[0/8] tree reading cleanups in mount

Message

Nikolay Borisov Oct. 15, 2019, 3:42 p.m. UTC

Hello, 

Here is the second version of the tree reading code which gets executed during 
mount. This goes a bit further than the previous posting in that it not only 
introduces a new function but also refactors the code which decides which backup
root to use. Overall I think the semantics are now much cleaner and centralised
in a single function - init_tree_roots rather than split between mount and 
backup root write out. 

The series starts gradually by simplifying find_newest_super_backup and its
callers (patches 1-2). 

It then paves the way forward by introducing read_backup_rooti (patch 3) which
supersedes (patch 4) next_root_backup, the latter is then removed (patch 6). While
at it I also remove the unnecessary objectid mutex during mount (patch 5). 

The final 2 patches streamlines how btrfs_fs_info::backup_root_index is being
initialised. 

This patchset has been tested by simulating (via btrfs-corrupt-block) corruption
of the primary root and resorting to using usebackuproot mount option. I've also 
added a regression test to btrfs-progs that will follow shortly. 

Nikolay Borisov (8):
  btrfs: Cleanup and simplify find_newest_super_backup
  btrfs: Remove newest_gen argument from find_oldest_super_backup
  btrfs: Add read_backup_root
  btrfs: Factor out tree roots initialization during mount
  btrfs: Don't use objectid_mutex during mount
  btrfs: Remove unused next_root_backup function
  btrfs: Rename find_oldest_super_backup to init_backup_root_slot
  btrfs: Streamline btrfs_fs_info::backup_root_index semantics

 fs/btrfs/disk-io.c | 252 ++++++++++++++++++++-------------------------
 1 file changed, 113 insertions(+), 139 deletions(-)

Comments

David Sterba Oct. 29, 2019, 4:42 p.m. UTC | #1

On Tue, Oct 15, 2019 at 06:42:16PM +0300, Nikolay Borisov wrote:
> Hello, 
> 
> Here is the second version of the tree reading code which gets executed during 
> mount. This goes a bit further than the previous posting in that it not only 
> introduces a new function but also refactors the code which decides which backup
> root to use. Overall I think the semantics are now much cleaner and centralised
> in a single function - init_tree_roots rather than split between mount and 
> backup root write out. 
> 
> The series starts gradually by simplifying find_newest_super_backup and its
> callers (patches 1-2). 
> 
> It then paves the way forward by introducing read_backup_rooti (patch 3) which
> supersedes (patch 4) next_root_backup, the latter is then removed (patch 6). While
> at it I also remove the unnecessary objectid mutex during mount (patch 5). 
> 
> The final 2 patches streamlines how btrfs_fs_info::backup_root_index is being
> initialised. 
> 
> This patchset has been tested by simulating (via btrfs-corrupt-block) corruption
> of the primary root and resorting to using usebackuproot mount option. I've also 
> added a regression test to btrfs-progs that will follow shortly. 
> 
> Nikolay Borisov (8):
>   btrfs: Cleanup and simplify find_newest_super_backup
>   btrfs: Remove newest_gen argument from find_oldest_super_backup
>   btrfs: Add read_backup_root
>   btrfs: Factor out tree roots initialization during mount
>   btrfs: Don't use objectid_mutex during mount
>   btrfs: Remove unused next_root_backup function
>   btrfs: Rename find_oldest_super_backup to init_backup_root_slot
>   btrfs: Streamline btrfs_fs_info::backup_root_index semantics

Moved from topic branch to misc-next. The cleaned up code looks much
better, thanks.