Merge branch 'for-linus-4.1' of git://git.kernel.org/pub/scm/linux/ke…

…rnel/git/mason/linux-btrfs

Pull btrfs fixes from Chris Mason:
 "The first commit is a fix from Filipe for a very old extent buffer
  reuse race that triggered a BUG_ON.  It hasn't come up often, I looked
  through old logs at FB and we hit it a handful of times over the last
  year.

  The rest are other corners he hit during testing"

* 'for-linus-4.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
  Btrfs: fix race when reusing stale extent buffers that leads to BUG_ON
  Btrfs: fix race between block group creation and their cache writeout
  Btrfs: fix panic when starting bg cache writeout after IO error
  Btrfs: fix crash after inode cache writeback failure

fs/btrfs/extent-tree.c

@@ -3180,8 +3180,6 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 fail:
 	btrfs_release_path(path);
-	if (ret)
-		btrfs_abort_transaction(trans, root, ret);
 	return ret;
 
 }
@@ -3487,8 +3485,30 @@ int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
 				ret = 0;
 			}
 		}
-		if (!ret)
+		if (!ret) {
 			ret = write_one_cache_group(trans, root, path, cache);
+			/*
+			 * Our block group might still be attached to the list
+			 * of new block groups in the transaction handle of some
+			 * other task (struct btrfs_trans_handle->new_bgs). This
+			 * means its block group item isn't yet in the extent
+			 * tree. If this happens ignore the error, as we will
+			 * try again later in the critical section of the
+			 * transaction commit.
+			 */
+			if (ret == -ENOENT) {
+				ret = 0;
+				spin_lock(&cur_trans->dirty_bgs_lock);
+				if (list_empty(&cache->dirty_list)) {
+					list_add_tail(&cache->dirty_list,
+						      &cur_trans->dirty_bgs);
+					btrfs_get_block_group(cache);
+				}
+				spin_unlock(&cur_trans->dirty_bgs_lock);
+			} else if (ret) {
+				btrfs_abort_transaction(trans, root, ret);
+			}
+		}
 
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)
@@ -3597,8 +3617,11 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 				ret = 0;
 			}
 		}
-		if (!ret)
+		if (!ret) {
 			ret = write_one_cache_group(trans, root, path, cache);
+			if (ret)
+				btrfs_abort_transaction(trans, root, ret);
+		}
 
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)

fs/btrfs/extent_io.c

@@ -4772,6 +4772,25 @@ struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info,
 			       start >> PAGE_CACHE_SHIFT);
 	if (eb && atomic_inc_not_zero(&eb->refs)) {
 		rcu_read_unlock();
+		/*
+		 * Lock our eb's refs_lock to avoid races with
+		 * free_extent_buffer. When we get our eb it might be flagged
+		 * with EXTENT_BUFFER_STALE and another task running
+		 * free_extent_buffer might have seen that flag set,
+		 * eb->refs == 2, that the buffer isn't under IO (dirty and
+		 * writeback flags not set) and it's still in the tree (flag
+		 * EXTENT_BUFFER_TREE_REF set), therefore being in the process
+		 * of decrementing the extent buffer's reference count twice.
+		 * So here we could race and increment the eb's reference count,
+		 * clear its stale flag, mark it as dirty and drop our reference
+		 * before the other task finishes executing free_extent_buffer,
+		 * which would later result in an attempt to free an extent
+		 * buffer that is dirty.
+		 */
+		if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) {
+			spin_lock(&eb->refs_lock);
+			spin_unlock(&eb->refs_lock);
+		}
 		mark_extent_buffer_accessed(eb, NULL);
 		return eb;
 	}

fs/btrfs/free-space-cache.c

@@ -3466,18 +3466,28 @@ int btrfs_write_out_ino_cache(struct btrfs_root *root,
 	struct btrfs_free_space_ctl *ctl = root->free_ino_ctl;
 	int ret;
 	struct btrfs_io_ctl io_ctl;
+	bool release_metadata = true;
 
 	if (!btrfs_test_opt(root, INODE_MAP_CACHE))
 		return 0;
 
 	memset(&io_ctl, 0, sizeof(io_ctl));
 	ret = __btrfs_write_out_cache(root, inode, ctl, NULL, &io_ctl,
 				      trans, path, 0);
-	if (!ret)
+	if (!ret) {
+		/*
+		 * At this point writepages() didn't error out, so our metadata
+		 * reservation is released when the writeback finishes, at
+		 * inode.c:btrfs_finish_ordered_io(), regardless of it finishing
+		 * with or without an error.
+		 */
+		release_metadata = false;
 		ret = btrfs_wait_cache_io(root, trans, NULL, &io_ctl, path, 0);
+	}
 
 	if (ret) {
-		btrfs_delalloc_release_metadata(inode, inode->i_size);
+		if (release_metadata)
+			btrfs_delalloc_release_metadata(inode, inode->i_size);
 #ifdef DEBUG
 		btrfs_err(root->fs_info,
 			"failed to write free ino cache for root %llu",

fs/btrfs/ordered-data.c

@@ -722,6 +722,7 @@ void btrfs_start_ordered_extent(struct inode *inode,
 int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 {
 	int ret = 0;
+	int ret_wb = 0;
 	u64 end;
 	u64 orig_end;
 	struct btrfs_ordered_extent *ordered;
@@ -741,9 +742,14 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 	if (ret)
 		return ret;
 
-	ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
-	if (ret)
-		return ret;
+	/*
+	 * If we have a writeback error don't return immediately. Wait first
+	 * for any ordered extents that haven't completed yet. This is to make
+	 * sure no one can dirty the same page ranges and call writepages()
+	 * before the ordered extents complete - to avoid failures (-EEXIST)
+	 * when adding the new ordered extents to the ordered tree.
+	 */
+	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 
 	end = orig_end;
 	while (1) {
@@ -767,7 +773,7 @@ int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 			break;
 		end--;
 	}
-	return ret;
+	return ret_wb ? ret_wb : ret;
 }
 
 /*