++ed by:
1 non-PAUSE user
Author image Tomas Podermanski


Net::NfDump - Perl API for manipulating with nfdump files based on libnf.net library


  use Net::NfDump;

  # Example 1: reading nfdump file(s)
  $flow = new Net::NfDump(
              InputFiles => [ 'nfdump_file1', 'nfdump_file2' ], 
              Filter => 'icmp and src net',
              Fields => 'proto, bytes' ); 


  while (my ($proto, $bytes) = $flow->fetchrow_array() )  {
      $h{$proto} += $bytes;

  foreach ( keys %h ) {
      printf "%s %d\n", $_, $h{$_};

  # Example 2: reading nfdump file(s) with aggregation and sorting
  $flow = new Net::NfDump(
              InputFiles => [ 'nfdump_file1', 'nfdump_file2' ], 
              Filter => 'icmp and src net',
              Fields => 'srcip/24/64, bytes', 
              Aggreg => 1, OrderBy => "bytes" ); 


  while (my ($ip, $bytes) = $flow->fetchrow_array() )  {
      printf "%s %d\n", $ip, $bytes;
      $h{$proto} += $bytes;

  # Example 3: creating and writing records to nfdump file
  $flow = new Net::NfDump(
              OutputFile => 'output.nfcap',
              Fields => 'srcip,dstip' );

  $flow->storerow_arrayref( [ txt2ip(''), txt2ip('') ] );


  # Example 4: reading/writing (merging two input files) and swap
  #            source and destination address if the destination port 
  #            is 80/http (I know it doesn't make much sense).

  $flow1 = new Net::NfDump( 
               InputFiles => [ 'nfdump_file1', 'nfdump_file2' ], 
               Fields => 'srcip, dstip, dstport' ); 

  $flow2 = new Net::NfDump( 
               OutputFile => 'nfdump_file_out', 
               Fields => 'srcip, dstip, dstport' ); 


  while (my $ref = $flow->fetchrow_arrayref() )  {

      if ( $ref->[2] == 80 ) { 
          ($ref->[0], $ref->[1]) = ($ref->[1], $ref->[0]);





Nfdump http://nfdump.sourceforge.net/ is a very popular toolset for collecting, storing and processing NetFlow/SFlow/IPFIX data. One of the key tools is a command line utility bearing the same name as the whole toolset (nfdump). Although this utility can process data very fast, it is cumbersome for some applications.

This module implements basic operations and allows to read, create and write flow records on binary files produced with nfdump tool. The module tries to keep the same naming conventions for methods as are used in DBI modules/API, so developers who got used to work with such interface should remain familiar with the new one.

The module uses the original nfdump sources to implement necessary functions. This enables to keep the compatibility with the original nfdump quiet easily and to cope with future versions of the nfdump tool with a minimal effort.

The architecture is following:

   |                        |  Implements all methods and functions 
   | Net::NfDump API (perl) |  described in this document.
   |                        |
   |                        |  The code converts internal nfdump 
   | libnf - glue code (C)  |  structures into perl and back to C.
   |                        |  See http://libnf.net for more information.
   |                        |  All original nfdump source files. There  
   |   nfdump sources (C)   |  are no changes in these files. All  
   |                        |  changes are placed into libnf code.

We always try to update Net::NfDump te lastest version of nfdump available on https://github.com/phaag/nfdump. Support for NSEL code is enabled.


The files created by Net::NfDump version >= 0.13 can be read only with nfdump 1.6.12 and newer. For reading it supports all formats starting with nfdump 1.6.



Options can be handled by various methods. The basic options can be handled by the constructor and then modified by methods such as $obj->query() or $obj->create().

The values after => indicate the default value for the item.

  • InputFiles => []

    List of files to read (arrayref).

  • Filter => 'any'

    Filter that is applied on input records. It uses nfdump/tcpdump syntax.

  • Fields => '*'

    List of fields to read or to update. Any supported field can be used here. See the chapter "Supported Fields" for the full list. Special field * can be used to define all fields.

  • Aggreg => 0

    Create aggregated result. When the method ->query() is called the library loads data into memory structure and perform aggregation according the Fields attribute.

  • OrderBy => '<none>'

    Sort the final result according the field specified. It can by used only for aggregated results.

  • TimeWindowStart, TimeWindowEnd => 0

    Filter flows that start or end in the specific time window. The options use unix timestamp values or 0 if the filter should not be applied.

  • OutputFile => undef

    Output file for storerow_* methods. Default: undef

  • Compressed => 1

    Flag indicating whether the output files should be compressed or not.

  • Anonymized => 0

    Flag indicating that output file contains anonymized data.

  • Ident => ''

    String identificator of files. The value is stored in the file header.

  • CompatMode => 0

    Enable nfdump compatibility features. Some features are implemented differently comparing to original nfdump. Currently thi option enables only LNF_OPT_COMP_STATSCMP for aggregated statistics computation.

Constructor, status information methods

  • $obj = new Net::NfDump( %opts )

      my $obj = new Net::NfDump( InputFiles => [ 'file1']  );

    The constructor. It defines the way the parameter options can be specified.

  • $ref = $obj->info()

      my $i = $obj->info();
      print Dumper($i);

    informs about the current state of processing input files. It returns information about already processed files, blocks and records. The information may be useful for estimating the time of processing the whole dataset. Hashref returns following items:

      total_files           - total number of files to process
      elapsed_time          - elapsed time 
      remaining_time        - estimated remaining time to process all records
      percent               - estimated percentage of processed records
      processed_files       - total number of processed files
      processed_records     - total number of processed records
      processed_blocks      - total number of processed blocks
      processed_bytes       - total number of processed bytes 
                              number of bytes read from file 
                              system after uncompressing 
      current_filename      - the name of the file currently processed
      current_total_blocks  - the number of blocks in the currently 
                              processed file 
      current_processed_blocks -  the number of processed blocks in the 
                              currently processed file
  • $obj->finish()


    closes all open file handles. It is necessary to call the method especially when a new file is created. The method flushes the file records which remained in the memory buffer and updates file statistics in the header. Without calling this method the output file might be corrupted.

Methods for reading data

  • $obj->query( %opts )

      $obj->query( Filter => 'src host' );

    This method has to be applied before any of the fetchrow_* methods is used. Any option described before can be used as a parameter of the method.

    After executing query command it possible to access $flow->{NUM_OF_FIELDS} and $flow->{NAME} variable to get returnd number of fields and field names. Here is an exmaple of code to acces field names:

        foreach $colno (0..$flow->{NUM_OF_FIELDS}-1) {
            print $flow->{NAME}->[$colno]."\t";
  • $ref = $obj->fetchrow_arrayref()

      while (my $ref = $obj->fetchrow_arrayref() ) {
          print Dumper($ref);

    This method has to be used after the query method. The method $obj->query() is called automatically if it has not been called before.

    It returns array reference with the record and skips to next record. It returns "true" if there are more records to read or "undef" if the end of a record set has been reached.

  • @array = $obj->fetchrow_array()

      while ( @array = $obj->fetchrow_arrayref() ) { 
        print Dumper(\@array);

    It has the same function as fetchrow_arrayref; however, it returns items in array instead.

  • $ref = $obj->fetchrow_hashref()

      while ( $ref = $obj->fetchrow_hashref() ) {
         print Dumper($ref);

    The same case as fetchrow_arrayref; however, the items are returned in the hash reference as the key => vallue tuples.

    NOTE: This method can be very ineffective in some cases, please, see PERFORMANCE section.

Methods for writing data

  • $obj->create( %opts )

      $obj->create( OutputFile => 'output.nfcapd' );

    This method creates a new nfdump file and has to be applied before any of $obj->storerow_* method is called.

  • $obj->storerow_arrayref( [ @array ] )

      $obj->storerow_arrayref( [ $srcip, $dstip ] );

    The method inserts data defined in arrayref to the file opened by the method $obj->create(). The number of fields and their order have to follow the order defined in the Fields option handled during $obj->new() or $obj->create() method.

  • $obj->storerow_array( @array )

      $obj->storerow_array( $srcip, $dstip );

    The same case as storerow_arrayref; however, the items are handled as a single array.

  • $obj->storerow_hashref ( \%hash )

      $obj->storerow_hashref( { 'srcip' =>  $srcip, 'dstip' => $dstip } );

    It inserts the structure defined as hash reference into output file.

    NOTE: This method can be very ineffective in some cases, please, see PERFORMANCE section.

  • $obj->clonerow( $obj2 )

      $obj->clonerow( $obj2 );

    This method copies the full content of the row from the source object (instance). This method is useful for writing effective scripts. See above the PERFORMANCE chapter.

Extra conversion and support functions

The module also provides extra convertion functions which allow to convert binnary format of IP address, MAC address and MPLS labels tag into text format and back.

Those functions are not exported by default, therefore it has to be either called with full module name or imported when the module is loaded. To import all support function :all a synonym may be used.

  use Net::NfDump qw ':all';
  • $txt = ip2txt( $bin )

  • $bin = txt2ip( $txt )

      $ip = txt2ip('');
      print ip2txt($ip);

    Converts both IPv4 and IPv6 addresses into text form and back. The standard inet_ntop/inet_pton functions can be used instead to provide the same results.

    Function txt2ip returns binnary format of IP address or "undef" if the conversion is not possible.

  • $txt = mac2txt( $bin )

  • $bin = txt2mac( $txt )

      $mac = txt2mac('aa:02:c2:2d:e0:12');
      print mac2txt($mac);

    It converts MAC address to xx:yy:xx:yy:xx:yy format and back. The function mac2txt accepts an address of any following format:


    It returns the binnary format of an address or "undef" if the conversion is not possible.

  • $txt = family2txt( $bin )

  • $bin = txt2family( $txt )

      $fam = txt2family('ipv6');
      print family2txt($fam);

    It converts internall address family (AF_INET, AF_INET6) to ipv4 or ipv6 string (or back).

    Function txt2family returns the binnary format of the family representation on the particular platform or "undef" if the conversion is not possible.

  • $txt = mpls2txt( $mpls )

  • $mpls = txt2mpls( $txt )

      $mpls = txt2mpls('1002-6-0 1003-6-0 1004-0-1');
      print mpls2txt($mpls);

    It converts label information into format Lbl-Exp-S and back.


      Lbl - Value given to the MPLS label by the router. 
      Exp - Value of the experimental bit. 
      S   - Value of the end-of-stack bit: Set to 1 for the oldest 
            entry in the stack and to zero for all other entries. 
  • $ref = flow2txt( \%row )

  • $ref = txt2flow( \%row )

    The function flow2txt gets hash reference to the items returned by fetchrow_hashref and converts all items into text format readable for human. It applies functions ip2txt, mac2txt, mpl2txt to the items for which it makes sense. The function txt2flow does the exact opossite.

  • $ref = file_info( $file_name )

      $ref = file_info('file.nfcap');
      print Dumper($ref);

    It reads information from the nfdump file header and provides various attributes such as number of blocks, version, flags, statistics, etc. As the result, the following items are returned:

      flows, bytes, packets
      flows_tcp, flows_udp, flows_icmp, flows_other
      bytes_tcp, bytes_udp, bytes_icmp, bytes_other
      packets_tcp, packets_udp, packets_icmp, packets_other


Up to date list of supported items is available on Net::NfDump::Fields

  Time items
  first - Timestamp of the first packet seen (in miliseconds)
  last - Timestamp of the last packet seen (in miliseconds)
  received - Timestamp regarding when the packet was received by collector 

  Statistical items
  bytes - The number of bytes 
  pkts - The number of packets 
  outbytes - The number of output bytes 
  outpkts - The number of output packets 
  flows - The number of flows (aggregated) 

  Layer 4 information
  srcport - Source port 
  dstport - Destination port 
  tcpflags - TCP flags  

  Layer 3 information
  srcip - Source IP address 
  dstip - Destination IP address 
  nexthop - IP next hop 
  srcmask - Source mask 
  dstmask - Destination mask 
  tos - Source type of service 
  dsttos - Destination type of service 
  srcas - Source AS number 
  dstas - Destination AS number 
  nextas - BGP Next AS 
  prevas - BGP Previous AS 
  bgpnexthop - BGP next hop 
  proto - IP protocol  

  Layer 2 information
  srcvlan - Source vlan label 
  dstvlan - Destination vlan label 
  insrcmac - In source MAC address 
  outsrcmac - Out destination MAC address 
  indstmac - In destination MAC address 
  outdstmac - Out source MAC address 

  MPLS information
  mpls - MPLS labels 

  Layer 1 information
  inif - SNMP input interface number 
  outif - SNMP output interface number 
  dir - Flow directions ingress/egress 
  fwd - Forwarding status 

  Exporter information
  router - Exporting router IP 
  systype - Type of exporter 
  sysid - Internal SysID of exporter 

  NSEL fields, see: http://www.cisco.com/en/US/docs/security/asa/asa81/netflow/netflow.html
  eventtime - NSEL The time that the flow was created
  connid - NSEL An identifier of a unique flow for the device 
  icmpcode - NSEL ICMP code value 
  icmptype - NSEL ICMP type value 
  xevent - NSEL Extended event code
  xsrcip - NSEL Mapped source IPv4 address 
  xdstip - NSEL Mapped destination IPv4 address 
  xsrcport - NSEL Mapped source port 
  xdstport - NSEL Mapped destination port 
 NSEL The input ACL that permitted or denied the flow
  iacl - Hash value or ID of the ACL name
  iace - Hash value or ID of the ACL name 
  ixace - Hash value or ID of an extended ACE configuration 
 NSEL The output ACL that permitted or denied a flow  
  eacl - Hash value or ID of the ACL name
  eace - Hash value or ID of the ACL name
  exace - Hash value or ID of an extended ACE configuration
  username - NSEL username

  NEL (NetFlow Event Logging) fields
  ingressvrfid - NEL NAT ingress vrf id 
  eventflag -  NAT event flag (always set to 1 by nfdump)
  egressvrfid -  NAT egress VRF ID

  NEL Port Block Allocation (added 2014-04-19)
  blockstart -  NAT pool block start
  blockend -  NAT pool block end 
  blockstep -  NAT pool block step
  blocksize -  NAT pool block size

  Extra/special fields
  cl - nprobe latency client_nw_delay_usec 
  sl - nprobe latency server_nw_delay_usec
  al - nprobe latency appl_latency_usec


It is obvious that performance of the perl interface is lower in comparison to highly optimized nfdump utility. While nfdump is able to process up to 2 milion of records per second, the Net::NfDump is not able to process more than 1 milion. However, there are several rules to keep the code optimised:

  • Use $obj->fetchrow_arrayref() and $obj->storerow_arrayref() instead of *_array and *_hashref equivalents. Arrayref handles only the reference to the structure with data. Avoid using *_hashref functions, it can be 5-times slower.

  • Handle to the perl API only items which are necessary to be used in the code. It is always more effective to define in Fields => 'srcip,dstip,...' instead of in Fields => '*'.

  • Preference to using $obj->clonerow($obj2) method is highly recommended. This method copies data between two instances directly in the C code in the libnf layer.

    Following code:

      $obj1->exec( Fields => '*' );
      $obj2->create( Fields => '*' );
      while ( my $ref = $obj1->fetchrow_arrayref() ) {
          # do something with srcip 

    can be written in a more effective way (several times faster):

      $obj1->exec( Fields => 'srcip' );
      $obj2->create( Fields => 'srcip' );
      while ( my $ref = $obj1->fetchrow_arrayref() ) {
          # do something with srcip 


Nfdump primary uses 64 bit counters and other items to store single integer value. However, the native 64 bit support is not compiled in every perl. For those cases where only 32 integer values are supported, the Net::NfDump uses Math::Int64 module.

The build scripts detect the platform automatically and math::Int64 module is required only on platforms where an available perl does not support 64bit integer values.


There are several examples in the examples and bin directory.

nfasnupd - Is script for updating the information about AS numbers and country codes based on BGP and geolocation database. Every flow can be extended with src/dst AS number and alco can be extended with src/dst country code.

The nfasnupd periodically checks and downloads the BGP database which is available as part of libn.net project. After that it updates the AS (or country code) information in the nfdump file. It can be run as the extra command (-x option of nfcapd) to update information when the new file is available.

The information about src/dst country works in a similar way. It uses maxmind database and Geo::IP module. However, nfdump does not support any field to store such kind of information; the xsrcport and xdstport fields are used instead. The country code is converted into 16 bit information (8 bits for the first character of a country code and another 8 bits for the second one).


nfdump project - https://github.com/phaag/nfdump libnf C interface http://libnf.net/


Tomas Podermanski, <tpoder@vut.cz>, Brno University of Technology NetX Networks a.s., <info@netx.as<gt>


Copyright (C) 2012 - 2019 by Brno University of Technology Copyright (C) 2020 by NetX Networks a.s.

This library is free software; you can redistribute it and modify it under the same terms as Perl itself.

If you are satisfied with using Net::NfDump, please, send us a postcard, preferably with a picture of your location / city to:

  Brno University of Technology 
  Tomas Podermanski 
  Antoninska 1
  601 90 
  Czech Republic 

2 POD Errors

The following errors were encountered while parsing the POD:

Around line 1398:

=back without =over

Around line 1423:

=back without =over