[v2,0/8] btrfs: separate/simplify/unify subpage handling

Message

Qu Wenruo Feb. 7, 2025, 4:25 a.m. UTC

[CHANGELOG]
v2:
- Fix a crash in 4K page sized system caused by
  btrfs_clear_buffer_dirty()
  It looks like using btrfs_meta_folio_set_writeback() and
  btrfs_meta_folio_clear_writeback() is screwing up the btree inode map.
  
  Fix it by reverting to the old manually clearing of
  PAGECACHE_DIRTY_TAG.

[METADATA SUBPAGE CHECKS]
Currently we only have one btrfs_is_subpage() check, utilized by data
and metadata.

But the truth is, btrfs_is_subpage() can return incorrect result if the
target folio belongs to a dummy extent buffer.

This means we have quite some metadata call sites doing their own
metadata specific subpage check.

[SUBPAGE SPECIFIC HANDLINGS]
There are several functions that split the metadata subpage handling
into a dedicated path.

But the truth is, a lot of such paths only have minor differences
compared to the regular routine.

[THIS SERIES]
This series address the above problems and prepare for the incoming data
larger folio support by:

- Remove btrfs_fs_info::sectors_per_page
  This is mostly a preparation for the future data larger folio support.

- Introduce a dedicated metadata specific subpage check

- Unify/simplify metadata subpage handling
  So that we have a single unified path for metadata

Qu Wenruo (8):
  btrfs: remove btrfs_fs_info::sectors_per_page
  btrfs: extract metadata subpage detection into a dedicated helper
  btrfs: make subpage attach and detach to handle metadata properly
  btrfs: use metadata specific helpers to simplify extent buffer helpers
  btrfs: simplify btrfs_clear_buffer_dirty()
  btrfs: simplify write_one_eb()
  btrfs: simplify read_extent_buffer_pages_nowait()
  btrfs: require strict data/metadata split for subpage check

 fs/btrfs/disk-io.c   |   5 +-
 fs/btrfs/extent_io.c | 184 +++++++++++++++-------------------------
 fs/btrfs/fs.h        |   7 +-
 fs/btrfs/subpage.c   | 196 +++++++++++++++++++++++++++----------------
 fs/btrfs/subpage.h   |  52 ++++++++++--
 5 files changed, 247 insertions(+), 197 deletions(-)

Comments

David Sterba Feb. 7, 2025, 7:30 p.m. UTC | #1

On Fri, Feb 07, 2025 at 02:55:59PM +1030, Qu Wenruo wrote:
> [CHANGELOG]
> v2:
> - Fix a crash in 4K page sized system caused by
>   btrfs_clear_buffer_dirty()
>   It looks like using btrfs_meta_folio_set_writeback() and
>   btrfs_meta_folio_clear_writeback() is screwing up the btree inode map.
>   
>   Fix it by reverting to the old manually clearing of
>   PAGECACHE_DIRTY_TAG.
> 
> [METADATA SUBPAGE CHECKS]
> Currently we only have one btrfs_is_subpage() check, utilized by data
> and metadata.
> 
> But the truth is, btrfs_is_subpage() can return incorrect result if the
> target folio belongs to a dummy extent buffer.
> 
> This means we have quite some metadata call sites doing their own
> metadata specific subpage check.
> 
> [SUBPAGE SPECIFIC HANDLINGS]
> There are several functions that split the metadata subpage handling
> into a dedicated path.
> 
> But the truth is, a lot of such paths only have minor differences
> compared to the regular routine.
> 
> [THIS SERIES]
> This series address the above problems and prepare for the incoming data
> larger folio support by:
> 
> - Remove btrfs_fs_info::sectors_per_page
>   This is mostly a preparation for the future data larger folio support.
> 
> - Introduce a dedicated metadata specific subpage check
> 
> - Unify/simplify metadata subpage handling
>   So that we have a single unified path for metadata
> 
> Qu Wenruo (8):
>   btrfs: remove btrfs_fs_info::sectors_per_page
>   btrfs: extract metadata subpage detection into a dedicated helper
>   btrfs: make subpage attach and detach to handle metadata properly
>   btrfs: use metadata specific helpers to simplify extent buffer helpers
>   btrfs: simplify btrfs_clear_buffer_dirty()
>   btrfs: simplify write_one_eb()
>   btrfs: simplify read_extent_buffer_pages_nowait()
>   btrfs: require strict data/metadata split for subpage check

All look good to me.

Reviewed-by: David Sterba <[e-mail]>

Please update the subjects in the simplify patches in what regard they
simplify the functions. For example "sipmlify subpage handling in ...".
Thanks.