[PATCH 00/16] btrfs: make open_ctree() init/exit sequence strictly matched

[Qu Wenruo <wqu@suse.com>]

Just like init_btrfs_fs(), open_ctree() also has tons of different
labels for its error handling.

And unsurprisingly the error handling labels are not matched correctly,
e.g. we always call btrfs_mapping_tree_free() even we didn't reach
sys chunk array read.

And every time we need to add some new function, it will be a disaster
just to understand where the new function should be put and how the
error handling should be done.

This patchset will follow the init_btrfs_fs() method, by introducing
an open_ctree_seq[] array, which contains the following sections:

- btree_inode init/exit
- super block read and verification
- mount options and features check
- workqueues init/exit
- chunk tree init/exit
- tree roots init/exit
- mount time check and various item load
- sysfs init/exit
- block group tree init/exit
- subvolume trees init/exit
- kthread init/exit
- qgroup init/exit

The remaining part of open_ctree() is only less than 50 lines, and are
all related to the very end of the mount progress, including log-replay,
uuid tree check.

Also to do better testing, for DEBUG build there will be a new mount
option, "fail_mount=%u" to allow open_ctree() to fail at certain stage
of open_ctree_seq[].

Unfortunately since that mount option can only be parsed in
open_ctree_features_init(), this means we can only fail after stage 2.
But this should still provide much better testing coverage.

To David:

This serious is going to conflict with the regression fix on block group
tree ("btrfs: loosen the block-group-tree feature dependency check"),
please let me rebase before merging.

To Johannes:

Not 100% sure about the zoned code, thus some review/advice would be very
appreciated.

Maybe I can further extract all the zoned code into an init/exit pair?

Qu Wenruo (16):
  btrfs: make btrfs module init/exit match their sequence
  btrfs: initialize fs_info->sb at the very beginning of open_ctree()
  btrfs: remove @fs_devices argument from open_ctree()
  btrfs: extract btree inode init code into its own init/exit helpers
  btrfs: extract super block read code into its own init helper
  btrfs: extract mount options and features init code into its own init
    helper
  btrfs: move btrfs_init_workqueus() and btrfs_stop_all_workers() into
    open_ctree_seq[]
  btrfs: extract chunk tree read code into its own init/exit helpers
  btrfs: extract tree roots and zone info initialization into init/exit
    helpers
  btrfs: extract mount time checks and items load code into its init
    helper
  btrfs: extract sysfs init into its own helper
  btrfs: extra block groups read code into its own init/exit helpers
  btrfs: move the fs root related code into its own init/exit helpers
  btrfs: extract kthread code into its own init/exit helpers
  btrfs: move qgroup init/exit code into open_ctree_seq[] array
  btrfs: introduce a debug mount option to do error injection for each
    stage of open_ctree()

 fs/btrfs/compression.c |   3 +-
 fs/btrfs/compression.h |   2 +-
 fs/btrfs/ctree.h       |   7 +
 fs/btrfs/disk-io.c     | 660 ++++++++++++++++++++++++++---------------
 fs/btrfs/disk-io.h     |   4 +-
 fs/btrfs/props.c       |   3 +-
 fs/btrfs/props.h       |   2 +-
 fs/btrfs/super.c       | 229 +++++++-------
 8 files changed, 557 insertions(+), 353 deletions(-)

-- 
2.37.3