[PATCH 0/3] pNFS/flexfiles: honor clora_changed and report cancelled I/O

[Benjamin Coddington <ben.coddington@hammerspace.com>]

RFC 8881 section 20.3.3 lets CB_LAYOUTRECALL carry clora_changed: when it
is FALSE the client may flush modified data to the data servers before
returning the layout; only when it is TRUE should it stop writing to the
DS and redirect through the MDS.  The client decodes cbl_layoutchanged but
has ignored it since commit b739a5bd9d9f ("NFSv4/flexfiles: Cancel I/O if
the layout is recalled or revoked"), cancelling in-flight DS I/O
unconditionally -- which can leave a write incomplete at the DS with no
write layout held.

This series teaches the client to honor clora_changed: on an unchanged
recall it drains its in-flight DS writes before LAYOUTRETURN instead of
cancelling them; when it does cancel (changed recall, or revoke) the
cancelled I/O is reported to the server as NFS4ERR_NXIO so it can
reconcile the affected mirror instance.  Patch 1 adds clora_changed to the
recall tracepoint.

Based on v7.1.

Benjamin Coddington (3):
  pNFS: report clora_changed in the cb_layoutrecall_file tracepoint
  pNFS: honor clora_changed when recalling a layout
  NFSv4/flexfiles: report cancelled I/O as a layout error

 fs/nfs/callback_proc.c                 |  5 ++-
 fs/nfs/flexfilelayout/flexfilelayout.c | 23 ++++++++++-
 fs/nfs/nfs4trace.h                     | 55 +++++++++++++++++++++++++-
 fs/nfs/pnfs.c                          | 21 +++++-----
 fs/nfs/pnfs.h                          |  2 +-
 5 files changed, 92 insertions(+), 14 deletions(-)

-- 
2.53.0