fsys_reiserfs.c

i386的bootloader源码grub

		{
		  journal_table += j_len - i;
		  goto found;
		}
	    }
	}
      else
	{
	  /* This is the end of cached journal marker.  The remaining
	   * transactions are still on disk.
	   */
	  struct reiserfs_journal_desc   desc;
	  struct reiserfs_journal_commit commit;

	  if (! journal_read (desc_block, sizeof (desc), (char *) &desc))
	    return 0;

	  j_len = desc.j_len;
	  while (i < j_len && i < JOURNAL_TRANS_HALF)
	    if (desc.j_realblock[i++] == blockNr)
	      goto found;
	  
	  if (j_len >= JOURNAL_TRANS_HALF)
	    {
	      int commit_block = (desc_block + 1 + j_len) & journal_mask;
	      if (! journal_read (commit_block, 
				  sizeof (commit), (char *) &commit))
		return 0;
	      while (i < j_len)
		if (commit.j_realblock[i++ - JOURNAL_TRANS_HALF] == blockNr)
		  goto found;
	    }
	}
      goto not_found;
      
    found:
      translatedNr = INFO->journal_block + ((desc_block + i) & journal_mask);
#ifdef REISERDEBUG
      printf ("block_read: block %d is mapped to journal block %d.\n", 
	      blockNr, translatedNr - INFO->journal_block);
#endif
      /* We must continue the search, as this block may be overwritten
       * in later transactions.
       */
    not_found:
      desc_block = (desc_block + 2 + j_len) & journal_mask;
    }
  return devread (translatedNr << INFO->blocksize_shift, start, len, buffer);
}

/* Init the journal data structure.  We try to cache as much as
 * possible in the JOURNAL_START-JOURNAL_END space, but if it is full
 * we can still read the rest from the disk on demand.
 *
 * The first number of valid transactions and the descriptor block of the
 * first valid transaction are held in INFO.  The transactions are all 
 * adjacent, but we must take care of the journal wrap around. 
 */
static int
journal_init (void)
{
  unsigned int block_count = INFO->journal_block_count;
  unsigned int desc_block;
  unsigned int commit_block;
  unsigned int next_trans_id;
  struct reiserfs_journal_header header;
  struct reiserfs_journal_desc   desc;
  struct reiserfs_journal_commit commit;
  __u32 *journal_table = JOURNAL_START;

  journal_read (block_count, sizeof (header), (char *) &header);
  desc_block = header.j_first_unflushed_offset;
  if (desc_block >= block_count)
    return 0;

  INFO->journal_first_desc = desc_block;
  next_trans_id = header.j_last_flush_trans_id + 1;

#ifdef REISERDEBUG
  printf ("journal_init: last flushed %d\n", 
	  header.j_last_flush_trans_id);
#endif

  while (1) 
    {
      journal_read (desc_block, sizeof (desc), (char *) &desc);
      if (substring (JOURNAL_DESC_MAGIC, desc.j_magic) > 0
	  || desc.j_trans_id != next_trans_id
	  || desc.j_mount_id != header.j_mount_id)
	/* no more valid transactions */
	break;
      
      commit_block = (desc_block + desc.j_len + 1) & (block_count - 1);
      journal_read (commit_block, sizeof (commit), (char *) &commit);
      if (desc.j_trans_id != commit.j_trans_id
	  || desc.j_len != commit.j_len)
	/* no more valid transactions */
	break;
      
#ifdef REISERDEBUG
      printf ("Found valid transaction %d/%d at %d.\n", 
	      desc.j_trans_id, desc.j_mount_id, desc_block);
#endif

      next_trans_id++;
      if (journal_table < JOURNAL_END)
	{
	  if ((journal_table + 1 + desc.j_len) >= JOURNAL_END)
	    {
	      /* The table is almost full; mark the end of the cached
	       * journal.*/
	      *journal_table = 0xffffffff;
	      journal_table = JOURNAL_END;
	    }
	  else
	    {
	      int i;
	      /* Cache the length and the realblock numbers in the table.
	       * The block number of descriptor can easily be computed.
	       * and need not to be stored here.
	       */
	      *journal_table++ = desc.j_len;
	      for (i = 0; i < desc.j_len && i < JOURNAL_TRANS_HALF; i++)
		{
		  *journal_table++ = desc.j_realblock[i];
#ifdef REISERDEBUG
		  printf ("block %d is in journal %d.\n", 
			  desc.j_realblock[i], desc_block);
#endif
		}
	      for (     ; i < desc.j_len; i++)
		{
		  *journal_table++ = commit.j_realblock[i-JOURNAL_TRANS_HALF];
#ifdef REISERDEBUG
		  printf ("block %d is in journal %d.\n", 
			  commit.j_realblock[i-JOURNAL_TRANS_HALF], 
			  desc_block);
#endif
		}
	    }
	}
      desc_block = (commit_block + 1) & (block_count - 1);
    }
#ifdef REISERDEBUG
  printf ("Transaction %d/%d at %d isn't valid.\n", 
	  desc.j_trans_id, desc.j_mount_id, desc_block);
#endif

  INFO->journal_transactions
    = next_trans_id - header.j_last_flush_trans_id - 1;
  return errnum == 0;
}

/* check filesystem types and read superblock into memory buffer */
int
reiserfs_mount (void)
{
  struct reiserfs_super_block super;
  int superblock = REISERFS_DISK_OFFSET_IN_BYTES >> SECTOR_BITS;

  if (part_length < superblock + (sizeof (super) >> SECTOR_BITS)
      || ! devread (superblock, 0, sizeof (struct reiserfs_super_block), 
		(char *) &super)
      || (substring (REISER3FS_SUPER_MAGIC_STRING, super.s_magic) > 0
	  && substring (REISER2FS_SUPER_MAGIC_STRING, super.s_magic) > 0
	  && substring (REISERFS_SUPER_MAGIC_STRING, super.s_magic) > 0)
      || (/* check that this is not a copy inside the journal log */
	  super.s_journal_block * super.s_blocksize
	  <= REISERFS_DISK_OFFSET_IN_BYTES))
    {
      /* Try old super block position */
      superblock = REISERFS_OLD_DISK_OFFSET_IN_BYTES >> SECTOR_BITS;
      if (part_length < superblock + (sizeof (super) >> SECTOR_BITS)
	  || ! devread (superblock, 0, sizeof (struct reiserfs_super_block), 
			(char *) &super))
	return 0;

      if (substring (REISER3FS_SUPER_MAGIC_STRING, super.s_magic) > 0
	  && substring (REISER2FS_SUPER_MAGIC_STRING, super.s_magic) > 0
	  && substring (REISERFS_SUPER_MAGIC_STRING, super.s_magic) > 0)
	{
	  /* pre journaling super block ? */
	  if (substring (REISERFS_SUPER_MAGIC_STRING, 
			 (char*) ((int) &super + 20)) > 0)
	    return 0;
	  
	  super.s_blocksize = REISERFS_OLD_BLOCKSIZE;
	  super.s_journal_block = 0;
	  super.s_version = 0;
	}
    }

  /* check the version number.  */
  if (super.s_version > REISERFS_MAX_SUPPORTED_VERSION)
    return 0;
  
  INFO->version = super.s_version;
  INFO->blocksize = super.s_blocksize;
  INFO->fullblocksize_shift = log2 (super.s_blocksize);
  INFO->blocksize_shift = INFO->fullblocksize_shift - SECTOR_BITS;
  INFO->cached_slots = 
    (FSYSREISER_CACHE_SIZE >> INFO->fullblocksize_shift) - 1;

#ifdef REISERDEBUG
  printf ("reiserfs_mount: version=%d, blocksize=%d\n", 
	  INFO->version, INFO->blocksize);
#endif /* REISERDEBUG */

  /* Clear node cache. */
  memset (INFO->blocks, 0, sizeof (INFO->blocks));

  if (super.s_blocksize < FSYSREISER_MIN_BLOCKSIZE
      || super.s_blocksize > FSYSREISER_MAX_BLOCKSIZE
      || (SECTOR_SIZE << INFO->blocksize_shift) != super.s_blocksize)
    return 0;

  /* Initialize journal code.  If something fails we end with zero
   * journal_transactions, so we don't access the journal at all.  
   */
  INFO->journal_transactions = 0;
  if (super.s_journal_block != 0 && super.s_journal_dev == 0)
    {
      INFO->journal_block = super.s_journal_block;
      INFO->journal_block_count = super.s_journal_size;
      if (is_power_of_two (INFO->journal_block_count))
	journal_init ();

      /* Read in super block again, maybe it is in the journal */
      block_read (superblock >> INFO->blocksize_shift, 
		  0, sizeof (struct reiserfs_super_block), (char *) &super);
    }

  if (! block_read (super.s_root_block, 0, INFO->blocksize, (char*) ROOT))
    return 0;
  
  INFO->tree_depth = BLOCKHEAD (ROOT)->blk_level;
  
#ifdef REISERDEBUG
  printf ("root read_in: block=%d, depth=%d\n", 
	  super.s_root_block, INFO->tree_depth);
#endif /* REISERDEBUG */

  if (INFO->tree_depth >= MAX_HEIGHT)
    return 0;
  if (INFO->tree_depth == DISK_LEAF_NODE_LEVEL)
    {
      /* There is only one node in the whole filesystem, 
       * which is simultanously leaf and root */
      memcpy (LEAF, ROOT, INFO->blocksize);
    }
  return 1;
}

/***************** TREE ACCESSING METHODS *****************************/

/* I assume you are familiar with the ReiserFS tree, if not go to
 * http://www.namesys.com/content_table.html
 *
 * My tree node cache is organized as following
 *   0   ROOT node
 *   1   LEAF node  (if the ROOT is also a LEAF it is copied here
 *   2-n other nodes on current path from bottom to top.  
 *       if there is not enough space in the cache, the top most are
 *       omitted.
 *
 * I have only two methods to find a key in the tree:
 *   search_stat(dir_id, objectid) searches for the stat entry (always
 *       the first entry) of an object.
 *   next_key() gets the next key in tree order.
 *
 * This means, that I can only sequential reads of files are
 * efficient, but this really doesn't hurt for grub.  
 */

/* Read in the node at the current path and depth into the node cache.
 * You must set INFO->blocks[depth] before.
 */
static char *
read_tree_node (unsigned int blockNr, int depth)
{
  char* cache = CACHE(depth);
  int num_cached = INFO->cached_slots;
  if (depth < num_cached)
    {
      /* This is the cached part of the path.  Check if same block is
       * needed.  
       */
      if (blockNr == INFO->blocks[depth])
	return cache;
    }
  else
    cache = CACHE(num_cached);

#ifdef REISERDEBUG
  printf ("  next read_in: block=%d (depth=%d)\n",
	  blockNr, depth);
#endif /* REISERDEBUG */
  if (! block_read (blockNr, 0, INFO->blocksize, cache))
    return 0;
  /* Make sure it has the right node level */
  if (BLOCKHEAD (cache)->blk_level != depth)
    {
      errnum = ERR_FSYS_CORRUPT;
      return 0;
    }

  INFO->blocks[depth] = blockNr;
  return cache;
}

/* Get the next key, i.e. the key following the last retrieved key in
 * tree order.  INFO->current_ih and 
 * INFO->current_info are adapted accordingly.  */
static int
next_key (void)
{
  int depth;
  struct item_head *ih = INFO->current_ih + 1;
  char *cache;
  
#ifdef REISERDEBUG
  printf ("next_key:\n  old ih: key %d:%d:%d:%d version:%d\n", 
	  INFO->current_ih->ih_key.k_dir_id, 
	  INFO->current_ih->ih_key.k_objectid, 
	  INFO->current_ih->ih_key.u.v1.k_offset,
	  INFO->current_ih->ih_key.u.v1.k_uniqueness,
	  INFO->current_ih->ih_version);
#endif /* REISERDEBUG */
  
  if (ih == &ITEMHEAD[BLOCKHEAD (LEAF)->blk_nr_item])
    {
      depth = DISK_LEAF_NODE_LEVEL;
      /* The last item, was the last in the leaf node.  
       * Read in the next block 
       */
      do
	{
	  if (depth == INFO->tree_depth)
	    {
	      /* There are no more keys at all.
	       * Return a dummy item with MAX_KEY */
	      ih = (struct item_head *) &BLOCKHEAD (LEAF)->blk_right_delim_key;
	      goto found;
	    }
	  depth++;
#ifdef REISERDEBUG
	  printf ("  depth=%d, i=%d\n", depth, INFO->next_key_nr[depth]);
#endif /* REISERDEBUG */
	}
      while (INFO->next_key_nr[depth] == 0);
      
      if (depth == INFO->tree_depth)
	cache = ROOT;
      else if (depth <= INFO->cached_slots)
	cache = CACHE (depth);
      else 
	{
	  cache = read_tree_node (INFO->blocks[depth], depth);
	  if (! cache)
	    return 0;
	}
      
      do
	{
	  int nr_item = BLOCKHEAD (cache)->blk_nr_item;
	  int key_nr = INFO->next_key_nr[depth]++;
#ifdef REISERDEBUG
	  printf ("  depth=%d, i=%d/%d\n", depth, key_nr, nr_item);
#endif /* REISERDEBUG */
	  if (key_nr == nr_item)
	    /* This is the last item in this block, set the next_key_nr to 0 */
	    INFO->next_key_nr[depth] = 0;

	  cache = read_tree_node (DC (cache)[key_nr].dc_block_number, --depth);
	  if (! cache)
	    return 0;
	}
      while (depth > DISK_LEAF_NODE_LEVEL);
      
      ih = ITEMHEAD;
    }
 found:
  INFO->current_ih   = ih;
  INFO->current_item = &LEAF[ih->ih_item_location];
#ifdef REISERDEBUG
  printf ("  new ih: key %d:%d:%d:%d version:%d\n", 
	  INFO->current_ih->ih_key.k_dir_id, 
	  INFO->current_ih->ih_key.k_objectid, 
	  INFO->current_ih->ih_key.u.v1.k_offset,
	  INFO->current_ih->ih_key.u.v1.k_uniqueness,
	  INFO->current_ih->ih_version);
#endif /* REISERDEBUG */
  return 1;
}

/* preconditions: reiserfs_mount already executed, therefore 
 *   INFO block is valid
 * returns: 0 if error (errnum is set), 
 *   nonzero iff we were able to find the key successfully.
 * postconditions: on a nonzero return, the current_ih and 
 *   current_item fields describe the key that equals the
 *   searched key.  INFO->next_key contains the next key after
 *   the searched key.
 * side effects: messes around with the cache.
 */
static int
search_stat (__u32 dir_id, __u32 objectid) 
{
  char *cache;
  int depth;
  int nr_item;
  int i;
  struct item_head *ih;