Btrfs: Fix bookend extent race v2

When dropping middle part of an extent, btrfs_drop_extents truncates
the extent at first, then inserts a bookend extent.

Since truncation and insertion can't be done atomically, there is a small
period that the bookend extent isn't in the tree. This causes problem for
functions that search the tree for file extent item. The way to fix this is
lock the range of the bookend extent before truncation.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>

fs/btrfs/extent-tree.c

@@ -3379,11 +3379,13 @@ static int noinline relocate_data_extent(struct inode *reloc_inode,
 	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
 	struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
 	struct extent_map *em;
+	u64 start = extent_key->objectid - offset;
+	u64 end = start + extent_key->offset - 1;
 
 	em = alloc_extent_map(GFP_NOFS);
 	BUG_ON(!em || IS_ERR(em));
 
-	em->start = extent_key->objectid - offset;
+	em->start = start;
 	em->len = extent_key->offset;
 	em->block_len = extent_key->offset;
 	em->block_start = extent_key->objectid;
@@ -3391,7 +3393,7 @@ static int noinline relocate_data_extent(struct inode *reloc_inode,
 	set_bit(EXTENT_FLAG_PINNED, &em->flags);
 
 	/* setup extent map to cheat btrfs_readpage */
-	mutex_lock(&BTRFS_I(reloc_inode)->extent_mutex);
+	lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
 	while (1) {
 		int ret;
 		spin_lock(&em_tree->lock);
@@ -3401,13 +3403,11 @@ static int noinline relocate_data_extent(struct inode *reloc_inode,
 			free_extent_map(em);
 			break;
 		}
-		btrfs_drop_extent_cache(reloc_inode, em->start,
-					em->start + em->len - 1, 0);
+		btrfs_drop_extent_cache(reloc_inode, start, end, 0);
 	}
-	mutex_unlock(&BTRFS_I(reloc_inode)->extent_mutex);
+	unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
 
-	return relocate_inode_pages(reloc_inode, extent_key->objectid - offset,
-				    extent_key->offset);
+	return relocate_inode_pages(reloc_inode, start, extent_key->offset);
 }
 
 struct btrfs_ref_path {
@@ -3831,7 +3831,6 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
 			 * the file extent item was modified by someone
 			 * before the extent got locked.
 			 */
-			mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
 				      lock_end, GFP_NOFS);
 			extent_locked = 0;
@@ -3896,8 +3895,12 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
 			lock_start = key.offset;
 			lock_end = lock_start + num_bytes - 1;
 		} else {
-			BUG_ON(lock_start != key.offset);
-			BUG_ON(lock_end - lock_start + 1 < num_bytes);
+			if (lock_start > key.offset ||
+			    lock_end + 1 < key.offset + num_bytes) {
+				unlock_extent(&BTRFS_I(inode)->io_tree,
+					      lock_start, lock_end, GFP_NOFS);
+				extent_locked = 0;
+			}
 		}
 
 		if (!inode) {
@@ -3951,7 +3954,6 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
 			if (ordered)
 				btrfs_put_ordered_extent(ordered);
 
-			mutex_lock(&BTRFS_I(inode)->extent_mutex);
 			extent_locked = 1;
 			continue;
 		}
@@ -4073,7 +4075,6 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
 		}
 
 		if (extent_locked) {
-			mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
 				      lock_end, GFP_NOFS);
 			extent_locked = 0;
@@ -4091,7 +4092,6 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
 	if (inode) {
 		mutex_unlock(&inode->i_mutex);
 		if (extent_locked) {
-			mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 			unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
 				      lock_end, GFP_NOFS);
 		}
@@ -4180,10 +4180,8 @@ static int noinline invalidate_extent_cache(struct btrfs_root *root,
 
 		lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
 			    key.offset + num_bytes - 1, GFP_NOFS);
-		mutex_lock(&BTRFS_I(inode)->extent_mutex);
 		btrfs_drop_extent_cache(inode, key.offset,
 					key.offset + num_bytes - 1, 1);
-		mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 		unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
 			      key.offset + num_bytes - 1, GFP_NOFS);
 		cond_resched();

fs/btrfs/extent_io.c

@@ -946,8 +946,12 @@ int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end,
 
 	err = set_extent_bit(tree, start, end, EXTENT_LOCKED, 1,
 			     &failed_start, mask);
-	if (err == -EEXIST)
+	if (err == -EEXIST) {
+		if (failed_start > start)
+			clear_extent_bit(tree, start, failed_start - 1,
+					 EXTENT_LOCKED, 1, 0, mask);
 		return 0;
+	}
 	return 1;
 }
 EXPORT_SYMBOL(try_lock_extent);

fs/btrfs/file.c

@@ -364,6 +364,7 @@ int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 		       u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
 {
 	u64 extent_end = 0;
+	u64 locked_end = end;
 	u64 search_start = start;
 	u64 leaf_start;
 	u64 ram_bytes = 0;
@@ -479,12 +480,6 @@ int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			goto next_slot;
 		}
 
-		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;
 
@@ -501,6 +496,26 @@ int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 			if (found_inline && start <= key.offset)
 				keep = 1;
 		}
+
+		if (bookend && found_extent && locked_end < extent_end) {
+			ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
+					locked_end, extent_end - 1, GFP_NOFS);
+			if (!ret) {
+				btrfs_release_path(root, path);
+				lock_extent(&BTRFS_I(inode)->io_tree,
+					locked_end, extent_end - 1, GFP_NOFS);
+				locked_end = extent_end;
+				continue;
+			}
+			locked_end = extent_end;
+		}
+
+		if (found_inline) {
+			u64 mask = root->sectorsize - 1;
+			search_start = (extent_end + mask) & ~mask;
+		} else
+			search_start = extent_end;
+
 		/* truncate existing extent */
 		if (start > key.offset) {
 			u64 new_num;
@@ -638,6 +653,10 @@ int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
 	}
 out:
 	btrfs_free_path(path);
+	if (locked_end > end) {
+		unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
+			      GFP_NOFS);
+	}
 	btrfs_check_file(root, inode);
 	return ret;
 }

fs/btrfs/inode.c

@@ -246,7 +246,6 @@ static int cow_file_range_inline(struct btrfs_trans_handle *trans,
 		return 1;
 	}
 
-	mutex_lock(&BTRFS_I(inode)->extent_mutex);
 	ret = btrfs_drop_extents(trans, root, inode, start,
 				 aligned_end, aligned_end, &hint_byte);
 	BUG_ON(ret);
@@ -258,7 +257,6 @@ static int cow_file_range_inline(struct btrfs_trans_handle *trans,
 				   compressed_pages);
 	BUG_ON(ret);
 	btrfs_drop_extent_cache(inode, start, aligned_end, 0);
-	mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 	return 0;
 }
 
@@ -437,9 +435,7 @@ static int cow_file_range(struct inode *inode, struct page *locked_page,
 	BUG_ON(disk_num_bytes >
 	       btrfs_super_total_bytes(&root->fs_info->super_copy));
 
-	mutex_lock(&BTRFS_I(inode)->extent_mutex);
 	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
-	mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 
 	while(disk_num_bytes > 0) {
 		unsigned long min_bytes;
@@ -477,8 +473,6 @@ static int cow_file_range(struct inode *inode, struct page *locked_page,
 		em->block_start = ins.objectid;
 		em->block_len = ins.offset;
 		em->bdev = root->fs_info->fs_devices->latest_bdev;
-
-		mutex_lock(&BTRFS_I(inode)->extent_mutex);
 		set_bit(EXTENT_FLAG_PINNED, &em->flags);
 
 		if (will_compress)
@@ -495,7 +489,6 @@ static int cow_file_range(struct inode *inode, struct page *locked_page,
 			btrfs_drop_extent_cache(inode, start,
 						start + ram_size - 1, 0);
 		}
-		mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 
 		cur_alloc_size = ins.offset;
 		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
@@ -1016,8 +1009,6 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 
 	INIT_LIST_HEAD(&list);
 
-	mutex_lock(&BTRFS_I(inode)->extent_mutex);
-
 	ret = btrfs_drop_extents(trans, root, inode,
 				 ordered_extent->file_offset,
 				 ordered_extent->file_offset +
@@ -1059,7 +1050,6 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 	btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 				ordered_extent->file_offset +
 				ordered_extent->len - 1, 0);
-	mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 
 	ins.objectid = ordered_extent->start;
 	ins.offset = ordered_extent->disk_len;