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