Re: [PATCH 0/6] Refactor nocow path

[David Sterba <dsterba@suse.cz>]

On Mon, Aug 05, 2019 at 05:47:02PM +0300, Nikolay Borisov wrote:
> This series aims at making the nocow path code more understanble. This done by 
> doing the following things: 
> 
> 1. Re-arranging and renaming some variables so that they have more expressive
> names, as well as reducing their scope. Patch 1 does this. 
> 
> 2. Since run_delalloc_nocow open-codes traversal of the btree it contains a lot
> of checks which do not pertain to the nocow logic per-se, but are there to 
> ensure the code has found the correct EXTENT_ITEM. The nocow logic itself 
> contains some subtle checks which are non-obvious at first. Patch 2 rectifies 
> this by adding appropriate comments.
> 
> 3. Patch 3 duplicates the call to btrfs_add_ordered_extent into each branch for 
> REGULAR or PREALLOC extents. Despite this duplication I think the code flow 
> becomes more streamlined and easier to understand. It also does away with one 
> of the local variables. 
> 
> 4. Patch 4 moves extent checking code into the branch it pertains to. 
> 
> 5. Patch 5 simplifies the conditions of the main 'if' in that function 
> 
> 6. Finally, patch 6 removes the BUG_ON that will be triggered in case 
> btrfs_add_ordered_extent returned ENOMEM. Now it's replaced with proper graceful
> error handling. 
> 
> This patchset has been tested with a full xfstest run with -onodatacow option
> mount options set. 
> 
> Nikolay Borisov (6):
>   btrfs: Refactor run_delalloc_nocow
>   btrfs: Improve comments around nocow path
>   btrfs: Simplify run_delalloc_nocow
>   btrfs: Streamline code in run_delalloc_nocow in case of inline extents
>   btrfs: Simplify extent type check
>   btrfs: Remove BUG_ON from run_delalloc_nocow

The patchset has been in for-next, no problems so far so I'd like to
promote it to misc-next. Patch 2 has some nontrivial changes suggested,
please update and resend. Thanks.