ofproto/ofproto-dpif-ipfix.c

   1 /*
   2  * Copyright (c) 2012, 2013, 2014, 2015 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 "ofproto-dpif.h"
  29 #include "dp-packet.h"
  30 #include "packets.h"
  31 #include "poll-loop.h"
  32 #include "sset.h"
  33 #include "util.h"
  34 #include "timeval.h"
  35 #include "util.h"
  36 #include "openvswitch/vlog.h"
  37
  38 VLOG_DEFINE_THIS_MODULE(ipfix);
  39
  40 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
  41 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
  42
  43 /* Cf. IETF RFC 5101 Section 10.3.4. */
  44 #define IPFIX_DEFAULT_COLLECTOR_PORT 4739
  45
  46 /* Cf. IETF RFC 5881 Setion 8. */
  47 #define BFD_CONTROL_DEST_PORT        3784
  48 #define BFD_ECHO_DEST_PORT           3785
  49
  50 /* The standard layer2SegmentId (ID 351) element is included in vDS to send
  51  * the VxLAN tunnel's VNI. It is 64-bit long, the most significant byte is
  52  * used to indicate the type of tunnel (0x01 = VxLAN, 0x02 = GRE) and the three
  53  * least significant bytes hold the value of the layer 2 overlay network
  54  * segment identifier: a 24-bit VxLAN tunnel's VNI or a 24-bit GRE tunnel's
  55  * TNI. This is not compatible with STT, as implemented in OVS, as
  56  * its tunnel IDs is 64-bit.
  57  *
  58  * Two new enterprise information elements are defined which are similar to
  59  * laryerSegmentId but support 64-bit IDs:
  60  *     tunnelType (ID 891) and tunnelKey (ID 892).
  61  *
  62  * The enum dpif_ipfix_tunnel_type is to declare the types supported in the
  63  * tunnelType element.
  64  * The number of ipfix tunnel types includes two reserverd types: 0x04 and 0x06.
  65  */
  66 enum dpif_ipfix_tunnel_type {
  67     DPIF_IPFIX_TUNNEL_UNKNOWN = 0x00,
  68     DPIF_IPFIX_TUNNEL_VXLAN = 0x01,
  69     DPIF_IPFIX_TUNNEL_GRE = 0x02,
  70     DPIF_IPFIX_TUNNEL_LISP = 0x03,
  71     DPIF_IPFIX_TUNNEL_STT = 0x04,
  72     DPIF_IPFIX_TUNNEL_IPSEC_GRE = 0x05,
  73     DPIF_IPFIX_TUNNEL_GENEVE = 0x07,
  74     NUM_DPIF_IPFIX_TUNNEL
  75 };
  76
  77 struct dpif_ipfix_port {
  78     struct hmap_node hmap_node; /* In struct dpif_ipfix's "tunnel_ports" hmap. */
  79     struct ofport *ofport;      /* To retrieve port stats. */
  80     odp_port_t odp_port;
  81     enum dpif_ipfix_tunnel_type tunnel_type;
  82     uint8_t tunnel_key_length;
  83 };
  84
  85 struct dpif_ipfix_exporter {
  86     struct collectors *collectors;
  87     uint32_t seq_number;
  88     time_t last_template_set_time;
  89     struct hmap cache_flow_key_map;  /* ipfix_flow_cache_entry. */
  90     struct ovs_list cache_flow_start_timestamp_list;  /* ipfix_flow_cache_entry. */
  91     uint32_t cache_active_timeout;  /* In seconds. */
  92     uint32_t cache_max_flows;
  93 };
  94
  95 struct dpif_ipfix_bridge_exporter {
  96     struct dpif_ipfix_exporter exporter;
  97     struct ofproto_ipfix_bridge_exporter_options *options;
  98     uint32_t probability;
  99 };
 100
 101 struct dpif_ipfix_flow_exporter {
 102     struct dpif_ipfix_exporter exporter;
 103     struct ofproto_ipfix_flow_exporter_options *options;
 104 };
 105
 106 struct dpif_ipfix_flow_exporter_map_node {
 107     struct hmap_node node;
 108     struct dpif_ipfix_flow_exporter exporter;
 109 };
 110
 111 struct dpif_ipfix {
 112     struct dpif_ipfix_bridge_exporter bridge_exporter;
 113     struct hmap flow_exporter_map;  /* dpif_ipfix_flow_exporter_map_node. */
 114     struct hmap tunnel_ports;       /* Contains "struct dpif_ipfix_port"s.
 115                                      * It makes tunnel port lookups faster in
 116                                      * sampling upcalls. */
 117     struct ovs_refcount ref_cnt;
 118 };
 119
 120 #define IPFIX_VERSION 0x000a
 121
 122 /* When using UDP, IPFIX Template Records must be re-sent regularly.
 123  * The standard default interval is 10 minutes (600 seconds).
 124  * Cf. IETF RFC 5101 Section 10.3.6. */
 125 #define IPFIX_TEMPLATE_INTERVAL 600
 126
 127 /* Cf. IETF RFC 5101 Section 3.1. */
 128 OVS_PACKED(
 129 struct ipfix_header {
 130     ovs_be16 version;  /* IPFIX_VERSION. */
 131     ovs_be16 length;  /* Length in bytes including this header. */
 132     ovs_be32 export_time;  /* Seconds since the epoch. */
 133     ovs_be32 seq_number;  /* Message sequence number. */
 134     ovs_be32 obs_domain_id;  /* Observation Domain ID. */
 135 });
 136 BUILD_ASSERT_DECL(sizeof(struct ipfix_header) == 16);
 137
 138 #define IPFIX_SET_ID_TEMPLATE 2
 139 #define IPFIX_SET_ID_OPTION_TEMPLATE 3
 140
 141 /* Cf. IETF RFC 5101 Section 3.3.2. */
 142 OVS_PACKED(
 143 struct ipfix_set_header {
 144     ovs_be16 set_id;  /* IPFIX_SET_ID_* or valid template ID for Data Sets. */
 145     ovs_be16 length;  /* Length of the set in bytes including header. */
 146 });
 147 BUILD_ASSERT_DECL(sizeof(struct ipfix_set_header) == 4);
 148
 149 /* Alternatives for templates at each layer.  A template is defined by
 150  * a combination of one value for each layer. */
 151 enum ipfix_proto_l2 {
 152     IPFIX_PROTO_L2_ETH = 0,  /* No VLAN. */
 153     IPFIX_PROTO_L2_VLAN,
 154     NUM_IPFIX_PROTO_L2
 155 };
 156 enum ipfix_proto_l3 {
 157     IPFIX_PROTO_L3_UNKNOWN = 0,
 158     IPFIX_PROTO_L3_IPV4,
 159     IPFIX_PROTO_L3_IPV6,
 160     NUM_IPFIX_PROTO_L3
 161 };
 162 enum ipfix_proto_l4 {
 163     IPFIX_PROTO_L4_UNKNOWN = 0,
 164     IPFIX_PROTO_L4_TCP_UDP_SCTP,
 165     IPFIX_PROTO_L4_ICMP,
 166     NUM_IPFIX_PROTO_L4
 167 };
 168 enum ipfix_proto_tunnel {
 169     IPFIX_PROTO_NOT_TUNNELED = 0,
 170     IPFIX_PROTO_TUNNELED,  /* Support gre, lisp and vxlan. */
 171     NUM_IPFIX_PROTO_TUNNEL
 172 };
 173
 174 /* Any Template ID > 255 is usable for Template Records. */
 175 #define IPFIX_TEMPLATE_ID_MIN 256
 176
 177 /* Cf. IETF RFC 5101 Section 3.4.1. */
 178 OVS_PACKED(
 179 struct ipfix_template_record_header {
 180     ovs_be16 template_id;
 181     ovs_be16 field_count;
 182 });
 183 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_record_header) == 4);
 184
 185 enum ipfix_entity_id {
 186 /* standard IPFIX elements */
 187 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_ID_##ENUM = ID,
 188 #include "ofproto/ipfix-entities.def"
 189 /* non-standard IPFIX elements */
 190 #define IPFIX_SET_ENTERPRISE(v) (((v) | 0x8000))
 191 #define IPFIX_ENTERPRISE_ENTITY(ENUM, ID, SIZE, NAME, ENTERPRISE) \
 192     IPFIX_ENTITY_ID_##ENUM = IPFIX_SET_ENTERPRISE(ID),
 193 #include "ofproto/ipfix-enterprise-entities.def"
 194 };
 195
 196 enum ipfix_entity_size {
 197 /* standard IPFIX elements */
 198 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_SIZE_##ENUM = SIZE,
 199 #include "ofproto/ipfix-entities.def"
 200 /* non-standard IPFIX elements */
 201 #define IPFIX_ENTERPRISE_ENTITY(ENUM, ID, SIZE, NAME, ENTERPRISE) \
 202     IPFIX_ENTITY_SIZE_##ENUM = SIZE,
 203 #include "ofproto/ipfix-enterprise-entities.def"
 204 };
 205
 206 enum ipfix_entity_enterprise {
 207 /* standard IPFIX elements */
 208 #define IPFIX_ENTITY(ENUM, ID, SIZE, NAME)  IPFIX_ENTITY_ENTERPRISE_##ENUM = 0,
 209 #include "ofproto/ipfix-entities.def"
 210 /* non-standard IPFIX elements */
 211 #define IPFIX_ENTERPRISE_ENTITY(ENUM, ID, SIZE, NAME, ENTERPRISE) \
 212     IPFIX_ENTITY_ENTERPRISE_##ENUM = ENTERPRISE,
 213 #include "ofproto/ipfix-enterprise-entities.def"
 214 };
 215
 216 OVS_PACKED(
 217 struct ipfix_template_field_specifier {
 218     ovs_be16 element_id;  /* IPFIX_ENTITY_ID_*. */
 219     ovs_be16 field_length;  /* Length of the field's value, in bytes.
 220                              * For Variable-Length element, it should be 65535.
 221                              */
 222     ovs_be32 enterprise;  /* Enterprise number */
 223 });
 224 BUILD_ASSERT_DECL(sizeof(struct ipfix_template_field_specifier) == 8);
 225
 226 /* Cf. IETF RFC 5102 Section 5.11.6. */
 227 enum ipfix_flow_direction {
 228     INGRESS_FLOW = 0x00,
 229     EGRESS_FLOW = 0x01
 230 };
 231
 232 /* Part of data record flow key for common metadata and Ethernet entities. */
 233 OVS_PACKED(
 234 struct ipfix_data_record_flow_key_common {
 235     ovs_be32 observation_point_id;  /* OBSERVATION_POINT_ID */
 236     uint8_t flow_direction;  /* FLOW_DIRECTION */
 237     struct eth_addr source_mac_address; /* SOURCE_MAC_ADDRESS */
 238     struct eth_addr destination_mac_address; /* DESTINATION_MAC_ADDRESS */
 239     ovs_be16 ethernet_type;  /* ETHERNET_TYPE */
 240     uint8_t ethernet_header_length;  /* ETHERNET_HEADER_LENGTH */
 241 });
 242 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_common) == 20);
 243
 244 /* Part of data record flow key for VLAN entities. */
 245 OVS_PACKED(
 246 struct ipfix_data_record_flow_key_vlan {
 247     ovs_be16 vlan_id;  /* VLAN_ID */
 248     ovs_be16 dot1q_vlan_id;  /* DOT1Q_VLAN_ID */
 249     uint8_t dot1q_priority;  /* DOT1Q_PRIORITY */
 250 });
 251 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_vlan) == 5);
 252
 253 /* Part of data record flow key for IP entities. */
 254 /* XXX: Replace IP_TTL with MINIMUM_TTL and MAXIMUM_TTL? */
 255 OVS_PACKED(
 256 struct ipfix_data_record_flow_key_ip {
 257     uint8_t ip_version;  /* IP_VERSION */
 258     uint8_t ip_ttl;  /* IP_TTL */
 259     uint8_t protocol_identifier;  /* PROTOCOL_IDENTIFIER */
 260     uint8_t ip_diff_serv_code_point;  /* IP_DIFF_SERV_CODE_POINT */
 261     uint8_t ip_precedence;  /* IP_PRECEDENCE */
 262     uint8_t ip_class_of_service;  /* IP_CLASS_OF_SERVICE */
 263 });
 264 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ip) == 6);
 265
 266 /* Part of data record flow key for IPv4 entities. */
 267 OVS_PACKED(
 268 struct ipfix_data_record_flow_key_ipv4 {
 269     ovs_be32 source_ipv4_address;  /* SOURCE_IPV4_ADDRESS */
 270     ovs_be32 destination_ipv4_address;  /* DESTINATION_IPV4_ADDRESS */
 271 });
 272 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ipv4) == 8);
 273
 274 /* Part of data record flow key for IPv6 entities. */
 275 OVS_PACKED(
 276 struct ipfix_data_record_flow_key_ipv6 {
 277     uint8_t source_ipv6_address[16];  /* SOURCE_IPV6_ADDRESS */
 278     uint8_t destination_ipv6_address[16];  /* DESTINATION_IPV6_ADDRESS */
 279     ovs_be32 flow_label_ipv6;  /* FLOW_LABEL_IPV6 */
 280 });
 281 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_ipv6) == 36);
 282
 283 /* Part of data record flow key for TCP/UDP/SCTP entities. */
 284 OVS_PACKED(
 285 struct ipfix_data_record_flow_key_transport {
 286     ovs_be16 source_transport_port;  /* SOURCE_TRANSPORT_PORT */
 287     ovs_be16 destination_transport_port;  /* DESTINATION_TRANSPORT_PORT */
 288 });
 289 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_transport) == 4);
 290
 291 /* Part of data record flow key for ICMP entities. */
 292 OVS_PACKED(
 293 struct ipfix_data_record_flow_key_icmp {
 294     uint8_t icmp_type;  /* ICMP_TYPE_IPV4 / ICMP_TYPE_IPV6 */
 295     uint8_t icmp_code;  /* ICMP_CODE_IPV4 / ICMP_CODE_IPV6 */
 296 });
 297 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_icmp) == 2);
 298
 299 /* For the tunnel type that is on the top of IPSec, the protocol identifier
 300  * of the upper tunnel type is used.
 301  */
 302 static uint8_t tunnel_protocol[NUM_DPIF_IPFIX_TUNNEL] = {
 303     0,              /* reserved */
 304     IPPROTO_UDP,    /* DPIF_IPFIX_TUNNEL_VXLAN */
 305     IPPROTO_GRE,    /* DPIF_IPFIX_TUNNEL_GRE */
 306     IPPROTO_UDP,    /* DPIF_IPFIX_TUNNEL_LISP*/
 307     IPPROTO_TCP,    /* DPIF_IPFIX_TUNNEL_STT*/
 308     IPPROTO_GRE,    /* DPIF_IPFIX_TUNNEL_IPSEC_GRE */
 309     0          ,    /* reserved */
 310     IPPROTO_UDP,    /* DPIF_IPFIX_TUNNEL_GENEVE*/
 311 };
 312
 313 OVS_PACKED(
 314 struct ipfix_data_record_flow_key_tunnel {
 315     ovs_be32 tunnel_source_ipv4_address;  /* TUNNEL_SOURCE_IPV4_ADDRESS */
 316     ovs_be32 tunnel_destination_ipv4_address;  /* TUNNEL_DESTINATION_IPV4_ADDRESS */
 317     uint8_t tunnel_protocol_identifier;  /* TUNNEL_PROTOCOL_IDENTIFIER */
 318     ovs_be16 tunnel_source_transport_port;  /* TUNNEL_SOURCE_TRANSPORT_PORT */
 319     ovs_be16 tunnel_destination_transport_port;  /* TUNNEL_DESTINATION_TRANSPORT_PORT */
 320     uint8_t tunnel_type;  /* TUNNEL_TYPE */
 321     uint8_t tunnel_key_length;  /* length of TUNNEL_KEY */
 322     uint8_t tunnel_key[];  /* data of  TUNNEL_KEY */
 323 });
 324 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_flow_key_tunnel) == 15);
 325
 326 /* Cf. IETF RFC 5102 Section 5.11.3. */
 327 enum ipfix_flow_end_reason {
 328     IDLE_TIMEOUT = 0x01,
 329     ACTIVE_TIMEOUT = 0x02,
 330     END_OF_FLOW_DETECTED = 0x03,
 331     FORCED_END = 0x04,
 332     LACK_OF_RESOURCES = 0x05
 333 };
 334
 335 /* Part of data record for common aggregated elements. */
 336 OVS_PACKED(
 337 struct ipfix_data_record_aggregated_common {
 338     ovs_be32 flow_start_delta_microseconds; /* FLOW_START_DELTA_MICROSECONDS */
 339     ovs_be32 flow_end_delta_microseconds; /* FLOW_END_DELTA_MICROSECONDS */
 340     ovs_be64 packet_delta_count;  /* PACKET_DELTA_COUNT */
 341     ovs_be64 layer2_octet_delta_count;  /* LAYER2_OCTET_DELTA_COUNT */
 342     uint8_t flow_end_reason;  /* FLOW_END_REASON */
 343 });
 344 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_aggregated_common) == 25);
 345
 346 /* Part of data record for IP aggregated elements. */
 347 OVS_PACKED(
 348 struct ipfix_data_record_aggregated_ip {
 349     ovs_be64 octet_delta_count;  /* OCTET_DELTA_COUNT */
 350     ovs_be64 octet_delta_sum_of_squares;  /* OCTET_DELTA_SUM_OF_SQUARES */
 351     ovs_be64 minimum_ip_total_length;  /* MINIMUM_IP_TOTAL_LENGTH */
 352     ovs_be64 maximum_ip_total_length;  /* MAXIMUM_IP_TOTAL_LENGTH */
 353 });
 354 BUILD_ASSERT_DECL(sizeof(struct ipfix_data_record_aggregated_ip) == 32);
 355
 356 /*
 357  * support tunnel key for:
 358  * VxLAN: 24-bit VIN,
 359  * GRE: 32-bit key,
 360  * LISP: 24-bit instance ID
 361  * STT: 64-bit key
 362  */
 363 #define MAX_TUNNEL_KEY_LEN 8
 364
 365 #define MAX_FLOW_KEY_LEN                                        \
 366     (sizeof(struct ipfix_data_record_flow_key_common)           \
 367      + sizeof(struct ipfix_data_record_flow_key_vlan)           \
 368      + sizeof(struct ipfix_data_record_flow_key_ip)             \
 369      + MAX(sizeof(struct ipfix_data_record_flow_key_ipv4),      \
 370            sizeof(struct ipfix_data_record_flow_key_ipv6))      \
 371      + MAX(sizeof(struct ipfix_data_record_flow_key_icmp),      \
 372            sizeof(struct ipfix_data_record_flow_key_transport)) \
 373      + sizeof(struct ipfix_data_record_flow_key_tunnel)         \
 374      + MAX_TUNNEL_KEY_LEN)
 375
 376 #define MAX_DATA_RECORD_LEN                                 \
 377     (MAX_FLOW_KEY_LEN                                       \
 378      + sizeof(struct ipfix_data_record_aggregated_common)   \
 379      + sizeof(struct ipfix_data_record_aggregated_ip))
 380
 381 /* Max length of a data set.  To simplify the implementation, each
 382  * data record is sent in a separate data set, so each data set
 383  * contains at most one data record. */
 384 #define MAX_DATA_SET_LEN             \
 385     (sizeof(struct ipfix_set_header) \
 386      + MAX_DATA_RECORD_LEN)
 387
 388 /* Max length of an IPFIX message. Arbitrarily set to accommodate low
 389  * MTU. */
 390 #define MAX_MESSAGE_LEN 1024
 391
 392 /* Cache structures. */
 393
 394 /* Flow key. */
 395 struct ipfix_flow_key {
 396     uint32_t obs_domain_id;
 397     uint16_t template_id;
 398     size_t flow_key_msg_part_size;
 399     uint64_t flow_key_msg_part[DIV_ROUND_UP(MAX_FLOW_KEY_LEN, 8)];
 400 };
 401
 402 /* Flow cache entry. */
 403 struct ipfix_flow_cache_entry {
 404     struct hmap_node flow_key_map_node;
 405     struct ovs_list cache_flow_start_timestamp_list_node;
 406     struct ipfix_flow_key flow_key;
 407     /* Common aggregated elements. */
 408     uint64_t flow_start_timestamp_usec;
 409     uint64_t flow_end_timestamp_usec;
 410     uint64_t packet_delta_count;
 411     uint64_t layer2_octet_delta_count;
 412     uint64_t octet_delta_count;
 413     uint64_t octet_delta_sum_of_squares;  /* 0 if not IP. */
 414     uint16_t minimum_ip_total_length;  /* 0 if not IP. */
 415     uint16_t maximum_ip_total_length;  /* 0 if not IP. */
 416 };
 417
 418 static void dpif_ipfix_cache_expire(struct dpif_ipfix_exporter *, bool,
 419                                     const uint64_t, const uint32_t);
 420
 421 static void get_export_time_now(uint64_t *, uint32_t *);
 422
 423 static void dpif_ipfix_cache_expire_now(struct dpif_ipfix_exporter *, bool);
 424
 425 static bool
 426 ofproto_ipfix_bridge_exporter_options_equal(
 427     const struct ofproto_ipfix_bridge_exporter_options *a,
 428     const struct ofproto_ipfix_bridge_exporter_options *b)
 429 {
 430     return (a->obs_domain_id == b->obs_domain_id
 431             && a->obs_point_id == b->obs_point_id
 432             && a->sampling_rate == b->sampling_rate
 433             && a->cache_active_timeout == b->cache_active_timeout
 434             && a->cache_max_flows == b->cache_max_flows
 435             && a->enable_tunnel_sampling == b->enable_tunnel_sampling
 436             && a->enable_input_sampling == b->enable_input_sampling
 437             && a->enable_output_sampling == b->enable_output_sampling
 438             && sset_equals(&a->targets, &b->targets));
 439 }
 440
 441 static struct ofproto_ipfix_bridge_exporter_options *
 442 ofproto_ipfix_bridge_exporter_options_clone(
 443     const struct ofproto_ipfix_bridge_exporter_options *old)
 444 {
 445     struct ofproto_ipfix_bridge_exporter_options *new =
 446         xmemdup(old, sizeof *old);
 447     sset_clone(&new->targets, &old->targets);
 448     return new;
 449 }
 450
 451 static void
 452 ofproto_ipfix_bridge_exporter_options_destroy(
 453     struct ofproto_ipfix_bridge_exporter_options *options)
 454 {
 455     if (options) {
 456         sset_destroy(&options->targets);
 457         free(options);
 458     }
 459 }
 460
 461 static bool
 462 ofproto_ipfix_flow_exporter_options_equal(
 463     const struct ofproto_ipfix_flow_exporter_options *a,
 464     const struct ofproto_ipfix_flow_exporter_options *b)
 465 {
 466     return (a->collector_set_id == b->collector_set_id
 467             && a->cache_active_timeout == b->cache_active_timeout
 468             && a->cache_max_flows == b->cache_max_flows
 469             && sset_equals(&a->targets, &b->targets));
 470 }
 471
 472 static struct ofproto_ipfix_flow_exporter_options *
 473 ofproto_ipfix_flow_exporter_options_clone(
 474     const struct ofproto_ipfix_flow_exporter_options *old)
 475 {
 476     struct ofproto_ipfix_flow_exporter_options *new =
 477         xmemdup(old, sizeof *old);
 478     sset_clone(&new->targets, &old->targets);
 479     return new;
 480 }
 481
 482 static void
 483 ofproto_ipfix_flow_exporter_options_destroy(
 484     struct ofproto_ipfix_flow_exporter_options *options)
 485 {
 486     if (options) {
 487         sset_destroy(&options->targets);
 488         free(options);
 489     }
 490 }
 491
 492 static void
 493 dpif_ipfix_exporter_init(struct dpif_ipfix_exporter *exporter)
 494 {
 495     exporter->collectors = NULL;
 496     exporter->seq_number = 1;
 497     exporter->last_template_set_time = TIME_MIN;
 498     hmap_init(&exporter->cache_flow_key_map);
 499     list_init(&exporter->cache_flow_start_timestamp_list);
 500     exporter->cache_active_timeout = 0;
 501     exporter->cache_max_flows = 0;
 502 }
 503
 504 static void
 505 dpif_ipfix_exporter_clear(struct dpif_ipfix_exporter *exporter)
 506 {
 507     /* Flush the cache with flow end reason "forced end." */
 508     dpif_ipfix_cache_expire_now(exporter, true);
 509
 510     collectors_destroy(exporter->collectors);
 511     exporter->collectors = NULL;
 512     exporter->seq_number = 1;
 513     exporter->last_template_set_time = TIME_MIN;
 514     exporter->cache_active_timeout = 0;
 515     exporter->cache_max_flows = 0;
 516 }
 517
 518 static void
 519 dpif_ipfix_exporter_destroy(struct dpif_ipfix_exporter *exporter)
 520 {
 521     dpif_ipfix_exporter_clear(exporter);
 522     hmap_destroy(&exporter->cache_flow_key_map);
 523 }
 524
 525 static bool
 526 dpif_ipfix_exporter_set_options(struct dpif_ipfix_exporter *exporter,
 527                                 const struct sset *targets,
 528                                 const uint32_t cache_active_timeout,
 529                                 const uint32_t cache_max_flows)
 530 {
 531     collectors_destroy(exporter->collectors);
 532     collectors_create(targets, IPFIX_DEFAULT_COLLECTOR_PORT,
 533                       &exporter->collectors);
 534     if (exporter->collectors == NULL) {
 535         VLOG_WARN_RL(&rl, "no collectors could be initialized, "
 536                      "IPFIX exporter disabled");
 537         dpif_ipfix_exporter_clear(exporter);
 538         return false;
 539     }
 540     exporter->cache_active_timeout = cache_active_timeout;
 541     exporter->cache_max_flows = cache_max_flows;
 542     return true;
 543 }
 544
 545 static struct dpif_ipfix_port *
 546 dpif_ipfix_find_port(const struct dpif_ipfix *di,
 547                      odp_port_t odp_port) OVS_REQUIRES(mutex)
 548 {
 549     struct dpif_ipfix_port *dip;
 550
 551     HMAP_FOR_EACH_IN_BUCKET (dip, hmap_node, hash_odp_port(odp_port),
 552                              &di->tunnel_ports) {
 553         if (dip->odp_port == odp_port) {
 554             return dip;
 555         }
 556     }
 557     return NULL;
 558 }
 559
 560 static void
 561 dpif_ipfix_del_port(struct dpif_ipfix *di,
 562                       struct dpif_ipfix_port *dip)
 563     OVS_REQUIRES(mutex)
 564 {
 565     hmap_remove(&di->tunnel_ports, &dip->hmap_node);
 566     free(dip);
 567 }
 568
 569 void
 570 dpif_ipfix_add_tunnel_port(struct dpif_ipfix *di, struct ofport *ofport,
 571                            odp_port_t odp_port) OVS_EXCLUDED(mutex)
 572 {
 573     struct dpif_ipfix_port *dip;
 574     const char *type;
 575
 576     ovs_mutex_lock(&mutex);
 577     dip = dpif_ipfix_find_port(di, odp_port);
 578     if (dip) {
 579         dpif_ipfix_del_port(di, dip);
 580     }
 581
 582     type = netdev_get_type(ofport->netdev);
 583     if (type == NULL) {
 584         goto out;
 585     }
 586
 587     /* Add to table of tunnel ports. */
 588     dip = xmalloc(sizeof *dip);
 589     dip->ofport = ofport;
 590     dip->odp_port = odp_port;
 591     if (strcmp(type, "gre") == 0) {
 592         /* 32-bit key gre */
 593         dip->tunnel_type = DPIF_IPFIX_TUNNEL_GRE;
 594         dip->tunnel_key_length = 4;
 595     } else if (strcmp(type, "ipsec_gre") == 0) {
 596         /* 32-bit key ipsec_gre */
 597         dip->tunnel_type = DPIF_IPFIX_TUNNEL_IPSEC_GRE;
 598         dip->tunnel_key_length = 4;
 599     } else if (strcmp(type, "vxlan") == 0) {
 600         dip->tunnel_type = DPIF_IPFIX_TUNNEL_VXLAN;
 601         dip->tunnel_key_length = 3;
 602     } else if (strcmp(type, "lisp") == 0) {
 603         dip->tunnel_type = DPIF_IPFIX_TUNNEL_LISP;
 604         dip->tunnel_key_length = 3;
 605     } else if (strcmp(type, "geneve") == 0) {
 606         dip->tunnel_type = DPIF_IPFIX_TUNNEL_GENEVE;
 607         dip->tunnel_key_length = 3;
 608     } else if (strcmp(type, "stt") == 0) {
 609         dip->tunnel_type = DPIF_IPFIX_TUNNEL_STT;
 610         dip->tunnel_key_length = 8;
 611     } else {
 612         free(dip);
 613         goto out;
 614     }
 615     hmap_insert(&di->tunnel_ports, &dip->hmap_node, hash_odp_port(odp_port));
 616
 617 out:
 618     ovs_mutex_unlock(&mutex);
 619 }
 620
 621 void
 622 dpif_ipfix_del_tunnel_port(struct dpif_ipfix *di, odp_port_t odp_port)
 623     OVS_EXCLUDED(mutex)
 624 {
 625     struct dpif_ipfix_port *dip;
 626     ovs_mutex_lock(&mutex);
 627     dip = dpif_ipfix_find_port(di, odp_port);
 628     if (dip) {
 629         dpif_ipfix_del_port(di, dip);
 630     }
 631     ovs_mutex_unlock(&mutex);
 632 }
 633
 634 bool
 635 dpif_ipfix_get_tunnel_port(const struct dpif_ipfix *di, odp_port_t odp_port)
 636     OVS_EXCLUDED(mutex)
 637 {
 638     struct dpif_ipfix_port *dip;
 639     ovs_mutex_lock(&mutex);
 640     dip = dpif_ipfix_find_port(di, odp_port);
 641     ovs_mutex_unlock(&mutex);
 642     return dip != NULL;
 643 }
 644
 645 static void
 646 dpif_ipfix_bridge_exporter_init(struct dpif_ipfix_bridge_exporter *exporter)
 647 {
 648     dpif_ipfix_exporter_init(&exporter->exporter);
 649     exporter->options = NULL;
 650     exporter->probability = 0;
 651 }
 652
 653 static void
 654 dpif_ipfix_bridge_exporter_clear(struct dpif_ipfix_bridge_exporter *exporter)
 655 {
 656     dpif_ipfix_exporter_clear(&exporter->exporter);
 657     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 658     exporter->options = NULL;
 659     exporter->probability = 0;
 660 }
 661
 662 static void
 663 dpif_ipfix_bridge_exporter_destroy(struct dpif_ipfix_bridge_exporter *exporter)
 664 {
 665     dpif_ipfix_bridge_exporter_clear(exporter);
 666     dpif_ipfix_exporter_destroy(&exporter->exporter);
 667 }
 668
 669 static void
 670 dpif_ipfix_bridge_exporter_set_options(
 671     struct dpif_ipfix_bridge_exporter *exporter,
 672     const struct ofproto_ipfix_bridge_exporter_options *options)
 673 {
 674     bool options_changed;
 675
 676     if (!options || sset_is_empty(&options->targets)) {
 677         /* No point in doing any work if there are no targets. */
 678         dpif_ipfix_bridge_exporter_clear(exporter);
 679         return;
 680     }
 681
 682     options_changed = (
 683         !exporter->options
 684         || !ofproto_ipfix_bridge_exporter_options_equal(
 685             options, exporter->options));
 686
 687     /* Configure collectors if options have changed or if we're
 688      * shortchanged in collectors (which indicates that opening one or
 689      * more of the configured collectors failed, so that we should
 690      * retry). */
 691     if (options_changed
 692         || collectors_count(exporter->exporter.collectors)
 693             < sset_count(&options->targets)) {
 694         if (!dpif_ipfix_exporter_set_options(
 695                 &exporter->exporter, &options->targets,
 696                 options->cache_active_timeout, options->cache_max_flows)) {
 697             return;
 698         }
 699     }
 700
 701     /* Avoid reconfiguring if options didn't change. */
 702     if (!options_changed) {
 703         return;
 704     }
 705
 706     ofproto_ipfix_bridge_exporter_options_destroy(exporter->options);
 707     exporter->options = ofproto_ipfix_bridge_exporter_options_clone(options);
 708     exporter->probability =
 709         MAX(1, UINT32_MAX / exporter->options->sampling_rate);
 710
 711     /* Run over the cache as some entries might have expired after
 712      * changing the timeouts. */
 713     dpif_ipfix_cache_expire_now(&exporter->exporter, false);
 714 }
 715
 716 static struct dpif_ipfix_flow_exporter_map_node*
 717 dpif_ipfix_find_flow_exporter_map_node(
 718     const struct dpif_ipfix *di, const uint32_t collector_set_id)
 719     OVS_REQUIRES(mutex)
 720 {
 721     struct dpif_ipfix_flow_exporter_map_node *exporter_node;
 722
 723     HMAP_FOR_EACH_WITH_HASH (exporter_node, node,
 724                              hash_int(collector_set_id, 0),
 725                              &di->flow_exporter_map) {
 726         if (exporter_node->exporter.options->collector_set_id
 727             == collector_set_id) {
 728             return exporter_node;
 729         }
 730     }
 731
 732     return NULL;
 733 }
 734
 735 static void
 736 dpif_ipfix_flow_exporter_init(struct dpif_ipfix_flow_exporter *exporter)
 737 {
 738     dpif_ipfix_exporter_init(&exporter->exporter);
 739     exporter->options = NULL;
 740 }
 741
 742 static void
 743 dpif_ipfix_flow_exporter_clear(struct dpif_ipfix_flow_exporter *exporter)
 744 {
 745     dpif_ipfix_exporter_clear(&exporter->exporter);
 746     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 747     exporter->options = NULL;
 748 }
 749
 750 static void
 751 dpif_ipfix_flow_exporter_destroy(struct dpif_ipfix_flow_exporter *exporter)
 752 {
 753     dpif_ipfix_flow_exporter_clear(exporter);
 754     dpif_ipfix_exporter_destroy(&exporter->exporter);
 755 }
 756
 757 static bool
 758 dpif_ipfix_flow_exporter_set_options(
 759     struct dpif_ipfix_flow_exporter *exporter,
 760     const struct ofproto_ipfix_flow_exporter_options *options)
 761 {
 762     bool options_changed;
 763
 764     if (sset_is_empty(&options->targets)) {
 765         /* No point in doing any work if there are no targets. */
 766         dpif_ipfix_flow_exporter_clear(exporter);
 767         return true;
 768     }
 769
 770     options_changed = (
 771         !exporter->options
 772         || !ofproto_ipfix_flow_exporter_options_equal(
 773             options, exporter->options));
 774
 775     /* Configure collectors if options have changed or if we're
 776      * shortchanged in collectors (which indicates that opening one or
 777      * more of the configured collectors failed, so that we should
 778      * retry). */
 779     if (options_changed
 780         || collectors_count(exporter->exporter.collectors)
 781             < sset_count(&options->targets)) {
 782         if (!dpif_ipfix_exporter_set_options(
 783                 &exporter->exporter, &options->targets,
 784                 options->cache_active_timeout, options->cache_max_flows)) {
 785             return false;
 786         }
 787     }
 788
 789     /* Avoid reconfiguring if options didn't change. */
 790     if (!options_changed) {
 791         return true;
 792     }
 793
 794     ofproto_ipfix_flow_exporter_options_destroy(exporter->options);
 795     exporter->options = ofproto_ipfix_flow_exporter_options_clone(options);
 796
 797     /* Run over the cache as some entries might have expired after
 798      * changing the timeouts. */
 799     dpif_ipfix_cache_expire_now(&exporter->exporter, false);
 800
 801     return true;
 802 }
 803
 804 void
 805 dpif_ipfix_set_options(
 806     struct dpif_ipfix *di,
 807     const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
 808     const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
 809     size_t n_flow_exporters_options) OVS_EXCLUDED(mutex)
 810 {
 811     int i;
 812     struct ofproto_ipfix_flow_exporter_options *options;
 813     struct dpif_ipfix_flow_exporter_map_node *node, *next;
 814     size_t n_broken_flow_exporters_options = 0;
 815
 816     ovs_mutex_lock(&mutex);
 817     dpif_ipfix_bridge_exporter_set_options(&di->bridge_exporter,
 818                                            bridge_exporter_options);
 819
 820     /* Add new flow exporters and update current flow exporters. */
 821     options = (struct ofproto_ipfix_flow_exporter_options *)
 822         flow_exporters_options;
 823     for (i = 0; i < n_flow_exporters_options; i++) {
 824         node = dpif_ipfix_find_flow_exporter_map_node(
 825             di, options->collector_set_id);
 826         if (!node) {
 827             node = xzalloc(sizeof *node);
 828             dpif_ipfix_flow_exporter_init(&node->exporter);
 829             hmap_insert(&di->flow_exporter_map, &node->node,
 830                         hash_int(options->collector_set_id, 0));
 831         }
 832         if (!dpif_ipfix_flow_exporter_set_options(&node->exporter, options)) {
 833             n_broken_flow_exporters_options++;
 834         }
 835         options++;
 836     }
 837
 838     ovs_assert(hmap_count(&di->flow_exporter_map) >=
 839                (n_flow_exporters_options - n_broken_flow_exporters_options));
 840
 841     /* Remove dropped flow exporters, if any needs to be removed. */
 842     if (hmap_count(&di->flow_exporter_map) > n_flow_exporters_options) {
 843         HMAP_FOR_EACH_SAFE (node, next, node, &di->flow_exporter_map) {
 844             /* This is slow but doesn't take any extra memory, and
 845              * this table is not supposed to contain many rows anyway. */
 846             options = (struct ofproto_ipfix_flow_exporter_options *)
 847                 flow_exporters_options;
 848             for (i = 0; i < n_flow_exporters_options; i++) {
 849               if (node->exporter.options->collector_set_id
 850                   == options->collector_set_id) {
 851                   break;
 852               }
 853               options++;
 854             }
 855             if (i == n_flow_exporters_options) {  // Not found.
 856                 hmap_remove(&di->flow_exporter_map, &node->node);
 857                 dpif_ipfix_flow_exporter_destroy(&node->exporter);
 858                 free(node);
 859             }
 860         }
 861     }
 862
 863     ovs_assert(hmap_count(&di->flow_exporter_map) ==
 864                (n_flow_exporters_options - n_broken_flow_exporters_options));
 865     ovs_mutex_unlock(&mutex);
 866 }
 867
 868 struct dpif_ipfix *
 869 dpif_ipfix_create(void)
 870 {
 871     struct dpif_ipfix *di;
 872     di = xzalloc(sizeof *di);
 873     dpif_ipfix_bridge_exporter_init(&di->bridge_exporter);
 874     hmap_init(&di->flow_exporter_map);
 875     hmap_init(&di->tunnel_ports);
 876     ovs_refcount_init(&di->ref_cnt);
 877     return di;
 878 }
 879
 880 struct dpif_ipfix *
 881 dpif_ipfix_ref(const struct dpif_ipfix *di_)
 882 {
 883     struct dpif_ipfix *di = CONST_CAST(struct dpif_ipfix *, di_);
 884     if (di) {
 885         ovs_refcount_ref(&di->ref_cnt);
 886     }
 887     return di;
 888 }
 889
 890 uint32_t
 891 dpif_ipfix_get_bridge_exporter_probability(const struct dpif_ipfix *di)
 892     OVS_EXCLUDED(mutex)
 893 {
 894     uint32_t ret;
 895     ovs_mutex_lock(&mutex);
 896     ret = di->bridge_exporter.probability;
 897     ovs_mutex_unlock(&mutex);
 898     return ret;
 899 }
 900
 901 bool
 902 dpif_ipfix_get_bridge_exporter_input_sampling(const struct dpif_ipfix *di)
 903     OVS_EXCLUDED(mutex)
 904 {
 905     bool ret = true;
 906     ovs_mutex_lock(&mutex);
 907     if (di->bridge_exporter.options) {
 908         ret = di->bridge_exporter.options->enable_input_sampling;
 909     }
 910     ovs_mutex_unlock(&mutex);
 911     return ret;
 912 }
 913
 914 bool
 915 dpif_ipfix_get_bridge_exporter_output_sampling(const struct dpif_ipfix *di)
 916     OVS_EXCLUDED(mutex)
 917 {
 918     bool ret = true;
 919     ovs_mutex_lock(&mutex);
 920     if (di->bridge_exporter.options) {
 921         ret = di->bridge_exporter.options->enable_output_sampling;
 922     }
 923     ovs_mutex_unlock(&mutex);
 924     return ret;
 925 }
 926
 927 bool
 928 dpif_ipfix_get_bridge_exporter_tunnel_sampling(const struct dpif_ipfix *di)
 929     OVS_EXCLUDED(mutex)
 930 {
 931     bool ret = false;
 932     ovs_mutex_lock(&mutex);
 933     if (di->bridge_exporter.options) {
 934         ret = di->bridge_exporter.options->enable_tunnel_sampling;
 935     }
 936     ovs_mutex_unlock(&mutex);
 937     return ret;
 938 }
 939
 940 static void
 941 dpif_ipfix_clear(struct dpif_ipfix *di) OVS_REQUIRES(mutex)
 942 {
 943     struct dpif_ipfix_flow_exporter_map_node *exp_node, *exp_next;
 944     struct dpif_ipfix_port *dip, *next;
 945
 946     dpif_ipfix_bridge_exporter_clear(&di->bridge_exporter);
 947
 948     HMAP_FOR_EACH_SAFE (exp_node, exp_next, node, &di->flow_exporter_map) {
 949         hmap_remove(&di->flow_exporter_map, &exp_node->node);
 950         dpif_ipfix_flow_exporter_destroy(&exp_node->exporter);
 951         free(exp_node);
 952     }
 953
 954     HMAP_FOR_EACH_SAFE (dip, next, hmap_node, &di->tunnel_ports) {
 955         dpif_ipfix_del_port(di, dip);
 956     }
 957 }
 958
 959 void
 960 dpif_ipfix_unref(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
 961 {
 962     if (di && ovs_refcount_unref_relaxed(&di->ref_cnt) == 1) {
 963         ovs_mutex_lock(&mutex);
 964         dpif_ipfix_clear(di);
 965         dpif_ipfix_bridge_exporter_destroy(&di->bridge_exporter);
 966         hmap_destroy(&di->flow_exporter_map);
 967         hmap_destroy(&di->tunnel_ports);
 968         free(di);
 969         ovs_mutex_unlock(&mutex);
 970     }
 971 }
 972
 973 static void
 974 ipfix_init_header(uint32_t export_time_sec, uint32_t seq_number,
 975                   uint32_t obs_domain_id, struct dp_packet *msg)
 976 {
 977     struct ipfix_header *hdr;
 978
 979     hdr = dp_packet_put_zeros(msg, sizeof *hdr);
 980     hdr->version = htons(IPFIX_VERSION);
 981     hdr->length = htons(sizeof *hdr);  /* Updated in ipfix_send_msg. */
 982     hdr->export_time = htonl(export_time_sec);
 983     hdr->seq_number = htonl(seq_number);
 984     hdr->obs_domain_id = htonl(obs_domain_id);
 985 }
 986
 987 static void
 988 ipfix_send_msg(const struct collectors *collectors, struct dp_packet *msg)
 989 {
 990     struct ipfix_header *hdr;
 991
 992     /* Adjust the length in the header. */
 993     hdr = dp_packet_data(msg);
 994     hdr->length = htons(dp_packet_size(msg));
 995
 996     collectors_send(collectors, dp_packet_data(msg), dp_packet_size(msg));
 997     dp_packet_set_size(msg, 0);
 998 }
 999
1000 static uint16_t
1001 ipfix_get_template_id(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
1002                       enum ipfix_proto_l4 l4, enum ipfix_proto_tunnel tunnel)
1003 {
1004     uint16_t template_id;
1005     template_id = l2;
1006     template_id = template_id * NUM_IPFIX_PROTO_L3 + l3;
1007     template_id = template_id * NUM_IPFIX_PROTO_L4 + l4;
1008     template_id = template_id * NUM_IPFIX_PROTO_TUNNEL + tunnel;
1009     return IPFIX_TEMPLATE_ID_MIN + template_id;
1010 }
1011
1012 static void
1013 ipfix_define_template_entity(enum ipfix_entity_id id,
1014                              enum ipfix_entity_size size,
1015                              enum ipfix_entity_enterprise enterprise,
1016                              struct dp_packet *msg)
1017 {
1018     struct ipfix_template_field_specifier *field;
1019     size_t field_size;
1020
1021     if (enterprise) {
1022         field_size = sizeof *field;
1023     } else {
1024         /* No enterprise number */
1025         field_size = sizeof *field - sizeof(ovs_be32);
1026     }
1027     field = dp_packet_put_zeros(msg, field_size);
1028     field->element_id = htons(id);
1029     if (size) {
1030         field->field_length = htons(size);
1031     } else {
1032         /* RFC 5101, Section 7. Variable-Length Information Element */
1033         field->field_length = OVS_BE16_MAX;
1034     }
1035     if (enterprise) {
1036         field->enterprise = htonl(enterprise);
1037     }
1038
1039 }
1040
1041 static uint16_t
1042 ipfix_define_template_fields(enum ipfix_proto_l2 l2, enum ipfix_proto_l3 l3,
1043                              enum ipfix_proto_l4 l4, enum ipfix_proto_tunnel tunnel,
1044                              struct dp_packet *msg)
1045 {
1046     uint16_t count = 0;
1047
1048 #define DEF(ID) \
1049     { \
1050         ipfix_define_template_entity(IPFIX_ENTITY_ID_##ID, \
1051                                      IPFIX_ENTITY_SIZE_##ID, \
1052                                      IPFIX_ENTITY_ENTERPRISE_##ID, msg); \
1053         count++; \
1054     }
1055
1056     /* 1. Flow key. */
1057
1058     DEF(OBSERVATION_POINT_ID);
1059     DEF(FLOW_DIRECTION);
1060
1061     /* Common Ethernet entities. */
1062     DEF(SOURCE_MAC_ADDRESS);
1063     DEF(DESTINATION_MAC_ADDRESS);
1064     DEF(ETHERNET_TYPE);
1065     DEF(ETHERNET_HEADER_LENGTH);
1066
1067     if (l2 == IPFIX_PROTO_L2_VLAN) {
1068         DEF(VLAN_ID);
1069         DEF(DOT1Q_VLAN_ID);
1070         DEF(DOT1Q_PRIORITY);
1071     }
1072
1073     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1074         DEF(IP_VERSION);
1075         DEF(IP_TTL);
1076         DEF(PROTOCOL_IDENTIFIER);
1077         DEF(IP_DIFF_SERV_CODE_POINT);
1078         DEF(IP_PRECEDENCE);
1079         DEF(IP_CLASS_OF_SERVICE);
1080
1081         if (l3 == IPFIX_PROTO_L3_IPV4) {
1082             DEF(SOURCE_IPV4_ADDRESS);
1083             DEF(DESTINATION_IPV4_ADDRESS);
1084             if (l4 == IPFIX_PROTO_L4_TCP_UDP_SCTP) {
1085                 DEF(SOURCE_TRANSPORT_PORT);
1086                 DEF(DESTINATION_TRANSPORT_PORT);
1087             } else if (l4 == IPFIX_PROTO_L4_ICMP) {
1088                 DEF(ICMP_TYPE_IPV4);
1089                 DEF(ICMP_CODE_IPV4);
1090             }
1091         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
1092             DEF(SOURCE_IPV6_ADDRESS);
1093             DEF(DESTINATION_IPV6_ADDRESS);
1094             DEF(FLOW_LABEL_IPV6);
1095             if (l4 == IPFIX_PROTO_L4_TCP_UDP_SCTP) {
1096                 DEF(SOURCE_TRANSPORT_PORT);
1097                 DEF(DESTINATION_TRANSPORT_PORT);
1098             } else if (l4 == IPFIX_PROTO_L4_ICMP) {
1099                 DEF(ICMP_TYPE_IPV6);
1100                 DEF(ICMP_CODE_IPV6);
1101             }
1102         }
1103     }
1104
1105     if (tunnel != IPFIX_PROTO_NOT_TUNNELED) {
1106         DEF(TUNNEL_SOURCE_IPV4_ADDRESS);
1107         DEF(TUNNEL_DESTINATION_IPV4_ADDRESS);
1108         DEF(TUNNEL_PROTOCOL_IDENTIFIER);
1109         DEF(TUNNEL_SOURCE_TRANSPORT_PORT);
1110         DEF(TUNNEL_DESTINATION_TRANSPORT_PORT);
1111         DEF(TUNNEL_TYPE);
1112         DEF(TUNNEL_KEY);
1113     }
1114
1115     /* 2. Flow aggregated data. */
1116
1117     DEF(FLOW_START_DELTA_MICROSECONDS);
1118     DEF(FLOW_END_DELTA_MICROSECONDS);
1119     DEF(PACKET_DELTA_COUNT);
1120     DEF(LAYER2_OCTET_DELTA_COUNT);
1121     DEF(FLOW_END_REASON);
1122
1123     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1124         DEF(OCTET_DELTA_COUNT);
1125         DEF(OCTET_DELTA_SUM_OF_SQUARES);
1126         DEF(MINIMUM_IP_TOTAL_LENGTH);
1127         DEF(MAXIMUM_IP_TOTAL_LENGTH);
1128     }
1129
1130
1131 #undef DEF
1132
1133     return count;
1134 }
1135
1136 static void
1137 ipfix_init_template_msg(void *msg_stub, uint32_t export_time_sec,
1138                         uint32_t seq_number, uint32_t obs_domain_id,
1139                         struct dp_packet *msg, size_t *set_hdr_offset)
1140 {
1141     struct ipfix_set_header *set_hdr;
1142
1143     dp_packet_use_stub(msg, msg_stub, sizeof msg_stub);
1144
1145     ipfix_init_header(export_time_sec, seq_number, obs_domain_id, msg);
1146     *set_hdr_offset = dp_packet_size(msg);
1147
1148     /* Add a Template Set. */
1149     set_hdr = dp_packet_put_zeros(msg, sizeof *set_hdr);
1150     set_hdr->set_id = htons(IPFIX_SET_ID_TEMPLATE);
1151 }
1152
1153 static void
1154 ipfix_send_template_msg(const struct collectors *collectors,
1155                         struct dp_packet *msg, size_t set_hdr_offset)
1156 {
1157     struct ipfix_set_header *set_hdr;
1158
1159     /* Send template message. */
1160     set_hdr = (struct ipfix_set_header*)
1161               ((uint8_t*)dp_packet_data(msg) + set_hdr_offset);
1162     set_hdr->length = htons(dp_packet_size(msg) - set_hdr_offset);
1163
1164     ipfix_send_msg(collectors, msg);
1165
1166     dp_packet_uninit(msg);
1167 }
1168
1169 static void
1170 ipfix_send_template_msgs(struct dpif_ipfix_exporter *exporter,
1171                          uint32_t export_time_sec, uint32_t obs_domain_id)
1172 {
1173     uint64_t msg_stub[DIV_ROUND_UP(MAX_MESSAGE_LEN, 8)];
1174     struct dp_packet msg;
1175     size_t set_hdr_offset, tmpl_hdr_offset;
1176     struct ipfix_template_record_header *tmpl_hdr;
1177     uint16_t field_count;
1178     enum ipfix_proto_l2 l2;
1179     enum ipfix_proto_l3 l3;
1180     enum ipfix_proto_l4 l4;
1181     enum ipfix_proto_tunnel tunnel;
1182
1183     ipfix_init_template_msg(msg_stub, export_time_sec, exporter->seq_number,
1184                             obs_domain_id, &msg, &set_hdr_offset);
1185     /* Define one template for each possible combination of
1186      * protocols. */
1187     for (l2 = 0; l2 < NUM_IPFIX_PROTO_L2; l2++) {
1188         for (l3 = 0; l3 < NUM_IPFIX_PROTO_L3; l3++) {
1189             for (l4 = 0; l4 < NUM_IPFIX_PROTO_L4; l4++) {
1190                 if (l3 == IPFIX_PROTO_L3_UNKNOWN &&
1191                     l4 != IPFIX_PROTO_L4_UNKNOWN) {
1192                     continue;
1193                 }
1194                 for (tunnel = 0; tunnel < NUM_IPFIX_PROTO_TUNNEL; tunnel++) {
1195                     /* When the size of the template packet reaches
1196                      * MAX_MESSAGE_LEN(1024), send it out.
1197                      * And then reinitialize the msg to construct a new
1198                      * packet for the following templates.
1199                      */
1200                     if (dp_packet_size(&msg) >= MAX_MESSAGE_LEN) {
1201                         /* Send template message. */
1202                         ipfix_send_template_msg(exporter->collectors,
1203                                                 &msg, set_hdr_offset);
1204
1205                         /* Reinitialize the template msg. */
1206                         ipfix_init_template_msg(msg_stub, export_time_sec,
1207                                                 exporter->seq_number,
1208                                                 obs_domain_id, &msg,
1209                                                 &set_hdr_offset);
1210                     }
1211
1212                     tmpl_hdr_offset = dp_packet_size(&msg);
1213                     tmpl_hdr = dp_packet_put_zeros(&msg, sizeof *tmpl_hdr);
1214                     tmpl_hdr->template_id = htons(
1215                         ipfix_get_template_id(l2, l3, l4, tunnel));
1216                     field_count =
1217                         ipfix_define_template_fields(l2, l3, l4, tunnel, &msg);
1218                     tmpl_hdr = (struct ipfix_template_record_header*)
1219                         ((uint8_t*)dp_packet_data(&msg) + tmpl_hdr_offset);
1220                     tmpl_hdr->field_count = htons(field_count);
1221                 }
1222             }
1223         }
1224     }
1225
1226     /* Send template message. */
1227     ipfix_send_template_msg(exporter->collectors, &msg, set_hdr_offset);
1228
1229     /* XXX: Add Options Template Sets, at least to define a Flow Keys
1230      * Option Template. */
1231
1232 }
1233
1234 static inline uint32_t
1235 ipfix_hash_flow_key(const struct ipfix_flow_key *flow_key, uint32_t basis)
1236 {
1237     uint32_t hash;
1238     hash = hash_int(flow_key->obs_domain_id, basis);
1239     hash = hash_int(flow_key->template_id, hash);
1240     hash = hash_bytes(flow_key->flow_key_msg_part,
1241                       flow_key->flow_key_msg_part_size, hash);
1242     return hash;
1243 }
1244
1245 static bool
1246 ipfix_flow_key_equal(const struct ipfix_flow_key *a,
1247                      const struct ipfix_flow_key *b)
1248 {
1249     /* The template ID determines the flow key size, so not need to
1250      * compare it. */
1251     return (a->obs_domain_id == b->obs_domain_id
1252             && a->template_id == b->template_id
1253             && memcmp(a->flow_key_msg_part, b->flow_key_msg_part,
1254                       a->flow_key_msg_part_size) == 0);
1255 }
1256
1257 static struct ipfix_flow_cache_entry*
1258 ipfix_cache_find_entry(const struct dpif_ipfix_exporter *exporter,
1259                        const struct ipfix_flow_key *flow_key)
1260 {
1261     struct ipfix_flow_cache_entry *entry;
1262
1263     HMAP_FOR_EACH_WITH_HASH (entry, flow_key_map_node,
1264                              ipfix_hash_flow_key(flow_key, 0),
1265                              &exporter->cache_flow_key_map) {
1266         if (ipfix_flow_key_equal(&entry->flow_key, flow_key)) {
1267             return entry;
1268         }
1269     }
1270
1271     return NULL;
1272 }
1273
1274 static bool
1275 ipfix_cache_next_timeout_msec(const struct dpif_ipfix_exporter *exporter,
1276                               long long int *next_timeout_msec)
1277 {
1278     struct ipfix_flow_cache_entry *entry;
1279
1280     LIST_FOR_EACH (entry, cache_flow_start_timestamp_list_node,
1281                    &exporter->cache_flow_start_timestamp_list) {
1282         *next_timeout_msec = entry->flow_start_timestamp_usec / 1000LL
1283             + 1000LL * exporter->cache_active_timeout;
1284         return true;
1285     }
1286
1287     return false;
1288 }
1289
1290 static void
1291 ipfix_cache_aggregate_entries(struct ipfix_flow_cache_entry *from_entry,
1292                               struct ipfix_flow_cache_entry *to_entry)
1293 {
1294     uint64_t *to_start, *to_end, *from_start, *from_end;
1295     uint16_t *to_min_len, *to_max_len, *from_min_len, *from_max_len;
1296
1297     to_start = &to_entry->flow_start_timestamp_usec;
1298     to_end = &to_entry->flow_end_timestamp_usec;
1299     from_start = &from_entry->flow_start_timestamp_usec;
1300     from_end = &from_entry->flow_end_timestamp_usec;
1301
1302     if (*to_start > *from_start) {
1303         *to_start = *from_start;
1304     }
1305     if (*to_end < *from_end) {
1306         *to_end = *from_end;
1307     }
1308
1309     to_entry->packet_delta_count += from_entry->packet_delta_count;
1310     to_entry->layer2_octet_delta_count += from_entry->layer2_octet_delta_count;
1311
1312     to_entry->octet_delta_count += from_entry->octet_delta_count;
1313     to_entry->octet_delta_sum_of_squares +=
1314         from_entry->octet_delta_sum_of_squares;
1315
1316     to_min_len = &to_entry->minimum_ip_total_length;
1317     to_max_len = &to_entry->maximum_ip_total_length;
1318     from_min_len = &from_entry->minimum_ip_total_length;
1319     from_max_len = &from_entry->maximum_ip_total_length;
1320
1321     if (!*to_min_len || (*from_min_len && *to_min_len > *from_min_len)) {
1322         *to_min_len = *from_min_len;
1323     }
1324     if (*to_max_len < *from_max_len) {
1325         *to_max_len = *from_max_len;
1326     }
1327 }
1328
1329 /* Add an entry into a flow cache.  The entry is either aggregated into
1330  * an existing entry with the same flow key and free()d, or it is
1331  * inserted into the cache. */
1332 static void
1333 ipfix_cache_update(struct dpif_ipfix_exporter *exporter,
1334                    struct ipfix_flow_cache_entry *entry)
1335 {
1336     struct ipfix_flow_cache_entry *old_entry;
1337
1338     old_entry = ipfix_cache_find_entry(exporter, &entry->flow_key);
1339
1340     if (old_entry == NULL) {
1341         hmap_insert(&exporter->cache_flow_key_map, &entry->flow_key_map_node,
1342                     ipfix_hash_flow_key(&entry->flow_key, 0));
1343
1344         /* As the latest entry added into the cache, it should
1345          * logically have the highest flow_start_timestamp_usec, so
1346          * append it at the tail. */
1347         list_push_back(&exporter->cache_flow_start_timestamp_list,
1348                        &entry->cache_flow_start_timestamp_list_node);
1349
1350         /* Enforce exporter->cache_max_flows limit. */
1351         if (hmap_count(&exporter->cache_flow_key_map)
1352             > exporter->cache_max_flows) {
1353             dpif_ipfix_cache_expire_now(exporter, false);
1354         }
1355     } else {
1356         ipfix_cache_aggregate_entries(entry, old_entry);
1357         free(entry);
1358     }
1359 }
1360
1361 static void
1362 ipfix_cache_entry_init(struct ipfix_flow_cache_entry *entry,
1363                        const struct dp_packet *packet, const struct flow *flow,
1364                        uint64_t packet_delta_count, uint32_t obs_domain_id,
1365                        uint32_t obs_point_id, odp_port_t output_odp_port,
1366                        const struct dpif_ipfix_port *tunnel_port,
1367                        const struct flow_tnl *tunnel_key)
1368 {
1369     struct ipfix_flow_key *flow_key;
1370     struct dp_packet msg;
1371     enum ipfix_proto_l2 l2;
1372     enum ipfix_proto_l3 l3;
1373     enum ipfix_proto_l4 l4;
1374     enum ipfix_proto_tunnel tunnel = IPFIX_PROTO_NOT_TUNNELED;
1375     uint8_t ethernet_header_length;
1376     uint16_t ethernet_total_length;
1377
1378     flow_key = &entry->flow_key;
1379     dp_packet_use_stub(&msg, flow_key->flow_key_msg_part,
1380                        sizeof flow_key->flow_key_msg_part);
1381
1382     /* Choose the right template ID matching the protocols in the
1383      * sampled packet. */
1384     l2 = (flow->vlan_tci == 0) ? IPFIX_PROTO_L2_ETH : IPFIX_PROTO_L2_VLAN;
1385
1386     switch(ntohs(flow->dl_type)) {
1387     case ETH_TYPE_IP:
1388         l3 = IPFIX_PROTO_L3_IPV4;
1389         switch(flow->nw_proto) {
1390         case IPPROTO_TCP:
1391         case IPPROTO_UDP:
1392         case IPPROTO_SCTP:
1393             l4 = IPFIX_PROTO_L4_TCP_UDP_SCTP;
1394             break;
1395         case IPPROTO_ICMP:
1396             l4 = IPFIX_PROTO_L4_ICMP;
1397             break;
1398         default:
1399             l4 = IPFIX_PROTO_L4_UNKNOWN;
1400         }
1401         break;
1402     case ETH_TYPE_IPV6:
1403         l3 = IPFIX_PROTO_L3_IPV6;
1404         switch(flow->nw_proto) {
1405         case IPPROTO_TCP:
1406         case IPPROTO_UDP:
1407         case IPPROTO_SCTP:
1408             l4 = IPFIX_PROTO_L4_TCP_UDP_SCTP;
1409             break;
1410         case IPPROTO_ICMPV6:
1411             l4 = IPFIX_PROTO_L4_ICMP;
1412             break;
1413         default:
1414             l4 = IPFIX_PROTO_L4_UNKNOWN;
1415         }
1416         break;
1417     default:
1418         l3 = IPFIX_PROTO_L3_UNKNOWN;
1419         l4 = IPFIX_PROTO_L4_UNKNOWN;
1420     }
1421
1422     if (tunnel_port && tunnel_key) {
1423        tunnel = IPFIX_PROTO_TUNNELED;
1424     }
1425
1426     flow_key->obs_domain_id = obs_domain_id;
1427     flow_key->template_id = ipfix_get_template_id(l2, l3, l4, tunnel);
1428
1429     /* The fields defined in the ipfix_data_record_* structs and sent
1430      * below must match exactly the templates defined in
1431      * ipfix_define_template_fields. */
1432
1433     ethernet_header_length = (l2 == IPFIX_PROTO_L2_VLAN)
1434         ? VLAN_ETH_HEADER_LEN : ETH_HEADER_LEN;
1435     ethernet_total_length = dp_packet_size(packet);
1436
1437     /* Common Ethernet entities. */
1438     {
1439         struct ipfix_data_record_flow_key_common *data_common;
1440
1441         data_common = dp_packet_put_zeros(&msg, sizeof *data_common);
1442         data_common->observation_point_id = htonl(obs_point_id);
1443         data_common->flow_direction =
1444             (output_odp_port == ODPP_NONE) ? INGRESS_FLOW : EGRESS_FLOW;
1445         data_common->source_mac_address = flow->dl_src;
1446         data_common->destination_mac_address = flow->dl_dst;
1447         data_common->ethernet_type = flow->dl_type;
1448         data_common->ethernet_header_length = ethernet_header_length;
1449     }
1450
1451     if (l2 == IPFIX_PROTO_L2_VLAN) {
1452         struct ipfix_data_record_flow_key_vlan *data_vlan;
1453         uint16_t vlan_id = vlan_tci_to_vid(flow->vlan_tci);
1454         uint8_t priority = vlan_tci_to_pcp(flow->vlan_tci);
1455
1456         data_vlan = dp_packet_put_zeros(&msg, sizeof *data_vlan);
1457         data_vlan->vlan_id = htons(vlan_id);
1458         data_vlan->dot1q_vlan_id = htons(vlan_id);
1459         data_vlan->dot1q_priority = priority;
1460     }
1461
1462     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1463         struct ipfix_data_record_flow_key_ip *data_ip;
1464
1465         data_ip = dp_packet_put_zeros(&msg, sizeof *data_ip);
1466         data_ip->ip_version = (l3 == IPFIX_PROTO_L3_IPV4) ? 4 : 6;
1467         data_ip->ip_ttl = flow->nw_ttl;
1468         data_ip->protocol_identifier = flow->nw_proto;
1469         data_ip->ip_diff_serv_code_point = flow->nw_tos >> 2;
1470         data_ip->ip_precedence = flow->nw_tos >> 5;
1471         data_ip->ip_class_of_service = flow->nw_tos;
1472
1473         if (l3 == IPFIX_PROTO_L3_IPV4) {
1474             struct ipfix_data_record_flow_key_ipv4 *data_ipv4;
1475
1476             data_ipv4 = dp_packet_put_zeros(&msg, sizeof *data_ipv4);
1477             data_ipv4->source_ipv4_address = flow->nw_src;
1478             data_ipv4->destination_ipv4_address = flow->nw_dst;
1479         } else {  /* l3 == IPFIX_PROTO_L3_IPV6 */
1480             struct ipfix_data_record_flow_key_ipv6 *data_ipv6;
1481
1482             data_ipv6 = dp_packet_put_zeros(&msg, sizeof *data_ipv6);
1483             memcpy(data_ipv6->source_ipv6_address, &flow->ipv6_src,
1484                    sizeof flow->ipv6_src);
1485             memcpy(data_ipv6->destination_ipv6_address, &flow->ipv6_dst,
1486                    sizeof flow->ipv6_dst);
1487             data_ipv6->flow_label_ipv6 = flow->ipv6_label;
1488         }
1489     }
1490
1491     if (l4 == IPFIX_PROTO_L4_TCP_UDP_SCTP) {
1492         struct ipfix_data_record_flow_key_transport *data_transport;
1493
1494         data_transport = dp_packet_put_zeros(&msg, sizeof *data_transport);
1495         data_transport->source_transport_port = flow->tp_src;
1496         data_transport->destination_transport_port = flow->tp_dst;
1497     } else if (l4 == IPFIX_PROTO_L4_ICMP) {
1498         struct ipfix_data_record_flow_key_icmp *data_icmp;
1499
1500         data_icmp = dp_packet_put_zeros(&msg, sizeof *data_icmp);
1501         data_icmp->icmp_type = ntohs(flow->tp_src) & 0xff;
1502         data_icmp->icmp_code = ntohs(flow->tp_dst) & 0xff;
1503     }
1504
1505     if (tunnel == IPFIX_PROTO_TUNNELED) {
1506         struct ipfix_data_record_flow_key_tunnel *data_tunnel;
1507         const uint8_t *tun_id;
1508
1509         data_tunnel = dp_packet_put_zeros(&msg, sizeof *data_tunnel +
1510                                              tunnel_port->tunnel_key_length);
1511         data_tunnel->tunnel_source_ipv4_address = tunnel_key->ip_src;
1512         data_tunnel->tunnel_destination_ipv4_address = tunnel_key->ip_dst;
1513         /* The tunnel_protocol_identifier is from tunnel_proto array, which
1514          * contains protocol_identifiers of each tunnel type.
1515          * For the tunnel type on the top of IPSec, which uses the protocol
1516          * identifier of the upper tunnel type is used, the tcp_src and tcp_dst
1517          * are decided based on the protocol identifiers.
1518          * E.g:
1519          * The protocol identifier of DPIF_IPFIX_TUNNEL_IPSEC_GRE is IPPROTO_GRE,
1520          * and both tp_src and tp_dst are zero.
1521          */
1522         data_tunnel->tunnel_protocol_identifier =
1523             tunnel_protocol[tunnel_port->tunnel_type];
1524         data_tunnel->tunnel_source_transport_port = tunnel_key->tp_src;
1525         data_tunnel->tunnel_destination_transport_port = tunnel_key->tp_dst;
1526         data_tunnel->tunnel_type = tunnel_port->tunnel_type;
1527         data_tunnel->tunnel_key_length = tunnel_port->tunnel_key_length;
1528         /* tun_id is in network order, and tunnel key is in low bits. */
1529         tun_id = (const uint8_t *) &tunnel_key->tun_id;
1530         memcpy(data_tunnel->tunnel_key,
1531                &tun_id[8 - tunnel_port->tunnel_key_length],
1532                tunnel_port->tunnel_key_length);
1533     }
1534
1535     flow_key->flow_key_msg_part_size = dp_packet_size(&msg);
1536
1537     {
1538         struct timeval now;
1539         uint64_t layer2_octet_delta_count;
1540
1541         /* Calculate the total matched octet count by considering as
1542          * an approximation that all matched packets have the same
1543          * length. */
1544         layer2_octet_delta_count = packet_delta_count * ethernet_total_length;
1545
1546         xgettimeofday(&now);
1547         entry->flow_end_timestamp_usec = now.tv_usec + 1000000LL * now.tv_sec;
1548         entry->flow_start_timestamp_usec = entry->flow_end_timestamp_usec;
1549         entry->packet_delta_count = packet_delta_count;
1550         entry->layer2_octet_delta_count = layer2_octet_delta_count;
1551     }
1552
1553     if (l3 != IPFIX_PROTO_L3_UNKNOWN) {
1554         uint16_t ip_total_length =
1555             ethernet_total_length - ethernet_header_length;
1556         uint64_t octet_delta_count;
1557
1558         /* Calculate the total matched octet count by considering as
1559          * an approximation that all matched packets have the same
1560          * length. */
1561         octet_delta_count = packet_delta_count * ip_total_length;
1562
1563         entry->octet_delta_count = octet_delta_count;
1564         entry->octet_delta_sum_of_squares = octet_delta_count * ip_total_length;
1565         entry->minimum_ip_total_length = ip_total_length;
1566         entry->maximum_ip_total_length = ip_total_length;
1567     } else {
1568         entry->octet_delta_sum_of_squares = 0;
1569         entry->minimum_ip_total_length = 0;
1570         entry->maximum_ip_total_length = 0;
1571     }
1572 }
1573
1574 /* Send each single data record in its own data set, to simplify the
1575  * implementation by avoiding having to group record by template ID
1576  * before sending. */
1577 static void
1578 ipfix_put_data_set(uint32_t export_time_sec,
1579                    struct ipfix_flow_cache_entry *entry,
1580                    enum ipfix_flow_end_reason flow_end_reason,
1581                    struct dp_packet *msg)
1582 {
1583     size_t set_hdr_offset;
1584     struct ipfix_set_header *set_hdr;
1585
1586     set_hdr_offset = dp_packet_size(msg);
1587
1588     /* Put a Data Set. */
1589     set_hdr = dp_packet_put_zeros(msg, sizeof *set_hdr);
1590     set_hdr->set_id = htons(entry->flow_key.template_id);
1591
1592     /* Copy the flow key part of the data record. */
1593
1594     dp_packet_put(msg, entry->flow_key.flow_key_msg_part,
1595                entry->flow_key.flow_key_msg_part_size);
1596
1597     /* Put the non-key part of the data record. */
1598
1599     {
1600         struct ipfix_data_record_aggregated_common *data_aggregated_common;
1601         uint64_t export_time_usec, flow_start_delta_usec, flow_end_delta_usec;
1602
1603         /* Calculate the negative deltas relative to the export time
1604          * in seconds sent in the header, not the exact export
1605          * time. */
1606         export_time_usec = 1000000LL * export_time_sec;
1607         flow_start_delta_usec = export_time_usec
1608             - entry->flow_start_timestamp_usec;
1609         flow_end_delta_usec = export_time_usec
1610             - entry->flow_end_timestamp_usec;
1611
1612         data_aggregated_common = dp_packet_put_zeros(
1613             msg, sizeof *data_aggregated_common);
1614         data_aggregated_common->flow_start_delta_microseconds = htonl(
1615             flow_start_delta_usec);
1616         data_aggregated_common->flow_end_delta_microseconds = htonl(
1617             flow_end_delta_usec);
1618         data_aggregated_common->packet_delta_count = htonll(
1619             entry->packet_delta_count);
1620         data_aggregated_common->layer2_octet_delta_count = htonll(
1621             entry->layer2_octet_delta_count);
1622         data_aggregated_common->flow_end_reason = flow_end_reason;
1623     }
1624
1625     if (entry->octet_delta_sum_of_squares) {  /* IP packet. */
1626         struct ipfix_data_record_aggregated_ip *data_aggregated_ip;
1627
1628         data_aggregated_ip = dp_packet_put_zeros(
1629             msg, sizeof *data_aggregated_ip);
1630         data_aggregated_ip->octet_delta_count = htonll(
1631             entry->octet_delta_count);
1632         data_aggregated_ip->octet_delta_sum_of_squares = htonll(
1633             entry->octet_delta_sum_of_squares);
1634         data_aggregated_ip->minimum_ip_total_length = htonll(
1635             entry->minimum_ip_total_length);
1636         data_aggregated_ip->maximum_ip_total_length = htonll(
1637             entry->maximum_ip_total_length);
1638     }
1639
1640     set_hdr = (struct ipfix_set_header*)((uint8_t*)dp_packet_data(msg) + set_hdr_offset);
1641     set_hdr->length = htons(dp_packet_size(msg) - set_hdr_offset);
1642 }
1643
1644 /* Send an IPFIX message with a single data record. */
1645 static void
1646 ipfix_send_data_msg(struct dpif_ipfix_exporter *exporter,
1647                     uint32_t export_time_sec,
1648                     struct ipfix_flow_cache_entry *entry,
1649                     enum ipfix_flow_end_reason flow_end_reason)
1650 {
1651     uint64_t msg_stub[DIV_ROUND_UP(MAX_MESSAGE_LEN, 8)];
1652     struct dp_packet msg;
1653     dp_packet_use_stub(&msg, msg_stub, sizeof msg_stub);
1654
1655     ipfix_init_header(export_time_sec, exporter->seq_number++,
1656                       entry->flow_key.obs_domain_id, &msg);
1657     ipfix_put_data_set(export_time_sec, entry, flow_end_reason, &msg);
1658     ipfix_send_msg(exporter->collectors, &msg);
1659
1660     dp_packet_uninit(&msg);
1661 }
1662
1663 static void
1664 dpif_ipfix_sample(struct dpif_ipfix_exporter *exporter,
1665                   const struct dp_packet *packet, const struct flow *flow,
1666                   uint64_t packet_delta_count, uint32_t obs_domain_id,
1667                   uint32_t obs_point_id, odp_port_t output_odp_port,
1668                   const struct dpif_ipfix_port *tunnel_port,
1669                   const struct flow_tnl *tunnel_key)
1670 {
1671     struct ipfix_flow_cache_entry *entry;
1672
1673     /* Create a flow cache entry from the sample. */
1674     entry = xmalloc(sizeof *entry);
1675     ipfix_cache_entry_init(entry, packet, flow, packet_delta_count,
1676                            obs_domain_id, obs_point_id,
1677                            output_odp_port, tunnel_port, tunnel_key);
1678     ipfix_cache_update(exporter, entry);
1679 }
1680
1681 static bool
1682 bridge_exporter_enabled(struct dpif_ipfix *di)
1683 {
1684     return di->bridge_exporter.probability > 0;
1685 }
1686
1687 void
1688 dpif_ipfix_bridge_sample(struct dpif_ipfix *di, const struct dp_packet *packet,
1689                          const struct flow *flow,
1690                          odp_port_t input_odp_port, odp_port_t output_odp_port,
1691                          const struct flow_tnl *output_tunnel_key)
1692     OVS_EXCLUDED(mutex)
1693 {
1694     uint64_t packet_delta_count;
1695     const struct flow_tnl *tunnel_key = NULL;
1696     struct dpif_ipfix_port * tunnel_port = NULL;
1697
1698     ovs_mutex_lock(&mutex);
1699     if (!bridge_exporter_enabled(di)) {
1700         ovs_mutex_unlock(&mutex);
1701         return;
1702     }
1703
1704     /* Skip BFD packets:
1705      * Bidirectional Forwarding Detection(BFD) packets are for monitoring
1706      * the tunnel link status and consumed by ovs itself. No need to
1707      * smaple them.
1708      * CF  IETF RFC 5881, BFD control packet is the UDP packet with
1709      * destination port 3784, and BFD echo packet is the UDP packet with
1710      * destination port 3785.
1711      */
1712     if (is_ip_any(flow) &&
1713         flow->nw_proto == IPPROTO_UDP &&
1714         (flow->tp_dst == htons(BFD_CONTROL_DEST_PORT) ||
1715          flow->tp_dst == htons(BFD_ECHO_DEST_PORT))) {
1716         ovs_mutex_unlock(&mutex);
1717         return;
1718     }
1719
1720     /* Use the sampling probability as an approximation of the number
1721      * of matched packets. */
1722     packet_delta_count = UINT32_MAX / di->bridge_exporter.probability;
1723     if (di->bridge_exporter.options->enable_tunnel_sampling) {
1724         if (output_odp_port == ODPP_NONE && flow->tunnel.ip_dst) {
1725             /* Input tunnel. */
1726             tunnel_key = &flow->tunnel;
1727             tunnel_port = dpif_ipfix_find_port(di, input_odp_port);
1728         }
1729         if (output_odp_port != ODPP_NONE && output_tunnel_key) {
1730             /* Output tunnel, output_tunnel_key must be valid. */
1731             tunnel_key = output_tunnel_key;
1732             tunnel_port = dpif_ipfix_find_port(di, output_odp_port);
1733         }
1734     }
1735
1736     dpif_ipfix_sample(&di->bridge_exporter.exporter, packet, flow,
1737                       packet_delta_count,
1738                       di->bridge_exporter.options->obs_domain_id,
1739                       di->bridge_exporter.options->obs_point_id,
1740                       output_odp_port, tunnel_port, tunnel_key);
1741     ovs_mutex_unlock(&mutex);
1742 }
1743
1744 void
1745 dpif_ipfix_flow_sample(struct dpif_ipfix *di, const struct dp_packet *packet,
1746                        const struct flow *flow, uint32_t collector_set_id,
1747                        uint16_t probability, uint32_t obs_domain_id,
1748                        uint32_t obs_point_id) OVS_EXCLUDED(mutex)
1749 {
1750     struct dpif_ipfix_flow_exporter_map_node *node;
1751     /* Use the sampling probability as an approximation of the number
1752      * of matched packets. */
1753     uint64_t packet_delta_count = USHRT_MAX / probability;
1754
1755     ovs_mutex_lock(&mutex);
1756     node = dpif_ipfix_find_flow_exporter_map_node(di, collector_set_id);
1757     if (node) {
1758         dpif_ipfix_sample(&node->exporter.exporter, packet, flow,
1759                           packet_delta_count, obs_domain_id, obs_point_id,
1760                           ODPP_NONE, NULL, NULL);
1761     }
1762     ovs_mutex_unlock(&mutex);
1763 }
1764
1765 static void
1766 dpif_ipfix_cache_expire(struct dpif_ipfix_exporter *exporter,
1767                         bool forced_end, const uint64_t export_time_usec,
1768                         const uint32_t export_time_sec)
1769 {
1770     struct ipfix_flow_cache_entry *entry, *next_entry;
1771     uint64_t max_flow_start_timestamp_usec;
1772     bool template_msg_sent = false;
1773     enum ipfix_flow_end_reason flow_end_reason;
1774
1775     if (list_is_empty(&exporter->cache_flow_start_timestamp_list)) {
1776         return;
1777     }
1778
1779     max_flow_start_timestamp_usec = export_time_usec -
1780         1000000LL * exporter->cache_active_timeout;
1781
1782     LIST_FOR_EACH_SAFE (entry, next_entry, cache_flow_start_timestamp_list_node,
1783                         &exporter->cache_flow_start_timestamp_list) {
1784         if (forced_end) {
1785             flow_end_reason = FORCED_END;
1786         } else if (entry->flow_start_timestamp_usec
1787                    <= max_flow_start_timestamp_usec) {
1788             flow_end_reason = ACTIVE_TIMEOUT;
1789         } else if (hmap_count(&exporter->cache_flow_key_map)
1790                    > exporter->cache_max_flows) {
1791             /* Enforce exporter->cache_max_flows. */
1792             flow_end_reason = LACK_OF_RESOURCES;
1793         } else {
1794             /* Remaining flows haven't expired yet. */
1795             break;
1796         }
1797
1798         list_remove(&entry->cache_flow_start_timestamp_list_node);
1799         hmap_remove(&exporter->cache_flow_key_map,
1800                     &entry->flow_key_map_node);
1801
1802         if (!template_msg_sent
1803             && (exporter->last_template_set_time + IPFIX_TEMPLATE_INTERVAL)
1804                 <= export_time_sec) {
1805             ipfix_send_template_msgs(exporter, export_time_sec,
1806                                      entry->flow_key.obs_domain_id);
1807             exporter->last_template_set_time = export_time_sec;
1808             template_msg_sent = true;
1809         }
1810
1811         /* XXX: Group multiple data records for the same obs domain id
1812          * into the same message. */
1813         ipfix_send_data_msg(exporter, export_time_sec, entry, flow_end_reason);
1814         free(entry);
1815     }
1816 }
1817
1818 static void
1819 get_export_time_now(uint64_t *export_time_usec, uint32_t *export_time_sec)
1820 {
1821     struct timeval export_time;
1822     xgettimeofday(&export_time);
1823
1824     *export_time_usec = export_time.tv_usec + 1000000LL * export_time.tv_sec;
1825
1826     /* The IPFIX start and end deltas are negative deltas relative to
1827      * the export time, so set the export time 1 second off to
1828      * calculate those deltas. */
1829     if (export_time.tv_usec == 0) {
1830         *export_time_sec = export_time.tv_sec;
1831     } else {
1832         *export_time_sec = export_time.tv_sec + 1;
1833     }
1834 }
1835
1836 static void
1837 dpif_ipfix_cache_expire_now(struct dpif_ipfix_exporter *exporter,
1838                             bool forced_end)
1839 {
1840     uint64_t export_time_usec;
1841     uint32_t export_time_sec;
1842
1843     get_export_time_now(&export_time_usec, &export_time_sec);
1844     dpif_ipfix_cache_expire(exporter, forced_end, export_time_usec,
1845                             export_time_sec);
1846 }
1847
1848 void
1849 dpif_ipfix_run(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
1850 {
1851     uint64_t export_time_usec;
1852     uint32_t export_time_sec;
1853     struct dpif_ipfix_flow_exporter_map_node *flow_exporter_node;
1854
1855     ovs_mutex_lock(&mutex);
1856     get_export_time_now(&export_time_usec, &export_time_sec);
1857     if (bridge_exporter_enabled(di)) {
1858       dpif_ipfix_cache_expire(
1859           &di->bridge_exporter.exporter, false, export_time_usec,
1860           export_time_sec);
1861     }
1862     HMAP_FOR_EACH (flow_exporter_node, node, &di->flow_exporter_map) {
1863         dpif_ipfix_cache_expire(
1864             &flow_exporter_node->exporter.exporter, false, export_time_usec,
1865             export_time_sec);
1866     }
1867     ovs_mutex_unlock(&mutex);
1868 }
1869
1870 void
1871 dpif_ipfix_wait(struct dpif_ipfix *di) OVS_EXCLUDED(mutex)
1872 {
1873     long long int next_timeout_msec = LLONG_MAX;
1874     struct dpif_ipfix_flow_exporter_map_node *flow_exporter_node;
1875
1876     ovs_mutex_lock(&mutex);
1877     if (bridge_exporter_enabled(di)) {
1878         if (ipfix_cache_next_timeout_msec(
1879                 &di->bridge_exporter.exporter, &next_timeout_msec)) {
1880             poll_timer_wait_until(next_timeout_msec);
1881         }
1882     }
1883     HMAP_FOR_EACH (flow_exporter_node, node, &di->flow_exporter_map) {
1884         if (ipfix_cache_next_timeout_msec(
1885                 &flow_exporter_node->exporter.exporter, &next_timeout_msec)) {
1886             poll_timer_wait_until(next_timeout_msec);
1887         }
1888     }
1889     ovs_mutex_unlock(&mutex);
1890 }