FastNetMon

Author: Pavel Odintsov pavel.odintsov at gmail.com License: GPLv2

FastNetMon - High Performance Network DDoS and Load Analyzer with PCAP/ULOG2/PF_RING support. But I recommends only PF_RING variant because other variants is so slow and use big amount of CPU and expected big packetloss.

What we do? We can detect hosts in OUR network with big amount of packets per second (30 000 pps in standard configuration) incoming or outgoing from certain host. And we can call external bash script which can send notify, switch off server or blackhole this client.

Why we write it? Because we can't find any software for solving this problem not in proprietary world not in open sourcу. NetFlow based solutions is so slow and can't react on atatck with acceptable speed.

Main programm screen image:

Example for cpu load for Intel i7 2600 with Intel X540 NIC on 250 kpps load:

At first you should install PF_RING (you can install any latest version):

cd /usr/src
wget 'http://downloads.sourceforge.net/project/ntop/PF_RING/PF_RING-6.0.1.tar.gz?r=http%3A%2F%2Fsourceforge.net%2Fprojects%2Fntop%2Ffiles%2FPF_RING%2F&ts=1402307916&use_mirror=cznic' -OPF_RING-6.0.1.tar.gz
tar -xf PF_RING-6.0.1.tar.gz 
cd PF_RING-6.0.1
# Debian way
apt-get install build-essential bison flex linux-headers-$(uname -r) libnuma-dev
# CentOS
yum install -y make bison flex kernel-devel 
# CentOS openvz case 
yum install -y make bison flex vzkernel-devel

Build PF_RING kernel module:

cd kernel
make 
make install
modprobe pf_ring

You can use precompiled and statically linced version of this tool without any compiling:

mkdir /root/fastnetmon
cd /root/fastnetmon
wget  https://raw.githubusercontent.com/FastVPSEestiOu/fastnetmon/master/fastnetmon -Ofastnetmon
chmod +x fastnetmon
./fastnetmon eth0

If you want to use static version you can skip this guide to part about "networks_list".

Build lib (We disabled bpf because it requires linking to PCAP):

cd /usr/src/PF_RING-5.6.2/userland/lib
./configure  --disable-bpf --prefix=/opt/pf_ring
make
make install

Install FastNetMon:

   # Debian 7 Wheezy
   apt-get install -y git libpcap-dev g++ gcc libboost-all-dev make

   # If you need traffic counting
   apt-get install -y libhiredis-dev

   # Support libs
   apt-get install -y libgpm-dev libncurses5-dev liblog4cpp5-dev

   # If you need PF_RING abilities 
   apt-get install -y libnuma-dev

   # If you need ASN/geoip stats
   apt-get install -y libgeoip-dev 

   cd /usr/src
   git clone https://github.com/FastVPSEestiOu/fastnetmon.git
   cd fastnetmon

You should start fastnetmon using this options:

LD_LIBRARY_PATH=/opt/pf_ring/lib/ ./fastnetmon eth3,eth4

If you want to avoid LD_LIBRARY_PATH on every call you should add pf_ring path to system:

echo "/opt/pf_ring/lib" > /etc/ld.so.conf.d/pf_ring.conf
ldconfig -v

Select backend, we use PF_RING as default, if you need PCAP/ULOG2 you must change variable ENGINE in Makefile.

Compile it:

make

Download GeoIP database to current folder:

http://dev.maxmind.com/geoip/legacy/geolite/
http://download.maxmind.com/download/geoip/database/asnum/GeoIPASNum.dat.gz
gunzip GeoIPASNum.dat.gz

It's REQUIRED to add all your networks in CIDR form to file /etc/networks_list if form when one subnet on one line. Please aggregate your networks because long networks list will significatly slow down programm. And please change REDIS_SUPPORT = yes to no in Makefile if you do not need traffic counting feature. When you running this software in OpenVZ node you may did not specify networks explicitly, we can read it from file /proc/vz/veip.

You can add whitelist subnets in similar form to /etc/networks_whitelist (CIDR masks too).

Copy standard config file to /etc:

cp fastnetmon.conf /etc/fastnetmon.conf

Start it:

./fastnetmon eth1,eth2

Example program screen:

FastNetMon v1.0 all IPs ordered by: packets

Incoming Traffic        96667 pps 240 mbps
xx.xx.xx.xx             7950 pps 3 mbps
xx.xx.xx.xx             5863 pps 65 mbps
xx.xx.xx.xx             2306 pps 1 mbps
xx.xx.xx.xx             1535 pps 16 mbps
xx.xx.xx.xx             1312 pps 14 mbps
xx.xx.xx.xx             1153 pps 0 mbps
xx.xx.xx.xx             1145 pps 0 mbps

Outgoing traffic        133265 pps 952 mbps
xx.xx.xx.xx             7414 pps 4 mbps
xx.xx.xx.xx             5047 pps 4 mbps
xx.xx.xx.xx             3458 pps 3 mbps
xx.xx.xx.xx             2959 pps 35 mbps
xx.xx.xx.xx             2612 pps 29 mbps
xx.xx.xx.xx             2334 pps 26 mbps
xx.xx.xx.xx             1906 pps 21 mbps

Internal traffic        0 pps

Other traffic           1815 pps

Packets received:       6516913578
Packets dropped:        0
Packets dropped:        0.0 %

Ban list:
yy.yy.yy.yy/20613 pps incoming

Enable programm start on server startup, please add to /etc/rc.local this lines:

cd /root/fastnetmon && screen -S fastnetmon -d -m ./fastnetmon eth3,eth4

When incoming or outgoing attack arrives programm call bash script (when it exists): /usr/local/bin/notify_about_attack.sh two times. First time when threshold exceed (at this step we know IP, direction and power of attack). Second when we collect 100 packets for detailed audit what did happens.

Example of first notification:

subject: Myflower Guard: IP xx.xx.xx.xx blocked because incoming attack with power 120613 pps
body: blank

Example of second notification:

subject: Myflower Guard: IP xx.xx.xx.xx blocked because incoming attack with power 120613 pps
body:
IP: xx.zz.xx.1
Attack power: 95267 packets per second
Peak attack power: 269017 packets per second
Attack direction: incoming
Incoming traffic: 43 mbps
Outgoing traffic: 15 mbps
Incoming pps: 95267 packets per second
Outgoing pps: 31119 packets per second

2014-07-04 13:59:54.778872 xx.xx.xx.xx:80 > xx.xx.xx.xx:46804 protocol: tcp size: 233 bytes
2014-07-04 13:59:54.778874 xx.xx.xx.xx:80 > xx.xx.xx.xx:46804 protocol: tcp size: 233 bytes
2014-07-04 13:59:54.778875 xx.xx.xx.xx:80 > xx.xx.xx.xx:46804 protocol: tcp size: 233 bytes
2014-07-04 13:59:54.778877 xx.xx.xx.xx:46804 > xx.xx.xx.xx:80 protocol: tcp size: 52 bytes
2014-07-04 13:59:54.778878 xx.xx.xx.xx:46804 > xx.xx.xx.xx:80 protocol: tcp size: 52 bytes
2014-07-04 13:59:54.778882 xx.xx.xx.xx:80 > xx.xx.xx.xx:46804 protocol: tcp size: 233 bytes
2014-07-04 13:59:54.778884 xx.xx.xx.xx:80 > xx.xx.xx.xx:46804 protocol: tcp size: 233 bytes
2014-07-04 13:59:54.778885 xx.xx.xx.xx:46804 > xx.xx.xx.xx:80 protocol: tcp size: 52 bytes

I recommend you to disable CPU freq scaling for gain max performance (max frequency):

echo performance | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

You can use this script for irq balancing on heavy loaded networks:

#!/bin/bash

# from http://habrahabr.ru/post/108240/
ncpus=`grep -ciw ^processor /proc/cpuinfo`
test "$ncpus" -gt 1 || exit 1

n=0
for irq in `cat /proc/interrupts | grep eth | awk '{print $1}' | sed s/\://g`
do
    f="/proc/irq/$irq/smp_affinity"
    test -r "$f" || continue
    cpu=$[$ncpus - ($n % $ncpus) - 1]
    if [ $cpu -ge 0 ]
            then
                mask=`printf %x $[2 ** $cpu]`
                echo "Assign SMP affinity: eth queue $n, irq $irq, cpu $cpu, mask 0x$mask"
                echo "$mask" > "$f"
                let n+=1
    fi
done

You can find more info and graphics here

Running tool without root permissions:

useradd fastnetmon
setcap cap_net_admin+eip fastnetmon
su fastnetmon
./fastnetmon eth0,eth1

Please keep in mind when run tool on OpenVZ because without root permissions tool can't get all VE ips and you should pass it explicitly.

Debugging flags.

DUMP_ALL_PACKETS will enable all packets dumping to console. It's very useful for testing tool on non standard platforms.

DUMP_ALL_PACKETS=yes ./fastnetmon eth3,eth4

Recommended configuration options for ixgbe Intel X540 driver:

cat /etc/modprobe.d/ixgbe.conf 
options ixgbe IntMode=2,2 MQ=1,1 DCA=2,2 RSS=8,8 VMDQ=0,0 max_vfs=0,0 L2LBen=0,0 InterruptThrottleRate=1,1 FCoE=0,0 LRO=1,1 allow_unsupported_sfp=0,0

If you saw intel_idle in perf top with red higlihting you can disable it with following kernel params (more details you can find Performance_Tuning_Guide_for_Mellanox_Network_Adapters.pdf):

intel_idle.max_cstate=0 processor.max_cstate=1