Return-Path: linux-nfs-owner@vger.kernel.org Received: from bombadil.infradead.org ([198.137.202.9]:59452 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751132AbaDSOu1 (ORCPT ); Sat, 19 Apr 2014 10:50:27 -0400 Date: Sat, 19 Apr 2014 07:50:26 -0700 From: Christoph Hellwig To: Trond Myklebust Cc: Bruce Fields , linux-nfs@vger.kernel.org Subject: Re: [PATCH 36/70] NFSd: Add reference counting to find_stateid Message-ID: <20140419145026.GE25682@infradead.org> References: <1397846704-14567-28-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-29-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-30-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-31-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-32-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-33-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-34-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-35-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-36-git-send-email-trond.myklebust@primarydata.com> <1397846704-14567-37-git-send-email-trond.myklebust@primarydata.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <1397846704-14567-37-git-send-email-trond.myklebust@primarydata.com> Sender: linux-nfs-owner@vger.kernel.org List-ID: > +static void nfs4_put_stateid(struct nfs4_stid *s) > +{ > + if (s == NULL) > + return; > + switch (s->sc_type) { > + case NFS4_OPEN_STID: > + case NFS4_LOCK_STID: > + case NFS4_CLOSED_STID: > + put_generic_stateid(openlockstateid(s)); > + break; > + case NFS4_DELEG_STID: > + case NFS4_REVOKED_DELEG_STID: > + case NFS4_CLOSED_DELEG_STID: > + nfs4_put_delegation(delegstateid(s)); > + } > +} I really don't like the way the inheritance for the stateids works, a pure put operation shouldn't need this. I think all this can be fixed by adding a ->free function pointer to struct nfs4_stid. At this point the braindamage of passing a kmem_cache pointer to various function can be removed (similar to how nfs4_alloc_stid should be replaced with a nfs4_init_stid that takes an already allocated stid), and nothing in the normal refcounting path should need these switches. > @@ -3804,26 +3823,33 @@ static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid) > return nfserr_bad_stateid; > status = check_stateid_generation(stateid, &s->sc_stateid, 1); > if (status) > - return status; > + goto out_put_stid; > switch (s->sc_type) { > case NFS4_DELEG_STID: > - return nfs_ok; > + status = nfs_ok; > + break; > case NFS4_REVOKED_DELEG_STID: > - return nfserr_deleg_revoked; > + status = nfserr_deleg_revoked; > + break; > case NFS4_OPEN_STID: > case NFS4_LOCK_STID: > ols = openlockstateid(s); > if (ols->st_stateowner->so_is_open_owner > && !(openowner(ols->st_stateowner)->oo_flags > & NFS4_OO_CONFIRMED)) > - return nfserr_bad_stateid; > - return nfs_ok; > + status = nfserr_bad_stateid; > + else > + status = nfs_ok; > + break; Not quite as urgent as for the refcounting, but I think moving more of these switches on the type into proper methods would improve the stateid code a lot.