network: Add support to configure DHCPv4 route MTU

This is useful for transitioning systems from small frames to jumbo frames.

Closes #14302

man/systemd.network.xml

@@ -1618,6 +1618,13 @@
           </listitem>
         </varlistentry>
 
+        <varlistentry>
+          <term><varname>RouteMTUBytes=</varname></term>
+          <listitem>
+            <para>Specifies the MTU for the DHCP routes. Please see the [Route] section for further details.</para>
+          </listitem>
+        </varlistentry>
+
         <varlistentry>
           <term><varname>ListenPort=</varname></term>
           <listitem>
@@ -2450,6 +2457,14 @@
         </listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>ControlledDelayPacketLimit=</varname></term>
+        <listitem>
+          <para>Specifies the hard lmit on the queue size in number of packets. When this limit is reached, incoming packets are
+          dropped. An unsigned integer ranges 0 to 4294967294. Defaults to unset and kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
       <varlistentry>
         <term><varname>FairQueuingControlledDelayPacketLimit=</varname></term>
         <listitem>

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

@@ -745,6 +745,10 @@ static const NLTypeSystem rtnl_nexthop_type_system = {
        .types = rtnl_nexthop_types,
 };
 
+static const NLType rtnl_tca_option_data_codel_types[] = {
+        [TCA_CODEL_LIMIT]            = { .type = NETLINK_TYPE_U32 },
+};
+
 static const NLType rtnl_tca_option_data_fq_types[] = {
         [TCA_FQ_PLIMIT]             = { .type = NETLINK_TYPE_U32 },
         [TCA_FQ_FLOW_PLIMIT]        = { .type = NETLINK_TYPE_U32 },
@@ -783,6 +787,7 @@ static const NLType rtnl_tca_option_data_tbf_types[] = {
 };
 
 static const char* const nl_union_tca_option_data_table[] = {
+        [NL_UNION_TCA_OPTION_DATA_CODEL] = "codel",
         [NL_UNION_TCA_OPTION_DATA_FQ] = "fq",
         [NL_UNION_TCA_OPTION_DATA_FQ_CODEL] = "fq_codel",
         [NL_UNION_TCA_OPTION_DATA_TBF] = "tbf",
@@ -791,6 +796,8 @@ static const char* const nl_union_tca_option_data_table[] = {
 DEFINE_STRING_TABLE_LOOKUP(nl_union_tca_option_data, NLUnionTCAOptionData);
 
 static const NLTypeSystem rtnl_tca_option_data_type_systems[] = {
+        [NL_UNION_TCA_OPTION_DATA_CODEL] =       { .count = ELEMENTSOF(rtnl_tca_option_data_codel_types),
+                                                   .types = rtnl_tca_option_data_codel_types },
         [NL_UNION_TCA_OPTION_DATA_FQ] =          { .count = ELEMENTSOF(rtnl_tca_option_data_fq_types),
                                                    .types = rtnl_tca_option_data_fq_types },
         [NL_UNION_TCA_OPTION_DATA_FQ_CODEL] =    { .count = ELEMENTSOF(rtnl_tca_option_data_fq_codel_types),

src/libsystemd/sd-netlink/netlink-types.h

@@ -92,6 +92,7 @@ const char *nl_union_link_info_data_to_string(NLUnionLinkInfoData p) _const_;
 NLUnionLinkInfoData nl_union_link_info_data_from_string(const char *p) _pure_;
 
 typedef enum NLUnionTCAOptionData {
+        NL_UNION_TCA_OPTION_DATA_CODEL,
         NL_UNION_TCA_OPTION_DATA_FQ,
         NL_UNION_TCA_OPTION_DATA_FQ_CODEL,
         NL_UNION_TCA_OPTION_DATA_TBF,

src/network/meson.build

@@ -107,6 +107,8 @@ sources = files('''
         networkd-util.h
         networkd-wifi.c
         networkd-wifi.h
+        tc/codel.c
+        tc/codel.h
         tc/fq.c
         tc/fq.h
         tc/fq-codel.c

src/network/networkd-dhcp4.c

@@ -310,6 +310,7 @@ static int link_set_dhcp_routes(Link *link) {
                 assert_se(sd_dhcp_route_get_destination_prefix_length(static_routes[i], &route->dst_prefixlen) >= 0);
                 route->priority = link->network->dhcp_route_metric;
                 route->table = table;
+                route->mtu = link->network->dhcp_route_mtu;
                 route->scope = route_scope_from_address(route, &address);
                 if (IN_SET(route->scope, RT_SCOPE_LINK, RT_SCOPE_UNIVERSE))
                         route->prefsrc.in = address;
@@ -353,6 +354,7 @@ static int link_set_dhcp_routes(Link *link) {
                 route_gw->protocol = RTPROT_DHCP;
                 route_gw->priority = link->network->dhcp_route_metric;
                 route_gw->table = table;
+                route_gw->mtu = link->network->dhcp_route_mtu;
 
                 r = dhcp_route_configure(&route_gw, link);
                 if (r < 0)
@@ -368,6 +370,7 @@ static int link_set_dhcp_routes(Link *link) {
                 route->protocol = RTPROT_DHCP;
                 route->priority = link->network->dhcp_route_metric;
                 route->table = table;
+                route->mtu = link->network->dhcp_route_mtu;
 
                 r = dhcp_route_configure(&route, link);
                 if (r < 0)

src/network/networkd-network-gperf.gperf

@@ -178,6 +178,7 @@ DHCPv4.SendDecline,                     config_parse_bool,
 DHCPv4.BlackList,                       config_parse_dhcp_black_listed_ip_address,       0,                             0
 DHCPv4.IPServiceType,                   config_parse_ip_service_type,                    0,                             offsetof(Network, ip_service_type)
 DHCPv4.SendOption,                      config_parse_dhcp_send_option,                   0,                             offsetof(Network, dhcp_client_send_options)
+DHCPv4.RouteMTUBytes,                   config_parse_mtu,                                AF_INET,                       offsetof(Network, dhcp_route_mtu)
 DHCPv6.UseDNS,                          config_parse_bool,                               0,                             offsetof(Network, dhcp6_use_dns)
 DHCPv6.UseNTP,                          config_parse_bool,                               0,                             offsetof(Network, dhcp6_use_ntp)
 DHCPv6.RapidCommit,                     config_parse_bool,                               0,                             offsetof(Network, rapid_commit)
@@ -277,6 +278,7 @@ TrafficControlQueueingDiscipline.FairQueueTrafficPolicingBuckets,            con
 TrafficControlQueueingDiscipline.FairQueueTrafficPolicingOrphanMask,         config_parse_tc_fair_queue_traffic_policing_u32,             0, 0
 TrafficControlQueueingDiscipline.FairQueueTrafficPolicingPacing,             config_parse_tc_fair_queue_traffic_policing_bool,            0, 0
 TrafficControlQueueingDiscipline.FairQueueTrafficPolicingCEThresholdSec,     config_parse_tc_fair_queue_traffic_policing_usec,            0, 0
+TrafficControlQueueingDiscipline.ControlledDelayPacketLimit,                 config_parse_tc_controlled_delay_u32,                        0, 0
 /* backwards compatibility: do not add new entries to this section */
 Network.IPv4LL,                         config_parse_ipv4ll,                             0,                             offsetof(Network, link_local)
 DHCP.ClientIdentifier,                  config_parse_dhcp_client_identifier,             0,                             offsetof(Network, dhcp_client_identifier)

src/network/networkd-network.h

@@ -95,6 +95,7 @@ struct Network {
         uint64_t dhcp_max_attempts;
         unsigned dhcp_route_metric;
         uint32_t dhcp_route_table;
+        uint32_t dhcp_route_mtu;
         uint16_t dhcp_client_port;
         int dhcp_critical;
         int ip_service_type;

src/network/tc/codel.c

@@ -0,0 +1,95 @@
+/* SPDX-License-Identifier: LGPL-2.1+
+ * Copyright © 2019 VMware, Inc. */
+
+#include <linux/pkt_sched.h>
+
+#include "alloc-util.h"
+#include "conf-parser.h"
+#include "netlink-util.h"
+#include "parse-util.h"
+#include "qdisc.h"
+#include "string-util.h"
+
+static int controlled_delay_fill_message(Link *link, QDisc *qdisc, sd_netlink_message *req) {
+        ControlledDelay *cd;
+        int r;
+
+        assert(link);
+        assert(qdisc);
+        assert(req);
+
+        cd = CODEL(qdisc);
+
+        r = sd_netlink_message_open_container_union(req, TCA_OPTIONS, "codel");
+        if (r < 0)
+                return log_link_error_errno(link, r, "Could not open container TCA_OPTIONS: %m");
+
+        if (cd->packet_limit > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_CODEL_LIMIT, cd->packet_limit);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_CODEL_LIMIT attribute: %m");
+        }
+
+        r = sd_netlink_message_close_container(req);
+        if (r < 0)
+                return log_link_error_errno(link, r, "Could not close container TCA_OPTIONS: %m");
+
+        return 0;
+}
+
+int config_parse_tc_controlled_delay_u32(
+                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) {
+
+        _cleanup_(qdisc_free_or_set_invalidp) QDisc *qdisc = NULL;
+        ControlledDelay *cd;
+        Network *network = data;
+        int r;
+
+        assert(filename);
+        assert(lvalue);
+        assert(rvalue);
+        assert(data);
+
+        r = qdisc_new_static(QDISC_KIND_CODEL, network, filename, section_line, &qdisc);
+        if (r == -ENOMEM)
+                return log_oom();
+        if (r < 0)
+                return log_syntax(unit, LOG_ERR, filename, line, r,
+                                  "More than one kind of queueing discipline, ignoring assignment: %m");
+
+        cd = CODEL(qdisc);
+
+        if (isempty(rvalue)) {
+                cd->packet_limit = 0;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = safe_atou32(rvalue, &cd->packet_limit);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r,
+                           "Failed to parse '%s=', ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+
+        qdisc = NULL;
+
+        return 0;
+}
+
+const QDiscVTable codel_vtable = {
+        .object_size = sizeof(ControlledDelay),
+        .tca_kind = "codel",
+        .fill_message = controlled_delay_fill_message,
+};

src/network/tc/codel.h

@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: LGPL-2.1+
+ * Copyright © 2019 VMware, Inc. */
+#pragma once
+
+#include "conf-parser.h"
+#include "qdisc.h"
+#include "time-util.h"
+
+typedef struct ControlledDelay {
+        QDisc meta;
+
+        uint32_t packet_limit;
+
+} ControlledDelay;
+
+DEFINE_QDISC_CAST(CODEL, ControlledDelay);
+extern const QDiscVTable codel_vtable;
+
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_controlled_delay_u32);

src/network/tc/qdisc.c

@@ -14,6 +14,7 @@
 #include "string-util.h"
 
 const QDiscVTable * const qdisc_vtable[_QDISC_KIND_MAX] = {
+        [QDISC_KIND_CODEL] = &codel_vtable,
         [QDISC_KIND_FQ] = &fq_vtable,
         [QDISC_KIND_FQ_CODEL] = &fq_codel_vtable,
         [QDISC_KIND_NETEM] = &netem_vtable,

src/network/tc/qdisc.h

@@ -8,6 +8,7 @@
 #include "networkd-util.h"
 
 typedef enum QDiscKind {
+        QDISC_KIND_CODEL,
         QDISC_KIND_FQ,
         QDISC_KIND_FQ_CODEL,
         QDISC_KIND_NETEM,
@@ -64,6 +65,7 @@ DEFINE_NETWORK_SECTION_FUNCTIONS(QDisc, qdisc_free);
 
 CONFIG_PARSER_PROTOTYPE(config_parse_tc_qdiscs_parent);
 
+#include "codel.h"
 #include "fq-codel.h"
 #include "fq.h"
 #include "netem.h"

test/fuzz/fuzz-network-parser/directives.network

@@ -99,6 +99,7 @@ MaxAttempts=
 IPServiceType=
 SendOption=
 SendDecline=
+RouteMTUBytes=
 [DHCPv6]
 UseNTP=
 UseDNS=
@@ -297,3 +298,4 @@ FairQueueTrafficPolicingBuckets=
 FairQueueTrafficPolicingOrphanMask=
 FairQueueTrafficPolicingPacing=
 FairQueueTrafficPolicingCEThresholdSec=
+ControlledDelayPacketLimit=