Merge pull request #14305 from ssahani/tc-fq

network tc: inroduce FQ - Fair Queue traffic policing

12 files changed, 177 insertions, 12 deletions

diff --git a/man/systemd.network.xml b/man/systemd.network.xml
index acafa8a712..8509a8485a 100644
--- a/man/systemd.network.xml
+++ b/man/systemd.network.xml
@@ -2450,6 +2450,14 @@
         </listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingPacketLimit=</varname></term>
+        <listitem>
+          <para>Specifies the hard limit on the real queue size. When this limit is reached, incoming packets are
+          dropped. Defaults to unset and kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
     </variablelist>
   </refsect1>
 
diff --git a/src/libsystemd/sd-netlink/netlink-types.c b/src/libsystemd/sd-netlink/netlink-types.c
index cadaa750af..636e00de4c 100644
--- a/src/libsystemd/sd-netlink/netlink-types.c
+++ b/src/libsystemd/sd-netlink/netlink-types.c
@@ -734,14 +734,18 @@ static const NLTypeSystem rtnl_nexthop_type_system = {
        .types = rtnl_nexthop_types,
 };
 
-static const NLType rtnl_tca_option_data_tbf_types[] = {
-        [TCA_TBF_PARMS]   = { .size = sizeof(struct tc_tbf_qopt) },
-        [TCA_TBF_RTAB]    = { .size = TC_RTAB_SIZE },
-        [TCA_TBF_PTAB]    = { .size = TC_RTAB_SIZE },
-        [TCA_TBF_RATE64]  = { .type = NETLINK_TYPE_U64 },
-        [TCA_TBF_PRATE64] = { .type = NETLINK_TYPE_U64 },
-        [TCA_TBF_BURST]   = { .type = NETLINK_TYPE_U32 },
-        [TCA_TBF_PBURST]  = { .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 },
+        [TCA_FQ_QUANTUM]            = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_INITIAL_QUANTUM]    = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_RATE_ENABLE]        = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_FLOW_DEFAULT_RATE]  = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_FLOW_MAX_RATE]      = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_BUCKETS_LOG]        = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_FLOW_REFILL_DELAY]  = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_LOW_RATE_THRESHOLD] = { .type = NETLINK_TYPE_U32 },
+        [TCA_FQ_CE_THRESHOLD]       = { .type = NETLINK_TYPE_U32 },
 };
 
 static const NLType rtnl_tca_option_data_fq_codel_types[] = {
@@ -756,18 +760,31 @@ static const NLType rtnl_tca_option_data_fq_codel_types[] = {
         [TCA_FQ_CODEL_MEMORY_LIMIT]    = { .type = NETLINK_TYPE_U32 },
 };
 
+static const NLType rtnl_tca_option_data_tbf_types[] = {
+        [TCA_TBF_PARMS]   = { .size = sizeof(struct tc_tbf_qopt) },
+        [TCA_TBF_RTAB]    = { .size = TC_RTAB_SIZE },
+        [TCA_TBF_PTAB]    = { .size = TC_RTAB_SIZE },
+        [TCA_TBF_RATE64]  = { .type = NETLINK_TYPE_U64 },
+        [TCA_TBF_PRATE64] = { .type = NETLINK_TYPE_U64 },
+        [TCA_TBF_BURST]   = { .type = NETLINK_TYPE_U32 },
+        [TCA_TBF_PBURST]  = { .type = NETLINK_TYPE_U32 },
+};
+
 static const char* const nl_union_tca_option_data_table[] = {
-        [NL_UNION_TCA_OPTION_DATA_TBF] = "tbf",
+        [NL_UNION_TCA_OPTION_DATA_FQ] = "fq",
         [NL_UNION_TCA_OPTION_DATA_FQ_CODEL] = "fq_codel",
+        [NL_UNION_TCA_OPTION_DATA_TBF] = "tbf",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(nl_union_tca_option_data, NLUnionTCAOptionData);
 
 static const NLTypeSystem rtnl_tca_option_data_type_systems[] = {
-        [NL_UNION_TCA_OPTION_DATA_TBF] =         { .count = ELEMENTSOF(rtnl_tca_option_data_tbf_types),
-                                                   .types = rtnl_tca_option_data_tbf_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),
                                                    .types = rtnl_tca_option_data_fq_codel_types },
+        [NL_UNION_TCA_OPTION_DATA_TBF] =         { .count = ELEMENTSOF(rtnl_tca_option_data_tbf_types),
+                                                   .types = rtnl_tca_option_data_tbf_types },
 };
 
 static const NLTypeSystemUnion rtnl_tca_option_data_type_system_union = {
diff --git a/src/libsystemd/sd-netlink/netlink-types.h b/src/libsystemd/sd-netlink/netlink-types.h
index 2702409147..effc8f4f70 100644
--- a/src/libsystemd/sd-netlink/netlink-types.h
+++ b/src/libsystemd/sd-netlink/netlink-types.h
@@ -92,8 +92,9 @@ 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_TBF,
+        NL_UNION_TCA_OPTION_DATA_FQ,
         NL_UNION_TCA_OPTION_DATA_FQ_CODEL,
+        NL_UNION_TCA_OPTION_DATA_TBF,
         _NL_UNION_TCA_OPTION_DATA_MAX,
         _NL_UNION_TCA_OPTION_DATA_INVALID = -1,
 } NLUnionTCAOptionData;
diff --git a/src/network/meson.build b/src/network/meson.build
index 41e482e7a1..5f0e0890de 100644
--- a/src/network/meson.build
+++ b/src/network/meson.build
@@ -107,6 +107,8 @@ sources = files('''
         networkd-util.h
         networkd-wifi.c
         networkd-wifi.h
+        tc/fq.c
+        tc/fq.h
         tc/fq-codel.c
         tc/fq-codel.h
         tc/netem.c
diff --git a/src/network/networkd-network-gperf.gperf b/src/network/networkd-network-gperf.gperf
index e80f0afd8f..5a1e0c0dad 100644
--- a/src/network/networkd-network-gperf.gperf
+++ b/src/network/networkd-network-gperf.gperf
@@ -259,6 +259,7 @@ TrafficControlQueueingDiscipline.TokenBufferFilterPeakRate,                  con
 TrafficControlQueueingDiscipline.TokenBufferFilterLatencySec,                config_parse_tc_token_buffer_filter_latency,                 0, 0
 TrafficControlQueueingDiscipline.StochasticFairnessQueueingPerturbPeriodSec, config_parse_tc_stochastic_fairness_queueing_perturb_period, 0, 0
 TrafficControlQueueingDiscipline.FairQueuingControlledDelayPacketLimit,      config_parse_tc_fair_queuing_controlled_delay_limit,         0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingPacketLimit,        config_parse_tc_fair_queue_traffic_policing_packet_limit,    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)
diff --git a/src/network/tc/fq.c b/src/network/tc/fq.c
new file mode 100644
index 0000000000..d9142c747b
--- /dev/null
+++ b/src/network/tc/fq.c
@@ -0,0 +1,93 @@
+/* SPDX-License-Identifier: LGPL-2.1+
+ * Copyright © 2019 VMware, Inc. */
+
+#include <linux/pkt_sched.h>
+
+#include "alloc-util.h"
+#include "conf-parser.h"
+#include "fq.h"
+#include "netlink-util.h"
+#include "parse-util.h"
+#include "string-util.h"
+
+static int fair_queue_traffic_policing_fill_message(Link *link, QDisc *qdisc, sd_netlink_message *req) {
+        FairQueueTrafficPolicing *fq;
+        int r;
+
+        assert(link);
+        assert(qdisc);
+        assert(req);
+
+        fq = FQ(qdisc);
+
+        r = sd_netlink_message_open_container_union(req, TCA_OPTIONS, "fq");
+        if (r < 0)
+                return log_link_error_errno(link, r, "Could not open container TCA_OPTIONS: %m");
+
+        r = sd_netlink_message_append_u32(req, TCA_FQ_PLIMIT, fq->limit);
+        if (r < 0)
+                return log_link_error_errno(link, r, "Could not append TCA_FQ_PLIMIT 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_fair_queue_traffic_policing_packet_limit(
+                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;
+        FairQueueTrafficPolicing *fq;
+        Network *network = data;
+        int r;
+
+        assert(filename);
+        assert(lvalue);
+        assert(rvalue);
+        assert(data);
+
+        r = qdisc_new_static(QDISC_KIND_FQ, 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");
+
+        fq = FQ(qdisc);
+
+        if (isempty(rvalue)) {
+                fq->limit = 0;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = safe_atou32(rvalue, &fq->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 fq_vtable = {
+        .object_size = sizeof(FairQueueTrafficPolicing),
+        .tca_kind = "fq",
+        .fill_message = fair_queue_traffic_policing_fill_message,
+};
diff --git a/src/network/tc/fq.h b/src/network/tc/fq.h
new file mode 100644
index 0000000000..ac6c29c409
--- /dev/null
+++ b/src/network/tc/fq.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: LGPL-2.1+
+ * Copyright © 2019 VMware, Inc. */
+#pragma once
+
+#include "conf-parser.h"
+#include "qdisc.h"
+
+typedef struct FairQueueTrafficPolicing {
+        QDisc meta;
+
+        uint32_t limit;
+} FairQueueTrafficPolicing;
+
+DEFINE_QDISC_CAST(FQ, FairQueueTrafficPolicing);
+extern const QDiscVTable fq_vtable;
+
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_packet_limit);
diff --git a/src/network/tc/qdisc.c b/src/network/tc/qdisc.c
index ee5aafe5e9..118754a91c 100644
--- a/src/network/tc/qdisc.c
+++ b/src/network/tc/qdisc.c
@@ -14,6 +14,7 @@
 #include "string-util.h"
 
 const QDiscVTable * const qdisc_vtable[_QDISC_KIND_MAX] = {
+        [QDISC_KIND_FQ] = &fq_vtable,
         [QDISC_KIND_FQ_CODEL] = &fq_codel_vtable,
         [QDISC_KIND_NETEM] = &netem_vtable,
         [QDISC_KIND_SFQ] = &sfq_vtable,
diff --git a/src/network/tc/qdisc.h b/src/network/tc/qdisc.h
index 8e0ae28855..cef540a4d5 100644
--- a/src/network/tc/qdisc.h
+++ b/src/network/tc/qdisc.h
@@ -8,6 +8,7 @@
 #include "networkd-util.h"
 
 typedef enum QDiscKind {
+        QDISC_KIND_FQ,
         QDISC_KIND_FQ_CODEL,
         QDISC_KIND_NETEM,
         QDISC_KIND_SFQ,
@@ -62,6 +63,7 @@ DEFINE_NETWORK_SECTION_FUNCTIONS(QDisc, qdisc_free);
 CONFIG_PARSER_PROTOTYPE(config_parse_tc_qdiscs_parent);
 
 #include "fq-codel.h"
+#include "fq.h"
 #include "netem.h"
 #include "sfq.h"
 #include "tbf.h"
diff --git a/test/fuzz/fuzz-network-parser/directives.network b/test/fuzz/fuzz-network-parser/directives.network
index c71fe5ce77..9d6e37484c 100644
--- a/test/fuzz/fuzz-network-parser/directives.network
+++ b/test/fuzz/fuzz-network-parser/directives.network
@@ -279,3 +279,4 @@ TokenBufferFilterPeakRate=
 TokenBufferFilterLatencySec=
 StochasticFairnessQueueingPerturbPeriodSec=
 FairQueuingControlledDelayPacketLimit=
+FairQueueTrafficPolicingPacketLimit=
diff --git a/test/test-network/conf/25-qdisc-fq.network b/test/test-network/conf/25-qdisc-fq.network
new file mode 100644
index 0000000000..10e0e5b859
--- /dev/null
+++ b/test/test-network/conf/25-qdisc-fq.network
@@ -0,0 +1,10 @@
+[Match]
+Name=dummy98
+
+[Network]
+IPv6AcceptRA=no
+Address=10.1.2.3/16
+
+[TrafficControlQueueingDiscipline]
+Parent=root
+FairQueueTrafficPolicingPacketLimit=1000
diff --git a/test/test-network/systemd-networkd-tests.py b/test/test-network/systemd-networkd-tests.py
index a72b9abf76..31ffba9dda 100755
--- a/test/test-network/systemd-networkd-tests.py
+++ b/test/test-network/systemd-networkd-tests.py
@@ -1509,6 +1509,7 @@ class NetworkdNetworkTests(unittest.TestCase, Utilities):
         '25-neighbor-ip-dummy.network',
         '25-neighbor-ip.network',
         '25-nexthop.network',
+        '25-qdisc-fq.network',
         '25-qdisc-netem-and-fqcodel.network',
         '25-qdisc-tbf-and-sfq.network',
         '25-route-ipv6-src.network',
@@ -2102,6 +2103,17 @@ class NetworkdNetworkTests(unittest.TestCase, Utilities):
         self.assertRegex(output, 'qdisc sfq')
         self.assertRegex(output, 'perturb 5sec')
 
+    def test_qdisc2(self):
+        copy_unit_to_networkd_unit_path('25-qdisc-fq.network', '12-dummy.netdev')
+        start_networkd()
+
+        self.wait_online(['dummy98:routable'])
+
+        output = check_output('tc qdisc show dev dummy98')
+        print(output)
+        self.assertRegex(output, 'qdisc fq')
+        self.assertRegex(output, 'limit 1000p')
+
 class NetworkdStateFileTests(unittest.TestCase, Utilities):
     links = [
         'dummy98',