From d474aa51bfd6437d74ebeda500f05f4149432109 Mon Sep 17 00:00:00 2001 From: Yu Watanabe Date: Tue, 9 Jun 2020 17:23:11 +0900 Subject: [PATCH] network: tc: introduce Enhanced Transmission Selection (ETS) Closes #15264. --- man/systemd.network.xml | 52 +++ src/network/meson.build | 2 + src/network/networkd-network-gperf.gperf | 6 + src/network/networkd-network.c | 1 + src/network/tc/ets.c | 338 ++++++++++++++++++ src/network/tc/ets.h | 25 ++ src/network/tc/qdisc.c | 1 + src/network/tc/qdisc.h | 2 + .../fuzz-network-parser/directives.network | 7 + 9 files changed, 434 insertions(+) create mode 100644 src/network/tc/ets.c create mode 100644 src/network/tc/ets.h diff --git a/man/systemd.network.xml b/man/systemd.network.xml index c0d8d979e6..002e96eb27 100644 --- a/man/systemd.network.xml +++ b/man/systemd.network.xml @@ -2929,6 +2929,58 @@ + + [EnhancedTransmissionSelection] Section Options + The [EnhancedTransmissionSelection] section manages the queueing discipline (qdisc) of + Enhanced Transmission Selection (ETS). + + + + + + + Bands= + + Specifies the number of bands. An unsigned integer ranges 1 to 16. This value has to be + at least large enough to cover the strict bands specified through the + StrictBands= and bandwidth-sharing bands specified in + QuantumBytes=. + + + + + StrictBands= + + Specifies the number of bands that should be created in strict mode. An unsigned integer + ranges 1 to 16. + + + + + QuantumBytes= + + Specifies the white-space separated list of quantum used in band-sharing bands. When + suffixed with K, M, or G, the specified size is parsed as Kilobytes, Megabytes, or Gigabytes, + respectively, to the base of 1024. This setting can be specified multiple times. If an empty + string is assigned, then the all previous assignments are cleared. + + + + + PriorityMap= + + The priority map maps the priority of a packet to a band. The argument is a white-space + separated list of numbers. The first number indicates which band the packets with priority + 0 should be put to, the second is for priority 1, and so on. There can be up to 16 numbers in + the list. If there are fewer, the default band that traffic with one of the unmentioned + priorities goes to is the last one. Each band number must be 0..255. This setting can be + specified multiple times. If an empty string is assigned, then the all previous assignments + are cleared. + + + + + [GenericRandomEarlyDetection] Section Options The [GenericRandomEarlyDetection] section manages the queueing discipline diff --git a/src/network/meson.build b/src/network/meson.build index 9d5cdfa516..97a164091a 100644 --- a/src/network/meson.build +++ b/src/network/meson.build @@ -115,6 +115,8 @@ sources = files(''' tc/codel.h tc/drr.c tc/drr.h + tc/ets.c + tc/ets.h tc/fifo.c tc/fifo.h tc/fq.c diff --git a/src/network/networkd-network-gperf.gperf b/src/network/networkd-network-gperf.gperf index fc756ef0fc..75ee8428f0 100644 --- a/src/network/networkd-network-gperf.gperf +++ b/src/network/networkd-network-gperf.gperf @@ -309,6 +309,12 @@ DeficitRoundRobinScheduler.Handle, config_parse_qdisc_handle, DeficitRoundRobinSchedulerClass.Parent, config_parse_tclass_parent, TCLASS_KIND_DRR, 0 DeficitRoundRobinSchedulerClass.ClassId, config_parse_tclass_classid, TCLASS_KIND_DRR, 0 DeficitRoundRobinSchedulerClass.Quantum, config_parse_drr_size, TCLASS_KIND_DRR, 0 +EnhancedTransmissionSelection.Parent, config_parse_qdisc_parent, QDISC_KIND_ETS, 0 +EnhancedTransmissionSelection.Handle, config_parse_qdisc_handle, QDISC_KIND_ETS, 0 +EnhancedTransmissionSelection.Bands, config_parse_ets_u8, QDISC_KIND_ETS, 0 +EnhancedTransmissionSelection.StrictBands, config_parse_ets_u8, QDISC_KIND_ETS, 0 +EnhancedTransmissionSelection.QuantumBytes, config_parse_ets_quanta, QDISC_KIND_ETS, 0 +EnhancedTransmissionSelection.PriorityMap, config_parse_ets_prio, QDISC_KIND_ETS, 0 PFIFO.Parent, config_parse_qdisc_parent, QDISC_KIND_PFIFO, 0 PFIFO.Handle, config_parse_qdisc_handle, QDISC_KIND_PFIFO, 0 PFIFO.PacketLimit, config_parse_pfifo_size, QDISC_KIND_PFIFO, 0 diff --git a/src/network/networkd-network.c b/src/network/networkd-network.c index f0fa5d0427..0e68b35e89 100644 --- a/src/network/networkd-network.c +++ b/src/network/networkd-network.c @@ -512,6 +512,7 @@ int network_load_one(Manager *manager, OrderedHashmap **networks, const char *fi "ControlledDelay\0" "DeficitRoundRobinScheduler\0" "DeficitRoundRobinSchedulerClass\0" + "EnhancedTransmissionSelection\0" "FairQueueing\0" "FairQueueingControlledDelay\0" "GenericRandomEarlyDetection\0" diff --git a/src/network/tc/ets.c b/src/network/tc/ets.c new file mode 100644 index 0000000000..c4fa5646dd --- /dev/null +++ b/src/network/tc/ets.c @@ -0,0 +1,338 @@ +/* SPDX-License-Identifier: LGPL-2.1+ */ + +#include + +#include "alloc-util.h" +#include "conf-parser.h" +#include "ets.h" +#include "memory-util.h" +#include "netlink-util.h" +#include "parse-util.h" +#include "qdisc.h" +#include "string-util.h" +#include "tc-util.h" + +static int enhanced_transmission_selection_fill_message(Link *link, QDisc *qdisc, sd_netlink_message *req) { + EnhancedTransmissionSelection *ets; + int r; + + assert(link); + assert(qdisc); + assert(req); + + ets = ETS(qdisc); + + r = sd_netlink_message_open_container_union(req, TCA_OPTIONS, "ets"); + if (r < 0) + return log_link_error_errno(link, r, "Could not open container TCA_OPTIONS: %m"); + + r = sd_netlink_message_append_u8(req, TCA_ETS_NBANDS, ets->n_bands); + if (r < 0) + return log_link_error_errno(link, r, "Could not append TCA_ETS_NBANDS attribute: %m"); + + if (ets->n_strict > 0) { + r = sd_netlink_message_append_u8(req, TCA_ETS_NSTRICT, ets->n_strict); + if (r < 0) + return log_link_error_errno(link, r, "Could not append TCA_ETS_NSTRICT attribute: %m"); + } + + if (ets->n_quanta > 0) { + r = sd_netlink_message_open_container(req, TCA_ETS_QUANTA); + if (r < 0) + return log_link_error_errno(link, r, "Could not open container TCA_ETS_QUANTA: %m"); + + for (unsigned i = 0; i < ets->n_quanta; i++) { + r = sd_netlink_message_append_u32(req, TCA_ETS_QUANTA_BAND, ets->quanta[i]); + if (r < 0) + return log_link_error_errno(link, r, "Could not append TCA_ETS_QUANTA_BAND attribute: %m"); + } + + r = sd_netlink_message_close_container(req); + if (r < 0) + return log_link_error_errno(link, r, "Could not close container TCA_ETS_QUANTA: %m"); + } + + if (ets->n_prio > 0) { + r = sd_netlink_message_open_container(req, TCA_ETS_PRIOMAP); + if (r < 0) + return log_link_error_errno(link, r, "Could not open container TCA_ETS_PRIOMAP: %m"); + + for (unsigned i = 0; i < ets->n_prio; i++) { + r = sd_netlink_message_append_u8(req, TCA_ETS_PRIOMAP_BAND, ets->prio[i]); + if (r < 0) + return log_link_error_errno(link, r, "Could not append TCA_ETS_PRIOMAP_BAND attribute: %m"); + } + + r = sd_netlink_message_close_container(req); + if (r < 0) + return log_link_error_errno(link, r, "Could not close container TCA_ETS_PRIOMAP: %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_ets_u8( + 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; + EnhancedTransmissionSelection *ets; + Network *network = data; + uint8_t v, *p; + int r; + + assert(filename); + assert(lvalue); + assert(rvalue); + assert(data); + + r = qdisc_new_static(QDISC_KIND_ETS, 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"); + + ets = ETS(qdisc); + if (streq(lvalue, "Bands")) + p = &ets->n_bands; + else if (streq(lvalue, "StrictBands")) + p = &ets->n_strict; + else + assert_not_reached("Invalid lvalue."); + + if (isempty(rvalue)) { + *p = 0; + + qdisc = NULL; + return 0; + } + + r = safe_atou8(rvalue, &v); + if (r < 0) { + log_syntax(unit, LOG_ERR, filename, line, r, + "Failed to parse '%s=', ignoring assignment: %s", + lvalue, rvalue); + return 0; + } + if (v > TCQ_ETS_MAX_BANDS) { + log_syntax(unit, LOG_ERR, filename, line, 0, + "Invalid '%s='. The value must be <= %d, ignoring assignment: %s", + lvalue, TCQ_ETS_MAX_BANDS, rvalue); + return 0; + } + + *p = v; + qdisc = NULL; + + return 0; +} + +int config_parse_ets_quanta( + 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; + EnhancedTransmissionSelection *ets; + Network *network = data; + int r; + + assert(filename); + assert(lvalue); + assert(rvalue); + assert(data); + + r = qdisc_new_static(QDISC_KIND_ETS, 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"); + + ets = ETS(qdisc); + + if (isempty(rvalue)) { + memzero(ets->quanta, sizeof(uint32_t) * TCQ_ETS_MAX_BANDS); + ets->n_quanta = 0; + + qdisc = NULL; + return 0; + } + + for (const char *p = rvalue;;) { + _cleanup_free_ char *word = NULL; + uint64_t v; + + r = extract_first_word(&p, &word, NULL, 0); + if (r == -ENOMEM) + return log_oom(); + if (r < 0) { + log_syntax(unit, LOG_ERR, filename, line, r, + "Failed to extract next value, ignoring: %m"); + continue; + } + if (r == 0) + break; + + r = parse_size(word, 1024, &v); + if (r < 0) { + log_syntax(unit, LOG_ERR, filename, line, r, + "Failed to parse '%s=', ignoring assignment: %s", + lvalue, word); + continue; + } + if (v == 0 || v > UINT32_MAX) { + log_syntax(unit, LOG_ERR, filename, line, 0, + "Invalid '%s=', ignoring assignment: %s", + lvalue, word); + continue; + } + if (ets->n_quanta >= TCQ_ETS_MAX_BANDS) { + log_syntax(unit, LOG_ERR, filename, line, 0, + "Too many quanta in '%s=', ignoring assignment: %s", + lvalue, word); + continue; + } + + ets->quanta[ets->n_quanta++] = v; + } + + qdisc = NULL; + + return 0; +} + +int config_parse_ets_prio( + 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; + EnhancedTransmissionSelection *ets; + Network *network = data; + int r; + + assert(filename); + assert(lvalue); + assert(rvalue); + assert(data); + + r = qdisc_new_static(QDISC_KIND_ETS, 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"); + + ets = ETS(qdisc); + + if (isempty(rvalue)) { + memzero(ets->prio, sizeof(uint8_t) * (TC_PRIO_MAX + 1)); + ets->n_prio = 0; + + qdisc = NULL; + return 0; + } + + for (const char *p = rvalue;;) { + _cleanup_free_ char *word = NULL; + uint8_t v; + + r = extract_first_word(&p, &word, NULL, 0); + if (r == -ENOMEM) + return log_oom(); + if (r < 0) { + log_syntax(unit, LOG_ERR, filename, line, r, + "Failed to extract next value, ignoring: %m"); + continue; + } + if (r == 0) + break; + + r = safe_atou8(word, &v); + if (r < 0) { + log_syntax(unit, LOG_ERR, filename, line, r, + "Failed to parse '%s=', ignoring assignment: %s", + lvalue, word); + continue; + } + if (ets->n_quanta > TC_PRIO_MAX) { + log_syntax(unit, LOG_ERR, filename, line, 0, + "Too many priomap in '%s=', ignoring assignment: %s", + lvalue, word); + continue; + } + + ets->prio[ets->n_prio++] = v; + } + + qdisc = NULL; + + return 0; +} + +static int enhanced_transmission_selection_verify(QDisc *qdisc) { + EnhancedTransmissionSelection *ets; + + assert(qdisc); + + ets = ETS(qdisc); + + if (ets->n_bands == 0) + ets->n_bands = ets->n_strict + ets->n_quanta; + + if (ets->n_bands == 0) + return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), + "%s: At least one of Band=, Strict=, or Quanta= must be specified. " + "Ignoring [EnhancedTransmissionSelection] section from line %u.", + qdisc->section->filename, qdisc->section->line); + + if (ets->n_bands < ets->n_strict + ets->n_quanta) + return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), + "%s: Not enough total bands to cover all the strict bands and quanta. " + "Ignoring [EnhancedTransmissionSelection] section from line %u.", + qdisc->section->filename, qdisc->section->line); + + for (unsigned i = 0; i < ets->n_prio; i++) + if (ets->prio[i] >= ets->n_bands) + return log_warning_errno(SYNTHETIC_ERRNO(EINVAL), + "%s: PriorityMap= element is out of bands. " + "Ignoring [EnhancedTransmissionSelection] section from line %u.", + qdisc->section->filename, qdisc->section->line); + + return 0; +} + +const QDiscVTable ets_vtable = { + .object_size = sizeof(EnhancedTransmissionSelection), + .tca_kind = "ets", + .fill_message = enhanced_transmission_selection_fill_message, + .verify = enhanced_transmission_selection_verify, +}; diff --git a/src/network/tc/ets.h b/src/network/tc/ets.h new file mode 100644 index 0000000000..c35d5976d8 --- /dev/null +++ b/src/network/tc/ets.h @@ -0,0 +1,25 @@ +/* SPDX-License-Identifier: LGPL-2.1+ */ +#pragma once + +#include + +#include "conf-parser.h" +#include "qdisc.h" + +typedef struct EnhancedTransmissionSelection { + QDisc meta; + + uint8_t n_bands; + uint8_t n_strict; + unsigned n_quanta; + uint32_t quanta[TCQ_ETS_MAX_BANDS]; + unsigned n_prio; + uint8_t prio[TC_PRIO_MAX + 1]; +} EnhancedTransmissionSelection; + +DEFINE_QDISC_CAST(ETS, EnhancedTransmissionSelection); +extern const QDiscVTable ets_vtable; + +CONFIG_PARSER_PROTOTYPE(config_parse_ets_u8); +CONFIG_PARSER_PROTOTYPE(config_parse_ets_quanta); +CONFIG_PARSER_PROTOTYPE(config_parse_ets_prio); diff --git a/src/network/tc/qdisc.c b/src/network/tc/qdisc.c index 6ba4325c9c..91cdd8917f 100644 --- a/src/network/tc/qdisc.c +++ b/src/network/tc/qdisc.c @@ -20,6 +20,7 @@ const QDiscVTable * const qdisc_vtable[_QDISC_KIND_MAX] = { [QDISC_KIND_CAKE] = &cake_vtable, [QDISC_KIND_CODEL] = &codel_vtable, [QDISC_KIND_DRR] = &drr_vtable, + [QDISC_KIND_ETS] = &ets_vtable, [QDISC_KIND_FQ] = &fq_vtable, [QDISC_KIND_FQ_CODEL] = &fq_codel_vtable, [QDISC_KIND_GRED] = &gred_vtable, diff --git a/src/network/tc/qdisc.h b/src/network/tc/qdisc.h index 802653efb0..0c9c0544b6 100644 --- a/src/network/tc/qdisc.h +++ b/src/network/tc/qdisc.h @@ -13,6 +13,7 @@ typedef enum QDiscKind { QDISC_KIND_CAKE, QDISC_KIND_CODEL, QDISC_KIND_DRR, + QDISC_KIND_ETS, QDISC_KIND_FQ, QDISC_KIND_FQ_CODEL, QDISC_KIND_GRED, @@ -87,6 +88,7 @@ CONFIG_PARSER_PROTOTYPE(config_parse_qdisc_handle); #include "cake.h" #include "codel.h" +#include "ets.h" #include "fifo.h" #include "fq-codel.h" #include "fq.h" diff --git a/test/fuzz/fuzz-network-parser/directives.network b/test/fuzz/fuzz-network-parser/directives.network index f97b1ba1b8..ade153a378 100644 --- a/test/fuzz/fuzz-network-parser/directives.network +++ b/test/fuzz/fuzz-network-parser/directives.network @@ -431,6 +431,13 @@ Handle= Parent= ClassId= Quantum= +[EnhancedTransmissionSelection] +Parent= +Handle= +Bands= +StrictBands= +QuantumBytes= +PriorityMap= [HeavyHitterFilter] Parent= Handle=