Merge pull request #24456 from yuwata/network-tcp-congctl

network: introduce TCPCongestionControlAlgorithm=

man/systemd.network.xml

@@ -1622,6 +1622,15 @@ Table=1234</programlisting></para>
         </listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>TCPCongestionControlAlgorithm=</varname></term>
+        <listitem>
+          <para>Specifies the TCP congestion control algorithm for the route. Takes a name of the algorithm,
+          e.g. <literal>bbr</literal>, <literal>dctcp</literal>, or <literal>vegas</literal>. When unset,
+          the kernel's default will be used.</para>
+        </listitem>
+      </varlistentry>
+
       <varlistentry>
         <term><varname>MultiPathRoute=<replaceable>address</replaceable>[@<replaceable>name</replaceable>] [<replaceable>weight</replaceable>]</varname></term>
         <listitem>

src/libsystemd/sd-netlink/netlink-types-rtnl.c

@@ -929,7 +929,7 @@ static const NLAPolicy rtnl_route_metrics_policies[] = {
         [RTAX_RTO_MIN]            = BUILD_POLICY(U32),
         [RTAX_INITRWND]           = BUILD_POLICY(U32),
         [RTAX_QUICKACK]           = BUILD_POLICY(U32),
-        [RTAX_CC_ALGO]            = BUILD_POLICY(U32),
+        [RTAX_CC_ALGO]            = BUILD_POLICY(STRING),
         [RTAX_FASTOPEN_NO_COOKIE] = BUILD_POLICY(U32),
 };
 

src/network/networkd-network-gperf.gperf

@@ -196,6 +196,7 @@ Route.Type,                                  config_parse_route_type,
 Route.InitialCongestionWindow,               config_parse_tcp_window,                                  0,                             0
 Route.InitialAdvertisedReceiveWindow,        config_parse_tcp_window,                                  0,                             0
 Route.TCPAdvertisedMaximumSegmentSize,       config_parse_tcp_advmss,                                  0,                             0
+Route.TCPCongestionControlAlgorithm,         config_parse_tcp_congestion,                              0,                             0
 Route.QuickAck,                              config_parse_route_boolean,                               0,                             0
 Route.FastOpenNoCookie,                      config_parse_route_boolean,                               0,                             0
 Route.TTLPropagate,                          config_parse_route_boolean,                               0,                             0

src/network/networkd-route.c

@@ -107,6 +107,8 @@ Route *route_free(Route *route) {
 
         sd_event_source_disable_unref(route->expire);
 
+        free(route->tcp_congestion_control_algo);
+
         return mfree(route);
 }
 
@@ -319,6 +321,7 @@ int route_get(Manager *manager, Link *link, const Route *in, Route **ret) {
 
 int route_dup(const Route *src, Route **ret) {
         _cleanup_(route_freep) Route *dest = NULL;
+        int r;
 
         /* This does not copy mulipath routes. */
 
@@ -336,6 +339,11 @@ int route_dup(const Route *src, Route **ret) {
         dest->manager = NULL;
         dest->multipath_routes = NULL;
         dest->expire = NULL;
+        dest->tcp_congestion_control_algo = NULL;
+
+        r = free_and_strdup(&dest->tcp_congestion_control_algo, src->tcp_congestion_control_algo);
+        if (r < 0)
+                return r;
 
         *ret = TAKE_PTR(dest);
         return 0;
@@ -1231,6 +1239,12 @@ static int route_configure(const Route *route, uint32_t lifetime_sec, Link *link
                         return r;
         }
 
+        if (!isempty(route->tcp_congestion_control_algo)) {
+                r = sd_netlink_message_append_string(m, RTAX_CC_ALGO, route->tcp_congestion_control_algo);
+                if (r < 0)
+                        return r;
+        }
+
         r = sd_netlink_message_close_container(m);
         if (r < 0)
                 return r;
@@ -2498,6 +2512,46 @@ int config_parse_route_type(
         return 0;
 }
 
+int config_parse_tcp_congestion(
+                const char *unit,
+                const char *filename,
+                unsigned line,
+                const char *section,
+                unsigned section_line,
+                const char *lvalue,
+                int ltype,
+                const char *rvalue,
+                void *data,
+                void *userdata) {
+
+        Network *network = userdata;
+        _cleanup_(route_free_or_set_invalidp) Route *n = NULL;
+        int r;
+
+        assert(filename);
+        assert(section);
+        assert(lvalue);
+        assert(rvalue);
+        assert(data);
+
+        r = route_new_static(network, filename, section_line, &n);
+        if (r == -ENOMEM)
+                return log_oom();
+        if (r < 0) {
+                log_syntax(unit, LOG_WARNING, filename, line, r,
+                           "Failed to allocate route, ignoring assignment: %m");
+                return 0;
+        }
+
+        r = config_parse_string(unit, filename, line, section, section_line, lvalue, ltype,
+                                rvalue, &n->tcp_congestion_control_algo, userdata);
+        if (r < 0)
+                return r;
+
+        TAKE_PTR(n);
+        return 0;
+}
+
 int config_parse_tcp_advmss(
                 const char *unit,
                 const char *filename,

src/network/networkd-route.h

@@ -50,6 +50,7 @@ struct Route {
         uint32_t initcwnd;
         uint32_t initrwnd;
         uint32_t advmss;
+        char *tcp_congestion_control_algo;
         unsigned char pref;
         unsigned flags;
         int gateway_onlink; /* Only used in conf parser and route_section_verify(). */
@@ -123,5 +124,6 @@ CONFIG_PARSER_PROTOTYPE(config_parse_route_type);
 CONFIG_PARSER_PROTOTYPE(config_parse_tcp_window);
 CONFIG_PARSER_PROTOTYPE(config_parse_route_mtu);
 CONFIG_PARSER_PROTOTYPE(config_parse_multipath_route);
+CONFIG_PARSER_PROTOTYPE(config_parse_tcp_congestion);
 CONFIG_PARSER_PROTOTYPE(config_parse_tcp_advmss);
 CONFIG_PARSER_PROTOTYPE(config_parse_route_nexthop);

test/fuzz/fuzz-network-parser/directives

@@ -189,6 +189,7 @@ Metric=
 TTLPropagate=
 MultiPathRoute=
 TCPAdvertisedMaximumSegmentSize=
+TCPCongestionControlAlgorithm=
 NextHop=
 [Network]
 KeepMaster=

test/test-network/conf/25-route-congctl.network

@@ -0,0 +1,16 @@
+# SPDX-License-Identifier: LGPL-2.1-or-later
+[Match]
+Name=dummy98
+
+[Network]
+IPv6AcceptRA=no
+Address=2001:1234:5:8f63::1/128
+Address=149.10.124.58/28
+
+[Route]
+Destination=2001:1234:5:8fff:ff:ff:ff:ff/128
+TCPCongestionControlAlgorithm=dctcp
+
+[Route]
+Destination=149.10.124.66
+TCPCongestionControlAlgorithm=dctcp

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

@@ -2800,6 +2800,24 @@ class NetworkdNetworkTests(unittest.TestCase, Utilities):
         self.assertRegex(output, '149.10.124.48/28 proto kernel scope link src 149.10.124.58')
         self.assertRegex(output, '149.10.124.66 via inet6 2001:1234:5:8fff:ff:ff:ff:ff proto static')
 
+    @expectedFailureIfModuleIsNotAvailable('tcp_dctcp')
+    def test_route_congctl(self):
+        copy_network_unit('25-route-congctl.network', '12-dummy.netdev')
+        start_networkd()
+        self.wait_online(['dummy98:routable'])
+
+        print('### ip -6 route show dev dummy98 2001:1234:5:8fff:ff:ff:ff:ff')
+        output = check_output('ip -6 route show dev dummy98 2001:1234:5:8fff:ff:ff:ff:ff')
+        print(output)
+        self.assertIn('2001:1234:5:8fff:ff:ff:ff:ff proto static', output)
+        self.assertIn('congctl dctcp', output)
+
+        print('### ip -4 route show dev dummy98 149.10.124.66')
+        output = check_output('ip -4 route show dev dummy98 149.10.124.66')
+        print(output)
+        self.assertIn('149.10.124.66 proto static', output)
+        self.assertIn('congctl dctcp', output)
+
     @expectedFailureIfModuleIsNotAvailable('vrf')
     def test_route_vrf(self):
         copy_network_unit('25-route-vrf.network', '12-dummy.netdev',