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