Merge pull request #30936 from yuwata/network-automatically-reconfigure-interface-on-failure

network: automatically reconfigure interface on failure

src/network/networkd-link.c

@@ -381,6 +381,8 @@ int link_stop_engines(Link *link, bool may_keep_dhcp) {
 }
 
 void link_enter_failed(Link *link) {
+        int r;
+
         assert(link);
 
         if (IN_SET(link->state, LINK_STATE_FAILED, LINK_STATE_LINGER))
@@ -390,7 +392,22 @@ void link_enter_failed(Link *link) {
 
         link_set_state(link, LINK_STATE_FAILED);
 
-        (void) link_stop_engines(link, false);
+        if (!ratelimit_below(&link->automatic_reconfigure_ratelimit)) {
+                log_link_warning(link, "The interface entered the failed state frequently, refusing to reconfigure it automatically.");
+                goto stop;
+        }
+
+        log_link_info(link, "Trying to reconfigure the interface.");
+        r = link_reconfigure(link, /* force = */ true);
+        if (r < 0) {
+                log_link_warning_errno(link, r, "Failed to reconfigure interface: %m");
+                goto stop;
+        }
+
+        return;
+
+stop:
+        (void) link_stop_engines(link, /* may_keep_dhcp = */ false);
 }
 
 void link_check_ready(Link *link) {
@@ -416,11 +433,9 @@ void link_check_ready(Link *link) {
         if (!link->activated)
                 return (void) log_link_debug(link, "%s(): link is not activated.", __func__);
 
-        if (link->iftype == ARPHRD_CAN) {
+        if (link->iftype == ARPHRD_CAN)
                 /* let's shortcut things for CAN which doesn't need most of checks below. */
-                link_set_state(link, LINK_STATE_CONFIGURED);
-                return;
-        }
+                goto ready;
 
         if (!link->stacked_netdevs_created)
                 return (void) log_link_debug(link, "%s(): stacked netdevs are not created.", __func__);
@@ -2554,6 +2569,7 @@ static int link_new(Manager *manager, sd_netlink_message *message, Link **ret) {
                 .n_ref = 1,
                 .state = LINK_STATE_PENDING,
                 .online_state = _LINK_ONLINE_STATE_INVALID,
+                .automatic_reconfigure_ratelimit = (const RateLimit) { .interval = 10 * USEC_PER_SEC, .burst = 5 },
                 .ifindex = ifindex,
                 .iftype = iftype,
                 .ifname = TAKE_PTR(ifname),

src/network/networkd-link.h

@@ -25,6 +25,7 @@
 #include "networkd-ipv6ll.h"
 #include "networkd-util.h"
 #include "ordered-set.h"
+#include "ratelimit.h"
 #include "resolve-util.h"
 #include "set.h"
 
@@ -106,6 +107,7 @@ typedef struct Link {
         LinkAddressState ipv4_address_state;
         LinkAddressState ipv6_address_state;
         LinkOnlineState online_state;
+        RateLimit automatic_reconfigure_ratelimit;
 
         unsigned static_address_messages;
         unsigned static_address_label_messages;

test/test-network/systemd-networkd-tests.py

@@ -12,6 +12,7 @@ import itertools
 import json
 import os
 import pathlib
+import random
 import re
 import shutil
 import signal
@@ -5109,9 +5110,7 @@ class NetworkdRATests(unittest.TestCase, Utilities):
 
         self.teardown_nftset('addr6', 'network6', 'ifindex')
 
-    def test_ipv6_token_static(self):
-        copy_network_unit('25-veth.netdev', '25-ipv6-prefix.network', '25-ipv6-prefix-veth-token-static.network')
-        start_networkd()
+    def check_ipv6_token_static(self):
         self.wait_online(['veth99:routable', 'veth-peer:degraded'])
 
         output = networkctl_status('veth99')
@@ -5121,6 +5120,26 @@ class NetworkdRATests(unittest.TestCase, Utilities):
         self.assertRegex(output, '2002:da8:2:0:1a:2b:3c:4d')
         self.assertRegex(output, '2002:da8:2:0:fa:de:ca:fe')
 
+    def test_ipv6_token_static(self):
+        copy_network_unit('25-veth.netdev', '25-ipv6-prefix.network', '25-ipv6-prefix-veth-token-static.network')
+        start_networkd()
+
+        self.check_ipv6_token_static()
+
+        for _ in range(20):
+            check_output('ip link set veth99 down')
+            check_output('ip link set veth99 up')
+
+        self.check_ipv6_token_static()
+
+        for _ in range(20):
+            check_output('ip link set veth99 down')
+            time.sleep(random.uniform(0, 0.1))
+            check_output('ip link set veth99 up')
+            time.sleep(random.uniform(0, 0.1))
+
+        self.check_ipv6_token_static()
+
     def test_ipv6_token_prefixstable(self):
         copy_network_unit('25-veth.netdev', '25-ipv6-prefix.network', '25-ipv6-prefix-veth-token-prefixstable.network')
         start_networkd()