Collector in v0.8.0 terminated by OOM killer

Python NetFlow/IPFIX library

This package contains libraries and tools for NetFlow versions 1, 5 and 9, and IPFIX. It is available on PyPI as "netflow".

Version 9 is the first NetFlow version using templates. Templates make dynamically sized and configured NetFlow data flowsets possible, which makes the collector's job harder. The library provides the netflow.parse_packet() function as the main API point (see below). By importing netflow.v1, netflow.v5 or netflow.v9 you have direct access to the respective parsing objects, but at the beginning you probably will have more success by running the reference collector (example below) and look into its code. IPFIX (IP Flow Information Export) is based on NetFlow v9 and standardized by the IETF. All related classes are contained in netflow.ipfix.

Licensed under MIT License. See LICENSE.

Using the library

If you chose to use the classes provided by this library directly, here's an example for a NetFlow v5 export packet:

Create a collector which listens for exported packets on some UDP port. It should then receive UDP packets from exporters.
Inside the UDP packets, the NetFlow payload is contained. For NetFlow v5 it should begin with bytes 0005 for example.
Call the netflow.parse_packet() function with the payload as first argument (takes string, bytes string and hex'd bytes).

Example UDP collector server (receiving exports on port 2055):

import netflow
import socket
sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
sock.bind(("0.0.0.0", 2055))
payload, client = sock.recvfrom(4096)  # experimental, tested with 1464 bytes
p = netflow.parse_packet(payload)  # Test result: <ExportPacket v5 with 30 records>
print(p.header.version)  # Test result: 5

Or from hex dump:

import netflow
p = netflow.parse_packet("00050003000379a35e80c58622a...")  # see test_netflow.py
assert p.header.version == 5  # NetFlow v5 packet
assert p.flows[0].PROTO == 1  # ICMP flow

In NetFlow v9 and IPFIX, templates are used instead of a fixed set of fields (like PROTO). See collector.py on how to handle these. You must store received templates in between exports and pass them to the parser when new packets arrive. Not storing the templates will always result in parsing failures.

Using the collector and analyzer

Since v0.9.0 the netflow library also includes reference implementations of a collector and an analyzer as CLI tools. These can be used on the CLI with python3 -m netflow.collector and python3 -m netflow.analyzer. Use the -h flag to receive the respective help output with all provided CLI flags.

Example: to start the collector run python3 -m netflow.collector -p 9000 -D. This will start a collector instance at port 9000 in debug mode. Point your flow exporter to this port on your host and after some time the first ExportPackets should appear (the flows need to expire first). After you collected some data, the collector exports them into GZIP files, simply named <timestamp>.gz (or the filename you specified with --file/-o).

To analyze the saved traffic, run python3 -m netflow.analyzer -f <gzip file>. The output will look similar to the following snippet, with resolved hostnames and services, transferred bytes and connection duration:

2017-10-28 23:17.01: SSH     | 4.25M    | 15:27 min | local-2 (<IPv4>) to local-1 (<IPv4>)
2017-10-28 23:17.01: SSH     | 4.29M    | 16:22 min | remote-1 (<IPv4>) to local-2 (<IPv4>)
2017-10-28 23:19.01: HTTP    | 22.79M   | 47:32 min | uwstream3.somafm.com (173...) to local-1 (<IPv4>)
2017-10-28 23:22.01: HTTPS   | 1.21M    | 3 sec     | fra16s12-in-x0e.1e100.net (2a00:..) to local-1 (<IPv6>)
2017-10-28 23:23.01: SSH     | 93.79M   | 21 sec    | remote-1 (<IPv4>) to local-2 (<IPv4>)
2017-10-28 23:51.01: SSH     | 14.08M   | 1:23.09 hours | remote-1 (<IPv4>) to local-2 (<IPv4>)

Please note that the collector and analyzer are experimental reference implementations. Do not rely on them in production monitoring use cases! In any case I recommend looking into the netflow/collector.py and netflow/analyzer.py scripts for customization. Feel free to use the code and extend it in your own tool set - that's what the MIT license is for!

Resources

Development environment

The library was specifically written in combination with NetFlow exports from Hitoshi Irino's fork of softflowd (v1.0.0) - it should work with every correct NetFlow/IPFIX implementation though. If you stumble upon new custom template fields please let me know, they will make a fine addition to the netflow.v9.V9_FIELD_TYPES collection.

Running and creating tests

The test files contain tests for all use cases in the library, based on real softflowd export packets. Whenever softflowd is referenced, a compiled version of softflowd 1.0.0 is meant, which is probably NOT the one in your distribution's package. During the development of this library, two ways of gathering these hex dumps were used. First, the tcpdump/Wireshark export way:

Run tcpdump/Wireshark on your public-facing interface (with tcpdump, save the pcap to disk).
Produce some sample flows, e.g. surf the web and refresh your mail client. With Wireshark, save the captured packets to disk.
Run tcpdump/Wireshark again on a local interface.
Run softflowd with the -r <pcap_file> flag. softflowd reads the captured traffic, produces the flows and exports them. Use the interface you are capturing packets on to send the exports to. E.g. capture on the localhost interface (with -i lo or on loopback) and then let softflowd export to 127.0.0.1:1337.
Examine the captured traffic. Use Wireshark and set the CFLOW "decode as" dissector on the export packets (e.g. based on the port). The data fields should then be shown correctly as Netflow payload.
Extract this payload as hex stream. Anonymize the IP addresses with a hex editor if necessary. A recommended hex editor is bless.

Second, a Docker way:

Run a softflowd daemon in the background inside a Docker container, listening on eth0 and exporting to e.g. 172.17.0.1:1337.
On your host start Wireshark to listen on the Docker bridge.
Create some traffic from inside the container.
Check the softflow daemon with softflowctl dump-flows.
If you have some flows shown to you, export them with softflowctl expire-all.
Your Wireshark should have picked up the epxort packets (it does not matter if there's a port unreachable error).
Set the decoder for the packets to CFLOW and copy the hex value from the NetFlow packet.

Your exported hex string should begin with 0001, 0005, 0009 or 000a, depending on the version.

The collector is run in a background thread. The difference in transmission speed from the exporting client can lead to different results, possibly caused by race conditions during the usage of the GZIP output file.

	# Assume the size that sent the most data is the source
	# TODO: this might not always be right, maybe use earlier timestamp?
	size1 = fallback(flow1, ['IN_BYTES', 'IN_OCTETS'])
	size2 = fallback(flow2, ['IN_BYTES', 'IN_OCTETS'])
	if size1 >= size2:
	src = flow1
	dest = flow2
	else:
	src = flow2
	dest = flow1

	# TODO: this next approach uses the lower port as the service identifier
	# port1 = fallback(flow1, ['L4_SRC_PORT', 'SRC_PORT'])
	# port2 = fallback(flow2, ['L4_SRC_PORT', 'SRC_PORT'])
	#
	# src = flow1
	# dest = flow2
	# if port1 > port2:
	# src = flow2
	# dest = flow1

bitkeks / python-netflow-v9-softflowd Goto Github PK

python-netflow-v9-softflowd's Introduction

Python NetFlow/IPFIX library

Using the library

Using the collector and analyzer

Resources

Development environment

Running and creating tests

python-netflow-v9-softflowd's People

Contributors

Stargazers

Watchers

Forkers

python-netflow-v9-softflowd's Issues

The packet (UDP data):

A Demonstration of the Infinite Loop

Proposed Solution

Recommend Projects

Recommend Topics

Recommend Org