Merge pull request #14337 from yuwata/network-tc-fq-more

network: tc: introduce more FQ settings

man/systemd.network.xml

@@ -2515,6 +2515,76 @@
         </listitem>
       </varlistentry>
 
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingFlowLimit=</varname></term>
+        <listitem>
+          <para>Specifies the hard limit on the maximum number of packets queued per flow. Defaults to
+          unset and kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingQuantum=</varname></term>
+        <listitem>
+          <para>Specifies the credit per dequeue RR round, i.e. the amount of bytes a flow is allowed
+          to dequeue at once. When suffixed with K, M, or G, the specified size is parsed as Kilobytes,
+          Megabytes, or Gigabytes, respectively, to the base of 1024. Defaults to unset and kernel's
+          default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingInitialQuantum=</varname></term>
+        <listitem>
+          <para>Specifies the initial sending rate credit, i.e. the amount of bytes a new flow is
+          allowed to dequeue initially. When suffixed with K, M, or G, the specified size is parsed as
+          Kilobytes, Megabytes, or Gigabytes, respectively, to the base of 1024. Defaults to unset and
+          kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingMaximumRate=</varname></term>
+        <listitem>
+          <para>Specifies the maximum sending rate of a flow. When suffixed with K, M, or G, the
+          specified size is parsed as Kilobytes, Megabytes, or Gigabytes, respectively, to the base of
+          1000. Defaults to unset and kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingBuckets=</varname></term>
+        <listitem>
+          <para>Specifies the size of the hash table used for flow lookups. Defaults to unset and
+          kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingOrphanMask=</varname></term>
+        <listitem>
+          <para>Takes an unsigned integer. For packets not owned by a socket, fq is able to mask a part
+          of hash and reduce number of buckets associated with the traffic. Defaults to unset and
+          kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingPacing=</varname></term>
+        <listitem>
+          <para>Takes a boolean, and enables or disables flow pacing. Defaults to unset and kernel's
+          default is used.</para>
+        </listitem>
+      </varlistentry>
+
+      <varlistentry>
+        <term><varname>FairQueueTrafficPolicingCEThresholdSec=</varname></term>
+        <listitem>
+          <para>Takes a timespan. This sets a threshold above which all packets are marked with ECN
+          Congestion Experienced (CE). Defaults to unset and kernel's default is used.</para>
+        </listitem>
+      </varlistentry>
+
     </variablelist>
   </refsect1>
 

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

@@ -756,6 +756,7 @@ static const NLType rtnl_tca_option_data_fq_types[] = {
         [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 },
+        [TCA_FQ_ORPHAN_MASK]        = { .type = NETLINK_TYPE_U32 },
 };
 
 static const NLType rtnl_tca_option_data_fq_codel_types[] = {

src/network/networkd-network-gperf.gperf

@@ -266,7 +266,15 @@ TrafficControlQueueingDiscipline.FairQueuingControlledDelayTargetSec,        con
 TrafficControlQueueingDiscipline.FairQueuingControlledDelayIntervalSec,      config_parse_tc_fair_queuing_controlled_delay_usec,          0, 0
 TrafficControlQueueingDiscipline.FairQueuingControlledDelayCEThresholdSec,   config_parse_tc_fair_queuing_controlled_delay_usec,          0, 0
 TrafficControlQueueingDiscipline.FairQueuingControlledDelayECN,              config_parse_tc_fair_queuing_controlled_delay_bool,          0, 0
-TrafficControlQueueingDiscipline.FairQueueTrafficPolicingPacketLimit,        config_parse_tc_fair_queue_traffic_policing_packet_limit,    0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingPacketLimit,        config_parse_tc_fair_queue_traffic_policing_u32,             0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingFlowLimit,          config_parse_tc_fair_queue_traffic_policing_u32,             0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingQuantum,            config_parse_tc_fair_queue_traffic_policing_size,            0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingInitialQuantum,     config_parse_tc_fair_queue_traffic_policing_size,            0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingMaximumRate,        config_parse_tc_fair_queue_traffic_policing_max_rate,        0, 0
+TrafficControlQueueingDiscipline.FairQueueTrafficPolicingBuckets,            config_parse_tc_fair_queue_traffic_policing_u32,             0, 0
+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
 /* 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/tc/fq.c

@@ -9,6 +9,20 @@
 #include "netlink-util.h"
 #include "parse-util.h"
 #include "string-util.h"
+#include "util.h"
+
+static int fair_queue_traffic_policing_init(QDisc *qdisc) {
+        FairQueueTrafficPolicing *fq;
+
+        assert(qdisc);
+
+        fq = FQ(qdisc);
+
+        fq->pacing = -1;
+        fq->ce_threshold_usec = USEC_INFINITY;
+
+        return 0;
+}
 
 static int fair_queue_traffic_policing_fill_message(Link *link, QDisc *qdisc, sd_netlink_message *req) {
         FairQueueTrafficPolicing *fq;
@@ -24,9 +38,62 @@ static int fair_queue_traffic_policing_fill_message(Link *link, QDisc *qdisc, sd
         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");
+        if (fq->packet_limit > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_PLIMIT, fq->packet_limit);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_PLIMIT attribute: %m");
+        }
+
+        if (fq->flow_limit > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_FLOW_PLIMIT, fq->flow_limit);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_FLOW_PLIMIT attribute: %m");
+        }
+
+        if (fq->quantum > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_QUANTUM, fq->quantum);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_QUANTUM attribute: %m");
+        }
+
+        if (fq->initial_quantum > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_INITIAL_QUANTUM, fq->initial_quantum);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_INITIAL_QUANTUM attribute: %m");
+        }
+
+        if (fq->pacing >= 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_RATE_ENABLE, fq->pacing);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_RATE_ENABLE attribute: %m");
+        }
+
+        if (fq->max_rate > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_FLOW_MAX_RATE, fq->max_rate);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_FLOW_MAX_RATE attribute: %m");
+        }
+
+        if (fq->buckets > 0) {
+                uint32_t l;
+
+                l = log2u(fq->buckets);
+                r = sd_netlink_message_append_u32(req, TCA_FQ_BUCKETS_LOG, l);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_BUCKETS_LOG attribute: %m");
+        }
+
+        if (fq->orphan_mask > 0) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_ORPHAN_MASK, fq->orphan_mask);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_ORPHAN_MASK attribute: %m");
+        }
+
+        if (fq->ce_threshold_usec != USEC_INFINITY) {
+                r = sd_netlink_message_append_u32(req, TCA_FQ_CE_THRESHOLD, fq->ce_threshold_usec);
+                if (r < 0)
+                        return log_link_error_errno(link, r, "Could not append TCA_FQ_CE_THRESHOLD attribute: %m");
+        }
 
         r = sd_netlink_message_close_container(req);
         if (r < 0)
@@ -35,7 +102,137 @@ static int fair_queue_traffic_policing_fill_message(Link *link, QDisc *qdisc, sd
         return 0;
 }
 
-int config_parse_tc_fair_queue_traffic_policing_packet_limit(
+int config_parse_tc_fair_queue_traffic_policing_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;
+        FairQueueTrafficPolicing *fq;
+        Network *network = data;
+        uint32_t *p;
+        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 (streq(lvalue, "FairQueueTrafficPolicingPacketLimit"))
+                p = &fq->packet_limit;
+        else if (streq(lvalue, "FairQueueTrafficPolicingFlowLimit"))
+                p = &fq->flow_limit;
+        else if (streq(lvalue, "FairQueueTrafficPolicingBuckets"))
+                p = &fq->buckets;
+        else if (streq(lvalue, "FairQueueTrafficPolicingOrphanMask"))
+                p = &fq->orphan_mask;
+        else
+                assert_not_reached("Invalid lvalue");
+
+        if (isempty(rvalue)) {
+                *p = 0;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = safe_atou32(rvalue, p);
+        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;
+}
+
+int config_parse_tc_fair_queue_traffic_policing_size(
+                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;
+        uint64_t sz;
+        uint32_t *p;
+        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 (streq(lvalue, "FairQueueTrafficPolicingQuantum"))
+                p = &fq->quantum;
+        else if (streq(lvalue, "FairQueueTrafficPolicingInitialQuantum"))
+                p = &fq->initial_quantum;
+        else
+                assert_not_reached("Invalid lvalue");
+
+        if (isempty(rvalue)) {
+                *p = 0;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = parse_size(rvalue, 1024, &sz);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r,
+                           "Failed to parse '%s=', ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+        if (sz > UINT32_MAX) {
+                log_syntax(unit, LOG_ERR, filename, line, r,
+                           "Specified '%s=' is too large, ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+
+        *p = sz;
+        qdisc = NULL;
+
+        return 0;
+}
+
+int config_parse_tc_fair_queue_traffic_policing_bool(
                 const char *unit,
                 const char *filename,
                 unsigned line,
@@ -67,13 +264,13 @@ int config_parse_tc_fair_queue_traffic_policing_packet_limit(
         fq = FQ(qdisc);
 
         if (isempty(rvalue)) {
-                fq->limit = 0;
+                fq->pacing = -1;
 
                 qdisc = NULL;
                 return 0;
         }
 
-        r = safe_atou32(rvalue, &fq->limit);
+        r = parse_boolean(rvalue);
         if (r < 0) {
                 log_syntax(unit, LOG_ERR, filename, line, r,
                            "Failed to parse '%s=', ignoring assignment: %s",
@@ -81,12 +278,132 @@ int config_parse_tc_fair_queue_traffic_policing_packet_limit(
                 return 0;
         }
 
+        fq->pacing = r;
+        qdisc = NULL;
+
+        return 0;
+}
+
+int config_parse_tc_fair_queue_traffic_policing_usec(
+                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;
+        usec_t sec;
+        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->ce_threshold_usec = USEC_INFINITY;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = parse_sec(rvalue, &sec);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r,
+                           "Failed to parse '%s=', ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+        if (sec > UINT32_MAX) {
+                log_syntax(unit, LOG_ERR, filename, line, 0,
+                           "Specified '%s=' is too large, ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+
+        fq->ce_threshold_usec = sec;
+        qdisc = NULL;
+
+        return 0;
+}
+
+int config_parse_tc_fair_queue_traffic_policing_max_rate(
+                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;
+        uint64_t sz;
+        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->max_rate = 0;
+
+                qdisc = NULL;
+                return 0;
+        }
+
+        r = parse_size(rvalue, 1000, &sz);
+        if (r < 0) {
+                log_syntax(unit, LOG_ERR, filename, line, r,
+                           "Failed to parse '%s=', ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+        if (sz / 8 > UINT32_MAX) {
+                log_syntax(unit, LOG_ERR, filename, line, 0,
+                           "Specified '%s=' is too large, ignoring assignment: %s",
+                           lvalue, rvalue);
+                return 0;
+        }
+
+        fq->max_rate = sz / 8;
         qdisc = NULL;
 
         return 0;
 }
 
 const QDiscVTable fq_vtable = {
+        .init = fair_queue_traffic_policing_init,
         .object_size = sizeof(FairQueueTrafficPolicing),
         .tca_kind = "fq",
         .fill_message = fair_queue_traffic_policing_fill_message,

src/network/tc/fq.h

@@ -8,10 +8,22 @@
 typedef struct FairQueueTrafficPolicing {
         QDisc meta;
 
-        uint32_t limit;
+        uint32_t packet_limit;
+        uint32_t flow_limit;
+        uint32_t quantum;
+        uint32_t initial_quantum;
+        uint32_t max_rate;
+        uint32_t buckets;
+        uint32_t orphan_mask;
+        int pacing;
+        usec_t ce_threshold_usec;
 } FairQueueTrafficPolicing;
 
 DEFINE_QDISC_CAST(FQ, FairQueueTrafficPolicing);
 extern const QDiscVTable fq_vtable;
 
-CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_packet_limit);
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_u32);
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_size);
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_bool);
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_usec);
+CONFIG_PARSER_PROTOTYPE(config_parse_tc_fair_queue_traffic_policing_max_rate);

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

@@ -287,3 +287,11 @@ FairQueuingControlledDelayIntervalSec=
 FairQueuingControlledDelayCEThresholdSec=
 FairQueuingControlledDelayECN=
 FairQueueTrafficPolicingPacketLimit=
+FairQueueTrafficPolicingFlowLimit=
+FairQueueTrafficPolicingQuantum=
+FairQueueTrafficPolicingInitialQuantum=
+FairQueueTrafficPolicingMaximumRate=
+FairQueueTrafficPolicingBuckets=
+FairQueueTrafficPolicingOrphanMask=
+FairQueueTrafficPolicingPacing=
+FairQueueTrafficPolicingCEThresholdSec=

test/test-network/conf/25-qdisc-fq.network

@@ -8,3 +8,11 @@ Address=10.1.2.3/16
 [TrafficControlQueueingDiscipline]
 Parent=root
 FairQueueTrafficPolicingPacketLimit=1000
+FairQueueTrafficPolicingFlowLimit=200
+FairQueueTrafficPolicingQuantum=1500
+FairQueueTrafficPolicingInitialQuantum=13000
+FairQueueTrafficPolicingMaximumRate=1M
+FairQueueTrafficPolicingBuckets=512
+FairQueueTrafficPolicingOrphanMask=511
+FairQueueTrafficPolicingPacing=yes
+FairQueueTrafficPolicingCEThresholdSec=100ms

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

@@ -2165,7 +2165,7 @@ class NetworkdNetworkTests(unittest.TestCase, Utilities):
         output = check_output('tc qdisc show dev dummy98')
         print(output)
         self.assertRegex(output, 'qdisc fq')
-        self.assertRegex(output, 'limit 1000p')
+        self.assertRegex(output, 'limit 1000p flow_limit 200p buckets 512 orphan_mask 511 quantum 1500 initial_quantum 13000 maxrate 1Mbit')
 
 class NetworkdStateFileTests(unittest.TestCase, Utilities):
     links = [