一:背景
公司最近exchange邮件系统来构建高可用邮件系统架构。前端负载均衡就是使用了微软的NLB来实现cas的负载均衡。但是方案实施工程中也发现了一些问题，使用NLB在网络中产生大量广播报，出现丢包严重问题。后改用haproxy替代NLB实现cas的负载均衡，但也这就导致了一个问题，最明显的就是用户通过Haproxy来访问邮件系统后 真正达到邮件系统的地址都是Haproxy的地址，在垃圾邮件过滤的时候就无法实现基于IP的过滤，并且也无法记录IP地址信息。为 了解决这个问题我google了很多方案，可以使用硬负载，硬负载可以实现全透明代理让后端邮件服务器获取到用户的真实IP，还有一种解决方案就是 Haproxy的全透明代理。接下来将会介绍下Haproxy的全透明代理的部署。
二:实验拓扑
三:方案部署
配置Haproxy
#!/bin/bash wget http://haproxy.1wt.eu/download/1.4/src/haproxy-1.4.25.tar.gz tar zxvf haproxy-1.4.25.tar.gz cd haproxy-1.4.25 yum install gcc gcc-c++ autoconf automake -y make TARGET=linux2628 arch=x86_64 USE_LINUX_TPROXY=1 #重点USE_LINUX_TPROXY=1用于编译支持TPTOXY make install mkdir /etc/haproxy cp examples/haproxy.cfg /etc/haproxy cp examples/haproxy.init /etc/init.d/haproxy chmod +x /etc/init.d/haproxy cp haproxy /usr/sbin/
修改配置文件/etc/haproxy/haproxy.cfg global log 127.0.0.1 local0 maxconn 409600 chroot /usr/local/share # uid 501 # gid 501 daemon nbproc 1 pidfile /usr/local/haproxy/logs/haproxy.pid # debug defaults log global maxconn 100000 contimeout 500000 clitimeout 3600000 srvtimeout 3600000 option redispatch retries 6 frontend mail.domain.com mode http bind 0.0.0.0:80 log global option tcplog redirect location https://mail.domain.com/owa frontend owa_443 mode tcp bind 0.0.0.0:443 default_backend pool_443 log global option tcplog backend pool_443 balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m source 0.0.0.0 usesrc clientip #重点,TPROXY需要加上这行。 server cas01 10.130.170.130:443 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:443 check inter 5000 weight 1 rise 2 fall 3 frontend smtp_25 mode tcp bind 0.0.0.0:25 default_backend pool_smtp log global option tcplog backend pool_smtp balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m source 0.0.0.0 usesrc clientip #重点,TPROXY需要加上这行。 server cas01 10.130.170.130:25 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:25 check inter 5000 weight 1 rise 2 fall 3 frontend pop_995 mode tcp bind 0.0.0.0:995 default_backend pool_pop log global option tcplog backend pool_pop balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas01 10.130.170.130:995 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:995 check inter 5000 weight 1 rise 2 fall 3 frontend pop_993 mode tcp bind 0.0.0.0:993 default_backend pool_993 log global option tcplog backend pool_993 balance source # option forwardfor # option originalto option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas01 10.130.170.130:993 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:993 check inter 5000 weight 1 rise 2 fall 3 frontend pop_135 mode tcp bind 0.0.0.0:135 default_backend pool_135 log global option tcplog backend pool_135 balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas01 10.130.170.130:135 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:135 check inter 5000 weight 1 rise 2 fall 3 frontend pop_593 mode tcp bind 0.0.0.0:593 default_backend pool_593 log global option tcplog backend pool_593 balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas01 10.130.170.130:593 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:593 check inter 5000 weight 1 rise 2 fall 3 frontend pop_60001 mode tcp bind 0.0.0.0:60001 default_backend pool_60001 log global option tcplog backend pool_60001 balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas02 10.130.170.130:60001 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:60001 check inter 5000 weight 1 rise 2 fall 3 frontend pop_55000 mode tcp bind 0.0.0.0:55000 default_backend pool_55000 log global option tcplog backend pool_55000 balance source option redispatch option abortonclose option persist stick on src stick-table type ip size 10240k expire 240m server cas02 10.130.170.130:55000 check inter 5000 weight 1 rise 2 fall 3 server cas02 10.130.170.131:55000 check inter 5000 weight 1 rise 2 fall 3 frontend vs_stats :8081 mode http log global option httplog default_backend stats_backend backend stats_backend mode http stats enable stats uri /stats stats auth admin:admin
由于ExchangeRPC的端口是动态端口，haproxy必须使用固定端口，我这里改成了60001和55000，修改注册表即可，这里不做介绍。
配置TProxy代码如下vi iptables.sh
#!/bin/bash /sbin/iptables -F /sbin/iptables -t mangle -N DIVERT /sbin/iptables -t mangle -A PREROUTING -p tcp -m socket -j DIVERT /sbin/iptables -t mangle -A DIVERT -j MARK --set-mark 1 /sbin/iptables -t mangle -A DIVERT -j ACCEPT /sbin/ip rule add fwmark 1 lookup 100 /sbin/ip route add local 0.0.0.0/0 dev lo table 100
上面的代码目的是为了让所有进入网卡的mangle表的包都打上标记，然后新增一条路由规则将这些打了标记的数据包发送至本地回环接口进行处理。
chmod +x iptables.sh ./iptables.sh
修改配置 /etc/sysctl.conf net.ipv4.ip_forward = 1 net.ipv4.conf.default.rp_filter = 2 net.ipv4.conf.all.rp_filter = 2 net.ipv4.conf.eth0.rp_filter = 0 net.ipv4.conf.all.send_redirects = 1 net.ipv4.conf.default.send_redirects = 1
到此为此基于TProxy的Haproxy全透明代理完成了。为真实实现高可用，haproxy也是单点故障，keepalived也不在这里介绍。
然后将两台cas server的网关改为haproxy的ip。现在可以查看到垃圾邮件的ip。
原文地址：http://246244.blog.51cto.com/236244/1568457
