Re: [PATCH 00/20] btrfs: remove plenty of redundant btrfs_mark_buffer_dirty() calls

[David Sterba <dsterba@suse.cz>]

On Wed, Dec 18, 2024 at 05:06:27PM +0000, fdmanana@kernel.org wrote:
> From: Filipe Manana <fdmanana@suse.com>
> 
> There's quite a lot of places calling btrfs_mark_buffer_dirty() when it
> is not necessary because we already have a path setup for writing, due
> to calls to btrfs_search_slot() with a 'cow' argument having a value of 1
> or anything that calls btrfs_search_slot() that way (and it's obvious
> from the context). These make the code more verbose, add some overhead
> and increase the module's text size unnecessarily. This patchset removes
> such unnecessary calls. Often people keep adding them because they copy
> the approach done from such places.

I've read the explict calls to btrfs_mark_buffer_dirty() as source-level
marker of the section pairing btrfs_search_slot(). There are also some
assertions and eb state checks that were done in set_extent_buffer_dirty(),
now we're losing them.

While the code gets reduced and the calls may be redundant for some
reasons I think we should keep something at least to do the assertions,
or eventually optimize btrfs_mark_buffer_dirty() to be faster if the
path is set up by the search slot with cow=1.