Re: [PATCH v2] SUNRPC: check current nsproxy before set of node nameon client creation

[Stanislav Kinsbursky]

08.09.2012 18:33, Myklebust, Trond ÐÐÑÐÑ:
> On Sat, 2012-09-08 at 08:59 +0300, Stanislav Kinsbursky wrote:
> > 08.09.2012 01:32, Myklebust, Trond ÐÐÑÐÑ:
> > > On Mon, 2012-08-13 at 08:10 -0400, Jeff Layton wrote:
> > > > On Mon, 13 Aug 2012 15:37:31 +0400
> > > > Stanislav Kinsbursky <skinsbursky@xxxxxxxxxxxxx> wrote:
> > > >
> > > > > v2:
> > > > > 1) rpc_clnt_set_nodename() prototype updated.
> > > > > 2) fixed errors in comment.
> > > > >
> > > > > When child reaper exits, it can destroy mount namespace it belongs to, and if
> > > > > there are NFS mounts inside, then it will try to umount them. But in this
> > > > > point current->nsproxy is set to NULL and all namespaces will be destroyed one
> > > > > by one. I.e. we can't dereference current->nsproxy to obtain uts namespace.
> > > > >
> > > > > Signed-off-by: Stanislav Kinsbursky <skinsbursky@xxxxxxxxxxxxx>
> > > > > ---
> > > > >    net/sunrpc/clnt.c |   16 +++++++++++++---
> > > > >    1 files changed, 13 insertions(+), 3 deletions(-)
> > > > >
> > > > > diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
> > > > > index 9a9676e..8fbcbc8 100644
> > > > > --- a/net/sunrpc/clnt.c
> > > > > +++ b/net/sunrpc/clnt.c
> > > > > @@ -277,8 +277,18 @@ void rpc_clients_notifier_unregister(void)
> > > > >    	return rpc_pipefs_notifier_unregister(&rpc_clients_block);
> > > > >    }
> > > > >
> > > > > -static void rpc_clnt_set_nodename(struct rpc_clnt *clnt, const char *nodename)
> > > > > +static void rpc_clnt_set_nodename(struct rpc_clnt *clnt)
> > > > >    {
> > > > > +	const char *nodename;
> > > > > +
> > > > > +	/*
> > > > > +	 * We have to protect against dying child reaper, which has released
> > > > > +	 * its nsproxy already and is trying to destroy mount namespace.
> > > > > +	 */
> > > > > +	if (current->nsproxy == NULL)
> > > > > +		return;
> > > > > +
> > > > > +	nodename = utsname()->nodename;
> > > > >    	clnt->cl_nodelen = strlen(nodename);
> > > > >    	if (clnt->cl_nodelen > UNX_MAXNODENAME)
> > > > >    		clnt->cl_nodelen = UNX_MAXNODENAME;
> > > > > @@ -365,7 +375,7 @@ static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args, stru
> > > > >    	}
> > > > >
> > > > >    	/* save the nodename */
> > > > > -	rpc_clnt_set_nodename(clnt, utsname()->nodename);
> > > > > +	rpc_clnt_set_nodename(clnt);
> > > > >    	rpc_register_client(clnt);
> > > > >    	return clnt;
> > > > >
> > > > > @@ -524,7 +534,7 @@ rpc_clone_client(struct rpc_clnt *clnt)
> > > > >    	err = rpc_setup_pipedir(new, clnt->cl_program->pipe_dir_name);
> > > > >    	if (err != 0)
> > > > >    		goto out_no_path;
> > > > > -	rpc_clnt_set_nodename(new, utsname()->nodename);
> > > > > +	rpc_clnt_set_nodename(new);
> > > > >    	if (new->cl_auth)
> > > > >    		atomic_inc(&new->cl_auth->au_count);
> > > > >    	atomic_inc(&clnt->cl_count);
> > > > >
> > > > > --
> > > > > To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> > > > > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > > > > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > > > Acked-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > OK, colour me confused (again).
> >
> > What color?
> >
> > > Why should a umount trigger an
> > > rpc_create() or rpc_clone_client()?
> >
> > It calls nsm_create().
> > Here is the trace (https://bugzilla.redhat.com/show_bug.cgi?id=830862,
> > comment 68):
>
> Right, but if we're using NFSv3 lock monitoring, we know in advance that
> we're going to need an nsm call to localhost. Why can't we just cache
> the one that we used to start lock monitoring in the first place?

Do you suggest to cache the call or the client for the call?

--
Best regards,
Stanislav Kinsbursky
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/