fs/reiserfs/file.c

   1 /*
   2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
   3  */
   4
   5 #include <linux/time.h>
   6 #include <linux/reiserfs_fs.h>
   7 #include <linux/reiserfs_acl.h>
   8 #include <linux/reiserfs_xattr.h>
   9 #include <linux/smp_lock.h>
  10 #include <asm/uaccess.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/swap.h>
  13 #include <linux/writeback.h>
  14 #include <linux/blkdev.h>
  15 #include <linux/buffer_head.h>
  16 #include <linux/quotaops.h>
  17
  18 /*
  19 ** We pack the tails of files on file close, not at the time they are written.
  20 ** This implies an unnecessary copy of the tail and an unnecessary indirect item
  21 ** insertion/balancing, for files that are written in one write.
  22 ** It avoids unnecessary tail packings (balances) for files that are written in
  23 ** multiple writes and are small enough to have tails.
  24 **
  25 ** file_release is called by the VFS layer when the file is closed.  If
  26 ** this is the last open file descriptor, and the file
  27 ** small enough to have a tail, and the tail is currently in an
  28 ** unformatted node, the tail is converted back into a direct item.
  29 **
  30 ** We use reiserfs_truncate_file to pack the tail, since it already has
  31 ** all the conditions coded.
  32 */
  33 static int reiserfs_file_release(struct inode *inode, struct file *filp)
  34 {
  35
  36         struct reiserfs_transaction_handle th;
  37         int err;
  38         int jbegin_failure = 0;
  39
  40         BUG_ON(!S_ISREG(inode->i_mode));
  41
  42         /* fast out for when nothing needs to be done */
  43         if ((atomic_read(&inode->i_count) > 1 ||
  44              !(REISERFS_I(inode)->i_flags & i_pack_on_close_mask) ||
  45              !tail_has_to_be_packed(inode)) &&
  46             REISERFS_I(inode)->i_prealloc_count <= 0) {
  47                 return 0;
  48         }
  49
  50         mutex_lock(&inode->i_mutex);
  51         reiserfs_write_lock(inode->i_sb);
  52         /* freeing preallocation only involves relogging blocks that
  53          * are already in the current transaction.  preallocation gets
  54          * freed at the end of each transaction, so it is impossible for
  55          * us to log any additional blocks (including quota blocks)
  56          */
  57         err = journal_begin(&th, inode->i_sb, 1);
  58         if (err) {
  59                 /* uh oh, we can't allow the inode to go away while there
  60                  * is still preallocation blocks pending.  Try to join the
  61                  * aborted transaction
  62                  */
  63                 jbegin_failure = err;
  64                 err = journal_join_abort(&th, inode->i_sb, 1);
  65
  66                 if (err) {
  67                         /* hmpf, our choices here aren't good.  We can pin the inode
  68                          * which will disallow unmount from every happening, we can
  69                          * do nothing, which will corrupt random memory on unmount,
  70                          * or we can forcibly remove the file from the preallocation
  71                          * list, which will leak blocks on disk.  Lets pin the inode
  72                          * and let the admin know what is going on.
  73                          */
  74                         igrab(inode);
  75                         reiserfs_warning(inode->i_sb,
  76                                          "pinning inode %lu because the "
  77                                          "preallocation can't be freed");
  78                         goto out;
  79                 }
  80         }
  81         reiserfs_update_inode_transaction(inode);
  82
  83 #ifdef REISERFS_PREALLOCATE
  84         reiserfs_discard_prealloc(&th, inode);
  85 #endif
  86         err = journal_end(&th, inode->i_sb, 1);
  87
  88         /* copy back the error code from journal_begin */
  89         if (!err)
  90                 err = jbegin_failure;
  91
  92         if (!err && atomic_read(&inode->i_count) <= 1 &&
  93             (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) &&
  94             tail_has_to_be_packed(inode)) {
  95                 /* if regular file is released by last holder and it has been
  96                    appended (we append by unformatted node only) or its direct
  97                    item(s) had to be converted, then it may have to be
  98                    indirect2direct converted */
  99                 err = reiserfs_truncate_file(inode, 0);
 100         }
 101       out:
 102         mutex_unlock(&inode->i_mutex);
 103         reiserfs_write_unlock(inode->i_sb);
 104         return err;
 105 }
 106
 107 static void reiserfs_vfs_truncate_file(struct inode *inode)
 108 {
 109         reiserfs_truncate_file(inode, 1);
 110 }
 111
 112 /* Sync a reiserfs file. */
 113
 114 /*
 115  * FIXME: sync_mapping_buffers() never has anything to sync.  Can
 116  * be removed...
 117  */
 118
 119 static int reiserfs_sync_file(struct file *p_s_filp,
 120                               struct dentry *p_s_dentry, int datasync)
 121 {
 122         struct inode *p_s_inode = p_s_dentry->d_inode;
 123         int n_err;
 124         int barrier_done;
 125
 126         BUG_ON(!S_ISREG(p_s_inode->i_mode));
 127         n_err = sync_mapping_buffers(p_s_inode->i_mapping);
 128         reiserfs_write_lock(p_s_inode->i_sb);
 129         barrier_done = reiserfs_commit_for_inode(p_s_inode);
 130         reiserfs_write_unlock(p_s_inode->i_sb);
 131         if (barrier_done != 1 && reiserfs_barrier_flush(p_s_inode->i_sb))
 132                 blkdev_issue_flush(p_s_inode->i_sb->s_bdev, NULL);
 133         if (barrier_done < 0)
 134                 return barrier_done;
 135         return (n_err < 0) ? -EIO : 0;
 136 }
 137
 138 /* I really do not want to play with memory shortage right now, so
 139    to simplify the code, we are not going to write more than this much pages at
 140    a time. This still should considerably improve performance compared to 4k
 141    at a time case. This is 32 pages of 4k size. */
 142 #define REISERFS_WRITE_PAGES_AT_A_TIME (128 * 1024) / PAGE_CACHE_SIZE
 143
 144 /* Allocates blocks for a file to fulfil write request.
 145    Maps all unmapped but prepared pages from the list.
 146    Updates metadata with newly allocated blocknumbers as needed */
 147 static int reiserfs_allocate_blocks_for_region(struct reiserfs_transaction_handle *th, struct inode *inode,     /* Inode we work with */
 148                                                loff_t pos,      /* Writing position */
 149                                                int num_pages,   /* number of pages write going
 150                                                                    to touch */
 151                                                int write_bytes, /* amount of bytes to write */
 152                                                struct page **prepared_pages,    /* array of
 153                                                                                    prepared pages
 154                                                                                  */
 155                                                int blocks_to_allocate   /* Amount of blocks we
 156                                                                            need to allocate to
 157                                                                            fit the data into file
 158                                                                          */
 159     )
 160 {
 161         struct cpu_key key;     // cpu key of item that we are going to deal with
 162         struct item_head *ih;   // pointer to item head that we are going to deal with
 163         struct buffer_head *bh; // Buffer head that contains items that we are going to deal with
 164         __le32 *item;           // pointer to item we are going to deal with
 165         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
 166         b_blocknr_t *allocated_blocks;  // Pointer to a place where allocated blocknumbers would be stored.
 167         reiserfs_blocknr_hint_t hint;   // hint structure for block allocator.
 168         size_t res;             // return value of various functions that we call.
 169         int curr_block;         // current block used to keep track of unmapped blocks.
 170         int i;                  // loop counter
 171         int itempos;            // position in item
 172         unsigned int from = (pos & (PAGE_CACHE_SIZE - 1));      // writing position in
 173         // first page
 174         unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;        /* last modified byte offset in last page */
 175         __u64 hole_size;        // amount of blocks for a file hole, if it needed to be created.
 176         int modifying_this_item = 0;    // Flag for items traversal code to keep track
 177         // of the fact that we already prepared
 178         // current block for journal
 179         int will_prealloc = 0;
 180         RFALSE(!blocks_to_allocate,
 181                "green-9004: tried to allocate zero blocks?");
 182
 183         /* only preallocate if this is a small write */
 184         if (REISERFS_I(inode)->i_prealloc_count ||
 185             (!(write_bytes & (inode->i_sb->s_blocksize - 1)) &&
 186              blocks_to_allocate <
 187              REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize))
 188                 will_prealloc =
 189                     REISERFS_SB(inode->i_sb)->s_alloc_options.preallocsize;
 190
 191         allocated_blocks = kmalloc((blocks_to_allocate + will_prealloc) *
 192                                    sizeof(b_blocknr_t), GFP_NOFS);
 193         if (!allocated_blocks)
 194                 return -ENOMEM;
 195
 196         /* First we compose a key to point at the writing position, we want to do
 197            that outside of any locking region. */
 198         make_cpu_key(&key, inode, pos + 1, TYPE_ANY, 3 /*key length */ );
 199
 200         /* If we came here, it means we absolutely need to open a transaction,
 201            since we need to allocate some blocks */
 202         reiserfs_write_lock(inode->i_sb);       // Journaling stuff and we need that.
 203         res = journal_begin(th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1 + 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb));   // Wish I know if this number enough
 204         if (res)
 205                 goto error_exit;
 206         reiserfs_update_inode_transaction(inode);
 207
 208         /* Look for the in-tree position of our write, need path for block allocator */
 209         res = search_for_position_by_key(inode->i_sb, &key, &path);
 210         if (res == IO_ERROR) {
 211                 res = -EIO;
 212                 goto error_exit;
 213         }
 214
 215         /* Allocate blocks */
 216         /* First fill in "hint" structure for block allocator */
 217         hint.th = th;           // transaction handle.
 218         hint.path = &path;      // Path, so that block allocator can determine packing locality or whatever it needs to determine.
 219         hint.inode = inode;     // Inode is needed by block allocator too.
 220         hint.search_start = 0;  // We have no hint on where to search free blocks for block allocator.
 221         hint.key = key.on_disk_key;     // on disk key of file.
 222         hint.block = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);    // Number of disk blocks this file occupies already.
 223         hint.formatted_node = 0;        // We are allocating blocks for unformatted node.
 224         hint.preallocate = will_prealloc;
 225
 226         /* Call block allocator to allocate blocks */
 227         res =
 228             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
 229                                        blocks_to_allocate, blocks_to_allocate);
 230         if (res != CARRY_ON) {
 231                 if (res == NO_DISK_SPACE) {
 232                         /* We flush the transaction in case of no space. This way some
 233                            blocks might become free */
 234                         SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
 235                         res = restart_transaction(th, inode, &path);
 236                         if (res)
 237                                 goto error_exit;
 238
 239                         /* We might have scheduled, so search again */
 240                         res =
 241                             search_for_position_by_key(inode->i_sb, &key,
 242                                                        &path);
 243                         if (res == IO_ERROR) {
 244                                 res = -EIO;
 245                                 goto error_exit;
 246                         }
 247
 248                         /* update changed info for hint structure. */
 249                         res =
 250                             reiserfs_allocate_blocknrs(&hint, allocated_blocks,
 251                                                        blocks_to_allocate,
 252                                                        blocks_to_allocate);
 253                         if (res != CARRY_ON) {
 254                                 res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
 255                                 pathrelse(&path);
 256                                 goto error_exit;
 257                         }
 258                 } else {
 259                         res = res == QUOTA_EXCEEDED ? -EDQUOT : -ENOSPC;
 260                         pathrelse(&path);
 261                         goto error_exit;
 262                 }
 263         }
 264 #ifdef __BIG_ENDIAN
 265         // Too bad, I have not found any way to convert a given region from
 266         // cpu format to little endian format
 267         {
 268                 int i;
 269                 for (i = 0; i < blocks_to_allocate; i++)
 270                         allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);
 271         }
 272 #endif
 273
 274         /* Blocks allocating well might have scheduled and tree might have changed,
 275            let's search the tree again */
 276         /* find where in the tree our write should go */
 277         res = search_for_position_by_key(inode->i_sb, &key, &path);
 278         if (res == IO_ERROR) {
 279                 res = -EIO;
 280                 goto error_exit_free_blocks;
 281         }
 282
 283         bh = get_last_bh(&path);        // Get a bufferhead for last element in path.
 284         ih = get_ih(&path);     // Get a pointer to last item head in path.
 285         item = get_item(&path); // Get a pointer to last item in path
 286
 287         /* Let's see what we have found */
 288         if (res != POSITION_FOUND) {    /* position not found, this means that we
 289                                            might need to append file with holes
 290                                            first */
 291                 // Since we are writing past the file's end, we need to find out if
 292                 // there is a hole that needs to be inserted before our writing
 293                 // position, and how many blocks it is going to cover (we need to
 294                 //  populate pointers to file blocks representing the hole with zeros)
 295
 296                 {
 297                         int item_offset = 1;
 298                         /*
 299                          * if ih is stat data, its offset is 0 and we don't want to
 300                          * add 1 to pos in the hole_size calculation
 301                          */
 302                         if (is_statdata_le_ih(ih))
 303                                 item_offset = 0;
 304                         hole_size = (pos + item_offset -
 305                                      (le_key_k_offset
 306                                       (get_inode_item_key_version(inode),
 307                                        &(ih->ih_key)) + op_bytes_number(ih,
 308                                                                         inode->
 309                                                                         i_sb->
 310                                                                         s_blocksize)))
 311                             >> inode->i_sb->s_blocksize_bits;
 312                 }
 313
 314                 if (hole_size > 0) {
 315                         int to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize) / UNFM_P_SIZE);   // How much data to insert first time.
 316                         /* area filled with zeroes, to supply as list of zero blocknumbers
 317                            We allocate it outside of loop just in case loop would spin for
 318                            several iterations. */
 319                         char *zeros = kmalloc(to_paste * UNFM_P_SIZE, GFP_ATOMIC);      // We cannot insert more than MAX_ITEM_LEN bytes anyway.
 320                         if (!zeros) {
 321                                 res = -ENOMEM;
 322                                 goto error_exit_free_blocks;
 323                         }
 324                         memset(zeros, 0, to_paste * UNFM_P_SIZE);
 325                         do {
 326                                 to_paste =
 327                                     min_t(__u64, hole_size,
 328                                           MAX_ITEM_LEN(inode->i_sb->
 329                                                        s_blocksize) /
 330                                           UNFM_P_SIZE);
 331                                 if (is_indirect_le_ih(ih)) {
 332                                         /* Ok, there is existing indirect item already. Need to append it */
 333                                         /* Calculate position past inserted item */
 334                                         make_cpu_key(&key, inode,
 335                                                      le_key_k_offset
 336                                                      (get_inode_item_key_version
 337                                                       (inode),
 338                                                       &(ih->ih_key)) +
 339                                                      op_bytes_number(ih,
 340                                                                      inode->
 341                                                                      i_sb->
 342                                                                      s_blocksize),
 343                                                      TYPE_INDIRECT, 3);
 344                                         res =
 345                                             reiserfs_paste_into_item(th, &path,
 346                                                                      &key,
 347                                                                      inode,
 348                                                                      (char *)
 349                                                                      zeros,
 350                                                                      UNFM_P_SIZE
 351                                                                      *
 352                                                                      to_paste);
 353                                         if (res) {
 354                                                 kfree(zeros);
 355                                                 goto error_exit_free_blocks;
 356                                         }
 357                                 } else if (is_statdata_le_ih(ih)) {
 358                                         /* No existing item, create it */
 359                                         /* item head for new item */
 360                                         struct item_head ins_ih;
 361
 362                                         /* create a key for our new item */
 363                                         make_cpu_key(&key, inode, 1,
 364                                                      TYPE_INDIRECT, 3);
 365
 366                                         /* Create new item head for our new item */
 367                                         make_le_item_head(&ins_ih, &key,
 368                                                           key.version, 1,
 369                                                           TYPE_INDIRECT,
 370                                                           to_paste *
 371                                                           UNFM_P_SIZE,
 372                                                           0 /* free space */ );
 373
 374                                         /* Find where such item should live in the tree */
 375                                         res =
 376                                             search_item(inode->i_sb, &key,
 377                                                         &path);
 378                                         if (res != ITEM_NOT_FOUND) {
 379                                                 /* item should not exist, otherwise we have error */
 380                                                 if (res != -ENOSPC) {
 381                                                         reiserfs_warning(inode->
 382                                                                          i_sb,
 383                                                                          "green-9008: search_by_key (%K) returned %d",
 384                                                                          &key,
 385                                                                          res);
 386                                                 }
 387                                                 res = -EIO;
 388                                                 kfree(zeros);
 389                                                 goto error_exit_free_blocks;
 390                                         }
 391                                         res =
 392                                             reiserfs_insert_item(th, &path,
 393                                                                  &key, &ins_ih,
 394                                                                  inode,
 395                                                                  (char *)zeros);
 396                                 } else {
 397                                         reiserfs_panic(inode->i_sb,
 398                                                        "green-9011: Unexpected key type %K\n",
 399                                                        &key);
 400                                 }
 401                                 if (res) {
 402                                         kfree(zeros);
 403                                         goto error_exit_free_blocks;
 404                                 }
 405                                 /* Now we want to check if transaction is too full, and if it is
 406                                    we restart it. This will also free the path. */
 407                                 if (journal_transaction_should_end
 408                                     (th, th->t_blocks_allocated)) {
 409                                         res =
 410                                             restart_transaction(th, inode,
 411                                                                 &path);
 412                                         if (res) {
 413                                                 pathrelse(&path);
 414                                                 kfree(zeros);
 415                                                 goto error_exit;
 416                                         }
 417                                 }
 418
 419                                 /* Well, need to recalculate path and stuff */
 420                                 set_cpu_key_k_offset(&key,
 421                                                      cpu_key_k_offset(&key) +
 422                                                      (to_paste << inode->
 423                                                       i_blkbits));
 424                                 res =
 425                                     search_for_position_by_key(inode->i_sb,
 426                                                                &key, &path);
 427                                 if (res == IO_ERROR) {
 428                                         res = -EIO;
 429                                         kfree(zeros);
 430                                         goto error_exit_free_blocks;
 431                                 }
 432                                 bh = get_last_bh(&path);
 433                                 ih = get_ih(&path);
 434                                 item = get_item(&path);
 435                                 hole_size -= to_paste;
 436                         } while (hole_size);
 437                         kfree(zeros);
 438                 }
 439         }
 440         // Go through existing indirect items first
 441         // replace all zeroes with blocknumbers from list
 442         // Note that if no corresponding item was found, by previous search,
 443         // it means there are no existing in-tree representation for file area
 444         // we are going to overwrite, so there is nothing to scan through for holes.
 445         for (curr_block = 0, itempos = path.pos_in_item;
 446              curr_block < blocks_to_allocate && res == POSITION_FOUND;) {
 447               retry:
 448
 449                 if (itempos >= ih_item_len(ih) / UNFM_P_SIZE) {
 450                         /* We run out of data in this indirect item, let's look for another
 451                            one. */
 452                         /* First if we are already modifying current item, log it */
 453                         if (modifying_this_item) {
 454                                 journal_mark_dirty(th, inode->i_sb, bh);
 455                                 modifying_this_item = 0;
 456                         }
 457                         /* Then set the key to look for a new indirect item (offset of old
 458                            item is added to old item length */
 459                         set_cpu_key_k_offset(&key,
 460                                              le_key_k_offset
 461                                              (get_inode_item_key_version(inode),
 462                                               &(ih->ih_key)) +
 463                                              op_bytes_number(ih,
 464                                                              inode->i_sb->
 465                                                              s_blocksize));
 466                         /* Search ofor position of new key in the tree. */
 467                         res =
 468                             search_for_position_by_key(inode->i_sb, &key,
 469                                                        &path);
 470                         if (res == IO_ERROR) {
 471                                 res = -EIO;
 472                                 goto error_exit_free_blocks;
 473                         }
 474                         bh = get_last_bh(&path);
 475                         ih = get_ih(&path);
 476                         item = get_item(&path);
 477                         itempos = path.pos_in_item;
 478                         continue;       // loop to check all kinds of conditions and so on.
 479                 }
 480                 /* Ok, we have correct position in item now, so let's see if it is
 481                    representing file hole (blocknumber is zero) and fill it if needed */
 482                 if (!item[itempos]) {
 483                         /* Ok, a hole. Now we need to check if we already prepared this
 484                            block to be journaled */
 485                         while (!modifying_this_item) {  // loop until succeed
 486                                 /* Well, this item is not journaled yet, so we must prepare
 487                                    it for journal first, before we can change it */
 488                                 struct item_head tmp_ih;        // We copy item head of found item,
 489                                 // here to detect if fs changed under
 490                                 // us while we were preparing for
 491                                 // journal.
 492                                 int fs_gen;     // We store fs generation here to find if someone
 493                                 // changes fs under our feet
 494
 495                                 copy_item_head(&tmp_ih, ih);    // Remember itemhead
 496                                 fs_gen = get_generation(inode->i_sb);   // remember fs generation
 497                                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);       // Prepare a buffer within which indirect item is stored for changing.
 498                                 if (fs_changed(fs_gen, inode->i_sb)
 499                                     && item_moved(&tmp_ih, &path)) {
 500                                         // Sigh, fs was changed under us, we need to look for new
 501                                         // location of item we are working with
 502
 503                                         /* unmark prepaerd area as journaled and search for it's
 504                                            new position */
 505                                         reiserfs_restore_prepared_buffer(inode->
 506                                                                          i_sb,
 507                                                                          bh);
 508                                         res =
 509                                             search_for_position_by_key(inode->
 510                                                                        i_sb,
 511                                                                        &key,
 512                                                                        &path);
 513                                         if (res == IO_ERROR) {
 514                                                 res = -EIO;
 515                                                 goto error_exit_free_blocks;
 516                                         }
 517                                         bh = get_last_bh(&path);
 518                                         ih = get_ih(&path);
 519                                         item = get_item(&path);
 520                                         itempos = path.pos_in_item;
 521                                         goto retry;
 522                                 }
 523                                 modifying_this_item = 1;
 524                         }
 525                         item[itempos] = allocated_blocks[curr_block];   // Assign new block
 526                         curr_block++;
 527                 }
 528                 itempos++;
 529         }
 530
 531         if (modifying_this_item) {      // We need to log last-accessed block, if it
 532                 // was modified, but not logged yet.
 533                 journal_mark_dirty(th, inode->i_sb, bh);
 534         }
 535
 536         if (curr_block < blocks_to_allocate) {
 537                 // Oh, well need to append to indirect item, or to create indirect item
 538                 // if there weren't any
 539                 if (is_indirect_le_ih(ih)) {
 540                         // Existing indirect item - append. First calculate key for append
 541                         // position. We do not need to recalculate path as it should
 542                         // already point to correct place.
 543                         make_cpu_key(&key, inode,
 544                                      le_key_k_offset(get_inode_item_key_version
 545                                                      (inode),
 546                                                      &(ih->ih_key)) +
 547                                      op_bytes_number(ih,
 548                                                      inode->i_sb->s_blocksize),
 549                                      TYPE_INDIRECT, 3);
 550                         res =
 551                             reiserfs_paste_into_item(th, &path, &key, inode,
 552                                                      (char *)(allocated_blocks +
 553                                                               curr_block),
 554                                                      UNFM_P_SIZE *
 555                                                      (blocks_to_allocate -
 556                                                       curr_block));
 557                         if (res) {
 558                                 goto error_exit_free_blocks;
 559                         }
 560                 } else if (is_statdata_le_ih(ih)) {
 561                         // Last found item was statdata. That means we need to create indirect item.
 562                         struct item_head ins_ih;        /* itemhead for new item */
 563
 564                         /* create a key for our new item */
 565                         make_cpu_key(&key, inode, 1, TYPE_INDIRECT, 3); // Position one,
 566                         // because that's
 567                         // where first
 568                         // indirect item
 569                         // begins
 570                         /* Create new item head for our new item */
 571                         make_le_item_head(&ins_ih, &key, key.version, 1,
 572                                           TYPE_INDIRECT,
 573                                           (blocks_to_allocate -
 574                                            curr_block) * UNFM_P_SIZE,
 575                                           0 /* free space */ );
 576                         /* Find where such item should live in the tree */
 577                         res = search_item(inode->i_sb, &key, &path);
 578                         if (res != ITEM_NOT_FOUND) {
 579                                 /* Well, if we have found such item already, or some error
 580                                    occured, we need to warn user and return error */
 581                                 if (res != -ENOSPC) {
 582                                         reiserfs_warning(inode->i_sb,
 583                                                          "green-9009: search_by_key (%K) "
 584                                                          "returned %d", &key,
 585                                                          res);
 586                                 }
 587                                 res = -EIO;
 588                                 goto error_exit_free_blocks;
 589                         }
 590                         /* Insert item into the tree with the data as its body */
 591                         res =
 592                             reiserfs_insert_item(th, &path, &key, &ins_ih,
 593                                                  inode,
 594                                                  (char *)(allocated_blocks +
 595                                                           curr_block));
 596                 } else {
 597                         reiserfs_panic(inode->i_sb,
 598                                        "green-9010: unexpected item type for key %K\n",
 599                                        &key);
 600                 }
 601         }
 602         // the caller is responsible for closing the transaction
 603         // unless we return an error, they are also responsible for logging
 604         // the inode.
 605         //
 606         pathrelse(&path);
 607         /*
 608          * cleanup prellocation from previous writes
 609          * if this is a partial block write
 610          */
 611         if (write_bytes & (inode->i_sb->s_blocksize - 1))
 612                 reiserfs_discard_prealloc(th, inode);
 613         reiserfs_write_unlock(inode->i_sb);
 614
 615         // go through all the pages/buffers and map the buffers to newly allocated
 616         // blocks (so that system knows where to write these pages later).
 617         curr_block = 0;
 618         for (i = 0; i < num_pages; i++) {
 619                 struct page *page = prepared_pages[i];  //current page
 620                 struct buffer_head *head = page_buffers(page);  // first buffer for a page
 621                 int block_start, block_end;     // in-page offsets for buffers.
 622
 623                 if (!page_buffers(page))
 624                         reiserfs_panic(inode->i_sb,
 625                                        "green-9005: No buffers for prepared page???");
 626
 627                 /* For each buffer in page */
 628                 for (bh = head, block_start = 0; bh != head || !block_start;
 629                      block_start = block_end, bh = bh->b_this_page) {
 630                         if (!bh)
 631                                 reiserfs_panic(inode->i_sb,
 632                                                "green-9006: Allocated but absent buffer for a page?");
 633                         block_end = block_start + inode->i_sb->s_blocksize;
 634                         if (i == 0 && block_end <= from)
 635                                 /* if this buffer is before requested data to map, skip it */
 636                                 continue;
 637                         if (i == num_pages - 1 && block_start >= to)
 638                                 /* If this buffer is after requested data to map, abort
 639                                    processing of current page */
 640                                 break;
 641
 642                         if (!buffer_mapped(bh)) {       // Ok, unmapped buffer, need to map it
 643                                 map_bh(bh, inode->i_sb,
 644                                        le32_to_cpu(allocated_blocks
 645                                                    [curr_block]));
 646                                 curr_block++;
 647                                 set_buffer_new(bh);
 648                         }
 649                 }
 650         }
 651
 652         RFALSE(curr_block > blocks_to_allocate,
 653                "green-9007: Used too many blocks? weird");
 654
 655         kfree(allocated_blocks);
 656         return 0;
 657
 658 // Need to deal with transaction here.
 659       error_exit_free_blocks:
 660         pathrelse(&path);
 661         // free blocks
 662         for (i = 0; i < blocks_to_allocate; i++)
 663                 reiserfs_free_block(th, inode, le32_to_cpu(allocated_blocks[i]),
 664                                     1);
 665
 666       error_exit:
 667         if (th->t_trans_id) {
 668                 int err;
 669                 // update any changes we made to blk count
 670                 mark_inode_dirty(inode);
 671                 err =
 672                     journal_end(th, inode->i_sb,
 673                                 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
 674                                 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb));
 675                 if (err)
 676                         res = err;
 677         }
 678         reiserfs_write_unlock(inode->i_sb);
 679         kfree(allocated_blocks);
 680
 681         return res;
 682 }
 683
 684 /* Unlock pages prepared by reiserfs_prepare_file_region_for_write */
 685 static void reiserfs_unprepare_pages(struct page **prepared_pages,      /* list of locked pages */
 686                                      size_t num_pages /* amount of pages */ )
 687 {
 688         int i;                  // loop counter
 689
 690         for (i = 0; i < num_pages; i++) {
 691                 struct page *page = prepared_pages[i];
 692
 693                 try_to_free_buffers(page);
 694                 unlock_page(page);
 695                 page_cache_release(page);
 696         }
 697 }
 698
 699 /* This function will copy data from userspace to specified pages within
 700    supplied byte range */
 701 static int reiserfs_copy_from_user_to_file_region(loff_t pos,   /* In-file position */
 702                                                   int num_pages,        /* Number of pages affected */
 703                                                   int write_bytes,      /* Amount of bytes to write */
 704                                                   struct page **prepared_pages, /* pointer to
 705                                                                                    array to
 706                                                                                    prepared pages
 707                                                                                  */
 708                                                   const char __user * buf       /* Pointer to user-supplied
 709                                                                                    data */
 710     )
 711 {
 712         long page_fault = 0;    // status of copy_from_user.
 713         int i;                  // loop counter.
 714         int offset;             // offset in page
 715
 716         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
 717              i++, offset = 0) {
 718                 size_t count = min_t(size_t, PAGE_CACHE_SIZE - offset, write_bytes);    // How much of bytes to write to this page
 719                 struct page *page = prepared_pages[i];  // Current page we process.
 720
 721                 fault_in_pages_readable(buf, count);
 722
 723                 /* Copy data from userspace to the current page */
 724                 kmap(page);
 725                 page_fault = __copy_from_user(page_address(page) + offset, buf, count); // Copy the data.
 726                 /* Flush processor's dcache for this page */
 727                 flush_dcache_page(page);
 728                 kunmap(page);
 729                 buf += count;
 730                 write_bytes -= count;
 731
 732                 if (page_fault)
 733                         break;  // Was there a fault? abort.
 734         }
 735
 736         return page_fault ? -EFAULT : 0;
 737 }
 738
 739 /* taken fs/buffer.c:__block_commit_write */
 740 int reiserfs_commit_page(struct inode *inode, struct page *page,
 741                          unsigned from, unsigned to)
 742 {
 743         unsigned block_start, block_end;
 744         int partial = 0;
 745         unsigned blocksize;
 746         struct buffer_head *bh, *head;
 747         unsigned long i_size_index = inode->i_size >> PAGE_CACHE_SHIFT;
 748         int new;
 749         int logit = reiserfs_file_data_log(inode);
 750         struct super_block *s = inode->i_sb;
 751         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
 752         struct reiserfs_transaction_handle th;
 753         int ret = 0;
 754
 755         th.t_trans_id = 0;
 756         blocksize = 1 << inode->i_blkbits;
 757
 758         if (logit) {
 759                 reiserfs_write_lock(s);
 760                 ret = journal_begin(&th, s, bh_per_page + 1);
 761                 if (ret)
 762                         goto drop_write_lock;
 763                 reiserfs_update_inode_transaction(inode);
 764         }
 765         for (bh = head = page_buffers(page), block_start = 0;
 766              bh != head || !block_start;
 767              block_start = block_end, bh = bh->b_this_page) {
 768
 769                 new = buffer_new(bh);
 770                 clear_buffer_new(bh);
 771                 block_end = block_start + blocksize;
 772                 if (block_end <= from || block_start >= to) {
 773                         if (!buffer_uptodate(bh))
 774                                 partial = 1;
 775                 } else {
 776                         set_buffer_uptodate(bh);
 777                         if (logit) {
 778                                 reiserfs_prepare_for_journal(s, bh, 1);
 779                                 journal_mark_dirty(&th, s, bh);
 780                         } else if (!buffer_dirty(bh)) {
 781                                 mark_buffer_dirty(bh);
 782                                 /* do data=ordered on any page past the end
 783                                  * of file and any buffer marked BH_New.
 784                                  */
 785                                 if (reiserfs_data_ordered(inode->i_sb) &&
 786                                     (new || page->index >= i_size_index)) {
 787                                         reiserfs_add_ordered_list(inode, bh);
 788                                 }
 789                         }
 790                 }
 791         }
 792         if (logit) {
 793                 ret = journal_end(&th, s, bh_per_page + 1);
 794               drop_write_lock:
 795                 reiserfs_write_unlock(s);
 796         }
 797         /*
 798          * If this is a partial write which happened to make all buffers
 799          * uptodate then we can optimize away a bogus readpage() for
 800          * the next read(). Here we 'discover' whether the page went
 801          * uptodate as a result of this (potentially partial) write.
 802          */
 803         if (!partial)
 804                 SetPageUptodate(page);
 805         return ret;
 806 }
 807
 808 /* Submit pages for write. This was separated from actual file copying
 809    because we might want to allocate block numbers in-between.
 810    This function assumes that caller will adjust file size to correct value. */
 811 static int reiserfs_submit_file_region_for_write(struct reiserfs_transaction_handle *th, struct inode *inode, loff_t pos,       /* Writing position offset */
 812                                                  size_t num_pages,      /* Number of pages to write */
 813                                                  size_t write_bytes,    /* number of bytes to write */
 814                                                  struct page **prepared_pages   /* list of pages */
 815     )
 816 {
 817         int status;             // return status of block_commit_write.
 818         int retval = 0;         // Return value we are going to return.
 819         int i;                  // loop counter
 820         int offset;             // Writing offset in page.
 821         int orig_write_bytes = write_bytes;
 822         int sd_update = 0;
 823
 824         for (i = 0, offset = (pos & (PAGE_CACHE_SIZE - 1)); i < num_pages;
 825              i++, offset = 0) {
 826                 int count = min_t(int, PAGE_CACHE_SIZE - offset, write_bytes);  // How much of bytes to write to this page
 827                 struct page *page = prepared_pages[i];  // Current page we process.
 828
 829                 status =
 830                     reiserfs_commit_page(inode, page, offset, offset + count);
 831                 if (status)
 832                         retval = status;        // To not overcomplicate matters We are going to
 833                 // submit all the pages even if there was error.
 834                 // we only remember error status to report it on
 835                 // exit.
 836                 write_bytes -= count;
 837         }
 838         /* now that we've gotten all the ordered buffers marked dirty,
 839          * we can safely update i_size and close any running transaction
 840          */
 841         if (pos + orig_write_bytes > inode->i_size) {
 842                 inode->i_size = pos + orig_write_bytes; // Set new size
 843                 /* If the file have grown so much that tail packing is no
 844                  * longer possible, reset "need to pack" flag */
 845                 if ((have_large_tails(inode->i_sb) &&
 846                      inode->i_size > i_block_size(inode) * 4) ||
 847                     (have_small_tails(inode->i_sb) &&
 848                      inode->i_size > i_block_size(inode)))
 849                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
 850                 else if ((have_large_tails(inode->i_sb) &&
 851                           inode->i_size < i_block_size(inode) * 4) ||
 852                          (have_small_tails(inode->i_sb) &&
 853                           inode->i_size < i_block_size(inode)))
 854                         REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
 855
 856                 if (th->t_trans_id) {
 857                         reiserfs_write_lock(inode->i_sb);
 858                         // this sets the proper flags for O_SYNC to trigger a commit
 859                         mark_inode_dirty(inode);
 860                         reiserfs_write_unlock(inode->i_sb);
 861                 } else {
 862                         reiserfs_write_lock(inode->i_sb);
 863                         reiserfs_update_inode_transaction(inode);
 864                         mark_inode_dirty(inode);
 865                         reiserfs_write_unlock(inode->i_sb);
 866                 }
 867
 868                 sd_update = 1;
 869         }
 870         if (th->t_trans_id) {
 871                 reiserfs_write_lock(inode->i_sb);
 872                 if (!sd_update)
 873                         mark_inode_dirty(inode);
 874                 status = journal_end(th, th->t_super, th->t_blocks_allocated);
 875                 if (status)
 876                         retval = status;
 877                 reiserfs_write_unlock(inode->i_sb);
 878         }
 879         th->t_trans_id = 0;
 880
 881         /*
 882          * we have to unlock the pages after updating i_size, otherwise
 883          * we race with writepage
 884          */
 885         for (i = 0; i < num_pages; i++) {
 886                 struct page *page = prepared_pages[i];
 887                 unlock_page(page);
 888                 mark_page_accessed(page);
 889                 page_cache_release(page);
 890         }
 891         return retval;
 892 }
 893
 894 /* Look if passed writing region is going to touch file's tail
 895    (if it is present). And if it is, convert the tail to unformatted node */
 896 static int reiserfs_check_for_tail_and_convert(struct inode *inode,     /* inode to deal with */
 897                                                loff_t pos,      /* Writing position */
 898                                                int write_bytes  /* amount of bytes to write */
 899     )
 900 {
 901         INITIALIZE_PATH(path);  // needed for search_for_position
 902         struct cpu_key key;     // Key that would represent last touched writing byte.
 903         struct item_head *ih;   // item header of found block;
 904         int res;                // Return value of various functions we call.
 905         int cont_expand_offset; // We will put offset for generic_cont_expand here
 906         // This can be int just because tails are created
 907         // only for small files.
 908
 909 /* this embodies a dependency on a particular tail policy */
 910         if (inode->i_size >= inode->i_sb->s_blocksize * 4) {
 911                 /* such a big files do not have tails, so we won't bother ourselves
 912                    to look for tails, simply return */
 913                 return 0;
 914         }
 915
 916         reiserfs_write_lock(inode->i_sb);
 917         /* find the item containing the last byte to be written, or if
 918          * writing past the end of the file then the last item of the
 919          * file (and then we check its type). */
 920         make_cpu_key(&key, inode, pos + write_bytes + 1, TYPE_ANY,
 921                      3 /*key length */ );
 922         res = search_for_position_by_key(inode->i_sb, &key, &path);
 923         if (res == IO_ERROR) {
 924                 reiserfs_write_unlock(inode->i_sb);
 925                 return -EIO;
 926         }
 927         ih = get_ih(&path);
 928         res = 0;
 929         if (is_direct_le_ih(ih)) {
 930                 /* Ok, closest item is file tail (tails are stored in "direct"
 931                  * items), so we need to unpack it. */
 932                 /* To not overcomplicate matters, we just call generic_cont_expand
 933                    which will in turn call other stuff and finally will boil down to
 934                    reiserfs_get_block() that would do necessary conversion. */
 935                 cont_expand_offset =
 936                     le_key_k_offset(get_inode_item_key_version(inode),
 937                                     &(ih->ih_key));
 938                 pathrelse(&path);
 939                 res = generic_cont_expand(inode, cont_expand_offset);
 940         } else
 941                 pathrelse(&path);
 942
 943         reiserfs_write_unlock(inode->i_sb);
 944         return res;
 945 }
 946
 947 /* This function locks pages starting from @pos for @inode.
 948    @num_pages pages are locked and stored in
 949    @prepared_pages array. Also buffers are allocated for these pages.
 950    First and last page of the region is read if it is overwritten only
 951    partially. If last page did not exist before write (file hole or file
 952    append), it is zeroed, then.
 953    Returns number of unallocated blocks that should be allocated to cover
 954    new file data.*/
 955 static int reiserfs_prepare_file_region_for_write(struct inode *inode
 956                                                   /* Inode of the file */ ,
 957                                                   loff_t pos,   /* position in the file */
 958                                                   size_t num_pages,     /* number of pages to
 959                                                                            prepare */
 960                                                   size_t write_bytes,   /* Amount of bytes to be
 961                                                                            overwritten from
 962                                                                            @pos */
 963                                                   struct page **prepared_pages  /* pointer to array
 964                                                                                    where to store
 965                                                                                    prepared pages */
 966     )
 967 {
 968         int res = 0;            // Return values of different functions we call.
 969         unsigned long index = pos >> PAGE_CACHE_SHIFT;  // Offset in file in pages.
 970         int from = (pos & (PAGE_CACHE_SIZE - 1));       // Writing offset in first page
 971         int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;
 972         /* offset of last modified byte in last
 973            page */
 974         struct address_space *mapping = inode->i_mapping;       // Pages are mapped here.
 975         int i;                  // Simple counter
 976         int blocks = 0;         /* Return value (blocks that should be allocated) */
 977         struct buffer_head *bh, *head;  // Current bufferhead and first bufferhead
 978         // of a page.
 979         unsigned block_start, block_end;        // Starting and ending offsets of current
 980         // buffer in the page.
 981         struct buffer_head *wait[2], **wait_bh = wait;  // Buffers for page, if
 982         // Page appeared to be not up
 983         // to date. Note how we have
 984         // at most 2 buffers, this is
 985         // because we at most may
 986         // partially overwrite two
 987         // buffers for one page. One at                                                 // the beginning of write area
 988         // and one at the end.
 989         // Everything inthe middle gets                                                 // overwritten totally.
 990
 991         struct cpu_key key;     // cpu key of item that we are going to deal with
 992         struct item_head *ih = NULL;    // pointer to item head that we are going to deal with
 993         struct buffer_head *itembuf = NULL;     // Buffer head that contains items that we are going to deal with
 994         INITIALIZE_PATH(path);  // path to item, that we are going to deal with.
 995         __le32 *item = NULL;    // pointer to item we are going to deal with
 996         int item_pos = -1;      /* Position in indirect item */
 997
 998         if (num_pages < 1) {
 999                 reiserfs_warning(inode->i_sb,
1000                                  "green-9001: reiserfs_prepare_file_region_for_write "
1001                                  "called with zero number of pages to process");
1002                 return -EFAULT;
1003         }
1004
1005         /* We have 2 loops for pages. In first loop we grab and lock the pages, so
1006            that nobody would touch these until we release the pages. Then
1007            we'd start to deal with mapping buffers to blocks. */
1008         for (i = 0; i < num_pages; i++) {
1009                 prepared_pages[i] = grab_cache_page(mapping, index + i);        // locks the page
1010                 if (!prepared_pages[i]) {
1011                         res = -ENOMEM;
1012                         goto failed_page_grabbing;
1013                 }
1014                 if (!page_has_buffers(prepared_pages[i]))
1015                         create_empty_buffers(prepared_pages[i],
1016                                              inode->i_sb->s_blocksize, 0);
1017         }
1018
1019         /* Let's count amount of blocks for a case where all the blocks
1020            overwritten are new (we will substract already allocated blocks later) */
1021         if (num_pages > 2)
1022                 /* These are full-overwritten pages so we count all the blocks in
1023                    these pages are counted as needed to be allocated */
1024                 blocks =
1025                     (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1026
1027         /* count blocks needed for first page (possibly partially written) */
1028         blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) + !!(from & (inode->i_sb->s_blocksize - 1));   /* roundup */
1029
1030         /* Now we account for last page. If last page == first page (we
1031            overwrite only one page), we substract all the blocks past the
1032            last writing position in a page out of already calculated number
1033            of blocks */
1034         blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT - inode->i_blkbits)) -
1035             ((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
1036         /* Note how we do not roundup here since partial blocks still
1037            should be allocated */
1038
1039         /* Now if all the write area lies past the file end, no point in
1040            maping blocks, since there is none, so we just zero out remaining
1041            parts of first and last pages in write area (if needed) */
1042         if ((pos & ~((loff_t) PAGE_CACHE_SIZE - 1)) > inode->i_size) {
1043                 if (from != 0) {        /* First page needs to be partially zeroed */
1044                         char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
1045                         memset(kaddr, 0, from);
1046                         kunmap_atomic(kaddr, KM_USER0);
1047                 }
1048                 if (to != PAGE_CACHE_SIZE) {    /* Last page needs to be partially zeroed */
1049                         char *kaddr =
1050                             kmap_atomic(prepared_pages[num_pages - 1],
1051                                         KM_USER0);
1052                         memset(kaddr + to, 0, PAGE_CACHE_SIZE - to);
1053                         kunmap_atomic(kaddr, KM_USER0);
1054                 }
1055
1056                 /* Since all blocks are new - use already calculated value */
1057                 return blocks;
1058         }
1059
1060         /* Well, since we write somewhere into the middle of a file, there is
1061            possibility we are writing over some already allocated blocks, so
1062            let's map these blocks and substract number of such blocks out of blocks
1063            we need to allocate (calculated above) */
1064         /* Mask write position to start on blocksize, we do it out of the
1065            loop for performance reasons */
1066         pos &= ~((loff_t) inode->i_sb->s_blocksize - 1);
1067         /* Set cpu key to the starting position in a file (on left block boundary) */
1068         make_cpu_key(&key, inode,
1069                      1 + ((pos) & ~((loff_t) inode->i_sb->s_blocksize - 1)),
1070                      TYPE_ANY, 3 /*key length */ );
1071
1072         reiserfs_write_lock(inode->i_sb);       // We need that for at least search_by_key()
1073         for (i = 0; i < num_pages; i++) {
1074
1075                 head = page_buffers(prepared_pages[i]);
1076                 /* For each buffer in the page */
1077                 for (bh = head, block_start = 0; bh != head || !block_start;
1078                      block_start = block_end, bh = bh->b_this_page) {
1079                         if (!bh)
1080                                 reiserfs_panic(inode->i_sb,
1081                                                "green-9002: Allocated but absent buffer for a page?");
1082                         /* Find where this buffer ends */
1083                         block_end = block_start + inode->i_sb->s_blocksize;
1084                         if (i == 0 && block_end <= from)
1085                                 /* if this buffer is before requested data to map, skip it */
1086                                 continue;
1087
1088                         if (i == num_pages - 1 && block_start >= to) {
1089                                 /* If this buffer is after requested data to map, abort
1090                                    processing of current page */
1091                                 break;
1092                         }
1093
1094                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1095                                 /* This is optimisation for a case where buffer is mapped
1096                                    and have blocknumber assigned. In case significant amount
1097                                    of such buffers are present, we may avoid some amount
1098                                    of search_by_key calls.
1099                                    Probably it would be possible to move parts of this code
1100                                    out of BKL, but I afraid that would overcomplicate code
1101                                    without any noticeable benefit.
1102                                  */
1103                                 item_pos++;
1104                                 /* Update the key */
1105                                 set_cpu_key_k_offset(&key,
1106                                                      cpu_key_k_offset(&key) +
1107                                                      inode->i_sb->s_blocksize);
1108                                 blocks--;       // Decrease the amount of blocks that need to be
1109                                 // allocated
1110                                 continue;       // Go to the next buffer
1111                         }
1112
1113                         if (!itembuf || /* if first iteration */
1114                             item_pos >= ih_item_len(ih) / UNFM_P_SIZE) {        /* or if we progressed past the
1115                                                                                    current unformatted_item */
1116                                 /* Try to find next item */
1117                                 res =
1118                                     search_for_position_by_key(inode->i_sb,
1119                                                                &key, &path);
1120                                 /* Abort if no more items */
1121                                 if (res != POSITION_FOUND) {
1122                                         /* make sure later loops don't use this item */
1123                                         itembuf = NULL;
1124                                         item = NULL;
1125                                         break;
1126                                 }
1127
1128                                 /* Update information about current indirect item */
1129                                 itembuf = get_last_bh(&path);
1130                                 ih = get_ih(&path);
1131                                 item = get_item(&path);
1132                                 item_pos = path.pos_in_item;
1133
1134                                 RFALSE(!is_indirect_le_ih(ih),
1135                                        "green-9003: indirect item expected");
1136                         }
1137
1138                         /* See if there is some block associated with the file
1139                            at that position, map the buffer to this block */
1140                         if (get_block_num(item, item_pos)) {
1141                                 map_bh(bh, inode->i_sb,
1142                                        get_block_num(item, item_pos));
1143                                 blocks--;       // Decrease the amount of blocks that need to be
1144                                 // allocated
1145                         }
1146                         item_pos++;
1147                         /* Update the key */
1148                         set_cpu_key_k_offset(&key,
1149                                              cpu_key_k_offset(&key) +
1150                                              inode->i_sb->s_blocksize);
1151                 }
1152         }
1153         pathrelse(&path);       // Free the path
1154         reiserfs_write_unlock(inode->i_sb);
1155
1156         /* Now zero out unmappend buffers for the first and last pages of
1157            write area or issue read requests if page is mapped. */
1158         /* First page, see if it is not uptodate */
1159         if (!PageUptodate(prepared_pages[0])) {
1160                 head = page_buffers(prepared_pages[0]);
1161
1162                 /* For each buffer in page */
1163                 for (bh = head, block_start = 0; bh != head || !block_start;
1164                      block_start = block_end, bh = bh->b_this_page) {
1165
1166                         if (!bh)
1167                                 reiserfs_panic(inode->i_sb,
1168                                                "green-9002: Allocated but absent buffer for a page?");
1169                         /* Find where this buffer ends */
1170                         block_end = block_start + inode->i_sb->s_blocksize;
1171                         if (block_end <= from)
1172                                 /* if this buffer is before requested data to map, skip it */
1173                                 continue;
1174                         if (block_start < from) {       /* Aha, our partial buffer */
1175                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1176                                                                    issue READ request for it to
1177                                                                    not loose data */
1178                                         ll_rw_block(READ, 1, &bh);
1179                                         *wait_bh++ = bh;
1180                                 } else {        /* Not mapped, zero it */
1181                                         char *kaddr =
1182                                             kmap_atomic(prepared_pages[0],
1183                                                         KM_USER0);
1184                                         memset(kaddr + block_start, 0,
1185                                                from - block_start);
1186                                         kunmap_atomic(kaddr, KM_USER0);
1187                                         set_buffer_uptodate(bh);
1188                                 }
1189                         }
1190                 }
1191         }
1192
1193         /* Last page, see if it is not uptodate, or if the last page is past the end of the file. */
1194         if (!PageUptodate(prepared_pages[num_pages - 1]) ||
1195             ((pos + write_bytes) >> PAGE_CACHE_SHIFT) >
1196             (inode->i_size >> PAGE_CACHE_SHIFT)) {
1197                 head = page_buffers(prepared_pages[num_pages - 1]);
1198
1199                 /* for each buffer in page */
1200                 for (bh = head, block_start = 0; bh != head || !block_start;
1201                      block_start = block_end, bh = bh->b_this_page) {
1202
1203                         if (!bh)
1204                                 reiserfs_panic(inode->i_sb,
1205                                                "green-9002: Allocated but absent buffer for a page?");
1206                         /* Find where this buffer ends */
1207                         block_end = block_start + inode->i_sb->s_blocksize;
1208                         if (block_start >= to)
1209                                 /* if this buffer is after requested data to map, skip it */
1210                                 break;
1211                         if (block_end > to) {   /* Aha, our partial buffer */
1212                                 if (buffer_mapped(bh)) {        /* If it is mapped, we need to
1213                                                                    issue READ request for it to
1214                                                                    not loose data */
1215                                         ll_rw_block(READ, 1, &bh);
1216                                         *wait_bh++ = bh;
1217                                 } else {        /* Not mapped, zero it */
1218                                         char *kaddr =
1219                                             kmap_atomic(prepared_pages
1220                                                         [num_pages - 1],
1221                                                         KM_USER0);
1222                                         memset(kaddr + to, 0, block_end - to);
1223                                         kunmap_atomic(kaddr, KM_USER0);
1224                                         set_buffer_uptodate(bh);
1225                                 }
1226                         }
1227                 }
1228         }
1229
1230         /* Wait for read requests we made to happen, if necessary */
1231         while (wait_bh > wait) {
1232                 wait_on_buffer(*--wait_bh);
1233                 if (!buffer_uptodate(*wait_bh)) {
1234                         res = -EIO;
1235                         goto failed_read;
1236                 }
1237         }
1238
1239         return blocks;
1240       failed_page_grabbing:
1241         num_pages = i;
1242       failed_read:
1243         reiserfs_unprepare_pages(prepared_pages, num_pages);
1244         return res;
1245 }
1246
1247 /* Write @count bytes at position @ppos in a file indicated by @file
1248    from the buffer @buf.
1249
1250    generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
1251    something simple that works.  It is not for serious use by general purpose filesystems, excepting the one that it was
1252    written for (ext2/3).  This is for several reasons:
1253
1254    * It has no understanding of any filesystem specific optimizations.
1255
1256    * It enters the filesystem repeatedly for each page that is written.
1257
1258    * It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
1259    * operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
1260    * to reiserfs which allows for fewer tree traversals.
1261
1262    * Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
1263
1264    * Asking the block allocation code for blocks one at a time is slightly less efficient.
1265
1266    All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
1267    use it, but we were in a hurry to make code freeze, and so it couldn't be revised then.  This new code should make
1268    things right finally.
1269
1270    Future Features: providing search_by_key with hints.
1271
1272 */
1273 static ssize_t reiserfs_file_write(struct file *file,   /* the file we are going to write into */
1274                                    const char __user * buf,     /*  pointer to user supplied data
1275                                                                    (in userspace) */
1276                                    size_t count,        /* amount of bytes to write */
1277                                    loff_t * ppos        /* pointer to position in file that we start writing at. Should be updated to
1278                                                          * new current position before returning. */
1279                                    )
1280 {
1281         size_t already_written = 0;     // Number of bytes already written to the file.
1282         loff_t pos;             // Current position in the file.
1283         ssize_t res;            // return value of various functions that we call.
1284         int err = 0;
1285         struct inode *inode = file->f_dentry->d_inode;  // Inode of the file that we are writing to.
1286         /* To simplify coding at this time, we store
1287            locked pages in array for now */
1288         struct page *prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME];
1289         struct reiserfs_transaction_handle th;
1290         th.t_trans_id = 0;
1291
1292         /* If a filesystem is converted from 3.5 to 3.6, we'll have v3.5 items
1293         * lying around (most of the disk, in fact). Despite the filesystem
1294         * now being a v3.6 format, the old items still can't support large
1295         * file sizes. Catch this case here, as the rest of the VFS layer is
1296         * oblivious to the different limitations between old and new items.
1297         * reiserfs_setattr catches this for truncates. This chunk is lifted
1298         * from generic_write_checks. */
1299         if (get_inode_item_key_version (inode) == KEY_FORMAT_3_5 &&
1300             *ppos + count > MAX_NON_LFS) {
1301                 if (*ppos >= MAX_NON_LFS) {
1302                         send_sig(SIGXFSZ, current, 0);
1303                         return -EFBIG;
1304                 }
1305                 if (count > MAX_NON_LFS - (unsigned long)*ppos)
1306                         count = MAX_NON_LFS - (unsigned long)*ppos;
1307         }
1308
1309         if (file->f_flags & O_DIRECT) { // Direct IO needs treatment
1310                 ssize_t result, after_file_end = 0;
1311                 if ((*ppos + count >= inode->i_size)
1312                     || (file->f_flags & O_APPEND)) {
1313                         /* If we are appending a file, we need to put this savelink in here.
1314                            If we will crash while doing direct io, finish_unfinished will
1315                            cut the garbage from the file end. */
1316                         reiserfs_write_lock(inode->i_sb);
1317                         err =
1318                             journal_begin(&th, inode->i_sb,
1319                                           JOURNAL_PER_BALANCE_CNT);
1320                         if (err) {
1321                                 reiserfs_write_unlock(inode->i_sb);
1322                                 return err;
1323                         }
1324                         reiserfs_update_inode_transaction(inode);
1325                         add_save_link(&th, inode, 1 /* Truncate */ );
1326                         after_file_end = 1;
1327                         err =
1328                             journal_end(&th, inode->i_sb,
1329                                         JOURNAL_PER_BALANCE_CNT);
1330                         reiserfs_write_unlock(inode->i_sb);
1331                         if (err)
1332                                 return err;
1333                 }
1334                 result = do_sync_write(file, buf, count, ppos);
1335
1336                 if (after_file_end) {   /* Now update i_size and remove the savelink */
1337                         struct reiserfs_transaction_handle th;
1338                         reiserfs_write_lock(inode->i_sb);
1339                         err = journal_begin(&th, inode->i_sb, 1);
1340                         if (err) {
1341                                 reiserfs_write_unlock(inode->i_sb);
1342                                 return err;
1343                         }
1344                         reiserfs_update_inode_transaction(inode);
1345                         mark_inode_dirty(inode);
1346                         err = journal_end(&th, inode->i_sb, 1);
1347                         if (err) {
1348                                 reiserfs_write_unlock(inode->i_sb);
1349                                 return err;
1350                         }
1351                         err = remove_save_link(inode, 1 /* truncate */ );
1352                         reiserfs_write_unlock(inode->i_sb);
1353                         if (err)
1354                                 return err;
1355                 }
1356
1357                 return result;
1358         }
1359
1360         if (unlikely((ssize_t) count < 0))
1361                 return -EINVAL;
1362
1363         if (unlikely(!access_ok(VERIFY_READ, buf, count)))
1364                 return -EFAULT;
1365
1366         mutex_lock(&inode->i_mutex);    // locks the entire file for just us
1367
1368         pos = *ppos;
1369
1370         /* Check if we can write to specified region of file, file
1371            is not overly big and this kind of stuff. Adjust pos and
1372            count, if needed */
1373         res = generic_write_checks(file, &pos, &count, 0);
1374         if (res)
1375                 goto out;
1376
1377         if (count == 0)
1378                 goto out;
1379
1380         res = remove_suid(file->f_dentry);
1381         if (res)
1382                 goto out;
1383
1384         file_update_time(file);
1385
1386         // Ok, we are done with all the checks.
1387
1388         // Now we should start real work
1389
1390         /* If we are going to write past the file's packed tail or if we are going
1391            to overwrite part of the tail, we need that tail to be converted into
1392            unformatted node */
1393         res = reiserfs_check_for_tail_and_convert(inode, pos, count);
1394         if (res)
1395                 goto out;
1396
1397         while (count > 0) {
1398                 /* This is the main loop in which we running until some error occures
1399                    or until we write all of the data. */
1400                 size_t num_pages;       /* amount of pages we are going to write this iteration */
1401                 size_t write_bytes;     /* amount of bytes to write during this iteration */
1402                 size_t blocks_to_allocate;      /* how much blocks we need to allocate for this iteration */
1403
1404                 /*  (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos */
1405                 num_pages = !!((pos + count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial
1406                                                                            pages */
1407                     ((count +
1408                       (pos & (PAGE_CACHE_SIZE - 1))) >> PAGE_CACHE_SHIFT);
1409                 /* convert size to amount of
1410                    pages */
1411                 reiserfs_write_lock(inode->i_sb);
1412                 if (num_pages > REISERFS_WRITE_PAGES_AT_A_TIME
1413                     || num_pages > reiserfs_can_fit_pages(inode->i_sb)) {
1414                         /* If we were asked to write more data than we want to or if there
1415                            is not that much space, then we shorten amount of data to write
1416                            for this iteration. */
1417                         num_pages =
1418                             min_t(size_t, REISERFS_WRITE_PAGES_AT_A_TIME,
1419                                   reiserfs_can_fit_pages(inode->i_sb));
1420                         /* Also we should not forget to set size in bytes accordingly */
1421                         write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
1422                             (pos & (PAGE_CACHE_SIZE - 1));
1423                         /* If position is not on the
1424                            start of the page, we need
1425                            to substract the offset
1426                            within page */
1427                 } else
1428                         write_bytes = count;
1429
1430                 /* reserve the blocks to be allocated later, so that later on
1431                    we still have the space to write the blocks to */
1432                 reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1433                                                       num_pages <<
1434                                                       (PAGE_CACHE_SHIFT -
1435                                                        inode->i_blkbits));
1436                 reiserfs_write_unlock(inode->i_sb);
1437
1438                 if (!num_pages) {       /* If we do not have enough space even for a single page... */
1439                         if (pos >
1440                             inode->i_size + inode->i_sb->s_blocksize -
1441                             (pos & (inode->i_sb->s_blocksize - 1))) {
1442                                 res = -ENOSPC;
1443                                 break;  // In case we are writing past the end of the last file block, break.
1444                         }
1445                         // Otherwise we are possibly overwriting the file, so
1446                         // let's set write size to be equal or less than blocksize.
1447                         // This way we get it correctly for file holes.
1448                         // But overwriting files on absolutelly full volumes would not
1449                         // be very efficient. Well, people are not supposed to fill
1450                         // 100% of disk space anyway.
1451                         write_bytes =
1452                             min_t(size_t, count,
1453                                   inode->i_sb->s_blocksize -
1454                                   (pos & (inode->i_sb->s_blocksize - 1)));
1455                         num_pages = 1;
1456                         // No blocks were claimed before, so do it now.
1457                         reiserfs_claim_blocks_to_be_allocated(inode->i_sb,
1458                                                               1 <<
1459                                                               (PAGE_CACHE_SHIFT
1460                                                                -
1461                                                                inode->
1462                                                                i_blkbits));
1463                 }
1464
1465                 /* Prepare for writing into the region, read in all the
1466                    partially overwritten pages, if needed. And lock the pages,
1467                    so that nobody else can access these until we are done.
1468                    We get number of actual blocks needed as a result. */
1469                 res = reiserfs_prepare_file_region_for_write(inode, pos,
1470                                                              num_pages,
1471                                                              write_bytes,
1472                                                              prepared_pages);
1473                 if (res < 0) {
1474                         reiserfs_release_claimed_blocks(inode->i_sb,
1475                                                         num_pages <<
1476                                                         (PAGE_CACHE_SHIFT -
1477                                                          inode->i_blkbits));
1478                         break;
1479                 }
1480
1481                 blocks_to_allocate = res;
1482
1483                 /* First we correct our estimate of how many blocks we need */
1484                 reiserfs_release_claimed_blocks(inode->i_sb,
1485                                                 (num_pages <<
1486                                                  (PAGE_CACHE_SHIFT -
1487                                                   inode->i_sb->
1488                                                   s_blocksize_bits)) -
1489                                                 blocks_to_allocate);
1490
1491                 if (blocks_to_allocate > 0) {   /*We only allocate blocks if we need to */
1492                         /* Fill in all the possible holes and append the file if needed */
1493                         res =
1494                             reiserfs_allocate_blocks_for_region(&th, inode, pos,
1495                                                                 num_pages,
1496                                                                 write_bytes,
1497                                                                 prepared_pages,
1498                                                                 blocks_to_allocate);
1499                 }
1500
1501                 /* well, we have allocated the blocks, so it is time to free
1502                    the reservation we made earlier. */
1503                 reiserfs_release_claimed_blocks(inode->i_sb,
1504                                                 blocks_to_allocate);
1505                 if (res) {
1506                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1507                         break;
1508                 }
1509
1510 /* NOTE that allocating blocks and filling blocks can be done in reverse order
1511    and probably we would do that just to get rid of garbage in files after a
1512    crash */
1513
1514                 /* Copy data from user-supplied buffer to file's pages */
1515                 res =
1516                     reiserfs_copy_from_user_to_file_region(pos, num_pages,
1517                                                            write_bytes,
1518                                                            prepared_pages, buf);
1519                 if (res) {
1520                         reiserfs_unprepare_pages(prepared_pages, num_pages);
1521                         break;
1522                 }
1523
1524                 /* Send the pages to disk and unlock them. */
1525                 res =
1526                     reiserfs_submit_file_region_for_write(&th, inode, pos,
1527                                                           num_pages,
1528                                                           write_bytes,
1529                                                           prepared_pages);
1530                 if (res)
1531                         break;
1532
1533                 already_written += write_bytes;
1534                 buf += write_bytes;
1535                 *ppos = pos += write_bytes;
1536                 count -= write_bytes;
1537                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
1538         }
1539
1540         /* this is only true on error */
1541         if (th.t_trans_id) {
1542                 reiserfs_write_lock(inode->i_sb);
1543                 err = journal_end(&th, th.t_super, th.t_blocks_allocated);
1544                 reiserfs_write_unlock(inode->i_sb);
1545                 if (err) {
1546                         res = err;
1547                         goto out;
1548                 }
1549         }
1550
1551         if (likely(res >= 0) &&
1552             (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))))
1553                 res = generic_osync_inode(inode, file->f_mapping,
1554                                           OSYNC_METADATA | OSYNC_DATA);
1555
1556         mutex_unlock(&inode->i_mutex);
1557         reiserfs_async_progress_wait(inode->i_sb);
1558         return (already_written != 0) ? already_written : res;
1559
1560       out:
1561         mutex_unlock(&inode->i_mutex);  // unlock the file on exit.
1562         return res;
1563 }
1564
1565 const struct file_operations reiserfs_file_operations = {
1566         .read = do_sync_read,
1567         .write = reiserfs_file_write,
1568         .ioctl = reiserfs_ioctl,
1569 #ifdef CONFIG_COMPAT
1570         .compat_ioctl = reiserfs_compat_ioctl,
1571 #endif
1572         .mmap = generic_file_mmap,
1573         .open = generic_file_open,
1574         .release = reiserfs_file_release,
1575         .fsync = reiserfs_sync_file,
1576         .sendfile = generic_file_sendfile,
1577         .aio_read = generic_file_aio_read,
1578         .aio_write = generic_file_aio_write,
1579         .splice_read = generic_file_splice_read,
1580         .splice_write = generic_file_splice_write,
1581 };
1582
1583 struct inode_operations reiserfs_file_inode_operations = {
1584         .truncate = reiserfs_vfs_truncate_file,
1585         .setattr = reiserfs_setattr,
1586         .setxattr = reiserfs_setxattr,
1587         .getxattr = reiserfs_getxattr,
1588         .listxattr = reiserfs_listxattr,
1589         .removexattr = reiserfs_removexattr,
1590         .permission = reiserfs_permission,
1591 };