Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 5 May 2023 01:23:21 +0200
From:   Petr Vorel <pvorel@suse.cz>
To:     Cyril Hrubis <chrubis@suse.cz>, ltp@lists.linux.it,
        NeilBrown <neilb@suse.de>, linux-nfs@vger.kernel.org,
        Jeff Layton <jlayton@kernel.org>
Subject: Re: [PATCH v5 5/5] nfs: Run on btrfs, ext4, xfs
Message-ID: <20230504232321.GA25208@pevik>
Reply-To: Petr Vorel <pvorel@suse.cz>
References: <20230504131414.3826283-1-pvorel@suse.cz>
 <20230504131414.3826283-6-pvorel@suse.cz>
 <ZFO6ywouPkmNKtkr@yuki>
 <20230504220037.GB4244@pevik>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230504220037.GB4244@pevik>
Precedence: bulk

Hi all,

...
> Tomorrow I'll test this:

> check_umount()
> {
> 	local i
> 	local dir="$1"
> 	local type="${2:-lhost}"
> 	local cmd="grep -q $dir /proc/mounts"

> 	for i in $(seq 50); do
> 		if [ "$type" = "lhost" ]; then
> 			$cmd || return
> 		else
> 			tst_rhost_run -c "$cmd" || return
> 		fi
> 		tst_sleep 100ms
> 		tst_res TWARN "failed to umount '$dir'"
> 	done
> }

> 	for i in $VERSION; do
> 		local_dir="$(get_local_dir $i $n)"
> 		grep -q "$local_dir" /proc/mounts && umount $local_dir

> 		# instead of 'sleep 2' below check here:
> 		check_umount "$local_dir"

> 		n=$(( n + 1 ))
> 	done

> 	n=0
> 	for i in $VERSION; do
> 		type=$(get_socket_type $n)
> 		remote_dir="$(get_remote_dir $i $type)"

> 		if tst_net_use_netns; then
> 			if test -d $remote_dir; then
> 				exportfs -u *:$remote_dir
> 				check_umount "$remote_dir" # instead of sleep 1
> 				rm -rf $remote_dir
> 			fi
> 		else
> 			tst_rhost_run -c "test -d $remote_dir && exportfs -u *:$remote_dir"
> 			check_umount "$remote_dir" rhost # instead of sleep 1
> 			tst_rhost_run -c "test -d $remote_dir && rm -rf $remote_dir"
> 		fi

> 		n=$(( n + 1 ))
> 	done

I tried various code (not only the one - wrong - above), but nothing works.
With fuser -vm, I wonder if there is some bug in older nfs versions:

                     USER        PID ACCESS COMMAND
/tmp/LTP_nfs08.fncaxJum2W/mntpoint/4/tcp:
                     root     kernel mount /tmp/LTP_nfs08.fncaxJum2W/mntpoint
                     root     kernel knfsd /tmp/LTP_nfs08.fncaxJum2W/mntpoint/4

Looking at c6c7f2a84da4 ("nfsd: Ensure knfsd shuts down when the "nfsd" pseudofs
is unmounted"), maybe there are other bugs, which causes files are used.

I also tried to use /proc/fs/nfsd/unlock_filesystem and
/proc/fs/nfsd/write_unlock_ip which made things better a bit (6.2.12), but still
it's not 100% solution. Also, "Release all locks on a local file system"
(unlock_filesystem) and Release all locks used by a client (write_unlock_ip) are
unlikely what I'm searching.

		umount $local_dir
		echo "$local_dir" > /proc/fs/nfsd/unlock_filesystem
		# $(tst_ipaddr rhost) => IP of the server
		echo "$(tst_ipaddr rhost)" > /proc/fs/nfsd/write_unlock_ip
		...

		exportfs -u *:$remote_dir
		echo "$remote_dir" > /proc/fs/nfsd/unlock_filesystem

=> any hint is welcome.

Kind regards,
Petr