alloc.c

	if (le16_to_cpu(el->l_tree_depth) == 0) {
		ocfs2_insert_at_leaf(insert_rec, el, type, inode);
		goto out_update_clusters;
	}

	right_path = ocfs2_new_inode_path(di_bh);
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	/*
	 * Determine the path to start with. Rotations need the
	 * rightmost path, everything else can go directly to the
	 * target leaf.
	 */
	cpos = le32_to_cpu(insert_rec->e_cpos);
	if (type->ins_appending == APPEND_NONE &&
	    type->ins_contig == CONTIG_NONE) {
		rotate = 1;
		cpos = UINT_MAX;
	}

	ret = ocfs2_find_path(inode, right_path, cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	/*
	 * Rotations and appends need special treatment - they modify
	 * parts of the tree's above them.
	 *
	 * Both might pass back a path immediate to the left of the
	 * one being inserted to. This will be cause
	 * ocfs2_insert_path() to modify the rightmost records of
	 * left_path to account for an edge insert.
	 *
	 * XXX: When modifying this code, keep in mind that an insert
	 * can wind up skipping both of these two special cases...
	 */
	if (rotate) {
		ret = ocfs2_rotate_tree_right(inode, handle,
					      le32_to_cpu(insert_rec->e_cpos),
					      right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	} else if (type->ins_appending == APPEND_TAIL
		   && type->ins_contig != CONTIG_LEFT) {
		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
					       right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	ret = ocfs2_insert_path(inode, handle, left_path, right_path,
				insert_rec, type);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out_update_clusters:
	ocfs2_update_dinode_clusters(inode, di,
				     le32_to_cpu(insert_rec->e_clusters));

	ret = ocfs2_journal_dirty(handle, di_bh);
	if (ret)
		mlog_errno(ret);

out:
	ocfs2_free_path(left_path);
	ocfs2_free_path(right_path);

	return ret;
}

static void ocfs2_figure_contig_type(struct inode *inode,
				     struct ocfs2_insert_type *insert,
				     struct ocfs2_extent_list *el,
				     struct ocfs2_extent_rec *insert_rec)
{
	int i;
	enum ocfs2_contig_type contig_type = CONTIG_NONE;

	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
		contig_type = ocfs2_extent_contig(inode, &el->l_recs[i],
						  insert_rec);
		if (contig_type != CONTIG_NONE) {
			insert->ins_contig_index = i;
			break;
		}
	}
	insert->ins_contig = contig_type;
}

/*
 * This should only be called against the righmost leaf extent list.
 *
 * ocfs2_figure_appending_type() will figure out whether we'll have to
 * insert at the tail of the rightmost leaf.
 *
 * This should also work against the dinode list for tree's with 0
 * depth. If we consider the dinode list to be the rightmost leaf node
 * then the logic here makes sense.
 */
static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
					struct ocfs2_extent_list *el,
					struct ocfs2_extent_rec *insert_rec)
{
	int i;
	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
	struct ocfs2_extent_rec *rec;

	insert->ins_appending = APPEND_NONE;

	BUG_ON(el->l_tree_depth);

	if (!el->l_next_free_rec)
		goto set_tail_append;

	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
		/* Were all records empty? */
		if (le16_to_cpu(el->l_next_free_rec) == 1)
			goto set_tail_append;
	}

	i = le16_to_cpu(el->l_next_free_rec) - 1;
	rec = &el->l_recs[i];

	if (cpos >= (le32_to_cpu(rec->e_cpos) + le32_to_cpu(rec->e_clusters)))
		goto set_tail_append;

	return;

set_tail_append:
	insert->ins_appending = APPEND_TAIL;
}

/*
 * Helper function called at the begining of an insert.
 *
 * This computes a few things that are commonly used in the process of
 * inserting into the btree:
 *   - Whether the new extent is contiguous with an existing one.
 *   - The current tree depth.
 *   - Whether the insert is an appending one.
 *   - The total # of free records in the tree.
 *
 * All of the information is stored on the ocfs2_insert_type
 * structure.
 */
static int ocfs2_figure_insert_type(struct inode *inode,
				    struct buffer_head *di_bh,
				    struct buffer_head **last_eb_bh,
				    struct ocfs2_extent_rec *insert_rec,
				    struct ocfs2_insert_type *insert)
{
	int ret;
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct ocfs2_path *path = NULL;
	struct buffer_head *bh = NULL;

	el = &di->id2.i_list;
	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);

	if (el->l_tree_depth) {
		/*
		 * If we have tree depth, we read in the
		 * rightmost extent block ahead of time as
		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
		 * may want it later.
		 */
		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
				       le64_to_cpu(di->i_last_eb_blk), &bh,
				       OCFS2_BH_CACHED, inode);
		if (ret) {
			mlog_exit(ret);
			goto out;
		}
		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
	}

	/*
	 * Unless we have a contiguous insert, we'll need to know if
	 * there is room left in our allocation tree for another
	 * extent record.
	 *
	 * XXX: This test is simplistic, we can search for empty
	 * extent records too.
	 */
	insert->ins_free_records = le16_to_cpu(el->l_count) -
		le16_to_cpu(el->l_next_free_rec);

	if (!insert->ins_tree_depth) {
		ocfs2_figure_contig_type(inode, insert, el, insert_rec);
		ocfs2_figure_appending_type(insert, el, insert_rec);
		return 0;
	}

	path = ocfs2_new_inode_path(di_bh);
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	/*
	 * In the case that we're inserting past what the tree
	 * currently accounts for, ocfs2_find_path() will return for
	 * us the rightmost tree path. This is accounted for below in
	 * the appending code.
	 */
	ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos));
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	el = path_leaf_el(path);

	/*
	 * Now that we have the path, there's two things we want to determine:
	 * 1) Contiguousness (also set contig_index if this is so)
	 *
	 * 2) Are we doing an append? We can trivially break this up
         *     into two types of appends: simple record append, or a
         *     rotate inside the tail leaf.
	 */
	ocfs2_figure_contig_type(inode, insert, el, insert_rec);

	/*
	 * The insert code isn't quite ready to deal with all cases of
	 * left contiguousness. Specifically, if it's an insert into
	 * the 1st record in a leaf, it will require the adjustment of
	 * e_clusters on the last record of the path directly to it's
	 * left. For now, just catch that case and fool the layers
	 * above us. This works just fine for tree_depth == 0, which
	 * is why we allow that above.
	 */
	if (insert->ins_contig == CONTIG_LEFT &&
	    insert->ins_contig_index == 0)
		insert->ins_contig = CONTIG_NONE;

	/*
	 * Ok, so we can simply compare against last_eb to figure out
	 * whether the path doesn't exist. This will only happen in
	 * the case that we're doing a tail append, so maybe we can
	 * take advantage of that information somehow.
	 */
	if (le64_to_cpu(di->i_last_eb_blk) == path_leaf_bh(path)->b_blocknr) {
		/*
		 * Ok, ocfs2_find_path() returned us the rightmost
		 * tree path. This might be an appending insert. There are
		 * two cases:
		 *    1) We're doing a true append at the tail:
		 *	-This might even be off the end of the leaf
		 *    2) We're "appending" by rotating in the tail
		 */
		ocfs2_figure_appending_type(insert, el, insert_rec);
	}

out:
	ocfs2_free_path(path);

	if (ret == 0)
		*last_eb_bh = bh;
	else
		brelse(bh);
	return ret;
}

/*
 * Insert an extent into an inode btree.
 *
 * The caller needs to update fe->i_clusters
 */
int ocfs2_insert_extent(struct ocfs2_super *osb,
			handle_t *handle,
			struct inode *inode,
			struct buffer_head *fe_bh,
			u32 cpos,
			u64 start_blk,
			u32 new_clusters,
			struct ocfs2_alloc_context *meta_ac)
{
	int status, shift;
	struct buffer_head *last_eb_bh = NULL;
	struct buffer_head *bh = NULL;
	struct ocfs2_insert_type insert = {0, };
	struct ocfs2_extent_rec rec;

	mlog(0, "add %u clusters at position %u to inode %llu\n",
	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);

	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
			(OCFS2_I(inode)->ip_clusters != cpos),
			"Device %s, asking for sparse allocation: inode %llu, "
			"cpos %u, clusters %u\n",
			osb->dev_str,
			(unsigned long long)OCFS2_I(inode)->ip_blkno, cpos,
			OCFS2_I(inode)->ip_clusters);

	rec.e_cpos = cpu_to_le32(cpos);
	rec.e_blkno = cpu_to_le64(start_blk);
	rec.e_clusters = cpu_to_le32(new_clusters);

	status = ocfs2_figure_insert_type(inode, fe_bh, &last_eb_bh, &rec,
					  &insert);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
	     "Insert.contig_index: %d, Insert.free_records: %d, "
	     "Insert.tree_depth: %d\n",
	     insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
	     insert.ins_free_records, insert.ins_tree_depth);

	/*
	 * Avoid growing the tree unless we're out of records and the
	 * insert type requres one.
	 */
	if (insert.ins_contig != CONTIG_NONE || insert.ins_free_records)
		goto out_add;

	shift = ocfs2_find_branch_target(osb, inode, fe_bh, &bh);
	if (shift < 0) {
		status = shift;
		mlog_errno(status);
		goto bail;
	}

	/* We traveled all the way to the bottom of the allocation tree
	 * and didn't find room for any more extents - we need to add
	 * another tree level */
	if (shift) {
		BUG_ON(bh);
		mlog(0, "need to shift tree depth "
		     "(current = %d)\n", insert.ins_tree_depth);

		/* ocfs2_shift_tree_depth will return us a buffer with
		 * the new extent block (so we can pass that to
		 * ocfs2_add_branch). */
		status = ocfs2_shift_tree_depth(osb, handle, inode, fe_bh,
						meta_ac, &bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		insert.ins_tree_depth++;
		/* Special case: we have room now if we shifted from
		 * tree_depth 0 */
		if (insert.ins_tree_depth == 1)
			goto out_add;
	}

	/* call ocfs2_add_branch to add the final part of the tree with
	 * the new data. */
	mlog(0, "add branch. bh = %p\n", bh);
	status = ocfs2_add_branch(osb, handle, inode, fe_bh, bh, last_eb_bh,
				  meta_ac);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

out_add:
	/* Finally, we can add clusters. This might rotate the tree for us. */
	status = ocfs2_do_insert_extent(inode, handle, fe_bh, &rec, &insert);
	if (status < 0)
		mlog_errno(status);

bail:
	if (bh)
		brelse(bh);

	if (last_eb_bh)
		brelse(last_eb_bh);

	mlog_exit(status);
	return status;
}

static inline int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
{
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;

	mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
			"slot %d, invalid truncate log parameters: used = "
			"%u, count = %u\n", osb->slot_num,
			le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
	return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
}

static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
					   unsigned int new_start)
{
	unsigned int tail_index;
	unsigned int current_tail;

	/* No records, nothing to coalesce */
	if (!le16_to_cpu(tl->tl_used))
		return 0;

	tail_index = le16_to_cpu(tl->tl_used) - 1;
	current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
	current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);

	return current_tail == new_start;
}

static int ocfs2_truncate_log_append(struct ocfs2_super *osb,
				     handle_t *handle,
				     u64 start_blk,
				     unsigned int num_clusters)
{
	int status, index;
	unsigned int start_cluster, tl_count;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	mlog_entry("start_blk = %llu, num_clusters = %u\n",
		   (unsigned long long)start_blk, num_clusters);

	BUG_ON(mutex_trylock(&tl_inode->i_mutex));

	start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
		status = -EIO;
		goto bail;
	}

	tl_count = le16_to_cpu(tl->tl_count);
	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
			tl_count == 0,
			"Truncate record count on #%llu invalid "
			"wanted %u, actual %u\n",
			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
			ocfs2_truncate_recs_per_inode(osb->sb),
			le16_to_cpu(tl->tl_count));

	/* Caller should have known to flush before calling us. */
	index = le16_to_cpu(tl->tl_used);
	if (index >= tl_count) {
		status = -ENOSPC;
		mlog_errno(status);
		goto bail;
	}

	status = ocfs2_journal_access(handle, tl_inode, tl_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Log truncate of %u clusters starting at cluster %u to "
	     "%llu (index = %d)\n", num_clusters, start_cluster,
	     (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index);

	if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
		/*
		 * Move index back to the record we are coalescing with.
		 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
		 */
		index--;

		num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
		mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n",
		     index, le32_to_cpu(tl->tl_recs[index].t_start),
		     num_clusters);
	} else {
		tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
		tl->tl_used = cpu_to_le16(index + 1);
	}
	tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);

	status = ocfs2_journal_dirty(handle, tl_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
					 handle_t *handle,
					 struct inode *data_alloc_inode,
					 struct buffer_head *data_alloc_bh)
{
	int status = 0;
	int i;
	unsigned int num_clusters;
	u64 start_blk;
	struct ocfs2_truncate_rec rec;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct buffer_head *tl_bh = osb->osb_tl_bh;

	mlog_entry_void();

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	i = le16_to_cpu(tl->tl_used) - 1;
	while (i >= 0) {
		/* Caller has given us at least enough credits to
		 * update the truncate log dinode */
		status = ocfs2_journal_access(handle, tl_inode, tl_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		tl->tl_used = cpu_to_le16(i);

		status = ocfs2_journal_dirty(handle, tl_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		/* TODO: Perhaps we can calculate the bulk of the
		 * credits up front rather than extending like
		 * this. */
		status = ocfs2_extend_trans(handle,
					    OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		rec = tl->tl_recs[i];
		start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
						    le32_to_cpu(rec.t_start));
		num_clusters = le32_to_cpu(rec.t_clusters);

		/* if start_blk is not set, we ignore the record as
		 * invalid. */
		if (start_blk) {
			mlog(0, "free record %d, start = %u, clusters = %u\n",
			     i, le32_to_cpu(rec.t_start), num_clusters);

			status = ocfs2_free_clusters(handle, data_alloc_inode,
						     data_alloc_bh, start_blk,
						     num_clusters);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}
		i--;
	}

bail:
	mlog_exit(status);
	return status;
}

/* Expects you to already be holding tl_inode->i_mutex */
static int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
{
	int status;
	unsigned int num_to_flush;
	handle_t *handle;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct inode *data_alloc_inode = NULL;
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct buffer_head *data_alloc_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

	BUG_ON(mutex_trylock(&tl_inode->i_mutex));

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
		status = -EIO;
		goto out;
	}

	num_to_flush = le16_to_cpu(tl->tl_used);
	mlog(0, "Flush %u records from truncate log #%llu\n",
	     num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno);
	if (!num_to_flush) {
		status = 0;
		goto out;
	}

	data_alloc_inode = ocfs2_get_system_file_inode(osb,
						       GLOBAL_BITMAP_SYSTEM_INODE,
						       OCFS2_INVALID_SLOT);
	if (!data_alloc_inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get bitmap inode!\n");
		goto out;
	}

	mutex_lock(&data_alloc_inode->i_mutex);

	status = ocfs2_meta_lock(data_alloc_inode, &data_alloc_bh, 1);
	if (status < 0) {
		mlog_errno(status);
		goto out_mutex;
	}

	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		mlog_errno(status);
		goto out_unlock;
	}

	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
					       data_alloc_bh);
	if (status < 0)
		mlog_errno(status);

	ocfs2_commit_trans(osb, handle);

out_unlock:
	brelse(data_alloc_bh);
	ocfs2_meta_unlock(data_alloc_inode, 1);

out_mutex:
	mutex_unlock(&data_alloc_inode->i_mutex);
	iput(data_alloc_inode);

out:
	mlog_exit(status);
	return status;
}

int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

	mutex_lock(&tl_inode->i_mutex);
	status = __ocfs2_flush_truncate_log(osb);
	mutex_unlock(&tl_inode->i_mutex);

	return status;
}

static void ocfs2_truncate_log_worker(struct work_struct *work)
{
	int status;
	struct ocfs2_super *osb =
		container_of(work, struct ocfs2_super,
			     osb_truncate_log_wq.work);

	mlog_entry_void();

	status = ocfs2_flush_truncate_log(osb);
	if (status < 0)
		mlog_errno(status);

	mlog_exit(status);
}

#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
				       int cancel)
{
	if (osb->osb_tl_inode) {
		/* We want to push off log flushes while truncates are
		 * still running. */
		if (cancel)
			cancel_delayed_work(&osb->osb_truncate_log_wq);

		queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
				   OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
	}
}

static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
				       int slot_num,
				       struct inode **tl_inode,
				       struct buffer_head **tl_bh)
{
	int status;
	struct inode *inode = NULL;
	struct buffer_head *bh = NULL;

	inode = ocfs2_get_system_file_inode(osb,
					   TRUNCATE_LOG_SYSTEM_INODE,
					   slot_num);
	if (!inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get load truncate log inode!\n");
		goto bail;
	}

	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
				  OCFS2_BH_CACHED, inode);
	if (status < 0) {
		iput(inode);
		mlog_errno(status);
		goto bail;
	}

	*tl_inode = inode;
	*tl_bh    = bh;
bail:
	mlog_exit(status);
	return status;
}

/* called during the 1st stage of node recovery. we stamp a clean
 * truncate log and pass back a copy for processing later. if the
 * truncate log does not require processing, a *tl_copy is set to
 * NULL. */
int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
				      int slot_num,
				      struct ocfs2_dinode **tl_copy)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	*tl_copy = NULL;

	mlog(0, "recover truncate log from slot %d\n", slot_num);

	status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di);
		status = -EIO;
		goto bail;
	}

	if (le16_to_cpu(tl->tl_used)) {
		mlog(0, "We'll have %u logs to recover\n",
		     le16_to_cpu(tl->tl_used));

		*tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
		if (!(*tl_copy)) {
			status = -ENOMEM;
			mlog_errno(status);
			goto bail;
		}

		/* Assuming the write-out below goes well, this copy
		 * will be passed back to recovery for processing. */
		memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);

		/* All we need to do to clear the truncate log is set
		 * tl_used. */
		tl->tl_used = 0;

		status = ocfs2_write_block(osb, tl_bh, tl_inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (tl_inode)
		iput(tl_inode);
	if (tl_bh)
		brelse(tl_bh);

	if (status < 0 && (*tl_copy)) {
		kfree(*tl_copy);
		*tl_copy = NULL;
	}

	mlog_exit(status);
	return status;
}

int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
					 struct ocfs2_dinode *tl_copy)
{
	int status = 0;
	int i;
	unsigned int clusters, num_recs, start_cluster;
	u64 start_blk;
	handle_t *handle;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

	if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
		mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
		return -EINVAL;
	}

	tl = &tl_copy->id2.i_dealloc;
	num_recs = le16_to_cpu(tl->tl_used);
	mlog(0, "cleanup %u records from %llu\n", num_recs,
	     (unsigned long long)tl_copy->i_blkno);

	mutex_lock(&tl_inode->i_mutex);
	for(i = 0; i < num_recs; i++) {
		if (ocfs2_truncate_log_needs_flush(osb)) {
			status = __ocfs2_flush_truncate_log(osb);
			if (status < 0) {
				mlog_errno(status);
				goto bail_up;
			}
		}

		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
		if (IS_ERR(handle)) {
			status = PTR_ERR(handle);
			mlog_errno(status);
			goto bail_up;
		}

		clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
		start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
		start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);

		status = ocfs2_truncate_log_append(osb, handle,
						   start_blk, clusters);
		ocfs2_commit_trans(osb, handle);
		if (status < 0) {
			mlog_errno(status);
			goto bail_up;
		}
	}

bail_up:
	mutex_unlock(&tl_inode->i_mutex);

	mlog_exit(status);
	return status;
}

void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

	mlog_entry_void();

	if (tl_inode) {
		cancel_delayed_work(&osb->osb_truncate_log_wq);
		flush_workqueue(ocfs2_wq);

		status = ocfs2_flush_truncate_log(osb);
		if (status < 0)
			mlog_errno(status);

		brelse(osb->osb_tl_bh);
		iput(osb->osb_tl_inode);
	}

	mlog_exit_void();
}

int ocfs2_truncate_log_init(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;

	mlog_entry_void();

	status = ocfs2_get_truncate_log_info(osb,
					     osb->slot_num,
					     &tl_inode,
					     &tl_bh);
	if (status < 0)
		mlog_errno(status);

	/* ocfs2_truncate_log_shutdown keys on the existence of
	 * osb->osb_tl_inode so we don't set any of the osb variables
	 * until we're sure all is well. */
	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
			  ocfs2_truncate_log_worker);
	osb->osb_tl_bh    = tl_bh;
	osb->osb_tl_inode = tl_inode;

	mlog_exit(status);
	return status;
}

/* This function will figure out whether the currently last extent
 * block will be deleted, and if it will, what the new last extent
 * block will be so we can update his h_next_leaf_blk field, as well
 * as the dinodes i_last_eb_blk */
static int ocfs2_find_new_last_ext_blk(struct inode *inode,
				       unsigned int clusters_to_del,
				       struct ocfs2_path *path,
				       struct buffer_head **new_last_eb)
{
	int next_free, ret = 0;
	u32 cpos;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *bh = NULL;

	*new_last_eb = NULL;

	/* we have no tree, so of course, no last_eb. */
	if (!path->p_tree_depth)
		goto out;

	/* trunc to zero special case - this makes tree_depth = 0
	 * regardless of what it is.  */
	if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
		goto out;

	el = path_leaf_el(path);
	BUG_ON(!el->l_next_free_rec);

	/*
	 * Make sure that this extent list will actually be empty
	 * after we clear away the data. We can shortcut out if
	 * there's more than one non-empty extent in the
	 * list. Otherwise, a check of the remaining extent is
	 * necessary.
	 */
	next_free = le16_to_cpu(el->l_next_free_rec);
	rec = NULL;
	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
		if (next_free > 2)
			goto out;

		/* We may have a valid extent in index 1, check it. */
		if (next_free == 2)
			rec = &el->l_recs[1];

		/*
		 * Fall through - no more nonempty extents, so we want
		 * to delete this leaf.
		 */
	} else {
		if (next_free > 1)
			goto out;

		rec = &el->l_recs[0];
	}

	if (rec) {
		/*
		 * Check it we'll only be trimming off the end of this
		 * cluster.
		 */
		if (le16_to_cpu(rec->e_clusters) > clusters_to_del)
			goto out;
	}

	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	eb = (struct ocfs2_extent_block *) bh->b_data;