examples/ip_reassembly/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  *
  33  */
  34
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <stdint.h>
  38 #include <inttypes.h>
  39 #include <sys/types.h>
  40 #include <string.h>
  41 #include <sys/queue.h>
  42 #include <stdarg.h>
  43 #include <errno.h>
  44 #include <getopt.h>
  45 #include <signal.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.h>
  53 #include <rte_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_malloc.h>
  73 #include <rte_ip.h>
  74 #include <rte_tcp.h>
  75 #include <rte_udp.h>
  76 #include <rte_string_fns.h>
  77
  78 #include "main.h"
  79 #include "ipv4_rsmbl.h"
  80
  81 #define APP_LOOKUP_EXACT_MATCH          0
  82 #define APP_LOOKUP_LPM                  1
  83 #define DO_RFC_1812_CHECKS
  84
  85 #ifndef APP_LOOKUP_METHOD
  86 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  87 #endif
  88
  89 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  90 #include <rte_hash.h>
  91 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  92 #include <rte_lpm.h>
  93 #include <rte_lpm6.h>
  94 #else
  95 #error "APP_LOOKUP_METHOD set to incorrect value"
  96 #endif
  97
  98 #ifndef IPv6_BYTES
  99 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 100                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 101 #define IPv6_BYTES(addr) \
 102         addr[0],  addr[1], addr[2],  addr[3], \
 103         addr[4],  addr[5], addr[6],  addr[7], \
 104         addr[8],  addr[9], addr[10], addr[11],\
 105         addr[12], addr[13],addr[14], addr[15]
 106 #endif
 107
 108
 109 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 110
 111 #define MAX_PORTS       RTE_MAX_ETHPORTS
 112
 113 #define MAX_JUMBO_PKT_LEN  9600
 114
 115 #define IPV6_ADDR_LEN 16
 116
 117 #define MEMPOOL_CACHE_SIZE 256
 118
 119 #define BUF_SIZE        2048
 120 #define MBUF_SIZE       \
 121         (BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 122
 123 #define MAX_FLOW_NUM    UINT16_MAX
 124 #define MIN_FLOW_NUM    1
 125 #define DEF_FLOW_NUM    0x1000
 126
 127 /* TTL numbers are in ms. */
 128 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 129 #define MIN_FLOW_TTL    1
 130 #define DEF_FLOW_TTL    MS_PER_S
 131
 132 #define DEF_MBUF_NUM    0x400
 133
 134 /* Should be power of two. */
 135 #define IPV4_FRAG_TBL_BUCKET_ENTRIES    2
 136
 137 static uint32_t max_flow_num = DEF_FLOW_NUM;
 138 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 139
 140 /*
 141  * RX and TX Prefetch, Host, and Write-back threshold values should be
 142  * carefully set for optimal performance. Consult the network
 143  * controller's datasheet and supporting DPDK documentation for guidance
 144  * on how these parameters should be set.
 145  */
 146 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 147 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 148 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 149
 150 /*
 151  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 152  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 153  * network controllers and/or network drivers.
 154  */
 155 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 156 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 157 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 158
 159 #define MAX_PKT_BURST 32
 160 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 161
 162 #define NB_SOCKETS 8
 163
 164 /* Configure how many packets ahead to prefetch, when reading packets */
 165 #define PREFETCH_OFFSET 3
 166
 167 /*
 168  * Configurable number of RX/TX ring descriptors
 169  */
 170 #define RTE_TEST_RX_DESC_DEFAULT 128
 171 #define RTE_TEST_TX_DESC_DEFAULT 512
 172 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 173 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 174
 175 /* ethernet addresses of ports */
 176 static struct ether_addr ports_eth_addr[MAX_PORTS];
 177
 178 /* mask of enabled ports */
 179 static uint32_t enabled_port_mask = 0;
 180 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 181 static int numa_on = 1; /**< NUMA is enabled by default. */
 182
 183 struct mbuf_table {
 184         uint16_t len;
 185         struct rte_mbuf *m_table[MAX_PKT_BURST];
 186 };
 187
 188 struct lcore_rx_queue {
 189         uint8_t port_id;
 190         uint8_t queue_id;
 191 } __rte_cache_aligned;
 192
 193 #define MAX_RX_QUEUE_PER_LCORE 16
 194 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 195 #define MAX_RX_QUEUE_PER_PORT 128
 196
 197 #define MAX_LCORE_PARAMS 1024
 198 struct lcore_params {
 199         uint8_t port_id;
 200         uint8_t queue_id;
 201         uint8_t lcore_id;
 202 } __rte_cache_aligned;
 203
 204 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 205 static struct lcore_params lcore_params_array_default[] = {
 206         {0, 0, 2},
 207         {0, 1, 2},
 208         {0, 2, 2},
 209         {1, 0, 2},
 210         {1, 1, 2},
 211         {1, 2, 2},
 212         {2, 0, 2},
 213         {3, 0, 3},
 214         {3, 1, 3},
 215 };
 216
 217 static struct lcore_params * lcore_params = lcore_params_array_default;
 218 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 219                                 sizeof(lcore_params_array_default[0]);
 220
 221 static struct rte_eth_conf port_conf = {
 222         .rxmode = {
 223                 .max_rx_pkt_len = ETHER_MAX_LEN,
 224                 .split_hdr_size = 0,
 225                 .header_split   = 0, /**< Header Split disabled */
 226                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 227                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 228                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 229                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 230         },
 231         .rx_adv_conf = {
 232                 .rss_conf = {
 233                         .rss_key = NULL,
 234                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 235                 },
 236         },
 237         .txmode = {
 238                 .mq_mode = ETH_MQ_TX_NONE,
 239         },
 240 };
 241
 242 static const struct rte_eth_rxconf rx_conf = {
 243         .rx_thresh = {
 244                 .pthresh = RX_PTHRESH,
 245                 .hthresh = RX_HTHRESH,
 246                 .wthresh = RX_WTHRESH,
 247         },
 248         .rx_free_thresh = 32,
 249 };
 250
 251 static const struct rte_eth_txconf tx_conf = {
 252         .tx_thresh = {
 253                 .pthresh = TX_PTHRESH,
 254                 .hthresh = TX_HTHRESH,
 255                 .wthresh = TX_WTHRESH,
 256         },
 257         .tx_free_thresh = 0, /* Use PMD default values */
 258         .tx_rs_thresh = 0, /* Use PMD default values */
 259         .txq_flags = 0x0,
 260 };
 261
 262 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 263
 264 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 265 #include <rte_hash_crc.h>
 266 #define DEFAULT_HASH_FUNC       rte_hash_crc
 267 #else
 268 #include <rte_jhash.h>
 269 #define DEFAULT_HASH_FUNC       rte_jhash
 270 #endif
 271
 272 struct ipv4_5tuple {
 273         uint32_t ip_dst;
 274         uint32_t ip_src;
 275         uint16_t port_dst;
 276         uint16_t port_src;
 277         uint8_t  proto;
 278 } __attribute__((__packed__));
 279
 280 struct ipv6_5tuple {
 281         uint8_t  ip_dst[IPV6_ADDR_LEN];
 282         uint8_t  ip_src[IPV6_ADDR_LEN];
 283         uint16_t port_dst;
 284         uint16_t port_src;
 285         uint8_t  proto;
 286 } __attribute__((__packed__));
 287
 288 struct ipv4_l3fwd_route {
 289         struct ipv4_5tuple key;
 290         uint8_t if_out;
 291 };
 292
 293 struct ipv6_l3fwd_route {
 294         struct ipv6_5tuple key;
 295         uint8_t if_out;
 296 };
 297
 298 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 299         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 300         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 301         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 302         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 303 };
 304
 305 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 306         {
 307                 {
 308                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 309                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 310                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 311                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 312                          1, 10, IPPROTO_UDP
 313                 }, 4
 314         },
 315 };
 316
 317 typedef struct rte_hash lookup_struct_t;
 318 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 319 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 320
 321 #define L3FWD_HASH_ENTRIES      1024
 322
 323 #define IPV4_L3FWD_NUM_ROUTES \
 324         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 325
 326 #define IPV6_L3FWD_NUM_ROUTES \
 327         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 328
 329 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 330 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 331 #endif
 332
 333 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 334 struct ipv4_l3fwd_route {
 335         uint32_t ip;
 336         uint8_t  depth;
 337         uint8_t  if_out;
 338 };
 339
 340 struct ipv6_l3fwd_route {
 341         uint8_t ip[16];
 342         uint8_t  depth;
 343         uint8_t  if_out;
 344 };
 345
 346 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 347         {IPv4(1,1,1,0), 24, 0},
 348         {IPv4(2,1,1,0), 24, 1},
 349         {IPv4(3,1,1,0), 24, 2},
 350         {IPv4(4,1,1,0), 24, 3},
 351         {IPv4(5,1,1,0), 24, 4},
 352         {IPv4(6,1,1,0), 24, 5},
 353         {IPv4(7,1,1,0), 24, 6},
 354         {IPv4(8,1,1,0), 24, 7},
 355 };
 356
 357 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 358         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 359         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 360         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 361         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 362         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 363         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 364         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 365         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 366 };
 367
 368 #define IPV4_L3FWD_NUM_ROUTES \
 369         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 370 #define IPV6_L3FWD_NUM_ROUTES \
 371         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 372
 373 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 374 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 375 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 376
 377 typedef struct rte_lpm lookup_struct_t;
 378 typedef struct rte_lpm6 lookup6_struct_t;
 379 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 380 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 381 #endif
 382
 383 struct lcore_conf {
 384         uint16_t n_rx_queue;
 385         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 386         uint16_t tx_queue_id[MAX_PORTS];
 387         struct mbuf_table tx_mbufs[MAX_PORTS];
 388         lookup_struct_t * ipv4_lookup_struct;
 389 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 390         lookup6_struct_t * ipv6_lookup_struct;
 391 #else
 392         lookup_struct_t * ipv6_lookup_struct;
 393 #endif
 394         struct ipv4_frag_tbl *frag_tbl[MAX_RX_QUEUE_PER_LCORE];
 395         struct rte_mempool *pool[MAX_RX_QUEUE_PER_LCORE];
 396 } __rte_cache_aligned;
 397
 398 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 399
 400 /* Send burst of packets on an output interface */
 401 static inline int
 402 send_burst(struct lcore_conf *qconf, uint16_t n, uint8_t port)
 403 {
 404         struct rte_mbuf **m_table;
 405         int ret;
 406         uint16_t queueid;
 407
 408         queueid = qconf->tx_queue_id[port];
 409         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 410
 411         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 412         if (unlikely(ret < n)) {
 413                 do {
 414                         rte_pktmbuf_free(m_table[ret]);
 415                 } while (++ret < n);
 416         }
 417
 418         return 0;
 419 }
 420
 421 /* Enqueue a single packet, and send burst if queue is filled */
 422 static inline int
 423 send_single_packet(struct rte_mbuf *m, uint8_t port)
 424 {
 425         uint32_t lcore_id;
 426         uint16_t len;
 427         struct lcore_conf *qconf;
 428
 429         lcore_id = rte_lcore_id();
 430
 431         qconf = &lcore_conf[lcore_id];
 432         len = qconf->tx_mbufs[port].len;
 433         qconf->tx_mbufs[port].m_table[len] = m;
 434         len++;
 435
 436         /* enough pkts to be sent */
 437         if (unlikely(len == MAX_PKT_BURST)) {
 438                 send_burst(qconf, MAX_PKT_BURST, port);
 439                 len = 0;
 440         }
 441
 442         qconf->tx_mbufs[port].len = len;
 443         return 0;
 444 }
 445
 446 #ifdef DO_RFC_1812_CHECKS
 447 static inline int
 448 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 449 {
 450         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 451         /*
 452          * 1. The packet length reported by the Link Layer must be large
 453          * enough to hold the minimum length legal IP datagram (20 bytes).
 454          */
 455         if (link_len < sizeof(struct ipv4_hdr))
 456                 return -1;
 457
 458         /* 2. The IP checksum must be correct. */
 459         /* this is checked in H/W */
 460
 461         /*
 462          * 3. The IP version number must be 4. If the version number is not 4
 463          * then the packet may be another version of IP, such as IPng or
 464          * ST-II.
 465          */
 466         if (((pkt->version_ihl) >> 4) != 4)
 467                 return -3;
 468         /*
 469          * 4. The IP header length field must be large enough to hold the
 470          * minimum length legal IP datagram (20 bytes = 5 words).
 471          */
 472         if ((pkt->version_ihl & 0xf) < 5)
 473                 return -4;
 474
 475         /*
 476          * 5. The IP total length field must be large enough to hold the IP
 477          * datagram header, whose length is specified in the IP header length
 478          * field.
 479          */
 480         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 481                 return -5;
 482
 483         return 0;
 484 }
 485 #endif
 486
 487 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 488 static void
 489 print_ipv4_key(struct ipv4_5tuple key)
 490 {
 491         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 492                         (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 493 }
 494 static void
 495 print_ipv6_key(struct ipv6_5tuple key)
 496 {
 497         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 498                 "port dst = %d, port src = %d, proto = %d\n",
 499                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 500                 key.port_dst, key.port_src, key.proto);
 501 }
 502
 503 static inline uint8_t
 504 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 505 {
 506         struct ipv4_5tuple key;
 507         struct tcp_hdr *tcp;
 508         struct udp_hdr *udp;
 509         int ret = 0;
 510
 511         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 512         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 513         key.proto = ipv4_hdr->next_proto_id;
 514
 515         switch (ipv4_hdr->next_proto_id) {
 516         case IPPROTO_TCP:
 517                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 518                                         sizeof(struct ipv4_hdr));
 519                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 520                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 521                 break;
 522
 523         case IPPROTO_UDP:
 524                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 525                                         sizeof(struct ipv4_hdr));
 526                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 527                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 528                 break;
 529
 530         default:
 531                 key.port_dst = 0;
 532                 key.port_src = 0;
 533                 break;
 534         }
 535
 536         /* Find destination port */
 537         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 538         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 539 }
 540
 541 static inline uint8_t
 542 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 543 {
 544         struct ipv6_5tuple key;
 545         struct tcp_hdr *tcp;
 546         struct udp_hdr *udp;
 547         int ret = 0;
 548
 549         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 550         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 551
 552         key.proto = ipv6_hdr->proto;
 553
 554         switch (ipv6_hdr->proto) {
 555         case IPPROTO_TCP:
 556                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 557                                         sizeof(struct ipv6_hdr));
 558                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 559                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 560                 break;
 561
 562         case IPPROTO_UDP:
 563                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 564                                         sizeof(struct ipv6_hdr));
 565                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 566                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 567                 break;
 568
 569         default:
 570                 key.port_dst = 0;
 571                 key.port_src = 0;
 572                 break;
 573         }
 574
 575         /* Find destination port */
 576         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 577         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 578 }
 579 #endif
 580
 581 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 582 static inline uint8_t
 583 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 584 {
 585         uint8_t next_hop;
 586
 587         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 588                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 589                         next_hop : portid);
 590 }
 591
 592 static inline uint8_t
 593 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 594 {
 595         uint8_t next_hop;
 596
 597         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 598                         ipv6_hdr->dst_addr, &next_hop) == 0)?
 599                         next_hop : portid);
 600 }
 601 #endif
 602
 603 static inline void
 604 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
 605         struct lcore_conf *qconf, uint64_t tms)
 606 {
 607         struct ether_hdr *eth_hdr;
 608         struct ipv4_hdr *ipv4_hdr;
 609         void *d_addr_bytes;
 610         uint8_t dst_port;
 611         uint16_t flag_offset, ip_flag, ip_ofs;
 612
 613         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 614
 615         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 616                 /* Handle IPv4 headers.*/
 617                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 618
 619 #ifdef DO_RFC_1812_CHECKS
 620                 /* Check to make sure the packet is valid (RFC1812) */
 621                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 622                         rte_pktmbuf_free(m);
 623                         return;
 624                 }
 625
 626                 /* Update time to live and header checksum */
 627                 --(ipv4_hdr->time_to_live);
 628                 ++(ipv4_hdr->hdr_checksum);
 629 #endif
 630
 631                 flag_offset = rte_be_to_cpu_16(ipv4_hdr->fragment_offset);
 632                 ip_ofs = (uint16_t)(flag_offset & IPV4_HDR_OFFSET_MASK);
 633                 ip_flag = (uint16_t)(flag_offset & IPV4_HDR_MF_FLAG);
 634
 635                  /* if it is a fragmented packet, then try to reassemble. */
 636                 if (ip_flag != 0 || ip_ofs  != 0) {
 637
 638                         struct rte_mbuf *mo;
 639                         struct ipv4_frag_tbl *tbl;
 640
 641                         tbl = qconf->frag_tbl[queue];
 642
 643                         /* prepare mbuf: setup l2_len/l3_len. */
 644                         m->pkt.vlan_macip.f.l2_len = sizeof(*eth_hdr);
 645                         m->pkt.vlan_macip.f.l3_len = sizeof(*ipv4_hdr);
 646
 647                         /* process this fragment. */
 648                         if ((mo = ipv4_frag_mbuf(tbl, m, tms, ipv4_hdr,
 649                                         ip_ofs, ip_flag)) == NULL)
 650                                 /* no packet to send out. */
 651                                 return;
 652
 653                         /* we have our packet reassembled. */
 654                         if (mo != m) {
 655                                 m = mo;
 656                                 eth_hdr = rte_pktmbuf_mtod(m,
 657                                         struct ether_hdr *);
 658                                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 659                         }
 660                 }
 661
 662                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid, qconf->ipv4_lookup_struct);
 663                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 664                         dst_port = portid;
 665
 666                 /* 02:00:00:00:00:xx */
 667                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 668                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 669
 670                 /* src addr */
 671                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 672
 673                 send_single_packet(m, dst_port);
 674         }
 675         else {
 676                 /* Handle IPv6 headers.*/
 677                 struct ipv6_hdr *ipv6_hdr;
 678
 679                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 680                                 sizeof(struct ether_hdr));
 681
 682                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 683
 684                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 685                         dst_port = portid;
 686
 687                 /* 02:00:00:00:00:xx */
 688                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 689                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 690
 691                 /* src addr */
 692                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 693
 694                 send_single_packet(m, dst_port);
 695         }
 696
 697 }
 698
 699 /* main processing loop */
 700 static int
 701 main_loop(__attribute__((unused)) void *dummy)
 702 {
 703         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 704         unsigned lcore_id;
 705         uint64_t diff_tsc, cur_tsc, prev_tsc;
 706         int i, j, nb_rx;
 707         uint8_t portid, queueid;
 708         struct lcore_conf *qconf;
 709         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 710
 711         prev_tsc = 0;
 712
 713         lcore_id = rte_lcore_id();
 714         qconf = &lcore_conf[lcore_id];
 715
 716         if (qconf->n_rx_queue == 0) {
 717                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 718                 return 0;
 719         }
 720
 721         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 722
 723         for (i = 0; i < qconf->n_rx_queue; i++) {
 724
 725                 portid = qconf->rx_queue_list[i].port_id;
 726                 queueid = qconf->rx_queue_list[i].queue_id;
 727                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 728                         portid, queueid);
 729         }
 730
 731         while (1) {
 732
 733                 cur_tsc = rte_rdtsc();
 734
 735                 /*
 736                  * TX burst queue drain
 737                  */
 738                 diff_tsc = cur_tsc - prev_tsc;
 739                 if (unlikely(diff_tsc > drain_tsc)) {
 740
 741                         /*
 742                          * This could be optimized (use queueid instead of
 743                          * portid), but it is not called so often
 744                          */
 745                         for (portid = 0; portid < MAX_PORTS; portid++) {
 746                                 if (qconf->tx_mbufs[portid].len == 0)
 747                                         continue;
 748                                 send_burst(&lcore_conf[lcore_id],
 749                                         qconf->tx_mbufs[portid].len,
 750                                         portid);
 751                                 qconf->tx_mbufs[portid].len = 0;
 752                         }
 753
 754                         prev_tsc = cur_tsc;
 755                 }
 756
 757                 /*
 758                  * Read packet from RX queues
 759                  */
 760                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 761
 762                         portid = qconf->rx_queue_list[i].port_id;
 763                         queueid = qconf->rx_queue_list[i].queue_id;
 764
 765                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 766                                 MAX_PKT_BURST);
 767
 768                         /* Prefetch first packets */
 769                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 770                                 rte_prefetch0(rte_pktmbuf_mtod(
 771                                                 pkts_burst[j], void *));
 772                         }
 773
 774                         /* Prefetch and forward already prefetched packets */
 775                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 776                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 777                                         j + PREFETCH_OFFSET], void *));
 778                                 l3fwd_simple_forward(pkts_burst[j], portid,
 779                                         i, qconf, cur_tsc);
 780                         }
 781
 782                         /* Forward remaining prefetched packets */
 783                         for (; j < nb_rx; j++) {
 784                                 l3fwd_simple_forward(pkts_burst[j], portid,
 785                                         i, qconf, cur_tsc);
 786                         }
 787                 }
 788         }
 789 }
 790
 791 static int
 792 check_lcore_params(void)
 793 {
 794         uint8_t queue, lcore;
 795         uint16_t i;
 796         int socketid;
 797
 798         for (i = 0; i < nb_lcore_params; ++i) {
 799                 queue = lcore_params[i].queue_id;
 800                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 801                         printf("invalid queue number: %hhu\n", queue);
 802                         return -1;
 803                 }
 804                 lcore = lcore_params[i].lcore_id;
 805                 if (!rte_lcore_is_enabled(lcore)) {
 806                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 807                         return -1;
 808                 }
 809                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 810                         (numa_on == 0)) {
 811                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 812                                 lcore, socketid);
 813                 }
 814         }
 815         return 0;
 816 }
 817
 818 static int
 819 check_port_config(const unsigned nb_ports)
 820 {
 821         unsigned portid;
 822         uint16_t i;
 823
 824         for (i = 0; i < nb_lcore_params; ++i) {
 825                 portid = lcore_params[i].port_id;
 826                 if ((enabled_port_mask & (1 << portid)) == 0) {
 827                         printf("port %u is not enabled in port mask\n", portid);
 828                         return -1;
 829                 }
 830                 if (portid >= nb_ports) {
 831                         printf("port %u is not present on the board\n", portid);
 832                         return -1;
 833                 }
 834         }
 835         return 0;
 836 }
 837
 838 static uint8_t
 839 get_port_n_rx_queues(const uint8_t port)
 840 {
 841         int queue = -1;
 842         uint16_t i;
 843
 844         for (i = 0; i < nb_lcore_params; ++i) {
 845                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 846                         queue = lcore_params[i].queue_id;
 847         }
 848         return (uint8_t)(++queue);
 849 }
 850
 851 static int
 852 init_lcore_rx_queues(void)
 853 {
 854         uint16_t i, nb_rx_queue;
 855         uint8_t lcore;
 856
 857         for (i = 0; i < nb_lcore_params; ++i) {
 858                 lcore = lcore_params[i].lcore_id;
 859                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 860                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 861                         printf("error: too many queues (%u) for lcore: %u\n",
 862                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 863                         return -1;
 864                 } else {
 865                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 866                                 lcore_params[i].port_id;
 867                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 868                                 lcore_params[i].queue_id;
 869                         lcore_conf[lcore].n_rx_queue++;
 870                 }
 871         }
 872         return 0;
 873 }
 874
 875 /* display usage */
 876 static void
 877 print_usage(const char *prgname)
 878 {
 879         printf ("%s [EAL options] -- -p PORTMASK -P"
 880                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
 881                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]"
 882                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 883                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 884                 "  -P : enable promiscuous mode\n"
 885                 "  --config (port,queue,lcore): rx queues configuration\n"
 886                 "  --no-numa: optional, disable numa awareness\n"
 887                 "  --enable-jumbo: enable jumbo frame"
 888                 " which max packet len is PKTLEN in decimal (64-9600)\n"
 889                 "  --maxflows=<flows>: optional, maximum number of flows "
 890                 "supported\n"
 891                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 892                 "flow\n",
 893                 prgname);
 894 }
 895
 896 static uint32_t
 897 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 898 {
 899         char *end;
 900         uint64_t v;
 901
 902         /* parse decimal string */
 903         errno = 0;
 904         v = strtoul(str, &end, 10);
 905         if (errno != 0 || *end != '\0')
 906                 return (-EINVAL);
 907
 908         if (v < min || v > max)
 909                 return (-EINVAL);
 910
 911         *val = (uint32_t)v;
 912         return (0);
 913 }
 914
 915 static int
 916 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 917 {
 918         char *end;
 919         uint64_t v;
 920
 921         static const char frmt_sec[] = "s";
 922         static const char frmt_msec[] = "ms";
 923
 924         /* parse decimal string */
 925         errno = 0;
 926         v = strtoul(str, &end, 10);
 927         if (errno != 0)
 928                 return (-EINVAL);
 929
 930         if (*end != '\0') {
 931                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 932                         v *= MS_PER_S;
 933                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 934                         return (-EINVAL);
 935         }
 936
 937         if (v < min || v > max)
 938                 return (-EINVAL);
 939
 940         *val = (uint32_t)v;
 941         return (0);
 942 }
 943
 944
 945 static int parse_max_pkt_len(const char *pktlen)
 946 {
 947         char *end = NULL;
 948         unsigned long len;
 949
 950         /* parse decimal string */
 951         len = strtoul(pktlen, &end, 10);
 952         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
 953                 return -1;
 954
 955         if (len == 0)
 956                 return -1;
 957
 958         return len;
 959 }
 960
 961 static int
 962 parse_portmask(const char *portmask)
 963 {
 964         char *end = NULL;
 965         unsigned long pm;
 966
 967         /* parse hexadecimal string */
 968         pm = strtoul(portmask, &end, 16);
 969         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 970                 return -1;
 971
 972         if (pm == 0)
 973                 return -1;
 974
 975         return pm;
 976 }
 977
 978 static int
 979 parse_config(const char *q_arg)
 980 {
 981         char s[256];
 982         const char *p, *p0 = q_arg;
 983         char *end;
 984         enum fieldnames {
 985                 FLD_PORT = 0,
 986                 FLD_QUEUE,
 987                 FLD_LCORE,
 988                 _NUM_FLD
 989         };
 990         unsigned long int_fld[_NUM_FLD];
 991         char *str_fld[_NUM_FLD];
 992         int i;
 993         unsigned size;
 994
 995         nb_lcore_params = 0;
 996
 997         while ((p = strchr(p0,'(')) != NULL) {
 998                 ++p;
 999                 if((p0 = strchr(p,')')) == NULL)
1000                         return -1;
1001
1002                 size = p0 - p;
1003                 if(size >= sizeof(s))
1004                         return -1;
1005
1006                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1007                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1008                         return -1;
1009                 for (i = 0; i < _NUM_FLD; i++){
1010                         errno = 0;
1011                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1012                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1013                                 return -1;
1014                 }
1015                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1016                         printf("exceeded max number of lcore params: %hu\n",
1017                                 nb_lcore_params);
1018                         return -1;
1019                 }
1020                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1021                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1022                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1023                 ++nb_lcore_params;
1024         }
1025         lcore_params = lcore_params_array;
1026         return 0;
1027 }
1028
1029 /* Parse the argument given in the command line of the application */
1030 static int
1031 parse_args(int argc, char **argv)
1032 {
1033         int opt, ret;
1034         char **argvopt;
1035         int option_index;
1036         char *prgname = argv[0];
1037         static struct option lgopts[] = {
1038                 {"config", 1, 0, 0},
1039                 {"no-numa", 0, 0, 0},
1040                 {"enable-jumbo", 0, 0, 0},
1041                 {"maxflows", 1, 0, 0},
1042                 {"flowttl", 1, 0, 0},
1043                 {NULL, 0, 0, 0}
1044         };
1045
1046         argvopt = argv;
1047
1048         while ((opt = getopt_long(argc, argvopt, "p:P",
1049                                 lgopts, &option_index)) != EOF) {
1050
1051                 switch (opt) {
1052                 /* portmask */
1053                 case 'p':
1054                         enabled_port_mask = parse_portmask(optarg);
1055                         if (enabled_port_mask == 0) {
1056                                 printf("invalid portmask\n");
1057                                 print_usage(prgname);
1058                                 return -1;
1059                         }
1060                         break;
1061                 case 'P':
1062                         printf("Promiscuous mode selected\n");
1063                         promiscuous_on = 1;
1064                         break;
1065
1066                 /* long options */
1067                 case 0:
1068                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1069                                 ret = parse_config(optarg);
1070                                 if (ret) {
1071                                         printf("invalid config\n");
1072                                         print_usage(prgname);
1073                                         return -1;
1074                                 }
1075                         }
1076
1077                         if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
1078                                 printf("numa is disabled \n");
1079                                 numa_on = 0;
1080                         }
1081
1082                         if (!strncmp(lgopts[option_index].name,
1083                                         "maxflows", 8)) {
1084                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
1085                                                 MAX_FLOW_NUM,
1086                                                 &max_flow_num)) != 0) {
1087                                         printf("invalid value: \"%s\" for "
1088                                                 "parameter %s\n",
1089                                                 optarg,
1090                                                 lgopts[option_index].name);
1091                                         print_usage(prgname);
1092                                         return (ret);
1093                                 }
1094                         }
1095
1096                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
1097                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
1098                                                 MAX_FLOW_TTL,
1099                                                 &max_flow_ttl)) != 0) {
1100                                         printf("invalid value: \"%s\" for "
1101                                                 "parameter %s\n",
1102                                                 optarg,
1103                                                 lgopts[option_index].name);
1104                                         print_usage(prgname);
1105                                         return (ret);
1106                                 }
1107                         }
1108
1109                         if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
1110                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1111
1112                                 printf("jumbo frame is enabled \n");
1113                                 port_conf.rxmode.jumbo_frame = 1;
1114
1115                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1116                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1117                                         ret = parse_max_pkt_len(optarg);
1118                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1119                                                 printf("invalid packet length\n");
1120                                                 print_usage(prgname);
1121                                                 return -1;
1122                                         }
1123                                         port_conf.rxmode.max_rx_pkt_len = ret;
1124                                 }
1125                                 printf("set jumbo frame max packet length to %u\n",
1126                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1127                         }
1128
1129                         break;
1130
1131                 default:
1132                         print_usage(prgname);
1133                         return -1;
1134                 }
1135         }
1136
1137         if (optind >= 0)
1138                 argv[optind-1] = prgname;
1139
1140         ret = optind-1;
1141         optind = 0; /* reset getopt lib */
1142         return ret;
1143 }
1144
1145 static void
1146 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1147 {
1148         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1149                 eth_addr->addr_bytes[0],
1150                 eth_addr->addr_bytes[1],
1151                 eth_addr->addr_bytes[2],
1152                 eth_addr->addr_bytes[3],
1153                 eth_addr->addr_bytes[4],
1154                 eth_addr->addr_bytes[5]);
1155 }
1156
1157 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1158 static void
1159 setup_hash(int socketid)
1160 {
1161         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1162                 .name = NULL,
1163                 .entries = L3FWD_HASH_ENTRIES,
1164                 .bucket_entries = 4,
1165                 .key_len = sizeof(struct ipv4_5tuple),
1166                 .hash_func = DEFAULT_HASH_FUNC,
1167                 .hash_func_init_val = 0,
1168         };
1169
1170         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1171                 .name = NULL,
1172                 .entries = L3FWD_HASH_ENTRIES,
1173                 .bucket_entries = 4,
1174                 .key_len = sizeof(struct ipv6_5tuple),
1175                 .hash_func = DEFAULT_HASH_FUNC,
1176                 .hash_func_init_val = 0,
1177         };
1178
1179         unsigned i;
1180         int ret;
1181         char s[64];
1182
1183         /* create ipv4 hash */
1184         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1185         ipv4_l3fwd_hash_params.name = s;
1186         ipv4_l3fwd_hash_params.socket_id = socketid;
1187         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1188         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1189                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1190                                 "socket %d\n", socketid);
1191
1192         /* create ipv6 hash */
1193         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1194         ipv6_l3fwd_hash_params.name = s;
1195         ipv6_l3fwd_hash_params.socket_id = socketid;
1196         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1197         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1198                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1199                                 "socket %d\n", socketid);
1200
1201
1202         /* populate the ipv4 hash */
1203         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1204                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1205                                 (void *) &ipv4_l3fwd_route_array[i].key);
1206                 if (ret < 0) {
1207                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1208                                 "l3fwd hash on socket %d\n", i, socketid);
1209                 }
1210                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1211                 printf("Hash: Adding key\n");
1212                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1213         }
1214
1215         /* populate the ipv6 hash */
1216         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1217                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1218                                 (void *) &ipv6_l3fwd_route_array[i].key);
1219                 if (ret < 0) {
1220                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1221                                 "l3fwd hash on socket %d\n", i, socketid);
1222                 }
1223                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1224                 printf("Hash: Adding key\n");
1225                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1226         }
1227 }
1228 #endif
1229
1230 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1231 static void
1232 setup_lpm(int socketid)
1233 {
1234         struct rte_lpm6_config config;
1235         unsigned i;
1236         int ret;
1237         char s[64];
1238
1239         /* create the LPM table */
1240         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1241         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1242                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1243         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1244                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1245                                 " on socket %d\n", socketid);
1246
1247         /* populate the LPM table */
1248         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1249                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1250                         ipv4_l3fwd_route_array[i].ip,
1251                         ipv4_l3fwd_route_array[i].depth,
1252                         ipv4_l3fwd_route_array[i].if_out);
1253
1254                 if (ret < 0) {
1255                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1256                                 "l3fwd LPM table on socket %d\n",
1257                                 i, socketid);
1258                 }
1259
1260                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1261                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1262                         ipv4_l3fwd_route_array[i].depth,
1263                         ipv4_l3fwd_route_array[i].if_out);
1264         }
1265
1266         /* create the LPM6 table */
1267         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
1268
1269         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
1270         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
1271         config.flags = 0;
1272         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
1273                                 &config);
1274         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1275                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1276                                 " on socket %d\n", socketid);
1277
1278         /* populate the LPM table */
1279         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1280                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
1281                         ipv6_l3fwd_route_array[i].ip,
1282                         ipv6_l3fwd_route_array[i].depth,
1283                         ipv6_l3fwd_route_array[i].if_out);
1284
1285                 if (ret < 0) {
1286                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1287                                 "l3fwd LPM table on socket %d\n",
1288                                 i, socketid);
1289                 }
1290
1291                 printf("LPM: Adding route %s / %d (%d)\n",
1292                         "IPV6",
1293                         ipv6_l3fwd_route_array[i].depth,
1294                         ipv6_l3fwd_route_array[i].if_out);
1295         }
1296 }
1297 #endif
1298
1299 static int
1300 init_mem(void)
1301 {
1302         struct lcore_conf *qconf;
1303         int socketid;
1304         unsigned lcore_id;
1305
1306         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1307                 if (rte_lcore_is_enabled(lcore_id) == 0)
1308                         continue;
1309
1310                 if (numa_on)
1311                         socketid = rte_lcore_to_socket_id(lcore_id);
1312                 else
1313                         socketid = 0;
1314
1315                 if (socketid >= NB_SOCKETS) {
1316                         rte_exit(EXIT_FAILURE,
1317                                 "Socket %d of lcore %u is out of range %d\n",
1318                                 socketid, lcore_id, NB_SOCKETS);
1319                 }
1320
1321 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1322                         setup_lpm(socketid);
1323 #else
1324                         setup_hash(socketid);
1325 #endif
1326                 qconf = &lcore_conf[lcore_id];
1327                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1328                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1329         }
1330         return 0;
1331 }
1332
1333 /* Check the link status of all ports in up to 9s, and print them finally */
1334 static void
1335 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1336 {
1337 #define CHECK_INTERVAL 100 /* 100ms */
1338 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1339         uint8_t portid, count, all_ports_up, print_flag = 0;
1340         struct rte_eth_link link;
1341
1342         printf("\nChecking link status");
1343         fflush(stdout);
1344         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1345                 all_ports_up = 1;
1346                 for (portid = 0; portid < port_num; portid++) {
1347                         if ((port_mask & (1 << portid)) == 0)
1348                                 continue;
1349                         memset(&link, 0, sizeof(link));
1350                         rte_eth_link_get_nowait(portid, &link);
1351                         /* print link status if flag set */
1352                         if (print_flag == 1) {
1353                                 if (link.link_status)
1354                                         printf("Port %d Link Up - speed %u "
1355                                                 "Mbps - %s\n", (uint8_t)portid,
1356                                                 (unsigned)link.link_speed,
1357                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1358                                         ("full-duplex") : ("half-duplex\n"));
1359                                 else
1360                                         printf("Port %d Link Down\n",
1361                                                 (uint8_t)portid);
1362                                 continue;
1363                         }
1364                         /* clear all_ports_up flag if any link down */
1365                         if (link.link_status == 0) {
1366                                 all_ports_up = 0;
1367                                 break;
1368                         }
1369                 }
1370                 /* after finally printing all link status, get out */
1371                 if (print_flag == 1)
1372                         break;
1373
1374                 if (all_ports_up == 0) {
1375                         printf(".");
1376                         fflush(stdout);
1377                         rte_delay_ms(CHECK_INTERVAL);
1378                 }
1379
1380                 /* set the print_flag if all ports up or timeout */
1381                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1382                         print_flag = 1;
1383                         printf("done\n");
1384                 }
1385         }
1386 }
1387
1388 static void
1389 setup_queue_frag_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1390         uint32_t queue)
1391 {
1392         uint32_t nb_mbuf;
1393         uint64_t frag_cycles;
1394         char buf[RTE_MEMPOOL_NAMESIZE];
1395
1396         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
1397                 max_flow_ttl;
1398
1399         if ((qconf->frag_tbl[queue] = ipv4_frag_tbl_create(max_flow_num,
1400                         IPV4_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
1401                         socket)) == NULL)
1402                 rte_exit(EXIT_FAILURE, "ipv4_frag_tbl_create(%u) on "
1403                         "lcore: %u for queue: %u failed\n",
1404                         max_flow_num, lcore, queue);
1405
1406         nb_mbuf = max_flow_num * MAX_FRAG_NUM;
1407         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
1408         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + MAX_PKT_BURST +
1409                 RTE_TEST_TX_DESC_DEFAULT;
1410         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)DEF_MBUF_NUM);
1411
1412         rte_snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
1413
1414         if ((qconf->pool[queue] = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
1415                         sizeof(struct rte_pktmbuf_pool_private),
1416                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
1417                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL)
1418                 rte_exit(EXIT_FAILURE, "mempool_create(%s) failed", buf);
1419 }
1420
1421 static void
1422 queue_frag_tbl_dump_stat(void)
1423 {
1424         uint32_t i, lcore;
1425         const struct lcore_conf *qconf;
1426
1427         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
1428                 if (rte_lcore_is_enabled(lcore) == 0)
1429                         continue;
1430
1431                 qconf = lcore_conf + lcore;
1432                 for (i = 0; i < qconf->n_rx_queue; i++) {
1433
1434                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
1435                                 "rxqueueid=%hhu frag tbl stat:\n",
1436                                 lcore,  qconf->rx_queue_list[i].port_id,
1437                                 qconf->rx_queue_list[i].queue_id);
1438                         ipv4_frag_tbl_dump_stat(stdout, qconf->frag_tbl[i]);
1439                 }
1440         }
1441 }
1442
1443 static void
1444 signal_handler(int signum)
1445 {
1446         queue_frag_tbl_dump_stat();
1447         if (signum != SIGUSR1)
1448                 rte_exit(0, "received signal: %d, exiting\n", signum);
1449 }
1450
1451 int
1452 MAIN(int argc, char **argv)
1453 {
1454         struct lcore_conf *qconf;
1455         int ret;
1456         unsigned nb_ports;
1457         uint16_t queueid;
1458         unsigned lcore_id;
1459         uint32_t n_tx_queue, nb_lcores;
1460         uint8_t portid, nb_rx_queue, queue, socketid;
1461
1462         /* init EAL */
1463         ret = rte_eal_init(argc, argv);
1464         if (ret < 0)
1465                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1466         argc -= ret;
1467         argv += ret;
1468
1469         /* parse application arguments (after the EAL ones) */
1470         ret = parse_args(argc, argv);
1471         if (ret < 0)
1472                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1473
1474         if (check_lcore_params() < 0)
1475                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1476
1477         ret = init_lcore_rx_queues();
1478         if (ret < 0)
1479                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1480
1481
1482         /* init driver(s) */
1483         if (rte_pmd_init_all() < 0)
1484                 rte_exit(EXIT_FAILURE, "Cannot init pmd\n");
1485
1486         if (rte_eal_pci_probe() < 0)
1487                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1488
1489         nb_ports = rte_eth_dev_count();
1490         if (nb_ports > MAX_PORTS)
1491                 nb_ports = MAX_PORTS;
1492
1493         if (check_port_config(nb_ports) < 0)
1494                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1495
1496         nb_lcores = rte_lcore_count();
1497
1498         /* initialize all ports */
1499         for (portid = 0; portid < nb_ports; portid++) {
1500                 /* skip ports that are not enabled */
1501                 if ((enabled_port_mask & (1 << portid)) == 0) {
1502                         printf("\nSkipping disabled port %d\n", portid);
1503                         continue;
1504                 }
1505
1506                 /* init port */
1507                 printf("Initializing port %d ... ", portid );
1508                 fflush(stdout);
1509
1510                 nb_rx_queue = get_port_n_rx_queues(portid);
1511                 n_tx_queue = nb_lcores;
1512                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1513                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1514                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1515                         nb_rx_queue, (unsigned)n_tx_queue );
1516                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1517                                         (uint16_t)n_tx_queue, &port_conf);
1518                 if (ret < 0)
1519                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1520                                 ret, portid);
1521
1522                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1523                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1524                 printf(", ");
1525
1526                 /* init memory */
1527                 ret = init_mem();
1528                 if (ret < 0)
1529                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1530
1531                 /* init one TX queue per couple (lcore,port) */
1532                 queueid = 0;
1533                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1534                         if (rte_lcore_is_enabled(lcore_id) == 0)
1535                                 continue;
1536
1537                         if (numa_on)
1538                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1539                         else
1540                                 socketid = 0;
1541
1542                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1543                         fflush(stdout);
1544                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1545                                                      socketid, &tx_conf);
1546                         if (ret < 0)
1547                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1548                                         "port=%d\n", ret, portid);
1549
1550                         qconf = &lcore_conf[lcore_id];
1551                         qconf->tx_queue_id[portid] = queueid;
1552                         queueid++;
1553                 }
1554                 printf("\n");
1555         }
1556
1557         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1558                 if (rte_lcore_is_enabled(lcore_id) == 0)
1559                         continue;
1560                 qconf = &lcore_conf[lcore_id];
1561                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1562                 fflush(stdout);
1563                 /* init RX queues */
1564                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1565                         portid = qconf->rx_queue_list[queue].port_id;
1566                         queueid = qconf->rx_queue_list[queue].queue_id;
1567
1568                         if (numa_on)
1569                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1570                         else
1571                                 socketid = 0;
1572
1573                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1574                         fflush(stdout);
1575
1576                         setup_queue_frag_tbl(qconf, lcore_id, socketid, queue);
1577
1578                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1579                                         socketid, &rx_conf, qconf->pool[queue]);
1580                         if (ret < 0)
1581                                 rte_exit(EXIT_FAILURE,
1582                                         "rte_eth_rx_queue_setup: err=%d,"
1583                                         "port=%d\n", ret, portid);
1584                 }
1585         }
1586
1587         printf("\n");
1588
1589         /* start ports */
1590         for (portid = 0; portid < nb_ports; portid++) {
1591                 if ((enabled_port_mask & (1 << portid)) == 0) {
1592                         continue;
1593                 }
1594                 /* Start device */
1595                 ret = rte_eth_dev_start(portid);
1596                 if (ret < 0)
1597                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1598                                 ret, portid);
1599
1600                 /*
1601                  * If enabled, put device in promiscuous mode.
1602                  * This allows IO forwarding mode to forward packets
1603                  * to itself through 2 cross-connected  ports of the
1604                  * target machine.
1605                  */
1606                 if (promiscuous_on)
1607                         rte_eth_promiscuous_enable(portid);
1608         }
1609
1610         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1611
1612         signal(SIGUSR1, signal_handler);
1613         signal(SIGTERM, signal_handler);
1614         signal(SIGINT, signal_handler);
1615
1616         /* launch per-lcore init on every lcore */
1617         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1618         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1619                 if (rte_eal_wait_lcore(lcore_id) < 0)
1620                         return -1;
1621         }
1622
1623         return 0;
1624 }