ofproto/ofproto-dpif-ipfix.c

   1 /*
   2  * Copyright (c) 2012 Nicira, Inc.
   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
  17 #include <config.h>
  18 #include "ofproto-dpif-ipfix.h"
  19 #include <sys/time.h>
  20 #include "byte-order.h"
  21 #include "collectors.h"
  22 #include "flow.h"
  23 #include "hash.h"
  24 #include "hmap.h"
  25 #include "list.h"
  26 #include "ofpbuf.h"
  27 #include "ofproto.h"
  28 #include "packets.h"
  29 #include "poll-loop.h"
  30 #include "sset.h"
  31 #include "util.h"
  32 #include "timeval.h"
  33 #include "util.h"
  34 #include "vlog.h"
  35
  36 VLOG_DEFINE_THIS_MODULE(ipfix);
  37
  38 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  39 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
  40
  41 /* Cf. IETF RFC 5101 Section 10.3.4. */
  42 #define IPFIX_DEFAULT_COLLECTOR_PORT 4739
  43
  44 struct dpif_ipfix_exporter {
  45     struct collectors *collectors;
  46     uint32_t seq_number;
  47     time_t last_template_set_time;
  48     struct hmap cache_flow_key_map;  /* ipfix_flow_cache_entry. */
  49     struct list cache_flow_start_timestamp_list;  /* ipfix_flow_cache_entry. */
  50     uint32_t cache_active_timeout;  /* In seconds. */
  51     uint32_t cache_max_flows;
  52 };
  53
  54 struct dpif_ipfix_bridge_exporter {
  55     struct dpif_ipfix_exporter exporter;
  56     struct ofproto_ipfix_bridge_exporter_options *options;
  57     uint32_t probability;
  58 };
  59
  60 struct dpif_ipfix_flow_exporter {
  61     struct dpif_ipfix_exporter exporter;
  62     struct ofproto_ipfix_flow_exporter_options *options;
  63 };
  64
  65 struct dpif_ipfix_flow_exporter_map_node {
  66     struct hmap_node node;
  67     struct dpif_ipfix_flow_exporter exporter;
  68 };
  69
  70 struct dpif_ipfix {
  71     struct dpif_ipfix_bridge_exporter bridge_exporter;
  72     struct hmap flow_exporter_map;  /* dpif_ipfix_flow_exporter_map_node. */
  73     atomic_int ref_cnt;
  74 };
  75
  76 #define IPFIX_VERSION 0x000a
  77
  78 /* When using UDP, IPFIX Template Records must be re-sent regularly.
  79  * The standard default interval is 10 minutes (600 seconds).
  80  * Cf. IETF RFC 5101 Section 10.3.6. */
  81 #define IPFIX_TEMPLATE_INTERVAL 600
  82
  83 /* Cf. IETF RFC 5101 Section 3.1. */
  84 OVS_PACKED(
  85 struct ipfix_header {
  86     ovs_be16 version;  /* IPFIX_VERSION. */
  87     ovs_be16 length;  /* Length in bytes including this header. */
  88     ovs_be32 export_time;  /* Seconds since the epoch. */
  89     ovs_be32 seq_number;  /* Message sequence number. */
  90     ovs_be32 obs_domain_id;  /* Observation Domain ID. */
  91 });
  92 BUILD_ASSERT_DECL(sizeof(struct ipfix_header) == 16);
  93
  94 #define IPFIX_SET_ID_TEMPLATE 2
  95 #define IPFIX_SET_ID_OPTION_TEMPLATE 3
  96
  97 /* Cf. IETF RFC 5101 Section 3.3.2. */
  98 OVS_PACKED(
  99 struct ipfix_set_header {
 100     ovs_be16 set_id;  /* IPFIX_SET_ID_* or valid template ID for Data Sets. */
 101     ovs_be16 length;  /* Length of the set in bytes including header. */
 102 });
 103 BUILD_ASSERT_DECL(sizeof(struct ipfix_set_header) == 4);
 104
 105 /* Alternatives for templates at each layer.  A template is defined by
 106  * a combination of one value for each layer. */
 107 enum ipfix_proto_l2 {
 108     IPFIX_PROTO_L2_ETH = 0,  /* No VLAN. */
 109     IPFIX_PROTO_L2_VLAN,
 110     NUM_IPFIX_PROTO_L2
 111 };
 112 enum ipfix_proto_l3 {
 113     IPFIX_PROTO_L3_UNKNOWN = 0,
 114     IPFIX_PROTO_L3_IPV4,
 115     IPFIX_PROTO_L3_IPV6,
 116     NUM_IPFIX_PROTO_L3
 117 };
 118 enum ipfix_proto_l4 {
 119     IPFIX_PROTO_L4_UNKNOWN = 0,
 120     IPFIX_PROTO_L4_TCP_UDP,
 121     NUM_IPFIX_PROTO_L4
 122 };
 123
 124 /* Any Template ID > 255 is usable for Template Records. */
 125 #define IPFIX_TEMPLATE_ID_MIN 256
 126
 127 /* Cf. IETF RFC 5101 Section 3.4.1. */
 128 OVS_PACKED(
 129 struct ipfix_template_record_header {
 130     ovs_be16 template_id;
 131     ovs_be16 field_count;
 132 });
 133 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_record_header) == 4);
 134
 135 enum ipfix_entity_id {
 136 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_ID_##ENUM = ID,
 137 #include "ofproto/ipfix-entities.def"
 138 };
 139
 140 enum ipfix_entity_size {
 141 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_SIZE_##ENUM = SIZE,
 142 #include "ofproto/ipfix-entities.def"
 143 };
 144
 145 OVS_PACKED(
 146 struct ipfix_template_field_specifier {
 147     ovs_be16 element_id;  /* IPFIX_ENTITY_ID_*. */
 148     ovs_be16 field_length;  /* Length of the field's value, in bytes. */
 149     /* No Enterprise ID, since only standard element IDs are specified. */
 150 });
 151 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_field_specifier) == 4);
 152
 153 /* Part of data record flow key for common metadata and Ethernet entities. */
 154 OVS_PACKED(
 155 struct ipfix_data_record_flow_key_common {
 156     ovs_be32 observation_point_id;  /* OBSERVATION_POINT_ID */
 157     uint8_t source_mac_address[6];  /* SOURCE_MAC_ADDRESS */
 158     uint8_t destination_mac_address[6];  /* DESTINATION_MAC_ADDRESS */
 159     ovs_be16 ethernet_type;  /* ETHERNET_TYPE */
 160     uint8_t ethernet_header_length;  /* ETHERNET_HEADER_LENGTH */
 161 });
 162 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_common) == 19);
 163
 164 /* Part of data record flow key for VLAN entities. */
 165 OVS_PACKED(
 166 struct ipfix_data_record_flow_key_vlan {
 167     ovs_be16 vlan_id;  /* VLAN_ID */
 168     ovs_be16 dot1q_vlan_id;  /* DOT1Q_VLAN_ID */
 169     uint8_t dot1q_priority;  /* DOT1Q_PRIORITY */
 170 });
 171 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_vlan) == 5);
 172
 173 /* Part of data record flow key for IP entities. */
 174 /* XXX: Replace IP_TTL with MINIMUM_TTL and MAXIMUM_TTL? */
 175 OVS_PACKED(
 176 struct ipfix_data_record_flow_key_ip {
 177     uint8_t ip_version;  /* IP_VERSION */
 178     uint8_t ip_ttl;  /* IP_TTL */
 179     uint8_t protocol_identifier;  /* PROTOCOL_IDENTIFIER */
 180     uint8_t ip_diff_serv_code_point;  /* IP_DIFF_SERV_CODE_POINT */
 181     uint8_t ip_precedence;  /* IP_PRECEDENCE */
 182     uint8_t ip_class_of_service;  /* IP_CLASS_OF_SERVICE */
 183 });
 184 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ip) == 6);
 185
 186 /* Part of data record flow key for IPv4 entities. */
 187 OVS_PACKED(
 188 struct ipfix_data_record_flow_key_ipv4 {
 189     ovs_be32 source_ipv4_address;  /* SOURCE_IPV4_ADDRESS */
 190     ovs_be32 destination_ipv4_address;  /* DESTINATION_IPV4_ADDRESS */
 191 });
 192 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ipv4) == 8);
 193
 194 /* Part of data record flow key for IPv6 entities. */
 195 OVS_PACKED(
 196 struct ipfix_data_record_flow_key_ipv6 {
 197     uint8_t source_ipv6_address[16];  /* SOURCE_IPV6_ADDRESS */
 198     uint8_t destination_ipv6_address[16];  /* DESTINATION_IPV6_ADDRESS */
 199     ovs_be32 flow_label_ipv6;  /* FLOW_LABEL_IPV6 */
 200 });
 201 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ipv6) == 36);
 202
 203 /* Part of data record flow key for TCP/UDP entities. */
 204 OVS_PACKED(
 205 struct ipfix_data_record_flow_key_tcpudp {
 206     ovs_be16 source_transport_port;  /* SOURCE_TRANSPORT_PORT */
 207     ovs_be16 destination_transport_port;  /* DESTINATION_TRANSPORT_PORT */
 208 });
 209 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_tcpudp) == 4);
 210
 211 /* Cf. IETF RFC 5102 Section 5.11.3. */
 212 enum ipfix_flow_end_reason {
 213     IDLE_TIMEOUT = 0x01,
 214     ACTIVE_TIMEOUT = 0x02,
 215     END_OF_FLOW_DETECTED = 0x03,
 216     FORCED_END = 0x04,
 217     LACK_OF_RESOURCES = 0x05
 218 };
 219
 220 /* Part of data record for common aggregated elements. */
 221 OVS_PACKED(
 222 struct ipfix_data_record_aggregated_common {
 223     ovs_be32 flow_start_delta_microseconds; /* FLOW_START_DELTA_MICROSECONDS */
 224     ovs_be32 flow_end_delta_microseconds; /* FLOW_END_DELTA_MICROSECONDS */
 225     ovs_be64 packet_delta_count;  /* PACKET_DELTA_COUNT */
 226     ovs_be64 layer2_octet_delta_count;  /* LAYER2_OCTET_DELTA_COUNT */
 227     uint8_t flow_end_reason;  /* FLOW_END_REASON */
 228 });
 229 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_aggregated_common) == 25);
 230
 231 /* Part of data record for IP aggregated elements. */
 232 OVS_PACKED(
 233 struct ipfix_data_record_aggregated_ip {
 234     ovs_be64 octet_delta_sum_of_squares;  /* OCTET_DELTA_SUM_OF_SQUARES */
 235     ovs_be64 minimum_ip_total_length;  /* MINIMUM_IP_TOTAL_LENGTH */
 236     ovs_be64 maximum_ip_total_length;  /* MAXIMUM_IP_TOTAL_LENGTH */
 237 });
 238 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_aggregated_ip) == 24);
 239
 240 #define MAX_FLOW_KEY_LEN                                 \
 241     (sizeof(struct ipfix_data_record_flow_key_common)    \
 242      + sizeof(struct ipfix_data_record_flow_key_vlan)    \
 243      + sizeof(struct ipfix_data_record_flow_key_ip)      \
 244      + sizeof(struct ipfix_data_record_flow_key_ipv6)    \
 245      + sizeof(struct ipfix_data_record_flow_key_tcpudp))
 246
 247 #define MAX_DATA_RECORD_LEN                                 \
 248     (MAX_FLOW_KEY_LEN                                       \
 249      + sizeof(struct ipfix_data_record_aggregated_common)   \
 250      + sizeof(struct ipfix_data_record_aggregated_ip))
 251
 252 /* Max length of a data set.  To simplify the implementation, each
 253  * data record is sent in a separate data set, so each data set
 254  * contains at most one data record. */
 255 #define MAX_DATA_SET_LEN             \
 256     (sizeof(struct ipfix_set_header) \
 257      + MAX_DATA_RECORD_LEN)
 258
 259 /* Max length of an IPFIX message. Arbitrarily set to accomodate low
 260  * MTU. */
 261 #define MAX_MESSAGE_LEN 1024
 262
 263 /* Cache structures. */
 264
 265 /* Flow key. */
 266 struct ipfix_flow_key {
 267     uint32_t obs_domain_id;
 268     uint16_t template_id;
 269     size_t flow_key_msg_part_size;
 270     uint64_t flow_key_msg_part[DIV_ROUND_UP(MAX_FLOW_KEY_LEN, 8)];
 271 };
 272
 273 /* Flow cache entry. */
 274 struct ipfix_flow_cache_entry {
 275     struct hmap_node flow_key_map_node;
 276     struct list cache_flow_start_timestamp_list_node;
 277     struct ipfix_flow_key flow_key;
 278     /* Common aggregated elements. */
 279     uint64_t flow_start_timestamp_usec;
 280     uint64_t flow_end_timestamp_usec;
 281     uint64_t packet_delta_count;
 282     uint64_t layer2_octet_delta_count;
 283     uint64_t octet_delta_sum_of_squares;  /* 0 if not IP. */
 284     uint16_t minimum_ip_total_length;  /* 0 if not IP. */
 285     uint16_t maximum_ip_total_length;  /* 0 if not IP. */
 286 };
 287
 288 static void dpif_ipfix_cache_expire(struct dpif_ipfix_exporter *, bool,
 289                                     const uint64_t, const uint32_t);
 290
 291 static void get_export_time_now(uint64_t *, uint32_t *);
 292
 293 static void dpif_ipfix_cache_expire_now(struct dpif_ipfix_exporter *, bool);
 294
 295 static bool
 296 ofproto_ipfix_bridge_exporter_options_equal(
 297     const struct ofproto_ipfix_bridge_exporter_options *a,
 298     const struct ofproto_ipfix_bridge_exporter_options *b)
 299 {
 300     return (a->obs_domain_id == b->obs_domain_id
 301             && a->obs_point_id == b->obs_point_id
 302             && a->sampling_rate == b->sampling_rate
 303             && a->cache_active_timeout == b->cache_active_timeout
 304             && a->cache_max_flows == b->cache_max_flows
 305             && sset_equals(&a->targets, &b->targets));
 306 }
 307
 308 static struct ofproto_ipfix_bridge_exporter_options *
 309 ofproto_ipfix_bridge_exporter_options_clone(
 310     const struct ofproto_ipfix_bridge_exporter_options *old)
 311 {
 312     struct ofproto_ipfix_bridge_exporter_options *new =
 313         xmemdup(old, sizeof *old);
 314     sset_clone(&new->targets, &old->targets);
 315     return new;
 316 }
 317
 318 static void
 319 ofproto_ipfix_bridge_exporter_options_destroy(
 320     struct ofproto_ipfix_bridge_exporter_options *options)
 321 {
 322     if (options) {
 323         sset_destroy(&options->targets);
 324         free(options);
 325     }
 326 }
 327
 328 static bool
 329 ofproto_ipfix_flow_exporter_options_equal(
 330     const struct ofproto_ipfix_flow_exporter_options *a,
 331     const struct ofproto_ipfix_flow_exporter_options *b)
 332 {
 333     return (a->collector_set_id == b->collector_set_id
 334             && a->cache_active_timeout == b->cache_active_timeout
 335             && a->cache_max_flows == b->cache_max_flows
 336             && sset_equals(&a->targets, &b->targets));
 337 }
 338
 339 static struct ofproto_ipfix_flow_exporter_options *
 340 ofproto_ipfix_flow_exporter_options_clone(
 341     const struct ofproto_ipfix_flow_exporter_options *old)
 342 {
 343     struct ofproto_ipfix_flow_exporter_options *new =
 344         xmemdup(old, sizeof *old);
 345     sset_clone(&new->targets, &old->targets);
 346     return new;
 347 }
 348
 349 static void
 350 ofproto_ipfix_flow_exporter_options_destroy(
 351     struct ofproto_ipfix_flow_exporter_options *options)
 352 {
 353     if (options) {
 354         sset_destroy(&options->targets);
 355         free(options);
 356     }
 357 }
 358
 359 static void
 360 dpif_ipfix_exporter_init(struct dpif_ipfix_exporter *exporter)
 361 {
 362     exporter->collectors = NULL;
 363     exporter->seq_number = 1;
 364     exporter->last_template_set_time = TIME_MIN;
 365     hmap_init(&exporter->cache_flow_key_map);
 366     list_init(&exporter->cache_flow_start_timestamp_list);
 367     exporter->cache_active_timeout = 0;
 368     exporter->cache_max_flows = 0;
 369 }
 370
 371 static void
 372 dpif_ipfix_exporter_clear(struct dpif_ipfix_exporter *exporter)
 373 {
 374     /* Flush the cache with flow end reason "forced end." */
 375     dpif_ipfix_cache_expire_now(exporter, true);
 376
 377     collectors_destroy(exporter->collectors);
 378     exporter->collectors = NULL;
 379     exporter->seq_number = 1;
 380     exporter->last_template_set_time = TIME_MIN;
 381     exporter->cache_active_timeout = 0;
 382     exporter->cache_max_flows = 0;
 383 }
 384
 385 static void
 386 dpif_ipfix_exporter_destroy(struct dpif_ipfix_exporter *exporter)
 387 {
 388     dpif_ipfix_exporter_clear(exporter);
 389     hmap_destroy(&exporter->cache_flow_key_map);
 390 }
 391
 392 static bool
 393 dpif_ipfix_exporter_set_options(struct dpif_ipfix_exporter *exporter,
 394                                 const struct sset *targets,
 395                                 const uint32_t cache_active_timeout,
 396                                 const uint32_t cache_max_flows)
 397 {
 398     collectors_destroy(exporter->collectors);
 399     collectors_create(targets, IPFIX_DEFAULT_COLLECTOR_PORT,
 400                       &exporter->collectors);
 401     if (exporter->collectors == NULL) {
 402         VLOG_WARN_RL(&rl, "no collectors could be initialized, "
 403                      "IPFIX exporter disabled");
 404         dpif_ipfix_exporter_clear(exporter);
 405         return false;
 406     }
 407     exporter->cache_active_timeout = cache_active_timeout;
 408     exporter->cache_max_flows = cache_max_flows;
 409     return true;
 410 }
 411
 412 static void
 413 dpif_ipfix_bridge_exporter_init(struct dpif_ipfix_bridge_exporter *exporter)
 414 {
 415     dpif_ipfix_exporter_init(&exporter->exporter);
 416     exporter->options = NULL;
 417     exporter->probability = 0;
 418 }
 419
 420 static void
 421 dpif_ipfix_bridge_exporter_clear(struct dpif_ipfix_bridge_exporter *exporter)
 422 {
 423     dpif_ipfix_exporter_clear(&exporter->exporter);
 424     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 425     exporter->options = NULL;
 426     exporter->probability = 0;
 427 }
 428
 429 static void
 430 dpif_ipfix_bridge_exporter_destroy(struct dpif_ipfix_bridge_exporter *exporter)
 431 {
 432     dpif_ipfix_bridge_exporter_clear(exporter);
 433     dpif_ipfix_exporter_destroy(&exporter->exporter);
 434 }
 435
 436 static void
 437 dpif_ipfix_bridge_exporter_set_options(
 438     struct dpif_ipfix_bridge_exporter *exporter,
 439     const struct ofproto_ipfix_bridge_exporter_options *options)
 440 {
 441     bool options_changed;
 442
 443     if (!options || sset_is_empty(&options->targets)) {
 444         /* No point in doing any work if there are no targets. */
 445         dpif_ipfix_bridge_exporter_clear(exporter);
 446         return;
 447     }
 448
 449     options_changed = (
 450         !exporter->options
 451         || !ofproto_ipfix_bridge_exporter_options_equal(
 452             options, exporter->options));
 453
 454     /* Configure collectors if options have changed or if we're
 455      * shortchanged in collectors (which indicates that opening one or
 456      * more of the configured collectors failed, so that we should
 457      * retry). */
 458     if (options_changed
 459         || collectors_count(exporter->exporter.collectors)
 460             < sset_count(&options->targets)) {
 461         if (!dpif_ipfix_exporter_set_options(
 462                 &exporter->exporter, &options->targets,
 463                 options->cache_active_timeout, options->cache_max_flows)) {
 464             return;
 465         }
 466     }
 467
 468     /* Avoid reconfiguring if options didn't change. */
 469     if (!options_changed) {
 470         return;
 471     }
 472
 473     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 474     exporter->options = ofproto_ipfix_bridge_exporter_options_clone(options);
 475     exporter->probability =
 476         MAX(1, UINT32_MAX / exporter->options->sampling_rate);
 477
 478     /* Run over the cache as some entries might have expired after
 479      * changing the timeouts. */
 480     dpif_ipfix_cache_expire_now(&exporter->exporter, false);
 481 }
 482
 483 static struct dpif_ipfix_flow_exporter_map_node*
 484 dpif_ipfix_find_flow_exporter_map_node(
 485     const struct dpif_ipfix *di, const uint32_t collector_set_id)
 486     OVS_REQUIRES(mutex)
 487 {
 488     struct dpif_ipfix_flow_exporter_map_node *exporter_node;
 489
 490     HMAP_FOR_EACH_WITH_HASH (exporter_node, node,
 491                              hash_int(collector_set_id, 0),
 492                              &di->flow_exporter_map) {
 493         if (exporter_node->exporter.options->collector_set_id
 494             == collector_set_id) {
 495             return exporter_node;
 496         }
 497     }
 498
 499     return NULL;
 500 }
 501
 502 static void
 503 dpif_ipfix_flow_exporter_init(struct dpif_ipfix_flow_exporter *exporter)
 504 {
 505     dpif_ipfix_exporter_init(&exporter->exporter);
 506     exporter->options = NULL;
 507 }
 508
 509 static void
 510 dpif_ipfix_flow_exporter_clear(struct dpif_ipfix_flow_exporter *exporter)
 511 {
 512     dpif_ipfix_exporter_clear(&exporter->exporter);
 513     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 514     exporter->options = NULL;
 515 }
 516
 517 static void
 518 dpif_ipfix_flow_exporter_destroy(struct dpif_ipfix_flow_exporter *exporter)
 519 {
 520     dpif_ipfix_flow_exporter_clear(exporter);
 521     dpif_ipfix_exporter_destroy(&exporter->exporter);
 522 }
 523
 524 static bool
 525 dpif_ipfix_flow_exporter_set_options(
 526     struct dpif_ipfix_flow_exporter *exporter,
 527     const struct ofproto_ipfix_flow_exporter_options *options)
 528 {
 529     bool options_changed;
 530
 531     if (sset_is_empty(&options->targets)) {
 532         /* No point in doing any work if there are no targets. */
 533         dpif_ipfix_flow_exporter_clear(exporter);
 534         return true;
 535     }
 536
 537     options_changed = (
 538         !exporter->options
 539         || !ofproto_ipfix_flow_exporter_options_equal(
 540             options, exporter->options));
 541
 542     /* Configure collectors if options have changed or if we're
 543      * shortchanged in collectors (which indicates that opening one or
 544      * more of the configured collectors failed, so that we should
 545      * retry). */
 546     if (options_changed
 547         || collectors_count(exporter->exporter.collectors)
 548             < sset_count(&options->targets)) {
 549         if (!dpif_ipfix_exporter_set_options(
 550                 &exporter->exporter, &options->targets,
 551                 options->cache_active_timeout, options->cache_max_flows)) {
 552             return false;
 553         }
 554     }
 555
 556     /* Avoid reconfiguring if options didn't change. */
 557     if (!options_changed) {
 558         return true;
 559     }
 560
 561     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 562     exporter->options = ofproto_ipfix_flow_exporter_options_clone(options);
 563
 564     /* Run over the cache as some entries might have expired after
 565      * changing the timeouts. */
 566     dpif_ipfix_cache_expire_now(&exporter->exporter, false);
 567
 568     return true;
 569 }
 570
 571 void
 572 dpif_ipfix_set_options(
 573     struct dpif_ipfix *di,
 574     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
 575     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
 576     size_t n_flow_exporters_options) OVS_EXCLUDED(mutex)
 577 {
 578     int i;
 579     struct ofproto_ipfix_flow_exporter_options *options;
 580     struct dpif_ipfix_flow_exporter_map_node *node, *next;
 581     size_t n_broken_flow_exporters_options = 0;
 582
 583     ovs_mutex_lock(&mutex);
 584     dpif_ipfix_bridge_exporter_set_options(&di->bridge_exporter,
 585                                            bridge_exporter_options);
 586
 587     /* Add new flow exporters and update current flow exporters. */
 588     options = (struct ofproto_ipfix_flow_exporter_options *)
 589         flow_exporters_options;
 590     for (i = 0; i < n_flow_exporters_options; i++) {
 591         node = dpif_ipfix_find_flow_exporter_map_node(
 592             di, options->collector_set_id);
 593         if (!node) {
 594             node = xzalloc(sizeof *node);
 595             dpif_ipfix_flow_exporter_init(&node->exporter);
 596             hmap_insert(&di->flow_exporter_map, &node->node,
 597                         hash_int(options->collector_set_id, 0));
 598         }
 599         if (!dpif_ipfix_flow_exporter_set_options(&node->exporter, options)) {
 600             n_broken_flow_exporters_options++;
 601         }
 602         options++;
 603     }
 604
 605     ovs_assert(hmap_count(&di->flow_exporter_map) >=
 606                (n_flow_exporters_options - n_broken_flow_exporters_options));
 607
 608     /* Remove dropped flow exporters, if any needs to be removed. */
 609     if (hmap_count(&di->flow_exporter_map) > n_flow_exporters_options) {
 610         HMAP_FOR_EACH_SAFE (node, next, node, &di->flow_exporter_map) {
 611             /* This is slow but doesn't take any extra memory, and
 612              * this table is not supposed to contain many rows anyway. */
 613             options = (struct ofproto_ipfix_flow_exporter_options *)
 614                 flow_exporters_options;
 615             for (i = 0; i < n_flow_exporters_options; i++) {
 616               if (node->exporter.options->collector_set_id
 617                   == options->collector_set_id) {
 618                   break;
 619               }
 620               options++;
 621             }
 622             if (i == n_flow_exporters_options) {  // Not found.
 623                 hmap_remove(&di->flow_exporter_map, &node->node);
 624                 dpif_ipfix_flow_exporter_destroy(&node->exporter);
 625                 free(node);
 626             }
 627         }
 628     }
 629
 630     ovs_assert(hmap_count(&di->flow_exporter_map) ==
 631                (n_flow_exporters_options - n_broken_flow_exporters_options));
 632     ovs_mutex_unlock(&mutex);
 633 }
 634
 635 struct dpif_ipfix *
 636 dpif_ipfix_create(void)
 637 {
 638     struct dpif_ipfix *di;
 639     di = xzalloc(sizeof *di);
 640     dpif_ipfix_bridge_exporter_init(&di->bridge_exporter);
 641     hmap_init(&di->flow_exporter_map);
 642     atomic_init(&di->ref_cnt, 1);
 643     return di;
 644 }
 645
 646 struct dpif_ipfix *
 647 dpif_ipfix_ref(const struct dpif_ipfix *di_)
 648 {
 649     struct dpif_ipfix *di = CONST_CAST(struct dpif_ipfix *, di_);
 650     if (di) {
 651         int orig;
 652         atomic_add(&di->ref_cnt, 1, &orig);
 653         ovs_assert(orig > 0);
 654     }
 655     return di;
 656 }
 657
 658 uint32_t
 659 dpif_ipfix_get_bridge_exporter_probability(const struct dpif_ipfix *di)
 660     OVS_EXCLUDED(mutex)
 661 {
 662     uint32_t ret;
 663     ovs_mutex_lock(&mutex);
 664     ret = di->bridge_exporter.probability;
 665     ovs_mutex_unlock(&mutex);
 666     return ret;
 667 }
 668
 669 static void
 670 dpif_ipfix_clear(struct dpif_ipfix *di) OVS_REQUIRES(mutex)
 671 {
 672     struct dpif_ipfix_flow_exporter_map_node *exp_node, *exp_next;
 673
 674     dpif_ipfix_bridge_exporter_clear(&di->bridge_exporter);
 675
 676     HMAP_FOR_EACH_SAFE (exp_node, exp_next, node, &di->flow_exporter_map) {
 677         hmap_remove(&di->flow_exporter_map, &exp_node->node);
 678         dpif_ipfix_flow_exporter_destroy(&exp_node->exporter);
 679         free(exp_node);
 680     }
 681 }
 682
 683 void
 684 dpif_ipfix_unref(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
 685 {
 686     int orig;
 687
 688     if (!di) {
 689         return;
 690     }
 691
 692     atomic_sub(&di->ref_cnt, 1, &orig);
 693     ovs_assert(orig > 0);
 694     if (orig == 1) {
 695         ovs_mutex_lock(&mutex);
 696         dpif_ipfix_clear(di);
 697         dpif_ipfix_bridge_exporter_destroy(&di->bridge_exporter);
 698         hmap_destroy(&di->flow_exporter_map);
 699         free(di);
 700         ovs_mutex_unlock(&mutex);
 701     }
 702 }
 703
 704 static void
 705 ipfix_init_header(uint32_t export_time_sec, uint32_t seq_number,
 706                   uint32_t obs_domain_id, struct ofpbuf *msg)
 707 {
 708     struct ipfix_header *hdr;
 709
 710     hdr = ofpbuf_put_zeros(msg, sizeof *hdr);
 711     hdr->version = htons(IPFIX_VERSION);
 712     hdr->length = htons(sizeof *hdr);  /* Updated in ipfix_send_msg. */
 713     hdr->export_time = htonl(export_time_sec);
 714     hdr->seq_number = htonl(seq_number);
 715     hdr->obs_domain_id = htonl(obs_domain_id);
 716 }
 717
 718 static void
 719 ipfix_send_msg(const struct collectors *collectors, struct ofpbuf *msg)
 720 {
 721     struct ipfix_header *hdr;
 722
 723     /* Adjust the length in the header. */
 724     hdr = msg->data;
 725     hdr->length = htons(msg->size);
 726
 727     collectors_send(collectors, msg->data, msg->size);
 728     msg->size = 0;
 729 }
 730
 731 static uint16_t
 732 ipfix_get_template_id(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
 733                       enum ipfix_proto_l4 l4)
 734 {
 735     uint16_t template_id;
 736     template_id = l2;
 737     template_id = template_id * NUM_IPFIX_PROTO_L3 + l3;
 738     template_id = template_id * NUM_IPFIX_PROTO_L4 + l4;
 739     return IPFIX_TEMPLATE_ID_MIN + template_id;
 740 }
 741
 742 static void
 743 ipfix_define_template_entity(enum ipfix_entity_id id,
 744                              enum ipfix_entity_size size, struct ofpbuf *msg)
 745 {
 746     struct ipfix_template_field_specifier *field;
 747
 748     field = ofpbuf_put_zeros(msg, sizeof *field);
 749     field->element_id = htons(id);
 750     field->field_length = htons(size);
 751 }
 752
 753 static uint16_t
 754 ipfix_define_template_fields(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
 755                              enum ipfix_proto_l4 l4, struct ofpbuf *msg)
 756 {
 757     uint16_t count = 0;
 758
 759 #define DEF(ID) \
 760     { \
 761         ipfix_define_template_entity(IPFIX_ENTITY_ID_##ID, \
 762                                      IPFIX_ENTITY_SIZE_##ID, msg); \
 763         count++; \
 764     }
 765
 766     /* 1. Flow key. */
 767
 768     DEF(OBSERVATION_POINT_ID);
 769
 770     /* Common Ethernet entities. */
 771     DEF(SOURCE_MAC_ADDRESS);
 772     DEF(DESTINATION_MAC_ADDRESS);
 773     DEF(ETHERNET_TYPE);
 774     DEF(ETHERNET_HEADER_LENGTH);
 775
 776     if (l2 == IPFIX_PROTO_L2_VLAN) {
 777         DEF(VLAN_ID);
 778         DEF(DOT1Q_VLAN_ID);
 779         DEF(DOT1Q_PRIORITY);
 780     }
 781
 782     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
 783         DEF(IP_VERSION);
 784         DEF(IP_TTL);
 785         DEF(PROTOCOL_IDENTIFIER);
 786         DEF(IP_DIFF_SERV_CODE_POINT);
 787         DEF(IP_PRECEDENCE);
 788         DEF(IP_CLASS_OF_SERVICE);
 789
 790         if (l3 == IPFIX_PROTO_L3_IPV4) {
 791             DEF(SOURCE_IPV4_ADDRESS);
 792             DEF(DESTINATION_IPV4_ADDRESS);
 793         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
 794             DEF(SOURCE_IPV6_ADDRESS);
 795             DEF(DESTINATION_IPV6_ADDRESS);
 796             DEF(FLOW_LABEL_IPV6);
 797         }
 798     }
 799
 800     if (l4 != IPFIX_PROTO_L4_UNKNOWN) {
 801         DEF(SOURCE_TRANSPORT_PORT);
 802         DEF(DESTINATION_TRANSPORT_PORT);
 803     }
 804
 805     /* 2. Flow aggregated data. */
 806
 807     DEF(FLOW_START_DELTA_MICROSECONDS);
 808     DEF(FLOW_END_DELTA_MICROSECONDS);
 809     DEF(PACKET_DELTA_COUNT);
 810     DEF(LAYER2_OCTET_DELTA_COUNT);
 811     DEF(FLOW_END_REASON);
 812
 813     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
 814         DEF(OCTET_DELTA_SUM_OF_SQUARES);
 815         DEF(MINIMUM_IP_TOTAL_LENGTH);
 816         DEF(MAXIMUM_IP_TOTAL_LENGTH);
 817     }
 818
 819 #undef DEF
 820
 821     return count;
 822 }
 823
 824 static void
 825 ipfix_send_template_msg(struct dpif_ipfix_exporter *exporter,
 826                         uint32_t export_time_sec, uint32_t obs_domain_id)
 827 {
 828     uint64_t msg_stub[DIV_ROUND_UP(MAX_MESSAGE_LEN, 8)];
 829     struct ofpbuf msg;
 830     size_t set_hdr_offset, tmpl_hdr_offset;
 831     struct ipfix_set_header *set_hdr;
 832     struct ipfix_template_record_header *tmpl_hdr;
 833     uint16_t field_count;
 834     enum ipfix_proto_l2 l2;
 835     enum ipfix_proto_l3 l3;
 836     enum ipfix_proto_l4 l4;
 837
 838     ofpbuf_use_stub(&msg, msg_stub, sizeof msg_stub);
 839
 840     ipfix_init_header(export_time_sec, exporter->seq_number, obs_domain_id,
 841                       &msg);
 842     set_hdr_offset = msg.size;
 843
 844     /* Add a Template Set. */
 845     set_hdr = ofpbuf_put_zeros(&msg, sizeof *set_hdr);
 846     set_hdr->set_id = htons(IPFIX_SET_ID_TEMPLATE);
 847
 848     /* Define one template for each possible combination of
 849      * protocols. */
 850     for (l2 = 0; l2 < NUM_IPFIX_PROTO_L2; l2++) {
 851         for (l3 = 0; l3 < NUM_IPFIX_PROTO_L3; l3++) {
 852             for (l4 = 0; l4 < NUM_IPFIX_PROTO_L4; l4++) {
 853                 if (l3 == IPFIX_PROTO_L3_UNKNOWN &&
 854                     l4 != IPFIX_PROTO_L4_UNKNOWN) {
 855                     continue;
 856                 }
 857                 tmpl_hdr_offset = msg.size;
 858                 tmpl_hdr = ofpbuf_put_zeros(&msg, sizeof *tmpl_hdr);
 859                 tmpl_hdr->template_id = htons(
 860                     ipfix_get_template_id(l2, l3, l4));
 861                 field_count = ipfix_define_template_fields(l2, l3, l4, &msg);
 862                 tmpl_hdr = (struct ipfix_template_record_header*)
 863                     ((uint8_t*)msg.data + tmpl_hdr_offset);
 864                 tmpl_hdr->field_count = htons(field_count);
 865             }
 866         }
 867     }
 868
 869     set_hdr = (struct ipfix_set_header*)((uint8_t*)msg.data + set_hdr_offset);
 870     set_hdr->length = htons(msg.size - set_hdr_offset);
 871
 872     /* XXX: Add Options Template Sets, at least to define a Flow Keys
 873      * Option Template. */
 874
 875     ipfix_send_msg(exporter->collectors, &msg);
 876
 877     ofpbuf_uninit(&msg);
 878 }
 879
 880 static inline uint32_t
 881 ipfix_hash_flow_key(const struct ipfix_flow_key *flow_key, uint32_t basis)
 882 {
 883     uint32_t hash;
 884     hash = hash_int(flow_key->obs_domain_id, basis);
 885     hash = hash_int(flow_key->template_id, hash);
 886     hash = hash_bytes(flow_key->flow_key_msg_part,
 887                       flow_key->flow_key_msg_part_size, hash);
 888     return hash;
 889 }
 890
 891 static bool
 892 ipfix_flow_key_equal(const struct ipfix_flow_key *a,
 893                      const struct ipfix_flow_key *b)
 894 {
 895     /* The template ID determines the flow key size, so not need to
 896      * compare it. */
 897     return (a->obs_domain_id == b->obs_domain_id
 898             && a->template_id == b->template_id
 899             && memcmp(a->flow_key_msg_part, b->flow_key_msg_part,
 900                       a->flow_key_msg_part_size) == 0);
 901 }
 902
 903 static struct ipfix_flow_cache_entry*
 904 ipfix_cache_find_entry(const struct dpif_ipfix_exporter *exporter,
 905                        const struct ipfix_flow_key *flow_key)
 906 {
 907     struct ipfix_flow_cache_entry *entry;
 908
 909     HMAP_FOR_EACH_WITH_HASH (entry, flow_key_map_node,
 910                              ipfix_hash_flow_key(flow_key, 0),
 911                              &exporter->cache_flow_key_map) {
 912         if (ipfix_flow_key_equal(&entry->flow_key, flow_key)) {
 913             return entry;
 914         }
 915     }
 916
 917     return NULL;
 918 }
 919
 920 static bool
 921 ipfix_cache_next_timeout_msec(const struct dpif_ipfix_exporter *exporter,
 922                               long long int *next_timeout_msec)
 923 {
 924     struct ipfix_flow_cache_entry *entry;
 925
 926     LIST_FOR_EACH (entry, cache_flow_start_timestamp_list_node,
 927                    &exporter->cache_flow_start_timestamp_list) {
 928         *next_timeout_msec = entry->flow_start_timestamp_usec / 1000LL
 929             + 1000LL * exporter->cache_active_timeout;
 930         return true;
 931     }
 932
 933     return false;
 934 }
 935
 936 static void
 937 ipfix_cache_aggregate_entries(struct ipfix_flow_cache_entry *from_entry,
 938                               struct ipfix_flow_cache_entry *to_entry)
 939 {
 940     uint64_t *to_start, *to_end, *from_start, *from_end;
 941     uint16_t *to_min_len, *to_max_len, *from_min_len, *from_max_len;
 942
 943     to_start = &to_entry->flow_start_timestamp_usec;
 944     to_end = &to_entry->flow_end_timestamp_usec;
 945     from_start = &from_entry->flow_start_timestamp_usec;
 946     from_end = &from_entry->flow_end_timestamp_usec;
 947
 948     if (*to_start > *from_start) {
 949         *to_start = *from_start;
 950     }
 951     if (*to_end < *from_end) {
 952         *to_end = *from_end;
 953     }
 954
 955     to_entry->packet_delta_count += from_entry->packet_delta_count;
 956     to_entry->layer2_octet_delta_count += from_entry->layer2_octet_delta_count;
 957
 958     to_entry->octet_delta_sum_of_squares +=
 959         from_entry->octet_delta_sum_of_squares;
 960
 961     to_min_len = &to_entry->minimum_ip_total_length;
 962     to_max_len = &to_entry->maximum_ip_total_length;
 963     from_min_len = &from_entry->minimum_ip_total_length;
 964     from_max_len = &from_entry->maximum_ip_total_length;
 965
 966     if (!*to_min_len || (*from_min_len && *to_min_len > *from_min_len)) {
 967         *to_min_len = *from_min_len;
 968     }
 969     if (*to_max_len < *from_max_len) {
 970         *to_max_len = *from_max_len;
 971     }
 972 }
 973
 974 /* Add an entry into a flow cache.  The entry is either aggregated into
 975  * an existing entry with the same flow key and free()d, or it is
 976  * inserted into the cache. */
 977 static void
 978 ipfix_cache_update(struct dpif_ipfix_exporter *exporter,
 979                    struct ipfix_flow_cache_entry *entry)
 980 {
 981     struct ipfix_flow_cache_entry *old_entry;
 982
 983     old_entry = ipfix_cache_find_entry(exporter, &entry->flow_key);
 984
 985     if (old_entry == NULL) {
 986         hmap_insert(&exporter->cache_flow_key_map, &entry->flow_key_map_node,
 987                     ipfix_hash_flow_key(&entry->flow_key, 0));
 988
 989         /* As the latest entry added into the cache, it should
 990          * logically have the highest flow_start_timestamp_usec, so
 991          * append it at the tail. */
 992         list_push_back(&exporter->cache_flow_start_timestamp_list,
 993                        &entry->cache_flow_start_timestamp_list_node);
 994
 995         /* Enforce exporter->cache_max_flows limit. */
 996         if (hmap_count(&exporter->cache_flow_key_map)
 997             > exporter->cache_max_flows) {
 998             dpif_ipfix_cache_expire_now(exporter, false);
 999         }
1000     } else {
1001         ipfix_cache_aggregate_entries(entry, old_entry);
1002         free(entry);
1003     }
1004 }
1005
1006 static void
1007 ipfix_cache_entry_init(struct ipfix_flow_cache_entry *entry,
1008                        struct ofpbuf *packet, const struct flow *flow,
1009                        uint64_t packet_delta_count, uint32_t obs_domain_id,
1010                        uint32_t obs_point_id)
1011 {
1012     struct ipfix_flow_key *flow_key;
1013     struct ofpbuf msg;
1014     enum ipfix_proto_l2 l2;
1015     enum ipfix_proto_l3 l3;
1016     enum ipfix_proto_l4 l4;
1017     uint8_t ethernet_header_length;
1018     uint16_t ethernet_total_length;
1019
1020     flow_key = &entry->flow_key;
1021     ofpbuf_use_stack(&msg, flow_key->flow_key_msg_part,
1022                      sizeof flow_key->flow_key_msg_part);
1023
1024     /* Choose the right template ID matching the protocols in the
1025      * sampled packet. */
1026     l2 = (flow->vlan_tci == 0) ? IPFIX_PROTO_L2_ETH : IPFIX_PROTO_L2_VLAN;
1027
1028     switch(ntohs(flow->dl_type)) {
1029     case ETH_TYPE_IP:
1030         l3 = IPFIX_PROTO_L3_IPV4;
1031         break;
1032     case ETH_TYPE_IPV6:
1033         l3 = IPFIX_PROTO_L3_IPV6;
1034         break;
1035     default:
1036         l3 = IPFIX_PROTO_L3_UNKNOWN;
1037     }
1038
1039     l4 = IPFIX_PROTO_L4_UNKNOWN;
1040     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1041         switch(flow->nw_proto) {
1042         case IPPROTO_TCP:  /* TCP */
1043         case IPPROTO_UDP:  /* UDP */
1044             l4 = IPFIX_PROTO_L4_TCP_UDP;
1045             break;
1046         }
1047     }
1048
1049     flow_key->obs_domain_id = obs_domain_id;
1050     flow_key->template_id = ipfix_get_template_id(l2, l3, l4);
1051
1052     /* The fields defined in the ipfix_data_record_* structs and sent
1053      * below must match exactly the templates defined in
1054      * ipfix_define_template_fields. */
1055
1056     ethernet_header_length = (l2 == IPFIX_PROTO_L2_VLAN)
1057         ? VLAN_ETH_HEADER_LEN : ETH_HEADER_LEN;
1058     ethernet_total_length = packet->size;
1059
1060     /* Common Ethernet entities. */
1061     {
1062         struct ipfix_data_record_flow_key_common *data_common;
1063
1064         data_common = ofpbuf_put_zeros(&msg, sizeof *data_common);
1065         data_common->observation_point_id = htonl(obs_point_id);
1066         memcpy(data_common->source_mac_address, flow->dl_src,
1067                sizeof flow->dl_src);
1068         memcpy(data_common->destination_mac_address, flow->dl_dst,
1069                sizeof flow->dl_dst);
1070         data_common->ethernet_type = flow->dl_type;
1071         data_common->ethernet_header_length = ethernet_header_length;
1072     }
1073
1074     if (l2 == IPFIX_PROTO_L2_VLAN) {
1075         struct ipfix_data_record_flow_key_vlan *data_vlan;
1076         uint16_t vlan_id = vlan_tci_to_vid(flow->vlan_tci);
1077         uint8_t priority = vlan_tci_to_pcp(flow->vlan_tci);
1078
1079         data_vlan = ofpbuf_put_zeros(&msg, sizeof *data_vlan);
1080         data_vlan->vlan_id = htons(vlan_id);
1081         data_vlan->dot1q_vlan_id = htons(vlan_id);
1082         data_vlan->dot1q_priority = priority;
1083     }
1084
1085     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1086         struct ipfix_data_record_flow_key_ip *data_ip;
1087
1088         data_ip = ofpbuf_put_zeros(&msg, sizeof *data_ip);
1089         data_ip->ip_version = (l3 == IPFIX_PROTO_L3_IPV4) ? 4 : 6;
1090         data_ip->ip_ttl = flow->nw_ttl;
1091         data_ip->protocol_identifier = flow->nw_proto;
1092         data_ip->ip_diff_serv_code_point = flow->nw_tos >> 2;
1093         data_ip->ip_precedence = flow->nw_tos >> 5;
1094         data_ip->ip_class_of_service = flow->nw_tos;
1095
1096         if (l3 == IPFIX_PROTO_L3_IPV4) {
1097             struct ipfix_data_record_flow_key_ipv4 *data_ipv4;
1098             data_ipv4 = ofpbuf_put_zeros(&msg, sizeof *data_ipv4);
1099             data_ipv4->source_ipv4_address = flow->nw_src;
1100             data_ipv4->destination_ipv4_address = flow->nw_dst;
1101         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
1102             struct ipfix_data_record_flow_key_ipv6 *data_ipv6;
1103
1104             data_ipv6 = ofpbuf_put_zeros(&msg, sizeof *data_ipv6);
1105             memcpy(data_ipv6->source_ipv6_address, &flow->ipv6_src,
1106                    sizeof flow->ipv6_src);
1107             memcpy(data_ipv6->destination_ipv6_address, &flow->ipv6_dst,
1108                    sizeof flow->ipv6_dst);
1109             data_ipv6->flow_label_ipv6 = flow->ipv6_label;
1110         }
1111     }
1112
1113     if (l4 != IPFIX_PROTO_L4_UNKNOWN) {
1114         struct ipfix_data_record_flow_key_tcpudp *data_tcpudp;
1115
1116         data_tcpudp = ofpbuf_put_zeros(&msg, sizeof *data_tcpudp);
1117         data_tcpudp->source_transport_port = flow->tp_src;
1118         data_tcpudp->destination_transport_port = flow->tp_dst;
1119     }
1120
1121     flow_key->flow_key_msg_part_size = msg.size;
1122
1123     {
1124         struct timeval now;
1125         uint64_t layer2_octet_delta_count;
1126
1127         /* Calculate the total matched octet count by considering as
1128          * an approximation that all matched packets have the same
1129          * length. */
1130         layer2_octet_delta_count = packet_delta_count * ethernet_total_length;
1131
1132         xgettimeofday(&now);
1133         entry->flow_end_timestamp_usec = now.tv_usec + 1000000LL * now.tv_sec;
1134         entry->flow_start_timestamp_usec = entry->flow_end_timestamp_usec;
1135         entry->packet_delta_count = packet_delta_count;
1136         entry->layer2_octet_delta_count = layer2_octet_delta_count;
1137     }
1138
1139     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1140         uint16_t ip_total_length =
1141             ethernet_total_length - ethernet_header_length;
1142
1143         entry->octet_delta_sum_of_squares =
1144             packet_delta_count * ip_total_length * ip_total_length;
1145         entry->minimum_ip_total_length = ip_total_length;
1146         entry->maximum_ip_total_length = ip_total_length;
1147     } else {
1148         entry->octet_delta_sum_of_squares = 0;
1149         entry->minimum_ip_total_length = 0;
1150         entry->maximum_ip_total_length = 0;
1151     }
1152 }
1153
1154 /* Send each single data record in its own data set, to simplify the
1155  * implementation by avoiding having to group record by template ID
1156  * before sending. */
1157 static void
1158 ipfix_put_data_set(uint32_t export_time_sec,
1159                    struct ipfix_flow_cache_entry *entry,
1160                    enum ipfix_flow_end_reason flow_end_reason,
1161                    struct ofpbuf *msg)
1162 {
1163     size_t set_hdr_offset;
1164     struct ipfix_set_header *set_hdr;
1165
1166     set_hdr_offset = msg->size;
1167
1168     /* Put a Data Set. */
1169     set_hdr = ofpbuf_put_zeros(msg, sizeof *set_hdr);
1170     set_hdr->set_id = htons(entry->flow_key.template_id);
1171
1172     /* Copy the flow key part of the data record. */
1173
1174     ofpbuf_put(msg, entry->flow_key.flow_key_msg_part,
1175                entry->flow_key.flow_key_msg_part_size);
1176
1177     /* Put the non-key part of the data record. */
1178
1179     {
1180         struct ipfix_data_record_aggregated_common *data_aggregated_common;
1181         uint64_t export_time_usec, flow_start_delta_usec, flow_end_delta_usec;
1182
1183         /* Calculate the negative deltas relative to the export time
1184          * in seconds sent in the header, not the exact export
1185          * time. */
1186         export_time_usec = 1000000LL * export_time_sec;
1187         flow_start_delta_usec = export_time_usec
1188             - entry->flow_start_timestamp_usec;
1189         flow_end_delta_usec = export_time_usec
1190             - entry->flow_end_timestamp_usec;
1191
1192         data_aggregated_common = ofpbuf_put_zeros(
1193             msg, sizeof *data_aggregated_common);
1194         data_aggregated_common->flow_start_delta_microseconds = htonl(
1195             flow_start_delta_usec);
1196         data_aggregated_common->flow_end_delta_microseconds = htonl(
1197             flow_end_delta_usec);
1198         data_aggregated_common->packet_delta_count = htonll(
1199             entry->packet_delta_count);
1200         data_aggregated_common->layer2_octet_delta_count = htonll(
1201             entry->layer2_octet_delta_count);
1202         data_aggregated_common->flow_end_reason = flow_end_reason;
1203     }
1204
1205     if (entry->octet_delta_sum_of_squares) {  /* IP packet. */
1206         struct ipfix_data_record_aggregated_ip *data_aggregated_ip;
1207
1208         data_aggregated_ip = ofpbuf_put_zeros(
1209             msg, sizeof *data_aggregated_ip);
1210         data_aggregated_ip->octet_delta_sum_of_squares = htonll(
1211             entry->octet_delta_sum_of_squares);
1212         data_aggregated_ip->minimum_ip_total_length = htonll(
1213             entry->minimum_ip_total_length);
1214         data_aggregated_ip->maximum_ip_total_length = htonll(
1215             entry->maximum_ip_total_length);
1216     }
1217
1218     set_hdr = (struct ipfix_set_header*)((uint8_t*)msg->data + set_hdr_offset);
1219     set_hdr->length = htons(msg->size - set_hdr_offset);
1220 }
1221
1222 /* Send an IPFIX message with a single data record. */
1223 static void
1224 ipfix_send_data_msg(struct dpif_ipfix_exporter *exporter,
1225                     uint32_t export_time_sec,
1226                     struct ipfix_flow_cache_entry *entry,
1227                     enum ipfix_flow_end_reason flow_end_reason)
1228 {
1229     uint64_t msg_stub[DIV_ROUND_UP(MAX_MESSAGE_LEN, 8)];
1230     struct ofpbuf msg;
1231     ofpbuf_use_stub(&msg, msg_stub, sizeof msg_stub);
1232
1233     ipfix_init_header(export_time_sec, exporter->seq_number++,
1234                       entry->flow_key.obs_domain_id, &msg);
1235     ipfix_put_data_set(export_time_sec, entry, flow_end_reason, &msg);
1236     ipfix_send_msg(exporter->collectors, &msg);
1237
1238     ofpbuf_uninit(&msg);
1239 }
1240
1241 static void
1242 dpif_ipfix_sample(struct dpif_ipfix_exporter *exporter,
1243                   struct ofpbuf *packet, const struct flow *flow,
1244                   uint64_t packet_delta_count, uint32_t obs_domain_id,
1245                   uint32_t obs_point_id)
1246 {
1247     struct ipfix_flow_cache_entry *entry;
1248
1249     /* Create a flow cache entry from the sample. */
1250     entry = xmalloc(sizeof *entry);
1251     ipfix_cache_entry_init(entry, packet, flow, packet_delta_count,
1252                            obs_domain_id, obs_point_id);
1253     ipfix_cache_update(exporter, entry);
1254 }
1255
1256 void
1257 dpif_ipfix_bridge_sample(struct dpif_ipfix *di, struct ofpbuf *packet,
1258                          const struct flow *flow) OVS_EXCLUDED(mutex)
1259 {
1260     uint64_t packet_delta_count;
1261
1262     ovs_mutex_lock(&mutex);
1263     /* Use the sampling probability as an approximation of the number
1264      * of matched packets. */
1265     packet_delta_count = UINT32_MAX / di->bridge_exporter.probability;
1266     dpif_ipfix_sample(&di->bridge_exporter.exporter, packet, flow,
1267                       packet_delta_count,
1268                       di->bridge_exporter.options->obs_domain_id,
1269                       di->bridge_exporter.options->obs_point_id);
1270     ovs_mutex_unlock(&mutex);
1271 }
1272
1273 void
1274 dpif_ipfix_flow_sample(struct dpif_ipfix *di, struct ofpbuf *packet,
1275                        const struct flow *flow, uint32_t collector_set_id,
1276                        uint16_t probability, uint32_t obs_domain_id,
1277                        uint32_t obs_point_id) OVS_EXCLUDED(mutex)
1278 {
1279     struct dpif_ipfix_flow_exporter_map_node *node;
1280     /* Use the sampling probability as an approximation of the number
1281      * of matched packets. */
1282     uint64_t packet_delta_count = USHRT_MAX / probability;
1283
1284     ovs_mutex_lock(&mutex);
1285     node = dpif_ipfix_find_flow_exporter_map_node(di, collector_set_id);
1286     if (node) {
1287         dpif_ipfix_sample(&node->exporter.exporter, packet, flow,
1288                           packet_delta_count, obs_domain_id, obs_point_id);
1289     }
1290     ovs_mutex_unlock(&mutex);
1291 }
1292
1293 static void
1294 dpif_ipfix_cache_expire(struct dpif_ipfix_exporter *exporter,
1295                         bool forced_end, const uint64_t export_time_usec,
1296                         const uint32_t export_time_sec)
1297 {
1298     struct ipfix_flow_cache_entry *entry, *next_entry;
1299     uint64_t max_flow_start_timestamp_usec;
1300     bool template_msg_sent = false;
1301     enum ipfix_flow_end_reason flow_end_reason;
1302
1303     if (list_is_empty(&exporter->cache_flow_start_timestamp_list)) {
1304         return;
1305     }
1306
1307     max_flow_start_timestamp_usec = export_time_usec -
1308         1000000LL * exporter->cache_active_timeout;
1309
1310     LIST_FOR_EACH_SAFE (entry, next_entry, cache_flow_start_timestamp_list_node,
1311                         &exporter->cache_flow_start_timestamp_list) {
1312         if (forced_end) {
1313             flow_end_reason = FORCED_END;
1314         } else if (entry->flow_start_timestamp_usec
1315                    <= max_flow_start_timestamp_usec) {
1316             flow_end_reason = ACTIVE_TIMEOUT;
1317         } else if (hmap_count(&exporter->cache_flow_key_map)
1318                    > exporter->cache_max_flows) {
1319             /* Enforce exporter->cache_max_flows. */
1320             flow_end_reason = LACK_OF_RESOURCES;
1321         } else {
1322             /* Remaining flows haven't expired yet. */
1323             break;
1324         }
1325
1326         list_remove(&entry->cache_flow_start_timestamp_list_node);
1327         hmap_remove(&exporter->cache_flow_key_map,
1328                     &entry->flow_key_map_node);
1329
1330         if (!template_msg_sent
1331             && (exporter->last_template_set_time + IPFIX_TEMPLATE_INTERVAL)
1332                 <= export_time_sec) {
1333             ipfix_send_template_msg(exporter, export_time_sec,
1334                                     entry->flow_key.obs_domain_id);
1335             exporter->last_template_set_time = export_time_sec;
1336             template_msg_sent = true;
1337         }
1338
1339         /* XXX: Group multiple data records for the same obs domain id
1340          * into the same message. */
1341         ipfix_send_data_msg(exporter, export_time_sec, entry, flow_end_reason);
1342         free(entry);
1343     }
1344 }
1345
1346 static void
1347 get_export_time_now(uint64_t *export_time_usec, uint32_t *export_time_sec)
1348 {
1349     struct timeval export_time;
1350     xgettimeofday(&export_time);
1351
1352     *export_time_usec = export_time.tv_usec + 1000000LL * export_time.tv_sec;
1353
1354     /* The IPFIX start and end deltas are negative deltas relative to
1355      * the export time, so set the export time 1 second off to
1356      * calculate those deltas. */
1357     if (export_time.tv_usec == 0) {
1358         *export_time_sec = export_time.tv_sec;
1359     } else {
1360         *export_time_sec = export_time.tv_sec + 1;
1361     }
1362 }
1363
1364 static void
1365 dpif_ipfix_cache_expire_now(struct dpif_ipfix_exporter *exporter,
1366                             bool forced_end)
1367 {
1368     uint64_t export_time_usec;
1369     uint32_t export_time_sec;
1370
1371     get_export_time_now(&export_time_usec, &export_time_sec);
1372     dpif_ipfix_cache_expire(exporter, forced_end, export_time_usec,
1373                             export_time_sec);
1374 }
1375
1376 void
1377 dpif_ipfix_run(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
1378 {
1379     uint64_t export_time_usec;
1380     uint32_t export_time_sec;
1381     struct dpif_ipfix_flow_exporter_map_node *flow_exporter_node;
1382
1383     ovs_mutex_lock(&mutex);
1384     get_export_time_now(&export_time_usec, &export_time_sec);
1385     if (di->bridge_exporter.probability > 0) {  /* Bridge exporter enabled. */
1386       dpif_ipfix_cache_expire(
1387           &di->bridge_exporter.exporter, false, export_time_usec,
1388           export_time_sec);
1389     }
1390     HMAP_FOR_EACH (flow_exporter_node, node, &di->flow_exporter_map) {
1391         dpif_ipfix_cache_expire(
1392             &flow_exporter_node->exporter.exporter, false, export_time_usec,
1393             export_time_sec);
1394     }
1395     ovs_mutex_unlock(&mutex);
1396 }
1397
1398 void
1399 dpif_ipfix_wait(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
1400 {
1401     long long int next_timeout_msec = LLONG_MAX;
1402     struct dpif_ipfix_flow_exporter_map_node *flow_exporter_node;
1403
1404     ovs_mutex_lock(&mutex);
1405     if (di->bridge_exporter.probability > 0) {  /* Bridge exporter enabled. */
1406         if (ipfix_cache_next_timeout_msec(
1407                 &di->bridge_exporter.exporter, &next_timeout_msec)) {
1408             poll_timer_wait_until(next_timeout_msec);
1409         }
1410     }
1411     HMAP_FOR_EACH (flow_exporter_node, node, &di->flow_exporter_map) {
1412         if (ipfix_cache_next_timeout_msec(
1413                 &flow_exporter_node->exporter.exporter, &next_timeout_msec)) {
1414             poll_timer_wait_until(next_timeout_msec);
1415         }
1416     }
1417     ovs_mutex_unlock(&mutex);
1418 }