Relevant patches backported from mainline

index 4791433..1a6455e 100644
@@ -3946,7 +3946,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 	struct ocfs2_merge_ctxt ctxt;
 	struct ocfs2_extent_list *rightmost_el;
 
-	if (!rec->e_flags & OCFS2_EXT_UNWRITTEN) {
+	if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) {
 		ret = -EIO;
 		mlog_errno(ret);
 		goto out;

index 41c76ff..ef09fd2 100644
@@ -670,7 +670,7 @@ static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *loc
 {
 	mlog_entry_void();
 
-	BUG_ON((!lockres->l_flags & OCFS2_LOCK_BUSY));
+	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
 	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
 
 	if (lockres->l_requested > LKM_NLMODE &&

index ef09fd2..4e97dcc 100644
@@ -980,18 +980,6 @@ again:
 		goto unlock;
 	}
 
-	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
-		/* lock has not been created yet. */
-		spin_unlock_irqrestore(&lockres->l_lock, flags);
-
-		ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0);
-		if (ret < 0) {
-			mlog_errno(ret);
-			goto out;
-		}
-		goto again;
-	}
-
 	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
 	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
 		/* is the lock is currently blocked on behalf of

@@ -1006,7 +994,14 @@ again:
 			mlog(ML_ERROR, "lockres %s has action %u pending\n",
 			     lockres->l_name, lockres->l_action);
 
-		lockres->l_action = OCFS2_AST_CONVERT;
+		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
+			lockres->l_action = OCFS2_AST_ATTACH;
+			lkm_flags &= ~LKM_CONVERT;
+		} else {
+			lockres->l_action = OCFS2_AST_CONVERT;
+			lkm_flags |= LKM_CONVERT;
+		}
+
 		lockres->l_requested = level;
 		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
 		spin_unlock_irqrestore(&lockres->l_lock, flags);

@@ -1021,7 +1016,7 @@ again:
 		status = dlmlock(osb->dlm,
 				 level,
 				 &lockres->l_lksb,
-				 lkm_flags|LKM_CONVERT,
+				 lkm_flags,
 				 lockres->l_name,
 				 OCFS2_LOCK_ID_MAX_LEN - 1,
 				 ocfs2_locking_ast,

index 014e739..9e76e3c 100644
@@ -318,9 +318,9 @@ out_attach:
 static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
 				   struct ocfs2_dentry_lock *dl)
 {
+	iput(dl->dl_inode);
 	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 	ocfs2_lock_res_free(&dl->dl_lockres);
-	iput(dl->dl_inode);
 	kfree(dl);
 }
 

index ad73b3c..a6a93b1 100644
@@ -2027,9 +2027,11 @@ static ssize_t __ocfs2_file_aio_write(struct kiocb *iocb,
 	ssize_t written = 0;
 	size_t ocount;		/* original count */
 	size_t count;		/* after file limit checks */
-	loff_t *ppos = &iocb->ki_pos;
+	loff_t old_size, *ppos = &iocb->ki_pos;
+	u32 old_clusters;
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = filp_dentry(file)->d_inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 
 	mlog_entry("(0x%p, %u, '%.*s')\n", file,
 		   (unsigned int)nr_segs,

@@ -2111,6 +2113,13 @@ relock:
 	 */
 	pos = *ppos;
 
+	/*
+	 * To later detect whether a journal commit for sync writes is
+	 * necessary, we sample i_size, and cluster count here.
+	 */
+	old_size = i_size_read(inode);
+	old_clusters = OCFS2_I(inode)->ip_clusters;
+
 	/* communicate with ocfs2_dio_end_io */
 	ocfs2_iocb_set_rw_locked(iocb, rw_level);
 

@@ -2136,6 +2145,21 @@ out_dio:
 	/* buffered aio wouldn't have proper lock coverage today */
 	BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
 
+	if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
+		/*
+		 * The generic write paths have handled getting data
+		 * to disk, but since we don't make use of the dirty
+		 * inode list, a manual journal commit is necessary
+		 * here.
+		 */
+		if (old_size != i_size_read(inode) ||
+		    old_clusters != OCFS2_I(inode)->ip_clusters) {
+			ret = journal_force_commit(osb->journal->j_journal);
+			if (ret < 0)
+				written = ret;
+		}
+	}
+
 	/* 
 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
 	 * function pointer which is called when o_direct io completes so that

index 93c01b7..4999180 100644
@@ -729,6 +729,27 @@ static void ocfs2_clear_page_regions(struct page *page,
 }
 
 /*
+ * Nonsparse file systems fully allocate before we get to the write
+ * code. This prevents ocfs2_write() from tagging the write as an
+ * allocating one, which means ocfs2_map_page_blocks() might try to
+ * read-in the blocks at the tail of our file. Avoid reading them by
+ * testing i_size against each block offset.
+ */
+static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
+				 unsigned int block_start)
+{
+	u64 offset = page_offset(page) + block_start;
+
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 1;
+
+	if (i_size_read(inode) > offset)
+		return 1;
+
+	return 0;
+}
+
+/*
  * Some of this taken from block_prepare_write(). We already have our
  * mapping by now though, and the entire write will be allocating or
  * it won't, so not much need to use BH_New.

@@ -781,6 +802,7 @@ int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
 				set_buffer_uptodate(bh);
 		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
 			   !buffer_new(bh) &&
+			   ocfs2_should_read_blk(inode, page, block_start) &&
 			   (block_start < from || block_end > to)) {
 			ll_rw_block(READ, 1, &bh);
 			*wait_bh++=bh;

index 6a91607..999e7e5 100644
@@ -275,7 +275,7 @@ static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
 		current_page = cs / spp;
 		page = reg->hr_slot_data[current_page];
 
-		vec_len = min(PAGE_CACHE_SIZE,
+		vec_len = min(PAGE_CACHE_SIZE - vec_start,
 			      (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
 
 		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",

index 6d822b7..f9533f7 100644
@@ -1105,9 +1105,16 @@ static int ocfs2_rename(struct inode *old_dir,
 		goto bail;
 	}
 
-	if (!new_de && new_inode)
-		mlog(ML_ERROR, "inode %lu does not exist in it's parent "
-		     "directory!", new_inode->i_ino);
+	if (!new_de && new_inode) {
+		/*
+		 * Target was unlinked by another node while we were
+		 * waiting to get to ocfs2_rename(). There isn't
+		 * anything we can do here to help the situation, so
+		 * bubble up the appropriate error.
+		 */
+		status = -ENOENT;
+		goto bail;
+	}
 
 	/* In case we need to overwrite an existing file, we blow it
 	 * away first */

index 43147c0..39cc000 100644
@@ -442,14 +442,14 @@ unlock_osb:
 	}
 
 	if (!ret) {
-		if (!ocfs2_is_hard_readonly(osb))
-			ocfs2_set_journal_params(osb);
-
 		/* Only save off the new mount options in case of a successful
 		 * remount. */
 		osb->s_mount_opt = parsed_options.mount_opt;
 		osb->s_atime_quantum = parsed_options.atime_quantum;
 		osb->preferred_slot = parsed_options.slot;
+
+		if (!ocfs2_is_hard_readonly(osb))
+			ocfs2_set_journal_params(osb);
 	}
 out:
 	return ret;

index c8e5d84..42ea5af 100644
@@ -907,7 +907,7 @@ lookup:
 		 * but they might own this lockres.  wait on them. */
 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 		if (bit < O2NM_MAX_NODES) {
-			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+			mlog(ML_NOTICE, "%s:%.*s:@least one node (%d) to "
 			     "recover before lock mastery can begin\n",
 			     dlm->name, namelen, (char *)lockid, bit);
 			wait_on_recovery = 1;

@@ -961,7 +961,7 @@ redo_request:
 		spin_lock(&dlm->spinlock);
 		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
 		if (bit < O2NM_MAX_NODES) {
-			mlog(ML_NOTICE, "%s:%.*s: at least one node (%d) to"
+			mlog(ML_NOTICE, "%s:%.*s:@least one node (%d) to "
 			     "recover before lock mastery can begin\n",
 			     dlm->name, namelen, (char *)lockid, bit);
 			wait_on_recovery = 1;

index 1d5e0cb..9e3e7df 100644
@@ -863,7 +863,7 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
 	status = ocfs2_try_open_lock(inode, 1);
 	if (status == -EAGAIN) {
 		status = 0;
-		mlog(0, "Skipping delete of %llu because it is in use on"
+		mlog(0, "Skipping delete of %llu because it is in use on "
 		     "other nodes\n", (unsigned long long)oi->ip_blkno);
 		goto bail;
 	}

index 75cd877..4c32d1c 100644
@@ -212,7 +212,7 @@ extern struct mlog_bits mlog_and_bits, mlog_not_bits;
 #define mlog_errno(st) do {						\
 	int _st = (st);							\
 	if (_st != -ERESTARTSYS && _st != -EINTR &&			\
-	    _st != AOP_TRUNCATED_PAGE)					\
+	    _st != AOP_TRUNCATED_PAGE && _st != -ENOSPC)		\
 		mlog(ML_ERROR, "status = %lld\n", (long long)_st);	\
 } while (0)
 

index 9e3e7df..ebb2bbe 100644
@@ -455,8 +455,8 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 	status = -EINVAL;
 	fe = (struct ocfs2_dinode *) bh->b_data;
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
-		mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
-		     (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
+		mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
+		     (unsigned long long)args->fi_blkno, 7,
 		     fe->i_signature);
 		goto bail;
 	}

index d272847..e29328a 100644
@@ -484,17 +484,6 @@ int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
 
 	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
 
-	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
-	    ocfs2_local_alloc_count_bits(alloc)) {
-		ocfs2_error(osb->sb, "local alloc inode %llu says it has "
-			    "%u free bits, but a count shows %u",
-			    (unsigned long long)le64_to_cpu(alloc->i_blkno),
-			    le32_to_cpu(alloc->id1.bitmap1.i_used),
-			    ocfs2_local_alloc_count_bits(alloc));
-		status = -EIO;
-		goto bail;
-	}
-
 	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
 		le32_to_cpu(alloc->id1.bitmap1.i_used);
 	if (bits_wanted > free_bits) {

@@ -712,9 +701,8 @@ static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
 	void *bitmap;
 	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
 
-	mlog_entry("total = %u, COUNT = %u, used = %u\n",
+	mlog_entry("total = %u, used = %u\n",
 		   le32_to_cpu(alloc->id1.bitmap1.i_total),
-		   ocfs2_local_alloc_count_bits(alloc),
 		   le32_to_cpu(alloc->id1.bitmap1.i_used));
 
 	if (!alloc->id1.bitmap1.i_total) {

index 9e76e3c..6ec9556 100644
@@ -344,12 +344,24 @@ static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
 {
 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 
-	mlog_bug_on_msg(!dl && !(dentry->d_flags & DCACHE_DISCONNECTED),
-			"dentry: %.*s\n", dentry->d_name.len,
-			dentry->d_name.name);
+	if (!dl) {
+		/*
+		 * No dentry lock is ok if we're disconnected or
+		 * unhashed.
+		 */
+		if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
+		    !d_unhashed(dentry)) {
+			unsigned long long ino = 0ULL;
+			if (inode)
+				ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
+			mlog(ML_ERROR, "Dentry is missing cluster lock. "
+			     "inode: %llu, d_flags: 0x%x, d_name: %.*s\n",
+			     ino, dentry->d_flags, dentry->d_name.len,
+			     dentry->d_name.name);
+		}
 
-	if (!dl)
 		goto out;
+	}
 
 	mlog_bug_on_msg(dl->dl_count == 0, "dentry: %.*s, count: %u\n",
 			dentry->d_name.len, dentry->d_name.name,

index 4999180..25338ab 100644
@@ -1514,7 +1514,7 @@ int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
 {
 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
 
-	if (new_size < le16_to_cpu(di->id2.i_data.id_count))
+	if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
 		return 1;
 	return 0;
 }

index a6a93b1..6d1ef46 100644
@@ -403,7 +403,7 @@ static int ocfs2_truncate_file(struct inode *inode,
 
 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
-					       i_size_read(inode), 0);
+					       i_size_read(inode), 1);
 		if (status)
 			mlog_errno(status);
 

@@ -1525,6 +1525,7 @@ static int ocfs2_remove_inode_range(struct inode *inode,
 	u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct address_space *mapping = inode->i_mapping;
 
 	ocfs2_init_dealloc_ctxt(&dealloc);
 

@@ -1533,10 +1534,20 @@ static int ocfs2_remove_inode_range(struct inode *inode,
 
 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
-					    byte_start + byte_len, 1);
-		if (ret)
+					    byte_start + byte_len, 0);
+		if (ret) {
 			mlog_errno(ret);
-		return ret;
+			goto out;
+		}
+		/*
+		 * There's no need to get fancy with the page cache
+		 * truncate of an inline-data inode. We're talking
+		 * about less than a page here, which will be cached
+		 * in the dinode buffer anyway.
+		 */
+		unmap_mapping_range(mapping, 0, 0, 0);
+		truncate_inode_pages(mapping, 0);
+		goto out;
 	}
 
 	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);

index aafd0a3..909a344 100644
@@ -1277,11 +1277,12 @@ static int ocfs2_queue_orphans(struct ocfs2_super *osb,
 				   ocfs2_orphan_filldir);
 	if (status) {
 		mlog_errno(status);
-		goto out;
+		goto out_cluster;
 	}
 
 	*head = priv.head;
 
+out_cluster:
 	ocfs2_meta_unlock(orphan_dir_inode, 0);
 out:
 	mutex_unlock(&orphan_dir_inode->i_mutex);

index 1a6455e..5714618 100644
@@ -6092,8 +6092,6 @@ start:
 	mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n",
 	     clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr);
 
-	BUG_ON(clusters_to_del == 0);
-
 	mutex_lock(&tl_inode->i_mutex);
 	tl_sem = 1;
 	/* ocfs2_truncate_log_needs_flush guarantees us at least one

index 909a344..9988abc 100644
@@ -193,11 +193,15 @@ int ocfs2_extend_trans(handle_t *handle, int nblocks)
 
 	mlog(0, "Trying to extend transaction by %d blocks\n", nblocks);
 
+#ifdef OCFS2_DEBUG_FS
+	status = 1;
+#else
 	status = journal_extend(handle, nblocks);
 	if (status < 0) {
 		mlog_errno(status);
 		goto bail;
 	}
+#endif
 
 	if (status > 0) {
 		mlog(0, "journal_extend failed, trying journal_restart\n");

index 5714618..cdbb607 100644
@@ -2389,6 +2389,18 @@ static int __ocfs2_rotate_tree_left(struct inode *inode,
 			goto out;
 		}
 
+		/*
+		 * Caller might still want to make changes to the
+		 * tree root, so re-add it to the journal here.
+		 */
+		ret = ocfs2_journal_access(handle, inode,
+					   path_root_bh(left_path),
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
 		ret = ocfs2_rotate_subtree_left(inode, handle, left_path,
 						right_path, subtree_root,
 						dealloc, &deleted);

@@ -3289,16 +3301,6 @@ static int ocfs2_insert_path(struct inode *inode,
 	int ret, subtree_index;
 	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
 
-	/*
-	 * Pass both paths to the journal. The majority of inserts
-	 * will be touching all components anyway.
-	 */
-	ret = ocfs2_journal_access_path(inode, handle, right_path);
-	if (ret < 0) {
-		mlog_errno(ret);
-		goto out;
-	}
-
 	if (left_path) {
 		int credits = handle->h_buffer_credits;
 

@@ -3323,6 +3325,16 @@ static int ocfs2_insert_path(struct inode *inode,
 		}
 	}
 
+	/*
+	 * Pass both paths to the journal. The majority of inserts
+	 * will be touching all components anyway.
+	 */
+	ret = ocfs2_journal_access_path(inode, handle, right_path);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
 	if (insert->ins_split != SPLIT_NONE) {
 		/*
 		 * We could call ocfs2_insert_at_leaf() for some types

@@ -3331,6 +3343,17 @@ static int ocfs2_insert_path(struct inode *inode,
 		 */
 		ocfs2_split_record(inode, left_path, right_path,
 				   insert_rec, insert->ins_split);
+
+		/*
+		 * Split might have modified either leaf and we don't
+		 * have a guarantee that the later edge insert will
+		 * dirty this for us.
+		 */
+		if (left_path)
+			ret = ocfs2_journal_dirty(handle,
+						  path_leaf_bh(left_path));
+			if (ret)
+				mlog_errno(ret);
 	} else
 		ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path),
 				     insert, inode);

@@ -3430,6 +3453,17 @@ static int ocfs2_do_insert_extent(struct inode *inode,
 			mlog_errno(ret);
 			goto out;
 		}
+
+		/*
+		 * ocfs2_rotate_tree_right() might have extended the
+		 * transaction without re-journaling our tree root.
+		 */
+		ret = ocfs2_journal_access(handle, inode, di_bh,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+		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,

@@ -3941,7 +3975,7 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 {
 	int ret = 0;
 	struct ocfs2_extent_list *el = path_leaf_el(path);
-	struct buffer_head *eb_bh, *last_eb_bh = NULL;
+	struct buffer_head *last_eb_bh = NULL;
 	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
 	struct ocfs2_merge_ctxt ctxt;
 	struct ocfs2_extent_list *rightmost_el;

@@ -3960,14 +3994,6 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 		goto out;
 	}
 
-	eb_bh = path_leaf_bh(path);
-	ret = ocfs2_journal_access(handle, inode, eb_bh,
-				   OCFS2_JOURNAL_ACCESS_WRITE);
-	if (ret) {
-		mlog_errno(ret);
-		goto out;
-	}
-
 	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, el,
 							    split_index,
 							    split_rec);

@@ -4029,8 +4055,6 @@ static int __ocfs2_mark_extent_written(struct inode *inode,
 			mlog_errno(ret);
 	}
 
-	ocfs2_journal_dirty(handle, eb_bh);
-
 out:
 	brelse(last_eb_bh);
 	return ret;

index 9988abc..051ffd5 100644
@@ -174,6 +174,12 @@ int ocfs2_commit_trans(struct ocfs2_super *osb,
  * transaction. extend_trans will either extend the current handle by
  * nblocks, or commit it and start a new one with nblocks credits.
  *
+ * This might call journal_restart() which will commit dirty buffers
+ * and then restart the transaction. Before calling
+ * ocfs2_extend_trans(), any changed blocks should have been
+ * dirtied. After calling it, all blocks which need to be changed must
+ * go through another set of journal_access/journal_dirty calls.
+ *
  * WARNING: This will not release any semaphores or disk locks taken
  * during the transaction, so make sure they were taken *before*
  * start_trans or we'll have ordering deadlocks.

index e29328a..84ed678 100644
@@ -75,18 +75,11 @@ static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
 static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
 					  struct inode *local_alloc_inode);
 
-/*
- * Determine how large our local alloc window should be, in bits.
- *
- * These values (and the behavior in ocfs2_alloc_should_use_local) have
- * been chosen so that most allocations, including new block groups go
- * through local alloc.
- */
 static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb)
 {
-	BUG_ON(osb->s_clustersize_bits < 12);
+	BUG_ON(osb->s_clustersize_bits > 20);
 
-	return 2048 >> (osb->s_clustersize_bits - 12);
+	return osb->local_alloc_size << (20 - osb->s_clustersize_bits);
 }
 
 /*

@@ -96,18 +89,23 @@ static inline int ocfs2_local_alloc_window_bits(struct ocfs2_super *osb)
 int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
 {
 	int la_bits = ocfs2_local_alloc_window_bits(osb);
+	int ret = 0;
 
 	if (osb->local_alloc_state != OCFS2_LA_ENABLED)
-		return 0;
+		goto bail;
 
 	/* la_bits should be at least twice the size (in clusters) of
 	 * a new block group. We want to be sure block group
 	 * allocations go through the local alloc, so allow an
 	 * allocation to take up to half the bitmap. */
 	if (bits > (la_bits / 2))
-		return 0;
+		goto bail;
 
-	return 1;
+	ret = 1;
+bail:
+	mlog(0, "state=%d, bits=%llu, la_bits=%d, ret=%d\n",
+	     osb->local_alloc_state, (unsigned long long)bits, la_bits, ret);
+	return ret;
 }
 
 int ocfs2_load_local_alloc(struct ocfs2_super *osb)

@@ -121,6 +119,19 @@ int ocfs2_load_local_alloc(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
+	if (ocfs2_mount_local(osb))
+		goto bail;
+
+	if (osb->local_alloc_size == 0)
+		goto bail;
+
+	if (ocfs2_local_alloc_window_bits(osb) >= osb->bitmap_cpg) {
+		mlog(ML_NOTICE, "Requested local alloc window %d is larger "
+		     "than max possible %u. Using defaults.\n",
+		     ocfs2_local_alloc_window_bits(osb), (osb->bitmap_cpg - 1));
+		osb->local_alloc_size = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
+	}
+
 	/* read the alloc off disk */
 	inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
 					    osb->slot_num);

@@ -181,6 +192,9 @@ bail:
 	if (inode)
 		iput(inode);
 
+	mlog(0, "Local alloc window bits = %d\n",
+	     ocfs2_local_alloc_window_bits(osb));
+
 	mlog_exit(status);
 	return status;
 }

@@ -508,6 +522,9 @@ bail:
 		iput(local_alloc_inode);
 	}
 
+	mlog(0, "bits=%d, slot=%d, ret=%d\n", bits_wanted, osb->slot_num,
+	     status);
+
 	mlog_exit(status);
 	return status;
 }

index 60a23e1..df27b52 100644
@@ -232,6 +232,7 @@ struct ocfs2_super
 	atomic_t needs_checkpoint;
 	struct ocfs2_journal *journal;
 
+	int local_alloc_size;
 	enum ocfs2_local_alloc_state local_alloc_state;
 	struct buffer_head *local_alloc_bh;
 	u64 la_last_gd;

index 6ef8767..9687b43 100644
@@ -256,6 +256,14 @@ struct ocfs2_space_resv {
 /* Journal limits (in bytes) */
 #define OCFS2_MIN_JOURNAL_SIZE		(4 * 1024 * 1024)
 
+/*
+ * Default local alloc size (in megabytes)
+ *
+ * The value chosen should be such that most allocations, including new
+ * block groups, use local alloc.
+ */
+#define OCFS2_DEFAULT_LOCAL_ALLOC_SIZE	8
+
 struct ocfs2_system_inode_info {
 	char	*si_name;
 	int	si_iflags;

index 8f09f52..4add3ee 100644
@@ -1519,8 +1519,9 @@ int __ocfs2_claim_clusters(struct ocfs2_super *osb,
 		if (min_clusters > (osb->bitmap_cpg - 1)) {
 			/* The only paths asking for contiguousness
 			 * should know about this already. */
-			mlog(ML_ERROR, "minimum allocation requested exceeds "
-				       "group bitmap size!");
+			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
+			     "group bitmap size %u!\n", min_clusters,
+			     osb->bitmap_cpg);
 			status = -ENOSPC;
 			goto bail;
 		}

index 39cc000..dfb14d1 100644
@@ -87,6 +87,7 @@ struct mount_options
 	unsigned long	mount_opt;
 	unsigned int	atime_quantum;
 	signed short	slot;
+	unsigned int	localalloc_opt;
 };
 
 static int ocfs2_parse_options(struct super_block *sb, char *options,

@@ -154,6 +155,7 @@ enum {
 	Opt_data_writeback,
 	Opt_atime_quantum,
 	Opt_slot,
+	Opt_localalloc,
 	Opt_err,
 };
 

@@ -169,6 +171,7 @@ static match_table_t tokens = {
 	{Opt_data_writeback, "data=writeback"},
 	{Opt_atime_quantum, "atime_quantum=%u"},
 	{Opt_slot, "preferred_slot=%u"},
+	{Opt_localalloc, "localalloc=%d"},
 	{Opt_err, NULL}
 };
 

@@ -601,6 +604,7 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 	osb->s_mount_opt = parsed_options.mount_opt;
 	osb->s_atime_quantum = parsed_options.atime_quantum;
 	osb->preferred_slot = parsed_options.slot;
+	osb->local_alloc_size = parsed_options.localalloc_opt;
 
 	sb->s_magic = OCFS2_SUPER_MAGIC;
 

@@ -754,6 +758,7 @@ static int ocfs2_parse_options(struct super_block *sb,
 	mopt->mount_opt = 0;
 	mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
 	mopt->slot = OCFS2_INVALID_SLOT;
+	mopt->localalloc_opt = OCFS2_DEFAULT_LOCAL_ALLOC_SIZE;
 
 	if (!options) {
 		status = 1;

@@ -820,6 +825,15 @@ static int ocfs2_parse_options(struct super_block *sb,
 			if (option)
 				mopt->slot = (s16)option;
 			break;
+		case Opt_localalloc:
+			option = 0;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option >= 0 && (option <= ocfs2_local_alloc_size(sb) * 8))
+				mopt->localalloc_opt = option;
+			break;
 		default:
 			mlog(ML_ERROR,
 			     "Unrecognized mount option \"%s\" "

@@ -868,6 +882,9 @@ static int ocfs2_show_options(struct seq_file *s, struct vfsmount *mnt)
 	if (osb->s_atime_quantum != OCFS2_DEFAULT_ATIME_QUANTUM)
 		seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
 
+	if (osb->local_alloc_size != OCFS2_DEFAULT_LOCAL_ALLOC_SIZE)
+		seq_printf(s, ",localalloc=%d", osb->local_alloc_size);
+
 	return 0;
 }