Re: [PATCH] nfsd4: handle failure to find backchannel

["J. Bruce Fields" <bfields@fieldses.org>]

On Thu, Jul 17, 2014 at 09:24:13PM +0800, Kinglong Mee wrote:
> On 7/17/2014 06:05, J. Bruce Fields wrote:
> > On Sat, Jul 12, 2014 at 10:02:17PM +0800, Kinglong Mee wrote:
> >> On 7/12/2014 05:16, J. Bruce Fields wrote:
> >>> The local variable "ses" will be left NULL here in the case we fail to
> >>> find a connection.  Spotted by a coverity scan.
> >>>     
> >>> Signed-off-by: J. Bruce Fields <bfields@redhat.com>
> >>>
> >>> diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
> >>> index 2c73cae9899d..fe22cd5c42d3 100644
> >>> --- a/fs/nfsd/nfs4callback.c
> >>> +++ b/fs/nfsd/nfs4callback.c
> >>> @@ -1001,14 +1001,18 @@ static void nfsd4_process_cb_update(struct nfsd4_callback *cb)
> >>>  	}
> >>>  	spin_unlock(&clp->cl_lock);
> >>>  
> >>> +	if (!c)
> >>> +		goto out_no_connection;
> >>
> >> Setting err to -EINVAL maybe better.
> >> Otherwise, nfsd4_mark_cb_down will be called with err == 0.
> >>
> >>>  	err = setup_callback_client(clp, &conn, ses);
> >>
> >> setup_callback_client also return -EINVAL when ses == NULL with conn.cb_xprt == NULL.
> > 
> > Thanks, yes, after looking over this carefully I don't believe we can
> > call setup_callback_client with ses NULL but conn->cb_xprt non-NULL, so
> > this is just a false positive from coverity.
> 
> ses and conn->cb_xprt will be set in the same condition before
> calling setup_callback_client,
> 
>  996         c = __nfsd4_find_backchannel(clp);
>  997         if (c) {
>  998                 svc_xprt_get(c->cn_xprt);
>  999                 conn.cb_xprt = c->cn_xprt;
> 1000                 ses = c->cn_session;
> 1001         }
> 1002         spin_unlock(&clp->cl_lock);
> 1003
> 1004         err = setup_callback_client(clp, &conn, ses);
> 
> ses and conn->cb_xprt will be NULL or non-NULL in the same time,
> so that, call setup_calback_client with ses NULL but conn->cb_xprt non-NULL will not appear.

Agreed but you also need to check what happens in the case where c is
NULL.

This is much more confusing than necessary.  But I'm inclined to hold
off on any cleanup here while jlayton's patches are still pending.

--b.

> 
> thanks,
> Kinglong Mee
> 
> >>> -	if (err) {
> >>> -		nfsd4_mark_cb_down(clp, err);
> >>> -		return;
> >>> -	}
> >>> +	if (err)
> >>> +		goto out_no_connection;
> >>>  	/* Yay, the callback channel's back! Restart any callbacks: */
> >>>  	list_for_each_entry(cb, &clp->cl_callbacks, cb_per_client)
> >>>  		run_nfsd4_cb(cb);
> >>> +	return;
> >>> +out_no_connection:
> >>> +	nfsd4_mark_cb_down(clp, err);
> >>> +	return;
> >>>  }
> >>>  
> >>>  static void nfsd4_do_callback_rpc(struct work_struct *w)
> >>> --
> >>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> >>> the body of a message to majordomo@vger.kernel.org
> >>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >>>
> >