[v3,0/7] large atomic writes for xfs

Message

John Garry Jan. 2, 2025, 2:04 p.m. UTC

Currently the atomic write unit min and max is fixed at the FS blocksize
for xfs and ext4.

This series expands support to allow multiple FS blocks to be written
atomically.

To allow multiple blocks be written atomically, the fs must ensure blocks
are allocated with some alignment and granularity. For xfs, today only
rtvol provides this through rt_extsize. So initial support for large
atomic writes will be for rtvol here. Support can easily be expanded to
regular files through the proposed forcealign feature.

An atomic write which spans mixed unwritten and mapped extents will be
required to have the unwritten extents pre-zeroed, which will be supported
in iomap.

Based on bf354410af83 ("Merge tag 'xfs-6.13-fixes_2024-12-12' of
https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into next-rc")

Patches available at the following:
https://github.com/johnpgarry/linux/tree/atomic-write-large-atomics-v6.13-v3

Changes since v2:
- Don't zero unwritten for single block atomic write
- Store RT atomic write unit max in FSBs

Changes since v1:
- Add extent zeroing support
- Rebase

John Garry (6):
  iomap: Increase iomap_dio_zero() size limit
  iomap: Add zero unwritten mappings dio support
  xfs: Add extent zeroing support for atomic writes
  xfs: Switch atomic write size check in xfs_file_write_iter()
  xfs: Add RT atomic write unit max to xfs_mount
  xfs: Update xfs_get_atomic_write_attr() for large atomic writes

Ritesh Harjani (IBM) (1):
  iomap: Lift blocksize restriction on atomic writes

 fs/iomap/direct-io.c   | 100 +++++++++++++++++++++++++++++++++++---
 fs/xfs/libxfs/xfs_sb.c |   3 ++
 fs/xfs/xfs_file.c      | 108 ++++++++++++++++++++++++++++++++++++++---
 fs/xfs/xfs_iops.c      |  21 +++++++-
 fs/xfs/xfs_iops.h      |   2 +
 fs/xfs/xfs_mount.h     |   1 +
 fs/xfs/xfs_rtalloc.c   |  23 +++++++++
 fs/xfs/xfs_rtalloc.h   |   4 ++
 include/linux/iomap.h  |   3 ++
 9 files changed, 248 insertions(+), 17 deletions(-)