Stable Release v1.6.17
See the Changelog file for all changes in release 1.6.17
nfdump is a toolset in oder to collect and process netflow and sflow data, sent from netflow/sflow compatible devices. The toolset supports netflow v1, v5/v7,v9,IPFIX and SFLOW. nfdump supports IPv4 as well as IPv6.
nfdump is used as backend toolset for NfSen.
NSEL/ASA, NEL/NAT support
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.
Note: The older nfdump-1.5.8-2-NSEL is not compatible with nfdump > 1.6.9 which supports NSEL/NEL.
nfdump contains an IPFIX module for decoding IPFIX flow data. It does not support the full IPFIX definition.
- Supports basically same feature set of elements as netflow_v9 module
- Only UDP traffic is accepted no tCP/SCTP
- If you would like to see more IPFIX support, please contact me.
Building and config options
The toolset is build upon the autotools framework. Run
make install should do the trick.
The following config options are available:
Compile nfdump, to read and process NSEL/NEL event data; default is NO
Build the flow-tools to nfdump converter; default is NO
Build sflow collector sfcpad; default is NO
Build nfprofile used by NfSen; default is NO
Build nftrack used by PortTracker; default is NO
Build nfdump, to read nfdump data files created with nfdump 1.5.x; default is NO
Development and beta options
Insert lots of debug and development code into nfdump for testing and debugging; default is NO
Add code to nfcapd to read flow data also from pcap files; default is NO
Build experimental nfpcapd collector to create netflow data from interface traffic or precollected pcap traffic, similar to softflowd; default is NO
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 methode.
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 comaptible files. "sfcapd includes sFlow(TM) code, freely available from https://github.com/sflow/sflowtool.
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.
ft2nfdump - flow-tools flow converter
ft2nfdump converts flow-tools data into nfdump format.
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
parse_csv.pl - Simple reader, written in Perl.
parse_csv.pl reads nfdump csv output and print the flows to stdout. This program is intended to be a framework for post processing flows for any other purpose.
Notes for sflow users:
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 using either the fast LZO1X-1 compression, LZ4 or the efficient but slow bzip2 methode. If you compress automatically flows while they are collected, LZO1X-1 or LZ4 methodes are recommended. 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:
You can check the compression speed for your system by running ./nftest <path/to/an/existing/netflow/file>.
General Operation of nfdump
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
- typically 5 min - nfcapd rotates and renames the output file with the timestamp nfcapd.YYYYMMddhhmm of the interval e.g. nfcapd.200907110845 contains data from July 11th 2009 08:45 onward. Based on a 5min time interval, this results in 288 files per day.
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:
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 -w -D -S 2 -B 1024000 -l /flow_base_dir/router1 -p 23456 nfcapd -w -D -S 2 -B 1024000 -l /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 -w -D -S 2 -l /flow_base_dir/routers -p 23456
Collected on one port and split per source:
nfcapd -w -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 firstname.lastname@example.org
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 192.168.92.162 255.255.255.224 interface fastethernet 0/0 ip 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 9995 ip flow-export version 5 ip 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 Ethernet1/0 ip address 192.168.92.162 255.255.255.224 duplex half flow-sampler my-map ! ! flow-sampler-map my-map mode random one-out-of 5 ! ip flow-cache timeout inactive 60 ip flow-cache timeout active 1 ip flow-capture fragment-offset ip flow-capture packet-length ip flow-capture ttl ip flow-capture vlan-id ip flow-capture icmp ip flow-capture ip-id ip flow-capture mac-addresses ip flow-export version 9 ip flow-export template options export-stats ip flow-export template options sampler ip flow-export template options timeout-rate 1 ip flow-export template timeout-rate 1 ip 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: nfcapd supports a large range of netflow v9 tags. Version 1.6 nfdump supports the following fields. This list can be found in netflow_v9.h
Flowset record types
CISCO ASA NSEL extension - Network Security Event Logging__
Cisco ASR 1000 series NEL extension - Nat Event Logging__
latency extensions for nfpcapd and nprobe__
32 and 64 bit counters are supported for any counters. However, internally nfdump stores packets and bytes counters always as 64bit counters. 16 and 32 bit AS numbers are supported.
Extensions: nfcapd supports a large number of v9 tags. In order to optimise disk space and performance, v9 tags are grouped into a number of extensions which may or may not be stored into the data file. Therefore the v9 templates configured on the exporter may be tuned with the collector. Only the tags common to both are stored into the data files. Extensions can be switch on/off by using the -T option. If you want to collect all data, use -Tall
###Sampling 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 option templates (tags #48, #49, #50 ), (tag #34, #35) or in the unofficial v5 header hack. Note: Not all platforms (or IOS 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)
###InfluxDB You can send nfprofile stats data to an influxdb database. The data are the same of rrd files. For enable this option you need libcurl dev package installed, use --enable-influxdb for configure the project and the nfprofile command should be invoked with option: -i . Example: -i http://localhost:8086/write?db=mydb&u=user&p=pass The parameters for auth (&u=user&p=pass) are optional. Then you get the stats data on influxdb mydb in the measurement nfsen_stats.
For put the stats of live profile you need to apply a patch to nfsen (in extra/nfsen) and add in nfsen.conf the option: $influxdb_url="http://mydbhost.local:8086/write?db=nfsen"; as example I added a preconfigured grafana dashboard in extra/grafana/Nfsen_Stats.json .
For more information, see the GitHub Wiki