Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.36 as permitted sender) client-ip=23.128.96.36;
From:   Benjamin Coddington <bcodding@redhat.com>
To:     trond.myklebust@hammerspace.com, anna@kernel.org
Cc:     linux-nfs@vger.kernel.org
Subject: [PATCH 0/2] NFSv4 READDIR d_type fixup
Date:   Tue, 17 Oct 2023 17:30:42 -0400
Message-ID: <cover.1697577945.git.bcodding@redhat.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

This is a combined posting of two previously posted patches.  The first
unconditionally adds the type attribute to the list of requested attributes
for v4 READDIR.  The second patch enables a per-mount modification via sysfs
of any of the attributes the client will currently decode for v4 READDIR.

The first patch solves a real problem but may cause a performance regression
for some servers that require extra processing to return inode information
along with the namespace.  We performed an informal survey of most of the
major NFSv4 server vendors and although we did not learn of any that may be
impacted, the potential remains.

The second patch gives us a way to disable this new READDIR behavior should
we find a serious impact in an existing setup.  I would appreciate serious
consideration of this patch in light of the number of claimed performance
regressions that have been reported almost every time we touch this
sensitive operation on the client.

For example:

echo 0x800 0x800000 0x0 > /sys/fs/nfs/0\:57/v4_readdir_attrs

.. will revert the behavior change from patch 1.

Benjamin Coddington (2):
  NFSv4: Always ask for type with READDIR
  NFSv4: Allow per-mount tuning of READDIR attrs

 fs/nfs/client.c           |  2 +
 fs/nfs/nfs4client.c       |  4 ++
 fs/nfs/nfs4proc.c         |  1 +
 fs/nfs/nfs4xdr.c          |  9 ++---
 fs/nfs/sysfs.c            | 81 +++++++++++++++++++++++++++++++++++++++
 include/linux/nfs_fs_sb.h |  1 +
 include/linux/nfs_xdr.h   |  1 +
 7 files changed, 94 insertions(+), 5 deletions(-)

-- 
2.41.0