mbox series

[0/7] Start moving write_begin/write_end out of aops

Message ID 20240528164829.2105447-1-willy@infradead.org (mailing list archive)
Headers show
Series Start moving write_begin/write_end out of aops | expand

Message

Matthew Wilcox May 28, 2024, 4:48 p.m. UTC
Christoph wants to remove write_begin/write_end from aops and pass them
to filemap as callback functions.  Here's one possible route to do this.
I combined it with the folio conversion (because why touch the same code
twice?) and tweaked some of the other things (support for ridiculously
large folios with size_t lengths, remove the need to initialise fsdata
by passing only a pointer to the fsdata pointer).  And then I converted
ext4, which is probably the worst filesystem to convert because it needs
three different bwops.  Most fs will only need one.

Not written yet: convert all the other fs, remove wrappers.

v2:
 - Redo how we pass fsdata around so it can persist across multiple
   invocations of filemap_perform_write()
 - Add ext2
 - Minor tweak to iomap

This is against 2bfcfd584ff5 (Linus current head) and will conflict with
other patches in flight.

Matthew Wilcox (Oracle) (7):
  fs: Introduce buffered_write_operations
  fs: Supply optional buffered_write_operations in buffer.c
  buffer: Add buffer_write_begin, buffer_write_end and
    __buffer_write_end
  fs: Add filemap_symlink()
  ext2: Convert to buffered_write_operations
  ext4: Convert to buffered_write_operations
  iomap: Return the folio from iomap_write_begin()

 Documentation/filesystems/locking.rst |  23 ++++
 fs/buffer.c                           | 158 ++++++++++++++++++--------
 fs/ext2/ext2.h                        |   1 +
 fs/ext2/file.c                        |   4 +-
 fs/ext2/inode.c                       |  55 ++++-----
 fs/ext2/namei.c                       |   2 +-
 fs/ext4/ext4.h                        |  24 ++--
 fs/ext4/file.c                        |  12 +-
 fs/ext4/inline.c                      |  66 +++++------
 fs/ext4/inode.c                       | 134 ++++++++++------------
 fs/iomap/buffered-io.c                |  35 +++---
 fs/jfs/file.c                         |   3 +-
 fs/namei.c                            |  25 ++++
 fs/ramfs/file-mmu.c                   |   3 +-
 fs/ufs/file.c                         |   2 +-
 include/linux/buffer_head.h           |  28 ++++-
 include/linux/fs.h                    |   3 -
 include/linux/pagemap.h               |  23 ++++
 mm/filemap.c                          |  77 ++++++++-----
 19 files changed, 417 insertions(+), 261 deletions(-)

Comments

Christoph Hellwig May 29, 2024, 5:20 a.m. UTC | #1
On Tue, May 28, 2024 at 05:48:21PM +0100, Matthew Wilcox (Oracle) wrote:
> Christoph wants to remove write_begin/write_end from aops and pass them
> to filemap as callback functions.  Here's one possible route to do this.
> I combined it with the folio conversion (because why touch the same code
> twice?) and tweaked some of the other things (support for ridiculously
> large folios with size_t lengths, remove the need to initialise fsdata
> by passing only a pointer to the fsdata pointer).  And then I converted
> ext4, which is probably the worst filesystem to convert because it needs
> three different bwops.  Most fs will only need one.

Hopefully ext4 will get convert to iomap before we need this.. :)

More seriously, there is an ext4 iomap conversion in progress and a
ext2 one, which is a really good copy & paste model for a lot of the
simple file systems.  Maybe just wait for some of this to settle
to avoid a lot of duplicate work?