Btrfs: Fix extent replacment race

Data COW means that whenever we write to a file, we replace any old
extent pointers with new ones.  There was a window where a readpage
might find the old extent pointers on disk and cache them in the
extent_map tree in ram in the middle of a given write replacing them.

Even though both the readpage and the write had their respective bytes
in the file locked, the extent readpage inserts may cover more bytes than
it had locked down.

This commit closes the race by keeping the new extent pinned in the extent
map tree until after the on-disk btree is properly setup with the new
extent pointers.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.h

@@ -2292,7 +2292,7 @@ extern struct file_operations btrfs_file_operations;
 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct inode *inode,
 		       u64 start, u64 end, u64 locked_end,
-		       u64 inline_limit, u64 *hint_block);
+		       u64 inline_limit, u64 *hint_block, int drop_cache);
 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 			      struct btrfs_root *root,
 			      struct inode *inode, u64 start, u64 end);

fs/btrfs/extent_map.c

@@ -198,6 +198,56 @@ static int mergable_maps(struct extent_map *prev, struct extent_map *next)
 	return 0;
 }
 
+int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len)
+{
+	int ret = 0;
+	struct extent_map *merge = NULL;
+	struct rb_node *rb;
+	struct extent_map *em;
+
+	write_lock(&tree->lock);
+	em = lookup_extent_mapping(tree, start, len);
+
+	WARN_ON(em->start != start || !em);
+
+	if (!em)
+		goto out;
+
+	clear_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+	if (em->start != 0) {
+		rb = rb_prev(&em->rb_node);
+		if (rb)
+			merge = rb_entry(rb, struct extent_map, rb_node);
+		if (rb && mergable_maps(merge, em)) {
+			em->start = merge->start;
+			em->len += merge->len;
+			em->block_len += merge->block_len;
+			em->block_start = merge->block_start;
+			merge->in_tree = 0;
+			rb_erase(&merge->rb_node, &tree->map);
+			free_extent_map(merge);
+		}
+	}
+
+	rb = rb_next(&em->rb_node);
+	if (rb)
+		merge = rb_entry(rb, struct extent_map, rb_node);
+	if (rb && mergable_maps(em, merge)) {
+		em->len += merge->len;
+		em->block_len += merge->len;
+		rb_erase(&merge->rb_node, &tree->map);
+		merge->in_tree = 0;
+		free_extent_map(merge);
+	}
+
+	free_extent_map(em);
+out:
+	write_unlock(&tree->lock);
+	return ret;
+
+}
+
 /**
  * add_extent_mapping - add new extent map to the extent tree
  * @tree:	tree to insert new map in

fs/btrfs/extent_map.h

@@ -59,4 +59,5 @@ struct extent_map *alloc_extent_map(gfp_t mask);
 void free_extent_map(struct extent_map *em);
 int __init extent_map_init(void);
 void extent_map_exit(void);
+int unpin_extent_cache(struct extent_map_tree *tree, u64 start, u64 len);
 #endif

fs/btrfs/file.c

@@ -177,10 +177,10 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 		}
 		flags = em->flags;
 		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
-			write_unlock(&em_tree->lock);
 			if (em->start <= start &&
 			    (!testend || em->start + em->len >= start + len)) {
 				free_extent_map(em);
+				write_unlock(&em_tree->lock);
 				break;
 			}
 			if (start < em->start) {
@@ -190,6 +190,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 				start = em->start + em->len;
 			}
 			free_extent_map(em);
+			write_unlock(&em_tree->lock);
 			continue;
 		}
 		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
@@ -269,7 +270,7 @@ int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       struct btrfs_root *root, struct inode *inode,
 		       u64 start, u64 end, u64 locked_end,
-		       u64 inline_limit, u64 *hint_byte)
+		       u64 inline_limit, u64 *hint_byte, int drop_cache)
 {
 	u64 extent_end = 0;
 	u64 search_start = start;
@@ -294,7 +295,8 @@ noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	int ret;
 
 	inline_limit = 0;
-	btrfs_drop_extent_cache(inode, start, end - 1, 0);
+	if (drop_cache)
+		btrfs_drop_extent_cache(inode, start, end - 1, 0);
 
 	path = btrfs_alloc_path();
 	if (!path)

fs/btrfs/inode.c

@@ -232,7 +232,8 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 	}
 
 	ret = btrfs_drop_extents(trans, root, inode, start,
-				 aligned_end, aligned_end, start, &hint_byte);
+				 aligned_end, aligned_end, start,
+				 &hint_byte, 1);
 	BUG_ON(ret);
 
 	if (isize > actual_end)
@@ -241,7 +242,7 @@ static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 				   inline_len, compressed_size,
 				   compressed_pages);
 	BUG_ON(ret);
-	btrfs_drop_extent_cache(inode, start, aligned_end, 0);
+	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 	return 0;
 }
 
@@ -1455,9 +1456,19 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	BUG_ON(!path);
 
 	path->leave_spinning = 1;
+
+	/*
+	 * we may be replacing one extent in the tree with another.
+	 * The new extent is pinned in the extent map, and we don't want
+	 * to drop it from the cache until it is completely in the btree.
+	 *
+	 * So, tell btrfs_drop_extents to leave this extent in the cache.
+	 * the caller is expected to unpin it and allow it to be merged
+	 * with the others.
+	 */
 	ret = btrfs_drop_extents(trans, root, inode, file_pos,
 				 file_pos + num_bytes, locked_end,
-				 file_pos, &hint);
+				 file_pos, &hint, 0);
 	BUG_ON(ret);
 
 	ins.objectid = inode->i_ino;
@@ -1485,7 +1496,6 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 
 	inode_add_bytes(inode, num_bytes);
-	btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
 
 	ins.objectid = disk_bytenr;
 	ins.offset = disk_num_bytes;
@@ -1596,6 +1606,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 						ordered_extent->len,
 						compressed, 0, 0,
 						BTRFS_FILE_EXTENT_REG);
+		unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
+				   ordered_extent->file_offset,
+				   ordered_extent->len);
 		BUG_ON(ret);
 	}
 	unlock_extent(io_tree, ordered_extent->file_offset,
@@ -2940,7 +2953,7 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
 						 cur_offset,
 						 cur_offset + hole_size,
 						 block_end,
-						 cur_offset, &hint_byte);
+						 cur_offset, &hint_byte, 1);
 			if (err)
 				break;
 			err = btrfs_insert_file_extent(trans, root,
@@ -5086,6 +5099,8 @@ static int prealloc_file_range(struct btrfs_trans_handle *trans,
 						  0, 0, 0,
 						  BTRFS_FILE_EXTENT_PREALLOC);
 		BUG_ON(ret);
+		btrfs_drop_extent_cache(inode, cur_offset,
+					cur_offset + ins.offset -1, 0);
 		num_bytes -= ins.offset;
 		cur_offset += ins.offset;
 		alloc_hint = ins.objectid + ins.offset;

fs/btrfs/ioctl.c

@@ -597,9 +597,8 @@ static int btrfs_defrag_file(struct file *file)
 		clear_page_dirty_for_io(page);
 
 		btrfs_set_extent_delalloc(inode, page_start, page_end);
-
-		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 		set_page_dirty(page);
+		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 		unlock_page(page);
 		page_cache_release(page);
 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
@@ -977,7 +976,7 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 
 	/* punch hole in destination first */
 	btrfs_drop_extents(trans, root, inode, off, off + len,
-			   off + len, 0, &hint_byte);
+			   off + len, 0, &hint_byte, 1);
 
 	/* clone data */
 	key.objectid = src->i_ino;

fs/btrfs/tree-log.c

@@ -534,7 +534,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 	saved_nbytes = inode_get_bytes(inode);
 	/* drop any overlapping extents */
 	ret = btrfs_drop_extents(trans, root, inode,
-			 start, extent_end, extent_end, start, &alloc_hint);
+			 start, extent_end, extent_end, start, &alloc_hint, 1);
 	BUG_ON(ret);
 
 	if (found_type == BTRFS_FILE_EXTENT_REG ||