and it
1513  * does a make_bad_inode when things go wrong.  But, we need to make sure
1514  * and clear the key in the private portion of the inode, otherwise a
1515  * corresponding iput might try to delete whatever object the inode last
1516  * represented.
1517  */
1518 static void reiserfs_make_bad_inode(struct inode *inode)
1519 {
1520         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1521         make_bad_inode(inode);
1522 }
1523
1524 /*
1525  * initially this function was derived from minix or ext2's analog and
1526  * evolved as the prototype did
1527  */
1528 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1529 {
1530         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1531         inode->i_ino = args->objectid;
1532         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1533         return 0;
1534 }
1535
1536 /*
1537  * looks for stat data in the tree, and fills up the fields of in-core
1538  * inode stat data fields
1539  */
1540 void reiserfs_read_locked_inode(struct inode *inode,
1541                                 struct reiserfs_iget_args *args)
1542 {
1543         INITIALIZE_PATH(path_to_sd);
1544         struct cpu_key key;
1545         unsigned long dirino;
1546         int retval;
1547
1548         dirino = args->dirid;
1549
1550         /*
1551          * set version 1, version 2 could be used too, because stat data
1552          * key is the same in both versions
1553          */
1554         _make_cpu_key(&key, KEY_FORMAT_3_5, dirino, inode->i_ino, 0, 0, 3);
1555
1556         /* look for the object's stat data */
1557         retval = search_item(inode->i_sb, &key, &path_to_sd);
1558         if (retval == IO_ERROR) {
1559                 reiserfs_error(inode->i_sb, "vs-13070",
1560                                "i/o failure occurred trying to find "
1561                                "stat data of %K", &key);
1562                 reiserfs_make_bad_inode(inode);
1563                 return;
1564         }
1565
1566         /* a stale NFS handle can trigger this without it being an error */
1567         if (retval != ITEM_FOUND) {
1568                 pathrelse(&path_to_sd);
1569                 reiserfs_make_bad_inode(inode);
1570                 clear_nlink(inode);
1571                 return;
1572         }
1573
1574         init_inode(inode, &path_to_sd);
1575
1576         /*
1577          * It is possible that knfsd is trying to access inode of a file
1578          * that is being removed from the disk by some other thread. As we
1579          * update sd on unlink all that is required is to check for nlink
1580          * here. This bug was first found by Sizif when debugging
1581          * SquidNG/Butterfly, forgotten, and found again after Philippe
1582          * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1583
1584          * More logical fix would require changes in fs/inode.c:iput() to
1585          * remove inode from hash-table _after_ fs cleaned disk stuff up and
1586          * in iget() to return NULL if I_FREEING inode is found in
1587          * hash-table.
1588          */
1589
1590         /*
1591          * Currently there is one place where it's ok to meet inode with
1592          * nlink==0: processing of open-unlinked and half-truncated files
1593          * during mount (fs/reiserfs/super.c:finish_unfinished()).
1594          */
1595         if ((inode->i_nlink == 0) &&
1596             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1597                 reiserfs_warning(inode->i_sb, "vs-13075",
1598                                  "dead inode read from disk %K. "
1599                                  "This is likely to be race with knfsd. Ignore",
1600                                  &key);
1601                 reiserfs_make_bad_inode(inode);
1602         }
1603
1604         /* init inode should be relsing */
1605         reiserfs_check_path(&path_to_sd);
1606
1607         /*
1608          * Stat data v1 doesn't support ACLs.
1609          */
1610         if (get_inode_sd_version(inode) == STAT_DATA_V1)
1611                 cache_no_acl(inode);
1612 }
1613
1614 /*
1615  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1616  *
1617  * @inode:    inode from hash table to check
1618  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1619  *
1620  * This function is called by iget5_locked() to distinguish reiserfs inodes
1621  * having the same inode numbers. Such inodes can only exist due to some
1622  * error condition. One of them should be bad. Inodes with identical
1623  * inode numbers (objectids) are distinguished by parent directory ids.
1624  *
1625  */
1626 int reiserfs_find_actor(struct inode *inode, void *opaque)
1627 {
1628         struct reiserfs_iget_args *args;
1629
1630         args = opaque;
1631         /* args is already in CPU order */
1632         return (inode->i_ino == args->objectid) &&
1633             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1634 }
1635
1636 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1637 {
1638         struct inode *inode;
1639         struct reiserfs_iget_args args;
1640         int depth;
1641
1642         args.objectid = key->on_disk_key.k_objectid;
1643         args.dirid = key->on_disk_key.k_dir_id;
1644         depth = reiserfs_write_unlock_nested(s);
1645         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1646                              reiserfs_find_actor, reiserfs_init_locked_inode,
1647                              (void *)(&args));
1648         reiserfs_write_lock_nested(s, depth);
1649         if (!inode)
1650                 return ERR_PTR(-ENOMEM);
1651
1652         if (inode->i_state & I_NEW) {
1653                 reiserfs_read_locked_inode(inode, &args);
1654                 unlock_new_inode(inode);
1655         }
1656
1657         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1658                 /* either due to i/o error or a stale NFS handle */
1659                 iput(inode);
1660                 inode = NULL;
1661         }
1662         return inode;
1663 }
1664
1665 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1666         u32 objectid, u32 dir_id, u32 generation)
1667
1668 {
1669         struct cpu_key key;
1670         struct inode *inode;
1671
1672         key.on_disk_key.k_objectid = objectid;
1673         key.on_disk_key.k_dir_id = dir_id;
1674         reiserfs_write_lock(sb);
1675         inode = reiserfs_iget(sb, &key);
1676         if (inode && !IS_ERR(inode) && generation != 0 &&
1677             generation != inode->i_generation) {
1678                 iput(inode);
1679                 inode = NULL;
1680         }
1681         reiserfs_write_unlock(sb);
1682
1683         return d_obtain_alias(inode);
1684 }
1685
1686 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1687                 int fh_len, int fh_type)
1688 {
1689         /*
1690          * fhtype happens to reflect the number of u32s encoded.
1691          * due to a bug in earlier code, fhtype might indicate there
1692          * are more u32s then actually fitted.
1693          * so if fhtype seems to be more than len, reduce fhtype.
1694          * Valid types are:
1695          *   2 - objectid + dir_id - legacy support
1696          *   3 - objectid + dir_id + generation
1697          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1698          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1699          *   6 - as above plus generation of directory
1700          * 6 does not fit in NFSv2 handles
1701          */
1702         if (fh_type > fh_len) {
1703                 if (fh_type != 6 || fh_len != 5)
1704                         reiserfs_warning(sb, "reiserfs-13077",
1705                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1706                                 fh_type, fh_len);
1707                 fh_type = fh_len;
1708         }
1709         if (fh_len < 2)
1710                 return NULL;
1711
1712         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1713                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1714 }
1715
1716 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1717                 int fh_len, int fh_type)
1718 {
1719         if (fh_type > fh_len)
1720                 fh_type = fh_len;
1721         if (fh_type < 4)
1722                 return NULL;
1723
1724         return reiserfs_get_dentry(sb,
1725                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1726                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1727                 (fh_type == 6) ? fid->raw[5] : 0);
1728 }
1729
1730 int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1731                        struct inode *parent)
1732 {
1733         int maxlen = *lenp;
1734
1735         if (parent && (maxlen < 5)) {
1736                 *lenp = 5;
1737                 return FILEID_INVALID;
1738         } else if (maxlen < 3) {
1739                 *lenp = 3;
1740                 return FILEID_INVALID;
1741         }
1742
1743         data[0] = inode->i_ino;
1744         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1745         data[2] = inode->i_generation;
1746         *lenp = 3;
1747         if (parent) {
1748                 data[3] = parent->i_ino;
1749                 data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1750                 *lenp = 5;
1751                 if (maxlen >= 6) {
1752                         data[5] = parent->i_generation;
1753                         *lenp = 6;
1754                 }
1755         }
1756         return *lenp;
1757 }
1758
1759 /*
1760  * looks for stat data, then copies fields to it, marks the buffer
1761  * containing stat data as dirty
1762  */
1763 /*
1764  * reiserfs inodes are never really dirty, since the dirty inode call
1765  * always logs them.  This call allows the VFS inode marking routines
1766  * to properly mark inodes for datasync and such, but only actually
1767  * does something when called for a synchronous update.
1768  */
1769 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1770 {
1771         struct reiserfs_transaction_handle th;
1772         int jbegin_count = 1;
1773
1774         if (sb_rdonly(inode->i_sb))
1775                 return -EROFS;
1776         /*
1777          * memory pressure can sometimes initiate write_inode calls with
1778          * sync == 1,
1779          * these cases are just when the system needs ram, not when the
1780          * inode needs to reach disk for safety, and they can safely be
1781          * ignored because the altered inode has already been logged.
1782          */
1783         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1784                 reiserfs_write_lock(inode->i_sb);
1785                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1786                         reiserfs_update_sd(&th, inode);
1787                         journal_end_sync(&th);
1788                 }
1789                 reiserfs_write_unlock(inode->i_sb);
1790         }
1791         return 0;
1792 }
1793
1794 /*
1795  * stat data of new object is inserted already, this inserts the item
1796  * containing "." and ".." entries
1797  */
1798 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1799                                   struct inode *inode,
1800                                   struct item_head *ih, struct treepath *path,
1801                                   struct inode *dir)
1802 {
1803         struct super_block *sb = th->t_super;
1804         char empty_dir[EMPTY_DIR_SIZE];
1805         char *body = empty_dir;
1806         struct cpu_key key;
1807         int retval;
1808
1809         BUG_ON(!th->t_trans_id);
1810
1811         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1812                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1813                       TYPE_DIRENTRY, 3 /*key length */ );
1814
1815         /*
1816          * compose item head for new item. Directories consist of items of
1817          * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1818          * is done by reiserfs_new_inode
1819          */
1820         if (old_format_only(sb)) {
1821                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1822                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1823
1824                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1825                                        ih->ih_key.k_objectid,
1826                                        INODE_PKEY(dir)->k_dir_id,
1827                                        INODE_PKEY(dir)->k_objectid);
1828         } else {
1829                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1830                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1831
1832                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1833                                     ih->ih_key.k_objectid,
1834                                     INODE_PKEY(dir)->k_dir_id,
1835                                     INODE_PKEY(dir)->k_objectid);
1836         }
1837
1838         /* look for place in the tree for new item */
1839         retval = search_item(sb, &key, path);
1840         if (retval == IO_ERROR) {
1841                 reiserfs_error(sb, "vs-13080",
1842                                "i/o failure occurred creating new directory");
1843                 return -EIO;
1844         }
1845         if (retval == ITEM_FOUND) {
1846                 pathrelse(path);
1847                 reiserfs_warning(sb, "vs-13070",
1848                                  "object with this key exists (%k)",
1849                                  &(ih->ih_key));
1850                 return -EEXIST;
1851         }
1852
1853         /* insert item, that is empty directory item */
1854         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1855 }
1856
1857 /*
1858  * stat data of object has been inserted, this inserts the item
1859  * containing the body of symlink
1860  */
1861 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1862                                 struct inode *inode,
1863                                 struct item_head *ih,
1864                                 struct treepath *path, const char *symname,
1865                                 int item_len)
1866 {
1867         struct super_block *sb = th->t_super;
1868         struct cpu_key key;
1869         int retval;
1870
1871         BUG_ON(!th->t_trans_id);
1872
1873         _make_cpu_key(&key, KEY_FORMAT_3_5,
1874                       le32_to_cpu(ih->ih_key.k_dir_id),
1875                       le32_to_cpu(ih->ih_key.k_objectid),
1876                       1, TYPE_DIRECT, 3 /*key length */ );
1877
1878         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1879                           0 /*free_space */ );
1880
1881         /* look for place in the tree for new item */
1882         retval = search_item(sb, &key, path);
1883         if (retval == IO_ERROR) {
1884                 reiserfs_error(sb, "vs-13080",
1885                                "i/o failure occurred creating new symlink");
1886                 return -EIO;
1887         }
1888         if (retval == ITEM_FOUND) {
1889                 pathrelse(path);
1890                 reiserfs_warning(sb, "vs-13080",
1891                                  "object with this key exists (%k)",
1892                                  &(ih->ih_key));
1893                 return -EEXIST;
1894         }
1895
1896         /* insert item, that is body of symlink */
1897         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1898 }
1899
1900 /*
1901  * inserts the stat data into the tree, and then calls
1902  * reiserfs_new_directory (to insert ".", ".." item if new object is
1903  * directory) or reiserfs_new_symlink (to insert symlink body if new
1904  * object is symlink) or nothing (if new object is regular file)
1905
1906  * NOTE! uid and gid must already be set in the inode.  If we return
1907  * non-zero due to an error, we have to drop the quota previously allocated
1908  * for the fresh inode.  This can only be done outside a transaction, so
1909  * if we return non-zero, we also end the transaction.
1910  *
1911  * @th: active transaction handle
1912  * @dir: parent directory for new inode
1913  * @mode: mode of new inode
1914  * @symname: symlink contents if inode is symlink
1915  * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1916  *         symlinks
1917  * @inode: inode to be filled
1918  * @security: optional security context to associate with this inode
1919  */
1920 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1921                        struct inode *dir, umode_t mode, const char *symname,
1922                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1923                           strlen (symname) for symlinks) */
1924                        loff_t i_size, struct dentry *dentry,
1925                        struct inode *inode,
1926                        struct reiserfs_security_handle *security)
1927 {
1928         struct super_block *sb = dir->i_sb;
1929         struct reiserfs_iget_args args;
1930         INITIALIZE_PATH(path_to_key);
1931         struct cpu_key key;
1932         struct item_head ih;
1933         struct stat_data sd;
1934         int retval;
1935         int err;
1936         int depth;
1937
1938         BUG_ON(!th->t_trans_id);
1939
1940         depth = reiserfs_write_unlock_nested(sb);
1941         err = dquot_alloc_inode(inode);
1942         reiserfs_write_lock_nested(sb, depth);
1943         if (err)
1944                 goto out_end_trans;
1945         if (!dir->i_nlink) {
1946                 err = -EPERM;
1947                 goto out_bad_inode;
1948         }
1949
1950         /* item head of new item */
1951         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1952         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1953         if (!ih.ih_key.k_objectid) {
1954                 err = -ENOMEM;
1955                 goto out_bad_inode;
1956         }
1957         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1958         if (old_format_only(sb))
1959                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1960                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1961         else
1962                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1963                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1964         memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1965         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1966
1967         depth = reiserfs_write_unlock_nested(inode->i_sb);
1968         err = insert_inode_locked4(inode, args.objectid,
1969                              reiserfs_find_actor, &args);
1970         reiserfs_write_lock_nested(inode->i_sb, depth);
1971         if (err) {
1972                 err = -EINVAL;
1973                 goto out_bad_inode;
1974         }
1975
1976         if (old_format_only(sb))
1977                 /*
1978                  * not a perfect generation count, as object ids can be reused,
1979                  * but this is as good as reiserfs can do right now.
1980                  * note that the private part of inode isn't filled in yet,
1981                  * we have to use the directory.
1982                  */
1983                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1984         else
1985 #if defined( USE_INODE_GENERATION_COUNTER )
1986                 inode->i_generation =
1987                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1988 #else
1989                 inode->i_generation = ++event;
1990 #endif
1991
1992         /* fill stat data */
1993         set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
1994
1995         /* uid and gid must already be set by the caller for quota init */
1996
1997         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1998         inode->i_size = i_size;
1999         inode->i_blocks = 0;
2000         inode->i_bytes = 0;
2001         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
2002             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
2003
2004         INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
2005         REISERFS_I(inode)->i_flags = 0;
2006         REISERFS_I(inode)->i_prealloc_block = 0;
2007         REISERFS_I(inode)->i_prealloc_count = 0;
2008         REISERFS_I(inode)->i_trans_id = 0;
2009         REISERFS_I(inode)->i_jl = NULL;
2010         REISERFS_I(inode)->i_attrs =
2011             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2012         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2013         reiserfs_init_xattr_rwsem(inode);
2014
2015         /* key to search for correct place for new stat data */
2016         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2017                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2018                       TYPE_STAT_DATA, 3 /*key length */ );
2019
2020         /* find proper place for inserting of stat data */
2021         retval = search_item(sb, &key, &path_to_key);
2022         if (retval == IO_ERROR) {
2023                 err = -EIO;
2024                 goto out_bad_inode;
2025         }
2026         if (retval == ITEM_FOUND) {
2027                 pathrelse(&path_to_key);
2028                 err = -EEXIST;
2029                 goto out_bad_inode;
2030         }
2031         if (old_format_only(sb)) {
2032                 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
2033                 if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2034                         pathrelse(&path_to_key);
2035                         err = -EINVAL;
2036                         goto out_bad_inode;
2037                 }
2038                 inode2sd_v1(&sd, inode, inode->i_size);
2039         } else {
2040                 inode2sd(&sd, inode, inode->i_size);
2041         }
2042         /*
2043          * store in in-core inode the key of stat data and version all
2044          * object items will have (directory items will have old offset
2045          * format, other new objects will consist of new items)
2046          */
2047         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2048                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2049         else
2050                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2051         if (old_format_only(sb))
2052                 set_inode_sd_version(inode, STAT_DATA_V1);
2053         else
2054                 set_inode_sd_version(inode, STAT_DATA_V2);
2055
2056         /* insert the stat data into the tree */
2057 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2058         if (REISERFS_I(dir)->new_packing_locality)
2059                 th->displace_new_blocks = 1;
2060 #endif
2061         retval =
2062             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2063                                  (char *)(&sd));
2064         if (retval) {
2065                 err = retval;
2066                 reiserfs_check_path(&path_to_key);
2067                 goto out_bad_inode;
2068         }
2069 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2070         if (!th->displace_new_blocks)
2071                 REISERFS_I(dir)->new_packing_locality = 0;
2072 #endif
2073         if (S_ISDIR(mode)) {
2074                 /* insert item with "." and ".." */
2075                 retval =
2076                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2077         }
2078
2079         if (S_ISLNK(mode)) {
2080                 /* insert body of symlink */
2081                 if (!old_format_only(sb))
2082                         i_size = ROUND_UP(i_size);
2083                 retval =
2084                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2085                                          i_size);
2086         }
2087         if (retval) {
2088                 err = retval;
2089                 reiserfs_check_path(&path_to_key);
2090                 journal_end(th);
2091                 goto out_inserted_sd;
2092         }
2093
2094         /*
2095          * Mark it private if we're creating the privroot
2096          * or something under it.
2097          */
2098         if (IS_PRIVATE(dir) || dentry == REISERFS_SB(sb)->priv_root) {
2099                 inode->i_flags |= S_PRIVATE;
2100                 inode->i_opflags &= ~IOP_XATTR;
2101         }
2102
2103         if (reiserfs_posixacl(inode->i_sb)) {
2104                 reiserfs_write_unlock(inode->i_sb);
2105                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2106                 reiserfs_write_lock(inode->i_sb);
2107                 if (retval) {
2108                         err = retval;
2109                         reiserfs_check_path(&path_to_key);
2110                         journal_end(th);
2111                         goto out_inserted_sd;
2112                 }
2113         } else if (inode->i_sb->s_flags & SB_POSIXACL) {
2114                 reiserfs_warning(inode->i_sb, "jdm-13090",
2115                                  "ACLs aren't enabled in the fs, "
2116                                  "but vfs thinks they are!");
2117         }
2118
2119         if (security->name) {
2120                 reiserfs_write_unlock(inode->i_sb);
2121                 retval = reiserfs_security_write(th, inode, security);
2122                 reiserfs_write_lock(inode->i_sb);
2123                 if (retval) {
2124                         err = retval;
2125                         reiserfs_check_path(&path_to_key);
2126                         retval = journal_end(th);
2127                         if (retval)
2128                                 err = retval;
2129                         goto out_inserted_sd;
2130                 }
2131         }
2132
2133         reiserfs_update_sd(th, inode);
2134         reiserfs_check_path(&path_to_key);
2135
2136         return 0;
2137
2138 out_bad_inode:
2139         /* Invalidate the object, nothing was inserted yet */
2140         INODE_PKEY(inode)->k_objectid = 0;
2141
2142         /* Quota change must be inside a transaction for journaling */
2143         depth = reiserfs_write_unlock_nested(inode->i_sb);
2144         dquot_free_inode(inode);
2145         reiserfs_write_lock_nested(inode->i_sb, depth);
2146
2147 out_end_trans:
2148         journal_end(th);
2149         /*
2150          * Drop can be outside and it needs more credits so it's better
2151          * to have it outside
2152          */
2153         depth = reiserfs_write_unlock_nested(inode->i_sb);
2154         dquot_drop(inode);
2155         reiserfs_write_lock_nested(inode->i_sb, depth);
2156         inode->i_flags |= S_NOQUOTA;
2157         make_bad_inode(inode);
2158
2159 out_inserted_sd:
2160         clear_nlink(inode);
2161         th->t_trans_id = 0;     /* so the caller can't use this handle later */
2162         if (inode->i_state & I_NEW)
2163                 unlock_new_inode(inode);
2164         iput(inode);
2165         return err;
2166 }
2167
2168 /*
2169  * finds the tail page in the page cache,
2170  * reads the last block in.
2171  *
2172  * On success, page_result is set to a locked, pinned page, and bh_result
2173  * is set to an up to date buffer for the last block in the file.  returns 0.
2174  *
2175  * tail conversion is not done, so bh_result might not be valid for writing
2176  * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2177  * trying to write the block.
2178  *
2179  * on failure, nonzero is returned, page_result and bh_result are untouched.
2180  */
2181 static int grab_tail_page(struct inode *inode,
2182                           struct page **page_result,
2183                           struct buffer_head **bh_result)
2184 {
2185
2186         /*
2187          * we want the page with the last byte in the file,
2188          * not the page that will hold the next byte for appending
2189          */
2190         unsigned long index = (inode->i_size - 1) >> PAGE_SHIFT;
2191         unsigned long pos = 0;
2192         unsigned long start = 0;
2193         unsigned long blocksize = inode->i_sb->s_blocksize;
2194         unsigned long offset = (inode->i_size) & (PAGE_SIZE - 1);
2195         struct buffer_head *bh;
2196         struct buffer_head *head;
2197         struct page *page;
2198         int error;
2199
2200         /*
2201          * we know that we are only called with inode->i_size > 0.
2202          * we also know that a file tail can never be as big as a block
2203          * If i_size % blocksize == 0, our file is currently block aligned
2204          * and it won't need converting or zeroing after a truncate.
2205          */
2206         if ((offset & (blocksize - 1)) == 0) {
2207                 return -ENOENT;
2208         }
2209         page = grab_cache_page(inode->i_mapping, index);
2210         error = -ENOMEM;
2211         if (!page) {
2212                 goto out;
2213         }
2214         /* start within the page of the last block in the file */
2215         start = (offset / blocksize) * blocksize;
2216
2217         error = __block_write_begin(page, start, offset - start,
2218                                     reiserfs_get_block_create_0);
2219         if (error)
2220                 goto unlock;
2221
2222         head = page_buffers(page);
2223         bh = head;
2224         do {
2225                 if (pos >= start) {
2226                         break;
2227                 }
2228                 bh = bh->b_this_page;
2229                 pos += blocksize;
2230         } while (bh != head);
2231
2232         if (!buffer_uptodate(bh)) {
2233                 /*
2234                  * note, this should never happen, prepare_write should be
2235                  * taking care of this for us.  If the buffer isn't up to
2236                  * date, I've screwed up the code to find the buffer, or the
2237                  * code to call prepare_write
2238                  */
2239                 reiserfs_error(inode->i_sb, "clm-6000",
2240                                "error reading block %lu", bh->b_blocknr);
2241                 error = -EIO;
2242                 goto unlock;
2243         }
2244         *bh_result = bh;
2245         *page_result = page;
2246
2247 out:
2248         return error;
2249
2250 unlock:
2251         unlock_page(page);
2252         put_page(page);
2253         return error;
2254 }
2255
2256 /*
2257  * vfs version of truncate file.  Must NOT be called with
2258  * a transaction already started.
2259  *
2260  * some code taken from block_truncate_page
2261  */
2262 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2263 {
2264         struct reiserfs_transaction_handle th;
2265         /* we want the offset for the first byte after the end of the file */
2266         unsigned long offset = inode->i_size & (PAGE_SIZE - 1);
2267         unsigned blocksize = inode->i_sb->s_blocksize;
2268         unsigned length;
2269         struct page *page = NULL;
2270         int error;
2271         struct buffer_head *bh = NULL;
2272         int err2;
2273
2274         reiserfs_write_lock(inode->i_sb);
2275
2276         if (inode->i_size > 0) {
2277                 error = grab_tail_page(inode, &page, &bh);
2278                 if (error) {
2279                         /*
2280                          * -ENOENT means we truncated past the end of the
2281                          * file, and get_block_create_0 could not find a
2282                          * block to read in, which is ok.
2283                          */
2284                         if (error != -ENOENT)
2285                                 reiserfs_error(inode->i_sb, "clm-6001",
2286                                                "grab_tail_page failed %d",
2287                                                error);
2288                         page = NULL;
2289                         bh = NULL;
2290                 }
2291         }
2292
2293         /*
2294          * so, if page != NULL, we have a buffer head for the offset at
2295          * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2296          * then we have an unformatted node.  Otherwise, we have a direct item,
2297          * and no zeroing is required on disk.  We zero after the truncate,
2298          * because the truncate might pack the item anyway
2299          * (it will unmap bh if it packs).
2300          *
2301          * it is enough to reserve space in transaction for 2 balancings:
2302          * one for "save" link adding and another for the first
2303          * cut_from_item. 1 is for update_sd
2304          */
2305         error = journal_begin(&th, inode->i_sb,
2306                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2307         if (error)
2308                 goto out;
2309         reiserfs_update_inode_transaction(inode);
2310         if (update_timestamps)
2311                 /*
2312                  * we are doing real truncate: if the system crashes
2313                  * before the last transaction of truncating gets committed
2314                  * - on reboot the file either appears truncated properly
2315                  * or not truncated at all
2316                  */
2317                 add_save_link(&th, inode, 1);
2318         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2319         error = journal_end(&th);
2320         if (error)
2321                 goto out;
2322
2323         /* check reiserfs_do_truncate after ending the transaction */
2324         if (err2) {
2325                 error = err2;
2326                 goto out;
2327         }
2328
2329         if (update_timestamps) {
2330                 error = remove_save_link(inode, 1 /* truncate */);
2331                 if (error)
2332                         goto out;
2333         }
2334
2335         if (page) {
2336                 length = offset & (blocksize - 1);
2337                 /* if we are not on a block boundary */
2338                 if (length) {
2339                         length = blocksize - length;
2340                         zero_user(page, offset, length);
2341                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2342                                 mark_buffer_dirty(bh);
2343                         }
2344                 }
2345                 unlock_page(page);
2346                 put_page(page);
2347         }
2348
2349         reiserfs_write_unlock(inode->i_sb);
2350
2351         return 0;
2352 out:
2353         if (page) {
2354                 unlock_page(page);
2355                 put_page(page);
2356         }
2357
2358         reiserfs_write_unlock(inode->i_sb);
2359
2360         return error;
2361 }
2362
2363 static int map_block_for_writepage(struct inode *inode,
2364                                    struct buffer_head *bh_result,
2365                                    unsigned long block)
2366 {
2367         struct reiserfs_transaction_handle th;
2368         int fs_gen;
2369         struct item_head tmp_ih;
2370         struct item_head *ih;
2371         struct buffer_head *bh;
2372         __le32 *item;
2373         struct cpu_key key;
2374         INITIALIZE_PATH(path);
2375         int pos_in_item;
2376         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2377         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2378         int retval;
2379         int use_get_block = 0;
2380         int bytes_copied = 0;
2381         int copy_size;
2382         int trans_running = 0;
2383
2384         /*
2385          * catch places below that try to log something without
2386          * starting a trans
2387          */
2388         th.t_trans_id = 0;
2389
2390         if (!buffer_uptodate(bh_result)) {
2391                 return -EIO;
2392         }
2393
2394         kmap(bh_result->b_page);
2395 start_over:
2396         reiserfs_write_lock(inode->i_sb);
2397         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2398
2399 research:
2400         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2401         if (retval != POSITION_FOUND) {
2402                 use_get_block = 1;
2403                 goto out;
2404         }
2405
2406         bh = get_last_bh(&path);
2407         ih = tp_item_head(&path);
2408         item = tp_item_body(&path);
2409         pos_in_item = path.pos_in_item;
2410
2411         /* we've found an unformatted node */
2412         if (indirect_item_found(retval, ih)) {
2413                 if (bytes_copied > 0) {
2414                         reiserfs_warning(inode->i_sb, "clm-6002",
2415                                          "bytes_copied %d", bytes_copied);
2416                 }
2417                 if (!get_block_num(item, pos_in_item)) {
2418                         /* crap, we are writing to a hole */
2419                         use_get_block = 1;
2420                         goto out;
2421                 }
2422                 set_block_dev_mapped(bh_result,
2423                                      get_block_num(item, pos_in_item), inode);
2424         } else if (is_direct_le_ih(ih)) {
2425                 char *p;
2426                 p = page_address(bh_result->b_page);
2427                 p += (byte_offset - 1) & (PAGE_SIZE - 1);
2428                 copy_size = ih_item_len(ih) - pos_in_item;
2429
2430                 fs_gen = get_generation(inode->i_sb);
2431                 copy_item_head(&tmp_ih, ih);
2432
2433                 if (!trans_running) {
2434                         /* vs-3050 is gone, no need to drop the path */
2435                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2436                         if (retval)
2437                                 goto out;
2438                         reiserfs_update_inode_transaction(inode);
2439                         trans_running = 1;
2440                         if (fs_changed(fs_gen, inode->i_sb)
2441                             && item_moved(&tmp_ih, &path)) {
2442                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2443                                                                  bh);
2444                                 goto research;
2445                         }
2446                 }
2447
2448                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2449
2450                 if (fs_changed(fs_gen, inode->i_sb)
2451                     && item_moved(&tmp_ih, &path)) {
2452                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2453                         goto research;
2454                 }
2455
2456                 memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2457                        copy_size);
2458
2459                 journal_mark_dirty(&th, bh);
2460                 bytes_copied += copy_size;
2461                 set_block_dev_mapped(bh_result, 0, inode);
2462
2463                 /* are there still bytes left? */
2464                 if (bytes_copied < bh_result->b_size &&
2465                     (byte_offset + bytes_copied) < inode->i_size) {
2466                         set_cpu_key_k_offset(&key,
2467                                              cpu_key_k_offset(&key) +
2468                                              copy_size);
2469                         goto research;
2470                 }
2471         } else {
2472                 reiserfs_warning(inode->i_sb, "clm-6003",
2473                                  "bad item inode %lu", inode->i_ino);
2474                 retval = -EIO;
2475                 goto out;
2476         }
2477         retval = 0;
2478
2479 out:
2480         pathrelse(&path);
2481         if (trans_running) {
2482                 int err = journal_end(&th);
2483                 if (err)
2484                         retval = err;
2485                 trans_running = 0;
2486         }
2487         reiserfs_write_unlock(inode->i_sb);
2488
2489         /* this is where we fill in holes in the file. */
2490         if (use_get_block) {
2491                 retval = reiserfs_get_block(inode, block, bh_result,
2492                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2493                                             | GET_BLOCK_NO_DANGLE);
2494                 if (!retval) {
2495                         if (!buffer_mapped(bh_result)
2496                             || bh_result->b_blocknr == 0) {
2497                                 /* get_block failed to find a mapped unformatted node. */
2498                                 use_get_block = 0;
2499                                 goto start_over;
2500                         }
2501                 }
2502         }
2503         kunmap(bh_result->b_page);
2504
2505         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2506                 /*
2507                  * we've copied data from the page into the direct item, so the
2508                  * buffer in the page is now clean, mark it to reflect that.
2509                  */
2510                 lock_buffer(bh_result);
2511                 clear_buffer_dirty(bh_result);
2512                 unlock_buffer(bh_result);
2513         }
2514         return retval;
2515 }
2516
2517 /*
2518  * mason@suse.com: updated in 2.5.54 to follow the same general io
2519  * start/recovery path as __block_write_full_page, along with special
2520  * code to handle reiserfs tails.
2521  */
2522 static int reiserfs_write_full_page(struct page *page,
2523                                     struct writeback_control *wbc)
2524 {
2525         struct inode *inode = page->mapping->host;
2526         unsigned long end_index = inode->i_size >> PAGE_SHIFT;
2527         int error = 0;
2528         unsigned long block;
2529         sector_t last_block;
2530         struct buffer_head *head, *bh;
2531         int partial = 0;
2532         int nr = 0;
2533         int checked = PageChecked(page);
2534         struct reiserfs_transaction_handle th;
2535         struct super_block *s = inode->i_sb;
2536         int bh_per_page = PAGE_SIZE / s->s_blocksize;
2537         th.t_trans_id = 0;
2538
2539         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2540         if (checked && (current->flags & PF_MEMALLOC)) {
2541                 redirty_page_for_writepage(wbc, page);
2542                 unlock_page(page);
2543                 return 0;
2544         }
2545
2546         /*
2547          * The page dirty bit is cleared before writepage is called, which
2548          * means we have to tell create_empty_buffers to make dirty buffers
2549          * The page really should be up to date at this point, so tossing
2550          * in the BH_Uptodate is just a sanity check.
2551          */
2552         if (!page_has_buffers(page)) {
2553                 create_empty_buffers(page, s->s_blocksize,
2554                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2555         }
2556         head = page_buffers(page);
2557
2558         /*
2559          * last page in the file, zero out any contents past the
2560          * last byte in the file
2561          */
2562         if (page->index >= end_index) {
2563                 unsigned last_offset;
2564
2565                 last_offset = inode->i_size & (PAGE_SIZE - 1);
2566                 /* no file contents in this page */
2567                 if (page->index >= end_index + 1 || !last_offset) {
2568                         unlock_page(page);
2569                         return 0;
2570                 }
2571                 zero_user_segment(page, last_offset, PAGE_SIZE);
2572         }
2573         bh = head;
2574         block = page->index << (PAGE_SHIFT - s->s_blocksize_bits);
2575         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2576         /* first map all the buffers, logging any direct items we find */
2577         do {
2578                 if (block > last_block) {
2579                         /*
2580                          * This can happen when the block size is less than
2581                          * the page size.  The corresponding bytes in the page
2582                          * were zero filled above
2583                          */
2584                         clear_buffer_dirty(bh);
2585                         set_buffer_uptodate(bh);
2586                 } else if ((checked || buffer_dirty(bh)) &&
2587                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2588                                                        && bh->b_blocknr ==
2589                                                        0))) {
2590                         /*
2591                          * not mapped yet, or it points to a direct item, search
2592                          * the btree for the mapping info, and log any direct
2593                          * items found
2594                          */
2595                         if ((error = map_block_for_writepage(inode, bh, block))) {
2596                                 goto fail;
2597                         }
2598                 }
2599                 bh = bh->b_this_page;
2600                 block++;
2601         } while (bh != head);
2602
2603         /*
2604          * we start the transaction after map_block_for_writepage,
2605          * because it can create holes in the file (an unbounded operation).
2606          * starting it here, we can make a reliable estimate for how many
2607          * blocks we're going to log
2608          */
2609         if (checked) {
2610                 ClearPageChecked(page);
2611                 reiserfs_write_lock(s);
2612                 error = journal_begin(&th, s, bh_per_page + 1);
2613                 if (error) {
2614                         reiserfs_write_unlock(s);
2615                         goto fail;
2616                 }
2617                 reiserfs_update_inode_transaction(inode);
2618         }
2619         /* now go through and lock any dirty buffers on the page */
2620         do {
2621                 get_bh(bh);
2622                 if (!buffer_mapped(bh))
2623                         continue;
2624                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2625                         continue;
2626
2627                 if (checked) {
2628                         reiserfs_prepare_for_journal(s, bh, 1);
2629                         journal_mark_dirty(&th, bh);
2630                         continue;
2631                 }
2632                 /*
2633                  * from this point on, we know the buffer is mapped to a
2634                  * real block and not a direct item
2635                  */
2636                 if (wbc->sync_mode != WB_SYNC_NONE) {
2637                         lock_buffer(bh);
2638                 } else {
2639                         if (!trylock_buffer(bh)) {
2640                                 redirty_page_for_writepage(wbc, page);
2641                                 continue;
2642                         }
2643                 }
2644                 if (test_clear_buffer_dirty(bh)) {
2645                         mark_buffer_async_write(bh);
2646                 } else {
2647                         unlock_buffer(bh);
2648                 }
2649         } while ((bh = bh->b_this_page) != head);
2650
2651         if (checked) {
2652                 error = journal_end(&th);
2653                 reiserfs_write_unlock(s);
2654                 if (error)
2655                         goto fail;
2656         }
2657         BUG_ON(PageWriteback(page));
2658         set_page_writeback(page);
2659         unlock_page(page);
2660
2661         /*
2662          * since any buffer might be the only dirty buffer on the page,
2663          * the first submit_bh can bring the page out of writeback.
2664          * be careful with the buffers.
2665          */
2666         do {
2667                 struct buffer_head *next = bh->b_this_page;
2668                 if (buffer_async_write(bh)) {
2669                         submit_bh(REQ_OP_WRITE, 0, bh);
2670                         nr++;
2671                 }
2672                 put_bh(bh);
2673                 bh = next;
2674         } while (bh != head);
2675
2676         error = 0;
2677 done:
2678         if (nr == 0) {
2679                 /*
2680                  * if this page only had a direct item, it is very possible for
2681                  * no io to be required without there being an error.  Or,
2682                  * someone else could have locked them and sent them down the
2683                  * pipe without locking the page
2684                  */
2685                 bh = head;
2686                 do {
2687                         if (!buffer_uptodate(bh)) {
2688                                 partial = 1;
2689                                 break;
2690                         }
2691                         bh = bh->b_this_page;
2692                 } while (bh != head);
2693                 if (!partial)
2694                         SetPageUptodate(page);
2695                 end_page_writeback(page);
2696         }
2697         return error;
2698
2699 fail:
2700         /*
2701          * catches various errors, we need to make sure any valid dirty blocks
2702          * get to the media.  The page is currently locked and not marked for
2703          * writeback
2704          */
2705         ClearPageUptodate(page);
2706         bh = head;
2707         do {
2708                 get_bh(bh);
2709                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2710                         lock_buffer(bh);
2711                         mark_buffer_async_write(bh);
2712                 } else {
2713                         /*
2714                          * clear any dirty bits that might have come from
2715                          * getting attached to a dirty page
2716                          */
2717                         clear_buffer_dirty(bh);
2718                 }
2719                 bh = bh->b_this_page;
2720         } while (bh != head);
2721         SetPageError(page);
2722         BUG_ON(PageWriteback(page));
2723         set_page_writeback(page);
2724         unlock_page(page);
2725         do {
2726                 struct buffer_head *next = bh->b_this_page;
2727                 if (buffer_async_write(bh)) {
2728                         clear_buffer_dirty(bh);
2729                         submit_bh(REQ_OP_WRITE, 0, bh);
2730                         nr++;
2731                 }
2732                 put_bh(bh);
2733                 bh = next;
2734         } while (bh != head);
2735         goto done;
2736 }
2737
2738 static int reiserfs_readpage(struct file *f, struct page *page)
2739 {
2740         return block_read_full_page(page, reiserfs_get_block);
2741 }
2742
2743 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2744 {
2745         struct inode *inode = page->mapping->host;
2746         reiserfs_wait_on_write_block(inode->i_sb);
2747         return reiserfs_write_full_page(page, wbc);
2748 }
2749
2750 static void reiserfs_truncate_failed_write(struct inode *inode)
2751 {
2752         truncate_inode_pages(inode->i_mapping, inode->i_size);
2753         reiserfs_truncate_file(inode, 0);
2754 }
2755
2756 static int reiserfs_write_begin(struct file *file,
2757                                 struct address_space *mapping,
2758                                 loff_t pos, unsigned len, unsigned flags,
2759                                 struct page **pagep, void **fsdata)
2760 {
2761         struct inode *inode;
2762         struct page *page;
2763         pgoff_t index;
2764         int ret;
2765         int old_ref = 0;
2766
2767         inode = mapping->host;
2768         *fsdata = NULL;
2769         if (flags & AOP_FLAG_CONT_EXPAND &&
2770             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2771                 pos ++;
2772                 *fsdata = (void *)(unsigned long)flags;
2773         }
2774
2775         index = pos >> PAGE_SHIFT;
2776         page = grab_cache_page_write_begin(mapping, index, flags);
2777         if (!page)
2778                 return -ENOMEM;
2779         *pagep = page;
2780
2781         reiserfs_wait_on_write_block(inode->i_sb);
2782         fix_tail_page_for_writing(page);
2783         if (reiserfs_transaction_running(inode->i_sb)) {
2784                 struct reiserfs_transaction_handle *th;
2785                 th = (struct reiserfs_transaction_handle *)current->
2786                     journal_info;
2787                 BUG_ON(!th->t_refcount);
2788                 BUG_ON(!th->t_trans_id);
2789                 old_ref = th->t_refcount;
2790                 th->t_refcount++;
2791         }
2792         ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2793         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2794                 struct reiserfs_transaction_handle *th = current->journal_info;
2795                 /*
2796                  * this gets a little ugly.  If reiserfs_get_block returned an
2797                  * error and left a transacstion running, we've got to close
2798                  * it, and we've got to free handle if it was a persistent
2799                  * transaction.
2800                  *
2801                  * But, if we had nested into an existing transaction, we need
2802                  * to just drop the ref count on the handle.
2803                  *
2804                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2805                  * and it was a persistent trans.  Otherwise, it was nested
2806                  * above.
2807                  */
2808                 if (th->t_refcount > old_ref) {
2809                         if (old_ref)
2810                                 th->t_refcount--;
2811                         else {
2812                                 int err;
2813                                 reiserfs_write_lock(inode->i_sb);
2814                                 err = reiserfs_end_persistent_transaction(th);
2815                                 reiserfs_write_unlock(inode->i_sb);
2816                                 if (err)
2817                                         ret = err;
2818                         }
2819                 }
2820         }
2821         if (ret) {
2822                 unlock_page(page);
2823                 put_page(page);
2824                 /* Truncate allocated blocks */
2825                 reiserfs_truncate_failed_write(inode);
2826         }
2827         return ret;
2828 }
2829
2830 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2831 {
2832         struct inode *inode = page->mapping->host;
2833         int ret;
2834         int old_ref = 0;
2835         int depth;
2836
2837         depth = reiserfs_write_unlock_nested(inode->i_sb);
2838         reiserfs_wait_on_write_block(inode->i_sb);
2839         reiserfs_write_lock_nested(inode->i_sb, depth);
2840
2841         fix_tail_page_for_writing(page);
2842         if (reiserfs_transaction_running(inode->i_sb)) {
2843                 struct reiserfs_transaction_handle *th;
2844                 th = (struct reiserfs_transaction_handle *)current->
2845                     journal_info;
2846                 BUG_ON(!th->t_refcount);
2847                 BUG_ON(!th->t_trans_id);
2848                 old_ref = th->t_refcount;
2849                 th->t_refcount++;
2850         }
2851
2852         ret = __block_write_begin(page, from, len, reiserfs_get_block);
2853         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2854                 struct reiserfs_transaction_handle *th = current->journal_info;
2855                 /*
2856                  * this gets a little ugly.  If reiserfs_get_block returned an
2857                  * error and left a transacstion running, we've got to close
2858                  * it, and we've got to free handle if it was a persistent
2859                  * transaction.
2860                  *
2861                  * But, if we had nested into an existing transaction, we need
2862                  * to just drop the ref count on the handle.
2863                  *
2864                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2865                  * and it was a persistent trans.  Otherwise, it was nested
2866                  * above.
2867                  */
2868                 if (th->t_refcount > old_ref) {
2869                         if (old_ref)
2870                                 th->t_refcount--;
2871                         else {
2872                                 int err;
2873                                 reiserfs_write_lock(inode->i_sb);
2874                                 err = reiserfs_end_persistent_transaction(th);
2875                                 reiserfs_write_unlock(inode->i_sb);
2876                                 if (err)
2877                                         ret = err;
2878                         }
2879                 }
2880         }
2881         return ret;
2882
2883 }
2884
2885 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2886 {
2887         return generic_block_bmap(as, block, reiserfs_bmap);
2888 }
2889
2890 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2891                               loff_t pos, unsigned len, unsigned copied,
2892                               struct page *page, void *fsdata)
2893 {
2894         struct inode *inode = page->mapping->host;
2895         int ret = 0;
2896         int update_sd = 0;
2897         struct reiserfs_transaction_handle *th;
2898         unsigned start;
2899         bool locked = false;
2900
2901         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2902                 pos ++;
2903
2904         reiserfs_wait_on_write_block(inode->i_sb);
2905         if (reiserfs_transaction_running(inode->i_sb))
2906                 th = current->journal_info;
2907         else
2908                 th = NULL;
2909
2910         start = pos & (PAGE_SIZE - 1);
2911         if (unlikely(copied < len)) {
2912                 if (!PageUptodate(page))
2913                         copied = 0;
2914
2915                 page_zero_new_buffers(page, start + copied, start + len);
2916         }
2917         flush_dcache_page(page);
2918
2919         reiserfs_commit_page(inode, page, start, start + copied);
2920
2921         /*
2922          * generic_commit_write does this for us, but does not update the
2923          * transaction tracking stuff when the size changes.  So, we have
2924          * to do the i_size updates here.
2925          */
2926         if (pos + copied > inode->i_size) {
2927                 struct reiserfs_transaction_handle myth;
2928                 reiserfs_write_lock(inode->i_sb);
2929                 locked = true;
2930                 /*
2931                  * If the file have grown beyond the border where it
2932                  * can have a tail, unmark it as needing a tail
2933                  * packing
2934                  */
2935                 if ((have_large_tails(inode->i_sb)
2936                      && inode->i_size > i_block_size(inode) * 4)
2937                     || (have_small_tails(inode->i_sb)
2938                         && inode->i_size > i_block_size(inode)))
2939                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2940
2941                 ret = journal_begin(&myth, inode->i_sb, 1);
2942                 if (ret)
2943                         goto journal_error;
2944
2945                 reiserfs_update_inode_transaction(inode);
2946                 inode->i_size = pos + copied;
2947                 /*
2948                  * this will just nest into our transaction.  It's important
2949                  * to use mark_inode_dirty so the inode gets pushed around on
2950                  * the dirty lists, and so that O_SYNC works as expected
2951                  */
2952                 mark_inode_dirty(inode);
2953                 reiserfs_update_sd(&myth, inode);
2954                 update_sd = 1;
2955                 ret = journal_end(&myth);
2956                 if (ret)
2957                         goto journal_error;
2958         }
2959         if (th) {
2960                 if (!locked) {
2961                         reiserfs_write_lock(inode->i_sb);
2962                         locked = true;
2963                 }
2964                 if (!update_sd)
2965                         mark_inode_dirty(inode);
2966                 ret = reiserfs_end_persistent_transaction(th);
2967                 if (ret)
2968                         goto out;
2969         }
2970
2971 out:
2972         if (locked)
2973                 reiserfs_write_unlock(inode->i_sb);
2974         unlock_page(page);
2975         put_page(page);
2976
2977         if (pos + len > inode->i_size)
2978                 reiserfs_truncate_failed_write(inode);
2979
2980         return ret == 0 ? copied : ret;
2981
2982 journal_error:
2983         reiserfs_write_unlock(inode->i_sb);
2984         locked = false;
2985         if (th) {
2986                 if (!update_sd)
2987                         reiserfs_update_sd(th, inode);
2988                 ret = reiserfs_end_persistent_transaction(th);
2989         }
2990         goto out;
2991 }
2992
2993 int reiserfs_commit_write(struct file *f, struct page *page,
2994                           unsigned from, unsigned to)
2995 {
2996         struct inode *inode = page->mapping->host;
2997         loff_t pos = ((loff_t) page->index << PAGE_SHIFT) + to;
2998         int ret = 0;
2999         int update_sd = 0;
3000         struct reiserfs_transaction_handle *th = NULL;
3001         int depth;
3002
3003         depth = reiserfs_write_unlock_nested(inode->i_sb);
3004         reiserfs_wait_on_write_block(inode->i_sb);
3005         reiserfs_write_lock_nested(inode->i_sb, depth);
3006
3007         if (reiserfs_transaction_running(inode->i_sb)) {
3008                 th = current->journal_info;
3009         }
3010         reiserfs_commit_page(inode, page, from, to);
3011
3012         /*
3013          * generic_commit_write does this for us, but does not update the
3014          * transaction tracking stuff when the size changes.  So, we have
3015          * to do the i_size updates here.
3016          */
3017         if (pos > inode->i_size) {
3018                 struct reiserfs_transaction_handle myth;
3019                 /*
3020                  * If the file have grown beyond the border where it
3021                  * can have a tail, unmark it as needing a tail
3022                  * packing
3023                  */
3024                 if ((have_large_tails(inode->i_sb)
3025                      && inode->i_size > i_block_size(inode) * 4)
3026                     || (have_small_tails(inode->i_sb)
3027                         && inode->i_size > i_block_size(inode)))
3028                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3029
3030                 ret = journal_begin(&myth, inode->i_sb, 1);
3031                 if (ret)
3032                         goto journal_error;
3033
3034                 reiserfs_update_inode_transaction(inode);
3035                 inode->i_size = pos;
3036                 /*
3037                  * this will just nest into our transaction.  It's important
3038                  * to use mark_inode_dirty so the inode gets pushed around
3039                  * on the dirty lists, and so that O_SYNC works as expected
3040                  */
3041                 mark_inode_dirty(inode);
3042                 reiserfs_update_sd(&myth, inode);
3043                 update_sd = 1;
3044                 ret = journal_end(&myth);
3045                 if (ret)
3046                         goto journal_error;
3047         }
3048         if (th) {
3049                 if (!update_sd)
3050                         mark_inode_dirty(inode);
3051                 ret = reiserfs_end_persistent_transaction(th);
3052                 if (ret)
3053                         goto out;
3054         }
3055
3056 out:
3057         return ret;
3058
3059 journal_error:
3060         if (th) {
3061                 if (!update_sd)
3062                         reiserfs_update_sd(th, inode);
3063                 ret = reiserfs_end_persistent_transaction(th);
3064         }
3065
3066         return ret;
3067 }
3068
3069 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3070 {
3071         if (reiserfs_attrs(inode->i_sb)) {
3072                 if (sd_attrs & REISERFS_SYNC_FL)
3073                         inode->i_flags |= S_SYNC;
3074                 else
3075                         inode->i_flags &= ~S_SYNC;
3076                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
3077                         inode->i_flags |= S_IMMUTABLE;
3078                 else
3079                         inode->i_flags &= ~S_IMMUTABLE;
3080                 if (sd_attrs & REISERFS_APPEND_FL)
3081                         inode->i_flags |= S_APPEND;
3082                 else
3083                         inode->i_flags &= ~S_APPEND;
3084                 if (sd_attrs & REISERFS_NOATIME_FL)
3085                         inode->i_flags |= S_NOATIME;
3086                 else
3087                         inode->i_flags &= ~S_NOATIME;
3088                 if (sd_attrs & REISERFS_NOTAIL_FL)
3089                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
3090                 else
3091                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3092         }
3093 }
3094
3095 /*
3096  * decide if this buffer needs to stay around for data logging or ordered
3097  * write purposes
3098  */
3099 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
3100 {
3101         int ret = 1;
3102         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3103
3104         lock_buffer(bh);
3105         spin_lock(&j->j_dirty_buffers_lock);
3106         if (!buffer_mapped(bh)) {
3107                 goto free_jh;
3108         }
3109         /*
3110          * the page is locked, and the only places that log a data buffer
3111          * also lock the page.
3112          */
3113         if (reiserfs_file_data_log(inode)) {
3114                 /*
3115                  * very conservative, leave the buffer pinned if
3116                  * anyone might need it.
3117                  */
3118                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3119                         ret = 0;
3120                 }
3121         } else  if (buffer_dirty(bh)) {
3122                 struct reiserfs_journal_list *jl;
3123                 struct reiserfs_jh *jh = bh->b_private;
3124
3125                 /*
3126                  * why is this safe?
3127                  * reiserfs_setattr updates i_size in the on disk
3128                  * stat data before allowing vmtruncate to be called.
3129                  *
3130                  * If buffer was put onto the ordered list for this
3131                  * transaction, we know for sure either this transaction
3132                  * or an older one already has updated i_size on disk,
3133                  * and this ordered data won't be referenced in the file
3134                  * if we crash.
3135                  *
3136                  * if the buffer was put onto the ordered list for an older
3137                  * transaction, we need to leave it around
3138                  */
3139                 if (jh && (jl = jh->jl)
3140                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3141                         ret = 0;
3142         }
3143 free_jh:
3144         if (ret && bh->b_private) {
3145                 reiserfs_free_jh(bh);
3146         }
3147         spin_unlock(&j->j_dirty_buffers_lock);
3148         unlock_buffer(bh);
3149         return ret;
3150 }
3151
3152 /* clm -- taken from fs/buffer.c:block_invalidate_page */
3153 static void reiserfs_invalidatepage(struct page *page, unsigned int offset,
3154                                     unsigned int length)
3155 {
3156         struct buffer_head *head, *bh, *next;
3157         struct inode *inode = page->mapping->host;
3158         unsigned int curr_off = 0;
3159         unsigned int stop = offset + length;
3160         int partial_page = (offset || length < PAGE_SIZE);
3161         int ret = 1;
3162
3163         BUG_ON(!PageLocked(page));
3164
3165         if (!partial_page)
3166                 ClearPageChecked(page);
3167
3168         if (!page_has_buffers(page))
3169                 goto out;
3170
3171         head = page_buffers(page);
3172         bh = head;
3173         do {
3174                 unsigned int next_off = curr_off + bh->b_size;
3175                 next = bh->b_this_page;
3176
3177                 if (next_off > stop)
3178                         goto out;
3179
3180                 /*
3181                  * is this block fully invalidated?
3182                  */
3183                 if (offset <= curr_off) {
3184                         if (invalidatepage_can_drop(inode, bh))
3185                                 reiserfs_unmap_buffer(bh);
3186                         else
3187                                 ret = 0;
3188                 }
3189                 curr_off = next_off;
3190                 bh = next;
3191         } while (bh != head);
3192
3193         /*
3194          * We release buffers only if the entire page is being invalidated.
3195          * The get_block cached value has been unconditionally invalidated,
3196          * so real IO is not possible anymore.
3197          */
3198         if (!partial_page && ret) {
3199                 ret = try_to_release_page(page, 0);
3200                 /* maybe should BUG_ON(!ret); - neilb */
3201         }
3202 out:
3203         return;
3204 }
3205
3206 static int reiserfs_set_page_dirty(struct page *page)
3207 {
3208         struct inode *inode = page->mapping->host;
3209         if (reiserfs_file_data_log(inode)) {
3210                 SetPageChecked(page);
3211                 return __set_page_dirty_nobuffers(page);
3212         }
3213         return __set_page_dirty_buffers(page);
3214 }
3215
3216 /*
3217  * Returns 1 if the page's buffers were dropped.  The page is locked.
3218  *
3219  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3220  * in the buffers at page_buffers(page).
3221  *
3222  * even in -o notail mode, we can't be sure an old mount without -o notail
3223  * didn't create files with tails.
3224  */
3225 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3226 {
3227         struct inode *inode = page->mapping->host;
3228         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3229         struct buffer_head *head;
3230         struct buffer_head *bh;
3231         int ret = 1;
3232
3233         WARN_ON(PageChecked(page));
3234         spin_lock(&j->j_dirty_buffers_lock);
3235         head = page_buffers(page);
3236         bh = head;
3237         do {
3238                 if (bh->b_private) {
3239                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3240                                 reiserfs_free_jh(bh);
3241                         } else {
3242                                 ret = 0;
3243                                 break;
3244                         }
3245                 }
3246                 bh = bh->b_this_page;
3247         } while (bh != head);
3248         if (ret)
3249                 ret = try_to_free_buffers(page);
3250         spin_unlock(&j->j_dirty_buffers_lock);
3251         return ret;
3252 }
3253
3254 /*
3255  * We thank Mingming Cao for helping us understand in great detail what
3256  * to do in this section of the code.
3257  */
3258 static ssize_t reiserfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3259 {
3260         struct file *file = iocb->ki_filp;
3261         struct inode *inode = file->f_mapping->host;
3262         size_t count = iov_iter_count(iter);
3263         ssize_t ret;
3264
3265         ret = blockdev_direct_IO(iocb, inode, iter,
3266                                  reiserfs_get_blocks_direct_io);
3267
3268         /*
3269          * In case of error extending write may have instantiated a few
3270          * blocks outside i_size. Trim these off again.
3271          */
3272         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
3273                 loff_t isize = i_size_read(inode);
3274                 loff_t end = iocb->ki_pos + count;
3275
3276                 if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3277                         truncate_setsize(inode, isize);
3278                         reiserfs_vfs_truncate_file(inode);
3279                 }
3280         }
3281
3282         return ret;
3283 }
3284
3285 int reiserfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
3286                      struct iattr *attr)
3287 {
3288         struct inode *inode = d_inode(dentry);
3289         unsigned int ia_valid;
3290         int error;
3291
3292         error = setattr_prepare(&init_user_ns, dentry, attr);
3293         if (error)
3294                 return error;
3295
3296         /* must be turned off for recursive notify_change calls */
3297         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3298
3299         if (is_quota_modification(inode, attr)) {
3300                 error = dquot_initialize(inode);
3301                 if (error)
3302                         return error;
3303         }
3304         reiserfs_write_lock(inode->i_sb);
3305         if (attr->ia_valid & ATTR_SIZE) {
3306                 /*
3307                  * version 2 items will be caught by the s_maxbytes check
3308                  * done for us in vmtruncate
3309                  */
3310                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3311                     attr->ia_size > MAX_NON_LFS) {
3312                         reiserfs_write_unlock(inode->i_sb);
3313                         error = -EFBIG;
3314                         goto out;
3315                 }
3316
3317                 inode_dio_wait(inode);
3318
3319                 /* fill in hole pointers in the expanding truncate case. */
3320                 if (attr->ia_size > inode->i_size) {
3321                         error = generic_cont_expand_simple(inode, attr->ia_size);
3322                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3323                                 int err;
3324                                 struct reiserfs_transaction_handle th;
3325                                 /* we're changing at most 2 bitmaps, inode + super */
3326                                 err = journal_begin(&th, inode->i_sb, 4);
3327                                 if (!err) {
3328                                         reiserfs_discard_prealloc(&th, inode);
3329                                         err = journal_end(&th);
3330                                 }
3331                                 if (err)
3332                                         error = err;
3333                         }
3334                         if (error) {
3335                                 reiserfs_write_unlock(inode->i_sb);
3336                                 goto out;
3337                         }
3338                         /*
3339                          * file size is changed, ctime and mtime are
3340                          * to be updated
3341                          */
3342                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3343                 }
3344         }
3345         reiserfs_write_unlock(inode->i_sb);
3346
3347         if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3348              ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3349             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3350                 /* stat data of format v3.5 has 16 bit uid and gid */
3351                 error = -EINVAL;
3352                 goto out;
3353         }
3354
3355         if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3356             (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3357                 struct reiserfs_transaction_handle th;
3358                 int jbegin_count =
3359                     2 *
3360                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3361                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3362                     2;
3363
3364                 error = reiserfs_chown_xattrs(inode, attr);
3365
3366                 if (error)
3367                         return error;
3368
3369                 /*
3370                  * (user+group)*(old+new) structure - we count quota
3371                  * info and , inode write (sb, inode)
3372                  */
3373                 reiserfs_write_lock(inode->i_sb);
3374                 error = journal_begin(&th, inode->i_sb, jbegin_count);
3375                 reiserfs_write_unlock(inode->i_sb);
3376                 if (error)
3377                         goto out;
3378                 error = dquot_transfer(inode, attr);
3379                 reiserfs_write_lock(inode->i_sb);
3380                 if (error) {
3381                         journal_end(&th);
3382                         reiserfs_write_unlock(inode->i_sb);
3383                         goto out;
3384                 }
3385
3386                 /*
3387                  * Update corresponding info in inode so that everything
3388                  * is in one transaction
3389                  */
3390                 if (attr->ia_valid & ATTR_UID)
3391                         inode->i_uid = attr->ia_uid;
3392                 if (attr->ia_valid & ATTR_GID)
3393                         inode->i_gid = attr->ia_gid;
3394                 mark_inode_dirty(inode);
3395                 error = journal_end(&th);
3396                 reiserfs_write_unlock(inode->i_sb);
3397                 if (error)
3398                         goto out;
3399         }
3400
3401         if ((attr->ia_valid & ATTR_SIZE) &&
3402             attr->ia_size != i_size_read(inode)) {
3403                 error = inode_newsize_ok(inode, attr->ia_size);
3404                 if (!error) {
3405                         /*
3406                          * Could race against reiserfs_file_release
3407                          * if called from NFS, so take tailpack mutex.
3408                          */
3409                         mutex_lock(&REISERFS_I(inode)->tailpack);
3410                         truncate_setsize(inode, attr->ia_size);
3411                         reiserfs_truncate_file(inode, 1);
3412                         mutex_unlock(&REISERFS_I(inode)->tailpack);
3413                 }
3414         }
3415
3416         if (!error) {
3417                 setattr_copy(&init_user_ns, inode, attr);
3418                 mark_inode_dirty(inode);
3419         }
3420
3421         if (!error && reiserfs_posixacl(inode->i_sb)) {
3422                 if (attr->ia_valid & ATTR_MODE)
3423                         error = reiserfs_acl_chmod(inode);
3424         }
3425
3426 out:
3427         return error;
3428 }
3429
3430 const struct address_space_operations reiserfs_address_space_operations = {
3431         .writepage = reiserfs_writepage,
3432         .readpage = reiserfs_readpage,
3433         .readahead = reiserfs_readahead,
3434         .releasepage = reiserfs_releasepage,
3435         .invalidatepage = reiserfs_invalidatepage,
3436         .write_begin = reiserfs_write_begin,
3437         .write_end = reiserfs_write_end,
3438         .bmap = reiserfs_aop_bmap,
3439         .direct_IO = reiserfs_direct_IO,
3440         .set_page_dirty = reiserfs_set_page_dirty,
3441 };