fix(controller): keep IOL runner containers off the kernel datapath

The unix-socket container gate matched GNS3_UNIX_SOCKET_NIO in the node
environment only, but the IOL runner reaches the same wiring through the
GNS3_IOL_RUNNER marker: the compute selects IOLDockerVM on it, and that
class forces the AF_UNIX socket pairs whatever the generic knob says —
and the documented environment for those nodes carries only the marker.
Such a container was therefore handed a kernel link its compute cannot
serve. The plain case failed at link creation (no veth end to enslave);
on the switch fast path it was worse — the switch defers a join against a
missing anchor by design (a link to a stopped node must not fail), so the
cable was silently dead instead of merely slow.

Match both markers, through the shared is_iol_runner_environment helper so
the controller and the compute select the same way. Verified live on a
real server: docker↔IOL-XE and switch↔IOL-XE links stay relay while the
docker↔switch control link still takes the kernel datapath.

gns3server/controller/udp_link.py

@@ -20,6 +20,7 @@ import asyncio
 import logging
 
 from gns3server.config import Config
+from gns3server.utils.application_id import is_iol_runner_environment
 from gns3server.utils.kernel_anchor import kernel_anchor_name
 from gns3server.utils.packet_filter_validation import (
     KERNEL_UNSUPPORTED_FILTERS,
@@ -56,14 +57,18 @@ log = logging.getLogger(__name__)
 def _is_unix_socket_docker(node):
     """
     Best-effort detection of vendor containers bridged through AF_UNIX
-    socket pairs (GNS3_UNIX_SOCKET_NIO in their environment): they have no
-    host-side interface to enslave into a kernel bridge. The compute side
+    socket pairs: they have no host-side interface to enslave into a kernel
+    bridge. Two ways in — the generic GNS3_UNIX_SOCKET_NIO knob, or the IOL
+    runner marker, whose class (IOLDockerVM, selected on GNS3_IOL_RUNNER)
+    forces the unix-socket wiring whatever the knob says. The compute side
     independently rejects kernel-datapath NIOs on these containers, so a
-    missed detection here only surfaces as a clearer-late error.
+    missed detection surfaces as a clearer-late error — except on the switch
+    fast path, where an absent anchor is deferred forever by design (a link
+    to a stopped node must not fail), making the miss a silently dead cable.
     """
 
     environment = (node.properties or {}).get("environment") or ""
-    return "GNS3_UNIX_SOCKET_NIO" in environment
+    return "GNS3_UNIX_SOCKET_NIO" in environment or is_iol_runner_environment(environment)
 
 
 class UDPLink(Link):

tests/controller/test_kernel_datapath_link.py

@@ -419,6 +419,47 @@ async def test_kernel_datapath_not_eligible_for_unix_socket_containers(project):
     assert link._kernel_datapath_eligible(node1, node2) is False
 
 
+@pytest.mark.asyncio
+async def test_kernel_datapath_not_eligible_for_iol_runner_containers(project):
+    """
+    The IOL runner marker is the other way into the unix-socket datapath: the
+    compute selects IOLDockerVM on GNS3_IOL_RUNNER and that class wires the
+    adapters through AF_UNIX socket pairs whatever the generic
+    GNS3_UNIX_SOCKET_NIO knob says — and the documented environment for
+    those nodes carries only the marker. Matching the knob alone let them
+    through to a kernel link their compute cannot serve.
+    """
+
+    link, node1, node2 = await _kernel_link(project)
+    node2._properties = {"environment": "GNS3_IOL_RUNNER=1"}
+    assert link._kernel_datapath_eligible(node1, node2) is False
+
+    # the marker travels in a multi-line environment like any other knob
+    node2._properties = {"environment": "GNS3_IOL_RUNNER=1\nGNS3_IOL_STARTUP_CONFIG=cfg.txt"}
+    assert link._kernel_datapath_eligible(node1, node2) is False
+
+
+@pytest.mark.asyncio
+async def test_kernel_datapath_not_eligible_for_an_iol_runner_container_on_a_switch(project):
+    """
+    The switch fast path absorbs the peer's anchor, and a container bridged
+    through unix sockets has none — nor will it ever. The switch defers a
+    join against a stopped peer by design (it must not fail the link), so
+    such a cable would be silently dead instead of merely slow: the gate has
+    to run before the switch branch, keeping the link on the relay the
+    switch's own port TAP serves.
+    """
+
+    compute = MagicMock()
+    compute.id = "compute-1"
+    compute.capabilities = {"ubridge_tap": True, "ubridge_iol_tap": True}
+    switch = _node(project, compute, "sw1", node_type="ethernet_switch")
+    peer = _node(project, compute, "iol1", environment="GNS3_IOL_RUNNER=1")
+    link = await _link(project, switch, peer)
+
+    assert link._kernel_datapath_eligible(switch, peer) is False
+
+
 @pytest.mark.asyncio
 async def test_kernel_datapath_disabled_by_config(project):
 
@@ -915,9 +956,7 @@ async def test_switch_link_repushes_the_switch_nio_on_node_start(project):
     switch.put = AsyncioMagicMock()
 
     await link.node_started(peer)
-    switch.put.assert_called_once_with(
-        "/adapters/0/ports/0/nio", data=entries[0][3], timeout=120
-    )
+    switch.put.assert_called_once_with("/adapters/0/ports/0/nio", data=entries[0][3], timeout=120)
 
     # a relay link never re-pushes anything
     switch.put.reset_mock()