[0/2] iomap: flush dirty cache over unwritten mappings on zero range

Message

Brian Foster Aug. 22, 2024, 2:59 p.m. UTC

Hi all,

This is the alternative flushing solution to the iomap zero range
problem with dirty pagecache over unwritten mappings. This is done in
two steps. Patch 1 lifts the XFS workaround into iomap, flushing the
range unconditionally and providing an easily backportable fix for
stable kernels. Patch 2 buries the flush further down into iomap, making
it conditional on the combined presence of dirty cache and unwritten
mappings in the target range. This may be reasonable backportable as
well, but is only required if performance is a concern.

I still have to look into the improved revalidation approach discussed
in the RFC thread, but given that requires validation support and this
is intended to be a generic fallback, I wanted to get this nailed down
first.

fstests coverage for this problem is posted here [1]. Thoughts, reviews,
flames appreciated.

Brian

v1:
- Alternative approach, flush instead of revalidate.
rfc: https://lore.kernel.org/linux-fsdevel/[e-mail]/

[1] https://lore.kernel.org/fstests/[e-mail]/

Brian Foster (2):
  iomap: fix handling of dirty folios over unwritten extents
  iomap: make zero range flush conditional on unwritten mappings

 fs/iomap/buffered-io.c | 52 ++++++++++++++++++++++++++++++++++++++----
 fs/xfs/xfs_iops.c      | 10 --------
 2 files changed, 48 insertions(+), 14 deletions(-)