fix(esw): harden the cascade end the create race leaves unhardened

docker create_veth hardens only its FIRST end (the Docker shape: host
anchor plus a container leg that keeps its L3 life). A cascade end is
that second, unhardened end whenever the peer switch's create won the
race, and the kernel gives it an IPv6 link-local — a silent L3 identity
on the switch fabric (ubridge-l2-anchor-spec §E). Each switch now
hardens its OWN end with 'link l2only <end> on' once the pair exists,
whatever branch brought it there (fresh create, lost race, reuse);
idempotent, and best-effort with a warning on a uBridge without the
command.
This commit is contained in:
YueGuobin 2026-10-02 21:49:14 +08:00
parent 6e36b58a5d
commit cb7f74c77b
No known key found for this signature in database
2 changed files with 47 additions and 0 deletions

View File

@ -476,6 +476,22 @@ class EthernetSwitch(KernelDatapathMixin, BaseNode):
else:
with contextlib.suppress(UbridgeError):
await self._ubridge_send(f'tc reset "{anchor}"')
# ``docker create_veth`` hardens only its FIRST end (the Docker
# shape: host anchor plus a container leg that keeps its L3
# life). Here BOTH ends are anchors, and whichever switch loses
# the create race owns the second, unhardened end — its kernel
# IPv6 link-local would give the switch fabric a silent L3
# identity (ubridge-l2-anchor-spec §E). Each switch hardens its
# OWN end once the pair exists; idempotent, and it covers a
# reused pre-existing end as well.
try:
await self._ubridge_send(f'link l2only "{anchor}" on')
except UbridgeError as e:
log.warning(
'Ethernet switch "{name}" [{id}]: could not harden cascade end "{anchor}" L2-only: {error}'.format(
name=self._name, id=self._id, anchor=anchor, error=e
)
)
# Register before _kernel_attach: the marker reconcile recognises
# kernel anchors through _kernel_anchors().
self._kernel_ports[port_number] = anchor

View File

@ -956,6 +956,37 @@ class TestEthernetSwitchNodesRoutes:
node._ubridge_send.assert_any_call(f'brctl addif "{br}" "{anchor}"')
assert node._kernel_ports[0] == anchor
async def test_ethernet_switch_create_cascade_nio_hardens_own_end_l2only(
self, app: FastAPI, compute_client: AsyncClient, compute_project: Project, ethernet_switch: dict
) -> None:
"""
``docker create_veth`` hardens only its FIRST end (the Docker shape:
host anchor plus a container leg that keeps its L3 life) — but a
cascade end the peer's winning create minted is that second,
unhardened end, and the kernel gives it an IPv6 link-local (a silent
L3 identity on the switch fabric, ubridge-l2-anchor-spec §E). Each
switch therefore hardens its OWN end once the pair exists, whatever
branch brought it there (fresh create, lost race, reuse).
"""
anchor, peer = "gs1a2b3c4d5e0", "gs1a2b3c4d5e1"
url = app.url_path_for(
"compute:create_ethernet_switch_nio",
project_id=ethernet_switch["project_id"],
node_id=ethernet_switch["node_id"],
adapter_number="0",
port_number="0",
)
# pre-existing end: the branch whose end create_veth did NOT harden
with self._cascade_sysfs(anchor, {"value": True}):
response = await compute_client.post(url, json=self._cascade_params(anchor, peer))
assert response.status_code == status.HTTP_201_CREATED
node = compute_project.get_node(ethernet_switch["node_id"])
node._ubridge_send.assert_any_call(f'link l2only "{anchor}" on')
# only its own end — the peer's end is the peer switch's to harden
assert not any(f'"{peer}"' in str(c) and "l2only" in str(c) for c in node._ubridge_send.call_args_list)
async def test_ethernet_switch_delete_cascade_nio_destroys_the_pair(
self, app: FastAPI, compute_client: AsyncClient, compute_project: Project, ethernet_switch: dict
) -> None: