NAME nfflowd - App to generate Cisco NetFlow from Linux netfilter VERSION version 1.113260 PURPOSE If you run a Linux based firewall, the "nfflowd" application in this distribution will generate a stream of Cisco NetFlow logging data for all TCP, UDP and ICMP connections passing through. What makes this different from other solutions is that it uses Linux netfilter's own connection tracking data, rather than observing a traffic capture. This is more efficient for your CPU, and allows tracking of NAT which would otherwise not be possible. No reconfiguration of the firewall ruleset or network is required. SYNOPSIS Make sure you have the "conntrack" and "flow-tools" packages installed on your Linux firewall system. write a small configuration describing your NetFlow collector: ~# cat > /etc/nfflowd.conf # source interface on this server args 172.16.1.1 # netflow collector address args 192.168.0.1 # netflow collector port number args 65001 ^D to start the logger as a daemon: ~# nfflowd start to specify a custom configuration other than /etc/nfflowd.conf: ~# nfflowd -c /other/config/file.ini DESCRIPTION The application "nfflowd" included in this distribution is designed to run as a daemon (background) process on a Linux based firewall running IPtables. Its purpose is to capture data from the Linux netfilter subsystem about each tracked data connection passing through the firewall, and log this in Cisco NetFlow format to a remote NetFlow collector server. We recommend the Argus 3 software for that collector (). With the default configuration you will also receive a local copy of the same data via "syslog", under the "local6" facility. "nfflowd" was written for NAT environments where you have a requirement to trace the IP address on the "private" side which made a particular connection or flow to a port/destination on the "public" (Internet) side. The NetFlow streams logged will allow you to trace back to that private IP. The daemon requires no reconfiguration of your firewall ruleset or network, but you may need to install one or two additional helper programs. These are commonly available in most Linux distributions - see "DEPENDENCIES", below. "nfflowd" is also efficient and accurate, as it uses Linux netfilter's own connection logging and tracking data. This is preferable to observing a promiscuous traffic capture, which may be CPU intensive and certainly cannot track NAT connections. What's special about NAT? An issue that's specific to NAT (rather than general Netflow from a Linux router) is that you really need a good and accurate record of the IP and port number translation. So there are *two* traffic flows, one before and one after the NAT. Both need logging, together with a way to correlate them. Traditionally such things are attempted using two instances of "softflowd" or "argus", one each side of the NAT. With accurate timestamping the two flows (pre and post-NAT) can be correlated and then the IP and port translation deduced. However this doesn't work for very long running flows, or where there are numerous flows to a destination, some of which are legitimate and some not. There exist cases where it is impossible to be sure that you are correlating the correct two flow records. Seeing as netfilter is already accurately tracking every connection, with NAT information, we can make use of that data. LIMITATIONS * It's assumed that Source IP NAT is taking place. If not, you may receive additional unecessary log lines, sorry. * Currently only TCP, UDP and ICMP traffic is logged. * The tool supports only IPv4 tracking. Is there a "conntrack" for IPv6? USAGE The "nfflowd" application included in this distribution is designed to run as a daemon (background) process. It also supports running in the foreground. You will need to provide a small amount of configuration, and otherwise there are sane defaults for all options. Here is the help text for the current version of "nfflowd": Usage: nfflowd [ -c file ] [ -f ] { start | stop | restart | install | check | help | version } -c file Specify configuration file (instead of /etc/nfflowd.conf) -f Run in the foreground (don't detach) start Starts a new nfflowd if there isn't one running already stop Stops a running nfflowd restart Stops a running nfflowd if one is running. Starts a new one. install Setup nfflowd to run automatically after reboot check Check the configuration file and report the daemon state help Display this usage info version Display the version of nfflowd Although the "reload" option is also available, it currently has no effect. Please note that the "install" option has not been tested by the author. Flow Output For each flow observed by "conntrack", three flow records are output by this tool. As explained below, three records are required to correlate the NAT operation which took place. The assumption of source IP NAT will be relaxed in a future release of this software. Here's an example of the three records generated by an HTTP requesst, which we'll walk through: 1248952480,737235000,279737,172.16.1.1,720,38052,136315,279737,10.0.195.106,74.125.6.225,0.0.0.0,0,0,62380,80,6,0,0 1248952480,737235000,279737,172.16.1.1,720,38052,136315,279737,192.76.7.154,74.125.6.225,0.0.0.0,0,0,62380,80,6,0,0 1248952480,737235000,279737,172.16.1.1,1432,2027040,136315,279737,74.125.6.225,192.76.7.154,0.0.0.0,0,0,80,62380,6,0,0 The format of these lines is "flow-tools" CSV for NetFlow version 5. Of course your NetFlow collector receives a proper UDP NetFlow payload, but here we are looking at the default Syslog output. The fields are listed in order in the table below: 1248952480 UNIX_SECS - Current count of seconds since 0000 UTC 1970 737235000 UNIX_NSECS - Residual nanoseconds since 0000 UTC 1970 279737 SYSUPTIME - Current time in milliseconds since the export device booted 172.16.1.1 EXADDR - Export device IP address 720 DPKTS - Packets in the flow 38052 DOCTETS - Total number of Layer 3 bytes in the packets of the flow 136315 FIRST - SysUptime at start of flow 279737 LAST - SysUptime at the time the last packet of the flow was received 10.0.195.106 SRCADDR - Source IP address 74.125.6.225 DSTADDR - Destination IP address 0.0.0.0 NEXTHOP - IP address of next hop router 0 INPUT - SNMP index of input interface 0 OUTPUT - SNMP index of output interface 62380 SRCPORT - TCP/UDP source port number or equivalent 80 DSTPORT - TCP/UDP destination port number or equivalent 6 PROT - IP protocol type (for example, TCP = 6; UDP = 17) 0 TOS - IP type of service (ToS) 0 TCP_FLAGS - Cumulative OR of TCP flags The first line of the flow triplet describes the connection (IP/port) source and desination from the perspective of the originating host on the "private" side (or behind) your NAT. The second line describes how the source NAT was applied by netfilter. In this case the original host's source IP was translated from 10.0.195.106 to 192.76.7.154, and the source port stayed the same (62380). The final line describes the return flow from that connection, so you can see how much traffic passed in each direction (NetFlow flows are unidirectional, so you do always get two flows for each connection anyway). We only log the public (post-NAT) IPs and port numbers, as that's enough data to complete the correlation. So in this example, host 10.0.195.106 make a TCP connection on port 80 to 74.125.6.225, sending 38052 bytes and receiving 2027040 bytes back. CONFIGURATION At minimum you need to provide a small configuration to let "nfflowd" know where your NetFlow collector server is. This by default can be located at "/etc/nfflowd.conf" although you can specify an alternate location using the "-c" command line option. The format of the file can be anything supported by Config::Any (i.e. YAML, JSON, etc), although we recommend using Config::General format, as in the examples below. The required configuration is the source IP on your firewall from which NetFlow packets are sent, and the IP/port numbers of the NetFlow collector server: # source interface on this server args 172.16.1.1 # netflow collector address args 192.168.0.1 # netflow collector port number args 65001 These three options must appear *in the exact order given above*. The other thing you might want to configure is whether and how a local copy of the NetFlow data is stored on the Linux firewall itself, via Syslog. The application uses Log::Dispatch and by default will log to Syslog facility "local6" with priority . Below is the default configuration, which you can override in your local "nfflowd" configuration file, but be warned! You *must* retain the "screen" dispatcher and configuration section, otherwise the application will stop working. You have the option to remove, change, or add to the "syslog" dispatcher only: dispatchers screen dispatchers syslog stderr 0 class Log::Dispatch::Screen min_level debug class Log::Dispatch::Syslog ident nfflowd min_level debug facility local6 "nfflowd" records the start and end of each connection reported by netfilter. To preserve memory, the application clears up its records of "unfinished" connections which are older than a set TTL. This defaults to one week, but can be changed by setting a seconds value using the following configuration: ttl 604800 In addition, you can control how each of the helper applications are loaded and configured. You would generally not need to do this, but here are the config options anyway: progname conntrack init_format -L format -E -e NEW,DESTROY -o timestamp -n progname flow-import format -z0 -f2 -V5 -m0xFF31EF progname flow-send format -V5 -s %s/%s/%s DEPENDENCIES Other than the standard contents of a Perl 5.8 distribution, you will need: * conntrack at * flow-tools at * Log::Dispatch::Configurator::Any version 1.0005 or later * Daemon::Generic * File::Slurp version 9999.06 or later * IPC::Run * Config::General * Config::Any version 0.15 or later SEE ALSO "pfflowd" at This is the BSD pf equivalent of this tool. "softflowd" at From the "pfflowd" stable, this tool requires a promiscuous traffic capture to generate data. It is therefore not fully stateful and cannot track NAT. http://www.cisco.com/en/US/tech/tk812/tsd_technology_support_protocol_ho me.html Cisco homepage for their NetFlow technology. THANKS David Ford of the University of Oxford's OxCERT team provided valuable input and testing infrastructure, many thanks. AUTHOR Oliver Gorwits COPYRIGHT AND LICENSE This software is copyright (c) 2011 by University of Oxford. This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.