Community ID Flow Hashing

When processing flow data from a variety of monitoring applications (such as Zeek and Suricata), it's often desirable to pivot quickly from one dataset to another. While the required flow tuple information is usually present in the datasets, the details of such "joins" can be tedious, particular in corner cases. This spec describes "Community ID" flow hashing, standardizing the production of a string identifier representing a given network flow, to reduce the pivot to a simple string comparison.

Pseudo code

function community_id_v1(ipaddr saddr, ipaddr daddr, port sport, port dport, int proto, int seed=0)
{
    # Get seed and all tuple parts into network byte order
    seed = pack_to_nbo(seed); # 2 bytes
    saddr = pack_to_nbo(saddr); # 4 or 16 bytes
    daddr = pack_to_nbo(daddr); # 4 or 16 bytes
    sport = pack_to_nbo(sport); # 2 bytes
    dport = pack_to_nbo(dport); # 2 bytes

    # Flip the endpoints as needed to abstract away directionality
    saddr, daddr, sport, dport = order_endpoints(saddr, daddr, sport, dport);

    # Produce 20-byte SHA1 digest. "." means concatenation. The
    # proto value is one byte in length and followed by a 0 byte
    # for padding.
    sha1_digest = sha1(seed . saddr . daddr . proto . 0 . sport . dport)

    # Prepend version string to base64 rendering of the digest.
    # v1 is currently the only one available.
    return "1:" + base64(sha1_digest)
}

function community_id_icmp(int seed, ipaddr saddr, ipaddr daddr, int type, int code, int seed=0)
{
    port sport, dport;

    # ICMP / ICMPv6 endpoint mapping directly inspired by Zeek
    sport, dport = map_icmp_to_ports(type, code);

    # ICMP is IP protocol 1, ICMPv6 would be 58
    return community_id_v1(saddr, daddr, sport, dport, 1, seed); 
}

Technical details

The Community ID is an additional flow identifier and doesn't need to replace existing flow identification mechanisms already supported by the monitors. It's okay, however, for a monitor to be configured to log only the Community ID, if desirable.
The Community ID can be computed as a monitor produces flows, or can also be added to existing flow records at a later stage assuming that said records convey all the needed flow endpoint information.
Collisions in the Community ID, while undesirable, are not considered fatal, since the user should still possess flow timing information and possibly the monitor's native ID mechanism (hopefully stronger than the Community ID) for disambiguation.
The hashing mechanism uses seeding to enable additional control over "domains" of Community ID usage. The seed defaults to 0, so this mechanism gets out of the way so it doesn't affect operation for operators not interested in it.
In version 1 of the ID, the hash algorithm is SHA1. Future hash versions may switch it or allow additional configuration.
The binary 20-byte SHA1 result gets base64-encoded to reduce output volume compared to the usual ASCII-based SHA1 representation. This assumes that space, not computation time, is the primary concern, and may become configurable in a later version.
The resulting flow ID includes a version number to make the underlying Community ID implementation explicit. This allows users to ensure they're comparing apples to apples while supporting future changes to the algorithm. For example, when one monitor's version of the ID incorporates VLAN IDs but another's does not, hash value comparisons should reliably fail. A more complex form of this feature could allow capturing configuration settings in addition to the implementation version.

The versioning scheme currently simply prefixes the hash value with ":", yielding something like this in the current version 1:

1:hO+sN4H+MG5MY/8hIrXPqc4ZQz0=
The hash input is aligned on 32-bit-boundaries. Flow tuple components use network byte order (big-endian) to standardize ordering regardless of host hardware.
This version includes the following protocols and fields:
- TCP / UDP / SCTP:
  
  IP src / IP dst / IP proto / source port / dest port
- ICMPv4 / ICMPv6:
  
  IP src / IP dst / IP proto / ICMP type + "counter-type" or code
  
  The exact handling of ICMP type & code is taken from Zeek; see implementations here:
- Other IP-borne protocols:
  
  IP src / IP dst / IP proto
The above does not currently cover how to handle nesting (IP in IP, v6 over v4, etc) as well as encapsulations such as VLAN and MPLS.
If a network monitor doesn't support any of the above protocol constellations, it can safely report an empty string (or another non-colliding value) for the flow ID.
Consider v1 a prototype. Feedback from the community, particularly implementers and operational users of the ID, is greatly appreciated. Please create issues directly in the GitHub project at https://github.com/corelight/community-id-spec, or contact Christian Kreibich (christian@corelight.com).
Many thanks for helpful discussion and feedback to Victor Julien, Johanna Amann, and Robin Sommer, and to all implementors and supporters.

Reference implementation

A complete implementation is available in the pycommunityid package. It includes a range of tests to verify correct computation for the various protocols. We recommend it to guide new implementations.

A smaller implementation is also available via the community-id.py script in this repository, including the byte layout of the hashed values (see packet_get_comm_id()). See --help and make.sh to get started:

  $ ./community-id.py --help
  usage: community-id.py [-h] [--seed NUM] PCAP [PCAP ...]

  Community flow ID reference

  positional arguments:
    PCAP         PCAP packet capture files

  optional arguments:
    -h, --help   show this help message and exit
    --seed NUM   Seed value for hash operations
    --no-base64  Don't base64-encode the SHA1 binary value
    --verbose    Show verbose output on stderr

For troubleshooting, the implementation supports omitting the base64 operation, and can provide additional detail about the exact sequence of bytes going into the SHA1 hash computation.