Return-Path: linux-nfs-owner@vger.kernel.org Received: from mail-wg0-f44.google.com ([74.125.82.44]:58753 "EHLO mail-wg0-f44.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755911Ab3JJNM4 (ORCPT ); Thu, 10 Oct 2013 09:12:56 -0400 Received: by mail-wg0-f44.google.com with SMTP id n12so2072559wgh.35 for ; Thu, 10 Oct 2013 06:12:55 -0700 (PDT) Message-ID: <5256A7D4.2050509@primarydata.com> Date: Thu, 10 Oct 2013 16:12:52 +0300 From: Benny Halevy MIME-Version: 1.0 To: Nadav Shemer CC: bfields@redhat.com, linux-nfs@vger.kernel.org Subject: Re: [PATCH] nfsd: unhash client before expiring it References: <1381409248-1858-1-git-send-email-nadav@primarydata.com> In-Reply-To: <1381409248-1858-1-git-send-email-nadav@primarydata.com> Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-nfs-owner@vger.kernel.org List-ID: On 2013-10-10 15:47, Nadav Shemer wrote: > Some client_expire operations (pnfs_expire_client) may release the state > lock. To prevent two threads from concurrently expiring the same client, > the client is unhashed before any other operation > > Signed-off-by: Nadav Shemer > --- > Related to the recent patchset from Benny Halevy, expire_client releases the state lock (expire_client->destroy_client->pnfs_expire_client). > A different thread processing a SET_CLIENTID/EXCHANGE_ID/etc. might find the client on the confirmed/unconfirmed hash table and also call expire_client on it. > > To correct this, I reshuffled destroy_client to remove the client from hash tables first thing, before the state lock may be released > > Benny: my previous version of this patch had some plays with reference counts (taking a reference for the duration of the destroy). > In the current version of the tree, destroy_client assumes cl_refcount==0 and is the only caller of free_client besides create_client. So I dropped it OK. In my state lock elimination branch this hunk is factored out and done under a spin lock while the heavy lifting of managing the tracking record will be done under a mutex. Benny > > fs/nfsd/nfs4state.c | 15 +++++++++------ > 1 file changed, 9 insertions(+), 6 deletions(-) > > diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c > index b4a28ef..d42434b 100644 > --- a/fs/nfsd/nfs4state.c > +++ b/fs/nfsd/nfs4state.c > @@ -1154,57 +1154,60 @@ unhash_client_locked(struct nfs4_client *clp) > list_del(&clp->cl_lru); > spin_lock(&clp->cl_lock); > list_for_each_entry(ses, &clp->cl_sessions, se_perclnt) > list_del_init(&ses->se_hash); > spin_unlock(&clp->cl_lock); > } > > static void > destroy_client(struct nfs4_client *clp) > { > struct nfs4_openowner *oo; > struct nfs4_delegation *dp; > struct list_head reaplist; > struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id); > > + list_del(&clp->cl_idhash); > + if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags)) > + rb_erase(&clp->cl_namenode, &nn->conf_name_tree); > + else > + rb_erase(&clp->cl_namenode, &nn->unconf_name_tree); > + spin_lock(&nn->client_lock); > + unhash_client_locked(clp); > + spin_unlock(&nn->client_lock); > + > INIT_LIST_HEAD(&reaplist); > spin_lock(&recall_lock); > while (!list_empty(&clp->cl_delegations)) { > dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt); > list_del_init(&dp->dl_perclnt); > list_move(&dp->dl_recall_lru, &reaplist); > } > spin_unlock(&recall_lock); > while (!list_empty(&reaplist)) { > dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru); > destroy_delegation(dp); > } > while (!list_empty(&clp->cl_openowners)) { > oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient); > release_openowner(oo); > } > pnfs_expire_client(clp); > nfsd4_shutdown_callback(clp); > if (clp->cl_cb_conn.cb_xprt) > svc_xprt_put(clp->cl_cb_conn.cb_xprt); > - list_del(&clp->cl_idhash); > - if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags)) > - rb_erase(&clp->cl_namenode, &nn->conf_name_tree); > - else > - rb_erase(&clp->cl_namenode, &nn->unconf_name_tree); > spin_lock(&nn->client_lock); > - unhash_client_locked(clp); > WARN_ON_ONCE(atomic_read(&clp->cl_refcount)); > free_client(clp); > spin_unlock(&nn->client_lock); > } > > static void expire_client(struct nfs4_client *clp) > { > nfsd4_client_record_remove(clp); > destroy_client(clp); > } > > static void copy_verf(struct nfs4_client *target, nfs4_verifier *source) > { > memcpy(target->cl_verifier.data, source->data, > sizeof(target->cl_verifier.data)); >