boot: T3871: rework interface renaming and ordering

python/vyos/ethtool.py

@@ -217,6 +217,8 @@ def check_speed_duplex(self, speed, duplex):
 
     def check_flow_control(self):
         """ Check if the NIC supports flow-control """
+        if self.get_driver_name() in _drivers_without_speed_duplex_flow:
+            return False
         return bool(self._flow_control)
 
     def get_flow_control(self):

smoketest/scripts/cli/test_interfaces_ethernet.py

@@ -229,6 +229,10 @@ def test_ethtool_coalesce(self):
                     msg = 'Driver does not fully support coalesce configuration'
                     with self.assertRaisesRegex(ConfigSessionError, msg):
                         self.cli_commit()
+                    # the failed commit leaves rx-usecs/tx-usecs staged in
+                    # the candidate config (commit() does not auto-rollback) -
+                    # discard it so it doesn't leak into the next interface
+                    self.cli_discard()
                     continue
 
                 # To find out the supported features
@@ -287,17 +291,25 @@ def test_ethtool_coalesce(self):
 
     def test_ethtool_flow_control(self):
         for interface in self._interfaces:
+            ethtool = Ethtool(interface)
             # Disable flow-control
             self.cli_set(self._base_path + [interface, 'disable-flow-control'])
-            # Check current flow-control state on ethernet interface
-            out, err = popen(f'sudo ethtool --json --show-pause {interface}')
-            # Flow-control not supported - test if it bails out with a proper
-            # this is a dynamic path where err = 1 on VMware, but err = 0 on
-            # a physical box.
-            if bool(err):
+
+            # Ask the same capability check the CLI commit itself uses,
+            # rather than a raw ethtool --show-pause probe: some drivers
+            # (virtio_net, vmxnet3, xen_netfront, ...) support querying
+            # pause parameters but not changing them, so a bare --show-pause
+            # exit code is not a reliable predictor of whether the commit
+            # will succeed.
+            if not ethtool.check_flow_control():
                 with self.assertRaises(ConfigSessionError):
                     self.cli_commit()
+                # the failed commit leaves disable-flow-control staged in
+                # the candidate config (commit() does not auto-rollback) -
+                # discard it so it doesn't leak into the next interface
+                self.cli_discard()
             else:
+                out, err = popen(f'sudo ethtool --json --show-pause {interface}')
                 out = loads(out)
                 # Flow control is on
                 self.assertTrue(out[0]['autonegotiate'])

src/helpers/vyos-interface-rescan.py

@@ -97,7 +97,8 @@ def get_interface_type(intf: str) -> str:
     return intf_type
 
 def get_new_interfaces() -> dict:
-    """ Read any new interface data left in /run/udev/vyos by vyos_net_name
+    """ Read any new interface data left in /run/udev/vyos by
+    vyos-net-name-resolve.py for interfaces without a configured hw-id
     """
     interfaces = {}
 

src/init/vyos-router

@@ -367,6 +367,26 @@ update_interface_config ()
     fi
 }
 
+# warn (but do not fail boot) about configured hw-id interfaces whose
+# hardware was not found by vyos-net-name-resolve.service
+warn_missing_interface_hardware ()
+{
+    local status_file="/run/vyos-net-name-resolve.json"
+    [ -f "${status_file}" ] || return 0
+
+    jq -r '.missing // {} | to_entries[] | "\(.value) \(.key)"' "${status_file}" \
+        2>/dev/null | while read -r intf hwid; do
+        [ -n "${intf}" ] || continue
+        log_failure_msg "hw-id ${hwid} configured as '${intf}' was not found during boot - check cabling, hardware and driver/firmware"
+    done
+
+    jq -r '.pending_unresolved // [] | .[]' "${status_file}" 2>/dev/null | \
+        while read -r intf; do
+        [ -n "${intf}" ] || continue
+        log_failure_msg "interface '${intf}' still has no hw-id configured after this boot's naming pass - bind it manually, e.g. 'set interfaces ethernet ${intf} hw-id <mac>', or remove the node if it is no longer needed"
+    done
+}
+
 cleanup_post_commit_hooks () {
     # Remove links from the post-commit hooks directory.
     # note that this approach only supports hooks that are "configured",
@@ -610,13 +630,20 @@ start ()
 
     disabled migrate || migrate_bootfile || overall_status=1
 
+    # config.boot is only guaranteed to exist/be current from this point on
+    # (tmpfs mount + encrypted config unlock + syntax migration have all
+    # already happened above) - hw-id based interface naming must be
+    # resolved against it before anything below applies interface config.
+    systemctl start vyos-net-name-resolve.service || overall_status=1
+
     restore_if_missing_preconfig_script
 
     run_preconfig_script
 
     run_postupgrade_script
 
     update_interface_config || overall_status=1
+    warn_missing_interface_hardware
 
     disabled system_activate || system_activate || overall_status=1