Re: [pnfs] [PATCH] nfsd41: renew_client must be called under the state lock

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

On Thu, Aug 27, 2009 at 07:37:02PM +0300, Benny Halevy wrote:
> On Aug. 27, 2009, 0:02 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
> > On Wed, Aug 26, 2009 at 11:20:48AM +0300, Boaz Harrosh wrote:
> >> On 08/26/2009 12:59 AM, J. Bruce Fields wrote:
> >>> commit fdf7875040506ca7e595dffef56cbd81ae6b384b
> >>> Author: Benny Halevy <bhalevy@panasas.com>
> >>> Date:   Thu Aug 20 03:21:56 2009 +0300
> >>>
> >>>     nfsd41: renew_client must be called under the state lock
> >>>     
> >>>     Until we work out the state locking so we can use a spin lock to protect
> >>>     the cl_lru, we need to take the state_lock to renew the client.
> >>>     
> >>>     Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> >>>     [nfsd41: Do not renew state on error]
> >>>     Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
> >>>     Signed-off-by: Benny Halevy <bhalevy@panasas.com>
> >>>     Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
> >>>
> >>> diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> >>> index d2a0524..8e5ac49 100644
> >>> --- a/fs/nfsd/nfs4state.c
> >>> +++ b/fs/nfsd/nfs4state.c
> >>> @@ -1463,12 +1463,16 @@ nfsd4_sequence(struct svc_rqst *rqstp,
> >>>  	spin_lock(&sessionid_lock);
> >>>  	status = nfserr_badsession;
> >>>  	session = find_in_sessionid_hashtbl(&seq->sessionid);
> >>> -	if (!session)
> >>> -		goto out;
> >>> +	if (!session) {
> >>> +		spin_unlock(&sessionid_lock);
> >>> +		goto err;
> >>> +	}
> >>>  
> >>>  	status = nfserr_badslot;
> >>> -	if (seq->slotid >= session->se_fchannel.maxreqs)
> >>> -		goto out;
> >>> +	if (seq->slotid >= session->se_fchannel.maxreqs) {
> >>> +		spin_unlock(&sessionid_lock);
> >>> +		goto err;
> >>> +	}
> >>>  
> >> A spin_unlock is a big and delicate inlined code site
> >> Why not do a "goto err_unlock" or something
> > 
> > Yes, we could add an
> > 
> > 	err_unlock:
> > 		spin_unlock(&sessionid_lock);
> > 		goto err;
> > 
> > at the end.  It still seems little delicate.
> > 
> > How about the following (on top of Benny's patch), which sends all the
> > unlock cases to one label?  (Disclaimer: untested.  And this depends on
> > the assumption that cstate->session is NULL on entry to this function,
> > which I haven't checked.)
> > 
> > --b.
> > 
> > diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
> > index 8e5ac49..5f634d2 100644
> > --- a/fs/nfsd/nfs4state.c
> > +++ b/fs/nfsd/nfs4state.c
> > @@ -1463,16 +1463,12 @@ nfsd4_sequence(struct svc_rqst *rqstp,
> >  	spin_lock(&sessionid_lock);
> >  	status = nfserr_badsession;
> >  	session = find_in_sessionid_hashtbl(&seq->sessionid);
> > -	if (!session) {
> > -		spin_unlock(&sessionid_lock);
> > -		goto err;
> > -	}
> > +	if (!session)
> > +		goto out;
> >  
> >  	status = nfserr_badslot;
> > -	if (seq->slotid >= session->se_fchannel.maxreqs) {
> > -		spin_unlock(&sessionid_lock);
> > -		goto err;
> > -	}
> > +	if (seq->slotid >= session->se_fchannel.maxreqs)
> > +		goto out;
> >  
> >  	slot = &session->se_slots[seq->slotid];
> >  	dprintk("%s: slotid %d\n", __func__, seq->slotid);
> > @@ -1487,10 +1483,8 @@ nfsd4_sequence(struct svc_rqst *rqstp,
> >  		cstate->status = nfserr_replay_cache;
> >  		goto out;
> >  	}
> > -	if (status) {
> > -		spin_unlock(&sessionid_lock);
> > -		goto err;
> > -	}
> > +	if (status)
> > +		goto out;
> >  
> >  	/* Success! bump slot seqid */
> >  	slot->sl_inuse = true;
> > @@ -1510,10 +1504,11 @@ nfsd4_sequence(struct svc_rqst *rqstp,
> >  out:
> >  	spin_unlock(&sessionid_lock);
> >  	/* Renew the clientid on success and on replay */
> > -	nfs4_lock_state();
> > -	renew_client(session->se_client);
> > -	nfs4_unlock_state();
> > -err:
> > +	if (cstate->session) {
> > +		nfs4_lock_state();
> > +		renew_client(session->se_client);
> > +		nfs4_unlock_state();
> > +	}
> >  	dprintk("%s: return %d\n", __func__, ntohl(status));
> >  	return status;
> >  }
> 
> The code looks correct.
> Coming to think about it, I hope that cstate is initialized to zeroes
> since we're not explicitly clearing it. Should we?
> cstate comes from the struct nfsd4_compoundres * nfsd4_proc_compound
> is being called with...

net/sunrpc/svc.c:svc_process_common() has this:

	/* Initialize storage for argp and resp */
	memset(rqstp->rq_argp, 0, procp->pc_argsize);
	memset(rqstp->rq_resp, 0, procp->pc_ressize);

OK!

--b.