From mboxrd@z Thu Jan 1 00:00:00 1970 From: Martin KaFai Lau Subject: [PATCH v3 net 5/5] ipv6: Replace spinlock with seqlock and rcu in ip6_tunnel Date: Fri, 11 Sep 2015 11:06:21 -0700 Message-ID: <1441994781-2718585-6-git-send-email-kafai@fb.com> References: <1441994781-2718585-1-git-send-email-kafai@fb.com> Mime-Version: 1.0 Content-Type: text/plain Cc: "David S. Miller" , Eric Dumazet , Hannes Frederic Sowa , Kernel Team To: netdev Return-path: Received: from mx0b-00082601.pphosted.com ([67.231.153.30]:22528 "EHLO m0041696.ppops.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1751540AbbIKSHe (ORCPT ); Fri, 11 Sep 2015 14:07:34 -0400 Received: from pps.filterd (m0041696 [127.0.0.1]) by m0041696.ppops.net (8.14.5/8.14.5) with SMTP id t8BI7OtD032497 for ; Fri, 11 Sep 2015 11:07:32 -0700 Received: from mail.thefacebook.com ([199.201.64.23]) by m0041696.ppops.net with ESMTP id 1wv08dgqbr-3 (version=TLSv1/SSLv3 cipher=AES128-SHA bits=128 verify=NOT) for ; Fri, 11 Sep 2015 11:07:32 -0700 Received: from facebook.com (2401:db00:11:d0a6:face:0:33:0) by mx-out.facebook.com (10.212.236.87) with ESMTP id f0d9db9258af11e5b3290002c9521c9e-9b7f7240 for ; Fri, 11 Sep 2015 11:07:19 -0700 In-Reply-To: <1441994781-2718585-1-git-send-email-kafai@fb.com> Sender: netdev-owner@vger.kernel.org List-ID: This patch uses a seqlock to ensure consistency between idst->dst and idst->cookie. It also makes dst freeing from fib tree to undergo a rcu grace period. Signed-off-by: Martin KaFai Lau --- include/net/ip6_tunnel.h | 4 ++-- net/ipv6/ip6_fib.c | 9 +++++++-- net/ipv6/ip6_tunnel.c | 51 +++++++++++++++++++++++++----------------------- 3 files changed, 36 insertions(+), 28 deletions(-) diff --git a/include/net/ip6_tunnel.h b/include/net/ip6_tunnel.h index 60b4f40..65c2a93 100644 --- a/include/net/ip6_tunnel.h +++ b/include/net/ip6_tunnel.h @@ -33,8 +33,8 @@ struct __ip6_tnl_parm { }; struct ip6_tnl_dst { - spinlock_t lock; - struct dst_entry *dst; + seqlock_t lock; + struct dst_entry __rcu *dst; u32 cookie; }; diff --git a/net/ipv6/ip6_fib.c b/net/ipv6/ip6_fib.c index e68350b..8a9ec01 100644 --- a/net/ipv6/ip6_fib.c +++ b/net/ipv6/ip6_fib.c @@ -155,6 +155,11 @@ static void node_free(struct fib6_node *fn) kmem_cache_free(fib6_node_kmem, fn); } +static void rt6_rcu_free(struct rt6_info *rt) +{ + call_rcu(&rt->dst.rcu_head, dst_rcu_free); +} + static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt) { int cpu; @@ -169,7 +174,7 @@ static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt) ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu); pcpu_rt = *ppcpu_rt; if (pcpu_rt) { - dst_free(&pcpu_rt->dst); + rt6_rcu_free(pcpu_rt); *ppcpu_rt = NULL; } } @@ -181,7 +186,7 @@ static void rt6_release(struct rt6_info *rt) { if (atomic_dec_and_test(&rt->rt6i_ref)) { rt6_free_pcpu(rt); - dst_free(&rt->dst); + rt6_rcu_free(rt); } } diff --git a/net/ipv6/ip6_tunnel.c b/net/ipv6/ip6_tunnel.c index 851cf6d..d27312a 100644 --- a/net/ipv6/ip6_tunnel.c +++ b/net/ipv6/ip6_tunnel.c @@ -126,45 +126,48 @@ static struct net_device_stats *ip6_get_stats(struct net_device *dev) * Locking : hash tables are protected by RCU and RTNL */ -static void __ip6_tnl_per_cpu_dst_set(struct ip6_tnl_dst *idst, - struct dst_entry *dst) +static void ip6_tnl_per_cpu_dst_set(struct ip6_tnl_dst *idst, + struct dst_entry *dst) { - dst_release(idst->dst); + write_seqlock_bh(&idst->lock); + dst_release(rcu_dereference_protected( + idst->dst, + lockdep_is_held(&idst->lock->lock))); if (dst) { dst_hold(dst); idst->cookie = rt6_get_cookie((struct rt6_info *)dst); } else { idst->cookie = 0; } - idst->dst = dst; -} - -static void ip6_tnl_per_cpu_dst_set(struct ip6_tnl_dst *idst, - struct dst_entry *dst) -{ - - spin_lock_bh(&idst->lock); - __ip6_tnl_per_cpu_dst_set(idst, dst); - spin_unlock_bh(&idst->lock); + rcu_assign_pointer(idst->dst, dst); + write_sequnlock_bh(&idst->lock); } struct dst_entry *ip6_tnl_dst_get(struct ip6_tnl *t) { struct ip6_tnl_dst *idst; struct dst_entry *dst; + unsigned int seq; + u32 cookie; idst = raw_cpu_ptr(t->dst_cache); - spin_lock_bh(&idst->lock); - dst = idst->dst; - if (dst) { - if (!dst->obsolete || dst->ops->check(dst, idst->cookie)) { - dst_hold(idst->dst); - } else { - __ip6_tnl_per_cpu_dst_set(idst, NULL); - dst = NULL; - } + + rcu_read_lock(); + do { + seq = read_seqbegin(&idst->lock); + dst = rcu_dereference(idst->dst); + cookie = idst->cookie; + } while (read_seqretry(&idst->lock, seq)); + + if (dst && !atomic_inc_not_zero(&dst->__refcnt)) + dst = NULL; + rcu_read_unlock(); + + if (dst && dst->obsolete && !dst->ops->check(dst, cookie)) { + ip6_tnl_per_cpu_dst_set(idst, NULL); + dst_release(dst); + dst = NULL; } - spin_unlock_bh(&idst->lock); return dst; } EXPORT_SYMBOL_GPL(ip6_tnl_dst_get); @@ -204,7 +207,7 @@ int ip6_tnl_dst_init(struct ip6_tnl *t) return -ENOMEM; for_each_possible_cpu(i) - spin_lock_init(&per_cpu_ptr(t->dst_cache, i)->lock); + seqlock_init(&per_cpu_ptr(t->dst_cache, i)->lock); return 0; } -- 2.5.1