| Message ID | 20240613050055.854323-1-trond.myklebust@hammerspace.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | LAYOUTRETURN on reboot | expand |
On Thu, 2024-06-13 at 01:00 -0400, trondmy@gmail.com wrote: > From: Trond Myklebust <trond.myklebust@hammerspace.com> > > Now that https://datatracker.ietf.org/doc/draft-ietf-nfsv4-layrec/ is > mostly done with the review process, I'd like to move the final patches > for the client implementation upstream. > > The following patch series therefore adds support to the flexfiles pNFS > driver so that if a metadata server reboot occurs while a client has > layouts outstanding, and is performing I/O, then the client will report > layoutstats and layout errors through a LAYOUTRETURN during the grace > period, after the metadata server comes back up. > This has implications for mirrored workloads, since it allows the client > to report exactly which mirror data instances may have been corrupted > due to the presence of errors during WRITEs or COMMITs. > > Trond Myklebust (11): > NFSv4/pnfs: Remove redundant list check > NFSv4.1: constify the stateid argument in nfs41_test_stateid() > NFSv4: Clean up encode_nfs4_stateid() > pNFS: Add a flag argument to pnfs_destroy_layouts_byclid() > NFSv4/pnfs: Add support for the PNFS_LAYOUT_FILE_BULK_RETURN flag > NFSv4/pNFS: Add a helper to defer failed layoutreturn calls > NFSv4/pNFS: Handle server reboots in pnfs_poc_release() > NFSv4/pNFS: Retry the layout return later in case of a timeout or > reboot > NFSv4/pnfs: Give nfs4_proc_layoutreturn() a flags argument > NFSv4/pNFS: Remove redundant call to unhash the layout > NFSv4/pNFS: Do layout state recovery upon reboot > > fs/nfs/callback_proc.c | 5 +- > fs/nfs/flexfilelayout/flexfilelayout.c | 2 +- > fs/nfs/nfs4_fs.h | 3 +- > fs/nfs/nfs4proc.c | 53 ++++-- > fs/nfs/nfs4state.c | 4 +- > fs/nfs/nfs4xdr.c | 7 +- > fs/nfs/pnfs.c | 223 +++++++++++++++++++------ > fs/nfs/pnfs.h | 30 +++- > include/linux/nfs_fs_sb.h | 1 + > include/linux/nfs_xdr.h | 2 +- > 10 files changed, 249 insertions(+), 81 deletions(-) > These have been used for a while inside of Meta vs. Hammerspace's servers and have been behaving. Reviewed-by: Jeff Layton <jlayton@kernel.org>
From: Trond Myklebust <trond.myklebust@hammerspace.com> Now that https://datatracker.ietf.org/doc/draft-ietf-nfsv4-layrec/ is mostly done with the review process, I'd like to move the final patches for the client implementation upstream. The following patch series therefore adds support to the flexfiles pNFS driver so that if a metadata server reboot occurs while a client has layouts outstanding, and is performing I/O, then the client will report layoutstats and layout errors through a LAYOUTRETURN during the grace period, after the metadata server comes back up. This has implications for mirrored workloads, since it allows the client to report exactly which mirror data instances may have been corrupted due to the presence of errors during WRITEs or COMMITs. Trond Myklebust (11): NFSv4/pnfs: Remove redundant list check NFSv4.1: constify the stateid argument in nfs41_test_stateid() NFSv4: Clean up encode_nfs4_stateid() pNFS: Add a flag argument to pnfs_destroy_layouts_byclid() NFSv4/pnfs: Add support for the PNFS_LAYOUT_FILE_BULK_RETURN flag NFSv4/pNFS: Add a helper to defer failed layoutreturn calls NFSv4/pNFS: Handle server reboots in pnfs_poc_release() NFSv4/pNFS: Retry the layout return later in case of a timeout or reboot NFSv4/pnfs: Give nfs4_proc_layoutreturn() a flags argument NFSv4/pNFS: Remove redundant call to unhash the layout NFSv4/pNFS: Do layout state recovery upon reboot fs/nfs/callback_proc.c | 5 +- fs/nfs/flexfilelayout/flexfilelayout.c | 2 +- fs/nfs/nfs4_fs.h | 3 +- fs/nfs/nfs4proc.c | 53 ++++-- fs/nfs/nfs4state.c | 4 +- fs/nfs/nfs4xdr.c | 7 +- fs/nfs/pnfs.c | 223 +++++++++++++++++++------ fs/nfs/pnfs.h | 30 +++- include/linux/nfs_fs_sb.h | 1 + include/linux/nfs_xdr.h | 2 +- 10 files changed, 249 insertions(+), 81 deletions(-)