Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2019-present Open Networking Foundation |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
| 16 | package org.onosproject.k8snetworking.impl; |
| 17 | |
| 18 | import org.onlab.packet.ARP; |
| 19 | import org.onlab.packet.Ethernet; |
| 20 | import org.onlab.packet.Ip4Address; |
| 21 | import org.onlab.packet.IpPrefix; |
Jian Li | b121844 | 2020-09-03 13:12:14 +0900 | [diff] [blame^] | 22 | import org.onlab.packet.MacAddress; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 23 | import org.onlab.packet.TpPort; |
| 24 | import org.onosproject.cluster.ClusterService; |
| 25 | import org.onosproject.cluster.LeadershipService; |
| 26 | import org.onosproject.cluster.NodeId; |
| 27 | import org.onosproject.core.ApplicationId; |
| 28 | import org.onosproject.core.CoreService; |
| 29 | import org.onosproject.k8snetworking.api.K8sFlowRuleService; |
| 30 | import org.onosproject.k8snetworking.api.K8sNetwork; |
| 31 | import org.onosproject.k8snetworking.api.K8sNetworkEvent; |
| 32 | import org.onosproject.k8snetworking.api.K8sNetworkListener; |
| 33 | import org.onosproject.k8snetworking.api.K8sNetworkService; |
| 34 | import org.onosproject.k8snetworking.api.K8sPort; |
| 35 | import org.onosproject.k8snetworking.util.RulePopulatorUtil; |
| 36 | import org.onosproject.k8snode.api.K8sNode; |
| 37 | import org.onosproject.k8snode.api.K8sNodeEvent; |
| 38 | import org.onosproject.k8snode.api.K8sNodeListener; |
| 39 | import org.onosproject.k8snode.api.K8sNodeService; |
| 40 | import org.onosproject.mastership.MastershipService; |
| 41 | import org.onosproject.net.Device; |
| 42 | import org.onosproject.net.DeviceId; |
| 43 | import org.onosproject.net.PortNumber; |
| 44 | import org.onosproject.net.device.DeviceService; |
| 45 | import org.onosproject.net.driver.DriverService; |
| 46 | import org.onosproject.net.flow.DefaultTrafficSelector; |
| 47 | import org.onosproject.net.flow.DefaultTrafficTreatment; |
| 48 | import org.onosproject.net.flow.TrafficSelector; |
| 49 | import org.onosproject.net.flow.TrafficTreatment; |
| 50 | import org.onosproject.net.flow.instructions.ExtensionTreatment; |
| 51 | import org.osgi.service.component.annotations.Activate; |
| 52 | import org.osgi.service.component.annotations.Component; |
| 53 | import org.osgi.service.component.annotations.Deactivate; |
| 54 | import org.osgi.service.component.annotations.Reference; |
| 55 | import org.osgi.service.component.annotations.ReferenceCardinality; |
| 56 | import org.slf4j.Logger; |
| 57 | |
| 58 | import java.util.Objects; |
| 59 | import java.util.concurrent.ExecutorService; |
| 60 | |
| 61 | import static java.util.concurrent.Executors.newSingleThreadExecutor; |
| 62 | import static org.onlab.util.Tools.groupedThreads; |
| 63 | import static org.onosproject.k8snetworking.api.Constants.DEFAULT_GATEWAY_MAC; |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 64 | import static org.onosproject.k8snetworking.api.Constants.EXT_ENTRY_TABLE; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 65 | import static org.onosproject.k8snetworking.api.Constants.K8S_NETWORKING_APP_ID; |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 66 | import static org.onosproject.k8snetworking.api.Constants.POD_RESOLUTION_TABLE; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 67 | import static org.onosproject.k8snetworking.api.Constants.PRIORITY_EXTERNAL_ROUTING_RULE; |
| 68 | import static org.onosproject.k8snetworking.api.Constants.PRIORITY_STATEFUL_SNAT_RULE; |
| 69 | import static org.onosproject.k8snetworking.api.Constants.ROUTING_TABLE; |
| 70 | import static org.onosproject.k8snetworking.util.RulePopulatorUtil.CT_NAT_SRC_FLAG; |
| 71 | import static org.onosproject.k8snetworking.util.RulePopulatorUtil.buildMoveArpShaToThaExtension; |
| 72 | import static org.onosproject.k8snetworking.util.RulePopulatorUtil.buildMoveArpSpaToTpaExtension; |
| 73 | import static org.onosproject.k8snetworking.util.RulePopulatorUtil.buildMoveEthSrcToDstExtension; |
Jian Li | b121844 | 2020-09-03 13:12:14 +0900 | [diff] [blame^] | 74 | import static org.onosproject.k8snode.api.Constants.DEFAULT_EXTERNAL_GATEWAY_MAC; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 75 | import static org.slf4j.LoggerFactory.getLogger; |
| 76 | |
| 77 | /** |
| 78 | * Provides POD's internal to external connectivity using source NAT (SNAT). |
| 79 | */ |
| 80 | @Component(immediate = true) |
| 81 | public class K8sRoutingSnatHandler { |
| 82 | |
| 83 | private final Logger log = getLogger(getClass()); |
| 84 | |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 85 | private static final int HOST_PREFIX = 32; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 86 | |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 87 | // we try to avoid port number overlapping with node port (30000 ~ 32767) |
| 88 | // in case the user has customized node port range, the following static |
| 89 | // value should be changed accordingly |
| 90 | private static final int TP_PORT_MINIMUM_NUM = 32768; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 91 | private static final int TP_PORT_MAXIMUM_NUM = 65535; |
| 92 | |
| 93 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 94 | protected CoreService coreService; |
| 95 | |
| 96 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 97 | protected DeviceService deviceService; |
| 98 | |
| 99 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 100 | protected DriverService driverService; |
| 101 | |
| 102 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 103 | protected LeadershipService leadershipService; |
| 104 | |
| 105 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 106 | protected MastershipService mastershipService; |
| 107 | |
| 108 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 109 | protected ClusterService clusterService; |
| 110 | |
| 111 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 112 | protected K8sNetworkService k8sNetworkService; |
| 113 | |
| 114 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 115 | protected K8sNodeService k8sNodeService; |
| 116 | |
| 117 | @Reference(cardinality = ReferenceCardinality.MANDATORY) |
| 118 | protected K8sFlowRuleService k8sFlowRuleService; |
| 119 | |
| 120 | private final InternalK8sNetworkListener k8sNetworkListener = |
| 121 | new InternalK8sNetworkListener(); |
| 122 | private final InternalK8sNodeListener k8sNodeListener = |
| 123 | new InternalK8sNodeListener(); |
| 124 | private final ExecutorService eventExecutor = newSingleThreadExecutor( |
| 125 | groupedThreads(this.getClass().getSimpleName(), "event-handler", log)); |
| 126 | |
| 127 | private ApplicationId appId; |
| 128 | private NodeId localNodeId; |
| 129 | |
| 130 | @Activate |
| 131 | protected void activate() { |
| 132 | appId = coreService.registerApplication(K8S_NETWORKING_APP_ID); |
| 133 | |
| 134 | localNodeId = clusterService.getLocalNode().id(); |
| 135 | leadershipService.runForLeadership(appId.name()); |
| 136 | k8sNetworkService.addListener(k8sNetworkListener); |
| 137 | k8sNodeService.addListener(k8sNodeListener); |
| 138 | |
| 139 | log.info("Started"); |
| 140 | } |
| 141 | |
| 142 | @Deactivate |
| 143 | protected void deactivate() { |
| 144 | k8sNodeService.removeListener(k8sNodeListener); |
| 145 | k8sNetworkService.removeListener(k8sNetworkListener); |
| 146 | leadershipService.withdraw(appId.name()); |
| 147 | eventExecutor.shutdown(); |
| 148 | |
| 149 | log.info("Stopped"); |
| 150 | } |
| 151 | |
| 152 | private void setContainerToExtRule(K8sNode k8sNode, boolean install) { |
| 153 | |
| 154 | K8sNetwork net = k8sNetworkService.network(k8sNode.hostname()); |
| 155 | |
| 156 | if (net == null) { |
| 157 | return; |
| 158 | } |
| 159 | |
| 160 | TrafficSelector.Builder sBuilder = DefaultTrafficSelector.builder() |
| 161 | .matchEthType(Ethernet.TYPE_IPV4) |
| 162 | .matchTunnelId(Long.valueOf(net.segmentId())) |
| 163 | .matchEthDst(DEFAULT_GATEWAY_MAC); |
| 164 | |
| 165 | TrafficTreatment.Builder tBuilder = DefaultTrafficTreatment.builder() |
| 166 | .setOutput(k8sNode.intgToExtPatchPortNum()); |
| 167 | |
| 168 | k8sFlowRuleService.setRule( |
| 169 | appId, |
| 170 | k8sNode.intgBridge(), |
| 171 | sBuilder.build(), |
| 172 | tBuilder.build(), |
| 173 | PRIORITY_EXTERNAL_ROUTING_RULE, |
| 174 | ROUTING_TABLE, |
| 175 | install); |
| 176 | } |
| 177 | |
| 178 | private void setExtToContainerRule(K8sNode k8sNode, |
| 179 | K8sPort k8sPort, boolean install) { |
| 180 | |
| 181 | K8sNetwork net = k8sNetworkService.network(k8sPort.networkId()); |
| 182 | |
| 183 | if (net == null) { |
| 184 | return; |
| 185 | } |
| 186 | |
| 187 | TrafficSelector.Builder sBuilder = DefaultTrafficSelector.builder() |
| 188 | .matchEthType(Ethernet.TYPE_IPV4) |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 189 | .matchIPDst(IpPrefix.valueOf(k8sPort.ipAddress(), HOST_PREFIX)); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 190 | |
| 191 | TrafficTreatment.Builder tBuilder = DefaultTrafficTreatment.builder() |
| 192 | .setOutput(k8sNode.extToIntgPatchPortNum()); |
| 193 | |
| 194 | k8sFlowRuleService.setRule( |
| 195 | appId, |
| 196 | k8sNode.extBridge(), |
| 197 | sBuilder.build(), |
| 198 | tBuilder.build(), |
| 199 | PRIORITY_STATEFUL_SNAT_RULE, |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 200 | POD_RESOLUTION_TABLE, |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 201 | install); |
| 202 | } |
| 203 | |
| 204 | private void setSnatDownstreamRule(K8sNode k8sNode, |
| 205 | boolean install) { |
| 206 | DeviceId deviceId = k8sNode.extBridge(); |
| 207 | |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 208 | TrafficSelector.Builder sBuilder = DefaultTrafficSelector.builder() |
| 209 | .matchEthType(Ethernet.TYPE_IPV4) |
| 210 | .matchIPDst(IpPrefix.valueOf(k8sNode.extBridgeIp(), HOST_PREFIX)); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 211 | |
| 212 | ExtensionTreatment natTreatment = RulePopulatorUtil |
| 213 | .niciraConnTrackTreatmentBuilder(driverService, deviceId) |
| 214 | .commit(false) |
| 215 | .natAction(true) |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 216 | .table((short) POD_RESOLUTION_TABLE) |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 217 | .build(); |
| 218 | |
| 219 | TrafficTreatment treatment = DefaultTrafficTreatment.builder() |
| 220 | .setEthSrc(DEFAULT_GATEWAY_MAC) |
| 221 | .extension(natTreatment, deviceId) |
| 222 | .build(); |
| 223 | |
| 224 | k8sFlowRuleService.setRule( |
| 225 | appId, |
| 226 | deviceId, |
| 227 | sBuilder.build(), |
| 228 | treatment, |
| 229 | PRIORITY_STATEFUL_SNAT_RULE, |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 230 | EXT_ENTRY_TABLE, |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 231 | install); |
| 232 | } |
| 233 | |
| 234 | private void setSnatUpstreamRule(K8sNode k8sNode, |
| 235 | boolean install) { |
| 236 | |
| 237 | K8sNetwork net = k8sNetworkService.network(k8sNode.hostname()); |
| 238 | |
| 239 | if (net == null) { |
| 240 | return; |
| 241 | } |
| 242 | |
| 243 | TrafficSelector selector = DefaultTrafficSelector.builder() |
| 244 | .matchEthType(Ethernet.TYPE_IPV4) |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 245 | .matchInPort(k8sNode.extToIntgPatchPortNum()) |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 246 | .matchEthDst(DEFAULT_GATEWAY_MAC) |
| 247 | .build(); |
| 248 | |
| 249 | TrafficTreatment.Builder tBuilder = DefaultTrafficTreatment.builder(); |
| 250 | |
| 251 | if (install) { |
| 252 | ExtensionTreatment natTreatment = RulePopulatorUtil |
| 253 | .niciraConnTrackTreatmentBuilder(driverService, k8sNode.extBridge()) |
| 254 | .commit(true) |
| 255 | .natFlag(CT_NAT_SRC_FLAG) |
| 256 | .natAction(true) |
| 257 | .natIp(k8sNode.extBridgeIp()) |
| 258 | .natPortMin(TpPort.tpPort(TP_PORT_MINIMUM_NUM)) |
| 259 | .natPortMax(TpPort.tpPort(TP_PORT_MAXIMUM_NUM)) |
| 260 | .build(); |
| 261 | |
| 262 | tBuilder.extension(natTreatment, k8sNode.extBridge()) |
| 263 | .setEthSrc(k8sNode.extBridgeMac()) |
Jian Li | b121844 | 2020-09-03 13:12:14 +0900 | [diff] [blame^] | 264 | .setEthDst(k8sNode.extGatewayMac()); |
| 265 | |
| 266 | if (MacAddress.valueOf(DEFAULT_EXTERNAL_GATEWAY_MAC).equals( |
| 267 | k8sNode.extGatewayMac())) { |
| 268 | tBuilder.setOutput(k8sNode.extIntfPortNum()); |
| 269 | } else { |
| 270 | tBuilder.setOutput(k8sNode.extBridgePortNum()); |
| 271 | } |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 272 | } |
| 273 | |
| 274 | k8sFlowRuleService.setRule( |
| 275 | appId, |
| 276 | k8sNode.extBridge(), |
| 277 | selector, |
| 278 | tBuilder.build(), |
| 279 | PRIORITY_STATEFUL_SNAT_RULE, |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 280 | EXT_ENTRY_TABLE, |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 281 | install); |
| 282 | } |
| 283 | |
| 284 | private void setExtIntfArpRule(K8sNode k8sNode, boolean install) { |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 285 | k8sNodeService.completeNodes().forEach(n -> { |
| 286 | Device device = deviceService.getDevice(n.extBridge()); |
| 287 | TrafficSelector selector = DefaultTrafficSelector.builder() |
| 288 | .matchEthType(Ethernet.TYPE_ARP) |
| 289 | .matchArpOp(ARP.OP_REQUEST) |
| 290 | .matchArpTpa(Ip4Address.valueOf(k8sNode.extBridgeIp().toString())) |
| 291 | .build(); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 292 | |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 293 | TrafficTreatment treatment = DefaultTrafficTreatment.builder() |
| 294 | .setArpOp(ARP.OP_REPLY) |
| 295 | .extension(buildMoveEthSrcToDstExtension(device), device.id()) |
| 296 | .extension(buildMoveArpShaToThaExtension(device), device.id()) |
| 297 | .extension(buildMoveArpSpaToTpaExtension(device), device.id()) |
Jian Li | 232a32c | 2020-02-04 00:32:21 +0900 | [diff] [blame] | 298 | .setEthSrc(k8sNode.extBridgeMac()) |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 299 | .setArpSha(k8sNode.extBridgeMac()) |
Jian Li | 232a32c | 2020-02-04 00:32:21 +0900 | [diff] [blame] | 300 | .setArpSpa(Ip4Address.valueOf(k8sNode.extBridgeIp().toString())) |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 301 | .setOutput(PortNumber.IN_PORT) |
| 302 | .build(); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 303 | |
Jian Li | 140d8a2 | 2019-04-24 23:41:44 +0900 | [diff] [blame] | 304 | k8sFlowRuleService.setRule( |
| 305 | appId, |
| 306 | n.extBridge(), |
| 307 | selector, |
| 308 | treatment, |
| 309 | PRIORITY_STATEFUL_SNAT_RULE, |
| 310 | EXT_ENTRY_TABLE, |
| 311 | install); |
| 312 | }); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 313 | } |
| 314 | |
| 315 | private class InternalK8sNodeListener implements K8sNodeListener { |
| 316 | |
| 317 | private boolean isRelevantHelper() { |
| 318 | return Objects.equals(localNodeId, leadershipService.getLeader(appId.name())); |
| 319 | } |
| 320 | |
| 321 | @Override |
| 322 | public void event(K8sNodeEvent event) { |
| 323 | switch (event.type()) { |
| 324 | case K8S_NODE_COMPLETE: |
| 325 | eventExecutor.execute(() -> processNodeCompletion(event.subject())); |
| 326 | break; |
Jian Li | 1b08d65 | 2019-05-02 17:28:09 +0900 | [diff] [blame] | 327 | case K8S_NODE_UPDATED: |
| 328 | eventExecutor.execute(() -> processNodeUpdate(event.subject())); |
| 329 | break; |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 330 | case K8S_NODE_INCOMPLETE: |
| 331 | default: |
| 332 | break; |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | private void processNodeCompletion(K8sNode k8sNode) { |
| 337 | if (!isRelevantHelper()) { |
| 338 | return; |
| 339 | } |
| 340 | |
| 341 | setExtIntfArpRule(k8sNode, true); |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 342 | setSnatDownstreamRule(k8sNode, true); |
| 343 | setContainerToExtRule(k8sNode, true); |
| 344 | } |
Jian Li | 1b08d65 | 2019-05-02 17:28:09 +0900 | [diff] [blame] | 345 | |
| 346 | private void processNodeUpdate(K8sNode k8sNode) { |
| 347 | if (k8sNode.extGatewayMac() != null) { |
| 348 | setSnatUpstreamRule(k8sNode, true); |
| 349 | } |
| 350 | } |
Jian Li | eb488ea | 2019-04-16 01:50:02 +0900 | [diff] [blame] | 351 | } |
| 352 | |
| 353 | private class InternalK8sNetworkListener implements K8sNetworkListener { |
| 354 | |
| 355 | private boolean isRelevantHelper() { |
| 356 | return Objects.equals(localNodeId, leadershipService.getLeader(appId.name())); |
| 357 | } |
| 358 | |
| 359 | @Override |
| 360 | public void event(K8sNetworkEvent event) { |
| 361 | switch (event.type()) { |
| 362 | case K8S_PORT_ACTIVATED: |
| 363 | eventExecutor.execute(() -> processPortActivation(event.port())); |
| 364 | break; |
| 365 | case K8S_PORT_REMOVED: |
| 366 | eventExecutor.execute(() -> processPortRemoval(event.port())); |
| 367 | break; |
| 368 | default: |
| 369 | break; |
| 370 | } |
| 371 | } |
| 372 | |
| 373 | private void processPortActivation(K8sPort port) { |
| 374 | if (!isRelevantHelper()) { |
| 375 | return; |
| 376 | } |
| 377 | |
| 378 | k8sNodeService.completeNodes().forEach(n -> |
| 379 | setExtToContainerRule(n, port, true)); |
| 380 | } |
| 381 | |
| 382 | private void processPortRemoval(K8sPort port) { |
| 383 | if (!isRelevantHelper()) { |
| 384 | return; |
| 385 | } |
| 386 | |
| 387 | k8sNodeService.completeNodes().forEach(n -> |
| 388 | setExtToContainerRule(n, port, false)); |
| 389 | } |
| 390 | } |
| 391 | } |