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 "byte-order.h"
  20 #include "collectors.h"
  21 #include "flow.h"
  22 #include "hash.h"
  23 #include "hmap.h"
  24 #include "ofpbuf.h"
  25 #include "ofproto.h"
  26 #include "packets.h"
  27 #include "sset.h"
  28 #include "util.h"
  29 #include "timeval.h"
  30 #include "util.h"
  31 #include "vlog.h"
  32
  33 VLOG_DEFINE_THIS_MODULE(ipfix);
  34
  35 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  36 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
  37
  38 /* Cf. IETF RFC 5101 Section 10.3.4. */
  39 #define IPFIX_DEFAULT_COLLECTOR_PORT 4739
  40
  41 struct dpif_ipfix_exporter {
  42     struct collectors *collectors;
  43     uint32_t seq_number;
  44     time_t last_template_set_time;
  45 };
  46
  47 struct dpif_ipfix_bridge_exporter {
  48     struct dpif_ipfix_exporter exporter;
  49     struct ofproto_ipfix_bridge_exporter_options *options;
  50     uint32_t probability;
  51 };
  52
  53 struct dpif_ipfix_flow_exporter {
  54     struct dpif_ipfix_exporter exporter;
  55     struct ofproto_ipfix_flow_exporter_options *options;
  56 };
  57
  58 struct dpif_ipfix_flow_exporter_map_node {
  59     struct hmap_node node;
  60     struct dpif_ipfix_flow_exporter exporter;
  61 };
  62
  63 struct dpif_ipfix {
  64     struct dpif_ipfix_bridge_exporter bridge_exporter;
  65     struct hmap flow_exporter_map;  /* dpif_ipfix_flow_exporter_map_nodes. */
  66     atomic_int ref_cnt;
  67 };
  68
  69 #define IPFIX_VERSION 0x000a
  70
  71 /* When using UDP, IPFIX Template Records must be re-sent regularly.
  72  * The standard default interval is 10 minutes (600 seconds).
  73  * Cf. IETF RFC 5101 Section 10.3.6. */
  74 #define IPFIX_TEMPLATE_INTERVAL 600
  75
  76 /* Cf. IETF RFC 5101 Section 3.1. */
  77 OVS_PACKED(
  78 struct ipfix_header {
  79     ovs_be16 version;  /* IPFIX_VERSION. */
  80     ovs_be16 length;  /* Length in bytes including this header. */
  81     ovs_be32 export_time;  /* Seconds since the epoch. */
  82     ovs_be32 seq_number;  /* Message sequence number. */
  83     ovs_be32 obs_domain_id;  /* Observation Domain ID. */
  84 });
  85 BUILD_ASSERT_DECL(sizeof(struct ipfix_header) == 16);
  86
  87 #define IPFIX_SET_ID_TEMPLATE 2
  88 #define IPFIX_SET_ID_OPTION_TEMPLATE 3
  89
  90 /* Cf. IETF RFC 5101 Section 3.3.2. */
  91 OVS_PACKED(
  92 struct ipfix_set_header {
  93     ovs_be16 set_id;  /* IPFIX_SET_ID_* or valid template ID for Data Sets. */
  94     ovs_be16 length;  /* Length of the set in bytes including header. */
  95 });
  96 BUILD_ASSERT_DECL(sizeof(struct ipfix_set_header) == 4);
  97
  98 /* Alternatives for templates at each layer.  A template is defined by
  99  * a combination of one value for each layer. */
 100 enum ipfix_proto_l2 {
 101     IPFIX_PROTO_L2_ETH = 0,  /* No VLAN. */
 102     IPFIX_PROTO_L2_VLAN,
 103     NUM_IPFIX_PROTO_L2
 104 };
 105 enum ipfix_proto_l3 {
 106     IPFIX_PROTO_L3_UNKNOWN = 0,
 107     IPFIX_PROTO_L3_IPV4,
 108     IPFIX_PROTO_L3_IPV6,
 109     NUM_IPFIX_PROTO_L3
 110 };
 111 enum ipfix_proto_l4 {
 112     IPFIX_PROTO_L4_UNKNOWN = 0,
 113     IPFIX_PROTO_L4_TCP_UDP,
 114     NUM_IPFIX_PROTO_L4
 115 };
 116
 117 /* Any Template ID > 255 is usable for Template Records. */
 118 #define IPFIX_TEMPLATE_ID_MIN 256
 119
 120 /* Cf. IETF RFC 5101 Section 3.4.1. */
 121 OVS_PACKED(
 122 struct ipfix_template_record_header {
 123     ovs_be16 template_id;
 124     ovs_be16 field_count;
 125 });
 126 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_record_header) == 4);
 127
 128 enum ipfix_entity_id {
 129 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_ID_##ENUM = ID,
 130 #include "ofproto/ipfix-entities.def"
 131 };
 132
 133 enum ipfix_entity_size {
 134 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_SIZE_##ENUM = SIZE,
 135 #include "ofproto/ipfix-entities.def"
 136 };
 137
 138 OVS_PACKED(
 139 struct ipfix_template_field_specifier {
 140     ovs_be16 element_id;  /* IPFIX_ENTITY_ID_*. */
 141     ovs_be16 field_length;  /* Length of the field's value, in bytes. */
 142     /* No Enterprise ID, since only standard element IDs are specified. */
 143 });
 144 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_field_specifier) == 4);
 145
 146 /* Part of data record for common metadata and Ethernet entities. */
 147 OVS_PACKED(
 148 struct ipfix_data_record_common {
 149     ovs_be32 observation_point_id;  /* OBSERVATION_POINT_ID */
 150     ovs_be64 packet_delta_count;  /* PACKET_DELTA_COUNT */
 151     ovs_be64 layer2_octet_delta_count;  /* LAYER2_OCTET_DELTA_COUNT */
 152     uint8_t source_mac_address[6];  /* SOURCE_MAC_ADDRESS */
 153     uint8_t destination_mac_address[6];  /* DESTINATION_MAC_ADDRESS */
 154     ovs_be16 ethernet_type;  /* ETHERNET_TYPE */
 155     ovs_be16 ethernet_total_length;  /* ETHERNET_TOTAL_LENGTH */
 156     uint8_t ethernet_header_length;  /* ETHERNET_HEADER_LENGTH */
 157 });
 158 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_common) == 37);
 159
 160 /* Part of data record for VLAN entities. */
 161 OVS_PACKED(
 162 struct ipfix_data_record_vlan {
 163     ovs_be16 vlan_id;  /* VLAN_ID */
 164     ovs_be16 dot1q_vlan_id;  /* DOT1Q_VLAN_ID */
 165     uint8_t dot1q_priority;  /* DOT1Q_PRIORITY */
 166 });
 167 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_vlan) == 5);
 168
 169 /* Part of data record for IP entities. */
 170 OVS_PACKED(
 171 struct ipfix_data_record_ip {
 172     uint8_t ip_version;  /* IP_VERSION */
 173     uint8_t ip_ttl;  /* IP_TTL */
 174     uint8_t protocol_identifier;  /* PROTOCOL_IDENTIFIER */
 175     uint8_t ip_diff_serv_code_point;  /* IP_DIFF_SERV_CODE_POINT */
 176     uint8_t ip_precedence;  /* IP_PRECEDENCE */
 177     uint8_t ip_class_of_service;  /* IP_CLASS_OF_SERVICE */
 178 });
 179 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_ip) == 6);
 180
 181 /* Part of data record for IPv4 entities. */
 182 OVS_PACKED(
 183 struct ipfix_data_record_ipv4 {
 184     ovs_be32 source_ipv4_address;  /* SOURCE_IPV4_ADDRESS */
 185     ovs_be32 destination_ipv4_address;  /* DESTINATION_IPV4_ADDRESS */
 186 });
 187 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_ipv4) == 8);
 188
 189 /* Part of data record for IPv4 entities. */
 190 OVS_PACKED(
 191 struct ipfix_data_record_ipv6 {
 192     uint8_t source_ipv6_address[16];  /* SOURCE_IPV6_ADDRESS */
 193     uint8_t destination_ipv6_address[16];  /* DESTINATION_IPV6_ADDRESS */
 194     ovs_be32 flow_label_ipv6;  /* FLOW_LABEL_IPV6 */
 195 });
 196 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_ipv6) == 36);
 197
 198 /* Part of data record for TCP/UDP entities. */
 199 OVS_PACKED(
 200 struct ipfix_data_record_tcpudp {
 201     ovs_be16 source_transport_port;  /* SOURCE_TRANSPORT_PORT */
 202     ovs_be16 destination_transport_port;  /* DESTINATION_TRANSPORT_PORT */
 203 });
 204 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_tcpudp) == 4);
 205
 206 static bool
 207 ofproto_ipfix_bridge_exporter_options_equal(
 208     const struct ofproto_ipfix_bridge_exporter_options *a,
 209     const struct ofproto_ipfix_bridge_exporter_options *b)
 210 {
 211     return (a->obs_domain_id == b->obs_domain_id
 212             && a->obs_point_id == b->obs_point_id
 213             && a->sampling_rate == b->sampling_rate
 214             && sset_equals(&a->targets, &b->targets));
 215 }
 216
 217 static struct ofproto_ipfix_bridge_exporter_options *
 218 ofproto_ipfix_bridge_exporter_options_clone(
 219     const struct ofproto_ipfix_bridge_exporter_options *old)
 220 {
 221     struct ofproto_ipfix_bridge_exporter_options *new =
 222         xmemdup(old, sizeof *old);
 223     sset_clone(&new->targets, &old->targets);
 224     return new;
 225 }
 226
 227 static void
 228 ofproto_ipfix_bridge_exporter_options_destroy(
 229     struct ofproto_ipfix_bridge_exporter_options *options)
 230 {
 231     if (options) {
 232         sset_destroy(&options->targets);
 233         free(options);
 234     }
 235 }
 236
 237 static bool
 238 ofproto_ipfix_flow_exporter_options_equal(
 239     const struct ofproto_ipfix_flow_exporter_options *a,
 240     const struct ofproto_ipfix_flow_exporter_options *b)
 241 {
 242     return (a->collector_set_id == b->collector_set_id
 243             && sset_equals(&a->targets, &b->targets));
 244 }
 245
 246 static struct ofproto_ipfix_flow_exporter_options *
 247 ofproto_ipfix_flow_exporter_options_clone(
 248     const struct ofproto_ipfix_flow_exporter_options *old)
 249 {
 250     struct ofproto_ipfix_flow_exporter_options *new =
 251         xmemdup(old, sizeof *old);
 252     sset_clone(&new->targets, &old->targets);
 253     return new;
 254 }
 255
 256 static void
 257 ofproto_ipfix_flow_exporter_options_destroy(
 258     struct ofproto_ipfix_flow_exporter_options *options)
 259 {
 260     if (options) {
 261         sset_destroy(&options->targets);
 262         free(options);
 263     }
 264 }
 265
 266 static void
 267 dpif_ipfix_exporter_clear(struct dpif_ipfix_exporter *exporter)
 268 {
 269     collectors_destroy(exporter->collectors);
 270     exporter->collectors = NULL;
 271     exporter->seq_number = 1;
 272     exporter->last_template_set_time = TIME_MIN;
 273 }
 274
 275 static bool
 276 dpif_ipfix_exporter_set_options(struct dpif_ipfix_exporter *exporter,
 277                                 const struct sset *targets)
 278 {
 279     collectors_destroy(exporter->collectors);
 280     collectors_create(targets, IPFIX_DEFAULT_COLLECTOR_PORT,
 281                       &exporter->collectors);
 282     if (exporter->collectors == NULL) {
 283         VLOG_WARN_RL(&rl, "no collectors could be initialized, "
 284                      "IPFIX exporter disabled");
 285         dpif_ipfix_exporter_clear(exporter);
 286         return false;
 287     }
 288     return true;
 289 }
 290
 291 static void
 292 dpif_ipfix_bridge_exporter_clear(struct dpif_ipfix_bridge_exporter *exporter)
 293 {
 294     dpif_ipfix_exporter_clear(&exporter->exporter);
 295     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 296     exporter->options = NULL;
 297     exporter->probability = 0;
 298 }
 299
 300 static void
 301 dpif_ipfix_bridge_exporter_set_options(
 302     struct dpif_ipfix_bridge_exporter *exporter,
 303     const struct ofproto_ipfix_bridge_exporter_options *options)
 304 {
 305     bool options_changed;
 306
 307     if (!options || sset_is_empty(&options->targets)) {
 308         /* No point in doing any work if there are no targets. */
 309         dpif_ipfix_bridge_exporter_clear(exporter);
 310         return;
 311     }
 312
 313     options_changed = (
 314         !exporter->options
 315         || !ofproto_ipfix_bridge_exporter_options_equal(
 316             options, exporter->options));
 317
 318     /* Configure collectors if options have changed or if we're
 319      * shortchanged in collectors (which indicates that opening one or
 320      * more of the configured collectors failed, so that we should
 321      * retry). */
 322     if (options_changed
 323         || collectors_count(exporter->exporter.collectors)
 324             < sset_count(&options->targets)) {
 325         if (!dpif_ipfix_exporter_set_options(&exporter->exporter,
 326                                              &options->targets)) {
 327             return;
 328         }
 329     }
 330
 331     /* Avoid reconfiguring if options didn't change. */
 332     if (!options_changed) {
 333         return;
 334     }
 335
 336     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 337     exporter->options = ofproto_ipfix_bridge_exporter_options_clone(options);
 338     exporter->probability =
 339         MAX(1, UINT32_MAX / exporter->options->sampling_rate);
 340 }
 341
 342 static struct dpif_ipfix_flow_exporter_map_node*
 343 dpif_ipfix_find_flow_exporter_map_node(
 344     const struct dpif_ipfix *di, const uint32_t collector_set_id)
 345 {
 346     struct dpif_ipfix_flow_exporter_map_node *exporter_node;
 347
 348     HMAP_FOR_EACH_WITH_HASH (exporter_node, node,
 349                              hash_int(collector_set_id, 0),
 350                              &di->flow_exporter_map) {
 351         if (exporter_node->exporter.options->collector_set_id
 352             == collector_set_id) {
 353             return exporter_node;
 354         }
 355     }
 356
 357     return NULL;
 358 }
 359
 360 static void
 361 dpif_ipfix_flow_exporter_clear(struct dpif_ipfix_flow_exporter *exporter)
 362 {
 363     dpif_ipfix_exporter_clear(&exporter->exporter);
 364     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 365     exporter->options = NULL;
 366 }
 367
 368 static bool
 369 dpif_ipfix_flow_exporter_set_options(
 370     struct dpif_ipfix_flow_exporter *exporter,
 371     const struct ofproto_ipfix_flow_exporter_options *options)
 372 {
 373     bool options_changed;
 374
 375     if (sset_is_empty(&options->targets)) {
 376         /* No point in doing any work if there are no targets. */
 377         dpif_ipfix_flow_exporter_clear(exporter);
 378         return true;
 379     }
 380
 381     options_changed = (
 382         !exporter->options
 383         || !ofproto_ipfix_flow_exporter_options_equal(
 384             options, exporter->options));
 385
 386     /* Configure collectors if options have changed or if we're
 387      * shortchanged in collectors (which indicates that opening one or
 388      * more of the configured collectors failed, so that we should
 389      * retry). */
 390     if (options_changed
 391         || collectors_count(exporter->exporter.collectors)
 392             < sset_count(&options->targets)) {
 393         if (!dpif_ipfix_exporter_set_options(&exporter->exporter,
 394                                              &options->targets)) {
 395             return false;
 396         }
 397     }
 398
 399     /* Avoid reconfiguring if options didn't change. */
 400     if (!options_changed) {
 401         return true;
 402     }
 403
 404     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 405     exporter->options = ofproto_ipfix_flow_exporter_options_clone(options);
 406
 407     return true;
 408 }
 409
 410 void
 411 dpif_ipfix_set_options(
 412     struct dpif_ipfix *di,
 413     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
 414     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
 415     size_t n_flow_exporters_options) OVS_EXCLUDED(mutex)
 416 {
 417     int i;
 418     struct ofproto_ipfix_flow_exporter_options *options;
 419     struct dpif_ipfix_flow_exporter_map_node *node, *next;
 420     size_t n_broken_flow_exporters_options = 0;
 421
 422     ovs_mutex_lock(&mutex);
 423     dpif_ipfix_bridge_exporter_set_options(&di->bridge_exporter,
 424                                            bridge_exporter_options);
 425
 426     /* Add new flow exporters and update current flow exporters. */
 427     options = (struct ofproto_ipfix_flow_exporter_options *)
 428         flow_exporters_options;
 429     for (i = 0; i < n_flow_exporters_options; i++) {
 430         node = dpif_ipfix_find_flow_exporter_map_node(
 431             di, options->collector_set_id);
 432         if (!node) {
 433             node = xzalloc(sizeof *node);
 434             dpif_ipfix_exporter_clear(&node->exporter.exporter);
 435             hmap_insert(&di->flow_exporter_map, &node->node,
 436                         hash_int(options->collector_set_id, 0));
 437         }
 438         if (!dpif_ipfix_flow_exporter_set_options(&node->exporter, options)) {
 439             n_broken_flow_exporters_options++;
 440         }
 441         options++;
 442     }
 443
 444     ovs_assert(hmap_count(&di->flow_exporter_map) >=
 445                (n_flow_exporters_options - n_broken_flow_exporters_options));
 446
 447     /* Remove dropped flow exporters, if any needs to be removed. */
 448     if (hmap_count(&di->flow_exporter_map) > n_flow_exporters_options) {
 449         HMAP_FOR_EACH_SAFE (node, next, node, &di->flow_exporter_map) {
 450             /* This is slow but doesn't take any extra memory, and
 451              * this table is not supposed to contain many rows anyway. */
 452             options = (struct ofproto_ipfix_flow_exporter_options *)
 453                 flow_exporters_options;
 454             for (i = 0; i < n_flow_exporters_options; i++) {
 455               if (node->exporter.options->collector_set_id
 456                   == options->collector_set_id) {
 457                   break;
 458               }
 459               options++;
 460             }
 461             if (i == n_flow_exporters_options) {  // Not found.
 462                 hmap_remove(&di->flow_exporter_map, &node->node);
 463                 dpif_ipfix_flow_exporter_clear(&node->exporter);
 464                 free(node);
 465             }
 466         }
 467     }
 468
 469     ovs_assert(hmap_count(&di->flow_exporter_map) ==
 470                (n_flow_exporters_options - n_broken_flow_exporters_options));
 471     ovs_mutex_unlock(&mutex);
 472 }
 473
 474 struct dpif_ipfix *
 475 dpif_ipfix_create(void)
 476 {
 477     struct dpif_ipfix *di;
 478     di = xzalloc(sizeof *di);
 479     dpif_ipfix_exporter_clear(&di->bridge_exporter.exporter);
 480     hmap_init(&di->flow_exporter_map);
 481     atomic_init(&di->ref_cnt, 1);
 482     return di;
 483 }
 484
 485 struct dpif_ipfix *
 486 dpif_ipfix_ref(const struct dpif_ipfix *di_)
 487 {
 488     struct dpif_ipfix *di = CONST_CAST(struct dpif_ipfix *, di_);
 489     if (di) {
 490         int orig;
 491         atomic_add(&di->ref_cnt, 1, &orig);
 492         ovs_assert(orig > 0);
 493     }
 494     return di;
 495 }
 496
 497 uint32_t
 498 dpif_ipfix_get_bridge_exporter_probability(const struct dpif_ipfix *di)
 499     OVS_EXCLUDED(mutex)
 500 {
 501     uint32_t ret;
 502     ovs_mutex_lock(&mutex);
 503     ret = di->bridge_exporter.probability;
 504     ovs_mutex_unlock(&mutex);
 505     return ret;
 506 }
 507
 508 static void
 509 dpif_ipfix_clear(struct dpif_ipfix *di) OVS_REQ_WRLOCK(mutex)
 510 {
 511     struct dpif_ipfix_flow_exporter_map_node *node, *next;
 512
 513     dpif_ipfix_bridge_exporter_clear(&di->bridge_exporter);
 514
 515     HMAP_FOR_EACH_SAFE (node, next, node, &di->flow_exporter_map) {
 516         hmap_remove(&di->flow_exporter_map, &node->node);
 517         dpif_ipfix_flow_exporter_clear(&node->exporter);
 518         free(node);
 519     }
 520 }
 521
 522 void
 523 dpif_ipfix_unref(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
 524 {
 525     int orig;
 526
 527     if (!di) {
 528         return;
 529     }
 530
 531     atomic_sub(&di->ref_cnt, 1, &orig);
 532     ovs_assert(orig > 0);
 533     if (orig == 1) {
 534         ovs_mutex_lock(&mutex);
 535         dpif_ipfix_clear(di);
 536         hmap_destroy(&di->flow_exporter_map);
 537         free(di);
 538         ovs_mutex_unlock(&mutex);
 539     }
 540 }
 541
 542 static void
 543 ipfix_init_header(uint32_t seq_number, uint32_t obs_domain_id,
 544                   struct ofpbuf *msg)
 545 {
 546     struct ipfix_header *hdr;
 547
 548     hdr = ofpbuf_put_zeros(msg, sizeof *hdr);
 549     hdr->version = htons(IPFIX_VERSION);
 550     hdr->length = htons(sizeof *hdr);  /* Updated in ipfix_send_msg. */
 551     hdr->export_time = htonl(time_wall());
 552     hdr->seq_number = htonl(seq_number);
 553     hdr->obs_domain_id = htonl(obs_domain_id);
 554 }
 555
 556 static void
 557 ipfix_send_msg(const struct collectors *collectors, struct ofpbuf *msg)
 558 {
 559     struct ipfix_header *hdr;
 560
 561     /* Adjust the length in the header. */
 562     hdr = msg->data;
 563     hdr->length = htons(msg->size);
 564
 565     collectors_send(collectors, msg->data, msg->size);
 566     msg->size = 0;
 567 }
 568
 569 static uint16_t
 570 ipfix_get_template_id(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
 571                       enum ipfix_proto_l4 l4)
 572 {
 573     uint16_t template_id;
 574     template_id = l2;
 575     template_id = template_id * NUM_IPFIX_PROTO_L3 + l3;
 576     template_id = template_id * NUM_IPFIX_PROTO_L4 + l4;
 577     return IPFIX_TEMPLATE_ID_MIN + template_id;
 578 }
 579
 580 static void
 581 ipfix_define_template_entity(enum ipfix_entity_id id,
 582                              enum ipfix_entity_size size, struct ofpbuf *msg)
 583 {
 584     struct ipfix_template_field_specifier *field;
 585
 586     field = ofpbuf_put_zeros(msg, sizeof *field);
 587     field->element_id = htons(id);
 588     field->field_length = htons(size);
 589 }
 590
 591 static uint16_t
 592 ipfix_define_template_fields(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
 593                              enum ipfix_proto_l4 l4, struct ofpbuf *msg)
 594 {
 595     uint16_t count = 0;
 596
 597 #define DEF(ID) \
 598     { \
 599         ipfix_define_template_entity(IPFIX_ENTITY_ID_##ID, \
 600                                      IPFIX_ENTITY_SIZE_##ID, msg); \
 601         count++; \
 602     }
 603
 604     DEF(OBSERVATION_POINT_ID);
 605     DEF(PACKET_DELTA_COUNT);
 606     DEF(LAYER2_OCTET_DELTA_COUNT);
 607
 608     /* Common Ethernet entities. */
 609     DEF(SOURCE_MAC_ADDRESS);
 610     DEF(DESTINATION_MAC_ADDRESS);
 611     DEF(ETHERNET_TYPE);
 612     DEF(ETHERNET_TOTAL_LENGTH);
 613     DEF(ETHERNET_HEADER_LENGTH);
 614
 615     if (l2 == IPFIX_PROTO_L2_VLAN) {
 616         DEF(VLAN_ID);
 617         DEF(DOT1Q_VLAN_ID);
 618         DEF(DOT1Q_PRIORITY);
 619     }
 620
 621     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
 622         DEF(IP_VERSION);
 623         DEF(IP_TTL);
 624         DEF(PROTOCOL_IDENTIFIER);
 625         DEF(IP_DIFF_SERV_CODE_POINT);
 626         DEF(IP_PRECEDENCE);
 627         DEF(IP_CLASS_OF_SERVICE);
 628
 629         if (l3 == IPFIX_PROTO_L3_IPV4) {
 630             DEF(SOURCE_IPV4_ADDRESS);
 631             DEF(DESTINATION_IPV4_ADDRESS);
 632         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
 633             DEF(SOURCE_IPV6_ADDRESS);
 634             DEF(DESTINATION_IPV6_ADDRESS);
 635             DEF(FLOW_LABEL_IPV6);
 636         }
 637     }
 638
 639     if (l4 != IPFIX_PROTO_L4_UNKNOWN) {
 640         DEF(SOURCE_TRANSPORT_PORT);
 641         DEF(DESTINATION_TRANSPORT_PORT);
 642     }
 643
 644 #undef DEF
 645
 646     return count;
 647 }
 648
 649 static void
 650 ipfix_send_template_msg(struct dpif_ipfix_exporter *exporter,
 651                         uint32_t obs_domain_id)
 652 {
 653     uint64_t msg_stub[DIV_ROUND_UP(1500, 8)];
 654     struct ofpbuf msg;
 655     size_t set_hdr_offset, tmpl_hdr_offset;
 656     struct ipfix_set_header *set_hdr;
 657     struct ipfix_template_record_header *tmpl_hdr;
 658     uint16_t field_count;
 659     enum ipfix_proto_l2 l2;
 660     enum ipfix_proto_l3 l3;
 661     enum ipfix_proto_l4 l4;
 662
 663     ofpbuf_use_stub(&msg, msg_stub, sizeof msg_stub);
 664
 665     ipfix_init_header(exporter->seq_number, obs_domain_id, &msg);
 666     set_hdr_offset = msg.size;
 667
 668     /* Add a Template Set. */
 669     set_hdr = ofpbuf_put_zeros(&msg, sizeof *set_hdr);
 670     set_hdr->set_id = htons(IPFIX_SET_ID_TEMPLATE);
 671
 672     /* Define one template for each possible combination of
 673      * protocols. */
 674     for (l2 = 0; l2 < NUM_IPFIX_PROTO_L2; l2++) {
 675         for (l3 = 0; l3 < NUM_IPFIX_PROTO_L3; l3++) {
 676             for (l4 = 0; l4 < NUM_IPFIX_PROTO_L4; l4++) {
 677                 if (l3 == IPFIX_PROTO_L3_UNKNOWN &&
 678                     l4 != IPFIX_PROTO_L4_UNKNOWN) {
 679                     continue;
 680                 }
 681                 tmpl_hdr_offset = msg.size;
 682                 tmpl_hdr = ofpbuf_put_zeros(&msg, sizeof *tmpl_hdr);
 683                 tmpl_hdr->template_id = htons(
 684                     ipfix_get_template_id(l2, l3, l4));
 685                 field_count = ipfix_define_template_fields(l2, l3, l4, &msg);
 686                 tmpl_hdr = (struct ipfix_template_record_header*)
 687                     ((uint8_t*)msg.data + tmpl_hdr_offset);
 688                 tmpl_hdr->field_count = htons(field_count);
 689             }
 690         }
 691     }
 692
 693     set_hdr = (struct ipfix_set_header*)((uint8_t*)msg.data + set_hdr_offset);
 694     set_hdr->length = htons(msg.size - set_hdr_offset);
 695
 696     /* XXX: Add Options Template Sets, at least to define a Flow Keys
 697      * Option Template. */
 698
 699     ipfix_send_msg(exporter->collectors, &msg);
 700
 701     ofpbuf_uninit(&msg);
 702 }
 703
 704 static void
 705 ipfix_send_data_msg(struct dpif_ipfix_exporter *exporter, struct ofpbuf *packet,
 706                     const struct flow *flow, uint64_t packet_delta_count,
 707                     uint32_t obs_domain_id, uint32_t obs_point_id)
 708 {
 709     uint64_t msg_stub[DIV_ROUND_UP(1500, 8)];
 710     struct ofpbuf msg;
 711     size_t set_hdr_offset;
 712     struct ipfix_set_header *set_hdr;
 713     enum ipfix_proto_l2 l2;
 714     enum ipfix_proto_l3 l3;
 715     enum ipfix_proto_l4 l4;
 716
 717     ofpbuf_use_stub(&msg, msg_stub, sizeof msg_stub);
 718
 719     ipfix_init_header(exporter->seq_number, obs_domain_id, &msg);
 720     exporter->seq_number++;
 721     set_hdr_offset = msg.size;
 722
 723     /* Choose the right template ID matching the protocols in the
 724      * sampled packet. */
 725     l2 = (flow->vlan_tci == 0) ? IPFIX_PROTO_L2_ETH : IPFIX_PROTO_L2_VLAN;
 726
 727     switch(ntohs(flow->dl_type)) {
 728     case ETH_TYPE_IP:
 729         l3 = IPFIX_PROTO_L3_IPV4;
 730         break;
 731     case ETH_TYPE_IPV6:
 732         l3 = IPFIX_PROTO_L3_IPV6;
 733         break;
 734     default:
 735         l3 = IPFIX_PROTO_L3_UNKNOWN;
 736     }
 737
 738     l4 = IPFIX_PROTO_L4_UNKNOWN;
 739     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
 740         switch(flow->nw_proto) {
 741         case IPPROTO_TCP:  /* TCP */
 742         case IPPROTO_UDP:  /* UDP */
 743             l4 = IPFIX_PROTO_L4_TCP_UDP;
 744             break;
 745         }
 746     }
 747
 748     /* Add a Data Set. */
 749     set_hdr = ofpbuf_put_zeros(&msg, sizeof *set_hdr);
 750     set_hdr->set_id = htons(ipfix_get_template_id(l2, l3, l4));
 751
 752     /* The fields defined in the ipfix_data_record_* structs and sent
 753      * below must match exactly the templates defined in
 754      * ipfix_define_template_fields. */
 755
 756     /* Common Ethernet entities. */
 757     {
 758         struct ipfix_data_record_common *data_common;
 759         uint16_t ethernet_total_length;
 760         uint8_t ethernet_header_length;
 761         uint64_t layer2_octet_delta_count;
 762
 763         ethernet_total_length = packet->size;
 764         ethernet_header_length = (l2 == IPFIX_PROTO_L2_VLAN)
 765             ? VLAN_ETH_HEADER_LEN : ETH_HEADER_LEN;
 766
 767         /* Calculate the total matched octet count by considering as
 768          * an approximation that all matched packets have the same
 769          * length. */
 770         layer2_octet_delta_count = packet_delta_count * ethernet_total_length;
 771
 772         data_common = ofpbuf_put_zeros(&msg, sizeof *data_common);
 773         data_common->observation_point_id = htonl(obs_point_id);
 774         data_common->packet_delta_count = htonll(packet_delta_count);
 775         data_common->layer2_octet_delta_count =
 776             htonll(layer2_octet_delta_count);
 777         memcpy(data_common->source_mac_address, flow->dl_src,
 778                sizeof flow->dl_src);
 779         memcpy(data_common->destination_mac_address, flow->dl_dst,
 780                sizeof flow->dl_dst);
 781         data_common->ethernet_type = flow->dl_type;
 782         data_common->ethernet_total_length = htons(ethernet_total_length);
 783         data_common->ethernet_header_length = ethernet_header_length;
 784     }
 785
 786     if (l2 == IPFIX_PROTO_L2_VLAN) {
 787         struct ipfix_data_record_vlan *data_vlan;
 788         uint16_t vlan_id = vlan_tci_to_vid(flow->vlan_tci);
 789         uint8_t priority = vlan_tci_to_pcp(flow->vlan_tci);
 790
 791         data_vlan = ofpbuf_put_zeros(&msg, sizeof *data_vlan);
 792         data_vlan->vlan_id = htons(vlan_id);
 793         data_vlan->dot1q_vlan_id = htons(vlan_id);
 794         data_vlan->dot1q_priority = priority;
 795     }
 796
 797     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
 798         struct ipfix_data_record_ip *data_ip;
 799
 800         data_ip = ofpbuf_put_zeros(&msg, sizeof *data_ip);
 801         data_ip->ip_version = (l3 == IPFIX_PROTO_L3_IPV4) ? 4 : 6;
 802         data_ip->ip_ttl = flow->nw_ttl;
 803         data_ip->protocol_identifier = flow->nw_proto;
 804         data_ip->ip_diff_serv_code_point = flow->nw_tos >> 2;
 805         data_ip->ip_precedence = flow->nw_tos >> 5;
 806         data_ip->ip_class_of_service = flow->nw_tos;
 807
 808         if (l3 == IPFIX_PROTO_L3_IPV4) {
 809             struct ipfix_data_record_ipv4 *data_ipv4;
 810             data_ipv4 = ofpbuf_put_zeros(&msg, sizeof *data_ipv4);
 811             data_ipv4->source_ipv4_address = flow->nw_src;
 812             data_ipv4->destination_ipv4_address = flow->nw_dst;
 813         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
 814             struct ipfix_data_record_ipv6 *data_ipv6;
 815
 816             data_ipv6 = ofpbuf_put_zeros(&msg, sizeof *data_ipv6);
 817             memcpy(data_ipv6->source_ipv6_address, &flow->ipv6_src,
 818                    sizeof flow->ipv6_src);
 819             memcpy(data_ipv6->destination_ipv6_address, &flow->ipv6_dst,
 820                    sizeof flow->ipv6_dst);
 821             data_ipv6->flow_label_ipv6 = flow->ipv6_label;
 822         }
 823     }
 824
 825     if (l4 != IPFIX_PROTO_L4_UNKNOWN) {
 826         struct ipfix_data_record_tcpudp *data_tcpudp;
 827
 828         data_tcpudp = ofpbuf_put_zeros(&msg, sizeof *data_tcpudp);
 829         data_tcpudp->source_transport_port = flow->tp_src;
 830         data_tcpudp->destination_transport_port = flow->tp_dst;
 831     }
 832
 833     set_hdr = (struct ipfix_set_header*)((uint8_t*)msg.data + set_hdr_offset);
 834     set_hdr->length = htons(msg.size - set_hdr_offset);
 835
 836     ipfix_send_msg(exporter->collectors, &msg);
 837
 838     ofpbuf_uninit(&msg);
 839 }
 840
 841 static void
 842 dpif_ipfix_sample(struct dpif_ipfix_exporter *exporter,
 843                   struct ofpbuf *packet, const struct flow *flow,
 844                   uint64_t packet_delta_count, uint32_t obs_domain_id,
 845                   uint32_t obs_point_id)
 846 {
 847     time_t now = time_wall();
 848     if ((exporter->last_template_set_time + IPFIX_TEMPLATE_INTERVAL) <= now) {
 849         ipfix_send_template_msg(exporter, obs_domain_id);
 850         exporter->last_template_set_time = now;
 851     }
 852
 853     ipfix_send_data_msg(exporter, packet, flow, packet_delta_count,
 854                         obs_domain_id, obs_point_id);
 855 }
 856
 857 void
 858 dpif_ipfix_bridge_sample(struct dpif_ipfix *di, struct ofpbuf *packet,
 859                          const struct flow *flow) OVS_EXCLUDED(mutex)
 860 {
 861     uint64_t packet_delta_count;
 862
 863     ovs_mutex_lock(&mutex);
 864     /* Use the sampling probability as an approximation of the number
 865      * of matched packets. */
 866     packet_delta_count = UINT32_MAX / di->bridge_exporter.probability;
 867     dpif_ipfix_sample(&di->bridge_exporter.exporter, packet, flow,
 868                       packet_delta_count,
 869                       di->bridge_exporter.options->obs_domain_id,
 870                       di->bridge_exporter.options->obs_point_id);
 871     ovs_mutex_unlock(&mutex);
 872 }
 873
 874 void
 875 dpif_ipfix_flow_sample(struct dpif_ipfix *di, struct ofpbuf *packet,
 876                        const struct flow *flow, uint32_t collector_set_id,
 877                        uint16_t probability, uint32_t obs_domain_id,
 878                        uint32_t obs_point_id) OVS_EXCLUDED(mutex)
 879 {
 880     struct dpif_ipfix_flow_exporter_map_node *node;
 881     /* Use the sampling probability as an approximation of the number
 882      * of matched packets. */
 883     uint64_t packet_delta_count = USHRT_MAX / probability;
 884
 885     ovs_mutex_lock(&mutex);
 886     node = dpif_ipfix_find_flow_exporter_map_node(di, collector_set_id);
 887     if (node) {
 888         dpif_ipfix_sample(&node->exporter.exporter, packet, flow,
 889                           packet_delta_count, obs_domain_id, obs_point_id);
 890     }
 891     ovs_mutex_unlock(&mutex);
 892 }