Btrfs: Fix a number of inline extent problems that Yan Zheng reported.

The fixes do a number of things:

1) Most btrfs_drop_extent callers will try to leave the inline extents in
place.  It can truncate bytes off the beginning of the inline extent if
required.

2) writepage can now update the inline extent, allowing mmap writes to
go directly into the inline extent.

3) btrfs_truncate_in_transaction truncates inline extents

4) extent_map.c fixed to not merge inline extent mappings and hole
mappings together
Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.c

@@ -1930,7 +1930,7 @@ static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			struct btrfs_path *path,
-			u32 new_size)
+			u32 new_size, int from_end)
 {
 	int ret = 0;
 	int slot;
@@ -1946,13 +1946,17 @@ int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 
 	slot_orig = path->slots[0];
 	leaf = path->nodes[0];
+	slot = path->slots[0];
+
+	old_size = btrfs_item_size_nr(leaf, slot);
+	if (old_size == new_size)
+		return 0;
 
 	nritems = btrfs_header_nritems(leaf);
 	data_end = leaf_data_end(root, leaf);
 
-	slot = path->slots[0];
 	old_data_start = btrfs_item_offset_nr(leaf, slot);
-	old_size = btrfs_item_size_nr(leaf, slot); BUG_ON(old_size <= new_size);
+
 	size_diff = old_size - new_size;
 
 	BUG_ON(slot < 0);
@@ -1984,9 +1988,45 @@ int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 	}
 
 	/* shift the data */
-	memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
-		      data_end + size_diff, btrfs_leaf_data(leaf) +
-		      data_end, old_data_start + new_size - data_end);
+	if (from_end) {
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start + new_size - data_end);
+	} else {
+		struct btrfs_disk_key disk_key;
+		u64 offset;
+
+		btrfs_item_key(leaf, &disk_key, slot);
+
+		if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
+			unsigned long ptr;
+			struct btrfs_file_extent_item *fi;
+
+			fi = btrfs_item_ptr(leaf, slot,
+					    struct btrfs_file_extent_item);
+			fi = (struct btrfs_file_extent_item *)(
+			     (unsigned long)fi - size_diff);
+
+			if (btrfs_file_extent_type(leaf, fi) ==
+			    BTRFS_FILE_EXTENT_INLINE) {
+				ptr = btrfs_item_ptr_offset(leaf, slot);
+				memmove_extent_buffer(leaf, ptr,
+				        (unsigned long)fi,
+				        offsetof(struct btrfs_file_extent_item,
+						 disk_bytenr));
+			}
+		}
+
+		memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
+			      data_end + size_diff, btrfs_leaf_data(leaf) +
+			      data_end, old_data_start - data_end);
+
+		offset = btrfs_disk_key_offset(&disk_key);
+		btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
+		btrfs_set_item_key(leaf, &disk_key, slot);
+		if (slot == 0)
+			fixup_low_keys(trans, root, path, &disk_key, 1);
+	}
 
 	item = btrfs_item_nr(leaf, slot);
 	btrfs_set_item_size(leaf, item, new_size);

fs/btrfs/ctree.h

@@ -907,7 +907,7 @@ int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			struct btrfs_path *path,
-			u32 new_size);
+			u32 new_size, int from_end);
 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_key *key, struct btrfs_path *p, int
 		      ins_len, int cow);

fs/btrfs/dir-item.c

@@ -249,7 +249,7 @@ int btrfs_delete_one_dir_name(struct btrfs_trans_handle *trans,
 		memmove_extent_buffer(leaf, ptr, ptr + sub_item_len,
 			item_len - (ptr + sub_item_len - start));
 		ret = btrfs_truncate_item(trans, root, path,
-					  item_len - sub_item_len);
+					  item_len - sub_item_len, 1);
 	}
 	return 0;
 }

fs/btrfs/extent_map.c

@@ -263,7 +263,12 @@ int add_extent_mapping(struct extent_map_tree *tree,
 		if (prev && prev->end + 1 == em->start &&
 		    ((em->block_start == EXTENT_MAP_HOLE &&
 		      prev->block_start == EXTENT_MAP_HOLE) ||
-			     (em->block_start == prev->block_end + 1))) {
+		     (em->block_start == EXTENT_MAP_INLINE &&
+		      prev->block_start == EXTENT_MAP_INLINE) ||
+		     (em->block_start == EXTENT_MAP_DELALLOC &&
+		      prev->block_start == EXTENT_MAP_DELALLOC) ||
+		     (em->block_start < EXTENT_MAP_DELALLOC - 1 &&
+		      em->block_start == prev->block_end + 1))) {
 			em->start = prev->start;
 			em->block_start = prev->block_start;
 			rb_erase(&prev->rb_node, &tree->map);
@@ -1618,13 +1623,13 @@ int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
 	u64 extent_offset;
 	u64 last_byte = i_size_read(inode);
 	u64 block_start;
+	u64 iosize;
 	sector_t sector;
 	struct extent_map *em;
 	struct block_device *bdev;
 	int ret;
 	int nr = 0;
 	size_t page_offset = 0;
-	size_t iosize;
 	size_t blocksize;
 	loff_t i_size = i_size_read(inode);
 	unsigned long end_index = i_size >> PAGE_CACHE_SHIFT;
@@ -1684,7 +1689,7 @@ int extent_write_full_page(struct extent_map_tree *tree, struct page *page,
 			clear_extent_dirty(tree, cur, page_end, GFP_NOFS);
 			break;
 		}
-		em = get_extent(inode, page, page_offset, cur, end, 0);
+		em = get_extent(inode, page, page_offset, cur, end, 1);
 		if (IS_ERR(em) || !em) {
 			SetPageError(page);
 			break;

fs/btrfs/extent_map.h

@@ -24,7 +24,6 @@
  * page->private values.  Every page that is controlled by the extent
  * map has page->private set to one.
  */
-
 #define EXTENT_PAGE_PRIVATE 1
 #define EXTENT_PAGE_PRIVATE_FIRST_PAGE 3
 

fs/btrfs/file-item.c

@@ -291,7 +291,7 @@ int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
 	new_item_size = blocks * BTRFS_CRC32_SIZE;
 	if (new_item_size >= btrfs_item_size_nr(leaf, slot))
 		return 0;
-	ret = btrfs_truncate_item(trans, root, path, new_item_size);
+	ret = btrfs_truncate_item(trans, root, path, new_item_size, 1);
 	BUG_ON(ret);
 	return ret;
 }

fs/btrfs/file.c

@@ -115,8 +115,20 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 		goto fail;
 	}
 	if (ret == 1) {
+		struct btrfs_key found_key;
+
+		if (path->slots[0] == 0)
+			goto insert;
+
 		path->slots[0]--;
 		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+		if (found_key.objectid != inode->i_ino)
+			goto insert;
+
+		if (found_key.type != BTRFS_EXTENT_DATA_KEY)
+			goto insert;
 		ei = btrfs_item_ptr(leaf, path->slots[0],
 				    struct btrfs_file_extent_item);
 
@@ -152,6 +164,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 			ret = btrfs_search_slot(trans, root, &key, path,
 						offset + size - found_end, 1);
 			BUG_ON(ret != 0);
+
 			ret = btrfs_extend_item(trans, root, path,
 						offset + size - found_end);
 			if (ret) {
@@ -292,7 +305,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	 */
 	inline_size = end_pos;
 	if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
-	    inline_size > 8192 ||
+	    inline_size > 32768 ||
 	    inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
 		u64 last_end;
 
@@ -312,7 +325,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 		aligned_end = (pos + write_bytes + root->sectorsize - 1) &
 			~((u64)root->sectorsize - 1);
 		err = btrfs_drop_extents(trans, root, inode, start_pos,
-					 aligned_end, end_pos, &hint_byte);
+					 aligned_end, aligned_end, &hint_byte);
 		if (err)
 			goto failed;
 		err = insert_inline_extent(trans, root, inode, start_pos,
@@ -456,13 +469,15 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			goto next_slot;
 		}
 
-		/* FIXME, there's only one inline extent allowed right now */
 		if (found_inline) {
 			u64 mask = root->sectorsize - 1;
 			search_start = (extent_end + mask) & ~mask;
 		} else
 			search_start = extent_end;
 
+		if (end <= extent_end && start >= key.offset && found_inline) {
+			*hint_byte = EXTENT_MAP_INLINE;
+		}
 		if (end < extent_end && end >= key.offset) {
 			if (found_extent) {
 				u64 disk_bytenr =
@@ -479,8 +494,10 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 					BUG_ON(ret);
 				}
 			}
-			if (!found_inline)
-				bookend = 1;
+			bookend = 1;
+			if (found_inline && start <= key.offset &&
+			    inline_end < extent_end)
+				keep = 1;
 		}
 		/* truncate existing extent */
 		if (start > key.offset) {
@@ -510,7 +527,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 				new_size = btrfs_file_extent_calc_inline_size(
 						   inline_end - key.offset);
 				btrfs_truncate_item(trans, root, path,
-						    new_size);
+						    new_size, 1);
 			}
 		}
 		/* delete the entire extent */
@@ -551,6 +568,13 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			if (!bookend)
 				continue;
 		}
+		if (bookend && found_inline && start <= key.offset &&
+		    inline_end < extent_end) {
+			u32 new_size;
+			new_size = btrfs_file_extent_calc_inline_size(
+						   extent_end - inline_end);
+			btrfs_truncate_item(trans, root, path, new_size, 0);
+		}
 		/* create bookend, splitting the extent in two */
 		if (bookend && found_extent) {
 			struct btrfs_key ins;

fs/btrfs/inode.c

@@ -89,6 +89,9 @@ static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 	ret = btrfs_drop_extents(trans, root, inode,
 				 start, start + num_bytes, start, &alloc_hint);
 
+	if (alloc_hint == EXTENT_MAP_INLINE)
+		goto out;
+
 	ret = btrfs_alloc_extent(trans, root, inode->i_ino, num_bytes, 0,
 				 alloc_hint, (u64)-1, &ins, 1);
 	if (ret) {
@@ -558,6 +561,7 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 	u64 item_end = 0;
 	int found_extent;
 	int del_item;
+	int extent_type = -1;
 
 	btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 	path = btrfs_alloc_path();
@@ -597,10 +601,15 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
 			fi = btrfs_item_ptr(leaf, path->slots[0],
 					    struct btrfs_file_extent_item);
-			if (btrfs_file_extent_type(leaf, fi) !=
-			    BTRFS_FILE_EXTENT_INLINE) {
+			extent_type = btrfs_file_extent_type(leaf, fi);
+			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 				item_end +=
 				    btrfs_file_extent_num_bytes(leaf, fi);
+			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
+				struct btrfs_item *item = btrfs_item_nr(leaf,
+							        path->slots[0]);
+				item_end += btrfs_file_extent_inline_len(leaf,
+									 item);
 			}
 		}
 		if (found_type == BTRFS_CSUM_ITEM_KEY) {
@@ -608,7 +617,7 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 						  inode->i_size);
 			BUG_ON(ret);
 		}
-		if (item_end < inode->i_size) {
+		if (item_end <= inode->i_size) {
 			if (found_type == BTRFS_DIR_ITEM_KEY) {
 				found_type = BTRFS_INODE_ITEM_KEY;
 			} else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
@@ -629,9 +638,10 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 		found_extent = 0;
 
 		/* FIXME, shrink the extent if the ref count is only 1 */
-		if (found_type == BTRFS_EXTENT_DATA_KEY &&
-			   btrfs_file_extent_type(leaf, fi) !=
-			   BTRFS_FILE_EXTENT_INLINE) {
+		if (found_type != BTRFS_EXTENT_DATA_KEY)
+			goto delete;
+
+		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 			u64 num_dec;
 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 			if (!del_item) {
@@ -659,7 +669,15 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 					inode->i_blocks -= num_dec;
 				}
 			}
+		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
+			   !del_item) {
+			u32 newsize = inode->i_size - found_key.offset;
+			newsize = btrfs_file_extent_calc_inline_size(newsize);
+			ret = btrfs_truncate_item(trans, root, path,
+						  newsize, 1);
+			BUG_ON(ret);
 		}
+delete:
 		if (del_item) {
 			ret = btrfs_del_item(trans, root, path);
 			if (ret)
@@ -769,7 +787,7 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 		u64 pos = (inode->i_size + mask) & ~mask;
 		u64 block_end = attr->ia_size | mask;
 		u64 hole_size;
-		u64 alloc_hint;
+		u64 alloc_hint = 0;
 
 		if (attr->ia_size <= pos)
 			goto out;
@@ -786,8 +804,11 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 					 pos, pos + hole_size, pos,
 					 &alloc_hint);
 
-		err = btrfs_insert_file_extent(trans, root, inode->i_ino,
-					       pos, 0, 0, hole_size);
+		if (alloc_hint != EXTENT_MAP_INLINE) {
+			err = btrfs_insert_file_extent(trans, root,
+						       inode->i_ino,
+						       pos, 0, 0, hole_size);
+		}
 		btrfs_end_transaction(trans, root);
 		mutex_unlock(&root->fs_info->fs_mutex);
 		unlock_extent(em_tree, pos, block_end, GFP_NOFS);
@@ -1531,8 +1552,8 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
 		em->end = EXTENT_MAP_HOLE;
 	}
 	em->bdev = inode->i_sb->s_bdev;
-	ret = btrfs_lookup_file_extent(NULL, root, path,
-				       objectid, start, 0);
+	ret = btrfs_lookup_file_extent(trans, root, path,
+				       objectid, start, trans != NULL);
 	if (ret < 0) {
 		err = ret;
 		goto out;
@@ -1627,15 +1648,23 @@ struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
 			((u64)root->sectorsize -1);
 		map = kmap(page);
 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
-		read_extent_buffer(leaf, map + page_offset, ptr, copy_size);
-		
-		if (em->start + copy_size <= em->end) {
-			size = min_t(u64, em->end + 1 - em->start,
-				PAGE_CACHE_SIZE - page_offset) - copy_size;
-			memset(map + page_offset + copy_size, 0, size);
+		if (create == 0 && !PageUptodate(page)) {
+			read_extent_buffer(leaf, map + page_offset, ptr,
+					   copy_size);
+			flush_dcache_page(page);
+		} else if (create && PageUptodate(page)) {
+			if (!trans) {
+				kunmap(page);
+				free_extent_map(em);
+				em = NULL;
+				btrfs_release_path(root, path);
+				trans = btrfs_start_transaction(root, 1);
+				goto again;
+			}
+			write_extent_buffer(leaf, map + page_offset, ptr,
+					    copy_size);
+			btrfs_mark_buffer_dirty(leaf);
 		}
-
-		flush_dcache_page(page);
 		kunmap(page);
 		set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
 		goto insert;