From: Neil Brown <neilb@suse.de>
Subject: Confused about BUG_ON in rpcb_getport_async
Date: Tue, 12 Aug 2008 12:27:22 +1000
Message-ID: <18592.62730.840231.108375@notabene.brown>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
To: linux-nfs@vger.kernel.org
Sender: linux-nfs-owner@vger.kernel.org


Hi 
 I have a report of a the BUG_ON in rpcb_getport_clnt being
 triggered.  This is:
	/* Autobind on cloned rpc clients is discouraged */
	BUG_ON(clnt->cl_parent != clnt);

 It looks to me that while they might be discouraged, they are not
 prevented and so having the BUG_ON is wrong.

 When rpc_clone_client creates a clone, it sets cl_autobind to 0,
 and gives the new client a reference to the same cl_xprt as the
 original client.

 The only effect of cl_autobind is to prevent rpc_force_rebind from
 clearing the BOUND flag on ->cl_xprt.
 So while a call to rpc_force_rebind on the clone will not clear that
 flag, a call on the original client will clear that flag.

 So a cloned client can still end up with a ->cl_xprt with the BOUND
 flag clear.

 So call_bind (which is present in the call trace under the oops) can
 find that !xprt_bound, even when the client is a cloned client.

 When this happens, ->rpcbind, which is rpcb_getport_clnt, goes BOOM.

 What should happen when a clone client finds that its transport is no
 longer bound?  Should rpc_getport_async just do
    clnt = task->tk_client->cl_parent;
 ??

Perplexed,
NeilBrown