Skip to content

btrfs: migrate IO path to folios

Qu Wenruo linux-btrfs 2 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v2 Latest3 patches 1 replyIn review
  2. v13 patches 0 repliesSuperseded

Changes in v2 from the cover letter

- Fix a PAGE_SHIFT usage in the 3rd patch on the data read path I know this won't be touched any time soon, but considering it's really one patch away from enabling higher order folios for metadata, let's make the cleanup closer to perfection.

Patches in v2 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/3btrfs: migrate get_eb_page_index() and get_eb_offset_in_page() to foliosfs/btrfs/accessors.c fs/btrfs/ctree.c fs/btrfs/disk-io.c +2 more+145 -1160 replies
2/3btrfs: migrate subpage code to folio interfacesfs/btrfs/compression.c fs/btrfs/defrag.c fs/btrfs/disk-io.c +9 more+279 -2750 replies
3/3btrfs: migrate various btrfs end io functions to foliosfs/btrfs/compression.c fs/btrfs/extent_io.c+81 -770 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v2 yet.

Read the v2 thread

Cover letter [PATCH v2 0/3] btrfs: migrate IO path to folios

[CHANGELOG]
v2:
- Fix a PAGE_SHIFT usage in the 3rd patch on the data read path I know this won't be touched any time soon, but considering it's really one patch away from enabling higher order folios for metadata, let's make the cleanup closer to perfection.

One critical problem I hit the most during my initial higher order folios tests are, incorrect access to the pages which conflicts with the page flag policy.

Since folio flags are only set to certain pages according to their policies (PF_ANY, PF_HEAD, PF_ONLY_HEAD, PF_NO_TAIL, PF_NO_COMPOUND and the most weird on PF_SECOND), setting page flags violating the policy would immedate lead to VM_BUG_ON().

Thus no matter if we go compound page or folio, we can not go the page-by-page iteration helpers that easily.
One of the hot spots which can lead to VM_BUG_ON()s are the endio helpers.

So this patch would:

- Make metadata set/get helpers to utilize folio interfaces

- Make subpage code to accept folios directly
This is to avoid btrfs_page_*() helpers to accept page pointers, which is another hot spot which uses page pointer a lot.

- Migrate btrfs bio endio functions to utilize bio_for_each_folio_all() This completely removes the ability to direct access page pointers. Although we still need some extra folio_page(folio, 0) calls to keep compatible with existing helpers.

And since we're here, also fix the choas of btrfs endio functions' naming scheme, now it would always be:

    end_bbio_<target>_(read|write)

The <target> can be:

- data
For non-compressed and non-encoeded operations

- compressed
For compressed IO and encoded IO.

- meta

And since compressed IO path is utilizing unmapped pages (pages without an address_space), thus they don't touch the page flags. This makes compressed IO path a very good test bed for the initial introduction of higher order folio.

Qu Wenruo (3):
btrfs: migrate get_eb_page_index() and get_eb_offset_in_page() to
folios
btrfs: migrate subpage code to folio interfaces
btrfs: migrate various btrfs end io functions to folios

 fs/btrfs/accessors.c        |  80 ++++----
 fs/btrfs/compression.c      |  15 +-
 fs/btrfs/ctree.c            |  13 +-
 fs/btrfs/defrag.c           |   3 +-
 fs/btrfs/disk-io.c          |   4 +-
 fs/btrfs/extent_io.c        | 363 +++++++++++++++++++-----------------
 fs/btrfs/extent_io.h        |  40 ++--
 fs/btrfs/file.c             |  13 +-
 fs/btrfs/free-space-cache.c |   4 +-
 fs/btrfs/inode.c            |  34 ++--
 fs/btrfs/ordered-data.c     |   5 +-
 fs/btrfs/reflink.c          |   6 +-
 fs/btrfs/relocation.c       |   5 +-
 fs/btrfs/subpage.c          | 304 ++++++++++++++----------------
 fs/btrfs/subpage.h          |  74 ++++----
 15 files changed, 500 insertions(+), 463 deletions(-)
Signature · 1 line
2.43.0