From mboxrd@z Thu Jan 1 00:00:00 1970 From: Pablo Neira Ayuso Subject: [PATCH 23/38] netfilter: nf_conntrack: resolve clash for matching conntracks Date: Fri, 20 Jul 2018 15:08:51 +0200 Message-ID: <20180720130906.27687-24-pablo@netfilter.org> References: <20180720130906.27687-1-pablo@netfilter.org> Cc: davem@davemloft.net, netdev@vger.kernel.org To: netfilter-devel@vger.kernel.org Return-path: Received: from mail.us.es ([193.147.175.20]:45524 "EHLO mail.us.es" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1731842AbeGTN5p (ORCPT ); Fri, 20 Jul 2018 09:57:45 -0400 Received: from antivirus1-rhel7.int (unknown [192.168.2.11]) by mail.us.es (Postfix) with ESMTP id A0DA2C516E for ; Fri, 20 Jul 2018 15:07:34 +0200 (CEST) Received: from antivirus1-rhel7.int (localhost [127.0.0.1]) by antivirus1-rhel7.int (Postfix) with ESMTP id 8B5A1DA4CA for ; Fri, 20 Jul 2018 15:07:34 +0200 (CEST) In-Reply-To: <20180720130906.27687-1-pablo@netfilter.org> Sender: netdev-owner@vger.kernel.org List-ID: From: Martynas Pumputis This patch enables the clash resolution for NAT (disabled in "590b52e10d41") if clashing conntracks match (i.e. both tuples are equal) and a protocol allows it. The clash might happen for a connections-less protocol (e.g. UDP) when two threads in parallel writes to the same socket and consequent calls to "get_unique_tuple" return the same tuples (incl. reply tuples). In this case it is safe to perform the resolution, as the losing CT describes the same mangling as the winning CT, so no modifications to the packet are needed, and the result of rules traversal for the loser's packet stays valid. Signed-off-by: Martynas Pumputis Signed-off-by: Pablo Neira Ayuso --- net/netfilter/nf_conntrack_core.c | 30 ++++++++++++++++++++++-------- 1 file changed, 22 insertions(+), 8 deletions(-) diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c index 5123e91b1982..4ced7c7102b6 100644 --- a/net/netfilter/nf_conntrack_core.c +++ b/net/netfilter/nf_conntrack_core.c @@ -632,6 +632,18 @@ nf_ct_key_equal(struct nf_conntrack_tuple_hash *h, net_eq(net, nf_ct_net(ct)); } +static inline bool +nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2) +{ + return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple, + &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) && + nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple, + &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) && + nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) && + nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) && + net_eq(nf_ct_net(ct1), nf_ct_net(ct2)); +} + /* caller must hold rcu readlock and none of the nf_conntrack_locks */ static void nf_ct_gc_expired(struct nf_conn *ct) { @@ -825,19 +837,21 @@ static int nf_ct_resolve_clash(struct net *net, struct sk_buff *skb, /* This is the conntrack entry already in hashes that won race. */ struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h); const struct nf_conntrack_l4proto *l4proto; + enum ip_conntrack_info oldinfo; + struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct)); if (l4proto->allow_clash && - ((ct->status & IPS_NAT_DONE_MASK) == 0) && !nf_ct_is_dying(ct) && atomic_inc_not_zero(&ct->ct_general.use)) { - enum ip_conntrack_info oldinfo; - struct nf_conn *loser_ct = nf_ct_get(skb, &oldinfo); - - nf_ct_acct_merge(ct, ctinfo, loser_ct); - nf_conntrack_put(&loser_ct->ct_general); - nf_ct_set(skb, ct, oldinfo); - return NF_ACCEPT; + if (((ct->status & IPS_NAT_DONE_MASK) == 0) || + nf_ct_match(ct, loser_ct)) { + nf_ct_acct_merge(ct, ctinfo, loser_ct); + nf_conntrack_put(&loser_ct->ct_general); + nf_ct_set(skb, ct, oldinfo); + return NF_ACCEPT; + } + nf_ct_put(ct); } NF_CT_STAT_INC(net, drop); return NF_DROP; -- 2.11.0