Netflow processing tools
nfdump-1.7.x or nfdump unicorn is the current release of nfdump.
nfdump is a toolset in order to collect and process netflow/ipfix and sflow data, sent from netflow/sflow compatible devices.
The toolset contains several collectors to collect flow data:
The collected flow data is stored into files and can be process afterwards.
nfdump processes and lists the flows in many different output formats and
can create a wide range of statistics.
nfdump has a very powerful flow filter to process flows. The filter syn‐
tax is very similar to tcpdump, but adapted and extended for flow filter‐
ing. A flow filter may also contain arrays of many thousand IP addresses
etc. to search for specific records.
nfdump can aggreagte flows according to a user defined number of ele‐
ments. This masks certain elements and allows to sum up flow records
matching the same values.
The combination of flow filtering and aggregation as input for any flow
statistics allows complex flow processing. Pre‐filtered and aggregated
flow data may also be written back into a binary flow file, which again
may be processed with nfdump
nfdump can enrich the listing of flows with geo location information,
AS information and TOR exit node information. AS information is enriched only,
if it is not available in the original flow record. IP addresses can be tagged with a
two letter country code, or with a longer location label containing the geographic
region, country and city. The geo location and AS information is retrieved from the
optional geoDB database, created by the geolookup program from the nfdump
tools. geolookup uses the Maxmind database GeoDB or GeoLite2 to create a
binary lookup database for nfdump. Please check the geolooup(1) man page
for more details. IP adresses can be tagged as TOR exit nodes, from the optional
torDB database, created by the torlookup program. Please chaeck the torlookup(1)
man page for details.
There is also a go-nfdump module to read nfdump flows files in Golang.
nfdump-1.7.x is compatible to nfdump-1.6.18, which means it can read files
created with nfdump-1.6.18 or newer. Flow files created with earlier nfdump
versions may not contain all flow elements. If you have older files, it is
recommended to use nfdump-1.6.17 to update the records.
If you have lots of flows files from nfdump-1.6.x, it is recommended to convert
these to the new format. You can convert any old files from nfdump-1.6.x to nfdump-1.7
format by reading/writing files: ./nfdump -r old-flowfile -y -w new-flowfile
Please note, that only nfdump may read nfdump-1.6.x flow files. All other programs understand
the new file format only.
Note for NfSen users: If you use NfSen, you must upgrade NfSen to the latest Github version https://github.com/phaag/nfsen.
All specific binaries such as nfprofile and nftrack are still available with nfdump-1.7 but may be removed in future.
nfdump is now a multi-threaded program and uses parallel threads mainly for
reading, writing and processing flows as well as for sorting. This may result
in faster flow processing, depending on the tasks. The speedimprovement
also heavily depends on the hardware (SSD/HD) and flow compression
option.
For netflow v9 and IPFIX, nfdump now supports FNF and flexible length fields. This
improves compatibility with some exporters such as yaf and others.
Support for Cisco Network Based Application Recognition (NBAR).
Supports Maxmind geo location information to tag/geolocate IP addresses
and AS numbers.
Supports TOR exit node information to IP addresses as TOR exit nodes.
nfpcapd automatically uses TPACKETV3 for Linux or direct BPF sockets for
*BSD. This improves packet processing. It adds new options to collect MAC and
VLAN information if requested as well as the payload of the first packet. This
creates a lot of new possibilities in order to process and filter flows, such
as _nfdump -r flowfile ‘payload content “POST”‘
nfpcapd can now store flow files locally or can sent them to a remote nfcapd
collector.
Metric exports: By default, every 60s a flow summary statistics can be sent
to a UNIX socket. The corresponding program may be nfinflux to insert
these metrics into an influxDB or nfexporter for Prometheus monitoring.
nfdump supports a default config file tipically /usr/local/etc/nfdump.conf to
store user defined paths for the geolookup and torlookup database files as well as for
user defined named output formats ( -o ‘fmt:%ts .. ‘, -o ‘csv:%ts ..’). See the default
/usr/local/etc/nfdump.conf.dist file for an example.
The nfdump program suite also contains geolookup. It allows either
to enrich IP addresses by country codes/locations and may add potential
missing AS information. Flows may be filtered according to country codes.
( ex: src geo US ). geolookup may also be used as standalone program to lookup
IPs for AS/Geo information, similar to the famous Team Cymru whois service.
geolookup uses a local database, which allows to process as many requests as you have.
In order to use geolookup, you need either a free or paid Maxmind account
in order to convert the Maxmind .csv files into an nfdump vector data file.
geolookup needs to be enabled when running configure: —enable-maxmind
The nfdump program suite also contains torlookup. It allows either
to enrich IP addresses by a TOR exit flag. Flows may be filtered according to
TOR IP addresses ( ex: src ip tor ). torlookup may also be used as standalone program
to lookup IPs for TOR exit node intervals with as many requests as you have.
In order to use torlookup or the nfdump output enrichment , you need to create the
tordb first. See also the toorlookup(1) man page. torlookup needs to be enabled when
running configure: —enable-tor
NSEL (Network Event Security Logging) as well as NEL (NAT Event Logging) are technologies invented by CISCO and also use the netflow v9 protocol. However, NSEL and NEL are not flows as commonly known but rather Events! exported from specific devices such as CISCO ASA. nfdump supports Event looging as part of netflow v9.
Jun OS NAT Event Logging is mostly compatible with CISCO’s NAT Event Logging - mostly - it needs another data interpretation.
See —enable-jnat below
The toolset is build upon the autotools framework. Run ./autogen.sh first.
Afterwards ./configure make and make install should do the trick.
For various older Linuxes need a more modern compiler:
% yum install centos-release-scl
Then you can install GCC 8 and its C++ compiler:
% yum install devtoolset-8-gcc devtoolset-8-gcc-c++
To switch to a shell which defaults gcc and g++ to this GCC version, use:
% scl enable devtoolset-8 -- bash
% sudo apt-get install clang-10% CC=clang-10 ./configure ...
The following config options are available: ( see ./configure —help for the complete list)
Development and beta options
nfcapd - netflow collector daemon.
Collects the netflow data, sent from exporters and stores the flow records
into files. Automatically rotates files every n minutes. ( typically
every 5 min ) The netflow versions mentioned above are read transparently
Multiple netflow streams can be collected by a single or collector.
nfcapd can listen on IPv6 or IPv4. Furthermore multicast is supported.
nfdump - process collected netflow records.
Nfdump reads the netflow data from one or many files stored by nfcapd.
It’s filter syntax is similar to tcpdump ( pcap like ) but adapted for netflow.
If you like tcpdump you will like nfdump. nfdump displays netflow
data and/or creates top N statistics of flows, bytes, packets. nfdump
has a powerful and flexible flow aggregation including bi-directional
flows. The output format is user selectable and also includes a simple
csv format for post processing.
nfanon - anonymize netflow records
IP addresses in flow records are anonimized using the CryptoPAn method.
nfexpire - expire old netflow data
Manages data expiration. Sets appropriate limits. Used by NfSen.
nfreplay - netflow replay
Reads the netflow data from the files stored by nfcapd and sends it
over the network to another host.
sfcapd - sflow collector daemon
scfapd collects sflow data and stores it into nfcapd compatible files.
“sfcapd includes sFlow(TM) code, freely available from https://github.com/sflow/sflowtool.
nfpcapd - pcap to netflow collector daemon
nfpcapd listens on a network interface, or reads precollected pcap traffic.
It either stores locally flow records into nfcapd compatible files or sends
the flows to a remote nfcapd collector. It is nfcapd’s companion to convert
traffic directly into nfdump records. Nfpcap can optionally integrate lots of
meta data as well as prt of the payload. ( -o fat, payload)
geolookup - Geo location lookup program.
geolookup converts Maxmind’s .csv files into the nfdump vector DB. The
converted DB may be used as a standalone lookup tool, or be be used by
nfdump in order to automatically lookup country and location.
Please note: You need a legitimate Maxmind account (free or paid) in
order to download the files.
torlookup - TOR location lookup program.
torlookup converts tor information files into the nfdump vector DB. The
converted DB may be used as a standalone lookup tool, or be be used by
nfdump in order to automatically flag tor exit node IPs.
ft2nfdump - flow-tools flow converter
ft2nfdump converts flow-tools data into nfdump format.
nfprofile - netflow profiler. Required by NfSen
Reads the netflow data from the files stored by nfcapd. Filters the
netflow data according to the specified filter sets ( profiles ) and
stores the filtered data into files for later use.
nftrack - Port tracking decoder for NfSen plugin PortTracker.
nfreader - Framework for programmers
nfreader is a framework to read nfdump files for any other purpose.
Own C code can be added to process flows. nfreader is not installed
sfcapd and nfcapd can be used concurrently to collect netflow and sflow
data at the same time. Generic command line options apply to both
collectors likewise. sfcapd’s sflow decoding module is based on InMon’s
sflowtool code and supports similar fields as nfcapd does for netflow v9,
which is a subset of all available sflow fields in an sflow record.
More fields may be integrated in future versions of sfcapd.
Binary data files can optionally be compressed by using either LZO1X-1, LZ4, ZSTD or bzip2 compression
LZO is fastest but less efficient, LZ4 and ZSTD are fast and pretty efficient and bzip2 is slow but efficient.
By default LZO and LZ4 embedded without external dependancies. Bzip2 and ZSTD are optional libraries, which are automatically added, if they are found while configuring and compiling.
The standard configure process checks for the installed libraries lz4, bz2 and zstd and enables them if they are found.
configure understands the following options:
--with-lz4path=PATH Expect liblz4 installed in PATH; default /usr/local--with-zstdpath=PATH Expect libzstd installed in PATH; default /usr/local--with-bz2path=PATH Expect libbz2 installed in PATH; default /usr/local
If no option is given and no library is found that compression algorithm is disabled. For LZ4, if no library is found, the embedded version is used. Compression algorithms can also be explicitly disabled by setting with-xxxpath=no In the case of lz4, it disables the system installed version and uses the embedded one.
Recommendation
If you compress automatically flows while they are collected, use LZ4 -z=lz4 as a standard.
Notes: Bzip2 uses about 30 times more CPU than LZO1X-1. Use bzip2 to archive netflow
data, which may reduce the disk usage again by a factor of 2. The compression of flow files
can be changed any time with nfdump -J
For more details on each methde, see:
LZO1X-1: http://www.oberhumer.com/opensource/lzo
LZ4: https://github.com/lz4/lz4
ZSTD: https://github.com/facebook/zstd
bzip2: http://www.bzip.org
The goal of the design is to able to analyze netflow data from
the past as well as to track interesting traffic patterns
continuously. The amount of time back in the past is limited only
by the disk storage available for all the netflow data. The tools
are optimized for speed for efficient filtering. The filter rules
should look familiar to the syntax of tcpdump ( pcap compatible ).
All data is stored to disk, before it gets analyzed. This separates
the process of storing and analyzing the data.
The data is organized in a time-based fashion. Every n minutes
Analyzing the data can be done for a single file, or by concatenating
several files for a single output. The output is either ASCII text
or binary data, when saved into a file, ready to be processed again
with the same tools.
You may have several netflow sources - let’s say ‘router1’ ‘router2’
and so on. The data is organized as follows:
/flow_base_dir/router1/flow_base_dir/router2
which means router1 and router2 are subdirs of the flow_base_dir.
Although several flow sources can be sent to a single collector,
It’s recommended to have multiple collector on busy networks for
each source.
Example: Start two collectors on different ports:
nfcapd -D -S 2 -B 1024000 -w /flow_base_dir/router1 -p 23456nfcapd -D -S 2 -B 1024000 -w /flow_base_dir/router2 -p 23457
nfcapd can handle multiple flow sources.
All sources can go into a single file or can be split:
All into the same file:
nfcapd -D -S 2 -w /flow_base_dir/routers -p 23456
Collected on one port and split per source:
nfcapd -D -S 2 -n router1,172.16.17.18,/flow_base_dir/router1 \-n router2,172.16.17.20,/flow_base_dir/router2 -p 23456
See nfcapd(1) for a detailed explanation of all options.
Security: none of the tools requires root privileges, unless you have
a port < 1024. However, there is no access control mechanism in nfcapd.
It is assumed, that host level security is in place to filter the
proper IP addresses.
See the manual pages or use the -h switch for details on using each of
the programs. For any questions send email to peter@people.ops-trust.net
Configure your router to export netflow. See the relevant documentation
for your model.
A generic Cisco sample configuration enabling NetFlow on an interface:
ip address 192.168.92.162 255.255.255.224interface fastethernet 0/0ip route-cache flow
To tell the router where to send the NetFlow data, enter the following
global configuration command:
ip flow-export 192.168.92.218 9995ip flow-export version 5ip flow-cache timeout active 5
This breaks up long-lived flows into 5-minute segments. You can choose
any number of minutes between 1 and 60;
Netflow v9 full export example of a cisco 7200 with sampling enabled:
interface Ethernet1/0ip address 192.168.92.162 255.255.255.224duplex halfflow-sampler my-map!!flow-sampler-map my-mapmode random one-out-of 5!ip flow-cache timeout inactive 60ip flow-cache timeout active 1ip flow-capture fragment-offsetip flow-capture packet-lengthip flow-capture ttlip flow-capture vlan-idip flow-capture icmpip flow-capture ip-idip flow-capture mac-addressesip flow-export version 9ip flow-export template options export-statsip flow-export template options samplerip flow-export template options timeout-rate 1ip flow-export template timeout-rate 1ip flow-export destination 192.168.92.218 9995
See the relevant documentation for a full description of netflow commands
Note: Netflow version v5 and v7 have 32 bit counter values. The number of
packets or bytes may overflow this value, within the flow-cache timeout
on very busy routers. To prevent overflow, you may consider to reduce the
flow-cache timeout to lower values. All nfdump tools use 64 bit counters
internally, which means, all aggregated values are correctly reported.
The binary format of the data files is netflow version independent.
For speed reasons the binary format is machine architecture dependent, and
as such can not be exchanged between little and big endian systems.
Internally nfdump does all processing IP protocol independent, which means
everything works for IPv4 as well as IPv6 addresses.
See the nfdump(1) man page for details.
netflow version 9 and infix:
nfcapd supports a large range of netflow v9 and ipfix tags. Version 1.7 of nfdump
also supports FNF - flexible netflow, subtemplates and understands a few specific exporters such as yaf.
This collector collects sflow https://www.sflow.org exports. It is largely identical to nfcapd except it only understands sflow packets.
This collector is able to listen on a host interface and generates netflow data from the network data stream on that interface. It make use of PACKET_RX_RING to read packets on an interface device level (TPACKETV3) on Linux hosts or of the BPF interface - Berkeley Packet Filter on *BSD hosts which provides raw access to data link layers. Nfpcapd builds an internal netflow cache which is periodically written to disk or forwarded to an nfcapd server. As a special feature, nfpcpad may collect the first few bytes of a network connection, if requested to do so ( -o payload), which allows filter and evaluate the flows with nfdump later.
Listen on eth0 and store the flows locally. Set flow cache active timeout to 60s, inactive tiemout to 30s:
nfpcapd -D -S 2 -w /var/flows -i eth0 -e 60,30 -u daemon -g daemon
Listen on eth0 and forward flow data to nfcapd running on a remote host. Add tunnel infos, MAC addr, vlan labels and first packet payload to the flows:
nfpcapd -D -S 2 -H 192.168.168.40 -i eth0 -e 60,30 -o fat,payload -u daemon -g daemon
In order to evaluate the payload, nfdump has some simple payload decoders for DNS, ja3, ja3s, ja4 and a few other.
Alternatively nfpcapd can also convert existing cap files into flow data:
nfpcapd -S 2 -w /var/flows -r pcapfile.pcap -e 60,30 -o fat,payload
Links
Extensions: nfcapd supports a large number of v9 tags. It automatically add
extensions to store data for v9/IPFIX elements which are supported.
By default, the sampling rate is set to 1 (unsampled) or to any given value specified
by the -s cmd line option. If sampling information is found in the netflow stream,
it overwrites the default value. Sampling is automatically recognised when announced
in v9/ipfix option templates with tags set (#302, #304, #305, #306), ( #48, #49, #50 ), ( #34, #35)or in the unofficial v5 header hack. The sampling data is stored in the sampling PacketInterval/PacketSpace model. If announced differently, it is converted accordingly.
Note: Not all platforms (or vendor software versions) support exporting sampling information
in netflow data, even if sampling is configured. The number of bytes/packets in each
netflow record is automatically multiplied by the sampling rate. The total number of
flows is not changed as this is not accurate enough. (Small flows versus large flows)
If you like this project your company may consider sponsoring it :) https://github.com/sponsors/phaag