mcast-tools 0.2.1

Friendly multicast send/receive tools for lab environments

mcast-tools

Friendly multicast send/receive tools for lab environments — IPv4 and IPv6. Hides the complexity of iperf / raw socket plumbing behind three simple commands, and adds capabilities iperf can't provide — source-specific (SSM) joins, IGMP/MLD version control, hop-count and DSCP-remark visibility via embedded metadata, and live dashboards that update in real time.

The address family is auto-detected from the group address — no flags: mcast-send 239.0.10.101 is IPv4, mcast-send ff3e::8000:1 is IPv6.

Why not just iperf?

iperf is great, but for a teaching lab a few limitations matter:

• No SSM joins. iperf -s -B <group> only does any-source (*,G) joins. No flag fixes this — it's a socket-level capability iperf doesn't expose.
• No IGMP version control. That's a kernel sysctl, not an iperf knob.
• No visibility into TTL or DSCP. iperf doesn't tell you which interface the join went out on, what IGMP version was actually sent, what TTL or DSCP arrived, or how many hops away the sender is.
• Output is built for log scraping, not live UX.

mcast-tools keeps iperf's "two commands, easy to remember" feel while adding the things you actually want in a lab.

Install

pip install mcast-tools

Or from source:

git clone https://github.com/mitchv85/mcast-tools.git
cd mcast-tools
pip install .

Provides three commands: mcast-send, mcast-receive, mcast-status.

Root is recommended for mcast-receive so it can force the IGMP/MLD version via /proc/sys/net/ipv4/conf/<iface>/force_igmp_version or /proc/sys/net/ipv6/conf/<iface>/force_mld_version. It will still run without root, but the kernel will use whatever version it's currently configured for.

Quick start

# Sender — defaults to 10 minutes, TTL 100, 25pps, DSCP=BE
mcast-send 239.0.10.101

# Receiver — defaults to ASM (*,G) join, IGMPv3
mcast-receive 239.0.10.101

# What groups is this host joined to?
mcast-status

# Same pair over IPv6 — SSM with MLDv2
mcast-receive ff3e::8000:1 --source 2001:db8::5 --interface eth1
mcast-send ff3e::8000:1 --rate 50pps --interface eth1

mcast-send

mcast-send <group> [duration] [ttl]
mcast-send <group> --rate 1Mbps --dscp ef --interface eth1

Option	Default	Purpose
group	(required)	IPv4 or IPv6 multicast group, e.g. 239.0.10.101 or ff3e::8000:1
duration	600	Seconds to send. 0 = forever
ttl	100	IPv4 TTL / IPv6 hop limit (--hop-limit is an alias of --ttl)
--rate	25pps	Npps, Nkbps, NMbps, etc.
--size	200	Datagram size in bytes (header + payload)
--dscp	be	0-63 or name (ef, af41, cs5, ...)
--source	(kernel-selected)	Bind the outgoing source address. Pin it for SSM demos — see below.
--interface	(routing default)	Egress interface override
--no-loopback	off	Disable IP_MULTICAST_LOOP
--quiet / --json	off	Suppress dashboard / emit JSON summary on exit

Each packet carries a 32-byte custom header with magic MCAS, a sequence number, a monotonic nanosecond timestamp, a random per-invocation sender ID, and the original TTL/DSCP the sender configured. mcast-receive uses these fields to compute loss, duplicates, reordering, hop count, and DSCP remarks.

mcast-receive

mcast-receive <group>                                  # ASM (*,G)
mcast-receive <group> <source>                         # SSM (S,G), IGMPv3
mcast-receive <group> --igmp-version 2 --interface eth1
mcast-receive ff3e::8000:1 --source 2001:db8::5 -i eth1   # IPv6 SSM, MLDv2

Option	Default	Purpose
group	(required)	IPv4 or IPv6 multicast group
source	(none)	Source IP for SSM (S,G) join — same family as the group
igmp_version	3 / 2	Positional version: IGMP 1/2/3 for IPv4 groups, MLD 1/2 for IPv6 groups. SSM needs IGMPv3 / MLDv2.
--mld-version	2	MLD version for IPv6 groups (--igmp-version is IPv4-only)
--interface	(routing default)	Interface to join on
--stall-timeout	3.0	Seconds without a packet before flagging STALLED
--duration	0	Auto-stop after N seconds (0 = run forever)
--quiet / --json	off	Suppress dashboard / emit JSON summary on exit

The receiver continuously displays:

• Status: WAITING → RECEIVING → STALLED based on packet arrival
• Received TTL with hop-count interpretation (recv 96, sender set 100 — 4 hops away)
• Received DSCP with remark detection (BE, sender set EF — REMARKED)
• Loss / Duplicates / Reordered counters
• Sender restart detection (new sender_id → restart counter increments)
• Gap log — recent stall events with duration

Why IGMPv3 / MLDv2 is mandatory for SSM

Only IGMPv3 Membership Reports carry the per-source state needed to communicate (S,G) filtering up to the querier. v1/v2 reports only express (*,G). The IPv6 story is identical with the names shifted: MLDv1 maps to IGMPv2 (ASM only) and MLDv2 maps to IGMPv3 (adds the source lists). There is no MLD analog of IGMPv1. mcast-receive enforces this with a clear error message rather than silently doing the wrong thing.

mcast-status

mcast-status              # current memberships, excluding kernel defaults
mcast-status --all        # include 224.0.0.1 etc.
mcast-status -i eth0      # one interface only

Reads /proc/net/igmp and /proc/net/igmp6 and pretty-prints which interfaces are joined to which groups. The IPv4 table includes the operational IGMP version and the force_igmp_version setting. The IPv6 table shows only the forced MLD state — the kernel doesn't expose the operational MLD version in /proc/net/igmp6, so we don't invent one. By default the v6 view hides all-nodes (ff02::1, ff01::1) and solicited-node (ff02::1:ffxx:xxxx) groups — NDP plumbing, not joins — which means a freshly-booted host shows an empty v6 table until something actually joins. --all reveals everything.

End-to-end example

In one terminal:

mcast-receive 239.0.10.101

In another:

mcast-send 239.0.10.101 --rate 50pps --ttl 64 --dscp ef

The receiver will show 50 pps, EF DSCP preserved, and a hop count derived from the difference between the sender-stamped original TTL (64) and the TTL observed on arrival. If anything in the path remarks DSCP or decrements TTL more than expected, you'll see it immediately.

IPv6 specifics

• Family auto-detection. The group address decides everything; there is no --ipv6 flag. Sources must match the group's family.
• Pin the sender's source for SSM. Without --source, the kernel picks the source via RFC 6724 source-address selection — and when an interface carries multiple same-prefix globals (say, a static address plus a SLAAC EUI-64 from a router's RA), every selection rule ties and the pick is unspecified. It can differ across runs. An SSM receiver filtering on "the other one" then silently receives nothing while both ends work perfectly. mcast-send G --source S binds the socket, which removes selection from the picture; pair it with the matching mcast-receive G --source S so the same S flows through the whole (S,G) demo. (The bind sets the source, not the egress — add --interface for fully deterministic egress. The same flag works on IPv4, where it matters once a host carries secondaries or VIPs.)
• Scopes and zones. Interface-local (ff01::) and link-local (ff02::) scope groups never appear usefully in the routing table, so the kernel must be told which interface — give a zone (ff02::42%eth1) or --interface. Conflicting zone + --interface is an explicit error. Routable scopes (site ff05::, global ff0e::, the SSM range ff3x::) resolve via the route table like IPv4 does.
• Hop limit & traffic class. --ttl/--hop-limit sets IPV6_MULTICAST_HOPS; --dscp sets the traffic class, whose byte layout matches the IPv4 TOS byte — so remark detection works identically across families. The wire header is unchanged from v0.1: orig_ttl simply carries the hop limit on IPv6.
• Joins. ASM uses IPV6_JOIN_GROUP (struct ipv6_mreq). SSM uses the protocol-independent MCAST_JOIN_SOURCE_GROUP with a hand-packed struct group_source_req — there is no IPV6_ADD_SOURCE_MEMBERSHIP.
• MLD forcing mirrors IGMP forcing via /proc/sys/net/ipv6/conf/<iface>/force_mld_version (saved and restored on exit). 0 means auto, which on modern kernels is MLDv2.
• Group hygiene on the receive socket uses IPV6_MULTICAST_ALL=0 (kernel 4.20+; silently best-effort on older kernels).

Behind-the-scenes details (worth knowing for a lab)

• Destination UDP port is 19779 (0x4D43 = "MC") for both families. Filter in tcpdump: tcpdump 'udp port 19779 and host 239.0.10.101' or tcpdump 'udp port 19779 and ip6 host ff3e::8000:1'.
• Default datagram size is 200 bytes (32-byte header + 168-byte payload). Configure with --size. Minimum is 32 bytes (header only).
• Packet padding is 0xA5 repeated — easy to eyeball in a capture.
• Sender ID randomization means restarting a sender looks like a brand-new sender on the receiver, which is exactly the right semantic — we don't want the receiver to interpret a restart as a giant burst of packet loss.
• Per-sender sequence tracking means a receiver listening to multiple senders (or one sender across restarts) doesn't conflate their sequence spaces.
• Stall detection is interval-based, not packet-rate-based — we don't need a control channel to know the sender's target rate.
• Foreign packet counter counts UDP datagrams on the group that don't start with the MCAS magic header. Useful for detecting when something else (real video, iperf, etc.) is sharing the group.

License

MIT. See LICENSE.