Cflow::find - find "interesting" flows in raw IP flow files

   use Cflow;

   Cflow::verbose(1);
   Cflow::find(\&wanted, <*.flows*>);

   sub wanted { ... }

or:

   Cflow::find(\&wanted, \&perfile, <*.flows*>);

   sub perfile { 
      my $fname = shift;
      ...
   }

This module implements an API for processing IP flow accounting information which as been collected from routers and written into flow files by one of the various flow collectors listed below.

It was originally conceived and written for use by FlowScan:

   http://net.doit.wisc.edu/~plonka/FlowScan/

This package is of little use on its own. It requires input in the form of time-stamped raw flow files produced by other software packages. These flow sources either snoop a local ethernet (via libpcap) or collect flow information from IP routers that are configured to export said information. The following flow sources are supported:

argus by Carter Bullard:

   http://www.qosient.com/argus/

flow-tools by Mark Fullmer (with NetFlow v1, v5, v6, or v7):

   http://www.splintered.net/sw/flow-tools/

CAIDA's cflowd (with NetFlow v5):

   http://www.caida.org/tools/measurement/cflowd/
   http://net.doit.wisc.edu/~plonka/cflowd/

lfapd by Steve Premeau (with LFAPv4):

   http://www.nmops.org/

Cflow::find() will iterate across all the flows in the specified files. It will call your wanted() function once per flow record. If the file name argument passed to find() is specified as -, flows will be read from standard input.

The wanted() function does whatever you write it to do. For instance, it could simply print interesting flows or it might maintain byte, packet, and flow counters which could be written to a database after the find subroutine completes.

Within your wanted() function, tests on the current flow can be performed using the following variables:

$Cflow::unix_secs

secs since epoch (deprecated)

$Cflow::exporter

Exporter IP Address as a host-ordered long

$Cflow::exporterip

Exporter IP Address as dotted-decimal string

$Cflow::localtime

$Cflow::unix_secs interpreted as localtime with this strftime(3) format:

   %Y/%m/%d %H:%M:%S

$Cflow::srcaddr

Source IP Address as a host-ordered long

$Cflow::srcip

Source IP Address as a dotted-decimal string

$Cflow::dstaddr

Destination IP Address as a host-ordered long

$Cflow::dstip

Destination IP Address as a dotted-decimal string

$Cflow::input_if

Input interface index

$Cflow::output_if

Output interface index

$Cflow::srcport

TCP/UDP src port number or equivalent

$Cflow::dstport

TCP/UDP dst port number or equivalent

$Cflow::ICMPType

high byte of CW$Cflow::dstport Undefined if the current flow is not an ICMP flow.

$Cflow::ICMPCode

low byte of CW$Cflow::dstport Undefined if the current flow is not an ICMP flow.

$Cflow::ICMPTypeCode

symbolic representation of CW$Cflow::dstport The value is a the type-specific ICMP code, if any, followed by the ICMP type. E.g.

   ECHO
   HOST_UNREACH

Undefined if the current flow is not an ICMP flow.

$Cflow::pkts

Packets sent in Duration

$Cflow::bytes

Octets sent in Duration

$Cflow::nexthop

Next hop router's IP Address as a host-ordered long

$Cflow::nexthopip

Next hop router's IP Address as a dotted-decimal string

$Cflow::startime

secs since epoch at start of flow

$Cflow::start_msecs

fractional portion of startime (in milliseconds) This will be zero unless the source is flow-tools or argus.

$Cflow::endtime

secs since epoch at last packet of flow

$Cflow::end_msecs

fractional portion of endtime (in milliseconds) This will be zero unless the source is flow-tools or argus.

$Cflow::protocol

IP protocol number (as is specified in /etc/protocols, i.e. 1=ICMP, 6=TCP, 17=UDP, etc.)

$Cflow::tos

IP Type-of-Service

$Cflow::tcp_flags

bitwise OR of all TCP flags that were set within packets in the flow; 0x10 for non-TCP flows

$Cflow::TCPFlags

symbolic representation of CW$Cflow::tcp_flags The value will be a bitwise-or expression. E.g.

   PUSH|SYN|FIN|ACK

Undefined if the current flow is not a TCP flow.

$Cflow::raw

the entire packed flow record as read from the input file This is useful when the wanted subroutine wants to write the flow to another FILEHANDLE. E.g.:

   syswrite(FILEHANDLE, $Cflow::raw, length $Cflow::raw)

$Cflow::reraw

the entire re-packed flow record formatted like CW$Cflow::raw. This is useful when the wanted subroutine wants to write a modified flow to another FILEHANDLE. E.g.:

   $srcaddr = my_encode($srcaddr);
   $dstaddr = my_encode($dstaddr);
   syswrite(FILEHANDLE, $Cflow::reraw, length $Cflow::raw)

These flow variables are packed into CW$Cflow::reraw:

   $Cflow::index, $Cflow::exporter,
   $Cflow::srcaddr, $Cflow::dstaddr,
   $Cflow::input_if, $Cflow::output_if,
   $Cflow::srcport, $Cflow::dstport,
   $Cflow::pkts, $Cflow::bytes,
   $Cflow::nexthop,
   $Cflow::startime, $Cflow::endtime,
   $Cflow::protocol, $Cflow::tos,
   $Cflow::src_as, $Cflow::dst_as,
   $Cflow::src_mask, $Cflow::dst_mask,
   $Cflow::tcp_flags,
   $Cflow::engine_type, $Cflow::engine_id

$Cflow::Bps

the minimum bytes per second for the current flow

$Cflow::pps

the minimum packets per second for the current flow

The following variables are undefined if using NetFlow v1 (which does not contain the requisite information):

$Cflow::src_as

originating or peer AS of source address

$Cflow::dst_as

originating or peer AS of destination address

The following variables are undefined if using NetFlow v1 or LFAPv4 (which do not contain the requisite information):

$Cflow::src_mask

source address prefix mask bits

$Cflow::dst_mask

destination address prefix mask bits

$Cflow::engine_type

type of flow switching engine

$Cflow::engine_id

ID of the flow switching engine

Optionally, a reference to a perfile() function can be passed to Cflow::find as the argument following the reference to the wanted() function. This perfile() function will be called once for each flow file. The argument to the perfile() function will be name of the flow file which is about to be processed. The purpose of the perfile() function is to allow you to periodically report the progress of Cflow::find() and to provide an opportunity to periodically reclaim storage used by data objects that may have been allocated or maintained by the wanted() function. For instance, when counting the number of active hosts IP addresses in each time-stamped flow file, perfile() can reset the counter to zero and clear the search tree or hash used to remember those IP addresses.

Since Cflow is an Exporter, you can request that all those scalar flow variables be exported (so that you need not use the Cflow:: prefix):

   use Cflow qw(:flowvars);

Also, you can request that the symbolic names for the TCP flags, ICMP types, and/or ICMP codes be exported:

   use Cflow qw(:tcpflags :icmptypes :icmpcodes);

The tcpflags are:

   $TH_FIN $TH_SYN $TH_RST $TH_PUSH $TH_ACK $TH_URG

The icmptypes are:

   $ICMP_ECHOREPLY     $ICMP_DEST_UNREACH $ICMP_SOURCE_QUENCH
   $ICMP_REDIRECT      $ICMP_ECHO         $ICMP_TIME_EXCEEDED
   $ICMP_PARAMETERPROB $ICMP_TIMESTAMP    $ICMP_TIMESTAMPREPLY
   $ICMP_INFO_REQUEST  $ICMP_INFO_REPLY   $ICMP_ADDRESS
   $ICMP_ADDRESSREPLY

The icmpcodes are:

   $ICMP_NET_UNREACH  $ICMP_HOST_UNREACH $ICMP_PROT_UNREACH
   $ICMP_PORT_UNREACH $ICMP_FRAG_NEEDED  $ICMP_SR_FAILED
   $ICMP_NET_UNKNOWN  $ICMP_HOST_UNKNOWN $ICMP_HOST_ISOLATED
   $ICMP_NET_ANO      $ICMP_HOST_ANO     $ICMP_NET_UNR_TOS
   $ICMP_HOST_UNR_TOS $ICMP_PKT_FILTERED $ICMP_PREC_VIOLATION
   $ICMP_PREC_CUTOFF  $ICMP_UNREACH      $ICMP_REDIR_NET
   $ICMP_REDIR_HOST   $ICMP_REDIR_NETTOS $ICMP_REDIR_HOSTTOS
   $ICMP_EXC_TTL      $ICMP_EXC_FRAGTIME

Please note that the names above are not necessarily exactly the same as the names of the flags, types, and codes as set in the values of the aforemented CW$Cflow::TCPFlags and CW$Cflow::ICMPTypeCode flow variables.

Lastly, as is usually the case for modules, the subroutine names can be imported, and a minimum version of Cflow can be specified:

   use Cflow qw(:flowvars find verbose 1.031);

Cflow::find() returns a hit-ratio. This hit-ratio is a string formatted similarly to that of the value of a perl hash when taken in a scalar context. This hit-ratio indicates ((# of wanted flows) / (# of scanned flows)). A flow is considered to have been wanted if your wanted() function returns non-zero.

Cflow::verbose() takes a single scalar boolean argument which indicates whether or not you wish warning messages to be generated to STDERR when problems occur. Verbose mode is set by default.

Here's a complete example with a sample wanted function. It will print all UDP flows that involve either a source or destination port of 31337 and port on the other end that is unreserved (greater than 1024):

   use Cflow qw(:flowvars find);

   my $udp = getprotobyname('udp');
   verbose(0);
   find(\&wanted, @ARGV? @ARGV : <*.flows*>);

   sub wanted {
      return if ($srcport < 1024 || $dstport < 1024);
      return unless (($srcport == 31337 || $dstport == 31337) &&
                      $udp == $protocol);

      printf("%s %15.15s.%-5hu %15.15s.%-5hu %2hu %10u %10u\n",
             $localtime,
             $srcip,
             $srcport,
             $dstip,
             $dstport,
             $protocol,
             $pkts,
             $bytes)
   }

Here's an example which demonstrates a technique which can be used to pass arbitrary arguments to your wanted function by passing a reference to an anonymous subroutine as the wanted() function argument to Cflow::find():

   sub wanted {
      my @params = @_;
      # ...
   }

   Cflow::find(sub { wanted(@params) }, @files);

Argus uses a bidirectional flow model. This means that some argus flows represent packets not only in the forward direction (from source to destination), but also in the reverse direction (from the so-called destination to the source). However, this module uses a unidirection flow model, and therfore splits some argus flows into two unidirectional flows for the purpose of reporting.

Currently, using this module's API there is no way to determine if two subsequently reported unidirectional flows were really a single argus flow. This may be addressed in a future release of this package.

Furthermore, for argus flows which represent bidirectional ICMP traffic, this module presumes that all the reverse packets were ECHOREPLYs (sic). This is sometimes incorrect as described here:

   http://www.theorygroup.com/Archive/Argus/2002/msg00016.html

and will be fixed in a future release of this package.

Timestamps ($startime and CW$endtime) are sometimes reported incorrectly for bidirectional argus flows that represent only one packet in each direction. This will be fixed in a future release.

Argus flows sometimes contain information which does not map directly to the flow variables presented by this module. For the time being, this information is simply not accessible through this module's API. This may be addressed in a future release.

Lastly, argus flows produced from observed traffic on a local ethernet do not contain enough information to meaningfully set the values of all this module's flow variables. For instance, the next-hop and input/output ifIndex numbers are missing. For the time being, all argus flows accessed throught this module's API will have both the CW$input_if and CW$output_if as 42. Althought 42 is the answer to life, the universe, and everthing, in this context, it is just an arbitrary number. It is important that CW$output_if is non-zero, however, since existing FlowScan reports interpret an CW$output_if value of zero to mean that the traffic represented by that flow was not forwarded (i.e. dropped). For similar reasons, the CW$nexthopip for all argus flows is reported as 127.0.0.1.

Currently, only NetFlow version 5 is supported when reading cflowd-format raw flow files.

When built with support for flow-tools and attempting to read a cflowd format raw flow file from standard input, you'll get the error:

   open "-": No such file or directory

For the time being, the workaround is to write the content to a file and read it from directly from there rather than from standard input. (This happens because we can't close and re-open file descriptor zero after determining that the content was not in flow-tools format.)

When built with support for flow-tools and using verbose mode, Cflow::find will generate warnings if you process a cflowd format raw flow file. This happens because it will first attempt to open the file as a flow-tools format raw flow file (which will produce a warning message), and then revert to handling it as cflowd format raw flow file.

Likewise, when built with support for argus and attempting to read a cflowd format raw flow file from standard input, you'll get this warning message:

   not Argus-2.0 data stream.

This is because argus (as of argus-2.0.4) doesn't seem to have a mode in which such warning messages are supressed.

The CW$Cflow::raw flow variable contains the flow record in cflowd format, even if it was read from a raw flow file produced by flow-tools or argus. Because cflowd discards the fractional portion of the flow start and end time, only the whole seconds portion of these times will be retained. (That is, the raw record in CW$Cflow::raw does not contain the CW$start_msecs and CW$end_msecs, so using CW$Cflow::raw to convert to cflowd format is a lossy operation.)

When used with cflowd, Cflow::find() will generate warnings if the flow data file is invalid as far as its concerned. To avoid this, you must be using Cisco version 5 flow-export and configure cflowd so that it saves all flow-export data. This is the default behavior when cflowd produces time-stamped raw flow files after being patched as described here:

   http://net.doit.wisc.edu/~plonka/cflowd/

The interface presented by this package is a blatant ripoff of File::Find.

Dave Plonka <plonka@doit.wisc.edu>

Copyright (C) 1998-2002 Dave Plonka. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

The version number is the module file RCS revision number ($Revision: 1.51 $) with the minor number printed right justified with leading zeroes to 3 decimal places. For instance, RCS revision 1.1 would yield a package version number of 1.001.

This is so that revision 1.10 (which is version 1.010), for example, will test greater than revision 1.2 (which is version 1.002) when you want to require a minimum version of this module.

perl(1), Socket, Net::Netmask, Net::Patricia.