Return-Path: linux-nfs-owner@vger.kernel.org Received: from mailhub.sw.ru ([195.214.232.25]:45924 "EHLO relay.sw.ru" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751309Ab3FZGP3 (ORCPT ); Wed, 26 Jun 2013 02:15:29 -0400 Subject: [PATCH v4] SUNRPC: fix races on PipeFS UMOUNT notifications To: Trond.Myklebust@netapp.com From: Stanislav Kinsbursky Cc: linux-nfs@vger.kernel.org, devel@openvz.org, linux-kernel@vger.kernel.org, jlayton@redhat.com Date: Wed, 26 Jun 2013 10:15:14 +0400 Message-ID: <20130626061448.32202.192.stgit@localhost.localdomain> In-Reply-To: <20130624075245.17104.6279.stgit@localhost.localdomain> References: <20130624075245.17104.6279.stgit@localhost.localdomain> MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Sender: linux-nfs-owner@vger.kernel.org List-ID: CPU#0 CPU#1 ----------------------------- ----------------------------- rpc_kill_sb sn->pipefs_sb = NULL rpc_release_client (UMOUNT_EVENT) rpc_free_auth rpc_pipefs_event rpc_get_client_for_event !atomic_inc_not_zero(cl_count) atomic_inc(cl_count) rpc_free_client rpc_clnt_remove_pipedir To fix this, this patch does the following: 1) Calls RPC_PIPEFS_UMOUNT notification with sn->pipefs_sb_lock being held. 2) Removes SUNRPC client from the list AFTER pipes destroying. 3) Doesn't hold RPC client on notification: if client in the list, then it can't be destroyed while sn->pipefs_sb_lock in hold by notification caller. Signed-off-by: Stanislav Kinsbursky Cc: stable@vger.kernel.org --- net/sunrpc/clnt.c | 5 +---- net/sunrpc/rpc_pipe.c | 2 +- 2 files changed, 2 insertions(+), 5 deletions(-) diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c index b827a4b..41f180c 100644 --- a/net/sunrpc/clnt.c +++ b/net/sunrpc/clnt.c @@ -236,8 +236,6 @@ static struct rpc_clnt *rpc_get_client_for_event(struct net *net, int event) continue; if (rpc_clnt_skip_event(clnt, event)) continue; - if (atomic_inc_not_zero(&clnt->cl_count) == 0) - continue; spin_unlock(&sn->rpc_client_lock); return clnt; } @@ -254,7 +252,6 @@ static int rpc_pipefs_event(struct notifier_block *nb, unsigned long event, while ((clnt = rpc_get_client_for_event(sb->s_fs_info, event))) { error = __rpc_pipefs_event(clnt, event, sb); - rpc_release_client(clnt); if (error) break; } @@ -641,8 +638,8 @@ rpc_free_client(struct rpc_clnt *clnt) rcu_dereference(clnt->cl_xprt)->servername); if (clnt->cl_parent != clnt) rpc_release_client(clnt->cl_parent); - rpc_unregister_client(clnt); rpc_clnt_remove_pipedir(clnt); + rpc_unregister_client(clnt); rpc_free_iostats(clnt->cl_metrics); kfree(clnt->cl_principal); clnt->cl_metrics = NULL; diff --git a/net/sunrpc/rpc_pipe.c b/net/sunrpc/rpc_pipe.c index c512448..9b4ec5b 100644 --- a/net/sunrpc/rpc_pipe.c +++ b/net/sunrpc/rpc_pipe.c @@ -1165,12 +1165,12 @@ static void rpc_kill_sb(struct super_block *sb) goto out; } sn->pipefs_sb = NULL; - mutex_unlock(&sn->pipefs_sb_lock); dprintk("RPC: sending pipefs UMOUNT notification for net %p%s\n", net, NET_NAME(net)); blocking_notifier_call_chain(&rpc_pipefs_notifier_list, RPC_PIPEFS_UMOUNT, sb); + mutex_unlock(&sn->pipefs_sb_lock); put_net(net); out: kill_litter_super(sb);