Example #1
0
/*
 * Split leaf, then insert to parent.
 * key:  key to add after split (cursor will point leaf which is including key)
 * hint: hint for split
 *
 * return value:
 *   0 - success
 * < 0 - error
 */
static int btree_leaf_split(struct cursor *cursor, tuxkey_t key, tuxkey_t hint)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	trace("split leaf");
	struct btree *btree = cursor->btree;
	struct buffer_head *newbuf;

	newbuf = new_leaf(btree);
	if (IS_ERR(newbuf))
		return PTR_ERR(newbuf);
	log_balloc(btree->sb, bufindex(newbuf), 1);

	struct buffer_head *leafbuf = cursor_leafbuf(cursor);
	tuxkey_t newkey = btree->ops->leaf_split(btree, hint, bufdata(leafbuf),
						 bufdata(newbuf));
	assert(cursor_this_key(cursor) < newkey);
	assert(newkey < cursor_next_key(cursor));
	if (key < newkey)
		mark_buffer_dirty_non(newbuf);
	else
		mark_buffer_dirty_non(leafbuf);
	return insert_leaf(cursor, newkey, newbuf, key < newkey);
}
Example #2
0
static loff_t unatom_dict_write(struct inode *atable, atom_t atom, loff_t where)
{
	unsigned delta = tux3_get_current_delta();
	struct buffer_head *buffer, *clone;
	loff_t old;
	unsigned offset;

	buffer = blockread_unatom(atable, atom, &offset);
	if (!buffer)
		return -EIO;

	/*
	 * The atable is protected by i_mutex for now.
	 * blockdirty() should never return -EAGAIN.
	 * FIXME: need finer granularity locking
	 */
	clone = blockdirty(buffer, delta);
	if (IS_ERR(clone)) {
		assert(PTR_ERR(clone) != -EAGAIN);
		blockput(buffer);
		return PTR_ERR(clone);
	}

	__be64 *unatom_dict = bufdata(clone);
	old = be64_to_cpu(unatom_dict[offset]);
	unatom_dict[offset] = cpu_to_be64(where);
	mark_buffer_dirty_non(clone);
	blockput(clone);

	return old;
}
Example #3
0
/* Modify buffer of refcount, then release buffer */
static int update_refcount(struct sb *sb, struct buffer_head *buffer,
			   unsigned offset, u16 val)
{
	unsigned delta = tux3_get_current_delta();
	struct buffer_head *clone;
	__be16 *refcount;

	/*
	 * The atable is protected by i_mutex for now.
	 * blockdirty() should never return -EAGAIN.
	 * FIXME: need finer granularity locking
	 */
	clone = blockdirty(buffer, delta);
	if (IS_ERR(clone)) {
		assert(PTR_ERR(clone) != -EAGAIN);
		blockput(buffer);
		return PTR_ERR(clone);
	}

	refcount = bufdata(clone);
	refcount[offset] = cpu_to_be16(val);
	mark_buffer_dirty_non(clone);
	blockput(clone);

	return 0;
}
Example #4
0
File: dir.c Project: Zkin/tux3
static void tux_update_entry(struct buffer_head *buffer, tux_dirent *entry,
			     inum_t inum, umode_t mode)
{
	entry->inum = cpu_to_be64(inum);
	entry->type = tux_type_by_mode[(mode & S_IFMT) >> STAT_SHIFT];
	mark_buffer_dirty_non(buffer);
	blockput(buffer);
}
Example #5
0
File: btree.c Project: Zkin/tux3
/*
 * Split leaf, then insert to parent.
 * key:  key to add after split (cursor will point leaf which is including key)
 * hint: hint for split
 *
 * return value:
 *   0 - success
 * < 0 - error
 */
static int btree_leaf_split(struct cursor *cursor, tuxkey_t key, tuxkey_t hint)
{
	trace("split leaf");
	struct btree *btree = cursor->btree;
	struct buffer_head *newbuf;

	newbuf = new_leaf(btree);
	if (IS_ERR(newbuf))
		return PTR_ERR(newbuf);
	log_balloc(btree->sb, bufindex(newbuf), 1);

	struct buffer_head *leafbuf = cursor_leafbuf(cursor);
	tuxkey_t newkey = btree->ops->leaf_split(btree, hint, bufdata(leafbuf),
						 bufdata(newbuf));
	assert(cursor_this_key(cursor) < newkey);
	assert(newkey < cursor_next_key(cursor));
	if (key < newkey)
		mark_buffer_dirty_non(newbuf);
	else
		mark_buffer_dirty_non(leafbuf);
	return insert_leaf(cursor, newkey, newbuf, key < newkey);
}
Example #6
0
static void add_maps(struct inode *inode, block_t index,
		     struct block_segment *seg, int nr_segs)
{
	unsigned delta = tux3_get_current_delta();

	for (int i = 0; i < nr_segs; i++) {
		struct block_segment *s = &seg[i];
		for (unsigned j = 0; j < s->count; j++) {
			struct buffer_head *buf;
			buf = blockget(inode->map, index + j);
			buf = blockdirty(buf, delta);
			memset(buf->data, 0, inode->i_sb->blocksize);
			*(block_t *)buf->data = s->block + j;
			mark_buffer_dirty_non(buf);
			blockput(buf);
		}
		index += s->count;
	}
}
Example #7
0
File: dir.c Project: Zkin/tux3
int tux_delete_entry(struct inode *dir, struct buffer_head *buffer,
		     tux_dirent *entry)
{
	unsigned delta = tux3_get_current_delta();
	tux_dirent *prev = NULL, *this = bufdata(buffer);
	struct buffer_head *clone;
	void *olddata;

	while ((char *)this < (char *)entry) {
		if (this->rec_len == 0) {
			blockput(buffer);
			tux_zero_len_error(dir, bufindex(buffer));
			return -EIO;
		}
		prev = this;
		this = next_entry(this);
	}

	/*
	 * The directory is protected by i_mutex.
	 * blockdirty() should never return -EAGAIN.
	 */
	olddata = bufdata(buffer);
	clone = blockdirty(buffer, delta);
	if (IS_ERR(clone)) {
		assert(PTR_ERR(clone) != -EAGAIN);
		blockput(buffer);
		return PTR_ERR(clone);
	}
	entry = ptr_redirect(entry, olddata, bufdata(clone));
	prev = ptr_redirect(prev, olddata, bufdata(clone));

	if (prev)
		prev->rec_len = tux_rec_len_to_disk((void *)next_entry(entry) - (void *)prev);
	memset(entry->name, 0, entry->name_len);
	entry->name_len = entry->type = 0;
	entry->inum = 0;

	mark_buffer_dirty_non(clone);
	blockput(clone);

	return 0;
}
Example #8
0
int btree_write(struct cursor *cursor, struct btree_key_range *key)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct btree *btree = cursor->btree;
	struct btree_ops *ops = btree->ops;
	tuxkey_t split_hint;
	int err;

	/* FIXME: we might be better to support multiple leaves */

	err = cursor_redirect(cursor);
	if (err)
		return err;

	while (key->len > 0) {
		tuxkey_t bottom = cursor_this_key(cursor);
		tuxkey_t limit = cursor_next_key(cursor);
		void *leaf = bufdata(cursor_leafbuf(cursor));
		int need_split;

		assert(bottom <= key->start && key->start < limit);
		assert(ops->leaf_sniff(btree, leaf));

		need_split = ops->leaf_write(btree, bottom, limit, leaf, key,
					     &split_hint);
		if (need_split < 0)
			return need_split;
		else if (!need_split) {
			mark_buffer_dirty_non(cursor_leafbuf(cursor));
			continue;
		}

		err = btree_leaf_split(cursor, key->start, split_hint);
		if (err)
			return err;	/* FIXME: error handling */
	}

	return 0;
}
Example #9
0
File: btree.c Project: Zkin/tux3
int btree_write(struct cursor *cursor, struct btree_key_range *key)
{
	struct btree *btree = cursor->btree;
	struct btree_ops *ops = btree->ops;
	tuxkey_t split_hint;
	int err;

	/* FIXME: we might be better to support multiple leaves */

	err = cursor_redirect(cursor);
	if (err)
		return err;

	while (key->len > 0) {
		tuxkey_t bottom = cursor_this_key(cursor);
		tuxkey_t limit = cursor_next_key(cursor);
		void *leaf = bufdata(cursor_leafbuf(cursor));

		assert(bottom <= key->start && key->start < limit);
		assert(ops->leaf_sniff(btree, leaf));

		err = ops->leaf_write(btree, bottom, limit, leaf, key,
				      &split_hint);
		if (!err) {
			mark_buffer_dirty_non(cursor_leafbuf(cursor));
			continue;
		}
		assert(err == -ENOSPC);

		err = btree_leaf_split(cursor, key->start, split_hint);
		if (err)
			break;	/* FIXME: error handling */
	}

	return err;
}
Example #10
0
/*
 * This is range deletion. So, instead of adjusting balance of the
 * space on sibling nodes for each change, this just removes the range
 * and merges from right to left even if it is not same parent.
 *
 *              +--------------- (A, B, C)--------------------+
 *              |                    |                        |
 *     +-- (AA, AB, AC) -+       +- (BA, BB, BC) -+      + (CA, CB, CC) +
 *     |        |        |       |        |       |      |       |      |
 * (AAA,AAB)(ABA,ABB)(ACA,ACB) (BAA,BAB)(BBA)(BCA,BCB)  (CAA)(CBA,CBB)(CCA)
 *
 * [less : A, AA, AAA, AAB, AB, ABA, ABB, AC, ACA, ACB, B, BA ... : greater]
 *
 * If we merged from cousin (or re-distributed), we may have to update
 * the index until common parent. (e.g. removed (ACB), then merged
 * from (BAA,BAB) to (ACA), we have to adjust B in root node to BB)
 *
 * See, adjust_parent_sep().
 *
 * FIXME: no re-distribute. so, we don't guarantee above than 50%
 * space efficiency. And if range is end of key (truncate() case), we
 * don't need to merge, and adjust_parent_sep().
 *
 * FIXME2: we may want to split chop work for each step. instead of
 * blocking for a long time.
 */
int btree_chop(struct btree *btree, tuxkey_t start, u64 len)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = btree->sb;
	struct btree_ops *ops = btree->ops;
	struct buffer_head **prev, *leafprev = NULL;
	struct chopped_index_info *cii;
	struct cursor *cursor;
	tuxkey_t limit;
	int ret, done = 0;

	if (!has_root(btree))
		return 0;

	/* Chop all range if len >= TUXKEY_LIMIT */
	limit = (len >= TUXKEY_LIMIT) ? TUXKEY_LIMIT : start + len;

	prev = malloc(sizeof(*prev) * btree->root.depth);
	if (prev == NULL)
		return -ENOMEM;
	memset(prev, 0, sizeof(*prev) * btree->root.depth);

	cii = malloc(sizeof(*cii) * btree->root.depth);
	if (cii == NULL) {
		ret = -ENOMEM;
		goto error_cii;
	}
	memset(cii, 0, sizeof(*cii) * btree->root.depth);

	cursor = alloc_cursor(btree, 0);
	if (!cursor) {
		ret = -ENOMEM;
		goto error_alloc_cursor;
	}

	down_write(&btree->lock);
	ret = btree_probe(cursor, start);
	if (ret)
		goto error_btree_probe;

	/* Walk leaves */
	while (1) {
		struct buffer_head *leafbuf;
		tuxkey_t this_key;

		/*
		 * FIXME: If leaf was merged and freed later, we don't
		 * need to redirect leaf and leaf_chop()
		 */
		if ((ret = cursor_redirect(cursor)))
			goto out;
		leafbuf = cursor_pop(cursor);

		/* Adjust start and len for this leaf */
		this_key = cursor_level_this_key(cursor);
		if (start < this_key) {
			if (limit < TUXKEY_LIMIT)
				len -= this_key - start;
			start = this_key;
		}

		ret = ops->leaf_chop(btree, start, len, bufdata(leafbuf));
		if (ret) {
			if (ret < 0) {
				blockput(leafbuf);
				goto out;
			}
			mark_buffer_dirty_non(leafbuf);
		}

		/* Try to merge this leaf with prev */
		if (leafprev) {
			if (try_leaf_merge(btree, leafprev, leafbuf)) {
				trace(">>> can merge leaf %p into leaf %p", leafbuf, leafprev);
				remove_index(cursor, cii);
				mark_buffer_dirty_non(leafprev);
				blockput_free(sb, leafbuf);
				goto keep_prev_leaf;
			}
			blockput(leafprev);
		}
		leafprev = leafbuf;

keep_prev_leaf:

		if (cursor_level_next_key(cursor) >= limit)
			done = 1;
		/* Pop and try to merge finished nodes */
		while (done || cursor_level_finished(cursor)) {
			struct buffer_head *buf;
			int level = cursor->level;
			struct chopped_index_info *ciil = &cii[level];


			/* Get merge src buffer, and go parent level */
			buf = cursor_pop(cursor);

			/*
			 * Logging chopped indexes
			 * FIXME: If node is freed later (e.g. merged),
			 * we dont't need to log this
			 */
			if (ciil->count) {
				log_bnode_del(sb, bufindex(buf), ciil->start,
					      ciil->count);
			}
			memset(ciil, 0, sizeof(*ciil));

			/* Try to merge node with prev */
			if (prev[level]) {
				assert(level);
				if (try_bnode_merge(sb, prev[level], buf)) {
					trace(">>> can merge node %p into node %p", buf, prev[level]);
					remove_index(cursor, cii);
					mark_buffer_unify_non(prev[level]);
					blockput_free_unify(sb, buf);
					goto keep_prev_node;
				}
				blockput(prev[level]);
			}
			prev[level] = buf;
keep_prev_node:

			if (!level)
				goto chop_root;
		}

		/* Push back down to leaf level */
		do {
			ret = cursor_advance_down(cursor);
			if (ret < 0)
				goto out;
		} while (ret);
	}

chop_root:
	/* Remove depth if possible */
	while (btree->root.depth > 1 && bcount(bufdata(prev[0])) == 1) {
		trace("drop btree level");
		btree->root.block = bufindex(prev[1]);
		btree->root.depth--;
		tux3_mark_btree_dirty(btree);

		/*
		 * We know prev[0] is redirected and dirty. So, in
		 * here, we can just cancel bnode_redirect by bfree(),
		 * instead of defered_bfree()
		 * FIXME: we can optimize freeing bnode without
		 * bnode_redirect, and if we did, this is not true.
		 */
		bfree(sb, bufindex(prev[0]), 1);
		log_bnode_free(sb, bufindex(prev[0]));
		blockput_free_unify(sb, prev[0]);

		vecmove(prev, prev + 1, btree->root.depth);
	}
	ret = 0;

out:
	if (leafprev)
		blockput(leafprev);
	for (int i = 0; i < btree->root.depth; i++) {
		if (prev[i])
			blockput(prev[i]);
	}
	release_cursor(cursor);
error_btree_probe:
	up_write(&btree->lock);

	free_cursor(cursor);
error_alloc_cursor:
	free(cii);
error_cii:
	free(prev);

	return ret;
}
Example #11
0
int alloc_empty_btree(struct btree *btree)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = btree->sb;
	struct buffer_head *rootbuf = new_node(btree);
	if (IS_ERR(rootbuf))
		goto error;
	struct buffer_head *leafbuf = new_leaf(btree);
	if (IS_ERR(leafbuf))
		goto error_leafbuf;

	assert(!has_root(btree));
	struct bnode *rootnode = bufdata(rootbuf);
	block_t rootblock = bufindex(rootbuf);
	block_t leafblock = bufindex(leafbuf);
	trace("root at %Lx", rootblock);
	trace("leaf at %Lx", leafblock);
	bnode_init_root(rootnode, 1, leafblock, 0, 0);
	log_bnode_root(sb, rootblock, 1, leafblock, 0, 0);
	log_balloc(sb, leafblock, 1);

	mark_buffer_unify_non(rootbuf);
	blockput(rootbuf);
	mark_buffer_dirty_non(leafbuf);
	blockput(leafbuf);

	btree->root = (struct root){ .block = rootblock, .depth = 1 };
	tux3_mark_btree_dirty(btree);

	return 0;

error_leafbuf:
	(btree->ops->bfree)(sb, bufindex(rootbuf), 1);
	blockput(rootbuf);
	rootbuf = leafbuf;
error:
	return PTR_ERR(rootbuf);
}

/* FIXME: right? and this should be done by btree_chop()? */
int free_empty_btree(struct btree *btree)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct btree_ops *ops = btree->ops;

	if (!has_root(btree))
		return 0;

	assert(btree->root.depth == 1);
	struct sb *sb = btree->sb;
	struct buffer_head *rootbuf = vol_bread(sb, btree->root.block);
	if (!rootbuf)
		return -EIO;
	assert(bnode_sniff(bufdata(rootbuf)));
	/* Make btree has no root */
	btree->root = no_root;
	tux3_mark_btree_dirty(btree);

	struct bnode *rootnode = bufdata(rootbuf);
	assert(bcount(rootnode) == 1);
	block_t leaf = be64_to_cpu(rootnode->entries[0].block);
	struct buffer_head *leafbuf = vol_find_get_block(sb, leaf);

	if (leafbuf && !leaf_need_redirect(sb, leafbuf)) {
		/*
		 * This is redirected leaf. So, in here, we can just
		 * cancel leaf_redirect by bfree(), instead of
		 * defered_bfree().
		 */
		bfree(sb, leaf, 1);
		log_leaf_free(sb, leaf);
		assert(ops->leaf_can_free(btree, bufdata(leafbuf)));
		blockput_free(sb, leafbuf);
	} else {
		defer_bfree(&sb->defree, leaf, 1);
		log_bfree(sb, leaf, 1);
		if (leafbuf) {
			assert(ops->leaf_can_free(btree, bufdata(leafbuf)));
			blockput(leafbuf);
		}
	}

	if (!bnode_need_redirect(sb, rootbuf)) {
		/*
		 * This is redirected bnode. So, in here, we can just
		 * cancel bnode_redirect by bfree(), instead of
		 * defered_bfree().
		 */
		bfree(sb, bufindex(rootbuf), 1);
		log_bnode_free(sb, bufindex(rootbuf));
		blockput_free_unify(sb, rootbuf);
	} else {
		defer_bfree(&sb->deunify, bufindex(rootbuf), 1);
		log_bfree_on_unify(sb, bufindex(rootbuf), 1);
		blockput(rootbuf);
	}

	return 0;
}

int replay_bnode_redirect(struct replay *rp, block_t oldblock, block_t newblock)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = rp->sb;
	struct buffer_head *newbuf, *oldbuf;
	int err = 0;

	newbuf = vol_getblk(sb, newblock);
	if (!newbuf) {
		err = -ENOMEM;	/* FIXME: error code */
		goto error;
	}
	oldbuf = vol_bread(sb, oldblock);
	if (!oldbuf) {
		err = -EIO;	/* FIXME: error code */
		goto error_put_newbuf;
	}
	assert(bnode_sniff(bufdata(oldbuf)));

	memcpy(bufdata(newbuf), bufdata(oldbuf), bufsize(newbuf));
	mark_buffer_unify_atomic(newbuf);

	blockput(oldbuf);
error_put_newbuf:
	blockput(newbuf);
error:
	return err;
}

int replay_bnode_root(struct replay *rp, block_t root, unsigned count,
		      block_t left, block_t right, tuxkey_t rkey)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = rp->sb;
	struct buffer_head *rootbuf;

	rootbuf = vol_getblk(sb, root);
	if (!rootbuf)
		return -ENOMEM;
	bnode_buffer_init(rootbuf);

	bnode_init_root(bufdata(rootbuf), count, left, right, rkey);

	mark_buffer_unify_atomic(rootbuf);
	blockput(rootbuf);

	return 0;
}

/*
 * Before this replay, replay should already dirty the buffer of src.
 * (e.g. by redirect)
 */
int replay_bnode_split(struct replay *rp, block_t src, unsigned pos,
		       block_t dst)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = rp->sb;
	struct buffer_head *srcbuf, *dstbuf;
	int err = 0;

	srcbuf = vol_getblk(sb, src);
	if (!srcbuf) {
		err = -ENOMEM;	/* FIXME: error code */
		goto error;
	}

	dstbuf = vol_getblk(sb, dst);
	if (!dstbuf) {
		err = -ENOMEM;	/* FIXME: error code */
		goto error_put_srcbuf;
	}
	bnode_buffer_init(dstbuf);

	bnode_split(bufdata(srcbuf), pos, bufdata(dstbuf));

	mark_buffer_unify_non(srcbuf);
	mark_buffer_unify_atomic(dstbuf);

	blockput(dstbuf);
error_put_srcbuf:
	blockput(srcbuf);
error:
	return err;
}

/*
 * Before this replay, replay should already dirty the buffer of bnodeblock.
 * (e.g. by redirect)
 */
static int replay_bnode_change(struct sb *sb, block_t bnodeblock,
			       u64 val1, u64 val2,
			       void (*change)(struct bnode *, u64, u64))
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct buffer_head *bnodebuf;

	bnodebuf = vol_getblk(sb, bnodeblock);
	if (!bnodebuf)
		return -ENOMEM;	/* FIXME: error code */

	struct bnode *bnode = bufdata(bnodebuf);
	change(bnode, val1, val2);

	mark_buffer_unify_non(bnodebuf);
	blockput(bnodebuf);

	return 0;
}

static void add_func(struct bnode *bnode, u64 child, u64 key)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct index_entry *entry = bnode_lookup(bnode, key) + 1;
	bnode_add_index(bnode, entry, child, key);
}

int replay_bnode_add(struct replay *rp, block_t parent, block_t child,
		     tuxkey_t key)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	return replay_bnode_change(rp->sb, parent, child, key, add_func);
}

static void update_func(struct bnode *bnode, u64 child, u64 key)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct index_entry *entry = bnode_lookup(bnode, key);
	assert(be64_to_cpu(entry->key) == key);
	entry->block = cpu_to_be64(child);
}

int replay_bnode_update(struct replay *rp, block_t parent, block_t child,
			tuxkey_t key)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	return replay_bnode_change(rp->sb, parent, child, key, update_func);
}

int replay_bnode_merge(struct replay *rp, block_t src, block_t dst)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct sb *sb = rp->sb;
	struct buffer_head *srcbuf, *dstbuf;
	int err = 0, ret;

	srcbuf = vol_getblk(sb, src);
	if (!srcbuf) {
		err = -ENOMEM;	/* FIXME: error code */
		goto error;
	}

	dstbuf = vol_getblk(sb, dst);
	if (!dstbuf) {
		err = -ENOMEM;	/* FIXME: error code */
		goto error_put_srcbuf;
	}

	ret = bnode_merge_nodes(sb, bufdata(dstbuf), bufdata(srcbuf));
	assert(ret == 1);

	mark_buffer_unify_non(dstbuf);
	mark_buffer_unify_non(srcbuf);

	blockput(dstbuf);
error_put_srcbuf:
	blockput(srcbuf);
error:
	return err;
}

static void del_func(struct bnode *bnode, u64 key, u64 count)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct index_entry *entry = bnode_lookup(bnode, key);
	assert(be64_to_cpu(entry->key) == key);
	bnode_remove_index(bnode, entry, count);
}

int replay_bnode_del(struct replay *rp, block_t bnode, tuxkey_t key,
		     unsigned count)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	return replay_bnode_change(rp->sb, bnode, key, count, del_func);
}

static void adjust_func(struct bnode *bnode, u64 from, u64 to)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	struct index_entry *entry = bnode_lookup(bnode, from);
	assert(be64_to_cpu(entry->key) == from);
	entry->key = cpu_to_be64(to);
}

int replay_bnode_adjust(struct replay *rp, block_t bnode, tuxkey_t from,
			tuxkey_t to)
{
	if(DEBUG_MODE_K==1)
	{
		printf("\t\t\t\t%25s[K]  %25s  %4d  #in\n",__FILE__,__func__,__LINE__);
	}
	return replay_bnode_change(rp->sb, bnode, from, to, adjust_func);
}