nf_conntrack: table full, dropping packet 连接跟踪表已满,开始丢包 的解决办法
中午业务说机器不能登录,我通过USM管理界面登录单板的时候发现机器没有僵死,然后一看日志,g一下子就明白了
tail -2000 /var/log/messages
Apr 10 12:48:35 bj-push-pushserver83 kernel: [95129.138804] __ratelimit: 16523 callbacks suppressed (“连接跟踪表已满,开始丢包”!相信不少用iptables的同学都会见过这个错误信息吧)
Apr 10 12:48:35 bj-xx kernel: [95129.138806] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.138974] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.139142] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.139566] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.139747] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.139823] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.140188] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.140435] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.140508] nf_conntrack: table full, dropping packet.
Apr 10 12:48:35 bj-xx kernel: [95129.141133] nf_conntrack: table full, dropping packet.
Apr 10 12:48:38 bj-xx kernel: [95131.483097] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:49:01 bj-xx /usr/sbin/cron[9492]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:49:38 bj-xx kernel: [95191.382486] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:50:01 bj-xx /usr/sbin/cron[9761]: (root) CMD (/opt/huawei/logs/LoadRst/suseRst.sh 2>/dev/null)
Apr 10 12:50:01 bj-xx /usr/sbin/cron[9762]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:50:38 bj-xx kernel: [95251.283552] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:51:01 bj-xx /usr/sbin/cron[9990]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:51:38 bj-xx kernel: [95311.185024] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:52:01 bj-xx /usr/sbin/cron[10232]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:52:38 bj-xx kernel: [95371.082714] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:52:59 bj-xx sshd[9994]: pam_unix2(sshd:auth): conversation failed
Apr 10 12:52:59 bj-xx sshd[9994]: error: ssh_msg_send: write
Apr 10 12:53:01 bj-xx /usr/sbin/cron[10891]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:53:38 bj-xx kernel: [95430.983871] possible SYN flooding on port 443. Sending cookies.
Apr 10 12:54:01 bj-xx /usr/sbin/cron[11097]: (root) CMD (/usr/bin/tsar –cron > /dev/null 2>&1)
Apr 10 12:54:04 bj-xx sshd[11094]: pam_tally(sshd:account): unknown option: reset
Apr 10 12:54:04 bj-xx sshd[11094]: Accepted publickey for root from 183.62.156.75 port 16959 ssh2
Apr 10 12:54:38 bj-xx kernel: [95490.883402] possible SYN flooding on port 443. Send
都是脚本和任务计划惹的祸
脚本内容
cat /opt/xx/logs/LoadRst/suseRst.sh
!/bin/bash
cd dirname $0
loadnum=uptime|awk -F':' '{print $4}'|awk -F',' '{print $1*1000}'
fileDate=date +"%Y%m%d_%H:%M:%S"
echo $fileDate
echo $loadnum
loadnum_ora=uptime|awk -F':' '{print $4}'|awk -F',' '{print $2}'
softirq=top -bn 1|awk '/ksoftirqd/ {print $9}'|head -1
echo -e $fileDate >>log
echo $softirq >>log
if [ $loadnum -ge “900” ]
then
echo “asdfasdf”
echo -e $fileDate >>log
/sbin/rcSuSEfirewall2 restart >> log 2>&1
else
echo -e “${fileDate}:success” >>log
fi
任务计划
crontab -l
DO NOT EDIT THIS FILE - edit the master and reinstall.
(/tmp/crontab.XXXXWNPsHE installed on Wed Apr 9 20:10:57 2014)
(Cron version V5.0 –
Id:crontab.c,v1.122004/01/2318:56:42vixieExp
)
/5 * * * /opt/xx/logs/LoadRst/suseRst.sh 2>/dev/null
0 0 * * * /opt/xx/logs/Firewall_log/tar-firewall.sh >/dev/null 2>&1
解决办法
一、关闭防火墙。 简单粗暴,直接有效
/etc/init.d/SuSEfirewall2_init stop
/etc/init.d/SuSEfirewall2_setup stop
切记:在防火墙关闭状态下,不要通过iptables指令(比如 iptables -nL)来查看当前状态!因为这样会导致防火墙被启动,而且规则为空。虽然不会有任何拦截效果,但所有连接状态都会被记录,浪费资源且影响性能并可能导致防火墙主动丢包!
二、加大防火墙跟踪表的大小,优化对应的系统参数
1、状态跟踪表的最大行数的设定,理论最大值
CONNTRACK_MAX = RAMSIZE (in bytes) / 16384 / (ARCH / 32)
以64G的64位操作系统为例
CONNTRACK_MAX = 64*1024*1024*1024/16384/2 = 2097152
即时生效请执行:
sysctl –w net.netfilter.nf_conntrack_max = 2100000
或者
vi /etc/sysctl.conf
net.netfilter.nf_conntrack_max = 2100000
sysctl -p
2、其哈希表大小通常为总表的1/8,最大为1/2。
CONNTRACK_BUCKETS = CONNTRACK_MAX / 8
同样64G的64位操作系统,哈希最佳范围是 262144 ~ 1048576 。
运行状态中查看
sysctl net.netfilter.nf_conntrack_buckets
通过文件 /sys/module/nf_conntrack/parameters/hashsize 进行设置。
或者新建 /etc/modprobe.d/iptables.conf,重新加载模块才生效:
options nf_conntrack hashsize=262144
3、还有些相关的系统参数sysctl -a | grep nf_conntrack
可以调优(/etc/sysctl.conf ):
net.netfilter.nf_conntrack_max = 1048576
net.netfilter.ip_conntrack_tcp_timeout_established = 3600
net.netfilter.nf_conntrack_tcp_timeout_close_wait = 60
net.netfilter.nf_conntrack_tcp_timeout_fin_wait = 120
net.netfilter.nf_conntrack_tcp_timeout_time_wait = 120
三、使用祼表,添加“不跟踪”标识。如下示例更适合桌面系统或随意性强的服务器。因为它开启了连接的状态机制,方便和外部通信。修改 /etc/sysconfig/iptables 文件:
*raw
对TCP连接不启用追踪,解决ip_contrack满导致无法连接的问题
-A PREROUTING -p tcp -m tcp –dport 80 -j NOTRACK
-A PREROUTING -p tcp -m tcp –dport 22 -j NOTRACK
-A PREROUTING -p tcp -m tcp –dport 21 -j NOTRACK
-A PREROUTING -p tcp -m tcp –dport 11211 -j NOTRACK
-A PREROUTING -p tcp -m tcp –dport 60000:60100 -j NOTRACK
-A PREROUTING -p tcp -s 192.168.10.1 -j NOTRACK
-A OUTPUT -p tcp -m tcp –sport 80 -j NOTRACK
-A OUTPUT -p tcp -m tcp –sport 22 -j NOTRACK
-A OUTPUT -p tcp -m tcp –sport 21 -j NOTRACK
-A OUTPUT -p tcp -m tcp –sport 11211 -j NOTRACK
-A OUTPUT -p tcp -m tcp –sport 60000:60100 -j NOTRACK
-A OUTPUT -p tcp -s 192.168.10.1 -j NOTRACK
COMMIT
*filter
允许ping
-A INPUT -p icmp -j ACCEPT
对本地回路、第5张网卡放行
-A INPUT -i lo -j ACCEPT
-A INPUT -i eth4 -j ACCEPT
连接状态跟踪,已建立的连接允许传输数据
-A INPUT -m state –state ESTABLISHED,RELATED,INVALID,UNTRACKED -j ACCEPT
filter表里存在但在raw里不存在的,默认会进行连接状态跟踪
-A INPUT -s 192.168.10.31 -p tcp –dport 2669 -j ACCEPT
-A INPUT -j REJECT –reject-with icmp-host-prohibited
-A FORWARD -j REJECT –reject-with icmp-host-prohibited
COMMIT
或者干脆对所有连接都关闭跟踪,不跟踪任何连接状态。不过规则就限制比较严谨,进出都需要显式申明。示例如下:
*raw
对TCP/UDP连接不启用追踪,解决nf_contrack满导致无法连接的问题
-A PREROUTING -p tcp -j NOTRACK
-A PREROUTING -p udp -j NOTRACK
-A OUTPUT -p tcp -j NOTRACK
-A OUTPUT -p udp -j NOTRACK
COMMIT
*filter
允许ping
-A INPUT -p icmp -j ACCEPT
对本地回路和eth1放行
-A INPUT -i lo -j ACCEPT
-A INPUT -i eth1 -j ACCEPT
只允许符合条件的连接进行传输数据
-A INPUT -p tcp –dport 22 -j ACCEPT
-A INPUT -p tcp –sport 80 -j ACCEPT
-A INPUT -p udp –sport 53 -j ACCEPT
-A INPUT -p udp –sport 123 -j ACCEPT
出去的包都不限制
-A OUTPUT -p tcp -j ACCEPT
-A OUTPUT -p udp -j ACCEPT
输入和转发的包不符合规则的全拦截
-A INPUT -j REJECT –reject-with icmp-host-prohibited
-A FORWARD -j REJECT –reject-with icmp-host-prohibited
COMMIT
效果如下图:
四、删除连接跟踪模块lsmod | grep nf_conntrack
,不使用连接状态的跟踪功能。
1、删除nf_conntrack和相关的依赖模块,示例:
rmmod nf_conntrack_ipv4
rmmod nf_conntrack_ipv6
rmmod xt_state
rmmod xt_CT
rmmod xt_conntrack
rmmod iptable_nat
rmmod ipt_REDIRECT
rmmod nf_nat
rmmod nf_conntrack
2、禁用跟踪模块,把它加到黑名单(/etc/modprobe.d/blacklist.conf ):
禁用 nf_conntrack 模块
blacklist nf_conntrack
blacklist nf_conntrack_ipv6
blacklist xt_conntrack
blacklist nf_conntrack_ftp
blacklist xt_state
blacklist iptable_nat
blacklist ipt_REDIRECT
blacklist nf_nat
blacklist nf_conntrack_ipv4
3、去掉防火墙里所有和状态相关的配置(比如state状态,NAT功能),示例:
*filter
允许ping
-A INPUT -p icmp -j ACCEPT
对本地回路和第2张网卡放行
-A INPUT -i lo -j ACCEPT
-A INPUT -i eth1 -j ACCEPT
对端口放行
-A INPUT -p tcp –dport 1331 -j ACCEPT
对IP放行
-A INPUT -s 192.168.10.31 -j ACCEPT
允许本机进行DNS查询
-A INPUT -p udp –sport 53 -j ACCEPT
-A OUTPUT -p udp -j ACCEPT
-A INPUT -j REJECT –reject-with icmp-host-prohibited
-A FORWARD -j REJECT –reject-with icmp-host-prohibited
COMMIT
另外,防火墙的配置文件最好也改下,不要加载任何额外模块(/etc/sysconfig/iptables-config):
IPTABLES_MODULES=”” # 不需要任何附加模块
IPTABLES_MODULES_UNLOAD=”no” # 避免iptables重启后sysctl中对应的参数被重置为系统默认值
IPTABLES_SAVE_ON_STOP=”no”
IPTABLES_SAVE_ON_RESTART=”no”
IPTABLES_SAVE_COUNTER=”no”
IPTABLES_STATUS_NUMERIC=”yes”
IPTABLES_STATUS_VERBOSE=”no”
IPTABLES_STATUS_LINENUMBERS=”no”
往往我们对连接的跟踪都是基于操作系统的(netstat / ss ),防火墙的连接状态完全是它自身实现产生的。
本文出自 “俊” 博客,请务必保留此出处http://2364821.blog.51cto.com/2354821/1393736