examples/ip_reassembly/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <signal.h>
  45
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_malloc.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76 #include "main.h"
  77
  78 #define APP_LOOKUP_EXACT_MATCH          0
  79 #define APP_LOOKUP_LPM                  1
  80 #define DO_RFC_1812_CHECKS
  81
  82 #ifndef APP_LOOKUP_METHOD
  83 #define APP_LOOKUP_METHOD             APP_LOOKUP_LPM
  84 #endif
  85
  86 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
  87 #include <rte_hash.h>
  88 #elif (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
  89 #include <rte_lpm.h>
  90 #include <rte_lpm6.h>
  91 #else
  92 #error "APP_LOOKUP_METHOD set to incorrect value"
  93 #endif
  94
  95 #define MAX_PKT_BURST 32
  96
  97 #include "rte_ip_frag.h"
  98
  99 #ifndef IPv6_BYTES
 100 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 101                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 102 #define IPv6_BYTES(addr) \
 103         addr[0],  addr[1], addr[2],  addr[3], \
 104         addr[4],  addr[5], addr[6],  addr[7], \
 105         addr[8],  addr[9], addr[10], addr[11],\
 106         addr[12], addr[13],addr[14], addr[15]
 107 #endif
 108
 109
 110 #define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1
 111
 112 #define MAX_PORTS       RTE_MAX_ETHPORTS
 113
 114 #define MAX_JUMBO_PKT_LEN  9600
 115
 116 #define IPV6_ADDR_LEN 16
 117
 118 #define MEMPOOL_CACHE_SIZE 256
 119
 120 #define BUF_SIZE        2048
 121 #define MBUF_SIZE       \
 122         (BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
 123
 124 #define MAX_FLOW_NUM    UINT16_MAX
 125 #define MIN_FLOW_NUM    1
 126 #define DEF_FLOW_NUM    0x1000
 127
 128 /* TTL numbers are in ms. */
 129 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 130 #define MIN_FLOW_TTL    1
 131 #define DEF_FLOW_TTL    MS_PER_S
 132
 133 #define DEF_MBUF_NUM    0x400
 134
 135 /* Should be power of two. */
 136 #define IPV4_FRAG_TBL_BUCKET_ENTRIES    2
 137
 138 static uint32_t max_flow_num = DEF_FLOW_NUM;
 139 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 140
 141 /*
 142  * RX and TX Prefetch, Host, and Write-back threshold values should be
 143  * carefully set for optimal performance. Consult the network
 144  * controller's datasheet and supporting DPDK documentation for guidance
 145  * on how these parameters should be set.
 146  */
 147 #define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
 148 #define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
 149 #define RX_WTHRESH 4 /**< Default values of RX write-back threshold reg. */
 150
 151 /*
 152  * These default values are optimized for use with the Intel(R) 82599 10 GbE
 153  * Controller and the DPDK ixgbe PMD. Consider using other values for other
 154  * network controllers and/or network drivers.
 155  */
 156 #define TX_PTHRESH 36 /**< Default values of TX prefetch threshold reg. */
 157 #define TX_HTHRESH 0  /**< Default values of TX host threshold reg. */
 158 #define TX_WTHRESH 0  /**< Default values of TX write-back threshold reg. */
 159
 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
 173 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 174 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 175
 176 /* ethernet addresses of ports */
 177 static struct ether_addr ports_eth_addr[MAX_PORTS];
 178
 179 /* mask of enabled ports */
 180 static uint32_t enabled_port_mask = 0;
 181 static int promiscuous_on = 0; /**< Ports set in promiscuous mode off by default. */
 182 static int numa_on = 1; /**< NUMA is enabled by default. */
 183
 184 struct mbuf_table {
 185         uint32_t len;
 186         uint32_t head;
 187         uint32_t tail;
 188         struct rte_mbuf *m_table[0];
 189 };
 190
 191 struct lcore_rx_queue {
 192         uint8_t port_id;
 193         uint8_t queue_id;
 194 } __rte_cache_aligned;
 195
 196 #define MAX_RX_QUEUE_PER_LCORE 16
 197 #define MAX_TX_QUEUE_PER_PORT MAX_PORTS
 198 #define MAX_RX_QUEUE_PER_PORT 128
 199
 200 #define MAX_LCORE_PARAMS 1024
 201 struct lcore_params {
 202         uint8_t port_id;
 203         uint8_t queue_id;
 204         uint8_t lcore_id;
 205 } __rte_cache_aligned;
 206
 207 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 208 static struct lcore_params lcore_params_array_default[] = {
 209         {0, 0, 2},
 210         {0, 1, 2},
 211         {0, 2, 2},
 212         {1, 0, 2},
 213         {1, 1, 2},
 214         {1, 2, 2},
 215         {2, 0, 2},
 216         {3, 0, 3},
 217         {3, 1, 3},
 218 };
 219
 220 static struct lcore_params * lcore_params = lcore_params_array_default;
 221 static uint16_t nb_lcore_params = sizeof(lcore_params_array_default) /
 222                                 sizeof(lcore_params_array_default[0]);
 223
 224 static struct rte_eth_conf port_conf = {
 225         .rxmode = {
 226                 .mq_mode        = ETH_MQ_RX_RSS,
 227                 .max_rx_pkt_len = ETHER_MAX_LEN,
 228                 .split_hdr_size = 0,
 229                 .header_split   = 0, /**< Header Split disabled */
 230                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 231                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 232                 .jumbo_frame    = 0, /**< Jumbo Frame Support disabled */
 233                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 234         },
 235         .rx_adv_conf = {
 236                 .rss_conf = {
 237                         .rss_key = NULL,
 238                         .rss_hf = ETH_RSS_IPV4 | ETH_RSS_IPV6,
 239                 },
 240         },
 241         .txmode = {
 242                 .mq_mode = ETH_MQ_TX_NONE,
 243         },
 244 };
 245
 246 static const struct rte_eth_rxconf rx_conf = {
 247         .rx_thresh = {
 248                 .pthresh = RX_PTHRESH,
 249                 .hthresh = RX_HTHRESH,
 250                 .wthresh = RX_WTHRESH,
 251         },
 252         .rx_free_thresh = 32,
 253 };
 254
 255 static const struct rte_eth_txconf tx_conf = {
 256         .tx_thresh = {
 257                 .pthresh = TX_PTHRESH,
 258                 .hthresh = TX_HTHRESH,
 259                 .wthresh = TX_WTHRESH,
 260         },
 261         .tx_free_thresh = 0, /* Use PMD default values */
 262         .tx_rs_thresh = 0, /* Use PMD default values */
 263         .txq_flags = 0x0,
 264 };
 265
 266 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 267
 268 #ifdef RTE_MACHINE_CPUFLAG_SSE4_2
 269 #include <rte_hash_crc.h>
 270 #define DEFAULT_HASH_FUNC       rte_hash_crc
 271 #else
 272 #include <rte_jhash.h>
 273 #define DEFAULT_HASH_FUNC       rte_jhash
 274 #endif
 275
 276 struct ipv4_5tuple {
 277         uint32_t ip_dst;
 278         uint32_t ip_src;
 279         uint16_t port_dst;
 280         uint16_t port_src;
 281         uint8_t  proto;
 282 } __attribute__((__packed__));
 283
 284 struct ipv6_5tuple {
 285         uint8_t  ip_dst[IPV6_ADDR_LEN];
 286         uint8_t  ip_src[IPV6_ADDR_LEN];
 287         uint16_t port_dst;
 288         uint16_t port_src;
 289         uint8_t  proto;
 290 } __attribute__((__packed__));
 291
 292 struct ipv4_l3fwd_route {
 293         struct ipv4_5tuple key;
 294         uint8_t if_out;
 295 };
 296
 297 struct ipv6_l3fwd_route {
 298         struct ipv6_5tuple key;
 299         uint8_t if_out;
 300 };
 301
 302 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 303         {{IPv4(100,10,0,1), IPv4(200,10,0,1), 101, 11, IPPROTO_TCP}, 0},
 304         {{IPv4(100,20,0,2), IPv4(200,20,0,2), 102, 12, IPPROTO_TCP}, 1},
 305         {{IPv4(100,30,0,3), IPv4(200,30,0,3), 103, 13, IPPROTO_TCP}, 2},
 306         {{IPv4(100,40,0,4), IPv4(200,40,0,4), 104, 14, IPPROTO_TCP}, 3},
 307 };
 308
 309 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 310         {
 311                 {
 312                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 313                          0x02, 0x1b, 0x21, 0xff, 0xfe, 0x91, 0x38, 0x05},
 314                         {0xfe, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
 315                          0x02, 0x1e, 0x67, 0xff, 0xfe, 0x0d, 0xb6, 0x0a},
 316                          1, 10, IPPROTO_UDP
 317                 }, 4
 318         },
 319 };
 320
 321 typedef struct rte_hash lookup_struct_t;
 322 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 323 static lookup_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 324
 325 #define L3FWD_HASH_ENTRIES      1024
 326
 327 #define IPV4_L3FWD_NUM_ROUTES \
 328         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 329
 330 #define IPV6_L3FWD_NUM_ROUTES \
 331         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 332
 333 static uint8_t ipv4_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 334 static uint8_t ipv6_l3fwd_out_if[L3FWD_HASH_ENTRIES] __rte_cache_aligned;
 335 #endif
 336
 337 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 338 struct ipv4_l3fwd_route {
 339         uint32_t ip;
 340         uint8_t  depth;
 341         uint8_t  if_out;
 342 };
 343
 344 struct ipv6_l3fwd_route {
 345         uint8_t ip[16];
 346         uint8_t  depth;
 347         uint8_t  if_out;
 348 };
 349
 350 static struct ipv4_l3fwd_route ipv4_l3fwd_route_array[] = {
 351         {IPv4(1,1,1,0), 24, 0},
 352         {IPv4(2,1,1,0), 24, 1},
 353         {IPv4(3,1,1,0), 24, 2},
 354         {IPv4(4,1,1,0), 24, 3},
 355         {IPv4(5,1,1,0), 24, 4},
 356         {IPv4(6,1,1,0), 24, 5},
 357         {IPv4(7,1,1,0), 24, 6},
 358         {IPv4(8,1,1,0), 24, 7},
 359 };
 360
 361 static struct ipv6_l3fwd_route ipv6_l3fwd_route_array[] = {
 362         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 363         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 364         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 365         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 366         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 367         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 368         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 369         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 370 };
 371
 372 #define IPV4_L3FWD_NUM_ROUTES \
 373         (sizeof(ipv4_l3fwd_route_array) / sizeof(ipv4_l3fwd_route_array[0]))
 374 #define IPV6_L3FWD_NUM_ROUTES \
 375         (sizeof(ipv6_l3fwd_route_array) / sizeof(ipv6_l3fwd_route_array[0]))
 376
 377 #define IPV4_L3FWD_LPM_MAX_RULES         1024
 378 #define IPV6_L3FWD_LPM_MAX_RULES         1024
 379 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
 380
 381 typedef struct rte_lpm lookup_struct_t;
 382 typedef struct rte_lpm6 lookup6_struct_t;
 383 static lookup_struct_t *ipv4_l3fwd_lookup_struct[NB_SOCKETS];
 384 static lookup6_struct_t *ipv6_l3fwd_lookup_struct[NB_SOCKETS];
 385 #endif
 386
 387 struct tx_lcore_stat {
 388         uint64_t call;
 389         uint64_t drop;
 390         uint64_t queue;
 391         uint64_t send;
 392 };
 393
 394 #ifdef IPV4_FRAG_TBL_STAT
 395 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 396 #else
 397 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 398 #endif /* IPV4_FRAG_TBL_STAT */
 399
 400 struct lcore_conf {
 401         uint16_t n_rx_queue;
 402         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 403         uint16_t tx_queue_id[MAX_PORTS];
 404         lookup_struct_t * ipv4_lookup_struct;
 405 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 406         lookup6_struct_t * ipv6_lookup_struct;
 407 #else
 408         lookup_struct_t * ipv6_lookup_struct;
 409 #endif
 410         struct rte_ip_frag_tbl *frag_tbl[MAX_RX_QUEUE_PER_LCORE];
 411         struct rte_mempool *pool[MAX_RX_QUEUE_PER_LCORE];
 412         struct rte_ip_frag_death_row death_row;
 413         struct mbuf_table *tx_mbufs[MAX_PORTS];
 414         struct tx_lcore_stat tx_stat;
 415 } __rte_cache_aligned;
 416
 417 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 418
 419 /*
 420  * If number of queued packets reached given threahold, then
 421  * send burst of packets on an output interface.
 422  */
 423 static inline uint32_t
 424 send_burst(struct lcore_conf *qconf, uint32_t thresh, uint8_t port)
 425 {
 426         uint32_t fill, len, k, n;
 427         struct mbuf_table *txmb;
 428
 429         txmb = qconf->tx_mbufs[port];
 430         len = txmb->len;
 431
 432         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 433                 fill += len;
 434
 435         if (fill >= thresh) {
 436                 n = RTE_MIN(len - txmb->tail, fill);
 437
 438                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 439                         txmb->m_table + txmb->tail, (uint16_t)n);
 440
 441                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 442                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 443
 444                 fill -= k;
 445                 if ((txmb->tail += k) == len)
 446                         txmb->tail = 0;
 447         }
 448
 449         return (fill);
 450 }
 451
 452 /* Enqueue a single packet, and send burst if queue is filled */
 453 static inline int
 454 send_single_packet(struct rte_mbuf *m, uint8_t port)
 455 {
 456         uint32_t fill, lcore_id, len;
 457         struct lcore_conf *qconf;
 458         struct mbuf_table *txmb;
 459
 460         lcore_id = rte_lcore_id();
 461         qconf = &lcore_conf[lcore_id];
 462
 463         txmb = qconf->tx_mbufs[port];
 464         len = txmb->len;
 465
 466         fill = send_burst(qconf, MAX_PKT_BURST, port);
 467
 468         if (fill == len - 1) {
 469                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 470                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 471                 if (++txmb->tail == len)
 472                         txmb->tail = 0;
 473         }
 474
 475         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 476         txmb->m_table[txmb->head] = m;
 477         if(++txmb->head == len)
 478                 txmb->head = 0;
 479
 480         return (0);
 481 }
 482
 483 #ifdef DO_RFC_1812_CHECKS
 484 static inline int
 485 is_valid_ipv4_pkt(struct ipv4_hdr *pkt, uint32_t link_len)
 486 {
 487         /* From http://www.rfc-editor.org/rfc/rfc1812.txt section 5.2.2 */
 488         /*
 489          * 1. The packet length reported by the Link Layer must be large
 490          * enough to hold the minimum length legal IP datagram (20 bytes).
 491          */
 492         if (link_len < sizeof(struct ipv4_hdr))
 493                 return -1;
 494
 495         /* 2. The IP checksum must be correct. */
 496         /* this is checked in H/W */
 497
 498         /*
 499          * 3. The IP version number must be 4. If the version number is not 4
 500          * then the packet may be another version of IP, such as IPng or
 501          * ST-II.
 502          */
 503         if (((pkt->version_ihl) >> 4) != 4)
 504                 return -3;
 505         /*
 506          * 4. The IP header length field must be large enough to hold the
 507          * minimum length legal IP datagram (20 bytes = 5 words).
 508          */
 509         if ((pkt->version_ihl & 0xf) < 5)
 510                 return -4;
 511
 512         /*
 513          * 5. The IP total length field must be large enough to hold the IP
 514          * datagram header, whose length is specified in the IP header length
 515          * field.
 516          */
 517         if (rte_cpu_to_be_16(pkt->total_length) < sizeof(struct ipv4_hdr))
 518                 return -5;
 519
 520         return 0;
 521 }
 522 #endif
 523
 524 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
 525 static void
 526 print_ipv4_key(struct ipv4_5tuple key)
 527 {
 528         printf("IP dst = %08x, IP src = %08x, port dst = %d, port src = %d, proto = %d\n",
 529                         (unsigned)key.ip_dst, (unsigned)key.ip_src, key.port_dst, key.port_src, key.proto);
 530 }
 531 static void
 532 print_ipv6_key(struct ipv6_5tuple key)
 533 {
 534         printf( "IP dst = " IPv6_BYTES_FMT ", IP src = " IPv6_BYTES_FMT ", "
 535                 "port dst = %d, port src = %d, proto = %d\n",
 536                 IPv6_BYTES(key.ip_dst), IPv6_BYTES(key.ip_src),
 537                 key.port_dst, key.port_src, key.proto);
 538 }
 539
 540 static inline uint8_t
 541 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 542 {
 543         struct ipv4_5tuple key;
 544         struct tcp_hdr *tcp;
 545         struct udp_hdr *udp;
 546         int ret = 0;
 547
 548         key.ip_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
 549         key.ip_src = rte_be_to_cpu_32(ipv4_hdr->src_addr);
 550         key.proto = ipv4_hdr->next_proto_id;
 551
 552         switch (ipv4_hdr->next_proto_id) {
 553         case IPPROTO_TCP:
 554                 tcp = (struct tcp_hdr *)((unsigned char *) ipv4_hdr +
 555                                         sizeof(struct ipv4_hdr));
 556                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 557                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 558                 break;
 559
 560         case IPPROTO_UDP:
 561                 udp = (struct udp_hdr *)((unsigned char *) ipv4_hdr +
 562                                         sizeof(struct ipv4_hdr));
 563                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 564                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 565                 break;
 566
 567         default:
 568                 key.port_dst = 0;
 569                 key.port_src = 0;
 570                 break;
 571         }
 572
 573         /* Find destination port */
 574         ret = rte_hash_lookup(ipv4_l3fwd_lookup_struct, (const void *)&key);
 575         return (uint8_t)((ret < 0)? portid : ipv4_l3fwd_out_if[ret]);
 576 }
 577
 578 static inline uint8_t
 579 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup_struct_t * ipv6_l3fwd_lookup_struct)
 580 {
 581         struct ipv6_5tuple key;
 582         struct tcp_hdr *tcp;
 583         struct udp_hdr *udp;
 584         int ret = 0;
 585
 586         memcpy(key.ip_dst, ipv6_hdr->dst_addr, IPV6_ADDR_LEN);
 587         memcpy(key.ip_src, ipv6_hdr->src_addr, IPV6_ADDR_LEN);
 588
 589         key.proto = ipv6_hdr->proto;
 590
 591         switch (ipv6_hdr->proto) {
 592         case IPPROTO_TCP:
 593                 tcp = (struct tcp_hdr *)((unsigned char *) ipv6_hdr +
 594                                         sizeof(struct ipv6_hdr));
 595                 key.port_dst = rte_be_to_cpu_16(tcp->dst_port);
 596                 key.port_src = rte_be_to_cpu_16(tcp->src_port);
 597                 break;
 598
 599         case IPPROTO_UDP:
 600                 udp = (struct udp_hdr *)((unsigned char *) ipv6_hdr +
 601                                         sizeof(struct ipv6_hdr));
 602                 key.port_dst = rte_be_to_cpu_16(udp->dst_port);
 603                 key.port_src = rte_be_to_cpu_16(udp->src_port);
 604                 break;
 605
 606         default:
 607                 key.port_dst = 0;
 608                 key.port_src = 0;
 609                 break;
 610         }
 611
 612         /* Find destination port */
 613         ret = rte_hash_lookup(ipv6_l3fwd_lookup_struct, (const void *)&key);
 614         return (uint8_t)((ret < 0)? portid : ipv6_l3fwd_out_if[ret]);
 615 }
 616 #endif
 617
 618 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
 619 static inline uint8_t
 620 get_ipv4_dst_port(struct ipv4_hdr *ipv4_hdr,  uint8_t portid, lookup_struct_t * ipv4_l3fwd_lookup_struct)
 621 {
 622         uint8_t next_hop;
 623
 624         return (uint8_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
 625                         rte_be_to_cpu_32(ipv4_hdr->dst_addr), &next_hop) == 0)?
 626                         next_hop : portid);
 627 }
 628
 629 static inline uint8_t
 630 get_ipv6_dst_port(struct ipv6_hdr *ipv6_hdr,  uint8_t portid, lookup6_struct_t * ipv6_l3fwd_lookup_struct)
 631 {
 632         uint8_t next_hop;
 633
 634         return (uint8_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 635                         ipv6_hdr->dst_addr, &next_hop) == 0)?
 636                         next_hop : portid);
 637 }
 638 #endif
 639
 640 static inline void
 641 l3fwd_simple_forward(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
 642         struct lcore_conf *qconf, uint64_t tms)
 643 {
 644         struct ether_hdr *eth_hdr;
 645         struct ipv4_hdr *ipv4_hdr;
 646         void *d_addr_bytes;
 647         uint8_t dst_port;
 648
 649         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 650
 651         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 652                 /* Handle IPv4 headers.*/
 653                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 654
 655 #ifdef DO_RFC_1812_CHECKS
 656                 /* Check to make sure the packet is valid (RFC1812) */
 657                 if (is_valid_ipv4_pkt(ipv4_hdr, m->pkt.pkt_len) < 0) {
 658                         rte_pktmbuf_free(m);
 659                         return;
 660                 }
 661
 662                 /* Update time to live and header checksum */
 663                 --(ipv4_hdr->time_to_live);
 664                 ++(ipv4_hdr->hdr_checksum);
 665 #endif
 666
 667                  /* if it is a fragmented packet, then try to reassemble. */
 668                 if (rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)) {
 669
 670                         struct rte_mbuf *mo;
 671                         struct rte_ip_frag_tbl *tbl;
 672                         struct rte_ip_frag_death_row *dr;
 673
 674                         tbl = qconf->frag_tbl[queue];
 675                         dr = &qconf->death_row;
 676
 677                         /* prepare mbuf: setup l2_len/l3_len. */
 678                         m->pkt.vlan_macip.f.l2_len = sizeof(*eth_hdr);
 679                         m->pkt.vlan_macip.f.l3_len = sizeof(*ipv4_hdr);
 680
 681                         /* process this fragment. */
 682                         if ((mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms,
 683                                         ipv4_hdr)) == NULL)
 684                                 /* no packet to send out. */
 685                                 return;
 686
 687                         /* we have our packet reassembled. */
 688                         if (mo != m) {
 689                                 m = mo;
 690                                 eth_hdr = rte_pktmbuf_mtod(m,
 691                                         struct ether_hdr *);
 692                                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 693                         }
 694                 }
 695
 696                 dst_port = get_ipv4_dst_port(ipv4_hdr, portid,
 697                         qconf->ipv4_lookup_struct);
 698                 if (dst_port >= MAX_PORTS ||
 699                                 (enabled_port_mask & 1 << dst_port) == 0)
 700                         dst_port = portid;
 701
 702                 /* 02:00:00:00:00:xx */
 703                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 704                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 705
 706                 /* src addr */
 707                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 708
 709                 send_single_packet(m, dst_port);
 710         }
 711         else {
 712                 /* Handle IPv6 headers.*/
 713                 struct ipv6_hdr *ipv6_hdr;
 714
 715                 ipv6_hdr = (struct ipv6_hdr *)(rte_pktmbuf_mtod(m, unsigned char *) +
 716                                 sizeof(struct ether_hdr));
 717
 718                 dst_port = get_ipv6_dst_port(ipv6_hdr, portid, qconf->ipv6_lookup_struct);
 719
 720                 if (dst_port >= MAX_PORTS || (enabled_port_mask & 1 << dst_port) == 0)
 721                         dst_port = portid;
 722
 723                 /* 02:00:00:00:00:xx */
 724                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 725                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 726
 727                 /* src addr */
 728                 ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 729
 730                 send_single_packet(m, dst_port);
 731         }
 732
 733 }
 734
 735 /* main processing loop */
 736 static int
 737 main_loop(__attribute__((unused)) void *dummy)
 738 {
 739         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 740         unsigned lcore_id;
 741         uint64_t diff_tsc, cur_tsc, prev_tsc;
 742         int i, j, nb_rx;
 743         uint8_t portid, queueid;
 744         struct lcore_conf *qconf;
 745         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 746
 747         prev_tsc = 0;
 748
 749         lcore_id = rte_lcore_id();
 750         qconf = &lcore_conf[lcore_id];
 751
 752         if (qconf->n_rx_queue == 0) {
 753                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 754                 return 0;
 755         }
 756
 757         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 758
 759         for (i = 0; i < qconf->n_rx_queue; i++) {
 760
 761                 portid = qconf->rx_queue_list[i].port_id;
 762                 queueid = qconf->rx_queue_list[i].queue_id;
 763                 RTE_LOG(INFO, L3FWD, " -- lcoreid=%u portid=%hhu rxqueueid=%hhu\n", lcore_id,
 764                         portid, queueid);
 765         }
 766
 767         while (1) {
 768
 769                 cur_tsc = rte_rdtsc();
 770
 771                 /*
 772                  * TX burst queue drain
 773                  */
 774                 diff_tsc = cur_tsc - prev_tsc;
 775                 if (unlikely(diff_tsc > drain_tsc)) {
 776
 777                         /*
 778                          * This could be optimized (use queueid instead of
 779                          * portid), but it is not called so often
 780                          */
 781                         for (portid = 0; portid < MAX_PORTS; portid++) {
 782                                 if ((enabled_port_mask & (1 << portid)) != 0)
 783                                         send_burst(qconf, 1, portid);
 784                         }
 785
 786                         prev_tsc = cur_tsc;
 787                 }
 788
 789                 /*
 790                  * Read packet from RX queues
 791                  */
 792                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 793
 794                         portid = qconf->rx_queue_list[i].port_id;
 795                         queueid = qconf->rx_queue_list[i].queue_id;
 796
 797                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 798                                 MAX_PKT_BURST);
 799
 800                         /* Prefetch first packets */
 801                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 802                                 rte_prefetch0(rte_pktmbuf_mtod(
 803                                                 pkts_burst[j], void *));
 804                         }
 805
 806                         /* Prefetch and forward already prefetched packets */
 807                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 808                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 809                                         j + PREFETCH_OFFSET], void *));
 810                                 l3fwd_simple_forward(pkts_burst[j], portid,
 811                                         i, qconf, cur_tsc);
 812                         }
 813
 814                         /* Forward remaining prefetched packets */
 815                         for (; j < nb_rx; j++) {
 816                                 l3fwd_simple_forward(pkts_burst[j], portid,
 817                                         i, qconf, cur_tsc);
 818                         }
 819
 820                         rte_ip_frag_free_death_row(&qconf->death_row,
 821                                 PREFETCH_OFFSET);
 822                 }
 823         }
 824 }
 825
 826 static int
 827 check_lcore_params(void)
 828 {
 829         uint8_t queue, lcore;
 830         uint16_t i;
 831         int socketid;
 832
 833         for (i = 0; i < nb_lcore_params; ++i) {
 834                 queue = lcore_params[i].queue_id;
 835                 if (queue >= MAX_RX_QUEUE_PER_PORT) {
 836                         printf("invalid queue number: %hhu\n", queue);
 837                         return -1;
 838                 }
 839                 lcore = lcore_params[i].lcore_id;
 840                 if (!rte_lcore_is_enabled(lcore)) {
 841                         printf("error: lcore %hhu is not enabled in lcore mask\n", lcore);
 842                         return -1;
 843                 }
 844                 if ((socketid = rte_lcore_to_socket_id(lcore) != 0) &&
 845                         (numa_on == 0)) {
 846                         printf("warning: lcore %hhu is on socket %d with numa off \n",
 847                                 lcore, socketid);
 848                 }
 849         }
 850         return 0;
 851 }
 852
 853 static int
 854 check_port_config(const unsigned nb_ports)
 855 {
 856         unsigned portid;
 857         uint16_t i;
 858
 859         for (i = 0; i < nb_lcore_params; ++i) {
 860                 portid = lcore_params[i].port_id;
 861                 if ((enabled_port_mask & (1 << portid)) == 0) {
 862                         printf("port %u is not enabled in port mask\n", portid);
 863                         return -1;
 864                 }
 865                 if (portid >= nb_ports) {
 866                         printf("port %u is not present on the board\n", portid);
 867                         return -1;
 868                 }
 869         }
 870         return 0;
 871 }
 872
 873 static uint8_t
 874 get_port_n_rx_queues(const uint8_t port)
 875 {
 876         int queue = -1;
 877         uint16_t i;
 878
 879         for (i = 0; i < nb_lcore_params; ++i) {
 880                 if (lcore_params[i].port_id == port && lcore_params[i].queue_id > queue)
 881                         queue = lcore_params[i].queue_id;
 882         }
 883         return (uint8_t)(++queue);
 884 }
 885
 886 static int
 887 init_lcore_rx_queues(void)
 888 {
 889         uint16_t i, nb_rx_queue;
 890         uint8_t lcore;
 891
 892         for (i = 0; i < nb_lcore_params; ++i) {
 893                 lcore = lcore_params[i].lcore_id;
 894                 nb_rx_queue = lcore_conf[lcore].n_rx_queue;
 895                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 896                         printf("error: too many queues (%u) for lcore: %u\n",
 897                                 (unsigned)nb_rx_queue + 1, (unsigned)lcore);
 898                         return -1;
 899                 } else {
 900                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 901                                 lcore_params[i].port_id;
 902                         lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 903                                 lcore_params[i].queue_id;
 904                         lcore_conf[lcore].n_rx_queue++;
 905                 }
 906         }
 907         return 0;
 908 }
 909
 910 /* display usage */
 911 static void
 912 print_usage(const char *prgname)
 913 {
 914         printf ("%s [EAL options] -- -p PORTMASK -P"
 915                 "  [--config (port,queue,lcore)[,(port,queue,lcore]]"
 916                 "  [--enable-jumbo [--max-pkt-len PKTLEN]]"
 917                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 918                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 919                 "  -P : enable promiscuous mode\n"
 920                 "  --config (port,queue,lcore): rx queues configuration\n"
 921                 "  --no-numa: optional, disable numa awareness\n"
 922                 "  --enable-jumbo: enable jumbo frame"
 923                 " which max packet len is PKTLEN in decimal (64-9600)\n"
 924                 "  --maxflows=<flows>: optional, maximum number of flows "
 925                 "supported\n"
 926                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 927                 "flow\n",
 928                 prgname);
 929 }
 930
 931 static uint32_t
 932 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 933 {
 934         char *end;
 935         uint64_t v;
 936
 937         /* parse decimal string */
 938         errno = 0;
 939         v = strtoul(str, &end, 10);
 940         if (errno != 0 || *end != '\0')
 941                 return (-EINVAL);
 942
 943         if (v < min || v > max)
 944                 return (-EINVAL);
 945
 946         *val = (uint32_t)v;
 947         return (0);
 948 }
 949
 950 static int
 951 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 952 {
 953         char *end;
 954         uint64_t v;
 955
 956         static const char frmt_sec[] = "s";
 957         static const char frmt_msec[] = "ms";
 958
 959         /* parse decimal string */
 960         errno = 0;
 961         v = strtoul(str, &end, 10);
 962         if (errno != 0)
 963                 return (-EINVAL);
 964
 965         if (*end != '\0') {
 966                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 967                         v *= MS_PER_S;
 968                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 969                         return (-EINVAL);
 970         }
 971
 972         if (v < min || v > max)
 973                 return (-EINVAL);
 974
 975         *val = (uint32_t)v;
 976         return (0);
 977 }
 978
 979
 980 static int parse_max_pkt_len(const char *pktlen)
 981 {
 982         char *end = NULL;
 983         unsigned long len;
 984
 985         /* parse decimal string */
 986         len = strtoul(pktlen, &end, 10);
 987         if ((pktlen[0] == '\0') || (end == NULL) || (*end != '\0'))
 988                 return -1;
 989
 990         if (len == 0)
 991                 return -1;
 992
 993         return len;
 994 }
 995
 996 static int
 997 parse_portmask(const char *portmask)
 998 {
 999         char *end = NULL;
1000         unsigned long pm;
1001
1002         /* parse hexadecimal string */
1003         pm = strtoul(portmask, &end, 16);
1004         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1005                 return -1;
1006
1007         if (pm == 0)
1008                 return -1;
1009
1010         return pm;
1011 }
1012
1013 static int
1014 parse_config(const char *q_arg)
1015 {
1016         char s[256];
1017         const char *p, *p0 = q_arg;
1018         char *end;
1019         enum fieldnames {
1020                 FLD_PORT = 0,
1021                 FLD_QUEUE,
1022                 FLD_LCORE,
1023                 _NUM_FLD
1024         };
1025         unsigned long int_fld[_NUM_FLD];
1026         char *str_fld[_NUM_FLD];
1027         int i;
1028         unsigned size;
1029
1030         nb_lcore_params = 0;
1031
1032         while ((p = strchr(p0,'(')) != NULL) {
1033                 ++p;
1034                 if((p0 = strchr(p,')')) == NULL)
1035                         return -1;
1036
1037                 size = p0 - p;
1038                 if(size >= sizeof(s))
1039                         return -1;
1040
1041                 rte_snprintf(s, sizeof(s), "%.*s", size, p);
1042                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') != _NUM_FLD)
1043                         return -1;
1044                 for (i = 0; i < _NUM_FLD; i++){
1045                         errno = 0;
1046                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1047                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1048                                 return -1;
1049                 }
1050                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1051                         printf("exceeded max number of lcore params: %hu\n",
1052                                 nb_lcore_params);
1053                         return -1;
1054                 }
1055                 lcore_params_array[nb_lcore_params].port_id = (uint8_t)int_fld[FLD_PORT];
1056                 lcore_params_array[nb_lcore_params].queue_id = (uint8_t)int_fld[FLD_QUEUE];
1057                 lcore_params_array[nb_lcore_params].lcore_id = (uint8_t)int_fld[FLD_LCORE];
1058                 ++nb_lcore_params;
1059         }
1060         lcore_params = lcore_params_array;
1061         return 0;
1062 }
1063
1064 /* Parse the argument given in the command line of the application */
1065 static int
1066 parse_args(int argc, char **argv)
1067 {
1068         int opt, ret;
1069         char **argvopt;
1070         int option_index;
1071         char *prgname = argv[0];
1072         static struct option lgopts[] = {
1073                 {"config", 1, 0, 0},
1074                 {"no-numa", 0, 0, 0},
1075                 {"enable-jumbo", 0, 0, 0},
1076                 {"maxflows", 1, 0, 0},
1077                 {"flowttl", 1, 0, 0},
1078                 {NULL, 0, 0, 0}
1079         };
1080
1081         argvopt = argv;
1082
1083         while ((opt = getopt_long(argc, argvopt, "p:P",
1084                                 lgopts, &option_index)) != EOF) {
1085
1086                 switch (opt) {
1087                 /* portmask */
1088                 case 'p':
1089                         enabled_port_mask = parse_portmask(optarg);
1090                         if (enabled_port_mask == 0) {
1091                                 printf("invalid portmask\n");
1092                                 print_usage(prgname);
1093                                 return -1;
1094                         }
1095                         break;
1096                 case 'P':
1097                         printf("Promiscuous mode selected\n");
1098                         promiscuous_on = 1;
1099                         break;
1100
1101                 /* long options */
1102                 case 0:
1103                         if (!strncmp(lgopts[option_index].name, "config", 6)) {
1104                                 ret = parse_config(optarg);
1105                                 if (ret) {
1106                                         printf("invalid config\n");
1107                                         print_usage(prgname);
1108                                         return -1;
1109                                 }
1110                         }
1111
1112                         if (!strncmp(lgopts[option_index].name, "no-numa", 7)) {
1113                                 printf("numa is disabled \n");
1114                                 numa_on = 0;
1115                         }
1116
1117                         if (!strncmp(lgopts[option_index].name,
1118                                         "maxflows", 8)) {
1119                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
1120                                                 MAX_FLOW_NUM,
1121                                                 &max_flow_num)) != 0) {
1122                                         printf("invalid value: \"%s\" for "
1123                                                 "parameter %s\n",
1124                                                 optarg,
1125                                                 lgopts[option_index].name);
1126                                         print_usage(prgname);
1127                                         return (ret);
1128                                 }
1129                         }
1130
1131                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
1132                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
1133                                                 MAX_FLOW_TTL,
1134                                                 &max_flow_ttl)) != 0) {
1135                                         printf("invalid value: \"%s\" for "
1136                                                 "parameter %s\n",
1137                                                 optarg,
1138                                                 lgopts[option_index].name);
1139                                         print_usage(prgname);
1140                                         return (ret);
1141                                 }
1142                         }
1143
1144                         if (!strncmp(lgopts[option_index].name, "enable-jumbo", 12)) {
1145                                 struct option lenopts = {"max-pkt-len", required_argument, 0, 0};
1146
1147                                 printf("jumbo frame is enabled \n");
1148                                 port_conf.rxmode.jumbo_frame = 1;
1149
1150                                 /* if no max-pkt-len set, use the default value ETHER_MAX_LEN */
1151                                 if (0 == getopt_long(argc, argvopt, "", &lenopts, &option_index)) {
1152                                         ret = parse_max_pkt_len(optarg);
1153                                         if ((ret < 64) || (ret > MAX_JUMBO_PKT_LEN)){
1154                                                 printf("invalid packet length\n");
1155                                                 print_usage(prgname);
1156                                                 return -1;
1157                                         }
1158                                         port_conf.rxmode.max_rx_pkt_len = ret;
1159                                 }
1160                                 printf("set jumbo frame max packet length to %u\n",
1161                                                 (unsigned int)port_conf.rxmode.max_rx_pkt_len);
1162                         }
1163
1164                         break;
1165
1166                 default:
1167                         print_usage(prgname);
1168                         return -1;
1169                 }
1170         }
1171
1172         if (optind >= 0)
1173                 argv[optind-1] = prgname;
1174
1175         ret = optind-1;
1176         optind = 0; /* reset getopt lib */
1177         return ret;
1178 }
1179
1180 static void
1181 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1182 {
1183         printf ("%s%02X:%02X:%02X:%02X:%02X:%02X", name,
1184                 eth_addr->addr_bytes[0],
1185                 eth_addr->addr_bytes[1],
1186                 eth_addr->addr_bytes[2],
1187                 eth_addr->addr_bytes[3],
1188                 eth_addr->addr_bytes[4],
1189                 eth_addr->addr_bytes[5]);
1190 }
1191
1192 #if (APP_LOOKUP_METHOD == APP_LOOKUP_EXACT_MATCH)
1193 static void
1194 setup_hash(int socketid)
1195 {
1196         struct rte_hash_parameters ipv4_l3fwd_hash_params = {
1197                 .name = NULL,
1198                 .entries = L3FWD_HASH_ENTRIES,
1199                 .bucket_entries = 4,
1200                 .key_len = sizeof(struct ipv4_5tuple),
1201                 .hash_func = DEFAULT_HASH_FUNC,
1202                 .hash_func_init_val = 0,
1203         };
1204
1205         struct rte_hash_parameters ipv6_l3fwd_hash_params = {
1206                 .name = NULL,
1207                 .entries = L3FWD_HASH_ENTRIES,
1208                 .bucket_entries = 4,
1209                 .key_len = sizeof(struct ipv6_5tuple),
1210                 .hash_func = DEFAULT_HASH_FUNC,
1211                 .hash_func_init_val = 0,
1212         };
1213
1214         unsigned i;
1215         int ret;
1216         char s[64];
1217
1218         /* create ipv4 hash */
1219         rte_snprintf(s, sizeof(s), "ipv4_l3fwd_hash_%d", socketid);
1220         ipv4_l3fwd_hash_params.name = s;
1221         ipv4_l3fwd_hash_params.socket_id = socketid;
1222         ipv4_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv4_l3fwd_hash_params);
1223         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1224                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1225                                 "socket %d\n", socketid);
1226
1227         /* create ipv6 hash */
1228         rte_snprintf(s, sizeof(s), "ipv6_l3fwd_hash_%d", socketid);
1229         ipv6_l3fwd_hash_params.name = s;
1230         ipv6_l3fwd_hash_params.socket_id = socketid;
1231         ipv6_l3fwd_lookup_struct[socketid] = rte_hash_create(&ipv6_l3fwd_hash_params);
1232         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1233                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd hash on "
1234                                 "socket %d\n", socketid);
1235
1236
1237         /* populate the ipv4 hash */
1238         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1239                 ret = rte_hash_add_key (ipv4_l3fwd_lookup_struct[socketid],
1240                                 (void *) &ipv4_l3fwd_route_array[i].key);
1241                 if (ret < 0) {
1242                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1243                                 "l3fwd hash on socket %d\n", i, socketid);
1244                 }
1245                 ipv4_l3fwd_out_if[ret] = ipv4_l3fwd_route_array[i].if_out;
1246                 printf("Hash: Adding key\n");
1247                 print_ipv4_key(ipv4_l3fwd_route_array[i].key);
1248         }
1249
1250         /* populate the ipv6 hash */
1251         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1252                 ret = rte_hash_add_key (ipv6_l3fwd_lookup_struct[socketid],
1253                                 (void *) &ipv6_l3fwd_route_array[i].key);
1254                 if (ret < 0) {
1255                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the"
1256                                 "l3fwd hash on socket %d\n", i, socketid);
1257                 }
1258                 ipv6_l3fwd_out_if[ret] = ipv6_l3fwd_route_array[i].if_out;
1259                 printf("Hash: Adding key\n");
1260                 print_ipv6_key(ipv6_l3fwd_route_array[i].key);
1261         }
1262 }
1263 #endif
1264
1265 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1266 static void
1267 setup_lpm(int socketid)
1268 {
1269         struct rte_lpm6_config config;
1270         unsigned i;
1271         int ret;
1272         char s[64];
1273
1274         /* create the LPM table */
1275         rte_snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
1276         ipv4_l3fwd_lookup_struct[socketid] = rte_lpm_create(s, socketid,
1277                                 IPV4_L3FWD_LPM_MAX_RULES, 0);
1278         if (ipv4_l3fwd_lookup_struct[socketid] == NULL)
1279                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1280                                 " on socket %d\n", socketid);
1281
1282         /* populate the LPM table */
1283         for (i = 0; i < IPV4_L3FWD_NUM_ROUTES; i++) {
1284                 ret = rte_lpm_add(ipv4_l3fwd_lookup_struct[socketid],
1285                         ipv4_l3fwd_route_array[i].ip,
1286                         ipv4_l3fwd_route_array[i].depth,
1287                         ipv4_l3fwd_route_array[i].if_out);
1288
1289                 if (ret < 0) {
1290                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1291                                 "l3fwd LPM table on socket %d\n",
1292                                 i, socketid);
1293                 }
1294
1295                 printf("LPM: Adding route 0x%08x / %d (%d)\n",
1296                         (unsigned)ipv4_l3fwd_route_array[i].ip,
1297                         ipv4_l3fwd_route_array[i].depth,
1298                         ipv4_l3fwd_route_array[i].if_out);
1299         }
1300
1301         /* create the LPM6 table */
1302         rte_snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
1303
1304         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
1305         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
1306         config.flags = 0;
1307         ipv6_l3fwd_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
1308                                 &config);
1309         if (ipv6_l3fwd_lookup_struct[socketid] == NULL)
1310                 rte_exit(EXIT_FAILURE, "Unable to create the l3fwd LPM table"
1311                                 " on socket %d\n", socketid);
1312
1313         /* populate the LPM table */
1314         for (i = 0; i < IPV6_L3FWD_NUM_ROUTES; i++) {
1315                 ret = rte_lpm6_add(ipv6_l3fwd_lookup_struct[socketid],
1316                         ipv6_l3fwd_route_array[i].ip,
1317                         ipv6_l3fwd_route_array[i].depth,
1318                         ipv6_l3fwd_route_array[i].if_out);
1319
1320                 if (ret < 0) {
1321                         rte_exit(EXIT_FAILURE, "Unable to add entry %u to the "
1322                                 "l3fwd LPM table on socket %d\n",
1323                                 i, socketid);
1324                 }
1325
1326                 printf("LPM: Adding route %s / %d (%d)\n",
1327                         "IPV6",
1328                         ipv6_l3fwd_route_array[i].depth,
1329                         ipv6_l3fwd_route_array[i].if_out);
1330         }
1331 }
1332 #endif
1333
1334 static int
1335 init_mem(void)
1336 {
1337         struct lcore_conf *qconf;
1338         int socketid;
1339         unsigned lcore_id;
1340
1341         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1342                 if (rte_lcore_is_enabled(lcore_id) == 0)
1343                         continue;
1344
1345                 if (numa_on)
1346                         socketid = rte_lcore_to_socket_id(lcore_id);
1347                 else
1348                         socketid = 0;
1349
1350                 if (socketid >= NB_SOCKETS) {
1351                         rte_exit(EXIT_FAILURE,
1352                                 "Socket %d of lcore %u is out of range %d\n",
1353                                 socketid, lcore_id, NB_SOCKETS);
1354                 }
1355
1356 #if (APP_LOOKUP_METHOD == APP_LOOKUP_LPM)
1357                         setup_lpm(socketid);
1358 #else
1359                         setup_hash(socketid);
1360 #endif
1361                 qconf = &lcore_conf[lcore_id];
1362                 qconf->ipv4_lookup_struct = ipv4_l3fwd_lookup_struct[socketid];
1363                 qconf->ipv6_lookup_struct = ipv6_l3fwd_lookup_struct[socketid];
1364         }
1365         return 0;
1366 }
1367
1368 /* Check the link status of all ports in up to 9s, and print them finally */
1369 static void
1370 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
1371 {
1372 #define CHECK_INTERVAL 100 /* 100ms */
1373 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1374         uint8_t portid, count, all_ports_up, print_flag = 0;
1375         struct rte_eth_link link;
1376
1377         printf("\nChecking link status");
1378         fflush(stdout);
1379         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1380                 all_ports_up = 1;
1381                 for (portid = 0; portid < port_num; portid++) {
1382                         if ((port_mask & (1 << portid)) == 0)
1383                                 continue;
1384                         memset(&link, 0, sizeof(link));
1385                         rte_eth_link_get_nowait(portid, &link);
1386                         /* print link status if flag set */
1387                         if (print_flag == 1) {
1388                                 if (link.link_status)
1389                                         printf("Port %d Link Up - speed %u "
1390                                                 "Mbps - %s\n", (uint8_t)portid,
1391                                                 (unsigned)link.link_speed,
1392                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1393                                         ("full-duplex") : ("half-duplex\n"));
1394                                 else
1395                                         printf("Port %d Link Down\n",
1396                                                 (uint8_t)portid);
1397                                 continue;
1398                         }
1399                         /* clear all_ports_up flag if any link down */
1400                         if (link.link_status == 0) {
1401                                 all_ports_up = 0;
1402                                 break;
1403                         }
1404                 }
1405                 /* after finally printing all link status, get out */
1406                 if (print_flag == 1)
1407                         break;
1408
1409                 if (all_ports_up == 0) {
1410                         printf(".");
1411                         fflush(stdout);
1412                         rte_delay_ms(CHECK_INTERVAL);
1413                 }
1414
1415                 /* set the print_flag if all ports up or timeout */
1416                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1417                         print_flag = 1;
1418                         printf("done\n");
1419                 }
1420         }
1421 }
1422 static void
1423 setup_port_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1424         uint32_t port)
1425 {
1426         struct mbuf_table *mtb;
1427         uint32_t n;
1428         size_t sz;
1429
1430         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
1431         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
1432
1433         if ((mtb = rte_zmalloc_socket(__func__, sz, CACHE_LINE_SIZE,
1434                         socket)) == NULL)
1435                 rte_exit(EXIT_FAILURE, "%s() for lcore: %u, port: %u "
1436                         "failed to allocate %zu bytes\n",
1437                         __func__, lcore, port, sz);
1438
1439         mtb->len = n;
1440         qconf->tx_mbufs[port] = mtb;
1441 }
1442
1443 static void
1444 setup_queue_tbl(struct lcore_conf *qconf, uint32_t lcore, int socket,
1445         uint32_t queue)
1446 {
1447         uint32_t nb_mbuf;
1448         uint64_t frag_cycles;
1449         char buf[RTE_MEMPOOL_NAMESIZE];
1450
1451         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
1452                 max_flow_ttl;
1453
1454         if ((qconf->frag_tbl[queue] = rte_ip_frag_table_create(max_flow_num,
1455                         IPV4_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
1456                         socket)) == NULL)
1457                 rte_exit(EXIT_FAILURE, "ipv4_frag_tbl_create(%u) on "
1458                         "lcore: %u for queue: %u failed\n",
1459                         max_flow_num, lcore, queue);
1460
1461         /*
1462          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM - 1)>
1463          * mbufs could be stored int the fragment table.
1464          * Plus, each TX queue can hold up to <max_flow_num> packets.
1465          */
1466
1467         nb_mbuf = 2 * RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) *
1468                         RTE_LIBRTE_IP_FRAG_MAX_FRAG;
1469         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
1470         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
1471
1472         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)DEF_MBUF_NUM);
1473
1474         rte_snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
1475
1476         if ((qconf->pool[queue] = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
1477                         sizeof(struct rte_pktmbuf_pool_private),
1478                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
1479                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL)
1480                 rte_exit(EXIT_FAILURE, "mempool_create(%s) failed", buf);
1481 }
1482
1483 static void
1484 queue_dump_stat(void)
1485 {
1486         uint32_t i, lcore;
1487         const struct lcore_conf *qconf;
1488
1489         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
1490                 if (rte_lcore_is_enabled(lcore) == 0)
1491                         continue;
1492
1493                 qconf = lcore_conf + lcore;
1494                 for (i = 0; i < qconf->n_rx_queue; i++) {
1495
1496                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
1497                                 "rxqueueid=%hhu frag tbl stat:\n",
1498                                 lcore,  qconf->rx_queue_list[i].port_id,
1499                                 qconf->rx_queue_list[i].queue_id);
1500                         rte_ip_frag_table_statistics_dump(stdout, qconf->frag_tbl[i]);
1501                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
1502                                 "TX packets _queued:\t%" PRIu64 "\n"
1503                                 "TX packets dropped:\t%" PRIu64 "\n"
1504                                 "TX packets send:\t%" PRIu64 "\n",
1505                                 qconf->tx_stat.call,
1506                                 qconf->tx_stat.queue,
1507                                 qconf->tx_stat.drop,
1508                                 qconf->tx_stat.send);
1509                 }
1510         }
1511 }
1512
1513 static void
1514 signal_handler(int signum)
1515 {
1516         queue_dump_stat();
1517         if (signum != SIGUSR1)
1518                 rte_exit(0, "received signal: %d, exiting\n", signum);
1519 }
1520
1521 int
1522 MAIN(int argc, char **argv)
1523 {
1524         struct lcore_conf *qconf;
1525         int ret;
1526         unsigned nb_ports;
1527         uint16_t queueid;
1528         unsigned lcore_id;
1529         uint32_t n_tx_queue, nb_lcores;
1530         uint8_t portid, nb_rx_queue, queue, socketid;
1531
1532         /* init EAL */
1533         ret = rte_eal_init(argc, argv);
1534         if (ret < 0)
1535                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1536         argc -= ret;
1537         argv += ret;
1538
1539         /* parse application arguments (after the EAL ones) */
1540         ret = parse_args(argc, argv);
1541         if (ret < 0)
1542                 rte_exit(EXIT_FAILURE, "Invalid L3FWD parameters\n");
1543
1544         if (check_lcore_params() < 0)
1545                 rte_exit(EXIT_FAILURE, "check_lcore_params failed\n");
1546
1547         ret = init_lcore_rx_queues();
1548         if (ret < 0)
1549                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1550
1551
1552         if (rte_eal_pci_probe() < 0)
1553                 rte_exit(EXIT_FAILURE, "Cannot probe PCI\n");
1554
1555         nb_ports = rte_eth_dev_count();
1556         if (nb_ports > MAX_PORTS)
1557                 nb_ports = MAX_PORTS;
1558
1559         if (check_port_config(nb_ports) < 0)
1560                 rte_exit(EXIT_FAILURE, "check_port_config failed\n");
1561
1562         nb_lcores = rte_lcore_count();
1563
1564         /* initialize all ports */
1565         for (portid = 0; portid < nb_ports; portid++) {
1566                 /* skip ports that are not enabled */
1567                 if ((enabled_port_mask & (1 << portid)) == 0) {
1568                         printf("\nSkipping disabled port %d\n", portid);
1569                         continue;
1570                 }
1571
1572                 /* init port */
1573                 printf("Initializing port %d ... ", portid );
1574                 fflush(stdout);
1575
1576                 nb_rx_queue = get_port_n_rx_queues(portid);
1577                 n_tx_queue = nb_lcores;
1578                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1579                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1580                 printf("Creating queues: nb_rxq=%d nb_txq=%u... ",
1581                         nb_rx_queue, (unsigned)n_tx_queue );
1582                 ret = rte_eth_dev_configure(portid, nb_rx_queue,
1583                                         (uint16_t)n_tx_queue, &port_conf);
1584                 if (ret < 0)
1585                         rte_exit(EXIT_FAILURE, "Cannot configure device: err=%d, port=%d\n",
1586                                 ret, portid);
1587
1588                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1589                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1590                 printf(", ");
1591
1592                 /* init memory */
1593                 ret = init_mem();
1594                 if (ret < 0)
1595                         rte_exit(EXIT_FAILURE, "init_mem failed\n");
1596
1597                 /* init one TX queue per couple (lcore,port) */
1598                 queueid = 0;
1599                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1600                         if (rte_lcore_is_enabled(lcore_id) == 0)
1601                                 continue;
1602
1603                         if (numa_on)
1604                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1605                         else
1606                                 socketid = 0;
1607
1608                         printf("txq=%u,%d,%d ", lcore_id, queueid, socketid);
1609                         fflush(stdout);
1610                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1611                                                      socketid, &tx_conf);
1612                         if (ret < 0)
1613                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1614                                         "port=%d\n", ret, portid);
1615
1616                         qconf = &lcore_conf[lcore_id];
1617                         qconf->tx_queue_id[portid] = queueid;
1618                         setup_port_tbl(qconf, lcore_id, socketid, portid);
1619                         queueid++;
1620                 }
1621                 printf("\n");
1622         }
1623
1624         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1625                 if (rte_lcore_is_enabled(lcore_id) == 0)
1626                         continue;
1627                 qconf = &lcore_conf[lcore_id];
1628                 printf("\nInitializing rx queues on lcore %u ... ", lcore_id );
1629                 fflush(stdout);
1630                 /* init RX queues */
1631                 for(queue = 0; queue < qconf->n_rx_queue; ++queue) {
1632                         portid = qconf->rx_queue_list[queue].port_id;
1633                         queueid = qconf->rx_queue_list[queue].queue_id;
1634
1635                         if (numa_on)
1636                                 socketid = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1637                         else
1638                                 socketid = 0;
1639
1640                         printf("rxq=%d,%d,%d ", portid, queueid, socketid);
1641                         fflush(stdout);
1642
1643                         setup_queue_tbl(qconf, lcore_id, socketid, queue);
1644
1645                         ret = rte_eth_rx_queue_setup(portid, queueid, nb_rxd,
1646                                         socketid, &rx_conf, qconf->pool[queue]);
1647                         if (ret < 0)
1648                                 rte_exit(EXIT_FAILURE,
1649                                         "rte_eth_rx_queue_setup: err=%d,"
1650                                         "port=%d\n", ret, portid);
1651                 }
1652         }
1653
1654         printf("\n");
1655
1656         /* start ports */
1657         for (portid = 0; portid < nb_ports; portid++) {
1658                 if ((enabled_port_mask & (1 << portid)) == 0) {
1659                         continue;
1660                 }
1661                 /* Start device */
1662                 ret = rte_eth_dev_start(portid);
1663                 if (ret < 0)
1664                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1665                                 ret, portid);
1666
1667                 /*
1668                  * If enabled, put device in promiscuous mode.
1669                  * This allows IO forwarding mode to forward packets
1670                  * to itself through 2 cross-connected  ports of the
1671                  * target machine.
1672                  */
1673                 if (promiscuous_on)
1674                         rte_eth_promiscuous_enable(portid);
1675         }
1676
1677         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1678
1679         signal(SIGUSR1, signal_handler);
1680         signal(SIGTERM, signal_handler);
1681         signal(SIGINT, signal_handler);
1682
1683         /* launch per-lcore init on every lcore */
1684         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1685         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1686                 if (rte_eal_wait_lcore(lcore_id) < 0)
1687                         return -1;
1688         }
1689
1690         return 0;
1691 }