btrfs: cleanup the error handling labels for open_ctree()

It's a disaster in open_ctree() for its error handling, it has TEN
labels, and under a lot of cases, those labels don't help much to
understand the correct timing to call them.

Furthermore, there are a lof of cleanup functions can be called with
internal NULL pointer checks, thus there is not much need to use so many
labels.

This patch will reduce the number of error handling labels from 10 to 2:

- fail:
  Before successful btrfs_sysfs_add_mounted() call

- fail_sysfs:
  After successful btrfs_sysfs_add_mounted() call

Then the complete lists of manual inspection on the functions called in
old labels:

- btrfs_free_qgroup_config()
  Freeing rbtree entries in fs_info->qgroup_tree, initialized by
  btrfs_init_fs_info(), thus safer to call at anytime.

- kthread_stop() for transaction_kthread and cleaner_kthread
  Just add extra NULL pointer checks before calling kthread_stop().

- btrfs_cleanup_transaction()
  It cleans up entries in:
  * fs_info->trans_list list
  * fs_info->ordered_roots list
  * fs_info->delayed_root structure
  * fs_info->delalloc_roots list
  * fs_info->fs_roots_radix radix tree
  Those members are all initialized in btrfs_init_fs_info() (except
  delalloc_roots which is initialized in init_mount_fs_info()), and
  eventually will be freed by btrfs_free_fs_info().
  Thus they are all safe to call at any timing of error path.

- btrfs_free_fs_roots()
  It cleans up fs_info->dead_roots list and fs_roots_radix tree,
  both are already properly initialized in btrfs_init_fs_info(), thus
  safe to call at any timing of error path.

- fiemap_write_and_wait() for btree_inode
  This must be called after we have initialized btree_inode, although at
  the timing at btrfs_sysfs_add_mounted() we should already have a valid
  btree_inode. Just add an ASSERT() in case.

- btrfs_sysfs_remove_mounted()
  This requires a successful call on btrfs_sysfs_add_mounted()

- btrfs_sysfs_remove_fsid()
  It has extra NULL pointer check on
  fs_devices->devinfo_kobj/devices_kobj. Thus can be called at any
  timing of error path.

- btrfs_put_block_group_cache()
  It cleans up the block group cache in fs_info->block_group_cache_tree,
  which is initialized in btrfs_init_fs_info() already.
  Thus can be called at any timing of error path.

- btrfs_drop_and_free_fs_root() on data_reloc_root
  It already has a NULL pointer check.

- free_root_pointers()
  It calls free_root_extent_buffers() on each root, which has NULL
  pointer check on the passed root.

  Furthermore, in free_root_extent_buffers() it calls
  free_extent_buffer() which also has NULL pointer check on the extent
  buffer.
  Thus it's completely safe to call on half-initialized root or NULL
  root.

- btrfs_stop_all_workers()
  It calls btrfs_destroy_workqueue() on btrfs workqueues, which already
  has NULL pointer check.
  For regular workqueues, it has manual NULL pointer check.

- btrfs_mapping_tree_free()
  It cleans up the extent mapping in fs_info->mapping_tree, which can
  handle empty rb tree all by itself.

- btrfs_close_devices()
  It's called for all error path anyway.

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
 fs/btrfs/disk-io.c | 88 +++++++++++++++++++++++-----------------------
 1 file changed, 44 insertions(+), 44 deletions(-)

Message ID	4836d9630867bd65390ce92fc9242df9e3af0faa.1658474716.git.wqu@suse.com (mailing list archive)
State	New, archived
Headers	show Return-Path: <linux-btrfs-owner@kernel.org> From: Qu Wenruo <wqu@suse.com> To: linux-btrfs@vger.kernel.org Subject: [PATCH] btrfs: cleanup the error handling labels for open_ctree() Date: Fri, 22 Jul 2022 15:26:34 +0800 Message-Id: <4836d9630867bd65390ce92fc9242df9e3af0faa.1658474716.git.wqu@suse.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: bulk
Series	btrfs: cleanup the error handling labels for open_ctree() \| expand btrfs: cleanup the error handling labels for open_ctree()

btrfs: cleanup the error handling labels for open_ctree()

Commit Message

Patch