Received-SPF: pass (google.com: domain of linux-nfs+bounces-2925-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
From: Dai Ngo <dai.ngo@oracle.com>
To: chuck.lever@oracle.com, jlayton@kernel.org
Cc: linux-nfs@vger.kernel.org
Subject: PATCH [v2 0/3] NFSD: drop TCP connections when NFSv4 client enters courtesy state
Date: Mon, 22 Apr 2024 20:12:30 -0700
Message-Id: <1713841953-19594-1-git-send-email-dai.ngo@oracle.com>
Precedence: bulk

When a v4.0 client enters courtesy state all its v4 states remain valid
and its fore and back channel TCP connection remained in ESTABLISHED
state until the TCP keep-alive mechanism timed out and shuts down the
back channel connection. The fore channel connection remains in ESTABLISHED
state between 6 - 12 minutes before the NFSv4 server's 6-minute idle timer
(svc_age_temp_xprts) shuts down the idle connection.

Since NFSv4.1 mount uses the same TCP connection for both fore and back
channel connection there is no TCP keep-alive packet sent from the server
to the client. The server's idle timer does not shutdown an idle v4.1
connection since the svc_xprt->xpt_ref is more than 1:  1 for sv_tempsocks
list, one for the session's nfsd4_conn and 1 for the back channel.
 
To conserve system resources in large configuration where there are lots
of idle clients, this patch series drop the fore and back channel connection
of NFSv4 client as soon as it enters the courtesy state. The fore and back
channel connections are automatically re-established when the courtesy
client reconnects.

v1 - v2:
[patch 3/3]: a more reliable way to check if the connection needs to be dropped. 

 fs/nfsd/nfs4callback.c | 14 ++++++++++++--
 fs/nfsd/nfs4state.c    | 32 ++++++++++++++++++++++++++++++--
 fs/nfsd/state.h        |  2 ++
 3 files changed, 44 insertions(+), 4 deletions(-)