Merge pull request #346 from nuclearcat/ipoe-stale-session-flush

Ipoe stale session flush fixes

2 files changed, 179 insertions, 0 deletions

diff --git a/tests/accel-pppd/ipoe/test_ipoe_link_ops.py b/tests/accel-pppd/ipoe/test_ipoe_link_ops.py
new file mode 100644
index 00000000..00d06af4
--- /dev/null
+++ b/tests/accel-pppd/ipoe/test_ipoe_link_ops.py
@@ -0,0 +1,82 @@
+import pytest
+from common import process, ipoe_iface
+
+
+@pytest.fixture()
+def accel_pppd_config(veth_pair_netns):
+    return (
+        """
+    [modules]
+    connlimit
+    radius
+    ipoe
+    ippool
+
+    [ip-pool]
+    gw-ip-address=192.0.2.1
+    192.0.2.2-255
+
+    [cli]
+    tcp=127.0.0.1:2001
+
+    [core]
+    log-error=/dev/stderr
+
+    [log]
+    log-debug=/dev/stdout
+    log-file=/dev/stdout
+    log-emerg=/dev/stderr
+    level=5
+
+    [radius]
+
+    [ipoe]
+    noauth=1
+    shared=1
+    gw-ip-address=192.0.2.1/24
+    interface=re:."""
+        + veth_pair_netns["veth_a"][1:]
+    )
+
+
+# sessions carry private state set up over generic netlink, a device made by
+# rtnetlink would have none of it
+@pytest.mark.dependency(depends=["ipoe_driver_loaded"], scope="session")
+@pytest.mark.ipoe_driver
+def test_ipoe_interface_cannot_be_created_by_iproute():
+    (exit_code, out, err) = process.run(
+        ["ip", "link", "add", "ipoetest0", "type", "ipoe"]
+    )
+    print("ip link add: exit=%d out=%s err=%s" % (exit_code, out, err))
+
+    # remove it again in case it got created anyway, so that the rest of the
+    # suite is not affected
+    process.run(["ip", "link", "del", "ipoetest0"])
+
+    assert exit_code != 0
+
+
+# stale interfaces have to be removable without restarting accel-pppd, which
+# would drop every remaining session
+@pytest.mark.dependency(depends=["ipoe_driver_loaded"], scope="session")
+@pytest.mark.ipoe_driver
+def test_ipoe_interface_can_be_deleted_by_iproute(
+    accel_pppd_instance, dhclient_instance, accel_cmd, veth_pair_netns
+):
+    assert accel_pppd_instance, "accel-pppd did not start"
+    assert dhclient_instance["is_started"]
+    assert ipoe_iface.wait_for_session(accel_cmd, veth_pair_netns["veth_a"])
+
+    before = ipoe_iface.list_ipoe()
+    print("ipoe interfaces: " + str(before))
+    assert len(before) > 0
+
+    (ifindex, name) = before[0]
+
+    (exit_code, out, err) = process.run(["ip", "link", "del", name])
+    print("ip link del %s: exit=%d out=%s err=%s" % (name, exit_code, out, err))
+    assert exit_code == 0
+
+    after = ipoe_iface.list_ipoe()
+    print("ipoe interfaces after delete: " + str(after))
+    assert (ifindex, name) not in after
diff --git a/tests/accel-pppd/ipoe/test_ipoe_stale_cleanup.py b/tests/accel-pppd/ipoe/test_ipoe_stale_cleanup.py
new file mode 100644
index 00000000..b7cfd533
--- /dev/null
+++ b/tests/accel-pppd/ipoe/test_ipoe_stale_cleanup.py
@@ -0,0 +1,97 @@
+import pytest
+from common import accel_pppd_process, ipoe_iface
+
+
+@pytest.fixture()
+def accel_pppd_config(veth_pair_netns):
+    return (
+        """
+    [modules]
+    connlimit
+    radius
+    ipoe
+    ippool
+
+    [ip-pool]
+    gw-ip-address=192.0.2.1
+    192.0.2.2-255
+
+    [cli]
+    tcp=127.0.0.1:2001
+
+    [core]
+    log-error=/dev/stderr
+
+    [log]
+    log-debug=/dev/stdout
+    log-file=/dev/stdout
+    log-emerg=/dev/stderr
+    level=5
+
+    [radius]
+
+    [ipoe]
+    noauth=1
+    shared=1
+    gw-ip-address=192.0.2.1/24
+    interface=re:."""
+        + veth_pair_netns["veth_a"][1:]
+    )
+
+
+# accel-pppd killed with SIGKILL has no chance to remove the session
+# interfaces it created, so they are left in the kernel. The next instance is
+# supposed to drop them while starting up.
+@pytest.mark.dependency(depends=["ipoe_driver_loaded"], scope="session")
+@pytest.mark.ipoe_driver
+def test_ipoe_stale_interfaces_removed_after_sigkill(
+    accel_pppd_instance,
+    dhclient_instance,
+    accel_cmd,
+    accel_pppd,
+    accel_pppd_config_file,
+    veth_pair_netns,
+    pytestconfig,
+):
+    assert accel_pppd_instance, "accel-pppd did not start"
+    assert dhclient_instance["is_started"]
+    assert ipoe_iface.wait_for_session(accel_cmd, veth_pair_netns["veth_a"])
+
+    # the session must have created an interface, otherwise there is nothing
+    # for this test to check
+    before = ipoe_iface.list_ipoe()
+    print("ipoe interfaces before the crash: " + str(before))
+    assert len(before) > 0
+
+    assert ipoe_iface.kill_accel_pppd() > 0
+
+    # nothing removes them while no accel-pppd is running (not asserted, the
+    # kernel module is free to start doing it on its own one day)
+    print("ipoe interfaces after the crash: " + str(ipoe_iface.list_ipoe()))
+
+    # start again, it replies to 'show version' only once the startup flush is
+    # done, so there is no need to wait for anything else
+    (is_started, thread, control) = accel_pppd_process.start(
+        accel_pppd,
+        ["-c" + accel_pppd_config_file],
+        accel_cmd,
+        pytestconfig.getoption("accel_pppd_max_wait_time"),
+    )
+
+    try:
+        assert is_started
+
+        after = ipoe_iface.list_ipoe()
+        print("ipoe interfaces after the restart: " + str(after))
+
+        # compared by ifindex: dhclient may get a new session in the meantime,
+        # and the fresh interface would reuse the ipoe0 name
+        stale = set(before) & set(after)
+        assert not stale, "interfaces left over from the killed instance: " + str(stale)
+    finally:
+        accel_pppd_process.end(
+            thread,
+            control,
+            accel_cmd,
+            pytestconfig.getoption("accel_pppd_max_finish_time"),
+        )