(17e) dataplane: advertise the gateway with LLDP on every port

[daniel-noland]

Native LLDP for the gateway. The dataplane owns its NICs, so nothing else can speak LLDP for them, and lldpd/lldpad are hostile to automation. LLDP itself is small, so the dataplane transmits and receives it.

This used to be one PR. It is now a stack of seven, each building on the one before:

• (17a) (17a) net: LLDP TLV framing and the mandatory TLVs #1906 net: TLV framing and the mandatory TLVs
• (17b) (17b) net: model the optional LLDP TLVs #1920 net: the optional TLVs
• (17c) (17c) net: assemble LLDP data units and decode them in Headers::parse #1921 net: the LLDPDU and Headers::parse
• (17d) (17d) lldp: the transmitter and the neighbour table #1922 lldp: the transmitter and the neighbour table
• (17e) (17e) dataplane: advertise the gateway with LLDP on every port #1923 dataplane: advertise on every port ← this PR
• (17f) (17f) dataplane: hear LLDP neighbours and show them #1924 dataplane: hear neighbours, show lldp neighbors, all-multicast
• (17g) (17g) lldp: fast start, shutdown ordering, and a required datapath namespace #1925 fast start, shutdown ordering, required datapath namespace

Stacked on (16) #1905, capture. (18) #1907, rate limiting, and (19) #1919, traffic protection, build on top.

This PR: advertise the gateway on every port

What a port advertises

TLV	Value
Chassis ID	locally assigned: the gateway's name from its k8s config
Port ID	the interface name from the command line
Port Description	the device: bus address and driver, e.g. 0000:03:00.0 (mlx5_pci)
System Name	the gateway's name
System Description	dataplane and version
System Capabilities	router, enabled
Management Address	the host's default-route source address (host namespace)
802.3 Max Frame Size	MTU plus Ethernet header and FCS

Nothing is advertised until a config has named the gateway. A chassis ID must not change while neighbours are listening, and the k8s name is the one a person recognises.

Wiring

• Per-port handles: the control-plane bridge gives each port an injection handle, a link-state watch and an identity watch.
• Identities are held until shutdown, not just until the datapath drops its end. The kernel driver reports each port once and then drops its end, and LLDP stops advertising on a port whose identity closes. vlab found this one.
• Link state is polled once a second from the driver supervisors: rte_eth_link_get_nowait for DPDK and sysfs carrier for the kernel driver. DPDK used to read link state only once, at bring-up.
• Inputs from elsewhere: mgmt publishes the gateway's name, and the host runtime finds the management address.

Testing

• n-vm, against real taps: the bridge, agent and transmitter together produce LLDP frames naming the gateway and the port. A reported identity outlives the datapath's end.
• vlab (kernel driver, two gateways): the spines' show lldp neighbors shows each gateway with the TLVs above.

🤖 Generated with Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Essentials
• Run ID: f47bbdd4-b9ed-4d63-a8e4-a3c1974db28f

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

Codecov Report

❌ Patch coverage is 18.16239% with 383 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
dataplane/src/drivers/cpbridge.rs	0.00%	117 Missing ⚠️
dataplane/src/lldp_agent.rs	42.26%	105 Missing and 7 partials ⚠️
dataplane/src/drivers/dpdk/port.rs	0.00%	46 Missing ⚠️
dataplane/src/drivers/dpdk/mod.rs	0.00%	31 Missing ⚠️
dataplane/src/drivers/kernel/mod.rs	0.00%	20 Missing ⚠️
dataplane/src/drivers/kernel/kif.rs	0.00%	19 Missing ⚠️
mgmt/src/processor/proc.rs	0.00%	18 Missing ⚠️
dataplane/src/runtime.rs	0.00%	16 Missing ⚠️
mgmt/src/tests/mgmt.rs	0.00%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!

[mvachhar]

Note

AI-assisted review: This review was drafted with Claude and reviewed by @mvachhar.

The wiring looks right: the management address is looked up from the host namespace, and nothing new runs on the packet workers beyond one injected frame per port per interval. Requesting changes for one thing: LLDP should start as soon as the dataplane does, rather than waiting for a config, so it's there to help debug a gateway whose config never arrives (inline).

[mvachhar]

Note

AI-assisted review: This comment was drafted with Claude and reviewed by @mvachhar.

No LLDP until a config arrives means no LLDP exactly when we'd want it most. If the config can't be fetched (k8s unreachable, wrong name, config error), the gateway never advertises, and nothing on the switch side shows that a gateway is even cabled up. LLDP should start as early as possible so it can help with debugging.

The name doesn't actually have to wait for the config: init_name() in runtime.rs already settles it at startup (--name, else the hostname), and it's the name the dataplane uses to fetch its own GatewayAgent config (MgmtParams.hostname). In the fabric the pod runs with hostNetwork and is pinned with kubernetes.io/hostname = gw.Name, so that hostname is the gateway name. Could the agent:

• advertise from startup with the chassis ID taken from init_name(), which is stable across restarts;
• use the config's name for System Name once it arrives (System Name can change freely);
• say (awaiting configuration) in System Description until a config is applied, so show lldp neighbors detail on the switch tells you the gateway is up but unconfigured?

If the config's name ever differs from the startup name, #1922's transmitter already handles that cleanly (shutdown frame for the old chassis ID, reinitDelay, then the new one), so there's no risk of a stale neighbour.