mbox series

[v7,00/16] netfs, cifs: Delegate high-level I/O to netfslib

Message ID 20240430140930.262762-1-dhowells@redhat.com (mailing list archive)
Headers show
Series netfs, cifs: Delegate high-level I/O to netfslib | expand

Message

David Howells April 30, 2024, 2:09 p.m. UTC
Hi Steve,

Here are patches to convert cifs to use the netfslib library.  I've tested
them with and without a cache.  There appears to be a signifcant
performance improvement in buffered writeback (around 50% throughput rate
with fio tests).

The patches remove around 2000 lines from CIFS.

Notes:

 (1) CIFS is made to use unbuffered I/O for unbuffered caching modes and
     write-through caching for cache=strict.

 (2) Various cifs fallocate() function implementations have issues that
     aren't easily fixed without enhanced protocol support.

 (3) It should be possible to turn on multipage folio support in CIFS now.

 (4) The then-unused CIFS code is removed in three patches, not one, to
     avoid the git patch generator from producing confusing patches in
     which it thinks code is being moved around rather than just being
     removed.

The patches can be found here also:

	https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=cifs-netfs

Changes
=======
ver #7)
 - Rebased on -rc6 to absorb some fixes that have gone upstream, including
   a fix for a leak of the fscache volume cookie.
 - Added a patch to turn on multipage folio support in cifs.

ver #6)
 - The netfslib write helpers got rewritten and, consequently, some changes
   were made here.
 - Rearranged the patch order a little to put the smaller changes first.
 - Use a different way to invalidate that doesn't use ->launder_folio().
 - Attempt to open the handle on the server READ+WRITE, even if the user
   asks for O_WRONLY, if the mount was with "-o fsc" so that we can fill in
   the gaps in partial folio writes; if we can't fall back to WRITE-only
   and disable caching.
 - Fixed the cookie key to match the key used by iget5_locked() to avoid
   "Duplicate cookie" errors.
 - Made add_credits_and_wake_if() clear the number of returned credits,
   allowing it to be called multiple times in cleanup/error handling.
 - Made ->async_writev() not return an error directly, but always set it on
   the subreq.
 - Fixed error handling from writethrough writing.

ver #5)
 - Rebased to -rc3 plus SteveF's for-next branch as netfslib is now
   upstream, as are a couple of patches from this series.
 - Replace the ->replay bool Shyam added with a flag on the netfs
   subrequest.  This is tested by the code, but not currently set (see
   above).

ver #4)
 - Slimmed down the branch:
   - Split the cifs-related patches off to a separate branch (cifs-netfs)
   - Deferred the content-encryption to the in-progress ceph changes.
   - Deferred the use-PG_writeback rather than PG_fscache patch
 - Rebased on a later linux-next with afs-rotation patches.

ver #3)
 - Moved the fscache module into netfslib to avoid export cycles.
 - Fixed a bunch of bugs.
 - Got CIFS to pass as much of xfstests as possible.
 - Added a patch to make 9P use all the helpers.
 - Added a patch to stop using PG_fscache, but rather dirty pages on
   reading and have writepages write to the cache.

ver #2)
 - Folded the addition of NETFS_RREQ_NONBLOCK/BLOCKED into first patch that
   uses them.
 - Folded addition of rsize member into first user.
 - Don't set rsize in ceph (yet) and set it in kafs to 256KiB.  cifs sets
   it dynamically.
 - Moved direct_bv next to direct_bv_count in struct netfs_io_request and
   labelled it with a __counted_by().
 - Passed flags into netfs_xa_store_and_mark() rather than two bools.
 - Removed netfs_set_up_buffer() as it wasn't used.

David

Link: https://lore.kernel.org/r/20231213152350.431591-1-dhowells@redhat.com/ [1]
Link: https://lore.kernel.org/r/20231013160423.2218093-1-dhowells@redhat.com/ # v1
Link: https://lore.kernel.org/r/20231117211544.1740466-1-dhowells@redhat.com/ # v2
Link: https://lore.kernel.org/r/20231207212206.1379128-1-dhowells@redhat.com/ # v3
Link: https://lore.kernel.org/r/20231213154139.432922-1-dhowells@redhat.com/ # v4
Link: https://lore.kernel.org/r/20240205225726.3104808-1-dhowells@redhat.com/ # v5

David Howells (16):
  cifs: Use alternative invalidation to using launder_folio
  cifs: Replace cifs_readdata with a wrapper around netfs_io_subrequest
  cifs: Replace cifs_writedata with a wrapper around netfs_io_subrequest
  cifs: Use more fields from netfs_io_subrequest
  cifs: Make wait_mtu_credits take size_t args
  cifs: Replace the writedata replay bool with a netfs sreq flag
  cifs: Move cifs_loose_read_iter() and cifs_file_write_iter() to file.c
  cifs: Set zero_point in the copy_file_range() and remap_file_range()
  cifs: Add mempools for cifs_io_request and cifs_io_subrequest structs
  cifs: Make add_credits_and_wake_if() clear deducted credits
  cifs: Implement netfslib hooks
  cifs: Cut over to using netfslib
  cifs: Remove some code that's no longer used, part 1
  cifs: Remove some code that's no longer used, part 2
  cifs: Remove some code that's no longer used, part 3
  cifs: Enable large folio support

 fs/netfs/buffered_write.c    |    6 +
 fs/netfs/io.c                |    7 +-
 fs/smb/client/Kconfig        |    1 +
 fs/smb/client/cifsfs.c       |  124 +-
 fs/smb/client/cifsfs.h       |   11 +-
 fs/smb/client/cifsglob.h     |   65 +-
 fs/smb/client/cifsproto.h    |   12 +-
 fs/smb/client/cifssmb.c      |  120 +-
 fs/smb/client/file.c         | 2898 ++++++----------------------------
 fs/smb/client/fscache.c      |  109 --
 fs/smb/client/fscache.h      |   54 -
 fs/smb/client/inode.c        |   45 +-
 fs/smb/client/smb2ops.c      |   10 +-
 fs/smb/client/smb2pdu.c      |  186 ++-
 fs/smb/client/smb2proto.h    |    5 +-
 fs/smb/client/trace.h        |  144 +-
 fs/smb/client/transport.c    |   17 +-
 include/linux/netfs.h        |    1 +
 include/trace/events/netfs.h |    1 +
 19 files changed, 886 insertions(+), 2930 deletions(-)

Comments

David Howells April 30, 2024, 2:47 p.m. UTC | #1
Hi Christian,

With Steve's agreement, could you pick this set of patches up also?

Thanks,
David
Steve French April 30, 2024, 2:58 p.m. UTC | #2
Yes - I think it is easier for now if Christian picks up the series in
his tree due to the VFS dependencies - if we start hitting merge
conflicts with my tree later we can consider changing that.

On Tue, Apr 30, 2024 at 9:47 AM David Howells <dhowells@redhat.com> wrote:
>
> Hi Christian,
>
> With Steve's agreement, could you pick this set of patches up also?
>
> Thanks,
> David
>
Christian Brauner May 2, 2024, 2:13 p.m. UTC | #3
On Tue, Apr 30, 2024 at 03:47:43PM +0100, David Howells wrote:
> Hi Christian,
> 
> With Steve's agreement, could you pick this set of patches up also?

Pulled both branches from you. Thank you!