Re: [PATCH 4.4 014/190] lockd: fix lockd shutdown race

[Ben Hutchings]

On Wed, 2018-04-11 at 20:34 +0200, Greg Kroah-Hartman wrote:
> 4.4-stable review patch.ÂÂIf anyone has any objections, please let me know.
> 
> ------------------
> 
> From: "J. Bruce Fields" <bfields@xxxxxxxxxx>
> 
> 
> [ Upstream commit efda760fe95ea15291853c8fa9235c32d319cd98 ]
[...]
> --- a/fs/lockd/svc.c
> +++ b/fs/lockd/svc.c
> @@ -129,6 +129,8 @@ lockd(void *vrqstp)
> Â{
> Â	int		err = 0;
> Â	struct svc_rqst *rqstp = vrqstp;
> +	struct net *net = &init_net;
> +	struct lockd_net *ln = net_generic(net, lockd_net_id);
> Â
> Â	/* try_to_freeze() is called from svc_recv() */
> Â	set_freezable();
> @@ -173,6 +175,8 @@ lockd(void *vrqstp)
> Â	if (nlmsvc_ops)
> Â		nlmsvc_invalidate_all();
>Â	nlm_shutdown_hosts();
> +	cancel_delayed_work_sync(&ln->grace_period_end);
> +	locks_end_grace(&ln->lockd_manager);
> Â	return 0;
> Â}
> Â
> @@ -267,8 +271,6 @@ static void lockd_down_net(struct svc_se
> Â	if (ln->nlmsvc_users) {
> Â		if (--ln->nlmsvc_users == 0) {
> Â			nlm_shutdown_hosts_net(net);
> -			cancel_delayed_work_sync(&ln->grace_period_end);
> -			locks_end_grace(&ln->lockd_manager);

Aren't these statements still needed for non-initial namespaces?

Ben.

> Â			svc_shutdown_net(serv, net);
> Â			dprintk("lockd_down_net: per-net data destroyed; net=%p\n", net);
> Â		}
> 
> 
> 
-- 
Ben Hutchings
Software Developer, Codethink Ltd.