首先我们要明白,Openwrt 是一个 GNU/Linux 的发行版, 和其他大多数的发行版一样,Openwrt 的防火墙同样也是基于 iptables .

其次,Openwrt支持两种途径配置 iptables ,一种就是 Openwrt 自己的 UCI 方式,另一种就是传统的 Linux 方式.

本文要探讨的是如何通过 UCI 方式来配置 iptables , UCI 的方式就是通过配置 /etc/config/firewall 这个文件来完成的.

至于传统的 Linux 方式,请参考篇文章,netfilter.

firewall 文件的结构

  default
  　这是 firewall 文件的第一个小节
  zone
  　可以有数个 zone , zone 又可以包含数个 network interfaces
  forwarding
  　位于的 zone 下面, 主要作用是允许数据封包转发
  rule 以及 redirect
  　可以看作是 zone 子集, 用来扩展进一步的封包限制.

　　我们随便打开一个 Openwrt 的分支版本软件包下载页面,例如 http://downloads.openwrt.org//backfire/10.03.1/brcm63xx/packages/, 我们会发现有很多相关的软件包.

• iptables_1.4.6-3.1_brcm63xx.ipk

这个软件包是 iptables 本身,其实 iptables 早已被大多数官方打包好并公开发行的 firmware 之中.

• kmod-ipt-*

这个开头的软件包,是 kernel space 层级的模块 vanilla)

• iptables-mod-*

这种开头的软件包,是 user space 层级的模块,有一部分也被包含进官方打包好并公开发行的 firmware 之中. 　比如 iptables-mod-nat_1.4.6-3.1_brcm63xx.ipk 这个软件就是为了提供基本的NAT功能,针对 SNAT DNAT MASQUERADE 这三个 Targets,以及针对 nat 这个 Tables 的.

下面是防火墙可以定义的配置项概要。最小的防火墙配置通常包含一个default节，至少两个zones（lan 和 wan）和一个forwarding允许数据包由lan转发到wan。

defaults节定义了不依赖于特定区域的防火墙全局设置。 本节可以定义以下选项：

A zone section groups one more interfaces and serves as a source or destination for forwardings, rules and redirects. Masquerading (NAT) of outgoing traffic is controlled on a per-zone basis.

The options below are defined within zone sections:

名称	类型	是否必需	缺省	描述
name	zone name	yes	(none)	Unique zone name
network	list	no	(none)	List of interfaces attached to this zone, if ommitted, the value of name is used by default
masq	boolean	no	0	Specifies whether outgoing zone traffic should be masqueraded – this is typically enabled on the wan zone
masq_src	list of subnets	no	0.0.0.0/0	Limit MASQUERADING to the given source subnets. Negation is possible by prefixing the subnet with !, multiple subnets are allowed. Only supported by the Firewall v2 package, version 11 and above or Firewall v1, version 13 and above
masq_dest	list of subnets	no	0.0.0.0/0	Limit MASQUERADING to the given destination subnets. Negation is possible by prefixing the subnet with !, multiple subnets are allowed. Only supported by the Firewall v2 package, version 11 and above or Firewall v1, version 13 and above
conntrack	boolean	no	1 if masquerading is used, 0otherwise	Force connection tracking for this zone (see Note on connection tracking)
mtu_fix	boolean	no	0	Enable MSS clamping for outgoing zone traffic
input	string	no	DROP	Default policy (ACCEPT, REJECT, DROP) for incoming zone traffic
forward	string	no	DROP	Default policy (ACCEPT, REJECT, DROP) for forwarded zone traffic
output	string	no	DROP	Default policy (ACCEPT, REJECT, DROP) for outgoing zone traffic
family	string	no	any	Protocol family (ipv4, ipv6 or any) to generate iptables rules for. Only supported by the Firewall v2 package
log	boolean	no	0	Create log rules for rejected and dropped traffic in this zone. Only supported by the Firewall v2 package, version 10 and above
log_limit	string	no	10/minute	Limits the amount of log messages per interval. Only supported by the Firewall v2 package, version 10 and above

The forwarding sections control the traffic flow between zones and may enable MSS clamping for specific directions. Only one direction is covered by a forwarding rule. To allow bidirectional traffic flows between two zones, two forwardings are required, with src and dest reversed in each.

Below is a listing of allowed option within forwardings:

名称	类型	是否必需	缺省	描述
src	zone name	yes	(none)	Specifies the traffic source zone, must refer to one of the defined zone names
dest	zone name	yes	(none)	Specifies the traffic destination zone, must refer to one of the defined zone names
mtu_fix	boolean	no	0	Enable MSS clamping for traffic flowing from the source zone to the destination zone (Deprecated and moved to zone sections in 8.09.2+)
family	string	no	any	Protocol family (ipv4, ipv6 or any) to generate iptables rules for. Only supported by the Firewall v2 package

The iptables rules generated for this section rely on the state match which needs connection tracking to work. At least one of the src or dest zones needs to have connection tracking enabled through either the masq or the conntrack option.

Port forwardings (DNAT) are defined by redirect sections. All incoming traffic on the specified source zone which matches the given rules will be directed to the specified internal host.

The options below are valid for redirects:

名称	类型	是否必需	缺省	描述
src	zone name	yes for DNATtarget	(none)	Specifies the traffic source zone, must refer to one of the defined zone names. For typical port forwards this usually is wan
src_ip	ip address	no	(none)	Match incoming traffic from the specified source ip address
src_dip	ip address	yes for SNATtarget	(none)	For DNAT, match incoming traffic directed at the given destination ip address. For SNAT rewrite the source address to the given address.
src_mac	mac address	no	(none)	Match incoming traffic from the specified mac address
src_port	port or range	no	(none)	Match incoming traffic originating from the given source port or port range on the client host
src_dport	port or range	no	(none)	For DNAT, match incoming traffic directed at the given destination port or port range on this host. For SNATrewrite the source ports to the given value.
proto	protocol name or number	yes	tcpudp	Match incoming traffic using the given protocol
dest	zone name	yes for SNATtarget	(none)	Specifies the traffic destination zone, must refer to one of the defined zone names.
dest_ip	ip address	yes for DNATtarget	(none)	For DNAT, redirect matched incoming traffic to the specified internal host. For SNAT, match traffic directed at the given address.
dest_port	port or range	no	(none)	For DNAT, redirect matched incoming traffic to the given port on the internal host. For SNAT, match traffic directed at the given ports.
target	string	no	DNAT	NAT target (DNAT or SNAT) to use when generating the rule Only supported by the Firewall v2, version 12 and above or Firewall v1, version 15 and above
family	string	no	any	Protocol family (ipv4, ipv6 or any) to generate iptables rules for. Only supported by the Firewall v2 package
reflection	boolean	no	1	Disables NAT reflection for this redirect if set to 0 – applicable to DNAT targets. Only supported by the Firewall v2, version 11 and above or Firewall v1, version 13 and above

Sections of the type rule can be used to define basic accept or reject rules to allow or restrict access to specific ports or hosts. Like redirects the rules are tied to the givensource zone and match incoming traffic occuring there.

Valid options for this section are:

名称	类型	是否必需	缺省	描述
src	zone name	yes	(none)	Specifies the traffic source zone, must refer to one of the defined zone names.
src_ip	ip address	no	(none)	Match incoming traffic from the specified source ip address
src_mac	mac address	no	(none)	Match incoming traffic from the specified mac address
src_port	port or range	no	(none)	Match incoming traffic originating from the given source port or port range on the client host if tcp or udp is specified as protocol
proto	protocol name or number	no	tcpudp	Match incoming traffic using the given protocol. Can be one of tcp, udp, tcpudp, udplite, icmp, esp, ah, sctp, or all or it can be a numeric value, representing one of these protocols or a different one. A protocol name from /etc/protocols is also allowed. The number 0 is equivalent to all.
dest	zone name	no	(none)	Specifies the traffic destination zone, must refer to one of the defined zone names. If specified, the rule applies to forwarded traffic else it is treated as input rule.
dest_ip	ip address	no	(none)	Match incoming traffic directed to the specified destination ip address
dest_port	port or range	no	(none)	Match incoming traffic directed at the given destination port or port range on this host if tcp or udp is specified as protocol
target	string	yes	DROP	Firewall action (ACCEPT, REJECT, DROP) for matched traffic
family	string	no	any	Protocol family (ipv4, ipv6 or any) to generate iptables rules for. Only supported by the Firewall v2 package

It is possible to include custom firewall scripts by specifying one or more include sections in the firewall configuration.

There is only one possible parameter for includes:

Included scripts may contain arbitary commands, for example advanced iptables rules or tc commands required for traffic shaping.

When writing custom iptables rules remember to use -I (insert) instead of -A (append) to ensure that the created rules appear before the generic ones.

The default configuration accepts all LAN traffic, but blocks all incoming WAN traffic on ports not currently used for connections or NAT. To open a port for a service, add a rulesection:

config rule
        option src              wan
        option dest_port        22
        option target           ACCEPT
        option proto            tcp

This example enables machines on the internet to use SSH to access your router.

This example forwards http (but not HTTPS) traffic to the webserver running on 192.168.1.10:

config redirect
        option src       wan
        option src_dport 80
        option proto     tcp
        option dest_ip   192.168.1.10

Source NAT changes an outgoing packet outgoing packet destined for the system so that is looks as though the system is the source of the packet.

Define source NAT for UDP and TCP traffic directed to port 123 originating from the host with the IP address 10.55.34.85. The source address is rewritten to 63.240.161.99.

config redirect
        option src              lan
        option dest             wan
        option src_ip           10.55.34.85
        option src_dip          63.240.161.99
        option dest_port        123
        option target           SNAT

When used alone, Source NAT is used to restrict a computer’s access to the internet, but allow a it to access a few services my manually forwarding what appear to be a few local services, e.g. NTP to the internet. While DNAT hides the local network from the internet, SNAT hides the internet from the local network.

Source NAT and destination NAT are combined and used dynamically in IP masquerading to make computers with private (192.168.x.x, etc.) IP address to appear on the internet with the system’s public WAN ip address.

Most users won’t want this. It’s usage is similar to SNAT, but as the the destination IP address isn’t changed, machines on the destination network need to be aware that they’ll receive and answer requests from a public IP address that isn’t necessarily theirs. Port forwarding in this fashion is typically used for load balancing.

config redirect
        option src              wan
        option src_dport        80
        option dest             lan
        option dest_port        80
        option proto            tcp

The following rule blocks all connection attempts to the specified host address.

config rule
        option src              lan
        option dest             wan
        option dest_ip          123.45.67.89
        option target           REJECT

The following rule blocks all connection attempts from the client to the Internet.

config rule
        option src              lan
        option dest             wan
        option src_mac          00:00:00:00:00
        option target           REJECT

The example below creates a forward rule rejecting traffic from lan to wan on the ports 1000-1100.

config rule
        option src              lan
        option dest             wan
        option dest_port        1000-1100
        option proto            tcpudp
        option target           REJECT

The rule below redirects all outgoing HTTP traffic from lan through a proxy server listening at port 3128 on the router itself.

config redirect
	option src              lan
	option proto            tcp
	option src_dport        80
	option dest_port        3128

The following rule redirects all outgoing HTTP traffic from lan through an external proxy at 192.168.1.100 listening on port 3128.

config redirect
        option src              lan
        option proto            tcp
        option src_ip           !192.168.1.100
        option src_dport        80
        option dest_ip          192.168.1.100
        option dest_port        3128

The following rule redirects all WAN ports for all protocols to the internal host 192.168.1.2.

config redirect
	option src              wan
	option proto            all
	option dest_ip          192.168.1.2

This example enables proper forwarding of IPSec traffic through the wan.

# AH protocol
config rule
        option src              wan
        option dest             lan
        option proto            ah
        option target           ACCEPT

# ESP protocol
config rule
        option src              wan
        option dest             lan
        option proto            esp
        option target           ACCEPT

For some configurations you also have to open port 500/UDP.

# ISAKMP protocol
config rule
        option src              wan
        option dest             lan
        option proto            udp
        option src_port         500
        option dest_port        500
        option target           ACCEPT

This example is for IPv6 tunnels only, and does not apply to native dual-stack interfaces.
This example only applies to the firewall v2 package and later (only in Trunk, not yet in Backfire).

In order to be able to address the IPv6 tunnel interface with uci firewall rules, you need to cover it in the network configuration first. Assuming an established SixXs tunnel, the example below declares a logical interface wan6 on top of the sixxs tunnel interface:

config interface wan6
	option proto            none
	option ifname           sixxs

IPv6 packets are by default not forwarded from lan to your wan6 interface, make sure to add net.ipv6.conf.all.forwarding=1 in /etc/sysctl.conf to enable it per名称nently. Add the following sections to the firewall config to cover the wan6 interface:

config zone
	option name             wan6
	option input            REJECT
	option output           ACCEPT
	option forward          REJECT
	option family           ipv6
	option mtu_fix          0

config forwarding
	option src              lan
	option dest             wan6

The family option ensures that the zone and all associated entries (rule, forwarding and redirect sections) are only added to ip6tables but not iptables.

传统的 iptables 规则, 是指在 Shell 命令行模式下指定一个包含规则定义的外部文件, 其实可以在本配置文件中, 指定多个配置文件, 如下:

config include
       option path /etc/firewall.user

config include
       option path /etc/firewall.vpn

需要注意的是, Openwrt 是以 UCI 方式来配置 iptables 的, 这与传统的 Linux 习惯是不一样的, 上面那两个指定的配置文件, 里面包含的规则写法必须与传统的 Linux 习惯一样. 那么如何配置上面那两个规则文件呢,请参考这篇文章 → netfilter.

After a configuration change, firewall rules are rebuilt by executing /etc/init.d/firewall restart; calling /etc/init.d/firewall stop will flush all rules and set the policies to ACCEPT on all standard chains. To manually start the firewall, call /etc/init.d/firewall start.

The firewall can be permananently disabled by executing /etc/init.d/firewall disable. Note that disable does not flush the rules, so it might be required to issue a stopbefore. Use enable to activate the firewall again.

In addition to includes it is possible to let the firewall execute hotplug handlers when interfaces are added to a zone or removed from it. This is useful to create rules for interfaces with dynamic ip configurations (dhcp, pppoe) on the fly.

Each time an interface is added or removed from a zone, all scripts in the /etc/hotplug.d/firewall/ directory are executed. Scripts must be named in the form NN-name with NN being a numeric index between 00 and 99. The name can be freely choosen.

Once a handler script is invoked, the informations about the event are passed through the environment. The table below lists defined variables and their meaning.

The decision whether to drop or to reject traffic should be done on a case-by-case basis. Many people see dropping traffic as a security advantage over rejecting it because it exposes less information to a hypothetical attacker. While dropping slightly increases security, it can also complicate the debugging of network issues or cause unwanted side-effects on client programs.

If traffic is rejected, the router will respond with an icmp error message (“destination port unreachable”) causing the connection attempt to fail immediately. This also means that for each connection attempt a certain amount of response traffic is generated. This can actually harm if the firewall is “attacked” with many simultanous connection attempts, the resulting “backfire” of icmp responses can clog up all available upload and make the connection unusable (DoS).

When connection attempts are droppped the client is not aware of the blocking and will continue to re-transmit its packets until the connection eventually times out. Depending on the way the client software is implemented, this could result in frozen or hanging programs that need to wait until a timeout occurs before they’re able to continue.

DROP

• less information is exposed
• less attack surface
• client software may not cope well with it (hangs until connection times out)
• may complicate network debugging (where was traffic dropped and why)

REJECT

• may expose information (like the ip at which traffic was actually blocked)
• client software can recover faster from rejected connection attempts
• network debugging easier (routing and firewall issues clearly distinguishable)

By default, the firewall will disable connection tracking for a zone if no masquerading is enabled. This is achieved by generating NOTRACK firewall rules matching all traffic passing via interfaces referenced by the firewall zone. The purpose of NOTRACK is to speed up routing and save memory by circumventing resource intensive connection tracking in cases where it is not needed. You can check if connection tracking is disabled by issuing iptables -t raw -vnL, it will list all rules, check for NOTRACK target.

NOTRACK will render certain ipables extensions unusable, for example the MASQUERADE target or the state match will not work!

If connection tracking is required, for example by custom rules in /etc/firewall.user, the conntrack option must be enabled in the corresponding zone to disable NOTRACK. It should appear as option 'conntrack' '1' in the right zone in /etc/config/firewall. For further information see http://security.maruhn.com/iptables-tutorial/x4772.html .

If you make a mistake you can delete a rule this way. First, issue this command:

# iptables -L -t raw --line-numbers

Now to delete, e.g. the third rule from chain OUTPUTexecute:

# iptables -t raw -D OUTPUT 3

1. http://www.netfilter.org/
netfilter框架的官方主页

2. http://zh.wikipedia.org/zh/TCP/IP%E5%8D%8F%E8%AE%AE
整个互联网的网络协议都以 TCP/IP 为基础,对这个协议簇有一般浅层的了解,会有助于掌握 iptables.

3. <TCP/IP Illustrated, Volume 1: The Protocols>
ISBN-10: 0321336313
作者: W. Richard Stevens
国内译作<TCP/IP协议详解 卷1>，目前已经有第二版．

4. <Modern Operating Systems>
作者: Andrew S. Tanenbaum
如果继续深入学习或应用 iptables ,必然会涉及到操作系统的内核知识,这本书可以说是一个相当的权威.例如会理解上文提及的 kernel space 与 user space 有何不同