Commit Graph

7 Commits

Author SHA1 Message Date
Donavan Fritz 9b777ca7d1 bird: per-peer import filter rejects connected subnet
Build flock Image / build (push) Successful in 2m17s
Without a filter, crt001's `network 2602:817:3000:A25::/64` gets
re-advertised to every peer on that subnet. bird installs the BGP /64
with metric 32, beating the kernel-connected route at 256, and all
inter-host VLAN-25 traffic hairpins through the gateway — losing PMTU
9000 and ~30x throughput. Broke Plex 2026-05-04: NFS to nas002 capped
at 7 MB/s, jumbo blackholed.

Add LocalSubnetV6/V4 (CIDR) to NodeBGP. Agent populates by masking the
peer's address to /64 (v6) or /24 (v4) — same fritzlab convention
already in localAddrSameSubnet. Render emits `import where net !=
<subnet>;` per BGP channel when set, falls back to `import all;`
otherwise so existing tests stay green.

Defence in depth: with the matching outbound route-map on crt001
(ROUTE_MAP_CLUSTER_OUT_V{4,6}) the agent now refuses the leak on its
own if the router filter ever drifts.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-04 21:03:59 -05:00
Donavan Fritz 71e584cf96 NodeConfig defaults + code-quality pass + fuzz tests + README
NodeConfig.Spec.Defaults adds per-node IPv6/IPv4 family defaults that pod
annotations can override; built-in baseline (v6=true, v4=false) still
applies when the field is omitted.

bird.Render now validates every operator-supplied value (peer addresses,
CIDRs, anycast IPs, source addresses) before templating — fuzz found a
peer address containing `}` produced unbalanced braces in bird.conf.
Failing input preserved as a regression seed.

Fuzz targets added for ParseAnnotations, ParseCNIArgs, HostIfaceName,
canonical, IPAM allocate sequences, embed.Embed, and bird.Render.
Hardened canonical/ipToU32 against nil and non-IPv4 inputs.

README rewritten for outside readers — quickstart, NodeConfig + annotation
reference with worked examples, anycast use cases, comparison vs Calico
and Cilium, requirements, limitations.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-25 09:25:45 -05:00
Donavan Fritz 3117d00210 bird: declare anycast as protocol static; filter static→kernel export
Build flock Image / build (push) Has been cancelled
Two coupled changes that fix the anycast advertisement path:

1. Add anycast /128 + /32 prefixes as `route … blackhole` lines in the
   protocol static stanzas. BIRD's master tables pick them up at
   preference 200 — higher than kernel-learned routes — so they're the
   ones the BGP export filter sees.

2. The kernel protocol's export filter now rejects RTS_STATIC. Without
   this, BIRD would push its blackhole back into the kernel, clobbering
   the agent-installed `<anycast> via <pod-eth0> dev flock<8hex>` route
   that's actually responsible for forwarding to the pod.

Result: BIRD has the route to advertise via BGP; the kernel has the
right route to forward; nothing fights over the kernel table.

Replaces the abandoned `gateway recursive` attempt — that's a BIRD 1.x
keyword, not BIRD 2.15.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-25 08:16:45 -05:00
Donavan Fritz 7ac497249f bird: gateway recursive on BGP protocols
Build flock Image / build (push) Has been cancelled
Default is `gateway direct` — BIRD silently rejects kernel routes whose
via address isn't on a directly-connected network interface. Our anycast
host routes use a pod /128 (or /32) as via, which is itself a kernel
route on a flock veth, not a connected network. With `gateway
recursive`, BIRD does a recursive lookup and accepts the kernel route.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-25 08:09:00 -05:00
Donavan Fritz 3f6dfd3e88 bird: add source address + next hop self (v6 anycast fix)
Build flock Image / build (push) Has been cancelled
Cisco IOS rejects IPv6 BGP advertisements whose next-hop is link-local-
only. BIRD2 was synthesising a link-local next-hop for kernel-learned
routes whose dev had no via gateway (our anycast /128s). Symptom: v4
anycast worked (Cisco doesn't have the same constraint for /32s), v6
anycast didn't make it past crt001.

- pkg/routing/bird/config.go: NodeBGP.LocalV6/LocalV4. Template now
  emits `local <addr> as <asn>` and `next hop self;` in the BGP
  channel for both families, mirroring Calico's `source address` +
  `next hop self` pattern.
- pkg/agent/bird.go: localAddrSameSubnet picks an interface address
  on the peer's /64 or /24 to use as source.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-25 07:45:51 -05:00
Donavan Fritz a1222f13cc bird: add learn + explicit static blackhole protocols
Build flock Image / build (push) Has been cancelled
BIRD2's protocol kernel does not import kernel routes by default; the
import filter on the channel is just for what BIRD has already learned.
Added `learn;` so the kernel-installed blackholes (from the agent's
SummaryRoutes) are picked up.

Also added explicit `protocol static static6/static4` with one
`route <cidr> blackhole;` per NodeConfig CIDR. This is belt-and-
suspenders: even if `learn` doesn't capture the kernel blackhole, BIRD
has the route directly and exports it via the BGP filter.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-24 23:06:25 -05:00
Donavan Fritz eb1f5e0d8d M2: netlink, IPAM/handler wiring, BIRD sidecar, CNI installer
Build flock Image / build (push) Has been cancelled
Code (Linux build, with no-op stubs for macOS dev):
- pkg/agent/netns_linux.go: ensureVeth → host-side configure (addrgenmode
  none, fe80::1/64, proxy_arp, forwarding) → move peer to pod ns →
  configure pod side (addr, default route via fe80::1, v4 169.254.1.1
  on-link gateway) → host /128 + /32 routes. Idempotent.
- pkg/agent/hostiface.go: deterministic host iface name flock<8hex> from
  FNV-1a-32(containerID).
- pkg/agent/annotations.go: parse flock.fritzlab.net/{ipv6,ipv4,cidr6,
  cidr4,ip-algo,anycast} with design-doc defaults; ParseCNIArgs for the
  K8S_POD_* keys kubelet sets.
- pkg/agent/podinfo.go: shared informer scoped to spec.nodeName==NODE,
  WaitForPod helper for ADD-vs-informer-sync race.
- pkg/agent/handlers.go: PodHandler does
    cache lookup → annotations → IPAM → store(pending) → SetupFunc →
    store(committed) → Result. Idempotent on retry. Del symmetric.
- pkg/routing/bird/config.go: text/template render with stable ordering;
  golden tests for host001 + anycast injection + sort stability.
- pkg/agent/bird.go: writes /etc/flock/bird/bird.conf, debounces 500ms,
  execs `birdc -s /run/flock/bird.ctl configure`. Installs blackhole
  kernel routes for the node summary CIDRs so BIRD's protocol kernel
  imports them.
- pkg/agent/runtime_linux.go: at startup, waits up to 60s for the per-
  node NodeConfig, reconciles committed allocations into IPAM.used,
  garbage-collects pending entries, builds PodHandler, swaps RPC
  handlers in.
- cmd/flock-installer: init-container binary that copies /opt/cni/bin/
  flock and writes 01-flock.conflist (lex-first so kubelet picks it
  over Calico's 10-calico.conflist on flock-labeled nodes).

Deploy:
- Dockerfile: alpine + iproute2 + bird2; multi-binary image.
- deploy/daemonset.yaml: install-cni init container; bird sidecar
  sharing /etc/flock/bird + /run/flock with the agent; ConfigMap-seeded
  bootstrap bird.conf so the sidecar boots before the agent renders.
  Privileged on flock-agent + install-cni; bird sidecar uses
  NET_ADMIN/RAW only.
- RBAC: pods + networkpolicies get/list/watch (the latter is reserved
  for M8 — harmless to grant now).

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
2026-04-24 22:33:48 -05:00