Return-Path: linux-nfs-owner@vger.kernel.org Received: from mail-vc0-f175.google.com ([209.85.220.175]:55811 "EHLO mail-vc0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754025AbaLAXgR (ORCPT ); Mon, 1 Dec 2014 18:36:17 -0500 Received: by mail-vc0-f175.google.com with SMTP id hy10so5201120vcb.20 for ; Mon, 01 Dec 2014 15:36:16 -0800 (PST) MIME-Version: 1.0 In-Reply-To: <20141201180533.7c8a7587@tlielax.poochiereds.net> References: <1416597571-4265-1-git-send-email-jlayton@primarydata.com> <1416597571-4265-2-git-send-email-jlayton@primarydata.com> <20141201224407.GD30749@fieldses.org> <20141201180533.7c8a7587@tlielax.poochiereds.net> Date: Mon, 1 Dec 2014 18:36:16 -0500 Message-ID: Subject: Re: [PATCH 1/4] sunrpc: add a rcu_head to svc_rqst and use kfree_rcu to free it From: Trond Myklebust To: Jeff Layton Cc: "J. Bruce Fields" , Chris Worley , Linux NFS Mailing List Content-Type: text/plain; charset=UTF-8 Sender: linux-nfs-owner@vger.kernel.org List-ID: On Mon, Dec 1, 2014 at 6:05 PM, Jeff Layton wrote: > On Mon, 1 Dec 2014 17:44:07 -0500 > "J. Bruce Fields" wrote: > >> On Fri, Nov 21, 2014 at 02:19:28PM -0500, Jeff Layton wrote: >> > ...also make the manipulation of sp_all_threads list use RCU-friendly >> > functions. >> > >> > Signed-off-by: Jeff Layton >> > Tested-by: Chris Worley >> > --- >> > include/linux/sunrpc/svc.h | 2 ++ >> > include/trace/events/sunrpc.h | 3 ++- >> > net/sunrpc/svc.c | 10 ++++++---- >> > 3 files changed, 10 insertions(+), 5 deletions(-) >> > >> > diff --git a/include/linux/sunrpc/svc.h b/include/linux/sunrpc/svc.h >> > index 5f0ab39bf7c3..7f80a99c59e4 100644 >> > --- a/include/linux/sunrpc/svc.h >> > +++ b/include/linux/sunrpc/svc.h >> > @@ -223,6 +223,7 @@ static inline void svc_putu32(struct kvec *iov, __be32 val) >> > struct svc_rqst { >> > struct list_head rq_list; /* idle list */ >> > struct list_head rq_all; /* all threads list */ >> > + struct rcu_head rq_rcu_head; /* for RCU deferred kfree */ >> > struct svc_xprt * rq_xprt; /* transport ptr */ >> > >> > struct sockaddr_storage rq_addr; /* peer address */ >> > @@ -262,6 +263,7 @@ struct svc_rqst { >> > #define RQ_SPLICE_OK (4) /* turned off in gss privacy >> > * to prevent encrypting page >> > * cache pages */ >> > +#define RQ_VICTIM (5) /* about to be shut down */ >> > unsigned long rq_flags; /* flags field */ >> > >> > void * rq_argp; /* decoded arguments */ >> > diff --git a/include/trace/events/sunrpc.h b/include/trace/events/sunrpc.h >> > index 5848fc235869..08a5fed50f34 100644 >> > --- a/include/trace/events/sunrpc.h >> > +++ b/include/trace/events/sunrpc.h >> > @@ -418,7 +418,8 @@ TRACE_EVENT(xs_tcp_data_recv, >> > { (1UL << RQ_LOCAL), "RQ_LOCAL"}, \ >> > { (1UL << RQ_USEDEFERRAL), "RQ_USEDEFERRAL"}, \ >> > { (1UL << RQ_DROPME), "RQ_DROPME"}, \ >> > - { (1UL << RQ_SPLICE_OK), "RQ_SPLICE_OK"}) >> > + { (1UL << RQ_SPLICE_OK), "RQ_SPLICE_OK"}, \ >> > + { (1UL << RQ_VICTIM), "RQ_VICTIM"}) >> > >> > TRACE_EVENT(svc_recv, >> > TP_PROTO(struct svc_rqst *rqst, int status), >> > diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c >> > index 5d9a443d21f6..4edef32f3b9f 100644 >> > --- a/net/sunrpc/svc.c >> > +++ b/net/sunrpc/svc.c >> > @@ -616,7 +616,7 @@ svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) >> > serv->sv_nrthreads++; >> > spin_lock_bh(&pool->sp_lock); >> > pool->sp_nrthreads++; >> > - list_add(&rqstp->rq_all, &pool->sp_all_threads); >> > + list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads); >> > spin_unlock_bh(&pool->sp_lock); >> > rqstp->rq_server = serv; >> > rqstp->rq_pool = pool; >> > @@ -684,7 +684,8 @@ found_pool: >> > * so we don't try to kill it again. >> > */ >> > rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); >> > - list_del_init(&rqstp->rq_all); >> > + set_bit(RQ_VICTIM, &rqstp->rq_flags); >> > + list_del_rcu(&rqstp->rq_all); >> > task = rqstp->rq_task; >> > } >> > spin_unlock_bh(&pool->sp_lock); >> > @@ -782,10 +783,11 @@ svc_exit_thread(struct svc_rqst *rqstp) >> > >> > spin_lock_bh(&pool->sp_lock); >> > pool->sp_nrthreads--; >> > - list_del(&rqstp->rq_all); >> > + if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags)) >> > + list_del_rcu(&rqstp->rq_all); >> >> Both users of RQ_VICTIM are under the sp_lock, so we don't really need >> an atomic test_and_set_bit, do we? >> > > No, it doesn't really need to be an atomic test_and_set_bit. We could > just as easily do: > > if (!test_bit(...)) { > set_bit(...) > list_del_rcu() > } Isn't there a chance that the non-atomic version might end up clobbering one of the other bits that is not set/cleared under the sp_lock? -- Trond Myklebust Linux NFS client maintainer, PrimaryData trond.myklebust@primarydata.com