Linux协议栈优化之Netfilter分类conntrack
浏览数：50 /
时间：2015年06月09日
首先，如果你不同意我以下的观点，本文就不必看了：如今内存不值钱，空间换时间很划算，要知道，一万年前的1秒和一万年后1秒是一样的，你要是觉得人们做的事情可能不同，但是请记住，永远都会发生“安迪给你的被比尔拿走”之类的事情，对于你而言，什么都没有改变！
虽然空间可以被拓展，但是要讲技巧。以下是一张截图，测试的是http服务器的常规性能，我依然使用相对值比较，因为我的电脑压不出真实性能，另外，为了将网络影响降低，我使用了本地环回地址，我的目的是测试conntrack对新建连接数的影响，截图如下：
如果你稍微明白一些conntrack的原理，就会明白它的开销主要集中在两个方面，一个是查找开销，在内存不值钱的年代，可以通过设置很大的hash bucket来缓解，另外一个开销就是分配conn结构体内存的开销。如果在一台压力很大的设备上，短连接特别多，所有的conntrack将会在一张hash表内，我们所能指望的就是针对一个五元组计算的hash值足够散列了，如果对于一些攻击流量，特别是当他研究了Linux计算conntrack hash值的算法后，他会构造很多hash一致的不同五元组的数据包，这会使得hash表的冲突链表过长，遍历开销过大，在当前的内核conntrack模块实现中，这是无法避免的，因为所有鸡蛋都在一个篮子里面。现在换一个思路。
将不同类的数据流分到不同的hash表中如何？也就是说将一张大表拆分成几张小表，或者说如果你真的不在乎内存的话，设置多张大表也行！那么剩下的问题就是将数据包进行分类了。可以有两种实现：
1.在RAW表中做一个模块
根据skb的特征设置其conntrack表索引，值得注意的是，为了避免同一个流分配到不同的hash表，match项必须是计算hash的元素，比如源/目标IP，源/目标端口，协议等。对于IP分片直接通过，不予考虑。
2.在MANGLE表中做一个优先级大于conntrack小于defrag的模块
这个就不多说了，和1一样，就是解决了IP分片的问题。
iptables规则如下：
iptables -t mangle -A PREROUTING -s 1.1.1.0/24 -j CONNMAP 2
以上规则将源IP为1.1.1.0/24段的数据包的conntrack设置在表2中。你也可以区分TCP协议和UDP协议，每一个协议一张表，这样就可以避免恶意UDP流量攻击问题。在加载模块的时候，你需要设置一个参数，即hash表的数量。
为了快速测试效果，我依然是先将代码写死，设置了3张hash表，基于内核3.2.5，两个patch如下：
C文件patch：
重新编译内核之后，效果十分可观，按照下面的脚本加入巨量的conntrack也丝毫不会影响最大连接数：
除了分了多张hash之外，另外的好处在于锁的粒度变细了，每次在操作conntrack的时候只需要锁住和自己相关的hash表的锁即可，当然，在本次的优化中没有涉及这一点。和ipset结合起来就更猛了，本身ipset就可以定义一个集合，而该集合则可以用来计算conntrack的hash值，多表版本的conntrack和ipset结合，这在效果上不就是多级hash表吗？而且表面上还是iptables配置的。
虽然空间可以被拓展，但是要讲技巧。以下是一张截图，测试的是http服务器的常规性能，我依然使用相对值比较，因为我的电脑压不出真实性能，另外，为了将网络影响降低，我使用了本地环回地址，我的目的是测试conntrack对新建连接数的影响，截图如下：
如果你稍微明白一些conntrack的原理，就会明白它的开销主要集中在两个方面，一个是查找开销，在内存不值钱的年代，可以通过设置很大的hash bucket来缓解，另外一个开销就是分配conn结构体内存的开销。如果在一台压力很大的设备上，短连接特别多，所有的conntrack将会在一张hash表内，我们所能指望的就是针对一个五元组计算的hash值足够散列了，如果对于一些攻击流量，特别是当他研究了Linux计算conntrack hash值的算法后，他会构造很多hash一致的不同五元组的数据包，这会使得hash表的冲突链表过长，遍历开销过大，在当前的内核conntrack模块实现中，这是无法避免的，因为所有鸡蛋都在一个篮子里面。现在换一个思路。
将不同类的数据流分到不同的hash表中如何？也就是说将一张大表拆分成几张小表，或者说如果你真的不在乎内存的话，设置多张大表也行！那么剩下的问题就是将数据包进行分类了。可以有两种实现：
1.在RAW表中做一个模块
根据skb的特征设置其conntrack表索引，值得注意的是，为了避免同一个流分配到不同的hash表，match项必须是计算hash的元素，比如源/目标IP，源/目标端口，协议等。对于IP分片直接通过，不予考虑。
2.在MANGLE表中做一个优先级大于conntrack小于defrag的模块
这个就不多说了，和1一样，就是解决了IP分片的问题。
iptables规则如下：
iptables -t mangle -A PREROUTING -s 1.1.1.0/24 -j CONNMAP 2
以上规则将源IP为1.1.1.0/24段的数据包的conntrack设置在表2中。你也可以区分TCP协议和UDP协议，每一个协议一张表，这样就可以避免恶意UDP流量攻击问题。在加载模块的时候，你需要设置一个参数，即hash表的数量。
为了快速测试效果，我依然是先将代码写死，设置了3张hash表，基于内核3.2.5，两个patch如下：
C文件patch：
diff -uNr linux-source-3.2/net/ipv4/netfilter/ipt_MASQUERADE.c linux-source-3.2.new/net/ipv4/netfilter/ipt_MASQUERADE.c
--- linux-source-3.2/net/ipv4/netfilter/ipt_MASQUERADE.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/ipt_MASQUERADE.c 2014-07-11 10:32:57.736666273 +0800
@@ -85,7 +85,7 @@
mr->range[0].min, mr->range[0].max });
/* Hand modified range to generic setup. */
- return nf_nat_setup_info(ct, &newrange, IP_NAT_MANIP_SRC);
+ return nf_nat_setup_info(ct, &newrange, IP_NAT_MANIP_SRC, 0);
}
static int
diff -uNr linux-source-3.2/net/ipv4/netfilter/ipt_NETMAP.c linux-source-3.2.new/net/ipv4/netfilter/ipt_NETMAP.c
--- linux-source-3.2/net/ipv4/netfilter/ipt_NETMAP.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/ipt_NETMAP.c 2014-07-11 10:35:14.976667434 +0800
@@ -67,7 +67,7 @@
mr->range[0].min, mr->range[0].max });
/* Hand modified range to generic setup. */
- return nf_nat_setup_info(ct, &newrange, HOOK2MANIP(par->hooknum));
+ return nf_nat_setup_info(ct, &newrange, HOOK2MANIP(par->hooknum), 0);
}
static struct xt_target netmap_tg_reg __read_mostly = {
diff -uNr linux-source-3.2/net/ipv4/netfilter/ipt_REDIRECT.c linux-source-3.2.new/net/ipv4/netfilter/ipt_REDIRECT.c
--- linux-source-3.2/net/ipv4/netfilter/ipt_REDIRECT.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/ipt_REDIRECT.c 2014-07-11 10:36:45.760668202 +0800
@@ -82,7 +82,7 @@
mr->range[0].min, mr->range[0].max });
/* Hand modified range to generic setup. */
- return nf_nat_setup_info(ct, &newrange, IP_NAT_MANIP_DST);
+ return nf_nat_setup_info(ct, &newrange, IP_NAT_MANIP_DST, 0);
}
static struct xt_target redirect_tg_reg __read_mostly = {
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c
--- linux-source-3.2/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4.c 2014-07-11 15:03:50.596093049 +0800
@@ -151,6 +151,12 @@
const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
+ struct iphdr *hdr = ip_hdr(skb);
+ if (ipv4_is_loopback(hdr->saddr)||ipv4_is_loopback(hdr->daddr)) {
+ skb->ij = 1;
+ } else {
+ skb->ij = 0;
+ }
return nf_conntrack_in(dev_net(in), PF_INET, hooknum, skb);
}
@@ -160,6 +166,12 @@
const struct net_device *out,
int (*okfn)(struct sk_buff *))
{
+ struct iphdr *hdr = ip_hdr(skb);
+ if (ipv4_is_loopback(hdr->saddr)||ipv4_is_loopback(hdr->daddr)) {
+ skb->ij = 1;
+ } else {
+ skb->ij = 0;
+ }
/* root is playing with raw sockets. */
if (skb->len < sizeof(struct iphdr) ||
ip_hdrlen(skb) < sizeof(struct iphdr))
@@ -254,7 +266,7 @@
getorigdst(struct sock *sk, int optval, void __user *user, int *len)
{
const struct inet_sock *inet = inet_sk(sk);
- const struct nf_conntrack_tuple_hash *h;
+ /*const struct nf_conntrack_tuple_hash *h;*/
struct nf_conntrack_tuple tuple;
memset(&tuple, 0, sizeof(tuple));
@@ -276,7 +288,7 @@
*len, sizeof(struct sockaddr_in));
return -EINVAL;
}
-
+/*
h = nf_conntrack_find_get(sock_net(sk), NF_CT_DEFAULT_ZONE, &tuple);
if (h) {
struct sockaddr_in sin;
@@ -297,6 +309,7 @@
else
return 0;
}
+*/
pr_debug("SO_ORIGINAL_DST: Can‘t find %pI4/%u-%pI4/%u.\n",
&tuple.src.u3.ip, ntohs(tuple.src.u.tcp.port),
&tuple.dst.u3.ip, ntohs(tuple.dst.u.tcp.port));
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c
--- linux-source-3.2/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_l3proto_ipv4_compat.c 2014-07-11 09:11:02.144624681 +0800
@@ -38,7 +38,7 @@
st->bucket < net->ct.htable_size;
st->bucket++) {
n = rcu_dereference(
- hlist_nulls_first_rcu(&net->ct.hash[st->bucket]));
+ hlist_nulls_first_rcu(&net->ct.hash[0][st->bucket]));
if (!is_a_nulls(n))
return n;
}
@@ -58,7 +58,7 @@
return NULL;
}
head = rcu_dereference(
- hlist_nulls_first_rcu(&net->ct.hash[st->bucket]));
+ hlist_nulls_first_rcu(&net->ct.hash[0][st->bucket]));
}
return head;
}
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_conntrack_proto_icmp.c linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_proto_icmp.c
--- linux-source-3.2/net/ipv4/netfilter/nf_conntrack_proto_icmp.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_conntrack_proto_icmp.c 2014-07-11 09:55:31.704647269 +0800
@@ -148,7 +148,7 @@
*ctinfo = IP_CT_RELATED;
- h = nf_conntrack_find_get(net, zone, &innertuple);
+ h = nf_conntrack_find_get(net, zone, &innertuple, skb->ij);
if (!h) {
pr_debug("icmp_error_message: no match\n");
return -NF_ACCEPT;
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_core.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_core.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_core.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_core.c 2014-07-11 10:24:16.544661863 +0800
@@ -74,7 +74,7 @@
struct nf_conntrack_tuple reply;
nf_ct_invert_tuplepr(&reply, tuple);
- return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
+ return nf_conntrack_tuple_taken(&reply, ignored_conntrack, 0);
}
EXPORT_SYMBOL(nf_nat_used_tuple);
@@ -206,7 +206,7 @@
const struct nf_conntrack_tuple *orig_tuple,
const struct nf_nat_range *range,
struct nf_conn *ct,
- enum nf_nat_manip_type maniptype)
+ enum nf_nat_manip_type maniptype, int ij)
{
struct net *net = nf_ct_net(ct);
const struct nf_nat_protocol *proto;
@@ -268,7 +268,7 @@
unsigned int
nf_nat_setup_info(struct nf_conn *ct,
const struct nf_nat_range *range,
- enum nf_nat_manip_type maniptype)
+ enum nf_nat_manip_type maniptype, int ij)
{
struct net *net = nf_ct_net(ct);
struct nf_conntrack_tuple curr_tuple, new_tuple;
@@ -296,7 +296,7 @@
nf_ct_invert_tuplepr(&curr_tuple,
&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
- get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
+ get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype, ij);
if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
struct nf_conntrack_tuple reply;
@@ -670,7 +670,7 @@
if (nf_nat_initialized(ct, manip))
return -EEXIST;
- return nf_nat_setup_info(ct, &range, manip);
+ return nf_nat_setup_info(ct, &range, manip, 0);
}
#else
static int
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_h323.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_h323.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_h323.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_h323.c 2014-07-11 10:08:16.712653742 +0800
@@ -411,14 +411,14 @@
/* Change src to where master sends to */
range.flags = IP_NAT_RANGE_MAP_IPS;
range.min_ip = range.max_ip = new->tuplehash[!this->dir].tuple.src.u3.ip;
- nf_nat_setup_info(new, &range, IP_NAT_MANIP_SRC);
+ nf_nat_setup_info(new, &range, IP_NAT_MANIP_SRC, 0);
/* For DST manip, map port here to where it‘s expected. */
range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
range.min = range.max = this->saved_proto;
range.min_ip = range.max_ip =
new->master->tuplehash[!this->dir].tuple.src.u3.ip;
- nf_nat_setup_info(new, &range, IP_NAT_MANIP_DST);
+ nf_nat_setup_info(new, &range, IP_NAT_MANIP_DST, 0);
}
/****************************************************************************/
@@ -504,13 +504,13 @@
/* Change src to where master sends to */
range.flags = IP_NAT_RANGE_MAP_IPS;
range.min_ip = range.max_ip = new->tuplehash[!this->dir].tuple.src.u3.ip;
- nf_nat_setup_info(new, &range, IP_NAT_MANIP_SRC);
+ nf_nat_setup_info(new, &range, IP_NAT_MANIP_SRC, 0);
/* For DST manip, map port here to where it‘s expected. */
range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
range.min = range.max = this->saved_proto;
range.min_ip = range.max_ip = this->saved_ip;
- nf_nat_setup_info(new, &range, IP_NAT_MANIP_DST);
+ nf_nat_setup_info(new, &range, IP_NAT_MANIP_DST, 0);
}
/****************************************************************************/
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_helper.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_helper.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_helper.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_helper.c 2014-07-11 10:08:55.008654066 +0800
@@ -439,13 +439,13 @@
range.flags = IP_NAT_RANGE_MAP_IPS;
range.min_ip = range.max_ip
= ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC, 0);
/* For DST manip, map port here to where it‘s expected. */
range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
range.min = range.max = exp->saved_proto;
range.min_ip = range.max_ip
= ct->master->tuplehash[!exp->dir].tuple.src.u3.ip;
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST, 0);
}
EXPORT_SYMBOL(nf_nat_follow_master);
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_pptp.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_pptp.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_pptp.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_pptp.c 2014-07-11 10:08:35.424653900 +0800
@@ -95,7 +95,7 @@
range.flags |= IP_NAT_RANGE_PROTO_SPECIFIED;
range.min = range.max = exp->saved_proto;
}
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC, 0);
/* For DST manip, map port here to where it‘s expected. */
range.flags = IP_NAT_RANGE_MAP_IPS;
@@ -105,7 +105,7 @@
range.flags |= IP_NAT_RANGE_PROTO_SPECIFIED;
range.min = range.max = exp->saved_proto;
}
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST, 0);
}
/* outbound packets == from PNS to PAC */
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_rule.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_rule.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_rule.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_rule.c 2014-07-11 10:09:31.024654370 +0800
@@ -56,7 +56,7 @@
ctinfo == IP_CT_RELATED_REPLY));
NF_CT_ASSERT(par->out != NULL);
- return nf_nat_setup_info(ct, &mr->range[0], IP_NAT_MANIP_SRC);
+ return nf_nat_setup_info(ct, &mr->range[0], IP_NAT_MANIP_SRC, 0);
}
static unsigned int
@@ -74,7 +74,7 @@
/* Connection must be valid and new. */
NF_CT_ASSERT(ct && (ctinfo == IP_CT_NEW || ctinfo == IP_CT_RELATED));
- return nf_nat_setup_info(ct, &mr->range[0], IP_NAT_MANIP_DST);
+ return nf_nat_setup_info(ct, &mr->range[0], IP_NAT_MANIP_DST, 0);
}
static int ipt_snat_checkentry(const struct xt_tgchk_param *par)
@@ -115,7 +115,7 @@
&ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.u3.ip :
&ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.u3.ip);
- return nf_nat_setup_info(ct, &range, HOOK2MANIP(hooknum));
+ return nf_nat_setup_info(ct, &range, HOOK2MANIP(hooknum), 0);
}
int nf_nat_rule_find(struct sk_buff *skb,
diff -uNr linux-source-3.2/net/ipv4/netfilter/nf_nat_sip.c linux-source-3.2.new/net/ipv4/netfilter/nf_nat_sip.c
--- linux-source-3.2/net/ipv4/netfilter/nf_nat_sip.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv4/netfilter/nf_nat_sip.c 2014-07-11 10:07:40.744653437 +0800
@@ -259,7 +259,7 @@
range.flags = (IP_NAT_RANGE_MAP_IPS | IP_NAT_RANGE_PROTO_SPECIFIED);
range.min = range.max = exp->saved_proto;
range.min_ip = range.max_ip = exp->saved_ip;
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_DST, 0);
/* Change src to where master sends to, but only if the connection
* actually came from the same source. */
@@ -268,7 +268,7 @@
range.flags = IP_NAT_RANGE_MAP_IPS;
range.min_ip = range.max_ip
= ct->master->tuplehash[!exp->dir].tuple.dst.u3.ip;
- nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC);
+ nf_nat_setup_info(ct, &range, IP_NAT_MANIP_SRC, 0);
}
}
diff -uNr linux-source-3.2/net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c linux-source-3.2.new/net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c
--- linux-source-3.2/net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/ipv6/netfilter/nf_conntrack_proto_icmpv6.c 2014-07-11 09:12:04.320625207 +0800
@@ -165,7 +165,7 @@
*ctinfo = IP_CT_RELATED;
- h = nf_conntrack_find_get(net, zone, &intuple);
+ h = nf_conntrack_find_get(net, zone, &intuple, skb->ij);
if (!h) {
pr_debug("icmpv6_error: no match\n");
return -NF_ACCEPT;
diff -uNr linux-source-3.2/net/netfilter/ipvs/ip_vs_nfct.c linux-source-3.2.new/net/netfilter/ipvs/ip_vs_nfct.c
--- linux-source-3.2/net/netfilter/ipvs/ip_vs_nfct.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/ipvs/ip_vs_nfct.c 2014-07-11 11:10:42.224685433 +0800
@@ -270,7 +270,7 @@
__func__, ARG_TUPLE(&tuple), ARG_CONN(cp));
h = nf_conntrack_find_get(ip_vs_conn_net(cp), NF_CT_DEFAULT_ZONE,
- &tuple);
+ &tuple, 0);
if (h) {
ct = nf_ct_tuplehash_to_ctrack(h);
/* Show what happens instead of calling nf_ct_kill() */
diff -uNr linux-source-3.2/net/netfilter/nf_conntrack_core.c linux-source-3.2.new/net/netfilter/nf_conntrack_core.c
--- linux-source-3.2/net/netfilter/nf_conntrack_core.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/nf_conntrack_core.c 2014-07-11 10:48:19.948674076 +0800
@@ -316,7 +316,7 @@
*/
static struct nf_conntrack_tuple_hash *
____nf_conntrack_find(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple, u32 hash)
+ const struct nf_conntrack_tuple *tuple, u32 hash, int ij)
{
struct nf_conntrack_tuple_hash *h;
struct hlist_nulls_node *n;
@@ -327,7 +327,7 @@
*/
local_bh_disable();
begin:
- hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[bucket], hnnode) {
+ hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[ij][bucket], hnnode) {
if (nf_ct_tuple_equal(tuple, &h->tuple) &&
nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
NF_CT_STAT_INC(net, found);
@@ -352,24 +352,24 @@
struct nf_conntrack_tuple_hash *
__nf_conntrack_find(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple)
+ const struct nf_conntrack_tuple *tuple, int ij)
{
return ____nf_conntrack_find(net, zone, tuple,
- hash_conntrack_raw(tuple, zone));
+ hash_conntrack_raw(tuple, zone), ij);
}
EXPORT_SYMBOL_GPL(__nf_conntrack_find);
/* Find a connection corresponding to a tuple. */
static struct nf_conntrack_tuple_hash *
__nf_conntrack_find_get(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple, u32 hash)
+ const struct nf_conntrack_tuple *tuple, u32 hash, int ij)
{
struct nf_conntrack_tuple_hash *h;
struct nf_conn *ct;
rcu_read_lock();
begin:
- h = ____nf_conntrack_find(net, zone, tuple, hash);
+ h = ____nf_conntrack_find(net, zone, tuple, hash, ij);
if (h) {
ct = nf_ct_tuplehash_to_ctrack(h);
if (unlikely(nf_ct_is_dying(ct) ||
@@ -390,26 +390,27 @@
struct nf_conntrack_tuple_hash *
nf_conntrack_find_get(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple)
+ const struct nf_conntrack_tuple *tuple, int ij)
{
return __nf_conntrack_find_get(net, zone, tuple,
- hash_conntrack_raw(tuple, zone));
+ hash_conntrack_raw(tuple, zone), ij);
}
EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
static void __nf_conntrack_hash_insert(struct nf_conn *ct,
unsigned int hash,
- unsigned int repl_hash)
+ unsigned int repl_hash,
+ int ij)
{
struct net *net = nf_ct_net(ct);
hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
- &net->ct.hash[hash]);
+ &net->ct.hash[ij][hash]);
hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
- &net->ct.hash[repl_hash]);
+ &net->ct.hash[ij][repl_hash]);
}
-void nf_conntrack_hash_insert(struct nf_conn *ct)
+void nf_conntrack_hash_insert(struct nf_conn *ct, int ij)
{
struct net *net = nf_ct_net(ct);
unsigned int hash, repl_hash;
@@ -419,7 +420,7 @@
hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
- __nf_conntrack_hash_insert(ct, hash, repl_hash);
+ __nf_conntrack_hash_insert(ct, hash, repl_hash, ij);
}
EXPORT_SYMBOL_GPL(nf_conntrack_hash_insert);
@@ -479,12 +480,12 @@
/* See if there‘s one in the list already, including reverse:
NAT could have grabbed it without realizing, since we‘re
not in the hash. If there is, we lost race. */
- hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
+ hlist_nulls_for_each_entry(h, n, &net->ct.hash[skb->ij][hash], hnnode)
if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
&h->tuple) &&
zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
goto out;
- hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
+ hlist_nulls_for_each_entry(h, n, &net->ct.hash[skb->ij][repl_hash], hnnode)
if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
&h->tuple) &&
zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
@@ -514,7 +515,7 @@
* guarantee that no other CPU can find the conntrack before the above
* stores are visible.
*/
- __nf_conntrack_hash_insert(ct, hash, repl_hash);
+ __nf_conntrack_hash_insert(ct, hash, repl_hash, skb->ij);
NF_CT_STAT_INC(net, insert);
spin_unlock_bh(&nf_conntrack_lock);
@@ -537,7 +538,7 @@
for NAT). */
int
nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
- const struct nf_conn *ignored_conntrack)
+ const struct nf_conn *ignored_conntrack, int ij)
{
struct net *net = nf_ct_net(ignored_conntrack);
struct nf_conntrack_tuple_hash *h;
@@ -550,7 +551,7 @@
* least once for the stats anyway.
*/
rcu_read_lock_bh();
- hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
+ hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[ij][hash], hnnode) {
ct = nf_ct_tuplehash_to_ctrack(h);
if (ct != ignored_conntrack &&
nf_ct_tuple_equal(tuple, &h->tuple) &&
@@ -571,7 +572,7 @@
/* There‘s a small race here where we may free a just-assured
connection. Too bad: we‘re in trouble anyway. */
-static noinline int early_drop(struct net *net, unsigned int hash)
+static noinline int early_drop(struct net *net, unsigned int hash, int ij)
{
/* Use oldest entry, which is roughly LRU */
struct nf_conntrack_tuple_hash *h;
@@ -582,7 +583,7 @@
rcu_read_lock();
for (i = 0; i < net->ct.htable_size; i++) {
- hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash],
+ hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[ij][hash],
hnnode) {
tmp = nf_ct_tuplehash_to_ctrack(h);
if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
@@ -636,7 +637,7 @@
__nf_conntrack_alloc(struct net *net, u16 zone,
const struct nf_conntrack_tuple *orig,
const struct nf_conntrack_tuple *repl,
- gfp_t gfp, u32 hash)
+ gfp_t gfp, u32 hash, int ij)
{
struct nf_conn *ct;
@@ -651,7 +652,7 @@
if (nf_conntrack_max &&
unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
- if (!early_drop(net, hash_bucket(hash, net))) {
+ if (!early_drop(net, hash_bucket(hash, net), ij)) {
atomic_dec(&net->ct.count);
if (net_ratelimit())
printk(KERN_WARNING
@@ -713,9 +714,9 @@
struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
const struct nf_conntrack_tuple *orig,
const struct nf_conntrack_tuple *repl,
- gfp_t gfp)
+ gfp_t gfp, int ij)
{
- return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
+ return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0, ij);
}
EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
@@ -753,7 +754,7 @@
}
ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
- hash);
+ hash, skb->ij);
if (IS_ERR(ct))
return (struct nf_conntrack_tuple_hash *)ct;
@@ -840,7 +841,7 @@
/* look for tuple match */
hash = hash_conntrack_raw(&tuple, zone);
- h = __nf_conntrack_find_get(net, zone, &tuple, hash);
+ h = __nf_conntrack_find_get(net, zone, &tuple, hash, skb->ij);
if (!h) {
h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
skb, dataoff, hash);
@@ -1170,7 +1171,7 @@
spin_lock_bh(&nf_conntrack_lock);
for (; *bucket < net->ct.htable_size; (*bucket)++) {
- hlist_nulls_for_each_entry(h, n, &net->ct.hash[*bucket], hnnode) {
+ hlist_nulls_for_each_entry(h, n, &net->ct.hash[0][*bucket], hnnode) {
ct = nf_ct_tuplehash_to_ctrack(h);
if (iter(ct, data))
goto found;
@@ -1297,6 +1298,7 @@
static void nf_conntrack_cleanup_net(struct net *net)
{
+ int i = 0;
i_see_dead_people:
nf_ct_iterate_cleanup(net, kill_all, NULL);
nf_ct_release_dying_list(net);
@@ -1305,7 +1307,9 @@
goto i_see_dead_people;
}
- nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
+ for (i = 0; i < 3; i++) {
+ nf_ct_free_hashtable(net->ct.hash[i], net->ct.htable_size);
+ }
nf_conntrack_ecache_fini(net);
nf_conntrack_tstamp_fini(net);
nf_conntrack_acct_fini(net);
@@ -1364,7 +1368,7 @@
{
int i, bucket;
unsigned int hashsize, old_size;
- struct hlist_nulls_head *hash, *old_hash;
+ struct hlist_nulls_head *hash[3], *old_hash[3];
struct nf_conntrack_tuple_hash *h;
struct nf_conn *ct;
@@ -1378,10 +1382,19 @@
hashsize = simple_strtoul(val, NULL, 0);
if (!hashsize)
return -EINVAL;
-
- hash = nf_ct_alloc_hashtable(&hashsize, 1);
- if (!hash)
- return -ENOMEM;
+ {
+ int k = 0;
+ for (k = 0; k < 3; k++) {
+ hash[k] = nf_ct_alloc_hashtable(&hashsize, 1);
+ if (!hash[k]) {
+ int j = 0;
+ for (j = 0; j < k; j++) {
+ //free hash[j];
+ }
+ return -ENOMEM;
+ }
+ }
+ }
/* Lookups in the old hash might happen in parallel, which means we
* might get false negatives during connection lookup. New connections
@@ -1390,24 +1403,38 @@
*/
spin_lock_bh(&nf_conntrack_lock);
for (i = 0; i < init_net.ct.htable_size; i++) {
- while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
- h = hlist_nulls_entry(init_net.ct.hash[i].first,
+ int k = 0;
+ for (k = 0; k < 3; k++) {
+ while (!hlist_nulls_empty(&init_net.ct.hash[k][i])) {
+ h = hlist_nulls_entry(init_net.ct.hash[k][i].first,
struct nf_conntrack_tuple_hash, hnnode);
ct = nf_ct_tuplehash_to_ctrack(h);
hlist_nulls_del_rcu(&h->hnnode);
bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
hashsize);
- hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
+ hlist_nulls_add_head_rcu(&h->hnnode, &hash[k][bucket]);
+ }
}
}
old_size = init_net.ct.htable_size;
- old_hash = init_net.ct.hash;
+ old_hash[0] = init_net.ct.hash[0];
+ old_hash[1] = init_net.ct.hash[1];
+ old_hash[2] = init_net.ct.hash[2];
init_net.ct.htable_size = nf_conntrack_htable_size = hashsize;
- init_net.ct.hash = hash;
+ {
+ int k = 0;
+ for (k = 0; k < 3; k++) {
+ init_net.ct.hash[k] = hash[k];
+ }
+ }
spin_unlock_bh(&nf_conntrack_lock);
-
- nf_ct_free_hashtable(old_hash, old_size);
+ {
+ int k = 0;
+ for (k = 0; k < 3; k++) {
+ nf_ct_free_hashtable(old_hash[k], old_size);
+ }
+ }
return 0;
}
EXPORT_SYMBOL_GPL(nf_conntrack_set_hashsize);
@@ -1494,6 +1521,7 @@
static int nf_conntrack_init_net(struct net *net)
{
int ret;
+ int i = 0;
atomic_set(&net->ct.count, 0);
INIT_HLIST_NULLS_HEAD(&net->ct.unconfirmed, UNCONFIRMED_NULLS_VAL);
@@ -1520,11 +1548,13 @@
}
net->ct.htable_size = nf_conntrack_htable_size;
- net->ct.hash = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
- if (!net->ct.hash) {
- ret = -ENOMEM;
- printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
- goto err_hash;
+ for (i = 0; i < 3; i++) {
+ net->ct.hash[i] = nf_ct_alloc_hashtable(&net->ct.htable_size, 1);
+ if (!net->ct.hash[i]) {
+ ret = -ENOMEM;
+ printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
+ goto err_hash;
+ }
}
ret = nf_conntrack_expect_init(net);
if (ret < 0)
@@ -1548,7 +1578,9 @@
err_acct:
nf_conntrack_expect_fini(net);
err_expect:
- nf_ct_free_hashtable(net->ct.hash, net->ct.htable_size);
+ for (i = 0; i < 3; i++) {
+ nf_ct_free_hashtable(net->ct.hash[i], net->ct.htable_size);
+ }
err_hash:
kmem_cache_destroy(net->ct.nf_conntrack_cachep);
err_cache:
diff -uNr linux-source-3.2/net/netfilter/nf_conntrack_helper.c linux-source-3.2.new/net/netfilter/nf_conntrack_helper.c
--- linux-source-3.2/net/netfilter/nf_conntrack_helper.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/nf_conntrack_helper.c 2014-07-11 10:57:25.424678691 +0800
@@ -227,7 +227,7 @@
hlist_nulls_for_each_entry(h, nn, &net->ct.unconfirmed, hnnode)
unhelp(h, me);
for (i = 0; i < net->ct.htable_size; i++) {
- hlist_nulls_for_each_entry(h, nn, &net->ct.hash[i], hnnode)
+ hlist_nulls_for_each_entry(h, nn, &net->ct.hash[0][i], hnnode)
unhelp(h, me);
}
}
diff -uNr linux-source-3.2/net/netfilter/nf_conntrack_netlink.c linux-source-3.2.new/net/netfilter/nf_conntrack_netlink.c
--- linux-source-3.2/net/netfilter/nf_conntrack_netlink.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/nf_conntrack_netlink.c 2014-07-11 11:02:54.168681473 +0800
@@ -692,7 +692,7 @@
last = (struct nf_conn *)cb->args[1];
for (; cb->args[0] < net->ct.htable_size; cb->args[0]++) {
restart:
- hlist_nulls_for_each_entry(h, n, &net->ct.hash[cb->args[0]],
+ hlist_nulls_for_each_entry(h, n, &net->ct.hash[0][cb->args[0]],
hnnode) {
if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
continue;
@@ -920,7 +920,7 @@
if (err < 0)
return err;
- h = nf_conntrack_find_get(net, zone, &tuple);
+ h = nf_conntrack_find_get(net, zone, &tuple, 0);
if (!h)
return -ENOENT;
@@ -986,7 +986,7 @@
if (err < 0)
return err;
- h = nf_conntrack_find_get(net, zone, &tuple);
+ h = nf_conntrack_find_get(net, zone, &tuple, skb->ij);
if (!h)
return -ENOENT;
@@ -1336,7 +1336,7 @@
struct nf_conntrack_helper *helper;
struct nf_conn_tstamp *tstamp;
- ct = nf_conntrack_alloc(net, zone, otuple, rtuple, GFP_ATOMIC);
+ ct = nf_conntrack_alloc(net, zone, otuple, rtuple, GFP_ATOMIC, 0);
if (IS_ERR(ct))
return ERR_PTR(-ENOMEM);
@@ -1446,7 +1446,7 @@
if (err < 0)
goto err2;
- master_h = nf_conntrack_find_get(net, zone, &master);
+ master_h = nf_conntrack_find_get(net, zone, &master, 0);
if (master_h == NULL) {
err = -ENOENT;
goto err2;
@@ -1460,7 +1460,7 @@
tstamp->start = ktime_to_ns(ktime_get_real());
add_timer(&ct->timeout);
- nf_conntrack_hash_insert(ct);
+ nf_conntrack_hash_insert(ct, 0);
rcu_read_unlock();
return ct;
@@ -1503,9 +1503,9 @@
spin_lock_bh(&nf_conntrack_lock);
if (cda[CTA_TUPLE_ORIG])
- h = __nf_conntrack_find(net, zone, &otuple);
+ h = __nf_conntrack_find(net, zone, &otuple, 0);
else if (cda[CTA_TUPLE_REPLY])
- h = __nf_conntrack_find(net, zone, &rtuple);
+ h = __nf_conntrack_find(net, zone, &rtuple, 0);
if (h == NULL) {
err = -ENOENT;
@@ -2020,7 +2020,7 @@
return err;
/* Look for master conntrack of this expectation */
- h = nf_conntrack_find_get(net, zone, &master_tuple);
+ h = nf_conntrack_find_get(net, zone, &master_tuple, 0);
if (!h)
return -ENOENT;
ct = nf_ct_tuplehash_to_ctrack(h);
diff -uNr linux-source-3.2/net/netfilter/nf_conntrack_pptp.c linux-source-3.2.new/net/netfilter/nf_conntrack_pptp.c
--- linux-source-3.2/net/netfilter/nf_conntrack_pptp.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/nf_conntrack_pptp.c 2014-07-11 11:04:11.824682130 +0800
@@ -148,7 +148,7 @@
pr_debug("trying to timeout ct or exp for tuple ");
nf_ct_dump_tuple(t);
- h = nf_conntrack_find_get(net, zone, t);
+ h = nf_conntrack_find_get(net, zone, t, 0);
if (h) {
sibling = nf_ct_tuplehash_to_ctrack(h);
pr_debug("setting timeout of conntrack %p to 0\n", sibling);
diff -uNr linux-source-3.2/net/netfilter/nf_conntrack_standalone.c linux-source-3.2.new/net/netfilter/nf_conntrack_standalone.c
--- linux-source-3.2/net/netfilter/nf_conntrack_standalone.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/nf_conntrack_standalone.c 2014-07-11 10:56:08.276678039 +0800
@@ -59,7 +59,7 @@
for (st->bucket = 0;
st->bucket < net->ct.htable_size;
st->bucket++) {
- n = rcu_dereference(hlist_nulls_first_rcu(&net->ct.hash[st->bucket]));
+ n = rcu_dereference(hlist_nulls_first_rcu(&net->ct.hash[0][st->bucket]));
if (!is_a_nulls(n))
return n;
}
@@ -80,7 +80,7 @@
}
head = rcu_dereference(
hlist_nulls_first_rcu(
- &net->ct.hash[st->bucket]));
+ &net->ct.hash[0][st->bucket]));
}
return head;
}
diff -uNr linux-source-3.2/net/netfilter/xt_connlimit.c linux-source-3.2.new/net/netfilter/xt_connlimit.c
--- linux-source-3.2/net/netfilter/xt_connlimit.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/xt_connlimit.c 2014-07-11 11:07:55.712684024 +0800
@@ -117,7 +117,7 @@
/* check the saved connections */
hlist_for_each_entry_safe(conn, pos, n, hash, node) {
found = nf_conntrack_find_get(net, NF_CT_DEFAULT_ZONE,
- &conn->tuple);
+ &conn->tuple, 0);
found_ct = NULL;
if (found != NULL)
diff -uNr linux-source-3.2/net/netfilter/xt_CT.c linux-source-3.2.new/net/netfilter/xt_CT.c
--- linux-source-3.2/net/netfilter/xt_CT.c 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/net/netfilter/xt_CT.c 2014-07-11 11:06:38.648683372 +0800
@@ -81,7 +81,7 @@
goto err1;
memset(&t, 0, sizeof(t));
- ct = nf_conntrack_alloc(par->net, info->zone, &t, &t, GFP_KERNEL);
+ ct = nf_conntrack_alloc(par->net, info->zone, &t, &t, GFP_KERNEL, 0);
ret = PTR_ERR(ct);
if (IS_ERR(ct))
goto err2;
H文件patch：
diff -uNr linux-source-3.2/include/linux/skbuff.h linux-source-3.2.new/include/linux/skbuff.h
--- linux-source-3.2/include/linux/skbuff.h 2014-02-01 21:18:39.000000000 +0800
+++ linux-source-3.2.new/include/linux/skbuff.h 2014-07-11 07:00:38.696558485 +0800
@@ -474,6 +474,7 @@
unsigned char *head,
*data;
unsigned int truesize;
+ int ij;
atomic_t users;
};
diff -uNr linux-source-3.2/include/linux/version.h linux-source-3.2.new/include/linux/version.h
--- linux-source-3.2/include/linux/version.h 1970-01-01 08:00:00.000000000 +0800
+++ linux-source-3.2.new/include/linux/version.h 2014-07-11 07:01:03.972558699 +0800
@@ -0,0 +1,2 @@
+#define LINUX_VERSION_CODE 197174
+#define KERNEL_VERSION(a,b,c) (((a) << 16) + ((b) << 8) + (c))
diff -uNr linux-source-3.2/include/net/netfilter/nf_conntrack_core.h linux-source-3.2.new/include/net/netfilter/nf_conntrack_core.h
--- linux-source-3.2/include/net/netfilter/nf_conntrack_core.h 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/include/net/netfilter/nf_conntrack_core.h 2014-07-11 08:59:04.872618612 +0800
@@ -50,7 +50,7 @@
/* Find a connection corresponding to a tuple. */
extern struct nf_conntrack_tuple_hash *
nf_conntrack_find_get(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple);
+ const struct nf_conntrack_tuple *tuple, int ij);
extern int __nf_conntrack_confirm(struct sk_buff *skb);
diff -uNr linux-source-3.2/include/net/netfilter/nf_conntrack.h linux-source-3.2.new/include/net/netfilter/nf_conntrack.h
--- linux-source-3.2/include/net/netfilter/nf_conntrack.h 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/include/net/netfilter/nf_conntrack.h 2014-07-11 09:05:35.412621916 +0800
@@ -176,7 +176,7 @@
conntrack). */
extern int
nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
- const struct nf_conn *ignored_conntrack);
+ const struct nf_conn *ignored_conntrack, int ij);
/* Return conntrack_info and tuple hash for given skb. */
static inline struct nf_conn *
@@ -207,9 +207,9 @@
extern struct nf_conntrack_tuple_hash *
__nf_conntrack_find(struct net *net, u16 zone,
- const struct nf_conntrack_tuple *tuple);
+ const struct nf_conntrack_tuple *tuple, int ij);
-extern void nf_conntrack_hash_insert(struct nf_conn *ct);
+extern void nf_conntrack_hash_insert(struct nf_conn *ct, int ij);
extern void nf_ct_delete_from_lists(struct nf_conn *ct);
extern void nf_ct_insert_dying_list(struct nf_conn *ct);
@@ -284,7 +284,7 @@
nf_conntrack_alloc(struct net *net, u16 zone,
const struct nf_conntrack_tuple *orig,
const struct nf_conntrack_tuple *repl,
- gfp_t gfp);
+ gfp_t gfp, int ij);
static inline int nf_ct_is_template(const struct nf_conn *ct)
{
diff -uNr linux-source-3.2/include/net/netfilter/nf_nat.h linux-source-3.2.new/include/net/netfilter/nf_nat.h
--- linux-source-3.2/include/net/netfilter/nf_nat.h 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/include/net/netfilter/nf_nat.h 2014-07-11 10:05:22.568652268 +0800
@@ -53,7 +53,7 @@
/* Set up the info structure to map into this range. */
extern unsigned int nf_nat_setup_info(struct nf_conn *ct,
const struct nf_nat_range *range,
- enum nf_nat_manip_type maniptype);
+ enum nf_nat_manip_type maniptype, int ij);
/* Is this tuple already taken? (not by us)*/
extern int nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
diff -uNr linux-source-3.2/include/net/netns/conntrack.h linux-source-3.2.new/include/net/netns/conntrack.h
--- linux-source-3.2/include/net/netns/conntrack.h 2014-01-03 12:33:36.000000000 +0800
+++ linux-source-3.2.new/include/net/netns/conntrack.h 2014-07-11 08:56:06.816617105 +0800
@@ -13,7 +13,7 @@
unsigned int expect_count;
unsigned int htable_size;
struct kmem_cache *nf_conntrack_cachep;
- struct hlist_nulls_head *hash;
+ struct hlist_nulls_head *hash[3];
struct hlist_head *expect_hash;
struct hlist_nulls_head unconfirmed;
struct hlist_nulls_head dying;重新编译内核之后，效果十分可观，按照下面的脚本加入巨量的conntrack也丝毫不会影响最大连接数：
for((i=1;i<220;i++));do for((j=1;j<254;j++));do conntrack -I conntrack -s 172.129.$j.$i -d $j.$i.192.$j -p udp -t 4000 --sport 245 --dport 2001;conntrack -I conntrack -s 192.13.$i.$j -d $i.$j.24.19 -p tcp --state ESTABLISHED -t 4000 --sport 67 --dport 505;conntrack -I conntrack -s 172.129.$j.$i -d $j.$i.192.$j -p icmp -t 4000;done;done >/dev/null 2>&1
除了分了多张hash之外，另外的好处在于锁的粒度变细了，每次在操作conntrack的时候只需要锁住和自己相关的hash表的锁即可，当然，在本次的优化中没有涉及这一点。和ipset结合起来就更猛了，本身ipset就可以定义一个集合，而该集合则可以用来计算conntrack的hash值，多表版本的conntrack和ipset结合，这在效果上不就是多级hash表吗？而且表面上还是iptables配置的。
郑重声明：本站内容如果来自互联网及其他传播媒体，其版权均属原媒体及文章作者所有。转载目的在于传递更多信息及用于网络分享，并不代表本站赞同其观点和对其真实性负责，也不构成任何其他建议。
